(Michele Cini) Topics and Methods in Condensed Mat

Michele Cini
Topics and Methods in Condensed Matter Theory
Michele Cini
Topics and Methods

in Condensed Matter Theory
From Basic Quantum Mechanics
to the Frontiers of Research
With 73 Figures and 4 Tables
123
Prof. Michele Cini

University of Rome
Department of Physics
Via della Ricerca Scientica
00133 Rome
Italy
e-mail: [email protected]
Library of Congress Control Number: 2007925062
ISBN-13 978-3-540-70726-4 Springer Berlin Heidelberg New York

This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microlm or in any other way, and storage in data banks. Duplication of
this publication or parts thereof is permitted only under the provisions of the German Copyright Law
of September 9, 1965, in its current version, and permission for use must always be obtained from
Springer. Violations are liable for prosecution under the German Copyright Law.
Springer is a part of Springer Science+Business Media
springer.com
Springer-Verlag Berlin Heidelberg 2007
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specic statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
Typesetting: Digital data supplied by Author
Production: LE-TEX Jelonek, Schmidt & Vckler GbR, Leipzig
Cover Design: eStudio Calamar S.L., F. Steinen-Broo, Girona, Spain
SPIN 11367314
57/3180/YL
543210
Printed on acid-free paper
To my wife Anna and to my son Massimo
Preface
Quod super est, vacuas auris animumque sagacem

semotum a curis adhibe veram ad rationem,
ne mea dona tibi studio disposta deli,
intellecta prius quam sint, contempta relinquas.
Titus Lucretius Carus, De Rerum Natura Liber Primus
This book is structured as a course for students with a good knowledge of basic Quantum Mechanics who want to specialize in Condensed Matter Theory.
Indeed it takes the reader to advanced levels in several topics, but there is an
obvious trick: no teacher can hope to cover all this material in one semester
without facing a serious and rightful student revolt. I am oering alternative
advanced topics that I have been developing in dierent years of teaching and
updated to the present time. Yet, I had to remove much important material
which is a condicio sine qua non for condensed matter theorists; the most serious sacrice (wisely urged by the Editor) was the removal of all the chapters
in Relativistic Quantum Mechanics and QED that I have been teaching for
many years in Atomic Physics courses but would have made the size of this
work unacceptable as one book. Such topics are covered in excellent Springer
books, like the series by Greiner. Even so, there is no encyclopedic attitude,
or attempt to cover all the hottest advanced topics, but I privileged those
arguments where I did some work and the methods that I used most often
or where I gave some contribution in my long research activity. The central
part of the book is devoted to the theory and applications of symmetry and
Greens functions. In the interest of the class, I present them is such a way
that one can easily separate an introductory part, that might be of interest
to a broader audience including experimentalists, and a more advanced and
demanding part.
Nobody really knows how knowledge grows, particularly in theoretical
physics (a rather special environment to study) but my impression is that
VIII
Preface
it grows by a vivid (if often not very precise initially) understanding of particular problems, that are later combined in more powerful units and made
more exact in the process. More complex phenomena require more mathematical ingenuity, new concepts arise, deep unexpected similarities between
diverse problems are discovered. By such processes in which mathematics and
physics merge into the reasoning process and are a source of inspiration for
each other, one can sometimes predict new facts that are later veried experimentally. In condensed matter physics, predicting new experimental features
is not reserved to geniuses, and one can really hope to achieve such results,
and although the success is not uncommon, it is certainly rewarding. I believe
that there is still lot of space for pioneers and while computing is important
in a quantitative science like physics most real discoveries will continue to
come from intuition and original thought.
Most times it is experiment that brings some unexpected and surprising
result. Superconductivity is a typical example: common sense would have
predicted that the resistance will unavoidably be present as a part of the imperfection of reality. This is the magic of Quantum Mechanics, that continues
to make stunning yet tangible reality by the eect of phases. The role of Berry
phases, ux quantization an the like is discussed mainly in Chapters 16 and
17. There are many phenomena but if we wish to go from phenomenology to
principles progress is conditioned by methods. A large variety of methods have
been generated by many theoreticians. Some are time honored, but enlighten
new problems in new ways: to make practical predictions in the presence of
symmetry we take advantage of abstract Group theory. However, important
new methods continue to be invented, and there is no sign that the gold mine
of ingenuity is exhausted. On the other hand, while an increasing number
of problems are successfully dealt with by the existing codes and computational methods, the ingenuity continues to be required, since new interesting
materials and processes are discovered which require fresh modeling.
Some of the subject matter is included because in my carrier I happened
to work in the subject or to make extensive use of the methods; in this way
I suppose I can provide in some measure a rst-hand presentation of some
topics. In particular, Chapters 6, 15, 13, 14 and 17 are the most original in the
sense of presenting results from my own research; elsewhere I tried to present
mature results in a fresh and attractive way. At least, I hope I will transmit
some of the fun I have to work using such results. The discussion of general
topics like Group theory or Feynman diagrams, the Keldysh theory, which
are discussed in many textbooks, but are traditionally considered rather hard
aims to help the reader by many examples and by direct procedures and
intuitive arguments. I tend to prove everything and therefore the reader will
nd sentences like it can be shown that ... only seldom and for relatively
unimportant side matters. Indeed, in most cases when something is really
understood, there is little diculty in nding a proof; on the other hand,
the converse is also true: having no clear justication often means having no
Preface
IX
real grasp or lacking mastery of the use of the results. However in several
cases, lengthy formal proofs are readily found in the literature and there is
no strong reason to reproduce them here: for instance, Wicks theorem and
the linked cluster theorem fall in this category. Then I prefer to give my own
intuitive arguments, that reect the way I visualize the result for myself, and
reference the literature where the canonical proofs are published.
Finally, I hope the readers will understand that this work costed me a
considerable eort and at some point I had to force the writing process to
converge. It is always possible to improve the book in many ways, but alas at
some point the writing must stop. It was this appeal to the readers understanding, besides the common interest in the Nature of Things, that prompted
to me the above quotation from Titus Lucretius Carus.
It is a pleasure to thank Professor Giancarlo Rossi, Doctor Gianluca Stefanucci and Doctor Yassen Stanev, all at the Physics Department of Rome
Tor Vergata University, and Doctor Claudio Verdozzi, currently at Lund, for
reading and discussing part of the manuscript and giving useful advice and
encouragement.
Rome, May 7, 2007
Michele Cini
Contents
Part I I-Introductory Many-Body Physics

Basic Many-Body Quantum Mechanics . . . . . . . . . . . . . . . . . . .
1.1 Slater Determinants and Matrix Elements . . . . . . . . . . . . . . . . .
1.1.1 Many-electron Matrix Elements . . . . . . . . . . . . . . . . . . . .
1.1.2 Derivation of the Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Second Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 Bosons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.2 Field Quantization and Casimir Eect . . . . . . . . . . . . . .
1.2.3 Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.4 Basis Change in Second Quantization and Field
Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.5 Hubbard Model for the Hydrogen Molecule . . . . . . . . . .
1.3 Schrieer-Wol Canonical Transformation . . . . . . . . . . . . . . . . .
1.4 Variational Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.5 Variational Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.6 Non-degenerate Perturbation Theory . . . . . . . . . . . . . . . . . . . . . .
11
13
15
16
18
18
Adiabatic Switching and Time-Ordered series . . . . . . . . . . . .

2.1 Time-dependent is Better: start from the Golden Age . . . . . . .
2.2 Evolution in Complex Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 Heisenberg Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2.2 Thermal Averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 The Interaction Picture and the Viable Expansion . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
19
21
21
23
25
26
Atomic Shells and Multiplets . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.1 Shell Structure of Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2 Hartree-Fock Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2.1 Physical Meaning of Exchange: the Cohesion of a
Simple Metal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Virial Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.4 Hellmann-Feynman Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5 Central Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

0 Limit) . . . . . . . . . . . . . . . . . . . .
3.5.1 L-S Multiplets (Hrel
29
29
29
1
1
2
4
5
5
7
9
35
37
38
38
40
XII
Contents
3.5.2 Hunds First Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.6 Atomic Coulomb Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.1 Hunds Second Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.2 J-J Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.6.3 Intermediate Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.7 Meitner-Auger Eect and Spectroscopy . . . . . . . . . . . . . . . . . . .
3.7.1 Auger Selection Rules and Line Intensities . . . . . . . . . . .
42
43
46
46
46
48
50
Greens Functions as Thought Experiments . . . . . . . . . . . . . . .

4.1 Greens Theorem for one-Body Problems . . . . . . . . . . . . . . . . . .
4.2 How Many-Body Greens Functions Arise . . . . . . . . . . . . . . . . . .
4.2.1 Correlation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.2 Quantum Greens Functions . . . . . . . . . . . . . . . . . . . . . . .
4.2.3 Quantum Averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.4 Greens functions at Finite Temperature . . . . . . . . . . . .
4.3 Non-interacting Propagators for Solids . . . . . . . . . . . . . . . . . . . .
4.3.1 Greens Functions for Tight-binding Hamiltonians . . . .
4.3.2 Lippmann-Schwinger Equation . . . . . . . . . . . . . . . . . . . . .
4.3.3 t matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.4 Ingleseld Embedding Method . . . . . . . . . . . . . . . . . . . . .
4.4 Kubo Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5 Vacuum Amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6 Lehmann Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6.1 Zero-Temperature Fermi Case . . . . . . . . . . . . . . . . . . . . . .
4.6.2 Finite Temperatures, Fermi and Bose . . . . . . . . . . . . . . .
4.6.3 Fluctuation-Dissipation Theorem . . . . . . . . . . . . . . . . . .
55
55
55
55
57
58
59
61
66
70
70
71
73
75
76
76
78
79
Hopping Electron Models: an Appetizer . . . . . . . . . . . . . . . . . .

5.1 Fano Resonances and Resolvents . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.1 Resonances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.2 Fano Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.3 One-body Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.4 Many-Body Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.5 The Resolvent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1.6 Self-Energy Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Magnetic Impurities and Chemisorption on Transition Metal
Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.3 Strong Coupling and the Kondo Peak . . . . . . . . . . . . . . . . . . . . .
5.3.1 Narrow-Band Anderson Model . . . . . . . . . . . . . . . . . . . . .
5.3.2 Anderson Model, s-d Model and Kondo Model . . . . . . .
5.3.3 Fermi Level Singularity and Kondo Minimum . . . . . . . .
5.4 The Nf 1 Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.1 Kondo Temperature in the Spin-Fluctuation Case . . . .
5.4.2 Density of Occupied States . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
81
81
82
82
83
86
89
90
95
95
98
100
101
105
106
107
Contents
Many-body Eects in Electron Spectroscopies . . . . . . . . . . . .

6.1 Electron Spectroscopy for Chemical Analysis (ESCA) . . . . . . .
6.1.1 Chemical Shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.1.2 Core-Level Splitting in Paramagnetic Molecules . . . . . .
6.1.3 Shake-up, Shake-o, Relaxation . . . . . . . . . . . . . . . . . . .
6.1.4 Lundqvist Model of Phonon and Plasmon Satellites . . .
6.2 Auger CVV Line Shapes: Two-Hole Resonances . . . . . . . . . . . .
6.2.1 Desorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Two Interacting Fermions in a Lattice . . . . . . . . . . . . . . . . . . . .
6.4 Quadratic Response Formalism and Spectroscopies . . . . . . . . .
6.4.1 One-Step Theory of Auger Spectra . . . . . . . . . . . . . . . . .
6.4.2 Plasmon Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
XIII
109
109
111
111
113
115
118
122
123
127
129
130
130
Part II Symmetry in Quantum Physics

7
Group Representations for Physicists . . . . . . . . . . . . . . . . . . . .

7.1 Abstract Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2 Point Symmetry in Molecules and Solids . . . . . . . . . . . . . . . . . .
7.2.1 Symmetry operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.2 Dirac characters and Irreducible Representations . . . . .
7.2.3 Schurs lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.4 Continuous Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.3 Accidental degeneracy and hidden symmetries . . . . . . . . . . . . .
7.4 Great Orthogonality Theorem (GOT) . . . . . . . . . . . . . . . . . . . . .
7.5 Little Orthogonality Theorem (LOT) . . . . . . . . . . . . . . . . . . . . .
7.6 Projection operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.7 Regular representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
133
133
135
135
140
143
144
145
149
152
154
155
156
Simpler Uses of Group Theory . . . . . . . . . . . . . . . . . . . . . . . . . . .

8.1 Molecular Orbitals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.1.1 Molecular Orbitals of N H3 . . . . . . . . . . . . . . . . . . . . . . . .
8.1.2 Molecular Orbitals of CH4 . . . . . . . . . . . . . . . . . . . . . . . .
8.1.3 Characters of Angular Momentum Eigenstates . . . . . . .
8.1.4 Examples: Oh Group, Ligand Group Orbitals, Crystal
Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.2 Normal Modes of vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.3 Space-Time Symmetries of Bloch States . . . . . . . . . . . . . . . . . . .
8.4 Space groups of solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.4.1 Symmorphic and Nonsymmorphic Groups . . . . . . . . . . .
8.5 Young Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
157
157
157
158
160
160
163
169
173
174
177
179
XIV
Contents
Product of Representations and Further Physical

Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.1 Irreducible Tensor Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2 Direct Product Representation . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2.1 Selection Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3 Reduction of the Direct Product Representation . . . . . . . . . . . .
9.4 Spin-Orbit Interaction and Double Groups . . . . . . . . . . . . . . . . .
9.5 Static and Dynamical Jahn-Teller Eect . . . . . . . . . . . . . . . . . . .
9.5.1 The Born-Oppenheimer (BO) Approximation . . . . . . . .
9.5.2 The Jahn-Teller Theorem . . . . . . . . . . . . . . . . . . . . . . . . .
9.6 Non-Adiabatic Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.6.1 Dynamical Jahn-Teller Eect . . . . . . . . . . . . . . . . . . . . . .
9.6.2 How the E Hamiltonian arises . . . . . . . . . . . . . . . . . .
9.6.3 Nuclear Wave Functions Cannot be Taken Real . . . . . .
9.7 Wigner-Eckart Theorem with Applications . . . . . . . . . . . . . . . .
9.8 The Symmetric Group and Many-Electron States . . . . . . . . . . .
9.9 Seniority Numbers in Atomic Physics . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
181
181
183
184
185
186
188
188
189
193
194
196
197
197
199
201
204
Part III More on Green Function Techniques

10 Equations of Motion and Further Developments . . . . . . . . . .
10.1 Equations of motion for the interacting propagator . . . . . . . . .
10.1.1 Equations of Motion and Ground-State Energy . . . . . . .
10.2 Time-Dependent Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.2.1 Auger Induced Ionic Desorption: Knotek-Feibelman
Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.3 Hierarchy of Greens Functions . . . . . . . . . . . . . . . . . . . . . . . . . . .
10.4 Composite Operator Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
207
207
208
208
11 Feynman Diagrams for Condensed Matter Physics . . . . . . .

11.1 Diagrams for the Vacuum Propagator . . . . . . . . . . . . . . . . . . . . .
11.1.1 Wicks Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.1.2 Goldstone Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.1.3 Diagram Rules for the Thermodynamic Potential . . . . .
11.2 Linked Cluster Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2.1 Valence Electron and Core Hole . . . . . . . . . . . . . . . . . . . .
11.2.2 H2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.2.3 The Linked Cluster Expansion and Greens Functions .
11.3 Diagrams for the Dressed Propagator . . . . . . . . . . . . . . . . . . . . .
11.3.1 Adiabatic Switching and Perturbation Theory . . . . . . .
11.3.2 Diagrams for the Propagator in frequency space . . . . . .
11.4 Dyson Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
217
217
219
221
224
225
227
229
230
231
232
235
238
210
212
213
216
Contents
XV
11.4.1 External Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Self-Energy from Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Skeleton Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Two-body Greens Function: the Bethe-Salpeter Equation . . .
Self-Energy and Two-Body Greens Function . . . . . . . . . . . . . . .
Functional Calculus and Diagrams . . . . . . . . . . . . . . . . . . . . . . . .
11.9.1 The Self-Energy as a Functional . . . . . . . . . . . . . . . . . . . .
11.9.2 Polarization Bubble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.9.3 The Vertex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11.10 Hedins Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
241
242
245
245
247
248
249
251
253
254
Many-Body Eects and Further Theory . . . . . . . . . . . . . . . . .

12.1 High Density Electron Gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.1.1 More Physical Insight about the RPA . . . . . . . . . . . . . . .
12.2 Low Density Electron gas: Ladder Approximation . . . . . . . . . .
12.3 Ladder Approximations in Electron Spectroscopies . . . . . . . . .
12.3.1 XPS and Auger Spectra from Metals . . . . . . . . . . . . . . .
12.3.2 The U<0 Phenomenon . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.3.3 Correlation in Early Transition Metals . . . . . . . . . . . . . .
12.4 Conserving Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.4.1 Continuity Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.4.2 The Functional . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.4.3 Gauge Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.4.4 Ground-State Energy and Grand Potential . . . . . . . . . .
12.4.5 Luttinger-Ward and ABL Variational Principles . . . . . .
12.5 Generalized Ward Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.6 Connection of Diagrams to D F T . . . . . . . . . . . . . . . . . . . . . . . .
12.6.1 Highlights on Density Functional Theory . . . . . . . . . . . .
12.6.2 Sham - Schl
uter Equation . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
257
257
259
263
264
264
267
270
272
272
274
275
276
277
278
279
279
282
283
13 Non-Equilibrium Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.1 Time-Dependent Probes and Nonlinear Response . . . . . . . . . . .
13.2 Kadano-Baym and Keldysh Methods . . . . . . . . . . . . . . . . . . . .
13.3 Complex-Time Integrals by Langreths Technique . . . . . . . . . . .
13.3.1 Finite temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.4 Keldysh-Dyson Equation on the Contour . . . . . . . . . . . . . . . . . .
13.5 Evolution on Keldysh Contour . . . . . . . . . . . . . . . . . . . . . . . . . . .
13.5.1 Contour Evolution of Bosons . . . . . . . . . . . . . . . . . . . . . .
13.6 Selected Applications of the Keldysh Formalism . . . . . . . . . . . .
13.6.1 Atom-Surface Scattering . . . . . . . . . . . . . . . . . . . . . . . . . .
13.6.2 Quantum Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
285
285
286
288
290
291
294
298
298
299
304
309
11.5
11.6
11.7
11.8
11.9
12
XVI
Contents
Part IV Non-Perturbative Approaches and Applications

14 Some Recursion Techniques with Applications . . . . . . . . . . . .
14.1 Lanczos-Haydock Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.1.1 Local Greens Function for a Chain . . . . . . . . . . . . . . . . .
14.1.2 Greens Function of any System . . . . . . . . . . . . . . . . . . . .
14.1.3 Terminator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.1.4 Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.2 Spin-Disentangled Diagonalization . . . . . . . . . . . . . . . . . . . . . . . .
14.3 Method of Excitation Amplitudes . . . . . . . . . . . . . . . . . . . . . . . .
14.4 Feenberg Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.4.1 Solving Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . .
14.4.2 Homogeneous systems . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
313
313
313
315
316
322
323
326
329
332
333
333
15 Aspects of Nonlinear Optics and Many-Photon Eects . . .

15.1 Diusion of Radiation in Dipole Approximation . . . . . . . . . . . .
15.1.1 Second-Order Processes . . . . . . . . . . . . . . . . . . . . . . . . . . .
15.2 Sum Frequency and Second-Harmonic Generation (SHG) . . . .
15.3 Diusion of Coherent Light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15.3.1 Eective mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15.3.2 Dynamical Stark Eect . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
335
335
336
342
343
343
346
351
Part V Selected Exact Results in Many-Body Problems

16 Quantum Phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16.0.3 Gauge Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . .
16.0.4 Spinor Rotations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16.0.5 Galilean Transformations . . . . . . . . . . . . . . . . . . . . . . . . .
16.1 Topologic phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16.1.1 Parametric Hamiltonians and Berry Phase . . . . . . . . . . .
16.1.2 Polarization of Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
355
355
356
356
357
358
362
17 Pairing from repulsive interactions . . . . . . . . . . . . . . . . . . . . . . .

17.0.3 W = 0 Pairing in Cu-O Clusters . . . . . . . . . . . . . . . . . . .
17.0.4 Pairing Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17.0.5 The W = 0 theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17.0.6 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17.1 Superconducting ux quantization and Josephson eect . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
367
369
373
373
375
378
381
Contents
XVII
18 Algebraic Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18.1 Lieb Theorems on the Half Filled Hubbard Model . . . . . . . . . .
18.2 Bethe Ansatz for the Heisenberg Chain . . . . . . . . . . . . . . . . . . .
18.3 Bethe Ansatz for Interacting Fermions 1 Dimension . . . . . . . . .
18.3.1 -Function Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18.3.2 The Hubbard Model in 1d . . . . . . . . . . . . . . . . . . . . . . . . .
18.3.3 The Periodic Boundary Conditions . . . . . . . . . . . . . . . . .
18.3.4 Spin Chain Analogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
383
383
387
392
392
396
401
403
406
Part VI Appendices
19 Appendix 1: Zero-point Energy in a Pillbox . . . . . . . . . . . . . . 409
20 Appendix II-Character Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
21 Proof of the Wigner-Eckart Theorem . . . . . . . . . . . . . . . . . . . . . 413
Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
Part I
I-Introductory Many-Body Physics
1 Basic Many-Body Quantum Mechanics
1.1 Slater Determinants and Matrix Elements

The solutions of eigenvalue equations like the time-independent one-electron
Schr
odinger equation hwi = i wi form a a complete set of spin-orbitals
{wi i (x)i } , where i (x) are normalized space orbitals and i = or
. The set can be taken orthogonal and ordered in ascending energy or in
any other arbitrary way. Any one-electron state can be expanded as a linear
combination of the wi . Moreover, we can think of a state for N electrons obtained as follows. Choose in any way N spin-orbitals out of the set {wi }, keep
them in the original order but call them v1 , v2 . . . vN ; now let |v1 , v2 . . . vN |
be the state with one electron in each. Imagine labeling1 the indistinguishable electrons with numbers 1, 2, N. In this many-body state one has an
amplitude (1, 2, . . . N ) ((x1 , 1 ), (x2 , 2 ), . . . (xN , N )) of having electron i in the one-particle state (xi , i ). How to calculate ? A product like

v1 v2 . . . vN = N
k vk is in conict with the Pauli principle because it fails to
be antisymmetric in the exchange of two particles. However, the remedy is
easy, because anti-symmetrized products are a basis for the antisymmetric
states. To this end, let P : {1, 2, . . . N } {P1 , P2 , . . . PN } be one of the N !
permutations of N objects. If N = 3, the set of 6 permutations comprises the
rotations {(1, 2, 3), (2, 3, 1), (3, 1, 2)} and {(2, 1, 3), (3, 2, 1), (1, 3, 2)}.
Anticipating some Group Theory
The last three are just transpositions, that is, they are obtained from the fundamental permutation (1,2,3) by one exchange. One can multiply two permutations Q and P; the product QP is the permutation obtained by performing
P and then Q and the result is:
{Q1 , Q2 , . . . QN }{P1 , P2 , . . . PN } = {QP1 , QP2 , . . . QPN }.
(1.1)
For instance2 , (3,2,1)(2,1,3)=(2,3,1) . All permutations can be obtained from

transpositions by multiplication. The inverse of a permutation is the one that
1
The electrons are identical, but this does not prevent us from labeling them;
rather it imposes that the wave function changes sign for each exchange of labels.
2
in words,P sends 1 2 and then Q sends 2 2, and in the same way 2
1 3 and 3 3 1.
upon multiplication restores the standard (ascending) order (1, 2, . . . N ). Any

permutation can be shown to be a product of transpositions, usually in more
than one way: for example, (2,3,1) is obtained from the standard order by
exchanging 1 and 2 and then 1 and 3 but also exchanging 1 and 3 and then
2 and 3. P has a parity or signature ()P dened such that an exchange is
odd, two are even, and so on, so in the above example (1,2,3), (2,1,3)and
(3,1,2) are even and the others odd. In Chapter 7 we shall see that such
simple observations have far reaching consequences.
Determinants
The antisymmetrizer operator
A=
1
()P P
N!
(1.2)
converts any product into a normalized Slater determinant, so we may write

a physically acceptable solution as
(1, 2, . . . N ) = A
N

k
1
vk =
()P vQ1 (1)vQ2 (2) . . . vQN (N ),
N! Q
(1.3)
or, equivalently,
v1 (1) v2 (1)
1
v1 (2) v2 (2)
(1, 2, . . . N ) =
...
...
N!
v1 (N ) v2 (N )
. . . vN (1)
. . . vN (2)
.
... ...
. . . vN (N )
(1.4)
Note that the transposed matrix is equivalent since the determinant is the
same. Exchanging two rows, that is, two electrons, one gets a - sign. A permutation of the electrons is equivalent to the inverse permutation of the
spinorbitals, and P (1, 2, . . . N ) = (P1 , P2 , . . . PN ) = ()P (1, 2, . . . N ).
The set of all determinants is complete provided an arbitrary order is
xed for the one-electron states (otherwise the set is overcomplete).
one electron and then consider the same
Suppose we solve hwi = i wi for
N
problem with N electrons and H = i h(i). This problem is no harder than
for a single electron, and the N body Schr
odinger equation is solved by (1.4)
with energy eigenvalue 1 + 2 + . . . + N .
1.1.1 Many-electron Matrix Elements
The matrix element |F | of operators F (1, 2, . . . N ) between determinantal states, when we expand the determinants, means a sum of (N !)2 terms.
This grows disastrously with N ; however there are simple rules to calculate
1.1 Slater Determinants and Matrix Elements
such matrix elements. Let (1, . . . N ) be a determinant made by N spinorbitals u1 , u2 . . . uN ; taken out of the orthogonal set {wi } (they may be same
as in , in which case we are dealing with expectation values). One can
readily observe these rules by working out a 2 2 example, while the proof
requires a trick which is explained in Sect. 1.1.2. The simplest case is f = 1,
and the rule is: the overlap between determinants is the determinant of the
matrix with elements the one-electron overlaps ui |vj :
| = Det [{ui |vj }] .
(1.5)
This useful result holds even if u and v spinorbitals are taken from dierent
sets w and w . The overlap of a determinant with itself is indeed 1, as it
should, which veries the normalization of determinants. The one-electron
matrix elements also imply a spin scalar product.
N
For one-body operators F (1, 2, . . . N ) = i f (i), where f acts on one
electron, the rule is simple: determinants gives the same results as simple
product wave functions, and antisymmetry has no consequences. The expectation values are given by
|F | =
N
ui |f |ui .
(1.6)
For example, if we pick f (i) = (x xi ), F = (x) is the number density

2
and one nds (x) = N
i |ui (x)| . O diagonal elements vanish if and
dier by more than 1 spinorbital; if they dier only by one spinorbital,vk in
and uk in ,
(1.7)
vk
= uk |F | = uk |f |vk .
pairs
N
1
Two-body operators of the form F = i=j fij = 2 i,j,i=j fij , like the
Coulomb interaction, have vanishing matrix elements when the two determinants dier by more than 2 spin-orbitals. For the rest, the best way to recall
the result is by the interaction vertices in Figure 1.1.1 below (embryos of the
Feynman diagrams that we introduce later). One must just note carefully
which lines enter at 1 and (left and right ) and which are outgoing. The order
of labels is: the one entering at 1, the one entering at 2, the one outgoing at
1. the one outgoing at 2.
If vi and vj in replace ui and uj in , ui
= vi , uj
= vj , then
|F | = vi (1)vj (2)|f (1, 2)|ui (1)uj (2) vi (1)vj (2)|f (1, 2)|ui (2)uj (1)
(1.8)
where the second, exchange term comes from the antisymmetry. These correspond to vertices a) and b) below, respectively. If f does not depend on spin
the exchange term vanishes for opposite spins.
The two-body operator matrix element describes the collision of a couple
of electron, while all the others are spectators. If and are the same, except

ui
ui
uj
vj
vi
uj
vj
vi
a)
b)
uk u uk
i
vk
c)
uj
ui
ui
d)
vk
e)
ui
uj
f)
Fig. 1.1. Interaction vertices
for ui
= vi , we get the vertex c) and its exchange companion d), representing
the two terms in the expression
|F | =
N
[vk (1)ui (2)|f (1, 2)|uk (1)ui (2)
i=k
ui (1)vk (2)|f (1, 2)|uk (1)ui (2) ].
(1.9)
This is similar to (1.8), but there is a summation over all the spinorbitals
present in both determinants, which act as background particles while the
i-th electrons jumps from v(i) to u(i). Finally, for the expectation value we
get the vertices e) and f) , that is,
|F | =
N
[ui (1)uj (2)|f (1, 2)|ui (1)uj (2)
j=i
uj (1)ui (2)|f (1, 2)|ui (1)uj (2) ].
(1.10)
1.1.2 Derivation of the Rules

A direct expansion of
|F | =
1
()P+Q
N!
P,Q
uP1 (1)uP2 (2) . . . uPN (N )|f |vQ1 (1)vQ2 (2) . . . vQN (N )
(1.11)
involves N !2 terms and is formidable unless N is small. However, the proof

of the above rules is easily obtained by a trick, that I exemplify in the case
of one-body operators. The matrix element
uP1 (1)uP2 (2) . . . uPN (N )|f (1, , N )|vQ1 (1)vQ2 (2) . . . vQN (N )
is a multiple integral; we permute the names of dummy variables and get
uP1 (P1 )uP2 (P2 ). . . uPN (PN )|f |vQ1 (P1 )vQ2 (P2 ) . . . vQN (PN ) ; f does not change
since it must
depend onparticles in a symmetric way. The bra is independent
of P, since i uPi | = i ui |, and we obtain
1.2 Second Quantization
u1 (1)u2 (2) . . . uN (N )|f |vQ1 (P1 )vQ2 (P2 ) . . . vQN (PN ) .
Now the Q summation yields back the determinant with a permutation
P of the electrons, that is, ()P ; the ()P factor cancels the one already
present in (1.11). Hence,
v1 (1) v2 (1) . . . vN (1)

v1 (2) v2 (2) . . . vN (2)
. (1.12)
|F | = u1 (1)u2 (2) . . . uN (N )|f |
...
... ... ...
v1 (N ) v2 (N ) . . . vN (N )
More explicitly,

1
|f | = u1 (1)u2 (2) . . . uN (N )|f |
()Q vQ1 (1)vQ2 (2) . . . vQN (N ).
N!
Q
(1.13)

1.2.1 Bosons
Since the time-independent Schr
odinger equation for the Harmonic Oscillator,
2 d2
h
m 2 x2
= E
(1.14)
+
2m dx2
2

h
has a characteristic length x0 = m

, one introduces the annihilation operator

x
ix0 p
1
+
a=
.
(1.15)
h
2 x0
this is equivalent to
a + a
x
= ,
x0
2
ix0 p
a a
= ,
h
(1.16)
with the commutation relation

[a, a ] = 1;
the Hamiltonian can be rewritten

1
H = a a +
.
h
2
(1.17)
(1.18)
If is a solution of (1.14) with eigenvalue E, a must be solution with

eigenvalue E
h. The conclusions are: 1) a0 = 0 if 0 is the ground state
and 2) a is a creation operator that in fact creates excitations like a destroys

them. One then learns that i) this represents a boson eld with one degree of
freedom (the x) ii) when dealing with real physical elds one never observes
the oscillators but only the excitations, e.g. photons for the electromagnetic
eld. The noninteracting bosons in a eld mode can be created in any number,
and each adds the same energy to the eld. The oscillator does not exist at
all, but the unique property of the oscillator potential which has innitely
many states with uniform spacing h
makes it a perfect representation for
the eld.
Example: Coupled Boson Representation of Angular Momentum
Schwinger [8] has shown how one can build a representation of the angular
momentum operators including components Ji , shift J and more exotic K
operators that conserve m but raise or lower j. All this was obtained using
creation and annihilation operators of a couple of modes, and everything
and
comes from a simple observation. For instance let j = 32 in units of h
2
consider the following scheme: For any j, we can write j = n1 +n
in
terms
of
2
2
2
n2 j = n1 +n
jz = n1 n
2
2
3
3/2
3/2
2
3/2
1/2
1
3/2
1/2
0
3/2
3/2
n1
0
1
2
3
2 integers 0 in several ways, and each entry corresponds to a choice of jz ;

n2 increases from 0 in 2j + 1 steps. So, a J+ should add 1 to n1 and remove
1 from n2 , while K should add 1 to both. Two harmonic oscillators can
provide the occupation number operators to represent that. So,
j=
n
1 + n
2
;
2
jz =
2
n
1 n
.
2
(1.19)
To extend this idea, one can observe that introducing a spinor operator

a1
=
(1.20)
a2
we may write
j=
This extends naturally to

h
;
2
jz =

h
z .
2
j =
(1.21)
(1.22)
which implies trivially

jx =

h
(a a2 + a2 a1 );
2 1
and hence
j+ = h
a1 a2 ;
jy =
i
h
(a1 a2 + a2 a1 )
2
j = h
a2 a1 .
(1.23)
(1.24)
Indeed, it is a simple matter to verify that

[jx , jy ] = i
hjz ,
(1.25)
2 j(j + 1).
j2 = h
(1.26)
One can change j by

K+ = h
a1 a2 ;
K = h
a 1 a2 .
(1.27)
Using j and m = jz in (1.19) one nds n1 and n2 , hence

(a )j+m (a2 )jm
|jm = 1
|0 .
(j + m)!(j m)!
(1.28)
This is an alternative way to derive results like Clebsh-Gordan coecients

and the like.
1.2.2 Field Quantization and Casimir Eect
The electromagnetic eld uctuations in vacuo have a macroscopic consequence named Casimir eect. This is of interest for fundamental physics but
also for potential applications.
Let two square mirrors of side L be put in front of each other at a distance
s. Roughly speaking, this causes boundary conditions E = 0 of vanishing
parallel electric eld component on both surfaces, at frequencies below the
plasma frequency p of the metal. The eld between the mirrors is constrained and has a reduced zero point energy; thus, the radiation pressure
is lower than in vacuum and a macroscopic attraction between the mirrors
appears. The eect was discovered by Casimir [6] theoretically and then veried experimentally [7],[200]. It is important at m distances, so it is longer
ranged than Van der Waals forces, which are mainly due to the uctuating
instantaneous dipoles on non-polar systems.
To understand this in more detail consider a metallic pillbox,with reecting walls, a square basis of side L and hight s. How much is the zero-point
b c
energy U (s) in the pillbox? Each wave-vector k = ( a
s , L , L ), (with integer
a, b and c) contributes h
ck and the sum diverges of course, so we impose an
exponential ultraviolet cuto , removing the short wavelengths such that
k 1. They are not involved anyhow because at ultraviolet frequencies the
mirrors are transparent.
U (s) = h
c

a 2
abc
b
L
2
+
c 2
L
2
a
b 2
c 2
e ( s ) +( L ) +( L ) .
(1.29)
We can calculate this exactly for large enough L and the result diverges as
0. One nds (see Appendix 1)

2
hc 2 L2
1
d
1
(1.30)
U (s) =
2
d
es 1
Using the expansion
Bn y n
1
1 y2
1 y4
y
=
1
y
+
+
...
=
ey 1
2
6 2!
30 4!
n!
n
(the Bn are called Bernoulli numbers), one obtains:

2
s
d
1
2 1
hc 2 L2
1
+
U (s) =
2
d
2
2 12s 30 4!s3
(1.31)
(1.32)
The rst two terms lead to the aforementioned divergence: should we try to
remove all the radiation from the cavity, including the high frequency modes,
that would cost us innite energy. However, the divergence disappears if we
ask: what changes if we shift one side of the cavity by 1 cm? To better answer
this question, suppose a cavity of length R is divided in two equal halves by a
mirror: evidently the energy of the vacuum is the diverging quantity 2U (R/2).
If instead the mirror is at distance s from one end and R s from the other,
the vacuum energy must be U (s) + U (R s), which also diverges. The nite
dierence

R
}
(1.33)
E(s) = lim {U (s) + U (R s) 2U
R
2
has the physical meaning of an energy that must be supplied to the system
in order to shift the mirror to the middle of the cavity. If the cavity is large,
this can be identied with the interaction energy at distance s. Eventually
one can let 0. The zero point energy decrease per unit surface is thus
hc
2
,
(1.34)
720 s3
and since the radiation pressure is proportional to the energy density one
observes an attractive force
E =
hc
2
.
240 s4
Measuring distances s in m, one nds
F =
(1.35)
0.013
dyne/cm2 .
(1.36)
s4
This force and its dependence on material and surface properties is actively
investigated and could be used to operate nano-machines.
F =
1.2.3 Fermions
The second quantization formalism for Fermions was invented in order to
deal with phenomena like neutron decay n p + e + or pair creation in
particle physics, but to create an electron-positron pair one needs about a
million eV. In condensed matter physics the typical energy scale is much
less than that, yet many important phenomena are naturally described in
terms of the creation (or annihilation) of fermion quasi-particles. Electronhole pairs can be created very much like electron-positron ones. In scattering
processes, when all the particles are conserved, one can proceed with Slater
determinants in rst quantization; however, second quantization formalism
is much easier to work with.
The change from bosons to fermions replaces permanents with determinants. In place of a N-times excited oscillator representing N bosons in a given
mode, we now consider N -fermion determinants |u1 u2 . . . uN |, where the spinorbitals are chosen from a complete orthonormal set {wi }. The index i can
be discrete or continuous but implies a xed ordering of the complete set. In
this way, one can convene e.g. that in |u1 u2 . . . uN | the indices 1 N are in
increasing order thereby avoiding multiple counting of the same state. The
zero-particles or vacuum state |vac replaces the oscillator ground state. For
the determinants, it is generally preferable
notation like
to use a compact

u
(1)
u
(2)
m
m
|um un | rather than the explicit 12 Det
which contains the
un (1) um (2)
3
same information. Consider the following correspondence between determinants and states of the Hilbert space with various numbers of electrons:
First Quantization
No electrons state(vacuum)
1 body state uk
2 body determinant |um un |
3 body determinant |um un up |
...
Second Quantization
|vac
ck |vac
cm cn |vac

cm cn cp |vac
...
(1.37)
Up to now the second-quantization side looks very similar to the compact

notation for determinants: the new idea is using the operator cm , clearly
deserving the name of electron creation operator in spin-orbital m, in order
to express all other operators. The left column introduces an occupation
number representation of the basis of the Hilbert space; second quantization
builds such a representation by creation operators cm . Adding a particle to
any state cannot lead to the vacuum state,
vac|cm = 0.
(1.38)
3
mathematically, it is an isomorphism; it can be thought of as a change in
notation.
10
Moreover, since a determinant is odd when columns are exchanged, we want

an anticommutation rule
[cm , cn ]+ cm cn + cn cm = 0.
(1.39)
It follows that the square of a creation operator vanishes. By denition,

cm {cn cr |vac } = cm cn cr |vac
(1.40)
The notation suggests that cm is the Hermitean conjugate of cm ; this is called

annihilation operator. Taking the conjugate of (1.40)
{vac|cr cn }cm = vac|cr cn cm
(1.41)
and taking the scalar product with cm cn cr |vac , we deduce that

{vac|cr cn }cm | cm cn cr |vac = 1.
(1.42)
If now we consider cm as acting on the right, we see that it is changing the

3-body state cm cn cr |vac into the 2-body one cn cr |vac . Thus, annihilation
operator is a well deserved name: an annihilation operator cm for a fermion
in the spin-orbital state um removes the leftmost state in the determinant
leaving a N 1 state determinant:
c1 |u1 u2 . . . uN | = |u2 . . . uN |
(1.43)
cm |vac = 0.
(1.44)
and
It obeys the conjugate of the anticommutation rules (1.39), namely,
Next consider
[cm , cn ]+ cm cn + cn cm = 0, c2m = 0.
(1.45)
cn cm cn cr |vac , n, m, r all dierent.
(1.46)
Since the creation operators anticommute, we get

cn cn cm cr |vac = cm cr |vac
since the m state is created at the leftmost place in the determinant but is
annihilated at once. This shows that creation and annihilation operators also
anticommute,
(1.47)
[cn , cm ]+ = 0, n
= m.
As long as the indices are dierent c and c all anticommute, so the pairs
cn cm ,cn cm ,cn cm and cn cm can be carried through any product of creation or
annihilation operators where the indices n,m do not occur.
Next we note that cp |vac |p is a one-body wave function; cp cp |vac =
|vac and cp cp cp |vac = cp |vac . Now one can check that
np cp cp
11
(1.48)
is the occupation number operator, having eigenvalue 1 on any determinant

where p is occupied and 0 if p is empty. On the other hand, cp cp having
eigenvalue 0 on any determinant where p is occupied and 1 if p is empty. thus
in any case cp cp + cp cp = 1. Since this holds on all the complete set it is an
operator identity and we may complete the rules with
[cp , cq ]+ = pq .
(1.49)
Note that np = np and n2p = np .

1.2.4 Basis Change in Second Quantization and Field Operators
We can readily go from basis set {an } to a new set {bn }; since

|bn >=
|ak >< ak |bn >
(1.50)
the rule is
bn =

k
ak < ak |bn >,
bn =
ak < bn |ak > .
(1.51)
It is often useful to go from any set {un } to the coordinate representation

introducing the creation and annihilation eld operators

(x) = n cn un (x)
(1.52)
(x) = n cn un (x),
(here un denotes the conjugate spinor). Note that cp |vac is a one-electron
state and corresponds to the rst-quantized spinor up (x); (y)|vac is a
one-electron state
corresponds to the rst-quantized spinor with spatial
and
u
(y)u
wave function
n (x) = (x y); thus it is a perfectly localized
n n
electron. The rules are readily seen to be
[ (x), (y)]+ = 0, [ (x), (y)]+ = 0,
(1.53)
and
[ (y), (x)]+ =

p,q
[cp , cq ]+ up (x)uq (y) =
up (x)up (y) = (x y)
p,q
(1.54)
where the also imposes the same spin for both spinors.
A one-body operator V (x) in second-quantized form becomes

V = dx (x)V (x) (x) =
Vp,q cp cq .
p,q
(1.55)
12
This gives the correct matrix elements between determinantal states, as one
can verify.
The above expressions imply spin sum along with the space integrals,
although this was not shown explicitly; let me write the spin components, for
one-body operators:

V =
dx V, (x)
(1.56)
,
For the spin

setting h
= 1, and using the Pauli matrices, Sz =
operators,

0
1
1
+
and the rule (1.55) one nds
2 z , S =
00

1
Sz =
dx (x) (x) (x) (x) , S + = dx (x) (x). (1.57)
2
Often we shall use a discrete basis and notation and we shall write

S+ =
ck ck
(1.58)
which is obtained from (1.57) by taking a Fourier transform in discrete notation. A two-body operator U (x, y) becomes

U = dx dy (x) (y)U (x, y) (y) (x) =

Uijkl ci cj cl ck
(1.59)
ijkl
(please note the order of indices carefully). The Hamiltonian for N interacting
electrons in an external potential (x) is the true many-body Hamiltonian
in the non-relativistic limit that we shall often regard as the full many-body
problem for which approximations must be sought. It may be written
H (r1 , r2 , . . . , rN ) = H0 (r1 , r2 , . . . , rN ) + U (r1 , r2 , . . . , rN )
(1.60)
where H0 is the free part

H0 = T + Vext =

1
h0 (i)
2i + V (ri ) =
2
i
i
(1.61)
with T the kinetic energy and Vext the external potential energy while
U=
1
uC (ri ri )
2
(1.62)
i=j
is the Coulomb interaction. This Hamiltonian may be written in secondquantized form
13
H = H0 + U,

H0 =
dr (r)h0 (r),

1
dxdy (x) (y)uC (x y), (y) (x). (1.63)
U =
2
,,,
Often the spin indices are understood as implicit in the integrations. It should
be kept in mind that relativistic corrections are needed in most problems with
light elements and the relativistic formulation is needed when heavy elements
are involved. Fortunately, the ideas that we shall develop lend themselves to
a direct generalization to Diracs framework.
1.2.5 Hubbard Model for the Hydrogen Molecule
The Hubbard Model is a lattice of atoms or sites that can host one electron
per spin; there is a hopping term between nearest neighbors like in a tightbinding model and a repulsion U between two electrons on the same atom.
The Hubbard Hamiltonian

cj ci + U
ni ni ,
(1.64)
H =K +W =t
i,j,
where K stands for the kinetic energy while W accounts for the on-site repulsive interaction. The summation on i, j runs over sites i and j which
are nearest neighbors in a cubic lattice. This is often called trivial Hubbard
Model to distinguish it from its extensions, involving degenerate orbitals and
o-site interactions, that have been studied for many purposes.4
To model H2 in the same spirit we represent the 1s orbitals of both atoms
= T + W
with
by two sites a and b and H

T = th
ca cb + cb ca
(1.65)
the kinetic energy, with th > 0 the hopping integral;

= U (
W
na n
a + n
b n
b ) .
4
(1.66)
Some people blame the Hubbard Model and its extensions as too idealized to
be realistic. Indeed nobody would use them to rene well-understood properties of
Silicon. However, there are lots of problems involving strong correlations and e.g.
transport, spectroscopies, time-dependent perturbations, which are far too hard for
an ab-initio description. Hubbard-like models are primarily conceptual tools aimed
at a semi-quantitative understanding. We shall see particularly in Chapters 4, 5 and
10 that often they allow to deal with highly excited states of strongly interacting
system very successfully. The Bosonic Hubbard Model is also important, e.g. in the
rapidly developing subject of Cold Bosonic Atoms in Optical Lattices (see Ref. [15].
14
We wish to solve with two electrons of opposite spin (the ms = 0 sector) so

we take

=
N
(
na + n
b ) = 2.
This is conserved. If U = 0, one solves the single-electron problem, and nds

the orbitals
|a |b
(1.67)
=
2
with energy eigenvalues
= th
(1.68)
and the ground state = has energy E = 2th . In the interacting

case, we choose a basis
|v1 >= |a a >, |v2 >= |a b >,
|v3 >= |b a >, |v4 >= |b b > .
U
th
Fig. 1.2. Singlet eigenvalues of the Hydrogen molecule model versus
U
.
th
There is a single state in the ms = 1 sector, so out of the 4 states in the

ms = 0 sector we expect one triplet and 3 singlets. We form the matrices
W = U Diag (1, 0, 0, 1) and
0110
1 0 0 1
T = th
(1.69)
1 0 0 1.
0110
One
the eigenvalues:
E = 0 for the triplet, and E0 = U, E =
nds

1
2
2
for the singlets, with E the ground state (remark16th + U
2 U
ably) for any U > 0. Magnetism never obtains in this model.
1.3 Schrieer-Wol Canonical Transformation
15
1.3 Schrieer-Wol Canonical Transformation

One often meets problems with Hamiltonians
H = H0 + V
(1.70)
such that the interaction V takes the system to an enlarged Hilbert space,
involving extra degrees of freedom not in action in the simple problem described by H0 . Let A denote the restricted space and B the enlargement.
Typically,

HA 0
H0 =
,
(1.71)
0 HB

and
V =
0 v
v 0

(1.72)
is the mixing term. A standard way to solve such problems, that we shall
meet several times in this book, is by a canonical transformation
= U HU 1
H H
is block-diagonal:
where U is designed such that H

A 0
H
H=
B
0 H
(1.73)
(1.74)
The transformation must be unitary in order to preserve the norm of states,

to this end we want U 1 = U ; this is granted if U = eS with S = S .
Thus, expanding the exponentials,
= eS HeS = H + [S, H] + 1 [S, [S, H]] +
H
2
(1.75)
Now we insert (1.70) with S = S1 +2 S2 + and separate orders. Including

up to second-order,
[S, H] = [S1 , H0 ] + 2 ([S1 , V ] + [S2 , H0 ] ),
(1.76)
[S, [S, H] ] = 2 [S1 , [S1 , H0 ] .
(1.77)
At order , we want to have nothing and we require that S1 be such that

V + [S1 , H0 ] = 0,
(1.78)
that is,

0 v
v 0

HA 0
0 s
+[
,
] = 0.
0 HB
s 0
(1.79)
16
where we tried the solution

S1 =
0 s
s 0

.
(1.80)
We immediately obtain two conditions, v = sHA + HB s and v =

HA s + s HB . Picking H0 eigenstates |m in the A subspace and | in
(A)
(B)
the B subspace, with eigenvalues Em and E , we obtain
sn =
vn
(B)
E
(A)
En
, (s )m =
(v )m
(B)
E
(A)
Em
(1.81)
The second-order contribution to (1.75), using (1.76),(1.77) and (1.79), is

] + [S2 , H0 ]. We may set S2 = 0 since

(s v + v s)
0
[S1 , V ] =
(1.82)
0
s v + v s
2
2 [S1 , V
already gives a Hermitean, block-diagonal result. Thus,to second order,

A = HA + Hint
H
(1.83)
where
2
[S1 , V ].
(1.84)
2
The eect of V can be obtained by working within the A subspace with
a renormalized Hamiltonian (see (10.48),( 1.82)) with elements

vm
vn
vn
1
vm
(Hint )mn =
+
.
(1.85)
(A)
(B)
(A)
(B)
2
En E
B Em E
Hint =
If the energy separation of A and B is large, the dependence of the energy

denominators on m, n is negligible, and we may write
Hint =
v | |v
(B)
E (A)
(1.86)
the denominator is a positive excitation energy.
1.4 Variational Principle

The energy of a quantum system is a quadratic functional of the wave function
. Consider a small variation + where ia an arbitrary function of
the same variables on which depends, while 0 is a complex parameter,
E stationary {E = 0, arbitrary};
(1.87)
1.4 Variational Principle
17
subject to the condition that the norm is conserved, namely,

(E N ) = 0.
(1.88)
The Lagrange multiplier is xed by the condition N =< ()|() >= 1.

Applying Lagranges method, one nds that the following statements are
equivalent:
{H = E, < | >= 1} {(EN ) = 0, = E} {(E) = 0, N = 1}.
(1.89)
This is an exact refurmulation of Quantum Mechanics
Example
Given the Hamiltonian
3 1
1 0
H =
1 0
1 0
1 0
11
00
00
00
00
1
0
0
0
nd variationally
the eigenfunctions of the form

2
2
=
. Normalizzation requires N = | = + 4 = 1 while E =

|H| = 832 , thus we must look for the extrema of f (, ) = EN.

One nds
f
= 0 = 4 = (3 + )
= 0 = =
The compatibility condition (3 + )

= 4yields = 4, = 1. For = 4
4
1
da = 4one nds 4 = 120

1 which is the ground state with
1
1

1
1
eigenvalue = 4. For = 1 da = one nds 1 = 15

1 which is the
1
1
exact excited state witheigenvalue = 1.
18
1.5 Variational Approximations

One can choose a trial function (x, {1 , 2 , n }) depending on parameters
{1 , 2 , n } and look for the extremum. If is not already normalized,
the normalization condition can be enforced by Lagranges method. If the
exact ground state belongs to the class of functions, it corresponds to the
minimum energy, otherwise the minimum always overestimates the ground
state energy. Some variational approximations, like the Hartree-Fock scheme
and the Bardeen- Cooper - Schrieer theory of superconductivity, have been
highly successful.
The excited states also correspond to extrema of the functional, however
there are severe limitations to the method.
The trouble is that the true eigenstates are orthogonal, but this
cannot be granted in general in a limited class of functions.
We need the orthogonality. For example we cannot give any meaning to
an excited state which fails to be orthogonal to the ground state. However,
the lowest state of any symmetry can always be found variationally, since it
is automatically orthogonal to the ground state.
A symmetry is an operator X, which is unitary (that is XX = 1), such
that [H, X] = 0. The eigenstates of an unitary operator X belong to dierent
eigenvalues are orthogonal. Indeed, if X1 = ei 1 aand X2 = ei 2 ,
(1 , 2 ) = (1 , X X2 ) = ei() (1 , 2 )
and with
= , this requires (1 , 2 ) = 0.
1.6 Non-degenerate Perturbation Theory

The standard perturbation series yields [25] the corrected eigenvalues
(0)
Em = Em
+ m|H |m +

|m|H |n |2
(0)
(0)
Em En
(1.90)
(0)
where Em are unperturbed eigenvalues and m|H |n are perturbation ma

trix elements; m excludes the terms with zero denominators. The perturbed
wave functions are:

m|H |m
(0) k|H |m
(0)
k
(1
)
n = n +
(0)
(0)
(0)
(0)
Em Ek
Em Ek
k

k|H |n n|H |m
+
+
(1.91)
(0)
(0)
(0)
(0)
n (Em Ek )(Em En )
(0)
where n are the unperturbed ones. A much more general form of perturbation theory will be developed starting from Chapter 11.
2 Adiabatic Switching and Time-Ordered

series
2.1 Time-dependent is Better: start from the Golden

Age
If we can solve a problem with a time independent Hamiltonian H0 , we surely
meet many interesting but hard problems with a Hamiltonian
H(t) = H0 + V (t)
(2.1)
where an extra term V (t) appears: sometimes the complication V (t) depends
on time, but in other cases it is static. Here, H(t) is in the Schrodinger picture,
the one that comes directly from classical physics with px x

, and so on,
and sometimes we shall write HS (t) and S for extra clarity. So, we have the
task of solving
i
h |S (t) = HS (t)|S (t)
(2.2)
t
which is notoriously dicult. We can solve formally by introducing the unitary time evolution operator US such that
|S (t) = US (t, t0 )|S (t0 ) .
(2.3)
The time t0 is arbitrary, and in time-independent problems one can choose

t0 = 0, but this is dull. In general, it is a much better idea taking HS (t)
which depends on time, with the condition that for t < t0 , V 0. For static
problems this appears lunatic, but it is a useful formal device. Dynamical and
static problems are best discussed with the adiabatic switching technique.
This is a cool revival of the old tale of the golden age, a happy era in the far
past. Assume that the interaction is added very slowly, starting from a time
t t0 , when the Hamiltonian was just H0 and everything was easy; if we let
t0 we may reasonably assume that the system does not warm up and
evolves adiabatically. Integrating (2.2) over time we nd

i t
US (t, t0 ) = 1
dt1 HS (t1 )US (t1 , t0 );
(2.4)
h t0
the advantage is that this may be solved by iteration:
20
2 Adiabatic Switching and Time-Ordered series
US (t, t0 ) = 1
i
h
dt1 HS (t1 ) + (
t0
i 2
)
h
t
t1
dt1 dt2 HS (t1 )HS (t2 ) + . . .

t0
t0
Iterating, we face the nested integral

tn1
t
t1
i
In (t) = ( )n
dt1
dt2 . . .
dtn H(t1 ) . . . H(tn )
h
t0
t0
t0
(2.5)
(2.6)
where t the domain is limited to t > t1 > . . . tn > t0 . If H does not depend on
n
n
n H (tt0 )
and the resulting exponential
time, one gets trivially In (t) = ( i
h
)
n!
series is immediately summed. This n! denominator proves useful and we can
bring it out in the time dependent case by a trick. Perform any permutation
of the time variables t1 . . . tn in In (t); in the new ordering the earlier times
will remain on the right of the later ones, while the value of the integral
remains unaltered. Thus one can sum all the n! identical replicas obtained
by permutation and divide by n!; accordingly, one denes the time ordering
operator P which puts earlier times on the right. For two terms
P [HS (t1 )HS (t2 )] = HS (t1 )HS (t2 )(t1 t2 ) + HS (t2 )HS (t1 )(t2 t1 );
with more operators,
P [HS (t1 )HS (t2 ) . . . HS (tn )] =

HS (tQ1 )HS (tQ2 ) . . . HS (tQn )
(tQn1 tQn ).
(2.7)
Under the action of P the operators can be permuted freely as if they commuted. Actually when dealing with electron operators one uses Wicks time
ordering operator T which is dened like P except that any exchange of
fermion operators which is needed to go from the given order to the standard
earlier to the right order brings a sign. Thus, if A ad B are fermion creation
or annihilator operators, T is such that
T [A(t)B(t )] = A(t)B(t )(t t ) B(t )A(t)(t t).
When acting on Hamiltonians where Fermi operators occur in pairs P and
T have the same eect, but T permits simplifying the denition of fermion
Greens functions. Summing over the permutations, that give identical contributions, and dividing by their number, the integral (2.6) can be rewritten
t
t
1 i
In (t) = ( )n
dt1 . . .
dtn T [H(t1 ) . . . H(tn )]
(2.8)
n!
h
t0
t0
and we may formally sum the series:

US (t, t0 ) = T exp
i
h
t
t0

d HS ( ) ;
(2.9)
here T exp is a conventional notation that means nothing but the exponential
series of time-ordered products.
2.2 Evolution in Complex Time
21

2.2.1 Heisenberg Picture
In most cases of interest, the wave function is a false target and W. Kohn
deserved a Nobel price for inventing the Density Functional Method (Section
12.6) that dispenses us from calculating it in important cases. Simply, is too
complicated to be computed or even approximated in really hard problems1 .
It depends from a huge number of variables and involves all the information
on a system, including 15-electron correlations that nobody cares about. Indeed, our task is another: doing experiments of some kind and interpreting
them never gives a complete information, so almost all the information that
the function contains is actually pointless. The interesting information is
expressed in the experiment-oriented Heisenberg correlation functions, that
we discuss starting from Chapter 10. In this Section we shall pretend that we
are still primarily interested in expanding the wave function, but the work is
actually oriented towards the Greens functions. In the Heisenberg picture,
A(t) = H |AH (t)|H
(2.10)
|H |S (t0 )
(2.11)
where by denition
is the t-independent snapshot of the golden-age , while any operator A,
including c and c , has an extra dynamic time dependence
AH (t) = US (t, t0 )AS (t)US (t, t0 ).
(2.12)
For a strictly time-independent H (no adiabatic switch) this simplies to read

A(t) = e
iHt
h
Ae
iHt
h
(2.13)
but in general, from (2.12) one gets:

dU (t, t0 )
dAH
= i
h S
AS US (t, t0 )
dt
dt
dUS (t, t0 )
dAS
+ US (t, t0 )ih
US (t, t0 );
h
+US (t, t0 )AS i
dt
dt
i
h
(2.14)
since
i
h
dU (t, t0 )
dUS (t, t0 )
= HS (t) US (t, t0 ), i
= US (t, t0 )HS (t)
h S
dt
dt
(2.15)
the result is
1
In 1 dimension, however, the Bethe Ansatz allows solving exactly some important models, see Chapter 18.
22
dAH
= US (t, t0 )HS (t) AS US (t, t0 )
dt

dAS
+US (t, t0 )AS HS (t) US (t, t0 ) + ih
.
dt H
i
h
(2.16)
It is natural to introduce the Heisenberg picture Hamiltonian and write:

dAS
dAH
= [AH (t) , HH (t)] + ih
i
h
.
(2.17)
dt
dt H
In the following, unless otherwise stated, we shall normally use the Heisenberg picture, and A(t) = H |AH (t)|H with H eigenstate of H0 (in
practice,the ground state in most cases). In terms of the Schrodinger picture,
this corresponds to evolving the state from t0 to time t, applying the operator
and then evolving back to t0 . We can merge the two US evolution operators
by introducing an oriented path C in complex time from t0 to t and back (see
Figure). This needs a generalized T such that (letting h
= 1)
AH (t) = US (t, t0 )AS (t)US (t, t0 ) =

t

t0
dt H(t )
AS (t) T exp i
dt H(t )
T exp i
t

= TC

exp i
t0
t0

dt H(t ) AS (t)
(2.18)
where C is an oriented path and TC is the time ordering operator on C. Note

that AS (t) is under the action of TC that places it appropriately.
Im z
Re(z)
t0
t0
t+
t
Re z
t
a)
b)
Fig. 2.1. a) A contour on the complex time z plane for obtaining A(t) from a
single Schr
odinger-picture evolution. Since along a closed path starting and ending
at t one collects US (t , t ) = 1, the path can be deformed freely as long as it starts
and ends at t0 and goes through t. b)The Keldysh contour; its main use will be
shown in Chapter (13)
23
We shall see in Chapter 10 that this is the most natural way to make
contact with experiment, and is also important for the connection with thermal physics that follows. The evolution operator satises the group propertyU (t, t1 )U (t1 , t2 ) = U (t, t2 ); hence the path can be deformed freely as
long as it starts and ends at t0 and goes through t. In time-dependent problems, the most common contour is the Keldysh one from t0 to t0 +
and back to t0 ; there are an ascending or positive branch and a descending or
negative branch, and a physical time can be taken on any of the two. I shall
write t+ and t the times taken on the ascending and descending branch
respectively; however A(t) = A(t+ ) = A(t ) .
2.2.2 Thermal Averages
The initial state of the experiment on a solid is never an eigenvector of the
Hamiltonian; it is described [118] by a Hermitean density matrix that we may
denote

wi |i i|,
(2.19)
=
i
where |i is a completeorthonormal set, wi 0 is the probability of nding

the system in |i with i wi = 1. The statistical average of any operator A is

where Tr
by denition A = i wi Aii and may be obtained as A = T rA,
denotes the trace (the sum of all the diagonal elements, which is independent
of the basis set.) In temperature-dependent problems, adopting the Grand
Canonical ensemble, is the Boltzmann distribution,
=
where =
1
KB T
eK
, K = H N,
Z
(2.20)
, N the number operator;

Z = T reK
(2.21)
is the partition function. As detailed e.g. in [117], this yields the maximum
entropy S = KB ln() with the constraints that T r = 1, particle number
and energy must be kept xed. Thus,
Finite T rule: A(t)

= T rA(t).
(2.22)

For an independent-electron system with Hamiltonian H0 = p p cp cp ,
any energy eigenstate is specied by the set of occupation numbers np of
the one-electron levels, and the trace sums over all possible choices of np .
Each term of the sum is a product of factors e(p) from lled states
and factors 1 from empty states. Let us pick a particular level k and let X(k)
denote the contribution to Z from all the congurations with nk = 0; then, the
contribution to Z from all the congurations with nk = 1 is e(k ) X(k),
24
since the other levels give the same contribution regardless the population
of level k. We can factor (1 + e(k ) ) from the trace. Therefore we may
conclude that

(1 + e(k ) ).
(2.23)
Z=
k
This is readily worked out for independent Fermions (Problem 2.1).)

The formal similarity between quantum averages and statistical ones
stems from the fact that eK = eiK(i) looks like a quantum propagation at imaginary time t = i. The system must be considered in thermal equilibrium at a (real) time t0 earlier than any time dependence of the
Hamiltonian. Then
Z = T r[eN e
t0 i
t0
dt H
] T r[eN T e
t0 i
t0
dt H
];
i
(2.24)
t0 i
dt H
here T has no eect but was inserted to emphasize that T e t0

is an evolution operator along the vertical track in Figure 2.2.2 a). In the
complex t plane one draws the so called imaginary-time axis; = it along the
axis corresponding to a 90 degrees rotation of the plane is the real variable
= Imt. The statistical average of a time-independent operator may be
written
t0 i
i
dt H0
1
t0
A(t0 )].
(2.25)
A = T r[eN T e
Z
If H is constant and no adiabatic switching is assumed, one can take t0 = 0.
However when using the adiabatic switching we take t0 as the golden age
dened above. The switching of the interaction is so slow that no heating
of the system is caused, so at modern times t we have the full Hamiltonian
H(t) with any further explicit time dependence that may be necessary. This
allows us to perform the thermal average with the simple, particle-number
conserving H0 , that is,
eK0
,
Z
K0 = H0 N, [H0 , N ] = 0.
A(t)
= T r0 AH (t), =
(2.26)
For evolving along the vertical track with a temperature-independent H one

can introduce the temperature Heisenberg representation
A( ) = e(HN ) Ae(HN ) ;
(2.27)
this is obtained from (2.13) by the usual substitution it and by H

H N, in order to introduce the grand-canonical ensemble. The statistical
average of a time-dependent operator may be written

1
i
dt H0
A(t)],
(2.28)
A(t) = T r[eN T e C
Z
where C is the contour of Figure 2.2.2 b). The utility of such contours will
be more evident in the following, and particularly in Chapter 13.
2.3 The Interaction Picture and the Viable Expansion
Im(t)
25
Im(t)
t0
t0
tRe(t)
t = t0 i
Re(t)
t = t0 i
a)
b)
Fig. 2.2. a) Contour for thermal averages at time t0 . The or imaginary time axis
is shown: = Im(t) is actually a real variable. b) Contour for thermal averages
at time t. Complex times in the vertical track are latest.

In the Heisenberg picture, states are xed while the operators AH (t) carry the
time evolution; however,AH (t) requires US (t, t0 ); it is ironic that generally we
cannot even write down the Hamiltonian itself
HH (t) = US (t, t0 )HS (t)US (t, t0 ).
Life is easy in the only case of stationary problems, when
AH (t) = exp [iHS (t t0 )] AS exp [iHS (t t0 )] ,
and in particular HH = HS . In order to be able to expand in powers of V one
introduces the Interaction Picture in which the operators evolve only with
H0 :
(2.29)
AI (t) = eiH0 t AS eiH0 t
while the wave function is dened by
I (t) = eiH0 t S (t).
(2.30)
The physics is unaltered since

S (t)|e
iH0 t iH0 t
AS (t)e
A(t) = S (t)|AS (t)|S (t) =

e
|S (t) = I (t)|AI (t)|I (t) . (2.31)
iH0 t iH0 t
Note that H = S at t = t0 and I = S at t = 0; moreover, during the

golden age when the system is unperturbed,
One nds:
i
I (t) = eiH0 t S (t) = eiE0 t S (t).
(2.32)
I (t) = eiH0 t V (t)S (t) = VI (t)I (t)

t
(2.33)
26
Golden age: V=0

I = eiE0 t S
Today
H = I
System in
ground state
of H0
S = I
t0
Fig. 2.3. Adiabatic switching and the three quantum pictures.
which is like Scr

odinger equation without the obvious part of the dynamics.
The evolution operator in the interaction picture from an arbitrary time
satises
and
I (t) = UI (t, )I ( )
(2.34)
UI (t, ) = eiH0 (t ) US (t, )
(2.35)
UI (t, ) = VI (t)UI (t, ).

t
t
i
dt VI (t )
UI (t, ) = T e h
i
The solution
(2.36)
(2.37)
is at the basis of all perturbation theory.

One must be able to switch representation. In the Heisenberg picture,
A = H |AH |H , where H = S (t0 ); moreover, using (2.32), I (t0 )) =
eiE0 t0 S (t0 ) and we may write A = I (t0 )|AH |I (t0 ) . In the interaction
picture,
A(t) = I (t)|AI (t)|I (t) = I (t0 )|UI (t, t0 )AI (t)UI (t, t0 )|I (t0 ) .
(2.38)
Therefore,
(2.39)
AH = UI (t, t0 )AI (t)UI (t, t0 ),
which is just (2.12) with S replaced by I . Having succeeded in writing Heisenberg operator in terms of interaction ones, we can expand them in series of
VI .
Problems
2.1. For H0 =
p p cp cp ,
show that nk is given by the Fermi distribution,
np =
1
1
Tr(np ) =
.
Z
1 + e(p )
(2.40)
27
2.2. Let H = H0 + H1 and U0S (t, t ) the evolution operator for H0 . Write an
equation for the evolution operator US (t, t ) for H.
2.3. Let A1 , A2 denote fermion creation or annihilation operators and an
operator such as a density which commutes with A1 , A2 under Wicks T
d
ordering. Find dt
T {A1 (t1 )A2 (t2 )(t)}. How is the result generalized to several
operators A1 , An ?
2.4. Derive the useful identity [205] holding for any Hamiltonian H that
depends on a parameter
d iH (t t)
e
= i
d

t
t
d eiH (t )
dH iH ( t)
e
.
d
(2.41)
3 Atomic Shells and Multiplets
3.1 Shell Structure of Atoms

X-ray spectroscopy and Photoemission (see Chapter 6) show a correspondence between the electronic levels in many-electron atoms and in Hydrogen. The next Table shows the inner levels of Fe; the rst column presents
the spectroscopic notation, the second the corresponding Hydrogen quantum
numbers and the third the measured binng energy.
Shell
K
LI
LII
LIII
MI
MII
MIII
nlj EB (eV)
1s 21 7112.0
2p 12 842.0
2p 32 719.9
3s 21 706.8
3p 12 94.0
3p 32 52.7
3d 23 52.7
It is evident that the shells correspond to the principal quantum numbers

and the subshells to the spin-orbit split levels. This Chapter is devoted to
the elementary theory of the shell structure and the atomic multiplets, which
requires the use of introductory many-body concepts and the inclusion of
relativistic eects.
3.2 Hartree-Fock Method

Consider the true non-relativistic N-body Hamiltonian for an atom or molecule
(1.60,1.61,1.62), namely, H (r1 , r2 , . . . , rN ) = H0 + V, with

1
h(i)
(3.1)
H0 = T + Hw =
2i + w(ri ) ,
2
i
i
V =
1
v (ri ri )
2
i=j
(3.2)
30
(or (1.63) in second quantization). The external potential w is due to the

nucleus (or nuclei). Solving the Schr
odinger equation is a formidable task;
the Hartree-Fock (HF) Method uses as a natural rst approximation the
determinantal wave function of the form (1.4)
u1 (1) u2 (1) . . . uN (1)

1 u1 (2) u2 (2) . . . uN (2)
(3.3)
(1, 2, . . . N ) =
... ... ...
N! . . .
u1 (N ) u2 (N ) . . . uN (N )
written in terms of spin-orbitals ui (i) to be determined. This enables us to
write the average of the one-body operator according to (1.7)
|F | =
N
ui |f |ui .
and the average of the two-body operator according to (1.10)

|F | =
N
[ui (1)uj (2)|f (1, 2)|ui (1)uj (2) uj (1)ui (2)|f (1, 2)|ui (1)uj (2) ].
j=i
The average of the N-electron Hamiltonian yields

|H| = EN =

i
Ii +
1
(Cij Eij ),
2
(3.4)
ij
where Ii = ui |h(i)|ui , while Cij and Eij are respectively the Coulomb and
exchange integrals involving the Coulomb interaction and orbitals i and j.
We are facing two main problems: on one hand, the determinantal wave
function is appropriate for independent electrons and on the other we need to
specify the orbitals somehow. We nd a way out of both problems if we use the
determinantal form as a variational ansatz and seek for optimal spin-orbitals;
we impose normalization by a Lagrange multiplier for each spin-orbital. If we
wish, we may use other Lagrange multipliers to enforce orthogonality, but if
we forget this requirement, orthogonal spin-orbitals are obtained anyhow.
The main reason why the HF approximation is important is that the
equations are written in the same way for all systems; we can illustrate the
procedure starting with the ground state of He. Then the two spin-orbitals
are u1s = u1s (r) and u1s = u1s (r), in obvious notation, and
E2 = u1s (1)| h(1) |u1s (1) + u1s (2)| h(2) |u1s (2)
+ u1s (1)u1s (2)| r112 |u1s (1)u1s (2)
(3.5)
does not involve exchange terms. We vary the orbital u1s (r) requiring E2 =
N, where N is the norm of u1s (r) and is the Lagrange multiplier. We
obtain the Hartree equation

h(1)u1s (1) + u1s (1)
d3 r2 u1s (2)
1
u1s (2) = u1s (1).
r12
31
(3.6)
(we speak of Hartree-Fock when exchange terms appear). Thus the optimal
orbital is obtained by a Schr
odinger-like equation where the electrons feels,
along with the nuclear potential contained in h, the Hartree potential

1
u1s (2);
(3.7)
V H (1) = d3 r2 u1s (2)
r12
V H is just the electrostatic potential due to the charge cloud of the oppositespin electron. Having chosen an independent-electron form of we are trying
to compensate the neglect of the Coulomb interaction by including its average
as an eective potential or mean eld. The Hartree approximation takes into
account the quantum nature of the electron and is self-consistent, that is, it
accounts for electrostatics. Indeed, HF is also called the self-consistent eld
method. Yet, it is still far from exact. The He ground state energy turns out
to be E2 = 77.866 eV, which is more than 1 eV too high compared to the
exact result ( 79 eV). This large discrepancy is the correlation energy and
is due exclusively to the determinantal form of . In other terms, the electron
does not see the average cloud of the other one, but a point particle with
which it can correlate its motion. The discrepancy may be small compared
to the binding energy of the system and also compared to core-level binding
energies, but since 1eV is the scale of chemical binding energy the accuracy
of the HF method is questionable if one wants to predict chemical trends.
The excited state 1s2s3 S of He can also be approximated variationally
as discussed in Sect. 1.4. Triplet He is called Orthohelium and converts to
the singlet Parahelium after a long time (spin-orbit coupling is small). The
relevant congurations are u1s u2s , with mS = 1, u1s u2s , with mS = 1.
Of course, there is also mS = 0, with the embarassing non-determinantal
conguration 12 [u1s u2s + u1s u2s ]. This state would be outside the scope
of the HF method, but we know that its energy is the same and anyhow it can
be reached from the determinantal states by a 90 degrees rotation. Hence we
can concentrate on u1s u2s , and repeating the above argument and setting
a = u1s , b = u2s we nd:
E2 = Ia + Ib + Cab Eab .
(3.8)
The extremal condition E = 0 is subject to the further conditions a|a =

b|b = 1, and we need two Lagrange multipliers a and b . So, we get, varying
a:

1
1
h(1)a(1) + a(1) d2 b (2)

b(2) b(1) d2 b (2)
a(2) = a a(1). (3.9)
r12
r12
Varying b, we nd the same with a and b interchanged. These are the HF
equations for the problem. For parallel spins a non-local exchange potential
32
arises. However, this is a non-local potential (the r.h.s does not depend just
on the local value of a).

V ex ( x)a( x) = b( x)
b (y)a( y )
|x y|

dy
(3.10)
The physical signicance of the exchange term will be further discussed in

the next subsection.
It is worth mentioning that had we started with a simple product, without
anti-symmetrization, we should have found the direct term but we would have
missed the exchange one. The Hartree method, which neglects exchange, has
also been widely applied. Its results are not always worse than those of HF.
Indeed, since the orbitals are lled according to the aufbau method, the shell
structure is reproduced and the Pauli principle is not totally ignored; both
methods neglect correlation, and this is the most serious limitation for both.
Let us take the scalar product of Equation (3.9) by |b :

b (1)a(1)|b(2)|2 |b(1)|2 b (2)a(2)
= a b | a .
b| h |a + d1d2
r12
The formidable-looking integral vanishes and one is left with
b| h |a = a b | a .
(3.11)
Exchanging a and b,
a| h |b = b a | b .
Taking the complex conjugate and subtracting from (3.11)we get
0 = (a b ) b | a ,
and non-degenerate orbitals are orthogonal.
The analogy of the HF equations to Schrodingers suggests that the so
called Koopmans eigenvalue a is the energy eigenvalue of the electron moving in spin-orbital a. However, one should not give any physical signicance
to the individual orbitals. One can introduce unitary linear transformations
of orbitals. Then, the determinant does not change, and nothing changes
since has a physical meaning, the spin-orbitals do not possess any by themselves. No physical observable corresponds to the energy of an orbital. Let us
take the scalar product of Equation (3.9) by |a :

a(1) b(1)a(2)b(2)
|a(1)|2 |b(2)|2
d1d2
= a ,
a(1)| h(1) |a(1) + d1d2
r12
r12
that is,
a = Ia + Cab Eab .
Using
(3.12)
33
b = Ib + Cba Eba , Cba = Cab , Eba = Eab ,

we may conclude that
a + b = Ia + Ib + 2 (Cab Eab )
= E2 = Ia + Ib + Cab Eab
and the energy of the atom is dierent from the sum of the Koopmans
eigenvalues.
Rather, they can be thought of as approximations to ionization potentials. Suppose the atom is photoionized, and the photo-electron is sent to the
threshold of the continuum (also called vacuum level). The ionization potential is the dierence between the energy of the initial state He 1s2s3 S and the
energy of the nal state He+ . The initial energy is E2 = Ia + Ib + Cab Eab ,
where a=1s, b=2s; indeed on removing the electron b, we must remove
b = Ib + Cab Eab . This fact is known as Koopmans theorem, and shows that
the HF calculations do bring some information about the excited states, after
all. The main weakness of this approximation is evident from the above He
1s2s 3 S example, too. The nal state He+ is hydrogen-like, with Z=2. If we
estimate the ionization potential by Koopmans eigenvalue, we pretend that
in the nal state the electron keeps its unrelaxed orbital a, which is computed
including the potential due to the electron in b. Thus, Koopmans eigenvalues imply a frozen-orbital approximation. One can x this problem by doing
separate HF calculations for the initial and nal congurations. This is called
the SCF method ; the errors due to correlation eects, however, cannot be
removed within the HF approach, and require the methods of Chapter 11.
The advantage of the HF approach over more accurate variational methods is that the equations are system-independent. The generalization of Equation ( 3.8) to the N-electron problem reads
EN =
N
1
[Cij Eij ] .
2
N
Ii +
(3.13)
i=j
The Koopmans eigenvalues are

(i)
(Cij Eij ) = EN EN 1
i = Ii +
(3.14)
j
(i)
where EN 1 refers to the system ionized in spin-orbital i, while the other spinorbitals remain frozen. Looking for the extremum of energy constrained by
normalization one nds the HF equations. We introduce the direct potential
d
V =
N

i

Vid (r),
Vid (r)
dr
|ui (r )|2

|r r |
summed over all electrons, and the exchange potential
34
V ex =

Viex (r), Viex (r)f (r) = ui (r)
where the summation
dr
ui (r ) f (r )
|r r |
(3.15)
runs over the spin-orbitals with the same spin as i.
For an atom with atomic number Z, the Fock operator

f=
p2
Z
+ V d (r) V ex (r)
2m |r|
(3.16)
allows to write the HF equations in the deceptively simple form

f ui (r) = i ui (r).
(3.17)
The HF equations have a complete othonormal set of solutions. The lowest N

allow to build the determinantal wave functions. The rest are called virtual
orbitals ; they are not directly related to any experiment involving excited
states, but they are often useful to generate multi-determinantal developments, like the Conguration Interaction expansion.
In open-shell system one does not know a priori which conguration will
give the lowest energy, and problems are generally harder than in closedshell systems. For instance, the u1s and u2s orbitals of Be (Z=4) are spinindependent and are determined by a pair of coupled HF equations.The problem is more involved with Li (Z=3). We may arbitrarily set the unpaired 2s
electron with spin up. This implies that the up-spin 1s electrons has the exchange interaction while the down spin electron does not have such a term,
hence there are 2 1s levels and 3 HF equations must be solved. The splitting
of the core level is physically correct. This is the HF method in its general
form, which is often termed unrestricted HF, or spin-polarized HF. In order
to simplify the computations the restricted HF method has been introduced.
One then imposes the same orbital for both spins, with half exchange interaction for both spin directions. This approximation, however, leads to incorrect
molecular dissociation at large distances and cannot describe magnetism.
The ground state conguration of every atomic species, which is reported
in Mendeleje tables, is the one which yields the minimum energy with the
HF method (or its relativistic extension). The Koopmans eigenvalues for the
closed-shell Cu+ ions (658.4 eV for 1s, 82.3 eV for 2s, 71.83 eV for 2p, 10.65
eV for 3s, 7.27 eV for 3p and 1.6 eV for 3d) may be compared with the
experimental binding energies (662 eV, 81.3 eV, 61.6 eV, 11.6 eV, 6.1 eV,
0.71 eV, respectively). In this case the relativistic eects are small and there
is no doubt that the general trend is correct, however the relative error for
the external shell exceeds 100%. The methods of Chapter 11 are needed in
order to improve the situation. While the true ground-state energy is always
lower than in the HF approximation, there is no such relation between the
Koopmans eigenvalues and the true levels of the system; in more advanced
theories, actually, they are resonances (see Chapter 5 ).
35
The relativistic extension of the HF method is called Dirac-Fock method.

One replaces the one-body Hamiltonian h by Diracs Hamiltonian [203][202]
hD (i) = ci .pi + mc2
Ze2
,
|r i |
(3.18)
where i is Diracs velocity [203] of electron i, = 4 is Diracs matrix and a

four-component wave function is sought. For light atoms, the Coulomb interaction continues to be used, although a better alternative is Breits interaction
[204]
e2
exp[ikr12 ](1 (1) (2)), ck =
WB (1, 2) =
r12
where h is the absolute value of the energy jump of each electron in the
collision. Breits interaction is particularly needed for the inner shells of
he3avy elements. With such changes, the self-consistent equations work basically as in the non-relativistic case, and Dirac-Fock atomic codes are available since a long time. GRASP (General Purpose Relativistic Atomic Structure Program[158]) labels Dirac-Fock solutions by J 2 , Jz , parity and seniority
number 9.9(See Chapter 9.9) does a partial conguration interaction, keeps
into account the nite nuclear dimensions and includes the main corrections
due to Quantum Electrodynamics.
3.2.1 Physical Meaning of Exchange: the Cohesion of a Simple
Metal
In all metals, the conduction electrons shield the electric eld of the Ions.
The shielded potential seen by an electron at the Fermi level is considerably attenuated, and in Aluminum and other metals with s and p conduction
bands, it is nearly at. These metals are called simple since many of their
properties can be explained with the model of free electrons. The theory of
Sommerfeld considers a gas of N electrons conned to volume V, with large N
3
kF
and V such that N
V = n = 3 2 , where n is the number density of conduction
electrons, that occupy all the states up to the Fermi level. The Fermi sphere
exists indeed and the Sommerfeld theory explains some facts correctly: the
contribution of electrons to the specic heat of the metals is correctly predicted to grow linearly with the temperature T . But the energy density of
F
is positive, and the electrons are prevented from
the electron gas VE = 3nE
5
escaping from the metal by a box. The real nature of this box is unexplained.
For the metal to be stable, an electron must be attracted by it. The Jellium is a hypothetical metal in which a Fermi liquid, that is an interacting
electron gas, is neutralized by positive uniform charge. The nuclear charges
are smeared out in uniform way and the system have complete translational
symmetry; however oversimplied, this model is still hard and nobody knows
the exact solution. It is a traditional bench mark of the many-body theory ,
36
and a constant source of ideas that then are applied to the realistic calculations. The electrostatic eects are included in the theory, and they improve
it considerably. We see that what we can learn from SCF approximation.
The most fundamental issue is the cohesion of a metal piece. We consider
a large cube of Jellium of volume V containing N electrons, with N/V = n
and impose periodic boundary conditions. The method of Hartree describes
to the state of the Jellium with a product wave function of the form
(1, 2, , N ) = u1 (r1 )u2 (r2 ) uN (rN )
(3.19)
where ui are unknown spin-orbitals; the Hartree potential is Vb + Vd , where

Vb is the positive background potential and Vd is the direct potential.
We look for a translationally invariant solution1 as a natural rst choice.
Then the space function in the u spin-orbitals is a plane-wave; the density
is a constant and cancels exactly the background density. Thus, Vb + Vd = 0
and the electrons are left with the mere kinetic energy, which is positive; the
result (a piece of metal cannot exist) is very unrealistic.
In the Hartree-Fock approximation, we can still nd plane-wave orbital
solutions, but since the Hartree term just cancels the background, the only
(exc)
. To obtain its expression, one can
potential is the exchange potential Vk

e2

write down the direct term Vd uk (r) = uk (r) k dr |rr
| uk (r ) uk (r) and
perform the charateristic exchange

4e2
e2

ikr 1
(r )uk (r) = e
Vex uk =
u
.
(3.20)
dr
k
|r r |
V |k k|2

k
The exchange term also goes like eikr , and the Hartree-Fock equations read
[
p2
eikr
e2
4
eikr
.
=
(k)
]
2m
V |k k|2 V
V
(3.21)
The Koopmans eigenvalue is

4e2
2 k2
h
(k) =
2m
(2)3
d3 k (kF k )
|k
1
k|2 .
(3.22)
For small k, 2e kF . The electron moves in a potential well in k space;

there is a singularity for k = kF . The exchange term is attractive, and this
fact has a simple physical interpretation. Every electron travels encircled by a
space region (Fermi hole) in which there is deciency of electrons of its same
spin. The electron and the Fermi hole constitute with a quasi-particle, very
1
Spin density wave solution of the Hartree-Fock equations are energetically favored at low n, however the energy dierence is far too small to be relevant o the
cohesion issue. Cs has low enough n to have a spin density wave in the ground state
according to Hartree-Fock equations; the actual metal, however, is not magnetic.
3.3 Virial Theorem
37
dierent from that of a free electron. The Jellium is stable if an electron at the
Fermi level is bound. This depends on the competition between the negative
contribution of the exchange term and that positive one of the kinetic energy.
The outcome of the competition depends on kF , that is an the density n of the
Jellium. For suciently small density (kF a0 << 1) the attraction prevails,
and the metal exists.
The function (k) has a logarithmic singularity for k = kF , where its
derivative diverges. This is physically wrong, no such behavior is observed,
but we have understood the reason for the existence of metals, and if we want
to understand more we must go into the many-body problem.
3.3 Virial Theorem

For a classical particle moving in a closed orbit in a potential V, one nds
p2
d
dt r p = 2m + r F , where F = V is the force. Averaging over a period,
d
dt
r p = 0, yields the well known Virial theorem 2T = r V where T
is the kinetic energy.
d
r p = [r p, H] ; averaging over an
For a Schr
odinger particle i
h dt
eigenstate of H, the r.h.s. vanishes; since the quantum commutators give
dp
p
dr
dt = V, dt = m we obtain the same Virial theorem 2T = r V .
In Diracs theory[203], if is an eigenspinor of H = c p + mc2 + V (r),
(using standard notation, with Diracs velocity) a similar procedure leads
to |c p| = |cr | .
Here we are particularly interested in a system of N non-relativistic particles, including electrons and nuclei, in Coulomb interaction; let ei and mi
denote charges and masses of the charged particles; the Hamiltonian is of the
form
H =
N
N

ei ej
2 2i
h
+
T + V.
2 i=1 mi
rij
i<j
(3.23)
The ground state wave function ({r i }) depends on the set {ri } of the N
position vectors. Consider the rescaled, normalized wave function
({ri }) =
3N
2
({r i }).
(3.24)
One checks easily that kinetic and Coulomb energies scale dierently, that is,
|T | = 2 |T | ,
(3.25)
|V | = |V | .
(3.26)
|H| = 2 |T | + |V | .
(3.27)
while
Hence,
38
Treating as a variational parameter, we nd the condition

2|T | + |V | = 0,
(3.28)
but since the optimum value is = 1 we end up with the Virial theorem
2T + V = 0.
(3.29)
This is one of the few known exact statements about interacting many-body
problems. The Hartree-Fock ground state satises the Virial Theorem, and
an approximate self-consistent calculation can be improved by scaling.
3.4 Hellmann-Feynman Theorem

Let H = H() denote a Hamiltonian which depends on a parameter and
H = E. Then
dH
dE
= |
| .
(3.30)
d
d
This follows from E() = |H()| and from the fact that a small d
produces a rst-order variation d which is orthogonal to . This is another
one of the few exact results for interacting many-body problems.
3.5 Central Field

In this Section we study the ground state and the low excited states of the
atom with atomic number Z, including the most important relativistic eects.
Using symmetry (exact and approximate) we can tell a lot without heavy
computing. The model Hamiltonian is

Htot = H0 + HC + Hrel
with
H0 =
Z 2

p

Ze2
,
2m
ri
i
HC =

=
Hrel
Z

e2
,
r
i<j ij
Z
(ri )Li S i
(3.31)
(3.32)
(3.33)
(3.34)
where pi and ri are electron momenta and coordinates, rij are distances;

contains the spin-orbit coupling between orbital angular momentum Li
Hrel
and spin Si of electron i which is the most notable relativistic correction
3.5 Central Field
39
(other corrections involving the orbital currents are actually larger, but less
evident since they fail to split levels). In the central eld model, the states of
the atom are constructed using 1-electron orbitals computed with a suitable
central V (r); the wave functions dier from the Hydrogen-like orbitals only in
the radial functions RnL (r). In terms of the one-electron basis states labeled
by n, L, m, mS quantum numbers2 , one starts assuming a conguration of the
atom (see Table 3.1).
H
He
Li
Be
B
C
N
O
F
Ne
Na
Mg
Al
Si
P
S
Cl
Ar
K
Ca
Sc
Ti
V
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Cr 24 3d5 4s Ag 47 4dl0 5s
ls
2
Mn 25 3d5 4s2 Cd 48 4dl0 5s2
1s
Fe 26 3d6 4s2 In 49 5s2 5p
2s
Co 27 3d7 4s2 Sn 50 5s2 5p2
2s2
2s2 2p Ni 28 3d8 4s2 Sb 51 5s2 5p3
2s2 2p2 Cu 29 3d10 4s Te 52 5s2 5p4
2s2 2p3 Zn 30 3d10 4s2
I 53 5s2 5p5
2s2 2p4 Ga 31 4s2 4p
Xe 54
5p6
2
2
2
5
2s 2p Ge 32 4s 4p
Cs 55
6s
As 33 4s2 4p3 Ba 56
2p6
6s2
Se 34 4s2 4p4 La 57 5d6s2
3s
Br 35 4s2 4p5 Ce 58 4f 5d6s2
3s2
3s2 3p Kr 36 4p6
Pr 59 4f 3 6s2
2
2
5s
3s 3p Rb 37
Nd 60 4f 4 6s2
2
2
3
Sr 38 5s
3s 3p
Pm 61 4f 5 6s2
3s2 3p4 Y 39 4d5s2 Sm 62 4f 6 6s2
3s2 3p5 Z r 40 4d2 5s2 Eu 63 4f 7 6s2
Nb 41 4d4 5s Gd 64 4f 7 5d6s2
3p6
Mo
42 4d5 5s Tb 65 4f 9 6s2
4s
Tc 43 4d5 5s2 Dy 66 4f l0 6s2
4s 2
3d4s2 Ru 44 4d7 5s Ho 67 4f 11 6s2
3d2 4s2 Rh 45 4d8 5s
Er 68 4f 12 6s2
3d3 4s2 Pd 46 4dl0
Tm 69 4f 13 6s2
Yb
Lu
Hf
Ta
W
Re
Os
Ir
Pt
Au
Hg
Tl
Pb
Bi
Po
At
Rn
Fr
Ra
Ac
Th
Pa
U
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
4f 14 6s2
5d6s2
5d2 6s2
5d3 6s2
5d4 6s2
5d5 6s2
5d6 6s2
5d7 6s2
5d9 6s
5d10 6s
5d10 6s
6s2 6p
6s2 6p2
6s2 6p3
6s2 6p4
6s2 6p5
6p6
7s
7s2
6d7s2
6d2 7s2
5f 2 6d7s2
5f 3 6d7s2
Table 3.1. The atomic numbers and the ground state conguration of the elements.
A shell comprises all the orbitals of a given n, and the inner ones, entirely
occupied by electrons, are core shells. The inner shells contribute most of the
binding energy; their charge screens the nuclear potential and contributes in
an important way to the V (r), in which the external electrons
move. The

M
=
0, the z
magnetic quantum number of a closed shell is ML =
L
i
i
component of spin is zero and thus all angular momentum quantum numbers
2
This is a very convenient basis, but individual electron quantum numbers do not
represent observable quantities; any measurement will give the quantum numbers
of the atom.
40
vanish; the parity is +1. Thus, the parity and angular momentum quantum
numbers of the atom are determined by the outer (or valence) electrons. This
suggests that Htot can be simplied. The valence electrons are modeled by

,
H = Hv + HC + Hrel
(3.35)
where the sums are restricted to incomplete shells. The residual Coulomb
interaction HC does not contain the core electron contributions and can be

dealt with approximately as a perturbation; for light atoms Hrel
is also a

perturbation. In the limit HC = Hrel = 0 all the states in a given conguration would be degenerate; for instance, a C atom in the fundamental
2 2
2
conguration
1s 2s 2p has 2 electrons in the six p spin-orbitals and would
6
have a
= 15 times degenerate ground state. This degeneracy stems from
2
the invariance of Hv for independent rotations of electrons orbitals in space
and of the total spin. The orbital angular momentum of each electron and
the total spin are good quantum numbers in this limit. In the presence of
interactions, the only conservation laws are
[H, J ] = 0
(3.36)
where J = L + S is the total angular momentum (invariance under rigid

rotations) and the parity
[H, ] = 0
(3.37)
(neglecting quite tiny eects of electro-weak mixing).

3.5.1 L-S Multiplets (Hrel
0 Limit)

In light atoms, one can neglect Hrel
to a good approximation; HC breaks the
invariance for independent orbital rotations of the electrons, that change rij .
i
One is left with the invariance under rigid rotations R(a) = e h aL . Hence
H commutes with the components of the total orbital angular momentum
L, and the conguration gives raise to a Russell-Saunders multiplet, with
dierent energies for dierent L. Besides, H commutes with the i-th electron
spin S i . Is S i a good quantum number? Not at all, since individual electrons

do not possess observables. What we can measure is the total spin S = i S i
of the atom. Moreover, L and the total electronic angular momentum
J =L+S
(3.38)
can be measured. But what quantum numbers are compatible? H commutes

with L2 , S 2 , J 2 , Lz , Sz , Jz . However, since J 2 fails to commute with Li and
Si , these operators are not all compatible and we can label states in two
alternative ways: 1) diagonalizing H, L2 , S 2 , Lz , S z , J z = Lz + S z , dening the |LSMLMS basis with quantum numbers E, L, S, ML , MS , MJ =
3.5 Central Field
41
ML + MS . Since H is invariant for independent rigid rotations in ordinary space and spin space, E is independent of ML and MS . 2) diagonalizing H, L2 , S 2 , J 2 , J z dening the |LSJMJ basis with quantum numbers
E, L, S, J, MJ . The two bases are connected by a unitary transformation, and
both schemes are referred to in literature as L-S or Russell-Saunders scheme.
The energy levels of this approximation, or atomic terms, are denoted with
symbols of the type 2S+1 L: as an example, 2 P has L=1, S=1/2; they are
degenerate (2L+1)(2S+1) times.
L-S Terms Inside a Given Conguration
The L-S terms belonging to a given conguration are resolved by HC . One
can easily nd what terms arise. For closed shells there is only 1 S. If a single
electron optical electron moves outside lled shells the entire atom has the
quantum numbers L, ML and MS = 12 of the electron. If a single electron
lacks in order to make lled shells, there is a single hole, this counts like
an electron with opposite values of ML and MS . If there are two inequivalent electrons (dierent principal quantum number n) outside lled shells,
one must sum their L and S. For instance, from two electron labeled np, n p
one builds the atomic terms 1 S,3 S,1 P,3 P,1 D,3 D. With 2 or more equivalent
electrons, the possible terms are limited by from the Pauli principle, and the
best thing is to proceed by examples.
Example: C atom (conguration 1s2 2s2 2p2 ). The closed shells may be ignored, and we must consider the conguration p2 . The 1-electron states available are (m, ) with m mL = 1, 0, 1 and = 21 ; there are 6 spin-orbitals
involved. Without the Pauli principle we would nd 1 S,3 S,1 P,3 P,1 D,3 D,
that is 1 + 3 + 3 + 9 + 5 + 15 = 36 states. Many of those terms are forbidden, since only the 2 electron determinants
(m1 1 , m2 2 ) involving dierent

6
(m, ) are allowed. There are
= 15 pairs of dierent spin-orbitals; by
2
linear combination on can form 15 allowed 2-electron states with well dened
L, S, ML , MS . The (m1 1 , m2 2 ) determinants are labelled by ML and MS ,
and those with parallel spins belong to S=1; MS = 0 is compatible with
singlet and triplet. Not all determinants have well dened L; however the determinant with the largest ML must belong to the maximum L. In this case
the largest ML is 2, and corresponds to (m1 1 , m2 2 ) = (1+, 1), having
MS = 0. Thus there is a 1 D term. 1 D has 5 states, ML = 2, 1, 0, 1, 2,
and we must nd 10 more orthogonal states, with L < 2. With ML = 1 we
can nd the 4 determinants (1, 0) having mixed L. A linear combination,
which might be found using the shift operator S on the state with ML = 2,
belongs to 1 D; 3 ML = 1 states are left, with L = 1 and MS 1; these belong
to 3 P (9 states). we have found 5+9=14 states out of 15; there is room only
for a 1 S. Thus the conguration is resolved as follows:
p2 1 S,3 P,1 D.
(3.39)
42
(See the Problems section for other examples.)

The parity of the atom is the product of those ()Li of the orbitals in
the determinant. The ground state conguration of C is even, and all terms
including 3 P are. The excited conguration 2s1 2p3 is odd and yields a term
denoted by 3 P o , where o stands for odd.
The L-S wave functions can be obtained by combining the angular momenta with the Clebsh-Gordan coecients. For two electrons, one forms determinants |m1 ms1 m2 ms2 , recalling that in order to avoid double counting
the orbitals must be ordered in some way; then,
|LSMLMS =

|m1 m
m1 ms 1 m2 ms 2
(3.40)
1
1
s1 m2 ms2 L1 m1 L2 m2 |LML ms1 ms2 |SMS ;
2
2
one has to normalize the result again in general (see Problem 3.3, 3.4). A
third electron can then be added if needed by multiplying by a spin-orbital
with the Clebsh-Gordan coecients, and antisymmetrizing the result. For
many electrons this build-up process becomes very time consuming, but one
can make the process faster by using fractional parentage coecients [147]
[148]. These table present the wave functions coded in a special way; this
saves labor, but introduces no new physical concepts.
3.5.2 Hunds First Rule
Hund established two empirical rules, that hold with no exception in atomic
physics, and are very popular with the students because there is no proof to
learn. This is the rst.
The lowest L-S level of the atomic conguration has the lowest S and the highest L
compatibly with S.
This is no theorem, but is true and reasonable, since high spin implies a
very antisymmetrical orbital wave function and therefore a reduced repulsion; increasing L also lowers the energy because higher L wave functions are
more diuse. For Z=6 ( C atom), the conguration is 2p2 : the terms are
1 3
S, P,1 D and the ground state is 3 P . For Z=74 (Tungsten ) the conguration is 5p6 5d4 6s2 ; the incomplete shell is 5d4 . With ML = 2, 1, 0, 1, 2, all
spins can be parallel, so the Hund rule wants S = 2. The maximum ML is
2 + 1 + 0 + (1) = 2. The ground term is 5 D. In the L-S limit (no spin-orbit
interaction) the states of a term are all degenerate. Combining the states of a
LS term with various ML and MS by means of the Clebsh-Gordan coecients
one builds |L, S, J, MJ states. The allowed values of J are obtained from L
with S by the usual angular momentum summation rule, while MJ (J, J).
Thus a term 3 F, (L=3, S=1), gives raise to 3 F4 ,3 F3 and 3 F2 , where the index to the right is J. The energy E of the atomic levels is independent of
3.6 Atomic Coulomb Integrals

1
p2
S0
D2
43
3
3
P2
P1
P0
Fig. 3.1. The p2 conguration splits into the L-S terms 1 S,1 D and 3 P , separated
by the Coulomb interaction. The relativistic correction then produces the atomic
levels as shown.

J and MJ . If we introduce Hrel
as a perturbation, the degeneracy is partly
removed; E depends on J but not on MJ . Here is the pattern of the levels
for the Carbon atom (see Fig. 3.1).

The interaction matrix elements ua ub | r112 |uc ud between spin-orbitals factor
into spin scalar products and Coulomb integrals a(1)b(2)| r112 |c(1)d(2) ina
(r)YLM (, ), and the like. One
volving the central eld orbitals a(r) = RnL
1
1
can expand r12 |r1 r2 | in polar coordinates such that r i (ri , i , i ).
Imposing that the result be the Greens function of Poissons equation one
nds, in terms of spherical harmonics and of the shorthand notations

r> = r1 (r1 r2 ) + r2 (r2 r1 ),
(3.41)
r< = r2 (r1 r2 ) + r1 (r2 r1 ),
K
K

r<
YKm
(1 , 1 )YKm (2 , 2 )
1
.
= 4
K+1
r12
(2K + 1)
r
K=0 >
m=K
(3.42)
To carry out this calculation, one needs integrals involving three spherical
harmonics; these are easily obtained by codes like Mathematica or Maple;
they are tabulated in the literature [144] as cK tables, where

dYLa ma ()YKm ()YLc mc ()
One easily nds that
2K + 1 K
c (La ma , Lc mc )(m, ma mc ).
4
(3.43)
44
ab|
1
|cd = (ma + mb , mc + md )
r12
cK (La , ma , Lc , mc )cK (Ld , md , Lb , mb )RK (ab, cd)
(3.44)
K=0
Actually this summation has in any case a few nonzero terms since they are
restricted by the triangular rules
|Ld Lb | K Ld + Lb
|La Lc | K La + Lc .
(3.45)
The radial integrals contain the specics of the problem, the rest is geometry.
For the diagonal elements (c,d)=(a,b) one uses the Slater integrals
F K (na La , nb Lb ) RK (a, b, a, b) (Coulomb interaction)
GK (na La , nb Lb ) RK (a, b, b, a) (Exchange interaction).
Using other widespread notations,

1
|ab =
aK (La ma , Lb mb )F K (na La , nb Lb )
ab|
r12
(3.46)
(3.47)
with
aK (La ma , Lb mb ) = cK (La ma , La ma )cK (Lb mb , Lb mb )
and
ab|
(3.48)

1
|ba =
bK (La ma , Lb mb )GK (na La , nb Lb )
r12
(3.49)
with
bK = |cK |2 .
(3.50)
HC .
The reIn this way, one can calculate the multiplet splittings due to
pulsion is invariant for space rotations and (separately) spin rotations of the
atom, which are exponentials in L, S and the HC matrix is diagonal in the
|LSML MS basis.
A closed shell contains of angular momentum L contains 2(2L + 1) electrons; the conguration with 2(2L + 1) n electrons is said to contain n
holes and to be conjugate to the one with n electrons. The expressions of
the matrices of the Coulomb interactions in terms of the Slater integrals are
the same for the conjugate congurations; however the one for n electrons
refers to the empty shell as the energy zero, and the one for n holes refers to
the lled shell. Thus,if the Slater parameters are the same in both cases, the
term separations are the same. The direct diagonalization of the Coulomb
matrix by standard methods becomes painstaking with many electrons; efcient methods based on tensor operators are reviewed by Weissbluth [147]
who also reports the results for the most common congurations. The multiplet energies for many more congurations are given by Condon and Shortley
[144] and by Slater [145].
45
Relativistic Correction to the L-S Coupling

Hydrogen is a special case, since L remains a good quantum number3 despite

= LS and the states can be labeled |LSJMJ .
the spin-orbit coupling Hrel

2
Besides, Hrel = 2 (J L2 S 2 ) yields the rst-order correction E(J) =
2 (J(J+1)L(L+1)S(S+1)). Landes interval rule E(J)E(J1) = J

is obeyed.

= i i Li S i fails4 to commute with L2
In many-electron atoms Hrel
and S 2 ; so, L S terms sharing the same J are mixed. In the C case, 3 P2
and 1 D2 contaminate each other, and 1 S0 mixes with 3 P0 . Spins and orbits
exchange angular momentum; only in the H case this does not happen (the

is small, and the mixing
spin of 1 electron is always 1/2.) For light atoms,Hrel
of dierent L-S terms is negligible. So,

EJ = LSJMJ |
i Li S i |LSJMJ = KLS LSJMJ |L S|LSJMJ ,
i
(3.51)
where we have used the Wigner Eckart theorem. for example, the ne structure of the 3 P and 3 F terms of the d2 conguration depends on dierent K
constants. The Lande rule can again be derived.
The LSJM scheme is unitarily equivalent to the LSML MS scheme:

|LSJM =
|LSMLMS LSML MS |LSJM ;
(3.52)
ML MS
since the |LSMLMS vectors have the correct normalization and antisymmetry property the result is automatically normalized. For example, let us
calculate |d2 3 P0 . The relevant Clebsh-Gordan Table is the one for summing
an angular momentum 1 to angular momentum J1 ;
J
J1 + 1
J1
J1 1
m2 = 1
(J1 +m)(J1 +m+1)
(2J1 +1)(2J1 +2)
(J1 +m)(J1 m+1)
2J1 (J1 +1)

(J1 m)(J1 m+1)

2J1 (2J1 +1)
m2 = 0
(J1 m+1)(J1 +m+1)
(2J1 +1)(J1 +1)

m
J1 (J1 +1)
(J1 m)(J1 +m)
J1 (2J1 +1)
m2 = 1

(J1 m)(J1 m+1)

(2J1 +1)(2J1 +2)
(J1 m)(J1 +m+1)
2J1 (2J1 +1)
(J1 +m+1)(J1 +m)
2J1 (2J1 +1)
We identify the momenta as follows: L J1 , S J2 , MS m2 . Since we

want J = 0 we use the last line, yielding 13 , 13 , 13 . So,
1
|d2 3 P0 = {|3 P ML = 1, MS = 1
3
3
| P ML = 0, MS = 0 + |3 P ML = 1, MS = 1. }
3
(3.53)
Even in the full Dirac theory the L of the large component can be used to label
the (4-component) spinor although L fails to commute with Diracs Hamiltonian.
N
4
developing L2 = ( k Lk )2 one obtains squares of angular momenta that commute with the components and cross products that do not; similar considerations
apply to the spin operators.
46
If one substitutes the determinantal expansion derived in Problem 3.4 the

result
1
|d2 3 P0 = {2|1+ , 2+ | 6|0+ , 1+ | 2|2+ , 2 |
30
2|2, 2+ | + |1+ 1 | + |1 1+ | + 2|2 , 1 | 6|1 , 0 |}
(3.54)
is no eigenstate of ML , MS but is eigenstate of J, MJ . The interaction HC is

diagonal in both the LSML MS and the LSJMJ basis, and both fall within
the L-S or Russell-Saunders scheme . However, LSJMJ remains a suitable
basis even in the presence of the spin-orbit interaction.
3.6.1 Hunds Second Rule
The ground level can be found by the Second Hunds rule, also empirical:
The level with the lowest J is lowest if the
shell is less than half lled. The level with
the highest J is lowest if the shell is more
than half lled. There is no spitting (in rst
order) for half lled shells.
For the C atom the ground state has J = 0, as shown in Figure 3.1.
3.6.2 J-J Coupling
The J-J approximation is opposite to L-S coupling: one neglects the Coulomb
interaction HC . This scheme is preferable for large Z and for excited states
involving diuse wave functions. The Hamiltonian reads
H0 =
p2
[ i + V (ri ) + Li S i ]
2m
i
(3.55)
The one-electron eigen-spinors are proportional to the generalized spherical

harmonics |L, J, MJ . These are obtained combining the spherical harmonics
YmL and the spin states and with the Clebsh-Gordan coecients. By the
angular momentum rules one can combine the Slater determinants that we
can form with these spinors to get total J eigenfunctions.
3.6.3 Intermediate Coupling

. This is necIn intermediate coupling one treats exactly both HC and Hrel
essary e.g. for the valence states of heavy atoms where both interactions are
important. With the 1-body eigenspinors of (3.55)one can form Slater determinants |uj1 m1 uj2 m2 | that have MJ quantum number; in general dierent
47

J values can be mixed. In the basis of such determinants the HC + Hrel
matrix is computed and diagonalized; the eigenvectors can be labeled with J
and energy eigenvalues are MJ independent. Alternatively, one can start with
the determinantal wave functions |uL1 mL1 mS1 uL2 mL2 mS2 | and compute the

Hrel
, then perform the unitary transformation to the |LSJM basis where
HC is already diagonal (its diagonal matrix elements depend only on L and S).
We exemplify the latter procedure in the case of the d2 conguration; L = 2

for both electrons and we denote the determinants as |m1 , m2 |; for example,
|1+ , 2+ | has two spin-up electrons with m = 1 and m = 2, respectively. .

To calculate the matrix of Hrel
= [L1 S 1 + L2 S 2 ] , where is the spin
+
orbit parameter, one uses the expansion Li S i = Liz Siz + 12 {L+
i Si +Li Si };

we know how to act on the |m1 , m2 | determinants, using Equation (6.1.1).
So,
1
1
1
[Li S i ]|1+ , 2+ | = {(1 +(2) )|1+ , 2+ |+ [2|2 , 2+ |+2|1, 1|]},
2
2
2
i
(3.56)
and so on. We can calculate the eect of the spin-orbit Hamiltonian on the
|LSJMJ basis. Using the results of the problems above,
HSO |d2 3 P0 = [2|1+ , 2+ | + 8|2 , 2+ | +

30
+
6|0 , 1+ | 7|1 , 1+ | 6|0+ , 0 | 4|2+ , 2 |
2|2, 1 | + 5|1+ , 1 | + 6|1 , 0 |]
Hence,
d2 3 P0 |HSO |d2 3 P0 = .
(3.57)
(3.58)
Besides, since
1
|d2 1 S0 = [|2+ , 2 | |2 , 2+ | |1+ , 1 | + |1 , 1+ | + |0+ , 0 |],
5
(3.59)
2 1
2 3
(3.60)
d S0 |HSO |d P0 = 6.
In this way the full HSO matrix can be built; it is a block matrix since
there is no coupling of dierent J. Letting = 2, one nds the following
Hamiltonians.
For J=3 there is only 3 F3 with energy . For J=1 there is only 3 P
1 , also
0
2
with energy . For J=4 there is the basis 1 G4 ,3 F4, and H =
. For
2 3

4 4 35 0

42 .
J=2, there is the basis 3 F2 , 1 D2 and 3 P2 and H = 4 35
0
5

42
0
1
5
48

6
1
. The degeneracy
6 0
is reduced according to the simple pattern 1 S 1 S0 , 3 P 3 P0 ,3 P1 ,3 P2 ,
1
D 1 D2 , 3 F 3 F2 ,3 F3 ,3 F4 ,1 G 1 G0 . More quantum numbers are
needed with more than 2 equivalent electrons, as discussed in Chapter 9.9.

Finally, if J=0 on the 3 P0 ,1 S0 basis, H = 2
3.7 Meitner-Auger Eect and Spectroscopy

An important spectroscopy is based on the Auger Eect. Actually, this eect
was rst reported in 1923 in Zeitschrift f
ur Physik by the Austrian Physicist
Lise Meitner (1878-1968), whose great contributions to physics tended to
be forgotten on the grounds that she was a lady, was of Jewish origin and
lived in the pre-war Germany. In 1925, the great French physicist Pierre
Auger (1899-1993) independently discovered the eect while investigating in
a bubble chamber the emission of an electron from an atom that absorbs a Xray photon, that is, photoemission. The photoelectrons have kinetic energy
Ek = h EB , where EB () is one of the binding energies of the atom
inner levels. However the Auger electrons have -indepedent energies given
approximately by the empirical Auger law
EA () EB () EB () EB ()
(3.61)
as a combination of 3 binding energies. Thus, EA () is characteristic of the

atomic species and is related to 3 atomic levels. This suggested that X-rays
of adequate energy produce a primary hole in the state a of binding energy
of EB () and a photoelectron; since the ion with a core hole is unstable, the
primary holes is lled up by an electron in a less bound level , and the energy
gained in the process is taken by an electron in level , than is emitted as
the Auger electron.5 For particular transition to happen, it is necessary that
the primary hole is deep enough (its binding energy must exceed the sum of
those of b and c). Of course, energy conservation is compatible with both the
processes a) and b) in Figure 3.2. The Auger transitions are denoted by the
spectroscopic symbols of the shells (or, more precisely, sub-shells) involved:
thus a KL1 M2 transition is due to a primary hole in the K shell that decays
leaving in the nal state L1 and M2 holes and an Auger electron. One speaks
about Core-Core-Core, Core-valence-valence, etc.,transitions depending on
the levels that are involved.
The Auger eect is caused by the Coulomb interaction: two electrons of
the system collide and while one lls up the primary hole, the other is shot
out as the Auger electron. Wentzel already in 1927 proposed a theory of the
Auger eect based on the independent electron model[204]. In the nal state,
the atom has two holes in the spin-orbitals and . In the initial state, one
5
The present two-step description is not completely right, but the one-step one
is more involved and is presented in Sect. 6.4.1.
a)
49
b)
Fig. 3.2. The Auger process. The primary hole is lled by either the or
the electron while the other one is emitted as the Auger electron.
hole was in the deep level ; the other was in the free particle spin-orbital
k that will be occupied by the Auger electron in the nal state. In this way,
the Auger process becomes a collision between two holes. The initial state
|i and the nal state |f of the atom are represented by 2 2 Slater
determinants; they have the same energy and are coupled by the Coulomb
interaction. The transition probability is given by the Fermi golden rule
Pif =
2
i |HC |f |2 ,
h
(3.62)
where HC is the Coulomb interaction. A deep hole can decay X or Auger.

The nal-state holes have same the two ways to decay, until a stable ion is
formed, in which all the holes are in the most external levels. X decay is
faster when the involved levels are distant in energy, because of the 3 factor
in the transition probability due to the density of photon nal states. On
the other hand, the Auger decay prevails if the states a, b and c are near
in energy, as the Auger matrix element is particularly large when the spinorbitals have similar sizes. For this reason, the Auger eect is the dominant
decay mechanism for inner shell holes of light atoms, while in heavy ones X
decay is much more likely.
An alternative mechanism that could take from |i to the nal state
|f comes to mind. One of electrons in the upper states , could ll up
the deep hole via a normal radiative process, emitting a X-ray photon; this
photon could then cause the photoemission of the other electron. Wouldnt
the nal state be the same? Indeed, this alternative process does exist, and
has the name of internal photoemission; unlike the Auger eect, it obeys to
optical selection rules; e.g. it cannot cause a KL1 L1 transition that instead
is observed and is about as probable as any other allowed transition (see
Sect. 3.7.1 for the Auger selection rules). The transition probability of the
internal photoemission can be calculated by perturbation theory, but since
it is a second-order process it turns out to be quite small compared to the
Auger eect.
50
The Auger spectrum is plot of the current versus the kinetic energy of electrons. The Auger instruments used to measure the spectra of molecules and
solid surfaces keep the sample, the source and the electron detector-analyzer
in vacuum. The exciting source can be an electron gun or a X-ray source, like
in the ESCA (Electron Spectroscopy for Chemical Analysis) machines. When
the sample is a solid surface, in order to limit the problem of the contamination from residual gases one needs the ultra-high vacuum. The mean free
path of electrons in a solid depends on their energy, but for the transitions
that are observed commonly it is just several Angstroms; the technique is
sensitive to the surface. The peaks of the spectrum are characteristic of the
atomic species, and the Auger technique lends itself to the surface chemical
analysis using a tiny amount of material in a non-destructive way. With scanning techniques, magnied images of the surface can be obtained in which
the distribution of elements is visualized .
Actually, the atoms that belong to molecules and solids have transition
energies somewhat dierent from free atoms, and a detailed analysis of these
chemical shifts supplies further information. This can be done using CoreCore-Core features, but much more can be learned from the study of the
shapes of Core-Valence-Valence Auger lines, since the external shells are the
most sensitive to the chemical bonds. The analysis of the Auger line shapes
is currently an interesting research topic (see also Chapters 6.4,6.2 12.3. ).
The original Auger formula yields the transition energies with an error
that can be relatively small, but is typically of the order of some tens of eV.
Such an error is very large compared to the precision with which routine
measurements can be done in a modern apparatus. The main problem is that
in the nal state there are two holes, and is necessary to account for their
repulsion; the hole-hole interaction shifts the peaks of the spectrum to lower
kinetic energy and splits them into multiplets. Moreover, every diagrammatic
transition (that is, a line which is predicted from the previous arguments)
has in reality various satellites, that correspond to excites states of the nal
ion. The Wentzel theory is based on a number of assumptions that limit its
validity; it neglects the correlation eects and Relativity. The detailed theory
of the Auger spectra is necessarily involved, even in the case of free atoms.
Nevertheless, the Auger spectroscopy is one of the most important and direct
methods for the study of the correlation eects in molecules and solids.
3.7.1 Auger Selection Rules and Line Intensities
Selection rules arise in the two-step model from the conservation of J 2 , Jz
and parity between the initial, core-hole state |i and the nal state |f
including the Auger electron. In the L-S approximation, L2 , Lz and S 2 , Sz are
also conserved, while in the jj scheme, the states are labeled by the j quantum
M5 N4,5 N4,5
M4 N4,5 N4,5
3
G4 + D2
3
3
S0
51
F4 +1 S0
1
F2,3
P0,1
G4 + D2
P2
P0,1
360
Kinetic energy (eV)
365
370
Fig. 3.3. Sketch of the M5 N4,5 N4,5 and M4 N4,5 N4,5 spectrum of Cd vapor, in
arbitrary units, from measurements by H. Aksela and S. Aksela (Ref. [161]). Many
of the multiplet terms are well resolved. The assignements were done by intermrdiate
coupling calculations of line positions and intensities.
numbers of both holes. Remarkably, the predicted number of lines 6 depends

on the scheme used for |f : consider for example the KLL transitions. In
the pure jj scheme one would predict 6 transitions, namely, KL1 L1 ,KL1 L2 ,
KL1 L3 ,KL2 L2 ,KL2 L3 , and KL3 L3 . In the LS scheme, the 2s0 2p6 conguration yields a 1 S term, 2s1 2p5 gives 1 P and 3 P ; from 2s2 2p4 the coupling of
angular momenta and the Pauli principle would allow 1 S,3 P and 1 D. However, 2s2 2p4 3 P is forbidden by parity conservation; indeed, the primary hole
has L=0 and is even, 2s2 2p4 3 P has L=1 and is even, and by L conservation
the Auger electron must be in a p state, which is odd. The transition to
the odd 2s1 2p5 3 P nal state is allowed, and one predicts 7 lines. When the
spin-orbit interaction is introduced, in intermediate coupling the forbidden
nal-state splits 2s2 2p4 3 P 3 P2 , 3 P1 , 3 P0 ; then, 3 P2 mixes with 1 D2 and
3
P0 mixes with 1 S0 from the same 2s2 2p4 conguration. Thus, 2s2 2p4 3 P2
and 2s2 2p4 3 P0 become more and more allowed as Z grows. Instead, 3 P1 remains purely P and forbidden (as long as one can neglect the mixing with
higher congurations). Thus, the number of lines grows to 9.
If there are open valence shells, each two-hole nal state becomes a multiplet
when valence and core-hole angular momenta are recoupled; to a rst approximation
we can often neglect such complications since usually the splitting is small.
52
In the non-relativistic limit the Auger intensities are computed by evaluating the matrix elements of HC with Pauli spinors; as a simple approximation,
Coulomb waves are used for the Auger electron states (since the Auger electron leaves a doubly charged ion behind) and Hartree-Fock orbitals for the
discrete levels. However, conguration mixing is important. For instance, in
computing KLL spectra, the mixing of 2s2 2p4 1 S0 and 2s0 2p6 1 S0 has an
important eect on the wave functions and hence on the intensities. The calculation simplies if one can approximate |i and |f as two-hole states,
and this is often useful (for instance, in noble gas and in transition metal
spectra). One then often uses as a further simplication the mixed coupling
scheme which consists in treating |i in the jj coupling scheme but |f in
the LS one. This is useful for core-valence-valence spectra in a wide range of
Z, when the spin-orbit interaction is small for valence holes but is important
for deep states. For quantitative work, we need (particularly for intermediate
and high Z) the relativistic theory with the Breit interaction
WB (1, 2) =
e2
exp[ikr12 ][1 (1) (2)], k =

r12
c
(3.63)
where is Diracs velocity and h is the energy dierence between the scattering states (namely, the deep hole and Auger electron states). Codes are
now available; GRASP[158] (General-purpose Relativistic Atomic Structure
Program) computes Dirac-Fock orbitals, takes linear combinations of determinants with the correct J 2 .Jz , parity and seniority number labels; further
it takes linear combinations of such states, thus doing a partial conguration
mixing, and includes corrections like the eects of nuclear size and the main
QED corrections.
Problems
3.1. Find how the ground conguration of N is resolved into L-S terms.
3.2. Find how the ground conguration of Ti is resolved into L-S terms.
3.3. For the conguration d2 obtain |3 P ML = 0, MS = 0 as a combination
of determinants |m
L1 , mL2 | of one-electron wave functions, using the ClebshGordan coecients.

Note that using JMJ 1|J |JMJ = h
J(J + 1) MJ (MJ 1) one nds
for d states
L+ |2 = 0
L |2 =2|1
+
L |1 =2|2
L |1 = 6|0
+
L |0 =
6|1 L |0 = 6| 1
+
L | 1 = 6|0 L | 1 = 2| 2
L+ | 2 = 2| 1 L | 2 = 0
(3.64)
53
3.4. For the conguration d2 show how to obtain |3 P ML , MS for the other
values of ML , MS using the results of the previous problem, the shift operators
and (6.1.1). Write down all the 9 states.
3.5. Using the results of the previous problem for |3 P ML = 0, MS = 1 in
the conguration d2 verify explicitly that it is really 3 P with the specied
quantum numbers.
4 Greens Functions as Thought Experiments
4.1 Greens Theorem for one-Body Problems

Greens theorem

! 2
"

n dS
2 d3 x =
V
(4.1)
(where
n is the outgoing normal to the surface S bounding the volume V )
is obtained from the divergence theorem

div A d3 x =
A
n dS,
(4.2)
V
with A = . Consider the one-particle Schrodinger equation

1
r ) )(
r)=0
(4.3)
( 2
+ V (
2 r
dened in some volume V with some boundary conditions; it is often convenient to change it into an integral equation by a method which is familiar
from classical physics. One introduces a Greens function satisfying
1
( 2
r ) = (
r
r ),
(4.4)
r ) )G(
r ,
+ V (
2 r
multiplies (4.3) by G(
r ,
r ) , (4.4) by (
r ) and subtracts; the result is
(exchanging r with r )

1

2
r , r )2
r G(
d3
(
r
)
(
r
)
G(
r
,
r ) . (4.5)
(
r)=

r
r
2 V
The V integral can be changed to a surface integral by Equation (4.1).
4.2 How Many-Body Greens Functions Arise

4.2.1 Correlation Functions
In terms of the Schr
odinger picture, the Heisenberg representation (Section
2.2.1) corresponds to evolving the state from t0 to time t, applying the operator and then evolving back to t0 . This appears insane but it is not so.
56
Consider for instance a measurement of the light absorption from a system

of Hamiltonian H = H0 + V , where H0 is the kinetic energy and V is the
electron-electron interaction which is a large perturbation and makes the
problem dicult. However, the transitions are due to the time dependent,
further perturbation H eit driven by the oscillating eld. If H is weak we
can calculate the absorption cross section () from an initial state |i by the
Fermi golden rule, introducing a complete set of nal states {|f }. In obvious
notation,

2
P = f Pif = f 2
h
|f |H |i | (Ei + Ef )
2

= h
f i|H |f (Ei + Ef )f |H |i

(4.6)

= 2
f i|H (Ei + H)|f f |H |i
h

= 2
h
i|H (Ei + H)H |i ().
Transforming from frequency to time,

d it
(t) =
e
() = i|H ei(HEi )t H |i
2

= i|eiHt H eiHt H |i = i|HH
(t)HH
(0)|i .
(4.7)
This (t) is a correlation function and is written as an average over |i of a

product of Heisenberg operators. It may appear that there is little to gain in
this approach, since |i is an interacting many-body state that can be quite
hard or impossible to calculate. Here is the real breakthrough:
The calculation of many-body states can be completely
avoided! There are powerful methods for expanding the correlation functions, or better, their combinations which are called
the Greens functions.
In
the optical absorption case, the perturbation is a one-body operator,
H = mn Mmn am an and the cross section can be expressed in terms of the
matrix elements Mmn and of a correlation function
ph (t) = i|ap (t)aq (t)ar (0)as (0)|i .
(4.8)
ph (t) is a particle-hole correlation function and tells us about a thought

experiment: a particle-hole pair ( created in the interacting ground state by
the absorption of a photon) is propagated to time t and then destroyed, and
we want the amplitude that the system returns in the ground state. We shall
discuss photoemission spectra in terms of one-hole correlation functions. In
dealing with Auger spectroscopy we shall have to do in Section 6.2 with 2-hole
correlation functions like
hh (t) = i|ap (t)aq (t)ar (0)as (0)|i ;
(4.9)
we shall also need more complicated operator averages, e.g. in Sect. 6.4.1 .
57
4.2.2 Quantum Greens Functions

It is natural to start assuming for simplicity that the initial state |i = |0 is
the ground state of H. Temperature eects are postponed until Section 4.2.4.
Thus, the initial state averages are understood to be taken over the ground
state |0 , and summarizing we shall write:
0 , with H|0 = E0 |0 .
= 0 |O|
T=0 rule: O
(4.10)
Along with ph (t) = ap(t)aq (t)ar(0)as (0) and hh (t) = ap(t)aq(t)ar (0)as (0)
we shall need more complicated objects; however, simpler objects are also extremely useful. Let us start with the one-body ones. Such are the so called
lesser and greater Greens functions (ground state averages)
#
<
(t, t ) = i (t ) j (t)
(4.11)
gi,j
$
>
gi,j
(t, t ) = j (t)i (t )
(4.12)
One is obtained from the other by exchanging the operators, and in equilibrium g < informs us about the lled states, while g > knows about the empty
ones. From these, we can build retarded and advanced Greens functions; now
what matters is the order of times:
<
>
igri,j (t, t ) = (gi,j
(t, t ) + gi,j
(t, t ))(t t )
=< [j (t ),i (t)]+ >(t t )
while
igai,j (t, t ) = (gi,j (t, t ) + gi,j (t, t ))(t t).

<
(r)
>
(4.13)
(4.14)
(a)
The two are related by gi,j (t, t ) = gj,i (t , t). How do they depend on
|0 ? In no way! Averaging on the vacuum, g < = 0 and
$
#
r
i gi,j
(t, t ) = vac |cj (t)ci (t ) |vac (t t )
(4.15)
and since H acting on the vacuum yields 0, we obtain
$
#

r
i gi,j
(t, t ) = vac |cj eiH(tt ) ci |vac (t t ) ,
(4.16)
which shows that g (r) is actually a function of t t ; so we can drop t and

write simply
#
$
(r)
i gi,j (t) = vac |cj eiHt ci |vac (t) .
The Fourier transform yields:
%
&

(r)
eiHt+izt dt ci |vac
i gi,j (z) = vac |cj
0
%
&
i
= vac |cj
ci |vac .
z H + i
(4.17)
58
This
is a one-body
problem, that can be rewritten in rst quantization
#
$
i
j| zH+i |i . Averaging on a completely lled system, g > = 0 and creation and annihilation operators are interchanged, but the result is the same
one-hole amplitude. For a partially lled system, g < brings information on
the occupied states and g > on the empty ones, but the end result is again
i
a one-body matrix element of zH+i
between spin-orbitals irrespective of
(r) (a)
can always be computed with the
where the Fermi level is. Thus, g , g
average performed over the vacuum, and are a one-particle property. Both
are characterized by the fact that they do not know where is the Fermi level
(for non-interacting systems at least; in the presence of interactions they may
have some smell of a change of occupation numbers through the change in
potential it introduces).
g (r) , g (a) are not suited for perturbation theory; the diagrammatic method
(Chapter 11 ) works with the time-ordered ones
(T )
igk,k (t, t ) = T ak (t)ak (t )
(4.18)
where T is Wicks time ordering operator such that

(T )
igkk (t, t ) ig (T ) (kt, k t ) = (t t )ak (t)ak (t )|0

(t t)ak (t )ak (t) .
(4.19)
This discontinuity is needed; we shall see (Chapter 10) that it implies a source
term in their equation of motion. In eld theory every scattering event is
represented as the annihilation of the ingoing particle and the creation of
the outgoing one: here is where the source is necessary. In real space the
time-ordered Greens function is
(T )
g (xt, x t ) = iT [ (x, t) (x t )]
(4.20)
with the operators in the Heisenberg picture (the spin indices are often omitted when they are not needed).
4.2.3 Quantum Averages

The average of one-body densities f(x) = f (x) (x) (x) is

(T )
f = i lim+ lim
f (x)g (xt, x t ).
(4.21)

t t
x x
The number density of spin at

x at time t is given by
(
x , t) = i lim
lim
t t+0
x
x
the current operator is
(T )
(xt, x t );
g
(4.22)
[px (x x ) + (x x )px ]
j(x)
=
2m
59
(4.23)
and its expectation value is given by

1
(T )

lim
x , t) =
lim (
J (
x
x )g (xt, x t ).
2m t t+0
x
x
(4.24)

The average of one-body operators A = dxa (x) (x) (x) is

(T )
= i lim lim
A
a (x)g (xt, x t ),
(4.25)
dx

t t+ x x
upper sign for Bosons. This may be thought of as a trace over space and spin
variables:

= i lim lim T r a (x)g (T ) (xt, x t ) .
A
(4.26)

t t+ x x
4.2.4 Greens functions at Finite Temperature

We recall from Section (2.2.2) that
= T rA.
Finite T rule: A
(4.27)
(T )
The ground state average g (xt, x t ) of Equation (4.20) must be replaced

by a thermal average,
(T )
g (xt, x t ) =
i
T rT [ (x, t) (x , t )], Z = T r;
Z
(4.28)
this is the time-ordered nite temperature propagator . When g(T ) is used in

(4.26), all temperature eects are included. It is stated in some books that
g(T ) does not posses a diagrammatic expansion, but in fact it can be obtained
by the methods in Chapter 13 and corresponds to the contour of Figure 2.2
b).
For time-independent problems the temperature Greens function G (T )
dened along the vertical track of the contour of Figure 2.2 a) oers the
same information most directly:
G (T ) (x, x ) = T [ (x, ) (x , )]
= T r{T [ (x, ) (x , )]}
(4.29)
where eld operators are in the temperature Heisenberg representation (2.27),

T is the Wick operator
that orders along the track. The thermal average (4.27)

= dxTra(x) (x)(x) where spin indices are understood and
means A
traced over by Tr, and can be rewritten as
60

=
A
dx
lim
x x, +
Tr[a(x)G (T ) (x, x )].
(4.30)
To see that, one notes that,by the cyclic property of the trace,

dxa (x)Tr[eK eK (x)(x)];
Tr[aeK (x)(x)eK ] =

K
eK commutes with = e Z and
dxa (x)Tr[ (x)(x)] =

In this way the common value of and disappears from the calT rA.
culation of the observables and the behavior of G for far from has no
physical meaning. G (T ) (x, x ) is periodic:
G (T ) (x, x ) = G (T ) (x, x + ), >
G (T ) (x, x ) = G (T ) (x + , x ), <
(4.31)
(upper sign for bosons). Indeed, let > ; then Z G (T ) (x, x ) is given
by T r{eK (x, ) (x , )}; using the cyclic property of trace this may
be transformed to read
T r{ (x , )eK (x, )} = T r{eK e+K (x , )eK (x, )}.
Thus, we got (x , + ), standing on the left of , in agreement with the
fact that + > , but this implies a sign change for Fermions, a - sign
comes from the denition of G (T ) and the rst line results; the second follows
in a similar way.
G (T ) is useful only in time-independent problems; then it is actually
G (T ) (x, x , ). According to the above discussion it is a function of
one variable dened in (, ) which can be extended to the neighboring
intervals like a periodic (antiperiodic) function for Bose (Fermi) systems. For
all practical considerations, we may extend G (T ) in this way to the real axis,
since this has no physical implications but allows to write the Fourier series
G
(T )
1 (T )
(x, x , ) =
G (x, x , n )ein t ,
n=
(4.32)
with the Matsubara frequencies

n =
and
G (T ) (x, x , n ) =
1
2
n
,
d ein G (x, x , ).
(4.33)
(4.34)
For Fermi (Bose) system, G (T ) is odd (even) and G (T ) (x, x , n )

= 0 for
odd (even) n.
4.3 Non-interacting Propagators for Solids
61

Non-relativistic Electrons
Consider the non-interacting electron system with Hamiltonian H0 = 12 2
r+
V
( r ), such that H0 k (x) = k k (x); in second-quantized form, H0 =
= 1, the annihilation operator
k k nk ; in the Heisenberg picture, setting h
for spin-orbital a evolves with
ca (t) = eiH0 t ca eiH0 t .
From the equation of motion ca = i[H0 , ca ] , since [nk , ck ] = kk ck , one
obtains for energy eigenstates ck = ik ck and so
ck (t) = ck eik t .
(4.35)
iga,b (t2 , t1 ) iga,b (t2 t1 ) = T [ca (t2 )cb (t1 )] ,
(4.36)
The propagator is dened by
where the average is taken over the ground state. For an empty (totally
full) system, this reduces to the retarded (advanced) Greens function, but
otherwise ga,b propagates both electrons and holes; for t1 < t2 an electron
is added at t1 , propagates forwards in time and is annihilated at t2 , for
t1 > t
2 a hole is introduced, propagates backwards and is annihilated. Using
bn = ak < ak |bn > (Equation 1.51) one readily nds that
k
ga,b (t) =
a|k k|b gk,k (t)
(4.37)
where gk,k (t) = kk gk (t) and

igk (t) = eik t {(t)[1 nk ] (t)nk }
(4.38)
where nk is the occupation number. The propagator evolves with the same
phase factor as the wave functions but has a discontinuity at t = 0 and
satises the equation of motion
gk,k (t) = k gk,k (t) + k,k (t).

(4.39)
t
This useful result is due to the denition of T . The Fourier component at
frequency is found by the frequently used integrals
i

0
ei(+x)t dt =
i
+x+i ,
0
ei(+x)t dt =
i
+xi ,
+0,
(4.40)
62

0
G (k, k ,) (k, k ) G (k,)=
dtG0 (k, k , t) eit =
k,k
,
k + ik
(4.41)
with k = + for empty states and k = for lled ones, and stands
for a positive innitesimal. Thus one has an electron for k > 0 and a hole
otherwise. In the space representation the retarded Greens function (4.17)
reads
k (
1
r )k (
r )
=
r|
|
r
r ,
r , ) =
(4.42)
g (r) (
k + i
H + i
k
It obeys the same equation of motion as the time-ordered one, the dierence
arises from the boundary conditions. Indeed, setting z = + i, using
1
(z H)|
r
zH
r
r ) = z
r|
=
r |
r = (

r|
1
1
|
r
H|
r (4.43)
r|
zH
zH
and exchanging
r with
r one nds
1
r ))g (r) (
r ,
r , ) = (
r
r )
( ( 2
+ V (
2 r
(4.44)
which is like (4.39) in another representation. Comparing with (4.4) we see

that g (r) = G.
For the temperature Greens function (4.29) we nd similar results.
G(k, ) = TrT ck ( )ck ( ).
(4.45)
Using ck (t) = eik t with the substitution it and introducing the chemical potential one gets: ck ( ) = e(k ) ck , ck ( ) = e(k ) ck and so
G(k, ) = e(k )( ) {( )Trck ck ( )Trck ck }. (4.46)

Now using (2.40) and setting = 0 for short we obtain
G(k, ) = e(k ) {( )(1 n(k)) ( )n(k)},
(4.47)
where n(k) is the Fermi distribution (2.40). This is similar to (4.38); the
changes are: i becomes -1 (a mere convention), energies are referenced to the
chemical potential, it at the exponent, the energy step functions are
smoothed according to the Fermi distribution and the Wick T now acts on
the vertical track. Next, we nd the thermal Greens function in frequency
space by (4.34) and (2.40)
G(k, n ) =
1
,
in + k
and in this case the only change (apart from using ) is in .
(4.48)
63
Non-relativistic Bosons
Consider a Bose eld which obeys the harmonic wave equation
2
k (t) = k2 k (t) .
t2
(4.49)
2

2
2 2
2 2
c (r, t) = 0.
[ 2 +c k ]k (t) = 0
t
t2
(4.50)
If k = ck, this is just
In second quantization is dealt with like the coordinate of an oscillator of

(arbitrary) mass m=1, that is
k (t) = h
ak (t) + ak (t)
ak exp [ik t] + ak exp [ik t]
=h
2k
2k
(4.51)
The canonically conjugated momentum is

k = k
(4.52)
and yields the Hamiltonian

H=
2 + 2 2
k
k k

k

1
hk ak ak +
.
2
(4.53)
The propagator is dened as:

Dkk (t) = 0|P [k (t) k (0)]| 0
(4.54)
where ak |0 = 0, P is the time ordering operator (earlier times to the right

but no sign changes on permuting operators like in T .) Now I show that for
t
= 0 #the D and$ have same wave equation, but at t = 0 the source acts.
Since 0|ak ak |0 = kk , obviously Dkk (t) = kk Dk (t) , where
Dk (t) =
#
$
#
$
(t) 0|ak exp [ik t] ak |0 + (t) 0|ak ak exp [ik t]|0
2k
. (4.55)
Therefore,
Dk (t) =
1
exp[ik | t | ]
{(t)exp[ik t]+(t)exp[ik t]} =
.
2k
2k
(4.56)
This is continuous, however
1
t Dk (t) = 2k {ik (t)exp[ik t] +
= i
2 {(t)exp[ik t](t)exp[ik t]}
ik (t)exp[ik t]}
(4.57)
64
is not, and
ik t
i t
2
i
ik (t)e k } 2i
t2 Dk (t) = 2 { (ik ) (t)e
ik t
i t
k
=
+ (t)e k } i (t)
2 {(t)e
{ (t) + (t)}
that is,
2
Dk (t) +k2 Dk (t) = i (t)
(4.58)
t2
Therefore,the propagator is Greens function of the wave equation. Fourier
transforming, one nds
[ 2 +k2 ]Dk () = i.
(4.59)
However the solution is problematic, because of poles on the real axis, and
the transform of (4.56)

0
1
dt ei(k )t + dt ei(+k )t
dt Dk (t) ei t =
Dk () =
2k
does not converge. Therefore one must introduce convergence factors

0
1
Dk () =
dt ei(k +i )t + dt ei(+k i )t
2k
0

1
1
1
+
=
2k i ( k + i) i ( + k i)

1
1
i
+
(4.60)
=
2k ( k + i) ( + k i)
and nally
Dkk () =
ikk
.
2 +2k i
(4.61)
For phonons and other non-relativistic bosons commonly one denes, (understanding the vacuum average)
#
$
k (t) = i P k (t) k (0)
D
(4.62)
in terms of
k (t) = ak exp [ik t] + ak exp [ik t]
without the normalization factor
1
2k
(4.63)
in (4.51); then
iDk (t) = exp[ik | t | ].
(4.64)
65
Relativistic Bosons, Photons

The Klein-Gordon equation
2

mc 2
2
)
+
(
=0
c2 t2
h
(4.65)
describes a spinless relativistic paricle. Introducing the relativistic dispersion

2
(4.66)
k = (chk) + m2 c4
one obtains the propagator
Dk () =
i
2 + (chk)2
+ m2 c4 i
(4.67)
In a 4-dimensional notation with
p =
h(k, i )
c
D (p) =
(4.68)
i
2
(cp) + m2 c4 i
(4.69)
For a massless scalar particle (if it existed) one would write

D (k, ) = D (p) =
p2
i
.
i
(4.70)
For a photon, the propagator between two points x, x in a 4-dimensional

notation is naturally given by
D, (x, x ) = P [A (x)A (x )] ,
(4.71)
whereA = (A, i) are the four-vector potential components; however a gauge

transformation A (x) A + (x)
x with arbitrary (x) will change the
potential and the propagator. In the Lorentz gauge such that
A = A +
1
= 0,
c t
(4.72)
the wave equation is

4
J .
c
thus for the free eld the wave equation is

1 2
2
2 2 A = 0.
c t
A =
which suggests writing
(4.73)
(4.74)
66
D, (x, x ) = i,
d4 p 4 eip(xx )
)
;
(2 p2 i0
(4.75)
this is evidently a Greens function of the wave equation. Actually the Greens
function is not unique and the gauge invariance allows to add to the r.h.s. of
2
(4.75) xx f (x x ), where f is an arbitrary function. The gauge in which
(4.75) holds as it stands is called the Feynman gauge.
4.3.1 Greens Functions for Tight-binding Hamiltonians
The tight-binding model Hamiltonian

ci cj , th > 0
H = th
(4.76)
<i,j>
is dened in terms of a d-dimensional graph or lattice; i and j stand for two

sets
of d coordinates for two sites or nodes of the lattice and the notation
means that the sum is over all i and overall j that are nearest neighbors
<i,j>
of i. Here I consider regular linear, square and cubic lattices. Using continuum normalization, and setting the lattice parameter to 1, the Bloch energy
eigenfunctions are
1 iki
|k =
e |i .
(4.77)
2 i
In d dimensions, the energy egenvalues are:
k = 2th
d
cos k .
(4.78)
The local density of states at the site at the origin is
d () =

k
|0 | k | ( k ) =
1
(2)d

d
d k
BZ
2th
d
,
cos (k ) .
(4.79)
where the integral extends to the Brillouin Zone. Converting the ( k )function into a sum of momentum functions, a factor |k k | appears in
the
To
better understand the physical meaning, dierentiate
denominator.
(
p + k )2 + V (
x ) u
x ) = k u
x ). (Equation 8.23:)
(
(
2m
k
k

(
p + k )2
(
p + k )2
+
+ V ( x ) (k )u
u
x)
( x )(k )
(
k
k
2m
2m
x ) + k k u
x ). (4.80)
= (k )k u
(
(
k
k
67
Multiplying by u
| and recalling (8.22) one obtains
k
p
1

x )| |
x ) = k k .
(
(
k ,
m k ,
h
(4.81)
1 Dimension
The band edges = 2th are the extrema of k = 2th cos k at k = and
k = 0 and = 0 if || > 2th . The argument of the in (4.79) vanishes for
);
k = k = arccos( 2h
1 () =

dk
0
1
(k k )
=
.
2th sin(k)
2th sin k
This can be rewritten

1 () =
(4t2h 2 )

,
4t2h 2
(4.82)
(4.83)
which in fact when integrated over all yields 1. The band-edge divergence
1 ()
0.1
0.05
-2
Fig. 4.1. Density of states on the 1d tight-binding lattice. Note the Van Hove
singularities at the band edges .
is characteristic: it reects the fact that at the band edges the group veloc
vanishes; this, combined with the one-dimensionality, leads to
ity vg = k
the inverse-square-root singularity. The symmetry around the band centre is
typical of bipartite graphs. These are lattices with the property that all
sites can be painted red or blue in such a way that any red (blue) site has
only blue (red) rst neighbors; in the linear chain these are odd and evennumbered sites. Changing the sign to all the blue orbitals is just a gauge
transformation and cannot change any physical quantity, yet it is equivalent
to sending the o-diagonal one-electron matrix element th to th . However,
k is proportional to th and must change sign as well. This can happen in
68
just one way: the spectrum is symmetric and the eigenfunctions at k and
k get exchanged by the gauge transformation.
The o-diagonal elements gm,n of the resolvent may be found by taking
1
1
= 1+ H
H between site 0 and n
= 0.
matrix elements of the identity H
One nds
g0,n = g0,n+1 + g0,n1

(4.84)
th
which suggests the solution
g0,n = g0,0 q |n|
with
q + 1 = 0.
th
Some care is needed to choose between the roots
q2
q =
where
2th
2
) 1
2th
(4.85)
(4.86)
(4.87)

( 2th )2 1 is imaginary for in the spectrum between 2th and 2th .
One has to consider the analytic continuation

from real to complex z,with

a branch cut on the real axis ; taking z = |z| exp[ 2i arg(z)] with the cut
along the negative Re(z) axis,

i |( )2 1| above the axis
z 2
2th
(
) 1=
i |( )2 1|
2th
below
2th
In the continuum |q | = 1, but q is discontinuous across the cut. With >

2th , we must prefer q which is < 1 and implies an exponential attenuation
of the amplitude g with distance whereas q+ > 1 would imply that the
amplitude
grows exponentially.

( 2th ) 1 > 0, > 2th becomes negative for < 2th and so |q | 1
everywhere. Therefore, everywhere,
g0,n () = g0,0 ()q ()|n| .
(4.88)
2 Dimensions
Using Equations (4.79,4.82), one nds
2 () =
1
(2)2
dkx
dky ( 2th [cos (kx ) + cos (ky )])
that is
1
2 () =
2
(4.89)

dk1 ( 2th cos (k)).
(4.90)
69
This integral can be written in terms of the complete elliptic integral [44]
of the rst kind. The band now extends from 4th to 4th , and is symmetric
(the graph is bipartite). Singularities become milder when integrated over,
however they are still evident (see Figure ??). The worst is the logarithmic
1
for k 0.
Van Hove singularity ( at = 0 the integrand of (4.90) goes like |k|
In other terms, this is just the singularity of 1 in integrated form, and is a
universal feature of d=2 lattices.) In addition, 2 jumps discontinuously to
2 ()
0.1
0.05
-4
-2
Fig. 4.2. Density of states on the 2d tight-binding square lattice. Note the Van
Hove singularities: jumps at the band edges and a diverging cusp a the centre.
0 at the band edges, which represents a milder singularity (in general, this
term involves a point where the function is not analytic, not necessarily a
divergence). This can again be read o (4.90): 2 = 0, for > 4, but setting
there is an
= (4 )th the band-edge singularity of 1 enters again since
interval near k = 0 where cos k > 1 /2; this interval is of order and the
integrand there is of order 1 , so the result is nonzero up to the edge.
3 ()
0.1
0.05
-6
-4
-2
Fig. 4.3. Density of states on the tight-binding cubic lattice.
70
3 Dimensions
The simple cubic lattice is bipartite, hence the density of states is again
symmetric around = 0. We can nd it by calculating
1
3 () =
2

dk2 ( 2th cos (k)).
(4.91)
From this integral representation and from the behavior of 2 it is easy to

show the nature of the singularities . The band-edge singularities at 6th
are of the square-root type, as for free particles. In addition,there are two
3
points where d
d is discontinuous. More details may be found in the book by
Economou[44].
4.3.2 Lippmann-Schwinger Equation
When the Hamiltonian can be decomposed in a free term and a perturbation,
H = H0 + H1 , the perturbed resolvent operator function can be written in
terms of the unperturbed one via the often useful identity
1
1
1
1
=
.
+
H1
zH
z H0
z H0
zH
(4.92)
In one-body problems this translates directly into Greens functions. The

usual way of treating impurity problems is via the Lippmann-Schwinger equation,

G( r , r ) = G0 ( r , r ) d3
r G0 (
r ,
r )V (
r )G(
r ,
r ) (4.93)
where V is the perturbing potential, which is nothing but (4.92) and a special
case of the Dyson equation (see Section 11.4 below).
4.3.3 t matrix
Another form of (4.92)
1
1
1
1
=
H1
+
zH
z H0
zH
z H0
(4.94)
is also useful; substituting in the r.h.s. of (4.92) one obtains

1
1
1
1
=
+
t
,
zH
z H0
z H0 z H0
(4.95)
where the t matrix obeys

t = H1 + H1
1
t.
z H0
This accounts at once for all the repeated scattering from H1 .
(4.96)
71
4.3.4 Ingleseld Embedding Method

The Lippmann-Schwinger equation is less convenient when the perturbation
is extended. Typical examples are surface problems, when one wants to treat
the eects of surface creation, reconstruction, or contamination. In such cases
one can resort to slab models, but there are drawbacks in any attempt to
represent a bulk by a few atomic layers, with quantized normal momenta.
The only practical alternative is the method of embedding. By this term one
understands that 1) in an extended system, there is a more or less localized
perturbing potential. A surface S (see Figure 4.4 left) divides the perturbed
region I from the far region II where the potential is negligible. Over S the
potential is also negligible 2) solving the problem in I is easier than in the full
system; the region I might be nite, or, in the case of surfaces, it is the nite
thickness that helps 3) the extended unperturbed system could be treated
easily because it is highly symmetric 4) then one wants to solve the problem
in I in such the way that the wave functions match those of the extended
system on S.
dS
II
Fig. 4.4. Left:Embedding of region I into II; Right: a pill-box element V between
S and its slightly inated, dotted version.
Following J.E. Ingleseld [114] we let the wave function
(
r)
r I
( r )
r II
( r ) =

(
r ) = (
r)
r S;
(4.97)
must make the functional

E=
|H|
|
(4.98)
stationary, with
1
H = 2 + V (r).
(4.99)
2
We write in II as a functional of and reformulate the problem such that
the latter is the only unknown. Thus, the solution in I yields the solution
everywhere. Let us see how this is done in practice.
72

|H| =
d r (
r )H(
r)+
3
d r (
r )(
r)+
3
II

dS
(4.100)
The surface element dS contributes

1
dS =
V H =
2 V
2
V
V
where the integral is over a pill-box of volume V around dS (Fig. 4.4 right)
and we kept only the kinetic contribution since the potential one goes to 0
as 0; since

2 =
2 = (
n outer dS +
n inner dS),
V
we are left with
1
dS =
2
S

(
r ) ( )
n dS.
(4.101)
We wish to write everything in terms of , so we must eliminate from

(4.101) and from the normalization condition. The latter is

3

d3
r )(
r ).
(4.102)
r (
| = d r ( r )( r ) +
I
II
To rewrite the normalization integral over II in terms of only, we vary ,

multiply the varied Schr
odinger equation H = + by and the
complex conjugate Schr
odinger equation by :

H = +
H = .
Dividing by and subtracting, one gets:
1 2
1
( 2 )
( ) = ||2 .
2

2
Next, integrating over II and using Greens theorem one nds

n .dS
||2 d3 r =
2

II
S
(4.103)
Here a - sign comes from the outgoing normal from I which is the opposite
of the outgoing normal from II. To simplify matters, we vary in such a way
that over S; then
0 in S and we dispose the rst term on the

r.h.s. The presence in (4.101,4.103) of

n is the only diculty left, that
now we proceed to remove. This requires a clever trick: in order to get the
gradient on the surface in terms of , we rst nd in II and on the surface
in terms of the gradient, and then invert the relation.
4.4 Kubo Formulae
73
Since the perturbation is localized in I, we know already from Equations

( 4.5 ,4.1) how to solve H = for in II and for an eigenvalue if is
assigned on S. This can be achieved using Greens functions, and Ingleseld
uses the one satisfying
1
r ))g(
r ,
( ( 2
r , ) = (
r
r )
(4.104)
+ V (
2 r
which is the negative of the retarded Greens function satisfying (4.44).Therefore,
he nds for
r II

g(
r , r )
(
r)=
r ( r )
2 S

2

(
r )
(4.105)
r g( r , r ) n d r
and in II is a functional of the boundary values of and . Now we are

in position to write down n as another functional of the surface values

of . In (4.105) we choose g having zero derivative on S, which eliminates the
second term, obtaining

1

( r ) =
g(
r ,
r ) (
r )
n dS
(4.106)
2 S
then put
r on S and invert the relation to read

n = 2 dS g 1 (
r ,
r )(
r ),
r ,
rS
(4.107)
(
r )
S
with g 1 a matrix inverse.

At this point is the only unknown; the numerator and the denominator
of the energy functional are written as functionals of and the problem can
be solved in region I. From the variational conditions Ingleseld obtained an
eective Schr
odinger equation for . One nds (Problem 4.5) that the particle
sees the eective potential
Vef f (r) = V (r) + U (r)();
(4.108)
besides the potential V (r) this comprises an embedding surface potential

U (r)() where is a curvilinear coordinate such that = 0 on S.
The method was intended rst for independent-electron problems but
lends itself to iterative self-consistent approaches like density functional theory . The extension of the method to the Dirac equation has been worked
out recently by Crampin [115].
4.4 Kubo Formulae

Particle-hole Greens functions also arise through the Kubo formulae[29] of
linear response theory. An interacting system with Hamiltonian HS is probed
with a weak time-dependent perturbation
74
(t),
H (t) = AF
(4.109)
where F (t) is an adiabatically switched time dependence. The total hamiltonian is

(4.110)
H = HS + H (t).
of an observable B;
by deWe seek the linear response B(t) = T r(t)B
nition, B is a dierence from equilibrium and vanishes for A = 0 ( that is,

= 0). Similar to the familiar relation P = E of electric eld to
T req B
polarization, the linear response equation is
t
B(t) =
dt BA (t t )F (t );
(4.111)
the role of the polarizability is played by the response function BA (t). We

need the density matrix (t), which deviates a little bit from the equilibrium
one eq . The perturbed (t) obeys the Heisenberg equation of motion
i
h
d
(t) = [H(t), (t)] .
dt
(4.112)
Setting (t) eq + (t), one nds a linearized equation

i
h
d
eq ].
(t) = [HS , (t)] + F (t)[A,
dt
(4.113)
The solution is

i
h(t) =
e h
HS (tt )
eq ]e h HS (tt ) F (t )dt

[A,
i
as one can readily check by calculating the derivative. Hence

t
i

1
(t )dt .
eq ]e hi HS (tt ) BF
Tr
e h HS (tt ) [A,
B(t) =
i
h
(4.114)
(4.115)
Now we move the left exponential to the right (by the cyclic property of the
t ) in the Heisenberg picture (with HS ):
trace) thereby obtaining B(t
t
1
eq ]B(t
t )F (t )dt
Tr
B(t) =
[A,
(4.116)
i
h
Now for the linear response function BA (t) of (4.111) we obtain

eq AB(t)};
i
hBA (t) = T r{Aeq B(t)
(4.117)
using again the cyclic property, we can change this into a more elegant equilibrium thermal average, namely the Kubo formula[29]
4.5 Vacuum Amplitudes
A eq AB(t)}
i
hBA (t) = T r{eq B(t)
= [B(t),
A(0)] .
75
(4.118)
By the identity (Problem 4.1)

eH ] = eH
[A,

0
h )
dA(i
d,
dt
(4.119)
and Equation (4.114),
eHS ] B(t)},
i
hBA (t) = T r{[A(0),
one nds the further Kubo formula

dA(i
h )
BA (t) =
B(t) d.

dt
0
(4.120)
This formalism has many applications; for example, if the system is exposed
to an electric eld, A is proportional to the position vector components of the
is the curelectron and the time derivative is proportional to the current, B
rent and the conductivity tensor. For the frequency-dependent conductivity,
H and using the cyclic property again, and
starting from A(i

) = eH Ae
the fact that it is retarded, one can easily show that, up to a factor,

i,j (, T ) =
dt
d ji (0)jj (t + ih ) eit .
(4.121)
0
4.5 Vacuum Amplitudes

In stationary problems one often uses a zero-body propagator or vacuum
amplitude
(4.122)
R (t) = UI (t) 0 ,
where the thermal average of the interacting picture evolution operator
UI (t) = eiH0 t eiHt is taken with the non-interacting distribution function
1
H0
. At zero temperature this becomes
Z0 Tre
R (t) = | UI (t) | = eiW0 t | eiHt | ,
(4.123)
where W0 is the ground state energy of H0 . Note that unlike all the other
Greens functions, here the average is taken on the non-interacting ground
state. R is related to the interacting ground state energy E0 through the
following device. Inserting a complete set of H eigenstates,

2
| |n | ei(En W0 )t ,
(4.124)
R (t) =
n
76
we can isolate the ground state energy from (4.124) by giving the time t an
imaginary part: t t i, > 0. In this way, the exponent has a real part
t (W0 En )
which is largest for n = 0. In the long run, the ground state dominates:
R (t) = | |0 |2 ei(E0 W0 )t ,
t (1 + i) .
There is a restriction: we need

|0
= 0;
therefore V should not change the ground state symmetry . Special care is
needed when is degenerate. When the overlap does not vanish,
R (t) = i (E0 W0 ) | |0 |2 ei(E0 W0 )t ,
and we get the closed formula
E0 = W0 + i
d
log(R(t)).
t(1i) dt
lim
(4.125)
Since UI admits the T exp expansion (2.37), the methods of Chapter 11 will
provide a practical way to compute R.
4.6 Lehmann Representation

In non-interacting models the poles of the Greens functions close to the real
axis correspond to eigenstates of the Hamiltonian where a particle can exist
or where one can put a particle. Here we see how to generalize this notion to
the interacting case. In this section we let h = 1.
4.6.1 Zero-Temperature Fermi Case
Denote the complete set of many-body eigenstates of H by |M, n , where the
integer M = 0, 1, 2, runs over the electron numbers and n = 0, 1, 2, runs
over the M-body eigenstates, such that
|M, n = M |M, n .
H|M, n = EM,n |M, n , N
Let us write the time-ordered Greens function as
G(N ; xt, x t ) = iN, 0|T [(x, t) (x t )]|N, 0 ,
(4.126)
emphasizing that it is the average over the N-body interacting ground state
of the Hamiltonian H and that the operators are in the Heisenberg picture. For time-independent H, setting h
= 1, (x, t) = eiHt (x)eiHt , and
77
G(xt, x t ) G(x, x , t t ); inserting the complete set we obtain the very

useful Lehmann representation: for t > t ,

ei(EN +1,n EN,0 )(tt )
iG(x, x , t t ) =
n
N, 0|(x)|N + 1, n N + 1, n| (x )|N, 0 ,
(4.127)
and the eigenstates with one more particle come into play; for t < t ,

iG(x, x , t t ) =
ei(EN 1,n EN,0 )(tt )
n
N, 0| (x)|N 1, n N 1, n|(x )|N, 0 ,
(4.128)
the N-1-particle states appear. By the standard integrals (4.40) we obtain

G(x, x , ) =
+
N, 0|(x)|N + 1, n N + 1, n| (x )|N, 0

(EN +1,n EN,0 ) + i
n
N, 0| (x)|N 1, n N 1, n|(x )|N, 0

. (4.129)
(EN,0 EN 1,n ) i
n
The poles of G provide information on the excitation spectrum, namely, on

the ionization potentials and electron anities, or in other terms, on quasielectron and quasi-hole excitations. Now we simplify the notation dropping
N and N 1. The structure of (4.129) invites us to write

(x, x , )

d
(4.130)
G(x, x , ) =
+ i
where w is an innitesimal quantity, with
(x, x , ) =
+
0|(x)|n n| (x )|0 ( (EN +1,n EN,0 ))
0| (x)|n n|(x )|0 ( (EN 1,n EN,0 )).
(4.131)
The positive spectral function (x, x, ) generalizes the density of states

(5.16): In normal systems the ionization potentials cannot be smaller than
the electron anities and innitesimal w = sgn( ) where =
M inn,n [E(N + 1, n) E(N, n )] is the chemical potential. So,
1
() = ImG()sign( ).
(4.132)
The real and imaginary parts of Greens functions are Hilbert transform pairs.
78
4.6.2 Finite Temperatures, Fermi and Bose

Let us extend the above to nite T , writings = 1 for bosons, s = 1 for
fermions. For the retarded Greens function,
g (r) (x, x , t t ) = iT r{[(x, t), (x , t )]s },
(4.133)
one nds, with K = H N and h

nm = En Em ,
g (r) (x, x , t t ) =

i Kn
(e
n|(x)|m m| (x )|n eimn (tt )
Z mn
+s eKn n| (x )|m m|(x)|n einm (t t) ) (t t ).

Now,let
Rmn (x, x ) = n|(x)|m m| (x )|n ;
(4.134)
(4.135)
exchanging m with n in the last line and Fourier transforming one readily
arrives at
g (r) (x, x , ) =
1 Rmn (x, x ) Kn
(e
+ seKm );
Z mn + mn + i
(4.136)
with a positive innitesimal; in the same way the advanced function is

g (a) (x, x , ) =
1 Rmn (x, x ) Kn
(e
+ seKm ).
Z mn + mn i
(4.137)
The same manipulations on the time-ordered Greens function yield

G(x, x , ) =
s eKm
1
eKn
).(4.138)
Rmn (x, x )(
Z mn
+ mn + i + mn i
Equations (4.130,4.131) can be extended to nite T, introducing the Tdependent spectral function
(x, x , ) =
2 Kn
e
Rmn (x, x )( + mn )(1 + se ) (4.139)
Z m,n
that allows to write down the retarded and advanced functions

d (x, x , )
g (r) (x, x , ) =
,
2 + i

d (x, x , )
(a)

g (x, x , ) =
;
2 i
the time-ordered one (4.138) may be written
(4.140)
(4.141)
G(x, x , ) =
d ns ( )(x, x , )
[
2 ( + i))
(x, x , )
ns ( )
s
]
( i)
79
(4.142)
with n
() 1 ns () and ns () = (1 + se )1 The knowledge of (also
a function of T ) determines all the one-body Greens functions. For independent electrons, g (r) and (4.139) are purely one-body properties independent
of the Fermi level; therefore the temperature dependence of (4.139) is neglected usually in Density Functional and other self-consistent calculations.
This simplication may be wrong when dealing with strongly correlated systems.
4.6.3 Fluctuation-Dissipation Theorem
The above results imply a number of relationships involving and the real
and imaginary parts of the various Greens functions. One shows (Problem
4.2)the following. Assuming that Rmn (x, x ) is real (it must be positive for
x = x and it can be taken real anyhow in the absence of magnetic elds)
show that

s
d ImG(x, x , )

ReG(x, x , ) =
(tanh
) , s = . (4.143)

2
Historically such relations are known as Fluctuation-dissipation theorem.
Problems
4.1. Prove the identity
eH ] = eH
[A,

0
e H [H, A] e H d,
(4.144)
4.2. Show that (4.143) holds.

4.3. In the continuous case of equation (4.44) determine g(x , x, z) in 1 dimension
4.4. In the continuous case of equation (4.44) determine g(x , x, z) in 3 dimensions
4.5. Derive the enbedding potential U (r) of Equation (4.108).
5 Hopping Electron Models: an Appetizer
5.1 Fano Resonances and Resolvents

5.1.1 Resonances
Often we treat excited levels as discrete, as they appear in low resolution
or in some simple approximation. Sooner or later, however a closer analysis
always shows that when all the degrees of freedom are considered they are
in a continuum. Thus, the 2p level of H, H(2p), is coupled to a continuum of
H(1s) + 1 photon, and thereby gains a width and a structure. The 2p state
of H is discrete only if you accept to neglect its interaction with a continuum
of photon modes that eventually take the H atom to the ground state while
producing photons. In 1952, Fermi discovered a peak in the pion-proton ( +
p) elastic cross section for center-of-mass kinetic energies 1.2 to 1.4 GeV; since
the half width at half maximum ( 100 MeV) implies a lifetime 1023 s,
which is very short, the strong interaction was implied in the formation and
also in the decay. This was christened the doubly charged ++ resonance.
Such high-energy Physics contents are pertinent to this book: hundreds of
resonances are familiar by now to particle physicists, but Fermis concept of
resonances is important in quantum problems at all energy scales. Firing 500
eV electrons on Helium and measuring the loss spectrum, one observes[31]
an asymmetric resonance in the ionization continuum al 60 eV which has
been identied as the 2s2p1 P state of He. Many more are known by now, and
they are all due to the neutral, twice excited He atoms in auto-ionizing states.
Using soft electrons (with not enough energy to produce ionizations) one sees
narrow asymmetric resonances in the elastic cross section, due to temporary
He ions. In all cases, the transition probability is described by some operator
T , like T Ap for electromagnetic transitions or, for fast electron scattering,
2
iqrj
T 4e
, where rj suns over the target electron positions. However
je
q2
thre are features that do not depend on T ; for instance, the asymmetric He
resonances are seen also in optical absorption.
We now develop techniques for dealing with such problems.
82
5.1.2 Fano Model

Here we consider a system consisting of very unlike parts: let |k stand for a
continuum of states of energy k and |0 , 0 represent a localized state and its
energy. Typically, |0 might represent atomic spin-orbital and |k a freeparticle states or Bloch states in a solid state problem from which an electron
can jump into the continuum of single-particle states |k (free-particle states
or Bloch states in a solid state problem.) Some perturbation produces hopping
matrix elements Vk0 coupling the continuum to the discrete state. The Fano
model is

H = H0 + Hh , H0 =
k, nk, +
0, n0,
(5.1)
kC,
while
Hh =
{Vk ak a0 + h.c.}
(5.2)
k,
The continuous spectrum of H0 corresponds to a set C of the real k, axis;

C can be bound or unbound, and also it can be compact or consist of several
pieces; for simplicity I assume that the continuum is not degenerate (only one
state corresponds to k ). This assumption can be removed later by a direct
extension of the present treatment[30].
There are no interaction terms (involving 2 creation and 2 annihilation
operators)in the Fano model. In rst quantization, it would be

k, |k k, | +
0, |0 0, | +
{Vk |k 0| + h.c.} (5.3)
h=
kC,
k,
and would represent 1 electron instead of representing any number of noninteracting elements. The two problems are closely related, and, for pedagogical reasons, I rst present the elementary one-body term and then the
many-body equation of motion approach (this is actually a one-body problem
but the equation of motion method lends itself to the many-body case.)
5.1.3 One-body Treatment
In the elementary approach one writes:

H |0 = 0 |0 + V0k |k
k
H |k = k |k + Vk0 |0 .
The H eigenstates are expandable on the old basis:

| = |0 0 | +
|k k | .
k
(5.4)
(5.5)
83
Taking the scalar product with | one obtains:

|k V0k = 0
( 0 )|0
(5.6)
( k )|k |0 Vk0 = 0.
(5.7)
We discuss the solution below.

5.1.4 Many-Body Treatment
In the many-body case, we must still diagonalize the Hamiltonian, that is,
write it in the form
=
H
n ;
(5.8)
the basis change (5.5) entails new creation operators for states such that

a = a0 0| + k ak k|
a = a0 |0 + k ak |k
with the inverse
(5.9)
ak = < k|> a
a0 = < 0|> a
We can nd the states by the equation of motion method. Note that

[a , n ]/
=(, )[a
, n0 ] = (, )a a a =
(, )
1 a a a =(, )a ,
here is Kroneker if we quantize in a big box, or Diracs if we normalize

on a continuum. In the diagonal basis,
= a .
[a , H]
This corresponds to the equation of motion
i

da
= a , H
= a
dt
a (t) = a (0) ei t
(5.10)
Written in the old basis, using (5.9), this is

[<|0 > a0 +
<|k > ak , H] =

(<|0 > a0 + <|k > ak )
k
(5.11)
84
Now,

a 0 , a0 a k = a 0 a 0 a k a 0 a k a 0 = a 0 a 0 a k + a 0 a 0 a k = a k
and the required commutators read:

[a0 , H] = 0 a0 +
Vk0 ak ,
[ak , H] = k ak + V0k a0 .
(5.12)
The equality holds separately for the coecients of a0 and of each ak ,

since these are all linearly independent states. So, we nd (5.6,5.7) again.
Thus, the many-body formulation yields the same system as the one-body
one, as it should always happen in non-interacting problems. The system
allows to obtain the spin-orbitals and thus build the many-electron Slater
determinants.
Now the discrete and continuous solution methods of Equations (5.6,5.7)
have quite dierent characters.
1-Discrete Method of Solution
Placing the system in a box, the continuous spectrum becomes discrete; one
can assume l
= k and write
|k =
Vk0
|0 ;
k
(5.13)
substituting into the rst one gets an algebraic equation for the discrete
eigenvalues
0
|Vk0 |2
= 0.
k
(5.14)
Note that here only real quantities appear.

|Vk0 |2
The Figure shows a graphical solution. y =
k k is plotted versus 0 . For illustration, I have taken k-independent V0k coecients and
evenly spaced, positive k values. The roots are the values where the sum
crosses y = 0 . Between two unperturbed k eigenvalues there is always one ; in addition, there is one root below the discretized continuum. Once the roots
are found, one normalizes the wave functions by writing
11 Vk0 112
2
||0 | {1 + k 1 k 1 } = 1 and the problem is completely solved. This solution is of little use, however, if we wish to understand what happens in the
continuum limit. By increasing the number of k, the degree of the equation
grows; a smaller and smaller interval separates a 0 from an of the sum, and
more signicant digits must be included to approximate the roots. This large
amount of information is not about the system, but its interaction with the
85
ctitious box. This is not the way to the continuum. A brute force approach
fails because of the irrelevant complications it produces.
0 -
|Vk0 |2
k k

0
Fig. 5.1. Graphical solution of Equation (5.25). The true continuum C, before
placing the system in the box, extends from = 0 to = 6; a dot marks a root
below it. Here the continuum is replaced by equally spaced roots.
2- Solution in the Continuum

At this stage the discrete and continuous methods of quantization become
dierent, with the latter which is much more suitable. Now the k summations must be read as integrals; now perturbed and unperturbed eigenvalues
must coincide, and we want tools to handle the problem, starting with some
denitions.
The density of states of the system is

() =
( ) .
(5.15)
E
Hence, E12 d() is the number of states in the interval E1 < < E2 ;
it may diverge in the continuum limit, when N ,
but we can convert
3
the sum to an integral with a suitable measure, like k (2)

d k,
3
where is the volume of the system. At the end all physical quantities
have a nite value when .
86
The local (or projected) density of states (LDOS) measures the degree of
mixing of the eigenstates at energy with the 0 orbital,

( )|0| |2 .
(5.16)
0 () =
(0)
0 ()
= ( 0 ). The V coupling changes

The unperturbed quantity is
the discrete delta-like level into a virtual level;
this has a width, and can
have structure. The set is complete, hence d0 () = 1.
The resolvent, or Green operator

G() = ( H + i)1 ,
(5.17)
where stands for a positive innitesimal, or if you like = +0 (intelligenti pauca). This is the Fourier transform of causal1 operator G(t) =
eiHt (t), and is analytical in the upper half plane.
Note that
G00 () 0|G()|0 =
|0| |2
,
+ i
(5.18)
and so
1
0 () = ImG00 ().
(5.19)

Here it is understood that C d + D , where D the possible
discrete eigenvalues outside C. If |0
= 0, a pole exists just below the real
2
axis at = i, and the residue is ||0 | . In second quantization one
denes the Greens function as the vacuum average
G00 = vac|a0
1
a |vac .
H + i 0
(5.20)

Inserting a0 = a |0 , one nds (5.18) again. However the second quantized formulation lends itself to the many-body treatment when interactions
are included.
5.1.5 The Resolvent
= 1, that is,
We obtain all the elements of G from the identity ( H + i)G
|0 = 0 | 0 = 1
0| ( H + i)G
|0 = k | 0 = 0
k| ( H + i)G
|k = 0 | k = 0
0| ( H + i)G
|k = k | k = (k k ) .
k| ( H + i)G
1
i.e. proportional to (t)
Setting z = i, since Hkk = 0,

( 0 )G00 () k V0k Gk0 ()
=1
Vk0 G00 ()
=0
( k )Gk0 ()
( 0 )G0k () k V0k Gk k ()
=0
Vk0 G0k () + ( k )Gkk () = (k, k ).
87
(5.21)
Now, thanks to , nothing diverges, and we obtain

G00 =
where
() =
1
,
0 ()

k
(5.22)
|V0k |2
k + i
(5.23)
is the self-energy. For the other elements of the resolvent matrix see Problem
5.3.

Unlike the sum in (5.25) this is a smooth complex function of ,
k
stands for an integral;
1 () Re() = P
2 () Im() =
|Vk0 |2
k
k
|Vk0 | ( k ).
(5.24)
As a function of z = + i, is analytic outside the real axis. If C = {a

b}, then there is a cut just below the axis, with a and b as branch points.
Outside C, 2 = 0. is a Herglotz function, that is, 1 Im() 0; it
follows that G00 () is also Herglotz, which is important to ensure 0 0.
(0)
We can deduce more analytic properties of G. For Hh 0, G00 G00 =
1
0 +i has a real pole which marks the discrete eigenvalue. For Hh
= 0,
G00 (z) is analytic except for the cut along C and possible poles at the roots
of
z 0 = (z).
(5.25)
Real roots are possible outside the unperturbed continuum where 2 = 0. If
there are roots, equation (5.18) really means

|0| |2
|0| |2
+
;
(5.26)
G00 () =
d ( )
+ i
+ i
C
D
the residue at the pole is the probability that a particle in | is found in

|0 .
Complex roots of (5.24) have C and Imz < 0 by causality (G(t)
(t)); they represent resonances. If were constant, 0 would be Lorentzian;
although this an be a poor approximation, qualitatively, 1 is a shift and 2
88
a broadening; indeed, 2 = = constant, fot t > 0 we obtain G00 (t) by

closing the integral in the lower half plane:
2

dz eizt
d eit
=
= ieit t .
2 z + i
2 + i
To calculate the wave functions, we use the assumption of a non-degenerate
continuum that allows to make the replacement

dl ( )
and write

ImG00 =
d ( ) |0| |2 ( ) = ()|0| |2 ,
(5.27)
where | is the eigenstate at energy h. Hence,

Im z
Re z
cut
Resonant
state
Localized
state
Fig. 5.2. The singularities of G00 (z) must be in the lower half of the complex z
plane. Outside the continua one can nd poles close to the real axis at localized
states. Branch cuts correspond to the continua; a and b are branch points; poles
below the cuts yield resonant states.
|0| |2 =
0 ()
ImG00 ()
=
,
()
()
(5.28)
and the wave function at site 0 is obtained (the phase is arbitrary). In a

similar way, using Fano notations, Vk0 V V () and one obtains

2 () =
d(0) (k )|Vk0 |2 ( k ) = (0) ()V 2 () = ()V 2 ()

(5.29)
89
where the last equality holds in the thermodynamic limit. Now we can solve
(5.7) in the sense of distributions. The most general real solution of xf (x) = C
is

1
f (x) = C P + Z(x) ,
x
whereZ is a constant. Thus we nd a family of solutions:

1
|k = |0 Vk0 P
+ Z( )(k )
k
(5.30)
Putting into (5.6) we nd

Z() =
0 1 ()
,
2 ()
(5.31)
and the solution is complete.

5.1.6 Self-Energy Operator
The above Anderson method to calculate the resolvent can be cast in operator
form and is very useful in problems when it is convenient to separate the
Hilbert space in two subspaces A and B writing

HAA HAB
H=
(5.32)
HBA HBB
As in Sect. 5.1.2 take the AA matrix element of the identity ( H)G = 1 :
GAA =
1
1
+
HAB GBA .
HAA
HAA
(5.33)
Then we take the BA element closing the equations:

1
HBA GAA .
HBB
(5.34)
1
1
HAA HAB H
HBA .
BB
(5.35)
GBA =
Hence,
GAA =
Thus we can work within subspace A provided that we work with an energydependent eective hamiltonian
Hef f () = HAA + AA , AA = HAB
1
HBA .
HBB
(5.36)
AA is called self-energy operator. To get the other matrix elements one

simply exchanges A and B in (5.35,5.34). If several subspaces B, C, are
coupled to A but not among themselves, they yield additive contributions to
AA .
90
5.2 Magnetic Impurities and Chemisorption on

Transition Metal Surfaces
The Anderson model

H=
k nk +
00 n0 +
{Vk ck c0 +h.c.} + Un0 n0
k
(5.37)
is just a Fano Hamiltonian with the addition of the innocent-looking but

far-reaching U term, which represents an oversimplied electron-electron interaction. It was introduced in a fundamental paper[9] on magnetic impurities
in metals. The main question is: in what conditions will the localize state be
magnetic? The continuum tends equalize the populations n0 , n0 magnetic
sub-levels in order to lower the kinetic energy, but there is a price to pay to
the local interaction.
Vacuum level
d band
-A
Fermi
level
-I
Fig. 5.3. Relative position of energy levels in Newns model (not to scale). is the
work function. The bending of the Vacuum level is due to the surface electric eld.
Here I mainly discuss a 1969 paper by D.M. Newns[10] dealing with H

chemisorption on Cu and Ni by an Anderson model2 because important points
were made there. In his scheme, the localized atomic state |0 > is the 1s H
(0)
level with energy 0 and the continuum is due to the metal d band states
|k >, delocalized in the bulk but exponentially damped outside the metal.
The chemical bond is due to hopping matrix elements V0k aa in the Fano
model. Unlike the Anderson paper, Newns considered the implications of the
dependence of the self-energy () in detail.
U=0 limit
If U is unimportant, we expect localized states if there are real roots of
Equation (5.25). The new feature is the presence of a Fermi level. Instead of
local elements of the resolvent we now handle Greens functions. The occupied
states are described by the one-hole Greens function
2
Chemisorption is the chemical bond between an atom (molecule) and a surface).

Now it is studied mainly by ab-initio methods, but some key concepts developed
from models remain fundamental.
5.2 Magnetic Impurities and Chemisorption on Transition Metal Surfaces
G00 () = < |c0
1
c0 | ,
H
91
(5.38)
where | is the ground state. For U = 0, | is a slater determinant of

solutions | of the Fano model, and changing the basis with c0 = |0 c ,
we nd
G00 () =
|0 | |
;
+ i
(5.39)
this is already familiar except for the appearance of the Fermi function f .
For the empty states, one denes the one-electron Greens function
1
(5.40)
c0 | .
H
From the Fano model we can obtain interesting quantities like the level population

G00 () = < |c0
n0 = 2
EF
d0 ()
(5.41)
where EF is the Fermi energy and the factor 2 is due to the spin.
Interacting Anderson Model in Mean Field Approximation
For the Hydrogen atom, the ionization potential I = 13.6eV and the electron
anity A = 0.7eV, are so dierent that it is impossible to set up a sensible
2
model without the interaction U 0 0 | re12 |0 0 |I A|. A work
function (Typically 4.5 eV) separates the Fermi level from the Vacuum
Level (minimum energy of a free electron far from the metal); so even is
smaller than I A. We need a simple calculation to orient ourselves. In the
Hartree-Fock approximation, one assumes a ground state of the determinantal
form
| =
<EF , <EF c+ c |vac .
Averaging the one-body operator

n0 =
c+
c | 0 >< 0| >
(5.42)
on | , only diagonal terms contribute and one nds

f ||0 |2 .
n0, =
(5.43)
Moreover, since the anti-commutation relations allow to lump all the up spin
operators on one side, it is readily seen that
|n0 n0 | = |n0 | |n0 | n0 n0 .
(5.44)
92
Thus,
E = |H| =
f , |h|, + U n0 n0 .
(5.45)
where h is the one-body (Fano) part of H of (5.3). We Set up the variational

calculation of in rst quantization, treating bra and ket as independent
unknowns. We use (5.43), and vary the bra, for occupied states:
n0
=
|0 0 | = |0 0 | .
(5.46)
This contribution multiplied by 0 appears in the h term and multiplied

by U n0 in the interaction term. Thus, in the unrestricted3 Hartree-Fock
E
equation
= | , the U term produces a shift
(0)
(0)
0 0 + U n0 .
(5.47)
Back in second quantization, the spin electrons have their Fock Hamiltonian
H =0 n0 +
k nk +
{Vk ck c0 +h.c.}
(5.48)
where < n > is a parameter; thus the equations for are coupled.
No privileged spin direction exists in this problem, and one could have the
impression that n0 is equal to n0 for symmetry reasons: remarkably,
this is wrong if U is large, and the level is about half lled, because the
symmetry is spontaneously broken (see Fig. 3.3). There is a couple of ground
states, namely
n0, >> n0, ,
and another one with . The overall symmetry of the problem is
not respected by each ground states. The chemisorbed atom has a magnetic
moment. The existence of a localized spin requires strong enough U and
partial occupation; the Hartree-Fock approximation is known to overestimate
somewhat the occurrence of magnetism compared to more rened approaches.
Despite this possibility, Newns found a nonmagnetic solution for H
chemisorbed on Cu and Ni; let us consider more closely the case when
n0 = n0 = 12 n where n0 is the total population of the ad-atom.
Then, 0 does not depend on spin but according to (5.47) it depends linearly
on n0 , with 0 I for n0 0 and 0 A for n0 2. On the other
hand, the Fano formalism applied to the Fock Hamiltonian yields another
functional dependence through

n0 = 2
EF
d0 ().
(5.49)
5.2 Magnetic Impurities and Chemisorption on Transition Metal Surfaces
93
n n
n n
a)
b)
Fig. 5.4. Qualitative dependence of the energy E on local spin when the virtual
level is about half lled: a) for U small compared to level width b) strong coupling
case.
In this way a self-consistent solution can be obtained. Newns found n0 =

1.06 for Cu and n0 = 1.16 for Ni; this small negative charge on the H atom
produces an electrostatic dipole at the surface. The result is an increase in
the surface dipole due to the spill-over of the conduction electrons past the
edge of the positive charge and thus an increase in the work function, which
has been observed.
Local View
It is instructive to illustrate this theory by a simple LCAO4 model. By expanding the continuum states on a localized basis,

|k =
|i i|k
(5.50)
i
one can expand the hopping integrals as well:

Vi i| .
Vk =
(5.51)
The most important Vi are those connecting to nearest neighbors. Various

chemisorption geometries can be modeled ,including Atop, Bridge (bridging
between 2 atoms) and Center (above the center of a triangle) geometries. The
3
this is conventional term meaning that we are not assuming that all quantities
be spin-independent.
4
Linear Combination of Atomic Orbitals.
94
resulting virtual level shapes depend qualitatively on the number of nearest

neighbors. Here I consider the Atop case, denoting in this paragraph a the H
adatom orbital and 0 the top surface atom;
Vk Va0 0|k ,
(5.52)
is hopping matrix element, and

a () =

k
Vk2
|Va0 |2 GS ()
k + i
(5.53)
where GS () is the local Greens function of the surface atom, an intrinsic

property of the undisturbed surface. For the sake of argument, let us represent
the d band by a tight binding N + 1 atom chain, with orthogonal site orbitals
i|j = ij and Hij = between nearest neighbor sites, Hij = 0 otherwise.
Here, the sites of the chain are 0,1,2, and chemisorption occurs above
site 0. In this case, GS G00 . Like a moment ago, the self-energy of the
top atom is related to the local Greens function of the chain with the top
atom removed, that is the 1,2, atom chain; but removing an atom from
a semi-innite chain makes no dierence, and 00 = 2 G00 . Hence,
G00 (z) =
1
, z = + i.
z 2 G00 (z)
We nd the solutions
G00 (z) =
(5.54)

z 2 4 2
.
2 2
Taking
the cut of z along the positive real axis, z = ||ei arg /2 . For

3i
4
z 2 4 2 = | 2 4 2 | exp 2 [arg( 2) + arg( + 2)] the cut is between 2 and
real axis, and corresponds to the
2 along the
band. Just
2 4 2 = i | 2 4 2 |, just below the cut, z 2 4 2 =
above
the
cut,
z

i | 2 4 2 |. The density of states is () = 1 ImG() computed on the
real axis, that is, just above the cut, and in order to have it positive, we must
choose

z z 2 4 2
.
(5.55)
G00 (z) =
2 2
Thus one obtains the semi-elliptic density of states

4 2 2
(4 2 2 ).
n() =
2 2
(5.56)
Re[G00 ] is the Hilbert transform of n() and is odd; it is linear in the band
and outside

2 4 2
, > 2.
(5.57)
Re[G00 ] =
2 2
5.3 Strong Coupling and the Kondo Peak
95
The chemisorption is weak (strong) when V is small (large) compared to .

For weak chemisorption, the self-energy is small and a 1 () = 0 is
solved by a + 1 (a ). So, if a is well inside the band, the solution is
a resonance having width 2 (a ); if a is well outside the band, a discrete
state obtains. a in the continuum can give a discrete solution only if it is close
to the edge. For strong chemisorption, one may have two roots straddling the
band. Then,

1
1 d 2 ( )
2 ( )
a
0 = a
d
( )
( C )
with C the center of the band. The two roots obtained in this way are the
bonding and anti-bonding levels of a localized surface molecule.

J. Kondo [34] in 1964 succeeded in explaining the strong eect of a small
amount of magnetic impurities on the resistivity of non-magnetic metals 5 .
A magnetic impurity with incomplete d or f shells that at room temperature
behaves like a magnetic dipole at very low temperatures appears to lose its
spin; lowering the temperature further, the resistivity does not continue to
decrease as in pure metals but starts to increase. Even much less than 0.01
atom % Fe in Au suce to produce the resistivity minimum at the Kondo
temperature TK (several 0 K). The Kondo temperature TK (a few millivolts)
is very unlike the characteristic energies (band width, distance of the d or f
level from Fermi level, on-site repulsion.) Here I wish to show that idealized
models do bring many of the characters of the Kondo physics, that cannot
currently be reproduced by the common ab initio methods. The Andersonlike models explain how such a small energy scale as TK arises and how we
can monitor the electronic structure by electron spectroscopy. Various regions
in parameter space are relevant to dierent experimental situations. When 0
is close enough to the Fermi level, like the f levels in SmB6 , one speaks of
mixed valence regime because strong charge uctuations occur and the dierent ions can have dierent, non-integer valence. When 0 + U is close enough
to the Fermi level similar phenomena occur. By constrast in this Section we
investigate the Kondo scenario, when the localized level is deep but multiple
occupation is hindered by large U.
5.3.1 Narrow-Band Anderson Model

In the narrow band limit of the Anderson Hamiltonian the continuum is
replaced by a single level k at the Fermi energy F = 0; we shall have in
5
For an introduction to the Kondo eect see e.g. [5]; for ne recent reviews see
Hewsons book [20] and Ref. [19].
96
mind applications to d (or f ) levels of impurities in metals. Accordingly in

this Section we denote the localized state as the d level. A very interesting
situation arises in the strongly correlated case U |d | when the d level is
deep (d < 0 with V << |d |.) For a single particle, the basis set is {|d , |k }
and the Hamiltonian matrix is

d V
(5.58)
H(1) =
V 0

with eigenvalues = d2r , r = 2d + 4V 2 . The spin-degenerate ground
2
state g has energy d Vd . We are interested in the one-electron
Greens function
Gdd () = g |cd
1
c |g ,
H d
(5.59)
which brings information in the two-particle states in the Sz = 0 sector where

U can act. To predict the poles of Gdd let us examine the expansion of twobody state cd |g into stationary states. With 2 electrons of opposite spin,
on the basis
{|1 = |d d , |2 = |d k , |3 = |k d |, |4 = |k k },
(5.60)
the Hamiltonian matrix is
2d + U
V
H(2) =
V
0
V
d
0
V
V
0
d
V
0
V
.
V
0
(5.61)
The triplet component brings structure right at d . Indeed, the 2-particle

triplet states are
1
{|d k , |d k }, (|d k |k d )
2
but only the last one is in the Sz = 0 sector; they have eigenvalue d .
In the singlet sector, one nds 3 eigenvalues of H(2). Referring to the basis
(5.60), the G44 singlet is clearly mainly peaked at 0, but does not enter the
calculation of Gdd . The G11 element gives structure mainly at high energies
U . The singlet |s = 12 (|2 + |3 ) is more interesting, and gives the
combination
G22 + G33 + G32 + G23
Gss =
.
2
One can readily compute the resolvent matrix G() = ( H(2))1 and Gss .
The density of states derived from Gss has an interesting structure near the
Fermi level which persists at high U , as shown in Figure 5.4 This is called
theAbrikosov-Suhl or Kondo peak and has a striking physical interpretation.
97
Fig. 5.5. The singlet density of states as obtained from Gss with 0 = 1, U = 6
and V0 = 0.1; the eigenvalues of H(2) are at = 1.023, 1.0, 0.0195, 4.004 and a
triplet at = 1. The singlet has little intensity in the high energy peak (here at
= 4); it is practically insensitive to changes in U in the low energy sector as long
as U is large. The peak at 0 = 0.0195 close to the Fermi level is the Kondo peak
due to spin ip. The lines are broadened into Lorentzians (width 0.02 |0 |).
For V = 0 the d level is singly occupied in the ground state and the
impurity has a net free spin; the deep level has a SU(2) degree of freedom.
For small V, the d level remains essentially singly occupied in the ground
state, but the transfer of spectral weight shows that it manages to interact
with the Fermi level. The only way this can occur is by an exchange of the
conduction electron with the localized one, while the impurity ips its spin.
To simplify the algebra, we can chop from the Hilbert space the uninteresting state with a doubly occupied d level, since it stands alone at high
energy, and write
d 0 V
(5.62)
H (2) = 0 d V
V V 0
on the reduced basis
{|1 , |2 , |3 } = {|d k , |d k |, |k k }.
(5.63)
The triplet noted above, with eigenvalue d , is (r, r, 0), r = 12 . It is easy

to nd 2 normalized vectors orthogonal to (r, r, 0) and to each other,
for
in
r0 r
stance (r, r, 0) and (0, 0, 1.) Using the orthogonal matrix c = r 0 r ,
01 0
2 0

d V
= cT hc = V 2 0 0 ;
we nd the transformed Hamiltonian h
0
0 d
d 2d +8V 2
(2)
. While
hence we obtain two singlets with eigenvalues =
2
98
(2)
(2)
V
+ 2 Vd , the ground state energy d 2
says that the
d
singlet is slightly lower than the triplet for small V , and a bound state
forms in which the spin is screened; the eigenvector (on the basis (5.63)
(
(2)
(+ )2 +2V 2
,
(2)
(+ )2 +2V 2
(2)
+
(2) 2
(+ ) +2V 2
,
) shows that |d k and
|d k | components are the important ones. The low energy sector (singlet ground state and triplet excited level) is well represented by an eective
Hamiltonian
(5.64)
Hef f = 2JS1 S2
which commutes with the square of the total spin and describes a spin-spin
2
coupling between the two electrons. The small separation 2 FVd between
the ground state and the triplet implies that the low-energy excitation is
a spin-ip, while charge excitations are frozen. The coupling between the
conduction electron and the localized one through spin produces a singlet
(2)
level in the valence region at + .
The narrow-band model is extremely simple, but calculations on larger
systems conrm and extend the scenario. Recent exact-diagonalization calculations of Kondo clusters alloyed with mixed-valence impurities in the presence of disorder are of great interest [36] and show T=0 phase diagrams very
rich in structure.
5.3.2 Anderson Model, s-d Model and Kondo Model
Consider an Anderson model with d < 0 and U chosen such that for Vk 0
the d level is singly occupied in the ground state. At small Vk , the lowenergy subspace, say, subspace A, corresponds to singly occupied d level;
there are two high-energy subspaces, say B and B, with excitation energies
E(d1 d2 ) = d + U and E(d1 d0 ) = d . Let us use the SchrieerWol transformation ( Sect. (1.3)) with the hopping term as v; Equation
(1.86)yields the approximate renormalized interaction

(ck d )(d ck ) (d ck )(ck d )
.
(5.65)
+
Hint =
Vk Vk
d + U
d

k,k
Besides a conventional scattering of conduction electrons by an impurity

potential, Hint produces diusion with spin ip (of both conduction and
impurity spins). The eective Hamiltonian is clearly a particular form of the
s-d model proposed long ago by Zener[28], that we now recall.
s-d model
We need a slight generalization of Equation (1.58) in which we write

01
ck
( )x ck ck ,
(5.66)
(kk )x = ck ck
=
ck
10

kk
99
and so on; if k were equal to k this would be the representation of the spin
operator in the k basis, but we are allowing o-diagonal elements. In this
way,
(kk )z = ck ck ck ck , (kk )+ = 2ck ck .
The s-d model proposed by Zener[28] for an impurity spin S in a metal reads

6
15 +
+

H = HF +
S (kk ) + S (kk ) + Sz (kk )z (5.67)
Jk,k
2

k,k
where HF is a Fano Hamiltonian[32]. For the impurity we denote the creation

operator by d and use d 0 ; we set
2Sx = d d + d d , 2Sy = i(d d d d ), 2Sz = d d d d ;
(5.68)
one can readily verify the angular momentum commutator relations in the
subspace with n n = 0, n + n = 1.
Spin-ip and the Kondo Model
Now in Equation (5.65) we consider the spin-ip (
= ) terms ignoring the
dull potential scattering terms; for instance we have a term ck d d ck =
1
+

2 (kk ) Sd which ts the s-d model. So we arrive at the Kondo model[35]

H = H0 + HK =
k ck ck +
Jk,k ( S) ck ck
(5.69)
k,k
where
Jk,k =
Vk Vk

1
1
.
+
d + U
d
(5.70)
This describes the net eect of the virtual valence uctuations; they couple
the local spin density of the conduction electrons with the impurity spin. The
d electron is reduced to its spin degree of freedom, while the charge is xed.
The k dependence is often neglected. In this case, the Kondo interaction can
be written
(5.71)
Hint = HK = J (0) (0)( S d ),

1
where (0) = N k ck is the electron eld operator at the origin. Since

(0, 0, 1) (1, i, 0)
{ }
(5.72)
(1, i, 0) (0, 0, 1)
we may also write
+
HK = J
Sz (0) (0) S (0) (0)

S+ (0) (0) Sz (0) (0)
,
.
(5.73)
This is the second quantized form of a Hamiltonian HK = J( S d )(x).
100
5.3.3 Fermi Level Singularity and Kondo Minimum

The resistivity of metals decreases with decreasing temperature since the
phonon contribution drops; the Kondo minimum at TK arises because the
magnetic impurity contribution increases as T 0. The impurity resistivity
is due to k k processes where the conduction electrons scatter from one
Bloch state to another in the impurity potential and is proportional to the
square modulus of the scattering amplitude. The amplitude is an average over
the thermal distributions of electrons, and the distribution peaks at the Fermi
level as T 0. For K or Ca impurities in a metal, nothing special happens
with lowering T, but for impurities such as Fe or Co the temperature is very
important. This is because the the impurity cross section grows logarithmically when the Fermi energy is approached, and the eect is entirely due to
the spin. To see how this is borne out by the Kondo model, let
H = H0 + HK , HK =
J
( S) ck ck
N
(5.74)
k,k
with the conduction electron spin; let | denote a lled Fermi sphere,
with the energy origin such that H0 | = 0, ci and cf creation operators
for spin-up electrons near the Fermi level,
|i = ci | , |f = cf | .
One obtains the scattering amplitude Uf i = i |U |f , still an operator in
impurity spin space, by the T exp formula in the interaction picture (2.36)

Uf i = if i
i|HK ( )|f d
1

(i)2
+
d1
d2 i|HK (1 )HK (2 )|f +
(5.75)
2
The rst-order integral gives 2(if )(HK )if , that is, an energy-independent
object. The second-order contribution, introducing a complete set, reads

1

1
(2)
Uf i =
d1
d2 ei(f 1 i 2 )
(HK )f (HK )i eiE (2 1 ) ;
2
using a convergence factor at t0 ,

(2)
1
Uf i (i)2 2(f i )
d2 ei(E i )2 =
ei(E i )1
i(E i )
1 (HK )f (HK )i
.
i
E i
Two kinds of matrix elements f |HK | |HK |i occur, with the ci

operator in the right HK factor or in the left. The terms with ci in the right
5.4 The Nf 1 Expansion
f |( S)cf cq | |( S)cq ci |i

E i
101
where | has an electron in a state q, yield, using (5.72),

(1 fq )
q,
q i
( S)( S) =
(1 fq ) 5
q
q i
6
Sz2 + S S+ .
(5.76)
In the terms with ci in the left, the operators appear in the order cq ci cf cq ,
so q must refer to a lled state and | involve two electrons and a hole:
f |( S)cq ci | |( S)cf cq |i
E i

q
6
fq 5 2
Sz + S+ S .
q i
Since [S+ , S ] = 2Sz the second-order amplitude has a contribution

fq
proportional to q q
Sz . Neglecting the energy dependence of the density
i

1
1
of states, the q summation gives q q
FFW dq q
where W is the
i
i
band width.
|i F |
(2)
)
(5.77)
Uif Sz log(
W
This gives a logarithmic singularity of the scattering amplitude when i
approaches the the Fermi energy. For the physical meaning of the Sz factor
see Problem 5.4.

Spectroscopically, Ce compounds and heavy fermion materials show deep f
levels having widths 0.1 eV, while U is several eV. The above discussion
indicates that such a special region of parameter space of the Anderson model
is particularly intriguing for studies of magnetism: a deep impurity level is
weakly coupled to the Fermi level (Vk << F d ) and is singly occupied
in the limit Vk 0. This requires d well below the bottom of the band
and U large; actually there is a mathematically well dened U limit to
study. The density of states will contain structure about d and 2d + U,
but also about F . Since U is large one should resort to some expansion in
Vk . Here I shall introduce with some changes and simplications a method
proposed by Gunnarsson and Sch
onhammer (in a comprehensive paper [27])
to calculate the relevant densities of states. Consider an Anderson-like model,
with a local impurity level which is Nf times degenerate
H=

k,
k nk + f
Nf
nf
102
Nf

nf nf , {m, }.
Vkm c0 ck + h.c. + U
(5.78)
,
The bottom of the band is at B and the Fermi energy is F = 0. In this

problem it is expedient to use a discrete formalism as in Section 5.1.4, i.e.
with {k } = 1 , 2 , and Kroneker deltas, and to go to the continuum limit
at the end. Let us assume

()|2 mm ;
V V
|W
k mk km ,k
(5.79)
2
m Vmk Vk m ,k |W ()| kk

linear in the hopping parameters. The U term fails to
with the energy W
account for the multiplet eects, but this has no consequence since the occupation of the impurity is nd 2. Introducing the degeneracy, the model becomes both more realistic and more tractable. The reason is that (as expected
a priori and veried below) each degenerate state gives equal contributions
to the ground state energy shift E, so for deep impurity levels
E = Nf
2
W
,
f
(5.80)
2 is some average of W
(). Thus, E is xed (it corresponds anyhow
where W
to a weak chemical interaction), Nf is a weak coupling limit, and we
can set up an expansion in inverse powers of Nf . We need the right combinations of conduction states that couple to the impurity, and introduce new
conduction operators with the same symmetry as the impurity m orbital6 :
1
m
=
Vmk ,k ck
(5.81)
W () k
and a combined index {m, }. Equations (5.79) yields
5
6
m , m + = , , .

Then, letting d stand for the discrete energy summation (that will become
continuous at the end)
k1 Vk1m (k2 k1 )Vmk2

m =
dm
ck1 ck2 ; (5.82)
2 (k1 )
W
m
m k1k2
now using the second of (5.79), one arrives at the useful form
H=
Nf
d

+

()c + h.c.
d W
0

nf nf + extra terms
+f c0 c0 + U
{, }
here unlike Ref [27] I use dimensionless the creation operators
(5.83)
103

where {, } is over distinct pairs and the extra terms describe conduction
states that are not coupled to the impurity. We shall see that the ground
state can be calculated exactly for U , Nf . We start by calling
|0 the ground state of H and | the singlet state with a lled shell of k
states up to F and empty impurity states; let
0 |H|0 = E0 , |H| = E00 , E = E0 E00 .
(5.84)
Henceforth the many-body energies will be referenced to the constant E00 . By

acting on | with the hopping term in the hamiltonian we produce electronhole states |f,
with one electron on a superposition of the impurity states
|0m and a hole in the conduction band at some occupied energy ; denoting
holes with bars one writes a particle-hole state as
d band
Fermi level
V
f
Fig. 5.6. Left: lowest state | with no occupation of the f state . Center: hopping
leads from | to a state with a single occupation of the f state. The full circle
represents a hole and the empty one a f electron: this conguration would be the
ground state for V 0 for the hole at the Fermi level. Right: further hopping may
lead to two kind of congurations: 1) a state |E, with no electrons in the f state
and an electron-hole pair in the valence band (upper panel) 2) a state |f, f , ,
with a double f hole (lower panel). For large U , however the latter is excluded, while
the coupling to |E, is negligible for large Nf . Thus, when both conditions hold,
one can solve the problem exactly.
1
|f, =
c0 | .
Nf
(5.85)
Note that f, |f, = 1, f,

|H|f, = f . To nd the coupling to | we
insert (5.81) in (5.85), obtain

Vkm c0 ck |
(5.86)
H| = E00 | +
k
104
by the form (5.78) of the Hamiltonian, and take the scalar product. The
nonvanishing contributions are those where the creation and annihilation operators match; then, using (5.79) one performs the k summation introducing
|2 ; the m summation then introduces a Nf factor yielding
|W
Nf

1 |Vkm |2 ( k )
() Nf ;
=W
|H|f, =
Nf k
W ()
(5.87)
If we act over (5.85) with the hopping term of the hamiltonian we generate
new states:
(5.88)
H|f,
|E,
, |f, f , , ;
here,
1
|E, =
|
Nf m, E
(5.89)
has a conduction electron with energy E > EF and keeps the hole while
|f, f ,
, has two impurity electrons and two conduction holes and is ruled
out as U . Since the normalized states |E,
and |f, both bring a 1
Nf
|2 and thus vanishes as Nf because of

factor, their coupling goes like |W
(5.80).Therefore, in that case the problem reduces to the one hole subspace,
that is, the ground state may be sought in the form

0
a()|f d ;
(5.90)
|0 = A | +
B
0
here B d provisionally stands for a discrete summation over the band energies and everything is dimensionless. We write the Schrodinger equation on
a basis set
{| , |f, 1 , |f, 2 , . . .}
with eigenvectors
v = (1, a(1 ), a(2 ), . . .)
acted by the Hamiltonian matrix (recalling (5.87))

(1 ) Nf W
(2 )
Nf W
0
Nf W
(1 )
f 1
0
HM =
Nf W
(2 )
0
f 2
(5.91)
(5.92)
The rst line of HM v = Ev gives

E = Nf
the second and any other line yield
0
B
()a(),
dW
(5.93)

() = (E f + )a().
Nf W
Hence,

E = Nf
d
B
()|2
|W
E f +
105
(5.94)
(5.95)
is indeed proportional to Nf , as anticipated. Moreover, the normalization

condition of (5.91) reads
0
0
()
W
1
2
2
|A| =
, C=
)2 ,
da () = Nf
d(
(5.96)
1+C
E f +
B
B
where again B is the bottom of the band. The occupation of the impurity
is
C
,
(5.97)
nf =
1+C
and when C 1 (spin uctuation limit) nf is close to 1. Equations
(5.95,5.96,5.97) remain the same in the continuum limit, when however the
Kroneker deltas over energy become Diracs delta functions and hence
2
= W
(5.98)
is the inverse life time due to hopping, a the inverse root of an energy, and
so on. Typically, 0.1eV.
5.4.1 Kondo Temperature in the Spin-Fluctuation Case
E is the energy shift with respect to | , but one is more interested in the
Kondo temperature
KB TK = = f E > 0,
(5.99)
which is the correlation-induced energy gain with respect to the non-interacting
lowest eigenvalue of |f, levels. In the spin uctuation case,when nf 1, a
2 and assumes that B ; then, setting
reasonable model takes constant W
in (5.95,5.96) W = , we nd
E =
and
Nf
E f
Nf
ln |
|=
ln | |
B
nf
Nf
.
=C =
1 nf
||
(5.100)
(5.101)
One obtains from (5.100),

E
= Be Nf ;
we expect the result to be roughly proportional to Nf so we introduce this
dependence, using (5.99) to eliminate E :
106
=(
(f )
( )
Nf B
Nf ln[ NB
]+ Nf
f
f
)(
)e Nf = (
)e
.
Nf
This can be rewritten

Nf Nf {(f )+ Nf ln[( NB
Nf
)]}
f
=(
= (
)e
)e
( )
f
Nf
(5.102)
with the renormalized f level

f = f +
In the Kondo problem the exponent
the Kondo temperature is given by
B
Nf
ln(
).
Nf
(
f )
Nf

f
Nf
(5.103)
is large and negative and
|e |
Nf Nf f
)e
.
=(
(5.104)
The population of the f level can be deduced from (5.101), which says
nf
Nf
=
. Since nf is close to 1, the solution is well approximated
that 1n
f
by
1 nf = e
| |
f
f
(5.105)
5.4.2 Density of Occupied States

To obtain the density of lled states relevant e.g. for valence photoemission
(see next Chapter) we need the g < (z) Greens function. So, we introduce a
(suciently) complete set {j} of many body states and write
g < (z) =

i,j
0 |cm |i i|
1
|j j|cm |0 .
z E0 (N ) + H
(5.106)
One can compute the matrix elements i|z H|j without diculty and then
1
obtain i| zE0 (N
)+H |j by a matrix inversion in analogy with Andersons
procedure used above in Sect. (5.1.2). For large U and Nf one can limit
the analysis to the above approximation; the relevant states are the one-hole
states
(5.107)
|m = m | ,
such that m|m = ( ), and impurity electron -two hole states,
namely,
c
2 m2 2 1 m1 1 | .
|0m , 1 m1 1 , 2 m2 2 = 0m
Nf 1 pairs
(5.108)
107

Here, pairs sums over dierent m, channels (the normalization is to Nf 1
since the case of {m1 , 1 } = {m2 , 2 } is excluded, but 1 is negligible compared
f
with eigenvalue E00 ;

d
to Nf ). |m is eigenstate of N
inspection of (5.83) and (5.84) then shows that

m|z E0 + H|m = ( )(z E )
(5.109)
0m , 1 m1 1 , 2 m2 2 |z E0 + H|0m , 1 m1 1 , 2 m2 2

= (1 1 )(2 2 )(z E 1 2 + f ) (5.110)
The hopping Hamiltonian takes from the one-hole subspace to a two-holesimpurity-electron one; so the matrix inversion is easy, using the method of
Sect. 5.1.6 which puts a self-energy in the denominator of the inverse of
(5.109); giving
m|(z E0 + H)1 |m =
( )
. (5.111)
0
()|2
d|W
z E Nf B zE+
f
The Kondo peak is borne out by this analysis.

Lacroix [21] using the equation of motion approach computed the density
of states of the Anderson model for U by the equation of motion
technique, and obtained the Kondo peak and its disappearence above the
Kondo temperature.
Problems
5.1. Find by direct matrix inversion G00 () as the 00 element of ( h)1 ,
with the matrix of h from Equation (5.3). This is instructive and very easy!
5.2. Prove Equation (13.136).
5.3. Find the other elements of the resolvent matrix by the method used in
the text for G00 .
5.4. What does the Sz factor in (5.77) mean?
6 Many-body Eects in Electron

Spectroscopies
6.1 Electron Spectroscopy for Chemical Analysis

(ESCA)
Electron spectroscopies and their rich phenomenology convey information on
molecules and solids and specically on their excitations. To describe excited
states we shall develop techniques based on modeling the spectra in terms of
Greens functions and also more general expectation values. ESCA is a set of
spectroscopies, UPS (Ultraviolet Photoemission Spectroscopy),XPS (X-Ray
Photoemission Spectroscopy)and AES (Auger Electron Spectroscopy). They
can be angular resolved (ARUPS and the like) and time resolved; APECS
(Auger Photoelectron Coincidence Spectroscopy) detects the photoelectron
and the Auger electron most probably coming from the same atom. ESCA
can make a chemical analysis with tiny samples and is able to discriminate
the oxidation states and give information about the electronic structure.
In an ESCA apparatus, the sample is excited with monochromatic radiation ( X-rays in the case of X-ray Photoemission Spectroscopy or XPS,
Ultraviolet light in the case of UPS) and the potoelectrons are analyzed in
energy by a spherical or cylindric analizer. A photoemission spectrum is a
plot of the photoelectron current versus photoelectron kinetic energy. In [94]
the XPS spectra of noble gas atoms and of several molecules are reported and
discussed. Core levels, with a well resolved spin-orbit separation, and valence
levels are observed; they often (but not always) agree qualitatively with the
results of Hartree-Fock calculations. However, correlation eects, (that require the methods of Chapter 11 and following Chapters for a microscopic
description) are most often clearly seen. Many core and valence levels show a
complex structure due to shake-up eects (excitations) or shake-o eects,
if the excitations involve a continuum. The Xe 4p level seen in photoemission
(see e.g. [95]) shows a broad continuum with sharp superimposed features in
striking contrast with one-particle descriptions, predicting a doublet of lines.
The interaction of the sample with the radiation is
e
[A(xi ) pi + pi A(xi )] ,
2mc i
N
H =
(6.1)
110
6 Many-body Eects in Electron Spectroscopies
where N is the number of electrons and A the vector potential. One also
often nds an alternative formulation, that comes directly from the relativistic
theory, namely

1
H = 2mc
c
i
N
i) =
d3 xA(xi ) j(x
Mmn am an
(6.2)
mn

where the current operator is (4.23); in the second-quantized form,
mn
runs over an arbitrary complete orthonormal set of one-body states (the set of
Hartree-Fock spin-orbitals is a convenient choice). Let |i , Ei denote the initial
state of the sample and its energy, h and k the photon and photoelectron
energies. The cross section can be worked out starting from the Fermi
golden rule
2
=
|f |H |i |2 (
h + Ei Ef )
(6.3)
h
where |f , f are the nal state of the sample and its energy. For the sake
of simplicity we shall consider the case of fast photoelectrons, when we can
neglect the post-collisional interaction between the photoelectron and the
be the N-1-electron Hamiltonian of the sample
ion left behind. We let H
f = E
f |f and write Ef = k + E
f . The
after photoionization, with H|
photoelectron state can be approximately described as a plane-wave of wave
vector k, and |f =ak |f ,where

|f is the N-1-electron nal state of the
1
sample . Moreover, f = f k , where the sum over
the wave vectors
accepted by the detector in the solid angle , namely k (k )
is the density of nal states of the photoelectron k .

Then the cross section reads

2
=
(k )
|
f|am |i |2 (Ei Ef k h).
h
m
(6.4)
The highest kinetic energy features in the photoemission spectrum correspond

to events when the system is left at or near the ground state; lower kinetic
energy features correspond to excited states of the system left behind, and
this occurs by energy conservation, without the need for a post-collisional
interaction2 . Proceeding as in (4.6) one can write this in terms of the hole
Greens function
Gmn () = i|an
1
1
am |i .
H + i0+
(6.5)
Due to the long range of the Coulomb potential, one should use Coulomb waves,
but at high kinetic energy one has a good excuse for using plane waves instead.
2
The outgoing electron can lose energy by collisions, producing secondary electrons; this is not contained in the above simplied description; the spectrum far
from threshold when such eects are important becomes dicult to analyze.
6.1 Electron Spectroscopy for Chemical Analysis (ESCA)
111
6.1.1 Chemical Shifts

The core level binding energies depend on the charge on the atom and change
with the chemical state. For example, the Xenon uorides XeF2 , XeF4 and
XeF6 are colorless solids melting at 140, 114 and 46 o C, respectively; their
F and Xe core levels are shifted by some eV along the series. The F 1s
level is 5.48 eV less bound in XeF2 than in F2 ; this dierence reduces to
4.6 eV in XeF4 and to 3.38 eV, in XeF6 , while the negative charge on the
Fluorine is reduced. The Xe levels shift in the opposite way, even more: the
binding energy grows if there is positive charge on the atom. A clear linear
correlation exists between the binding energies of the levels and the overall
charges of atoms, determined by self-consistent calculations. In a similar way,
the 1s levels of C in CH3 Br, CH2 Br2 , CHBr3 , CBr4 shift compared to CH4 .
Many organic molecules contain in-equivalent C atoms and their core peaks
can be resolved .
The chemical shifts are due to a combination of initial state eects and
nal state eects . The former are those just mentioned above due to the
mean electrostatic potential on the atom in the ground state; depending on
the chemical environment, a core level can be more or less tightly bound
than in the free atom. The nal state eects are due to the polarization of
the system around the core hole and are of the same order of magnitude,
i.e. several eV. However, they always reduce the binding energy since they
stabilize the nal state of the sample and the energy is taken o by the
photoelectron. From the the chemical shifts it is often possible to deduce
the valence of given atom in a compound. This is relevant to elements like
transition metals that can have dierent valence in dierent compounds.
6.1.2 Core-Level Splitting in Paramagnetic Molecules
The ground electronic conguration of the NO molecule is
O1s2 N 1s2 1 2 2 2 1 4 3 2 2 1 ;
the ground state is a spin doublet and N O+ has singlet and triplet states.
The ground state of O2 is 3 g , with a conguration
1s2 g (2s)2 u (2s)2 g (2p)2 u (2p)4 [g (2p)1 ]2
with two partially lled g states. In both cases this produces a core-level
splitting[94]. In N2 the 1s level has a binding energy 410 eV and is 0.9
eV wide at half maximum; the same level in NO has one component which
is less bound than in N2 by a fraction of eV and is about as wide, but there
is also a second component which is more bound by 1.5 eV. The intensity
ratio is 3:1. The O level is not resolved in Siegbahns data but is 0.3 eV
broader than in O2 . To simplify the notation, let us write the degenerate
ground determinantal states by
112

(N O) =
|ss|, sz = 12
|ss|, sz = 1
2
(6.6)
omitting lled shells; s denotes the 1s level of O or N, according to the process

that we wish to describe. For N O+ we must consider
|s|, sz = 1
(N O+ ) = |s|, |s|, sz = 0
(6.7)
|s|, sz = 1.
All these states have = |Lz | = 1, where z is the internuclear axis, but not
all are spin eigenstates. The singlet is
1
(1 ) = [|s| |s|];
2
(6.8)
the triplet states are:

(3 ) =
|s|, sz = 1
+ |s|, sz = 0
|s|, sz = 1.
1 |s|
2
(6.9)
Let us write the total energy of the states with sz =0; since
H=
hi +
1
,
r
i<j ij
(6.10)
does not depend on spin and |H| = |H| , one nds:

E(1 ) = |s||H||s| |s||H||s| ,
.
E(3 ) = |s||H||s| + |s||H||s| .
In the calculation of |s||H||s| , since the two determinants dier by
2 spin-orbitals, there is no contribution from the one-body Hamiltonian; the
splitting is E = 2J, where
J = |s||
1
||s|
r12
(6.11)
is the exchange integral. Hartree-Fock calculations give E(N ) = 0.88

eV,E(O) = 0.68 eV. The lowest state is triplet. This agrees with the observation that the peak at lower binding energy is 3 times more intense.
In a similar way, the 1s spectrum of O2 (binding energy 547 eV ) has
two components separated by 1.1 eV with an intensity ratio 2:1. From the
3
g ground state one can go to 4 and 2 ions. Let represent the
orbitals with Lz = 1. From the ground state with components (3 g ) =
| + |, sz = 1

3
( g ) =
| + |,
1 [| + | + | + |],
2
+
| |,
113
sz = 1
sz = 0
sz = 1
To write the nal state, let us include the unpaired 1s level (that in the initial
state was understood with the other closed-shell states). 6 states are found,
including the sz = 32 component of 4 , namely, (4 ) = | + s|. The
other components of 4 are easily obtained by the S operator and the
doublet 2 by orthogonality. In this way one explains the splitting, which is
again due to an exchange interaction.
6.1.3 Shake-up, Shake-o, Relaxation
In the nal state of core-level photoemission, a localized hole exists, producing
a eld and a polarization of the system. The polarization phenomena are
not included in independent particles theory. However, they are commonly
observed: they shift the levels towards lower binding energies. The nal state
eects show up in the spectrum also with the presence of satellite peaks.
The electron contribute to the relaxation shift, since the spectator electrons
actually are involved in some measure in the photoemission process; there
is some probability that the ion is left, in the nal state, excited. We can
adapt the independent particles theory in order to to include some correlation
eects, by using dierent orbitals for the initial and nal states. Neglecting the
energy dependence of the density of states (k ) and of the matrix elements,
the shape of the photoelectron spectrum from a deep level |c is given by
1
() = ImGc () = i|ac ( H)ac |i = i; c|( H)|i; c ,
(6.12)
where |i; c = ac |i is the N-1 electrons state that is obtained from the initial
state |i by creating the core hole but holding the orbitals frozen. We calculate
|i in the Hartree-Fock approximation for the neutral sample; under known
conditions |i; c is a single Slater determinant. Introducing the eigenstates |
and eigenvalues of H with N-1 electrons, we nd

() =
|i; c| |2 ( ).
(6.13)
However, the | eigenstates must be calculated in the presence of the corehole, that is, as determinants formed with relaxed spin-orbitals. We can obtain them too by the Hartree-Fock approximation using an excited N-1 electron conguration with the core electron missing3 . Since the overlap of determinants is the determinant of overlaps (Equation 1.5) all the many-body |
3
Unfortunately, this approximation has the shortcoming that the excited state
is not orthogonal to the ground N-1 electron state or to the lower states of the same
symmetry. This is a general drawback of the Hartree-Fock approximation.
114
states contribute to the summation, each yielding a peak at its eigenvalue.

The discrete ones yield narrow shake-up peaks, while those in the continuum
contribute broad shake-o structures. Remarkably, the rst two moments of
the spectrum give

d
|i; c| |2 ( ) = 1
(6.14)
and
|i; c| |2 ( ) = i; c|H|i; c .
(6.15)
The excitation of satellites exactly balances the relaxation shift. The 1s

spectrum of Ne is rich of satellites (see next Figure).
Ne 1s main line
J (a.u.)
320
340
360
380
Kinetic Energy (eV)
Fig. 6.1. Sketch of the satellite region of the Ne 1s photoemission spectrum (data
from Ref. ([94])). The peak in the extreme right at 383 eV kinetic energy is due
to 1s holes in the nal state; the satellites occur at lower kinetic energy because
the N e+ ion is left in an excited state. Several satellites are due to states of the
2p5 np1 2 S nal congurations, with n=3, 4, 5, . The most intense satellites are
some 20 times smaller than the main line, since Ne is a lled-shell system with low
polarizability.
Besides the single particle excitations, other excitations of the system,

like vibrations and plasmons, that can be treated like harmonic oscillators
(bosons) contribute to the relaxation energy and to the shake-up spectrum.
The time scale of vibrations is 1013 s; typically the dipole-allowed electronic transitions take much shorter than that and according to the FranckCondon principle the nuclei do not have time to move during the transition.
Since the equilibrium bond lengths and angles of the ions are dierent from
the initial neutral species, the ion is left in an excited vibrational state. This
is easily visualized in the case of bi-atomic molecules by plotting the potential energy surfaces of the neutral molecule and of the ion versus internuclear
115
distance. The energy of the transition between the vibrational ground states
of two electronic states of the neutral species and of the ion is called adiabatic ionization energy. The satellites are on the high binding energy side
and appear as energy losses; they correspond to vibrationally excited ions in
the nal state. The energy spacing between satellites is the vibration energy
of the ion, which diers from that of the neutral species.
When the equilibrium distances dier slightly compared to the widths of
the Gaussian wave functions, there is a high probability to end up in the
vibrational ground state of the ion, and this corresponds the most intense
peak in the photoemission spectrum. The satellites are small and fast decreasing. On the other hand, if the change in equilibrium distance is large,
the intensity of the adiabatic transition is small, the spectrum has many lines
of comparable intensities and appears almost continuous, with an intensity
maximum at the vertical ionization energy, i.e. the energy dierence between
the two potential energy surfaces at the ground state bond length. To assign
the peaks in the photoemission spectrum, an important hint comes from the
vibrational structure; this is much more intense for bonding than nonbonding
or core levels.
6.1.4 Lundqvist Model of Phonon and Plasmon Satellites
The coupling of a core electron to a vibrational mode aects the Photoemission line shape. To model the vibration, consider a single Boson mode
0
of frequency 0 =
2 with vibration coordinate x; in this way we disregard
the dierence between the initial and nal state frequencies, but such details
are readily xed if necessary. Before the photoemission event, the harmonic
potential of the vibration of the ion can be written V (x) = 12 M 02 x2 and has
a minimum for x = 0. The initial state Hamiltonian of the vibration is
H0 = 0 d d,
(6.16)
with [d, d ] = 1; the initial state |i is the vacuum, |i = |0d , and the excited
eigenstates are
1
|nd = dn |0d .
(6.17)
n!
The photo-ionization produces a sudden change in the Hamiltonian; the
potential is still harmonic, but the minimum is shifted. For a harmonic potential, the shift x x + produces a change V (x) V (x) + M 02 x
apart from a constant 2 . In second quantization, the new interaction
term proportional to x, may be written in the form
H1 = g(d + d ).
(6.18)
This is the Lunqvist model discussed by Langreth in a very enlightening

paper[39]. In the nal-state Hamiltonian H = H0 +H1 we perform a canonical
transformation to new bosons
116
s = d , s = d ,
(6.19)
with to be found; since the canonical transformation is time-independent,

is just H written in terms of the s operators,
the transformed Hamiltonian H
= 0 [s s + s + s + ||2 ] + g[s + + s + ].
H
We require that there be no terms linear in s or s , and obtain
=
g
;
0
(6.20)
the new Hamiltonian is diagonal:

2
= 0 s s g .
H
0
(6.21)
2
Note the ground state energy is lowered by the relaxation shift E = g 0

that pushes the threshold of the photoemission spectrum to higher kinetic
energy. Remarkably, treating H1 in second-order perturbation theory one
would obtain the exact result foe E. Note that the ground state |0s of H
4
is the s vacuum while |i is the d vacuum . Since
d|0s = |0s
(6.22)
that is, |0s is an eigenstate of the annihilation operator, we say that it is a

coherent state for the d bosons.
Next I present a very simple way to solve Lundqvists model for the whole
spectrum (see 11.49 for a more sophisticated and general method). The eigen with eigenvalues En = E + n0 are
states of H
1
|ns = (s )n |0s .
n!
(6.23)
We seek the density of states
L() i|( H)|i

=
|i|n |2 ( En ),
(6.24)
n=0
so we need the Franck-Condon factors |i|n |2 |0d |ns |2 . Let 0d |0s = C;
then, 0d |1s = 0d |s |0s = 0d |d + g0 |0s = g0 C, since the d contribution
is 0. In general,
g
1
1
0d |ns = 0d |(s )n |0s = ( )n C.
n!
n! 0
4
(6.25)
The inactionary cosmology regards the creation of the universe as a transition

from a false vacuum. In Physics one often meets amazing connections between very
unlike phenomena.
117
The normalization condition gives

1
C = e 2 a , a =
Hence,
g2
E
=
.
2
0
0

an
( E n0 ).
n!
n=0
L() = ea
(6.26)
(6.27)
The probability of n excited bosons Pn = an! ea follows the Poisson statistics, reecting the statistical independence of the non-interacting bosons, and
the average number of excited bosons is

n = a = 2 .
(6.28)
The density of states is normalized to 1 and its rst moment is

dL() = ea

an
(n0 + E) .
n!
n=0
Using (6.26) we readily nd that this vanishes since

an
an1
n0 =
a0 = Eea .
n!
(n
1)!
n=1
n=1
The center-of-mass of the line remains where it would be for g=0; the threshold is at lower binding energy, at a0 below the center-of-mass. Fourier transforming L() one nds the correlation function

dL()eit = i|eiHt |i
2L(t) =
= ea
Thus,

an ia0 tin0 t
e
.
n!
n=0
(6.29)
i|eiHt |i = eC(t) ,
(6.30)
C(t) = ia0 t + a(ei0 t 1).
(6.31)
where
Note the characteristic exponential at the exponent. For strongly coupled,
slow modes, the a 1 case is relevant. Formally, we let 0 0 with g 2 = a02
nite. Then, eiHt e
g2 t2
2
and one nds the Gaussian line shape
2
1
L() = e 2g2 .
g 2
(6.32)
118
The relaxation energy diverges in the Gaussian limit, which is a serious overestimate; however the Gaussian line shape is often a good approximation
for phonon broadened core levels in solids and the width provides a sensible
measure of the electron-phonon interaction, although the Gaussian should be
convolved with a Lorentzian (producing a so called Voigt prole) to account
for the core hole lifetime. The result
lim ea
2

an
1
( + a0 n0 ) = e 2g2
n!
g 2
n=0
(6.33)
will be useful in Section 15.3.2.
6.2 Auger CVV Line Shapes: Two-Hole Resonances

Electron spectroscopies are a most interesting window over the various and
apparently fantastic scenarios of many-body theory including the strong correlation case and the appearing of bound states. This is particularly true for
the theory of the Auger eect[107] in solids (for a review of the Auger line
shape theory see Ref. [109]). In the core-valence valence transitions, the primary core hole decays into the Auger electrons and a pair of valence holes. In
the earliest theory developed by Lander[108] within the band theory of solids,
the Auger spectrum is proportional to the self-convolution of the density of
occupied states. However, Powell[110] pointed out that while the Al spectrum
was qualitatively consistent with the Lander predictions, the Ag spectrum is
quasi-atomic (it shows atomic multiplet peaks much narrower than twice the
band-width). Other authors [111] proposed a classication of spectra after
their atomic-like or band-like shape.
Strong deviations from one-electron theory arose in the rst correlated
theory[75][76] of Auger CVV line shapes (now known as Cini-Sawatzky theory). Let us consider an atom interacting with a solid, assuming for simplicity
that the valence orbitals of the atom and the solid band are completely lled;
the model Hamiltonian[76] in the hole representation is
H = Ha + HS + Hint + Hf
(6.34)
where the atomic contribution

Ha = dh cdh cdh + l
cm cm + Hr
(6.35)
m,
comprises a spin-up deep hole term for the primary hole, cm creates a
valence hole with magnetic quantum number m; Hr is the hole-hole repulsion Hamiltonian with screened direct and exchange matrix elements
U (m1 , m2 , m3 , m4 ). The crystal Hamiltonian HS and the atom-crystal hopping term Hint are one-body contributions that can be chosen according to
any model deemed appropriate. Finally,
Hf =
k ck ck
119
(6.36)
k,
describes the free particle continuum above the vacuum level; in the ground
state | of H only the k holes exist and H| = 0. Our initial state is

|i = cdh | . Core-Valence-Valence Auger transitions are produced by the
matrix elements
A(m, m , k) = dh (1)k (2)|
1
|m (1)m (2)
r12
(6.37)
of the Coulomb interactions and their exchange counterparts. Inter-atomic

Auger transitions, producing holes on nearby atoms, are neglected because
their matrix elements are very small. Thus, the transitions lead to a set of
states
|mn = cn cm |
(6.38)
with two holes localized on the valence orbitals of the same atom that was
ionized initially. At this point Hf may be forgotten.
By the Fermi golden rule one nds[76] that the spectrum is proportional
to the free-electron density of states and to S(Ei Ek , k), given by

Amn (k)Apq (k)Dmnpq ()
(6.39)
S(, k) =
mnpq
where the local density of states (LDOS) for the two nal-state valence holes
is
h Hr H1 )|mn ;
(6.40)
Dmnpq () = pq|(
here, H1 is the sum of all the one-body terms of H Hf . The dependence
on the angles of k can be taken care of by multiple scattering techniques, in
the spirit of Ref. [86], but here we concentrate on the dependence.
The fact that the local density of states appears, rather than the band
density of states as in the Lander theory, is one of the main points of Ref.[76]
and qualies Auger spectroscopy as a local probe of valence states. There
are two trivial special cases. In the atomic limit Hint = 0, and HS = 0;
then the Hamiltonian is diagonal in the L-S or |LSM MS representation;
the spectrum consists of unbroadened multiplet terms. The other simple
limiting case is the non-interacting (Hr 0) case; density of states ma(0)
trix Dmnpq () is readily worked out in terms of the one-body local denh H)|n = nm () using their transforms;
sity of states mn () = m|(
here we dene the correlation functions Dmnpq (t) = pq|eiHt |mn and
mn (t) = m|eiHt |n , both with diagonal elements equal to 1 for t=0.
Indeed,
(0)
Dmnpq (t) = mp (t)nq (t)

(0)
Dmnpq (t) = pm (t)qn (t) pn (t)qm (t).
(6.41)
120
In frequency space these become convolutions of one-hole densities of states.

To calculate the interacting density of states including the solid one can
1
1
1
1
use the identity (4.92) zH
= zH
+ zH
H1 zH
that holds if H = H0 +H1 .
0
0
In this case, H1 is identied with the repulsion Hr and H0 is the one-body part
of H Hf . Taking matrix elements in the basis (6.38) one obtains
a matrix

equation for the Fourier transform of Dmnpq (t)(t), using 0 dt exp[i(
i
H)t t] = H+i
. However, if the solid does not signicantly perturb
the spherical symmetry of the atom, the the D matrices are diagonal in the
|LM SMS representation, where Hr is diagonal, and obey uncoupled the
equations.
Then, we may drop the L,M indices and dene
(0) () = i00|
1
|00 = i[I (0) () iD(0) ()],
H0 + i
with
I (0) () =
d
D(0) ( )
.

(6.42)
(6.43)
One obtains
() =
(0) ()
,
1 + iU (0) ()
(6.44)
and within the bands

1
D(0) ()
Re =
.
(6.45)
(1 U I (0) ())2 + 2 U 2 D(0) ()2

The center-of-mass of D computed from (6.40) is dD() = H1 +Hr
averaged over the appropriate multiplet L,M component; understanding the
term, it is shifted by U = Hr , so correlation must deform the line shape
within the continuum. However no line shape can ensure this if U is too large,
and something must grow outside. Poles of outside the bands are two-hole
resonances.
To better understand the result, let us consider the rectangular local density of states
( ||)
.
(6.46)
() =
2
To do self-convolutions of functions that vanish outside an interval, it is expedient to use the identities
(a |x|) = (a + x)(a x),

i)
ii) (x + a)(x + b) = (b a)(x + a) + (a b)(x + b),
(6.47)
iii)
(x + a)(c x) = (a + c) [(x + a) (x c)]
D() =
where iii) comes from ii) and (x) = 1 (x). One nds the triangularshaped result
D0 () = (2 ||)(
||
1
2 ).
2 4
121
(6.48)
0
As an alternative route to (6.48) we can use (t) = sin(t)
t , D (t) =
2
sin(t
, and as this is symmetric we can compute the real Fourier transform
t
by calculating the cosine transform:

0
2D () =
dt cos(t)D0 (t).
(6.49)
Hence,by direct integration,

0
I () =
dyD0 (y)
y
1
+ 2
2 42
=
log |
| + 2 log |
|.
2
2
4
2
(6.50)
This, with (6.45) shows the distortion of band-like line shapes as = U

grows. However, for > 1.44, the intensity of the continuous line shape drops.
The overall intensity cannot drop, however. The missing intensity is in the
split-o the two-hole resonances, that however come into the theory as 00
singularities. To clarify the situation, a trick is in order. One changes the
above I 0 () to a complex function by setting z = + i, and writing (see
6.42)
i(0) (z) =
z + 2
z
1
z 2 42
log(
)+ 2 log(
) = I 0 ()iD0 (). (6.51)
2
z 2
4
z2
This is equivalent to convolving the line shape with a Lorentzian of width ,

which, incidentally, is always appropriate to describe broadenings [124] which
are present in reality even if they were not considered in the model. Hence
the two-hole singularities show up as Lorentzian peaks.
Two-hole resonances are also observed when core-holes interact with valence holes and the core-valence repulsion is comparable with the valence band
width. For example,in Zn, Cu,Fe, Co, Ni metals there is evidence[191] that
a L2 hole can decay in a normal L2 M45 M45 process or in a Coster-Kronig
L2 L2 L3 M45 M45 M45 M45 process. It is possible to observe the decay
spectrum in coincidence with the L2 photoelectron in APECS (= AugerPhotoelectron Coincidence spectroscopy) and compare with the L3 M45 M45
line shapes. In some of these metals the line shape is much narrower; this suggests that in the L2 L3 M45 process the nal-state M45 hole is localized around
the L3 hole and thus can inuence the 3-hole nal state in the valence band.
The relation between the quasi-atomic states seen in Auger spectroscopy
and the Kanamori[67] paper was understood only later, after the Sawatzky
paper[112] using the Hubbard model with the Kanamori solution. Indeed a
122
D()
-2
D()
-2
Fig. 6.2. For the rectangular band model, the two-hole line shape is triangular
for U = 0 (left panel,dotted), and is shown here broadened with = 0.01. With
increasing U the maximum becomes sharper and shifts to higher , as shown for
U = 1.1 (left panel,solid). At U = 1.8 a well developed split-o state develops
(right panel, dotted). The solid line in the right panel shows the result for U = 3
when most of the intensity is in the two-hole resonance, but a residual band-like
continuum can still be seen.
footnote in Kanamoris paper on magnetism pointed out a narrow band of

solutions of the Hubbard model outside the band continuum which exist for
large enough U. In a periodic solid, the split-o states themselves form a
band. However Kanamoris paper was in a dierent physical context, using
a dierent Hamiltonian than (6.34), and gave no clue about how to observe
these solutions.
Further extensions are presented below and in Chapter 12.
To deal with the Auger CVV spectra of covalently bonded semiconductors,
one has to include overlap eects[135]; the theory has also been extended to
include relativistic eects in intermediate coupling ([201]) and metallic bands
(Sections 12.3.1,12.3.2,12.3.3 ), but is still rather incomplete for strongly correlated nearly half lled bands.
6.2.1 Desorption
Later, Knotek and Feibelman[113] discovered that the two-hole resonances
cause ionic desorption5, e.g. of O+ ions from Ti Oxide surfaces. The primary
5
Desorption is the breaking of an atom-surface chemical bond followed by the

emission of a neutral species or an ion; it can be caused e.g. by an increase of the
temperature, or by ionizing radiation.
6.3 Two Interacting Fermions in a Lattice
123
ionization occurs in the Ti 2p level, but the hole has little chance to Auger
decay via intra-atomic transitions since the Ti ion (nominally, T i4+ ) has
no valence electrons left. So, the electrons must come from the surrounding
closed-shell O ions. One can consider electron transfer to the empty Ti
valence shell and/or direct inter-atomic decay. When two electrons come from
dierent O ions no desorption occurs; indeed, desorption takes several phonon
periods (of the order of 1013 s) and bond healing by hole delocalization
should be much faster. However, there is some chance of two holes in the
same O. In this case, a two-hole resonance forms and lives long enough to
allow the O+ ion to escape. The generalization of the theory to partially lled
bands is discussed in Section 12.3.

The so called Cini-Sawatzky theory has been applied[153] to the N6,7 O4,5 O4,5
Auger spectrum of Au metal. The multiplet structure was interpreted in the
intermediate coupling scheme using free atom experimental data[154], and
using as the only parameter the F0 Slater integral, that is modied with
respect to the free atom by the solid state screening. The top panel (a) in
Figure 6.3 shows the experimental data calculated with a value of U(1 G4 )=3.4
eV (dashed); the solid line in (a) is obtained by shifting the dashed line
to higher binding energy by 1.2 eV. The prole in in excellent agreement
with experiment but its absolute position is wrong by 1.2 eV. On the other
hand, the prole calculated with U(1 G4 )=4.6 eV is shown in b). Now it is in
agreement with the experimental position of the main features but the shape
is in rather poor agreement with experiment. The prole of (c) is obtained
from the one in (b) by arbitrarily increasing the width of the Lorentzian
lifetime broadening to 2.0 eV. It is clear that this does not x any problems.
So, it appears that the Cini-Sawatzky theory involves a systematic error on
the position. A similar shift between the best line shape and the experimental
position was found in an accurate analysis of the Ag M4,5 N4,5 N4,5 line[155],
where it is possible to get a very good agreement with the experimental prole
but the theoretical results must be shifted 2.2 eV to lower binding energy.
In both cases, we traced back the origin of the problem to the extreme
on-site repulsion used in the Cini-Sawatzky model. This prompted the following analysis aiming to generalize the calculation to an arbitrary distance
dependence of the hole-hole interaction.
Consider two electrons (or holes) in a otherwise empty (full) lattice of
N sites, interacting through an arbitrary potential V (). Lets write
|R1 R2 for the determinantal two-body state with a up-spin electron at site
1 and a down-spin one at site 2; the sites belong to a periodic lattice in d
dimensions. Here I report exact results by Verdozzi and the writer[66] on the
two-hole Greens function,
124
Fig. 6.3. Sketch of the Au N6,7 O4,5 O4,5 Auger spectrum (Ref. ([153])) compared
with theoretical proles (see text). The solid line is a drawing of the experimental
prole, on a binding energy scale, where 0 corresponds to both holes from the Fermi
level; the theory used the Slater integral F 0 , or equivalently one of the U values,
as an adjustable parameter. The dotted line is the best shift to the theoretical
line shape obtained with U (1 G4 ) = 3.4eV, but appears to be shifted compared to
experiment. By increasing U one can adjust the line position, but the line shape
agreement is lost (dashed line). To x this problem, one must account for the o-site
interactions (see below).
G(R1 , R2 , R3 , R4 , z) = R1 R2 |
1
|R3 R4
zH
(6.52)
with z = E i0+, as a function of energy E (for two electrons,z = E +i0+ ,

but nothing else changes). The Hamiltonian is:

k n
k +
V ()|R , R + R , R + |. (6.53)
H = H0 + H1 =
k
R,
Here, belongs to the lattice of relative motion, which we call the lattice,
isomorphous to . For V = 0, the problem is trivial because H = H + H ;
then the time evolution operator factors and the non-interacting Greens
function that we denote by a small g is given by
g(R1 , R2 , R3 , R4 , z) = g (1) (R1 , R3 , z) g (1) (R2 , R4 , z)
(6.54)
(1)
where g is the one-particle Greens function and denotes a convolution.

Let us deal with the interacting case. The center-of-mass vector
1
(6.55)
Rcm = R +
2
belongs to a {Rcm } lattice which besides the sites also includes the
midpoints of all sides.
The translational symmetry allows to take R1 as the origin without loss
of generality. The key step to exploit the periodicity is a Fourier transformation: the total crystal momentum Q of the pair is conjugate to Rcm and
is conserved, and one is left with an eective one-body problem over the
lattice to be solved.
125
Fourier Transforming the Center-of-mass Motion

Transforming with xed , and understanding the spin arrows,
i
eiQRcm
e 2 Q iQR
|; Q =
|Rcm , Rcm + =
e
|R, R +
2
2
N
N R
Rcm
(6.56)
where the integration variable was shifted using (6.55). Moreover,
1 iQ[R+ 12 ]
|R, R + =
e
|; Q .
N Q
(6.57)
One can immediately check that this {|; Q } set is a correctly normalized
basis for two particles.
Fourier Transforming the Relative Motion
The trick is
eiQR
eiQR
|R, R + =
(, )|R, R +
N
N
R
R
where we can use (, ) =
1
N
1
N
but

R,
3
2
3
2
eiq
eiq
(6.58)
eiq( )
.
N
So, this becomes

eiQR+iq |R, R + =
R,
ei[Qq]R+iq[R+ ] |R, R + ,
(6.59)
R,

R,R+
and the result is
eiQR
1 iq
|R, R + =
e
|Q q, q ,
N
N q
R
(6.60)
where
|Q q, q =
1
N
ei[Qq]R eiq[R+ ] |R, R + .
(6.61)
R,R+
Substituting (6.60 )in (6.56), one obtains

|, Q =
exp( 2i Q ) iq
e
|Q q, q .
N
q
(6.62)
126
Projected Hamiltonian
In the |, Q basis,
H=
H Q , H Q = H0Q + H1Q ,
(6.63)
where the interaction is diagonal,

H1Q =
V ()|, Q Q, |,
(6.64)
while
H0Q =

,
Q

|, Q W
Q, |;
(6.65)
Q
we determine W
using (6.62): one nds
Q
2 Q( )
W,
= e
i
1 iq( )
e
[(Q q) + (q)].
N q
(6.66)
The Greens Function

Due to the translational symmetry, G(R1 , R2 , R3 , R4 , z) can be computed
as G(0, , R, R + , z); we change basis using (6.57) and write
0, |
1 iQ[R+ 12 ( )] Q
1
G (z),
|R, R + =
e
zH
N
(6.67)
terms of a Q-projected Greens function

GQ
(z) = Q, |
1
|Q, .
z HQ
(6.68)
The non-interacting two-particle function g Q has elements

Q
2 Q[ ]
g,
(z) = e
i

q
eiq[ ]
.
z (Q q) (q)
(6.69)
The operator identity

1
1
1
1
=
+
V
;
Q
Q
Q
zH
z H0
z H0 z H Q
can be iterated:
(6.70)

1
1
1
1
1
1
=
+
V
+
V
z HQ
z H Q z H0Q
z H0Q
z H0Q
z H0Q

1
1
1
1
=
V +V
+
V
.
(6.71)
Q
Q
Q
zH
z H0
z H0
z H0Q
6.4 Quadratic Response Formalism and Spectroscopies
127
Expanding the inverse operator inside the curly brackets and summing the
geometric series, we obtain
1
1
1
1
1
=
+
V
.
1
Q
Q
z HQ
1
V
z H0
z H0
z H0Q
zH Q
(6.72)
We take matrix elements of (6.72) on the |; Q basis and we nd, adopting matrix notation on the lattice, with V diagonal,
5
61
V gQ.
GQ = g Q + g Q 1 V g Q
(6.73)
This is the solution. The size of the matrix that must be inverted depends
on the range of the potential. In the zero-range case of the on-site potential
(Hubbard Model with local interaction U ) we set = , and write
g Q (z) =

q
1
.
z (Q q) (q)
(6.74)
Then, Equation (6.73) is scalar and the solution can be simplied to read:
GQ =
gQ
.
1 U gQ
(6.75)
This result by Kanamori[67] was proposed by Sawatzky [112] in the present

context. In the on-site case, a characteristic feature of the Kanamori theory
is the existence of split-o states for each Q for strong enough interaction U
compared to the band width. This single state can develop to a full discrete
spectrum for long-range V (see Ref. [66].) The results of the analysis in Ref.
[66] showed that the above described shift between the Cini-Sawatzy line
shapes and the experimental prole e.g. in Au and Ag could be accounted
for by a realistic screened hole-hole interaction. Sect. (12.2) below shows the
connection of the Kanamori theory to the Galitzkii self-energy in the low
density case and its relation to diagram methods. Some physical applications
of this theory to electron spectroscopy are presented in Sect. 6.2

Consider a weak, adiabatically switched periodic perturbation
V (t) = f V eit+t ,
(6.76)
with a small positive constant acting on an interacting system with Hamiltonian H; here f is a parameter and V a time-independent operator. We
want the wave function (t) that reduces to the ground state eiEt g for
r . In the interaction picture,
128
I (t) = VI (t)I (t)

t
(6.77)
with the starting condition I (t) g in the remote past. In rst approximation, we write (setting h = 1)
i
I (t) = VI (t)I (t) = f ei[HEi]t V g .

t
(6.78)
Let z = E + + i. The solution with the correct initial conditions is

I (t) = g + f ei[Hz]t
1
V g .
zH
(6.79)
In the Schr
odinger picture,
(t) = eiEt g + f eizt
1
V g .
zH
(6.80)
Photoemission

e
Now let V = mc
i A(xi ) pi (summed over electrons) the operator which

produces photoemission. In second quantization,V = p ,d (p , c)cp cd where
deep a electron is annihilated and a photoelectron created, and (p , c) are the
matrix
elements. We assume that H includes the photoelectron kinetic energy
Tpe = p p np but no post-collisional interaction with the photoelectrons.
The photoelectron current is
Jp =
d
(t)|np | (t) .
dt
(6.81)
in the above approximation, the current is a quadratic response in the perturbation V (which is logical, since it is a d.c. response to an a.c. perturbation.)
Since np g = 0,
np = f 2 e2t g |V
1
1
np
V |g .
H zH
(6.82)
1
1
np
V |g .
z H z H
(6.83)
Hence,
Jp = 2f 2 e2t g |V
Now since [np , H] = 0, we develop:
Jp = 2f 2 e2t

dd
letting 0,
(p, d ) (p, d)g |cd
1
1
cd |g
H zH
(6.84)
Jp = 2f 2
(p, d ) (p, d)g |cd
dd
cd |g
|z p H|2
129
(6.85)
where the photoelectron energy p has appeared (up to now, Tpe was part
of H, but now we use H for the system Hamiltonian without Tpe .) This is
actually
Jp = 2f 2
(p, d ) (p, d)g |cd (E + p H)cd |g
(6.86)
dd
so the result depends on the one-hole density of states matrix dd () of the
system. This formalism generalizes the elementary theory of photoemission,
but oers the possibility of a clear enhancement of the understanding of the
Auger eect.
6.4.1 One-Step Theory of Auger Spectra
Let k label the Auger electron; for the current Jk =
like above
Jk = 2f 2 e2t g |V
d
dt (t)|nk | (t)
we nd
1
1
nk
V |g , (6.87)
(H + W + T ) z (H + W + T )
where now, with a slight change in notation, we made explicit the operator

(6.88)
W =
Whh dk cd ck ch ch + h.c.
producing the Auger transitions; ck and cd create the Auger electron and the
deep electron, respectively, and the annihilation operators create the pair of
nal-state holes; also, T = Tpe + TA the kinetic energies of photoelectrons
and Auger electrons. Using
1
1
1
1
=
+
W
zH W T
zH T
zH T zH W T
and
1
1
1
1
=
+
W
zH W T
zH T
zH W T zH T
1
V |g = 0 one is left with
since nk zHT
1
1
W
nk
H W T z H T
1
1
W
V |g
zH T zH W T
Jk = 2f 2 e2t g |V
(6.89)
130
But nk commutes with H and T, for 0,
1
1
nk
= nk (E + H T ).
H T zH T
So, we reach the general expression of the one-step description of the Auger
eect by Gunnarsson and Sch
onhammer [22]:
1
W nk
z H W T
1
V |g .
(E + H T )W
zH W T
Jk = 2f 2 g |V
(6.90)
The most important terms are diagonal in the deep hole; writing V explicitly
and replacing T by TA + p , where p is the photoelectron energy,
Jk = 2f 2
| (p, d)|2 g |ad
p,d
(E + H T )W
1
W nk
z H W p T A
1
ad |g .
z H W p T A
(6.91)
The W operators in the denominators produce any number of virtual Auger

transitions after the deep hole creation and before the real transition takes
place; this can mix dierent decay channels giving interference eects. For
lled bands, it is reasonable to neglect interband excitations and deep hole
mixing letting (up to a constant)
1
1
V |g V |g g |V
V |g ,
zH W T
zH W T
recovering the Cini-Sawatzky theory.
Auger CVV spectra of transition metals with incomplete bands and the
so-called U<0 phenomenon are discussed in Section 12.3.
6.4.2 Plasmon Gain
Gunnarsson et al. [26] analyzed by their theory the plasmon gain satellites
which are observed experimentally in the KL2 L3 spectra of Na and Mg by a
failure of the two-step model.
Problems
6.1. Verify that j(x)

is the current operator and that (6.2) yields the same
matrix elements as (6.1) and so the two formulations are equivalent.
6.2. Prove Equation 13.136.
Part II
Symmetry in Quantum Physics
7 Group Representations for Physicists
7.1 Abstract Groups

Groups are central to Theoretical Physics, not only as mathematical aids
to solve problems, but above all as conceptual tools. We shall develop the
Group Theory that should be known by Condensed Matter theorists using
a physical language and building on what the readers know already about
quantum mechanical operators. In this Section, however, we need to introduce
some abstract mathematical denitions.
A Group G is a set with an operation or multiplication between any two
elements satisfying the following conditions:
G is closed, i.e. a G, b G = ab G.
The product is associative : a(bc) = (ab)c.
an identity e G such that ea = ae = a, a G.
a G, a1 that is every element has an inverse, such that a1 a =
aa1 = e.
It is not necessary that G be commutative and generally ab

= ba. Commutative Groups are called Abelian. Quantum Mechanical operators do not
generally commute, and we are mainly interested in non-Abelian Groups. The
discrete Groups may have a nite number NG of elements; NG is called the
order of G.
Many Groups of interest have a nite NG , like: the Group C3v of symmetry
operations of an equilateral triangle, the Group S(N ) of permutations of N
objects.
Innite-order Groups are also important. The translations that leave a
Bravais lattice are an Abelian Group; the set of symmetry translations and
rotations of a crystal are is Space Group, and is not Abelian; both are innite discrete Groups. Among the continuous Groups, the most useful are the
Lie Groups , introduced by Sophus Lie in 1870 to discus the symmetries of
dierential equations; by denition, Lie Groups are such that the elements depend smoothly on some parameters. For example, GL(n) (theGeneral Linear
Group in n dimensions) is the set of linear operations xi = j aij xj , where
134
A = aij is suchthat DetA

= 0. The simplest case is GL(2), the set of 2 2
ab
matrices
with adbc
= 0 depends on 4 parameters (such a smooth decd
pendence denes a 4-dimensional manyfold ). The SL(n) Group is a particular
case of GL(n), being the Special Linear Group in n dimensions or unimodu
lar Group in n dimensions and is the set of linear operations xi = j aij xj ,
where A = aij is such that DetA = 1. Other frequently used Groups are: the
rotation Group O(n) of orthogonal transformations, that leave Euclidean distances invariant1, and also of the orthogonal matrices such that AT = A1 .
We shall also need the Group U(n) of the unitary nn matrices such that
A = A1 and the Group SU (n) of special (determinant equal to +1) unitary
transformations.
Let A and B denote two Groups with all the elements dierent, that is,
a A a
/ B, (except the identity, of course.) We also assume that all
the elements of A commute with those of B. This is what happens if the
two Groups have nothing to do with each other, for instance one could do
permutations of 7 objects and the other spin rotations. In such cases it is
often useful to dene a direct product C = A B, which is a Group whose
elements are ab = ba.
If H G is a Group itself it is a subgroup of the Group G. For instance,
GL(2) has a subgroup O(2) which leaves distances invariant; the translations are a subgroup of the Space Group. Another subgroup of GL(2) is
dened such that given two points
(x, y) and (, ) in the plane, the opxy
erations do not change Det
= x y. Writing the transformation

ab
x
ab
x
=
and
=
, one nds the condition
y

cd
y
cd

ab
Det
= 1, so one is left with SL(2). Usually one proceeds the other
cd
way: one knows a Group H, discovers new operations and so builds G.
Let H be a subgroup of G; for A G consider the set
H(a)
= {aha1 , h H}.
Since ah1 a1 ah2 a1 = ah1 h2 a1 , this is a subgroup of G, the conjugate
subgroup with respect to a. Occasionally, it may coincide with H itself (as
a set, not element by element), in which case we write aH = Ha. If a
G, aH = Ha, then H is called invariant subgroup of G, or normal divisor of
G.
in 3d, the rotation operators R() = ei

tum operators are generators of rotations.
L
h
are familiar; the angular momen-
7.2 Point Symmetry in Molecules and Solids
135

In this section I give a brief account of Point Groups with the most frequently
used information; for further details, one can use crystallography books, like
Ref. [24]. The operations of the 32 point Groups are rotations (proper and
improper) and reections. In the Sch
onies notation, which is frequently used
in molecular Physics, proper rotations by an angle 2
n are denoted Cn and
reections by ; improper rotations Sn are products of Cn and , the reection plane being orthogonal to the rotation axis. The molecular axis is one
of those with highest n. A symmetry plane can be vertical (i.e. contain the
molecular axis) or horizontal (i.e. orthogonal to it), and the reections are v
or h accordingly.
7.2.1 Symmetry operators
Let G = {R, S, T, } denote the symmetry Group. Unlike the operators of
observables, that must be Hermitean, those of symmetries are unitary:
R G, R1 = R
(7.1)
in order to preserve normalization: R|R = |R R = 1; this is evident

from the familiar rotation operators

i
(7.2)
R = exp L .
h
The same holds for reections: the z : (x, y, z) (x, y, z) reection is

represented by the unitary matrix diag(1,1,-1). Since R| = | |R =
| ||2 = 1, the eigenvalues are phase factors: = ei , with real .
The rule that eigenstates belonging to dierent eigenvalues are orthogonal
holds for unitary operators as well. Let R| = ei | , R| = ei | .
Then,
R = |R| = ei | ;
(7.3)
on the other hand, taking the complex conjugate
= |R | = |R1 | = ei | ;
R
(7.4)
thus if the phases are dierent the scalar product is 0.

We can represent the operator S by the matrix
D (S) = |S|
(7.5)
on an arbitrary basis; then

S| =
| D (S).
(7.6)
136
Then, for the symmetry RS G

RS| = R
| D (S) =
| D (R)D (S) =
| D (RS)
(7.7)
with
D (RS) =
D (R)D (S).
(7.8)
thus, the matrices are multiplied like the operators and in the same order,
providing a representation of the Group.
In simple cases, all the symmetry operators commute; then we can diagonalize them simultaneously with the Hamiltonian matrix |H| . We can
diagonalize R, then with the new basis diagonalize S, and continue until we
have a new basis labeled with all the symmetry-related quantum numbers.
This breaks up the Hilbert space into several orthogonal subspaces. Then,
we can diagonalize H separately in each subspace. In such cases,we get the
maximum simplication from symmetry without Group Theory. The names
of Groups are just nicknames for the symmetry of the problem.
Blochs Theorem
In band structure calculations one solves

2
p
+ V (x) (x) = (x)
2m
(7.9)
ipti
in a periodic V (x). The translation operators Ti = e h , p = ih where

ti primitive translations of the Bravais lattice, dene an Abelian Group GT .
The periodic boundary conditions TiN = 1 for some large integer N allow
to use a nite, cyclic GT .Let G be a reciprocal lattice vector; by denition,
ei G t = 1 for any lattice translation t . We solve the eigenvalue equation

for all the unitary operators Ti
Ti (x) = (x + ti ) = eii (x)
(7.10)
at a time, by the Bloch Theorem:

Theorem 1. Equation (7.9) is solved by
(x) = k (x) = eikx uk (x)
(7.11)
with
uk (x) = uk (x + ti )
periodic; i = k ti with the wave vector k = G

N .
(7.12)
137
Indeed, the conditions TiN = 1 require that eN kti = 1, so N k must be a

reciprocal lattice vector. On the Bloch basis one has a sub-problem for each
k which reduces to seeking periodic solutions of
[
(p + k)2
+ V (x)]uk (x) = k uk (x)
2m
(7.13)
Water Molecule
Suppose want to calculate the molecular orbitals of H2 O in a LCAO model
neglecting overlap. Putting the molecule on the xz plane, with z as the molecular axis, we see that it belongs to the C2v point Group, with the operations
C2 : (x, y, z) (x, y, z); (xz) : (x, y, z) (x, y, z); (yz) : (x, y, z)
(x, y, z) We can form the Multiplication Table:
E
C2 (xz) (yz)
C2
E (yz) (xz)
(xz) (yz) E
C2
(yz) (xz) C2
E
the Group is Abelian and the square of each operation is the identity E.
Therefore the eigenstates of H will be even or odd under any of the operations.
2 | ,
From an arbitrary basis {| } we can generate 2 bases |C2 = 1C
2
then 4 bases |C2 (xz) =
1(xz) 1C2
| ,
2
2
and 8 bases |C2 (xz)
1yz 1(xz) 1C2
| .
2
2
2
One can diagonalize H separately on these

(yz) =
bases. Here is a representation of the Group. If we replace every operator by
1, the multiplication Table is trivially veried; -1 choices are also allowed, as
follows:
C2v
A1
A2
B1
B2
I
1
1
1
1
C2 xz
1 1
1 1
1 1
1 1
yz
1
1
1
1
g=4
z
xy, Rz
x, Ry
y, Rx
Such are the possible symmetry types of the solutions,that have conventional names shown in the rst column.
The C3v Group (NH3 )
The C3v (or 3m) Group is the symmetry Group of an equilateral triangle (or
the Group S(3) of the permutations of 3 objects). Let the vertices be labeled
(a,b,c). The operations are a C3 rotation, its square (or inverse, which is the
same) and three vertical planes a , b , c through the center and the vertices.
The sense of rotation is arbitrary and we may choose C3 as the operation
(a, b, c) (c, a, b); clearly a (a, b, c) (a, c, b). Next, we need a convention
138
about the multiplication of two operation. a C3 (a, b, c) = a (c, a, b) The commonly adopted one states that the result is (c, b, a), that is, a keeps the rst
entry xed, although the rst operation sent a elsewhere. In other terms, it is
the position of symmetry elements that matters, while the label can change.
In this way, one notices that a C3 = b , but C3 a = c , thus the Group
is not Abelian. All the information about the abstract Group is in the multiplication table.
E C3 C32 a b c
C3 C32 E c a b
C32 E C3 b c a
(7.14)
a b c E C3 C32
2
b c a C3 E C3
c a b C3 C32 E
The rearrangement theorem holds:
Theorem 2. Each line and each column contain all R G.
This follows from the denition of Group. In any line or column there are g
elements, and all are distinct since for instance C3 R = C3 S R = S. Every
operation does a permutation of the vertices. In C3v the converse is also true,
so C3v is isomorphous to S(3). Already in C4v the 8 operations are fewer
than the 24 permutations of 4 objects. The Cayley theorem states that
Theorem 3. Any group of order NG is isomorphous to a subgroup of S(NG ).
Z3 = {E, C3 , C32 } C3v is an invariant Abelian subgroup. A Group G is
called simple if it does not have invariant subgroups. So, Z3 is simple, but
C3v is not. A Group G is called semi-simple if it does not have abelian
invariant subgroups. So, Z3 is semi-simple, but C3v is not.
It is interesting to nd representations of Z3 using the multiplication table
above. Besides the trivial one, with all the operators represented by 1 (the
2i
A1 representation below in Table 7.1 ) one can represent C3 by = e 3
and C32 by the third cubic root of 1, namely ; in a third representation,
C3 , C32 . Thus, there are two complex conjugate representations.
The p orbitals of an atom centered at the origin behave dierently under Z3 : the z orbital, that we assume parallel to the rotation axis, is not
transformed and belongs to A1 . The combination x + iy is multiplied by
and x iy is multiplied by ; thus the (x,y) pair is a basis for the conjugate
representations. This is reported in the last column of Table 7.1 )
The Quotient Group
Let H be a subgroup of order NH of G and g G. the sets
Hg = {hg, h H}, gH = {gh, h H}
are respectively right and left cosets. Both have exactly NH elements; for
instance, h1 g = h2 g multiplied on the right by g 1 becomes h1 = h2 .

C3 C32
C3 = Z3 I
A1
1 1 1
1
=e
z
139
2i
3
(x, y)
Table 7.1. The Z3 Group and its representations.
For example, in C3v , using Z3 and a one forms the right coset
{E, C3 , C32 }a = {a , b , c }.
Using b or c one nds the same coset, while a rotation gives Z3 . Actually,
we have split C3v in two subsets of 3 elements each as follows:
C3v = Z3 + Z3 a .
(7.15)
Let us generalize these ndings to any G. If we form a right coset using a

subgroup H G of order NH and g H we only get elements hg H. On
the other hand, if g1
/ H the coset Hg1 is totally disjoint from H since for
/ H; otherwise from hg1 = h H we
all its NH elements we conclude hg1
could deduce g1 H. If G is the union of H and Hg, we may conclude that
NG = 2NH . Otherwise, we form the coset Hg2 with g2
/ H, g2
/ Hg; now
Hg2 is disjoint from H and from Hg1 (since hg2 = h g1 g2 = h1 h g1
Hg1 ). We go on until G has been totally partitioned:

Hgj .
(7.16)
G=
j
This proves the famous Lagrange Theorem:

Theorem 4. The order of any subgroup of G is a divisor of NG .
Let H be an invariant subgroup of G. All its right cosets are equal to the
left cosets, i.e. aH = Ha, a. It is useful to dene some multiplications of
sets. We may say that aH H = aH, meaning that for any element of the
coset ahh = ah is again in aH.
For example,Z3 is an invariant subgroup of C3v and the left coset a Z3 =
{a , b , c } is the set of reections; multiplying by the rotations in Z3 we
get nothing more and {a , b , c } Z3 = {a , b , c } . In a similar way, we
can dene an abstract set multiplication, treating the cosets as elements; for
instance aH bH is the set of all the elements ah1 bh2 , h1 H, h2 H; it is
understood that every element may occur more than once and we remove all
duplicates. Since H is invariant, h1 b = bh3 for some h3 H; so, aH bH =
abH and aH a1 H = H. Thus, the cosets are elements of the quotient
Group G/H, where H is the identity. Broadly speaking C3v /Z3 reduces to
two elements, the rotations and the reections; rotation rotation= rotation,
rotation reection= reection, and reection reection= rotation. This
is a powerful synthesis of the multiplication table of C3v .
140
7.2.2 Dirac characters and Irreducible Representations

When G is not Abelian, the elementary method allows diagonalizing H simultaneously with one or a few R G. In the C3v example, one can choose
C3 (and C32 , but this adds nothing) or one reection. In this way, one is neglecting most of the symmetry related information. Although this does not
cause errors, the use of Group Theory is much more rewarding.
We must nd functions of the operators R G such that they commute
with all G elements (besides H; but this we take for granted, by denition of
a symmetry Group). We could think about powers or products, but since G
is closed the most general function is a linear combination. For any abstract
Group, the linearcombinations of the elements constitute the Group algebra.
NG
T should be a symmetric object. This is a good idea,
Intuitively, = T G
NG
NG
G
since, by the rearrangement theorem (2) N
T G T S =
T G T = S
T G T,
so commutes with all G.
Actually, we can do much better than that; intuitively, the sum of all
rotations of a given angle is already a symmetric object, the sum of all reections another one. To proceed, we need to introduce the class concept, that
allows to classify together operations that do essentially the same thing: such
operators are called conjugated2 . Mathematically, two operators A and B are
conjugated
(7.17)
A " B = {X G : A = X 1 BX}
which is tantamount to say that X is a 3change 4of reference that converts
B into A. Thus, C3v has 3 classes {E} , C3 , C32 , {a , b , c } . In Abelian
Groups each element is conjugated to itself. The Identity E is trivially a
class. For every class C, with nC elements, we dene Diracs character
C =
nC
T,
(7.18)
T C
that commutes
all G, that is X G, C = X 1 C X. Indeed,
nwith
C
1
1
X C X =
T X has nC terms conjugated with T and such
T C X
terms are all dierent because X 1 T X = X 1 T X T = T . So, we have a
quantum number for each class.
For C3v , the Dirac characters are
E = E, R = C3 + C32 , = a + b + c .
These are simultaneously diagonal and their eigenvalues E = 1, R , (except the rst, which is trivial) are useful wavefunction labels. They are not
independent, but occur in combinations. Using the multiplication table,
2
this term is rather strange, since an element can be conjugated to any number
of elements.

2
R
= 2 + R ;
2
R
(7.19)

hence
141
= 2 + R R =
2
1
2 = 3 + 3R 2 = 3 + 3R .
(7.20)
(7.21)
The allowed combinations correspond to the possible symmetry types allowed

by C3v ; they are called Irreducible Representations or Irreps for short:
Irrep R
A1 2
A2 2
E -1
3
-3
0
(7.22)
We can label H eigenstates by the irreps; remarkably, they are never

mixed by symmetry operations since
X G, C | = C | C X| = C X| .
(7.23)
A1 is a trivial representation with all operators represented by 1 and Dirac

characters equal to nC for each class. This totalsymmetric representation
exists for any Group, since any multiplication table is satised. A2 is similar
but odd for reections. In all non-Abelian Groups there are irreps that require
matrices, rather than numbers, to represent operators, and the irrep E above
is an example. Just consider transforming a point (x, y) in the plane ((px , py )
y
1
Fig. 7.1. The chosen geometry, with the 1 reection that changes x into -x.
orbitals or any pair of functions transforming like (x,y) will do the same):

10
c s
1 0
D(E) =
; D(C3 ) =
; D(a ) =
(7.24)
01
s c
0 1
1
2
3
with c = cos( 2
3 ) = 2 , s = sin( 3 ) = 2 . These are the generators, i.e., the
others can be obtained by multiplication:

c s
c s
; D(b ) = D(a )D(C3 ) =
;
D(C32 ) = D(C3 )2 =
s c
s c

c s
D(c ) = D(a )D(C32 ) =
.
(7.25)
s c
142
The multiplication table is obeyed.

Howcan the Dirac

characters commute
1 0
00
with every D? One nds R =
, =
. The size m of the
0 1
00
representative matrices of the Irreps (m=1 for A1 and A2 , m=2 for E) grows
with the size and complexity of the Group. In the 32 point Groups one nds
at most vector or pseudovector representations having m=3, but much larger
sizes occur in more advanced applications.
The whole Hilbert space is partitioned into subspaces labelled by the
irreps; besides the symmetry operators, H does not have matrix elements between states with dierent good quantum numbers and the diagonalization
problem is broken. Diagonalizing H within each subspace one nds eigenstates carrying the Irrep label. Consider a pair of eigenfunctions (x , y )
belonging to the Irrep E; Hx = x x , Hy = y y . However, x = y = ,
because every operation S G maps x into a linear combination Sx =
x + y , but since [S, H] = 0, the transformed function Sx must belong
to the same eigenvalue as before. The same reasoning can be repeated for the
matrices. To sum up, the matrices of H and those of Dirac characters have
the remarkable property of being constants, that is, identity matrices I(mm
multiplied by numbers:
H = Imm , C = C Imm
(7.26)

x
x
By a unitary transformation
= U
we can change basis
y
y
and matrices within the E irrep. Let U be a rotation: then R, D(R)
U D(R)U 1 is consistent with the multiplication table and does not modify
the Dirac characters. Thus, one of the non-trivial D(R) matrices can always
be chosen diagonal. Up to this unitary transformation, the Irreps with the
same characters are to be identied. For instance, we may wish
+ to have
, a new
in which D(C
3 ) is diagonal. Picking U =
representation D
1 i
2
2
i
1
2
2
2i/3
3 ) = U D(C3 )U = diag(e
is unitary (U U =unit matrix), D(C
, which
2i/3
, e
)
0
1
2
3 ) is also diagonal of course, but U D(a ) U =

is diagonal; D(C
1 0
and the other reections are o-diagonal.
By joining the basis of an Irreps of size m1 with basis of an Irreps of
size m2 we obtain a basis of size m1 + m2 with D matrix in block form. By
unitary transformations we can generate m1 + m2 representatives which are
no longer in block form. In general, by taking linear combinations of several
bases of Irreps, we can form bigger representatives with no simple pattern
of vanishing elements. These are called reducible representations to imply
that the intelligent thing to do is just the reverse of this process. Starting
with an arbitrary basis that yield big and useless D matrices one would like
to nd linear transformations that put the D matrices in block form, and
allow separating the basis in symmetry-adapted parts; when the blocks can
143
no longer be reduced in size, the Irreps are obtained. Therefore we cannot

put all the matrices of an irrep in the same block form.
Traces and Characters of an Irrep
Consider a basis and the representative matrices of a m-dimensional irrep i
(in this paragraph I omit the index i for short). The trace

D (R), R C
(7.27)
(C) = T rD(R) =
is the same for all D matrices in the given class C, since the Tr operation
is invariant under unitary transformations like conjugation. (C) is called
character, and is related to Diracs character. The matrices of the C are
constants3 , hence, recalling the denition of C
T rC = mC = nC (C).
Thus,
(C) =
m
C .
nC
(7.28)
(7.29)
The characters (C) are very useful, as we shall see, and are tabulated for
the point Groups (See Appendix II; note that these tables are square (the
number of irreps is equal to the number of classes).). In particular T rE = m
can be read o the Tables.
Actually, the tables were computed (and can be found for new Groups
when necessary) from the multiplication Table; one rst deduces the classes
and c as in the above example. One does not know m a priori but shall
derive below
NG
(7.30)
m2 = 2 .
C
C nC
With this result, the reader is already able to calculate the character tables.
7.2.3 Schurs lemma
We have shown above that the matrices of Dirac characters and of H within
an irrep are multiples of the identity matrix. Due to the importance of this
crucial point Ill present a second proof, the traditional one based on Shurs
lemma. The two arguments enlighten dierent facets of the problem; it will
be apparent below that this is just a matter of algebra.
Lemma 1. Any m m matrix M commuting with all the representative matrices D(Ri ), 1 = 1, NG ) of an m-dimensional Irrep must be proportional
to the Identity matrix Imm .
3
H is also constant and T rH = m.
144
We rst prove the lemma for a 2 2 diagonal matrix, then generalize to any
matrix of any size.
Proof for a diagonal matrix
a11 a12
a21 a22

0
a12 (d2 d1 )
00
=
.
[A, ] =
0
00
a21 (d2 d1 )
For 2 2 matrices, we let M = = diag(d1 , d2 ), with A =

, and
Now, if d2 were
= d1 , this would require a diagonal A. Since one cannot
diagonalize simultaneously all the D(Ri ), 1 = 1, NG ), we conclude that
d2 = d1 .
In the 33 case, we let M = = diag(d1 , d2 , d3 ), and the same reasoning
leads to
0
a12 (d2 d1 ) a13 (d3 d1 )
0
a23 (d3 d2 ) .
[A, ] = a21 (d1 d2 )
a31 (d1 d3 ) a32 (d2 d3 )
0
If all the di are dierent, this vanishes only for diagonal A. If d1 is dierent
from the other diagonal elements, the a11 element must be one block, separated from the rest, which cannot be true for all the representative matrices
in an irrep. This is formally generalized formally to matrices of any size with
the conclusion that = Imm for some .
Proof for a Hermitean M
Assuming M hermitean we know that by a unitary transformation U it can
be diagonalized U M U = = Imm and actually this implies that
M = Imm in any basis.
Proof for any M
We can do without the assumption that M is Hermitean. Indeed, H1 =
M + M and H2 = i(M M ) are Hermitean at any rate. Now, we
show that M also commutes, for, by taking the Hermitean conjugate of
D(Ri )M = M D(Ri ), one nds M D(Ri ) = D (Ri )M dag . Now we multiply on the left and on the right by D(Ri ) and since it is unitary we nd
D(Ri )M = M D(Ri ). Then H1 and H2 also commute, and are constants.
Therefore, M = 12 (H1 iH2 ) = Imm ,
q.e.d.
7.2.4 Continuous Groups
Many Groups of fundamental importance in Physics are Lie Groups. They
are continuous (elements can be labeled by parameters) and continuously
7.3 Accidental degeneracy and hidden symmetries
145
connected ( for every pair of elements a continuous path in parameter space

can be found that joins them). Moreover, the parameters of the products are
C 1 functions of those of the factors.
A compact Lie Group has all parameters that vary over a closed interval;
the Lorentz Group and the Group of all translations are noncompact Lie
Groups, while the rotations are a compact Lie Group.
The phase factors ei constitute the Group U (1). The SO(3) Group of
rotations in 3d is a continuous Group. An element may be represented as a
vector directed along the axis and with length equal to the angle of (say,
counterclockwise) rotation ; this corresponds to a sphere of radius where,
however, each point of the surface is equivalent to the opposite one. All the
rotations with the same || belong to the same class. the angular momentum
operator L is the generator of innitesimal rotations. For integer L one nds

2L+1 spherical harmonics YLM (, ) that are simultaneous eigenvectors of
L2 and Lz . Only the harmonics of a given L mix under rotations; they are
the basis of an irrep.
For angular momentum J = 12 , the rotation operators (7.2) build a double
valued representation, since R = R
; they make up SU(2), the
+ 2
covering Group of O(3) corresponding to a sphere of radius 2. The Wigner

matrices are dened by
R |Jm =
J
J
|Jm Dm
m ().
(7.31)
m =J
The SU(3) Group is a paradigm of Particle Physics. SO(4) is the rotation

group in 4d;
are in the planes ij with i and j = 1 to 4. Therefore
rotations

4
there are
= 6 generators A1 , A2 , A3 , B1 , B2 , B3 ; it can be shown that
2
[Ai , Aj ] = iijk Ak ,[Bi , Bj ] = iijk Bk and [Ai , Bj ] = 0.

Let H be a Hamiltonian of C3v symmetry; if one succeeds to separate the
Hilbert space according to the irreps, the problem breaks down into independent subproblems of A1 , A2 and E symmetry; in the E subspace one can
decide to have a diagonal; if a is the reection that sends x to x one has
(x , y ) pairs that are not mixed by H since they form a m = 2 subspace
where H is constant. If one diagonalizes H in the x subspace nds all the x ;
from each x one then nds the corresponding y without need of further
diagonalizations; one could apply any o-diagonal operator like C3 and then
orthogonalize to x . In general, the knowledge of one eigenfunction is enough
to determine all the m degenerate eigenfunctions belonging to the same irrep.
Sometimes one nds unexpected degeneracies. Eigenfunctions belonging
to non-degenerate irreps are found to be degenerate; or the actual degeneracy
146
d is higher than m. The most popular case occurs in the Schrodinger theory
of the H atom, when energy depends only on the principal quantum number
n and orbitals have a n2 fold degeneracy, while the spherical symmetry only
allows the energy to be independent on magnetic quantum number. In such
cases, one cannot produce the 2p orbitals by rotating the 2s one. In such
cases, one speaks of accidental degeneracy.
It is possible that by accident two states unrelated by symmetry come so
close in energy to appear degenerate in low resolution experiment; however
a mathematically exact degeneracy with no symmetry reason is miraculous.
Simply, one was unaware of using a Subgroup of the actual Group, because
some symmetry had still to be discovered, as in the following examples.
Hydrogen Atom
The n2 fold degeneracy of the non-relativistic H atom Hamiltonian
H=
k
p2
2m r
is explained by a dynamical symmetry, specic of the
(7.32)
1
r
potential.
The Classical Runge-Lenz Vector
The fact that the force F = k rr3 is central explains the conservation of
angular momentum: ddtL =

v
p +
r F = 0. We take the z axis along L ,
and the motion occurs in the z = 0 plane. The special form of the potential
entails the conservation of the Runge-Lenz vector
To see that, using
d 1
dt r
r
p L
k .
R=
m
r
= rr3v , one works out
r2
d
v
r (
r
v)
r
=
.
dt r
r3
(7.33)
(7.34)
L
The x component of the numerator is (x2 + y 2 )vx x(xvx + yvy ) = m
y.
Developing in this way, one nds
L
d
r
=
(y, x, 0).
dt r
mr3
(7.35)
The conservation of R follows comparing this with
p L
L
d
kL
= (Fy , Fx , 0) =
(y, x, 0).
dt m
m
mr3
(7.36)
147
During the motion,

p L is in the xy plane. At perihelion,
p
r
p L

r R
r . Since it is conserved, R is pinned at the aphelion-perihelion
4
direction and this is why the orbits are closed5 .
Quantum Runge-Lenz Vector
The quantum Runge-Lenz vector

p L L
p
r
R =
k .
2m
r
(7.37)
commutes with H; the calculation can be carried out like above, with
replaced by hi [H, A] . Hence,
dA
dt
HLM = ELM H R LM = E R LM ;
but one can check that R does not commute with L2 , so R LM does not
belong to L and the energy must be L independent. More details and the relation to the O(4) Group worked out by Pauli can be found on Schis book[25].
In the relativistic case, the Dirac-Coulomb Hamiltonian HDC commutes with
the Biedernharn-Johnson-Lippman (BJL) pseudoscalar operator

i
Ze2
r
2
(7.38)
K
(H
mc
)
5
DC
2
mc
c
r

where K = L + h
, =
, = 4 , 5 = 1 2 3 4 . This is
the reason why the levels do not depend on the sign of K and for instance
the pairs (2s1/2 , 2p1/2 ), (3s1/2 , 3p1/2 ), (3p3/2 , 2d3/2 ) are degenerate in Diracs
theory.
B=
Appearence is Deceptive: Simple Example

The square tight-binding cluster described by
0110
1 0 0 1
h=
1 0 0 1
0110
(7.39)
has the following spectrum:

2
For circular orbits, using mv

= rk2 one fonds that R vanishes.
r
5
The 2d isotropic harmonic oscillator also does; the energy is conserved along
both axes, so there is an extra conserved quantity also in that case.
148
-2 2 = ( 12 , 21 , 12 , 12 )
0 0,1 = ( 12 , 0, 0, 12 )
0 0,2 = (0, 12 , 12 , 0)
2
2 = ( 12 , 12 , 12 , 12 )
and the degeneracy is expected because the irrep E of C4v has to be represented. The eigenvalues come in pairs (||) since this graph is bipartite.
a)
3
2
b)
Fig. 7.2. llustrative example of dynamical symmetry. a) square tight-binding cluster b) deformed cluster with heavy lines denote doubled matrix elements. The symmetry Group of b) is isomorphic to the Group of the square.
Now we deform the square, by changing the hopping integrals to and

from an atom: letsay, we double those connecting site 3 to 1 and 4. Now, the
Hamiltonian reads
0120
= 1 0 0 1.
h
(7.40)
2 0 0 2
0120
The deformed cluster has much less geometrical symmetry but as much dynamical symmetry as before. Was the degeneracy removed? No. The rst
and last lines are equal and the other two are proportional, thus two zero
has the following spectrum:
eigenvalues are still there. Indeed, h
1
1
10 ( 2 , 10 , 25 , 12 )
0
( 12 , 0, 0, 12 )
0
(0, 25 , 15 , 0)

10 ( 12 , 110 , 25 , 12 )
There is accidental degeneracy. One symmetry element is the reection
7.4 Great Orthogonality Theorem (GOT)
0
0
=
0
1
149
001
1 0 0
0 1 0
000
with 0,1 = 0,1 ; 0,2 = 0,1 . To explain this degeneracy we want a

non-Abelian Group. Indeed, the 90 degree rotation is broken, but we seek a
= h. We seek it such that
convenient generalization S, with S T = S 1 , S T hS
opposite sites are not coupled ( S14 = S23 = 0) and S12 = S24 , S13 = S34 .
One solution is the generalized 900 rotation
0010
0 1 3 0
1 0 0 0
3 0 0 1
S=
(7.41)
0 0 0 1 S = 10 1 0 0 3
0100
0 3 1 0
Note that S 4 = 1 and S does not produce a permutation of sites but
is a genuine generalization. Moreover S mixes 0,1 and 0,2 , does not commute with and explains the degeneracy. The deformed problem has a lower
geometrical symmetry, but actually is still C4v symmetric because of a hidden dynamical symmetry. More dynamical symmetries will be discussed in
Sect.17.0.6.

The theorem states that the D matrices are orthogonal over all their indices.
Theorem 5. (Great Orthogonality Theorem)
Let
be the element of the matrix of the irrep i representing the
element R G. Let NG be the order of the Group and let mi be the dimension
of the irrep i. Then
(i)
D (R)
(i)
RG
(j)
D (R) D (R) =
NG
mi ij .
(7.42)
The GOT is the central result of Group representation theory, and I prepare
the proof by a few remarks.
Remark 1: I am going to present the proof for discrete Groups, however
the GOT extends to continuous ones. The following orthogonality holds for
the Wigner matrices[23]

8 2
J
a,a b,b J,J
()DaJ b () =
(7.43)
dDab
2J + 1

2

2
where , , are Euler angles and d = 0 d 0 sin d 0 d.
150
Remark 2: Let us resume the degenerate (m = 2) irrep of C3v and the

matrices

E
10
01
+
C3
1
3
2
2
3
1
2
2
,+
C32
12
23
3
2
12
c
, a + b , +
,
1
3
1
3
1 0
2
2
2
2
3
3
0 1
1
1
2
(7.44)
Note that C32 is the inverse of C3 and its matrix is D(C3 )T . More generally,
the operators are unitary and
(i)
(i)
D
(R) = D
(R1 ).
(7.45)
Remark 3: Let i be basis function belonging to the -th component of

the irrep i and O represent a total-symmetric operator, that is,
[O, R] = 0, R G.
Theorem 6. The matrix elements obey the following rule:
i |O|j = O(i)ij .
(7.46)
For i
= j, i and j are orthogonal because they belong to dierent eigenvalues of at least some c . The invariant operator O commutes with the
Dirac characters, so O| j has the same characters as | j . For i = j the
matrix {O } represents an operator that commutes with everything, so it
must commute with all the D matrices; so Schurs lemma applies.
For example, a spherically symmetric potential has vanishing matrix elements between states of dierent L and within L it has vanishing matrix
elements between states of dierent ML ; the diagonal matrix elements are
independent of ML and depend on L.
Remark 4:
Theorem 7. Consider an arbitrary operator .
O=
R1 R
(7.47)
RG
is an invariant operator.
This follows
immediately from

the rearrangement theorem: T G, OT =
T T 1 RG R1 RT = T RG (T R)1 (T R) = T O.
Remark 5:
151
(example of GOT)
Let us form 6-component lists with the elements of the D matrices; if we
treat them like vectors and compute the norms we nd 6 for the irreps with
m=1 and 3 for those with m=2.
element
11 of irrep
12 of irrep
21 of irrep
22 of irrep
irrep A1
irrep A2
list
1 1
E (1,
21 ,
12 , 1,
, )
2 2
3
3
3
3
E (0,
,
,
0,
,
)
2 2
2
2
3
3
3
3
E (0, 2 , 2 , 0, 2 , 2 )
E (1, 21 , 12 , 1, 21 , 21 )
(1, 1, 1, 1, 1, 1)
(1, 1, 1, 1, 1, 1)
of squares
3
3
3
3
6
6
Moreover these vectors are all orthogonal.

Proof of GOT
We insert (7.47) into (7.46)but do not specify yet :

i |R1 R|j = O(i)ij .
(7.48)
RG
Letting R operate with

R|j =
(j)
|j |D
(R),
i |R =
(i)
i |D
(R) ,
we nd
(i) ||(j) =
(j)
(i)
i ||j D
(R)D
(R) = O(i)ij .
(7.49)
RG

To prove the GOT we must eliminate i ||j . Now we take advantage
of the total freedom to choose as we like. If we assume that for some pair
of components, and , say ,
= |i j |,
(7.50)
then i ||j = , we end up with

RG
(j)
(i)
D (R)D
(R) = O (i)ij .
(7.51)
152
The matrix element O (i) could in principle have acquired a dependence

on the and indices, but we have proven the orthogonality on the second
indices. The orthogonality on the rst indices and is obtained by taking
conjugate, and noting that this exchanges the indices and

the complex
=
.
Thus,
1
R
R
(j)
(i)
D (R)D
(R) = O(i)ij .
(7.52)
RG
To determine O(i) we set i = j, = and = . We get

(i)
(i)
(i)
(i)
D
(R) D
(R) = O(i) =
D
(R1 )D
(R).
RG
(7.53)
RG
Summing over from 1 to mi ,

(i)
mi O(i) =
D
(E) =
1 = NG .
RG
(7.54)
RG
This completes the proof.

The GOT is an orthogonality relation between vectors having NG com(i)
ponents, one for each component of D (R); each irrep contributes m2i components, and in order that they be orthogonal

m2i NG .
(7.55)
i
This is an important restriction on the number and on the dimensions of

irreps. We prove below Burnsides theorem

m2i = NG .
(7.56)
i
7.5 Little Orthogonality Theorem (LOT)

From GOT we may obtain an orthogonality theorem between characters
which is called LOT. The diagonal elements obey
(i)
(j)
D
(R) D
(R) =
RG
NG
ij .
mi
(7.57)
Summing on , one obtains the LOT

RG
In particular,
(i) (R) (j) (R) = NG ij .
(7.58)
7.5 Little Orthogonality Theorem (LOT)
|(i) (R)|2 = NG ,
153
(7.59)
RG
which gives a method to verify that a given representation is indeed irreducible; if we mix two irreps i and j with coecients
ni and nj , then
5
6
(R) = ni (i) + nj (j) and the r.h.s. becomes NG n2i + n2j + 2ni nj ij .
Rewriting the LOT in terms of classes,

nC |(i) (C)|2 = NG ,
(7.60)
C
and inserting (i) (C) =
mi C
nC
we obtain
7
8
8 NG
mi = 9 2 .
C
(7.61)
C
nC
A reducible representation can be reduced to block form and the block of the
i-th irrep appears ni times along the diagonal; on any basis, the trace is

nj (j) (R).
(R) =
j
Now multiply by (i) (R) and sum over R:

(i) (R) (R) =
nj
(i) (R) (j) (R)
j
which implies
1 (i)
(R) (R).
NG
(7.62)
1
nC (i) (C) (C).
NG
(7.63)
ni =
Also,
ni =
(i)
(j)
The LOT
C (C) (C) = NG ij , so the vectors with compo becomes:
nents NG (i) (C) are also orthogonal. They have component for each class
and there is one of them for each irrep. Thus,the number of irreps does not
exceed the number of classes. Actually one can prove that they are equal, that
is, the character tables must be square [156]. This is related to the second
character orthogonality theorem

i
(i) (C)(i) (C ) =
also demonstrated in Ref.[156].
NG

nC CC ,
(7.64)
154
7.6 Projection operators

From any f (x) one can obtain functions of well-dened parity using the inversion operator and its eigenvalues to write f () (x) = f (x) f (x). The GOT
generalizes this trick converting an arbitrary basis set into symmetry-adapted
basis functions without any need to diagonalize any matrix. Multiply (7.42)

(i)
NG
(j)
j
RG D (R) D (R) = mi ij by and sum over , using the fact

(j)
that D (R)j = Rj :
that is,
(j)
(i)
D
(R) D (R) =
RG
i
D
(R) Rj =
mi (i)
D (R) Rj = ij j .
NG
NG
ij
mi
(7.65)
Hence
i
=
P
mi (i)
D (R) R
NG
(7.66)
i
is a generalized projection operator6such that P
= |i i |. In other terms,
i
P
j = ij i .
(7.67)
i
The diagonal P
operators are all what we want, and require only the

diagonal elements of the D matrices. Let = j c(j, )j belong to a
reducible representation:
i
= c(i, )i
(7.68)
P
is projected on i , unless c(i, ) = 0 (if the choice was unlucky never mind,
we shall succeed with another of the reducible set). To transform the {}
set to the new {j } basis we just need to know how every R G operates on
(i)
(i)
the old set and the diagonal elements of the D (R) matrices. The D (R)
matrices are properties of the abstract Group,and we know them in principle:
they are mi mi matrices and having xed which operators R we can and wish
to have diagonal, we have enough relations from the multiplication table to
build them; in the most common problems this is just a matter of geometry to
determine how the components of a point or other simple functions transform.
Moreover, the matrices are not really needed to project to the new basis.
Indeed,

i
P
=
c(i, )i .
(7.69)
In the combination on the r.h.s. all functions belong to the irrep i, so the
l.h.s. is the projection operator,
6
i
i
The name is deserved since the diagonal terms are such that (P
)2 = P
.
7.7 Regular representation
P (i) =
i
P
=
i (R) R.
155
(7.70)
The characters are enough to build P (i) and to obtain P (i) which belongs to
irrep i. Repeating this with enough , one builds a basis for the irrep. These
bases are typically small sets, where one can easily orthonormalize; usually
one readily performs the unitary transformations needed to have as many
diagonal operators as possible, in order to increase the number of quantum
numbers.
7.7 Regular representation

The next theorem has a very elegant proof:
Theorem 8.
m2i = NG .
(7.71)
Recall the rearrangement theorem: left-multiplying the group elements by

one of them produces a permutation. In any Group of order NG = g with
elements R1 Rg , we associate to each element a vector
R1 (1, 0, 0, 0, 0, 0), R2 (0, 1, 0, 0, 0, 0),
and a g g matrix that does the permutation. Thus, for each R G there
is a vector and a matrix; what matters is the Multiplication Table, and the
vectors are just a way to get the matrices. Let us rewrite the multiplication
Table of C3v .
E C3 C32 a b c
C3 C32 E c a b
C32 E C3 b c a
a b c E C3 C32
b c a C32 E C3
c a b C3 C32 E
Let us denote the operations R1 R6 with R1 = E, R2 = C3 , and so on; the
eect of the left-multiplication by C3 is 1 2, 2 3, 3 1, 4 6, 5
4, 6 5; the matrix is
001000
1 0 0 0 0 0
0 1 0 0 0 0
,
D(C3 ) =
0 0 0 0 1 0
0 0 0 0 0 1
000100
156
while D(E) = Diag(1, 1, 1, 1, 1, 1). This regular representation exists for any
G; (E) = NG while for all the other classes = 0 since no other element
can have 1 on the diagonal (RX =
X R = E ) . Irrep i is contained in
the regular representation ni = N1G R i (R) (R) = N1G i (E) (E) = mi
times. Reducing the representation, we shall have on the diagonal mi times
the mi mi block, and the length of the diagonal is NG . This proves the
Burnside theorem.
Problems
7.2. Build the character table of C4v .
7.3. Build the projection operator for the irrep E, componet y, of C4v .
8 Simpler Uses of Group Theory
8.1 Molecular Orbitals

To exemplify Group theory methods, I present LCAO calculations of molecular orbitals for simple molecules starting with a minimal basis set (only
the atomic orbitals which are most directly involved in covalent bonding)neglecting overlap.
8.1.1 Molecular Orbitals of N H3
y
2
1
x
3
Fig. 8.1. The H atoms of ammonia are at the vertices of an equilateral triangle.
N H3 belongs to the C3v Group (see the character table in Appendix). A

minimal basis set comprises 2s, 2px, 2py , 2pz for N and the ground state orbitals s1 , s2 , s3 for the H atoms. The geometry is sketched in Figure 8.1.1; the
N atom (not shown) is above the origin and is not shifted by any operations.
Therefore 2s and 2pz are assigned to A1 while (2px , 2py ) is a basis for E. Let
us write a representation on the basis

1
0
0
s 1 0 , s2 1 , s 3 0 .
0
0
1
one nds:
2
E C3 C3 a b c
100
001
010
100
001
010
0 1 0 1 0 0 0 0 1 0 0 1 0 1 0 1 0 0
001
010
100
010
100
001
158
This representation can be reduced by the LOT into A1 E. The A1

2 +s3 ; to nd a basis for E one can use projection
combination is 1 = s1 +s
3
operators. The operator that projects onto E is
2 1 1

P (E) =
(E) (R)R = 1 2 1 .
(8.1)
R
1 1 2
The that one obtains depends on the s orbital; if we start from s1 , which
2 s3 which is
is v (1) invariant, we get (upon normalization) 2 = 2s1 s
6
also v (1) invariant. This is the x component, which is invariant under the
exchange of 2 and 3. The orthogonal function is odd under such an exchange
3
. One can get this result by the projection operator. Starting
and is 3 = s2s
2
1 s3 which is not orthogonal to 2 but has
from s2 , one obtains 3 = 2s2 s
6
the orthogonal component 3 . Alternatively, one can obtain those results by
projecting directly on the x and y components of the irrep E, by
(E)
Pxx
=
2 (E)
Dxx (R) R.
6
(8.2)
In the current geometry, unlike the last

Chapter, the reection 1 leaves x
1 0
. The other reections are obtained
invariant, so we must take 1 =
0 1
+
+
,
,
1
3
1
3
2
2
2
and 3 = 1 C3 = 23 12
. Thus, one
as 2 = 1 C3 =
3
1
2
2
2
2
obtains
2 1 1
1
(E)
Pxx
= 1 12 12 .
(8.3)
3
1 12 12
In a similar way,
0 0 0
= 0 12 21 .
0 12 21
(E)
Pyy
(8.4)
Thus, we have determined the symmetry orbitals; 2s, 2pz , 1 belong to A1 and
the pairs (2px , 2py ), (2 , 3 ) belong to E. The 7 7 determinant is broken
into a 3 3 determinant for the A1 basis and a pair of 2 2 determinants
for the x, y components of E; moreover, by the Schur lemma, the latter are
identical.
8.1.2 Molecular Orbitals of CH4
CH4 belongs to the tetrahedral symmetry Group Td ; we may put the C atom
at the origin and H atoms at (-1,1,1),(1,-1,1),(-1,-1,-1),(1,1,-1) in appropriate
units. There are NG =24 operations. There are C3 and C32 rotations around
159
CH bonds,C2 and S4 rotations around the x,y,z axes, and 6 d reexions in

the planes that contain two CH bonds. The C orbitals (2s,2px , 2py , 2pz ) are
classied at once under A1 and T2 .
z
1
2
y
4
3
Fig. 8.2. Tetrahedral CH4 molecule inscribed in a cube; the open circle denotes
the C position.
Td
A1
A2
E
T1
T2
E
1
1
2
3
3
8C3
1
1
1
0
0
3C2
1
1
2
1
1
6d
1
1
0
1
1
6S4
g = 24
1
r2
1
0 (3z 2 r2 , x2 y 2 )
1
(Rx , Ry , Rz )
1
(x, y, z)
For the Hydrogens, let d(i, C) denote the list of the destinations of atom 1
under the operations in class C. Here is the situation:
Class moved unmoved

d(1,C)
E
0
4
4
1
8C3
3
1
1 {1,1,2,2,3,3,4,4}
3C2
4
0
0
{2,3,4}
d
2
2
2 {1,1,1,2,3,4}
6S4
4
0
0 {2,2,3,3,4,4}
Table 8.1. Eects of the operations of Td on atom 1 and characters of the representation of Hydrogens.
Under each operation, every atom which changes position contributes 0 to

the character (we are neglecting overlaps), while unmoved atoms contribute
1. Using the LOT with NG = 24 we nd that (H4 ) = A1 T2 . The A1
combination is trivially A1 = |1+|2+|3+|4
. The destinations in the above
2
160
Table 8.1.2 and the Character Table allow to project on T2 by the operator
(7.70):
P (T2 ) |1 = 3|1 (|2 + (|3 + |4 ) + [3|1 + (|2 + |3 + |4 )] 2(|2 + (|3 + |4 )
= 2[3|1 (|2 + (|3 + |4 )].
If we wish to obtain z the fastest way is probably to symmetrize between
1 and 2. Indeed, if 3|1 (|2 + (|3 + |4 ) is in T2 , 3|2 (|1 + (|3 + |4 )
also is; we sum and normalize obtaining
|z =
Also,
|1 + |2 |3 |4
.
2
|1 |2 |3 + |4
,
2
|1 |2 + |3 |4
|y =
.
2
|x =
8.1.3 Characters of Angular Momentum Eigenstates

The set {|J, MJ } is the basis of an irrep of O(3) but reduces in its subgroups. A rotation Rz () by an angle about the z axis multiplies |J, MJ
by exp(iMJ ), thus the character of Rz () is
() =
J

MJ =J
exp(iMJ ) =
sin((J + 12 ))
.
sin( 2 )
(8.5)
All rotations by an angle belong to the same class, thus the character is
this, independently of the rotation axis.
8.1.4 Examples: Oh Group, Ligand Group Orbitals, Crystal Field
The Oh Group
We consider a cube with a system of cartesian axes through the centers of the
faces. The centers of the faces are vertices of an octahedron. The operations
of symmetry of the cube and octahedron are the same. We label the vertices
of the octahedron as in Figure 8.1.4. There are 48 operations in the cubic
Group Oh . The C4 rotations around the cartesian axes and their inverses
form the 6C4 class, and the C2 rotations form the 3C2 class. The 2
3 rotations
in both senses around the 4 cube diagonals give a class 8C3 ; these axes are
perpendicular to faces of the octahedron. Joining the center of EC with the
center of FB we have a C2 axis. There are 12 edges, so there are 6 axes and the
class is 6C2 . The inversion forms a class; the 3h class includes reections like
161
the one in the BCEF plane. The C4 rotations followed by a h reection in the
plane perpendicular to the rotation axis form the 6S4 improper rotation class.
The plane containing A,D and the midpoints of EC and BF is a symmetry
plane, and the reection belongs to a 6d class. Finally,there are S6 improper
rotations: the axes are those of the C3 rotations; if we look along a C3 axis,
opposite faces appear as concentric triangles; the gure is invariant under
a C6 rotation followed by a reection in the orthogonal plane through the
center (Figure 6.2). With 4 axes and two senses of rotations one builds a 8S6
class.
z
A
C
y
D
Fig. 8.3. Octahedron
Ligand Group Orbitals

Oh E 6C4
A1g 1 1
A1u 1 1
A2g 1 1
A2u 1 1
Eg 2 0
Eu 2 0
T1g 3 1
T1u 3 1
T2g 3 1
T2u 3 1
6 2
3C2
1
1
1
1
2
2
1
1
1
1
2
6C2
1
1
1
1
0
0
1
1
1
1
0
8C3
1
1
1
1
1
1
0
0
0
0
0
i 6S4
1 1
1 1
1 1
1 1
2 0
2 0
3 1
3 1
3 1
3 1
0 0
3h
1
1
1
1
2
2
1
1
1
1
4
6d
1
1
1
1
0
0
1
1
1
1
2
8S6
g = 48
1
x2 + y 2 + z 2
1
1
1
1 (x2 y 2 , 2z 2 x2 + y 2 )
1
0
(Rx , Ry , Rz )
0
(x, y, z)
0
(xy, xz, yz)
0
0
The Oh Group is the point Group of many interesting solids, including complexes like CuSO4 5H2 O and FeCl3 where a transition metal ion at the center
162
of an octahedron; most often the octahedral symmetry is only approximate,

as in CuSO4 5H2 O where the central Cu++ is bound to 4 H2 O molecules
and 2 SO
ions. A LCAO model of their properties is often called ligand
4
eld theory.
If A F represent 6 s orbitals, they are the basis of a representation
with the characters shown above. One nds that = A1g Eg T1u . Let us
work out the projection of orbital A into T1u :

Class C
(T1u )
contribution
RC RA
E
A
3
3A
6C4 2A+B+C+E+F 1
2A+B+C+E+F
3C2
A+2D
-1
-(A+2D)
6C2 2D+B+C+E+F -1 -( 2D+B+C+E+F)
8C3
2(B+C+E+F)
0
0
I
D
-3
-3D
6S4 2D+B+C+E+F -1 -(2D+B+C+E+F)
3h
2A+D
1
2A+D
6d 2A+B+C+E+F 1
2A+B+C+E+F
8S6
2(B+C+E+F)
0
0
The normalized T1u projection is 1 =
AD
.
2
Operating in the same way
on D we again get 1 . Operating on the other functions, we obtain 2 =

and 3 =
CF
BE
. In this way one easily builds the ligand group orbitals.
Crystal Field
The number of d electrons in transition metal ions is:
Ti V Cr Mn Fe Co Ni Cu
M ++ 2 3 4 5 6 7 8 9
M +++ 1 2 3 4 5 6 7 8
Hunds rule allows to nd the ground state of partially lled d shells and
the corresponding number of unpaired electrons as follows:
d1 d2 d3 d4 d5 d6 d7 d8 d9
ground state D 3 F 4 F 5 D 6 S 5 D 4 F 3 F 2 D
unpaired
1 2 3 4 5 4 3 2 1
2
for isolated transition ions one would always predict paramagnetism, but the
compounds can be paramagnetic or not. The bivalent Fe ( 3d6 conguration)
8.2 Normal Modes of vibration
163
forms the complexes [F e(H2 O)6 ]2+ , which is green, and paramagnetic, but
also [F e(CN )6 ]4 , which is yellow and diamagnetic. These facts can already
been understood by the crystal eld theory, in which the ligands behave like
point charges, or anyhow they generate a eld to octahedral symmetry that
resolves the degeneracy of the atomic terms (for the rst transition series the
spin-orbit interaction is a small perturbation to be introduced subsequently).
A more quantitative treatment is obtained then from the ligand eld theory.
For d (or D) states, from (8.5) and from the even parity we nd the
characters
Oh E 6C4 3C2 6C2 8C3 i 6S4 3h 6d 8S6
d 5 1 1 1 1 5 1 1 1 1
and conclude d = Eg T2g . This is already clear from the Character Table,
reporting (dxy , dxz , dyz ) as a basis for T2g and (dx2 y2 , z 2 ) for Eg . In transition
metal complexes usually = E(Eg ) E(T2g ) > 0, since the T2g orbitals stay
far from the negative ligands. Thus in the absence of Coulomb interactions
one would ll the available levels according to the aufbau principle, starting
with T2g . In crystal eld theory one tries to predict the magnetic properties by
diagonalizing a many-electron Hamiltonian which is the sum of the isolated
ion Hamiltonian and the one of the crystal eld.
There are two simple limiting cases. If
U, where U represent the
order of magnitude of the multiplet splitting, due mainly to the Coulomb
interaction, one treats as a perturbation of the isolated ion multiplet. The
atomic Hund rule holds, and paramagnetism obtains. If U, Hunds rule
holds (high spin is preferred) within the degenerate T2g and Eg levels, but Eg
starts being lled only after T2g is full, and 6 electrons yield a diamagnetic
complex.

In the Born-Oppenheimer approximation, the potential that governs the motion of electrons depends parametrically on the positions of the nuclei. The
total energy U of the molecule is calculated in function of the nuclear coordinates and the conguration of equilibrium correspond to a minimum energy.
The vibrations can then be studied by considering U as the potential energy
of the nuclei. The conguration of the molecule is given by a vector

v that
species the shift from equilibrium of the coordinates of all the N nuclei,
v = (x1 , y1 , z1 , x2 , xN , yN , zN ) (v1 , , v3N )
(8.6)
where I introduced a notation vi for the generic cartesian coordinate. All the
possible motions of the nuclei are described classically from the equations of
Newton,
164
mi vi =
U
vi
(8.7)
where mi is the mass associated to v in the obvious way x1, y1, z1 m1 and
so on. In the harmonic approximation, expanding around to one conguration
of equlibrium,
U=
1
Uij vi vj ,
2 ij
(8.8)
and the Fourier transform of the equations of the motion is

2 vi =
We introduce
1
Uij vj .
mi j
(8.9)
Qi = vi mi
(8.10)
Uij
Wij =
mi mj
(8.11)
and the symmetrized matrix
then the equations of the motion read:

2 Qi =
Wij Qj .
(8.12)
The eigenvectors Q of the W matrix are the normal modes, and their eigenfrequencies are obtained from the secular equation. Three modes, of null
frequency, correspond to rigid translations of the molecule, and 3 others (or
2, for linear molecules) to rigid rotations. Rotations also have = 0, since
the energy of the molecule does not depend on its orientation. The remaining
3N-6 (or 3N-5) frequencies are vibrational. The Group theory is helpful to
simplify the solution of the secular equation. Every R operation of the Group
of the molecule sends every nucleus in itself or to another identical nucleus in
an equivalent position; meantime, it produces a (proper or improper) rotation the system of cartesian axes that we can imagine xed at every nucleus.
So, R maps each Qi into a linear combination of the components. In such
a way, we associate to every R a matrix D(R), and obtain a representation
of the Group in the space of the vectors Q. Every D(R) commutes with
W , since R is a symmetry and W must be invariant under R; thus if Qa
is a solution of the secular equation, also D(R)Qa must be a solution, and
with the same . Therefore, we can diagonalize W simultaneously with the
maximum number of commuting D(R) matrices. The theory of the Groups
adds the Dirac characters; this means that in a new basis whose elements
are vibrations pertaining to irreps of the Group, W is block diagonal. Thus,
the normal modes can be assigned to irreps of the Group; the reduction of
165
the Q, (or equivalently v) representation, allows to gain in simple way the

symmetry and the degeneracy of the normal ways, without having to resolve
the secular equation.
The H2 O molecule (C2v Group), and that requires a 9-dimensional v. We
place the molecule on the xz plane. The cartesian components of the shift
from the position of equilibrium of each atom transform like its p orbitals
and can be thought of as numbered arrows,parallel to the x,y and z axes
respectively; th earrows (1,2,3) from one H, (4,5,6) from O and (7,8,9) from
the other H, constitute a basis. The operations are C2 , (xz), and (yz). The
C2 operation has the representative
0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 1
0 0 0 1 0 0 0 0 0
(8.13)
D(C2) =
0 0 0 0 1 0 0 0 0 .
0 0 0 0 0 1 0 0 0
1 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0
D(C2 ) and D(v (yz)) have the same block structure
00b
D(C2 ) = 0 b 0
b00
where now 0 is for a null 3 3 block, and
1 0 0
1 0
b(C2 ) = 0 1 0 , b((yz)) = 0 1
0 0 1
0 0
(8.14)
0
0;
1
(8.15)
D((xz)) is block diagonal with b = diag(1, 1, 1). From the matrices we

gain the characters:
: (E) = 9, (C2 ) = 1, ((yz)) = 1, ((xz)) = 3.
(8.16)
We could arrive to this result without writing the D matrices, taking into
account that for each operation:
the atoms that change position contribute 0 to the character; 2)

each arrow (cartesian movement) that it remains invariant contributes
+1, and every arrow that changes sign contributes -1,
more generally, the cartesian shifts of an atom that does not change position behave like (x, y, z), so if the arrow is rotated by the contribution
is cos().
166
C2v
A1
A2
B1
B2
I
1
1
1
1
C2 v v
1 1 1
1 1 1
1 1 1
1 1 1
g=4
z
xy, Rz
x, Ry
y, Rx
The representation is clearly reducibile. It includes all the possible

movements of atoms, therefore not only the vibrations, but also the rigid rotations and translations of the molecule. In order to classify the vibrations,
it is in the rst place necessary to separate them from the rigid motions.
The rigid translations along the (x,y,z) axes transform like (x,y,z); they
are marked in the character tables and are a basis for the representation of
polar vectors (reducible or not, according to the Group).
Adding the characters of x, y and z, we get for the 3 translations the
characters:
C2v E C2 (xz) (yz)
(8.17)
trasl 3 -1 1
1
Having separated 3 coordinated of the center of mass, N-3 remain The
motion that they describe is to be decomposed in a rigid rotation and a
vibration. So, we can separate to others 3 degrees of freedom (2 for linear
molecules) in order to describe the rigid rotation. An innitesimal rigid rotation transforms the coordinate ri of the atom i according to r i = d ri ;
one can expand d as a linear combination of rotations Rx , Ry and Rz around
the axes. These are mixed among themselves under the symmetry operations,
and are a basis of a representation rot . Since is an axial vectors, rot
is the axial vectors representation. The rotation about the x axis transforms
like the Rx generator, that is, like yz, and so on; the tables of the characters report the classication of the rotations. In the Abelian Groups, there
are only one-dimensional irreps with characters 1; R is odd (even) if the
sense of rotation changes (does not change). For the water molecule, using
the character table, we nd
C2v E C2 (xz) (yz)
rot 3 -1 -1
-1
We can proceed to separate the vibrations as it follows:
C2v
trasl
rot
vibr
E
9
3
3
3
C2 (xz) (yz)
-1 3
1
-1 1
1
-1 -1
-1
1
3
1
The reduction of the representation yields
(8.18)
167
vibr = 2A1 B1 .
Finally. The water molecule has two total-symmetric vibrations and one of
B1 symmetry.
A total-symmetric vibration, that all the molecules have, is the breathing
mode. In the symmetric stretch the angle between the OH bonds varies. In
order to discover what kind of vibration is the one labeled B1 that goes like
x one can construct the projection operator
P (B1 ) = E C2 + (xz) (yz);
in terms of block representative matrices,
1 + b((xz))
0 b(C2 ) b((yz))
,
0
0
P (B1 ) =
1 + b((xz))
b(C2 ) b((yz)) 0
(8.19)
(8.20)
where = 1 b(C2 ) + b((xz)) b((yz)). Projecting, one nds a vibration

where the only arrows are 3 and 9, and are opposite. One H shifts up along
the molecular axis and the other goes down; such a vibration indeed changes
sign under C2 and (yz).
Vibrations of N H3
Here the complete table of the characters of the Group.
C3v
A1
A2
E
I
1
1
2
2C3
1
1
1
3v g = 6
1
z
1 Rz
0 (x, y)
The Cartesian movements of N are transformed like the coordinates. Therefore for N we have:
C3v E 2C3 3v
(8.21)
(N ) 3 0 1
Lets nd the characters of (H3 ). (E) = 9 (no arrows move) (C3 ) = 0
(all the H move and contribute 0) (v ) = 1 (two H reected one in the
other: character 0; for the other H, two arrows in the reection plane and one
orthogonal: character 1) Therefore:
C3v
E 2C3
(N ) 3 0
(H3 ) 9 0
(N H3 ) 12 0
trasl 3 0
rot
3 0
vibr 6 0
Finally, vibr = 2A1 + 2E.
3v
1
1
2
1
-1
2
168
Vibrations of Methane (CH4 )

Methane is a tetrahedral molecule.
With 5 atoms (E) = 15.

8C3 : all the atoms move except a H and the C: for each an arrow
is along the rotation axis (+1),while the others two, on the plane,
are transformed as the coordinates (x,y) of this plane and contribute
T rD(R) = 2 cos( 2
3 ) = 1. Therefore (8C3 ) = 0.
3C2 : 2 arrows of the C change sign and the third does not move: = 1
6d CH2 remains in place; each atom has 2 arrows in plane and one
reected and = 3.
S4 H moved; for the atom of C: 900 rotation around z, (x, y, z)
(y, x, z); then reection (y, x, z). So, = 1.
Thus,
Td
A1
A2
E
T1
T1
tot
trasl
rot
vibr
E 8C3
1 1
1 1
2 1
3 0
3 0
15 0
3 0
3 0
9 0
3C2
1
1
2
1
1
1
1
1
1
6d
1
1
0
1
1
3
1
1
3
6S4
g = 24
1
r2
1
0 (3z 2 r2 , x2 y 2 )
1
(Rx , Ry , Rz )
1
(x, y, z)
1
1
= T2
1
= T1
1
A1 E 2T2
Vibrations of Benzene (C6 H6 )

There are 12 atoms and 36 coordinates, therefore (E) = 36. The rotations
C2 , C3 and C6 around the vertical axis move all the atoms and have character
0. Rotation C2 around to a diagonal of the hexagon leaves 4 atoms in place:
for each one arrow is invariant and the others two change sign. Therefore,
(C2 ) = 4. Rotation C2 around an axis to opposite sides and S3 and
S6 move all the atoms and have character 0. The reection h in the plane of
the hexagon leaves two arrows invariant for every atom and changes sign to
the third, therefore (h ) = 12. The reection for a plane containing the C2
axis has character 0 . The reection for a plane containing a axis C2 leaves 4
atoms in place, with two arrows invariant and one changed of sign for every
atom. Therefore (v ) = 4. The characters of trasl are the sums of those of
A2u and E1u ; those of rot adding those of A2g and E1g .
8.3 Space-Time Symmetries of Bloch States
D6h
A1g
A2g
B1g
B2g
E1g
E2g
A1u
A2u
B1u
B2u
E1u
E2u
tot
trasl
rot
vibr
E 2C6
1 1
1 1
1 1
1 1
2 1
2 1
1 1
1 1
1 1
1 1
2 1
2 1
36 0
3 2
3 2
30 4
2C3
1
1
1
1
1
1
1
1
1
1
1
1
0
0
0
0
C2 3C2
1 1
1 1
1 1
1 1
2 0
2 0
1 1
1 1
1 1
1 1
2 0
2 0
0 4
1 1
1 1
2 2
3C2
1
1
1
1
0
0
1
1
1
1
0
0
0
1
1
2
i 2S3
1 1
1 1
1 1
1 1
2 1
2 1
1 1
1 1
1 1
1 1
2 1
2 1
0 0
3 2
3 2
0 0
2S6
1
1
1
1
1
1
1
1
1
1
1
1
0
0
0
0
h 3d
1 1
1 1
1 1
1 1
2 0
2 0
1 1
1 1
1 1
1 1
2 0
2 0
12 4
1 1
1 1
12 4
169
3v
g = 24
1
1
Rz
1
1
0 ! (Rx , Ry ) "
x2 y 2 , xy
0
1
1
z
1
1
0
(x, y)
0
0
1
1
0
The Group has 24 elements, and it is found thatvibr = 2A1g A2g A2u
2B1u 2B2g 2B2u E1g 3E1u 4E2g 2E2u .

Let t1 , t2 , t3 be the primitive translation vectors of a crystal lattice, {Ti } the
set of lattice translation operators and V (

x ) the periodic crystal potential,
such that [Ti , V ] = 0. Periodic boundary conditions are assumed, that is,
for some N 1, TiN = 1. The simultaneous eigenfunctions

x ) of H and
(
k
any Ti are such that Ti

x ) =
x + t i ) = C
x ) with C N = 1.
(
(
(
k
k
k
Now, TiN ei k x = ei k x requires k t i = 2

N integer; therefore,
pg1 + qg2 + rg2
k =
N
with p, q, r Z, and the reciprocal lattice vectors dened by ti gj = 2ij .
We know from Sect. 7.2.1 that Blochs functions are
x ),
(8.22)
x ) = ei k x u
(
(
k
k ,
where u
x ) are lattice periodic. Moreover, since (letting h = 1)
(
k
p ei k x = ei k x (
p + k ),
we may write
170
(
p + k )2
x ).
+ V ( x ) u
x ) = k u
(
(
k
k
2m
(8.23)
No degeneracy is predicted, since Abelian Groups have only one-dimensional

representations. Each translation T by a vector t is a class, while each k
vector labels a dierent irrep. The LOT of Equation(7.58) yields the useful
relation
1 i(kk )t
e
= (k k ),
(8.24)
NC t
where NC is the number of cells. The second character orthogonality theorem
7.64 gives:

ei(tt )k = NC (t t ),
(8.25)
kBZ
with the sum extended to the Brillouin zone.

The trivial spin degeneracy is lifted if the spin-orbit interaction is included,
p2
+ V (
x ) + HSO
(8.26)
H=
2m
with
1
V
p.
(8.27)
HSO
=
4m2 c2
Eigen-spinors can be taken of the Bloch form,
x)
(
x
x
i k
i k
k ,+
u
(8.28)
(x) =e
(x)= e
u
k ,
k ,
(x)
k ,
One nds spinors with = + and - that reduce to up and down spin for
c , but otherwise spin and orbital degrees of freedom mix. The periodic
functions are obtained by solving

1
(
p + k )2
+V(x)+
V ( p + k ) u
x)
(
k ,
2m
4m2 c2
x ). (8.29)
=
(
u
k , k ,
Space Inversion
The inversion is the operator P 0 : P 0

x =
x . Let
x ) solve the
(
k ,
Schr
odinger equation with a periodic potential V ( x ); an inverted crystal
x ) = P 0 V (
x ) = V (
x ) instead. Since
poses the same problem with V (i) (
(P 0 )2 is the identity, from Hk (x) = k k (x) one gets
P 0 HP 0 P 0 k = H(x)P 0 k = k P 0 k ;
171
thus the solution the inverted crystal is the spinor

x ) and the energy
(
k
(i)

x ), one cannot
is the same as in the original problem. If V ( x )
= V (
k
speak of degeneracy, since the corresponding problems remain distinct. If
however the crystal has inversion symmetry, that is, [P 0 , H] = 0, adding
this element to the translations produced a non-Abelian Group which implies
degeneracy. Indeed, k (x) and k (x) both belong to the eigenvalue
.
k
0
ikx
However, P k (x) k (x) = e
uk (x) belongs to k since it gets
multiplied by eikt under a translation by t. Therefore we must make the
identications
(8.30)
P 0 k (x) = k (x),
and uk (x) = uk (x); so,

=
.
k
k
(8.31)
Time Reversal
The antilinear1 Kramers operator K takes the complex conjugate, and is
useful to discuss time reversal symmetry, as I show in a moment. In the
Schr
odinger theory, assume one knows how to solve a time-dependent problem
i
h
(t)
= H(t)(t)
t
(8.32)
but wants to solve the Schr

odinger equation with a time-reversed Hamiltonian
i
h
(t )
= H(t ) (t ).
t
(8.33)
= H(t)(t) and setting

Applying K to both sides of (8.32) we get i
h (t)
t

t = t in the operators,
(t)
= H(t )(t) .
(8.34)
t
Comparing with (8.33) we conclude that (t ) = (t) ; introducing a time
reversal operator T this may be interpreted by saying that (t ) = T (t) is
the time reversed wave function with T = K.
i
h
Things are more involved with the Pauli equation; I write H0 the spinindependent part of the Hamiltonian, which I assume real, and separate the
spin-eld coupling:
i
h
1
eh
(t)
= [H0 (t) + B(t)](t), =
.
t
2mc
is antilinear if O(a
+ b O).
An operator O
+ b) = a O
(8.35)
172

Taking the complex conjugate and multiplying both sides by i2 =
0 1
1 0
one nds
(i2 )(t) = H0 (t)(i2 )(t) + (i2 ) B(t)(t)

(t)
= H0 (t)(i2 )(t) + (1){(i2 ) B(t)(i2 )}(i2 )(t) .(8.36)
i
The familiar anticommutation relations give

(i2 ) B(t)(i2 ) = B(t).
Now let t = t in the operators.
(i2 )(t) = H0 (t )(i2 )(t) + (1) B(t )(i2 )(t) .

t
(8.37)
which says that now (i2 )(t) is the time reversed spinor with B(t ) =
B (t ) . But this sign is wanted: indeed the currents do change sign under
time reversal, hence the vector potential and the magnetic eld also change
sign. Thus,
(8.38)
T = iy K.
i
Note that
K, K = , ,
(8.39)
T , T = , .
(8.40)
and
Then
T 2 = ,
thus
T 1 = T .
T reverses the sign of

p (indeed T
p T 1 =
p because of the complex
conjugation) and of the angular momenta L and S (for instance, 2 1 =

1
1 2 T 1 T
= 1 ). Thus the scalar product L is time-reversal
invariant.
Kramers Theorem for one-electron states says that a stationary Hamiltonian like (8.26), without magnetic elds, even in the presence of the spinorbit interaction, has a twofold degeneracy since and T have the same
energy and are orthogonal.
Since T commutes with H, if is
a stationary state of H, T also is.
Moreover T ; indeed, if =
,
8.4 Space groups of solids
T = iy K
173

0 1

=
=
1 0
T | = (, )
= 0.
then
Since [T , H] = 0, the time reversed Bloch eigenspinors
T
x ) = iei k x y u
x)
(8.41)
(
(
k
k ,
have the same energy and belong to k , ( they are periodic spinors times
ei k x ). Besides, they belong to since time reversal reverses spin. Using

(8.40),
x ) = T z
x ), T
x ) ;

x ), z
(
(
(
(
k
k
k
k
the anticommutation rules give
x ), T
x ) ;
= z T
(
(
k
k
since z is Hermitean,

x ), z
x ) = T
x ), z T
x ) .
(
(
(
(
k
k
k
k
In summary the time reversal invariance requires
x) =
x );
T
(
.
=
(
k
k ,
k ,
k ,
(8.42)
Conjugation
If P 0 and T are simultaneous symmetries,the conjugation
C = P 0T
(8.43)
also is conserved. Using equations (8.30),(8.42) we see that this is such that
C
x ) =
x );
(
(
=
k ,
k ,
k ,
k ,
(8.44)
and there is degeneracy at every k point.

The Space Group elements are [93]
r +
a
(|a) :
r =
(8.45)
174
where denotes an operation of the Point Group while

a is a translation (as
we will see, not always a Bravais one). This is a Group with the multiplication
rule
(| b )(|
a ) = (| b +
a)
(8.46)
and it has the Point Group as a subgroup. We can write the faithful representation2

a
r
r
=
.
(8.47)
1
0 1
1
The inverse operation is
Since
(|
a )1 = (1 | 1
a ).
(8.48)
a ) = (1 |1 ( b +
a
a ))
(|
a )1 (| b )(|
(8.49)
the translations (unlike rotations) are an invariant subgroup.

A Space Group generated them from the Bravais translations and from the
Point Group is symmorphic. The denition (8.46) (8.47) does not correspond
to the direct product, that it would give instead (| b )(|

a ) = (| b +
a ).
It is called a semidirect product. The translation depends on the choice of the
origin: consider the operation
r +
a;
r =
(8.50)
let us rewrite it in a reference with the origin shifted to b . The starting
point in the new reference is

s =
r + b and the transformed one is s =

r + b ; therefore the transformation is described by

s b +
a + b.
s =
(8.51)
So, the rotation is the same, but

a
a =
a b + b . The operation
can be made homogeneous, that is,

a = 0, if we can nd b such that
a.
b = ( 1)1
(8.52)
8.4.1 Symmorphic and Nonsymmorphic Groups

The symmorphic Groups have the rotations of the point Group and the translations of the Bravais lattice; nonsymmorphic Groups have extra symmetry
elements are called screw axes and glide planes . These new operations
depend on special relations between the dimensions of the basis (that is, of
the unit which is periodically repeated) and of the Bravais translations. For
2
dierent operations have dierent matrices.
175
example, CdS (hexagonal close-packed structure) has both kinds of extra

symmetries. A plane a b contains alternated Cd and S ions arranged in a
regular hexagonal pattern and above this plane at a distance 2c another a b
plane with Cd and S exchanged, and so on. The C6 rotation about the center
of the hexagon exchanges Cd and S; for the same reason, a translation by 2c
along the c axis does not belong to the Bravais lattice and is no symmetry;
however the combined operation (C6 and translation by 2c ) leaves the system
invariant. The v reection of the hexagon is broken since it leads to the exchange of Cd and S; however, the v plane is a glide plane: reection in this
plane becomes a symmetry when accompanied by the 2c translation along the
screw axis.
It is natural to ask whether one can eliminate the 2c translation from the
screw axis operation by simply shifting the origin to b . According to (8.51)

s , if 0 = (1 ) b +
a.
the transformation becomes homogeneous,
s =
This requires satisfying (8.52) but if

a =
a , there is no solution, and this
is the case since the translation

a is along the axis of the rotation . So, the
screw axes cannot be eliminated.
Iterating a screw axis operation we must obtain a Bravais translation.
Since belongs to the Point Group, n must be the identity for some integer
n ; then (8.46) implies (|

a )2 = (2 |
a +
a ), (|
a )3 = (3 |2
a +
a +
a)
and iterating we nd
(|
a )n = (n |
n1
a ) = (1, t ).
k
(8.53)
k=0
where t is a pure translation.
a where
a is normal to the rotation axis. The
Let us write
a =
a +
condition (8.52) can always be solved to eliminate

a . Setting
a =
a , it
n1 k
follows that k=0 a = n a and
.
a =
n
(8.54)
For instance, a screw-axis with an angle = 2 can have a translation equal to

1/4, 2/4 o 3/4 a Bravais vector. In the case of a glide plane, is a reection;
the above holds with n = 2, and

a +
a = 2
a = t ; hence a glide plane
has a translation of half a Bravais vector.
International Notation
In the international notation3 , a screw-axis with an angle = 2 and a
translation equal to 1/4, 2/4 o 3/4 a Bravais vector is denoted by 41 , 42 o
43 . The international notation for a Space Group starts with a letter ( P
3
International Tables for X-Ray Crystallography (1952)
176
for primitive, I for body-centered, F for face centered, R per rombohedric)

followed by the indication of the Point Group. Thus,
42
3
m m
is the face centered Group with the Oh Point Group, with a C4 axis with
an horizontal symmetry plane, a C3 axis and inversion symmetry; this is
symmorphic,while the diamond Group
F
41 2
3
d m
is not; d denotes a glide plane with a translation 1/4 Bravais vector.
F
Representations of the Translation Group and of the Space Group
Every k labels a representation of the Translation Group , with eigenvalue
r ); so
ei k t . In general, one denes Rf (
r ) = f (R1
(,
a )ei k r = ei k (, a ) r
and using the inverse operation
(,
a )ei k r = ei k ( r a ) .
Rotating two vectors by the same angle the scalar product does not change;
so we may write
(,
a )ei k r = ei( k )( r a ) .
(8.55)
In terms of Bloch functions, (,

a )n ( k ,
r ) yields a linear combination of

n ( k , r ), where n n because in general Point Group operations mix

degenerate bands.
One denes Star of k k the set { k }. Higher symmetry points have
stars with fewer elements. For non-degenerate bands, k points of low symmetry correspond to a single , but in general there is a subspace associated
to a given k point.
The Group of the wave vector k or little Group is the Subgroup G

G
k
which consists of the operations (, a ) (with a that may be a Bravais vector
or not) such thet does not change k :

k = k + G
with G in the reciprocal lattice (k vectors that dier by G are equivalent).
G
has the invariant Subgroup T
T G of the translations t such that
k
k
ei k t = 1. The set of the basis functions of a representation of G

for all
k
the points of a star provide a basis for a representation of the Space Group
G. For further reading on this subject see [93].
8.5 Young Diagrams
177
8.5 Young Diagrams

Young Tableaux for S(3)
Let us consider again the Group C3v . All operators can be expressed as
reections or products of reections (C3 = 1 2 , C32 = 1 3 ); moreover since
1 exchanges 2 with 3, etc., all operations are permutations and C3v S(3).
In the irrep A1 all the operators are represented by 1; in A2 the reections
i are represented by -1. The totally symmetric functions belong to A1 while
those of A2 are totally antisymmetric under any i . Young diagrams represent
the irreps of S(n) by schemes with n boxes. Horizontal lining of boxes is
associated with symmetrization, vertical lining with antisymmetrization, and
the length of lines does not increase. So the rst diagram refers to A1 , the last
to A2 and the intermediate one to E. Here are the Young diagrams for S(3).
Each diagram correspond to an irrep, and this fact is general. E corresponds
to a mixed symmetry; in the geometry of Figure 7.1 with 1 on the y axis,

the projector on the x component is
1
1
1
1
(E)
= 1 (C3 + C32 ) 1 + (2 + 3 ) = 1 1 (2 + 3 ) 1 + (2 + 3 ),
Pxx
2
2
2
2
(8.56)
that is, introducing the anti-symmetrizer A(2, 3) = 1 1 and the symmetrizer S(2, 3) = 1 + 1
(E)
Pxx
= A(2, 3)[S(1, 3) + S(1, 2)].
(8.57)
A(2, 3)S(1, 3) and A(2, 3)S(1, 2) are mixed symmetries and project on E.
These projectors written in terms of symmetrizers and antisymmetrizers are
called Young projectors. They may be thought of as projectors on the various
irreps based on the regular representation. The three diagrams exhaust the
possible partitions of 3 in not increasing integers, that is, 3 =2+1=1+1+1.
The Young tables or Young tableaux are obtained from the Young diagrams by inserting numbers from 1 to 3 so that every line and every column
grow along. The tableau
123
178
represents the projection operator S(1, 2, 3), the symmetrizer ; the Young
12
but one can also
projection operator A(1, 3)S(12) is given by
3
1
13
antisymmetrize with respect to 2, getting
. Finally, 2
2
3
projects on A2 .
The fact that there are two tables with mixed permutation symmetry is
due to degeneracy 2 of the irrep E. In general, in the Young tables for S(N ),
the m-dimensional irreps occur m times.
Young Tableaux for S(4)
S(4), has 24 elements and the following 5 irreps that may be found by the
above stated rules.
1234
123
4
is the 1-dimensional A1 representation.

124
3
134
2
are 3 occurrences of a 3-dimensional irrep, while

12
34
13
24
is 2-dimensional,
1
2
is 3d and nally
is the totally antisymmetric ir3
4
rep. These correspondences are useful as it extends to all groups S(N) of
permutations of N objects. Thus, the irreps of S(N) are known for any N.
12
3
4
13
2
4
14
2
3
Young Tableaux for Spin Eigenfunctions

Consider a system consisting of N spins 1/2 and a set of eigenstates |S, MS
obtained e.g. by solving the eigenvalue problems for S 2 and Sz . Any permutation of the spins sends an eigenfunction into a linear combination of the
eigenfunctions with the same eigenvalues S, MS ; in other terms, the S, MS
quantum numbers label subspaces of functions that do not mix under permutations. Within each subspace, one can use projection operators to produce S
and MS eigenfunctions that form a basis of irreps of S(N). For each symmetry
type (i.e. for each component of each irrep) there exists one solution.
For instance, for N=3 a quartet and two doublets exist. With MS = 32
the only state is | , which is invariant for any permutation of the arrows;
8.5 Young Diagrams
179
by the shift operators one nds that | 32 , 12 = 13 (| + | |+ ). This

is invariant for spin permutations, too, and belongs to the A1 irrep of S(3).
The shift operators preserve the permutation symmetry, and all the 2MS + 1
states belong to the same irrep. The ortogonal subspace involving one down
spin yields two doublets; one is | 12 , 12 = 12 [| | ] ; another one which
is not orthogonal is | 12 , 12 = 12 [| | ]; orthonormalizing we obtain
| 12 , 12 = 16 [2| | | ]. These doublets are not invariant for
spin permutations; they are just the two components of the familiar irrep E of
the C3v permutation symmetry. Moreover, by the shift operators each yields
its | 12 , 12 companion. A quarted and 2 doublets exhaust all the 23 states
available for N=3, and there is no space for the A2 irrep. This is general:
since spin 1/2 has two states available, any spin wave function belongs to a
Young diagram with 1 or 2 rows.
Thus we realize that rather than solving eigenvalue problems we can be
smarter and obtain the eigenfunctions directly by Young projectors. The permutation symmetry of the full many-electron wave function will be discussed
in Sect. 9.8.
Problems
3
2
5
a)
b)
Fig. 8.4. a) The CuO4 Hubbard model cluster. b) The Cu5 O4 Hubbard model
cluster.
8.1. The CuO4 Hubbard model cluster

Consider the Hubbard model cluster in Figure (8.5 a)), belonging to C4v
symmetry. (1) Find the irreps of the one-electron orbitals.
(2) Consider this cluster with 4 fermions, in the Sz = 0 sector. Classify the
4-body states with the irreps of the Group.
8.2. The Cu5 O4 Hubbard model cluster
For the cluster in Figure (8.5 b)) classify the 4-fermion states in the Sz = 0
with the irreps of C4v .
9 Product of Representations and Further

Physical Applications
9.1 Irreducible Tensor Operators

where is a wave function and A some operator. Then, actLet = A,
R, where the
ing with a unitary operator R, one nds R = A R = RAR
transformed operator is A = RAR . Thus, functions are transformed accord ;

ing to R while operators are transformed according to A RAR
actually the two rules dier by a matter of notation. R acts on everything
on its right, so in the case of operators the last factor R = R1 is there
while functions are at the
just to ensure that the action of R is limited to A,
extreme right and there is no need for that. We can consider (x1 , x2 , x3 ) as a
set of functions or as the components of an operator. As functions that transform as a basis of a representation of some symmetry Group, they transform
according to the rule
(x1 , x2 , x3 ) (Rx1 , Rx2 , Rx3 ) = (D(R)k1 xk , D(R)k2 xk , D(R)k3 xk );
this is the vector representation, which is irreducible in cubic and higher
Groups. If we treat them as a set of operators, we write
(x, y, z) (RxR , RyR , RzR ).
This denes a vector operator, but the linear combinations that result are
the same. A tensor is a set of operators Ti (the components) that are mapped
into linear combinations by every R, that is,

S
Ti STi S =
Tj Dji (S);
(9.1)
j
the multiplication Table is followed since

RS
Ti RSTi S R =
Tj Dji (RS).
(9.2)
If the representation is the irrep , we can speak of the irreducible tensor

operator T () and in any representation all its components are mixed by the
Group operations. In GL(n) a tensor of rank r is a set of operators T such
that
182
9 Product of Representations and Further Physical Applications

T(1),(2),,(r)
=
a(1)(1) a(2)(2) a(r)(r) T(1),(2),,(r) .
(1),(2),,(r)
(9.3)
Given a second-rank tensor T , one can build a symmetric tensor Sij =
Tij + Tji and an antisymmetric one Aij = Tij Tji . Obviously a symmetric
(antisymmetric) tensor remains symmetric (antisymmetric) under the transformations of GL(n). In general, the tensors of GL(n) are reduced into irreducible parts by taking linear combinations according to the irreps of the
permutation Group S(r). Further reduction is possible in subgroups of GL(n).
Tensors in Polar Form
Under the operators of O+ (3) the ITO transform like spherical harmonics.
Therefore, the following polar form is expedient. For a vector, the polar components are xm , m = 0, 1, where, using a traditional notation,
x + iy
x iy
x+1 = , x0 = z, x1 = ,
2
2
and transform according to
Rxm R1 =
1
xn Dnm
(R),
(9.4)
(9.5)
n
1
(R) represents the rotation in the basis of the spherical harmonics
where Dnm
with l = 1. It is natural to put the basis vectors in polar form too,
e1 + ie2
e1 ie2
, e0 = e3 , e1 =
.
(9.6)
2
2
3
In this way the expansion of a vector in the basis, V = i=1 Vi ei is replaced
by
1

()m Vm em .
(9.7)
V =
e+1 =
m=1
The scalar product of two real vectors V and W in polar form is

V W =
1
()m Vm Wm ,
(9.8)
m=1
as one can readily verify. The advantage of the polar form is that now we can
generalize to an ITO of any rank of O+ (3). The denition of such an ITO is

()
Tq() Dqp
(R).
(9.9)
RTp() R1 =
q
For instance if denotes the representation with J = 0, then q = 0 and T is

a scalar; if denotes the representation with J = 1, then q = 0, 1 and T (1)
(l)
(l)
is a vector in the polar representation; in general, Tq transforms like Yq .
9.2 Direct Product Representation
183
9.2 Direct Product Representation

Direct Product of Matrices
From the pair of matrices

a11 a12
b11 b12
A=
, B=
,
a21 a22
b21 b22
one can build the direct product, thet is, the block matrix
a11 b11 a11 b12 a12 b11

a11 b21 a11 b22 a12 b21
a11 B a12 B
D =AB =
=
a21 b11 a21 b12 a22 b11
a21 B a22 B
a21 b21 a21 b22 a22 b21
a12 b12
a12 b22
.
a22 b12
a22 b22
Note that T rD = T rAT rB.

We can number rows and columns as usual and say for instance D24 =
a12 b22 , but for direct-product matrices a dierent rule is used. One denotes
the matrix element by a pair if row indices and a pair of column indices, so
a12 b22 = D12,22 . In general, Dkp,ij = aki bpj is the element of a mn mn
matrix D = A B, where m and n are the sizes of A nd B respectively.
Direct Product of Representations
Further important developments of the theory of symmetry are needed for
systems that consist of parts (e.g. two electrons, spin and orbit of an electron,
one electron and a vibration mode). One starts with two parts 1 and 2.
The basis can always be chosen as if they were independent, and any state
()
()
is a linear combination of products f () (1)g () (2), where f1 , fm and
()
()
g1 , gn are bases for irreps of dimensions m and n respectively:
m
()
()
()
Rfi = k=1 fk (1)Dki (R),

(9.10)
()
()
()
n
Rgj = p=1 gp (2)Dpj (R).
are a convenient basis; they transform according to
R|ij =
n
m

p
k
n
m

k
()
()
|kp Dki (R)Dpj (R)

()
|kp Dkp,ij (R).
(9.11)
Thus, the direct product matrix D() = D() D() with elements
()
()
= Dki (R)Dpj (R) is the representative matrix of R in what is
called the direct product representation. Its characters are
()
Dkp,ij (R)
184
( (R) =
()
Dkp,kp (R) = ( (R)( (R).
(9.12)
kp
It is almost a tongue-twister, but is a far reaching result: the character of

the direct product of two representations is the product of the characters of
the two representations. Great! We can now perform the usual analysis of
the product into irreps by the character orthogonality theorem. Consider for
instance the Group of the Square, and suppose that the two parts 1 and 2
belong to known irreps: their products dene a representation of G and we
wish to reduce it:
C4v
A1
A2
B1
B2
E
A2 B1
B2 B1
E A2
EE
I
1
1
1
1
2
1
1
2
4
C2 2C4
1 1
1 1
1 1
1 1
2 0
1 1
1 1
2 0
4 0
2 v
1
1
1
1
0
1
1
0
0
2d
g=8
1
z
1
Rz
1
x2 y 2
1
xy
0
(x, y)
1
B2
1
A2
0
E
0 A1 A2 B1 B2
(9.13)
In this way we can build a multiplication table for the irreps of C4v .
C4v
A1
A2
B1
B2
E
A1
A1
A2
B1
B2
A2
A2
A1
B2
B1
B1
B1
B2
A1
A2
B2
B2
B1
A2
A1
E
E
E
E
E
A1 + A2
E E E E
+B1 + B2
(9.14)
Here A1 occurs in the diagonal and only there because

n A1 =
1
1 () ()
1 () (R) =
(R) = .
NG
NG
R
(9.15)
9.2.1 Selection Rules

Suppose the amplitude of a physical transition governed by some operator T
depends on the matrix element |T| which is cumbersome and expensive
to compute. It would be disappointing at the end of a computational tour de
force to discover that the matrix element is 0, or that it is identical to another
one that we knew already. Group theory predicts which matrix elements must
be equal and which must vanish by establishing selection rules. Indeed, T is
some tensor component and T generates a representation (T ) ; the states also
9.3 Reduction of the Direct Product Representation
185
belong to representations () and () of the Group G. A matrix element

|T | is an integral, which is invariant, therefore it vanishes unless the
integrand gets some contribution from A1 . If the reduction of () (T )
() to irreps yields (among others ) A1 , then |T |
= 0, otherwise we
can dispense from the calculation of the matrix element since it vanishes by
symmetry. In other terms, the condition for a non-vanishing result is that the
reductions of (T ) and of () () have at least one irrep in common. For
example, if one is interested in electromagnetic transitions in C4v symmetry,
the dipole operator transforms like a vector (x, y, z), so one component is
in A1 and two are in E. Thus one nds that a transition from B1 to B2
is forbidden, while from the E states all the other states can be reached
by the in-plane component of the dipole, while E E is allowed by the z
component.
Most molecules with an even number of electrons have lled shells conguration in the ground state, which is total-symmetric1 . Then, the excited
states that can be reached by photon absorption are those where one of the
components of the dipole is classied. A similar analysis can be done for the
infrared transitions between vibrational states that belong to the same electronic level. Clearly, the phonon vacuum belongs to A1 and the one-phonon
states have the symmetry of the normal mode which is excited. On the other
hand, the Raman eect is a two-photon process in which the system goes
from an initial state |i to a nal state |f by absorbing a photon of polarization vector 1 and emitting another one of polarization vector 2 and
dierent frequency. Emission and absorption are coherent, that is, they are
one quantum process, and the amplitude is given by the matrix element of
the operator

R=
Rpq (2 )p (1 )q
pq
where Rpq xp xq is the Raman tensor. From the symmetry viewpoint, what
matters is that the components of the Raman tensor transform like xp xq ,
and this determines the selection rules. In systems with inversion symmetry,
the normal modes must be gerade or ungerade (even or odd). Only ungerade
modes are infrared active and gerade ones are Raman active. For instance,
infrared and Raman spectra of Benzene have no frequencies in common.
9.3 Reduction of the Direct Product Representation

From the m -times degenerate irrep and the m -times degenerate irrep
one forms a direct product representation () () of dimension m m .
We have seen how the direct product representation is reduced; but it remains to be seen how the basis separates in symmetry-adapted bases. By
1
See the Section 9.5.2on the Jahn-Teller eect for a justication of this statement.
186
inserting a complete set one can go, with a unitary transformation, from the
basis {|ij } of the direct product () () to a basis of functions that
transform according to irrep () of the Group G.
|ij =
m
irreps

g
|r r|ij .
(9.16)
The coecients r|ij are the Clebsh-Gordan (CG) coecients of the

Group; in the case of the rotation Group O(3) they are the well known coecients LM |L1 M1 L2 M2 . For a point group, the CG coecients are even
easier to nd than for O(3). One forms a table R|ij , i = 1, m , j =
1, m , R G, of the eects of the operations on the direct-product basis,
then combines them according to the projectors. In order to decompose the
D matrices, we want a basis change, obtained by inserting complete sets:
()
()
()
Dki (R)Dpj (R) = Dkp,ij (R) = kp|Rij

()
kp|s Dsr
(R)r|ij .
=
(9.17)
rs
This corresponds to the familiar use of CG coecients for the sum of angular
momenta (for an example, see Problem 9.1).
9.4 Spin-Orbit Interaction and Double Groups

Up to now we have considered only the Group of transformations of the
space orbitals, ignoring spin. The spin-orbit interaction makes the problem
less symmetrical: as an example, a Hydrogen level of angular L
= 0, 2(2L+1)
times degenerate, separates in two levels J = L 1/2, degenerate 2J + 1
times.
The spin, alone, or in an A1 orbital, yields a representation of the operators. The rotation around to the z axis by an angle is done by
i

e 2 0
i
z
2
(9.18)
R = e
=
= cos( ) iz sin( )
0 ei 2
2
2
and belongs to the SU(2) covering group of SO(3). For = 2, R = 1. If
we rotate around axis n, since ( n)2 = 1,
R = ei 2 n = cos( ) + i( n) sin( )
(9.19)
2
2
and for = 2, R = 1 anyway2; this is a rotation that commutes with any
other symmetry.
2
a closely similar formula with the 4 4 matrices in place of is the starting

point of the Group theory using Diracs equation. In this section we assume that
the Pauli theory with the relativistic corrections is adequate.
9.4 Spin-Orbit Interaction and Double Groups
187
Following Teller (1929) we indicate with E the 2 rotation around an

arbitrary axis; E = E for a function without spin and E = E and for the
commutes with all the R G.
spinor representation: in both the cases, E
Adding E to the generators of the symmetry Group we obtain the so-called
double Group G which has, along with every R G, also ER.
E alone is a class. A rotation by forms a class with 4 i.e. its inverse

in G, since the two operations are conjugated by a vertical reection or by
may form a separate class.
another rotation. Reections times E
2
=
Since E = E the eigenvalues are 1. By Schurs lemma, D() (E)

()
()
()
D (E) , so (ER) = (R).
Among the new irreps, where the - sign holds, the spinor representation is
=
always there; it is suitable for an electron in a total-symmetric orbital. (E)
2; rotation by about n are represented by the SU(2) matrices D() =
i
e 2 . Since the characters are invariant for unitary
transformations, we
ei 2 0
may take n along the axis z: thus D() =

. Thus the character
0 ei 2
for spin 1/2 is
(1/2) () = 2 cos( ).
(9.20)
2
If the inversion i is in G, it leaves spin and any angular momentum invariant,
so D(i) = D(E) and (i) = 2. The reections and all the improper rotations can be written like products iR .3 In such a way, we can complete the
characters of the spinorial representation4. This describes an electron with
an orbital A1.
The above information is enough to build easily the character table for
G from that of G, without having to work out everything from the multiplication table. Having listed the classes, one can append the irreps of G,
= +() (R). Then one appends the spinor representation,

with () (ER)
and knows how many irreps are missing to reach the number of classes. The
sizes are found by the Burnside theorem, and the LOT allows to nd the

.
characters. As an example, we can extend from C3v to C3v
C3v
E E
A1
A2
E
E1/2
5
6
3
1
1
2
2
1
1
1
1
2
2
1
1
C3 C32
C3 E 3v 3v E
C32 E
1
1
1
1
1
1 1 1
1 1 0
0
1
1 0
0
1
1
i
i
1
1 i
i
Example: a reection in (xyz) (xy z) in the (x,y) plane can be obtained

as a rotation (xyz) (x yz) followed by i.
4
The character tables for the most common double Groups are available in the
literature, and some are reported in Appendix B
188
The A1 , A2 and E irreps have the same charcters as in C3v , and E1/2 is
the spinor representation. Since the classes are 6 two irreps are missing, and
since the sum of the squares of the dimensions must be 12, the new ones
are one-dimensional. By orthogonality, we nd that they are conjugate representations as shown above. At this point the reader could solve Problem
9.4.
In a physical problem, one can begin by classifying the space orbitals
according to G and then extend the theory to G including spin. The direct
product of the orbital irrep by the spinor representation will include the
spin. In general , the result will be reducible in the double Group applying
the LOT. We will be able to thus establish how the spin-orbit interaction
reduces the degeneracy in the problem in issue. For example, if the orbital
belongs to A1 or A2 of C3v , the spinor belongs to E1/2 , and no level splitting
occurs; if the orbital belongs to E the product representation has dimension

, E E1/2 = E1/2 + 5 + 6 .
4, but one nds that in C3v
9.5 Static and Dynamical Jahn-Teller Eect

9.5.1 The Born-Oppenheimer (BO) Approximation
The total Hamiltonian of a system of electrons and nuclei may be written,
with R for set of nuclear and r for the electronic cordinates,
Htot (r, R) = Te (r) + TN (R) + V (r, R)
(9.21)
where Te (r) and TN (R) are the electronic and nuclear kinetic terms, and
V (r, R) = Vee + VeN + VN N
(9.22)
contains all interactions. The Schr

odinger equation
Htot (r, R)tot (r, R) = W tot (r, R)
(9.23)
is intractable. In order to separate variables approximately, one keeps the

nuclei xed (TN 0) introducing the adiabatic electronic Hamiltonian
He (r; R) = Te + V (r, R)
(9.24)
where the R dependence is only parametric; the Schrodinger equation

He n (r; R) = En (R)n (r; R)
(9.25)
yields the adiabatic eigenstates n (r; R) and the potential energy surfaces
En (R). This is the BO approximation, which further assumes that if nuclei
move their evolution is conned to an adiabatic surface, and the harmonic
oscillations about equilibrium correspond to the minimum of E0 (R).
189
Below we adopt a compact notation, writing the nuclear kinetic energy

h2 2
TN =
2M R2 and understanding indices and summations. The nuclear wave
function is then expected to be given by5
2 2
h
+ E0 (R)(R) = W (R).
2M R2
(9.26)
Summarizing the BO approximation:
the positions of the nuclei are external parameters that determine the
Coulomb external potential in which the electrons move, and the Hamiltonian H(R)
the momenta i
hR canonically conjugated to the nuclear positions are
ignored. This assumes that the nuclear masses are innitely large
The equilibrium conguration corresponds to a minimum of the total
energy:
(9.27)
E0 (R) = 0 |H|0
E0
= 0.
R
(9.28)
0
( E
R = 0 represents extremum conditions in all the components of the
nuclear position vectors)
the electronic states are calculated at the equilibrium conguration and
E0 (R) is the potential energy for the nuclear motion.
9.5.2 The Jahn-Teller Theorem

CH4 is a tetrahedron (Td Group), SF6 an octahedron (Oh Group), and
snowakes have beautiful regular shapes. Why are so many molecules and
solids highly symmetric in Nature? Is a maximum symmetry principle to
be discovered? Although the concept of symmetry is central to quantum
physics, from solids down to subnuclear particles, the answer is denitely no.
Bloch waves that carry crystal momentum and Hydrogen states with angular momentum are less symmetrical then the respective Hamiltonians. Many
atoms and nuclei have non-spherical ground states. Yet, the regular shapes
of molecules and solids is striking since there would be an innite number of
ways to move the nuclei and seek for a lower energy conguration; we must understand why such molecules cannot gain energy from any distortion. Within
the Born-Oppenheimer approximation, the Jahn-Teller Theorem provides an
answer, for the 32 point Groups: what matters is degeneracy. The electronic
cloud needs to be degenerate, otherwise the electrons do not have enough
degrees of freedom to lower the symmetry; on the other hand, if there is degeneracy, a suitable deformation always exists, except for linear molecules.
5
In Equation (9.41) below we shall see that this statement although reasonable
is a bit too simple-minded and actually something is missing.
190
The CH4 ground state is non-degenerate in the Td symmetry; but by removing a bonding t2 electron the CH4+ ion distorts until the symmetry Group
becomes Abelian.
Initially proposed as a computational aid which exploits a given symmetry
of the Hamiltonian, Group Theory eventually dictates which symmetries are
allowed or forbidden at all.
Mathematical Formulation of the Problem
of the nuclei in
We want a criterion to decide if a given conguration R
the BO approximation can be the equilibrium one, according to (9.28). We
must minimize with respect to the shifts of the nuclei from to the reference
but rst, we eliminate the rigid shifts of the molecule by
conguration R;
using the normal modes of vibration of Section (8.2) instead of the nuclear
positions. These modes are labeled by the index of the frequency and by
an index i for the degeneracy and multiplicity (the same symmetry can occur
several times). The amplitude of the motion according to a normal mode
is specied by a normal coordinate qi , and we must minimize E(q), where
q {qi } stands for the whole set.
Actually, we do not have any analytic expression of E(q) to dierentiate.
in powers of q, letting
Therefore, we expand the Hamiltonian H around R
H0 = H(R); this enables us to nd the correction to E(q) by perturbation

theory. The correction terms of H yield the interactions between electrons and
vibrations, also called the vibronic couplings. Including up to the quadratic
terms, we nd

()
Vi qi +
Wik qi qk .
(9.29)
H = H0 +
i
ik
The equilibrium condition is that the average rst-order corrections to the

energy vanish.
In the Born-Oppenheimer approximation, Vi operates on electrons; the
true Hamiltonian depends quadratically on the momenta p
i canonically conjugated to the qi . If we take into account the momenta, we go beyond the
Born-Oppenheimer approximation; one then speaks about the dynamical
Jahn-Teller eect, see next Section. Note that Vi and qi must transform
in the same way, as H must be a scalar and in the table of the direct products A1 appears only on the diagonal. We do not mind ViA1 here: such terms
do not distort the symmetry. 6 The equilibrium condition is

()
Vi qi = 0.
(9.30)
H =
i
()
Since the qi
are linearly independent we really need

0 |Vi |0 = 0.
(9.31)
6
Actually, they could in principle only increase it: with a A1 motion the water
molecule could be straightened and become a Dh molecule.
191
Non-degenerate Case
If the ground state is not degenerate, (0 0 ) = A1 and 0 |Vi |0 = 0 for

= A1 . So, all the nuclei move, keeping the symmetry of the molecule, until
(A )
0 |Vi 1 |0 = 0. This 0 is not due to the symmetry, but to the existence of
a minimum of energy versus breathing mode coordinate: we can vary a bond
length or an angle until the condition is satised.
Degenerate Case
(0)
(0)
(0)
When H0 = E0 is solved by several , we must apply degenerate perturbation theory, diagonalizing the perturbation matrix with elements
(0)
(0)
|H | . If the eigenvalues do not vanish all identically, one predicts
()
a splitting linear in the qi of the degenerate level, a lowering of the symmetry and a distortion of the molecule. The only Hermitean matrix that has
all eigenvalues equal to 0 is the null one. Thus, for equilibrium we need to
satisfy the strong condition
(0) |Vi |(0) = 0 , , .
(9.32)
Now we must examine the normal modes occurring in the assumed geometry
in order to see if any generates matrix elements that destroy the symmetry.
(0)
Let 0 denote the irrep where belongs. The Jahn-Teller eect is caused
by a normal mode belonging to an irrep
= A1 which is contained in 0 0 .
In such cases there is no reason why the matrix elements vanish, and the
molecular conguration is unstable7 .
In 1937, Jahn and Teller demonstrated that for a non-linear molecule
with degenerate ground state irrep 0 , a vibration always exists such that

= A1 is contained in 0 0 ; this implies that the fundamental electronic
terms of non-linear molecules are not degenerate (even if not necessarily totalsymmetric). The proof is obtained by repeating for all the 32 point Groups8
the same analysis that we now exemplify in the case of Td .
Example
We resume CH4 (Sections 8.1.2,8.2);
Td
A1
A2
E
T1
T1
7
E
1
1
2
3
3
8C3
1
1
1
0
0
3C2
1
1
2
1
1
6d
1
1
0
1
1
6S4
g = 24
1
r2
1
0 (3z 2 r2 , x2 y 2 )
1
(Rx , Ry , Rz )
1
(x, y, z)
One speaks about pseudo-Jahn-Teller eect when the distortion is caused by a

vibronic coupling of close energy levels of the same symmetry.
8
when other Groups are appropriate the situation needs verication.
192
The degenerate irreps of the Td Group they are E, T1 and T2 , and

E E = A1 A2 E, T1 T1 = T2 T2 = A1 E T1 T2
(9.33)
We saw vibr = A1 + E + 2T2 . Therefore at least one Jahn-Teller active

vibration , able to distort the molecule, exists for any degenerate state. The
considerable width of the t2 level of CH4 seen in photoemission is due just to
the fact that the ionization excites the vibrations of the molecule compelling
the nuclei to seek a new equilibrium new position. The distortion must remove
the degeneracy completely. The basis of an irrep of a Group G is also a basis
of a representation of every subgroup of G, since by eliminating some matrices
the remaining ones continue to obey the multiplication Table of the subgroup;
but usually the representation is reducible. As an example, from Td one can
go to D2d , e.g. by stretching the tetrahedron along the z axis. This stretching
completely resolves the degeneracy of electronic states of symmetry E, but is
not enough for T2 states. Under Td , the characters are:
g = 24
Td E 8C3 3C2 6d 6S4
E 2 1 2 0 0 (3z 2 r2 , x2 y 2 )
T2 3 0 1 1 1
(x, y, z)
while the Character Table of the Subgroup is:
D2d
A1
A2
B1
B2
E
E
1
1
1
1
2
C2 2C2
1 1
1 1
1 1
1 1
2 0
2d
1
1
1
1
0
2S4
1
1
1
1
0
C3 is broken; for Td there is no distinction between C2 and C2 , therefore

one of the 3C2 one goes under C2 , and two under C2 . We then consider the
operations of Td that survive in the D2d subgroup. The D(R) are the same,
even if they are not irreducible any more, and they have the same traces. We
place the characters of the surviving operations under the classes of D2d , and
analyze the representation. For the degenerate ones one nds
D2d E C2 2C2 2d 2S4 analysis
E(Td) 2 2 2 0 0 A1 B1
T2 (Td ) 1 1 1 1 1 E B2
therefore the distortion is enough in the rst case but not in the second. This
analysis cannot determine the type of distortion uniquely.
Linear Molecules
Why is CO2 straight? For the linear molecules the theorem does not apply.
Let denote the electron angular momentum parallel to the molecular axis.
9.6 Non-Adiabatic Operator
193
In molecules with = 0 degenerate the electronic ground state is unique and

there is no problem. In those with || 1, the states are degenerate.
Therefore between the unperturbed states = 2, 4, 6 . . . In order to resolve
the degeneracy it would be necessary to fold the molecule;
but no fold can

occur since the vibronic matrix element vanishes: V = 0. In fact,
a shift of a nucleus outside of the axis is a vector (cos , sin ); the matrix
elements are subject to the rule of selection rule = 1. Therefore the
matrix elements vanish and the degeneracy remains.

In the BO approximation the evolution is conned to an adiabatic surface9
and one could expect to derive Equation (9.26) for the nuclear motion from
the ansatz:
trial (r; R) = (R)0 (r; R).
(9.35)
This is not exactly true, as we shall see shortly. Let us look for the best
solution of the form (9.35) variationally, with a given 0 (r; R), looking for
the (R) that yields the minimum of the energy
E = Htot = TN + He .
(9.36)
Again we adopt a compact notation, writing the nuclear kinetic energy

h2 2
TN =
2M R2 and understanding indices and summations. One looks for
the unconditional minimum of

h2
2
+
dR (R)E0 (R)(R)
F [] =
drdR 0
0
2M
R2

(9.37)
W dR (R)(R)
where W is a Lagrange multiplier that ensures normalization. Using
2
2
2
|0 |
+
|
|

=
|

|
+
2
|0
0
0
R2
R2
R R
R2
9
If in R0 one has degenerate wave functions,

n (r; R0 ), n = 1 . . .
(9.34)
with
He (r; R0 )n (r; R0 ) = E(R0 )n (r; R0 ),
the topology of the surfaces is important. For conical intersections, when the surfaces cross each other, one speaks of Jahn-Teller eect; when surfaces touch at extremal points one speaks of Renner-Teller eect [5]. However, the JT theorem limits
the occurrence of glancing intersections to linear molecules (and to cases when the
gradient is accidentally vanishing or particularly small for reasons independent of
Group theory).
194
one nds
F [] =

+

+
2
h
2M

dR
dR

2
+2
R2
0
dr0
R2
drdR 0
0
R R
(9.38)
dR (R)E0 (R)(R) W
dR (R)(R).
We vary the bra (variation of bra and ket produces identical results).

h2 2
h2
0
F = dR (R)
dr0
2
2M R
M
R R

2
2
h
(R)
W
)(R)
= 0. (9.39)
(R)
+
(E
dr0
0
2M
R2
This implies:

h2
2 2
h
0
+ E0 (R)(R)
dr0
2M R2
M
R R

2
h2
(R) = W (R)
dr0
2M
R2
(9.40)
The minimum condition is

2 2
h
+ E0 (R)(R) + (R)(R) = W (R)
2M R2
where the non adiabatic operator appears

2
h2
h2
0
0
(R) =
.
dr0
dr0
M
R R 2M
R2
(9.41)
(9.42)
This has been often ignored in the literature; the reasons are that 1) the rst
contribution, averaged over real electronic wave functions vanishes since

1
0 (r; R) =
dr0 (r; R)
dr0 (r; R)0 (r; R);
R
2 R
m
). It will be apparent shortly
2) the second contribution is small (of order M
that such reasons are not generally as safe as they may appear to be.
9.6.1 Dynamical Jahn-Teller Eect

At strong vibronic coupling , the energy surfaces have Ndeg deep and distant
minima and the nuclear degrees of freedom can hardly tunnel between them.
195
Then, the kinetic energy of the nuclei does not play a role, and one can
observe a static JT eect with broken symmetry. At weak coupling, one
speaks about dynamic JT eect and the system oscillates between several
minima; the overall symmetry remains unbroken10. The JT approximation
to the solutions of Htot tot = W tot ( Equation 9.23) uses as a reduced basis
set the Ndeg degererate adiabatic functions in the symmetric conguration
R0 :
Ndeg

n (R)n (r)
(9.43)
tot (r, R)
n
here n (r) are assumed known, and one seeks the nuclear amplitudes n (R).
Substituting into Equation (9.23) and taking the scalar product by m | leads
to

(r; R0 ) [Htot W ]
n (R)n (r) = 0,
(9.44)
drm
that is,

drm
n (R)n (r) = 0,
TN + Te (r) + V (r, R) W
:
;<
=
where Te (r) + V (r, R) = He (r, R). By orthogonality,

2 2
h
m (R) +
n (R)
2
2M R
n
drm
(r; R0 )He (r, R)n (r) = W m (R).
(9.45)
Dening

Vmn (R) =
drm
(r; R0 )He (r, R)n (r)
(9.46)
we get the coupled problem

2 2
h
(R)
+
Vmn (R)n (R) = W m (R).
m
2M R2
n
(9.47)
or in matrix form
10
The distinction between dynamical and static JT eect usually depends only
on the time scale of the experiment. For example, (Cu6H2 O)++ ions look perfectly
octahedral when observed at room temperatures in EPR experiments, but below
20 0 K or with fast spectroscopies it is seen that this symmetrical conguration is
the time average of stretched and compressed ones as the top and bottom H2 O
molecules oscillate up and down. On the other hand, X-Ray diraction at room
temperature shows that (CuBr6 )4 ions is a tetragonally distorted, stretched octahedron; the latter is classied as an example of static JT eect. It should be kept
in mind, however, that the JT approximation may fail completely, as it does in
strongly correlated models with strong electron-phonon coupling [71].
196
V11 (R) .
.
.
=
+
.
.
2M R2
V1 (R) .
2
HJT
.
.
.
.
. V1 (R)
.
.
.
.
.
. V (R)
(9.48)
9.6.2 How the E Hamiltonian arises

In the popular E problem two degenerate potential energy surfaces with
electronic wave functions (x , y ) of E symmetry interact with degenerate
vibrations of the same symmetry 11 ; let (qx , qy ) be the normal coordinates.
Neglecting anharmonic terms, (9.29) reads
2.
He = qx V (x) + qy V (y) + Kq
(9.49)
Here, V (x), V (y) transform according to E and represent the operator poten must belong to A1 and
tial due to the phonons acting on the electrons; K
2
2
2
q = qx + qy . Then, (9.48) becomes
HJT =

2 2
h
h2 2
(V (x))xx (V (x))xy
+
q
x
(V (x))yx (V (x))yy
2M qx2
2M qy2

(V (y))xx (V (y))xy
2.
+qy
+ K q
(V (y))yx (V (y))yy
(9.50)
Group theory dictates the form of the vibronic interaction; for illustration, we
adopt the geometry of Figure 7.1 of Section 7.2.2. (x, y) is a basis for E; for
the present purpose, however, we shall use an alternative basis for E in C3v ,
with the same D matrices, namely, (fx , fy ) = (2xy, x2 y 2 ). It is is evident
that 2xy transforms like x and x2 y 2 like y in the chosen geometry, since
2xy is odd and x2 y 2 even under the 1 reection. For any basis (fx , fy ) of
E, we know from (7.46) that fx |y = fy |x = 0 and that fx |x = fy |y .
Therefore, if now (x , y ) are electronic states and are a basis for E, x2 + y2
belongs to A1 and for the V (x) elements we nd:

2 drx y V (x),
dr(x2 y2 )V (y) = 2 =

2
2
dr(x2 + y2 )V (y) = 0 = dr(
(9.51)
x + y )V (x),

2
2
dr(x y )V (x) = 0 = 2 drx y V (y).
Thus, V (x) has equal o-diagonal matrix elements on (x , y ) and nothing
on the diagonal and V (y) has opposite diagonal elements and 0 o-diagonal;
this is clear already when one considers the a parity. Thus, the form of the
JT hamiltonian in the E problem is:
HJT =
11
2 2
h
h2 2
x2 + qy2 ).
+ [qx x + qy z ] + K (q
2M qx2
2M qy2
The N a3 molecule oers a simple example of this sutuation.
(9.52)
9.7 Wigner-Eckart Theorem with Applications
197
The second-quantized version is:

5
6
h + (ay ay + 1/2)
h + (ax + ax )x + (ay + ay )z ;
HJT = (ax ax + 1/2)
(9.53)
(two levels and two-bosons problem) and is among those exactly solved by
the method of Excitation Amplitudes (See ([77]) and Section 14.77 ).
9.6.3 Nuclear Wave Functions Cannot be Taken Real
The BO wave functions are readily found in the M limit, and the result
is very interesting. Letting qx = q cos(), qy = q sin()),
2 = qM + K q
2.
HJT + [qx x + qy z ] + K q

with
M=
sin() cos()
cos() sin()
(9.54)

.
2 is an additive constant, and HJT is the Hamiltonian 12 for a spin

K q
in a magnetic eld B = (qx , 0, qy ). The above matrix M has eigenvalues

1, and the potential energy surfaces are obtained by rotating two intersecting parabolas around the energy axis. The eigenstates corresponding to the
eigenvalues 1 of M are

cos(/2 + /4)
sin(/2 + /4)
,
+ () =
. (9.55)
() =
sin(/2 + /4)
cos(/2 + /4)
These wave functions change sign under a 2 rotation. This is normal in
spin-dependent problems (see Section 9.4) but is striking and unacceptable
in the present problem; of course, the nuclear wave function must be unique.
Indeed, one can have the unique wave function, by inserting a phase factor,

cos(/2 + /4)
sin(/2 + /4)
i/2
i/2
() = e
, + () = e
.
sin(/2 + /4)
cos(/2 + /4)
(9.56)
but then the solution is complex. This means that the Berry phase (See Sect.
??) is coming into play. For further information on this and other JT systems
one may consult reference [5].
9.7 Wigner-Eckart Theorem with Applications

Let |i > and |j > denote the components of bases of irreps and
()
(that could also coincide) and Tp the p component of an irreducible tensor
12
The Hamiltonian given in Ref. [5] is obtained by a harmless rotation qx
qy , qy qx in q space.
198
RTp() R1 =
()
Tq() Dqp
(R).
(9.57)
As the components i,p and k vary, one nds a number of matrix elements
()
< i|Tp |k > that are all connected by the Wigner-Eckart theorem
Theorem 9.
i|Tq() |k = T () i|pk ,
(9.58)
where the reduced matrix element < T () > does not depend on the
components;i|pk is the Clebsh-Gordan coecient.
Since the Clebsh-Gordan coecients are mere geometry, the dynamics
enters through the reduced matrix element. A formal proof is given in Appendix C, but an intuitive argument is also useful. A very crucial point in
that the tensor and the function spaces must be irreducible. If you know one
basis function of an irrep you can build all of them by the Group operations
and orthogonalization; and the same is true of the tensor components; then
it is at least very plausible that all the matrix elements can in principle be
obtained by symmetry from the knowledge of any non-vanishing one of them.
Based on this, the theorem is then most simply understood. Suppose that by
()
direct computation we obtain a particular element i0 |Tp0 |k0 = Q
= 0.
Next, we decide to look at the system from a new transformed reference ob()
tained by some R G; then, |i0 , |k0 and Tp0 transform to linear
combinations of all the components, still remaining in their irreps; however
the matrix element is still Q. Varying R, we can write a system of linear
equations linking the components, all with the same right-hand-side. Are all
those equations must be compatible, and since there are enough relations to
()
determine < i|Tq |k >, one can solve and each matrix element must be
proportional to the only r.h.s. Q. Any two tensors T and T of the same irrep generate the same system, except that the r.h.s. are Q and Q; so they
must yield proportional results. There is a particularly simple tensor dened
by Tp |k = |pk which yields the theorem with < T () > 1. For all
the other tensors the theorem holds with some reduced matrix element.
Simple Applications
The theorem reduces the calculation of a tensor to one of one its components.
It implies less that in a symmetry adapted basis all the irreducible tensor
operators have the same elements matrix elements, up to a multiplicative
constant. We can choose the most comfortable operator (as long as it does
not have a null reduced matrix element). As an example, in O(3), j = 1 labels
(1)
the irrep of vectors and Tq , q = 0, 1 is a vector operator in polar form.
Therefore, its matrix elements are proportional them to those of j, that they
are easy to calculate in this basis:
9.8 The Symmetric Group and Many-Electron States
jm|T |j m = Cjm|J |j m .
199
(9.59)
To determine the constant C, it is enough to compute explicitly jm|T |jm .

Often it is easier to calculate the matrix elements of the scalar T J . Indeed,

jm|T J |jm =
jm|T |jm jm |J |jm
(9.60)
m
but this is simply

C

jm|J|jm jm |J|jm = CJ(J + 1).
(9.61)
m
For example the nuclear quadrupole moment is the tensor

Qik =
3xpi xpk ik rp2 ,
(9.62)
where xpi is a Cartesian component of the radius vector r p of proton p in

the nucleus. It is a traceless symmetric tensor and we can replace it with the
tensor with components
Qik =
3Q
2
Ii Ik + Ik Ii ik I 2 ,
2I(2I 1)
3
(9.63)
built with a constant Q (named the quadrupole moment) and withthe components of the nuclear spin I. In a similar way one can build an atomic
quadrupole moment tensor from J .

The spin-orbit interaction is HSO = i l(i)s(i) in many-electron atoms,
where l(i) is the orbital angular momentum of electron i , s(i) is its spin;.
Since l(i) is a vector like the total orbital angular momentum L and s(i) is
a vector in spin space like the total spin angular momentum S, the WignerEckart theorem allows to write
LM LS MS |HSO |LM LS MS = ALM LS MS |L S|LM LS MS ,
(9.64)
where A is a constant.
9.8 The Symmetric Group and Many-Electron States

()
For an N electron system, let i (x1 . . . xN ) denote an amplitude depending

on the space coordinates only and transforming according the component i
of irrep (): if P denote a permutation,
()
Pi
m

j
()
()
j Dj,i (P).
(9.65)
200

()
Let q (1 . . . N ) denote an amplitude depending on the spin coordinates

only and transforming according the component q of irrep (), such that
P()
=
q
m
()
()
n Dn,q (P).
(9.66)
When permuting the electrons, i.e. doing the same permutation on spins and
coordinates, the wave function must go into itself up to a sign, therefore
components must enter symmetrically, like in
=
m
() ()
k k .
(9.67)
Is a fully antisymmetrized wave function? One preliminary condition is

that both irreps must have the same dimension m. Next, we impose
P = ()P =
m
()
()
[Pk ][Pk ];
(9.68)
using (9.65,9.66) we nd that this requires

m
()
()
Dj,k (P)Dn,k (P) = ()P nj .
(9.69)
Since the identity permutation is represented by the identity matrix, it is

readily seen that this holds if
()
()
Dj,k (P) = ()P Dk,j (P 1 ).
(9.70)
Two irreps that satisfy the condition (9.70) are conjugate or associate. We
saw in Sect. 8.5 that the spin eigenfunction have a Young tableau consisting
of up to two lines. The following [N M, M ] two-line tableau (left) with
N M integers denoted ai aN M on the top line and M integers denoted
bi bM on the bottom line is suitable for a spin eigenfunction symmetry of
S = 126(N 2M ) (Equation 9.71, left); then, the conjugate tableau is
5spin
M
2 , 1N M as shown below (right).
a1 a2 . . . . . . . aN M
b1 b2 . . . . . bM
a1
a2
.
.
.
.
.
.
.
.
aN M
b1
b2
.
.
.
.
.
bM
(9.71)
9.9 Seniority Numbers in Atomic Physics
201

Let m (x) denote the d orbitals of an atom; under the rotation of an angle
iL
, that we denote R = e h O+ (3), they transform according to

l
R m (x) =
n (x)Dmn
(R ),
(9.72)
n
l
(R ) is the Wigner matrix; therefore {m (x)} is a basis
where l = 2 and Dmn
l
(R ) also play a second role: the
of the irrep l = 2 of O+ (3). However, Dnm
transformation

l
m (x) Rm (x) =
n (x)Dmn
(R ),
(9.73)
n
qualies m (x) at xed x as polar components of a vector under SU (2l+1) =

l
SU (5), and so Dnm
(R) SU (2l + 1). For the same reason
Tmn (x1 , x1 ) = m (x2 )n (x2 )
Tmnp (x1 , x2 , x3 ) = m (x1 )n (x2 )p (x3 )
(9.74)
for xed points xi , are components of tensors of rank 2, 3, , respectively.

The permutation of two indices of a tensor commutes with the operations of
the Group; so the tensors that belong to irrep of SN of permutation of the
indices are the bases of the irreps of SU(5). With 2 indices, the possibilities
are
= Tmn + Tnm
T
mn
T
m
n
= Tmn Tnm
(9.75)
is associated with triplet states, while T

to singlets. Let us see how
m
mn
n
we can classify the singlet states of d2 conguration using the polar Tmn of
. In obvious notation (m1 , m2 ) for the two(9.74) projected into
electron determinants, one nds by the shift operators
T
1
1
GM=4 = (2, 2)
GM=3 = (1,2)+(2,1)
2
GM=2 =
6[(0,2)+(2,0)]+4(1,1)
28
(9.76)
and by orthogonality
1
DM=2 =
2[(0, 2) + (2, 0)] 3(1, 1)
;
5
(9.77)
202
then
1
[(1,2)+(2,1)]+ 6[(0,1)+(1,0)]
14
[(2,2)+(2,2)]+4[(1,1)+(1,1)]+6(0,0)
70
2[(2,2)+(2,2)]+[(1,1)+(1,1)]2(0,0)
14
GM=1 =
GM=0 =
DM=0 =
(9.78)
and nally
1
S=
[(2, 2) + (2, 2)] [(1, 1) + (1, 1)] + (0, 0)
.
5
(9.79)
The form of 1 S is remarkably simple: it is just a scalar product in the

polar form (9.8). Let us see how this calculation extends to the case of more
electrons, and higher rank tensors. With 3 indices, the irrep .. .. .. does not
take part in the construction of the states to 3 electrons since the function of
..
spin would have to be .. which cannot be antisymmetrized totally on and
..
..
.. ...
. Thus one is left with .. (for quartets) and
(for doublets). Actually
..
..
there are two doublets:
T
mn
p
= A13 S12 Tmnp = Tmnp + Tnmp Tpnm Tnpm

(9.80)
mp
n
= A12 S13 Tmnp = Tmnp + Tpnm Tnmp Tpmn
SU (5) has the subgroup O+ (5) of proper rotations in 5 dimensions; on a

cartesian basis its representatives are orthogonal matrices such that aT = a1
that is amx amy = xy . Transforming any cartesian tensor according to

Tmn
= amx any Txy
and then taking the trace, that is, setting m=n and summing over n, one gets

Tnn
= anx any Txy = xy Txy = Txx ;
in other terms, the trace Tmm is invariant. For a tensor whose cartesian
components Tmn = vm wn are the products of cartesian vector components
the trace becomes the scalar product Tmm = vm wm . Thus, in O+ (5) the
traces acquire particular meaning. The invariance remains obviously true with
any number of components; for instance taking the trace of

= amx any apz Txyz
Tmnp
203
one nds

= amx amy apz Txyz = xy apz Txyz = apz Txxz ,
Tmmp
thus Txxz is a vector. On the other hand a cartesian tensor of the form Txyz =

= amx any apz xy vz = amx anx apz vz =
xy vz is transformed as follows: Txyz
mn apz vz , and retains its form.
The and the T of Equation (9.74) are tensor components also for the
subgroup, except that since they are polar tensors the rule must be slightly
modied. In order to agree with the scalar product (9.8), the polar form of
the trace reads

()m Tm,m .
(9.81)
T rT =
m
For a cartesian tensor Tmnp of rank 3, one denes the traces:

T r(1,2) T = Tmmp
T r(1,3) T = Tmnm
T r(2,3) T = Tmpp .
(9.82)
Since the Tr operation commutes with the operations of the Group O+ (5),
many irreps of SU(5) are no longer irreducible. Vectors are sent to vectors and
are bases of an irreducible representation of O+ (5). Traceless tensors are sent
to traceless tensors and are bases of irreducible representations of O+ (5).
T
is traceless, while T
must be reducible. In fact, all the d2 states
m
mn
n
above belong to
of SU(5), but in O+ (5) the invariant 1 S is classied
in the (0, 0) irrep (the notation means: no boxes in either the rst and the
second line). The state with no electrons is already invariant, so this singlet
of SU(5), in O+ (5) one
is assigned the seniority number v = 0. From
can extract a traceless tensor which is classied in the irrep (2,0). All irreps
of O+ (5) are labelled by (1 , 2 ), 1 2 ; the parentheses denote the space
of traceless tensors. Since (2,0) cannot be made with fewer than 2 electrons
1
D and 1 F receive v = 2.
12
With
the 1-3 trace vanishes and the other two are equal in absolute
3
is a 5-component vector, so it deserves
value; clearly, vp = T r(1,2) T
mn
p
to be classied in (1, 0); it is obviously a doublet, so it corresponds to 21 D.
The other doublet states 23 P, 23 D, 23 F, 23 G, 23 H will correspond to the traceless
v
tensor T(2,0) T
mn5 p . This does not occur with less than 3 electrons
mn
p
nor with orbital angular momentum < 2. These states deserve the seniority
number v = 3 along with 43 P and 43 F that are classied in (1,1) and are also
unprecedented.
204
Problems
9.1. In a molecule of C3v symmetry, two electrons are in orbitals of the irrep
E. Find the space part of the singlet wave function of A1 symmetry (if any
exists) and the relevant CG coecients.
E. Find the space part of the singlet wave function of A2 symmetry and the
relevant Clebsh-Gordan coecients.
E. Find the space part of the singlet wave function of E symmetry and the
relevant Clebsh-Gordan coecients.

9.4. Build the character table for C4v
.
9.5. Consider the Cu++ ion, with one conguration 3d9 . How are the J =
3/2, 5/2 levels split by a square planar D4 environment?
Part III
More on Green Function Techniques
10 Equations of Motion and Further

Developments
10.1 Equations of motion for the interacting propagator

We used EOM several times (see Sections 4.3,4.4,5.1.2); now we extend the
approach used in Equation (4.39) for the free propagator to interacting problems. Using the many-body Hamiltonian (1.63) one readily obtains

[ (x), H] = h0 (x) (x) +

dy + (y) v (x, y) , (y) (x) ,
(10.1)
with all the operators in the Heisenberg representation (h0 is a rst-quantized

one-body operator), with the notation x = (x, tx ). We multiply on the left
by (z) and perform an interacting ground state average:

i h0 (x) (z) (x) =

t

dy w (x, y) , (z) (y) (y) (x) , tz > tx . (10.2)
The order of operators in the l.h.s. is appropriate for g (T ) (Equation (4.20))

if z is later, that is, tz > tx , and we assume this for the moment. For the
r.h.s. we must dene the time-ordered two-particle Greens function1 and
understanding the spin indices, we write, still in terms of Heisenberg operators
G2 (x1 , x2 , x3 , x4 ) = T [(x1 )(x2 ) (x3 ) (x4 )] .
(10.3)
Comparison with (10.2) is easier using the fact that Heisenberg operators
anticommute under T ordering , hence it holds that G2 (x1 , x2 , x3 , x4 ) =
T [ (x4 ) (x3 )(x2 )(x1 )] and
(z) (y) (y) (x) = ()G2 (x, y, y + , z)
(10.4)
where the notation y + means that although ty = tx that particular y is just

later.
1
Some authors use dierent orderings of the arguments.
208
10 Equations of Motion and Further Developments

i

H0 (x) ig (T ) (x, z) =
t
dy w (x, y) , ()G2 (x, y, y + , z).
(10.5)
Extending the calculation to tz tx one nds that the Green function

obeys
[i
H0 ]g (T ) (xt, x t ) = (xx )(tt )i dx1 v(xx1 )G2 (xt, x1 t, x1 t+ , x t )

t
(10.6)
which is often written in the literature using lighter notations like

d
H0 (1) G(1; 1 ) = (11 ) i d2v(1, 2)G2 (1; 2|2+ ; 1 )
(10.7)
i
dt1
One has also the adjoint equation

22
d
U (2) G(1; 2) = (1 2) i G2 (1; 1 |1+ ; 2)V (1 2).
+
i
dt2
2m
(10.8)
10.1.1 Equations of Motion and Ground-State Energy
As a byproduct, using (10.4, 10.5) one obtains an expression yielding the
ground state energy in terms of the one-body Greens function. Setting z = x
with tz = t+
x and integrating over x one nds:

H
dx lim+ lim
(x)
g (T ) (x, t, x , t )
i
0
t
t t r r

= i dxdyv(x, y) (x) (y)(y)(x) = 2iv .

(10.9)
The average of H0 can be obtained from Equation (4.26). So, we get the
exact ground state energy (lower sign for fermions):

i
E = H0 + V =
i + h0 (x) g (T ) (x, t, x , t )
dx lim+ lim
2
t
t t r r
(10.10)
10.2 Time-Dependent Problems

The EOM are rewarding when the Hamiltonian depends in time. Let
H = H0 + H1 (t) =
k nk +
Vk,k (t)ak ak .
209
(10.11)
k,k
r

For igk,k
(t, t ) = (t t )[ak (t) , ak (t )]+ one nds
gk,k (t, t ) = (t t )k,k + (i)(t t )[ iak (t) , ak (t )]+ . (10.12)
t
t
Since
ia k = [ak , H] = k ak +
Vk,k (t)ak
(10.13)
k
one nds the EOM

i
r

gk,k

r

r

= (t t )k,k + k gk,k
Vk,p (t)gp,k
(t, t ) +
(t, t ).
t
p
(10.14)
This must be solved with the initial condition

r
igk,k
(t, t 0) = k,k .
(10.15)
r
No information about the lling enters the problem, and gk,k
is actually a
one-body quantity which does not depend on the lling. It is no harder to
calculate than a one-body wave function, even in time-dependent problems.
a

The same is true for advanced one, igk,k
(t, t ) = (t t)[ak (t) , ak (t )]+ .
For later use, we are interested in the time-ordered Greens function, dened
as usual by
(T )
(10.16)
igk,k (t, t ) = T ak (t)ak (t ) .
Now, the initial conditions

(T )
gk,k (0, 0 ) = ikk (1 fk ),
(T )
gk,k (0, 0+ ) = ikk fk
know where is the Fermi level; note the characteristic discontinuity

(T )
(T )
i gk,k (t, t + 0) gk,k (t, t 0) = ak ak + ak ak = k,k .
(10.17)
(10.18)
Let us calculate the t derivative of Eq.(4.19), taking into account that the
functions contribute
(t t )ak ak + ak ak = (t t )k,k .
Using (10.13), one nds
(T )
gk,k
t
(T )
= (t t )k,k + k gk,k (t, t ) +

p
(T )
Vk,p (t)gp,k (t, t ).
(10.19)
210
This EOM is the same as (10.14), the dierence is in the initial conditions. I
stress that g (T ) depends on the Fermi level and can describe genuine manybody eects, so it is important to be able to express it in terms of the much
simpler g (r) . This is achieved as follows:

(T )
r
a

gk,q
(t, )gq,k
(10.20)
igk,k (t, t ) =
(, t )q (t, t ),
q
where is any time such that < t, < t and

q (t, t ) = (1 fq )(t t ) fq (t t) = (t t ) fq .
(10.21)
This solution was rst obtained by Ref. ([48]) by the Keldysh fomalism [84];
I found the simple derivation below and used the result to re-formulate the
theory of transport[63] (See Chapter 13.6.2). Since retarded and advanced
functions do not depend on lling, we may take the system empty, denoting
the vacuum by |0 , and write
r

ik (tt )
gk,k
,
(t, t ) = i(t t )0|ak (t)a (t )|0 = i(t t )kk e
k
a

ik (tt )
.
gk,k
(t, t ) = i(t t)0|ak (t)a (t )|0 = i(t t)kk e
k
Hence,

r
a
gk,q
(t, )gq,k
(, t) = k,k ,
(10.22)
r
a
fq gk,q
(t, )gq,k
(, t) = k,k fk .
(10.23)
Thus, the initial conditions are obeyed; the EOM are also satised because
r
(t, ): the (t ) term does not arise because < t,
they are obeyed by gk,q
but the required comes in from the derivative. Thus, (10.20) is readily
seen to satisfy EOM and initial conditions and is the exact solution.
10.2.1 Auger Induced Ionic Desorption: Knotek-Feibelman
Mechanism
Desorption (Section 6.2) is a process of emission of atoms, molecules or ionic
species that previously belonged to a surface or were adsorbed (i.e. chemically
bound to it). The amount of ions emitted depends on the electronic properties
of the species in a striking way. For example, if a Ag surface is bombarded
with Ar+ in the KeV range, some Ag + ions can be collected by a mass
spectrometer, but the ion yield is often much lower than the K + signal arising
from traces of K on the surface. This is because Ag + has a much lower
probability than K + of escaping without being neutralized.
When ionic surfaces are irradiated with X-rays one observes the desorption
of O+ or other positive ions that originally were anions[113]. When O
211
initially bound by the Madelung potential becomes O+ by Auger processes, it

starts being repelled from the surface; however, the copious ion yield remained
quite a puzzle for some time. The reason is that the desorption process occurs
on the time scale " 1013 s of vibrations; electrons move on the time scale of
the inverse band width W and have a lot of time for bond healing.
Knotek and Feibelman pointed out[113] that O+ remains O+ due to
the same mechanism that localizes the two-hole resonances[76] (Section 6.2).
Sometimes it is the easy production of neutrals that needs explanation. For
example, F desorbs from uorides as F and F ; in the solid it is basically F ,
but if some process produces neutral F , the Madelung force holding the ion
in the solid disappears and desorption starts; however desorption is a slow
process and one would expect the ion to quickly resume the electron. The
writer proposed[13] to compute the ion yield by a time-dependent Anderson
model

k nk,
H(t) = (t)(n0+ + n0 ) + U (t)n0+ n0 +
k,

+
[Vk (t)a0, ak, + h.c.];
(10.24)
k,
at time t=0 two holes are created in the adatom orbital 0 and the consequent
desorption causes the time dependence. The band was assumed initially lled
and the amplitude for the two holes to be still on-site at time t is
N (t) = a0 (t)a0+ (t)a0+ (0)a0 (0) (t),
(10.25)
averaged over the hole vacuum. I solved the two-body problem by the EOM
method. Consider the correlation functions (averaged over the vacuum)
(t, ) =
k (t, ) =
(t, ) = ia0+ (t)a0+ ( ) ,
(10.26)
k (t, ) = iak+ (t)a0+ ( ) ,

ia0 (t)a0+ ( )a0+ (0)a0 (0) ,
iak (t)ak+ ( )ak+ (0)a0 (0) ,
(10.27)
(10.28)
(10.29)
where the left hand sides are spin-independent; we want to compute

N (t) = i (t, t)(t).
(10.30)
From the equations of motion
one obtains

ia 0 = (t)a0 + k Vk (t)ak + U (t)a0 n0 ,
ia k = k ak + Vk (t)a0 (t),
(10.31)
i t
(t, ) = (t)(t, ) + k Vk (t)k (t, ),
k (t, ) = k k (t, ) + Vk (t)(t, ).

i t
(10.32)
212
The initial conditions are (t, t) = i, k (t, t) = 0. It is convenient (although

not essential) to set Vk (t) = Vk (0)u(t); then, introducing the self-energy of
the local Greens function of H(0),
|V(0)|2
, = 0+
() =
(10.33)
k + i
k
we obtain
(t, ) = (t)(t, ) + u(t)

t
dt u(t )(t t )(t , ).
(10.34)
Moreover,
i
(t, ) = ( ) (t, ) +
Vk ( )k (t, )
+ iU ( )g(t, ) (, ),
k (t, ) = k k (t, ) + Vk ( )k (t, ).
k (t, = 0) = 0, and we nd the closed equation

i
(10.35)
(10.36)
(t, ) = ( ) (t, ) + iU ( )(t, ) (, )
+u( )
0
d u( )( ) (t, ).
(10.37)
This can be integrated[13]:

(t, )( ) = i(t, 0(, 0)( )
+i
d (t, )(, )U ( )G( , )( ).
(10.38)
Finally,
2
N (t) = (t, 0)(t) iU
dt 2 (t, t )U (t )N (t ).
(10.39)
The theory accounts for the Knotek-Feibelman mechanism. The theory of

desorption and atom-surface scattering with a partially lled band will be
discussed in Chapter 13.6.
10.3 Hierarchy of Greens Functions

We can use the EOM to generate approximate non-perturbative solutions of
open-shell interacting models. Consider for instance the time-dependent Anderson model obtained by adding an on-site interaction to the time-dependent
Fano model of Equation (10.11):
10.4 Composite Operator Method
H = H0 + H1 (t) + W =
k nk +
Vk,k (t)ak ak +
k,k
213
Us ns, ns, . (10.40)
denote the one-electron wave function labeled by k at site s, such

Let k (s)
that ak = s as k (s). Since

ia k, = [ak, , H] = k ak, +
Vk,k (t)ak , +
Us k (s)as, ns, (10.41)
k
the EOM read:

(T )
gk,k
(T )
= (t t )k,k + k gk,k (t, t )

(T )
+
Vk,p (t)gp,k (t, t ) +
Us k (s)D,s,k (t, t )
i
(10.42)
where the two-body propagator

D,s,k (t, t ) = T as, (t)ns, (t)ak , (t ) .
(10.43)
appears. The interaction makes the problem hard and the equations do not
close any more. If one can be content with a mean-eld approximation, then
one can try
?
(10.44)
D,s,k (t, t ) T as, (t)ak , (t ) ns, (t)
?
which closes the equations again. Otherwise, one can generate an equation of
motion for D,s,k (t, t ) in order to truncate the hierarchy of Greens functions
at an higher level. When we cannot achieve the exact solution, one should
compare various approximations.

For nite systems, one can in principle nd the exact solution by the EOM
method. As a trivial example, the one-site Hubbard model
H1site = E(n+ + n ) + U n+ n
(10.45)
yields the closed set of equations:

i
da+
= Ea+ + U a+ n ,
dt
da+ n
= (E + U )a+ n .
dt
Without interactions the particles are those annihilated by a , while in the

presence of interactions one has particles (1) = a+ , (2) = a+ n and
(3) = a , (4) = a n+ ; the zoological garden contains no more beasts.
214
This idea of the hierarchy can be generalized (see Ref.([49],[50][51]) and references therein). These authors call a composite operator, and write
a+
J(1)
(1)
a+ n (2)
J(2)
d
=
(10.46)
a = (3) , J = i dt = J(3)
J(4)
a n +
(4)
adopting a spinor notation (these are not spinors, they are just lists.) Anyhow, since the equation (10.46) is closed in this case, we may write

J(, t) =
(, ) (, t),
(10.47)
where for the sake of generality I imply a possible time dependence of the
operators. For H1site , the 44 matrix is block diagonal, since the rst two
entries are not connected to the others; each block is

E U
.
=
0 E+U
Following Ref.([49],[50][51]), we introduce
S, (t, t ) = (t t )0 |[ (t) , (t )]+ |0 ,
(10.48)
which is just i times the retarded Greens function (4.13), possibly averaged
over the Grand-Canonical Ensemble; we are using [, ]+ for Fermi particles,
but it would be [, ] for Bosons. The EOM read
i
S(t, t )
|t =t = [J(t), (t)]+ .
t
(10.49)
We dene the m matrix by

m=i
S(t, t )
|t =t ,
t
(10.50)
use Eq.(10.47) and the so called normalization matrix I = {I, } with elements
I, (t) = S, (t, t).
(10.51)
The nonzero diagonal blocks of I are

[a , a ]+
[a , a n ]+
1 n
=
for = +, .
[a n , a ]+ [a n , a n ]+ ,
n ns
Finally the EOM become
m = I.
(10.52)
215
Let us extend the treatment to many sites, denoted by latin indices; we

start with a Nop -component spinor (p, t) for each site involving cp and the
operators of nearby sites. We write Greens functions
S, (p, t, p, t ) = (t t )[ (p, t) , (q, t )]+ ,
(10.53)
and the initial conditions

I, (p, q, t) = S, (p, t, q, t);
(10.54)
the normalization matrix has site indices as well. From now on spinor indices
can be understood; we write the m matrix
m(p, q) = [J(p, t), (q, t)]+ .
(10.55)
Formally, the argument proceeds as above: understanding the site indices as

well the EOM are still given by Equation (10.52); the eigenvalues of could
be interpreted as the eigen-energies of the system. The trouble is that when
one calculates J = i d
dt , new operators arise; Equation (10.47) does not hold,
but is replaced by
J(p, t) =
Nop
(p, q) (q, t) + J(p),
(10.56)
with J is a nonlinear rest. In other terms, the set of operators is not closed,
and the spinors have more components than we can aord.
To keep the calculation manageable, one could trivially truncate the
spinors at Nop operators, throwing away the unwanted ones and J; the
matrix gives a solvable approximation. This is less accurate than the standard EOM method, which prescribes to replace the extra operators by some
approximation like (10.44).
However, there is a more clever alternative criterion for sorting out the
terms linear in in the r.h.s. of Equation (10.56). From the Nop -component
one can work out the truncated I matrix. The map (p), (q) Ip,q
sends two operators into a number like a scalar product in operator space. In
this Section I denote this scalar product by , and write
(p), (q) = I(p, q) = [ (p, t), (q, t)]+ .
Taking the scalar product of (10.56)
J(p), (s) =
Nop
(p, q) (q, t), (s) + J(p), (s) .
(10.57)
Now in order to drop the last term we do not need to assume that J is
small; we make the milder assumption that it is orthogonal to the spinors.
Dropping J we nally obtain (10.52), that is,
216
= mI 1
(10.58)
and achieve a well dened, solvable approximation. In this way, the new
depends on those complicated operators that we are not considering explicitly. Although this is not a unique way to proceed, it is certainly appealing.
Fourier transforming to momentum space, the retarded Greens function is
approximated by
Nop

i (k)
,
(10.59)
g (r) (, k) =
Ei (k) + i
i
where Ei (k) are the eigenvalues of (k) and the spectral functions i (k) can
be derived[199] from I(k).
For a review on the applications to the Hubbard Model, see [198].
Problems
10.1. Verify (10.10) explicitly in the non-interacting limit.
10.3. Use the tight-binding method to calculate the retarded Greens functions of the tight-binding Hamiltonians in d=1,2,3 dimensions.
10.4. Dealing with X-Ray absorption and emission in Metals, Langreth[205]
considered the model Hamiltonian

H=
q aq aq + E0 b b +
Vqq aq aq bb
(10.60)
q
qq
where b and aq annihilate core and conduction electrons, respectively; the

Greens functions g(t t ) = T b(t)b (t ) describes the deep state dynamics
and Fkk (, ; t, t ) = T ak ( )ak ( )b(t)b (t ) the absorption and emission
processes. For absorption, the average is a ground-state expectation value in
the presence of the core electron. Derive the EOM for Fkk in the absorption
case.
11 Feynman Diagrams for Condensed Matter

Physics
11.1 Diagrams for the Vacuum Propagator

Consider a many-body system, like a moleculeor solid, with Hamiltonian
H = H0 + V , where V is the interaction, H0 = j j nj is the kinetic energy
with ground state | and eigenvalue W0 . I shall write

ck ck |0
(11.1)
| =
k <EF
to mean that the product is over occupied spin-orbitals, and produces a Fermi
sphere in the thermodynamic limit. How can we nd the ground state energy
E0 of H? Standard perturbation theory fails unless V is small compared to
the unperturbed energy dierence; in practice, it fails almost always, since in
most interesting problems, the spectrum H0 is continuous.
Continuous spectra can be perturbed in such way that discrete spectra
arise. Suppose one wants to nd the bound states of the Hydrogen atom by
treating the Coulomb interaction as a perturbation: for any Z an innity of
bound levels exists, but there is no hope to get any sensible result with any
nite number of terms. The very existence of bound states requires poles of
1
= 1 + z + z 2 + . . . can only be
Greens functions to form. The pole of 1z
found at innite order. The formula (4.125) for E0 from the vacuum amplitude
(4.123)
(11.2)
R (t) = | UI (t) | = eiW0 t | eiHt |
is useful because we know R at least formally, recalling (2.36)
t
i
dt VI (t )
UI (t) = eiH0 t eiHt = T e h 0

t1
i t
i 2 t
=1+
dt1 VI (t1 ) + ( )
dt1
dt2 VI (t2 ) + ,
h 0
0
0
(11.3)
Note that in this section we use the telescope form (2.6) of the nested
integrals. To illustrate the method, I shall use the simplied interaction

V =
U (i, j, k, l)ci cj cl ck .
(11.4)
ijkl
218
11 Feynman Diagrams for Condensed Matter Physics
In this way, we assume that only opposite-spin electrons interact, and this is
the simplication1 . Since |VI (t)| = |V | , the rst-order contribution
is
R
= i t | V | = i t
(1)
t
(t) = i
dt1 | V |
0
U (i, j, i, j)|ni nj |
(11.5)
ij
where n is the occupation number. In (11.5) we must average a product of

i
k
t1
j
l
Fig. 11.1. The rst-order contribution, with i=k, j=l. One says that i is contracted
with k and j with l. A more explicit denition follows shortly.
occupation numbers, and for each spin-orbital the creation operator is represented by a line leaving a vertex and the annihilation operator by a line
entering the vertex. Graphically, if we represent the up-spin operators as
the ends of the left line, and the interaction to the dashed line, this corresponds to the pattern of Fig. (15.4). The occupation number product over
the non-interacting | can be written as a vacuum average and using the
anticommutation rules we can separate the two spins: thus it factors:
|ni nj | = |ni | |nj | = ni nj
(11.6)
where the average occupation numbers have been assumed spin-independent

for the sake of simplicity. However we did not achieve much, yet: substitution
in the energy formula (4.125) yields 0. In second-order,
R(2) (t) =
ijkl i j k l
t
t
U (i, j, k, l)U (i j k l ) dt1 0 1 dt2
0
| T [ci (t1 ) cj (t1 ) cl (t1 ) ck (t1 ) ci (t2 ) cj (t2 ) cl (t2 ) ck (t2 )] | .
(11.7)
More generally, we need to calculate the ground state average
M = | T [A1 (t1 )A2 (t2 )A3 (t3 ) . . .]|
(11.8)
where A1 , A2 , A3 . . . are interaction-picture operators - either creation or annihilation operators - dened on an orthonormal basis in the interaction picture and | is the Fermi sphere. This calculation can be worked out by
1
Later we can easily restore the full interaction by adding exchange terms.
219
writing down all the operators that make up the Fermi sphere according to
Equation (11.1); then this becomes an average over the vacuum |0 ; then,
by the anticommutation relations, we can bring the annihilation operators
to act on |0 to produce 0; all what remains from the anticommutators is
the result. However, we naturally ask if we can do something to reduce all
this groundwork. Fortunately, we can, in terms of contractions. A contraction of two operators A and B which are either creation or annihilation
interaction-picture operators, is dened as
A (t) B (t ) = | T [A (t) B (t )] | .
(11.9)
For this to be non zero one of the two operators must create and the other
annihilate the same one-body state, and the contractions yield (up to a sign)
the non-interacting propagators ( Equation (4.38))
igk (t) = T [ck, (t)ck, (0)] = eik t {(t)[1 nk ] (t)nk },
(11.10)
which propagates a hole for t < 0 and an electron for t > 0. For equal time
contractions one denes:
c (t) c (t) c (t)c(t) .
(11.11)
11.1.1 Wicks Theorem

Wicks theorem, basic for all sorts of perturbation expansions, reads:
M = T [A1 (t1 )A2 (t2 )A3 (t3 ) . . .] =
P
P () AP 1 AP 2
AP 3 AP 4 . . . ;
(11.12)
in words, M is the sum of the products ()P AP 1 AP 2 AP 3 AP 4 . . . of the
contractions; P is the permutation that takes from the initial expression
A1 A2 A3 . . . to the nal one, and ()P its signature. I propose an intuitive
proof2 .
As a warm-up, we calculate the ground state expectation value of the
product in the second-order term (11.7); for an easy start, we do so without
the time ordering T, and for the moment we let
M = | ci (t1 ) cj (t1 ) cl (t1 ) ck (t1 ) ci (t2 ) cj (t2 ) cl (t2 ) ck (t2 ) | .
Also, we assume provisionally that the canonical basis is the one of H0 eigenstates; then the time dependences are given by c-number phase factors as in
Equation (4.35), ck (t) = ck eik t . The operators come in pairs at each time
and since Fermi operators anticommute, one can lump all the spins on the
left without changing sign: we obtain
M = | ci (t1 ) ck (t1 ) ci (t2 ) ck (t2 ) cj (t1 ) cl (t1 )cj (t2 ) cl (t2 ) | .
2
the (much longer) standard one by induction may be found in textbooks[2].
220
Now the and are dierent species (mathematically, linearly independent

operators); since actually | = | | the matrix element M factors,
M = | ci (t1 ) ck (t1 ) ci (t2 ) ck (t2 ) |
| cj (t1 ) cl (t1 )cj (t2 ) cl (t2 ) | .
(11.13)
The assumption that the canonical basis is the one of H0 eigenstates can be
removed, since anyhow H0 = H0 + H0 , and since H0 and H0 commute,
we can write c (t) = eiH0 t c eiH0 t and the time-dependent creation and
annihilation operators simply anticommute. While using Bloch waves to dene the Fermi sphere, we may wish to work with contractions in a site basis.
Next, we restore the T operation, and consider
M = | T [ci (t1 ) ck (t1 ) ci (t2 ) ck (t2 ) cj (t1 ) cl (t1 )cj (t2 ) cl (t2 )] |
= (t1 t2 )M12 + (t2 t1 )M21
(11.14)
M12 = | ci (t1 ) ck (t1 ) cj (t1 ) cl (t1 )ci (t2 ) ck (t2 ) cj (t2 ) cl (t2 ) | ,
M21 = | ci (t2 ) ck (t2 ) cj (t2 ) cl (t2 ) ci (t1 ) ck (t1 ) cj (t1 ) cl (t1 ) | .
Again, the operators come in pairs at each time, so in each term of the
sum ( 11.14) we can lump the spins on the left and factorize like above:
M12 = | ci (t1 ) ck (t1 ) ci (t2 ) ck (t2 ) |
|cj (t1 ) cl (t1 )cj (t2 ) cl (t2 ) |
M21 = | ci (t2 ) ck (t2 ) ci (t1 ) ck (t1 ) |
|cj (t2 ) cl (t2 ) cj (t1 ) cl (t1 ) |
and the whole time-ordered matrix element ( 11.14) breaks down into the
time ordered factors:
M = | T [ci (t1 ) ck (t1 ) ci (t2 ) ck (t2 )] |
| T [cj (t1 ) cl (t1 )cj (t2 ) cl (t2 )] | .
(11.15)
Thus, the time-ordering does not prevent the factorization, and the argument
works independently of the number of creation and annihilation operators.
Now we can breath a little bit, but can we do any better? We can actually
use the dierent spin-orbitals of any canonical basis as the dierent species,
but now we must keep track of the signs. Consider the matrix element for
spins. Particle i created at time t1 must be annihilated either by ck (t1 ) or by
221
ck (t2 ); so, two contractions contribute, but in each case the matrix element
breaks as before3 . Therefore the ground state average can be obtained as a
sum over permutations. However, there is more. In the above discussion, we
used no specic property of the ground state | ; we can average the matrix
element on any state in the same way (the contractions are then redened
accordingly). We can analytically continue Equation (11.12) from the real
axis to the verical track of Figure 2.2.2 b) (Sect.2.2.2). So, Wicks theorem
holds at nite temperatures as well; then stands for a thermal average.
Comments
This is a very general theorem that is often understated in books. It is obvious
that Wicks theorem also holds for averages over the vacuum |0 . By the way,
is itself a vacuum; for states above the Fermi level we can create electrons
like on |0 ; for occupied states the annihilation of an electron is the creation
of a hole. The transformation to hole operators c = b is canonical, or if you
like it is just a change in notation. So Wicks theorem works in both cases
for the same reason.
Wicks theorem for bosons works in the same way, and for the same reasons, except that the Bose operators commute under T and there is no sign
nuisance in this case.
11.1.2 Goldstone Diagrams
We represent the second-order terms diagrammatically by drawing two interaction lines, labeled t1 and t2 with t2 < t1 according to the scheme of
Fig. (11.1.2). These are time-ordered or Goldstone diagrams. Two diagrams
must be identied if they are topologically equivalent in the Goldstone sense,
that is, if they can be deformed into one another without changing the time
ordering. In this example with no interaction for parallel spins the vertices
on the left refer to spin up and there are no exchange terms. If we contract
only equal-time operators we get the a) diagram of Figure 11.1.2. All the nonpropagating lines (those that begin and end at the same time) must refer to
occupied orbitals (equal-time Wicks rule).
Diagrams consisting of two separated pieces, like a), are called unlinked.
Diagrams b),c) and d) are linked and b)and c) have the same value. The
contribution of diagram a) is obtained setting k = i, l = j, k = i , l = j in
R(2) (t) :
3
The possibility of contracting 4 or more operators with the same indices can
also be included. For averages on either |0 or , c c are either 0 or 1, so the Wick
factorization is granted since 0*0=0 and 1*1=1. So, these terms do not modify
the result. As emphasized in the Landau series[3], in the thermodynamic limit the
theorem holds for averages on any state, since taking the same index for two c c
factors means selecting a set of null measure.
222
j
i
t1
t2
k
l
time
Fig. 11.2. Starting the setup of the second-order contribution.
R(a) (t) =
U (i, j, i, j)U (i j i j )
ij i j
+
| T [c+
(t
)
i+ 1 cj (t1 ) cj
t
0
dt1
t1
0
dt2
+
(t1 ) ci+ (t1 ) c+
i + (t2 ) cj (t2 ) cj (t2 ) ci + (t2 )] | .
(11.16)
and contracting one nds
2

t
R(a) (t) =
U (i, j, i, j)ni nj
U (i j i j )ni nj .
2

ij
(11.17)
ij
Here again we note that a line labeled i closing on itself, or tadpole, simply
contributes ni .
Diagram b) arises from the identications k = i, l = j, k = i , l = j and
to the contractions
+
+
+
c+
i+ (t1 ) ck+ (t1 ).ci + (t2 ) ck + (t2 ).cj (t1 ) cl (t2 ).cl (t1 ) cj (t2 ). (11.18)
The equal-time contractions
and
ci+ (t1 ) ck+ (t1 ) = (i, k) ni
(11.19)

c+
i (t2 ) ck + (t2 ) = (i , k ) ni
(11.20)
are immediate;let us consider the lines that represent propagators (see (4.37)).
The descending line yields
cj (t1 ) cl (t2 ) = (j, l ) eij (t1 t2 ) [nj (t1 t2 ) (1 nj ) (t2 t1 )]
= (j, l ) (i)gj (t2 t1 )
(11.21)
The - sign in front of (i)gj comes from the order of operators which is
opposite to the convention of ( 4.36 ).
223
The ascending line gives:

cl (t1 ) cj (t2 ) = (j , l) eil (t2 t1 ) [(1 nl ) (t1 t2 ) nl (t2 t1 )]
= (j , l) (i)gl (t1 t2 )
(11.22)
In both cases, the time argument of g is the nal time minus the initial time
of the directed line. Since t2 < t1 , the contribution of the descending l = j
line represents a hole in a occupied state, while the for the ascending line is
an electron in an empty one. In both cases the directed line brings a factor
ig.
a)
i=k
b)
i=k
c)
d)
j=l
j=l
t1
j=l
j=l
i=k
j=l
i=k
j=l
t2
i=k
i=k
j=l
i=k
j=l
i=k
Fig. 11.3. Second-order contributions to R(t).
Thus, diagram b) ( like the identical one c)) yields the following expression:
R(b) =

U (i, j, k, l)U (i , j , k , l )ik i k l j j l ni ni
ijkl i j k l
i gj (t2 t1 )igl (t1 t2 ).
dt1
t1
0
dt2
(11.23)
What is coming from these observations is a convention for representing

mathematical expressions graphically; each fermion line labeled j and directed from t to t yields igj (t t); each interaction line brings a factor
(i)U (i, j, k, l) (where the arguments refer to the top interaction in Figure
11.1.2 and each closed fermion loop brings a - sign; all internal indices are
summed over; the nested time integrals are then evaluated. To evaluate the
ground state energy, one takes the t (1 i) limit and uses the energy
formula (4.125). Working out (11.23) one nds
R(b) =

ijli
U (i, j, i, l)U (i , l, i , j)ni ni

0
dt1
t1
dt2
ei(l j )(t1 t2 ) ()nj (1 nl ).
(11.24)
224
The integral is elementary:

t1
t
1 + it(j l ) eit(j l )
dt1
dt2 ei(l j )(t1 t2 ) =
.
(j l )2
0
0
(11.25)
The term linear in t contributes to the ground state energy. Note however
that j is a hole, l is an electron,hence j l < 0 and the exponential in the
r.h.s. drops out as t (1 i). Problem 11.1 deals with diagram d).
Despite the heuristic appeal of cartoons showing the basic scattering
events between particles, the diagrammatic expansion does not look promising at all at this stage. The number of dierent diagrams and their complexity grows catastrophically with the order, and we need to go to order
to get useful results. Worse still, diagrams may defy intuition severely.
Their heuristic appeal can become very misleading, since the processes they
describe include counter-intuitive ones. Diagrams obtained from d) by appending tadpoles or bubbles freely to each line do belong to the expansion,
and must be included, even when the lines at a given time over-number the
electrons in the sample and/or violate the Pauli principle. No such principle
holds in diagrammatics, and several lines can bear the same quantum numbers. In one-particle problems, diagrams arise that appear to describe several
fermions propagating at a time.
11.1.3 Diagram Rules for the Thermodynamic Potential
Using imaginary time, the above results extend directly to the calculation of
the thermodynamic potential
where
= KB T lnZG
(11.26)
ZG = T re(HN )
(11.27)
is the grand partition function, the chemical potential and N the particle number operator. may be found as the sum of ring diagrams (see
=
Sect. 12.4.5) .The mean energy and particle number are related by E
+ N + T S, and since the entropy S goes to 0 at T=0 the ground-state

energy may be found from the zero-temperature limit. Luttinger and Ward
[55] derived and stated the rules for in frequency space as follows. 1) Draw
all possible n-th order diagrams. Put labels on each line (a label like r stands
for one-body labels including spin, and the non-interacting Hamiltonian is
supposed diagonal with eigenvalues r ) and associate a Matsubara frequency
l = + (2l+1)i
, with integer l. Conserve frequency at each vertex. 2) For
each diagram: insert a factor rs|v|r s for each interaction (labeled by onen+1
body states, r and s entering and r,s leaving), a factor ()
2n n! and a (-) sign
1
for each loop 3) For each line labelled by r write a factor Gr (l ) = l
. This
r
11.2 Linked Cluster Theorem
225
factor is called
propagator. Next, sum over all internal labels and all frequen+
cies l with 1 l , including a convergence factor el 0 for loops. Examples
may be found in Ref.[55]. The T=0 version of the expansion
isobtained by

removing , replacing l by real frequencies and 1 l by d. Please
note that 1) simple diagrams may be easily computed without the diagram
rules directly from the T exp formula, 2) nobody ever computes very complex
diagrams, which are too many and too complicated to be of any use. The real
point is: how to avoid any heavy-duty use of the diagram rules.
= 2
Fig. 11.4. Combining diagrams. Those summed in the rst row consist of dierent
subdiagrams, while those of the second row are made up with like subdiagrams.
The only cheap and ecient way to produce and compute lots of highorder diagrams is by combining smaller ones. In a linked diagram one can
go from any vertex to any other one by following propagator or interaction
lines. The diagram a) of Figure 11.3 is an example of an unlinked diagram
made up of two simpler linked ones. Consider doing the same with any pair of
dierent connected diagrams. They give raise to dierent combined graphs for
R as shown in Figure 11.4. The diagrams on the l.h.s., top row, dier by the
order of interactions in time and are to be counted as dierent contributions
because the ordering is enforced by dierent functions; their sum is just the
product of the original diagrams. If however the same diagram is considered
twice, (bottom row and diagram 11.3 a) ) the contribution to R comes only
i } the set of linked
once so the result must be divided by 2. Denoting by {D
diagrams, consider a particular unlinked diagram of order n containing, say,
, D
, D
. Each of its n interaction lines labeled by times
3 linked parts D
tn < tn1 < < t1 belongs to one of the 3 sub-diagrams. Within each subdiagram, the ordering of the interaction lines is xed by the fact that it is a
226
i (t), however by varying the relative ordering of two

replica of one of the D
interaction lines belonging to dierent sub-diagrams we get a diagram which
belongs to the series for R. Summing over all such interchanges we simply
get rid of the relative orderings; if the three sub-diagrams are dierent the
result is
(t)D
(t);
(t)D
D
however, if there are identical subdiagrams we must pay attention. If, say,
(t) = D
(t), then each contribution to R is counted twice (each has a
D
ghost with the two sub-diagrams exchanged, but the series has one of them)
and the result is
1
(t).
D (t)2 D
2!
These observations extend to any order and to any number of linked sub i . The sum of all the unlinked diagrams of any order containing
diagrams D
i } is
p linked parts chosen in any way in {D

D
ni (t)
i
;
(11.28)
(p,
ni )
Rp (t) =
n
i!
i
i
{ni }
here {ni } = n1 , n2 , is a list of non-negative integers specifying the num i , while the total number of parts is xed by the
ber ni of occurrences of D
Kroneker . This is just

1
ni (t)
D
(n1 , n2 , , nk , )!
(11.29)
i
p!
i
{np }

(
ni )!
where (n1 , n2 , , nk , )! = n1 !n2 !... is the multinomial coecient ; thus
Rp (t) =
Rp (t) ==
1
1
Di (t))p = (L(t))p
(
p! i
p!
(11.30)

Di (t) is the sum of all linked diagrams.
where L =
Thus we have obtained the simple but far reaching Linked Cluster Theorem:
R(t) = eL(t)
(11.31)
Every approximation to L(t) takes us to innite order in the series for R(t),
and a way to achieve real progress is open.
This exact result is the basis of the so called cumulant expansion4 . Now
our task is calculating L and the unlinked second-order diagram 11.3 a) must
be discarded, while the rst-order generates a partial series:
4
Note that a Linked Cluster Theorem exists[42] even if the interactions are
mediated by bosons; each sub-diagram must then be divided by its order. The
combinatorial argument is similar, but is modied by the fact that each line gets
two labels instead of one.
ln R(1) (t) = it
227
U (i, j, i, j)ni nj ,
ij
and dierentiating with respect to t and multiplying by i we nd

U (i, j, i, j)ni nj
E (1) =
i,j
which is the Hartree-Fock approximation, a marked improvement over the

previous null result.
11.2.1 Valence Electron and Core Hole
To see the Linked Cluster Theorem and the diagram rules at work, consider
the simplest molecular-orbital model with sites a and b with hopping matrix
element V

0 V
(11.32)
H0 =
V 0
with eigenvalues V ; now suppose that a core electron sits at site a; the
valence Hamiltonian becomes

U V
H=
;
(11.33)
V 0
where U is the valence electron-core electron interaction. The new eigenvalues
are
U
,
(11.34)
=
2
with = U 2 + 4V 2 . The core occupation number is nc 1 and the core

Greens function enters only through nc . The unperturbed (U=0) valence
propagator on site a reads
(t)eiV t + (t)eiV t
,
(11.35)
2
where the rst term is the contribution of the lled ground state orbital, and
the second comes from the excited state. The unperturbed lling is n0a = 1/2.
The rst-order diagram is itU na . There is one second-order diagram (Fig.
11.3 b) or c)).The descending line brings a - sign, another one is due to the
ring; one obtains
igaa (t) =
(iU )2
4

0
dt1
t1
dt2 e
2iV (t2 t1 )
U
= ( )2
V
1 + e2iV t + 2iV t
16

. (11.36)
228
In third order we may add a third tadpole on the electron or on the

hole lines; the contributions cancel (since the descending line brings a - sign)
and there is no third-order term. In fourth-order one could consider adding
two tadpoles to the second-order bubble, but there is a similar cancellation
between those attached to electron and hole lines. The contribution comes
from the diagram in Figure 11.5 and from a similar diagram with t3 < t4 .
t1
t2
t3
t4
Fig. 11.5. The fourth-order contributions to R(t). Note that two valence electron
lines propagate between t3 and t2 , although in this problem there is just 1 valence
electron. This illustrates that particle-number-violating diagrams must be included.
We get
t1
t2
t3

(iU )4 t
dt1
dt2
dt3
dt4 e2iV (t4 +t3 t1 t2 )
8
0
0
0
0

1 + 2e2iV t
U 4 5 e4iV t 4e2iV t
iV t
=( )
.
(11.37)
V
512
128
( )
Since the problem is simply solved exactly, we can check these results
by calculating R(t) and its expansion. One easily derives the Hamiltonian
on the basis of the U = 0 eigenstates and the ground-state-diagonal
matrix H
element its exponential:
=
H
U
2
V
U
2
U
2
U
2
+V
R(t) = e
iV t
2
{cos(
2iV
t
t
)+
sin( )}. (11.38)
2
where we choose V > 0 by setting

R(t) = eiV t eiHt
11
Expanding C(t) = log(R(t)) one nds

1 + e2iV t + 2iV t
U
i
C(t) = iV t + iV t U t + ( )2
2
V
16

4iV t
2iV t
2iV t
5e
1 + 2e
4e
U
U
iV t
+( )4
+ O( )6 . (11.39)
V
512
128
V
229
Using (4.125) we compute the time derivative and let the exponential die out
as t (1 i); we obtain, in agreement with the exact solution,
U2
U4
U
+
+ O(U 6 )
2
8V
128V 3
W0 = V +
(11.40)
11.2.2 H2 Model
As a further example, we shall use the H2 model of Sect. 1.2.5.
igaa (t) =
1
{(t)eith t + (t)eith t }
2
(11.41)
From (4.37) one gets

iga,a = igb,b =
6
1 5 ith t
e
(t) eith t (t) ,
2
6
1 5 ith t
e
(t) + eith t (t) .
2
The net second-order contribution comes from only 11.3 d). It is
iga,b = igb,a =
4 (iu)2

0
dt1
t1
dt2 [ga,a (t2 t1 )ga,a (t1 t2 )]2

it
1 e4ith t
2
=u
16th
64t2h
(11.42)
where a factor 4 comes from the fact that each interaction line can be labeled a
or b indierently. The diagrams 11.3 b) and c) vanish because the terms with
the like interaction lines are canceled by those on interactions on dierent
atoms. In third-order there is no contribution because shifting any vertex
from an electron to a hole line changes sign to the diagram. For the same
reason, there is considerable cancelation in fourth order, the only surviving
terms being the one of Figure (11.2.2) and the one which results from the
exchange of t3 and t4 . Direct calculation shows that

it
itu
1 e4ith t
+ u2
2
16th
64t2h

e4ith t
itth
e4ith t
itth
.
1024
2048
512
C(t) = 2ith t
+
u4
t4h

+
5 e8ith t
8192
(11.43)
230

t1 , s1
t2 , s2
t3 , s3
t4 , s4
Fig. 11.6. One of the fourth-order diagrams that contribute to R(t) for the Hubbard H2 model. The interaction lines are labeled with time and site (si = a or b)
indices.
t
+
=
t
+
t
Fig. 11.7. Contributions to the core propagator
11.2.3 The Linked Cluster Expansion and Greens Functions

The Lundqvist model[38]
H = c c +

q
q aq aq + cc
gq (aq + aq )
(11.44)
has been used to discuss plasmon eects in core photoelectron spectra. The
removal of the core electron shifts the plasmon coordinates and the density
of states relevant to the spectrum is obtained from the correlation function

d it
e + |c (0)c(t)|+ ,
(11.45)
N () =
2
where + is the plasmon vacuum with the core electron present. N () is
obtained from the core propagator
g(t) = i+ |T c(t)c (0)|+ .
(11.46)
By expanding the propagator as above we generate diagrams where the boson

propagator (4.64) iDk (t) = exp[ik | t | ] appears; each electron-plasmon
vertex contributes a factor igq ; plasmon lines ( iDq (t) factors ) may be
emitted and adsorbed any number of times according to any pattern.
11.3 Diagrams for the Dressed Propagator
231
Langreth[39] noted that g (0) (t t ) is a simple phase factor for t < t while
g (tt ) 0 for t > t ; thus the ascending time line must be labeled t at the
bottom
and t at the top. From each diagram one can collect the same phase
factor: g (0) (ti tj ) = g (0) (t t ). Hence, g(t t ) = g (0) (t t )(t t ).
Erasing the deep hole propagator from the diagrams for g we nd those for ,
a succession of interaction lines ordered in a particular way; one can classify
these diagrams as linked and unlinked. Repeating the above argument one
nds that Linked Cluster Theorem applies,
(0)
g(t t ) = g (0) (t t )eC(tt ) ,
(11.47)
where C is the only unlinked diagram, namely the second-order one: a boson
is emitted at time t2 > t and later re-adsorbed at t1 < t . Hence,
t
t1

2
(igq )
dt1
dt2 Dq (t1 t2 ).
(11.48)
C(t t ) =
t
and one achieves the exact solution5

C(t ) =
(igq )2
t
dt1

q
t1
dt2 eiq (t1 t2 )

gq2
1 iq t eiq t
.
q2
(11.49)
In addition,the Linked Cluster Theorem has been invoked as an ansatz

which is useful for special problems[40], [41]. The idea is simple. Start with
the ansatz (11.47) and expand both sides in powers of the interaction using
the cumulant expansion

a2
a3
exp[
an xn ] = 1 + a1 x + ( 1 + a2 )x2 + ( 1 + a1 a2 + a3 )x3 + . . .
2
3!
1
From a few diagrams for g that one can compute directly we can obtain
an equal number of an coecients that perform a particular innite resummation of the series.

The time-ordered propagator (4.20) averaged over the interacting ground
state or over the grand-canonical ensemble lends itself to a perturbation expansion. In this section we consider in detail the zero temperature case, using
5
This agrees with Equations (6.30 6.31) taking into account that there the
Fourier transform involved ei(+H)t and here we are using ei(+H)t instead.
232
a slightly lighter, discrete notation and the fact that for time-independent
problems the dependence is on a single time, namely, t = t1 t2 . In the
non-interacting problem with Hamiltonian H0 , the propagator is

g0 (a, b, t) = i | T ca (t) cb (0) | ,

(11.50)
where | is the ground state and cb creates an electron in a single-particle
state |b . We know that g 0 is diagonal on the basis of the H0 eigenstates and
is Greens function of the Schr
odinger equation; moreover,
g0 (k, k ,) =
kk
kk g0 (k,),
k + ik
(11.51)
with k = +0 for empty states, k = 0 for lled ones.

With H = H0 + V , where V is the interaction, the propagator is

(11.52)
ig(a, b, t) = 0 | T ca (t)cb (0) |0
averaged over the unknown interacting ground state, and the operators
are in the Heisenberg Picture. We use the standard denition (2.12) for the
Heisenberg operators but switch to the interaction picture, using (2.39)
AH = UI (t, t0 )AI (t)UI (t, t0 ),
in order to obtain g from an innite order series expansion of the evolution
operator in powers of V :
11.3.1 Adiabatic Switching and Perturbation Theory
To obtain g we do not really need 0 , as the denition seems to suggest. We
can resort to the trick of the adiabatic switching of V. Here we assume6 a
time-independent H and write
UI (t1 , t2 ) = UI (t1 , 0) UI (t2 , 0) = eiH0 t1 eiH(t1 t2 ) eiH0 t2
(11.53)
Assume that at time t = t0 in the remote past the system was in the unperturbed ground state , with energy eigenvalue W0 ; then at time 0 the
interaction picture state is a wave packet containing 0 :

|n n | eiEn t0 | (11.54)
UI (0, t0 ) | = eiHt0 eiH0 t0 | = eiWn t0
n
For t0 we can distil from it the interacting ground state by shifting

the path in the complex t plane with a small tilt setting t t(1i), = 0+ .
In this way, among the exponentials
6
The general case is presented in Chapter 13.
eiEn t0 (1i) = eiEn t0 eEn |t0 | ,
233
(11.55)
the one of the ground state dominates, and one is left with
UI (0, t0 ) | = |0 0 | eiE0 t0 | eiW0 t0 = |0 0 | UI (0, t0 ) | ; (11.56)
formally, (provided that the denominators do not vanish)
Im(t)
t0
Re(t)
t0
Fig. 11.8. The complex t plane with the tilted Gell-Mann and Low path.
|0 =
UI (0, t0 ) |
,
0 | UI (0, t0 ) |
0 | =
| UI (t0 , 0)
.
| UI (t0 , 0) |0
(11.57)
This is the Gell-Mann and Low Theorem [83]7 . Any expectation value
0 |A|0 on the interacting ground state can be obtained from non-interacting
ground state averages. One obtains:
| UI (t0 , 0) AUI (0, t0 ) |
;
| UI (t0 , 0) |0 0 | UI (0, t0 ) |

but along the tilted path, |0 0 | is equivalent to n |n n |; so
0 | A |0 =
(11.58)
| UI (t0 , 0) AUI (0, t0 ) |

| UI (t0 , 0) AUI (0, t0 ) |
=
.
| UI (t0 , 0) UI (0, t0 ) |
| UI (t0 , t0 ) |
(11.59)
We take advantage to set the propagator in this form. Consider g for t > 0,
ig(a, b, t) = 0 | ca (t)cb (0) |0 with Heisenberg operators. In order to use the

T exp expansion we must go over to the interaction picture8 with caH (t) =
UI (t, 0)ca (t)UI (t, 0), hence
0 | A |0 =
The original proof reported by Ref. [2] is based on the perturbation series, so
it depends on its validity; also, it shows that numerators and denominators bear a
phase factor that diverges in the adiabatic limit.
8
Recall that with our convention the Schr
odinger, Heisenberg and interaction wave functions coincide at t = 0 and that at any other time the
change of representation is obtained by A (t) = I (0)| AH (t) |I (0) =
I (0)| UI (0, t) AI (t) UI (t, 0) |I (0) .
234
ig(a, b, t) = 0 |UI (t, 0)ca (t)UI (t, 0)cb (0)|0

=
|UI (t0 , 0)UI (t, 0)ca (t)UI (t, 0)cb (0)UI (0, t0 )|
, t > 0. (11.60)
|UI (t0 , t0 )|
Using the Group property of U , the unitarity property UI (t, 0) = UI (0, t), and
the fact that under T the fermion operators anticommute, we shall manoeuvre
to obtain a single U in the numerator as well:
ig(a, b, t) =
|UI (t0 , 0)UI (0, t)ca (t)UI (t, 0)cb (0)UI (0, t0 )|
, t > 0.
|UI (t0 , t0 )|
(11.61)
In other terms,
ig(a, b, t) =
|UI (t0 , t)ca (t)UI (t, 0)cb (0)UI (0, t0 )|

, t > 0.
|UI (t0 , t0 )|
More generally we can write
(11.62)
, setting S = UI (, ),
ig(a, b, t1 , t2 ) =
a (t1 )c (t2 )]|

|T [Sc
b
|S|
(11.63)
and expand in powers of V using the T exp formula (2.36). At each order
one obtains a sum of partial amplitudes, involving V and the bare propagator g 0 ; these expressions are best handled when represented as diagrams.
The key point is that dierent partial amplitudes give topologically inequivalent10 diagrams; at order n there is a nite set of possible topologies and all
correspond to partial amplitudes. In all diagrams, an oriented g 0 line enters
at a point b, goes through some interaction vertices and then reaches the exit
point a. Between two interaction vertices, the propagator line is labeled by
a spin-orbital: the one entering from outside will correspond to spin-orbital
b and the outgoing one to a. Dotted interaction lines start at each vertex;
at order n the diagram contains n interaction lines. A properly oriented and
labeled propagator line must pass by every vertex; each graph presents a path
which takes from b to a, and circuits attached to it (and possibly to other
circuits) by interaction lines.
The denominator of Equation (11.63) is like the vacuum amplitude in
(11.2) and yields all diagrams not connected to the main line; the only dierence is that in (11.2) the interactions are between times 0 and t and here
they can take place at any time. In the time representation (11.63), the
mathematical expression or amplitude of a disconnected diagram consisting of two parts is the product of the two amplitudes. Therefore, we may
9
We shall nd the result (11.63) more in general in Equation (13.56).

Two diagrams are topologically equivalent if they can be deformed with continuity to the same shape; in complex graphs this criterion requires a talent for the
ne arts. I shall give a more practical prescription below.
10
235
a (t1 )c (t2 )]| = |T [Sc

a (t1 )c (t2 )]| L |S|
write |T [Sc

where the L
b
b
sux means that only linked diagrams must be kept (no parts must be disconnected from the b a line). The denominator cancels with the unlinked
diagrams and we are left with
a (t1 )c (t2 )]| L .
ig(a, b, t1 , t2 ) = |T [Sc
b
(11.64)
11.3.2 Diagrams for the Propagator in frequency space

The diagram rules will be derived in the Chapter 13, and can be veried
by working out low-order terms from the T exp formula (2.36). I present the
Feynman method in frequency space, that is, the expansion of the interacting
(0)
time-ordered gab () in terms of gab () and interaction vertices; the interaction is a two-body operator (typically, the Coulomb potential). We must label
a)
a,
b)
q,
b,
k,
q,
a
k,
q,
k
k,
k,
c)
q,
+
b
d)
Fig. 11.9. Diagrams for g: a) and b) exhaust the rst order, c) and d) are secondorder contributions.
the g 0 lines and the interaction lines by frequencies such that a and b have
frequency and at all vertices the sum of ingoing frequencies equals the sum
of outgoing ones. When momentum or crystal momentum are also good quantum numbers (translationally invariant or periodic systems) four-momentum
is conserved at each vertex. must pay attention to avoid double counting. To
see if two diagrams with an apparent correspondence between the vertices
are indeed topologically equivalent, one can start considering a path on one
diagram and the corresponding path on the other. If one meets the corresponding points in the same order in both diagrams, and this remains true
236
for all possible paths, the diagrams are topologically equivalent. For example,
the two diagrams in Figure 11.3.2 left are equivalent; I have added letters to
show that all vertices can be labeled in such a way that all paths correspond.
The orientation of arrows does not point to the evolution in time and each
propagator corresponds to electrons and holes, so the fact that one arrow
goes back in the second diagram is not meaningful.
Fig. 11.10. Left panel: two diagrams that look dierent but are the same. Right
panel: two ways to represent the interaction iVkpmn .
Topologically equivalent diagrams are the same diagram and we

Next, I list the rules to translate a diagram into a formula in the frequency
representation. Write a factor
ig0 (k,) =
i
,
k +ik
(11.65)
for each electronic line with k = +0 for empty states and k = 0 for lled
ones (0 stands for a positive innitesimal, as usual). For every interaction line
labeled like in the left diagram a) include a factor

iVkpmn = i
d r
(r) p (r ) m (r) n (r )

= iVpknm . (11.66)
d r k
|r r |
3
As we see in the right panel of Figure 11.3.2, what matters is the identity
of the ingoing and outgoing lines, and seemingly dierent pictures yield the
same expression. Multiply by (1) for each closed electron ring, sum over
intermediate
spin-orbitals and integrate over intermediate frequencies , , . . .
d

with d
2
2 . . .. This rule must be better specied when there are nonpropagating lines, like the tadpole. The contribution of the line (excluding
q,
(0)
nq
Fig. 11.11. A tadpole is a line that starts and ends at the same vertex.
the vertex) is:

d
ig0 (q,) = i
g0 (q, t = 0).
2
q
q
237
(11.67)

ig0 (q, t) = 0| T cq (t) cq (0) |0 ,
Since
(11.68)
is discontinuous, we introduce the equal-times rule

T cq ( )cq ( ) = cq (+ )cq ( )
(11.69)
where + is just after and just before. The transform must be taken for
t = 0 (the non-propagating lines simply have no time to propagate). Since
ig0 (q, t = 0 ) = n(0)
q ,
(0)
(0)
where nq is the unperturbed occupation number, the result is k nk .
(0)
Including a () for the circuit, the tadpole yields + q nq , which is
ready for inserting the q dependence of the interaction vertex. Let us see the
rst-order diagram (Figure (11.3.2), a)) and its value D[a)]: one obtains

0
0
0
D[a)] = ig (a,)ig (b,) (i)
(11.70)
Va k b k nk .
k
If k stands for a one-electron wave-vector which is conserved in the absence

of interactions, the matrix element Va,k,b,k brings a (a, b) factor (the tadpole cannot exchange energy and momentum). In rst-order there is also the
exchange diagram (Figure (11.3.2), b)) which is obtained by exchanging the
outgoing interaction lines. The (-) sign must not be inserted; the arc is a
non-propagating line; one obtains
d
ig 0 (k, ) (i) Vk a b k .
D[b)] = ig (a,)ig (b,)
2
0
(11.71)
(11.3.2), c)) shows a second-order diagram with a couple of tadpoles inserted into a propagating line. In this case,

0
0
(11.72)
D[c)] = D[a)]ig (k,) (i)
Va k b k nk ,
k
where the new term in parentheses may be identied with the added tadpole
and the new ig 0 represents the new propagator piece.
So, we can calculate a subset of a diagram for later use, knowing that
it will appear in many other diagrams and bring some type of process into
play. As another example of the diagram rules, we calculate the pair bubble
238
that appears as an inset in the second-order diagram (11.3.2), d)). This is

a prototype polarization part, that is, a graph that can be inserted into an
interaction line; it will be useful later, and the big bubble unlike the tadpole
does exchange energy and momentum.
i0 ()= (1) 2
d
ig0 (m,)ig0 (n, + ),
2
m,n
(11.73)
where the (-1) factor is due to the closed circuit and the 2 factor to the
sum over the spins of the circuit. We start evaluating the bubble with the
integral

d 0
g (m,)g0 (n, + ) =
2
1
d
1
2 m +im + n +in
(11.74)
by the residue method. When the poles are on the same side of the real axis
(m n > 0,) one closes the path on the other side and gets 0. Otherwise we
integrate in the upper half plane, and there are two cases: a) n occupied and
m empty b) m occupied and n empty. These represent respectively hole and
electron propagation. One nds:
ifn [1fm ]

n +m i0 case a
d
1
1
=
(11.75)
2 m +im + n +in
ifm [1fn ]
case
b
n +m +i0
So, we end up with
i0 () = 2

m,n

fm (1 fn )
fn (1 fm )
+i
.
+ m n i0
+ m n + i0
(11.76)
It is already evident at this stage that as the number of interaction lines

increases and progressively more complex diagrams arise the labor involved
tends to increase in a prohibitive way, except for cases like the one shown
in (11.72), when the complex diagrams arise as combinations of simple ones.
Now, combining simple things to build structures is something that we may
do by using ingenuity, while the most general graph at order n requires computing power. If the crucial physics is hidden in some monster-diagram that
rst appears at some high order, we have little hope to understand. However,
no such cases are known, and ingenuity has been rewarding.
11.4 Dyson Equation

Like the Linked Cluster method, Dysons leads to a summation to innite
order of selected classes of diagrams. A look to Figure 11.12 , where g with
11.4 Dyson Equation
239
Fig. 11.12. Some of the lowest-order diagrams for g; there are two kinds in rstorder, but the variety grows with order in an impressive way.
the thick line, shows that as we proceed the number and complexity of the
diagrams grows in an inordinate way. However, although normally we cannot
obtain g exactly, we can use topology to make exact statements about g. All
terms except the rst have a factor ig (0) (a, )ig (0) (b, ), that is, an incoming
and an outgoing line. The stu in between is a self-energy part. The latter
does not have external lines, but is usually drawn with short ones to show
where they belong in the full diagram. Therefore the mathematical expression
for the self-energy part is just the one for ig (0) divided by ig (0) (a, )ig (0) (b, ).
Having computed a simple self-energy part, like the tadpole or the bubble,
one can conceive iterating it indenitely. The diagrammatic series is summed
Fig. 11.13. Left: Dysons equation. Right: iterative solution.
by writing the Dyson equation (Figure 11.13). The simplest approximation

to includes in g the repeated scattering of the simplest type any number
of times; any approximation is a partial re-summation of the series to innite
order. We can also sum two dierent self-energy parts, e.g. tadpole + bubble,
and the iteration will give all sorts of diagrams in which such insertions occur
in any sequence; all such diagrams will occur once. However, if we iterate
a diagram showing a bubble followed by a tadpole we miss all the diagrams
where two consecutive bubbles (or tadpoles) occur. To avoid missing diagrams
in this silly way it is important to use irreducible self-energy . A self-
240
energy part is called reducible if it can be split into two diagrams by cutting
a single propagator. The left diagram in Figure(11.4) is reducible, the right
one is not.
Fig. 11.14. Reducible (left) self-energy part and irreducible (right) self-energy
parts.
The Irreducible (or proper) Self-Energy is the sum of all the irreducible self-energy parts. The analytic expression of is i(), and the
function is also called self-energy. The rst-order self-energy part reads:

(11.77)
(i) (1) = (i)
Va k b k n0k
Vk a b k n0k .
k
The Dyson equation yields the exact diagrammatic expansion of the greens
function if the exact is known. Indeed, each diagram containing whatever
insertions in any number of times in any order comes out of by the
iteration of Figure 11.13 exactly once. Figure 11.13 reads:
ig = ig0 +ig0 (i)ig g = g0 +g0 g; g = g0 +gg0 ;

with x = (r, t) and dx = d3 rdt, these are
g(x, x ) = g0 (x, x ) +
g(x, x ) = g0 (x, x ) +

(11.78)
dx1 dx2 g0 (x, x1 )(x1 , x2 )g(x2 , x )

dx1 dx2 g(x, x1 )(x1 , x2 )g0 (x2 , x ).
(11.79)
In analogy with (10.7) we can rewrite Dysons equation in dierential

form. This analogy will be exploited in Section 11.8. Since the noninteracting
g (0) obeys

d
H0 (1) g (0) (1; 1 ) = (11 )
(11.80)
i
h
dt1
/
0
by applying i dtd1 H0 (1) we obtain
[i H0 (x)]g(x, x )
t
dx1 (x, x1 )g(x1 , x ) = h(x x ).
(11.81)
The solution of Dysons equation is simple when the symmetry is so high

that g is diagonal on the basis of H0 eigenstates. Jellium is a theorists toy
11.4 Dyson Equation
241
metal; an innite electron system with a uniform neutralizing background of

positive charge. Both g0 and g are diagonal on a plane-wave basis, and the
scalar Dysons equation yields
g(k,) =
1
;
0 (k) (k,)
(11.82)
thus can be thought of as a complex correction to the energy eigenvalue.

The Dyson equation leads to the notion of a quasi-particle; as far as g is concerned, the many-body interactions can be summarized in a self-energy correction to the one-body dispersion, leading to a useful picture of an eective,
modied or dressed electron moving around. One can produce many-body
states that behave like a quasi-electron or a quasi-hole added to the system,
although the imaginary part of eventually damps the single-particle character and redistributes the energy. In this way some of the simplicity of the
independent-electron picture is retained; moreover, the independent-electron
model is partly validated; thus one understands the fact that band-structure
calculations and the Sommerfeld theory are useful in many cases although
electrons are far from being independent.
11.4.1 External Potential
a
=
a
+p
a
q +
a
+
+
b
Fig. 11.15. Propagator in external potential (big dot).
For non-interacting electrons in an external potential the diagram rules

are similar. The series is shown in Figure 11.15 where the heavy oriented
line is ig(a, b, ), the light one represents
ig0 (a, b,) =
iab
,
a + ia
and the big dot between two lines , one ending with a label p and the second
starting with label q stands for for iVpq . In the Fano model one has only
the V0k matrix elements; to calculate the local Greens function g00 () one
can use two ways (see Figure 11.16).
(0)
(0)
() + g00
() 00 () g00 ()
g00 () = g00
242
i)
ii)
Fig. 11.16. Two ways to derive for the Fano model: i) the Dyson equation for
g00 ii) the Dyson equation for the g matrix.
One can write the Dyson equation using the self-energy i) for g00 (),
(0)
which is ready for inserting g00 lines; since
i() =
(iV0k )igk ()(iVk0 ) = i

k
|V0k |
k + ik
(11.83)
we nd back the self-energy of Chapter 5, but with the important inclusion

of the Fermi level, where Im changes sign. The matrix Dyson equation
g = g +g g
could be solved by matrix inversion, but it is simpler to write

g00 = g000 +g000 V0k gk0 (),
k
gk0 = g0kk Vk0 g00 =
Vk0
k +ik g00
In the alternative method ii) we are seeking a matrix self-energy, where one
(0)
should insert not g00 lines, but any two lines.which is readily solved.
Another interesting example occurs in the theory of resistivity of metals;
the external potential is due to impurities. At second and higher order, the
self-energy is complex. One has the problem of calculating the Greens function and then averaging over a random impurity distribution; however, in the
dilute case, repeated scattering against the same impurity dominates. The
inverse quasi-particle lifetime 1 is given by Im; actually this can be taken
as 1 in Drude theory and is the dominant contribution to the resistivity at
low temperature, while at higher T phonon scattering becomes important.
11.5 Self-Energy from Interactions

First-Order
Suppose we know the spin-orbitals {a, b, } that diagonalize the free-particle
Hamiltonian H0 and the Greens function g (0) ; we wish the Greens function
11.5 Self-Energy from Interactions
243
g (1) including the eects of the direct rst-order Coulomb self-energy, i.e. the
tadpole (see Fig. 11.5a)). The self-energy is a matrix with elements (from

(0)
Equation 11.77) ab () = k Vakbk fk , that is,
+
b)
a)
Fig. 11.17. a) Direct Hartree self-energy b) Hartree-Fock self-energy.
ab () =
(0))
fk
d3 rd3 r
a(r) k(r ) b(r)k(r )

.
|r r |
(11.84)
(0)
ab () = Wab is just the ab matrix element of an eective potential

W (0) (r) =

dr
(0)
fk |k(r )|2
;
|r r |
(11.85)
but this is nothing else than the electrostatic potential produced by the tadpole charge density. Had we considered non-interacting electrons in an effective external potential W (0) (r), we should have obtained the same ab .
Note that ab includes the eects of W (0) (r) exactly, as if we had found the
eigenstates {a(1) , b(1) , } of H0 + W (0) (r) and computed the new Greens
function g (1) , taking the matrix elements in the old basis {a, b, } We can
still improve the approximation using just the tadpole, but this time with
(1)
g (1) . The new correction ab can then be interpreted as if the electrons did
not interact, but moved in an additional potential W (1) (r). By iterating the
argument, one reaches the self-consistency, as shown in Fig. 11.5b) where
the exchange term has also been included, and the internal propagators are
dressed, renormalized, fully interacting propagators, shown as the thick lines.
This corresponds to the Hartree-Fock approximation. Hence, if one starts
with the Hartree-Fock basis as {a, b, } the series for starts with the
second order.
Second-Order
The proper (2) has two diagrams ( direct and an exchange).
The direct one is shown in Figure 11.18 a). It is convenient to introduce
the pair bubble (11.114), and to write:
244
b
b)
a)
Fig. 11.18. Second-order self-energy: a) direct b) exchange

i() =
d 0
ig (k,)(iVknbm )(iVamkn )(i0 ())
2
(11.86)
The calculation is an exercise in contour integration similar to the calculation of the bubble, and the contribution of Equation (11.76 ) to (11.86)
is

1
i
d
.
fn [1 fm ]Vknbm Vamkn
i
2 k + ik n + m i0
k,m,n
Again, we get zero unless the poles = n m + i0 and = k + ik are

on opposite subplanes, which requires k < 0, that is, k must be occupied;
the pole in the upper half plane comes from the second factor and yields the

n (1fm )Vknbm Vamkn
. Adding the second contribution
contribution i klm fk f+
m n k im
with m lled and k, n empty,

ab () =
[(1 fm ) fk fn + fm (1 fk ) (1 fn )]Vknbm Vamkn
m,k,n
n +m k im
(11.87)
To get the exchange term from Figure 11.18 a) one cuts the lines raising from the bottom vertices and exchanges their upper ends, labeling in
such way that the exchange is done in the upper interaction resulting in
Vamkn Vamnk . The value of the diagram is obtained from (11.86) by
Vamkn Vamnk , (no closed circuit any more). Thus, the total second-order
self-energy is:
(2)
ab () =
[(1 fm ) fk fn + fm (1 fk ) (1 fn )]
m,k,n
Vknbm [Vamkn Vamnk ]

.
n +m k im
(11.88)
11.7 Two-body Greens Function: the Bethe-Salpeter Equation
245
Unlike the rst-order result, this correction is beyond Hartree-Fock, complex and -dependent. Starting from Hartree-Fock, we now obtain corrections to Koopman ionization potentials and electron anities and lifetimes
without resorting to the often prohibitive Conguration-Interaction computations. This is very useful in atomic and molecular calculations [4]. Typically
the corrections are of the order of 1 eV.
11.6 Skeleton Diagrams

We can do better for free (no other diagrams to compute) by considering the
self-energy of Figure 11.19. where the heavy lines are dressed propagators.
This is self-consistent perturbation theory, in analogy with the diagrammatic
version of the Hartree-Fock approximation (by the way, no tadpoles and no
open-oysters appear, since we assume that the self-consistent potential is our
one-body potential). This procedure is equivalent to a summation to innite
Fig. 11.19. By expanding the right-hand side, one nds an innite series of selfenergy corrections.
order of a series of more and more complex self-energy diagrams, and can
simplify enormously the task of summing the most relevant parts of the series
for .
Skeleton diagrams are those with no self-energy insertions. Since all the
internal propagators are dressed, to avoid double counting only skeleton diagrams are allowed. A few skeleton diagrams can replace an innity of selfenergy ones, if the self-consistency can be carried out by numerical iteration.
This can be a very good solution, depending on the choice of skeletons.
11.7 Two-body Greens Function: the Bethe-Salpeter

Equation
One can perform a Dyson type analysis on the series for the two-particle
Greens function G2 (10.3). Working in the time representation, we sort out
the direct and exchange contributions to G2 consisting of diagrams with only
self-energy insertions. In obvious shorthand notation,
246
G2 (1, 2, 3, 4) = ig(3, 1)g(4, 2) ig(3, 2)g(4, 1)

+
d5d6d7d8ig(3, 5)ig(7, 1)ig(4, 6)ig(8, 2)(5, 6, 7, 8),
(11.89)
where is the scattering amplitude. The corresponding diagrammatic equation is shown in Figure 11.20. is the sum of all two-body diagrams such that
each ingoing line starts with an interaction, and each outgoing line leaves an
interaction. Internal lines can have all sorts of self-energy insertions.
1
3
G2
=
4
+
2
+
2
Fig. 11.20. The two-body Greens function G2 and the scattering amplitude .
The dressed incoming and outgoing lines belong to G2
Fig. 11.21. is the scattering amplitude; the heavy lines represent dressed propagators, the rst two diagrams shown are irreducible, the other two are not; J the
irreducible scattering amplitude.
In one can separate out the irreducible interaction diagrams, that cannot
be split in two by cutting only two dressed lines; let J denote their sum. All
terms in can be obtained by iteration from those of J, and this enables us
to write down a Dyson-like equation for .
Finally, the whole series is obtained from the Bethe-Salpeter equation

G2 (1234) = g(31)g(42)g(32)g(41)+ d5d6d7d8J(5678)ig(35)ig(46)G2(7812)
(11.90)
which is shown in Figure 11.23.
11.8 Self-Energy and Two-Body Greens Function
247
Fig. 11.22. is obtained by iterating J.

1
3
G2
1
+
7
G2
8
2
4
2
4
4
6
2
Fig. 11.23. The Bethe-Salpeter equation for the two-body Greens function.
2
=
1
1
Fig. 11.24. The relation of the self-energy to the scattering amplitude. Replacing
by its diagrammatic expansion one gets the expansion for . This diers from
Ref. [116], Chapter 10.6, since we assume that the Hartree-Fock approximation is
embodied in our bare propagator.
11.8 Self-Energy and Two-Body Greens Function

Let us rewrite the equation of motion (10.7),

H0 (1) g(1; 1 ) = (11 ) i d2v(1, 2)G2 (1; 2|2+ ; 1 )

i
t1
(11.91)
where H0 is the non-interacting Hamiltonian. We compare with the Dyson

equation (11.81)

[i
H0 (1)]g(1, 1 ) d2(1, 2)g(1, 1) = h(1, 1 ).
(11.92)
t1
is a non-local potential that can have some local contribution (proportional
to (r r )), so there is some freedom in the denition of . One can decide
to include the Hartree potential VH (x) due to the charge ig(x, x+ ) as a local
part of or as a potential in H0 (x). We shall write the one-body term in
both the above equations as H0 (x) + VH (x), where the second term actually
comes from the local part of . Thus,
248
[i
H0 (1) VH (1)]g(1, 1 ) = (1, 1 ) +

t1
d2(1, 2)g(2, 1).
(11.93)
We nd

h d2v(1, 2)G2 (1, 2|2+ , 1 ).
d2 [VH (1)(1, 2) + (1, 2)] g(2, 1 ) = i
(11.94)
We have obtained g, not just , since G2 comprises the incoming and
outgoing legs; a formal relation between self-energy and two-body function is
obtained by amputating the outgoing one, that is,

(1; 1 ) + VH (1)(1, 1 ) = i
h d2d3v(1, 2)G2 (1, 2|2+ , 3)g 1 (3; 1 ) (11.95)
where g 1 is the inverse of g in the matrix sense.
It is clear that the approximations for the two-body function and for the
self-energy cannot be chosen independently. The relation to the scattering
amplitude is given by the gure 11.23.
11.9 Functional Calculus and Diagrams

Given a functional F which depends on a function (x), one denes the
functional derivative11
d
F [(x)]
=
F [(x) + (x y)].
(11.96)
(y)
d

For instance, if F [] = d3 xf ((x), (x)), then it turns out that
f
f
F
=
.
(x)
If we think of the integrals in discrete form, and the integrand does nor depend on derivatives of , the functional derivative is just a partial derivative
with respect to the value of in a particular space-time point. Functional
dierentiation is a powerful tool that we shall use in parallel with the diagrammatics. Functional derivatives with respect to the propagator g are
particularly easy because they undo the integrations which are prescribed by
the rules (see Figure 11.25. ) This remark will be useful later.
11
We shall often use a 4-dimensional notation x (x, t) and write

F []
d
=
F [(x, t) + (x x3 )(t t3 )].
(x3 , t3 )
d
249
4
4
4
2
5
+
1
5
2
= (1, 2)
(1,2)
g(4,5)
Fig. 11.25. Diagrammatic interpretation of the functional derivative.
11.9.1 The Self-Energy as a Functional

Functional derivatives also naturally arise for another reason. To study an
interacting many-body system, one can introduce an external probe potential (x, t) and look for a one-body response. The response arises from the
dependence of the propagator g on small changes in the perturbation

H1 = dx(x)(x, t)
(11.97)
where (x) = (x) (x) is the density operator. The most direct way to
study the dependence of g on is through the functional derivative g(1,2)
(3) .
Rewriting (11.63) in the form
ig(1, 2) =
(2)]|
|T [S(1)
,
|S|
(11.98)
since |S|
= lim,0 UI (, t3 + )UI (t3 + , t3 )UI (t3 , ), and
t +
UI (t3 + , t3 ) = 1 hi t33 H1 ( )d + one nds
S = lim UI (, t3 + )
,0
(x3 )

t3 +
i
dx
dt(x)(x x3 )(t t3 )
t3
i
UI (t3 , ) =
S(3),
h
(11.99)

i |T S(3)(1) (2)|

|T S(3)|
g(1, 2) =
ig(1, 2)
.
(3)
h
|S|
|S|
(11.100)
250
We may conclude that

i
h
g(1, 2)
= iG2 (1, 3|3+ , 2) + ig(1, 2)g(3, 3+).
(3)
(11.101)
This yields a new, useful link between g and G2 , but we are interested in
involving . We multiply by v(1, 3) and integrate over 3 (that is, over dx3 dt3 ,
of course). Since ig(3, 3+) = (3) is the density, we may introduce the
Hartree potential VH (1) = d3v(1, 3)n(3) and write

g(1, 2)
= i d3v(1, 3)G2 (1, 3|3+ , 2) VH (1)g(1, 2).
i
h d3v(1, 3)
(3)
(11.102)
Now replace 2 by 1 and 3 by 2. The result may be used again in (11.91)

H0 (1) g(1; 1 ) = (11 ) i d2v(1, 2)G2 (1; 2|2+ ; 1 )
i
t1
with added to H0 , and yields
H0 (1)Vef f (1))g(1, 1 ) = (1, 1 )+i

h
(i
t1
g(1, 1)
; (11.103)
(2)

= (1) + d3v(1, 3)(3)
d2v(1, 2)
here the screened potential Vef f (1) = (1) + VH

appears.
Comparison with (11.93),

[i
H0 (1) VH (1)]g(1, 1 ) = (1, 1 ) + d2(1, 2)g(2, 1).
t1
again with VH changed to Vef f , yields

g(1, 2)
(1, 3)g(3, 2)d3 = i
h v(1, 3)
d3
(3)
(11.104)
where the instantaneous Coulomb potential v xes t3 = t1 . We can solve this

for , introducing the (matrix) inverse g 1 of g, such that

1
(11.105)
d3g(1, 3) g(3, 2) = d3g(1, 3)g 1(3, 2) = (1, 2).
Dierentiating (11.105) we obtain

g 1 (1, 3)
g(3, 2)
g(3, 2)d3 = g 1 (1, 3)
d3
(2)
(2)
and substituting in (11.104) we get

g 1(4, 5)
g(1, 2)
= d4d5g(1, 4)
g(5, 2);
(3)
(3)
(11.106)
251
now the rhs has a comfortable g factor on the right. Hence, post-multiplying
by g 1 we obtain the desired result

g 1(4, 6)
(1, 6) = i
h d(3, 4)v(1, 3)g(1, 4)
.
(11.107)
(3)
11.9.2 Polarization Bubble
The dielectric function
x ) is dened by the relation (using a notation

(x,
3
with x (r, t), dx d rdt,)

D(x, t) = dx (x, x , t t )E(x , t )
between displacement vector and electric eld which holds in linear media.
The external charges are the sources of D; hence, the eective potential
Vef f (r) due to an external source having bare potential is

(11.108)
Vef f (r) = dr dt 1 (r, t, r , t )(r , t ),
the inversion implied in the notation 1 is matrix inversion in the r, t, r
indices. Hence,adopting a lighter notation,
1 (1, 2) =
Vef f (1)
.
(2)
In the microscopic theory,

Vef f (1) = (1) + d3v(1, 3)(3), (3) = ig(3, 3+)
(11.109)
(11.110)
where v(1, 2) is the Coulomb interaction v(x) = 1r (t). When is produced

by a point charge at r1 (t), Vef f (2) is the screened interaction W (1, 2);

Vef f (2)
W (1, 2) = d3
v(1, 3).
(11.111)
(3)
By a Dyson equation, one can determine the screened interaction W in
terms of the bare (Coulomb) one v. In Figure 11.26 the light dotted line
stands for iV and the bold dotted line stands for iW . Therefore, W is
=
Fig. 11.26. The Dyson equation for the screened interaction W (heavy dashed
line) in terms of the Coulomb interaction (light dashed line)and the full polarization
propogator .
252
determined by the polarization according to Figure 11.26:

W(1, 2) = v(1, 2) v(1, 3)(3, 4)W(4, 2)d(3, 4).
(11.112)
Here, is the sum of all the irreducible polarization parts that cannot be
split by cutting a single V line. Since the series for (1, 2) is symmetric
under a mirror reection that exchanges 1 and 2, (2, 1) = (1, 2), and also
W (2, 1) = W (2, 1); moreover the alternative form exists

W (1, 2) = v(1, 2) d(74)W (1, 7)(7, 4)v(4, 2).
(11.113)
The general structure of the diagrams for (Figure 11.27 ) shows that it can
be obtained from g and a vertex :

(1, 2) = i
h
g(2, 3)g(4, 2+) (341)d (34)
(11.114)
3
Fig. 11.27. the structure of , according to Equation (11.114), showing the points
where interaction lines can be inserted.
Any approximation for yields the corresponding approximation to ,

and we shall give an exact expression for this in the next Section. A similar
analysis can be done on the diagrams for .
3
=
2
1
Fig. 11.28. The relation of the self-energy to the vertex function, after Equation
(11.115).
A formal expression for is:

(1, 2) = i
h
g(1, 4) (4, 2; 3)W (1+, 3)d(34)
253
(11.115)
in terms of the dressed interaction W and of the vertex . The fact that
the same function indeed appears in the expressions for and will be
apparent shortly.
11.9.3 The Vertex
We evaluate (11.111) and obtain a new form of Dysons equation (11.113)
for W , and a new formula to comparewith (11.114). Taking the functional
derivative of (11.110) Vef f (1) = (1) + d3v(1, 3)(3), one nds

Vef f (2)
g(4, 4+ )
= (2, 3) i
h d4
v(2, 4).
(3)
(3)
+
)
Since (11.114) says that we want two g factors, we evaluate g(4,4
by the
(3)
trick (11.106),

g 1 (5, 6)
g(4, 4+ )
= d(56)g(4, 5)
g(6, 4+ ),
(3)
(3)
and then substitute into (11.111): this gives the screened interaction

W (1, 2) = v(1, 2) + i
h v(1, 3)g(4, 5)
g 1 (5, 6)
g(6, 4+ )v(2, 4)d(3456).
(3)
(11.116)
The integral on the r.h.s. is a functional that begins with a bare interaction
v(1, 3) and ends with another bare v(2, 4). If we aim at (11.113) we must
convert the rst to a screened interaction: we need to screen it by and
we can if we screen as well. The correct way to do this is by changing
the independent variable from the external to the eective potential via the
functional chain rule

Vef f (7)
= d(7)
= d(7)1 (7, 3)
. (11.117)
(3)
(3) Vef f (7)
Vef f (7)
This produces

i
h
:
W (1, 2) = v(1, 2) + d(47)v(2, 4)

g 1 (5, 6)
d(56)g(4, 5)g(6, 4+)
d3v(1, 3)1 (7, 3) . (11.118)
Vef f (7)
;<
=
254
Here the square brackets contain W (1, 7) (see Equation 11.111); the result
agrees with Equation (11.113) if the under-braced quantity is (7, 4):

g 1 (5, 6)
g(6, 4+).
(7, 4) = d(56)g(4, 5)
(11.119)
Vef f (7)
This yields the following expression for the vertex of Equation (11.114):
(1, 2, 3) =
g 1 (1, 2)
.
Vef f (3)
(11.120)
Using the solution to Problem 11.4,

g 1 (1, 2) = (i
H0 Vef f (1))(1, 2) (1, 2),

t
(11.121)
we obtain the exact result

(1, 2, 3) = (1, 2)(1, 3) +
(1, 2)
.
Vef f (3)
(11.122)
We can re-derive Equation (11.115) for in terms of the dressed interaction

W and of the vertex using (11.117) and (11.120) in (11.107).
11.10 Hedins Equations

Hedins equations fully determine g, in principle at least. Dysons equation

(11.123)
g(1, 1 ) = g0 (1, 1 ) + d2d3g0 (1, 2)(2, 3)g(3, 1)
requires the knowledge of ; this is given by Equation (11.115), namely,

(1, 2) = i
h g(1, 4) (4, 2; 3)W (1+, 3)d(34),
(11.124)
in terms of the screened interaction W and the vertex . Equation (11.112)

W(1, 2) = v(1, 2) v(1, 3)(3, 4)W(4, 2)d(3, 4)
(11.125)
allows to calculate W if one knows the polarization , which by Equation
(11.114)

(1, 2) = i
h
g(2, 3)g(4, 2+) (341)d (34)
(11.126)
again requires the vertex. To close the equations we have to nd one for
the vertex. To lowest order, (123) (1, 2)(1, 3) (this is called the GW
11.10 Hedins Equations
255
approximation); Migdal proposed plausible arguments to show that the vertex

corrections can be neglected to a good approximation in Jellium and weakly
correlated solids. This Migdal theorem is often violated in strongly correlated
systems, but Hedin found a rigorous equation for . The starting point is
Equation (11.122)
(1, 2, 3) = (1, 2)(1, 3) +
(1, 2)
.
Vef f (3)
(11.127)
The functional derivative is done via the chain rule

(1, 2)
(1, 2) g(4, 5)
= d(4, 5)
.
Vef f (3)
g(4, 5) Vef f (3)
By dierentiating the identity g 1 g = 1 and left-multiplying by g one obtains
the analogue of (11.106), namely,

g(1, 2)
g 1 (4, 5)
= g(1, 4)
g(5, 2)d(4, 5).
(11.128)
Vef f (3)
Vef f (3)
Thus,
(1, 2)
=
Vef f (3)

d(45)
(1, 2)
g 1(6, 7)
g(4, 6)
g(7, 5).
g(4, 5)
Vef f (3)
1
(1,2)
Now inserting Equation (11.120) (1, 2, 3) = g
Vef f (3) we obtain the least
obvious of Hedins equations:

(123) = (1, 2)(1, 3) +
(1, 2)
g(46)g(75) (673)d(4567). (11.129)
g(4, 5)
Hedin[53] obtained these exact equations that formally determine selfenergy, polarization, vertex function and Greens function. Although they
were not solved exactly, they lie at the heart of powerful approximate methods for rst-principle calculations [54]. This equation lends itself to a diagrammatic interpretation (see Figure 11.10). Indeed, functional dierentiation may be understood as removing a propagator line from a self-energy
diagram, which then becomes a for-point function with 2 incoming and 2 outgoing lines. By the Hedin equation, all possible combinations and iterations
of such scattering diagrams give raise to the most general vertex.
It was pointed out recently[139] that Hedins equations are helpful in

counting Feynman diagrams.
256

4
4
5
6
5
7
1 2
1 2
Fig. 11.29. A skeleton Self-energy diagram contributing to (1, 2), the corresponding contribution to (1,2)
, obtained by deleting the g(4, 5) line (Figure 11.9), and
g(4,5)
the contribution to (2, 1, 3) arising from the Hedin equation (11.129).
Problems
11.1. Compute diagram d).
11.2. Express the improper self-energy in terms of .

11.3. Are there any skeleton diagrams in the next Figure?
a)
b)
c)
Fig. 11.30. Are there any skeleton diagrams?.
11.4. Evaluate g 1 in terms of .
12 Many-Body Eects and Further Theory
12.1 High Density Electron Gas

k
m
k-q
m+q
+
k
Fig. 12.1. Diagram for g with a self-energy insertion involving 0 in Jellium.
For Jellium, due to momentum conservation, we re-label the diagram

11.10 d) involving the bubble 0 as in Figure 12.1 and (11.73) becomes
3
d m
d
ig0 (m, ) ig0 (m + q, + )
i0 (q, ) = 2
(12.1)
3
2
(2)
and (11.76) becomes, restoring h,

0 (q, ) = 2
d3 m
(2)3

fm (1 fm+q )
fm+q (1 fm )
h + m m+q + i
h + m m+q i
(12.2)
with

fm fm+q
,
Re0 (q, ) = 2 m h +
m m+q

Im0 (q, ) = 2 m fm (1 fm+q )(
h + m m+q ).
(12.3)
So, the self-energy (11.86) becomes

i() =
d3 q
(2)3
d 0
ig (k q, )(iVq )2 (i)0 (q, )
2
(12.4)
258

2
and since Vq = 4e
q2 , this diverges at small q. The electron gains self-energy by
exciting the medium and then re-adsorbing the excitations, but the process
runs out of control for the long-wavelength ones. What is going wrong at long
distances? It is the Coulomb interaction V , which is causing the divergence by
its long range, but should actually be replaced by a shorter ranged screened
interaction W .
Equation (11.112) is solved by Fourier transformation thanks to the translational invariance and becomes:
W (q, ) =
Vq
,
(q, )
(12.5)
where the exact dielectric function is given by

(q, ) = 1 + Vq (q, )
(12.6)
in terms of the exact irreducible polarization part. The cheapest approximation prompts itself: it consists of using instead of the unknown the lowest polarization part; the resulting approximation is popular as the Random
Phase Approximation (RPA) based on
RP A (q, ) = 1 + Vq 0 (q, ).
(12.7)
As usual, simplicity brings extra benets; in this case, we can identify the
RPA as asymptotically exact in the case of high density (or perfect Fermi gas).
Dimensionally 0 = [E 1 ], that is, it is inverse energy, and any polarization
part is clearly
the same; to check this, recall that momentum integrals are

actually q summations and carry no dimension, interaction lines V and

frequency integrals d carry E and Greens functions g bring E 1 . Thus,
inserting a new V into a polarization
part to build
a more complicated one
brings a dimensionless factor q dVq g 2 q Vq g = [1]. However, if we
scale the density of the liquid, we are changing the Fermi wave vector kF ;
then for every interaction line Vq scaling like kF2 there is a q summation
and an energy denominator kF2 (each energy must be counted kF2 );
(kF3 )
dVq g 2 kF3 kF2 kF2 kF4 kF1 , and the factor scales like kF1 ,
thus,
q
that is, like the Wigner-Seits radius rs . Thus 0 is the dominant polarization
part at high density, and the RPA is good in that case. The second fraction
in Equation (12.2) may be transformed setting m + q = n, which gives
fm+q = fn , fm = fn+q ; now renaming with n m one nds
3
d m
0 (q, ) = 2
3 fm (1 fm+q )
(2)

1
1
.
(12.8)
h + m m+q + i
h + m+q m i
This was rst evaluated by Lindhard [52]. Writing q for
h
, a = q 2q , a+ = q + q2 ,
= 2E
F
q
kF
and setting:
mkF
[1
2 2 h2
1 + a
1 + a+
1
| [1 a2+ ] log |
|} ,
+ {[1 a2 ] log |
2q
1 a
1 a+
q2
q2
2
[1 a ], q > 2, 22 + q > > 2 2 q
mkF
[1 a2 ], q < 2, q2 + q > > q2 + q
Im(0 ) =
2
2
4q
h
2,
q < 2, q2 + q > > 0
0
otherwise.
259
Re(0 ) =
(12.9)
12.1.1 More Physical Insight about the RPA

It is fairly common that important results are rst obtained the hard way
by the advances of the general theory and after a while somebody nds that
they could have been arrived at by a much simpler, but smart, method.
This sort of re-derivation is important as it oers a fuller understanding of
the result. H. Ehrenreich and M. H. Cohen [43] have shown that one can
derive the Lindhard dielectric function in a purely one-electron formulation.
Accordingly, in this Subsection we let H be a one-body Hamiltonian and look
for the density response to a weak potential W ()eit .
The average of any operator A can be obtained from the density
matrix

as T rA. The density operator at x is n
(x) = (x xe ) = 1 q eiq(xe x)

with xe the electron position operator; hence n(x) = 1 q eiqx T r[eiqxe ]
and the Fourier transform is n(q, ) = 1 T r[eiqxe ]. Writing the trace on a
plane-wave basis,
2
n(q, ) =
k||k + q .
(12.10)
A factor of 2 is due to the spin trace. For W = 0 one is left with a free-particle
problem and the zeroth approximation (0) such that its action on the plane
waves is (0) |k = f (Ek )|k . We consider the equation of motion
i
h
= [H, ]
t
(12.11)
for the one-electron density matrix in the presence of the perturbation W and
expand the density matrix = (0) +(1) + in powers of W . Linearizing the
equation of motion for the linear response (1) and taking matrix elements
between plane-wave states one readily arrives at
k|(1) |k + q =
t
(Ek Ek+q )k|(1) |k + q + [f (Ek+q ) f (Ek )]k|W |k + q . (12.12)
ih
Using k|W |k+q = W (q, )eit and dropping eit in all terms we obtain:
260
k|(1) |k + q = W (q, )
f (Ek+q ) f (Ek )
Ek+q Ek + h
(12.13)
Inserting into (12.10),

n = W (q, )
2 f (Ek+q ) f (Ek )
.
Ek+q Ek + h
(12.14)
W (q, ) is the screened potential induced by introducing external charge

with number density n(ext) (q, ):
W (q, ) =

4e (ext)
n
(q, ) + n(q, )
2
q
(12.15)
where n(q, ) is the induced density. n(ext) (q, ) alone would produce a
potental Vq, and with a slight generalization of Equation (12.5) we write
4e (ext)
n
(q, ) = Vq, = (q, )W (q, );
q2
(12.16)
hence,
q2
((q, ) 1)W (q, )
(12.17)
4e2
and comparing with (12.12) we obtain the Lindhard dielectric function,
n(q, ) =
(q, ) = 1 +
8e2 f (k ) f (k+q )

.
q2
k+q k h i
(12.18)
In the static limit, this may be evaluated to read

(q, 0) = 1 +
2kF + q
2me2 kF
2me2 2
q2
) ln |
|.
+
(k
F
2 2
2 3
4
2k
h q
h q
F q
(12.19)
It can be shown that for q 0,

(q, 0) 1 +
KT2 F
q2
(12.20)
4kF
is the Thomas-Fermi wave vector. Using (12.20), one nds
where KT F = a
B
that the screened potential of a point charge becomes a Yukawa potential
1r exp(KT F r). Actually, the asymptotic behavior of Fourier transform
reects the singularities [92]; at long distance the behavior of the screening
charge is oscillatory (Friedel oscillations):
n(r)
cos(2kF r)
.
r3
(12.21)
261
This is due to a divergence in d(q,0)

at q = 2kF , the largest q that can be
dq
transferred to an electron in the Fermi sphere without increasing its energy.
Note that for > 0 we may write
2 Im((q, )) =
8e2
f (k ) [1 f (k+q )] (k+q k h) (12.22)
q2
k
which is positive in the part of the q plane where h = k+q k =

5
6
h
2
2
can be satised with k inside the Fermi sphere and k = q
2m 2k q + q
outside. For a given , the shortest q vector is obtained by setting k right at
the Fermi surface and q parallel to k, and the longest with q antiparallel to
k.
qmin
k
qmax
Fig. 12.2. The longest and shortest q vectors for electron-hole excitations with a
given excitation energy. One starts with a Fermi wave vector k and must reach the
h.
energy surface at EF +
Electron-hole excitations and 2 are conned to strip between the two

h
2
h
2
parabolas h
= 2m
(2kF q + q 2 ). and h
= 2m
(2kF q + q 2 ).
The term in f (k+q ) in the summation in Equation (12.18) can be rewritten
f (k )
k k+q
h i
by a change of variable k + q k ; then, we can restyle as
(q, ) = 1 +
16e2
k+q k
f (k )
.
q2
(k+q k )2 (
h + i)2
(12.23)
The Drude dielectric function

The classical treatment by Drude also gave raise to a dielectric function.
Starting from the electron equation of motion in a periodic eld,m

r = m
r
i(qrt)
eE 0 e
were is the electron mean free time, one neglects q and solves
e E0
with r(t) = Aeit , obtaining A = m
+i . Then, the current density is
j = enr =
ne2 E0
it
.
m 1i e
Thus,
262
J = E,
(12.24)
2
0
with 0 = nem . Assuming that everywith the conductivity () = 1i
thing depends only on z, Maxwells equations give:
2
4i
d2 E(z)
=
E 2 J(z).
dz 2
c2
c
Then (12.24) gives

d2 E(z)
2
4i()
= 2 1+
E(z).
dz 2
c
this is tantamount to say that the medium produces a refractive index nref
c
and c nref
with n2ref = = 1 + 4i()
. Thus we arrive at
=1
p2
( + i )
(12.25)

2
is the Plasma frequency. This is qualitatively correct for
where P = 4ne
m
simple metals, but looks very dierent from the Lindhard result.
Plasmons and the Lindhard dielectric function
Some resemblance of the Lindhard to the classical calculation is recovered by
a long-wavelength (q 0) expansion. The q 0 limit of Equation ( 12.23)is
obtained by inspection since the k q term in k+q k averages to 0 and
2

P
f (k ) = n2 , with the result that (q, ) 1 (+i)
2 . A more accurate
analysis readily gives:
(q, ) 1
hkF 2
P2
P2
3
(
)
q2 +
( + i)2
5 m
( + i)4
(12.26)
Plasmon modes are dened by (q, ) = 0, which is the condition for selfsustained oscillations (one can have W (q, ) nite with Vq, = 0. They are
collective modes of the electron liquid with

hkF 2 q 2
3
) 2 +
(12.27)
(q) = P 1 + (
10 m
P
With increasing q, eventually the plasmon branch enters the electron-hole
continuum and becomes unstable against converting into a pair. Actually they
are sharply dened as q 0 but are damped with increasing q; this Landau
damping is due to decay in multiple electron-hole pairs. Thus, the plasmons
dominate the high-frequency, long wavelength screening, while electron-hole
pairs in metals are slower and act as shorter distance (higher q) screening
12.2 Low Density Electron gas: Ladder Approximation
263
modes. Plasmon satellites are typically observed in electron spectroscopies of

metals in the 10 to 25 eV range (see Chapter 6.4). In semiconductors and
insulators plasmons are seen in a similar way, since the gap is often small
compared to the P value that can be deduced from the valence electron
concentration. The phenomenon of ultraviolet transparency is well known:
metals become transparent at frequencies above the plasma frequency because
the electrons cannot follow the eld.
The Jellium energy in the high density case can be found by summing
over the ring diagrams. Gell-Mann and Brueckner found the result

E = dx[t[n] + x [n] + c [n]]n,
(12.28)
where I have introduced the following energies per particle (small rS ):
2
3
(3 2 n) 3 2.21
kinetic t[n] = 10
rs2 Ry
exchange
0.916
Ry
rs
correlation 0.062 ln(rr ) 0.096Ry.
(12.29)
12.2 Low Density Electron gas: Ladder Approximation

In the low density case the quantum electron gas is far from perfect, that is,
the kinetic energy is dominated by the interaction energy. Yet things again
simplify, since many-electron collisions become unlikely and one approaches
a two-body problem. So,we start adding two particles to an empty lattice (or
holes to a lled one; these two problems can reect quite dierent physics,
but are related by a canonical particle-hole transformation). The solution for
G2 is very interesting and can be obtained in several ways.
In Sect.(6.3) I show how one can write the two-particle Greens function
for two electrons in the empty lattice for any interaction potential; the solution is simplest when the potential in on-site (Hubbard interaction). Let Rcm
be the center-of mass of the pair and g Q be the Fourier transform of the twoparticle Greens function with respect to Rcm , with g Q the non-interacting
limit. The two-particle Greens function is given by the Kanamori [67]result,
gQ =
gQ
,
1 U gQ
(12.30)
where, as in (6.74),
g Q (z) =

q
1
.
z (Q q) (q)
(12.31)
This formula is readily obtained as the sum of a series of ladder diagrams,

which in the empty lattice gives the exact solution, and is a useful approximation at low density. Each scattering introduces an interaction and a couple
264
of propagators; if the pair carries momentum Q and frequency , the propagators may be labeled Q q, and q, . According to the diagram
drules,
the interaction brings a factor iU and we must perform the q 2
sum
on internal labels. Doing the frequency integral by countour integration, we
can show that introducing an interaction and a couple of propagators brings
a factor U g Q (see Equation 6.69 with z = + i0).
+ ...
+ ...
Fig. 12.3. Ladder approximation for the scattering amplitude and for the selfenergy (the rst term is the Hartree contribution). Note that and are simply
related; there is a general link, as discussed in Section 12.5 below.
Then, the sum of the geometrical series gives the Kanamori result, that
we obtained in Sect. 6.3 by quite dierent means. Then the one-particle selfenergy at low density can be gained by Equation of Figure 12.3. This Low
Density Approximation (LDA) was rst proposed by V. Galitzkii [133].
12.3 Ladder Approximations in Electron Spectroscopies

12.3.1 XPS and Auger Spectra from Metals
The theory of the Auger CVV spectra of closed-band solids outlined in Section
(6.2) brings valuable information on the local electronic structure, but needs
to be extended to conduction bands. The presence of a Fermi level in the
band where the two holes are produced by the transition complicates the
problem in several ways. The most evident are: 1) the band can be polarized
by the primary hole, so a two-hole Greens function cannot tell the whole
story 2) the existence of electron-hole excitations oers an intra-band decay
channel for the two-hole resonances. The writer[91] proposed the following
model Hamiltonian:
(12.32)
H = H0 + H 1 + H 2 .
The independent-particle part is
H0 =
ij ci cj + c b b,
265
(12.33)
ij
with the rst term (in obvious notation) describes the band structure, b
creates the core electron;
H1 = U n0+ n0
(12.34)
introduces the local repulsion at site 0, and
H2 = Ucv (n0+ + n0 )b b
(12.35)
introduces the on-site core-electron valence-electron repulsion.

The theory based on assumption that the band is almost full, that is the
number of holes per quantum state in the band satises
nh = a0 ao
1;
(12.36)
this is satised e.g. for the Ni d bands. The nh

1 assumption makes it easier
to justify the 2-step model, treating the core-hole decay separately from its
creation (by the way, the 1 step approach was not yet available in 1979).
Due to the band polarizability, one needs a 3-body Greens function (the
third body is the core electron)
g(3) (t) = T b(t)c0+ (t)c0 (t)c0 (0)c0+ (0)b (0) .
(12.37)
1 assumption makes it possible to factor the 3-body

Then, the nh
Greens function as a core-propagator times a two-hole Greens function,
g(3) (t) T b(t)b (0) c0+ (t)c0 (t)c0 (0)c0+ (0) ,
(see the original paper for details). The core propagator can be expressed
a-la Langreth as shown in Section 11.2.3. Thus, the CVV spectrum from the
almost lled band can be modeled as a two-hole spectrum convoluted with an
1 assumption justies the low
asymmetric core line shape. Finally,the nh
density approximation (for holes); this is expected to be a good approximation
up to nh 0.1 if U is comparable to the band width, but this range shrinks
at stronger coupling, while at weak coupling any approximation works. The
LDA self-energy was calculated analytically for a band structure having an
arbitrary spectral density
B(), B < < 0
A(), 0 < < T
where B is the bottom and T the top of the band. Thus,nh =
and the non-interacting propagator is given by
B
T
A()
B()
+
.
ig0 () =
d
d

+
i
+
i
0
0
(12.38)
T
0
dA))
(12.39)
266
The results were displayed numerically for the rectangular band model of
Section (6.2). It was necessary to remove the rst-order (tadpole) contribution
(1) = iU n :
(12.40)
it would make the site where the Auger decay takes place more repulsive than
the rest of the sites. Thus,

() = iU n +
d g0 ( )T ( + ),
(12.41)
where the T matrix is given by

T () =
iU
(2)
1 + iU g0 ()
(2)
, g0 ()
d
g0 ( )g0 ( ).
2
(12.42)
The integrals can be done with the help of the Lehmann representation.
When U is large enough to produce split-o states in the closed-band theory,
there is structure outside the continuum. One nds
Re() U 2 A( U )(T + + U )( + U ),
(12.43)
that is, a bump between = U T and = U . As a result, there is

a bump in the density of states too, due to broadened split-o states. This
explained[91] a satellite peak observed in the Ni photoemission spectrum; the
physical picture is that the hole produced in the photoemission process has
a chance to nd another of opposite spin already there; a two-hole resonance
forms outside the band continuum, and decays into band holes plus a number
of electron-hole pairs. The Auger spectrum was also obtained in this model,
in qualitative agreement with the experiments.
The next step appeared to be the self-consistent LDA, obtained as in
Section (11.6); it had been applied to the Anderson Model [132] and was
expected to have an enhanced range of validity. To our surprise, the writer
and Verdozzi in cluster studies found[134] that the self-consistent procedure
actually ruins the approximation. Comparing with exact results we found
that the non-selfconsistent version, keeping only the diagrams that remain
in the closed-band limit, was much superior; we called this the Bare Ladder
Approximation (BLA). Actually, the self-consistent LDA puts weight in processes where many electron-hole pairs are virtually excited simultaneously,
but the on-site interaction does not allow this, and a vertex correction tends
to remove the dressing of the internal lines. The BLA has the further advantage that the Herglotz property is granted, that is, the density of states
is non-negative. This is of course quite necessary, but most approximations
have no built-in device to ensure this.
The property of vertex corrections that tend to undress internal propagators was later observer by people working on the GW approximation (see
Section 11.10).
267
12.3.2 The U<0 Phenomenon

Measurements of Auger spectra of compounds of Sc, Ti and Cr [159] showed
that the peak kinetic energy was higher than expected by Landers theory
(Section 6.2); shifts as large as 4 eV were reported. This led the authors to
conclude that in such cases the on-site interaction is U < 0 and to propose an
explanation based on a dynamic bipolaron eect. An alternative, U > 0 explanation was put forth by V
aclav Drchal and the writer[128]. We think that
the on-site interaction is always repulsive; the repulsion can indeed produce
an eective attraction by a correlation eect (W=0 pairing, see Chapter 17)
but this is o-site and the binding energies are expected to be of the order of
tens of meV. The probability [22] of measuring a photoelectron of momentum p and an Auger electron of momentum k in APECS (Sect. 6.4) is We
approximate the probability of emitting by

dt
dt f (t, t , p, k)eiE(tt ) ,
(12.44)
Itot (E, p, k) =
0
where, introducing complete set summations
over the states of the neutral
,
system,
f (t, t , p, k) =

,
c,c
|ac ei[H(0)+iop ]t ac |

(k)eiH(1)(tt ) HA (k) |
| HA
i[H(0)iop ]t
| ac e
ac | V (c , p) V (c, p)
(12.45)
Here, V (c, p) is the matrix element of the electro-magnetic Hamiltonian between the core-electron state c and the photoelectron state, H() is the
hamiltonian of the system with core electrons in the primary-hole state,
the operator op describes virtual Auger transitions, but is often replaced by
a constant

HA (k) =
M (k)a0 a0
is the operator that describes the valence hole creation at site 0 in the local
spin-orbitals denoted by greek letters; M (k) are Auger matrix elements.
Moreover, | is the true ground state with no core-hole ( = 1), and the sums
run over complete sets. For degenerate core levels a set of relaxed ground states
must be included. We considered a non-degenerate core state and assumed
that these sums are saturated by two main contributions, namely | = |
and | = | , where | is the relaxed ground state with the core-hole and its
screening cloud. In the case it is necessary one could easily extend the theory
including plasmon satellites and other excited states. One of the ingredients

is the core-hole Greens function gc (t) = i |ac ei[H(0)+iop ]t ac | . One
can observe the core density of states in Photoemission experiments as an
asymmetric line shape. We argue that the terms with
= are negligible,
since the factor
268

iH(0)(tt )
| HA
e
HA |
represents the evolution without the core-hole and it is unlikely that a screening cloud can be created taking from | to | if there in nothing to screen.
Let denote the inverse lifetime of the core hole and denote the inverse characteristic time of the electron-hole pairs screening the core hole
and contributing the asymmetry. We assume that both are small compared to the other relevant energies. thus, we may model the situation with
gc (t) = iei[ci( +)]t , where c is the core energy level, while
c (t) = i | ac ei[H(0)+iop ]t ac | ei tt
(12.46)
involves a screened core energy . Collecting these results,

Itot = |V (p)|2 [I + I ]
with the unrelaxed contribution
2
I = |M (k)|
d D(E )
2 ( c )2 + 2
written in terms of the two-holes density of states D as in closed-band theory,

and an average square matrix element. On the other hand,

d Irel (E )
I =
2
2
2 ( E) +
where the relaxed line shape requires modeling the screening cloud. We modeled the ground state of the system in the presence of the core hole, using a
Hamiltonian of the form
H = H0 + Hee + Hch ,
(12.47)
where H0 is the usual band term,

Hee = U
ni ni
(12.48)
is an approximate electron-electron interaction with i running over the lattice

sites and the greek indices over the local atomic valence orbitals, and

n0
(12.49)
Hch = W
is the core hole-valence electron interaction at site i=0 (W > 0). Letting |
denote the ground state with no core-hole ( determinantal state in HF and
Local Density approximations) we looked for the ground state | with the
core-hole in the form
| + (a +
269
bq a0 aq ) |
with q as the wavevector and bq variational parameters.

By minimizing the energy one nds a complicated but physically relevant
solution until it is clear that the solution must be found in the space of
distributions in terms of a wave packet

(h, ) = (1 n)(EF )hd (EF )(EF + + h)
as the following singular limit for h 0:
|F =

1
(1 n)d lim
(h, E(q))|q .
h0
N q
(12.50)
The relaxed state turns out to be

1
| =
a0 aF | .
(1 n)d
(12.51)
The screening electron comes from the Fermi surface. Now within the twostep model one can obtain the Fermi golden rule expression for the relaxed
spectrum

1 2|M (k)|2
Im[
dtei(Ei0)t
gc(hhe) (, t)]
(12.52)
Irel (E) =
(1 n)d
0
where
gc(hhe) (, t) = (i)3 | T [a0 (t)a0 (t)a0 (t)a0 a0 (t)a0 ] |
(12.53)
is a three-body causal Greens function averaged over the unrelaxed ground

(hhe)
state. In order to compute gc
we proposed the approximation 1 was shown
i Figure 12.3.2 for , and all dierent.
This means:
(hhe)
(hh)
(e)
(he)
(h)
(, t) = gc (, t)gc (, t) + gc (, t)gc (, t)
gc
(he)
(h)
(h)
(h)
(e)
+gc (, t)gc (, t) 2gc (, t)gc (, t)gc (, t),
, , all dierent
(note the factor 2, which is needed to give the correct U=0 limit.) When two
indices coincide, we nd
(hhe)
(h)
(hh)
(h)
(, t) = (1 n2 )gc (, t) + gc (, t)gc (, t)
gc
(he)
(h)
(h)
(h)
(e)
+gc (, t)gc (, t) gc (, t)gc (, t)gc (, t),
1
The exact three-body wave function for a system of interacting identical particles can be obtained by a method by Faddeev [206] but here we were looking for
a simple approximation for the three-body Greens function.
270
(hhe)
Fig. 12.4. The approximation for gc

introduced in ref. [128] for , and all
dierent. The black box represent ladder interactions involving pairs of Fermions,
while all diagrams involving all Fermions are neglected.
and a similar expression if = . The Auger intensity is

1
Irel (E) = ImA(E)(EF E)
(he)
(he)
the correlation part of gc

where, calling gc
(he)
between gc
and its U=0 limit,
(h)
A(E) = (1 n)2 gc (E) + (d 1)

d (hh)
(e)
+ d2 2i
gc (E )gc ().
(12.54)
, that is, the dierence
(he)
d
(E
2i gc
)gch ()
(12.55)
For partially lled bands, a one-body contribution arises, along with the rest
of 3-body contributions and the 2-body ones in the unrelaxed contributions;
we called our approach the 1-2-3 theory. In was shown in Ref. [128] (see
Figure 12.5 ) that this theory qualitatively reproduces the experimental trend
for early transition metals.
12.3.3 Correlation in Early Transition Metals
Part of the intensity in Auger CVV and APECS spectra from transition
metals comes from the decay of core-holes that are unscreened in the initial
state; this is obtained by a two-body Greens function, as in the lled band
case. However, much of the intensity comes from core holes that are screened
when the Auger decay occurs.
Some of these spectral features are amenable to the one-body g, but the
rest requires[128] a two-hole-one-electron propagator, that is a 3-body function g (3) . In the strongly correlated case the approximation of the last Section
runs into trouble because the Herglotz property is not granted (the density of
states may turn negative.) The Core-Ladder approximation [127]is an extension of the ladder approximation to this problem that has several attractive
features. It is based on the idea of formally rewriting the problem in terms of
271
kinetic energy, eV
-6
Fig. 12.5. Thick solid line:schematic drawing of the experimental Ti L2 M45 M45 line
shape (see Ref. [160]) on a kinetic energy scale centered at the maximum. Dashed
line: the theory of Ref. [128] (dashed), with U =1.684 eV, band width w=7.14 eV
and n=0.254. Thin solid line: the closed-band theory result with the same U and w,
which predicts too low kinetic energy and is clearly not applicable in this case.
a ctitious three-body interaction, where the third body is forward-scattered

by an interaction x.
x
a)
b)
Fig. 12.6. a) typical second-order diagram for the 1 hole- 2-electron function g (3) (t);
b) the same diagram with a ctitious 3-body interaction that allows to carry on a
partial summation of the 3-line ladder series to innite order.
272

The third body is ltered through a x= i |i i|type projector, that is,
it is left undisturbed. In core problems, one can exactly dispose of the innite
summation in favour of a local projector. The same applies to localized splito states that occur at strong coupling. Thus one can base an approximation
that can be systematically improved by including more terms if needed. It
treats electrons and holes at equal footing, carries on a partial summation of
the perturbation series to innite order, and becomes exact both at weak and
at strong coupling (in the sense that it becomes equivalent to a full Ladder
approximation). A particularly interesting feature is that our approach grants
the Herglotz property, that is, the density of states is granted to be nonnegative. Indeed it is a common drawback of perturbation approaches that
potentially powerful diagram summations become untenable by the failure to
guarantee this zero-order requirement of positive probability. In this case we
achieve the result by proving that there exists a model Hamiltonian for which
our Core-Ladder series gives the exact answer. The theory naturally explains
the apparent negative-U behaviour of the early transition metal spectra.
12.4 Conserving Approximations

When working with approximate self-energies, we must be ready to get some
quantitative details wrong, but we can make no use of approximations that violate the basic conservation laws. The continuity equation must hold; energy
and momentum conservation laws, when applicable, must also be obeyed. An
approximation that respects these fundamental symmetries is called conserving. The Hartree-Fock approximation is, but if we pick a general this
is not granted automatically.
12.4.1 Continuity Equation
Let us see how the Greens function formalism embodies current conservation.
The Dyson equation has two forms:

(12.56)
g(x, x ) = g0 (x, x ) + dx1 dx2 g0 (x, x1 )(x1 , x2 )g(x2 , x )
g(x, x ) = g0 (x, x ) +
dx1 dx2 g(x, x1 )(x1 , x2 )g0 (x2 , x )
By applying the g01 operator one obtains two equations:

21
U (1) g(1; 2) = (1 2) + d1(1; 1)g(1; 2)

+
i
t1
2m
2
recall from (4.20) that g(1, 2) = iT [(1) (2)].
(12.57)
(12.58)

2
+ 2 U (2) g(1; 2) = (1 2) + d1g(1; 1)(1; 2).

i
t2
2m
273
(12.59)
Taking the dierence and writing 21 22 = (1 + 2 ) (1 2 ), we

get:
1 2
g(1; 2) =
2m
[U (1) U (2)]g(1; 2) + Z(1; 2),
(it1 + it2 )g(1; 2) + (1 + 2 )
(12.60)
where
Z(1; 2) =
d
1 {(1;
1)g(
1; 2) g(1; 1)(1; 2)} .
(12.61)
Now we set 2 = 1+ , which means t2 = t+

1 , r 2 = r 1 . In the left hand side,
ig(1; 1+ ) = (1) (1) =
n(1)
(12.62)
is the particle density and

1 2
g(1; 2)
2m
2=1+
1
(1)1 (1) 1 (1) (1) = j(1)
2mi
(12.63)
is the current density. Hence we end up with the continuity equation, which
is the number conservation law in a dierential form, provided that the right
hand side vanishes, and this requires
Z(1, 1+ ) = 0.
(12.64)
The exact Dyson equations (12.56,12.57) must imply the continuity equation
and (12.64). The best way to check this is to use the equations of motion
(10.7) involving the two-particle Greens function

2
d
+ 1 U (1) g(1; 1 ) = (1, 1 ) i d1V (1, 1)G2 (1; 1 |1+ ; 1 )
i
dt1
2m
(12.65)
and

21
d

U (1 ) g(1; 1 ) = (1, 1 ) i d1G2 (1; 1 |1+ ; 1 )V (1, 1 ).
+
i
dt1
2m
(12.66)
In this way, taking the dierence, we get
274
(it1 + it1 )g(1; 1 ) +

i
1
(1 + 2 ) (1 1 )g(1; 1 ) =
2m
= [U (1) U (1 )]g(1; 1 )
d
1[V (1,
1) V (1 , 1)]G2 (1; 1 |1+ ; 1 ). (12.67)
When we set 1 = 1+ the r.h.s. vanishes.

This is an obvious statement about the exact and g; however it is also a
condition on approximations, which can only be tenable if they are conserving. The condition on approximate and g is that both forms (12.56,12.57)
of the Dyson equation be obeyed and (12.64) holds. Next, we discuss how to
build conserving approximations.
12.4.2 The Functional
There is a simple diagrammatic prescription for creating conserving approxi(n)
mations. Let skel. denote a skeleton diagram of order n for the self-energy
(1, 2) and imagine joining the two ends by a propagator g(2, 1) . The result
is the ring diagram in the next Figure:
(n)
skel.
1
2
g
Fig. 12.7. Building a contribution (n) to the functional (occurring 2n times)

(n)
from a skeleton diagram skel. for the self-energy, by closing the ends by an interacting propagator g.
(n)

=
(1, 2)g(2, 1)d1d2
(12.68)
similar to one of those bubble diagrams that contribute to the no-particle

propagator or ground-state energy (Section 11.1) , except that the 2n internal propagators are dressed g and there are no self-energy insertions. The
analytical expression of any diagram is a functional of g. As illustrated in
Figure 11.10 the inverse operation (opening the g(2, 1) line to restore the
original self-energy) is a functional dierentiation. We introduce an interaction strength parameter (which is 1 in the fully interacting case). Any
graph with 2n g lines must be divided by 2n to avoid multiple counting of
contributions to (1, 2) (see Figure 12.8). Let n represent the sum of all the
skeleton diagrams with n vertices, and
[g, ] =
n
n
2n
T r[g()n ()],
275
(12.69)
where the Tr operation now sums over spin and all one-electron labels and
integrates in d
. This leads to the following result.
Theorem 10. (Luttinger-Ward theorem) [55] The exact self-energy is given
by (note the order of arguments!)
(1, 2) =
.
g(2, 1)
(12.70)
A further theorem by Baym and Kadano [56] states that the continuity
equation and the momentum, angular momentum and energy conservation
laws are embodied in the functional;
Theorem 11. If (and only if ) a self-energy is -derivable, that is, comes according to (12.70) from some approximate , the approximation is conserving.
In other terms, even for approximate Equation (12.70) is equivalent to the
conservation laws.
The GW approximation is also -derivable and hence conserving.
1
2
+ 14
1
4
Fig. 12.8. Diagrams for and the corresponding . The rst-order example refers
to the Hartree-Fock approximation.
12.4.3 Gauge Transformation

We prove Theorem 11 for the continuity equation (the other proofs are somewhat longer but similar). Under a gauge transformation (x) (x)ei(x)
or
276
g(1, 2) ei(1) g(1, 2)ei(2) ,
(12.71)
is invariant, since for any line entering a vertex there is another one leaving
it. In an innitesimal gauge transformation , g changes by the amount
g = [(1) (2)]g(1, 2). Then to rst order as a functional of g will
change by the amount

0 = = d1d2
g(1, 2) = d1d2 (2, 1)g(1, 2)
g(1, 2)

= i d1d2 (2, 1)[(1) (2)]g(1, 2). (12.72)
Exchanging the dummy variables in the second term, Eq.(12.72) yields

0 = d1d2 [(2, 1)g(1, 2) g(2, 1)(1, 2)](1)
(12.73)
The coecient of (1) must vanish identically. that is

d2 {(1, 2)g(2, 1) g(1, 2)(2, 1)} = 0.
(12.74)
and we are back to Equation (12.64): the approximation is conserving.

12.4.4 Ground-State Energy and Grand Potential
(n)
2
g (0)
Fig. 12.9. Building a contribution (n) to the grand potential (occurring 2n times)
from a diagram for the self-energy. This looks very much like Fig. 12.7, but
please observe the dierences: propagators are bare, and the self-energy improper.
The grand potential is the sum of ring diagrams. One may work out the
grand-potential
and the ground-state energy
in parallel; just, in the former
case one uses 1 l , while in the latter case d. We use a short-hand notation valid for both cases 3 . Let (n) denote any such diagram of order n;
it may be thought of as the result of closing a self-energy diagram (improper,
in general) with a bare propagator. Mutatis mutandis, this is similar to the
3
Note however that some expressions are ambiguous if taken literally at T=0
and then one should take the T 0 limit at the end (see below).
277
procedure for building from the skeleton diagrams for the self-energy, since
there are 2n bare lines in (n) . J.M.Luttinger and J.C. Ward [55], derived
the diagram rules for the grand-potential in the frequency-momentum representation for Jellium. They wrote the contribution of order n, summed over
one-body states r, such that g (0) is diagonal,

(n)
1
1
(12.75)
T rg (0) ;
dgr(0) () r
n =
2n r
2n

here d stand for KB T times the summation over Matsubara frequencies
in the nite T case, and the Tr operation
implies summing over indices and
integrating over frequency. To compute n n one can use the fact that

an

a()d
n
n
an
a() =
=
.
(12.76)
n
0
n
n
Thus,writing r (, ) for at coupling constant ,

d (0)
1
= 0 +
gr () r (, ),
d
2 r
(12.77)
where 0 is the non-interacting limit. Recalling (2.23) one nds

0 =
1
ln(1 + e(r ) ).
r
(12.78)
Using the solution to Problem 11.2 and the Dyson equation we end up with

d
1
gr ()r (, ),
= 0 +
d
(12.79)
2 r
0
which is a new expression for the ground-state energy in terms of the Greens
function, involving a coupling-constant integration. This implies
d
1
= T rg(, )(, ).
d
2
(12.80)
12.4.5 Luttinger-Ward and ABL Variational Principles

From Equation (12.70) we know that =
g ; hence Luttinger and Ward
obtained a new functional with the interesting property of being stationary
with respect to the correct g, that is,
= 0,
g
(12.81)
provided that g obeys the Dyson equation g = (g01 )1 . Indeed, using

the same T r notation as above for the sums of ring diagrams
278
5
6
1
T r ln( g01 ) + g =
++g
= ,
1
g
g
g0 g
(12.82)
and so one nds that (12.81) holds for the functional

= T r{ln( g01 ) + g} .
(12.83)
Luttinger and Ward [55] further proved that, as the notation suggests, veries (12.80) and coincides with 0 for = 0; so, it is indeed the grand potential
(at nite temperatures) or the ground-state energy (at T=0). These results
have been generalized in the nineties; the Lund Group (C. -O.Almbladh and
which depends on g
U. von Barth) [136][137] have constructed a functional
and on the screened interaction W and is variational in both variables; they
by using simple approximations for
tested numerically the performance of
g and W and found that the results where competitive. The derivation in
Ref.[137] also points out the connection with Kohn-Sham theory (see below).
12.5 Generalized Ward Identities

Still, we have to exploit the continuity equation
+ divJ = 0
t
(12.84)
as a source of exact relations betwen the Greens

functions. Here, in obvious

h

notation, (x) = (x)(x), Ji (x) = i
(
. Let
2m
xi
xi ) (x )(x)
x =x
A1 , A2 denote Heisenberg operators such as fermion creation or annihilation
operators and an operator such as a density which commutes with A1 , A2
under Wicks T ordering. We know (see Problem 2.3) that
d
T {A1 (t1 )A2 (t2 )(t)} = T {A1 (t1 )A2 (t2 )(t)}
dt
+ (t t1 )T {[(t), A1 (t1 )] A2 (t2 )}
+ (t t2 )T {A1 (t1 )[(t), A2 (t2 )] }.
(12.85)
Following [140] (see also [1]) we set A1 (t1 ) = (x1 ), A2 (t2 ) = (x2 ), take
a (ground-state or thermal ) average, use the continuity equation to replace
and obtain
t T {(x1 ) (x2 ) (x)(x)} = T {(x1 ) (x2 )J(x)}
+(t t1 )T {[(x), (x1 )] (x2 )}
+(t t2 )T {(x1 )[(x), (x2 )] } .
(12.86)
Since [(x), (x1 )] = (x x1 )(x) and [(x), (x2 )] = (x x1 ) (x),

one obtains the Ward identity [141]
12.6 Connection of Diagrams to D F T
h
( )G2 (x, x1 , x2 , x ) =
2mi
= i (4) (x x1 )g(x, x2 ) i (4) (x x2 )g(x, x1 )
279
t G2 (x, x1 , x2 , x) +
(12.87)
where (4) (x) = (3) (r)(t) is the 4-dimensional and the limit x x is understood. The continuity equation is an expression of charge conservation and
is associated by Noethers theorem to the gauge invariance. Ward identities
generally arise from invariance Groups of the theory.

12.6.1 Highlights on Density Functional Theory
As W. Kohn emphasized in his Nobel lecture, the many-body problem is
rather ill posed in terms of the many-body wave function 0 (x1 , x2 , ), with
xi {ri , i }, and it of paramount practical and conceptual importance that
it can be formulated in terms of the density,

n(r) = 0 |
n|0 = N
dx2 dxN |(r, , x2 , , xN |2 . (12.88)
Indeed, 0 requires an immense amount of information for large N and a

very small change in it gives an orthogonal state, while n(r) is observable. I
summarize the topic, starting with the Hohenberg-Kohn theorem [121] .
Theorem 12. For an N-electron liquid in an external potential Vext (x), the
ground state energy E0 is an universal functional of the density n(r).
This means that n determines E0 and also the wave function 0 : symbolically,
n E0 , 0 .
Proof. Suppose we know the ground state wave function 0 (x1 , x2 , ) and
energy E0 of an interacting electron liquid in an external potential Vext
(Equations 1.60,1.61,1.62) with Hamiltonian H = T + U + Vext . Since
H0 = (T + U + Vext )0 = E0 0 , we can infer Vext (apart from an

, one gets
additive constant). Thus, for a dierent potential energy Vext

H 0 = (T + U + Vext )0 = E0 0 with 0
= 0 . Assuming 0 = 0 one

)0 = (E0 E0 )0 and Vext and Vext
should dier by
should nd (Vext Vext
a constant. If Vext is such that the ground state is degenerate, several wave
functions belong to E0 , but this conclusion is not aected.Thus, 0 is a wave
packet of eigenstates of H involving excited states.
Having shown that 0 determines Vext , that is, 0 Vext ; we next show
that we can do with much less information, since n Vext , too. Assume
= Vext such that n is the same; the change

the converse is true: then Vext
280

Vext Vext
produces a new hamiltonian H , with a dierent ground state
wave function 0 and eigenvalue E0 ; by the variational principle

E0 = 0 |H|0 < 0 |H|0 = 0 |H + Vext Vext
|0 .
Thus,
E0 < E0 +

d3 x[Vext (x) Vext
(x)]n(x).
But this is absurd! If one starts by considering H one nds

E0 < E0 + d3 x[Vext
(x) Vext (x)]n(x),
(12.89)
(12.90)
(12.91)
and summing the two one nds the contradictory result E0 + E0 < E0 + E0 .
Note that in the case of degeneracy the ground state density is not unique,
but this does not change the conclusion that the same n cannot be compatible
with two potentials. Thus, n Vext , or (in the DFT parlance) Vext (r) in an
unique functional of n(r). So, n Vext H and since in principle one can
solve the Schr
odinger equation, we may conclude that n 0 , E0 .
The functional E0 [n] is unknown: it would be the solution of all kinds of
ground state problems in a formula. We can think physically, and separate
out some obvious contributions. We set

E0 [n] = dxVext (x)n(x) + VH [n] + E[n]

(12.92)
where
VH [n] =
1
2
dxdx
n(x)n(x )
|x x |
(12.93)
is the Hartree term; this shifts the problem to a new unknown universal
The electrostatic potential is
functional E.

n(x )
=
[E0 [n] E[n]].
(x) = Vext (x) + dx
(12.94)

|x x |
n
Since
E0
n
= we are left with

(x) +
E
= .
n
(12.95)
One can try a simple approximation by extrapolating the high-density Jellium

result (12.28,12.29) (see also Problem 23). This is not accurate outside the
Jellium case. Gradient corrections involving n have been considered but do
not take far: the approach remains semi-classical, closely related to the (1927)
Thomas-Fermi scheme.
281
Kohn and Sham [122] invented the fully quantum mechanical version,
just in time to take advantage of the computer revolution. To nd a quantum kinetic energy functional T [n], rst of all one must realize that it exists. As noted above n Vext H. Now, H determines 0 , E0 and also
T = 0 |T |0 . We cannot carry on this program because we cannot solve
the Schr
odinger equation for the interacting system. However for a noninteracting system S of electrons moving in some external potential VS we
can. Let TS [n] be the kinetic energy of a ctitious N-electron system S having
the same density as the real one; Kohn and Sham wrote

(12.96)
E0 [n] = TS [n] + dxVext (x)n(x) + VH [n] + Exc [n],
where Exc [n] is a new unknown exchange-correlation functional. Indeed, we
can apply to S the above argument showing that the density yields the external potential: n VS . Therefore, the required potential ensuring that S
has the exact density must exist; having separated out the kinetic energy of
S (that will dier from the exact kinetic energy of the interacting system)
we can write (12.96) and pretend we know all but Exc [n] potential that must
be approximated in some way, usually drawing from the Jellium theory at
the density n(x) prevailing at a given point. This is the Local Density Approximation. Among the popular approximations, I mention the interpolation
formula
1

4
4
3 3 3
n3
3
Exc [n] =
(12.97)
n 3 d3 r 0.056
1 d r
10
0.079 + n 3
and the extensions including gradient corrections

/
0
2
(n)|n(r)|
+
.
Exc [n] = d3 r xc (n)n(r) + (2)
xc
(12.98)
Exc [n] produces the exchange-correlation potential

Vxc =
Exc
.
n
(12.99)
This contributes to the chemical potential, along with the electrostatic potential , since
E0
TS
==
+ + Vxc .
(12.100)
n
n
Thus in the ctitious system the electrons do not interact but feel a potential
VS = (x) +
Exc
,
n
(12.101)
which does not coincide with the screened potential but includes Vxc . One
must solve self-consistently the set of N equations, the Kohn-Sham equations
282
1
( 2 + VS )i (x) = i i (x),
2
(12.102)
and compute TS and the density

n(x) =
N
|i (x)|2
(12.103)
i=1
from the determinant of the orbitals i (x). The exchange-correlation potential can be improved over the jellium estimate (12.29) by including the
generalized gradient corrections[138] and the accuracy of the results strongly
depends on the system. In metals, typical error may be 0.3 eV per atom
but for nite systems they are usually much worse. Bond lengths are usually
reproduced to better than 109 cm.
Further Developments
The density functional perturbation theory is a technique that allows to calculate e.g. phonon dispersion in solids[149]. I quote the following useful result
by Janak[146] without reporting the proof.
Theorem 13. In density functional theory,
E
= i ,
ni
(12.104)
that is, the derivative of the total energy with respect to the occupation number
of a Kohn-Sham orbital is equal to the eigenvalue of that orbital.
Vignale and Rasolt [119]formulated the current-density-functional theory,
an extension which is needed to include magnetic elds; the current couples
to the vector potential A and a term in A2 appears. A few pioneers [129],[130]
tried time-dependent versions of Kohn-Sham equations in atomic problems;
the results were rather encouraging, but a justication was lacking untill
Runge and Gross[131] found a suitable formalism. This was based on the
t2
H(t)|(t) , but for

variational principle A = 0, where A = t1 (t)|i t
the rest the theory parallels the static one. For further details see[54].
This is the basis of Density Functional method on which most ab-initio
calculations are performed, [125] and is of paramount importance as a practical calculational scheme when the correlation eects are not too severe. Very
good comprehensive reviews on this subject are available [125], [54]; however
I wish to discuss less well known consequences.
12.6.2 Sham - Schl
uter Equation
There is no diagrammatic expansion for either Exc or vxc , however L.J. Sham
and M. Schl
uter [123] devised a scheme that in principle allows to deduce vxc
283
from the self-energy. The exchange-correlation self-energy xc , which is all the

self-energy minus the Hartree term, enters the Dyson equation for the Greens
function g in terms of the Hartree approximation gH (which also accounts for
the external potential); in a shorthand self-evident matrix notation reads:
1
xc .
g = gH + gH xc g g 1 = gH
(12.105)
In the ctitious non-interacting system the electrons see an eective potential

vef f = Vext + VH + vxc .
(12.106)
Thus the Greens function gKS of the ctitious Kohn-Sham system obeys
1
1
gKS = gH + gH Vxc gKS gKS
= gH
Vxc .
(12.107)
rather than Equation (12.105). Hence,

1
g 1 gKS
= (xc Vxc ) g = gKS + gKS (xc Vxc )g.
(12.108)
Now, we observe that gKS

= g, but both yield the same density; thus, introducing as usual the positive innitesimal ,
g(r, t, r = r, t + ) = gKS (r, t, r = r, t + ) = n(r).
(12.109)
Hence,
[gKS (xc Vxc )g]r =r,t =t+ = 0;
(12.110)
more explicitly, letting x (r, t), x (r, t + ), this means

dx dx gKS (x, x ) {xc (x , x ) (x x )Vxc (x )} g(x , x+ ) = 0.
(12.111)
If an approximation to xc is given, the corresponding g can be obtained
from the Dyson equation and this may be considered as an equation for
uter equation is awkward and highly non-linear
Vxc (x ). This Sham - M. Schl
(the unknown is also contained in gKS ) however it is considered [126] as an
important source of information on Vxc . Like the density-functional approach,
the Sham - M. Schl
uter equation also holds generally for time-dependent
problems.
Problems
12.1. If as an approximate E[n] one keeps only t[n] (see Equation 12.28),
what kind of approximation is obtained?
12.2. Evaluate the diagram of Figure 11.2.2 with all si = a.
13 Non-Equilibrium Theory
13.1 Time-Dependent Probes and Nonlinear Response

Phenomena like electron transport, electronic transitions in chemisorption,
desorption1 , molecule-surface collision processes, and sputtering 2 are naturally described by time-dependent electronic Hamiltonians. What happens in
the Fano-Anderson model (Chapter 5.1.2) if the and V parameters depend
on time? Such a time-dependence occurs because the Hamiltonian of electrons depends on the position of the nuclei, and, through this, from the time
3
. Such problems are completely outside the scope of the diagram method discussed so far; the equation of motion method (EOM) can work but depends
on some approximate truncation procedure. This Chapter is devoted to the
quantum theory of such non-equilibrium processes. A powerful generalized
perturbation method enables us to deal with time dependent problems and
processes that are far from equilibrium, generalizing the Kubo approach [29]
to all orders. It is a general and in principle exact technique; the Feynman
method is recovered as a particular case. There is no need to assume that
the perturbation is small or that the system deviates little from equilibrium;
one does not need to make any assumption that the system evolves in a
reversible way. As one could expect, the study of the excited states and outof-equilibrium situations involves special diculties, however this generalized
theoretical framework is also powerful for exploring the equilibrium nonlinear
response to strong perturbations, that can be treated to all orders.
1
See Chapters 6.2, 10.2.1.

This is an important process in industry, and generates a clean surface under
vacuum. The surface is bombarded with ions with energies in the KeV range, and
can typically disintegrate with the speed of some atomic layers per second.
3
In a truly complete theory, the Hamiltonian would be time-independent and
there would be no external elds, but then one should simultaneously deal with
all the degrees of freedom, including light sources, electron beams and all that.
Already in the static case this turns out to be dicult and one tries to apply, when
possible, the Born-Oppenheimer approximation. The obvious generalization of such
approximation to the dynamic case consists in treating the nuclei like classical
particles, with a well dened trajectory: the electrons then are subject to a timedependent potential.
2
286
To this end we will develop a formalism introduced many years ago by

L.P.Kadano and G.Baym[82], above all in the version elaborated little later
by L.V.Keldysh[84]. A somewhat generalized formulation in which the initial
state is completely arbitrary was provided by Wagner [101]. In the literature,
the application of this method is not yet very frequent, although it is gaining
ground in recent years e.g. in the eld of transport theory; it is still used
much less than its potential and importance would deserve; this is due to a
the relative shortage of examples which constitutes an obstacle to its spread,
and surely to its undeniable complexity. The technique generalizes the diagrammatic development of the Green functions to time-dependent problems
and systems that are far from the thermodynamic equilibrium. Several kinds
of Green functions are indeed needed, each with its place in the theory. Since
there are several ways to choose which functions to dene and since all possible choices have been made in the literature, the reader of the original articles
nds a true zoo of such functions. I ll introduce all the most common species,
but having care to specify their relations; Ill try to avoid the notation inconsistencies that frequently transform those exotic beasts into monsters.
13.2 Kadano-Baym and Keldysh Methods

Kadano and Baym[56] and Keldysh[84] devised dierent versions of the
generalized diagram method; ultimately they are equivalent but the KadanoBaym formalism is more popular in Statistical Mechanics while the Keldysh
has been mostly used for dynamics. Here Ill follow the Keldysh scheme.
As in Equation (2.1) we split the Hamiltonian in two parts, H(t) = H0 +
V (t), with a free part H0 , and a perturbation V (t) which is not necessarily
small nor even time-dependent. In Section 2.3 we found how to expand the
Interaction Picture evolution operator UI in powers of VI ,
t

UI (t, t0 ) = T exp i
dt VI (t ) , UI (t, t0 ) = UI (t0 , t).
(13.1)
t0
The operators in the Greens functions are in the Heisenberg picture; however
for any operator A we can switch representation by starting the evolution
from a golden age t0 when Heisenberg and Schrodinger pictures are the same,
according to
(13.2)
AH (t) = UI (t, t0 )AI (t)UI (t, t0 ).
There are four UI factors to expand in series in

g < (t, t ) = 0 (t0 )|H
(t )H (t)|0 (t0 ) =
0 |UI (t , t0 )I (t )UI (t , t0 )UI (t, t0 )I (t)UI (t , t0 )|0 ,
(13.3)
although we can reduce them to 3 since UI (t , t0 )UI (t, t0 ) = UI (t , t0 )UI (t0 , t)
and this is UI (t , t) by the group property. It would still be cumbersome to
13.2 Kadano-Baym and Keldysh Methods
287
expand the three UI factors, but we can do with just one expansion, since for
each operator A
AH (t) = UI (t, t0 )AI (t)UI (t, t0 ) =
t
t0
i
dt VI (t )
i
dt VI (t )
t0
t
= (T e
)AI (t)(T e
)=

t
dt VI (t ) AI (t)
= TC exp i
(13.4)
t0
where C is any oriented path in complex time through t0 and t , using the
generalized time-ordering TC along C that we met ins Section (2.2.1). Note
that AI (t) is under the action of TC that places it appropriately. In a similar
Im(t)
t0
Re(t)
Fig. 13.1. A contour on the complex t plane for inserting A(t) is a single
interaction-picture evolution.

way, we can read from left to right g < (t, t ) = 0 (t0 )|H
(t )H (t)|0 (t0 ) as
one story: the system starts at the golden age t0 , evolves to t, is acted
on by , then evolves to receive the action of at time t and eventually
it evolves back to the golden age. Physically, t can be before or after t. In
this story, we meet after because g < is dened with on the left of
. Thus along the path C = C1 U C2 , t precedes t , I write t <C t , and
Im(t)
t0
Re(t)
t

Fig. 13.2. The contour C for g < on
> the t plane with t <C t . Note that C can be
analyzed as a two-step path C = C1 C2 , C1 starts from t0 and returns there after
visiting t, and C2 starts from t0 and returns there after visiting t. The rst return
and restart from t0 can be avoided thanks to the group property.
288

i
d VI ( )
i
d VI ( )
C1
C2
g < (t, t ) = TC1 e
I (t )
TC2 e
I (t)

i
d VI ( )
C
I (t )I (t)
= TC e

i
d VI ( )
C
= TC e
I (t)I (t ) , t <C t .
(13.5)
Because of the group property, the contour C is largely arbitrary. It can go
back to t0 between t and t any number of times, including 0. The terms
arising in the series development of the operators are ordered automatically
by TC with earlier times (on C) to the right. Moreover,

>

d VI ( )} I (t)I (t )
g (t, t ) = TC exp{i
C

= TC exp{i
d VI ( )} I (t )I (t) , t >C t
(13.6)
C
We can also use the same C in both cases, placing t and t in opposite orders.
As we know, the knowledge of both g > an g < gives access to the physically
important retarded and advanced Greens functions. We also need to dene
a time-ordered (on C) Greens function:

ig(t, t ) = TC H (t)H
(t ) = g > (t, t )C (t t ) g < (t, t )C (t t), (13.7)
where the contour C goes through t and t and C (t t ) = 1 if C is such

that t is met rst and 0 otherwise.
13.3 Complex-Time Integrals by Langreths Technique

Let A and B denote contour-time-ordered Greens functions analyzed in >
and < parts as in Equations ( 13.5, 13.6)
A(t, t ) = ia> (t, t )C (t t ) + ia< (t, t )C (t t)
B(t, t ) = ib>(t, t )C (t t ) + ib< (t, t )C (t t);
(13.8)
following Langreth we must develop their combinations in series and in parallel which are needed to calculate diagrams in this theory. The combination
in series D = AB is dened by

d A(t, )B(, t )
D(t, t ) = AB(t, t ) =
C
= id> (t, t )C (t t ) + id< (t, t )C (t t).
(13.9)
and must be rewritten as a combination of ordinary real-axis integrals. To

calculate
d< (t, t ) we want t to be earlier on the contour, so we adopt C =
>
C1 C2 like in Figure (13.2). C1 starts from t0 and returns there after visiting
13.3 Complex-Time Integrals by Langreths Technique
289
t, C1 starts from t0 and returns there after visiting t. Along C1 , <C t

B = ib< , while along C2 , >C t A = ia< , hence
d< (t, t ) = iD = d< [C1 ] + d< [C2 ]

d [C1 ] = i

<
d [C2 ] = i
(13.10)
d A(t, )ib< (, t )
<
C1
d ia< (t, )B(, t ).
(13.11)
C2
On C1 , A = ia> on the t0 < < t branch, A = ia< on the back trip, so

t
t0
d< [C1 ] =
d (i)a> (t, )b< (, t ) +
d ia< (t, )b< (, t ) =
t0
t
t
6
d (i) a> (t, ) a< (t, ) b< (, t ).(13.12)
t0
Now we let t0 , and formally extend the integration to the full real
axis introducing a theta function. The result is

3
5
6
4
<
d [C1 ] =
d (i) a> (t, ) a< (t, ) (t ) b< (, t )

=
d ar (t, )b< (, t ) = ar b<
(13.13)
using Langreths convenient shorthand notation (product of small letters for

real axis integrals). One nds d< [C2 ] and d> in a similar way and gets
d< = ar b< + a< ba ; d> = ar b> + a> ba .
(13.14)
From d< , d> one nds dr , da :

dr = i(t t )[d< + d> ] = i(t t )[ar b< + a< ba + ar b> + a> ba ], (13.15)
that is, more explicitly,
dr (t, t ) = i(t t )
3
5
6
d ar (t, ) b> (, t ) + b< (, t ) +
6
4
5 >
a (, t ) + a< (, t ) ba (, t ) . (13.16)
Since g < + g > = i(gr ga ), we can simplify this to read

d {ar (t, )br (, t ) aa (t, )ba (, t )} .
dr (t, t ) = (t t )
(13.17)
However, the second integrand vanishes unless t > > t but then the in
front of the integral vanishes; therefore, we conclude that
290
dr = ar br , da = aa ba .
(13.18)
The series combination E = ABC is immediately obtained:

e> = (AB)r c> + (AB)> ca = ar br c> + (ar b> + a> ba )ca ,
(13.19)
e< = (AB)r c< + (AB)< ca = ar br c< + (ar b< + a< ba )ca ,
(13.20)
and so on. Langreth also introduced the parallel combination

F (t, t ) = A(t, t )B(t , t),
(13.21)
where no integration is understood. Recalling (13.8) and splitting F in the

same way, one nds
if > (t, t ) = a> (t, t )b< (t , t)
if < (t, t ) = a< (t, t )b> (t , t)
(13.22)
where again no integration is understood. Hence,

fr (t, t ) = i(t t )(f > + f < ) = (t t )[a> b< a< b>] .
(13.23)
Other forms are useful. Substituting a> (t t ) by (iar a> )(t t ), and
using b> + b< = ibr , we obtain
fr (t, t ) = i(t t )[ar b< a< br ];
(13.24)
alternatively one can eliminate b (t t ) in favor of (ibr b )(t t ) to

write
(13.25)
fr (t, t ) = i(t t )[a> br ar b> ];
<
>
the advanced part is dealt with in the same way.

13.3.1 Finite temperatures
For the nite temperature extension of this technique we must insert at the
end of the contour C the vertical track in Figure 2.2.2 a). Then one can
introduce k (, t), k (t, ) and the Matsubara function k M (, ) where t is
on the real time axis and along C comes before , which are imaginary, on
the vertical track. Moreover let us detote the integrals on the vertical track
i
f $g =
dzf (z)g(z).
(13.26)
0
Equations (13.14) become

d< = ar b< + a< ba + a $ b , d> = ar b> + a> ba + a $ b ;
(13.18) remains as it is. The new functions obey (see [101], [120]):
dM = aM $ bM
d = ar b + a $ bM

d = aM $ b + a b a .
(13.27)
(13.28)
13.4 Keldysh-Dyson Equation on the Contour
291

The Feynman method of Chapter 11 is generalized by expanding

i
d VI ( )

C
ig(t, t ) = TC H (t)H (t ) = TC e
I (t)I (t )
(13.29)
where the evolution in the interaction picture is on C and4

e

C
d VI ( )

(i)n
=
d1
dn V (1 ) V (n ).
n!
C
C
n=0
(13.30)
The diagrams are the same as in the Feynman scheme, but the time integrals
must be done on C, e.g. by the Langreth technique.
Example: Independent Particles
In independent-particle problems, V is a one-body operator. Lets write
V (t) = v(t) ,
(13.31)
where a matrix multiplication on spin-orbital indices such as in

v(t)kk k k
V (t) =
kk
is understood. In rst-order,

5
6
2
d v( )TC ( )( )(t) (t ) .
g = (i)
(13.32)
Using Wicks theorem (Section 11.1.1) we may write

TC [ ( )( )(t) (t )] = TC ( )( ) TC (t) (t ) +
TC (t) ( ) TC ( ) (t ) (13.33)
with the equal-times rule (11.69)
TC ( )( ) = (+ )( )
(13.34)
where + is just after and just before. Each contraction TC (t)(t ) =
ig 0 (t, t ) brings an unperturbed Greens function; each product of contractions
brings a factor ()P which is the signature of the permutation of operators
implied in doing the contraction. So, 13.32 becomes:

3
4
d v( ) ig 0 (, )ig 0 (t, t ) + ig 0 (t, )ig 0 (, t )
(13.35)
g = (i)2
C
292
+
t
Fig. 13.3. First-order contribution of a one-body perturbation
The rst contribution in Figure 13.3 is disconnected. It contains a factor

d ( ), where is independent of t and t . The contour C may be taken
on the real axis from to and then back to ; the integral vanishes
independently of the positions of t and t on the contour. This result extends
to all orders, the Linked Cluster Theorem of Section 11.9.1 again applies,
and the unlinked diagrams yield nothing. So, one obtains the same rules as
in Section 11.9.1 with the only dierence that the time integrals are on C.
Thus, one obtains the Dyson equation,
g = g 0 + g 0 g
(13.36)
again with the time integrals on C.

Solving the Dyson equation
If is known, one can develop the Dyson equation where g 0 g is a series
combination. Indeed, the self-energy, like g, depends on two times, and we
write:
i(t, t ) = > (t, t )C (t t ) < (t, t )C (t t).
(13.37)
Naturally, for t
= t one also denes
ir (t, t ) = [ > (t, t ) + < (t, t )] (t t ),
ia (t, t ) = [ > (t, t ) + < (t, t )] (t t).
(13.38)
So, from (13.36) one obtains:

gr = gr0 + gr0 r gr
(13.39)
ga = ga0 + ga0 a ga
>
g =g
4
0>
gr0 r g >
gr0 > ga
(13.40)
+g
0>
a ga
(13.41)
from now on, all operators are understood to be in the Interaction Picture, and
I drop the index I. Moreover Ill often understand
h to simplify the writing of the
equations.
g < = g 0< + gr0 r g < + gr0 < ga + g 0< a ga .
293
(13.42)
However all the information is contained in g > and g < ; we need just
two equations. The equations for g (r) and g (a) are equivalent and provide
[ > (t, t ) + < (t, t )], so all we need is
f = g> g<
(13.43)
= > < .
(13.44)
and to this eect we dene
The dierence of the equations for g > and g < gives

f = f 0 + gr0 r f + gr0 ga + f 0 a ga .
(13.45)
We can also give a closed solution in terms of the self-energy. This equation
is a matrix equation in the indices t,t with the formal solution
"
5
61 ! 0
f = 1 gr0 r
f [1 + a ga ] + gr0 ga . .
(13.46)
The matrix inversion is readily carried out:

5
1 gr0 r
61
moreover,
So,
5
6
= 1+gr0 r +gr0 r gr0 r + = 1+ gr0 + gr0 r gr0 + r = 1+gr r ;
5
1 gr0 r
61
gr0 = gr0 + gr0 r gr0 + = gr .
f = [1 + gr r ] f 0 [1 + a ga ] + gr ga .
(13.47)
In summary, once is known, a clear cut procedure yields g. One can use a
and r to solve for ga and gr like in one-electron problems. The results and
are then used in Equation 13.47 for f , where the dependence on the Fermi
level enters. Eventually, from ga , gr and f one also obtains g < and g > .
Dynamical Independent-Particle Problems
The special case of independent electrons can be solved exactly but is far
from trivial due to many-body eects; this case also lends itself to include
interactions approximately in a dynamical-Hartree-Fock or time-dependent
density-functional scheme. In this case all the self-energy comes from the
connected rst-order diagram on the right of Figure 13.3, that is,

g =
d g 0 (t, ) v( ) g 0 (, t );
(13.48)
C
this must be interpreted as
294
g =
d
C
which implies that
d g 0 (t, ) (, ) g 0 (, t )
(13.49)
(, ) = v( )C ( )
(13.50)
that is, the time-dependent self-energy is instantaneous and the propagator

line comes out of it at the very same point along the contour where it entered
the interaction. This implies that < , > and vanish, so r = a = 0,
but at equal times equations (13.38) do not hold, so we may have r and a
proportional to (t t ). From g = g 0 vg 0 one gets

gr = gr0 vgr0
(13.51)
<
g = gr0 vg 0< + g 0< vga0

where gr0 vgr0 = d gr0 (t, )v( )gr0 (, t ), while the Dyson equation is
gr = gr0 r gr0
g < = gr0 r g 0< + gr0 < ga0 + g 0< vga0 .
(13.52)
This means that the self-energy is given by

= 0; r (t, t ) = a (t, t ) = (t t )v(t).
(13.53)
13.5 Evolution on Keldysh Contour

The contour C can be deformed freely as long as it starts and ends at the
golden age t0 and goes through t and t. We have used one with two wings
ad hoc for series combinations. We could keep introducing new contours to
calculate the various combinations arising in the diagrams; however it is more
practical to have a standard contour and consider the various position of the
times on it. In time-dependent problems, one usually prefers the Keldysh contour of Figure (15.3.2) from t0 to t0 and back to t0 ; there are an ascending
or positive branch and a descending or negative branch, and g(t, t ) can have
each of its arguments on any of the two. Thus, there are 4 dierent Greens
functions dened on the real axis, that as above I shall denote with small
letters. Taking both times on the positive branch, one gets
(t+ ) .
g c (t, t ) = iTC H (t+ )H
(13.54)
Note that g c (t, t ) = iT H (t)H

(t ) is nothing but the time-ordered
Greens function of the usual theory
ig c (t, t ) = g > (t, t )(t t ) g < (t, t )(t t).
(13.55)
The traditional diagram expansion, that holds for constant V , is a special case
of the general one. The time sequence of Figure 13.5 in the interaction picture yields ()(i)U (, )U (, t )I (t )U (t , t)I (t)U (t, ) where a
295
(-) factor arises from the interchange of and . Including the other ordering of t and t, one obtains iU (, )T I (t)I (t )U (, ) with
the descending or backwards evolution operator U (, ) relegated on the
far left. The expectation value is taken at the golden age t0 over the noninteracting ground state (Chapter 1).
t
t0
Re t
Fig. 13.4. The Keldysh contour with t and t on the ascending branch.
At nite temperature one completes the contour as follows:

t
t0
Re t
i
Fig. 13.5. The Keldysh contour with t and t on the ascending branch. The vertical
track is necessary for nite-temperature averages as in Figure 2.2.2.
It is interesting to note that if V is static one can adopt the adiabatic

switching on at t0 and o at t0 ; then the adiabatic theorem states that
|U (, ) = ei |. So,
5
6
|T (t) (t )U (, ) |
c

g (t, t ) = i
,
(13.56)
|U (, )|
which is just Equation (11.63), the starting point of the usual expansion.
If V = V (t), the adiabatic theorem does not apply, and the new formalism
is needed, however even if V is constant there is no need for the adiabatic
theorem in the new formalism. So, the general scheme is more powerful and
conceptually simpler.
Taking both times on the negative branch one obtains
gc (t, t ) = iTC H (t )H
(t ) = iTC H (t)H
(t ) ,
(13.57)
with T that orders times as the reverse of T ; instead of (13.55) we write
296
i
g c (t, t ) = g > (t, t )(t t) g < (t, t )(t t )
(13.58)
The other two combinations, where the order on C is xed, are
g + (t, t ) = iTC H (t+ )H

(t ) = ig < (t, t )

g (t, t ) = iTC H (t )H (t+ ) = ig > (t, t ).
(13.59)
g + (t, t ) + g (t, t ) = g c (t, t ) + gc (t, t ),
(13.60)
g c (t, t ) g + (t, t ) =
6
5
i g > (t t ) g < (t t) ig < [(t t ) (t t)]
6
5
= i g > + g < (t t ) = gr (t, t ),
(13.61)
Note that
and in a similar way5

g c (t, t ) g (t, t ) = ga (t, t ).
(13.62)

On the Keldysh contour, C d = dt+ + dt ; so one can write all
integrals
on the real axis by introducing a matrix notation with the pattern

++ +
of time labels, that is, by introducing the 2 2 matrix Greens

function
G(t, t ) =
gc g+
g gc

.
(13.63)

However, since we wish to rewrite dt as dt we must insert a
- sign for each integration in dt . This is conveniently done by inserting a
Pauli z matrix factor between any two matrix Greens functions in series
combinations. For example, the linked rst-order diagram due to a one-body
perturbation V (t) given by Equation (13.48) becomes

d g (t, ) v( ) g (, t )
g =
C
d v( ) G 0 (t, )z G 0 (, t ). (13.64)
Here
0
G (t, )z G (, t ) =
g c (t, ) g + (t, )
g (t, )
g c (t, )
g c (, t ) g + (, t )
g (, t ) gc (, t )

(13.65)
5
These are the correct equations, although in the literature gr and ga are often
confused.
297
Check: Dysons equation from Keldysh matrix formalism

The Keldysh formalism the reproduces the results of Sect. 13.3 (although it
may be more friendly when performing the diagram expansion). To see that,
we dene
c +
x x
X (t, t , ) =
(13.66)
= G 0 (t, )z G 0 (, t )
x x
c
and do the matrix multiplications; X is given by
c

g (t, )g c (, t ) g + (t, )g (, t ) g c (t, )g + (, t ) g + (t, )
g c (, t )
.
g (t, )g c (, t ) gc (t, )g (, t ) g (t, )g + (, t ) gc (t, )
g c (, t )
Thus for instance x = ix> = g (t, )g c (, t ) gc (t, )g (, t ) that is,
x> = [g > (t, )g c (, t ) gc (t, )g > (, t )] .
Eliminating g c by Equation (13.62) and gc by Equation (13.60), one nds
>
x = g > ga + gr g > , in agreement with (13.14). The Dyson equation reads

G(t, t ) = G 0 (t, t ) +
d1
d2 G 0 (t, 1 )z (1 , 2 )z G(2 , t ), (13.67)
where of course
=
c +

c

.
(13.68)
Again, this can be cast as in Sect. 13.3 in terms of advanced and retarded
functions. To this end, we use the unitary matrix
+
+
,
,
1 1
1
1
1 + iy
1
2
2
2
2
= 1 1
U=
, U =
(13.69)
12 12
2
2
2
leading to
+
U G U
gc g g+ +
gc gc g +g+
gc
2 +
2 +
c
c
c
g g +g +g +
gc
g +g g
2
2
0 ga
gr f

(13.70)
where f = if = g + + g and
U U
0 a
r

(13.71)
= ++ . To transform the Dyson equation we also want

where = i
01
U z U 1 = x =
; the self-energy appears in the combination
10
U z z U 1 = x
0 a
r

x =
r
a 0

,
(13.72)
298
and in the literature sometimes is dened as the latter matrix, omitting

the problem that leads to introducing the z factors. Using the Langreth
shorthand notation of Section 13.3 the Dyson equation becomes (since the
(-i) factors in f and cancel)

0 ga0
0 ga
0 ga
0 ga0
r
=
,
(13.73)
+
gr f
gr0 f 0
gr0 f 0
a 0
gr f
which agrees with the previous result (Equations 13.39 and following ones).
13.5.1 Contour Evolution of Bosons
In the case of bosons, TC does not imply any sign change when two creation or
annihilation operators are exchanged. Greens function are written in terms of
eld operators . For instance, for phonons with eld operators m = dm +dm
one denes[62] the Greens functions
Dmn (t, t ) = iTC m (t)n (t )
(13.74)
>
with Dmn
(t, t ) = m (t)n (t ) and so on, such that
>
<
iDmn (t, t ) = Dmn
(t, t )C (t t ) + Dmn
(t, t )C (t t).
(13.75)
Moreover, on the Keldysh contour

D+ = iD< , D = iD>
(13.76)
without the sign change, and so

+
Dmn
(t, t ) = in (t )m (t) = Dnm
(t , t).
(13.77)
Moreover, in the denitions of the retarded and advanced functions,

r
Dmn
(t, t ) = i(t t )[m (t), n (t )] ,
(13.78)
a
Dmn
(t, t ) = i(t t)[m (t), n (t )] ,
(13.79)
commutators rather than anticommutators occur. With these modications,

the Keldysh formalism extends to bosons.
13.6 Selected Applications of the Keldysh Formalism

Being a general theoretical set-up for the many-body problem, this diagram
method is virtually unlimited in scope. The examples discussed below were
selected among the most instructive known to the writer.
299
13.6.1 Atom-Surface Scattering

Consider processes like atom-surface scattering, desorption or sputtering,
when a moving atom interacts with a surface. As in Section 10.2.1 we model
the situation by a Fano model (Chapter 5) with a non-degenerate atomic
energy level d interacting via time-dependent hopping matrix elements Vdk (t)
with a solid having a continuum of Bloch eigenvalues k ; spin indices are
suppressed in this problem. For the sake of deniteness we consider below a
problem, where the atom is scattered or is being emitted from the surface,
Vdk (t) 0 for t , and we are interested in the probability of detecting a
neutral atom or a ion. We want the long-time limit of the mean occupation
number nd (t) = cd (t)cd (t) . We split the hamiltonian H = H0 + Hh , H0 =

kC, k nk, + d n0, with Hh =

k, {Vk ak a0 + h.c.} like in ( 5.1) but
here it is convenient to write

H0 =
c c
(13.80)
with greek indices running over d and all the k states. Up to time t0
the system is in an eigenstate of H0 specied by occupation numbers n0 ;
this sets the initial condition of the problem. In some cases we wish to use
adiabatic switching to introduce Hh , as in desorption problems; in scattering
problems on the other hand the initial condition does not necessarily coincide
with the ground state oh H0 . The unperturbed Greens functions are readily
obtained.

0<
(t, t ) = c (t )c (t) = n0 ei (tt )
(13.81)
g
0>
g
(t, t ) = c (t)c (t ) = (1 n0 )ei (tt ) .
(13.82)
Hence,the unperturbed retarded and advanced functions are

0
gr,
(t, t ) = i ei (tt ) ,
0
(t, t ) = +i ei (t t) ,
ga,
(13.83)
(13.84)
both independent of occupation numbers. The perturbation is, with a slight

change in notation,

HI (t) =
Vdk (t)cd ck + h.c.,
(13.85)
k
and we want
<
c
nd (t) = cd (t)cd (t) = gdd
(t, t) = igdd
(t, t + 0), t
where t + 0 is just more than t.
(13.86)
300
The self-energy is given by (13.53) where a and r ar matrices in the

state indices; he Dyson equation may be simplied using the instantaneous
character of the self-energy to read

0

0
gr, (t, t ) = gr, (t, t ) +
d gr, (t, )V ( )gr,
(, t ). (13.87)
0
But gr,
and we can further simplify 6

0

0
gr, (t, t ) = gr, (t, t ) +
d gr, (t, )V ( )gr,
(, t ). (13.89)
The local (dd) component has a closed Dyson equation; indeed, using (in
matrix notation)

0
0
gr,dd = gr,dd
+ gr,dd
Vdk gr,kd
(13.90)
k
and inserting
one nds
0
gr,kd = gr,kk
Vkd gr,dd
(13.91)
0
0
+ gr,dd
r(d) gr,dd ,
gr,dd = gr,dd
(13.92)
with the dynamical self-energy

r(d) (t, t ) =
0
Vdk (t)gr,kk
(t, t )Vkd (t ).
(13.93)
These equations solve the problem of an electron in a empty system (and

the conjugate problem of a hole in a lled one). Please note carefully that
the instantaneous self-energy becomes retarded when a closed (dd) Dyson
equation is written; this is similar to what happens in Section 5.1.2 where
the proper self-energy of the localized state is -dependent.
As an example, we can come back to the desorption model of Section
10.2.1. For instance, if at time t = 0, a neutral F atom starts desorbing from
a Fluoride surface, where it is F , we describe it in the hole picture with
n0d = nd (0) = 1, n0k = nk (0) = 0; the problem requires the retarded hole
Greens function.
For simplicity one usually introduces the assumption that the time dependence of Vdk (t) is the same for all k, that is,
Vdk (t) = u(t)Vdk
6
(13.94)
The advanced function parallels the retarded one:
0
ga, (t, t ) = ga,
(t, t ) +

0
d ga, (t, )V ( )ga,
(, t ).
(13.88)
where u(t) is some time-dependent function. Then,

r(d) (t, t ) = i(t, t )u (t)u(t )
|Vdk |2 eik (tt ) .
301
(13.95)
In static conditions (time-independent H), the virtual level is characterized

by a self-energy (), that here Ill denote as S (). So, a Fourier transformation gives
S () =

k

|Vdk |2
S (t) = i(t)
|Vdk |2 eik t ,
k + i
(13.96)
leading to
r(d) (t, t ) = u (t)u(t ) S (t t ).
(13.97)
The dynamics of the virtual levels is factored from the dynamics due to
the time-dependent Hamiltonian. Using (13.83), the Dyson equation for the
retarded function reads
d1
gr,dd(t, t ) = ieit { eit (t t ) +

1
i1
S

d2 e u(1 )u(2 ) (1 2 )gr,dd (2 , t ;
(13.98)
(13.99)
t
the limits of integration are the eect of the functions. By dierentiation

one can check that7
t
i gr,dd (t, t ) = gr,dd (t, t ) + u (t)

d u( ) S (t )gr,dd (, t ). (13.100)
t

t
In the theory of desorption from an ionic crystal, gr,dd (t, t ) is the probability
amplitude that a hole created on the desorbing species at time t is found
there at time t. Let us calculate the probability
in a simple model with
P = gr,dd(, 0)|2
(13.101)
u(t) = et ;
(13.102)
this may be reasonable if the hopping integrals decrease exponentially with

the atom-surface distance and the normal component v of the speed is about
constant. The frequency associated with bond-breaking is vd , where d
o
1 A is the bond length. For any atom desorbing with a kinetic energy of some
eV one nds that h
1 eV is much less than a typical band width.

7
this integro-dierential equation remains valid even if depends on time (see
Ref. [99])
302
For a Lorentzian virtual level, S = 2i is a complex constant, with

the real part a mere shift of the level that we can re-adsorb in the denition
of d . The lifetime width is signicant. Then one can assume8
S (t) = 2i(t)(t).
(13.103)
Thanks to the equation (13.100) become dierential,

i
5
6
gr,dd (t, t ) = 2ie2t gr,dd (t, t ),

t
(13.104)
and its solution is straightforward:

2t
gr,dd (t, 0) = i(t)eit (1e
(13.105)
Thus, the probability P that a neutral F is detected at innity is

P = e
(13.106)
v0
The Lorenzian model predicts a velocity dependence P (v) e v , where v0

is a constant; for v
v0 , P is close to 0. However, P (v) is not analytic as
1.
v 0, and any development in powers of fails despite the fact that
An evident weakness of the Lorentzian approximation is that P in (13.106)
is independent of the position of d with respect to the band continuum, while
the shape of the virtual level depends on it. The solutions of Equation (13.100)
(Ref. [47] ) show that P may be enhanced by orders of magnitude if the level
is near the band edges.
Including the Fermi Level
The general formalism is necessary when the band is partly occupied. f is the
quantity that depends explicitly on the occupation numbers, and the Dyson
equation for = 0 is solved by
f = [1 + gr r ] f 0 [1 + a ga ]
(13.107)
where it is understood that

0
0>
0<
(t, t ) = g
(t, t ) g
(t, t ) = (1 2n0 )ei (tt ) .
f
(13.108)
The equation for f is a matrix equation in the time indices (t,t and intermediate times) and in those denoting spin-orbital states (d, k and intermediate states). Let me make such dependence explicit by the notation change
f [f ] (t, t ), and the like, when necessary; then, Equation (13.107) reads
8
The self-energy must have a factor (t) because it is retarded. However, the
product (t)(t) is ambiguous; here it is understood that (0) = 1 and therefore the
transform is S () = 2i.
f (t, t ) =

303
d1 [1 + gr r ] (t, 1 )
0
d2 f
(1 , 2 ) [1 + a ga ] (2 , t ).
(13.109)
The following device is very useful. Introducing t0 and the trivial

factorization
ei (tt ) = (i)ei (tt0 ) (t t0 )(i)ei (tt ) (t t0 ),

one can write
0
0
0
f
(t, t ) = (1 2n0 )gr,
(t, t0 ga,
(t0 , t ).
(13.110)
When this is inserted into (13.109) the result is

!
"
0
f (t, t ) =
1 2n0
d1 [1 + gr r ] (t, 1 )gr,
(1 , t0 )
0
d2 ga,
(t0 , 2 ) [1 + a ga ] (2 , t ).
(13.111)
Now we recall the Dyson equations [1 + gr r ]gr0 = gr ,[1 + ga a ]ga0 = ga , and

simplify to
!
"
1 2n0 gr, (t, t0 )ga, (t0 , t ).
(13.112)
f (t, t ) =
This compact result completes the general solution of the many-electron problem with time-dependent Hamiltonian which is exact in the absence of interactions. The advanced and retarded are merely single-electron problems,
while the many body eects are accounted for by the equation for f.
Once we know gr , ga and f, since
2g < = g < + g > + g < g > = i(gr ga ) f,
(13.113)
<
(t). Applying Equation(13.113) to the vacuum,
we readily gain nd (t) = gdd
<
g 0 and i(gr ga) f = 0, hence (13.112) yields the rather unusual
expression of a matrix dierence as a matrix product

gr, (t, t0 )ga, (t0 , t ).
(13.114)
i(gr ga ) =
Remarkably, for a noninteracting system gr and ga and do not depend on

the lling, so this expression must hold for any9 n0 . Inserting (13.114) into
(13.113) one gets
9
It is easy to check this for the totally lled system, writing 2g > = i(gr ga )+f =
0 and using (13.112) again.
304
<
2g
=
gr, (t, t0 )ga, (t0 , t ) f ,
that is,
<
g
(t, t ) =
n0 gr, (t, t0 )ga, (t0 , t ).
(13.115)
To solve the atom-surface scattering problem we set = = d,

nd (t) =
n0 gr,d (t, t0 )ga,d (t0 , t) =
n0 |gr,d (t, t0 )|2 .
(13.116)
If an atom with an occupied d level scatters against the surface and the k
states of interest are empty, we shall consider n0d = 1, n0k = 0; then this is
essentially a one-particle problem and the result says nd (t) = |gr,dd (t, t0 )|2 . In
the case of F desorption from an alkali uoride, approximately n0d = 1, n0k = 0
in the hole picture. the many-body complications arise in the study of the
ionization probability of a positive ion hitting a metal surface. In this case,
n0d = 0 and

n0k |gr,dk (t, t0 |2 ,
(13.117)
nd (t) =
k
n0k
where
is the Fermi distribution. We can write the result in terms of the
local component gr,dd by using the appropriate component of the Dyson equation, namely,

0
gr,dk (t, t0 =
d gr,dd (t, )Vdk ( )gr,kk
(, t0 )

d gr,dd (t, )Vdk ( )eik ;
(13.118)
= ieik t
t0
this is convenient as gr,dd can be determined by a closed equation in terms

(d)
of r .
A.Blandin, A.Nourtier and D.W.Hone[48] developed the theory for an
atom which is initially in equilibrium with a metal surface with Fermi energy
EF and is suddenly sputtered away with speed v; they assumed a Lorentzian
virtual level. The deviation from the adiabatic (v 0) result nd () =
v0
(EF ) is still of the singular form e v , with v0 a characteristic speed. This
looks like the one-body result (13.106), but should change when including a
less pathological virtual level shape.
13.6.2 Quantum Transport
Partitioned Static Transport Theories
The earliest quantum theory of transport through a junction which is still
used today after 50 years is the pioneering one due to Landauer [61]. I It deals
305
with non-interacting electrons and allows to calculate the current-voltage

characteristics, that is, J(V ), where J is the stationary (d.c.) current due to
a static potential dierence V. Consider a system S between two electrodes;
we let the L and R subscripts or superscripts refer to left and right leads, e.g.
(L) (), (R) () are the respective densities of states, and VmR , VmL the hopping
matrix elements between the lead states at energy and the eigenstate m
of S. (We shall omit these labels when unnecessary). In the presence of a
bias, the chemical potentials L , R dier, and therefore there are two Fermi
distributions fL , fR . The Landauer formula for the current is

e
tm,n tn,m
(13.119)
J=
d[fL () fR ()]
h
m,n
where m, n label eigenstates of the isolated S and the transmission coecients
are

(r)
().
tm,n = 2 (L) ()(R) ()VmR VnL gm,n
Phenomenological approaches, like the one due to Beenakker, [69] based
on a master equation, are very useful, particularly when dealing with strongly
correlated systems, like for instance single-electron transistors and related
devices. For an application to superconducting systems see for example [70]
where the main issue is the Coulomb blockade pattern of two-electron tunneling in the framework of W=0 pairing (Chapter 17).
The application of the Keldysh formalism to the calculation of stationary tunneling currents was pioneered by C. Caroli,R. Combescot, P. Nozieres
and D.Saint-James in a series of fundamental joint papers [62]. They considered a 1d tight-binding model for a metal-insulator-metal (MIM) junction.
The main feature of their approach is the use of an ideal partition which
enables them to reach the stationary state adiabatically: to zeroth order,
the applied bias is already there, but J = 0 since left and right subsystems
are disconnected. The perturbation to be accounted for to all orders is a
static pseudo-Hamiltonian that establishes the connection. In principle, the
pseudo-Hamiltonian can be calculated ab initio. It is switched on adiabatically in the presence of the bias and a stationary state is obtained in which
a time-independent current ows. The theory was originally formulated in a
tight-binding model and the expression for the current was established from
the continuity equation and written in terms of Greens functions:

eT d <
eT
<
(c ca ca c ) =
{g () ga
()},
(13.120)
J=
i
h
h 2 a
where a, stand for tho sites connected by the hopping T . The theory was
then extended to a continuous formulation. In both cases, the current-voltage
characteristics were obtained in terms of local densities of states calculated
on both sides of the partition. However, the left and right unperturbed subsystems cannot exist in insulation because the zero-order wave functions obey
306
special boundary conditions at the idealized partition. The partitioned theory of transport through junctions was put on a more general foundation
by Feuchtwang[85], who obtained the current-voltage characteristics of onedimensional model systems, rigorous within the Schrodinger theory in the
non-interacting limit; he also used the Keldysh formalism but questioned
some technical aspects of the previous formulations.
The partitioned approaches manage to work out the theory without introducing the time by an artice. One considers that the left and right halves
of the system are already biased while an ideal partition prevents the electrical continuity. The adiabatic removal of the partition eventually allows a
stationary current to ow. Thus, the partitioned approach describes a situation that is quite dierent form the experimental one, when the connection is
there when the bias is applied. The static treatments were based on a crucial
assumption of equivalence with the physical situation; this assumption can
fail as we know today (see below).
Partition-free Dynamical Transport Theory
A time-dependent, partition-free framework was proposed in 1980 by the
writer[63]; the new scheme was found to be powerful to calculate the currentvoltage characteristics, but above all for the rst time allowed to obtain transient responses as well. The dynamical theory using the Keldysh formalism
and the equation of motion method is no harder than the static one and
yields much more information; it has not received much attention in the past
but it is gaining ground in recent years for its clear advantages. One crucial
point in favour of my approach is that it is partition-free: the whole system including the leads is in thermal equilibrium at some temperature until
the bias is applied. This is physically much more appealing, and reects the
experimental situation. I worked out the theory in parallel for discrete and
continuous models using time-ordered Greens functions and Equation (4.24)
for the current. In the discrete one-dimensional case, when the unperturbed
Hamiltonian is of the form
H0 =
cs (cs+1 + cs1 )
(13.121)
s=
the perturbation, which is dealt with exactly,is

H1 = (t)
Vs (t)cs cs
(13.122)
s=
and the current owing through site s in the 1d electrodes is

Js =
5 T
6
eTs
T
lim gs,s+1
(t, t ) gs,s1
(t, t ); ,

h t t+0
(13.123)
307
this is physically equivalent if not identical to (13.120). where Ts is the hopping matrix element between site s and its neighbors. For the continuous
model,in obvious notation,

H0 =
k ck ck ,
(13.124)
k
H1 (t) =
(t)Vk,k (t)ck ck ,
(13.125)
k,k
and the current density was obtained by Equation (4.24). As we saw below
Equation (10.16), the equation of motion
i

T
T

T

gkk (t, t ) = kk (t t ) + k gkk
Vkp (t)gpk
(t, t ) +
(t, t ),
t
p
(13.126)
T
T
with the initial conditions gkk
(0, 0 ) = ikk (1 fk ), gkk (0, 0+ ) = ikk fk
T
T
( fk is the Fermi function) and gkk (t, t + 0+ ) gkk (t, t + 0 ) = ikk , are
satised by

(r)
(a)
T

gkk
fq gkq (t, )gqk (, t )
(t, t ) = i(t t)
i(t t )
q
(r)
(a)
(1 fq )gkq (t, )gqk (, t )
(13.127)
provided that is earlier than t and t. This is essentially the Blandin et al.
solution[48] with a much simpler derivation. Here we need g T (t, t ) with t > 0,
r,a
r,a
and t just large than t, so we may set = 0. Writing gkk
(t) = gkk (t, 0), one
nds

e
h
(r)
(r)
Im
J (x, t) =
gx,q
(t) gx,q
(t).
(13.128)
m
q
The corresponding solution for the current owing at site {i, j, k} in a lead
oriented along the x axis in a 3d discrete model is
Jijk =
2eT
(r)
r
#
fq gijk,q (t)gi1jk,q
(t).
h
(13.129)
The solution is valid for any time dependence of V , but in order to make
the results more explicitly in Reference [63] I assumed H1 (t) = (t)V (x),,
with V (x) 0 deep in the left electrode and V (x) constant deep in the
right one. Now the retarded and advanced Greens functions can be computed
simply using a constant nal-state Hamiltonian.
In the non-interacting case, the current-voltage characteristics are obtained by taking the t limit by asymptotic methods[63] [92] and agree
with the results by Feuchwang[85]. It turns out that the current response is
308
very indirectly related to the local Greens functions and is much more naturally written in terms of the asymptotic behavior of the wave function at long
distance in the free-electron wires. For small enough V , both the continuous
and the discrete models become ohmic and
J =
eV
.
(13.130)
It was shown in Ref. [63] and proved formally by Stefanucci and Ambladh [103] that for the non-interacting system the current-voltage characteristics obtained by the partitioned approach must coincide generally with
those of the partition-free approach; moreover, these Authors considered including correlation eects within the Time Dependent Density Functional
Theory (TDLDA) scheme. In strongly correlated systems and in the presence of bound states the proof does not apply and the current can oscillate
for long times, as in the Josephson eect . Examples are discussed in Ref.
[192] where Hubbard clusters produce Josephson supercurrents by the W=0
mechanism (Chapter 17). A new implementation of the time-dependent quantum transport theory within the TDLDA scheme has proposed quite recently
([102]).
Meir-WinGreen Weak-Coupling Formula
In 1992, Yigal Meir and Ned S. WinGreen [64] proposed a Laudauer formula
for the current through an interacting electron region. The system S is placed
between two non-interacting electrodes characterized by
<
>
() = 2i f () ( k ) , gk,k
() = 2i [1 f ()] ( k ) .
gk,k
(13.131)
They started from

e
d L <
L <
J=
VkL ,n gn,kL () Vk,n
gkL ,n () .
h
2
(13.132)
kL ,n
Here, gk<L ,n links the k state in the left electrode to the S state annihilated
by dn , and once integrated in yields ckL dn . Let us calculate the current

which enters from L. One can write down these mixed Greens functions in
terms of those based in S through

<
T
<
T
Vk,m gk,k
() gm,n
() gk,k
() gm,n
() ,
m

<
T
T
<
gk,k
= Vk,m
() gn,m
() gk,k
() gn,m
() .
<
=
gk,n
<
gn,k
(13.133)
Substituting in (13.132), we are free to exchange m and n in the second term;

in this way, we can collect terms and get a summand proportional to
309
/
! T
"
! T
"0
<
T
<
T
;
gk,k
Vk,n Vk,m
gn,m + gn,m
gn,m
gk,k + gk,k
T
T
<
>
now we replace gn,m
+ gn,m
by gn,m
+ gn,m
, simplify according to
!
"
!
"
g< g> + g< g< g> + g< = g<g> g> g< ,

and use g > = g < + g (r) g (a) . Next, we perform the sum with k = d()
and introduce
m,n = 2
d()Va,n Va,m
,
(13.134)
a

where a adds some generality by summing over possible independent channels in the L electrode, like for instance angular momentum or spin.
J=
ie
h

d
/
0
5 r
6
(L)
a
<
m,n () f (L) () gn,m
gn,m
.
+ gn,m
(13.135)
m,n
The same formula with R in place of L will give the current entering from
the right electrode, which is the opposite in stationary conditions; thus one
can symmetrize the expression with
(L)
(L) (R) (L) (L)

(L) f (L) (R) f (R)
, f
.
2
2
(L)
(R)
In the special case m,n (E) = m,n (E), that is when the couplings are
the same apart from a constant, the formula simplies to (see Problem 15.1)
J=
ie
h
d[f (L) () f (R) ()]
(r)
(a)
m,n [gm,n
gm,n
].
(13.136)
m,n
where
=
Since
(L) (R)
.
+ (R)
(r)
(a)
m,n [gm,n
gm,n
] = 2iIm
m,n
(13.137)
(L)
(r)
m,n gm,n
,
m,n
this expression has a strong intuitive appeal if m,n can be taken real: contributions to the current arise from that slice of the density of states which
is occupied at the one side of the junction and empty at of the other
side. The density of states must be computed in the presence of the connections; however, the theory of [64] is particularly suited for small , when the
coupling to the leads is so weak that one can approximate g as the Greens
function of the isolated S.
Problems
Part IV
Non-Perturbative Approaches and

Applications
14 Some Recursion Techniques with

Applications
14.1 Lanczos-Haydock Recursion

14.1.1 Local Greens Function for a Chain
Any stationary quantum problem, with any number of degrees of freedom,
can be mapped into a solvable linear chain one-body tight-binding model
exactly. From an abstract mathematical viewpoint, the method for putting
a symmetric matrix in three-diagonal form which is readily diagonalized was
invented by Lanczos[58] in 1950. This is enough to calculate the Greens function of systems with a nite number of states, and the algorithm is powerful
and convenient even for large matrices provided that they are sparse, that is,
most elements vanish. Haydock and coworkers[59] extended the method and
developed applications to physical systems with many degrees of freedom.
Let un , n = 0, 1, 2 . . . be the site orbitals of a semi-innite chain, with local
levels an . The Hamiltonian is dened by
Hun = an un + bn+1 un+1 + bn un1 ,
where bs are hopping integrals, and its matrix is three-diagonal:
a0 b 1 0 0 . . .
b 1 a1 b 2 0 . . .
H=
0 b 2 a2 b 3 0 .
0 0 b 3 a3 b 4
... ... ... ... ...
(14.1)
(14.2)
For this model, we want the local retarded Greens function is

G0 (E) = u0 |
1
|u0 ,
EH
which is the 00 element of
E a0 b1
0
0
b1 E a1 b2
0
b2 E a2 b3
(E H)1 =
0
0
0
b3 E a3
...
...
...
...
(14.3)
1
...
...
0
.
b4
...
(14.4)
314
14 Some Recursion Techniques with Applications
1
Now, zH
with z = E + i, = +0 is the Fourier transform of eiHt (t),
hence G0 (z) is causal; moreover it vanishes for z in all directions,
G0 (z) = G0 (z ) and is a Herglotz function, i.e. yields a non-negative density
of states n(E) = 1 ImG0 (z = E) 0. According to the standard rule, the
00 element is given by
D1 (E)
G0 (E) =
,
(14.5)
D0 (E)
where the denominator is det(E H) and the numerator is obtained from it

by cropping the rst row and column,
E a0 b1
0
0
...
b1 E a1 b2
0
...
b2 E a2 b3
0
D0 = Det 0
,
0
0
b3 E a3 b4
...
...
...
... ...
0
0
...
E a1 b2
b2 E a2 b3
0
...
0
b3 E a3 b4
D1 = Det 0
(14.6)
.
0
0
b4 E a4 b5
...
...
...
... ...
Both are secular determinants of matrices representing chain Hamiltonians.
Similarly, we dene Dn as obtained by a crop of the rst n rows and columns;
it refers to a shortened chain with the rst n sites removed. Later, we shall
also consider shortening the chain the other way, dening N = D0 and
n , n < N the shortened chain with the rst n sites left. Now expand D0
by Laplaces rule along the rst line:
b1 b2
0
0 ...
0 E a2 b3
0 ...
.
D0 = (E a0 )D1 (b1 )Det
(14.7)
0
b3 E a3 b4 . . .
... ...
... ... ...
The b2 term in the rst line does not contribute, and
D0 (E) = (E a0 )D1 (E) b21 D2 (E)
which inserted into Eq.(14.5) yields
G0 (E) =
D1
D1
1
=
=
.
2
D0
(E a0 )D1 b21 D2
(E a0 ) b21 D
D1
Then,
D2
1
=
3
D1
(E a1 ) b22 D
D2
and iterating,
G0 (E) =
1
E a0
b21
Ea1
315
(14.8)
b2
2
b2
3
Ea2
Ea3 ...
For the uniform chain, with an 0 and bn b, one gets

G0 (E) =
1
.
E b2 G0 (E)
(14.9)
This yields the semi-elliptic density of states
(4b2 E 2 ) 4b2 E 2
n(E) =
.
2b2
14.1.2 Greens Function of any System
Consider an arbitrary Hamiltonian H, possibly representing a complicated
many-body system, and any normalized state u0 ; if this is an eigenstate we
are nished, otherwise we may always write :
H|u0 = a0 |u0 + b1 |u1 , u0 |u1 = 0, u1 |u1 = 1.
(14.10)
This is tantamount to call b1 |u1 the part of H|u0 which is orthogonal to

|u0 , and implies that a0 = u0 |H|u0 , and b1 = u1 |H|u0 ; since
u1 =
1
(H a0 )|u0
b1
(14.11)
is normalized, we get (up to an arbitrary phase factor)

b1 = (H a0 )|u0 ,
(14.12)
where v denotes the norm of v. This can be computed from the given H
and u0 ; now we also know u1 and we can proceed with
H|u1 = a1 |u1 + b1 |u0 + b2 |u2 , u0 |u2 = 0, u1 |u2 = 0.
(14.13)
Here, b1 is the same as above since H is Hermitean and u2 is taken to be normalized. Thus, a1 = u1 |H|u1 , and b2 |u2 = (H a1 )|u1 b1 |u0 . Imposing
that u2 is indeed normalized, we obtain, since the phase of b2 can be chosen
at will,
(14.14)
b2 = (H a1 )|u1 b1 |u0 ;
again, we can compute that, write
u2 =
and expect the next step
1
[(H a1 )|u1 b1 |u0 ]
b2
(14.15)
316
H|u2 = a2 |u2 + c|u0 + b2 |u1 + b3 |u3

This might appear a boring way to get equations of increasing complexity,
but here the clever idea rst comes into play, since H|u2 does not contain
anything like c|u0 . This is because c = u0 |H|u2 , but Eq. (14.10) shows
that H|u0 is orthogonal to all un with n > 1. Hence, continuing to introduce
orthogonal un states, we always get three-term recursion relations
H|un = an |un + bn |un1 + bn+1 |un+1
(14.16)
with
an = un |H|un
(14.17)
bn+1 = (H an )|un bn |un1

1
|un+1 =
[(H an )|un bn |un1 ] .
bn+1
(14.18)
(14.19)
Hence, we have a general algorithm to nd the local Greens function of any

state u0 , mapping it to site 0 of a chain of un that can be generated stepwise,
with their levels an and hoppings to their nearest neighbors. The chain is
semi-innite, that is does not end, unless the system has a nite Hilbert
space or the choice of u0 was extremely clever.
14.1.3 Terminator
Let N be the number of sites that we can aord to compute; the N -th denominator of the continued fraction ( 14.8) reads E aN 1 b2N / . . . and one gets
an approximation by setting bN = 0. By truncating the continued fraction in
this way we get a discrete spectrum; this can then be broadened by shifting
the poles a bit below the real axis with E z = E + i, which converts the
functions into Lorentzians. As an illustration, consider the rectangular level
with band width W = 2, n(E) = (2 E 2 )/2, which we met in Section
6.2 and arises from

1
z+
G(z) =
log
.
(14.20)
2
z
We shall see in Sect. (14.1.4) how to calculate the a and b coecients from
G. The comparison with the N = 5, 10 and 40 continued fractions (Figure
14.1) shows that such an articial procedure, while allowing a comparison of
some qualitative kind, is too rough for many purposes; it is mathematically
pathological, e.g. the second and all the even moments of the approximate
n(E) diverge.
As an alternative to the Lorentzian broadening, one can compute the
integrated local dos
E
n(E)dE,
N (E) =
317
which is a bar diagram but converges to the exact result in norm. One can
smooth it by tting with a continuous function and then dierentiate to nd
an approximate n(E); the results are much better than before, and at least for
one-electron problems converge quickly almost everywhere. Interestingly, for
regular solids there are always special intervals centered at E values where the
approximation fails badly; these singular values, where n is a bad function, are
dn
is discontinuous.
band-edge singularities or Van Hove singularities , where dE
If one is interested in the overall line shape of some spectroscopy, this may
be of minor importance, since such singularities tend to be masked somewhat
by the experimental broadenings. However there is a message there: the longtime behavior of the Fourier transform of any function of frequency depends
[92] precisely on such details, and the transform is a correlation function,
which can be observed in appropriate experiments. We know from Section
0.2
0.2
0.2
Fig. 14.1. The rectangular n(E) with W = 0.5 and a broadening = 0.05 (dashed)
compared with the continued fraction approximations using N=5 denominators
(left), N=10 denominators (centre), N=40 (right). The continued fraction coecient
are discussed in the next Sections.
4.3.1 that the band-edge singularities depend only on dimensionality; some

weaker singularities inside the band depend on the lattice symmetry, too. The
Van-Hove singularities are universal features, independent of the details of
the Hamiltonian. Stubborn computing fails, but one can gain them directly
from a simpler model in the same universality class. We do not need the model
to be realistic, but to have the correct singularities at the correct places. Such
a model is called terminator. The local Greens function of the terminator is
exactly known and we can calculate, say, N denominators of the continued
fraction expansion, using Greek letters for the coecients we write
318
0 (E) =
1
E 0
12
E1
E2
(14.21)
2
2
2
3
E3 .........
At level N the full continued fraction truncated ends with

E N 1
2
N
1
2 t (E) .
E N 1 N
N
Here, tN (E) denes the tail of the continued fraction. The idea is: pick the
tail and attach it to G. Doing this by brute force is not rewarding, while
deepening the mathematics is a far better idea.
Pade Representation
Orthogonal Polynomials
It is evident that the continued fraction truncated at level N is the ratio of
two polynomials. To nd them, consider the linear chain eigenstates,
N
fn() |n
(14.22)
fn fn = , ,
(14.23)
| =
n=0
that satisfy
N

n=0
and the Schr

odinger equation
N
fn() [an |n + bn+1 |n + 1 + bn |n 1 ]
n=0
= E ()
fn() |n , b0 = 0.
(14.24)
n=0
N
N
fn bn+1 |n + 1 =
n=0 fn1 bn |n , and
n=0 fn bn |n 1 =
f
b
|n ,
(f
=
0),
one
nds
n+1
n+1
N
+1
n=0
Since
N
n=0

()
()
(an E () )fn() + bn+1 fn+1 + bn fn1 |n = 0.
(14.25)
n=0
This implies the recurrence relations (r.r.)

()
()
(an E () )fn() + bn+1 fn+1 + bn fn1 = 0,
()
f1 = 0.
(14.26)
319
f0 0| does not vanish (otherwise fn 0 ) and we may dene N independent functions
f
Pn (E () ) = n ,
(14.27)
f0
having the same r.r.
(an E () )Pn + bn+1 Pn+1 + bn Pn1 = 0,
P1=0 .
(14.28)
but P0 (E) = 1; all the others are determined by the r.r. and depend on the
a, b coecients of the previous sites. Evidently, they are just polynomials in
E. Changing the sum
N

fm
fn = m,n
(14.29)
to an energy integral,
m,n =
N

dE(E E () )fm
fn
(14.30)
and using (14.27) we nd

m,n =
dE
N
0| |0 (E E () )Pm

(E)Pn (E)
(14.31)
that is, the P are orthogonal polynomials

m,n =
dE n(E)Pm
(E)Pn (E)
(14.32)
with a weight n(E), the local density of states at |0 (compare with Section
5.1.4).
Polynomials, Secular Determinants and G
The secular determinant N (E) consists of the rst N rows and columns of
D0 (Equation 14.6)
0
0
...
E a0 b1
b1 E a1 b2
0
...
0
b2 E a2 b3
N = Det 0
;
(14.33)
0
0
b3 E a3 b4
...
...
...
. . . . . . N N
let us consider then the succession of determinants n (E) and expand
n+1 (E) using the last row:
320
n+1
... ...
...
...
...
. . . E a3 bn2
0
0
0
= Det . . . bn2 E an2 bn1
...
0
bn1 E an1 bn
...
0
0
bn
E an n+1n+1
... ...
...
...
. . . E a3 bn2
0
= (E an )n + bn Det
;(14.34)
. . . bn2 E an2 0
...
0
bn1 bn nn
and we are left with

n+1 (E) = (E an )n (E) b2n n1 (E).

Multiplying (14.28) by ni=1 bn we nd (14.35) with
n (E) = Pn (E)
n
bn .
(14.35)
(14.36)
i=1
Thus, we have got D0 = N (E) for the N -site chain in terms of the Pn and
in view of (14.5) a new expression for G0 is within reach if we can also express
D1 (E). However, D1 (E) is the counterpart of D0 (E) in a chain where site 0
has been cut o. The polynomials of that chain satisfy the same r.r.(14.28)
but they vanish on site 0. We can specify them as follows:
Q0 = 0, Q1 (E) = 1, (an E)Qn + bn+1 Qn+1 + bn Qn1 = 0.
(14.37)
Pn and Qn are the independent solutions for the N site chain that correspond
to dierent initial conditions. Therefore,
D1 (E) = QN (E)
N
bn .
(14.38)
i=2
Finally, we get for the N site chain

(N )
G0 (E) =
QN (E)
.
b1 PN (E)
(14.39)
Such a representation as a ratio of polynomials is exact here, but is popular

as a way to approximate functions and is called Pade approximant.
How to Append the Terminator s Tail to G.
We compute G up to the N -th denominator, and continue the continued
fraction with tN (E) taken from the terminator. That will be much better
than nothing, and will bring much of the missing information. We need to
321
pick up tN (E) from the terminator continued fraction, but we are not so
silly to use it directly. For the truncated terminator, we write, in place of
(14.39,14.28,14.37)
N (E)
(N )
0 (E) =
.
(14.40)
1 N (E)
where s are the counterparts of the Qs in the terminators chain; the
polynomials,
(n E () )n + n+1 n+1 + n n1 = 0,
1=0 .
0 = 0, 1 (E) = 1, (n E)n + n+1 n+1 + n n1 = 0.
(14.41)
(14.42)
The rule for inserting the tail is:

2
tN (E)
N 1 N 1 + N
(14.43)
In order to use (14.43) in (14.40,) we need to show up N 1 , so we obtain

n+1 and n+1 from (14.41,14.42), then set n + 1 = N and obtain:
(N )
1 E N 1 N 1 (E) N 1 N 2 (E)
.
1 E N 1 N 1 (E) N 1 N 2 (E)
(14.44)
Now, using (14.43), the full terminator reads:
0 (E) =
2
tN (E)]N 1 (E) N 1 N 2 (E)
1 E [N 1 + N
.
2 t (E)]
1 E [N 1 + N
N
N 1 (E) N 1 N 2 (E)
(14.45)
We can clean it up using (14.41,14.42) again to eliminate the N 2 items:

0 =
1 N N tN (E)N 1
.
1 N N tN (E)N 1
(14.46)
Finally, this is a compact tail indeed:

tN (E) =
N 1 0 N
1
.
N 1 N 1 0 N 1
(14.47)
The transplantation is obtained by replacing Greek by Latin letters in (14.46):

G0 (E) =
1 QN bN tN (E)QN 1
.
b1 PN bN tN (E)PN 1
(14.48)
This method can give excellent results already with N a few tens if the
terminator has the correct edges of the continua with the right edge and Van
Hove singularities.
322
14.1.4 Moments
The above method is in a way a rened version of the moments method [60],
which is itself worth considering for its conceptual and practical importance.
Let |0 be any state of the system, and
1
n() = 0|( H)|0 = ImG()
(14.49)
the local density of states. Even for complicated H we can obtain useful
results by computing the moments

n
n = 0|H |0 =
d n n();
(14.50)
if we could obtain them all and sum an exponential series we would build
the Fourier transform of (14.49). Now, 0 = 1, 1 is the center-of-mass of
the virtual level, 2 informs us about its width, 3 characterizes its skewness, and with increasing n ner details are revealed. The moments allow
computing
the Greens function: if Re z is larger than the eigenvalues of H,

k
G(z) =
k=0 z k+1 . The short time behavior of n(t) depends on the rst
moments, that we can compute; the long-time n(t)is often needed and even
the asymptotic trend is of interest, but that information must be sought elsewhere. Letting N denote the number of moments that we can compute, we
face the problem of using them in the best way. One possibility would be
choosing a functional form which depends on N parameters and imposing the
values of N moments. For instance, if the edges of the continuum are known
one can pick a polynomial or an expansion in Tchebychev polynomials. However, the Van Hove singularities will prevent the uniform convergence of the
procedure. A more powerful method exists. By the M + 1-site chain Hamilp
)00 by
tionian1 HM of the form (14.2) one computes the moments p = (HM
matrix multiplication; from p , p = 1, N one can deduce the ak and bk
coecients for k N 1. In fact, 1 = a0 ; inserting into 2 = a20 + b21 we
have one unknown and solve immediately obtaining b21 = 2 21 . The third
moment is
3 = a30 + (2a0 + a1 )b21 ;
inserting the previous results we obtain
a1 =
31 + 21 2 3
.
21 2
We can continue in this way as long as we please:

4 = (a0 + a1 )2 b21 + (a20 + b21 )2 + b21 b22 ;
1
The N moments obtained by matrix multiplication using HM are obviously
exact if M is large enough.
14.2 Spin-Disentangled Diagonalization
323
again we have only one unknown and nd

b22 =
32 + 21 2 3 23 21 4 + 2 4
.
(21 2 )2
General explicit formulas are also available from the theory of Pade approximants [157].
In this way we solve the inverse of Haydocks problem: given a normalized
Herglotz function n(), to determine a semi-innite chain Hamiltonian such
that n() is its density of states. For a symmetric n(), like those arising from
bipartite lattices (Section 4.3.1 ) we may take ak 0. For the rectangular
band model (14.20) one readily determines the coecients (Problem 14.2);
in this way I performed the comparison between the truncated fraction and
the exact density of states following Equation(14.20).

If the Hamiltonian matrix H is too large to store, and is not assumed to
be sparse2, we may still be able to nd eigenvalues and eigenvectors by the
recently introduced Spin-Disentangled technique [18]. I nd that it is fast
and easy to use and works very well with the Lanczos-Haydock algorithm
discussed above. Let M be the number of electrons of spin , and M +
M = N their total number. Consider a real orthonormal basis {| }, =
1 m , of dimension m , for the M electrons of spin . The dimension
of the Hilbert space is NH = m m and H is a NH NH matrix; now I
show how to solve by matrices no larger than the largest between m m
and m m . For m = m , a 1, 000, 000 1, 000, 000 problem is solved by
1, 000 1, 000 matrices. One can write the ground state wave function in the
interesting form
| =
L | |
(14.51)
which shows how the and congurations are entangled. The particles of
one spin are treated as the bath for those of the opposite spin: this form also
enters the proof of a famous theorem by Lieb[37] discussed in Section 18.6.
In Equation (14.51) L is a m m rectangular matrix.
Schr
odinger Equation
Let K denote the kinetic energy m m square matrix in the basis {| },
()
()
and ns the spin- occupation number matrix at site s in the same basis (ns
2
If it is sparse, one can store and manipulate the nonzero elements Hij together
with i and j; this procedure is annoying and leads to a slowing down of the computations. Moreover, in many problems of interest H is not sparse.
324
is a symmetric matrix since the | s are real). The Hamiltonian (in a

lattice model, or the realistic Hamiltonian
on a discrete grid) is H = H0 + U ,

with
H
=
K
+
K
,
while
U
=
U
(s, s )ns ns . Using (K ) =

0
s,s

(K ) and the like, one nds that
H0 | =
L [| | (K ) + | | (K ) ]
(K L) | | +
5
(LKT ) | |
6
T
K L + LK
| |
(14.52)
where T stands for Transpose. Since ns | =

U | =
U (s, s )
s,s
| |ns | ,
L, | | (ns ) (ns ) ;
(14.53)
, ,
but (ns ) = (ns ) and one can write

U | =
U (s, s )
s,s
(ns Lns ), | |
(ns Lns ), | | .
(14.54)
Recalling Equation (14.51) we see how H acts:

H : L K L + LKT +
U (s, s )ns Lns .
(14.55)
s,s ,
s,s
For illustration, below I adopt the on-site Hubbard interaction, when this
reduces to Liebs rule [37]

()
n()
(14.56)
H[L] = [K L + LKT ] + U
s Lns .
s
In the absence of magnetic elds, KT can be taken real and symmetric, hence
we drop the transposition sign. In particular in the Sz = 0 sector (M = M )
()
()
and K = K , ns = ns . The action of H is obtained in a spin-disentangled
way, in terms of operators acting in the spin subspaces. The generality of
the method is not spoiled by the fact that it is fastest in the Sz = 0 sector,
because it is useful provided that the spins are not totally lined up; on the
other hand, Sz = 0 can always be assumed, as long as the hamiltonian is
SU (2) invariant. For example, consider the Hubbard Model with 2 sites a
and b and 2 electrons (H2 molecule, Section 1.2.5). The intersite hopping
325
is t and the on-site repulsion U . In the standard method, one sets up basis
vectors for the Sz = 0 sector
|v1 >= |a a >, |v2 >= |a b >,
|v3 >= |b a >, |v4 >= |b b > .
One then looks for eigenstates (three singlets and one triplet)
=
4
i |vi >
(14.57)
i=1
using the Hamiltonian
U
t
H=
t
0
t t
00
00
t t
0
t
.
t
U
(14.58)
We can do with 2
2 (rather
bythe spin-disentangled

than 4 4)
matrices
10
00
0 t
1 2
, nb =
.
,K =
, na =
method with L =
00
01
3 4
t0
By (14.56),
a,b
a,b

(H[L]) | |
(14.59)
H| =

U 1 + t(2 + 3 )
t(1 + 4 )
.
t(1 + 4 )
U 4 + t(2 + 3 )
The reader can readily verify that this is the same as applying H in the form
(14.58) to the standard wave function (14.57) and then casting the result
in the form (14.51). Since we can apply H we can also diagonalize it. By
this device, we can work with matrices
whose dimensions is the square root
of those of the Hilbert space: NH NH matrices solve the NH NH

problem.
with (H[L]) =
The Practical Numerical Recipe

Starting from a trial wave function of the form (14.51), preferably having the
proper symmetry, one can avoid computing the Hamiltonian matrix, since its
operation is given by Eq. (14.56). Each new application of the Hamiltonian
takes us to a new Lanczos site and we can proceed by generating a Lanczos
chain. To this end we need to orthogonalize to the previous sites by the scalar
product
(14.60)
1 |2 = Tr(L1 L2 ).
In this way one can put the Hamiltonian matrix in a tri-diagonal form. This
method is better suited if one is mainly interested in the low-lying part of the
spectrum. A severe numerical instability sets in when the chain is longer than
a few tens of sites, i.e. well before the Lanczos method converges. On several
occasions, I found it preferable to use repeated two-site chains alternated
with moderate-size ones.
326
14.3 Method of Excitation Amplitudes

In 1977 I found the exact solution [77] of a model introduced by Hewson and
Newns [79] to discuss photoemission spectroscopy from valence states3 ; the
same idea can solve exactly a class of models and later I published many
extensions and applications to photoemission and Auger spectroscopy[124]
and non-linear optics4 . The simplest Hewson-Newns model is a Fano model
interacting with a boson degree of freedom. Consider a spin-less Fano model,
(Chapter 5),

hF =r nr+
k nk +
{Vkr ck cr +h.c.},
k
where as usual nr is the occupation number operator of some resonant orbital,

while the index k runs over the electron continuum; let
Hv =0 d d,
(14.61)
be some harmonic oscillator (vibration or or plasmon). The Hewson-Newns

[79] model Hamiltonian reads
H = hF + Hv + Hev ,
(14.62)
where a linear coupling is assumes between the oscillator and the resonant
state:
Hev = g(d + d )nr .
(14.63)
The electron can be captured by the resonance and excite the plasmons, but if
eventually the resonance decays, the interaction with the plasmons is turned
o, otherwise a permanent shift of the oscillator coordinate may result. The
interaction can be very strong if the boson frequency resonates with dressed
electronic excitations. The purpose of the method is to calculate exactly the
the excitation amplitudes
v (E) = 0| dv cr
1
c+ |0
E H + i r
(14.64)
where |0 is the electron and plasmon vacuum. The amplitude of exciting v

plasmons is Av (E) = 1v! v (E) . The identity (4.92)
3
This approach should not be confused with other useful recursion methods,
which are completely dierent, despite the fact that they lead to continued fraction solutions. Among the most important examples, I recall the Mori generalized
Langevin Equation Method [208] and the Lee solution technique [209]; for a review
see Ref. [210]. However, quite recently, I became aware of a related solution of a
boson-boson model had been put forth previously by Sumi[100] developing a theory
of exciton polaritons.
4
For applications to many-photon eects, see Reference [78] and next Chapter.
14.3 Method of Excitation Amplitudes
327
1
1
1
1
=
+
Hev
EH + i Eh + i Eh + i
EH + i
yields
1
c |0 +
E h + i r
1
1
0| dv cr
Hev
c |0 =
E h + i
E H + i r
1
1
Hev
c |0
v0 G0rr (E) + 0| dv cr
E h + i
E H + i r
v (E) = 0| dv cr
(14.65)
where G0rr (E) is the retarded Greens function for g=0. In other terms, this
is the solution of the Fano problem,
G0rr (E) =< 0|cr
1
cr |0 >.
Eh + i
(14.66)
One veries by induction that

dv hn = (h + v0 )n dv
(14.67)
with the consequence that

dv
1
1
=
dv .
E h + i
E h v0 + i
(14.68)
Thus,
v (E) = v0 G0rr (E) + 0| cr
1
1
dv Hev
c |0 (14.69)
E h v0 + i
E H + i r
!
"
1
1
c gdv d + d cr
c |0 .
E h v0 + i r
E H + i r
(14.70)
Now note that, by reading the matrix element from right to left, the electron
is created, annihilated, created and annihilated again, thus the rst annihilation produces the electron vacuum; moreover, on the left of dv the only
boson operators are those contained in h. Therefore, the matrix element gets
factored.
= v0 G0rr (E) + 0| cr
1
c |0
v (E) = v0 G0rr (E) + g 0| cr
E h v0 + i r

1
0| dv d + d cr
cr |0 .
E H + i
Since
(14.71)
328
! "n
! "n1 ! + "n
d d+ = n d+
+ d
d,
(14.72)
0|d d = 0|d1
(14.73)
one gets
and nds the recursion relations

v (E) = v0 G0rr (E) + gG (E v0 ) {v+1 (E) + vv1 (E)} .
(14.74)
To solve this system, it is convenient to dene
(E)
1
E
-1
Fig. 14.2. The interaction of a slightly broadened ( = 0.01 ) rectangular level of

Equation (14.20)(dotted line, width=1 , arbitrary units) with a plasmon mode of
frequency p = 1.3 with coupling g=0.5, according to Equation(14.77). Note that
the local density of states is deformed by interaction with the plasmon mode, while
the plasmon satellites are not identical in shape to the main line. At strong coupling
(stronger g) localized resonances also arise.
v (E) = gv v (E)
(14.75)
casting the r.r. in the form

3
4
v (E) = v0 G0rr (E) + G0 (E v0 ) v+1 (E) + vg2 v1 (E) .
(14.76)
For v = 0, the solution is given by the continued fraction

0 (E) = G(E)
G0 (E)
1
g2 G0 (E)G0 (E
0)
2g2 G0 (E0 )G0 (E20 )
3g2 G0 (E20 )G0 (E30 )
1
1.................................
(14.77)
from which, using the r.r. again, the other amplitudes with are easily derived. The interacting density of states (E) = Im()
gets satellites and
14.4 Feenberg Method
329
is deformed in a characteristic way (see Fig. 14.2). Note that G0rr (E) is not
a simple energy denominator as in Haydocks method and contains the full
band-theory information. This method has already been applied to a variety
of problems involving few electrons interacting with bosons, leading to exact
solutions in many cases. The present is a simple example, and more generally
the solution([124])[46]) is not given by a continued fraction.

Feenberg[97] long ago introduced a very general method for solving linear
problems, later reviewed by Swain[98] who aimed at applications in the eld
of Nonlinear Optics. Let A denote a square matrix of order N and aij its
elements. By denition,

Q
() aq(1)1 aq(2)2 ...aq(N )N
(14.78)
Det A =
Q
is the sum of all the products of N elements in the matrix, one for each line
and column; each product is multiplied by the sign of the permutation Q
of the rst indices, when the product is ordered with the second indices in
ascending order 1 to N. Below Ill write simply A for Det A. Everybody knows
the Laplace development; choosing an arbitrary row index i,

()(i+j) aij Aij ,
(14.79)
A=
j
rc
where A is obtained from A by removing row r and column c. This is not the
only useful expansion, however. It often happens that the diagonal elements
are nite, that is, O(1), while the o-diagonal ones are O(V), where V is some
small parameter. In such cases, the Feenberg development is specially suited,
since besides being exact, it can serve as an expansion in powers of V.
Let Ai be the determinant obtained from A by removing the i-th row
and column; Aij the determinant of the matrix that one obtains from A by
removing rows and columns i and j; and so on. Chosen at will a diagonal
element aii , in each term of A it is multiplied by elements coming from rows
A
and columns
= i; thus a
is an antisymmetrized product of all the elements
ii
A
from other rows and columns, and this is just a
= Ai . Given a pair of
ii
A
indices i and j and the product pij = aij aji , pij is an antisymmetrized
A
= Aij ; one
product of the elements from other rows and columns, so p
ij
can continue in this way, ending with A123...N 1. Thus, one obtains a closed
development:

aij aji Aij +
aij ajk aki Aijk
A = aii Ai
j=i
k=j,i j=i

l=k,j,i k=j,i j=i
aij ajk akl ali Aijkl + ....
(14.80)
330
A convenient notation is

A = aii Ai
aij aji Aij +
aij ajk aki Aijk
j
j
k

aij ajk akl ali Aijkl + ....

l
(14.81)

where j means that the index j must be dierent from any other index
in the summand. Equating the Feenberg and the Laplace developments one
nds for i
= j; summarizing,
Aij = Ai , i = j

= ()(i+j) [aji Aij
ajk aki Aijk +
k

ajk akl ali Aijkl ], i
= j.
l
(14.82)
The sequences of dierent indices like those in ajk akl ali are called by
Swain[98] irreducible processes. To write down a general term of such an
expansion one can start from enumerating all the irreducible processes that
occur at a given order. The elements of the inverse matrix are:
(i+j)
(A1 )ij = ()
Aji
A
(14.83)
and the diagonal ones read5

Gi (A1 )ii =
Ai
.
A
(14.84)
Let us introduce:
Di = AAi , Dij =
i
=
Dji = AAiji , Djk
A
A
Aij , Dijk = Aijk ,
A
Ai
il
il
Aijk , Djk = Aijkl ,
(14.85)
where where each D brings lower and possibly upper indices; the upper ones
are those of the A above, while the A below bears the upper and the lower
indices. Directly from the Feenberg development,
a a A

A
Di = AAi = aii j ij Ajii ij + j k aij ajk aki Aijk
i

Aijkl
l k j aij ajk akl ali Ai + ....
a a
a a aki
= aii j ijDiji + j k ijDjk
i
j
jk
aij ajk akl ali
l k j
+ ....
Di
(14.86)
jkl
and similarly
5
one uses the notation Gi since if A = z H these are resolvent matrix elements.
Dki =
akj ajk akj ajm amk

Ai
= akk
+
+
ik
j D ik
jm
Aik
Djm
j
331
(14.87)
where the summation indices cannot be equal to i.

We can expand the D ratios: from the formula for A one obtains for
instance

Auv = aii Aiuv j aij aji Aijuv + j k aij ajk aki Aijkuv

(14.88)
l k j aij ajk akl ali Aijkluv + ....
hence we obtain a D with a single low index:

A
A
uv
Diuv = AAuvi
= aii j aij aji Aijuv
+ j k aij ajk aki Aijkuv
uvi
uvi

A
l k j aij ajk akl ali Aijkluv
+
....
uvi
that is,
a a

Diuv = aii j Dijiji + j k
a a ja a
l k j ij Djkikl li + ....
aij ajk aki

i
Djk
(14.89)
(14.90)
jkl
D objects with several low indices can be written down as products of D

objects having fewer low indices:
Dij =
A
A Ai
=
= Di Dji = Dj Dij
Aij
Ai Aij
(14.91)
By raising indices one generates a recursive solution.

Example: (A1 )11 for a Three-Diagonal Matrix A.
! 1 "
A1
1
= G1 =
A
=
11
A
D1
(14.92)
where
D1 = a11
a1j aj1 a1j ajk ak1

+
...
1
j
j
k
Dj1
Djk
we realize that a1j = 0 unless j = 2; then, in the second summation we must

read a12 a2k ak1 , where k
= 1, k
= 2. Then since ak1 = 0, only the rst sum
remains, and
D1 = a11
D21 =
a12 a21
.
D21
a2j aj2 a2j ajk ak2

A1
= a22
+
....
12
j D 12
jk
A12
Djk
j
(14.93)
(14.94)
332
where j=1 is forbidden, and so j=3; then, no k is acceptable and we are left
with
a23 a32
D21 = a22
.
(14.95)
D312
So,
G1 =
1
a11
a12 a21
a a
a22 231232
(14.96)
D
3
In this way, one nds the recurrence relations

Ci = aii
aii+1 ai+1i
,
Ci+1
Ci = Di1,2,...i1
(14.97)
(14.98)
and the result is a continued fraction.

14.4.1 Solving Linear Systems
Given the linear non-homogeneous system

aij xj = bi
(14.99)
Cramers formula yields the well-known solution

xi =
()(i+j) Aji
j
bj
(14.100)
and we can expand the numerator by (14.82)

b ()(i+j)
xi 5= AAi bi + j j A
()()(i+j)
6

aji Aij k ajk aki Aijk + l k ajk akl ali Aijkl ...
that is, since Aijl... are symmetric in the exchange of indices,
a A b
xi = AAi bi j ij Aij j +
aik akj Aijk bj aik akl alj Aijkl bj
jkl
...
jk
A
A
or
xi =
aij bj aij ajk bk aij ajk akl bl

bi
...
j Dij
jk
jkl
Di
Dijk
Dijkl
(14.101)
333
14.4.2 Homogeneous systems

Homogeneous linear systems are also of interest, for instance to nd the
eigenvectors of a matrix H:

aij xj = 0, i = 1, ..., N.
(14.102)
j
This case reduces to the previous one: provided r is such that the determinant
Ar does not vanish, one can assign an arbitrary value to xr and the reduced
system

aij xj = air xr , i = 1, ..., N, i
= r
(14.103)
j=r
with bi = air xr (excluding row and column r from all determinants) has a
unique solution
xi
air aij ajr aij ajk akr
= r +
+ ...
(14.104)
r
j Dr
jk
xr
Di
Dijk
ij
Problems
14.1. Show that if we can calculate the local Greens function, we can also
obtain any matrix element Gmn (E).
14.2. For the rectangular band model (14.20) n() =
continued fraction coecients.
(W 2 2 )
,
2W
nd the
14.3. For the semiellyptic band model, nd the continued fraction coecients.
14.4. The 33 linear system Ax = b with a non-singular matrix
a11 a12 a13

A = a21 a22 a23
a31 a32 a33
is notoriously solved by x = A1 b, where
11
11
1
1 a22 a23 1 1 a13 a12 1 1 a12
1
11
11
1 a32 a33 1 1 a33 a32 1 1 a22
11
11
1
1 a23 a21 1 1 a11 a13 1 1 a13
1
1
1
1
1
11
A =
1
11
11
detA
a33 a31 a31 a33 a23
1
11
11
1 a21 a22 1 1 a12 a11 1 1 a11
1
11
11
1 a31 a32 1 1 a32 a31 1 1 a21
1
a13 11
a23 1
1
a11 11
.
a21 1
1
a12 11
a22 1
But how does the solution work by Feenbergs method? Try to nd x1 .
15 Aspects of Nonlinear Optics and

Many-Photon Eects
15.1 Diusion of Radiation in Dipole Approximation

For strong electromagnetic elds, like those produced by a laser, the linear
response theory fails. Interesting processes leading to color changes become
important, and this has implications and applications. In this Chapter we
shall study the Raman eect, Second Harmonic Generation, the diusion of
Coherent light and the Dynamical Stark eect.
By diusion or scattering we mean those 2-photon processes where one
photon is annihilated and one is created. In the elastic (Rayleigh) diusion
the frequency of the two photons is the same, in the inelastic (Raman) scattering a shift of frequency brings further information about the scatterer.
The Quadratic Response formalism of Section 6.4 is well suited to describe
two-photon processes (not only the scattering processes, but also others, like
the decay of the H 2s level).
Classical Model of Diusion by a Molecule
Classically, the elastic diusion is caused by the induced currents. The inelastic diusion is due to the following mechanism. An oscillating dipole of
frequency 0 radiates a power
I=
2 04
2
d .
3
3c
(15.1)
Now let an electromagnetic wave of pulsation induce in the molecule a

dipole
d =
E cos(t);
(15.2)
the polarizability
is a second-rank tensor. If m is the pulsation of the
molecular vibrations,
=
0 +
1 cos(m t + )
(15.3)
where is an arbitrary phase. Thus, one expects elastic radiation at and inelastic components at m . The one at m and +m are called Stokes
and anti-Stokes radiation, respectively. Classically, their intensity should be
the same, actually the Stokes light is more intense, but this requires Quantum
Theory to explain.
336
15 Aspects of Nonlinear Optics and Many-Photon Eects
Photon
We use the trasverse gauge, without scalar potential and vector potential
A = N
ei k r it ;
(15.4)
for the normalization we ask that a unit volume V contains a photon energy,
that is,

1

.
(15.5)
E E + B B d3 r = h
4 V

A ,
From E = t
B = rot A , one nds E E = B B = N 2 2 ; putting into
2
2
(15.5), 2N
and
4 V = h
A =
2
h
ei k r it ;
V
(15.6)
We shall often work with V = 1 and h

= 1.
Dipole Approximation
In the semiclassical theory, the dipole moment e

r = d is a time-independent
operator, while E (t) is an external eld. The radiation-matter interaction

is introduced by adding to the Lagrangian e e A

v , where v is the
velocity. Here, = 0 and we are considering long wavelengths
k 0, when

2
h
e
e d( A r )
dA
it
e
. Then, using
r
A r =
we
A
V
dt
dt
dt
drop the total derivative ending up with the Hamiltonian
H (t) = E d = VA eit + VE eit .
(15.7)
Here,
2
h
(15.8)
d , VE = VA ,
V
describe photon absorption and emission, respectively.
VA = i
where VA and VE
15.1.1 Second-Order Processes

Let |i and |f denote two stationary states of the unperturbed molecule. An
incident photon of pulsation 1 can induce a |i |f transition ending up
as a photon of pulsation 2 in the solid angle d2 leading to a detector. We
wish to obtain the dierential cross section of this two-photon process. The
answer is contained in the second-order term of Equation (2.36) or in the
more intuitive, but equivalent argument below.
As a preliminary, simpler calculation we seek the probability amplitude
Aif (t) = f |t that the molecule makes the transition to f | assuming 1) that
337
both photons belong to sharply specied modes 2) that photon 1 is absorbed

at time t1 and photon 2 is emitted at time t2 . There are two time orderings
and two contributions, represented by a) and b) in Figure 15.1. Quantum
theory also yields diagram b), contrary to the classical intuition, because the
eect (photon emission) precedes the cause. The quantum uncertainty also
aects time.
t= |f
t= |f
1
t=t2
t=t1
t=t1
t= |i
t=t2
t= |i
a)
b)
Fig. 15.1. A photon scattering second order process: in a) the photon is absorbed
and then re-emitted, in b) it is re-emitted and then absorbed.
The interaction picture state at time t corresponding to diagram a) is

|t = [iVE (2 , t2 )]I ei2 t2 [iVA (2 , t2 )]I ei1 t1 |i .
(15.9)
This would be the correct result if the times t1 and t2 were assigned by the
experiment. Since this is not the case, we must allow for the interference of
dierent values, by integrating over the intermediate times. A part Aif of
the amplitude comes from events in which t2 > t1 : the contribution to the
process is f |t where

|t =
dt2
t2
dt1 [VE (2 , t2 )]I [VA (1 , t1 )]I ei(2 t2 1 t1 ) ,
(15.10)
thus

a)
Aif (t) =
dt2
t2
dt1 f | [VE (2 , t2 )]I [VA (1 , t1 )]I |i ei(2 t2 1 t1 ) .
Expanding the interaction picture operators
(15.11)
338

a)
Aif (t) =
t2
dt2
dt1
f |eiHt2 VE (2 )ei(H2 )t2 eiHt1 VA (1 )ei(H1 )t1 |i
(15.12)
and working out the calculations, one nds

Aif (t) =
f |VE (2 )ei(HEf 2 )t2 e
dt2
i(HEi 1 )t1
t2
dt1
VA (1 )|i .
(15.13)
We do the t1 integral giving the integrand the usual adiabatic switch:

t2
eixt2
.
(15.14)
dt1 eixt1 =
ix
Thus,

a)
Aif (t)
dt2 f |VE (2 )ei(HEf 2 )t2 ei(HEi 1 )t2

=
i
VA (1 )|i
H Ei 1 i0
t
dt2 ei(Ef 2 )t2 +i(Ei 1 )t2
f |VE (2 )
i
VA (1 )|i (15.15)
H Ei 1 i0
and nally
i
Ef Ei + 2 1 i0
i
f |VE (2 )
VA (1 |i .
H Ei 1 i0
Aif (t) = ei(Ef +2 Ei 1 )t

a)
(15.16)
The contribution of diagram b) is

i
Ef Ei + 2 1 i0
i
f |VA (1 )
VE (2 |i ,
H Ei 2 i0
Aif (t) = ei(Ef +2 Ei 1 )t

b)
(15.17)
and the total amplitude is

a)
b)
Aif (t) = Aif (t) + Aif (t).
(15.18)
339
Energy Conservation
The 1 photon is incoming, and must be switched on adiabatically; therefore
Aif (t) must have an innitesimal exponential growth. Letting x = Ef Ei +
2 1 ,
et
e2t
|Aif |2 2
, 0.
Aif
x i
x + 2
The transition probability per unit time is
2e2t
dPif
2
2(x), 0.
dt
x + 2
(15.19)
In this way, the energy conserving (Ef Ei + 2 1 ) factor arises much

more elegantly than in more elementary treatments.
Density of States for the Outgoing Photon
The experiment measures the energy h2 of the outgoing photon and its
direction within a solid angle d2 : instead, Pif refers to a well determined
photon mode with given four-momentum h

( k , ). Therefore, we must sum
dPif
+ Ei Ef having k within d2 . The sum
dt on the modes with 2 = 1
(2 |k|), that is,
is the density of states d(2 )
k d2

V d2
V d2 2
d(2 ) =
.
(15.20)
d3 k(2 |k|) =
3
2)
(2)3 2
The incident photon is normalized to 1 in volume V ; the probability per unit
time to observe the outgoing photon is the dierential cross section
1
1
i
VA (1 )|i +
dif = 2 11f |VE (2 )
H Ei 1 i0
12
1
i
VE (2 )|i 11 d(2 ).
f |VA (1 )
(15.21)
H Ei 2 i0
Recalling (15.8),
2
dif = |M |
1 23 d2
,
h2 c4
(15.22)
where
i
1 d |i +
H Ei 1 i0
i
2 d |i .
f |
1 d
H Ei 2 i0
M = f |
2 d
(15.23)
340
In the second denominator, we introduce a complete set and since 2 =

1 + Ei Ef we obtain

1 df n
1 dni
2 dni
2 df n
(15.24)
M=
+
ni
1
nf
1
n
provided that no denominator vanishes (otherwise one speaks of resonant
Raman scattering which requires a separate treatment). Following Kramers
and Heisenberg, we may write

Rpq (2 )p (1 )q ,
(15.25)
M=
pq
in terms of the Raman tensor
Rpq

f |
d p |n n| d q |i f | d q |n n| d p |i
=
+
.
ni 1
nf 1
n
(15.26)
In the elastic f = i case, the tensor describes the Rayleigh scattering. Otherwise if 1 > 2 then Ef > Ei (Stokes transition); in the opposite case the
system gives energy to the photon (anti-Stokes transition).
Inversion Symmetry
We saw in Chapter 8 that in systems like the Benzene molecule possessing
inversion symmetry, Rpq = 0 unless |i and |f are of the same parity. Thus,
Raman-active modes are not seen in infrared absorption, and conversely infrared active modes yield no Raman scattering. infrared.
Low-frequency Scattering
For 1 0, the Raman tensor has a nite limit. Hence, the cross-section
(15.22) goes like 1 23 . The elastic scattering goes with 14 . This explains
why the sky is blue and the advantage of radio-astronomy over infrared or
optical observations of objects that are beyond interstellar clouds.
High-frequency Scattering
When 1 is large compared to the main absorption frequencies, we insert into
(15.26) the expansions
1
ni
1
2 + ,
ni 1
1
1
1
nf
1
2 + ,
nf + 1
1
1
and nd
(1)
(2)
Rpq = Rpq
+ Rpq
+
(15.27)

1
f | d p |n n| d q |i + f | d q |n n| d p |i = 0
1 n
(15.28)

1
ni f | d p |n n| d q |i + nf f | d q |n n| d p |i .
2
1 n
(15.29)
d
ni n| d q |i = n|[H, d q ] |i = in| dt
d q |i ,
d
nf f | d q |n = if | dt d q |n ,
(15.30)
with
(1)
=
Rpq
and
(2)
Rpq
=
341
Since
keeping up to second-order terms,

i
d
d
Rpq = 2
f | d p |n n| d q |i f | d q |n n| d p |i
1 n
dt
dt

d
i
d q d p |i .
= 2 f | d p d q
(15.31)
1
dt
dt

2
For one electron, [ d p , dt

d q ] = iem pq . With N electrons, d = e N
k=1 r k
and

d
N ie2
[ d p,
d q ] =
pq .
(15.32)
dt
m
Thus, for large 1 ,
N e2
Rpq =
if pq .
(15.33)
m12
No Raman scattering occurs. For the elastic scattering, remarkably the crosssection (15.22) becomes frequency independent:
1
12 4
2 2
1 N e2
1 1 d2
e
2
2
1
d = 11
(

)
(

)
=
N
|
2
1 | d2
pq
2 p
1 q1
m12
c4
mc2
2
= N 2 r2 |
2
1 | d2 ,
(15.34)
e
e2
mc2
where re =
1 and
2,
is the classical radius of the electron. If is the angle between
1 |2 = cos2 ().
|
2
The outgoing photon polarization

e 2 must be orthogonal to k 2 but accord
ing to (15.34), cannot be orthogonal to the incoming photon polarization
1.

Therefore 2 is in the k 2 1 plane. Let be the 1 k 2 angle; we obtain

the classical Thomson formula
d = N 2 re2 sin2 ()d2 .
(15.35)
342
15.2 Sum Frequency and Second-Harmonic Generation

(SHG)
t=
|f
|f
|f
|f
3
2
1
t= |i
a)
|i
b)
|f
|f
|i
|i
d)
c)
|i
e)
|i
f)
Fig. 15.2. The lowest-order second-harmonic generation diagrams.
As discussed in the last Section, the inelastic emission of radiation can be

treated by perturbation theory, unless of course the incident radiation elds
are comparable to those that exist in the unperturbed crystal and can reach
109 Volt/cm. Such elds can alter and even destroy the specimen1 . So,
the method used above for the Raman scattering is useful again and yields
the probability of absorbing 1 and 2 photons and emitting 3 . Since there
are 3! orderings, 6 diagrams contribute in lowest order (gure 15.2) . The
amplitude is
(15.36)
B = X(1 , 2 ) + X(2 , 1 )
where in obvious notation, with fn f (n ) the Fermi distribution, and f =
1 f,
6

X(1 , 2 ) =
fs fn (s, n, r, i, 1 , 2 ),
(15.37)
i
(s, n, r, 1, 1 , 2 ) = fr
snr
[VE (3 )]sn [VA (2 )]nr [VA (1 )]rs

,
(rs 1 i0) (ns 1 2 i0)
(15.38)
Blowing up the sample is not the best way to study it; however the next Section
deals with interesting cases that cannot be treated at low order because the eld is
so strong that all the degrees of freedom are dressed and deeply aected by photons.
15.3 Diusion of Coherent Light
343
(s, n, r, 2, 1 , 2 ) = fr
[VA (2 )]sn [VE (3 )]nr [VA (1 )]rs

,
(rs 1 i0) (3 1 + ns i0)
(15.39)
(s, n, r, 3, 1 , 2 ) = fr
[VA (2 )]sn [VA (1 )]nr [VE (3 )]rs

,
(3 + rs i0) (3 1 + ns i0)
(15.40)
(s, n, r, 4, 1 , 2 ) = fr
[VE (3 )]rn [VA (2 )]sr [VA (1 )]ns

,
(ns 1 i0) (nr 1 2 i0)
(15.41)
(s, n, r, 5, 1 , 2 ) = fr
[VA (2 )]rn [VE (3 )]sr [VA (1 )]ns

,
(ns 1 i0) (3 1 rn i0)
(15.42)
[VA (2 )]rn [VA (1 )]sr [VE (3 )]ns

.
(3 + ns i0) (3 1 + rn i0)
(15.43)
and
(s, n, r, 6, 1 , 2 ) = fr
The transition probability is proportional to

= (3 1 2 )|B(VE ; VA )|2 .
(15.44)
This was my starting point for a calculation2 of SHG spectra from interfaces
[104], an unexplored subject at the time, in the European Esprit Project
EPIOPTICS. Experiments were planned to observe the SHG signal from
Si surfaces and interfaces. EPIOPTICS had the main motivation that SHG
may carry information from buried interfaces, without exposing them, which
is important for materials characterization. A photon is odd under inversion,
and it can be shown that to second order no SHG occurs from systems with
inversion symmetry such as Si. The surface breaks the inversion symmetry,
and with the available radiation sources a tiny fraction (even 1015 or less)
of the incident power was predicted and observed to go into the SHG signal
????.
With a similar formalism one can study other nonlinear eects like the
two-photon decay of the 2s level of the H atom. In the following, we shall
deal with many photon eects that occur when the elds are so strong that
low order diagrams cannot work.

15.3.1 Eective mode
When polarized monochromatic radiation (e.g. a laser beam) is shed over the
sample, one can often formulate the scattering problem to a good approximation as if the photons belonged to one eective mode. Suppose there are
2
In atomic and molecular problems, when no medium aects the eld propagation, my formulation is an alternative to a more traditional one for bulk
materials[105], which was based on the second-order susceptibility (2) .
344
M modes of frequency and polarization

in the beam; let He denote the
many-electron Hamiltonian and N the dimensionality of the Hilbert
(k)
and creation operator bk ,

space. For each mode k with vector potential A
treating the vector potential as a constant in the interaction region in the
spirit of the dipole approximation, the radiation-matter coupling reads

(k)
Hk =
Mmn
am an (bk + bk )
(15.45)
m,n,
with
e (k)
e

m| A
p |n =
A m|

p |n
mc
mc
where
p is the momentum operator. Therefore,
(k)
=
Mmn
k + b ),
Hk = A(k) L(b
k
(15.46)
(15.47)
operates on the electrons but not on the photons. The part of the
where L
Hamiltonian that describes the eld and its interaction is
H=
M /
0
k + b ) .
bk bk + A(k) L(b
k
(15.48)
By a real rotation mk in the mode space, such that mk nk = mn , I

introduce new operators d:

A(k) bk =
A(k)
mk dm ,
(15.49)
k
the rotated vector with components k A(k) mk has the same length as the
original one,

(A(k) )2 .
||A|| =
k
We need not specify the rotation further provided that

A(k) mk = ||A||m1 ;
(15.50)
then,
H=h
1 + d ).
dm dm + ||A||L(d
1
Thus we can use a single mode with eective coupling ||A||.
(15.51)
345
Mode Normalization
For a single mode, with A = c

||A||2 =
r it , that is, A(k) = c 2h ,

k
V
2
h
i
V e
2
hc2 2
h c2 V
=
(3) k,
V
V (2)3
(15.52)
(3)
k
where (3) k is the spread of the laser beam in k space. Actually,
(2)3 is
comparable to the inverse of the volume where the laser beam is coherent;
hence we can safely work with one eective mode if V is the coherence volume.
Coherent Light Scattering

The diusion of laser radiation from a sample can be studied by a model
Hamiltonian of the form
H = He + Hph + HI + HI ,
(15.53)
where He represents the electronic system, Hph the free photons, HI and
HI the interactions of the electrons with the laser and scattered photons,
respectively. The free eld may be described by the Hamiltonian
Hph = 0 d d + 0 dscatt dscatt ,
(15.54)
where d annihilates photons 0 from the laser while dscatt destroys scattered
photons. The interaction terms are:
(d + d )
HI = M
(15.55)
(dscatt + d ),
HI = M
scatt
(15.56)
and M
are operators acting on the electronic degrees of freedom.
where M
The laser photons belong to one mode and are in a coherent state such that
d|c = |c
(15.57)
s = d .
(15.58)
that is, s|c = 0 with

We take real for simplicity. We met a coherent state already in Section
6.1.4; in that case the shift was produced by the g coupling and could be
found by Equation (6.20), while now is a property of the laser. Equation
(6.28) n
= 2 , still holds, however. In the case of laser radiation, can vary in
a wide range: it can reach 104 in continuously operated lasers and largely
exceed 108 in pulsed lasers (pulses last a few tens of a nanosecond).
It is often convenient to perform a canonical transformation to the s
representation,
346
=H
e + H
ph + H
I + H ,
H
I
(15.59)
e = He + 0 2 + 2 M
(15.60)
ph = 0 s s + dscatt dscatt ,
H
0
(15.61)

I = 0 + M
(s + s ) = M
(s + s )
H
(15.62)
where
The initial state of the system is |g; 0 |g |0 , where |g stands for the
electronic ground state and |0 for the photon vacuum.
The scattered spectrum S(0 ) is proportional to the dscatt photon emission rate. Instead of deriving a general formal expression for te rate (see
Ref. [46]), here I consider a special case by a simple, physically motivated
treatment.
15.3.2 Dynamical Stark Eect
A photon impinging on a molecule or a solid can give raise to lots of processes,
but one can choose the symmetry of the system and the photon such that
just one transition is possible. Then the system-radiation interaction can be
understood by a two-level model. F Schuda, C R Stroud Jr and M Hercher
[106] using a dye laser and a beam of Na atoms realized this situation: the
o
laser was tuned at 0 5890 Angstrom in resonance with a dipole allowed

transition; this connects the F=2 hyperne level arising from the 2 S 12 ground
state and the F=3 hyperne level arising from the 2 P 32 state3 . These two levels
form an ideal two-level system because by the F = 0, 1 selection rules no
other levels can be reached by the exciting radiation. They discovered that if a
F=3
3 2P 3
2
F=2
F=3
3 2P 3
2
F=2
2
3 S1
2
F=1
a)
F=2
F=2
3 S1
2
F=1
b)
Fig. 15.3. Hyperne levels of Na with a) the two-level system b) a three-level

system.
3
of
Here, F is the total angular momentum resulting from the nuclear spin I =

and the electronic J = L + S angular momentum.
21
11 N a
3
2
347
coherent eld is in resonance with a two-level system, the scattered radiation

consists of a central quasi-elastic radiations with two symmetric satellites at
higher and lower frequencies, like in the Stark eect. The separation of the
10
,a
side bands from the central lines is 100 MHz, thus
0 2 10
very small eect which however is clearly seen (a typical situation in atomic
physics). The Hamiltonian4 which describes the two-level system interacting
with the laser eld is:
H0 = He + Hph + HI
He = 1 n1 + 2 n2 , Hph = 0 d d
HI = (a1 a2 + a2 a1 )g(d + d ).
(15.63)
Here, m , am denote the unperturbed atomic levels and annihilation operators, nm = am am , g the electron-photon matrix elements, 0 the laser frequency. The scattered photons and their emission
H = dscatt dscatt + HI
HI = (a1 a2 + a2 a1 )g(dscatt + dscatt )
(15.64)
will be introduced later and treated perturbatively.

Shifted Representation
We now shift the laser photon operators by d s + , with =
the following eects:
g
0
obtaining
Hph 0 s s + 0 (s + s ) + 0 2 ,
HI (a1 a2 + a2 a1 )g(s + s ) + 2g(a1 a2 + a2 a1 ).
We rearrange the terms of the transformed H0 as follows:
e + H
I + 0 s s,
0 = H
H
with
(15.65)

e = 1 n1 + 2 n2 + 0 2 + 2G a a2 + a a1 , G = g
H
1
2
(15.66)
!

!
"
"
I = 0 + g a a2 + a a1 s + s = M
s + s .
H
1
2
(15.67)
the initial state of the system is a1 |0 , where |0 is the vacuum for s and
dscatt photons. Note that g enters as G = g in Equation(15.66) and as g
in in equation (15.67). The two entries have quite dierent consequences.
4
For simplicity, we ignore the MF multiplicity and model this as a two-level
system for a spin-less eective particle.
348
Without the g term in equation (15.67) there would be no dynamical Stark

eect. However, to understand the situation, we need to evaluate the relative
importance of the various couplings.
The atomic target has the characteristic length a0 = 1 Bohr radius, with
2
a characteristic energy ae 0 =1 Hartree. We can get reasonable estimates of the
coupling energy g and the number by the following arguments: 1) g must
is proportional
be proportional to the uctuating atomic dipole ea0 2) 2 = n
to the beam energy density nVh in the coherence volume V 1cm3 . This
reasoning leads to
e2
a3
(15.68)
n
h 0 .
(g)2 g 2 n
a0
V
2
At optical or near ultraviolet frequencies such that h

ae 0 , for n
= 108
4
9
( = 10 ) we may have g 10
h or even less. Then in Equation (15.67)
the g term is quite tiny compared to the one in 0 and we can expand
around the limiting case5 g 0, with constant G = g. For g = 0 we
e , that becomes
diagonalize H
e = 1 n
H
1 + 2 n
2 + 0 2 ;
(15.69)
the unperturbed (g 0 ) Hamiltonian is:

g0 = 1 n
H
1 + 2 n
2 + 0 2 + 0 (s + s ).
(15.70)
Back to the Unshifted Representation

Here n
1 and n
2 are conserved; there are eigenstates with n
1 = 1, n
2 = 0
and with n
1 = 0, n
2 = 1. At this point it is simplest to work in the old
representation setting d = s + and getting in the two cases the unperturbed
Hamiltonians
i + 0 d d , i = 1, 2.
(15.71)
h0,i = i n
In this round trip among the representations we did not simply take the
g 0 limit, since we are keeping the corrections of order G that renormalize
the levels. The |i, n eigenstates have eigenvalues i + n0 . In the resonant
1, n + 1 is almost degenerate with |2, n .
case h0 = 2 1 , |
The perturbation is the term that we had neglected,
g(a1 a2 + a2 a1 )(s + s ) = g(a1 a2 + a2 a1 )(d + d 2).
(15.72)
Since G
2 1 , states are little aected by the tildes, and the matrix
element between almost degenerate states is6

1, n+1|g(a1 a2 +a2 a1 )(d+d 2)|
2, n gn+1|d |n = g n + 1. (15.73)
5
Note that the limit alone would correspond to classical elds.

Note that the most important n values are 2 , and in this sense a high degree
of coherence leads to stronger radiation-matter coupling.
6
|2
0
|1
|
1, 3
|
2, 2
|(2)
|
1, 2
|
2, 1
|(1)
|
1, 1
|
2, 0
|(0)
|
1, 0
a)
349
|
1, 0
b)
c)
Fig. 15.4. Resonant radiation interacting with a two-level system. The energy
spectrum is represented a) according to a simple one-photon view b) in the manyphoton description, involving an innite ladder of states, neglecting the g(a1 a2 +
a2 a1 )(s + s ) term c) with the g(a1 a2 + a2 a1 )(s + s ) term included.
First-order perturbation theory yields a very accurate description of the

eect of the tiny perturbation (15.72). The perturbed spectrum consists of the
2,n
|1, 0 ground state and an innity of close doublets; the doublet |1,n+1|
2
n + 1 |2,
n and the splitting is 2g n + 1.
arises from |1,
Scattered Light
Up to now we have dressed the atomic levels with photons from the laser obtaining an innite double ladder of states, but the scattered photons still have
to be considered. The scattered light arises from the spontaneous emission
between the dressed states, according to A. Einsteins A coecient
Amn =
3
4mn
d2 ,
3hc3 mn
(15.74)
where d is the dipole operator.

d connects |
2, n with |
1, n with the same n, hence there are transitions
only between consecutive doublets, with 0 .The frequencies of the transitions in Figure 15.5 a) are given by the scheme:
transition
+ (n) (n 1) 0 + g( n +
1
+
n) 0 + 2g n
+ (n) + (n 1)
0 + g( n + 1 n) 0
(n) (n 1)
0 +
g( n + 1 + n) 0
(n) + (n 1) 0 + g( n + 1 + n 0 2g n)
(15.75)
350
|+ (n)
n0
| (n)
|+ (n 1)
(n 1)0
| (n 1)
a)
b)
Fig. 15.5. The contribution of two consecutive doublets to the scattered radiation.
a) the spontaneous emission lines b) the resulting spectral intensity versus .
Simple Explanation of the Dynamic Stark Triplet Spectrum

0
We wish to plot the spectral intensity S(x) versus x =
2g . Since the matrix
elements are equal, and the frequencies are so close that the variation of the
3 factor is negligible, the 4 transitions in (15.75) have practically the same
intensity. The width of the central peak is of the order of the inverse
lifetime g of the excited state, and we represent it by a Lorentzian. The
system is prepared in a state that for g 0 corresponds to the atom in 1
and a coherent state of d photons. The doublet of states corresponding to
1 n a
a e , where a = 2 . The most
|1, n is occupied with probability Pn = 2n!
important contributions arise from n . Now note that
6
ea an 5
( 0 2g n) + ( 0 + 2g n)
4 n=0 n!
| 0 |ea an
(( 0 )2 4g 2 n).
2
n!
n=0
(15.76)
To obtain the shape of the side peaks, one can rewrite Equation (6.33) in
the form:
(zaz0 )2
n

2g2
a
e
(z nz0 )
, a 1,
(15.77)
ea
n!
z0 2a
n
with z = ( )2 and z0 = 4g 2 . Finally, the spectrum may be approximated
by
(x2 a)2
1
|x|
exp[
]+
.
(15.78)
S(x) =
2
2a
2 + (2x)2
4g 2a
351
The spectral maximum at = 0 is due to the unresolved central transitions,

while half of the intensity belongs to the two satellites at = 0 2g. This
pattern agrees with experiments.
S(x)
0.15
-10
10
Fig. 15.6. The characteristic shape of S(x) versus x for a = 100 and = g.
Problems
Part V
Selected Exact Results in Many-Body

Problems
16 Quantum Phases
16.0.3 Gauge Transformations

Without the gauge invariance, any theory is untenable. In classical theory,
the Hamiltonian of a charged particle is
H=
2
(p eA
c )
+ eV (x)
2m
(16.1)
where p is the kinetic momentum and A the vector potential. Both are unobservable; one could have started with new potentials A , V such that
A = A + , V = V
1
c t
(16.2)
where (x, t) is a completely arbitrary function. Only elds appear in the

classical equations of motion; quite the other way, the gauge transformed
Schr
odinger equation in terms of the old potentials reads

e
1
1

(p {A + })2 ) + e{V
} = ih
.
(16.3)
2m
c
c t
t
Thus, the changes are:
e
e
p (p ) , i
(ih +
) .
c
t
t c t
(16.4)
One nds that the new wave function is related to the old by
(x, t) = (x, t) exp[
ie(x, t)
],
hc
(16.5)
that is, substitution into (16.3) gives the correct equation for . The gauge
changes the phase at any point and at any time in an arbitrary way. The
matrix elements of the coordinates are trivially invariant, while those of the
canonical momentum are shifted by pmn (p + ec )mn ; the Schrodinger
theory is invariant because what matters is the mechanical momentum p ec A
which remains unshifted. Thus, the theory is gauge invariant. Equation (16.5)
also holds in Klein-Gordon and Dirac theories.
356
16 Quantum Phases
16.0.4 Spinor Rotations

Consider the Pauli equation
( B) =
(16.6)
for a spin in a magnetic eld in a Cartesian reference K. Rotating by an angle

counterclockwise about any axis one obtains a reference K. Choosing the
z = z axis parallel to the rotation axis, xik = aik xk with
cos() sin() 0
(16.7)
{aik } = sin() cos() 0
0
0 1
and x appears like x rotated clockwise. How does (16.6 ) transform to K ?
Since the Pauli equations must hold in the rotated system,
( B ) = ,
(16.8)
with the same Pauli matrices as before. Let = R where R rotates

clockwise;
= and ( B) is not a scalar. Writing (16.8) in K one obtains
(RR B ) = ;
this agrees with (16.6) if
Ri R = aik k
(16.9)
iz
2
since then (RR B ) = ( B). This is satised by R = e

. Thus, R can
be considered as an operator phase, that imparts vector-like transformations
on the Pauli matrices, although they are not the components of any vector.
The same reasoning applies to Diracs matrices under rotations and boosts:
they are not four-vector components, yet Diracs equation is covariant.
16.0.5 Galilean Transformations
In order to simplify the writing, let us consider a one-dimensional problem
for a spinless particle in a potential U ,
2 2
h
(x, t) + U (x, t) (x, t) = ih (x, t).
2
2m x
t
(16.10)
We go from the reference K of coordinates (x, t) to a reference K of coordinates (x , t) moving with speed v, via the Galilei transformation x =
x vt, t = t. Galilean covariance requires that the problem in K becomes
2 2
h
(x , t) + U (x , t) (x , t) = ih (x , t).

2m (x )2
t
(16.11)
16.1 Topologic phases
357
We need to specify how U transforms and we assume that it is a scalar, that

is,
U (x , t) = U (x vt, t) = U (x, t).
(16.12)
We need to write (16.11) in K and in order to change variables we note that
for any f (x, t) it holds that
f (x, t) = f (x + vt , t ) = f (x, t) + v f (x, t).

t
t
t
x
Hence, the rule is:
t
x
= v x
(16.13)
and (16.11) becomes

[
2 2
h
+ U (x, t)] (x vt, t) = [i

h + ihv ] (x vt, t).
2
2m x
t
x
(16.14)
We get back (16.10) by introducing a phase factor:

(x , t) = (x, t)ei(x,t) .
(16.15)
(x, t) = 0 would mean that the wave function is a scalar, as in fact it is

in the relativistic Klein Gordon-theory1 ; in the present case, instead, the
invariance is obtained by
(x, t) =
mvx mv 2 t
.
h
2h
(16.16)
The probability density is scalar although the wave function is not.

Pancharatnam phase
The Indian physicist S. Pancharatnam in 1956 introduced[14] the concept of
a gemetrical phase. Let H( ) be an Hamiltonian which depends from some
parameters, represented by ; let |( ) be the ground state. We can dene
the phase dierence 12 between two ground states |( 1 ) and |( 2 )

(provided that they are not orthogonal):
1
Klein-Gordon particles may be pseudoscalar (change sign under parity) as pions

do. In Diracs theory a Lorentz transformation (boost) corresponds to a rotation
by an imaginary angle in 4-dimensional space-time, and the 4-component spinor
undergoes such a rotation. Thus, the transformation law depends on the kind of
particle.
358
16 Quantum Phases
( 1 )|( 2 ) = |( 1 )|( 2 ) |ei12 .

However, this is gauge dependent and cannot have any physical meaning.
Now consider 3 points and compute the total phase in a closed circuit
1 2 3 1 ; remarkably,
= 12 + 23 + 31
(16.17)
is gauge independent! Indeed, the phase of any can be changed at will by

a gauge transformation, but such arbitrary changes cancel out in computing
. This clearly holds for any closed circuit with any number of . Therefore
is entitled to have physical meaning. This revealed that there may
be observables that are not given by Hermitean operators. For instance, consider a Linear Combination of Atomic Orbitals (LCAO) model for
a molecule or cluster (or a Hubbard Model[17]). Thus, th is the matrix element
of the one-body Hamiltonian between site orbitals. We want a rough estimate,
because in order to understand the essentials we are ready to neglect details.
Thus we neglect overlaps between dierent orbitals ; phase dierences are
implied from (16.2). Therefore, the inter-atomic hopping terms are modied
by the vector potential according to the Peierls prescription
th th e
2i
0
b

A d
r
a
(16.18)
where
hc
= 4 107 Gauss cm2
(16.19)
e
is the ux quantum or uxon. In the case of H2 this can be gauged away,
but with three or more atoms the physical meaning is that a magnetic ux
is concatenated with the molecule; changing by a uxon has no physical
meaning, however.
0 =
16.1.1 Parametric Hamiltonians and Berry Phase

Let H[R] be an Hamiltonian which depends on a set R of 3 parameters, as
for example the components of a vector 2 . A basis of stationary eigenstates
is dened (up to the phases, which are arbitrary) by:
H[R]an [R] = En [R]an [R]
(16.20)
Now suppose the system is prepared in an energy eigenstate of the Hamil
tonian at time t = 0 but R = R (t) depends on the time. It is assumed that

H[R] has a discrete spectrum with no degeneracy and the evolution of the
Hamiltonian H[ R (t)] adiabatic, i.e. it is very slow; the involved frequencies

are small compared to the resonance frequencies that a stationary copy of
2
With more parameters the problem requires more complicated mathematics.
359
the system at any time would have3 . To be specic, let the motion start at
time 0 and end at time T ; therefore the duration T must be long enough to
meet the adiabatic assumptions. We wish to solve
i
h
n = H[R(t)]n
t
(16.21)
with the initial condition

(t = 0) n (t = 0) = an [R(0)].
(16.22)
We shall nd that the integral of (16.21) bears a real Berry phase [65] n (t),
which turns out to be due to the topology of H in parameter space. This phase
can then be dropped or altered by a gauge transformation at any t, but a
net phase acquired in a closed path cannot be gauged away. To calculate the
phase, lets expand on the instantaneous basis:

n (t) = an [R(t)]cn (t) +
cm am [R(t)], cn (0) = 1.
(16.23)
m=n
By the Kato adiabatic theorem 4 under our assumptions (nondegenerate discrete spectrum and slow enough evolution) only the rst term remains: the
system does not move from the n-th eigenstate. The adiabatic solution has
|cn (t)| 1, and the only eect of the evolution is on the phase, namely,
cn (t) = ei(n (t)+n (t)) ;
here

hn (t) =
(16.24)
dt En [R(t )]
(16.25)
is the familiar dynamic phase, which exists in stationary problems as well.
n (t) using (16.23,16.24) we need to keep in mind that

To compute i
h t
t an = R R an where R R is a scalar product in parameter space; hence

i
n (t) = ( En + i n + R R )an e h
t
h

+
cm am +
cm a m .
m=n
t
0
dt [En [R(t )]+in (t)
(16.26)
m=n
We drop the last term which is negligible, because the coecients of the
m
= n terms are small and the derivative is also small: these are second
order terms. Using (16.24), 0 = (i
h t
H)n becomes
3
The original paper by Berry is based on the adiabatic hypothesis which makes
the analysis easier, however the existence of the Berry phase does not depend on
this assumption.
4
The proof is clearly discussed in Ref.[194].
360
16 Quantum Phases
t

i
dt E [R(t )]+in (t)]
0 =
h n + ih R R an e h 0 [ n
[cm Em ]am
+
m=n
Now, a scalar product by an removes completely all states with m

= n and
leaves us with an equation for the Berry phase:
n = i R an [R(t)]|R |an [R(t)] .
(16.27)
The matrix element looks similar to a momentum average, but the gradient
is in parameter space. The overall phase change is a line integral
T

dtan |R |an R = i an |R |an dR.

n = i
0
The phase dierence between two an [R] at dierent R points ia arbitrary; so

the integrand is undened and this result is quite un-physical, except that in
a closed circuit C, the Berry phase, given by

(16.28)
n (C) = i an |R |an d R
C
is a property of H. In trivial topologies when C can be contracted to a point

the phase vanishes. In multiply connected problems the phase does not vanish
in general and is a geometric property which characterizes the Hamiltonian
as a function of the parameter space. Discretizing a continuous path in parameter space, the Pancharatnam phase picks contributions
ei =
()|( + )
|()|( + ) |
(16.29)
at each step and we arrive at Berrys phase as a limit.

Open-path Berry Phase
Let a (discrete or continuous) path in parameter space go from 1 to an
equivalent point N connected to 1 by a symmetry: H(N ) = U H(1 )U ,
with unitary U ; adopting the natural choice (N ) = U (1 ) one makes the
phase dierence along the path
=
N
i,i+1
i=1
gauge invariant. This is an open path Berry phase; |(1 )|U |(1 ) | = 1
because the spectrum is not degenerate. In the case of a symmetry connecting
1 to itself, we speak of a single-point Berry phase ; by (16.29),
ei = |U | = arg(|U | ).
(16.30)
361
Vector Potential Analogy

One naturally writes

n (C) =
An dR
(16.31)
introducing a sort of vector potential (which depends on n, however) A n =

ian |R |an . The gauge invariance arises in the familiar way, that is, if we
modify the basis with
an [R] ei(R) an [R],
then A n A n R , and the extra term, being a gradient, does not

contribute to n (C). The Berry phase is gauge independent and real since
an |an = 1 = R an |an = 0
that is R an |an + an |R an = 0, or an |R an + c.c. = 0; this implies
that an |R an is imaginary. We can also grant the reality of
A n = ian [R]|R an [R] = Iman [R]|R an [R] .

We prefer to work with a manifestly real and gauge independent integrand;
going on with the electromagnetic analogy, we introduce the eld as well:

n (C) =
Bn
rot A n
n dS
n dS.
(16.32)
S
Let us derive an explicit formula for B. We start by computing [ (a|a)]i =

ijk j (a|k a). When performing the derivative, the term in j k a is symmetric in j and k and therefore vanishes when antisymmetrized. Hence,
[ (a|a)]i = ijk (j a|k a) = [(a| |a)]i . Next we insert a complete
set |am am | and obtain
B n = ImR an | |aR an = Im
R an |am am |aR an .
m
Note that
an |R an [R]
is imaginary, (the terms with m
= n have arbitrary phases). Thus under the
Im operator we may remove the real m=n term. If you think that saving one
term is not an achievement, wait a moment. We get:
R an |am am |aR an .

(16.33)
B n = Im
m=n
To simplify (16.33), we apply R to H[R]an [R] = En [R]an [R] :
362
16 Quantum Phases
(R H[R])an [R] + H[R]R an [R] = (R En [R])an [R] + En [R]R an [R];

next we multiply by am [R], with m
= n. We obtain:
am |R |an =
am |R H|an
.
Em En
(16.34)
There are severe divergence problems if there are degeneracies along C,

and we wisely limited ourselves to discrete non-degenerate spectra.
Example
For example, consider a particle with any spin s in a magnetic eld B (a

physical one, this time) with H = k

hB
s , with k a constant. The energy
eigenvalues are
hkB, n (s, s).
En (B) = n
Initially the particle is in a stationary state. Let us shift the direction of
B adiabatically, sweeping a solid angle (C). Berry has shown that the
geometric phase is n (C) = n(C).
16.1.2 Polarization of Solids
The modern theory of polarization in solids is based on these ideas; for a
review, see [16]. Consider a periodic solid with Hamiltonian
1 2
p + V.
2m i=1 i
N
H=
(16.35)
The usual denition of the dipole moment of a system in terms of the density
(r), which applies to molecules, is:

d = e d
r
r (
r ) = e0 | R |0
(16.36)
where |0 is the ground state and
eR = e
r i,
N
(16.37)
i=1
summed over the N electrons,is the dipole operator. There are problems in
applying this to solids because: 1) in solids we should like to use periodic
boundary conditions, and the dipole operator is forbidden, since it takes
outside the Hilbert space of periodic functions 2) we should like to deal with
a bulk property, while (16.36) depends crucially on the surfaces 3) one never
measures anything like (16.36). What one can do is rather to measure the
d
(e R ) ) coming out from the crystal under
current (which is proportional to dt
some time-dependent probe, like a mechanical stress; we suppose that the
experiment is done adiabatically.
363
Adiabatic Current
One can dene the current density j = T rj,
j=
N

P
,
P
=
pi
mL3
i=1
(16.38)
in terms of the total momentum P , the electron mass m and the super-cell
size L; the polarization change in the experiment is

P = dtj(t) .
(16.39)
The density matrix (t) in the adiabatic limit was given by Niu and Thouless
[80] in terms of the eigenstates |n, t ; the instantaneous density matrix
i = |0, t 0, t|
(16.40)
(t) = i (t) + (t).
(16.41)
commutes with H(t); we write
By the adiabatic theorem , the correction is small, and to a good approximation,

(0n |0, t n, t| + n0 |n, t 0, t|).
(16.42)
= |0, t 0, t| +
n>0
Since does not depend on time explicitly,

i
d
[i + ] = [H, ];
dt
(16.43)
d
in the adiabatic case dt
is negligible (time derivatives are small, the correction is small). Hence, we are left with
d
i = [H, ].
dt
(16.44)
For n > 0, since (16.40) implies 0, t|i |n, t 0,

0=
d
d
d
d
0, t|i |n, t = 0, t| i |n, t + 0, t|i |n, t +0, t|i | n, t ; (16.45)
dt
dt
dt
dt
d
d
but dt
0, t|i |n, t = 0 also vanishes (since dt
0, t|0, t = 0) and
0, t|
d
d
d
d
i |n, t = 0, t|i | n, t = 0, t| n, t = 0, t|n, t ,
dt
dt
dt
dt
(16.46)
364
16 Quantum Phases
where the last equality follows from 0, t|n, t = 0. Taking matrix elements of
(16.44) we obtain the desired correction
0n = i
t|n, t
0,
.
E0 En
(16.47)
Taking into account Equations (16.38 ) and (16.42) the average current is
j =
ie
h 0 |n n |P |0
+ c.c.
mL3 n>0
En E0
(16.48)
Idealized Experiment
Rather than a mechanical stress, we can more simply (from the theorists
point of view) use ingenuity to introduce a tiny electric eld without disturbing the periodicity of the solid. Consider modifying the Hamiltonian (16.35)
to read
N
1
(pi
hk)2 + V.
(16.49)
H=
2m i=1
This device [81] corresponds to a constant vector potential (that will vary
slowly in time) and would represent a gauge change, were it not for the
periodic boundary conditions; running through the supercell of length L an
electron collects a magnetic ux. Indeed, this is called Hamiltonian with a
ux. In order to make this compatible with the periodicity, we must choose
()
with components
k appropriately; we introduce vectors k
()
2
;
L
(16.50)
()
now if k is a combination with integer coecients of k , the Resta operator
U (k) = ei k R ,
(16.51)
is allowed by the periodic boundary conditions. The vector potential requires

a phase factor which is the many-body version of (16.5); in this case this is just
()
U ( k ): picking k = k , which makes U allowed, we satisfy the modied
Schr
odinger equation and the periodic boundary conditions by writing the
ground state of (16.49)
()
()
|0 ( k ) = U ( k )|0
(16.52)
Relation to Berry Phase

To see how this has to do with the Berry phase, the relevant quantity is
()
z () = 0 |U ( k )|0 = |z () |ei
365
(16.53)
where , that may depend on , is a single-point Berry phase. Note that U

is unitary and replacing the ground state by a sum over the complete set,

()
()
0 |U ( k ) |k k |U ( k )|0 ;
|z () |2
k
however the r.h.s. is unity and we may conclude |z () |2 1. We may note

that z () is a true and highly correlated many-body quantity, that unlike the
dipole (16.36) cannot be expressed in terms of the density.
We need to deal with this problem by perturbation theory, since then the
expressions of the current and of turn out to be closely related. Using (16.49)
with the periodic boundary conditions we have k-dependent eigenvalues and
to rst-order the modied ground state U |0 reads
|0 (k)(1) |0 +
|n n | h kP |0 |
m
.
E
E0
n
n>0
(16.54)
Normally in writing this rst-order expansion one does not care about the
phase, but here we must keep track of (16.53); so we write using (16.50)
|0 (k) = ei {|0 +
2
h |n n |P |0 |
}.
mL n>0
En E0
(16.55)
Multiplying by 0 | one nds

z = ei ,
(16.56)
and multiplying by 0 |, since 0 |0 = 0,

0 |U (k () )|0 = ei {
2
h 0 |n n |P |0 |
}.
mL n>0
En E0
(16.57)
From (16.56) one nds = Im log(z ); hence

= Im[
0 |U |0 + 0 |U | 0
]
0 |U |0
and then (16.57,16.53) yield

0 |U (k () ) |0
d
= i
+ c.c.
dt
0 |U (k () ) |0
(16.58)
and eventually, comparing with (16.48),

j =
e
.
2L2
(16.59)
366
16 Quantum Phases
The connection with the Berry phase is established, and the polarization
change P is proportional to .
The actual calculation can be accomplished by expanding in determinants.
Indeed, if |0 = Det|1 N | is a determinant,

U (k () ) |0 = exp ik
ri Det|1 N |
i
also is, with spin-orbital i multiplied by exp [ik ri ] . The overlap (16.53) is
computed as a determinant of one-body overlaps.
Problems
16.1. Given the two reference systems of Section 16.0.5, if in K
k2 t
(x, t) = eikxih 2m
what does the transformed state look like in K ?
17 Pairing from repulsive interactions
Long ago, Kohn and Luttinger [174] proposed that a superconducting instability should occur in Jellium. They focussed on pairs of parallel spin electrons
of large relative angular momentum L. The large L and the triplet state keep
the electrons apart, so the Coulomb repulsion is particularly mild in such
pairs. At long distances, the screened interaction undergoes Friedel oscillations due to the singularity of the dielectric function at 2kF (See Section
12.1.1). This means a slight over-screening of the repulsion in some distance
ranges: one gets an eective attraction from the repulsion via a quantum
mechanical correlation eect. Attraction in some distance ranges does not
necessarily imply binding, but Kohn and Luttinger suggested that this effect could indeed produce pairing. While no such superconducting instability
appears to be relevant to ordinary metals, the paradoxical theoretical idea
that attraction could result from repulsion is fascinating and the discovery of
high-Tc superconductivity [175] has stimulated a hot discussion on the possibility that something similar is realized in the Cuprates (although the pairs
are singlets with L=2, so some important modication is needed). The fact
that TC can be above liquid N2 (rather than liquid He) temperatures, and the
evidence that this occurs in strongly correlated materials suggests to part of
the community that the mechanism must be dierent from the conventional
phonon-assisted BCS one[181]. Other Authors prefer a conventional, or an
enhanced[182] phonon-assisted mechanism. New superconductors including
doped Fullerenes and Carbon nanotubes[197] complicate the riddle.
Anyhow, strong correlations and phonons are there, so they are likely to
be involved, with a complex trade between the various degrees of freedom;
some modeling is in order. Then, it looks natural to start by the repulsive
trivial Hubbard model (1.64), or some of its variants, in 2 dimensions, while
keeping in mind the claims that such a model cannot superconduct at all [168]
[169] or at least cannot have superconducting long-range order [170]. An effective attractive force comes from second-order eects, e.g., the exchange of
spin uctuations, but it must overcome the strong direct Hubbard repulsion
to give pairing. Bickers and co-workers[162] explored the consequences of a
spin density wave instability within the RPA approximation using the trivial
Hubbard Hamiltonian. They found a spin density wave phase and a superconducting phase with pair-wavefunctions of dx2 y2 symmetry. These ndings
368
were conrmed by using the FLEX approximation[163] which treats the uctuations in the magnetic, density and pairing channels in a self-consistent
and conserving way. Renormalization Group (RG) methods[164] are a well
controlled alternative approach to deal with Fermi systems having competing singularities. The RG has been used by several authors[165],[166],[167] to
study the coupling ows at dierent particle densities. In agreement with the
previous ndings, RG calculations show a d-wave superconducting instability away from half lling driven by the exchange of spin- or charge-density
uctuations.
States of dierent symmetry may be separated by tiny energies per electron, so it is very desirable to have exact data to rely on and observe pairing
or get clear-cut results. The 1d Hubbard model is solved[73] by Bethe Ansatz
techniques (Chapter 18) but does not show pairing. In 2d one can compute
the exact ground state of small clusters by Lanczos techniques (Section 14.1).
A pairing criterion that we shall discuss below was given by Richardson in
the context of nuclear physics[176]. Dening
(N
+ 2) = E(N + 2) + E(N ) 2E(N + 1)
(17.1)
where E(N ) is the N -body ground state energy,
(N
+ 2) < 0
(17.2)
is insignals a bound pair in the ground state with N + 2 particles and ||

terpreted as the binding energy of the pair.
The writer and Balzarotti[171][172] discovered how one can systematically
Fig. 17.1. Pictorial representation of Equation (17.1);the large circles stand for
the interacting N-body system, the black circles represents the added particle, and
the wiggly line is the interaction, which remains after the simplication.
build repulsive models that yield pairing and superconducting ux quantization, and that the two properties have a strong mathematical connection to
each other. Both require special symmetric clusters ( it was known that pairing occurs in the 4 4 system[177] but the simpler examples like CuO4 and
the general case had not been discovered yet) and derived an analytic theory of pairing interactions (W=0 pairing). We reported several detailed case
studies where the formulas are validated by comparison with the numerical
data. The results support general qualitative criteria for pairing induced by
the on-site repulsion only. Our approach successfully predicts the formation
of bound pairs, and explains why other ingredients like strong o-site interactions are needed to force pairing in other geometries. The analytic approach
369
is also needed to predict what happens for large systems and in the thermodynamic limit. Understanding the implications of the small cluster results is
not trivial since the computed pair binding energies show a rapid decrease
with increasing the cluster size; so several authors on the basis of the numerical data on 44 or even 66 clusters wrongly concluded that pairing in the
Hubbard model as a size eect. However, much larger cells (at least 3030)
are needed to estimate the asymptotic behaviour and the analytic treatment
suggests that pairing with a reduced but substantial binding energy persists
in the full plane. Details may be fond in a recent review paper[173].
17.0.3 W = 0 Pairing in Cu-O Clusters
In this Section, we illustrate the concept of W = 0 pairs and the way they
become bound states, by using examples with a geometry relevant for the
Cuprates. Nevertheless, the simple square Hubbard plane and its symmetric
clusters like the 4 4 cluster with periodic boundary conditions also present
W=0 pairing[189]. Moreover, Hubbard models with the geometry of Carbon
CuO4
Cu5 O4
Cu5 O16
Fig. 17.2. Allowed clusters that were used to show pairing and superconducting
ux quantization by exact diagonalization methods in Ref. [172]; the holes were 4,
and the number of congurations involved in the case of Cu5 O16 is 44,100.
Nanotubes (CNT) [188] allow W=0 pairing away from half lling, opening the
possibility that even there the mechanism of superconductivity is basically
electronic [207]. Here we focus on the three-band Hubbard Hamiltonian
H = K + W + Wosite
(17.3)
where introducing the site energies p and d refering to Oxygen and Copper,
the kinetic term is
370
K=t
(pj di + h.c.) + tpp
jj ,
ij,
pj pj + d
ni + p
i,
nj ; (17.4)
j,
here ts are hopping parameters; pj (di ) destroy holes at the Oxygen (Copper)
ions labeled j (i) and n are the number operators; ij refers to pairs of nearest
neighbors. The interaction terms are

W = Ud
ni ni +Up
nj nj ,
Wosite = Upd
ni nj . (17.5)
i
ij,
the positive U parameters represent repulsion, Ud and Up on-site and Upd

o-site on adiacent Cu and O sites. In line with literature estimates[183] we
assume the following values (in eV): p d = 3.5, t = 1.3, tpp = 0.65,
Ud = 5.3, Up = 6 and, most probably, Upd < 1.2. Note that we are using
the same t for all Cu-O bonds, while some authors use more complicated
conventions. For instance, F. C. Zhang and T. M. Rice[184] use an alternating
sign prescription for t, which may be obtained by changing the sign of all the
O orbitals in the horizontal lines containing Cu ions. The two pictures are
related by a gauge transformation, under which the orbital symmetry labels
A1 and B1 are interchanged. Some of the symmetry related information is
gauge dependent and unobservable, while some is physical (e.g. degeneracies
are).
As Balzarotti and the writer[171][172] pointed out, highly symmetric clusters possess 2-holes singlet eigenstates of H which do not feel the on-site
repulsion W ; such eigenstates were called W = 0 pairs and play a crucial
role for pairing. In order to have W = 0 solutions, the clusters must possess
the full C4v (square) symmetry, and must be centered around a Cu site1 . In
particular, the geometries, examined by Hirsch et al.[179], and Balseiro et
al.[178] are forbidden from our viewpoint. The Cu4 O4 geometry considered
by Ogata and Shiba[180] has the C4v symmetry, but lacks the central Cu,
and therefore it is forbidden, too [171].
The aufbau principle is strictly valid for non-interacting electrons, yet
here it is useful in a wide range of parameters to predict the symmetry of
the ground state in the allowed clusters. Concerning the many-body ground
state, an interesting situation arises when the ground state can be thought
of as consisting in W=0 pairs moving on a closed-shell background. W=0
pairs can arise when there are degenerate one-body levels, and a preliminary
analysis can be done assuming that the symmetry Group is C4v (below we
shall discuss more general groups). In the Character Table of the C4v Group
( Appendix 20) we see that there are 4 one-dimensional irreducible representations (irreps) A1 , A2 , B1 , B2 and 1 two-dimensional irrep E. The smallest
1
For large enough clusters, the local physics is well described by an innite plane,
and the cluster shape is no longer important, but such large systems are outside the
scope of diagonalization methods. Yet, our approach emphasizes the importance of
the symmetry in the mechanism and predicts that any important distortion of the
lattice readily destroys pairing.
371
square Cu-O cluster, CuO4 , see Fig. 17.2, is also the simplest2 where W = 0
pairing occurs, and we shall use it as a prototype. The one-body levels and
their symmetry labels are reported in Table 17.1. Let us build a 4-hole state
A
Ex Ey B1

1
A1
2
4 + 0 0 2 + 4 + 2
Table 17.1. One-body levels of the CuO4 cluster in units of t as a function of the
dimensionless parameter tpp /t.
in CuO4 (see Fig. 17.3 b)). The rst two holes go into a bonding level of A1
symmetry; this is a totally symmetric (1 A1 ) pair. For negative , the other
two holes go into a non-bonding level of E(x, y) symmetry, which contains 4
spin-orbital states. Labelling sites as in Figure
17.3 a),

the creation
operators

p3
p1

4
Using the E orbitals according to the Pauli principle one obtains
=6
2
dierent pair-states. The irrep multiplication table allows
for labeling
them
?
?
?
A2 B1 B2 . Proaccording with their space symmetry: E E = A1
jecting according to usual rules we see that A2 is a spin-triplet, 3 A2 , while
the remaining irreps are spin-singlets, 1 A1 , 1 B1 and 1 B2 . One can readily
verify that the B2 singlet operator

1
bB 2 = Ex,
(17.6)
Ey,
+ Ey,
Ex,
2
=
for the degenerate orbitals are Ey
1
2
p 2 p4
=
Ex
1
2
is a W = 0 pair (no double occupation) 3 . To rst order in perturbation

theory, the 4-body singlet state of B2 symmetry is degenerate with the A2
triplet; Hunds rule would have predicted a 3 A2 ground state. However, the
true ground state turns out to be singlet. The numerical results on the CuO4
cluster show that (4)

is negative for 0 > > 0.5 and that its minimum
value occurs at = 0, when the non-bonding orbitals B1 and E become
degenerate: the paired interacting ground state is also degenerate in this
particular case.This accidental degeneracy is is due to the fact that any permutation of the oxygens is a symmetry and the symmetry Group becomes
S4 C4v ; the existence of two dierent W=0 pairs has a deep inuence on
the magnetic properties (Section 17.1 below). In Fig.17.3 c) we plot (4)

for = 0, p d = 0, Upd = 0 and Up = Ud = U .
(4)
has a minimum at U 5 t and it is negative when 0 < U < 34.77 t.
2
< 0 at half lling with 4 fermions in a 4-site cluster, omitting

One can obtain
the Cu site; we regard this as a limiting case with a strongly repulsive central site.
3
Note that 1 B2 is the symmetry label of the pair wave function in the gauge we
are using, and must not be confused with the symmetry of the order parameter.
372

A1
(4)
2
B1 (x2 y 2 )
0
E(x, y)
U
t
.01
12
.02
.03
4
a)
A1
b)
.04
c)
Fig. 17.3. a) The CuO4 unit. b) One-particle energy levels for < 0 with their
symmetry labels and their llings in the 4-particle ground state according to the
aufbau principle. Two fermions belong to the doubly degenerate E(x, y) representation. For 0, the Group becomes S4 C4v and the B1 (x2 y 2 ) level merges
with E(x, y) giving raise to triple degeneracy. Note that among the irreps of C4v
(see the Character Table in Appendix II) A2 and B2 are not represented in this
scheme. Below we shall see that this observation is crucial for W=0 pairing theory.
c)(4)
versus Ut for this cluster is negative in a wide range (it becomes positive
U
for t > 34.77 ) and the maximum pairing occurs at U 5t; this is not a weak
coupling phenomenon.
I emphasize that (4)

becomes positive for large values of U/t and hence
pairing disappears in the strong coupling regime. In the present problem U
must exceed several tens of times t before the asymptotic strong coupling
regime sets in. A perturbation theory will strictly apply at weak coupling
where the second derivative of the curve is negative. However, qualitatively
a weak coupling approach is rewarding in all the physically interesting range
of parameters. The sign of depends on U and and its magnitude is
unlike any of the input parameters. Since the opening Chapters of this book,
dealing with Kondo physics, two-hole resonances, and plasmon eects, we
have appreciated that the occurrence of bound states giving raise to new
energy scales is the most intriguing signature of strong correlation; actually,
it could be its denition.
On the other hand, at positive s the B1 non-bonding level is pushed be
low the degenerate one and (4)
becomes large and positive (at = +0.65,
(4)
= 0.53 eV). The dependence of (4)
on the input parameters has been
studied in detail; also, we have found that any lowering of the symmetry is
reected by a corresponding increase of (4)[171].

We have performed numerical explorations in other fully symmetric clusters like Cu5 O4 and Cu5 O16
and we have found negative values of (4),

of the order of few meV, using
physical parameter values.
373
17.0.4 Pairing Mechanism

In all the allowed clusters up to 21 atoms, the lowest one-hole level belongs to
A1 symmetry, and the next E level yields the W = 0 pair. The interactions
produce a non-degenerate 1 B2 4-hole ground state having the same symmetry
as the W = 0 pair. The interested reader may see Refs.[186],[187] for the
details.
In symmetric clusters as well as in the full plane, (4)

< 0 arises[172] from
an eective attractive interaction between the holes of the W = 0 pair; the
same interaction is repulsive for triplet pairs. In the lowest approximation,
the attraction stems from the second-order exchange diagram.The eective
Fig. 17.4. The second-order exchange interaction that at weak coupling can be the
leading contribution to the eective attraction, depending on the lattice and on the
parameters (see Ref. [172]).
interaction can be found by a canonical transformation (basically a SchrieerWol-type argument) which is long and not easily summarized, so I defer the
reader to the original paper [196]. The results reduce to those of perturbation
theory in the weak coupling case. Moreover, the analytical results compared
semi-quantitatively with those from exact diagonalization of several clusters
at various llings including the 4 4 Hubbard cluster[190]. Being predictive
in strongly correlated problems is the exception, rather than the rule, but
in this case simplicity has emerged from symmetry. Applying the formulas
to supercells of increasing size, it was possible to extrapolate the results to
the thermodynamic limit, showing that < 0 is not a size eect and could
well contribute a pair binding energy in the 10 meV range. The inclusion of
phonon eects in the W=0 scheme was envisaged in Ref. [70].
17.0.5 The W = 0 theorem
A general theorem puts useful restrictions on the many-body ground state
symmetry. Consider a Hubbard model4 H = H0 + W on a lattice , where

tij cj ci
(17.7)
H0 =
ij,
4
that is, a model with on-site, possibly site-dependent, interactions Ui .
374
is the free part and

W =
Ui ni ni ,
(17.8)
with an irrep of G0 . Let G0 be the optimal symmetry Group of H0 , that is, a

Group containing enough symmetry operations that no degeneracy between
one-body eigenstates is accidental. In other terms, G0 must contain enough
operations to justify all such degeneracies by mixing all the degenerate subspaces.
We can construct[185] all the W = 0 pair eigenstates by using projection
operators of G0 . We start by using the irreps of G0 to label all the one-body
eigenstate of H0 . In this way, we are splitting the irreps of G0 according to
the following denition.
Denition. An irrep is represented in the onebody spectrum of H if at least one of the one-body
levels belongs to .
Let E be the set of the irreps of G0 which are represented in the onebody spectrum of H. Let | be a two-body eigenstate of the non-interacting
Hamiltonian with spin S z = 0 and P () the projection operator on the irrep
. We wish to prove the W = 0 Theorem:
/ E W P () | = 0.
(17.9)
Let us rst translate this into words:

- Any nonvanishing projection of | on an irrep
not contained in E, is an eigenstate of H with no
double occupancy. The singlet component of this
state is a W = 0 pair. Conversely, any pair belonging to an irrep represented in the spectrum must
have non-vanishing W expectation value:
Proof : Let us consider a two-body state of opposite spins belonging to
the irrep of G0 :

| () =
() (i, j)ci cj |0 .
ij
Then we have
ni ni | () = () (i, i)ci ci |0 () (i, i)|i , i .
Consider the projection operator P () on the irrep : since

() (i, i)|i , i =
() (i, i)|i , i ,
P ()
i
375
if P () |i , i = 0 i , then () (i, i) = 0 i . It is worth to

note that this condition is satised if and only if P () |i = 0 i ,
where |i
i |0 . Now it is always possible to write |i in the form
= c()
|i =
c
(i)| where | is the one-body eigenstate of H with
E

spin belonging to the irrep . Hence, if
/ E, P ( ) |i = 0 and so

P ( ) |i , i = 0. Therefore, if | () is a two-hole singlet eigenstate of the
kinetic term and
/ E, then it is also an eigenstate of W with vanishing
eigenvalue, that means a W = 0 pair.
Q.E.D.
Remark : Suppose we perform a gauge change in H such that G0 is preserved; clearly, a W = 0 pair goes to another W = 0 pair. Thus, the theorem
implies a distinction between symmetry types which is gauge-independent.
Remark :The complete characterization of the symmetry of W = 0 pairs
requires the knowledge of G0 . A partial use of the theorem is possible if
one does not know G0 but knows a Subgroup G0< . It is then still granted
that any pair belonging to an irrep of G0< not represented in the spectrum
has the W = 0 property. On the other hand, accidental degeneracies occur
with a Subgroup of G0 , and by mixing degenerate pairs belonging to irreps
represented in the spectrum one can nd W = 0 pairs also there.
17.0.6 Examples
CuO4
The CuO4 cluster is a simple example of the W=0 theorem. If
= 0,G0 = C4v ;
of the W = 0 pair belongs to the irrep B2 which is not represented in the
spectrum (see Table 17.1). A2 is also not represented, but projecting any twobody state one nds nothing. If = 0, a three-times degenerate one-body
level exists due to an accidental degeneracy between E(x, y) and B1 orbitals;
with 4 interacting fermions, pairing occurs in two ways, as A1 and B2 are
both degenerate ground states. The extra degeneracy cannot be explained
in terms of C4v , whose irreps have dimension 2 at most. For = 0 any
permutation of the four Oxygen sites is actually a symmetry and therefore
G0 is enlarged to S4 (the group of permutations of four objects). S4 has the
irreducible representations A1 (total-symmetric), B2 (total-antisymmetric),
E (self-dual), T1 and its dual T2 , of dimensions 1, 1, 2, 3 and 3, respectively.
These irreps break in C4v as follows
A1 = A1 , T1 = B1 E, T2 = A2 E, B2 = B2 , E = A1 B2 .
Labelling the one-body levels with the irreps of S4 , one nds that E is not
contained in the spectrum and thus yields W = 0 pairs:
1
E =
A1
B2 .
(17.10)
376
The corresponding pair-creation operators reads:

2
1
= B1
B1 + Ex
Ex + Ey
Ey
bA
1
3
3
(17.11)
for the total-symmetric pair and Eq. (17.6) for the B2 component.
The 4 4 Hubbard cluster with periodic boundary conditions
Energy Irrep of G Degeneracy

2
4
1
B
2
4
4
0
4
6
-2
1
4
-4
A1
1
Table 17.2. One-body spectrum for t = 1. The irrep symbols will be explained
shortly.
The one-body spectrum of the model (see Table ( 17.0.6)) involves a 6

times degenerate shell, whereas the Space Group G does not have irreps with
dimensions bigger than 4 (see Ref.[156]). The optimal Group G0 is larger than
the Space Group and includes symmetries which are not isometries[190].
Consider the transformation d of the 4 4 lattice shown in Equation
(17.12):
1 2 3 4
5 1 4 8
5 6 7 8
6 2 3 7
d:
9 10 11 12
10 14 15 11
13 14 15 16
9 13 16 12
(17.12)
Because of the periodic boundary conditions, the nearest neighbours of 1 are

2, 5, 4 and 13. The dynamical symmetry operation d can be obtained by
rotating the plaquettes 1,2,5,6 and 11,12,15,16 clockwise and the other two
counterclockwise by 90 degrees. This preserves nearest neighbours (and so,
each order of neighbours) but is no isometry ( for example, the distance between 1 and 3 changes). Thus, this symmetry operation d is a new, dynamical
symmetry. Including d and closing the multiplication table one obtains the
required Optimal Group G with 384 elements in 20 classes (like G) as shown
in Table .
377
C1
C2
C3
C4
C5
C6
C7
C8
C9
C10
I
t22 C4 [2] x C2

C2 d C2 t22 d C2 [1] C43 t02 d
C11
C12
C13 C14 C15
C16
C17
C18
C19
C20
C43 t20 d C2 t01 C2 x [1] C4 C2 t01 d C2 t12 d C2 x [1]d C4 [1]d C2 [1]d C4 [1]
Table . Here, we report one operation for each of the 20 classes Ci ; the others can
be obtained by conjugation. The operations are: the identity I, the translation tmn
of m steps along x and n along y axis; the other operations C2 , C4 , , are those of
the Group of the square and are referenced to the centre; however,C2 [i], C4 [i], [i],
and [i] are centered on site i.
The complete Character Table of G is shown below.
G
A1
A1
B2
2
B
1
2
1
2
3
4
1
2
3
4
1
2
3
4
1
2
C1
1
1
1
1
2
2
3
3
3
3
4
4
4
4
6
6
6
6
8
8
C2
1
1
1
1
2
2
3
3
3
3
-4
-4
-4
-4
6
6
6
6
-8
-8
C3
1
1
-1
-1
-2
2
3
3
-3
-3
-2
-2
2
2
0
0
0
0
-4
4
C4
1
1
-1
-1
-2
2
3
3
-3
-3
2
2
-2
-2
0
0
0
0
4
-4
C5
1
1
1
1
2
2
3
3
3
3
0
0
0
0
-2
-2
-2
-2
0
0
C6
1
1
1
1
2
2
-1
-1
-1
-1
0
0
0
0
-2
-2
2
2
0
0
C7
1
-1
1
-1
0
0
-1
1
-1
1
-2
2
-2
2
-2
2
0
0
0
0
C8
1
-1
1
-1
0
0
-1
1
-1
1
2
-2
2
-2
-2
2
0
0
0
0
C9
1
1
1
1
2
2
-1
-1
-1
-1
0
0
0
0
2
2
-2
-2
0
0
C10
1
-1
-1
1
0
0
-1
1
1
-1
-2
2
2
-2
0
0
-2
2
0
0
C11
1
-1
-1
1
0
0
-1
1
1
-1
2
-2
-2
2
0
0
-2
2
0
0
C12
1
-1
1
-1
0
0
-1
1
-1
1
0
0
0
0
2
-2
0
0
0
0
C13
1
-1
-1
1
0
0
-1
1
1
-1
0
0
0
0
0
0
2
-2
0
0
C14
1
1
-1
-1
-2
2
-1
-1
1
1
0
0
0
0
0
0
0
0
0
0
C15
1
1
1
1
-1
-1
0
0
0
0
-1
-1
-1
-1
0
0
0
0
1
1
C16
1
1
1
1
-1
-1
0
0
0
0
1
1
1
1
0
0
0
0
-1
-1
C17
1
1
-1
-1
1
-1
0
0
0
0
-1
-1
1
1
0
0
0
0
1
-1
C18
1
1
-1
-1
1
-1
0
0
0
0
1
1
-1
-1
0
0
0
0
-1
1
C19
1
-1
1
-1
0
0
1
-1
1
-1
0
0
0
0
0
0
0
0
0
0
C20
1
-1
-1
1
0
0
1
-1
-1
1
0
0
0
0
0
0
0
0
0
0
Character Table of the Optimal Group G of the 4 4 model.
As the notation suggests, the irreps A1 and A1 both reduce to A1 , in C4v ,

2 both reduce to B2 . One can easily determine how the irreps
while B2 and B
of G split in C4v (see problem17.1). We rightly called G the Optimal Group
378
because no accidental degeneracy of orbitals occurs using G. All the W = 0

pairs can be found by projecting on the appropriate irreps, determined using
the W = 0 Theorem(see Problem 17.2).
17.1 Superconducting ux quantization and Josephson

eect
Magnetic ux can be trapped in the hole of a hollow superconducting cylinder
with thick walls compared to the penetration depth and persist in the absence of an applied eld. As explained by Schrieer[181] the ux quantum is
hc
7
gausscm2, (see 16.19) but the ux quantum in superconductors
e = 410
hc
is actually 2e . The Superconducting Flux Quantization (SFQ) is a signature
of superconductivity, since it clearly implies that the external magnetic eld
excites diamagnetic currents in a system where the carriers have charge 2e.
The perturbation due to the eld must be gentle enough, otherwise the superconductor becomes normal; the eld is screened by supercurrents at the
surface of the superconductor and its perturbation on the system cannot be
too strong otherwise the supercurrent exceeds the critical current.
The above models with on-site repulsive interactions and W=0 pairing
give this signature. The vector potential A enters through the Peierls prescription (16.18)

2i
tab tab e 0
Adr
In small clusters, W=0 pairing requires full C4v symmetry; moreover, the ux
tubes must be inserted in such a way that the symmetry is not distorted. We
must arrange the thought experiment in such a way that the perturbation
due to the eld is gentle, and does not destroy the pairing. In small clusters,
that we may stud more easily, the current carrying bonds must be so weak
that the current screening the eld is not excessive and the change in the
or less. When this is
ground state energy due to the eld is of the order
done, the SFQ already obtains with a few pairs and even one pair yields a
well dened double-minimum pattern. This means that although in textbooks
SFQ is explained in terms of the order parameter, it is basically a single-pair
symmetry property. Going to the thermodynamic limit is necessary to achieve
macroscopically large barriers that localize the system in one of the minima.
The simplest case is CuO4 and the insertion of ux tubes is shown in
Figure 17.4 a). In order to preserve the symmetry, a ux is inserted in each
of the four Cu-O-O triangles or plaquettes of the cluster. The dotted bonds
must represent negative hopping integrals tpp of the order of 50 meV or less,
otherwise the pairing mechanism breaks down and paramagnetism prevails.
Figure 17.4 b) shows the ground state energy of the cluster versus . The
ux dependence was computed with p d = 3.5, t = 1.3, Ud = 5.3, Up = 6.
This pattern is characteristic of SFQ. The system is diamagnetic at small
17.1 Superconducting ux quantization and Josephson eect
379
a)
E4 () (eV)
Ud = 5.3, Up = 6 eV
Ud = Up = 0
E4 () (eV)
8.006
5.15
1
4
1
2
3
4
7.986
5.05
7.966
1
4
1
2
b)
3
4
1
c)
Fig. 17.5. Correlation-induced ux quantization in the CuO4 cluster. a) The CuO4

unit with 4 inserted ux tubes . A ux is inserted in each of the four Cu-O-O
triangles or plaquettes of the cluster. b) Flux dependence of the ground state energy of the cluster with the standard parameter values p d = 3.5, t = 1.3, Ud =
5.3, Up = 6 but tpp = 0.05eV. Note that the system is diamagnetic at small ux
and exhibits the characteristic minimum at half uxon. c) Ground state energy of
the cluster with Ud = Up = 0. without correlation the cluster shows a normal,
paramagnetic behaviour. Similar phenomena have been observed in exact diagonalization work with larger clusters (up to 21 atoms) in Ref. [172] and with rings of
clusters.
Fig. 17.6. Rings of clusters threaded by ux tubes allow to study the propagation
of bound W=0 pairs. The dotted line represents hopping between two consecutive
clusters that must be designed in such a way that the square symmetry of each
cluster is preserved.
380
(its energy increases with ux) but reaches a maximum due to a level crossing
and goes trough a second minimum at = 20 ; = 0 is equivalent to = 0.
The two ground states at = 0 and = 20 are of comparable depth and are
due to W=0 pairs of dierent symmetry.
Fig. 17.7. Inserting 4 ux tubes, each carrying a ux in a CuO4 unit in the plane,
the magnetic ux through any path in the plane must equal times the number
of tubes encircled by the path. This is achieved by choosing the vector potential
according to the above pattern, with A dr = 0 along each bond without arrow,

A dr = 2 along each oriented bond with the integration path parallel to the
arrow. For a generic ux in each tube any reection of C4v reverses the arrows
( ) and thus C4v is broken and only the commutative Z4 Group of the
rotations survives. W=0 pairs are forbidden in the eld and pairing is destroyed.
However, at = 20 , one nds A dr = 40 and the Peierls prescription is t it
on any bond with arrow; now any reection implies which is just a gauge
change, replacing it by it. This is equivalent to a unitary transformation S that
changes the signs to all the Cu orbitals along the diagonals except the central one.
Since S reverses the arrows, S is a symmetry.
Indeed, as we saw earlier, two W=0 irreps exist for tpp = 0. SFQ is a
correlation eect and disappears for Ud = Up = 0; without correlation the
cluster shows a normal, paramagnetic behaviour, shown in Figure 17.4 c)
Similar phenomena have been observed in exact diagonalization work with
larger clusters (up to 21 atoms) in Ref. [172] and with rings of clusters. Larger
systems can be obtained by joining several clusters in such a way that bound
pairs can propagate. Rings of clusters threaded by ux tubes also show SFQ,
and Figure 17.5) illustrates the general idea of how such systems can be
17.1 Superconducting ux quantization and Josephson eect
381
built,in such a way that the square symmetry of each cluster is preserved.
The SFQ pattern is quite similar[18],[?] to the one discussed above; this time
the inter-cluster hopping must be a weak link in order to make the magnetic
perturbation gentle enough. By projecting on the low energy sector, we can
solve very large clusters in an essentially analytic way and show that the
superconducting behavior remains independent of size.
This suggests that the SFQ pattern must be a property of the Hubbard
model which extends to the thermodynamic limit and is essentially related
to the symmetry. The minimum at = 0 is clearly due to the W=0 pair,
yielding a nondegenerate ground state (when there s degeneracy, the ground
state must be paramagnetic, since the system interacts with the eld by a
Zeeman mixing of the degenerate components and gains energy). The eld
tends to destroy the pairs and does so by lowering the symmetry; indeed the
reections are broken. The second minimum at = 20 is due to the lattice
which for this particular ux recovers the ux symmetry. This is actually a
theorem, and the proof was obtained by a construction shown in gure 17.6
. One can see how one can choose the gauge for the vector potential when the
4 ux tubes are inserted like in gure 17.4 in a CuO4 unit belonging to a full
plane. From the construction in the Figure one sees that a general ux in the
ux tubes destroys the W=0 pairing by breaking the symmetry (C4v Z4 )
but at = 20 the full symmetry is restored, because the reections are
equivalent to a gauge transformation. The ux quantization pattern arises
from a level crossing between the ground states at = 0 and = 20 .
Models systems capable of superconducting ux quantization such as rings
of CuO4 clusters can also be operated[192] as Josephson junctions. One needs
to include a barrier and introduce a mild violation of the symmmetry to
produce avoided crossings; Josephson and inverse-Josephson (Shapiro) eects
then appear as adiabatic responses to slowly changing ux (t).
Recently, exact studies on small Hubbard clusters at nite temperatures
by Fernando et al.[193] have given useful insight on phase separation and
pairing.
Problems
17.1. Find how the irreps of the optimal Group G split in C4v .
17.2. Using the W=0 theorem and the Optimal Group, nd the irreps with
no double occupation in the 4 4 lattice .
18 Algebraic Methods
18.1 Lieb Theorems on the Half Filled Hubbard Model

Consider a repulsive Hubbard model, dened on a d-dimensional lattice or
graph of || sites, with nearest-neighbor
hopping.
No symmetry is assumed;
however we can use as an example a || || simple square Hubbard
Model with periodic boundary conditions and hopping between nearest neighbors (n.n.):
H =T +W =t
n.n.

x,y
cx cy + U
nx nx ;
U > 0.
(18.1)
Let us specialize at half lling with

n N = N = ||
2 electrons for each spin.
||
For each spin, there are m =
congurations, and the dimensionality of
n
the problem is m2 . Here I present a simplied proof that the Hubbard Model
at half lling has a unique ground state; this could serve as an introduction to
Liebs paper [37] which contains the proofs of more general exact statements.
For the total spin, which is conserved, the second quantization operators
(1.57,1.58) become in the lattice case
Sz =

1
(n (x) n (x)) , S + =
c (x)c (x),
2 x
x
(18.2)
and one can work out

1
S 2 = Sz2 + (S + S + S S + ).
2
A symmetric determinantal conguration like
|sites = |i , j , k , . . . , i , j , k , . . . |
(18.3)
in the site representation has S=0 and is singlet. A two-site system is the
most trivial bipartite lattice; with two electrons it becomes the model system
that we met in Sect. 14.2; there, I used it for illustrating the form (14.51),
384
| =
L | |
(18.4)
of the many-electron the wave function. The eigenvector given by the matrix
L = i y2 represents the only triplet, and has nothing on the diagonal. Out

4
of the
= 6 states of the model, three are singlets. At half lling L is a
2
square m m matrix. A glance to (18.4) suggests that there would be much
to gain in complex problems if L were diagonal, because fewer numbers would
represent | , and one knows that L can be diagonalized if it is Hermitean.
If L can be taken Hermitean, it remains Hermitean when changing the basis.
Thus it is natural to ask if this advantage is free or we must pay for it.
Hermitean L
There is no magnetic eld, the Hamiltonian
is real and we take the basis
functions real as well. Then, if = , L, is an eigenstate of
H with eigenvalue , L also belongs to ; the transposed matrix LT has the
up and down spin amplitudes interchanged, and since this is a symmetry
operation it leaves the subspace of invariant. Hence, if L belongs to ,
this applies to L and for the Hermitean combinations L + L , i(L L ).
So we shall take L Hermitean1 . Hermiticity implies that we may impose the
normalization condition (14.60) | = 1 by writing T rL2 = 1. If all the
eigenvalues of L are non-negative, we write L 0. Then, we must have
T rL > 0 strictly, because otherwise L = 0 and the state vanishes. One can
infer about the spin of the many-body state2 if L 0. In fact, some diagonal
elements must be positive, and this holds true on the site basis like in (18.3);
therefore, is singlet, or has some mixture with a singlet. In the two-site
two-electron example one can see that for large negative U the eigenvalues of
L are of the same sign, while for large positive U they are of opposite signs:
attraction favors singlets.
In his proof, Lieb used the Scr
odinger Equation in the form (14.56):

ns Lns .
(18.6)
H : L K L + LKT + U
s
He had obtained (14.54)(for the real Hamiltonian case) from the variational
principle starting from the energy
1
As a two-electron example, writing x, y for sites or orbitals, we may use the

singlet
|x y | + |y x |
x
;
(18.5)
L=
=
2
2
incidentally, this shows that the vanishing of the diagonal elements does not prevent
L from representing a singlet. However a singlet can give a non-zero contribution
to TrL while triplets and higher spins cannot.
2
It is not the sign of the wave function that brings information, of course, but
a non-vanishing TrL.
18.1 Lieb Theorems on the Half Filled Hubbard Model
E = 2T r(KLL) + U
T r(Lnx Lnx ).
385
(18.7)
Canonical Transformation and Pseudospin

Electron-electron repulsion is the same as electron-hole attraction; by a
canonical transformation starting from the model (18.1) we now get a Hamiltonian with electron-electron attraction. We need to change the up spin operators only, according to
(18.8)
dx = c ,
whereby the transformed kinetic energy for up-spin electrons reads
= t
K K
n.n.
dx dy
x,y
n.n.
= t
dy dx .
x,y
Having t for down spins and t for up spins is denitely ugly; so we change
signs to the orbitals in a sublattice (here the bipartite hypothesis enters)
replacing (18.8) by
dx = sign(x)cx .
(18.9)
Now, the transformed Hamiltonian reads
=t
H

n.n.

dx dy + cx cy U
dx dx nx + U N ; U > 0, (18.10)
x,y
where N = x nx is conserved. At half lling, we keep dealing with N =

N = ||
2 Fermions for each spin, and the amplitudes L in a site picture are
the same, except that is replaced by the conguration with the sites not
in . For example, the repulsive H2 molecule model
H(t, U ) (14.58) has the
Ur
energy eigenvalues (U ) = Ur
,
0,
U,
,
r
=
16t2 + U 2 ; the eigenvalues
2
2
of
0 t t0
= H(t, U ) + U = t U 0 t , U > 0
H
(18.11)
t 0 U t
0 t t0
are the same as those of H(t, U ).
With the above substitution of d for c, Sz of Equation (18.2) becomes
1
1
Sz =
([1 dx dx nx ]) = [ N N ]
2 x
2
and S + reads
S+ =
(x)dx cx .
(18.12)
(18.13)
Both are still conserved of course although they do not mean spin any more
in the transformed problem and are therefore called pseudo-spin.
386
Unique Ground State: Attractive Case

Let us assume H as in (18.1) but U < 0. If the Hermitean L matrix is a
ground state, then we can diagonalize L and write C LC = diag(i ), where
C is an eigenvector matrix, and i are the eigenvalues. Thus we can form a
positive semi-denite matrix |L| = diag(|i |), which is granted to be singlet
or a linear combination including a singlet.
Now it is easy to show that |L| is also a ground state. To this end, we
calculate the energy accordingto (18.7) on a basis of eigenvectors of L. For
the kinetic energy T rKL2 = i Kii 2i the result
is the same for L and |L|.
For the potential energy the result for L is U x,i i j (nx )2i,j and for |L|

U x,i |i j |(nx )2i,j ; since U<0 the latter result cannot be larger than the
former, thus
case (18.5)
one nds
|L|is a ground state. In the
two-electron

1
1
1
0
10
1
1
1
C= 2
and gets C LC = 2
, hence |L| = C 2
=
1 1
0 1
01

10
; this yields the ground state
C = 12
01
=
|x x | + |y y |
,
2
(18.14)
which is degenerate with (18.5) for U = 0 but is actually better for U < 0.
The question arises: is |L| really dierent from L? If it is, we get a new
ground state, namely the dierence, which we write in terms of the eigenvectors:

|i [|i | i ] i |.
(18.15)
R = |L| L =
i
Being a ground state, R > 0, unless R = 0. We now show that this is the case,
that is, |i | = i i. The latter equality must be true for some i, otherwise
L = |L| and we obtain the same ground state with a uninteresting (-)
sign in front; thus, the kernel of R is not empty, that is, RV = 0 for some
m-component vector V . The Schr
odinger equation (18.6) for R

K R + RKT + U
ns Rns = R.
(18.16)
s
averaged over V has no kinetic terms because RV = 0; it follows that

V |nx Rnx |V = 0.
x
This implies that either nx |V = 0 or nx |V is also in the kernel of R. Since V

does not vanish, nx |V
= 0 for some x, then the kernel extends to nx |V . The
proliferation of the kernel never stops; applying the Schrodinger equation to
V,

nx Rnx V = RV.
(18.17)
KR + RK + U
x
18.2 Bethe Ansatz for the Heisenberg Chain
387
Now since RV = 0 and Rnx V = 0 we nd RKV = 0, that is, every conguration connected to V by K is in the kernel; but any two conguration can
be reached from each other in a nite number of steps, that is K p has matrix
elements between them for some p. Finally the kernel has swallowed all the
Hilbert space! Thus, L = |L| (although, of course, we remain free to choose
L = |L|. Thus there is a singlet component in the ground state and L 0.
This implies that the ground state is unique. If L1 and L2 were dierent
ground states any combination L = L1 + L2 would be a ground state, but
we would be free to choose such that T rL = 0; this is impossible since any
ground state must be positive semidenite.
Repulsive Case
Now we assume H as in (18.1)with U > 0. The ground state is the transform
of the one with U < 0. (See Problem 18.1 for an example). This implies that
the eigenvalue is the same; the eigenvector is also the same, albeit it is written
in a reshued basis, and is therefore a unique singlet.
More generally, one can study bipartite lattices with sublattices A and B
with |A| and |B| sites per cell. Lieb showed that the ground state has spin
S = |B| |A| per cell; thus, the trivial Hubbard model has a singlet ground
state, while the three-band Hubbard model has S = 12 per cell.

For an electron in magnetic eld in classical canonical ensemble,

(px eHy/c)2 + p2y + p2z

] = V (2mKB T )3/2
2mKB T
(18.18)
is actually independent of the eld H, since the eld dependence can be eliminated by a change of variable; this extends to any system. Thus the very existence of magnetism cannot be understood classically. The Heisenberg model
was one of the rst quantum mechanical theories aimed at describing ferromagnetism. Consider N spins s = 1/2 arranged on a ring (SN +1 = S1 ); n =
and n = are the two possible states at site n, and a conguration may be
specied by assigning |1 , 2 , , N ; the Hilbert space has 2N dimensions.
The Heisenberg Hamiltonian is
Z=
d3 x
d3 p exp[
HH = J
N
Sn Sn+1 = H0 + HXY ,
(18.19)
n=1
H0 = J
N

n=1
z
Snz Sn+1
(18.20)
388
while the so-called XY model hamiltonian, which has been studied by its own
sake, reads
N
"
J ! +
+
HXY =
Sn Sn+1 + Sn Sn+1
.
(18.21)
2 n=1
Let N r be the number of spins; HH is invariant for translations along
the chain (which implies conservation of a momentum K) and for rotation
in spin space (conservation of the total spin S and of its z component, Sz =
N N
, N = N N . For r = 0 the problem is trivial because
2
HH | . . . = E (N =0) | . . . , E (N =0) = J
N
4
(18.22)
and this is the only state. Here I assume ferromagnetic coupling, i.e. J > 0
for deniteness. Note that
H0 |1 , 2 , , N = {E (N =0) +
(J)
(2)N }|1 , 2 , , N
4
(18.23)
where N is the number of reversed scalar products (number of times or

nearest neighbors are encountered instead of or ); the (2) factor enters
because the scalar products are counted as positive in E (N =0) and must be
reversed. We may think of HXY as moving spins to the left or to the right
provided that the neighboring spins are ; this important limitation causes
the interaction between excitations and makes the problem interesting. The
method invented by Bethe in 1931 [68] to solve the Schrodinger equation for
this model for any N and r has been later applied with great success to several
many-body problems. Excellent reviews of recent work are available[72]; here
I show its principle.
One Magnon
For r N = 1 a conguration can be written |n , where n is the only
spin; since N = 2, n|H0 E (N =0) |n = J, n|HXY |n 1 = J2 and
(HH E (N =0) )|n = J|n
J
(|n + 1 + |n 1 ).
2
(18.24)
Taking the scalar product f (n) = E |n with the solution of H|E =

E|E , one nds the recurrence relation
(E E
(N =0)

f (n + 1) + f (n 1)
f (n)
)f (n) = J
2

(18.25)
which is readily solved to give the magnon solution

N
1 ikj n
E j =
e
|n , E = E (N =0) + J [1 cos(kj )]
N n=1
(18.26)
389
with the wave vector
2j
, j integer
(18.27)
N
reecting the translation invariance of the problem. This represents a spin
wave excitation of a ferromagnet.
kj =
Two Magnons
For N = r = 2 it is natural write a basis {|n1 , n2 }, n2 > n1 specifying the
positions of spins. If n2 > n1 + 1, N = 4 and one nds:
(HH E (N =0) )|n1 ,n2
= 2|n1 , n2
J
|n1 +1,n2 +|n1 1,n2 +|n1 ,n2 +1+|n1 ,n2 1
;
+
2
(18.28)
let us take the scalar product f (n1 , n2 ) = E |n1 , n2 with the solution of
H|E = E|E , and write the recurrence formula:
(E E (N =0) )f (n1 , n2 )
= 2f (n1 , n2 )
J
f (n1 + 1, n2 ) + f (n1 1, n2 ) + f (n1 , n2 + 1) + f (n1 , n2 1)
(18.29)
.
+
2
This is satised by f (n1 , n2 ) = ei(k1 n1 +k2 n2 ) with energy eigenvalue
E = E (N =0) + J(1 cos(k1 ) + 1 cos(k2 ))
(18.30)
that would be the same as the sum of two magnon energies and would represent the superposition of two magnons if the momenta ki were given by
(18.27). However, the periodic boundary conditions do not apply separately
to the two momenta. Instead, we must have
or also
ei(k1 +k2 )N = 1,
(18.31)
eik1 N = ei , eik2 N = ei .
(18.32)
The interaction of the two magnons produces a relative phase shift, which
also changes the energy although the eigenvalue (18.30) is formally the sum
of two magnon energies. In terms of this (yet unknown) phase, we may write
N k1 = + 21 , N k2 = + 22 .
(18.33)
The Bethe quantum numbers i are integers and range from 0 to N 1.

However, one must pay more attention. First, there is a degenerate solution with the k1 and k2 exchanged and must be included for generality and
also to satisfy the boundary conditions, as I show in a moment; so one should
write
f (n1 , n2 ) = Aei(k1 n1 +k2 n2 ) + Bei(k2 n1 +k1 n2 ) , 0 < n1 < n2 < N.
Note that:
(18.34)
390
the ordering 0 < n1 < n2 < N is needed otherwise the basis would be
over-complete.
The two contributions must enter with the same probability, A and B
dier by a phase (I have the right to use the same symbol, as I show
shortly) and we rewrite (18.34) as
f (n1 , n2 ) = ei(k1 n1 +k2 n2 + 2 ) +ei(k2 n1 +k1 n2 2 ) , 0 < n1 < n2 < N (18.35)
This is not normalized but we do not care.

may be identied with the same real phase as in (18.32) above. Indeed,
we must be free to decide that the numbering of spins runs in the range
(n2 , n2 + N 1) rather than (1, N ); now n2 < n1 + N ; therefore
f (n1 , n2 ) = f (n2 , n1 + N )
=e
i{k1 n2 +k2 (n1 +N )+ 2 }
+ ei{k2 n2 +k1 (n1 +N ) 2 } ,
n2 < n1 + N.
(18.36)
This shift of the origin exchanges n1 and n2 , and the two terms in (18.35)
are therefore exchanged; the second term of (18.35) becomes the rst of
(18.36), but the phase /2 changes sign and an extra N k2 term appears in the exponent. These dierences must cancel because the boundary condition holds. Comparing with ( 18.35 ) we nd the conditions
ei 2 eik2 N = ei 2 , ei 2 eik1 N = ei 2 , which are satised, according to

(18.32); is indeed the same.

, when n2 =
n1 n2
n1 + 1 and the down spin in n1 cannot jump to the right and the one in n2
cannot jump to the left; Equation (18.29) must change because N = 2 and
f (n, n) does not occur in the equation since |n, n has no meaning, therefore
It is time worry about congurations like
(E E (N =0) )f (n1 , n1 + 1)
=
J
f (n1 1, n2 ) + f (n1 , n1 + 2)
= f (n1 , n1 + 1) +
.
2
(18.37)
However the ansatz is still (18.35); the ansatz allows f (n, n)

= 0 and grants
that (18.29) continues to hold formally for n2 = n1 + 1 as well,
(E E (N =0) )f (n1 , n1 + 1)
f (n1 + 1, n1 + 1)
= 2f (n1 , n1 + 1) +
J
2
f (n1 1, n1 + 1) + f (n1 , n1 + 2) + f (n1 , n1 )
.
(18.38)
+
2
Comparing the last two equations we get the condition
f (n1 , n1 + 1) =
1
[f (n1 , n1 ) + f (n1 + 1, n1 + 1)].
2
(18.39)
391
When we substitute the ansatz (18.35),we are free to set n1 = 0 to obtain

more readily the n1 -independent condition
ei(k1 +k2 ) + 1 2eik1
N
(18.40)
.
i(k
+k
)
ik
1
2
2
D
e
+ 1 2e
One moment, is this is really a phase factor? To check that, one observes
that |N|2 = 6 + 2 cos(k1 + k2 ) 6[cos(k1 ) + cos(k2 )] is symmetric in k1 and
k2 , hence |N| = |D|. Now the identity
ei =
ei + 1
cot( ) = i i
2
e 1
(18.41)
is very useful to cast (18.40) in real form:
k1
k2
2 cot( ) = cot( ) + cot( ).
2
2
2
(18.42)
The Bethe Ansatz equations (18.33 , 18.42), can be solved for k1 , k2 and
by combined analytic and numerical techniques for N some tens[72]. Most
solutions are scattering states with real momenta and a nite ; however when
one of the Bethe quantum numbers vanishes, the corresponding k and
also vanish, and there is no interaction. Other solutions with 1 and 2 close
to each other have complex momenta with k1 = k2 and represent magnon
bound states in which the two ipped spins propagate at a close distance.
Many Magnons
It is stunning that the Bethe ansatz works for any system size, and is generalized for any r. Again one chooses the origin and orders the spins with
ni < ni+1 ; a basis is {|n1 , n2 . . . nr }, specifying the positions of spins. We
rst seek a possible H eigenstate as a product of r magnon wave functions of
momenta kj . As the natural extension of (18.32) we allow each k to pick up
a phase from each of the others;

i
eiki N = e j=i ij , ij = ji .
(18.43)
Equation (18.33) becomes
N ki =
ij + 2i
(18.44)
j=i
and the integers i ranging from 0 to N-1 are Bethe quantum numbers. The
r! permutations P of the momenta yield degenerate r-magnon plane waves
that must be summed,so the Bethe Ansatz reads
!i r kPj nj + i PiPj "
2
j=1
i<j
(18.45)
e
f (n1 , . . . , nr ) =
P
392
Again, we must consider two cases: 1) the reversed spins are all far away
2) there are neighboring reversed spins. In case 1) Schrodinger equation is
satised identically and yields the direct extension of (18.30) for the energy
eigenvalue:
r

(1 cos kj ).
(18.46)
E E0 = J
j=1
We can count our spins starting anywhere without changing the amplitude,
so we impose
f (n1 , , nr ) = f (n2 , , nr , n1 + N ).
This causes a permutation of the terms of the sum, and in each exponent the
term changes sign and an extra contribution proportional to N appears.
Exactly as in the two-magnon case, these changes are xed by (18.44) and
we may conclude that the phases are the same in (18.44) and (18.45).
Some changes occur in the Schr
odinger equation when there are neighboring reversed spins because N is reduced and some hoppings do not exist; the
same reasoning that took us to (18.40) can be extended directly and yields
eiij =
ei(ki +kj ) + 1 2eiki

.
ei(ki +kj ) + 1 2eikj
(18.47)
which in real form can be written, as above,

2 cot
ij
ki
kj
= cot cot ,
2
2
2
i, j = 1, . . . , r.
(18.48)
The Bethe equations (18.44,18.48) are hard to solve for many ipped
spins in large systems. In recent years solutions have been obtained which
are far from trivial; among others, bound states of several magnons have
been reported.
The Bethe Ansatz is a rare example of exact solution of an interacting
many-body problem; it keeps the same form independent of the size of the
system. The root of the (relative) simplicity that allows this solution is onedimensionality: the evolution does not allow overtakings, and spins always
keep a xed order. However, without the ingenuity of Hans Bethe, the solution
could have remained undetected; who knows how many important model
problems are solvable, but still unsolved.
18.3 Bethe Ansatz for Interacting Fermions 1 Dimension

18.3.1 -Function Interaction
Let N electrons in 1 dimension on a line of length L whose ends are identied
interact with short-range repulsion. We seek the space wave function belonging to a given irrep of SN , having separated the spin variables according to
393
Sect. 9.8. The Hamiltonian, which can be shown [142] to be related to the
Kondo Hamiltonian, is:
N

2
H=
+ 2c
(xi xj ),
2
xi
i=1
i<j
c > 0.
(18.49)
Here x1 . . . xi . . . xN are the electron coordinates, in short {xi }. Indeed, as in

Section 1.1, one pretends that rst of all the electrons receive a priori labels
1, 2, N ; one runs though the line and notes down where one meets electrons
and in what order. The order denes a permutation Q SN ; the i-th electron
along the line is the one bearing the a priori label Qi , we label its position
xQi , and say that we are in in sector Q. The Bethe ansatz solution[74] yields
the many-electron wave function in the sector, which is just the ordering
sector Q : 0 < xQ1 < . . . < xQN
(18.50)
It is:
(x1 . . . xi . . . xN ) = Q (x1 . . .
xi . . . xN ), 0 < xQ1 < . . . < xQN ,
. . . xi . . . xN ),
Q (x1 . . . xi . . . xN ) = P Q,P (x1
Q,P (x1 . . . xi . . . xN ) = [Q, P ]ei
kPi xQi
(18.51)
Here, P SN while k1 . . . kN denote dierent numbers, and [Q, P ] are amplitudes to be determined.
Thus each electron has 2 labels, the physical one is i, which is the order
along the line, while Qi depends on a ctitious (for indistinguishable particles)
and unobservable order, that is on the sector. However, if we remain in a
sector Q, we cannot describe any overtaking: when an electron i overtakes
the one on its right, the description automatically switches to a new sector.
Qi1
Qi Qi+1
Qi1
Qi+1 Qi
i1
i1
i+1
i i+1
i
Fig. 18.1. If a classical electron (triangle) overtakes another one they exchange
their physical labels and keep their sector labels. Since actual electrons are indistinguishable, one ends up in a neighbor sector Q = P (ii+1) Q, where P is a
transposition.
Now we study under what conditions

H =
N
ki2
i=1
and how one can determine the momenta ki .
(18.52)
394

Thus, Q =
P Q,P , where QP is [Q, P ] times a plane-wave state
where the ordering of electrons sector labels Q and the ordering of momenta
are xed. We are in a given sector Q as long as xQi increases as i increases,
but if two electrons with consecutive physical labels sit at the same x, we
are also in the sector Q = P (ii+1) Q, in which the images of i and i+1
are exchanged. This fact allows to impose sector boundary conditions: at the
boundary each plane wave must have the same amplitude in both neighboring
sectors. For example, in the Q sector the contribution to Q proportional to
[Q, P ] when xQ3 = xQ4 is

i(k +k )x +i
pPi xQi
i=3,4
Q,P = [Q, P ]e P3 P4 Q3
;
(18.53)
both in Q and in Q the contribution of P is summed to the contribution
of the permutation P = P (34) P bringing the same exponential:

i(k +k )x +i
pPi xQi
i=3,4
Q,P + Q,P = ([Q, P ] + [Q, P ])e P3 P4 Q3
. (18.54)
This amplitude must be the same as
Q ,P + Q ,P = ([Q , P ] + [Q , P ])e
i(kP3 +kP4 )xQ3 +i

i=3,4
pPi xQi
and the continuity condition is

[Q, P ] + [Q, P ] = [Q , P ] + [Q , P ].
(18.55)
Next, we integrate the Schr

odinger equation for over xQ3 at the sector
boundary, recalling that xQ3 xQ4 by denition:
x Q4
x Q4
dxQ3 H({xQi }) = E
dxQ3 .
(18.56)
xQ4
xQ4
Since this is O() and goes to 0,

,
+
x Q4
2
dxQ3 2 + 2c(xQ3 xQ4 ) ({xQi }) = 0,
xQ3
xQ4
and using
dx(x) =
1
2
one nds a discontinuity in the derivative of the full amplitude of (18.51):
]xQ3 =xQ4 [
]xQ3 =xQ4 = c({xQi , xQ3 = xQ4 }).
xQ3
xQ3
Identifying the solution in each sector with Q we can specify
(18.57)
395
Q ]xQ3 =xQ4 [
Q ]xQ3 =xQ4 = cQ ({xQi , xQQ3 = xQ4 }).
xQ3
xQ3
(18.58)
We need to write this in terms of the [P, Q] unknowns. In sector Q, xQ3 xQ4 ,
from (18.54)

i(k +k )x +i
pPi xQi
i=3,4
;
c({xi , x3 = x4 }) = c([Q, P ] + [Q, P ])e P3 P4 Q3
(18.59)
the contribution to [ xQ ]xQ3 =xQ4 proportional to [Q, P ] is
[

= ikP3 [Q, P ]e
Q,P
ikP3 xQ3 +i

i=3,4
kPi xQi
(18.60)
still in sector Q we must include the contribution of the permutation P =

P (34) P bringing the same plane wave:

+ QP
QP
= i(kP3 [Q, P ] + kP [Q, P ])e
3
= i(kP3 [Q, P ] + kP4 [Q, P ])e
i(kP3 +kP4 )xQ3 +i
i(kP3 +kP4 )xQ3 +i
i=3,4
i=3,4
kPi xQi
pPi xQi
(18.61)
.
On the other hand, in the neighboring sector Q = P (34) Q, we have

ikP3 xQ +ikP4 xQ +i
kPi xQi

i=3,4
3
4

Q ,P + Q ,P = [Q , P ]e

ik x +ik x +i
kPi xQi
i=3,4
+[Q , P ]e P3 Q3 P4 Q4
.
(18.62)
We need to write this in the xQi variables,
Q ,P + Q ,P = [Q , P ]e
+[Q , P ]e
ikP3 xQ4 +ikP4 xQ3 +i
ikP4 xQ4 +ikP3 xQ3 +i
i=3,4
i=3,4
kPi xQi
kPi xQi
(18.63)
Evaluating [ xQ ]xQ3 =xQ4 one nds

3

Q
P + Q P = i(kP4 [Q , P ] + kP3 [Q , P ])e
i(kP3 +kP4 )xQ3 +i

i=3,4
kPi xQi
.
(18.64)
Thus (18.58) gives

kP4 [Q , P ] + kP3 [Q , P ] kP3 [Q, P ] kP4 [Q, P ])
= ic([Q, P ] + [Q, P ]).
(18.65)
We eliminate [Q , P ] from (18.65,18.55) and get:

[Q, P ] =
(kP4 kP3 )[Q P ]+ic[QP ]

kP4 kP3 ic
Besides, we must add periodic boundary conditions,
(18.66)
396
(x1 . . . xi . . . xN ) = (x1 . . . xi + L . . . xN ), i = 1, 2, N
(18.67)
which, like in the c = 0 case, determine the spectrum. However, unlike the
non-interacting problem, the N momenta are not independent. Before tackling this problem I show that the 1d Hubbard Model leads to similar equations.
18.3.2 The Hubbard Model in 1d
Consider N Fermions moving on a linear Hubbard chain of Ns sites

H =T
ci cj + U
ni ni ,
(18.68)
<ij>
with nearest-neighbor hopping T = 1 and periodic boundary conditions.

This problem was solved by Lieb and Wu [73] by a Bethe Ansatz and is a
discrete variant of the one of the previous Section. For N = 1,

y|
ci+1 ci
f (x)|x = y|
f (i)|i + 1
=
f (i)y|i + 1 =
f (i)(i, y 1) = f (y 1);
(18.69)
then the elementary Schr

odinger equation
[f (x + 1) f (x 1)] = Ef (x)
(18.70)
is solved by f (x) = eikx , eikNs = 1, E = 2 cos(x). For N Fermions, let us

denote an eigenfunction of H explicitly by
|M , M =
Ns

x1 =1
Ns
...
...
xi =1
Ns

xN =1
f (x1 . . . xN )| x1 . . . xM xM+1 . . . xN ,
: ;< = :
;<
=
(18.71)
where xi is the site of electron i; more simply, we shall also write
|x1 . . . xM xM+1 . . . xN | x1 . . . xM xM+1 . . . xN .
: ;< = :
;<
=
The rst-quantized schr

odinger equation reads:
N
[f (x1 , . . . , xi + 1, . . . xN ) + f (x1 , . . . , xi 1, . . . xN )]
i=1
i(1,M),j(M+1,N )
+U
(xi , xj )f (x1 , . . . , xi , . . . xj . . . xN )
(xi ,xj )
= Ef (x1 , . . . , xi , . . . xN ).
(18.72)
397
where the spin in xi 1 is the same as in xi . In conguration space there are

sectors
RQ = {X : 1 xQ1 xQ2 . . . xQN . . . Ns }
(18.73)
where Q is a permutation. The inequalities imply a partial superposition:
for instance both RQ and P (12) RQ = {X : 1 xQ2 xQ1 . . . xQN . . .
Ns } include the congurations with {1 xQ1 = xQ2 . . . xQN . . . Ns }.
Each electron carries two labels; label i marks the electron position along the
chain, while Qi keeps track of its place in the fundamental sequence 1 N
and of its spin, which is for n M + 1.
For U = 0 momenta would arise because of the need of each one-body
orbital to be periodic. In the interacting case things are less simple, but in
the Bethe Ansatz we consider the electron momenta
k1 < k2 < . . . kN ,
with their permutations
P : {k1 , k2 . . . kN } {kP1 , kP2 . . . kPN }.
The Bethe Ansatz for the wave function is the same as in Equation (18.51):

fQ (x1 . . . xN ) =
[QP ]ei i kPi xQi .
(18.74)
P
fQ is the amplitude in the sector RQ in terms of N !2 coecients [QP ]. The

ki and fQ are unknown, but we know that if Q and Q dier by the exchange
of two electrons of the same spin, fQ = fQ .
For example, in the 2-body case, let k1 < k2 ; the possible permutations
are Q = P = (1, 2) and Q = P = (2, 1), the only sectors are RQ = {X :
1 x1 x2 Ns }, RQ = {X : 1 x2 x1 Ns }, and the Bethe ansatz
reads
amplitude
i(k1 x1 +k2 x2 )
i(k2 x1 +k1 x2 )
fQ (x1 , x2 ) = [QP ]e
+ [QP ]e
fQ (x1 , x2 ) = [Q P ]ei(k1 x2 +k2 x1 ) + [Q P ]ei(k2 x2 +k1 x1 )
sector
Q : x1 x2 (18.75)
Q : x1 x2
If the spins are parallel, U does not act, and the two sectors are disjoint.
The Schr
odinger equation Equation (18.72) links the amplitude f (X) of
a given conguration X = (x1 , . . . , xi , . . . xj . . . xN ) to those that can be
reached from X in one step, that is, applying H once. If X RQ , applying
H we remain in RQ , if X is an interior point, that is |xi xj | > 1, i
= j;
otherwise X is a frontier point. We get dierent conditions from in the two
cases.
398
X is an Interior Point

Let fQ (x1 . . . xN ) = P [QP ]ei n kPn xQn ; consider the contribution from a
particular P to the i = r term in the i sum of Equation (18.72): we obtain
the kinetic term
[fQ (x1 . . . xr + 1, . . . xN ) + fQ (x1 . . . xr 1, . . . xN )]

= (eikPr + eikPr )[QP ]ei n kPn xQn ;
(18.76)
summing over i we get, independently of P ,

E = 2
N
cos(kj ).
(18.77)
j=1
Note that U enters indirectly, by modifying the k values. There are no interior points if the lling is too high, because then double occupations are
unavoidable; however (18.77) holds in general.
X is a Frontier Point
Consider any conguration in (18.74) with at least a double occupation on
a site that must be among the rst M and among the last M . For a given
Q, in (18.73) for some i, xQi = xQi+1 ; then, in Equation (18.74) for given
Q, P one nds at the exponent a contribution (kPi + kPi+1 )xQi . However,
if P is the permutation obtained from P by exchanging i and i + 1, the
resulting exponential is the same. For X Q, the permutations P and P
both contribute to fQ with the same exponential factor,

i
k
x
([Q, P ] + [Q, P ])ei(kPi +kPi+1 )xQi e m=i,i+1 Pm Qm .
The same exponential also arises from the permutation Q obtained from Q
by exchanging i, i + 1. Indeed, the contribution to fQ from P and P is

i
k
x
([Q , P ] + [Q , P ])ei(kPi +kPi+1 )xQi e m=i,i+1 Pm Qm .
We must ensure that f is single-valued, that is, {X RQ , X RQ } =
fQ (X) = fQ (X). This reproduces the condition (18.55)
[QP ] + [QP ] = [Q P ] + [Q P ].
(18.78)
Nesting of a Two-Body Problem in the Many-Body One

From the above, we may conclude that 1) any double occupancy arises as an
equality xi = xi+1 2) congurations with double occupancies are at the border
between sectors and belong to neighboring ones, dened by xQi xQi+1 , and
399
by xQi xQi+1 . 3) (18.77) must hold in both sectors. Let site i be the
same as i + 1 (double occupation). In sector Q dened by xQi xQi+1 ,
the two electrons are those labelled Qi , Qi+1 ; there is superposition with the
sector Q = P (ii+1) Q where the permutations dier by Qi Qi+1 (P
is a transposition). The contribution to the Bethe function arising from a
permutation P of momenta is

i
k
x
[QP ]ei(kPi xQi +kPi+1 xQi+1 ) e m=i,i+1 Pm Qm ,
however in the case xQi = xQi+1 , also belonging to the sector there is mixing
of the permutations
P, P that dier by i i + 1; the exponential is the

same, e
m=,i,i+1
kPm xQm
. To proceed, we introduce the amplitude
i(kPi xQi +kPi+1 xQi+1 )

P
P
Q (xQi , xQi+1 ) [QP ]e

+ [QP ]e
i kP xQi +kP
i
i+1
xQi+1
= [QP ]ei(kPi xQi +kPi+1 xQi+1 ) + [QP ]ei(kPi+1 xQi +kPi xQi+1 ) , (18.79)
such that the contribution to fQ arising from P and P is:

kPm xQm
PP
P i
m=,i,i+1
= P
e
, xQi xQi+1 .
Q
Q
(18.80)
The sector Q dened by xQi xQi+1 , gets from permutations P, P the

amplitude

kPm xQm
PP
PP i
m=,i,i+1
Q
=
e
, xQi xQi+1
(18.81)

Q
(same exponential as in (18.80) ) and

P
P
Q
= [Q P ]e
i kPi xQ +kPi+1 xQ

i
i+1

+ [Q P ]e
= [Q P ]ei(kPi xQi+1 +kPi+1 xQi ) + [Q P ]e
i kP xQ +kP
i
i+1
x Q
i+1
i kPi+1 xQi+1 +kPi xQ
(18.82)
.
There is superposition of sectors, so in order to write (18.72) and obtain

relations among the coecients we must consider

kPm xQm
PP
PP i
m=,i,i+1
QQ
,
(18.83)
= QQ e
where one understands that
P
P
QQ
P
P
xQi xQi+1
Q
PP
Q xQi > xQi+1 .
(18.84)
We dene the set of the indices that in a given sector correspond to electrons that do not participate to double occupations: SQ := {i : xQi
=
400
xQri1 , xQri+1 } Now, the rst member of (18.72) involves one sum on electrons; the contribution of those that do not participate to double occupations
is the correspondent piece of the energy eigenvalue (18.77):
N

PP
PP
QQ
(x1 . . . , xQ + 1, . . . xN ) + QQ (x1 , . . . , xQr 1, . . . , xN )
r
r,rSQ
= 2
N
PP
cos(kPr )QQ
.
(18.85)
r,rSQ
P
Since the eigenvalue is (18.77), the two-body function P
QQ must contribute,
and obeys in xQi = xQi+1 = x to (18.72) with eigenvalue 2 cos(kPi )
2 cos(kPi+1 ). Equation (18.72) grants that
P
PP
PP
[P
QQ (x + 1, x) + QQ (x 1, x) + QQ (x, x + 1)
P
PP
+P
QQ (x, x 1)] + U QQ (x, x)
P
= 2(cos(kPi ) + cos(kPi+1 ))P
QQ (x, x).
(18.86)
The 2 body problem is nested into the N body one in such the way that
everything generalizes. In view of (18.84), Equation (18.86) means
P
PP
[P
Q (x + 1, x) + Q (x 1, x)
P
PP
PP
+P
Q (x, x + 1) + Q (x, x 1)] + U Q (x, x)
P
= 2(cos(kPi ) + cos(kPi+1 ))P
Q (x, x).
(18.87)
Substituting (18.79,18.82), eliminating the common factor ei(kPi +kPi+1 )x and

making the substitutions:
ikPi+1
P

P
+ [Q P ]eikPi
Q (x + 1, x) [Q P ]e
ikPi
PP
Q (x 1, x) [QP ]e
+ [QP ]eikPi+1

P
P
[QP ]eikPi+1 + [QP ]eikPi
Q (x, x + 1)
PP
Q (x, x 1) [Q P ]eikPi + [Q P ]eikPi+1
P
P
[QP ] + [QP ]
Q (x, x)
(18.88)
we obtain
[QP ](eikPi + eikPi+1 + U ) + [QP ](eikPi + eikPi+1 + U )
= [Q P ](eikPi+1 + eikPi ) + [Q P ](eikPi + eikPi+1 ).
We eliminate [Q P ] using (18.78), and nd
(18.89)
[QP ] =
U
2i
[QP ]+(sin(kPi )sin(kPi+1 ))[Q P ]
U
sin(kPi )sin(kPi+1 )+ 2i
401
(18.90)
This looks like a slightly modied version of Equation (18.66). It is instructive

to work out the explicit solution of the problem with 2 electrons, see Problem
18.2; it is somewhat harder than one could expect.
18.3.3 The Periodic Boundary Conditions
The problems of the last two Subsections are closely related and can be
carried out by similar means; we choose to refer to the continuous case in
the following. We need to nd ecient ways to impose the periodic boundary
conditions (18.67). Following Yang[74], we arrange the [Q, P ] amplitudes as
a N ! N ! matrix and consider the column with some particular P :
[1, 2, 3, . . . N , P ]
[Q2 , P ]
P =
(18.91)
...
[QN ! , P ]
Let 0 denote the column corresponding to the fundamental (or identity)
permutation I = {1, 2, 3, . . . N }. We cast the equations (18.66) in a more
convenient form,
[QP ] =
[Q P ] ij [QP ]
ic
, ij =
.
1 + ij
ki kj
(18.92)
This is a relation involving two columns, in shorthand3 :

P = Yiji,j P , P = P (ij) P, ; j = i + 1.
(18.93)
To write down the column-switching operator Yija,b which performs a transposition of consecutive indices, we also need to introduce a row-switching
ba
operator PQQ
which acts on the Qi and exchanges a and b. Indeed,
Yija,b =
ba
ij + PQQ

aij + bij P ab , b = a + 1.
1 + ij
(18.94)
The upper indices of Yija,b give the position of the consecutive electron pair in
the list, the lower indices the identity of the elements; for example, for N=4,
3,4
exchanges the third and fourth elements in the list, provided that they
Y24
3
When it is clear from the context which are the i and j indices we are referring
to, we shall also continue to use the shorthand notation P for the permutation
of momenta obtained from P by exchanging ki and kj , and shall use P for the
column corresponding to it.
402
3,4
2,3 3,4
are 2 and 4, giving Y24
3124 = 3142 ; moreover, Y14
Y24 3124 = 3412 , and
so on; any P can be obtained from 0 by applying products of Y operators.
The action of Yija,b is the same as the action of Pij on any permutation
P. From this very fact one that can prove some identities, that can of course
be checked starting from the denition of Y . Since Pij Pji = 1,
Yija,b Yjia,b = 1
(18.95)
Yijab Yklcd = Yklcd Yijab .
(18.96)
and since Pij Pkl = Pkl Pij ,
Moreover, one can check that P12 P13 P23 = P23 P13 P12 ; in this way one nds
that
a,b b,c a,b
b,c
Yik Yij = Yijb,c Yika,b Yjk
.
(18.97)
Yjk
Here is the operator that exchanges both rows and columns:
ij
i,j
Xij = PQQ
=
Yij
ij
1 PQQ
ij
1 + ij
, j = i + 1.
(18.98)
This operator lets the electron i to overtake the electron j=i+1 (recall that
the phase factors have already been dealt with). For example if N = 3,
[{1, 2, 3}, {1, 2, 3}]

[{2, 1, 3}, {1, 2, 3}]
[{1, 3, 2}, {1, 2, 3}]
(18.99)
0 =
[{3, 2, 1}, {1, 2, 3}]

[{2, 3, 1}, {1, 2, 3}]
[{3, 1, 2}, {1, 2, 3}].
and one nds
X1,3 0 =
[{3, 2, 1}, {3, 2, 1}]
[{2, 3, 1}, {3, 2, 1}]
[{3, 1, 2}, {3, 2, 1}]
,
X
X
=
2,1 1,3 0
[{1, 2, 3}, {3, 2, 1}]
[{2, 1, 3}, {3, 2, 1}]

[{1, 3, 2}, {3, 2, 1}]
[{3, 1, 2}, {3, 1, 2}]

[{1, 3, 2}, {3, 1, 2}]
[{3, 2, 1}, {3, 1, 2}]
[{2, 1, 3}, {3, 1, 2}]
[{1, 2, 3}, {3, 1, 2}]

[{2, 3, 1}, {3, 1, 2}]
(18.100)
Now if we consider the amplitude

[Q, P ]ei
kPi xQi
in sector Q ( xQi xQi+1 , ) and consider the transformation

xQ1 xQ1 + L,
403
: xQ2 xQ3 . . . xQ1 with P : kP2 , kP3 , . . . kP1 with

we land in the sector Q
an amplitude

P ]ei i kPi xQi .
eikP1 L [Q,
Since the wave function must be single valued, this implies
[Q, P ] = eikP1 L
P ].
[Q,
(18.101)
Consider for instance the last but one component [{2, 3, 1}, {1, 2, 3}] of 0
in (18.99) and its evolution in (18.100): the boundary condition is
[{1, 2, 3}, {3, 1, 2}] = eik1 L [{2, 3, 1}, {1, 2, 3}].
(18.102)
In general, by the same reasoning,

eikj L 0 = Xj+1,j Xj+2,j XN,j X1,j X2,j Xj1,j 0 .
(18.103)
Thus, 0 is a simultaneous eigenfunction of N operators; these however may

be shown to commute, so we can limit ourselves to consider one of them, e.g.
letting j = N.
18.3.4 Spin Chain Analogy
Usually a problem is practically solved when reduced to a linear system, but
this is not the case here. Since there are N !2 amplitudes [P, Q] and for each
amplitude there are N-1 consecutive i, i + 1 pairs, this is a large homogeneous
system and cannot practically be solved by the usual methods, unless N is
rather small. Yang achieved a solution by the use of the symmetry, a very
smart and profound one.
ij
If all the electrons have parallel spins, PQQ
= 1, Xij 1 and
ikj L
= 1. This is nothing but the condition of the non(18.103) yields e
interacting theory, as one should expect, since parallel spin electrons cannot
be at the same position. Recalling Sect.(9.8), more generally 0 is associated
with a Yang diagram with one or two columns, and the spin function then
belongs to the conjugate representation. We go to the conjugate representaba
ba
tion by change PQQ
PQQ , and to this end we introduce

Xij
ij
1 + PQQ
ij
1 + ij
(18.104)
where P now operates on the spin variables. The eigenvalue equation is

Zj Xj+1,j
Xj+2,j
XN,j
X1,j
X2,j
Xj1,j
= j ,
(18.105)
but, since since we are looking for the spin part of the same wave function,
404
j = eikj L .
(18.106)
These are N equations, but each yields all the eigenvalues since they can be
shown to commute; so we can limit ourselves to solve

X2N
XN
ZN = N , ZN = X1N
1,N .
(18.107)
One can interpret as the wave function of N spins on a ring, =

(y1 , y2 , , ym ; z1 , z2 , zN M ) where the integers y1 , y2 , , ym denote the
positions along the chain of the M reversed spins, and the zs those of the
N-M up spins. This may be taken to be be symmetric in the interchange
of ys and separately in the interchange of zs while any attempt to symmetrize simultaneously in variables belonging to the two sets must give 0.
The same information is contained in a reduced wave function that we denote by (y1 , y2 , , yM ), with the ys in increasing order, which depends
on M variables
only; for N M spins up and M spins down the components
N
are just
, that is, the size of the problem is substantially reduced.
M
Three Spins,, ,
Since for parallel spins the problem is trivial, let us start considering M =
1 reversed spin. For example if N = 3, we may replace the unknown 6component amplitude (18.99) by the 3-component reduced wave function
(1)
= (2) ,
(18.108)
(3)
where the components can be interpreted as the amplitude that the overturned spin comes rst, second and third, respectively. In terms of the full
wave function for the three spins, (1) stands for [{, 2, 3}, {1, 2, 3}] = [{
, 3, 2}, {1, 2, 3}]; F (2) is the amplitude that spin 2 is down and is symmetric
in 1 and 3, and so on. The full 0 can be obtained from the reduced . Let
us work out4 the eigenvalue equation (18.107)

X23
.
Z3 = 3 , Z3 = X13
(18.109)
Recalling (18.104), and setting

aij =
1
23
bij =
1 + 23
1 + 23
(18.110)
ij

we see that aij + bij = 1 and Xij
= aij + bij PQQ
; X23 (1) = (1), but

X23 mixes the other components:
4
Here and below this section involves some algebra which is quite elementary
but somewhat tedious; using a computer code is a good idea.
a23 b23
b23 a23
(2)
(3)

=
(3)
]
(2)[a23 + b23 (2)
405
,
.
(2)
(3)[a23 + b23 (3)
]
(18.111)
The solution is gained by the following inspired ansatz,with some real :

k2 +
(3)
=
(2)
k3
ic
2
ic
2
which means, the ratio of two consecutive components is obtained as the ratio
of two linear functions. Indeed, now we show that we can solve the boundary
condition eigenvalue problem by writing
1
k1 +ic/2
=
(18.112)
.
k2 ic/2
k1 +ic/2 k2 +ic/2
k2 ic/2 k3 ic/2
It holds:
a23 b23
b23 a23
(2)
(3)

=
3 (2)
(3)
2
,
,
(18.113)
where I have introduced

j =
kj + ic/2
;
kj ic/2
(18.114)
a simplify command yields immediately

j aij
ki + ic/2
.
=
bij
kj ic/2

Next, to apply X13
to the result, we compute
, +
,
+

3 (1)
(1)
a13 b13
=
.
(3)
(3)
b13 a13

2
Thus,
1
Z3 = 3 (2) .
(18.115)
(18.116)
(18.117)
(3)
3 2 1
Now we can solve the eigenvalue equation by requiring that j = j and that
1 2 3 = 1;
(18.118)
this means that a translation by L of the whole system has no eect. Putting
together (18.106) and (18.114) one gets
eikj L =
kj + ic/2
kj ic/2
(18.119)
and with (18.118) one can determine and the momenta. The beauty of this
approach is that it can be fully extended.
406
N Spins,1 Overturned
The above results extend immediately to arbitrary N ; thus,
(, y) =
y1

j=1
kj + ic/2
kj+1 ic/2

,
(18.120)
where the integer y species the position of the overturned spin along the
chain.
General Case
The above solution extends to any N and M via the following generalized
Bethe ansatz involving M dierent parameters:

=
a(P )(P1 , y1 )(P2 , y2 ) (PM , yM ),
(18.121)
P SM
with the same as before; here given two permutations P and P that dier
by the exchange of elements and + 1 the coecients obey[142]
a(P ) = a(P )
P P (+1) + ic
.
P P (+1) ic
(18.122)
The extension of (18.119) is

eikj L =
kj + ic/2
.
kj ic/2
(18.123)
The general proof that the solution has the [N M, M ] permutation

symmetry may be found in [142]. Note that the ansatz (18.121) describes
one kind of particles, the down-spin electrons, while the up-spin electrons
are reduced to sites in an eective chain; Yangs procedure reduced by 1 the
number of particle species.
Problems
18.1. Work out the L matrix (Section 18.6) for the ground state of the repulsive H2 molecular model H(t, U ) (14.58) and for the canonically transformed,
attractive model (21.32) and compare the results for Ut = 7.
18.2. Work out the detailed solution for the Hubbard chain with N = 2
electrons. Verify the solution by writing down explicit energy eigenvalues for
Ns = 3.
Part VI
Appendices
19 Appendix 1: Zero-point Energy in a Pillbox
Setting
b
L
= qx ,
c
L
d2 q =
= qy ,
U (s) = h
cL2
! a "2

a 3
d2 k
d2 k, (1.29) yields:

a
2
1 + k 2 e( s ) 1+k .
(19.1)
kx ,ky >0
The angle integral from 0 to
yields
n
hc 2 L2 3
a
dz 1 + z e( s ) 1+z
3
4s
a
U (s) =
hc 2 L2
4s3
d
d
the summation is performed by
3
a
a
dz
e( s ) 1+z ;
1+z
en =
U (s) =
hc 2 L2
d
d
3
0
1
1e
(19.2)
1 = e11 ; thus
1
dz
.
1 + z e s 1+z 1
(19.3)
By the substitution u = 1 + z, and doing one of the derivatives one gets
2

su
d
1
hc 2 L2
(19.4)
U (s) =
du !
"2 ;
2
d
s
esu 1
1
and putting

1
esu
1
du !
"2 =
u
s
e 1
Finally,
hc 2 L2
U (s) =
2

dx
es
d
d
1
(x 1)2
2
1
1
.
es 1
1
1
es 1
(19.5)

.
(19.6)
20 Appendix II-Character Tables
C3 = Z3 I
C3 C32
A1
1 1 1
1
C3v
A1
A2
E
I
1
1
2
2C3
1
1
1
C6v
A1
A2
B1
B2
E1
E2
I
1
1
2
2
(x, y)
C4v
A1
A2
B1
B2
E
I 2C5
2C52
1
1
1
1
1
1
2 2 cos 2 cos 2
2 2 cos 2 2 cos
I
1
1
1
1
2
2
C2
1
1
1
1
2
2
C2 2C
1
1
1
1
2 2 cos
2 2 cos 2
2C3
1
1
1
1
1
1
2C 6
1
1
1
1
1
1
I
1
1
1
1
C2v
A1
A2
B1
B2
2i
3
3v g = 6
1
z
1 Rz
0 (x, y)
C5v
A1
A2
E1
E2
C!v "
A1 ! + "
A2
E1 ()
E2 ()
=e
z
I
1
1
1
1
2
C2
1
1
1
1
2
2C4
1
1
1
1
0
C2
1
1
1
1
2 v
1
1
1
1
0
v
1
1
1
1
v
1
1
1
1
g=4
z
xy, Rz
x, Ry
y, Rx
2d g = 8
1
z
1
Rz
1 x2 y 2
1
xy
0
(x, y)
5v g = 10, = 2
5
1
z
1
Rz
0
(x,
y) "
!
0
xy, x2 y 2
3d
1
1
1
1
0
0
3v
g = 12
1
z
1
Rz
1
1
0 (x,
x , Ry
! y), (R
")
0
xy, x2 y 2
v
g=
1
z
1
Rz
0 ! (x, y) "
0
xy, x2 y 2
D2
A1
B1
B2
B3
I
1
1
1
1
C2z
1
1
1
1
C2y
1
1
1
1
C2x
1
1
1
1
g=4
x2 , y 2 , z 2
z, xy, Rz
y, xz, Ry
x, yz, Rx
412
D3
A1
A2
E
20 Appendix II-Character Tables

D4
A1
A2
B1
B2
E
2C3 2C2
I
g=6
1 1
1
z 2 , x2 + y 2
1 1 1
z, Rz
2 1 0 (x, y) , (Rx , Ry )
E
1
1
1
1
2
2
1
1
1
1
2
2
2C6
1
1
1
1
1
1
1
1
1
1
1
1
C2
1
1
1
1
2
2C4
1
1
1
1
0
C2
1
1
1
1
0
2C2
g=8
1
z 2 , x2 + y 2
1
z
1
x2 y 2
1
xy
0 (x, y) , (Rx , Ry )
I
1
1
1
1
2
2
h
1
1
1
1
2
2
2C3
1
1
1
1
1
1
2S3
1
1
1
1
1
1
3C2
1
1
1
1
0
0
3v
g = 12
1
z2
1
Rz
1
1
! z
"
0 (x, y) , xy, x2 y 2
0
(xz, yz) , (Rx , Ry )
2C3
1
1
1
1
1
1
1
1
1
1
1
1
C2
1
1
1
1
2
2
1
1
1
1
2
2
3C2
1
1
1
1
0
0
1
1
1
1
0
0
3C2
1
1
1
1
0
0
1
1
1
1
0
0
i
1
1
1
1
2
2
1
1
1
1
2
2
2S3
1
1
1
1
1
1
1
1
1
1
1
1
D3h
A1
A2
A1
A2
E
E
D6h
A1g
A2g
B1g
B2g
E1g
E2g
A1u
A2u
B1u
B2u
E1u
E2u
I
1
1
1
1
2
2S6
1
1
1
1
1
1
1
1
1
1
1
1
h
1
1
1
1
2
2
1
1
1
1
2
2
3d
1
1
1
1
0
0
1
1
1
1
0
0
3v
g = 24
1
1
Rz
1
1
0 ! (Rx , Ry ) "
0
x2 y 2 , xy
1
1
z
1
1
0
(x, y)
0
Double Groups

C3v
A1
A2
E
E1/2
5
6

E E
C4v
A1
A2
B1
B2
E1
E2
E3
1
1
1
1
2
2
2
1
1
1
1
2
2
2
E E
1
1
2
2
1
1
1
1
2
2
1
1
C32
C3
2
3v 3v E
C3 E C3 E
1
1
1
1
1
1 1 1
1 1 0
0
1
1 0
0
1
1
i
i
1
1 i
i
C43 C2
2C2
2C2
C4
3

C4 E C4 E C2 E 2C2 E 2C2 E
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
2
0
0
0
2
0
0
0
2
2
0
0
0
21 Proof of the Wigner-Eckart Theorem
Here I reinforce the intuitive discussion given in Chapter 9 by a proof of the

Wigner-Eckart theorem :
i|Tq() |k = T () i|pk ,
where the reduced matrix element < T () > does not depend on the
components;i|pk is the Clebsh-Gordan coecient.
Here, |i > and |j > denote the components of bases of irreps and
()
(that could also coincide) and Tp the p component of an irreducible
tensor

()
Tq() Dqp
(R).
(21.1)
RTp() R1 =
q
As the components i,p and k vary, one nds a number of matrix elements
()
< i|Tp |k > that are all connected by the theorem.
Proof
R G, R = R
implies
i|Tq() |k = i|R RTq()R R|k =

()
i|R
Tq() Dqp
(R) R|k =
j|Tq() |r >
()
Dji (R) Dqp ()(R)Drk ()(R).
(21.2)
qjr
In order to use the GOT, we write the product of two D matrices as one
()
()
D. Actually,Dqp (R)Drk (R) is the direct product of two irreps,

()
()
()
(R)Drk (R) =
qr|s Dst t|pk ,
Dqp
st
where we have used Clebsch-Gordan coecients, and inserting in (21.2 ),

()
()
i|Tp() |k >=
j|Tq() |r > Dji (R)
qr|s > Dst t|pk > .
qjr
st
(21.3)
414
21 Proof of the Wigner-Eckart Theorem
Here R is arbitrary; summing over R, in the l.h.s. we get a factor NG

(=Group order);in the r.h.s. the GOT yields
()
()
Dji (R) Dst (R) =
Ng
js it ,
m
and nally we obtain

NG i|Tp() |k >=
= j|Tq() |r
qjr
qr|s t|pk
st
NG
js it ,
m
that is
i|Tq() |k = T () i|pk ,
where the reduced matrix element
1
T () =
j|Tq() |r > qr|j
m qjr
does not depend on the components.
q.e.d.
(21.4)
Solutions
Problems of Chapter 2

2.1 np = Z1 T r(np ) has a denominator Z = k (1 + e(k ) ) and a

numerator k=p (1 + e(k ) )e(p ) , since p is lled in all states that
contribute.
2.2 One readily veries by computing the time derivative that

i t
U (t, t ) = U0 (t, t )
U (t, )H1 ( )U (, t )
h t
is equivalent to the Schr

odinger equation.
2.3 From the denition one can easily verify that
d
T {A1 (t1 )A2 (t2 )(t)} = T {A1 (t1 )A2 (t2 )(t)}
+
dt
(t t1 )T {[(t), A1 (t1 )] A2 (t2 )}
+(t t2 )T {A1 (t1 )[(t), A2 (t2 )] }
(21.5)
where the functions arise from (t ti ) factors (for (t) standing on the left
of Ai (ti ) and from (ti t) factors (for (t) standing on the right of Ai (ti );
hence one gets the commutators. The argument extends immediately to any
n yielding the term in and an extra term
n
(t tq )T {A1 (t1 ) [(t), Aq (tq )] An (tn )}.
(21.6)
q=1
2.4 One can thinkof the (t, t ) interval divided into N 1 intervals and

write eiH (t t) = n eiH (tn tn1 ) ; after dierentiating, one lets N .

6
3.1 The ground conguration 1s 2s 2p has
= 20 states. The largest
3
ML is reached by (1+, 1, 0) thus ML = 2, MS = 12 . The resulting
2
416
Solutions
D occupies 10 states. There are 4 determinants with ML = 1, namely,

(1, 1, 1), (1, 0+, 0, ).
p3 4 S,2 P,2 D.

10
2 2
6
2
3.2 The ground conguration of Ti is 1s 2s 2p 3d . There are
= 45
2
states. The largest ML is 4. One nds:
d2 1 S,3 P,1 D,3 F,1 G.
3.3 The tables of Clebsh-Gordan coecients yield

(ms1 , ms2 )
1
1
| ms1 ms2 |S = 1, MS = 0 =
2
2
2
and for |l1 = 2m1 l2 = 2m2 |L = 1, ML = 0 the following values:

m2 = 2 25

1
m2 = 1
10
m2 = 0 0
m2 = 1 m2 = 2
10
2
5
hence normalizing again (which is generally necessary when using ClebshGordan coecients)
|3 P ML = 0, MS = 0 = 110 {2|2+ , 2 | + 2|2 , 2+ | |1+ , 1 |
|1 , 1+ |}.
3.4 For two-electron

states, L = L
1 + L2 . For L=1 one learns from (6.1.1)
that L |0 = 2|1 , L |0 = 2| 1 .
So,
1
|3 P ML = 1, MS = 0 = L+ |3 P ML = 0, MS = 0 =
2
+
L+
1
1 + L2
{2|2+ , 2 | + 2|2 , 2+ | |1+ , 1 | |1 , 1+ |}.

2
10
Bringing the one-body operators to act on the respective states,using (6.1.1)

and the spin shift operators one nds
1
|3 P ML = 0, MS = 1 = {2|2+ , 2+ | |1+ , 1+ |}.
5
Moreover,
1
|3 P ML = 1, MS = 1 = {2|2+, 1+ | 6|1+ , 0+ |},
10
Solutions
417
1
|3 P ML = 1, MS = 1 = {2|2 , 1 | 6|1 , 0 |},
10
1
|3 P ML = 0, MS = 0 = {2|2+ , 2 |+2|2 , 2+ ||1+ , 1 ||1 , 1+ |},
10
1
|3 P ML = 0, MS = 1 = {2|2 , 2 | |1 , 1 |},
5
1
|3 P ML = 1, MS = 1 = {2|1+, 2+ | 6|0+ , 1+ |},
10
1
|3 P ML = 1, MS = 1 = {2|1 , 2 | 6|0 , 1 |},
10
3.5 We found
1
|3 P ML = 0, MS = 1 = {2|2+ , 2+ | |1+ , 1+ |}.
5
Il is clearly a triplet since the spin conguration is , anf ML = m1 +m2 = 0.
In order to verify that it is a P state, we use
+
L2 = L21 + L22 + 2L1z L2z + L+
1 L2 + L1 L2
where L21 = L22 = 2 3 Now, 22L1z L2z {2|2+ , 2+ | |1+ , 1+ |} = 2 2 2

+
+ +
+
+
(2)|2+ , 2+ |21(1)|1+, 1+ |; L+
1 L2 |2 , 2 | = 0, L1 L2 |1 , 1 | =
+
+
(2) 2|1 , 1 |, and in this way the check is readily completed.
4.1 Integrate
4.2 Let
d
d
! H
"
e [A, eH ] = eH [H, A] eH .
fmn (x, x , ) =
eKn Rmn (x, x )( mn ),
mn
which incidentally is just the spectral density (4.139) without the 1 s e

factor. Equation (4.138) implies

d

(1 + s e )fmn (x, x , ),
ZReG(x, x , ) = P

(21.7)
ZImG(x, x , ) = (1 s e )fmn (x, x , ),
and the results follows by substitution.
418
Solutions
4.3 In the continuous case, equation (4.44) reads

2 d2
h
)g(x , x, z) = (x x )
2m dx2
(z +
and is solved by
g(x , x, z) =
2m e
h2
2mz
h
2
(21.8)
|xx |
i z
(21.9)
|
= sign(x x ) ] and the band-edge singularity is similar to
[note that d|xx
dx
the discrete case (with one band edge, however).
4.4 In the continuous case, for 3d, the solution of (4.44), namely,
(z +
2 2
h
)g(r , r, z) = (r r )
2m
(21.10)
is readily veried to be
2mz
i
|rr |
h
2
e
m

g(r , r, z) =
.
|r r |
2
h2
Therefore,
1

Im limr r
g(r, r , z) yields the well known
(21.11)
z singularity.

1
4.5 U (r) = 12 n + S d2 r (g 1 (r, r , )+(E)
g (r, r , ))(r ), where
and E are the unperturbed and perturbed energy eigenvalues, respectively.
5.3 We found already that
Gk0 () =
Vk0
G00 ().
k
(21.12)
The o-diagonal element of the last of 5.21 yields

Gkk () =
Vk0
G0k (), k
= k ,
k
(21.13)
and by substituting into the third of (5.21) we obtain

G0k () =
V0k Gkk
.
0 ()
(21.14)
Combining with the -diagonal element of the last of (5.21) we obtain

Gkk () =
1
k
|V0k |2
0 ()
(21.15)
Solutions
G0k () =
V0k
,
( k )( 0 ()) |V0k |2
419
(21.16)
and nally
Gkk () =
V0k
Vk0
.
k ( k )( 0 ()) |V0k |2
(21.17)
5.4 The Sz factor implies that 1) the scattering conduction spin direction is
the quantization axis for the impurity spin, 2) the singular scattering does not
ip the impurity spin 3) for S > 12 the scattering probability grows with the
square of the spin component mS 4) opposite mS gives opposite amplitude
(2) (2)
5) The resistivity goes with T rUif Uif .
6.1 Averaging over a one-electron wave function,

e
j(x)
=
d3 x (x ) [px (x x ) + (x x )px ] (x ),
2m
but the rst term in the integrand can be rewritten (px (x )) (x
x )(x ) = i
hx (x ) (x x )(x ) and the expectation value is the current. Then
N

1

i)
d3 xA(xi ) j(x
H = 2mc
c
i

e
3

(21.18)
=
d x [A(x) px (x x ) + A(x)(x x )px ]
2mc

. The
where A(x) and px commute. Let us take the matrix element Hmn
rst term contributes

e
3

d x m (x ) d3 xpx A(x)(x x )n (x ).
2mc
Caution
is needed when integrating with the functions if there are operators.
3
d x(x x )(x ) = (x) holds even if contains operators, but must
stand on the left. So we cannot integrate over d3 x directly, but we can over
d3 x and we nd

e
d3 x m
(x )px A(x )n (x ).
2mc
The second contribution

e
3

d x m (x ) d3 xA(x)(x x )px n (x )
2mc
is (integrating in either way)

e
(x )A(x)px n (x).

d3 xm
2mc
Therefore the two formulations are equivalent.
420
Solutions
8.1
1) The characters of the representation (1) with one electron are
(E) = 5, (C2 ) = 1, (2C4 ) = 1, (2v ) = 3, (2d = 1. Applying the
LOT one nds (1) = 2A1 E B1 .
2) We choose a basis whose elements are determinants (i, j, k, m) |i
j k m | with, say, i > j, k > m, numbering the sites as in Figure 8.5.
2
5
= 100. One must determine how many
The number of congurations is
2
remain invariant or change sign under the operations of the Group. Under C2 ,
the only contributions come from (4, 2, 4, 2), (4, 2, 5, 3), (5, 3, 4, 2), (5, 3, 5, 3),
which remain invariant. Under x , one nds invariant congurations
(2, 1, 2, 1), (2, 1, 4, 1), (2, 1, 4, 2), (4, 1, 2, 1), (4, 1, 4, 1), (4, 1, 4, 2), (4, 2, 2, 1),
(4, 2, 4, 1), (4, 2, 4, 2), (5, 3, 5, 3) while the conguration that change sign are
(2, 1, 5, 3), (4, 1, 5, 3), (4, 2, 5, 3), (5, 3, 2, 1), (5, 3, 4, 1), (5, 3, 4, 2). The congurations (5, 2, 5, 2), (5, 2, 4, 3), (4, 3, 5, 2), (4, 3, 4, 3) remain invariant under
one of the d reections. Proceeding in this way, one nds the characters of
the reducible representation (4) for 4 particles:
C4v I C2 2C4
A1
1 1 1
A2
1 1 1
B1
1 1 1
B2
1 1 1
2 2 0
E
(4) 100 4 0
2 v
1
1
1
1
0
4
2d g = 8
1
z
1
Rz
1 x2 y 2
1
xy
0
(x, y)
4
Applying the LOT one nds (4) = 15A1 11A2 24E 13B1 13B2 .
8.2 The characters of (4) are 1296, 16, 0, 64, 16; hence, (4) = 184A1
144A2 320E 176B1 152B2 .
9.1
In E E = A1 A2 E all the irreps are contained. Using the set of
matrices (7.44),
E
10
01
+
C3
1
3
2
2
3
1
2
2
,+
C32
12
23
3
2
12
c
, a + b , +
,
1
3
1
3
1 0
2
2
2
2
3
3
0 1
1
1
2
Solutions
421
from the one-electron basis (x, y) we form the direct-product basis

(x1 x2 , x1 y2 , y1 x2 , y1 y2 ).
y+x 3
3
, C32 x = x+y
, C32 y =
2
2
x+y 3
y+x 3
x+y 3
, 3 x = 2 , 3 y
x, 1 y = y, 2 x = 2 , 2 y =
2

the projection operator P (A1 ) = RG R, one nds
Using C3 x = x+y2
, C3 y =
3
, 1 x =
y+x 3
and
2
y+x2
=
x2 +y2
1 3
PA1 x1 x2 = x1 x2 +( x1 +y21 3
)( x2 +y22 3 ) + ( x1 +y
2 )(
2
x1 +y1 3
x2 +y2 3
x1 y1 3 x2 y2 3
+x1 x2 + (
)(
)+(
)(
)
2
2
2
2
= 3 (x1 x2 + y1 y2 ) .
This is the wave function, which is even in the exchange on 1 and 2, so it is

singlet. Normalizing,
x1 x2 + y1 y2
1
.
(21.19)
A1 =
2
9.2
|x1 y2 y1 x2
,
(21.20)
|1 A2 =
2
to be multiplied for the singlet spin function. Therefore the non-vanishing
coecients are: 1 A1 |Ex1 Ey1 = 1 A1 |Ey1 Ey1 = 12 .
9.3 Using the set of matrices (7.44),

E
10
01
+
C3
23
2
3
21
2
,+
C32
12
23
3
2
12
c
, a + b , +
,
1
3
1
3
1 0
2
2
2
2
3
3
0 1
1
1
2
we nd that the contribution of the reections is the same as that of the

rotations for P(E,y) and opposite for P(E,x) . Consequently,
x1 +y1 3 x2 +y2 3
)(
)
P(E,y) x1 x2 = x1 x2 12 (
2
2
1 x1 +y1 3 x2 +y2 3
2(
)( 2
)
2
1 x1 +y1 3 x2 +y2 3
+x1 x2 2 (
)(
) 12 ( x1 y21 3 )( x2 y22 3 )
2
2
= 32 (x1 x2 y1 y2 ) ;
P(E,x) x1 x2 = 0. Hence,
|1 Y =
|x1 x2 |y1 y2
.
2
Thus,
@1 1
@ 1
1
X 1 E1 x1 E2 x2 = 1 X 1 E1 y1 E2 y2 = .
2
422
Solutions
9.4 One can check the results with Appendix II.

9.5 To nd the characters of the 3/2 and 5/2 representations in C4v
, or
similarly in D4 , under rotations by an angle we choose to rotate around
the z axis; then |LSJM |LSJM eiM . So, the D matrix is diagonal and
the character is given by the formula
(J) () =
sin[(J + 12 )]
sin( 2 )
(21.21)
which which generalizes (9.20) above and results from using the identity
J

m=J
1
x
cos(mx) = csc( ) sin[(J + )x].
2
2
One nds
(3/2) () =
sin(2)
= 4 cos()) cos( ).
sin( 2 )
2
(21.22)
The reections and all the improper rotations can be written like products
iR . Proceeding in this way,
3

C43 C4 C2 2C 2 2C2
D 4 E E
2C2 E
C4 E C4 E C2 E 2C 2 E

A1 1 1
1
1
1
1
1
1
1
1
1
1
A 2 1 1
B 1 1 1 1 1
1
1
1
1
1
1
B 2 1 1 1 1
E 1 2 2 0
0
2
0
0
E 2 2 2 2 2 0
0
0
E 3 2 2 2
2
0
0
0
3/2
4 4 0
0
0
0
0
2
0
0
0
5/2 6 6 2
and 3/2 = E2 E3 , 5/2 = E2 2E3 .
10.1 This result is most easily obtained
in the time representation, keeping

(x )m (x)eim (tt ) , where m (x)

in mind that for t > t G(t, t ) = m m
are the energy eigenfunctions. Thus, each term yields half of the answer.
10.3 One easily recovers the results of Sect. 4.3.1.
10.4 As outlined in Ref. [205], one nds:
Solutions
+ ik ]Fkk (, , t, t ) = kk ( )g(t t )

Vkq Fqk (, , t, t )( ; t, t ),
i
423
(21.23)
were ( ; t, t ) = 1 for t > > t and vanishes otherwise.
11.1
R
(d)

(i)U (ilkj)(i)U (kjil)
0
iklj
dt1
t1
dt2
igi (t2 t1 )igk (t1 t2 )igl (t2 t1 )igj (t1 t2 ).
(21.24)
The integral yields (dropping an exponential term which goes to zero as

U(ilkj)2
t (1 i)) R(d) (t) = it iklj k +
.
j i l
11.2 From = + G0 + G0 G0 + one nds
1
.
1 G0
(21.25)
11.3 b)is the only skeleton.

11.4 Putting the Hartree potential and in the one-body term, G(0)1 (1, 2) =
i t
H0 Vef f (1))(1, 2) and we may rewrite (11.81) as G(0)1 = G1 + ,
that is,
G1 (1, 2) = (i
H0 Vef f (1))(1, 2) (1, 2).

t
(21.26)
12.1 One gets
2
E
n
(3 2 ) 3
2
n 3 , and inserting into (12.95) one nds n =
23
3 2 (
(x)) 3 , that is, one is left with the semi-classical Thomas-Fermi

method.
12.2

t1
t2
t3
(iU )4 t
dt
dt
dt
dt4 (e2iV (t1 +t2 t3 t4 ) )2
1
2
3
28
0
0
0
0

1 + 2e2iV t
1 U 4 5 e4iV t 4e2iV t
( )
iV t
=
. (21.27)
32 V
512
128
424
Solutions
13.1 Symbolically, one writes Equation (13.135)
J = (L) f (L) (g r g a ) + (L) g <
but for the other electrode one writes
J = (R) f (R) (g r g a ) (R) g < ;
the solution arrives eliminating g < algebraically; this is legitimate for every
innitesimal energy interval d.
14.1 Dening the combinations
p = um + us , p = um us , r = um + ius , s = um ius ,
one obtains
Gmn (E) =
Gpp Gqq i [Grr Gss ]

.
4
prob1
14.2 One gets the moments and nds an = 0 and
b2n = W 2
n2
.
4n2 1
(21.28)
14.3 One gets the moments and nds an = 0 and

b2n =
W2
.
4
(21.29)
14.4 From Equation (14.101),

x1 =
a12 A12 b2 + a13 A13 b3

a12 a23 A123 b3 + a13 a32 A123 b2
A1 b1
+
A
A
A
b1
a12 b2 + a13 b3
a12 a23 b3 + a13 a32 b2
=
+
(21.30)
.
D1
D12
D123
We may proceed by noting that D123 = D12 D312 , D312 = a33 D12 = D1 D21 ,
D21 = a22 a23 a32 /D312 , and we have all the ingredients to apply 14.86 and
carry on the calculation.
Solutions
425
16.1 Using (16.15),(16.16), the exponent transforms to
hk 2
mv 2
mvx
)x + t(
+
]
h
2m
2h
but this must be written in terms of x . Thus one obtains
i[(k
(x , t) = eik x ih
(k )2 t
2m
with hk = h
kmv. Thus the plane-wave becomes a plane-wave with Galilean
transformation of the momentum.
17.1 Table III shows how the irreps of G split in C4v .
G
A1
A1
B2
2
B
1
2
1
2
3
4
1
2
3
4
1
2
3
4
1
2
C4v
A1
A1
B2
B2
2B2
2A1
A1 + 2B1
A1 + 2B1
2A2 + B2
2A2 + B2
A1 + B1 + E
A1 + B1 + E
A2 + B2 + E
A2 + B2 + E
A2 + B1 + 2E
A2 + B1 + 2E
A1 + B2 + 2E
A1 + B2 + 2E
2A1 + 2B1 + 2E
2A2 + 2B2 + 2E
Table III. Reduction of the irreps of Optimal Group G of the 4 4 model in the
point Group.
17.2 The W = 0 Theorem ensures that no double occupancy is possible for

pairs in the irreps A1 , B2 , 1 , 2 , 1 , 2 , 3 , 4 , 2 , 3 , 1 , 2 , 3 , 1 and
2 .
426
Solutions
18.1 In the repulsive case, the Hamiltonian (14.58) reads
U t t 0
t 0 0 t
H=
t 0 0 t.
0 t tU
(21.31)
on the basis (a a , a b , b a , b b )) where a

and b are the two sites.
For U > 0, the ground state eigenvalue is = 12 (U 16 + U 2 ) (we measure
energies in t units);

2q
L=Q
q2
1
with Q =
, q = (U + 16 + U 2 ). For U = 7, =
2
2
16+U
+U 16+U

0.181492 0.683418
0.531129, L =
.
0.683418 0.181492
In the attractive case (21.32)
0 t t0
= H(t, U ) + U = t U 0 t , U > 0
H
(21.32)
t 0 U t
0 t t0
on the basis ( a , b , a , b )) where = b and = a for the
new Fermions. The transformed matrix is

=Q
2 q
L
q 2
1
=
with Q
, q = (U + 16 + U 2 ). For U = 7, =
2
2
16+U
U 16+U

0.683418 0.181492
=
0.531129, L
.
0.181492 0.683418
18.2 With 2 fermions, Q = P = {1, 2}, Q = P = {2, 1}, let k1 < k2 denote
the momenta, and we may write

fQ (x1 , x2 ) = [QP ]ei(k1 x1 +k2 x2 ) + [QP ]ei(k2 x1 +k1 x2 ) x1 x2
fQ (x1 , x2 ) = [Q P ]ei(k1 x2 +k2 x1 ) + [Q P ]ei(k2 x2 +k1 x1 ) x1 x2
con
[QP ] =
U
[QP ] + {sin(k1 ) sin(k2 )} [Q P ]
2i
sin(k1 ) sin(k2 ) +
U
2i
(21.33)
and so on. For parallel spins U cannot act, sectors Q and Q are disjoint and
the coecients are all determined by [QP ] = [Q P ] = [Q P ] = [QP ] and
normalization.
Solutions
427
For antiparallel spins, the periodicity conditions are also needed. In sector
Q, with 1 x1 x2 Ns let the amplitude be fQ (x1 , x2 ). Let the electron
1 sit at be site x1 ; if we decide to relabel x1 + Ns that site, the amplitude
does not change, but we are now in sector Q . So,
fQ (x1 , x2 ) = fQ (x1 + Ns , x2 ).
(21.34)
Now,
[QP ]ei(k1 x1 +k2 x2 ) + [QP ]ei(k2 x1 +k1 x2 )
= [Q P ]ei(k1 x2 +k2 (x1 +Ns )) + [Q P ]ei(k2 x2 +k1 (x1 +Ns ))
implies
[QP ] = [Q P ]eik1 Ns

[QP ] = [Q P ]eik2 Ns .
(21.35)
(21.36)
We eliminate [Q P ] by (18.78) and get

[QP ] = ([QP ] + [QP ] [Q P ])eik2 Ns ,
which yields:
[ QP ](1 eik2 Ns ) = ([QP ] [Q P ])eik2 Ns
= ([Q P ]eik1 Ns [Q P ])eik2 Ns .
(21.37)
Total momentum is conserved; hence

ei[k1 +k2 ]Ns = 1
and so
(21.38)
[Q P ] = [QP ].
Introduce the phase

(0, 2),
and since k1 < k2 ,, let
ik N
e 1 s = ei , Ns k1 = 21 , 1 Z
eik2 Ns = ei , Ns k2 = 22 + , 2 Z.
(21.39)
Integers 1 2 are Bethe quantum numbers that label the state with
[QP ] = [QP ]ei .
(21.40)
In sector Q one nds:

[Q P ]ei(k1 x2 +k2 x1 ) + [Q P ]ei(k1 x1 +k2 x2 )
= [Q P ]e
i(k1 x1 +k2 x2 +k2 Ns )
+ [Q P ]ei(k1 x2 +k2 x1 +k1 Ns ) .
(21.41)
428
Solutions
This yields

[Q P ] = [Q P ]ei

[Q P ] = [Q P ]ei .
(21.42)
Using (21.40,21.42, 18.78 ) one gets

[QP ] = [Q P ],
[QP ] = [Q P ] = [QP ]ei .
(21.43)
Then, (21.33) becomes

ei =
sin(k2 ) sin(k1 )
sin(k2 ) sin(k1 ) +
U
2i
U
2i
(21.44)
and the (un-normalized) wave function reads

fQ (x1 , x2 ) = ei(k1 x1 +k2 x2 ) + ei(k2 x1 +k1 x2 +) x1 x2
.
fQ (x1 , x2 ) = ei(k1 x2 +k2 x1 ) + ei(k2 x2 +k1 x1 +) x1 x2
This may be taken real if s are opposite; then we may write
k1 = k2 = k > 0;
besides,
ei =
i sin(k)
i sin(k) +
U
4
U
4
We can simplify by an algebraic transformation based on

ei + 1
.
cot
= i i
2
e 1
(21.45)
(21.46)
This leads to
U cot( ) = 4 sin(k).
2
For odd n we may also write
4
tan( n ) = sin(k).
2
2
U
(21.47)
(21.48)
Introducing
0 (p) = 2 arctan(
2p
),
U
(21.49)
we get
= n + 0 (2sin(k)).
One can verify the above results for Ns= 3, in the singlet sector.Direct di2
2
(2 times), 1+U+ 92U+U
agonalization yields the eigenvalues 1+U 92U+U
2
2
2
2
(2 times), 2+U+ 36+4U+U
and for the ground state 2+U 36+4U+U
.
2
2
Solutions
429
These results are reproduced by the Bethe ansatz with Equation (18.77).
From = 0, that is k = 3 the solution of (21.47)

= 6 arccos[
U
5
+
8 16
36 + 4U + U 2
]
16
leads to the ground state and to the last but one. The other states arise from
= 1 and = 2, and each choice yields both eigenvalues.
References
1. Charles P. Enz,A Course on Many-Body Theory Applied to Solid-State Physics,

World Scientic (1992)
2. A.L. Fetter and J. D. Walecka, Quantum theory of Many-particle systems,
McGraw Hill, New York (1971)
v
3. Evgenij M. Lifsits and Lev P. Pitaevskij, Teoria dello Stato Condensato, Ed.
Mir, Moskow (1981)
4. L S Cederbaum 1975 J. Phys. B: At. Mol. Phys. 8 290-303
5. G. Grosso e G. Pastori Parravicini Solid State Physics, academic Press Cambridge (2000)
6. H.B.G. Casimir, Proc. Kon. Ned. Akad. Wetensch. B51, 793 (1948)
7. S. Lamoreaux, Phys Rev Lett, 78, p5 (1996)
8. J. Schwinger, U.S. Atomic Energy Commission Rept. NYO-3071 (1952)
9. P. W. Anderson, Phys. Rev. 124, 41 (1961)
10. D.M. Newns, Phys. Rev. 179, 1123 (1969)
11. P. Coleman, cond-mat/0206003
12. M.L. Knotek and P.J. feibelman, Phys. Rev.Lett. 40, 964 (1978)
13. Michele Cini, Phys. Rev. B32,1945 (1985)
14. S. Pancharatnam, Proc. Indian Acad. Sci. A 44, 247 (1956)
15. D. Jaksch, C. Bruder, J.I. Cirac, C. W. Gardiner and P. Zoller, Phys.Rev
Letters 81, 3108 (1998).
16. R.Resta, J. Phys.: Condens. Matter 12 R107 (2000)
17. J. Hubbard, Proc, R. Soc. Lond. A276, 238 (1963)
18. Michele Cini, Gianluca Stefanucci, Enrico Perfetto and Agnese Callegari, J.
Phys.: Condens Matter 14, L709 (2002).
19. P. Coleman, cond-mat/0206003
20. A.C. Hewson,The Kondo problem to Heavy fermions ,Cambridge University
Press (1997)
21. C. Lacroix, J. Phys. F: Metal Phys. 11(1981) 2389
22. O. Gunnarsson and K. Sch
onhammer, Phys. Rev. B22, 3710 (1980)
23. Brian L. Silver, Irreducible Tensor Methods, Academic Press New York (1976)
24. W. Borchardt-Ott,Crystallography, Springer -VerlagBerlin (1995)
25. Leonard I. Schi, Quantum Mechanics,McGraw-Hill New York
26. O. Gunnarsson,K. Sch
onhammer, J.C. Fuggle and R. L
asser, Phys. Rev. B23,
4350 (1981)
27. O. Gunnarsson and K. Sch
onhammer, Phys. Rev. B28 , 4315 (1983)
28. C. Zener, Phys. Rev. 81, 440 (1951).
29. Ryogo Kubo, J. Phys. Soc. Japan12, 570 (1957).
30. L.C. Davis and L.A. Feldkamp,Phys. Rev.B 15 2961 (1977)
432
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
63.
64.
65.
66.
References
E.N. Lassettre, Suppl. Radiation Research 1, 530 (1959)
U. Fano,Phys. Rev.124 1866 (1961)
P.W. Anderson,Phys. Rev.124 41 (1961)
J. Kondo, Prog. Theor. Phys. 32, 37 (1964).
J. Kondo, Theory of dilute Magnetic Alloys, in Solid State Physics, Ed. Seits
23, (1969).
C. Verdozzi, Y. Luo, and Nicholas Kioussis, Phys. Rev.B 70, 132404 (2004)
E. H. Lieb, Phys. Rev. Lett. 62, 1201 (1989).
B. I. Lundqvist, Phys. Kondensierten Materie 9, 236 (1969).
David C. Langreth, Phys. Rev. B 1, 471 (1970).
R. Brout and P. Carruthers, Lectures on the Many-Electron Problem (WileInterscience, NewYork, 1963)
Gerald D. Mahan, Phys. rev. B 11 4814 (1975 )
Gerald D. Mahan, Many-Particle Physics, Plenum Press, New york and London
(1990)
H. Ehrenreich and M. H. Cohen, Phys. Rev.115, 786 (1959)
E. N. Economou : Greens functions in Quantum Physics, Springer - Veriag,
1979. Second edition 1983
Michele Cini and Andrea Dandrea, J. Phys. C 21, 193 (1988)
M.Cini, A. DAndrea and C. Verdozzi, International Journal of Modern Physics
B9, 1185 (1995)
M.Cini, Il Nuovo Cimento 9D, 1515 (1987)
A. Blandin, A. Nourtier and D.W. Hone, J. Phys. (paris) 37, 369 (1976)
Adolfo Avella, Ferdinando Mancini and Roland M
unzner, Phys. Rev. B63,
645117 (2001)
Ferdinando Mancini and Adolfo Avella, cond-mat/0006377
Ferdinando Mancini and Adolfo Avella,The Hubbard model within the equation
of motion approach, advances in Physics 53, 537 (2004)
J. Lindhard, Kgl. Danske Videnskab. Selskab, Mat.-Fys. Medd.,28, no.8 (1954)
Lars Hedin, Phys.Rev.139, A796 (1965).
G.Onida,L.Reining and A.Rubio, Rev. Modern Phys. 74, 601 (2002)
J.M.Luttinger and J.C. Ward,Phys. Rev. 118, 1417 (1960.
G.Baym and L. Kadano, Phys. Rev. 124, 287 (1961);
G. Baym, Phys. Rev. 127, 1391 (1962).
C. Lanczos, J. Res. Natl. Bur. Stand. 45, 255 (1950).
R. Haydock, V. Heine and M.J. Kelly, J. Phys. C 5, 2845 (1972), R. Haydock, Solid State Physics Ed. Eherenreich, F. Seitz and D. Turnbull (London,
Academic Press, 1980), R. Haydock and C.M.M. Nex, J. Phys. C 18, 2235
(1985).
F. Cyrot-Lackmann, J. Phys. C:Solid State Phys. 5, 300 (1972); F. CyrotLackmann, M.C. Desjonqueres and J.P. Gaspard, J. Phys. C:Solid State Phys.
7, 925 (1974)
R. Landauer, IBM J. Res. Dev. 1, 233 (1957); Philosof. Mag. B21, 863 (1970)
C. Caroli,R. Combescot, P. Nozieres and D.Saint-James, J.Phys. C:Solid State
Phys. 4, 916 (1971); ibidem 5, 21 (1972)
Michele Cini, Phys. Rev. B22, 5887 (1980)
Yigal Meir and Ned S. Wingreen, Phys. Rev. Letters 68, 2512 (1992).
M.V. Berry, Proc. R. Soc. Lond. A392, 45 (1984)
C. Verdozzi and M. Cini, Phys. Rev. 51, 7412 (1995).
References
433
67. J. Kanamori, Prog. Theor. Phys. 30, 275 (1963).

68. H. Bethe, Z. Phys. 71, 205 (1931), Zur Theorie der Metalle. Eigenwerte und
Eigenfunktionen der linearen Atomketten.
69. C.W. Beenakker, Phys. Rev.B44, 1646 (1991)
70. E. Perfetto and Michele Cini, Phys. Rev. B 71, 014504 (2005)
71. Enrico Perfetto and Michele Cini, Phys. Rev.B 69, 092508 (2004)
72. Michael Karbach and Gerhard M
uller, cond-mat/9809162; cond-mat/0008018
73. E.H. Lieb and F.Y. Wu, cond-mat/0207529
74. C.N.Yang, Phys. Rev. Letters 19, 1312 (1967).
75. Michele Cini, Solid State Commun. 20, 605 (1976);
76. Michele Cini, Solid State Commun. 24, 681 (1977)
77. Michele Cini, Phys.Rev. B17, 2486 (1978)
78. M.Cini, A. DAndrea and C. Verdozzi, International Journal of Modern Physics
B9, 1185 (1995).
79.
80. Q Niu and D J Thouless, J. Phys. A:MAth . Gen. 17 (1984) 2453
81. Walter Kohn, Phys. Rev. 133 (1964) A171
82. L.P.Kadano and G.Baym, Quantum Statistical Mechanics, W.A.Benjamin,
Inc., New York (1962)
83. M. Gell-Mann and F. Low, Phys. Rev. 84, 350 (1951)
84. L.V. Keldysh, J.Exptl. Theoret. Phys. (U.S.S.R.) 47, 1515 (1964) [English
translation Soviet Phys. JETP 20, 1018 (1965)]
85. T.E.Feuchtwang,Phys. Rev. B10, 4121 (1974) and Phys. Rev. B12, 3979
(1975)
86. C.R. Natoli et al., Phys. Rev. B 42, 1944 (1990)
87. D.C. Langreth and J.W.Wilkins, Phys.Rev.B6, 3189 (1972); D.C. Langreth,
1975 N.A.T.O. Advanced Study Institute, Antwerp, Belgium, on Linear and
Nonlinear Electron Transport in Solids;
88. Jhy-Jiun Chang and D.C. Langreth, Phys. Rev. B8, 4638 (1973)
89. L.D.Landau e E.M.Lifshitz, Fisica Cinetica, Editori Riuniti, Roma (1984)
90. R.D. King-Smith and D. Vanderbilt, Phys. Rev. B 47, 1651 (1993)
91. M.Cini, Surface Sci. 87, 483 (1979)
92. M.J. Lighthill, An introduction to Fourier analysis and generalised functions,
Cambridge University Press Cambridge 1958
93. Melvin Lax, Symmetry principles in Solid State and molecular Physics, ISBN:
0486420019 Dover Publications
94. K.Siegbahn et al. Esca applied to free molecules,North-Holland Publishing
Company Amsterdam London (1969).
95. P. Weightman, J. Phys. C (1982)
96. Michele Cini, Corso di Fisica Atomica e Molecolare-Introduzione alla Fisica
Teorica per Strutturisti,II Edizione, Edizioni Nuova Cultura, Roma (1998)(in
Italian)
97. E. Feenberg, Phys.Rev.74,206 (1948); Phys.Rev.74,664 (1948)
98. S.Swain, continued fraction methods in atomic physics, in Advances in atomic
and molecular Physics, 22, 387 (1986)
99. Michele Cini, Il Nuovo Cimento 8D, 333 (1986)
100. Hitoshi Sumi, J. Phys. Soc. Japan 36, 770 (1974)
101. Matias Wagner, Phys. Rev. B44, 6104 (1991)
102. S. Kurth,G.Stefanucci,C.O. Ambladh, A. Rubio and E.K.U. Gross, Phys. Rev.
B72, 035308 (2005)
434
References
103. G.Stefanucci and C.O. Ambladh, Phys. Rev.B 69, 195318 (2004)
104. Michele Cini, Phys. Rev. B43,4792 (1991); Michele Cini,R. Del Sole, Jang Guo
Ping and L. Reining, Calculation of the linear and nonlinear optical properties
of Si(111)1x1-As, Proceedings of the Epioptic meeting, Berlin, June 1991.
105. Y.R. Shen, The principles of Nonlinear Optics, John Wiley and Sons, New
York (1984).
106. Schuda, C R Stroud Jr and M Hercher, J. Phys. B: Atom. Molec. Phys. 7,
L198 (1974)
107. P.Auger, J. Physique Radium 6, 205 (1925)
108. J.J.Lander, Phys. Rev. 91, 1382 (1953)
109. Claudio Verdozzi, Michele Cini and Andrea Marini, Jnl. Electron Spectroscopy
and Related Phen VOL.117-118, 41-55 (2001)
110. C.J.Powell,Phys. Rev. Letters 30, 1179 (1973)
111. L.I. Yin, T.Tsang and I. Adler,Physics Letters 57A,193 (1976); Phys. Rev. B
15, 2974 (1977)
112. G.A. Sawatzky, Phys. Rev. Letters 39, 504 (1977)
113. Knotek and Feibelman, Phys. Rev. 1978
114. J.E. Ingleseld, J. Phys. C:Solid State Phys. 14, 3795 (1981)
115. S. Crampin, J. Phys.: Condens. Matter 16, 8875 (2004)
116. Richard D. Mattuck, A Guide to Feynman Diagrams in the Many-Body Problem,Dover, New York
117. John Inkson, Many-Body theory of Solids - An Introduction,Plenum Press New
York 1984
118. R. P. Feynman,Statistical Mechanics - A set of lectures, W.A. BENJAMIN,
INC, Reading, Massachussetts (1972)
119. G. Vignale and Mark Rasolt, Phys. Rev. Letters 59, 2360 (1987)
120. Robert van Leeuwen, Nils Erik Dahlen, Gianluca Stefanucci, Carl-Olof Almbladh and Ulf von Barth, Lectures Notes in Physics, Springer Verlag, 2005 (to
appear); also cond-mat/0506130.
121. P. Hohenberg and W.Kohn, Phys. Rev. 136, B864 (1964)
122. W. Kohn and L.J. Sham, Phys. Rev. 140, A1133 (1965)
123. L.J. Sham and M. Schl
uter, Phys. Rev. Letters 51, 1888 (1983)
124. Michele Cini and A. DAndrea, J. Phys. C 16, 4469 (1983)
125. R.M. Dreizler and E.K.U. Gross, Density Functional Theory, Springer Verlag
1990
126. Ulf von Barth, Niels Erik Dahlen, Robert van Leeuwen, and Gianluca Stefanucci,Phys. Rev. B72, 235109-1 (2005)
127. Andrea Marini and Michele Cini, Phys. Rev. B 60, 11391 (1999)
128. M. Cini and V. Drchal, J. Phys.: Condens. Matter 6, 8549 (1994);M. Cini and
V. Drchal, J. Electron Spectrosc. Relat. Phen.72, 151 (1995)
129. M.J. Stott and E. Zaremba, Phys Rev A 21, 12 (1980)
130. A. Zangwill and P. Soven, Phys Rev A 21, 1561 (1980)
131. E. Runge and E.K.U. Gross, Phys. Rev. Letters 52, 997 (1984).
132. J.R. Schrieer and D. C. Mattis, Phys. Rev 140, A1412 (1965);B. Kj
ollerstr
om, D.J. Scalapino and J.R. Schrieer,Phys. Rev 18, 665 (1966)
133. V. Galitzkii, Soviet Phys. JETP 7, 104 (1958).
134. M. Cini and C. Verdozzi, Solid State Commun.57, 657 (1986);M. Cini and C.
Verdozzi, Nuovo Cimento 9D, 1 (1987).
135. A. Pernaselci and M. Cini, J. Electron Spectrosc. and Relat. Phen. 82,79
(1996).
References
136.
137.
138.
139.
140.
141.
142.
143.
144.
145.
146.
147.
148.
149.
150.
151.
152.
153.
154.
155.
156.
157.
158.
159.
160.
161.
162.
163.
164.
165.
166.
167.
168.
169.
170.
171.
172.
435
M. Hindgren, Ph.D. Thesis Lund University, 1997.

Niels Eric Dahlen and Ulf von Barth, Phys. Rev. B 69, 195102 (2004)
J.P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Letters 77, 3865 (1996)
L.G.Molinari, Phys. Rev. B 71, 113102 (2005)
M. Revzen,T. Toyoda, Y. Takahashi and F. C. Khanna, Phys. Rev. B 40, 769
(1989)
J.C. Ward, phys. Rev. 78, 182 (1950); Y. Takahashi, Nuovo Cimento 6, 371
(1957).
Andrei, Furuya and Lowenstein, Rev. Modern Phys. 55, 331 (1983)
H.B. Thacker,Rev. Modern Phys. 53, 253 (1981)
E.U. Condon and G.H. Shortley, The theory of Atomic Spectra, Cambridge
University Press, Cambridge (1935)
John C. Slater, Quantum theory of Atomic Structure, Volume II, McGraw-Hill
Book company, New York (1960)
J. F. Janak, Phys. Rev. B 18, 7165 (1978)
Mitchel Weissbluth, Atoms and Molecules, Academic Press San Diego (1978)
Nielsen C.W. and Koster G.F. Spectroscopic coecients for pn ,dn and f n
Congurations, MIT press, Camb`
uridge, Massachussetts (1963)
Andrea Dal corso, Phys. Rev. B 64 235118-1 (2001).
M.Cini, R.Del Sole and L. Reining, Surface Sci. 287/288, 693 (1993)
M. Cini, Surface Review and Letters 1,4 , 443 (1994)
K.G. Dyall et al., Computer Physics Communications B 55 425 (1989).
C. Verdozzi, M. Cini, J.F. McGilp, G. Mondio, D. Norman, J.A. Evans, A.D.
Laine, P.S. Fowles, L. Duo and P. Weightman, Phys. Rev. B 43 9550 (1991)
S. aksela, M. Harkoma, M. Pohjola and H. Aksela, J. Phys. B 17, 2227 (1984)
R.J. Cole, C. Verdozzi, M. Cini and P. Weightman, Phys. Rev. B 49 13329
(1994)
Morton Hamermesh, Group Theory and its application to physical problems,Dover Publications, Inc., New York (1962).
G.A. Baker Jr and P. Graves-Morris, Pade approximants, Encyclopaedia of
Mathematics and its Applications Volume 14, Cambridge MA Addison-Wesley
(1981).
K.G. Dyall et al., Computer Physics Communications 55,425 (1989)
DKG de Boer, C Haas and G A Sawatzky, J. Phys. F: Met.Phys. 14, 2769
(1984)
P. Hedegard and F Hillebrecht, Phys. Rev. B34 3045 (1986)
H.ksela and S. Aksela, J.Phys. B 7,1262 (1974).
N.E. Bickers, D.J. Scalapino and S.R. White, Int.J. Mod. Phys. 1, 687 (1987)
N.E. Bickers, D.J. Scalapino and S.R. White, Phys. Rev. Letters 62, 961 (1989)
R. Shankar, rev. Mod. Phys. 66, 129 (1994)
D. Zanchi and HJ Schulz, Phys. Rev. B63, 13609 (2000)
C.J. Halboth and W. Metzner Phys. Rev. Letters 61, 7364 (2000)
C. Honerkamp and M. Salmhofer, Phys. Rev. Letters 64, 184516 (2001)
P.W.Anderson, G. Baskaran, Z. Zou and T. Hsu, Phys. Rev. Letters 58, 2790
(1987)
P.W.Anderson and Z. Zou, Phys. Rev. B 60, 132 (1988)
G. Su and M. Suzuki, Phys. Rev. B58, 117 (1998)
Michele Cini and Adalberto Balzarotti, Nuovo Cimento D 18, 89 (1996)
Michele Cini and Adalberto Balzarotti, Phys. Reb B56, 14711 (1997)
436
References
173. Adalberto Balzarotti, Michele Cini, Enrico Perfetto and Gianluca Stefanucci,
J. of Phys.: Condens. Matter 16 R1387 (2004)
174. W.Kohn and J.M. Luttinger, Phys. Rev. Letters 15, 524 (1965)
175. J.G. Bednorz and K.A. M
uller, Z. Phys. B64, 189 (1986)
176. RW Richardson, Phys. Rev. 141, 949 (1966)
177. G. Fano, F. Ortolani and A. Parola, Phys. Rev. B 42, 6877 (1990); A. Parola,
S. Sorella, M. Parrinello and E. Tosatti, Phys. Rev. B 43, 6190 (1991); G.
Fano, F. Ortolani and A. Parola, Phys. Rev. B 46, 1048 (1992).
178. C. A. Balseiro, A. G. Rojo, E. R. Gagliano and B. Alascio, Phys. Rev. B 38,
9315 (1988).
179. J. E. Hirsch, S. Tang, E. Loh and D. J. Scalapino, Phys. Rev. Lett. 60, 1668
(1988).
180. M. Ogata and H. Shiba, J. Phys. Soc. Jpn. 57, 3074 (1988).
181. J. Bardeen, L.N. Cooper and J. R. Schrieer, Phys. Rev. 108, 1175 (1957):
J.R. Schrieer, The theory of superconductivity, Perseus Books, Reading, Massachussets (1964)
182. E. Cappelluti, B. cerruti and L. Pietronero, Phys. Rev.B 69 , 161101 (2004)
183. M. S. Hybertsen, M. Schl
uter and N. Christensen, Phys. Rev. B 39, 9028
(1988); M. Schl
uter, in Superconductivity and Applications, edites by H.
S. Kwok et al. (Plenum Press, New York, N. Y.) p. 1 (1990).
184. F. C. Zhang and T. M. Rice, Phys. Rev. B 37, 3759 (1988)
185. M. Cini, A. Balzarotti, R. Brunetti, M. Gimelli and G. Stefanucci, Int. J. Mod.
Phys. B 14, 2994, (2000).
186. M. Cini and A. Balzarotti, Solid State Comm. 101, 671 (1997); M. Cini, A.
Balzarotti, J. Tinka Gammel and A. R. Bishop, Il Nuovo Cimento D 19, 1329
(1997).
187. M. Cini, A. Balzarotti and G. Stefanucci Eur. Phys. J. B 14, 269 (2000).
188. E. Perfetto, G. Stefanucci, and M. Cini, Phys. Rev. B 66, 165434 (2002); E.
Perfetto , G. Stefanucci , M. Cini Eur. Phys. J.B 30, 139 (2001).
189. M. Cini, E. Perfetto and G. Stefanucci, Eur. Phys. J. B 20, 91 (2001).
190. Michele Cini, Adalberto Balzarotti, Raaella Brunetti, Maria Gimelli and Gianluca Stefanucci, International J. Modern Phys. 14, Nos. 25-27, pag. 2994
(2000)
191. C.P. Lund, S.M. Thurgate and A.B. Wedding,Phys Rev. B55, 5455 (1997)
192. Stefano Bellucci, Michele Cini, Pasquale Onorato and Enrico Perfetto, J.
Phys.: Condens. Matter 18,S2069 (2006)
193. G.W. Fernando, A.N. Kocharian, K. Palandage, Tun Wang and J.W. Davenport, Cond-mat/0608579
194. David J. Griths, Introduction to Quantum Mechanics,Prentice Hall , upper
Sadle River, New Jersey 07458 (1994)
195. A.Callegari, M. Cini, E. Perfetto and G.Stefanucci, E Eur. Phys. J. B 34, 455
(2003)
196. M. Cini, G. Stefanucci and A. Balzarotti, Eur. Phys. J. B 10, 293 (1999).
197. M. Kociak et al., Phys. Rev. Lett. 86, 2416 (2001). A. Kasumov et al., Phys.
Rev. B 68, 214521 (2003)
198. F. Mancini and A. Avella, Advances in Physics 53, 537 (2004)
199. Satoro Odashima, Adolfo Avella and Ferdinando Mancini, Phys. Rev. B 72,
205121 (2005)
200. G. Bressi, G. Carugno, R. Onofrio and G. Ruoso, Phys. Rev. Lett. 88 041804
(2002)
References
437
201. M. Cini and C. Verdozzi, J.Phys.(Condensed Matter) 1, 7457 (1989)

202. Bernd Thaller, The Dirac Equation, Springer-Verlag Berlin Heidelberg (1992)
203. W. Greiner, Relativistic Quantum Mechanics- Wave Equations, SpringerVerlag Berlin Heidelberg (1990)
204. D. Chattarji, The theory of Auger Transitions, Academic Press London (1976)
205. D.C. Langreth, Phys. Rev. 182, 973 (1969)
206. L.D. Faddeev, Sov. Phys. JETP 12, 1014 (1961)
207. Stefano Bellucci, Michele Cini, Pasquale Onorato and Enrico Perfetto, Phys.
Rev. B 75, 014523 (2007)
208. H. Mori, Prog. Theor. Phys. 33, 423 (1965); 34, 399 (1965)
209. M. Howard Lee, Phys. Rev.E61, 3571 (2000)
210. Umerto Balucani,M. Howard Lee and Valerio Tognetti, Dynamical Correlations, Physics Reports 373, 409 (2003)
Index
C3v , 137
C3v , 137
GL(n), 133
O(n) , 134
O+ (3), 201
SL(n) , 134
SO(3), 145
SO(4), 145
SU (5), 201
U (1), 145
SCF method, 33
++ , 81
1-2-3 theory, 270
Abelian Group, 133
Abrikosov-Suhl peak , 96
accidental degeneracy, 371
adiabatic electronic Hamiltonian, 188
adiabatic ionization energy, 115
adiabatic theorem, 359
advanced Greens functions, 57, 288
AES, 109
ammonia, 158
Anderson, 285
anti-Stokes, 335, 340
APECS, 109, 267, 270
APEX, 121
ARUPS, 109
atomic levels, 42
atomic terms, 41
Auger, 210
Auger eect, 48
Auger Electron Spectroscopy, 109
Auger matrix elements, 52
Auger Photoelectron Coincidence
Spectroscopy, 109
Auger selection rules, 50
Auger spectroscopy, 56
Bare Ladder Approximation, 266

Baym-Kadano theorem, 275
Berry phase, 358
Bethe ansatz, 391, 393, 406
Bethe-Salpeter equation, 246
Biedernharn-Johnson-Lippman
operator, 147
bipartite graph , 67
bipartite lattice, 383
Bloch, 136, 169
Bohr-van Leeuwen theorem, 387
Born-Oppenheimer approximation, 190
boson propagator, 63
Breit interaction , 52
Brillouin zone, 66, 170
Burnside theorem, 155
Carbon Nanotubes , 369
Casimir eect, 7
Cayley theorem, 138
central eld, 38, 39
characters, 143
characters of angular momentum
eigenstates, 160
chemical shift, 50, 111
chemisorption, 285
Cini-Sawatzky theory, 118, 127, 264
classical radius, 341
Clebsh-Gordan, 42
Clebsh-Gordan coecients, 186
Clebsh-Gordan coecients, 186
coherent light, 345
coherent state, 116
Compact Lie Group, 145
Complex time, 287
composite operators, 213
conguration, 39
conguration interaction, 34
440
Index
conjugate irreps, 200

conserving approximation, 272
continuity equation, 273
correlation energy , 31
coset, 138, 139
Coster-Kronig, 121
Covering Group, 145
crystal eld, 162
current-voltage characteristics, 308
Density Functional Theory, 73, 282
desorption, 122, 210, 299
diagram rules, 224
dielectric function, 251, 259
dierential cross section, 339
dipole approximation, 336, 344
Dirac characters, 140
Dirac equation, 73
Dirac-Fock , 52
direct product, 134
double groups, 186
dynamical Jahn-Teller eect, 190
Dyson equation, 70, 239, 241, 292
Einstein coecient, 349
electron anity, 77
embedding, 70, 71
equal-times rule, 237, 291
equation of motion, 61, 62, 74, 82, 83,
207, 211, 285, 307
ESCA, 109
Excitation Amplitudes, 326
Faddeev, 269
Fano, 81, 212, 241, 242, 285, 299, 326
Feenberg, 329
fermion annihilation operator, 9
fermion creation operator, 9
Feynman, 285
eld operators, 11
nal state eects, 111
uctuation-dissipation theorem, 79
uxon, 358
fractional parentage coecients, 42
Franck-Condon , 116
Friedel oscillations, 260, 367
functional chain rule, 253
functional derivative, 249
Galilean covariance, 356

Galitzkii, 264
Gell-Mann and Low Theorem, 233
generalized Bethe ansatz, 406
glide plane, 175
Goldstone diagrams, 221
Greens functions hierarchy, 212
ground state energy, 208, 277
Group, 133
Group property, 23
Group of the wave vector, 176
GW approximation, 255, 266, 275
Hartree potential, 31
Hartree-Fock, 29, 272
Haydock, 313
Hedins equations, 255
Heisenberg model, 387
Heisenberg picture, 287
Herglotz, 87, 266, 270, 272, 314, 323
Hewson-Newns model, 326
Hohenberg-Kohn theorem, 279
Hubbard, 13, 212
Hund, 42
Hydrogen, 13
infrared, 340
Ingleseld, 71
initial state eects, 111
interaction picture, 287
Intermediate Coupling, 46
International notation, 175
inversion symmetry, 171
ionization potential, 77
irreducible interaction, 246
Irreducible Representations, 140, 141
irreducible self-energy, 239
irreducible tensors, 181
irreducile tensors, 181
irrep, 141
Jahn-Teller Theorem, 189
Janak theorem, 282
Jellium, 35, 240, 281
Josephson, 381
Josephson eect, 308, 378
junctions, 306
Kadano-Baym method, 286
Index
Kanamori solution, 122, 127
Keldysh, 210, 286, 294
Klein-Gordon equation , 65
Knotek-Feibelman mechanism, 121,
122, 210, 212, 270
Kohn-Sham equations, 281, 282
Kondo, 95, 393
Kondo Model, 99
Kondo peak , 96
Kondo temperature, 105
Koopmans, 33
Kramers, 340
Kramers Theorem, 172
Kubo, 75, 285
Kubo formulae, 73
L-S multiplets, 40
Lagrange theorem, 139
Lanczos, 313, 325
Landauer, 304
Langreth, 288
laser, 345
LCAO, 93
Lehmann, 76, 77
lesser and greater Greens functions, 57
Lie Group, 133, 144
Lieb, 324, 383
Ligand group orbitals, 161
Linked Cluster Theorem, 225, 226, 231
linked diagrams, 225
Lippmann-Schwinger equation, 70
little Group, 176
Local Density Approximation, 281
local density of states, 66, 86
LOT, 152
Low Density Approximation (LDA),
264
magnon, 388, 391
many-body Hamiltonian, 12
master equation, 305
Matsubara, 60, 290
mean eld, 31
Migdal theorem, 255
mixed coupling scheme , 52
mixed valence , 95
molecular orbitals, 158
moments, 322
non-adiabatic operator, 194

Normal divisor, 134
O(4) Group, 147
O(n), 134
Open-path geometric phase, 360
optical electron, 41
orbitals of CH4 , 159
orbitals of NH3 , 157
Orthogonal polynomials, 318
Orthohelium, 31
Pade, 318, 323
Pancharatnam phase, 357
partition function, 224
partition-free transport theory, 306
Peierls, 358
permanent, 9
permutations, 1, 2
photoemission, 106
photon propagator , 65
photon scattering, 337
plasmon, 326
plasmon satellites, 114
Point Groups, 135
Poisson statistics, 117
polar form, 182
Polarization bubble, 251
polarization part, 238
projection operators, 158
propagator, 61
pseudo-Jahn-Teller eect, 191
Quadratic Response Formalism, 127
quasi-particle, 241
quotient Group, 139
Raman, 335
Raman eect, 185
Rayleigh, 335
rearrangement theorem, 138
rectangular level, 120, 316, 328
Regular representation, 155
representation, 138
resonances, 81
resonant Raman scattering, 340
Resta operator, 364
restricted HF, 34
retarded Greens functions, 57, 288
441
442
Index
Runge-Lenz Vector, 146, 147

Russell-Saunders, 41, 46
s-d model, 98
scattering amplitude, 246
Sch
onies, 135
Schrieer-Wol transformation., 98
screw axis, 175
second character orthogonality theorem,
170
selection rules, 184
self-consistent eld, 31
self-energy, 89, 90, 94, 107, 127, 212,
241245, 247, 249, 258, 265,
274276, 283, 293, 300, 302
self-energy part, 239
semi-elliptic density of states, 94, 315
semi-simple Group, 138
semidirect product, 174
seniority number, 35, 201, 203
shake-o , 113
shake-up, 113
Sham - Schl
uter equation, 283
Shapiro, 381
simple Group, 138
sky, 340
Slater integrals, 44
Sommerfeld, 35
spectral function, 77, 78
spin eigenfunctions, 178
spin-disentangled diagonalization, 323
spin-orbit interaction, 39, 186
spin-polarized HF, 34
spontaneous emission, 349
sputtering, 285, 299
star of k , 176
Stokes, 335, 340
SU(n), 134
subgroup, 134
superconducting ux quantization, 368,
369, 378
Symmetric Group, 199
symmetry, 135
symmorphic Group, 174
t matrix, 70
temperature Greens function, 59
temperature Greens function , 62
tensor, 340
thermodynamic potential, 224

Thomson, 341
three-diagonal matrix, 331
tight-binding Hamiltonian, 66
Time Dependent Density Functional
Theory, 308
time reversal, 171
Time-dependent Density Functional
theory, 282
time-dependent problems, 209
time-ordered nite temperature
propagator, 59
time-ordered Greens function, 58
Time-ordered propagator, 231
Topologically equivalent diagrams, 236
Tr, 23, 275
transport, 210
transpositions, 1, 2
tunneling, 305
two-hole resonances, 120, 121, 270
U(n), 134
U<0, 270
Ultraviolet Photoemission Spectroscopy, 109
unrestricted HF, 34
UPS, 109
vacuum amplitude, 75
vacuum level, 33
Van Hove, 317
variational principle, 279
Variational principle, 16, 277
Vector potential analogy, 361
vertex, 252
vertical ionization energy, 115
vibration, 326
vibrations of Ammonia, 167
vibrations of Benzene, 168
vibrations of Methane, 168
vibrations of Water, 166
vibronic couplings, 190
Vibronic States, 193
virtual orbitals, 34
W=0 pairing, 369
W=0 theorem, 373
Ward identities, 278
water, 137
Index
Wicks theorem, 219
Wigner Eckart theorem, 45
Wigner matrices, 145
Young diagrams, 177

Young tables, 177, 178
Yukawa potential, 260
X-Ray Photoemission Spectroscopy,

109
XPS, 109
Zener model, 98
443

(Michele Cini) Topics and Methods in Condensed Mat

Uploaded by

Copyright:

Available Formats

(Michele Cini) Topics and Methods in Condensed Mat

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Michele Cini) Topics and Methods in Condensed Mat

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics does the book cover?

What topics does the book cover?

Michele Cini

Topics and Methods in Condensed Matter Theory

Topics and Methods

With 73 Figures and 4 Tables

Prof. Michele Cini

Library of Congress Control Number: 2007925062

ISBN-13 978-3-540-70726-4 Springer Berlin Heidelberg New York

Printed on acid-free paper

To my wife Anna and to my son Massimo

Quod super est, vacuas auris animumque sagacem

Rome, May 7, 2007

Part I I-Introductory Many-Body Physics

Adiabatic Switching and Time-Ordered series . . . . . . . . . . . .

Atomic Shells and Multiplets . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.5.2 Hunds First Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Greens Functions as Thought Experiments . . . . . . . . . . . . . . .

Hopping Electron Models: an Appetizer . . . . . . . . . . . . . . . . . .

Many-body Eects in Electron Spectroscopies . . . . . . . . . . . .

Part II Symmetry in Quantum Physics

Group Representations for Physicists . . . . . . . . . . . . . . . . . . . .

Simpler Uses of Group Theory . . . . . . . . . . . . . . . . . . . . . . . . . . .

Product of Representations and Further Physical

Part III More on Green Function Techniques

11 Feynman Diagrams for Condensed Matter Physics . . . . . . .

11.4.1 External Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Many-Body Eects and Further Theory . . . . . . . . . . . . . . . . .

Part IV Non-Perturbative Approaches and Applications

15 Aspects of Nonlinear Optics and Many-Photon Eects . . .

Part V Selected Exact Results in Many-Body Problems

17 Pairing from repulsive interactions . . . . . . . . . . . . . . . . . . . . . . .

I-Introductory Many-Body Physics

1 Basic Many-Body Quantum Mechanics

1.1 Slater Determinants and Matrix Elements

For instance2 , (3,2,1)(2,1,3)=(2,3,1) . All permutations can be obtained from

1 Basic Many-Body Quantum Mechanics

upon multiplication restores the standard (ascending) order (1, 2, . . . N ). Any

converts any product into a normalized Slater determinant, so we may write

1.1 Slater Determinants and Matrix Elements

For example, if we pick f (i) = (x xi ), F = (x) is the number density

1 Basic Many-Body Quantum Mechanics

Fig. 1.1. Interaction vertices

[vk (1)ui (2)|f (1, 2)|uk (1)ui (2)

ui (1)vk (2)|f (1, 2)|uk (1)ui (2) ].

[ui (1)uj (2)|f (1, 2)|ui (1)uj (2)

uj (1)ui (2)|f (1, 2)|ui (1)uj (2) ].

1.1.2 Derivation of the Rules

 uP1 (1)uP2 (2) . . . uPN (N )|f |vQ1 (1)vQ2 (2) . . . vQN (N )

involves N !2 terms and is formidable unless N is small. However, the proof

1.2 Second Quantization

v1 (1) v2 (1) . . . vN (1)

1.2 Second Quantization

has a characteristic length x0 = m

with the commutation relation

If is a solution of (1.14) with eigenvalue E, a must be solution with

1 Basic Many-Body Quantum Mechanics

[vk (1)ui (2)|f (1, 2)|uk (1)ui (2)

ui (1)vk (2)|f (1, 2)|uk (1)ui (2) ].

[ui (1)uj (2)|f (1, 2)|ui (1)uj (2)

uj (1)ui (2)|f (1, 2)|ui (1)uj (2) ].

uP1 (1)uP2 (2) . . . uPN (N )|f |vQ1 (1)vQ2 (2) . . . vQN (N )

and the ground state = has energy E = 2th . In the interacting

|H| = 832 , thus we must look for the extrema of f (, ) = EN.

where Em are unperturbed eigenvalues and m|H |n are perturbation ma