Untitled

T h eo ry of
S u p e r c o n d u c t iv it y
Revised Printing
T h eo ry of
S u perco n d u ctiv ity
R e v is e d P r i n t i n g
J. R . ScHRIEFFER
Florida State University
Tallahassee, Florida
C R C Press
Taylor & Francis G ro u p
Boca Raton London New York
C R C Press is an im print of the

Taylor & Francis Group, an inform a business
ABP
A dvanced B o o k P rogram
The publisher is pleased to acknowledge the assistance of Paul Orban, who produced
the illustrations for the original edition. The revised edition (198.3) includes an
Appendix containing Nobel Lectures, December 11, 1972, by J. R. Schrieffer, Leon N.
Cooper, and John Bardeen, reprinted with permission of the Nobel Foundation.
Copyright © 1973 by the Nobel Foundation.
First published 1964 by Westview Press

Published 2018 by CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
,
CRC Press is an imprint of the Taylor & Francis Group an informa business
Copyright © 1999, 1964 by J. R. Schrieffer

No claim to original U.S. Government works
This book contains information obtained from authentic and highly regarded
sources. Reasonable efforts have been made to publish reliable data and
information, but the author and publisher cannot assume responsibility for
the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material
reproduced in this publication and apologize to copyright holders if
permission to publish in this form has not been obtained. If any copyright
material has not been acknowledged please write and let us know so we may
rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be
reprinted, reproduced, transmitted, or utilized in any form by any electronic,
mechanical, or other means, now known or hereafter invented, including
photocopying, microfilming, and recording, or in any information storage or
retrieval system, without written permission from the publishers.
Trademark Notice: Product or corporate names may be trademarks or
registered trademarks, and are used only for identification and explanation
without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Library of Congress Catalog Card Number: 99^60035
ISBN 13: 978-0-7382-0120-7 (pbk)
Cover design by Suzanne Heiser
CONTENTS
Editor’s Foreword vii
Preface xv
Preface to the Revised Printing xvi i
I
Ch a pter in t r o d u c t io n 1
1-1 Simple Experimental Facts 4
1-2 Phenomenological Theories 9
Ch a pter 2 t h e p a ir in g t h e o r y o f s u p e r c o n d u c t iv it y 24
2-1 Physical Nature of the Superconducting State 24
2-2 The One-Pair Problem 28
2-3 Landau’s Theory of a Fermi Liquid 34
2-4 The Pairing Approximation 36
2-5 Quasi-Particle Excitations 44
2-6 Linearized Equations of Motion 49
2-7 Concluding Remarks 57
ix
x C o n te n ts
Ch a pter 3 a p p l ic a t io n s o f t h e p a ir in g th eo ry 61
3-1 Justification of the Pairing Hypothesis 61
3-2 Acoustic Attenuation Rate 62
3-3 Nuclear-Spin Relaxation Rate 69
3-4 Electromagnetic Absorption 72
3-5 Physical Origin of the Coherence Factors 74
3-6 Electron Tunneling 78
3-7 Other Applications of the Pairing Theory 87
Ch a pter 4 -
electro n io n s y s t e m 89
4-1 The Electron-ion Hamiltonian 89
4-2 Bare Phonons 92
4-3 Bare Electrons 95
4-4 Bare Electron-Phonon Interaction 98
4-5 The Electron-Phonon Hamiltonian 102
Ch a pter 5 f ie l d -t h e o r e t ic m e t h o d s in t h e
M A N Y-BO D Y PRO BLEM 103

5-1 The Schrodinger, Heisenberg, and
Interaction Pictures 103
5-2 The Green’s Function Approach 105
5-3 The Free Fermi Gas 108
5-4 Spectral Representation of (?(p, r) 112
5-5 Analytic Properties of G 115
5-6 Physical Interpretation of 6r(p, p 0) 116
5-7 Interpretation of A(p, co) 119
5-8 The One-Phonon Green’s Function 124
5-9 Perturbation Series 126
6
Ch a p t er elem en ta r y e x c it a t io n s in norm al m eta ls 137
6-1 The Electron Gas with Coulomb Interactions 137
6-2 The Coupled Electron-Phonon System 148
C o n te n ts xi
Ch a pter 7 f ie l d -t h e o r e t ic m eth o d s a p p l ie d to
SUPERCO N D UCTIVITY 164

7-1 Instability of the Normal Phase 164
7-2 Nambu-Gor’kov Formalism 169
7-3 Zero-Temperature Excitation Spectrum 180
7-4 Extension to Finite Temperature 193
Ch apter 8 e l e c t r o m a g n e t ic p r o p e r t ie s of
SUPERCONDUCTORS 203
8-1 London Rigidity 203
8-2 Weak-Field Response 206
8-3 The Meissner-Ochsenfeld Effect 212
8-4 Electromagnetic Properties for Finite q and w 220
8-5 Gauge Invariance 224
8-6 The Vertex Function and Collective Modes 233
8-7 Flux Quantization 240
8-8 The Knight Shift 244
8-9 The Ginsburg-Landau-Gor’kov Theory 248
C o n c l u s io n 254
A p p e n d ix second q u a n t iz a t io n f o r m a l is m 257
A-l Occupation-Number Representation 257
A-2 Second Quantization for Bosons 259
A-3 Second Quantization for Fermions 265
A N
p p e n d ix L
o bel , 1972
ectu res 267
Macroscopic Quantum Phenomena from Pairing
in Superconductors
J. R. Schrieffer 267
Microscopic Quantum Interference Effects in the
Theory of Superconductivity
Leon N. Cooper 279
Electron-Phonon Interactions and Superconductivity
John Bardeen 300
xii C o n te n ts
N o tes and R eferen ces 317
In d ex 329
t \ b / A 4 * c e J - Q o s o k Q , Im 4 * c 4
Anderson: Basic Notions of Condensed M atter Physics, ABC ppbk,

ISBN 0-201-32830-5
Atiyah: K-Theory, ABC ppbk, ISBN 0-201-40792-2
Bethe: Interm ediate Q uantum Mechanics, ABC ppbk, ISBN 0-201-32831-3
Clemmow: Electrodynamics of Particles and Plasmas, ABC ppbk,
ISBN 0-20147986-9
Davidson: Physics of Nonneutral Plasmas, ABC ppbk
ISBN 0-201-57830-1
DeGennes: Superconductivity of Metals and Alloys, ABC ppbk,
ISBN 0-7382-0101-4
d ’Espagnat: Conceptual Foundations Q uantum Mechanics, ABC ppbk,
ISBN 0-7382-0104-9
Feynman: Photon-Eladron Interactions, ABC ppbk, ISBN 0-201-36074-8
Feynman: Q uantum Electrodynamics, ABC ppbk, ISBN 0-201-36075-4
Feynman: Statistical Mechanics, ABC ppbk, ISBN 0-201-36076-4
Feynman: Theory of Fundamental Processes, ABC ppbk, ISBN 0-201-36077-2
Forster: Hydrodynamic Fluctuations, Broken Symmetry, and Correlation Functions,
ABC ppbk, ISBN 0-201-41049-4
G ell-M ann/N e’eman: The Eightfold Way, ABC ppbk, ISBN 0-7382-0299-1
Gottfried: Q uantum Mechanics, ABC ppbk, ISBN 0-201-40633-0
Kadanoff/Baym: Q uantum Statistical Mechanics, ABC ppbk, ISBN 0-201-41046-X
Khalatnikov: An Intro to the Theory of Superfluidity, ABC ppbk,
ISBN 0-7382-0300-9
Ma: M odern Theory of Critical Phenomena, ABC ppbk, ISBN 0-7382-0301-7
Migdal: Qualitative Methods in Q uantum Theory, ABC ppbk, ISBN 0-7382-0302-5
N egele/O rland: Q uantum Many-Particle Systems, ABC ppbk, ISBN 0-7382-0052-2
Nozieres/Pines: Theory of Q uantum Liquids, ABC ppbk, ISBN 0-7382-0229-0
Nozieres: Theory of Interacting Fermi Systems, ABC ppbk, ISBN 0-201-32824-0
Parisi: Statistical Field Theory, ABC ppbk, ISBN 0-7382-0051-4
Pines: Elementary Excitations in Solids, ABC ppbk, ISBN 0-7382-0115-4
Pines: The Many-Body Problem, ABC ppbk, ISBN 0-201-32834-8
Quigg: Gauge Theories of the Strong, Weak, and Electromagnetic Interactions,
ABC ppbk, ISBN 0-201-32832-1
Richardson: Experimental Techniques in Condensed M atter Physics at Low
Temperatures, ABC ppbk ISBN 0-201-36078-0
Rohrlich: Classical Charges Particles, ABC ppbk ISBN 0-201-48300-9
Schrieffer: Theory of Superconductivity, ABC ppbk ISBN 0-7382-0120-0
Schwinger: Particles, Sources, and Fields Vol. 1, ABC ppbk
ISBN 0-7382-0053-0
ISBN 0-7382-0054-9
ISBN 0-7382-0055-7
Schwinger: Quantum Kinematics and Dynamics, ABC ppbk, ISBN 0-7382-0303-3
Thom: Structural Stability and Morphogenesis, ABC ppbk, ISBN 0-201-40685-3
Wyld: Mathematical Methods for Physics, ABC ppbk, ISBN 0-7382-0125-1
PREFACE
The material presented here is an outgrowth of a series of lectures

I gave at the University of Pennsylvania during the fall of 1962.
I have stressed the fundamentals of the microscopic theory of
superconductivity rather than attempting to give a broad survey
of the field as a whole. As a result, a number of highly interesting
and important areas are not discussed here; an example is the
application of the microscopic theory to type II (or “ hard”)
superconductors. The material presented here is primarily in
tended to serve as a background for reading the literature in which
detailed applications of the microscopic theory are made to specific
problems.
A variety of formal techniques have been used in the litera
ture to describe the pairing correlations basic to superconductivity.
For this reason I have developed a number of these techniques in
the text and it is hoped that the inelegance of this approach will
be justified by the usefulness of the material.
A brief review of the simple experimental facts and several
phenomenological theories of superconductivity is given in the first
chapter. This is followed in Chapter 2 by an account of the
original pairing theory advanced by Bardeen, Cooper, and the
author. A number of applications of this theory are worked out
xv
xvi Preface
in Chapter 3. This first portion of the book uses only the tech
niques of quantum mechanics which are covered in a standard
graduate course on quantum theory. While the notation of
second quantization is used as a convenient shorthand, this
formalism is reviewed in the appendix.
In Chapters 4 and 5 the many-body aspects of the coupled
electron-ion system are developed with a view to treating in a
more realistic manner the effective interaction between electrons
which brings about superconductivity. In addition, the basis for
treating strong quasi-particle damping effects important in strong
coupling superconductors is developed. In Chapter 6 a discussion
of elementary excitations in normal metals is given, which lays
the ground work for the field-theoretic treatm ent of the super
conducting state given in Chapter 7. There, the noninstantaneous
nature of the interaction bringing about superconductivity is
treated as well as the breakdown of the quasi-particle approxima
tion and the resolution of this difficulty. In the final chapter the
electromagnetic properties of superconductors are treated, as well
as the collective excitations of the system.
I should like to thank Drs. P. W. Anderson, J. Bardeen,
L. P. Kadanoff, D. J. Scalapino, Y. Wada, and J. W. Wilkins for
many helpful discussions during the preparation of this manuscript.
I am also indebted to Drs. F. Bassani and J. E. Robinson, who
prepared a set of notes covering a lecture series I gave at Argonne
National Laboratory during the spring of 1961. Much of the
material in Chapters 4 and 5 and in the appendix is related to their
notes. In addition, I would like to express my sincere apprecia
tion to Mrs. Dorothea Hofford for the speed and accuracy with
which she typed the manuscript. Finally, I should like to thank
my wife for her considerable help in preparing this book.
J . R . Sc h r ie f f e r
Philadelphia , Pennsylvania
J u l y 1964
PREFACE TO THE
REVISED PRINTING
Since the first appearance of this book in 1964, the field of

superconductivity has undergone dramatic growth in scope and
level of activity. The two-volume treatise, Superconductivity,
edited by R. D. Parks, gave an account of the field at the close of
the 60s. Collections of papers describing current research on
superconductivity can be found in proceedings of conferences,
such as the International Conference on Low Temperature Phy
sics and the Applied Superconductivity Conference.
Since 1964 many significant areas of research have devel
oped including type II superconductivity and the Abrikosov vor
tex lattice. This fundamental understanding led to high field
magnets having im portant technological applications. Josephson
tunnel junctions proved to exhibit a wide variety of interesting
phenomena which have led to a spectrum of devices of use in high
precision measurements and in computers. The long sought after
organic superconductor has been discovered, and major advances
in fabricating new superconducting materials have been made.
3He was discovered to be a spin triplet superconductor, with
many remarkable properties.
x v iii P r e fa c e to th e R e v is e d P rin tin g
On the theoretical side, there was early recognition that the

pairing correlations of metallic superconductors also played a
fundamental role in the structure of atomic nuclei. The theory of
superconductivity also forms the basis for understanding the
structure of neutron stars, despite their enormously high temper
ature. Finally, the broken symmetry concept inherent in the pair
ing theory has been helpful in setting up gauge theories of ele
m entary particles. Possibly other developments stemming from
the pairing theory lie ahead.
This revised printing does not attem pt to give an account of
the above topics since this would be impossible in a single vol
ume. Rather, we note that the fundamentals of the theory of
superconductivity as discussed have remained unchanged over
the past two decades, and the recent developments have been
built on that theoretical foundation. Hopefully, the reader will
find the text of continuing value as an introduction to this fasci
nating and active field.
J. R
o bertS c h r ie f f e r
Santa Barbara, 1983

T h eo ry of
S u p e r c o n d u c t iv it y
Revised Printing
CHAPTER 1
INTRODUCTION
The phenomenon of superconductivity is a remarkable ex

ample of quantum effects operating on a truly macroscopic
scale.1 In a superconducting material, a finite fraction of the
electrons are in a real sense condensed into a “ macromolecule” (or
“ superfluid” ) which extends over the entire volume of the system
and is capable of motion as a whole. At zero temperature the
condensation is complete and all the electrons participate in form
ing this superfluid, although only those electrons near the Fermi
surface have their motion appreciably affected by the conden
sation. As the temperature is increased, a fraction of the electrons
evaporate from the condensate and form a weakly interacting
gas of excitations (or “ normal fluid”), which also extends through
out the entire volume of the system, interpenetrating the super
fluid.2 As the temperature approaches a critical value T c, the
fraction of electrons remaining in the superfluid tends to zero and
the system undergoes a second-order phase transition from the
superconducting to the normal state. This two-fluid picture of a
superconductor is formally analogous to that which characterizes
superfluid He4, although there are important differences between
these systems.1*3
2 T h e o ry of Sup ercon d u ctivity
The amazing properties of superconductors (e.g., perfect

diamagnetism, zero d-c electrical resistance, etc.4) are related to
the peculiar excitation spectrum of the superfluid. As we shall
see, the superfluid can carry out potential (or irrotational) flow
with little change of its “ internal energy” (i.e., energy associated
with forces binding the superfluid together). On the other hand,
the superfluid cannot support rotational flow. In analogy with
superfluid He4, if one tries to force the superfluid into motion
having vorticity (i.e., a nonvanishing curl of its linear momentum),
a fraction of the superfluid is necessarily converted into normal
fluid. Since the normal fluid does not take advantage of the
forces binding the superfluid together, there is in general a large
increase in energy associated with creating this vorticity. It is
reasonable, therefore, that the superfluid possesses a rigidity or
stiffness with respect to perturbations which, like the magnetic
field, tends to impart vorticity (i.e., angular momentum) to the
system. On the basis of this assumed rigidity, London1*5 was
able to account theoretically for the perfect diamagnetism of
bulk superconductors in weak magnetic fields (the Meissner
effect6) and for the apparent lack of d-c electrical resistance, as
first observed by Kamerlingh Onnes in 1911.7
As we shall see, the microscopic theory of superconductivity
proposed by Bardeen, Cooper, and the author8 can be thought of
in terms of this sort of two-fluid picture.9 In the lowest approxi
mation the superfluid is formed from pairs of electrons which are
bound together by lattice polarization forces. The pairs greatly
overlap with each other in space, and it is the strong p air -p a ir
correlations in addition to correlations between mates of a pair
which are ultimately responsible for the rigidity of the super
fluid wave function discussed above. More generally, these cor
relations are responsible for an energy gap in the elementary
excitation spectrum of a superconductor from which many prop
erties of the superconductor (in addition to its electromagnetic
behavior) follow as a consequence. In the theory, the normal
fluid is composed of the gas of elementary excitations of the
system.
In trod u ction 3
It is perhaps not surprising that the microscopic theory of

superconductivity followed Onnes’ remarkable discovery of this
phenomenon by almost fifty years, considering the physical and
mathematical complications of the problem. It was not until
1950 that the basic forces responsible for the condensation were
recognized, through the insight of Frohlich.10 He suggested that
an effective interaction between electrons arising from their inter
action with crystal lattice vibrations (phonons) was of primary
importance in bringing about the condensation. At this time,
independent experiments on the isotope effect in superconductors
were being carried out by Reynolds et al.11 and by Maxwell12
which gave experimental support to Frohlich’s point of view.
Early theories of Frohlich10 and Bardeen13 based on a perturbation
treatm ent of the electron-phonon interaction ran into mathe
matical difficulties. The significance of these difficulties was em
phasized by Schafroth’s14 proof that the Meissner effect cannot be
obtained in any finite order of perturbation theory, beginning
with the uncoupled system. Later, Migdal15 showed that there
is no energy gap in the electronic excitation spectrum within the
perturbation theory. In the BCS theory, the electron-phonon
coupling constant g enters in the nonanalytic fashion e ~llg2, in
agreement with Schafroth’s and Migdal’s results.
The microscopic theory explains essentially all of the general
features of superconductivity. In addition to this qualitative
explanation, it is in remarkably good quantitative agreement
with experiment considering the crudeness of the approximations
necessitated by our uncertainties regarding electronic and phononic
band structure, electron-phonon matrix elements, etc., in real
metals.
In this book we shall attem pt to give an account of the under
lying physical ideas of the theory. While some of the discussion is
couched in the language of the many-body problem, much of this
formalism is developed in the text. In general, we shall not give
a detailed discussion of the relation between theory and experi
ment and the reader is referred to a number of books and review
articles9,16 covering this area. We list below a few of the most
4 T h e o ry o f Sup e rcon d u ctiv ity
important simple experimental facts about superconductors. One

conventionally distinguishes between the behavior of type I (or
soft) superconductors and type II (or hard) superconductors.
M SIM PLE E X P E R IM E N T A L FA CT S
Electromagnetic Properties
The d-c electrical resistivity of materials in the soft super
conducting state is zero. This fact is established to better than
one part in 1015 of the resistance of the normal state at the corre
sponding temperature.161 At T = 0 the resistivity of a super
conductor ideally remains zero up to a critical frequency
Hwg ~ 3.5JcBT c (presumably the threshold for creating excitations
out of the condensate). In practice, the edge of the gap is
smeared and a precursor electromagnetic absorption is observed
below the edge of the gap in certain cases. At finite tempera
tures, there is a finite a-c resistivity for all a> > 0 (presumably
because of absorption by the thermally excited normal fluid if
cd < cDg). For cd » cDg, the resistivities of the normal and super
conducting states are essentially equal, independent of
temperature.
In 1933, Meissner and Oschenfeld6 discovered that a bulk
superconductor is a perfect diamagnet. Thus the magnetic field
B penetrates only to a depth A ~ 500 A and is excluded from the
main body of the material. If one (incorrectly) argues that the
vanishing zero-frequency electrical resistance implies that there can
be no electric field (of any frequency) in a superconductor,
Maxwell’s equation
V x E = - -c ^dt (1-1)
shows that the magnetic field present in the normal metal will be
“ frozen in ” when the metal becomes superconducting. This is
contrary to the Meissner effect, which states that the field is
expelled in the superconducting phase. The point is that the
superfluid gives rise to a purely inductive impedance which
vanishes only at zero frequency.9 It is this nonzero impedance
In tro d u ctio n 5
which permits the expulsion of B. This point is discussed further

below.
The perfect exclusion of the magnetic flux in bulk soft super
conductors increases the free energy per unit volume of the super
conductor by H^“j877,5 if H is the externally applied field. Since
there is only a finite amount of energy reduction due to the con
densation into the superconducting phase, there must be a critical
field H C(T) at which the total free energy of the superconducting
and normal states are equal. The critical field is a maximum H 0
at T = 0 and falls to zero at T — T c as shown in Figure 1-1. For
typical “ soft” superconductors like Al, Sn, In, Pb, etc., H 0 is of
order a few hundred gauss. In “ hard ” superconductors like Nb3Sn,
superconductivity can persist to an “ upper” critical field H c2 of
order 105 gauss presumably due to the magnetic flux penetrating
into the bulk of the material for H larger than a “ lower” critical
field H cl.17' 18 Thus, as opposed to a soft superconductor, a perfect
Meissner effect does not exist above H cl in a hard superconductor.
F IG U R E l-l The deviations of the critical field from T u /n ’s law

# C( T ) = / / 0[l — ( T / T c)2], i.e., the prediction of the G o rte r-C asim ir
model.
If one has a multiply connected superconductor, e.g., a

hollow cylinder, the flux passing through the hole cannot have
an arbitrary value, but is quantized to multiples of hc/2e ~
4 x 10~7 gauss cm2. Quantization of flux in units twice this
size was predicted by London1 while the experimental observation
of the effect and the establishment of the correct unit of flux
was carried out independently by Deaver and Fairbank20a and by
Doll and Nabauer.20b
Thermodynamic Properties
In zero magnetic field, there is a second-order phase transition
at T c.21 The jump in specific heat is generally about three times
the electronic specific heat y T c in the normal state just above the
transition. In well-annealed pure specimens the width of the
transition can be as small as 10” 4 °K although this is not believed
to be the intrinsic width of the transition.22 As T /T c -> 0, the
electronic specific heat generally falls as ae~ blT, presumably due
to the energy gap for creating elementary excitations. The ratio
of the energy gap 2A(0) at T = 0 to k BT c is usually of order 3.5,
the ratio being larger for stronger coupling superconductors like
Pb and Hg. A plot of the specific heat of Sn is shown in Figure
1-2. For T ^ T cl 2, the curve is reasonably well fitted by a T 3.
10
Ces 1\
yTc tin
10"1
- N .
10"2
T JT
1 1 .J---------1---------
F IG U R E 1-2 The electronic specific heat of Sn.

In tro d u ctio n 7
In the presence of a magnetic field the A-#-phase transition

for a bulk specimen is first order, i.e., a latent heat is involved.4
Isotope Effect
As we discussed above, the isotope effect shows that lattice
vibrations play an essential role in bringing about superconduc
tivity. In particular, one finds that the critical field at zero
temperature H 0 and the transition temperature T c vary as
Tc ~ ^ ~ H0 (a ~ |) (1-2)
when the isotopic mass M of the material is varied. Thus, T c and
H0 are larger for lighter isotopes. If lattice vibrations were not
important in the phenomenon there is no reason why T c should
change as neutrons are added to the nuclei since their main effect
is to change the mass of the ions. While the value a = 0.45 to
0.50 is approximately correct for many superconductors there are
a number of notable exceptions, for example Ru, Mo, Nb3Sn, and
Os23 which have small or vanishing isotope effects. As Garland24
has shown, this does not preclude the phonons from causing the
transition. Although the actual mechanism in these materials
is not firmly established at present, it is not unlikely that the
electron-phonon interaction is the appropriate mechanism even
in these exceptional cases.
Energy Gap
There are several direct ways of observing the energy gap in
the elementary excitation spectrum of superconductors.16dj e As
we mentioned above, the threshold for absorbing electromagnetic
radiation gives a value for the energy gap.25 An even simpler
method26 (due to Giaever) is to measure the electron tunneling
current between two films of a superconducting material separated
by a thin (~ 20 A) oxide layer. As T -> 0, no current flows until
the applied voltage (times the electronic charge) V exceeds the
energy gap 2A. As the temperature is increased, a finite current
flows for V < 2A(T); however, a break in the curve persists for
8 T h e o ry o f Su p ercon d u ctivity
F IG U R E 1-3 The temperature-dependent energy gap of Al as deter

mined by electron tunneling.
V = 2A(T). The temperature-dependent energy gap observed

in this way is plotted in Figure 1-3. The temperature dependence
of the energy gap can also be simply determined from the rate of
attenuation of sound waves,27 the rate of decay of nuclear
magnetization,28, 29 and the impurity-limited electronic thermal
conductivity.30 All of these methods give essentially the same
results.
Coherence Effects
If one attempts to account for the rate of electromagnetic and
acoustic absorption as well as the rate of nuclear spin relaxation
in superconductors on the basis of a simple two-fluid energy-gap
model, one quickly discovers inconsistencies. Experimentally the
rate of acoustic absorption decreases monotonically as T decreases
below Tc,27 while the nuclear spin relaxation rate initially rises,
passing through a peak before dropping to zero at low tem
perature.28 If one takes the same matrix elements as in the
normal state for the coupling of the excitations with phonons
In tro d u ctio n 9
and with the nuclear spins, the two processes should have identical
temperature dependences. Therefore, at least some of these
matrix elements differ from those in the normal metal. As we
shall see, the matrix elements appropriate to the superconducting
state are linear combinations of those in the normal state.8 Since
the coefficients in the linear combination depend upon the nature
of the coupling (scalar, vector, spin), the square of the matrix
elements in the superconducting state differ for coupling the
excitations to acoustic (scalar) and electromagnetic (vector) or
nuclear magnetization (spin) variables.
1-2 P H E N O M E N O L O G IC A L T H E O R IE S
Gorter-Casimir Model
In 1934, Gorter and Casimir2 advanced a two-fluid model
along the lines which we discussed above. If x represents the
fraction of electrons which are in the “ normal” fluid and (1 — x)
the fraction condensed into the superfluid, they assumed the free
energy for the electrons is of the form
F(x, T) = x * f n(T) + (1 - x)f,(T ) (1-3)
where f n and f s were chosen to be

fn(T) = - & T 2 (1-4)
and
f s(T) = - 0 = const. (1-5)
In the normal metal the electronic free energy is just (1-4) so that
the free energy of the S- and A-phases agree when (1 — x) —* 0,
i.e., at T c. The energy —£ represents the condensation energy
associated with the superfluid. By minimizing F(x, T) with
respect to x for fixed T, one finds that the fraction x of “ normal”
electrons at a temperature T is given by
10 T h e o ry o f S u p ercon d u ctivity
From the thermodynamic relation

= F n(T) - F S(T) (1-7)
one finds from (l-3)-(l-6) the expression
H C(T) = H o 1 ( 1-8)
for the temperature-dependent critical field. Thus, H c is pre

dicted to be a parabolic function of ( T / T c), in rough agreement
with experiment. In addition, the free energy gives the electronic
specific heat in the S -phase as
(1-9)
so that the relative jump in the electronic specific heat at T c is 3,
again in general agreement with experiment. This agreement is
not completely surprising since the theory was constructed in
what appears to be a rather artificial manner in order to obtain
agreement with experiment. In particular, one would expect the
exponent r to be unity rather than one-half if x represents the
fraction of electrons which are normal. Furthermore, the con
densation energy /3 would be expected to increase as more particles
condense into the ordered phase. Nonetheless, the Gorter-
Casimir theory leads to nontrivial predictions which, when com
bined with the London theory, are in reasonably good agreement
with experiment. Unfortunately, there is little detailed connec
tion between the expression (1-3) and that given by the
microscopic theory.
The London Theory

In the year following Gorter and Casimir’s work, F. and H.
London advanced a phenomenological theory of the electro
magnetic behavior of superconductors.1,31 Their scheme is
based on a two-fluid type concept with superfluid and normal
fluid densities ns and nn plus the associated velocities vs and vn.
Owing to local charge neutrality, the densities are restricted by
In tro d u c tio n 11
ns + nn where n is the average number of electrons per unit

= n,
volume. The super and normal fluid current densities are
postulated to satisfy
§(jit = ^771- 2 E (Js = -er*svs) (l-10a)
J n = <jnE (Jn = — ennv n) (l-10b)
The first of these equations is nothing more than F = ma applied
to a set of free particles of charge —e and density n s. Apparently
the superfluid is unaffected by the usual scattering mechanisms
which produce the finite conductivity a n associated with the normal
fluid.
The second (and most famous) equation of the London theory is
V x Js = — me
—B (1-11)
71 P ^
This latter equation leads to the Meissner effect. One can see this
by considering the curl of one of Maxwell’s equations:
VxVxB = yVxJj (1-12)
where we have neglected the displacement current and the normal
fluid current J n since we are interested in the static Meissner
effect. On combining (1-11) and (1-12) one has
V2B = ! ^ B = A ?B
where London’s penetration depth AL is defined by
/ me2 \,3 1/2
(1-14)
\4:7Tnse2j
If (1-13) is applied to a plane boundary located at x = 0, the
magnetic field (parallel to the surface) decreases into the super
conductor according to
B(x) = B(0)e~XIAL (1-15)
Therefore the magnetic field vanishes in the bulk of the material
and one obtains perfect diamagnetism, as required. For solu

tions of London’s equations for other geometries see London’s
book.1
To understand the relation between London’s two equations,
we notice that the curl of (l-10a) is the time derivative of (1-11).
Therefore, outside of a constant of integration, the Meissner effect
follows from the “ perfect” conductivity of the superfluid, i.e.,
(l-10a). By postulating (1-11), the Londons added the all-
important restriction that B = 0 inside the superconductor
regardless of its history, which is the essence of the Meissner effect.
If one combines the result (1-6) of the Gorter-Casimir model
(1-16)
for the temperature dependence of the superfluid density, with
London’s expression (1-14) for the penetration depth, one finds
A(T) [1 - (T/T,)4]1' 2
(1-17)
Thus, for T — T c, A = oo so that no flux is excluded at T c.
As T drops infinitesimally below T cy A decreases rapidly, thereby
establishing the Meissner effect in bulk specimens for all T < T c.
This temperature dependence is surprisingly close to that observed
experimentally although the results of the microscopic theory are
in somewhat better agreement with experiment than is (1-17).
The fact that the supercurrents are uniquely determined by
the magnetic-field configuration (according to the Meissner effect)
guarantees that one can apply reversible thermodynamics to
quasi-static processes in superconductors, an important fact.4
If we introduce the vector potential A, the second London
equation (1-11) can be written as
V x Js = Vx A (1-18)
As London pointed out, this equation can be satisfied by taking
In trod u ction 13
if A is properly gauged. For the supercurrent to be conserved

we must require
V* A = 0 (1-20)
We are, however, still free to add to A the gradient of any function

X which satisfies Laplace’s equation V 2X = 0. For an isolated
simply connected body, the normal component J sl must vanish
on the surface. Therefore, A ± must also vanish on the surface.
This condition determines (V X )± over the entire surface and
therefore determines X to within an additive constant (which of
course cannot contribute to A or Js). For a massive body these
conditions ensure that A = 0 in the bulk of the material. If
current is flowing through the boundaries (e.g., when a super
conductor is an element in a circuit), the current on the boundary
uniquely specifies A. Therefore, while (1-19) does not appear to
be gauge-invariant, the theory is in fact gauge-invariant since
one is told to throw away any part of A which does not satisfy
the London gauge conditions. In this way, physical predictions
are independent of the choice of gauge.
In a multiply connected body, the restrictions
V 2X ' = 0 = 0 ( 1- 21 )
surface
no longer require that an added gauge potential VX' be zero.
Therefore A is not uniquely determined by the boundary condition
A l = 0, in this case. If we form the line integral of A around a
loop surrounding a hole in the multiply connected body, Stokes’s
theorem gives
( 1-22 )
for the flux through the loop. If the path is taken within the
interior of the superconductor where B = 0, then V x A = 0 and
we may write A as the gradient of a scalar
A = VX (1-23)
14 T h e o ry of Su p e rco n d u ctiv ity
While Y X must be single valued, X will not be single valued in

general, since
<j> A • dl = <j> V X • dl = AX = 0 (1-24)
where J X is the change of X as one travels once around the hole.
By specifying the flux 0 through each hole, one can uniquely
determine A, as required.
F. London’s Justification of the London Theory

F. London5 pointed out that the equation (1-19)
JS = - ^meA
71 P ^
(1-25)'
could be deduced from first principles if one assumed that the
many-body wave function Ws describing the superfluid is rigid
with respect to perturbations due to a transverse vector potential
(V • A = 0). One can see this as follows. The current density
Js0 in the absence of A
Jso(r) = J - W * ) S(r, - r) d ^ - ■-d*rni
ph n$ C
- m 2
(1-26)
clearly vanishes. If a weak magnetic field is applied to the
system and Ws is unaffected to first order by this perturbation,
the paramagnetic current (1-26) continues to vanish, while the
diamagnetic current is given by
= - j=i i mc A<r) jf - r>d*r' ■• d3r»s

= mc A(r)
71 P ^
(1-27)
in agreement with (1-25). More accurately, it is assumed that
in the long wavelength limit, the paramagnetic and diamagnetic
currents of the normal fluid exactly cancel each other (as they do
in the Landau diamagnetism of the normal state) while the para
In trod u ction 15
magnetic current of the superfluid vanishes, leaving the dia

magnetic supercurrent. We have suggested above that the origin
of the London “ rigidity” is the energy gap in the excitation spec
trum of the system. This somewhat imprecise statement is not
in conflict with the fact that insulators also possess an energy gap
in their excitation spectra. This follows since interband matrix
elements of the magnetic perturbation are large in this case so
that the paramagnetic current is nonzero and just cancels the
diamagnetic current (see Chapter 8).
As we shall see below, the microscopic theory reduces exactly
to the form (1-27) in the limit of fields which vary slowly in space.
On the basis of London’s quantum interpretation of the Lon
don equations,1 he concluded that the flux 0 trapped through a
hole of a multiply connected superconducting body must be an
integral multiple of hcje ~ 4 x 10“7 gauss cm2. To understand
this result, consider two concentric superconducting cylinders, as
shown in Figure 1-4. Suppose that the thickness of the cylinders
is large compared to the penetration depth A and that a flux 0
is trapped within the hole of the inner cylinder. Furthermore,
assume that there is no magnetic field in the region between the
F IG U R E 1-4 Tw o concentric superconducting cylinders with a flux 0

trapped within the smaller cylinder.
16 T h e o ry of S u p ercon d u ctivity
inner and outer cylinders, so that the flux through the hole of the
outer cylinder is also equal to 0. The inner cylinder acts only
as a shield to ensure that no magnetic field touches the physically
interesting outer cylinder. Let be the wave function for the
outer cylinder when there is no flux trapped, 0 = 0. To deter
mine the wave function xf/0 in the case 0 / 0, we note that the
vector potential in the outer ring is in the 6 direction and has the
value
>1 / x 0 1 0 /0 9 \ _ / 0 0\
( 1 -2 8 )
Since A in the outer cylinder is the gradient of the scalar (00/277-),

it follows that WQand W# are related by the gauge transformation
= e~ie<p f 9JlhcXF 0 (1-29)
where 0; is the azimuthal coordinate of the ^’th electron. If
W# and W0 are to be single-valued functions of the coordinates
dj, one must have
—
60
he
= integer (1-30)
or 0 is quantized to the London values
tf>n = (n = 0, ±1, ± 2 ,...) (1-31)
To complete the argument, suppose that the inner cylinder is
made normal so that the magnetic field fills the entire hole in the
outer cylinder. Owing to the Meissner effect, the magnetic field
will penetrate only a small distance (~ 5 .10“6 cm) into the outer
cylinder. Therefore the above argument should continue to hold
since this small perturbation should not affect the wave function
W appreciably (particularly if London’s “ rigidity” is effective).
On the basis of this argument London concluded that the flux
trapped through any hole in a massive specimen is quantized to
multiples of hcje.
In 1953 Onsager32 suggested that the actual value of the
flux quantum might be one-half this value, presumably because
In tro d u ctio n 17
of the effective charge of the entities making up the superfluid

being 2e. In a beautiful set of experiments carried out by Deaver
and Fairbank20a and independently by Doll and Nabauer,20b
Onsager’s suggestion was verified. In essence the difficulty in
London’s argument is that there is another series of low-lying
states which are distinct from London’s state Wn and cannot be
generated from the ground state W0 by a gauge transformation.
This second series of states, first discussed by Byers and Yang,19
leads to the quantized flux values
* „ = (« + ^ 7 (n = 0, ±1, + 2 ,...) (1-32)
By taking the London series (1-31) and the Byers-Yang series
(1-32) together, one obtains the result suggested by Onsager
<?n = ri( t i ) (w = o, ± 1 , ± 2 , . . . ) (1-33)
in agreement with experiment. The fact that these are the only
allowed values for 0 follows from the BCS pairing theory since
other values of 0 lead to an extraordinarily high energy of the
electron system and are therefore unstable. The problem of flux
quantization is discussed further in Chapter 8.
Pippard’s Nonlocal Generalization of the London Theory
The basic equations (1-10) and (1-11) of the London theory
are “ local” in the sense that they relate the current densities and
the electromagnetic potentials at the same point in space. On
the basis of numerous experimental results, Pippard33 concluded
that these local relations must be replaced by nonlocal relations
giving the currents at a given point in space as a space average
of the field strengths taken over a region of extent £0 ~ 10" 4 cm
about the point in question. One of the most compelling argu
ments for this generalization is that the penetration depth A
increases appreciably if a sufficient amount of impurity is intro
duced into the material. This effect sets in when the mean free
path I of electrons in the normal state falls below a distance £0,
known as Pippard’s “ coherence” length. As we shall see, £0 is a
18 T h e o ry o f Su p e rcon d u ctivity
measure of the size of the pair bound state from which the super
fluid wave function is constructed. In the microscopic theory it
is related to the energy gap 2A by £0 = Hvf IttA, where vF is the
Fermi velocity. On the other hand, in the London theory A is
not expected to be appreciably affected by impurities, particularly
near T = 0, where all of the electrons are condensed. In choosing
a form for the nonlocal relations, Pippard was guided by Chamber’s
nonlocal expression,34 relating the current density and electric
field strength in the normal metal
R = r - r' (1-34)
where a is the long wavelength electrical conductivity. Chamber’s
expression is a solution of Boltzmann’s transport equation if the
scattering mechanism is characterized by a mean free path I.
For fields varying slowly over a mean free path Z, (1-34) reduces
to Ohm’s law J = oE. With Chamber’s expression in mind,
Pippard assumed that London’s equation
should be replaced by
(1-36)
The effective coherence length £ is given by
where a is an empirical constant of order unity and £0 is a length

characteristic of the material. For a pure material, Pippard’s
equation reduces to London’s equation if A(r) varies slowly over
a coherence length. For an impure material, Pippard’s equation
leads to an extra factor < 1 multiplying (1/cA) in London’s
equation in this long wavelength limit, thereby increasing the
effective penetration depth. In most cases distances of order
A « £ are of importance in penetration phenomena so that the full
reduction £/£0 is not effective. In highly impure specimens A is

of order or greater than f and one has A ~ (£0/O1/2*
That the effective coherence length £ should be bounded by
the mean free path I is certainly reasonable from a physical point
of view. It is a tribute to Pippard’s insight into the physics of
the problem that his equation is almost identical to that given
by the microscopic theory.8
A good deal of the qualitative aspects concerning the electro
magnetic properties of superconductors can be understood on the
basis of a simple energy-gap model. Prior to the BCS theory,
Bardeen160 gave a theoretical derivation of the nonlocal electro
dynamics. He assumed that the single-particle matrix elements
of the magnetic perturbation were unaltered by the condensation
and that the single-particle excitation spectrum was altered only
by adding a constant to the excitation energy, thereby creating
an energy gap. Subsequent to the work of BCS, Ferrell, Glover,
and Tinkham35 employed the Kramers-Kronig relation to give
quite a general discussion of how the electrodynamic behavior of
a superconductor comes about, because of its energy gap. For a
review of their arguments, the reader is referred to Tinkham’s
review article.161
Ginsburg-Landau Theory
In 1950 Ginsburg and Landau36 proposed an extension of the
London theory which takes into account the possibility of the
superfluid density ns varying in space. They phrased the theory
in terms of an effective wave function !P(r) which we normalize
such that the local density of condensed electrons is given by
|<F(r)|2 = (1.38)
Tl
where n is the total number of electrons per unit volume. Roughly

speaking, *F(r) corresponds to the center-of-mass wave function
of the BCS pairs. Ginsburg and Landau treated ^(r) as an order
parameter which is to be determined at each point in space by
20 T h e o ry o f Su p e rco n d u ctiv ity
minimizing the free-energy functional F(W , T) of the system.

The problem is then one of guessing an appropriate form for F .
Suppose that /('f', T) is the difference of free energy per unit
volume between the S- and iV-phases when W is uniform. Then
it is natural to include in F the term
Jf[W { r), T ] d 3r (1-39)

While f(*F, T) is not known a priori , Ginsburg and Landau
determined this function for small W (which is all that is needed
when T is near T c) by expanding / as a power series in I'f'l2 and
retaining the first two nonvanishing terms; thus
T) ~ a(T)\W\2 + \b{T)\V\* (1-40)
for \W\2 « 1. The equilibrium value |^ e|2 is determined by
minimizing /:
df
8 \W
= 0= a(T ) + b(T)\We\2 (1-41)
and therefore
l^ l2 = I1' 42)
From (1-40) and (1-42) one finds the (zero-field) free-energy
difference per unit volume between the S - and iV-phases is given
by
f s ( T) - h ( T ) = f ( T ) = - \ ^ - (1*43)
where we have used the thermodynamic relation between the

critical field and the N - S free-energy difference. If we use the
fact that in the London theory X2(T) ~ 1/ns(T), we obtain a
second relation between a(T ) and b(T):
From (1-43) and (1-44) we find

a(T) =
H C2(T) X2(T)
4tr A2(0) (1-45)
H C*(T) A*(T)
b(T) =
4 tt A4(0)
and therefore f(W , T) given by (1-40) can be expressed in terms of
experimentally measurable quantities.
If W{r) is not uniform in space, Ginsburg and Landau argue
that extra terms should be included in F which involve the rate
of change of W in space. Presumably these terms would come
from (a) the kinetic energy associated with extra wiggles in the
many-body wave function describing ns and/or vs changing in
space and (b) the interaction energy density being influenced by
the variations of the superfluid density in a region surrounding
the point in question. If | \P|2 varies slowly in space, it should be
sufficient to keep the leading term in |grad W]2. On the basis of
gauge invariance, one would expect that this term, when combined
with the effect of a vector potential A(r), would lead to a free-
energy contribution of the form
f n* 5 Vf'(r) + — A(r)!F(r) d3r (1-46)
J 2m* I c
where e* is the effective charge of the “ entities” forming the
superfluid. (As we shall see, 2n* = n, e* = 2e, and m* = 2m,
consistent with the pairing theory.)
By minimizing the total free-energy difference
w T) - V’F(r) + - A(r)<P(r) d 3r
+ d3r +
J 8tt
d 3r (1-47)
with respect to W(r), one finds the constitutive equation of the
Ginsburg-Landau theory
3{T) A2(T) A2(T)

+ H4nn* A2(0) A2(0) ^(r) = 0 (1-48)
[ -
The current density is given by
,s(r) = — b ^ ~ e*2A(r) _ 2 ^ r {^ (r)V^ {r) - y (r)v y *w}

(1-49)
with our normalization of W. As in the London theory one is to
use the gauge V • A = 0. Therefore, (1-48) and (1-49) together
with Maxwell’s equation V x V x A = 477-J/c lead to two non
linear differential equations which determine the functions W(r)
and A(r).
We note that if A = 0 and W is uniform in space, (1-48)
reduces to the condition
1- O T . o
which states that W is equal to its equilibrium value (1-44), as

required. If W is perturbed slightly from its equilibrium value
at some point, say r = 0, then the linearized Ginsburg-Landau
equation
w # O M * m p _ 0 n . 51)
2m* ( ) 27m * A2(0) ( '
for the deviation W(r) leads to

~ ----- (1-52)
p -rld
W
r
Thus the perturbation dies away exponentially, with the charac

teristic length
Trn*h2 1,2 M &— _ {1.53)
m * H 2( T ) A ( T) [1 - T/TJ
where the last estimate uses the microscopic theory to relate

H 0 and £0. We see that even though the relation between J s
and A is approximated by a local expression, the Ginsburg-
Landau theory definitely includes nonlocal effects and the co
herence length appears in a natural way.
In tro d u ctio n 23
Gor’kov37 has given a derivation of the Ginsburg-Landau

theory starting from the microscopic theory. He finds that the
GL wave function W is proportional to the local value of the energy -
gap parameter J . His derivation is outlined in Chapter 8.
The GL theory is particularly useful in calculations where one
cannot treat the magnetic field by perturbation theory. Typical
examples of such situations include thin films in strong magnetic
fields, N - S phase boundaries, the intermediate state, etc. One
can give a simple derivation of flux quantization on the basis
of the current equation (1-49), and one finds the flux quantum
to be hc/e*. The experimentally observed value hc\2e leads to
the value e* = 2e, as mentioned above. The GL theory has
recently played an important role in explaining the magnetic
behavior of so-called “ hard” superconductors, which are particu
larly interesting materials, due to their high critical fields (~ 105
gauss). The fundamental theoretical work in this area is due to
Abrikosov,17 who established the vortex picture to account for
this new magnetic behavior. Each vortex carries one quantum
of flux.
Unfortunately, the original Ginsburg-Landau theory is re
stricted to the temperature range (T c — T ) / T c « 1, although it
has recently been extended to all temperatures under suitable
conditions by Werthamer and by Tewordt.38
CHAPTER 2
THE PAIRING TH EO RY OF
SUPERCONDUCTIVITY
In analogy with a free electron gas, normal (N) metals exhibit a

single-particle excitation spectrum which, in the limit of a large
system, is a continuum starting at zero energy. The degeneracy
associated with this spectrum leads to the linear temperature
dependence of the electronic specific heat near 0°K, and to the
large electrical and thermal conductivities of these materials. In
the superconducting (S ) phase, the single-particle excitation is
radically different from that of normal metals. In super
conductors a minimum energy 2/1, called the energy gap, is required
to make a single-particle excitation from the ground state.
2-1 P H Y S IC A L N A T U R E O F T H E S U P E R C O N D U C T IN G
STA TE
This qualitative difference in the excitation spectra is paral
leled by a qualitative difference in the wave functions of the N-
and ^-phases of metals. In the AT-phase, the probability that
two single-particle states i and j are simultaneously occupied
24
T h e Pairing T h e o ry o f Su p ercon d u ctivity 25
is a smoothly varying function of the quantum numbers i and j.

For example, in a pure single crystal, the expectation value
P kV» = < A | n u % n | A > (2-1)
is a smoothly varying function of k and k' (so long as one does not
cross the Fermi surface in varying k or k'). Here \N} represents
a typical state in the normal phase and nfcT is the operator which
measures the number of electrons in k f , etc. (see the Appendix).
In the superconducting phase,8 the corresponding probability
P kk's = O SK tT vJS) (2-2)
is also a smoothly varying function of k and k' except when k
and k' are related by the “ pairing” condition. This condition
states that for a given state k, there exists a single mate k such
that the probability P kks is larger than P kk,s by a finite amount,
for all states k' in the vicinity of k. This singular behavior of
the two-particle correlation function, which has been stressed by
Yang,39 is no doubt the sort of picture F. London had in mind
when he suggested that superconductivity is due to a condensation
of the electrons in momentum space.1 When proper account is
taken of residual interactions conventionally neglected in the
description of the normal state, these “ pairing correlations”
leading to superconductivity emerge in a natural manner. Above
the superconducting transition temperature, the pairing correla
tions are broken up by thermal fluctuations and play no important
role in the normal phase.
It is essential to realize at the outset that the lowering in
energy of the $-phase due to interactions between mates (say
k | and k j ) of a given pair depends critically on the choice of
mates (k' f and k' | ) for other pairs. In fact, the energy gap and
most of the observed properties of the superconducting phase
would be absent were it not for strong correlations between the
pairs. The reason for the simple BCS model working so well is
that in real metals these pair-pair correlations are almost entirely
due to Pauli principle restrictions rather than correlations due to
true dynamical interactions between the pairs. This fact allows
one to treat the system in lowest order as if dynamical interactions
existed only between mates of a pair. The pair-pair correlations

would then be accounted for by working out this reduced problem
consistently with Fermi-Dirac statistics so as to include the crucial
Pauli principle correlations between the pairs. We shall call this
scheme the pairing (or BCS) approximation.
For a translationally invariant system, we shall see that the
pairing (k f , —k j ) of Bloch states leads to the lowest energy of
the system. Supercurrent-carrying states are generated by trans
lating this state of the system by an amount q/2 in k-space. The
pairing would then be (k -I- q/2 f , —k + q/2 | ) and the electrons
would have a net drift velocity vd = Sq/2m. More generally,
corresponding to each physical system and each state of that
system there is a choice of pairing of single-particle states which
minimizes the energy (or free energy, at finite temperature). For
example, in a superconductor with nonmagnetic impurities present
one should pair one-electron states <pn which include the impurity
scattering potential, as Anderson 40 first pointed out. He showed
that one should pair a state <pn and its time-reversed mate
<pn* to form the ground state of the system in this case. For
a uniform hollow cylinder in the absence of a magnetic field,
one would pair the state (n, m, k) with its time-reversed partner
(n, —m, —k ), where n and m are the radial and azimuthal
quantum numbers, respectively, and k is the wave number for
motion along the axis of the cylinder. In the presence of a
magnetic field, the best pairing depends on the thickness of
the cylinder and the strength of the field. For a thickness
d » X (the penetration depth) one would pair (n, m + v,k) with
(n, —m + v, k) or (n, m + v,k) with (n, —m + v -f 1, k) depend
ing on whether the flux trapped in the hole is an even or odd
multiple of the flux quantum hcj2e, that is, vhcje or (v + \)hc/e.
As we shall see, these are the only allowed values of the flux
trapped within a thick-walled superconducting cylinder. We shall
study in detail these various possible pairings in later chapters
when we apply the basic theory to physical problems.
While the above “ pairing” approximation gives a good
account of the single-particle excitation spectrum, there exist
The Pairing T h e o ry o f Su p e rco n d u ctiv ity 27
collective modes, such as the plasmons, arising from residual inter

actions neglected within this approximation. In addition there
may be small momentum exciton-like collective modes which lie
within the gap. For larger momentum, the exciton states rise
above the gap edge and pass into the continuum, thereby becoming
heavily damped. The nature of the collective states and their
effect on system properties is discussed in Chapter 8.
Also neglected within the simplest pairing approximation are
damping effects. In the strong-coupling superconductors, such
as lead and mercury, it is essential to include these effects on the
same footing as the pairing correlations to obtain a reasonable
description of these “ bad actors.”
The above discussion suggests that the excited states of a
superconductor can be represented by a two-fluid model, one for
the condensed electrons and one for the excitations. As we
mentioned in the introduction, phenomenological two-fluid models
(most notably the Gorter-Casimir model2 and the Ginsburg-
Landau model36) have played an extremely important role in
laying the ground work for our present understanding of super
conductivity. While there are important differences between the
predictions of the pairing theory and the earlier two-fluid models,
the theories share the basic idea that the superfluid electrons
(i.e., the strongly correlated pairs in our case) can be described
by a local density ps(r) and a local flow velocity vs(r). The
excited electrons then form an interpenetrating normal fluid
which in local thermal equilibrium can be described by the local
quantities pn(r) and vn(r). As we shall see, the superfluid can
only carry out potential flow, that is curl vs(r) = 0, a condition
emphasized by F. London.1 (No such restriction holds for the
normal fluid.) Many of the observed properties of superconduc
tors can be understood in terms of a two-fluid model having a
temperature-dependent energy gap for creating normal fluid
(excitations) from the superfluid component.160 As we shall see,
one can often interpret the results of the microscopic theory in
terms of such a model.9
2-2 T H E O N E -P A IR P R O B L E M
To understand the origin and consequence of pairing corre
lations, it is helpful to consider the problem, first studied by
Cooper,41 of a pair of electrons interacting above a noninteracting
Fermi sea of electrons via a velocity-dependent nonretarded two-
body potential V. Thus, all but two of the electrons are assumed
to be noninteracting. The background electrons enter the total
problem only through the Pauli principle by blocking states below
the Fermi surface from participating in the remaining two-particle
problem. If we measure the kinetic energy ek relative to its
value at the Fermi surface, only states with ek > 0 are available
to the interacting pair of electrons. Since the system is assumed
to be translationally invariant and one neglects spin-dependent
forces, the center-of-mass momentum hq of the pair and the total
spin S are constants of motion. The orbital wave function of the
pair can then be written as
^(r,, r2) = <p,(p)e">-R (2-3)
where the relative and center-of-mass coordinates are defined
by p = r: - r2 and R = (rx + r2)/2, respectively. The relative
coordinate wave function is symmetric for the singlet spin state
(S = 0) and antisymmetric for the triplet states (S = 1). In
the limit q —> 0 the relative coordinate problem is spherically
symmetric so that <p(p) is an eigenfunction of angular momentum
and can be labeled by the angular momentum quantum numbers
I and ni. For q / 0, the component of angular momentum
along q and parity remain good quantum numbers but I is no
longer sharp.
For simplicity, we first consider the zero momentum states
q = 0, so that ip can be expanded as
</'(ri> r2) = <p(p) = 2k afceik'p = 2k akeik''ie ~ik'r2 (2-4)
In (2-4), the sum is restricted to the available states (ek > 0).
Since the factors eik ri and e can be thought of as single
~ i k r 2
particle states of momentum k and —k, we see that the pair wave
function is a superposition of configurations in each of which a
definite pair state (k, —k) is occupied.
The Pairing T h e o ry o f Su p e rcon d u ctivity 29
To find the zero-spin eigenstates of the pair, we write

Schrodinger’s equation as
(W - H 0)ip = Vi/j (2-5a)
and from (2-4) one has
(W - 2ek)ak = 2 v kk’a *' (2-5b)
k'
where the matrix element Fkk' is defined by

Fkk> = <k, —k| F|k', -k '> (2-6)
In Figure 2-1 a typical scattering process caused by V is
illustrated.
While the Schrodinger equation (2-5b) cannot be analytically
solved in general, the solution is immediate if Fkk^ is taken to be
a factorizable potential Fkk> = Awk,*wk- More generally, if the
system is isotropic Fkk> can be expanded into its partial wave
components
Fkk. = 1=20 m=2 - IFj(|k |,|k 1 ) W k ) V " W (2-7)
F IG U R E 2-1 A typical transition occurring in C o o p e r’s problem in

which one pair of electrons interacts above a quiescent Fermi sea. The
center-of-mass momentum of the pair is chosen to be zero in this
drawing.
and the (I, m) eigenstates of the pair can be determined if Vx is

taken to be factorizable,
îdk l> lk 'D = \ w h w k ‘l* (2-8)
In this case we have from (2-5b)

(W,m - 2€k)ak = Alwkl ^ wk.l*ak. (2-9a)
«k = a k Y,m(Qk) (2-9b)
Equation (2-9a) can be written as
(2-10a)
where the constant C is defined as
(2-10b)
By substituting (2-10a) into the definition (2-10b) one obtains
the equation
1 = A, 2 IV I 2
k VV I m ~
53 <2-11>
determining the energy eigenvalues Wlm. If we work in a large
but finite box the single-particle energies ek form a discrete set
so that when W passes from below to above 2ek, &(W) jumps
from —oo to oo. As IT moves toward the next higher value of
2ek, O(W) again approaches —oo and jumps to + oo as W passes
through this higher value. The function <P(W) is shown sche
matically in Figure 2-2. As W passes through the origin to
negative values (i.e., the region of bound states) 0 (W ) increases
from — oo to zero as shown. The eigenvalues Wim are given
by the intersections of &(W) with the constant function 1/A*,
as shown for both positive (repulsive) and negative (attractive)
A,. While the eigenvalues in the continuum are trapped between
the unperturbed energies 2tk and approach the unperturbed
energies as the size of the box goes to infinity, a state is bound
The Pairing T h e o ry o f Su p ercon d u ctivity 31
split off from the continuum for an attractive /-wave potential,

as shown in the figure. For the simple case
1 0 < €k < wc (2-12a)
10 otherwise
and A, < 0, the binding energy \WXm\ of the pair in the split-off
state is given by
N( 0)
log I WlmI\W+ 2lOc (2-12b)
ImI
or
2cor
WlmI = (2-12c)
exp ' 2 ' 1
U(0)|A,iJ
-
We have assumed the density of states N(ek) is slowly varying

in the interval 0 < ek < cdc and have approximated it by iV(0),
the density of single-electron states of one-spin orientation,
evaluated at the Fermi surface. From (2-12c) one has for weak
coupling [iV(0)|Aj| « 1]
IWlmI 2toc exp iV(0)|A,|J
2
(2-13a)
while for strong coupling [iV(0)|Aj| » 1] one obtains
\Wim\ ~ N ( 0)1 A , K (2-13b)
F IG U R E 2-2 A plot of the function <P(W) [see (2-1 I)] which determines
the eigenenergies in C o o p e r’s one-pair problem. For a repulsive
interaction (A, > 0), all states are trapped in the continuum, while for
an attractive interaction, a bound state is split off.
32 T h e o ry o f Su p ercon d uctivity
From (5-13a) we see that the binding energy is an extremely

sensitive function of the coupling strength for weak coupling;
however, a bound state exists for arbitrarily weak coupling so
long as the potential is attractive near the Fermi surface. This
important result was discovered by Cooper,41 who suggested that
the instability of the normal phase, because of pairs of electrons
entering this type of bound state, was associated with the occur
rence of the superconducting phase.
Earlier work of Schafroth, Blatt, and Butler (SBB)42 is
closely related to Cooper’s discussion. Schafroth43 had suggested
that the superconducting state might correspond to a Bose-
Einstein condensation of pairs of electrons into localized bound
states. An attem pt to develop a theory along these lines was
made by SBB using what they call the quasi-chemical equilibrium
approach for evaluating the partition function of the system.
Owing to mathematical difficulties, they were not able to carry
out calculations based on their general formulation for any model
which exhibited superconducting properties. For a qualitative
picture they suggested a model with localized pairs such that the
size of the pair bound state is small compared to the average
spacing between pairs. The bound pairs would presumably be
capable of translational motion relative to the other pairs and
one would obtain a continuum of Bose-Einstein excitations above
the ground state without an energy gap, in contrast with the
pairing theory. If the pairs were indeed well separated they could
be treated as independent and Cooper’s discussion would be
appropriate. It should be pointed out, however, that, subsequent
to the work of Bardeen, Cooper, and the author, Blatt and
Matsubara extended the Bose condensation approach to give the
results of the pairing theory.
Actual superconductors differ in a fundamental manner from
a bound pair model in which the pairs are either well separated in
space and/or weakly interacting. As we shall see below, there
are on the average about one million bound pairs which have their
centers of mass falling within the extent of a given pair function.
Thus, rather than weakly overlapping pairs, one has just the reverse
T h e Pairing T h e o ry o f S u p ercon d u ctivity 33
limit, very strongly overlapping pairs. As we mentioned above,

it is surprising that one can meaningfully treat such a system in
zero order by including only dynamical interactions between mates
of a pair and neglect all but the Pauli principle restrictions when
treating interactions between the pairs. It is intended that the
discussion below and in the following chapters will help clarify
this point.
Returning to the one-pair problem of Cooper, it is interesting
to see how the energy of the bound state varies with the center-
of-mass momentum %q. If we assume only the 5 -wave (I = 0)
part of V is important (as appears to be the case except for the
crystalline anisotropy effects), one finds that the binding energy
lf(q) satisfies
where |k + q/2| and |k — q/2| are required to be greater than kFi

and the sum is to be cut off at = ek For small q one finds
coc .
w q\ = \w0\ -
vFhq
2
(2-15)
where
as above. Thus, the pair energy increases linearly with the center-
of-mass momentum in the limit q -> 0, rather than as q 2, as one
might expect. As Cooper pointed out, the drift of the pair with
respect to the noninteracting Fermi sea strongly reduces the
binding energy of the pair due to the reduced density of low-
energy states available to the pair. This effect dominates the
q2 increase of kinetic energy for small q.
If | W01is imagined to be of order k T c, (2-15) shows that the
pair would have lost most of its binding energy when
This number is roughly numerically equal to the reciprocal of

34 T h e o ry o f Sup ercon d u ctivity
Pippard’s coherence length £0 ~ 10“4 cm,33 about which we

shall have more to say. If one calculates the pair function <p(p)
from (2-4) and (2-10a) one finds the size of the bound state is of
order f0. Thus, one would be required to have an extremely small
density of bound pairs if an isolated pair model were to be appro
priate. In fact, the density would be so small that the predicted
N - S energy difference at zero temperature would be many orders
of magnitude too smali to agree with experiment.
Thus far we have considered only the singlet spin state of the
bound pair. If there is a strong attractive odd I potential, the
triplet state will have the largest binding energy and one might
expect the pairing in the superconducting state to be in a triplet
state. There is no experimental evidence to support other than
singlet pairing at present.
In closing this section we emphasize that the single-pair
model exhibits a continuous spectrum above the ground state,
without an energy gap.
2-3 L A N D A U ’S T H E O R Y O F A FERMI L IQ U ID
Looking back at Cooper’s argument, one might raise several
objections to the conclusion that the bound state in the two-
particle problem has anything to do with the occurrence of super
conductivity. For example, one knows that Coulomb and
phonon interactions between electrons in the normal state lead
to a correlation energy of order one electron-volt per electron,44
compared with the negligibly small binding energy W ~ 10“4 ev
of a bound pair. Is it not possible that the strong correlations
between all the electrons in the normal state will necessarily
lead to fluctuations which break up the weakly bound state of a
given pair of electrons? In addition, even if such a bound state
could exist in a metal, would not the strong overlap of the pairs
required to fit the observed condensation energy lead to inter
actions which would destroy the concept of bound pairs?
In answering the first objection, it is important to recognize
that Landau’s theory of a Fermi liquid45 gives a good account of
the low-lying single-particle excitations of the normal state. In
T he Pairing T h e o ry o f S u p ercon d u ctivity 35
this theory, the excited states of a normal metal are placed in

one-to-one correspondence with those of a free-electron gas. In
Landau’s theory, one asserts that the essential effect of the inter
actions between the electrons in the normal state of a metal is
to shift the effective mass of an electron (now called a “ quasi
particle”) by an amount which is observed to be of order 10 to 50
per cent. An important feature of the theory is that a quasi
particle, as opposed to a “ bare” electron, is a stable excitation in
the immediate vicinity of the Fermi surface (at sufficiently low
temperature). There is, however, a coupling between quasi
particles which arises from interactions neglected within this Fermi
liquid approximation. This residual coupling leads to super
conductivity.
The basis for Landau’s theory has been extensively investi
gated and one knows that the theory is correct in all orders of
perturbation theory46 starting from the noninteracting system.
The theory no doubt has a wider range of validity than that of the
perturbation series itself, although its exact limitations are not
known at present. Empirically, Landau’s theory works very well
in the normal state.
The remarkably small energy difference between the normal
and superconducting states of a metal (10“8 ev per electron)
strongly suggests that there is only a subtle shift of the electron-
electron correlations between the two states. Since Landau’s
theory gives a good account of the normal state, it is reasonable to
use the complete set of wave functions given by this theory as a
basis for constructing the wave functions of the superconducting
state. This procedure is particularly appealing because the super
conducting wave functions primarily involve normal state con
figurations in which quasi-particle excitations are present only
near the Fermi surface. However, these are just the configurations
which are best described by the Landau theory. Therefore,
Cooper’s result is to be understood in the sense that his two-
particle problem is actually a two-quasi-particle problem.
A difficulty with the above approach is that one knows little
about the interaction between quasi-particles in the normal phase
from an experimental point of view. While the quasi-particle’s

effective mass involves the forward scattering amplitude of two
quasi-particles, the effective mass also involves band-structure
effects which are difficult to estimate accurately. More important
is the fact that one needs the interaction for finite momentum and
energy transfer so that one is forced to estimate the interaction
theoretically. While this problem is not completely settled at
present, it appears that one understands the essential features of
the interaction, the remaining complications arising primarily
from detailed crystalline effects (see Chapter 7).
In regard to the second objection mentioned above, it is true
that the simple picture of bound pairs of electrons forming the
ground state of the superconductor is impaired by their great over
lap. Nevertheless there remains the strongly correlated occupancy
of a given quasi-particle state (say k f ) with its mate (say —k j )
as in Cooper’s problem. Interactions between quasi-particles
which tend to break up this correlated occupancy are presumably
already included in Landau’s description of the normal state.
Thus a simplified model in which one includes pairing correlations
between otherwise noninteracting quasi-particles is not at the
outset an unreasonable starting point. It is this point of view
Bardeen, Cooper, and the author took in constructing the micro
scopic theory of superconductivity.
2-4 T H E P A IR IN G A P P R O X IM A T IO N
We saw above that for a translationally invariant normal
system carrying no current, the q = 0 pair state is the most un
stable, in the sense that it is the pair with the largest binding
energy W. In Chapter 7 a time-dependent treatm ent of the
normal state instability is given and one finds the greatest growth
rate is for q = 0 pairs, in agreement with Cooper’s result. Thus it
is natural to solve the reduced problem in which interactions are
considered only between electrons of opposite momentum. One
hopes that the resolution of the strongest instability will modify
the system so as to remove the q ^ 0 pair instabilities as well.
This is exactly what happens. We restrict our attention to
pairing electrons of opposite spin orientation.
T he Pairing T h e o ry o f S u p e rc o n d u c tiv ity 37
We shall use the formalism of second quantization to describe

the interacting electron system; this scheme is reviewed in the
Appendix. The creation and destruction operators for electrons
of wavevector k and ^-component of spin s ( f or | ) are denoted
by c ks+ and c ks> respectively. They satisfy the usual Fermi
anticommutation rules. The Hamiltonian for the reduced problem
of the q = 0 pairs is
tfred = 2 e*n» + 2 + (2-17)
ks k k'
where the pairing matrix element V is given by k >k
VWk = <*', ~k'\ V\k, - k > (2-18)

and the operator bk + creates a pair of electrons in the single
particle states k f and —k \ , that is,
h + — r +r
°k - ckt c -k i
+ (2-19)
bk — c - k l ck t
This type of Hamiltonian forms the basis of the theory of super
conductivity proposed by Bardeen, Cooper, and the author.8
Further argument for concentrating on these particular inter
actions in describing the ground state and the low-lying excited
states are given in the original paper by Bardeen, Cooper, and the
author and in a review article by Bardeen and the author.9 There
it is argued, on the basis of phase-space considerations as well as
effects due to the antisymmetry of the wave functions, that the
q = 0 pair state should be macroscopically occupied in cases
where the superfluid momentum density is zero (although vs need
not be zero in the presence of magnetic fields). We expect the
matrix element Vk,k to be predominantly negative near the Fermi
surface for superconductivity to occur. As we shall see in Chapter
7, this attraction is due to the ions overscreening the Coulomb
repulsion, thereby reversing the sign of the effective interaction.
While (2-17) is written in terms of bare single-particle operators
c ks, the reduced Hamiltonian can be formally viewed as a model
Hamiltonian describing residual interactions (Vk'k) between the
quasi-particles in the normal phase as discussed above. Since the
pairing correlations constitute a fractional change of only ~ 10“8
in the total correlation energy of a metal, it is clear that this more
liberal interpretation of H red must be adopted. When the quasi

particle picture of the normal phase excitations is insufficient, as in
lead and mercury, other techniques must be used in treating the
superconducting phase (see Chapters 6 and 7). Note that the
form of H red depends upon the choice of pairing, as discussed in
Section 2-1.
Since the pairing interaction maintains the pairing condition,
it follows that the eigenstates of H red can be labeled by those
states k, s which are occupied, their mates —k, —s being
unoccupied. This labeling of states leads to a one-to-one corre
spondence of the eigenstates of H red with the eigenstates of a
noninteracting Fermi gas (or the normal state). As we shall see,
if excitations happen to be in states k f and —k j , special care
must be taken so that these configurations are properly orthogonal
to the ground-state wave function.
It is clear that if V is attractive, the ground state of H red has
no pair state (k f , —k | ) occupied by a single electron. In this
case the operator wfcT + n _ ki can be replaced by 2bk+ bk that is,
twice the pair occupation number. The reduced Hamiltonian is
then
tfred° = 2 + 2 (2-20)
k k .k ’
It might be argued that eigenstates of H red° follow immediately

by forming new operators B n which are linear combinations of the
bk s such that H red is of the form ]>n +Bn- This argument is
incorrect. If the operators bk and bk + described true Bose particles
(rather than pairs of fermions) the B n’s and B n+’s would also
describe Bose particles and the ground state would be formed by
placing all the bosons in the lowest state. Rather, one finds by
direct calculation the commutation relations
[bk, b k. +] = 0 for k + k' (2-21a)
A , V +] = 1 - (wu + n -k i) for k = k' (2-2lb)
and
[bk, b k.] = 0 = [bk +, b k. +] (2-21c)
These are not of the form required by Bose-Einstein statistics.
The factor (nfct + n _ k i ) in (2-21b) represents the effect of the
T h e Pairing T h e o ry o f Su p e rcon d u ctivity 39
Pauli principle acting on the individual electrons forming the pair.

Perhaps it is simplest to view the “ pairon” operators bk and bk +
as satisfying Bose-Einstein statistics for k ^ k' and satisfying the
Pauli principle bk +2 = 0 = 6fc2fork = k \ It is the latter relation
which ruins a simple Bose gas picture.47
In attempting to find a variational estimate of the ground-
state energy and wave function of H red, the author tried to adapt
the intermediate coupling approximation of Tomonaga, familiar
in the coupled meson-nucleon problem and the polaron
problem.48,49 In these problems one assumes that successive
bosons (mesons or phonons) are emitted into the same orbital state
(about the proton or the electron, respectively). The form of the
orbital state <p and the weight A N of the N/2 boson state are then
determined by minimizing the system energy. Lee, Low, and
Pines49a simplified the procedure for the polaron by assuming
what is equivalent to a parameterized form for the weights A N.
Their wave function, after a canonical transformation has been
made to eliminate the electron’s coordinate from the problem, is
i<Ao>a nk cMa*+ +a"fc)i°> (2_22)
where the a + ’s are phonon creation operators. The function gk
is essentially the Fourier transform of the orbital wave function <p
of the phonons surrounding the electron.
The application of this physical idea to superconductivity is
complicated by several features. First, the “ pairon” operators
do not truly satisfy Bose statistics and, second, the number of
electrons is a definite number N 0 in our system, rather than being
a probability distribution | ^ | 2 of finite width about the value
N 0. The author tried to describe the ground state of H red by
i<Ao>“ n 9 e * K + 1°> = n u + £ a +)|o> (2-23)

k k
where we have used the fact that bk +2 = 0 in expanding out the

exponential. The normalization integral is easily seen to be
< ^ o > = n (i + b*i2) (2-24)
k
so that
(2-25)
is a properly normalized state. By expanding out the infinite
product one sees that |0O> has a nonvanishing amplitude for all
even numbers of electrons, 0, 2, 4, .... However, by choosing gk
appropriately the mean number of particles described by |0O> can
be adjusted to be the required number N 0. As in the grand canon
ical ensemble, one can show that the width of the distribution is
of order N 01/2 so that particle number fluctuations cause no
difficulty in a large system.
Since we want to minimize the ground-state energy subject
to the constraint
<-£0|2\Uô> = <<Ao| I nks\4<0> = N 0 (2-26)
On combining (2-20) and (2-27) one finds the quantity to be

minimized is8
W = ^ 2(ek - n)vk2 + ^ Vk.kukvkuk.vk. (2-28)
k k. k'
where uk and vk are defined by

1
(2-29a)
(1 + 9 k 2 ) 112
and
9k
(2-29b)
= ( T W 2
thus
(2-29c)
We have assumed phases are chosen so that Vk>k and gk are real
quantities. On minimizing W one finds
1
T he Pairing T h e o ry o f Su p ercon d u ctivity 41
and
u kv k = 2 j |“ (2-30c)
where E k is defined by
E k = +[(cfc - /x)2 + ^ 2]1/2 (2-30d)
As we shall see, E k turns out to be the energy required to create a
quasi-particle of momentum k in the superconducting state. The
“ energy-gap” parameter A k satisfies the integral equation
| m~, ( 2 - 3 0 e )
One must simultaneously solve (2-30e) and the constraint condition

<<Ao|ôp|-Ao> = 2 £ »fca = (2-31)
k
to determine A k and p,. If the single-particle energy ek is measured

relative to the Fermi energy in the normal state, /z is just the shift
of the chemical potential between the normal and superconducting
states. For a system possessing particle-hole symmetry in the
vicinity of the Fermi surface, one finds /z = 0. In general, this is
an excellent approximation and we shall assume /z = 0 in solving
for A k. Once the energy-gap equation (2-30e) is solved, one can
obtain the energy difference WN — Ws between the N- and S -
states by inserting the expressions for uk and vk back into (2-28).
An explicit solution of (2-30e) is easily obtained if VkW is approxi
mated by the s-wave potential
V = f ~ V < 0 for M and kte-l < /9 q9X
kk' - \ 0 otherwise ( )
so that Vkk>is attractive in a shell of width 2<oc centered at the
Fermi surface. In this case one finds
fj0 for |ek| < coc
k 10 otherwise ( )
where
A q = .1
Jo ’ HsinhE Z T f “ 2" c exp |2' 34’
[N (0)V \
The right-hand equality holds in the weak-coupling limit

N ( 0)F < J. By substituting this result into the expression for
the ground-state energy (2-28) and subtracting the ground-state
energy of the normal phase (i.e., the unperturbed Fermi sea in
this model) one finds the condensation energy
WN - Ws = iiV ( 0)A 02 2 2N(0)u>cz exp [ ~ iV(0) f ] ^2‘35*
Since thermodynamics 4 gives the relation
WN - Ws = 07T
(2-36)
where H 0 is the critical magnetic field for destroying super
conductivity at zero temperature, we find
H 0 = 2[ttN (0)]1/2 A q (2-37)
By using experimental values of iVÔ) and AQone finds values of
H 0 which are in reasonably good agreement with experiment.9,16
We note that the condensation energy (2-35) is not an
analytic function of the coupling constant iVÔJF so that a
perturbation treatment starting from the normal phase could not
give this result unless one sums an infinite number of graphs of a
selected class.50
Returning to the ground-state wave function (2-23), we
would expect on the basis of the Tomonaga scheme that the
projection of |(/f0> onto the iV-particle space would lead to a
function (in the coordinate representation) of the form
S2> ' **rN> = *AoN
= ^ 9 ( ri - r2)xi2?(r3 - r4)x34- ’ * ^ - 1 “ tn)Xn - i . n (2-38)
The function cp is the relative coordinate wave function of a pair
(the same function fo r all p a ir s ) and yi; is corresponding spin
function f (i) ^ (j). Thus within the pairing approximation all
pairs are in the same state in the ground-state wave function.
The operator s / in (2-38) antisymmetrizes the entire function.
T he Pairing T h e o ry o f Su p e rco n d u ctiv ity 43
This result, first noted by Dyson, can be obtained by taking the

inner product of |t/r0> with the basis vector
1*1, « i; *2, «2 ; • • • r N, «w> = K + (r i)«As2 + (**2) • • •</'s„+ ( M ° >
and expanding the ip+’s in terms of the creation operators ckS{ +

by
*ASl + (r() = 2 e - ik r‘CteS(+ (2-39)
k
The A-particle state (2-38) has been discussed by B latt51a in the

case where cp is a general function of rx and r2. This generalization
corresponds to considering pairings between states other than k \
and —k | .
The orbital function cp in (2-38) is given by
9>(p) = 2 3 ,fceik'p
k
(2’4°)
thus gk is the Fourier transform of <p(p), as stated above. While
there is a formal similarity between (2-38) and the wave function
for a condensed Bose-Einstein gas of pairs of electrons with
opposite spin, the antisymmetrization operator stf is all important
in real superconductors. In fact, the (unnormalized) ground
state of the noninteracting Fermi gas can be written inthis form
with
* - {J !*! I X (2-41>
so that antisymmetrization removes the correlations between
opposite spin electrons implied by cp in this case. In the super
conducting state gk differs from (2-41) only for values of k in the
immediate vicinity of the Fermi surface. This difference is
reflected in a long-range tail of <p(p) which increases the probability
of two antiparallel spin electrons being near each other out to a
range £0 = %vf IttAq ~ 10“4 cm, that is, Pippard’s coherence
length. As we mentioned above, this quasi-bound state has such
a long range in space that on the average about 106 other pairs
have their centers of mass in this region. [In this estimate
electrons deep within the Fermi sea have not been counted since
they behave essentially as if the material were in the normal
phase.] Thus an isolated pair picture has little meaning here.
2-5 Q U A S I-P A R T IC L E E X C IT A T IO N S
To find the excited states of the BCS reduced Hamiltonian
(2-17) we consider adding an electron to the system in the state
p f (its mate —p j being empty). The only effect of this process
is to block the pair state (p f , —p | ) from participating in the
pairing interaction (due to the Pauli principle). Since —p j is
assumed to be empty, the electron in p f cannot be scattered out
of this state, due to the form of the pairing interaction (2 -2 0 ).
Of course, residual interactions not explicitly included in H red will
allow this process to take place; however, these interactions appear
to have a small effect on the excitation spectrum (since they are
implicitly included in the quasi-particles of the normal state).
The quasi-particle energy is defined to be the total excitation
energy of the system when the extra electron is added to the sys
tem. From (2-28) we see that by deleting the pair state (p f ,
—p | ), the energy of the interacting pairs is increased by
(2-42)
To this we must add the single-particle energy €p of the added
electron. The total excitation energy is given by
ep[l - 2vp2] + 2 ApUpVj
p^pVp (2-43)
where we have used the gap equation (2-30e) to simplify the
interaction energy term. If we use the results (2-30) for uk and
vk (with ft = 0 ), we find the excitation energy
(2-44)
Thus, the parameter E p defined by (2-30c) is just the energy
required to create a quasi-particle in state p f . A plot of E p vs. p
is given in Figure 2-3. The minimum energy required to add an
electron to the system is A kf = A 0 ~ 1 0 " 3 — 1 0 " 4 ev. In prin
ciple the chemical potential /z should be shifted a small amount to
ensure <A> = N 0 + 1 in the excited state; however, this correc
tion has negligible effect in a large system.
The Pairing T h e o ry o f Sup e rcon d u ctiv ity 45
F IG U R E 2-3 The quasi-particle energy E k in the superconducting

state plotted as a function of the wavevector k. The energy
Ek = (ek2 + Ak2)112 differs from the corresponding energy \ck\ of the
normal state only in the vicinity of the Fermi surface. The energy gap
observed in experiments which do not inject or withdraw electrons
from the system is 2 A 0t a minimum energy A0 being required to create
each quasi-particle produced in a one-electron transition. Note: All
energies are measured with respect to the Fermi energy.
In the above calculation, nothing has been said about p being

above (or below) the Fermi surface. Since the pairing interactions
smooth out the jump in the single-particle occupation numbers
(n k} in the normal phase at the Fermi surface, as shown in Figure
2-4, there is a finite probability of being able to add an electron
to the system in a state p below the Fermi surface. The excitation
energy is positive in this case as it is for |p| > p f. In an analogous
way one can calculate the energy required to remove an electron
in the state p f from the ground state. One again finds the
energy E p( > 0 ) regardless of whether |p| is greater or less than the
Fermi momentum.
Therefore, the minimum energy required to create a single
particle-like excitation from the superconducting ground state is
2 A0, A0 for removing an electron from one state and A0 for placing
it in another state.
It is important to realize that the states created by adding an
electron to | 0 On) in state p f or removing an electron from state
—p | in ]0 OAr> are identical within the pairing approximation
except that the number of superfluid pairs in the two states
differs by unity. If instead of |^0Ar) we work with the state |</r0>
(2-23), which represents an ensemble of ground-state wave

functions averaged over systems with • • • N — 2 , N, N + 2 • • •
particles, these two states generated by cpT+ and by c _ pi are
truly identical, aside from a normalization factor. This result
is established by noting that
Cpt +|(Ao) = cpt + n (u k + v k^k + ) |0)
k
= u pcp1 + n ( uk + wfc6„ + )|0> (2-45a)
k*p
= «p|lApt>
and
c-pi|*Ao) = c - p l n (Uk 4" v kPk+ ) |^)
k
- V p t + k*p
= n (uk + ^ A +)|°> (2 -4 5 b)
= -^pI'Apt)
where |^pT) is the normalized one-quasi-particle state
IfApt) = CP,+ k*p
n K + « A + )| 0 > (2-46)
F IG U R E 2-4 A plot of the average occupation number <nk> of the

Bloch states in the superconducting state if one makes a single-particle
model for the normal state. The occupation number in the normal
state, shown here as I for k < kF and 0 for k > kF, is also rounded due to
normal state interactions, although a discontinuity of <n k) presumably
remains. The “ sm earing” of the Fermi surface by the pairing correla
tions occurs only over a range ~ I0 “4A:F about the Fermi surface.
The Pairing T h e o ry o f Sup e rcon d u ctiv ity 47
An important mathematical simplification occurs if one

considers the linear combination
y v\ + = ^p^pt ^p^-pi (2-47a)
of the two equivalent operators. From (2-45) we see that y pT +
applied to | 0 O) creates the normalized state
rpt + |0o> = I*Ajpr> (2-47b)
The orthogonal combination
Y - pi = upc -pi + vpcp i + (2-48a)
when applied to |t/r0> leads to the null-state vector (not to be
confused with the vacuum | 0 >)
y-pi |<Ao> = 0 (2-48b)
The relations (2-47a) and (2-48a), and their Hermitian conjugates
yPt = VpT - vpc- pI + (2-49a)
y-pi + = ^ - p i + + vpcpt (2-49b)
were introduced independently by Bogoliubov 52 and by Valatin . 53
These relations are known as the B-V transformation. As the
notation suggests y p t + and y . p i + create quasi-particles in states
p f and —p \ , respectively, while ypt and y_pi destroy quasi
particles in these states. Thus one has the relations
ypr + |ô> = I'/'pt) (2-50a)
y-pj +|<Ao> = I'/'-pi) (2-50b)
YpM o) = 0 (2-50c)
y-pj|<Ao> = 0 (2-50d)
The last two relations are equivalent to the statement that
is the vacuum state for quasi-particles. It follows by direct
computation that the quasi-particle operators satisfy Fermi-Dirac
statistics:
{yPs> Vp's' } ^ ^pp' $ss' (2-51a)
{yPS. Vp's-} = {yPS+. yP-S' +} = o (2-5ib)
and can be thought of as leading to excitations that form a weakly
interacting Fermi gas.
It is important to remember that the y and y + ’s must operate

on the ensemble-averaged states, e.g., |0 O), not on the ^V-particle
projection of these states \^0N}. If one thinks of ypT + as acting
on an iV-particle state, it would appear that a quasi-particle is a
linear combination of a particle and a hole. This is not correct.
In the iV-particle system a quasi-particle of momentum p and
spin s is nothing more than an electron definitely occupying the
state p, s with its mate —p, —s being definitely empty. In
configuration space, the N + 1 particle wave function correspond
ing to 1 0 pT> is
0 pt(ri> 5 i» ’ ' ‘ r 2V+ l> sn +1 )
= s / ( p ' ( r-L — T2 )Xl 2 cP (r3 4 34
— r )X ***
<P(tn- i — tn)Xn- 1, Ne X P (^P * TN+l) f N+1 (2 -5 2 )
where 9>'(p) is given by (2-40) with the term k = p deleted. For
some purposes it is convenient to discuss the excitation in terms
of the empty state —p, — s and call the excitation a “ hole.” In
other cases one prefers to concentrate on the occupied state p, s.
The wave function for the pair state (p f , —p j ) is the same in
either case regardless of the words used to describe it. However,
one must keep in mind that the number of superfluid pairs differs
by unity in the two descriptions of the same state.
An excited state having quasi-particles in k ^ , k 2 <s2- • -knsn
is given by
|<Afc1s1.k2S2.-fc„s„> = y/cjs, +yk2s2 + ■■’yk„s„+ |o> (2-53)
The excitation energy is E kiSi +E k2S2 + • • •E knSn. The Bogoliubov-
Valatin operators have the important property that the excited
StatC \<PPu -pj> = yPT+y-Pi +l‘/'o> (2-54)
is orthogonal to the ground state. This is not true if one generates
the excitations by applying cpT+cpl+ to |ip0). In the original
BCS treatm ent these doubly excited pair states (called “ real”
pairs as opposed to the “ virtual” pairs occurring in the ground
state) were treated separately and were represented by the factor
(vp - upbp +) (2-55)
The Pairing T h e o ry o f Su p e rc o n d u c tiv ity 49
in the wave function. However, this is just the factor one

obtains by expressing the y + ’s in (2-54) in terms of the c and c +’s
and simplifying the factors involving bp+ in the wave function.
It has been shown by several groups 54 that for H red the
variational solutions for \ifj0} and the quasi-particle spectrum are
exact in the limit of a large system so long as the number of
excitations present is small compared to the number of electrons
participating in the pairing interactions. In the next section we
shall see how these results are generalized to finite temperature
where the latter condition is not satisfied.
2-6 L IN E A R IZ E D E Q U A T IO N S O F M O T IO N
In the original work of Bardeen, Cooper, and the author a
complete discussion of the thermodynamic properties of the
superconducting state was carried out within the pairing approxi
mation. As for T = 0 , their treatment of the system described
by H red is exact in the limit of large volume. In agreement with
experiment, they obtained a second-order phase transition at T c
and an exponentially vanishing electronic specific heat for
T < \Tc. For a discussion of this work and its comparison with
experiment, the reader is referred to the original BCS paper 8
and to a review article by Bardeen and the author . 9
Rather than repeating the BCS finite-temperature treatment,
we would like to illustrate an alternative procedure based on a
linearization of the equations of motion for the single-particle
operators cpt+ and c_pi. The discussion follows closely a treat
ment given by Valatin , 55 and leads to results identical to those
of BCS.
To fix ideas, we begin with the reduced Hamiltonian (2-24)
-®red = k2,s' k2, k' ^k'k^k'^^k (^"^)
(although, owing to our approximations, the scheme gives
essentially the same results if the full two-body interaction is
considered). The basic idea is to find eigenoperators, say /za +
and up, which satisfy
[#red > Ha + ] = Qa Pa * (2-56a)
and the Hermitian conjugate relation

[-®red> /h?] = (2-56b)
where the Q’s are positive quantities. It follows that the eigen-
operators (xa + and jjl0 create and destroy excitations of the system
since by applying the operator equation (2-56b) to the ground
state \ifj0} of H red we find
[ ^ r e d> K + ]|^0> = (# r e d ~ M 'o K ^ o ) = |<Ao> (2 -5 7 )
Thus 11pay= p.a +\ipoy is an eigenstate of H red with excitation

energy Qa.In an analogous manner one finds that ^ lowers the
energy of the system by Q#, from which it follows that the ground
state (or the excitation vacuum) satisfies
V-Mo) = 0 (2-58)
for all j8 . Operators which approximately satisfy (2-56) pre
sumably give an approximate description of the excitations.
Except in extremely simple systems, the exact operators jjua + can
neither be found nor are of great interest since physically interest
ing probes (i.e., external fields, injected particles, etc.) create
complicated superpositions of such excitations (see Chapter 5).
Suppose we try to find an operator which adds a quasi
particle of momentum p and spin f to the ground state of H recl.
The simplest fermion operator which will add this momentum and
spin to the system is cpT + , so we try
[^red5Cp!+] = €pCpl + + 2k' ^ k'p^k, +c-pi (2-59)
In the absence of the interaction cpT+ satisfies (2-56a) with the
excitation energy Qa = ep. The “ excitation” energy is just the
energy required to add an electron to the system in state p. In
the presence of V, cpt+ is no longer an eigenoperator. In fact
we must go out of the operator subspace of ckt+ and include
products of the forms c +cc in constructing fjLa+. When this more
complicated guess for fia+ is commuted with H rca, still higher
order polynomials in c + and c appear. In most cases the series
continues on to infinite order, just as the series of equations
determining the Green’s functions, which are discussed in Chapter
5. To obtain a tractable problem we must cut off the chain at a
The Pairing T h e o ry o f Su p e rco n d u ctiv ity 51
certain order by approximating the commutators. Whether the

termination is meaningful clearly depends on the physics of the
problem. Fortunately, the interactions in H red are sufficiently
simple that one can cut off the series by including in a sense only
a linear combination of c and c +.
By taking matrix elements of (2-59) between the iV-particle
ground state | 0 , N } and the N + 1 particle state \p f , N + 1 )
with one quasi-particle present in state p f , we have
(•Qpt - cpK p f , N + l|cpr + |0, N>
= 2 Vk,p( p ] , N + 1 |c _pj |a, N + 2 > <a, N + 2\bk, + \0,N)
a.k'
(2-60)
where the sum is over the eigenstates of the 2^ + 2 particle
system. If we measure all energies relative to the chemical
potential, /x = lim (W0 N+n — W0 N)jn , where n » 1 , £?pT is the
nlN-^0
energy required to add a quasi-particle in p f to | 0 , N }. We
argue that for a large system the intermediate state sum is given
by retaining only the N -f 2 particle ground state. It is not that
the matrix elements of bk>+ for all other a are small compared to
the one for a = 0 , but as we shall see the matrix element of
c_pl entering the equation is small when the a ^ 0 matrix element
of bk,+ is large, and therefore the product is negligible. Thus
(2-60) becomes
(Qp - ep) F p = 2 VkpB kGp (2-61)
where
F p = ( p \ , N + l|cpt + | 0 , N ) (2-62a)
Gp = ( p \ , N + l|c_ pl|0, N + 2 > (2-62b)
B k = < 0 , N + 2|6te+10, N > (2-62c)
Another relation between F and G can be obtained by taking the
matrix element of (H red, c _ p l) between the states 10, N -f 2 ) and
\p f , N + 1 ). If the intermediate state sum is again replaced
by the single term a = 0 , one finds
(Qp + ep)Gp = 2 VkpB kF p (2-63)
52 T h e o ry of Su p ercon d u ctivity
where we have chosen phases so that all quantities are real. The
secular equation for (2-61) and (2-63) is
Qp €p Ap
= Qp2 - € P2 - Ap2 = 0 (2-64)
Ap Qp 4- €p
where the parameter Ap is defined by
dp = “ 2 (2-65)
k
(The matrix element B k is still to be determined.) From equation

(2-64) we find the (positive) excitation energy Qp of the state
\p f , N + 1 ) is given by
tip = + ( * P2 + ^ P2 ) 1/2 = Ep (2 - 6 6 )
The negative energy root —E p corresponds to the process in
which a quasi-particle in the time-reversed state —p j is destroyed.
The eigenoperators /za+ = ypT+ and ^ = y_pi corresponding to
the positive and negative energy solutions are of the form
ypt + = Uptpi* — VpC-piR + (2-67a)
and
7 - Pi = UpC-p | + VpRCpt + (2-67b)
The operator R + transforms a given state in an N -particle system
into the corresponding state in the N + 2 particle system; thus
R +\0, N ) = |0, N + 2 > (2 - 6 8 )
and
R +\lc,s]N> = \k,8-,N + 2 >
while
R\0,N + 2 > = | 0 ,^ > , etc.
By inserting the eigenvalues back into (2-61) or (2-63) and
requiring that the y + and y’s satisfy Fermi anticommutation
relations it follows that up and vp are given by
T he Pairing T h e o ry o f S u p ercon d u ctivity 53
The formal similarity between these results and those of the last
section is complete if we require that | 0 , A> be the ground state
of the system, that is,
ypt| 0 , A> = 0 (2-70a)
and
y _pi | 0 , A ) = 0 (2-70b)
Thus, the y + ’s create noninteracting fermion excitations from the
“ vacuum state” 10 , A>.
By inverting (2-67a), (2-67b), and their Hermitian conjugates
to solve for the c-operators in terms of the y’s, one finds from the
definition of B k (2-62c) the relation
bh = ukvk =A (2-71)
On combining this result with che definition of Zlp (2-65) we find
an equation determining the parameter zJp:
( 2 ' 7 2 )
which is just the energy-gap equation (2-30e). Thus the excita

tion energies are identical in the two approaches and the quasi
particle operators differ only by the presence of A .56
As in the BCS treatment, it is straightforward to generalize
these results to finite temperature. The only change is that
instead of the ground state | 0 , A ) appearing one has a typical
state | T, A> excited at the temperature T. All goes through as
above except for the relation (2-71). For T # 0 one has
B k = (T , N + 2\bk +\T, N> = A (i _ f k1 _ f _ kl) (2 - 7 3 )
where / ks is the expectation value of the quasi-particle occupation
number yks +yks in the state \T, A ). Since the quasi-particles
are essentially independent fermions (whose properties change
slowly with temperature) f ks is given by the Fermi distribution

function
(2-74)
1
A e$Ek _j_ i ( E k > 0)
and
Bk = tanh p E k (2-75)
By inserting this result into (2-65) we obtain the finite-temperature
BCS gap equation
-2 v
A k i . PEk
/” 2E k U n h - r
‘
(2-76)
This finite-temperature treatm ent of the pairing theory is
entirely equivalent to the BCS treatment, which, as we mentioned
above, gives an exact account of the system described by the
reduced Hamiltonian (in the limit of large volume). If Vkp is
approximated by (2-32), A k is again of the form (2-33) where
A0(j3) satisfies
1 _ fWc de
tanh A + V ) 1' (2-77)
A0)T = Jo > 2 + ô 2 ) U2
As T increases from zero, A 0 decreases as shown in Figure 2-5,
vanishing at the transition temperature T c. Thus, T c is given by
de
tanh 2 kRT r
(2-78)
F IG U R E 2-5 A plot of the temperature dependence of the energy-gap

parameter A0(T). Note that A0 vanishes with infinite slope as T T c,
leading to the second-order phase transition.
The Pairing T h e o ry o f Su p ercon d u ctivity 55
In the weak-coupling limit this gives

(2-79)
so that the ratio 2 A 0(T = 0 )jkBT c is 3.52 in this limit. While
this ratio is in reasonably good agreement with experiment for
weak-coupling superconductors ,9 , 16 the ratio is too small to account
for the observed ratio for lead and mercury. It now appears that
temperature-dependent damping effects account for the dis
crepancy .57
The free energy of the superconducting state
F s = Ws - T S (2-80)
can be obtained by calculating the expectation value of i / red
with respect to the typical state | T, N ) and using the standard
expression
s = - 2kB 2 {/fc log f k + ( 1 - f k) log ( 1 - /,)} (2-81)
k
for the entropy of the quasi-particle (normal) fluid, where f k
is given by (2-74). The energy Ws is easily seen to be 8
The bulk critical magnetic field H C( T ) is given by

^ = F n (T) - F S(T) (2-83)
where the free energy of the normal state F N(T) is given by
(2-80), (2-81), and (2-82) with A k = 0 . H 0 is plotted in Figure
2 - 6 for the potential (2-32). The electronic specific heat can be
calculated from
ces = 2kBp* 2 / k( 1 - A ) [ E | ^ - 2] (2-84)
56 T h e o ry of Sup ercon d uctivity
F IG U R E 2-6 A plot of the critical magnetic field versus temperature.
and is plotted in Figure 2-7. The jump of the electronic-specific

heat at T c is due to A2 being proportional to (T c — T) near T c
so that the derivative in (2-84) is discontinuous at T c.
The reader is referred to the literature 9 16 for a detailed
discussion of the thermodynamic properties of the system.
F IG U R E 2-7 A plot of the electronic-specific heat as a function of T JT .

The Pairing T h e o ry o f Su p ercon d uctivity 57
2-7 C O N C L U D IN G REM ARKS

Before closing this chapter we would like to make a few
remarks.
1. While there is a formal similarity between the ground
state (2-38) of H red and a condensed Bose-Einstein gas, the
analogy must be used with care due to the strong overlap of the
pair functions. As a result of this overlap the excitation spectrum
in real metals exhibits an energy gap rather than a continuous
spectrum characteristic of a Bose gas. If the treatment is ex
tended to include the interactions neglected in H red and one as
sumes all interactions to be of short range, a continuous boson
spectrum, starting at zero energy appears in the energy gap
corresponding to density fluctuations in the electron system. In
real metals these low-lying boson modes are pushed up to the
plasmon energy (~ 10 4 x 2 zl0) due to the long-range Coulomb
interaction between electrons so that there are no low-lying
boson modes except for dressed lattice vibrations (phonons) in
cases of physical interest .58
2 . The discussion in this chapter has emphasized states for
which the momentum density of the superfluid is zero, i.e., the
pairing (k f , —k \ ) was treated. If the Hamiltonian of the
electron system were Galilean-invariant, states with finite super
fluid momentum could be formed by translating the zero-
momentum eigenfunction by an amount q / 2 in momentum space.
The transformed wave function would be
exP (| 2 9 ' r,)ô(ri«i. r2s2. •••
= - r2) exp iq • 2 + r2) t l i 2

J^<p(r1
•9>(rAT-i - rw) exp iq ’ (iw-i + rN) t jv-1 i n (2-85)

so that the “ center-of-mass ” wave function of a pair would go
from the q = 0 state to the plane wave state of momentum q.
For states involving mass flow of this sort, a condensed Bose gas
picture may be helpful; however, one must be cautious in using
this picture in detailed calculations. In particular since the fixed

ions will create a magnetic field in the moving frame, the function
9? will change its form. Also, if the wavevector q varies appreci
ably over a coherence length the separation into center-of-mass
and relative coordinates is of questionable value.
3. We have concentrated on singlet spin pairing with <p
being an «s-state in the absence of crystal anisotropy. The prob
lems of triplet spin pairing and I ^ 0 orbital states have been
treated by a number of authors 59 and we refer the reader to the
literature for a discussion of these questions. In addition,
pairing of states other than Bloch functions can be easily handled,
since the basic scheme does not rely on the form of the states
being paired.
4. In Section 2 - 2 we saw that not only the q = 0 pairs are
unstable if we consider fluctuations about the normal state, but
the q # 0 pairs are also unstable. In Chapter 8 we investigate
the stability of the ground state |</r0> given by the pairing approxi
mation. As mentioned above, there are no unstable pair fluctu
ations about this state. This result is due to the finite energy
required to create from the superfluid the quasi-particles which
one tries to bind together by the residual interactions.
5. In Section 2 - 6 we stated that the matrix element
<a, N -f 2\bk +\0, N } is large not only for the N + 2 particle
ground state 10 , N + 2 ) but that there is another state a giving
a large matrix element. Specifically, the two-quasi-particle state
|A: f , - k | , N + 2 ) gives
|< * t , - k ] r , N + 2 |&fc+|0 ,iV>| = V = ^ ( l + (2' 86)
compared to the matrix element
|<0, N + 2\bk + 10, N )I = ukvk =A (2-87)
which we retained. For k on the Fermi surface, both of matrix
elements are equal to 1 /( 2 )1/2. However, the matrix element
-f l\c_p l\k\, —k I , N + 2 > which multiplies (2 -8 6 )
in (2-60) is zero since c _ pi does not affect the quasi-particles
The P airing T h e o ry o f Su p e rco n d u ctiv ity 59
in A: | and —k | . On the other hand, the matrix element

( p f , N + l|c_ p4 | 0 , N + 2 ) which multiplies (2-87) in (2-60) is
equal to —vp which is 1 /( 2 ) 1/2 for p on the Fermi surface. There
fore we are justified in retaining the single term a = 0 in the
intermediate state sum.
6. While the ground-state wave function \ip0} [see (2-25)]
represents an ensemble average of ground-state wave functions
| 0 on> f°r systems having an even number of electrons,
|<Ao> = 2 A^ > (2-88)
N(even)
we can obtain \>p0Ny from \ifj0} if A N is arranged to be of the form
\Aj<\eiN*. Then
|^> = 2 I^ I« w1 *on> (2-89)

N (e ve n)
and
1 C2n
\An.\ Itos-y = 2^ Jo 0>dcp (2-90)
By our choice of phases (2-32) is just |^0°) so ^hat
given by
lô®) = n K + c2‘*»A +)|0> (2-91)
k
that is, a factor of e i(P is contributed by each creation operator c +.
Therefore the normalized N' particle ground state is given by
1 f 271
^ 0 N = 2tt|A n\ J0 e~ iN V>I~I A + e2'“’vA +)|0> d<P (2-92)
where the amplitude of the N' particle state is given by
i r 2n
\An.I2 = — (uk2 + e2l*vk2)d<p (2-93)
Z7r Jo k
The probability \AN,\2 is sharply peaked about the average
number N 0, having a w idth 60 of the order of iV01/2. The fact
that the average energy (ô^l^redl'Ao0’) is independent of 9? should
not be interpreted as a degeneracy of the ground state of a physical
N particle system. Since the N — 2 , N, N + 2 , . . . particle
systems are completely independent, the average energy of these
systems should be independent of the relative phase of their wave

functions.
For a large system, whether the physical system has an even
or an odd total number of electrons makes no difference in its
macroscopic properties; thus the wave functions above apply for
any N. The situation is distinctly different for pairing corre
lations in atomic nuclei, where these differences lead to the well-
known even-odd effects.61
CHAPTER 3
APPLICATIONS OF THE
PAIRING TH EO RY
Since the BCS theory was originally proposed, attempts to

justify the pairing correlations basic to the theory have pro
ceeded along two lines. The first approach has been to apply
the BCS theory to a wide variety of phenomena in superconductors
and check the theoretical predictions of the pairing approximation
against experiment. The second approach has been to treat by
various approximate methods the residual interactions neglected
within the pairing scheme, hoping to show that these residual
interactions introduce no major change in the predicted properties
of the system. Both approaches have enjoyed considerable
success. Owing to the remarkably good agreement between the
pairing theory and a broad class of experimentally observed
phenomena, it would appear that the first approach has success
fully established the validity of the pairing concept upon which
the theory is based .9 , 16
3-1 J U S T IF IC A T IO N O F T H E P A IR IN G H Y P O T H E S IS
In this chapter we shall follow the first approach and review
the calculation of a number of system properties within the pairing
61
approximation. In the few cases where a discrepancy between

theory and experiment exists, one can often attribute the differ
ence to limitations of our understanding of normal state effects
(details of band structure, phonon spectra, electron-phonon inter
actions, etc.). In Chapters 7 and 8 we take the second approach
and discuss several theoretical advances which take proper account
of retardation and damping effects. We also treat certain classes
of residual interactions, neglected within the simple pairing
approximation. Within the framework of these more elaborate
treatments the predicted system properties are essentially in
agreement with those given by the pairing model; where differ
ences appear, the agreement between theory and experiment is
generally improved by the more elaborate treatment.
3-2 A C O U S T IC A T T E N U A T IO N RAT E
While the majority of the electron-phonon interaction has
been accounted for in forming the wave functions for the normal
and superconducting states, there remains the part corresponding
to resonant phonon absorption and emission processes. These
resonant processes give rise to attenuation of acoustic waves
(dressed phonons). To calculate the time rate of change of
( N qoô), the number of phonons of wave vector q 0 and polarization
A0, we consider a typical state |/> excited at temperature T .62
Within the pairing approximation |/> is of the form
K> = - k, s(occ.)
n y*s+-J L nqt x K | 0 . (3-1)
L Q.A T >
where the quasi-particle operators are defined as in Section

y ks+
2 - 6 appropriate to the temperature T and act on states with a

fixed number of particles. The state 10 , T> of N Q electrons is
the vacuum state for these operators, that is,
yks(T)| 0 , T ) = 0 for all k and s (3-2)
The average of the quasi-particle occupation numbers in a small
region r of k-space taken about a particular value k is given by the
Fermi distribution
2
k in r
<7 IYks + Y k s \ I ) =
A p p lic atio n s o f the Pairing T h e o ry 63
although the occupation number vks = </|y/cs+y/cs| - 0 f°r a given

state k, s is either one or zero for the state vector given by (3-1).
We assume the dressed electron-phonon interaction is of the
form
tfei-ph = 2 9ppuK a + a - vi +)cP's+cps (3-4)
PP'S/l
where q = p' - p + K (see Chapter 4). We shall use the golden

rule 63 to calculate the transition rate ( N qoAoy so that we need
matrix elements of H el_ph between |/> and all states |F> which
are degenerate with |/> and differ from it by having a single
particle change its state and a phonon q 0A0 either absorbed or
emitted. If the phonon energy -ojQqAo is smaller than the energy
gap 2 A(T), which is the case in most acoustic attenuation experi
ments, additional quasi-particles cannot be created from the super
fluid so that only quasi-particle scattering processes enter. The
final states are of the form
]Fy = fyp2 5 +yplS«a0 A„K> absorption
lyplS+yPa A0 + 17> emission
where q0 = p 2 — Pi + K . Since H el_ph conserves the total num
ber of electrons (as does yp>S'+yps), |/> and | F } both describe
states of the N 0 particle system. The matrix elements are
readily evaluated by transforming (3-4) to the y-representation
(i.e., make a particle conserving B-V transformation). From
(2-67) we find
C p t + = u p Y p r + + v pY - p i r + (3-6a)
c Pt = u p Y p X + v p R Y - p i + (3-6b)
c - p I + = « Py - p i + ~ ^ p Y p 1 R + (3-6c)
c - p i = « Py - p i - v p R y p t + (3-6d)
Since the y’s are linear combinations of c’s and c + ,s it follows that
there are two terms in H ei_ph which connect |/> with a given final
state |F>; in particular, the factors cP2 t +cPit and c _ P ii+c _ P2i
lead to the same quasi-particle transition. The matrix elements
arising from these two factors must be added before squaring the
total matrix element. Note that if one makes a single-particle

approximation for the normal state, only a single term contributes
in the normal state. This difference is characteristic of the super
conducting state and appears in most dynamical properties of the
system. Thecross terms which enter in squaring the total
matrix element are known as “ coherence effects” and have
important experimental consequences, as we shall see below.
From (3-4) we see that the factors multiplying cp,t +cpt and
c _ p i +c..p,i are identical so that we are interested in the combina
tion
c p ' T + c pt + c - p i + c -p'i = n (P> P ' ) [ y p ' t + y P t + y - p i + y - P' i ]
+ ra(p,-p,)[yp't+y-pi +^ - yPry -p 'i^ +] (3“7)
where we have used the transformation (3-6). The so-called
“ coherence factors” m( p, p') and n(p, p') are defined by
m(p, p') = upvp, + vpup, (3-8a)
and
n( p, p') = upup. - vpvp. (3-8b)
(We shall meet two more coherence factors I and p below in dis
cussing spin-flip processes and the electromagnetic response of the
system.) Returning to (3-5), the matrix element for phonon
absorption is given (for s = \ ) by
= »(Pl, p2)fPlt(l - PP2t)[^,0A0]1/a 9M o (3-9)
< F \ H el_vh\ I}
where fPit and pPzT are the quasi-particle occupation numbers

for the initial state and are either one or zero, as mentioned above.
N QoAo is the phonon occupation number for this state. The rate
of absorption of ^0A0 phonons is then
MW = 2tt X 2 PlP'
2 \gPP-A0\2n2(p, p'K tC 1 - V , ) ^ o ,
x 8(EP. - E p - wQoXo) (3-10)
where the sum is restricted by momentum conservation p' =
p + q 0 + K. The factor of two accounts for the absorption by
A p p lic atio n s of the Pairing T h e o ry 65
quasi-particles of both spin orientations. In a similar way one

finds the emission rate is given by
êmiss = ^l7T * ^ 2
PP'
I^PP'ôl ^ P )^p't(^ ^pt)(^<?0ô^
x 8(EP. - E p - a>QoJ (3-11)
so that the net rate of absorption is
dN -
= SoAoôAo - (3-12)
where the acoustic attenuation rate is given by
«?oAo = 477 2 l£jpp'A0|2n2(P> P'Km - V t)
pp'
S ( E P- - E p - w,oAo)
(3-13)
and S isthe spontaneous emission rate. When the sum is
performed, the occupation numbers can be replaced by their local
average values and one has
ato*o = 4 n pp'
2 |£pp-a„|M p, P')(/p - fp') H E P. - E p - w?oAo)
(3-14)
where p' = p -f q0 + K and K is a reciprocal lattice vector. The
spontaneous emission rate S'qoAo is given by (3-11) with N QoAo = 0.
The expression for a simplifies if we assume that g depends
only on the momentum transfer (i.e., q 0 for normal processes).
For | q01 small compared to kF only normal processes enter the
sum with appreciable weight, so that a is given by
«<oA0 = 27r |^ oAo|2 2 (X+ ~ /P‘>
x 8(EP. - E p - u>qoXo) (3-15)
where p' = p + q0, and the relation
»’(p, p') = (upup. - vpvp. f = \ (l + €p€p'E J ’ Ap) (3-16)
has been used. Since the speed of sound is small compared to
the Fermi velocity, —\0~3vF in typical cases, energy and momen
tum conservation require that q0 be essentially tangent to the
F IG U R E 3-1 Coordinate system with q0 chosen as the polar axis and

p' = p + q0-
Fermi surface for those states p which contribute to (3-15). For

spherical energy surfaces, only those states near the Fermi surface
lying near the equatorial plane (perpendicular to q0) enter the
problem. If we approximate the energy contours of the Bloch
states by spheres with the effective mass m*, the sum in (3-15)
can be expressed as
1 f m *2 f
2 — J ( s (3‘17)
where q0 is the polar axis as shown in Figure 3-1. The epep. term
in the coherence factor vanishes upon integration since the re
maining factors in the integral are even in ep and €p ;64 thus
(3-15) reduces to
a q0 *o
I 1/7 . l ^ |q0|2
2ir
x j d e p dep. ( l - ^ ; ) (]j(fp
/ p ~- U )) 8(EP' -~ E p -
fp aj,o/lo)
(3-18)
x [f(E ) - f ( E ’)]
where E' = E + a)QoAq and we have taken A to be independent of

p. For most experiments wqoAo « A so that we may set E = E f
in (3-18) except in the Fermi factor and obtain
QOAq
(3-19)
Qq
1)
77 I9qo*o I
^
ho\
This expression applies to the normal state if we set A = 0 , so
that the ratio of the acoustic attenuation rates in the S- and N-
phases at the temperature T is
« s(T ) m 2______ (3-20)
« n(T ) m
+ 1exp A (T )
!cb T
Therefore the temperature-dependent energy gap can be obtained
0.2 0 .4 0 .6 0.8 1.0
T /T c
F IG U R E 3-2 The longitudinal acoustic-attenuation coefficient in the
superconducting state relative to that in the normal state compared
with the result of the simple pairing theory.
from ocs/aN. In Figure 3-2 the theoretical ratio is compared

with experimental results for longitudinal phonons in tin and
indium.27ai 65 The rapid drop near T c reflects the rapid decrease
in the number of excitations as the energy gap opens up below
T c. We also note that the large density-of-states factors
EE'
{E2 - A2)1I2(E'2 - A2)112
are cancelled by the coherence factor ( 1 — A2/ E 2) in (3-18) for

cjQoAo « A so that only the Fermi factors enter the attenuation
rate in this limit.
The energy bands and the gap parameter are anisotropic in
real metals so that as/aN measures a complicated average of A
over the regions discussed above. Variations of the averaged
gap of the order of 1 0 per cent have been observed in single
crystals as q 0 is rotated relative to the crystal axes . 27
F IG U R E 3-3 The relative acoustic-attenuation coefficient for trans

verse waves in tin as measured by Bohm and Morse.
In the above discussion it was tacitly assumed that the

electron-phonon matrix element is the same in the N- and S-
phases. Although this is most likely true for purely longitudinal
phonons , 66 this assumption is incorrect for transverse phonons.
A transverse phonon sets up transverse electromagnetic fields in
addition to crystal potential effects.273- 67 While the screening
of the crystal potential is essentially the same in the N- and S-
states, the Meissner effect drastically reduces the electromagnetic
coupling between the transverse phonons and the electrons.
Thus, one expects an essentially discontinuous drop of the trans
verse acoustic attenuation rate upon entering the superconducting
state at T c due to the Meissner effect eliminating the electro
magnetic coupling. The remaining coupling should be accurately
treated by the above analysis and one indeed finds this to be the
case, as shown in Figure 3-3.67b
3-3 N U C L E A R -S P IN R E L A X A T IO N RA T E
The above calculation of the acoustic attenuation rate is
easily modified to account for the relaxation rate of oriented
nuclear spins due to their hyperfine coupling to the valence elec
trons. The interaction for a given nuclear spin I is of the form
Hi s = ^ 2 a k * a k[h(ck't X t - c^+Cfci)
k, k'
+ I +Ck'i+Ck\ + I - Ck"\+Ckl] (3-21)
where a k is proportional to the amplitude of the Bloch function
Xk(r ) the nuclear site in question, so that a _ k = afc*, and
I ± = Ix ± H y 66 To calculate the rate at which a given nucleus
decreases its 2 -component of spin we observe that the Zeeman and
hyperfine energies are in general small compared to the energy
gap so that only quasi-particle spin-flip processes enter. We
consider a typical initial state |/> excited at the temperature T ,
as in the preceding section, and notice that the final states are of
the form
7p2
t + 7pi i 1 -0 nuclear spin flips down (3-22a)
Vpi i +Vp21 1^) nuclear spin flips up (3-22b)
70 T h e o ry of Su p e rcon d u ctivity
As before, H l s contains two terms which connect | T } and a

given final state; for the final state (3-22a) one has the terms
A I - ( a p 2 * a P l CP2 1 + C P l l + « - p , * « - p aC - p l t + C -p a i)
= A I _ a P2*aPi(cP2l+cPii + c_Pif+c_P2l) (3-23)
By transforming to the y-representation these terms become
A I - a P2 *^pi[^(Pi >P2 )(Y p 2 t Y p i I “h 7 - Plt Y - p 2 i )
+ 7>(Pi>P2)(rp2t +r - Plt + ^ - y Pli Y - P2i ^ + )l (3_24)
where the coherence factors I and p are defined by
HPuVz) = uPiuP2 + vPivP2
p ( P 1 . P 2 ) = UP VP2 - VP UP2
The nuclear spin-flip-down matrix element is then
A ar2*a pJ(Vi’ P2 K 1 i ( 1 - vP2^){I_)n (3-26)
where the last factor gives the nuclear matrix element. The
transition rate for flipping down the nuclear spins is proportional
to
^down = 2 tt\A\2 2 | « p J 2 | « p J 2*2(Pl>P2)
X f Pl( l - /p2) S(EP2 - E Pl - co)N, (3-27)
Thus the rate of decay of the z-component of nuclear magnetiza
tion is proportional to
P 1 .P 2 * \ ± j P x £ j P2 >
x / Pld ~ f P2)Z (E P2 - E Pl - w) (3-28)
if we neglect crystalline anisotropy, since l2(px, p2) is given by
<’ (P „ P .) - I ( l + ’ ’ ' ' “ e * e 2 A ” ) (3 ' 29>

and the term in ee' vanishes on integration as above. If A is
independent of p near the Fermi surface, (3-28) can be written
for w « A as
as = 47r| A | 2 |a | 4Ar2 (0 )
E ( E + w)kBT( — 8f /dE) d E
x E ( E + a>) ( E 2 - A 2)1I2[(E + oj)2 - A 2]112 1 ’
Ja
A p p lic atio n s of the Pairing T h e o ry 71
so that the ratio of the relaxation rates in the S- and jV-phases is

, r [E(E + CO) + A * ] ( - d f l d E ) d E
«N ~ L [E 2 ~ ^ 2]1,2[(E + O, ) 2 - J 2] 1/2 1 >
If QJ is set equal to zero in the (3-31) the integral is logarithmically

singular at the lower limit. If a> is calculated from the Zeeman
energy, Hebei and Slichter 28 found the ratio increased to about
1 0 before falling to zero as T - > 0 . Experimentally they found
the ratio increases to about 2 in aluminum so that gap ani
sotropy and spacial inhomogeneity may well be the limiting
feature. Perhaps the most interesting and important feature of
F IG U R E 3-4 (a) Ratio of the nuclear-spin relaxation rates in the normal

and superconducting states in aluminum. Solid curve calculated by
L. C. Hebei for a smeared energy gap. (b) The nuclear relaxation time
T 1 of superconducting aluminum. The solid curves are based on the
pairing theory with the density of states smeared by folding the density
of states with a square function of width 2d and height ( 2 d ) -1 . The
dotted and solid curves were calculated with 2A(0)/kBT c = 3.52 and
3.25, respectively, the latter value being that found by Biondi and
Garfunkel from microwave experiments.
the ratio ccsjaN for spin relaxation rate is its predicted rise as T
drops below T c as shown in Figure 3-4. This result is distinctly
different from the result for the acoustic attenuation rate. The
only difference in the predicted rates for the two types of processes
is that the coherence factor n2 appears in the acoustic case while
I2 appears in the spin relaxation case. As we mentioned, the
anomalously small matrix element for quasi-particles near the
Fermi surface being scattered by phonons exactly cancels out
the large density of quasi-particle states in this vicinity. On the
other hand, the quasi-particles are coupled to the nuclear spins
with essentially the -same strength as single particles in the
normal state so that the large density of quasi-particle states near
the Fermi surface leads to an increased relaxation rate. Of
course, at low enough temperatures few quasi-particles are excited
so that the relaxation rate goes to zero as T -> 0 . It is clear that
a simple energy-gap form of a two-fluid model could not account
for the sharp drop in the acoustic attenuation rate near T c and
simultaneously a rapid rise of the nuclear-spin relaxation rate.
It is interesting to note that the beautiful experiments of Hebei
and Slichter were being carried out during the period when the
BCS theory was being formulated, and that their experiments
gave one of the first substantiations of the detailed nature of the
pairing correlations which are basic to the theory.
3-4 E L E C T R O M A G N E T IC A B S O R P T IO N
Another example of resonant energy absorption is the real
part of the electrical conductivity oq in a thin film. If we de
scribe the electromagnetic field by the vector potential
A(r, t) = A0ei(«'r- w0 + c.c. (3-32)
the first-order coupling is of the form
H A{t) = - ^ Jj(r) • A(r, t) <Pr
= 2 Ao • (2p + q K + a . / v -1"* + H c - (3‘33)

p, s
As above, there are two terms in a given sum which lead to the
same quasi-particle transition; for example, the combination
CP'T+CPT - c_pi+c_pU = Z(p, p')(yPM+ypt - y - p i +y - p^)
- p(v> P')(yp't+y-pi + ^ + yPiy - v ’\R + ) (3-34)
enters here due to (2k -b q) -> —(2k -f q) as k —> —(k + q) and
(k + q) -» -k . The first terms on the right-hand side of (3-34)
lead to quasi-particle scattering and contribute at T ^ 0 while
the last terms lead to creation or destruction of two quasi-particles.
They contribute only if to ^ 2A.
For simplicity we consider only T = 0 so that absorption
occurs only for to ^ 2 A. By calculating the rate of photon
absorption just as we did for phonon absorption, one finds
<71S 1 E) - A2] dE
I f f (a , -
a1N to ) A (E 2 - A 2)ll2[(w - E)2 - A 2]112 1 ’

Mattis and Bardeen 69 have carried out the integral in terms of the
complete elliptic integrals E and K and find the ratio of the
conductivities in the S- and iV-phases is
+ (” 6)
where x — oj/2 A ^ I. A plot of the theoretical ratio is shown in
Figure 3-5 and is in quite good agreement with experiment.
For general w and temperature, the integrals must be done
numerically. For to « 2 A, a shows a rise as T decreases below
T c, as in the case of nuclear spin relaxation, followed by the low-
temperature exponential drop. For to > k T c/2, the ratio no
longer shows a peak. Several cases have been worked out by
Miller.70 Note: At the time of the first printing of this volume, it
had been reported that a precursor absorption (co/2A~0.85) was
observed at low temperature.71,72 Subsequent measurements and
improved processing of the data showed that the precursor was
an artifact. Collective modes which were proposed to account for
the precursor gave too weak an absorption to agree with experi
ment (see Ch. 8).
hco/2 J ( 0)
F IG U R E 3-5 The frequency dependence of o1/<jn and a2jan at T = 0
as calculated by Tinkham from the w ork of Mattis and Bardeen.
3-5 P H Y S IC A L O R IG IN O F T H E C O H E R E N C E
FACTORS
Aside from the coherence factors, one might have guessed
the results of this chapter on the basis of a simple single-particle
energy-gap model for the superconductor. The physical origin
of the coherence factors is, however, fairly simple.
Suppose we are interested in a process in which a quasi
particle is scattered from an initial state, say k f , to a final state,
say k' f , by absorbing a boson (a phonon or photon) of momentum
k' — k. For simplicity we assume there are no quasi-particles
in the states —k j , k' f , and —k' j initially. (The argument is
easily generalized to include excitations in these states.)
As we saw in Chapter 2 , a quasi-particle in k f (and none in
—k j ) corresponds to an electron definitely occupying the Bloch
state k f (i.e., with unit probability) and the mate state —k |
being definitely empty. The pair state (k' f , —k' j ), with no
quasi-particles in it, has a probability amplitude uk>of k' f and
—k '| being empty, and an amplitude vk> of these states being
simultaneously occupied. Since u k>2 + vk 2 = 1 , there is zero

amplitude for other possible occupancies of this pair state. There
fore, the initial state of the system can be viewed as having an
k' t
a m p l it u d e u k> a m p l it u d e vk>
k k' t
- k'l (c) k1 k l (d) k ,i

a m p l it u d e u k a m p l it u d e vk
a m p l it u d e u k a m p l it u d e vk
F IG U R E 3-6 (a) and (b) The two configurations entering the wave func
tion for a state with a quasi-particle in k f . (c) and (d) The tw o config
urations entering the wave function for the state with a quasi-particle
in &f , showing how a) -> c)
and b ) - > f ) when the electrons couple
to a field which does not flip the electronic spin (acoustic or electromag
netic fields are examples), (e) and (f) The tw o configurations entering
for a state with a quasi-particle in — j , showing how a)->e) and
b) /)w h en the electrons couple to a field which flips the electronic spin
(the hyperfine coupling involved in nuclear spin relaxation is an example).
amplitude uk> for the single-particle occupancy shown in Figure

3-6a and an amplitude vk> for that shown in Figure 3 -6 b. For
clarity we do not show how states other than the four mentioned
above are occupied.
In analogy, the final state has k' f definitely occupied and
—k' | definitely empty, while there is an amplitude uk that k \
and —k^ are empty, and vk that these states are occupied, as
shown in Figure 3-6c and 3-6d.
The interaction Hamiltonian, being a one-body operator, can
at most change the state of one electron. Consider the operator
c/c't+c/c!: ft transforms the portion of the initial state shown in
Figure 3-6a into the portion of the final state shown in 3-6c, as
indicated by the solid line. The probability amplitude for this
process is clearly nkuk>. In addition, the operator c _ k i +c _ k>i
transforms the portion of the initial state shown in Figure 3-6b into
the final-state configuration illustrated in 3-6d. The amplitude
for this process is —vkvk>, the minus sign arising from the fact that
when the operator is applied to the state shown in 3-6b it produces
the negative of that in 3-6d because of the ordering of the operators
which describe the two states, as one can easily verify. Now if
the coefficients of ck^ +ck] and c _ k i +c _ k>l in the interaction
Hamiltonian are identical (including sign), the overall amplitude
for the quasi-particle transition is ukuk. — vkvk> — m(k, k') as in
acoustic attenuation due to scattering of quasi-particles. Al
ternatively, if the coefficients in the Hamiltonian are equal in
magnitude but opposite in sign, one obtains the total amplitude
ukuk> + vkvk> = Z(k, k'), as in electromagnetic absorption by the
excitations.
In the case of spin flip, as for the nuclear spin-relaxation
problem, the corresponding final state would be a quasi-particle
in —k']{ with the configurations shown in Figure 3-6e and f.
In this case the operator c _ kfi+ck] transforms 3-6a into 3-6e
with amplitude nkuk, while c _ k i +ck,T transforms 3-6b into 3-6f
with amplitude vkvk,. Since the coefficients of these operators
are identical in the hyperfine interaction, the ^-coherence factor
enters here.
A p p lic a tio n s o f the Pairing T h e o ry 77
When two quasi-particles are created in k f and —k' j

from the superfluid with no quasi-particles present initially, one
has the unique final-state configuration shown in Figure 3-7a.
The only configurations in the initial state which are connected
to this final state by a one-body operator are shown in Figure
3-7b and c, which are transformed into 3-7a by cfcf+cfc»t and
c _ k,i +c _ kl with amplitudes ukvk. and uk.vk, respectively. For
acoustic attenuation (i.e., coefficients of the same sign) one has
the total amplitude ukvk, -f uk.vk = m( k, k'), while for electro
magnetic absorption (opposite signs) one has ukvk> — uk.vk =
2 ?(k, k'), in agreement with the results above. The spin-flip pair-
creation process follows in an analogous manner.
From the above discussion it is clear that there are in general
only two terms in the c, c + representation of a one-body operator
which contribute to a given quasi-particle process and that the
contribution due to each term can be understood by these simple
pictures. Notice that the number of particles is explicitly
a m p l it u d e 1
a m p litu d e u kvk> a m p litu d e u k.vk
F IG U R E 3-7 (a) The only configuration entering a state with quasi

particles in and — k' j . (b) A configuration which is connected to
that in (a) by a spin-independent one-body operator, (c) The only
state other than that shown in (b) connected to that shown in (a).
conserved between the initial- and final-state configurations,

although the number of particles in single-particle states other than
(k \ , —k j ) and (k' f , —k' j ) depend upon which configuration
of these states one is considering. For example, the number of
electrons not shown in Figure 3-6a is two larger than that shown
in 3-6b since the total number of electrons in each configuration
is exactly N 0.
3-6 E L E C T R O N T U N N E L IN G
The three examples discussed above, acoustic and electro
magnetic absorption and nuclear spin relaxation, all involve
transitions between states of the N 0 particle system. The
problem of electron tunneling between two metals which are
separated by an insulating layer involves transitions between N 0
and N 0 + n particle states of each metal. In his pioneering ex
periments, Giaever 26 observed that at sufficiently low temperature
no current flowed between a normal metal and a superconducting
metal separated by a thin oxide layer, unless the applied voltage
V (multiplied by the electronic charge) exceeded the energy-gap
parameter A of the superconductor. One might intuitively expect
this result on the basis of an energy-gap model. On the other
hand, one might argue that if two electrons tunneled simultane
ously, they could be bound together by pairing correlations once
they entered the superconductor and no excitation energy would
be required, suggesting that a current flow is possible even for
very small voltages. Clearly, as the oxide thickness vanishes,
this situation would obtain. However, for thick layers one might
expect a very small probability for two electrons to tunnel simul
taneously so that in this case little current would flow until the
one-particle threshold voltage is reached. That this is not the
case was pointed out by Josephson , 83 who showed that the super
fluid pair tunneling rate is of the same order of magnitude as the
single-particle tunneling rate. We shall discuss two-particle pro
cesses below and concentrate on one-particle processes for the
moment.
The foundation for a Hamiltonian formulation of the tunnel
ing problem was laid by Bardeen 74 and refined by Cohen,
A p p licatio n s o f the Pairing T h e o ry 79
Falicov, and Phillips ,75 and more recently by Prange .76 In this
approach, one describes the system by an effective Hamiltonian
H = Hx + Hr + Ht (3-37)
where H x and H r are the full many-body Hamiltonians for the
left and right metals in the absence of tunneling, and H T is a
one-body operator which transfers electrons between the two
metals,
H t = 2 {r ^kk’ck'sr+cksl + H.c.} (3-38)
kk's
Bardeen has shown that T kk>is given by the matrix element
of the current density at center of the oxide taken between single
particle states which decay exponentially as one moves into the
oxide layer. Harrison 77 has evaluated T kw within the WKB
approximation and finds
I\Tmkk]219 = 1
7 2 r
>fc|<T' eX P
Pi Pi
where p± is the one-dimensional density of states for motion
r
J xl
Jc±(x) dx (3-39)
normal to the barrier which has boundaries of xx and xr.

To calculate the tunneling current,78, 79 we begin with the
zero-temperature case and treat H T by first-order time-dependent
perturbation theory. The rate of transferring electrons from I to
r is
Wl^r — 277 2 K a lK/?r| 2 -^fcfc'Ck'sr + Cfcs*|0z)|0r)|2 ~ V)

a, d kk’s
(3-40)
if the electrons decrease their potential energy by V in moving
from I to r due to the applied bias. In (3-40) the state vectors
are the exact many-body eigenstates of H x and H r, that is,
= *«h> ,3 . n
H r \Pr> = ‘ B \Pr> ^ }
and the energies ea and are measured relative to the ground-
state energies in I and r, respectively. At zero temperature
electrons cannot tunnel in the reverse direction due to energy
conservation. From (3-40) one can readily see that the current
density is proportional to
rv
I(V ) oc j V N T+r(E)N T_ l(V - E )dE (3-42)
where
N T+r(E) = ^ |<|Sr|<v + |0r>|2 8(e/, - E)
k .0
s iVr(0) P dekPS +\ k , E ) (3-43a)
and
J - 00
N T_ l(E) = 2 Kai|cfc|0/ ) | 2 S(ea — E)

k, a
s JV,(0) Jf*- 00
dekPr \ K E ) (3-43b)
The spectral weight functions p(+) and p(~) are discussed in Section
5-7 and are related to the one-electron Green’s function G by
p( +)(k, to) = —- Im 7T
G (k, w) to ^ 0 (3-44a)
p(_ )(k, at) = - Im
7T
G(k, —w) co ^ 0 (3-44b)
(See Chapter 5.) Thus a knowledge of G(k, co) for each metal
suffices to determine the tunneling current. We note that in
deriving (3-42) wehave assumed T kk> = const, in an energy
region V ( « E F) about the Fermi surface, which is a very good
approximation for voltages of interest in investigating the super
conductor aspects of the tunneling characteristic.
The effective tunneling density of states for a superconducting
metal can be calculated in the pairing approximation with a non
retarded two-body potential directly from the definition (3-43).
By using the particle conserving B-V transformation (3-6) we
find
where we have used the fact that if a given ek makes the argu
ment of the delta function vanish, so does €k> = —ek so that
uk2 + uk>2 = 1 . This simple manner in which the coherence
factors vanish in the expression for the tunneling current was first
pointed out by Cohen, Falicov, and Phillips . 75 It is interesting to
note that N T+ is just the density of quasi-particle states which
we used earlier in this chapter, as one would have guessed on the
basis of a simple energy-gap model without coherence effects.
In a similar manner one finds for a nonretarded pairing
potential model
(3-46)
where the coherence factors vk2 vanishes in the result just as u 2
did in N T+.
As we shall see in Chapter 7, the expressions (3-45) and (3-46)
are incorrect in real metals due to the strong retardation effects
associated with the phonon interaction between electrons. There
one finds the simple result
(3-47)
as opposed to N(0)(E — \ dA2jdE)l[E2 — A2(E)]112, which follows
from (3-46).
Returning to expression (3-42) for I(V ) we find for tunneling
between a normal and a superconducting metal that
(3-48)
if we use (3-47), where I s and I N are the currents flowing
when the superconductor is in the S- or A-state, respectively.
Therefore, the tunneling experiment can give detailed information
about the energy dependence of the gap parameter.
The finite-temperature tunneling current can be treated in an
analogous manner if one includes a thermodynamic average over
initial states rather than using | 0 f >|0 r>, and further one includes
currents from I to r and r to Z.78, 79 The tunneling current can
then be expressed in terms of the spectral weight functions for the
thermodynamic Green’s functions. In the simple case of a non
retarded two-body pairing potential, the expression reduces to the
golden rule result for a simple energy-gap model without coherence
effects, as for T = 0 . A typical I - V characteristic for tunneling
between a normal and a superconducting metal is shown in Figure
3-8 for several temperatures for an energy independent A . Experi
mental curves are in general agreement with theory although small
deviations exist, some of which we shall discuss below.
A pictorial view of the one-particle tunneling process between
a normal and a superconducting metal is illustrated in Figure
3-9a and b. In 3-9a, an electron in k f , beneath the Fermi
surface in the normal metal tunnels through the oxide to state
k ' f above the Fermi surface in the superconductor. In the pro
cess a hole is left behind in Zgiving an excitation energy ea = |efc|
for this metal. In addition, a quasi-particle is placed in k ' f
giving an excitation energy e# = E w = (ek 2 + Ak 2)112 for the
F IG U R E 3-8 The tunneling current I between a normal and a super

conducting metal as a function of the applied voltage V (multiplied by
the electronic charge).
superconductor. This process can go only if the pair state

(k' f — k' | ) is initially empty; this occurs with the probability
uk 2. Energy is conserved if \ek\ + E k> = V. Another energy-
conserving process shown in Figure 3-9b is identical to that shown
in 3 -9 a, except for the quasi-particle being placed in a state k" f
beneath the Fermi surface. F o rJ = constant, the two states are
related by ek> = —e^, as we saw above. The probability that
(k" f , —k" j ) is initially empty is uk»2 = vk>2 so that the total
probability for the process to go is uk 2 + vk 2 = 1 , as far as the
Pauli principle restrictions are concerned. The tunneling current
is then given by summing only over states above (or below) the
Fermi surface in the superconductor and replacing the coherence
factor by unity, as we saw above.
Figure 3-9 (a) and (b) Tw o final states for a given initially occupied
state k \ . These processes enter the expression for the single-particle
tunneling rate between a normal and a superconducting metal.
Figure 3-10 (a) and (b) Processes analogous to those in Figure 3-9,
however the tunneling is between superconductors here.
A semiconductor model of the excitations in the super

conductor is often helpful for discussing tunneling phenomena.
However, the model must be used with caution since states “ above
the energy gap” in this model are really linear combinations of
quasi-particle states above and below the Fermi surface, i.e., as
are k' f and k" f in our example. The reader is referred to the
work of Bardeen 74 for further details on the semiconductor point
of view.79, 80
Single-particle tunneling between two superconductors can be
understood by similar diagrams, as shown in Figure 3-10. An
electron from k f in I can tunnel to either k ' f or k " f as before,
or an electron in k f can tunnel to the same final states, where
ek = —€k. Since vk2 and v 2 = u 2 are the probabilities that
(k \ , — k j ) and (k f , —k | ) are initially occupied, we see that
the total probability that an electron is available for tunneling is
u k2 + v k2 = Thus the coherence factors drop out in both the
1-
initial final states if sums are restricted to be above (or below)

the Fermi surface. Notice that the current begins at V = + Ar
in this case at T = 0 .
We turn now to the superfluid pair tunneling process proposed
by Josephson .83 He points out that in the absence of an applied
voltage a tunnel current can flow between two superconductors if
a superfluid pair is transferred from one side of the junction to the
other without creation of quasi-particles on either side. A simple
derivation of Josephson’s effect, due to Josephson 83 and to
Anderson 84 makes the physics of the situation quite clear. In

the absence of both an applied potential across the barrier and the
tunneling Hamiltonian, no energy is required to transfer v super
fluid pairs from one side of the barrier to the other. If the state
with v pairs transferred from left to right (relative to a standard
state) is denoted by <£v, the “ tight-binding approximation” for
the exact eigenstates in the presence of the tunneling operator is
wa = 2 e'“v (3-49)
V
The canonical momentum a plays the role of the wave number k in

band theory. Since H T can only transfer one electron between
the materials, the coupling between the 0 v’s is second order in
T and one finds for the energy shift due to H T of the eigenstates
— (W \W ) — 2 (3-50)
where H T{2) is the second-order tunneling Hamiltonian given
by
= Ht e ^ it 0Ht (3’51)
and
h J y = 4|<0v+1|^ r<2)|0v>| (3_52)
To find the current, note that the rate of transfer of pairs is
d<» IdEA J. ,
~ w " \ a f ) ■t <sm “> <3-53*>
where the average is taken in a wave-packet state formed from
the s,incomplete analogy with the tight-binding approach
to the one-electron theory of metals. In the absence of an
applied bias, the momentum ha (canonically conjugate to the
pair number v) is a constant of motion; however, for V ^ 0 one
has
d(hay
-V =2F (3-53b>
From (3-53a) and (3-53b) it follows that the rate at which electrons
flow across the barrier is
J (t) = = J 1 sin + «0 (3-54)
so that an alternating current of frequency 2V/h = 483.6 Mc/sec/
/ivolt is expected to flow for V ^ 0 , while for V = 0 , a steady
current is expected, according to (3-53a). The d-c effect has been
observed by Rowell and Anderson .85 As Josephson pointed out ,83
the current is sharply reduced when a magnetic field is applied
to the junction of such a strength that a multiple of the flux
quantum occurs in the junction. This effect has been observed
by Rowell. 86 The a-c effect has been observed by Shapiro .860
Tunneling experiments by Burstein and Taylor 81 show that
in many cases an excess current between two superconductors at
low reduced temperature begins at an applied bias Ax or Ar, that
F IG U R E 3-11 (a) and (b) Processes contributing to the two-particle

tunneling mechanism which leads to current onsets at Ax and Ar.
A p p licatio n s o f the Pairing T h e o ry 87
is, below the one-particle threshold d, 4- d r. Wilkins and the

author 82 have proposed a two-particle tunneling mechanism in
volving the superfluid electrons to explain the phenomenon.
Their processes, illustrated in Figure 3-1 la and b involve a second-
order matrix element in which a pair of superfluid electrons in I
is withdrawn without creating excitations in Z, and the electrons
tunnel to r, where they occupy quasi-particle states. This pro
cess is shown in 3 - 1 la and gives a current onset at Ar -f V. A
second process shown in 3-1 lb involves removing two superfluid
electrons from Z but leaving two extra quasi-particles behind.
The two electrons tunnel to r, where they recombine to enter the
superfluid, thereby creating no excitations in r. This process has
an onset at Al = V. The processes give a polarity-independent,
temperature-insensitive excess current, as observed. To fit the
magnitude of the observed currents one must assume the oxide
films are patchy, having a small fraction of thin regions so that the
fourth-order matrix element \T\^ which appears in the two-
particle tunneling rate does not lead to a drastic reduction of the
excess current relative to the single-particle rate which involves
\T\2. The oxide thicknesses and the ratio of the areas of thick
and thin regions required to obtain agreement with experiment
do not appear to be unreasonable, considering the imperfect
nature of the oxide films.
3-7 O T H E R A P P L IC A T IO N S O F T H E P A IR IN G
THEORY
In this chapter only a few of the simplest applications of the
pairing theory have been discussed. In Chapter 8 we shall discuss
the electromagnetic properties of superconductors, i.e., the Meiss
ner effect, the persistence of supercurrents, magnetic flux quanti
zation, etc., as well as the paramagnetic spin susceptibility and the
resultant Knight shift. Further support for the pairing theory
comes from the distinctly different effects magnetic and nonmag
netic impurities have on the energy gap. As Anderson showed,
nonmagnetic impurities do not smear the gap edge as one might
intuitively expect; on the contrary, nonmagnetic impurities remove
the crystalline anisotrophy of the gap and thereby effectively

sharpen the observed energy gap. In Anderson’s discussion the
time-reversal invariance of the impurity scattering potential plays
an essential role. Abrikosov and Gor’kov have discussed the
case of magnetic impurities. Owing to the lack of time-reversal
invariance, they find a broadening of the gap edge. As the
impurity concentration is increased to a critical value (~ 1 %) the
energy gap vanishes, although the density of states is smaller
than that in the normal phase near the Fermi surface. For a
small range of concentration above this critical value, the material
exhibits ‘‘gapless ’’ superconductivity. This effect was discovered
experimentally by Reif and Woolf in tunneling experiments.
The reader is referred to the literature 9 , 16 for a discussion of
numerous other applications which give further empirical support
for the theory, e.g., thermal conductivity, boundary effects and
small specimens, type II superconductors, etc. In general the
agreement between theory and experiment is remarkably good, as
mentioned above, considering the simplicity of the models used in
applying the basic ideas of the pairing theory.
CHAPTER 4
ELECTRON-ION SYSTEM
As we have seen above, many of the observed properties of super

conductors can be explained on the basis of a model in which an
attractive velocity-dependent potential acts between pairs of
quasi-particles of the normal metal. However, from the isotope
effect1 1 , 12 and from electron-tunneling current anomalies87, 88 we
know that the electron-phonon interaction plays an essential
role in bringing about superconductivity in most (if not all)
superconductors.
4-1 T H E E L E C T R O N - IO N H A M IL T O N IA N
To obtain a more complete understanding of superconduc
tivity, we should study the full electron-ion system and show how
the results of the simplified model discussed in Chapters 2 and 3
emerge from “ first principles.” In particular, one would hope
to be able to (a) explain the isotope effect; (b) determine why
certain metals are superconductors and others are not; (c) isolate
those parameters which determine the transition temperature;
(d) check the basic nature of the pairing interaction by explaining
the observed tunneling density-of-states anomalies. In addition,
one would like to account for the observed deviations from the
89
empirical “ law of corresponding states.” This “ law” states that

all superconductors have identical properties when these proper
ties are expressed in reduced units; e.g., the critical field curve
H C(T)/H 0 is a universal function of the reduced temperature T / T c,
etc. Although it is not possible to reach all of these goals at
present, the basic theoretical framework which is required to
treat manv-bodv effects in real metals is developing rapidly. It
is likely that the above questions will be resolved in the foreseeable
future.
In our discussion we concentrate on a simplified model of a
metal in which ions, with their core electrons rigidly attached,
interact with a sea of conduction electrons. Of course, the ions
interact among themselves, as do the conduction electrons. Thus
we assume that the core electrons adiabatically follow the vibrating
nuclei but are otherwise unexcited. While our approximation
neglects core polarization, this effect presumably has a minor
influence on the dynamics of the conduction electrons in most
superconductors because of the large energy required to excite
the cores. Core polarization may well play a more important
role in determining the ion-ion interaction; however, this effect
can be included approximately in the ion-ion potential.
For simplicity we shall also neglect spin-orbit interactions,
although as we shall see later they have been invoked as a possible
explanation for the nonvanishing Knight shift observed in the
superconducting state .8 9 , 90 The hyperfine and electron spin-
spin interactions will also be neglected for the moment; the former
were treated as a perturbation when we considered nuclear-spin
relaxation processes (see Chapter 3).
It is particularly convenient to omit for the moment the
magnetic interaction between conduction electrons due to their
orbital motion. This interaction is exceedingly important in
understanding superconductivity because it leads to the Meissner
effect. The reason that the omission is not damaging at the
outset is that in the absence of external fields (and/or net currents
in the material) the magnetic forces between electrons almost
exactly cancel each other and therefore lead to weak effects.
E le c tro n -io n System 91
When external fields and/or net currents exist there is a coherent

addition of the fluctuating magnetic fields between electrons; the
resultant field can be conveniently treated within a self-consistent
field approximation and fluctuations about the average field again
neglected. We shall adopt this view in treating the magnetic
properties of superconductors in Chapter 8 .
The Hamiltonian of the conduction electron-ion system is
then
5 .2 ,1 ^ + ? &
+ \ 2 W (Rv, Rv,) + 2 U(r„ Rv) (4-1)
V*V ' I, V
where r, is the position of the ^th conduction electron, and Rv the

position of the vth ion whose equilibrium position is at Rv°. The
index v labels both the cell n and the site a in this cell at which
B v° is located. For a crystal with one atom per unit cell the label
a is superfluous and the ions will be labeled by n. The first
two terms in (4-1) represent the kinetic energy of the conduction
electrons and the Coulomb interactions between them. The
third and fourth terms give the kinetic energy of the ions and
the ion-ion interaction. The last term in (4-1), representing the
conduction electron-ion interaction, is in general nondiagonal
in the electron coordinate representation because of exchange
interactions between rhe conduction and core electrons.
We work with a system of unit volume and use periodic
boundary conditions. Our basic approach is to cast this compli
cated many-body Hamiltonian into a form which will allow us to
use well-known techniques of quantum field theory. There, one
traditionally begins with a set of “ bare” particles and through
a consistent treatment of the coupling between these particles
one derives a set of “ dressed” particles in terms of which proper
ties of the physical system can be described. The richness of
various approximation procedures and the relative ease of cal
culation make this approach highly attractive. An introduction
to field-theoretic techniques in the many-body problem 91 is given

in Chapters 5 and 6 .
The “ bare” particles in our case will be of two types: single
electrons occupying Bloch states, and quantized vibrations of the
ionic lattice. The coupling between these excitations causes
drastic changes in the system which cannot be treated by elemen
tary perturbation theory. Nevertheless, the methods of quantum
field theory are sufficiently powerful to allow us to understand
the origin of the superconducting state and make detailed theo
retical predictions regarding its properties.
4-2 BARE P H O N O N S
We introduce bare-phonon coordinates Qq x of wavevector q
and polarization A which describe the deviations §RVof the ions
from their equilibrium positions Rv°. This is done by performing
the canonical transformation
R"“ = R"“° + (NCM CY >2 I <4-2>

The Q’s are normal coordinates of the vibrating ion system when
the ion-ion interaction is treated within the harmonic approxi
mation. This procedure is discussed in detail by Peierls.92 In
(4-2), N c is the number of unit cells per unit volume and M c the
total ionic mass within a unit cell. The polarization vectors
€g A(a) are determined by solving the above normal-mode problem.
It is convenient to normalize these vectors by requiring
2 l / a|€,,(«)| 2 = M c (4-3)
a
For one atom per unit cell, this reduces to \eqt A\ = 1. The
normalization condition and an orthogonality condition for the
e’s can be written as
2 = M c &Ar (4-4)
a
The wavevectors q are restricted to the firstBrillouin zone
(which contains Nc points). Since the density of states inq-space
of phonons with a given polarization is l/(277)3, we shall often

replace sums over q by integrals through the relation
(4-5)
The number a0 of independent polarizations (i.e., branches in
the phonon spectrum) is three times the number of ions per unit
cell, so that the total number of phonon modes is equal to the
number of degrees of freedom of the ionic lattice, as expected.
There also exists a completeness relation of the form
2 *qAa ) ■C-,A(«')e,(l'<Rn“0 R,,'“'0>= Snn' (4'6)
It is conventional to call the three branches, which in the
long wavelength limit have all ions in a unit cell moving in phase,
acoustic branches. The remaining branches are called optical.
If the wave vector q is along certain symmetry directions in
the crystal, the polarization vectors eq>k are either parallel (for
longitudinal phonons) or perpendicular (for transverse phonons)
to q. In general, the polarization vectors bear no simple relation
to q although one continues to speak of longitudinal and transverse
phonons, keeping the same designation of a given branch as q
moves away from a symmetry direction.
While the transverse acoustic modes tend to zero frequency as
q 0 , the longitudinal modes tend to the ionic-plasma frequency
Qp = (±ttNcz 2e21Mc)112 due to the long-range nature of the
Coulomb force .93 We know that in real solids the frequency of a
longitudinal acoustic sound wave is proportional to q and vanishes
as q -> 0 ; clearly, the difference comes from the conduction electrons
responding to the electric field set up by the ionic oscillations and
screening out the long-range force. In our scheme, the screening
is due to the electron-phonon and electron-electron interactions.
The main point is that, while the shift in the phonon frequency
is large (from Qp -+ ~ 0 ), the problem is simply treated by standard
field-theoretic techniques. Of course, simpler methods could be
used for this problem (for example, the Thomas-Fermi approxi
mation or a time-independent self-consistent field approach);
94 T h e o ry of Su p ercon d uctivity
however, these methods are not general enough to treat super

conductivity. By working out simple problems with the more
elaborate scheme we shall be better able to understand how to
approach the problem of superconductivity.
To complete the dynamics of the bare phonons, the momen
tum part of the canonical transformation (4-2) is given by
(M \1/2
P„a = y (4-7)
\ i y c/ q,A
where 17QAis the phonon momentum. It follows from the canoni
cal commutation relations for the ion variables,
[Pna,R n^ ] = ^ 8 nn- 8aa' I
%
(4-8a)
[Pna> PB.B.] = [Rbb, Rb.„.] = 0 (4-8b)
that the phonon variables also satisfy canonical commutation
relations
[HqA’ Qq'A'I = &qq' ÂA' V
(4-9a)
[77,„ 77,.,.] = [QqA, = 0 (4-9b)
In (4-8a), 1 is the unit tensor. As we shall see, the fact that
phonons are bosons has nothing to do with the spin of the indi
vidual ions. This is clear since we have treated the ions as being
distinguishable by localizing them near lattice sites. The Bose
character of a phonon simply reflects the quantum-mechanical
commutation rules applied to the individual ions.
With the aid of above relations it is straightforward to trans
form the ionic kinetic energy plus the ion-ion interaction (treated
to second order in SR) into phonon coordinates. One finds
2 v i r + \ 2 w* - ^ \ l
~ q , A
+
+ const- W Q 'S Q J
(4-10)
V
where Qq A is the normal-mode frequency and the constant term

is the energy of the ion system when the ions are located on
their equilibrium sites. We introduce bare-phonon creation and

destruction operators a q k + and a QtAby
(4-1 la)
and
(4-llb)
The formalism of second quantization, which deals with these
operators, is discussed in the Appendix. It follows from (4-9)
that the a ’s satisfy Bose commutation relations:
\P'qA'> ^q'A'
J ^ qq' ÂA' (4-12a)
KA> %'A'] = K a + ? Uq'A'+] = 0 (4-12b)
The bare-phonon Hamiltonian (4-10) then becomes
H ph = 2 M qA(Nqi + h) (4-13)
where N QtA = a q A+a q A is the phonon number operator and we
have dropped the constant term in (4-10). H ph will be one of the
two terms in our zeroth-order Hamiltonian for the system.
The anharmonic terms neglected in (4-10) presumably have
little effect on superconductivity since the volume change between
the A7- and ^-states is small.4 Also, since the transition tempera
ture T c is small compared to the Debye temperature, the amplitude
of the ionic vibration is expected to be small below T c, as our
approximation requires.
4-3 BARE ELEC T R O N S
We would like to introduce a set of one-electron eigenstates
Xk to describe the bare conduction electrons; however, a difficulty
arises. If we introduce a one-body potential U0 and require that
Xk satisfies the Schrodinger equation
(4-14)
then the states Xk are n°t in general orthogonal to the core states.
Even if we arrange U0 so that the “ conduction band” solutions of
(4-14) are orthogonal to the core states when the ions are in their
equilibrium positions, the orthogonality is not maintained when
the ions vibrate. A partial solution to this difficulty has been
given by Wilkins,94 using the pseudo-potential method of Kleinman
and Phillips. This work was generalized to the many-body
problem in the work of Bassani, Robinson, Goodman, and the
author , 95 who treat the case of rigid cores described within the one-
electron approximation. We shall not discuss this treatment here
because of the mathematical complications necessary to carry
through the analysis. It suffices to say that an auxiliary wave
field describing the conduction electrons can be introduced in
such a manner that the conduction and core states are properly
orthogonal even if the cores vibrate. The equations of motion
of this auxiliary wave field are the same as for the original wave
field except for a redefinition of the potentials involved. Since
these potentials are difficult to estimate from first principles at
present, we shall simply disregard the above complication and
proceed using the one-electron states (4-14) as the bare conduction
electron states.
To make the states Xk precise, we must define U0. In order
that the electrons are not scattered by the lattice when no phonons
are present, UQshould include the electron-ion interaction with the
ions fixed on their equilibrium positions. Since the ion system
has a large positive charge, this leads to a very large negative
potential acting on a conduction electron. Since the Coulomb
interactions with the remaining conduction electrons cancel most
of this interaction, we include in U0 the potential due to the re
maining conduction electrons occupying a standard configuration.
This configuration could be a uniform distribution of electronic
charge or the distribution given by treating the conduction elec
trons within the Hartree-Fock approximation. Of course the
better one does in choosing U0, the less there is to take into account
as coupling between the bare particles. In any event, U0 should
be chosen to have the periodicity of the lattice (although it may not
be diagonal in the coordinate representation) so that Bloch’s
theorem 96 holds
Xk(r + a) = eikaxte(r) (4-15)
E le c tro n -lo n System 97
where a is any translation under which the crystal lattice is in

variant. This relation defines the “ crystal momentum” hk of
the state in question. While Xk an eigenfunction of crystal
momentum, it is not in general an eigenfunction of physical
momentum (i.e., a plane wave), since it can be represented as a
linear combination of plane waves of wave vectors k -f K n. The
vectors K n are reciprocal lattice vectors defined by
Kn • a = 27r x integer (4-16)
where a is an arbitrary allowed translation of the lattice. These
reciprocal lattice vectors play an important role in the theory
of solids since crystal momentum is conserved mod h K n in dynami
cal processes, and it is crystal rather than physical momentum we
shall usually encounter.
While the curves of e versus k are fairly simple for alkali
metals, the situation becomes quite involved for polyvalent
metals. The reader is referred to Ziman’s book 96 for a discussion
of theoretical and experimental results in this rapidly developing
field. The over-all picture is greatly simplified if the energy states
are represented in an extended zone scheme as one does for a
free electron, rather than folding the curves back into the first
Brillouin zone. In many cases one finds that a large fraction of
the energy surfaces resemble the free-electron case except for
discontinuities of the energy as one crosses zone boundaries.
Since a major part of our calculations will emphasize states near
the Fermi surface and since in general a small fraction of the Fermi
surface which is effective lies near zone boundaries, one hopes
that an effective mass approximation will adequately represent
the gross features of the band structure. We shall often make this
approximation to simplify the mathematics although the approxi
mation is not essential.
We shall use the formalism of second quantization to treat
the conduction electrons; this scheme is reviewed in the Appendix.
The creation and destruction operators for an electron in state
k with z-component of spin s are defined to be ck s+ and cfc s,
98 T h e o ry o f Sup ercon d uctivity
respectively. These operators satisfy Fermi anticommutation

relations:
^k's' } = &kk' ^ss' (4-17a)
K s, Ck,8.} = {cks +, cfcV+} = 0 (4-17b)
The bare electron Hamiltonian is then
H el = 2
k, S
€kn*s (4-1S)
where N ks = cks+cks is the electron number operatorfor state
k, s. We work in an extended zone scheme with the k-sum
running over all but the core states.
The spin-orbit interaction in the periodic lattice could be
included in (4-18) by working with spin-orbitals rather than the
orbital functions The electron-phonon interaction would
then contain spin-flip terms. We shall neglect these complications
for the moment.
Thus, the total zero-order Hamiltonian is the sum of the bare-
particle energies,
Ho = 2 ^ +2 + i) (4-19)
k. s q, X
4-4 B A R E E L E C T R O N - P H O N O N IN T E R A C T IO N
We included in H 0 the interaction of the electrons with the
ions in their equilibrium positions. The difference between this
potential and the full electron-ion potential remains as a pertur
bation (along with several other terms). It turns out that it is
sufficient for most purposes to expand this difference in powers of
the ionic displacements 8 RV, and retain only the leading term.
Thus, the bare electron-phonon interaction is of the conventional
form for a boson-fermion coupling, that is, linear in the boson
field, bilinear in the fermion field.
A reliable first-principles calculation of the coupling is not
possible at present for most superconductors, since one requires
accurate one-electron wave functions as well as reliable ionic
potentials. Unfortunately, the most reliable calculations of the

interaction are for small-momentum transfers; the phase space
for these processes is small and they contribute little in bringing
about superconductivity. It is rather the large-momentum
transfers (of order the Fermi momentum hkF) which are important,
but their coupling is difficult to estimate accurately since, in
this case, one is sampling short-range details of the core potential
rather than the Coulomb tail. In addition, details of the Bloch
functions near the cores become important for these processes
and it is in this vicinity that a free-electron approximation for Xk
is poorest. In practice one can carry through calculations without
specifying details of the electron-phonon matrix elements and
replace certain averages of these matrix elements by parameters
to be determined from electrical resistivity, thermal conductivity,
superconducting transition temperature, etc.
We can make a few general statements about the coupling.
The perturbing potential acting on the ith electron is
Z[U„ - uiv°] = - 2 § r v- vtu„
v v (4-20)
(NcM cy>* 2 2 ' V i C7ive,<‘ -R»
where we have used (4-2) and have denoted U(rt, Rv) by UltV as
well as U (rp Rv°) by U%tv°. For given values of q and Athe matrix
element of this potential between bare electron states k and k' is
- ( N M
Q c) v 2 2 <*'l V . t g * ) • e » e * - * . 0 (4-21)
If we introduce a cell location Rn° and the relative position

pa° such that
R n a ° = R n° + 9a° (4 -2 2 )
we can reduce the matrix element to

( N \1/2 2 <*'l W « l* > • 2 ^ -k .a +K„
c/ a Kn
4 23 )
( -
In the reduction we used Bloch’s theorem (4 -1 5 ) plus the fact that

Uu v depends only on the relative separation between the electron
and ion. In (4 -2 3 ) Uu a refers to ions in the unit cell located at
the origin, i.e., at R n° = 0 , and K n is a reciprocal lattice vector.
We see from (4 -2 3 ) that the electron-phonon interaction conserves
momentum mod MC, as mentioned above. The number of scripts
in (4 -2 3 ) is somewhat confusing so that we introduce the
abbreviation
( hN \ 1/2
(4' 24>
Since q isrestricted to the first zone we use the convention that
if the momentum transfer k ' — k falls outside the first zone we
shall use the corresponding reduced wave vector for q and suppress
K. We note that
Qk'kA= Qkk'iI* (4 -2 5 )
Going over to the second-quantization language for electrons, the

bare electron-phonon interaction becomes
^ e l-p h =
k,k2',s,AQkk'tSPk'-k . Ack's+cks (4 -2 6 )
where the phonon field amplitude is defined by

9 V = aQA+ (4 -2 7 )
As we mentioned above, despite a large amount of good work, our

first-principles understanding of the matrix elements gk Ais in
a rough state at present. Ziman 96 gives a detailed discussion of
all but the most recent work.
An oversimplified but useful model of a solid consists of
smearing the ions out into a continuous charged “ jelly” ; in the
absence of vibrations the jelly is taken to be uniform so that the
Bloch functions degenerate into simple plane waves. The bare
phonons are quantized vibrations of the jelly. In this “ jellium”
model, the bare electron-phonon interaction is easily calculated
and one finds for longitudinal phonons 93
.477C2 / h Z 2N c\112
where Zce is the total ionic charge per unit cell and Qp =
(477-ArcZc2e2/M c)112 is the ionic plasma frequency. In the long
wavelength limit (q —> 0 ), where the model is presumably reason
able, the coupling of the electrons to bare longitudinal phonons is
singular. The singularity is clearly due to the long-range Cou
lomb force. When screening is taken into account, the dressed
interaction vanishes as q -> 0 , again showing the importance of
screening in metals. We note the relation
2g 2 4:7re2
(4-29)
hOP = ~Y~
holds for longitudinal phonons in jellium. This relation has
significance in superconductivity since the left-hand side turns
out to be related to the bare electron-electron interaction resulting
from the exchange of virtual phonons. The equality (4-29) states
that the phonon attraction is exactly cancelled by the Coulomb re
pulsion if the electrons scatter without changing their energy (i.e.,
the net interaction vanishes in the static limit). This result is
peculiar to jellium; however, the relative scale of the phonon and
Coulomb interactions in real metals is roughly set by (4-29).
The jellium model has the added simplicity that the electrons
and transverse phonons are uncoupled since their interaction is
proportional to q • e qAi which vanishes in this case.
Actual metals are considerably more complicated than
jellium since transverse phonons play a strong role in umklapp
processes96 (transitions in which the momentum transfer k' — k
lies outside of the first Brillouin zone). Transverse phonons can
also enter normal (non-umklapp) processes if the electronic energy
contours in k-space are not spherical or if q is not in a symmetry
direction. Also, there is an electromagnetic coupling between
the transverse phonons and the electrons .67 In addition, the
bare longitudinal matrix elements are certainly more complicated
than (4-28) since they will reflect crystalline anisotropy as well as
details of the core potential and the behavior of the Bloch functions
near the cores. The reader is referred to Ziman’s book for further
details about the electron-phonon interaction in real metals. It
appears that the orthogonalized plane wave method will prove to

be very useful in gaining insight into this difficult problem.94*97*98
4-5 T H E E L E C T R O N - P H O N O N H A M IL T O N IA N
To complete the program of expressing the electron-ion
Hamiltonian (4-1) in terms of bare electron and phonon operators
we must include the Coulomb interaction between conduction
electrons. As shown in the Appendix, this can be expressed in
second-quantization language as
tfei-ei = £ ki 2 s' <k 3 >k*l V|ki, K ) c k.jS +cICtS, +cIC2S.ckiS (4-30)
••• /C4 , S,
where the Coulomb matrix element is given by

<k3, k4| F|kx, k2> = J x te3*(r)xfe4*(r') jf f,| yfcl ( - W r') d3r d3r’
*_
(4-31)
It follows from Bloch’s theorem (4-15) that the Coulomb inter
action conserves crystal-momentum mod reciprocal-lattice vector
K; thus the matrix element vanishes unless
ki + k 2 = k 3 + k 4 + K (4-32)
The final term in H is the difference between the one-body potential
UQintroduced in defining the Bloch functions [see (4-14)], and the
interaction Ux v° of the electrons with the ions fixed on their
equilibrium positions. This contribution is
H f j = 2 <k + K|&|k>ck +K.s +c,s (4-33)
k, K, s
where
U = 2 Utv° - U0 (4-34)
V
The full electron-phonon Hamiltonian is then

H = H el + H ph + tfel_ph + H el. el + H d (4-35)
The system is a complicated one from a field-theoretic point of
view since it involves the interaction of a Bose field with a self
coupled Fermi field, and, as we shall see, the coupling constants
are not small. In the next chapter we shall discuss how field-
theoretic methods can be applied to this system.
CHAPTER 5
FIELD-THEORETIC
METHODS IN THE
MANY-BODY PROBLEM
We shall introduce field-theoretic methods in the many-body

problem by discussing three well-known “ pictures” or represen
tations used in discussing quantum mechanical problems.
5-1 T H E S C H R O D IN G E R , H E ISE N B E R G , A N D
IN T E R A C T IO N P IC T U R E S
In elementary discussions of quantum mechanics one usually
works in the “ Schrodinger picture” in which the dynamical
variables are taken to be time independent so that the wave
function contains the time dependence of the problem. In this
picture, the wave function ¥$(£) satisfies
= H{t^ {t) {5'l)
While the dynamical variables are independent of time, the
Hamiltonian may contain an explicit time dependence because
103
of an external field acting on the system; (5-1) includes this possi

bility. If we consider an isolated system, H is independent of
time and an exact solution of (5-1) is given by
Ws(t) = > ^ 0) (5-2)
It is convenient for many purposes to make a unitary transforma
tion to the “ Heisenberg picture” in which the wave function WH
is time independent and the time dependence of the problem is
transferred to the operators. If we choose phases so that the
wave functions Ws and WHare identical at a time t0, these functions
are related by
WH(t) = WH = e imt- to>i*'Fs (t) (5-3)
Thus the unitary operator factors out the time dependence of *FS.
In order that all observable quantities (i.e., matrix elements) be
unaltered by the transformation, the operators 0 in the two
pictures must be related by
QH(t) = eiHU-t0)inQsê-imt-t0)in ( . )
5 4
The Hamiltonian H(p, q) has the same form in each picture

although the time dependence of the p ’s and q’s which express H
differ according to (5-4). From (5-4) it follows that the time
dependence of a Heisenberg operator is given by
ih d^ = [0h(0, H] +
ir i h ( 5 ' 5 )
where the partial derivative accounts for any time dependence of

the operator 0 S(O-
For the purpose of perturbation expansions it is often con
venient to write the Hamiltonian as
H = H + H' 0 (5-6)
and define an interaction picture through the relations
Wj(t) = (5-7a)
and
Qj(t) = e*"o<«-«o)/»0 s(«)e-iHo^-«o)/» (5-7b)
Field-Theoretic M e th o d s in the M a n y -B o d y P roblem 105
If H' vanishes, the Heisenberg and interaction pictures are

identical. In the presence of H\ the time dependence of the
operator 0 ; is given by the zero-order Hamiltonian H 0, and the
wave function Yj is time-dependent solely due to the perturbation
H\ Weshall choose H so that the time dependence of the
0
operators 0 7 is very simple; this allows one to construct a simple

set of rules for treating H ' in a perturbation series corresponding
to Feynman diagrams.
By inserting (5-7a) into Schrodinger’s equation (5-1) we find
ih ^ P - =
ct
(5-8)
where H / is the perturbing Hamiltonian expressed in the inter
action picture,
H /(t) = eM oU -toM H s'e-'W -W * (5-9)
From now on we shall work with units such that % = 1 .
5-2 T H E G R E E N ’S F U N C T IO N A P P R O A C H
In the many-body problem we are ultimately interested in pre
dicting such quantities as the thermodynamic and mechanical
properties of the system as well as nonequilibrium properties such
as electrical and thermal conductivities, and absorption of quanta
of external fields. It is clear that we cannot determine these
quantities by solving for the exact many-body eigenfunctions
except for extremely simple systems; even if these functions were
available they would be hopelessly complicated unless expressed
in a form suitable for calculating a particular property of the system.
One would prefer to work with dynamical quantities which are
more closely related to experiment and contain less information
than the full wave functions. One would then approximate
these dynamical quantities directly rather than working with the
Y 9*. Quantities which satisfy these conditions are the Green’s
functions of quantum field theory .9 1 , 99 The one-electron Green’s
function gives information about the spin and charge densities
and the momentum distribution of the electrons as well as in

formation about the excitation spectrum of the system. When
the system consists only of fermions interacting via an instanta
neous two-body potential, the one-particle Green’s function also
suffices to determine the ground-state energy (or more generally
the free energy) of the system . 100 The electrical conductivity,
magnetic susceptibility, and many other nonequilibrium properties
can be obtained from the two-particle Green’s function.
Perhaps the greatest advantage of the Green’s function
approach is that approximation schemes are readily developed
which allow one to use physical insight in solving many-body
problems. By studying simple problems we shall learn how to
transcribe our physical ideas into field-theoretic language.
Alternatively, mathematical approximations can often be better
understood by reversing this process. The interplay of these two
possibilities will allow us to go a long way in understanding complex
systems.
We consider for the moment a system of fermions of spin \
interacting via a two-body spin-independent potential. The
one-particle Green’s function G is defined by
where | 0 > represents the exact ground state of the interacting

system expressed in the Heisenberg picture and the field operators
i/ js and 0S+ (see the Appendix) are also expressed in the Heisenberg
picture. The time-ordering symbol T is defined to order
chronologically the operators inside the braces so that operators
with earlier times are placed to the right with minus signs arising
from anticommutation of the operators included. Thus
We note that G depends on time only through r = t - t2. If1
the system is translationally invariant, G depends only on the

relative coordinates r = rx — r 2 and r = tx — t2:
£ s(r 1 , G; r 2 >t2) = Gs(r, t) (5-12)
Field-Theoretic M e th o d s in the M an y -B o d y P roblem 107
(The latter simplification does not hold for electrons in solids

because of the crystal lattice.) The Fourier transform of G(r, r),
defined by
Gs(V,Po) = j*e“ i<p'r_poI,Grs(r, r) d 3r dr (5-13)
is particularly useful since as we shall see later itspoles as a
function of p are related to the elementary excitation spectrum
0
of the system. The Fourier inverse of (5-13) is

<?s(r> *) = J e i(pr- po^gs(p, p 0) (5-14)
We are again using periodic boundary conditions and a box of unit
volume so that we may use the relation
To save writing we use the abbreviations

x = (r, t)
V = (P> Po)
p x = p • r - p 0r
d*x = d3r dr
d*p = d3p dp 0
The Fourier transforms (5-13) and (5-14) then become
Gs(p) = j e - ' pxGs(x) d*x (5-13')
and
Gs(x) = j e ' * xGs( p ) (5-14')
For a translationally invariant system one often works with the
spatial Fourier transform of G(r, t ),
T) = j e ~ ' p'rOs(r, r) d3r (5-15)
By using the relation (see the Appendix)
0 .(r, 0 = 2
P
Cpse‘P'r (5‘16)
where cp s destroys an electron of spin s in plane wave state p,

one readily establishes
Gs(p, r) = -i<0|T{cps(r)CpS+ (0)}|0> (5-17)
5-3 T H E FREE FERM I G A S
To illustrate the structure of G we study a system of non
interacting fermions of spin contained in our box of unit
volume. In the ground state | 0 >, all plane wave states with
momentum less than the Fermi momentum p F are occupied with
an up and a down spin electron, while all other states are empty.
The Hamiltonian of the system is
H = 2 ePcP* +cPs (5-18)
P. S
where ep = p 2 / 2 m. Since the system is symmetric in the spin

variable, we shall concentrate on the up-spin electrons and sup
press s. From the anticommutation relations (see the Appendix)
{ C P S > CP ' S ' + } = S p p ' ^SS' /* 1Q\
{cp„ c p.,} = {cpa\ c p., +} = 0 j
it follows that in the Heisenberg picture one has
cP(r) = cp(0)e-‘V (5-20)
With the definition (3-17) for G( p, r ) we find
r(n ^ - f - * ' < 0 | c pc p + | 0 > e - ^ r > 0
G(P’ T )- \ i<0 |cp*cp| 0 > e -V r < 0 (" “ !)
With the aid of (5-19) this can be written as
T>° (5-22)
I >/»«"'■'
where the Fermi function f p is one forp p F.
In a nontrivial many-body problem, the time Fourier trans
form of G is of great interest since as we mentioned its singularities
determine the elementary excitation spectrum of the system.
For the free Fermi gas one finds
G ( p , p 0) = G(p) = (5-23)
P 0 €p ' ly]p
Field-Theoretic M e th o d s in the M a n y -B o d y Problem 109
where rjp is a positive (negative) infinitesimal for p greater (less)

than p F. S is a positive infinitesimal. Note that G(p) has a pole
at p = €p, the energy required to add an electron of momentum
0
p to the system. To establish (5-23) we consider the inverse

Fourier transform.
r*oo j e lpo dp
G(p, r ) =
"t"
6
2?t
-P (5-24)
0
J~< Po irjp
The integral is evaluated by using Cauchy’s theorem. For r ^ 0
the contour can be closed around an infinite semicircle in the
upper half of the complex p 0-plane, as shown in Figure 5-1,
since e ~ ipo(l~d) vanishes on this added piece of contour. For
r > 0 the contour may be closed in the lower half-plane for similar
reasons. Because of the factor irjp, the pole is in the upper half
plane for p p F. Thus,
for r > 0 we obtain
G (p . t ) =
0 P < Pf
-ie ~ i€pl p > pF
while for t ^ Owe find
l€pT p Pf
t
as required by (5-22). While the factor e ipo in (5-23) can be6
replaced by unity for most purposes and hence we suppress it,

there are cases where we shall need this factor (an example is in
calculating the exchange energy within the Hartree-Fock
approximation).
F IG U R E 5-1 Integration contour for evaluating (5-24) when r < 0.

I 10 T h e o ry of S u p ercon d u ctivity
Another method to obtain G(p, p 0) for the free Fermi gas is

to construct an equation of motion for 6r(p, t ). This is easily
done by writing (5-17) as
£(p, r) = - ; < O | cp ( t )cp + (O)|O>0( t ) + i<O|cp+ (O)cp(T)|O>0(-T)
By differentiating (5-25) with respect to r we find
the delta-function terms coming from differentiating the - 8
functions. With the aid of the anticommutation relations (5-19)

and the expression (5-20), it follows that 6r(p, r) satisfies the
differential equation
(5-27)
It is clear from this equation why G is called a “ Green’s function” :
it is in fact the Green’s function for the operator (id/dr — ep).
We can solve (5-27) by taking its Fourier transform with respect
to r; then
(5-28)
The essential difference between this expression and the correct
result (5-23) is that (5-28) does not specify how one is to treat the
singularity at p = ep. The difficulty is that (5-27) is a first-
0
order differential equation in r and we must specify a boundary

condition in solving the equation. The most general solution of
(5-27) can be written as the sum of a particular solution
T> 0 (5-29)
T< 0
plus the general solution of the homogeneous equation
HP, t ) = ifPe~'(”T (for all r)
Field-Theoretic M e th o d s in the M a n y -B o d y Problem
The arbitrary constant f p is determined, for example, from the

boundary condition G(p, 0 ") = i f p. With this choice of f p, the
Fourier transform of the full solution P + h agrees with (5-23), as
required. Thus the boundary conditions uniquely specify how
we are to treat the singularity of G(p).
At this point the reader may wonder how G is to be determined
for the interacting system, since in the derivation (5-20) to (5-23)
for the free system, we explicitly used the ground-state wave
function; yet we argued that the Green’s functions free us from
considering the many-body wave function. There are two
principal methods for determining G (and higher-order Green’s
functions). The first is to construct an equation of motion for G
analogous to (5-27); in the presence of a two-body interaction an
extra term appears on the right-hand side of this equation involv
ing the “ two-particle” Green’s function, defined in the x-represen-
tation by
G (xx, x2i x3, x j = ( ~ i ) \ o m i x ^ x ^ i x ^ i x ^ l O }
2 (5-30)
We have suppressed spin variables. Since G is also unknown, we
2
write an equation of motion for it and find the three-particle

Green’s function appearing. In this way one can generate a
hierarchy of equations which unfortunately does not terminate.
The chain is often truncated by neglecting correlations between
more than two or three particles at a time. In principle, one can
then solve the coupled equations for the functions that are
retained. While the scheme sounds straightforward, the mathe
matical problem is exceedingly complicated, even if one neglects
all but two-body correlations. One is usually satisfied by re
taining only pieces of the two-body correlations. This scheme
has been extensively discussed by Schwinger, Martin, Kadanoff,
etc . 9 9 , 10 1
The second method is the perturbation expansion of Feyn
man, Tomonaga, and Dyson , 102 familiar in quantum electro
dynamics. While the most naive version of this scheme is
formally limited to problems where the expansion converges, this
approach is so closely related to one’s physical intuition that partial
summations of the diverging series can often be found which lead

to physically meaningful results. The discussion below follows
the second line of approach.
5-4 S P E C T R A L R E P R E S E N T A T IO N O F G(p, r)
The one-particle Green’s function G for the interacting Fermi
gas is closely related to G for the noninteracting system. To see
this, consider the Green’s function for spin-up particles:
<?(P, t ) = -t'< 0 |T { Cp(r)cp + (0)}|0>
—t<0|cp(0)e ~ iH,c„ + (0) 10>ei£°nt r > 0 (5-31)
i<0|cp + (0)eiHtcp(0 )|0 > e-i£o"' T sj 0
where E 0n is the ground-state energy of the interacting n-particle
system. By inserting between c and c+ a complete set of eigen
states |!Fmn±1> of H for the n ± 1 particle system, we find
H 2 l(«P +)m.o|2« -i(£"n+1- £»")l T> 0
0 ( P, r) = m (5-32)
i ' 2 \ ( c p ) m . o \ 2 e i < E m" ~ 1 ~ E ° n ) l T<°
The matrix elements are defined by
(cP+L.o = < ^ mn +1 |Cp+lôn> (5-33a)
K ) m.o = <Vm" - 1 |cp|'F0") (5-33b)
that is, matrix elements of the operators for bare electrons taken
between exact eigenstates of the full Hamiltonian. For con
venience we introduce the energies o>mn ± 1 defined by
£ m" +1 - E 0n = +1 (5-34a)
and
E J - E 0* = —o>mn_ 1 - ^
- 1 (5-34b)
where /zn = E 0n+ — E 0n is the chemical potential of the n-
1
particle system. Since we are concerned with large systems we

have fin ^ = p. The energy tomn + 1 is necessarily non-
negative while the extra minus signs introduced for convenience

in (5-34b) require that o>mn - 1 be nonpositive. By inserting these
expressions into (5-32) we obtain
f - ; 2 K O m.oi2e - i(“" +w)t r >0
G(p, t ) = < m (5-35)
2e - i ( c o m n ~ 1 + u ) i T ^ 0
2 I 0
An important quantity is the spectral weight function A( p, oo),

defined by
^ (P . w ) = 2m K CP + ) m . o | 2 S ( " ~ < V + 1)
+ 2 Kcp)m.o|2 S(“ - c o /- 1) (5-36)

m
The one-pafticle Green’s function for the interacting system can

then be expressed as
nt \ f°° A(p, w) doj
0 (V > P o )= \
J - Po — W — (JL + lex)■
00 0* (5‘37)
where is a positive infinitesimal. To ensure the correct defini
8
tion of G for r = 0 , the factor eipo should multiply the right-hand

6
side of (5-37) as in (5-23). The spectral representation (5-37)

for G(p, p 0) can be checked by calculating its Fourier inverse
G(p, r) and noticing that the result is identical with (5-32). We
see from (5-37) that G(p, p 0) is simply a weighted sum of Green’s
functions (5-23) for the noninteracting system.
For the free Fermi gas G is given by taking A to be a delta
function,
A( p, oj) = (1- f p) S(o, - [ep - ,i]) + f p 8 (0 , - [ep - /x])
= S(o, - [ep - /a]) (5-38)
since for p > p F the first term when inserted into (5-37) gives
l/(Po ~~ €p + and fhe second term vanishes; for p < p F the
second term gives 1 l ( p Q — ep — iS) while the first term vanishes,
in agreement with (5-23).
I 14 T h e o ry o f S u p ercon d u ctivity
According to (5-36) the spectral weight function is a positive

real quantity
A(p, ) = A*(p,
co (jo) (5-39)
and satisfies the sum rule
(5-40)
The latter follows by inserting the expression (5-36) under the
integral
m m
and using the completeness of the states |^ mn±1) to write this as
— 00
A(p, w) dw = <0 |cpcp + + cp +cp| 0 > = 1
where we have used the anticommutation relation (5-19) in the

last step. We note that the free gas expression (5-38) trivially
satisfies the sum rule.
From the definition of A it is clear that its positive-frequency
part (co > 0 ) contains information about processes involving the
addition of an electron to the system while its negative frequency
part is related to hole-injection (electron-extraction) processes.
Thus the sum rule connects these two types of processes.
As we mentioned above, one usually sets up equations to
determine G directly rather than first finding A. A simple
relation between the two functions is given by taking the imaginary
part of the spectral representation (5-37):
CO > 0 (5-41)
CO < 0
where we have used the well-known relation for factors occurring
under an integral
x ± ir) x
(5-42)
Field-T heoretic M e th o d s in the M a n y -B o d y Problem 115
P denotes the principal part of the singularity is to be taken in

performing the integral. A dispersion relation connecting the
real and imaginary parts of G follows by inserting (5-41) into the
spectral representation,
Re G(p, p 0) - - P P Im £(p, p 0') dpQ'
” Ju Vo “ Vo
Im G(p, p 0') dp0' (5-43)
+
Vo ~ Vo
5-5 A N A L Y T IC P R O P ER T IES O F G
We can neglect the infinitesimal ia> in the spectral represen

8
tation (5-37) if we allow the co-integration to be along the contour

(7, shown in Figure 5-2; hence
f A(p,aj)daj (5-44)
<?(P> Vo)Jc 7---------:------
=
(Vo ~~ p) ~ a
Thus far, (7(p, p 0) has been defined only for real values of p 0 — fi.
We now extend the definition by writing
r ^(p> du)
(5-45)
G(V,Vo)
Jc (Vo ~~ ~
F IG U R E 5-2 Integration contour for the spectral representation of

G(V> Vo)•
as a function of the complex variable p Integrals of this

0 . 1 0 0
type define functions of p which are analytic over the entire

0
plane except possibly for values of ( p — p) lying on the contour

0
C . In our case the contour C cuts the plane into two regions so
that two different functions / : and f u can be defined. As shown
in Figure 5-2, f^Po - p) is analytic in region I and f u ( p - p) 0
is analytic in region II. For real p 0, G(p, p 0) coincides with

fi(Po - p) for p - n > 0 and with f u ( p - fx) for p - p < 0.
0 0 0
The function / x can be defined in region II by its analytic contin

uation across the cut. This continued function is not in general
analytic in region II. Similarly, the analytic continuation of / n
into region I is not analytic in region I. The functions / x and f u
will play an important role in later calculations.
Since A( p, co) is real, it is seen from (5-45) that at points
lying infinitely close to the contour, / x in region I and f n in region
II are complex conjugates. This property will also be useful
later in our discussion.
5-6 P H Y S IC A L IN T E R P R E T A T IO N O F G(p, p 0)
There are many ways of interpreting G. If we w ork in the

^-representation, and let x = x , we have
1 2
Gs(r, t ; r, t) = »'<0 |</<s + (r, <)<As(r, <)|0> = *'<0|ps(r, <)|0 > (5-46)
In this limit, —iG gives the expected value of the density of

particles writh spin s. For a translationally invariant system,
—i G (p , r = 0") gives the momentum distribution of the bare
particles, since
Gs(p, 0 ") = ;< 0 |cps +cps| 0 > = i<0|nps|0> (5-47)
Since G is a function of r, it contains more information than
is given by (5-46) and (5-47), which only use its value at r = 0".
G also gives the probability amplitude for finding the system in
the same state at time r > 0 as it wras in at r = 0+. This can
be shown as followrs. We choose phases so that the Heisenberg
and Schrodinger pictures are identical at t = 0 . For the moment
we speak in the Schrodinger language. Let us create at t = 0 a
particle in the bare state p. The wave function for the system is
then
cP+ | ^ iO(0 )>
if ¥ 5 o represents the ground state. At a later time r the wave
function has evolved into
e~iHzcp +\ 0(0)>
If the particle were created in state p at t — r rather than t = 0 ,
the wave function would be
cp+e - ‘"' | ^ , o(0 )>
Thus, the probability amplitude for finding the system at t = r > 0
in the same state as it was in at t = 0 is the scalar product of the
two Schrodinger states at time r:
< ^ o ( 0 ) | c ,HV ’ ' % + l ^ o ( 0 ) >
which in the Heisenberg picture becomes
<0| cp ( t )cp + (0)|0> = -t<?(p,T) (r > 0)
On physical grounds we might expect this probability amplitude
to oscillate with some frequency (Ep + /x) and decrease in magni
tude with a decay rate |-Tp|:
<?(p, t) e~ i(Ep +u)re~ |rp|T
oc
As is apparent from (5-22), this form is exact for the free Fermi
gas with E p + /x = ep and r p = 0 .
Let us now investigate the behavior of G(p, r) for large
positive times. By taking the Fourier inverse of the spectral
representation (5-37) one finds for r > 0
/•00
<?(P> T) = - i J A(p, w) e- i(a)+li)z doj (5-48)
While A( p, co) is a sum of delta functions for a finite system, it is
a continuous function in the limit of a large system. We carry
out this integral by pushing the contour into the lower half-plane
as shown in Figure 5-3 and we use the analytic continuation of
A in this region. If the continuation of A is analytic in the lower
co-plane
F IG U R E 5-3 Integration contour for determining G{pt t) for large

positive t.
half-plane except for a simple pole at a> = E p — i\Tp\ with

residue rp the integral can be written as the sum of the pole
contribution
— -nrve~ hbp +«>le - |rpil
2
plus the line integral

- t oo
where the path of integration is along the negative imaginary

axis. For times r such that E p t » 1 but e ~ | rp|T > 1 / E p t , the
pole term dominates and we have the simple result guessed
above. If we interpret the pole contribution in terms of the
contribution of a quasi-stationary state called a quasi-particle,
we see that the poles of the analytic continuation of 4(p, co) into
the lower half-plane gives the energy and damping rate of these
quasi-particles . 1 0 0 , 101
It is often convenient to express this result in terms of the
Green’s function itself, rather than in terms of ^4(p, to). Consider
^(P* Po) f°r real values of p which are greater than p. The
0
analytic continuation of this function into the complex j90-plane

^(P» Po ) is given by
0 ( p , p 0) = G ( p ,p 0) Im p 0 > 0 ( 5 49a)
= G(p,;>0) - 2TTiA{p, p 0 - fi) Im j)0 < 0
Since G(p, p 0) has singularities only along the real axis [see (5-45)],
Field-T heoretic M e th o d s in the M a n y -B o d y P roblem 119
the poles of A(p, p — p) in the lower half-plane coincide with those

0
of the analytic continuation of Cr(p, p Q) for p > p. Thus, if 0
there exists only a simple pole of the analytic continuation of

G(p, p 0) into the lower half-plane for p — p > 0 , this pole may
0
be interpreted as giving the energy and decay rate of a quasi

particle. If we assume there are several poles in the lower half
plane, the pole nearest the axis will presumably give the dominant
contribution for large r. The pole contribution may also be
interpreted as stating that if a bare particle is created at time
t = 0 , it has a probability amplitude n rp of being in the quasi
particle state of momentum p.
In a completely analogous way one can show that the poles
of the analytic continuation of ^f(p, a>) into the upper half-plane
for at < 0 may be interpreted as giving the energy and damping
rate of quasi-holes. Again, it is convenient to express this result
in terms of G rather than A. Consider G(p, p 0) for real values
of p which are less than p. The analytic continuation of this
0
function into the complex p 0-plane 6r(p, p 0) is given by

G(f,Po) = G(P,Po) + n iA ( p ,p — n) 2 Im p 0 > 0
0
= G (p ,p 0) I m p , < 0 ( ' b)
Since (5(p, p 0) has poles only along the real axis, the poles of A
and G in the upper half-plane coincide. Therefore, the poles of
the analytic continuation of 6r(p, p 0) (for real p < p) into the 0
upper half-plane give the energies and damping rates of the

quasi-holes.
5-7 IN T E R P R E T A T IO N O F A ( p, a,)
The above discussion is made clearer if we try to understand
the spectral weight function
A (P> <*>) = 2 K CP + ) m . o | 2 8 (t o - <Omn + 1 )
m
+ 2m Kcp)m. o|2 S(“ " “ m"-1) (5-37a)
in greater detail. Consider first the positive frequency part of
A (co > 0 ). If the system is initially in its ground state |^0n)
and a bare electron is placed in the state p at t = 0 , the state

vector just after the electron has been added is given by
0 P> = cp+|¥V> (5-50)
We should keep in mind that unlike the states | l/ymv), |0 p> is not
in general normalized to unity. This is easily understood if
|¥V> is expanded in occupation number states; according to the
Pauli principle any component of |*P0n> in which the state p is
occupied gives a vanishing contribution to (5-50), thereby chang
ing the normalization of | 0 P>.
Suppose we want to know the relative probability that the
system described by | 0 p> (not an eigenfunction of H in general)
is in the eigenstate |0 mn+r >. Aside from a normalization factor,
the probability is
^m(P) = \ <^ mn + 1\ ^ y \ 2 = K O m . o l 2
Thus the strength of the delta function in the first term of (5-37)
is just this relative probability. [For a translationally invariant
system, P m(p) vanishes unless the momentum of state m is p.]
Suppose we want the relative probability that just after the
electron is injected, the system is in the group of eigenstates
whose energy is in the interval a> —> a> + w. Again, aside from
8
the normalization factor, the probability is
Therefore, A( p, a>) gives the relative probability per unit energy

that the system described by |0 p> has an energy a> 4- /x greater
than that of | 0 on)*
If |0 P) were normalized, the completeness of the states
| 0 mn + 1> for the n + 1 particle system would require the total
probability
Field-Theoretic M e th o d s in the M an y -B o d y Problem 121
to be unity. Instead, one finds the integral is

oo
A(V,w)do> = <<£p|<£p> = <ôn|(l - «P)|ôn> = 1 - <»„>
(5-51)
0
The free-gas weight function (5-38) again trivially satisfies this

condition, since for p > p F the excitation energy ep — p, is greater
than zero, so that the integral on the left-hand side of (5-51) is
unity and <np> = 0 in this case, while the integral vanishes for
p = cp|¥V>
which describes the system just after a bare electron of momentum

p has been withdrawn. Again, 10P> is in general not normalized
to unity. If we ask for the relative probability that the system
which gives the strength of the delta function in the second

term of (5-37). Owing to our choice of signs in the definition of
comn _ 1 [see (5-34b)], the excitation energy i £mn - 1 — E 0n~1 is
defined to be the negative of Then, if ^4(p, to) is plotted
as a function of to, one should think of the excitation energy of
states involved in electron extraction as being measured along the
negative co-axis. Our choice of signs unfolds the electron and
hole parts of A; otherwise we would have to introduce two
functions, say p( +) and p(~\ defined on the interval 0 ^ oo by
p<+)(p,«) = 2 |(cp+)m.o| 2 S(co - o)mn +1) (5-36a)
m
and
p<->(p,o») = 2 |(cp)m,o| 2 S(co - K " - 1!) (5-36b)
m
so that
G(P» Po) = Jo[qr Po p(~+ )(p,«o)dq>
w ~ P + *8
+ f" P( )(P> w) dco
1Jo Po + OJ - ft ~ i 8
(5-37b)
The p-functions are somewhat more convenient than A for finite-

temperature problems where the excitation energies are not
necessarily positive.
To connect these spectral weight functions with those dis
cussed by Kadanoff and Baym, we note that our A, p( +), and
p(_) are related to their A BK, G > , and 6 ?< by
A bk(P, u>) = nA(p, oj)
2
^?>(p) = 27T/>( + )(P>

G <(V’ OJ) = 27rp(_)(p, —co)
The quasi-particle approximation is now simply understood:
the probability distribution for the energy of the system just
after a bare electron that has either been added to or subtracted
from the ground state is approximated by one or more Lorentzian
functions:
I r <+>i ir <->i
u p2 — + v p2 P7T
= [w _ + / y +>2 [w + e„<->]* + r p<->2 (5'52)

This expression satisfies the sum rule (5-40) if u -f v = 1 . p 2 2
This weight function is more complicated than that for the free
Fermi gas, 8 (o> - [ep - /*]), for three reasons.
1 . The quasi-particle energy E p(±) is not in general equal to
± |cp — fx| but includes “ self-energy” effects arising from inter
actions of the injected particle (or hole) with the medium.
2 . The delta function has been smeared out into a finite-
width Lorentzian function, owing to |0 P> not being an eigenstate
of the interacting system in general.
3. There are two peaks in the weight function (5-52), one for
positive and one for negative o>, both of which have finite weights
up 2 and vp2, respectively. In accordance with our discussion
above, the two peaks reflect the possibility of either adding an
electron to momentum state p (leading to the positive energy
peak) or removing an electron from this state (giving the negative
energy peak). Both processes are possible in the interacting
system for a given state p since the probability that this state is
occupied is neither one nor zero.
Field-T heoretic M e th o d s in the M a n y -B o d y Problem 123
It is interesting to note that the presence of large self-energy

effects does not necessarily imply that the level widths T are large.
In fact, for a system described by the BCS reduced Hamiltonian
(see Chapter 2 ), the spectral weight function is exactly of the form
(5-52) with / y +) = r pi ~) —> 0:
^bcs(P> w) - up 2 - E p) + v 8 (o> + Ep)
p 2 (5-53)
Thus, the reduced Hamiltonian has the peculiar property in
common with the free-electron gas that cp| 0 > and cp+| 0 ) are
eigenstates of the system. The quasi-particle energy E p is given
by
Ep = [(cp - + J p2] 1/2 (5-54)
where Ap is the energy-gap parameter. For €p ^ /z, E p is drastic
ally different from ep — /z. The functions u and v are 2 2
< - 1 (■ + t t ) (5-55a)
^ - 5 (* ” V ) (5' “ b)
and give the probability that the bare particle state p is unoccu
pied or occupied, respectively. For Ap -> 0 , the BCS spectral
function goes over to that of the free Fermi gas. If one goes
beyond the BCS approximation, finite level widths appear (as well
as more complicated “ continuum” contributions, which will be
discussed later). While ^4(p, o>) is in general an extremely com
plicated function, its form for small ep — /z and small oj is expected
to be simply the sum of two Lorentzian functions as in (5-52);
however, the sum of the residues may be less than unity. This
simplicity is exploited in Landau’s theory of the Fermi liquid,
discussed in Chapter 2 , although a correspondingly simple picture
holds for the superconducting state. Since most transport prop
erties emphasize this low-energy part of the spectral function, a
quasi-particle approximation is often sufficient for calculating the
electrical and thermal conductivities, etc.
5-8 T H E O N E - P H O N O N G R E E N ’S F U N C T IO N
In a completely analogous manner we introduce the one-
phonon Green’s function D defined by
, t 1; r2, t2) = —^ 0 | T{<pA(ri, tx)(pA+ (r2, G)}|0/ (5-56)
where the T-product is given by
n r, - M r,"
(r 2 J ^)(?)/l(ri’ *2!
IVa ^l) > <!
< ^2
D depends on time only through the difference r = tx —t2, and
fora translationally invariant system depends on the spatial
variables only through r = r — r2. With the aid of the
1
expansion
<pAr, 0 = 2
Q
we see that the propagator of phonons of wave vector q and
polarization A is
A,(q, r) = —i(0| T{<pqA(r)q>qA+ (0)}|0> (5-57)
for a translationally invariant system. We recall that the
phonon-field amplitude is related to the creation and destruction
operators by
*PqA ® - q?.
As for G, we can write a spectral representation of D by introducing

the weight function
B x(q. w) = 2 I<wl9v +I°>l2S(w - a ij
m
- 2 l<m k A | 0 > | 2 S(cu + o>m) (5-58)
m
where tum is the excitation energy E mn - E 0n of the ?i-particle
system. Then the time Fourier transform of D ?(q, T) *s given by
where S = 0 + . Since by definition B is a real quantity, the

imaginary part of (5-59) gives the relation
Im D,(q, q0) = - ^ ( q , Qo) sgn 9o (5-60)
so that a dispersion relation holds for D :
^ v I f 00 Im DA q, w) sgn co _ ^
Ai(q> *U/
?o = 77 J_ oo q0 - W + lOJdo- 5-61
Just as for G(p,p0), we can extend the definition of D( q, q0)
into the complex go-plane with the aid of the spectral representa
tion (5-59) and show that the poles of the analytic continuation
of D into the lower half-plane for > 0 may be interpreted as
giving the energy and decay rate of phonons (or phonon “ holes”)
of wave vector q (or —q) and polarization A. The situation is
somewhat more complicated here than in the electron case since
it is the field amplitude cpqA which is used in defining D rather than
the destruction operator a q A, as was used for the electron Green’s
function. We recall that the electron injection and extraction
processes were separated by our definition of A. We see from
the definition of B that both phonon creation and destruction
processes contribute to the positive (and to the negative) fre
quency parts of B. In fact, B is an antisymmetric function of
co for a system with inversion symmetry since cpqA+ = <p_Qt;
In this case we can write D as
£t(q, 7o) = Jfo B„{ q.cu) q— 2 —rco^ ~ mI du>
0
2
0
<5' 62)
For a system consisting only of bare phonons (i.e., uncoupled
to the electrons), the spectral function is
B x{q, o>) = 8 (|w| - A) sgn a> (5-63)
so that the bare phonon propagator is
20
D °A q ’ qo) = q02 - + »S (5' 64)
In the next section we shall see how to build up the true

one-electron and one-phonon Green’s functions as a perturbation
series involving the corresponding functions for the noninteracting
system.
5-9 P E R T U R B A T IO N SERIES
We shall not go into the details of deriving the perturbation-
series expansions for the electron and phonon Green’s functions
for two reasons: there exist several good discussions of this
derivation in print103; also one need not keep the derivation in
mind when using the simple rules that follow from this derivation.
The perturbation expansion is based on two main assump
tions. If the zero-order Hamiltonian which describes the bare
particles (i.e., our Bloch electrons and bare phonons) is designated
by H q, then the ground state of H is assumed to go over adia-
0
batically to the ground state of the interacting system as the inter

action is turned on adiabatically in time. In addition, it is
assumed that the resulting time dependence of the ground state
and the r-dependence of G and D because of the presence of the
interactions can be expanded as a power series in the strength of
the interaction. These assumptions are less restrictive than they
appear, since selected terms of these divergent expansions can
often be summed to give physically meaningful results, as men
tioned earlier. Unfortunately, one is often open to the criticism
that neglected terms could drastically modify the results.
For simplicity we begin by assuming the total crystal
momentum (i.e., the momentum k of a Bloch function, q of a
phonon) is conserved bv the interactions; that is, we neglect
umklapp processes and all but the diagonal part of the one-body
potential U. We can easily extend the treatm ent to include these
effects. The rules for calculating the one-electron Green’s func
tion (?s(p, p 0) for the interacting system described by the Hamil
tonian (4-35) are listed below.
1. Draw all diagrams in which an electron of spin s and four-
momentum p (represented by a directed solid line) enters from
the right and leaves from the left, undergoing all topologically
distinct interactions in the process. Only “ connected diagrams,”
those which cannot be separated into two or more unconnected
pieces without breaking any lines, are to be included. The Cou
lomb interaction between electrons is represented by a dashed
line and the one-body potential U is represented by a dotted line
Field-Theoretic M e th d d s in the M an y -B o d y P roblem 127
connecting the electron line to a cross (i.e., the source of the

one-body potential). A phonon is represented by a wavy line.
Label all the lines so that four-momentum and spin are conserved
at each vertex. The “ spin” of the phonon is taken to be zero
in our definitions.
2 . Associate with each bare electron line of four-momentum
p the factor iG (p), where G ( p ) is the bare electron Green’s
0 0
function
G (P) = --------------------------------------- (5-23a)
Po - «p +
0
3 . Associate with each bare phonon line of four-momentum

q and polarization A the factor iD A(q), where D A(q) is the bare
0 0
phonon Green’s function

* “ <»> - -% :■ + a (5 '6 4 a )
4. Associate with each vertex in which a phonon of polariza

tion A and momentum q is emitted (or —q is absorbed), scattering
an electron from p to p', the factor gPtP>tx-
5. Associate with each Coulomb line which involves one-
electron scattering from p to p' and the other scattering from
k to k ' the factor <p', k'| F|p, k ). If the bare electron states are
approximated by plane waves the factor is
t7{ _ 4?re2
p p - (p - p')2
6 . Associate with each one-body line connected to an electron
of momentum p the factor <p|t/|p>. The four-momentum is
conserved in the process.
7. Include a factor ( —i)n, where n is the number of inter
actions (4), (5), and (6 ) taking place in the graph, and a factor
( —l)1, where I is the number of closed electron loops in the graph.
8 . Multiply the above factors together and integrate over all
free internal four-momenta k x • • •k n according to
f d*k 2 d*kn
J (27r)4 (2 tt)4' ’ ’ (2 tr) 4 t ^ 3 ^ 4 5 ^ 6 7]
where F 2, F 3, . . . are the factors arising from rules 2 , 3, . . ..
9. Sum the resulting contribution over all phonon polariza

tions and spin orientations of the internal electron lines, and sum
the contributions from all topologically distinct graphs.
We illustrate the rules by calculating a few low-order contri
butions to iG (p)y shown graphically in Figure 5-4a, b, c, d, and e.
These are
(a) iG 0 (p) = (5-65a)
(b) [iG0 (p)]
X iG 0 (p') liGo(p)] (5-65b)

(c) (5-65c)
(d) [iG0 (F)] - *) j ^ 4 <P'- Pi F IP’ P'>iG0(P') [iGo(p)}
(5-65d)
d^p'
(e) [tGW?)] 2 f \9w A H D ox( P - P' )
X iO0 (p') liGo ip)} (5-65e)

The term (a) neglects all interactions and approximates G for the
interacting system by its value G for the noninteracting system.
0
Term (b) describes the interaction of the added electron (or hole)
with the average charge distribution of the bare conduction elec
trons of both spin orientations. While this term is formally
infinite in the limit of a system with infinite volume, this infinity
is cancelled by an infinite contribution of opposite sign arising
from the one-body potential U as shown in Figure 5-4c (see
Sections 4 - 4 and 4-5 for the definition of U). By thinking through
the definition of U, it becomes clear that the cancellation simply
reflects the over-all electrical neutrality of the electron-ion system.
A similar cancellation occurs whenever a diagonal matrix element
of the Coulomb interaction appears, so that we need not worry
(d) (e)
F IG U R E 5-4 Terms entering the perturbation series for iG(p, p 0).

(a) The zeroth approximation is ô)- (b) The lowest-order
“ direct” contribution from the Coloum b interactions, (c) The lowest-
order contribution from the one-body potential U. This cancels the
divergent Coulom b contribution from (b). (d) The lowest-order
“ exchange” contribution from the C ou lom b interaction, (e) The
lowest-order contribution from the phonon field.
about infinities o f this nature. In passing we note that there is no

term corresponding to (b) in which the Coulomb line is replaced
by a phonon line, since by definition there is no q = 0 phonon
mode. Term (d) represents the lowest-order exchange contribu
tion due to the Coulomb potential. No spin sum appears in (d),
as we expect. As we shall see, when the perturbation series is
suitably rearranged, graphs which look like (b) and (c) will include
the interactions within the Hartree approximation. By including
the counterpart of (d), one gets the Hartree-Fock approximation.
Term (e), the lowest-order phonon process entering G, will
lead to an effective mass correction and a finite decay rate of the
added electron (or hole) due to phonon emission.
After inserting the expressions for G and D0, one can easily
0
carry out the p 0'-integrals by the method of residues; however,

the three-momentum integrals cannot be performed unless we
have explicit expressions for the matrix elements involved. The
result of the p 0'-integrations will be given below.
It is important to notice that the above contributions to G

can be written as
G(P) = O (p) + G (p)ZR(p)G (p)
0 0 (5-66)
0
where the “ reducible self-energy” ZR(p) is given to this order by
x 2 p'l V|p, p'> + <p\ p|F|p, p'>

^ S
+ 2 \0pp >.\2 D ox(P ~ P ')| + <P|#|P> (5-67)

Since every graph for iG starts and ends with a bare electron line,
it follows that G(p) can be exactly expressed in the form (5-66),
where Z R(p) contains no external lines entering or leaving the
graph. An important observation is that G(p) can also be
expressed in terms of the “ irreducible self-energy” Z(p):
= Go (P) + G (p )I(p )G (p )
G(P) 0 (5-68)
where Z(p) si given by the sum of all graphs giving Z R except
those whichcan be separated into two unconnected pieces by
cutting a single bare electron line. Thus, while the graphs
shown in Figure 5-4b, c, d, and e contribute to Z (and Z R), the
graphs shown in Figure 5-5a and b contribute only to Z R and
must not be included in computing Z since these terms are already
accounted for in (5-68) as one can see by solving this equation by
iteration. On the other hand Figure 5-5c shows a valid con
tribution to Z. Equation (5-68) is known as Dyson’s equation
(for G) . 102 We note that
W p) = G j f ) ~ I{P )
or using the expression for G 0 (p) we find the simple relation

G(p) = --------------------- r— ( - 5 69)
Po - *p ~ 2(P ) + ly)v
Field-T h eoretic M e th o d s in the M a n y -B o d y Problem 131
The power of the graphical methods is in part due to the fact

that one can calculate E (p) to low order in perturbation theory
and often obtain a reasonable approximation for G, even though
a straightforward series expansion of G in powers of the inter
action strength is meaningless. As we shall see, the effect of
electron-phonon interactions on G in normal metals comes almost
entirely from the lowest-order graph for E, as was first pointed
out by Migdal. 15
In Section 5-5 we noted that the energy E v and damping
rate |FP| of a quasi-particle are given by the pole of the analytic
continuation of G into the lower right (upper left) half-plane . 100
From (5-69) we have
E p = €p -H Re Z?(p, E p + i r p) (5-70a)
rp = Im E( p, Ep + %rp) (5-70b)
where E is the appropriately continued function. While such
poles usually exist, one must be careful to check that they give
(a) (b)
-L
(c)
F IG U R E 5-5 Higher-order corrections to iG(p, p 0). (a) This graph is
automatically included if one includes the contribution shown in
Figure 3-4b in the irreducible self-energy E. (b) This graph is also
included automatically if Figures 3-4d and 3-4e are included in E. (c)
This graph is not automatically included since it has an irreducible self
energy part. Its contribution must be included explicitly in E.
a good representation of the spectral weight function A( p, oj) for

real w, and hence a good approximation of (?(p, p 0). In general,
the analytic behavior of G is more complicated than the idealized
quasi-particle picture, as we shall see in Chapter 6 .
Returning to the low-order contribution to E considered
above, we have from (5-67):
z (p) - J ^ 3 ( 2 p'If Ip>p'> - <p'> pI f Ip> p'»/p'
+ + 2 * / ^ lsw*l2
X (Po - Po - ®<,A + iSPo' - V +

' ) 2 2 iVp' (5 71)
where q = p — p\ and we have used the relation
ptPn'5
dpo ________
27T p 0' - f p. + ^ p'
- 1 IP'I P f
(5-72)
The first two terms on the right-hand side of (5-71) give the direct
and exchange Coulomb contributions; the former is largely
cancelled by the third term (p\U\p}, as we mentioned above.
The last term in (5-71) is most easily evaluated by writing
i /%■ D° ^ p ~ p ')G°{p,)
_ i [ dJ E l ( 1______________ I______ )
J 2n \p0' - Po - QqA + 1 8 Po - Po + Qqx - *8/
x Po
—;------ -------------- (5-73)
- ep. + ir)p.
and noticing that the first factor in D does not contribute for
0
IP'\ > P f since neither this factor nor G (p') has poles in the 0
upper half-plane; therefore, by closing the jp0'-contour in the

upper half-plane we obtain zero in this case. Similarly, the second

term in D does not contribute for \p'\ < p F, since neither this
0
factor nor GQ( p ‘) has poles in the lower half-plane, and we obtain
zero by closing in the lower half-plane in this case. The remaining
contributions are
r l
J 2 tt p 0' - p 0 - QqA + p0 €p> i8
Po - V +1 |P'|
1
< Pf
(5_74)
_i r ^pp_ _________ i 1
J 2tr p o - Po ++ ®qx ~ i&Po' ~ V + iS
Po - «P- - Qq* + I? I > P f
Thus, the last term in (5-71), which arises from the lowest-order
phonon process, is
* -0» - ^pp' a|
1 - fp - , fp- (5-75)
Po - - Q qA + i8 Po - «p. + Qqx - ih
The two terms in this expression can be easily understood
within the framework of time-independent (Brillouin-Wigner)
perturbation theory. Consider adding to the system an electron
in the state p above the Fermi surface. The familiar second-order
process shown in Figure 5-6, in which a phonon of energy Qq A
is virtually emitted and reabsorbed by the added electron, leads
to the energy shift
dzP' IsWa | 2 ( 1 ~
AE == Z^ P
r \
TTTTS
( 2
r £iLlU
po - v -
fy ,
7T
QqX1 (5-76)
Here P indicates the principal part of the integral is to be taken;

the factor ( 1 —f p.) reflects the Pauli principle requirement that
the intermediate state p' is initially unoccupied. Equation (5-76)
agrees with the real part of the first term of (5-75).
The second term in (5-75) is somewhat more subtle but

illustrates a general feature of many-body perturbation theory.
Consider for the moment the system in the absence of the added
electron. Owing to the electron-phonon interaction there will be
processes contributing to the ground-state energy of the system in
which an electron, initially in a state p' within the Fermi sea,
virtually emits a phonon of energy Qq k and jumps into an un
occupied state p above the Fermi surface. The excited electron
then returns to its initial state by reabsorbing the phonon, as
shown in Figure 5-7. For a given intermediate state p (above the
Fermi surface) these processes contribute the energy shift
(5-77)
to the ground state. Note the symbol P is not important here
since the integrand is never singular. If an electron is now
added to the system in state p, the virtual excitation of the Fermi
sea leading to (5-77) is forbidden by the Pauli principle. There
fore, this term must be subtracted in calculating the excitation
energy. Thus, the total excitation energy of the system, when
an electron is added in state p, is given by
We know from the discussion above that the excitation energy of

the system is also given by the pole of G(p, p 0); thus to order g 2
Ep = €p + Re Z\p, E p) = €p Re27(p, ep) (5-79)

If one uses (5 -7 5 ) for Z, the two expressions (5-78) and (5-79) are
identical, as desired.
Throughout our application of many-body perturbation
theory we shall see analogous examples of the Pauli principle
bringing about modifications of the virtual fluctuations of the
system when a particle (or hole) is added to the system. In the
q> A
F IG U R E 5-6 A typical contribution to the second-order time-

independent perturbation series for the ground-state energy in the
presence of an added particle of momentum p. The Pauli principle
restricts the intermediate state to |p'| > pF.
superconducting state a major fraction of the energy gap arises

from this type of effect. In a real sense the Pauli principle is
ultimately responsible for absence of low-lying elementary exci
tations in superconducting metals.
F IG U R E 5-7 A typical contribution to the ground-state energy in the

absence of the extra particle. The Pauli principle restricts the inter
mediate states to |p| > pF. This “ vacuum-fluctuation ” process is
suppressed if a particle is added to p, thereby leading to a shift of the
quasi-particle energy.
136 T h e o ry of Sup e rcon d u ctiv ity
To complete the interpretation of (5-75), we observe that

—2 Im i^p, €p) = \Tp\ gives the damping rate of the added
2
electron. From (5-75) we have

- 2 Im Z(p, €p)
= IflW*!2*1 - } S(£> - e*- -

2 n l f j S f (5'80)
a result identical with that of conventional first-order time-
dependent perturbation theory, i.e., the golden rule . 104 A similar
argument gives the decay rate of an added hole.
CHAPTER 6
ELEMENTARY
EXCITATIONS IN NORMAL
METALS
If one attempts to calculate the electronic self-energy in normal

metals by the perturbation series described above, one quickly
discovers that the series diverges, owing to the singular nature
of the Coulomb matrix elements evaluated for small momentum
transfer q.
6-1 T H E E L E C T R O N G A S W IT H C O U L O M B
IN T E R A C T IO N S
This singularity, because of the long-range nature of the
Coulomb potential, can be circumvented by summing a set of
graphs which physically represents the screening of this potential
by the valence electrons. We neglect the phonons for the moment.
From the pioneering work of Bohm and Pines 105 and from the
more recent work of Gell-Mann and Brueckner , 106 one knows that
the most important screening effects in the limit q -> 0 are given
137
»■— - + - 0 - + - - o - o —
+ — c ^ ) — — O ) — +,tc
F IG U R E 6-1 The random phase approximation (RPA ) for the effective
interaction in the electron gas.
by summing the so-called bubble graphs. In effect, one can re

place the original bare Coulomb interaction V by an effective
potential c given by the series shown in Figure 6 - 1 . This
approximation for the screening is known as the random phase
approximation (RPA). To simplify the algebra we neglect
momentum nonconserving processes and approximate the Bloch
states by plane wave states so that matrix elements of the Cou
lomb interaction are given by V{q) = 47re lq2. By using the 2
rules given in Chapter 3 for evaluating Feynman graphs, we have

r ? ™ ( q ) = V(q) - V(q)P™A(q)*~?PA(q) (6-1)
where the “ irreducible polarizability” P{q), evaluated within the
RPA, is given by
P RPA = 2 i j G o {p + q)G oip) (6-2)
the factor of two coming from the spin sum. By rearranging

(6 - 1 ) we have
^-RPA/„\ = Ylul = Z iil (6 -3 )
' c (q) 1 + V(q)P*™{q) Ko(q) K ’
where q = (q, q0) and V(q) is a function of q 2 alone. From the

form of (6 -3 ), it is clear that the denominator is the wave vector
and frequency-dependent dielectric function K-0 (q, #o) °f
valence electrons, evaluated within the random phase approxi
mation.103a’ 107 By inserting the expression for G into (6-3), it 0
immediately follows that

Elementary Excitations in Normal Metals 139
Re *0 (q, q0) = 1 + V(q) Re P EPA(q, q0)

= 1 - 2 V(q)P
+q ^p)
J[ —(2-rr)3 f (1 - / p + «)
?o2 - (^p +q ^p)
(6-4)
where P indicates the principal part of the integral is to be taken.
In addition we have
Im K0(q, 9o) = 2 ” V(q) ~ f P+q)
X [% 0 — €P + q + €P ) + 5(g0 + €P + q ~~ ep)l (^“5)

The integrations (6-4) and (6-5) are straightforward and lead to the
results first given by Lindhard108:
/ q0) = 1 +
Re #c0 (q, 2 e 2 m kF i _ + j_ y
24f/
h
Trq 2 q
/mqo _q_\
1 +
x In \qkF 2 k FJ
(6 -6 a)
i _ M * +
\?&F 2k F)
/Lp + M go 3_\
_ M go <7_\
1
2 kFJ
2? \qkF 2 kF) _ im o g\
\qkF 2 kFJ
and
f0 for 2 m|g0| > q2 -j- 2 qkF
0 for q > 2kF and 2m\q0\< q2 — 2qkF
e2 m 2 ^q for q < and kF

Im K0 (q, qQ)
2 2
6
= i
2 m\q0\ < |q 2 - 2 gkF\ (6 -6 b)
e 2 mkp 2 _ (m o _ JL Y X
\qkF kp) j 2
for |q - 2 2 qkp\ < 2 m\q0\ < | q 2 + 2 qkF\

For q Q = 0, (6 -6 ) gives Im k = 0 so that the static dielectric

constant is given by
*o(q> 0) = 1 + 0.66rs^ ) ^6'7a^
where
“w - ^ [ 1 + i L i r i ) l n | r ^ | ] <6-7b'
and
= 1
477rs3 a 0 3
3 n
(6-7c)
a being the Bohr radius. The quantity rs is a dimensionless
0
measure of the strength of the Coulomb interaction. A plot of

1/*o(^ 0) as obtained from (6-7) is given in Figure 6 -2 , along
with a plot of the Thomas-Fermi result. The main features of
the plot are given below.
1 . kq approaches the Thomas-Fermi result
*o(q. 0) = 1 + ^ (6-8)
as q -> 0 , where the screening wave number is given by
fan e2
Ep
F IG U R E 6-2 A plot of the zero-frequency R P A dielectric function k

as a function of the wave number q. Also shown is the Thom as-Ferm i
approximation for k.
Elem entary Excitation s in N o rm a l M etals 141
2. d K 0/dq -> oo as q -> 2kF. This fact leads to the result,

first discussed by Kohn, Langer, and Vosko,109a b that the asymp
totic form of the screened Coulomb potential for large distance is
not a Yukawa potential but rather the oscillatory function
oc cos J (6-9)
There is good experimental evidence to support this result . 1 1 0
3 . k 0 -> 1 as (/-> oo. Thus, screening is ineffective for very
large momentum transfers.
For g0 ^ 0 , the imaginary part of k is nonzero only when q 0
and q are related so that the argument of the delta function in

(6-5) can vanish for some |p| p F, that is to
say,
q — qkF ^ 2 m|g0| ^ q + 2 qkF
2 2 2 (6 - 1 0 )
The situation is illustrated in Figure 6-3. For large frequency,
i.e., |g0| » q + (2 g£F/ 2 ra), we have the familiar limiting form
2
2
Re K0 (q>?0) = 1 - ^ (6 - 1 1 )
HQ
where o>p2 = 4rrne lm . The two expressions (6 -8 ) and (6 - 1 1 ) are

2
useful limiting forms for making physical arguments.
F IG U R E 6-3 A plot showing the behavior of the imaginary part of the

R P A dielectric function for general wave number q and frequency q0.
While the RPA is valid for small momentum transfers, it is

desirable to introduce an effective potential 'V'dq) which includes
all vacuum polarization processes.103a A partial summation of
the series can be performed if we introduce the irreducible
polarizability P (q ), defined as the sum of all graphs in which a
single interaction line enters and leaves the graph and which
cannot be divided into two disconnected graphs by cutting a
single Coulomb line. P(q) is shown in Figure 6-4. Then,
r M - = YM (6 - i 2 )'
YM) + V{q)
V(q)P(q) ~ K(q)
1
The RPA result is given by retaining only the first term in the
series for P(q).
To obtain a compact expression for /c(g), we consider the
quantit}' (evaluated in the Heisenberg picture)
<0 |T { p _ ,(-r)p ,( 0 )}|0 > (6-13a)
Here pq is the gth Fourier component of the electron density
operator,
pq{t) = 2 f <ri'1',Vrs+(r><)&(r>0 d3r = ^ c p^ + (t)cp + q,s(t)

s " P. s
(6-13b)
Again, the phonons are neglected for the moment. The time
Fourier transform of (6-13a) can be expanded as a perturbation
P(„ = < ^ > + < ^ 7 > + < 2 > + < ^ >

N /
+ <3 >+ + + + e tc '
F IG U R E 6-4 The perturbation series for the irreducible polarization

propagator P(q). The R P A retains only the first term in the series.
Elem entary E xcitations in N o rm a l M etals 143
series in the Coulomb interaction strength and is easily seen to be

related to the series for F~c(q):
r c(q) - V(q) = _ L _ ,
F(g) K(q)
= f e,flo*<0| T { p - q( - T)pq(0)}\0> dr
oo J -
(6-13c)
This expression allows us to study the analytic structure of
*(q> 0 o) as a function of q for fixed q. As for G(p, p 0), we can
0
insert a complete set of eigenstates of H between pq and p _ q

in (6-13) and obtain the spectral representation
—— - = f" ^ (q’ w) <loJ (6-14a)
*(<!> 0 o) J - oo q 0 - CO + iq0b
1
where
F(q, <o) = V(q) 2 |<«|p_ 9 | 0 ) | 2 S(o> - con0) for «, > 0 (6-14b)
n
and
F(q,o>) = - F ( - q , |w|) for < 0 (6-14c)
For a system with inversion symmetry (6-14c) is equivalent to
F(q,<v) = - F ( q, —co) so that (6-14a) becomes
— — - 1 = f °° F( q, co) (--------------------- r-g) da>
*(q> 0o) Jo Wo - ^ + Wo° 0o + w + Wo<7
(6-15)
The right-hand side of (6-15) is of the same form as the spectral
representation of the phonon (boson) Green’s function. The
essential difference between the two is that the weight function
F(q, co) involves matrix elements of the electronic density fluctua
tion operator, while the longitudinal phonon weight function
0 (q, co) involves matrix elements of the ionic density fluctuation
operator. Thus, in analogy with the phonon Green’s function,
the poles of 1 //c(q, g0) glve boson-like excitations of the
electron gas which are excited by a longitudinal field.
The nature of these Bose excitations is easily understood

within the RPA. By taking the imaginary part of (6-14a) one
finds
for qQ > 0
(6-16a)
I 7T *(q, q0)
for q o c 0
or
= 1# *2(9)sgn 9o (6 - 16b)
7T Kl (q) + 22
2 K (q)
where k ± and k 2 are the real and imaginary parts of k , respectively.

If the RPA expressions (6 -6 a) and (6 -6 b) are used for
and k 2 , we see that, for a given momentum q, the boson excitations
form a continuum over the interval
q - q k F < q ^ q + 2 qkF
2 2 0 2
since k 2 is finite on this interval but vanishes elsewhere, as stated

by (6 - 1 0 ). These excitations are in one-to-one correspondence
with the excitations of the noninteracting system, in which an
electron initially in state p within the Fermi sea is raised above the
Fermi surface to a state p + q.
While the RPA predicts the energies of these single-particle
like excitations to be identical with corresponding excitation
energies in the noninteracting system, the 4'wave functions” of
the excitations are markedly different because of correlations
existing between the excited electron (or hole) and the back
ground electrons. Crudely speaking, the excited electron pushes
neighboring electrons away, making a 44correlation hole” which
follows the excited particle as it travels through the system. This
local depletion of background electrons allows the fixed positive
ions to terminate the electric field lines of the excited electron so
that the net electric field associated with the excitation, i.e., the
excited electron plus its correlation hole, vanishes far outside of
the correlation hole. Since there is no long-range electric field
associated with the excitation, Gauss 5 law states that the net
Elem entary Excitations in N o r m a l M etals 145
effective charge of the excitation is zero. Therefore, if the excited

electron moves through the system, it follows that a backflow
of background electrons must accompany the excitation in such a
way that the net charge transported is zero.
This backflow is the analog of the backflow discussed by
Feynman and Cohen 1 1 1 in the case of excitations in superfluid
He4. In our case, the backflow is essential in obtaining consistent
expressions for the response of the system to external fields. In
particular, we shall see that the Meissner effect can be treated in a
gauge-invariant manner once backflow is properly taken into
account in the superconducting state . 1 1 2 - 1 1 4 W ithout back
flow one violates the continuity equation for electronic charge
and therefore violates gauge invariance. This problem is discussed
in detail in Chapter 8 .
As we mentioned above, k 2 vanishes outside of the interval
given by (6 - 1 0 ); however, 1 /k can still be singular if k x and k 2 both
vanish. In the long wavelength, high-frequency limit, (6 - 1 1 )
shows that k x vanishes when
q = co =
0 2 p 2
4tm e2
(6-17)
Thus, there is an excited state of the system of excitation energy
equal to the electronic plasma frequency. Bohm and Pines 105
were the first to discover these excitations in metals and named
them plasmons. Physically, they are simply density-fluctuation
waves of the electron system. For sufficiently large momentum
q, the plasmons pass into the single particle-like continuum, as
shown in Figure 6-5 and become so heavily damped that they are
no longer useful entities. For small q they play a very important
role in determining the response of the system to longitudinal
perturbations.
Having discussed the problem of screening, we now return
to the problem of determining the one-electron Green’s function
for the interacting electron gas. Since the system is electrically
neutral and (we assume) translationally invariant, the Hartree
term in the self-energy Z vanishes. From our above discussion
F IG U R E 6-5 A plot of the excitation spectrum of the electron gas

treated within the RPA.
of screening, it would appear natural to next include the exchange

self-energy, not with the bare Coulomb matrix element V(q) but
rather with the screened potential Y c(q) evaluated within the
RPA, as shown in Figure 6 -6 ; thus,
£(p) —i j G 0
dfiq
(2 ^ (6-18)
While G and Y c are known explicitly, the integration is difficult
0
to carry out for general p. Calculations were first carried out by

Quinn and Ferrell, 1 1 5 who found the energy of a quasi-particle
near the Fermi surface is given by91b
E n = E F*\bPz
~ 2
- 0.166^ ICj?
(In rs + 0.203) + ln rs 1.80 (6-19a)
// r(-q) \
H * A
GQ{p + q)
F IG U R E 6-6 The electron self-energy evaluated within the screened
exchange approximation. This graph presumably gives the most
important contribution to E(p) from small momentum transfer pro
cesses (|q| « kF).
Elem entary Excitations in N o rm a l M etals 147
and the damping rate due to electron-hole pair production is

|2rpair(^)| = (0
2EF(0.252rs112) 1 (6-19b)
The effective mass of the quasi-particle at the Fermi surface
is given by differentiating (6-19a):
1 1 EP
8
= - [ 1 - 0.083rs(ln rs + 0.203)] (6-19c)
m* kF dp -kF m
Since the electronic specific heat C is proportional to ra*, one
obtains
C = 1 + 0.083rs(ln rs + 0.203) (6-20)
cv
a result first derived by Gell-Mann. 11 6 From (6-19b) it follows
that the quasi-particles are well defined, in the sense that
| E (p ) - E f |» | r(p)\, so long as |ep — E F\ < \EF. As we shall
see, this is no longer the case once the phonons are included.
The results (6-19a, b, c) are strictly valid only in the high-
density limit (rs < 1 ). Since rs is typically 2 < rs < 5 for real
metals, one must use these results with care. The long wave
length part of the Coulomb potential is probably well accounted
for in Z by the exchange graph screened within the RPA. For
very short wavelengths (i.e., short-range part of the potential),
parallel spin electrons do not interact since they are kept apart
by the Pauli principle. One can presumably treat the interaction
between antiparallel spin electrons by the second-order graph
shown in Figure 6-7. 103 1 1 7 Silverstein 11 8 and Pines have
k — q,
F IG U R E 6-7 The leading contribution to Z for large momentum

transfers (|g| » kF), after Figure 5-4d has been included.
exploited these limiting forms (i.e., large and small g-limits) by

interpolating between these limits to obtain an estimate of E.
The reader is referred to the literature for more details on this
complicated problem . 91
6-2 T H E C O U P L E D E L E C T R O N - P H O N O N SY STEM
Screening also plays an important role in determining the
dressed phonon frequencies and the dressed electron-phonon
interaction, since the long-range Coulomb forces between ions
and between ions and electrons are screened out by the conduction
electrons. For long wavelength effects we can again use the
random phase approximation since it gives the leading corrections
in this limit. For shorter wavelengths, processes neglected by the
RPA begin to play a role. However, the RPA gives roughly
correct answers for all wavelengths, since in the region where it is
least accurate it gives small corrections to the bare quantities so
that errors in these small corrections can often be neglected.
We begin by treating the entire electron-phonon system
within the RPA and, for simplicity, neglect umklapp processes
in both the Coulomb interaction and the electron-phonon inter
action. The (irreducible) longitudinal phonon self-energy n x(q)
is given by the series of graphs shown in Figure 6 -8 . The string
of bubbles, characteristic of the RPA, simply represents the
screening of the ion-ion potential. From the series for r c(q)
shown in Figure 6 - 1 we see that i^ c(q) and TJ^q) differ in two
respects: (a) the incoming and outgoing bare Coulomb lines in
y ' c[q) have been replaced by electron-phonon matrix elements in
n x{q) and (b) the leading term V(q) in the series for y c{q) is
missing in n x(q). Therefore, we have the relation
9ql* Qql 9ql* 9ql 9ql* ^

Figure 6-8 The random phase approximation for the self-energy of
longitudinal phonons.
Elem entary E xcitation s in N o r m a l M etals 149
- Hg) - i = kail n t{q) (6 - 2 1 a)

.*(?)
2
V2 (q) V(q)
where we have used (6-3) in the left equality and have assumed
gv p. A is a function only of the momentum transfer for longi
tudinal phonons. By rearranging (6 -2 1 ) we have
n,(q) =
ball2 1
1 (6 - 2 lb)
_«(?)
-
V(q)
From Dyson’s equation,

A ( ? ) _1 = D 0l ( q ) - X - n t(q)
we find
DM) = bail2'Qa 2|gai 2Qoi ( 6 - 22 )
qo -
2 O2
•^q/ —
V(q)K(q) V(q)
As we mentioned in Chapter 4, the relation

O ql
2 —— 2 bq{|2AjI (4-29a)
V(q)
holds for jellium, so that in this case D t(q ) simplifies to £

DM) = 2 Q„
.Q ,2 (6-23)
q o2 - -x (nq ) +
Since the poles of D give the dressed phonon frequencies ojq t
we find
Oql 2
M (6-24)
*(q> Uql)
where k is the analytic continuation of k across the single-particle
like cut along the g0-axis. Equation (6-24) agrees with the in
tuitive result that the dressed frequencies should be given by
reducing the effective force between ions (or the square of the ionic
charge) by the electronic dielectric constant. Since Q is Q ti2
proportional to the ionic charge squared, (6-24) is the expected

result. In calculating the real part of coQ t it is sufficient to use

the static dielectric constant K(q, 0 ) since typical phonon and
electron frequencies differ by ~ (m/M ~ 1 0 ~3. In the long
) 1 12
wavelength limit we have, using (6 -8 ),

(6-25a)
or
(6-25b)
From this expression we see that the dressed longitudinal phonons
have a sound-wave type of dispersion law with sound velocity
(mZl3M)ll vF, where vF is the Fermi velocity. This result for
2
jellium was first derived by Bohm and Staver . 1 19 The situation

in real metals is considerably more complicated since the bare
phonon frequencies QqA and the bare electron-phonon matrix
elements depend on crystallographic orientation. In addition,
umklapp processes are important. Despite these complications,
screening reduces the bare ion frequencies to sound-wave modes
(i.e., ojq oc q) so that jellium is not an unrealistic model on which
one can test methods.
Suppose we are interested in coupling the dressed phonons to
the electrons. While the ion-ion interactions have been screened,
the electron-ion interactions and therefore the electron-phonon
interactions remain unscreened. Intuitively, one would expect
the screened matrix element g to be given by
(6-26)
that is, the bare matrix element divided by the wave number and
frequency-dependent dielectric function of the conduction elec
trons. This result is easily obtained by noting that the
screened interaction is given by the series shown in Figure 6-9,
which is just the expansion of (6-26) in powers of the irreducible
polarizability.
i,,= 4 . 5 0 ^ - " + . 0 — O — * + e te
F IG U R E 6-9 The random phase approximation for the screened inter
action between electrons and longitudinal phonons.
If one uses the dressed D- function [given by (6 -2 2 )] and the

screened electron-phonon matrix element in calculating other
dynamical quantities, one must be careful not to double count by
including vacuum polarization processes already accounted for
in these functions. The prescription for using D and g is clear
from their definition.
We now determine G (p ) within the screened exchange
approximation of (6-18) but include longitudinal phonons as well
as the Coulomb interaction in the calculation. The screened
Coulomb piece of £ (p ) is still given by (6-18), while the one-
phonon process shown in Figure 6 - 1 0 gives
Z ph(p) = i j o 0(p + q ^ t f D A - q ) ^ (6-27)
where {gQtJ 2 = gqig _ qi. The total self-energy within this approxi
mation is
Z(p) = i j o 0(p + q ) \ r c{q) + { g ^ D ^ - q ) ] (6-28)
£>i(-q)
F IG U R E 6-10 The lowest-order dressed phonon contribution to the

electron self-energy.
If we use the relation 2\gql\2IQ q l = V(q ), which holds for

jellium, we find the simple result
q <> d4g (6-29a)
(?)
+ is (2 w)4
Z(p) = i J*G o(P + O 2
?o2 -
*(?)
or
= * J^ o r(?)___________ (6-29b)
1 + 7(g)P(g) - ^ + iS (277)4
vo
The denominator in (6-29b) is just the total dynamical dielectric
constant of the system including electronic and ionic polariza-
bilities, since the ionic polarizability is given by the high-frequency
F IG U R E 6-11 The real part of the effective interaction between

electrons due to the screened Coulom b interaction and the exchange
of a dressed phonon, plotted as a function of the energy transfer qQ
for a fixed momentum transfer q. The plot is shown for the R PA treat
ment of the “ jellium ” model of a metal. The resonance occurs at the
dressed phonon frequency wqt illustrating the effect of ionic over
screening of the bare Coulom b interaction for q0 < wq and under
screening for q0 > ojq. For high-frequency q0 » wq, the ions do not
respond and ^"(q, q0) approaches the bare Coulom b interaction reduced
by the electronic dielectric function /c(q, q0).
form —Qqti lq in analogy with the limiting form (6 - 1 1 ) of the

2 0 2
electronic polarizability. Thus, the effective screened potential

between conduction electrons is the bare Coulomb potential V(q)
divided by the total dielectric function of the medium. From
(6-29a) we see that if q 2
corresponding to an overscreening of the Coulomb repulsion between

electrons by the vibrating ions. For q > ay the effective
0 2 2
potential is underscreened by the ions vibrating out of phase with

the electrons so that the effective repulsion is stronger than the
bare interaction. This phenomenon is just a dielectric anomaly,
familiar in the theory of dielectrics. For q » to the ionic
0 2 2
polarization is negligible and the bare potential is screened only

by the conduction electrons. A plot of this effective potential
as a function of q is shown in Figure 6 - 1 1 for fixed q. It is the
0
attractive region to the left of the dielectric anomaly which is

primarily responsible for superconductivity.
It is perhaps best to insert a word of caution at this point,
in that we have used the word “potential” in a broader sense
than the formalism of Hamiltonian dynamics allows. In particu
lar, one cannot use this “potential” in a straightforward Hamil
tonian formalism since the strong frequency dependence of the
effective interaction implies strong retardation effects. Since a
two-body potential local must be in time in the Hamiltonian
scheme, the retardation effects must be simulated by a velocity-
dependent potential, a somewhat misleading procedure. In
addition the effective interaction is not a real function so that
H will not be Hermitian in general, unless damping effects are
neglected. For these reasons it is best to treat the superconductor
by Green’s function methods, which allow retardation and damping
effects to be included.
To obtain an explicit expression for E(p) within the screened
exchange approximation (6-28), we need only consider the phonon
contribution (6-27) since the Coulomb piece (6-18) has been dis
cussed above. The integrations in (6-27) are best performed by
first carrying out the three-momentum integral. If the vari
ables |p'| = jp -I- q|, |q|, and <p (the azimuthal angle of p' around
the polar axis p) are introduced, as shown in Figure 6-12, the

phonon contribution becomes
^ {p) - ( 4 l i l f . dq‘ / * ' dp' Tpo + qM* + »8) -

x J\ d q { g q^ D A - q ) (6-30)
For convenience we measure all energies with respect to the

Fermi energy E F so that ePF = 0. Since D decreases as 1/q02 for
large q0, the dominant part of the integral comes from |g0| < wav
[a typical phonon energy, i.e., ~ (mlM)ll E F ~ 10 ~ E F]. We 2 2
shall be interested in electron energies |j?0| < wav so that the

most important values of |ep,| are also of order coav or less.
For this reason the ^'-integral can be replaced by an integral
over €p> with the limits extending from —oo to oo. Thus,
im f°° r°° i
E ^ = (2 nfp J_ „ dq° J _ co dfp' (Po + ? o )(l + *8) - V
r2kF
x q dq{gql}2D,(q) (6-31)
The limits on the q integral have been simplified by using the

fact that only states with \p'\ ~ kF contribute strongly to the
F IG U R E 6-12 Coordinate system for carrying out the momentum

integral in the expression for 2 vh.
integral, as mentioned above. If the maximum phonon wave

number qm is less than kF (as it usually is), transitions involving
2
q > qm are to be interpreted as umklapp processes (which we

include here) and the appropriate reduced-phonon wave vector
is to be used in calculating D . After carrying out the ep,-integral
by the method of residues one finds
771 C2kF
= 3^ J sgn (Po + ?o) <ko Jo
f 00
Z ph(p) Q dq{gql} 2 D,(q)
= JQ (6-32)
m CP o
dq 0 q dq{gqlf D,(q)
where we have used the fact that {g}2D is an even function of

q0. The most important terms in Im Z ph(p) come from taking
{gqi to be real and making a pole approximation (with a real
} 2
frequency coQl) for D^q). Within these approximations we have

yv) PQ(Pq) O
Im Z'ph(7 >) = 4—n p—sgnp 0 J I0 q d q {g ql} ojql (6-33) 2
where q (p 0) is the wave number such that ojql = \p0\. For the
jellium model with K(q) given by the static long wavelength limit
k /q [see (6 - 8 )], we find the damping rate of electrons, because of
2 2
real phonon emission, is given by

mil/ sgnp 0 I to2 dcu
Im i^ph(^) = —-—
4TTj-fC—
mM
l 2 v kFn P o°
v ^ h/ x f lpol
2
pTl Jq
(6-34)
or
where Q = ±TrNZe2jM is the ionic plasma frequency and we have

2
set p = k F since Z varies slowly with p. While (6-34) holds for

p « Qp, one obtains a reasonably good approximation (within
0
the jellium model) for all p by using (6-34) for |_p0| less than the
0
maximum phonon energy o>max ~ Qp and a constant given by

setting |p 0| = a»max in (6-34) for |p 0| > o>max.
By combining (6-19b) and (6-34) one obtains the total damp

ing rate due to electron-hole pair production and real phonon
emission:
(6-35)
where {a, b} means the smaller quantity, a or 6 . Since Qv ~
1 0~ E F phonon processes dominate the damping rate up to
3
|p 0| ~ . l E Fyexcept for extremely small energies |p 0| ~ 10~342p,

0
which are of no real importance in superconductivity.

From (6-32) the real part of U vh(p) is given by
1
For jellium this leads to the effective mass correction near the
Fermi surface,
(6-37)
From (6-37) we see that the phonon cloud surrounding the
electron increases its effective mass. The numerical value of
Smph given by this expression is not to be trusted in real metals
[nor is the damping rate (6-34) to be trusted] since umklapp pro
cesses have been treated as if they were normal processes in our
jellium model and transverse phonons have been neglected al
together. In addition, the bare electron-phonon matrix elements
in real metals no doubt differ widely from the jellium results for
large momentum transfer, as discussed in Chapter 4. Neverthe
less, these expressions suggest that damping effects and mass
corrections due to phonon processes are quite important, a result
supported by experiment.
We turn now to an important observation made by Migdal. 15
Consider for the moment the class of graphs for E, in which no
Coulomb interactions appear, other than those implicitly included
by working with the screened electron-phonon vertex and the

dressed-phonon line. Transverse phonons are neglected for
convenience. This is essentially the model studied by Migdal.
Within the framework of this model one has the exact relation
Z*Hp) = i J gqO(p + -q)r{P , q)D(q) (6-38)
where G is expressed in terms of the unknown function E by
Dyson’s equation. The vertex function JH(p, q) is given by the
sum of all graphs in which an electron of four-momentum p + q
leaves and an electron and phonon of four-momenta p and q ,
respectively, enter, subject to the condition that the graph cannot
be separated into two unconnected pieces by cutting one electron
or one phonon line. A few low-order terms of r are shown in
Figure 6-13a. External lines are connected to the graphs for
clarity but are not to be included in the definition of r . By
estimating the magnitude of the second and higher terms in this
series Migdal argues that the corrections to the leading term gq
are of order (mlM)ll gq ~ 1 0 ~ gq and are therefore negligible.
2 2
P + ?
functionr(p,
F IG U R E 6-13 The perturbation series for the electron-phonon vertex
for clarity.
q). The external electron and phonon lines are included
X
(a)
(b)
F IG U R E 6-14 The one-electron spectral weight function ^4(p, w) for
electrons Interacting with dressed phonons of a constant frequency cj0.
The Bloch energy €p is measured relative to the Fermi surface, and
A is expressed in the form A(€pt xw0). The coupling constant was
chosen such that g2N(0)lwo = The plots are given for (a) ep = 0,
(b) €p = OJS wq, (c) €p = 2w0, and (d) cp = 5ca0. The vertical heavy line
represents a delta function in each plot. (See p. 163.)
(d )
FIGURE 6-14 (continued )
x
(a)
».75a>£>* x w d )
(b)
F IG U R E 6-15 The one-electron spectral weight function ^4(p, w) for
electrons interacting with dressed phonons having a D eb/e spectrum
(oq oc q with a maximum frequency toD. The Bloch energy ep is measured
relative to the Fermi surface, (a) Spectral weight function
for €p = 0. There is a delta-function contribution at x = 0. (b)
Spectral weight function for cp = 0.75o>D. (c) Spectral weight function
for €p = 2o)D. (d) Spectral weight function for ep = 5wD. (see p. 163.)
(c)
(d)
FIGURE 6-15 (continued)

His result depends strongly on the ratio of the sound velocity

divided by the Fermi velocity being small compared to unity.
Since the phase velocity of optical phonons is large compared to
the Fermi velocity as q -> 0 , the result does not hold in this case . 120
One can understand Migdal’s result by considering the first
correction to the vertex shown in Figure 6-13a. The interesting
values of p 0 and q0 are of order or less than the Debye energy
wD. Since the propagator of the phonon which is exchanged across
the vertex varies as ( p 0 — k0)~2 for p 0 - k 0 » wD, the important
values of k0 are also of order wD. Now in order for the electron
propagators to be sufficiently large to give a relative correction
to the vertex of order unity [i.e., of order g2N(Q)lcoD] the inter
mediate states k and k + q must both be within an energy of
order coD of the Fermi surface (unless q « kF\ this case requires
special care). Therefore, instead of a relative correction g2N ( 0 )/
wDi one has the result [g2N(0)l<x)D](a)D/E F) ~ (m/M)112. The same
reasoning holds for more complicated vertex corrections.
If we accept Migdal’s result we can easily solve the integral
equation
£ ph(p) = i j G ( p + q){gq}2D(q) (6-39)
for Z(p). A similar equation, with G replaced by G0, was solved
above, the result being (6-32). As Migdal pointed out, the solu
tion is independent of whether G or G0 appears in the equation.
This follows since Z ph(p) is essentially independent of the three-
momentum in the region where Z affects the integral so that the
€p, -integral in (6-31) is unaltered by subtracting Z(p') from the
denominator to form G (pf) rather than G0(p'). In fact, we have
used Migdal’s integration procedure in deriving (6-31) to illustrate
this point. Thus, the solution of (6-39) is given by (6-32) and
the results following from (6-32) still hold.
While the above result is attractive because of its simplicity,
one becomes suspicious of Migdal’s argument since it states that
the phonon-induced electron self-energy is given correctly to
order (m /M)112 ~ 1 0 " 2 by (6-32). As we have seen, (6-32) leads
to a continuous excitation spectrum without an energy gap, while

superconductors exhibit an energy gap which in some cases is
1 0 _1^ p, i.e., ten times the error limit set by Migdal’s argument.
The difficulty is simply that certain sums of corrections to F,
while formally being of order (m/M)112, are actually divergent.
The resolution of these singularities leads to a qualitatively new
state of the system which exhibits an energy gap in the elementary
excitation spectrum of the electrons and corresponds to the super
conducting state of the metal.
We close this chapter by noting that the solution (6-32) of
Migdal’s problem (i.e., the system described by what is sometimes
known as “ Frohlich’s ” Hamiltonian) allows one to determine the
spectral weight function A(j>, cu) for this problem if gq and ajq are
given. Engelsberg and the author 120 have studied two models
for gq and ajq. The first assumes an Einstein spectrum of phonons,
that is, toq = tu0, a constant, and assumes a general function for
gq. The second model takes gq a,ndwq proportional to q , as in the
Frohlich model. The weight functions for these two models are
shown in Figures 6-14 and 6-15 for several values of kinetic energy
above the Fermi surface. The coupling was chosen to be roughly
appropriate to the strong-coupling superconductors, although the
situation is similar for weaker coupling. The main lesson to be
learned is that, while the quasi-particle approximation is good
for states very near to or very far from the Fermi surface, the
weight function is far from a Lorentzian or a simple sum of
Lorentzians for energies ~o >a v / 2 -> 2 a>av, where coav is the average
phonon frequency. However, it is just this region in which most
of the interaction bringing about superconductivity occurs, as we
mentioned above. Fortunately, one can solve for most properties
of a superconductor without explicitly making a quasi-particle
(or pole) approximation for G.
CHAPTER 7
FIELD-THEORETIC
METHODS APPLIED TO
SUPERCOND UCTIVITY
In Chapters 2 and 3, the superconducting state was treated as

resulting from nonretarded interactions between undamped
quasi-particles of the normal phase. In real metals, retardation
and damping effects play a strong role in a fundamental under
standing of the superconducting phase. In addition, collective
modes are important in understanding the electromagnetic
properties of a superconductor in a gauge-invariant manner. The
methods of quantum field theory outlined in Chapters 4 to 6 are
useful in treating these effects.
7-1 IN S T A B IL IT Y O F T H E N O R M A L P H A SE
In Chapter 2 we saw how the normal state becomes unstable
with respect to bound-pair formation in the presence of an attrac
tive two-body potential. It is instructive to go back to Migdal’s
argument and see which graphs lead to the instability. We would
like to isolate a set of graphs in which two electrons multiply-
164
Field-Theoretic M e th o d s A p p lie d to S u p ercon d u ctivity 165
—j— e tc .
FIG U R E 7-1 The ladder-graph series.
scatter each other corresponding to the two-particle problem

discussed by Cooper. 41 It is clear that the series of ladder
graphs shown in Figure 7-1 represents such an effect. If the
effective potential is attractive, one would expect to find a bound
state of the two-particle system, or an instability of the system
as a whole. To understand how these graphs would enter E(p),
one need only close one of the electron lines onto itself as shown in
Figure 7-2. If we reinterpret the graphs of this figure in terms of
G, D, and r it follows that the instability is due to the singular
behavior of r resulting from the graphs shown in Figure 7-3
[see (6-38)].
In order to understand the instability in more detail we
consider the ladder graph series of Figure 7-1. Except for the
external electron lines, the sum of the series is given by the
“ C’-matrix, defined by
(k' -f q, —k'\t\Jc + q, —k ) = ^ ( k ' — k)
x <k” + q, -k"\t\k + q, - k - ) - — (7-1)

( Z tt)
+ + + + -*■
- ( - e tc .
F IG U R E 7-2 The ladder approximation for E.

\
+ + Q e tc .
/
F IG U R E 7-3 A class of ladder graphs which contributes to the electron-
phonon vertex function T and leads to the superconductor instability.
One can readily establish this relation by performing an iteration

solution of (7-1) in powers of 'V and noticing that one generates
the desired series. While this integral equation cannot be solved
in general, a solution is immediate if we replace q ) by a non
retarded factorizable 5 -wave potential which is finite in a shell
around the Fermi surface
n p k) = A0wp*wk (7-2)
where
1 < cun
wk =
0 otherwise
\e k \
Then one finds the solution

<fc' + q, - k '\ t \ k + q, -fc>
A0Wk' +,*M,k+, (7-3)
d*p
1 - i \ 0 J | w p +q|2<?0(i> + 9)Go ( - p ) ^ 2 „ ) 4
which can be checked by direct substitution.
If we are able to show that t is ill-behaved, then r will also
be ill-behaved since in our approximation r is given by
T f
F(P> q) = 9q 1 - » J k + q\t\p +
d*k 1
kyG0(k)G0(k + q) ^ j 5 j
(7-4)
(see Figure 7-3). From (7-3) we see that t will be singular when
the denominator vanishes, that is, when
Y0 = * J k + q I 2^ * + q)Go ( - k )
d*k
(7-5)
For simplicity we consider the case of zero center-of-mass momen

tum for the interacting pair of electrons (i.e., a q = 0 pair, as in
the Cooper problem). We then seek the values of q0 such that
(7-5) is satisfied. By inserting the expression for G0 and carrying
out the &0-integral by residues, one readily finds that (7-5)
reduces to
1 v' IWuI^ v' IWf, I^ x
A0
T = 2 - 2
\ k \ > k F VO — * € k
s 0 {q °]
|k |< k F HO L€k
( 7 ' 6 )
where we have replaced J d3kl(27t)3 by 2 k since we are interested

in drawing an analogy with the solutions of the Cooper problem
(2 - 1 1 ) for q = 0 . If \wk\2 were zero for states below the Fermi
surface, (2 - 1 1 ) and (7-6) would be identical. By allowing the
interaction to extend down into the Fermi sea, the states below
the Fermi surface play a role in determining the singularities of t.
A plot of the right-hand side of (7-6) is shown in Figure 7-4. For
A0 > 0 (i.e., a repulsive potential) the perturbed states are again
trapped between the unperturbed levels and no bound states
< £(7 o )
F IG U R E 7-4 A plot of the function <P(q0) which determines the poles

of the ^-matrix in the many-body system. For a repulsive s-wave
interaction (A > 0),
0 all poles are real. For an attractive interaction
(A < 0),
0 two pure imaginary poles of t appear (regardless of the strength
|A0| if one considers the limit of a large system). These poles illustrate
the instability of the normal phase at T =0
for an attractive tw o-body
interaction.
appear. For A0 < 0 and very small, again no bound states appear.
However as A0 becomes a large negative number (i.e., stronger
attractive interaction) the 1 /A0 line slides up toward the g0-axis
until it reaches the critical value such that it is tangent to the
peak in 0 ( q Q) near q0 = 0 . Thus far, no bound states have
appeared. Beyond this point, however, two pure imaginary roots
appear corresponding to the two roots which disappeared for A0
greater than the critical value. These pure imaginary roots 10 1
can be obtained by converting the sums in (7-6) to integrals again
and assuming the density of Bloch states in energy near the Fermi
surface is a constant N ( 0 ). Then if we set q0 = i a , we have
(7-7)
For iV(0)A0 < 0 (and larger than the critical value mentioned
above, which is vanishingly small in the limit of a large system)
one has the pair of pure imaginary roots
a =
where the last equality is valid for weak coupling, that is,
|iV(0 )A0| < 14 ’
There are several important differences between the results
of Cooper’s two-particle problem and the ^-matrix problem dis-
cussed above. The solutions of the two-particle problem gave
a real energy corresponding to a stable bound state. When inter
actions are included within the Fermi sea, as in t, there is no longer
a stable bound state formed from the normal phase, but there is a
rapid unstable growth in amplitude of correlated pair states.
Unfortunately, one cannot trace the time evolution of the un
stable system long enough to discover directly what state it is
tending toward, since the ^-approximation is not accurate enough
to treat the correlations in the superconducting phase.
Another difference is that the binding energy of the one-pair
problem involves e~2IN{°)iAo1 5 while the growth rate a involves the

exponentially larger quantity e ~ llNi0)lxoim As we saw in Chapter
2 , the energy gap in the superconducting state involves the latter
quantity, a result which depends on a consistent treatment of
correlations both above and below the Fermi surface.
If one carries out a solution of (7-5) for finite center-of-mass
momentum q, one finds the growth rate a(q) decreases for increas
ing q. Thus, the q = 0 pairs are the most unstable if there is no
net current in the normal state, and one would expect these pairs
to play an important role in forming the ground state of the super
conducting phase. (For an average drift velocity v d of the elec
trons in the normal state the q = 2 m v d pair state has the most
rapid growth rate.)
7-2 N A M B U - G O R ’K O V F O R M A L IS M
While the perturbation series discussed in Chapter 5 for the
Green’s functions is not adequate for resolving the instability of
the normal state, a modified scheme due to Nambu , 114 which
preserves the simplicity of the Feynman-Dyson series, is applic
able. A closely related scheme was developed by Gor’kov 12 1
prior to Nambu’s work. Since the Nambu formalism is particu
larly convenient for carrying out calculations, we shall discuss
this approach in some detail.
Perhaps the simplest way to understand the Nambu formalism
is to consider a system of electrons interacting via a two-body
nonretarded potential V. The Hamiltonian for the system is
h =2k, s
ekn kS + i 2 <k + q>k' - ql Flk’ k'>
k , k ’ ,q
* ^k + q, s C k '-q .s ' ^k's'^ks
= # 0 + # in t (7-9)
In the Hartree-Fock approximation one linearizes the inter
action term with respect to a given state | 0 > so that the two-
body operator H lnt is replaced by a one-body operator. In the
linearization one replaces a typical operator product cx+c2 +c3c4 by
c1 +c2 +c3c4 => <(0 |c1 +c4 | 0 )>c2 +c3 — +c3 | 0 )>c2 +c4
+ (Olca +CalO^ +c* - <0|c2 +c4 |0>c1 +c3 (7-10)
170 T h e o ry o f Sup ercon d u ctivity
The state | 0 > is then determined self-consistently in terms of the

eigenstates of the linearized Hamiltonian. An equivalent, but
for our purposes, more convenient way of formulating this
approximation is to introduce a modified zero-order Hamiltonian
Ho = H 0 + H x - f*N (7-11)
where
H x = 2 X*nks (7-12)
k, s
will turn out to be the Hartree-Fock potential which, for simplicity,
we assume to be translationally invariant and spin independent.
The factor
H-N = M 2 (7' 13)
k, s
is included as a convenience to shift the origin of energy, where /z
is the chemical potential. The modified interaction Hamiltonian is
then
H[nt = H ,nt - H x (7-14)
so that
H' = H 0' + H i’nt = H — fxN (7-15)
Therefore, the energy spectrums of H and H' are identical except
for the energy shift implied by fiN. In this language the Hartree-
Fock (HF) approximation is equivalent to requiring that the
elementary excitation spectrum of H 0' is to first order unaffected
by the residual interactions H[nt.
To carry out this prescription, we note that the poles of the
one-particle Green’s function Cr0 (P> ô) appropriate to H 0' give
the elementary excitation energies ep of H 0'. Restricting the
discussion to zero temperature, we have
G0S(P> t) = - i< 0 |T{cps(<)cps + (0 )}|0 > (7-16)
where | 0 > is the A 0-particle ground state of H 0' and
Cps(t) = eiHo'tcps(0)e~iHo't = cps( 0 ) e - « ' p (7-17)
From (7-17) we find that

Field-Theoretic M e th o d s A p p lie d to Su p e rco n d u ctiv ity 171
where S = 0 + . The factor e i6po in (7-18) ensures that the ô'

integral of G0s is the average occupation number:
- i G 0s(p,t = 0 ) = - i j o os(V,Po )^ -° = <0 |wps| 0 > (7-19)
By inserting the expression (7-18) into (7-19) and closing the

j 90-contour in the upper half-plane we find
< » k .io > = { ; ' i m <«*»
Since the total number of electrons must be N 0, we find the
restriction
= 2 j j s , +xP<u = N 0 (7-21)
d3p
<0 1 2 :
which determines the chemical potential /z for a given H artree-
Fock potential yp. The pole of Cr0s(p, p 0) is at p 0 = ep -f yp — /z
= ep. As we saw in Chapter 5, l v is the excitation energy of the
N o + 1 particle system when an electron is added to the N 0-
particle system, if ep > 0 . If ep < 0 , —ep = |€p| is the excitation
energy of the N 0 — 1 system when an electron is removed from
the jV0-particle system.
To find the effect of H[nt on the excitation spectrum we
require the self-energy i7(p, p 0) due to these residual interactions.
By using the standard Feynman-Dyson rules, one finds to first
order in H[nt
d*pf
^(P. Po) = - 2i J <p. p'l V|p, p')(?o(p', Po') (27T
+ * J <P'» P| V |p, p')Go(p', P o ) - xP (7-22)

The terms correspond to the direct and exchange contributions
and the subtracted Hartree-Fock potential, respectively, as
shown in Figure 7-5. Since the matrix elements of the two-body
potential are independent of p 0f, we can use (7-19) to obtain the
simple result
^(P.2»o) = f 7 1 3 5 <2<P’ P'|F|P» P'> - <P'> Pl^|P> P '» - Xp
Jcp><0
(7-23)
F IG U R E 7-5 Contributions to E within the Hartree-Fock approxi

mation. The electron lines include the self-consistently determined
self-energy E.
If the poles of G0(pf p 0) are to be unaffected by E, we must

require the self-consistency condition
E( p, i p) = 0 (7-24)
to be satisfied. Since in our particular problem E is independent
of po, we have the familiar Hartree-Fock relation
J€p' <0 \l7T)

= f ^ 3 3 {2<P’ P 'If Ip-p'> - <p'»p|f|p.p'» (7-25)
determining yp.
While this appears to be a complicated way of phrasing the
H F approximation, the scheme is easily extended to include the
pairing correlations. The idea is simply to generalize the lineari
zation (7-10) to include Hartree-like terms involving (0\c1+c2 + 10>
and <0 |c3c4 | 0 >. The state | 0 > is then to be determined self-
consistently as in the standard Hartree-Fock approach.
At this point one might argue that since the full Hamiltonian
commutes with the total number of particles operator Aop, the
exact eigenstates of H are eigenfunctions of N op. One might then
argue that our approximate state | 0 > should also be chosen to be
an eigenfunction of N op and therefore the Hartree-like terms
<0 |c1 +c2 +| 0 ) and <0 |c3c4 | 0 > vanish identically. We can make
two counter arguments. First, in the limit of a large system the
ground states of the N 0 and the N 0 + v particle system are
degenerate for |v| « N 0, if the origin of energy is shifted by
Field-T heoretic M e th o d s A p p lie d to Su p e rco n d u ctiv ity 173
measuring the kinetic energies relative to the exact chemical

potential fi. Therefore, we are free to choose the ground state
of H as a linear combination of these degenerate states, and
therefore the ground state (or more generally, the eigenstates) of
H need not be chosen to be eigenstates of N op. Furthermore,
even if the exact ground states describing different numbers of
particles were not degenerate, one might be able to obtain with a
given amount of effort a more accurate ground-state energy and
excitation spectrum if the approximate states were not required
to be eigenfunctions of N op. We saw an illustration of this in the
BCS theory. In any event, one can view the generalized Hartree
scheme as one which defines a zero-order Hamiltonian H 0' with
the difference between the true Hamiltonian and H 0' remainng
as a perturbation to be treated by other means. The essential
point is that it is simpler to include the nonanalytic pairing
correlations in zero-order (and violate the requirement that | 0 >
be an eigenfunction of N op) than to require the states to be
strict eigenfunctions of N op at each state of the calculation.
A similar situation occurs in nuclear structure theory, where in
lowest order one introduces a nonspherical Hartree potential
which gives a ground state that is not an eigenfunction of angular
momentum.
As in the HF scheme, an alternate way of viewing this
generalized HF approximation is to consider the modified zero-
order Hamiltonian
Ho = H 0 -f (H x + H 0) - fxN (7 -2 6 )
where the added term H # is of the form

# * = 2 + H.c.} (7 -2 7 )
k
Here one desires to describe pairing correlations between k f and

—k \ . Other pairings can be handled in a similar manner.
Owing to the presence of H 0, H 0' is no longer a one-particle opera
tor in the conventional sense, a fact which complicates a perturba
tion treatment of the modified interaction Hamiltonian,
ff.n t = tfin t - (Hx + H 0) (7 -2 8 )
Nambu discovered a neat way of getting around this formal

complication. He pointed out that if one introduces a two-
component (spinor) field operator
^ = L CfcT♦) ^ ^ = c-*< + (7-29)

and its Hermitian adjoint
v k + = (ck, + , c _ k l) = 8kk,1, {V k, Wk.) = 0 (7-30)
{Wk, Wk. +}
H q can be formally written as a one-body operator in the xF k

field. To see this we introduce the four Pauli matrices
-C 3 H,° '3 -G -3 *-G (7-31)3

from which it follows that the Hermitian operators xF k + rllf /k are
given by
= +c_■ki + + c - k i c k t (7-32a)
+
V k = —i[ckf ~ c - k i ckt]
c - k i + (7-32b)
+
V k ■» f c t + n _ k j - 1
T *V k (7-32c)
V k = »kt - n _ kl + 1 (7-32d)
These relations allow one to write H 0' as
Ho = 2k Pk +[*kT3
X + 4>kiTi + j>k2 T2 W k
k
+ 2 *k(7-33)
where <f>kl and <f>k2 are the real and imaginary parts, respectively,
of <f>k and €k = €k + Xk ~~ P as before. The last term in (7-33)
is included to compensate for the —1 occurring in (7-32c).We
can easily remove this annoying (infinite) term once we go over
to the Green’s function scheme. The form (7-33) of H 0' is just
that required to make the Feynman-Dyson perturbation series
rules work if H int takes the form of a one- and/or two-body
potential in the xF k language. One readily finds that H ifnt

takes the form
tflnt = i 2
k,k,’ q
<k + q,k' - q | F | k ,k ')(^ +(I + T n ) 3
X -q+r3X
Pk') ~~ X
f/k+(XkT3
2 + (l)klrl + <f>k2r2)^k
(7 -3 4 a )
if the matrix elements of V satisfy the symmetry requirements
< k 4, k 2|F|k3, k 4> = < — k 3, - k 4|F| - k 4, - k 2>
= < — k 3, k 2|F| - k i , ^ ) = < k l5 — k 4|F|k3, - k 2>
(7 -3 4 b )
These symmetry conditions are satisfied for potentials arising
from the electron-phonon and Coulomb interactions. To make
the first term of (7-34a) agree precisely with H int, one must
include extra terms, as in (7-33); however, these terms will be
automatically included by our choice of G0, and we shall disregard
them for now.
To determine the chemical potential p, and the Hartree
fields Xk and <f>k we imitate the Hartree-Fock treatment given
above. The one-particle Green’s function (matrix) is defined as
< W p , t) = — i< 0 | T{Wva(t)Wv^ + ( 0 )} 10> (7 -3 5 )
where | 0 > is the ground state of H 0' for an average number of
electrons N 0 and
Wv(t) = eiHo'txFp(0)e~ iHo'1 (7 -3 6 )
The quantity
Gou(p, t) = -i< 0 |r{ cpt(<)Cpt +(0)}|0> (7-37)
is the Green’s function for spin-up electrons, while
G „ 2 2 (P ,0 = - » < 0 | r { c - Pi + ( O c - p i (0 )}|0 > (7 -3 8 )
can be thought of as the Green’s function for spin-down holes.
The off-diagonal matrix elements of G
G oia(P ,<) = - * < 0 |T { cp T(<)c _ p J(0 )}|0 > (7 -3 9 a )
G oai(P. t) = -» < 0 | T {c_ pi + (Ocpt + (0)}|0> (7 -3 9 b )
are related to the amplitude for subtracting or adding a pair of

particles to the system without creating excitations. In order to
remove the infinite c-number term from H 0' it is convenient to
define G0(p, t = 0 ) as
G0u(p, <= 0) = lim G0 u ( 2 M ) (7-40a)
1-+0-
Gcmtp-1= 0 ) = lim G0 2 2 (p, t) (7-40b)
*-►0 +
With these conditions it follows that the time Fourier trans
form of Cr0 (p, t) is
**0(P>Po) — ~~2 ^~2
p /n » X _ (P o l + €pT3 + <f>plTl + <t>p2 T 2 ) e ,<SPo,3
7 2 7 2 , V £-----
Vo - *p “ 9vl - rp2 + I 6
Since the pole of 6r(p, p 0) for p 0 > 0 gives the quasi-particle
energy E p, we see that E p is given by
E p = (ep2 + <j>pl2 + 4>p2 2 ) 1 '2 (7-42)
More explicitly, one has the various matrix components of G0
given by
r, , x
G0 n(p) = Po2 (Po + j?t p2) e t i, p °• s ,n
(7-43a)
— 2
+ i8
<Nft
1
~^p 0
G022(P) = Po2 „(Po2 -- E «P)e
p2+ iS
(7-43b)
<t>p*
= Po2
V -- VUp E 2 + ii88 - ,7 -4 3 o )
As in the HF approximation we require the total (average)

number of electrons to be N 0. Owing to the limiting forms (7-40),
we have
<0 | y wps| 0 > = 2 (-*)[Goii(j>. t = 0 ) - G022(p, t = 0 )]
P.s p
= 2 ( - *) Tr [t 3 G0(P, t = 0 )] = N 0 (7-44)
V
From (7-41) we find that
- iG 0(p, t = 0) = -i ; 2n
J g 0(p ,'Po)
(Ep fp)f3 <f>Pl t 1 <f>P2 t 2 ^7 . 4 5 ^
2E p
Field-Theoretic M e th o d s A p p lie d to Su p ercon d u ctivity 177
Thus, the restriction determining the chemical potential fi is
where we have used r t2 = 1, Tr r t = 0 , and Tr 1 = 2 . The

reader will recognize (7-46) as the result we found in Chapter 2 .
We turn now to the problem of fixing ank• Since we
can use the Feynman-Dyson rules to calculate the effects of
H'int, we find the lowest-order self-energy contribution, in analogy
with (7-22), is
S (p,^0) = -*r3 J<P, P'l^lP, P'> T r[r 3 G0 (p')]
+*J <P'. Pi F|P. P '> t3G0(1>>3 ^3
- ( X p t 3 + </>P1 T 1 + <£p2T 2 ) (7-47)
The only change in the rules of Chapter 5 is that each vertex
coupling an electron line to an interaction line of a two-body
potential or to a phonon includes a factor of r 3 because the
coupling is of the form W+t3xP [see (7-34a)]. One must also
remember to heep the matrices in their proper order as given by
the graphs, and to interpret a closed loop as implying a trace over
the free matrix scripts, as in the first term of (7-47). By using
(7-45) we can reduce (7-47) to
S(p. Po) = t3 J^ 3 (2 p'l FIp> p'>
- <P', Pi F|p, P ') K - 2 “ * ]
If we require that the quasi-particle energy E v is unaffected to

first order in H[nt, we obtain the self-consistency condition
L(p, Ep) = 0 (7-49)
as a matrix equation. Since the Pauli matrices are linearly

independent, (7-49) requires the coefficients of the r ’s in (7-48)
to be separately zero and we obtain three equations determining
X, and 0 2:
Xp = J ^ { p'|^|p>p') - <p'> P|F|p. P ')K '2
3 2 (7-50a)
(7-50b)
(7-50c)
where the pairing matrix Vpp>is given by
VPV = <p'. - P 'I F Ip> - P> = <p'. p |r|p , p'> (7-51)
[see (7-34b)]. Since the total Hamiltonian is invariant to rota
tions in T-space about the r3-axis (rx and r 2 never enter H ), we
are free to choose phases so that </>2 = 0 and (7-50b) reduces to the
BCS energy-gap equation, where <f>p is identified as the energy-gap
parameter Ap. It is interesting to note that the Hartree-Fock
potential yP is °f the expected form since vp>2 gives the average
occupancy of states p', s. Thus, this generalized HF scheme is
equivalent to the BCS theory discussed in Chapter 2 if yP is
lumped with the energy ep of that chapter.
To make connection with the spectral weight function ^4(p, co)
of Chapter 5 , we note that G011(p) can be written
where
Field-Theoretic M e th o d s A p p lie d to Su p e rc o n d u c tiv ity 179
as usual. Since we are measuring all energies relative to /i, we

have from (5-41)
A( p, = - - sgn
oj) 2 7T Im G011(p, co)
co T _
= V ( , - E p) + v 8 (0 , + E p)
8 0 p 2 (7-54)
The physical interpretation of the two peaks in A(p, co) was given
in connection with (5-53) and will not be repeated here.
Another way of viewing the generalized HF scheme, which is
easily extended to include retardation and damping effects, etc.,
is that of self-consistent perturbation theory. In this approach
one calculates Z(^p) as a perturbation series in which one uses one-
particle propagators which themselves include the self-energy
being calculated. Therefore, one obtains an approximate integral
equation determining Z. In carrying out the procedure one must
be careful not to double count graphs. If we continue to use the
Nambu notation, the most general form for 'L(p) is
H(p) = [1 - Z (p )]p 01 + X(P)rs + HV)?! + f c p ) r (7-55) 2
We again choose phases so that the coefficient of r 2 is equal to

zero. In contrast to the above treatment, the coefficient of t 1
need not be real; in fact, the imaginary part of <f> contributes to

the damping rate of quasi-particles, as we shall see. Furthermore,
the quantities Z, y, and <f>in (7-55) are functions of the four-momen
tum (p, p 0) rather than the three-momentum p as in (7-50).
This generality is required to treat retardation effects as we did
in Chapter 6 for the normal state. The generalized HF scheme
then corresponds to first-order self-consistent perturbation theory
since within this approximation Z (7-55) is given by
(7-56)
(7-57)
and (7-55) we have

Z (P)Po + H p )t3 + 4>iP)î
*w - -[Z(p)p-.r- m p f + is (7'58)
where by G0(^>) we now mean the true free-electron Green’s
function in the Nambu notation,
Go(2>) = Po
—p ~1 €p
/ z T Iy10£ = ^ ol " ePT3 + (7-59)
and 6P is measured relative to /x. As before we use the notation
e(P) = + x(P) (7-60a)
W = IHp)2 + t i p ) 2]112 (7-60b)
Thus, (7-56) represents a set of coupled integral equations de
termining Z, y, and 0. Since the coefficient of 1 vanishes on the
right-hand side (7-56), one finds Z(p) = 1 within this approxima
tion, and the remaining terms in (7-56) reduce to (7-50), as
required.
In the general case, the Nambu formalism is equivalent to a
self-consistent perturbation approach to determine the coefficients
of the Pauli matrices in the expression for £ (7-55).
7-3 Z E R O -T E M P E R A T U R E E X C IT A T IO N S P E C T R U M
We are now in the position of being able to handle damping
and retardation effects in determining the quasi-particle spectrum
of a superconductor at zero temperature. N am bu114 and Eliash-
berg122 have treated the problem by self-consistent perturbation
theory and have retained the lowest-order dressed-phonon and
dressed-Coulomb contributions to Z, as shown in Figure 7-5.
Within this approximation one finds the (matrix) equation
E(p) = i JT 3G(fOT3|^2 f a p - P') + ^ c ( P - p')
(7-61)
determining 27, where G is given by (7-58). From the above
equation one can determine the unknown complex functions
Field-Theoretic M e th o d s A p p lie d to Sup e rcon d u ctiv ity 181
Z(p), x ( p )> and <f>(p)- If there are no further singularities in the

electron-phonon vertex function (other than those accounted for
in the resolution of the pair instability), this integral equation
treats the electron-phonon interaction exactly to order (m/M)112 ~
10 “2. The quantity 'Vc(p — p') is the Coulomb potential screen
ed by the electronic-dielectric function k [see (6-12)]. While in
principle we should construct an equation to determine the
dressed-phonon propagator D, which enters (7-61), numerical
solutions of this equation show that the general features of the
solutions are insensitive to details of the phonon spectrum. Thus,
one can expect to obtain reasonable results with a fairly simple
form for D, which is best obtained from experimentally determined
phonon spectra. Since the phonon frequencies are essentially the
same in the normal and superconducting states,66 one need not
include this small shift (typically So)QlcoQ < 10 ~4) in the calculation
of S. The effective interaction in (7-61)
2A {9pp *y d a p - p ’) + ^ d p - p i (7-62)
is the generalization of the interaction occurring in (6-28), the
only difference being that (7-62) includes transverse phonons as
well as umklapp processes.
In Chapter 6 we saw that in the normal state the phonon
contribution to E could be obtained analytically since 2?ph(p, p Q)
was essentially independent of p in the region where E played a
role in the form of G [see (6-39)]. As we saw, if the three-momen
tum integral is carried out first, the residue at the pole of G is
independent of E so that the integral equation for E is reduced to
a quadrature. The situation is distinctly different for the super
conducting state, since Migdal’s trick of integrating first with
respect to the three-momentum does not lead to an expression
independent of E as in the iV-state. Nevertheless, the trick is
highly useful and reduces (7-61) to a one-dimensional rather than a
four-dimensional integral equation, once the screened Coulomb
interaction i^ c is handled properly. The one-dimensional in
tegral equations can then be solved numerically, as we shall see
below.
To reduce (7-61) to one-dimensional form we begin with the

term coming from the phonon part of the interaction,
Z ph(p) = i J t 3G ( p ') t 3 2 {gpp.,Y D ,{p - p') ^ (7-63)
The right-hand side of this equation is a function of p only
through the momentum transfer |p' — p| (in the absence of crys
talline anisotropy effects). Since the momentum transfer is
averaged over in performing the angular part of the p'-integral,
E ph varies slowly with |p| about p F, the variation being appreciable
when |p| varies by ~ \ p F- Now we shall be interested in
aaph(P, p 0) for energies |_p0| < ojd « E F. Since the phonon propa
gator decreases as l / ( p 0f - Po)2 for | p 0f - p 0\ » it is
clear that the dominant contribution from the ^'-integral comes
from energies p 0' < o>c, where the cutoff energy a>c is several
times the Debye energy a>D « E F. However, this means that
the major contribution to the integral over |p| comes from states
with kinetic energy <tuc, owing to the form of G. Therefore, in
evaluating Z ph we can approximate 2(p, p 0) occurring under the
integral by E ( p F, p 0), i.e., a function of p 0 alone. By using the
variables p\ g, and </>, as in (6-30), we obtain the expression
s,h()>» - ( w w r . iv ° r . *■"
[Z(p0')p0'l -
[Z(Po')Po'? ~ V2- 4>2(Po') + i 8
(*2k
X2 X JO
F 1 d q iV q ifD M ’ Po - Po') (7-64)
where we have used the same approximations as in (6-31) and for
simplicity have assumed particle-hole symmetry to be valid near
the Fermi surface in order to make the r 3-term in Z ph vanish.
We use the integration procedure of Eliashberg122 and break up
D (p - p ) into two parts, Du which is analytic in the upper
half of the g?0'-plane and D l analytic in the lower half-plane [see
the spectral representation of D (5-59)]. For the term Du, the
portion of the ^ 0'-contour originally along the positive real axis
Field-Theoretic M e th o d s A p p lie d to Sup ercon d u ctivity 183
is folded through the upper half-plane back along the negative

real axis as shown in Figure 7-6, the cuts representing the singu
larities of G(p'). Since Du is analytic in the upper half-plane,
it has no discontinuity across the left G-cut. By using the
relation
G(p, Po + i 8) = G*(p, p 0 - i 8) (7-65)
that is, the values of G on opposite sides of the cut are related by
complex conjugation (see Section 5-4), the deformed contour can
be replaced by an integral along the lower side of the cut if G(p)
is replaced by G(p) — G*(p) = 2i Im G(p). Therefore the Du-
piece of 27ph is given by
On performing the e'-integral one obtains the expression
In a similar manner the term coming from D l can be handled

by folding into the lower half-plane the portion of the p 0'-contour
originally along the negative real axis, as shown in Figure 7-7.
co m p le x to — p lan e
—Aq Aq
F IG U R E 7-6 Folded contour for evaluating 27uph.
com plex co-plane
■ M M iiiiiw £ w m m m m^m--------------------
m m m mm,
—
F IG U R E 7-7 Folded contour for evaluating Z'/ph.
Again, Dl has no discontinuity across the right G-cut and one

finds
m r°° Z 'p o 'l -
= M il 1 dp° Re
1*2
l(Z 'P o r - f 2] 1/2
2kk
X 2 F9 d q { g qi } 2D x ‘(q, p 0 - p 0 ') (7-68)
A JO
By sending p 0' -> —Vo in (7-67) and using the fact that Z ( p )
and </>(p) are even functions of p 0, a fact which follows from (7-61),
one finds for |p | ~ p F,
& * ( p ) = X fH p ) + Xuph(p)
= m f dp.' Re {[ ( Z P .)
(7-69)
The interaction kernels K +vh and A _ph are defined by
K ±*h(Po’ P o) = 2 J0F^|0 d w B *te’

1 (7-70)
+ Po + °> -
Po' Vo - Vo + u - i S.
where K _ph is to be used with the 1-component of (7-69) and
A+ph is associated with the -^-component. In (7-70), B A(q, a>)
is the phonon spectral weight function defined by (5-58). As we
shall see below, K +ph plays the role of the phonon portion of the
pairing interaction. In the static limit (p 0 and p 0' —> 0) A +ph
reduces to the form given by Bardeen and Pines,93 if one includes
transverse phonons in their analysis and sets ep = ep> in their
expressions. For general p 0 and p 0\ K +vh differs from their
Field-Theoretic M e th o d s A p p lie d to Sup ercon d uctivity 185
velocity-dependent, nonretarded interaction. K +ph gives the

correct form in the region where the expressions differ, since as
we mentioned above it arises from an equation which is exact to
order (m/M)112 ~ 10-2.
We now turn to the problem of reducing the Coulomb term in
the equation for Z, (7-61), to one-dimensional form.123 Un
fortunately, the potential Y'c(p — p ) does not decrease rapidly
for \p0 — p 0'\ > wD, as is the case for the phonon interaction.
For this reason the p 0'-integral is not limited to the region
|po'| < o>c, as it was above, and the trick of integrating first with
respect to the three-momentum does not work here in a straight
forward way. To get around this complication we would like
to introduce a pseudo-potential which accounts for the Coulomb
interaction in (7-61) outside of the energy interval —coc <
p 0' < wc. This general approach was first discussed by Bogoliu-
bov, Tolmachev, and Shirkov52 and reformulated by Morel and
Anderson124 to treat pairing correlations in He3. To determine
the pseudo-potential we consider the Coulomb part of the electron
self-energy
Zc(p) = i J t 3G( p ^ r ^ c i P ~ P ’) (7’71)
If we define </>c, yc, and (1 — Z)cp 0 to be the coefficients of r l5
r 3, and 1, respectively, in Zc we have
4>c{p) = - *J ( 2 ^ (Z’p 0')2 - e 2 - <f>'2 ~ P>)

(7-72a)
= 1 J
f d4'X), €;
(2tt)4 ( Z ' p 0')2 - e'2 - f 2 ~ P ) (7’72b)
[1 - Z W P , - ■ J g ; ( Z W / _ Pi - 2 _ f . n ( p - V ) (7-720)
For simplicity we assume that i^ c{p - p') is well represented by

a statically screened potential so that it is independent of p 0
and P q . Since the left-hand side of (7-72c) is antisymmetric in
p 0 and the right-hand side is independent of p Q, we find
[1 — Z(p)]c = 0. We also neglect yc since its main effect is to
give an unimportant shift to the chemical potential and change

the effective mass associated with the Bloch states. The latter
effect can be included in the density of states N(e). Therefore
we are interested in the Eq. (7-72a) for </>c, which can be written
as
n p ) _ 2 j " *?£ j | v lm { _ _ 5# _ _ y c(p _
(7-73)
In the reduction we have folded the jo0'-contour as in Figure 7-7
and have used the fact that 2 Im G is the discontinuity across the
6r-cut. Since we want the pseudo-potential Uc to reduce the
^'-interval in (7-73) to A0 -> wc, we define
[i{Z'Por - €-'2 - <r. [ U c{ p ,p ' )

^
(7-74)
For (7-73) and (7-74) to agree, Uc(p ,p ') must satisfy
iU r, p ) = n ( P - p'> + 2 f i t JiS? ^ - J-')
* lm v » - g - « + i8 r ‘(i,' iy ') (1 -75)
where we have simplified this equation by using

Z (p )^ l
for p 0 > wc (7-76)
which follows from the form ofK +ph (7-70) and the integral
equation (7-69). That Uc defined by (7-75) actually leads to the
correct <f>c can be seen by substituting the formal solution for Uc
into (7-74) and rearranging the resulting expression so that it
agrees with (7-73).
To estimate the magnitude of Uc we assume 5-wave pairing
in the superconductor. In this case only the spherical average
of p - p') enters:
i J ^ d P ~ P ')dP = Vc(p ,p ') (7-77)
where /z is the cosine of the angle between p and p'. The pseudo
potential equation (7-75) now becomes
0x >
(7-78b)
0x <
The physical interpretation of the pseudo-potential is clear
from the form of this equation since it is just the equation satisfied
by the 5 -wave part of the ^-matrix for two-particles scattering in
the region outside of —coc —> cuc. It is reasonable that the
effective potential to use between particles near the Fermi surface
is given by the sum of all multiple scatterings of the particles in
the region away from the Fermi surface, plus the Born term for
scattering near the Fermi surface, in agreement with (7-78a).
The pseudo-potential can be obtained explicitly if Vc(p, p ') is
approximated by a factorizable potential:
otherwise (7-79)
where the maximum energy a>m is of order the Fermi energy E F.
While this is a rough approximation, it should give the correct
order of magnitude for C7c, and one finds
U c(p ,p ') = (7-80)
1+ N (0 )V c In
if the density of Bloch states is taken constant for \ep\ < ajm.
This result, first given by Bogoliubov, Tolmachev, and Shirkov,52
shows that the effective Coulomb repulsion to be used near the
Fermi surface is weaker than the true screened Coulomb inter-
action due to the factor [1 + N ( 0) Vc In (wm/wc)], which is typically
of order 2 or 3. Physically, this reduction is associated with
the fact that scatterings far from the Fermi surface lead to a
smaller probability for two electrons being within the range of the
screened Coulomb potential. Therefore, in the region \p0"\ <

wc, the matrix elements of the screened Coulomb interaction
taken between these correlated states are smaller than the
corresponding plane-wave matrix elements.
We are now in a position to perform the three-momentum
integral since the dominant part of this integral comes from
Bloch states for which Uc(p, p') is a constant Uc. The reduction
goes through just as for the phonon contribution and one finally
obtains, for the total 2 ,
Z (p) = N ( 0) J“cdp0' R e{ [(| ^ j\+ * Z ^ K ±(Po>Po') (7-81)
As before i f + is to be used with the r x-component and K _ with
the 1-component, where
K +(Po>Po) = K +Fh(Po>Po) - Uc (7-82a)
K -(P o ,P o ') = K - pt,{ P o ,P o ) (7-82b)
and K ±ph are given by (7-70). Had we included the dynamic
dielectric function /c(q, q0) and not assumed particle-hole sym
metry, the Coulomb interaction would enter K _ as well. These
effects have been discussed by Schrieffer, Scalapino, and Wilkins.78
The integral equations given by (7-81) can be simplified by intro
ducing the energy-gap parameter A, defined by
(7-83)
It is this quantity which most closely corresponds to the BCS
parameter Av. We then obtain the integral equations
which determine A and Z where A0, the value of the gap parameter
at the edge of the gap, is defined by
A o = A(A0) (7-84c)
Numerous approximate calculations have been made in an attem pt
to solve these equations. The first numerical calculations were
performed by Swihart,125 who set Z = 1 and in effect took K +
to be a square-well potential in the variable p 0 — p 0', attempting
to approximate the Bardeen-Pines potential. From his results he
concluded that A (p 0) decreases monotonically and changes sign
as p 0 increases from zero. Morel and Anderson124 and Culler
et al.126 used the correct Eliashberg potential K appearing in
(7-84) but took Z = 1, as before. The former group treated
analytically the case of constant-frequency phonons (an Einstein
spectrum), while the latter group treated the Debye spectrum
(wQoc q). In both cases it was found that A increased initially
with increasing p 0 before changing sign near the average phonon
frequency, the peak being due to the resonant nature of the
Eliashberg interaction.
The most complete calculations at present are those of
Scalapino, Wilkins, and the author.78 In attempting to explain
the tunnel-current anomalies observed in superconducting lead
by Rowell, Anderson, and Thomas,88 and earlier by Giaever,
Hart, and Megerle,87 they assumed that the weight function
determining K ±ph (7-61) is of the form
" (o) ? J T ' 0 "> - ? ( « - („,.)* <7-86>

The Lorentzian functions were adjusted to represent in a gross
manner the phonon density of states for each polarization. The
values = 4.4 x 10“3 ev and aq* = 8.5 x 10“3 were used for
the average transverse and longitudinal frequencies with the half
widths oj21 = 0.75 x 10“3 ev and w2l = 0.5 x 10"3 ev. These
values were chosen to give a rough fit to the phonon spectrum
observed by Brockhouse in lead and to approximately reproduce
the positions of the tunnel-current anomalies in this material.
The coupling strengths wA were chosen to be independent of A, a
reasonable approximation since the dominant part of the interac

tion involves umklapp processes in this case. The value of w was
adjusted to make A0 be 1.34 x 10 _ 3 ev, the value appropriate for
lead. The real and imaginary parts of A(E) are shown in Figure
7-8. Notice that Re A = Ax goes through a maximum as E
increases toward a>/ 4- A0 or to/ -f A0, reflecting the resonance in
Arph near these frequencies. The imaginary part o fJ = A2 is small
until p 0 approaches the “ emission threshold” for transverse
phonons ~ to/ -f A0, a second rise appearing near to/ -f A0
because of longitudinal phonon emission. In these calculations
the Coulomb pseudo-potential was set equal to zero, while plots
of A(E) and Z(E) for N(0)Uc = 0.11, a value which is reasonable
for lead, are shown in Figures 7-9a and 7-9b. To compare these
results with experiment, we note that according to (3-43a) and
(3-44a) the theoretical tunneling density of states is given by
im = - 1f . d“ Ime<k’E) =Re W - ,7'86)

where we have used the relation
G(k, E) = Gn (jfc, E) =
1 E + (7-87)
Z(E) £2 _ £2 _ A2(E) + ig
for G where e = ek/Z(E). The tunneling density of states is
plotted using the above solutions for A (E ), along with the experi
mental results of Rowell, Anderson, and Thomas. The agree
ment is remarkably good in spite of the simple model of the phonon
spectrum and the electron-phonon coupling used. The gross
structure of N T(E) can be understood by expanding (7-86) to
first order in J 2:
N t (E) , , [(R eJ)2 — (Im J ) 2]
i m = + ----------- 2 w ( 7 ' 8 8 )
As pointed out above, just below the phonon-emission thresholds

Re A increases with increasing E, causing a tendency for N T to
rise, as seen by the knee just below (E — J 0)/oj/ = 1. After the
threshold, Re A decreases while Im A becomes large, thereby
F IG U R E 7-8 The real and imaginary parts of the energy-gap parameter

A plotted as a function of energy for lead, setting the Coulom b pseudo
potential equal to zero.
producing a sharp drop in N T. A similar situation obtains near

the longitudinal phonon emission threshold. Thus a peak in the
phonon density of states is reflected by a knee or peak in N T
followed by a rapid drop in this function. A good deal of detailed
information about the phonon density of states can be gained in
this manner from the I - V tunneling curves. In particular, van
Hove singularities in the phonon spectrum along with more general
singularities are reflected in d2//d F 2. These have been discussed
by Scalapino and Anderson.127
It is interesting to note that the electron-phonon coupling is
so strong for lead that the quasi-particle picture is meaningless
over much of the energy spectrum. Nevertheless, the Green’s
function approach is sufficiently powerful and simple to allow
this problem to be treated in detail. Furthermore, one finds
structure in measurable quantities, e.g., N T(E ), over energy inter
vals which are small compared to the level width one would calcu
late in perturbation theory. This is due to the fact that the
spectral weight function ^4(p, oj) is distinctly non-Lorentzian in
form in this case.
(b)
F IG U R E 7-9 (a) The same as Figure 7-8 except the Coulom b pseudo
potential is set equal to a value roughly appropriate for lead, N(0)Uc =
0.11, (b) The real and imaginary parts of the renormalization function
Z(w) for this case, (c) Tunneling density of states.
Thus far we have said nothing about the isotope effect

emerging from these calculations. Estimates of the isotope-effect
coefficient
_3 1 n J0
“ = -T bM ( 7 ' 8 9 )
by Morel and Anderson124 are in reasonably good agreement with

experiment for a large group of superconductors, with a few
exceptions. Within their approximations they found the small
value ~ 0.35 for zinc (a very weak coupling superconductor), while
more complete calculations by Garland24 removed the discrep
ancy. In addition, Garland argues that in “ dirty” transition
metals (i.e., those which are sufficiently impure that the single
particle states are best represented as admixtures of s- and (in
states) the large d-band density of states strongly affects the
effective Coulomb potential to be used in solving the gap equa
tion and that very large deviations from a = \ can be obtained
even though the attraction is due to phonon exchange. This is a
possible explanation of the vanishing isotope effect in ruthenium
and the small value a ~ 0.2 for osmium. For highly purified tran
sition metals, an s-s Colomb interaction made attractive by d-
or /-band polarization effects cannot be ruled out as a possible
mechanism leading to superconductivity without an isotope effect;
however, this mechanism appears to be highly unlikely from both
a theoretical and experimental point of view.
7-4 E X T E N S IO N T O FIN ITE T E M P E R A T U R E

For finite-temperature problems one would like to define the
one-particle Green’s function as the statistical average of Green’s
functions defined for the exact excited states of the system.91c> 99b
More generally, it is convenient to include an ensemble average
over systems with different total number of particles and use the
grand canonical ensemble as the appropriate weighting factor.
Thus we define
G s (t u r2> T> p, m) = -* Tr [«(/?, r, r)«£s+(r2) 0)}] (7-90)
where the grand canonical density matrix is given by

e ~ G(H — u N ) i
U(P> /*) = rpr e - 0{H- uN) P = JTTp (7"9I)
and the T-product is defined as before. We suppress the j3- and
^-dependence of G. For convenience we define H — pN = A and
let the time dependence develop by K rather than H,
i/js(r, t) = e iKrJjs(r, 0)e~iKx (7-92)
For a translationally invariant system we are interested in the
Green’s function
Gt(p, r) = - t Tr [«(j8, h.)T{ cpi( t )cp, +(0)}] (7-93)
To gain insight into this function we note that a spectral repre
sentation for the time Fourier transform of G(p, t) can be derived
just as in Chapter 5 and one finds
J -» p 0 - O ) + l S J - „ ^0 + a>-t8
where the spectral functions are given by
P<+>(P, co) = 2 Mn|<m |cp+ |w>|2 - E n - W) (7-95a)
n, m
p<->(p, <o) = 2 Mn|<m lcp|w>|2 S(^m - E n -w) (7-95b)

71, 771
The states |n) are the eigenstates

K\n} = (H — pN)\n) = E n\n> (7-96a)
and un is the diagonal element of the density matrix,
u, - ^ e~0En (7.96b)
771
It follows from (7-95) and (7-96) that p(+) and p<_) are related by
p<->(p,co) = e « V +)(P. - « ) (7‘97)
Field-T heoretic M e th o d s A p p lie d to Su p e rco n d u ctiv ity 195
as can be seen by interchanging dummy indices in (7-95b).

Thus, the spectral representation (7-94) can be rewritten as
Re G(p, p 0) = P [ — — dco
J - oo Po — to
(7-98a)
and
Im G(p, p 0) = p 0) t a nh^ (7-98b)
The spectral weight function A(p, p 0) is defined by
A( p,co) = )(p, co) + p(->(p, - t o ) =
Pi+ +)(p,to)(l + e-*» ) p(
and reduces to the corresponding function introduced in Chapter 5

at zero temperature. As opposed to the zero-temperature case,
however, p( +) and pi ~) are in general finite over the entire co-axis,
since the system may lower its energy relative to p. when a particle
is added to or subtracted from the system if T # 0. Therefore,
the simple picture that the positive and negative frequency
parts of A( p, co) correspond to particle- and hole-injection pro
cesses, respectively, no longer holds if T ^ 0.
As for T = 0, A( p, co) satisfies the sum rule
This is seen by calculating

r. A( p, co) d c o — 1 (7-99a)
/•oo /•00
A(p, t o ) d w =[/>< + ) ( p , o i ) + / ( _ ) ( p , a> )] ( i ai
3
J—
00 J— 00
= 2 w„{<n|cpcp+|n> + <n|cp+cp|w»
n, m
= Tr u = 1 (7-99b)
where we have used the anticommutation relations for the c’s.
As pointed out in Chapter 5, the functions A, p( +), and p(~ )
are related to the functions A BK, G >, and G <, discussed by
Kadanoff and Baym,91c by
A bk{]>, a>) = 2ttA( p , w) (7-100a)
G > (p , to) = 2 n p ( + )( p , w ) (7-100b)
( ^ ( p , to) = 277/)( - ) ( p , - t o ) (7-100c)
The highly automatic nature of the zero-temperature per

turbation series for G can be carried over to finite temperature by
an elegantly simple procedure introduced by Abrikosov, Gor’kov,
and Dzyaloshinskii,128 who extended the pioneering work of
Matsubara in this area. Similar techniques were advanced in
dependently by Martin and Schwinger99b in their fundamental
work on Green’s function techniques in the many-body problem.
The basic result of their development is that G(p, p 0) can be de
termined by the analytic continuation of a Green’s function
^(p, iajn) defined over a discrete set of pure imaginary frequencies
io)n. The function ^(p, iajn) can be constructed by the usual
Feynman-Dyson rules if all energy variables p 0 associated with
fermion lines occurring in the zero-temperature expansion are
formally replaced by
i(2n + 1)^ .
Poî<x)n = jg (n = integer) (7-101)
and the corresponding energy integrals are replaced by
The phonon energy variables are replaced by

= i2U7T (7-103)
/7 1AQ\
that is, an even multiple of 7n//? (while the electronic energies are
replaced by odd multiples of 7ri/j3). The difference between the
boson and fermion rules comes from the definition of the T-product.
In analogy with (7-102), g0-integrals are replaced by
r. i H . l « » •*>
The zero-order propagators to be used in the perturbation series
Once the function ^(p, icon) defined on the integers n = 0, ± 1,

± 2, . . . is determined, this function is then to be analytically
continued to the real-energv axis iajn - » p 0 in such a way that
^(p, z) is a bounded function as z -> oo in the complex plane.
In practice the continuation usually consists of setting functions
such as e0lO)n and e0iv™equal to - 1 and + 1, respectively, and re
placing the remaining imaginary energies by real energies, i.e.,
icon —> p 0 (or %vm -> q0). Once the continued function ^(p, p Q) has
been determined, the actual Green’s function G(p, p 0) is given by
Re ff(p, p 0) = ^(P> Po) (7-106a)
and
Im 6r(p, Po) = tanh j Im ^(P> Vo + i 8) (7-106b)
where 8 = 0 + as usual.
The procedure is made particularly convenient by using the
Poisson summation formulas
-FM dco
i .
,00)+ 1
27n J c e + 1 dw 0(0 = (2?l ^ 1)7T] (7-107)
where the contour c encircles the imaginary co-axis as shown in
Figure 7-10 and does not enclose any singularities of F(co).
This relation follows from Cauchy’s theorem since the denominators
(e003 + 1) and (e~0co + 1) have first-order poles at co = icon. For
sums involving phonon energies one has
F{v)
1= — co v c
,0V - 1 dv
p f
27ti ] c e~0v
dv
(7-108)
The trick is then to expand the co- or y-contour to infinity, picking
up contributions from singularities of F ( co) as the contour moves
outward. In most cases the remaining integral around the
F IG U R E 7-10 The contour of integration to be used with finite-

temperature calculations.
infinite circle vanishes because of the vanishing of F on this

contour.
To illustrate the procedure we calculate G for the Frohlich
model of the electron-phonon system in the normal state, keeping
only the lowest-order contribution to E. For zero temperature
we had
^(P, Po) = * J |0«|20o(p'. Po') A)(q, Po - P o ) (7-109)
where q = p' — p [see (6-27)]. To calculate we use the pre
scription given above to transcribe this expression for E into one
for the self-energy E appropriate to ^ ; thus
_ 1 00 f d?p'
E( p, iwn) = - ^ 2 J (2^)5 1^12^ o(p,’ iftV >ô(q. ~ »«v)
(7-110)
Field-Theoretic M e th o d s A p p lie d to Sup e rcon d u ctiv ity 199
By using (7-107) the w'-sum can be written as

1 00
2 ô(p') *«v)ô(q> ia>n -
P n' = - oo
= ^ J W , w)@0{q, iwn - oj)f(to) dco (7-111)
where f(w ) = l/(e5w + 1 ) is the Fermi function. As the o>-

contour is expanded to infinity one picks up contributions to the
integral from the poles of and at a> = ep, and to = ia>n ± Qq,
respectively. (In the Frohlich model Qq is assumed to include
electron-screening effects so that Qq oc q.) The deformed contour
is shown in Figure 7-11 for €p, and a>n > 0. Notice that the
poles are encircled in the clockwise direction over the entire
plane. By Cauchy’s theorem one obtains
-2a«/M _ f(j< on + a 9) , /(**, - a,) ,7_112.
(ep. - iajn)2 - Q 2 io)n + Q q - ep. iwn - Q q - ep.
F IG U R E 7-11 The deformed version of the contour shown in Figure

7-10. Notice that all poles are encircled in the clockwise direction.
The poles shown are for the lowest-order phonon contribution to the
electron self-energy in the normal state.
for the sum (7-111). In order that ^(p, z) be bounded for large
z we must set
f(icon ± Qq)= eiPa}ne±pn(i + j = i _ e ± w q (7“113)
before replacing ia>n by 2 . Therefore, the properly continued
self-energy is given by
As T —> 0, the first term in the brackets contributes only for

€p' > 0, that is, states above the Fermi surface. The second
term enters for ep, < 0, in agreement with our earlier result
(5-78). The physically interesting Green’s function G can now
be obtained from the relations (7-112) and Dyson’s equation for
S?(p, z) = ----------L_=.----- (7-115)

2 - €p - Z(p, 2)
From this example it is clear that finite-temperature calculations
require essentially the same amount of effort as the corresponding
zero-temperature calculations, except that the resulting momen
tum integrals are more difficult to perform at finite temperature.
The frequency sum scheme also applies in the Nambu
formalism. As a simple example, we consider the finite tempera
ture generalization of (7-56) in which one is concerned with a
nonretarded two-body potential V. The quantities E and ^ are
now 2 x 2 matrices. For simplicity we keep only the energy-gap
portion of E (i.e., the ^-component) and obtain
^(p, ia>n) s p = 1J ^3 <p\ p| F|p, p'>
4>* (7-116)
„ ..4 „ (*«v)2 - V 2 - K 2
The Poisson summation formula (7-113) allows us to convert the
w'-sum to the form
where the contour encircles the imaginary cu-axis. By expanding

the contour to infinity we pick up residues from poles at w =
+ (tp/2 + <f>p 2)112 = ± E P' as shown in Figure 7-12 and obtain
A - [ / ( £ P0 - f ( - E p.)] = - A _ [ l - 2f ( E p,)]
(7-118)
for the sum (7-117). Therefore the energy-gap equation becomes
where Vpp> = <p\ p| F|p, p'>, in agreement with the result of BCS
and of the linearized equation of motion treatment (2-76). The
finite-temperature generalization of the retarded interaction
problem discussed in Section 7-3 is also straightforward to carry
out by these techniques.78
Although we have emphasized the calculation of G at finite
temperature, the phonon Green’s function follows in a similar
manner.
As a general conclusion of this chapter, one can say that, on
refining the BCS theory to include noninstantaneous interactions
and damping effects, only details of the quasi-particle spectrum
are altered. In particular, the energy gap in the elementary
(*}
E v. E v.
F IG U R E 7-12 The poles contributing to the electron self-energy In

the superconducting state if one assumes an instantaneous tw o-body
potential.
excitation spectrum continues to exist within the above approxi

mations. The effective density of quasi-particle states in energy
is well approximated by the simple BCS model near the edge of
the gap with small deviations of order or less than 1 to 5 per cent
occurring in the vicinity of the Debye frequency. In the next
chapter we shall see that collective effects do not alter these
conclusions; however, collective modes can exist in the energy
gap under suitable conditions.
CHAPTER 8
ELECTROMAGNETIC
PROPERTIES OF
SUPERCONDUCTORS
Among the many surprising properties of superconductors, their

response to externally applied electric and magnetic fields is
the most striking. In 1911, while measuring the electrical re
sistivity of metals at liquid helium temperature, Kamerlingh
Onnes7 discovered that certain metals passed into a radically
new phase in which the voltage drop across the metallic specimen
vanished even though a finite current was flowing through it.
He characterized the new state as one of infinite electrical con
ductivity or “ superconductivity.” Equally striking was Meissner
and Ochenfeld’s 6 discovery that under ideal conditions a super
conductor is a perfect diamagnet, that is, the magnetic field
strength B vanishes within the bulk of the superconductor.
8-1 LONDON R IG ID IT Y
In Chapter 1 we gave a qualitative discussion of the origin
of these unique properties. Here we take a more formal approach
203
204 T h e o ry o f Su p erco n d u ctiv ity
and show how these effects follow from microscopic considerations.

While the mathematical aspects of the discussion become involved
at points, the underlying physics which accounts for the phe
nomena was clearly stated by London5 in 1935. He suggested
that the wave function Ws of the “ superfluid” electrons is “ rigid”
or “ stiff” with respect to perturbations due to the presence of a
weak magnetic field. Then, as in the problem of diamagnetism
of atoms, the vector potential leads to a finite-current density
since = 0, owing to the rigidity of Ws. This induced current

then gives rise to a magnetic field which screens out the external
field and leads to perfect diamagnetism in a large system.
The microscopic implications of London’s interpretation of
the Meissner effect can be seen in the following manner. Since
Maxwell’s equations ensure that the magnetic field is necessarily
a transverse field (V • B = 0), the magnetic perturbation H'
only affects the transverse excitations of the system. If XF S is to
be essentially unaffected by this (weak) perturbation, the sum of
the squares of the first-order perturbation series amplitudes
82
( - )
must be vanishingly small. Here Wa is the state with transverse

excitation a present. Presumably, this anomalously small value
of the sum will occur if the matrix elements <¥/a|//'|¥ /s> tend to
zero while the excitation energies E a — E s remain finite. Clearly,
this will not be the situation for magnetic fields which vary
rapidly in space. Such fields create excitations involving elec
tronic states far from the Fermi surface. These states are
presumably unaffected by superconducting correlations and there
fore lead to finite matrix elements as in the normal state. Fortu
nately, in establishing the Meissner effect, one only requires the
response of the system to magnetic fields which vary slowly in
space. In this limit only electronic states near the Fermi surface
enter, and there is no reason to expect the matrix elements not to
Electrom agn etic Pro pe rties o f Su p e rco n d u cto rs 205
vanish in this case. Therefore, London’s interpretation of the

Meissner effect leads one to suspect that (1) the matrix elements
for creating transverse excitations from the superfluid by a
magnetic field tend to zero for fields which vary slowly in space;
(2) there is an energy gap in the transverse excitation spectrum
of the superfluid. As we shall see below, these conditions are
satisfied by the BCS theory. It is possible, however, that the
matrix elements and the energy denominators both vanish in the
long wavelength limit, in such a way as to give a finite sum. If
this sum does not exactly cancel the diamagnetic current in this
limit, a Meissner effect is still obtained, as in superconductors
with I / 0 pairing170 and “ gapless” superconductors.172 The
essential difference between the normal and superconducting
states in metals is that the paramagnetic and diamagnetic currents
do not exactly cancel in the long wavelength limit in the latter.
On the basis of these arguments one might wonder why an
insulator, which has as an energy gap for creating transverse
electronic excitations, is not also a perfect diamagnet. The point
here is that the energy gap arises from the one-body crystal
potential in this case, rather than the effective electron-electron
interaction. One can derive a sum rule which shows that for
insulators the wave function shifts just enough to make cancel
the diamagnetic term eA/c in (8-1) except for a weak diamagnetism,
as in normal metals.
If one can explain the Meissner effect, the “ infinite con
ductivity” observed by Onnes can also be explained. This
follows because one can show the currents flowing in his con
figuration (i.e., a superconducting section in an otherwise normal
circuit) were diamagnetic in the superconducting section.1 That
is, the currents in the superconductor were due to electrons
described by a wave function which was essentially the same as
in the absence of the current. The finite current then arose
self-consistently from the magnetic field which was generated
by the current itself.
In addition to the Meissner effect, one must understand the
stability of persistent currents in a multiply connected body,
206 T h e o ry o f S u p e rc o n d u c tiv ity
for example, a superconducting ring. While the Meissner effect

plays a role in the details of the phenomenon, the currents are not
primarily diamagnetic in origin in this case. On the contrary,
the wave function for the current-carrying state differs greatly
from that in the absence of currents, in contrast with the situation
described above. However, the effect is again due to a “ rigidity”
of the wave function with respect to all fluctuations which occur
with a finite thermodynamic probability. That is, essentially all
fluctuations lead to states of higher free energy and therefore
regress without leading to decay of the current.
8-2 W E A K -F IE L D R E S P O N S E
In the beginning of Chapter 2 we argued that transverse
electromagnetic fields need not be included directly in calculating
the detailed pairing interactions which bring about superconduc
tivity. Their effect can be taken into account in terms of a
space- and time-dependent average field which is calculated self-
consistentlv from the external field and the currents flowing in the
material. While the externally applied magnetic field generally
represents a large perturbation on the system, the induced field
arising from the supercurrents cancels the external field over most
of the material, as we know from the Meissner effect. Therefore,
the net field acts only very near the surface and can often be
treated as a weak perturbation on the system as a whole. Thus,
we shall formally treat the total transverse electromagnetic field
as an externally applied field and solve for self-consistency as a
separate problem.
As we have seen above, the Coulomb potential plays an es
sential role in the pairing theory. It cannot be treated by the
self-consistent field scheme we use for the transverse field and
therefore we include the total Coulomb interaction in the zero-
order Hamiltonian.
We begin by considering a simply connected bulk super
conductor of unit volume in the presence of a weak externally
applied electromagnetic field described by the vector and scalar
Electrom agnetic P roperties of Su p e rco n d u cto rs 207
potentials A (r, /) and cp(r,t), respectively. As usual, we use

periodic boundary conditions. For convenience, we write
jA ,(r ) 0* = i - 1,2,3)
A- {X) ~ W (x) („ - 0) (8' 3)
where x = (r, t). To first order in A u, the coupling of the electrons
to the electromagnetic field is
H ” = ~ l j ' E j u p(x )A u(x ) d3r
=- 1 J [ jp(z) ’ A(x) - pe(x)c<p{x)] d 3r (8-4)

where we use the metric (1, 1, 1, —1) in fi-sums for fx — 1, 2, 3,
and 0, respectively. We call H p the paramagnetic coupling.
The paramagnetic four-current is defined by
jup(x) (p = i = 1, 2, 3) (8-5)
Pe(x) = - e ^ <Ps+(x )>Ps(x ) = ~ep{x) {p. = 0)
the (1, 2, 3) components giving the electronic current density

operator in the absence of A and the last component being the
electronic charge density operator. The physical current density
Cu{x) in the presence of A is the sum
j uix ) = j A x ) + j A x ) (8-6)
where the diamagnetic current density j d is given by
• d(~\
J u ( x ) = <mc
Pe(x )î(x ) (n = i = 1, 2, 3)
(8-7)
I 0 (/x = 0)
The full coupling of the electrons to the perturbing electromagnetic
field is then
H' = H> + H d
where the diamagnetic coupling is defined by
nd = -
J
f p»{x) i=i
2 A >2(x ) d3r (8-8)
Therefore, the total system Hamiltonian is

& = H + Hf
If we work in an interaction representation where H' is
taken to be the perturbation, and assume that ^ 0 as £ —> —oo,
the ground state of the system in the presence of A evolves in
time according to
= Texp - i H'(t')dt' |0> = U(t, - oo)|0> (8-9)
—00
Here |0> is the ground state of H and all quantities are expressed
in the interaction representation. Therefore, the expectation
value of the current density in the state | <£(£)> is given by
JM ) = 0|3W > = <0| U +(t, - oo)ju(r, t)U{t, —co)|0>
( 8 - 10)
Since we are interested in the terms of J u which are first order in
A u we have
The zeroth-order terms in J u vanish except for the average

electronic-charge density ( j 0(x)), which does not interest us here.
By using the expressions (8-4) and (8-7) we find that the linear
response J u and the externally applied potential A u are related
by a nonlocal kernel K uv:
J M) = - £ 2 J A'«v(r, t; r', t')Av(r', V) d 3r' dt' (8-12)
where the spatial integral runs over the unit volume and the time
integral extends from - oo to oo. The electromagnetic response
kernel K uv is given by
Electrom agn etic P ro pe rties o f S u p e rco n d u cto rs 209
where the theta function is defined by
- *'>- {o V< n ( 8 ' 1 3 b |
If the system is translationally invariant, K uv depends only

on the difference x - x' = (r - r', t - t'). In this case it is
convenient to work with the spatial Fourier transform of K uv
defined by
Jf„(q, t - t') = j K „ ( x - , x ' ) e - d * r d ar'
= - 7 ? < o | t i / ( q . <), j v p( - q . «')]|o>0(< - n
+ 8(t - t') 8UV(1 - 8V, 0) (8-14)

where n is the number of electrons per unit volume. Since the
diamagnetic (second) term in (8-14) is known explicitly, we
concentrate on the paramagnetic (first) term in this expression,
and define
RuM, r) = -;< 0 |j/(q , t), j P ( - q, 0)]|0>^(r) (8-15)
If the ground-state wave function were “ rigid” with respect to
all perturbations (rather than only those which lead to transverse
excitations) B uv would be identically zero and (8-12) would reduce
to London’s equation
TYIC
(#* = * = 1, 2, 3) (8-16)
This relation is clearly not gauge-invariant since the predicted
current depends upon the choice of gauge. In London’s equation,
only the transverse part of A is to be used1 and therefore J is
properly gauge-invariant. Since the longitudinal part of A
couples to longitudinal excitations, the wave function is not
“ rigid” with respect to this type of perturbation and the para
magnetic term does not vanish in this case. In fact, if A is purely
a gauge potential, the paramagnetic and diamagnetic terms
exactly cancel as required by gauge invariance. In carrying out
an approximate evaluation of K uv, one may be able to accurately

treat only excitations which enter the transverse response of the
system. In this case the longitudinal part of the paramagnetic and
diamagnetic currents will not cancel in general and the resultant
current will not be manifestly gauge invariant. Nevertheless, if
one recognizes the difficulty and uses only that part of K uv which
is accurately calculated, correct physical predictions would be
obtained for transverse fields. This is exactly the situation we
shall meet when K uv is evaluated within the pairing (BCS)
approximation. The inclusion of longitudinal collective modes
or superfluid flow then restores gauge invariance by correcting the
longitudinal part of the paramagnetic term.
In our discussion thus far, we have always dealt with time-
ordered products of operators rather than retarded commu
tators of operators as appear in (8-15). It is the former that we
can more readily handle by the Green’s function scheme. Fortu
nately, R uv can be expressed in terms of a time-ordered product
of current densities if one works with the time Fourier transforms
of these quantities. To see this we note that R uv(q, q0) defined by
= J°^ K M ,
K M , r) <lo)e-iq°l ^ (8-17)
can be expressed in the spectral form
k m , ? 0) = r
J-oo?0 —tu + l °§ <8-i8 >
The spectral weight function Cuv (q, a>) is given by

c wv(q>w) = 2 <°|j/(q)lw><wljvp(-q)|o>
n
&(En - e 0 - M
-2 <0Uvp(-q)|w><w|jHp(q)|°> H E n - E 0 + w)
where " (8-19)
H\n) = E n\n> (8-20)
This spectral representation can be checked by inserting the
complete set of intermediate states \ri) between the operators
in (8-15) and comparing this result with the expression given by
combining (8-17) to (8-19).
Consider the corresponding time-ordered product expression
P «,(q> r) = p uv(q, q0)e-"o' (8-22)

is given by the spectral representation
(8-23)
as one can check by direct calculation. By comparing the spectral
forms (8-18) and (8-23) we see that in the case that Cuv(q, co) is real,
the real parts of P uv and R uv are identical while the imaginary
parts differ by a minus sign for q0 < 0; thus
Re P uv(q, q0) = Re JR(1V(q, q0) (8-24a)
Im P uv(q, q0) = sgn q0 Im ^ v(q, q0) (8-24b)
More generally, the discontinuity of P wv(q, q0) across the cut
determines Cuv from which R llv can be obtained with the use of
(8.18). Therefore, R uv is known once P uv is determined. (Since
the expression for K uv involves only the system in the absence of A,
the operators j u and j up are identical in this case and we shall
often suppress the script p in the operator j up.)
Summarizing the results obtained thus far, we find the
response of the system to a weak externally applied potential
A u(q) = [A(q), c<p{q)] is given by
— 2 K „t(q)At{q) - K u0(q)A0(q) (8-25)

L t= i J
where q = (q, q0). The kernel K uv is given by combining the

expressions (8-14) and (8-15),
the two terms giving rise to the paramagnetic and diamagnetic

currents, respectively. The quantity AL2 = mc2/47me2 is the
square of the London penetration depth. The paramagnetic

kernel R uv is given in terms of the time-ordered quantity
P UM = r)jv( - q , 0)}|0>e*.’ d r (8-27)
by the relations (8-24).
8-3 T H E M E IS S N E R -O C H S E N F E L D EFFECT
As Schafroth has shown,14 the Meissner effect requires that the
transverse part of the kernel K uv remain finite in the long wave
length limit (q —^ 0) for zero frequency (q0 = 0). Now gauge
invariance and charge conservation require that
3
2 Kitfj = 0 (gauge invariance) (8-28a)
j= i
and
^ Qiîj = 0 (charge conservation) (8-28b)
i
for q0 = 0. When these relations are combined with the rotational

invariance of the ground state |0>, it follows that K tJ is of the
form
if(q2) (8-29)
The Meissner effect then requires that
K ( q 2) > 0 as q2 0 (8-30)
since the factor [Si; - <M;/q2] guarantees that K tJ is purely
transverse in this case.
The original BCS calculation of J\was carried out in the
transverse gauge, that is, q • A(q) = 0.In this gauge onlythe
transverse part of K tj is calculated, and one does not try to ensure
that the longitudinal part of K i} vanishes, as required by (8-29).
It is, however, instructive to calculate the entire kernel within
the BCS approximation so that we can understand the role of
collective modes or superfluid flow in giving the correct result for
Electrom agnetic Prope rties o f Su p e rco n d u cto rs 213
the longitudinal part of K . To calculate the paramagnetic portion

of K we require
The current-density operator j(q) is given by the Fourier trans

form of the expression (8-5) and one finds
(8-32)
Therefore P XJ becomes
X <0|T{chs +( r ) c k +q.s(r)ck.+q,s. +(0)chV(0)}|0> (8-33)

We could evaluate this expression within the pairing approxima
tion by replacing the exact ground state by the BCS ground state
(2-33)
(8-34)
k
and expressing the c’s in terms of the quasi-particle operators
by the Bogoliubov-Valatin transformation (2-56). The time
dependence is then approximated by that of free quasi-particles
and the vacuum expectation value is evaluated in the standard
way.
An equivalent procedure, which is more easily generalized
beyond the pairing scheme, is to express (8-31) in terms of the
Nambu field 'F,,.114 One then makes a Hartree factorization of
the expectation value. In the Nambu notation, j(q) takes the
form
(8-35)
and P t; becomes
X < 0 |T { ¥ V (r)in + 1
< ( T ) ^ k. +/ ( 0 ) iq 'k.(0)}|0> (8-36)
Within the Hartree factorization, the expectation value becomes

- T r [<0|T{^k +Q(r)^ k+q + (0)}|0><0|T{^k( - T ) ^ k+(0)}|0>] 8kik.
= Tr [G(k + q, r)G(k, - r)] 8k k, (8-37)
To make connection with Gor’kov’s formulation,121 we note that
in carrying out the trace in (8-37), terms of the form G 11G 11'
and G22G22 correspond to GG' in Gor’kov’s notation, while
G 12G21' and G21G 12 correspond to products of his ^-functions.
Within this Hartree-like approximation, the time Fourier
transform of P XJ is given by
PM = “5 J ( 0 ( k + l)t(k + + tjTr^k W
where q = (q, q0). If the pairing potential is nonretarded, we
saw in Chapter 7 that within the pairing approximation G(A;) is
given by
G(*) = * *0
°,V ~ + Io = (^ 2 + V ) 1'2] (8-39)
[see (7-41)]. Since we are interested in the static Meissner effect,
we set ô = 0 and P XJ reduces to
*>>■ ° » - - 2( 3 ' ( k + i ) , ( k + 1) i i k - q) <8‘4°a)
where the function L( k, q) is defined by

V r® rjb
L ( k, q) = j ^ Tr [G(k + q)G(k)]
_ • f °° dk0 ____ (kg2 + Îce k + Q ^ k ^ k + ci)____ /g 4Qbl
“ 1 } _ x ^ (kg2 - E k2 + i 8 ) ( k 02 - E k + * + i 8)
In the reduction, we have used the relations
T,2 = 1
Tr 1 = 2
Tr r, = 0 = Tr t.t, (t # j) (8-40c)
Electrom agn etic P roperties of Su p e rco n d u cto rs 215
The integral is performed bv closing the contour in the upper

(or lower) half-plane and one finds for L (k , q) the real quantity
L(k, q) = ^ (i - ^ k+EqkE+ k~k-
+q
k+9'\J E-------1-------
k + E k +q
p 2(k,: lr
k -I-
+ nO
E k + E,k+q
q)
(8-41)
where j9 (k, k + q) is the coherence factor we met in Chapter 3.
This is the result of BCS.8, 9
To establish the Meissner effect within the pairing approxi
mation, wenote that L( k, q) -> 0 as q -> 0, owingto the coherence
factor jt?2(k, q) vanishing in this limit and the energy denominator
remaining finite (Ek -I- E k +q ^ 2zJ0). Therefore,
lim P tj(q, 0) = lim B lj(q, 0) = 0 (8-42)
q-» 0 q -+ 0
and the electromagnetic kernel reduces to the London kernel

lim
q -> 0
K ti(q,
0) = LAL 8,, ( i , j = 1, 2, 3) (8-43)
The transverse part of this expression is
lim K ti(q, 0) = 8i’ ~ qiq’ ± (8-44)
q-+o q al
By comparing this result with the general form (8-29) we see that
lim
= -^ > 0 K ( q2) (8-45)
q2 AL
-» 0
which establishes the Meissner effect at zero temperature. Un

fortunately, the longitudinal part of K does not vanish in this
approximation, but is given by (<M;/<72)(1/AL2); however, this un
physical longitudinal response will be eliminated below.
The above derivation emphasizes the role of the energy gap
in bringing about the Meissner effect. Aside from scale factors,
the quantity L ( k, q), given by (8-41), is the square of magnetic
perturbation matrix element taken between the ground state and
a transverse excited state, divided by the excitation energy for
the two quasi-particles which are excited. In the superconducting

state, the matrix element vanishes as q 2 —> 0 [i.e., p 2(k, q) -* 0]
and the energy denominator remains finite, in accordance with
the discussion in the beginning of this chapter. Therefore, only
the diamagnetic term in K uv survives in this long wavelength limit.
In the normal metal L ( k, q) goes over to the conventional result
of second-order perturbation theory:
r
J ^ (k , q) = ]--------------- (8-46)
\€k+q €kI
if the k and k + q are on opposite sides of the Fermi surface, and
zero otherwise, as required by the Pauli principle. As we argued
above, a finite value of P lj(q, 0) arises in the normal metal as
q2 —>■0 despite the fact that most of the matrix elements vanish,
because of the vanishingly small-energy denominator in this
case. If one calculates the magnitude of P tJ one finds that it
almost exactly cancels the diamagnetic term, leaving the weak
Landau diamagnetism of the normal state.
In “ gapless” superconductors172 both the matrix elements
and energy denominators vanish, but the density of states near
the Fermi surface is small enough to ensure that P XJ does not
cancel the diamagnetic term as q -> 0.
To extend this calculation to finite temperature, we use the
prescription discussed in Chapter 7 to convert the zero tempera
ture form (8-38) to one involving the discrete frequency sums.
The only change is that L ( k, q) becomes
L( k, q) = - A 2 Tr [G(k + q, iwn)G(k, icon)] (8-47)

~P n= - oo
where con = (2n 4- 1 )777/8. As before we convert the sum to an

integral using (7-107) and obtain
L ( k, q) = - -l j j £ Tr [G(k + q, co)G(k, w)]f(w) (8-48)

Electrom agnetic P roperties o f Su p ercon d u cto rs 217
where the integral encircles the entire imaginary axis in a counter

clockwise sense and f ( w) = [e003 + I]-1 is the Fermi function.
By expanding the contour to infinity we pick up residues at the
four poles ± E k = ± E and ± E k +q = ± E \ as shown in Figure
8-1. These residues lead to the real expression
+ ( 8 ' 4 9 )
which can be written as

L(k, q) = / 2(k „q)- ■[1 - f ( E k) - f ( E k +J ]
+ k+q
’ F ( k ’ q) U ( E w+J - f ( E k)} (8-50)

~ ^k +q
The coherence factors p 2 and I2 are given by
1 ( M l,
the upper and lower signs applying to p and Z, respectively.

Since we are interested in establishing the Meissner effect at finite
temperature we consider the limit of (8-50) as q -> 0. The first
F IG U R E 8-1 Poles contributing to the static electromagnetic kernel.

term gives the contribution of the “ superfluid” electrons (i.e., the

superfluid component in the two-fluid model), and vanishes in this
limit as it did at zero temperature. The second term gives the
contribution of the thermally excited quasi-particles (i.e., the
“ normal” fluid component) and does not vanish as q —> 0 since
the denominator vanishes in this limit. The essential physical
difference between the two terms is that the superfluid term
involves creation of two quasi-particles, with the minimum excita
tion energy being 2A 0. On the other hand, the normal fluid term
involves scattering of quasi-particles already present and the
excitation energy in this case can be arbitrarily small, as in the
normal metal. Therefore L(k, 0) becomes
i( k - -» = W ’ w " (8' 52>

and from (8-40a) we find
ip \ 2 C d 3k
( m) ) ( (8_53)
It is convenient to define the effective density ps( T ) of superfluid

electrons at temperature T by
Ps(T) _ 2/3E F r « e0E* dk
Ps(0) k/ Jo ** (e0E*
(eBE* ++ l)2 (8'54a)
where the Fermi energy is given by E F = kF2j2m. The relation
Ps(0) =n = ^ (8-54b)
states that all the valence electrons act as superfluid electrons at

T = 0. On combining (8-53) and (8-54) we find the simple form
lim P „ (q, 0, - [l - K (8-55)

E lectrom agn etic Prope rties o f Su p e rco n d u cto rs 219
By using this result in the expression (8-24) for K xv one finds

within the pairing approximation that
Ps(T) So
lim K xj(q, 0) = Al2(0) LPs(O) (8-56)
Thus, as long as ps(T ) is nonzero, K (q 2) [defined by (8-29)], is
nonzero as q2 -> 0 and the Meissner effect is obtained. A plot
of ps(T )/ps(0) is shown in Figure 8-2. As T T c, the density
of superfluid electrons vanishes and one goes over to the normal
state with its weak Landau diamagnetism.92
In summary, we find within the pairing approximation the
following phenomena:
1. The Meissner effect is obtained for all T ^ T c.
2. The transverse part of the electromagnetic response kernel
K goes to the London form in the long wavelength limit (8-56).
3. By recognizing that only the superfluid electrons give a
finite contribution to the transverse part of K as q —> 0, we
obtain an expression for the density of superfluid electrons ps(T )
as a function of T (8-54).
F IG U R E 8-2 The superfluid density as a function of temperature. A t

T = 0 all the electrons are in the superfluid, while at T ^
Tc all elec
trons are in the normal fluid.
4. The kernel K is not manifestly gauge invariant within this

approximation.
8-4 E L E C T R O M A G N E T IC P R O P ER T IES F O R FINITE

q A N D o>
Although it is gratifying to see the Meissner effect emerge
within the pairing approximation, one would like to know the
kernel K tj(q, co, T ) for general values of q, co, and T , as well
as the effect of impurities on this function. This problem has
been worked out by Mattis and Bardeen.69a Rather than re-
deriving their results by the Green’s function formalism,69bi c we
simply state their conclusions. For many purposes it is more
convenient to express the kernel in coordinate space rather than
in q-space. If one works in the transverse gauge
V • A(r, co) = 0 (8 -5 7 )
they find
J(r, co) = - a JdVR [ R ^ r ')] I(io, R, T ) e - Rl1 (8 -5 8 )
Here R = r — r' and the constant a is given by

e2N(0)vF (8 -5 9 )
2 tt2 K c
vF being the Fermi velocity (we do not set K = 1 in this section).

The form (8 -5 8 ) is arranged to resemble the forms given by
Pippard33 and by Chambers34 for the Meissner effect and the
anomalous skin effect, respectively. The factor e~ Rn accounts
for impurity scattering effects, where I is the electronic mean free
path in the normal state. The all important function I (co, R , T )
is given by
r*oo r*oo c
I(w , R , T ) = J J | L(o), e, e' } _ [/(^) - /(Q]
X cos de d e 1 (8-60)
Ele ctrom agn e tic P ro pe rties o f Su p e rco n d u cto rs 221
where as usual / is the Fermi function. The function L(w, e, e')

is the generalization of the function L(k, q) of the previous
section, and is given by
1
|2(e,€ \ E + E ' + ha> - i8 E + E ' — hu> + i 8
x [1 - f ( E ) — f { E ') ]
1
E — E ' -1- hex) — i 8
1
+ Ê -----
— Eit,'—— hoj -f- i 8
X [ f ( E ') — f ( E ) ] (8-61)
The coherence factors are defined by
(8-62)
and E = (e2 + A 2)112. In several limiting cases /(to, R, T ) takes

a simple form:
1. %oj » A 0\ In this limit, which includes the normal metal
as a special case, I becomes
/ (co, R, T ) = i7rHu)elRo)lvF (8-63)
and (8-58) reduces to Chambers’ expression for the anomalous
skin effect. This allows the coefficient a to be evaluated in terms
of the surface impedance of the normal metal in the extreme
anomalous limit.
2. co = 0: In this low-frequency limit (8-58) reduces to a
form closely related to Pippard’s equation. It is conventional
to introduce the function J ( R , T ) (not to be confused with the
current density J) by the relation
1(0, R , T ) = ^ J(R, T) (8-64)
where Pippard’s coherence length is defined in terms of micro

scopic parameters by
to = -774Z J-q (8-65)
In this limit (8-58) becomes
where A (T ) = mlps(T ) e 2 is London’s parameter.1 This expres

sion agrees with Pippard’s equation except for the factor e~Rlio
being replaced by J ( R , T ) in (8-66). The definition (8-64)
ensures that the J (/?, T ) and e~RI*o have the same integral
f°° J ( R , T ) d R = £0 = f°° e~RI*o dR (8-67)
Jo Jo
for all T ^ T c. One finds that not only are the integrals of the
two functions the same, but also the functions themselves resemble
each other over the entire range of R and T. For example
J ( R , 0) is within 5 per cent of e~RI*o for all R and J(0, 0) = 1,
J(0, T c) = 1.33.
3. q£0 « 1, co = 0: In this long wavelength, zero-frequency
limit we have already seen that one has the London expression
, -<r> = - £€ r !A (r> - W ) A(r» <8' 68>

for the pure superconductor, in agreement with (8-66) as can be
seen by taking A outside of the integral in this limit. For a
short mean free path I « £0, one obtains an extra factor
J ( 0, T )lj£ o ^ Z/f0, which shows that the London penetration
depth increases with impurity concentration.
4. 22/fo « i (g£o » 1): If the field is well localized on space,
for example, by a skin depth A « £0 or by the geometry of a thin
film where d « £0, one can evaluate /(co , R, T ) at R = 0 and
take it outside of the integral in (8-58). Since the remaining in
tegral is the same as in the normal state of the metal in this
limit (i.e., the factor ~ 1), we can normalize the current
e l(oRIVF
Electrom agnetic P roperties o f Su p e rcon d u cto rs 223
to that in the normal state and express the ratio in terms of the
complex surface conductivities cr in the two states. Thus,
ct, = a 1 + ia 2 = 1(oj, 0, T ) _g g
Gn crn inhaj
The expression for G1/crn was given in Chapter 3 while the expres
sion for o 2 is
cr 1 [ '
2 [1 - 2f ( E + hoj)][E2 + hwE + A2] dE
^ ^ L-HO.-A {[^2 - E 2I(E + M 2 - ^2]}1/2
(8-70)
the lower limit being the larger of the two quantities A — %a>
and —A. At zero temperature, the ratio cr2/crn is
°z - j ( > + - 1 (■ - (8' 71)

while the absorptive part is given by
<7n
= ( l + p \ E (k) ~ ^ - K ( k )
\ f id ) } fid)
(8-72)
In these expressions E and K are the complete elliptic integrals
and
Jc' = (1 — k 2)112 where k = 22AA — heo
+ hd) (8-73)
The functions o x and cr2 have been calculated by Tinkham 129 for
T = 0. For T ^ 0 numerical calculations are necessary to
determine the surface conductivity; however, a simple low-
frequency limit is
CTo 7tA . A
^ = az t a n h < 8 ' 7 4 )
Calculations for a wide range of frequency and temperature have

been carried out by Miller.130a
In general one finds remarkably good quantitative agreement
between these predictions of the pairing theory and experiment.
As mentioned in Ch. 3, a precursor absorption had been observed

below the gap edge col2A~ 0.85. While ?=h 0 collective modes give
absorption in this region, the absorption is too weak to account
for these experimental results, which were later found to be spuri
ous.
8-5 G A U G E IN V A R IA N C E
While the simple pairing approximation gives an accurate
account of the response of the system to transverse electromagnetic
fields, it does not in general give the correct response to longi
tudinal fields. In particular, we saw in a previous section that
it predicts unphysical longitudinal currents which depend on the
choice of gauge of the electromagnetic potentials. The physical
origin of this difficulty was first recognized by Bardeen,131 who
pointed out that a (longitudinal) gauge potential couples primarily
to the collective density fluctuation mode of the electron system
(i.e., the plasmons of the charged electron gas). He argued that
if one generalizes the pairing scheme to include this mode in a
consistent way, a gauge-invariant theory would be obtained.
While a number of authors have contributed to the detailed
resolution of this problem, the pioneering work of Anderson47
followed by that of Rickayzen132 gave the essentials of a general
ized pairing scheme which includes these effects. In essence,
their approach is to extend the random phase approximation to
include pairing correlations.
It is well known that a gauge-invariant response is a conse
quence of local charge conservation in the system. By local charge
conservation we mean that the electronic current and charge
density operators satisfy the continuity equation at each point in
space and time,
V -j(M ) + = 0 (8-75)
In Fourier transform variables this becomes
q • j(q. %) - goMq. 9o) = 0 (8-76a)
Electrom agn etic P roperties o f Su p e rcon d u cto rs 225
With the definitions (8-5) and (8-6) plus the metric (1, 1, 1, —1)
used previously, the continuity equation for the four-current
becomes
2 = 0 (8-76b)
u= o
where q = (q, q0) as usual. From these relations, it follows that
the expected current
•J/iM ) = <j*(r»0> (8‘77)
satisfies the continuity equation
V • J(r, t) + at/° ^ ’ ° = 0 (8-78a)
or in Fourier transform space,
2 qJM ) = 0 (8-78b)
U=0
If we concentrate on the linear response of the system to the
potential A u, we define [see (8-25)]
= ~ h 2 K u M A v(q)
77 v= 0
(8-79)
It follows from the continuity equation (8-78b) that the response
kernel K uv(q) must satisfy the equation
3
uI =o
[For qQ = 0, this condition reduces to the condition (8-28b),
used in discussing the static Meissner effect.]
Turning now to the restrictions imposed on K by gauge
invariance, we note that the most general gauge transformation
is of the form
A(r, t) => A(r, t) + Wl(r, t)
The observable fields

*■/_ *\
E(r>t) = - -10A(r,<)
— Tjjj „
V<p(r, „<)
B(r, 0 = V x A(r,<) (8-81b)
are invariant under this transformation. In Fourier-transform
space, the gauge transformation becomes
A M ) => A M ) + k v A (q) (8-81c)
If the observable current is to be unaffected by the gauge
transformation, we must require that K satisfies
v= 0
2 K uM qv =0 (8-82)
To show the equivalence between the restrictions of local
charge conservation (8-80) and of gauge invariance (8-82), we
note that K uv(q) satisfies the symmetry relations
Re K uv(q) = Re K vu( - q ) (8-83a)
Im K uv(q) = - I m K vu( - q ) (8-83b)
as a consequence of the definition (8-26) and the spectral repre
sentation (8-18) for the retarded commutator E uv. Therefore by
changing dummy indices, the real part of the charge-conservation
restriction (8-80) can be written as
2 qv Re K vu(q) = 2 Re K uA = 0 (8-84a)
V V
or sending qu —qu
2 Re K liv(q)qv= 0 (8-84b)
V
which is the real part of the gauge-invariance restriction (8-82).

In a similar way one finds for imaginary part
2 q v Im K vu(q) = - 2 Im K uv( - q)qv = 0 (8-84c)
V V
or
2 Im K uv( q ) q v = 0 (8-84d)
which agrees with the imaginary part of the gauge-invariance

restriction (8-82). Therefore, in an exact calculation, gauge invar
iance would follow as a consequence of local charge conservation.
Unfortunately, approximate calculations of K uv may not
maintain local charge conservation and therefore not lead to
gauge-invariant results. This difficulty, however, is not peculiar
to the superconducting state. It is a commonly held view that
the gauge-invariance problem of the simple pairing approximation
is due to the use of wave functions which do not describe a system
with a fixed number of particles. That this is not the source of
difficulty is seen by realizing that the matrix elements entering
the kernel K involve the operator j u, which only connects states
|a, N } with the same number of particles. Therefore, if the states
|a) used in calculating these matrix elements are an ensemble
average of states |cc, A7>, each with a fixed number of particles,
as in the BCS approach, one simply obtains an ensemble average
of matrix elements, each of which is evaluated between two
states with the same number of particles. Since these fixed N
matrix elements are slowly varying functions of N, the ensemble
average does not affect the over-all result.
The actual source of error is that the quasi-particle excitations
are not treated accurately enough in the simple pairing scheme to
ensure local charge conservation under all conditions. Exactly
the same situation exists in the normal state if one does not work
within a “ conserving” approximation, as Baym and Kadanoff133
put it, even if one uses states which explicitly describe iV-particles.
In treating the motion of an electron in the medium, one must
include the “ backflow” of other electrons around the electron in
question,132*134 as Feynman and Cohen135 stressed in their work
on excitations in superfluid helium. This backflow has a dipolar
form at large distance. As one can show, the backflow cancels
itself out if the quasi-particles are excited by a transverse field.9
Therefore, in calculating the response of the system to transverse
fields, the backflow currents play no role and one can obtain correct
results by an approximation which neglects these complicating
effects, as we did in the previous sections.
On the other hand, for the longitudinal response of the system,

the backflow around a quasi-particle has a coupling to the external
potential which in the long wavelength limit is equal in magnitude
but opposite in sign to that given by the bare quasi-particle.
Thus the dressed quasi-particle (i.e., the bare quasi-particle plus
its associated backflow cloud) is very weakly coupled to a slowly
varying longitudinal potential, as Pines and the author have
discussed.134 There is however an extra longitudinal mode of the
system which occurs once backflow is properly taken into account
and this is the collective density fluctuation mode.47*52 Physi
cally, one can think of the collective mode simply as a compres-
sional wave in the superfluid. From this physical picture it is
reasonable that the current and particle densities associated with
this mode will satisfy the continuity equation. Since in the long
wavelength limit only the density fluctuation mode is coupled
to a longitudinal potential, it is reasonable that a gauge-invariant
response will be obtained once these effects are included.
Kadanoff and Ambegaokar have shown that in the long wave
length limit the collective mode can be described as a state in
which the phase of the energy-gap parameter varies periodically
in space and time, while the magnitude of the gap parameter
remains fixed. Since the phase of the gap parameter gives the
mean local center-of-mass momentum of the superfluid pairs, a
periodically varying phase is exactly what one would expect if the
superfluid momentum density varies periodically.
There are now a number of formalisms for including the
backflow and the collective mode. One of the simplest ways to
handle the problem is to make use of a “ generalized W ard’s
identity,” which is the Green’s function analog of the continuity
equation. By making approximations which are consistent with
this identity one can ensure local charge conservation and therefore
gauge invariance. This approach was first discussed by Nambu,114
and we follow his line of argument below.
We consider the time-ordered quantity Au(x} y , z) defined in
terms of the Nambu field W by
A„(*, y , Z) = { 0 \ T { j u{ z m x ) W ^ { y ) } \ O y (8-85)
Ele ctrom agn e tic Pro pe rties o f Su p e rc o n d u c to rs 229
The four-current density j u is defined by (8-6). It is clear that

the paramagnetic kernel R uv can be calculated from by taking
the appropriate gradients and traces of (8-85), as we shall see
below. We define the vertex function Tu(x', y\ z) by the integral
relation
A(x, y,z) = e J G(z, x ^ T ^ x , y ’, z)G(y’, y) d 4x ’ d*y' (8-86)
where
G(x,x') = -i< 0 |T { ^ (a :)y +(^)}|0) (8-87)
is Nambu’s one-particle Green’s function. We assume the system
to be translationally invariant so that we can write
+ ( 8- 88 )
The generalized W ard’s identity for the superconductor is then

3
+ 2 - ?0r 0 ( p
r? 1 (8-89)
y q uT u ( p q,p) = g J 'd p + q>p) + q,p)
U
= T3G_1(p) - G_1(2> + q)T3
To prove this identity, we take the four-divergence of Au
with respect to 2 = (z, z0 = t2):
i t +1 - ^ { [ i W +%rrH),f" (!')}|0>
+ <0|T{[jo(2), ^ ) ] ^ +(y)}|0> S(z0 - x0)
+ < 0|T {ne)[jo(*), ^ +(2/)]}|0> 8(z0 - Vo) (»-90)
The last two terms on the right-hand side arise from differentiating
the time dependence due to the time-ordering symbol T. Now
the first term on the right-hand side of this expression vanishes
by virtue of the continuity equation (8-75). If we use the equal
time anticommutation relations of the W s (7-21), the commutators
in (8-90) can be reduced to
[jo(2)> ¥V )] s(2o - x o) = eT3W(z) 84(z - x) (8-91a)
and
[j0(z), Xf/+ (y)} S(20 - Vo) = - e f /+ (y)x3 s4(z - y)
(8-9 lb)
By inserting these expressions into (8-90) and using the definition

(8-86), and (8-87) we find
iG(x - z)r3 S4(z - y) - i r 3G(z - y) 84(z - x)
- I
G(a: — x') 3 s r, +, dT0
dz,o 2
G(y' - y) d*x' d*y' (8-92)
Going over to Fourier transform variables we find (8-92)

reduces to
3
G(l> + q)r3 - t 3G ( p ) = G(p + q) 2 quTu{p + q, p)G(p) (8-93a)
u =o
or finally by operating with G - 1 (p + q) on the left and with
G-1(2 >) on the right we obtain the generalized W ard’s identity
2 + q>P) = t3 - G -^ p + q)r3 (8-93b)
U =0
as stated.
What is the physical significance of Tu and why is this
identity of interest? The significance of Tu can be best under
stood by noticing that the four-current density operator j up((\)
can be written in the Nambu notation as
Î)+<, (,* = *=1,2,3)
j/(q) (8-94)
- ern 0* = °)
If we define the “ free” vertex function y w(p + q, p) as
— (p + l) 1 (/* = * = 1, 2, 3)
Y«(P + q, P) = \ m\ 2'> (8-95)
. t3 (P = °)
then j up(q) can be written as
j/(q) = - « 2 ^ p+y«(p + q> P ^ . +a (8-96)
Electrom agn etic Prope rties o f Su p e rco n d u cto rs 231
The vertex function Tu(p + q, p) is then a “ dressed ” version of

the free vertex y u{p + q, p). To make this plausible, we apply
the generalized W ard’s identity to a system of noninteracting
electrons. Since G~1(p) becomes
G o '1^ = Po* ~ «pr3 (8-97)
in this limit, where ep = ( p 2/2m) — /x, (8-93b) reduces to
2 + I ’ P) = (eP+« - s ) 1 - ?o-^3 (8-98)
U
However, this relation is identically satisfied if Tu is the free
vertex y u. Therefore, we can think of the dressed electrons as
interacting with the electromagnetic field through the dressed
vertex (—eTp).
As to why the generalized W ard’s identity is of interest, we
shall now see that the paramagnetic kernel can be simply ex
pressed in terms of G and Tu. Furthermore, if we approximate
G and T in a way which maintains W ard’s identity, the full
electromagnetic kernel K uv will be manifestly gauge-invariant.
From the definitions of the time-ordered kernel P uv (8-21) and the
vertex function Tu (8-85) and (8-86), it is straightforward to show
that P uv(q) is given by
p uM = ~ ie2 J T r[yu(2 >. p + q)G(p + ? )r v(p + q ,p ) G (p )]-j^
(8-99)
If we consider only the components of P uv with p and v ^ 0,
and approximate the vertex function Tv by the bare vertex yv,
we retrieve the expression (8-38) given by the pairing approxima
tion. The lack of gauge invariance within this approximation
is a consequence of calculating P uv with dressed G ’s but bare
vertex functions, thereby violating the generalized W ard’s
identity.
It is convenient to represent the relation (8-99) in graphical
form. In Figure 8-3 we show P uv(q) represented in terms of the
dressed electron lines and the vertex parts. While P uv appears
p +q r v(p,p + q)
y»(p + q>p)
P uv (q ) = — i e 2 x -----
F IG U R E 8-3 The polarizability kernel P uv represented in terms of

the bare and dressed vertices y u and T v, respectively.
to be unsymmetrical in /x and v, the expression (8-99) is equally

valid if Tu and Tv -> yv.
We now prove that K uv is gauge-invariant if the generalized

W ard’s identity is satisfied. This follows by writing
2v p uv(q)qv = - i e 2 JI Tr L|y«(p>p + q)G(p + q) 2v v(p + q , p )
* H <S?
= ie2 j*Tr [y„( 2>,2> + q)G(p + q ^ G ' ^ p + q)r 3
- T3G -M ^)} G (p )]|^;

= ie2 |*Tr[Y(i(p, p + q){T3G(p) - G(p + ?)r3}]
- (2
^
tt
(8-100)
In the second equality we have used W ard’s identity (8-93b).

Since r3 commutes with , the last equality in (8-100) reduces
y u
to
2 p uv(q)qv = i<? |Tr[{Y w(l> + q ,p ) ~ y u( p , P ~ ?)}
U J
x r 3G ( y )(Z7
] |^T) (8-101)
when we use cyclic invariance of the trace. From the definition of

y u (8-95)
we see that
Yu(p + 9>p) ~ Y A P ’P - q) = T
^ c - Sp.o)
fl ( 8- 10 2 )
Therefore, we obtain the simple expression

% P uM q v = - — qA 1 - V o) (8-103)
V
where we have used the relation (7-35) giving the number of

electrons per unit volume in terms of G. Since the right-hand
side of this equation is real, (8-24) allows us to replace P uv in
(8-103) by the physically relevant kernel R uv and we finally obtain
lR » M q , = - 5 «.o] (8-104)
To check that K uv is manifestly gauge-invariant due to this result,

we use (8-26) to write the gauge-invariance condition as
2 K u M ) q v = 0 = 7 7 2 R ^ ( q ) q v + v \ q » W - 8«.o] (8-105)
v C v AL
Since 1/AL2 = 47me2/rac2, we see from (8-104) that the gauge-

invariance condition is satisfied identically,
[
4,jtw62 1 1
mc?~ + W " [1 " **-°] = ° ( 8 ' 1 0 6 )
In the next section we shall discuss the generalization of the pairing

scheme which is required in order that the generalized W ard’s
identity is to be satisfied.
8-6 T H E V E R T E X F U N C T IO N A N D C O L L E C T IV E
M ODES
In seeking a gauge-invariant generalization of the pairing
scheme, Nambu used a well-known prescription of quantum-
field theory for constructing approximations which satisfy the
generalized W ard’s identity (GWI) (8-89).137 If G is described
f / S
U______L- -(- -JL .u____ L—i__ |___ Li----Li----U- + -J-1-L—L___\ 1 .1
. \ / \ "“ v \
FIG U R E 8-4 The pairing approximation sums all no-line-crossing

graphs contributing to Z.
by a certain set of perturbation series graphs, the corresponding

vertex function Tu (which satisfies the GWI) is given by the sum
of all graphs in which the free vertex y u is inserted in each bare
electron line in this set. The pairing approximation for G,
G -1( p ) = p 0l - e pr 3 - Z ( p ) (8-107)
S (p) = i j T 3G(p')T3r ( p - p ’) (8-108)
can be formally thought of as the sum of all graphs in which no
two interaction lines cross, as shown in Figure 8-4. If the vertex
y u is inserted at all possible places in these graphs, the resultant
series is summed by the ladder graph approximation for F u, as
shown in Figure 8-5. Therefore F u satisfies the linear integral
equation
ru(p + q, p) = yu(p + q<p)
+ i J T3G(k + q)Tu(k + q, k)G{k)r3i/ '(p - k) (8-109)
p + q
+ r.
s
F IG U R E 8-5 An equation for the vertex function Tu which leads to
manifestly gauge-invariant results for the electromagnetic kernel
within the pairing scheme.
Ele ctrom agn etic Pro pe rties o f Su p e rco n d u cto rs 235
where the G’s are the Nambu functions evaluated within the
pairing approximation (8-108). To check that the solution of
this vertex equation is in fact consistent with the GWI, we form
the quantity
2 + 9>P) = 2 A p + 9 ,p )
+ i f r 3G{k + q) 2 quTu(k + q, k)
Jj *■rH " "'
x—
(2tt)«* (8-110)
d'k
G(k)r3-T(p - k)
The right-hand side of this equation can be reduced by use of the

assumed GWI. By using the relation
2 quTu(k + q , k ) = t 3G -'(k) - G ~\k + q)r3 (8-111)
V
we see that the second term becomes
r d*k
[ i J t3G(& + q ) j 3r { V - k)
(2w) J ^3
d*k
3
(2tt)4J
TBG ( k ) T r ( p k)
= Z(p + q)r3 - r 3E(p) (8-112)
where we have used the equation determining Z(^) (8-108). In
addition, from (8-95) we see that the free vertex satisfies
2
u
quy»(p + q>p) = (er+« - e?)1 3
- ?ot (8-113)
On combining these results, (8-93b) reduces to

2 q»TAP + q>P) = r 3[ p 0l - epT3 - Z(|))] - [(ô + q0)l
- «p+«t3 - E(p + q)Jt3

u
= T ^ ~ 1(P) - G - ^ p + q)r3
3 (8-114)
which is the required GWI. Therefore, if G and Tu are given by
solutions of Eqs. (8-107), (8-108), and (8-109), the electromagnetic
kernel K uv determined through (8-99) will be manifestly gauge-
invariant.
To understand the mechanism by which gauge invariance has
been restored in this rather formal scheme, we again look at the
GWI. If we assume that the dressed vertex is a well-behaved

function in the limit q and q0 go to zero, the left-hand side of
(8-114) vanishes while the right-hand side becomes the finite
quantity
r 3E ( p ) - E ( p )t3 = = 2 ir a J p (8-115)
(the second equality holding for a nonretarded pairing potential).
Therefore, Tu(p H- q, p) must be singular for q = 0. If one
thinks of the coupling as going through a set of excited states
of the system (i.e., one thinks of r u as written in a spectral form),
one is tempted to argue that the q = 0 singularity in T reflects
the existence of a low-lying collective mode whose frequency Qq
goes to zero in the long-wavelength limit. To check this idea
we would like to obtain an explicit solution of the vertex equation
(8-109) and see if r u is actually singular for q0 = Qq and q # 0.
As in Chapter 7, this f-matrix-like equation can be solved if
i^(k — p) is approximated by a factorizable potential
i^(k — p) = Aw*(k)w(p) (8-116)
In solving (8-109) it is convenient to think of the 2 x 2 matrices
and y u as being represented by four-component column
vectors. Thus we replace the matrix component <Z|rjr> by the
column vector component (Tu)lr and (8-109) becomes
r v(p + q ,p ) = Yn(p + q ,p )
+ iXw*( p) | T 3'G '( * + ?)T 3rG r(t)
d*k
x T u(k + q, k)w(k)
( Z tt)
(8-117)
The scripts I and r indicate which part of the double script (Ir) the
matrices act and G means the transpose of G. Since the last
term in this expression is a constant (matrix) multiple of w *(p),
we have
r u(p + q,p) = y u( p + q>p) + w *( p )c q (8-ii8)
where the constant Cq is defined by
C - d*k
C Q = iX I t 3 'G '( * + q)Tj&(k)Tu(k + q, k)w(k) ^ 4 (8- 119)
By inserting the relation (8-118) into (8-119) and solving for CQi
one finds the explicit solution
r u(p + q,p) = Yu{p + q,p) + [1 - A<J>(g)]~1x w(?)w*(p) (8- 120)
where <|>(g) is a generalization of the function &(q) [see (5-21)] used
in discussing the instability of the normal state. The matrix
function <j>(g) is given by
4(9) = i J t 3‘Gl(k + 9)T3'G'(*)|W(k)|a (8-121)
The matrix x(Q) is defined by

XM = iXJ T 'G'(fc +
3 q)r3rGr(k)yu(k + q, k)w(k) (8-122)
Since x» is regular for q0 < 2 A0, the singularity of Tu must arise
from a singularity of [1 — A<|>(g)]_1. For this matrix to be
singular, the determinant of its inverse must be zero. Therefore,
the dispersion law for the collective mode (or modes) is
det [1 - A<J>(q, Qq)] = 0 (8-123)
where Qq is the frequency of the mode in question. If we assume
5-state pairing and as in Chapter 7 take w(k) to be
” <*> ■{J otherwise (8‘1241

there is a root of (8-123) which for |q |f0 « 1, satisfies the dispersion
law
V* = (^TT2 |q| (8-125)
This sound-wave mode was first discovered by Bogoliubov.52
Physically, it corresponds to long wavelength density fluctuations
of the electron system as a whole. Since the pairing correlations
are not expected to be changed appreciably by slowly varying the
electron density in space and time, one might expect such a
collective mode on physical grounds. In fact (8-125) follows if
one uses the standard hydrodynamic expression for the speed of
sound s:
52« = —dP
ap
(8-126)
238 T h e o ry o f S u p e rc o n d u c tiv ity
where p and P are the mass density and pressure of the free-
electron gas. Therefore within this approximation pairing corre
lations play no direct role in determining the velocity of this mode,
their main function being to remove low-lying single-particle
states which would otherwise lead to damping of the wave.
Returning to the solution (8-120) for the vertex function, if
the two-body potential is purely 5-wave, the transverse part of
%u vanishes, as one can easily see on symmetry grounds from
(8-122). Therefore vertex corrections do not affect the Meissner
kernel in this case. If there is a finite attractive d-wave part of the
potential, d-state excitons exist and will contribute to the vertex
function. Calculations by Rickayzen show that these collective
corrections to the Meissner kernel are in general small.
If there is a strongly attractive d -wave potential one should
see a precursor for infrared absorption below the gap edge, owing
to creation of d-state excitons.138,139 Such anomalies were first
observed by Ginsberg, Richards, and Tinkham,71, 72 although as
Tsuneto138 has shown, the predicted absorption is an order of
F IG U R E 8-6 Energy of a p-state exciton as a function of momentum.

Ele ctrom agn e tic P ro pe rties o f Su p e rc o n d u c to rs 239
P + q P + q
p + q
P P p k P
F IG U R E 8-7 Vacuum-polarization correction to the equation shown

in Figure 8-5. This correction is all important in the longitudinal
part of Puv and leads to plasma oscillations.
magnitude weaker than that observed experimentally. Since

the calculations were carried out on a continuum model, momen
tum was conserved in the absorption process. Owing to the fact
that the excitons pass into the single-particle continuum for
q > l/f 0, as shown in Figure 8-6, only the long wavelength com
ponents of the penetrating field contribute to exciton creation,
and these small ^-components are small compared to those for
q ~ 1/A » l/f0, thereby giving a small absorption. Experi
mentally, the precursor is essentially unaffected by impurities.13013
Since the exciton state is destroyed by impurities,1300 it appears
that the precursor is due to another mechanism. [Subsequent ex
periments have shown the precursor absorption to be spurious].
Thus far we have neglected vacuum-polarization processes in
the vertex function Tu. As we saw in Chapter 6, these processes
dominate the long wavelength polarizability of the electron gas
because of the long-range Coulomb potential and therefore must
be included in I\ The vertex equation including vacuum polariza
tion processes is illustrated in Figure 8-7. The only change in
(8-109) is to add the term
f
- iV B{q) J Tr {r3G(k + q)Tu{k + q, k)G(k)}
d*k
(8-127)
to the right-hand side. Here VB(q) is the sum of the bare Coulomb
and bare longitudinal phonon interactions:
F b (?) = - ^ 2" + M 2 A > l( < 7 ) (8-128)
To make connection with the RPA for the electron gas, we note
that P RPA (4-2) is proportional to P 00 in (8-99) when we (1)
include only the polarization term (8-127) in the equation for
(2) replace all G’s by Gq and (3) set gq
s , = It is easily seen
0.
that when (8-127) is included, the solution Tu(p + q, p) continues

to satisfy the GWI and K uv is still manifestly gauge-invariant.
By explicitly solving this improved vertex equation, one can see
that the Bogoliubov sound-wave mode continues to exist if
V l ) approaches a finite value as q -> 0, as Anderson47 first
showed. In the presence of the Coulomb potential, which of
course always exists in real metals, the Bogoliubov-Anderson
mode is pushed up to high energy and becomes the plasma oscilla
tion of the electron system. Therefore, the q = 0 singularity of
Tu(p + q , p) required by (8-115) does not imply a low-lying boson
mode in physical metals due to the long-range Coulomb inter
action between electrons.
8-7 F L U X Q U A N T IZ A T IO N
A qualitatively new effect arises when we investigate the
electromagnetic behavior of a multiply connected superconducting
system, e.g., a long, hollow cylinder. In this case, flux can be
trapped in the hole and persist in the absence of an externally
applied field. On the basis of London’s “ rigidity” concept he
concluded that for a cylinder with walls thick compared to the
penetration depth A, flux could be trapped only in multiples of
hc/e = 4 x 10 ~7 gauss cm2.1 This value of the flux quantum
follows if one assumes that the only low-lying current-carrying
states of the superfluid are those given by multiplying the super
fluid ground state by a single-valued phase factor, as we saw in
Chapter 1. We shall see below that there are two distinct sets
of low-lying states, one being the set considered by London, the
other arising from phase factors multiplying a basic state which
is not included in London’s set. Owing to these two sets of
states, the flux quantum in superconductors is actually hcj2e,
i.e., one-half the London unit. The even multiples are associated
with London-type states, while the odd multiples are due to the
Electrom agnetic P ro pe rties o f Su p e rco n d u cto rs 241
other series of states, as Byers and Y ang19 first pointed out.

The value hc/2e was observed experimentally by Deaver and
Fairbank20a and by Doll and Nabauer20b prior to Byers and
Yang’s work.
Perhaps the simplest way of understanding flux quantiza
tion is by considering a long, hollow cylinder of inner and outer
radii a and b, respectively. Suppose the cylinder is initially in
the normal state in the presence of a magnetic field and the
temperature is lowered so that the cylinder becomes supercon
ducting. Owing to the Meissner effect, the magnetic field will be
expelled from the material and in general there will be a finite
magnetic flux 0 trapped in the hole. If we assume the wall
thickness b — a is much greater than the penetration depth, the
magnetic field in the penetration layer will be a small perturbation
on the system as a whole and cannot affect our results. If we
use cylindrical coordinates (r, 9, z), the vector potential is given
by
(8-129)
where the line integral is taken around a circle of radius r and
0(r) is the flux enclosed by the path. Since 0(r) goes to a constant
0 (the total trapped flux) for r — a » A, we write
0 @(r) — 0
M r ) = ^ ~ + — 9~ = A e°\r) + A„^\r) (8-130)
and only include A ei0)(r) in zero order, ^4(1) being treated as a
perturbation. We first consider the single-particle states defined
in the presence of ^4(0) and then pair up these states to form the
superconducting phase. The azimuthal part of the single-particle
eigenfunctions satisfy
Pe + c h2
2m e * m(6) = 2 ^ 7 2 {M + vW hW (8’131a)
where
e<P
is the flux measured in units of London’s flux quantum and

0 m(0) = eiM0 (8-131c)
In order that be single-valued, M must be an integer. The
dynamics is simplified if we assume the thickness of the cylinder
b — a is small compared to the radius a. The angular kinetic
energy is then h2(M + cp)2\2mea 2, i.e., a parabolic function of M ,
centered about M = —cp. If we are to obtain a low-energy
state of the system we must pair single-particle states which are
(a) degenerate with each other and (b) coupled to other paired
states by the two-body potential. Condition (a) means that the
paired states M and M must satisfy
\M + <p\ = \M + <p\ (8-132)
This condition can be satisfied if M — M but conservation of
angular momentum forbids the two-body potential from connecting
states paired in this manner. The other choice is M -f cp =
—(M + cp), that is, M = m — cp and M — —m — cp are paired,
so that the pairing is symmetric in M -space aboutthe value
M = —cp. Since M and M are required to be integers, it follows
that m and cp are both integers or both half-odd integers. There
fore, we conclude from this result and (8-13lb) that one obtains a
large pairing energy and therefore a low-energy state of the system
only if the trapped flux is given by
0 = (8-133)
where n is an integer. In Figure 8-8a, b, c, and d we illustrate
the pairing for n = 0, 1, 2, and 3, respectively. Geometrically,
one simply pairs symmetrically about the value M = —nj 2.
Notice that the states being paired for n = 0 and n = 2 (Figure
8-8a and c) differ only by a shift of all the angular momentum
quantum numbers by the fixed amount —1. Thus, the system
wave functions for the n = 0 and n = 2 states are related by the
phase factor
</<2(ri, r2- • -r N) = c "’f ®^0(ri> ra' • ' rw) (8-134)
Ele ctrom agn e tic Pro pe rties of Su p e rc o n d u c to rs 243
F IG U R E 8-8 Pairings of azimuthal quantum numbers for flux quantum

numbers n = 0, I, 2, 3 are shown in (a), (b), (c), and (d), respectively
[see Eq. (8-133)].
in agreement with London’s argument. The states for n = 1 and

n = 3 are related in the same manner:
ifj3(rx, r2. • -rN) = r l9 r2- • -r„) (8-135)
The essential point is that the even n- and odd n-states are not
related by a phase factor. For example, the n = 0 and n — 1
states differ by sliding only the mates to the right of M = 0
left one notch in forming the n = 1 state from n = 0. This
cannot be done by a phase factor, yet the states differ little in
energy, contrary to London’s assumption. It is this “extra

degree of freedom” which was missing in London’s argument
and leads to the flux quantum being hc/2e rather than hcje.
The quantization of flux leads naturally to the stability of
persistent supercurrents. In order for the supercurrent in the
cylinder to “decay,” the system must make transitions between
macroscopically distinct states with different flux quantum
number n. Thus the current cannot slowly “ dribble” away,
but it must decrease in macroscopic steps. The probability for
such a macroscopic thermodynamic fluctuation is presumably
vanishingly small since typical fluctuations involve the excitation
of a few particles at a time rather than the entire system as a
whole. This point of view has been emphasized by Bohr and
Mottelson,141 and while it is not an airtight theory, it is a com
pelling argument in the absence of a more complete calculation.
8-8 T H E K N IG H T SHIFT
On the basis of our earlier discussion it would appear that
the electronic spin susceptibility should vanish in a superconductor
as T —> 0 if one uses s-state pairing. In this case the Pauli
principle ensures that the spins are paired in the singlet state
so that a finite spin magnetization can result only if the spin
Zeeman energy 2fxBH is greater than 2 J 0, the minimum energy to
break up a pair. A means of checking this prediction is the
Knight shift142 (the change in nuclear magnetic resonance fre
quency due to the coupling of the nuclear spins with the polariza
tion of the electrons). If only the electronic spin polarization
(as opposed to orbital effects) is important in the shift, the Knight
shift gives a measure of the electronic spin susceptibility. Reif143
found that the shift in superconducting mercury, when extrap
olated to 0°K was about two-thirds of the value in the normal
state, contrary to the simple pairing theory. Androes and
Knight144 found similar results in tin, while the shift in vanadium
is nearly the same in the N - and S-phases.
Ele ctrom agn e tic Pro pe rties o f Su p e rc o n d u c to rs 245
There have been several attempts to explain this discrepancy,

none of which is widely accepted at present as providing the essen
tial mechanism. It may well be that a fraction of the observed
shift comes from several of the following mechanisms.
Parallel Spin Pairing

If one uses p-state (or any odd Z-state) pairing, the anti
symmetry of the wave-function forces the spins to be paired in
triplet states. Fisher59a studied the model in which one pairs
states with equal z-components of spin (i.e., up paired with up,
down with down). The reduced Hamiltonian is then the sum of
two noninteracting parts. In this case there is no energy gap
for creating spin polarization since a down-spin pair can be changed
into an up-spin pair with no change of energy. Unfortunately,
this type of pairing leads to an anisotropic energy gap which
vanishes in certain directions and a nonexponential electronic
specific heat at low temperature. A more general triplet pairing
has been treated by Balian and W erthamer590, who include all
three components of the triplet state. In a pure material they
obtain an isotropic energy gap; however, a small amount of dis
order or impurity (which was certainly present in the above
experiments) destroys their state as well as that of Fisher.
Spin-Orbit Coupling at the Surface
Since the Meissner effect screens out the magnetic field within
a distance A ~ 5 x 10"6 cm of the surface, the Knight shift
experiments are carried out on specimens whose dimensions are
small compared to A in order to eliminate line broadening due to
the field inhomogeniety. The smallness of the particles led
Ferrell145 to suggest that the spin-orbit coupling near the surface
might be sufficiently strong to mix the one-electron spin states
during scattering and thereby lead to a finite spin susceptibility.
A semiquantitative theory of this effect was worked out by Ferrell
and by Anderson,146 however, we shall not discuss the calculations
here. Unfortunately, as mentioned above, recent measurements
on vanadium and aluminum,147 which are light metals and should
have appreciably weaker spin-orbit coupling than mercury and

tin, give essentially the same Knight shift in the normal and
superconducting states. This rules out the spin-orbit mechanism,
at least in this case.
Collective Magnetization
In the original BCS paper8 it was suggested that there might
be low-lying collective spin wave states in a superconductor
which give rise to the observed Knight shift. Bardasis and
Schrieffer139 found that by retaining the p-wave part of the two-
body potential, spin wave states can exist in the energy gap.
However, to obtain a nonzero long wavelength susceptibility the
spectrum must go down to zero energy in this limit. These authors
found that two situations can exist. If the p-wave part of the
two-body potential is weaker than the 5-wave part, the spin
waves possess a finite energy as their momentum goes to zero.
If the jo-wave potential is stronger than the 5-wave potential, the
spin wave states are unstable and the ground state is then formed
by p-state pairing. One is then led back to the difficulties of the
first proposal.
Modified Antiparallel Spin Pairing

Soon after the BCS theory was proposed, Heine and Pippard148
suggested that one might relax the strict BCS rule of pairing a given
state k with only one other state k. They argued that if k is
paired with a group of states centered about k, a finite spin sus
ceptibility might result. Their argument was based on an assumed
form of the two-particle density matrix which is not consistent
with that given by the most general form of the pairing theory.
Since it has not been possible to construct a wavefunction which
gives their density matrix, it appears that their basic assumption
cannot be justified and we are forced to reject this point of view.
Subsequent to Heine and Pippard’s proposal, Schrieffer149
argued that a finite spin susceptibility would be obtained if the
pairing condition were modified in the magnetized state. In
Electrom agn etic Prope rties of Su p erco n d u cto rs 247
particular, he suggested pairing states k f and —k' j , which

are degenerate when the spin Zeeman energy is included, rather
than the BCS prescription of pairing degenerate orbital states k f
and —k | . This scheme then leads to a net spin magnetization.
In a pure unbounded specimen the modified pairing gives essen
tially no pairing energy since momentum conservation forbids
a pair with center-of-mass momentum k — k' from scattering
into a pair state with center-of-mass momentum k — k', owing to
these momenta being different in general. Under experimental
conditions, impurity and surface scattering is sufficiently strong
to spread the single-particle states “k ” in momentum space by
an amount large compared to the “ center-of-mass” momentum
%|k — k'| ~ 2/jlb H / v f . Therefore, the actual single-particle eigen
states (including these one-particle scattering effects) can be
formed into pairs in the above manner and coupled by the two-
body potential. The pairing still occurs between two definite
single-particle states. It is reasonable to assume that the reduc
tion in the pairing energy will be a smoothly varying function of
the net spin magnetic moment and therefore a finite spin sus
ceptibility will result. This scheme is very different in point of
view from that proposed by Heine and Pippard, who argue that
a group of single-particle states are strongly correlated in occu
pancy (even in a pure, unbounded specimen).
Recently Cooper150 has reinvestigated Schrieffer’s idea by
introducing a phenomenological two-body potential whose matrix
elements are taken to be a slowly varying function of the center-
of-mass momentum of each pair. This leads to a pairing energy
which varies slowly as a function of the spin magnetization and
gives a finite spin paramagnetism. Cooper stressed the added
possibility of momentum nonconservation being due to a non-
translationally invariant two-body potential.
Orbital Paramagnetism
Clogston, Gossard, Jaccarino, and Y afet151 have argued
convincingly that in vanadium essentially all of the Knight
shift is due to Kubo-Obata temperature-independent orbital
paramagnetism.152 Since this orbital paramagnetism should be

the same in the N- and $-states, the lack of change in the Knight
shift in vanadium is presumably explained without modifying the
pairing theory. It appears unlikely that this mechanism can
account for the observed shift in all cases.
8-9 T H E G IN S B U R G - L A N D A U - G O R ’K O V T H E O R Y
Thus far we have concentrated on the response of a super
conductor to weak electromagnetic fields. There are many im
portant problems, e.g., N-S phase boundary, the intermediate
and mixed states, etc., in which the magnetic field enters in a
nonperturbative manner. These problems typically involve the
energy-gap parameter A varying as a function of position in the
materials. As we discussed in Chapter 1, the phenomenological
theory of Ginsburg and Landau36 (proposed in 1950) gives in
many instances a good account of these strong field situations.
An important advance in the microscopic theory was made by
Gor’kov,37 who showed how the GL equations follow from the
pairing theory when T is near T c and the magnetic field varies
slowly in space over a coherence length. Gor’kov found that the
GL effective wave function ^(r) is proportional to the local
value of the gap parameter A(r) and the effective charge e* of the
GL theory is equal to 2e, the charge of a pair of electrons. It is
interesting to note that these results were guessed prior to
Gor’kov’s work, the identification of ^(r) andd(r) being suggested
by Bardeen8 and the effective charge e* = 2e being suggested
by Ginsburg153 prior to the BCS theory in order to fit the GL
theory with experiment.
We give a brief summary of Gor’kov’s derivation below.
To familiarize the reader with Gor’kov’s scheme, we use his
notation. For simplicity Gor’kov used a nonretarded zero-range
attractive potential to describe the pairing interactions. Since
this singular potential leads to divergences, it is cut off in momen
tum space at the appropriate point in the derivation. The vector
potential A(r) is treated self-consistently as we did above in the

weak-field case. The system Hamiltonian is then
n - - Z J V lo j i — A/ + Mj<As(r) d3r
16 A(r
c
JV t +(r),/,t(r)<Ai +(r)'/'i(r) d3r (8-136)

where e = —\e\ is the charge of an electron, and we measure
single-particle energies relative to the chemical potential p. The
thermodynamic Green’s function
C ir r ,v _ Tr [ e - * " T { U x ) h +{x')}]
Xf e ~PH
= - ( ^ ( a # , +(*')}> (8-137a)
is defined for pure imaginary time
ifj(x) = 0(r, t) = eHli/j(r, 0)e~Hl (8-137b)
and satisfies the equation of motion
d 1 -A
te
(r) 4- pjG(x, x )
dr 2m
+ V<T{'P^(x ')'Pi +(x )'Pi(x )'P\(x )}> = S(^ ~ *') (8-138)
This result follows from (8-136) and (8-137) since
1 V - - A(r)
= [H, # r)]
2m c
+ F0i +W i(^ )JAr(^) (8-139)
In Gor’kov’s scheme, the pairing approximation for this zero-
range potential corresponds to factorizing the four-point function
in (8-138),
<T{«Af +(*')(/'i +(a;)i/-|(x)</rf(a;)}) => <T{i/.t +(x')ifi{ +(x)}}(<Pi(x )>/'i(x )'>
(8-140)
Therefore the equation for G becomes
250 T h e o ry of Su p ercon d uctivity
where the “ anomalous” Green’s function F +(x, x ) is defined by

F +(x,x') = -<T{0i + (x)0t + (*')}> (8-142)
and the energy-gap parameter J(r) is given by
A*( r) = F<i/q(r)0T(r)>* = V F +(x, x') (8-143)
The functions G and correspond to Nambu’s G1X and G21.
Since F+ is an unknown function, it must be determined from its
equation of motion. By making a factorization similar to that in
(8-140), (except that the four-point function now contains three
j/ and one one finds
C ’s 0),
(8-144)
As we discussed in Chapter 7, the pure imaginary time
Green’s functions can be expressed in the Fourier series variable
ajn = (2n -f l)nlp (n = integer) and one finds the Fourier com
ponents ^ w(r, r') and S' J r , r') satisfy
+ A (r)^ Jr, r') = 8(r - r')
- J* (r)S U r,r') = 0 (8-145)

These equations together with the condition (8-143)
(8-146)
in principle determine the behavior of the superconductor in the
presence of an arbitrarily strong potential A at any temperature
kBT = l/p.
The nonlinearity of the coupled equations makes them
difficult to handle. Gor’kov restricted his attention to the region
T near T c, where the gap parameter is small so that a perturbation
expansion in powers of A can be carried out (in the spirit of the
GL theory). Furthermore, the penetration depth A becomes
Ele ctrom agn etic P ro pe rties o f S u p e rco n d u cto rs 251
large compared to Pippard’s coherence length £0 for T ~ T c and

therefore A will vary slowly over a coherence length. In, this
limit the linear relation between the current density and the vector
potential reduces to London’s equation. To carry out the series
solution of (8-145) in powers of J , Gor’kov wrote these equations
in integral form
where ^ is the Green’s function for an electron in the normal metal

in the presence of the magnetic field
If one solves the equations for F +, accurate to terms of order Zl4,

one finds that the equation (8-146) determining A is
x Zl*(l) J*(m) d3s d3l d3m (8-149)

The first term on the right-hand side of the form
(8-150a)
where the kernel K ( r, r')
(8-150b)
is given by
and
if A = 0. Since A is assumed to vary slowly over a coherence

length and ^ ( r , r') decreases exponentially for |r — r'| > vF/aj
~ (a) ~ A), a WKB-like approximation can be used to
include A and one finds
(8-150e)
and
** ^ %P
^ w(r, r') = - r') exp - (r - r') • A (r)
(8-150f)
The singularity of K 0 as r r' arises from the zero-range
two-body potential. If one cuts off the potential Vkk> outside
the energy range —cu0 oj0 (centered about the Fermi surface),
one has
J K 0(R) d R = N { 0) J “° i tanh ^ de
(8-151)
In the reduction we have used the equation determining k BT c =
1l(3c. By expanding the normal metal Green’s functions in powers
of the small quantity (e/c)(r — r') • A(r) and assuming that A(r)
varies slowly over a coherence length, Gor’kov obtains the
equation
lG 12{-nkBT cf (8-153)
and i(x) is the Riemann zeta function.
Ele ctrom agn e tic P ro pe rties o f S u p e rc o n d u c to rs 253
By introducing the “ wave function”

0(r) = A (r)[U (3 )n y 2
4t7rTn
(8-154)
one obtains the Ginsburg-Landau-like equation
f±
[2m
V- te*
A(r) (‘ ~ r j
= 0
- 1 Wr)|i
H(r)
(8-155)
where e* = 2e. One can also calculate the current density
7P (Vr< - Vr)G(x,x') 2e2 A(r )G(x
—
J(r) = mc ,*')] +
J t'=t ,r = r'
(8-156)
to second order in A by the perturbation expansion used above
and one finds on using the relation between i/j and A :
p*2
ie*
J ( r ) = - <jm + - (8-157)
— A (r)\4>(r ) \ 2
as in the GL theory.
Recently, the derivation of Gor’kov has been extended to all
temperatures by W ertham er154 and by Tewordt,155 who continue
to assume the system is such that A and A vary slowly over a
coherence length. Their equations are somewhat more compli
cated than the GLG form, as one might expect. Gor’kov has
extended his treatm ent to include finite mean-free-path effects.
He finds the equations have the same form as above, except that
the “mass” m is increased relative to that of the pure material.
CONCLUSION
In this book we have mainly discussed the microscopic aspects

of the theory of superconductivity. The microscopic theory has
of course many macroscopic consequences, the discussion of which
is beyond the scope of our present treatment. Of particular
interest in this respect is the theory of type II or “ hard” super
conductors. As we mentioned in the introduction these materials
do not show a sharp drop of the magnetization curve at the critical
field, which is characteristic of ideal type I or “ soft” super
conductors. Rather, they exhibit a perfect Meissner up to a lower
critical field H cl, after which the magnetization drops continuously
to zero at the upper critical field H c2, owing to flux penetration.
Since the Ginsburg-Landau parameter k which determines
,
whether a material falls in class I or II, can be influenced

by the concentration of impurities, cold work, etc., one can in
vestigate the transition region between these two distinctly
different types of magnetic behavior.157,158
There is also a large body of literature dealing with the effect of
magnetic and nonmagnetic impurities on the properties of super
conductors.159, 160 An important advance in this area was made
by Anderson,146 who argued that in the presence of nonmagnetic
(time-reversal-invariant) impurities one should pair single-particle
states which already include the effects of scattering from
the impurities. On the basis of this idea he showed that the
254
C o n c lu sio n 255
dominant effect of a small concentration of nonmagnetic impuri

ties is to remove the crystalline anisotropy of the energy gap and
thereby reduce the transition temperature.161 The gap edge is not
broadened, however. For larger impurity concentrations, numer
ous other effects such as shifts of valence electron concentration,
electron and phonon band structure, electron-phonon interaction,
etc., become important and the problem of determining Tc is quite
involved.
Magnetic impurities generally have a tendency to lower the
superconducting transition temperature, because the antiparallel
spin correlations in the superconducting state prevents the valence
electron-magnetic ion interaction from entering as favorably as
in the normal state.162*163 There are examples,160 however, where
no localized moment occurs and T c is increased. In addition,
magnetic impurities can, within a limited range of concentration
lead to “ gapless” superconductivity.172 This important effect
was predicted by Abrikosov and Gor’kov, and observed by Reif and
Woolf. The problem of superconductivity in transition metals
and the relation between ferromagnetism and superconductivity
will no doubt receive considerable attention in the future.
Another problem which has received a good deal of attention
is the behavior of small specimens and thin films in the presence
of strong magnetic fields.164 A related problem is the structure
of the superconductor normal-metal phase boundary. In addi
tion there are many interesting problems associated with super
imposed films of normal and superconducting metals.165 It is
well known that a superconductor can make a nearby normal metal
become superconducting by allowing the electrons of the two
metals to intermingle. The range over which this effect can take
place, the effect of magnetic fields and magnetic impurities on the
interaction between the metals, etc., are worthy of further study.
Another area in which the pairing theory has met with con
siderable success is in the theory of nuclear structure. Following
the original suggestion of Bohr, Mottelson, and Pines,166 numerous
workers have used the pairing theory to calculate the single
particle and collective excitation spectra in heavy nuclei. The
theory has been remarkably successful in predicting the difference

of the low-lying single-particle spectra between even and odd
nuclei. In even-even nuclei there is an energy gap of order
1 mev for exciting a neutron or a proton, because of the existence
of the pairing interactions. In even-odd or odd-odd nuclei the
corresponding excitation energy is smaller by a factor of four or
more due to the unpaired particles being present in these nuclei.
In addition, the rotational and vibrational spectra are generally
brought into close agreement with experiment once the pairing
correlations are included. Pick-up and stripping reactions measure
the energy distribution of the bare particles and reasonable agree
ment is obtained between the smeared Fermi surface characteristic
of the pairing theory and these experiments. Although the effect
of the pairing correlations in nuclei is not as striking as in super
conductors, it is clear that these correlations play an important
role in determining the properties of nuclei.167
Another area where the pairing concept has been applied is
in the mass spectrum of elementary particles. It is difficult to
resist drawing the analogy between the quasi-particle energy in
a superconductor E p = (ep2 + A2)112 and the relativistic form
E p = ( p 2 + m 2)112. Nam bu168 and co-workers have based a
model of elementary particles on the pairing scheme, as has
Fisher.169 Whether these attempts will play a role in the ultimate
resolution of the mass spectrum problem is unclear at present.
The pairing theory has also been applied170 to the possibility
of a superfluid phase of He3. Thus far no such transition has
been observed down to temperatures below 0.01°K.
A problem that has received little attention to date is whether
there are systems having strong correlations involving clusters of
more than two particles. One knows that alpha-particle corre
lations are important in light nuclei and L ittle171 has experimental
data which might be interpreted in terms of large clusters. Never
theless, we must conclude that the pairing correlations, upon
which the pairing theory is based, are the essential correlations
required to explain the basic phenomena observed in the super
conducting state.
A PPEN D IX
SECOND-QUANTIZATION
FORMALISM
In this appendix we shall give a brief summary of second

quantization.
A-1 O C C U P A T IO N - N U M B E R R E P R E S E N T A T IO N
Let us consider a system of n identical particles described
in the Schrodinger representation by the Hamiltonian
H {x1 • • -xn) = 2 ^ + 2 CiOCj) + \ y v 2(x{, X j ) (A-l)
The coordinate xt labels the position and spin of particle i.
Three-body potentials and higher interactions can be included in
a straightforward manner, but for the moment we shall confine
ourselves to two-body interactions.
The many-body Schrodinger equation is
H(x, •• xnm x , •••*„,«) = ih 8W(Xldt 'Xn’ t] (A-2)
257
We introduce a complete set of n-particle wave functions 0.

These are constructed as a properly symmetrized product of one-
particle wave functions uk(x), which form a complete orthonormal
set
^ u k.*(x)uk(x) dx = hkk' (orthonormality) (A-3)
2 uk*(z ')u k(x) = 8(x — x') (completeness) (A-4)
k
The function is then given by

0
0 • -ukn(xn) (A-5)
where i f = (1/n!) 2 R in Bose statistics and i f = (1/n!) 2 (~ 1)pP
in Fermi statistics and the summation is over all n! possible
permutations of the coordinates • -xn and p is the order of
the permutation. Rather than labeling 0 by the quantum
numbers k x, k2 - k n, we may specify the state by stating how
many times each single-particle state enters the product. Let
this occupation number be n k for state k. Then the set of numbers
n l9 n 2 - • -nk uniquely specifies the symmetrized state 0 nitn2• -nk-
If we describe a system of zi-particles, we have clearly n k = n.
For Fermi statistics the occupation numbers n k are restricted to
the values 0 and 1, whereas for Bose statistics they can have all
possible positive integer values (as well as 0). The functions
form a complete orthonormal set of n-particle
^ r t ! ■■■ nk - - ( x i ‘ * ’ x n)
functions for fermions when they are multiplied by the factor

(n!)1/2, whereas a multiplication factor (n!/n1!n2!)1/2 must be
included to obtain a complete orthonormal set of Bose functions.
(Note: 0! is defined as unity.) The orthonormality condition is
J ’ •*»)4>»i.«a-••(*!• • -Xn ) d X i " dxn

= V .n ,* (A-6)
In general, the total Schrodinger wave function may be expanded
in the complete set of the 0 Ui... ...:
xP(x1 • • •xn, t) — ^ A(n1 • • •nk • • •, t ) 0 ni... nk. . \xx • • •xn) (A-7)
S e c o n d -Q u an tiz atio n Form alism 259
The coefficients A{n1 • • -nk • • •, t) are now to be interpreted as

the wave functions in the occupation-number representation.
Their norm gives the probability of finding nfc-particles in state k.
A-2 S E C O N D Q U A N T IZ A T IO N F O R B O S O N S
For Bose statistics we introduce a set of operators a k and a k +
defined by
“ k + t f V - . n * . •■(*!• ' - * n ) = K + - - ■ -Xn + 1)
(A-8)
•(*!• ' 'xn) = (nk)m ^ ni---nk - l - - i x l - ' ' - 1)
The operator a k + (creation operator) creates an additional particle
in state k , and a k (annihilation operator) destroys a particle in
state k. If nk = 0 in 0, the operator a k gives 0.
We can see that a k + is the Hermitian conjugate of a k by noting
that the only nonvanishing matrix element of a k is
(a ) n * Uk ~ 1
which is equal to (nk)112, and its Hermitian conjugate operator will
have as its only nonvanishing matrix element
* = K ) 1'2
The operator which has only this nonvanishing matrix element is
indeed a k + by definition. If we define a new operator N k = a k +a k,
it follows from definition (A-8) that its eigenvalue equation is
Nk®ni...nir..(xi- ■•*„) = nk0 ni...nk...(x1- • -xn) (A-9)
Therefore, N k may be interpreted as the operator which measures
the number of particles of state k (the number operator). We
may now construct the operator N which measures the total
number of particles in the system:
N = 2 = 2 a k +Clk (A-10)
The commutation relation between the creation and destruction

operators is easily established to be
[®7c> ®k' ] ^ kk' M [P'k > ak' ] ^ ( A “l l )
For instance,
(aka k + - rii •••n k ••• 10
= [(% + 1) - nk]0,rix •••n k •• . = , •••n k
rii
The Hamiltonian can be expressed in the occupation number

representation, and it can be seen that it becomes
(.ki , k 2 \V2\ki, k^)a ki. + + a kia k2 (A-12)

ki'k2'ki kz
where
<k'\Hi\k} =
and
<&i , k 2 | V2\kl9 k 2y
j u ki.*(x1)uk2.*(x2)V2(x1, x 2)uki(x1)uk2(x)2dx1 dx2
This can be proved by noticing that all matrix elements ofHamil-

tonian (A-12) with the complete set &ni. . are equal to the matrix
elements as evaluated from the original Hamiltonian (A-l) in
configuration space. We shall not give the complete proof but
will outline the way in which it can be obtained.
First, we evaluate all matrix elements of the Hamiltonian
(A-12). The basic rule is to use the orthonormality of the set <P
and the definition (A-8) for creation and destruction operators.
The wave function obtained by operating with (A-12) on the
wave function to the right of the matrix element must be equal to
the wave function on the left of the matrix element.
1. In diagonal elements only terms with k = k' and k x = k x',
k 2 = k 2 or k x = k 2 , k 2 = k x' or k x = k 2 = k x’ = k 2 give nonzero
Se con d -Q u an tiz atio n Form alism 261
contribution because only these terms leave the wave function

to the right unchanged. We have, then,
k
+ i 2 V kl{<k1k 2\V2\k1k 2y
ki * k 2
-f ( k 2k i | V2\kik<^))
fci
+ \ ^ Uk1 {n k1 ~
(A-13)
2. We now illustrate off-diagonal elements between wave
functions which differ in the occupation numbers of two states
i and j. Let the wave functions be
• rt,rty • • • and 0
with
n i + nj = m i + ra;
because the number of particles is fixed. The only matrix
elements different from zero are those for which either
(1) n} = rrij + 1 ft* = m i ± 1
or
(2) +2 = ra* + 2
On choosing the upper signs, the matrix elements for the first
case are
l \k,y[(ml + 1)mj]112 + + 1)m,]112
x V\k,k,} + ( k lk i\V\kjk l')) (A-14)
and for the second case
- l)(m, + !)("h + 2)]ll2( k jkj \V2\kik i')
On choosing the lower signs, we have the same expressions with
i and j interchanged.
3. We show here off-diagonal elements between wave
functions which differ in the occupation numbers of only three
states i, j , and I. Those different from zero obey conditions of the
type
ni — mx ± 1 7ij = mj ± 1 nx = m x + 2
The matrix element with the upper choice of signs is

|[(W| + 2 ){ml + I)mjm i]ll2{ ( k lk l\V2\kjJci) + ( k lk l\V2\kikJ}}
With the lower choice of signs, we have
+ 1){mi + l)(m, - 1)ml]112
x {<kikj \V2\klk ly + (kjktl V2\klk l}} (A-15)
4. The following are off-diagonal elements which differ in the
occupational numbers of four states k t, kv k t, and k s; those different
from zero obey the condition
n{ = mi + 1 rij = rrij + 1 nl = ml ± 1 ns = ms ± 1
The matrix element is

[m lm j(m l + 1){ms + 1)]1/2{<*;^S| F 2 | M ( > } (A ' 1 6 )
with the upper choice of signs. A matrix element of the same
type, but with Z, s , and i, j interchanged is obtained with the
lower choice of signs.
The same matrix elements as in 1, 2, 3, and 4 are obtained
from Hamiltonian (A-l). Let us consider as an example the last
case; the same procedure will apply to the other cases. We want
to evaluate
<0- • rrii — 1, rrij — 1, m l + 1, ms + 1|
\ 2 V (m)\- ■ 2 rrij, m l9 ras>
pq
First, we can simplify the normalization constants by integrating

over all variables other than p and q whatever p and q may be.
We have
[(rrii - 1)!(m; - l ) ! ( r a , + l)!(ms +
x <ttt(l)- • - 1 )w,(m,)' • -f- m l)uj { m i + m, -f- 1) • • •
x u j ( m i + rrij -f m l — + m 3 + m ,)• • •
X us(mt + m, + m t + ms)\\ 2 V2iP> ••
pq
x + 1)- • •ul(mi + m l)uj{mi + m l + 1) - • •
x w;(mj + ra; + mÛsinii + m }r + m l + 1) - • •
x u^rrii + rrij + m l + ras)>
Se con d -Q u an tiz atio n Form alism 263
where P indicates all possible permutations of particles. When

p = m i and q — m i -|- ra; + m x, or vice versa, we obtain from the
integral the quantity
( k lks\V2\kik,') +
multiplied by a factor because of the permutations
which interchange particles on the same state on the wave
function to the right. Because of the summation over p and q
in the operator, we must consider the sum of all terms with
m i ^ p ^ {rrii + m ,)
and
(mf + rrij + ra,) < (vni + + m, + ms)
this introduces another multiplicative factor (ml + l)(ms + 1).
By multiplication of all factors with the normalization constants,
we obtain
+ 1)(ms + l)]1/2{<fc,&s| + (k ,k s\V2\kik i)}
which is the same as Eq. (A-16).
Let us define some new operators which do not depend on
state k but depend on the variable x as
4>{z) = 2 uk(x)ak
k
and
>P+ (X) = 2 «fc(*)*«k + (A-17)
k
They are called “ wave field” operators and satisfy the commuta
tion relations
M x), tfj+(x')] = 2 uk{x)uk.*{x')[ak, a k. +] = 8(x - x')
kk'
MX), 4>{x')} = M x), ftx')]* = o (A-18)
Toclarify the usefulness of the wave field operators, we may
notice that p(x) = ip+(x)i/j (x) represents the density of particles in
x-space and the number operator is
N — p(x) dx = 2 ak'+ak \uk, +(x)uk(x)dx = 2 ak +a k (A-19)
J kk' j k
In many instances it is useful to have the Fourier transform of p ,

given by
pq = |e iqxp(x) dx = ^ a k + a k l elQ'xuk,*(x)uk(x) dx (A-20)
J kk' J
When the uk(x) are given simply by plane waves eik xand we
normalize in a box of unit volume, we have k' = k —q and,
consequently,
p = 2 a k, + a k' + q (A- 2 1 )
k'
We may express the Hamiltonian operator in terms of the
field variable i(j (x ); from Eq. (A-12) and the definition of the iff we
obtain
H = j i/j +(x )H1(x )i/j (x ) dx
+ i J V + (x )lfJ +(x ) V2(%, x')ifj(x')ifj(x) dx dx' (A- 22)
The order of the operators ensures that the term i = j has been
omitted in the two-body potential. If V2{x, x ) = V2(x — #'),
that is to say, if our two-body operator is translationally invariant,
the Hamiltonian can be written as
H = [ 4 , + {x)H1{x)>li(x)dx + V 2 ( 9 ) v ( P q + Pq) ( A- 2 3 )
J q
where rj is the normally ordered product such that all the i/j + are
placed to the left and all ifj to the right in the product. The
proof is obtained by expanding V2(x — x ) as its Fourier trans
form and using the definition of p q.
The prescription for expressing an n-body interaction in the
occupation-number representation is now clear:
Vv = t J V + (*i)- • - f W F v f e - • - xv)4i(xv)- ■-^(x^ dx1- ■■dxv
( A- 2 4 )
From this point on we shall suppress the variables • • •xn in 0

and represent 0 by
• •*„) = |*H-••«(••■> ( A- 2 5 )
Second-Quantization Formalism 265
Equally well, |nx • • •nx• • •) can be considered to be a vector in a

Hilbert space. This vector has components • • -xn\n1 • • •nx• • •)
along a complete set of position eigenvectors \x1- ■-xny. These
components are indentified with the function
• -nt- ■- ( x r ■-Xn)
The n-body Schrodinger equation may now be expressed
in the a^-language as
HY(t) =
where
W(t) = 2 A{n1, n 2 - ■■t)\n1, n 2- ■-y (A-26)
ni,n2•••
and
H = H , + V2 + • • • + Vv + ■. .
as given in second-quantization formalism.
A-3 S E C O N D Q U A N T IZ A T IO N F O R F E R M IO N S
For Fermi statistics we introduce creation and destruction
operators ck + and ck, formally given by Eq. (A-8), referred to the
antisymmetric wave function. These operators satisfy an anti
commutation relation
{ c k + > c k'} = &kk'
and
{ck, c k.} = 0 = {ck \ c k, +} (A-27)
where
-f BA
{A, B) = A B
It can be seen that this choice of commutation relations restricts
the occupation number of the states i to 0 or 1 as required by
Fermi statistics. In fact, ckck&nk = 0 = (nk — 1)ll2(nk)ll2&Uk _2*
With this restriction in mind, the definition of the operator ck is
The Hamiltonian operator in the second-quantization formalism is

given by
H = 2 <fc'sl ^ il ^ X - 5+cfc5 + 5 2 < ij\ n ik }c isU is, +cks.cls (A-29)
k'ks Ijkl SS'
where s and s' label the spin of the particle.

The wave fields are defined as
>P(z) = 2
k
and they satisfy the anticommutation equations
{ifj(x), <A+(x')} = &(x - x')
{^(x)^(x')} = 0 = { r { x ) , ^ { x ' ) }
Note that in (A-29) the order of the destruction operators with

regard to the matrix-element indices is the inverse of that of the
creation operators. The v-body operator of Eq. (A-24) is
Fv = i 2 <*v'- - - * i'|r ,|* v•••*!>
V'kv'sv'• ••fci 'si 'kvsv•••kiSi
* sv' * ' *ckl' Sl Si Ckvsv (A-24 )
The ordering is not entirely arbitrary because of sign changes
arising from anticommutation of fermion operators; that of
(A-29), (A-24'), (A-24), and (A-22) gives the correct sign for either
fermion or boson operators.
If one is dealing with particles of nonzero spin, the co
ordinate x represents both sp&ce and spin variables, as does the
variable k. Integrals over x represent integrals over space and
sums over spin variables.
MACROSCOPIC QUANTUM PHENOMENA FROM
PAIRING IN SUPERCONDUCTORS
Nobel Lecture, December 11,1972
by
J. R. S c h r ie ffe r
University of Pennsylvania, Philadelphia, Pa.
I. I ntroduction
It gives me great pleasure to have the opportunity to join my colleagues John
Bardeen and Leon Cooper in discussing with you the theory of superconduct
ivity. Since the discovery of superconductivity by H. Kamerlingh Onnes in
1911, an enormous effort has been devoted by a spectrum of outstanding scien
tists to understanding this phenomenon. As in most developments in our branch
of science, the accomplishments honored by this Nobel prize were made
possible by a large number of developments preceding them. A general under
standing of these developments is important as a backdrop for our own contri
bution.
On December 11, 1913, Kamerlingh Onnes discussed in his Nobel lecture (1)
his striking discovery that on cooling mercury to near the absolute zero of tem
perature, the electrical resistance became vanishingly small, but this dis
appearance “did not take place gradually but abruptly.” His Fig. 17 is re
produced as Fig. 1. He said, “Thus, mercury at 4.2 K has entered a new state
Fig. 1
267
which owing to its particular electrical properties can be called the state of
superconductivity.55 H e found this state could be destroyed by applying a
sufficiently strong magnetic field, now called the critical field Hc. In April —
June, 1914, Onnes discovered that a current, once induced in a closed loop of
superconducting wire, persists for long periods without decay, as he later graphi
cally demonstrated by carrying a loop of superconducting wire containing a
persistent current from Leiden to Cambridge.
In 1933, W. Meissner and R. Ochsenfeld (2) discovered that a superconductor
is a perfect diamagnet as well as a perfect conductor. The magnetic field van
ishes in the interior of a bulk specimen, even when cooled down below the
transition temperature in the presence of a magnetic field. The diamagnetic
currents which flow in a thin penetration layer near the surface of a simply
connected body to shield the interior from an externally applied field are stable
rather than metastable. On the other hand, persistent currents flowing in a
multiply connected body, e.g., a loop, are metastable.
An important advance in the understanding of superconductivity occurred
in 1934, when C. J. Gorter and H. B. G. Casimir (3) advanced a two fluid
model to account for the observed second order phase transition at Tc and
other thermodynamic properties. They proposed that the total density of
electrons g could be divided into two components
e= (i)
where a fraction Qs/Qn of the electrons can be viewed as being condensed into a
“superfluid,55 which is primarily responsible for the remarkable properties of
superconductors, while the remaining electrons form an interpenetrating
fluid of “normal” electrons. The fraction QslQn grows steadily from zero at Tc
to unity at T = 0, where “all of the electrons55 are in the superfluid condensate.
A second important theoretical advance came in the following year, when
Fritz and Hans London set down their phenomenological theory of the electro
magnetic properties of superconductors, in which the diamagnetic rather than
electric aspects are assumed to be basic. They proposed that the electrical
current density j s carried by the superfluid is related to the magnetic vector
potential A at each point in space by
j8= - LAc a (2)
where A is a constant dependent on the material in question, which for a free
electron gas model is given by A — m lg se 2, m and e being the electronic mass
and charge, respectively. A is to be chosen such that p • A = 0 to ensure cur
rent conservation. From (2) it follows that a magnetic field is excluded from a
superconductor except within a distance
— J Ac214:71
which is of order 10-6 cm in typical superconductors for T well below Tc.
Observed values of A are generally several times the London value.
In the same year (1935) Fritz London (4) suggested how the diamagnetic
268
property (2) might follow from quantum mechanics, if there was a “rigidity”
or stiffness of the wavefunction ip of the superconducting state such that ip was
essentially unchanged by the presence of an externally applied magnetic field.
This concept is basic to much of the theoretical development since that time,
in that it sets the stage for the gap in the excitation spectrum of a supercon
ductor which separates the energy of superfluid electrons from the energy of
electrons in the normal fluid. As Leon Cooper will discuss, this gap plays a
central role in the properties of superconductors.
In his book published in 1950, F. London extended his theoretical conjec
tures by suggesting that a superconductor is a “quantum structure on a macro
scopic scale [which is a] kind of solidification or condensation of the average
momentum distribution” of the electrons. This momentum space condensation
locks the average momentum of each electron to a common value which ex
tends over appreciable distance in space. A specific type of condensation in
momentum space is central to the work Bardeen, Cooper and I did together.
It is a great tribute to the insight of the early workers in this field that many
of the important general concepts were correctly conceived before the micro
scopic theory was developed. Their insight was of significant aid in our own
work.
The phenomenological London theory was extended in 1950 by Ginzburg
and Landau (5) to include a spatial variation of q8. They suggested that
Qs/ q be written in terms of a phenomenological condensate wavefunction ip(r)
as g«(r)/p = |v;(r)|2 and that the free energy difference AF between the
superconducting and normal states at temperature T be given by
AF . , _ (F +
. . h*
“ A(r) ) V>(r) - a ( T ) \v (r)\* + bS Il ip(r) •d*r (3)
' - / { i2m
where e, m, a and b are phenomenological constants, with a(Tc) = 0.
They applied this approach to the calculation of boundary energies between
normal and superconducting phases and to other problems.
As John Bardeen will discuss, a significant step in understanding which forces
cause the condensation into the superfluid came with the experimental discov
ery of the isotope effect by E. M axwell and, independently, by Reynolds, et al.
(6). Their work indicated that superconductivity arises from the interaction
of electrons with lattice vibrations, or phonons. Quite independently, Herbert
Frohlich (7) developed a theory based on electron-phonon interactions which
yielded the isotope effect but failed to predict other superconducting properties.
A somewhat similar approach by Bardeen (8) stimulated by the isotope effect
experiments also ran into difficulties. N. Bohr, W. Heisenberg and other
distinguished theorists had continuing interest in the general problem, but met
with similar difficulties.
An important concept was introduced by A. B. Pippard (9) in 1953. On the
basis of a broad range of experimental facts he concluded that a coherence
length £ is associated with the superconducting state such that a perturbation
of the superconductor at a point necessarily influences the superfluid within a
distance £ of that point. For pure metals, £ ~ 10~4 cm. for T Tc. H egener-
269
alized the London equation (3) to a non-local form and accounted for the
fact that the experimental value of the penetration depth is several times
larger than the London value. Subsequently, Bardeen (10) showed that
Pippard’s non-local relation would likely follow from an energy gap model.
A major problem in constructing a first principles theory was the fact that
the physically important condensation energy AF amounts typically to only
10-8 electron volts (e.V.) per electron, while the uncertainty in calculating
the total energy of the electron-phonon system in even the normal state
amounted to of order 1 e.V. per electron. Clearly, one had to isolate those
correlations peculiar to the superconducting phase and treat them accurately,
the remaining large effects presumably being the same in the two phases and
therefore cancelling. Landau’s theory of a Fermi liquid (11), developed to
account for the properties of liquid H e3, formed a good starting point for such a
scheme. Landau argued that as long as the interactions between the particles
(H e3 atoms in his case, electrons in our case) do not lead to discontinuous
changes in the microscopic properties of the system, a “quasi-particle” de
scription of the low energy excitations is legitim ate; that is, excitations of the
fully interacting normal phase are in one-to-one correspondence with the
excitations of a non-interacting fermi gas. The effective mass m and the Fermi
velocity vp of the quasi-particles differ from their free electron values, but aside
from a weak decay rate which vanishes for states at the Fermi surface there is
no essential change. It is the residual interaction between the quasi-particles
which is responsible for the special correlations characterizing superconductivi
ty. The ground state wavefunction of the superconductor \p0is then represented
by a particular superposition of these normal state configurations, &n.
A clue to the nature of the states @n entering strongly in yi0 is given by com
bining Pippard’s coherence length £ with Heisenberg’s uncertainty principle
Ap ~ H/g ~ 10-*pF (4)
where pp is the Fermi momentum. Thus, W0 is made up of states with quasi
particles (electrons) being excited above the normal ground state by a
momentum of order Ap. Since electrons can only be excited to states which are
initially empty, it is plausible that only electronic states within a momentum
10-4 pp of the Fermi surface are involved significantly in the condensation,
i.e., about 10-4 of the electrons are significantly affected. This view fits nicely
with the fact that the condensation energy is observed to be of order 10_4@-
ksTc. Thus, electrons within an energy ~ vpAp ~ kT c of the Fermi surface
have their energies lowered by of order kT c in the condensation. In summary,
the problem was how to account for the phase transition in which a condensa
tion of electrons occurs in momentum space for electrons very near the Fermi
surface. A proper theory should automatically account for the perfect conduc
tivity and diamagnetism, as well as for the energy gap in the excitation
spectrum.
II. T h e P a ir in g C o n c e p t
In 1955, stimulated by writing a review article on the status of the theory
of superconductivity, John Bardeen decided to renew the attack on the problem.
270
H e invited Leon Cooper, whose background was in elementary particle physics
and who was at that time working with C. N. Yang at the Institute for Advanced
Study to join in the effort starting in the fall of 1955. I had the good fortune
to be a graduate student of Bardeen at that time, and, having finished my
graduate preliminary work, I was delighted to accept an invitation to join them.
W e focused on trying to understand how to construct a ground state W0
formed as a coherent superposition of normal state configurations 0 n,
W0 = Z a n0 n (5)
n
such that the energy would be as low as possible. Since the energy is given in
terms of the Hamiltonian H by
E0 = (W0, HWo) = 2 an'*On H 0 n)
n,n
(0n'9 (6)
we attempted to make E0minimum by restricting the coefficients an so that only
states which gave negative off-diagonal matrix elements would enter (6). In
this case all terms would add in phase and E0 would be low.
By studying the eigenvalue spectrum of a class of matrices with off-diagonal
elements all of one sign (negative), Cooper discovered that frequently a single
eigenvalue is split off from the bottom of the spectrum. H e worked out the
problem of two electrons interacting via an attractive potential- V above a
quiescent Fermi sea, i.e., the electrons in the sea were not influenced by V and
the extra pair was restricted to states within an energy Hcdd above the Fermi
surface, as illustrated in Fig. 2. As a consequence of the non-zero density of
quasi-particle states JV(0) at the Fermi surface, he found the energy eigenvalue
spectrum for two electrons having zero total momentum had a bound state
split off from the continuum of scattering states, the binding energy being
2
E b = Hcdbc --------- (7)
JV( 0 ) v K '
if the matrix elements of the potential are constant equal to V in the region of
interaction. This important result, published in 1956 (12), showed that, re
gardless of how weak the residual interaction between quasi-particles is, if the
interaction is attractive the system is unstable with respect to the formation of
bound pairs of electrons. Further, if E b is taken to be of order ksTc, the un
certainty principle shows the average separation between electrons in the bound
state is of order 10-4 cm.
W hile Cooper’s result was highly suggestive, a major problem arose. If,
as we discussed above, a fraction 10-4 of the electrons is significantly involved
in the condensation, the average spacing between these condensed electrons
271
is roughly 10-# cm. Therefore, within the volume occupied by the bound state
o f a given pair, the centers o f approximately (lCM/lO-6)8 £ 10® other pairs will
be found, on the average. Thus, rather than a picture of a dilute gas of strongly
bound pairs, quite the opposite picture is true. The pairs overlap so strongly
in space that the mechanism of condensation would appear to be destroyed
due to the numerous pair-pair collisions interrupting the binding process of
a given pair.
Returning to the variational approach, we noted that the matrix elements
(&n'i H<Pn) in (6) alternate randomly in sign as one randomly varies n and ri
over the normal state configurations. Clearly this cannot be corrected to obtain
a low value of E0 by adjusting the sign of the an’s since there are JV2 matrix
elements to be corrected with only j\f parameters an. We noticed that if the
sum in (6) is restricted to include only configurations in which, if any quasi
particle state, say k , s} is occupied (s = j or j is the spin index), its “m ate”
state k , s is also occupied, then the matrix elements of H between such states
would have a unique sign and a coherent lowering of the energy would be
obtained. This correlated occupancy of pairs of states in momentum space is
consonant with London’s concept of a condensation in momentum.
In choosing the state k , s to be paired with a given state kf s , it is important
to note that in a perfect crystal lattice, the interaction between quasi-particles
conserves total (crystal) momentum. Thus, as a given pair of quasi-particles
interact, their center of mass momentum is conserved. To obtain the largest
number of non-zero matrix elements, and hence the lowest energy, one must
choose the total momentum o f each pair to be the same, that is
k-{-k = q. (8)
States with q # 0 represent states with net current flow. The lowest energy
state is for q = 0, that is, the pairing is such that if any state A:j is occupied in
an admissible 0 n , so is—A;j occupied. The choice of jj spin pairing is not
restrictive since it encompasses triplet and singlet paired states.
Through this reasoning, the problem was reduced to finding the ground state
of the reduced Hamiltonian
tfred = 2 e* 71*. - 2 V f t bk'+bk. (9)
ks kk'
The first term in this equation gives the unperturbed energy of the quasi
particles forming the pairs, while the second term is the pairing interaction
in which a pair of quasi-particles in (A;j, —A:j) scatter to (£'J, —k’J). The
operators b\ = ckJ C- b e i n g a product of two fermion (quasi-particle)
creation operators, do not satisfy Bose statistics, since bk+2 = 0. This point is
essential to the theory and leads to the energy gap being present not only for
dissociating a pair but also for making a pair move with a totalmomentum
different from the common momentum of the rest of the pairs. It is this feature
which enforces long range order in the superfluid over macroscopic distances.
III. T h e G ro u n d S t a t e
In constructing the ground state wavefunction, it seemed clear that the average
occupancy of a pair state (Arf, —£[) should be unity for k far below the Fermi
272
surface and 0 for k far above it, the fall off occurring symmetrically about kp
over a range of momenta of order
~ 104 cm-1.
One could not use a trial W0 as one in which each pair state is definitely oc
cupied or definitely empty since the pairs could not scatter and lower the
energy in this case. Rather there had to be an amplitude, say vjc, that (£j,
—A;j) is occupied in and consequently an amplitude ujc = J\ —VJ* that the
pair state is empty. After we had made a number of unsuccessful attempts
to construct a wavefunction sufficiently simple to allow calculations to be
carried out, it occurred to me that since an enormous number ( ~ 1019) of pair
states (A;'|, —A;'j) are involved in scattering into and out of a given pair state
(yfcj, —&[), the “instantaneous” occupancy of this pair state should be essen
tially uncorrelated with the occupancy of the other pair states at that “instant”.
Rather, only the average occupancies of these pair states are related.
On this basis, I wrote down the trial ground state as a product of operators
—one for each pair state—acting on the vacuum (state of no electrons),
= n {uk+vkbk) |0>, (10)
where m* = J i _ ^ 2 Since the pair creation operators bk+ commute for different
k?s, it is clear that represents uncorrelated occupancy of the various pair
states. I recall being quite concerned at tfye time that was an admixture of
states with different numbers of electrons, a wholly new concept to me, and as
I later learned to others as well. Since by varying vjc the mean number of
electrons varied, I used a Lagrange multiplier (the chemical potential) to
make sure that the mean number of electrons (JVop) represented by was the
desired number N. Thus by minimizing
E0—/j,N = (!P0, [//r e d -^ o p ]^ o )
with respect to vjc, I found that vjc was given by
z>* —1
where
Ek —J (ek—fiy+Ajc2 (12)
and the parameter Ak satisfied what is now called the energy gap equation:
(13)
From this expression, it followed that for the simple model
{
F, |e* —//| CncoB and |e*'—/J < Hcojy
0 ..
Vk'K u .
, otherwise
Z,= ' <0De“ J ^ ) F (14)
273
and the condensation energy at zero temperature is
AF = \N{0)A* (15)
The idea occurred to me while I was in New York at the end of January,
1957, and I returned to Urbana a few days later where John Bardeen quickly
recognized what he believed to be the essential validity of the scheme, much to
my pleasure and amazement. Leon Cooper will pick up the story from here to
describe our excitement in the weeks that followed, and our pleasure in un
folding the properties of the excited states.
IV. Q u an tu m P h en om en a o n a M a c r o sc o p ic S c a le
Superconductors are remarkable in that they exhibit quantum effects on a
broad range of scales. The persistence of current flow in a loop of wire many
meters in diameter illustrates that the pairing condensation makes the super
fluid wavefunction coherent over macroscopic distances. On the other hand,
the absorption of short wavelength sound and light by a superconductor is
sharply reduced from the normal state value, as Leon Cooper will discuss.
I will concentrate on the large scale quantum effects here.
The stability of persistent currents is best understood by considering a cir
cular loop of superconducting wire as shown in Fig. 3. For an ideal small
diameter wire, one would use the eigenstates e<m0, (m = 0 ,± 1 ,± 2 , . . .), of the
angular momentum Lz about the symmetry axis to form the pairing. In the
ground state no net current flows and one pairs m f with — mj, instead of
with —£[ as in a bulk superconductor. In both cases, the paired states are
time reversed conjugates, a general feature of the ground state. In a current
carrying state, one pairs (m-fv)J with (— (v — 0 ,=b l , i 2 . . .), so
that the total angular momentum of each pair is identical, 2 hv. It is this com
m onality of the center of mass angular momentum of each pair which preserves
the condensation energy and long range order even in states with current flow.
Another set of flow states which interweave with these states is formed by
pairing (m +v)J with (—m + r + l)J, (v = 0 ,± 1 ,± 2 . . .), with the pair
angular momentum being (2v-{-l)H. The totality of states forms a set with all
integer multiples n of h for allowed total angular momentum of pairs. Thus,
even though the pairs greatly overlap in space, the system exhibits quantiza
tion effects as if the pairs were well defined.
There are two important consequences of the above discussion. First, the
fact that the coherent condensate continues to exist in flow states shows that
to scatter a pair out of the (rotating) condensate requires an increase of energy.
274
Crudely speaking, slowing down a given pair requires it ot give up its binding
energy and hence this process will occur only as a fluctuation. These fluctua
tions average out to zero. The only way in which the flow can stop is if all pairs
simultaneously change their pairing condition from, say, v to v — 1. In this
process the system must fluctate to the normal state, at least in a section of the
wire, in order to change the pairing. This requires an energy of order the
condensation energy AF. A thermal fluctuation of this size is an exceedingly
rare event and therefore the current persists.
The second striking consequence of the pair angular momentum quantization
is that the magnetic flux 0 trapped within the loop is also quantized,
<Pn= n ~
2e (* = 0,±1,±2...). (16)
This result follows from the fact that if the wire diameter d is large compared
to the penetration depth A, the electric current in the center of the wire is
essentially zero, so that the canonical angular momentum of a pair is
2e
•f-pair == —C fpair X A (17)
where rpair is the center of mass coordinate of a pair and A is the magnetic
vector potential. If one integrates Lpair, around the loop along a path in the
center of the wire, the integral is nh, while the integral of the right hand side of
2e ^
(17) is — 0 .
c
A similar argument was given by F. London (4b) except that he considered
only states in which the superfluid flows as a whole without a change in its
internal strucutre, i.e., states analogous to the (iw +v)|, (— w + r) J set. H e found
0 z = n-hcje. The pairing (m + r)j, (m -fv + l)j cannot be obtained by adding
v to each state, yet this type of pairing gives an energy as low as the more
conventional flow states and these states enter experimentally on the same basis
as those considered by London. Experiments by Deaver and Fairbank (13),
and independently by Doll and Nabauer (14) confirmed the flux quantization
phenomenon and provided support for the pairing concept by showing that
2e rather than e enters the flux quantum. Following these experiments a clear
discussion of flux quantization in the pairing scheme was given by Beyers and
Yang (15).
The idea that electron pairs were somehow important in superconductivity
has been considered for some time (16, 17). Since the superfluidity of liquid
H e4 is qualitatively accounted for by Bose condensation, and since pairs of
electrons behave in some respects as a boson, the idea is attractive. The
essential point is that while a dilute gas of tightly bound pairs of electrons might
behave like a Bose gas (18) this is not the case when the mean spacing between
pairs is very small compared to the size of a given pair. In this case the inner
structure of the pair, i.e., the fact that it is made o f fermions, is essential;
it is this which distinguishes the pairing condensation, with its energy gap for
single pair translation as well as dissociation, from the spectrum of a Bose con-
275
OXIDE BARRIER = 0
densate, in which the low energy exictations are Bose-like rather than Fermi-
like as occurs in acutal superconductors. As London emphasized, the con
densation is an ordering in occupying momentum space, and not a space-like
condensation of clusters which then undergo Bose condensation.
In 1960, Ivar Giaever (19) carried out pioneering experiments in which elec
trons in one superconductor (Sx) tunneled through a thin oxide layer ( ~ 20 —
30 A) to a second superconductor (S2) as shown in Fig. 4. Giaever’s experi
ments were dramatic evidence of the energy gap for quasi-particle excitations.
Subsequently, Brian Josephson made a highly significant contribution by
showing theoretically that a superfluid current could flow between Sx and S2
with zero applied bias. Thus, the superfluid wavefunction is coherent not only
in St and S2 separately, but throughout the entire system, Sx—0 —S2, under
suitable circumstances. W hile the condensate amplitude is small in the oxide,
it is sufficient to lock the phases of Sx and S2 together, as has been discussed in
detail by Josephson (20) and by P. W. Anderson (21).
To understand the meaning of phase in this context, it is useful to go back
to the ground state wavefunction Wo, (10). Suppose we write the parameter v*
as |z>*| exp ito W0. Let this wavefunc
tion, say IfV1* represent Sl3 and have phase <px. Similarly, let !?V2) represent St
and have phase angle q>2. If we write the state of the combined system as a
product
(18)
then by expanding out the double product we see that the phase of that part of
!F0<1»2) which has N x pairs in Sx and N 2 pairs in S2 is N x <px+ N 2<p2. For a truly
isolated system, 2(AiH-A2) = 2JV is a fixed number of electrons; however N x
and JV2 are not separately fixed and, as Josephson showed, the energy of the
combined system is minimized when gol = (p2 due to tunneling of electrons
between the superconductors. Furthermore, if (px = cp2, a current flows between
Sj and S2
j = j x sin(<px—<p2) (19)
If <px—<p2 = <p is constant in time, a constant current flows with no voltage
applied across the junction. If a bias voltage is V applied between St and S2,
then, according to quantum mechanics, the phase changes as
276
2 eV d q>
h ~ dt
(20)
Hence a constant voltage applied across such a junction produces an
alternating current of frequency
v = ------= 483 TH z/V . (21)
h
These effects predicted by Josephson were observed experimentally in a
series of beautiful experiments (22) by many scientists, which I cannot discuss
in detail here for lack of time. I would mention, as an example, the work of
Langenberg and his collaborators (23) at the University of Pennsylvania on
the precision determination of the fundamental constant ejh using the fre
quency-voltage relation obeyed by the alternating Josephson supercurrent.
These experiments have decreased the uncertainty in our experimental knowl
edge of this constant by several orders of magnitude and provide, in combina
tion with other experiments, the most accurate available value of the Sommerfeld
fine structure constant. They have resulted in the resolution of several dis
crepancies between theory and experiment in quantum electrodynamics and
in the development of an “atomic” voltage standard which is now being used
by the United States National Bureau of Standards to maintain the U .S. legal
volt.
V. C o n clu sio n
As I have attempted to sketch, the development of the theory of superconduct
ivity was truly a collaborative effort, involving not only John Bardeen, Leon
Cooper and myself, but also a host of outstanding scientists working over a
period of half a century. As my colleagues will discuss, the theory opened up
the field for many exciting new developments, both scientific and technological,
many of which no doubt lie in the future. I feel highly honored to have played
a role in this work and I deeply appreciate the honor you have bestowed on me
in awarding us the Nobel prize.
R eferences
1. Kamerlingh Onnes, H., Nobel Lectures, Vol. 1, pp. 306 —336.
2. Meissner, W. and Ochsenfeld, R., Naturwiss. 21, 787 (1933).
3. Gorter, C. J. and Gasimir, H. B. G., Phys. Z. 35, 963 (1934); Z. Techn. Phys. 15,
539 (1934).
4. London, F., [a] Phys. Rev. 24, 562 (1948); [b] Superfluids, Vol. 1 (John Wiley & Sons,
New York, 1950).
5. Ginzburg, V. L. and Landau, L. D., J. Exp. Theor. Phys. (U.S.S.R.) 20, 1064 (1950).
6. Maxwell, E., Phys. Rev. 78, M l (1950); Reynolds, C. A., Serin, B., Wright W. H. and
Nesbitt, L. B., Phys. Rev. 78, 487 (1950).
7. Frohlich, H., Phys. Rev. 79, 845 (1950).
277
8. Bardeen, J., Rev. Mod. Phys. 23, 261 (1951).
9. Pippard, A. B., Proc. Royal Soc. (London) A 216, 547 (1953).
10. Bardeen, J., [a] Phys. Rev. 97, 1724 (1955); [b] Encyclopedia of Physics, Vol. 15
(Springer-Verlag, Berlin, 1956), p. 274.
11. Landau, L. D., J. Exp. Theor. Phys. (U.S.S.R.) 30 (3), 1058 (920) (1956); 32 (5),
59 (101) (1957).
12. Cooper, L. N ., Phys. Rev. 104, 1189 (1956).
13. Deaver, B. S. Jr., and Fairbank, W. M., Phys. Rev. Letters 7, 43 (1961).
14. Doll, R. and Nabauer, M ., Phys. Rev. Letters 7, 51 (1961).
15. Beyers, N. and Yang, C. N ., Phys. Rev. Letters 7, 46 (1961).
16. Ginzburg, V. L., Usp. Fiz. Nauk 48, 25 (1952); transl. Fortsch. d. Phys. 1, 101 (1953).
17. Schafroth, M. R., Phys. Rev. 96, 1442 (1954); 100, 463 (1955).
18. Schafroth, M. R., Blatt, J. M. and Butler, S. T., Helv. Phys. Acta 30, 93 (1957).
19. Giaever, I., Phys. Rev. Letters 5, 147 (1960).
20. Josephson, B. D., Phys. Letters 1, 251 (1962); Advan. Phys. 14, 419 (1965).
21. Anderson, P. W ., in Lectures on the Many-body Problem, edited by E. R. Caianiello
(Academic Press, Inc. New York, 1964), Vol. II.
22. See Superconductivity, Parks, R. D., ed. (Dekker New York, 1969).
23. See, for example, Parker, W. H. Taylor B. N. and Langenberg, D. N. Phys. Rev.
Letters 18, 287 (1967); Finnegan, T. F. Denenstein A. and Langenberg, D. N. Phys.
Rev. B4, 1487 (1971).
278
M ICROSCOPIC QUANTUM INTERFERENCE
EFFECTS IN THE THEORY OF
SUPERCONDUCTIVITY
Nobel Lecture, December 11, 1972
by
L eon N C oo per
Physics Department, Brown University, providence, Rhode Island
It is an honor and a pleasure to speak to you today about the theory of super
conductivity. In a short lecture one can no more than touch on the long history
of experimental and theoretical work on this subject before 1957. Nor can one
hope to give an adequate account of how our understanding of superconductivi
ty has evolved since that time. The theory (1) we presented in 1957, applied
to uniform materials in the weak coupling limit so defining an ideal supercon
ductor, has been extended in almost every imaginable direction. To these
developments so many authors have contributed (2) that we can make no
pretense of doing them justice. I will confine myself here to an outline of some
of the main features of our 1957 theory, an indication of directions taken since
and a discussion of quantum interference effects due to the singlet-spin pairing
in superconductors which might be considered the microscopic analogue of the
effects discussed by Professor Schrieffer.
N ormal M eta l
Although attempts to construct an electron theory of electrical conductivity

date from the time of Drude and Lorentz, an understanding of normal metal
conduction electrons in modern terms awaited the development of the quantum
theory. Soon thereafter Sommerfeld and Bloch introduced what has evolved
into the present description of the electron fluid. (3) There the conduction
electrons of the normal metal are described by single particle wave functions.
In the periodic potential produced by the fixed lattice and the conduction
electrons themselves, according to Bloch’s theorem, these are modulated
plane waves:
(r) e“-',
where uK(r) is a two component spinor with the lattice periodicity. W e use
K to designate simultaneously the wave vector Ac, and the spin state cj :K = k, T ;
—K = —Ac, 1 . The single particle Bloch functions satisfy a Schrodinger equa
tion
where F0(r ) is the periodic potential and in general might be a linear operator
to include exchange terms.
The Pauli exclusion principle requires that the many electron wave function
be antisymmetric in all of its coordinates. As a result no two electrons can be
279
The normal ground state wavefunction, An excitation of the normal system.
0 O> is a filled Fermi sphere for both spin
directions.
in the same single particle Bloch state. The energy of the entire system is
2JV
W = E £{
1= 1
where £* is the Bloch energy of the ith single electron state. The ground state
of the system is obtained when the lowest JV Bloch states of each spin are
occupied by single electrons; this can be pictured in momentum space as the
filling in of a Fermi sphere, Fig. 1. In the ground-state wave function there is
no correlation between electrons of opposite spin and only a statistical correla
tion (through the general anti-symmetry requirement on the total wave func
tion) of electrons of the same spin.
Single particle excitations are given by wave functions identical to the ground
state except that one electron states k i < k p are replaced by others k j < kp.
This may be pictured in momentum space as opening vacancies below the
Fermi surface and placing excited electrons above, Fig. 2. The energy difference
between the ground state and the excited state with the particle excitation kj
and the hole excitation ki is
S j — Si = £j — £ p — (£i — £ p ) = £ j — Si = |ty| + |£*|
where we define e as the energy measured relative to the Fermi energy
£i = £i — £ p .
W hen Coulomb, lattice-electron and other interactions, which have been
omitted in constructing the independent particle Bloch model are taken into
account, various modifications which have been discussed by Professor Schrief
fer are introduced into both the ground state wave function and the excitations.
These may be summarized as follows: The normal metal is described by a
ground state 0 O and by an excitation spectrum which, in addition to the
various collective excitations, consists of quasi-fermions which satisfy the usual
anticommutation relations. It is defined by the sharpness of the Fermi surface,
the finite density of excitations, and the continuous decline of the single particle
excitation energy to zero as the Fermi surface is approached.
280
E lectron C orrelations that P roduce S u perco nductivity
For a description of the superconducting phase we expect to include correla
tions that are not present in the normal metal. Professor Schrieffer has discussed
the correlations introduced by an attractive electron-electron interaction and
Professor Bardeen will discuss the role of the electron-phonon interaction in
producing the electron-electron interaction which is responsible for supercon
ductivity. It seems to be the case that any attractive interaction between the
fermions in a many-fermion system can produce a superconducting-like state.
This is believed at present to be the case in nuclei, in the interior of neutron
stars and has possibly been observed (4) very recently in H e3. We will therefore
develop the consequences of an attractive two-body interaction in a degenerate
many-fermion system without enquiring further about its source.
The fundamental qualitative difference between the superconducting and
normal ground state wave function is produced when the large degeneracy of
the single particle electron levels in the normal state is removed. If we visualize
the Hamiltonian matrix which results from an attractive two-body interaction
in the basis of normal metal configurations, we find in this enormous matrix,
sub-matrices in which all single-particle states except for one pair of electrons
remain unchanged. These two electrons can scatter via the electron-electron
interaction to all states of the same total momentum. We may envisage the
pair wending its way (so to speak) over all states unoccupied by other electrons.
[The electron-electron interaction in which we are interested is both weak
and slowly varying over the Fermi surface. This and the fact that the energy
involved in the transition into the superconducting state is small leads us to
guess that only single particle excitations in a small shell near the Fermi
surface play a role. It turns out, further, that due to exchange terms in the
electron-electron matrix element, the effective interaction in metals between
electrons of singlet spin is much stronger than that between electrons of triplet
spin— thus our preoccupation with singlet spin correlations near the Fermi
surface.] Since every such state is connected to every other, if the interaction
is attractive and does not vary rapidly, we are presented with submatrices of
the entire Hamiltonian of the form shown in Fig. 3. For purposes of illustration
we have set all off diagonal matrix elements equal to the constant— V and
the diagonal terms equal to zero (the single particle excitation energy at the
Fermi surface) as though all the initial electron levels were completely degener
ate. Needless to say, these simplifications are not essential to the qualitative
result.
Diagonalizing this matrix results in an energy level structure with M — 1
levels raised in energy to E = -f V while one level (which is a superposition
of all of the original levels and quite different in character) is lowered in energy
to
E = — (M — \)V.
Since M , the number of unoccupied levels, is proportional to the volume
of the container while V, the scattering matrix element, is proportional to
1/volume, the product is independent of the volume. Thus the removal of
281
M COLUMNS
0 -V -V -V
-V 0 -V •
-V-V 0 -V
M ROWS
-V 0 -V
-V -V -V-V 0
-V
- . . . 0
• • ■ • -(M-I)V
For V = 0, For V > 0,M — 1 levels at
M levels E = V and one level
at E = 0 at E = — (M—1) V
Fig. 3.
the degeneracy produces a single level separated from the others by a volume
independent energy gap.
To incorporate this into a solution of the full Hamiltonian, one must devise
a technique by which all of the electrons pairs can scatter while obeying the
exclusion principle. The wave function which accomplishes this has been dis
cussed by Professor Schrieffer. Each pair gains an energy due to the removal of
the degeneracy as above and one obtains the maximum correlation of the entire
wave function if the pairs all have the same total momentum. This gives a
coherence to the wave function in which forva combination of dynamical and
statistical reasons there is a strong preference for momentum zero, singlet spin
correlations, while for statistical reasons alone there is an equally strong
preference that all of the correlations have the same total momentum.
In what follows I shall present an outline of our 1957 theory modified by
introducing the quasi-particles of Bogoliubov and Valatin. (5) This leads to
a formulation which is generally applicable to a wide range of calculations
282
kt
Fig. 4.
The ground state of the superconductor is
a linear superposition of states in which pairs
(Atf —k l ) are occupied or unoccupied.
IS»T
“t
in a manner analogous to similar calculations in the theory o f normal metals.
We limit the interactions to terms which scatter (and thus correlate) singlet
zero-momentum pairs. To do this, it is convenient to introduce the pair
operators:
bk = C-gCK
, * * *
^k — CK C-K
and using these we extract from the full Hamiltonian the so-called reduced
Hamiltonian
•^reduced = E 2 |e | bk bk -j- X <2.ebk bk -f- 2J Vk,kbk,bk
k<kf k>kf kk'
where is the scattering matrix element between the pair states k and k'.
G round S tate
As Professor Schrieffer has explained, the ground state of the superconductor
is a linear superposition of pair states in which the pairs (k T , — k I ) are occupied
or unoccupied as indicated in Fig. 4. It can be decomposed into two disjoint
vectors—one in which the pair state k is occupied, 0 k and one in which it is
unoccupied, 0(ky
V»o =
The probability amplitude that the pair state k is (is not) occupied in the
ground state is then vk(uk). Normalization requires that |m|2 -(-|p |2 = 1 . The
phase of the ground state wave function may be chosen so that with no loss o
generality uk is real. We can then write
u = (1 - h ) 11*
v = h11* eiv
where
0 < h < 1.
A further decomposition of the ground state wave function of the supercon
ductor in which the pair states k and k ' are either occupied or unoccupied
Fig. 5 is:
ip0 = UkUV®{k),(V) + UkVV®(k), k' + VkUk’0k, (k') + VkVk,^ k tV •
This is a Hartree-like approximation in the probability amplitudes for the
occupation of pair states. It can be shown that for a fermion system the wave
283
Fig. 5.
A decomposition of the ground state of the superconductor into states in which the pair
states k and k' are either occupied or unoccupied.
function cannot have this property unless there are a variable number of
particles. To terms of order 1/JV, however, this decomposition is possible for
a fixed number of particles; the errors introduced go to zero as the number of
particles become infinite. (6 )
The correlation energy, WC) is the expectation value of HTe&for the state y)0
Wc = (yj0, Hre^xpo) = WQ [h,(p].
Setting the variation of Wc with respect to h and q>equal to zero in order to
minimize the energy gives
h = 1/2 (1 -e /E )
E = (£2+ |Zl|2) 1/2
where
A = |zl|ei,p
satisfies the integral equation
If a non-zero solution of this integral equation exists, Wc < 0 and the

“normal” Fermi sea is unstable under the formation of correlated pairs.
In the wave function that results there are strong correlations between pairs
of electrons with opposite spin and zero total momentum. These correlations
are built from normal excitations near the Fermi surface and extend over spatial
distances typically of the order of 10-4 cm. They can be constructed due to the
large wave numbers available because of the exclusion principle. Thus with
a small additional expenditure of kinetic energy there can be a greater gain
in the potential energy term. Professor Schrieffer has discussed some of the
properties of this state and the condensation energy associated with it.
S ing le -P article E xcitations

In considering the excited states of the superconductor it is useful, as for the
284
Fig. 6.
A single par icle excitation of the
superconduc or in one-to-one cor-
respondenc with an excitation of
the normal fermion system.
normal metal, to make a distinction between single-particle and collective

excitations; it is the single-particle excitation spectrum whose alteration is
responsible for superfluid properties. For the superconductor excited (quasi
particle) states can be defined in one-to-one correspondence with the excita
tions of the normal metal. One finds, for example, that the expectation value
of / / red for the excitation Fig. 6 is given by
Ek — J £l + |^|2-
In contrast to the normal system, for the superconductor even as e goes to
zero E remains larger than zero, its lowest possible value being E = |zl|.
One can therefore produce single particle excitations from the superconducting
ground state only with the expenditure of a small but finite amount of energy.
This is called the energy gap; its existence severely inhibits single particle
processes and is in general responsible for the superfluid behavior of the electron
gas. [In a gapless superconductor it is the finite value of A (r), the order para
meter, rather than the energy gap as such that becomes responsible for the
superfluid properties.] In the ideal superconductor, the energy gap appears
because not a single pair can be broken nor can a single element of phase
space be removed without a finite expenditure of energy. If a single pair is
broken, one loses its correlation energy; if one removes an element of phase space
from the system, the number of possible transitions of all the pairs is reduced
resulting in both cases in an increase in the energy which does not go to zero
as the volume of the system increases.
The ground state of the superconductor and the excitation spectrum de
scribed above can conveniently be treated by introducing a linear combination
of c* and c, the creation and annihilation operators of normal fermions. This
is the transformation of Bogoliubov and Valatin (5):
7*0 = UkCK ~ VkC- K
7* i = v kcK Jr u kc - K
It follows that
Vk Vo= 0
285
so that the yki play the role of annihilation operators, while the y*ki create
excitations
Yki • • • y*mj Wo = Vw • • • mj‘
T he y operators satisfy Fermi anti-commutation relations so that with them
we obtain a complete orthonormal set of excitations in one-to-one correspon
dence with the excitations of the normal metal.
We can sketch the following picture. In the ground state of the supercon
ductor all the electrons are in singlet-pair correlated states of zero total
momentum. In an m electron excited state the excited electrons are in “quasi
particle” states, very similar to the normal excitations and not strongly
correlated with any of the other electrons. In the background, so to speak, the
other electrons are still correlated much as they were in the ground state.
The excited electrons behave in a manner similar to normal electrons; they
can be easily scattered or excited further. But the background electrons—
those which remain correlated—retain their special behavior; they are difficult
to scatter or to excite.
Thus, one can identify two almost independent fluids. The correlated portion
of the wave function shows the resistance to change and the very small specific
heat characteristic of the superfluid, while the excitations behave very much
like normal electrons, displaying an almost normal specific heat and resistance.
When a steady electric field is applied to the metal, the superfluid electrons
short out the normal ones, but with higher frequency fields the resistive proper
ties of the excited electrons can be observed. [7]
T herm odynamic P roperties , th e I deal S uperco nductor
W e can obtain the thermodynamic properties of the superconductor using
the ground state and excitation spectrum just described. The free energy of
the system is given by
F[h,cpJl = Wc ( T ) - T S ,
where T is the absolute temperature and S is the entropy; f is the super
conducting Fermi function which gives the probability of single-particle ex
citations. The entropy of the system comes entirely from the excitations as
the correlated portion of the wave function is non-degenerate. The free energy
becomes a function of f(k) and h(k)9 where f(k) is the probability that the
state k is occupied by an excitation or a quasi-particle, and h(k) is the relative
probability that the state k is occupied by a pair given that is not occupied by
a quasi-particle. Thus some states are occupied by quasi-particles and the
unoccupied phase space is available for the formation of the coherent back
ground of the remaining electrons. Since a portion of phase space is occupied by
excitations at finite temperatures, making it unavailable for the transitions
of bound pairs, the correlation energy is a function of the temperature, Wc( T).
As T increases, WC(T) and at the same time A decrease until the critical tem
perature is reached and the system reverts to the normal phase.
Since the excitations of the superconductor are independent and in a one-
to-one correspondence with those of the normal metal, the entropy of an
286
excited configuration is given by an expression identical with that for the
normal metal except that the Fermi function, /(& ), refers to quasi-particle
excitations. The correlation energy at finite temperature is given by an expres
sion similar to that at T — 0 with the available phase space modified by the
occupation functions f(k). Setting the variation of F with respect to h, 9 , and
f equal to zero gives:
h = 1/2 ( l-e /E )
E = J e * + \A\‘
and
1
/ = 1+ exp (EjkBT)
where
A = |zl|ei»'
is now temperature-dependent and satisfies the fundamental integral equation
of the theory
d k(T) = - 1 ,,o „ A^ T—) ta
/ 2 2v Vkk.— . n h l— —— I.
*»- E k, ( 7 ) \ 2 k BT /
The form of these equations is the same as that at T — 0 except that the
energy gap varies with the temperature. The equation for the energy gap can
be satisfied with non-zero values of A only in a restricted temperature range.
The upper bound of this temperature range is defined as T c the critical
temperature. For T < TC) singlet spin zero momentum electrons are strongly
correlated, there is an energy gap associated with exciting electrons from the
correlated part of the wave function and E(k) is bounded below by |J|. In
this region the system has properties qualitatively different from the normal
metal.
In the region T > Tc, A — 0 and we have in every respect the normal solu
tion. In particular/, the distribution function for excitations, becomes just the
Fermi function for excited electrons k > h?, and for holes k < k?
^ 1 + exp(|e|/A:BT)'
If we make our simplifications of 1957, (defining in this way an ‘ideal*
superconductor)
FjYk == F jfij < H(d$v
= 0 otherwise
and replace the energy dependent density of states by its value at the Fermi
surface, JV(0), the integral equation for A becomes
fcwav de / /ca 4 . a U\
1 = jV(0) V J - = = = = = tanh l f - ± £ L
oJ + \ 2kBT )
The solution of this equation, Fig. 7, gives A (T ) and with this f and h.
We can then calculate the free energy of the superconducting state and obtain
the thermodynamic properties of the system.
287
0 0.2 0.4 0.6 0.8 1.0
T / Tc
Fig. 7.
Variation of the energy gap with temperature for the ideal superconductor.
In particular one finds that at Tc (in the absence of a magnetic field) there
is a second-order transition (no latent heat: Wc = 0 at T c) and a discontinuity
in the specific heat. At very low temperatures the specific heat goes to zero
exponentially. For this ideal superconductor one also obtains a law of cor
responding states in which the ratio
where
y = 2 / 37r2jV(0 )£B2.
The experimental data scatter about the number 0.170. The ratio of A
to kBT c is given as a universal constant
AlkBTc = 1.75.
There are no arbitrary parameters in the idealized theory. In the region
o f empirical interest all thermodynamic properties are determined by the quanti
ties y and hwav e The first, y, is found by observation of the normal
specific heat, while the second is found from the critical temperature, given by
kBTc = 1.14 t o w e - i'W ,
At the absolute zero
zl =foJav/Sin h ^ J _ ) .
Further, defining a weak coupling limit [JV(0)F 1] which is one region
of interest empirically, we obtain
A ~ 2fc»ave-1/*<°>F.
The energy difference between the normal and superconducting states be
comes (again in the weak coupling limit)
W s-W n = Wc = —2N(0)(ficosw) 2 e- 2 /*<0)F
288
The dependence of the correlation energy on (&oav) 2 gives the isotope effect,
while the exponential factor reduces the correlation energy from the dimen-
sionally expected 7V(0)(^coav) 2 to the much smaller observed value. This,
however, is more a demonstration that the isotope effect is consistent with our
model rather than a consequence of it, as will be discussed further by Professor
Bardeen.
The thermodynamic properties calculated for the ideal superconductor are
in qualitative agreement with experiment for weakly coupled superconductors.
Very detailed comparison between experiment and theory has been made by
many authors. A summary of the recent status may be found in reference (2).
W hen one considers that in the theory of the ideal superconductor the existence
of an actual metal is no more than hinted at (We have in fact done all the
calculations considering weakly interacting fermions in a container.) so that
in principle (with appropriate modifications) the calculations apply to neutron
stars as well as metals, we must regard detailed quantitative agreement as a
gift from above. We should be content if there is a single metal for which such
agreement exists. [Pure single crystals of tin or vanadium are possible candi
dates.]
To make comparison between theory and experiments on actual metals, a
plethora of detailed considerations must be made. Professor Bardeen will
discuss developments in the theory of the electron-phonon interaction and the
resulting dependence of the electron-electron interaction and superconducting
properties on the phonon spectrum and the range of the Coulomb repulsion.
Crystal symmetry, Brillouin zone structure and the actual wave function (S,
P or D states) of the conduction electrons all play a role in determining real
metal behavior. There is a fundamental distinction between superconduc
tors w ich always show a Meissner effect and those (type II) which allow mag
netic field penetration in units of the flux quantum.
When one considers, in addition, specimens with impurities (magnetic and
otherwise) superimposed films, small samples, and so on, one obtains a variety
of situations, developed in the years since 1957 by many authors, whose rich
ness and detail takes volumes to discuss. The theory of the ideal superconductor
has so far allowed the addition of those extensions and modifications necessary
to describe, in what must be considered remarkable detail, all of the experience
actually encountered.
M icroscopic I nterference E ffects
In its interaction with external perturbations the superconductor displays
remarkable interference effects which result from the paired nature of the
wave function and are not at all present in similar normal metal interactions.
Neither would they be present in any ordinary two-fluid model. These “co
herence effects” are in a sense manifestations of interference in spin and
momentum space on a microscopic scale, analogous to the macroscopic
quantum effects due to interference in ordinary space which Professor Schrieffer
discussed. They depend on the behavior under time reversal of the perturbing
fields. (8 ) It is intriguing to speculate that if one could somehow amplify them
289
Fig. 8.
Ultrasonic attenuation as a function of temperature across the superconducting transition
as measured by Morse and Bohm.
properly, the time reversal symmetry of a fundamental interaction might be
tested. Further, if helium 3 does in fact display a phase transition analogous to
the superconducting transition in metals as may be indicated by recent experi
ments (4) and this is a spin triplet state, the coherences effects would be
greatly altered.
Near the transition temperature these coherence effects produce quite dra
matic contrasts in the behavior of coefficients which measure interactions with
the conduction electrons. Historically, the comparison with theory of the be
havior of the relaxation rate of nuclear spins (9) and the attenuation of longi
tudinal ultrasonic waves in clean samples ( 1 0 ) as the temperature is decreased
through 7"c provided an early test of the detailed structure of the theory.
The attenuation of longitudinal acoustic waves due to their interaction
with the conduction electrons in a metal undergoes a very rapid drop ( 1 0 a)
as the temperature drops below T c. Since the scattering of phonons from
“normal” electrons is responsible for most of the acoustic attenuation, a drop
was to be expected; but the rapidity o f the decrease measured by Morse and
Bohm (1 Ob) Fig. 8 was difficult to reconcile with estimates of the decrease in
the normal electron component of a two-fluid model.
The rate of relaxation of nuclear spins was measured by Hebei and Slichter
(9a) in zero magnetic field in superconducting aluminum from 0.94 K to
4.2 K just at the time of the development of our 1957 theory. Redfield and
Anderson (9b) confirmed and extended their results. The dominant relaxation
mechanism is provided by interaction with the conduction electrons so that
one would expect, on the basis of a two-fluid model, that this rate should
290
decrease below the transition temperature due to the diminishing density of
“normal” electrons. The experimental results however show just the reverse.
The relaxation rate does not drop but increases by a factor of more than two
just below the transition temperature. Fig 13. This observed increase in the
nuclear spin relaxation rate and the very sharp drop in the acoustic attenuation
coefficient as the temperature is decreased through T c impose contradictory
requirements on a conventional two-fluid model.
To illustrate how such effects come about in our theory, we consider the transi
tion probability per unit time of a process involving electronic transitions from
the excited state k to the state k' with the emission to or absorption of energy from
the interacting field. W hat is to be calculated is the rate of transition between
an initial state |i > and a final state |/ > with the absorption or emission of
the energy Ha)\k,-k\ (a phonon for example in the interaction of sound waves
with the superconductor). All of this properly summed over final states and
averaged with statistical factors over initial states may be w ritten:
2n I e x p (-W ,lk BT)\ < f \ H m |f > |* d(Wf - W , )
co = —h —--------------------------------------------------------------------
E exp (— WilksT)
W e focus our attention on the matrix element i > . This typically
contains as one of its factors matrix elements between excited states of the
superconductor of the operator
B = E BK'KCK'CK
KK'
where c*K, and cK are the creation and annihilation operators for electrons in
the states K' and K , and BK,K is the matrix element between the states K' and
K of the configuration space operator B(r)
BK,K = < K '\B ( r ) \K > .
The operator B is the electronic part of the matrix element between the full
final and initial state
< /|M n t|* '> = mft< f \ B \ i > .
In the normal system scattering from single-particle electron states K to K'
is independent of scattering from — K' to — K. But the superconducting states
are linear superpositions of (K,—K) occupied and unoccupied. Because of this
states with excitations k f and k' j are connected not only by but also
by c*kic_k,i ; if the state |f > contains the single-particle excitation k' | while
the state |z > contains k t, as a result of the superposition of occupied and
unoccupied pair states in the coherent part of the wave function, these are
connected not only by BK,K c*K,cK but also by B-K_K, c*Kc_K,.
For operators which do not flip spins we therefore write:
B = E (B k ,k ck,ck -j- B -k-.k, c_kc_k,).
k k'
Many of the operators, B , we encounter (e.g., the electric current, or the
charge density operator) have a well-defined behavior under the operation
of time reversal so that
B r 'k = i B_k _k, = BWk.
291
1
I
I Js ll> *
P P
Ufc. | v
P
k#
I
~ I
P P
Fig. 9.
The two states |t > and < f\ shown are connected by with the amplitude uvuk.
Then B becomes
B — Z Bk,k (cxfCkf i c_k^c_k,i)
kk'
where the upper (lower) sign results for operators even (odd) under time
reversal.
The matrix element of B between the initial state, tp . . . . . ., and the
final state y) . . . *,t . . . contains contributions from c*k,^ck^ Fig. 9 and un
expectedly from Fig. 10. As a result the matrix element squared
i > |2 contains terms of the form
|s *'k|2 |K-“*T ivO |a>
where the sign is determined by the behavior of B under time reversal:
upper sign B even under time reversal
lower sign B odd under time reversal.
Applied to processes involving the emission or absorption of boson quanta
such as phonons or photons, the squared matrix element above is averaged
with the appropriate statistical factors over initial and summed over final
states; substracting emission from absorption probability per unit time, we
obtain typically
292
P\M
Fig. 10.
The two states |t > and < / 1 are also connected by c*u^C-vi with the amplitude —vyvi •
a = — |w |2 E (/»■-/») H Ev - Ek—h(o
n kw
where f k is the occupation probability in the superconductor for the excitation
let or lej. [In the expression above we have considered only quasiparticle or
quasi-hole scattering processes (not including processes in which a pair of
excitations is created or annihilated from the coherent part of the wave
function) since < A, is the usual region of interest for the ultrasonic
attenuation and nuclear spin relaxation we shall contrast.]
For the ideal superconductor, there is isotropy around the Fermi surface
and symmetry between particles and holes; therefore sums of the form E can
k
be converted to integrals over the superconducting excitation energy, E:
E
E - • 2JV(0) dE
k E 2— A 2
E E
where jV(0 ) - y = = = = jV(0 ) —is the density of excitations in the super
conductor, Fig. 11.
293
Fig. 11.
i Ratio of superconducting to normal den-
3 E/A sity of excitations as a function of EfA.
The appearance of this density of excitations is a surprise. Contrary to our

intuitive expectations, the onset of superconductivity seems initially to enhance
rather than diminish electronic transitions, as might be anticipated in a reason
able two-fluid model.
But the coherence factors \(u'u^v'v*)^ are even more surprising; they behave
in such a way as to sometimes completely negate the effect of the increased
density of states. This can be seen using the expressions obtained above for u
and v for the ideal superconductor to obtain
(;u'uTv'v ) 2 = -1( /,H ee'TA2\
------ |.
V ' 2\ EE )
In the integration over k and k f the ee' term vanishes. W e thus define
(u'u^fv’v) *; in usual limit where H c o A , e ~ e and E ~ E \ this be
comes
1/ £2\
(u2—v2)\ M £ 2) operators even under time reversal
(m2 + z;2)* E2*\

\ operators odd under time reversal.
For operators even under time reversal, therefore, the decrease of the co
herence factors near e = 0 just cancels the increase due to the density of states.
For the operators odd under time reversal the effect of the increase of the densi
ty of states is not cancelled and should be observed as an increase in the rate
of the corresponding process.
In general the interaction Hamiltonian for a field interacting with the super
conductor (being basically an electromagnetic interaction) is invariant under
the operation of time reversal. However, the operator B might be the electric
current j(r) (for electromagnetic interactions) the electric charge density
g(r) (for the electron-phonon interaction) or the z component of the electron
spin operator, oz (for the nuclear spin relaxation interaction). Since under
time-reversal
y(r, t) -> — j(r, —t) (electromagnetic interaction)
g(r, t) -+ -f g(r, —t) (electron-phonon interaction)
<Jz(t) -+ — az(—t) (nuclear spin relaxation interaction)
these show strikingly different interference effects.
294
Fig. 12.
Comparison of observed ultrasonic attenuation with the ideal theory. The data are due to
Morse and Bohtn.
Ultrasonic attenuation in the ideal pure superconductor for ql > 1 (the

product of the phonon wave number and the electron mean free path) depends
in a fundamental way on the absorption and emission o f phonons. Since the
matrix elements have a very weak dependence on changes near the Fermi
surface in occupation of states other than k or k' that occur in the normal to
superconducting transition, calculations within the quasi-particle model can
be compared in a very direct manner with similar calculations for the normal
metal, as Bk,k is the same in both states. The ratio of the attenuation in the
normal and superconducting states becom es:
. l/e
Since (m2—y2)s-> -l —\2
I , the coherence factors cancel the density of states
giving
^ = 2f { A ( T ) ) = ---------- -
Qn 1 + exp
Morse and Bohm (10b) used this result to obtain a direct experimental
determination of the variation of A with T. Comparison of their attenuation
data with the theoretical curve is shown in Figure 12.
295
In contrast the relaxation of nuclear spins which have been aligned in a
magnetic field proceeds through their interaction with the magnetic moment
o f the conduction electrons. In an isotropic superconductor this can be shown
to depend upon the z component of the electron spin operator
^ K 'K — B ( ck'\ck} c-klrc -k 'l)
so that
&KK — —B-k-k'-
This follows in general from the property of the spin operator under time
reversal
Oz(t) = — Gz( — t).
The calculation of the nuclear spin relaxation rate proceeds in a manner

not too different from that for ultrasonic attenuation resulting finally in a
ratio of nuclear spin relaxation rates in superconducting and normal states
in the same sample:
Rs * t / E \ 2 df(E )
- - -4 j d s o , a j .
But (k2 +z/2)s does not go to zero at the lower lim it so that the full effect of
the increase in density of states at E = A is felt. Taken literally, in fact, this
expression diverges logarithmically at the lower limit due to the infinite density
of states. W hen the Zeeman energy difference between the spin up and spin
down states is included, the integral is no longer divergent but the integrand
is much too large. Hebei and Slichter, by putting in a broadening of levels
phenomenologically, could produce agreement between theory and experi
ment. More recently Fibich (11) by including the effect of thermal phonons
has obtained the agreement between theory and experiment shown in Fig. 13.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Fig. 13.
Comparison of observed nuclear spin relaxation rate with theory. The circles represent
experimental data of Hebei and Slichter, the crosses data by Redfield and Anderson.
296
Interference effects manifest themselves in a similar manner in the interac
tion of electromagnetic radiation with the superconductor. Near T c the absorp
tion is dominated by quasi-particle scattering matrix elements of the type we
have described. Near T — 0, the number of quasi-particle excitations goes
to zero and the matrix elements that contribute are those in which quasi
particle pairs are created from y)0. For absorption these latter occur only
when H > 2A. For the linear response of the superconductor to a static
cd
magnetic field, the interference occurs in such a manner that the paramagnetic
contribution goes to zero leaving the diamagnetic part which gives the Meiss
ner effect.
The theory developed in 1957 and applied to the equilibrium properties
of uniform materials in the weak coupling region has been extended in numer
ous directions by many authors. Professor Schrieffer has spoken of Josephson
junctions and macroscopic quantum interference effects; Professor Bardeen
will discuss the modifications of the theory when the electron-phonon inter
actions are strong. The treatment of ultrasonic attenuation, generalized to
include situations in uniform superconductors in which ql < 1 , gives a sur
prisingly similar result to that above. (12) There have been extensive de
velopments using Green’s function methods (13) appropriate for type II super
conductors, materials with magnetic impurities and non-uniform materials or
boundary regions where the order parameter is a function of the spatial co
ordinates. (14) With these methods formal problems of gauge invariance and/or
current conservation have been resolved in a very elegant manner. (15) In
addition, many calculations (16) of great complexity and detail for type II
superconductors have treated ultrasonic attenuation, nuclear spin relaxation
and other phenomena in the clean and dirty limits (few or large numbers of
impurities). The results cited above are modified in various ways. For example,
the average density of excitation levels is less sharply peaked at T c in a type II
superconductor; the coherence effects also change somewhat in these altered
circumstances but nevertheless play an important role. Overall one can say
that the theory has been amenable to these generalizations and that agreement
with experiment is good.
It is now believed that the finite many-nucleon system that is the atomic
nucleus enters a correlated state analogous to that of a superconductor. (17)
Similar considerations have been applied to many-fermion systems as diverse
as neutron stars, (18) liquid H e3, (19) and to elementary fermions. (20) In
addition the idea of spontaneously broken symmetry of a degenerate vacuum
has been applied widely in elementary particle theory and recently in the
theory of weak interactions. (21) W hat the electron-phonon interaction has
produced between electrons in metals may be produced by the van der Waals
interaction between atoms in H e3, the nuclear interaction in nuclei and neutron
stars, and the fundamental interactions in elementary fermions. W hatever the
success of these attempts, for the theoretician the possible existence of this
correlated paired state must in the future be considered for any degenerate
many-fermion system where there is some kind of effective attraction between
fermions for transitions near the Fermi surface.
297
In the past few weeks my colleagues and I have been asked many tim es:
“W hat are the practical uses of your theory?” Although even a summary in
spection o f the proceedings of conferences on superconductivity and its appli
cations would give an immediate sense of the experimental, theoretical and
developmental work in this field as well as expectations, hopes and anticipa
tions—from applications in heavy electrical machinery to measuring devices
of extraordinary sensitivity and new elements with very rapid switching speeds
for computers — I, personally, feel somewhat uneasy responding. The discovery
of the phenomena and the development of the theory is a vast work to which
many scientists have contributed. In addition there are numerous practical
uses of the phenomena for which theory rightly should not take credit. A
theory (though it may guide us in reaching them) does not produce the trea
sures the world holds. And the treasures themselves occasionally dazzle our
attention; for we are not so wealthy that we may regard them as irrelevant.
But a theory is more. It is an ordering of experience that both makes ex
perience meaningful and is a pleasure to regard in its own right. Henri Poin-
car£ wrote (2 2 ):
Le savant doit ordonner; on fait la science
avec des faits comme une maison avec des
pierres; mais une accumulation de faits
n’est pas plus une science qu’un tas de
pierres n ’est une maison.
One can build from ordinary stone a humble house or the finest chateau.
Either is constructed to enclose a space, to keep out the rain and the cold.
T hey differ in the ambition and resources of their builder and the art by which
he has achieved his end. A theory, built of ordinary materials, also may serve
many a humble function. But when we enter and regard the relations in the
space of ideas, we see columns of remarkable height and arches of daring
breadth. They vault the fine structure constant, from the magnetic moment
of the electron to the behavior of metallic junctions near the absolute zero;
they span the distance from materials at the lowest temperatures to those in
the interior of stars, from the properties of operators under time reversal to the
behavior of attenuation coefficients just beyond the transition temperature.
I believe that I speak for my colleagues in theoretical science as well as
myself when I say that our ultimate, our warmest pleasure in the midst of one
o f these incredible structures comes with the realization that what we have
made is not only useful but is indeed a beautiful way to enclose a space.
R eferences and N otes

1. Bardeen, J., Cooper, L. N. and Schrieffer, J. R ., Phys. Rev. 108, 1175 (1957).
2. An account of the situation as of 1969 may be found in the two volum es: Superconductiv
ity, edited by R. D. Parks, Marcel Dekker, Inc., New York City (1969).
3. Sommerfeld, A., Z. Physik 47, 1 (1928). Bloch, F., Z. Physik 52 555 (1928).
4. Osheroff, D. D., Gully, W. J., Richardson R. C. and Lee, D. M., Phys. Rev. Letters 29,
920 (1972); Leggett, A. J., Phys. Rev. Letters 29, 1227 (1972).
298
5. Bogoliubov, N. N ., Nuovo Cimento 7, 794 (1958); Usp. Fiz. Nauk 67, 549 (1959);
Valatin, J. G., Nuovo Cimento 7, 843 (1958).
6. Bardeen, J. and Rickayzen, G., Phys. Rev. 118, 936 (1960); Mattis, D. C. and Lieb, E.,
J. Math. Phys. 2, 602 (1961); Bogoliubov, N. N ., Zubarev D. N. and Tserkovnikov, Yu.
A., Zh. Eksperim. i Teor. Fiz. 39, 120 (1960) translated: Soviet Phys. JETP 12, 88
(1961).
7. For example, Glover, R. E. I ll and Tinkham, M ., Phys. Rev. 108, 243 (1957); Biondi,
M. A. and Garfunkel, M. P., Phys. Rev. 116, 853 (1959).
8. The importance of the coupling of time reversed states in constructing electron pairs
was emphasized by P. W. Anderson,; for example, Anderson, P. W ., J. Phys. Chem.
Solids 11, 26 (1959)
9. a) Hebei, L. G. and Slichter, C. P., Phys. Rev. 113, 1504 (1959).
b) Redfield, A. G. and Anderson, A. G., Phys. Rev. 116,583 (1959).
10. a) Bommel, H. E., Phys. Rev. 96, 220 (1954).
b) Morse, R. W. and Bohm, H. V ., Phys. Rev. 108, 1094 (1957).
11. Fibich, M ., Phys. Rev. Letters 14, 561 (1965).
12. For example, Tsuneto T., Phys. Rev. 121, 402 (1961).
13. Gor’kov, L. P., Zh. Eksperim. i Teor. Fiz. 34, 735 (1958) translated: Soviet Physics
JETP 7, 505 (1958); also Martin P. G. and Schwinger J., Phys. Rev. 115, 1342 (1959);
Kadanoff L. P. and Martin P. C., Phys. Rev. 124, 670 (1961).
14. Abrikosov, A. A. and Gor’kov, L. P., Zh. Eksperim. i Teor. Fiz. 39, 1781 (1960) trans
lated: Soviet Physics JETP 12, 1243 (1961); de Gennes P. G., Superconductivity of Me
tals and Alloys, Benjamin, New York (1966).
15. For example, Ambegaokar V. and Kadanoff L. P., Nuovo Cimento 22, 914 (1961).
16. For example, Caroli C. and Matricon J., Physik Kondensierten Materie 3, 380 (1965);
Maki K., Phys. Rev. 141, 331 (1966); 156, 437 (1967); Groupe de Supraconductivit6
d’Orsay, Physik Kondensierten Materie 5, 141 (1966); Eppel D., Pesch W. and Te-
wordt L., Z. Physik 197; 46 (1966); McLean F. B. and Houghton A., Annals of Physics
48, 43 (1968).
17. Bohr, A., Mottelson, B. R. and Pines, D., Phys. Rev. 110, 936 (1958); Migdal, A. B.,
Nuclear Phys. 13, 655 (1959).
18. Ginzburg, V. L. and Kirzhnits, D. A., Zh. Eksperim. i Theor. Fiz. 47, 2006 (1964)
translated: Soviet Physics JETP 20, 1346 (1965); Pines D., Baym G. and Pethick C.,
Nature 224, 673 (1969).
19. Many authors have explored the possibility of a superconducting-like transition in He*.
Among the most recent contributions see reference 4.
20. For example. Nambu Y. and Jona-Lasinio G., Phys. Rev. 122, 345 (1961).
21. Goldstone, J., Nuovo Cimento 19, 154 (1961); Weinberg, S., Phys. Rev. Letters 19,
1264 (1967).
22. Poincar^, H., La Science et l’Hypoth£se, Flammarion, Paris, pg. 168 (1902). “The
scientist must order; science is made with facts as a house with stones; but an accumula
tion of facts is no more a science than a heap of stones is a house.”
299
ELECTRON-PHONON INTERACTIONS AND
SUPERCONDUCTIVITY
Nobel Lecture, December 11, 1972
By J ohn B a r deen
Departments of Physics and of Electrical Engineering
University of Illinois
Urbana, Illinois
1
I ntroductio n
Our present understanding of superconductivity has arisen from a close
interplay of theory and experiment. It would have been very difficult to have
arrived at the theory by purely deductive reasoning from the basic equations
of quantum mechanics. Even if someone had done so, no one would have be
lieved that such remarkable properties would really occur in nature. But, as
you well know, that is not the way it happened, a great deal had been learned
about the experimental properties of superconductors and phenomenological
equations had been given to describe many aspects before the microscopic
theory was developed. Some of these have been discussed by Schrieffer and
by Cooper in their talks.
M y first introduction to superconductivity came in the 1930’s and I greatly
profited from reading David Shoenberg’s little book on superconductivity, [ 1 ]
which gave an excellent summary of the experimental findings and of the
phenomenological theories that had been developed. At that time it was
known that superconductivity results from a phase change of the electronic
structure and the Meissner effect showed that thermodynamics could be
applied successfully to the superconductive equilibrium state. The two fluid
Gorter— Casimir model was used to describe the thermal properties and the
London brothers had given their famous phenomenological theory of the
electrodynamic properties. Most impressive were Fritz London’s speculations,
given in 1935 at a meeting of the Royal Society in London, [2] that super
conductivity is a quantum phenomenon on a macroscopic scale. He also gave
what may be the first indication of an energy gap when he stated that “the
electrons be coupled by some form of interaction in such a way that the
lowest state may be separated by a finite interval from the excited ones.”
H e strongly urged that, based on the Meissner effect, the diamagnetic aspects
of superconductivity are the really basic property.
M y first abortive attempt to construct a theory, [3] in 1940, was strongly
influenced by London’s ideas and the key idea was small energy gaps at the
Fermi surface arising from small lattice displacements. However, this work
was interrupted by several years of wartime research, and then after the war
I joined the group at the Bell Telephone Laboratories where my work turned
to semiconductors. It was not until 1950, as a result of the discovery of the
300
isotope effect, that I again began to become interested in superconductivity,
and shortly after moved to the University of Illinois.
The year 1950 was notable in several respects for superconductivity theory.
The experimental discovery of the isotope effect [4, 5] and the independent
prediction of H. Frohlich [6 ] that superconductivity arises from interaction
between the electrons and phonons (the quanta of the lattice vibrations) gave
the first clear indication of the directions along which a microscopic theory
might be, sought. Also in the same year appeared the phenomenological
Ginzburg— Landau equations which give an excellent description of super
conductivity near T c in terms of a complex order parameter, as mentioned
by Schrieffer in his talk. Finally, it was in 1950 that Fritz London’s book [7]
on superconductivity appeared. This book included very perceptive comments
about the nature of the microscopic theory that have turned out to be re
markably accurate. He suggested that superconductivity requires “a kind of
solidification or condensation of the average momentum distribution.” He
also predicted the phenomenon of flux quantization, which was not observed
for another dozen years.
The field of superconductivity is a vast one with many ramifications. Even
in a series of three talks, it is possible to touch on only a few highlights. In
this talk, I thought that it might be interesting to trace the development of
the role of electron-phonon interactions in superconductivity from its begin
nings in 1950 up to the present day, both before and after the development
of the microscopic theory in 1957. By concentrating on this one area, I hope
to give some impression of the great progress that has been made in depth
of understanding of the phenomena of superconductivity. Through develop
ments by many people, [8 ] electron-phonon interactions have grown from a
qualitative concept to such an extent that measurements on superconductors
are now used to derive detailed quantitative information about the interaction
and its energy dependence. Further, for many of the simpler metals and alloys,
it is possible to derive the interaction from first principles and calculate the
transition temperature and other superconducting properties.
The theoretical methods used make use of the methods of quantum field
theory as adopted to the many-body problem, including Green’s functions,
Feynman diagrams, Dyson equations and renormalization concepts. Following
Matsubara, temperature plays the role of an imaginary time. Even if you are
not familiar with diagrammatic methods, I hope that you will be able to
follow the physical arguments involved.
In 1950, diagrammatic methods were just being introduced into quantum
field theory to account for the interaction of electrons with the field of photons.
It was several years before they were developed with full power for application
to the quantum statistical mechanics of many interacting particles. Following
Matsubara, those prominent in the development of the theoretical methods
include Kubo, Martin and Schwinger, and particularly the Soviet physicists,
M igdal, Galitski, Abrikosov, Dzyaloshinski, and Gor'kov. The methods were
first introduced to superconductivity theory by Gor'kov [9] and a little later
in a somewhat different form by Kadanoff and Martin. [10] Problems of
301
superconductivity have provided many applications for the powerful Green’s
function methods of m any-body theory and these applications have helped to
further develop the theory.
Diagrammatic methods were first applied to discuss electron-phonon
interactions in normal metals by M igdal [11] and his method was extended
to superconductors by Eliashberg. [12] A similar approach was given by
Nambu. [13] The theories are accurate to terms of order (m /M )1/2, where m
is the mass of the electron and M the mass of the ion, and so give quite accurate
quantitative accounts of the properties of both normal metals and super
conductors.
W e will first give a brief discussion of the electron-phonon interactions as
applied to superconductivity theory from 1950 to 1957, when the pairing theory
was introduced, then discuss the M igdal theory as applied to normal metals,
and finally discuss Eliashberg’s extension to superconductors and subsequent
developments. W e will close by saying a few words about applications of the
pairing theory to systems other than those involving electron-phonon inter
actions in metals.
2
D evelopm ents from 1950— 1957
The isotope effect was discovered in the spring of 1950 by Reynolds, Serin,
et al, [4] at Rutgers University and by E. M axwell [5] at the U . S. National
Bureau of Standards. Both groups measured the transition temperatures of
separated mercury isotopes and found a positive result that could be interpreted
as T qM 112 ~ constant, where M is the isotopic mass. If the mass of the ions
is important, their motion and thus the lattice vibrations must be involved.
Independently, Frohlich, [ 6 ] who was then spending the spring term at
Purdue University, attempted to develop a theory of superconductivity based
on the self-energy of the electrons in the field of phonons. He heard about
the isotope effect in mid-M ay, shortly before he submitted his paper for
publication and was delighted to find very strong experimental confirmation
of his ideas. He used a Hamiltonian, now called the Frohlich Hamiltonian,
in which interactions between electrons and phonons are included but Cou
lomb interactions are omitted except as they can be included in the energies
of the individual electrons and phonons. Frohlich used a perturbation theory
approach and found an instability of the Fermi surface if the electron-phonon
interaction were sufficiently strong.
W hen I heard about the isotope effect in early M ay in a telephone call from
Serin, I attempted to revive my earlier theory of energy gaps at the Fermi
surface, with the gaps now arising from dynamic interactions with the phonons
rather than from small static lattice displacements. [14] I used a variational
method rather than a perturbation approach but the theory was also based on
the electron self-energy in the field of phonons. W hile we were very hopeful
at the time, it soon was found that both theories had grave difficulties, not
easy to overcome. [15] It became evident that nearly all of the self-energy is
included in the normal state and is little changed in the transition. A theory
302
involving a true many-body interaction between the electrons seemed to be
required to account for superconductivity. Schafroth [16] showed that starting
with the Frohlich Hamiltonian, one cannot derive the Meissner effect in any
order of perturbation theory. M igdal’s theory, [11] supposedly correct to
terms of order (m/M ) 1/2, gave no gap or instability at the Fermi surface and
no indication of superconductivity.
O f course Coulomb interactions really are present. The effective direct
Coulomb interaction between electrons is shielded by the other electrons and
the electrons also shield the ions involved in the vibrational motion. Pines and
I derived an effective electron-electron interaction starting from a Hamiltonian
in which phonon and Coulomb terms are included from the start. [1 7] As is the
case for the Frohlich Hamiltonian, the matrix element for scattering of a pair
of electrons near the Fermi surface from exchange of virtual phonons is
negative (attractive) if the energy difference between the electron states in
volved is less than the phonon energy. As discussed by Schrieffer, the attractive
nature of the interaction was a key factor in the development of the micro
scopic theory. In addition to the phonon induced interaction, there is the
repulsive screened Coulomb interaction, and the criterion for superconductivity
is that the attractive phonon interaction dominate the Coulomb interaction
for states near the Fermi surface. [18]
During the early 1950’s there was increasing evidence for an energy gap at
the Fermi surface. [19] Also very important was Pippard’s proposed non-local
modification [20] of the London electrodynamics which introduced a new length
the coherence distance, £0, into the theory. In 1955 I wrote a review article [17]
on the theory of superconductivity for the Handbuch der Physik, which was
published in 1956. The central theme of the article was the energy gap, and
it was shown that Pippard’s version of the electrodynamics would likely follow
from an energy gap model. Also included was a review of electron-phonon
interactions. It was pointed out that the evidence suggested that all phonons
are involved in the transition, not just the long wave length phonons, and
that their frequencies are changed very little in the normal-superconducting
transition. Thus one should be able to use the effective interaction between
electrons as a basis for a true many-body theory of the superconducting state.
Schrieffer and Cooper described in their talks how we were eventually able
to accomplish this goal.
3
G r e en ’s F unction M ethod for N ormal M etals
By use of Green’s function methods, Migdal [ 1 1 ] derived a solution of Frohlich’s
Hamiltonian, H = //ei+ #p h + 7Z ei-ph, for normal metals valid for abritrarily
strong coupling and which involves errors only of order (m\M )1/2. The Green’s
functions aredefined by thermal average of time ordered operators for the
electrons and phonons, respectively
G = —i< 7 y (l)y > + (2 )> (la)
D = —i< T ' 0 ( l ) 0 +(2 ) > (lb)
303
Here y)(r,t) is the wave field operator for electron quasi-particles and
0 (r,f) for the phonons, the symbols 1 and 2 represent the space-time points
(rn*i) and (r2,**) and the brackets represent thermal averages over an ensemble.
Fourier transforms of the Green’s functions for H0 = He\-\-Hv^ for non
interacting electrons and phonons are
°o(P) = ---------------------------------------------------------(
(On— Eo{k) + ldk
2 a)
= (-------— r v n
[V» — (Oo{q) + 1 0 Vn~\-(Oo{q) — lOj (2b)
where P = (k,con) and Q, = (q,vn) are four vectors, e0(k) is the bare electron
quasiparticle energy referred to the Fermi surface, co0(q) the bare phonon
frequency and a)n and vn the Matsubara frequencies
con = (2nJr \)n ik ^ T ; vn — 2nn\k^T (3)
for Fermi and Bose particles, respectively.
As a result of the electron-phonon interaction, i / ei-ph, both electron and
phonon energies are renormalized. The renormalized propagators, G and D ,
can be given by a sum over Feynman diagrams, each of which represents a
term in the perturbation expansion. W e shall use light lines to represent the
bare propagators, G0 and D0i heavy lines for the renormalized propagators,
G and D , straight lines for the electrons and curly lines for the phonons.
The electron-phonon interaction is described by the vertex
G(P+Q)
G(P)
which represents scattering of an electron or hole by emission or absorption
of a phonon or creation of an electron and hole by absorption of a phonon
by an electron in the Fermi sea. M igdal showed that renormalization of the
vertex represents only a small correction, of order (m /M )1/2, a result in accord
with the Born-Oppenheimer adiabatic-approximation. If terms of this order
are neglected, the electron and phonon self-energy corrections are given by
the lowest order diagrams provided that fully renormalized propagators are
used in these diagrams.
The electron self-energy E{P) in the Dyson equation:
„ — + — @------
G{P) = G0(P) +GJP)E(P)G{P) (4)
is given by the diagram
2. (5)
The phonon self-energy, 7r(QJ, defined by

«a a a ^0 a aa »+ s tA A /W (6 )
304
is given by
GIP+Q)
S>
G(P)
Since to order(m/M)1!* one can use an unrenormalized vertex function
a = ao, the Dyson equations form a closed system such that both Z(P) and
n (d ) can be determined. The phonon self-energy, n (QJ, gives only a small
renormalization of the phonon frequencies. As to the electrons, M igdal noted
that we are interested in states k very close to AF, s o that to a close approxima
tion Z(k,w) depends only on the frequency. For an isotropic system,
Z(k,c o) ~ Z(kv,(o) = E(oj) (7)
The renormalized electron quasi-particle energy, co*, is then given by a root
of
e(k) = co* = £oW +^(w jt) (8 )
In the thermal Green’s function formalism, one may make an analytic
continuation from the imaginary frequencies, cow, to the real co axis to determine
I{w ).
Although E((o) is small compared with the Fermi energy, E y , it changes
rapidly with energy and so can affect the density of states at the Fermi surface
and thus the low temperature electronic specific heat. The mass renormal
ization factor m*/m, at the Fermi surface may be expressed in terms of a par
ameter X :
m*lm = Z(kv) = 1 + X = (d£0/d*)F/(d£/d*)F (9)
In modern notation, the experession for X is
A= (10)
where F(co) is the density of phonon states in energy and a 2(co) is the square
of the electron-phonon coupling constant averaged over polarization directions
of the phonons. Note that X is always positive so that the Fermi surface is
stable if the lattice is stable. Values of X for various metals range from about
0.5 to 1.5. The parameter X corresponds roughly to the N(0) Fphonon of the
BCS theory.
4 N a m b u -E l ia s h b e r g T h e o r y f o r S u p e r c o n d u c t o r s
M igdal’s theory has important consequences that have been verified experi
mentally for normal metals, but gave no clue as to the origin of supercon
ductivity. Following the introduction of the BCS theory, Gor'kov showed
that pairing could be introduced through the anomalous Green’s function
F(P) = i < Ty)1y)i > , (1 1 )
Nambu showed that both types of Green’s functions can be conveniently
included with use of a spinor notation
305
M (r>0\
V= (12)
Vvl (*•»<)/
where y)^ and \pi are wave field operators for up and down spin electrons
and a matrix Green’s function with components
= - i < 7 > av » > (13)
Thus Gn and G22 are the single particle Green’s functions for up and down
spin particles and G12 ==G 21= F(P) is the anomalous Green’sfunction of
Gor'kov.
There are two self-energies, Z x and Z 2, defined by the matrix
(L\ EA
2 = 1 (14)
V* zJ
Eliashberg noted that one can describe superconductors to the same accuracy
as normal metals if one calculates the self-energies with the same diagrams that
M igdal used, but with N am bu matrix propagators in place of the usual
normal state Green’s functions. The matrix equation for G is
G = Go+GoEG (15)
The matrix equation for Z yields a pair of coupled integral equations for Z x
and Z 2. Again Z x and Z 2 depend mainly on the frequency and are essentially
independent of the momentum variables. Following Nam bu, [13] one may
define a renormalization factor £ s(g>) and a pair potential, A (co), for isotropic
systems through the equations:
oj^ s (oj) = co+ZAco) (16)
A(co) = Z 2(a>)IZ(o>)- (17)
Both Zs and A can be complex and include quasi-particle life-time effects.
Eliashberg derived coupled non-linear integral equations for Zs(^) and
A (cd) which involve the electron-phonon interaction in the function a2(co)F(co).
The Eliashberg equations have been used with great success to calculate the
properties of strongly coupled superconductors for which the frequency
dependence of Z and A is important. They reduce to the BCS theory and
to the nearly equivalent theory of Bogoliubov [21 ] based on the principle of
“compensation of dangerous diagrams” when the coupling is weak. By weak
coupling is meant that the significant phonon frequencies are very large
compared with A;b7*c, so that A{a>) can be regarded as a constant independent
of frequency in the important range of energies extending to at most a few
In weak coupling one may also neglect the difference in quasi-particle
energy renormalization and assume that Zs — Zn-
The first solutions of the Eliashberg equations were obtained by Morel and
Anderson [22] for an Einstein frequency spectrum. Coulomb interactions were
included, following Bogoliubov, by introducing a parameter /li* which re
normalizes the screened Coulomb interaction to the same energy range as the
phonon interaction, In weak coupling, N(0)V = %—ju*. They estimated A
from electronic specific heat data and /u* from the electron density and thus
the transition temperatures, 7"c, for a number of metals. Order-of-magnitude
306
agreement with experiment was found. Later work, based in large part on
tunneling data, has yielded precise information on the electron-phonon
interaction for both weak and strongly-coupled superconductors.
4
A nalysis of T unneling D ata
From the voltage dependence of the tunneling current between a normal
metal and a superconductor one can derive A(a>) and thus get direct infor
mation about the Green’s function for electrons in the superconductor. It
is possible to go further and derive empirically from tunneling data the
electron-phonon coupling, cl2((d)F(cjo), as a function of energy. That electron
tunneling should provide a powerful method for investigating the energy gap
in superconductors was suggested by I. Giaever, [23] and he first observed
the effect in the spring of 1960.
The principle of the method is illustrated in Fig. 1. At very low temperatures,
the derivative of the tunneling current with respect to voltage is proportional
to the density of states in energy in the superconductor. Thus the ratio of the
density of states in the metal in the superconducting phase, JVs, to that of the
same metal in the normal phase, JVn, at an energy eV above the Fermi surface
is given by
N s(eV) (d //d 7 )ns
An ~ (d //d F )nn
ftiv)
' 'n s
~ N SSM ~ V^ ^LA =Z
Tunneling from a normal metal into a superconductor
Fig. 1.
Schematic diagram illustrating tunneling from a normal metal into a superconductor near
T = 0°K . Shown in the lower part of the diagram is the uniform density of states in energy
of electrons in the normal metal, with the occupied states shifted by an energy eV from an
applied voltage V across the junction. The upper part of the diagram shows the density of
states in energy in the superconductor, with an energy gap 2 A . The effect of an increment of
voltage <5V giving an energy change dco is to allow tunneling from states in the range dco. Since
the tunneling probability is proportional to density of states Ng (to), the increment in current
<51 is proportional to Ns (co)<5V.
307
ENERGY (IN UNITS OF € )
Fig. 2.
Conductance of a Pb-Mg junction as a function of applied voltage (from reference 24).
The normal density is essentially independent of energy in the range

involved (a few m eV). In weak coupling superconductors, for a voltage V
and energy co = eV,
Ms(C0)~ , (19)
Nn Jay2—A2
As T -> 0 K, no current flows between the normal metal and the super
conductor until the applied voltage reaches A/e, when there is a sharp rise
in dljdV followed by a drop. This is illustrated in Fig. 2 for the case of Pb.
The first experiments of Giaever were on aluminum, which is a weak
coupling superconductor. Good agreement was found between theory and
experiment. In later measurements on tunneling into Pb, a strongly coupled
superconductor, Giaever, Hart and M egerle [24] observed anomalies in the
density of states that appeared to be associated with phonons, as shown in
Fig. 2. These results were confirmed by more complete and accurate tunneling
data on Pb by J. M . Rowell et al. [25]
In the meantime, in the summer of 1961, Schrieffer had derived numerical
solutions of the Eliashberg equations working with a group engaged in de
veloping methods for computer control using graphical display methods. [26]
H e and co-workers calculated the complex A{o) for a Debye frequency
308
spectrum. Later, at the University of Pennsylvania, he together with J. W.
Wilkins and D. J. Scalapino [27] continued work on the problem with a view
to explaining the observed anomalies on Pb. They showed that for the general
case of a complex A (co)
(d //d F )n3 = JV8(fi>) = Re f to (20)
(d //d F )n n Nn \ j a ) 2—A 2((o)
where Re represents the real part. From measurements of the ratio over the
complete range of voltages, one can use Kramers-Kronig relations to obtain
both the real and imaginary parts of A(co) = A r{w) + iz l 2(a>). From analysis
of the data, one can obtain the Green’s functions which in turn can be used
to calculate the various thermal and transport properties of superconductors.
This has been done with great success, even for such strongly-coupled super
conductors as lead and mercury.
For lead, Schrieffer et al, used a phonon spectrum consisting of two Lo-
rentzian peaks, one for transverse waves and one for longitudinal and obtained
a good fit to the experimental data for T < < Tc. The calculations were
extended up to T c for Pb, Hg, and Al by Swihart, Wada and Scalapino, [28]
again finding good agreement with experiment.
In analysis of tunneling data, one would like to find a phonon interaction
spectrum, a 2(co)F(co), and a Coulomb interaction parameter, //*, which when
inserted into the Eliashberg equations will yield a solution consistent with the
tunneling data. W. L. M cM illan devised a computer program such that one
could work backwards and derive a2(co)F(a>) and //* directly from the tunneling
data. His program has been widely used since then and has been applied to
a number of superconducting metals and alloys, including, Al, Pb, Sn, the
transition elements Ta and Nb, a rare earth, La, and the compound N b 3Sn.
In all cases it has been found that the phonon mechanism is dominant with
reasonable values of jn*. Peaks in the phonon spectrum agree with peaks
in the phonon density of states as found from neutron scattering data, as
shown in Fig. 3 for the case of Pb. In Fig. 4 is shown the real and imaginary
parts of A(co) for Pb as derived from tunneling data.
One can go further and calculate the various thermodynamic and other
properties.Good agreement with experiment is found forstrongly coupled
superconductors even when there are significantdeviations from the weak
coupling limits. For example, the weak-coupling BCS expression for the
condensation energy at T = 0 K is
EBCS=l-N(0)ZnAo2 (21)
where N(0)£n is the phonon enhanced density of states and A0 is the gap
parameter at T = 0 K. The theoretical expression with (co) and A (co)
derived from tunneling data, again for the case of Pb, gives [29, 30, 31]
Etheor = 0.78 E'bcS (22)
in excellent agreement with the experimental value
Fexp = (0.76-|-0.02) F'bcs- (23)
309
V - A ( M IL L IV O L T S )
Fig. 3.
Density of states versus energy for Pb. Solid line, calculated by Schrieffer et a l; long dashed
ine, observed from tunneling; short dashed line, BCS weak coupling theory.
E N E R G Y (m e V )
R e al and im a g in a ry p a r ts of A v e rsu s cj- A q for Pb.
Fig. 4.
Real and imaginary parts of A (to) = î(w) -fLd2(a>) versus energy for Pb. (After McMillan
& Rowell).
310
i— i------ 1— i-----1— r— r
E N E R G Y ( M I L L IE L E C T R O N V O L T S )
O
C o m p a riso n of a F ( a ) and F(cu) for
P b (a ft e r M c M illa n and R o w e ll)
Fig. 5.
Comparison of azF for Pb derived from tunneling data with phonon density of states from
neutron scattering data of Stedman et al. [8]
In Figs. 5, 6 , 7, and 8 are shown other examples of a 2(co)F(co) derived from
tunneling data for Pb, In, [31] La, [32] and N b 3Sn. [33] In all cases the
results are completely consistent with the phonon mechanism. Coulomb
interactions play only a minor role, with /i* varying only slowly from one metal
to another, and generally in the range 0 . 1 — 0 2 .
311
a F ( oj) for indium.
Fig. 6.
a*F for In (after M cMillan and Rowell).
As a further check, it is possible to derive the phonon density of states,

F(a>) from neutron scattering data and use pseudo-potential theory to calculate
the electron-phonon interaction parameter aq(co). From these values, one can
use the Eliashberg equations to calculate Zî0*) an(i A{ gj>) and the various
superconducting properties, including the transition temperature, Tc. Extensive
calculations of this sort have been made by J. P. Carbotte and co-workers
[34] for several of the simpler metals and alloys. For example, for the gap
edge, A0, in A1 at T = 0 K they find 0.19 meV as compared with an experi
mental value of 0.17. The corresponding values for Pb are 1.49 m eV from
theory as compared with 1.35 m eV from experiment. These are essentially
first principles calculations and give convincing evidence that the theory as
formulated is essentially correct. Calculations made for a number of other
metals and alloys give similar good agreement.
C onclusions
In this talk we have traced how our understanding of the role of electron-
phonon interactions in superconductivity has developed from a concept to a
precise quantitative theory. The self-energy and pair potential, and thus
the Green’s functions, can be derived either empirically from tunneling data
or directly from microscopic theory with use of the Eliashberg equations.
Physicists, both experimental and theoretical, from different parts of the
world have contributed importantly to these developments.
All evidence indicates that the electron-phonon interaction is the dominant
mechanism in the cases studied so far, which include many simple metals,
312
T
Fig. 7.
a*F for La (after Lou and Tomasch).
transition metals, a rare earth, and various alloys and compounds. Except
possibly for the metallic form of hydrogen, [35] which is presumed to exist
at very high pressures, it is unlikely that the phonon mechanism will yield
substantially higher transition temperatures than the present maximum of
about 21 K for a compound of Nb, Al and Ge.
Otjier mechanisms have been suggested for obtaining higher transition
temperatures. One of these is to get an effective attractive interaction between
electrons from exchange of virtual excitons, or electron-hole pairs. This re
quires a semiconductor in close proximity to the metal in a layer or sandwich
structure. At present, one can not say whether or not such structures are
feasible and in no case has the exciton mechanism been shown to exist. As
Ginzburg has emphasized, this problem (as well as other proposed mechanisms)
deserves study until a definite answer can be found. [36]
The pairing theory has had wide application to Fermi systems other than
electrons in metals. For example, the theory has been used to account for
313
0.6
0.5
0.4 o
II
CM CM
0.2
0.1
0 10 20 30
E (mV)
2
a F for N b 3S n ( a f t e r L.Y.L. Sh en)
Fig. 8.
arF for Nb3Sn (after Y. L. Y. Shen).
many aspects of nuclear structure. It is thought the nuclear matter in neutron

stars is superfluid. Very recently, evidence has been found for a possible pairing
transition in liquid H e3 at very low temperatures [37]. Some of the concepts,
such as that of a degenerate vacuum, have been used in the theory of ele
mentary particles. Thus pairing seems to be a general phenomenon in Fermi
systems.
The field of superconductivity is still a very active one in both basic science
and applications. I hope that these lectures have given you some feeling for
the accomplishments and the methods used.
R eferences
1. Shoenberg, D. Superconductivity, Cambridge Univ. Press, Cambridge (1938). Second
edition, 1951.
2. London, F. Proc. Roy. Soc. (London) 152A, 24 (1935).
3. Bardeen, J. Phys. Rev. 59, 928A (1941).
4. Reynolds, C.A., Serin, B. Wright W. H. and Nesbitt, L. B. Phys. Rev. 78, 487 (1950).
5. Maxwell, E., Phys. Rev. 78, 477 (1950).
6. Frohlich, H., Phys. Rev. 79, 845 (1950); Proc. Roy. Soc. (London) Ser. A 213, 291 (1952).
7. London, F., Superfluids, New York, John Wiley and Sons, 1950.
8. For recent review articles with references, see the chapters by D. J. Scalapino and by
W. L. M cMillan and J. M. Rowell in Superconductivity, R. D. Parks, ed., New York,
Marcel Bekker, Inc., 1969, Vol. 1. An excellent reference for the theory and earlier
experimental work is J. R. Schrieffer, Superconductivity, New York, W. A. Benjamin,
Inc., 1964. The present lecture is based in part on a chapter by the author in Cooperative
Phenomena, H. Haken and M. Wagner, eds. to be published by Springer.
314
9. Gor'kov, L. P., Zh. Eksper i. teor. Fiz. 34, 735 (1958). (English transl. Soviet Phys. —
JETP 7, 505 (1958)).
10. Kadanoff L. P. and Martin, P. G. Phys. Rev. 124, 670 (1961).
11. Migdal, A. B., Zh. Eksper i. teor. Fiz. 34, 1438 (1958). (English transl. Soviet Phys. —
JETP 7, 996 (1958)).
12. Eliashberg, G. M., Zh. Eksper i. teor. Fiz. 38, 966 (1960). Soviet Phys. — JETP 11,
696 (1960).
13. Nambu, Y., Phys. Rev. 117, 648 (1960).
14. Bardeen, J., Phys. Rev. 79, 167 (1950); 80, 567(1950); 81 829 (1951).
15. Bardeen, J., Rev. Mod. Phys. 23, 261 (1951).
16. Schafroth, M. R., Helv. Phys. Acta 24, 645 (1951); Nuovo Cimento,9, 291 (1952).
17. For a review see Bardeen, J., Encyclopedia of Physics, S. Flugge, ed., Berlin, Springer-
Verlag, (1956) Vol. X V , p. 274.
18. Bardeen, J., L. N. Cooper and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957).
19. For references, see the review article of M. A. Biondi, A. T. Forrester, M. B. Garfunkel
and C. B. Satterthwaite, Rev. Mod. Phys. 30, 1109 (1958).
20. Pippard, A. B., Proc. Roy. Soc. (London) A216, 547 (1954).
21. See N. N. Bogoliubov, V. V. Tolmachev and D. V. Shirkov, A New Method in the
Theory of Superconductivity, New York, Consultants Bureau, Inc., 1959.
22. Morel P. and Anderson, P. W., Phys. Rev. 125, 1263 (1962).
23. Giaever, I., Phys. Rev. Letters, 5, 147; 5, 464 (1960).
24. Giaever, I., Hart H. R., and Megerle K., Phys. Rev. 126, 941 (1962).
25. Rowell, J.M ., Anderson P. W. and Thomas D. E., Phys. Rev. Letters, 10, 334 (1963).
26. Culler, G .J., Fried, B. D., Huff, R. W. and Schrieffer, J. R ., Phys. Rev. Letters 8, 339
(1962).
27. Schrieffer, J. R., Scalapino, D .J. and Wilkins, J. W., Phys. Rev. Letters 10, 336 (1963);
D. J. Scalapino, J. R. Schrieffer, a n d j. W. Wilkins, Phys. Rev. 148, 263 (1966).
28. Scalapino, D. J., Wada, Y. and Swihart, J. C., Phys. Rev. Letters, 1 4,102 (1965); 1 4,106
(1965).
29. Eliashberg, G. M., Zh. Eksper i. teor. Fiz. 43, 1005 (1962). English transl. Soviet Phys. —
JETP 16, 780 (1963).
30. Bardeen, J. and Stephen, M., Phys. Rev. 136, A1485 (1964).
31. McMillan, W. L. and Rowell, J. M. in Reference 8.
32. Lou, L. F. and Tomasch, W. J., Phys. Rev. Lett. 29,858 (1972).
33. Shen, L. Y. L., Phys. Rev. Lett. 29, 1082 (1972).
34. Carbotte, J. P., Superconductivity, P. R. Wallace, ed., New York, Gordon and Breach,
1969, Vol. 1, p. 491; J. P. Carbotte and R. C. Dynes, Phys. Rev. 172, 476 (1968);
C. R. Leavens and J. P. Carbotte, Can. Journ. Phys. 49, 724 (1971).
35. Ashcroft N. W., Phys. Rev. Letters, 21, 1748 (1968).
36. See V. L. Ginzburg, “The Problem of High Temperature Superconductivity,” Annual
Review of Materials Science, Vol. 2, p. 663 (1972).
g7. Osheroff D. D., Gully W. J., Richardson R. C. and Lee, D. M. Phys. Rev. Lett. 29, 1621
(1972).
315
NOTES AND REFERENCES
1. F. London, Superfluids, Vol. I, Wiley, New York, 1950.

2. C. J. Gorter and H. G. B. Casimir, Phys. Z., 35, 963 (1934);
Z. Tech. Phys., 15, 539 (1934).
3. F. London, Superfluids, Vol. II, Wiley, New York, 1950.
4. D. Schoenberg, Superconductivity, Cambridge, New York,
1952.
5. F. London, Phys. Rev., 74, 562 (1948).
6. W. Meissner and R. Ochsenfeld, Naturwiss., 21, 787 (1933).
7. H. Kamerlingh Onnes, Comm. Phys. Lab. Univ. Leiden,
Nos. 119, 120, 122 (1911).
8. J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev.,
106, 162 (1957); 108, 1175 (1957).
9. J. Bardeen and J. R. Schrieffer, Progr. Low Temp. Phys.,
Vol. Ill, North-Holland, Amsterdam, 1961.
10. H. Frohlich, Phys. Rev., 79, 845 (1950).
11. C. A. Reynolds, B. Serin, W. H. Wright, and L. B. Nesbitt,
Phys. Rev., 78, 487 (1950).
12. E. Maxwell, Phys. Rev., 78, 477 (1950).
13. J. Bardeen, Rev. Mod. Phys., 23, 261 (1951).
14. M. R. Schafroth, Helv. Phys. Acta, 24, 645 (1951).
15. A. B. Migdal, Soviet Phys. J E T P , 1, 996 (1958).
16. (a) See Ref. 4.
(b) B. Serin, Handbuch der Physik, 15, 210, Springer, Berlin,
1950.
317
(c) J. Bardeen, Handbuch der Physik, 15, 274, Springer,

Berlin, 1950.
(d) M. A. Biondi, A. T. Forrester, M. P. Garfunkel, and C. B.
Satterthwaite, Rev. Mod. Phys., 30, 1109 (1958).
(e) E. A. Lynton, Superconductivity, Methuen, London, 1963.
(f) M. Tinkham, Low Temperature Physics, p. 149, Gordon
and Breach, New York, 1962.
(g) D. H. Douglass and L. M. Falicov, Progr. Low Temp.
Phys., Vol. IV, C. J. Gorter (ed.), North-Holland, Amster
dam, 1963.
17. A. A. Abrikosov, J. Exptl. Theoret. Phys. (U SSR ), 32,
1442 (1957), translated as Soviet Phys. J E T P , 5 , 1174
(1957).
18. (a) M. Tinkham, Phys. Rev., 129, 2413 (1963).
(b) P. W. Anderson, Proc. Ravello Spring School, 1963.
19. N. Byers and C. N. Yang, Phys. Rev. Letters, 7, 46 (1961).
20. (a) B. D. Deaver, Jr. and W. M. Fairbank, Phys. Rev.
Letters, 7, 43 (1961).
(b) R. Doll and M. Nabauer, Phys. Rev. Letters, 7, 51 (1961).
21. W. H. Keesom and J. H. van den Ende, Commun. Phys.
Lab. Univ. Leiden, No. 2196 (1932); W. H. Keesom and
J. A. Kok, Physica, 1, 175 (1934).
22. D. Mapother, private communication.
23. T. H. Geballe, B. T. Matthias, G. W. Hull, Jr., and E. Corenz-
wit, Phys. Rev. Letters, 6, 275 (1961); T. H. Geballe and B.T.
Matthias, I B M J. Res. Develop., 6, 256 (1962).
24. J. W. Garland, Phys. Rev. Letters, 11, 111, 114 (1963).
25. R. E. Glover, III, and M. Tinkham, Phys. Rev., 108, 243
(1957); M. A. Biondi and M. Garfunkel, Phys. Rev., 116,
853 (1959).
26. I. Giaever, Phys. Rev. Letters, 5, 147, 464 (1960).
27. (a) R. W. Morse, Progr. Cryog., Vol. I, p. 220, K. Mendelssohn
(ed.), Heywood, London, 1959.
(b) B. T. Geilikman and V. Z. Kresin, J. Exptl. Theoret.
Phys. (U SSR ), 41, 1142 (1961), translated in Soviet Phys.
J E T P , 14, 816 (1961).
(c) V. L. Pokrovskii, J. Exptl. Theoret. Phys. (U SSR ), 40,
143 (1961), translated in Soviet Phys. J E T P , 13, 100 (1961).
28. L. C. Hebei and C. P. Slichter, Phys. Rev., 113, 1504 (1959);
L. C. Hebei, Phys. Rev., 116, 79 (1959).
N o te s and References 319
29. A. G. Redfield, Phys. Rev. Letters, 3, 85 (1959); A. G. Red-

field and A. G. Anderson, Phys. Rev., 116, 583 (1959).
30. J. Bardeen, G. Rickayzen, and L. Tewordt, Phys. Rev., 113,
982 (1959).
31. F. London and H. London, Proc. Roy. Soc. {London),
A149, 71 (1935); Physica, 2, 341 (1935).
32. L. Onsager, Phys. Rev. Letters, 7, 50 (1961).
33. A. B. Pippard, Proc. Roy. Soc. {London), A216, 547 (1953).
34. R. G. Chambers, Proc. Roy. Soc. {London), A65, 458 (1952).
35. R. A. Ferrell and R. E. Glover, III, Phys. Rev., 109, 1398
(1958); M. Tinkham and R. A. Ferrell, Phys. Rev. Letters,
2, 331 (1959).
36. V. L. Ginsburg and L. D. Landau, J . Exptl. Theoret. Phys.
(U SSR ), 2 0 , 1064 (1950).
37. L. P. Gor’kov, J. Exptl. Theoret. Phys. (U SSR ), 36, 1918
(1959), translated in Soviet Phys. J E T P , 9, 1364 (1959).
38. (a) N. R. Werthamer, Phys. Rev., 132, 663 (1963).
(b) L. Tewordt, Phys. Rev., 132, 595 (1963).
39. C. N. Yang, Rev. Mod. Phys., 34, 694 (1962).
40. P. W. Anderson, Phys. Chem. Solids, 11, 26 (1959).
41. L. N. Cooper, Phys. Rev., 104, 1189 (1956).
42. M. R. Schafroth, J. M. Blatt, and S. T. Butler, Helv. Phys.
Acta, 30, 93 (1957). Subsequent to the work of Bardeen,
Cooper, and Schrieffer, Matsubara and Blatt attem pted to
improve the mathematical treatm ent of SBB. Owing to the
complexity of their formalism, they did not carry through
the calculation far enough to compare their results with the
BCS theory. In particular, a well-defined, simple equation
corresponding to the BCS energy-gap equation was not
derived, nor was the quasi-particle spectrum obtained.
See T. Marsubara and J. M. Blatt, Progr. Theoret. Phys.
(Kyoto), 23, 451 (1960). Later, this approach was extended
to obtain the results of the pairing theory; see J. M. Blatt,
Progr. Theoret. Phys. (Kyoto), 27, 1137 (1962). See also
M. Baranger, Phys. Rev., 130, 1244 (1963).
43. M. R. Schafroth, Phys. Rev., I l l , 72 (1958).
44. D. Pines, The M any-Body Problem, Benjamin, New York,
1962.
45. L. D. Landau, J. Exptl. Theoret. Phys. (U SSR ), 30, 1058
(1956).
46. J. M. Luttinger and P. Nozieres, Phys. Rev., 127, 1423, 1431

(1962).
47. Anderson has exploited the formal similarity between the
commutation relations (2-21) and those for a collection of
fictitious Pauli spin operator S k. The connection between
the two sets is
2 S zk = 1 - (nkT + n . k i)
S xk + i S yk = bk +
&xk “ v^yk ~
He was able to obtain the results of the BCS theory by a semi-
classical treatment of this fictitious spin system. See P. W.
Anderson, Phys. Rev., 110, 827 (1958); 112, 1900 (1958).
48. S. Tomonaga, Progr. Theoret. Phys. (Kyoto), 2, 6 (1947).
49. (a) T. D. Lee, F. Low, and D. Pines, Phys. Rev., 90, 297 (1953).
(b) T. D. Lee and D. Pines, Phys. Rev., 92, 883 (1953).
50. L. P. Kadanoff and P. C. Martin, Phys. Rev., 124, 670 (1961).
51. (a) J. M. Blatt, Progr. Theoret. Phys. (Kyoto), 27, 1137
(1962); Proc. Superconductivity Conference, Cambridge
1959, unpublished; Theory of Superconductivity, Academic,
New York, 1964.
(b) M. Baranger, Phys. Rev., 130, 1244 (1963).
(c) F. Bloch and H. E. Rorschach, Phys. Rev., 128, 1697
(1962).
52. N. N. Bogoliubov, Nuovo Cimento, 7, 6, 794 (1958); also see
N. N. Bogoliubov, V. V. Tolmachev, and D. V. Shirkov,
A New Method in the Theory of Superconductivity, Con
sultants Bureau, New York, 1959.
53. J. Valatin, Nuovo Cimento, 7, 843 (1958).
54. (a) N. N. Bogoliubov, Physica, S u p p l ., 26, 1 (1960).
(b) J. Bardeen and G. Rickayzen, Phys. Rev., 118, 936 (1960).
(c) B. Mtihlschlegel, J . Math. Phys., 3, 522 (1962).
(d) D. C. Mattis and E. Lieb, J . Math. Phys., 2, 602 (1961).
(e) R. Haag, private communication.
55. J. Valatin, private communication, 1957.
56. The operator R is particularly useful in tunneling calcula
tions; see J. Bardeen, Phys. Rev. Letters, 9, 147 (1962) and
B. D. Josephson, Phys. Letters, 1, 251 (1962).
57. (a) H. Suhl, Bull. Am. Phys. Soc ., 6 , 119 (1961).
(b) Y. Wada, Rev. Mod. Phys., 36, 253 (1964).
58. P. W. Anderson, Phys. Rev., 112, 1900 (1958).

59. (a) J. C. Fisher, private communication, 1959.
(b) K. A. Brueckner, T. Soda, P. W. Anderson, and P. Morel,
Phys. Rev., 118, 1442 (1960).
(c) R. Balian and N. R. Werthamer, Phys. Rev., 131, 1553
(1963).
60. B. Bayman, Nucl. Phys., 15, 33 (1960).
61. B. R. Mattelson, The N -Body Problem, Wiley, New York,
1959.
62. The point of view taken here in calculating finite-temperature
properties is due to J. Bardeen.
63. L. I. Schiff, Quantum M echanics, Chap. 8, McGraw-Hill,
New York, 1949.
64. In order that the limits on ep and ep> do not affect this
argument, one must satisfy the condition |q0|fo » I* If
|qo|£o « the term ep€p, ep2 and does not average to
zero. In this case the average coherence factor is twice as
large as that for |q0|fo » ^ however, only one-half as many
states contribute to the sum for small q0. Therefore, the
resultant expression for a is the same in the two cases.
65. (a) See Ref. 9.
(b) D. H. Douglass and L. M. Falicov, Progr. Low Tem p.
Phys., Vol. IV, C. J. Gorter (ed.), North-Holland, Amster
dam (to be published).
(c) J. Bardeen, Rev. Mod. Phys., 34, 667 (1962).
(d) M. Tinkham, Phys. Rev., 129, 2413 (1963).
(e) B. T. Matthias, T. H. Geballe, V. B. Compton, Rev.
Mod. Phys., 35, 1 (1963).
(f) E. A. Lynton, Superconductivity, Methuen, London, 1963.
66. S. B. Chandrasekhar and J. A. Rayne, Phys. Rev., 124, 1011
(1961). Anderson estimated a relative shift of the sound
velocity of order (m/M) ~ 10~5, in agreement with the
results of Chandrasekhar and Rayne [see Phys. Rev., 112,
1900 (1958)], while Ferrell [Phys. Rev. Letters, 6, 541 (1961)]
suggested that a phonon-frequency shift 8ojqIwq ~ 0.1 might
be obtained for q£0 > 1. Detailed calculations by Marko
witz (private communication) and by Toxin and Liu (private
communication) show that the shift is definitely less than
1 per cent over the entire frequency range. See also R. E.
Prange, Phys. Rev., 129, 2495 (1963).
67. A. B. Pippard, Phil. Mag., 46, 1104 (1955); Low Temperature

Physics, Gordon and Breach, New York, 1962; J. R. Liebo-
witz, Bull. Am. Phys. Soc., Ser. I I , 9, 267 (1964).
68. C. P. Slichter, Principles of Magnetic Resonance, Harper,
New York, 1963.
69. (a) D. C. Mattis and J. Bardeen, Phys. Rev., I l l , 412 (1958).
(b) A. A. Abrikosov andL. P. Gor’kov, Soviet Phys. J E T P , 35,
1558 (1958), translated in 8 , 1090 (1959); 36, 319 (1959),
translated in 9, 220 (1959).
(c) G. Rickayzen, in C. Fronsdal (ed.), The M any-Body
Problem, Benjamin, New York, 1961.
70. P. B. Miller, Phys. Rev., 113, 1209 (1959).
71. D. M. Ginsberg and M. Tinkham, Phys. Rev., 118, 990
(1960). See also D. M. Ginsberg, P. L. Richards, and M.
Tinkham, Phys. Rev. Letters, 3, 337 (1959).
72. P. L. Richards and M. Tinkham, Phys. Rev., 119, 575 (1960).
73. D. M. Ginsberg and J. D. Leslie, Rev. Mod. Phys., 36, 198
(1964); J. Bardeen, Rev. Mod. Phys., 36, 198 (1964). See
also D. J. Scalapino, J. R. Schrieffer, and J. W. Wilkins
(to be published).
74. J. Bardeen, Phys. Rev. Letters, 6 , 57 (1961); 9, 147 (1962).
75. M. H. Cohen, L. M. Falicov, and J. C. Phillips, Phys. Rev.
Letters, 8 , 316 (1962).
76. R. E. Prange, Phys. Rev., 131, 1083 (1963).
77. W. A. Harrison, Phys. Rev., 123, 85 (1961).
78. (a) J. R. Schrieffer, D. J. Scalapino, and J. W. Wilkins,
Phys. Rev. Letters, 10, 336 (1963);
(b) Phys. Rev. (to be published).
79. (a) J. W. Wilkins, Ph.D. thesis, University of Illinois, 1963.
(b) J. R. Schrieffer, Rev. Mod. Phys., 36, 200 (1964).
(c) J. R. Schrieffer, in T. Bak (ed.), Phonons and Phonon
Interactions, Benjamin, New York, 1964.
80. E. Burstein, D. N. Langenberg, and B. N. Taylor, Phys.
Rev. Letters, 6 , 92 (1961); Advances in Quantum Electronics,
J. R. Singer (ed.), Columbia Univ. Press, New York, 1961.
81. B. N. Taylor and E. Burstein, Phys. Rev. Letters, 10, 14(1963).
82. J. R. Schrieffer and J. W. Wilkins, Phys. Rev. Letters, 10,
17 (1963).
83. B. D. Josephson, Phys. Rev. Letters, 1, 251 (1962).
84. P. W. Anderson, Proc. Ravello Spring School, 1963. See also
R. A. Ferrell and R. E. Prange, Phys. Rev. Letters, 10,

479 (1963) and V. Amkegaokar and A. Baratoff, Phys. Rev.
Letters, 10, 486 (1963).
85. P. W. Anderson and J. M. Rowell, Phys. Rev. Letters, 10,
230 (1963).
86. (a) J. M. Rowell, Phys. Rev. Letters, 11, 200 (1963); Rev.
Mod. Phys., 36, 199 (1964).
(b) B. D. Josephson, Thesis, Cambridge University, 1962;
Phys. Rev. Letters, 1, 251 (1962); Rev. Mod. Phys., 36, 216
(1964).
(e) S. Shapiro et al., Rev. Mod. Phys., 36, 223 (1964).
87. I. Giaever, H. R. Hart, and K. Megerle, Phys. Rev., 126, 941
(1962).
88. J. M. Rowell, P. W. Anderson, and D. E. Thomas, P hys .
Rev. Letters, 10, 334 (1963).
89. R. A. Ferrell, Phys. Rev. Letters, 3, 262 (1959).
90. P. W. Anderson, Phys. Rev. Letters, 3, 325 (1959).
91. (a) D. J. Thouless, The Quantum Mechanics of M any-Body
Systems, Academic, New York, 1961.
(b) D. Pines, The M any-Body Problem, Benjamin, New York,
1962.
(c) L. P. Kadanoff and G. Baym, Quantum Statistical
Mechanics, Benjamin, New York, 1962.
(d) P. Nozieres, Le problem du N corpes, Dunod, Paris,
1963, translated as The Theory of Interacting Fermi Systems,
Benjamin, New York, 1963.
(e) T. D. Schultz, Quantum Field Theory and the M any-Body
Problem, Gordon and Breach, New York, 1963.
(f) V. L. Bonch-Bruevitch and S. V. Tyablikov, Method
Funcii Grina Statisticeskei Mexanike, Moscow, 1960, trans
lated by D. Ter Haar, North-Holland, Amsterdam, 1962.
(g) A. A. Abrikosov, L. P. Gor’kov, and I. E. Dzyaloshinskii,
Methods of Quantum Field Theory in Statistical Mechanics ,
Prentice-Hall, Englewood Cliffs, N.J., 1963.
(h) 1 9 6 2 Cargese Lectures in Theoretical Physics, M. Levy
(ed.), Benjamin, New York, 1962.
(i) The M any-Body Problem, C. Fronsdal (ed.), Benjamin,
New York, 1962.
(j) The M any-Body Problem (Les Houches Notes), Dunod,
Paris, 1959.
324 T h e o ry of Su p e rc o n d u c tiv ity
92. R. E. Peierls, Quantum Theory of Solids, Clarendon Press,

Oxford, 1956.
93. J. Bardeen and D. Pines, Phys. Rev., 99, 1140 (1953).
94. J. Wilkins, Ph.D. thesis, University of Illinois, 1963.
95. F. Bassani, J. Robinson, B. Goodman, and J. R. Schrieffer,
Phys. Rev., 127, 1969 (1962).
96. J. M. Ziman, Electrons and Phonons, Clarendon Press, Oxford,
1960. See also The Fermi Surface, W. A. Harrison and
M. B. Webb (eds.), Wiley, New York, 1960.
97. L. J. Sham, Proc. Phys. Soc. (London), 78, 895 (1961).
98. W. A. Harrison, Phys. Rev., 129, 2503 (1963); 129, 2512
(1963); 131, 2433 (1963).
99. (a) J. Schwinger, Proc. Natl. Acad.Sci. (U.S.), 37, 452 (1951).
(b) P. C. Martin and J. Schwinger, Phys. Rev., 115, 1342
(1959).
100. V. M. Galitskii and A. B. Migdal, Soviet Phys. J E T P , 7 ,
96 (1958).
101. (a) L. P. Kadanoff and P. C. Martin, Phys. Rev., 124, 670
(1961).
(b) L. P. Kadanoff, Proc. Ravello Spring School, 1963.
This time-dependent form of the instability was known to
many workers in the field in 1957-1958, e.g., Goldstone,
Anderson, Bogoliubov, etc.
102. (a) R. P. Feynman, Phys. Rev., 76, 769 (1949).
(b) S. Tomonaga, Progr. Theoret. Phys. (Kyoto), 1, 27 (1946).
(c) F. J. Dyson, Phys. Rev., 75, 486 (1949).
103. (a) J. Hubbard, Proc. Roy. Soc. (London), A240, 539 (1957).
(b) N. N. Bogoliubov and D. V. Shirkov, Introduction to the
Theory of Quantized Fields, Wiley-Interscience, New York,
1959.
104. L. I. Schiff, Quantum Mechanics, McGraw-Hill, New York,
1949.
105. D. Bohm and D. Pines, Phys. Rev., 92, 609 (1953).
106. M. Gell-Mann and K. Brueckner, Phys. Rev., 106, 364 (1957).
107. P. Nozieres and D. Pines, Phys. Rev., I l l , 442 (1958).
108. J. Lindhard, Kgl. Danske Videnskab, Selskab. Mat. Fys.
Medd., 28, 8 (1954).
109. (a) J. S. Langer and S. H. Vosko, Phys. Chem. Solids, 1 2 ,
196 (1959).
(b) W. Kohn and S. H. Vosko, Phys. Rev., 119, 912 (1960).
(c) J. Friedel, Phil. Mag., 43, 153 (1952); Nuovo Cimento

Suppl, 2, 287 (1958).
110. T. J. Rowland, Phys. Rev., 125, 459 (1962).
111. R. P. Feynman and M. Cohen, Phys. Rev., 102, 1189 (1956).
112. D. Pines and J. R. Schrieffer, Nuovo Cimento, 10, 496 (1958).
113. G. Rickayzen, Phys. Rev., 115, 765 (1959).
114. Y. Nambu, Phys. Rev., 117, 648 (1960).
115. J. J. Quinn and R. A. Ferrell, Phys. Rev., 112, 812 (1958).
116. M. Gell-Mann, Phys. Rev., 106, 359 (1957).
117. P. Nozieres and D. Pines, Phys. Rev., I l l , 442 (1958).
118. S. D. Silverstein, Phys. Rev., 128, 631 (1962).
119. T. Staver, Ph.D. thesis, Princeton University, 1952 (un
published).
120. S. Engelsberg and J. R. Schrieffer, Phys. Rev., 131, 993
(1963).
121. L. P. Gor’kov, J. Exptl. Theoret. Phys. (U SSR ), 34, 735
122. G. M. Eliashberg, J. Exptl. Theoret. Phys. (U SSR ), 38, 966
(1960), translated in Soviet Phys. J E T P , 1 1 , 696 (1960).
123. See Ref. 78b.
124. P. Morel and P. W. Anderson, Phys. Rev., 125, 1263 (1962).
125. J. C. Swihart, I B M J. Res. Develop., 6 , 14 (1962).
126. G. J. Culler, B. D. Fried, R. W. Huff, and J. R. Schrieffer,
Phys. Rev. Letters, 8, 399 (1962).
127. (a) D. J. Scalapino and P. W. Anderson, Phys. Rev., 133,
A291 (1964).
(b) D. J. Scalapino, Rev. Mod. Phys., 36, 205 (1964).
128. A. A. Abrikosov, L. P. Gor’kov, and I. E. Dzyaloshinskii,
Soviet Phys. J E T P , 9, 636 (1959).
129. D. M. Ginsberg and M. Tinkham, Phys. Rev., 118, 990 (1960).
130. (a) P. B. Miller, Phys. Rev., 118, 928 (1960).
(b) D. M. Ginsberg and J. D. Leslie, Rev. Mod. Phys., 36,
198 (1964).
(c) K. Make and T. Tsuneto, Progr. Theoret. Phys. (Kyoto),
28, 163 (1962).
131. J. Bardeen, Nuovo Cimento, 5 , 1766 (1957).
132. G. Rickayzen, Phys. Rev., 115, 795 (1959).
133. G. Baym and L. P. Kadanoff, Phys. Rev., 124, 287 (1961).
134. D. Pines and J. R. Schrieffer, Nuovo Cimento, 10, 496 (1958).
135. R. P. Feynman and M. Cohen, Phys. Rev., 102, 1189 (1956).
136. (a) L. P. Kadanoff and V. Ambegaokar, Nuovo Cimento,

22, 914 (1961).
(b) V. Ambegaokar and L. P. Kadanoff, The M any-Body
Problem, C. Fronsdal (ed.), Benjamin, New York, 1962.
(c) J. C. Ward, Phys. Rev., 78, 182 (1950).
137. Z. Koba, Progr. Theoret. Phys. (Kyoto), 6, 322 (1951).
138. T. Tsuneto, Phys. Rev., 118, 1029 (1960).
139. A. Bardasis and J. R. Schrieffer, Phys. Rev., 121, 1050 (1961).
140. Subsequent to the experiments of Deaver and Fairbank and
of Doll and Nabauer, Little and Parks performed a beautiful
experiment in which they observed a periodic variation in the
transition temperature of a superconducting cylinder as a
function of the enclosed magnetic flux <f>. Their results
confirm the quantized flux unit hc\2e\ see W. A. Little
and R. D, Parks, Phys. Rev. Letters, 9, 9 (1962); W. A.
Little, Rev. Mod. Phys., 36, 264 (1964).
141. A. Bohr and B. R. Mottelson, Phys. Rev., 125, 495 (1962).
142. W. D. Knight, Solid State Physics, Vol. 2, Seitz and Turnbull
(eds.), Academic, New York, 1956.
143. F. Reif, Phys. Rev., 106, 208 (1957).
144. G.M. Androes and W. D. Knight, Phys. Rev. Letters, 2,
386 (1959).
145. R. A. Ferrell, Phys. Rev. Letters, 3, 262 (1959).
146. P. W. Anderson, Phys. Rev. Letters, 3, 325 (1959).
147. R. J. Noer and W. D. Knight, Rev. Mod. Phys., 36, 177 (1964).
148. V.Heine and A. B. Pippard, Phil. Mag., 3, 1046 (1958).
149. J. R. Schrieffer, Phys. Rev. Letters, 3, 323 (1959). See also
Ref. 9, p. 262. A similar suggestion was advanced by J. M.
Blatt, Proc. Superconductivity Conference, Cambridge, 1959,
unpublished.
150. L. N. Cooper, Phys. Rev. Letters, 8, 367 (1962); B. B. Schwartz
and L. N. Cooper, Rev. Mod. Phys., 36, 280 (1964).
151. A. M. Clogston, A. C. Gossard, V. Jaccarino, and Y. Yafet,
Phys. Rev. Letters, 9, 262 (1962).
152. R. Kubo and Y. Obata, J. Phys. Soc. Japan, 11, 547 (1956).
153. V. L. Ginsburg, J. Exptl. Theoret. Phys. (U SSR ), 29, 748
154. See Ref. 38a.
155. See Ref. 38b.
156. P. W. Anderson, Proc. Ravello Spring School, 1963.
157. (a) T. Kinsel et al., Rev. Mod. Phys., 36, 105 (1964).
(b) L. Dubeck et al., Rev. Mod. Phys., 36, 110 (1964).
158. (a) T. G. Berlincourt, Rev. Mod. Phys., 36, 19 (1964).
(b) P. W. Anderson and Y. B. Kim, Rev. Mod. Phys., 36,
39 (1964).
159. G. Chanin, E. A. Lynton, and B. Serin, Phys. Rev., 114,
719 (1959).
160. (a) B. T. Matthias et al., Rev. Mod. Phys., 36, 155 (1964).
(b) B. T. Matthias, T. H. Geballe, and V. B. Compton,
Rev. Mod. Phys., 35, 1 (1963).
161. (a) T. Tsuneto, Progr. Theoret. Phys. (Kyoto), 28, 857 (1962).
(b) P. G. de Gennes et al., Phys. Condensed Matter, 1, 176
(1963).
(c) D. Markowitz and L. P. Kadanoff, Phys. Rev., 131, 563
(1963).
(d) L. Gruenberg (to be published).
162. H. Suhl and B. T. Matthias, Phys. Rev., 114, 977 (1959).
163. W. Baltenspenger, Rev. Mod. Phys., 36, 157 (1964).
164. J. Bardeen, Rev. Mod. Phys., 34, 667 (1962); D. H. Douglass,
Rev. Mod. Phys., 36, 316 (1964).
165. P. G. de Gennes, Rev. Mod. Phys., 36, 225 (1964).
166. A. Bohr, B. R. Mottelson, and D. Pines, Phys. Rev., 110, 936
(1958).
167. (a) A. Bohr and B. R. Mottelson (book to be published).
(b) M. Baranger, Cargese Lectures 1962 in Theoretical
Physics, M. Levy (ed.), Benjamin, New York, 1963.
168. Y. Nambu and G. Jona-Lasinio, Phys. Rev., 122, 345
(1961); 124, 246 (1961).
169. J. C. Fisher, Phys. Rev., 129, 1414 (1963).
170. K. A. Brueckner, T. Soda, P. W. Anderson, and P. Morel,
Phys. Rev., 118, 1442 (1960).
171. W. A. Little, Rev. Mod. Phys., 36, 264 (1964).
172. (a) A. A. Abrikosov and L. P. Gorkov, J . Exptl. Theoret.
Phys. (U SSR), 39, 1781 (1960); translated in Soviet Phys.
J E T P , 12, 1243 (1961).
(b) F. Reif and M. A. Woolf, Phys. Rev. Letters, 9, 315
(1962).
(c) J. C. Phillips, Phys. Rev. Letters, 10, 96 (1963).
(d) H. Suhl and D. R. Fredkin, Phys. Rev. Letters, 10, 131,
268 (1963).
INDEX
Abrikosov vorticies, 23 Collective magnetization, 246

Acoustic attenuation, 8, 62, 67, Collective modes, 27, 224, 237-
76, 77 240
Bogoliubov-Anderson mode,
237-240
Backflow, 226-227 Condensation energy, 42
Bardeen, J., and Pines, D., 184 Conductivity, infinite, 205
BCS theory, 2, 35 Cooper’s problem, 28
energy-gap equation, 39, 53 Critical magnetic field, 5, 42,
wave function, 39 55
Bogoliubov-Valatin transfor Current-carrying state, 57
mation, 47, 63 Current, paramagnetic and dia
Bose-Einstein gas, 39 magnetic, 207
condensation, 32, 57 Currents, persistent, 205, 244
Bose-Einstein statistics, 38,
257-265
Bubble graphs, 138 Damping effects, 27, 73
(see also Quasi-particle)
Dielectric function, 138-145
Coherence effects, 8, 64, 70, Dielectric function for electron-
74 phonon system, 152
329
330 Index
Dyson’s equation, for electrons, Gauge invariance, 13, 209, 224-

130 235
in Nambu formalism, 179 Ginsburg-Landau theory, 19-
for phonons, 149 23, 248
Gor’kov theory, 248
Gorter-Casimir two-fluid model,
Effective mass correction, due 9-10
to Coulomb interaction, Green’s function, approach to
147 the many-body problem,
due to electron-phonon inter 105
action, 156 one-electron, 80, 106
Eigenoperators, 49 analytic properties of, 115
Electrical conductivity, 4, 72 dispersion relation for, 115
Electromagnetic absorption, 4, for free Fermi gas, 108-111
72, 76, 77 physical interpretation of,
Electromagnetic response ker 116
nel, 208, 210, 220-224 spectral representation of,
Electrons, bare, 35, 95-98 112-115
Electron-ion Hamiltonian, 89- one-phonon, 124-125
91 perturbation series for, at
Electron-phonon interaction, finite temperature, 196
63, 98, 102 at zero temperature, 126-
screened matrix element, 150 131
Electron-phonon vertex func in superconductivity theory,
tion, 157, 181 164-202
Electronic specific heat, 55 two-electron, 111
Eliashberg interaction, 184
Energy gap, 7, 24
crystalline anisotropy of, 88 Hartree-Foch approximation,
equation, 41, 52-54, 188- 169-172
189 generalized, 172-178
Heisenberg picture, 103
Fermi liquid theory, 34
Flux quantization, 6, 15-17, Instability of the normal phase,
240-244 32, 164
Free energy, 55 Interaction picture, 103-104
Index 331
Irreducible polarizability, 138 Pairing approximation, 26, 36

Isotope effect, 3, 7, 193 Pairing condition, 25
Pairing correlations, 25
Pairing, modified antiparallel
Jellium, 100, 149 spins, 246
Josephson’s effect, 78, 84 parallel spins, 245
and time-reversal invariance,
Knight shift, 244-248 88
Pair-pair correlations, 2, 25
Particle number fluctuation,
Law of corresponding states, 90 40
Lead, 189 Pauli principle correlations, 39,
Linearized equations of motion, 134-135
49 Penetration depth, 11, 12
London’s rigidity, 2, 10-17, 203 Phenomenological theories, 9-
23
Phonons, bare, 92-95
Magnetic impurities, 88, 255 dressed, 149
Meissner-Oschenfeld effect, 4, longitudinal, 68, 189
11, 212, 215 in superconducting state, 181
Migdal’s theorem, 157-162 transverse, 189
breakdown of, 164-168 Phonon-electron interaction,
98-102
Nambu formalism, 169 Physical origin of coherence
Nonlocal electrodynamics, 17- factors, 74
19 Pippard’s coherence length, 34
Nonmagnetic impurities, 87, Pippard’s theory, 17-19
254 Precursor electromagnetic ab
Anderson theory, 254 sorption, 73
Normal fluid, 1 Pseudo-potential, 186-187
Nuclear pairing correlations,
255
Nuclear spin relaxation, 8, 69 Quasi-particle, 35
approximation, 122
breakdown of, 191
One-pair (Cooper) problem, 28 for electron-phonon sys
Orbital paramagnetism, 247 tem, 163
332 Index
energy and damping rate, 41, Spectral weight function, one-

118-119, 131 electron, 80, 113-115,
excitations, 44 123, 178
operators, 47 one-phonon, 124-125
physical interpretation of,
119-123
Random phase approximation, Spin-orbit interaction, 98, 245
for the electron gas, 138- Superconductors, gapless, 205,
148 216
for the electron-phonon sys type I, 5, 254
tem, 148-163 type II, 5, 254
generalized, for superconduc Superfluid, 1
tivity, 224
Retardation effects, 81
and damping effects, 153 Thermal conductivity, 8
Time-ordering symbol, 106
Schafroth, Blatt, and Butler Tunneling, 7, 78
theory, 32 density of states, 80, 190
Schrodinger picture, 103 Hamiltonian, 78
Screening, 137 Josephson, 78, 84
overscreening, 153 two-particle, 87
underscreening, 153 Two-fluid model, 27
Second-quantization formalism,
257-266
for bosons, 259-265 Vacuum fluctuations, 134
for fermions, 265-266 Van Hove singularities, 191
Self-energy, due to electron- Vertex function, electromag
phonon interaction, 200 netic, 229-240
electron (definition), 130 electron-phonon, 157
screened exchange approxi
mation (RPA), 150-151
Self-energy, phonon (defini W ard’s identity, generalized,
tion), 148-149 228-233

Untitled

Uploaded by

Copyright:

Available Formats

Untitled

Uploaded by

Document Information

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

Untitled

Uploaded by

Copyright:

Available Formats

T h eo ry of

C R C Press is an im print of the

First published 1964 by Westview Press

Copyright © 1999, 1964 by J. R. Schrieffer

Editor’s Foreword vii

M A N Y-BO D Y PRO BLEM 103

SUPERCO N D UCTIVITY 164

N o tes and R eferen ces 317

Anderson: Basic Notions of Condensed M atter Physics, ABC ppbk,

The material presented here is an outgrowth of a series of lectures

Since the first appearance of this book in 1964, the field of

On the theoretical side, there was early recognition that the

Santa Barbara, 1983

The phenomenon of superconductivity is a remarkable ex­

The amazing properties of superconductors (e.g., perfect

It is perhaps not surprising that the microscopic theory of

important simple experimental facts about superconductors. One

which permits the expulsion of B. This point is discussed further

F IG U R E l-l The deviations of the critical field from T u /n ’s law

If one has a multiply connected superconductor, e.g., a

F IG U R E 1-2 The electronic specific heat of Sn.

In the presence of a magnetic field the A-#-phase transition

F IG U R E 1-3 The temperature-dependent energy gap of Al as deter­

V = 2A(T). The temperature-dependent energy gap observed

where f n and f s were chosen to be

From the thermodynamic relation

for the temperature-dependent critical field. Thus, H c is pre­

The London Theory

ns + nn where n is the average number of electrons per unit

and one obtains perfect diamagnetism, as required. For solu­

if A is properly gauged. For the supercurrent to be conserved

We are, however, still free to add to A the gradient of any function

While Y X must be single valued, X will not be single valued in

F. London’s Justification of the London Theory

= - j=i i mc A<r) jf - r>d*r' ■• d3r»s

magnetic current of the superfluid vanishes, leaving the dia­

F IG U R E 1-4 Tw o concentric superconducting cylinders with a flux 0

Since A in the outer cylinder is the gradient of the scalar (00/277-),

of the effective charge of the entities making up the superfluid

where a is an empirical constant of order unity and £0 is a length

reduction £/£0 is not effective. In highly impure specimens A is

where n is the total number of electrons per unit volume. Roughly

minimizing the free-energy functional F(W , T) of the system.

Jf[W { r), T ] d 3r (1-39)

where we have used the thermodynamic relation between the

From (1-43) and (1-44) we find

3{T) A2(T) A2(T)

The current density is given by

,s(r) = — b ^ ~ e*2A(r) _ 2 ^ r {^ (r)V^ {r) - y (r)v y *w}

which states that W is equal to its equilibrium value (1-44), as

for the deviation W(r) leads to

Thus the perturbation dies away exponentially, with the charac­

where the last estimate uses the microscopic theory to relate

Gor’kov37 has given a derivation of the Ginsburg-Landau

In analogy with a free electron gas, normal (N) metals exhibit a

is a smoothly varying function of the quantum numbers i and j.

existed only between mates of a pair. The pair-pair correlations

collective modes, such as the plasmons, arising from residual inter­

To find the zero-spin eigenstates of the pair, we write

where the matrix element Fkk' is defined by

The phenomenon of superconductivity is a remarkable ex

F IG U R E 1-3 The temperature-dependent energy gap of Al as deter

for the temperature-dependent critical field. Thus, H c is pre

and one obtains perfect diamagnetism, as required. For solu

magnetic current of the superfluid vanishes, leaving the dia

,s(r) = — b ^ ~ e2A(r) _ 2 ^ r {^ (r)V^ {r) - y (r)v y w}

Thus the perturbation dies away exponentially, with the charac

collective modes, such as the plasmons, arising from residual inter

liberal interpretation of H red must be adopted. When the quasi

which is just the energy-gap equation (2-30e). Thus the excita

F IG U R E 3-3 The relative acoustic-attenuation coefficient for trans