Uploaded by choritrob

0.0 (0)

This page intentionally left blank

University of Rochester, USA

Ashok Das

World Scientific

NEW JERSEY

•

LONDON

•

SINGAPORE

•

BEIJING

•

SHANGHAI

•

HONG KONG

•

TA I P E I

•

CHENNAI

Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.

LECTURES ON QUANTUM FIELD THEORY Copyright © 2008 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.

ISBN-13 ISBN-10 ISBN-13 ISBN-10

978-981-283-285-6 981-283-285-8 978-981-283-286-3 (pbk) 981-283-286-6 (pbk)

Printed in Singapore.

To My friends and collaborators Josif and Susumu and to Ever caring and charming Kiron and Momo

This page intentionally left blank

Preface

Over the past several years I have taught a two-semester graduate course on quantum ﬁeld theory at the University of Rochester. In this course the ideas of quantum ﬁeld theory are developed in a traditional manner through canonical quantization. This book consists of my lectures in this course. At Rochester, we also teach a separate course on quantum ﬁeld theory based on the path integral approach and my lectures in that course have already been published by World Scientiﬁc in A. Das, Field Theory: A Path Integral Approach (Second Edition), World Scientiﬁc, Singapore (2006). The material in the present book should be thought of as complementary to this earlier book. In fact, in the present lectures, there is no attempt to develop the path integral methods, rather we use the results from path integrals with a brief discussion when needed. The topics covered in the present book contain exactly the material discussed in the two-semester course except for Chapter 10 (Dirac quantization) and Chapter 11 (Discrete symmetries) which have been added for completeness and are normally discussed in another course. Quantum ﬁeld theory is a vast subject and only selected topics, which I personally feel every graduate student in the subject should know, have been covered in these lectures. Needless to say, there are many other important topics which have not been discussed because of time constraints in the course (and space constraints in the book). However, all the material covered in this book has been presented in an informal (classroom like) setting with detailed derivations which should be helpful to students. A book of this size is bound to have many possible sources of error. However, since my lectures have already been used by various vii

viii

Preface

people in diﬀerent universities, I have been fortunate to have their feedback which I have incorporated into the book. In addition, several other people have read all the chapters carefully and I thank them all for their comments. In particular, it is a pleasure for me to thank Ms. Judy Mack and Professor Susumu Okubo for their tireless eﬀort in going through the entire material. I am personally grateful to Dr. John Boersma for painstakingly and meticulously checking all the mathematical derivations. Of course, any remaining errors and typos are my own. Like the subject itself, the list of references to topics in quantum ﬁeld theory is enormous and it is simply impossible to do justice to everyone who has contributed to the growth of the subject. I have in no way attempted to give an exhaustive list of references to the subject. Instead I have listed only a few suggestive references at the end of each chapter in the hope that the readers can get to the other references from these sources. The Feynman graphs in this book were drawn using Jaxodraw while most other ﬁgures were generated using PSTricks. I am grateful to the people who developed these extremely useful softwares. Finally, I would like to thank Dave Munson for helping out with various computer related problems. Ashok Das Rochester

Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Relativistic equations . . . . 1.1 Introduction . . . . . . 1.2 Notations. . . . . . . . 1.3 Klein-Gordon equation 1.3.1 Klein paradox . . 1.4 Dirac equation . . . . . 1.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vii 1 1 2 10 14 19 26 27 27 34 40 44 47 49 58 62

2

Solutions of the Dirac equation . . . . . . 2.1 Plane wave solutions . . . . . . . . 2.2 Normalization of the wave function 2.3 Spin of the Dirac particle. . . . . . 2.4 Continuity equation . . . . . . . . . 2.5 Dirac’s hole theory . . . . . . . . . 2.6 Properties of the Dirac matrices . . 2.6.1 Fierz rearrangement . . . . . 2.7 References . . . . . . . . . . . . . .

3

Properties of the Dirac equation . . . . . . . . . . 3.1 Lorentz transformations . . . . . . . . . . . 3.2 Covariance of the Dirac equation . . . . . . 3.3 Transformation of bilinears. . . . . . . . . . 3.4 Projection operators, completeness relation 3.5 Helicity . . . . . . . . . . . . . . . . . . . . . 3.6 Massless Dirac particle . . . . . . . . . . . . 3.7 Chirality . . . . . . . . . . . . . . . . . . . . 3.8 Non-relativistic limit of the Dirac equation. 3.9 Electron in an external magnetic ﬁeld . . . 3.10 Foldy-Wouthuysen transformation. . . . . . ix

. 65 . 65 . 72 . 82 . 84 . 92 . 94 . 99 . 105 . 107 . 111

x

Contents

3.11 Zitterbewegung . . . . . . . . . . . . . . . . . . . . . 117 3.12 References . . . . . . . . . . . . . . . . . . . . . . . . 122 4 Representations of Lorentz and Poincar´ groups . . . . . . e 4.1 Symmetry algebras . . . . . . . . . . . . . . . . . . . 4.1.1 Rotation . . . . . . . . . . . . . . . . . . . . . . 4.1.2 Translation . . . . . . . . . . . . . . . . . . . . 4.1.3 Lorentz transformation . . . . . . . . . . . . . 4.1.4 Poincar´ transformation . . . . . . . . . . . . . e 4.2 Representations of the Lorentz group . . . . . . . . . 4.2.1 Similarity transformations and representations 4.3 Unitary representations of the Poincar´ group . . . . e 4.3.1 Massive representation . . . . . . . . . . . . . . 4.3.2 Massless representation . . . . . . . . . . . . . 4.4 References . . . . . . . . . . . . . . . . . . . . . . . . Free 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 Klein-Gordon ﬁeld theory . . . . . . . . . Introduction . . . . . . . . . . . . . . . . Lagrangian density . . . . . . . . . . . . Quantization . . . . . . . . . . . . . . . . Field decomposition. . . . . . . . . . . . Creation and annihilation operators. . . Energy eigenstates . . . . . . . . . . . . Physical meaning of energy eigenstates . Green’s functions . . . . . . . . . . . . . Covariant commutation relations . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 125 125 129 130 133 135 140 147 151 155 160 161 161 163 167 171 175 186 190 194 205 209 211 211 215 219 223 229 233 241 246 254

5

6

Self-interacting scalar ﬁeld theory . . . . . . . . . . . 6.1 N¨ther’s theorem . . . . . . . . . . . . . . . . . o 6.1.1 Space-time translation . . . . . . . . . . . 6.2 Self-interacting φ4 theory. . . . . . . . . . . . . 6.3 Interaction picture and time evolution operator 6.4 S-matrix . . . . . . . . . . . . . . . . . . . . . . 6.5 Normal ordered product and Wick’s theorem . 6.6 Time ordered products and Wick’s theorem . . 6.7 Spectral representation and dispersion relation 6.8 References . . . . . . . . . . . . . . . . . . . . .

Contents

xi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . the Goldstone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 257 260 263 268 270 281 283 285 285 286 291 297 300 303 305 308 312 318 325 327 327 330 335 342 347 350 355 360 376 379 379 384 390 395 401 407

7

Complex scalar ﬁeld theory . . . . . . . . . 7.1 Quantization . . . . . . . . . . . . . . 7.2 Field decomposition. . . . . . . . . . 7.3 Charge operator . . . . . . . . . . . . 7.4 Green’s functions . . . . . . . . . . . 7.5 Spontaneous symmetry breaking and theorem . . . . . . . . . . . . . . . . 7.6 Electromagnetic coupling. . . . . . . 7.7 References . . . . . . . . . . . . . . . Dirac 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 8.9 8.10 8.11

8

ﬁeld theory. . . . . . . . . . . . . . . . . . . Pauli exclusion principle . . . . . . . . . . . Quantization of the Dirac ﬁeld. . . . . . . . Field decomposition. . . . . . . . . . . . . . Charge operator . . . . . . . . . . . . . . . . Green’s functions . . . . . . . . . . . . . . . Covariant anti-commutation relations . . . . Normal ordered and time ordered products Massless Dirac ﬁelds . . . . . . . . . . . . . Yukawa interaction . . . . . . . . . . . . . . Feynman diagrams . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . .

9

Maxwell ﬁeld theory . . . . . . . . . . . . . . 9.1 Maxwell’s equations. . . . . . . . . . . 9.2 Canonical quantization . . . . . . . . . 9.3 Field decomposition. . . . . . . . . . . 9.4 Photon propagator . . . . . . . . . . . 9.5 Quantum electrodynamics . . . . . . . 9.6 Physical processes . . . . . . . . . . . . 9.7 Ward-Takahashi identity in QED . . . 9.8 Covariant quantization of the Maxwell 9.9 References . . . . . . . . . . . . . . . . method for constrained systems . Constrained systems . . . . . . . Dirac method and Dirac bracket. Particle moving on a sphere . . . Relativistic particle . . . . . . . . Dirac ﬁeld theory . . . . . . . . . Maxwell ﬁeld theory . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10 Dirac 10.1 10.2 10.3 10.4 10.5 10.6

xii

Contents

10.7 References . . . . . . . . . . . . . . . . . . . . . . . . 413 11 Discrete symmetries . . . . . . . . . . . . . . . . . . . . . . 11.1 Parity. . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.1 Parity in quantum mechanics . . . . . . . . . . 11.1.2 Spin zero ﬁeld . . . . . . . . . . . . . . . . . . 11.1.3 Photon ﬁeld . . . . . . . . . . . . . . . . . . . . 11.1.4 Dirac ﬁeld . . . . . . . . . . . . . . . . . . . . . 11.2 Charge conjugation . . . . . . . . . . . . . . . . . . . 11.2.1 Spin zero ﬁeld . . . . . . . . . . . . . . . . . . 11.2.2 Dirac ﬁeld . . . . . . . . . . . . . . . . . . . . . 11.2.3 Majorana fermions . . . . . . . . . . . . . . . . 11.2.4 Eigenstates of charge conjugation . . . . . . . . 11.3 Time reversal . . . . . . . . . . . . . . . . . . . . . . 11.3.1 Spin zero ﬁeld and Maxwell’s theory . . . . . . 11.3.2 Dirac ﬁelds . . . . . . . . . . . . . . . . . . . . 11.3.3 Consequences of T invariance . . . . . . . . . . 11.3.4 Electric dipole moment of neutron . . . . . . . 11.4 CPT theorem . . . . . . . . . . . . . . . . . . . . . . 11.4.1 Equality of mass for particles and antiparticles 11.4.2 Electric charge for particles and antiparticles . 11.4.3 Equality of lifetimes for particles and antiparticles . . . . . . . . . . . . . . . . . . . . . . . . 11.5 References . . . . . . . . . . . . . . . . . . . . . . . . 12 Yang-Mills theory . . . . . . . . . . . . . . . . . . . 12.1 Non-Abelian gauge theories . . . . . . . . . . 12.2 Canonical quantization of Yang-Mills theory . 12.3 Path integral quantization of gauge theories . 12.4 Path integral quantization of tensor ﬁelds . . 12.5 References . . . . . . . . . . . . . . . . . . . . 13 BRST invariance and its consequences . . . 13.1 BRST symmetry . . . . . . . . . . . 13.2 Covariant quantization of Yang-Mills 13.3 Unitarity . . . . . . . . . . . . . . . . 13.4 Slavnov-Taylor identity . . . . . . . . 13.5 Feynman rules . . . . . . . . . . . . . 13.6 Ghost free gauges . . . . . . . . . . . 13.7 References . . . . . . . . . . . . . . . . . . . . . . . . . theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415 415 417 424 428 429 436 437 441 449 453 458 464 467 473 477 479 479 480 480 482 485 485 502 512 530 542 545 545 550 561 565 571 578 581

Contents

xiii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583 583 589 596 599 601 605 616 619 619 621 623 631 638 647 656 666 669 669 679 690 692 721 724 732 733 733 739 744 748 759 766

14 Higgs phenomenon and the standard model . 14.1 St¨ckelberg formalism . . . . . . . . . u 14.2 Higgs phenomenon . . . . . . . . . . . 14.3 The standard model. . . . . . . . . . . 14.3.1 Field content . . . . . . . . . . . 14.3.2 Lagrangian density . . . . . . . . 14.3.3 Spontaneous symmetry breaking 14.4 References . . . . . . . . . . . . . . . . 15 Regularization of Feynman diagrams . 15.1 Introduction . . . . . . . . . . . 15.2 Loop expansion . . . . . . . . . 15.3 Cut-oﬀ regularization . . . . . . 15.3.1 Calculation in the Yukawa 15.4 Pauli-Villars regularization . . . 15.5 Dimensional regularization . . . 15.5.1 Calculations in QED . . . 15.6 References . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

16 Renormalization theory . . . . . . . . . . 16.1 Superﬁcial degree of divergence . . 16.2 A brief history of renormalization . 16.3 Schwinger-Dyson equation . . . . . 16.4 BPHZ renormalization . . . . . . . 16.5 Renormalization of gauge theories . 16.6 Anomalous Ward identity . . . . . 16.7 References . . . . . . . . . . . . . .

17 Renormalization group and equation . . . . . . . . 17.1 Gell-Mann-Low equation . . . . . . . . . . . 17.2 Renormalization group . . . . . . . . . . . . 17.3 Renormalization group equation . . . . . . . 17.4 Solving the renormalization group equation 17.5 Callan-Symanzik equation . . . . . . . . . . 17.6 References . . . . . . . . . . . . . . . . . . .

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 769

This page intentionally left blank

in this case. ∂t i (1. t) represents the wave function of the system which corresponds to the probability amplitude for ﬁnding the particle at the coordinate x at a given time t and the Hamiltonian.2) with p denoting the momentum of the particle and V (x) representing the potential through which the particle moves. non-relativistic physical system and promote each of the observables to a Hermitian operator.1) takes the form 1 . 2m H= (1. The time evolution of the quantum mechanical system. has the generic form p2 + V (x). even for a free particle.1 Introduction As we know.1) Here ψ(x.Chapter 1 Relativistic equations 1. non-relativistic quantum mechanics. is given by the time dependent Schr¨dinger equation which has the form o ∂ψ = Hψ. we start with the Hamiltonian description of the corresponding classical.) This formalism is clearly non-relativistic (non-covariant) which can be easily seen by noting that. in single particle. (Throughout the book we will use a bold symbol to represent a three dimensional quantity. H. the dynamical equation (1.

in this case. we will study how we can systematically develop a quantum mechanical description of a single relativistic particle and the diﬃculties associated with such a description. the momentum operator has the form i p → −i ∇. space and time are not treated on an equal footing in this case and. J3 ). x2 .2 1 Relativistic equations p2 ∂ψ = ψ. on the other hand. J = (J1 . for the ground state electron is c of the order of the ﬁne structure constant).6) . the ground state electron is fairly relativistic ( v .3) ∂t 2m In the coordinate basis. In this chapter. (1. We note that in the three dimensional Euclidean space.) Thus. must treat space and time coordinates on an equal footing and remain form invariant in all inertial frames (Lorentz frames). J2 . (1. takes o the form 2 ∂ψ =− ∇2 ψ. which we are all familiar with. Let us also recall that. there is a need to generalize the quantum mechanical description to relativistic systems. a vector is labelled uniquely by its three components. (1. A = (A1 . Consequently. consequently. A2 . x = (x1 . (We denote three dimensional vectors in boldface. A relativistic equation. even for a simple fundamental system such as the Hydrogen atom.4) so that the time dependent Schr¨dinger equation. A3 ). the equation cannot have the same form (covariant) in diﬀerent Lorentz frames. Therefore.2 Notations Before proceeding any further. let us ﬁx our notations. i 1. (1. x3 ).5) ∂t 2m This equation is linear in the time derivative while it is quadratic in the space derivatives.

the scalar product of any two arbitrary vectors is deﬁned to be A · B = Ai Bi = δij Ai Bj = δij Ai Bj . Let us note from the deﬁnition of the length of a vector in Euclidean space that. unlike the case of the Euclidean space. Consequently. (µ = 0. represents the metric of the Euclidean space and is trivial (in the sense that all the nonzero components are simply unity). we can again deﬁne vectors in this manifold. there are now two distinct four vectors that we can deﬁne on this manifold.7) where repeated indices are assumed to be summed. namely. (1. 1. for any vector. as we know. A2 = 0. consequently. namely.1. 3 and we are being a little sloppy in representing the four vector by what may seem like its component) . In such a space. it is necessarily positive deﬁnite. However. any point in this manifold will be speciﬁed uniquely by four coordinates and. if and only if A = 0.9) When we treat space and time on an equal footing and enlarge our three dimensional Euclidean manifold to the four dimensional spacetime manifold. (1. The scalar product of two vectors is invariant under rotations of the three dimensional space which is the maximal symmetry group of the Euclidean space that leaves the origin invariant.2 Notations 3 where x and J represent respectively the position and the angular momentum vectors while A stands for any arbitrary vector. (1. these would now consist of four components.8) The Kronecker delta. in this case. δij . However. 2. This also allows us to deﬁne the length of a vector simply as A2 = A · A = Ai Ai = δij Ai Aj = δij Ai Aj . A2 ≥ 0. Namely. any vector would also have four components. it does not matter whether we write the indices “up” or “down”.

−x). namely. (1. On a more fundamental level.10) Here c represents the speed of light (necessary to give the same dimension to all the components) and we note that the two four vectors simply represent the two distinct possible ways space and time components can be embedded into the four vector. and the covariant metric tensor. x).13) The contravariant metric tensor. commonly known as the Minkowski space.11).4 1 Relativistic equations xµ = (ct. we can write them in the matrix form as 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . xµ = η µν xν . −). 0 0 0 −1 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . the two four vectors have distinct transformation properties under Lorentz transformations (in fact. (1. η µν .10) as well as using (1. are inverses of each other. The contravariant and the covariant vectors are related to each other through the metric tensor of the four dimensional manifold. we can immediately read out the metric tensors for the four dimensional Minkowski space which are diagonal with the signature (+. −.12) ηµν (1. since they satisfy .11) From the forms of the contravariant and the covariant vectors in (1. one transforms inversely with respect to the other) and are known respectively as contravariant and covariant vectors. ηµν . xµ = (ct. Namely. −. 0 0 0 −1 η µν (1. xµ = ηµν xν .

the nonuniqueness in the choice of the metric tensors reﬂects the nonuniqueness of the embedding of space and time components into a four vector. are independent of the choice of a metric. η µν = η νµ . namely.17) Since the contravariant and the covariant vectors transform in an inverse manner. B µ = (B 0 . (1.) . use diﬀerent metric conventions and you should be careful in reading the literature. (As is clear from the above discussion.1.15) This particular choice of the metric is conventionally known as the Bjorken-Drell metric and this is what we will be using throughout these lectures. any product of Lorentz tensors deﬁnes a scalar if all the Lorentz indices are contracted. (1. each is symmetric as they are expected to be. Diﬀerent authors. B). if there is no free Lorentz index. A). (Two Lorentz indices are said to be contracted if a contravariant and a covariant index are summed over all possible values.16) we can deﬁne an invariant scalar product of the two vectors as A · B = Aµ Bµ = Aµ B µ = A0 B 0 − A · B. ηµν = ηνµ . (1.) Given two arbitrary four vectors Aµ = (A0 .14) Furthermore.2 Notations 5 η µλ ηλν = δµν . however. = η µν Aµ Bν = ηµν Aµ B ν (1. This is the generalization of the scalar product of the three dimensional Euclidean space to the four dimensional Minkowski space and is invariant under Lorentz transformations which are the analogs of rotations in Minkowski space. In fact. however. Physical results. namely. such a product is easily seen to be invariant under Lorentz transformations.

6

1 Relativistic equations

Given this, we note that the length of a (four) vector in Minkowski space can be determined to have the form A2 = A · A = η µν Aµ Aν = ηµν Aµ Aν = (A0 )2 − A2 . (1.18)

Unlike the Euclidean space, however, here we see that the length of a vector need not always be positive semi-deﬁnite. In fact, if we look at the Minkowski space itself, we ﬁnd that x2 = xµ xµ = ηµν xµ xν = c2 t2 − x2 . (1.19)

This is the invariant length (of any point from the origin) in this space. The invariant length between two points inﬁnitesimally close to each other follows from this to be ds2 = c2 dτ 2 = ηµν dxµ dxν , (1.20)

where τ is known as the proper time. For coordinates which satisfy (see (1.19), we will set c = 1 from now on for simplicity) x2 = t2 − x2 > 0, (1.21)

we say that the region of space-time is time-like for obvious reasons. On the other hand, for coordinates which satisfy x2 = t2 − x2 < 0, (1.22)

the region of space-time is known as space-like. The boundary of the two regions, namely, the region for which x2 = t2 − x2 = 0, (1.23)

deﬁnes trajectories for light-like particles and is, consequently, known as the light-like region. (Light-like vectors, for which the invariant length vanishes, are nontrivial unlike the case of the Euclidean space.)

1.2 Notations

7

space-like

time-like

lig ht -li ke

t

ike t-l h lig

space-like x

Figure 1.1: Diﬀerent invariant regions of Minkowski space. Thus, we see that, unlike the Euclidean space, the Minkowski space-time manifold separates into four invariant wedges (regions which do not mix under Lorentz transformations), which in a two dimensional projection has the form shown in Fig. 1.1. The diﬀerent invariant wedges are known as

t > 0, t < 0, x2 < 0 :

x2 ≥ 0 :

future light cone, past light cone, (1.24)

x2 ≥ 0 :

space − like.

All physical processes are assumed to take place in the future light cone or the forward light cone deﬁned by

t>0

and x2 ≥ 0.

time-like

(1.25)

Given the contravariant and the covariant coordinates, we can deﬁne the contragradient and the cogradient respectively as (c = 1)

8

1 Relativistic equations

∂µ = ∂µ =

∂ = ∂xµ ∂ = ∂xµ

∂ , −∇ , ∂t ∂ ,∇ . ∂t (1.26)

From these, we can construct the Lorentz invariant quadratic operator ∂2 − ∇2 , ∂t2

= ∂ 2 = ∂ µ ∂µ =

(1.27)

which is known as the D’Alembertian. It is the generalization of the Laplacian to the four dimensional Minkowski space. Let us note next that energy and momentum also deﬁne four vectors in this case. (Namely, they transform like four vectors under Lorentz transformations.) Thus, we can write (remember that c = 1, otherwise, we have to write E ) c pµ = (E, p), pµ = (E, −p). (1.28)

Given the energy-momentum four vectors, we can construct the Lorentz scalar p2 = pµ pµ = E 2 − p2 . The Einstein relation for a free particle (remember c = 1) E 2 = p2 + m2 , (1.30) (1.29)

where m represents the rest mass of the particle, can now be seen as the Lorentz invariant condition p2 = E 2 − p2 = m2 . (1.31)

1.2 Notations

9

In other words, in this space, the energy and the momentum of a free particle must lie on a hyperbola satisfying the above relation. We already know that the coordinate representation of the energy and the momentum operators takes the forms

E → i

∂ , ∂t p → −i ∇.

(1.32)

We can combine these to write the coordinate representation for the energy-momentum four vector operator as

pµ = i ∂ µ = i pµ = i ∂µ = i

∂ = ∂xµ ∂ = ∂xµ

i i

∂ , −i ∇ , ∂t ∂ ,i ∇ . ∂t (1.33)

Finally, let us note that in the four dimensional space-time, we can construct two totally antisymmetric fourth rank tensors ǫµνλρ , ǫµνλρ , the four dimensional contravariant and covariant Levi-Civita tensors respectively. We will choose the normalization ǫ0123 = 1 = −ǫ0123 so that

ǫ0ijk = ǫijk = −ǫ0ijk ,

(1.34)

where ǫijk denotes the three dimensional Levi-Civita tensor with ǫ123 = 1. An anti-symmetric tensor such as ǫi jk is then understood to denote

ǫi jk = η iℓ ǫℓjk ,

(1.35)

and so on. This completes the review of all the essential basic notation that we will be using in this book. We will introduce new notations as they arise in the context of our discussions.

10 1.3 Klein-Gordon equation

1 Relativistic equations

With all these basics, we are now ready to write down the simplest of the relativistic equations. Note that in the case of a non-relativistic particle, we start with the non-relativistic energy-momentum relation p2 + V (x), 2m

E=

(1.36)

and promote the dynamical variables (observables) to Hermitian operators to obtain the time-dependent Schr¨dinger equation (see (1.1)) o ∂ψ = ∂t

2

i

−

2m

∇2 + V (x) ψ.

(1.37)

Let us consider the simplest of relativistic systems, namely, a relativistic free particle of mass m. In this case, we have seen that the energy-momentum relation is none other than the Einstein relation (1.30), namely,

E 2 = p2 + m2 , or, E 2 − p2 = pµ pµ = m2 . (1.38)

Thus, as before, promoting these to operators, we obtain the simplest relativistic quantum mechanical equation to be (see (1.33))

pµ pµ φ = m2 φ, or, or, (i ∂ µ )(i ∂µ )φ = m2 φ, −

2

φ = m2 φ.

(1.39)

Setting = 1 from now on for simplicity, the equation above takes the form + m2 )φ = 0.

(

(1.40)

1.3 Klein-Gordon equation

11

Since the operator in the parenthesis is a Lorentz scalar and since we assume the quantum mechanical wave function, φ(x, t), to be a scalar function, this equation is invariant under Lorentz transformations. This equation, (1.40), is known as the Klein-Gordon equation and, for m = 0, or when the rest mass vanishes, it reduces to the wave equation (recall Maxwell’s equations). Like the wave equation, the Klein-Gordon equation also has plane wave solutions which are characteristic of free particle solutions. In fact, the functions e∓ik·x = e∓ikµ x = e∓ik

µ µx

µ

= e∓i(k0 t−k·x) ,

(1.41)

with kµ = (k0 , k) are eigenfunctions of the energy-momentum operator, namely, using (1.33) (remember that = 1) we obtain ∂ ∓ik·x e = ±kµ e∓ik·x , ∂xµ

pµ e∓ik·x = i∂ µ e∓ik·x = i

(1.42)

so that ±kµ are the eigenvalues of the energy-momentum operator. (In fact, the eigenvalues should be ± kµ , but we have set = 1.) This shows that the plane waves deﬁne a solution of the Klein-Gordon equation provided

k2 − m2 = (k0 )2 − k2 − m2 = 0, or, k0 = ±E = ± k2 + m2 . (1.43)

Thus, we see the ﬁrst peculiarity of the Klein-Gordon equation (which is a relativistic equation), namely, that it allows for both positive and negative energy solutions. This basically arises from the fact that, for a relativistic particle (even a free one), the energy-momentum relation is given by the Einstein relation which is a quadratic relation in E, as opposed to the case of a non-relativistic particle, where the energy-momentum relation is linear in E. If we accept the Klein-Gordon equation as describing a free, relativistic, quantum mechanical particle of mass m, then, we will see shortly that the presence of the negative energy solutions would render the theory inconsistent.

12

1 Relativistic equations

To proceed further, let us note that the Klein-Gordon equation and its complex conjugate (remember that a quantum mechanical wave function is, in general, complex), namely, + m2 )φ = 0, + m2 )φ∗ = 0, (1.44)

( ( would imply

φ∗ φ − φ φ∗ = 0, or, or, ∂ ∂t φ∗ ∂φ∗ ∂φ −φ ∂t ∂t

∂µ (φ∗ ∂ µ φ − φ∂ µ φ∗ ) = 0, − ∇ · (φ∗ ∇φ − φ∇φ∗ ) = 0. (1.45)

Deﬁning the probability current density four vector as J µ = (j 0 , J) = (ρ, J), where 1 (φ∗ ∇φ − φ∇φ∗ ) , 2im i 2m φ∗ ∂φ∗ ∂φ −φ ∂t ∂t , (1.47) (1.46)

J = ρ =

we note that equation (1.45) can be written as a continuity equation for the probability current, namely, ∂µ J µ = ∂ρ + ∇ · J = 0. ∂t

(1.48)

The probability current density, 1 (φ∗ ∇φ − φ∇φ∗ ) , 2im

J=

(1.49)

1.3 Klein-Gordon equation

13

of course, has the same form as in non-relativistic quantum mechanics. However, we note that the form of the probability density (which results from the requirement of covariance) i 2m ∂φ∗ ∂φ −φ ∂t ∂t

ρ=

φ∗

,

(1.50)

is quite diﬀerent from that in non-relativistic quantum mechanics and it is here that the problem of the negative energy states shows up. For example, even for the simplest of solutions, namely, plane waves of the form φ(x) = e−ik·x , we obtain k0 E i (−ik0 − ik0 ) = =± . 2m m m

(1.51)

ρ=

(1.52)

Since energy can take both positive and negative values, it follows that ρ cannot truly represent the probability density which, by deﬁnition, has to be positive semi-deﬁnite. It is worth noting here that this problem really arises because the Klein-Gordon equation, unlike the time dependent Schr¨dinger equation, is second order in time o derivatives. This has the consequence that the probability density involves a ﬁrst order time derivative and that is how the problem of the negative energy states enters. (Note that if the equation is second order in the space derivatives, then covariance would require that it be second order in time derivative as well. This would, in turn, lead to the diﬃculty with the probability density being positive semideﬁnite.) One can, of course, ask whether we can restrict ourselves to positive energy solutions only in order to avoid the diﬃculty with the interpretation of ρ. Classically, we can do this. However, quantum mechanically, we cannot arbitrarily impose this for a variety of reasons. The simplest way to see this is to note that the positive energy solutions alone do not deﬁne a complete set of states (basis) in the Hilbert space and, consequently, even if we restrict the states to be of positive energy to begin with, negative energy states may be

14

1 Relativistic equations

generated through quantum mechanical corrections. It is for these reasons that the Klein-Gordon equation was abandoned as a quantum mechanical equation for a single relativistic particle. However, as we will see later, this equation is quite meaningful as a relativistic ﬁeld equation.

1.3.1 Klein paradox. Let us consider a charged scalar particle de-

scribed by the Klein-Gordon equation (1.40) in an external electromagnetic ﬁeld. We recall that the coupling of a charged particle to an electromagnetic ﬁeld is given by the minimal coupling

pµ → pµ − eAµ , or, ∂µ → ∂µ + ieAµ , (1.53) where we have used the coordinate representation for the momentum as in (1.33) and Aµ denotes the vector potential associated with the electromagnetic ﬁeld. In this case, therefore, the scalar particle will satisfy the minimally coupled Klein-Gordon equation (∂µ + ieAµ ) (∂ µ + ieAµ ) + m2 φ(x) = 0. (1.54)

As a result, the probability current density in (1.46) can be determined to have the form Jµ = ← → i φ∗ (x) ∂ µ φ(x) + 2ieAµ φ∗ (x)φ(x) , 2m

(1.55)

where we have deﬁned ← → A ∂ µ B = A(∂ µ B) − (∂ µ A) B. (1.56)

With this general description, let us consider the scattering of a charged scalar (Klein-Gordon) particle with positive energy from a constant electrostatic potential. In this case, therefore, we have A = 0, A0 = Φ = constant. (1.57)

1.3 Klein-Gordon equation

15

For simplicity, let us assume the constant electrostatic potential to be of the form 0, z < 0,

Φ(z) =

Φ0 , z > 0,

(1.58)

**and we assume that the particle is incident on the potential along the z-axis as shown in Fig. 1.2.
**

V

eΦ0

I z=0

II z→

Figure 1.2: Klein-Gordon particle scattering from a constant electrostatic potential. The dynamical equations will now be diﬀerent in the two regions, z < 0 (region I) and z > 0 (region II), and have the forms + m2 φI = 0,

ψ

ψ

z < 0,

ψ ψ

+ m2 + 2ieΦ0

∂ − e2 Φ2 φII = 0, 0 ∂t

z > 0.

(1.59)

In region I, there will be an incident as well as a reﬂected wave so that we can write φI (t, z) = e−iEt eipz + A e−ipz , z < 0, (1.60)

**while in region II, we only expect a transmitted wave of the form φII (t, z) = B e−iEt+ip z ,
**

′

z > 0,

(1.61)

16

1 Relativistic equations

where A, B are related respectively to reﬂection and transmission coeﬃcients. We note here that the continuity of the wave function at the boundary z = 0 requires that the energy be the same in the two regions. For the wave functions in (1.60) and (1.61) to satisfy the respective equations in (1.59), we must have E = p′ = ± = ± p2 + m2 , (E − eΦ0 )2 − m2 (E − eΦ0 + m) (E − eΦ0 − m). (1.62)

Here we have used the fact that the energy of the incident particle is positive and, therefore, the square root in the ﬁrst equation in (1.62) is with a positive sign. However, the sign of the square root in the second relation remains to be ﬁxed. Let us note from the second relation in (1.62) that p′ is real for both E − eΦ0 > m (weak potential) and for E − eΦ0 < −m (strong potential). However, for a potential of intermediate strength satisfying −m < E − eΦ0 < m, we note that p′ is purely imaginary. Thus, the behavior of the transmitted wave depends on the strength of the potential. As a result, in this second case, we must have p′ = i|p′ |, −m < E − eΦ0 < m, (1.63)

in order that the wave function is damped in region II. To determine the sign of the square root in the cases when p′ is real, let us note from the second relation in (1.62) that the group velocity of the transmitted wave is given by vgroup = p′ ∂E = . ∂p′ E − eΦ0 (1.64)

Since we expect the transmitted wave to be travelling to the right, we determine from (1.64) that p′ > 0, for E − eΦ0 > 0, p′ < 0

for E − eΦ0 < 0.

(1.65)

1.3 Klein-Gordon equation

17

This, therefore, ﬁxes the sign of the square root in the second relation in (1.62) for various cases. Matching the wave functions in (1.60) and (1.61) and their ﬁrst derivatives at the boundary z = 0, we determine p′ B, p

1 + A = B, so that we determine p − p′ , p + p′

1−A=

(1.66)

A=

B=

2p . p + p′

(1.67)

Let us next determine the probability current densities associated with the diﬀerent beams. From (1.55) as well as the form of the potential in (1.58) we obtain ˆ Jinc = z · Jinc = Jreﬂ = −ˆ · Jreﬂ z p , m p (p − p′ )(p − (p′ )∗ ) = , m (p + p′ )(p + (p′ )∗ ) 4p2 (p′ + (p′ )∗ ) , 2m (p + p′ )(p + (p′ )∗ ) (1.68)

ˆ Jtrans = z · Jtrans =

where we have used (1.67) as well as the fact that, while p is real and positive, p′ can be positive or negative or even imaginary depending on the strength of the potential (see (1.63) and (1.65)). We can now determine the reﬂection and the transmission coeﬃcients simply as (p − p′ )(p − (p′ )∗ ) Jreﬂ = , Jinc (p + p′ )(p + (p′ )∗ ) Jtrans 2p(p′ + (p′ )∗ ) = . Jinc (p + p′ )(p + (p′ )∗ ) (1.69)

R = T =

We see from the reﬂection and the transmission coeﬃcients that (p − p′ )(p − (p′ )∗ ) + 2p(p′ + (p′ )∗ ) = 1, (p + p′ )(p + (p′ )∗ )

R+T =

(1.70)

18

1 Relativistic equations

so that the reﬂection and the transmission coeﬃcients satisfy unitarity for all strengths of the potential. However, let us now analyze the diﬀerent cases of the potential strengths individually. First, for the case, E − eΦ0 > m (weak potential), we see that p′ is real and positive and we have R= p − p′ p + p′

2

< 1,

T =

4pp′ > 0, (p + p′ )2

R + T = 1, (1.71)

which corresponds to the normal scenario in scattering. For the case of an intermediate potential strength, −m < E − eΦ0 < m, we note from (1.63) that p′ is purely imaginary in this case. As a result, it follows from (1.69) that R= (p − p′ )(p + p′ ) = 1, (p + p′ )(p − p′ ) T = 0, R + T = 1, (1.72)

so that the incident beam is totally reﬂected and there is no transmission in this case. The third case of the strong potential, E − eΦ0 < −m, is the most interesting. In this case, we note from (1.65) that p′ is real, but negative. As a result, from (1.69) we have p + |p′ | p − |p′ |

2

R=

> 1,

T =−

4p|p′ | < 0, (p − |p′ |)2

R + T = 1. (1.73)

Namely, even though unitarity is not violated, in this case the transmission coeﬃcient is negative and the reﬂection coeﬃcient exceeds unity. This is known as the Klein paradox and it contradicts our intuition from the one particle scatterings studied in non-relativistic quantum mechanics. On the other hand, if we go beyond the one particle description and assume that a suﬃciently strong enough electrostatic potential can produce particle-antiparticle pairs, there is no paradox. For example, the antiparticles are attracted by the barrier leading to a negative charged current moving to the right which explains the negative transmission coeﬃcient. On the other hand, the particles are reﬂected from the barrier and add to the totally reﬂected incident particles (which is already seen for intermediate strength potentials) to give a reﬂection coeﬃcient that exceeds unity.

1.4 Dirac equation

19

1.4 Dirac equation As we have seen, relativistic equations seem to imply the presence of both positive as well as negative energy solutions and that quantum mechanically, we need both these solutions to describe a physical system. Furthermore, as we have seen, the Klein-Gordon equation is second order in the time derivatives and this leads to the deﬁnition of the probability density which is ﬁrst order in the time derivative. Together with the negative energy solutions, this implies that the probability density can become negative which is inconsistent with the deﬁnition of a probability density. It is clear, therefore, that even if we cannot avoid the negative energy solutions, we can still possibly obtain a consistent probability density provided we have a relativistic equation which is ﬁrst order in the time derivative just like the time dependent Schr¨dinger equation. The diﬀerence, of course, is that o Lorentz invariance would require space and time to be treated on an equal footing and, therefore, such an equation, if we can ﬁnd it, must be ﬁrst order in both space and time derivatives. Clearly, this can be done provided we have a linear relation between energy and momentum operators. Let us recall that the Einstein relation gives E 2 = p2 + m2 . The positive square root of this gives

(1.74)

E=

p2 + m2 ,

(1.75)

which is far from a linear relation. Although the naive square root of the Einstein relation does not lead to a linear relation between the energy and the momentum variables, a matrix square root may, in fact, lead to such a relation. This is exactly what Dirac proposed. Let us, for example, write the Einstein relation as

E 2 − p2 = m2 , or,

p 2 = p µ p µ = m2 .

(1.76)

20

1 Relativistic equations

Let us consider this as a matrix relation (namely, an n × n identity matrix multiplying both sides). Let us further assume that there exist four linearly independent n×n matrices γ µ , µ = 0, 1, 2, 3, which are space-time independent such that p = γ µ pµ , /

(1.77)

represents the matrix square root of p2 . If this is true, then, by deﬁnition, we have p p = p 2 ½, // or, or, or, γ µ pµ γ ν pν = p2 ½, γ µ γ ν pµ pν = p2 ½, 1 µ ν (γ γ + γ ν γ µ )pµ pν = p2 ½. 2 (1.78)

Here ½ denotes the identity matrix (in the appropriate matrix space, in this case, n dimensional) and we have used the fact that the matrices, γ µ , are constant to move them past the momentum operators. For the relation (1.78) to be true, it is clear that the matrices, γ µ , have to satisfy the algebra (µ = 0, 1, 2, 3) γµγν + γν γµ = γµ, γν = 2η µν ½.

+

(1.79)

Here the brackets with a subscript “+” stand for the anti-commutator of two quantities deﬁned in (1.79) (sometimes it is also denoted by curly brackets which we will not use to avoid confusion with Poisson brackets) and this algebra is known as the Cliﬀord algebra. We see that if we can ﬁnd a set of four linearly independent constant matrices satisfying the Cliﬀord algebra, then, we can obtain a matrix square root of p2 which would be linear in energy and momentum. Before going into an actual determination of such matrices, let us look at the consequences of such a possibility. In this case, the solutions of the equation (sign of the mass term is irrelevant and the mass term is multiplied by the identity matrix which we do not write explicitly)

1.4 Dirac equation

21

pψ = mψ, / would automatically satisfy the Einstein relation. Namely, p(pψ) = mpψ, // / or, p2 ψ = m2 ψ.

(1.80)

(1.81)

Furthermore, since the new equation is linear in the energy and momentum variables, it will, consequently, be linear in the space and time derivatives. This is, of course, what we would like for a consistent deﬁnition of the probability density. The equation (1.80) (or its coordinate representation) is known as the Dirac equation. To determine the matrices, γ µ , and their dimensionality, let us note that the Cliﬀord algebra in (1.79) γµ, γν = 2η µν ½, µ = 0, 1, 2, 3, (1.82)

+

can be written out explicitly as (γ 0 )2 = ½, (γ i )2 = −½, for any ﬁxed i = 1, 2, 3, γ 0 γ i + γ i γ 0 = 0, i = j. (1.83)

γ i γ j + γ j γ i = 0,

We can choose any one of the matrices to be diagonal and without loss of generality, let us choose b1 0 · · · 0 b2 · · · γ0 = . .. . . . 0 0 ··· 0 0 . . . .

(1.84)

bn

From the fact that (γ 0 )2 = ½, we conclude that each of the diagonal elements in γ 0 must be ±1, namely,

22

1 Relativistic equations

bα = ±1,

α = 1, 2, · · · , n.

(1.85)

Let us note next that using the relations from the Cliﬀord algebra in (1.83), for a ﬁxed i, we obtain Tr γ i γ 0 γ i = Tr γ i (−γ i γ 0 ) = −Tr (γ i )2 γ 0 = Tr γ 0 , (1.86)

where “Tr” denotes trace over the matrix indices. On the other hand, the cyclicity property of the trace, namely, Tr ABC = Tr CAB, leads to Tr γ i γ 0 γ i = Tr (γ i )2 γ 0 = −Tr γ 0 . Thus, comparing Eqs. (1.86) and (1.88), we obtain Tr γ i γ 0 γ i = Tr γ 0 = −Tr γ 0 , (1.88) (1.87)

or,

Tr γ 0 = 0.

(1.89)

For this to be true, we conclude that γ 0 must have as many diagonal elements with value +1 as with −1. Consequently, the γ µ matrices must be even dimensional. Let us assume that n = 2N . The simplest nontrivial matrix structure would arise for N = 1 when the matrices would be two dimensional (namely, 2 × 2 matrices). We know that the three Pauli matrices along with the identity matrix deﬁne a complete basis for 2 × 2 matrices. However, as we know, they do not satisfy the Cliﬀord algebra. Namely, if we deﬁne σ µ = (½, σ), then, σµ , σν = 2η µν ½. (1.90)

+

In fact, we know that in two dimensions, there cannot exist four anti-commuting matrices.

1.4 Dirac equation

23

The next choice is N = 2 for which the matrices will be four dimensional (4 × 4 matrices). In this case, we can ﬁnd a set of four linearly independent, constant matrices which satisfy the Cliﬀord algebra. A particular choice of these matrices, for example, has the form

γ0 = γi =

½

0 , 0 −½ i = 1, 2, 3, (1.91)

0 σi , −σi 0

where each element of the 4 × 4 matrices represents a 2 × 2 matrix and the σi correspond to the three Pauli matrices. This particular choice of the Dirac matrices is commonly known as the Pauli-Dirac representation. There are, of course, other representations for the γ µ matrices. However, the physics of Dirac equation is independent of any particular representation for the γ µ matrices. This can be easily seen by invoking Pauli’s fundamental theorem which says that if there are two sets of matrices γ µ and γ ′µ satisfying the Cliﬀord algebra, then, they must be related by a similarity transformation. Namely, if γµ, γν γ ′µ , γ ′ν = 2η µν ½, = 2η µν ½, (1.92)

+

+

then, there exists a constant, nonsingular matrix S such that (in fact, the similarity transformation is really a unitary transformation if we take the Hermiticity properties of the γ-matrices) γ ′µ = Sγ µ S −1 . Therefore, given the equation (γ ′µ pµ − m)ψ ′ = 0, we obtain (1.94) (1.93)

24

1 Relativistic equations

(Sγ µ S −1 pµ − m)ψ ′ = 0, or, or, or, (γ µ pµ − m)S −1 ψ ′ = 0, S(γ µ pµ − m)S −1 ψ ′ = 0,

(γ µ pµ − m)ψ = 0,

(1.95)

with ψ = S −1 ψ ′ . (The matrix S −1 can be moved past the momentum operator since it is assumed to be constant.) This shows that different representations of the γ µ matrices are equivalent and merely correspond to a change in the basis of the wave function. As we know, a change of basis does not change physics. To obtain the Hamiltonian for the Dirac equation, let us go to the coordinate representation where the Dirac equation (1.80) takes the form (remember = 1) (i∂ − m)ψ = (iγ µ ∂µ − m)ψ = 0, / (1.96)

or,

(iγ 0 ∂0 + iγ · ∇ − m)ψ = 0.

Multiplying with γ 0 from the left and using the fact that (γ 0 )2 = ½, we obtain ∂ψ = (−iγ 0 γ · ∇ + mγ 0 )ψ. ∂t

i

(1.97)

Conventionally, one denotes β = γ 0, α = γ 0 γ. (1.98)

In terms of these matrices, then, we can write (1.97) as ∂ψ = (−iα · ∇ + mβ)ψ = (α · p + βm)ψ. ∂t

i

(1.99)

This is a ﬁrst order equation (in time derivative) like the Schr¨dinger o equation and we can identify the Hamiltonian for the Dirac equation with

1.4 Dirac equation

25

H = α · p + βm.

(1.100)

In the particular representation of the γ µ matrices in (1.91), we note that

β = γ0 = α = γ0γ =

½

0 , 0 −½

½

0 0 −½

0 σ −σ 0

=

0 σ . σ 0

(1.101)

We can now determine either from the deﬁnition in (1.98) and (1.79) or from the explicit representation in (1.101) that the matrices α, β satisfy the anti-commutation relations = 2δij ½, = 0, (1.102)

αi , αj αi , β

+ +

with β 2 = ½. We can, of course, directly check from this explicit representation that both β and α are Hermitian matrices. But, independently, we also note from the form of the Hamiltonian in (1.100) that, in order for it to be Hermitian, we must have

β † = β, α† = α. In terms of the γ µ matrices, this translates to (1.103)

β = γ 0 = (γ 0 )† = β † , α = γ 0 γ = (γ 0 γ)† = α† . Equivalently, we can write (1.104)

26

1 Relativistic equations

(γ 0 )† = γ 0 , (γ i )† = −γ i . (1.105)

Namely, independent of the representation, the γ µ matrices must satisfy the Hermiticity properties in (1.105). (With a little bit of more analysis, it can be seen that, in general, the Hermiticity properties of the γ µ matrices are related to the choice of the metric tensor and this particular choice is associated with the Bjorken-Drell metric.) In the next chapter, we would study the plane wave solutions of the ﬁrst order Dirac equation. 1.5 References The material presented in this chapter is covered in many standard textbooks and we list below only a few of them. 1. L. I. Schiﬀ, Quantum Mechanics, McGraw-Hill, New York, 1968. 2. J. D. Bjorken and S. Drell, Relativistic Quantum Mechanics, McGraw-Hill, New York, 1964. 3. C. Itzykson and J-B. Zuber, Quantum Field Theory, McGrawHill, New York, 1980. 4. A. Das, Lectures on Quantum Mechanics, Hindustan Publishing, New Delhi, India, 2003.

Chapter 2

Solutions of the Dirac equation

2.1 Plane wave solutions The Dirac equation in the momentum representation (1.80) (p − m)ψ = (γ µ pµ − m)ψ = 0, / or in the coordinate representation (i∂ − m)ψ = (iγ µ ∂µ − m)ψ = 0, / (2.2) γµ, (2.1)

deﬁnes a set of matrix equations. Since the Dirac matrices, are 4 × 4 matrices, the wave functions, in this case, are four component column matrices (column vectors). From the study of angular momentum, we know that multicomponent wave functions suggest a nontrivial spin angular momentum. (Other nontrivial internal symmetries can also lead to a multicomponent wavefunction, but here we are considering a simple system without any nontrivial internal symmetry.) Therefore, we expect the solutions of the Dirac equation to describe particles with spin. To understand what kind of particles are described by the Dirac equation, let us look at the plane wave solutions of the equation (which are supposed to describe free particles). Let us denote the four component wave function as (x stands for both space and time) ψ1 (x) ψ2 (x) , ψ(x) = ψ3 (x) ψ4 (x) 27

(2.3)

28 with ψα (x) = e−ip·x uα (p),

2 Solutions of the Dirac equation

α = 1, 2, 3, 4.

(2.4)

Substituting this back into the Dirac equation, we obtain (we deﬁne A = γ µ Aµ for any four vector Aµ ) / (iγ µ ∂µ − m)ψ(x) = 0, or, or, (p − m)u(p) = 0, /

(iγ µ (−ipµ ) − m)u(p) = 0, (2.5)

where the four component function, u(p), has the form u1 (p) u2 (p) . u(p) = u3 (p) u4 (p)

(2.6)

Let us simplify the calculation by restricting to motion along the z-axis. In other words, let us set p1 = p2 = 0. In this case, the equation takes the form (γ 0 p0 + γ 3 p3 − m)u(p) = 0. (2.8) (2.7)

Taking the particular representation of the γ µ matrices in (1.91), we can write this explicitly as p0 − m 0 p0 − m 0 p3 p3 0 0 u1 (p)

0 −p 3 0

u2 (p) u (p) = 0.(2.9) −(p0 + m) 0 3 0 −(p0 + m) u4 (p) −p3

2.1 Plane wave solutions

29

This is a set of four linear homogeneous equations and a nontrivial solution exists only if the determinant of the coeﬃcient matrix vanishes. Thus, requiring,

p0 − m 0 p3 0 0 p0 − m 0 −p3 = 0, det −p3 0 −(p0 + m) 0 0 p3 0 −(p0 + m) we obtain,

(2.10)

(p0 − m) (p0 − m)(p0 + m)2 − p2 (p0 + m) 3 or, or, or, (p2 − m2 )2 − 2p2 (p2 − m2 ) + p4 = 0, 0 3 0 3 p2 − p2 − m2 = 0. 0 3

+p3 p3 + (p0 − m)(−p3 (p0 + m)) = 0, 3

(p2 − p2 − m2 )2 = 0, 0 3

(2.11)

Thus, we see that a nontrivial plane wave solution of the Dirac equation exists only for the energy values

p0 = ± E = ±

p 2 + m2 . 3

(2.12)

Furthermore, each of these energy values is doubly degenerate. Of course, we would expect the positive and the negative energy roots in (2.12) from Einstein’s relation. However, the double degeneracy seems to be a reﬂection of the nontrivial spin structure of the wave function as we will see shortly. The energy eigenvalues can also be obtained in a simpler fashion by noting that

30

2 Solutions of the Dirac equation

det(γ µ pµ − m) = 0, or, or, or, or, det (p0 − m)½ −σ3 p3 σ3 p 3 −(p0 + m)½ = 0,

det − (p2 − p2 − m2 )½ = 0, 0 3 p2 − p2 − m2 = 0. 0 3

det − (p2 − m2 )½ + p2 ½ = 0, 0 3 (2.13)

This is identical to (2.11) and the energy eigenvalues would then correspond to the roots of this equation given in (2.12). (Note that this method of evaluating a determinant is not valid for matrices involving submatrices that do not commute. In the present case, however, the submatrices ½, σ3 are both diagonal and, therefore, commute which is why this simpler method works out.) We can obtain the solutions of the Dirac equation by directly solving the set of four coupled equations in (2.9). Alternatively, we can introduce two component wave functions u(p) and v (p) and write ˜ ˜ u(p) ˜ v (p) ˜

u(p) = where

,

(2.14)

u(p) = ˜

u1 (p) u2 (p)

,

v (p) = ˜

u3 (p) u4 (p)

.

(2.15)

We note that for the positive energy solutions p0 = E+ = E = p 2 + m2 , 3 (2.16)

the set of coupled equations takes the form γ µ pµ − m u(p) = 0, or, (E+ − m)½ −σ3 p3 σ3 p 3 −(E+ + m)½ u(p) ˜ v (p) ˜ = 0. (2.17)

2.1 Plane wave solutions

31

Writing out explicitly, (2.17) leads to

(E+ − m)˜(p) + σ3 p3 v (p) = 0, u ˜ σ3 p3 u(p) + (E+ + m)˜(p) = 0. ˜ v (2.18)

The two component function v (p) can be solved in terms of u(p) and ˜ ˜ we obtain from the second relation in (2.18) σ3 p 3 u(p). ˜ E+ + m

v (p) = − ˜

(2.19)

Let us note here parenthetically that the ﬁrst relation in (2.18) also leads to the same relation (as it should), namely, (E+ − m) σ3 u(p) ˜ p3 (E+ − m)(E+ + m) σ3 u(p) ˜ p3 (E+ + m)

2 (E+ − m2 ) σ3 u(p) ˜ p3 (E+ + m)

v (p) = − ˜ = − = − = −

σ3 p 3 p2 3 σ3 u(p) = − ˜ u(p), ˜ p3 (E+ + m) E+ + m

(2.20)

where we have used the property of the Pauli matrices, namely, 2 σ3 = ½. Note also that if the relation (2.19) obtained from the second equation in (2.18) is substituted into the ﬁrst relation, it will hold identically. Therefore, the positive energy solution is completely given by the relation (2.19). Choosing the two independent solutions for u as ˜ 1 0 0 1

u(p) = ˜

,

u(p) = ˜

,

(2.21)

we obtain respectively

3 (2.9) becomes (E− − m)˜(p) + σ3 p3 v (p) = 0. (2.22) v (p) = − ˜ σ3 p 3 E+ + m 0 1 = 0 p3 E+ + m .25) (2.27) .24) and the set of equations (2. from the second of the two two-component Dirac equations in (2.26) Choosing the independent solutions as 1 0 0 1 v (p) = ˜ . (The question of which components can be chosen independently follows from an examination of the dynamical equations. for example. u ˜ σ3 p3 u(p) + (E− + m)˜(p) = 0.) ˜ Similarly. ˜ E− − m (2.23) This determines the two positive energy solutions of the Dirac equation (remember that the energy eigenvalues are doubly degenerate). ˜ v We can solve these as u(p) = − ˜ σ3 p 3 v (p). Thus. we note that v must vanish in the rest frame while u remains ˜ ˜ arbitrary. u can be thought of as the independent solution.32 2 Solutions of the Dirac equation σ3 p 3 v (p) = − ˜ E+ + m and 1 0 = p3 −E + m + 0 . v (p) = ˜ . (2. Thus.18). for the negative energy solutions we write p0 = E− = −E = − p 2 + m2 . (2.

0 1 (2. (2. (2. (2.1 Plane wave solutions 33 we obtain respectively p3 −E − m − 0 σ3 p 3 u(p) = − ˜ E− − m and 1 0 = .21) and (2. from the fact that we can write σ p ˜ − E 3−3m v (p) − . 1 0 0 p3 ↓ u− = E− − m . − E+ + m 0 u↓ + = .2.30) u+ (p) = u(p) ˜ .31) p3 −E − m − 0 ↑ u− (p) = . σ3 p 3 ˜ − E + m u(p) + u− (p) = the positive and the negative energy solutions have the explicit forms 1 0 1 0 p3 E+ + m u↑ (p) + 0 = p3 .27) are reminiscent of the spin up and spin down states of a two component spinor.29) and these determine the two negative energy solutions of the Dirac equation.28) u(p) = − ˜ σ3 p 3 E− − m 0 1 = 0 p3 E− − m . The independent two component wave functions in (2. v (p) ˜ (2. Thus.32) .

(The change in the sign in the dependent components comes from raising the index of the momentum. pi = −pi . the solutions take the forms (with p0 = E± = ± p2 + m2 ) u+ (p) = u(p) ˜ .) 2.) It is because of the presence of negative energy solutions that the wave function becomes a four component column matrix as opposed to the two component spinor we expect in non-relativistic systems. where p1 = p2 = 0.34) v (p) ˜ (2.35) then. namely.36) Here α and β are normalization constants to be determined.27) respectively are ˜ ˜ normalized as (for the same spin components) . we can write the solutions for motion along a general direction as u(p) ˜ .33) which can be explicitly veriﬁed. unlike the nonrelativistic counting (2s + 1). u− (p) = σ·p u(p) ˜ E+ + m σ·p ˜ E− − m v (p) . for the case of general motion. relativistic particle of spin s in the presence of both positive and negative energies follows to be 2(2s + 1).2 Normalization of the wave function Let us note that if we deﬁne E = E+ = p2 + m2 = −E− .) From the structure of the wave function. (2. (We will determine the spin of 2 the Dirac particle shortly. it is also clear that. (The correct counting for the number of components of the wave function for a massive. u− (p) = β σ·p ˜ E + m u(p) σ· − E + p v (p) m ˜ v (p) ˜ u+ (p) = α .21) and (2. (2. (2. The two component spinors u(p) and v (p) in (2.34 2 Solutions of the Dirac equation The notation is suggestive and implies that the wave function corresponds to that of a spin 1 particle.

˜ v E +m .2 Normalization of the wave function 35 u† (p)˜(p) = 1 = v † (p)˜(p). Similarly. this product vanishes.38) where we have used the familiar identity satisﬁed by the Pauli matrices (σ · A)(σ · B) = A · B + iσ · (A × B). Given this.2. ˜ u E+m (2.40) p2 +1 (E + m)2 2E |β|2 v † (p)˜(p).39) u† (p)u− (p) − v † (p) ˜ σ·p − E + m v (p) ˜ v (p) ˜ = |β|2 v † (p) ˜ = |β|2 = σ·p σ·p v (p) + v † (p)˜(p) ˜ ˜ v (E + m)2 v † (p)˜(p) ˜ v (2. for the negative energy solutions we have σ·p v = β β −˜† (p) E + m ∗ (2. ˜ u ˜ v For diﬀerent spin components. we can now calculate σ·p ˜ ˜ u† (p)u+ (p) = α∗ α u† (p) u† (p) E + m + = |α|2 u† (p)˜(p) + u† (p) ˜ u ˜ = |α|2 1 + = |α|2 = |α|2 = p2 (E + m)2 (2.37) u(p) ˜ σ·p ˜ E + m u(p) σ·p σ·p u(p) ˜ (E + m)2 u† (p)˜(p) ˜ u u† (p)˜(p) ˜ u E 2 + m2 + 2Em + p2 (E + m)2 2E(E + m) † u (p)˜(p) ˜ u (E + m)2 2E |α|2 u† (p)˜(p).

36 2 Solutions of the Dirac equation It is worth remarking here that although we have seen in (2.38) and (2. we see that the diﬀerence between the hermitian conjugate u† and u is in the relative sign in the second of the two-component spinors. another wave function (known as the adjoint spinor) that plays an important role is deﬁned to be u(p) = u† (p)γ 0 .40) simply because we have not speciﬁed their spin components.42) −˜† (p) .41) σ·p ˜ ˜ u+ (p) = α∗ u† (p) u† (p) E + m ½ 0 0 −½ σ·p ˜ u = α∗ u† (p) −˜† (p) E + m .37) that. for example. We can also calculate the product σ·p ˜ u u+ (p)u+ (p) = α∗ α u† (p) −˜† (p) E + m = |α|2 u† (p)˜(p) − u† (p) ˜ u ˜ = |α|2 1 − = |α|2 u(p) ˜ σ·p ˜ E + m u(p) σ·p σ·k u(p) ˜ (E + m)2 p2 u† (p)˜(p) ˜ u (E + m)2 E 2 + m2 + 2Em − p2 † u (p)˜(p) ˜ u (E + m)2 . v Thus. In dealing with the Dirac equation. u† u = 1 = v † v . we have carried along ˜ ˜ ˜ ˜ these factors in (2. Thus. for the same spin components. σ·p v u− (p) = β ∗ −˜† (p) E + m σ·p v = β ∗ −˜† (p) E + m v † (p) ˜ ½ 0 0 −½ (2. (2.

we can show that 2m |β|2 v † (p)˜(p). m † d3 x ψp (x)ψp′ (x) = (2.43) 2m |α|2 u† (p)˜(p). u† u.44) Our naive instinct will be to normalize the wave function. the motivation for such a normalization condition comes from the fact that. as we will see. In fact.2 Normalization of the wave function 37 = |α|2 = 2m(E + m) † u (p)˜(p) ˜ u (E + m)2 (2. this is not a relativistic normalization. would give (for the same spin components) E (2π)3 δ3 (p − p′ ). Thus. as we will see shortly.38) and (2. we determine from (2. in the rest frame of the particle. must be positive. a relativistically covariant normalization would be to require (for the same spin components) E = u† (p)u− (p).40) (for the same spin components when (2.46) (Remember that this will correspond to the probability density and. is related to the probability density which transforms like the time component of a four vector. − m u† (p)u+ (p) = + (2. therefore.45) However. ˜ v E+m u− (p)u− (p) = − (2. ψp (x) = e−ik·x u(p) with p0 = ±E.47) With the requirement (2.46). ˜ u E +m Similarly.2. this will reduce to u† u± = 1 which corresponds to ± the non-relativistic normalization. with this normalization condition. as in the non-relativistic case. By the way. by requiring (for the same spin components) u† (p)u+ (p) = 1 = u† (p)u− (p).) The independent wave functions for a free particle.37) holds) . + − (2.

with this normalization. appears to be a Lorentz invariant (scalar) and we will see later that this is indeed true. α = α∗ = 2E E |α|2 = . . therefore.44) (for the same spin components) u+ (p)u+ (p) = 2m 2m E + m |α|2 = = 1. E+m E + m 2m (2. For example.49) It is also clear that. u† (p)u− (p) = − E+m m β = β∗ = E+m . E+m E + m 2m 2m 2m E + m |β|2 = − = −1. with this normalization.50) u− (p)u− (p) = − This particular product. v (p) ˜ (2. σ·p ˜ E + m u(p) σ·p − E + m v (p) ˜ . we can write the normalized positive and negative energy solutions of the Dirac equation to be u+ (p) = E+m 2m E+m 2m u− (p) = u(p) ˜ . Therefore. E+m m E+m . 2m (2.48) or.43) and (2. 2m E 2E |β|2 = . we will obtain from (2.38 2 Solutions of the Dirac equation u† (p)u+ (p) = + or. Let us also note here that by construction the positive and the negative energy solutions are orthogonal.

E+m E+m .38) and (2. =α β ∗ u† (p) ˜ u† (p) ˜ σ·p E+m σ·p − E + m v (p) ˜ v (p) ˜ − u† (p) ˜ σ·p σ·p v (p) + u† (p) ˜ ˜ v (p) ˜ E+m E+m (2.40) (for the same spin components) u† (p)u+ (p) = + or. 2 (2.53) This still behaves like the time component of a four vector (m is a Lorentz scalar). In this case. we will obtain from (2. Correspondingly.2. (There is no rest frame for a massless particle.51) Therefore. ˜ E+m u+ (p)u− (p) = −2α∗ β u† (p) ˜ (2. that σ·p v (p) = 0. ˜ E+m σ·p u− (p)u+ (p) = −2β ∗ α v † (p) ˜ u(p) = 0.52) While we will be using this particular normalization for massive particles.54) or. orthonormal solutions of the Dirac equation. − + (2. Note. however. u† (p)u− (p) = − E+m β = β∗ = E+m . α = α∗ = 2E |α|2 = E. 2 2E |β|2 = E.) The probability density has to be well deﬁned. the solutions we have constructed correspond to four linearly independent. an alternative normalization which works well for both massive and massless particles is given by u† (p)u+ (p) = E = u† (p)u− (p).2 Normalization of the wave function 39 u† (p)u− (p) + = α∗ β = 0. . let us note that it becomes meaningless for massless particles.

d3 x ψ † (x)ψ(x) = 1. we obtain 2m |α|2 = m. Let us note once again that this is a particularly convenient normalization for massless particles. The true normalization is really contained in the total probability.56) where a(p) is a coeﬃcient which depends on the normalization of u(p) in such a way that the wave function would lead to a total probability normalized to unity. Let us deﬁne a four dimensional generalization of the Pauli matrices as (in this section. while the arbitrariness in the normalization of u(p) may seem strange. u− (p)u− (p) = − E+m u+ (p)u+ (p) = (2.57) Namely. (2. it can be understood in light of what we have already pointed out earlier as follows. the structure of the plane wave solutions of the Dirac equation is suggestive of the fact that the particle described by the Dirac equation has spin 1 .55) which vanishes for a massless particle. ψ(x) = (2. in this case. we will use the notations of three dimensional . That this is indeed true 2 can be seen explicitly as follows.3 Spin of the Dirac particle As we have argued repeatedly. 2.40 2 Solutions of the Dirac equation Correspondingly. Let us note here parenthetically that. E +m 2m |β|2 = −m. a particular choice of normalization for the u(p) is compensated for by a speciﬁc choice of the coeﬃcient function a(p) so that the total probability integrates to unity. We can write the solution of the Dirac equation for a general motion (along an arbitrary direction) as d4 p a(p)δ(p2 − m2 ) e−ip·x u(p). This product continues to be a scalar.

σj ] 0 2iǫijk σk = 2iǫijk αk . 3. be checked readily that this is related to the αi matrices deﬁned in (1.98) and (1. α ˜ (2. i = 1. σj ] 0 2iǫijk σk 0 0 [σi . αi .) Furthermore. that this deﬁnes a reducible representation of spin generators since the matrices are block diagonal. ˜ (2. of course.59) ρ= 0 ½ ½ 0 .62) αi . αj ˜ ˜ = = This shows that 1 αi satisﬁes the angular momentum algebra (remem2˜ ber = 1) and this is why we call the matrices. (2. however. let us note that . 2.2. the generalized ˜ Pauli matrices. (2.101) through the relation 0 σi σi 0 αi = where = ρ˜ i = αi ρ. (Note.59) and write αi = ραi = αi ρ.61) From the structures of the matrices αi and αi we conclude that ˜ [σi .3 Spin of the Dirac particle 41 Euclidean space since we will be dealing only with three dimensional vectors) σi 0 0 σi αi = ˜ .60) We note that ρ2 = ½ so that we can invert the deﬁning relation (2.58) It can. ˜ (2.

2. i. k = 1.63) With these relations at our disposal. 3. we obtain . ½] = 2iǫijk αk . β ˜ = = 0.100) (remember that we are using three dimensional Euclidean notations in this section) H = α · p + βm = αi pi + βm. αj ˜ = σi 0 0 σi 0 0 σj σj 0 − 0 σj σj 0 σi 0 0 σi = σi σj − σj σi 0 [σi . σj ] = 2iǫijk σk 0 −[σi . the Dirac equation transforms covariantly under a Lorentz transformation.42 2 Solutions of the Dirac equation αi . j.64) As we will see in the next chapter. therefore. αi . (2. (2. rotations which correspond to a subset of the Lorentz transformations must also be a symmetry of the Dirac Hamiltonian. (2. σj ] 0 2iǫijk σk [σi . ½] 0 0 0 σi σj − σj σi 0 = [σi . the angular momentum operators which generate rotations should commute with the Dirac Hamiltonian. Let us recall that the orbital angular momentum operator is given by Li = ǫijk xj pk .65) Calculating the commutator of this operator with the Dirac Hamiltonian. let us look at the free Dirac Hamiltonian in (1. Consequently. Lorentz transformations deﬁne a symmetry of the Dirac Hamiltonian and. In other words.

66) we obtain 1 αi .66) Here we have used the fact that since β is a constant matrix and m is a constant. the total angular momentum which should commute with the Hamiltonian must contain a spin part as well. H ˜ 2 Li + = [Li . H ˜ 2 1 αi . Consequently. the total angular momentum which should commute with the Hamiltonian. H] = [ǫijk xj pk .69) In this case. pℓ ]pk = ǫijk αℓ (iδjℓ ) pk = iǫijk αj pk . the second term in the Hamiltonian drops out of the commutator. we note that the orbital angular momentum operator does not commute with the Dirac Hamiltonian.3 Spin of the Dirac particle 43 [Li . Thus. so that combining this relation with (2. therefore.2. To determine the spin angular momentum. (2.68) In other words. β m ˜ (2. αj pj + αi . H ˜ = = αi . αℓ pℓ ] = ǫijk αℓ [xj . we can identify the spin angular momentum operator with . we note that αi . αj pj + βm ˜ ˜ αi . (2. αℓ pℓ + βm] = [ǫijk xj pk . can be identiﬁed with 1 αi . ˜ 2 Ji = Li + (2. if rotations are a symmetry of the system. H] + = iǫijk αj pk − iǫijk αj pk = 0.67) = 2iǫijk αk pj = −2iǫijk αj pk .

in particular. i ∂ (ψ † ψ) = −i∇ · ψ † αψ . ∂t −i (2.74) This is the continuity equation for the probability current density associated with the Dirac equation and we note that we can identify .70) Note.4 Continuity equation The Dirac equation. ∂t ∂ (ψ † ψ) = −∇ · ψ † αψ .72) with ψ † on the left and (2. Therefore. we obtain ∂ψ † ∂ψ +i ψ = −i ψ † α · ∇ψ + (∇ψ † ) · αψ .73) with ψ on the right and subtracting the second from the ﬁrst. (2.44 2 Solutions of the Dirac equation Si = 1 αi . Multiplying (2.71) which has doubly degenerate eigenvalues ± 1 .73) where the gradient is assumed to act on ψ † . 2 2. we obtain ← − ∂ψ † = ψ † iα · ∇ + βm). ∂t (2.72) Taking the Hermitian conjugate of this equation. is given by ∂ψ = Hψ = − iα · ∇ + βm ψ. ∂t ∂t iψ † or. or. ∂t i (2. we conclude 2 that the particle described by the Dirac equation corresponds to a spin 1 particle. written in the Hamiltonian form. ˜ 2 (2. that 1 1 α3 = ˜ 2 2 σ3 0 0 σ3 S3 = .

76) so that the continuity equation can be written in the manifestly covariant form ∂µ J µ = 0.4 Continuity equation 45 ρ = ψ † ψ = probability density.77)) and. J) = (ψ † ψ. ∂t This suggests that we can write the current four vector as J µ = (ρ. J = ψ † αψ = probability current density. (We are.2. therefore. P = (2.79) is a constant independent of any particular Lorentz frame.77) (2. is the time component of j µ (see (2. It is worth recalling that we have already used this Lorentz transformation property of ρ in deﬁning the normalization of the wave function.) On the other hand. the total probability d3 x ρ = d3 x ψ † ψ. yet to show that j µ transforms like a four vector which we will do in the next chapter. (2. ψ † αψ). of course. in fact. (2. shows that the probability density. must transform like the time coordinate under a Lorentz transformation.78) This.80) . to write the continuity equation as ∂ρ = −∇ · J. ρ.75) (2. and note that the Hermitian conjugate of ψ satisﬁes (2. An alternative and more covariant way of deriving the continuity equation is to start with the covariant Dirac equation iγ µ ∂µ − m ψ = 0.

84) This is.82) with ψ on the right and subtracting the second from the ﬁrst.82) where we have used the property of the gamma matrices that (for µ = 0.81) Multiplying this equation with γ 0 on the right and using the fact that (γ 0 )2 = ½. (2. 1. (2.77). in fact. 3) γ 0 γ µ γ 0 = (γ µ )† . or. ← − ψ − iγ µ ∂µ − m = 0.46 2 Solutions of the Dirac equation ← − ψ † − i(γ µ )† ∂µ − m = 0. (2. (2.86) (2. Note from the deﬁnition in (2. (2.80) with ψ on the left and (2. i∂µ ψγ µ ψ = 0. we obtain ← − i ψγ µ ∂µ ψ + ψγ µ ∂µ ψ = 0. we obtain ← − ψ − iγ 0 (γ µ )† γ 0 ∂µ − m) = 0. which is what we had derived earlier in (2. or. γ 0 (γ µ )† γ 0 = γ µ . 2. or. ∂µ ψγ µ ψ = 0.85) . J = ψγψ = ψ † γ 0 γψ = ψ † αψ.85) that J 0 = ψγ 0 ψ = ψ † γ 0 γ 0 ψ = ψ † ψ = ρ. the covariant continuity equation and we can identify J µ = ψγ µ ψ.83) Multiplying (2.

1 (as well as from the equation above) that the positive and the negative energy solutions are separated by a gap of magnitude 2m (remember that we are using c = 1). First of all. 2. for example.2. Even when the probability density is consistently deﬁned. 2. 2.40).38) and (2. Consequently. In the free particle case. the presence of negative energy solutions leads to many conceptual diﬃculties.87) Thus.1. the probability density is independent of time derivative much like the Schr¨dinger equation. m −m Figure 2. (2.5 Dirac’s hole theory 47 Let me conclude this discussion by noting that although the Dirac equation has both positive and negative energy solutions. in such a case. we note that the energy spectrum . because it is a ﬁrst order equation (particularly in the time derivative). This is rather diﬀerent from the case of the Klein-Gordon equation that we have studied in chapter 1. even for this simple case of a free particle the energy spectrum has the form shown in Fig. the energy eigenvalues are given by p0 = E± = ±E = ± p2 + m2 . as we o have seen explicitly in (2. can be deﬁned to be positive deﬁnite even in the presence of negative energy solutions.5 Dirac’s hole theory We have seen that Dirac’s equation leads to both positive and negative energy solutions.1: Energy spectrum for a free Dirac particle. We note from Fig. the probability density.

Dirac postulated that the ground state in such a theory is the state where all the negative energy states are ﬁlled with electrons. a positive energy electron can no longer drop down to a negative energy state without violating the Pauli exclusion principle since the negative energy states are already ﬁlled. Dirac assumed that the electrons in the negative energy states are passive in the sense that they do not produce any observable eﬀect such as charge. therefore.) This redeﬁnition of the vacuum automatically prevents the instability associated with matter. (Momentum and energy of these electrons are also assumed to be unobservable. there is a way out of 1 this diﬃculty. it does predict some new physical phenomena which are experimentally observed. quantum mechanically this is not acceptable. in fact. Furthermore. For exam- . In the case of Dirac particles. we can restrict ourselves to the subspace of positive energy solutions. of course. It is only because fermions obey Pauli exclusion principle that this works for the Dirac equation. this implies that any such physical system (of Dirac particles) would make a transition to these unphysical energy states thereby leading to a collapse of all stable systems such as the Hydrogen atom. To be speciﬁc. Thus. unlike the conventional picture of the ground state as being the state without any particle (quantum). even if we start out with a positive energy solution. contains an inﬁnite number of negative energy particles. however. For example. (Note that this would not work for a bosonic system such as particles described by the Klein-Gordon equation. Classically. Namely. let us assume that the particles described by the Dirac equation are the spin 1 electrons. Taking advantage of this fact.) On the other hand. any perturbation would cause the energy to lower. fermions. Namely. This simply means that one redeﬁnes the values of all these observables with respect to this ground state. electromagnetic ﬁeld etc. Since physical systems have a tendency to go to the lowest energy state available. here the ground state. any given energy state can at any time accommodate at the most two electrons with opposite spin projections. Let us recall that the Dirac particles carry spin 2 and are. destabilizing the physical system and leading to an ultimate collapse. Dirac postulated that the physical ground state (vacuum) in such a theory should be redeﬁned for consistency. 2 Since fermions obey Pauli exclusion principle. But as we have argued earlier within the context of the Klein-Gordon equation.48 2 Solutions of the Dirac equation is unbounded from below.

2. This “hole” state is what we have come to recognize as the anti-particle – in this case. the absence of a negative energy electron can be thought of as a “hole” which would have exactly the same mass as the particle but otherwise opposite internal quantum numbers. i = 1. (γ i )† = −γ i . µ = 0.) In spite of this. thereby. Thus. In this section. 3. The amount of energy necessary to excite a negative energy electron to a positive energy state is E ≥ 2m. (This unconventional deﬁnition of a vacuum state can be avoided in a second quantized ﬁeld theory which we will study later. a positron – and the process under discussion is commonly referred to as pair creation (production). in deﬁning a ﬁrst order equation. 1. 2.6 Properties of the Dirac matrices 49 ple. the Dirac equation passes as a one particle equation primarily because of the Pauli exclusion principle.) This is Dirac’s theory of electrons and works quite well. 2. However. if enough energy is provided to such a ground state. 2. the four Dirac matrices satisfy (in addition to the Cliﬀord algebra) (γ 0 )† = γ 0 . were crucial in taking a matrix square root of the Einstein relation and. . this is a general feature that combining quantum mechanics with relativity necessarily leads to a many particle theory. On the other hand. Furthermore. 3. we will study some of the useful properties of these matrices. a negative energy electron can make a transition to a positive energy state and can appear as a positive energy electron. γ µ . we must recognize that it is inherently a many particle theory in the sense that the vacuum (ground state) of the theory is deﬁned to contain inﬁnitely many negative energy particles.88) Tr γ µ = 0.6 Properties of the Dirac matrices The Dirac matrices. (2. As we have seen. (The absence of a negative energy electron in the ground state can be thought of as the ground state plus a positive energy “hole” state with exactly opposite quantum numbers to neutralize its eﬀects. the Dirac theory predicts an anti-particle of equal mass for every Dirac particle.

91) that we can identify this with the matrix ρ deﬁned earlier in (2. † γ5 = γ5 . we obtain γ5 = i ½ 0 σ1 0 0 −σ1 σ1 0 0 0 −σ2 −σ2 σ3 σ2 0 0 −σ3 σ3 0 0 −½ σ1 0 0 −σ1 σ2 σ3 0 −i½ −i½ 0 = i −σ2 σ3 0 = i −σ1 σ2 σ3 0 0 = i = ½ ½ 0 . To obtain the other basis matrices.93) . Of course.92) We recognize from (2. let us deﬁne the following sets of matrices. (2.91) where we have used the property of the Pauli matrices σ1 σ2 σ3 = i½. 4! (2.50 2 Solutions of the Dirac equation Since these are 4 × 4 matrices.89) represents the four-dimensional generalization of the Levi-Civita tensor.91). (2. a complete set of Dirac matrices must consist of 16 such matrices. Note that in our particular representation for the γ µ matrices given in (1.60).90) i ǫµνλρ γ µ γ ν γ λ γ ρ . Let γ5 = iγ 0 γ 1 γ 2 γ 3 = − where ǫ0123 = 1 = −ǫ0123 . Note that. by deﬁnition. (2. the identity matrix will correspond to one of them. 2 γ5 = ½. (2.

2. 2. = i η µν − γ ν γ µ whose explicit forms in our representation can be worked out to be (i.6 Properties of the Dirac matrices 51 and that. σ µν . 3) . ½ 0 = σi . ν = 0. (2. it anticommutes with any one of them. we can deﬁne four new matrices as γ5 γ µ . j.97) = −i η µν − γ µ γ ν .95) Since we know the explicit forms of the matrices ½. µ = 0. (2. 3. Namely. since it is the product of all the four γ µ matrices. k = 1. we can also deﬁne six anti-symmetric matrices.2. as (µ. 1. γ µ + = 0. γ5 . (2.94) Given the matrix γ5 . γ µ and γ5 in our representation.96) Finally. 1. 3) σ µν = −σ νµ = i i γ µ . let us write out the forms of γ5 γ µ also in this representation. 2. γ5 γ 0 = γ5 γ i = 0 ½ ½ ½ 0 −σi 0 ½ 0 0 0 −½ σi 0 = ½ 0 −½ 0 −σi 0 0 . γ ν = (γ µ γ ν − γ ν γ µ ) 2 2 (2.

¯ Here. From (2. σ µν . the notation is suggestive and stands for the fact that ψΓ(S) ψ . that the matrices 1 1 αi = ǫijk σ jk . ǫijk ǫℓjk = 2 δiℓ . γ5 . γµ. 1. (2.98) using the identity for products of Levi-Civita tensors. γµ.99) can be identiﬁed with the spin operators for the Dirac particle.98) 1 We have already seen in (2. ˜ (2.70) that the matrices 2 αi represent the ˜ spin operators for the Dirac particle. Γ(S) Γ(V ) Γ(A) = = ½. (This relation can be obtained from (2.) We have thus constructed a set of sixteen Dirac matrices. namely. 6. provide a basis for all the 4 × 4 matrices.52 2 Solutions of the Dirac equation σ 0i = iγ 0 γ i = i 0 iσi iσi 0 ½ 0 0 −σi σi 0 0 −½ = iαi .98) we conclude. 0 −σi 0 −σi σj 0 −iǫijk σk σi 0 = σ ij = iγ i γ j = i −σi σj 0 0 −σj σj 0 = i = i −iǫijk σk 0 = ǫijk αk . 1. ˜ 2 4 (2. namely. 4. therefore.100) Γ(T ) = Γ(P ) = = γ5 where the numbers on the right denote the number of matrices in each category and these. 4. in fact.

ψΓ(V ) ψ. In fact.103) γ 0 σ µν γ 0 = The Dirac matrices satisfy nontrivial (anti) commutation relations. all other matrices satisfy γ 0 Γ(α) γ 0 = (Γ(α) )† . it can be checked that except for γ5 . ψΓ(A) ψ and ψΓ(P ) ψ behave respectively like a vector. (2.102) α = S.104) (2. Finally. axial-vector and a pseudo-scalar under a Lorentz and parity transformations as we will see in the next chapter. from γ 0 γ µ γ ν γ 0 = γ 0 γ µ γ 0 γ 0 γ ν γ 0 = (γ ν γ µ )† . Let us note here that each of the matrices.2. it follows that i 0 µ ν γ (γ γ − γ ν γ µ )γ 0 2 i (γ ν γ µ )† − (γ µ γ ν )† = 2 i = − (γ µ γ ν − γ ν γ µ )† 2 = (σ µν )† . γ 0 γ µ γ 0 = (γ µ )† . We already know that .6 Properties of the Dirac matrices 53 transforms like a scalar under Lorentz and parity transformations. V. ¯ ¯ ¯ ¯ Similarly. A. ψΓ(T ) ψ. it follows easily that γ 0 ½γ 0 = (γ 0 )2 = ½ = (½)† . γ 0 γ5 γ µ γ 0 = −γ5 γ 0 γ µ γ 0 = −(γ5 )† (γ µ )† = (γ5 γ µ )† . even within a given class. (2. (2. has its own hermiticity property. However.101) where we have used the fact that γ5 is Hermitian and it anti-commutes with γ µ . T. tensor. which is deﬁned to be Hermitian.

105) We can also calculate various other commutation relations in a straightforward and representation independent manner. −i η νλ − γ ν γ λ γλ − γν γµ. (2. σ νλ = γ µ .108) However. γ µ . For example. for the commutator of two σ µν matrices.106) [A. since γ µ matrices satisfy simple anti-commutation relations. we obtain . γλ = i γ µ. (2. we have used the fact that (2. B] C + B [A.107) can also be expressed in terms of commutators (instead of anti-commutators) as [A. Similarly. BC] = [A. In this derivation.107) is more useful for our purpose. γ µ + + = 2η µν ½.54 2 Solutions of the Dirac equation γµ.107) We note here parenthetically that the commutator in (2. γν γ5 . = 0. C] . BC] = ABC − BCA = (AB + BA)C − B(AC + CA) = A. the form in (2. (2. γ ν γ λ = i γµ. B + C − B A. C + . γν + + = 2i η µν γ λ − η µλ γ ν .

+ ν − γµγν (2. γ µ = 2η νµ γµ − 4γ = 2γ ν − 4γ ν = −2γ ν . Thus. σ λρ = − i η µν − γ µ γ ν . and it follows now that. σ λρ + i γ µ . σ λρ = iγ µ γ ν .112) . σ λρ = i γ µ γ ν . σ λρ γ ν = iγ µ 2iη νλ γ ρ − 2iη νρ γ λ + i 2iη µλ γ ρ − 2iη µρ γ λ γ ν = −2i η µλ i(η νρ − γ ρ γ ν ) + η νρ − i(η µλ − γ µ γ λ ) −η µρ i(η νλ − γ λ γ ν ) − η νλ (−i(η µρ − γ µ γ ρ ) = −2i η µλ σ νρ + η νρ σ µλ − η µρ σ νλ − η νλ σ µρ . (these relations are true only in 4-dimensions) γµ γ ν γ µ = γµ γ ν .110) (2. γµ A µ = γµ Aν γ ν γ µ = Aν γµ γ ν γ µ = −2Aν γ ν = −2A /γ /.6 Properties of the Dirac matrices 55 σ µν . The various commutation and anti-commutation relations also lead to many algebraic simpliﬁcations in dealing with such matrices. for example. we see that the σ µν matrices satisfy an algebra in the sense that the commutator of any two of them gives back a σ µν matrix. Similarly.111) (2. We will see in the next chapter that they provide a representation for the Lorentz algebra. (2.2. where we have used (γµ = ηµν γ ν ) γµ γ µ = 4 ½.109) Thus. This becomes particularly useful in calculating various amplitudes involving Dirac particles.

For example. γ ν + and so on.115) Even more complicated traces can be evaluated by using the basic relations we have developed so far. γ ν 2 2 1 = Tr 2η µν ½ = η µν Tr ½ = 4η µν . or. = (2. we note that Tr γ µ γ ν γ λ γ ρ = Tr γµ. = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ − Tr γ ν γ λ γ ρ γ µ . Therefore. (2.116) .56 2 Solutions of the Dirac equation γµ γ ν γ λ γ µ = γµ γ ν γ λ . The commutation and anticommutation relations also come in handy when we are evaluating traces of products of such matrices. 2 = Tr γ µ γ5 = −Tr γ5 γ µ = 0. 2 Tr γ µ γ ν γ λ γ ρ = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ .114) Tr γ µ γ ν + Tr γ5 γ µ (2. γν + − γν γµ γλγρ − γλγµ γρ + = Tr 2η µν γ λ γ ρ − γ ν γ µ γ λ γ ρ = 8η µν η λρ − Tr γ ν γ µ .113) = 2η λµ γµ γ ν + 2γ ν γ λ = 2γ λ γ ν + 2γ ν γ λ = 2 γ λ . it follows (in 4-dimensions) that 1 1 Tr γ µ γ ν + Tr γ ν γ µ = Tr γ µ . − γργµ Tr γ µ γ ν γ λ γ ρ = 4 η µν η λρ − η µλ η νρ + η νλ η µρ . γ λ + = 8η µν η λρ − 8η µλ η νρ + Tr γ ν γ λ γ µ . γ µ + − γµγλ = 4η λν ½. we know from the cyclicity of traces that Tr γ µ γ ν = Tr γ ν γ µ . γ ρ or. (2. For example.

σ µ = (½. 1 S = √ γ0 2 ½ + γ2 . (2.2. ˜ (2. (2.118) Similarly. 0 σµ ˜ σµ 0 γW = where µ .6 Properties of the Dirac matrices 57 and so on. 1 γM = . This is known as the Majorana representation and is quite useful in the study of Majorana fermions which are charge neutral fermions. To conclude this section. However. Explicitly. there exists a representation for the Dirac matrices where γ µ are all purely imaginary. σ). For example.117) It can be checked that the Dirac matrices in the Pauli-Dirac representation and the Majorana representation are related by the similarity (unitary) transformation µ γM = Sγ µ S −1 . It can be checked . namely. 2 γM = . 3 γM = . −σ). the µ γM matrices have the forms 0 σ2 0 σ2 σ2 0 −σ2 0 iσ3 0 −iσ1 0 0 iσ3 0 −iσ1 0 γM = .119) σ µ = (½. (2.120) This is known as the Weyl representation for the Dirac matrices and is quite useful in the study of massless fermions. there are other equivalent representations possible which may be more useful for a particular system under study. there exists yet another representation for the γ µ matrices. We would use all these properties in the next chapter to study the covariance of the Dirac equation under a Lorentz transformation. let us note that we have constructed a particular representation for the Dirac matrices commonly known as the Pauli-Dirac representation.

−γ5 γµ .1 Fierz rearrangement.124) Tr Γ(a) Γ(b) = δ(a) . σµν . it follows now that if . 4 (b) (2. γ5 γ µ . given this set of matrices. the sixteen Dirac matrices Γ(a) . a = S. A. P deﬁne a complete basis for 4 × 4 matrices. γµ .126) With this. We have explicitly constructed the sixteen matrices to correspond to the set Γ(a) = {½. V.122) From the properties of the γ µ matrices. As a result.122) is straightforward. Explicitly. the linear independence of the set of matrices in (2. γ5 } . 1 S=√ 2 ½ + γ5 γ 0 . σ µν . This is easily demonstrated by showing that they are linearly independent which is seen as follows. γ µ . we can construct the inverse set of matrices as Γ(a) . (2.100).6. T. γ5 } .121) 2.125) Γ(a) = (2. it can be easily checked that Tr Γ(a) Γ(b) = 0. Tr Γ(a) Γ(a) Γ(a) = such that a not summed.58 2 Solutions of the Dirac equation that the Weyl representation is related to the standard Pauli-Dirac representation through the similarity (unitary) transformation µ γW = Sγ µ S −1 . For example. As we have pointed out in (2. (2. (2.123) where “Tr” denotes trace over the matrix indices. (2. we can write the inverse set of matrices as 1 {½. a = b.

129). V. (a) (2. any arbitrary 4 × 4 matrix M can be expanded as a linear superposition of these matrices. multiplying (2. P . A.127) then. we obtain . (2. (M ) C(a) δ(b) . Since the set of matrices in (2.127) implies that all the coeﬃcients of expansion must vanishing which shows that the set of sixteen matrices Γ(a) in (2. or. (2. and taking trace over the matrix indices and using (2. C(a) δ(b) = 0.125) we obtain C(a) Γ(a) = 0. namely. (a) (a) or.122) provide a basis for the 4 × 4 matrix space. C(b) = 0.128) for any b = S. M= (a) C(a) Γ(a) .2. (2.122) are linearly independent. (a) C(a) Tr Γ(b) Γ(a) = 0.130) C(b) = Tr Γ(b) M . As a result they constitute a basis for 4 × 4 matrices. Substituting (2. (M ) (2.127) with Γ(b) . Therefore.129) Multiplying this expression with Γ(b) and taking trace over the matrix indices. (a) Tr Γ(b) or. where b is arbitrary.6 Properties of the Dirac matrices 59 C(a) Γ(a) = 0. we obtain C(a) Tr Γ(b) Γ(a) = (a) (a) (M ) (M ) (a) Tr Γ(b) M = or. T.130) into the expansion (2.

it is now straightforward to obtain . 3. then using (2. Equation (2. γ. we will use the standard convention that the repeated index (a) as well as the matrix indices are being summed) Γ(a) M = γβ Γ(a) N ¯ γβ αδ ¯ αδ Γ(a) Mββ Γ(a) ¯ Nδδ = Mββ Nδδ Γ(a) ¯ ¯ ¯ ¯ γβ Γ(a) ¯ αδ = Mββ Nδδ δγ δ δβα = Mαβ Nγδ . it can be used eﬀectively in many ways.132) Here α.131) Introducing the matrix indices explicitly. (2. η = 1.132) we can derive (for simplicity. 4 and correspond to the matrix indices of the 4 × 4 matrices and we are assuming that the repeated indices are being summed. ¯ ¯ ¯ ¯ M Γ(a) αδ N Γ(a) ¯ δδ γβ = Mαδ Γ(a) ¯ Nγ β Γ(a) ¯ ¯ ββ = Mαδ Nγ β Γ(a) ¯ ¯ ¯ δδ Γ(a) ¯ ββ = Mαδ Nγ β δδβ δδβ = Mαβ Nγδ . (a) Γ(a) γη Γ(a) αβ = δαη δβγ .60 2 Solutions of the Dirac equation M= (a) Tr Γ(a) M Γ(a) . or. For example.133). we note that if M and N denote two arbitrary 4 × 4 matrices. Just like any other completeness relation. this leads to Mαβ = (a) Γ(a) (a) γη Mηγ Γαβ .133) Using the relations in (2. β. 2.132) describes a fundamental relation which expresses the completeness relation for the sixteen basis matrices. ¯ ¯ ¯ ¯ (2. (2.

134) are known as the Fierz rearrangement identities which are very useful in calculating cross sections. if they correspond to anti-commuting fermion operators.135) Since this is true for any matrices M. we can write the ﬁrst of the Fierz rearrangement identities in (2. ¯ = ψ1γ Γ(a) M ¯ ψ4β ψ3α Γ(a) N γβ αδ ψ2δ αδ ¯ ¯ = ψ1γ ψ4β ψ3α ψ2δ Γ(a) M Γ(a) N ¯ ¯ ¯ = ψ1γ ψ4β ψ3α ψ2δ Mαβ Nγδ = ψ3 M ψ4 ¯ ψ1 N ψ2 . Let us note that using the explicit forms for Γ(a) and Γ(a) in (2.134) as (assuming the spinors are ordinary functions) ¯ ¯ ψ1 M ψ2 ψ3 N ψ4 = 1 ¯ ¯ ¯ ¯ ψ1 M ψ4 ψ3 N ψ2 + ψ1 M γ µ ψ4 ψ3 N γµ ψ2 4 ¯ ¯ ¯ ¯ +ψ1 M σ µν ψ4 ψ3 N σµν ψ2 − ψ1 M γ5 γ µ ψ4 ψ3 N γ5 γµ ψ2 ¯ ¯ +ψ1 M γ5 ψ4 ψ3 N γ5 ψ2 . the right-hand sides of the identities in (2. we have assumed that the spinors are ordinary functions.6 Properties of the Dirac matrices 61 ¯ ψ1 M Γ(a) ψ4 ¯ ψ3 N Γ(a) ψ2 αδ ¯ = ψ1α M Γ(a) ¯ ψ4δ ψ3γ N Γ(a) αδ γβ ψ2β γβ ¯ ¯ = ψ1α ψ4δ ψ3γ ψ2β M Γ(a) N Γ(a) ¯ ¯ ¯ = ψ1α ψ4δ ψ3γ ψ2β Mαβ Nγδ = ψ1 M ψ2 ¯ ψ1 Γ(a) M ψ4 ¯ ψ3 Γ(a) N ψ2 γβ ¯ ψ3 N ψ4 .122) and (2. On the other hand.134) The two relations in (2.135) equivalently as . (2.2. N and any spinors.134) pick up a negative sign which arises from commuting the fermionic ﬁelds past one another. we can ˜ deﬁne a new spinor ψ4 = N ψ4 to write the identity in (2.124) respectively. (2. In deriving these identities.

137) then using various properties of the gamma matrices derived earlier as well as (2. 1964. D. we obtain from (2. 1968. L. Drell.7 References 1. choose M = (½ − γ5 ) γ µ . . McGraw-Hill. J. for example. McGraw-Hill.138) This is the well known fact from the weak interactions that the V −A form of the weak interaction Hamiltonian proposed by Sudarshan and Marshak is form invariant under a Fierz rearrangement (the negative sign is there simply because we are considering spinor functions and will be absent for anti-commuting fermion ﬁelds). Schiﬀ. New York.110) and (2. (2. New York. N = (½ − γ5 ) γµ . 2. (2.136) ¯ ¯ ψ1 (½ − γ5 ) γ µ ψ2 ψ3 (½ − γ5 ) γµ ψ4 ¯ ¯ = −ψ1 (½ − γ5 ) γ µ ψ4 ψ3 (½ − γ5 ) γµ ψ2 . if we which is often calculationally simpler. Bjorken and S.136) Thus.62 2 Solutions of the Dirac equation ¯ ¯ ¯ ¯ ˜ ψ1 M ψ2 ψ3 N ψ4 = ψ1 M ψ2 ψ3 ψ4 = 1 ¯ ˜ ¯ ¯ ˜ ¯ ψ1 M ψ4 ψ3 ψ2 + ψ1 M γ µ ψ4 ψ3 γµ ψ2 4 ¯ ˜ ¯ ¯ ˜ ¯ +ψ1 M σ µν ψ4 ψ3 σµν ψ2 − ψ1 M γ5 γ µ ψ4 ψ3 γ5 γµ ψ2 ¯ ˜ ¯ +ψ1 M γ5 ψ4 ψ3 γ5 ψ2 = 1 ¯ ¯ ¯ ¯ ψ1 M N ψ4 ψ3 ψ2 + ψ1 M γ µ N ψ4 ψ3 γµ ψ2 4 ¯ ¯ ¯ ¯ +ψ1 M σ µν N ψ4 ψ3 σµν ψ2 − ψ1 M γ5 γ µ N ψ4 ψ3 γ5 γµ ψ2 ¯ ¯ +ψ1 M γ5 N ψ4 ψ3 γ5 ψ2 . I.113). Quantum Mechanics. (2. Relativistic Quantum Mechanics. 2.

New Delhi.2. Lectures on Quantum Mechanics. Real representations of ﬁnite Cliﬀord algebras. Quantum Field Theory. 6. 5.7 References 63 3. Classiﬁcation. Physical Review 109. S. 2003. Sudarshan and R. Hindustan Publishing. . Itzykson and J-B. 1657 (1991). C. (1957). Das. 4. Okubo. E. Zuber. I. Marshak. Proceedings of PaduaVenice conference on mesons and newly discovered particles. E. India. Journal of Mathematical Physics 32. G. New York. McGrawHill. 1860 (1958). A. 1980. C.

This page intentionally left blank .

Thus.3) that a 65 . The rotation around the z-axis in (3.1 Lorentz transformations In three dimensions. where R represents the rotation matrix such that x′ = cos θ x1 − sin θ x2 .3) so that the coeﬃcient matrix on the right hand side can be identiﬁed with the rotation matrix in (3. we see from (3. we are well acquainted with rotations.2) (3. 3 Here we are using a three dimensional notation.Chapter 3 Properties of the Dirac equation 3. we know that a rotation of coordinates around the z-axis by an angle θ can be represented as the transformation x → x′ = R x.1). For example.1) x′ = sin θ x1 + cos θ x2 . but this can also be written in terms of the four vector notation we have developed.2) can also be written in matrix form as x1 cos θ − sin θ 0 ′ x2 = sin θ cos θ 0 x2 . x′ 1 x′ 3 0 0 1 x3 (3. 1 (3. 2 x′ = x3 .

with unit determinant. x′ 3 = x3 . (3. Under a Lorentz boost along the x-axis. (3. 3 x′ = ǫx1 + x2 .4) where we have identiﬁed θ = ǫ = inﬁnitesimal. 2 (3. such that (3.5) x′ 0 = γx0 − γβx1 . x′ 2 = x2 . R (RT R = ½).6) can also be written in the matrix form as . x′ 1 = −γβx0 + γx1 . We observe here that the matrix representing the inﬁnitesimal change under a rotation is anti-symmetric. we also know that the coordinates transform as (boost velocity v = β since c = 1) xµ → x′ µ .66 3 Properties of the Dirac equation rotation around the 3-axis (z-axis) or in the 1-2 plane is denoted by an orthogonal matrix.2) that an inﬁnitesimal rotation around the 3-axis (z-axis) takes the form x′ = x1 − ǫx2 .7) We recognize that (3. We also note from (3. 1 x′ = x3 .6) where the Lorentz factor γ is deﬁned in terms of the boost velocity to be 1 1 − β2 γ= .

= 0 0 1 0 x2 0 0 0 1 x3 x′ 0 γ (3.3.12) . 1 0 0 1 (3.10) Λµν − sinh ω = 0 0 0 0 . Thus. 1 − β2 1 − β2 (3. we conclude from (3. where cosh ω − sinh ω 0 0 cosh ω 0 0 (3. (3. .9) that −∞ ≤ ω ≤ ∞.11) 1 β2 − = 1.8) 1 1 − β2 β .1 Lorentz transformations 67 where we have deﬁned cosh ω = γ = sinh ω = γβ = so that 0 x −γβ 0 0 ′ 1 1 x −γβ γ 0 0 x = x′ 2 0 0 1 0 x2 ′3 x 0 0 0 1 x3 0 x cosh ω − sinh ω 0 0 1 − sinh ω cosh ω 0 0 x . we note that a Lorentz boost along the x-direction can be written as x′ µ = Λµν xν .9) 1 − β2 cosh2 ω − sinh2 ω = Since the range of the boost velocity is given by −1 ≤ β ≤ 1.

we can think of the Lorentz boost along the 1-axis as a rotation in the 0-1 plane with an imaginary angle (so that we have hyperbolic functions instead of ordinary trigonometric functions).12) or (3.) Furthermore. ω. (3. then we can write (ω = ǫ = inﬁnitesimal) −ǫ 0 0 −ǫ 1 0 0 = δµν + ǫµν . 0 1 0 0 0 1 sinh ω (3. µ (3. (or the parameter of boost) can take any real value and. x′ = Λµν xν . Let us ﬁnally note that if we are considering an inﬁnitesimal Lorentz boost along the 1-axis (or a rotation in the 0-1 plane).) Therefore. as a result. Lorentz boosts correspond to noncompact transformations unlike space rotations. is easily seen from (3. namely.16) . (That these rotations become complex is related to the fact that the metric has opposite signature for time and space components.15) also shows that the covariant vector transforms in an inverse manner compared with the contravariant vector. Λµ (or Λµν ). = 0 0 1 0 0 0 0 1 1 Λµν (3. the “angle” of rotation.68 From this.3). (Incidentally. we can obtain.13) to be an ν orthogonal matrix in the sense that µ (ΛT )µν Λνλ = Λνµ Λνλ = δλ . (3.13) The matrix representing the Lorentz transformation of the coordinates. as we have seen.15) and also has a unit determinant. much like the rotation matrix R in (3. cosh ω 3 Properties of the Dirac equation ν Λµ = ηµλ η νρ Λλρ which would lead to the transformation of the covariant coordinate vector.14) sinh ω = 0 0 0 0 cosh ω 0 0 .

0 0 0 0 (3. ρ Λµρ Λµσ = δσ . General Lorentz transformations are deﬁned as the transformations x′ µ = Λµν xν . namely. or.20) . 0 −ǫ 0 0 0 0 0 ǫµν −ǫ = 0 0 0 0 . (3. therefore. In a general language. which leave the length of the vector invariant.19) ηµν x′ µ x′ ν = ηµν xµ xν . ηµν Λµρ Λνσ xρ xσ = ηρσ xρ xσ . which can be thought of as rotations in the four dimensional space-time.1 Lorentz transformations 69 where. or. Λµρ Λµσ = ηρσ . ηµν Λµρ xρ Λνσ xσ = ηµν xµ xν .18) In other words. = 0 0 0 0 0 0 0 0 0 ǫµν = η νλ ǫµλ (3.17) It follows from this that ǫ 0 0 −ǫ 0 0 0 = −ǫνµ . the matrix representing the change under an inﬁnitesimal Lorentz boost is anti-symmetric just like the case of an inﬁnitesimal rotation. we note that we can combine rotations and Lorentz boosts into what are known as the homogeneous Lorentz transformations.3. (3. or. or.

25) . Lorentz transformations deﬁne the maximal symmetry of the space-time manifold which leaves the origin invariant. det Λ = 1. and (3.70 3 Properties of the Dirac equation This is. “gonia” in Greek means an angle or a corner and. we note that ΛT as can be seen from (3.21) Therefore. then the transformation is called orthochronous. (3. what we have seen before in (3.20) explicitly as Λ00 Λ00 + Λi 0 Λi 0 = 1.20)) (det Λ)2 = 1. such a Lorentz transformation does not change the direction of time.) Note also that since ΛT Λ = ½. of course. or.22) If Λ00 ≥ 1. (3. Namely. 2 = 1 + Λi 0 ≥ 1. orthochronous means straight up in time. orthogonal means the corner that is straight up (perpendicular). Λ00 Λ00 2 2 − Λi 0 2 = 1. or. (The Greek preﬁx “ortho” means straight up.23) = Λνµ and ΛT ν µ = Λνµ The set of homogeneous Lorentz transformations satisfying Λ00 ≥ 1. Choosing ρ = σ = 0. or. we can write out the relation (3.15). therefore. Incidentally.24) µ ν (3. det Λ = ±1. or. we conclude that Λ00 ≥ 1. we obtain (for clarity. Thus. an orthodontist is someone who can make your teeth straight. (3. Λ00 ≤ −1. In the same spirit.

det Λ = −1 (which is orthochronous but no longer proper). t → −t. namely. Finally. then by adding space reﬂection. xµ → −xµ . orthochronous Lorentz transformations and constitute a set of continuous transformations that can be connected to the identity matrix. we obtain a Lorentz transformation 1 0 0 0 0 −1 0 0 Λprop . if we add both space and time reﬂections. to a proper Lorentz tranformation.28) satisfying Λ00 ≤ −1 and det Λ = −1 (which is neither proper nor orthochronous). = 0 0 1 0 0 0 0 1 Λ = Λtime Λprop (3. = 0 0 −1 0 0 0 0 −1 Λ = Λspace Λprop (3.3. Λ00 ≥ 1. (3. x → −x. we can obtain the other Lorentz transformations by simply appending space reﬂection or time reﬂection or both (which are discrete transformations).) In general. det Λ = ±1. Thus. if Λprop denotes a proper Lorentz transformation. If we add time reversal. however. we obtain a Lorentz transformation .1 Lorentz transformations 71 are known as the proper. to a proper orthochronous Lorentz transformation. then we obtain a Lorentz transformation −1 0 0 0 0 1 0 0 Λprop . det Λ = ±1.27) This would correspond to having Λ00 ≥ 1. Given the proper orthochronous Lorentz transformations. (Just to emphasize.26) Λ00 ≤ −1. there are four kinds of Lorentz transformations. we note that the set of transformations with det Λ = 1 are known as proper transformations and the set for which Λ00 ≥ 1 are called orthochronous.

covariance implies that a given equation is form invariant under a particular transformation (has the same form in diﬀerent reference frames).33) (3.2 Covariance of the Dirac equation Given any dynamical equation of the form Lψ = 0. 0 −1 0 (3. With this general deﬁnition. we say that it is covariant under a given transformation provided the transformed equation has the form L′ ψ ′ = 0. 3. however. (3.29) where L is a linear operator.30) −1 0 0 0 −1 0 = 0 0 −1 0 0 0 0 Λprop . (3. These additional transformations. In these lectures. In simple terms.32) . (3. we would refer to proper orthochronous Lorentz transformations as the Lorentz transformations.72 3 Properties of the Dirac equation Λ = Λspace−time Λprop with Λ00 ≤ −1 and det Λ = 1 (which is proper but not orthochronous). the operator L. let us now consider the Dirac equation iγ µ ∂µ − m ψ(x) = 0. Under a Lorentz transformation xµ → x′ µ = Λµν xν .31) where ψ ′ represents the transformed wavefunction and L′ stands for the transformed operator (namely. cannot be continuously connected to the identity matrix since they involve discrete reﬂections. with the transformed variables).

2 Covariance of the Dirac equation 73 if the transformed equation has the form ′ iγ µ ∂µ − m ψ ′ (x′ ) = 0.35) where S(Λ) is a 4 × 4 matrix. then the Dirac equation would be covariant under a Lorentz transformation. the Lorentz transformation can also mix up the spinor components (much like angular momentum/rotation). we can deﬁne an operator L(Λ) to represent the Lorentz transformation on the coordinate states as (with indices suppressed) |x → |x′ = |Λx = L(Λ)|x .3. the transformed wavefunction has the form ψ ′ (x′ ) = ψ ′ (Λx) = S(Λ)ψ(x). (with S(Λ) representing the 4 × 4 matrix which rotates the matrix components of the wave function) |ψ → |ψ ′ = L(Λ)S(Λ)|ψ = = L(Λ)S(Λ) dx |x x|ψ dx L(Λ)|x S(Λ)ψ(x) . In the notation of other symmetries that we know from studies in non-relativistic quantum mechanics. what this means is that we are ﬁnding a representation of the Lorentz transformation on the Hilbert space. Let us assume that. under a Lorentz transformation. (3.36) However. are a set of four space-time independent matrices and. do not change under a Lorentz transformation. (3. therefore. since ψ(x) is a four component spinor. we can deﬁne the Lorentz transformation acting on the Dirac Hilbert space (Hilbert space of states describing a Dirac particle) as. since the Dirac wavefunction is a four component spinor. in addition to the change in the coordinates. γ µ . Parenthetically.34) where ψ ′ (x′ ) is the Lorentz transformed wave function. Thus. Note that the Dirac matrices. (3.

35)) ψ ′ (x′ ) = x′ |ψ ′ = S(Λ)ψ(x). can be represented by a matrix S(Λ) which depends only on the parameter of transformation Λ and not on the space-time coordinates. ′ iΛµν γ ν ∂µ − m S −1 (Λ)ψ ′ (x′ ) = 0. (3.37) we obtain (see (3. A more physical way to understand this is to note that the Dirac wave function simply consists of four functions which do not change.39) (3. but get rotated by the S(Λ) matrix.42) .33) that ∂x′ µ = Λµν . from (3. on the wave function.37) where the wave function is recognized to be ψ(x) = x|ψ . (3. Since the Lorentz transformations are invertible. we can write iγ µ ∂µ − m ψ(x) = 0. the matrix S(Λ) must possess an inverse so that from (3. ′ iΛµν Sγ ν S −1 ∂µ − m ψ ′ (x′ ) = 0.40) (3. so that. or. (3. Let us also note from (3.74 = 3 Properties of the Dirac equation dx |Λx S(Λ)ψ(x).38) Namely.35) we can write ψ(x) = S −1 (Λ)ψ ′ (x′ ). ∂xµ ∂x′ ν iγ µ Λνµ ∂ν′ − m S −1 (Λ)ψ ′ (x′ ) = 0.41) ∂x′ ν ∂ − m S −1 (Λ)ψ ′ (x′ ) = 0. or. ∂xν deﬁne a set of real quantities. iγ µ (3. or. Thus. the eﬀect of the Lorentz transformation. or.

Therefore.44) + = + + µν = Λµρ Λνσ γ ρ . (3. Λνσ γ σ (3. or.15)). Λµν γ ν = S −1 γ µ S.43) Let us note that if we deﬁne γ ′ µ = Λµν γ ν . Next. γ σ = 2Λµρ Λνρ ½ = 2η = Λµρ Λνσ 2η ρσ ½ ½. (3. then. there must exist a matrix connecting the two representations. let us note that since the parameters of Lorentz transformation are real (namely. such that Λµν Sγ ν S −1 = γ µ .40). or. or.43) that the matrix S exists and all we need to show is that it also generates Lorentz transformations in order to prove that the Dirac equation is covariant under a Lorentz transformation. Therefore. by Pauli’s fundamental theorem. Sγ 0 S † γ 0 γ 0 γ µ γ 0 γ 0 S −1 γ 0 S −1 = γ µ . the matrices γ ′ µ satisfy the Cliﬀord algebra and. (3. generating Lorentz transformations (for the Dirac wavefunction). (Λ∗ )µν = Λµν ) γ 0 Λµν γ ν γ 0 = Λµν γ 0 (γ ν )† γ 0 = Λµν γ ν . γ ′ µ. γ µ and γ ′µ . γ 0 (S −1 γ µ S)† γ 0 = S −1 γ µ S. It now follows from (3.2 Covariance of the Dirac equation 75 where we have used (3. γ ′ ν Λµρ γ ρ . (Sγ 0 S † γ 0 )γ µ (Sγ 0 S † γ 0 )−1 = γ µ .3. we see from (3.45) where we have used the orthogonality of the Lorentz transformations (see (3. or.42) that the Dirac equation will be form invariant (covariant) under a Lorentz transformation provided there exists a matrix S(Λ).46) † † † .

As a result. (3. we determine det (Sγ 0 S † γ 0 ) = det (b ½).48) which.49) We also note that det γ 0 = 1 and since we are interested in proper Lorentz transformations.47) Taking the Hermitian conjugate of (3. or. therefore. we can denote Sγ 0 S † γ 0 = b ½. b4 = 1.76 3 Properties of the Dirac equation Here we have used (3. . Sγ 0 S † γ 0 = b∗ ½ = b ½. (3. Using these in (3. determines that the parameter b is real.46) that the matrix Sγ 0 S † γ 0 commutes with all the γ µ matrices and. It is clear from (3. or. γ 0 (γ 0 Sγ 0 S † )γ 0 = b∗ ½. or. b = b∗ . det S = 1. (3. namely.43) and the relations (γ 0 )† = γ 0 = (γ 0 )−1 as well as γ 0 (γ µ )† γ 0 = γ µ . therefore. γ 0 Sγ 0 S † = b∗ ½.51) In fact.50) The real roots of this equation are b = ±1. S † γ 0 = b γ 0 S −1 .47). or.47). we can determine the unique value of b in the following way. (3. (3.100) and calculating the commutator with γ µ ). we obtain (Sγ 0 S † γ 0 )† = b∗ ½. must be proportional to the identity matrix (this can be easily checked by taking a linear combination of the sixteen basis matrices in (2. or.

it follows from (3. that S †S = S †γ 0γ 0S = bγ 0 S −1 γ 0 S = bγ 0 Λ0ν γ ν = bγ 0 Λ00 γ 0 + Λ0i γ i = b Λ00 + Λ0i γ 0 γ i .47). or.52) which follows since S † S represents a non-negative matrix. we are assuming Λ00 ≥ 1. . Since we are dealing with proper Lorentz transformations. Thus. S † γ 0 = γ 0 S −1 . Tr S † S = 4bΛ00 0.55) These are some of the properties satisﬁed by the matrix S which will be useful in showing that it provides a representation for the Lorentz transformations. or.54) which implies (see (3.53) or. (3. b > 0.3. b < 0. The two solutions of this equation are obvious Λ00 ≥ 1.47) that Sγ 0 S † γ 0 = ½.56) (3.51) that b = 1. we conclude from (3.53)) that b > 0 and. Λ00 ≤ −1.43) and (3. (3.2 Covariance of the Dirac equation 77 Let us note. therefore. using (3. (3. (3.

we recall that the inﬁnitesimal transformation matrix is antisymmetric. since ǫµν = −ǫνµ .57) From our earlier discussion in (3. let us consider an inﬁnitesimal Lorentz transformation of the form (ǫµν inﬁnitesimal) x′µ = Λµν xν = (δµ + ǫµ )xν = xµ + ǫµ xν . Mµν denote matrices in the Dirac space). 4 (3.60) ½+ ½+ i Mµν ǫµν 4 ½− i Mστ ǫστ 4 i i Mµν ǫµν − Mµν ǫµν + O(ǫ2 ) 4 4 (3. namely.78 3 Properties of the Dirac equation Next.61) (3.18). We can also write S −1 (ǫ) = ½ + so that S −1 (ǫ)S(ǫ) = = = i Mµν ǫµν . 4 (3.62) (3. ǫµν = −ǫνµ .59) where the matrices Mµν are assumed to be anti-symmetric in the Lorentz indices (for diﬀerent values of the Lorentz indices. (3.63) ½ + O(ǫ2 ). Mµν = −Mνµ . .58) For an inﬁnitesimal transformation. ν ν ν (3. we can expand the matrix S(Λ) as S(Λ) = S(ǫ) = ½ − i Mµν ǫµν . therefore.

64) i ǫλρ γ µ .64). σ νλ = 2i η µν γ λ − η µλ γ ν . 4 4 (3.64) that if we identify M λρ = σ λρ . i ǫλρ γ µ . M λρ 4 i ǫλρ γ µ . we have determined the form of S(ǫ) to be S(ǫ) = ½ − i i Mµν ǫµν = ½ − σµν ǫµν .106) γ µ . ½+ γµ + − i Mλρ ǫλρ γ µ 4 ½− i Mστ ǫστ 4 = γ µ + ǫµν γ ν . i i λρ ǫ Mλρ γ µ − ǫλρ γ µ Mλρ + 0(ǫ2 ) = γ µ + ǫµν γ ν .2 Covariance of the Dirac equation 79 To the leading order. let us recall the commutation relation (2. therefore. then. The deﬁning relation for the matrix S(Λ) in (3.68) . or. or.3. M λρ = ǫµν γ ν . λ 2 ρ = − (3. or. we see that for inﬁnitesimal transformations.43) now takes the form S −1 (ǫ)γ µ S(ǫ) = Λµν γ ν = δµν + ǫµν γ ν . 4 At this point.67) which coincides with the right hand side of (3.65) − (3. S −1 (ǫ) indeed represents the inverse of the matrix S(ǫ).66) (3. and note from (3. σ λρ 4 i = − ǫλρ × 2i η µλ γ ρ − η µρ γ λ 4 1 µ ρ = ǫ γ + ǫµ γ λ = ǫµν γ ν . 4 4 (3. Therefore.

we have shown that there exists a S(Λ) which satisies (3. let us note that inﬁnitesimal rotations around the 3-axis or in the 1-2 plane would correspond to choosing ǫ12 = ǫ = −ǫ21 . S(ǫ) = ½ − i 1 i σ12 ǫ = ½ + γ1 γ2 ǫ = ½ − α3 ǫ.69) with the generators of inﬁnitesimal Lorentz transformations for the 1 Dirac wave function. For completeness.) We will see in the next chapter (when we study the representations of the Lorentz group) that the algebra (2. ˜ 2 2 2 (3. satisfy can 2 be identiﬁed with the Lorentz algebra (which also explains why they are closed under multiplication).70) with all other components of ǫµν vanishing. 2 (3.80 3 Properties of the Dirac equation Let us note here from the form of S(ǫ) that we can identify 1 σµν . be obtained from an inﬁnite sequence of inﬁnitesimal transformations resulting in an exponentiation of the inﬁnitesimal generators as ˜ S(θ) = e− 2 α3 θ .98)). i (3. the matrix S(Λ) for a ﬁnite Lorentz transformation will be the product of a series of such inﬁnitesimal matrices which leads to an exponentiation of the inﬁnitesimal generators with the appropriate parameters of transformation. (3. (The other factor of 2 is there to avoid double counting. be constructed out of a series of inﬁnitesimal transformations and.72) .71) A ﬁnite rotation by angle θ in the 1-2 plane would. 1 σµν . at least for inﬁnitesimal Lorentz transformations. consequently. of course. In such a case (see also (2. the Dirac equation is form invariant (covariant) under such a Lorentz transformation. Thus.43) and generates Lorentz transformations and as a result. then. A ﬁnite transformation can.109) which the generators of the inﬁnitesimal transformations.

so that we can write (see also (2. therefore. we are considering an inﬁnitesimal boost along the 1-axis. consistent with the fact that the Dirac equation describes spin 1 2 particles. of course. we can determine S(θ) = cos This shows that S(θ + 2π) = − cos S(θ + 4π) = S(θ).3. (3. is double valued and.74) and. 0 ½ (3.77) with all other components of ǫµν vanishing. α 2 2 (3. This is. namely. in this case. Furthermore. namely. Let us next consider an inﬁnitesimal rotation in the 0-1 plane. we can identify ǫ01 = ǫ = −ǫ10 .98)) S(ǫ) = i 1 σ01 ǫ = ½ + γ0 γ1 ǫ 2 2 1 = ½ − α1 ǫ. rotations ˜i ˜ deﬁne unitary transformations. recalling that σ3 0 0 σ3 α3 = ˜ we have 2 .73) α3 ˜ = ½ 0 = ½.76) θ θ − i˜ 3 sin . we have S † (θ) = S −1 (θ).2 Covariance of the Dirac equation 81 Note that since α† = αi . corresponds to a spinor representation. 2 ½− (3. (3.78) . the rotation operator.75) That is. (3. θ θ − i˜ 3 sin α 2 2 = −S(θ). therefore. In this case.

α2 = 1 ½ 0 = ½. there does not exist any ﬁnite dimensional unitary representation. 0 ½ (3. (3.79) α1 = . as 4 × 4 matrices). This is related to the fact that Lorentz boosts are non-compact transformations and for such transformations. operators deﬁning boosts are not unitary. Let us note next that. since under a Lorentz transformation .83) That is. the matrix for a ﬁnite boost. All the unitary representations are necessarily inﬁnite dimensional. ω. 3. in this four dimensional space (namely. recalling that 0 σ1 σ1 0 1 (3.82 3 Properties of the Dirac equation In this case. Furthermore. we have shown how to construct the matrix S(Λ) for ﬁnite Lorentz transformations (for both rotations and boosts). can be obtained through exponentiation as S(ω) = e− 2 α1 ω .3 Transformation of bilinears In the last section.82) We note here that since α† = α1 .80) and. 2 2 S(ω) = cosh (3. (3. 1 S † (ω) = S −1 (ω).81) we can determine ω ω − α1 sinh . therefore.

84) (3.50) and (2.3 Transformation of bilinears 83 ψ ′ (x′ ) = S(Λ)ψ(x). it follows that ψ ′ † (x′ ) = ψ † (x)S † (Λ).3. we see that if we deﬁne a current of the form j µ (x) = ψ(x)γ µ ψ(x). (3.55)).87) ′ where we have used (3. namely.85) where we have used the relation (3. This implies that a bilinear product such as ψψ would transform under a Lorentz transformation as ψ (x′ )ψ ′ (x′ ) ψ(x)S −1 (Λ)S(Λ)ψ(x) = ψ(x)ψ(x). Similarly.88) . (3. such a product will not change under a Lorentz transformation – would behave like a scalar – which is what we had discussed earlier in connection with the normalization of the Dirac wavefunction (see (2. J µ (x) = ψ(x)γ µ ψ(x) → Λµν J ν (x).86) ′ ψ(x)ψ(x) → = Namely. under a Lorentz transformation ψ(x)γ µ ψ(x) → = = ψ (x′ )γ µ ψ ′ (x′ ) ψ(x)S −1 (Λ)γ µ S(Λ)ψ(x) ψ(x)Λµν γ ν ψ(x) = Λµν ψ(x)γ ν ψ(x). we see that the adjoint wave function ψ(x) transforms inversely compared to the wave function ψ(x) under a Lorentz transformation. it would transform as a four vector under a proper Lorentz transformation. (3. ′ (3. Thus. ψ (x′ ) = ψ ′ † (x′ )γ 0 = ψ † (x)S † (Λ)γ 0 = ψ † (x)γ 0 S −1 (Λ) = ψ(x)S −1 (Λ).56). In other words.43).

the probability current density (see also (2. p) = 0. 3.93) . p) = 0.92) so that the equations satisﬁed by u(p) and v(p) (positive and negative energy solutions). completeness relation Let us note that the positive energy solutions of the Dirac equation satisfy (p − m)u+ (p) = γ 0 p0 − γ · p − m u+ (p0 . what we had observed earlier. −p) = u− (−p) = . E+m 2m σ ·p E+m v ˜ v ˜ v(p) = u− (−p0 .89) and (3. can be written as (p − m)u(p) = 0. It is customary to identify (see (2. / where p0 = ω = p2 + m2 . the reason for this will become clear when we discuss the quantization of Dirac ﬁeld theory later) E+m 2m u ˜ σ·p E+m u ˜ u(p) = u+ (p) = .89) while the negative energy solutions satisfy − γ 0 p0 − γ · p − m u− (−p0 .91). Finally.90) (3.4 Projection operators.49).84 3 Properties of the Dirac equation This is. we can determine the transformation properties of the other bilinears under a Lorentz transformation in a straightforward manner. Namely.85)) transforms like a four vector so that the probability density transforms as the time component of a four vector.90). / (3. (3.91) with the same value of p0 as in (3. (3. (3. (3. we note that in this way. of course.

u (p)v (p) = 0 = v r (p)us (p). the adjoint equations are easily obtained to be (taking the Hermitian conjugate and multiplying γ 0 on the right) u† (p) (p)† − m γ 0 = 0. (3.97) . With these notations. as we had seen earlier. 2. / v † (p) (p)† + m γ 0 = 0. / (p + m)v(p) = 0.3. / or. 4.83)). As we have seen earlier there are two positive energy solutions and two negative energy solutions of the Dirac equation. v(p)(p + m) = 0. or. Let us also note that each of the four solutions really represents a four component spinor.95) where we have used (γ µ )† γ 0 = γ 0 γ µ (see (2. 3. or. / (3. 2.94) Given these equations. (3. (−p − m)v(p) = 0. r = 1. we can write down the Lorentz invariant conditions we had derived earlier from the normalization of a massive Dirac particle as (see (2. u(p)(p − m) = 0. / or. α v r (p)v s (p) = α=1 r s r s v α (p)vα (p) = −δrs . Let us denote them by ur (p) and v r (p).4 Projection operators. completeness relation 85 and − γ 0 p0 + γ · p − m v(p) = 0.50)) 4 u r (p)us (p) = α=1 4 r uα (p)us (p) = δrs . Let us denote the spinor index by α = 1. / (3.96) where r. can represent the spin projection of the two component spinors (in terms of which the four component solutions were obtained).

86 3 Properties of the Dirac equation Although we had noted earlier that u+ (p)u− (p) = 0. (3. From the form of the equations satisﬁed by the positive and the negative energy spinors. it is clear that we can deﬁne projection operators for such solutions as p+m / . (3. 2m −p + m / . (3. (3. 2m −p + m r / u (p) = 0.101) . This also allows us to write 2 r=1 [u r (p)ur (p) − v r (p)v r (p)] = 4.100) Λ− (p) = 2m These are. Λ+ (p) = Λ+ (p)ur (p) = = Λ+ (p)v r (p) = Λ− (p)ur (p) = Λ− (p)v r (p) = = p+m r / u (p) 2m p − m + 2m r / u (p) = ur (p). because the direction of momentum changes for v(p) (see the derivation in (2. −p). 4 × 4 matrices and their eﬀect on the Dirac spinors is quite clear.93) and (3. 2m (3. namely.99) for any two spin components of the positive and the negative energy spinors.52)). 2m p+m r / v (p) = 0. of course. the last relation in (3.94).97) can be checked to be true simply because v(p) = u− (−p0 .98) For completeness we note here that it is easy to check u† (p)v(−p) = 0 = v † (−p)u(p). 2m −p + m r / v (p) 2m −p − m + 2m r / v (p) = v r (p).

2m / p + m −p + m / × 2m 2m 1 1 −p2 + m2 = −m2 + m2 2 4m 4m2 0 = Λ− (p)Λ+ (p). we see that / Λ± (p) are indeed projection operators and they are orthogonal to each other.4 Projection operators. 2m 2m Λ+ (p) + Λ− (p) = (3. Let us deﬁne a 4 × 4 matrix P with elements . Let us note that Λ+ (p)Λ+ (p) = = = Λ− (p)Λ− (p) = = = Λ+ (p)Λ− (p) = = = p+m / 2m 2 = p2 + 2mp + m2 / 4m2 2m(p + m) / m2 + 2mp + m2 / = 4m2 4m2 p+m / = Λ+ (p). Furthermore. (3.103) as it should be since all the solutions can be divided into either positive or negative energy ones. Let us next consider the outer product of the spinor solutions. while Λ− (p) projects only on to the space of negative energy ones. Thus.102) where we have used (p)2 = p2 and p2 = m2 .3. 2m −p + m / 2m 2 = p2 − 2mp + m2 / 4m2 2m(−p + m) / m2 − 2mp + m2 / = 2 4m 4m2 −p + m / = Λ− (p). let us also note that p + m −p + m / / + = ½. completeness relation 87 Similar relations also hold for the adjoint spinors and it is clear that Λ+ (p) projects only on to the space of positive energy solutions.

105) Thus. we see that the matrix P projects only on to the space of positive energy solutions and. Namely.106) Similarly. αβ (3. 4. β = 1. α α 4 s uβ (p)Pβα (p) β=1 2 2 s r uβ (p)ur (p)uα (p) β β=1 r=1 r s δrs uα (p) = uα (p). we can identify Pαβ (p) = (Λ+ (p))αβ . 2. 2 or. 3. α α.104) This matrix has the property that acting on a positive energy spinor it gives back the same spinor. if we deﬁne .88 3 Properties of the Dirac equation 2 Pαβ (p) = r=1 r ur (p)uβ (p). (3. r=1 4 (u (p)P (p))α = s = = (P (p)v s (p))α = β=1 4 s Pαβ (p)vβ (p) 2 s r ur (p)uβ (p)vβ (p) = 0. α β=1 r=1 = (3. r=1 r ur (p)uβ (p) = α p+m / 2m . therefore. 4 4 2 r ur (p)uβ (p)us (p) β α (P (p)us (p))α = β=1 2 Pαβ (p)us (p) = β β=1 r=1 = r=1 4 ur (p)δrs = us (p).

109) The completeness relation for the solutions of the Dirac equation now follows from the observation that (see (3. αβ (3.108) Namely. the matrix Q projects only on to the space of negative energy solutions with a phase (a negative sign).103)) .107) then.4 Projection operators. it is straightforward to see that 4 (Q(p)us (p))α = β=1 4 Qαβ (p)us (p) β 2 r r vα (p)v β (p)us (p) = 0. Hence we can identify Qαβ = −(Λ− (p))αβ .3. (3. β β=1 r=1 4 4 s Qαβ (p)vβ (p) = β=1 2 β=1 r=1 r s vα (p)(−δrs ) = −vα (p). 4 s v β (p)Qβα (p) = β=1 2 β=1 r=1 2 s r r v β (p)vβ (p)v α (p) 2 r s r vα (p)v β (p)vβ (p) = (Q(p)v s (p))α = = r=1 4 (v s (p)Q(p))α = = r=1 r s (−δrs )v α = −v α . (3. 2 or. r=1 r r vα (p)v β (p) = p−m / 2m . completeness relation 89 2 Qαβ (p) = r=1 r r vα (p)v β (p).

112) where M stands for a 4 × 4 matrix. (3. uu and vv have opposite sign.110) In a matrix notation.110) or in (3. M (3. rather it involves a sum with the metric structure of the space built in. These relations are particularly useful in simplifying the evaluations of transition amplitudes and probabilities. this may represent the expectation value of a given operator in a given electron state and will have the form (r not summed) = u r (p)M ur (p). we can think of the space of solutions of the Dirac equation as an indeﬁnite metric space.113) If we are not interested in the expectation value in a particular electron state. r=1 r r r ur (p)uβ (p) − vα (p)v β (p) = δαβ . 2 or. For example. the latter being negative while the former is positive. the completeness relation does not involve a sum of terms with positive deﬁnite sign. As we have seen. the completeness relation (3. In such a space. let us suppose that we are interested in a transition amplitude which has the form u r (p)M us (p′ ).111) We note here that the relative negative sign between the two terms in (3.110) can also be written as 2 r=1 (ur (p)u r (p) − v r (p)v r (p)) = ½. Hence. but rather wish to obtain an average over the two possible .90 3 Properties of the Dirac equation Pαβ − Qαβ = (Λ+ (p))αβ + (Λ− (p))αβ = δαβ .111) can be understood as follows. (3. If the initial and the ﬁnal states are the same. α (3.

λ=1 r.β.s=1 1 2 4 2 4 r s uα (p)Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α. if we have a transition from a given electron state to another and if we are interested in a process where we average over the initial electron states and sum over the ﬁnal electron states (for example.β=1 1 Tr M Λ+ (p). 2 (3.s=1 σλ r ur (p)uα (p) λ σλ = ur (p) λ = .σ. think of an experiment with unpolarized initial electron states where the ﬁnal spin polarization is not measured).s=1 2 u s (p′ )M † (u r )† (p) † = 1 u r (p)M us (p′ )u s (p′ )γ 0 M † γ 0 ur (p) 2 r. the probability for such a transition will be determined from 1 u r (p)M us (p′ ) 2 r. completeness relation 91 electron states (in experiments we may want to average over the spin polarization states).s=1 1 2 r.β=1 r=1 4 = = = Mαβ (Λ+ (p))βα = α.114) Similarly.3.s=1 2 2 u r (p)M us (p′ ) † = 1 u r (p)M us (p′ ) 2 r.4 Projection operators.σ. then we will have 1 2 1 2 1 2 1 2 2 M = u r (p)M ur (p) r=1 2 4 r uα (p)Mαβ ur (p) β r=1 α.β.β=1 4 2 r Mαβ ur (p)uα (p) β α.λ=1 2 s Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α.

Note. it commutes with the total angular momentum). 2 The trace is over the 4×4 matrix indices and can be easily performed using the properties of the Dirac matrices that we have discussed earlier in section 2. unlike the case of non-relativistic systems where we specify a given energy state by the projection of spin along the z-axis (namely. . 3.115) = 1 Tr M Λ+ (p′ )γ 0 M † γ 0 Λ+ (p) .116) does not commute either with the orbital angular momentum or with spin (rather. In fact.6. H] = (3. in the relativistic case this is not useful since spin is not a constant of motion.118) As a consequence. On the other hand.92 1 2 4 3 Properties of the Dirac equation = Mαβ Λ+ (p′ α.67)) 1 αi .5 Helicity As we have seen. H = −iǫijk αj pk . Thus. spin commutes with the Hamiltonian (since in this frame p = 0) and such solutions can be labelled by the spin projection. (3. however. it can be easily checked that the plane wave solutions which we had derived earlier are not eigenstates of the spin operator. we note that since momentum commutes with the Dirac Hamiltonian. ˜ 2 [Si . the Dirac Hamiltonian H = α · p + βm.λ=1 βσ γ 0M †γ 0 σλ (Λ+ (p))λα (3. for a particle at rest. 2 σi 0 0 σi S= αi = ˜ . namely. by the eigenvalue of Sz ). that.β.117) satisﬁes the commutation relation (see (2. (3. we have already seen that the spin operator 1 ˜ α.σ.

124) . for a Dirac parti1 cle. the eigenstates of energy can also be labelled by the helicity eigenvalues.119) the operator S · p does also (momentum and spin commute and. therefore. Note that S·p |p | 2 h2 = = 1 ½. (3.122) where we have used (this is the generalization of the identity satisﬁed by the Pauli matrices) (S · p )(S · p ) = 1 1 ˜ ˜ (α · p) (α · p) = p2 ½. H]pi = −iǫijk αj pk pi = 0. can only be ± 2 and we can label the positive and the negative energy solutions also as u(p. (3. h′ ).121) measures the longitudinal component of the spin of the particle or the projection of the spin along the direction of motion.5 Helicity 93 [pi . H] = [Si . h′ ) = δhh′ = −v(p. α · p + βm] = 0.3. h)v(p. Namely. h)u(p. 4 (3. h′ ) = 0 = v(p. H] = [pi . this operator is a constant of motion.123) Therefore. h)u(p. [Si pi . |p | (3. h). (3. 4 4 (3. The normalization relations in this case will take the forms u(p. This is known as the helicity operator and we note that since the Hamiltonian commutes with helicity. v(p. h) with h = ± 1 (the two helicity 2 eigenvalues). the order of these operators in the product is not relevant). h′ ). the eigenvalues of the helicity operator. The normalized operator h= S·p . h)v(p. u(p.120) Therefore.

σ · p u1 (p) − p0 + m u2 (p) = 0. (3. the completeness relation. h)] = ½.126) where we are not assuming any relation between p0 and p as yet. (3.130) . h) − v(p.128) Explicitly. (3. Let us represent the four component spinor (as before) as u1 (p) u2 (p) u(p) = .127) where u1 (p) and u2 (p) are two component spinors.110) or (3.125) 1 h=± 2 3.94 3 Properties of the Dirac equation Furthermore. h)v(p. the Dirac equation takes the form (p0 − m)½ σ·p −σ · p u1 u2 −(p0 + m)½ = 0. In terms of u1 (p) and u2 (p). 1 2 particle. γ 0 p0 − γ · p − m u(p) = 0. or. this leads to the two (2-component) coupled equations p0 − m u1 (p) − σ · p u2 (p) = 0.6 Massless Dirac particle Let us consider the free Dirac equation for a massive spin (γ µ pµ − m) u(p) = 0. can now be written as [u(p. which can also be written as p0 u1 (p) − σ · p u2 (p) = mu1 (p). (3. (3.129) p0 u2 (p) − σ · p u1 (p) = −mu2 (p). (3.111). h)u(p. (3.

2 1 (u1 (p) + u2 (p)) .131) p0 + σ · p (u1 (p) − u2 (p)) = m (u1 (p) + u2 (p)) . (3. (3.133) We note that it is the mass term which couples the two equations. are not invariant under parity or space reﬂection and are known as the Weyl equations. the two equations in (3.134) These two equations. p0 uL (p) + σ · p uL (p) = muR (p). Let us note that in the limit m → 0. since vanishing of the mass which is a Lorentz scalar should not change the behavior of the equation under proper Lorentz transformations). 2 uL (p) = uR (p) = (3. These equations. p0 uL (p) = −σ · p uL (p). the equations in (3.132) then. We note that if we deﬁne two new (2-component) spinors as 1 (u1 (p) − u2 (p)) . can be shown to be covariant under proper Lorentz transformations (as they should be. (3. we obtain p0 − σ · p (u1 (p) + u2 (p)) = m (u1 (p) − u2 (p)) .3.131) can be rewritten as a set of two coupled (2-component) spinor equations of the form p0 uR (p) − σ · p uR (p) = muL (p).130).6 Massless Dirac particle 95 Taking the sum and the diﬀerence of the two equations in (3. like the Dirac equation. however.133) reduce to two (2-component) spinor equations which are decoupled and have the simpler forms p0 uR (p) = σ · p uR (p). . The corresponding two component spinors uL and uR are also known as Weyl spinors.

96 Let us note that 3 Properties of the Dirac equation or. σ·p uL (p) = −uL (p). (3. (3. p0 uL (p) = −σ · p uL (p).138) For p0 = |p |. (3. we must have p0 = ±|p |.137) which is the Einstein relation for a massless particle. |p | (3. It is clear. the two diﬀerent Weyl equations really describe particles with opposite helicity. (3. while (3. for the positive energy solutions. namely.139) or.140) In other words. therefore. σ·p uR (p) = uR (p).136) Thus. for a nontrivial solution of these equations to exist. that for such solutions. = (σ · p) (σ · p) uR (p) = p 2 uR (p). (p0 )2 uR (p) = σ · p p0 uR (p) (p0 )2 − p 2 uR (p) = 0. we can show that uL (p) also satisﬁes (p0 )2 − p 2 uL (p) = 0. p0 uR (p) = σ · p uR (p). we must have (p0 )2 − p 2 = 0. we note that p0 uR (p) = σ · p uR (p). Recalling that 1 σ denotes the spin 2 . Similarly. |p | or.135) or.

On the other hand. We also know. namely.6 Massless Dirac particle 97 operator for a two component spinor.) p p Left-handed Right-handed Figure 3. the electron neutrino emitted in a beta decay A X Z+1 → A Y Z + e+ + νe . (3. . uR (p) describes a particle with helicity + 1 or a particle with spin parallel to its direction of 2 motion. Correspondingly. we note that uL (p) describes a particle with helicity − 1 or a particle with spin anti-parallel to its 2 direction of motion. therefore. experimentally. 3. This is shown in Fig. Such a particle is known as a right-handed particle since its spin motion would correspond to that of a right-handed screw. can be described by a two component Weyl equation. If we think of spin as arising from a circular motion. then we conclude that for such a particle.1: Right-handed and left-handed particles with spins parallel and anti-parallel to the direction of motion. that νe is left-handed. (Handedness is also referred to as chirality and these spinors can be shown to be eigenstates of the γ5 matrix which can also be understood more easily from the chiral symmetry associated with massless Dirac systems. As we know. such a particle is called a left-handed particle (which is the reason for the subscript L).3. the circular motion would correspond to that of a left-handed screw.141) is massless (present experiments suggest they are almost massless) and.1 and we note here that this nomenclature is opposite of what is commonly used in optics.

would violate parity. hence. |p | → |p |. The neutrino is described by the equation σ·p uL (p) = −uL (p). (3. we conclude that it must transform under parity like L. will have opposite helicity or will be right-handed.146) Consequently. . |p | (3. p → −p. In the hole theoretic language. observed in experiments such as n → p + e− + ν e . |p | |p | |p | (3. therefore. (3. L = x × p → (−x ) × (−p ) = x × p = L.142) and has negative helicity. and processes involving neutrinos.98 3 Properties of the Dirac equation 1 its helicity is − 2 .143) A very heuristic way to conclude that parity is violated in processes involving neutrinos is as follows.144) Under parity or space reﬂection. (3. the neutrino equation is not invariant under parity.145) Since σ represents an angular momentum. This has been experimentally veriﬁed in a number of processes. the absence of a negative energy neutrino would appear as a “hole” with opposite helicity. x → −x. Alternatively. Therefore. It is helicity which is the conserved quantum number and. the neutrino is left-handed and hence satisﬁes the equation p0 uL = −σ · p uL . the absence of a negative energy neutrino would appear as a “hole” with the momentum reversed. then. in this description. That the antineutrino is right-handed is. so that under a space reﬂection σ · (−p) σ·p σ·p → =− . of course. the anti-neutrino.

(3. we note that if u(p) (or v(p)) is a solution. γ5 uR (p) = uR (p). (3.54).61)). namely.148) From the structure of the massless Dirac equation (3.7 Chirality With the normalization for massless spinors discussed in (2.147) u(p) = E 2 E 2 = |p| 2 |p| 2 u(p) ˜ σ·p u(p) . the solutions of the massless Dirac equation (m = 0) pu(p) = 0 = pv(p).53) and (2.147). see. (3.3.150) it follows that the eigenvalues of γ5 are ±1 and spinors with the eigenvalue +1.92)) u(p) ˜ σ·p u(p) ˜ E E (3. the solutions of the massless Dirac equation can be classiﬁed according to the eigenvalues of γ5 also known as the chirality or the handedness. (2.149) commutes with γ5 (in fact. γ5 vR (p) = vR (p). / / can be written as (see also (3. then γ5 u(p) (or γ5 v(p)) is also a solution. Since 2 γ5 = ½.7 Chirality 99 3.151) .100)) H = α · p. for example.60) and ρ commutes with α. (3. in the Pauli-Dirac representation γ5 = ρ deﬁned in (2. This can also be seen from the fact that the Hamiltonian for a massless Dirac fermion (see (1. ˜ |p| v(p) = σ·p v (p) ˜ v (p) ˜ = σ ·p v (p) ˜ |p| v (p) ˜ . Therefore.

(3. γ5 vL (p) = −vL (p). = 1 2 ˜ ½ + σ ·p u(p) 2 |p| uL (p) = PL u(p) = = |p| 2 1 (½ − γ5 )u(p) 2 1 ½ − σ ·p u(p) ˜ 2 −1 2 ½ |p| − σ·p |p| u(p) ˜ . Given a general spinor. ½ + σ ·p v(p) |p| ˜ vL (p) = PL v(p) = = |p| 2 1 (½ − γ5 )v(p) 2 ˜ − 1 ½ − σ·p v (p) 2 1 2 ½ − σ ·p v(p) |p| ˜ |p| . the right-handed and the left-handed components can be obtained through the projection operators (½ denotes the identity matrix in the appropriate space) 1 uR (p) = PR u(p) = (½ + γ5 )u(p) 2 σ ·p ˜ 1 |p| 2 ½ + |p| u(p) .152) are called left-handed (negative chirality) spinors. namely.100) would no longer commute with γ5 and in this case chirality would not be a good quantum number to label the states with. γ5 uL (p) = −uL (p). then the Dirac Hamiltonian (1. (3. We note that if the fermion is massive (m = 0).153) where we have deﬁned . vR (p) = PR v(p) = = |p| 2 1 2 1 2 1 (½ + γ5 )v(p) 2 ½ + σ ·p v(p) ˜ |p| .100 3 Properties of the Dirac equation are known as right-handed (positive chirality) spinors while those with the eigenvalue −1.

hχ(±) (p) = σ·p 1 2|p| 2 ½± ½± 1 1 = ± × 2 2 σ·p χ ˜ |p| 1 σ·p χ = ± χ(±) (p).3.6) that in the massless limit. we note that 2 spinors of the form χ(±) (p) = 1 σ·p 1± χ. This is connected with our earlier observation (see section 3.154) 2 2 We note that by deﬁnition these projection operators satisfy PR 2 = PR .91)) 1 2 PR = ½ ½ 1 ½ −½ . ˜ 2 |p| (3. the four component spinors can be eﬀectively described by two component spinors. 2 −½ ½ ½ ½ (3. (3.7 Chirality 101 1 1 PR = (½ + γ5 ). we will continue our discussion in the Pauli-Dirac representation of the Dirac matrices which we have used throughout.158) . PL = . From the deﬁnition of the helicity operator in (3. (In the Pauli-Dirac representation.157) correspond to states with deﬁnite helicity.119). PL = (½ − γ5 ). However.155) PR PL = 0 = PL PR . The reducibility of the spinors is best seen in the Weyl representation for the Dirac matrices discussed in (2. (3. the Dirac equation reduces to two decoupled two component Weyl equations (recall that it is the mass term which generally couples these two spinors). which implies that any four component spinor can be uniquely decomposed into a right-handed and a left-handed component. PL 2 = PL .153) that in the massless Dirac theory. PR + PL = 1. ˜ |p| 2 (3.156) We note from (3. namely. these projection operators have the explicit forms (see (2.121) (for the two component spinors S = 1 σ).

−ˆ ). vL (p) = −v (−) (p) v (−) (p) .117)) P (±) 0 1 2( P4×4 (±) = 0 P (±) 0 1 2( = ˆ ½ ± σ · p) 0 ˆ ½ ± σ · p) . therefore.163) . deﬁne projection operators into the space of positive and negative helicity two component spinors. ˆ 2 (3.153) are described by two component spinors with positive helicity while the lefthanded (four component) spinors are described in terms of two component spinors of negative helicity.102 3 Properties of the Dirac equation so that the right-handed (four component) spinors in (3. It is straightforward ˜ ˆ p to check that they satisfy the relations (P (+) )2 = P (+) . we see from (3. ˜ ˆ 2 (3.160) can also be written in a covariant notation as P (+) = 1 µ σ pµ . (P (−) )2 = P (−) . uL (p) = . P (+) + P (−) = 1.70) or (3.(3.159) We note here that the operators P (±) = 1 2 ½± σ·p 1 ˆ = (½ ± σ · p). vR (p) = . (3. (3. σ µ deﬁned in (2. They can be easily generalized to a reducible representation of operators acting on the four component spinors and have the form (see (2.153) and (3. P (+) P (−) = 0 = P (−) P (+) .162) and.161) with σ µ .157) that we can identify |p| 2 |p| 2 u(+) (p) u(+) (p) u(−) (p) −u(−) (p) |p| 2 |p| 2 v (+) (p) v (+) (p) uR (p) = .120) and pµ = (1. Explicitly. |p| 2 P (−) = 1 µ σ pµ .

164) which is the reason the spinors can be simultaneous eigenstates of chirality and helicity (when mass vanishes).163) that PL. we can choose u= ˜ 1 .159) as well as (3.166) For example.167). For completeness as well as for later use. (3.R . We note from (3. 0 v= ˜ 0 . For example. In fact. with the choice of the basis in (3. ˜˜ ˜˜ (3. (3.7 Chirality 103 and it is straightforward to check from (3. ˜ ˜ ˜ ˜ uu† + v v † = ½.3. v(+) (p) = ½ + σ · p v. P4×4 = 0. (±) (3.167) such that when p1 = p2 = 0. we can also deﬁne normalized spinors u(+) and v (+) . v (+) (p) = .165) 2 2 Each of these spinors is one dimensional and together they span the two dimensional spinor space. We can choose u and v to be ˜ ˜ normalized so that we have u(+) (p) = u† u = 1 = v † v .157) that we can write the positive and the negative energy solutions as (we will do this in detail for the right-handed spinors and only quote the results for the left-handed spinors) 1 1 ˆ ˜ ˆ ˜ ½ + σ · p u. let us derive some properties of these spinors. 1 (3. the helicity spinors simply reduce to eigenstates of σ3 .168) . Furthermore.163) we see that the right-handed spinors with chirality +1 are characterized by helicity +1 while the left-handed spinors with chirality −1 have helicity −1. the normalized spinors take the forms (here we are using the three dimensional notation so that pi = (p)i ) u(+) (p) = 1 2|p|(|p| + p3 ) 1 2|p|(|p| − p3 ) |p| + p3 p1 + ip2 p1 − ip2 |p| − p3 .156) and (3. from (3.

169) which can be checked from the explicit forms of the spinors in (3.159). as well as (3. We note that with p0 = |p|. we can write 1 0 pγ (½ + γ5 ) / 4 = 1 4 |p| 4 |p| 2 σ·p |p| −σ · p −|p| ˆ σ·p ½ 0 0 −½ ½ ½ ½ ½ = ½ ˆ σ·p ½ ½ ½ ½ ½ . (3. (3.171) = P (+) P (+) P (+) P (+) . u(+) (p)u(+)† (p) = v (+) (p)v (+)† (p) = 1 2 ˆ ½ + σ · p . R † u† (p)vR (−p) = vR (−p)uR (p) = 0.170) can be simpliﬁed by noting the following identity. the positive helicity spinors satisfy u(+)† (p)u(+) (p) = v (+)† (p)v (+) (p) = 1. we do not need to use any particular representation for our discussions. R † uR (p)u† (p) = vR (p)vR (p) = R |p| 2 P (+) P (+) P (+) P (+) . Here we note that the second relation follows from the fact that a positive helicity spinor changes into an orthogonal negative helicity spinor when the direction of the momentum is reversed (which is also manifest in the projection operator).170) The completeness relation in (3. In general. u(+)† (p)v (+) (−p) = 0 = v (+)† (−p)u(+) (p). Given the form of the right-handed spinors in (3.169) it now follows in a straightforward manner that † u† (p)uR (p) = |p| = vR (p)vR (p).168).104 3 Properties of the Dirac equation However. (3.

L † u† (p)vL (−p) = 0 = vL (−p)uL (p). 2m E + m E+m (3.4. ˜ 2m σ·p E+m σ·p u= ˜ uL (p).170) as † uR (p)u† (p) = vR (p)vR (p) = R 1 0 pγ (½ + γ5 ). / 4 (3. L † uL (p)u† (p) = vL (p)vL (p) = L 1 0 pγ (½ − γ5 ).3. (3.176) .175) In (3.174) we have deﬁned uL (p) = uS (p) = E+m u.8 Non-relativistic limit of the Dirac equation Let us recall that the positive energy solutions of the Dirac equation have the form (see (2. We conclude this section by noting (without going into details) that similar relations can be derived for the left-handed spinors and take the forms † u† (p)uL (p) = |p| = vL (p)vL (p).172) which can also be derived using the methods in section 3. (3.174) while the negative energy solutions have the form σ·p ˜ −E + m v v ˜ vS (p) vL (p) u− (p) = E+m 2m = .49)) u ˜ σ·p ˜ E+mu uL (p) uS (p) u+ (p) = E+m 2m = .173) 3. / 4 (3.8 Non-relativistic limit of the Dirac equation 105 so that we can write the completeness relation in (3.

in (3. Similarly. (α · p + βm) u+ (p) = Eu+ (p). (3. ˜ 2m σ·p E +m σ·p v=− ˜ vL (p). (3.106 3 Properties of the Dirac equation and we emphasize that the subscript “L” here does not stand for the left-handed particles introduced in the last section. when |p | ≪ m. σ · p uL (p) = (E + m)uS (p). which satisfy the equation Hu+ (p) = Eu+ (p). In the non-relativistic limit.179) We note that the second equation in (3. uL (p) and uS (p) are known as the large and the small components of the positive energy Dirac solution. vL (p) and vS (p) are also known as the large and the small components of the negative energy solution. E+m uS (p) = (3. m½ σ · p −m½ σ·p uL (p) uS (p) =E uL (p) uS (p) . or.174). Similarly.175) we have denoted vL (p) = vS (p) = − E+m v.180) . we expect the large components to give the dominant contribution to the wave function.179) gives the relation σ·p uL (p). or. 2m E + m E+m (3. Let us next look at the positive energy solutions in (3.178) This would lead to the two (2-component) equations σ · p uS (p) = (E − m)uL (p).177) It is clear that in the non-relativistic limit. the component uS (p) is much smaller than (of the order of v ) uL (p) and c correspondingly.

equation (3.182) (3. what we know for a free non-relativistic electron (spin 1 particle).3. E ≈ m (recall that we have set c = 1). then.183) Namely.9 Electron in an external magnetic ﬁeld The coupling of a charged particle to an external electromagnetic ﬁeld can be achieved through what is conventionally known as the minimal coupling. (3. therefore. This is. |p| ≪ m. E+m (3. with the substitution of this.184) where e denotes the charge of the particle and Aµ represents the four vector potential of the associated electromagnetic ﬁeld.181) has the form p2 uL (p) = ENR uL (p). if we identify the non-relativistic energy (without the rest mass term) as ENR = E − m.33) and remember that we are choosing = 1) . 2 3. 2m where we have used the fact that for a non-relativistic system. Since the coordinate representation of pµ is given by (see (1. 2m (3. the ﬁrst equation in (3.179) takes the form (σ · p) or.9 Electron in an external magnetic field 107 while.181) p2 uL (p) ≃ (E − m)uL (p). σ·p uL (p) = (E − m)uL (p). of course. and. the Dirac equation in this case reduces to the Schr¨dinger o equation for a two component spinor which we are familiar with. Furthermore. This preserves the gauge invariance associated with the Maxwell’s equations and corresponds to deﬁning pµ → pµ − eAµ .

The Dirac equation for the positive energy electrons.187) where we are assuming that A = A(x). In this case. we have used |p| ≪ m in the non-relativistic limit.185) the minimal coupling prescription also corresponds to deﬁning (in the coordinate representation) ∂µ → ∂µ + ieAµ . In this case.189) uS (p) = (3. we obtain . takes the form (α · (p − eA ) + βm) u(p) = Eu(p). in this case. σ · (p − eA ) uL (p) = (E + m)uS (p). (3. or. B = (∇ × A ) . Substituting this back into the ﬁrst equation in (3. (3.188) Explicitly.189) leads to σ · (p − eA) σ · (p − eA) uL (p) ≃ uL (p).189).108 3 Properties of the Dirac equation pµ → i∂µ . we have A0 = 0 = φ. (3. we can write the two (2-component) equations as σ · (p − eA ) uS (p) = (E − m)uL (p).190) where in the last relation. E+m 2m (3. the second equation in (3.186) Let us next consider an electron interacting with a time independent external magnetic ﬁeld. (3. m½ σ · (p − eA ) σ · (p − eA ) −m½ uL (p) uS (p) =E uL (p) uS (p) .

we are going to use purely three dimensional notation for simplicity) (p × A + A × p )i = ǫijk (pj Ak + Aj pk ) = ǫijk (pj Ak − Ak pj ) = ǫijk [pj . (3. (3. 1 e (p − eA)2 − σ · B uL (p) ≃ ENR uL (p). in the non-relativistic limit.192) Note that (here. 2m or.191) using the following identity for the Pauli matrices (σ · (p − eA)) (σ · (p − eA)) = (p − eA) · (p − eA ) + iσ · ((p − eA) × (p − eA)) = (p − eA)2 − ieσ · (p × A + A × p) . Ak ] = −iǫijk [∇j . Ak ] = −i (∇ × A )i = −i(B)i .193) = (p − eA )2 − ieσ · (−iB) (3.194) Consequently.195) 2m 2m where we have identiﬁed (as before) . 2m (3.3. when we can approximate the Dirac equation by that satisﬁed by the two component spinor uL (p). We can use this in (3.9 Electron in an external magnetic field 109 (σ · (p − eA)) (σ · (p − eA)) uL (p) ≃ (E − m)uL (p).191) Let us simplify the expression on the left hand side of (3. (3. equation (3.191) takes the form 1 (p − eA)2 − eσ · B uL (p) = (E − m)uL (p).192) to write (σ · (p − eA)) (σ · (p − eA)) = (p − eA )2 − eσ · B.

(3. this shows that a point Dirac particle has a magnetic moment corresponding to a gyro-magnetic ratio g = 2. a minimally coupled Dirac particle automatically leads.196) We recognize (3. (3. can have g-factors quite diﬀerent from 2. Thus. however. In this case. for example. in the non-relativistic limit.195) to be the Schr¨dinger equation for a charged o electron with a minimal coupling to an external vector ﬁeld along with a magnetic dipole interaction with the external magnetic ﬁeld. comparing with (3.197) Of course.197) we obtain g = 2. one says that there is an anomalous contribution to the magnetic moment.198) Let us recall that the magnetic moment of a particle is deﬁned to be (c = 1) e S. we have to add such an interaction by hand) and we can identify the magnetic moment operator associated with the electron to correspond to e σ. change this value slightly and the experimental deviation of g from the value of 2 (g − 2 experiment) for the electron agrees exceptionally well with the theoretical predictions of quantum electrodynamics. Particles with a nontrivial structure (that is particles which are not point like and have extended structures). 2m µ=g (3. Quantum mechanical corrections (higher order corrections) in an interacting theory such as quantum electrodynamics. 2m µ= (3. however. we know that the magnetic moments are given by . to a magnetic dipole interaction (recall that in the non-relativistic theory. for the proton and the neutron.199) 1 Since S = 2 σ for a two component electron.110 3 Properties of the Dirac equation ENR = E − m. Namely.

3. This is commonly known as the Pauli coupling or the Pauli interaction. this method runs into diﬃculty.203) eκ µν σ Fµν . (3.202) denotes the electromagnetic ﬁeld strength tensor and κ represents the anomalous magnetic moment of the particle. In the non-relativistic limit. (3.10 Foldy-Wouthuysen transformation In the last two sections. we have described how the non-relativistic limit of a Dirac theory can be taken in a simple manner.79 µnm . at higher orders. the relevant expansion parameter is |p| and m the method works quite well in the lowest order of expansion. 2m (3. 2mP µnm = (3. 3. The process of eliminating the “small” .10 Foldy-Wouthuysen transformation 111 µN µP = −1. as we have seen explicitly. the Hamiltonian becomes non-Hermitian).200) where the nuclear magneton is deﬁned to be |e| . = 2.201) with mP denoting the mass of the proton.91 µnm . if we were to calculate the electric dipole interaction of an electron in a background electromagnetic ﬁeld using the method described in the earlier sections. the electric dipole 1 moment becomes imaginary at order m2 (namely. However. For example. Anomalous magnetic moments can be accommodated through an additional interaction Hamiltonian (in the Dirac system) of the form (this is known as a non-minimal coupling) HI = where Fµν = ∂µ Aν − ∂ν Aµ . This puzzling feature can be understood in a simple manner as follows.

206) It is clear from the form of the matrix in (3.204) where the matrix A. In this case. for example.180)) σ·p .100) as well as (1. Thus. let us look at the free Dirac theory where we know that the Hamiltonian has the form (see (1.207) Let us next look for a unitary transformation that will diagonalize the Hamiltonian in (3. for example. (3. the main idea in the works of Foldy-Wouthuysen as well as Tani is to ensure that the relevant transformation used in going to the non-relativistic limit is manifestly unitary.101)) H = α · p + βm = γ 0 (γ · p + m) .206) is the source of the problem in taking the non-relativistic limit consistently.207). has the form (see (3. Since the lack of unitarity in (3. in the case of the free Dirac equation. E+m A= (3. such a transformation would . The matrix T that takes us to the two component “large” spinors in (3. This diﬃculty in taking a consistent non-relativistic limit 1 to any order in the expansion in m was successfully solved by Foldy and Wouthuysen and also independently by Tani which we describe below. (3.206) that it is not unitary and this is the reason that the Hamiltonian becomes non-Hermitian at higher orders in the inverse mass expansion (non-relativistic expansion).205) for the positive energy spinors.112 3 Properties of the Dirac equation components from the Dirac equation described in the earlier sections can be understood mathematically as uL (p) uS (p) uL (p) AuL (p) uL (p) 0 u(p) = = −→ T . (3.204) has the form T = ½ −A 0 ½ .

(3. (3.83) or (1.10 Foldy-Wouthuysen transformation 113 also transform the spinor into two 2-component spinors that will be decoupled and we do not have to eliminate one in favor of the other (namely. γ · pθ 2m 2n+1 = cos γ ·p sin |p| |p|θ 2m (3.209) From the properties of the gamma matrices in (1. Let us consider a transformation of the type U (θ) = e 2m γ ·p θ . (γ · p)2 = −p2 = −|p|2 . we note that (γ)† = −γ.3. m).208) where the real scalar parameter of the transformation is a function of p and m.91). θ = θ(|p|.212) .210) and using this we can simplify and write U (θ) = ∞ n=0 1 (2n)! |p|θ 2m γ · pθ 2m + 2n + 1 (2n + 1)! . (3. avoid the problem with “large” and “small” spinors).211) It follows now that U † (θ) = cos which leads to |p|θ 2m − γ ·p sin |p| |p|θ 2m . 1 (3.

214) |p| |p|θ 2m |p|θ m 2 × cos = γ 0 cos × cos = = = γ ·p sin |p| γ ·p sin |p| γ ·p sin |p| |p|θ 2m |p|θ m (γ · p + m) − γ 0 (γ · p + m) cos γ 0 (γ · p + m) cos γ0 m cos + |p| cos |p|θ m |p|θ m − − + |p| sin − m sin .114 3 Properties of the Dirac equation U (θ)U † (θ) = cos |p|θ 2m + |p|θ 2m − γ·p sin |p| − |p|θ 2m |p|θ 2m |p|θ 2m (3.208).208) is indeed unitary. Namely. Under the unitary transformation (3. the free Dirac Hamiltonian (3. (3.213) × cos = cos 2 γ·p sin |p| 2 |p|θ 2m |p|θ 2m γ ·p |p| sin2 = cos2 + sin2 |p|θ 2m = ½.207) would transform as H → H ′ = U (θ)HU † (θ) = cos |p|θ 2m + |p|θ 2m |p|θ 2m |p|θ 2m γ·p sin |p| − − |p|θ 2m γ 0 (γ · p + m) |p|θ 2m |p|θ 2m |p|θ 2m γ ·p sin |p| γ ·p sin |p| |p|θ m |p|θ m γ·p . the transformation (3.

tan θ= − m sin |p| . However.218) would be decoupled in the energy eigenvalue equation and we can without any diﬃculty restrict ourselves to the positive energy sector where the energy eigenvalue equation takes the form p2 + m2 u′ (p) = Eu′ (p). our discussion has been quite general and the parameter of the transformation.217) We see from (3.215) In this case.216) which. leads to the diagonalized Hamiltonian m2 p2 + m2 p2 p2 + m2 H ′ = γ0 + = γ0 p2 + m2 . (3. we have |p|θ m m p2 + m2 |p|θ m |p| cos = . from (3. if we want the transformation to diagonalize the Hamiltonian.214). (3. m |p| m . it is clear from (3. 1 1 (3.3. the two components of the transformed spinor u′ (p) 1 u′ (p) 2 u′ (p) = U (θ)u(p) = . sin = p2 + m2 . As a result.10 Foldy-Wouthuysen transformation 115 So far.217) that the Hamiltonian is now diagonalized in the positive and the negative energy spaces.214) that we can choose the parameter of transformation to satisfy |p|θ m = |p|θ m |p| cos or. or. θ. = 0. has been arbitrary. |p|θ m m arctan |p| (3. (3.219) .

215).214) will have the form . = 0. the unitary transformation in (3. (3. the ultrarelativistic limit |p| ≫ m.220) UFW (θ) = cos + which has a natural non-relativistic expansion in powers of |p| .211) takes the form 1 arctan 2 |p| m γ·p sin |p| 1 arctan 2 |p| .211) is also quite useful. the transformed Hamiltonian (3. In this case.214) that if we choose the parameter of transformation to satisfy |p|θ m |p|θ m |p|θ m m cos or. in this case. θ=− m arctan |p| (3.116 3 Properties of the Dirac equation For |p| ≪ m.183) to 1 the lowest order and it can be expanded to any order in m without any problem. There is a second limit of the Dirac equation. This m analysis can be generalized even in the presence of interactions and the higher order terms in the interaction Hamiltonian are all well behaved without any problem of non-hermiticity.221) this would lead to |p|θ m |p|θ m |p| cos sin = = − p2 + m2 m . tan + |p| sin =− m . We also note that with the parameter θ determined in (3. m (3. |p| m |p| . for which the generalized Foldy-Wouthuysen transformation (3. . namely.222) p2 + m2 As a result. or. this leads to the non-relativistic equation in (3. the transformation is known as the Cini-Touschek transformation and is obtained as follows. Let us note from (3.

where operators carry time dependence and states are time independent. 3.207).224) m which clearly has a natural expansion in powers of |p| (ultrarelativistic expansion). For example.223) p2 + m2 γ·p= |p| p2 + m2 α · p.220) as transforming away the α · p term in the Hamiltonian (3.11 Zitterbewegung The presence of negative energy solutions for the Dirac equation leads to various interesting consequences.225) In the Heisenberg picture. we can think of the Foldy-Wouthuysen transformation (3.3.207) while the Cini-Touschek transformation rotates away the mass term βm from the Hamiltonian (3. let us consider the free Dirac Hamiltonian (1. the unitary transformation (3. In fact. (3. Therefore. in this case. the Heisenberg equations of motion take the forms ( = 1) . |p| (3.211) has the form UCT (θ) = cos 1 arctan 2 m |p| − γ ·p sin |p| 1 arctan 2 m .100) H = α · p + βm.11 Zitterbewegung 117 H ′ = γ0 = γ0 p2 p2 + m2 + m2 p2 + m2 γ·p |p| (3. |p| m which has a natural expansion in powers of |p| .

on the other hand.101) we conclude that [α. α(t) = eiHt α e−iHt . using (1. for example. even though the momentum of a free particle is a constant of motion.118 3 Properties of the Dirac equation 1 i 1 [x . it follows that the . The second equation in (3. Secondly.228) (3. pj ] = i i 1 j ij α (iδ ) = αi .230) In other words. As a result. the velocity is not. α p ] = αj [xi .229) [α(t). since the eigenvalues of α are ±1 (see.226) shows that the momentum is a constant of motion as it should be for a free particle. Furthermore. by deﬁnition. αj pj + βm] = 0.226) Here a dot denotes diﬀerentiation with respect to time. it follows that (3. (1. (3. The ﬁrst equation. (3. i i ˙ or. xi = ˙ 1 1 i [p . identiﬁes α(t) with the velocity operator. α · p + βm] = 0. H] = 0. H] = [pi . pi = ˙ (3. = i ˙ or. p = 0. αj pj + βm] i i 1 1 i j j [x .227) where we have denoted the operator in the Schr¨dinger picture by o α(0) = α. H] = [α. H] = [xi . Let us recall that. x = α.101)).

it follows that eiHt αe−iHt eiHt |ψ = λ eiHt |ψ . We note from Heisenberg’s equations of motion that the time derivative of the velocity operator is given by 1 [α(t). we note that if α|ψ = λ|ψ .231) where λ denotes the eigenvalue of the velocity operator α.232) shows that the eigenvalues of α(t) are the same as those of α (only the eigenfunctions are transformed) and.102)) αi . or. α(t)|ψ ′ = λ |ψ ′ . (3. αj αi . β = 2δij ½. (3. = 0. These peculiarities of the relativistic theory can be understood as follows.232) or. More explicitly.11 Zitterbewegung 119 eigenvalues of α(t) are ±1 as well. = λ eiHt |ψ . H = −2iα(t)H + 2ip. therefore. (3.233) Equation (3. then.235) + + . are ±1. α(t) eiHt |ψ where we have identiﬁed |ψ ′ = eiHt |ψ . This is easily understood from the fact that the eigenvalues of an operator do not change under a unitary transformation.234) Here we have used the relations (see (1. This would seem to imply that the velocity of an electron is equal to the speed of light which is unacceptable even classically. (3. H] i + ˙ α(t) = = −i 2α(t)H − α(t). since the electron is a massive particle. (3.3.

E mγ (3.238) Furthermore.241) p = mγv. Substituting this back into (3.120 3 Properties of the Dirac equation as well as the fact that momentum commutes with the Hamiltonian. so that mγv p = = v. We note that.239) is quite expected.237) (3.236). which is unexpected. For example. α(t) = pH −1 + (α(0) − pH −1 )e−2iHt . however.234).236) (3. On the other hand. (3. (3. α(t) = α(0)e−2iHt . diﬀerentiating (3. + α(0) − α(t) = H H The ﬁrst term in (3. p −2iHt p e . Let us note next that both p and H are constants of motion. we obtain ˙ α(t) = −2i(α(0) − p H −1 )He−2iHt . we obtain ¨ ˙ α(t) = −2iα(t)H. or. we ﬁnally determine −2i(α(0) − p H −1 )He−2iHt = −2iα(t)H + 2ip. It represents an additional component to the . in an p eigenstate of momentum it would have the form E which is the true relativistic expression for velocity. using this relation in (3. ˙ ˙ or. (c = 1) E = mγ. It is the second term. Therefore.234) with respect to time. (3.239) or. for a relativistic particle.234) we have ˙ α(0) = −2iα(0)H + 2ip = −2i(α(0) − pH −1 )H.239).240) which is the ﬁrst term in (3. from (3.

243) where a is a constant.11 Zitterbewegung 121 velocity which is oscillating at a very high frequency (for an electron at rest. The unconventional operator relations in (3.244) u† (p) α(0) − E+m = 2m = p u(p) H u† ˜ u ˜ σ·p ˜ σ E+mu p σ·p ˜ ˜ −E u + σ E + m u σ·p u† E + m ˜ p σ·p σu − E E + m u ˜ ˜ u† ˜ σ·p E+m σ p −E p −E E+m † u ˜ 2m . The ﬁrst two terms in (3. since there is no potential whatsoever in the problem. Let us also note from (3.5MeV corresponding to a frequency of the order of 1021 /sec. However.243) can be shown in the Schr¨dinger picture to arise from the presence of o negative energy solutions. In fact. the energy is ≈ .239) and (3.3. H x(t) = a + α(0) − (3. the third term represents an additional contribution to the electron trajectory which is oscillatory with a very high frequency. + α(0) − H H (3. for example.226) that since ˙ x(t) = α(t) = p −2iHt p e .243) are again what we will expect classically for uniform motion. it is easy to check that for a positive energy electron state E+m 2m u ˜ . Its occurrence is quite surprising. we obtain i p t+ H 2 p H −1 e−2iHt .) and gives a time dependence to α(t).242) integrating this over time. This quivering motion of the electron was ﬁrst studied by Schr¨dinger o and is known as Zitterbewegung (“jittery motion”). σ·p ˜ E+mu u(p) = we have (3.

Relativistic Quantum Mechanics. 1968. 1980. Quantum Field Theory. 1964. This shows that even though the eigenvalues of the operator α(t) are ±1 corresponding to motion with the speed of light. New York. ˜ ˜ This shows that even though the operator relations are unconventional. (3. New York. 3.) 3. McGraw-Hill. Itzykson and J-B. . C. New York. This also shows that the eigenstates of the velocity operator. (This makes clear that neglecting the negative energy solutions of the Dirac equation would lead to inconsistencies. α(t).246) as we should expect from the Ehrenfest theorem.12 References 1. Zuber. 2. Schiﬀ. McGraw-Hill. Bjorken and S. the physical velocity of the electron (observed velocity which is the expectation value of the operator in the positive energy electron state) is what we would expect. McGrawHill. D. which are not simultaneous eigenstates of the Hamiltonian must necessarily contain both positive and negative energy solutions as superposition and that the extra terms have non-zero value only in the transition between a positive energy and a negative energy state. I.122 = E+m 2m − +˜† u = = E+m 2m E+m 2m − − 3 Properties of the Dirac equation σ·p p † u u + u† σ ˜ ˜ ˜ u ˜ E E+m σ·p (σ · p)2 p σ˜ − u† u ˜ u ˜ E+m E (E + m)2 p E 1+ p2 (E + m)2 + 2p E+m u† u ˜ ˜ (3. Quantum Mechanics. in a positive energy electron state p H α(t) + = + .245) 2E 2p p × + E E+m E+m u† u = 0. L. Drell. J.

Okubo.3. Physical Review 78. L. 422 (1958). L. Foldy and S. Progress of Theoretical Physics 12. 7.12 References 123 4. Cini and B. S. Touschek. Progress of Theoretical Physics 6. 102 (1954). A. Wouthuysen. 29 (1950). . 6. 12. 603 (1954). M. Tani. Nuovo Cimento 7. 5. S. ibid. 267 (1951).

This page intentionally left blank .

the symmetry algebras can be easily obtained from the coordinate representation of the symmetry generators and that is the approach we will follow in our discussions. 3. let us start with rotations which we have already discussed brieﬂy in the last chapter. therefore. we will study the symmetry algebras of the Lorentz and the Poincar´ groups as well as their representations e which are essential in constructing physical theories. for space-time symmetries.Chapter 4 Representations of Lorentz and Poincar´ e groups 4. i. Let us consider an arbitrary. 2. inﬁnitesimal three di- mensional rotation of the form δxi = ǫi jk xj αk .1) where αk represents the inﬁnitesimal constant parameter of rotation around the k-th axis (there are three of them). it is suﬃcient to study the behavior of inﬁnitesimal transformations since any ﬁnite transformation can be built out of inﬁnitesimal transformations. In addition. experimentally we know that space-time translations also deﬁne a symmetry of physical theories. should be invariant under Lorentz transformations. But. In studying the symmetry algebras of continuous symmetry transformations. In this chapter. k = 1. j. (4.1 Rotation. 4. It is worth noting here that. Furthermore.1 Symmetry algebras Relativistic theories. (Let us recall our 125 .1. the symmetry algebra associated with a continuous symmetry group is given by the algebra of the generators of inﬁnitesimal transformations. as we have discussed.

) ω i j = ǫi jk αk . (4. (4. Let us next deﬁne an inﬁnitesimal vector operator (also known as the tangent vector ﬁeld operator) for rotations (an operator in the coordinate basis) of the form 1 1 ij ω Mij = ω ij (xi ∂j − xj ∂i ) 2 2 (4. we can write the inﬁnitesimal rotations also as δxi = −R(ω)xi = ω i j xj .) If we now identify (In the last chapter we had denoted the inﬁnitesimal transformation matrices by ǫi j . ǫi jk = η iℓ ǫℓjk . (4. the form of the rotation that we had discussed in the last chapter.35) here for clarity. then.34) and (1. we note that ω ij = −ω ji .2) and that the inﬁnitesimal rotation around the k-th axis can also be represented in the form (taking place in the i-j plane) δxi = ω i j xj .7) . ǫµν .126 4 Representations of Lorentz and Poincar´ groups e notation from (1.6) In other words. etc.3) (4. of course. It follows now that i R(ω)xi = −ω kj xj ∂k xi = −ω kj xj δk = −ω ij xj = −ω i j xj . (4. ǫijk denotes the three dimensional Levi-Civita tensor with ǫ123 = 1. but here we denote them by ω ij .4) This is.5) R(ω) = = ω ij xi ∂j = −ω ij xj ∂i . ω µν in order to avoid confusion.

(4.8) = R(ω). ω]ij .10) we can obtain the algebra satisﬁed by the generators of inﬁnitesimal rotations. we note that R(ω). The Lie algebra of the group of rotations can be obtained from the algebra of the vector operators themselves. R(ω) = ω ij xi ∂j . are known as the generators of inﬁnitesimal rotations. ω kℓ xk ∂ℓ = ω ij ω kℓ [xi ∂j . Mij .8). where we have identiﬁed ω ij = ω ik ω kj − ω ik ωkj = −ω ji = [ω. xk ∂ℓ ] = ω ij ω kℓ (xi [∂j . they give back a rotation. namely. we can calculate them directly as . R(ω) in (4. from the algebra of the vector operators in (4.9) Namely. Such an algebra is called a non-Abelian (non-commutative) algebra. the vector operator. 2 (4. Mij .5). Thus. rather. R(ω) = 1 ij ω Mij .4. Using the form of R(ω) in (4.5) generates inﬁnitesimal rotations and the operators. Alternatively. xk ] ∂ℓ + xk [xi .1 Symmetry algebras 127 Namely. two rotations do not commute. ∂ℓ ] ∂j ) = ω ij ω kℓ (ηjk xi ∂ℓ − ηiℓ xk ∂j ) = ω ij ω jℓ xi ∂ℓ − ω ij ω ki xk ∂j = ω ij ωjk − ω ij ωjk xi ∂k (4.

xk ∂ℓ − xℓ ∂k ] = [xi ∂j . But up to a rescaling. Mrs ] 4 i j 1 = ǫi pq ǫjrs (−ηpr Mqs − ηqs Mpr + ηps Mqr + ηqr Mps ) 4 1 1 = − ǫir q ǫjrs Mqs − ǫi p s ǫjrs Mpr 4 4 . r. (4. (4. Mkℓ ] = [xi ∂j − xj ∂i . then. Jj ] = = 1 pq rs ǫ ǫ [Mpq .11) This is the Lie algebra for the group of rotations.12) 1 1 Ji = − ǫi jk Mjk = − ǫijk (xj ∂k − xk ∂j ). then we may deﬁne the operators. 2 2 which gives the familiar orbital angular momentum operators. with a factor of “i”. 2. Mij . xk ∂ℓ ] + [xj ∂i . we obtain (p. (4. s = 1. xℓ ∂k ] − [xj ∂i . To obtain the familiar algebra of the angular momentum operators. xℓ ∂k ] = (ηjk xi ∂ℓ − ηiℓ xk ∂j ) − (ηjℓ xi ∂k − ηik xℓ ∂j ) − (ηik xj ∂ℓ − ηjℓ xk ∂i ) + (ηiℓ xj ∂k − ηjk xℓ ∂i ) = −ηik (xj ∂ℓ − xℓ ∂j ) − ηjℓ (xi ∂k − xk ∂i ) +ηiℓ (xj ∂k − xk ∂j ) + ηjk (xi ∂ℓ − xℓ ∂i ) = −ηik Mjℓ − ηjℓ Mik + ηiℓ Mjk + ηjk Miℓ . If we would like the generators to be Hermitian quantum mechanical operators corresponding to a unitary representation. q. = (xi ∂j − xj ∂i ) = ǫijk Jk . ǫjrs Mrs 2 2 [Ji .128 4 Representations of Lorentz and Poincar´ groups e [Mij . 3) 1 pq 1 ǫi Mpq . we note that we can deﬁne (recall that in the four vector notation Ji = −(J)i ) Mij or. Using this.11) represents the Lie algebra of the group SO(3) or equivalently SU (2). xk ∂ℓ ] − [xi ∂j .

let us note that a constant in- ﬁnitesimal space-time translation of the form δxµ = ǫµ . leads to (see (4.17) (4.1 Symmetry algebras 129 1 p rs ǫ ǫ Mps 4 i r j ǫi q r ǫjsr Mqs 1 q sr ǫ ǫ Mqs 4 i r j (4. 4.12)) 1 q ǫri ǫjsr Mqs = − ǫijr ǫrsq Mqs = −ǫijr Jr . ǫri q ǫjsr + ǫrsq ǫij r + ǫrjq ǫsi r = 0. so that µ R(ǫ)xµ = −ǫν ∂ν xµ = −ǫν δν = −ǫµ . the Lie algebra of SO(3) or SU (2) (or the familiar algebra of angular momentum operators) up to a rescaling. namely. (4. can be generated by the inﬁnitesimal vector operator R(ǫ) = −ǫµ Pµ = −ǫµ ∂µ . In the same spirit.11) or (4. (4. then.16) (4.18) . Jj ] = ǫij r Jr .14) which. 2 (4.1. where in the last step we have used the Jacobi identity for the structure constants of SO(3) or SU (2) (or the identity satisﬁed by the Levi-Civita tensors).4.13) is. = −ǫri q ǫjsr Mqs . [Ji .15) The algebra of the generators in (4. of course.2 Translation.13) 1 + ǫis q ǫjrs Mqr + 4 1 1 = − ǫriq ǫjsr Mqs − 4 4 1 − ǫri q ǫjsr Mqs − 4 or.

21) Namely. µ.20) In other words. (4. Pν ] = [∂µ .19) The Lie algebra associated with translations is then obtained from R(ǫ).18). (4. ǫ ν ∂ν ] = ǫµ ǫ ν [∂µ . let us note that if we deﬁne an inﬁnitesimal vector operator 1 µν 1 ω Mµν = ω µν (xµ ∂ν − xν ∂µ ) 2 2 µν = ω xµ ∂ν = −ω µν xν ∂µ . ∂ν ] = 0. (4.1.22) where. the inﬁnitesimal. a proper Lorentz transformation can be thought of as a rotation in the four dimensional Minkowski space-time and has the inﬁnitesimal form δxµ = ω µν xν . constant parameters of transformation satisfy ω µν = −ω νµ .23) As in the case of rotations. ν = 0. as we have seen in (3. R(ω) = (4. R(ǫ) = [ǫµ ∂µ . (4.3 Lorentz transformation. 3. 2. 1.24) then.130 and we can write 4 Representations of Lorentz and Poincar´ groups e δxµ = −R(ǫ)xµ . we obtain . (4. As we have seen in the last chapter. 4. two translations commute and the corresponding relation for the generators is [Pµ . ∂ν ] = 0. translations form an Abelian (commuting) group while the three dimensional rotations form a non-Abelian group.

R(ω) = = ω µν ω λρ [xµ ∂ν .4.26) M0i = −Mi0 = x0 ∂i − xi ∂0 = Ki . we can think of R(ω) as the vector operator generating inﬁnitesimal proper Lorentz transformations and the operators. where. Thus. and the generators of inﬁnitesimal boosts with (4. (4.12)) Mij = −Mji = xi ∂j − xj ∂i = ǫij k Jk . as the generators of the inﬁnitesimal transformations. λ (4.1 Symmetry algebras 131 µ R(ω)xµ = −ω λν xν ∂λ xµ = −ω λν xν δλ = −ω µν xν = −ω µν xν = −δxµ . we can determine the group properties of the Lorentz transformations from the algebra of the vector operators generating the transformations. We also note that we can identify the inﬁnitesimal generators of spatial rotations with (see (4. ω λρ Mλρ 2 2 R(ω). ω]µν . as in the case of rotations.27) As before.29) .25) Therefore. 1 µν 1 ω Mµν . xλ ∂ρ ] = ω µν ω λρ (ηνλ xµ ∂ρ − ηµρ xλ ∂ν ) = ω µλ ω λρ xµ ∂ρ − ω µν ω λµ xλ ∂ν ω µλ ω λν − ωµλ ω λν xµ ∂ν (4. (4.28) = = ω µν xµ ∂ν = R(ω). we have ω µν = −ω νµ = ω µλ ω λν − ω µ ω λν = [ω. Mµν = −Mνµ .

three rotations and three boosts). This is what we had seen explicitly in (3. As we have seen these transformations correspond to rotations. Mλρ ] = [xµ ∂ν − xν ∂µ . in this case. xρ ∂λ ] = (ηνλ xµ ∂ρ − ηµρ xλ ∂ν ) − (ηνρ xµ ∂λ − ηµλ xρ ∂ν ) − (ηµλ xν ∂ρ − ηνρ xλ ∂µ ) + (ηµρ xν ∂λ − ηνλ xρ ∂µ ) = −ηµλ (xν ∂ρ − xρ ∂ν ) − ηνρ (xµ ∂λ − xλ ∂µ ) +ηµρ (xν ∂λ − xλ ∂ν ) + ηνλ (xµ ∂ρ − xρ ∂µ ) = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ . xλ ∂ρ ] − [xµ ∂ν .31) coincides exactly with the six generators we have (namely.132 4 Representations of Lorentz and Poincar´ groups e This shows that the algebra of the vector operators is closed and that Lorentz transformations deﬁne a non-Abelian group. therefore. In fact. The algebra of the generators can also be calculated directly and has the form [Mµν .) We end this section by pointing out that the algebra in (2. 2 (4. the matrices σµν provide a representation for the generators of the Lorentz group.69) in connection with the discussion of covariance of the Dirac equation. However.30) This. (Namely. the Lie algebra of the generators is isomorphic to that of the group SO(4). Lorentz transformations (boosts) are non-compact unlike rotations in Euclidean space.109) coincides with (4. up to a scaling. it is n(n−1) ) 2 1 × 4 × (4 − 1) = 2 × 3 = 6. therefore. since the rotations are in Minkowski space-time whose metric is not Euclidean it is more appropriate to identify the Lie algebra as that of the group SO(3. xλ ∂ρ ] + [xν ∂µ . in four dimensions and. (4. xρ ∂λ ] − [xν ∂µ . xλ ∂ρ − xρ ∂λ ] = [xµ ∂ν . This implies that. we note that the number of generators for SO(4) which is (for SO(n).30) (up to a scaling). gives the Lie algebra associated with Lorentz transformations. 1). .

µ µ R(ǫ. the general transformation of the coordinates takes the form δxµ = ǫµ + ω µν xν .4.33) then.32) are known as the (inﬁnitesimal) Poincar´ transformations or the inhomoe geneous Lorentz transformations. If. Namely. ω) = −ǫµ Pµ + 1 µν ω Mµν 2 = −ǫµ ∂µ + ω µν xµ ∂ν = −ǫµ ∂µ − ω µν xν ∂µ = − (ǫµ ∂µ + ω µν xν ∂µ ) . (4. we also consider inﬁnitesimal translations. The transformations in (4. Clearly. ǫ λ ∂λ + ω λρ xρ ∂λ = ǫµ ω λρ [∂µ . ω µν denote respectively the parameters of inﬁnitesimal translation and Lorentz transformation.1. (4. ω) . xρ ∂λ ] = ǫµ ω λρ ηµρ ∂λ + ω µν ǫ λ (−ηνλ ) ∂µ + ω µν ω λρ (ηµρ xν ∂λ − ηνλ xρ ∂µ ) = − (ω µν ǫ ν − ω µν ǫν ) ∂µ + ω µλ ωλν − ω µλ ωλ ν xµ ∂ν = − (ǫµ ∂µ + ω µν xν ∂µ ) = R (ǫ. (4. xρ ∂λ ] + ω µν ǫ λ [xν ∂µ . (4. ω).1 Symmetry algebras 133 4. acting on the coordinates. in this case. ω) = ǫµ ∂µ + ω µν xν ∂µ . in addition to inﬁnitesimal Lorentz e transformations. ∂λ ] + ω µν ω λρ [xν ∂µ .35) . if we deﬁne an inﬁnitesimal vector operator as R(ǫ.34) The algebra of the vector operators for the Poincar´ transformae tions can also be easily calculated as R(ǫ. it generates inﬁnitesimal Poincar´ e transformations.32) where ǫµ .4 Poincar´ transformation. ω)xµ = − ǫν ∂ν + ω λν xν ∂λ xµ = −(ǫν δν + ω λν xν δλ ) = −(ǫµ + ω µν xν ) = −δxµ . R(ǫ.

(4. [Pµ . The commutator of a generator (multiplied by the appropriate transformation parameter) with any operator gives the inﬁnitesimal change in that operator under the transformation generated by that particular generator.22) and (3.36) We can also calculate the algebra of the generators of Poincar´ group. (4.30). combining with our earlier results on the algebra of the translation group.21)). Pν ] (see (4. Note that [Pµ . for example. For change in the coordinate four vector under an inﬁnitesimal Lorentz transformation. (4.) Thus. Mνλ ] = ηµν Pλ − ηµλ Pν . Pν ] = 0. since the generae tors of translations do not commute with the generators of Lorentz .57). We note that the algebra of translations deﬁnes an Abelian subalgebra of the Poincar´ algebra (4.38).37) which simply shows that under a Lorentz transformation. ω µν = ω µλ ω λν − ω µλ ωλ ν = [ω.30) and (4. the only relation that needs to be calculated is the commutator between the generators of translation and Lorentz transformations. (This is seen by recalling that 2 ω µν Mµν generates inﬁnitesimal Lorentz transformations. Mλρ ] as well as [Pµ . Mνλ ] = [∂µ .134 4 Representations of Lorentz and Poincar´ groups e where we have identiﬁed ǫµ = (ω µν ǫ ν − ω µν ǫν ) . see. e We already know the commutation relations [Mµν .21). (4. xν ∂λ − xλ ∂ν ] = ηµν ∂λ − ηµλ ∂ν = ηµν Pλ − ηµλ Pν . However. Therefore. as well as the homogeneous Lorentz group. (4. (4.38) [Mµν . ω]µν . we conclude that the Lie algebra associated with the Poincar´ e transformations (inhomogeneous Lorentz group) is given by [Pµ . Mλρ ] = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ . Pµ behaves 1 like a covariant four vector.

Kj ] . 2.40) We recall from (4. e (4. Kj ] = [M0i . (4. Poincar´ algebra cannot be written as a direct sum e of those for translations and Lorentz transformations. M0j ] = −Mij = −ǫij k Jk . Poincar´ algebra = t4 ⊕ so(3. Jj ] = M0i .26) and (4.27) that we can identify the angular momentum and boost operators as 1 Ji = − ǫi jk Mjk . Mλρ ] = (−ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ ). [Ki .42) .4.2 Representations of the Lorentz group Let us next come back to the homogeneous Lorentz group and note that the Lie algebra in this case is given by (4. 3. 2 Ki = M0i . (4. i = 1. (4.30) [Mµν . (The general convention is to denote groups by capital letters while the algebras are represented by lower case letters. [Ji . Namely.41) Written out in terms of these generators.12).39) Rather. it is what is known as a semi-direct sum of the two algebras. (4. the Lorentz algebra takes the form [Ki .2 Representations of the Lorentz group 135 transformations. Jj ] = ǫij k Jk . − 1 kℓ ǫ Mkℓ 2 j 1 = − ǫjkℓ (−ηiℓ M0k + ηik M0ℓ ) 2 1 1 k ǫ Kk − ǫji ℓ Kℓ = 2 ji 2 = ǫij k Kk = [Ji .) 4. 1).

(4. if we deﬁne the generators with a factor of “i”. Ki . as can be seen from the hermiticity of the generators in the coordinate basis. (4. As a consequence of (4. (Ji − iKi ) . we have A† = −Ai . Ki = −i (Ai − Bi ) . inﬁnite dimensional representations are unitary. † Mµν = i(xµ ∂ν − xν ∂µ ) = Mµν . Ji† = −Ji . in choosing the coordinate representation for the generators.13) in the last relation. (4.136 4 Representations of Lorentz and Poincar´ groups e where we have used (4.43) which also leads to the inverse relations Ji = (Ai + Bi ) .46) Namely. namely. However.44) Parenthetically. This is a set of coupled commutation relations. consequently. (4. the generators in the new basis are all anti-Hermitian.47) . namely. (compared to Ji ) is connected with the fact that such transformations are non-compact and.42) that we can assign the following hermiticity properties to the generators. Ki† = Ki . the ﬁnite dimensional representations of boosts are non-unitary (hence the opposite Hermiticity of Ki ). we have not been particularly careful about choosing Hermitian operators.45). i † Bi = −Bi .45) This unconventional hermiticity for Ji arises because. (4. The opposite hermiticity property of the generators of boosts. Let us deﬁne a set of new generators as linear superpositions of Ji and Ki as (this is also known as changing the basis of the algebra) Ai = Bi = 1 2 1 2 (Ji + iKi ) . let us note from the form of the algebra in (4.

Therefore. (4. the Lorentz group is double valued (doubly connected). Kj ] + i [Ki .4.49) Incidentally.48) In other words. one says that the Lorentz algebra is isomorphic to the direct sum of two angular momentum algebras. Bj ] = = 1 ([Ji . (4. Mathematically.43). in this new basis. 2 ij 1 1 (Ji − iKi ) . Jj ] − [Ki . (Jj − iKj ) 2 2 [Ai .2 Representations of the Lorentz group 137 In the new basis (4. Jj ] + i [Ji . 2 ij 1 1 (Ji + iKi ) . as we have already seen in the last chapter. Kj ]) 4 1 ǫ k Jk − iǫij k Kk + iǫij k Kk − ǫij k Jk = 4 ij = 0. Jj ] − i [Ji . Jj ] − i [Ji .42) takes the form 1 1 (Ji + iKi ) . the algebra separates into two angular momentum algebras which are decoupled. the Lorentz algebra (4. Kj ] − i [Ki . so(3. Kj ]) 4 1 ǫ k Jk − iǫij k Kk − iǫij k Kk + ǫij k Jk 4 ij 1 k ǫ (Jk − iKk ) = ǫij k Bk . (Jj − iKj ) 2 2 1 ([Ji . Aj ] = = = = [Bi . Kj ] + i [Ki . 1) ≃ so(3) ⊕ so(3) ≃ su(2) ⊕ su(2). Jj ] − [Ki . it is more . Bj ] = = = = [Ai . Kj ]) 4 1 ǫ k Jk + iǫij k Kk + iǫij k Kk + ǫij k Jk 4 ij 1 k ǫ (Jk + iKk ) = ǫij k Ak . Jj ] + [Ki . (Jj + iKj ) 2 2 1 ([Ji .

. 2 A (4. we can write (This is the generalization of the S(Λ) matrix that we studied in (3. Denoting by jA and jB the eigenvalues of the Casimir operators A2 and B2 respectively for the two algebras.50) An irreducible nonunitary representation of the homogeneous Lorentz group. we have 3 1 jA = 0. Explicitly. .35) in connection with the covariance of the Dirac equation. (4.jB) (Λ)ψ (jA .51) where Λ represents the ﬁnite Lorentz transformation parameter. 1) which is SL(2. .j ) −i θ i Ji A B +δi Ki A B D (Λ) = e = e . . 2 2 3 1 jB = 0. (4.) (jA . The ﬁnite dimensional unitary representations of each of the angular momentum algebras is well known. .jB ) (just as the representation of the rotation group is denoted by D(j) ).jB ) (x) = D(jA ) (Λ)D(jB ) (Λ)ψ (jA . 2 2 (4.jB ) (x′ ) = D(jA . .53) .j ) (j . 1. this represents the operator implementing ﬁnite transformations on the Hilbert space of states or wave functions as ψ ′ (jA . ) to describe the Lorentz transformations.52) where the ﬁnite parameters of rotation and boost can be identiﬁed with θi = 1 i i θ + θB . 2 A δi = i i i θ − θB .jB ) “ ” (j ) (j ) i i −i θA Ai A +θB Bi B “ ” (j . .jB ) (x). . . . . . much the same way we consider the universal covering group SU (2) of SO(3) to describe rotations.138 4 Representations of Lorentz and Poincar´ groups e meaningful to consider the simply connected universal covering group of SO(3. therefore. 1. can be speciﬁed uniquely once we know the values of jA and jB and is labelled as D(jA . Namely.

we conclude that the values of the spin in a given representation characterized by (jA . These two representations are inequivalent and. 2 ) . D(0. For jA = jB = 0. For jA = jB = 2 . .0) . dimensionality = 1. in fact. dimensionality (4. 2 (4. jB = 0. . 2 (4. jB = 2 .44)) J = A + B. |jA − jB | + 1.55) Consequently. are complex conjugates of each other and can be identiﬁed to act on the wave functions of the two kinds of massless Dirac particles (Weyl 1 fermions) we had discussed in the last chapter. jB ) will have the dimensionality D (jA . For jA = 1 . . (4. jB ) can lie between j = |jA − jB |. 1 1 D(0.54) and its spin content follows from the fact that (see (4. 2 1 1 D( 2 .2 Representations of the Lorentz group 139 Such a representation labelled by (jA .jB ) = (2jA + 1) (2jB + 1) . jA + jB .56) The ﬁrst few low lying representations of the Lorentz group are as follows. j = .57) corresponds to a scalar representation with zero spin (which acts on the wave function of a Klein-Gordon particle).0) . . from our knowledge of the addition of angular momenta. 1 Similarly.4. . dimensionality = 2. (4. dimensionality = 2. j = .58) 1 corresponds to a two component spinor representation with spin 2 .59) also corresponds to a two component spinor representation with spin 1 2 . for jA = 0. (4. j = 0.

2 ) . ··· .) It may be puzzling as to where the four component Dirac spinor ﬁts into this description. It actually corresponds to a reducible representation of the Lorentz group of the form 1 1 1 D(0. Bi (0) 1 (1) Ai 2 = − σi . we now consider Hermitian generators by letting Mµν → iMµν as in (4.) From (4. 4. 2 ) ⊗ D ( 2 . (Namely. 2 1 1 ( ) Bi 2 = − σi . we have (We note here that the negative sign in the spin 2 k representation in (4.) Ai (0) = 0. 2 ··· . dimensionality = 2 × 2 = 4.47). Let us now con- struct explicitly a few of the low order representations for the generators of the Lorentz group. = 0. j = 0. Ai . 1.140 4 Representations of Lorentz and Poincar´ groups e D( 2 . Bi by a factor of i.63) . (4. j = .2. 2 (4. Ki . 2 (1) ( 1 .0) . dimensionality = 4. 2 (4.0) (0) = −i Ai 2 − Bi Ki 2 = i σi . this leads to the ﬁrst two nontrivial representations for the angular momentum and boost operators of the forms 1 ( 1 . namely.50).0) (1) (0) Ji 2 = Ai 2 + Bi = − σi .44).1 Similarity transformations and representations.62) Using (4.60) is known as a four component vector representation and can be identiﬁed with a spin content of 0 and 1 for the components.61) This discussion can similarly be carried over to higher dimensional representations. we note that for the ﬁrst few low order represen1 tations. we scale all the generators Ji .48). t and a spin 1 component x (under rotations) and the same is true for any other four vector.62) arises because ǫij = −ǫijk in (4. 1 1 (4. (Note that a four vector such as xµ has a spin zero component. To compare with the results that we had derived earlier.

65) would imply the existence of an invertible matrix S such that S −1 σi S = σi .0) 2 ⊕ Ki (0. 1 ) (0) Ji 2 = Ai + Bi 2 = − σi .71) and (3.98) (or (3. 2 (0.65) 1 then. we would say that the two representations 2 . (Dirac) Ki i 2 σi 0 0 i − 2 σi . if there exists a similarity transformation relating the two. Two representations are said to be equivalent. . if we can ﬁnd a similarity transformation S leading to (0. 0 and 0. 0 and 0. S −1 σi S = −σi .64) we see that the condition (4. 1 ) Ki 2 = S −1 Ki 2 S.63) and (4.2 Representations of the Lorentz group 141 and 1 (1) (0. They 2 provide the representations of angular momentum and boost for the left-handed and the right-handed Weyl particles.66) which is clearly impossible.0) Ji 2 = S −1 Ji 2 S.64) give the two inequivalent representations of dimensionality 2 as we have noted earlier. From (4. 1 are 2 equivalent. 2 (4. from (4. (4.78)). 2 are inequivalent representations. the two representations la1 belled by 1 . we note that these do not resemble the generators of the Lorentz algebra deﬁned in (3. In fact. For example.4. 1 ) ( 1 .64).64) Equations (4.0) = Ji 2 ⊕ Ji 2 = = Ki ( 1 .0) (0. we can obtain the representation of the Lorentz generators for the reducible four component Dirac spinors as (0. (4.69) and (2. 1 ) ( 1 . 1 ) 2 = 1 − 2 σi Ji (Dirac) 0 1 − 2 σi 0 . (4.63) and (4. 1 ) (1) (0) Ki 2 = −i Ai − Bi 2 i = − σi . ( 1 .67) However.63) and (4. Therefore.

2 = S −1 Ki = 1 2 (Dirac) S i 2 ½ ½ −½ ½ 0 i −2 σi 0 = 0 i −2 σi ½ −½ ½ ½ (4.67) and in our earlier discussion in (3.142 4 Representations of Lorentz and Poincar´ groups e This puzzle can be understood as follows. (4. we can apply the inverse similarity transformation in (2. we note that the generators in (4.98) (see also (3. if we would like the generators in the standard Pauli-Dirac representation (which is what we had used in our earlier discussions).71) and (3.70) = σi i − 2 σi 0 1 PD σ . On the other hand.69) and (2. we note that the angular momentum and boost operators in (4.121) to obtain Mµν = Ji (Dirac) PD = S −1 Ji = 1 2 (Dirac) S − 1 σi 0 2 1 0 − 2 σi 0 σi ½ ½ −½ ½ − 1 σi 2 0 −1 2 ½ −½ ½ ½ = Ki (Dirac) PD 1 PD = − ǫi jk σjk . i 0 i γ . γW = ǫijk 2 W .γ = 2 W W −iσi 0 σk 0 0 iσi 0 σk 0i σW = .68) As a result. 2 0i Therefore. We note that in the Weyl representation for the gamma matrices deﬁned in (2. (4.78)) are equivalent since they are connected by a similarity transformation that .69) 2 µν and. consequently.119). ij σW = i i j γ . give a representation of the Lorentz generators in the Weyl representation.67) are obtained from 1 W σ .

Let us illustrate this for the simpler case of rotations. as we discussed earlier. On the other hand. there must be a similarity transformation relating the two representations. therefore.2 Representations of the Lorentz group 143 relates the Weyl representation of the Dirac matrices in (2. we can deduce the matrix form of the corresponding generators to be . The case for Lorentz transformations follows in a parallel manner. we can determine a representation for the generators of the Lorentz transformations belonging to the representation for the four vectors.4) the matrix structure of the generator J3 to be 0 0 i 0 0 0 J3 = −i 0 0 . (Here we will use 3dimensional Euclidean notation without worrying about raising and lowering of the indices. For example. from the Lie algebra point of view the four vector representation corresponds to jA = jB = 1 (see (4. from the inﬁnitesimal change in the coordinates under a Lorentz transformation (see. (4.119) to the standard Pauli-Dirac representation.18)). Surprisingly.60)) and we can 2 construct the representations for J and K in this case as well from a knowledge of the addition of angular momenta. (4. Let us consider a three dimensional inﬁnitesimal rotation of coordinates around the z-axis as described in (3. considering inﬁnitesimal rotations of the coordinates around the x-axis and the y-axis respectively. (3. the two representations for the generators constructed from two diﬀerent perspectives (for the same four vector representation) appear rather diﬀerent and.) Representing the inﬁnitesimal change in the coordinates as δxi = iǫ (J3 )ij xj . There is yet another interesting example which sheds light on similarity transformations between representation.72) Similarly.4.71) we can immediately read out from (3. for example.4).

k = 1.73) in spite of the fact that they belong to the same representation for j = 1. page 182 (note there is a typo in the sign of the 23 element for L2 in this reference) 1 . This puzzle can be resolved by noting that there is a similarity transformation that connects the two representations and. j.72) and (4. they are equivalent. therefore. Jj ] = iǫijk Jk . 2 0 1 0 1 0 0 = 0 0 0 . This is.73) and. J2 = 0 0 0 i −i 0 0 0 0 . New Delhi). 0 0 −1 0 −1 0 i = √ 1 0 −1 .74) J1 = 0 0 0 0 0 0 −i 0 i .75) (LA) J1 (LA) J2 (LA) J3 which look really diﬀerent from the generators in (4.144 4 Representations of Lorentz and Poincar´ groups e It can be directly checked from the matrix structures in (4. (4. the representation in the space of three vectors which would correspond to j = 1. 3. it is well known from the study of the representations of the angular momentum algebra that the generators in the representation j = 1 have the forms1 0 1 0 1 = √ 1 0 1 . (4.72) and (4. in fact. Quantum Mechanics: A Modern Introduction. To construct the similarity transformation (which actually is a unitary transformation). Das and A. page 289 or Lectures on Quantum Mechanics. C. let us note that the generators obtained (LA) from the Lie algebra are constructed by choosing the generator J3 See. 2 0 1 0 (4. Melissinos (Gordon and Breach).73) that they satisfy [Ji . i. Das (Hindustan Book Agency. for example. A. 2. therefore provide a representation for the generators of rotations. A. On the other hand.

Let us note from (4.76) 0 Let us construct a unitary matrix from the three eigenstates in (4. 1 √ 2 i − √ 2 0 0 −1 1 − √2 U= 0 i − √2 .4. S = U.76) which will diagonalize the matrix J3 .78) then. 0 U† = 0 1 − √2 1 √ 2 i √ 2 0 i √ 2 −1 . it is straightforward to check S −1 J1 S = 0 1 − √2 1 2 = 0 1 − √2 √ 1 √ 2 i √ 2 0 i √ 2 i √ 2 0 i √ 2 √ 1 1 0 − √2 2 i i −1 0 0 i − √2 0 − √2 0 −i 0 0 0 −1 0 0 0 0 0 −i 0 −1 0 1 1 − √2 0 − √2 0 0 0 0 0 .72) that the three normalized eigenstates of J3 have the forms 1 √ 2 i − √ . (4.77) If we now deﬁne a similarity transformation (unitary) S −1 = U † .2 Representations of the Lorentz group 145 to be diagonal. −1 (4. 0 0 (4. 2 ψ (j3 =1) = 0 ψ (j3 =−1) 1 −√ i2 = − √2 . 0 ψ (j3 =0) = 0 .

79) This shows explicitly that the two representations for J corresponding to j = 1 in (4.146 4 Representations of Lorentz and Poincar´ groups e = S −1 J2 S 0 1 0 1 (LA) √ 1 0 1 = J1 . related by a similarity transformation and. therefore. (4. . 2 0 1 0 1 √ 2 i √ 2 S −1 J3 S √ 1 2 i = 0 0 −1 0 0 0 − √2 i 1 √ √ −i 0 0 0 − 2 0 2 1 i √ √ 0 0 −i 0 2 2 0 0 0 −1 0 = 0 i i 1 i − √2 0 √2 − √2 √2 0 0 −1 0 i (LA) = √ 1 0 −1 = J2 .75) which look rather diﬀerent are. 2 0 1 0 0 0 0 i 1 √ 2 i √ 2 0 0 −1 1 − √2 i − √2 0 = 0 1 − √2 1 = 0 1 − √2 1 0 = 0 0 √ 2 0 i √ 2 i √ 2 0 i √ 2 0 0 0 −1 (LA) 0 = J3 .73) and (4. √ 1 2 i −1 −i 0 0 − √2 0 0 0 0 0 1 1 √ 0 √2 0 2 i i −1 − √2 0 √2 0 0 0 0 0 0 i 0 0 0 −1 1 − √2 i − √2 0 (4. are equivalent. in fact.72).

Mµν ] P λ = P λ (ηλµ Pν − ηλν Pµ ) + (ηλµ Pν − ηλν Pµ ) P λ = Pµ Pν − Pν Pµ + Pν Pµ − Pµ Pν = 0. Therefore. P λ Pλ . which is supposed to characterize the inﬁnitesimal transformation of P 2 . Let us deﬁne a new vector operator. Mµν ] + [Pλ .38) e since it commutes with all the ten generators.4. let us note that the operator P 2 = η µν Pµ Pν = P µ Pµ . namely.3 Unitary representations of the Poincar´ group e 147 4. Pµ = 0.80) deﬁnes a quadratic Casimir operator of the Poincar´ algebra (4. Mµν P λ Pλ . The operators Mµν generate inﬁnitesimal Lorentz transformations through commutation relations and the relation above. Mµν act as Hermitian operators. we seek to ﬁnd unitary representations in some inﬁnite dimensional Hilbert space where the generators Pµ . it is useful to study the representations of the Poincar´ group.81) can be intuitively understood as follows. (4. from the generators of the Poincar´ group as e . it is known that it has only inﬁnite dimensional unitary representations except for the trivial representation that is one dimensional. P 2 . Mµν = = = P λ [Pλ . In order to determine the unitary representations. simply implies that P 2 does not change under a Lorentz transformation (it is a Lorentz scalar) which is to be expected since it does not have any free Lorentz index. Pµ P 2 . (4. Since Poincar´ e group is non-compact (like the Lorentz group). known as the Pauli-Lubanski operator.3 Unitary representations of the Poincar´ group e Since we are interested in physical theories which are invariant under translations as well as homogeneous Lorentz transformations. This e would help us in understanding the kinds of theories we can consider and the nature of the states they can have.81) The last relation in (4.

85) where the order of the operators is once again not important. 2 (4.82) The commutator between Pµ and Mνλ introduces metric tensors (see (4. 2 1 = − ǫµνλρ M λρ .82) also as W µ = Pν M µν = M µν Pν . (4. in general.83) which follows from the fact that the generators of translation commute. It follows from (4. (However. Furthermore. therefore. 2 (4. Let us deﬁne the dual of the generators of Lorentz transformation as M µν = Mµν 1 µνλρ ǫ Mλρ .38)) which vanish when contracted with the anti-symmetric LeviCivita tensor.83) that. we have e . 2 2 (4.) In general. we can write (4.83) implies that the PauliLubanski operator has only three independent components (both in the massive and massless cases). Let us next calculate the commutators between W µ and the ten generators of the Poincar´ group.84) With this. the order of Pµ and Mνλ are irrelevant in the deﬁnition of the Pauli-Lubanski operator.148 4 Representations of Lorentz and Poincar´ groups e Wµ = 1 1 µνλρ ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . First. the vector Wµ is orthogonal to Pµ . this is not true for massless theories as we will see shortly. As a result. (4. we note that Pµ W µ = 1 µνλρ ǫ Pµ Pν Mλρ = 0.

(4. Consequently. Mστ ] 2 µν 1 λρ ǫ (−ηλσ Mρτ − ηρτ Mλσ + ηλτ Mρσ + ηρσ Mλτ ) = 2 µν 1 1 1 1 = − ǫµνσρ Mρτ + ǫµντ ρ Mρσ + ǫµντ ρ Mρσ − ǫµνσρ Mρτ 2 2 2 2 ρ ρ = −ǫµνσ Mρτ + ǫµντ Mρσ = −ǫµνσρ − 1 ǫ δζ Mδζ 2 ρτ 1 + ǫµντ ρ − ǫρσδζ Mδζ 2 1 1 = − ǫµνσρ ǫτδζ ρ Mδζ + ǫµντ ρ ǫσδζ ρ Mδζ 2 2 1 ζ ζ δ δ ζ δ ζ ηµτ δν δσ − δσ δν + δµ δν ητ σ − δσ ητ ν = 2 δ δ ζ +δµ ηντ δσ − ηστ δν − σ ↔ τ Mδζ = 1 ηµτ Mνσ − Mσν + Mµν ητ σ − Mµσ ητ ν 2 +ηντ Mσµ − ηστ Mνµ − σ ↔ τ = 1 2ηµτ Mνσ − 2ηντ Mµσ − 2ηµσ Mντ + 2ηνσ Mµτ 2 .86) which follows from the the fact that momenta commute.3 Unitary representations of the Poincar´ group e 149 [W µ . Pσ 2 1 µνλρ ǫ Pν [Mλρ . Mστ = = 1 λρ ǫ Mλρ . Pσ ] 2 1 µνλρ ǫ Pν (−ηλσ Pρ + ηρσ Pλ ) = 2 = 0.4. any function of Wµ and. Mστ 2 µν 1 λρ ǫ [Mλρ . Pσ ] = = 1 µνλρ ǫ Pν Mλρ . in particular Wµ W µ . We also note that Mµν . will also commute with the generators of translation.

. Mστ ] M µν + Pν M µν . (4. Mστ ] = Pν M µν . it transforms like a vector which we should expect since it has a free Lorentz index. the operator Mµν behaves exactly like the generators of Lorentz transformation (see (4. Using this. that [Wµ . we see that the operator W µ transforms precisely the same way as does the generator of translation or the P µ operator under a Lorentz transformation. Mστ = [Pν .30)).90) = −ǫµν λρ Wλ Pρ .88) Equation (4. then. Mρσ ] Pλ 2 2 1 λρσ ǫ (ηµρ Wσ − ηµσ Wρ ) Pλ 2 ν (4. (4. Mστ = (ηνσ Pτ − ηντ Pσ ) M µν µ ν µ ν +Pν −δσ M ντ − δτ M µ + δτ M νσ + δσ M µ σ τ µ = Pτ M µ − Pσ M µ − δσ Pν M ντ − Pτ M µ σ τ σ µ +δτ Pν M νσ + Pσ M µ τ µ µ µ µ = δσ Pν Mτ ν − δτ Pν Mσν = δσ Wτ − δτ Wσ . Namely.87) Here we have used the identity satisﬁed by the four dimensional LeviCivita tensors. for completeness as well as for later use. (4. Namely. we can now evaluate [W µ . 1 1 λρσ ǫν Mρσ Pλ = ǫνλρσ [Wµ . Let us note here. Wν ] = = Wµ . ǫµνλρ ǫστ ζρ = −ηµσ (ηντ ηλζ − ηνζ ηλτ ) − ηµτ (ηνζ ηλσ − ηνσ ηλζ ) −ηµζ (ηνσ ηλτ − ηντ ηλσ ) .87) simply says that under a Lorentz transformation. it behaves like a second rank antisymmetric tensor under a Lorentz transformation.89) In other words.150 4 Representations of Lorentz and Poincar´ groups e = −ηµσ Mντ − ηντ Mµσ + ηµτ Mνσ + ηνσ Mµτ .

(4. (4. cannot represent Casimir operators of the algebra. There are other Lorentz scalars that can be constructed from Pµ and Mµν such as Mµν M µν . of course.1 Massive representation. satisfying . To ﬁnd unitary irreducible representa- tions of the Poincar´ algebra. Mστ ] W µ = W µ (ηµσ Wτ − ηµτ Wσ ) + (ηµσ Wτ − ηµτ Wσ ) W µ = Wσ Wτ − Wτ Wσ + Wτ Wσ − Wσ Wτ = 0. Mστ ] + [Wµ . The irreducible representations of the Poincar´ group can be clase siﬁed into two distinct categories. the representations can be labelled by the eigenvalues of these operators.89) that [W µ Wµ . It can be shown that P 2 and W 2 represent the only Casimir operators of the algebra and. we choose the basis vectors of the repe resentation to be eigenstates of the momentum operators. labelled by the momentum eigenvalues.3 Unitary representations of the Poincar´ group e 151 It follows now from (4. The eigenstates of the momentum operators |p are. we can choose the momentum operators.3. 4.4. (4. Namely. let us note from this analysis that a Casimir operator for the Poincar´ algebra e must necessarily be a Lorentz scalar (since it has to commute with Mµν ). without loss of generality.91) which is to be expected since Wµ W µ is a Lorentz scalar. we conclude that if we deﬁne an operator W 2 = W µ Wµ .93) However. it is easy to check that these do not commute with the generators of translation and. In fact.92) then. Mστ ] = W µ [Wµ . pµ . consequently. this would also represent a Casimir operator of the Poincar´ e algebra since Wµ commutes with the generators of translation (see (4. Pµ . therefore. to be diagonal (they satisfy an Abelian subalgebra).86)). L2 = Lµ Lµ with Lµ = P ν Mνµ . Therefore. which we treat separately. Mµν M µν .

namely. However.152 4 Representations of Lorentz and Poincar´ groups e Pµ |p = pµ |p .94) and in this basis. the eigenvalues of W 2 are not so obvious. (4. using (4. We recall that Wµ = 1 µνλρ 1 ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . where p 2 = p µ p µ = m2 . the eigenvalues of the operator P 2 = P µ Pµ are obvious. let us study this operator in some detail.95) (4. 2 2 (4.97) Therefore. P 2 |p = pµ pµ |p . we have 1 µνλρ ǫ Mλρ Pν ǫµστ ζ Mτ ζ Pσ 4 W 2 = W µ Wµ = = = 1 µνλρ στ ζ ǫ ǫµ Mλρ Pν Mτ ζ Pσ 4 1 −η νσ η λτ η ρζ − η λζ η ρτ − η ντ η λζ η ρσ − η λσ η ρζ 4 − η νζ η λσ η ρτ − η λτ η ρσ Mλρ Pν Mτ ζ Pσ 1 1 1 = − Mλρ Pν M λρ P ν − Mλρ Pν M νλ P ρ + Mλρ Pν M νρ P λ 2 4 4 1 1 − Mλρ Pν M ρν P λ + Mλρ Pν M λν P ρ 4 4 1 = − Mλρ Pν M λρ P ν − Mλρ Pν M νλ P ρ 2 .88).96) Here m denotes the rest mass of the single particle state and we assume the rest mass to be non-zero. Therefore. (4.

99). 2 (4. takes the form 1 W 2 = − m2 M λρ Mλρ − m2 Mλ0 M 0λ 2 1 = − m2 2M 0λ M0λ + M ij Mij + m2 M 0λ M0λ 2 1 (4.98) where we have simpliﬁed terms in the intermediate steps using the anti-symmetry of the Lorentz generators.101) where Jk represents the total angular momentum of the particle. p 2 = p µ p µ = m2 .99) and the operator W 2 acting on such a state.3 Unitary representations of the Poincar´ group e 153 1 λ ρ = − Mλρ M λρ P 2 + δν P ρ P ν − δν P λ P ν 2 λ −Mλρ M νλ Pν P ρ + 4P λ P ρ − δν P ν P ρ 1 = − M λρ Mλρ P 2 − Mλρ M νλ Pν P ρ .100) = − m2 M ij Mij . (4. the Pauli-Lubanski operator (4.102) The result in (4. (4.4. (4. To understand the meaning of this operator. 0. 0) . Let us note that in the rest frame (4. 0. pµ = (m. let us go to the rest frame of the massive particle. In this frame.102) can also be derived in an alternative manner which is simpler and quite instructive. 2 Recalling that (see (4.82) has the form . we obtain 1 W 2 = − m2 ǫij k J k ǫij ℓ Jℓ 2 1 ℓ = − m2 −2δk J k Jℓ 2 = m2 J k Jk = −m2 J 2 .26)) Mij = ǫij k Jk .

104) which is the result obtained in (4. this simply measures the spin of the particle.102). (m. for a massive particle at rest. we ﬁnd eigenvalues of W 2 : w2 = −m2 s(s + 1). (4. Technically.154 4 Representations of Lorentz and Poincar´ groups e W0 = Wi 1 0ijk ǫ Mjk pi = 0. It is not hard to see that all possible 3dimensional rotations would leave such a vector invariant. rotations around the x or the y or the z axis will not change the time component of a four vector (recall that it is the spin 0 component of a four vector) and. The dimensionality of the representation can also be understood in an alternative manner as follows. Namely. all the degenerate states can be labelled not . It follows now that W 2 = W µ Wµ = W i Wi = m2 J i Ji = −m2 J2 . therefore. For a state at rest with momentum of the form pµ = (m. 0.105) Thus. (4. one says that the 3-dimensional rotations deﬁne the “little” group of a time-like vector and this method of determining the representation is known as the “induced” representation. 0).103) where we have used (1. Clearly.12). Therefore. Therefore. 2 (4. would deﬁne the stability group of such vectors. these would deﬁne an invariant subgroup of the Lorentz group and will lead to the degeneracy of states. 2 m 1 iνλρ ǫ Mλρ pν = − ǫijk Mjk = 2 2 m jk = − ǫi Mjk = mJi . we see that the representations with p2 = 0 can be labelled by the eigenvalues of the Casimir operators.34) as well as (4. for a massive particle. namely the mass and the spin of a particle and the dimensionality of such a representation will be (2s + 1) (for both positive as well as negative energy). we can ask what Lorentz transformations would leave such a vector invariant. we can think of W 2 as being proportional to J 2 and in the rest frame of the particle. That is. 0. s).

s − 1.108) However.95)) pµ pµ = 0. acting on states in such a vector space. the eigenstates of P0 are invariant under three dimensional rotations and are simultaneous eigenstates of the angular momentum operators as well and such spaces are (2s+1) dimensional. −s. Consequently.3.106) Namely. up to a normalization factor.103) that. p = 0. a state at rest is an eigenstate of the P0 operator. From the Lorentz algebra. we note that (see (4. In closing. 1 . (4. Mij ] = 0. (4. (4. −s + 1.83) we see that our states in the representation should also satisfy Pµ W µ = 0. 1. we would have (see (4.2 Massless representation. commute with P0 . which generate 3-dimensional rotations and are related to the angular momentum operators. 0. the representations for a massless partie cle are slightly more involved. but also by the eigenvalues of three dimensional rotations. let us note from (4. 0.83)) P 2 = 0. pµ = (p. the operators Mij . namely.30)) [P0 . from (4. the three nontrivial Pauli-Lubanski operators correspond to the generators of symmetry of the “little group” in the rest frame. In contrast to the massive representa- tions of the Poincar´ group. Namely. In this case. This deﬁnes the 2s + 1 dimensional representation for a massive particle of spin s. This can also be seen algebraically. · · · and 2 ms = s.107) Consequently. s = 0.3 Unitary representations of the Poincar´ group e 155 just by the eigenvalue of the momentum. we note that (see (4. 4. −p). · · · .109) . (4. The basic reason behind this is that the “little” group of a light-like vector is not so obvious.4.

To determine h. W µ has the form W µ = −hpµ . acting on states in such a space. 2 2 (4. the simultaneous eigenstates of P 2 and W 2 would . or. p0 h= (4. W 2 = W µ Wµ = 0. then it can be shown (we will see this at the end of this section) that the representations are inﬁnite dimensional with an inﬁnity of spin values. consequently. (4. namely. (4.114) This is nothing other than the helicity operator (since L · p = 0) and. let us recall that Wµ = 1 µνλρ ǫ Pν Mλρ . we will not consider such representations. in the second case where W 2 = 0 on the states of the representation. therefore.110) In the ﬁrst case. W 2 = W µ Wµ = 0. namely.156 4 Representations of Lorentz and Poincar´ groups e There now appear two distinct possibilities for the action of the Casimir W 2 on the states of the representation. namely. On the other hand.112) from which it follows that acting on a general momentum basis state |p (not necessarily restricting to massless states). Such representations do not correspond to physical particles and.88)) 1 0ijk 1 ǫ pi Mjk = ǫijk pi ǫjkℓ Jℓ = −p · J = −J · p. we can easily show that the action of W µ in such a space is proportional to that of the momentum vector. for a massless particle if W 2 = 0.111) we conclude that in this space J·p .113) W0 = Comparing with (4. 2 (4.111) where h represents a proportionality factor (operator). it would lead to (see (4.

117) . It follows now from (4.115) We see from both (4.4. as we had pointed out earlier.107). in the massless case. indeed represents the helicity operator up to a normalization as we had noted in (4. −p) and we would like to determine the “little” group of symmetries associated with such a vector.3 Unitary representations of the Poincar´ group e 157 correspond to the eigenstates of momentum and helicity. let us note here that in the light-like frame (4. 2 1 2νλρ ǫ Mλρ pν = −p (M01 − M13 ) . 0. For completeness. This can be seen intuitively from the fact that the motion of the particle is along the z axis.103) and (4. We recognize that in this case. (4. 2 1 3νλρ ǫ Mλρ pν = −pM12 = W 0 .115) that the Pauli-Lubanski operator indeed has only three independent components because of the transversality condition (4. W 0 . the Pauli-Lubanski operator (4. Let us recall that we are considering a massless state with momentum of the form pµ = (p. (4.113).116) Let us now determine the dimensionality of the massless representations algebraically. We also note from (4. 2 2 1 1νλρ ǫ Mλρ pν = p (M02 − M23 ) .83). (P0 − P3 ) |p = 2p|p . 0. the set of Lorentz transformations which would leave this four vector invariant must include rotations around the z-axis. 2 W0 = W1 = W2 = W3 = (4.115) that.82) takes the form 1 0ijk 1 ǫ Mjk Pi = ǫijk Mjk pi = −pM12 . but also algebraically by recognizing that a light-like vector of the form being considered is an eigenstate of the operator P0 − P3 .115) that (the contributions from W 0 and W 3 cancel out) W 2 = W µ Wµ = W 1 W1 + W 2 W2 = −p2 (M01 − M13 )2 + (M02 − M23 )2 . namely.

These represent all the symmetries of the light-like vector (state).118) so that rotations around the z-axis deﬁne a symmetry of the lightlike vector (state) that we are considering. Clearly. as we had also seen in the massive case. Π2 ] |p = [P0 − P3 . M12 is the generator of rotations around the z axis or in the two dimensional plane and Π1 . Π2 ] = [M12 .115). also deﬁne symmetries of light-like states.90) that in the . M02 − M23 ] = 0. Thus. = [P0 − P3 . This may seem puzzling. We note that the algebra of the symmetry generators takes the form [M12 . M01 − M13 ] = Π2 . M01 − M13 ] |p = 2P1 |p = 0. [P0 − P3 .120) and. from the Poincar´ algebra in (4.158 4 Representations of Lorentz and Poincar´ groups e Furthermore. the three independent Pauli-Lubanski operators are. 2. Π2 have the same commutation relations as those of translations in this two dimensional space. (4. but can be easily understood as follows. comparing with W i . i = 1.119) p Π1 = M01 − M13 = Π2 = M02 − M23 = − It follows now that these operators commute with P0 − P3 in the space of light-like states. Π1 ] = [M12 . To determine the other symmetries of a light-like vector. in fact.38). let us deﬁne two new operators as W2 . Π1 ] |p [P0 − P3 . [Π1 . Furthermore. namely. we see that up to a normalization. p W1 . (4. it is isomorphic to the algebra of the Euclidean group in two dimensions. (4. (4. the generators of symmetry of the “little” group. We note from (4. M02 − M23 ] |p = 2P2 |p = 0. 3 in (4. we say that the stability group or the “little” group of a light-like vector is E2 .121) Namely. M12 ] = 0. M02 − M23 ] = −Π1 . therefore. [M12 . Π2 ] = [M01 − M13 . E2 (which consists of translations and rotation). we see that e [P0 − P3 .

3 Unitary representations of the Poincar´ group e 159 momentum basis states (where pρ is a number). Namely. namely. The meaning of the transformations. (4.123) that (h corresponds to the helicity quantum number) (Π1 ± iΠ2 ) |p. Π2 correspond to generators of “translation”.121) that Π1 ∓ iΠ2 correspond respectively to raising and lowering operators for M12 . Π1 ∓ iΠ2 ] = ±i (Π1 ∓ iΠ2 ) . Clearly. the Pauli-Lubanski operators satisfy an algebra and. see (3. we can say that Π1 |p. they generate transformations which will leave the momentum basis states invariant. such a state would correspond to the highest or the lowest helicity state.122) that spin can take an inﬁnite number of values which.123) and comparing with (4.122) Let us also note for completeness that the Casimir of the E2 algebra is given by Π2 = Π2 + Π2 = (Π1 ∓ iΠ2 )(Π1 ± iΠ2 ). As a result. then it follows from (4. massless theories with nontrivial spin that are parity invariant would have two dimensional representations corresponding to the highest and the . h and this is the reason for the assertion in (4. therefore.4.) This corresponds to the one dimensional representation of E2 known as the “degenerate” representation. what we have been investigating within the context of “little” groups. then. 1 2 (4. h = 0. As a result. follows from (4.124) (Alternatively. h = 0 = Π2 |p. Furthermore. Let us note from (4. if W 2 = 0 in this space of states.146)). Since Π1 . does not correspond to any physical system. we note from (4. if W 2 = 0 in this space. [M12 . if our theory is also invariant under parity (or space reﬂection).116). This is.111). as we have already pointed out. of course. the space of physical states would also include the state with the opposite helicity (recall that helicity changes sign under space reﬂection. (4. their eigenvalues can take any value. we see that in the space of light-like momentum states W 2 ∝ Π2 . On the other hand. can be thought of as generators of some transformations.86) as the transformations that leave pµ invariant.

can be seen in a heuristic way as follows. 4. In other words. in general. Academic Press. New York (1959).4 References 1. E. V. Bargmann. 3. Annals of Mathematics 48. . Irreducible unitary representations of the Lorentz group. Wigner. E. some component of the velocity would exceed the speed of light). However. spin can only be either parallel or anti-parallel to the direction of motion leading to the one dimensional nature of the representation. Annals of Mathematics 40.160 4 Representations of Lorentz and Poincar´ groups e lowest helicity states. Let us consider spin as arising from a circular motion. On unitary representations of the inhomogeneous Lorentz group. 2. the dimensionality of the representation will be one dimensional. On the other hand. in such a case. Incidentally. then we must have states corresponding to both the circular motions leading to the two dimensional representation. if parity (space reﬂection) is a symmetry of the system. the fact that the massless representations have to be one dimensional. as we have seen explicitly in the case of massless fermion theories describing neutrinos. the only consistent circular motion a massless particle can have. Then. is in a plane perpendicular to the direction of motion (otherwise. independent of the spin of the particle. Wigner. 568 (1947). it is clear that since a massless particle moves at the speed of light. 149 (1939). Group Theory and its Applications to the Quantum Mechanics of Atomic Spectra. if the theory is not parity invariant.

however. can lead to a particlelike behavior and. (5. If. It is clear from the study of classical electromagnetic ﬁelds or even from that of the displacement ﬁeld for the oscillations of a string that such quantities naturally lead to inﬁnitely many modes of oscillation each of which. it manages to manifest itself as a single particle theory only because of the Pauli principle. upon quantization. Even so. on the other hand. Combining relativity with quantum mechanics necessarily seems to lead to a many particle theory.1 Introduction It is clear by now that the quantum mechanical description of a single relativistic particle runs into diﬃculties. can describe many particles. The main question in dealing with ﬁelds. then the natural basic object for such a theory is a ﬁeld variable which is a continuous function of space and time. While the Dirac theory inherently describes a many particle theory. therefore. The answer to the ﬁrst question is quite easy from all of our discussions 161 . a consistent theory describing such a particle cannot truly be a single particle theory. what this means is that a relativistic particle has a large enough energy that it can create particles and.Chapter 5 Free Klein-Gordon ﬁeld theory 5. we want to have a theory which describes inﬁnitely many degrees of freedom. Intuitively.1) where distinct fermions are involved. is how do we choose dynamical equations of motion and how do we quantize such equations. it is not adequate to describe various decay processes such as n → p + e− + ν e . therefore.

namely. However. it is generally preferrable to work in the Lagrangian formulation which is manifestly covariant. the classical free Klein-Gordon theory where we assume that the ﬁeld variable φ(x) ≡ φ(x. each ﬁeld must belong to a speciﬁc representation of the Lorentz group. namely.1: Oscillating string ﬁxed at both ends. so far. (5. We would like to have a relativistic. The classical. e the quantization of the ﬁelds as operators can then be carried out in the Hamiltonian formalism. Secondly.4) . We also assume that the ﬁeld falls oﬀ rapidly at spatial inﬁnity (this is necessary for the theory to be well deﬁned). φ(x) = φ∗ (x). whenever any ambiguity arises in the quantization.2) where φ (x ) denotes the Lorentz transformed ﬁeld. since the Hamiltonian is not a manifestly covariant concept.162 5 Free Klein-Gordon field theory Figure 5.40)). one reverts back to the Hamiltonian formalism for its resolution. On the other hand. has the form + m2 φ(x) = ∂µ ∂ µ + m2 φ(x) = 0. Given a relativistic dynamical equation. (5. covariant theory for fundamental physical processes and. φ (x ) = φ(x). as we have seen (see (1. the dynamical equation involving the ﬁeld variables must be covariant under Poincar´ transformations. let us look at the simplest of ﬁeld theories.3) |x|→∞ The derivatives of the ﬁeld are also assumed to fall oﬀ at inﬁnity. With this in mind. therefore. lim φ(x) → 0. free Klein-Gordon equation with mass m. t) is real and that it behaves like a scalar under a Lorentz transformation. (5.

(5.2 Lagrangian density The next question that we would like to ask is whether there exists a Lagrangian or an action which would lead to the Klein-Gordon equation under a minimum action principle. of course. we assume that the Lagrangian density depends only on the ﬁeld variables and their ﬁrst derivatives. However. let us assume that there exists an action of the form tf tf tf ti S= ti dt L = ti dt d3 x L = d4 x L. P 2 denotes one of the Casimirs of the Poincar´ algebra and m is a constant.2 Lagrangian density 163 This equation is clearly invariant under Lorentz transformations. (Note that the operator in the parenthesis in the momentum space corresponds to (P 2 − m2 ).5) where we have written the Lagrangian in terms of a Lagrangian density as L= d3 x L. a Lagrangian density can depend on higher order derivatives.6) Furthermore. 5. we follow exactly what we normally do in point particle mechanics.7) In general. (5.) It is a second e order (hyperbolic) equation and can be classically solved uniquely ˙ if we are given the initial values (at t = 0) φ(x. 0) (a dot denotes a derivative with respect to t). Namely. To that end. 0) and φ(x. L = L (φ(x). As we have seen. namely. for equations which are at most second order in the derivatives.5. ∂µ φ(x)) . the Lagrangian density can depend at the . it can be easily shown that the initial values can be prescribed on any arbitrary space-like surface and the equation can still be uniquely solved. However. The speciﬁcation of the initial values would appear to single out a preferred hyper-surface which can possibly destroy covariance under a Lorentz transformation. (5.

8) δφ (x.9) δS = ti tf d4 x δL d4 x d4 x d4 x ∂L ∂L δφ(x) + δ(∂µ φ(x)) ∂φ(x) ∂∂µ φ(x) ∂L ∂L δφ(x) + ∂µ δφ(x) ∂φ(x) ∂∂µ φ(x) ∂L δφ(x) + ∂µ ∂φ(x) −∂µ ∂L δφ(x) ∂∂µ φ(x) = ti tf = ti tf = ti ∂L δφ(x) ∂∂µ φ(x) δφ(x) tf tf = ti d4 x ∂L ∂L − ∂µ ∂φ(x) ∂∂µ φ(x) + d3 x ∂L δφ(x) ˙ ∂ φ(x) .164 5 Free Klein-Gordon field theory most on the ﬁrst order derivatives of the ﬁeld variables. subject to the boundary conditions (5.10) Here we have used Gauss’ theorem (as well as the vanishing of ﬁeld variables asymptotically (5. ti (5. tf ) = 0. Note that under an inﬁnitesimal change tf (5. These are the kinds of equations we will be interested in and correspondingly we will assume this dependence of the Lagrangian density on the ﬁeld variables through out. We have . ti ) = δφ (x.3)) to simplify the surface term. We can now generalize the variational principle of point particle mechanics and ask under what conditions will the action be stationary if we change the ﬁelds arbitrarily and inﬁnitesimally as φ(x) → φ′ (x) = φ(x) + δφ(x).

(5. for the Klein-Gordon equation. ti ) = 0 = δφ (x. we can think of the dynamical equations as arising from the minimum principle associated with a given action. As in point particle mechanics. the action will be stationary under an arbitrary change in the ﬁeld variable subject to the boundary conditions provided (namely. (5. namely. tf ) . we conclude that under an arbitrary.2 Lagrangian density 165 also used the fact that δ(∂µ φ) = ∂µ (δφ) which is easily seen from (consider one dimension for simplicity) 1 ′ φ (x + ǫ) − φ′ (x) − (φ(x + ǫ) − φ(x)) ǫ 1 (φ′ (x + ǫ) − φ(x + ǫ)) − (φ′ (x) − φ(x)) = lim ǫ→0 ǫ 1 = lim (δφ(x + ǫ) − δφ(x)) ǫ→0 ǫ ǫ→0 δ(∂x φ) = lim = ∂x (δφ). δφ (x.14) This is known as the Euler-Lagrange equation associated with the action S or the Lagrangian density L(φ.9). (5. ∂µ ∂ µ φ + m2 φ = 0. inﬁnitesimal change in the ﬁeld variable. ∂φ(x) ∂∂µ φ(x) (5.10) – vanishes because of the boundary conditions (5.5. ∂L ∂L − ∂µ = 0. the change in the action is given by tf δS = ti d4 x ∂L ∂L − ∂µ ∂φ(x) ∂∂µ φ(x) δφ(x). (5.13) Therefore. ∂µ φ(x)).15) . Thus.12) Thus. δS = 0 only if).11) We note that the “surface term” – the last term in the expression (5.

166 5 Free Klein-Gordon field theory we can ask whether there exists a Lagrangian density whose EulerLagrange equation will lead to this dynamical equation. ∂∂µ φ(x) Therefore. ∂φ(x) ∂L = ∂ µ φ(x). real scalar ﬁeld as its Euler Lagrange equation.16) ∂L = −m2 φ(x). adding a total divergence to L also does not change the equations of motion.3)). Note that the Lagrangian density L in (5. we see that the Lagrangian density 1 m2 2 ∂µ φ∂ µ φ − φ . ∂µ ∂ µ + m2 φ(x) = 0. 2 2 L= then (5. in this case. It is worth noting here that one can add a constant to L without changing the equation of motion. in this case. the action does not change either.19) gives the Klein-Gordon equation for a free. . ∂φ(x) ∂∂µ φ(x) or. if we assume the ﬁelds to fall oﬀ rapidly at inﬁnite separation (see (5. This would simply correspond to adding a zero point energy. the Euler-Lagrange equation. Note that if we choose 1 m2 2 ∂µ φ∂ µ φ − φ .17) ∂L ∂L − ∂µ = −m2 φ(x) − ∂µ ∂ µ φ(x) = 0. 2 2 L= (5.19) is manifestly Lorentz invariant. Of course. (5. gives (5. Similarly.18) In other words.

x) .22) The classical canonical Poisson bracket relations between the coordinate and the conjugate momentum have the familiar form {x.3 Quantization Since we have a Lagrangian description for the classical Klein-Gordon ﬁeld theory. the Lagrangian is a function of the coordinate and the velocity of the particle (a dot denotes a derivative with respect to t). the basic variables are φ(x) and ∂µ φ(x) and we have from (5. ˙ (5. we can try to develop a Hamiltonian description for it (which we need for quantization) in complete parallel to what we do for point particle mechanics. x} = 0 = {p. ˙ ˙ (5. L = L (x.5. ˙ (5. x) . {x.19) (xµ in this case merely labels space-time coordinates) . in turn. p) = px − L (x.21) and this leads to the deﬁnition of the Hamiltonian in the form (this is an example of a Legendre transformation) H(x.3 Quantization 167 5. H} . p} = 1. ˙ p = {p. lead to the dynamical equations in the Hamiltonian form as x = {x. (5. In the case of a point particle. ∂x ˙ p= (5.24) In the case of the classical scalar (Klein-Gordon) ﬁeld theory.20) The conjugate momentum associated with the dynamical variable x is deﬁned as ∂L .23) These. p} . H} . on the other hand.

φ(y)}x0 =y0 {φ(x). we can show that the dynamical equation (namely.27) H= (5. ∂µ φ(x)) . the KleinGordon equation in this case) can be written in the Hamiltonian form as ˙ φ(x) = {φ(x). (5. (5.28) If we further assume the equal time canonical Poisson bracket relations between the ﬁeld variable φ(x) and the conjugate momentum Π(x) to be (see (5. = δ3 (x − y). (5. Note that the Lagrangian density in this case is given by .25) In an analogous manner to the point particle mechanics in (5. H} .30) Let us examine all of this in detail for the case of the free KleinGordon theory. (5.23)) {φ(x). we deﬁne a momentum canonically conjugate to the ﬁeld variable φ(x) as ∂L . Π(y)}x0 =y0 = 0 = {Π(x). Π(y)}x0 =y0 . and a Hamiltonian d3 x H.21). allow us to deﬁne a Hamiltonian density as ˙ H = Π(x)φ(x) − L.168 5 Free Klein-Gordon field theory L = L (φ(x).26) which will. H} . ˙ ∂ φ(x) Π(x) = (5.29) then. ˙ Π(x) = {Π(x). in turn.

φ − ∇φ · ∇φ − 2 2 2 (5.3 Quantization 169 L = = m2 2 1 ∂µ φ∂ µ φ − φ 2 2 1 ˙2 1 m2 2 φ .33) 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x).34) With the equal time canonical Poisson bracket relations deﬁned to be (see (5.5. Π(y)}x0 =y0 we obtain = 0 = {Π(x). we can identify the conjugate momentum with ∂L ˙ = φ(x). 2 2 2 (5. φ(y)}x0 =y0 {φ(x). ˙ ∂ φ(x) Π(x) = (5.32) Consequently. (5.31) Therefore. this leads to the Hamiltonian density ˙ H = Π(x)φ(x) − L ˙ = Π(x)φ(x) − = Π(x)Π(x) − = 1 ˙2 1 m2 2 φ (x) φ + ∇φ · ∇φ + 2 2 2 1 m2 2 1 2 Π (x) + ∇φ · ∇φ + φ (x) 2 2 2 (5. Π(y)}x0 =y0 .35) . 2 2 2 which gives the Hamiltonian for the theory H = = d3 x H d3 x 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x) . = δ3 (x − y).29)) {φ(x).

˙ Π(x) = ∇ · ∇φ(x) − m2 φ(x) = ∇ 2 φ(x) − m2 φ(x).36) where we have used the fact that. x0 δ3 (x − y) = Π(x). since the Hamiltonian is time independent (this can be checked easily using the equations of motion and will also be shown in (5. φ(y)} x0 =y0 = d3 y ∇y φ y. x0 · ∇y −δ3 (x − y) +m2 φ y. we can choose the time coordinate for the ﬁelds in its deﬁnition to coincide with x0 in order that we can use the equal time Poisson bracket relations (5. ∇y φ(y)} + m2 φ(y) {Π(x). d3 y 1 m2 2 1 2 Π (y) + ∇y φ · ∇y φ + φ (y) 2 2 2 x0 =y 0 = d3 y ∇y φ(y) ·{Π(x).37). the Hamiltonian equations (5. in this case.37) Thus. H} = Π(x). Π(y)} x0 =y 0 d3 y Π y. Furthermore.36) and (5.29).73)). −δ3 (x − y) (5.170 5 Free Klein-Gordon field theory ˙ φ(x) = {φ(x). we also have ˙ Π(x) = {Π(x). 1 2 d3 y 1 m2 2 1 2 Π (y) + ∇y φ · ∇y φ + φ (y) 2 2 2 x0 =y 0 x0 =y 0 d3 y φ(x). (5. lead to ˙ φ(x) = Π(x). x0 = ∇ · ∇φ(x) − m2 φ(x).38) . (5. Π2 (y) d3 y Π(y) {φ(x). H} = = = = φ(x).

we can expand the ﬁeld operator in this basis as .) 5.4 Field decomposition Given any complete set of solutions of the classical Klein-Gordon equation. In the quantum theory. This is. see (5. = iδ3 (x − y).5. or. (5. the quantization of such a system is quite straightforward. In the second quantization. Π(y)]x0 =y0 = 0 = [Π(x). we obtain the second order equation ¨ ˙ φ(x) = Π(x) = ∇ 2 φ(x) − m2 φ(x). in the ﬁrst quantization we quantize x’s and p’s. Once we have the classical Hamiltonian description of a physical system. ∂µ ∂ µ + m2 φ(x) = 0. This is conventionally referred to as second quantization. therefore. (Namely. the free Klein-Gordon equation (1. but not the wave function. ¨ φ − ∇ 2 φ(x) + m2 φ(x) = 0.4 Field decomposition 171 These are ﬁrst order equations as Hamiltonian equations should be and from these. We already know that the plane wave solutions (1. of course.40) Note that this is the same Klein-Gordon equation as we had studied earlier in chapter 1. φ(y)]x0 =y0 [φ(x). brings out the classical Hamiltonian description of the classical KleinGordon ﬁeld theory. we can expand the ﬁeld operator φ(x) in such a basis.40) (which is a second order Euler-Lagrange equation) and this discussion. therefore. we are supposed to treat φ(x) and Π(x) as Hermitian operators (because we are dealing with real ﬁelds. (5.2)) satisfying the equal time commutation relations (remember = 1) [φ(x).39) or. Π(y)]x0 =y0 . we quantize the “wavefunction”.41) of the KleinGordon equation deﬁne a complete basis and. However. here we are treating φ(x) as an operator which is quantized and not as a wavefunction as we had done earlier.

41).45) (5. φ(k) = 0.43) d4 k −k2 + m2 φ(k)e−ik·x = 0.41) where we recognize that the operators φ(x) and φ(k) are Fourier transforms of each other (up to a multiplicative factor). Namely. 1 (2π) 3 2 3 2 ∂µ ∂ µ + m2 d4 k e−ik·x φ(k) = 0.46) . (Recall that 1 k · x = k0 x0 − k · x and the factor (2π)3/2 is introduced for later convenience. (5. (5.172 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 d4 k e−ik·x φ(k).41) in the Klein-Gordon equation. we can denote φ(k) = δ k2 − m2 a(k).) Since φ(x) satisﬁes the Klein-Gordon equation ∂µ ∂ µ + m2 φ(x) = 0. diﬀerent. in general. Substituting this back into the expansion (5. (5. We also emphasize here that the hermiticity properties of a given ﬁeld and its Fourier transform are. As a result. This shows that unless k2 = m2 .44) where the operator a(k) is no longer constrained by the equation of motion. (5. we obtain 1 (2π) or.42) using the ﬁeld expansion (5. (5. the Fourier transform has non-vanishing contribution (support) only on the massshell of the particle deﬁned by k2 = (k0 )2 − k2 = m2 . we obtain 1 (2π) 3 2 φ(x) = d4 k δ k2 − m2 e−ik·x a(k).

5. (5.4 Field decomposition k0 173 k2 = m2. k0 < 0 Figure 5.2: Mass shell on which φ(k) has support. k0 = ± k2 + m2 ≡ ±Ek . (k0 )2 = k2 + m2 .46). (5. k0 > 0 k k 2 = m 2 . we can write δ k 2 − m2 2 = δ (k0 )2 − Ek = = 1 δ k0 − Ek + δ k0 + Ek 2|k0 | 1 δ k0 − Ek + δ k0 + Ek 2Ek . We note that the argument of the delta function vanishes for k 2 = m2 . or. or. the ﬁeld expansion becomes .48) Using this relation in (5.47) Correspondingly.

2k0 (5. so that we can write 1 (2π) 3 2 a† (−k) = a(k). k) .51) Comparing the left-hand side and the right-hand side of (5. φ† (x) = φ(x). or.49) Changing k → −k in the second term in (5.53) . (5. 2k0 (5. Therefore. 2k0 (5.52) φ(x) = d3 k −ik·x e a(k) + eik·x a† (k) . k) 0 x0 +ik·x = 1 (2π) 3 2 d3 k 1 0 e−iEk x +ik·x a(Ek . we obtain 1 (2π) 2 3 φ(x) = d3 k −ik·x e a(k) + eik·x a(−k) . k) 2Ek 0 +ik·x + eiEk x a(−Ek .2)). we obtain a† (k) = a(−k).50) Let us note that we are dealing with a Hermitian ﬁeld (see (5.51).174 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 dk0 d3 k × e−ik 1 δ k0 − Ek + δ k0 + Ek 2Ek a(k0 . 1 (2π) = 3 2 d3 k ik·x † e a (k) + e−ik·x a† (−k) 2k0 1 3 (2π) 2 d3 k −ik·x e a(k) + eik·x a(−k) . (5.49) and identifying k0 = Ek > 0.

2k0 (5. In fact.5. (5. conventionally one deﬁnes a(k) . a(k) = √ 2k0 a† (k) a† (k) = √ . we often denote the positive and the negative energy parts of the ﬁeld operator as d3 k (2π)3 2k0 d3 k (2π)3 2k0 φ(+) (x) = φ(−) (x) = e−ik·x a(k).5 Creation and annihilation operators Given the ﬁeld expansion (5. eik·x a† (k).56) . φ(x) = φ(+) (x) + φ(−) (x). (5. Thus.53) as d3 k (2π)3 2k0 φ(x) = e−ik·x a(k) + eik·x a† (k) .58) 5. (5.5 Creation and annihilation operators 175 Here we are supposed to understand that k0 = Ek = k2 + m2 > 0. (5.54) It is clear from (5.57) so that we can write the ﬁeld operator as a sum of its positive and negative energy parts.54) that a(k) and a† (k) are really functions of the three momentum k alone.56) This unique decomposition of the ﬁeld operator φ(x) into positive and negative energy (frequency) parts is quite signiﬁcant as we will see in the course of our discussions.55) and in terms of these operators we can write the ﬁeld operator (5.

56). (5. we note from these relations that we can write the operators a(k) and ˙ a† (k) in terms of φ(x) and Π(x) as (recall that Π(x) = φ(x)) a(k) = = a† (k) = = − 1 (2π)3 2k0 i (2π)3 2k0 1 (2π)3 2k0 i (2π)3 2k0 d3 x eik·x k0 φ(x) + iΠ(x) ← → d3 x eik·x ∂t φ(x). (5. For example. it is assumed that k0 = Ek (namely.176 5 Free Klein-Gordon field theory φ(x) = d3 k (2π)3 2k0 e−ik·x a(k) + eik·x a† (k) . d3 x e−ik·x k0 φ(x) − iΠ(x) ← → d3 x e−ik·x ∂t φ(x). It is clear from (5.60) 2(2π)3 The two deﬁning relations in (5.61) ← → where ∂t is deﬁned in (1.61).32) to be ˙ Π(x) = φ(x) = −i d3 k k0 e−ik·x a(k) − eik·x a† (k) .59) and (5. This can also be checked explicitly.61) that since the left-hand sides are independent of time.60) are invertible. k0 is not an independent variable). Namely. It is important to emphasize here that in the expressions in (5. (5.59) we obtain the expansion for the conjugate momentum in (5. the expressions on the righthand side must also be time independent.61) we obtain (neglecting the overall multiplicative factors) . taking the time derivative of the ﬁrst relation in (5.

a(k′ ) − e−ik·x+ik ·y a(k). the time independence of the second expression in (5. Π(y)]x0 =y0 = − ′ ′ ′ x0 =y 0 (5.5. a† (k′ ) + eik·x−ik ·y a† (k).3)) and have √ made the identiﬁcation k0 = Ek = k2 + m2 .61) can also be derived in a simple manner (or note that the second relation is the Hermitian conjugate of the ﬁrst and. a(k′ ) + e−ik·x+ik ·y a(k). Imposing now the quantization relations between φ(x) and Π(x) (see (5.61) are indeed time independent.5 Creation and annihilation operators 177 ∂t = = = = ← → d3 x eik·x ∂t φ(x) 2 2 d3 x eik·x (∂t φ(x)) − (∂t eik·x ) φ(x) 2 d3 x eik·x (∇2 − m2 )φ(x) − (∂t eik·x ) φ(x) 2 d3 x −∂t + ∇2 − m2 eik·x φ(x) d3 x (k0 )2 − k2 − m2 eik·x φ(x) = 0. must be time independent). a(k′ ) + eik·x+ik ·y a† (k). Similarly. a† (k′ ) = 0. integrated the space derivatives by parts assuming that the surface terms vanish (see (5. This shows explicitly that the integrals in (5. [Π(x).63) d kd k 3 3 ′ √ k0 k′ 0 2(2π)3 ′ × e−ik·x−ik ·y a(k). in the intermediate steps. we have used the fact that φ(x) satisﬁes the Klein-Gordon equation.62) Here. (5. φ(y)]x0 =y0 = ′ × e−ik·x−ik ·y a(k). a† (k′ ) .40)) we obtain d3 k (2π)3 2k0 d3 k ′ (2π)3 2k′ 0 ′ [φ(x). therefore.

recalling that the integrals are ime independent.65) From these. a† (k′ ) = 0.64) i (2π)3 d3 k d3 k ′ k′ 0 4k0 ′ × e−ik·x−ik ·y a(k). However. Thus.) a(k). a(k′ ) − eik·x+ik ·y a† (k).178 ′ 5 Free Klein-Gordon field theory − eik·x−ik ·y a† (k). Π(y)]x0 =y0 = − ′ ′ x0 =y 0 (5. a(k′ ) = 0 = a† (k). k′ 0 φ(y) − iΠ(y) x0 =y 0 . (5.66) can also be obtained more directly from the inversion formulae in (5.66) This shows that the operators a(k) and a† (k). Any other redeﬁnition will introduce a multiplicative factor into the commutation relations. [φ(x). a† (k′ ) + eik·x−ik ·y a† (k). a† (k′ ) = δ3 k − k′ . have commutation relations analogous to the annihilation and creation operators of a harmonic oscillator.61). a(k). ′ ′ x0 =y 0 (5. which are the coeﬃcients of expansion of the ﬁeld operator in a plane wave basis. we have a(k). it appears that there is an inﬁnite number of such operators in the present case – one for every value of the momentum k.55). The commutation relations in (5. a† (k′ ) = ′ d3 xd3 y √ eik·x−ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x). a† (k′ ) = iδ3 (x − y). a(k′ ) + eik·x+ik ·y a† (k). we can deduce the fundamental commutation relations between the coeﬃcients of expansion to be (This nice form arises because of the redeﬁnition (5. a† (k′ ) . for example. a(k′ ) − e−ik·x+ik ·y a(k).

5 Creation and annihilation operators 179 = d3 xd3 y ′ √ eik·x−ik ·y (2π)3 4k0 k′ 0 × −ik0 [φ(x). e 0 k′ 0 4k (5. Π(y)] + ik′ 0 [Π(x).66) can also be derived in the same manner a(k). The other two commutation relations in (5. k′ 0 φ(y) + iΠ(y) = d3 xd3 y ′ √ eik·x+ik ·y 3 4k 0 k ′ 0 (2π) × ik0 [φ(x).5. φ(y)] (k0 − k′ 0 ) = − √ 4k0 k′ 0 (k0 − k′ 0 ) = − √ 4k0 k′ 0 d3 xd3 y ik·x+ik′ ·y 3 e δ (x − y) (2π)3 d3 x i(k+k′ )·x e (2π)3 x0 =y 0 x0 =y 0 x0 =y 0 (k0 − k′ 0 ) i(k0 +k′ 0 )x0 3 = − √ δ (k + k′ ) = 0. a† (k′ ) = ′ d3 xd3 y √ e−ik·x−ik ·y 3 4k 0 k ′ 0 (2π) . Π(y)] + ik′ 0 [Π(x). a(k′ ) = d3 xd3 y ′ √ eik·x+ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x).67) where we have used the fact that for k′ = k. φ(y)] x0 =y 0 = = = (k0 + k′ 0 ) √ 4k0 k′ 0 (k0 + k′ 0 ) √ 4k0 k′ 0 d3 xd3 y ik·x−ik′ ·y 3 e δ (x − y) (2π)3 d3 x i(k−k′ )·x e (2π)3 x0 =y 0 (k0 + k′ 0 ) i(k0 −k′ 0 )x0 3 √ δ (k − k′ ) = δ3 (k − k′ ). e 4k0 k′ 0 a† (k). we have k′ 0 = k0 .

φ(y)] = = = (k0 − k′ 0 ) √ 4k0 k′ 0 (k0 − k′ 0 ) √ 4k0 k′ 0 d3 xd3 y −ik·x−ik′ ·y 3 e δ (x − y) (2π)3 d3 x −i(k+k′ )·x e (2π)3 x0 =y 0 x0 =y 0 (k0 − k′ 0 ) −i(k0 +k′ 0 )x0 3 √ δ (k + k′ ) = 0. We note that d3 k d3 k ′ √ 0 ′ 0 k k 2(2π)3 ′ d3 x Π2 (x) = − ′ d3 x × e−i(k+k )·x a(k)a(k′ ) − e−i(k−k )·x a(k)a† (k′ ) − ei(k−k )·x a† (k)a(k′ ) + ei(k+k )·x a† (k)a† (k′ ) = − d3 k d3 k′ √ 0 ′ 0 −i(k0 +k′ 0 )x0 3 k k e δ k + k′ a(k)a(k′ ) 2 0 −k ′ 0 )x0 ′ ′ −e−i(k −ei(k + ei(k = − 1 2 δ3 k − k′ a(k)a† (k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) 0 x0 0 +k ′ 0 )x0 d3 k k0 e−2ik a(k)a(−k) − a(k)a† (k) (5.180 5 Free Klein-Gordon field theory × k0 φ(x) − iΠ(x). k′ 0 φ(y) − iΠ(y) = ′ d3 xd3 y √ e−ik·x−ik ·y 3 4k 0 k ′ 0 (2π) x0 =y 0 × −ik0 [φ(x).68) where we have used the fact that k′ 0 = k0 for k′ = −k. Here. let us look at the Hamiltonian of the system. independent of whether k′ = ±k. e 4k0 k′ 0 (5. Π(y)] − ik′ 0 [Π(x). To understand the meaning of these operators further.69) − a† (k)a(k) + e2ik 0 x0 a† (k)a† (−k) . . we have used the fact that.

71) d3 x × e−i(k+k )·x a(k)a(k′ ) + e−i(k−k )·x a(k)a† (k′ ) +ei(k−k )·x a† (k)a(k′ ) + ei(k+k )·x a† (k)a† (k′ ) = d3 k d3 k ′ √ √ 2k0 2k′ 0 × e−i(k +e−i(k 0 +k ′ 0 )x0 ′ ′ δ3 k + k′ a(k)a(k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a(k)a† (k′ ) . (5.70) Similarly.5.5 Creation and annihilation operators 181 k′ 0 = (k′ )2 + m2 = k2 + m2 = k0 . d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ (5. we can calculate d3 x ∇φ(x) · ∇φ(x) = − ′ 1 (2π)3 d3 x ′ d3 k d3 k ′ √ √ k · k′ 2k0 2k′ 0 × e−i(k+k )·x a(k)a(k′ ) − e−i(k−k )·x a(k)a† (k′ ) − ei(k−k )·x a† (k)a(k′ ) + ei(k+k )·x a† (k)a† (k′ ) = − d3 k d3 k ′ √ √ k · k′ 0 ′0 2k 2k 0 +k ′ 0 )x0 ′ ′ × e−i(k −e−i(k −ei(k + ei(k = − 1 2 δ3 k + k′ a(k)a(k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a(k)a† (k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) 0 +k ′ 0 )x0 d3 k k2 0 0 −e−2ik x a(k)a(−k) − a(k)a† (k) 0 k 0 x0 − a† (k)a(k) − e2ik d3 x φ2 (x) = 1 (2π)3 ′ a† (k)a† (−k) .

(5. we obtain H = = 1 2 d3 x H d3 x Π2 (x) + ∇φ(x) · ∇φ(x) + m2 φ2 (x) 1 4 d3 k k0 e−2ik 0 x0 = − a(k)a(−k) − a(k)a† (k) − a† (k)a(k) + e2ik + 0 x0 a† (k)a† (−k) k2 0 0 −e−2ik x a(k)a(−k) − a(k)a† (k) 0 k 0 x0 − a† (k)a(k) − e2ik − a† (k)a† (−k) m2 −2ik0 x0 a(k)a(−k) + a(k)a† (k) e k0 0 x0 + a† (k)a(k) + e2ik = − 1 4 d3 k k0 0 x0 a† (k)a† (−k) (k0 )2 − k 2 − m2 a(k)a(−k) + e2ik 0 x0 × e−2ik a† (k)a† (−k) − (k0 )2 + k 2 + m2 = 1 2 a(k)a† (k) + a† (k)a(k) d3 k k0 a(k)a† (k) + a† (k)a(k) . (5.69).34).72) Substituting (5.182 +ei(k 0 −k ′ 0 )x0 5 Free Klein-Gordon field theory δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) + ei(k = 1 2 0 +k ′ 0 )x0 d3 k 0 0 1 e−2ik x a(k)a(−k) + a(k)a† (k) 0 k 0 x0 + a† (k)a(k) + e2ik a† (k)a† (−k) .71) and (5.72) into the expression for the Hamiltonian (5.

we can indeed think of them as the annihilation and creation operators of a harmonic oscillator system. we see from (5.73).73) that the Hamiltonian for the free KleinGordon ﬁeld theory is indeed time independent and is the sum of the Hamiltonians for an inﬁnite number of harmonic oscillators of frequency labelled by Ek . (5. H] = a(k). a(k′ ) a† (k′ ) + a† (k′ ) a† (k). the zero point energy of such a system would be inﬁnite. a† (k′ ) a(k′ ) 2 Ek′ a(k′ )δ3 k − k′ + δ3 k − k′ a(k′ ) 2 = Ek a(k).54) k0 = k2 + m2 = Ek .74) Thus. a(k′ ) d3 k ′ Ek′ −δ3 k − k′ a† (k′ ) − a† (k′ )δ3 k − k′ 2 (5.5 Creation and annihilation operators 183 (5. Thus. From the form of the Hamiltonian for the Klein-Gordon ﬁeld in (5. = = d3 k ′ d3 k ′ d3 k ′ Ek′ a(k′ ) a(k).73) = d3 k Ek a(k)a† (k) + a† (k)a(k) . It follows now that Ek′ a(k′ )a† (k′ ) + a† (k′ )a(k′ ) 2 [a(k). H = a† (k). Clearly. The two relations in (5. = = d3 k ′ Ek′ 2 d3 k ′ Ek′ a(k′ )a† (k′ ) + a† (k′ )a(k′ ) 2 a† (k). 2 where we have used the relation (5.75) show that the operators a(k) and a† (k) annihilate and create respectively a quantum of energy Ek . a† (k′ ) + a(k). But since this is an additive constant . we then conclude that we can think of this as the Hamiltonian for an inﬁnite collection of decoupled harmonic oscillators.75) = −Ek a† (k).5. a† (k).

= a† (k)a† (k′ ) = a† (k′ )a† (k). Thus.76) N.O.O. If we normal order the Hamiltonian using (5. = = = d3 k d3 k d3 k Ek a† (k)a(k) = where we have deﬁned the number operator for an oscillator with momentum k as in the case of a simple harmonic oscillator as N (k) = a† (k)a(k).78) and we can deﬁne the total number operator for all the oscillators (inﬁnite number of them) in the system as . N.184 5 Free Klein-Gordon field theory we can always rescale (shift) our energy to measure from zero which we take to be the ground state energy.77) H N.73) (namely. : a† (k)a(k′ ) : = a† (k)a(k′ ). : a† (k)a† (k′ ) : (5. : a(k)a(k′ ) : = a(k)a(k′ ) = a(k′ )a(k). then. the ordering of the operators is ambiguous. (5. Let us recall that in the passage from classical to a quantum theory.O. (5. N. we assume that the quantum theory is deﬁned by the normal ordered Hamiltonian) Ek : a(k)a† (k) + a† (k)a(k) : 2 Ek † a (k)a(k) + a† (k)a(k) 2 d3 k Ek N (k). a(k)a† (k′ ) a† (k)a(k′ ) a(k)a(k′ ) a† (k)a† (k′ ) = = = = : a(k)a† (k′ ) : = a† (k′ )a(k). Taking advantage of this. we have from (5.76). we deﬁne normal ordering as the ordering in which the creation operators stand to the left of the annihilation operators. This is conveniently done by normal ordering of the operators. (for bosons) N.O.O.

(5.5. .80) then leads to (5. N = = a(k). It now follows from the deﬁnition of the number operators in (5. a† (k′ )a(k′ ) = a† (k′ ) a† (k). (5. N ] = = a† (k). d3 k′ N (k′ ) d3 k′ −a† (k′ )δ3 (k − k′ ) = −a† (k). a† (k). Equation (5. a† (k′ )a(k′ ) a(k).5 Creation and annihilation operators 185 N= d3 k N (k) = d3 k a† (k)a(k).77) how the normal ordering has redeﬁned away an inﬁnite zero point energy (which basically would arise from commuting a(k)a† (k) to bring it to the normal ordered form). d3 k′ N (k′ ) d3 k′ a(k′ )δ3 (k − k′ ) = a(k). a† (k).80) [a(k). N (k′ ) = = a(k).79) It is worth noting from the derivation in (5. a(k′ ) = −a† (k′ )δ3 (k − k′ ).81) which reﬂects again the fact that a(k) and a† (k) merely lower or raise the number of quanta of momentum k by one unit.79) that a(k). a† (k′ ) a(k′ ) = a(k′ )δ3 (k − k′ ).78) and (5. N (k′ ) = a† (k).

E|H|E E| d3 k Ek a† (k)a(k)|E d3 k Ek E|a† (k)a(k)|E (5. which can be achieved by quantizing the system in a box.83) where we are assuming that the state |E is normalized and that the eigenvalue E is discrete for simplicity. (5. (5. (5.6 Energy eigenstates 5 Free Klein-Gordon field theory From now on.85) . H] = Ek a(k).75)) [a(k). consequently. a† (k). It is clear from (5. we do not have the problem of negative energy states that we had in the single particle theory.83) that E = = = ≥ 0. the inner product represents the norm of the state a(k)|E ) and.186 5.77) and (5. This shows that the eigenvalues of the Hamiltonian have to be positive semideﬁnite and in the second quantized ﬁeld theory. the integrand is positive semideﬁnite.84) This is because both Ek as well as the inner product of states is positive deﬁnite (namely. H = −Ek a† (k).77) (as describing the quantum free Klein-Gordon theory) of the form H= d3 k Ek a† (k)a(k).82) Let us consider an energy eigenstate |E of this Hamiltonian satisfying H|E = E|E . Let us next recall that the operators a(k) and a† (k) satisfy the commutation relations (see (5. let us consider the normal ordered Hamiltonian (5.

the energy eigenvalue is bounded from below (E ≥ 0).84). . or. there must exist a state (energy eigenstate) with a minimum energy. = −Ek a† (k)|E . But as we have seen in (5. clearly. therefore. then a† (k) acting on it gives another energy eigenstate with the higher value of energy E+Ek . Namely. (5. a(k)H|E = Ek a(k)|E . therefore. we note that a† (k). we can lower the energy eigenvalue arbitrarily. raises the energy value. = Ek a(k)|E . or.89) This is. |Emin .88) and we cannot lower the energy any further.86) − Ha(k)|E H {a(k)|E } = (E − Ek ) {a(k)|E } . (5. we had redeﬁned . (5.87) − Ha† (k)|E = (E + Ek ) a† (k)|E In other words. if |E is an energy eigenstate with energy eigenvalue E.6 Energy eigenstates 187 As a result. the ground state of the system or the vacuum state (recall that by normal ordering the Hamiltonian.5. then {a(k)|E } is also an eigenstate of energy with the lower energy value E − Ek . we have [a(k). H |E or. (5. The annihilation operator a(k) lowers the energy eigenvalue as it should. In such a state. such that a(k)|Emin = 0. if |E represents an energy eigenstate with eigenvalue E. Therefore. it would appear that by applying the annihilation operator successively. d3 k Ek Emin |a† (k)a(k)|Emin Emin = = 0. Similarly. H] |E or. Let us next note that since a(k) acting on an energy eigenstate lowers the energy. a† (k)H|E H a† (k)|E = −Ek a† (k)|E . The creation operator a† (k).

n2 . · · · . = 0. k1 . nℓ . · · · . k1 .) Given the vacuum state. k2 . . kℓ (5. n2 . n2 . k2 . This ground state or the vacuum state is denoted by |0 and satisﬁes (we assume that the state is normalized) |Emin a(k)|0 0|0 N |0 H|0 ≡ |0 .90) In other words. the vacuum state contains no quantum of energy or no particle if energy is related to that of particles. nℓ . k1 . (5.188 5 Free Klein-Gordon field theory the energy of the ground state to be zero). A general energy eigenstate with higher energy will have the form (up to normalizations) a† (k1 ) n1 a† (k2 ) n2 · · · a† (kℓ ) nℓ |0 . = 0. k2 . k1 . it is clear now that the expectation value of any normal ordered operator in the vacuum state. = 0 = 0|a† (k). kℓ = d3 k N (k)|n1 . nℓ . = 1.91) From our study of the harmonic oscillator we know that such states will be eigenstates of the number operator and we can denote such a state (normalized) as |n1 . · · · . would vanish. (Incidentally. (5. n2 . · · · . k2 .92) This will be an eigenstate of the total number operator satisfying N |n1 . (5. conventionally called the vacuum expectation value or vev. kℓ = a† (k1 ) √ n1 ! n1 a† (k2 ) √ n2 ! n2 ··· a† (kℓ ) √ nℓ ! nℓ |0 . one can build up states of higher energy by simply applying the creation operator.93) = (n1 + n2 + · · · + nℓ ) |n1 . nℓ . kℓ .

96) if we deﬁne the momentum operator as (This is.6 Energy eigenstates 189 This relation can be easily shown using the identity (see also (5.94) N (k)|n1 . Namely. kℓ (5. k1 . n2 quanta with four momentum k2 and so on. (5. k2 . k2 . (5. (5. However. · · · . and . · · · . Let us further note that since H= d3 k Ek a† (k)a(k) = d3 k Ek N (k).) P= then d3 k k a† (k)a(k) = d3 k k N (k). · · · . k1 . kℓ = (5. n2 . · · · .98) = (n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ ) |n1 . k1 . a† (k1 ) = n1 a† (k1 ) which also leads to n1 n1 δ3 (k − k1 ) . the natural deﬁnition from considerations of covariance. kℓ = d3 k Ek N (k)|n1 . nℓ . (a† (k1 ))n1 = a† (k) a(k). n2 . nℓ . nℓ .95) n1 δ3 (k − k1 ) + · · · + nℓ δ3 (k − kℓ ) |n1 . k1 . the state |n1 .80)) N (k).5. k1 . kℓ . n2 . of course. k1 . kℓ . kℓ contains n1 quanta with four µ µ momentum k1 . nℓ . k2 .97) H|n1 . nℓ . nℓ . · · · . the momentum operator can also be derived from N¨ether’s theorem and o coincides exactly with this in the normal ordered form as we will see later. · · · .

· · · . (5. let us analyze the state obtained from the vacuum by applying a single creation operator. n2 .190 5 Free Klein-Gordon field theory P|n1 .7 Physical meaning of energy eigenstates To obtain the physical meaning of these energy eigenstates. nℓ . k2 . n2 . p = n1 k1 + n2 k2 + · · · + nℓ kℓ . · · · .102) that H 2 − P2 |k 2 = (Ek − k2 )|k = m2 |k . = Ek |k .66) as well as the properties (5.101) This state satisﬁes N |k H|k P|k = |k .103) . k1 . kℓ (5. nℓ . k2 . |k = a† (k)|0 . k2 . kℓ . 5. kℓ = d3 k k N (k)|n1 . k1 . these states are eigenstates of H and P with total energy and momentum given by E = n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ . · · · .90)) that these states are orthonormal and deﬁne a complete basis of states for the Hilbert space of the Klein-Gordon theory. = k|k . nℓ . (5.99) = (n1 k1 + n2 k2 + · · · + nℓ kℓ ) |n1 . (5. (5. Namely.102) It follows from (5. n2 .100) We can show in a straightforward manner (using the fundamental commutation relations in (5. k1 .

104) µ can be thought of as the state with n1 particles with k1 = (Ek1 . it is clear that 0|φ(x)|0 = 0.106) In fact. ﬁeld theories naturally describe systems of many identical particles. k). nℓ . (5. This function satisﬁes ∂µ ∂ µ + m2 k|φ(x) = k| ∂µ ∂ µ + m2 φ(x)|0 = 0. (5. Similarly. of course.5. |φ(x) = φ(x)|0 = φ(−) (x)|0 . (5.105) as well as the fact that only the ﬁeld operator φ(x) depends on the space-time coordinates on which the . k1 ). we can think of this as the one particle state with four momentum kµ = (Ek . Note that the particles described by such states are necessarily identical (although with diﬀerent energy and momentum) and. we can show that the state |n1 . k2 .7 Physical meaning of energy eigenstates 191 This state. k2 ) and so on.107) This is. (5. as a result. therefore. (5. therefore. an ordinary function representing the projection of the state |φ(x) on to the one particle state |k . satisﬁes the single particle Klein-Gordon equation with positive energy (it provides a representation for the Poincar´ e group) and. kℓ . the only nonzero matrix element of the ﬁeld operator involving the vacuum state is given by k|φ(x)|0 = k|φ(x) . µ n2 particles with k2 = (Ek2 . · · · . n2 . Let us next consider the state that is produced by the ﬁeld operator acting on the vacuum. k1 .105) Since φ(x) is linear in the creation and the annihilation operators.108) where we have used (5.

(5. Let us note here parenthetically that the wave function for a single particle is really identiﬁed with φ(x)|k . |k represents an energy eigenstate of the system describing a single particle (namely. k|φ(x) deﬁnes a solution of the classical Klein-Gordon equation and we can relate this function to the single particle Klein-Gordon wave function with positive energy. |ψ = |k ). a† (k′ ) + a† (k′ )a(k) 0 = = d3 k ′ (2π)3 2k′ 0 1 (2π)3 2k0 e−ik ·x δ3 k − k′ ′ ′ ∗ e−ik·x . (5.109) to be the positive energy plane wave solutions of the single particle Klein-Gordon equation.109) √ with k0 = Ek = k 2 + m2 > 0. we note explicitly that φ(x)|k = = k|φ(x) ∗ = 0|a(k)φ(−) (x)|0 d3 k ′ (2π)3 2k′ 0 e ik ′ ·x † ′ ∗ ik ′ ·x ∗ 0 a(k) d3 k ′ a (k )e 0 ∗ = = (2π)3 2k′ 0 d3 k ′ (2π)3 2k′ 0 0|a(k)a (k )|0 † ′ × eik ·x 0 a(k).110) In the second quantized description. We recognize the function in (5. In the present case. then.192 5 Free Klein-Gordon field theory Klein-Gordon operator ( + m2 ) can act. . If |ψ denotes a state of the system. the wave function is given by ψ(x) = x|ψ . in the ﬁrst quantized description. This can be understood by noting that |φ(x) is like the coordinate basis state |x (which is the eigenstate of the operator representing the dynamical variable) of the ﬁrst quantized description. Therefore. This brings out the connection between the second quantized theory and the ﬁrst quantized theory.

φ (x2 ) = k. As we have seen.109). positive semi-deﬁnite and there is no problem of negative . k′ ) at the coordinates x1 and x2 at a given time. x0 = x0 = · · · = x0 ).114) would give the probability for ﬁnding two Klein-Gordon particles with four momenta kµ = (Ek . We can similarly construct multi-particle states in the conﬁguration space of the form |φ (x1 ) .112) can be thought of as the one particle state in the conﬁguration space – describing the quantum mechanical state of the single particle at the coordinate x. Thus. Such probabilities are. k′ |φ (x1 ) . such a state contains a superposition of all possible momentum states (which follows from the ﬁeld decomposition in (5. the wave functional for a single particle with a deﬁnite energymomentum has the form ψ(φ(x)) = φ(x)|ψ = φ(x)|k .7 Physical meaning of energy eigenstates 193 Thus. k′ |φ(−) (x1 ) φ(−) (x2 ) |0 . which is what we have seen explicitly in (5. These states naturally lead to quantum mechanical probabilities which are non-negative. (5. · · · . such conﬁguran 1 2 tion states are automatically symmetric as we would expect for a system of identical Bose particles. by construction. (5. Furthermore. Such states are physically meaningful only if the time coordinates are equal (namely. · · · . They can also be shown to deﬁne a complete basis and describe what is known as the Fock space for the system.111) |φ(x) (5. xn . φ (xn ) = φ(−) (x1 ) φ(−) (x2 ) · · · φ(−) (xn ) |0 . φ (x2 ) .113) which will describe a state with n particles at coordinates x1 . the absolute square of the amplitude (with x0 = x0 ) 2 1 k. for example. (5. k) and k′ µ = (Ek′ .5.56)). The state = φ(−) (x)|0 .

eventually we would like to study interactions of the system. for the simple case of the Klein-Gordon ﬁeld interacting with an external source J(x) (which is a c-number function or a classical function). In a realistic theory the ﬁeld can have self-interactions or can interact with other dynamical ﬁelds of the system.194 5 Free Klein-Gordon field theory probabilities in the second quantized description. this equation is covariant (in fact. 2 2 (5. which we will study in the following chapters. the translation invariance of the Green’s function arises because the right-hand side of the equation is invariant under translations. If we know the Green’s function in (5. the Klein-Gordon equation will modify. e . all the physical states have positive semideﬁnite energy and that the probabilities are non-negative as they should be. (5. For example. it is invariant under a Lorentz transformation) leading to the fact that the Green’s function is Poincar´ invariant.115) provided we know the Green’s function of the system.116) as the Euler-Lagrange equation (5.115) L= (5. Thus. namely. 5. The Green’s function for a given inhomogeneous equation is deﬁned as the solution of the equation with a delta source. the Green’s function G(x−y) satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y).117) Here. the second quantized noninteracting Klein-Gordon theory is free from the problems of the ﬁrst quantized theory that we had discussed earlier. Furthermore.117). which can be obtained from the Lagrangian density m2 2 1 ∂µ φ∂ µ φ − φ + Jφ. When there are interactions present. for the Klein-Gordon equation.8 Green’s functions Even though we have been examining the free Klein-Gordon ﬁeld theory so far.14). namely. Even classically we know that we can solve an inhomogeneous equation of the kind (5. the equation will take the form ∂µ ∂ µ + m2 φ(x) = J(x).

The Green’s function is. Thus.118) d4 y −δ4 (x − y) J(y) (5. therefore.120) d4 k −ik·(x−y) e G(k). (5. We note here that a homogeneous solution can always be added to the particular solution depending on the system under study (and sometimes in order to implement appropriate boundary conditions). . or. − m2 (5.119) = J(x).117). deﬁning the Fourier transforms δ4 (x − y) = G(x − y) = 1 (2π)4 d4 k e−ik·(x−y) . 4 (2π) (2π)4 G(k) = k2 1 .121) In other words. (2π)4 and substituting them back into (5.8 Green’s functions 195 then the particular solution of the inhomogeneous equation (5. an important concept in studying the solutions of a system when interactions are present and can be easily determined by going over to the momentum space. the Fourier transformation turns the partial diﬀerential equation into an algebraic equation which is trivial to solve. 1 1 −k2 + m2 G(k) = − . or. we obtain ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y).5.115) can be written as φ(x) = − so that ∂µ ∂ µ + m2 φ(x) = − = − µ d4 y ∂xµ ∂x + m2 G(x − y)J(y) d4 y G(x − y)J(y). (5.

4. or. the Green’s function is not uniquely deﬁned until we specify a contour of integration in the complex k0 -plane and specifying a contour simply corresponds to specifying a boundary condition for the Green’s function. (2π)4 k2 − m2 (5.3.122) G(x − y) = = This explicitly shows that the Green’s function is Poincar´ invariant.3: Poles of the integrand in the complex k0 -plane.120).122) has poles at k2 − m2 = 0. (5. or.121) back into (5. 5. If we choose a contour of the form shown in Fig. e Let us note that the integrand of the Green’s function in (5. Therefore. substituting (5. this is expressed by writing .196 5 Free Klein-Gordon field theory Thus. (k0 )2 = k2 + m2 . we determine the Green’s function for the Klein-Gordon equation to be d4 k −ik·(x−y) e G(k) (2π)4 d4 k e−ik·(x−y) . this is equivalent to moving the poles inﬁnitesimally into the upper half plane.123) which lie on the real axis in the complex k0 -plane as shown in Fig. then. 5. k0 = ±Ek . Im k 0 −Ek Ek Re k 0 Figure 5. Mathematically.

the integral will vanish. k0 = ±Ek + iǫ. we have to close the contour in the upper half plane (for the damping of the exponential) in which . or.8 Green’s functions 197 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek + iǫ Ek + iǫ Re k 0 Figure 5.125) where we have identiﬁed ǫ = η .126) For x0 − y 0 < 0.4: Choice of the contour in the complex k0 -plane for the advanced Green’s function. or. Therefore. (5. (5. We note that if x0 − y 0 > 0.124) are in the upper half plane. 2 Since in this case the poles in (5. if we enclose the contour in the lower half plane. then we must close the contour in the lower half plane for the damping of the exponential in (5. we conclude that GA (x − y) = 0.124).124) In this case. the poles of the integrand occur at k2 − m2 − ik0 η = (k0 )2 − k2 − m2 − ik0 η = 0.5. or. iη k − 2 0 2 ≃ k 2 + m2 . k0 − iǫ = ±Ek . GA (x − y) = lim η→0+ e−ik·(x−y) d4 k . on the other hand. (2π)4 k2 − m2 − ik0 η (5. for x0 − y 0 > 0.

129) would appear at . sin Ek x0 .5 or equivalently if we push both the poles to the lower half plane. if we choose the contour in the complex k0 -plane as shown in Fig. (2π)3 Ek This is known as the advanced Green’s function and has support only in the past light cone. For x0 − y 0 < 0. then. Explicitly. for x0 < 0. 5.127) d3 k eik·(x−y) sin Ek x0 − y 0 . we have d3 k (2π)4 d3 k (2π)4 e−ik·(x−y) (k0 − iǫ + Ek ) (k0 − iǫ − Ek ) GA (x − y) = lim = lim 2πi ǫ→0+ dk0 ǫ→0+ 0 0 e−i(Ek +iǫ)(x −y )+ik·(x−y) 2Ek 0 0 e−i(−Ek +iǫ)(x −y )+ik·(x−y) + −2Ek = = i 2 d3 k eik·(x−y) (2π)3 Ek e−iEk (x 0 −y 0 ) − eiEk (x0 −y0 ) (5. we can write (we denote the argument of the Green’s function by x for simplicity) G (x) = A 0.129) In this case. GA (x) = θ −x0 d3 k eik·x sin Ek x0 . 4 k 2 − m2 + ik 0 η (2π) GR (x) = lim η→0+ (5. the poles of the integrand in (5.198 5 Free Klein-Gordon field theory case we can evaluate the integral by the method of residues. we can express this mathematically by deﬁning e−ik·x d4 k .128) or. d3 k eik·x (2π)3 Ek for x0 > 0. (2π)3 Ek Similarly. (5.

x0 < 0. on the other hand. therefore. since both the 2 poles are in the lower half plane. we have to close the contour in the upper half plane for the damping of the exponential in (5.130) where we have again identiﬁed ǫ = η . k0 + iǫ ≃ ±Ek . we conclude that GR (x) = 0. or. k0 = ±Ek − iǫ. we have to close the contour in the lower half plane (for the damping of the exponential) and the method of residues gives GR (x) = ǫ→0+ lim d3 k (2π)4 dk0 e−ik·x (k0 + iǫ + Ek ) (k0 + iǫ − Ek ) . (5.131) For x0 > 0. or. (5. Note that if x0 < 0. k2 − m2 + ik0 η = (k0 )2 − k2 − m2 + ik0 η = 0.129) and. In this case.5: Choice of the contour in the complex k0 -plane for the retarded Green’s function. if we close the contour in the upper half plane. k0 + iη 2 2 2 ≃ k2 + m2 = Ek .8 Green’s functions 199 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek − iǫ Ek − iǫ Re k 0 Figure 5.5. or. then the integral will vanish.

the average of the two. (5.132) e−i(−Ek −iǫ)x + −2Ek 0 +ik·x The overall negative sign in (5. 1 GA (x) + GR (x) .135) would also deﬁne a Green’s function corresponding to the contour picking up the principal values at the poles.132) arises because the contour is clockwise in the lower half plane. namely. the diﬀerence between GA (x) and GR (x) would deﬁne a function which would satisfy the homogeneous (Klein-Gordon) equation and . (2π)3 Ek (5. Thus.) It is clear from (5.128) and (5. (2π)3 Ek Note that this Green’s function has support only in the future light cone and is known as the retarded Green’s function.133) or.133) that GR (x) = GA (−x).134) Both the retarded and the advanced Green’s functions satisfy the inhomogeneous equation (5.200 = lim (−2πi) 5 Free Klein-Gordon field theory ǫ→0+ d3 k e−i(Ek −iǫ)x (2π)4 2Ek 0 +ik·x = − = − i 2 d3 k eik·x −iEk x0 0 − eiEk x e 3 E (2π) k d3 k eik·x sin Ek x0 . − d3 k eik·x (2π)3 Ek GR (x) = for x0 < 0. sin Ek x0 . GR (x) = −θ x0 d3 k eik·x sin Ek x0 . 2 (5.117) with a delta function source. Therefore. On the other hand. we see that we can write 0. (5. (The advanced and the retarded Green’s functions have similar form except for their support and the overall sign. for x0 > 0.

139) The retarded and the advanced Green’s functions play an important role in many classical calculations as well as in many calculations in statistical mechanics. (2π)3 Ek This clearly satisﬁes the homogeneous Klein-Gordon equation ∂µ ∂ µ + m2 G(x) = 0. known as the Feynman Green’s function.136) + θ x0 = = θ x0 + θ −x0 d3 k eik·x sin Ek x0 . (5. and is antisymmetric. G(x) = −G(−x). namely.5.6.136) that the Schwinger function is real (G(x))∗ = G(x).137) Furthermore. it is easily seen from (5. however. which is more useful in the calculation of scattering matrix elements. 5. This is deﬁned with the contour in the complex k0 -plane as shown in Fig.138) (5. which corresponds to pushing one of the poles (on the left) to the upper half plane while moving the other (on the right) .8 Green’s functions 201 would not strictly correspond to a Green’s function of the theory. Conventionally such a function is known as the Schwinger function and is deﬁned to be G(x) = GA (x) − GR (x) = θ −x0 d3 k eik·x sin Ek x0 (2π)3 Ek d3 k eik·x sin Ek x0 (2π)3 Ek d3 k eik·x sin Ek x0 (2π)3 Ek (5. there is another Green’s function. (5. Quantum mechanically.

there will be a nontrivial contribution independent of whether we close the contour in the upper half or in the lower half plane.6: Choice of the contour in the complex k0 -plane for the Feynman Green’s function. or. the poles of the integrand in (5. or.140) occur at k2 − m2 + iη = 0.140) In this case. Since there are poles in both halves of the complex k0 -plane. or. to the lower half plane. 2 (k0 )2 − Ek + iη = 0.141) η where we have identiﬁed ǫ = 2Ek with η → 0+ . Mathematically. (5. = ± (Ek − iǫ) . (k0 )2 − Ek − k0 = ± Ek − iη 2Ek iη 2Ek 2 ≃ 0. (5.142) . this can be implemented by deﬁning d4 k e−ik·x . For x0 < 0 we have to close the contour in the upper half plane for the damping of the exponential and this will pick up the residue of the pole at k0 = −Ek + iǫ. in this case. (2π)4 k2 − m2 + iη GF (x) = lim η→0+ (5.202 Im k 0 5 Free Klein-Gordon field theory Im k 0 −Ek Ek Re k 0 → −Ek + iǫ Ek − iǫ Re k 0 Figure 5.

then we have to close the contour in the lower half plane which will pick up the residue of the pole at k0 = Ek − iǫ. we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 (5. we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 GF (x) = = ǫ→0+ lim ǫ→0+ lim 2πi d3 k e−i(−Ek +iǫ)x +ik·x (2π)4 2 (−Ek + iǫ) 0 i = − 2 = − i 2 d3 k eiEk x +ik·x (2π)3 Ek d3 k eik·x (2π)3 k0 (5. Therefore. for x0 < 0. On the other hand.145) = −G(+) (x).5. for x0 > 0. if x0 > 0.143) = G(−) (x). Here the contour is clockwise leading to an overall negative sign. The Feynman Green’s function clearly has support in both the future as well as the past light cones and can be written as .144) GF (x) = = ǫ→0+ lim ǫ→0+ lim (−2πi) i 2 i 2 d3 k e−i(Ek −iǫ)x +ik·x (2π)4 2 (Ek − iǫ) 0 = − = − d3 k e−iEk x +ik·x (2π)3 Ek d3 k e−ik·x (2π)3 k0 (5.8 Green’s functions 203 Therefore. where. we have used k ↔ −k in the last step (and we understand that k0 = Ek ).

136) as d3 k eik·x sin Ek x0 = G(+) (x) + G(−) (x).146) G(+) (−x) = (5. we will not go into this in more detail here.150) The functions G(+) (x) and G(−) (x) (see (5.143) and (5. Note from (5. the Feynman Green’s function is an even function.147) Therefore. another Green’s function that is quite useful in studies of ﬁnite temperature ﬁeld theory (as well as in the study of unitarity relations such as the cutting rules).149) for which the pole at k0 = −Ek is pushed down to the lower half plane while the pole at k0 = Ek is moved up to the upper half plane.143) and (5. However. (2π)3 d4 k θ(−k0 )δ(k2 − m2 ) e−ik·x .145) that we can write the Schwinger function (5. We note from the structure of G(−) (x) in (5. (2π)3 (5. (2π)3 Ek G(x) = (5. (2π)4 k2 − m2 − iη G(x) = lim η→0+ (5.145)) can be written in the manifestly covariant forms d4 k θ(k0 )δ(k2 − m2 ) e−ik·x .143) and G(+) (x) in (5. It is deﬁned as e−ik·x d4 k .151) G(+) (x) = i G(−) (x) = −i . (2π)3 k0 (5. (5. GF (−x) = GF (x). namely. therefore.148) There is yet another choice of the contour and.204 5 Free Klein-Gordon field theory GF (x) = −θ x0 G(+) (x) + θ −x0 G(−) (x).145) that i 2 d3 k eik·x = −G(−) (x).

150) that we can write GA (x) = θ(−x0 )G(x) = θ(−x0 ) G(+) (x) + G(−) (x) . (5.151) that the Schwinger function has the manifestly covariant representation G(x) = G(+) (x) + G(−) (x) = i d4 k ǫ(k0 )δ(k2 − m2 ) e−ik·x . (2π)3 (5.9 Covariant commutation relations 205 and are correspondingly known as the positive and the negative energy (frequency) Green’s functions. let us calculate some of the covariant commutation relations satisﬁed by the ﬁeld operators.9 Covariant commutation relations To make contact between the quantum ﬁeld theory and the various Green’s functions that we have constructed.58) that φ(x) = φ(+) (x) + φ(−) (x).154) which will be useful in our later discussions.151)) G(+) (x) ∗ = G(−) (x).133) and (5. They satisfy (see (5.57) that (we recall that k0 = Ek = 0) √ (5. Let us recall from (5.128). where we note from (5. For completeness. (5.5. GR (x) = −θ(x0 )G(x) = −θ(x0 ) G(+) (x) + G(−) (x) .153) where ǫ(x) = (θ(x) − θ(−x)) is known as the sign function (or the alternating step function). we note from (5. We note from (5.152) and all the Green’s functions can be expressed as linear combinations of these two fundamental Green’s functions.155) k2 + m2 > . (5. 5.

a† (k′ ) . a† (k′ ) e e 2k0 2k′ 0 = 0.157) d3 k d3 k′ −ik·x −ik′ ·y √ √ e e a(k).156) Since we know the commutation relations (5. it is straightforward to evaluate φ(+) (x). eik·x a† (k). (5. a(k). φ(−) (x). d3 k ′ (2π)3 2k′ 0 eik ·y a† (k′ ) ′ d3 k d3 k′ ik·x ik′ ·y † √ √ a (k). φ(+) (x). (2π)3 2k0 d3 k ′ ′ eik ·y a† (k′ ) (2π)3 2k′ 0 . a(k). d3 k ′ (2π)3 2k′ 0 e−ik ·y a(k′ ) ′ (5. a(k′ ) 0 ′0 2k 2k = 0. φ(+) (y) = = 1 (2π)3 d3 k (2π)3 2k0 e−ik·x a(k).66) between the creation and the annihilation operators. φ(−) (y) = d3 k e−ik·x a(k). a(k′ ) = 0 = a† (k).206 5 Free Klein-Gordon field theory φ(+) (x) = φ(−) (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k). φ(−) (y) = = 1 (2π)3 d3 k (2π)3 2k0 eik·x a† (k). a† (k′ ) = δ3 k − k′ .

.5.158) = −iG(+) (x − y).158) and (5.161) simply because the ﬁeld operator φ(x) satisﬁes the Klein-Gordon equation. φ(y)] = φ(+) (x) + φ(−) (x). φ(−) (y) + φ(−) (x). furthermore. φ(+) (y) = iG(+) (y − x) = −iG(−) (x − y). (2π)3 Ek (5. It is now clear that the Schwinger function satisﬁes the homogeneous equation µ ∂x µ ∂x + m2 G(x − y) = 0. a† (k′ ) e 2k0 2k′ 0 d3 k d3 k′ −ik·x+ik′·y 3 √ √ e δ k − k′ 2k0 2k′ 0 d3 k −ik·(x−y) e 2k0 (5.158) and (5.159) are not restricted to have equal time arguments any more (unlike in (5.160) Namely. (5. φ(+) (y) + φ(−) (y) = φ(+) (x). the commutator of two ﬁeld operators.40)) and. (5.159) Note that the ﬁeld operators in the commutators (5.159). at unequal times. is proportional to the Schwinger Green’s function. using the results in (5.9 Covariant commutation relations 207 = = = 1 (2π)3 1 (2π)3 1 (2π)3 d3 k d3 k′ −ik·x+ik′·y √ √ a(k). It follows now that φ(−) (x). φ(+) (y) = −iG(+) (x − y) − iG(−) (x − y) = −iG(x − y) = −i d3 k eik·(x−y) sin Ek x0 − y 0 . we can evaluate [φ(x).

(5. this relation can also be seen to imply that [φ(x). therefore. (5.160) also shows that two ﬁeld operators φ(x) and φ(y) do not commute for arbitrary values of the coordinates. the commutator of two ﬁeld operators vanishes.164) which is expected from the anti-symmetry of the Schwinger function (5.208 5 Free Klein-Gordon field theory ∂µ ∂ µ + m2 φ(x) = 0. a light signal cannot connect two space-like points and. This is consistent with our intuitive expectation. Equation (5.165) x0 =y 0 We note that by a Lorentz transformation. namely.138) and (5. This is known as the principle of microscopic causality.162) These commutation relations are known as covariant commutation relations simply because the (scalar) Green’s functions are invariant under Lorentz transformations.166) In other words. for (x − y)2 < 0. (5. two measurements at space-like separations should not inﬂuence each other. for space-like separations. (The non-commutativity is a reﬂection of the Hamiltonian dynamics. φ(y)]x0 =y0 = −iG(x − y) = 0. as can be checked even in the simple classical Poisson bracket relations for a one dimensional free particle or a harmonic oscillator.40)) [φ(x). (5. x0 =y 0 (5.) On the other hand. We also note that .163) where Λµν represents a Lorentz transformation. Namely. for x0 = y 0 d3 k eik·(x−y) sin Ek x0 − y 0 (2π)3 Ek G(x − y)|x0 =y0 = = 0.160) leads to the familiar relation (see (5. G(Λx) = G(x). φ(y)] = 0.

(2π)3 This is consistent with our earlier ﬁeld quantization rule (5. φ(y)] = −iθ(x0 − y 0 ) G(+) (x − y) + G(−) (x − y) = −iθ(x0 − y 0 )G(x − y) = iGR (x − y). (5. φ(y)] = −iθ(y 0 − x0 ) G(+) (x − y) + G(−) (x − y) = −iθ(y 0 − x0 )G(x − y) = −iGA (x − y). Bjorken and S.40). (5. φ(y) [φ(x).167) d3 k ik·(x−y) e = −δ3 (x − y). ˙ φ(x).136). 1964. namely.10 References 1. McGrawHill.160) as well as the relations in (5. 5. let us note from (5. Relativistic Quantum Fields. J. which will be useful later.5.154) that the retarded and the advanced Green’s function can also be expressed in terms of covariant commutation relations as θ(x0 − y 0 ) [φ(x). Π(y)]x0 =y0 = x0 =y 0 = −i ∂G(x − y) ∂y 0 x0 =y 0 = iδ3 (x − y).168) To conclude this section.150) and (5. D. θ(y 0 − x0 ) [φ(x). Drell.10 References 209 ∂G(x − y) ∂y 0 = − x0 =y 0 =− d3 k eik·(x−y) Ek cos Ek x0 − y 0 |x0 =y0 (2π)3 Ek (5. (5. New York.169) .

Lectures on Quantum Mechanics. Quantum Field Theory. Introduction to Relativistic Quantum Field Theory. John Wliley. 4. Itzykson and J-B.210 5 Free Klein-Gordon field theory 2. Peterson. New Delhi. Evanston (1961). A. New York (1969). . McGrawHill. Zuber. Schweber. Roman. Hindustan Publishing. Das. Introduction to Quantum Field Theory. 1980. 5. P. India. 2003. C. S. 3. Row. New York.

3) if the action does not change under these transformations. S= (6.1) is invariant under the transformations (6.1 N¨ther’s theorem o In trying to extend the free Klein-Gordon ﬁeld theory to include interactions.2) φ(x) → φ′ (x′ ) . in the present case. Under a general transformation of the form xµ → x′ µ . we have to be guided by some symmetry principles and the question we have to answer is how we can incorporate the concept of symmetry into the ﬁeld theoretic framework. Let us suppose that we have a dynamical system described by the action d4 x L.1) where. Obviously.3) ′ ∂µ φ′ (x′ ) . we say that the dynamics described by the action (6. if 211 . ∂µ φ(x)). Namely. the ﬁrst question that we face is how we can choose one interaction term over another. (6. we assume L = L(φ(x). ∂µ φ(x) → (6.Chapter 6 Self-interacting scalar ﬁeld theory 6.

4) then. d4 x L (φ(x). ∂µ φ(x)) = ′ d4 x′ L(φ′ x′ . d4 x L φ′ (x).3). let us consider an inﬁnitesimal transformation with ∂x′ = 1. as indicated in (6. (6. ∂µ φ′ (x) − L (φ(x). ∂µ φ′ x′ δS = or. Thus.3) includes a very interesting class of transformations where the space-time coordinates do not change. namely. or.212 6 Self-interacting scalar field theory S= d4 x L (φ(x). ∂µ φ(x)) = 0.3) would imply ′ d4 x′ L φ′ x′ . the Euler-Lagrange equations for the primed and the unprimed systems would remain form invariant.3) deﬁne a symmetry of the system. Our discussion of symmetries applies to both spacetimetransformations where space-time coordinates are transformed as well as internal symmetry transformations where space-time coordinates are unaﬀected by the transformation. the transformations in (6. ∂µ φ(x)) = 0. ∂x (6.5) and only the dynamical variables of the theory transform.6) which holds for most global space-time symmetries as well as internal symmetries. Symmetries have interesting consequences for continuous transformations. x′ µ = xµ .7) . (6. Such transformations are known as internal symmetry transformations to be contrasted with space-time transformations where space-time coordinates transform along with the dynamical variables. ∂µ φ(x)) = 0. It is worth emphasizing here that (6. ∂µ φ′ x′ ). It is clear that in such a case. ∂µ φ′ (x) − d4 x L φ′ (x). (6. In such a case. − d4 x L (φ(x). invariance of the action under the inﬁnitesimal forms of the transformations in (6.

keep only linear terms in δφ(x)) L φ′ (x). ∂µ φ(x)) = ∂µ K µ .8) which must hold independent of the use of equations of motion for the system under study.9) we can calculate explicitly (we remember that δφ(x) is inﬁnitesimal and. ∂µ φ(x)) = L (φ(x).7) will hold if L φ′ (x).¨ 6. (6. ∂µ φ(x)) + δφ(x) −L (φ(x). so that δ(∂µ φ(x)) = ∂µ φ′ (x) − ∂µ φ(x) = ∂µ δφ(x). (6. therefore. as we will see later.10) (6. ∂µ φ′ (x) − L (φ(x). On the other hand. ∂µ φ(x)) = δφ(x) ∂L ∂L + (∂µ δφ(x)) ∂φ(x) ∂∂µ φ(x) ∂L ∂L + (∂µ δφ(x)) ∂∂µ φ(x) ∂∂µ φ(x) ∂L ∂∂µ φ(x) . (6.11) ∂L ∂L + δ(∂µ φ(x)) ∂φ(x) ∂∂µ φ(x) = δφ(x)∂µ = ∂µ δφ(x) . Clearly. This is quite general and it is possible to have symmetry transformations for which K µ = 0. In fact. deﬁning the inﬁnitesimal change in the ﬁeld variable as (this is known as the Lie derivative of the ﬁeld variable up to a sign) φ′ (x) − φ(x) = δφ(x). ∂µ φ′ (x) − L (φ(x). (6. this is indeed the case for internal symmetry transformations.1 Nother’s theorem 213 where we have identiﬁed the integration variable in the ﬁrst term on the right-hand side to be x (instead of x′ ) in the intermediate step.

However.12) ∂L − Kµ ∂∂µ φ(x) This shows that whenever there is a continuous symmetry associated with a system. given a conserved current. φ′ must also satisfy the same equation of motion in the transformed frame as the original ﬁeld. Namely. we can deﬁne a charge Q= d3 x J 0 (t. In fact. we can deﬁne a current ∂L − K µ.214 6 Self-interacting scalar field theory Here we have used the Euler-Lagrange equation in the intermediate steps. not all such variations will deﬁne a symmetry.15) which will be a constant (in time) and will generate the symmetry transformations. ∂∂µ φ(x) J µ (x) = δφ(x) (6. there is no restriction on δφ and. namely. For a symmetry. therefore. we obtain ∂µ δφ(x) or. on φ′ . ∂µ J µ (x) = 0. (6. First.13) which is conserved. on the other hand.8) and (6. in general. This is just the principle of least action. for any inﬁnitesimal variation.11). (6. ∂µ δφ(x) ∂L ∂∂µ φ(x) = ∂µ K µ = 0. x) .) Comparing (6.14) Several comments are in order here. this is always true when the equations of motion are used. (6. Its tensor structure is determined by the tensor structure of the inﬁnitesimal parameter of transformation. the conserved current independent of the parameter of transformation is not always a vector. we note that . (Note that. Second.

then it generates inﬁnitesimal transformations (through commutation relations) which deﬁne a symmetry of the theory.16) Here we have used the fact that. (6. consequently. with the usual assumptions on the asymptotic fall oﬀ of the ﬁeld operators (see (5. The converse is also true.17) and.18) where the parameter of translation ǫµ is assumed to be inﬁnitesimal and constant (global). there exists a conserved charge which is the generator of these inﬁnitesimal symmetry transformations. let us consider the simple case of an inﬁnitesimal o global space-time translation deﬁned by xµ → x′µ = xµ + ǫµ . H] = 0. (6.1 Space-time translation. Q corresponds to the generator of the inﬁnitesimal symmetry transformations of the theory. or. Namely. . if we have a continuous symmetry in the theory (transformations under which the action is invariant). δxµ = x′µ − xµ = ǫµ .¨ 6.3)). if there exists a conserved charge in a given theory. the time independence of Q also means that it commutes with the Hamiltonian of the theory [Q. This is known as N¨ther’s theorem and is quite important in the study of symmeo tries in dynamical systems.1 Nother’s theorem 215 dQ dt = = = d3 x ∂0 J 0 d3 x ∂0 J 0 + ∂i J i d3 x ∂µ J µ = 0.1. In this case. In the operator language. As an example of the consequences of N¨ther’s theorem. (6. 6. Q and H can have simultaneous eigenstates. we are dealing with a scalar ﬁeld. Since. the integral of a total divergence (∇ · J) vanishes. Thus.

Next. (6.9) and this is how it corresponds to the negative of the Lie derivative) δφ(x) = φ′ (x) − φ(x) = φ′ (x) − φ′ (x′ ) = −ǫµ ∂µ φ(x).19) and in this case we obtain the change in the ﬁeld to correspond to (see (6.8)) L φ′ (x).22) .21) = −ǫµ ∂µ φ(x) = −ǫµ ∂µ L = ∂µ K µ . ∂µ φ(x)) = δφ(x) ∂L ∂L + (∂ν δφ(x)) ∂φ(x) ∂∂ν φ(x) ∂L ∂L − ǫµ (∂µ ∂ν φ(x)) ∂φ(x) ∂∂ν φ(x) (6. ∂µ φ′ (x) − L (φ(x). With this.7) and (6. where we have used (5.10).21) that (ǫµ is a constant parameter) K µ = −ǫµ L. = − φ′ x′ − φ′ (x) = −ǫµ ∂µ φ′ (x) (6. (6. the action is invariant under inﬁnitesimal translations which deﬁne a symmetry of the system.20) where in the last step.20).216 6 Self-interacting scalar field theory φ′ x′ = φ (x) . we can now calculate explicitly the change in the Lagrangian density (see (6. then. we note that under the transformation (6. we have identiﬁed φ′ (x) = φ(x) simply because the parameter ǫµ multiplying on the right-hand side is already inﬁnitesimal and any further correction coming from φ′ (x) will only be of higher order. We can now identify from (6. We see that since the change in the Lagrangian density is a total divergence.11) and (6.

(6. we can always deﬁne an improved symmetric stress tensor by coupling the theory to a gravitational background and taking variation with respect to the gravitational background. even when the naive N¨ther o procedure does not lead to a symmetric stress tensor. namely. that the conserved current independent of the parameter can be identiﬁed with ∂L − η µν L.¨ 6.1 Nother’s theorem 217 δφ(x) ∂L ∂L = −ǫν (∂ν φ(x)) . (6.25) which can be easily checked to satisfy (using the equations of motion) ∂µ T µν = 0. ∂∂µ φ(x) T µν = (∂ ν φ(x)) (6. (6.26) T µν is known as the stress tensor of the theory and can always be deﬁned to be symmetric (It is the source for the gravitational ﬁeld gµν just as the electromagnetic current j µ is the source for the electromagnetic potential Aµ .24) µ Jǫ (x) = δφ = −ǫν (∂ν φ(x)) = −ǫν (∂ ν φ(x)) It is clear.). depending on the parameter of transformation.23) so that we obtain the current associated with the symmetry transformation. T µν = T νµ . Therefore. to be (see (6. therefore.27) .13)) ∂L − Kµ ∂∂µ φ(x) ∂L + ǫµ L ∂∂µ φ(x) ∂L − η µν L ∂∂µ φ(x) = −ǫν T µν . ∂∂µ φ(x) ∂∂µ φ(x) (6.

We note that. let us recall that the Lagrangian density has the form L= m2 2 1 ∂µ φ∂ µ φ − φ . would correspond to the components of a four vector which we identify with the energy-momentum operator as Pµ = d3 x T 0µ . in this case. should generate inﬁnitesimal space-time translations. ∂∂µ φ(x) and this leads to the explicit form for the stress tensor (6. For the free Klein-Gordon theory.29) Therefore.31) = ∂ ν φ(x)∂ µ φ(x) − η µν L = T νµ . is manifestly symmetric.) As a result. which.30) T µν = ∂ ν φ(x) ∂L − η µν L ∂∂µ φ(x) (6. for the free theory. 2 2 (6. (6.28) This is consistent with our physical intuition that the energy-momentum. in the present case. .15)). is a second rank tensor simply because the parameter of transformation is a four vector. in this case. (In general.218 6 Self-interacting scalar field theory The conserved current. we obtain ∂L = ∂ µ φ(x). P µ . the conserved charges associated with the symmetry transformation (see (6.25) (6. the current is a tensor one rank higher than the parameter of transformation.

since the free Lae grangian density is invariant under the discrete transformation φ(x) ↔ −φ(x).34) . We had also given an expression for the momentum operator in terms of creation and annihilation operators in (5.2 Self-interacting φ4 theory 219 P0 = = = = H.34). (6.97) and it can be checked that P in (6. 6. Pi = = P = − d3 x T 00 d3 x d3 x ˙ φ(x) 2 − 2 1 ˙ φ(x) 2 2 m2 2 1 φ + ∇φ · ∇φ + 2 2 1 ˙ φ(x) 2 1 m2 2 + ∇φ · ∇φ + φ 2 2 d3 x T 0i ˙ d3 x ∂ i φ(x)φ(x).6. ˙ d3 x ∇φ(x)φ(x).33) To include interactions. Furthermore.2 Self-interacting φ4 theory Let us denote the free part of the Klein-Gordon Lagrangian density as m2 2 1 ∂µ φ∂ µ φ − φ . we note that the Lagrangian density for the interaction must be invariant under Lorentz transformations as well as translations (Poincar´ invariant). 2 2 L0 = (6.32) We see that P 0 = H coincides with the form of the Hamiltonian we had derived earlier in (5.32) corresponds exactly to the former expression when normal ordered and expressed in terms of creation and annihilation operators. (6.

4! LI = − λ > 0. Thus. in units of = c = 1 (which we have been using). in fact. (6. we can show that the canonical dimension of any variable can be expressed in powers of an arbitrary mass dimension [M ].220 6 Self-interacting scalar field theory we would like to preserve this symmetry in the interactions as well.) As a result. . (6. for various other reasons we can show that (6. let us introduce the concept of canonical dimensions. To explain this brieﬂy. With these conditions. (6. a Hamiltonian) which is unbounded from below for any value of the coupling constant g. In these units.35) represents the only meaningful interaction term for such a quantum ﬁeld theory.35) The restriction on the coupling constant.37) and has. Let us also g note here that an interaction Lagrangian density of the form − 3! φ3 . which does not respect the discrete symmetry (6. the same canonical dimension as .34). λ. then. would lead to a potential (and. is there so that the Hamiltonian will be positive deﬁnite. Consequently. therefore. which in turn allows us to deﬁne a vacuum state of the theory. Let us recall that the action for a quantum mechanical particle has the generic form S = = dt (p q + · · · ) ˙ (p dq + · · · ).36) While there are other interaction Lagrangian densities that we can construct consistent with our symmetry requirements. the action is dimensionless. (We will see this shortly. the simplest interaction Lagrangian density involving only the scalar ﬁelds which we can think of has the form λ 4 φ (x). the fully self-interacting theory of a real Klein-Gordon ﬁeld is described by a Lagrangian density L = L0 + LI .

or.39) In other words. the mass term in the Lagrangian density (the second term) automatically leads to a dimensionless action in these units with the canonical dimension of the ﬁeld already determined. [M ]−4 [M ][φ][M ][φ] = [M ]0 . From the canonical dimension of the interaction term in the Lagrangian density in (6. [M ]−4 [λ] [M ]4 = [M ]0 . [φ] = [M ]. In general.35) is dimensionless. let us note that for a monomial interaction action of the form . has a canonical dimension 1 and.6. we conclude that the coupling constant or the interaction strength for the φ4 -self-interaction in (6. we obtain λ 4 φ = [M ]0 . or. 4! [SI ] = − or. or. in these units.40) Thus. (The dimensionality of the ﬁeld variable depends on the number of space-time dimensions which we will see later. d4 x [∂µ ] [φ] [∂ µ ] [φ] = [M ]0 .) The mass parameter. the Klein-Gordon ﬁeld variable in four dimensions has a canonical dimension 1. or. d4 x d4 x [λ] ([φ])4 = [M ]0 . therefore. (6. ([φ])2 = [M ]2 . (6. of course.2 Self-interacting φ4 theory 221 [L] = [T ] = [xµ ] = [M ]−1 . (6.38) With this we can now study the canonical dimension of the free part of the Klein-Gordon action and this leads to 1 m2 2 ∂µ φ∂ µ φ − φ 2 2 [S0 ] = or. or.35). [λ] = [M ]0 . [∂µ ] = [M ]. d4 x = [M ]0 .

we can show that the transition amplitudes (scattering amplitudes) in the quantum theory will become divergent in such a way that meaningful physical results cannot be extracted from such theories. ˜ SI = [g] d4 x [φn ] = [M ]0 . Such theories are known as non-renormalizable theories.41) Therefore. ˙ ∂ φ(x) Π(x) = (6. therefore. [g] = [M ]4−n . (6. the φ4 interaction is the only physically allowed interaction in this case. the coupling constants in the theory cannot have dimensions of inverse mass.32) we obtain the Hamiltonian density for the interacting theory to be ˙ H = Πφ − L = Π2 − = Π2 − 1 ˙2 1 m2 2 λ 4 φ − φ φ − ∇φ · ∇φ − 2 2 2 4! m2 2 λ 4 1 2 1 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! . In such a case.222 6 Self-interacting scalar field theory g ˜ SI = − n! d4 x φn . Since the interaction Lagrangian density does not involve derivatives of ﬁelds. consequently. or. for n > 4.42) so that from (6. [g] [M ]−4 [M ]n = [M ]0 . restricts n ≤ 4 and.32)) ∂L ˙ = φ(x). the canonical momentum conjugate to the ﬁeld variable of the theory continues to be (see (5.) This. consistent with our symmetry requirements (see.34)). (6. or. the coupling constant of the interaction Lagrangian density will have inverse dimensions of mass. for example. (We will see this in a later chapter when we discuss renormalization of quantum ﬁeld theories. If we want to restrict to renormalizable theories which can give rise to meaningful physical predictions.

43) H = = d3 x H d3 x m2 2 λ 4 1 2 1 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! (6. t) carries time dependence. This leads to diﬀerent possible pictures for describing the same quantum mechanical system through distinct time evolutions. as we know. also the . neither the states in the Hilbert space nor the operators acting on state vectors are observables. therefore. There is. This is why we restrict to λ > 0 in the interacting theory. each term in the integrand in (6. the operators carry time dependence while the state vectors do not and the dynamical equations are given by the Heisenberg equations of motion. In this picture. = which leads to (6. For λ < 0. the Hamiltonian will be bounded from below leading to a meaningful vacuum state. the ground state of the free Hamiltonian will not be stable under perturbations. Consequently.3 Interaction picture and time evolution operator 223 1 2 1 m2 2 λ 4 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! = H0 + HI . As a result.6. the observables correspond to expectation values of Hermitian operators in quantum states.3 Interaction picture and time evolution operator In quantum mechanics as well as in quantum ﬁeld theory. the Hamiltonian is indeﬁnite because of the interaction term. of course.44) is positive deﬁnite and. Rather. on the other hand. Our discussion so far has been within the context of the Heisenberg picture where the ﬁeld operator φ(x. The other point that should be emphasized here is that our quantum Hamiltonian should be normal ordered (even though we are not indicating the normal ordering explicitly). the quantum description allows for a unitary change in the states as well as the operators without changing the expectation values which correspond to physical quantities. It is now clear that for λ > 0.44) = H0 + HI . 6.

we have assumed that at t = 0 both the pictures coincide.45) O(S) = e−iHt O(H) (t)eiHt . |ψ(0) (S) = |ψ (H) . for example. (6.45) that (H) ψ|O(H) (t)|ψ (H) = (S) ψ(t)|O(S) |ψ(t) (S) .46) In writing (6. The two pictures are related by the unitary transformation = e−iHt |ψ |ψ(t) (S) (H) . namely. Thus. for example. In the Schr¨dinger picture.224 6 Self-interacting scalar field theory Schr¨dinger picture where the operators are time independent but o the state vectors carry all the time dependence and the time evolution of states is given by the Schr¨dinger equation governed by the o total Hamiltonian of the system. t) eiHt = φ(H) (x.47) O(S) = O(H) (0). (6. (6.49) . time independent.48) The total Hamiltonian operator in the Schr¨dinger picture coino cides with that in the Heisenberg picture (see. the o derivation of the time evolution operator is exactly analogous to the discussion in non-relativistic quantum mechanics and has the form U (S) (t. where H denotes the total Hamiltonian of the system in the Heisenberg picture (we do not put a superscript denoting this to avoid possible confusion) which is time independent. It follows from (6. of course. (6. (6. the Klein-Gordon ﬁeld o operator will be given by φ(S) (x) = e−iHt φ(H) (x. in the Schr¨dinger picture. 0).45). t0 ) = e−iH(t−t0 ) .45)) and is. (6.

(IP ) (IP ) (6. In this case. o Here. the Schr¨dinger picture is not very desirable for relativiso tic theories since the ﬁeld operators in the Schr¨dinger picture are o not manifestly Lorentz covariant. both the operators as well as the state vectors carry time dependence. (IP H0 ) (t) = eiH0 t (S) H0 e−iH0 (S) t (S) = H0 .) Although the Heisenberg picture has a manifestly covariant description. we can determine that (remember that = 1) i ∂|ψ(t) ∂t (IP ) =i (S) ∂ eiH0 t e−iHt |ψ ∂t (H) (S) = −H0 eiH0 (S) t −iHt e |ψ (H) + eiH0 t (S) t He−iHt |ψ t iH0 (H) (IP = −H0 ) |ψ(t) (IP = −H0 ) |ψ(t) (IP = HI ) (t)|ψ(t) (IP ) (IP ) + eiH0 (S) He−iH0 (S) e (S) t −iHt e |ψ (H) + H (IP ) |ψ(t) . (6. in relativistic interacting ﬁeld theories.52) where we have used the fact that the total Hamiltonian H is the same in both the Schr¨dinger as well as the Heisenberg pictures and o that. once again. this picture is useful. . (6. that we can identify |ψ(0) (IP ) = |ψ(0) (S) (S) = |ψ (H) .51) In the interaction picture. we deﬁne |ψ(t) (IP ) = eiH0 (S) t t |ψ(t) (S) = eiH0 (S) (S) t −iHt e |ψ t (H) . by deﬁnition. intermediate between the Schr¨dinger and the Heisenberg pictures. O(IP ) (t) = eiH0 = eiH0 (S) O(S) e−iH0 e t (S) (S) t −iHt O(H) (t)eiHt e−iH0 . (However.6.3 Interaction picture and time evolution operator 225 However. the more convenient description goes under the name of interaction picture which is.50) Note. O(IP ) (0) = O(S) = O(H) (0). in some sense. therefore. it is worth emphasizing here that for certain applications.

In this picture. in this picture. H0 = 1 (IP ) (IP O (t). it is easy to show that the time evolution operator satisﬁes the properties. This shows that once we deﬁne commutation relations for the operators in the free theory. (6. . t0 ) |ψ (t0 ) .51) that the free Hamiltonian in the interaction picture is time independent and.55) can be seen from (6. H0 ) .53) Similarly. the time evolution of the state vectors. i (6.56) then. let us drop the superscript (IP ) with the understanding that we are nonetheless working in the interaction picture. (6.49) and (6.55) where the time evolution operator in the interaction picture in (6. is governed by the Schr¨dinger equation with the interaction o Hamiltonian playing the role of the Hamiltonian.54) In other words.50) to have the explicit form U (t. while the dynamical evolution of the operators is governed by the free Hamiltonian through the Heisenberg equations of motion. We note from (6. (6. the ﬁeld operators can have a plane wave expansion.226 6 Self-interacting scalar field theory (IP HI ) (t) = eiH0 (S) t (S) HI e−iH0 (S) t . we can derive the time evolution for the operators to be ∂O(IP ) ∂t = (S) (S) ∂ eiH0 t O(S) e−iH0 t ∂t (S) (S) = iH0 O(IP ) (t) − iO(IP ) (t)H0 (S) = −i O(IP ) (t). t0 ) = eiH0 (S) t −iH(t−t0 ) −iH0 e e (S) t0 . therefore. if we deﬁne the time evolution operator through the relation |ψ(t) = U (t. From now on. in the interaction picture they continue to hold even in the presence of interactions.

57) Furthermore.) We can now solve (6. (Since H0 is time independent in the interaction picture (see (6. t1 ) U (t1 .53) that the interaction Hamiltonian in the interaction picture is time dependent. t0 ) = HI (t)U (t. t0 ) = U (t2 .51)). t0 ) t0 ½−i ½−i t dt1 HI (t1 ) t0 t t0 ½−i t1 dt2 HI (t2 ) U (t2 . U † (t. (6. t) = ½. t0 ) .3 Interaction picture and time evolution operator 227 U (t. t0 ) t0 t t1 dt1 HI (t1 ) + (−i)2 dt1 t0 t0 dt2 HI (t1 ) HI (t2 ) +··· + ··· +(−i)n +··· .55) satisﬁes the equation (this also follows from (6.50)) ∂|ψ(t) = HI (t)|ψ(t) .57). t0 ) . let us look at the second order term in (6.59) To bring this to a more convenient form.59) and note that . t t1 t0 tn−1 t0 dt1 t0 dt2 · · · dtn HI (t1 ) HI (t2 ) · · · HI (tn ) (6. i ∂t i or. t0 ) = ½ − i = = dt1 HI (t1 ) U (t1 .58) iteratively subject to the initial condition in (6.58) We note from (6. U (t2 . t0 ) = U −1 (t. from (6.52) we see that the time evolution operator in (6. ∂t ∂U (t.6. (6. t) . it follows that the total Hamiltonian is time independent in the interaction picture. to obtain t U (t. t0 ) = U (t0 .56) with the use of (6.

t0 (6. we can show that the n-th term in the series in (6. In a similar manner.59) can be written as .61) with the operator at later time standing to the left of the operator at earlier time. (6.60) where T denotes the time ordering operator and is conventionally deﬁned for two bosonic operators as T (A(t1 )B(t2 )) = θ(t1 − t2 )A(t1 )B(t2 ) + θ(t2 − t1 )B(t2 )A(t1 ).228 6 Self-interacting scalar field theory (−i)2 t t1 dt1 t0 t0 t dt2 HI (t1 ) HI (t2 ) t2 = (−i)2 = (−i)2 2 t dt2 t0 t t0 dt1 HI (t2 ) HI (t1 ) t1 dt1 t0 t2 t0 dt2 HI (t1 ) HI (t2 ) dt1 HI (t2 ) HI (t1 ) + t0 dt2 t0 t = (−i)2 2 t t1 dt1 t0 t2 t0 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + t0 dt2 t0 t dt1 θ (t2 − t1 ) HI (t2 ) HI (t1 ) t = (−i)2 2 t dt1 t0 t t0 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + t0 dt2 t0 t dt1 θ (t2 − t1 ) HI (t2 ) HI (t1 ) = (−i)2 2 t0 dt1 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + θ (t2 − t1 ) HI (t2 ) HI (t1 ) = (−i)2 2! t t0 t dt1 dt2 T (HI (t1 ) HI (t2 )) .

t0 ) = T e t0 dt′ HI (t′ ) .63). we assume that the initial and the ﬁnal states are plane wave states corresponding to free particles.4 S-matrix 229 (−i)n n! t t0 t ··· t0 dt1 · · · dtn T (HI (t1 ) · · · HI (tn )) ..59) takes the form t U (t. This can clearly be implemented by assuming that the interaction switches oﬀ adiabatically at t = ±∞ (adibatic hypothesis). in the interaction picture the time evolution operator is given by the time ordered exponential involving the integral of only the interaction Hamiltonian. in the scattering of particles in relativistic quantum ﬁeld theory.4 S-matrix In the non-relativistic quantum mechanical scattering problems. (6. For example. t0 t0 dt1 · · · dtn HI (t1 ) · · · HI (tn ) U (t.. Similarly. the iterative solution for the time evolution operator in (6.6. t0 ) = ½ − i + (−i)2 T 2! dt1 HI (t1 ) t0 t t0 t dt1 dt2 HI (t1 ) HI (t2 ) t0 t t (−i)n +··· + T n! +··· or. we can implement this by modifying the interaction Hamiltonian as (η) HI (t) → lim HI (t) = lim e−η|t| HI (t). −i Rt . (6.62) Therefore.63) In other words. η→0+ η→0+ (6. we also assume that the incoming particles at t = −∞ as well as the outgoing particles at t = ∞ are described by free particle states.64) . (This is a formal deﬁnition which is to be understood in the sense of the expansion described in (6.) 6.

the state into which this will evolve at t = ∞ is deﬁned from (6. the S-matrix (or the scattering matrix) of the theory can be identiﬁed with the time evolution operator (6. −∞). Let us denote the initial state at inﬁnite past as the free particle state |ψi (−∞) = |i . (6. (6.67) which has .55) to be |ψi (∞) = U (∞. which is the deﬁnition of the scattering amplitude.68) Consequently. the interaction vanishes. −∞)|i = f |S|i .66) to be Sf i = f |ψi (∞) = f |U (∞.64) can be thought of as the relevant boundary condition for the scattering problem under study. We are essentially assuming here that any smooth function describing the adiabatic switching oﬀ of the interaction (and not just the speciﬁc form in (6.66) where we have identiﬁed S = U (∞. we can take the initial and the ﬁnal states to be the eigenstates of the free Hamiltonian which we know to be free particle energy-momentum states. the probability amplitude for an initial state |i to be in the ﬁnal free particle state |f at t = ∞. As a result. (6.65) Then.64)) leads to the same result for the rate of transition in the physical scattering of particles. −∞)|ψi (−∞) = U (∞. −∞)|i = S|i . The condition (6. (6. is obtained from (6.67) Therefore.230 6 Self-interacting scalar field theory so that in the inﬁnite past as well as in the inﬁnite future.

this state would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (in) = U (0. in general. −∞)|Ψ(−∞) (in) .63). not well deﬁned if we do not use an adiabatic interaction of the form. the linear contribution to the right-hand side of (6. Schr¨dinger and interaction) coincide at t = 0 (see (6. in (6. the S-matrix (the scattering matrix) is unitary since the time evolution operator is (see (6.69) through the appropriate boundary conditions. −∞) is.70) will have the form .4 S-matrix 231 the explicit perturbative expansion of the form (see (6. These states can be constructed by noting that the three pictures (Heisenberg.64) provides a regularized meaning to the oscillatory terms in (6.51)) and o since the Heisenberg states are time independent. Let us see this explicitly in the case of the linear term in the expansion of the time evolution operator in (6. In this case.6. Then. These are asymptotic free states as t → −∞ and t → ∞ respectively. −∞) = lim T e η→0+ −∞ dt HI (t) (η) = ½ − lim+ i η→0 ∞ (η) dt HI (t) −∞ ∞ −∞ (η) (η) dtdt′ T HI (t)HI t′ (−i)2 + lim 2! η→0+ +··· . It is clear from the expansion above that the adiabatic switching of the interaction in (6. they are uniquely given by the states (in any picture) at t = 0.70) where the time evolution operator U (0. Let us suppose that at t → −∞ we have a free incoming state denoted as |Ψ(−∞) (in) = |i .64). (6.57)).69) Furthermore. (6. This also leads to the notion of “in” and “out” states which are quite important in a formal description of scattering theory.63)) −i ∞ R S = U (∞. say.

72) we see that they are related to each other as . from their deﬁnitions in (6. We also note that the integral in (6. they are not identical because their deﬁnitions use diﬀerent regularizing factors.232 6 Self-interacting scalar field theory 0 η→0+ lim −i = = = (η) dt HI (t)|Ψ(−∞) (in) −∞ 0 η→0+ lim −i lim −i lim dt eηt eiH0 (S) t (S) HI e−iH0 (S) t −∞ 0 |i η→0+ dt e−i(Ei −H0 (S) +iη)t −∞ (S) HI |i η→0+ 1 (S) HI |Ψ(−∞) (S) Ei − H0 + iη (in) . (6.70) and (6.72) and an analysis as in (6.64) naturally provides a regularization for the formal deﬁnition of the time evolution operator U (0. the “in” and the “out” states deﬁne a complete space of states at t → −∞ and t → ∞ respectively. This clariﬁes how the boundary condition (6.72) we see that even though both the “in” and the “out” states are related to the same Heisenberg state. −∞). (6. the regularizing factor would have to be e−ηt without which U (0. Rather.71) is not deﬁned at the lower limit in the absence of the adiabatic factor eηt . we have used the fact that H0 ) = (S) H0 and that |Ψ(−∞) (in) = |i is a free state with Ei the energy (IP eigenvalue of H0 ) . In a parallel manner. Furthermore.71) would show that in this case. ∞)|Ψ(∞) (out) . ∞) would not be well deﬁned at the upper limit.71) (IP where in the intermediate steps.70) and (6. This shows that the asymptotically free “in” and “out” states can be deﬁned in a unique manner by relating them to the Heisenberg state. From (6. if we assume that as t → ∞. then this would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (out) = U (0. we have a free outgoing state |Ψ(∞) (out) .

simply corresponds to arranging factors in the product so that the creation operators stand to the left of the annihilation operators. this is equivalent to saying that the negative energy (frequency) parts of the ﬁeld operator stand to the left of the positive energy (frequency) parts (recall that the positive energy part of the ﬁeld operator. Correspondingly. we will derive a simple relation for the product of factors each of which is normal ordered and then using this relation.73) where we have used the properties of the time evolution operator given in (6.69) corresponds to no scattering at all. The ﬁrst term in the expansion of the S-matrix in (6. −∞)|Ψ(−∞) = U (∞. (6. ∞)|Ψ (H) = U (∞.67) (and. contains the annihilation operator while . This shows that the “in” and the “out” states are related by the S-matrix in (6. 0)|Ψ (in) (H) = U (∞. 0)U (0. (6. we can deﬁne S − ½ = T. we will simplify the time ordered product of factors of normal ordered terms. Furthermore. −∞)|Ψ(−∞) (in) = S|Ψ(−∞) (in) . The calculation can be carried out in two steps. Sf i . as we have deﬁned earlier in (5.5 Normal ordered product and Wick’s theorem 233 |Ψ(∞) (out) = U † (0.57). Later we will see that the Feynman rules are a wonderful and simple way of systematizing these results. such a calculation will involve time ordered products of normal ordered products and this is where the Wick’s theorem comes in handy. 6. First. In terms of ﬁeld operators. It is clear from this discussion that in calculating the S-matrix elements. φ(+) .74) where the T-matrix represents the true nontrivial eﬀects of scattering.6.76).5 Normal ordered product and Wick’s theorem The normal ordered product. we need to evaluate the matrix elements of time ordered products of operators between free particle states. therefore. by a phase since the S-matrix is unitary). since our (interaction) Hamiltonian is assumed to be normal ordered.

75).76) Since the vacuum state is. N φ(+) (x)φ(+) (y) ≡ : φ(+) (x)φ(+) (y) : = φ(+) (x)φ(+) (y). annihilated by the annihilation operator. For example.76)). for the bosonic scalar ﬁeld. using the relations in (6.234 6 Self-interacting scalar field theory the negative energy part. Thus. N φ(+) (x)φ(−) (y) ≡ : φ(+) (x)φ(−) (y) : = φ(−) (y)φ(+) (x). (6. we note from (6. the order of the factors inside a normal ordered product is not important.76) that . we can equivalently say that φ(+) (x)|0 =0= 0|φ(−) (x).75) We note that. N φ(−) (x)φ(−) (y) ≡ : φ(−) (x)φ(−) (y) : = φ(−) (x)φ(−) (y). φ(−) . (6. that the vacuum expectation value of any normal ordered product vanishes. Furthermore. therefore. as far as bosonic operators are concerned. (6.77) It is clear. N φ(−) (x)φ(+) (y) ≡ : φ(−) (x)φ(+) (y) : = φ(−) (x)φ(+) (y). by deﬁnition. has the creation operator so that this prescription is equivalent to the deﬁnition of normal ordering in (5. we have N (φ(x)) ≡ : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). we can write N (φ(x)φ(y)) ≡ : φ(x)φ(y) : = = : φ(+) (x) + φ(−) (x) φ(+) (y) + φ(−) (y) : : φ(+) (x)φ(+) (y) + φ(+) (x)φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) : = φ(+) (x)φ(+) (y) + φ(−) (y)φ(+) (x) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y).

in the simple example of the product of two ﬁeld operators. Let us denote (6.158) and (6. normal ordering is quite useful since we would like the vacuum expectation value of any observable to vanish simply because the vacuum contains no particles – it is empty and. (6. (6.81) .79) where we have used (5. let us note that. This gives another reason to choose the prescription of normal ordering for physical observables.76).5 Normal ordered product and Wick’s theorem 235 0|N (φ(x)φ(y)) |0 = 0|φ(+) (x)φ(+) (y) + φ(−) (y)φ(+) (x) + φ(−) (x)φ(+) (x) +φ(−) (x)φ(−) (y)|0 = 0. Furthermore. therefore. (6. (6.6.80) with the pairing of the two ﬁeld operators deﬁned to be related to the invariant positive energy Green’s function in (5. To see how we can write a product of operators in normal ordered form. φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) = : φ(x)φ(y) : −iG(+) (x − y). the calculation of S-matrix elements will simplify enormously. we have φ(x)φ(y) = φ(+) (x) + φ(−) (x) φ(+) (y) + φ(−) (y) = φ(+) (x)φ(+) (y) + φ(+) (x)φ(−) (y) + φ(−) (x)φ(+) (y) +φ(−) (x)φ(−) (y) = φ(+) (x)φ(+) (y) + φ(−) (y)φ(+) (x) + φ(+) (x). does not carry any nontrivial quantum number.79) as φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y).151) as φ(x)φ(y) = −iG(+) (x − y). this also shows that if we can rewrite any product of operators (including normal ordered factors) in terms of normal ordered terms.78) Therefore.

This brings out yet another connection between Green’s functions and the quantum ﬁeld theory. Therefore. we note from (6.80) is zero.85) Thus adding the two terms in (6. We know that by deﬁnition : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). (6. Since we choose the Hamiltonian to be normal ordered. but since it is a c-number function. (6.236 6 Self-interacting scalar field theory We note that G(+) (x − y) is not a symmetric function (so that the pairing should be read carefully as from left to right). : φ(x) : : φ(y) : = : φ(x) : φ(+) (y) + : φ(x) : φ(−) (y). φ(−) (y) = = : φ(x)φ(−) (y) : + φ(+) (x).84) simply corresponds to a normal ordered product : φ(x) : φ(+) (y) = φ(x)φ(+) (y) = : φ(x)φ(+) (y) : .84) gives : φ(x) : φ(−) (y) = φ(x)φ(−) (y) = φ(−) (y)φ(x) + φ(x).82) because the vacuum expectation value of the normal ordered product in (6. φ(−) (y) : φ(x)φ(−) (y) : − iG(+) (x − y). (6. the second term in (6. let us next see how we can combine a product of normal ordered factors into normal ordered terms.86) (6. Clearly.84) (6.83) We recognize that the ﬁrst term in (6. On the other hand.86) we obtain . this will be necessary in developing a perturbation expansion for the quantum ﬁeld theory where the interaction Hamiltonian is normal ordered.85) and (6.80) that we can also identify φ(x)φ(y) = 0|φ(x)φ(y)|0 = −iG(+) (x − y).

let us note that . Next. (6. (6. what we had seen earlier in (6.80). φ(−) (z) φ(y) +φ(x) φ(+) (y). (6. φ(−) (z) = : φ(x)φ(y)φ(z) : + φ(+) (x).89) Finally. of course. φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + : φ(x)φ(y)φ(−) (z) : + φ(x)φ(y) + iG(+) (x − y) .88) φ(x)φ(y) = = but (6.5 Normal ordered product and Wick’s theorem 237 : φ(x) : : φ(y) : = : φ(x) : φ(+) (y)+ : φ(x) : φ(−) (y) = : φ(x)φ(+) (y) : + : φ(x)φ(−) (y) : − iG(+) (x − y) = : φ(x)φ(y) : + φ(x)φ(y).87) This is. φ(−) (z) = : φ(x)φ(y)φ(z) : − iG(+) (x − z)φ(y) − iG(+) (y − z)φ(x) = : φ(x)φ(y)φ(z) : + φ(x)φ(z) : φ(y) : + φ(y)φ(z) : φ(x) : . : φ(x)φ(y) : −iG(+) (x − y) : φ(x)φ(y) : + φ(x)φ(y).6.87) shows how a product of two simple normal ordered factors can be expressed as a normal ordered term and a pairing.79) and (6. let us look at a product of the form : φ(x)φ(y) : : φ(z) : = : φ(x)φ(y) : φ(+) (z) + : φ(x)φ(y) : φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + φ(−) (z) : φ(x)φ(y) : + : φ(x)φ(y) : . namely.

φ(−) (w)] = : φ(x)φ(y)φ(z)φ(w) : −iG(+) (x − z)φ(y)φ(w) −iG(+) (y − w)φ(x)φ(+) (z) − iG(+) (x − w)φ(−) (z)φ(y) −iG(+) (y − w)φ(−) (z)φ(x) = : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(y)φ(w) + φ(y)φ(z) : φ(x)φ(w) : + φ(x)φ(w) + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(w) : φ(x)φ(z) : = : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(z) : φ(x)φ(w) : + φ(y)φ(w) : φ(x)φ(z) : + φ(x)φ(z) φ(y)φ(w) + φ(x)φ(w) φ(y)φ(z). (6. φ(−) (z)φ(−) (w)] = : φ(x)φ(y)φ(+) (z)φ(+) (w) : + : φ(x)φ(y)φ(−) (z)φ(+) (w) : + : φ(x)φ(y)φ(+) (z)φ(−) (w) : + : φ(x)φ(y)φ(−) (z)φ(−) (w) : +[: φ(x)φ(y) :.90) −iG(+) (y − z)φ(x)φ(w) − iG(+) (x − w)φ(y)φ(+) (z) These simple examples demonstrate how a product of normal ordered .238 6 Self-interacting scalar field theory : φ(x)φ(y) : : φ(z)φ(w) : = : φ(x)φ(y) : φ(+) (z)φ(+) (w)+ : φ(x)φ(y) : φ(−) (z)φ(+) (w) + : φ(x)φ(y) : φ(−) (w)φ(+) (z)+ : φ(x)φ(y) : φ(−) (z)φ(−) (w) = : φ(x)φ(y)φ(+) (z)φ(+) (w) : +φ(−) (z) : φ(x)φ(y) : φ(+) (w) +[: φ(x)φ(y) :. φ(−) (z)]φ(w) + [: φ(x)φ(y) :. φ(−) (w)]φ(+) (z) + φ(−) (z)φ(−) (w) : φ(x)φ(y) : +[: φ(x)φ(y) :. φ(−) (w)]φ(+) (z) +φ(−) (z)[: φ(x)φ(y) :. φ(−) (z)]φ(+) (w) + φ(−) (w) : φ(x)φ(y) : φ(+) (z) +[: φ(x)φ(y) :.

φ(x)φ(y)φ(z)φ(w) = = : φ(x)φ(y) : + φ(x)φ(y) : φ(z)φ(w) : + φ(z)φ(w) : φ(x)φ(y) :: φ(z)φ(w) : + φ(x)φ(y) : φ(z)φ(w) : + φ(z)φ(w) : φ(x)φ(y) : + φ(x)φ(y) φ(z)φ(w) = : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(z) : φ(x)φ(w) : . that (6. For example. This now leads us to an important theorem known as the Wick’s theorem (this is one form of it) which says that a product of factors of normal ordered ﬁeld operators can be written as a sum of terms with normal ordered terms and all possible pairings between ﬁeld operators except for pairings between operators within the same normal ordered factor. we already know from (6. (6.92) Similarly. Let us note that any product of operators (and not just a product of normal ordered factors) can be written in terms of normal ordered products.6.91) φ(x)φ(y)φ(z) = : φ(x)φ(y) : φ(z) + φ(x)φ(y)φ(z) = : φ(x)φ(y) : : φ(z) : + φ(x)φ(y) : φ(z) : = : φ(x)φ(y)φ(z) : + φ(x)φ(z) : φ(y) : + φ(y)φ(z) : φ(x) : + φ(x)φ(y) : φ(z) : . therefore.80) that φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y). It follows.5 Normal ordered product and Wick’s theorem 239 factors can be written in terms of normal ordered products with all possible pairings between ﬁelds in diﬀerent factors (but no pairing between two ﬁelds within the same factor).

This construction can also be extended to the case where only some of the factors are normal ordered. the product of any number of ﬁeld operators can be expressed as a sum of terms involving normal ordered products with all possible pairings (pairings are from left to right since the positive energy Green’s function G(+) (x−y) is not symmetric). (6.89) that . In general. we have already noted in (6. This has to be contrasted with the products of normal ordered terms which also had a similar expansion except that there were no pairings between terms within the same factor. For example. (6. Wick’s theorem for a product of ﬁeld operators says that φ (x1 ) φ (x2 ) · · · φ (xn ) =: φ (x1 ) · · · φ (xn ) : + φ (x1 ) φ (x2 ) : φ (x3 ) · · · φ (xn ) : + φ (x1 ) φ (x3 ) : φ (x2 ) φ (x4 ) · · · φ (xn ) : + · · · + φ (xn−1 ) φ (xn ) : φ (x1 ) · · · φ (xn−2 ) : + ··· + φ (x1 ) φ (x2 ) φ (x3 ) φ (x4 ) φ (x5 ) φ (x6 ) : φ (x7 ) · · · φ (xn ) : + ··· .93) In other words.240 6 Self-interacting scalar field theory + φ(y)φ(w) : φ(x)φ(z) : + φ(x)φ(y) : φ(z)φ(w) : + φ(z)φ(w) : φ(x)φ(y) : + φ(x)φ(y) φ(z)φ(w) + φ(x)φ(z) φ(y)φ(w) + φ(x)φ(w) φ(y)φ(z). (We are interested in vacuum expectation values because any matrix element can be written as a vacuum expectation value.94) This is quite signiﬁcant since it says that the vacuum expectation value of the product of any number of factors involving ﬁeld operators (whether the factors are normal ordered or not) is given by the terms where all the ﬁelds are completely paired.) An immediate consequence of this result is that the vacuum expectation value of an odd number of ﬁeld operators must vanish (since not all ﬁeld operators can be paired).

6.6 Time ordered products and Wick’s theorem

241

: φ(x)φ(y) : φ(z) =: φ(x)φ(y) :: φ(z) : = : φ(x)φ(y)φ(z) : + φ(x)φ(z) : φ(y) : + φ(y)φ(z) : φ(x) : . (6.95)

We can now extend this result as (using (6.79) as well as (6.90))

: φ(x)φ(y) : φ(z)φ(w) =: φ(x)φ(y) :

: φ(z)φ(w) : + φ(z)φ(w)

= : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(z) : φ(x)φ(w) + φ(y)φ(w) : φ(x)φ(z) : + φ(z)φ(w) : φ(x)φ(y) : + φ(x)φ(z) φ(y)φ(w) + φ(x)φ(w) φ(y)φ(z), and so on. 6.6 Time ordered products and Wick’s theorem There is a second kind of operator product, known as the time ordered product, which plays a fundamental role in quantum ﬁeld theories. We have already come across time ordering in the deﬁnition of the time evolution operator in (6.63) as well as in the deﬁnition of the S-matrix. The time ordered product of two bosonic operators is deﬁned to be (see (6.61)) φ(x)φ(y), x0 > y 0 , φ(y)φ(x), y 0 > x0 . (6.96)

T (φ(x)φ(y)) = T (φ(y)φ(x)) =

(6.97)

Note that the order of the factors, inside time ordering, is not important for bosons. More compactly, we can write the time ordered product of two ﬁeld operators in (6.97) as

T (φ(x)φ(y)) = θ x0 − y 0 φ(x)φ(y) + θ y 0 − x0 φ(y)φ(x),

(6.98)

242

6 Self-interacting scalar field theory

where θ(x) represents the step function, deﬁned to be

θ(x) =

1, for x > 0, 0, for x < 0.

(6.99)

We note here that, if the two time arguments are equal, there is no ambiguity in the deﬁnition of the time ordered product in (6.97) or (6.98) since the ﬁeld operators commute at equal times. However, for operators which do not commute at equal times, the time ordered product has to be deﬁned more carefully. Fortunately, in most of our discussions, this would not be a problem. We note, in this simple example, that the equal time limit can be 1 approached by assuming that θ(0) = 2 . However, in more complicated situations, the equal time limit has to be taken in a consistent limiting manner. Thus, for example, we may choose the equal time limit as x0 → y 0 + 0+ or y 0 → x0 + 0+ with the appropriate limiting behavior of the step function. But, whatever limiting procedure we choose must be consistent through out. Let us note here that an integral representation for the step function which has the limiting behavior built in is given by

θ(x) = lim −

→0+

dk e−ikx . 2πi k + i

(6.100)

Of course, by deﬁnition

T (φ(x)) = φ(x) =: φ(x) : .

(6.101)

If we have a product of three ﬁelds or more, we can generalize the definition of time ordered product in a straightforward manner. Thus, for the product of three ﬁeld operators, we have

6.6 Time ordered products and Wick’s theorem

243

T (φ(x)φ(y)φ(z)) = θ x0 − y 0 θ y 0 − z 0 φ(x)φ(y)φ(z)

+θ x0 − z 0 θ z 0 − y 0 φ(x)φ(z)φ(y)

+θ y 0 − z 0 θ z 0 − x0 φ(y)φ(z)φ(x)

+θ y 0 − x0 θ x0 − z 0 φ(y)φ(x)φ(z)

+θ z 0 − x0 θ x0 − y 0 φ(z)φ(x)φ(y)

+θ z 0 − y 0 θ y 0 − x0 φ(z)φ(y)φ(x),

(6.102)

and so on. Given the deﬁnition of time ordered products, we can now relate them to normal ordered products as follows.

T (φ(x)) = φ(x) = : φ(x) :, T (φ(x)φ(y)) = θ x0 − y 0 φ(x)φ(y) + θ y 0 − x0 φ(y)φ(x) = θ x0 − y 0 : φ(x)φ(y) : −i G(+) (x − y) : φ(y)φ(x) : − iG(+) (y − x) . (6.103) +θ y 0 − x0

Remembering that the order of the terms inside normal ordering does not matter for bosons, we obtain from (6.103)

T (φ(x)φ(y)) = θ x0 − y 0 + θ y 0 − x0 : φ(x)φ(y) :

−iθ x0 − y 0 G(+) (x − y) + iθ y 0 − x0 G(−) (x − y) = : φ(x)φ(y) : + i − θ x0 − y 0 G(+) (x − y) +θ y 0 − x0 G(−) (x − y) = : φ(x)φ(y) : + iGF (x − y), (6.104)

244

6 Self-interacting scalar field theory

where we have used (5.147) as well as the deﬁnition of the Feynman Green’s function in (5.146) in the intermediate step. Let us write (6.104) as

T (φ(x)φ(y)) =: φ(x)φ(y) : + φ(x)φ(y),

(6.105)

where the contraction of two ﬁelds is deﬁned as the c-number Feynman Green’s function, namely,

φ(x)φ(y) = iGF (x − y).

(6.106)

Note that, since the Feynman Green’s function is symmetric (it is an even function, see (5.148)), the order of the contraction is not important unlike the pairing of ﬁelds deﬁned in (6.81). Furthermore, because the vacuum expectation value of a normal ordered product vanishes, we immediately identify

0|T (φ(x)φ(y)) |0 = φ(x)φ(y) = iGF (x − y).

(6.107)

Namely, the Feynman Green’s function can be identiﬁed with the vacuum expectation value of the time ordered product of two ﬁeld operators which brings out yet another connection between the Green’s functions and the quantum ﬁeld theory. As we have seen earlier, the Feynman Green’s function, unlike the Schwinger function, satisﬁes

∂xµ ∂ xµ + m2 GF (x − y) = −δ4 (x − y).

(6.108)

Since we know how to express the product of any number of ﬁeld operators in terms of normal ordered products and pairings, we can carry through this construction to a time ordered product of any number of ﬁeld operators as well. Let us simply note the results here.

6.6 Time ordered products and Wick’s theorem

245

T (φ(x)φ(y)φ(z)) = : φ(x)φ(y)φ(z) : + φ(x)φ(y) : φ(z) : + φ(y)φ(z) : φ(x) : + φ(x)φ(z) : φ(y) :, T (φ(x)φ(y)φ(z)φ(w)) = : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(y) : φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(z) : φ(x)φ(w) : + φ(y)φ(w) : φ(x)φ(z) : + φ(z)φ(w) : φ(x)φ(y) : + φ(x)φ(y) φ(z)φ(w) + φ(x)φ(z) φ(y)φ(w) + φ(x)φ(w) φ(y)φ(z). Here, we have to remember that (6.109)

φ(x)φ(y) = iGF (x − y) = φ(y)φ(x),

(6.110)

namely, this is an even function and the two contractions are, therefore, not distinct. This again shows that the time ordered product of any number of ﬁelds can be written as a sum of normal ordered terms with all possible distinct contractions. Namely,

T (φ (x1 ) φ (x2 ) · · · φ (xn )) =: φ (x1 ) φ (x2 ) · · · φ (xn ) : + φ (x1 ) φ (x2 ) : φ (x3 ) · · · φ (xn ) : + ··· + φ (x1 ) φ (x2 ) φ (x3 ) φ (x4 ) : φ (x5 ) · · · φ (xn ) : + ··· . (6.111)

This is another form of Wick’s theorem and without going into derivations, we note here that a time ordered product of normal

246

6 Self-interacting scalar field theory

ordered products can be written as a sum of normal ordered terms with all possible contractions except for the contractions between ﬁeld operators within a given normal ordered factor. It follows now that since the vacuum expectation value of a normal ordered product vanishes, the vacuum expectation value of a time ordered product of ﬁeld operators (where the factors may or may not be normal ordered) is given only by the terms where all the ﬁeld operators have been pairwise contracted – namely, such a vacuum expectation value, if it is nonzero, will involve products of Feynman Green’s functions. Let us note here that the vacuum expectation value of a product of operators is a fundamental quantity in a quantum ﬁeld theory since any matrix element can be written as a vacuum expectation value. As a result, the scattering matrix elements (see (6.68)) can be written as vacuum expectation values of time ordered products of ﬁelds (coming from normal ordered interaction terms if the Hamiltonian is normal ordered as well as ﬁeld operators coming from the initial and the ﬁnal states if they do not correspond to vacuum states). Therefore, this clariﬁes why it is the Feynman Green’s function that is so important in the calculation of scattering matrix elements in a relativistic quantum ﬁeld theory.

6.7 Spectral representation and dispersion relation It is very rarely that we can solve an interacting ﬁeld theory exactly. When we cannot solve a theory, it is useful to derive as much information as we can about the theory from its invariance and symmetry properties. In this section, we will discuss brieﬂy two such interesting topics that often play an important role in studies of ﬁeld theory. We have already seen that vacuum expectation values of ﬁeld operators are quite important and, in fact, in the case of free ﬁelds we have already seen how the vacuum expectation values of two free ﬁeld operators can be related to various Green’s functions. For example, let us note from (5.160) and (5.153) that, for free ﬁelds we can write the Schwinger function as (the subsequent arguments hold for any other vacuum expectation value of two ﬁelds)

6.7 Spectral representation and dispersion relation

247

0| [φ(x), φ(y)] |0

= −iG(x − y) =

d4 k ǫ(k0 )δ(k2 − m2 ) e−ik·(x−y) . (6.112) (2π)3

Let us now consider two arbitrary operators A(x) and B(x) that do not necessarily obey the free ﬁeld equations (these could, for example, represent the ﬁeld operators in the fully interacting theory) and deﬁne 0| [A(x), B(y)] |0 = −iGAB (x − y), (6.113)

and we would like to derive as much information as is possible about this function from the known facts about the quantum ﬁeld theory. Let us recall that the vacuum and the one particle states of the theory (see (5.90), (5.101) and (5.102)) satisfy Pµ |0 = 0, Pµ |k = kµ |k , k0 = Ek = k2 + m2 > 0. (6.114) The true physical vacuum |0 is a Lorentz invariant state. Furthermore, we note that since Pµ generates translations of coordinates, we can write A(x) = eiP ·x A(0)e−iP ·x , B(y) = eiP ·y B(0)e−iP ·y . (6.115)

**As a result, inserting a complete set of energy-momentum basis states into (6.113) we obtain GAB (x − y) = i 0| [A(x), B(y)] |0 = i
**

n

[ 0|A(x)|n n|B(y)|0 − 0|B(y)|n n|A(x)|0 ] e−ikn ·(x−y) 0|A(0)|n n|B(0)|0

= i

n

−eikn ·(x−y) 0|B(0)|n n|A(0)|0

,

(6.116)

248

6 Self-interacting scalar field theory

where we have used (6.115) as well as the properties of the state vectors (6.114), in particular, Pµ |n = knµ |n . (6.117)

Note that since the operators A and B are not free ﬁeld operators, the complete intermediate states |n need not involve only single particle states. In particular, if we identify A(x) = B(x), and recall that the energy eigenstates are deﬁned for positive energy (kn0 > 0), then (6.116) leads to

GAA (x − y) = i 0| [A(x), A(y)] |0 = i

n

| 0|A(0)|n |2 θ(kn0 ) e−ikn ·(x−y) − eikn ·(x−y) d4 q (2π)3 θ(q0 )δ4 (q − kn )| 0|A(0)|n |2 × e−iq·(x−y) − eiq·(x−y)

=

i (2π)3

n

= where

i (2π)3

d4 q ρ(q) θ(q0 ) e−iq·(x−y) − eiq·(x−y) ,

(6.118)

ρ(q) = (2π)3

n

δ4 (q − kn )| 0|A(0)|0 |2 ,

(6.119)

is known as the spectral function associated with the vacuum expectation value of the commutator and measures all the contribution coming from the matrix elements for a given q. Because of the Lorentz invariance of the theory, it is straightforward to show that the spectral function is really a function of q 2 , namely, ρ(q) = ρ(q 2 ), so that we can write from (6.118) (6.120)

6.7 Spectral representation and dispersion relation

249

GAA (x − y) = = =

i (2π)3

d4 q ρ(q 2 )θ(q0 ) e−iq·(x−y) − eiq·(x−y) i ǫ(q0 ) e−iq·(x−y) (2π)3 d4 q ǫ(q0 )δ(q 2 − σ 2 ) e−iq·(x−y) (2π)3 (6.121)

d4 q ρ(q 2 )

∞ 0

dσ 2 ρ(σ 2 ) i

=

0

∞

dσ 2 ρ(σ 2 ) G(x − y, σ 2 ),

where we have used (5.153) (or (6.112)) in the last step. Relation (6.121) is known as the spectral representation of the vacuum expectation value of the commutator and expresses it in terms of the vacuum expectation value of the free ﬁeld commutator in (6.112). Here σ 2 can be thought of as the spectral parameter (eigenvalue of P 2 ) of the free Klein-Gordon equation and (6.121) represents the vacuum expectation value in terms of contributions coming from diﬀerent mass values (which is the reason for the name spectral representation or spectral decomposition). The second topic that we will discuss in this section goes under the name of dispersion relations and can be proﬁtably used in the study of scattering theory. This can be very simply understood as follows. Let us suppose that we have a retarded function of time which we denote by f (t) (see, for example, the retarded Green’s function in (5.133)). Since the function is a retarded function (namely, it can be written as f (t) = θ(t)g(t) and since θ(t)θ(t) = θ(t)), we can write f (t) = θ(t)f (t). (6.122)

We are suppressing here any dependence of the function f (t) on other arguments that are not relevant to our discussion. Let us assume that the Fourier transform f (k0 ) of the function exists and is given by

∞ −∞

f (t) =

dk0 −ik0 t e f (k0 ). 2π

(6.123)

Using the integral representation of the step function

250

6 Self-interacting scalar field theory

θ(t) = lim −

ǫ→0+

1 2πi

∞ −∞

dk0

e−ik0 t , k0 + iǫ

(6.124)

**from (6.122) we obtain 1 2πi
**

∞ −∞

f (t) =

ǫ→0+

lim

−

dk0

e−ik0 t k0 + iǫ

′

∞ −∞

′ dk0 −ik0 t ′ ′ e f (k0 ) 2π

∞

=

ǫ→0+

lim −

−∞ ∞

′ dk0 dk0 e−i(k0 +k0 )t ′ f (k0 ) (2π)2 i k0 + iǫ ′ e−ik0 t dk0 dk0 ′ f (k0 ) ′ (2π)2 i k0 − k0 + iǫ

=

ǫ→0+

lim − lim

−∞ ∞

=

ǫ→0+

−∞

dk0 −ik0 t e 2π

∞ −∞

′ ′ f (k0 ) dk0 ′ − k − iǫ . 2πi k0 0

(6.125)

Comparing (6.125)with (6.123), we conclude that the Fourier transform of a retarded function must satisfy a relation 1 2πi

∞ −∞ ′ dk0 ′ f (k0 ) , − k0 − iǫ

f (k0 ) = lim

ǫ→0+

′ k0

(6.126)

which is the basic dispersion relation for the Fourier transform of a the retarded function. (We note here that in the conventional discussion of dispersion relations, the conjugate variable k0 is identiﬁed with ω.) Decomposing the denominator in (6.126) in the standard manner lim 1 1 ′ = ′ + iπδ(k0 − k0 ), ′ k0 − k0 − iǫ k0 − k0 (6.127)

ǫ→0+

where the ﬁrst term on the right-hand side represents the principal part, the dispersion relation (6.126) can also be written as 1 πi

∞ −∞ ′ dk0 ′ f (k0 ) , ′ k0 − k0

f (k0 ) =

(6.128)

6.7 Spectral representation and dispersion relation

251

where the principal value for the pole is understood. Comparing the real and the imaginary parts in (6.128) we obtain 1 π 1 π

∞ −∞ ′ Im f (k0 ) , ′ k0 − k0

Re f (k0 ) =

′ dk0

Im f (k0 ) = −

∞

−∞

′ dk0

′ Re f (k0 ) . ′ k0 − k0

(6.129)

This is quite interesting because it says that the real part of the Fourier transform of the retarded function can be obtained from a knowledge of its imaginary part in the entire complex k0 plane and vice versa. This form of the dispersion relation, where the real part of an amplitude is related to the imaginary part (or the other way around), is known as the Kramers-Kronig relation. Let us recall from the optical theorem in scattering in quantum mechanics that the imaginary part of the forward scattering amplitude is related to the total scattering cross section. Therefore, the ﬁrst of the relations in (6.129) can also be seen to imply that the real part of a forward scattering amplitude at a ﬁxed energy (frequency) can be given as the integral over all energies of the total cross section. In optics, the ﬁrst of the relations in (6.129) can be used to determine the real part of the refractive index of a medium in terms of its imaginary parts (which is how Kramers and Kronig had used such a dispersion relation). Using (6.129) we can also write

f (k0 ) = Re f (k0 ) + iIm f (k0 ) = = 1 π 1 π

∞ −∞ ∞ ′ dk0 ′ dk0 ′ Im f (k0 ) ′ ′ + iπδ(k0 − k0 )Im f (k0 ) ′ k0 − k0

−∞

′ Im f (k0 ) , ′ k0 − k0 − iǫ

(6.130)

where ǫ → 0+ is assumed. This relation says that we can determine the complete function f (k0 ) (which is the Fourier transform of a retarded function f (t)) from a knowledge of its imaginary part in the entire complex k0 plane. There is a similar relation where the function can be written in terms of its real part alone. Without going

252

6 Self-interacting scalar field theory

into details we note here that dispersion relations inherently imply causality in a theory. Our derivation for the dispersion relations has been quite formal and it is clear that (6.130) will hold (and is meaningful) only if lim Im f (k0 ) → 0. (6.131)

|k0 |→∞

This is quite interesting for it says that, as long as only the imaginary part (or the real part in an alternate formulation) of the Fourier transform of a retarded function vanishes for large values of the argument, the dispersion relation (6.130) would hold. This can be contrasted with the vanishing of an analytic function for large values of its argument (which is what we use in complex analysis) which would imply that both the real and the imaginary parts of the function have to vanish in this regime. If Im f (k0 ) does not fall oﬀ fast enough, then we can deﬁne a subtracted dispersion relation that will ¯ be true and this is done as follows. Let k0 denote a ﬁxed point in the complex k0 plane and let us deﬁne a new function ¯ f (k0 ) − f (k0 ) , (6.132) ¯ k0 − k0 Clearly, the imaginary part of this function will vanish for large values of its argument provided Im f (k0 ) does not grow for large values of its argument. Therefore, for this function we can write a dispersion relation of the form (6.130) F (k0 ) = F (k0 ) = = = = lim 1 π 1 π 1 π

∞ −∞ ∞ ′ dk0 ′ Im F (k0 ) ′ k0 − k0 − iǫ

ǫ→0+

ǫ→0+

lim

−∞ ∞ −∞

′ dk0

′ ¯ Im (f (k0 ) − f (k0 )) ′ ′ ¯ (k0 − k0 )(k0 − k0 − iǫ)

ǫ→0+

lim

′ dk0

′ ¯ Im (f (k0 ) − f (k0 )) ′ ′ ¯ (k0 − k0 − iǫ)(k0 − k0 − iǫ) ∞ −∞ ′ ′ ¯ dk0 Im (f (k0 ) − f (k0 ))

ǫ→0+

lim

1 ¯ π(k0 − k0 )

′ k0

×

1 1 . − ′ ¯ − k0 − iǫ k0 − k0 − iǫ

(6.133)

6.7 Spectral representation and dispersion relation

253

**Using the deﬁnition in (6.132), we can also write this relation as 1 π
**

′ k0 ∞ −∞

¯ f (k0 ) − f (k0 ) =

ǫ→0+

lim ×

′ ′ ¯ dk0 Im (f (k0 ) − f (k0 ))

1 1 − ′ ¯ . − k0 − iǫ k0 − k0 − iǫ

(6.134)

In many cases of physical interest, the imaginary part of a retarded function is odd, namely,

Im f (k0 ) = −Im (−k0 ).

(6.135)

¯ If this is true and if we choose the point of subtraction k0 = 0, then the relation (6.134) takes the form 1 π

∞ −∞ ′ ′ Im f (k0 ) Im f (k0 ) − ′ . ′ k0 − k0 − iǫ k0 − iǫ (6.136)

f (k0 ) − f (0) = lim

ǫ→0+

′ dk0

In this case, we truly have a subtracted dispersion relation. However, if (6.135) does not hold, then the appropriate dispersion relation is given by (6.134). (Higher subtractions are needed if the function has a more singular behaviour at inﬁnity.) As a simple example of the discussions on the dispersion relation (6.130), let us look at the momentum space retarded propagator deﬁned in (5.129) (in our metric k0 = k0 ) GR (k0 , k) = lim 1 . 2 2 k0 − Ek + ik0 η

η→0+

(6.137)

This has the imaginary part given by

2 2 Im GR (k0 , k) = −πǫ(k0 )δ(k0 − Ek ) π = − [δ(k0 − Ek ) − δ(k0 + Ek )] . 2Ek

(6.138)

254

6 Self-interacting scalar field theory

We note that the imaginary part, in this case, satisﬁes (6.135). Furthermore, it vanishes for large values of k0 and, therefore, we expect (6.130) to hold in this case. Explicitly we note that

∞ −∞ ′ Im GR (k0 , k) ′ k0 − k0 − iǫ ′ dk0

ǫ→0+

lim = = = =

1 π

′ dk0

ǫ→0+

lim − lim −

π 2πEk

1 ′ ′ δ(k0 − Ek ) − δ(k0 + Ek ) ′ k0 − k0 − iǫ

ǫ→0+

ǫ→0+

lim

1 2 (k0 + iǫ)2 − Ek 1

2 k0

1 1 1 − 2Ek Ek − k0 − iǫ −Ek − k0 − iǫ

η→0+

lim

−

2 Ek

+ ik0 η

,

(6.139)

where we have used the fact that ǫ is inﬁnitesimal and have identiﬁed, as in (5.130), η = 2ǫ. Equation (6.139) can be compared with the dispersion relation in (6.130). 6.8 References 1. J. Schwinger, Physical Review 74, 1439 (1948); ibid. 75, 651 (1949); ibid. 76, 790 (1949); S. Tomonaga, Progress of Theoretical Physics 1, 27 (1946). 2. F. J. Dyson, Physical Review 75, 486 (1949); em ibid. 82, 428 (1951). 3. G. C. Wick, Physical Review 80, 268 (1950). 4. E. L. Hill, Reviews of Modern Physics 23, 253 (1957). 5. S. Schweber, Introduction to Relativistic Quantum Field Theory, Row, Peterson, Evanston (1961). 6. J. D. Bjorken and S. Drell, Relativistic Quantum Fields, McGrawHill, New York, 1964.

6.8 References

255

7. P. Roman, Introduction to Quantum Field Theory, John Wliley, New York (1969). 8. C. Itzykson and J-B. Zuber, Quantum Field Theory, McGrawHill, New York, 1980. 9. N. N. Bogoliubov and D. V. Shirkov, Introduction to the theory of Quantized Fields, Nauka, Moscow (1984). 10. A. Das, Lectures on Quantum Mechanics, Hindustan Publishing, New Delhi, India, 2003.

This page intentionally left blank .

2 (7.3) φ1 (x) = i φ2 (x) = − √ φ(x) − φ† (x) . The relations in (7. spin zero (scalar) ﬁelds which are Hermitian. the dynamical equations take the forms ∂µ ∂ µ + m2 φ(x) = 0.1) Therefore. 2 (7. 2 1 √ (φ1 (x) − iφ2 (x)) . there are now two distinct equations. φ† (x) = φ(x). ∂µ ∂ µ + m2 φ† (x) = 0. The physical meaning of this system can be understood by noting that we can express the complex ﬁeld operator in terms of two real (Hermitian) ﬁeld operators as 1 √ (φ1 (x) + iφ2 (x)) .1 Quantization The natural ﬁeld theoretic example to consider next is that of a free Klein-Gordon ﬁeld theory where the ﬁeld operator is not Hermitian.Chapter 7 Complex scalar ﬁeld theory 7. (7. 2 257 . In this case.2) can be inverted to give 1 √ φ(x) + φ† (x) .2) φ(x) = φ† (x) = where φ1 (x) and φ2 (x) are two distinct.

and is manifestly Hermitian. we can think of the complex Klein-Gordon ﬁeld theory as describing two independent ﬁeld degrees of freedom – either in terms of φ1 (x) and φ2 (x) or through φ(x) and φ† (x) – each of which is treated as an independent variable. two distinct particles) with degenerate mass. Therefore. the equal time canonical commutation relations follow to be [φi (x). The Lagrangian density which will give the dynamical equations in (7. independent spin zero Klein-Gordon ﬁelds (and. such a system equivalently describes two free. φj (y)]x0 =y0 = 0 = [Πi (x). ∂µ ∂ µ + m2 φ2 (x) = 0. (7.4) as the Euler-Lagrange equations is easily determined to be m2 2 1 m2 2 1 ∂µ φ1 ∂ µ φ1 − φ1 + ∂µ φ2 ∂ µ φ2 − φ 2 2 2 2 2 m2 1 ∂µ (φ1 − iφ2 ) ∂ µ (φ1 + iφ2 ) − (φ1 − iφ2 ) (φ1 + iφ2 ) 2 2 (7. ˙ ∂ φi (x) i = 1. therefore. (Note that we have not been careful about the order of the factors in rearranging terms in (7. (7.4) Namely. If we treat φ1 (x) and φ2 (x) as the independent variables. 2.) We can quantize this theory in one of the two equivalent ways. (7.6) Consequently.5) because distinct degrees of freedom are expected to commute.1) in terms of the real ﬁeld variables as ∂µ ∂ µ + m2 φ1 (x) = 0.5) L = = = ∂µ φ† ∂ µ φ − m2 φ† φ. then we can deﬁne the momenta conjugate to the two ﬁeld operators as Πi (x) = ∂L ˙ = φi (x). Πj (y)]x0 =y0 = iδij δ3 (x − y).1) or (7.7) .258 7 Complex scalar field theory It follows now that we can express the dynamical equations in (7. [φi (x). Πj (y)]x0 =y0 .

2 2 . if we treat φ(x) and φ† (x) as the independent dynamical variables. the Hamiltonian density is obtained in a straightforward manner to be of the form 2 H = i=1 ˙ Πi φi − L 1 1 m2 2 = Π2 + Π2 − Π2 + ∇φ1 · ∇φ1 + φ 1 2 1 2 2 2 1 1 m2 2 1 φ − Π2 + ∇φ2 · ∇φ2 + 2 2 2 2 2 = 1 2 1 m2 2 Π1 + ∇φ1 · ∇φ1 + φ 2 2 2 1 1 1 m2 2 + Π2 + ∇φ2 · ∇φ2 + φ 2 2 2 2 2 2 = i=1 1 2 1 m2 2 Πi + ∇φi · ∇φi + φ .9) Alternatively.7.6).8) where we have used the identiﬁcation in (7.5) to be Π(x) = = ∂L ˙ = φ(x) ˙ ∂ φ† (x) 1 1 ˙ ˙ √ φ1 (x) + iφ2 (x) = √ (Π1 (x) + iΠ2 (x)) . then we can deﬁne the momenta conjugate to the ﬁeld operators from (7. We see that the Hamiltonian density of the free complex scalar ﬁeld theory is simply the sum of the Hamiltonian densities for two non-interacting real Klein-Gordon ﬁeld theories with degenerate mass and H= d3 x H.1 Quantization 259 In this case. (7. 2 2 2 i (7.

2 Field decomposition As in the case of the Hermitian (real) Klein-Gordon ﬁeld. in this case.260 Π† (x) = ∂L ˙ = φ† (x) ˙ ∂ φ(x) 7 Complex scalar field theory 1 1 ˙ ˙ = √ φ1 (x) − iφ2 (x) = √ (Π1 (x) − iΠ2 (x)). 7.8) and H= d3 x H.10) 2 2 As a result. φ† (y) = φ† (x). although it is to be understood that x0 = y 0 in all of the relations in (7. take the forms (we do not write the equal time condition explicitly for simplicity. Π(y) = iδ3 (x − y).10). in this case. [Π(x).11)) [φ(x). (7. the equal time canonical commutation relations.56)). which can be compared with (7. Π† (y) = φ† (x). is manifestly Hermitian as it should be even though the dynamical variables are not. φ† (y) = 0.13) Note that the Hamiltonian. using (7.12) = Π† Π + ∇φ† · ∇φ + m2 φ† φ. (7. (7. Π† (y) = Π† (x). we can derive the Hamiltonian density in this formulation to have the form ˙ ˙ H = Πφ† + Π† φ − L ˙ ˙ = ΠΠ† + Π† Π − φ† φ + ∇φ† · ∇φ + m2 φ† φ = 2Π† Π − Π† Π + ∇φ† · ∇φ + m2 φ† φ (7. φ(y)] = φ(x).5) and (7. Π(y)] = Π(x). of the form . in the present case. Π† (y) = 0.11) Furthermore. we can expand the ﬁeld φ(x) in the basis of plane wave solutions of the Klein-Gordon equation (see (5. φ(x).

66)) ai (k). a† (k′ ) = δij δ3 (k − k′ ).15) Given the decomposition in (7.14) with k0 = Ek = k2 + m2 > 0. e−ik·x b(k) + eik·x a† (k) . 2. can then be interpreted as the annihilation and the creation operators for the two kinds of quanta associated with the ﬁelds φi (x). i i = 1. aj (k′ ) = 0 = a† (k). j (7. ai (k) and a† (k).2)) φ(x) = = 1 √ (φ1 (x) + iφ2 (x)) 2 1 √ 2 d3 k (2π)3 2k0 e−ik·x (a1 (k) + ia2 (k)) +eik·x a† (k) + ia† (k) 1 2 = φ† (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k) + eik·x b† (k) .17) . (7. i j ai (k). 2 1 √ (a1 (k) − ia2 (k)) . As before. The commutation relations between these creation and annihilation operators can then be obtained from the canonical quantization relations to be (see (5. 2.7. it is clear that the ﬁeld operators φ(x) and φ† (x) can also be expanded in the plane wave basis as (see (7. (7. a† (k′ ) . with i i = 1.16) where we have deﬁned a(k) = b(k) = 1 √ (a1 (k) + ia2 (k)) . 2 (7.14).2 Field decomposition 261 φi (x) = d3 k (2π)3 2k0 √ e−ik·x ai (k) + eik·x a† (k) .

we can write : H := d3 k Ek a† (k)a(k) + b† (k)b(k) . a† (k). a(k).15) and the deﬁnition in (7. a† (k′ ) = a† (k). b(k′ ) = 0.19) If we normal order the Hamiltonian in (7.16)) and take the form a(k).12)) can be written.262 7 Complex scalar field theory which correspond to the annihilation operators associated with the quanta for the two ﬁelds φ(x) and φ† (x) respectively. a(k).20) (We can equivalently normal order the Hamiltonian (7. b† (k′ ) = 0. b† (k′ ) = b(k).17) b (or directly from (7. (7. (7.11) and (7. a† (k′ ) = b(k). b† (k′ ) = δ3 (k − k′ ). b† (k′ ) = b† (k). b(k). b(k′ ) = a(k). in two equivalent ways as H= = d3 x H d3 k Ek † a1 (k)a1 (k) + a1 (k)a† (k) 1 2 + a† (k)a2 (k) + a2 (k)a† (k) 2 2 = d3 k Ek † a (k)a(k) + a(k)a† (k) 2 + b† (k)b(k) + b(k)b† (k) .18) The Hamiltonian for the system (see (7. The commutation relations between the a(k). (7.8) or (7. a† (k′ ) = 0.19) with respect to the a(k) and b(k) operators. a† (k) and † (k) can now be obtained from (7.19) with respect to the ai (k) operators and in that case the Hamiltonian would . in terms of the creation and the annihilation operators. a(k′ ) = b(k).

After all. of course. 7. The theory is. as was also the case for a real Klein-Gordon theory. H|0 = 0.) The vacuum state in such a case can be deﬁned to be the lowest energy state annihilated by both a(k) and b(k). = b† (k)|0 . (7.3 Charge operator 263 be a sum of two Klein-Gordon Hamiltonians that we have already studied in chapter 5. in the present case. the theory is invariant.22) ˜ |k Similarly. manifestly Lorentz invariant and translation invariant. The degeneracy can be understood as follows.7. Such a transformation is not a space-time symmetry transformation since . there is in addition an internal symmetry associated with the system. (7. However. there will now be two distinct one particle states degenerate in mass (and.23) where θ is a real constant parameter of transformation (a global transformation). degeneracies arise when there is a symmetry present in the system under study. let us analyze the symmetry properties of this system. we can also construct higher order states which will also show a degeneracy that was not present in the case of the real KleinGordon theory. therefore. under the transformation φ(x) → φ′ (x) = e−iθ φ(x). energy) given by |k = a† (k)|0 . Namely. as we know from our studies in non-relativistic quantum mechanics. φ† (x) → φ′ † (x) = φ† (x)eiθ .21) However. (7. as we will see shortly. Namely.3 Charge operator To understand the physical meaning of the two distinct and degenerate one particle states. a(k)|0 = 0 = b(k)|0 = 0|a† (k) = 0|b† (k).

Inﬁnitesimally. φ† (x). δφ(x) = φ′ (x) − φ(x) = −iǫφ(x).13)) .24)) showing that it is a symmetry of the system and in this case we have K µ = 0.8)) is given by L φ′ (x). ∂µ φ′ † (x) −L φ(x). the transformation in (7.23) (or the inﬁnitesimal form (7.26) On the other hand.264 7 Complex scalar field theory the space-time coordinates are not changed by this transformation. (7. the change in the Lagrangian density (see (6.25) The derivatives do not act on the exponentials since θ is a constant parameter (independent of space-time coordinates). such a transformation is known as an internal symmetry transformation. Thus. ∂µ φ′ (x).23) (or the inﬁnitesimal form of it in (7. (7. in this case. φ′ † (x). We note that under the global phase transformation (7.24)). for inﬁnitesimal transformations in (7. ∂µ φ† (x) = ∂µ φ′ † ∂ µ φ′ − m2 φ′ † φ′ − ∂µ φ† ∂ µ φ + m2 φ† φ = ∂µ φ† eiθ e−iθ ∂ µ φ − m2 φ† eiθ e−iθ φ − ∂µ φ† ∂ µ φ + m2 φ† φ = 0. we have (see (6.5) is invariant under the global phase transformation in (7. (7.24) where ǫ denotes the inﬁnitesimal constant parameter of transformation. the Lagrangian density (7.24). ∂µ φ(x). Rather.23) takes the form (α = ǫ = inﬁnitesimal) δφ† (x) = φ′ † (x) − φ† (x) = iǫφ† (x).

27) (Note here that classically the order of the factors is not important.29) This is a vector current since the parameter of transformation is a scalar. we have freely rearranged the order of factors in (7. (7. we can deﬁne the N¨ther current associated with this o symmetry to be (see (6. Even in a quantum theory. the current can be easily checked to be conserved using the equations of motion in (7. under normal ordering the order of factors does not matter (for a bosonic theory). (7.) Furthermore.) Therefore. (In fact. as we have seen. we normal order all observables such as currents and charges and.16). this is the probability current density (up to a normalization) which we had studied in the ﬁrst quantized theory.28) and the parameter independent current follows to be ← → J µ = iφ† (x) ∂ µ φ(x). therefore.30) Using the ﬁeld decomposition (7. But here it is an operator. (7.7. † ∂∂µ φ ∂∂µ φ µ Jǫ = δφ (7.27). Of course. we can show that the associated conserved charge of the theory can be written in terms of the creation and annihilation operators as .13)) ← → ∂L ∂L + δφ† − K µ = iǫφ† (x) ∂ µ φ(x).1) ∂µ J µ (x) = 0.3 Charge operator 265 δφ ∂L ∂L + δφ† ∂∂µ φ ∂∂µ φ† = −iǫφ(x)∂ µ φ† (x) + iǫφ† (x)∂ µ φ(x) = iǫ φ† (x)∂ µ φ(x) − φ(x)∂ µ φ† (x) = iǫ φ† (x)∂ µ φ(x) − (∂ µ φ† (x))φ(x) ← → = iǫφ† (x) ∂ µ φ(x).

32) annihilates the vacuum.32) Since this charge operator is associated with a global phase transformation of the kind associated with electromagnetic interactions (namely.33) so that the vacuum does not carry any charge. (7. we are suppressing the normal ordering symbol for simplicity with the understanding that the charge operator has the normal ordered form in (7.266 7 Complex scalar field theory Q = = i = d3 x J 0 (x) = i ← → d3 x φ† (x) ∂ 0 φ(x) ˙ ˙ d3 x φ† (x)φ(x) − φ† (x)φ(x) d3 k a† (k)a(k) − b(k)b† (k) . We note that the normal ordered charge operator (7. the charge operator gives Q|k = = = = d3 k′ a† (k′ )a(k′ ) − b† (k′ )b(k′ ) a† (k)|0 d3 k′ a† (k′ )a(k′ )a† (k)|0 d3 k′ a† (k′ ) a(k′ ). a U (1) phase transformation) we can identify this operator as the electric charge operator.34) d3 k′ a† (k′ )δ3 (k′ − k)|0 = a† (k)|0 = |k . (7.32).22). . Acting on the ﬁrst of the two one particle states in (7. Q|0 = d3 k a† (k)a(k) − b† (k)b(k) |0 = 0. (7.31) If we normal order. the charge operator takes the form : Q := d3 k a† (k)a(k) − b† (k)b(k) . Here. a† (k) + a† (k)a(k′ ) |0 (7.

the second quantized description is symmetric under the interchange of particles and anti-particles. They describe two distinct particles with degenerate mass but with exactly opposite electric charge.37) That is. The vacuum in this bosonic theory.22). the one particle state |k is an eigenstate of the charge operator with eigenvalue (+1). These are. ˜ Namely. This transformation is otherwise known as the charge conjugation symmetry which we will . of course. b† (k) + b† (k)b(k′ ) |0 = = − = − = − d3 k′ b† (k′ )δ3 (k′ − k)|0 = −b† (k)|0 (7. (7. acting on the second one particle state in (7.35) ˜ = −|k . In contrast. the second one particle state of the theory |k is also an eigenstate of the charge operator but with the eigenvalue (−1).36) we note that it is invariant under the discrete transformation a(k) ↔ b(k). the second quantized theory leads to a much more satisfactory description of particles and anti-particles. the charge operator leads to ˜ Q|k d3 k′ a† (k′ )a(k′ ) − b† (k′ )b(k′ ) b† (k)|0 d3 k′ b† (k′ )b(k′ )b† (k)|0 d3 k′ b† (k′ ) b(k′ ).7. H= (7.5. from the form of the Hamiltonian (7. in particular. Furthermore.3 Charge operator 267 In other words.20) d3 k Ek a† (k)a(k) + b† (k)b(k) . Therefore. the particle and the anti-particle states – but both of them now correspond to positive energy states. contains no particle and this should be contrasted with the deﬁnition of the Dirac vacuum discussed in section 2. The meaning of the two one particle states is now clear.

Since we have already constructed the Feynman Green’s function for the real Klein-Gordon ﬁeld theory in (5. we note that the complex Klein-Gordon ﬁeld theory describes spin zero particles carrying electric charge while the real Klein-Gordon ﬁeld theory describes charge neutral spin zero particles. let us discuss the Feynman Green’s function in the basis of the complex φ(x) and φ† (x) ﬁelds. Thus.146).4 Green’s functions In chapter 6 we have already seen that the Feynman Green’s functions play the most important role in the calculations of the S-matrix in a quantum ﬁeld theory. we will construct only the Feynman Green’s function for the complex KleinGordon ﬁeld theory.16) as well as the commutation relations in (7.18) we now obtain 0|φ(x)φ(y)|0 = d3 kd3 k′ √ 0| e−ik·x a(k) + eik·x b† (k) 3 k0 k′ 0 2(2π) × e−ik ·y a(k′ ) + eik ·y b† (k′ ) |0 = 0. in this section.107) that the Feynman Green’s function can be related to the vacuum expectation value of the time ordered product of two ﬁeld operators. Using the ﬁeld decomposition in (7. We note from (6. 0|φ† (x)φ† (y)|0 = d3 kd3 k′ √ 0| e−ik·x b(k) + eik·x a† (k) 2(2π)3 k0 k′ 0 × e−ik ·y b(k′ ) + eik ·y a† (k′ ) |0 = 0. 7. 0|φ(x)φ† (y)|0 = d3 kd3 k′ √ 0| e−ik·x a(k) + eik·x b† (k) 2(2π)3 k0 k′ 0 ′ ′ ′ ′ . to conclude this section. Therefore.268 7 Complex scalar field theory study in a later chapter.

7. 0|T (φ† (x)φ† (y))|0 = θ(x0 − y 0 ) 0|φ† (x)φ† (y)|0 + θ(y 0 − x0 ) 0|φ† (y)φ† (x)|0 = 0.145). Here we have used (7.39) .4 Green’s functions 269 × e−ik ·y b(k ) + eik ·y a† (k ) |0 = = = d3 kd3 k √ e−ik·x+ik ·y 0|a(k)a† (k )|0 2(2π)3 k0 k 0 d3 kd3 k √ 2(2π)3 k0 k 0 e−ik·x+ik ·y δ3 (k − k ) d3 k e−ik·(x−y) (2π)3 2k0 = −iG(+) (x − y). (7.38) = iG(−) (x − y).38). 0|φ† (y)φ(x)|0 = d3 kd3 k √ 2(2π)3 k0 k 0 0| e−ik·y b(k) + eik·y a† (k) × e−ik ·x a(k ) + eik ·x b† (k ) |0 = = = d3 kd3 k √ e−ik·y+ik ·x 0|b(k)b† (k )|0 2(2π)3 k0 k 0 d3 kd3 k √ 2(2π)3 k0 k 0 eik ·x−ik·y δ3 (k − k ) d3 k eik·(x−y) (2π)3 2k0 (7.21) as well as the deﬁnitions of the positive and negative energy Green’s functions in (5.143) and (5. it now follows that 0|T (φ(x)φ(y))|0 = θ(x0 − y 0 ) 0|φ(x)φ(y)|0 + θ(y 0 − x0 ) 0|φ(y)φ(x)|0 = 0. Given the relations in (7.

there is only one independent Feynman Green’s function that is related to the vacuum expectation value of the time ordered product of φ(x)φ† (y).2) (in order to be consistent with the standard notation used in the discussion of spontaneous symmetry breaking in the literature). Denoting φ1 = σ and φ2 = χ in the deﬁnition in (7. namely. 7. (7.42) .270 On the other hand.146).5 Spontaneous symmetry breaking and the Goldstone theorem We can introduce interactions into the complex Klein-Gordon ﬁeld theory much the same way we did for the real Klein-Gordon ﬁeld theory. 0|T (φ† (x)φ(y))|0 = 0|T (φ(y)φ† (x))|0 = iGF (y − x) = iGF (x − y). we see that in this case. we have 0|T (φ(x)φ† (y))|0 7 Complex scalar field theory = θ(x0 − y 0 ) 0|φ(x)φ† (y)|0 + θ(y 0 − x0 ) 0|φ† (y)φ(x)|0 = −iθ(x0 − y 0 )G(+) (x − y) + iθ(y 0 − x0 )G(−) (x − y) = iGF (x − y). deﬁning 1 φ(x) = √ (σ(x) + iχ(x)) .40) Here we have used the deﬁnition of the Feynman Green’s function in (5. 4 L = ∂µ φ† ∂ µ φ − m2 φ† φ − λ > 0. It is easy to check that the Lagrangian density with the quartic interaction λ † 2 (φ φ) . Thus. 2 (7. (7.23). However.41) describes the most general (renormalizable) self-interacting theory for the complex Klein-Gordon ﬁeld. in this case we should look for interactions that would preserve the internal global symmetry in (7.

this describes the theory of two real Klein-Gordon ﬁelds with quartic self-interactions and which are also interacting with each other.7. the inﬁnitesimal transformations (7.24) can be shown to have the form δσ(x) = ǫχ(x). ∂σ ∂χ (7.42). δχ(x) = −ǫσ(x). 16 L = (7. ∂V ∂V =0= . In the real basis of (7.) λ † 2 (φ φ) 4 2 V = m2 φ† φ + = λ m2 2 σ + χ2 + σ 2 + χ2 2 16 . the minimum of the potential is obtained by requiring ∂V ∂V =0= .47) .45) and we note that. The Lagrangian density (7. (7. ∂φ ∂φ† or.44) The Hamiltonian for the interacting theory (7.43) is invariant under the global phase transformation (7. the other (kinetic energy) terms would only add positively to the energy.41) or (7.41) has the form λ † 2 (φ φ) .41) can also be written as m2 2 1 (∂µ σ∂ µ σ + ∂µ χ∂ µ χ) − σ + χ2 2 2 λ 2 − σ 2 + χ2 .24).5 Spontaneous symmetry breaking and the Goldstone theorem 271 we note that the Lagrangian density (7. the minimum of energy coincides with the minimum of the potential (We note that if the ﬁeld φ is not constant.46) In general.43) Namely.23) or the inﬁnitesimal form in (7. 4 H= d3 x Π† Π + ∇φ† · ∇φ + m2 φ† φ + (7. for constant ﬁeld conﬁgurations. (7.

(7. (7. perturbation around the minimum of the potential is stable. c χc = 0|χ(x)|0 = 0.23) (or (7. In quantum theory. 16 L = (7. this Lagrangian density takes the form 1 m2 2 (∂µ σ∂ µ σ + ∂µ χ∂ µ χ) + σ + χ2 2 2 λ 2 − σ 2 + χ2 .46) is a sum of positive terms. equivalently σc = χc = 0.42). this translates to the fact that we can deﬁne the ground state of the theory such that φ† = 0|φ† (x)|0 = 0.24) and (7.50) and (7. σc = 0|σ(x)|0 = 0. a concept that we do not get into here) φc = φ† = 0. c or. (7.56) has exactly this property (see (5. and perturbation theory developed around such a vacuum would be stable.51) Both the Lagrangian densities in (7. it is clear that the minimum occurs at (the subscript c is supposed to denote a classical ﬁeld. since the potential in (7. Let us next consider the same Lagrangian density for the complex scalar ﬁeld as in (7.41) except that we change the sign of the mass term L = ∂µ φ† ∂ µ φ + m2 φ† φ − λ † 2 (φ φ) .49) φc = 0|φ(x)|0 = 0.272 7 Complex scalar field theory In the present case.48) As we know from studies in classical mechanics. We note from our discussion of the real Klein-Gordon theory that the plane wave expansion for the ﬁeld operator in (5. or.50) In terms of the real ﬁelds σ and χ deﬁned in (7.44)). Changing the sign of the mass term does not change the .51) are clearly invariant under the global phase transformation in (7. 4 λ > 0.106)).

c = 1) leading to an acausal propagation (and.7. Such particles are known as tachyons and a theory with tachyons.4)) − m2 φ(x) = 0. corresponding to the Einstein relation (7.51) has the form V = −m2 φ† φ + = − λ † 2 (φ φ) 4 2 m2 2 λ σ + χ2 + σ 2 + χ2 2 16 . 2 − m2 ∂|k| k (7.55) The minimum of this potential is obtained from (7. or. in this case. To begin with.54) Namely. but it does change the Hilbert space structure of the theory in a very interesting way.41) would lead to the dynamical equation (compare with (1. (7.50) or (7. therefore.5 Spontaneous symmetry breaking and the Goldstone theorem 273 symmetry behaviour. a violation of causality).48) .53) This corresponds to a particle with an imaginary mass and has the immediate consequence that the group velocity for the particle motion (this is also the same as the phase velocity in the present case) exceeds unity |k| ∂k0 =√ > 1. (7. let us note that in the absence of interaction (λ = 0). in general. k0 = k2 − m2 . for constant ﬁeld conﬁgurations the potential of the theory (7. Let us note that.40) or (5. the theory (7.52) k2 = −m2 . has many undesirable features. the particle will travel faster than the speed of light (recall that in our units.

57) λ For the second of these solutions. 4m2 . let us analyze the second derivatives of the potential at the two extrema in (7. = σ −m2 + ∂V ∂χ = −m2 χ + λ χ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.60) 2m σc =± √ .χc =0 λ = σc =0=χc ∂2V ∂χ2 σc =0=χc = −m2 .59) = 0.57) which lead to ∂2V ∂σ 2 ∂2V ∂σ∂χ while ∂2V ∂σ 2 ∂2V ∂χ2 ∂2V ∂σ∂χ 2m σc =± √ .χc =0 λ . for example. (7. λ χc = 0.56). we can choose σc = 0 = χc . (7. 2m σc =± √ . = 0. σc =0=χc = −m2 + = −m2 + = 0. (7. 2 σc + χ2 = c 2m σc = ± √ .χc =0 λ (7. (7.58) To determine the true minimum.274 7 Complex scalar field theory ∂V ∂σ = −m2 σ + λ σ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0. namely. 3λ 4 λ 4 4m2 λ 4m2 λ = 2m2 .56) = χ −m2 + There now appear to be two sets of solutions to the equations in (7.

1: Potential leading to spontaneous breaking of symmetry.57). There are several interesting things to note from this analysis.62) so that the second solution.7.1. indeed. All of this is more easily seen from the graph of the potential in Fig. the ﬁrst corresponds to a local maximum. However. First. the second solution in (7. . λ (7. Each point on this circle is related by the symmetry transformation (7.5 Spontaneous symmetry breaking and the Goldstone theorem 275 In other words we note that out of the two solutions in (7.55) that V (σc = χc = 0) = 0.57) is a true minimum and we conclude that the true minimum of the potential occurs for 2 2φ† φc = σc + χ2 = c c 4m2 .44). V (σ. λ λ λ (7.61) In fact.61). we note that there is an inﬁnity of minima (in the σ-χ ﬁeld space) of the potential given by the circle (7. leads to a lower energy. V 2 ((σc + χ2 ) c m2 4m2 ) = − = λ 2 = − 4m2 λ λ + 16 4m2 λ 2 2m4 m4 m4 + =− . On the other hand. χ) V =0 χ 2m −√ λ 2m √ λ σ Figure 7. 7. we note explicitly from the form of the potential in (7.

χ(x)] |0 (7. This has to be contrasted with the case where an interaction term in the Hamiltonian can break the symmetry of a theory explicitly.61) to correspond to the point (in the σ-χ ﬁeld space) 2m σc = √ . such a breaking manifests in the following manner. the action or the Hamiltonian of the theory is invariant under the symmetry transformation.63) then. it now follows that = −iǫ 0| [Q.66) which implies that . σ(x)] = ǫχ(x). if we choose the minimum in (7.65) 0|δχ(x)|0 2m = −ǫ 0|σ(x)|0 = − √ ǫ. for example. In the quantum theory. (7. Thus.49)) 2m 0|σ(x)|0 = √ . in the quantum theory this would imply (see (7.64) We recall that conserved charges are the generators of inﬁnitesimal symmetry transformations of the theory and generate the transformations through commutators.44) spontaneously. We recall that any choice for the minimum of the potential would translate into the properties of the vacuum state of the quantum theory. λ (7. χ(x)] = −ǫσ(x). δχ(x) = −iǫ [Q.64). the inﬁnitesimal transformations in (7. From (7.276 7 Complex scalar field theory the choice of any particular point on this circle as the minimum of the potential to develop perturbation theory (or to build the quantum theory) breaks the symmetry (7. In this case. λ χc = 0. In this case.44) can be written in terms of the commutators with the charge operator as δσ(x) = −iǫ [Q. but the ground state is not. λ 0|χ(x)|0 = 0. (7.

(7. Therefore. (7. or. H] = 0. although the Hamiltonian of the theory is. (7.69) that [Q.5 Spontaneous symmetry breaking and the Goldstone theorem 277 Q|0 = 0. t)Q|0 d3 x 0|eiP ·x J 0 (0)e−iP ·x Q|0 . or. HQ|0 = E0 Q|0 . (note that Q is Hermitian) χ|χ = = = 0|QQ|0 0| d3 x J 0 (x. it is suggestive that we can think of this state as another vacuum state.69) would appear to be degenerate with the vacuum state |0 in energy.31) and (7. This can be easily seen from (7. let us denote Q|0 = |χ It follows now from (7. H|χ = E0 |χ . (QH − HQ)|0 = 0. [Q.69) where we have denoted the energy of the vacuum |0 as E0 and this derivation shows that the state |χ deﬁned in (7. (7.68) Since Q does not annihilate the vacuum of the present theory.68) and (7.70) (7.69) as follows. namely.7. or.67) This explicitly shows that the vacuum state of the theory is not invariant under the symmetry transformation in this case. The problem with this interpretation is that the state is not normalizable.71) . H]|0 = 0.

(7. it follows that the momentum operator also commutes with Q and.74) In other words.75) takes a state out of the Hilbert space. (7. In fact. This analysis also shows that the ﬁnite transformation operator U (θ) = e−iθQ . the state |χ is not normalizeable and hence cannot be thought of as another vacuum. (7.72) χ|χ = = = d3 x 0|eiP ·x J 0 (0)Qe−iP ·x |0 d3 x 0|J 0 (0)Q|0 0|J 0 (0)Q|0 d3 x −→ ∞.66) that χ(x) corresponds to the Nambu-Goldstone boson of the theory.72) leads to (7. (7. therefore. we have [Pµ . Q] = 0. . and as a consequence.73) where we have used the property of the ground state Pµ |0 = 0.278 7 Complex scalar field theory We have already seen that the Hamiltonian commutes with Q expressing the fact that it is independent of time. Since Q does not depend on spatial coordinates. it is straightforward to show that the charge Q does not exist when there is spontaneous breakdown of the symmetry and the formally unitary operator in (7.75) does not act unitarily on the Hilbert space. The ﬁeld whose variation (under the inﬁnitesimal symmetry transformation) develops a nonzero vacuum expectation value is conventionally known as the Goldstone ﬁeld (Nambu-Goldstone boson in the case of spontaneous breakdown of a bosonic symmetry) and in this case we see from (7.

(7.77).1 that the massless mode would correspond to oscillations along the valley of minima while the massive mode would represent the orthogonal oscillations.77) so that the shifted ﬁelds have vanishing vacuum expectation values (vevs) 0|σ(x)|0 = 0 = 0|χ(x)|0 . λ χ = 0|χ(x)|0 = 0. (7. is given by σ(x) = σ ′ (x) + σ with the later identiﬁcation σ ′ (x) = σ(x) as the dynamical variable.60) shows that in this case of spontaneous breakdown of the symmetry.64) 2m σ = 0|σ(x)|0 = √ . λ χ(x) → χ(x) + χ = χ(x). √ one of the ﬁelds has become massive with a mass 2m while the other is massless (in spite of the fact that the free theory had tachyons (see (7. We note that the second derivatives of the potential (with respect to ﬁeld variables) give the mass (squared) parameters for the ﬁeld operators (particles) in a theory and (7. We note from Fig.60) is as follows. With the choice of the solutions for the minima as in (7.51) becomes .5 Spontaneous symmetry breaking and the Goldstone theorem 279 The second interesting observation that follows from the analysis in (7. say for the ﬁrst term.7. the shift in (7.54))). We can see this quantitatively in the quantum theory as follows.77). the Lagrangian density in (7.53) and (7.76) we can shift the ﬁeld variables as (in classical mechanics this corresponds to analyzing small oscillations around the minimum of a potential) 2m σ(x) → σ(x) + σ = σ(x) + √ . 7.78) (Explicitly. (7.) Under the shift (7.

79) This shows explicitly that perturbed around the true minimum. This leads to a general theorem known as the Goldstone theorem which says that if in a manifestly Lorentz invariant theory with a positive deﬁnite metric Hilbert space. one of the scalar ﬁelds (in this case χ(x)) becomes massless while the other √ scalar ﬁeld (in this case σ(x)) is massive with a nontachyonic mass 2m.66)). of course.280 7 Complex scalar field theory L = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 λ − 16 2m σ+ √ λ 2 2 2m σ+ √ λ 2 + χ2 +χ 2 = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 λ − 16 2 2 4m 4m2 σ 2 + χ2 + √ σ + λ λ 2 4m2 4m σ +χ + √ σ+ λ λ = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 − + λ 16 (σ 2 + χ2 )2 + 4m2 4m σ 2 + χ2 + √ σ + λ λ 8m 16m2 2 16m4 σ + + √ σ(σ 2 + χ2 ) 2 λ λ λ 8m2 2 32m3 (σ + χ2 ) + √ σ λ λ3 = 1 m4 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ − m2 σ 2 + 2 2 λ √ m λ λ − σ(σ 2 + χ2 ) − (σ 2 + χ2 )2 . aware of this phenomenon (known as the spontaneous symmetry breaking phenomenon) from studies in other areas of physics. The massless boson is known as the Nambu-Goldstone boson (this is the one whose ﬁeld variation picks up a nonvanishing vacuum expectation value as shown in (7. then there must appear massless excitations. a continuous symmetry is spontaneously broken. 2 16 (7. We are. if we look at the Hamiltonian describing a . For example.

(The π meson which is the lightest among the spin zero bosons has a mass mπ ∼ 140 MeV. Unfortunately.6 Electromagnetic coupling It is clear from our discussions so far that the complex Klein-Gordon ﬁeld theory describes spin zero charged particles whereas the real Klein-Gordon ﬁeld theory describes spin zero charge neutral particles. The Goldstone theorem is interesting in the sense that it can provide a possible reason for the existence of massless spin zero bosons if they are found in nature. the Hamiltonian is invariant under rotations. The prescription is again through the minimal coupling discussed in (1. Consequently the rotational symmetry is spontaneously broken. we have eﬀectively chosen a deﬁnite orientation in the physical space.6 Electromagnetic coupling 281 magnet (for example. Given the complex Klein-Gordon ﬁeld theory. The analogue of the massless Nambu-Goldstone bosons in this case correspond to the long range correlations associated with the spin waves. The ground state of the theory is clearly the state (spin conﬁguration) where all the spins are aligned. Thus. in the case of the Ising model). H = −g si · si+1 .7. there are two such vacuum conﬁgurations allowed in one dimension. ∂µ → ∂µ + ieAµ . (For example. we can ask how such a system can be coupled to a background electromagnetic ﬁeld. it is clear that once we have chosen a particular ground state (vacuum) where all the spins are pointing in a given direction.) However.81) where “e” denotes the charge carried by the particle.80) i where the negative sign in the Hamiltonian (g > 0) is for ferromagnets. namely. the Lagrangian density describing the interaction of charged spin zero ﬁelds with a background electromagnetic ﬁeld (without the quartic self-interaction term) is given by . (7.) 7. when the spins are all pointing up or down. (7.53). we do not observe any massless spin 0 bosons in nature.

note that under this transformation.82) Dµ φ = ∂µ φ + ieAµ φ.84) where the parameter of transformation θ(x) is now a local function of space-time coordinates (unlike in (7. φ† (x) → φ′ † (x) = φ† (x)eiθ(x) .82) is invariant under a local (U (1)) phase transformation of the form φ(x) → φ′ (x) = e−iθ(x) φ(x). As we can verify readily. ′ Dµ φ(x) → Dµ φ′ (x) = = ∂µ + ieA′ (x) φ′ (x) µ ∂µ + ie Aµ (x) + 1 (∂µ θ(x)) e e−iθ(x) φ(x) = e−iθ(x) (−i(∂µ θ(x)) + ∂µ + ieAµ (x) + i(∂µ θ(x))) φ(x) = e−iθ(x) (∂µ + ieAµ (x)) φ(x) = e−iθ(x) Dµ φ(x). (7. In fact. e (7.282 7 Complex scalar field theory L = ∂µ φ† ∂ µ φ − m2 φ† φ (∂µ − ieAµ ) φ† (∂ µ + ieAµ ) φ − m2 φ† φ (Dµ φ)† (D µ φ) − m2 φ† φ. → ((∂µ + ieAµ ) φ)† (∂ µ + ieAµ ) φ − m2 φ† φ = = (7.83) is known as the covariant derivative since it transforms covariantly under a local phase transformation (or equivalently under a gauge transformation) as we will see shortly.82) The modiﬁed derivative Dµ introduced in (7.85) . the Lagrangian density (7. Aµ (x) → A′ (x) = Aµ (x) + µ 1 ∂µ θ(x).23)). (7.

A. . Physical Review 127. Peterson. the electromagnetic ﬁeld in the present discussion has no dynamics (it is a background ﬁeld and not a dynamical ﬁeld) which we will develop in chapter 9.84) requires an additional ﬁeld (Aµ ). Of course. S. which is obtained by minimally coupling the complex scalar ﬁeld to the electromagnetic ﬁeld. unlike the global phase invariance in (7. Introduction to Relativistic Quantum Field Theory. Goldstone. Nambu and G. This is a very general feature in a quantum ﬁeld theory.7 References 1. Schweber.86) Consequently. Row. 345 (1961). 7. the theory described by (7. 965 (1962). Aµ .) It follows from this observation that under the local transformation (7. the covariant derivative acting on a ﬁeld transforms exactly like the ﬁeld itself (covariantly) under such a transformation (see (7. 2. J. φ′ † (x). 4.23). A′ (x)) µ = ′ Dµ φ′ (x) † D′ µ φ′ (x) − m2 φ′ † (x)φ′ (x) = (Dµ φ(x))† eiθ(x) e−iθ(x) D µ φ(x) − m2 φ† (x)eiθ(x) e−iθ(x) φ(x) = (Dµ φ(x))† (D µ φ(x)) − m2 φ† (x)φ(x) = L φ. Goldstone. J. φ† . Jona-Lasinio. is invariant under the local phase transformation (7. Y. 154 (1961). 3. Salam and S. Nuovo Cimento 19. φ† (x). the local phase invariance (7. namely.7.84)). (In contrast.84) L(φ(x). Local phase transformations are commonly known as gauge transformations and we note that.84). Aµ (x)) → L(φ′ (x).82). (7. note that ∂µ φ(x) does not transform covariantly. Physical Review 122. invariance under a local transformation necessarily requires additional ﬁelds known as gauge ﬁelds. Evanston (1961).7 References 283 In other words. Weinberg.

8. Bjorken and S. Quantum Field Theory. Bernstein. P. J. Itzykson and J-B. Reviews of Modern Physics 46. 1964. John Wliley. D. C. Drell. Introduction to Quantum Field Theory. New York (1969). 1980. McGrawHill. Relativistic Quantum Fields. 7. New York. 6. Zuber. McGrawHill. . J. 7 (1974). Roman.284 7 Complex scalar field theory 5. New York.

In trying to go beyond and study the theory of spin 1 ﬁelds.1 Pauli exclusion principle So far we have discussed ﬁeld theories which describe spin zero bosonic particles. 2 we note that spin 1 particles such as electrons are fermions. For example. a† F + + = a† F 2 + a† F 2 = 0.) Thus. Unlike 2 Bose particles. a† F F aF . in dealing with such systems. The number operator F for the system is given as usual by NF = a† aF .2) = aF a† + a† aF = 1.Chapter 8 Dirac ﬁeld theory 8. (8.1) It is easy to show that we can assign Fermi-Dirac statistics to such an oscillator (and. F F 285 . incorporate the Pauli principle) by requiring that the creation and the annihilation operators satisfy anticommutation relations as opposed to the conventional commutation relations for the bosonic oscillator. therefore. F (8. fermions obey the Pauli exclusion principle which simply says that there can at the most be one fermion in a given state. if we require [aF . Let us consider an oscillator described by the annihilation and the creation operators aF and a† respectively. aF ]+ = a2 + a2 = 0. F F a† . (There can be more fermions only if they are non-identical. we have to ﬁnd a mechanism for incorporating the Pauli principle into our theory.

2 Quantization of the Dirac ﬁeld As we have seen in (1.6) where ψα (x). if this system is analyzed in more detail. has the form (iγ µ ∂µ − m) ψ = 0. 1 2 (8.3) = a† 1 − a† aF aF = a† aF = NF . This equation is manifestly Lorentz covariant and the adjoint equation has the form . can only be nF = 0.2) automatically lead to NF (NF − 1) = 0. what the Pauli exclusion principle would say. the Dirac equation for a massive spin particle. (8. therefore. the wave function of the system will have the necessary antisymmetry property characteristic of fermions.5) This is.4) which shows that the eigenvalues of the number operator. of course.286 then. In fact. 1. 3. Namely. we obtain 2 NF = a† aF a† aF F F 8 Dirac field theory = a† F aF . It is clear. 2. the anti-commutation relations in (8. in such a theory. 4. there can be at the most one quantum in a given state. 8. α = 1. that in dealing with fermionic ﬁeld operators. F F F In other words. a† F + − a† aF aF F (8.96). it leads to the fact that with the anti-commutation relations. (8. is a four component complex spinor function. we will be naturally dealing with anti-commuting variables (these variables are commonly known as Grassmann variables) and anti-commutation relations.

8. ∂ψ ∂∂µ ψ (8.10) .9) give ∂L = (iγ µ ∂µ − m) ψ = 0.41)) ψ(x) = ψ † (x)γ 0 . However. however. let us note that since ψ and ψ are anti-commuting (Grassmann) variables. in this case. consequently. Clearly.9) is manifestly Lorentz invariant (recall the transformation of the Dirac bilinears under a Lorentz transformation discussed in section 3.61)) describing fermionic particles and. we will always use the convention of taking fermionic (Grassmann) derivatives from the left. With this. the 1 Dirac ﬁeld operators are expected to belong to the (reducible) spin 2 representation of the Lorentz group (see (4. the Euler-Lagrange equations following from the Lagrangian density (8.2 Quantization of the Dirac field 287 ← − ψ iγ µ ∂µ + m = 0. In general. In deriving the Euler-Lagrange equations.7) as the Euler-Lagrange equations. need to be treated as anti-commuting operators. we would like to decide on a Lorentz invariant Lagrangian density which would give rise to the two Dirac equations (8. In dealing with the Dirac ﬁeld theory. where the adjoint spinor is deﬁned to be (see also (2.3). (8.6) and (8. the Lagrangian density of the form / L = ψ (iγ µ ∂µ − m) ψ = iψ∂ ψ − mψψ. we note that we can treat ψ(x) and ψ(x) as independent ﬁeld operators as in the case of the complex Klein-Gordon ﬁeld theory. (8.8) In trying to second quantize the Dirac equation (theory). we have to be careful about signs that may arise in taking derivative operators (with respect to these variables) past other Grassmann variables. ∂ψ ← − ∂L ∂L − ∂µ = mψ − ∂µ (−iψγ µ ) = ψ iγ µ ∂µ + m = 0.7) (8. as we have seen in chapter 3 as well as in chapter 4.

9) is Lorentz invariant and gives the two Dirac equations as the Euler-Lagrange equations.83).9) only by a total divergence.12) However. It is worth noting here that even though the Lagragian density (8.6) and (8.11) † where we have used (2. (8. the action associated with it is since a total divergence does not contribute to the action.8) as well as the fact that γ 0 is Hermitian. 2 .9) is not Hermitian. it is not Hermitian. Hermitian and which can also be checked to give the two Dirac equations as the Euler-Lagrange equations has the form ← − 1 1 ψ (iγ µ ∂µ − m) ψ − ψ iγ µ ∂µ + m ψ. L′ = 1 1 1 ψ (iγ µ ∂µ − m) ψ − ∂µ iψγ µ ψ + ψ (iγ µ ∂µ − m) ψ 2 2 2 i = ψ (iγ µ ∂µ − m) ψ − ∂µ ψγ µ ψ 2 i (8.288 8 Dirac field theory which are exactly the two Dirac equations (8. Even though the Lagrangian density (8. An alternate Lagrangian density which is Lorentz invariant. Namely.12) diﬀers from L in (8. note that L† = ψ (iγ µ ∂µ − m) ψ ← − † = ψ † − iγ µ † ∂µ − m ψ ← − = ψ † − iγ µ † ∂µ − m γ 0 ψ ← − = ψ − iγ µ ∂µ − m ψ = ψ (iγ µ ∂µ − m) ψ − i∂µ ψγ µ ψ = L. we note that the Lagrangian density in (8. 2 2 L′ = (8.7) that we are interested in.13) = L − ∂µ ψγ µ ψ . (8. In fact.

ψβ (y) + + = 0. = 0. ˙ ∂ψ (8. the second and the third relations in (8. the equal time canonical anti-commutation relations for the ﬁeld variables and their conjugate momenta take the forms (we do not write the equal time condition explicitly for simplicity.15) can also be written as (Πψ )α (x).14) Πψ = Πψ = Consequently. (Πψ )β (y)]+ = −iδαβ δ3 (x − y).15) is a reﬂection of our choice of left derivatives. −iψβ (y) + ψα (x).2 Quantization of the Dirac field 289 For this reason as well as for simplicity. (8. ˙ ∂ψ ∂L = 0.8. However. (Πψ )β (y) or. ψβ (y) + = −iδαβ δ3 (x − y).15) and (8. in the presence of gravitation. it is more appropriate to use the Lagrangian density in (8.9) to describe the Dirac theory (equations). † ψα (x). (Πψ )β (y)]+ . . but we should understand that x0 = y 0 in all the relations in (8.14).16)) [ψα (x).16) = δαβ δ3 (x − y). † † ψα (x). (Πψ )β (y) or. we use L in (8.27)) for the theory. we can deﬁne momenta conjugate to the ﬁeld variables ψ and ψ as (we would be careful with taking derivatives from the left which is our convention for fermionic derivatives) ∂L = −iψγ 0 = −iψ † . + † = ψα (x). [ψα (x). ψβ (y)]+ = 0 = [(Πψ )α (x).15) The negative sign in the last relation in (8. Using the ﬁrst relation in (8.9) to describe the dynamics of the system. (8. If we work with the Lagrangian density (8.12) which naturally leads to a symmetric stress tensor (see (6.

) ˙ H = −Πψ ψ − L ˙ ˙ = − −iψ † ψ − iψ † ψ − iψγ · ∇ψ + mψψ = −iψγ · ∇ψ + mψψ. For example.17) Using the ﬁeld anti-commutation relations (8. leading to the Hamiltonian H= d3 x H = d3 x −iψγ · ∇ψ + mψψ .16) are all that we need for quantizing the Dirac theory.9) has the form (The negative sign in the ﬁrst term arises from our convention of choosing left derivatives for fermions.16). Such systems can be systematically studied through the method of Dirac brackets and such an analysis leads to the conclusion that the relations (8. H = d3 y † ψα (x). (8.19) .18) (8.) The Hamiltonian density for the Dirac ﬁeld theory (8. ˙ iψα (x) = ψα (x).15) and (8. We will discuss the method of Dirac in more detail in chapter 10.290 8 Dirac field theory (We note that we have not written any anti-commutation relation for ψ α (x) and (Πψ )α (x) simply because the relations in (8. ψβ (y) + −iγ 0 γ · ∇y ψ(y) + β † + m ψα (x). (8. ψβ (y) γ 0 ψ(y) β x0 =y 0 = d3 y δαβ δ3 (x − y) −iγ 0 γ · ∇y ψ(y) + mδαβ δ3 (x − y) γ 0 ψ(y) β α β x0 =y 0 = −i γ 0 γ · ∇ψ(x) α + m γ 0 ψ(x) . it is now easy to check that this Hamiltonian leads to the two Dirac equations as the Hamiltonian equations.15) and (8.14) can be thought of as constraints on the dynamics of the system.

. or. ψβ (y) + x0 =y 0 = d3 y i ψ(y)γ · ∇y β δαβ δ3 (x − y) x0 =y 0 −mψ β (y)δαβ δ3 (x − y) ← − = −i ψ(x)γ · ∇ α − mψ α (x). or.8. H d3 y i ψ(y)γ · ∇y β † ψα (x).20) Similarly. ← − ˙ iψγ 0 = −iψγ · ∇ − mψ. ψβ (y) + † −mψ β (y) ψα (x). or. (iγ µ ∂µ − m) ψ(x) = 0.22) Namely. (8.21) and this can be written in the matrix form as ← − ˙ iψ † = −iψγ · ∇ − mψ. the anti-commutation relations (8.7) as the Hamiltonian (Heisenberg) equations. (8.6) and (8. or.15) and (8.2 Quantization of the Dirac field 291 where we have used the fact that the Hamiltonian is independent of time to identify x0 = y 0 and this equation can be written in the matrix form as ˙ iψ(x) = −iγ 0 γ · ∇ψ(x) + mγ 0 ψ(x). ˙ iγ 0 ψ(x) = −iγ · ∇ψ(x) + mψ(x). ← − ψ iγ µ ∂µ + m = 0.18) indeed reproduce the two Dirac equations (8.16) as well as the Hamiltonian (8. we obtain ˙† iψα (x) = = † ψα (x). (8.

uα (k. s) are anti-commuting operators. / (8. s)uα (k.46).94) and the ﬁeld (8.93) and (3. s) and d† (k. Therefore. s)ψα (x) α (2π)3 k0 d3 x d3 k ′ m ′ √ ei(k−k )·x u† (k.23) × e−ik·x c(k. Taking advantage of our experience with the complex Klein-Gordon ﬁeld theory. s)vα (k′ . s′ ) α ′ . s) and vα (k. s′ ) α 3 (2π) k0 k′ 0 +ei(k+k )·x u† (k. As in the case of the Klein-Gordon ﬁeld (see section 5. s)uα (k′ .4). s) denote the positive and the negative energy spinors deﬁned in (3. where c(k. s) . we write √ (k0 = Ek = k2 + m2 > 0) ψα (x) = 1 s=± 2 d3 k m (2π)3 k0 (8.23) automatically satisﬁes the Dirac equation in this basis (i∂ − m) ψ(x) = 0.5). s) + eik·x d† (k. we can expand the Dirac ﬁeld operator in the basis of these plane wave solutions.24) The speciﬁc form of the prefactor in (8. let us note that the ﬁeld decomposition in (8. s′ ) c(k′ . s′ ) d† (k′ .23) can be inverted to give √ (k0 = Ek = k2 + m2 > 0) d3 x = s′ m eik·x u† (k.292 8. s)vα (k.3 Field decomposition 8 Dirac field theory We have already seen that the positive and the negative energy plane wave spinor solutions of the Dirac equation deﬁne a complete basis for the space of four component spinor functions (see the discussion in section 3.23) has been chosen keeping in mind the massive normalization of the plane wave solutions in (2.

s). s)ψα (x) = d† (k. ψβ (y) † ψα (x). s) = d(k.27).92)). we know the equal time quantization conditions for the ﬁeld operators from (8. = δαβ δ3 (x − y). using the inversion relations (8.16) to be ψα (x). s).28) Therefore. s)vα (k .29) + . s ) = c(k. we can now obtain the anti-commutation relations satisﬁed by the operators c(k. d† (k . In a similar manner.x0 =y 0 +. c† (k.8.26) that m † e−ik·x ψα (x)uα (k. ψβ (y) +.x0 =y 0 = 0.3 Field decomposition 293 δ3 (k − k ) u† (k. c† (k . s ) c(k . s). s ) α (8. s)uα (k . ψβ (y) † † ψα (x).25)-(8. s). d† (k. (8. (2π)3 k0 m † eik·x ψα (x)vα (k.27) c† (k. (2π)3 k0 (8.x0 =y 0 +. s). s ) d† (k .15) and (8.25) and (8. s).26) It follows from (8. s) = d3 x d3 x Furthermore. (8. it can be shown that m † e−ik·x vα (k. k0 m where we have used the massive normalization for the spinors discussed in (2.46) (see also (3. s) and d(k.28) and they correspond to c(k. s). s ) + . (2π)3 k0 d3 x (8. = 0. s).25) m k0 δss c(k. s ) α 0 x0 = s d3 k √ m k0 k 0 + δ3 (k + k ) e2ik = s u† (k. s) in a straightforward manner from the quantization conditions (8. s ) = δss δ3 (k − k ) = d(k.

s) + eik·x d† (k. s) = (γ 0 k0 − γ · k + m)v(k. s) as well as d(k. This shows that we can think of c(k. s′ )e−ik ·x × e−ik·x c(k. s)u(k. s) . / we can simplify the Hamiltonian in (8. s)(γ · k + m)u(k.s′ =± 2 ′ d3 x ψ (−iγ · ∇ + m) ψ m k0 ′ m 3 d k k′ 0 × u(k′ . d† (k.18) of the theory. s′ c† (k′ s′ )eik ·x + v k′ .31) . s′ )eik ·x + v(k′ . s) + eik·x d† (k. s). s) = (γ 0 k0 − γ · k − m)u(k. s). c† (k. s′ )d(k′ .30) (k + m) v(k. / (8. The Hamiltonian (8.93) and (3. in this case. s′ d(k′ . s) = 0. s)v(k. s) as annihilation and creation operators for the two kinds of fermionic quanta in this theory (much like the complex Klein-Gordon ﬁeld theory).s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m k0 ′ × u k′ . s)(−γ · k + m)v(k. s′ )e−ik ·x × (−iγ · ∇ + m) × e−ik·x c(k.294 8 Dirac field theory with all other anti-commutators vanishing. can be written out in terms of creation and annihilation operators as H= = d3 x −iψγ · ∇ψ + mψψ = d3 x (2π)3 d3 k ′ 1 s.30) and write (k − m) u(k. s) = 0. s) = 1 s. If we now use the relations (3. s′ )c† (k′ .94) (8.

s) . s) (8. s)c† (k′ . s′ u(k. s′ v(k.46) (see also (3. s)d† (k. s′ )e−ik ·x ×γ 0 e−ik·x c(k. s) − eik·x d† (k. s) )x0 v k′ . s′ )d† (k. s)u(k. s′ )d† (k. s)d(k.s′ =± 2 ′ 0 +k 0 ′ 0 +k 0 )x0 u k′ . s′ )d† (k. s′ )c(k. s)d(k′ .s′ =± 2 d3 kd3 k′ m k′ 0 m 0 k k0 δ3 (k − k′ ) u k′ . s) −δ3 (k + k′ ) ei(k − e−i(k = 1 s. s′ γ 0 v(k. −k. s) Ek m δss′ c† (k. s)c† (−k. s)c† (k′ . We are yet to normal order the Hamiltonian. s′ γ 0 u(k.92)). s) −e2ik 0 x0 u† k0 .32) + e−2ik = 0 x0 m d3 k 1 s.3 Field decomposition 295 H= 1 s. s) − v k′ . s′ u(k. s′ γ 0 v(k. s′ )c(k. −k. s′ γ 0 u(k. s′ )c(k. s) v † k0 . s) − d(k. s) − d(k.8. where we have used the orthogonality relations for the plane wave solutions following from (2. The normal ordering for products of fermionic operators is deﬁned exactly the same . s) d3 k u† k. s′ d(k′ . s′ )eik ·x + v k′ . s)c(k.s′ =± 2 = 1 s=± 2 d3 k Ek c† (k. s) m − v † k. s′ )d† (k. s′ )c(k. s′ c† (k′ . s′ )d† (k.s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m 0 k k0 ′ × u k′ . s′ )c(k. s)v(k. s) = 1 s. s)d(−k. s)d(k′ . s′ v(k. s)c† (k.

s′ ). Using (8. the normal ordered Hamiltonian for the free Dirac ﬁeld theory in (8. (8. However. s′ ). s) and d(k. s) + d† (k. we pick up a negative sign since they anti-commute (as opposed to bosonic variables). s) : = −d† (k. s)|0 d(k. s)d(k′ .33).34) even though we do not put the normal ordering symbol explicitly. s). s .35) c(k. namely. = 0. s)c(k. s)d(k′ . s |k. = d† (k. For example.22)) = c† (k. s)d(k. (8.33) so that the order of factors inside a product that is normal ordered becomes important. s) (The Hamiltonian is understood to be normal ordered as in (8.32) is obtained to have the form : H := s=± 1 2 d3 k Ek c† (k. s)d(k′ . s)|0 H|0 The one particle states of the theory can now be deﬁned as (see (7. s)|0 . s′ ) : = d† (k.36) |k. : d(k′ . = 0 = 0|d† (k. we now have : d† (k. (8. we write products with the creation operators standing to the left of the annihilation operators.34) As in the case of the complex (Klein-Gordon) scalar ﬁeld theory (see (7. s). s)|0 . s′ )d† (k. s) . (8. every time we move a fermionic operator past another in order to bring the product into its normal ordered form.) = 0 = 0|c† (k.296 8 Dirac field theory way as for the bosonic operators. we can deﬁne the vacuum state of the theory to correspond to the state which is annihilated by both c(k.21)).

the Lagrangian density (8. as in the case of the complex scalar ﬁeld theory. transforms as L(ψ. ′ (8. Therefore.40) .37) or the inﬁnitesimal form of it (ǫ is the constant inﬁnitesimal parameter of transformation) δǫ ψ = ψ ′ (x) − ψ(x) = −iǫψ(x).4 Charge operator The Dirac Lagrangian density (8.9) L = ψ (iγ µ ∂µ − m) ψ. ψ ) = ψ (iγ µ ∂µ − m) ψ ′ ′ ′ ′ (8. ψ) → L(ψ ′ .8. ψ). 8. (8.12)). δǫ ψ = ψ (x) − ψ(x) = iǫψ(x).39) = ψeiθ (iγ µ ∂µ − m) e−iθ ψ = ψ (iγ µ ∂µ − m) ψ = L(ψ.4 Charge operator 297 Both these states can be checked to be eigenstates of the Hamil√ tonian (8.34) with energy eigenvalue k0 = Ek = k2 + m2 > 0. We note that under the global phase transformation (α is the constant parameter of transformation) ψ(x) → ψ ′ (x) = e−iθ ψ(x). like the Lagrangian density for the complex Klein-Gordon ﬁeld theory can be seen to be invariant under a global U (1) phase transformation. we need to analyze the symmetry properties of the theory.9) (or (8. ψ(x) → ψ (x) = ψ(x)eiθ .38) (8. there are two distinct one particle states with degenerate mass (energy) and to understand the meaning of these states.

Thus.37) (or (8.13) and (7.27) and we note that the form of the expression in (8. (8.26). we note that K µ = 0.37) (or (8.7) that this current is conserved.43) where J µ (x) = ψ(x)γ µ ψ(x). (8.42) δǫ ψ α = (−iǫψ)α (−iψγ µ )α = ǫψ(x)γ µ ψ(x). we obtain (see (6.42) is consistent with the convention of taking left derivatives for the fermion ﬁelds) ∂L ∂L + δǫ ψα ∂∂µ ψα ∂∂µ ψ α (8. the Lagrangian density (8. the N¨ther current associated with the internal symmetry o transformation (8. It is a vector current as in the case of the complex Klein-Gordon ﬁeld theory (see (7.41) On the other hand.37) of the system is given by (note that in the ﬁrst quantized theory. this would correspond to the total probability) .13) to be µ Jǫ (x) = ǫψ(x)γ µ ψ(x) = ǫJ µ (x). as in the case of the complex scalar ﬁeld theory in (7.9) is invariant under the global phase transformation (8. ∂µ J µ (x) = 0.298 8 Dirac field theory Namely. using the inﬁnitesimal transformation (8.38).45) The conserved charge associated with the symmetry transformation (8.6) and (8.44) represents the current independent of the parameter of transformation. where we have rearranged the fermionic operators keeping in mind their anti-commuting nature. In this case. (8.29)) and it can be checked using the equations of motion (8.38)) is obtained from (6.38)). (8.

. s′ )c(k′ . s′ ) |0 = 1 s′ =± 2 d3 k′ c† (k′ . lepton number.23). s′ ) c(k′ . s . we can express (8. since we are dealing with a fermion system. s) . s) − d† (k. Therefore. But. c† (k. etc) which are associated with U (1) symmetry transformations as well. s′ )d(k′ . s′ ) c† (k.8.33) and (7. (8.47) As in the case of the complex Klein-Gordon ﬁeld. s′ )δss′ δ3 (k′ − k)|0 = c† (k. s)|0 = |k. we can now show that the charge operator (normal ordered) satisﬁes (see (7. s)|0 = 1 s′ =± 2 d3 k′ c† (k′ .37) (or (8.46) We note here that the charge (8. as in the case of the complex Klein-Gordon ﬁeld we can associate the charge operator with that of the electric charge. s) + − c† (k.38)). s′ ) − d† (k′ .34)) Q|0 Q|k.46) in terms of creation and annihilation operators as : Q := s=± 1 2 d3 k c† (k. Using the ﬁeld decomposition in (8. (8. s = 0. it is also possible to associate the charge operator with a fermion number (such as baryon number. s)c(k′ . s)d(k. s′ ). = 1 s′ =± 2 d3 k′ c† (k′ .4 Charge operator 299 Q= d3 x J 0 (x) = d3 x ψ † (x)ψ(x). s)c(k.46) is associated with an Abelian symmetry transformation (8.

d† (k. the vacuum does not contain any particle as we would intuitively expect (compare this with the ﬁrst quantized description of the vacuum in section 2. s′ ).49) namely. 8. s)|0 = − d3 k′ d† (k′ . the vacuum state of the theory is charge neutral while the two distinct one particle states. s). s = 1 s′ =± 2 8 Dirac field theory d3 k′ c† (k′ . in the second quantized description of the Dirac theory. which are degenerate in mass and are created by the operators c† (k. the second quantized description of the Dirac theory is symmetric under the interchange of particles and anti-particles. s′ ) 1 s′ =± 2 d(k′ .34) is symmetric under the discrete transformation (see also (7.5) and that the (normal ordered) Hamiltonian in (8. s) ↔ d(k. as we have also seen in the case of the complex Klein-Gordon ﬁeld theory. s′ )c(k′ . s′ ) |0 = − d3 k′ d† (k′ . s . we can deﬁne the Green’s function for the Dirac ﬁeld theory as satisfying the Dirac equation with a delta function potential (source) as . s) respectively. (8. s′ )δss′ δ3 (k − k′ )|0 (8. This shows that. s) and d† (k. s′ )d(k′ . we can identify them with the particle and the anti-particle states of the theory with positive energy. Namely. s)d(k′ . carry opposite charge.300 Q|k. Thus. s) + − d† (k.48) 1 s′ =± 2 = −d† (k. s′ ) − d† (k′ .37)) c(k. s)|0 = −|k. s′ ) d† (k.5 Green’s functions As we have discussed within the context of the Klein-Gordon theory.

advanced. / so that (8. / (8. we do not have to calculate the Green’s function for the Dirac ﬁeld theory separately. From these then.117) we know that the Green’s function for the Klein-Gordon equation satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y).53) (i∂ − m) S(x − y) = (i∂ − m) (i∂ + m) G(x − y) / / / = − ∂µ ∂ µ + m2 G(x − y) = δ4 (x − y).8. The right-hand side is also a matrix.γ 2 = −γ µ ∂µ γ ν ∂ν − m2 = −γ µ γ ν ∂µ ∂ν − m2 ∂ ∂ + µ ν − m2 = −η µν ∂µ ∂ν − m2 (8. but it is the trivial 4 × 4 identity matrix (which we do not write explicitly). Feynman.54) Therefore. we would concentrate mainly on the Feynman Green’s function since that is the most useful . Furthermore. Although the relation (8. S(x − y) is a 4 × 4 matrix function in the spinor space. from (5.50) Here.5 Green’s functions 301 (i∂ − m) S(x − y) = δ4 (x − y).52) S(x − y) = (i∂ + m) G(x − y).51) = − ∂µ ∂ µ + m2 . etc. we can identify (8. Let us recall the identity (i∂ − m) (i∂ + m) = (i∂ )2 − m2 / / / = − 1 µ ν γ .).53) holds for any Green’s function of the theory (retarded. (8.

145) √ where k0 = Ek = k2 + m2 > 0) S (+) (x) = (i∂ + m) G(+) (x) / = (i∂ + m) / = i 2 i 2 d3 k 1 −ik·x e (2π)3 k0 d3 k k + m −ik·x / e .56) SF (k) = lim ǫ→0+ (8. k − m + iǫ / ǫ→0+ (8. 3 (2π) k0 (8.143) and (5.58) We can express the Feynman Green’s function in terms of the positive and the negative energy Green’s functions as well (see (5.140)) k+m / .53) we can also obtain the positive and the negative energy Green’s functions for the Dirac ﬁeld theory as (see (5.53) can be written in the momentum space as S(k) = (k + m) G(k). 3 (2π) k0 S (−) (x) = (i∂ + m) G(−) (x) / = (i∂ + m) − / = − = i 2 i 2 i 2 d3 k 1 ik·x e (2π)3 k0 / d3 k (−k + m) ik·x e 3 (2π) k0 / d3 k (k − m) ik·x e .55) SF (k) = (k + m) GF (k) = lim / which can also be written as 1 . We note that the relation (8. for example. / so that we can write.302 8 Dirac field theory in the calculation of S-matrix elements.57) From (8. (see (5. k2 − m2 + iǫ (8.146)) .

8) also as ψ α (x) = ψ α + ψ α (x). s)uα (k. s)v α (k. (8.6 Covariant anti-commutation relations Let us decompose the ﬁeld operator into its positive and negative energy (frequency) parts as (+) (−) ψα (x) = ψα (x) + ψα (x). s)uα (k.8. s). s). (2π)3 k0 m eik·x c† (k. Similarly. (2π)3 k0 m eik·x d† (k.6 Covariant anti-commutation relations 303 SF (x) = (i∂ + m) GF (x) / = (i∂ + m) −θ x0 G(+) (x) + θ −x0 G(−) (x) / = −θ x0 S (+) (x) + θ −x0 S (−) (x).62) (+) d3 k m e−ik·x d(k.59) where we have used the fact that the terms involving derivatives of the step function cancel out (namely. (8. δ(x0 )(G(+) (x) + G(−) (x)) = 0). s)vα (k.63) (2π)3 k0 ψ α (x) = 1 s=± 2 (−) d3 k . s). we can decompose the adjoint ﬁeld operator (8. (8. where ψ α (x) = 1 s=± 2 (+) (−) (8. (8. 8.60) where (+) ψα (x) = 1 s=± 2 d3 k m e−ik·x c(k.61) (2π)3 k0 (−) ψα (x) = 1 s=± 2 d3 k Similarly. s). other Green’s functions for the Dirac ﬁeld theory can also be easily obtained from the corresponding functions for the KleinGordon ﬁeld theory.

ψ β (y) (+) + = s.29) that the only nontrivial anti-commutation relations among these four ﬁeld components will have the form (these are not at equal times) d3 kd3 k′ (2π)3 (+) ψα (x). s)uβ k′ . s) and d† (k. s)uβ (k.58).s′ =± 1 2 m2 k0 k′ 0 + × eik·x−ik ·y vα (k.s′ =± 1 2 ′ c(k. d(k. s′ δss′ δ3 (k − k′ ) = 1 s=± 2 d3 k m −ik·(x−y) e uα (k. ψ β (y) (−) + ′ = s. s) in (8.s′ =± 1 2 m2 k0 k′ 0 + × e−ik·x+ik ·y uα (k. s).106) as well as the identiﬁcation in (8. s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 × e−ik·x+ik ·y uα (k. s). s) (2π)3 k0 = = 1 2 / d3 k m (k + m)αβ −ik·(x−y) e 3 k0 (2π) 2m / d3 k (k + m)αβ −ik·(x−y) e 3 (2π) k0 (+) = −iSαβ (x − y). s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 . d(k′ . c† (k. (8. Similarly. s).64) where we have used the completeness relation in (3.s′ =± 2 ′ d† (k. s)v β k′ . s). c† (k′ .304 8 Dirac field theory It is clear from the basic anti-commutation relations for the creation and the annihilation operators c(k. we have d3 kd3 k′ (2π)3 (−) ψα (x). s)uβ k′ . s′ = s. s′ = 1 s.

s)v β k′ . αβ (8.65) ψα (x). in rearranging terms to bring them to a particular form.136) (see also (5. s) (2π)3 k0 = = 1 2 / d3 k m (k − m)αβ ik·(x−y) e 3 k0 (2π) 2m / d3 k (k − m)αβ ik·(x−y) e (2π)3 k0 (−) = −iSαβ (x − y). by deﬁnition the normal ordered product of two ﬁeld operators yields (note from (8.9. since fermionic ﬁeld operators anti-commute (Grassmann variables).8.66) which is the analogue of the Schwinger function (5.7 Normal ordered and time ordered products The deﬁnitions of normal ordered and time ordered products for Dirac ﬁeld operators are still the same as discussed in sections 6.109).6. s′ δss′ δ3 (k − k′ ) = 1 s=± 2 ′ d3 k m ik·(x−y) e vα (k.63) that the positive and the negative energy parts of the ﬁeld operators contain annihilation and creation operators respectively) . s)v β (k. However. Thus. we can write (+) = ψα (x). As a result. where we have used (3. ψ β (y) + + (−) + ψα (x).61) and (8. ψ β (y) (−) (+) + = −iSαβ (x − y) − iSαβ (x − y) = −i S (+) (x − y) + S (−) (x − y) = −iSαβ (x − y).160)) for the Dirac ﬁeld theory and these relations are analogous to the covariant commutation relations for the Klein-Gordon ﬁeld operators discussed in section 5.5 and 6.7 Normal ordered and time ordered products 305 × eik·x−ik ·y vα (k. ψ β (y) (+) (−) (8. 8. we need to be careful about negative signs that can arise from changing the order of any two such operators.

From the deﬁnition in (8. here the order of factors inside normal ordering can lead to additional signs. (+) (8. it is easy to show that ψ β (y)ψα (x) = : ψ β (y)ψα (x) : −iSαβ (x − y).70) (8. (+) (−) (+) (8.69) (8. Similarly. In other words. unlike the bosonic case. (+) (+) (+) (+) (+) (8.67) and we note that.306 8 Dirac field theory (−) (−) : ψα (x)ψβ (y) : = ψα (x)ψβ (y) (−) (−) : ψβ (y)ψα (x) : = −ψα (x)ψβ (y) (−) = − : ψα (x)ψ β (y) :.67).68) where in the last step we have made the identiﬁcation in (8.71) . using (8. it is clear that (+) (−) : ψα (x)ψ β (y) : = : ψα (x) + ψα (x) ψ β (y) + ψ β (y) : (+) (+) = : ψα (x)ψ β (y) + ψα (x)ψ β (y) (−) (−) + ψα (x)ψ β (y) + ψα (x)ψ β (y) : (+) (+) = ψα (x)ψ β (y) − ψ β (y)ψα (x) (−) (−) + ψα (x)ψ β (y) + ψα (x)ψ β (y) (+) = ψα (x)ψ β (y) − ψα (x).68) as ψα (x)ψ β (y) = : ψα (x)ψ β (y) : −iSαβ (x − y).65). ψ β (y) (−) + (+) (−) (+) (−) (+) (−) (+) (−) (+) (−) = ψα (x)ψ β (y) + iSαβ (x − y). As a result. we see that we can denote 0|ψα (x)ψ β (y)|0 = −iSαβ (x − y) = ψα (x)ψ β (y). we can write (8.64).

.72) This can be expressed in terms of the normal ordered products as T ψα (x)ψ β (y) = θ(x0 − y 0 ) : ψα (x)ψ β (y) : −iSαβ (x − y) −θ(y 0 − x0 ) : ψ β (y)ψα (x) : −iSαβ (x − y) = θ(x0 − y 0 ) : ψα (x)ψ β (y) : + θ(y 0 − x0 ) : ψα (x)ψ β (y) : +i − θ(x0 − y 0 )Sαβ (x − y) + θ(y 0 − x0 )Sαβ (x − y) = = θ(x0 − y 0 ) + θ(y 0 − x0 ) : ψα (x)ψ β (y) : + iSF .59) and which can be compared with (6.αβ (x − y) = ψα (x)ψ β (y). It follows now from (8.81). 0|T ψα (x)ψ β (y) |0 = iSF .αβ (x − y).6. The Wick’s theorem for normal ordered products of Dirac ﬁelds can now be developed in a way completely analogous to the discussions in section 6. (8.74) Once again. (8.αβ (x − y) : ψα (x)ψ β (y) : + iSF . (8. we can develop Wick’s theorem for time ordered products of Dirac ﬁelds in a straightforward manner as discussed in section 6. The time ordered product of Dirac ﬁeld operators is again deﬁned with time ordering of the operators from left to right (largest time on the left) keeping in mind the appropriate negative signs associated with rearranging fermionic operators as T (ψα (x)ψ β (y)) = θ(x0 − y 0 )ψα (x)ψ β (y) − θ(y 0 − x0 )ψ β (y)ψα (x).73) (+) (−) (−) (+) where we have used the deﬁnition in (8.7 Normal ordered and time ordered products 307 which can be compared with (6.5 which we will not repeat here.8.73) that the Feynman Green’s function is related to the vacuum expectation value of the time ordered product of ψα (x) and ψ β (y).104).

in this section we will discuss brieﬂy the quantization of such ﬁelds. Therefore. when the Dirac particle is massless.54) which is appropriate for massless spinors. 2 1 (½ − γ5 ).7. it is most natural to decompose it into its chirality states. Unlike in (8. Furthermore. The discussion for the left-handed ﬁeld can be carried out in an analogous manner. we can expand the ﬁeld as (see (8.170) we can invert the decomposition in (8.53) and (2. Using the orthonormality relations for the right-handed spinors described in (3. we note here that there is no sum over spin polarization because the chiral spinors are charaterized by their unique chirality or helicity (see.7. Correspondingly. 2 ψR (x) = ψL (x) = (8. for example. for example. (8.76) Such massless left-handed and right-handed Dirac ﬁelds (Weyl ﬁelds) arise naturally in physical theories (as we will see.77) where. the multiplicative factors in (8.164)).3).75) As a result.23)) d3 k (2π)3 |k| ψR (x) = e−ik·x c(k)uR (k) + eik·x d† (k)vR (k) . the discussion following (3. (8. as before.77) to obtain (here k0 = |k|) . in the case of the standard model in 14. we can also decompose a Dirac ﬁeld into a left-handed and a right-handed component as 1 (½ + γ5 )ψ(x). we have identiﬁed k0 = Ek = |k|. Let us consider the ﬁeld quantization of a right-handed Dirac ﬁeld (Weyl ﬁeld).308 8.23) because we are using here the normalization (2. Using the deﬁnition of the chiral spinors in section 3.23).8 Massless Dirac ﬁelds 8 Dirac field theory As we have seen in section 3.77) is diﬀerent from those in (8. the free Dirac Lagrangian density for a massless spinor ﬁeld naturally decomposes as / / / L = iψ∂ ψ = iψ R ∂ ψR + iψ L ∂ ψL .

c† (k′ ) = δ3 (k − k′ ) = d(k). (8. (8. .80) The Hilbert space of the theory can now be built with the vacuum deﬁned as satisfying c(k)|0 = 0 = d(k)|0 . for example. (8.8 Massless Dirac fields 309 d3 x (2π)3 |k| = eik·x u† (k)ψR (x) R 1 |k|k′ | ′ d3 k ′ d3 x (2π)3 ei(k−k )·x c(k′ )u† (k)uR (k′ ) R ′ +ei(k+k )·x d† (k′ )u† (k)vR (k′ ) R = d3 k ′ δ3 (k − k′ )c(k′ )u† (k)uR (k′ ) R ′| |k||k +δ3 (k + k′ ) e2i|k|x d† (k′ )u† (k)vR (k′ ) R = c(k).76)).79) As a result. it can be shown that the nontrivial anti-commutation relations between the creation and the annihilation operators take the form c(k). (8. d† (k′ ) + + . 3 |k| (2π) (8.81) and the higher particle states created by applying the creation operators and other quantities of interest can be derived in the standard manner. from the equal-time anti-commutation relations for the ﬁeld variables following from the Lagrangian density for the righthanded spinor ﬁeld (see.78) 0 Similarly. it can also be shown that (it is assumed that k0 = |k|) d† (k) = d3 x † e−ik·x vR (k)ψR (x).8.

74)) iSF .80) in evaluating the vacuum expectation values. for example. (8.Rαβ (x − y) = d3 kd3 k (2π)3 |k||k | × θ(x0 − y 0 )e−ik·x+ik ·y uRα (k)u† β (k ) 0|c(k)c† (k )|0 R † −θ(y 0 − x0 )eik·x−ik ·y vRα (k)vRβ (k ) 0|d(k )d† (k)|0 = d3 k e−ik·(x−y) θ(x0 − y 0 )uRα (k)u† β (k) R (2π)3 |k| † −eik·(x−y) θ(y 0 − x0 )vRα (k)vRβ (k) . (8.Rαβ (x − y) = † 0|T ψRα (x)ψRβ (y) |0 † = θ(x0 − y 0 ) 0|ψRα (x)ψRβ (y)|0 † −θ(y 0 − x0 ) 0|ψRβ (y)ψRα (x)|0 .83) Here we have used (8.84) . (8.172) this can also be written as d3 k 1 iSF .81) we obtain (we assume k0 = |k| in the ﬁeld decomposition) iSF . (8. Dropping the spinor indices and using (3. let us determine the propagator for a righthanded fermion.77) and using the properties of the vacuum in (8.310 8 Dirac field theory As an example. We recall that the Feynman propagator is deﬁned as the time ordered product (see.R (x − y) = (2π)3 4|k| × (|k|γ 0 − γ · k)θ(x0 − y 0 )e−i|k|(x −(|k|γ 0 − γ · k)θ(y 0 − x0 )ei|k|(x = 0 −y 0 )+ik·(x−y) 0 −y 0 )−ik·(x−y) γ 0 (½ + γ5 ) d3 k eik·(x−y) 0 0 (|k|γ 0 − γ · k)θ(x0 − y 0 )e−i|k|(x −y ) 3 (2π) 4|k| −(|k|γ 0 + γ · k)θ(y 0 − x0 )ei|k|(x 0 −y 0 ) γ 0 (½ + γ5 ).82) Putting in the ﬁeld decomposition in (8.

84) separately.8 Massless Dirac fields 311 Let us evaluate the two terms in (8. 2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 (8. The second term in the integrand can also be simpliﬁed in a similar manner − 0 0 1 θ(y 0 − x0 )(|k|γ 0 + γ · k)ei|k|(x −y ) 4|k| = = = = dk0 (|k|γ 0 + γ · k) ei(k +|k|)(x 2πi 4|k| k0 + i 0 0 0 0 −y 0 ) dk0 (|k|γ 0 + γ · k) eik (x −y ) 2πi 4|k| k0 − |k| + i dk0 (k0 γ 0 + γ · k) eik (x −y ) 2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 dk0 (−k0 γ 0 + γ · k) e−ik (x −y ) 2πi 4k0 k0 + |k| − i 0 . Using the integral representation for the theta function (see (6.100) and the limit → 0 is to be understood) we can write the ﬁrst term in the integrand as 1 0 0 θ(x0 − y 0 )(|k|γ 0 − γ · k)e−i|k|(x −y ) 4|k| = − = − = − = − dk0 (|k|γ 0 − γ · k) e−i(k +|k|)(x 2πi 4|k| k0 + i 0 0 0 0 0 −y 0 ) dk0 (|k|γ 0 − γ · k) e−ik (x −y ) 2πi 4|k| k0 − |k| + i dk0 (k0 γ 0 − γ · k) e−ik (x −y ) 2πi 4k0 k0 − |k| + i dk0 k e−ik (x −y ) / .8.85) Here we have used the fact that the integral has contributions only when k0 = |k| and correspondingly have rewritten some of the terms.

for example.88) by simply multiplying a factor of γ 0 on the right which leads to iS F .85) and (8.84) we obtain d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) / e 4 (2π) 4k0 × = k0 1 1 + 0 − |k| + iǫ k + |k| − iǫ (8. 2(k2 + iǫ) k + iǫ iSF . 2(k2 + iǫ) k + iǫ (8.R (k) = iSF . 4 (2π) 2(k2 + iǫ) It follows now that the fermion propagator in the momentum space has the form ikγ 0 PR / ikγ 0 (½ + γ5 ) / = 2 . 2πi 4k0 k0 + |k| − iǫ 0 0 0 8 Dirac field theory (8.R (x − y) = / d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) e . However.89) We note that although our discussion has been within the context of right-handed fermion ﬁelds.74)).87) iSF . 8.88) We note here that we have deﬁned the propagator in (8.312 = − / dk0 k e−ik (x −y ) .R (k)γ 0 = ik(½ − γ5 ) / ikPL / = 2 . (8. The reason for this is that the adjoint ﬁeld ψ R has negative chirality. everything carries over in a straightforward manner to left-handed fermion ﬁelds.86) Therefore. adding the two results in (8.9 Yukawa interaction Let us next consider the theory which describes the interaction between a Dirac ﬁeld and a charge neutral Klein-Gordon ﬁeld.86) and substituting into (8.R (k) = (8. For . the conventional covariant propagator (deﬁned as the time ordered product of ψR ψ R ) can be obtained from (8.82) as † the time ordered product of ψR ψR and not as the conventional time ordered product of ψR ψ R (see.

However. the correct parity invariant trilinear interaction for a pseudoscalar meson should be (−gψγ5 ψφ). since here we are only interested in developing calculational methods in a quantum ﬁeld theory.92) We note here that the π 0 meson is experimentally known to be a pseudoscalar meson (changes sign under a space reﬂection) which is not reﬂected in our trilinear interaction in (8.90) L = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 4! where we are assuming that the fermionic particle (say the proton) has mass m while the spin zero meson (e.86). Namely. in this case.) In this case. such a theory can represent the interaction between protons (if assumed to be fundamental) and the π 0 -meson as well as the self-interaction of the mesons.. has the form 1 λ M2 2 / φ − g ψψφ − φ4 . L = L0 + LI . we can separate the Lagrangian density (8. L0 = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 λ LI = −g ψψφ − φ4 . 4! (8. As in (6.36). ψψ is invariant under a Lorentz transformation and φ is a scalar ﬁeld) and is known as the Yukawa interaction. The Lorentz invariant Lagrangian density. (8. (The self-interaction can be switched oﬀ at the lowest order by setting λ = 0. where 1 M2 2 / φ .92). since it leads to the Yukawa potential in the Born approximation and the coupling constant g represents the strength of the interaction.91) (8. the π 0 meson) has mass M .8.g.9 Yukawa interaction 313 example. The trilinear coupling gψψφ is manifestly Lorentz invariant (as we have seen in (3. we can denote the interaction Lagrangian density as (remember everything is normal ordered) .90) into a free part and an interaction part. we are going to ignore this aspect of the meson and also switch oﬀ the self-interactions of the meson ﬁeld for simplicity.

94) in terms of normal ordered products as S = ½ − ig − g2 2 d4 x : ψ(x)ψ(x)φ(x) : d4 xd4 x′ : ψ(x)ψ(x)φ(x)ψ(x′ )ψ(x′ )φ(x′ ) : + : ψ(x)ψ(x)ψ(x′ )ψ(x′ ) : φ(x)φ(x′ ) + ψα (x)ψ β (x′ ) : ψ α (x)ψβ (x′ )φ(x)φ(x′ ) : − ψα (x′ )ψ β (x) : ψβ (x)ψ α (x′ )φ(x)φ(x′ ) : + φ(x)φ(x′ ) ψα (x)ψ β (x′ ) : ψ α (x)ψβ (x′ ) : − φ(x)φ(x′ ) ψα (x′ )ψ β (x) : ψβ (x)ψ α (x′ ) : − ψα (x)ψ β (x′ ) ψβ (x′ )ψ α (x) : φ(x)φ(x′ ) : − ψα (x)ψ β (x′ ) ψβ (x′ )ψ α (x) φ(x)φ(x′ ) + ······ . Using Wick’s theorem (see sections 6. in this theory the S-matrix will have the expansion (see (6. (8. the adiabatic switch-oﬀ is assumed) S = T e−i R∞ R −∞ LI = −g : ψψφ := −HI .95) .94) +··· .69). (8.93) dt HI = T e−i R d4 x HI = T e−ig = d4 x :ψψφ: ½ − ig − g2 2 d4 x : ψ(x)ψ(x)φ(x) : d4 xd4 x′ T : ψ(x)ψ(x)φ(x) :: ψ(x′ )ψ(x′ )φ(x′ ) : (8.314 8 Dirac field theory As a result. we can write (8.5 and 6.6).

9 Yukawa interaction 315 To see how perturbation theory works in the interacting quantum ﬁeld theory.96) Each term in this expression leads to a distinct physical process. This can be written out in detail as S (1) = −ig (−) d4 x (+) (−) ψ α (x)ψα (x) − ψα (x)ψ α (x) (−) (+) (+) (+) (−) + ψ α (x)ψα (x) + ψ α (x)ψα (x) φ(+) (x) (+) (−) + φ(−) (x) ψ α (x)ψα (x) − ψα (x)ψ α (x) (+) (−) + ψ α (x)ψα (x) + ψ α (x)ψα (x) . (There can be other nontrivial matrix elements where the initial state has one more proton.98) where the initial state contains a proton with momentum k and spin s. k′ . s′ . (−) (−) (+) (+) (8. let us look at the lowest order (order g) term in the expansion (8. a nontrivial matrix element would result for the case 0|S1 |k.97) Each positive energy operator in (8. let us look at the ﬁrst term in (8.95). (+) (8.97) contains an annihilation operator and. (1) (8. therefore. q . just to get a feeling for what is involved.8. anti-proton and π 0 meson than the ﬁnal state.98) we substitute the ﬁeld decomposition for each of the positive energy ﬁeld operators and write .96) S1 = −ig (1) (+) d4 x ψ α (x)ψα (x)φ(+) (x).) To evaluate the matrix element (8. an anti-proton with momentum k′ and spin s′ and a π 0 meson with momentum q and the ﬁnal state contains no particles. For example. s.

q = −ig d4 x s. k′ .99) can be satisﬁed only if the energy of each of the particles vanishes which is not possible since the particles are massive. s). energy conservation δ(k0 + k′ 0 + q 0 ) in (8. The delta function merely ensures the overall conservation of energy and momentum in the process. s. k′ 0 . an anti-proton along with a π 0 meson are annihilated. our goal in this section has been to describe the calculational methods in quantum ﬁeld theory. (For completeness. s) ˜ ˜ × 0|d(p′ . s′ )uα (p.˜′ =± 1 ˜s 2 ′ (1) d 3 p d 3 p d 3 p′ ˜ (2π)3 2˜ 0 p m (2π)3 p0 m (2π)3 p′ 0 p × e−i(p+p +˜)·x v α (p′ . (2π)3 k′ 0 (8. let us consider the second .95) also leads to a distinct physical process. note that since k0 . q 0 > 0. s)d† (k′ . However. We can similarly associate a physical process with every other term of S (1) in (8.316 8 Dirac field theory 0|S1 |k. s) ˜ ˜ ˜ p × δss δ3 (p − k) δs′ s′ δ3 (p′ − k′ ) δ3 (˜ − q) ˜ ˜ = −(2π)4 ig δ4 (k + k′ + q) × 1 (2π)3 2q 0 m (2π)3 k0 m v α (k′ . s)a(˜ )c† (k. s′ )c(p. In fact.99) represents the process where a proton. s′ )uα (k. we note here that this lowest order matrix element in (8.) Each term at order g2 in the expansion in (8.96).˜′ =± 2 ˜s (2π)3 2˜ 0 p m (2π)3 p0 m (2π)3 p′ 0 × (2π)4 δ4 (p + p′ + p) v α (p′ . The matrix element (8. s′ .99) where we have used the appropriate (anti) commutation relations between the creation and the annihilation operators in evaluating the vacuum expectation value. s′ )uα (p. s′ )a† (q)|0 ˜ ˜ p = −ig d 3 p d 3 p d 3 p′ ˜ 1 s.99) vanishes because energy-momentum conservation cannot be satisﬁed. For example.

k3 . s1 . the matrix element will have the form k4 .120)) d4 q −iq·(x−x ) e GF (q). In such a case. s3 )vα (k4 . s2 (−) (+) (+) = −ig2 × d4 xd4 x GF (x − x ) (−) (+) (+) (−) k4 . s4 . s2 = − ig2 2 d4 xd4 x GF (x − x ) (−) (+) (+) (2) (−) × k4 .95) (2) S2 = − =− g2 2 g2 2 d4 xd4 x φ(x)φ(x ) : ψ(x)ψ(x)ψ(x )ψ(x ) : d4 xd4 x iGF (x−x ) : ψ(x)ψ(x)ψ(x )ψ(x ) : . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x )|k1 .9 Yukawa interaction 317 term in the expansion at order g2 in (8. k3 . s2 = −ig2 × d4 xd4 x GF (x − x ) ei(k3 +k4 )·x e−i(k1 +k2 )·x m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 (8. k2 . Recalling that the Fourier transformation of the Green’s function is given by (see (5. s2 )uβ (k1 . s4 .100) A nontrivial matrix element of this operator would be obtained if the initial state contains a proton and an anti-proton of momenta k1 and k2 respectively and the ﬁnal state also contains a proton and an anti-proton of momenta k3 and k4 respectively. k2 .101) × uα (k3 . (8.102) . s3 |S2 |k1 . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x ) (−) + ψ α (x )ψα (x )ψ β (x)ψβ (x)|k1 . (2π)4 GF x − x = (8. k2 . s1 . k3 . s1 .8. s1 ). s4 )v β (k2 . s4 .

s3 )vα (k4 .101) can be written as (2) 8 Dirac field theory k4 .10 Feynman diagrams It is quite clear from the two simple examples that we have worked out in the last section that the evaluation of the S-matrix elements using the Wick expansion is quite tedious. Instead. s3 )vα (k4 . s3 |S2 |k1 . s1 ) = −(2π)4 g2 δ4 (k1 + k2 − k3 − k4 ) × m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 (8. s4 ) v β (k2 . The delta function merely ensures the overall energy momentum conservation in the process. It is clear from looking at the expansion of the S-matrix that it contains terms . s2 )uβ (k1 . Physically this matrix element corresponds to a proton and an antiproton annihilating to create a π 0 meson which subsequently pair produces a proton and an anti-proton. s1 . k2 . s1 ). k3 .318 the matrix element (8. s1 ) = −ig2 × d4 q (2π)4 δ4 (q − k3 − k4 )δ4 (q − k1 − k2 ) GF (q) m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 × uα (k3 . 8. s2 )uβ (k1 . s4 . s4 )iGF (k1 + k2 )v β (k2 . s2 )uβ (k1 . Feynman developed a graphical method for evaluating these matrix elements which is in one to one correspondence with the Wick expansion. s3 )vα (k4 . s4 )v β (k2 .103) × uα (k3 . s2 = −ig2 × d4 q 4 4 ′ ′ d xd x GF (q)e−i(q−k3 −k4 )·x ei(q−k1 −k2 )·x (2π)4 m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 × uα (k3 .

Let us deﬁne the two kinds of Feynman propagators that are possible for our theory (the Yukawa theory). Therefore.) It should also be remembered that the fermion propagator is a 4×4 matrix with α and β representing the (Dirac) matrix indices. We note that the scalar (meson) Feynman propagator is an even function and corresponds to that of a charge neutral ﬁeld. k = iGF (k) = lim ǫ→0 k 2 i . (This is the same as saying that a fermionic particle moves in the direction shown or an anti-particle moves in the opposite direction. as we have seen. On the other hand. going from from ψ β to ψα .10 Feynman diagrams 319 with a number of propagators each of which corresponds to the contraction of a pair of ﬁelds (Feynman Green’s functions) and normal ordered products. namely. simply give rise to normalization factors as well as spinor functions for the particles in the initial and the ﬁnal states. the line representing the fermion propagator has a direction associated with it. Hence. therefore. on the other hand. The propagators are completely independent of the initial and the ﬁnal states and. The normal ordered products have to give nonvanishing matrix elements between the initial and the ﬁnal state and. let us introduce a graphical representation for these two basic elements of Wick’s expansion. occur as internal lines in a diagram. are represented by external lines. αβ (8. the Feynman propagator for the fermions is not an even function and corresponds to that of a (Dirac) ﬁeld carrying charge. Thus. The initial and the ﬁnal states. For example. − M 2 + iǫ β i (k + m)αβ / lim α = iSF . an external boson line can represent either the annihilation or the creation of a particle at the interaction point (vertex).8.104) Several things are to be noted here. the line representing this propagator has no direction associated with it.αβ (k) = ǫ→0+ k2 − m2 + iǫ k = lim ǫ→0+ i k − m + iǫ / . simply because we have a charge neutral scalar ﬁeld .

320 8 Dirac field theory and in such a case. is represented as a vertex (point of interaction) with no meaning attached to the external lines. therefore.107) k. (particle creation). a fermion external line with an arrow away from the interaction point (vertex) can represent the creation of a particle or the annihilation of an anti-particle (since it corresponds to the ﬁeld ψ) with an external line factor = m (2π)3 k 0 m (2π)3 k 0 uα (k. (8. carries a direction for charge ﬂow). (anti-particle creation). in such a case is given by (the simplest way to visualize this is to note that the end of a fermion line along the direction of the arrow corresponds to a ψ ﬁeld which can either annihilate a particle or create an anti-particle) = m (2π)3 k 0 m (2π)3 k 0 uα (k.105) On the other hand. in a local quantum ﬁeld theory. vα (k. s α = However. s). s). (8. we .93). The interaction can be read out from the Lagrangian density LI and. s). (anti-particle annihilation). the normalization factor for the external line √ (state) is denoted by (k0 = k2 + M 2 ) k = 1 (2π)3 2k0 . for the Yukawa interaction (8. s α = Note that it is the interaction between the diﬀerent ﬁelds which gives rise to nontrivial scattering. a fermion external line with an arrow ﬂowing into an interaction point (vertex) can represent either the annihilation of a particle or the creation of an anti-particle at that point and the external line factor. s). v α (k. (8. Thus. (particle annihilation).106) k. since the fermion ﬁeld carries a charge (and.

(8. s′ Figure 8.108) These are the basic elements out of which Feynman diagrams are constructed by attaching internal and external lines to vertices. . We should also note for completeness that these are the Feynman rules in momentum space. calculations of S-matrix elements are simpler in momentum space. the interaction vertex is obtained from (iSI = i d4 x LI ) by taking appropriate ﬁeld derivatives and the factor of (2π)4 arises from transforming to momentum space) β k1 α k3 k2 = −(2π)4 ig δ4 (k1 + k2 + k3 ) δαβ . one has to have a factor of (−1) for every internal fermion loop. (There is one other rule. conventionally known as Feynman rules. with the normalization factor (2π)4 for every momentum integration.8. which we do not worry about at this point. We can also deﬁne analogous Feynman rules in coordinate space. With these rules.10 Feynman diagrams 321 have with the convention that all momenta are incoming at a vertex (basically. namely.) q α β k. Furthermore. we can show that every Feynman diagram corresponds to a unique physical process in one to one correspondence with Wick’s expansion. However. the rule for connecting vertices and lines is that the four momentum of every internal line (propagator) must be integrated 1 over all possible values. There are also some other subtleties like the symmetry factor associated with a diagram (graph).1: Lowest order Feynman diagram. s k ′ .

(2π)3 k′ 0 (8. s) (2π)3 k0 m v β (k′ . 8. s2 k3 . s3 Figure 8. k1 . an anti-proton and a π 0 meson are annihilated (which we considered in (8. therefore. k′ . the corresponding lowest order Feynman diagram will be given by Fig. s′ ) (2π)3 k′ 0 (1) = −(2π)4 ig δ4 (k + k′ + q) × 1 (2π)3 2q 0 m (2π)3 k0 m v α (k′ . where the external lines represent the respective physical particles that are annihilated.2: A second order Feynman diagram. 8.1.322 8 Dirac field theory In this language.2. s). s1 k4 . The value of this diagram can be obtained from the Feynman rules to be 0|S1 |k. with the external lines representing physical particles. the physical process where a proton. s4 α β q δ γ k2 . s.99). s′ )uα (k. Let us next note that if we are interested in the second order process where a proton and an anti-proton annihilate and create a π 0 meson which subsequently pair creates a proton and an anti-proton. q = −(2π)4 igδ4 (k + k′ + q)δαβ × 1 (2π)3 2q 0 m uα (k. s′ . The .109) This is exactly the value of the ﬁrst order S-matrix element which we had evaluated in (8.99)) will correspond in the lowest order to evaluating the Feynman diagram in Fig.

namely. This shows the power of the Feynman diagram method. from the Feynman rules we see that it will be given by (we are ignoring the overall energymomentum conserving delta function (2π)4 δ4 (k1 + k2 − k3 − k4 )) . s1 )v β (k2 . s4 )iGF (k1 + k2 )v β (k2 .103). s1 . Once again. it is worth emphasizing that whenever confusions are likely to arise in the Feynman method. this is exactly the second order S-matrix element which we had calculated from the Wick expansion in (8. 8. However. k2 . For example.2 can describe the scattering of two protons through the exchange of a π 0 meson.10 Feynman diagrams 323 value of the diagram can now be calculated using the Feynman rules to be (2) k4 . s3 |S2 |k1 . s4 . s2 )iGF (q)(−(2π)4 ig)δ4 (q − k3 − k4 )δγδ × m 0 (2π)3 k3 m 0 uγ (k3 . s2 )uβ (k1 .8. s3 )vα (k4 . it is always resolved by going back to Wick’s expansion (of importance are concepts like the symmetry factor associated with a graph). If we are interested in only the basic diagram (and not the external line factors since the diagram can represent various distinct processes). k3 . the Feynman diagram in Fig. To conclude this section. s2 = d4 q (−(2π)4 ig)δ4 (k1 + k2 − q)δαβ (2π)4 m 0 (2π)3 k1 m 0 (2π)3 k2 ×uα (k1 . s4 ) (2π)3 k4 = −(2π)4 g2 δ4 (k1 + k2 − k3 − k4 ) × m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 (8.110) × uα (k3 . a proton with initial momentum k1 scattering into a state with momentum k2 by emitting a π 0 meson with momentum q = k1 − k2 which is captured by an initial proton of momentum k4 scattering into a state with momentum k3 . let us observe that the same Feynman diagram can describe distinct physical processes depending on the external lines. s1 ). s3 )vδ (k4 .

4π |x| = (8.111) where V (q) represents the scattering amplitude in the Born approximation.114) . it follows from energy conservation that 0 0 k1 = k2 . Furthermore. (8. if we are in the center of mass frame where k1 = −k4 . we see that V (q) = V (q) and taking the Fourier transform of (8. 0 0 k2 = k3 . 0 0 k1 = k4 .113). q = (k1 − k2 ) = (0. we obtain d3 q iq·x e V (q) (2π)3 eiq·x d3 q (2π)3 −(q2 + M 2 ) + iǫ d|q||q|2 0 ∞ 0 ∞ −∞ π 2π −iV (x) = −ig2 = = = = = ǫ→0+ lim −ig2 ∞ 0 ig2 (2π)3 ig2 (2π)2 g2 dθ sin θ 0 dφ d|q| i|q||x|(|q|2 + M 2 ) zeiz|x| z2 + M 2 iM e−M |x| 2iM |q|2 ei|q||x| cos θ |q|2 + M 2 ei|q||x| − e−i|q||x| (2π)2 |x| dz g2 (2πi) (2π)2 |x| ig2 e−M |x| . k1 − k2 ). k2 = −k3 .113) Therefore.112) then.324 8 Dirac field theory −iV (q) = −ig2 GF (q) = ǫ→0+ lim − ig2 . (8. (8. q 2 − M 2 + iǫ q = k1 − k2 .

New York (1993). N. C. 3. Introduction to Quantum Field Theory. D.115) We recognize this as the Yukawa potential (for the exchange of a particle with mass M ) and this shows that the Yukawa interaction (8. John Wiley. 6. Relativistic Quantum Fields.11 References 325 where we have used the method of residues to evaluate the integral (the poles of the integrand are along the imaginary axis and closing the contour in the upper half or in the lower half of the complex zplane leads to the same result). New York. 5. 8. . Itzykson and J-B. New York. P. Evanston (1961).11 References 1. Quantum Field Theory. Peterson. McGrawHill. therefore. 4π |x| V (x) = − (8. Zuber. we obtain from (8. Schweber. 4.8. We recall from studies in scattering theory in non-relativistic quantum mechanics that the potential in the coordinate space corresponds to the Fourier transform of the Born amplitude and. Shirkov. Introduction to Relativistic Quantum Field Theory. S. Row.93) leads to the Yukawa potential in the Born approximation. Bjorken and S. Introduction to the theory of Quantized Fields. N. McGrawHill. 2. 1964.114) the potential between the fermions to be g2 e−M |x| . New York (1969). Gross. Roman. V. F. John Wliley. J. Moscow (1984). Drell. Bogoliubov and D. Nauka. 1980. Relativistic Quantum Mechanics and Field Theory.

This page intentionally left blank .

1) to write 327 . Such ﬁelds are commonly said to describe vector mesons (as opposed to (pseudo) scalar mesons described by the spin 0 Klein-Gordon ﬁelds) or gauge bosons if there is a gauge invariance associated with the corresponding theory.Chapter 9 Maxwell ﬁeld theory 9. It is more convenient from our point of view to rewrite Maxwell’s equations (9. ∂t ∂E ∇×B= . Thus. 1 ) representation of the Lorentz group (see sec2 2 tion 4. As we know. the dynamical ﬁeld would naturally describe particles with spin 1. ∂t ∇×E=− ∇ · B = 0. Let us start with the classical Maxwell’s equations in vacuum which are given by (remember c = 1) ∇ · E = 0.1) where E and B represent the electric and the magnetic ﬁelds respectively. (9.1 Maxwell’s equations The next ﬁeld theory to consider logically is that of an Abelian gauge theory and the Maxwell ﬁeld theory is a prototype of such a theory. this is described by the vector potential Aµ (x) which belongs to the ( 1 .1) in a manifestly Lorentz covariant form. ∂B . we note that we can solve the second and the third equations in (9.2). To that end.

328 9 Maxwell field theory B = ∇ × A. Aµ = (Φ. the second . (9. In other words. Therefore. A). denotes the electric ﬁeld vector. (9.6) (9. we note that F0i = ∂0 Ai − ∂i A0 = Ei . combine them into the four vector potential Aµ = (Φ. ∂t (9.3) If we now deﬁne the four dimensional curl of the four vector potential as Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . −A). E3 ). E = − ∂A − ∇Φ. Similarly. where E = (E1 .5) µ. ν = 0. 3. B3 ). Fij = ∂i Aj − ∂j Ai = −ǫijk Bk . We can. the six independent components of the tensor Fµν are given by (three components each of) the electric and the magnetic ﬁelds.7) (9.4) where ǫijk denotes the three dimensional Levi-Civita tensor and B = (B1 .2) where A(x) and Φ(x) are known as the vector and the scalar potentials respectively. of course. (9. (9. 2. 1.8) denotes the magnetic ﬁeld vector. B2 . E2 .

4)) ∂i F i0 = 0. (9.4) as Fµν = ∂µ Aν − ∂ν Aµ .1). ∂t (9.9) leads to ∂µ F µj = 0. ∂Ej + ∂ i (−ǫijk Bk ) = 0. (9. ∇ · E = 0. or.1).9) takes the form (F 00 = 0 by deﬁnition (9. equation (9. or. for ν = j.1) can be written in the manifestly covariant form as ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0.13) (9. ∂t ∂E = ∇ × B.11) which is the last equation in (9. or. or. Thus.1) are given by ∂µ F µν = 0.1 Maxwell’s equations 329 rank anti-symmetric tensor Fµν is also known as the ﬁeld strength tensor. ∂0 F 0j + ∂i F ij = 0. for ν = 0. (9.9. ∂t ∂Ej = −ǫjik ∂ i Bk .10) which coincides with the ﬁrst of the equations in (9. with the ﬁeld strength tensor deﬁned in (9. ∂ 0 F0j + ∂ i Fij = 0. It is now easy to check that the other two equations (the ﬁrst and the fourth) in (9. we see that the set of four Maxwell’s equations (9. Similarly. or. or.9) For example. equation (9.12) .

For example. let us examine whether there exists a Lagrangian density which will lead to Maxwell’s equations as its Euler-Lagrange equations. Fµν ′ → Fµν = ∂µ A′ − ∂ν A′ µ ν = = ∂µ (Aν + ∂ν θ) − ∂ν (Aµ + ∂µ θ) ∂µ Aν − ∂ν Aµ = Fµν . real. 9.14) where α(x) is an arbitrary. Maxwell’s equations (9. Consequently. Let us consider the Lagrangian density 1 1 L = − Fµν F µν = − (∂µ Aν − ∂ν Aµ )(∂ µ Aν − ∂ ν Aµ ) 4 4 1 (9.16) = − ∂µ Aν (∂ µ Aν − ∂ ν Aµ ).14) of the four vector potential Aµ (x). But ﬁrst.15) In other words. we note that under a gauge transformation Aµ (x) → A′ (x) = Aµ (x) + ∂µ θ(x). as an operator. which denotes the dynamical variable of the theory. the ﬁeld strength tensor is unchanged under the gauge transformation (9. 2 which leads to . (9.12) are also invariant under these transformations. local parameter of transformation. δAµ (x) = A′ (x) − Aµ (x) = ∂µ θ(x).330 9 Maxwell field theory Furthermore. µ (9. the gauge invariance of Maxwell’s equations is obvious in this formulation. µ or.2 Canonical quantization In trying to second quantize the Maxwell ﬁeld theory. we have to treat the four vector potential Aµ (x).

16) since the Lagrangian density (9. so that the Euler-Lagrange equations. 4 2 L= (9. Fµν = ∂µ Aν − ∂ν Aµ . in this case.2 Canonical quantization 331 ∂L ∂Aν ∂L ∂∂µ Aν = 0. The Lagrangian density (9. gives Maxwell’s equations as its Euler-Lagrange equations. ∂µ (∂ µ Aν − ∂ ν Aµ ) = ∂µ F µν = 0. take the form ∂L ∂L − ∂µ = 0.19) with both Aµ and Fµν treated as independent dynamical variables would also lead to Maxwell’s equations as its Euler-Lagrange equations.16). . However.9). ∂Fµν or. = − 1 1 1 (∂ µ Aν − ∂ ν Aµ ) − ∂ µ Aν + ∂ ν Aµ 2 2 2 (9. ∂∂µ Aν (9.9.20). ∂Aν ∂∂µ Aν or.19) is equivalent to (9. ∂L = 0.18) which coincides with the manifestly covariant Maxwell’s equations (9.16) when the ﬁeld variable Fµν is eliminated using the (ﬁrst) auxiliary ﬁeld equation in (9. ∂µ or.17) = − (∂ µ Aν − ∂ ν Aµ ) = −F µν . indeed. Note that a Lagrangian density 1 1 Fµν F µν − Fµν (∂ µ Aν − ∂ ν Aµ ) . It is manifestly Lorentz invariant. (9. ∂L = 0. we will work with the simpler Lagrangian density in (9.20) ∂µ F µν = 0. namely.

for canonical quantization of such a system. (9. This is a consequence of the gauge invariance of the theory. would then appear to be [Ai (x). we can think of A0 (x) as a classical function (c-number function) – not as an operator – and without loss of generality we can set it equal to zero A0 (x) = 0.15)).21) and the ﬁrst peculiarity of the Maxwell ﬁeld theory is now obvious from (9. we can deﬁne the associated conjugate momentum as (see (9.332 9 Maxwell field theory Furthermore. the momentum conjugate to the ﬁeld variable A0 (x) vanishes. x0 =y 0 .22) In other words.23) This is equivalent to saying that. we note that A0 (x) which would have a nontrivial commutation relation only with Π0 (x). namely. (9. Π0 (x) = −F 00 = 0. (9. (9. in this gauge the nontrivial components of the conjugate momentum are given by ˙ Πi (x) = −F 0i = Ei = −(∂ 0 Ai − ∂ i A0 ) = −Ai (x). As a result. ˙ ∂ Aµ (x) (9. Such a theory which is invariant under a (local) gauge transformation is known as a gauge theory and Aµ (x) is correspondingly known as the gauge ﬁeld. Πj (y) x0 =y 0 = 0 = Πi (x). this Lagrangian density is also invariant under the gauge transformation (9. Πj (y) j = iδi δ3 (x − y).25) . now must commute with all the ﬁeld variables in the theory. since the ﬁeld strength tensor Fµν is gauge invariant (see (9.21). in this gauge. we have to choose a gauge and this particular choice is known as the temporal (axial) gauge.14) of the vector potential. If we treat Aµ (x) as the independent ﬁeld variable.17)) Πµ (x) = ∂L = −F 0µ . Aj (y)]x0 =y0 Ai (x). Furthermore. Therefore.24) The naive canonical quantization conditions.

or.27). For example.25) leads to Ai (x).2 Canonical quantization 333 However. ∂ 0 (∂µ Aµ ) = 0. the dynamical equations (see (9.) On the other hand. in this gauge. (9.27) (9. namely. (9.23). x0 =y 0 (9. this is not quite consistent which can be seen as follows.28) and the condition (9. or. The proper quantization condition consistent with the transversality condition (9. x0 =y 0 x0 =y 0 (9.30) .23).26) Namely. the theory can be checked to have a residual gauge invariance under a time independent gauge transformation which preserves (9. j = −iδi δ3 (x − y). ∂ 0 (∇ · A) = 0. Πj (y) or. Aj (y) j = −iδi TR (x − y). lead to (for ν = 0) ∂µ (∂ µ A0 − ∂ 0 Aµ ) = 0.29) which does not satisfy the transversality condition (9. (As a parenthetical remark.18)) ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0. we note that even after imposing the gauge condition (9. ˙ Ai (x). Aj (y) j = iδi δ3 (x − y). the canonical quantization condition (9. or. ∇ · A = 0.9. in the gauge (9.23).27) can be thought of as the gauge choice for this residual invariance. the dynamical ﬁeld variable A(x) must be transverse.27) is given by ˙ Ai (x). ¯ A(x) → A′ (x) = A(x) + ∇θ(x).

the correct quantization condition (Dirac bracket) that results in the gauge (9. Here 1 2 stands for the inverse of the Laplacian operator and for∇ mally denotes the Green’s function for ∇2 .334 9 Maxwell field theory where the transverse delta function is formally deﬁned as the nonlocal j operator (whose coordinate representation is given by δi TR (x − y)) j j δi TR = δi + ∂i 1 ∂j .30) is consistent with the transversality condition in (9. there exists a systematic procedure for deriving the correct quantization conditions for such systems. When quantized systematically through the Dirac method.30).22).32) = ∂ j + ∂ i ∂i 1 j j j 2 ∂ = ∂ − ∂ = 0. which we will discuss in the next chapter.23) subject to (9.23). We note here that relations such as (9. the Hamiltonian density for the Maxwell ﬁeld theory can be obtained to be .33) In the gauge (9. Let us note here that in momentum space.30) has the explicit form j δi TR (x − y) = j x δi + ∂i ∂ jx 1 ∇2 x δ3 (x − y) = = d3 k ik·(x−y) j ki kj δi + 2 e (2π)3 k d3 k ik·(x−y) j e δi TR (k). (9.27) constitute constraints on the dynamics of the theory and systems with constraints are known as constrained systems.27). (2π)3 (9. known as the Dirac method. As we have pointed out in connection with the discussions in the Dirac ﬁeld theory. ∇2 (9.27) is given by (9. the transverse delta function in (9.23) and (9. ∇ and (9.31) so that (remember that ∂ i ∂i = −∇2 ) j j ∂ i δi TR = ∂ i δi + ∂i 1 j 2 ∂ ∇ (9.

38) (9.27)) A0 = 0.34) and leads to the Hamiltonian d3 x H = = 1 2 1 2 d3 x Π2 + B2 d3 x E2 + B2 .9.23) (see also (9. Maxwell’s equations (9. (9. (9.3 Field decomposition 335 ˙ H = Πi Ai − L = −Πi Πi + = = = = 1 Fµν F µν 4 1 2F0i F 0i + Fij F ij −Πi Πi + 4 1 1 −Πi Πi + Πi Πi + (−ǫijk Bk )(−ǫijℓ Bℓ ) 2 4 1 1 1 i Π2 + B2 − Π Πi + Bi Bi = 2 2 2 1 2 E + B2 . ∇ · A = 0. (9.18) take the form ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = Aν = 0.36) which also lead to ∂ µ Aµ = 0.35) H= Indeed this is what we know to be the Hamiltonian (energy) for the Maxwell theory even from studies in classical electrodynamics. 2 (9. 9.3 Field decomposition In the temporal gauge (9.37) .

in this case. since the vector potential is transverse.39) Thus. (9. or.23)) A0 ≡ 0. λ) · ǫ(k.41) The vector ǫ(k) represents the polarization (vector) of the plane ˆ wave solution travelling along k and carries the vector nature of the vector potential. for ν = i. λ′ ) = δλλ′ .40) (9. k · ǫ (k) = 0. complex satisfying k · ǫ (k.44) ǫ∗ (k. There can be two such independent directions. 2. the wave equation (9. therefore. the polarization vector characterizing the vector potential must be transverse to the direction of propagation of the plane wave. 9. λ = 1. satisfy the Klein-Gordon equation for a massless particle (wave equation) and.336 9 Maxwell field theory As we have seen. in general. (9. On the other hand. will have plane wave solutions of the form A(x) ∝ ǫ(k) e∓ik·x . with k0 = Ek = √ k2 = |k |. We can choose the two independent polarization vectors to be orthonormal and. for ν = 0.43) Namely. (9. the three components of the vector potential. for example. as shown in Fig. We note from (9. .1. (9. we have Ai = 0.42) (9.38) vanishes identically (because of the choice of the gauge condition (9. λ) = 0. we must have ∇ · A(x) = 0.27) that.

45) +ǫ∗ (k. 2) ǫ(k. λ) (9. quite analogous to that of the neutral Klein-Gordon ﬁeld except for the presence of the polarization vectors. The vector nature of the ﬁeld variable is now contained completely in the polarization vectors.3 Field decomposition 337 k ǫ(k. 1) and ǫ(k. λ) . It is also clear that any transverse vector ˆ (to the direction k) can now be expressed as a linear superposition of the two independent polarization vectors (9. This discussion is. Let us next note that . therefore. λ) are operators and as we have deﬁned. where k0 = Ek = |k |.44) (since they deﬁne a basis in the plane transverse to the direction of propagation).1: Two independent and orthogonal polarization vectors ǫ(k. λ) and a† (k. 2) transverse to the direction of propagation k. the ﬁeld operator A(x) is Hermitian as it should be.46) Note that a(k.9. 1) Figure 9. λ) eik·x a† (k. (9. λ) e−ik·x a(k. Given this. we can decompose the vector ﬁeld A(x) in the basis of the plane wave solutions as 2 A(x) = λ=1 d3 k (2π)3 2k0 ǫ(k.

47) and (9. λ) = i ← → (eik·x ∂0 A(x)) · ǫ∗ (k. we can show that d3 x (2π)3 2 ˙ eik·x A(x) 0 k0 ǫ(k. λ)ei(k +k)·x a† (k′ . λ) + ǫ∗ (−k. λ)a(k.50) . a(k. λ). λ)e 2 a(k′ . λ) . (9. λ)a(k′ . λ). λ=1 or. a† (k. (9. λ) . λ) 2k′ 0 0 +k ′ 0 )t +δ3 (k + k′ )ǫ∗ (k′ .49) where we have used the normalization for the polarization vectors in (9.44). λ) = −i (9. λ) = 1 0 √ ǫ(k. we conclude that i d3 x ← → 2k0 eik·x ∂0 A(x) = √ (2π)3 2k0 d3 x (2π)3 2k0 2 ǫ (k. λ) + ǫ∗ (k′ . λ)−ǫ∗ (−k. (9. λ)ei(k 2 a† (k′ . λ) = λ=1 d3 k ′ √ δ3 (k − k′ )ǫ(k′ .49) leads to d3 x (2π)3 2k0 ← → (e−ik·x ∂0 A(x)) · ǫ(k.48).338 9 Maxwell field theory d3 x (2π)3 2 e ik·x A(x) = λ=1 −i(k ′ −k)·x d3 k ′ d3 x √ 2k′ 0 (2π)3 ′ × ǫ(k′ .47) 2k0 λ=1 Similarly. Taking the Hermitian conjugate of (9. λ).48) 2 =−i λ=1 From (9. λ)a(k. λ)e2ik t a† (−k. λ)e2ik t a† (−k. λ)a(k.

54) The Hilbert space for this theory can be built up in the standard manner.49) and (9.61) in the case of the Klein-Gordon ﬁeld). λ′ )]. λ) behave exactly like the annihilation and the creation operators for a harmonic oscillator system.9. P = λ=1 d3 k k a† (k.50). ˙ ˙ = 0 = [Ai (x).51) using (9.50) denote the two inversion formulae for the Maxwell ﬁeld (analogous to (5. (9. λ)a(k.3 Field decomposition 339 Equations (9. Aj (y)]x0 =y0 (9. Given the quantization conditions (see (9. λ). j = −iδi TR (x − y). λ) and a† (k.46)) that Ek = |k|. a(k′ . we can easily show that [a(k. λ). [Ai (x). (9. The Hamiltonian and the momentum operators for the Maxwell theory can again be expressed in terms of the creation and the annihilation operators. (9. If we denote the vacuum state of the theory by |0 . then it satisﬁes . Aj (y)]x0 =y0 ˙ [Ai (x). λ′ )] = δλλ′ δ3 (k − k′ ).30)). λ′ )] = 0 = [a† (k. [a(k.77) and (5. a† (k′ . a† (k′ . The normal ordered forms for these operators can be shown to correspond to (see (5. Aj (y)]x0 =y0 .53) where we recall from the Einstein relation (see (9. λ). λ).97) and we do not show the normal ordering symbol explicitly) 2 H = λ=1 2 d3 k Ek a† (k. λ). λ)a(k. the operators a(k.52) In other words.49) and (9.

Later.58) This state.57) Therefore. Second. describes a one photon state and the index λ merely labels the polarization of the photon.56) which satisﬁes H|k.55) = 0 = P|0 . (9. λ = Ek |k. the true physical degrees of freedom are two in number. therefore. there are two distinct one photon states corresponding to the two possible independent transverse polarizations a photon can have. First of all. (9. λ = 0.340 9 Maxwell field theory a(k. λ P|k.23) and (9. λ)|0 . The one particle state is then obtained to be (λ = 1. we achieved quantization. the energy of the system is positive semideﬁnite and the Hilbert space is the standard harmonic oscillator space. show explicitly that the polarization is directly related to the spin of photon. (9. the canonically quantized description which involves only true degrees of freedom is not manifestly Lorentz covariant although the ﬁnal result of any physical calculation will be. namely the photons. we will also . λ = a† (k. do not carry any electric charge. = k |k. λ). by eliminating the extra degrees of freedom through non-covariant gauge choices (see (9. in fact. Furthermore. However. (9. shows that even though the vector potential Aµ (x) has four independent ﬁeld degrees of freedom. the particles of the theory. This. We can. As a result. involving the true degrees of freedom. λ)|0 H|0 = 0 = 0|a† (k. λ . λ . 2) |k. There are several things to note here. the one particle state satisﬁes (H 2 − P2 )|k.27)). λ 2 = (Ek − k2 )|k. since the A(x) ﬁeld is Hermitian like the charge neutral Klein-Gordon ﬁeld. consequently.

1) − ia† (k. 2 (9. 1) as creating a photon polarized along the x-axis and a† (k. 1) − ia† (k. 1) + ey a† (k.) A simple way to see that these would represent the creation operators for left and right circularly polarized photons is to note that 2 ˆ ˆ ǫ∗ (k. R L R L where we have used the familiar identiﬁcation that 1 ˆ ǫL (k) = √ ex − iˆy . (9. 2) as creating a photon polarized along the y-axis. (These one photon states are eigenstates of helicity with eigenvalue ±1. 2) λ=1 = 1 (ˆx + iˆy )(a† (k.62) describe respectively the left and the right circular polarization vectors.61) = ǫ∗ (k)a† (k) + ǫ∗ (k)a† (k). 1) + ia† (k. e 2 (9. e 2 1 ˆ ǫR (k) = √ ex + iˆy . .9.59) then. λ) = ex a† (k. λ)a† (k. 2) = ey . Let us note that if we choose the direction of propagation of the plane wave solution to be along the z-axis (k = kˆz ) and e ˆ ǫ(k.3 Field decomposition 341 quantize the Maxwell ﬁeld theory in a manifestly covariant manner which will bring out some other interesting aspects of such a theory. 2) . 2)) 2 (9. that the operators a† (k) = L a† (k) = R 1 √ a† (k. 2) . we can identify a† (k. 1) = ex . 1) + ia† (k. 2)) e e 2 1 e e + (ˆx − iˆy )(a† (k. in this language.60) would then create photons which are left and right circularly polarized respectively. 2 1 √ a† (k. It is clear. ˆ ǫ(k.

λ). (9.64) with k0 = |k|. for simplicity.52). (9.59) then we can write the ﬁeld decomposition (9. 3 2k 0 (2π) d3 k (2π)3 2k0 ǫµ (k. λ) = (0. λ)). λ)). This allows us to decompose the ﬁeld into its positive and negative energy parts as 2 A(+) (x) µ = λ=1 2 d3 k ǫµ (k. it follows that the only nontrivial covariant commutation relations have the form (these are not at equal time) 2 [A(+) (x).63) and furthermore. if we choose the polarization vectors to be real as in (9. λ′ )e−ik·x+ik ·y δλλ′ δ3 (k − k′ ) .65) A(−) (x) = µ λ=1 From the commutation relations for a(k.4 Photon propagator 9 Maxwell field theory Let us note that in the temporal gauge (9. a† (k′ . λ′ )] 2 = λ. λ)(e−ik·x a(k.23) (see also (9. λ)eik·x a† (k. (9. λ′ )e−ik·x+ik ·y [a(k. ǫ(k.λ′ =1 1 (2π)3 d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ ×ǫµ (k.45) as (recall (9. λ) and a† (k.λ′ =1 1 (2π)3 ′ d3 k d3 k ′ √ √ 2k0 2k′ 0 ×ǫµ (k.342 9. A(−) (y)] µ ν = λ. λ).27)). λ)+eik·x a† (k. λ). λ)ǫν (k′ . λ) in (9. 2 as four vectors ǫµ (k. if we deﬁne the two polarization vectors for λ = 1.23)) 2 Aµ (x) = λ=1 d3 k (2π)3 2k0 ǫµ (k. λ)e−ik·x a(k. λ)ǫν (k′ .

143) and (5.4 Photon propagator 2 343 d3 k ǫµ (k. . It follows now that Aµ (x)Aν (y) = (A(+) (x) + A(−) (x))(A(+) (y) + A(−) (y)) µ µ ν ν = A(+) (x)A(+) (y) + A(+) (x)A(−) (y) µ ν µ ν (−) +A(−) (x)A(+) (y) + A(−) (x)Aν (y) µ ν µ = = = Therefore. we obtain : Aµ (x)Aν (y) : +[A(+) (x).68) From this. µν [A(−) (x).66) which are the analogs of the positive and the negative energy Green’s functions for the Maxwell ﬁeld (see (5. namely. λ)ǫν (k. A(+) (y)] = −[A(+) (y).145)). λ) λ=1 = −iG(+) (x − y).67) Aµ (x)Aν (y) = 0|Aµ (x)Aν (y)|0 = −iG(+) (x − y). λ)e−ik·(x−y) 2k0 2 = λ=1 1 (2π)3 = 1 2 d3 k e−ik·(x−y) (2π)3 k0 ǫµ (k.9. λ)ǫν (k.107)). A(−) (y)] ν µ : Aµ (x)Aν (y) : −iG(+) (x − y) µν : Aµ (x)Aν (y) : + Aµ (x)Aν (y). A(−) (x)] = iG(+) (y − x) µ ν ν µ νµ = − 1 2 d3 k eik·(x−y) (2π)3 k0 2 ǫµ (k. µν (9. λ) λ=1 = −iG(−) (x − y). µν (9. (9. λ)ǫν (k. we can deﬁne the Feynman propagator in the standard manner (see (6.

µν (x − y) = −θ(x0 − y 0 )G(+) (x − y) + θ(y 0 − x0 )G(−) (x − y) µν µν = −i × d3 k (2π)3 2 ǫµ (k. for example.µν (x − y) = 0|θ(x0 − y 0 )Aµ (x)Aν (y) + θ(y 0 − x0 )Aν (y)Aµ (x)|0 = −iθ(x0 − y 0 )G(+) (x − y) − iθ(y 0 − x0 )G(+) (y − x) µν νµ = −iθ(x0 − y 0 )G(+) (x − y) + iθ(y 0 − x0 )G(−) (x − y) µν µν (−) = i(−θ(x0 − y 0 )G(+) (x − y) + θ(y 0 − x0 )Gµν (x − y)).344 9 Maxwell field theory 0|T (Aµ (x)Aν (y)) |0 = iGF . (9.(9.59)). (9. To simplify this expression.69) µν By deﬁnition. λ) λ=1 0 0 0 0 0 0 1 θ(x0 − y 0 )e−ik (x −y ) + θ(y 0 − x0 )eik (x −y ) .70) 0 2k where we have changed k → −k in the second term and have used the fact that the polarization vectors are unaﬀected by this change of variables (see. λ)ǫν (k.100) and write 1 0 0 0 0 0 0 θ(x0 − y 0 )e−ik (x −y ) + θ(y 0 − x0 )eik (x −y ) 0 2k = ǫ→0+ lim − dk′ 0 1 e−i(k +k )(x −y 2πi 2k0 k′ 0 + iǫ 0 ′0 0 0) − ei(k 0 −k ′ 0 )(x0 −y 0 ) k′ 0 − iǫ . the Feynman Green’s function has the form GF . λ)ǫν (k. we use the integral representation for the step function given in (6. λ) λ=1 1 θ(x0 − y 0 )e−ik·(x−y) + θ(y 0 − x0 )eik·(x−y) 2k0 d3 k ik·(x−y) e (2π)3 2 = −i × ǫµ (k. therefore.

λ) . λ)ǫν (k. k .140)). λ=1 (9. 0). we can identify the momentum space Feynman Green’s function for the Maxwell (photon) ﬁeld to be GF .72) Therefore. (9. Substituting this into (9. λ)ǫν (k.4 Photon propagator 345 ′0 0 0) = lim − ǫ→0+ dk′ 0 e−ik (x −y 2πi 2k0 ′0 = = ǫ→0+ lim − dk′ 0 e−ik (x −y ) 2πi (k′ 0 )2 − (|k| − iǫ)2 dk′ 0 e−ik (x −y ) . we obtain GF . λ)ǫν (k.µν (x − y) = lim ǫ→0+ d4 k e−ik·(x−y) (2π)4 k2 + iǫ 2 ǫµ (k.46). λ) λ=1 = GF (k) λ=1 ǫµ (k.73) where GF (k) denotes the Feynman Green’s function for a massless scalar ﬁeld (see (5.70) and denoting the variable of integration k′ 0 → k0 . 0. 2πi (k′ 0 )2 − k2 + iǫ ′0 0 0 0 1 1 − k′ 0 − k0 + iǫ k′ 0 + k0 − iǫ 0 ǫ→0+ lim − (9. |k| (9. ˆ kµ = kµ − (k · η)η µ (k · η)2 − k2 = 0.9.73) depends on the polarization sum and to evaluate the polarization sum.74) . λ) .71) where we have used the Einstein relation (9.µν (k) = ǫ→0+ lim 1 k2 + iǫ 2 2 ǫµ (k. let us note that if we introduce two new four vectors η µ = (1. The Feynman Green’s function in (9. 0.

λ)) (η ησ ) (kσ kσ ) λ=1 2 2 or.76) or. 2 deﬁne an orthonormal basis in four dimensions. λ)ǫσ (k.73). they satisfy ǫµ (k. also satisﬁes .346 9 Maxwell field theory ˆ then. η µ ηµ = 1. η µ . + iǫ iGF . λ)ǫµ (k. physical calculations carried out with this propagator do lead to Lorentz covariant results. (9.75) Consequently. we can write the completeness relation for these basis vectors as ˆ ˆ ηµ ην kµ kν ǫµ (k. ˆ ˆ (ǫσ (k. λ) = −ηµν + ηµ ην − kµ kν . λ) + ηµ ην − kµ kν = ηµν . λ)ǫν (k.78) and. kµ and ǫµ (k. ˆ ˆ kµ kµ = −1. ˆ η µ ǫµ (k. is manifestly transverse (see (9. λ)ǫν (k. (9. λ). λ=1 ˆ ˆ ǫµ (k. − 2 λ=1 ˆ ˆ ǫµ (k. λ) + σ + = ηµν . λ)ǫν (k. λ) = 0 = η µ kµ . we obtain the Feynman propagator for the photon ﬁeld in the temporal gauge to be i ˆ ˆ (ηµν − ηµ ην + kµ kν ). kµ GF . λ) = 0.µν (k) = 0. in this gauge. namely. Namely. However.75)).77) The Feynman propagator has a non-covariant look in this gauge. (9. as expected (since we are in the gauge A0 = 0). ˆ kµ ǫµ (k. λ = 1. λ′ ) = −δλλ′ . Substituting this back into (9.µν (k) = lim − ǫ→0+ k2 (9. The Feynman propagator.

which follows from (9. 4 = L0 + LI .82) The total Lagrangian density (9.80) (or (9.5 Quantum electrodynamics 347 GF .5 Quantum electrodynamics (9. Aµ (x) → A′ (x) = Aµ (x) + µ ′ 1 ∂µ θ(x).9. 9.83) .83)) and we have separated the Lagrangian density in (9. ψ(x) → ψ (x) = ψ(x)eiθ(x) . e (9.81) (9.80) denotes the covariant derivative (see (7.79) Quantum electrodynamics (QED) is the theory describing the interaction of electrons and positrons with the electromagnetic ﬁeld.0µ (k) = 0. where Dµ ψ(x) = (∂µ + ieAµ (x))ψ(x). The Lagrangian density for this theory is given by 1 L = − Fµν F µν + iψγ µ Dµ ψ − mψψ 4 1 = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ. (9.74).80) into the free and the interaction parts as 1 / L0 = − Fµν F µν + iψ∂ ψ − mψψ.82)) can be checked to be invariant under the ﬁnite local gauge transformations ψ(x) → ψ ′ (x) = e−iθ(x) ψ(x). 4 LI = −eψγ µ ψAµ . (9.

85)) ′ Dµ ψ(x) → Dµ ψ ′ (x) = = ∂µ + ieA′ (x) ψ ′ (x) µ ∂µ + ie Aµ (x) + 1 (∂µ θ(x)) e e−iθ(x) ψ(x) = e−iθ(x) (−i(∂µ θ(x)) + ∂µ + ieAµ (x) + i(∂µ θ(x))) ψ(x) = e−iθ(x) (∂µ + ieAµ (x)) ψ(x) = e−iθ(x) Dµ ψ(x). δǫ ψ(x) = ψ (x) − ψ(x) = iψ(x)ǫ(x).84) where ǫ(x) denotes the inﬁnitesimal local parameter of transformation. ψ. we need to examine the invariance of the minimally coupled Dirac Lagrangian density in (9. let us note that the covariant derivative in (9.83) (with θ(x) = ǫ(x) =inﬁnitesimal and keeping terms to linear order in ǫ) δǫ ψ(x) = ψ ′ (x) − ψ(x) = −iǫ(x)ψ(x).84)).81) transforms covariantly under a gauge transformation (see (7. ψ. Aµ ) → LDirac (ψ ′ . e (9. Aµ ). A′ ) µ ′ = iψ γ µ Dµ ψ ′ − mψ ψ ′ ′ ′ = iψγ µ Dµ ψ − mψψ = LDirac(ψ.83) (or (9. First. δǫ Aµ (x) = A′ (x) − Aµ (x) = µ ′ 1 ∂µ ǫ(x). the minimally coupled Dirac Lagrangian density transforms as ′ LDirac (ψ. In fact. Therefore.348 9 Maxwell field theory where θ(x) is the real and ﬁnite local parameter of transformation or under the inﬁnitesimal form of the transformation (9. we have already seen that the Lagrangian density for the Maxwell theory is invariant under the gauge transformation in (9. (9.80). = iψeiθ(x) γ µ e−iθ(x) Dµ ψ − mψeiθ(x) e−iθ(x) ψ (9. ψ .85) It follows from this that under a (ﬁnite) gauge transformation.86) .

λ µ = ǫ→0+ (9. µ k ν = iGF . therefore. (9. the complete QED Lagrangian density (9. J µ (x) = ψγ µ ψ.64) so that the external line in (9.107).80) r µβ α q p = −(2π)4 ie (γ µ )βα δ4 (p+q +r). λ).91) . the minimally coupled Dirac Lagrangian density is also invariant under the gauge transformation (9.80) is gauge invariant.90) can represent both the annihilation as well as the creation of a photon.77)) correponds to the temporal gauge (9.87) In momentum space the current conservation takes the form kµ J µ (k) = 0.5 Quantum electrodynamics 349 Namely. In this theory.106)-(8.89) lim − 1 ˆ ˆ i(ηµν − ηµ ην + kµ kν ) . (9. ∂µ J µ (x) = 0.27)).90) (2π)3 2k0 where we have chosen a real polarization vector as in (9. (9.23) (see also (9. k2 + iǫ ǫµ (k. there is a conserved current. We note here that the photon propagator is a gauge dependent quantity. (9. The form of the propagator in (9.88) The Feynman rules for QED can be derived in the standard manner.89) (or (9.83) and. In addition.9. The interaction vertex for QED can also be read out from the Lagrangian density (9. in QED we have the photon propagator as well as the photon external line factor which take the forms.µν (k) = k.104) and in (8. We have already derived the fermion propagator as well as the fermion external line factors in (8.

2. a) Photon pair production: For example. but not the vertices (points) where each of the two photons was produced. λ Figure 9.6 Physical processes 9 Maxwell field theory With these rules. the amplitude for this process would simply be the sum of the amplitudes (quantum superposition principle) given by the two diagrams in Fig.350 9. we only measure two photons in the ﬁnal state. let us calculate the matrix element for an electron and a positron to annihilate and produce a pair of photons e− (k1 ) + e+ (k2 ) → γ(k3 ) + γ(k4 ).92) To lowest order there are two Feynman diagrams which will contribute to such a process and they are shown in Fig. s2 k4 . 9. s1 ν + k1 − k4 k2 . .2. 9. we can now calculate the S-matrix elements for various physical processes in QED and let us see how some of these calculations are carried out at low orders. As a result. We note here that experimentally. the S-matrix element for this process can be worked out to be (we are assuming that the ﬁnal state photons are outgoing and have carried out the momentum integration for the internal line and are omitting the overall energy-momentum conserving delta function δ4 (k1 + k2 − k3 − k4 )) k3 . s1 k2 . λ′ ν µ k3 . λ′ k4 . λ µ k1 − k3 k1 . Therefore. s2 k1 . (9.2: The two lowest order Feynman diagrams that contribute to the pair production of photons as well as to the Compton scattering.

the γ-algebra can be simpliﬁed and a simple formula for the diﬀerential cross-section can be obtained (as discussed in section 3. λ′ ) +v (k2 . s1 ) / K − m + iǫ / 1 ǫ k4 . s2 ) ǫ (k3 . we have deﬁned ˜ K = k1 − k4 . s1 ) v β (k2 .6 Physical processes 351 γ(k3 .94) and the limit ǫ → 0+ is understood. depending on whether we are interested in summing over all spin states of the fermions or not.95) . λ). (9. γ(k4 . (9. λ)u (k1 . s2 ) × −(2π)4 ie (γ ν )βδ iSF . s2 ) ǫ (k4 .4). λ′ uα (k1 . s1 ) . b) Compton scattering: The Feynman diagrams for the elastic scattering of an electron by a photon e− (k1 ) + γ(k2 ) → e− (k3 ) + γ(k4 ). λ) / 1 ǫ (k3 . e+ (k2 ) = 1 1 4 (2π)6 (2π) 1 1 0 2k3 0 2k4 m 0 k1 m 0 k2 ×ǫµ (k3 . / ˜ − m + iǫ K / (9.9.δγ (K)(−(2π)4 ie) (γ ν )δα = − ie2 (2π)2 1 0 2k3 1 0 2k4 m 0 k1 m 0 k2 / × v (k2 . K = k1 − k3 . λ′ )|S (2) |e− (k1 ). λ′ u (k1 .93) where. The diﬀerential cross-section will be proportional to the absolute square of the matrix element. λ) ǫν k4 . Furthermore.δγ (K)(−(2π)4 ie) (γ µ )γα ˜ −(2π)4 ie (γ µ )βδ iSF . for simplicity.

γ(k4 . we will assume the ﬁnal state electron and photon (e− (k3 ). this process is described by the two Feynman diagrams shown in Fig. but . we can obtain the S-matrix element for Compton scattering directly from the results in (9. (9. 9. namely.2 except for the labeling of the momenta. γ(k4 )) to be outgoing.93) (or (9.352 9 Maxwell field theory are given to the lowest order by the same diagrams as in Fig. s3 )ǫ (k2 . Therefore.96)) not only in the external line factors. In this case.3. λ) = − ie2 (2π)2 1 0 2k2 1 0 2k4 m 0 k1 m 0 k3 / × u(k3 . the internal line represents a photon propagator signifying that the electrons scatter by exchanging (emitting and absorbing) a photon. this amplitude will differ from (9.93) to be e− (k3 ). The amplitude diﬀers from the case of the pair production of photons by only the external line factors. Thus. Here. λ′ )|S (2) |e− (k1 ). s3 )ǫ (k4 . (9.94) K = k1 − k4 and we have introduced Q = k1 + k2 . λ) / 1 ǫ (k2 . here we have the annihilation of an electron with momentum k1 and the creation of an electron with momentum k3 (the photon line factor in (9. s1 ) / Q − m + iǫ / 1 ǫ (k4 .98) To the lowest order. (9. λ′ )u (k1 . λ′ ) +u(k3 .90) is the same for the creation or the annihilation of a photon).96) / ˜ − m + iǫ K / ˜ where as deﬁned in (9. 9. s1 ) . λ)u(k1 .97) c) M¨ller scattering: M¨ller scattering is the study of elastic scato o tering of two electrons e− (k1 ) + e− (k2 ) → e− (k3 ) + e− (k4 ).93). instead of an electron and a positron being annihilated as was the case in (9. γ(k2 .

e− (k2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k4 . carrying out the momentum integration for the internal line and omitting the overall energy-momentum conserving delta function δ4 (k1 + k2 − k3 − k4 ).99) ˜ where K and K are deﬁned in (9.6 Physical processes 353 k3 k4 k4 k3 + k1 − k3 k1 − k4 k1 k2 k1 k2 Figure 9. This can be calculated in a straightforward manner as before. s1 ) 2 + iǫ K ˆ ˆ ˜ ˜ (ηµν − ηµ ην + Kµ Kν ) u(k4 . s1 ) .94). s2 ) ˜ 2 + iǫ K (9. s3 )γ ν u(k1 . +u(k3 . Assuming that the ﬁnal state electrons are outgoing. the S-matrix element takes the form e− (k3 ). also in the propagator of the diagram. s3 )γ u(k2 . s4 )γ µ u(k2 . s2 ) µ ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . d) Bhabha scattering: Finally. e− (k4 )|S (2) |e− (k1 ). the Bhabha scattering can be thought of as the process . s4 )γ ν u(k1 .9.3: The two lowest order Feynman diagrams for M¨ller scato tering.

e+ (k2 . e+ (k4 . s3 )γ µ v(k4 .354 9 Maxwell field theory e− (k1 ) + e+ (k2 ) → e− (k3 ) + e+ (k4 ). 9. s4 ) +v(k2 . k2 k4 k3 (9. s3 ).3. On the other hand.4: The two lowest order Feynman diagrams for Bhabha scattering.4 represents the annihilation of an electron and a positron to create a photon which subsequently creates an electron and a positron. In this case. We see that topologically these diagrams are the same as in Fig. s1 ). The lowest order Feynman diagrams for this process are shown in Fig. the second diagram simply describes the scattering of an electron and a positron by exchanging a photon. s3 )γ ν u(k1 . 2 + iǫ K (9. s2 )γ µ v(k4 . the ﬁrst diagram in Fig. 9. but in the present case we will treat the ﬁnal state electron and the positron as outgoing (e− (k3 ).100) k4 + k1 + k2 k1 k3 k1 k1 − k3 k2 Figure 9. s2 )γ ν u(k1 . e+ (k4 )). we have e− (k3 . 9. s1 ) 2 + iǫ Q ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . We can calculate this matrix element as before and omitting the overall delta function representing conservation of energy-momentum. s4 ) ˆ ˆ (ηµν − ηµ ην + Qµ Qν ) v(k2 . s2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k3 . s4 )|S (2) |e− (k1 . s1 ) .4.101) .

independent of the choice of gauge. This gives a ﬂavor of low order calculations in QED. of course. −p.103) that ∂ iSF (p) = ∂pµ = = ∂ ∂pµ i p−m / i 1 (−γµ ) p−m / p−m / i i (iγµ ) p−m / p−m / 1 = − (iSF (p)) Γµ (p.9. 4 (9. which we will study later.104) . the Lagrangian density (9.7 Ward-Takahashi identity in QED 355 where as deﬁned in (9.103) where the i factor in the denominator is understood. depends on the choice of gauge. 9.7 Ward-Takahashi identity in QED Before we discuss more complicated symmetries.97) K = k1 − k3 and Q = k1 + k2 . In simple language this identity implies that there exists a relation between the fermion two point function (self-energy) and the vertex function in QED and the relation can be traced to the gauge invariance of the theory. It follows from (9. (9.94) and (9. As we have seen. Here we would only give a very simple derivation of this relation in QED. e (9. Such relations can be derived more systematically and more formally for more complicated theories like the non-Abelian gauge theories through the BRST (Becchi-Rouet-Stora-Tyutin) symmetry.80) for QED has the form 1 ¯ ¯ L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ. let us discuss very brieﬂy the Ward-Takahashi identity in QED. However. 0) (iSF (p)) .102) The structure of the photon propagator in the theory. let us note that the fermion propagator in QED can be represented as i = p−m / iSF (p) = p .

e Consequently.104) we can obtain a relation between the fermion two point function and the three point vertex function as 1 Γµ (p. −p. As a result the right-hand side of (9. k = 0)(iSF (p)).107) ∂ (iSF (p)) (iSF (p))−1 ∂pµ = −i which can also be written as i 1 ∂ −1 S (p) = − Γµ (p. −p. µ F ∂p e (9. by inverting (9.356 9 Maxwell field theory where Γµ (p. −p. k = 0).108) .91) in QED with a zero momentum photon (without the delta function and the factor of (2π)4 ). −p.105) 1 = − (iSF (p))Γµ (p.106) In this case. ∂pµ F (9. 0) denotes the three point vertex (9. k = 0) = −(iSF (p))−1 e = ∂ (iSF (p))−1 ∂pµ ∂ −1 S (p).104) can be diagrammatically represented as i i iγµ p−m / p−m / 1 e = − p k=0 p (9. a diagrammatic representation for the basic identity (9. (9.104) in QED resulting from the structure of the fermion propagator has the form ∂ ∂pµ p =− 1 e p k=0 p .

(The trick in obtaining this relation in a simple manner lies in channeling the external momentum only through the fermion lines. Using the graphical identity in (9. let us look at the fermion self-energy at one loop. For example.109) Here we are looking only at the proper vertex parts of the diagrams (namely. we see that (9. The diagram on the left-hand side in (9. such relations are called Slavnov-Taylor identities which we will discuss later).109) represents the one loop correction to the fermion two point function or the self-energy (multiplied by a factor of i) whereas the right-hand side is the correction to the three point vertex function at one loop.106). without the external lines). but we understand that the external momentum is channelled only through the fermion lines) + Figure 9. .) At two loop order the fermion self energy diagrams have the forms shown in Fig.7 Ward-Takahashi identity in QED 357 This relates the three point vertex function to the fermion two point function (self-energy) in QED at the tree level and such relations are known as Ward-Takahashi identities (in non-Abelian gauge theories.108) holds for the one loop corrections to the amplitudes.9.5: Two loop corrections to the fermion self-energy in QED. Such a relation can be seen to hold at any higher order in perturbation theory as well. we can write k ∂ ∂pµ p k+p p =− 1 e p k p k+p k+p q=0 (9. Once again.5 (there are no external lines and we do not label the momenta for simplicity. 9.

Thus. We note that including quantum corrections (we will study renormalization in a later chapter).358 9 Maxwell field theory Using the basic graphical identity in (9.106) we can easily verify the validity of the identity (9.110) correspond precisely to the two loop corrections to the vertex function.110) We recognize that the graphs in the bracket on the right-hand side of (9. if the fermion self-energy and the vertex function change as .108) diagrammatically at every order in perturbation theory.106) at this order we obtain ∂ ∂pµ + =− 1 e + + + + + =− 1 e + + + + + . A simple and powerful consequence of the Ward-Takahashi identity can be derived as follows. (9. using (9.

then from (9.9.113).112) This physically implies that if there are divergent parts in the fermion self-energy graphs at higher order.113) −eCψA − Dψψ + · · · /ψ where A. Γµ −→ Z1 Γµ .85)) iaψD = iaψ(∂ + ieA /ψ / /)ψ. (9. D are constants and the dots denote other structures that may be induced due to quantum corrections. B. (9. then they must equal those present in the vertex correction graphs at the same order (we will study these issues in more detail within the context of renormalization later). Z2 are constants.114) .7 Ward-Takahashi identity in QED 359 −1 −1 SF (p. Invariance under the local gauge transformations (9. The Lagrangian density for QED including quantum corrections can be written as A 1 /ψ / L = − Fµν F µν + iψD − mψψ − Fµν F µν + iBψ∂ ψ 4 4 1 = − (1 + A)Fµν F µν + i(1 + B)ψ∂ ψ − e(1 + C)ψA / /ψ 4 −(m + D)ψψ + · · · . Thus comparing with (9. m = 0). m = 0) −→ Z2 SF (p. C.83). we can identify a = 1 + B = 1 + C. requires that the kinetic energy part of the fermion and the photon interaction term must combine into the form of a covariant derivative (see (9. on the other hand. That the Ward-Takahashi identity (9.111) where Z1 . (9.115) (9.108) is a consequence of gauge invariance can be seen heuristically in the following way. (9.108) we conclude that these constants must be related as Z1 = Z2 .

117) where Fµν denotes the ﬁeld strength tensor.116) We will discuss these topics in more detail when we study renormalization chapter 16. since the ﬁeld strength tensor Fµν does not change under a shift of the vector potential by a gradient. 4 Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . the Lagrangian density (9.117) as well as the action for this theory are invariant under the gauge transformation δAµ = ∂µ α(x). ˙ ∂ Aµ Πµ = which implies the constraint Π0 = −F 00 = 0.118) As we have already seen in (9. We recall that the Lagrangian density for Maxwell’s theory is given by 1 L = − Fµν F µν . The canonical momenta conjugate to the ﬁeld variables Aµ can be calculated from the Lagrangian density and take the forms (see (9.22)) ∂L = −F 0µ . we can choose a gauge condition and if we choose the gauge ∇·A = 0 (in our earlier discussion we had chosen the temporal gauge A0 = 0.8 Covariant quantization of the Maxwell theory As we have seen. then the equations of motion (see (9. (9.18)) .15). As a result. the present choice of gauge is known as the Coulomb gauge). Let us recapitulate brieﬂy how this arises.21) and (9. 9. (9.360 which is another way of saying 9 Maxwell field theory Z2 = 1 + B = 1 + C = Z1 . canonical quantization of Maxwell’s theory results in a lack of manifest Lorentz invariance. (9.

if sources (charges and currents) are present.8 Covariant quantization of the Maxwell theory 361 ∂µ F µν = 0. the two transverse physical degrees of freedom describe the true dynamics of the theory while the time-like degree of freedom is merely related to the charge density. however. ∇2 A0 = −J0 . . but we lose manifest Lorentz covariance in the process (because all components of Aµ are not being treated on a equal footing). A0 = − 1 J0 . The canonical quantization can now be carried out for the two physical degrees of freedom.9. Let us emphasize here that the ﬁnal result for the calculation of any physical amplitude. ∇2 A0 = 0. or.120) In either case. But. A0 = 0. or. or. remains manifestly Lorentz invariant in the canonical formalism. ∂i ∂ i A0 − ∂ 0 Ai = 0. ∇2 (9.119) On the other hand. or. (9. we lose manifest Lorentz covariance in the intermediate steps (which is highly desirable in any calculation). the dynamical equations ∂µ F µν = J ν . would lead to ∂µ F µ0 = J 0 . in the Coulomb gauge ∇ · A = 0. or. lead to (for ν = 0) ∂µ F µ0 = ∂i F i0 = 0. We recognize that this is a consequence of our choice of gauge.

. it projects on to the space of the components of any vector transverse (perpendicular) to the gradient operator ∂ µ .121) It follows from the explicit form of P µν that P µν Pνλ = (η µν = η µλ = = 2 λ − ∂ µ ∂ ν ) δν − ∂ν ∂ λ + ∂µ∂λ − ∂µ∂λ − ∂µ∂λ η µλ P µλ . namely. (This is relevant in the discussion within the context of a path integral quantization of the system which we will discuss brieﬂy in chapter 12.362 9 Maxwell field theory The need for a choosing a gauge can be seen in an alternative manner as follows.) Furthermore. (9. Consequently. 4 2 where P µν = η µν − ∂µ∂ν .123) With a suitable normalization.122) (9. we cannot carry out calculations in perturbation theory. (In fact.) We note that we can write the Lagrangian density (9.117) for the Maxwell theory also as 1 1 L = − Fµν F µν = Aµ P µν Aν + total derivatives. therefore. we see that P µν can be thought of as µν a projection operator. This implies that even if we can quantize the theory (say. the Feynman propagator for the theory cannot be deﬁned. in the naive path integral formalism). P = 1 P µν deﬁnes the normalized projection operator. (9. − ∂µ∂λ (9. we note that ∂µ P µν = ∂µ (η µν = (∂ ν − − ∂µ∂ν ) ∂ ν ) = 0 = ∂ν P µν . the inverse of P µν does not exist (this also means that the determinant of P µν vanishes) and the Green’s function and.124) so that this is the transverse projection operator.

we can try to modify the theory so as to make it nonsingular. Therefore. (This formulation of the theory is due to Fermi and considered as a gauge choice.125) . in this procedure we lose manifest Lorentz covariance since we single out the transverse degrees of freedom. However. the system is singular and contains constraints among the ﬁeld variables (as we will discuss in the next chapter). for example.125) would appear to be diﬀerent from Maxwell’s theory.9. the Lagrangian density 1 1 L = − Fµν F µν − (∂µ Aµ )2 + J µ Aµ . Let us consider. Namely. since we realize that the diﬃculties in quantization arise because of the singular nature of the Lagrangian density. In such a case.8 Covariant quantization of the Maxwell theory 363 We note here that whenever the determinant of the matrix of highest derivatives of the Lagrangian density vanishes.) This additional term breaks gauge invariance and consequently leads to a nonsingular theory. But to understand the issue better. But clearly the theory (9. 4 2 where J µ represents a conserved current (9.125) ∂µ J µ = 0.126) Here we have generalized Maxwell’s theory to include a conserved 1 current. the Cauchy initial value problem cannot be uniquely solved. this gauge is known as the Feynman-Fermi gauge. (9. Thus we see that the naive canonical quantization has unpleasant features in the case of Maxwell’s theory since the ﬁelds are constrained and the momentum conjugate to A0 vanishes. at this point there is no justiﬁcation for adding this new term to the Lagrangian density. We can solve for the constraints and quantize only the true dynamical degrees of freedom. simply because the Green’s function does not exist. let us look at the equations of motion following from the action in (9. But more than that we have also added a term − 2 (∂µ Aµ )2 to Maxwell’s Lagrangian density. without any further input. We can take an alternative approach.

the presence of this additional term in the Lagrangian density would not change the physics of Maxwell’s theory.364 9 Maxwell field theory ∂µ or.131) seems to modify the theory. this is just the Maxwell’s equations in the presence of conserved sources.127) takes the form ∂µ (∂ µ Aν − ∂ ν Aµ ) + ∂ ν (∂ · A) = −J ν . (9. An alternate way to see this is to note that if we take the divergence of the equations of motion in (9. Furthermore. ∂µ F µν + ∂ ν (∂ · A) = −J ν . or.130) where we have used the anti-symmetry of the ﬁeld strength tensor as well as the conservation of J µ . (9. (∂ · A) = ∂ν J ν . If we now write out the left-hand side explicitly.132) . (9. t = 0. (∂ · A) = 0. or. ∂∂µ Aν ∂Aν −∂µ F µν − ∂ ν (∂ · A) − J ν = 0. (9. we recognize that if we restrict classically to the initial value conditions ∂χ = 0.128) (9. therefore. Aν = −J ν . χ is a free ﬁeld and. or. ∂t χ = 0.127). (9.129) ∂ · A = −∂ · J = 0.127).125) where χ = ∂ · A.127) Without the second term on the left-hand side in (9. ∂L ∂L − = 0. (9. Thus we see that although the presence of the term − 1 χ2 in the 2 Lagrangian density (9. we have ∂ν ∂µ F µν + or.

137) x0 =y 0 . classically we can think of Maxwell’s theory as described by the modiﬁed Lagrangian density (9. (9. Thus. ˙ ∂ Aµ Πµ = (9.135) Clearly now all components of the momenta are well deﬁned without any constraint and hence we can quantize the theory as (Πν is conjugate to Aν ) ν [Aµ (x). Πν (y)]x0 =y0 = iδµ δ3 (x − y). 2 (9.136) with all other commutators vanishing. or.8 Covariant quantization of the Maxwell theory 365 then χ = 0 at all times and we recover the familiar Maxwell’s theory. We can deﬁne the canonical momenta as ∂L ˙ = −Aµ .134) It is now obvious that the coeﬃcient matrix of highest derivatives is nonsingular in this case. ˙ Aµ (x). Let us now rewrite this modiﬁed Lagrangian density as 1 1 L = − ∂µ Aν (∂ µ Aν − ∂ ν Aµ ) − (∂µ Aµ )2 − J µ Aµ 2 2 1 = − ∂µ Aν ∂ µ Aν − J µ Aµ + total divergence.9.125) with the supplementary condition ∂ · A = 0.40) and note that we are being slightly sloppy here in the sense that the commutation relation should really involve ﬁeld variables and their conjugate momenta) ˙ φ(x). (9.133) (9. φ(y) = iδ3 (x − y). We now see that this quantization relation is exactly like the quantization condition for four distinct scalar ﬁelds (see (5. Aν (y) x0 =y 0 = −iηµν δ3 (x − y).

we note that if we calculate the commutator of the ﬁeld variables with the Hamiltonian. 2 2 (9. Namely.138) It is worth pointing out at this point that the Hamiltonian density (and. But more serious than that is the fact that if we adopt the above convention. the coordinate and the momenta exchange roles. we ﬁnd 1 d3 x′ − Πν (x′ )Πν (x′ ) 2 Aµ (x). However. the Hamiltonian) for this theory does not seem to be positive semi-deﬁnite since the time components (µ = 0) contribute a negative amount. Let us construct the Hamiltonian density for this theory and see whether the above quantization relations are consistent (assume. the commutation relation between A0 ˙ and A0 has a relative negative sign. that would be against the spirit of Lorentz covariance since we no longer treat all the components of Aµ on the same footing. J µ = 0 and we will neglect the total divergence terms) ˙ H = Πµ Aµ − L 1 = −Πµ Πµ + ∂µ Aν ∂ µ Aν 2 1 1 = −Πµ Πµ + Πµ Πµ + ∂i Aµ ∂ i Aµ 2 2 1 1 = − Πµ Πµ + ∂i Aµ ∂ i Aµ . H = Aµ (x). Πν (x′ )Πν (x′ ) d3 x′ iηµν δ3 (x − x′ )Πν (x′ ) x0 =x′0 ˙ = −iΠµ (x) = iAµ (x). + 1 ′ ∂ Aν (x′ )∂ ′i Aν (x′ ) 2 i x0 =x′0 x0 =x′0 = − = − 1 2 d3 x′ Aµ (x). therefore. This problem can. for simplicity. (9. the Hamiltonian for Maxwell’s theory would become unbounded from below.366 9 Maxwell field theory except for one thing. be simply ﬁxed by postulating that for the time component.139) . However. of course.

136) would lead to Aµ (x). whereas the left-hand side of the above expression would be zero if the supplementary condition (9. = i∂ ν δ3 (x − x′ ). Thus.133) may be thought of as picking out the smaller physical subspace from the larger total vector space. We . or.133) as yet. but the derivation needs to be done carefully in a limiting manner x0 → x′0 .140) in the quantum theory. (9. = i∂ ν δ3 (x − x′ ). we can think of the supplementary condition as selecting out the subspace of the physical Hilbert space of the theory. the naive “Hilbert” space is much larger than the physical Hilbert space of the Maxwell theory consisting of only the transverse photon degrees of freedom. we can weaken the supplementary condition (9. ∂µ Aµ |phys = 0.133) were to hold as an operator equation.133) and say that this condition is true only on the space of physical states of the theory. however.) As we would see shortly.141) x0 =x′0 ∂µ Aµ (x).142) In other words. such a condition is hard to implement as an operator condition. (9. (Note that since all ﬁeld components are dynamical in this modiﬁed theory. even (9. namely. where Aµ ’s are operators.136) are consistent. Πν (x′ ) x0 =x′0 x0 =x′0 (This relation is essentially correct. = iη µν δ3 (x − x′ ).) We note that.8 Covariant quantization of the Maxwell theory 367 which is the correct Heisenberg equation of motion for the ﬁeld variables Aµ and hence the quantization relations (9.9. Πν (x′ ) or. It is easy to see that the commutation relation (9. the non vanishing right-hand side would lead to an inconsistency. (9. We have. Πν (x′ ) ∂µ Aµ (x). The supplementary condition (9.142) is a very stringent condition and we have to relax this further. not made use of the supplementary condition (9. A little bit of thinking tells us that although in the classical theory we can impose the condition ∂µ Aµ = 0.

−1.146) ǫµ (k. a(k′ . λ)fk (x)a† (k. 1. λ′ ) a(k. λ′ ) = η λλ = ′ As we have seen earlier (see (9. (9. 3. λ = λ′ = i = 1. To proceed further. λ)fk (x)a(k.74)). (9. Πν (x′ ) |ψ or.145) Here ǫµ (k.144) where fk (x) = 1 (2π)3 2ω e−ik·x . . λ′ ) . µ d3 k ǫµ (k. λ′ ) = 0 = ′ a† (k. λ). 3.143) which is inconsistent. 2. 0 = i∂ ν δ3 (x − x′ ). x0 =x′ 0 = ψ|i∂ ν δ3 (x − x′ )|ψ . (Unlike the earlier discussion. λ). k0 = ω(k) = |k|. a† (k′ . λ = λ′ = 0. we can always choose a basis such that the polarization vector for λ = 0 is time-like (recall η µ ). λ). λ) and a† (k.) They are normalized as 1. λ)ǫ∗µ (k. At this point we can quantize the coeﬃcients a(k. let us make a plane wave expansion for the ﬁeld variables Aµ (x) = λ ∗ + ǫ∗ (k. λ) . here they are not required to be transverse. then ψ| ∂ · A(x).141). a† (k′ . (9.368 9 Maxwell field theory can see this already from the commutator relation (9. λ) (9. λ)’s are the components of the polarization vector for a photon travelling along k and polarization index λ = 0. 2. Basically. we note that if |ψ represents a physical state.147) = −η λλ δ3 (k − k′ ). (9. λ) as annihilation and creation operators satisfying a(k. ˆ while that for λ = 3 is longitudinal (recall kµ ) and the other two polarization vectors are transverse to the momentum four vector.

λ′ ) ν ∗ d3 kd3 k′ ǫµ (k. λ)ǫ∗ (k′ .λ′ d3 kd3 k′ iω ′ ∗ × ǫµ (k. 2. λ′ )fk (x)fk′ (y) ν µ = λ d3 k(−iω) ǫµ (k. λ′ ) µ = λ. Aν (y) = ∗ ∗ d3 k(−iω)ηµν [fk (x)fk (y) + fk (x)fk (y)] = −iηµν δ3 (x − y).8 Covariant quantization of the Maxwell theory 369 where λ. λ). ǫσ (k. λ′ ) ν ∗ −ǫ∗ (k. λ′ )fk′ (y)a† (k′ .148) Using the completeness relation for the polarization vectors ǫµ (k. λ)ǫ∗ (k′ . λ′ ) − ǫ∗ (k′ . λ)ǫ∗σ (k. 1. λ) k (9. a† (k′ . Of course. a(k′ . λ′ )fk (x)fk′ (y) a(k. λ)ǫ∗σ (k. λ′ )fk (x)fk′ (y) a† (k. λ) ν ∗ fk (x)fk (y) ǫσ (k.150) . Aν (y) = λ.9. λ′ = 0. λ)ǫ∗σ (k. λ′ )fk′ (y)a(k′ . λ) + ǫ∗ (k. ǫσ (k. λ) + ǫ∗ (k. λ) (9.λ′ ∗ −iω ′ ǫν (k′ . λ)fk (x)a(k. λ′ )fk (x)fk′ (y) + ǫ∗ (k. λ)fk (x)a† (k. we do not write the equal-time arguments explicitly although they are assumed) ˙ Aµ (x). λ)ǫν (k′ . (9. λ)ǫν (k′ . λ)ǫ∗ (k. λ). we obtain the equal-time commutator to be ˙ Aµ (x). λ)ǫ∗ (k. λ) ν = ηµν .λ′ d3 kd3 k′ (−iω ′ η λλ )δ3 (k − k′ ) ′ ∗ ∗ × ǫµ (k. we have to check that these conditions are consistent with the equal-time quantization conditions for the ﬁeld variables (for simplicity. λ). λ) ∗ µ f (x)fk (y) . 3. λ)ǫν (k.148).149) λ in (9. µ = λ.

2)a(k. (9. 2. let us denote by |0 the vacuum state of the theory so that 0|0 = 1. λ) and a† (k. λ)|0 = 0. 0)a(k. If we calculate the normal ordered Hamiltonian of the theory (recall J µ = 0). 2) + a† (k.154) . We also note from (9. 1. a(k.370 9 Maxwell field theory Thus the commutation relations for the creation and the annihilation operators are indeed consistent with the quantization condition for the ﬁeld variables.152) Let us next consider the state with one time-like photon |1. λ = 0. 0) + a† (k.147). For example. λ) behave respectively like annihilation and creation operators for λ = 1. it is strange to note that there are four independent photon degrees of freedom.147) that although a(k. it has the form :H: = d3 k ω(k) −a† (k. 3.151) We can now develop the Hilbert space description for the photons in this theory which leads to some unusual features in the present case. 3)a(k. The norm of this one photon state can now be calculated (9.153) where F (k) is a suitable smearing function such that d3 k |F (k)|2 < ∞. 1) +a† (k. However. for the time-like component there is a relative negative sign in the commutation relation (9. (9. 3. λ = 0)|0 . (9. λ = 0 = d3 k F (k)a† (k. 2. 3) . 1)a(k.

We also realize that this negative norm is a consequence of the negative sign in the commutator (9.158) and the energy of this one photon state is given by . 1) +a† (k.155) This shows that the vector space of the theory has an indeﬁnite metric – the norm for the one time-like photon state is negative although states containing only space-like photons have positive norm.9. 2) + a† (k. the normal ordered Hamiltonian will have the form (compare with (9. so that the one time-like photon state is given by (9. Thus the natural modiﬁcation that comes to mind is to interchange the roles of these operators for the time-like photon. 3)a(k. Let us suppose that a† (k. 1)a(k. 0)|0 = 0. λ = 0|1. (9.147) for the annihilation and the creation operator for a time-like photon. this also leads to trouble as we see below.8 Covariant quantization of the Maxwell theory 371 1. 0 = a(k. 0)a(k. (9.156) |k.151)) :H: = d3 k ω(k) −a(k. λ = 0 = = = − d3 kd3 k′ F ∗ (k)F (k′ ) 0|a(k. 0)a† (k. 0)a† (k′ .157) In this case. 0)|0 d3 kd3 k′ F ∗ (k)F (k′ ) 0|a† (k′ . 2)a(k. 0)|0 . 0) + a† (k. (9. 0) − δ3 (k − k′ )|0 d3 k |F (k)|2 < 0. However. 3) .

In fact it is obvious that with this deﬁnition. 0). (9. 0)]|0 (9. But in the presence of interactions such states may be excited. 0 = Ha(k. the physical states must satisfy ∂µ Aµ (x)|phys = 0. the energy of the state with one time-like photon becomes negative. n3 |m0 . 0)a† (k′ . This is seen by using the supplementary condition (subsidiary condition) which we still have not imposed.160) where nλ . a(k. This not only demands that certain . 0)|0 = [H.159) = −ω(k)a(k. m2 . 1.161) However. where we have used the fact that H|0 = 0. 0 . Thus states containing an odd number of time-like photons have negative norm. The question that immediately comes to our mind is what happens to the probabilistic interpretation of the theory. n1 . n2 . physical results are well behaved. As we will see shortly in spite of the presence of negative norm states. deﬁnite states of arbitrarily high negative energy values are possible and the Hamiltonian becomes unbounded from below. 3. Namely. as discussed earlier this is too stringent a condition to allow any state of the radiation ﬁeld. 0)]|0 = d3 k′ ω(k′ )[−a(k′ . mλ denote number of photons for each of the polarizations λ = 0. m1 . (9. 2. In a free theory we may impose suitable conditions to prohibit any unwanted state.372 9 Maxwell field theory H|k. So we are stuck with an indeﬁnite metric space (with the standard interpretation for annihilation and creation operators for the timelike photon) and the normalization of states in this space is given by (we suppress the momentum label for simplicity) n0 . 0)|0 = −ω(k)|k. Thus we see that if we exchange the roles of the creation and the annihilation operators for the time-like photon. m3 = (−1)n0 δn0 m0 δn1 m1 δn2 m2 δn3 m3 . a(k.

3))|phys = 0. or. 2. λ or. 0) − a(k. (9. 3)a(k. where we have used the deﬁnitions and the properties of the polarization vectors in (9. Then.162) where ∂µ Aµ (+) (x) is the positive frequency part of the divergence of the vector potential (Maxwell’s ﬁeld) and contains only the destruction operator or the annihilation operator. λ = 3) |phys = 0. (9. (This is commonly known as the Gupta-Bleuler quantization. k0 a(k.74) as well as the fact that for positive energy photons k0 = ω = |k|. 0)a(k. λ) = 0 for λ = 1.164) the Gupta-Bleuler supplementary condition leads to kµ ǫµ (k.8 Covariant quantization of the Maxwell theory 373 kinds of photons are not present in the physical state.129) or (9.) We remark here that since (see (9. Recalling that kµ ǫµ (k. λ)|phys = 0. as we have seen. Let Vphys = {P } denote the set of all states which satisfy the supplementary condition and let |ψ represent such a state. Gupta and Bleuler weakened the supplementary condition to have the form ∂µ Aµ (+) (x)|phys = 0.163) ∂µ Aµ (x) is like a free scalar ﬁeld. 3) |phys = 0.165) implies that . or.9. λ = 0) − k2 a(k. it can be decomposed into positive and negative frequency parts uniquely in a relativistically invariant manner and this decomposition is preserved under time evolution.130)) ∂µ Aµ (x) = 0. 0) + ǫµ (k.165) (a(k. λ)a(k. (9. but it also requires that those photons cannot be emitted. Therefore. the supplementary condition (9. |k| (9. kµ ǫµ (k.

All the coeﬃcients Cn0 . (9.n2 . n1 .n1 .169) which leads to the relation in (9.n1 .n3 − n0 |n0 − 1. or.170) .n3 = 0.n2 .n3 in the expansion can be determined recursively from (9. √ Cn0 . n2 . a superposition of diﬀerent states of the form |ψ = n0 . 3)) |n0 . 0)|ψ = ψ|a† (k.n2 . 0)a(k.n2 .n1 . n3 − 1 = 0. n2 . (9.n3 Cn0 .n1 .168) with the condition C0. n2 .167) where nλ denotes the number of photons with polarization λ and the supplementary condition relates the number of time-like and longitudinal photon states as √ √ n0 Cn0 . A general physical state is. 0)|ψ = a(k.n3 ) ×|n0 − 1.n2 .n1 . n1 . (9. n3 = 0.n2 .0 = 1.n1 . n2 . n3 √ − n3 |n0 .n3 −1 + n3 Cn0 −1. 0) − a(k. ψ|a† (k.n3 |n0 .n2 .n1 . (9. n1 . 3)a(k. n1 . n2 . n3 .n1 .168). Cn0 .n1 . √ √ ( n0 Cn0 . 3)|ψ .n3 (a(k. (9. or. of course.168) This can be seen as follows or.374 9 Maxwell field theory a(k.n2 .166) This leads to the fact that the physical states must contain superpositions of states with an equal number of time-like and longitudinal photons.n2 . 3)|ψ .n1 . n3 − 1 = 0.n3 −1 + n3 Cn0 −1.n2 . n1 .

Thus although we can write a general physical state with a ﬁxed number n1 . n1 . n1 .172) and so on. n0 = n3 = 0. n2 . 2 − 2|1.174) we can assume that in a truly physical state of the radiation ﬁeld there are no time-like and longitudinal photons present. n1 . (9. 0 .n3 ’s determined recursively. For example. n1 . 0|1. (9. = |0. n1 .n2 . In other words. all such states except for |φ0 have zero norm.n1 . n2 . 1 + |2. = φ0 |φ0 . 1 + 2. n1 . 1|1. n2 of transverse photons as a linear superposition of states of the form |φ because φ|φ = |φ0 + b1 |φ1 + b2 |φ2 + . 0 . n2 .171) and so on where we have used the values of Cn0 . . (9. n1 . = 0. 0 . 2|0. n2 . n1 . n1 . n2 .9. n1 . 2 + 2 1. n1 . n1 . the physical states allowed by the supplementary condition have the forms |φ0 |φ1 |φ3 = |0. 0 = 1 − 2 + 1 = 0. . n2 . n2 .175) . 1 − |1. n2 . . n2 .8 Covariant quantization of the Maxwell theory 375 It follows now that for a ﬁxed number n1 . n2 .173) (9. n2 of transverse photons. n1 . Let us note that because of the negative norm of states containing an odd number of time-like photons. n2 . 0|2. n1 . n2 . n2 . 1|0. n2 . we have φ1 |φ1 φ2 |φ2 = 0. n1 . n2 . n2 . 0 = 1 − 1 = 0. n1 . (9. 1 + 1. √ = |0.

(9. We also point out here that the Gupta-Bleuler supplementary condition can be thought of as the quantum analog of the covariant Lorentz condition ∂µ Aµ (x) = 0.179) This is diﬀerent from (9.376 9 Maxwell field theory in such states. + iǫ (9.89) simply because this can be thought of as quantizing the theory in a diﬀerent gauge (FeynmanFermi gauge). there is no problem with a probabilistic interpretation). Without going into details we note here that the Feynman propagator for the photon in this theory has the simpler form iGF . if we further mod out the states by the zero norm states. Since the zero norm states are orthogonal to all the states including themselves. The physical subspace of the theory selected by the supplementary condition (9. However their eﬀect cancels out in the ﬁnal result. If interactions are present time-like and longitudinal states may occur as intermediate states.77) or (9. V0 V phys = (9. such states decouple. on the space of physical states since ψ|∂µ Aµ |ψ = 0. consequently. Vphys . we have the true physical subspace of the theory where the norm of states is positive deﬁnite. namely. The physical S-matrix elements. One way of heuristically saying this is that the probability for the emission of a longitudinal photon cancels out the negative probability for the emission of a time-like photon.177) (9.µν (k) = lim − ǫ→0+ k2 iηµν . are gauge .176) where |ψ represents a physical state.165) contains states with positive semi-deﬁnite norm (negative norm states are eliminated by the supplementary condition or the physical state condition and. A diﬀerent way of saying this is to note that being orthogonal to every other state. however.178) where V0 represents the set of states with zero norm.

John Wiley. 5. V.9 References 377 independent and do not depend on the form of the photon propagator in a particular gauge. K. 4. Shirkov. Evanston (1961). P.9. Bjorken and S. New York (1969). J. 567 (1950). S. S. Row. Moscow (1984). New York. Gross. 9. Drell. Bleuler. Relativistic Quantum Fields. Ward. Introduction to the theory of Quantized Fields. Physical Review 78. Zuber. C. Introduction to Relativistic Quantum Field Theory. New York. New York (1993). 8. Nuovo Cimento 6. Helvetica Physica Acta 23. D. J. Quantum Field Theory. Bogoliubov and D. 6. Schweber. Proceedings of the Physical Society (London) A63. C. 9. Peterson. Introduction to Quantum Field Theory. We will develop these ideas further when we study the covariant quantization of non-Abelian gauge theories in chapter 13. 1980. 371 (1957). 1964. McGrawHill. Takahashi. N. N.9 References 1. 3. Gupta. Roman. McGrawHill. 681 (1950). John Wliley. N. 2. . Y. Nauka. Relativistic Quantum Mechanics and Field Theory. Itzykson and J-B. 182 (1950). 7. 10. F.

This page intentionally left blank .

we deﬁne the momenta canonically conjugate to the coordinates as pi = ∂L . N .1) Equation (10.Chapter 10 Dirac method for constrained systems 10. The Lagrangian is a func˙ tion of N coordinates qi as well as N velocities qi which are assumed ˙ to be independent so that the conﬁguration space of the theory is 2N dimensional. in general. rather some of these variables have to satisfy constraints following from the structure of the theory. carry out the quantization of a theory) and in this chapter we will discuss the Dirac method in some detail. · · · . Given the Lagrangian of the theory. Such systems are known as constrained systems and the naive passage to the Hamiltonian description for such a system starting from the Lagrangian description fails. i = 1. thereby. In this case. there is a systematic procedure due to Dirac which allows us to go from the Lagrangian description of a theory to the Hamiltonian description (and. ∂ qi ˙ (10. To appreciate the diﬃculties associated with constrained systems. we can go from the conﬁguration space of the system to the phase space 379 .1). velocities can be expressed in terms of momenta (and coordinates).1 Constrained systems As we have seen in the study of the Dirac ﬁeld theory in chapter 8 as well as the Maxwell ﬁeld theory in chapter 9. 2. let us consider a classical system of point particles described by the Lagrangian L(qi . relates momenta to velocities (and coordinates) of the theory and if this relation is invertible. qi ). the dynamical phase space variables of these theories are not all independent. In this case.

5). pj } = δi = −{pj . ∂qi (10.4) or (10.6) qi = {qi . A2 } = (10. H} = − ˙ which can be shown. The phase space spanned by the independent coordinates and momenta is also 2N dimensional with a (equal time) canonical Poisson bracket structure {qi . the time evolution of any dynamical variable A(qi . (10. qj } = 0 = {pi . ˙ (10. qi ). it follows now that the 2N ﬁrst order dynamical equations (evolution equations) for the system can be written in the Hamiltonian form ∂H . using the deﬁnition (10. pi ). pi ) as ∂A1 ∂A2 ∂A1 ∂A2 − . to be equivalent to the N second order Euler-Lagrange equations following from the Lagrangian description. pi ) = pi qi − L(qi .4) which allows us to write the Poisson bracket between any two dynamical variables A1 (qi . pj }. pi ) in the phase space can be written in the Hamiltonian form . In fact. A2 (qi . qi ) → (qi . ˙ ˙ (10.5) Using (10. H} = ˙ pi = {pi . j {qi .3) where summation over repeated indices is understood.380 10 Dirac method for constrained systems (qi . ∂pi ∂H .2) and uniquely deﬁne the Hamiltonian of the system through the Legendre transformation H(qi .3). pi ). ∂qi ∂pi ∂pi ∂qi {A1 . qi }.

However.9) and as discussed in chapter 8 (see (8. (10.14)). (10. H}. Πα ). θα ) = −Πα θα − L(θα .9) However. β {θα .1 Constrained systems 381 ˙ A(qi . (10. (10.12) The Poisson bracket between any two dynamical variables can now be obtained from (10. For example. If the relations (10. we have to deﬁne the derivative in (10.12). we can deﬁne the momenta conjugate to the coordinates as Πα = ∂L . since the dynamical variables are now of Grassmann odd (fermionic) nature. (10. . θα ). Πα ) = θα Πα − L(θα .10) through the Legendre transformation (this is consistent with the convention of left derivatives) ˙ ˙ ˙ ˙ H(θα . θα }.11) The 2M dimensional phase space is endowed with a canonical Poisson bracket structure (with the convention of left derivatives) {θα .5).8) then. if ˙ we have a classical system described by the Lagrangian L(θα . · · · M with θα θβ = −θβ θα . Πβ } = −δα = {Πβ . 2.10. θα ). pi ) = {A(qi . Πβ }. θβ } = 0 = {Πα . we will choose the convention of taking derivatives from the left. ˙ ∂ θα (10. α = 1.7) This discussion can also be generalized to a classical system described by Grassmann (anti-commuting) variables. pi ).9) can be inverted to express velocities in terms of the momenta. we can go from the conﬁguration space to the phase space ˙ (θα . unlike the bosonic case in (10. θα ) → (θα .

The discussion so far assumes that the transformation to phase space in (10. ∂Π ∂H ˙ . H} = − ∂θα (10. In a similar manner. F2 } = − ∂F1 ∂F2 ∂F1 ∂F2 + α ∂θα ∂Π ∂Πα ∂θα . B2 } = − = ∂B ∂B where we have used (10.14) the 2M ﬁrst order dynamical equations for the system can now be written in the Hamiltonian form as ∂H ˙ θα = {θα . (10.15) which can be compared with (10. θβ } α ∂θα ∂Π ∂Π ∂θβ ∂B1 ∂B2 ∂B1 ∂B2 + ∂θα ∂Πα ∂Πα ∂θα . for example.2) or (10.14) Using (10. Πβ } β − {Π . the Poisson brackets between a Grassmann even and a Grassmann odd variable as well as the bracket between two Grassmann odd variables can be obtained to correspond to {F. B} = − {B.12) as well as the fact that ∂θα and ∂Πα are Grassmann odd (fermionic) variables. Πα ) takes the form ∂B2 ∂B1 α ∂B2 ∂B1 {θα . F } = ∂F ∂B ∂F ∂B + α ∂θα ∂Π ∂Πα ∂θα ∂B ∂F ∂B ∂F + ∂θα ∂Πα ∂Πα ∂θα . {F1 .6).382 10 Dirac method for constrained systems the structure of the Poisson bracket now depends on whether the dynamical variables are Grassmann even (bosonic) or Grassmann odd (fermionic).13) {B1 . Πα = {Πα . with the convention of left derivatives. B2 (θα .10) is invertible so that all the velocities can be expressed uniquely in terms of independent momenta leading to the . the Poisson bracket between two bosonic dynamical variables B1 (θα . Thus. . H} = − α . Πα ). (10.

19) bij (q) = (10.16) we note that pi = ˜ij (q)qj + bij (q)qj . it follows that b = b(q). To understand the diﬃculty. the transformation to phase space in (10.20) Therefore. b ˙ so that ∂pi ∂2L = . (10. ∂ qi ∂ qj ˙ ˙ det (10.16) where the elements b.16) can be easily seen to have the form A−1 = ½ 0 −b−1˜ b−1 b . it follows that the transformation is invertible (A−1 exists) if ∂2L = 0. (10. ˜=˜ b b(q).21) . ∂ qj ˙ ∂ qj ∂ qi ˙ ˙ (10.18) so that A−1 exists only if the matrix b is invertible. (10. From (10. the diﬃculty in passage to a Hamiltonian description arises when this transformation is noninvertible.17) The inverse of the matrix A in (10. However.10. In this case. let us concentrate on the bosonic theory for simplicity. For conventional b theories described by Lagrangians with quadratic velocity terms. ˜ denote N × N matrices.2) can be written in general in the matrix form as q p q q ˙ =A = ½ 0 ˜ b b q q ˙ .1 Constrained systems 383 unique Hamiltonian of the system.

we can determine only R of the N velocities in terms of coordinates and independent momenta as qa = fa (qi .384 10 Dirac method for constrained systems The naive passage to the Hamiltonian description can be carried out only in this case. therefore. it is not clear a priori how to deﬁne the Hamiltonian uniquely. R. the transformation (10.20) that in this case not all the conjugate momenta can be thought of as independent variables leading to the fact that not all of N independent velocities can be expressed in terms of independent momenta.2) is not invertible. In this case. · · · . It follows from (10. As a result.4) need not satisfy the constraints of the theory and. the passage to a Hamiltonian description is achieved systematically through the method of Dirac which we describe in this section. if the Lagrangian describing the theory satisﬁes ∂2L = 0. 2. ˙ a.2 Dirac method and Dirac bracket In a physical system with constraints.23) of the physical system is of rank R < N . it is also clear that the naive canonical Poisson brackets in (10. there exist relations or constraints between various dynamical variables and such systems are known as constrained systems. On the other hand. pb ). therefore. In this case.22) then. (10. ∂ qi ∂ qj ˙ ˙ det (10. ∂ qi ∂ qj ˙ ˙ (10. In other words. 10.24) and the other N − R velocities cannot be determined. b = 1. Let us assume that the N × N matrix ∂2L . may not represent the true brackets necessary for a Hamiltonian description of the system. Furthermore. we can write .

In fact.29) so that on the constrained hypersurface Γc .25). b = 1. N.25) deﬁne N − R constraints among dynamical variables which we denote by ϕα = pα − gα (qi . (10. · · · . pa ). qb ). 2. if (F − G) = 0. ˙ ˙ ˙ ˙ ˙ (10. (10. we note that the constraints in (10. · · · . (10. qi ). R + 2.28) that in spite of the fact that qα cannot be determined. pa ) = 0. ˙ pα = gα (qi . namely.2 Dirac method and Dirac bracket 385 pa = ga (qi . the Hamiltonian of the theory is no longer unique.26) and we recognize that such constraints will reduce the dimensionality of the true phase space of the system. On the other hand.) Two dynamical variables F. α = R + 1. the canonical Hamiltonian does not depend on velocities as we would expect. a. For example. we also note that because of the non-invertibility of (10. pa ). from the canonical . ∂ qα ˙ ∂ qα ˙ (10. G in Γ are said to be weakly equal. ˙ ∂L ∂Hcan = gα − = gα − pα ≈ 0. we can deﬁne the canonical Hamiltonian of the theory as the Legendre transform Hcan = pi qi − L(qi . Hcan = Hcan (qi . We note from (10.26) (following from the Lagrangian of the theory) are known as primary constraints of the theory and deﬁne a 2N −(N − R) = N +R dimensional hypersurface Γc in the naive 2N dimensional phase space Γ of the system.28) where we have used (10. the dimensionality of Γc would be further reduced.25) It is clear that the second set of equations in (10.10. if they are equal on the constrained hypersurface Γc . F ≈ G. namely.27) Γc As usual.2) (presence of constraints). (If there are further constraints in the theory. R. qi ) = pa qa + gα qα − L(qi .

(10. Hcan } + λβ {ϕα . Hp } = − ˙ and using (10. Hcan } + λβ {ϕα . ∂pi ∂pi ∂Hp ∂(Hcan + λα ϕα ) =− .31) Furthermore.34) The constraints of the theory should be invariant under time evolution and using (10.35) .26) as (it is known as the primary Hamiltonian since it incorporates only the primary constraints in (10.4).26)) Hp = Hcan + λα ϕα . it follows that we can identify the Lagrange multipliers in (10. we can deﬁne the primary Hamiltonian associated with the system by incorporating the primary constraints of the theory (10.29). ϕβ } ≈ 0. (10.32) qi ≈ {qi . Hp } = ˙ pi ≈ {pi . Hp } = {F. ∂qi ∂qi (10. (10.30) with velocities λα ≈ qα .33) which remain undetermined.386 10 Dirac method for constrained systems Hamiltonian of the theory. ϕβ } + {ϕα . (10. The time evolution of any dynamical variable F can now be written as ˙ F ≈ {F. ˙ (10. with the canonical Poisson brackets in (10. Hcan + λα ϕα }. we can write the Hamiltonian equations as ∂Hp ∂(Hcan + λα ϕα ) = .30) where λα denote undetermined Lagrange multipliers and it follows that Hp ≈ Hcan . Hcan + λβ ϕβ } ˙ = {ϕα . λβ }ϕβ ≈ {ϕα .34) this leads to ϕα ≈ {ϕα .

Let us denote all the constraints of the theory (primary. secondary. then there remain undetermined Lagrange multipliers in the primary Hamiltonian after all the . On the other hand.36)-(10. . · · · . the constraints which have at least one nonvanishing Poisson bracket with the constraints are known as second class constraints and are denoted as b φα ≈ 0. Requiring the secondary constraints to be invariant under time evolution as in (10.2 Dirac method and Dirac bracket 387 where we have used (10.36) which can be divided into two distinct classes.27)) Poisson bracket with all the constraints (including themselves) are known as ﬁrst class constraints and are denoted as e ψ α ≈ 0. · · · . The constraints which have weakly vanishing (see (10.38) Here we have used the observation due to Dirac that there are an even number of second class constraints in a theory and we note from (10. n1 . if there are primary constraints in the theory which are ﬁrst class.10. (10. It is clear that (10. α = 1. In general. We continue with this process until all the constraints are determined to be evolution free. 2. It is also worth noting that in this process of determining all the constraints that are time independent (evolution free). · · · . 2. α = 1.26) in the last step.35) (with Hp as the Hamiltonian) may determine some other Lagrange multipliers or generate newer constraints (tertiary etc). ) collectively as ϕα ≈ 0.35) may determine some of the Lagrange multipliers or may lead to new constraints known as secondary constraints. tertiary. . .38) that n1 + 2n2 = n < 2N . all of the Lagrange multipliers may be completely determined or some of them may remain undetermined.37) 2. n < 2N. 2. (10. α = 1. 2n2 . (10. 1.

) The ﬁrst class constraints have a special signiﬁcance in that they are associated with gauge invariances (local invariances) in the theory. a consistent Hamiltonian description.388 10 Dirac method for constrained systems constraints have been determined. in general. 2.41) the Poisson bracket of the constraints with any dynamical variable F does not. Although we have identiﬁed all the constraints of the theory. (10.40) as . n1 . α = 1.39) and are chosen such that they convert the ﬁrst class constraints (10. The gauge ﬁxing conditions are generally denoted as e χα ≈ 0. As we have seen in the case of Maxwell ﬁeld theory. requires gauge ﬁxing conditions and we must choose as many gauge ﬁxing conditions as there are ﬁrst class contraints. (10. First. (10.42) This issue of incompatibility of the Poisson brackets can be addressed through the use of the Dirac brackets which are constructed as follows. collectively as φA ≈ 0. in this case.40) so that the true phase space of the theory is 2(N − n1 − n2 ) dimensional. we cannot yet impose them directly in the theory since the canonical Poisson brackets in (10. even though φA ≈ 0.37) into second class constraints.4) are not compatible with them. A = 1. which are all second class. · · · . 2(n1 + n2 ) < 2N. vanish {F. after gauge ﬁxing we denote all the constraints of the theory (including the gauge ﬁxing conditions). φA } ≈ 0. 2. / (10. · · · . (The number of undetermined Lagrange multipliers equals the number of ﬁrst class constraints among the primary constraints. Thus. let us note that we can construct the matrix of Poisson brackets of all the constraints in (10. Namely.

φA }CAB {φB . φA } −1 ≈ {F. φA }D = {F. The Dirac bracket has the interesting property . in the deﬁnitions of the Hamiltonian. φA } − {F. Jacobi identity) and in addition satisﬁes −1 {F. (10. in contrast to the canonical Poisson brackets.45) which can be shown to have all the properties of a Poisson bracket (anti-symmetry.44) will involve integration over the intermediate space coordinate. we can show that the Dirac brackets are indeed the correct Poisson brackets of the theory subject to the constraints in (10. G}. In fact. φA } = 0.10. (10. (10.44) We note here parenthetically that in ﬁeld theories (where the variables depend on space coordinates as well). φA } − {F. through the use of Lagrange brackets. energy-momentum tensor and other observables of the theory since their Dirac bracket with any variable vanishes) and it is the Dirac brackets that can be quantized (promoted to commutators or anti-commutators) in a quantum theory. the Dirac bracket is compatible with the constraints in the sense any dynamical variable of the theory has a vanishing Dirac bracket with a constraint. φA } − {F. G} − {F. With this. φB } ≈ C AB . we can now deﬁne the Dirac bracket between any two dynamical variables as −1 {F.46) Namely.40). φB }CBD C DA = {F.43) will be coordinate dependent and the product in (10.2 Dirac method and Dirac bracket 389 {φA . we can work with the Dirac brackets and set the constraints to zero in the theory (namely. φB }CBD {φD .43) This is an even dimensional anti-symmetric matrix and Dirac had −1 shown that it is nonsingular so that its inverse CAB exists. the matrix of Poisson brackets in (10. As a result. A −1 −1 C AD CDB = δB = CBD C DA . (10. G}D = {F. namely.

raising and lowering of indices can be carried out trivially) qi qi = 1. (10. If we denote the coordinate of the particle as qi . 10. therefore.3 Particle moving on a sphere As a simple example of constrained systems. when the number of constraints is large. the inverse needs to be deﬁned carefully with the appropriate boundary condition relevant for the problem. i = 1. This simple system is an interesting example in the study of constrained quantization and is a prototype of the ﬁeld theoretic model known as the nonlinear sigma model.49) gives the constraint (10. n. 2. · · · . rather than dealing with a large matrix of Poisson brackets in (10. It is also worth emphasizing here that in dealing with ﬁeld theories where the matrix C AB is coordinate dependent. then the matrix of highest derivatives has the form . then the coordinates are constrained to satisfy (the metric in this simple theory is Euclidean and. ∂F 2 (10.390 10 Dirac method for constrained systems that it can be constructed in stages. We note that if we combine the dynamical variables qi . F into qa = (qi . we can equivalently choose a smaller set of even number of constraints and deﬁne an intermediate Dirac bracket and then construct the ﬁnal Dirac bracket of the theory building on this structure. F ). · · · . Namely.43) and its inverse.48) where F is a Lagrange multiplier ﬁeld (an auxiliary ﬁeld without independent dynamics) whose Euler-Lagrange equation 1 ∂L = − (qi qi − 1) = 0. a = 1. ˙ ˙ 2 L= (10.47) where summation over repeated indices is understood. let us consider the motion of a point particle constrained to move on the surface of a ndimensional sphere of radius unity. 2. n + 1. We note that the dynamics of the system can be described by the Lagrangian 1 (qi qi − F (qi qi − 1)) .47) on the dynamics.

51) Therefore. ˙ ∂F pi = pF = (10.51) has the form 1 i i ˙ Hcan = pi qi + pF F − L = ˙ p p + F (qi qi − 1) . {qi . F } = {pi .50) which is indeed singular (see (10. (10. {qi . pF } = 0.22)) reﬂecting the fact that there are constraints in the theory and the naive Hamiltonian description would not hold. 2 (10.30)) 1 i i p p + F (qi qi − 1) + λ1 pF .53) Hp = Hcan + λ1 ϕ1 = (10. pF } = 1. (10.54) The equal-time canonical Poisson brackets of the theory are given by j {qi .48) and (10.10. pj } = δi . pF } = {pi .52) The canonical Hamiltonian following from (10. the theory has a primary constraint given by ϕ1 = pF ≈ 0. (10. pj } = {F.55) . 2 leading to the primary Hamiltonian (see (10. F } = {pF . To understand the passage to the Hamiltonian description in this simple case. {F. F } = {qi . we note that the conjugate momenta of the theory are given by ∂L = qi . pF } = 0. ˙ ∂ qi ˙ ∂L = 0. qj } = {pi .3 Particle moving on a sphere 391 ∂2L = ∂ qa ∂ qb ˙ ˙ δij 0 0 0 .

we obtain ϕ3 ≈ {ϕ3 .60) . = pi pj {qi .56) Let us denote this secondary constraint as ϕ2 = 1 (qi qi − 1) ≈ 0. we obtain ϕ2 ≈ {ϕ2 . 2 (10. (pj pj + F (qj qj − 1) + λ1 pF )} 2 ≈ pi pi − F ≈ 0. requiring the primary constraint to be independent of time leads to ϕ1 ≈ {ϕ1 . (10.59) Requiring the new constraint (10. Hp } ˙ 1 = {pF .57) Requiring (10. Hp } ˙ 1 1 = { (qi qi − 1). (pj pj + F (qj qj − 1) + λ1 pF )} 2 2 = qi pj {qi . In particular. 2 (10. qj } = pi pi − F qi qi (10.58) which generates a new constraint ϕ3 = qi pi ≈ 0.59) to be time independent.57) to be time independent. (pi pi + F (qi qi − 1) + λ1 pF } 2 1 = − (qi qi − 1) ≈ 0. (10. pj } = qi pi ≈ 0.392 10 Dirac method for constrained systems With these we can now determine the time evolution of any dynamical variable. pj } + qi F qj {pi . Hp } ˙ 1 = {qi pi .

ϕ2 .64) = −{ϕ3 . = −{ϕ4 . qj } − λ1 {F. ϕ3 } = { (qi qi − 1).63) where we have identiﬁed pi pi = p2 .61) 1 = {pi pi − F. ϕ4 } = {pF .62) = 2pi F qj {pi . pj pj − F } = 2pi pj {qi . 2. qj pj } = qi qj {qi . which determines the Lagrange multiplier and the chain of constraints terminates. Collecting all the constraints into φA = (ϕ1 . In fact.10. ϕ2 }. ϕ1 }. pj } = 2pi pi = 2p2 = −{ϕ4 . 3.3 Particle moving on a sphere 393 so that we can identify ϕ4 = pi pi − F ≈ 0. which takes the form (10. pi pi − F } = 1 1 {ϕ2 . (10. ϕ3 .61) to be time independent leads to ϕ4 ≈ {ϕ4 . Hp } ˙ (10. pj } = qi qi ≈ 1 2 {ϕ3 . we can calculate the matrix of the (equal-time) Poisson brackets of constraints {φA . pF } = −2F qi pi − λ1 ≈ −λ1 ≈ 0. (10.52). A = 1. Requiring (10. ϕ4 ). ϕ4 } = {qi pi .57). ϕ3 }.61) are second class. It is easy to check that all the constraints of the theory (10. (10.59) and (10. . the nontrivial equal-time Poisson brackets between the constraints take the forms {ϕ1 . φB } = C AB . 4. (pj pj + F (qj qj − 1) + λ1 pF )} 2 (10.

pj } 2 −1 1 {qi . pj }D = {qi . pF }D = 0. F }D = {qi . In particular.66) (10.68) . F } = 2pi .66).67) −1 with CAB given in (10. = −qi pj + pi qj .394 10 Dirac method for constrained systems C AB The inverse of this matrix has the form 0 −2p2 0 1 0 0 −1 0 0 −1 0 0 1 0 . φA }CAB {φB . pF }D = 0. 2 2p = 0 1 0 . (qk qk − 1)}C23 {qℓ pℓ . pF }D = {pi . pj }D = {pi . 2 {qi . qk pk }C32 { (qℓ qℓ − 1). F } − {pi . G}. pj pj − F }C41 {pF . F }D = {pi . {qi . qj }D = 0. pj } − {pi . qk pk }C32 { (qℓ qℓ − 1).45)) −1 {F. pj } 2 −1 1 −{pi . pj } − {qi .65) −1 CAB The Dirac brackets between any two dynamical variables can now be deﬁned as (see (10. F } = −2qi p2 . −1 {qi . G}D = {F. the Dirac brackets between the fundamental dynamical variables take the forms 1 −1 {pi . (qj qj − 1)}C21 {pF . F } − {qi . pj } 2 j = δi − qi qj . 0 0 (10. = 0 −1 0 2p2 2 −1 0 −2p 0 0 0 0 1 (10. {F. F }D = {pF . G} − {F. 1 −1 {pi . pF }D = {F. (10.

Let us next study the Hamiltonian description of a free relativistic massive particle which is described by the equation d2 xµ = 0.4 Relativistic particle In the last section we studied a simple example of point particle dynamics which is constrained. This is the starting point for the quantization of this theory. we can set the constraints to zero so that the true independent dynamical variables are (qi .73) where λ is a parameter labelling the trajectory and we have identiﬁed . 10.61)) and the Hamiltonian for the theory is given by (see (10.10.70) can also be written as dpµ = 0.54)) 1 i i pp.71) then the dynamical equation (10. pi ) (we note that F = pi pi because of the constraint (10. (10.70) can be obtained as the EulerLagrange equation following from the action (recall that c = 1) dλ (xµ xµ ) 2 = ˙ ˙ 1 S=m ds = m dλ L. We note that if we deﬁne the relativistic four velocity and momentum as uµ = dxµ .69) where we have used (10.4 Relativistic particle 395 With the use of the Dirac brackets.70) where τ denoting the proper time is assumed to label the trajectory of the particle and m is the rest mass of the particle.72) The dynamical equation (10. dτ pµ = muµ . (10. dτ 2 m (10. dτ (10. The constraints of this theory were all second class.62). 2 Hp = (10.

73) that ∂2L = ∂ xµ xν ˙ ˙ m xρ xρ ˙ ˙ xµ xν ˙ ˙ xσ xσ ˙ ˙ Pµν = ηµν − .75) S = m dλ dλ dxµ dxµ dλ dλ 1 2 → m = = m m dξ dxµ dξ dxµ dλ dξ dλ dξ dxµ dxµ dξ dξ 1 2 1 2 1 2 dξ dλ dλ dξ dxµ dxµ dξ dξ = S.70) or (10.77) where τ denotes the proper time associated with the particle. the action (10. dλ xµ = ηµν xν . then (10. (10.73) is invariant under the diﬀeomorphism (10.396 10 Dirac method for constrained systems xµ = ˙ dxµ . (10.72) as the Euler-Lagrange equations.75) which is an example of a local gauge transformation much like in the Maxwell ﬁeld theory.78) which leads to .73) will lead to the dynamical equation (10. ˙ ˙ (10. If we choose a gauge λ = τ. We note from (10. let us proceed with the Hamiltonian description without choosing a gauge at this point. However. so that (10. (10. dλ dλ dξ λ → ξ = ξ(λ).74) We note that under a transformation xµ = ˙ dξ dxµ dxµ = .76) Namely.

we can rewrite the action (10. Let us note that the action (10.73) to describe the motion of a massless particle. g= xµ xµ ˙ ˙ .81).79) Therefore. We note that introducing an auxiliary variable g. the action (10.81) where because of the diﬀeomorphism invariance of (10.73) in the form 1 2 √ ˙ ˙ dλ g g−1 xµ xµ + m2 . it is more convenient for us to work with the action (10.81) using (10. S= dλ L = (10. ˙ ˙ ∂g 4 or. then we obtain (10.124)) and the naive passage to the Hamiltonian description fails (see (10.81) is manifestly diﬀeomorphism invariant and the massless limit can now be taken in a straightforward manner. Viewed in this manner.73) we can think of g as the metric on the one dimensional manifold of the trajectory of the particle. We can obtain the momenta conjugate to these variables to be .73).22)). For example.82).81) are xµ and g. The two dynamical variables of the theory in (10. To see that the two actions (10. m2 (10.81) are equivalent. ˙ (10.10. However.73) and (10.80) as in the case of the Maxwell theory (see (9. it is not clear how to take the m = 0 limit in (10. Pµν is a projection operator and ∂2L = 0. we note that the equation for the auxiliary ﬁeld g takes the form g−1/2 ∂L = −g−1 xµ xµ + m2 = 0.4 Relativistic particle 397 xµ Pµν = ˙ m xρ xρ ˙ ˙ xν − ˙ xµ xµ xν ˙ ˙ ˙ xσ xσ ˙ ˙ = 0 = Pµν xν . ∂ xµ xν ˙ ˙ det Pµν = det (10.73) can also be written in an alternate form that is more useful.82) If we eliminate g in (10.

1 1 = − √ pµ pµ − m2 ≈ 0. The canonical Hamiltonian of the theory follows to be √ g µ = pµ x + pg g − L = ˙ ˙ p p µ − m2 . pg } = 0. Hp = Hcan + λ1 ϕ = 2 1 (10. g} = {pg . pg } = {pµ .84) Hcan (10.398 10 Dirac method for constrained systems pµ = ∂L = g−1/2 xµ .86) The equal time canonical Poisson brackets for the theory are given by {xµ .88) ϕ ˙ 1 ≈ {ϕ . {xµ . Hp } = {pg . pg } = 1. xν } = {pµ .83) so that the primary constraint of the theory follows to be ϕ1 = pg ≈ 0. pν } = {g. we obtain the primary Hamiltonian for the theory to correspond to √ g µ p p µ − m 2 + λ1 p g . ∂g ˙ (10. pν } = δν . ˙ ∂ xµ ˙ pg = ∂L = 0.84) and requiring the constraint to be time independent. {g. g} = {xµ . we obtain √ g µ p pµ − m2 + λ1 pg } 2 (10. (10. pg } = 0. µ {xµ .84). 4 g which leads to the secondary constraint .85) Adding the primary constraint (10.87) With these we can now determine the evolution of the primary constraint (10. 2 µ (10. g} = {pµ .

if we combine all the constraints into φA = (ϕ1 . then it follows from (10. 2 (10. Let us choose the gauge ﬁxing conditions 1 ≈ 0. χ2 } = { 1 µ p pµ − m2 .89) is the familiar Einstein relation for a massive relativistic particle (the multiplicative factor is for later calculational simplicity). m2 {ϕ1 . λ − x0 } = p0 2 (10.89) We recognize that the secondary constraint (10. As a result.90) 2 ϕ ˙ 2 so that it is time independent and the chain of constraints terminates. The non-vanishing Poisson brackets between the constraints are given by 1 } = −1 = −{χ1 . 2 2 = 0.92) that the matrix of Poisson brackets of constraints has the form . (10. which render all the constraints to be second class. g − {ϕ2 . Time evolution of the secondary constraint leads to √ g ν 1 µ 2 p p ν − m 2 + λ1 p g } ≈ {ϕ . The two constraints of the theory can be easily checked to be ﬁrst class which is associated with the fact that the theory has a local gauge invariance and is also reﬂected in the fact that the Lagrange multiplier λ1 remains arbitrary (the primary constraint is ﬁrst class). χ1 . χ2 ). ϕ1 }. Hp } = { p pµ − m . m2 (10.4 Relativistic particle 399 ϕ2 = 1 µ p pµ − m2 ≈ 0.10. ϕ2 }. χ1 } = {pg . ϕ2 .92) = −{χ2 .91) χ1 = g − χ2 = λ − x0 ≈ 0.

93) 1 0 −1 CAB The equal time Dirac bracket between any two dynamical variables is given by (see (10.95) and with (10.400 10 Dirac method for constrained systems C AB whose inverse is given by 0 0 0 0 1 p0 0 0 = 1 0 0 −p0 0 0 −1 0 0 0 p0 . pν }D = {xµ . φ1 }C13 {φ3 . pg }{g − 1 .96) pµ 0 δ . pg }D = 0.45)) −1 {F. g}D = {xµ . 0 0 0 (10. pν }D = {g. pg } {xµ . pg } = 0.86) vanishes. pν } 2 p (10. xν }D = {pµ . pν } − {xµ . 0 = −1 0 1 0 − p0 . pν } µ = δν − {xµ . pg }D = 0. µ = δν − 1 λ 1 p pλ − m2 } − 0 {λ − x0 . φ2 }C24 {φ4 . (On . pg }D = {g. pg } − {g. G}D = {F. m2 −1 {xµ . g}D = {pg . pg }D = {pµ . −1 {g. G}. G} − {F.94) (10. = {g. p0 ν With the Dirac brackets we can set the constraints to zero in which case we see that the primary Hamiltonian (10. pg } − {g.94) we can calculate the Dirac bracket between the fundamental variables to be {xµ . φA }CAB {φB . 0 0 0 0 (10. g}D = {pµ .

dτ dτ (10. therefore. that the ﬁrst of the gauge ﬁxing conditions corresponds to choosing λ = τ . where α = 1.5 Dirac field theory 401 the other hand. if we set the constraints strongly to zero. (10.22)). the free Dirac theory is described by the Lagrangian density / L = iψ∂ ψ − mψψ. 4 denote the Dirac indices and since the Lagrangian density is ﬁrst order in derivatives. or. 10. ψα . dλ dλ xµ xµ = ˙ ˙ (10.20)).10. it is obvious that the matrix of second order dervatives vanishes and the system is singular (see (10.82) that the ﬁrst of the gauge ﬁxing conditions in (10.5 Dirac ﬁeld theory As we have seen in (8. where the adjoint spinor is given by ψ = ψ† γ 0 . 3.99) (10.97) On the other hand.91) would imply dxµ dxµ = 1. (10.100) † The dynamical variables of the theory can be chosen to be ψα . dxµ dxµ = 1. 2. This manifests in the deﬁnition of the conjugate momenta as (we have chosen the convention of left derivatives) .89) in particular leads to H = p0 = p2 + m2 which can be taken as the Hamiltonian. from the deﬁnition of the inﬁnitesimal invariant length in the Minkowski space (see (1.98) It follows.) Let us also note from (10.9). we have (remember that c = 1) dτ 2 = dxµ dxµ .

α α ρα = Πα ≈ 0.101) in the intermediate steps.103) where we have used (10. The canonical Hamiltonian density of the theory (consistent with the convention of left derivatives) is obtained to be (see (10. (10.) As a result. we obtain the primary Hamiltonian for the Dirac ﬁeld theory as (see (10. Rather they correspond to the momenta conjugate to ψ and ψ † respectively.102) which in this case correspond to fermionic constraints. α α (10.402 10 Dirac method for constrained systems Π† = α Πα = ∂L † = −i(ψγ 0 )α = −iψα . The canonical Hamiltonian is now obtained to be Hcan = d3 x Hcan = d3 x −iψγ · ∇ψ + mψψ .11)) ˙ ˙† Hcan = −Π† ψα + ψα Πα − L α = −iψγ · ∇ψ + mψψ. ˙ ∂ ψα ∂L = 0. † ˙ † ˙ = iψα ψα − iψα ψα − iψγ · ∇ψ + mψψ (10.105) .30)) Hp = Hcan + d3 x φ† ξα + λ† ρα .104) Adding the primary constraints to the canonical Hamiltonian.101) (Note that Π† is not the Hermitian conjugate of Π. we obtain the two sets of primary constraints for the theory to be † ϕ† = Π† + iψα ≈ 0. (10. ˙† ∂ ψα (10.

φ† (y)} α β = † † d3 y ξβ (y){Π† (x) + iψα (x). The equal-time canonical Poisson brackets for the ﬁeld variables take the forms (see (10. Using the fact that the primary Hamiltonian is independent of time. β (10. Hcan } α = = = = † d3 y {Π† (x) + iψα (x). Π† (y)} = −δαβ δ3 (x − y) = {ψα (x). Hp } α α ≈ {φ† (x).13)(10. −iψ(y)γ · ∇y ψ(y) + mψ(y)ψ(y)} α d3 y iψ(y)γ · ∇y − mψ(y) β {Π† (x).5 Dirac field theory 403 where the Lagrange multipliers ξα . (10.10. we can now calculate the time evolution of the primary constraints (10. λ† in the present case denote α fermionic variables.14)).106) with all other brackets vanishing. ˙ φ† (x) ≈ {φ† (x).12)) † {ψα (x).106) (see also (10. ρβ (y)} . For example. ψβ (y)} α d3 y iψ(y)γ · ∇y − mψ(y) i∇ψ(x) · γ + mψ(x) α β − δαβ δ3 (x − y) . Π† (y) + iψβ (y)} α β . d3 y ξβ (y){φ† (x). Hcan } + α d3 y ξβ (y){φ† (x). Πβ (y)}. φ† (y)} α β −λ† (y){φ† (x). Using (10. we can identify x0 with the time variable of the ﬁeld operators in Hp leading to (we do not show equal-time explicitly for simplicity) {φ† (x).107) α β Let us evaluate each of these terms separately.102).

109) which determines the Lagrange multiplier λ† = − ∇ψ(x) · γ + imψ(x) α α . Πβ (y)} α β d3 y λ† (y) − iδαβ δ3 (x − y) β (10.108) = iλ† (x).111) which determines the other Lagrange multiplier ξα = γ 0 (γ · ∇ + im)ψ(x) α . ρβ (y)} α β = − = − † d3 y λ† (y){Π† (x) + iψα (x).404 = 0.112) . − 10 Dirac method for constrained systems d3 y λ† (y){φ† (x). (10. requiring the second set of primary constraints to be time independent we obtain ρα (x) ≈ {ρα (x). (10.107) and requiring the constraint to be independent of time leads to ˙ φ† (x) ≈ i∇ψ(x) · γ + mψ(x) α + iλ† (x) = 0. Hp } = {Πα (x). ψβ (y)} γ 0 (−iγ · ∇y + m)ψ(y) + iξ(y) β d3 y − δαβ δ3 (x − y) γ 0 (−iγ · ∇y + m)ψ(y) + iξ(y) α β = − γ 0 (−iγ · ∇ + m)ψ(x) + iξ(x) = 0. α Substituting these results into (10. Hp } ˙ = = † d3 y {Πα (x). α α (10.110) Similarly. (10.

ρβ (y)} = {Π† (x) + iψα (x).5 Dirac field theory 405 In this case. Π† (y) + iψβ (y)} = 0. we see that requiring the two primary constraints to be time independent does not introduce any new constraint. Πβ (y)} α α = −iδαβ δ3 (x − y) = {ρα (x). (10.115) Thus. β {ρα (x). which constitute all the constraints of the theory.110) and (10.102). we have (at equal time) † † {φ† (x). (10.114) Let us note next that the two sets of primary constraints (10. (10. Substituting (10. we have deﬁned Π = Π† γ 0 . φ† (y)}. we obtain Hp = = d3 x −iψγ · ∇ψ + mψψ + φ† ξα + λ† ρα α α d3 x −iψγ · ∇ψ + mψψ † + (Π† + iψα ) γ 0 (γ · ∇ + im)ψ α α + = − ∇ψ · γ + imψ α Πα d3 x Πγ · ∇ψ + imΠψ + (−∇ψ · γ + imψ)Π . α α β β † {φ† (x). rather it determines both the Lagrange multipliers in the primary Hamiltonian (10. we can write α the matrix of Poisson brackets as . Πβ (y)} = 0.112) into the primary Hamiltonian.10. In fact. ρα ). for simplicity of notation. φ† (y)} = {Π† (x) + iψα (x). deﬁne second class constraints (this is consistent with the fact that there is no undetermined Lagrange multiplier in the theory).113) where. ρβ (y)} = {Πα (x).105). combining the constraints into ΦA = (φ† .

G(y)} −i † d3 z {F (x).45)) {F (x). G(y)} − d3 zd3 z {F (x).116) The inverse of the matrix is easily seen to be C −1 (x. Πγ (z)}{Π† (z) + iψγ (z).406 10 Dirac method for constrained systems C(x. G(y)} . ρβ (y)} α α β {ρα (x). ρβ (y)} β 0 = −i ½ ½ 0 δ3 (x − y). (10. Recalling that constraints can be set to zero inside the Dirac bracket. ργ (z)}(iδγδ δ3 (z − z )){φ† (¯).117) The equal-time Dirac bracket of any two dynamical variables can now be easily deﬁned (see (10. γ (10. ΦA (z)}CAB (z. φ† (z)}(iδγδ δ3 (z − z )){ρδ (¯). G(y)}D = {F (x). Π† (z) + iψγ (z)}{Πγ (z). (10. G(y)} ¯ δ z = {F (x). we obtain .118) Using (10. φ† (y)} {ρα (x). y) = i 0 ½ ½ 0 δ3 (x − y).118) we can calculate the Dirac bracket between the ﬁeld variables. z ){ΦB (¯). y) = {φ† (x). G(y)} ¯ ¯ z γ +{F (x). G(y)} γ † +{F (x). G(y)} − −1 ¯ z d3 zd3 z {F (x). φ† (y)} {φ† (x). G(y)} ¯ = {F (x).

4 µ.16)). (10.118) in evaluating this bracket because Πγ has a vanishing Poisson bracket with ψα . we can set the constraints to zero in the theory. ψβ (y)} −i = −i † † d3 z{ψα (x).121) which is the Hamiltonian we have used in the quantization of the Dirac ﬁeld theory (see (8. (10. Π† (z) + iψγ (z)}{Πγ (z). ν = 0. 2. Similarly.16). ψβ (y)} γ d3 z(−δαγ δ3 (x − z))(−δγβ δ3 (z − y)) (10. the Hamiltonian of the theory (10.122) where the ﬁeld strength tensor (see (9. As we have seen in (9.113) becomes Hp = d3 x −iψγ · ∇ψ + mψψ .10. 1. ψβ (y)}D = 0 = {ψα (x). ψβ (y)}D = {ψα (x). 3. (10. Furthermore.6 Maxwell field theory 407 † † {ψα (x). Therefore. where we have neglected the last term in (10.120) and these are the relations we have used in quantizing the Dirac theory (see (8.18)). we recall that when using the Dirac brackets. we can show that † † {ψα (x). ψβ (y)}D . Maxwell’s equations can be obtained as the Euler-Lagrange equations from the Lagrangian density 1 L = − Fµν F µν .6 Maxwell ﬁeld theory Let us next consider the Maxwell ﬁeld theory which as we know exhibits gauge invariance.119) = −iδαβ δ3 (x − y). 10.4)) .

(10. Fij = −ǫijk Bk . ˙ Π = E = −(A + ∇A0 ). As a result of the gauge invariance of the theory. 2. µ (10.128) We can now obtain the canonical Hamiltonian density of the theory to have the form .126) so that we have Π0 = 0. (10.123) is the four dimensional curl of the four vector potential and contains the electric and the magnetic ﬁelds as its components (see (9. k = 1. 3.5) and (9. the matrix of quadratic derivatives becomes singular (see (9.127) This determines that the theory has a primary constraint given by ϕ1 (x) = Π0 (x) ≈ 0.125) where α(x) denotes the local parameter of gauge transformation. ˙ ∂ Aµ Πµ = (10.122)) and this manifests in the deﬁnition of the conjugate momenta as ∂L = −F 0µ .124) We also know that Maxwell’s theory is invariant under gauge transformations Aµ (x) → A′ (x) = Aµ (x) + ∂µ α(x).408 10 Dirac method for constrained systems Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . (10. j. i.7)) F0i = Ei . (10.

Πν (y)}.10.127). Πν (y)} = δµ δ3 (x − y) = −{Πν (y). 2 (10. therefore. 2 (10.130) where we have neglected a surface term (total divergence) in integrating the last term by parts. Aµ (x)}. ν {Aµ (x).132) Using these we can calculate the time evolution of the primary constraint (10. Aν (y)} = 0 = {Πµ (x). is given by Hcan = = d3 x Hcan d3 x 1 2 (Π + B2 ) − A0 ∇ · Π .131) where λ1 denotes the Lagrange multiplier of the theory. = 2 (10.128) Hp = Hcan + = d3 x d3 x λ1 ϕ1 1 2 (Π + B2 ) − A0 ∇ · Π + λ1 Π0 . The equal-time canonical Poisson brackets of the theory are given by {Aµ (x).128) and requiring the constraint to be time independent we obtain (the Poisson brackets are at equal-time which can be achieved using the time independence of Hp ) .129) where we have used (10. (10.6 Maxwell field theory 409 1 ˙ ˙ Hcan = Πµ Aµ − L = −Π · A + Fµν F µν 4 1 = −Π · (−Π − ∇A0 ) + (−E2 + B2 ) 2 1 2 (Π + B2 ) + Π · ∇A0 . The canonical Hamiltonian. The primary Hamiltonian of the theory is now obtained by adding the primary constraint (10.

the Lagrange multiplier λ1 remains as yet undetermined (there is one primary constraint which is ﬁrst class). Let us choose . (10. Hp } ≈ 0.410 10 Dirac method for constrained systems ϕ1 (x) ≈ {ϕ1 (x). as we have seen. According to the Dirac procedure. In this case. ˙ so that the chain of constraints terminates. Therefore.133) = ∇ · Π(x) ≈ 0.134) It can now be checked that the secondary constraint is time independent. We see that Maxwell’s theory has two constraints ϕ1 (x) = Π0 (x) ≈ 0. ϕ2 (x) = ∇ · Π(x) ≈ 0. (Π2 (y) + B2 (y)) 2 − A0 (y)∇ · Π(y) + λ1 (y)Π0 (y)} ≈ − d3 y {Π0 (x). Hp } ˙ = 1 d3 y {Π0 (x). This is consistent with our earlier observation that not all Lagrange multipliers in the primary Hamiltonian are determined when there are ﬁrst class constraints present. ϕ2 (x) ≈ {ϕ2 (x). in the presence of ﬁrst class constraints we are supposed to add gauge ﬁxing conditions to convert them into second class constraints. A0 (y)}∇ · Π(y) (10. we have a secondary constraint in the theory given by ϕ2 (x) = ∇ · Π(x) ≈ 0.136) and it is clear that both these constraints are ﬁrst class constraints. ﬁrst class constraints signal the presence of gauge invariances (local invariances) in the theory which we know very well in the case of Maxwell ﬁeld theory. (10. Furthermore.135) (10. namely. Hp } = {∇ · Π(x).

we can calculate the matrix of the second class constraints from the fact that the only nonvanishing equal-time Poisson brackets between the constraints are given by {ϕ1 (x). χ2 (y)} = {∇ · Π(x).136) and (10. Therefore. ∇ · A(y)} = −δ3 (x − y) = −{χ1 (x). α = 1.6 Maxwell field theory 411 χ1 (x) = A0 (x) ≈ 0. (Π)j (y)} = (∇x )i (∇y )j (δij δ3 (x − y)) 2 = −∇x δ3 (x − y) = −{χ2 (x). ϕ2 (y)}. χ1 (y)} = {Π0 (x). ϕ1 (y)}. Combining the ﬁrst class constraints as well as the gauge ﬁxing conditions into φA = (ϕα . χ2 (x) = ∇ · A(x) ≈ 0. y) = {φA (x). χα ). 2. relations (10. = (∇x )i (∇y )j {(Π)i (x). (10.140) .10. = −1 0 0 0 −2 0 −∇x 0 0 (10.138) which leads to (x0 = y 0 ) C AB (x. = 1 0 0 0 2 0 ∇x 0 0 (10.136) into second class constraints. A0 (y)} {ϕ2 (x).137) determine all the constraints of the theory. φB (y)} 0 0 −1 0 2 0 0 0 −∇x 3 δ (x − y). (10. y) −2 0 0 0 ∇x 3 δ (x − y).139) The inverse of this matrix is easily calculated to be 0 0 1 0 −1 CAB (x.137) as the gauge ﬁxing conditions which clearly convert the constraints (10.

with the usual asymptotic condition that ﬁelds vanish at spatial inﬁnity. φA (z)}CAB (z. G(y)}D = {F (x). the relevant Dirac brackets for the Maxwell ﬁeld theory follow from (10. (10.143) to be {Ai (x). Πν (y)} ¯ ν 0 ν = δµ δ3 (x − y) − δµ δ0 δ3 (x − y) − d3 zd3 z δµ δj ¯ i ν −2 ×((∇z )i δ3 (x − z))(∇z δ3 (z − z ))((∇z )j δ3 (¯ − y)) ¯ z ¯ ν 0 ν i ν = (δµ − δµ δ0 ) + δµ δj (∇x )i 1 3 j 2 (∇x ) δ (x − y). G(y)} − −1 ¯ z d3 zd3 z {F (x).141) and.142) ¯ 1 1 3 2 δ (x − y) = − 4π|x − y| . the formal inverse of the Laplacian (Green’s function) has the explicit form −2 ∇x δ3 (x − y) = The equal-time Dirac bracket between any two dynamical variables can now be calculated from {F (x). Aj (y)}D = 0 = {Πi (x). ∇z · Π(z)}(∇z δ3 (z − z )){∇z · A(¯). Πν (y)}D = {Aµ (x). Πj (y)}D .136) and (10. Πν (y)}D . it leads to {Aµ (x). Πj (y)}D = δi + (∇x )i 1 3 j 2 (∇x ) δ (x − y) ∇x (10. d3 zd3 z {Aµ (x). for the dynamical ﬁeld variables. in particular. ∇x (10. j {Ai (x). . Π0 (z)}(δ3 (z − z )){A0 (¯).412 10 Dirac method for constrained systems where. Πν (y)} ¯ ¯ z {Aµ (x). z ){φB (¯).143) Since we can now set the constraints (10.144) j = δi TR (x − y).137) to zero. Πν (y)} − −2 ¯ z +{Aµ (x). G(y)}. Aν (y)}D = 0 = {Πµ (x). ∇x (10.

1 (1951). K. Hanson. Physical Review 83. M. we obtain the Hamiltonian for the theory (10. New York (1964). T. setting the constraints (10. Canadian Journal of Mathematics 2. Yeshiva University. Barcelos-Neto. Roma (1976).131) to be d3 x 1 Π2 + B2 . L. 269 (1987).10. P. 3. 2. 10. A.25)) in quantizing Maxwell’s theory. Anderson and P. Lectures on Quantum Mechanics. Scherer. Acta Physica Polonica B18. Springer-Verlag. A. J. Furthermore. M. . G.145) which is the Hamiltonian we have used in the study of Maxwell’s theory in chapter 9. Teitelboim. Dirac. Das and W. Dirac.7 References 413 which are the relations used in (9. A. J.30) (along with the ﬁrst two of (9. 3. 5. 1018 (1951). Constrained Dynamics.136) to zero. 2 Hp = (10. Berlin (1982). A. Regge and C.7 References 1. Accademia Nazionale dei Lincei. 4. Sundermeyer. P. Bergmann. Constrained Hamiltonian Systems. ibid. 6. 129 (1950).

This page intentionally left blank .

we cannot talk of an inﬁnitesimal form for such transformations. therefore. In this chapter.1 Parity The simplest of the discrete transformations is known as parity or space inversion (also known as space reﬂection or mirror reﬂection) where one reﬂects the spatial coordinates through the origin x −→ −x. By deﬁnition.1) while t remains unchanged. Note that (in three space dimensions) space inversion cannot be obtained through any rotation and. there are a few discrete symmetries which play a fundamental role in physics. however. the eﬀect of a parity transformation can be thought of as choosing a left-handed coordinate system as opposed to the conventional right-handed one. Classically. we can group objects into diﬀerent categories depending on their behavior under parity or space inversion. it is not a continuous transformation. parity (P). P (11. 415 . we will describe three such important symmetries. this represents a space-time transformation. 11. It is clear that a vector would change sign in a left-handed coordinate system. these are not continuous symmetries and. Therefore. consequently. Therefore. namely. In addition to continuous symmetries.Chapter 11 Discrete symmetries So far we have discussed mainly continuous symmetries of dynamical systems. time reversal (T ) and charge conjugation (C) where the ﬁrst two correspond to space-time symmetry transformations while the last is an example of an internal symmetry transformation. We note that even classically.

t) −→ −J(−x.1) that applying the parity transformation twice.2) and so on. t). namely. while a scalar (from the point of view of rotations. x −→ −x −→ x. A · (B × C) −→ (−A) · ((−B)×(−C)) = −A · (B × C). t) = J 0 (−x. t) −→ ρ(−x. we note from (11. t). p −→ −p. (11. L·p |p| −→ P P P P P P P P P L · (−p) L·p =− . of course. x −→ −x. This characterization carries over to the quantum domain as well. L = x × p −→ (−x) × (−p) = x × p = L. A × B −→ (−A) × (−B) = A × B. A · B −→ (−A) · (−B) = A · B. t) = ρ(x. Similarly. J 0 (x. We note from (11.4) . these variables would. P (11.3) In quantum mechanics. P P (11. In particular.2) that the volume (which behaves like a scalar under rotations) changes sign and is known as a pseudoscalar. be promoted to operators and they would satisfy the corresponding operator transformation properties. |p| |p| J(x. for example the length of a vector) is expected to remain unchanged under a parity transformation. let us note the transformation properties of some of the well known classical variables under parity. say. We see from (11.2) that even though a vector is expected to change sign under a parity transformation. the cross product of two vectors remains invariant and is correspondingly known as a pseudovector or an axial vector. returns the coordinates to their original value.416 11 Discrete symmetries A −→ −A.

1 Parity 417 so that that even classically parity operation deﬁnes a group with two elements (a group of order two).2)) and that expectation values of quantum operators behave like classical objects (Ehrenfest theorem). P −→ − P . For simplicity. P P (11. let us consider a one dimensional quantum mechanical system. simply corresponds to asking whether this group can be carried over to quantum mechanics without any obstruction.6) holds. let us assume that under the parity transformation. the parity transformation (11. (11.5) Deﬁning a symmetry in a quantum system. we know that dynamical laws of physics (such as Newton’s equation) do not depend on the handedness of the coordinate frame. 11. under space reﬂection an arbitrary quantum mechanical state transforms as . The only diﬃculty we may apprehend is in the case of spinor states since they are double valued. the classical laws of physics are invariant under parity or space inversion. For a long time it was believed that the microscopic systems are also invariant under parity. Classically. namely. of course. But a systematic analysis. we can analyze the parity transformation in (11.6) from two equivalent points of view. but we know now that there exist processes in nature which do not respect parity. First. shows that even that does not present any diﬃculty. but not the operators such that (11.6) where we have used the fact that classically vectors change sign under space reﬂection (see (11. In this case.11. ½ and P with P 2 = ½. let us assume that P represents the operator which implements the action of parity on the quantum mechanical states. as we will see.1 Parity in quantum mechanics. In other words. In this case. As is standard in quantum mechanics. Namely.1) would result in X −→ − X .1. quantum mechanical states change.

since P|x = |−x . P (11. (11. ψ P |P |ψ P = ψ|P † P P|ψ = − ψ|P |ψ . or.12) P 2 |x which shows that P2 = ½. the parity operator is unitary. (11. Since P has real eigenvalues we conclude that the parity operator is Hermitian and thus we have . It is now obvious from (11.9) δ(x − y) = x|P † P|y . P is an idempotent operator and the only eigenvalues of P are ±1. = P| − x = |x .8) Note that parity inverts space coordinates and.10) which leads to P † P = ½.7) ψ P |X|ψ P = ψ|P † XP|ψ = − ψ|X|ψ . therefore. such that ψ|X|ψ ψ|P |ψ → → P P (11.11) In other words.418 11 Discrete symmetries |ψ → |ψ P = P|ψ . (11. Furthermore. −x| − y = x|P † P|y . (11. or. (11. acting on the coordinate basis P must lead to |x → |xP = P|x = | − x .9) that xP |y P = x|P † P|y .13) Namely.

17) Let us note further that since the eigenvalues of P are ±1. if |ψ is an eigenstate of parity. ½. P P (11.16) Namely. (11. or. In other words. then P|ψ = ±|ψ . an eigenstate of parity has an associated wave function which is either even or odd depending on the eigenvalue of the parity . under a parity transformation the wave function transforms as ψ(x) −→ ψ P (x) = ψ(−x). for such a state |ψ −→ |ψ P = P|ψ = ±|ψ .11.7) that P |ψ → |ψ P = P|ψ = P = = P dx ψ(x)|x = dx ψ(x)| − x .19) Namely. (11. ψ(x) −→ ψ P (x) = ψ(−x) = ±ψ(x).1 Parity 419 P2 = P † = P = P −1 . dx |x x|ψ dx ψ(x)P|x (11.14) To see how a scalar (not a multi-component) wave function transforms under parity. we note from (11. P (11.15) so that we obtain x|ψ P = ψ P (x) = ψ(−x).18) (11.

however.24) . P ) −→ OP (X. X]+ = 0.20) ψ|X|ψ ψ|P |ψ → → P ψ|X P |ψ = − ψ|X|ψ . P = O P † XP. we may assume that the states do not transform under the parity transformation. rather under a transformation only the operators change as (this point of view may be more relevant within the context of ﬁeld theories) O −→ OP = P † OP. such that (11.14). Similarly. P ) = P † O(X. ψ|P P |ψ = − ψ|P |ψ . PX + XP = 0. namely.23) In general.22) where we have used (11. (11. P ]+ = 0. the corresponding wave function need not have any such symmetry behavior.420 11 Discrete symmetries operator for that state. A general state of the system. does not have to be an eigenstate of the parity operator and.6) holds. P † P P (11.21) It follows from this that P X −→ X P = P † XP = −X. one can show that for any operator in this simple one dimensional theory. [P. O(X. (11. or. we can obtain that [P. −P ). P P (11. therefore. In the other point of view. or. (11. P )P = O(−X.

P ) −→ P † H(X. (11. namely. P )] = 0. H] = 0. indeed.28) .26) that this is.27) where U (t) is know as the time evolution operator.29) (11. −P ). P )P = H(−X. what happens. which are eigenfunctions of the Hamiltonian are even or odd functions of x depending on the value of the index n. the eigenstates of the Hamiltonian would naturally decompose into even and odd states. (11. P and H can be simultaneously diagonalized and the eigenstates of the Hamiltonian would carry speciﬁc parity quantum numbers.1 Parity 421 We conclude from our general discussion of symmetries (in earlier chapters) that a quantum mechanical theory would be parity invariant if the Hamiltonian of the theory remains invariant under the transformation. or.25) As a result. Therefore. −P ) = H(X. (Recall that the Hermite polynomials. It is obvious from simple systems like the one dimensional harmonic oscillator 1 P2 + mω 2 X 2 = H(−X.) If the Hamiltonian H of a quantum mechanical system is time independent. [P. (11. H(X. in such a case.27) that [P. H(X. P ) = (11.11. U (t)] = 0. we see that if parity is a symmetry of a quantum mechanical system. and it follows from (11. 2m 2 P H(X. the solution of the time dependent Schr¨dinger equation o can be written as |ψ(t) = e−iHt |ψ(0) = U (t)|ψ(0) . then as we have seen [P. P ). If parity is a symmetry of the theory. Hn (x).

takes the form P= dx | − x x|. (Recall that N¨ther’s theorem holds for o continuous symmetries. ψ P (x) = P † XP = x|ψ P = ψ(−x). Before we go on to the discussion of parity in quantum ﬁeld theories. dxdy | − x x| − y y| dxdy δ(x + y)| − x y| dy |y y| = ½. (11. (11. we cannot construct any function of X and P which would represent the parity operator. in this one dimensional theory. a nontrivial representation of the parity operator is still possible and.30) so that an even parity state would continue to be an even state under time evolution just as an odd parity state would remain an odd state.) However. let us note that in this simple one dimensional quantum theory.422 As a result. This merely corresponds to the fact that classically there is no dynamical conserved quantity associated with this discrete symmetry.31) It is obvious from this deﬁnition that P† = P2 = = = |ψ P dx |x −x| = dx | − x x| = P. we conclude that 11 Discrete symmetries P|ψ(t) = PU (t)|ψ(0) = U (t)P|ψ(0) . = P|ψ = or. dx | − x x|ψ = dx ψ(x)| − x . This is another way to say that parity is conserved in such a theory. dxdy | − y y|X| − x x| .

Namely. P (11. Therefore. multi-component operators (recall the Dirac ﬁeld or the Maxwell ﬁeld). As we have seen. Let us now go over to relativistic quantum systems.11. the properties of the operator P such as in (11. Furthermore.35) for transformation of wavefunction). so that it truly gives a representation of the parity operator.1 Parity 423 = = − = −X dxdy (−xδ(x + y))| − y x| dx x|x x| = − dx X|x x| dx |x x| = −X. We note that under a parity transformation a multi-component object would transform as P β ψα (x. not only does x → −x. The important thing to note here is that such a representation is always nonlocal and not very useful from a practical point of view.33) β where the matrix Sα can. t). mix up the diﬀerent components of the object. such systems can be described consistently only in the language of quantized ﬁelds which are.32) d |x dx = −P. from the fact that two space inversions are equivalent to leaving the object unchanged.14) are more important. we are led to the requirement that . P †P P = = = dxdy | − y y|P | − x x| dxdy − i dx − i d δ(x + y) | − y x| dy x| = −P dx |x x| (11. under a parity transformation. t) −→ ψα (x. but the components of the object may mix with one another. On the other hand. in general. in principle. we need to generalize our discussion of parity transformation to multicomponent objects. We have already seen this in the case of Lorentz transformations (see (3. t) = ηψ Sα ψβ (−x.

so is φ(−x. t). we conclude that in this case.) Thus we see that ηψ really measures the intrinsic behavior (and not the space part) of the wave function of a particle under space inversion. Let us begin with the spin zero ﬁeld which is described by the Klein-Gordon equation ∂2 − ∇2 φ(x) = 0. (11.35). (11. (Remember that parity and the momentum operators do not commute and. the above relation represents an operator relation P β ψα (x. t) is a solution of the Klein-Gordon equation. t) −→ ψα (x. φ= (11. From our general discussion in (11. t) = φ(−x.1. the equation is clearly invariant under space inversion.424 11 Discrete symmetries 2 ηψ S 2 = ½.37) If ηφ = 1 or the intrinsic parity of the spin zero ﬁeld is positive.35) P The parameter ηψ is called the intrinsic parity of the ﬁeld ψα and denotes the parity eigenvalue of the one particle state at rest.2 Spin zero ﬁeld. t) = ηφ φ(−x. 11. (11. (11. ( + m2 )φ(x) = 0. cannot have simultaneous eigenstates unless the eigenvalue of momentum vanishes.38) P (11. Consequently. t) −→ φP (x.34) and (11. φ(x. In either of the cases. t)P = ηψ Sα ψβ (−x.34) In the language of quantized ﬁelds. t).36) depending on whether the ﬁeld is massless or massive. with ηφ = ±1. then φP (x. t). if φ(x.39) . therefore. t) = P † ψα (x. t). ∂t2 or.

(11. t) = ηφ φℓ (−x.m (π − θ. ϕ) −→ Yℓ. then φP (x. (11. π + ϕ) = (−1)ℓ Yℓ. a one particle state not at rest). we can deﬁne the total parity for the state (with orbital angular momentum ℓ) as . since parity commutes with angular momentum. t) = −φ(−x. for example.42) r −→ r. ϕ). Yℓ. t) = Furthermore.2)). ϕ −→ π + ϕ. a ﬁeld or a state may not have a well deﬁned parity value (consider.44) In such a case.11.41) φP (x. θ −→ π − θ. ℓ ℓ gℓ (r. θ. (11. θ. On the other hand. t) = φP (r. t). t). the spherical harmonics as φ(x.1 Parity 425 and we say that such a ﬁeld is a scalar ﬁeld and that the associated particles are scalar particles.40) and we say that such a ﬁeld is a pseudoscalar ﬁeld and that the associated particles are pseudoscalar particles (see.m (θ. t). we can expand such a ﬁeld (or a state) in the basis of the angular momentum eigenstates.m (θ. for example. ℓ P (11. (11. from the fact that P x −→ −x. In general. t)Yℓ. however. t) = ηφ (−1)ℓ φℓ (x. ϕ.43) so that φP (x. therefore. On the other hand. ϕ).m (θ. we have P P P (11. ϕ. if the intrinsic parity of the spin zero ﬁeld is negative or ηφ = −1. t) = ℓ φℓ (r. namely.

426 11 Discrete symmetries ηTOT = ηφ (−1)ℓ . P → P + S. B and C respectively and ℓ is the orbital angular momentum of the B-C system. if we have . P P → S + S. (11.45) If parity is a symmetry of the theory. It now follows that in a parity conserving theory.46) (11. (11. → P + P.48) S → P + P.49) will be forbidden. ηB and ηC are the intrinsic parities of the particles A. we must have ηA = ηB ηC (−1)ℓ .47) where ηA . (11. (11. Let us further note that for a complex spin zero ﬁeld. whereas processes such as (ℓ = 0) S → P + S. one spin zero particle can decay into two spin zero particles in an s-state (ℓ = 0) only through the channels S → S + S. the total parity quantum number must be conserved in a physical process and this leads to the fact that for a decay (in the rest frame of A) of a spin zero particle into two spin zero particles A → B + C.

t). This result is. As we have seen in (7.11. t) = i (φP )† (x. (11. t) = i φ† (x. t) = i|ηφ |2 φ† (−x. t).2). (11. t) − (∂ µ (φP )† (x. t) − ∂ µ φ† (x. quite general for the bosons and says that for bosons.53) P which is the correct transformation for the current four vector as we have seen in (11. t))φ(x. t) φ(x. t)∂ µ φP (x. t))φP (x. in fact. In quantum ﬁeld theory φ and φ† are associated with particles and antiparticles. the intrinsic parity of the ﬁeld φ† is the same as that of the ﬁeld φ. t)∂ µ φ(x. t) −→ J P µ (x. (φ destroys a particle or creates an antiparticle whereas φ† destroys an antiparticle or creates a particle. see discussion in sections 7. particles and antiparticles would have the same intrinsic parity.) Thus we conclude that a charged spin zero particle and its antiparticle have the same intrinsic parity.51) Under a parity transformation J µ (x.38)).(11. t) = ηφ φ† (−x. ∗ (φP )† (x. t) = ηφ φ(−x. t) = −J(x.52) which leads explicitly to (recall that |ηφ |2 = 1) JP (x. the current density has the explicit form J µ (x. J P 0 (x.29).2 and 7. t). t) − (∂ µ φ† (−x. . t). since ηφ is assumed to take only real values ±1 (see (11.1 Parity 427 φP (x.3. Let us next look at the electromagnetic current associated with a charged Klein-Gordon system. t) = J 0 (−x. t)∂ µ φ(−x. t) . (11. t) .50) On the other hand.

56) we conclude that under a parity transformation the components of the four vector potential would transform as A(x.54) under space reﬂection requires that under a parity transformation.428 11 Discrete symmetries 11. A0 (x.58) P P . ∂A ∂A − ∇φ = − − ∇A0 . we must have E(x. we would then generalize these as operator transformations A(x. t) −→ B(−x.54) we note that invariance of (11. A0 (x. P P (11. P P (11. ∂t ∂t B = ∇ × A. (11. t). from the deﬁnitions of the electric and the magnetic ﬁelds in (9. B(x. In the quantum theory. t) −→ −E(−x. t) = A0 (−x. t). ∇×B = ∂E + J. t). Furthermore. we see that whereas the electric ﬁeld would transform like a vector.1. t).2).55) Namely. the magnetic ﬁeld would behave like an axial vector if parity is a symmetry of the system. t).3 Photon ﬁeld. t) −→ AP (x. t). ∂t (11. namely. From Maxwell’s equations in the presence of sources (charges and currents) ∇ · E = ρ = J 0. t) = ηA A(−x. t) −→ −A(−x. t) = −A(−x. t) −→ A0 (−x.57) This is very much like the transformations (11. t) −→ AP 0 (x.53) of the components of the current four vector under a space reﬂection. t) = ηA0 A0 (−x. E = − (11.

61) where in the last step we have let x → −x under the integral. (11. From the parity transformation of the dynamical components. t). t) d3 x J P µ (x. t)) d3 x J µ (−x. t) = −A(−x. t) · AP (x. t)Aµ (x. Thus according to our general discussion in (11. under a parity transformation . t) − (−J(x. (11. namely A. t)Aµ (x. A0 is nondynamical). therefore. t) d3 x J 0 (−x. t)AP 0 (x. From the form of the electromagnetic interaction Hamiltonian Hem = d3 x J µ (x.1 Parity 429 Note that a physical photon has only transverse degrees of freedom (namely. t)) · (−A(x. This shows that the electromagnetic interaction is invariant under parity and. t)Aµ (−x. we see that AP (x. t) µ d3 x J P 0 (x. t)A0 (−x.11.35).4 Dirac ﬁeld. parity (quantum number) must be conserved in electromagnetic processes. the photon is a vector particle.59) from which we conclude that the intrinsic parity of the photon ﬁeld (or the one photon state) is negative.33) or (11. t)AP (x.1. t)Aµ (x. (11.60) we note that under a parity transformation Hem = → = = = = P d3 x J µ (x. t). t) d3 x J µ (x. 11. In other words. t) − JP (x. t) = Hem . Let us recall that the Dirac wave function or the Dirac ﬁeld is described by a four component object.

t). S 2 = ½. t) = ηψ Sα ψβ (−x. t) = ψ (x.430 11 Discrete symmetries P β ψα (x. see (8. P (11. t)γ 0 γ 0 S † γ 0 P = ηψ ψ(−x. t)S † γ 0 ∗ = ηψ ψ † (−x. t)γ 0 S † γ 0 γ µ Sψ(−x. α = 1. t) = ψ(x. (11. (11. we can write the transformation as ψ P (x.63) (11. t)γ 0 S † γ 0 . t) = (ψ P )† (x.68) . With this choice (which also ∗ implies ηψ = ηψ ) we note that ∗ ψ (x. t)γ µ ψ P (x. t)γ µ ψ(x. with (see (11. t)γ 0 = ηψ ψ † (−x. t)γ 0 S † γ 0 γ µ Sψ(−x.67) 2 = ηψ ψ(−x.34)) 2 ηψ S 2 = ½. t) = ηψ Sψ(−x. t). t). 3.44)) J µ (x. and under a parity transformation this would transform as J µ (x. t). we can always choose ηψ = ±1.62) In matrix notation. t) = ψ(−x.66) Let us recall that associated with the Dirac system is a conserved current (probability current or electromagnetic current. t) −→ J P µ (x. 2.64) Since we have assumed the parity eigenvalues to be ±1. 4. (11. t) −→ ψα (x. (11. t) P P (11.65) which will satisfy the above relation.

(11. t)γ 0 S † γ 0 γ i Sψ(−x. t).11. with ηψ = ±1. then the parity transformed function ψ P (x. t) = −ψ(−x. namely. Namely.68) and (11. t)γ 0 ψ(−x. ψ(x. J P 0 (x.73) (11.74) It is now quite easy to see that if ψ(x. we know the transformation properties of the current four vector under space inversion (see (11. t)γ 0 S † Sψ(−x. (11. S † S = ½. comparing (11. we obtain ψ(−x. t) also satisﬁes the Dirac equation in the new coordinates.69) Comparing (11. t). or. t) = −J i (−x. t) = ηψ ψ(−x. t) = J 0 (−x.65) we have S = S † = S −1 . t).53)). J P i (x. t) −→ ψ P (x. if we deﬁne .71) Similarly. t)γ i ψ(−x.68) and (11.69) for µ = 0. t)γ 0 .1 Parity 431 2 where we have used the fact that ηψ = 1. t) = ψ(−x.70) so that using (11.69) for µ = i.72) Thus we determine the transformation of the Dirac ﬁeld (wave function) under a space reﬂection to be ψ(x. P P P (11. (11. (11. we obtain ψ(−x. t) is a solution of the Dirac equation. S = γ 0. t) −→ ψ (x. t). On the other hand. or. t). t) = ηψ γ 0 ψ(−x.

(11. −x).432 11 Discrete symmetries y µ = (t. then ∂ − m ψ P (y) = ∂y µ (11. (11. ∂xµ (11. Since the electromagnetic interactions conserve parity.6. Let ǫ1 and ǫ2 denote the polarization vectors for the two photons while k represents their relative momentum. the two photons move with equal and opposite momentum. The polarization planes of the photons in this decay can.77) In the rest frame of the positronium. To determine the relative parity between Dirac particles and antiparticles.9). However. the two component Weyl fermions violate P. space inversion is a symmetry of the Dirac equation. let us next analyze the decay of the positronium in the 1 S state where we use the standard spectroscopic notation 2S+1 L 0 J for the states (the positronium is a bound state of an electron and a positron much like the Hydrogen with a ground state energy equal to −6. in fact. as we have discussed in section 3. the scalar transition amplitude will have the general form Mi→f = (a ǫ1 · ǫ2 + b k · (ǫ1 × ǫ2 )) ψpositronium .76) In other words. This can also be seen by noting that parity is a symmetry of the free Dirac Lagrangian density (8.78) where ‘a’ and ‘b’ are scalar functions of momentum and ψpositronium represents the wave function for the positronium.75) iγ µ iγ 0 ∂ ∂ − iγ i i − m ηψ γ 0 ψ(x) ∂t ∂x = ηψ γ 0 iγ 0 = ηψ γ 0 iγ µ ∂ ∂ + iγ i i − m ψ(x) ∂t ∂x ∂ − m ψ(x) = 0.8 eV) e− + e+ → γ + γ. be measured and it .

11. (11. for this decay we have ǫ1 · ǫ2 = 0. namely.1 Parity 433 is experimentally observed that the two are perpendicular. the total parity of the state is given by ηTOT = ηe− ηe+ . it can also be seen in an intuitive manner from the structure of the spinor functions as follows. a very general result for fermions.80) k · (ǫ1 × ǫ2 ) −→ −k · (ǫ1 × ǫ2 ) .82) and for this state to represent a pseudoscalar state.79) Consequently. we must have ηTOT = ηe− ηe+ = −1.83) In other words. the electron and the positron must have relative negative intrinsic parity. (11. the relative intrinsic parity between Dirac particles and antiparticles is negative. Since the transition amplitude is a scalar under parity while P (11. Let us choose two representative solutions (of positive and negative energy) of the Dirac equation in the rest frame (see (2. This result can be more directly seen when we study charge conjugation in the next section. However. (11. On the other hand. (11. in the rest frame of the positronium. Namely. in fact.49)) . the transition amplitude for this decay takes the form Mi→f = (b k · (ǫ1 × ǫ2 )) ψpositronium .81) we conclude that the positronium state must be a pseudoscalar under parity. Let us recall from our earlier discussions that antiparticles are related to negative energy states. This is.

namely. t)γ5 ψ(x. t) = = P P |η|2 ψ(−x. 1 1 0 0 1 1 (11. (11. t)γ 0 γ5 γ 0 ψ(−x.87) −ψ(−x. t)ψ(x. t)γ5 ψ(−x. 0 0 0 0 0 0 0 0 = −η . t). the only meaningful quantity as far as the fermions are concerned is the relative parity of a pair of fermions. t). t)ψ(−x.84) introduced in (2.3) under parity. ψ(x. .100) (or in section 3. t)γ5 ψ P (x. t) = = P P |η|2 ψ(−x. For example.85) 0 P 0 = −→ ηγ 0 0 0 0 P −→ ηγ 0 = 1 0 1 0 0 = η . t)γ 0 γ 0 ψ(−x. t) −→ ψ (x. t)ψ P (x.86) which shows that this combination behaves like a scalar under parity. we note that ψ(x. On the other hand. t) (11.434 11 Discrete symmetries ψp0 >0 ψp0 <0 This analysis also brings out another feature involving the fermions. (11. Let us next calculate the transformation properties of the sixteen Dirac bilinears ψΓ(α) ψ. angular momentum). t) −→ ψ (x. since fermions always appear in pairs in any process (because of conservation of fermion number. t) ψ(−x.

t).92)) . it is known as an axial vector (or a pseudovector).91) Since strong interactions conserve parity. the only allowed interaction in this case has the form (recall also the discussion in section 8. t). or ψγ5 ψφ. The transformation properties of the bilinears are important in constructing relativistic theories.89) Namely. t)γ5 γ 0 ψ(x. P P P P (11. the combination ψγ5 γ µ ψ behaves in an opposite way from the current four vector and consequently.88) (11. As it turns out. ψ(x. namely ψψφ. ψ(x. t)σ ij ψ(x. t)σ 0i ψ(−x. t)γ 0 ψ(−x.1 Parity 435 so that this combination behaves like a pseudoscalar under parity. t)γ5 γ 0 ψ(−x. We have already seen that the combination ψγ µ ψ behaves like a four vector. t) −→ ψ(−x. t) −→ −ψ(−x. t) −→ −ψ(−x. Finally. t)γ i ψ(−x. we can also derive that ψ(x. it was thought sometime ago that the strong force between the nucleons was mediated through the Yukawa interaction of spin zero mesons. t)γ i ψ(x. t). t). ψ(x. t).11. It can similarly be checked that ψ(x. t)σ ij ψ(−x. ψ(x. t) −→ ψ(−x. namely. Thus. the spin zero meson can be identiﬁed with the π-mesons which are known to be pseudoscalar mesons. (11.8 after (8. t). only one of the forms of the interaction can be allowed depending on the intrinsic parity of the spin zero meson. t) −→ −ψ(−x. t)σ 0i ψ(x. there are two possible Lorentz invariant interactions we can write down involving fermions and spin zero particles. P P (11. Therefore. t)γ5 ψ i ψ(x. However. t)γ 0 ψ(x. t) −→ ψ(−x. for example.90) so that the combination ψσ µν ψ behaves exactly like the ﬁeld strength tensor F µν under space inversion and hence is known as a tensor. t)γ5 γ i ψ(−x.

92) Conversely. we can determine its behavior under a parity transformation which would then tell us whether parity will be conserved in processes mediated through such interactions.94)) and is denoted by C. Furthermore.95) . (11. ∂t (11. B → −B. ∇ · B = 0.94) This discrete transformation can be thought of as charge reﬂection or charge conjugation (simply because reﬂecting the sign of the charge generates the transformations in (11. let us go back to Maxwell’s equations in the presence of charges and currents ∇ · E = ρ. ∂t ∂E + J. J → −J.93) We note that this set of equations is invariant under the discrete transformations ρ → −ρ. ∂t B = ∇ × A. since ∂A − ∇A0 .436 11 Discrete symmetries ψγ5 ψφ. ∇×E = − ∇×B = ∂B . 11. E → −E.2 Charge conjugation To understand the discrete symmetry known as charge conjugation. (11. if we know the interaction Lagrangian or Hamiltonian density for a given process. E = − classically we see that charge conjugation leads to (11.

opposite charges are assigned to particles and antiparticles.98) and we say that the charge conjugation parity (or simply the charge parity) of the photon as well as the current is −1. Therefore. In other words.11. C (11. 11.29) as . (11. For this reason. C (11. we can generalize the classical result (11.2 Charge conjugation 437 Aµ −→ −Aµ . Let us now analyze the complex Klein-Gordon ﬁeld theory which describes charged spin zero particles.96) by saying that C Aµ (x) −→ AC (x) = ηA Aµ (x) = −Aµ (x). it is not a space-time transformation. (As we have seen.) We have seen that for such a theory we can deﬁne an electromagnetic current (7.97) is invariant under this transformation. Let us also note that in a quantum theory. charge conjugation is sometimes also referred to as particleantiparticle conjugation. Note that the transformation of charge conjugation does not change the space-time coordinates and. therefore. therefore. in a quantum theory.2. charge conjugation is trivial in this theory as we will discuss shortly.1 Spin zero ﬁeld.96) It is obvious that the electromagnetic interaction Hamiltonian Hem = d3 x J µ Aµ . reﬂecting charges in a quantum theory is equivalent to saying that charge conjugation really interchanges particles and antiparticles. two charge conjugation operations (reﬂecting the charge twice) would bring back any variable to itself. C Jµ −→ −Jµ . In the quantum theory. µ C Jµ (x) −→ Jµ (x) = ηJ Jµ (x) = −Jµ (x). charge conjugation really interchanges every distinct quantum number between particles and antiparticles. Let us also note here that like parity. a real Klein-Gordon theory describes charge neutral spin zero particles and.

C (11. Therefore.99) We note that if we treat φ(x) as a classical function (even though the functions are classical. As a result. we continue to use “†” instead of “∗” so that the passage to quantum theory will be straightforward). under a charge conjugation. would deﬁne a transformation under which J µ (x) −→ i (φC )† ∂ µ φC − (∂ µ (φC )† )φC = = i|ηφ |2 φ∂ µ φ† − (∂ µ φ)φ† −i φ† ∂ µ φ − (∂ µ φ† )φ = −J µ (x). (11. (11. C C C Aµ (x) −→ −Aµ (x).2 and 7.3) that φ(x) and φ† (x) can be thought of as being associated with particles and antiparticles (φ annihilates a particle/creates an antiparticle while φ† annihilates an antiparticle/creates a particle).102) .100) with |ηφ |2 = 1.438 11 Discrete symmetries J µ = i φ† ∂ µ φ − (∂ µ φ† )φ .101) C Thus the ﬁeld transformations in (11. (11. ∗ φ† (x) −→ (φC )† (x) = ηφ φ(x).100) is the appropriate charge conjugation transformation for complex scalar ﬁeld. It follows now that under such a transformation. this further supports the fact that (11. We have already discussed (see sections 7. we have (|ηφ |2 = 1) φ(x) −→ ηφ φ† (x). then C φ(x) −→ φC (x) = ηφ φ† (x).100) would deﬁne the appropriate charge conjugation transformation for the current. ∗ φ† (x) −→ ηφ φ(x).

we have to choose an operator ordering which we have not taken care of in the above discussion. (11. All of the above discussion carries over to the quantum theory if we are careful about the operator ordering of φ and φ† .100) do not lead to the correct transformation properties for the current operator. C C (11. of course. spin zero particles interacting with photons described by 1 L = (Dµ φ)† (Dµ φ) − m2 φ† φ − Fµν F µν . As is known.2 Charge conjugation 439 Fµν (x) −→ −Fµν (x).103) Therefore. C ∗ ∗ (Dµ φ(x))† = (∂µ − ieAµ )φ† (x) −→ ηφ (∂µ + ieAµ )φ = ηφ Dµ φ. 4 (11. Dµ φ(x) = (∂µ + ieAµ )φ(x) −→ ηφ (∂µ − ieAµ )φ† = ηφ (Dµ φ)† . then under the transformation (11. For example.11. it would appear as if the ﬁeld transformations (11. can be traced to the fact that in going from a classical theory to the quantum theory. the operator ordering which works for the current involving a product of bosonic operators is to symmetrize the product (this can also be checked to be equivalent to . The problem.100) we would have J µ (x) = i φ† ∂ µ φ − (∂ µ φ† )φ −→ J Cµ (x) = i|ηφ |2 φ∂ µ φ† − (∂ µ φ)φ† = −i (∂ µ φ)φ† − φ∂ µ φ† = −i φ† ∂ µ φ − (∂ µ φ† )φ .105) C since the operators φ and φ† do not commute with ∂µ φ† and ∂µ φ respectively.99).104) is invariant under charge conjugation. the Lagrangian density of charged. we note that if we deﬁne the electromagnetic current in the quantum theory as in (11. As a result.

if a spin-zero ﬁeld theory describes charge neutral particles which are their own antiparticles. (11.106) 2 µ Jsym (x) = and it is easy to check that this current will transform properly (namely. If we symmetrize the products in the deﬁnition of the current. 2 µ Jsym (x) = (11. As we have discussed earlier. In such a case.107) µ the eﬀect of the inﬁnite constant in Jsym in (11.108) This is esthetically more pleasing and treats the vacuum in a more symmetrical way. Let us note here that the symmetrized current diﬀers from the original current merely by a constant which is inﬁnite (this is the value of the commutator).106) leads to (this should be compared with the normal ordered charge deﬁned in section 7. Note that we can also write the symmetrized current as 1 µ (J (x) − J Cµ (x)) . will change sign) under charge conjugation.109) which brings out the particle-antiparticle symmetry in a more direct way. But this has the interesting consequence that while (remember the Dirac vacuum picture) 0|J µ (x)|0 −→ ∞. (11.440 11 Discrete symmetries normal ordering the product).110) . it is described by a Hermitian ﬁeld operator φ† (x) = φ(x). we obtain i φ† ∂ µ φ + (∂ µ φ)φ† − (∂ µ φ† )φ − φ∂ µ φ† . (11.3) µ 0|Jsym (x)|0 = 0. under a charge conjugation (11.

But by analyzing various physical processes where charge conjugation is a symmetry or by analyzing the invariance of various interaction terms under charge conjugation. so that in these cases (11.115) since the charge parity of the photon is (−1). C (11. . one can show that for all Hermitian spin zero ﬁelds that we know of (scalar or pseudoscalar) ηφ = 1. we conclude that the π 0 meson must also have charge parity (+1) and similar arguments hold for other mesons. ηφ = ±1. (11. or.111) From the fact that two charge conjugation operations should return the ﬁeld to itself. If charge parity is to be conserved in this process (this corresponds to an electromagnetic process since photons are involved).11. we expect 2 ηφ = 1. (11.113) φ(x) −→ φC (x) = φ(x).112) In this case. As an example. the sign of the charge parity cannot be ﬁxed from theory. C (11. the total charge parity of the ﬁnal state is (+1).2 Charge conjugation 441 φ(x) −→ φC (x) = ηφ φ† (x) = ηφ φ(x).114) Such ﬁelds or particles are known as self-charge-conjugate and simply describe particles that are their own antiparticles. let us note that in the decay π 0 → 2γ.

we note that if we can ﬁnd a matrix C satisfying C(γ µ )T C −1 = −γ µ .118) and (11. (iγ µ (∂µ − ieAµ ) − m) Cψ = 0.2 Dirac ﬁeld. iC(γ µ )T C −1 (∂µ − ieAµ ) + m Cψ = 0. interacting with an electromagnetic ﬁeld.2. T (11. from (11. it would satisfy the equation (iγ µ (∂µ − ieAµ ) − m) ψ C (x) = 0.120) (11. we obtain i(γ µ )T (∂µ − ieAµ ) + m ψ = 0.119) that the two equations are very similar in structure. or. the equation of motion takes the form (iγ µ (∂µ + ieAµ ) − m) ψ(x) = 0. T T T (11. let us note from (11.117) Taking the transpose of the adjoint equation (11.442 11 Discrete symmetries 11. In fact. If the fermions are minimally coupled to the photons.117). Let us next analyze a system of Dirac fermions in- teracting with an electromagnetic ﬁeld. To see the relation between ψ C (x) and ψ(x). or. The adjoint equation is easily obtained to correspond to ← ψ(x) iγ µ ∂µ −ieAµ (11.116) +m = 0. (11. (11.121) .119) where we have identiﬁed the charge conjugate function ψ C as describing the positron.118) Let us recall that a positron which is the antiparticle of the electron is also a Dirac particle with the same mass but carries an opposite electric charge. Thus.118) we would have C i(γ µ )T (∂µ − ieAµ ) + m ψ = 0.

The important symmetry property that C must have is that it should be antisymmetric.) To see this.11. −(γ µ )T + = 2η µν ½.122) (11. Similarly.128) (11. (11.124) and hence from the generalized Pauli theorem (see (1. namely. we can identify (with |ηψ |2 = 1) ψ C (x) = ηψ Cψ . C −1 γ µ C = −(γ µ )T . Therefore we conclude that if we can ﬁnd a matrix C with the property C(γ µ )T C −1 = −γ µ . we can also show that σ µν C = − (σ µν C)T . (This property depends on the dimensionality of space-time. from C −1 γ µ C = −(γ µ )T . −(γ µ )T .92)). let us assume that instead C is symmetric. In this case.125) . namely. (11.2 Charge conjugation 443 which can be compared with (11.126) (11. In 4-dimension C T = −C.127) (11. it follows that there must exist a similarity transformation which relates −(γ µ )T and γ µ . C T = C. we obtain γ µ C = −C(γ µ )T = −C T (γ µ )T = − (γ µ C)T .123) That such a matrix exists can be deduced from the fact that −(γ µ )T also satisﬁes the Cliﬀord algebra.119). T (11. or.

C † C = 1) ψ(x) −→ ψ C (x) = ηψ Cψ . if we do not restrict to any particular representation. in the PauliDirac representation that we have been using. we have (with |ηψ |2 = 1.130) so that it satisﬁes (it can be veriﬁed explicitly using the properties of the γ µ matrices in our representation) C = C † = C −1 = −C T .129) Unlike the parity transformation. Therefore. Thus.444 11 Discrete symmetries so that if C T = C. ∗ ψ(x) −→ ψ (x) = −ηψ ψ T C −1 .131) (11. the structure of the matrix C depends on the representation of the Dirac matrices and. the appropriate operator ordering for the current is antisymmetrization (this is equivalent to normal ordering in this case).133) The fermion ﬁelds. there can be only six linearly independent antisymmetric matrices and C cannot be symmetric. given the classical current (see (8. it follows that C must be antisymmetric. satisfy anticommutation relations to reﬂect the Fermi-Dirac statistics that the underlying particles obey. it is easy to determine that C = γ 0γ 2. therefore. But these are 4 × 4 matrices and. however. It now follows that ψ C = ηψ Cψ = ηψ γ 0 γ 2 (γ 0 )T ψ ∗ = −ηψ γ 2 ψ ∗ . (11.44)) . T (11. In this case.132) In general. C C T (11. C T = −C. as quantum mechanical operators. (11. we can construct 10 linearly independent antisymmetric matrices.

(11. Therefore.2 Charge conjugation 445 J µ (x) = ψ(x)γ µ ψ(x).134) = 1 ψ α (x) (γ µ )αβ ψβ (x) − ψβ (x) (γ µ )αβ ψ α (x) 2 1 T ψ(x)γ µ ψ(x) − ψ T (x)(γ µ )T ψ (x) . (11. (11. Let us note here again that the antisymmetrized current diﬀers from the original current by an inﬁnite constant whose eﬀect is to make µ 0|Janti (x)|0 = 0. this current transforms properly under charge conjugation.122) in the intermediate steps.136) µ = −Janti (x). we can obtain the antisymmetrized current as µ Janti (x) = (11.137) whereas 0|J µ (x)|0 −→ ∞.138) Thus this current treats the vacuum more symmetrically which can also be seen from the fact that we can write the current in the form .135) 2 It can now be easily checked that under charge conjugation the antisymmetrized current transforms as µ Janti (x) −→ C 1 C C ψ (x)γ µ ψ C (x) − (ψ C )T (x)(γ µ )T (ψ )T (x) 2 = = = 1 T |ηψ |2 −ψ T (x)C −1 γ µ Cψ (x) + ψ C T (γ µ )T (C −1 )T ψ 2 1 T T ψ T (x)(γ µ )T ψ (x) + ψ(x) C −1 γ µ C ψ(x) 2 1 T T ψ T (x)γ µ ψ (x) − ψ(x)γ µ ψ(x) 2 (11. where we have used (11.11.

446 11 Discrete symmetries µ Janti (x) = 1 µ (J (x) − J Cµ (x)) . we obtain T C −1 γ5 γ µ C = C −1 γ5 CC −1 γ µ C = γ5 − (γ µ )T (11. . 2 (11. if we deﬁne properly antisymmetrized Dirac bilinear combinations as = − (γ µ γ5 )T = (γ5 γ µ )T . We have already seen in (11.142) 2 Therefore. = i γ 0γ 1γ 2γ 3 (11.140) − (γ 1 )T T − (γ 2 )T T − (γ 3 )T T = γ5 . It follows from this and the deﬁnition of γ5 that C −1 γ5 C = C −1 iγ 0 γ 1 γ 2 γ 3 C = iC −1 γ 0 CC −1 γ 1 CC −1 γ 2 CC −1 γ 3 C = i − (γ 0 )T = i γ 3γ 2γ 1γ 0 Similarly. (11.139) Let us next calculate how the Dirac bilinears transform under charge conjugation.141) i C −1 σ µν C = C −1 (γ µ γ ν − γ ν γ µ ) C 2 i = C −1 γ µ CC −1 γ ν C − C −1 γ ν CC −1 γ µ C 2 i − (γ µ )T − (γ ν )T − − (γ ν )T − (γ µ )T = 2 i (γ µ )T (γ ν )T − (γ ν )T (γ µ )T = 2 i ν µ = (γ γ − γ µ γ ν )T = −(σ µν )T .122) that the charge conjugation matrix C satisﬁes C −1 γ µ C = −(γ µ )T .

ψψ − ψ T ψ (11. (11.144) 2 (If we think for a moment. P. (ψψ)anti really measures the sum of probability densities (number densities) for particles and antiparticles and that should not change under charge conjugation (particle ↔ antiparticle). A.148) .146) ψγ5 γ µ ψ anti −→ − ψσ µν ψ anti In general we can denote ψΓ(α) ψ −→ ǫα ψΓ(α) ψ C anti anti . ǫα = −1. (11. for α = S.145) Similarly we can also show that ψγ5 ψ ψγ5 γ µ ψ ψσ µν ψ −→ −→ C C C anti anti anti ψγ5 ψ anti .2 Charge conjugation 447 ψΓ(α) ψ anti ≡ 1 2 ψΓ(α) ψ − ψ T Γ(α) T ψ T . (11. T.147) where the phase ǫα for the diﬀerent classes has the form 1.143) then it is clear that under charge conjugation the scalar combination would transform as C ψψ anti −→ = = 1 C C ψ ψ C − (ψ C )T (ψ )T 2 1 T |ηψ |2 ψ T C −1 Cψ − ψC T − (C −1 )T ψ 2 1 T = ψψ anti . α = V. (11. for (11. .11. .) We have already seen that ψγ µ ψ −→ − ψγ µ ψ C anti anti .

t). (11. T we obtain . t).448 11 Discrete symmetries Let us next recall from (11. t)γ 0 )T C P ηψ ηψ Cγ 0 C −1 C(γ 0 )∗ ψ ∗ (−x. t) −→ ψ (x. Next. let us note that since the charge conjugate wave function is deﬁned to be C ψ C = ηψ Cψ .73) that under parity a Dirac particle transforms as P ψ(x.150) under parity. t) −→ (ψ C )P (x. t) = ηψ γ 0 ψ(−x. T (11. P ψ(x. the antiparticle will transform as C ψ C (x.152) Of course.83).151) Here we have used the fact that γ 0 is Hermitian in our metric. t) C P −ηψ γ 0 ηψ Cψ (−x. the antiparP ticle will have the intrinsic parity −ηψ so that the relative particleantiparticle parity will be given by P P P ηψ −ηψ = −(ηψ )2 = −1.149) Since the antiparticle is described by C ψ C (x. t) = ηψ C(ψ )T (x. but this derivation is simpler and more general. t) = ηψ Cψ (x. t) P −ηψ γ 0 ψ C (−x. t) P C −ηψ ηψ γ 0 C(γ 0 )T ψ ∗ (−x. Thus P if ηψ represents the intrinsic parity of a Dirac particle. P P P (11. t) −→ ψ P (x. t) P P = = = = = C P ηψ ηψ C(ψ(−x. we have already seen this from the analysis of the positronium decay in (11. t)γ 0 . t). T T (11. t) = ηψ ψ(−x.

154) T (11.159) . / where (11.157) In other words.158) then from (11. then it follows that C γ5 ψ C = −ηψ C − ψ (11.156) T T C = ηψ Cψ = ψ C .3 Majorana fermions. (p − m)ψ C = 0. T (11.153) As a result. the antiparticle of a particle with a given handedness has the opposite handedness.119) we see that the antiparticle must also satisfy the same free equation. Let us recall that if a free Dirac particle satisﬁes the equation (p − m)ψ = 0. / (11. we have C γ5 ψ C = −ηψ Cψ = −ψ C . (11.155) On the other hand. namely. if γ5 ψ = −ψ. of course.3.6 and have used this in the construction of the standard model in section 14.11.2. (11. We have. we see that if γ5 ψ = ψ. 11.2 Charge conjugation C γ5 ψ C = ηψ γ5 Cγ 0 ψ ∗ C = ηψ CC −1 γ5 C(γ 0 )T ψ ∗ T 449 T T C C = ηψ Cγ5 (γ 0 )T ψ ∗ = −ηψ C(γ 0 )T γ5 ψ ∗ † C = −ηψ C(γ 0 )T γ5 ψ ∗ C = −ηψ C(γ 0 )T (γ5 ψ)∗ C C = −ηψ C(γ 0 )T ((γ5 ψ)† )T = −ηψ C(γ5 ψ) . already seen this in the study of neutrinos in section 3.

160) We have also seen that there are bosons in nature which are their own antiparticles. whether there are also fermionic particles that are their own antiparticles and if so how does one describe them. in fact.123) we see that a Dirac particle which is its own antiparticle must satisfy ψ = ψ C = ηψ Cψ . T 11 Discrete symmetries (11. .162) Such particles are known as self charge conjugate fermions or Majorana fermions. quite obvious.165) so that the Majorana fermions are charge neutral and hence cannot have any electromagnetic interaction. and would also satisfy the free Dirac equation (p − m)ψ = (p − m)ψ C = 0. / / (11. (11.450 ψ C = ηψ Cψ .161) We can now ask the corresponding question in the case of the Dirac particles. The answer to this question is. namely. Let us note from the deﬁnition of the antisymmetrized current in (11.135) µ Janti (x) = T 1 T ψ(x)γ µ ψ(x) − ψ T (x)γ µ ψ (x) 2 1 C ψ(x)γ µ ψ(x) − ψ (x)γ µ ψ C (x) .163) T (11. 2 = (11. These are charge neutral particles described by real quantum ﬁelds φ† = φ.164) that for Majorana fermions (ψ = ψ C ) µ Janti (x) = 0. They can be massive and will have only two independent degrees of freedom (because of the constraint in (11.162)). (11. From the deﬁnition of charge conjugation in (11.

In fact. (This is. it can in principle be complex). we can have Majorana-Weyl spinors. (11. the naive mass term would have the form / L = iψ L ∂ ψL − mψ L ψL . this implies that it cannot have a well deﬁned handedness. this would require P P ηψ = −(ηψ )∗ . (11. then χC = (ψL )C + ψL = χ. In particular for d = 2 mod 8. To answer this. however. let us note that if we try to write a Lagrangian density for a massive Weyl fermion (say left-handed). as we have seen a particle with a deﬁnite handedness will lead to an antiparticle with the opposite handedness. Since a Majorana fermion is its own antiparticle. not true if the space time dimensionality is not four.167) (11. Let us also note here that since both the Weyl and the Majorana fermions have only two degrees of freedom. then the antiparticle will have the intrinsic parity −(ηψ )∗ (although we had chosen the phase of the intrinsic parity for a Dirac fermion to be real.169) . However.168) so that we can express a Majorana fermion in terms of Weyl fermions.2 Charge conjugation 451 Secondly. the mass term would trivially vanish since (11. For a Majorana fermion then. This then raises the question that since the two component Majorana fermion can be massive whether it is possible to have a mass for the two component Weyl fermion.11.) We have also seen that if the intrinsic parity of a Dirac particle P P is ηψ . we can ﬁnd a simple relation between them.166) which can be satisﬁed only if the intrinsic parity of a Majorana fermion is imaginary. we note that if we deﬁne χ = ψL + (ψL )C . Thus we cannot talk of a Majorana particle with a given helicity.

Obviously.452 11 Discrete symmetries 1 1 ψ L ψL = ψ (1 + γ5 ) × (1 − γ5 ) ψ = 0.173) this would imply violation of lepton number which is also not seen yet. .170) This is. note that if we go beyond the conventional Dirac mass term in the Lagrangian density. It is now clear that the Majorana mass term would violate conservation of fermion number and this is what we give up in trying to write a mass term for the Weyl fermion. Similarly. a Weyl fermion cannot have mass. the mass term is not zero and we can give a mass to the Weyl fermion.172) then this would imply violation of baryon number which is not seen at present. / (11. But such an identiﬁcation is not compatible with the experimental results. the essence of the statement that a Weyl fermion is massless. 2 2 (11. of course. Such a mass therm is known as a Majorana mass term and in writing this term we have obviously given up something to which we will return in a moment. note that any majorana particle would necessarily imply some sort of violation of a fermion number which is not seen in nature. if (here n represents the antineutron) n = n. if (once again ν denotes the antineutrino) ν = ν. (11. In fact. the neutron and the neutrino are charge neutral and hence are prime candidates to be Majorana particles. Let us next ask whether there are any particles in nature which we can think of as Majorana particles.171) Clearly. if we require conservation of fermion number. Obviously. However. in this case. we can write ( we will choose C ηψ = 1) C / L = iψ L ∂ ψL + mψL ψL T = iψ L ∂ ψL − mψL C −1 ψL . namely. (11.

. mντ < 18. It follows now that ˆ C Q|Q. . = Q|Q. . = −Q| − Q. (11. . . . Similarly. mνµ < 190keV. .4 Eigenstates of charge conjugation. of course. 11. = Q| − Q. . . . let us note that if Q denotes the operator for charge (electric). so that comparing the two relations we obtain (11. On the other hand.2 Charge conjugation 453 We can now ask why anyone would then like to study such particles or mass terms. . = Q| − Q. . .11. its theoretical studies would require the study of the properties of such particles. . = QC|Q. . . It is. .2MeV) (the present trend is to quote the lowest mass bound along with the diﬀerences of mass squares for the neutrinos determined from the neutrino oscillation experiments) suggest that the neutrinos may indeed have small masses. we have also seen that a Hermitian spin zero ﬁeld has charge parity +1. we have ˆ ˆ QC|Q. This. the direct experimental limits (mνe < 2ev. up to the experiments to decide whether this is really what happens. We have seen that the pho- ton has odd charge conjugation parity.176) .2. . . . means that a one photon state is an eigenstate of charge conjugation with eigenvalue −1. . . . . Independent of whether supersymmetry is a symmetry of nature. Another reason for the study of Majorana particles is that they are fundamental in the study of supersymmetry which we will not go into. . of course.174) where . refer to other quantum numbers the state can depend on. . . . . The answer to this lies in the fact that while no one has really measured the mass of the neutrino. then acting on an eigenstate it gives ˆ Q|Q. The Majorana mass is one way to understand such small masses and can arise naturally in grand uniﬁed theories where both baryon and lepton numbers can be violated in small amounts. To understand the eigenstates of charge ˆ conjugation in general.175) (11.

ℓ = (−1)ℓ |π − π + . Q + = 0. they cannot have simultaneous eigenstates unless the charge of the state vanishes. It is easy to see that C|n C|π − p C|K 0 = |n . seen this already in the sense that the photon which does not carry charge has a well deﬁned charge parity and so does a Hermitian spin zero ﬁeld which describes charge neutral mesons. ℓ = |π + π − . Note that C|π − π + = |π + π − . Let us note. 0 (11. the charge operator does not commute with the charge conjugation operator and. consequently. not all charge neutral states would be eigenstates of charge conjugation operator. however. Similarly. On the other hand.177) In other words. or. that while all eigenstates of C must be charge neutral. . for the positronium with orbital angular momentum and spin values ℓ.454 11 Discrete symmetries ˆ ˆ C Q = −QC. ˆ C. we would have C|π − π + .180) so that such a state would be an eigenstate of charge conjugation with eigenvalue (−1)ℓ . a state such as |π − π + can be an eigenstate of C. s respectively.178) which clariﬁes the point. (11. ℓ . We have. (11.179) The ﬁnal state simply corresponds to the initial state with the two particles interchanged. of course. (11. = |π + p . = |K . Thus if the state has well deﬁned angular momentum quantum number.

consistent with our earlier discussion.181) = |e+ e− . describe the decay . 11.11. (11. s → nγ.186) Let us also note here that if we have an amplitude involving N photons shown in Fig. this is an eigenstate of charge conjugation with charge parity (−1)ℓ+s . ℓ. |e− e+ .182) then the conservation of charge parity would require (recall that the photon has charge parity −1) (−1)ℓ+s = (−1)n . ℓ.185) (11. Note. / (11. ℓ. namely. s = −(−1)ℓ (−1)s+1 |e− e+ . then the charge parity must be conserved. that upon requiring conservation of charge parity. then π 0 −→ 3γ. for example.183) This is.2 Charge conjugation 455 C|e− e+ . similarly. ℓ. If charge conjugation is a symmetry of the theory. (11. (11. ℓ.184) It would. Thus if we look at the decay of the positronium to n photons. would lead to the result that 1 = (−1)n . namely. say that if charge conjugation is a symmetry. s = (−1)ℓ+s |e− e+ . s In other words. that the positronium in the ℓ = 0 = s state decays into two photons. of course. (11.1 which can. in particular. s . the decay π 0 → nγ.

(11. As we have seen. then C † HC = H. note that if a Hamiltonian is invariant under charge conjugation.189) where U denotes the time evolution operator.456 11 Discrete symmetries nγ → (N − n)γ.190) (11. (11. and it follows from this that C † U C = U. (−1)n = (−1)N −n . Therefore. any amplitude involving an odd number of photons must vanish if charge conjugation is a symmetry. In fact.187) then conservation of charge parity would require (electromagnetic interactions are invariant under charge conjugation) Figure 11. it is also known that charge conjugation symmetry holds true in strong interactions also. electromagnetic interactions are invariant under charge conjugation. Similarly. (11. or.1: An amplitude with N photons.188) In other words. then we must have . if the strong interaction Hamiltonian respects charge conjugation symmetry. This result goes under the name of Furry’s theorem. (−1)N = (−1)2n = 1.

Namely. The simplest way to see this is to note that the weak current (see (14. We can again check with the charged weak current that + JµW K C −→ −→ P g C C √ (ηe )∗ ην ν eL γµ (1 + γ5 ) eL 2 2 g C C √ (ηe )∗ ην ν PL γµ (1 + γ5 ) eP e L 2 2 . On the other hand the product operation CP appears to be a good symmetry in most weak processes. the scattering cross-section for π − p elastic scattering would be identical to the cross-section for π + p scattering which is experimentally observed.2 Charge conjugation 457 π − p|U |π − p = π − p|C † U C|π − p = π + p|U |π + p .191) In other words.100)) g g C C + Jµ W K = − √ eL γµ (1 − γ5 ) νeL −→ − √ eC γµ (1 − γ5 ) νeL L 2 2 2 2 g C C √ (ηe )∗ ην eT C −1 γµ (1 − γ5 ) Cν TL = e L 2 2 g C C T T = − √ (ηe )∗ ην eT γµ 1 − γ5 ν TL L e 2 2 g C C √ (ηe )∗ ην (ν eL (1 − γ5 ) γµ eL )T = 2 2 g C C √ (ηe )∗ ην ν eL γµ (1 + γ5 ) eL = 2 2 g − = − √ ηj ν eL γµ (1 − γ5 ) eL = ηj JµW K . (11. Thus we see that P and C (P for parity) are conserved in strong and electromagnetic interactions but are violated in weak interactions.192) 2 2 where ηJ denotes a phase. (11. it is violated in weak processes just like parity.11. Let us note here without going into details that while charge conjugation is a good symmetry for both strong and electromagnetic interactions. the weak current does not have a well deﬁned transformation property under charge conjugation and hence will violate C-invariance.

It follows from this that + J0W K Ji+ K W g CP − −→ ηTOT √ ν eL γ0 (1 − γ5 ) eL = −ηTOT J0W K 2 2 g CP −→ −ηTOT √ ν eL γi (1 − γ5 ) eL = ηTOT Ji− K .193) 2 2 where ηTOT denotes the total phase under the transformation. Maxwell’s equations in the presence of sources ∇ · E(x) = ρ(x). T t −→ −t. namely. time reversal or time reﬂection corresponds to the space time transformation where we reverse only the sign of time.3 Time reversal Classically. (11.194) W 2 2 This has well deﬁned transformation properties under CP transformations and leads to CP invariance.458 = = = 11 Discrete symmetries g C C P P √ (ηe )∗ ην (ην )∗ ηe ν eL γ 0 γµ (1 − γ5 ) γ 0 eL 2 2 g C C P P √ (ηe )∗ ην (ην )∗ ηe ν eL γ 0 γµ γ 0 (1 − γ5 ) eL 2 2 g ηTOT √ ν eL γ 0 γµ γ 0 (1 − γ5 ) eL . Similarly. x −→ x.195) We note that the Newton’s equation is invariant under time reversal since it is second order in the time derivative. ∇ · B(x) = 0. ∇ × E(x) = − ∇ × B(x) = ∂B . 11. (11. ∂t (11. T (11.196) are also form invariant if we deﬁne . ∂t ∂E(x) + J.

11. −t).57). we know that there is a sense of direction for time. −t). however. t) = A0 (x. J(x.197) ∂A . ∂t (11. −t). The ﬁrst problem which we face. B(x. However. From the deﬁnition of the electric and the magnetic ﬁelds E = −∇A0 − B = ∇ × A. we determine that under time reversal A0 (x. Macroscopically. the entropy or disorder increases with time. in statistical mechanics we know that microscopic reversibility does hold. t) = E(x. t) = −J(x. t) −→ ρT (x.3 Time reversal 459 ρ(x. t) −→ ET (x. While some of the macroscopic equations are invariant under time reversal. t) −→ BT (x. t) = −B(x. t) = ρ(x. therefore. symmetry transformations normally are implemented by linear operators which are also unitary. Thus we would like to investigate whether time reversal can be a symmetry of quantum mechanical systems. T T T T T T (11. is opposite from their behavior under parity in (11. Note that in quantum mechanics. E(x.199) The transformation of the Aµ (x) potentials under time reversal. −t). is that the time dependent Schr¨dinger equation is ﬁrst order in the time derivative just o like the diﬀusion equation and hence cannot be invariant under a naive extension of the time inversion operation to the quantum theory. namely. t) −→ JT (x. A(x. t) −→ AT (x. Thus assuming that time inversion is also implemented by such an operator T . t) −→ A0T (x. t) = −A(x.198) (11. we will have . −t). we know that macroscopic physics is not. −t). The diﬀusion equation which is ﬁrst order in the time derivative is not symmetric under t → −t.

∂t (11. [T . H] = 0. In fact. (11.201) then it follows that under a time inversion. (11. (11. ∂t Thus we see that in such a case |ψ and T |ψ satisfy diﬀerent equations and hence this operation cannot deﬁne a symmetry of the time dependent Schr¨dinger equation. the crucial diﬀerence between the diﬀusion equation and the Schr¨diner equation is the fact that the time derivative in the o quantum mechanical case comes multiplied with a factor of “i” which makes it possible to deﬁne a time inversion in the quantum theory which will be a symmetry of the system. By deﬁnition. −i ∂|ψ ∂t = T H|ψ . let us assume that T is an antilinear (antiunitary) operator. o However.203) ∂ (T |ψ ) = H(T |ψ ).460 11 Discrete symmetries |ψ −→ |ψ T T = T |ψ . namely.202) would lead to iT or. an antilinear operator T satisﬁes . Let us for the moment give up the assumption that the operator T is linear. the time dependent Schr¨dinger equation o i ∂|ψ = H|ψ .200) If we assume that the time independent Hamiltonian is also invariant under time inversion.

T AT † = AT = ηA A. = HT |ψ .206) . ηB are phases. (11. φ|ψ −→ T φ|ψ T T |ψ −→ |ψ T T = T |ψ .205) or. the dynamical equation would transform as ∂|ψ ∂t i would lead to T or. i ∂ (T |ψ ) = H(T |ψ ). (11.3 Time reversal 461 T (α|ψ1 + β|ψ2 ) = α∗ T |ψ1 + β ∗ T |ψ2 . T ABT † = AT B T = ηA ηB AB. ∂t Therefore.204) where A and B are assumed to be Hermitian operators and ηA . we see that |ψ and T |ψ satisfy the same equation and hence such a transformation would deﬁne a symmetry of the time dependent Schr¨dinger equation.) Under such an antilinear (antiunitary) transformation. (11.11. with such a deﬁnition of T . = φ|ψ ∗ = ψ|φ . Thus the quantity we are interested in is the transition amplitude ψf |ψi . i −i = H|ψ . if the Hamiltonian is symmetric under time inversion. but we will not need this for our discussions. ∂|ψ ∂t ∂T |ψ ∂(−t) = T H|ψ . o Let us next assume that we are looking at a process which goes from the state |ψi to the state |ψf . (We can also show that T † = T −1 . T iT † = −i.

T p −→ −p. t) = x|ψ(t) −→ ψ T (x.209) Let us derive the transformation properties of some of the operators under time inversion.211). t) = x|ψ(−t) = ψ ∗ (x. the role of the initial and the ﬁnal states are interchanged and the transition amplitude of interest would be ψi |ψf . (11. we expect ψf |ψi −→ ψi |ψf . under time inversion. It is reassuring to note that the canonical commutation relations are unchanged by the ηX = 1. Pj ] . T (11.204) (see. in particular the third relation). ηP = −1. on the other hand. Pj ] T † = T (Xi Pj − Pj Xi ) T † = Xi (−Pj ) − (−Pj )Xi = − [Xi .210) Thus from Ehrenfest’s theorem we would expect the corresponding operators to transform as T XT † = ηX X = X.462 11 Discrete symmetries In the time reversed process. let us note from (11. T ∗ (11. It follows now that T [Xi .204) as well as (11. . −t).208) which is indeed consistent with the deﬁnition of time inversion transformations given in (11. (11.207) (11.211) T PT † = ηP P = −P. In particular.204) that ψ(x. Classically. we know that x −→ x. T (11. Thus.212) where we have used (11.

m . we obtain T Li T † = ǫijk T Xj Pk T † = −ǫijk Xj Pk = −Li .217) . it follows that T L2 T † = T Li Li T † = (−Li ) (−Li ) = Li Li = L2 . m is an eigenstate of L2 and L3 with eigenvalues 2 ℓ(ℓ + 1) and − m respectively so that we can identify T |ℓ. (11. m L3 |ℓ.3 Time reversal 463 time inversion transformation (note from (11. (11. (11. m = − m(T |ℓ. −m . m .204) that i → −i under the anti-linear time reversal transformation).214) so that the L2 operator remains unchanged under time inversion.215) m|ℓ. (11. (11.11. It follows now that for the angular momentum eigenstates satisfying L2 |ℓ. Therefore. m with T = Cm |ℓ. m L3 T |ℓ. ℓ(ℓ + 1)(T |ℓ. under time inversion each component of the angular momentum operator would change sign which is consistent with the classical result. m ).213) Namely. m = 2 2 = = ℓ(ℓ + 1)|ℓ. m we have L2 T |ℓ.216) = −T L3 |ℓ. Let us note here that the naive extension of the time inversion transformation with a linear operator would have led to an inconsistency in the commutation relations. m ). m = T L2 |ℓ. Since the angular momentum operators can be written as Li = ǫijk Xj Pk . We see that the state T |ℓ. m = |ℓ.

T (11.m (θ. t) = ηAµ Aµ (x. −mj . −m .218) (11.223) .3. m = (i)2m |ℓ. mj = (i)2mj |j. we can write T |j. (11. which follows because under time inversion (say for m > 0) (11. φ) = = (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. By convention. where we have used the fact that the spherical harmonics for m < 0 are deﬁned with a phase +1.220) (−1)m Yℓ.222) 11. (11. t) −→ AµT (x.−m (θ. m = T |ℓ. m (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. φ) = = T θ. We have already seen that under time reversal Aµ (x.219) Yℓ. −t).464 11 Discrete symmetries |Cm |2 = 1.m (cos θ)eimφ 4π ℓ + |m|)! ∗ −→ Yℓ. Thus. φ).1 Spin zero ﬁeld and Maxwell’s theory.m (cos θ)e−imφ 4π (ℓ + |m|)! (11.221) It can be shown that the above relation holds true even for half integer angular momentum values. we can indeed write T |ℓ.m (θ. φ|ℓ. namely. the constant (phase) is chosen to be Cm = (−1)m = i2m . mj = T |j. in general.

−t). −t) − J µ (x.224) It follows.229) = d4 x S . the current four vector transforms as J µ (x.227) ηJ 0 = 1. T T (11. −t). t) = T ∂µ Aν (x. t) − ∂ν AT (x. 1 − Fµν F µν − J µ Aµ 4 1 T µν T T − J µT AT −→ d4 x − Fµν F µ 4 1 = d4 x − Fµν (x.11. ν µ so that T F0i (x. (11. therefore. t) −→ Fij (x. t) = ηJ µ J µ (x. t) −→ F0i (x − t) = F0i (x. ηJ = −1. t) − ∂ν Aµ (x. (11. −t) F µν (x. we have seen that under time reversal. T Fij (x. of course. t) −→ J µT (x. where T (11. ηA = −1.226) Similarly. t) = −Fij (x. −t) 4 1 = d4 x − Fµν (x)F µν (x) − J µ (x)Aµ (x) 4 = S.228) The Maxwell theory is. −t)Aµ (x. (11. −t). that Fµν (x.3 Time reversal 465 where ηA0 = 1.225) T T −→ ∂µ AT (x. invariant under time reversal as we have already seen and as can be checked directly. t). t) (11.

(11. −t)) −→ i (φT ∗ (x)∂ µ φT (x) − (∂ µ φT ∗ (x))φT (x)) T = −i (φ∗ (x.231) (11. −t) ∂t2 (11. we have φ(x) −→ φT (x) = ηφ φ∗ (x.234) ∂2 − ∇2 + m2 φ(x) = 0. −t))φ(x. −t) − (∂ µ φ∗ (x. ∂t2 (11. Furthermore. −t)). note that under time inversion (assume classical behavior) J µ (x) = i (φ∗ (x)∂ µ φ(x) − (∂ µ φ∗ (x))φ(x)) = i (φ(x. −t).232) That is. −t))φ∗ (x.230) where the matrix T mixes the components of the function and |ηψ |2 = 1. ∂t2 ∂2 − ∇2 + m2 φ∗ (x) = 0. T (11.466 11 Discrete symmetries In general. For example. Note that since the complex spin zero ﬁeld satisﬁes + m2 φ(x) = + m2 φ∗ (x) = it is clear that in this case + m2 φT (x) = = = 0.233) T (11. −t). the Klein-Gordon equation is invariant under time inversion. ∂2 − ∇2 + m2 φT (x) ∂t2 ∂2 − ∇2 + m2 ηφ φ∗ (x.235) . −t) − (∂ µ φ(x. −t)∂ µ φ∗ (x. for a multicomponent function or ﬁeld we expect the behavior under time reversal to be T β ∗ ψα (x) −→ ψα (x) = ηψ Tα ψβ (x. for a spin zero ﬁeld. −t)∂ µ φ(x.

(Here ∂ we have used the fact that J 0 (x. −t)∂ i φ(x.230)). −t) = which is.2 Dirac ﬁelds. t) −→ ψ T (x. we obtain J 0 (x. −t))φ(x. leads to (upon complex conjugation) (−iγ µ∗ ∂µ − m) ψ ∗ (x) = 0. −t) − ( φ∗ (x. T (11. −t) is deﬁned with − ∂t .236) J i (x.) Thus we conclude that the spin zero charged particles interacting with photons can be described by a theory which will be time reversal invariant. ∂t ∂x ∂ ∂ − iγ i∗ i − m ψ ∗ (x. t) −→ −i φ∗ (x. −iγ 0∗ −iγ 0∗ iγ 0∗ ∂ ∂ − iγ i∗ i − m ψ ∗ (x.237) To determine when this function will be a solution of the Dirac equation (so that time reversal can be a symmetry of the system) we note that the Dirac equation (iγ µ ∂µ − m) ψ(x) = 0. −t). −t) ∂t ∂t J 0 (x.238) ∂ ∂ − iγ i∗ i − m ψ ∗ (x.197). (11. or.239) (11.11. 11. −t) = 0. −t). or. we expect that under time reversal ψ(x. −t) − (∂ i φ∗ (x.3. From our general discussion (see (11. t) −→ −i φ∗ (x. or. t) = 0. −t) = T T ∂ ∂ φ(x. −t))φ(x.3 Time reversal 467 Therefore. of course. −t) = 0. −J i (x. −t). the correct behavior of the current four-vector under time inversion as we have already seen in (11. ∂(−t) ∂x (11. t) = ηψ T ψ ∗ (x. ∂t ∂x .

(11.244) (11. γ ˜ (11. and the matrices γ µ deﬁned as (˜ µ = γ µ† ) ˜ γ (11. γ ν∗ } = 2η µν ½. if ψ T (x. or. we should have (iγ µ ∂µ − m) ψ T (x) = 0. −t). t) were to be a solution of the Dirac equation. iγ 0 iγ 0 ∂ ∂ + iγ i i − m ψ T (x) = 0. T γ i∗ T −1 = −γ i . ∂t ∂x ∂ ∂ + iγ i i − m T ψ ∗ (x.245) .241) ψ T (x) = ηψ T ψ ∗ (x.468 11 Discrete symmetries On the other hand.243) γ 0 = γ 0. γ ν } = 2η µν ½. −t) = 0. or. ˜ γ i = −γ i . We note that the matrices γ µ∗ satisfy the Cliﬀord algebra {γ µ∗ . t).242) would satisfy the same Dirac equation as ψ(x.240) Therefore. ∂t ∂x (11. comparing the two equations we see that if we can ﬁnd a matrix T such that T γ 0∗ T −1 = γ 0 . then the time reversed wave function (11. ˜ also satisfy the Cliﬀord algebra {˜ µ .

(11.11. T γ i∗ T −1 = −γ i . Combining the two results we obtain (11.251) In other words.250) CT −1 γ µ T C −1 = Cγ µT C −1 = −γ µ .122) that Cγ µT C −1 = −γ µ . CT −1 γ µ CT −1 −1 = −γ µ . Thus we can identify T γ µ∗ T −1 = γ µ . Choosing . (11. T = −γ i∗ = − γ i† = γ iT . They must also be related to each other through a similarity transformation. (11. ˜ leading to T γ 0∗ T −1 = γ 0 .248) Thus we see that the inverse relation can be written as T −1 γ µ T = γ µT . the matrix CT −1 anticommutes with all the γ µ ’s and hence must be a multiple of γ5 .249) (11.246) (11.3 Time reversal 469 Thus both γ µ∗ and γ µ must be related to γ µ through similarity trans˜ formations (by Pauli’s fundamental theorem). let us invert these relations and write T T −1 γ 0 T T −1 γ i T = γ 0∗ = γ 0† = γ 0T .247) To determine the explicit form of T . or. Let us also recall from (11.

−t)T −1 γ 0 .254) (11. we determine (in our representation) T = iγ 1 γ 3 . the T matrix satisﬁes T = T † = T −1 . as in the case of charge conjugation. Therefore we see that the time inversion transformations T ψ(x. and using the deﬁnition (11. t) −→ ψ (x. Let us next analyze the behavior of the current four-vector under time inversion .257) with (|ηψ |2 = 1) deﬁnes a symmetry of the Dirac equation. −t) = ηψ γ 0 T ψ (x. t) −→ ψ T (x. (11.256) so that the time inversion matrix in the spinor space is antisymmetric. t) = ηψ T ψ ∗ (x.255) Furthermore. T T T ∗ ψ(x. −t) = T ηψ T γ 0T ψ (x. (11. independent of any representation (the superscript denotes transposition) (T )T = −T. −t). t) = ηψ ψ T (x.470 11 Discrete symmetries CT −1 = −γ5 = −iγ 0 γ 1 γ 2 γ 3 . Note that with this choice.252) (11.130) C = γ 0γ 2. (11. we can show that in four space-time dimensions.253) (11.

−t) = = T T T ψ T (x. −t) ψ(x. −t)T −1 γ 0 γ µ γ 0 T ψ (x.11. −t)γ i ψ(x. we have .259) which is the behavior of the current four vector under a time inversion and we can identify the transformations in (11. −t) T T = −J i (x.260) (this is very much like the charge conjugation in (11. −t)T −1 γ 0 T ψ (x. (11. −t). −t)T −1 γ 0 γ µ γ 0 T ψ (x. let us note that since T −1 γ µ T = γ µT . −t)γ i ψ (x. −t)T −1 γ 0 γ i γ 0 T ψ (x.122) except for the −1 sign). (11. we also see that a Dirac system interacting with photons would be invariant under such a transformation. −t) ψ T (x. −t). −t)γ 0 ψ(x. −t). −t)γ 0 ψ (x. Before we examine the transformation properties of Dirac bilinears under time inversion. T J i (x. −t) T T T T = J 0 (x. (11. −t) = = −ψ T (x.257) to correspond to time inversion for a Dirac system. Furthermore. −t) − ψ(x.258) T J 0 (x) −→ ψ T (x. t) −→ ψ T (x.3 Time reversal 471 J µ (x) = T ψ(x)γ µ ψ(x) T −→ ψ (x)γ µ ψ T (x) = = so that we have T |ηψ |2 ψ T (x.

the 0 transformation of J5 = −ψ † γ5 ψ shows that under time inversion . −t) ψ(x. we can show that T ψ T (x. t) −→ ψ (x.262) ψ(x. t)ψ(x. −t)T −1 γ 0 γ 0 T ψ (x. ψ(x. −t)ψ(x. t) −→ −ψ(x. t)γ5 ψ(x.472 11 Discrete symmetries T −1 γ5 T T −1 γ5 γ µ T T −1 σ µν T T = γ5 . t) −→ ψ(x. −t)ψ(x. t)ψ T (x. −t)γ5 γ i ψ(x. (11. −t). −t). −t) −ψ(x. −t)γ5 ψ(x. T T (11. −t).261) The behavior of the Dirac bilinears under time reversal can now be obtained in a simple manner. −t). t)γ5 ψ 0 ψ(x. −t)γ5 ψ(x. t) = = = T ψ T (x. T = γ5 γ µ = (γ µ γ5 )T = − (γ5 γ µ )T . T T ψ(x. 2 (11. −t)γ5 γ 0 ψ(x. t)γ5 ψ i ψ(x. T = = = i −1 µ ν T (γ γ − γ ν γ µ ) T 2 i µT ν T T γ γ − γ ν γ µT 2 i ν µ T (γ γ − γ µ γ ν )T = −σ µν .263) Namely. −t)T −1 γ 0 γ5 γ 0 T ψ (x. T T T ψ(x. t)γ5 ψ T (x. In particular. −t) ψ(x. t) −→ ψ (x. −t) − ψ(x. t) = = = Similarly. the axial vector (current) behaves exactly the same way as the vector (current) under a time inversion.

T T (11.222)) T |α.268) . namely. −t)σ ij ψ(x. −t). (11. a state with given angular momentum transforms as (see (11.121)) s·p s · p T (−s) · (−p) −→ = = h. t) −→ −ψ(x. 11.3 Consequences of T invariance. t)σ 0i ψ(x.3 Time reversal 473 the helicity does not change. j. It is also easy to see now that under time reversal. then the transition amplitude can be represented as (11. t)σ ij ψ(x.3. p. the chirality of a spinor does not change. Let us also note that if we are looking at a reaction of the form a + b −→ c + d. (11. ψ(x. T = ηψ γ5 T ψ ∗ = ηψ T γ5 ψ ∗ † = ηψ T γ5 ψ ∗ γ5 ψ T = ηψ T (γ5 ψ)∗ .267) where α stands for all the other quantum numbers of the state. −mj .266) so that a left-handed spinor remains left-handed under time reversal and a right-handed spinor remains right-handed.264) We can also show that ψ(x. Let us recall that under time reversal. We could have expected this from the deﬁnition (see (3. j. −t)σ 0i ψ(x. −p. the tensor bilinears have precisely the same behavior under time reversal as the electromagnetic ﬁeld strength tensor F µν . mj −→ (i)2mj |α.11. −t). |p| |p| |p| h= (11. t) −→ ψ(x.265) In other words.

(11. if we are looking at the reciprocal reaction c + d −→ a + b.269) where |i stands for the initial state.271) (11. d .273) On the other hand. for example.272) (11. it does not follow a priori that | i|S|f |2 = | f |S|i |2 . S † = S. It is worth emphasizing here that the detailed balance does not say that the rates for the two reactions a + b −→ c + d. |i = |a. (11. to obtain the spin of π-meson etc.274) (11.270) Since the time evolution operator is not Hermitian. |f for the ﬁnal state and S denotes the S-matrix or the time evolution matrix. b . namely. This is known as the principle of detailed balance and has been used. and c + d −→ a + b.474 11 Discrete symmetries f |S|i .275) . then the transition amplitude can be represented as i|S|f . where we are still using the same notation. On the other hand. (11. (11. namely. |f = |c. experimentally this is seen to hold in many processes.

we can write (for distinct |i and |f states) f |S|i = −2πi f |Hint |i δ (Ei − Ef ) . It follows then (in the operator formulation) that the evolution operator or the S-matrix must also be invariant under time inversion. T ST † = S. = |β. That is. h (11. (11. In this case. (11. mi . Let us next try to understand. when the principle of detailed balance may hold. mf . therefore. Let us next assume that we are dealing with a system whose Hamiltonian is invariant under time reversal. from Fermi’s Golden rule we know that the rate for a process |i → |f is given by Wi→f ∝ 1 | f |S|i |2 ρf . pi . theoretically. Let us assume that we are dealing with a system where the interaction Hamiltonian can be treated as a perturbation.3 Time reversal 475 have to be the same. the principle of detailed balance will hold. then we have | f |S|i | = 2π| f |Hint |i |δ (Ei − Ef ) = 2π| i|Hint |f |δ (Ei − Ef ) = | i|S|f |.277) and furthermore.11. (11.279) Let us choose the initial and the ﬁnal states to correspond to |i |f = |α. pf . In fact. if the interaction Hamiltonian is Hermitian.276) where ρf denotes the density of states for the ﬁnal state which may not be the same for the two processes.278) In this case. (11.280) .

mf | = | α. as we have seen in (11. we have | β. −mf . pi . −pf . in addition. −pi . −mi |S|β. mf |S|α. mi = | α. mi = ± α. −mf |. mf |S|α. pf . −mf . −mi |S|β. time reversal invariance would imply that f |S|i β.476 11 Discrete symmetries where mi and mf denote the spin projections for the initial and the ﬁnal state respectively (corresponding to the same angular momentum).283) This is known as the reciprocity relation and holds for any system where time reversal invariance is a symmetry.222) T |i T |f = (i)2mi |α. pi . But this is not the same as the detailed balance. if we square this relation and sum over all the spin projections mi . pf . mi |P −1 PSP −1 P|β. (11. mi = = = (i)2mi −2mf α. or. (11. pi .282) where we have used (the third) relation in (11. mf |S|α.204). the phase factor can only take values ±1 and we have β. pf . mi | = | α. −mi |S|β. mf |S|α. pf . mf .281) so that in this case. the system is invariant under parity. Let us note that if. −pf . −pf . −mf . mi | β. pf . mf |T −1 T ST −1 T |α. −pf . pi . (11. −mi |S|β. −pi . | β. −pi . −mi . Under time reversal. pi .284) Therefore. −mf |. pf . −pi . Since the transitions can only be from integer spin states to integer spin states or half integer spin states to half integer spin states. pi . = (i)2mf |β. we obtain . −pi . (11. −pf .

Therefore. then from P and T invariance we have β. −m|S|β.3. the principle of detailed balance will hold. it has . The stringent limits on T -invariance in electromagnetic interactions comes from the measurement of the electric dipole moment of the neutron.mf | β. pi . pi . m = α. pi . pf . m = α. (11. m|S|β.3 Time reversal 477 mi . pf . P is violated – but the strength of the interaction Hamiltonian is assumed to be weak. Note that if we are looking for a subclass of processes where mi = mf = m. m|S|α. Let us recall that although the neutron is electrically neutral. −m . pf . (11. it is obvious that the principle of detailed balance or the semi-detailed balance must hold in practically all processes. mf |S|α. On the other hand. pi .mf This is very similar to the principle of detailed balance but not quite.285) = mi . mi |2 | α.11. Thus in this case we will have β.287) (11. strong and electromagnetic interactions are assumed to be P and T symmetric. m|S|α. since rotations are assumed to be a symmetry of physical systems. we can make a rotation on the right-hand side to bring the state | − m → |m . pi . Note that in weak interactions. mf |2 . pf . This relation is known as the principle of semi-detailed balance and holds true in every system for which P and T are symmetries. therefore. m (11. pf . mi |S|β. pi .4 Electric dipole moment of neutron.286) Furthermore. 11.288) In this case. pf . Time reversal invariance holds extremely well in strong and electromagnetic interactions.

this is the Pauli coupling) Hint = µe σ · E + µm σ · B.293) This basically measures the extent to which P and T invariances hold in electromagnetic interactions. T T T (11. (µe )exp < 0. σ −→ σ. σ · B −→ σ · B.289) Let us also note here that the only natural dimension associated with the neutron is its size which is of the order of 1F = 10−13 cm. Consequently. (11.478 11 Discrete symmetries a magnetic dipole moment. Let us assume that it also has an electric dipole moment. T T (11.291) σ · E −→ −σ · E. so that we have σ · E −→ −σ · E. B −→ −B. the electric dipole interaction changes sign under each one of the transformations. Thus the magnitude of any dipole moment (electric) will be of the order of µe ≃ ed ≃ e 10−13 cm = 10−13 e-cm. P P P P P E −→ E.292) In other words. whereas the interaction involving the magnetic dipole moment is invariant under P and T . (11. Thus the neutron would interact with an electromagnetic ﬁeld through the interaction Hamiltonian (there is no minimal coupling. B −→ B.290) We have already seen that under parity and time reversal E −→ −E. (11.63 × 10−25 e-cm. then the electric dipole moment of the neutron must vanish. . the electric dipole moment would be parallel to its spin. Since there is no natural axis for the neutron other than the spin axis. Experiments have obtained extremely small limits on µe . if electromagnetic interactions are invariant under P and T . σ −→ −σ. σ · B −→ σ · B.

−s = m. m. The operation of CP. −s|H|α. Thus we see that m = = = α. then it follows that m = α. (11. quantities with bars would refer to antiparticle parameters).4 CPT theorem We have seen that some of the discrete symmetries appear to be violated in weak interactions.297) . Thus under the CPT transformation. m. m. s . m. s α. s|H|α. C and T in any order even though the individual transformations may not be a symmetry of the system.294) where m denotes the mass of the particle. |α. m. of course. m. m. −s . Rather than going into the technical proof of this theorem. s its spin and α denotes the other quantum numbers.296) If the Hamiltonian is CPT invariant.4. The CPT theorem is basically an answer to this question and in essence says that any physical Hamiltonian would be invariant under the combined operation of P. −s|. We can naturally ask whether there exists any combination of these symmetries which will be a symmetry of the weak Hamiltonian. s|(CPT )−1 (CPT )H(CPT )−1 (CPT )|α. s −→ η α.1 Equality of mass for particles and antiparticles. s = m|α.4 CPT theorem 479 11. m.295) where m. m. m. m. Let us consider the state of a particle at rest which satisﬁes H|α. (11. CPT (11. m. 11.11. takes a particle to the antiparticle state. let us discuss the obvious consequences of this theorem. −s|(CPT )H(CPT )−1 |α. α denote the parameters associated with the antiparticle (in this section. s α. (11.

480 11 Discrete symmetries In other words. s|(CPT )−1 (CPT )Q(CPT )−1 (CPT )|α. q. so that ˆ CPT ˆ Q −→ −Q. for example.299) ˆ C ˆ Q −→ −Q.) Thus for a one particle state with charge q at rest. Let Q denote the operator which measures the charge of a state. q.2 Electric charge for particles and antiparticles. s ˆ α.4.3 Equality of lifetimes for particles and antiparticles.301) . Even though C-invariance is violated in weak interactions. (11. ˆ 11. s|Q|α. the equality of particle and antiparticle masses is a consequence of CPT invariance. let us decompose the weak Hamiltonian into a sum of Hermitian terms as (basically into a parity even and a parity odd part) HW K = H+ + H− . −s|(−Q)|α.4. In fact. q. q. ˆ T ˆ Q −→ Q. q.298) = −q. (11. Since the weak interactions violate parity. one can show that all the quantum numbers of antiparticles are equal in magnitude but opposite in sign from those of the particles. K 0 and K have the same mass. Namely. we have q = = = ˆ α. As we have seen ˆ P ˆ Q −→ Q. In this way. the ex0 perimental limit on the mass diﬀerence between the K 0 −K provides the best limit on CPT invariance. the electric charge for particles is equal in magnitude but opposite in sign from that of the antiparticles. −s (11. it is CPT invariance which guarantees 0 that.300) (11. 11. ˆ (Recall Q = d3 x ψγ 0 ψ for a Dirac particle. s ˆ α. q.

j. j. mj = α. mj −→ |β.304) = (i)2mj −2mj α. mj = (11. mj . and the corresponding antiparticle decay channel |α.303) The cross terms would have a pseudoscalar character and would vanish since the reaction rate is a scalar quantity. Let us consider the decay channel of a particle |α.307) . j. mj (11. j. Furthermore. the transition probability from an initial state |i to a ﬁnal state |f would be given by | f |S|i |2 ∝ | f |HW K |i |2 = | f |H+ + H− |i |2 = | f |H+ |i |2 + | f |H− |i |2 . j. j. mj −→ |β.305) (11. −mj . j. mj |(CPT )−1 (CPT )H± (CPT )−1 (CPT )|α. j. j. −mj |H± |β. mj . mj |H± |β. β. (11. −mj = α. j. using the rotational invariance of physical Hamiltonians. mj |H± |α. j.4 CPT theorem 481 where PH± P −1 = ±H± . j. (11. j.306) where we have used the fact that H± would be individually invariant under CPT according to the CPT theorem. j.11. j. (11. mj .302) In such a case. mj |H± |α. we can rewrite this relation as β. −mj |H± |β. j. We know that β.

mj |2 ρβ (11. mj |H± |α. mj |H− |α. j. β) which is not relevant for the subsequent discussion. This follows trivially from the equality of particle and antiparticle masses. j. and τ ) of particles and antiparticles. However. j. .309) which leads to a relation between the total life times (τ . mj |H+ |α. (11. mj = β. j. 984 (1958). There are many other interesting consequences of the CPT theorem including the connection between spin and statistics which we cannot go into. j. mj |H± |α. therefore.310) = 1 . j.482 11 Discrete symmetries Since the Hamiltonians H± are Hermitian we also have β. j. 11. From CPT invariance.5 References 1. mj |H+ |α. Let us simply note here that since CPT must be conserved in any physical theory and we know that CP is violated in weak interactions by a small amount. mj |2 . Okubo. S. mj |2 = | β. j. namely 2π 1 = τ = 2π β β | β. mj |2 + | β. j. j. mj |H± |α. mj |H± |α. we have obtained | β. mj ∗ . j. it follows that the weak interactions must violate T -invariance by a small amount as well so that CPT will hold. j. mj |2 ρβ | β. j. mj |2 + | β. Physical Review 109. mj |H− |α. j. τ Here we have assumed that the phase space for the decay products of the antiparticle is the same as that for the decay products of the particle. j. we note here that the individual decay channels may have diﬀerent life times for particle and anti-particle.308) Here we have neglected an overall phase factor (depending on α. (11. j.

3. Gasiorowicz. New York (1964). S. 4. Itzykson and J-B. Spin and Statistics and All That. New York (1993). Streater and A. Quantum Field Theory. 8. Zuber. C. D. 7. Princeton (1964). 9. Benjamin. Bjorken and S. McGrawHill. John Wliley. New York (1966). R. P CT . Drell. Peterson.11. 6. J. Introduction to Relativistic Quantum Field Theory. Roman. Gross. Evanston (1961). Invariance Principles and Elementary Particles. Elementary Particle Physics. 1964.5 References 483 2. John Wiley. Wightman. . Relativistic Quantum Fields. New York (1969). McGrawHill. Introduction to Quantum Field Theory. Schweber. Relativistic Quantum Mechanics and Field Theory. S. John Wiley. J. F. New York. J. S. Sakurai. F. New York. Princeton University Press. 5. Row. 1980. P.

This page intentionally left blank .

this Lagrangian density has the global symmetry ψ → e−iθ ψ.1) where ψ provides a representation of U (1) (namely. Let us recall that gauge ﬁeld theories necessarily arise when we try to promote a global symmetry to a local symmetry. it is a complex ﬁeld).37) and (8. (12. (12.Chapter 12 Yang-Mills theory 12. δψ δψ = = where θ is a real constant parameter and inﬁnitesimally. we have talked about the simplest of gauge theories.9) / L = iψ∂ ψ − mψψ. As we have seen in (8. let us quickly review how the gauge ﬁelds come into the theory in the case of quantum electrodynamics (QED).38). 485 (12. Such theories belonging to non-Abelian (non-commutative) gauge groups are commonly known as Yang-Mills theories and are fundamental building blocks in the construction of physical theories.1 Non-Abelian gauge theories So far. the Maxwell theory which is based on the Abelian gauge group U (1). we have −iǫψ. iǫψ. namely. For example.3) . We will now study gauge theories based on more complicated symmetries.2) ψ → ψeiθ . Let us start with the Dirac Lagrangian density for a free fermion (8.

(12.1) nor the corresponding action is invariant under the local phase transformation in (12.7) .6) Clearly this can be achieved if we introduce a new ﬁeld variable Aµ (x) (gauge ﬁeld) and deﬁne the covariant derivative as Dµ ψ(x) = (∂µ + ieAµ (x)) ψ(x).81)).4) = −ǫ(x)ψ∂ ψ + ψγ µ ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ / = (∂µ ǫ(x))ψγ µ ψ. The internal symmetry in this case arises from the fact that ψ is a complex ﬁeld variable while the Lagrangian density which is a functional of ψ is Hermitian.5) (12. δ (Dµ ψ) = −iǫ(x)Dµ ψ(x). Consequently. δψ = iǫ(x)ψ.4) since ∂µ ψ does not transform covariantly and there is no term in the Lagrangian density (12. the way out of this diﬃculty is to deﬁne a covariant derivative Dµ ψ such that it transforms covariantly under the local transformation. we note that δψ = −iǫ(x)ψ. (12. δ (∂µ ψ) = ∂µ (δψ) = ∂µ (−iǫ(x)ψ) = −i ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ(x). (12. If we now want to promote the symmetry to be local. namely.1) whose variation would cancel the ∂µ ǫ(x) term.486 12 Yang-Mills theory with θ = ǫ = inﬁnitesimal. Neither the Lagrangian density (12. we have δL = iδψ∂ ψ + iψγ µ ∂µ δψ − mδψψ − mψδψ / −imǫ(x)ψψ + imǫ(x)ψψ. namely ǫ = ǫ(x) (or θ = θ(x)). the phase of ψ remains arbitrary. As we have seen earlier (see (9. Consequently.

the ﬁeld Aµ known as the gauge ﬁeld.10). δAµ = 1 ∂µ ǫ(x). we note that since the gauge ﬁeld transforms longitudinally (by a gradient). e (12. if we deﬁne a ﬁeld strength tensor Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . (12.7) as describing the minimal coupling of the charged fermions to the electromagnetic ﬁeld and as we have seen in (9.8) We recognize (12.9) is invariant under the inﬁnitesimal local gauge transformations δψ = −iǫ(x)ψ(x). this would transform as δFµν = ∂µ δAν − ∂ν δAµ = 1 (∂µ ∂ν ǫ − ∂ν ∂µ ǫ) = 0. δψ = iǫ(x)ψ(x). is nondynamical in this theory. the Lagrangian density L = iψγ µ Dµ ψ(x) − mψψ.13) .12.11) under the gauge transformation (12. 4 (12. the curl of the ﬁeld would be unchanged by such a transformation. (12.86). e δAµ (x) = (12.1 Non-Abelian gauge theories 487 with the transformation property for the additional ﬁeld 1 ∂µ ǫ(x).10) However. e (12.12) Hence the ﬁeld strength tensor is gauge invariant and the gauge invariant Lagrangian density for the dynamical part of the gauge ﬁeld can be written as (quadratic in derivatives) 1 Lgauge = − Fµν F µν . That is. To introduce the kinetic energy part for the gauge ﬁeld (in order to give it dynamics) in a gauge invariant manner.

Let us consider a free Dirac theory described by the Lagrangian density (containing several complex fermion ﬁelds) L = iψ k ∂ ψk − mψ k ψk .16) where θ a denotes the real global parameter of transformation and inﬁnitesimally we have δψ k = iǫa ψ ℓ (T a )ℓk . a δψk = −iǫa Tkℓ ψℓ . dim G.17) with θ a = ǫa = inﬁnitesimal. . Let us now generalize these ideas to theories with more complicated symmetries. 2. dim G represents the dimensionality of the symmetry group. (12. we assume that ψk belongs to a nontrivial representation R of some internal symmetry group G and dim R denotes the dimensionality of the representation. 2.15) where k is an internal symmetry index.) The generators (matrices) T a satisfy the Lie algebra of the Lie group G (this is necessary for the transformations to form a group) . aT a k = ψU −1 = ψeiθ k . 4 (12. (We note here parenthetically that dim G = n2 − 1 for the group SU (n). dim G. . 2. . k a = 1. Furthermore.10). .488 12 Yang-Mills theory The total Lagrangian density for QED is then given by (see also (9. dim R. (12.14) and has the local gauge invariance described in (12. This Lagrangian density is invariant under the global phase transformations aT a ψk → ψk → Uψ ψU † k = e−iθ ψ k. · · · .80)) 1 LQED = − Fµν F µν + iψγ µ Dµ ψ − mψψ. / k = 1. Here the T a ’s represent the generators of the internal symmetry group G and are assumed to be Hermitian. . . . (12. a = 1. Namely. .

20) Once again. the structure constants of the group SU (2) can be identiﬁed with f abc = ǫabc .16)). 2. (T a )† = T a as well as the fact that the internal symmetry generators commute with the Dirac gamma matrices since the two act on diﬀerent spaces. (For semi-simple groups the structure constants can always be chosen to be completely antisymmetric.12.1 Non-Abelian gauge theories 489 T a .19) Here we have used the fact that the symmetry generators are Hermitian. namely. (12. the structure constants are nontrivial). dim G. .18) − 1 . (12. the symmetry group is known as a non-Abelian group. namely. f 458 = f 678 = 23 . we already know from the study of angular momenta. . a. 3 while for the group SU (3) the structure constants are more complicated and the nontrivial ones 1 are given by f 123 = 1. we note that the ordinary derivative acting on the fermion ﬁeld does not transform covariantly under this transformation. b. f 147 = f 246 = f 257 = f 345 = 2 . For example. a δψ k = iǫa (x)ψ ℓ Tℓk . That is. It is easy to see that the Lagrangian density is invariant under the transformations (12. Let us now try to make this symmetry a local symmetry of our theory. . .) When the generators of the symmetry group 2 do not commute (namely.17) (or (12. f 156 = f 367 = √ a / (i∂ − m)ψk − iǫa ψ k (i∂ − m)Tkℓ ψℓ / a / / = iǫa ψ k (T a )kℓ (i∂ − m)ψℓ − iǫa ψ k (i∂ − m)Tkℓ ψℓ = 0. let us consider the inﬁnitesimal local phase transformations a δψk = −iǫa (x)Tkℓ ψℓ . T b = if abc T c . . c = 1. δL = δψ k (i∂ − m)ψk + ψ k (i∂ − m)δψk / / = iǫa ψ ℓ T a ℓk where the real constants f abc which are completely antisymmetric (antisymmetric under the exchange of any pair of indices) are known as the structure constants of the symmetry group. a. (12. b. 2. c = 1.

where g denotes the appropriate coupling constant and Aµ = T a Aa . We note that the number of gauge ﬁelds (that are needed to deﬁne a covariant derivative) is the same as dim G.24) (12. We can compare this with QED where the generator is the identity matrix ½ which commutes with every operator. / / δL = δψ k (i∂ − m)ψk + ψ k (i∂ − m)δψk = (∂µ ǫa (x))ψ k γ µ (T a )kℓ ψℓ .22) As in the case of the U (1) theory. (12. We can also write the covariant derivative in (12. µ µ (12.26) . there is no other term in the Lagrangian density whose variation would cancel the ∂µ ǫa (x) term in (12. we write the covariant derivative (in the matrix notation) as Dµ ψ = (∂µ + igAµ ) ψ. Hence we deﬁne a covariant derivative (Dµ ψ)k such that under an inﬁnitesimal local transformation a δ (Dµ ψ)k = −iǫa (x)Tkℓ (Dµ ψ)ℓ .23) Introducing a new ﬁeld (gauge ﬁeld).490 12 Yang-Mills theory a δ (∂µ ψk ) = ∂µ (−iǫa (x)Tkℓ ψℓ ) a a = −i (∂µ ǫa (x)) Tkℓ ψℓ − iǫa (x)Tkℓ ∂µ ψℓ .25) This is the generalization of the minimal coupling (in QED) to the present case and the T a ’s are the generators of the symmetry group belonging to the same representation as the fermions. (12.24) explicitly as a a Dµ ψk = ∂µ ψk + igTkℓ Aa ψℓ = ∂µ δkℓ + igTkℓ Aa ψℓ .20). µ (12. the dimension of the Lie group. (12.21) so that under the local transformation (12.22).

T b Ab ψm µ (12.1 Non-Abelian gauge theories 491 Noting that we would like the covariant derivative to transform covariantly under an inﬁnitesimal local transformation.26) that we should have a igTkℓ δ Aa ψℓ = δ(Dµ ψk ) − ∂µ δψk µ a = −iǫa Tkℓ (Dµ ψ)ℓ − ∂µ δψk a b a = −iǫa Tkℓ ∂µ ψℓ + igTℓm Ab ψm − ∂µ (−iǫa Tkℓ ψℓ ) µ a b = −iǫa Tkℓ ∂µ ψℓ + igTℓm Ab ψm µ a a +i (∂µ ǫa ) Tkℓ ψℓ + iǫa Tkℓ ∂µ ψℓ a b a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa Tkℓ Tℓm Ab ψm . µ This can also be written as .30) a = i (∂µ ǫa ) Tkℓ ψℓ + igǫa f abc (T c )km Ab ψm . µ (12.28) can be written in the form a a a igTkℓ δ Aa ψℓ = igTkℓ δAa ψℓ + igTkℓ Aa δψℓ µ µ µ a a b = igTkℓ δAa ψℓ + igTkℓ Aa − iǫb Tℓm ψm µ µ a b a = igTkℓ δAa ψℓ + gǫa Tkℓ Ab Tℓm ψm .12.27) we conclude from (12. a δ (Dµ ψ)k = −iǫa Tkℓ (Dµ ψ)ℓ .28). (12. we obtain a igTkℓ δAa ψℓ µ a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a T b km Ab ψm µ −gǫa T b T a km Ab ψm µ km a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a .29) Substituting this into (12. µ µ (12.28) We note that the left-hand side in (12.

16)) .34) If we use the deﬁnition (see (12.31) which determines the transformation for the gauge ﬁeld to be δAa = µ 1 1 ∂µ ǫa − gf abc Ab ǫc = ∂µ ǫa − f abc Ab ǫc . µ g (12. 2. (12.25)) Aµ = T a Aa . a δψ k = iǫa (x)ψ ℓ Tℓk .492 12 Yang-Mills theory a igTkℓ ψℓ δAa − µ 1 ∂µ ǫa + f abc Ab ǫc = 0. (12. . . We can now deﬁne a unitary representation of the group as describing the ﬁnite symmetry transformation matrix (see (12. .35) where the gauge ﬁelds Aµ are matrices belonging to the same representation of the group as the fermions (since the generators T a belong to this representation). δAa = µ 1 ∂µ ǫa − f abc Ab ǫc . µ µ g g (12. (12. dim G. µ g (12.26) is invariant under the inﬁnitesimal local transformations a δψk = −iǫa (x)Tkℓ ψℓ (x). we can write the Lagrangian density in (12. the Lagrangian density is a scalar) L = iψγ µ Dµ ψ − mψψ.32) Thus the Lagrangian density L = iψ k γ µ (Dµ ψ)k − mψ k ψk .36) with normal product rules for matrices.33) in terms of matrices as (of course. µ a = 1. .33) with the covariant derivative deﬁned in (12.

the tensor representing the Abelian ﬁeld strength (see (12.38). U † = U −1 .1 Non-Abelian gauge theories 493 U = e−iθ a (x)T a .34)) for the ﬁeld variables in the matrix form as ψ → U ψ. ψ → ψU −1 .38) To construct the Lagrangian density for the dynamical part of the gauge ﬁeld in a gauge invariant manner.37) so that we can write the ﬁnite transformations (corresponding to (12.39) . Aµ → U Aµ U −1 − 1 (∂µ U ) U −1 . ig (12. (12.11)) would transform as fµν = ∂µ Aν − ∂ν Aµ 1 (∂ν U ) U −1 ig 1 (∂µ U ) U −1 ig 1 1 (∂µ ∂ν U ) U −1 − (∂ν U ) ∂µ U −1 ig ig 1 1 (∂ν ∂µ U ) U −1 + (∂µ U ) ∂ν U −1 ig ig → ∂µ U Aν U −1 − −∂ν U Aµ U −1 − = ∂µ U Aν U −1 − −∂ν U Aµ U −1 + = 1 1 (∂µ U ) ∂ν U −1 − (∂ν U ) ∂µ U −1 + ∂µ U Aν U −1 ig ig +U ∂µ Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 −U ∂ν Aµ U −1 − U Aµ ∂ν U −1 1 ∂µ U ∂ν U −1 − ∂ν U ∂µ U −1 + ∂µ U Aν U −1 ig = +U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 +U (∂µ Aν − ∂ν Aµ ) U −1 . we note that under the gauge transformation (12.12. (12.

42) so that Fµν transforms covariantly.38).41) (12. Let us also note that under the gauge transformation ig [Aµ . in the present case. if we deﬁne the ﬁeld strength tensor as Fµν = ∂µ Aν − ∂ν Aµ + ig [Aµ . it is clear that. Aν ] . (12.39) and (12.40). then under the gauge transformation (12. here fµν = ∂µ Aν − ∂ν Aµ is neither invariant nor does it have a simple transformation under the gauge transformation (12. Aν ] U −1 − − ∂µ U Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 . (12. we see that unlike in the case of QED. Fµν → U Fµν U −1 .494 12 Yang-Mills theory Namely. It is now easy to construct the gauge invariant Lagrangian density for the dynamical part of the gauge ﬁeld as (quadratic in derivatives) .38).40) Thus comparing (12. Aν ] = ig (Aµ Aν − Aν Aµ ) → ig U Aµ U −1 − 1 1 (∂µ U ) U −1 U Aν U −1 − (∂ν U )U −1 ig ig − U Aν U −1 − 1 1 (∂ν U )U −1 U Aµ U −1 − (∂µ U )U −1 ig ig 1 ∂µ U Aν U −1 − U Aµ ∂ν U −1 ig = ig U Aµ Aν U −1 − + + = 1 ∂µ U ∂ν U −1 − U Aν Aµ U −1 g2 1 1 ∂ν U Aµ U −1 − U Aν ∂µ U −1 − 2 ∂ν U ∂µ U −1 ig g 1 ∂µ U ∂ν U −1 − ∂ν U ∂µ U −1 ig igU [Aµ .

46) where we can set m = 0 if the fermions (quarks) are massless. µ ν ν µ (12.45) The complete Lagrangian density for QCD (quantum chromodynamics.43) where a particular normalization for the trace of the generators is assumed (namely. ν µ µ ν or.41) is easily seen to take the form. therefore.59) and the comments fol2 lowing that equation). 4 (12. more generally for fermions interacting with a non-Abelian gauge ﬁeld.38). or. This Lagrangian density is invariant under the non-Abelian gauge symmetry transformations (12. Fµν = ∂µ Aν − ∂ν Aµ + ig Aµ . Aν . T a T b = 1 δab .38) which have the inﬁnitesimal form (see (12. In components. a a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac = −Fνµ .44) so that the gauge invariant Lagrangian density for the gauge ﬁeld can also be written in terms of component ﬁelds as 1 a Lgauge = − Fµν F µν a 4 1 = − ∂µ Aa − ∂ν Aa − gf abc Ab Ac ν µ µ ν 4 × ∂ µ Aνa − ∂ ν Aµa − gf apq Aµp Aνq . From the cyclicity of trace this Lagrangian density is easily seen to be invariant under the gauge transformations (12. QCD corresponds to the particular case when G = SU (3)) is. the ﬁeld strength tensor (12. (12. 2 4 (12.1 Non-Abelian gauge theories 495 1 1 a Lgauge = − Tr Fµν F µν = − Fµν F µν a . given by 1 a LQCD = − Fµν F µν a + iψ k γ µ (Dµ ψ)k − mψ k ψk . see (12.12. a Fµν T a = ∂µ Aa − ∂ν Aa T a + igAb Ac T b .34)) . T c ν µ µ ν = ∂µ Aa − ∂ν Aa T a − gf abc Ab Ac T a .

T a = 0. f abp f pcq T q + f cap f pbq T q + f bcpf paq T q = 0. or.48) and the Jacobi identity associated with this algebra is given by T a. T c + if cap T p .37) as well) U (x) = eiθ a (x)T a .50) From the structure of the Lie algebra we know that we can write the ﬁnite group elements in a unitary representation as (recall ﬁnite rotations and the angular momentum operators and see (12. T c . T b = if abc T c . δAa = µ δψk = −iǫa (x)(T a )kℓ ψℓ . then the Lie algebra breaks up into two commuting non-Abelian subalgebras.496 12 Yang-Mills theory δψ k = iǫa (x)ψ ℓ (T a )ℓk .47) Let us discuss brieﬂy some of the properties of the Lie algebra of the symmetry group G. T b + if bcp T p . In this case . (12.51) If we can divide the generators of the non-Abelian Lie algebra into two non-Abelian subsets such that f abc = 0 when one index is in one set and another index is in the second set. T c + T c.49) Using (12. (12. as we have noted in (12. T b . The algebra of the Hermitian generators. (12. T a = 0.18). T a . µ g (12. 1 ∂µ ǫa − f abc Ab ǫc . f abp f pcq + f cap f pbq + f bcp f paq = 0. has the form T a . T b + T b . (12.48) it can be seen that the Jacobi identity (12.49) imposes a restriction on the structure constants of the group of the form if abp T p . or.

T b T c (12. we note that .53) Using the cyclicity property of the trace. is completely antisymmetric in all its indices.55) where the index c is ﬁxed (and not summed).52) We note that C2 is a normalization constant which determines the values of the structure constants.54) Tr Tr T a T b T c − Tr T b T a T c . A direct product of simple Lie groups is called semi-simple. using the commutation relation (12. let us next note that the quantity habc = = T a. Tr T a T b = (12. To prove this let us note that we can always diagonalize the tensor Tr T a T b such that (this is a symmetric real matrix in the “ab” space) 0 Ka if if a = b. A non-Abelian Lie group that cannot be so factorized is called a simple Lie group. we will assume that the symmetry group G is simple. (12. a = b. In all our discussions.1 Non-Abelian gauge theories 497 the group G is a direct product of two independent non-Abelian Lie groups. On the other hand. Furthermore.18) we obtain habc = Tr if abp T p T c = if abp Tr T p T c = if abp Kp δpc = iKc f abc . It depends on the representation but not on the indices a and b. (12.12. For any representation of a simple Lie group we can write Tr T a T b = C2 δab .

61) . we have (T a T a )mn = C(R) δmn .56) we conclude that Kb = Kc = K. (12.57) Using this and comparing (12.58) (12.59) which is used to show that the structure constants for a semi-simple group are completely antisymmetric in the indices. 2. · · · . (It is chosen to be 1 for the fundamental representation to which the fermions 2 belong in SU (n) and this is the normalization used in (12. a. since habc is completely antisymmetric.55) and (12. However. This shows that we can write Tr (T a T b ) = C2 δab . b = 1. Similarly. we have hacb = −habc .43).) We note that if we write Tr T a T b = T (R) δab . we note here that it is (12. dim R. dim G. (12. · · · .59) where the constant C2 depends only on the representation. (12.56) with the index b ﬁxed. 2. n = 1. m.60) then T (R) is known as the index of the representation R.498 12 Yang-Mills theory hacb = Tr T a. (12. (12. T c T b = if acp Tr T p T b = if acp Kp δpb = iKb f acb = −iKb f abc . Furthermore.

a particular representation that is very important as well as useful is given by Ta = −if abc . T b = = = T aT b − T bT a Ta cp cq cq cp Tb pq − Tb Ta pq − if acp − if bpq − − if bcp − if apq = −f acp f bpq + f bcp f apq = f abp f pcq = −f cap f pbq − f bcp f paq = if abp (−if pcq ) = if abp (T p )cq . (12. (12. we conclude that the identiﬁcation .63) bc This is consistent with the hermiticity requirement for the generators. (12.65) where we have used the anti-symmetry of the structure constants as well as (12.62) We can determine the generators of the group in various representations much like in the case of angular momentum. However. namely. T a.64) Furthermore. we can easily check that this representation satisﬁes the Lie algebra.50). T a† = = Ta ∗ ∗ bc cb − if acb = if acb = −if abc = T a bc . As a result. The two are clearly related as (this is easily seen by taking trace in the respective spaces) T (R) dim G = C(R) dim R. (12.1 Non-Abelian gauge theories 499 where C(R) is known as the Casimir of the representation R.12.

transforms according to the adjoint representation of the group. (12.66) indeed deﬁnes a representation of the Lie algebra known as the adjoint representation. Let us next look at the transformation for the gauge ﬁelds in (12. Aµ . T b .26)) δAa = µ 1 (adj) a D θ .500 12 Yang-Mills theory a T(adj) bc = −if abc . which inﬁnitesimally has the form Fµν → U Fµν U −1 so that = Fµν + iǫb Fµν . (12. g µ (12.68) which shows that the gauge ﬁeld. for example.42). (12. µ (12.67) so that we can equivalently write (see.69) .34) δAa = µ = = = 1 ∂µ ǫa − gf abc Ab ǫc µ g 1 ∂µ ǫa + gf bac Ab ǫc µ g 1 ∂µ ǫa + ig − if bac Ab ǫc µ g 1 b ∂µ ǫa + ig T(adj) g ac Ab ǫc . This can also be seen from the transformation of the ﬁeld strength tensor in (12.

or. T b a = iǫb Fµν T a . µ (12.70) Comparing this with the transformation of the fermions in (12.71) can also be written in the matrix form as (adj) Dµ ǫ = ∂µ ǫ + ig[Aµ .72) where Aµ = Aa T a and ǫ = ǫa T a (with T a in any representation). (12. (12. a a δFµν T a = iǫb Fµν if abc T c . the gauge ﬁeld must transform in the adjoint representation. let us scale the ﬁeld variables as .) For completeness. or. let us note here that the covariant derivative in the adjoint representation (12.34) we conclude that the ﬁeld strength Fµν as well as the gauge ﬁeld Aµ transform according to the adjoint representation of the group.1 Non-Abelian gauge theories 501 δFµν = iǫb Fµν .12. ǫ]. (It does not matter what representation the matter ﬁelds belong to. T b .73) Let us note the following features of this Lagrangian density. µ Let us now concentrate only on the gauge ﬁeld (Yang-Mills) part of the Lagrangian density 1 a LYM = − Fµν F µν a 4 1 = − ∂µ Aa − ∂ν Aa − gf abc Ab Ac ν µ µ ν 4 × ∂ µ Aνa − ∂ ν Aµa − gf apq Aµp Aνq . a b δFµν = −f abc Fµν ǫc b = −i − if cab ǫc Fµν c = −i T(adj) ab b ǫc Fµν c = −iǫc T(adj) Fb. ab µν (12.67) (adj) Dµ ǫ a = ∂µ ǫa − gf abc Ab ǫc . For example.

44)) Fµν a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac . ν µ µ ν = ∂µ Aν − ∂ν Aµ + ig[Aµ . 12. g µ It follows then that 1 a Fµν → ∂µ Aa − ∂ν Aa − f abc Ab Ac .502 12 Yang-Mills theory Aa → µ 1 a A . therefore.75) LYM = − Tr Fµν F µν = − Fµν F µν a .74) Thus we note that in this theory. the pure Yang-Mills theory is an interacting theory unlike the Maxwell theory. Since the gauge ﬁelds couple to any ﬁeld (particle) carrying charge of the symmetry group.41) and (12. (12. Aν ]. Physically we understand this in the following way. (12. 2 4 where the ﬁeld strength tensor is given by (see (12. The other feature to note is that unlike the photon ﬁeld. We note that the Lagrangian density for the Yang-Mills theory is given by (12. here the gauge ﬁelds have self interaction (they interact with themselves) and. the gauge ﬁelds carry the charge of the non-Abelian symmetry group (they have a nontrivial symmetry index) in contrast to the photon ﬁeld which is chargeless. in the case of non-Abelian symmetry they must couple to themselves and possess self interactions.45)) 1 a 1 (12. In the present case.76) .2 Canonical quantization of Yang-Mills theory Let us discuss the canonical quantization of Yang-Mills theory in the same spirit as the discussion of Maxwell’s theory in chapter 9. ν µ µ ν g which leads to LYM → − 1 ∂µ Aa − ∂ν Aa − f abc Ab Ac ν µ µ ν 4g2 × ∂ µ Aνa − ∂ ν Aµa − f apq Aµp Aνq . the coupling constant can be scaled out and written as an overall multiplicative factor in the Lagrangian density.43) (or (12.

80) Let us note that the coeﬃcient matrix of highest derivatives in (12. (12.79) where the covariant derivative is deﬁned to be in the adjoint representation of the group (see (12.67) and (12. In the matrix notation (see (12. we can identify the non-Abelian electric and magnetic ﬁelds from (12.76) as a a F0i = ∂0 Aa − ∂i Aa − gf abc Ab Ac = Ei . µ (12.68) and we are suppressing the symbol “(adj)” to denote the covariant derivative in the adjoint representation for simplicity) .72)).12.77) The Lagrangian density (12. the theory is singular implying that there are constraints which is evident from the fact that the YangMills Lagrangian density is invariant under the inﬁnitesimal gauge transformation (see (12. we can write the Euler-Lagrange equations as Dµ F µν = ∂µ F µν + ig[Aµ .2 Canonical quantization of Yang-Mills theory 503 As in the case of the Abelian theory (Maxwell theory). i 0 0 i a Fij a = ∂i Aa − ∂j Aa − gf abc Ab Ac = −ǫijk Bk . ∂∂µ Aa ∂Aa ν ν ∂µ or. −∂µ F µν a − −gf abc Ab F µν c = 0. 2 2 4 LYM = Tr Ei Ei − (12.78) The Euler-Lagrange equation for the theory are obtained from (12.75) can also be written explicitly in terms of the non-Abelian electric and magnetic ﬁeld strength tensors (12. j i i j (12.75) is the transverse projection operator just like in the case of Maxwell’s theory. µ (Dµ F µν )a = ∂µ F µν a − gf abc Ab F µν c = 0. Therefore. or. F µν ] = 0.68)).75) and have the form ∂L ∂L − = 0.77) as 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a .

Di F i0 a = 0.81) In fact. the constrained structure of the theory is already obvious in the Euler-Lagrange equation of motion (12. ν. we see that we have only 3N canonical momenta where N = dim G.84) (12. we choose Aa (x) = 0.86) If a gauge ﬁeld conﬁguration does not satisfy this condition we can always make a suitable gauge transformation so that the transformed ﬁeld would satisfy the condition. For example.82) which is a constraint relation. (12. as in the case of Maxwell’s theory and a Πi a (x) = −F 0i a = Ei (x).79) for ν = 0.77). A0 (x) = 0. Let us obtain the momenta canonically conjugate to the ﬁeld variables Aa from (12.85) a where the non-Abelian electric ﬁeld Ei (x) is deﬁned in (12. namely. Namely. Thus. we have Π0 a (x) = −F 00 a (x) = 0. This implies that 0 Aa is like a c-number quantity which commutes with every other 0 operator in the theory.75) and this leads to µ Πµ a (x) = ∂L = −F 0µ a (x). The momentum conjugate to Aa does not exist. (12. g (12. requiring that .504 12 Yang-Mills theory δAa = µ 1 (Dµ ǫ(x))a . 0 or. Thus we can choose a gauge condition and set it equal to zero. ˙ a (x) ∂ Aµ (12. (12.83) a Noting that the ﬁeld strength Fµν is antisymmetric in the indices µ.

(12. we note from (12.82) that Di Ei = ∂i Ei + ig[Ai .86) the non-Abelian electric ﬁeld in (12.75) takes the form 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a 2 2 4 1 ˙ a ˙ a 1 a ij a 1 ˙ ˙ = Tr Ai Ai − Fij F ij = Ai Ai − Fij F .91) can be inverted to give . ˙ Ei = Ai .2 Canonical quantization of Yang-Mills theory 505 A′ = U A0 U −1 − 0 or.87) we can determine the parameter of transformation and we see that a gauge ﬁeld can always be transformed to satisfy (12. ig (12. Ei ] = 0. Let us note that with the gauge condition (12.12. 1 (∂0 U ) U −1 = 0. On the other hand. i or. the term quadratic in the nonAbelian magnetic ﬁeld) is invariant under the ﬁeld redeﬁnition (gauge transformation) 1 ¯ 1 ¯ Ai = S Ai S −1 − (∂i S)S −1 = S Ai − S −1 (∂i S) S −1 .86) the Lagrangian density (12.88) 2 2 4 LYM = Tr Ei Ei − where we have used the fact that in the gauge (12.91) ig ig so that there is a cyclic variable in the theory that needs to be separated. in the present case. U −1 ∂0 U = igA0 . (12. we expect the physical dynamical degrees of freedom of a massless gauge ﬁeld to be transverse. (12. We note that the relation (12.89) From our discussion in the case of the Maxwell theory.86). (12.77) takes the form a ˙ Ei = Aa . let us also note that the potential term in (12.90) so that neither the gauge ﬁeld Ai nor the electric ﬁeld Ei is transverse ˙ (recall that Ei = Ai ). However.88) (namely.

namely.93). then we can deﬁne a new ﬁeld variable Ai satisfying the transversality condition (12. we obtain 1 ¯ ∂i Ai = ∂i S Ai S −1 − ∂i (∂i S)S −1 ig 1 ¯ ¯ = (∂i S)Ai S −1 + S Ai (∂i S −1 ) − ∂i (∂i S)S −1 ig = (∂i S) S −1 Ai − + Ai S + = − = − 1 (∂i S −1 ) ig (12.91) that .95) This shows that if we can ﬁnd a gauge transformation which satisﬁes ¯ (12. ig (12. ¯ ∂i Ai = 0.72)) Di (A)G = ∂i G + ig[Ai . We note next from the ﬁeld redeﬁnition in (12. (∂i S)S −1 ] ig 1 Di (A) (∂i S)S −1 . Imposing this condition in (12.92) in the intermediate step and the covariant derivative in the adjoint representation is deﬁned for a matrix function G as (see (12. G].91).93) 1 1 (∂i S) (∂i S −1 ) − ∂i (∂i S)S −1 ig ig 1 ∂i (∂i S)S −1 − [Ai . let us require the new ﬁeld variable Ai to be transverse. (12.94). (12.506 12 Yang-Mills theory 1 1 ¯ Ai = S −1 Ai S + S −1 ∂i S = S −1 Ai + (∂i S)S −1 S.92) ig ig Keeping in mind that we would like the dynamical variables to be ¯ transverse.94) where we have used (12.

S −1 S] − ∂i (S −1 S) S −1 ig 1 ˙ ¯ ¯ ˙ = S Ai − Di (A)(S −1 S) S −1 ig 1 ˙ ¯ ¯ = S Ai − Di (A)f S −1 .86) we can write 1 ˙ ˙ ¯ ¯ Ei = Ai = S Ai − Di (A)f S −1 = SE i S −1 .12. where (12.96) ˙ ¯ where we have used ∂0 S −1 = −S −1 SS −1 .100) .97) From (12.99) Furthermore. ig (12. decomposing E i into its transverse and longitudinal components as T E i = E i + ∂i φ.2 Canonical quantization of Yang-Mills theory 507 1 ˙ ¯ Ai = ∂0 S Ai S −1 − (∂i S)S −1 ig ˙ ˙¯ ¯ ¯ ˙ = S Ai S −1 + S Ai S −1 − S Ai S −1 SS −1 − 1 ˙ ˙ (∂i S)S −1 − (∂i S)S −1 SS −1 ig 1 ˙ ¯ ¯ ˙ ˙ = S Ai − [Ai . (12.98) (12. Here Di (A) is the covariant derivative in the adjoint representation (see (12.95)) with respect to ¯ the gauge ﬁeld Ai and we have identiﬁed ˙ f = S −1 S. ig (12.96) we note that in the gauge (12. ig where we can identify 1 ˙ ¯ ¯ E i = Ai − Di (A)f.

E i ]S −1 ig ¯ = S(∂i E i + ig[Ai . φ] + ig[Ai . ¯ ∂i Di (A)φ = ρ. E i ] = 0. the constraint equation (12.102) Let us next note that under the ﬁeld redeﬁnition (12. Di (A)Ei = ∂i Ei + ig[Ai . φ] = −ig[Ai . or.508 12 Yang-Mills theory ∂i E i = 0.104) T ¯ where we have used ∂i Ai = 0 in the intermediate step. Ei ] ¯ = ∂i SE i S −1 + ig[S(Ai − 1 −1 S (∂i S))S −1 . E i ] = ρ. (12. or. T ∂i E i = ∂i ∂i φ = ∇2 φ. φ=− (12.90) takes the form ¯ ¯ Di (A)E i = ∂i E i + ig[Ai . SE i S −1 ] ig = (∂i S)E i S −1 + S(∂i E i )S −1 − SE i S −1 (∂i S)S −1 ¯ +igS[Ai − 1 −1 S (∂i S).91). ig ∇2 ˙ ¯T ¯ Ei = Ai . ¯ ¯ ∂i ∂i φ + ig[Ai .98). E i + ∂i φ ] = 0. E i ])S −1 ¯ = S(Di (A)E i )S −1 . T ¯ ¯ ∂i ∂i φ + ig∂i [Ai . (12. or. We can invert this relation to determine . T ¯ ∂i ∂i φ + ig[Ai .101) we conclude from the deﬁnition in (12. As a result of this relation.99) that 1 1 ¯ ∂i Di (A)f. E i ] = 0.91) and (12. or. (12.103) where we have used (12.

∂i Di (A) 1 ¯ ρ.106) as well as the cyclicity of trace. ∂j Dj (A) (12.12.105) so that from (12. i ˙ ¯a ∂A i (12.100) we obtain T ˙ ¯ E i = E i + ∂i φ = Ai + ∂i (12.88) takes the form LYM = Tr Ei Ei − 1 Fij (A)F ij (A) 2 1 ¯ ¯ = Tr E i E i − Fij (A)F ij (A) 2 1 1 ˙ ˙ ¯ ¯ = Tr (Ai + ∂i ¯ ρ)(Ai + ∂i ∂k Dk (A) ρ) ¯ ∂j Dj (A) 1 ¯ ¯ − Fij (A)F ij (A) 2 1 ˙ ˙ ¯ ¯ ¯ ¯ = Tr Ai Ai − Fij (A)F ij (A) 2 1 1 2 −ρ ¯ ∇ ∂k Dk (A) ρ ¯ ∂j Dj (A) 1 ¯a ¯a 1 ˙ ˙ ¯ ¯ = A A − Fij (A)F ij (A) 2 i i 4 1 1 1 a 2 − ρa ¯ ∇ ∂k Dk (A) ρ .2 Canonical quantization of Yang-Mills theory 509 φ = 1 ¯ ρ.98) and the decomposition in (12. ¯ We have isolated the dynamical variables of the theory to be Aa i which are transverse and the conjugate momenta can be determined from (12. ∂i Di (A) = Di (A)∂i which has been used together with integration by parts).91).108) . ¯ 2 ∂j Dj (A) (12.107) to be Π ia = ∂LYM ˙ ¯ = Aa . (12. the Lagrangian density (12.93).107) where we have neglected total divergence terms in the intermediate ¯ ¯ steps (note also that because of (12.106) Using the deﬁnition in (12.

i ¯ ∂i Aa = 0.111) (12. i (12. We note that the Lagrangian density in (12. A′ a = Aa + i i g such that ¯i ∂i A′ a = 0. Here we see that since the gauge ﬁelds are self interacting. Here the transverse delta function corresponds to the one already deﬁned in the case of Maxwell’s theory in (9.110) then we can make a time independent inﬁnitesimal gauge transformation (under which the theory is invariant) 1 ¯ǫ a ¯ ¯ Di (A)¯ . This can be seen from the ¯ fact that if we have a ﬁeld conﬁguration Aa which is transverse. Let us note here that the long range behavior of the interaction (and. Thus the long range behavior of this theory is still not well deﬁned in the Coulomb gauge.31). The last term is like the long range Coulomb interaction term in the case of QED (in an Abelian theory where the structure constants vanish and the adjoint covariant derivative reduces to an ordinary derivative. i. of the theory) ¯ is. therefore.109) with all other brackets vanishing. i TR jb (12. this last term has the form ρ 1 2 ρ which corresponds to ∇ the long range Coulomb interaction). This is related to the question of gauge ﬁxing.510 12 Yang-Mills theory The equal-time canonical Poisson brackets for the theory can now be obtained as in the case of the Maxwell theory ¯ {Aa (x). namely the Coulomb gauge does not uniquely deﬁne the gauge ﬁelds. ¯ Gribov has shown that the operator ∂i Di (A) does possess zero modes or eigenvectors with zero eigenvalue.112) . therefore. even in the absence of other matter ﬁelds there is a long range interaction.e. Π (y)} = δab δi j (x − y).107) has exactly the same form as in the Abelian theory except for the last term. controlled by the eigenvalues of the operator ∂i Di (A). (12.

the hypersurface deﬁning the gauge choice intersects the gauge orbits more than once. known as the Gribov ambiguity.115) . therefore. We can show that this nonuniqueness. consequently. by modifying the theory. As long as we are within perturbation theory we can neglect such conﬁgurations. However. we recall that in QED (see (9. we would not worry about it within the context of perturbation theory.) g Hence such gauge ﬁeld conﬁgurations can only be reached nonperturbatively. we can also try to quantize the non-Abelian gauge theory covariantly.111) that this has a singularity at g = 0.125) with J µ = 0) 1 1 a L = − Fµν F µν a − (∂µ Aµa )2 . (It is of the form O 1 . On the other hand.2 Canonical quantization of Yang-Mills theory 511 ¯ provided ǫa corresponds to a zero mode of ∂i Di (A). classically we can impose the condition ∂ · A = 0. 4 2 (12. Thus.75) as (see (9. In other words. very much along the lines of the Abelian theory that we have discussed in chapter 9. There now exist prescriptions to ¯ take care of this problem. there are serious diﬀerences from Maxwell’s theory. makes the theory nonsingular.129)) (∂ · A) = 0. However. let us modify the theory (12. much like in the Maxwell theory. But let us note from the form of A′ a in i (12. Gribov ambiguity is a phenomenon associated with large gauge transformations and is an important issue since it leads to nonperturbative eﬀects. For example. The canonical quantization of Yang-Mills theories does not have manifest covariance very much like the Maxwell theory. in the present case. occurs in all gauges other than the axial gauge.114) namely. (12. namely. This shows that ¯ the Coulomb gauge does not really specify the gauge ﬁeld conﬁguration uniquely. Therefore.113) The additional term clearly breaks gauge invariance and. The other way of saying this is that within the framework of perturbation theory we are interested in inﬁnitesimal gauge transformations. for example.12. (12. ∂ · A behaves like a free ﬁeld and.

nor can we think of the supplementary condition ∂ · Aa(+) |phys = 0.119) in a physically meaningful manner. the additional term in (12. consequently.512 12 Yang-Mills theory which then translates to the Gupta-Bleuler condition on the physical states ∂ · A(+) |phys = 0.118) Thus. 12. (12. µ Contracting with ∂ν . it cannot be uniquely decomposed into a positive and a negative frequency part (in a time invariant manner). we need to analyze the question of modifying the theory in a more systematic and detailed manner. however. We would see next how we can derive intuition on this important question from the path integral quantization of the theory. we obtain (∂ · Aa ) = gf abc ∂ν Ab F µν µ c (12.117) = 0. even classically the equations of motion are given by ∂µ F µν a − gf abc Ab F µν c + ∂ ν (∂ · Aa ) = 0.) Correspondingly the naive analog of the Gupta-Bleuler condition for non-Abelian gauge theories does not seem to exist.116) In the case of the Yang-Mills theory. the physical subspace would keep changing with time which is not desirable. (Namely. In other words.113) does not seem to be suﬃcient in contrast to the Abelian gauge theory. (12. since it is not invariant under time evolution. just modifying the Lagrangian density as in (12. in contrast to the Abelian theory. (12. Therefore. since (∂ · Aa ) is not a free ﬁeld.3 Path integral quantization of gauge theories Path integrals provide an alternative to the canonical quantization of ﬁeld theories and are particularly useful in studying complicated . Furthermore.113) has truly modiﬁed the theory. we note that (∂ · Aa ) does not behave like a free ﬁeld and.

In scalar and fermion ﬁeld theories where the relation between the Hamiltonian and the Lagrangian is conventional the vacuum to vacuum transition amplitude can be written as a path integral. Since this is a complete topic in itself. the path integral description is given by (recall = 1) J 0|0 = Z[J] = eiW [J] = N = N Dφ ei(S[φ]+ Dφ eiS R d4 x J(x)φ(x)) (J ) [φ] . A functional is roughly deﬁned as a function of a function and it is clear that the action of a ﬁeld theory is naturally a functional. Z[J] and W [J] are known as generating functionals for the Green’s functions of the theory. (12.120) where N is a normalization constant.4 for the deﬁnitions of “in” and “out” states) in the presence of interactions.12.121) Furthermore. For example. δφ(y) ǫ ǫ→0+ (12. Let us recall that the primary goal in the study of relativistic ﬁeld theories is to calculate scattering matrix elements which can be derived from the vacuum to vacuum transition amplitude (namely. Like the derivative of a function. . for a real scalar ﬁeld interacting only with an external source. from the “in” vacuum to the “out” vacuum. The functional dependence of a quantity on a variable is generally denoted by a square bracket. J(x) is the external source to which the scalar ﬁeld is coupled and S[φ] = d4 x 1 ∂µ φ∂ µ φ − m2 φ2 . we will only go over some of the essential concepts involved in such a description. see section 6. 2 (12.3 Path integral quantization of gauge theories 513 gauge theories. rather. we will not go into the systematic details of this description. we can also deﬁne functional derivatives in a simple manner through the relation (in four dimensions) δF [φ(x)] F [φ(x) + ǫδ4 (x − y)] − F [φ(x)] = lim .122) and various Green’s functions (n-point functions) are obtained from the generating functionals Z[J] and W [J] through functional diﬀerentiation.

namely.120) can be evaluated in a straightforward manner. δVi φi = (12. (12.126) is independent of ﬁeld variables and is a constant (possibly divergent) and can be incorporated into the normalization constant N . this is a generalization of the Gaussian integral to the functional space and leads to Z[J] = eiW [J] = N det ∂µ ∂ µ + m2 −1 2 e− 2 i R d4 xd4 y J(x)G(x−y)J(y) .124) δVi →0 and in terms of these discretized ﬁeld variables. Basically. i (12.126) where the Green’s function G(x − y) is formally the inverse of the operator in the quadratic part of the action.125) With these basics a simple functional integral for a quadratic action such as in (12.127) (We note here that a Gaussian integral for fermions leads to positive powers of determinants in contrast to the negative powers for bosons which is related to the fact that fermion loops have a negative sign associated with them.120) is known as a functional integration and is deﬁned as follows. Let us divide the entire space-time into inﬁnitesimal cells labelled by “i” of volume δVi and deﬁne 1 δVi d4 x φ(x). (12. (12. G = (−∂µ ∂ µ − m2 )−1 .123) Clearly in the limit of vanishing volume we recover lim φi = φ(x).) The determinant in (12.514 12 Yang-Mills theory The integration in (12. When we have an . the functional integration is deﬁned (up to a normalization constant) as Dφ = dφi .

let us go back to the Maxwell theory 1 L(J) = − Fµν F µν + J µ Aµ . With this brief introduction to path integral description.120). as we have seen before (see (9. The advantage of using the path integral description lies in the fact that there are no operators in (12. (12. (12.122)) we have identiﬁed P µν (x − y) = (η µν (J µ .P −1 J ν µν ).129) µν A ν )+(J µ . 4 (12. cannot be obtained in a closed form.126) Z [Jµ ] = eiW [Jµ ] = N (det (−P µν ))− 2 e− 2 (J 1 i µ . everything is a classical function. in general.P 1 ] .12. This is a Gaussian functional integral and we can evaluate this using our earlier result in (12. Aµ ) = − ∂ µ ∂ ν ) δ4 (x − y).123) and (9. In this case.128) where J µ represents a conserved current (source). In this case.131) d4 x J µ (x)Aµ (x).131). the operator P µν is a projection . However.130) where N is a normalization constant and (see (9. we evaluate the functional integral perturbatively (expanding the interaction term in the exponent so that the functional integral becomes a series of integrals corresponding to diﬀerent moments of a Gaussian integration) and the perturbative expansion coincides with the perturbative expansion of amplitudes in the conventional canonical quantized ﬁeld theory.132) where we have used the notation in (12. the generating functional in the path integral formalism is given by (J ) [A Z [Jµ ] = eiW [Jµ ] = N = N DAµ eiS µ] (12.124)).A µ) DAµ ei[ 2 (Aµ .3 Path integral quantization of gauge theories 515 interacting theory (not quadratic in the ﬁeld variables) the functional integral. (12.

The longitudinal vectors kµ (or ∂µ ) are its eigenvectors with zero eigenvalue.133) (12. x (12. the inverse of the matrix cannot be deﬁned either. (As we have seen the form of transformation for the gauge ﬁelds (12. all the Aµ ’s that are obtained by making a gauge transformations with all possible θ(x) are said to lie on an “orbit” in the group space.134) We can immediately see that for a U (1) gauge group.133) in terms of U (θ) is quite general and holds for the non-Abelian theories as well. even in the absence of any sources.) (θ) For a ﬁxed Aµ .136) . (The operator possesses zero modes and. is proportional to the “volume” of the orbits denoted by dθ(x). consequently. (12.135) which is the familiar gauge transformation for Maxwell’s theory.) The source of the diﬃculty is not hard to see. The Lagrangian density for Maxwell’s theory is invariant under the gauge transformation 1 (∂µ U )U −1 . the determinant of P µν vanishes. This implies that the generating functional does not exist in this case.516 12 Yang-Mills theory operator for transverse photons. the generating functional. e Aµ → A(θ) = Aµ + µ (12. Therefore. on the other hand. ie Aµ → A(θ) = U Aµ U −1 − µ where U (θ) = e−iθ(x) . The action S. we can write the transformation as 1 ∂µ θ(x). is constant on such orbits. Clearly therefore. This is a U (1) symmetry and the symmetry is noncompact since the parameter of transformation θ(x) can take any real value.

we can ﬁnd a gauge trans(θ) formed Aµ which does and F [A(θ) (x)] = 0.3 Path integral quantization of gauge theories 517 (In the non-Abelian case. Thus. for example. namely. Rather we should integrate over each orbit only once. we pick up a representative gauge ﬁeld from each gauge orbit. the Lorentz/Landau gauge.) This is an inﬁnite factor (which is one of the reasons for the divergence in the naive evaluation of the functional integral) and must be extracted out before doing any calculations. the axial gauge. We recognize that we should not integrate over all gauge ﬁeld conﬁgurations because they are not really distinct. A3 (x) = 0. A0 (x) = 0. then even if Aµ does not satisfy the condition. We can . the temporal gauge.139) is known as the gauge condition (or gauge ﬁxing condition). Physical quantities are. The way this is carried out is by choosing a hypersurface which intersects each orbit only once. of course.12. here are a few of the familiar gauge ﬁxing conditions F [Aµ (x)] = ∂µ Aµ (x) = 0. In this way. (12. (12. (12. this should be replaced by the group invariant Haar measure x dU (x).138) has a unique solution for some θ(x). µ (12.140) and so on. if F [Aµ (x)] = 0. gauge invariant and do not depend on the choice of the hypersurface (gauge). The method for extracting this factor out of the path integral is due to Faddeev and Popov and relies on the method of gauge ﬁxing. the Coulomb gauge. ∇ · A(x) = 0.137) deﬁnes the hypersurface which intersects the gauge orbits once. This procedure is known as gauge ﬁxing and the condition F [Aµ (x)] = 0.

That is (for the Abelian . in general. Let us make a gauge transformation Aµ → Aµ in (12. let us do the following trick due to Faddeev and Popov. known as the Faddeev-Popov determinant. (12. Let us deﬁne ∆−1 [Aµ ] = FP x dθ(x) δ F [A(θ) (x)] . it would lead only to gauge invariant Green’s functions. gauge dependent (recall. (The integration measure should be dU (x) which is essential in the case of non-Abelian theories. the photon propagator) although the physical S-matrix (the scattering matrix) elements are gauge independent. therefore. is gauge invariant which can be (θ ′ ) seen as follows. Then ∆−1 [A(θ ) ] = FP µ x ′ dθ(x) δ F [A(θ+θ ) (x)] µ (θ) dθ(x) δ F [Aµ (x)] x −1 ′ = = ∆FP [Aµ ] .) To extract out the inﬁnite gauge volume factor. µ (12. for example.) Note that the quantity ∆FP [Aµ ]. (Only physical quantities need to be gauge independent. Therefore. be thought of as a completeness relation.143) This follows from the fact that the measure in the group space is invariant under a gauge transformation. we know from ordinary perturbation theory that the Green’s functions are.141) This can also be written as ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] = 1. (12.518 12 Yang-Mills theory already see the need for gauge ﬁxing from the fact that because the action is gauge invariant so is the generating functional (if sources are transformed appropriately in the non-Abelian case).141).142) which can. On the other hand. Thus we have to ﬁx a gauge without which even the Cauchy initial value problem cannot be solved.

we should have the Haar measure which is gauge invariant.147) Z[Jµ ] = N = N DAµ ∆FP [Aµ ] dθ(x) x DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS (J ) [A µ] = N DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS .148) where the (inﬁnite) gauge volume has been factored out and absorbed into the normalization constant N of the path integral. we still have to determine what ∆FP [Aµ ] is.12. To do this let us note from (12.141) that .3 Path integral quantization of gauge theories 519 group a translation of the transformation parameter corresponds to a gauge transformation) d θ(x) + θ ′ (x) . (12.146) Furthermore. (12. µ under which the generating functional takes the form dθ(x) δ F [Aµ (x)] eiS x (J ) [A µ] (J ) [A µ] (12.145) Remembering that ∆FP [Aµ ] is gauge invariant we can now insert this identity factor into the generating functional to write (J ) [A Z[Jµ ] = N DAµ ∆FP [Aµ ] (θ) dθ(x) δ F [Aµ (x)] eiS x µ] . However. namely. Therefore. let us make an inverse gauge transformation Aµ → A(−θ) . d(U U ′ ) = dU. this gives the correct functional form for the generating functional.144) In the non-Abelian case. (12. dθ(x) = (12.

151) θ=0 . We see that the Faddeev-Popov determinant can be calculated simply by restricting to inﬁnitesimal gauge transformations (since we take θ = 0 at the end). We note that since ∆−1 [Aµ ] is gauge invariant we can make an FP inverse gauge transformation to make F [Aµ (x)] = 0 in the above derivation. for gauge ﬁelds which satisfy the condition F [Aµ (x)] = 0. (Here we can completely ignore the problem of Gribov ambiguity associated with large gauge transformations. be thought of as the Jacobian that goes with a given gauge choice.153) .151) we determine δF [Aµ (x)] ∆FP [Aµ ] = det δθ(y) (θ) (12. therefore. we have θ(x) = 0.152) The Faddeev-Popov determinant can.520 12 Yang-Mills theory ∆−1 [Aµ ] = FP x dθ(x) δ F [A(θ) (x)] µ dF δ F [A(θ) (x)] µ x = = det det δθ δF (12. (12. (12.) We can further generalize our derivation by noting that a general equation of the hypersurface has the form (physical results are not sensitive to the choice of the hypersurface) F [Aµ (x)] = f (x). Thus.149) δθ δF F [Aµ (x)]=0 (θ) . On the other hand. using (12.150) (12.149)-(12.

(12. The Faddeev-Popov determinant is unchanged by this modiﬁcation because f (x) does not depend on Aµ (x).154) into the functional integral.156) where we have deﬁned a gauge ﬁxing action as SGF = d4 x LGF = − d4 x (F [Aµ (x)])2 .158) . we can now do the following (also known as the ’t Hooft trick). (12. Then we can insert the identity ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] − f (x) = 1.155) Following ’t Hooft. Thus. We note that physical quantities are independent of f (x).12. (12. (12. (12. Thus the generating functional in this case is given by Z[Jµ ] = N DAµ ∆FP [Aµ ]δ F [Aµ (x)] − f (x) eiS (J ) [A µ] . Hence we can multiply the generating functional by a weight factor and integrate over all f (x). the generating functional becomes Z[Jµ ] = N × = N = N DAµ ∆FP [Aµ ] Df δ F [Aµ (x)] − f (x) e DAµ ∆FP [Aµ ]e i − 2ξ R d4 x(f (x))2 iS (J ) [Aµ ] e h i R 1 i S (J ) [Aµ ]− 2ξ d4 x(F [Aµ (x)])2 (J ) +S DAµ ∆FP [Aµ ] ei(S 1 2ξ GF ). We note further that since δF [Aµ (x)] δθ(y) (θ) ∆FP [Aµ ] = det θ=0 .157) and ξ is known as the gauge ﬁxing parameter.3 Path integral quantization of gauge theories 521 where f (x) is independent of Aµ .

but opposite statistics.e. [c(x). [c(x).522 we can write this as δF [Aθ (x)] µ δθ(y) 12 Yang-Mills theory ∆FP [Aµ ] = det = = = where Sghost = − θ=0 δF Dc Dc e−i(c. (12. c(y)]+ = 0. Thus the generating functional now takes the form Z[Jµ ] = eiW [Jµ] = N DAµ Dc Dc eiSeﬀ (J ) [Aµ .161) Thus although these ﬁelds behave as scalar ﬁelds under Lorentz transformations. to write the Faddeev-Popov determinant in the form of a ghost action (action involving ghost ﬁelds). We note here that this is possible (since we have a determinant with a nonnegative power) only if the ghost ﬁelds c(x) and c(x) anticommute (ghost ﬁelds have the same Lorentz structure as the parameters of transformation. (12. (12. i. δF [Aθ (x)] µ d xd y c(x) δθ(y) 4 4 (12. These ﬁelds are known as Faddeev-Popov ghosts and as is obvious from their anticommutation relations. c(y)]+ = 0.( δθ )θ=0 c) Dc Dc e −i R d4 xd4 y c(x) δF [Aθ (x)] µ δθ(y) θ=0 c(y) Dc Dc eiSghost .160) Here we have introduced two independent ﬁctitious ﬁelds c(x) and c(x).162) . known as ghost ﬁelds. they obey anticommutation rules. graphs involving these ﬁctitious particles in closed loops must have an additional (−1) factor just like the fermions.. c(y)]+ = 0.c] .159) θ=0 c(y).c. [c(x).).

our responsibility to show that the physical interpretation of the theory has not changed. Let us now look at a simple gauge ﬁxing condition in Maxwell’s theory.142) that both these actions arose from inserting a factor of unity into the functional integral).3 Path integral quantization of gauge theories 523 where Seﬀ [Aµ . c] = S (J) [Aµ ] + SGF + Sghost = d4 x Leﬀ [Aµ . of course. for the moment let us emphasize that the purpose of a gauge ﬁxing Lagrangian density is to break gauge invariance. for example.153)) F [Aµ (x)] = ∂µ Aµ (x) = f (x). 2ξ 2ξ (12.164) In this case.12. c] . we start with a gauge invariant Lagrangian density and add to it a gauge ﬁxing Lagrangian density determined by the gauge ﬁxing condition that we want to work with. c. It is worth noting at this point that we have modiﬁed our starting Lagrangian density by a series of formal manipulations. However. the covariant condition (see (12. Otherwise it fails its purpose and will not correspond to an acceptable gauge ﬁxing condition. this modiﬁes the starting theory. (J) (J) (12. the gauge ﬁxing Lagrangian density (see (12.165) .163) Thus we can summarize what we have done so far.157)) has the form LGF = − 1 1 (F [Aµ (x)])2 = − (∂µ Aµ (x))2 . It is. that the S-matrix we obtain is independent of the gauge choice and is unitary and that this formulation leads naturally to the Gupta-Bleuler states as physical states. the gauge ﬁxing Lagrangian density should make the quadratic part of the eﬀective Lagrangian density nonsingular. We then add a ghost Lagrangian density which is determined by an inﬁnitesimal gauge variation of the gauge ﬁxing condition and this is expected to compensate for the change in the theory due to the gauge ﬁxing term (recall from (12. We would show this in the next chapter. (12. namely. To do covariant perturbation theory in the path integral formalism. c. Of course.

we may neglect them and then for ξ = 1 we recognize that our eﬀective Lagrangian density (12.160) for this gauge choice is obtained to be δF [Aθ (x)] µ δθ(y) xδ 4 Sghost = − = − = d4 xd4 y c(x) d4 xd4 y c(x) c(y) θ=0 (x − y) c(y) d4 x Lghost . however.524 12 Yang-Mills theory It is clear that this provides longitudinal components to the quadratic terms in ﬁelds and hence breaks gauge invariance. This . e (12. 4 2ξ (12. we note that F [A(θ) (x)] = ∂µ A(θ)µ (x) = ∂µ Aµ (x) + µ so that (see (12.168) d4 x ∂µ c(x)∂ µ c(x) = where we have neglected total divergence terms. therefore.152)) δF [Aµ (x)] δθ(y) (θ) 1 µ ∂ θ(x) . (12.167) Absorbing the factor 1 into the normalization factor (alternatively e scaling the ghost ﬁelds).125). the ghost ﬁelds are interacting and have to be present. the ghost action (12. (12. To obtain the corresponding ghost Lagrangian density for this gauge choice.169) We note here that the ghost ﬁelds are noninteracting in the case of the Maxwell theory in ﬂat space-time and. Thus our eﬀective Lagrangian density for the Maxwell theory with the choice of a covariant gauge ﬁxing condition becomes 1 1 (J) Leﬀ = − Fµν F µν − (∂µ Aµ )2 + ∂µ c∂ µ c + J µ Aµ . In non-Abelian gauge theories.169) is nothing other than Maxwell’s theory in the Feynman-Fermi gauge (9.166) = θ=0 1 e xδ 4 (x − y).

One way of looking at the ghost ﬁelds is as if they are there to subtract out the unphysical ﬁeld degrees of freedom.170) 4 2 Here we have introduced an auxiliary ﬁeld F which does not have any dynamics. On the other hand each of the ghost ﬁelds.169) as ξ 1 (J) Leﬀ = − Fµν F µν + F 2 −F (∂µ Aµ )+∂µ c∂ µ c+J µ Aµ .169). namely.171) . It is also true that when Maxwell’s theory is coupled to a gravitational ﬁeld. since the ghost ﬁelds and the ghost action are really necessary for the unitarity of the S-matrix.12. the Aµ ﬁeld has four ﬁeld degrees of freedom. Hence we can think of the eﬀective Lagrangian density as having two (4−2×1 = 2) eﬀective ﬁeld degrees of freedom which is the correct number of physical dynamical components as we have already seen within the context of canonical quantization of Maxwell’s theory. On the other hand. This naive counting works pretty well as we will see later.3 Path integral quantization of gauge theories 525 explains why the naive modiﬁcation in (12. To do this let us rewrite the eﬀective Lagrangian density (12. (The ghost degrees of freedom subtract because they have the unphysical statistics. The equation of motion for this ﬁeld leads to the gauge ﬁxing condition and elimination of this ﬁeld through its equations of motion leads to the familiar gauge ﬁxing Lagrangian density in (12. they anticommute even though they are scalar ﬁelds. For example. Furthermore. the ghost ﬁelds automatically couple to the geometry also. Hence omitting the ghost Lagrangian density in such a case would lead to incorrect results. has only one ﬁeld degree of freedom (we will discuss the question of hermiticity of the ghost ﬁelds in the next chapter). (12.164).169) and ask how we can recover the Lorentz gauge from the Lorentz like covariant gauge condition in (12. being a scalar. (12. we see that the equation of motion for F is given by ξF (x) = ∂µ Aµ (x).113) of the non-Abelian gauge theory failed to be suﬃcient unlike in the Maxwell theory (namely. the ghost Lagrangian density was missing).) Let us now go back to the eﬀective Lagrangian density (12. we cannot neglect them even if they are noninteracting particularly when we are doing calculations at ﬁnite temperature.

in the intermediate states we ﬁnd time-like photon states which contribute negatively whereas the longitudinal states contribute an equal positive amount. (12. As a . in the limit ξ → 0. First. The Hilbert space contains only photon states of physical polarization. (12. Here we explicitly eliminate the unphysical (dependent) ﬁeld degrees of freedom and then quantize the physical (independent) ﬁeld degrees of freedom. In this case. We maintain manifest Lorentz covariance and quantize all components of the ﬁeld as independent variables. However.174) and describes Maxwell’s theory in the Lorentz gauge. in the process of eliminating the dependent ﬁeld degrees of freedom we lose manifest Lorentz covariance. Thus the vector space that we work with in this case is much larger than the physical Hilbert space. Namely. this equation leads to the Lorentz condition (or the Landau gauge) ∂µ Aµ (x) = 0. We select out the physical Hilbert space by imposing supplementary conditions on the state vectors in a Lorentz covariant way. we have the operator quantization and again there appear to be two distinct possibilities in this case. From our discussions of the Abelian gauge theories thus far.172) Therefore. we can quantize the Abelian gauge theory canonically. therefore. we see that there are two distinct quantization procedures.170) with ξ = 0) 1 (J) Leﬀ = − Fµν F µν − F (∂µ Aµ ) + ∂µ c∂ µ c + J µ Aµ . The second possibility is to use the GuptaBleuler quantization method to quantize the theory in a manifestly covariant manner.173) ξ→0 4 2ξ which can also be written equivalently as (see (12. 4 (12. Here we modify the theory so that all the ﬁeld degrees of freedom are independent. the eﬀective Lagrangian density in the Lorentz gauge has the form 1 1 (J) Leﬀ = lim − Fµν F µν − (∂µ Aµ )2 + ∂µ c∂ µ c + J µ Aµ .526 12 Yang-Mills theory and. the larger vector space of the theory contains states of indeﬁnite norm and when we do perturbation theory in this formalism.

their contributions cancel out and eﬀectively we are left with only two physical transverse degrees of freedom. (12. In summary. (12. However. The µ correct description for the generating functional.176) In this way. Rather the extra contributions (from the unphysical degrees of freedom) are cancelled by the Faddeev-Popov determinant which we can write as a ghost action (namely. In the path integral formalism.3 Path integral quantization of gauge theories 527 result. therefore.12. As a result. the ghost contributions cancel those from the unphysical gauge ﬁeld degrees of freedom). as the Lagrangian density describing the . we also modify the theory (as in the Gupta-Bleuler method) so that there are no dependent variables and all components of the ﬁeld contribute to any Green’s function of the theory. The second method of quantizing the Abelian gauge theory is through the method of path integrals.175) where AT denotes the transverse physical degrees of freedom. the coeﬃcient matrix of the quadratic terms in the Lagrangian density is singular and. is given by (J ) [AT ] µ Z[Jµ ] = N = N DAT eiS µ DAµ ∆FP [Aµ ]δ(∂ · A(x)) eiS (J ) [A µ] . in this formalism there is no reference to the Hilbert space of the theory. we note that gauge invariance puts a very strong constraint on the structure of the Lagrangian density for the gauge ﬁeld. the Faddeev-Popov determinant or the ghost action can be thought of as the Jacobian in transforming from the physical dynamical ﬁeld variables to all components of the ﬁeld variables through the constraint relation. non-invertible. The generating functional for physical Green’s functions is given by (J ) [AT ] µ Z[Jµ ] = N = N DAT eiS µ DAµ δ(∂ · A(x))eiS (J ) [A µ] . In this formalism the ﬁeld variables are treated as classical variables. In particular. if we take Linv . according to Faddeev and Popov.

in general. then we cannot deﬁne propagators and the entire philosophy of doing perturbative calculations with Feynman diagrams breaks down.67) and (12. The standard covariant gauge ﬁxing (12. in the non-Abelian gauge theory. g (ǫ)a Aa (x) → Aµ (x) = Aa (x) + µ µ (12.178) where ǫa (x) is the inﬁnitesimal parameter of transformation. we add to the gauge invariant Lagrangian density a term which breaks gauge invariance and thereby allows us to deﬁne the propagator for the gauge ﬁeld. As we have seen.177) which is invariant under the inﬁnitesimal gauge transformation 1 Dµ ǫa (x). µ a Fµν (x) = ∂µ Aa (x) − ∂ν Aa (x) − gf abc Ab (x)Ac (x). the Lagrangian density for the gauge ﬁelds is given by (see (12. and to compensate for that we have to add a corresponding Lagrangian density for the ghost ﬁelds following the prescription of Faddeev and Popov.76)) Dµ ǫa (x) = ∂µ ǫa (x) − gf abc Ab (x)ǫc (x). (12. In order to circumvent this diﬃculty. Here the covariant derivative in the adjoint representation as well as the non-Abelian ﬁeld strength tensors are deﬁned as (see (12. adding a gauge ﬁxing Lagrangian density changes the theory.75)) 1 a Linv = − Fµν F µνa .528 12 Yang-Mills theory dynamics of the gauge ﬁeld theory.179) ν µ µ ν with g denoting the coupling constant of the theory. 4 (12. Let us now apply these ideas to the study of the non-Abelian gauge theory.164). Such a term is called a gauge ﬁxing term and any term which makes the coeﬃcient matrix of the quadratic terms (in the action) nonsingular and maintains various global symmetries of the theory is allowed for this purpose. consists of adding to the invariant Lagrangian density a gauge ﬁxing Lagrangian density of the form . On the other hand.

we can write 1 (ǫ)a F a [A(ǫ) (x)] = ∂ µ Aµ (x) = ∂ µ Aa (x) + Dµ ǫa (x) . we can write the ghost action corresponding to this gauge choice as (see (12.181) Here ξ represents a real arbitrary constant parameter known as the gauge ﬁxing parameter. µ µ g so that we have (see (12. 2ξ (12.183) = 1 µ ab 4 ∂ D δ (x − y) g x xµ µ = ∂x ∂xµ δab − gf acb Ac (x) δ4 (x − y). (12.12. rescaling the ghost ﬁelds (to absorb the factor of g ) we can write the ghost Lagrangian density for this choice of gauge ﬁxing to be . Following the prescription of Faddeev and Popov. it is redundant to set it equal to zero at the end since (ǫ) F a [Aµ (x)] is linear in ǫa ) Sghost = = − d4 x Lghost d4 xd4 y ca (x) δF a [Aµ (x)] b c (y). (12.3 Path integral quantization of gauge theories 529 LGF = − 1 (∂µ Aµa (x))2 .160) and note that since ǫa is an inﬁnitesimal parameter. δǫb (y) (ǫ) (12.181) that we are using.182) We note that for the covariant gauge choice (12.122)) δF a [Aµ (x)] δǫb (y) (ǫ) (12.184) µ 1 Consequently.180) which corresponds to a gauge ﬁxing condition of the form F a [Aµ (x)] = ∂µ Aµa (x) = f a (x).

the gauge ﬁxing and the ghost Lagrangian densities modify the original theory in a compensating manner which allows us to deﬁne the Feynman rules of the theory and carry out perturbative calculations. (12. merely correspond to a multiplicative factor of unity (see (12. let us consider the gauge theory of the antisymmetric tensor ﬁeld Aµν (x) .142)) which does not change the physical content of the theory. With all these modiﬁcations.185) where we have dropped total derivative terms (surface terms).4 Path integral quantization of tensor ﬁelds The method due to Faddeev and Popov gives a simple recipe for quantizing gauge theories in the path integral formalism. the total Lagrangian density for the non-Abelian gauge theory can be written in this covariant gauge as LTOT = Linv + LGF + Lghost 1 1 a (∂µ Aµa )2 + ∂ µ ca (Dµ c)a . the gauge ﬁxing and the ghost Lagrangian densities. However. In a deeper sense. 12. As we have mentioned earlier.186) = − Fµν F µνa − 4 2ξ We note that the ghost ﬁelds in the present case are interacting unlike in Maxwell’s theory and. in some cases we have to work through the details of this analysis rather carefully in order to obtain the correct result. in the path integral formulation. therefore.530 12 Yang-Mills theory Sghost = = − = d4 x Lghost µ ab d4 xd4 y ca (x) ∂x Dxµ δ4 (x − y) cb (y) d4 x ∂ µ ca (x) (Dµ c(x))a . cannot be neglected even in ﬂat space-time. As an example of how we should be careful in carrying out the Faddeev-Popov analysis in complicated gauge theories. (12.

since Fµνλ is completely antisymmetric in all the indices. (12. 3. we also note that this Lagrangian density is invariant under the gauge transformation δAµν (x) = ∂µ θν (x) − ∂ν θµ (x).12. µ. ν = 0. Furthermore. (12. 1. we conclude that .191) The independent ﬁeld variables of the theory are A0i . 6 (12. (12.4 Path integral quantization of tensor fields 531 Aµν (x) = −Aνµ (x).188) where the ﬁeld strength tensor corresponds to the totally antisymmetric tensor Fµνλ = ∂[µ A νλ] = ∂µ Aνλ + ∂ν Aλµ + ∂λ Aµν .189) Naive counting shows that Aµν has six ﬁeld degrees of freedom. Let us consider only the free Lagrangian density for this ﬁeld deﬁned by 1 L = − Fµνλ F µνλ . (12.190) so that not all ﬁeld variables are independent. the canonical momenta are antisymmetric (like the ﬁeld variables).192) namely. However. This can be seen from the deﬁnition of the conjugate momenta ∂L = −F 0µν (x).187) Such tensor ﬁelds have been studied in connection with the question of conﬁnement (of quarks in QCD) and are known as Kalb-Ramond ﬁelds. 2. Aij and we note that Πµν = −Πνµ = −F 0µν . ˙ ∂ Aµν (x) Πµν (x) = (12.

6 (12.198) actually represents only two independent conditions. 2. (12. j.194) (12.198) This would seem like three constraints and hence we would naively conclude that this theory has no dynamical degrees of freedom. we note that the theory still possesses a residual (static) gauge invariance under the transformation δAij = ∂i θj (x) − ∂j θi (x). k = 1. on closer inspection we notice that the Coulomb gauge condition (12. i. We can see this by writing out the gauge condition explicitly . Π0µ = −F 00µ = 0. (12. With these conditions we have ˙ Πij = −F0ij = −Aij . However. 3. 3.197) so that we can impose a Coulomb gauge condition of the form ∂i Aij (x) = 0.188) takes the form 1 1 L = − Fµνλ F µνλ = − 3F0ij F 0ij + Fijk F ijk 6 6 1 = − 3Πij Πij + Fijk F ijk .196) Thus it would seem that the theory has truly three degrees of freedom. Π0i = 0. However.532 12 Yang-Mills theory or. (12. 2. and the Lagrangian density (12.195) (12. the corresponding ﬁeld variables are like c-number quantities and we can choose the gauge condition A0i (x) = 0. i = 1.193) Since these momenta identically vanish.

199) ∂2 A21 + ∂3 A31 = 0. Since Fµνλ represent a totally antisymmetric third rank tensor in four space-time dimensions satisfying the Bianchi identity (see (12. ∂1 A12 + ∂3 A32 = 0.203) We recognize this as the Lagrangian density for a free. . Hence the theory has truly one physical degree of freedom and the antisymmetric tensor ﬁeld describes a scalar ﬁeld (it is a gauge theory describing a scalar degree of freedom).189)). 2. 12 2 (12.201) we can represent them also as (this is the dual of the ﬁeld strength tensor in four dimensions) 1 Fµνλ (x) = √ ǫµνλρ ∂ρ φ(x).12.4 Path integral quantization of tensor fields 533 ∂1 A1j + ∂2 A2j + ∂3 A3j = 0. ∂ρ Fµνλ − ∂λ Fρµν + ∂ν Fλρµ − ∂µ Fνλρ = 0. Thus. massless real scalar ﬁeld describing a single degree of freedom.202) where φ(x) represents a real scalar ﬁeld. The fact that the theory has only one degree of freedom can also be seen in the following manner. In this case.200) It is now clear that any two of the three conditions lead to the third condition so that there are only two independent conditions. (12. 3 are (12. 2 (12. the three conditions corresponding to j = 1. ∂1 A13 + ∂2 A23 = 0.188) takes the form 1 1 1 L = − Fµνλ F µνλ = − × ǫµνλρ ǫµνλσ ∂ρ φ∂ σ φ 6 6 2 1 1 ρ = − × (−6δσ )∂ρ φ∂ σ φ = ∂ρ φ∂ ρ φ. the Lagrangiandensity (12. (12.

2ξ LGF = − (12. (12. the starting gauge invariant Lagrangian density has the form 1 Fµνλ F µνλ .205) corresponding to the choice of gauge condition F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x).204) and we add to it the covariant gauge ﬁxing Lagrangian density (ξ is the gauge ﬁxing parameter) 1 (∂µ Aµν )2 .534 12 Yang-Mills theory Let us now look at the path integral quantization of this theory.207) (θ) δF ν [Aµσ (x)] δθ λ (y) (θ) θ=0 µ ν ν µ = ∂xµ δλ ∂x − δλ ∂x δ4 (x − y) ν = (δλ x ν − ∂xλ ∂x ) δ4 (x − y).206) To determine the Lagrangian density for the ghosts we note that F ν [Aµλ (x)] = ∂µ A(θ)µν (x) = ∂µ (Aµν (x) + ∂ µ θ ν (x) − ∂ ν θ µ (x)). According to our earlier discussions. 6 L=− (12. Hence to write the determinant as an action we need ghost ﬁelds which carry a vector index and we have .208) This is a matrix with two vector indices. (12. so that we have (12.

Let us start with the generating functional (J ) [A Z [Jµν ] = N DAµν eiS µν ] . (12. Thus the eﬀective Lagrangian density appears to have 6 − 2 × 4 = 6 − 8 = −2.209) Therefore. (12. the eﬀective Lagrangian density for the theory appears to be given by Leﬀ = L + LGF + Lghost 1 1 = − Fµνλ F µνλ − (∂µ Aµν )2 6 2ξ 1 + (∂µ cµ − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) . The ﬁeld variable Aµν has six degrees of freedom. (12. This does not seem right (does not agree with the analysis from canonical quantization) and we will show now that this is a consequence of our careless application of the Faddeev-Popov procedure.212) . The ghost ﬁelds being vectors have four degrees of freedom each and since they satisfy anticommutation relations.210) Let us now try our naive counting of degrees of freedom of the theory. 2 (12.12.211) eﬀective ﬁeld degrees of freedom. they subtract out ﬁeld degrees of freedom.4 Path integral quantization of tensor fields 535 Sghost = − = − = = 1 2 d4 xd4 y cν (x) δF ν [Aµσ (x)] λ c (y) δθ λ (y) x ν − ∂xλ ∂x ) δ4 (x − y) cλ (y) (θ) ν d4 xd4 y cν (x) (δλ d4 x (∂µ cν (x) − ∂ν cµ (x)) (∂ µ cν (x) − ∂ ν cµ (x)) d4 x Lghost .

(12.215) Thus inserting this identity element into the functional integral.216) Sghost = d4 x (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) . (12.152)).209) 1 2 ). (θ) ∆FP [Aµν ] = det = det ( (12.536 12 Yang-Mills theory The gauge symmetry allows us to choose a gauge condition and we have chosen a Lorentz like gauge condition. the generating functional can be written as Z [Jµν ] = N DAµν Dcµ Dcν (J ) +S ghost × δ (F ν [Aµλ (x)] − f ν (x)) ei(S where as we have seen before in (12. as we have seen (see (12.213) Therefore. (12. we introduce the identity element as (see (12. (12. we would use the ’t Hooft trick to write i − 2ξ Df µ δ (F ν [Aµλ (x)] − f ν (x)) e = e i − 2ξ R d4 x fν (x)f ν (x) R d4 x F ν [Aµλ (x)]Fν [Aµλ (x)] .154)) (θ) ∆FP [Aµν ] x dθ σ (x)δ F ν [Aµλ (x)] − f ν (x) = 1. (12.218) . F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x).217) In the naive application of the quantization procedure. δF ν [Aµλ (x)] δθ σ (y) ν x δσ θ=0 ν − ∂x ∂xσ ) δ4 (x − y) .214) where.

220) (12. we note that the gauge condition ∂µ Aµν (x) = f ν (x). (12. (12. We take into account the transverse nature of f µ (x) exactly like the Maxwell ﬁeld and we apply the ’t Hooft trick by integrating over a transverse weight factor i − 2ξ Df µ δ (F ν [Aµλ (x)] − f ν (x)) e µν R d4 xd4 y fµ (x)P µν (x−y)fν (y) .123)) as µν µ ν ∂x ∂x x P (x − y) = η µν − δ4 (x − y). (9.221) where P (x − y) is the normalized transverse projection operator deﬁned earlier (see. (12.223) Thus one has to use a gauge ﬁxing condition and we choose again a covariant gauge condition 1 1 2 F [fµ (x)] = ∂µ f µ (x) = f (x).4 Path integral quantization of tensor fields 537 However.12. (12. (12. for example.221) has a gauge invariance (just like Maxwell’s theory) of the form δfµ (x) = ∂µ θ(x). the action for fµ in (12. implies that ∂ν ∂µ Aµν = ∂ν f ν (x) = 0. the function f ν (x) has to be transverse and if we neglect to take this fact into account we may get an incorrect result.222) We see that because of the transverse projection operator.219) That is.224) so that we can write .

(12.226) Thus using the ’t Hooft trick we can write Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .P × = i µν fν ) Df ∆FP [fµ ] δ F [fµ (x)] − f (x) e i i − 2ξ (f. (12.228) det xδ 4 (x − y) 1 2 Dc(x) eiS ghost .229) .P ×∆FP [fµ ] e− 2ξ (F (fµ ).f µ )− 2 (c.Fν [Aµλ ])− 2 (c.225) where δF [fµ (x)] δθ(y) (θ) 1 2 ∆FP [fµ ] = det θ=0 = det xδ 4 (x−y) .227) the functional integral becomes Df µ Dc δ (F ν [Aµλ (x)] − f ν (x)) e = Dc e− 2ξ (F i ν [A i i − 2ξ (fµ . (12. Substituting this into (12. (12. (12. c) .f ) Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ . c) i µλ ]. c) = − 2 2 d4 x c(x) c(x).f i µ) ∆FP [fµ ].226)) ∆FP [fµ ] = = we determine 1 1 S ghost = − (c.F (fµ )) i µν fν ) = Df µδ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .227) Furthermore. remembering that (see (12.538 12 Yang-Mills theory ∆FP [fµ ] x (θ) dθ(x)δ F [fµ (x)] − f (x) = 1.

(12.4 Path integral quantization of tensor fields 539 Note here that the ﬁeld c(x) is a real anticommuting scalar ﬁeld.230) SGF = − = − Therefore. This is diﬀerent from the usual Faddeev-Popov ghosts in the sense that the Faddeev-Popov ghosts seem to come in pairs. In ﬂat space-time this ghost Lagrangian density can be seen to give a total divergence and.F µ[Aλσ ]) J +S GF +Sghost +S ghost ).12. (12. . the generating functional takes the form Z [Jµν ] =N =N where 1 2ξ 1 2ξ d4 x Fµ [Aλσ (x)]F µ [Aλσ (x)] d4 x (∂µ Aµν (x)) ∂ λ Aλν . (12. but in the presence of gravitation it cannot be written as a total divergence and is quite relevant.233) This is again not right and the agreement with canonical quantization seems to be worse than before. The ghosts of the present form are known as Nielsen ghosts.232) Counting the ﬁeld degrees of freedom we see that the eﬀective number of degrees of freedom seems to be 6 − 2 × 4 − 1 = −3. may be neglected. we can write the eﬀective Lagrangian density as (J) Leﬀ 1 1 (∂µ Aµν ) ∂ λ Aλν = − Fµνλ F µνλ + J µν Aµν − 6 2ξ + 1 1 (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) − c 2 2 c. (12. therefore.231) DAµν Dcµ Dcν Dc ei(S DAµν Dcµ Dcν Dc ei(S J +S ghost +S ghost i ) e− 2ξ (Fµ [Aλσ ]. Thus using this modiﬁed ’t Hooft weighting factor.

We can again use the FaddeevPopov trick and write ∆FP [cµ ] x (λ) dλ(x) δ F [cµ (x)] − f (x) = 1. the ghost Lagrangian density Lghost is invariant. and δcµ (x) = ∂µ λ(x).234) where λ(x) and λ(x) are anticommuting scalar parameters. µ (12.235) and ∆FP [cµ ] x dλ(x) δ F [c(λ) (x)] −f (x) = 1. (12. ∆−1 [cµ ] = FP x (λ) dλ(x)δ F [cµ (x)] − f (x) (λ) dF δ F [cµ (x)] − f (x) det (λ) = x δF [cµ (x)] δλ(y) (12. (12. under δcµ (x) = ∂µ λ(x). (12. (12.236) Let us choose for simplicity Lorentz like gauge conditions F and F = ∂µ cµ (x) =f (x). namely. the ghost Lagrangian density also possesses a gauge invariance.540 12 Yang-Mills theory We notice at this point that although we have ﬁxed up the gauge invariance of the gauge ﬁxing Lagrangian density. λ=0 .238) = ∂µ cµ (x) = f (x).237) We are now ready to calculate the Faddeev-Popov determinants associated with these gauge ﬁxing conditions.239) (λ) = det δF [cµ (x)] δλ(y) .

the Faddeev-Popov term is an inverse determinant and. (12. we cannot use weight factors of the forms i − 2χ e R e e d4 x f (x)f (x) . the appropriate weight factor in this case would correspond to .237) and (12. f and f are Grassmann (fermionic) functions. . e i − 2χ R e e e e d4 x f (x) f (x) . the ghost ﬁelds c and c behave like commuting scalars. (12. consequently. We note that contrary to the usual case. the Jacobian for a change of variables is the inverse of the determinant.241) We note here that since we are writing an inverse determinant as an action. the ’t Hooft trick needs to be carried out rather carefully. in the present case we can write 4 −1 ∆FP [cµ ] = = det xδ (x − y) c e) c Dc Dc e−i(e.12. Rather.4 Path integral quantization of tensor fields 541 so that (λ) −1 λ=0 −1 ∆FP [cµ ] = = δF [cµ (x)] det δλ(y) det xδ 4 (x − y) .242) The gauge conditions (12.238) correspond to fermionic conditions. Therefore.240) where we have used the fact that for anticommuting variables. namely. (12.243) since the exponents vanish (because of the fermionic nature of the variables). Namely. e . (12. Similarly we can show that 4 −1 ∆FP [cµ ] = = det xδ (x − y) Dc Dc e ” “ e c c e −i e.

c and c being commuting scalars contribute positively to the counting of the number of degrees of freedom. (12. commuting. (12.244) Thus putting in these identity elements into the functional integral for the generating functional and using the ’t Hooft trick with the weight factor (12. (12. c : 1 degrees of freedom. 2χ c (12.246) 1 1 = − Fµνλ F µνλ − (∂µ Aµν )2 + J µν Aµν 6 2ξ 1 1 + (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) − c 2 2 1 − (∂ν cν ) (∂µ cµ ) − c c − c c. c : −1 degrees of freedom. c : 1 degrees of freedom. cµ : −4 degrees of freedom. The other .245) d4 x Leﬀ . (12.248) The ghosts c. c : 1 degrees of freedom.244) the generating functional takes the form Z [Jµν ] = N where Seﬀ and Leﬀ = DAµν Dcµ Dcν DcDc Dc D c D c eiSeﬀ . cµ : −4 degrees of freedom. c. c : 1 degrees of freedom.542 12 Yang-Mills theory e i − 2χ R e e e d4 x f (x) f (x) . anticommuting.247) We are now ready to count the number of eﬀective ﬁeld degrees of freedom in the theory. Aµν : 6 degrees of freedom.

3. This matches exactly with the counting of the degrees of freedom from the canonical quantization. 5. Theoretical and Mathematical Physics 1. C. Physics Letters 93B. Faddeev and V. 378 (1971). Veltman. McGraw-Hill. Huang. Gribov.5 References 1. P. ’t Hooft and M. 6. 10. Veltman. Physical Review 96. N. Leptons and Gauge Fields. Physics Letters 25B. Diagrammar.249) degree of freedom. Nuclear Physics B139. Qualks. Zuber. V. Itzykson and J-B. New York (1965). Singapore (1982). D. Yang and R. McGrawHill. D. 4. R. (12. Mills. ’t Hooft and M. CERN preprint (1973). Hibbs and R. 2. 7.5 References 543 way of saying this is that these are ghosts of the ghosts cµ and cµ respectively and hence they contribute just the opposite way from cµ and cµ . Utiyama. N. . Siegel. 9. Feynman. G. Physical Review 101. 8. L. 11. 29 (1967). World Scientiﬁc. G. 1 (1978). L. 170 (1980). New York. W. we see that eﬀectively the theory has 6 − 2 × 4 − 1 + 4 × 1 = 6 − 8 − 1 + 4 = 1. Quantum Mechanics and Path Integrals. Nuclear Physics B44. 191 (1954). C. 1597 (1956).12. 12. 1980. 12. N. L. Mohapatra. 1 (1970). 189 (1972). Popov. Physical Review D4. Faddeev. R. Quantum Field Theory. N. K. R. A. Thus counting the degrees of freedom.

Gross. Singapore (1997). Finite Temperature Field Theory. Field Theory:A Path Integral Approach. 15. F. World Scientiﬁc.544 12 Yang-Mills theory 13. A. Singapore (2006). Das. New York (1993). 14. John Wiley. Das. A. Relativistic Quantum Mechanics and Field Theory. World Scientiﬁc. .

(13.1 BRST symmetry As we have seen in (12.Chapter 13 BRST invariance and its consequences 13.34) of the original theory. remembers the gauge invariance of the original theory. therefore. the total Lagrangian density for the non-Abelian gauge theory (Yang-Mills theory) has the form LTOT = Linv + LGF + Lghost . However. with gauge ﬁxing and ghost terms. µ (13.67)) (Dµ c)a = ∂µ ca − gf abc Ab cc .3) where g denotes the coupling constant of the theory.1) 2 + ∂ µ ca (Dµ c)a . The total Lagrangian density has been gauge ﬁxed and. develops a global fermionic symmetry which. in some sense.2) is invariant under the global transformations 545 . It is easy to check that the total Lagrangian density (13. the total Lagrangian density.181) is given by 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ (13. which for the covariant gauge in (12.2) Here ξ represents the arbitrary gauge ﬁxing parameter and the covariant derivative in the adjoint representation is deﬁned as (see (12. does not have the gauge invariance (12.186).

ca .546 13 BRST invariance and its consequences δAa = µ δca δca ω (Dµ c)a .2 respectively. (13. The invariance of the Lagrangian density (13.50)). only when the ghost equation of motion is used). .2) can be seen by ﬁrst noting that (recall that the ghost ﬁelds are Grassmann variables) δ (Dµ ca ) = Dµ δca − gf abc δAb cc µ = ω Dµ f abc cb cc − ωf abc Dµ cb cc = 0. while for ca it is µ true only on-shell (namely. g ω abc b c = f cc.7) for the ﬁelds φa = Aa . Similarly. ca independent of the parameters of transµ formations where δ1.2 correspond to transformations with the parameters ω1. We note that the nilpotency of the transformations holds oﬀ-shell only for the ﬁelds Aa . = 6 (13.5) Here we have used the Jacobi identity for the symmetry algebra (see (12.4) we have δ2 δ1 φa = 0. This shows that under two successive transformations of the kind (13.6) when the ghost equation of motion is used. = − µ gξ (13. 2 1 ω δ f abc cb cc = f abc δcb cc = f abc f bpq cp cq cc 2 2 ω abc bpq f f + f abp f bqc + f abq f bcp cp cq cc = 0. ca . we obtain δ ∂ µ Aa = µ ω µ ∂ Dµ ca = 0. g (13. 2 ω ∂ µ Aa .4) where ω is an arbitrary anti-commuting constant parameter of the global transformations.

δLinv = 0.2) under the transformations (13. therefore. is slightly unpleasant in the sense that the nilpotency of the anti-ghost ﬁeld transformation holds only on-shell.2) is invariant under the global transformations (13.4) can now be easily checked. Since this will also be quite useful for our latter discussions.13. the invariant Lagrangian density is trivially invariant under these transformations. In this derivation.4) with an anti-commuting constant parameter.8) Consequently. This is known as the BRST (Becchi-Rouet-Stora-Tyutin) transformation for a gauge theory and arises when the gauge ﬁxing and the ghost Lagrangian densities have been added to the original gauge invariant Lagrangian density. let us recall that the gauge ﬁxing Lagrangian density in (13. namely. we have used the fact that δ (Dµ c)a = 0 which we have seen in (13.1 BRST symmetry 547 The invariance of the Lagrangian density (13.170) we can write the gauge ﬁxing Lagrangian density by introducing an auxiliary ﬁeld.34)) with the parameter ǫa (x) = ωca (x) and. ν gξ (13. however. This shows that the action for the total Lagrangian density (13. we need to worry only about the changes in the gauge ﬁxing and the ghost Lagrangian densities which lead to 1 δ (LGF + Lghost ) = − (∂ ν Aa ) ∂ µ δAa + (∂ µ δca ) Dµ ca ν µ ξ ω µ ν a ω ∂ (∂ Aν ) Dµ ca = − (∂ ν Aa ) ∂ µ Dµ ca − ν gξ gξ ω = −∂ µ ∂ ν Aa Dµ ca .5). As we have seen earlier within the context of the Abelian gauge theory in (12. The present formulation of the BRST symmetry. First. Generally. this is a reﬂection of the fact that the theory is lacking in some auxiliary ﬁeld variables and once the correct auxiliary ﬁelds are incorporated the symmetry algebra will close oﬀ-shell (without the use of equations of motion).2) can also be written equivalently as . (13. we note that the transformation for Aa can really be thought of as an inﬁnitesiµ mal gauge transformation (see (12.9) so that the action is invariant.

2 ω = − F a. but it is this form that is very useful as we will see.) It is clear from the form of LGF in (13. we recover the original gauge ﬁxing Lagrangian density (up to a total divergence term). the BRST transformations in (13. g (13. µ (13. (13.13) δF a = 0. (13. Among other things LGF as written above allows us to take such gauge choices as the Landau gauge which corresponds to simply taking the limit ξ = 0. and it is straightforward to check that these transformations are nilpotent oﬀ-shell.548 13 BRST invariance and its consequences LGF = ξ a a F F + (∂ µ F a )Aa . δ2 δ1 φa = 0.14) .10) diﬀers from that in (12.10) that the equation of motion for the auxiliary ﬁeld takes the form ξF a = ∂ µ Aa .170) by a total divergence. g ω abc b c = f cc. namely.11) and when we eliminate F a from the Lagrangian density using this equation. µ 2 (13.4) take the form δAa = µ δca δca ω (Dµ c)a . The total Lagrangian density can now be written as LTOT = Linv + LGF + Lghost ξ 1 a = − Fµν F µνa + F a F a + ∂ µ F a Aa + ∂ µ ca (Dµ c)a .10) where F a is an auxiliary ﬁeld. (The form of the gauge ﬁxing Lagrangian density in (13.12) µ 4 2 In this case.

In some sense the BRST transformations. g ω F a + gf abc cb cc . F a . replace the original gauge invariance of the theory and play a fundamental role in the study of non-Abelian gauge theories.9)). g ω abc b c f cc. F a represents µ the missing auxiliary ﬁeld that we had alluded to earlier. ca .8) and the auxiliary ﬁeld F a does not transform at all which leads to δLTOT = δ (LGF + Lghost ) = = ∂ µ F a δAa + ∂ µ δca (Dµ c)a µ ω ω µ a ∂ F (Dµ c)a − ∂ µ F a (Dµ c)a = 0. 2 (13. We note here that the BRST and the anti-BRST transformations are not quite symmetric in the ghost and the anti-ghost ﬁelds which is a .1 BRST symmetry 549 for all the ﬁeld variables φa = Aa . we will not pursue this symmetry further in our discussions.12) invariant. There is also a second set of fermionic transformations involving the anti-ghost ﬁelds of the form ω (Dµ c)a . Therefore.4). since these do not lead to any new constraint on the structure of the theory beyond what the BRST invariance provides.15) Unlike in the formulation of BRST variations without the auxiliary ﬁeld in (13. These are known as the anti-BRST transformations. which can also be easily checked to leave the total Lagrangian density (13.16) δAa = µ δca = δca = δF a = −ωf abc F b cc . ca .13. However. which deﬁne a residual global symmetry of the full theory. g g (13. here we see that the total Lagrangian density is invariant under the BRST transformations (as opposed to the Lagrangian density changing by a total divergence in (13. We note that Linv is invariant under the BRST transformation as we had argued earlier in (13.

it allows us to carry out covariant quantization of the non-Abelian gauge theory. δca = −ǫca .2) or (13. commuting inﬁnitesimal parameter and the generator of this symmetry transformation merely corresponds to the ghost number operator.12). it corresponds to the operator that counts the ghost number of the ﬁelds. In the covariant gauge in Maxwell’s theory.2 Covariant quantization of Yang-Mills theory The presence of the BRST symmetry and the ghost scaling symmetry in the gauge ﬁxed Yang-Mills theory leads to many interesting consequences. has to do with the particular hermiticity properties that the ghost and the anti-ghost ﬁelds satisfy for a consistent covariant quantization of the theory which we will discuss in the next section. Therefore.181). we note here that these fermionic symmetries arise naturally in a superspace formulation of gauge theories. Let us recall that the vector space of the full theory in the covariant gauge (12. We recall that the physical space must be selected in such a way that it remains invariant under the time evolution of the system.) 13. the total Lagrangian density (13. for example. which is normally associated with the number operator. Without going into details. as we have emphasized several times by now. namely.17) with all other ﬁelds remaining inert. (13.550 13 BRST invariance and its consequences reﬂection of the asymmetric manner in which these ﬁelds occur in the ghost Lagrangian density in (13. the physical Hilbert space needs to be properly selected for a discussion of physical questions associated with the non-Abelian gauge theory.12) is also invariant under the inﬁnitesimal bosonic global symmetry transformations δca = ǫca . (The fact that these transformations are like scale transformations and not like phase transformations. Here ǫ represents a constant. For example. contains many more states than the physical states alone. In addition to these two anti-commuting symmetries. This is known as the ghost scaling symmetry of the theory. We have already seen that the naive Gupta-Bleuler quantization does not work in the nonAbelian case. we .

the conditions in (13.130) in the covariant gauge (12. therefore. In the Abelian theory. On the other hand.162)) ∂ µ A(+) (x)|phys = 0.18). (13.2 Covariant quantization of Yang-Mills theory 551 have seen that the states in the physical space are selected as the ones satisfying the Gupta-Bleuler condition (see (9.118). QBRST . whether they can reduce the vector space suﬃciently enough to coincide with the physical space). The corresponding operator in a non-Abelian theory. “(+)”. the Gupta-Bleuler condition works because ∂ µ Aµ satisﬁes the free Klein-Gordon equation (9. Qc .) o Thus. are conserved and hence can be used to deﬁne a physical Hilbert space which would remain invariant under the time evolution of the system. as we have seen in (12. not clear a priori if the conditions (13. we can identify the physical space of states of the gauge theory as satisfying (we note that at this point this only deﬁnes a subspace of the total vector space and we still have to show that this subspace indeed coincides with the physical Hilbert space) QBRST |phys Qc |phys = 0. does not satisfy a free equation and hence it is not a suitable operator for identifying the physical subspace in a time invariant manner.19) are suﬃcient to reproduce even the GuptaBleuler condition in the case of the Abelian theory (namely. the generators of the BRST symmetry. (QBRST and Qc are the charges constructed from the N¨ther o current for the respective transformations whose explicit forms can be obtained from the N¨ther procedure and will be derived below. and the ghost scaling symmetry.13. stands for the positive frequency part of the ﬁeld.19) would appear to correspond to only two conditions and not an inﬁnite number of conditions as we have seen is the case with the Gupta-Bleuler condition in (13.19) Note that even in the case of Maxwell’s theory. µ (13. = 0. We recognize that even though this looks like one condition.18) where the superscript. in reality it is an inﬁnite number of conditions since it has to hold for every value of the coordinates. .164) and hence the physical space so selected remains invariant under time evolution. It is.

552 13 BRST invariance and its consequences To see that these conditions indeed lead to the Gupta-Bleuler condition in Maxwell’s theory.20) where we have used the fact that the auxiliary ﬁeld does not transform under the BRST transformations and the ﬁelds Aa . From this. (13. 2 (13.22) d3 x − (∂ 0 F a )ca + F a (D 0 c)a + 0 d3 x Jc = ˙ d3 x ca ca − ca (D 0 c)a . 2 (13. . g abc µ a b c f (∂ c ) c c . g 2 µ(ǫ) Jc = δca ∂LTOT ∂LTOT + δca ∂∂µ ca ∂∂µ ca = −ǫca (D µ c)a − ǫca ∂ µ ca = ǫ ((∂ µ ca )ca − ca (Dµ c)a ) . we can obtain the BRST as well as the ghost scaling current densities without the parameters of transformation to correspond to µ JBRST = F µνa (Dν c)a + F a (D µ c)a + µ Jc = (∂ µ ca ) ca − ca (Dµ c)a . let us note that the N¨ther current o densities associated with the BRST transformation as well as the ghost scaling transformation have the forms (recall that we use left derivatives for Grassmann variables and that ω is an anticommuting parameter) JBRST (x) = (ω)µ ∂LTOT ∂LTOT a ∂LTOT a a ∂LTOT δAν + δF + δca a + δc ∂∂ ca a a ∂∂µ Aν ∂∂µ F ∂∂µ c µ = −F µνa δAa + δca (D µ c)a − δca (∂ µ ca ) ν ω g = − F µνa (Dν c)a + F a (Dµ c)a + f abc (∂ µ ca ) cb cc .21) The corresponding conserved charges can also be obtained from these current densities and take the forms QBRST = = = Qc = 0 d3 x JBRST d3 x F 0i a (Di c)a + F a (D 0 c)a + g abc ˙ a b c f c cc 2 g abc ˙ a b c f c cc . F a are inert µ under the scaling of ghost ﬁelds.

In particular. In principle. the BRST and the ghost scaling charge operators can be obtained from (13. (13.19) Qc |phys = 0. denoting the physical states of the theory as |phys = |Aµ ⊗ |n. (13.22) to correspond to ← → d3 x F ∂ 0 c.12). this allows for states containing an equal number of ghost and anti-ghost particles. (13.24) QBRST = Qc = ˙ d3 x − F c + F c = ˙ ˙ ˙ d3 x cc − cc = − If we use the ﬁeld decomposition for the ﬁelds and normal order the charges (so that the annihilation/positive frequency operators are to the right of the creation/negative frequency operators). ← → d3 x c ∂ 0 c.23) following from (13. we note that for the Abelian theory where f abc = 0 and there is no internal index “a”.26) implies that the physical states must have (net) zero ghost number. n .27) . (13. Thus.25) Let us note that the condition (13. the BRST charge has the explicit form (we do not show the normal ordering explicitly) QBRST = i d3 k c(−) (−k)F (+) (k) − F (−) (−k)c(+) (k) .2 Covariant quantization of Yang-Mills theory 553 where we have integrated by parts the ﬁrst term (in QBRST ) and have used the equation of motion for the gauge ﬁeld Dµ F µνa = −∂ ν F a . (13.13.

28) c(+) (k)|n. then this implies that (13.554 13 BRST invariance and its consequences where n.19) really select out the subspace of physical states.30) (13. |phys = |Aµ ⊗ |0. 0 = |Aµ . for every momentum mode k). n denote the (equal) number of ghost and anti-ghost particles. and F (+) (k)|Aµ = 0. we note that if the physical states have to further satisfy the condition QBRST |phys = i d3 k c(−) (−k)F (+) (k) − F (−) (−k)c(+) (k) |Aµ ⊗ |n.31) where we have used the equation of motion for the auxiliary ﬁeld (see (13.12) that we can obtain the canonical momenta conjugate to various ﬁeld variables of the theory to correspond to . the physical states should have no ghost particles (the number of ghost and anti-ghost particles have to be the same by the other physical condition). n = 0. and must further satisfy (with k0 = |k|) F (+) (k)|phys = 0 = kµ Aµ(+) (k) |phys . n = 0. This is precisely the Gupta-Bleuler condition in momentum space and this derivation shows how a single condition can give rise to an inﬁnite number of conditions (in this case.11)). Thus.29) Namely. we note from the form of the Lagrangian density (13. To investigate systematically whether the physical state conditions in (13.19) are the right ones even for the non-Abelian theory. we feel conﬁdent that the physical state conditions in (13. (13. (13.

(13. 0 ˙ ∂F a ∂LTOT ˙ = −ca . the (anti) commutation relations (13. t). Πb (y. the BRST parameter ω is seen to be anti-Hermitian ω † = −ω.34) With this choice. The equal-time 0 canonical (anti) commutation relations for the theory can now be written as ( = 1) Aa (x. t) c ca (x. In particular.13. ˙ ∂ ca (13. t) ca (x. Aa plays the role of the momentum conjugate to F a .32) Here we have used left derivatives for the anti-commuting ghost ﬁelds. t) i F a (x. Πb (y.2 Covariant quantization of Yang-Mills theory 555 Πia = Πa = Πa = c Πa = c ∂LTOT = −F 0ia . Πjb (y.33) have to satisfy the proper hermiticity properties etc. ∂ ca ˙ ∂LTOT = (D0 c)a . = iδab δ3 (x − y). t). (13. we see that in this formulation with the auxiliary ﬁeld.) are given by (ca )† = ca . Πb (y. (ca )† = −ca . = iδab δ3 (x − y). t).33) + The hermiticity conditions for the ghost ﬁelds which arise out of various consistency conditions (for example.35) . the Lagrangian density and the conserved charges have to be Hermitian. = iδab δ3 (x − y). (13. ˙ ∂ Aa i ∂LTOT = Aa . t). t) c + j = iδab δi δ3 (x − y).

we see that we can think of iQc as the ghost number operator (with the ghost number for ca being negative). (13. Qc ] = −ica (x).36) and we can calculate the algebra of charges as well as various other relations of interest using the (anti) commutation relations (13.38) is that . c QBRST = − Qc = − (13. QBRST ]+ = 2Q2 BRST = 0.13) and (13. therefore.17). The third is a statement of the fact that the BRST charge carries a ghost number of unity. Therefore.14)).34) for the ghost ﬁelds. Qc ] = 0. c c 2 a d3 x (Πa ca + ca Πc ) .33) [QBRST . that (this basically describes the behavior of the ghost ﬁelds under a scaling (13.38) This algebra can also be derived directly from the transformation laws for the ﬁeld variables in (13. The algebra of the conserved charges also follows in a straightforward manner from (13. see (13. for example.556 13 BRST invariance and its consequences and we note that the currents (13. We note. The conserved charges (13. [Qc . [QBRST .38) simply reiterates in an operator language the fact that the BRST transformations are nilpotent (see (13.22) are Hermitian with the assigned hermiticity conditions (13.36) and (13. Qc ] = −iQBRST . the conserved charges (13.17)) [ca (x). An immediate consequence of the algebra in (13. The second equation implies that the ghost scaling symmetry is Abelian.21) and.37) [ca (x). (13.33) for the ﬁeld variables.22)) g d3 x Πia (Di c)a − F a Πa + f abc Πa cb cc .22) can now be expressed in terms of the ﬁelds and the conjugate momenta as (before integrating the ﬁrst term by parts in QBRST . Qc ] = ica (x). The ﬁrst equation in (13.

Consequently. therefore. First. QBRST . [QBRST .40). (13. it is truly a BRST singlet (invariant) state. if a state |Ψ cannot be written as ˜ |Ψ = QBRST |Ψ . (13. (13. eπQc + = 0. (we assume that the vacuum state is a BRST singlet (invariant) state) .42) ˜ for some |Ψ and still satisﬁes the physical state condition.39) which is quite useful (particularly) in dealing with gauge theories at ﬁnite temperature. then. (13. or. namely. As we have discussed earlier.2 Covariant quantization of Yang-Mills theory 557 QBRST eπQc = eπ(Qc −i) QBRST = −eπQc QBRST . for every |Ψ ∈ Vphys .40) We emphasize again that such an identiﬁcation of the physical subspace is invariant under the time evolution of the system since QBRST commutes with the Hamiltonian since the BRST transformations deﬁne a symmetry of the full theory.13.12) contains various unphysical states as well as states with negative norm in addition to the physical states. Namely. H] = 0. As is clear from our earlier discussions. must represent a truly gauge invariant ﬁeld variable. the metric of this space and the inner product become indeﬁnite and a probabilistic description of the quantum theory is lost unless we can restrict to a suitable subspace with a positive deﬁnite inner product.41) We note that there are two possible kinds of states which will satisfy the physical state condition (13. The ﬁeld operator which creates such a state must necessarily commute with QBRST and. we can select out a subspace Vphys by requiring that states in this space are annihilated by QBRST . we have QBRST |Ψ = 0. the total vector space V of the complete gauge theory (13.

has a diﬀerent character. (13. Such states would. Ψ] = 0. therefore. therefore. as we have seen in (13. such states would have positive norm. Ψ|0 = |Ψ . The nilpotency of QBRST also implies that all such states would have zero norm because (QBRST is Hermitian with our choice of hermiticity conditions (13.45) and transverse ﬁelds would represent such operators which are gauge invariant asymptotically. However. (13. therefore. This implies that (13.47) Such a state would also be orthogonal to any physical state (either of the ﬁrst or the second kind) satisfying the physical state condition .34) for the ghost ﬁelds. correspond to truly physical states of the theory and since gauge invariant degrees of freedom have physical (non-negative) commutation relations. namely.38)).46) We see that the physical state condition is trivially satisﬁed in this case because of the nilpotency of QBRST (see (13. also satisﬁes the above relation. (Note that the auxiliary ﬁeld does not transform under a BRST transformation and.558 13 BRST invariance and its consequences QBRST |Ψ = QBRST Ψ|0 = [QBRST .13). (13. (13. it can be written as the BRST variation of the anti-ghost ﬁeld and.44) [QBRST .43) where we have assumed that the operator Ψ creates the state |Ψ from vacuum. Ψ] |0 = 0.40) can be written in the form ˜ |Ψ = QBRST |Ψ .) The second class of states which would satisfy the physical state condition (13.) ˜ ˜ Ψ|Ψ = Ψ|QBRST QBRST |Ψ = 0.

we note that we can write P (0) = ½ − P ′ . (13.13) that the auxiliary ﬁeld can be written as a BRST variation of the anti-ghost ﬁeld as (note that QBRST is the generator of the BRST transformations and generates the transformations in (13. The true physical states. If we denote all the states of Vphys of the second kind by V0 . deﬁning P (n) as the projection operator onto the n-unphysical particle sector.13. be identiﬁed as belonging to the V quotient space V phys = phys and will have positive deﬁnite norm. be decomposed into states containing ﬁxed numbers of unphysical particles. then ˜ Ψ′ |Ψ = Ψ′ |QBRST |Ψ = 0. µ We recall from (13. this would contain all the zero norm states which would be orthogonal to Vphys itself. Here. of course.13) through (anti) commutation with the appropriate ﬁeld variables) F a ∼ [QBRST .51) . which satisfy the physical state condition and are of the ﬁrst kind can. then. ca ]+ . V0 Every state in Vphys can. (13. with no unphysical particles) and from the completeness of the projection operators in (13.49) P (n) P (m) = δnm P (n) . we have ∞ n=0 P (n) = ½.48) which follows from the physical state condition.49).50) It is clear from the deﬁnition in (13. the longitudinal components of Aa and the auxiliary ﬁelds. by unphysical. therefore. (13.2 Covariant quantization of Yang-Mills theory 559 (13. we mean states containing quanta of the ghost ﬁelds. Thus.40) because if |Ψ represents a physical state (of either kind) while |Ψ′ is a state of the second kind. (13.49) that P (0) projects onto the truly physical states of the theory (namely.

= 0. R]+ = 0. the diﬀerence between the two lies in the fact that P ′ must necessarily involve unphysical ﬁelds since it projects onto unphysical states and. it must have the form (for some fermionic operator R. however. whose explicit form is not important for our discussions) P ′ = [QBRST . then. we can write P (n) = QBRST . R]+ . R(n) . P (0) QBRST . therefore. project onto the zero norm space V0 when acting on states |Ψ . we can even show that for every P (n) . (13. so that QBRST . P ′ = QBRST . (13.54) (13. Consequently. therefore.53) However. must be gauge invariant and will commute with QBRST . of course. n ≥ 1. In fact. but this is not very illuminating. P ′ = 0. (13.560 where ∞ n=1 13 BRST invariance and its consequences P = ′ P (n) . is the fact that P (0) projects onto the true physical states and.56) + P′ It is now easy to show that P (n) . and. Namely. for n ≥ 1. that P ′ must also commute with QBRST . It follows. What is interesting. cannot be truly gauge invariant. therefore. [QBRST .55) which follows from the nilpotency of QBRST . we have QBRST .52) We can. (13. construct the actual forms of all the projection operators explicitly.

R]+ |Ψ = 0. deﬁned by the physical state condition. This V0 would correspond to the true physical subspace of the total vector space. This also makes it clear that the V norm of a state |Ψ ∈ V phys = phys would be positive deﬁnite.40). shows that Vphys . and the norm of any such state is. We will only outline the proof of unitarity in this section. For example.59) This.57) + Ψ′ | [QBRST . which follows from the physical state condition (13. We note that. Ψ′ |P (n) |Ψ Ψ′ |P ′ |Ψ = = Ψ′ | QBRST . The question of the unitarity of the S-matrix in a gauge theory can be formulated in the following manner. then. 13.40) (as well as the hermiticity of QBRST ). This completes the covariant quantization of the non-Abelian theory and shows that the generalization of the supplementary condition (Gupta-Bleuler condition) to the non-Abelian theory can be achieved as a consequence of the BRST symmetry of the theory and is given by (13. let |Ψ .58) (13.13.19). R(n) |Ψ = 0.34). then.3 Unitarity The BRST invariance of a gauge theory is quite important from yet another consideration. (13. with the hermiticity assignments for the ghost ﬁelds in (13. Any vector |Ψ ∈ Vphys can now be written as |Ψ = P (0) |Ψ + P ′ |Ψ . therefore. has a positive semi-deﬁnite norm as we should have for a physical vector space and the value of the norm depends on the truly physical component of the state. obtained to be Ψ|Ψ = Ψ|P (0) |Ψ ≥ 0. It leads to a formal proof of unitarity of the theory when restricted to the subspace of the physical Hilbert space. |Ψ′ ∈ Vphys . (13. the total Lagrangian .3 Unitarity 561 which satisfy the physical state condition (13.

= = Ψ|AP (0) B|Φ Ψ|A(½ − P ′ )B|Φ = Ψ|AB|Φ . (13. |Ψ and |Ψ . any two operators which do not take us out of Vphys ) Ψ|P (0) |Φ Ψ|(½ − P ′ )|Φ = Ψ|Φ .64) The two states.e.. the S-matrix is BRST invariant since the full theory is. This question can be systematically analyzed as follows. P (0) |Ψ = P (0) |Ψ = P (0) |Ψ = |Ψ .63) where we note that in the space V phys .61) Given these the question that we would like to understand is whether one can deﬁne an operator Sphys which would act and correspond to the S-matrix in the physical subspace of states of the theory and which will also be unitary. so that [QBRST . (13. (13.60) Furthermore. S † S = SS † = ½.562 13 BRST invariance and its consequences density (13.12) is Hermitian so that the S-matrix of the theory is formally unitary. namely. diﬀer only by a zero norm state which is orthogonal to every state in Vphys so that the inner product of any two such states is the same .57). Furthermore. (13. |Φ ∈ Vphys and any two operators A and B acting on Vphys (i.62) which follow from (13. we can deﬁne a unique state |Ψ ∈ V phys as |Ψ = P (0) |Ψ . S] = 0. 2 (13. we note that given any state |Ψ ∈ Vphys . Let us note here from the discussions of the last section that for any |Ψ . P (0) acts as the identity operator. namely.

it is clear that we can deﬁne the S-matrix which acts on the physical space V phys as satisfying P (0) S = Sphys P (0) . (13.66) and (13.68) where we have used (13.62) as well as the formal unitarity of S (13. (13. In a similar manner. Given this.70) .13. Sphys |Ψ = Sphys P (0) |Ψ = P (0) S|Ψ = S|Ψ . (13. the physical state condition (13. also automatically leads to a formal proof of unitarity of the S-matrix in the subspace of the truly physical states of the theory.62). Furthermore. |Ψ = P (0) |Ψ ∈ V phys as deﬁned in (13. 2 (13.3 Unitarity 563 Ψ|Φ = Ψ| P (0) |Φ = Ψ|P (0) |Φ = Ψ|Φ . (13. (13.40) which naturally follows from the BRST invariance of the theory.63). It follows from (13.69) In other words.67) It is clear that since S is BRST invariant.65) where the last line follows from the ﬁrst of the conditions in (13.60). it will leave the space Vphys invariant and hence Sphys will not take a state out of V phys . we now have † Ψ|Sphys Sphys |Φ † Ψ|P (0) Sphys Sphys P (0) |Φ = = = Ψ|S † P (0) S|Φ Ψ|S † S|Φ = Ψ|Φ = Ψ|Φ .66) such that for |Ψ ∈ Vphys .68) that † Sphys Sphys = ½. we can also show that † Sphys Sphys = ½.

It now follows from the physical state condition (13. we note that these extra terms in the Lagrangian density (namely. − ca F a − ∂ µ ca Aa µ 2 + (13. we can show that the gauge ﬁxing and the ghost Lagrangian densities lead to no physical consequences. in fact. We can also show that all the physical matrix elements of the theory are independent of the choice of the gauge ﬁxing parameter ξ in the following manner (the BRST variation denoted is with the parameter of transformation taken out) . This is particularly important since (as we have emphasized before) we have modiﬁed the starting theory through a series of formal manipulations and we should show that this has not changed the physical results of the theory. = 0.40) that phys| (LGF + Lghost ) |phys′ = −g phys| QBRST . LGF + Lghost = ξ a a F F + (∂ µ F a )Aa + ∂ µ ca (Dµ c)a µ 2 ξ = gδ − ca F a − ∂ µ ca Aa µ 2 ξ . To show this. namely. be written as a BRST variation (with the parameter of transformation taken out). ξ a a c F + ∂ µ ca Aa µ 2 + |phys′ (13. the gauge ﬁxing and the ghost Lagrangian densities) can.564 13 BRST invariance and its consequences As another consequence of the BRST invariance of the theory. = g QBRST .71) Here we have used the fact that the BRST charge is the generator of the BRST transformations so that the transformations for any fermionic operator is generated through the anti-commutator of the operator with the generator.72) This shows that the terms added to modify the original Lagrangian density have no contribution to the physical matrix elements of the theory.

However.120).73) where we have used the fact that the vacuum belongs to the physical Hilbert space of the theory and as such is annihilated by the BRST charge. such a simple diagrammatic derivation does not carry over to non-Abelian gauge theories and they are best described systematically within the context of path integrals which we will do next. However. the vacuum functional is given by 0|0 J = Z[J] = eiW [J] = N Dφ eiS J [φ] . (13. (13. ca F a ]+ |0 J = 0.75) . let us recapitulate brieﬂy some of the essential concepts from path integrals (a detailed discussion is beyond the scope of these lectures). We have already given a simple diagrammatic derivation of such identities in the case of QED in section 9. These are known as the Ward-Takahashi identities (in the Abelian case) or the SlavnovTaylor identities (in the non-Abelian case) of the theory and are quite essential in establishing the renormalizability of gauge theories. In this case. before we go into the actual derivation.13.4 Slavnov-Taylor identity 565 ∂ 0|0 ∂ξ J = ∂Z[J] i = 0| ∂ξ 2 ig 2 ig 2 d4 x F a F a |0 J J = − = − d4 x 0|δ (ca F a ) |0 d4 x 0| [QBRST . For simplicity. (13. 13.7. as we have seen in (12. let us consider the self-interacting ﬁeld theory of a real scalar ﬁeld coupled to an external source described by the action S J [φ] = S[φ] + d4 x J(x)φ(x).4 Slavnov-Taylor identity The BRST invariance of the full theory leads to many relations between various scattering amplitudes of the theory.74) where S[φ] denotes the dynamical action of the self-interacting scalar ﬁeld.

78) is known as the classical ﬁeld (see (7. (13. In particular we note that 0|φ(x)|0 J δn W [J] δJ(x1 ) · · · δJ(xn ) J=0 .49)) and is a functional of J. Z[J] and W [J] are known as the generating functionals for the Green’s functions of the theory. the n-point Green’s functions would involve Feynman diagrams which consist of parts that are not connected and Z[J] (through (13. This is more fundamental since the general Green’s functions can be constructed in terms of these.76)) includes contributions of all diagrams (including the ones that are not connected) to a Green’s function.76) and (13.77) in the path integral formalism corresponds to a unique Feynman diagram in perturbation theory.76) J=0 deﬁnes the n-point time ordered Green’s function of the theory. the Green’s functions calculated from W [J] (the logarithm of Z[J]) through (13. δJ(x) Z[J] δJ(x) (13. The connected two point time ordered Green’s function which also corresponds to the Feynman propagator of the theory is similarly obtained from (13. 0|T φ(x1 ) · · · φ(xn ) |0 = δn Z[J] (−i)n Z[J] δJ(x1 ) · · · δJ(xn ) . 0|T φ(x1 ) · · · φ(xn ) |0 c = (−i)n−1 leads to the connected n-point time ordered Green’s functions of the theory. (13. On the other hand. let us note here that each term in the calculation of the Green’s function in (13. For example. To clarify this distinction further. In general.77) iGF (x − y) = 0|T φ(x)φ(y) |0 = (−i) δ2 W [J] δJ(x)δJ(y) J=0 c = (−i)2 δ2 Z[J] Z[J] δJ(x)δJ(y) J=0 . On the other hand.566 13 BRST invariance and its consequences As we have already mentioned in the last chapter.79) .77) involve only Feynman diagrams where all the parts of the diagram are connected. (13.77) = φc (x) = (−i) δZ[J] δW [J] = .

81) = −J(y). the more fundamental concept in perturbation theory is the diagram without the external propagators known as the vertex function. These are quite fundamental in the study of quantum ﬁeld theory and the generating functional which generates such vertex functions is constructed as follows.80) It can now be checked using (13. The diagrams that contribute to the n-point vertex function and which cannot be separated into two disconnected diagrams by cutting a single internal line (propagator) of the diagram lead to what are known as the 1PI (one particle irreducible) vertex functions or the proper vertex functions of the theory. However.82) . Let us Legendre transform W [J] as (this is like going from the Lagrangian to the Hamiltonian) Γ[φc ] = W [J] − d4 x J(x)φc (x).4 Slavnov-Taylor identity 567 The connected n-point time ordered Green’s functions consist of connected Feynman diagrams with external lines (propagators).13.78) that δΓ[φc ] δφc (x) = d4 y δW [J] δJ(y) δJ(y) − φc (y) − δ4 (x − y)J(y) δJ(y) δφc (x) δφc (x) (13. With these basic ideas from path integrals. 2 (13. The generating functional Γ[φc ] is known as the eﬀective action of the theory (including all quantum corrections) and generates the proper (1PI) vertex functions of the theory. (13.12) as well as source terms as follows Leﬀ = LTOT + J µa Aa + J a F a + i (η a ca − ca η a ) µ +K µa (Dµ c)a + K a g abc b c f cc . let us consider the eﬀective Lagrangian density (at the lowest order or tree level) which consists of LTOT for the Yang-Mills theory (13.

Denoting all the ﬁelds and the sources generically by A and J respectively. we obtain the change . In fact. can now be written as δW [J] . we have assumed the convention of left derivatives for the anticommuting ﬁelds.83) The vacuum expectation values of operators. for example. δη a 0| (Dµ c)a |0 0| = (Dµ c)a J = δW [J] . in the presence of sources. δη a 0|Aa |0 µ J = Aa µ J = A(c) a = µ 0|ca |0 J = ca J J = c(c)a = (−i) J δW [J] . but we have also added sources (K µa . δJ µa δW [J] . (7. δK a (13. The eﬀective Lagrangian density is no longer invariant under the BRST transformations (13.13) of the ﬁelds Aa .84) Here. K a ) for the composite variations under the BRST transformation (namely. for the variations in (13. we have not only introduced sources for all the ﬁeld variables in the theory. ca which are nonlinear in µ the ﬁeld variables). recalling that LTOT in (13. δK µa J g abc b c f c c |0 2 = g abc b c f cc 2 = δW [J] . η a ).568 13 BRST invariance and its consequences Here. 0|F a |0 J = F a J = F (c)a = δJ a δW [J] 0|ca |0 J = ca J = c(c)a = (−i) . we can write the generating functional for the theory as R d4 x Leﬀ 0|0 J = Z[J] = eiW [J] = N DA ei .49)) and in what follows we will ignore the superscript (c) for notational simplicity. The ﬁelds A(c) are known as the classical ﬁelds (see. (13.13) when the external sources are held ﬁxed.82) is BRST invariant and that the BRST transformations are nilpotent. It is worth noting here that the source K µa is of fermionic nature (in addition to the fermionic sources η a . The usefulness of this will become clear shortly.

85) g 2 We note that the generating functional is deﬁned by integrating over all possible ﬁeld conﬁgurations. − iη a (x) + iη a (x) a µa (x) a (x) δK δK δJ (x) (13. it should not change under any ﬁeld redeﬁnition).87) +iη a (x) The functional integration measure can be easily checked to be invariant under such a fermionic transformation and using (13. Therefore. since it does not depend on the ﬁeld variables.4 Slavnov-Taylor identity 569 in Leﬀ to be (remember that the parameter of the BRST transformation is anti-commuting) = J µa δAa + J a δF a + i (η a δca − δca η a ) µ = δLeﬀ g ω µa J (Dµ c)a + i − f abc η a cb cc + F a η a . if we redeﬁne the ﬁelds under the path integral as.88) d4 x J µa (x) This is the master equation from which we can derive all the identities relating the connected Green’s functions of the theory by taking functional derivatives with respect to sources. (13.87) can also be written as δW δW δW = 0.78)) d4 x Leﬀ δZ[J] = 0 = N = ω g DA i d4 x δLeﬀ ei d4 x J µa (x) δZ δZ − iη a (x) µa (x) δK δK a (x) δZ .85) to (see also (13.86) the generating functional should be invariant (namely. It is here that the . A → A + δBRST A.75) we note that (13. This immediately leads from (13.13. (13. δJ a (x) (13.

570 13 BRST invariance and its consequences usefulness of the sources for the composite BRST variations becomes evident.80) and the ﬁeld variables are really the classical ﬁelds and we are dropping the superscript (c) for simplicity). it is clear that δΓ δAa µ δΓ δF a δΓ δca δΓ δca δΓ a δKµ δΓ δK a = −J µa . however. δK a (13. K] = W [J. µ (13.89) where K stands generically for the sources for the composite BRST variations. From the deﬁnition of the generating functional for the proper vertices (see (13. = −J a . Most often. These can be obtained by passing from the generating functional for the connected Green’s functions W [J] to the generating functional for the proper vertices Γ[A] through the Legendre transformation involving the ﬁeld variables of the theory (see (13. a δKµ δW .90) Using these deﬁnitions. = iη a .88) in terms of the generating functional for the proper vertices as . K] − d4 x J µa Aa + J a F a + i (η a ca − ca η a ) . = = δW . we are interested in the identities satisﬁed by the proper (1PI) vertices of the theory. we see that we can rewrite the master equation (13.81)). = iη a . we have Γ[A.

for example. The BRST invariance. in this way.93) − c δF b (p)δAa (−p) δcc (−p)δK µa (p) δc (−p)δcb (p) µ A simple analysis of this relation shows that the mixed two point vertex function involving the ﬁelds F -Aµ is related to the two point function for the ghost ﬁelds and.92) and setting all Taking derivative of this with respect to ﬁeld variables to zero gives δ2 Γ δ2 Γ δ2 Γ = 0. This is essential in proving the renormalizability of gauge theories. (13.91) This is the master equation from which we can derive all the relations between various (1PI) proper vertices resulting from the BRST invariance of the theory by taking functional derivatives with respect to (classical) ﬁelds. the gauge dependence of the eﬀective potential in a gauge theory (with scalar ﬁelds) as well as the gauge independence of the physical poles of the propagator can be obtained systematically from the Nielsen identities which.4 Slavnov-Taylor identity 571 d4 x δΓ δΓ δΓ δΓ δΓ + a − F a (x) a = 0. consequently.) . For example. Thus. is very fundamental in the study of gauge theories as far as renormalizability and gauge independence of physical observables are concerned. (We would discuss renormalization of ﬁeld theories in a later chapter. let us note that we can write the master identity (13. the counter terms (quantum corrections) should satisfy such a relation.13.91) in the momentum space as δΓ δΓ δΓ δΓ + δAa (−k) δK µa (k) δca (−k) δK a (k) µ −F a (k) δΓ δca (k) = 0. a (x) δK µa (x) a (x) δAµ δc (x) δK δc (x) (13. follow from the BRST invariance of the theory. δ2 δF b (p)δcc (−p) d4 k (13. like the Slavnov-Taylor identities.

572 13. 2 µ where we have neglected total divergence terms and have identiﬁed Oµν ab = δab η µν 1 µ ν ∂ ∂ . − 1− (13.94) The propagators of the theory can be derived from the free part of the Lagrangian density which is quadratic in the ﬁeld variables 1 ∂µ Aa − ∂ν Aa (∂ µ Aνa − ∂ ν Aµa ) ν µ 4 1 2 ∂ µ Aa + ∂ µ ca ∂µ ca − µ 2ξ − 1− 1 µ ν a ∂ ∂ Aν − ca ca ξ (13. We note that since gauge ﬁxing breaks gauge invariance. (They are the Green’s functions of the free theory.96) take the forms .96) The propagators for the gauge and the ghost ﬁelds are the in−1ab verses Oµν and M −1ab respectively of the two point functions (up to multiplicative factors). These are the essential elements in carrying out any perturbative calculation in a quantum ﬁeld theory.2) 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ 2 + ∂ µ ca Dµ ca . it renders the theory nonsingular making it possible to deﬁne the propagator of the theory and derive the Feynman rules for the non-Abelian gauge theory. Let us recall that the total Lagrangian density after gauge ﬁxing (in the covariant gauge) has the form (13. ξ M ab = −δab .5 Feynman rules 13 BRST invariance and its consequences Understanding gauge ﬁxing for gauge theories in the path integral formalism is quite essential in developing the perturbation theory for gauge theories.95) LQ = − = = 1 a µν A η 2 µ 1 a µν ab b A O Aν + ca M ab cb . (13.) The inverses of the two point functions are easily calculated in the momentum space where the quadratic operators in (13.

Therefore. η µν p2 − 1 − 1 µ ν −1ac p p O νλ (p) = −δac δµ . p2 (13.97) is quite simple (it is like the propagator of a massless scalar ﬁeld) M −1ab (p) = δab .101) 1 β µ (αp2 + 1)δµ − α(1 − ) − p pλ = 0. let us derive this inverse systematically.97) The inverse of M ab (p) in (13. λ ξ (13. we can parameterize the most general form of the inverse as O−1ab (p) = δab α ηµν + β µν pµ pν .5 Feynman rules 573 M ab (p) = δab p2 . λ or. ηµν ). 1 µ ν p p ξ αηνλ + β pν pλ p2 = −δµ . λ αp2 δµ − α 1 − λ 1 1 µ p pλ + β pµ pλ − β 1 − pµ pλ = −δµ .100) Substituting the parameterization (13.99) where α. or. Oµν ab (p) = −δab η µν p2 − 1 − 1 µ ν p p . p2 (13.13.98) while the derivation of the inverse of Oµν ab (p) is a bit more involved. With this parameterization. λ ξ ξ . λ ξ ξ (13. Let us note that −1ab the Green’s function Oµν (p) is a symmetric second rank tensor (of rank 2) and with all the Lorentz structures available (pµ . β are arbitrary parameters to be determined (they are not necessarily constants and can be Lorentz invariant functions of the momentum). ξ (13. we note that the inverse is deﬁned to satisfy (repeated indices are summed) −1bc Oµν ab (p)Oνλ (p) = δac δµ .99) for the inverse and carrying out the multiplication explicitly we have η µν p2 − 1 − or.

104) where ( . see (12. p2 (13. . 2 p p (13.126)) h −1ab i − 2 J µa (−p).102) α=− α 1− or. it follows that 1 .79).99) as pµ pν δab ηµν + (ξ − 1) 2 .M −1ab (p)ηb (p) Z0 = N e i . ) represents the integral over p and we deduce from this that (see (13. (13. β = α(ξ − 1) = − so that we can write the inverse in (13.103) With these.Oµν (p)J νb (p) +i η a (−p). sometimes the gauge propagator is also denoted by ab Dµν (p)) δ2 Z0 (−i)2 Z0 δJ µa (−p)δJ νb (p) iGab (p) = µν J µa =η a =η a =0 −1ab −1ab = (−1) − iOµν (p) = iOµν (p) = − iδab pµ pν ηµν + (ξ − 1) 2 .574 13 BRST invariance and its consequences Setting the coeﬃcients of each distinct Lorentz structure to zero. we can write the generating functional for the free theory as (involving only the quadratic terms of the action as well as sources for the ﬁelds. p2 1 β − = 0.105) This deﬁnes the Feynman propagator (we do not explicitly write the subscript “F” and the iǫ in the denominator for simplicity) for the gauge ﬁeld which clearly depends on the gauge ﬁxing parameter ξ. 2 p p −1ab Oµν (p) = − (13. ξ ξ 1 (ξ − 1).

p2 (13.107) µ.108) = iGab (p) = We can now write the complete generating functional as Z[J] = exp i d4 x Lint 1 δ i δJ Z0 [J].5 Feynman rules 575 When ξ = 0.109) where we have used J generically for all the sources and the interaction Lagrangian density can now be identiﬁed with Lint = LTOT − LQ = gf abc ∂µ Aa Aµb Aνc − ν g2 abp cdp a b µc νd f f Aµ Aν A A 4 (13.106) which deﬁnes the ghost propagator (which is sometimes also denoted by Dab (p)).e.13. while for ξ = 1. (13. b pµ pν iδab ηµν + (ξ − 1) 2 . when we are in the Landau gauge this propagator is transverse. The propagators can be diagrammatically represented as = iGab (p) µν = − a p b (13.110) −gf abc Aa ∂ µ cc cb . Similarly we note that (once again we suppress the subscript “F” and the iǫ prescription which is assumed) iGab (p) = δ2 Z0 (−i)2 Z0 δη a (−p)δη b (p) J µa =η a =η a =0 = (−1) −iM −1ab (p) = iM −1ab (p) = iδab . µ . 2 p p iδab . the gauge is known as the Feynman gauge where the propagator has a much simpler form which is more suitable for perturbative calculations. a p ν. i. p2 (13..

p4 ) = i (13. V µνλρ abcd (p1 . p2 . p3 .576 13 BRST invariance and its consequences Let us now derive the interaction vertices for the theory which are more useful in the momentum space (where Feynman diagram calculations are primarily carried out). We note that we can write the interaction action in the momentum space as Sint = d4 x Lint d4 p1 d4 p2 d4 p3 δ4 (p1 + p2 + p3 ) = (2π)4 gf abc × (−ip1µ ) Aa (p1 ) Aµb (p2 ) Aνc (p3 ) ν − (2π)4 g2 abp cdp f f 4 d 4 p1 d 4 p2 d 4 p3 d 4 p4 × δ4 (p1 + p2 + p3 + p4 )Aa (p1 )Ab (p2 )Aµc (p3 )Aνd (p4 ) µ ν −(2π)4 gf abc d4 p1 d4 p2 d4 p3 δ4 (p1 + p2 + p3 ) (13.111) × (−ipµ ) Aa (p1 ) cc (p3 ) cb (p2 ) . p3 ) = i δ3 S δAa (p1 )δAb (p2 )δAc (p3 ) µ ν λ = i δ3 Sint δAa (p1 )δAb (p2 )δAc (p3 ) µ ν λ = (2π)4 gf abc δ4 (p1 + p2 + p3 ) × η µν (p1 − p2 )λ + η νλ (p2 − p3 )µ + η λµ (p3 − p1 )ν . p2 . µ 3 The tree level 3-point and 4-point interaction vertices of the theory follow from this to correspond to (the restriction | denotes setting all ﬁeld variables to zero) V µνλ abc (p1 .112) δ4 S δAa (p1 )δAb (p2 )δAc (p3 )δAd (p4 ) µ ν ρ λ = i δ4 Sint δAa (p1 )δAb (p2 )δAc (p3 )δAd (p4 ) µ ν ρ λ .

b = V µνλ abc (p1 .114) = −(2π)4 gf abc pµ δ4 (p1 + p2 + p3 ). p2 . p2 ν. p3 .112)-(13. d λ. p4 ). a ρ. c p3 p1 µ.115) .113) V µ abc (p1 . p2 . b c = V µ abc (p1 . 3 Therefore. (13. a p3 p1 µ. p3 ). a p2 b p2 ν. with (13. p3 ) = i δ3 S δAa (p1 )δcb (p2 )δcc (p3 ) µ = i δ3 Sint δAa (p1 )δcb (p2 )δcc (p3 ) µ (13. p2 . p2 . p3 ).13. p4 p3 p1 µ.5 Feynman rules 577 = −(2π)4 ig2 δ4 (p1 + p2 + p3 + p4 ) f abp f cdp η µλ η νρ − η µρ η νλ +f acpf dbp η µρ η νλ − η µν η λρ +f adp f bcp η µν η λρ − η µλ η νρ .114) we can represent all the interaction vertices of the theory as λ. c = V µνλρ abcd (p1 . (13.

we recall (see (12. such a gauge is known as the light-cone gauge. this would. When n2 = 0. LTOT = − Fµν F µνa − 4 2ξ (13. simplify the perturbative calculations enormously primarily because the number of diagrams to evaluate will be much smaller (namely. from two distinct points of view. 13. Let us ﬁrst discuss this from the diagrammatic point of view. If possible.578 13 BRST invariance and its consequences All the momenta in the above diagrams are assumed to be incoming and the arrow in the ghost diagram shows the direction of ﬂow of the ghost number. the diagrams with ghosts will be absent). A class of gauge conditions of the form n · Aa (x) = f a (x).6 Ghost free gauges In quantizing a non-Abelian gauge theory. or (13.114)). The dot over the ghost line denotes where the derivative acts in the interaction term in the Lagrangian density (namely. in what sense these gauges become ghost free in a non-Abelian gauge theory. n2 = 0 n2 = 0.116). we ﬁnd that a gauge ﬁxing term necessarily modiﬁes the theory and thereby requires the addition of a ghost Lagrangian density which as we have argued is necessary to balance the modiﬁcation induced by the gauge ﬁxing term.160)) that the total Lagrangian density for a non-Abelian gauge theory takes the form 1 1 a (n · Aa )2 − ca n · Dca . In a general axial-like gauge (13. see for example (13. of course.153) and (12.116) where nµ is an arbitrary vector does indeed achieve this. We will now discuss. These represent all the Feynman rules for the Yang-Mills theory and perturbative calculations can now be carried out using these Feynman rules.117) . Such a class of gauge choices is conventionally known as ghost free gauges. for n2 = 1 (normalized) it is called the temporal gauge while for n2 = −1 it is known as the axial gauge. the cc ﬁeld. It is interesting to ask if there exist gauge conditions in a non-Abelian theory where the ghost degrees of freedom may not be important much like in the Abelian theory.

µν (13.119) Similarly.1: Examples of diagrams with open and closed ghost lines. n·p iGab = (13. Therefore.120) and the antighost-ghost-gauge vertex involves a multiplicative factor of nµ . (n · p) nµ Gab = −ξδab µν nµ nν Gab = −ξδab . the integrand of any diagram involving ghosts.6 Ghost free gauges 579 With such a choice of gauge the gauge propagator can be calculated to have the form iGab µν = iδab n µ pν + n ν pµ n 2 pµ pν 1 − − ηµν + p2 (n · p) (n · p)2 pµ pν −ξ .116) in this theory has the form δab . . (n · p)2 (13. either open lines or closed loops.13. the ghost propagator in the axial-like gauge (13. will have the structure (we are not writing the possible multiplicative Lorentz factors involving nµ which do not eﬀect the integral and we are evaluating the integral in D dimensions) Figure 13.118) so that we have (independent of whether n2 = 0 or n2 = 0) pν .

(13. (n · k)m+1 δ(1 − x0 − x1 − · · · − xm ) m+1 = 0 i=0 dxi (13. This shows that all the diagrams involving ghost interactions can be regularized to zero so that the ghost degrees of freedom are irrelevant with such a choice of gauge. (n · (k + p1 + · · · + pm )) 1 m dxi 0 i=0 1 m dD k dD k n · (k + x1 p1 + · · · + xm pm ) δ(1 − x0 − x1 − · · · − xm ) . .122) and the vanishing of this integral (in the last step) is easily seen using dimensional regularization (that we will discuss in chapter 15. we can rewrite this also as 1 m dxi 0 i=0 δ(1 − x0 − x1 − · · · − xm ) (n2 )(m+1)/2 dD k = 0. (k2 )(m+1)/2 (13. Using the fact that the integral is Lorentz invariant. namely.580 13 BRST invariance and its consequences dD k = 1 (n · k) (n · (k + p1 )) . in (15.121) Here we have used the Feynman parameterization to combine the denominators (to be discussed in the next chapter) and have shifted the variable of integration in the last step to bring it to the simpler form.123) Here G[f a ] is an arbitrary functional of f a which we normally choose to correspond to the ’t Hooft weight factor .116) has the form Z = N = N DAa Df a det n · Dab δ n · Aa − f a G[f a ]eiS µ DAa Df a det n · ∂δab + gf abc n · Ac µ ×δ n · Aa − f a G[f a ]eiS .10)). . An alternative way to see this is to note that the generating functional in the axial-like gauge (13.

let us choose G[f a ] = det n · ∂δab + gf abc f c −1 − i R d4 x f a f a 2ξ e . the covariant gauge.7 References 581 G[f a ] = e i − 2ξ R d4 x f a f a . Theoretical and Mathematical Physics 10. 99 (1972). (13. (13.125) In this case. C. A. . therefore. A. 2. correspond to ghost free gauges very much like in the Abelian gauge theories. 436 (1971). 13. Physics Letters 52B. since the generating functional does not depend on the functional form of G [f a ] (up to irrelevant multiplicative constants). C. Taylor. A.13. It would not work in say.126) This shows that in this gauge the ghosts decouple completely and can be absorbed into the normalization factor. (13.116) in a non-Abelian gauge theory. Rouet and R. 344 (1974). 3. Becchi. J.124) However. Nuclear Physics B33. Slavnov. we can write the generating functional as Z = N DAa Df a det n · ∂δab + gf abc n · Ac δ n · Aa − f a µ × det n · ∂δab + gf abc f c −1 − i R d4 x f a f a iS 2ξ e e = N = N DAa Df a δ (n · Aa − f a) e µ DAa eiS− 2ξ µ i i − 2ξ R d4 x f a f a iS e R d4 x (n·Aa )2 . Stora.7 References 1. The axial-like gauge choices (13. Note that such a decoupling works only because nµ (in n · Aa inside the determinant) is a multiplicative operator.

Reviews of Modern Physics 59. C. Physical Review D13. A. 1067 (1987). Stora. Progress of Theoretical Physics 60. 66 (1979). G. A. Finite Temperature Field Theory. Kugo and I. T. . Leibbrandt. Becchi. 5. Progress of Theoretical Physics Supplement No. World Scientiﬁc. Ojima. Ojima. Kugo and I. 127 (1975). Rouet and R. Singapore (1997). 1869 (1978). 8. Frenkel.582 13 BRST invariance and its consequences 4. Communications in Mathematical Physics 42. 7. 9. 6. J. T. 2325 (1979). Das.

As we have mentioned earlier gauge ﬁelds can be thought of as the carriers of physical forces. a massless gauge ﬁeld is completely consistent with the observed fact that electromagnetic forces are long ranged. L = − Fµν F µν + 4 2 583 (14.34) respectively.Chapter 14 Higgs phenomenon and the standard model In chapters 9. Therefore.14) or (12. 2 M 2 a µa Aµ A . 2 or (14. 14. One of the striking features of these theories is that the gauge ﬁelds are massless simply because the action for a mass term for the gauge ﬁeld in the Lagrangian density M2 Aµ Aµ . Therefore.1) is not invariant under the gauge transformation (9. However. in this chapter we would discuss this important question of massive gauge ﬁelds concluding with the standard model of the electroweak interactions. 12 and 13 we have studied gauge theories (both Abelian and non-Abelian) in great detail.1 St¨ckelberg formalism u Let us consider the Lagrangian density for a charge neutral spin 1 ﬁeld Aµ given by M2 1 Aµ Aµ .2) . we also know of physical forces in nature (such as the weak force) that are short ranged and this would seem to suggest (intuitively) that the gauge ﬁelds associated with such forces may be massive.

we also note here that the theory (14. for example.2) has the form ∂L ∂L − = 0. we obtain ∂ · A = 0.5) eliminates one degree of freedom leaving us with three ﬁeld degrees of freedom which is the correct number of dynamical ﬁeld degrees of freedom for a massive spin 1 ﬁeld.5) and substituting this back into the Euler-Lagrange equation in (14.2) describes a massive photon which can be seen as follows. section 13. As has been mentioned earlier.3) The Lagrangian density (14. −∂µ F µν − M 2 Aν = 0. (14. or.5) deﬁnes the Proca equation (and (14. The propagator for this theory can be worked out (see. or.5) in a simple manner and in momentum space has the form . The Euler-Lagrange equation following from (14.2) does not have a gauge invariance (because of the mass term) unlike the Maxwell theory. (14.2) denotes the Proca Lagrangian density) and it is clear that this system of equations deﬁnes a massive spin 1 ﬁeld theory (a massive photon). (14. ∂µ F µν + M 2 Aν = 0.6) Equation (14. ∂∂µ Aν ∂Aν ∂µ or.4) yields ∂µ (∂ µ Aν − ∂ ν Aµ ) + M 2 Aν = 0. + M 2 Aν = 0.584 14 Higgs phenomenon and the standard model where the ﬁeld strength tensor is deﬁned as in the Maxwell theory Fµν = ∂µ Aν − ∂ν Aµ .4) with ∂ν and using the anti-symmetry of the ﬁeld strength tensor. Note that the ﬁeld variable Aµ has four ﬁeld degrees of freedom while the constraint (14.4) Contracting (14.6) together with the constraint (14. (14.

The ﬁeld strength tensor is. we see that the propagator in this theory does not fall oﬀ fast enough for large values of the momenta unlike in all other theories that we have studied so far. the St¨ckelberg theory u (14. e M θ(x). This leads to diﬃculties in establishing renormalizability of the (interacting) Proca theory for massive photons. but we have kept it for consistency with earlier discussion. of course. 2 −M M2 (14. in addition to the spin 1 ﬁeld. In contrast to the Proca theory (14. also contains a dynamical charge neutral spin zero ﬁeld (real scalar ﬁeld) which mixes with Aµ .9) where e denotes the coupling constant of QED (this can be set to unity.8) = − Fµν F µν + 4 2 2 which. St¨ckelberg considered the folu lowing generalized Lagrangian density 1 M2 1 Aµ + ∂µ χ L = − Fµν F µν + 4 2 M Aµ + 1 µ ∂ χ M pµ pν i ηµν − .83) as well as for discussions in connection with the Higgs phenomenon in the next section).9). it is suﬃcient to note that under the transformation (14.8) was constructed to be invariant under a Maxwell-like gauge transformation Aµ (x) → Aµ (x) + 1 ∂µ θ(x).¨ 14. (14. χ(x) → χ(x) − e (14. M (14. As a result of these diﬃculties. invariant under the gauge transformation (14. To see the invariance of the full theory.9) Aµ + M 1 1 1 ∂µ χ − ∂µ χ → Aµ + ∂µ θ + ∂µ θ M e M e 1 = Aµ + ∂µ χ.7) M2 1 1 Aµ Aµ + ∂µ χ∂ µ χ + M Aµ ∂µ χ.10) .1 Stuckelberg formalism 585 iGµν (p) = − p2 Thus. The limit M → 0 of the Proca theory is clearly quite subtle.2). say in (9.

8) takes the form M2 1 Aµ Aµ . the there is no diﬃculty in establishing renormalizability in such a gauge. we can choose a gauge and if we choose the unitary gauge χ(x) = 0. in this unitary gauge. This is typical of the behavior of theories in a unitary gauge. but would lead to propagators that are well behaved at high momentum. L = − Fµν F µν + 4 2 (14. section 12. As a result. However. (14.7)) i pµ pν ηµν − .586 14 Higgs phenomenon and the standard model Because of the gauge invariance of the theory.13) We recognize this to be the Proca theory (14. if we choose a covariant gauge condition which leads to a gauge ﬁxing Lagrangian density of the form (see. the gauge propu agator (see also (14. M θ(x) = (14. 2ξ − (14.3 and we note that this is known as the ’t Hooft gauge which we will discuss in the next section) 1 (∂µ Aµ − ξM χ)2 .15) this would induce a ghost action. Furthermore.12) then the Lagrangian density (14.11) which can be achieved by the choice of the gauge transformation parameter e χ(x). . the massless limit M → 0 is now straightforward.2) and this makes it clear that the Proca theory can be thought of as the gauge ﬁxed St¨ckelberg theory. On the other hand. for example. 2 −M M2 iGµν (p) = − p2 (14.14) has the unpleasant features described earlier.

To see that the complete Lagrangian density is also invariant. ig (14. (14.1 Stuckelberg formalism 587 All of this analysis can be carried over to non-Abelian gauge theories as well as to the Einstein theory of gravitation. SU (n) (as we have discussed earlier in section 12. This Lagrangian density can be easily checked to be invariant under the gauge transformations (see.16) ig ig where the trace is over the fundamental representation of the group.1). ig (14.18). the Lagrangian density for the gauge ﬁeld is invariant under the transformation (14. Let us consider the Lagrangian density L = Tr − 1 Fµν F µν 2 1 1 +M 2 Aµ + (∂µ V )V −1 Aµ + (∂ µ V )V −1 . Aν ] .17) where χ denotes a matrix valued scalar ﬁeld. (12. 1 (∂µ U )U −1 .41)) Aµ → U Aµ U −1 − V → U V. We will now brieﬂy describe the St¨ckelberg formalism for non-Abelian gauge u theories.¨ 14. for example. = e M χ.38) and (12. it is suﬃcient to note that . say. Here the ﬁeld strength tensor and the operator V are deﬁned as Fµν V = ∂µ Aν − ∂ν Aµ + ig [Aµ .18) As we have already seen in chapter 12.

(14.19) Using the cyclicity under trace it is now straightforward to see that the full theory is invariant under the gauge transformation in (14.588 14 Higgs phenomenon and the standard model Aµ + 1 (∂µ V )V −1 ig 1 1 (∂µ U )U −1 + (∂µ U V )(U V )−1 ig ig 1 (∂µ U )U −1 ig → U Aµ U −1 − = U Aµ U −1 − + = U Aµ + 1 (∂µ U )U −1 + U (∂µ V )V −1 U −1 ig 1 (∂µ V )V −1 U −1 .23) . ig (14. 2 (14.20) which can be achieved by the choice of the gauge transformation matrix U (x) = V −1 (x). the propagator has the form (recall that the quadratic part of the Lagrangian density has the similar form both in the Abelian as well as in the non-Abelian theories) pµ pν iδab ηµν − .22) This can be thought of as a generalization of the Proca theory (14.2) to the non-Abelian case and describes a massive spin 1 ﬁeld belonging to SU (n). In this gauge.16) becomes 1 L = − Tr Fµν F µν + M 2 Tr Aµ Aµ . If we choose the unitary gauge condition V = ½.18). then the Lagrangian density (14. as before. 2 − M2 p M2 iGab (p) = − µν (14.21) (14.

24) can be checked easily (see (7. (14. e δφ = −iǫ(x)φ(x).5 we studied the phenomena of spontaneous break down of a global symmetry in the self-interacting complex scalar ﬁeld theory where the mass term for the scalar ﬁeld had the “wrong” sign. but now interacting with an Abelian gauge ﬁeld as well (scalar QED).84) and (7. 14. see also section 7.2 Higgs phenomenon 589 with the unpleasant features discussed earlier that are characteristics of the unitary gauge choice.6) λ 1 2 L = − Fµν F µν + (Dµ φ)† (D µ φ) + m2 φ† φ − φ† φ .83) Dµ φ = ∂µ φ + ieAµ φ.26) .5). we saw that one of the real components of the scalar ﬁeld becomes massless while the other picks up a mass (with the right sign).15)). Let us next ask what happens if there is a spontaneous breakdown of a local symmetry. δAµ = 1 ∂µ ǫ(x).24) 4 4 where the theory has a “wrong” sign for the mass term and the covariant derivative is deﬁned as in (7. For simplicity we consider again the self-interacting theory of a complex scalar ﬁeld (as in section 7. (14. On the other hand. The Lagrangian density for such a theory is given by (with λ > 0. (14. the propagator will have the correct asymptotic behavior for large values of the momenta. In that case.2 Higgs phenomenon In section 7. if we choose a covariant gauge ﬁxing condition (a generalization of (14. This is known as the Nambu-Goldstone phenomena and the massless scalar ﬁeld is known as the Nambu-Goldstone boson.14.85)) to be invariant under the inﬁnitesimal local transformations δφ† = iǫ(x)φ† (x).25) The Lagrangian density (14.

28) (14. (14.77)) .5.) Rather the true vacuum of the theory satisﬁes (see (7. let us rewrite the Lagrangian density (14.42)) 1 φ = √ (σ + iχ). this actually corresponds to the local maximum of the potential. χ ﬁelds 1 λ † L = − Fµν F µν + (Dµ φ)† (Dµ φ) + m2 φ† φ − φφ 4 4 ← → 1 = − Fµν F µν + ∂µ φ† ∂ µ φ − ieAµ φ† ∂µ φ 4 λ † 2 φφ + e2 Aµ Aµ φ† φ + m2 φ† φ − 4 2 1 1 1 m2 2 = − Fµν F µν + ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + σ + χ2 4 2 2 2 ← → e2 λ 2 − eAµ χ ∂µ σ + Aµ Aµ σ 2 + χ2 − σ 2 + χ2 . there is actually an inﬁnity of minima of the potential.24) in terms of the σ. this is just one of them) 2m σc = σ = √ .29) For further analysis. since the mass term for the scalar ﬁeld has the “wrong” sign. (14. (14.5.30) 2 16 Let us now rewrite the theory around the true vacuum by shifting the ﬁeld variables as (see (7. On the other hand. the normal vacuum for which φ = φ† = 0. λ where (see also (7. as we have discussed in section 7. 2 χc = χ = 0.590 14 Higgs phenomenon and the standard model The gauge boson in this theory is massless because the gauge symmetry (14. (As we have seen in section 7.64).26) does not allow a mass term for the gauge ﬁeld.27) does not correspond to the true vacuum.

30) becomes ← → 1 L = − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) − eAµ χ ∂µ (σ + v) 4 + + e2 1 1 Aµ Aµ ((σ + v)2 + χ2 ) + ∂µ σ∂ µ σ + ∂µ χ∂ µ χ 2 2 2 (14. λ In this case.33) where we have used the form of v in (14.31) where we have identiﬁed the vacuum expectation value (vev) of the ﬁeld with 2m σ =v= √ . 16 (14.32) to simplify some of the terms. but it now mixes with the gauge ﬁeld Aµ so that we have . (14.14.34) 2 2 We note here that the χ ﬁeld appears to be massless in this theory as in (7. the Lagrangian density (14. λ χ → χ + χ = χ.33) which are quadratic in the ﬁeld variables e2 v 2 1 Aµ Aµ LQ = − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) + 4 2 1 1 + ∂µ χ∂ µ χ + evAµ ∂µ χ + ∂µ σ∂ µ σ − m2 σ 2 . (14. Let us look at only the terms in (14.79).32) m2 λ ((σ + v)2 + χ2 ) − ((σ + v)2 + χ2 )2 2 16 1 1 = − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) + ∂µ σ∂ µ σ 4 2 1 m2 v 2 e2 v 2 + ∂µ χ∂ µ χ − m2 σ 2 + + Aµ Aµ + evAµ ∂µ χ 2 4 2 ← → e2 − eAµ χ ∂µ σ + Aµ Aµ σ 2 + χ2 + 2vσ 2 λ 4 − (σ + 2σ 2 χ2 + χ4 + 4vσ 3 + 4vσχ2 ).2 Higgs phenomenon 591 2m σ → σ + σ = σ + √ = σ + v.

8).592 14 Higgs phenomenon and the standard model to diagonalize the theory in order to determine the true spectrum of the theory.79).35) the quadratic Lagrangian density (14.34) takes the form e2 v 2 1 Bµ B µ LQ = − (∂µ Bν − ∂ν Bµ ) (∂ µ B ν − ∂ ν B µ ) + 4 2 1 + ∂µ σ∂ µ σ − m2 σ 2 . 1 1 ∂µ χ Aµ + ∂ µ χ ev ev 2 1 = Aµ Aµ + Aµ ∂µ χ + 2 2 ∂µ χ∂ µ χ.) To achieve this. ev Bµ = Aµ + then we have (14. As in (7.34). (14. (The Lagrangian density in (14. ev e v Aµ + Bµ B µ = (14. σ (14.38) and a massive gauge ﬁeld Bµ with mass (this is the analog of the unitary gauge in (14. let us deﬁne 1 ∂µ χ. ignoring the σ ﬁeld should be compared with the St¨ckelberg Lagrangian density u (14.39) .11)) 4e2 m2 .36) Thus in terms of this new ﬁeld variable (14. it describes a massive scalar ﬁeld σ with mass m2 = 2m2 .35) ∂µ Bν − ∂ν Bµ = ∂µ Aν + 1 1 ∂ν χ − ∂ν Aµ + ∂µ χ ev ev = ∂µ Aν − ∂ν Aµ . λ m2 = M 2 = e2 v 2 = B (14.37) that the quadratic part of the shifted Lagrangian density is completely diagonalized.37) 2 We see from (14.

of course. the Goldstone mode of the complex scalar ﬁeld. Thus some states must have negative norm in such a description.14. When we are dealing with a gauge theory.2 Higgs phenomenon 593 The other real component χ of the spin 0 boson . The answer to this question lies in the fact that if a gauge theory is written in a manifestly Lorentz invariant way. why is there no massless particle even though we have a spontaneous breakdown of symmetry. we can see this happening in the following way. The scalar ﬁeld responsible for this phenomenon is conventionally known as the Higgs ﬁeld. we have a mechanism for giving masses to gauge bosons without violating renormalizability and unitarity of the theory. Guralnik. it is the incompatibility of these two conditions that avoids the Goldstone theorem. It is. the massless mode has combined with the massless gauge ﬁeld to give it an additional degree of freedom. This is consistent with the unitarity of the theory (from the point of view of ﬁeld degrees of freedom). this second aspect that is important in developing a gauge theory of the weak interactions.8) the metric of the Hilbert space becomes indeﬁnite. namely. That is there can arise states in the Hilbert space with negative norm. A crucial assumption in the Goldstone theorem is that the theory should be manifestly Lorentz invariant as well as the metric of the Hilbert space should be positive semi-deﬁnite. This is in a way a double beneﬁt in the sense that we do not have to worry about the absence of massless bosons in nature and furthermore. Hagen and Kibble. consist only of transverse polarization. In fact. as we have seen (recall Gupta-Bleuler quantization in 9. . Intuitively. namely. This phenomenon of the gauge ﬁeld acquiring a mass by absorbing a Goldstone particle is known as the Higgs phenomenon and was investigated independently by Higgs. Thus the other two states must have cancelling eﬀect with each other as we have seen earlier. The ﬁeld Aµ is a four vector and hence must have four independent polarization states. what has happened is that the original gauge ﬁeld which was massless had only two physical degrees of freedom just like the photon. (We can think of the physical degrees of freedom as the two transverse components.) However.the NambuGoldstone boson . Hence the gauge ﬁeld has become massive with three degrees of freedom without violating unitarity. however. The physical states. At this point we may ask why the Goldstone theorem fails in this case.has completely disappeared from the spectrum of the theory. Brout and Englert.

30) is invariant under the inﬁnitesimal gauge transformations (see (7. eliminate the mixing term between the Aµ and the χ ﬁelds). To introduce the gauge choice. (14.40) nonetheless is invariant the shifted local transformations (recall that it is the vacuum of the theory that breaks the symmetry) δAµ (x) = δσ(x) = ǫ(x)χ(x).33) which reﬂects a spontaneous breakdown of the local symmetry (14.42) .41) e where M is deﬁned in (14.39). δχ(x) = −ǫ(x)σ(x).84)) δσ(x) = ǫ(x)χ(x).41) in the theory allows us to choose a gauge ﬁxing Lagrangian density which we would prefer to be manifestly Lorentz invariant and to simultaneously diagonalize the quadratic Lagrangian density (14. e 1 ∂µ ǫ(x). This is achieved by the ’t Hooft gauge ﬁxing which is described by the gauge ﬁxing Lagrangian density LGF = − = − 1 1 µ (∂ Aµ − ξevχ)2 = − (∂ µ Aµ − ξM χ)2 .34) (namely. First we note that the unshifted theory (14. 2ξ 2ξ ξM 2 2 1 µ (∂ Aµ )2 + M (∂ µ Aµ )χ − χ . 1 ∂µ ǫ(x). it is not the right gauge to study renormalizability of the theory. that is more useful for this purpose let us proceed as follows. δχ(x) = −ǫ(x)(v + σ(x)) = −ǫ(x) δAµ (x) = M + σ(x) . The presence of the local symmetry (14. (14. as pointed out in the last section. commonly known as the ’t Hooft gauge or the Rξ gauge. 2ξ 2 (14.40) e The shifted theory (14.594 14 Higgs phenomenon and the standard model Although the unitary gauge is useful for the counting of true dynamical degrees of freedom in the theory.

34) together with the cross terms in (14. The complete Lagrangian density in this gauge has the form LTOT = L + LGF + Lghost . The gauge ﬁxing Lagrangian density (14.14. disappear from the action.42) combine into a total divergence and.46) and can be easily checked to be invariant under the BRST transformations . thereby. the ’t Hooft gauge choice in (14.170)) LGF = ξ 2 F − F (∂ · A − ξM χ).44)) leads to an interacting ghost Lagrangian density. (14. (14.15).2 Higgs phenomenon 595 which can be compared with (14.43) and we note that in this gauge the propagators fall oﬀ for large values of the momenta unlike in the unitary gauge. 2 −M p − ξM 2 (14.164).45) We see that unlike in the conventional covariant gauge ﬁxing (12.42) (or (14. − ξM 2 pµ pν i ηµν − (1 − ξ) 2 . As a result. 2 (14. It is clear that the mixing terms in (14.44) and the choice of this gauge ﬁxing leads to a ghost Lagrangian density of the form (scaling out a factor of 1 ) e Lghost = ∂ µ c∂µ c − ξM 2 cc − ξM eσcc. the propagators for the Aµ and the χ ﬁelds in this gauge are given respectively by iGµν (p) = − iGχ (p) = p2 p2 i .42) can also be written with an auxiliary ﬁeld as (see (12.

M + σ c. δAµ (x) δσ(x) δPσ (x) δχ(x) δPχ (x) δc(x) (14.3 The standard model The standard model or the gauge theory of electroweak interactions was built on the results of years of earlier work on weak interactions. e δχ = −ω δc = 0.) are described quite well by a speciﬁc form of the Fermi theory which involves four fermion interactions in the form of a current-current interaction Hamiltonian density GF † HI = √ Jµ J µ . This is the starting point for establishing the renormalizability of the gauge theory with a spontaneously broken symmetry (Higgs theory). 14. e δσ = ωχc. for example. δc = −ωF.596 14 Higgs phenomenon and the standard model δAµ = 1 ω∂µ c.48) where Pσ and Pχ denote respectively the sources for the composite BRST variations of the ﬁelds σ and χ in (14.4 (see.49) . (13. the β decay of the neutron etc. At low energies. Therefore. (14.91)) we can now derive the master equation for the 1PI generating functional following from this invariance which takes the form δΓ δΓ δΓ δΓ δΓ δΓ ∂µ c(x)+ + −F (x) = 0. let us review some of these essential earlier results before constructing the standard model (known as the Weinberg-SalamGlashow theory). 2 (14.47) Following the discussions in section 13. weak interactions (for example. δF = 0.47).

quite weak and were also known to be extremely short ranged. 2 Q = I3L + (14.50) with ψn . As a result. proton. Experimentally .52) where Q denotes the electric charge of the particle.3 The standard model 597 where the current is a sum of the hadronic and leptonic currents of the form (had) (lept) Jµ = Jµ + Jµ = ψ p γµ (1 − γ5 )ψn + ψ ν γµ (1 − γ5 )ψe + · · · .50) that only the left-handed components of the fermion ﬁelds participate in the weak interactions because of which the weak isospin group can be identiﬁed with the group SUL (2) (meaning that the symmetry transformation changes only the left-handed components of the ﬁelds (see (3.17 × 10−5 (GeV)−2 .50) have a weak isospin structure and are also known as the weak isospin currents. It was known that all the weakly interacting particles (leptons and hadrons) can be described by multiplets belonging to the weak isospin group of SU (2). I3L the quantum number associated with the 3-component of the weak isospin generator and Y its weak hypercharge quantum number. The coupling constant (interaction strength) GF is known as the Fermi coupling and is determined from low energy experiments to have the value GF = 1.140) for a deﬁnition of right and left-handed particles/ﬁelds). ψp . (The V − A structure is connected with the fact that parity is violated maximally in weak interactions. electron and neutrino respectively. the currents in (14. (14. therefore.139) and (3.14.) In addition to the weak isospin quantum number. it is clear from the V − A (vector minus axial vector) structure of the currents in (14. it was also known that we can assign an additive (U (1)) quantum number known as the weak hypercharge to all the weakly interacting particles such that the charge of any particle can be written as Y . ψν denoting the fermion ﬁelds associated with neutron. (14. ψe .51) The weak interactions are. Furthermore.

(14.) Therefore.598 14 Higgs phenomenon and the standard model it was known that weak interactions violate both weak isospin and weak hypercharge quantum numbers. As a result we may try to formulate the weak interactions as a gauge theory such that it reduces to the V − A theory in the low energy limit. it is logical to associate the Higgs mechanism to the spontaneous breakdown of these symmetries. a gauge theoretic description would naturally involve massive gauge bosons. but in such a way that the electric charge is conserved in any weak process. we have already seen that gauge theories can describe physical forces and do not have any problem of renormalizability (we will discuss renormalizability and renormalizability of gauge theories in a later chapter). On the other hand. . the V − A theory cannot be the fundamental theory underlying weak interactions (forces). (14. As we have discussed weak interactions violate both the weak isospin and the weak hypercharge symmetries and. it is clear that we should look for a gauge theory of weak interactions based on the local symmetry group SUL (2) × UY (1). Namely. Furthermore. Although the V − A theory with four fermion interactions works quite well at low energy. We have seen in the last section that gauge bosons can become massive through the Higgs mechanism which would involve the spontaneous breakdown of some local symmetry. we would like the gauge theory to describe the symmetry behavior (“SSB” denotes spontaneous symmetry breakdown) SSB SUL (2) × UY (1) −→ UEM (1). it is not renormalizable. since weak interactions are short ranged.51) is dimensional with inverse dimensions of mass (energy). As a result. let us construct the standard model in several steps.53) where SUL (2) and UY (1) denote respectively the weak isospin and the weak hypercharge symmetry groups and we would like this symmetry group to spontaneously breakdown (at the weak interaction scale) in a way such that only the electromagnetic symmetry survives as the true symmetry of the low energy theory.54) With these basics. therefore. (The nonrenormalizable character is already manifest in the fact that the coupling constant of the theory (14.

eR = R. The quarks as well as the other families can be introduced in a straightforward manner. . although the leptons are believed to be fundamental particles. its ﬁeld can be decomposed into a left-handed part and a right-handed part. Thus. In contrast. However. the left-handed doublets have Y = −1 while for the righthanded singlets carry a hypercharge of Y = −2. The matter ﬁelds associated with the electron family can be grouped as multiplets of SUL (2) as νe e L L= .55) where the left.7) 1 ψL = (½ − γ5 )ψ.50) where the hadronic current was given in terms of neutron and proton ﬁelds) the matter ﬁelds in the standard model consist of quarks and leptons and they come in three families. The quark families can also be described by such structures.14. (14. The other lepton families (µ and the τ families) can be introduced through an obvious generalization. the neutrino is experimentally known to be essentially left-handed (see section 3.3 The standard model 599 14. namely. therefore. unlike the V − A theory (see (14. Let us ﬁrst discuss the ﬁeld content of the stan- dard model. we now know that the constituents of hadronic matter are the quarks. (14. 2 1 ψR = (½ + γ5 )ψ. The standard model is expected to describe the weak and electromagnetic interactions of all fundamental particles that participate in the weak processes. The left-handed multiplets of the standard model are doublets of SUL (2) while the right-handed particles correspond to singlets. These include both hadrons and leptons as we have already mentioned.56) 2 ψ = ψL + ψR . the electron and its neutrino. there will be right-handed singlet components for every quark ﬁeld (as opposed to the absence of νR in the leptonic sector). However.3.and the right-handed components of a spinor ﬁeld are deﬁned by (see also section 3. We note that the electron is massive and. for simplicity we will consider the matter to consist of only one family of leptons.1 Field content. The only diﬀerence is that since all the quark ﬁelds are massive. Each multiplet of SUL (2) has a unique value of the weak hypercharge quantum number (corresponding to UY (1)) and for the leptonic matter ﬁelds.6).

52).57) Yµ . in fact.54)) is given by a doublet of charged ﬁelds. a Wµ . (14.) The gauge ﬁelds which are expected to correspond to the carriers of forces are determined from the local symmetry structure of the theory. Let us now write down the quantum numbers for the matter and the Higgs ﬁeld and verify explicitly that (14. (φ+ )† = φ− . the minimal multiplet of Higgs ﬁelds that can achieve the spontaneous breakdown of the symmetry (as in (14. the local symmetry of the electroweak group must be spontaneously broken through the Higgs mechanism and this requires that the theory should also contain scalar Higgs ﬁelds which would be described by multiplets of SUL (2). As we will see shortly.58) 0 Since the Higgs multiplet is charged. 3. it carries a nontrivial weak hypercharge quantum number given by Y = 1 which is.600 14 Higgs phenomenon and the standard model (The hypercharge quantum numbers for the quarks are also diﬀerent from those for leptons in order to be compatible with (14. so that the theory should contain four gauge ﬁelds in total. minimal Higgs ﬁeld : φ = φ+ φ0 .53) to describe the local (gauge) symmetry of the theory. 2. (14. a = 1. (φ0 )† = φ . . then the corresponding gauge ﬁelds are determined to be SUL (2) : 3 gauge ﬁelds. Furthermore. the opposite of that for the left-handed lepton doublet.52) holds. UY (1) : 1 gauge ﬁeld. If we assume (14. if we would like the gauge bosons to be massive.

let us note that the gauge invariant Lagrangian density for the gauge ﬁelds follows from our earlier discussions (see (9.2 Lagrangian density. 2. (The Abelian gauge ﬁelds are not self-interacting which is why g′ does not appear in the Lagrangian density for the gauge ﬁelds.14.16) and (12. 3.45)) to correspond to 1 1 a LG = − Fµν F µν a − Fµν F µν . 4 4 a = 1.60) where the ﬁeld strength tensors for the Abelian as well as the nonAbelian gauge ﬁelds are deﬁned as (see (9. It is now a straightforward matter to write down the complete Lagrangian density for the theory which has the local gauge invariance SUL (2) × UY (1) and we will do this in several steps. but it will be present in the matter Lagrangian density.3 The standard model 601 Fields νe e eR φ+ φ0 L I3L 1 2 Y 2 1 −2 1 2 1 −2 I3L + − − 1 2 1 2 Y 2 Q 0 −1 −1 1 0 (14. First.3. (14.) The Lagrangian density (14.44)) Fµν a Fµν = ∂µ Yν − ∂ν Yµ . a a b c = ∂µ Wν − ∂ν Wµ − gǫabc Wµ Wν . Note that ǫabc denotes the structure constant of SUL (2) as we know from the study of angular momentum. (14.59) =0 = −1 −1 2 0 1 2 −1 2 −1 1 2 1 2 0 − 1 = −1 1 2 =1 1 −2 + 1 = 0 2 1 2 + −1 2 This shows that all the quantum numbers assigned to various ﬁelds are consistent with the relation (14.61) We are assuming that g and g′ are the coupling constants for the gauge interactions associated with the groups SUL (2) and UY (1) respectively.52). 14.60) .4) and (12.

63) is that the right-handed fermions couple twice as strongly to the Yµ . therefore. An alternative a way to see this is to note that the ﬁelds Wµ do not carry weak hypercharge quantum number just as Yµ does not carry weak isospin quantum number and. 2 2 (14.602 14 Higgs phenomenon and the standard model is clearly invariant under the inﬁnitesimal gauge transformations of SUL (2) and UY (1) deﬁned by 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) . We have also used a vector notation here to represent the three components of the isospin vectors which is often used interchangeably. The gauge invariant Lagrangian density for the matter ﬁelds (leptons) can be obtained through minimal coupling.) The other thing to note from the structure of (14. Since the two symmetry groups are commuta ing (direct products). g′ (14. the gauge ﬁeld Wµ for the group SUL (2) does not transform under the group UY (1) and vice versa. they are inert under the corresponding transformations. g g 1 ∂µ ǫ(x). sometimes it is conventional to denote the generators as τ by making the identiﬁcation 2 σ = τ . the Lagrangian density for the leptons takes the form ig′ σ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R. 2. a = 1.63) Lf = iLγ µ ∂µ + ig Here we have used the fact that the generators of angular momentum 1 (SU (2)) are related to the Pauli matrices as 2 σ in the fundamental representation to which the left-handed doublets belong.62) a δWµ (x) = δYµ (x) = where ǫa (x). (We note here that within the context of isospin. we simply have to remember that there are two local symmetries present in the theory. 3 and ǫ(x) denote the local inﬁnitesimal parameters of symmetry transformations associated with the groups SUL (2) and UY (1) respectively. Since the left-handed fermions belong to an isospin doublet while the right-handed fermions are isospin singlets.

65) can be checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)φ(x) + ǫ(x)φ(x). The transformations (14. since we .64). Besides the gauge ﬁelds and the leptons. We note that we have combined the three components of ǫa (x). The Lagrangian density (14. 2 2 (14. the coupling to the UY (1) gauge ﬁeld is correspondingly with the opposite sign. 3 into a vector ǫ(x) in (14.62). 2. we also need a Lagrangian density for the scalar Higgs ﬁelds which are expected to lead to a spontaneous breakdown of the local symmetries.63) can be easily checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). The minimally coupled Lagrangian density consistent with the assignment of the SUL (2) and UY (1) quantum numbers takes the form ig′ σ · Wµ − Yµ φ 2 2 † LH = ∂µ + ig ∂ µ + ig σ ig′ · Wµ − Y µ φ.65) We note that since the hypercharge of the Higgs doublet is just the opposite of that of the left-handed lepton doublet. 2 2 δφ(x) = −i (14. δL(x) = −i (14. 2 2 δR(x) = −iǫ(x)R(x). massless in this theory to begin with). therefore.3 The standard model 603 ﬁeld as the left-handed ones simply reﬂecting the Y quantum numbers for the two ﬁelds.62).66) together with the gauge ﬁeld transformations in (14. A mass term as well as a potential for the Higgs ﬁeld can be introduced consistent with the SUL (2) symmetry. Note also that the SUL (2) invariance forbids a mass term for the lepton ﬁelds (the electron and the neutrino are. In fact.14.64) show explicitly that the right-handed fermions do not change under the SUL (2) transformations.64) together with the transformations in (14. a = 1. The minimally coupled fermion Lagrangian density (14.

φ† .65). collecting all the terms in (14. 2 2 δR(x) = −iǫ(x)R(x).24).67) which is invariant under the local transformations in (14. g′ (14.70) δYµ (x) = . here the Higgs ﬁeld is represented by an isospin doublet) V φ. the fermions can have Yukawa interactions with the scalar ﬁelds which can be described by the Lagrangian density LY = −h Rφ† L + LφR . the Yukawa interaction leads to masses for fermions after the spontaneous breakdown of the symmetry.604 14 Higgs phenomenon and the standard model would like the potential for the Higgs ﬁeld to lead to the spontaneous breakdown of the local symmetries we choose it to have the form (see (14. (14. φ† = −m2 φ† φ + λ † 2 φφ .60). (14. g g 1 ∂µ ǫ(x). In addition. of course.68). (14.63).66). Therefore. (14. δL(x) = −i δφ(x) = −i a δWµ (x) = i σ · ǫ(x)φ(x) + ǫ(x)φ(x). it is worth noting that in spite of the similarity in their forms for the potential.67) and (14. (14. (14.68) can be checked to be invariant under the transformations (14.66).64) and (14. 2 2 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) . 4 λ > 0. As we will see shortly.69) which is invariant under the inﬁnitesimal local gauge transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). the complete weak interaction Lagrangian density describing just one family of leptons (electron family) can be written as L = LG + Lf + LH + LY − V φ.68) where h denotes the Yukawa coupling and the Lagrangian density (14.

φ† = 0. λ (14.29)) 0 v √ 2 φ = .2) that the ﬁrst solution in (14. We know that both SUL (2) and UY (1) symmetries are violated in weak interactions (the associated quantum numbers are not conserved in weak processes). λ (14. the weak interactions are short ranged so that we would like the gauge bosons to be massive. φ† = −m2 φ† φ + λ > 0. there are two solutions to (14.74) .71) so that the minimum of the potential occurs at ∂V ∂φ ∂V ∂φ† λ † φφ 2 = φ† − m2 + = − m2 + = 0. namely.72). Furthermore.3.3 Spontaneous symmetry breaking. 2m2 .67) in the form λ † 2 φφ .72) λ † φφ 2 As we have seen in the last section. 4 V φ. the factor of √2 below is a consequence of the deﬁnition in (14. (14.5 (and also in section 14. (14. 2m v=√ . we have noted earlier in section 7. Both these issues can be addressed simultaneously if the local symmetries are spontaneously broken by the Higgs phenomenon and this is what we discuss next.14. We have chosen the Higgs potential (14. φ = 0.73) φ φ† φ = = Furthermore.3 The standard model 605 14. furthermore. Let us choose the solution for the minimum to be of the form (recall that there is an inﬁnity of possible minima lying on a 1 circle.73) represents a local maximum while the second solution corresponds to the true minimum of the potential.

58) we obtain . (14. to determine the spectrum of particles we need to expand the theory around the true ground state (vacuum) by shifting the scalar ﬁeld as 0 v √ 2 φ →φ+ φ =φ+ .75) However. (14.77) σ · Wµ = 2 = 1 2 √ + 2Wµ φ+ φ0 .65).68)). it follows that 3 Wµ √ − 2Wµ (14. before shifting the ﬁelds let us rewrite the Lagrangian densities for the Higgs and the Yukawa sector of the theory for simplicity (see (14. (14. √ − 3 2Wµ −Wµ (14.76) where we have deﬁned 1 ± 1 2 Wµ = √ (Wµ ∓ iWµ ).) We note that we can write explicitly in the matrix form 3 Wµ 2 1 Wµ − iWµ σ · Wµ φ = 2 = 1 2 1 2 1 2 3 Wµ + iWµ −Wµ √ + 3 2Wµ Wµ . as we have discussed earlier. this is the sector of the theory that involves the scalar ﬁeld and will be aﬀected by the shift. 2 As a result.606 14 Higgs phenomenon and the standard model Therefore. (Basically.67) and (14. the other parts will not change under the shift.78) 3 −Wµ 1 1 3 + + 0 √ 2 Wµ φ + 2 Wµ φ 1 √ W − φ+ − 1 W 3 φ0 µ µ 2 2 Using these as well as the explicit doublet representation of the Higgs ﬁeld in (14.

φ† = ∂µ + ig σ ig′ · Wµ − Yµ φ 2 2 † ∂ µ + ig σ ig′ µ · Wµ − Y φ 2 2 2 −h Rφ† L + LφR + m2 φ† φ − = ∂µ + × + λ † φφ 4 ig − 0 ig′ ig 3 Yµ − Wµ φ− − √ Wµ φ 2 2 2 ∂µ − ig′ µ ig µ3 + ig φ + √ W µ+ φ0 Y + W 2 2 2 ∂µ + × ig 3 0 ig′ ig + Yµ + Wµ φ − √ Wµ φ− 2 2 2 ig ig′ µ ig µ3 0 φ + √ W µ− φ+ Y − W 2 2 2 0 ∂µ − −h eR φ− νeL + φ eL + ν eL φ+ + eL φ0 eR +m2 φ− φ+ + φ φ0 − 0 λ − + 0 2 φ φ + φ φ0 .79) in the Higgs and the Yukawa sector becomes LH + LY − V φ.79) The expansion around (14. 2 v 0 0 φ → φ +√ .80) under which the Lagrangian density (14. φ† = ∂µ + ig − 0 ig 3 ig′ v Yµ − Wµ φ− − √ Wµ φ + √ 2 2 2 2 .14.3 The standard model 607 LH + LY − V φ. 4 (14. 2 (14.74) can be done equivalently by letting v φ0 → φ0 + √ .

the complete quadratic Lagrangian density is obtained to be .60) and (14.82) hv λv 2 0 0 2 φ +φ . However.81) gives ∂µ φ− ∂ µ φ+ + ∂µ φ ∂ µ φ0 + + + 0 igv + µ − igv − µ + Wµ ∂ φ − Wµ ∂ φ 2 2 iv g2 v 2 + µ− 0 3 Wµ W − √ (g′ Yµ + gWµ )∂ µ (φ − φ0 ) 4 2 2 v2 ′ 3 g Yµ + gWµ 8 g′ Y µ + gW µ3 (14. Adding the quadratic terms from the gauge and the fermion Lagrangian densities in (14.63) respectively. we cannot truly determine the spectrum of the theory until we diagonalize the quadratic part of the Lagrangian density. The spectrum of the theory can be obtained from the quadratic part of the Lagrangian density and the quadratic part of (14.81) . (g′ Yµ + gWµ ) as well ± appear to have become massive. − √ (eR eL + eL eR ) − 8 2 0 3 We note that the ﬁeld combinations (φ0 + φ ). since as the ﬁelds Wµ there is mixing between various ﬁelds.608 × + × ∂µ − ∂µ + ∂µ − 14 Higgs phenomenon and the standard model ig v ig′ µ ig µ3 + φ + √ W µ+ φ0 + √ Y + W 2 2 2 2 ig 3 ig′ Yµ + Wµ 2 2 ig′ µ ig µ3 Y − W 2 2 ig + v 0 φ + √ − √ Wµ φ− 2 2 ig v φ0 + √ + √ W µ− φ+ 2 2 v 0 − heR νeL φ− − heR eL φ + √ − hν eL eR φ+ 2 v v 0 −heL eR φ0 + √ + m2 φ− φ+ + φ + √ 2 2 − v λ − + 0 φ φ + φ +√ 4 2 v φ0 + √ 2 2 v φ0 + √ 2 (14.

85) . (14.83).84) to write 3 Wµ = Zµ cos θW − Aµ sin θW .84) where θW is known as the Weinberg angle (also called the weak mixing angle). 8 2 It is now a straightforward matter to diagonalize the Lagrangian density (14. Aµ = g g2 + g′2 Yµ − g′ g2 + g′2 3 Wµ 3 = cos θW Yµ − sin θW Wµ .14.83) + λv 2 0 hv 0 2 − √ (eR eL + eL eR ) − φ +φ . We can invert the relations in (14.3 The standard model 609 LQ = − 1 + + ∂µ Wν − ∂ν Wµ ∂ µ W ν − − ∂ ν W µ − 2 1 3 3 − ∂µ Wν − ∂ν Wµ ∂ µ W ν 3 − ∂ ν W µ 3 4 1 / / − (∂µ Yν − ∂ν Yµ ) (∂ µ Y ν − ∂ ν Y µ ) + ieL ∂ eL + iν eL ∂ νeL 4 +ieR ∂ eR + ∂µ φ− ∂ µ φ+ + ∂µ φ ∂ µ φ0 + / 0 g2 v 2 + µ− Wµ W 4 + igv + − (Wµ ∂ µ φ− − Wµ ∂ µ φ+ ) 2 iv 0 3 − √ g′ Yµ + gWµ ∂ µ φ − φ0 2 2 v2 ′ 3 g Yµ + gWµ 8 g′ Y µ + gW µ3 (14. Yµ = Zµ sin θW + Aµ cos θW . Let us deﬁne g′ g2 + g′2 g g2 + g′2 Zµ = Yµ + 3 Wµ 3 = sin θW Yµ + cos θW Wµ . (14.

(14. for example.86) 1 √ (σ 2 ∂µW ν 3 − ∂ν W µ 3 Similarly. 2 (14. if we deﬁne the parameters gv . 2 cos θW hv √ . 2 i 0 √ φ − φ0 . 2 λ 4m2 λv 2 = 2 2 λ = 2m2 . (14.89). 2 2 (14.89) MW MZ = = me = 2 MH = then together with (14. we can deﬁne (this is equivalent to deﬁning φ0 = iχ).87) so that we can write 0 ∂µ φ ∂ µ φ0 = 1 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ.29)) σ = χ = 1 0 √ φ + φ0 .83) as . 2 MW g2 + g′2 v = . see. we can write the quadratic part of the Lagrangian density (14. (14.610 14 Higgs phenomenon and the standard model With this.84)-(14. φ = 0 1 √ (σ 2 + − iχ).88) Furthermore. it is easy to check that 3 3 ∂µ Wν − ∂ν Wµ = (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) + (∂µ Yν − ∂ν Yν ) (∂ µ Y ν − ∂ ν Y µ ) + (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) .

91) .3 The standard model 611 LQ = − 1 + + ∂µ Wν − ∂ν Wµ ∂ µ W ν − − ∂ ν W µ − 2 1 − (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) 4 1 − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) 4 / / / + ieL ∂ eL + ieR ∂ eR + iν eL ∂ νeL +∂µ φ− ∂ µ φ+ + 1 1 M2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ − H σ 2 2 2 2 2 MZ Zµ Z µ − me (eR eL + eL eR ) 2 + − 2 + +iMW (Wµ ∂ µ φ− − Wµ ∂ µ φ+ ) + MW Wµ W µ − −MZ Zµ ∂ µ χ + = − 1 + + ∂µ Wν − ∂ν Wµ ∂ µ W ν − − ∂ ν W µ − 2 i i + 2 ∂µ φ+ W µ − + ∂ µ φ− +MW Wµ − MW MW 1 − (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) 4 + 2 1 MZ Zµ − ∂µ χ 2 MZ Zµ − 1 µ ∂ χ MZ 1 − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) 4 1 M2 / / + ∂µ σ∂ µ σ − H σ 2 + e(i∂ − me )e + iν eL ∂ νeL .14.90) 2 2 where we have combined the two chirality states eL .56)) e = eL + eR . eR to deﬁne the massive electron ﬁeld as (see (14. If we now redeﬁne the (massive) gauge ﬁelds as (14.(14.

namely.612 14 Higgs phenomenon and the standard model ± Wµ ∓ i ± ∂µ φ± → Wµ . The charged ± vector (spin 1) bosons Wµ are massive with mass MW = gv . The Higgs √ particle has the usual mass MH = 2m (see (14.95) One of the two neutral vector bosons.38)).90) becomes completely diagonal with the form LQ = − 1 + + ∂µ Wν − ∂ν Wµ 2 2 + ∂ µ W ν − − ∂ ν W µ − + MW Wµ W µ − 1 M2 − (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) + Z Zµ Z µ 4 2 1 1 − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) + ∂µ σ∂ µ σ − 4 2 / / +e(i∂ − me )e + iν eL ∂ νeL . we see that after spontaneous breakdown of symmetry the electron has become massive with mass hv me = √ . MW 1 ∂µ χ → Zµ .94) while the neutrino remains massless as we would like.92) Zµ − then the quadratic Lagrangian density (14. 1 2 2 M σ 2 H (14. 2 (14.93) is now completely diagonalized and shows some very desirable features. MZ (14.93) The quadratic Lagrangian density (14. Although the starting theory had a massless electron. 2 (14. Zµ is also massive with mass .

93). That this is indeed true can be seen as follows. This suggests that there is still a residual unbroken gauge symmetry in the theory which can possibly be identiﬁed with the low energy electromagnetic gauge symmetry UEM (1). Let us look at the minimally coupled fermion Lagrangian density (14. (14.14. 2 cos θW (14. . namely.63) Lf = iLγ µ ∂µ + ig = i ν eL e L σ ig′ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R 2 2 ′ ig + 3 ∂µ + ig Yµ + ig Wµ νeL + √2 Wµ eL 2 2 γµ ig ′ ig ig 3 − ∂µ + 2 Yµ − 2 Wµ eL + √2 Wµ νeL ig + ig 3 ig′ Yµ + Wµ νeL + √ Wµ eL 2 2 2 ig′ ig − ig 3 Yµ − Wµ eL + √ Wµ νeL 2 2 2 + ieR γ µ ∂µ + ig′ Yµ eR = iν eL γ µ + ieL γ µ ∂µ + ∂µ + + ieR γ µ ∂µ + ig′ Yµ eR / / / = iν eL ∂ νeL + ieL ∂ eL + ieR ∂ eR g g + − − √ Wµ ν eL γ µ eL − √ Wµ eL γ µ νeL 2 2 g 3 − Wµ (ν eL γ µ νeL − eL γ µ eL ) 2 .3 The standard model 613 MZ = MW g2 + g′2 v = . Aµ which is massless.84)) g′ g2 + g′2 g g2 + g′2 sin θW = cos θW = .97) There is another neutral vector boson in the theory (14.96) where the Weinberg angle θW is deﬁned by (see (14.

indeed we can think of the spontaneous symmetry breaking as leaving a residual UEM (1) gauge invariance as the low energy symmetry of the electroweak theory. We also note from (14. (14. 2 2 2 .614 − 14 Higgs phenomenon and the standard model g′ Yµ (ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) 2 / / / = iν eL ∂ νeL + ieL ∂ eL + ieR ∂ eR g − + − √ Wµ ν eL γ µ eL + Wµ eL γ µ νeL 2 g − (Zµ cos θW − Aµ sin θW ) (ν eL γ µ νeL − eL γ µ eL ) 2 g′ − (Zµ sin θW + Aµ cos θW )(ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) 2 / / = iν eL ∂ νeL + ie∂ e g − + − √ Wµ ν eL γ µ eL + Wµ eL γ µ νeL 2 − Zµ 2 g2 + g′2 gg′ g2 + g′2 g2 (ν eL γ µ νeL − eL γ µ eL ) + g′2 (ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) − Aµ eγ µ e. (14. in addition to the correct electromagnetic current. 2 2 2 g g + Jµ = − √ eL γµ νeL = − √ eL γµ (1 − γ5 ) νeL .98) We note from (14.98) that the massless neutral vector boson Aµ couples only to the charged fermions as the photon should and we can identify the electric charge as gg′ g2 + g′2 e= = g sin θW = g′ cos θW . both charged as well as neutral weak currents given by g g − Jµ = − √ ν eL γµ eL = − √ ν eL γµ (1 − γ5 ) eL .98) that the standard model has.99) Thus.

8MW (14.103) Experimental measurements lead to a value of the Weinberg angle as . 4 2GF 4 2GF sin2 θW = (14.102) which leads to 2 MW = (14. (14.100) We recall that the V − A theory describes only charged current interactions. they would lead to a low energy eﬀective current-current interation Hamiltonian density in the lowest order of the form (basically.49) (for just one family of leptons) and comparing the two we determine G √F 2 g2 .14. 2 8MW g2 e2 √ = √ . 14.101) This has exactly the form of the V − A four-fermion interaction in (14. W Figure 14. two charged currents interact through the charged W gauge boson propagator (exchange) as shown in Fig.98).3 The standard model 0 Jµ = − 615 g2 (ν eL γ µ νeL − eL γ µ eL ) 1 2 g2 + g′2 +g′2 (ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) . If we look at the charged current interaction terms in (14.1: Charged current-current interaction in the lowest order.1 and the propagator reduces 2 to a multiplicative factor of M12 in the low energy limit p2 ≪ MW ) W HI = g2 µ 2 ν eL γµ (1 − γ5 ) eL eL γ (1 − γ5 ) νeL .

et le Radium 7.105) These gauge bosons have been discovered and their masses are determined very close to the theoretically predicted values. Such processes have also been experimentally observed. Furthermore. for example. The standard model compares very well with the experimental results. J. de Phys. 14. the standard model diﬀers from the conventional Fermi theory in (14. cos θW 1 2 ≃ 80 GeV.51) and recall that we have set = c = 1) e2 1 ≃ .616 14 Higgs phenomenon and the standard model sin2 θW ≃ 0. although bounds on the value of its mass can be put from various other arguments.49) in that it predicts weak neutral currents (in addition to the conventional charged weak currents) and hence processes where a neutral heavy vector boson (Zµ ) is exchanged. Search for Higgs particles is a top priority in many experiments at the upcoming LHC. 4π 137 we can now determine MW MZ = = GF ≃ 10−5 (GeV)−2 . Proca.23. 347 (1936). e2 √ 4 2GF sin2 θW MW ≃ 90 GeV. the WeinbergSalam-Glashow theory does not predict a unique mass for the Higgs particle since its mass depends on the quartic coupling λ. (14. A. . Once again these can be accomodated into extensions of the standard model that we will not go into here. (14.4 References 1. The standard model also suﬀers from the fact that the neutrino is massless in this theory while we now believe that the neutrino may have a small mass (from neutrino oscillation experiments). (14. However.104) and using the values of other constants (see.

321 (1964).4 References 617 2. Englert and R. Physical Review 145. Physical Review Letters 13. u 3. 11. W. 5. Physical Review Letters 19. Kibble. G. Physical Review 109. 4. . 209 (1957). 1264 (1967). S. 1156 (1966). P. F. 1860 (1958). 10. E. C. Physical Review 109. E. R. R. W. G. G. H. Nuclear Physics B35. Higgs. Bernstein. Helvetica Physica Acta 30.14. J. Guralnik. 6. G. Feynman and M. 7 (1974). Marshak. C. 167 (1971). S. ’t Hooft. Proceedings of PaduaVenice conference on mesons and newly discovered particles. E. 3265 (2004). 193 (1958). Ruegg and M. International Journal of Modern Physics A19. P. 8. Weinberg. (1957). 9. Reviews of Modern Physics 46. Physical Review Letters 13. Brout. 7. Hagen and T. St¨ckelberg. Gell-Mann. 585 (1964). Sudarshan and R. C. B. Ruiz-Altaba.

This page intentionally left blank .

We have also carried out a few calculations of simple Feynman diagrams describing physical processes in section 9. However. as we go to higher orders in perturbation.1 Introduction So far we have discussed the basic structures of various quantum ﬁeld theories. The simple diagrams in 9. As a result.6 are known as tree diagrams where the momenta of all the internal propagators are uniquely determined (through energy-momentum conserving delta functions) in terms of the external momenta (in the diagrams in 9. As we have discussed earlier (see section 13. in general.4).Chapter 15 Regularization of Feynman diagrams 15. any scattering matrix element (S matrix element) in a given quantum ﬁeld theory can be built out of 1PI (one particle irreducible) graphs. Instead some of the internal propagators may involve additional momenta (loop momenta) that are integrated over all possible values. the evaluation of Feynman diagrams may lead to divergences depending on the structure of the integral that is being evaluated. This procedure is known. We will study this question in a systematic manner developing the necessary ideas in this chapter as well as in the next. as renormalization of a quantum ﬁeld theory and involves several steps. the Feynman diagrams become topologically more complicated and may contain internal propagators whose momenta are not uniquely determined in terms of the external momenta. we have to ﬁnd a way of extracting meaningful results from such a quantum ﬁeld theory. In such a case.6 there was only one internal propagator). For example. let us look at one of the simplest interacting ﬁeld 619 . We recall that an 1PI graph is deﬁned to be a (vertex) graph which cannot be separated into two disconnected graphs by cutting a single internal line.6.

which are 1PI graphs whereas the graph in Fig. therefore. (15. 15. given these Feynman rules. With these Feynman rules. we can calculate any perturbative amplitude in the φ4 theory.2. it is suﬃcient to concentrate on the 1PI graphs. The Feynman rules for this theory are given by i .1) where the coupling constant λ is assumed to be positive.36)) M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ .620 15 Regularization of Feynman diagrams theories. 2 2 4! L= (15. 15. − M2 p p4 p3 = iGF (p) = p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). . Clearly the 1PI diagrams are fundamental since any other diagram can be built using them as basic elements.1 (which topologically involve loops and. is not. for example. are called loop diagrams) Figure 15.2) p1 p2 where all the momenta in the vertex are assumed to be incoming and we have not written the “iǫ” term in the propagator explicitly although it should be understood. In fact. namely.1: A few examples of 1PI diagrams in the φ4 theory. in studying quantum ﬁeld theories at higher orders. Therefore. we can construct the graphs in Fig. the scalar φ4 theory described by the Lagrangian density (see (6.

Let us note that if our theory is described by a Lagrangian density such that (although we are assuming the basic ﬁeld variable to be a scalar ﬁeld. let us discuss the notion of loop expansion in a quantum ﬁeld theory in this section.15. a) = a−1 L (φ. we are forced to give a meaning to the Feynman amplitudes in some manner and this process is known as regularization of the diagrams. the integrals over the arbitrary loop momenta may not. 15. and P is the power of “a” associated with this graph.2 Loop expansion As we have mentioned repeatedly. it is suﬃcient to study the divergence behavior of only the 1PI graphs. Namely. to study the divergence structure of any amplitude (graph).1 and Fig. problems with divergences arise in a quantum ﬁeld theory at the loop level. 15. Since the 1PI graphs are fundamental. we invariably run into divergences. while the propagator of the theory which is the inverse of the tree level two point function will have a factor “a” associated with it.3) where a is a constant parameter. 15. then it is clear that each interaction vertex in the theory will have a factor a−1 multiplied to it. (15.2: An example of a diagram in the φ4 theory which is not 1PI. this discussion applies to any quantum ﬁeld theory) L (φ.2 Loop expansion 621 Figure 15. In such a case. In the discussions below we will introduce a number of regularization schemes that are commonly used in studying divergent Feynman amplitudes. As we have mentioned. Therefore. be convergent. in evaluating scattering amplitudes particularly at the loop level (where the topology of the diagram involves loops as in Fig. ∂µ φ. in general. ∂µ φ) .2). then it follows easily that . Thus if we are considering a proper vertex graph (1PI graph).

therefore. (15. but for a clear understanding of the loop expansion we are going to restore in this section only) i Z[J] = N Dφ e SJ . . On the other hand the number of loops in a graph is deﬁned as the number of independent momenta which we have to integrate over (after all the energy-momentum conserving delta functions have been taken care of). we can eliminate I − V to obtain L = P + 1. Realizing that with every internal line there is a momentum which is integrated and that each vertex has a delta function enforcing the conservation of energy and momentum at that vertex.5) where the term (+1) reﬂects the fact that every 1PI vertex function has an overall energy-momentum conserving delta function associated with it. (15.120). From the two relations in (15. (15.622 15 Regularization of Feynman diagrams P = I − V.4) and (15. is given by L = I − V + 1.6) In other words.3). be thought of as the parameter of expansion in the number of loops.5). (15. the number of loops (independent momenta) in such a graph. we have set = 1 throughout our discussion. therefore. the generating functional is deﬁned as (see. Let us recall that in the path integral description of a quantum ﬁeld theory. for example. (12. the power of “a” in a diagram in such a theory also determines the number of loops in the diagram and can.4) where I denotes the number of internal lines and V the number of interaction vertices in the graph.7) where represents the Planck’s constant and we see that it can be thought of as an overall multiplicative parameter like “a” in (15.

for example. for example. We are usually accustomed to the idea of perturbation in powers of the coupling constant in the theory.3 Cut-oﬀ regularization To get an idea of how divergences arise in a quantum ﬁeld theory at the loop level. This is quite important. the entire perturbation expansion has to be rearranged which is nontrivial. when we study theories where there is spontaneous breakdown of a symmetry.6). let us calculate the simplest nontrivial graph in the φ4 theory (15. In such a case. The reason why such an expansion is advantageous can be seen from the fact that since this parameter multiplies the whole action. In addition the small value of makes it a legitimate expansion (perturbation) parameter. sections 7. Therefore.1) which gives the one loop correction to the two point function. 15. as we have seen (see. Factoring out an overall momentum conserving delta function (2π)4 δ4 (p1 −p2 ) and identifying p1 = p2 = p we have (we will always factor out the overall momentum conserving delta function along with the factor of (2π)4 ) k iλ 2 d4 k i (2π)4 k2 − M 2 p p =I=− . the scalar ﬁeld picks up a vacuum expectation value which depends on the coupling constant of the theory.3 Cut-off regularization 623 As a result. the expansion is unaﬀected by how one divides the total action into a free part and an interaction part.5 and 14. On the other hand.2). when we perturb around the true vacuum (by shifting the ﬁeld variable by its vacuum expectation value). expanding in powers of in a quantum ﬁeld theory also corresponds to expanding in the number of loops as described in (15.15. the loop expansion is unaﬀected by shifting of ﬁelds since the parameter of expansion is a multiplicative factor in front of the total action and as a result the loop expansion is quite useful. Loop expansion is a diﬀerent perturbative expansion.

8) has 1 an associated symmetry factor of 2 and have rotated the momentum to Euclidean space by letting k0 → ik0E in the intermediate step to facilitate the evaluation of the integral.8) dy 1 − Here we have used the fact that the Feynman diagram in (15.8) takes the form . we have used the value of the angular integral in four dimensions which can be easily determined as π π 2π dΩ = 0 π dθ1 sin2 θ1 0 dθ2 sin θ2 0 1 −1 dθ3 = 0 dθ1 1 (1 − cos 2θ1 ) 2 d cos θ2 × (2π) (15. therefore.8) that this integral is divergent and. We see from (15. Furthermore. With this regularization. y + M2 (15. we deﬁne (regularize) this integral by cutting oﬀ the momentum integration at some higher value Λ and then taking the limit Λ → ∞ at the end of the calculation. the two point function in (15. such divergences are also known as ultraviolet divergences.) Since large values of momentum correspond to small values of coordinate separation.624 = − = − = − = − = − = − iλ 2 15 Regularization of Feynman diagrams id4 kE i 4 −k 2 − M 2 (2π) E d4 kE dΩ 0 ∞ 0 iλ 2(2π)4 iλ 2(2π)4 k2 ∞ 1 2 E + M 3 kE dkE k2 1 2 E + M iλ (2π 2 ) 2(2π)4 iλ 32π 2 iλ 32π 2 ∞ 0 ∞ 0 1 2 k2 dkE 2 E 2 2 kE + M dy y + M2 − M2 y + M2 M2 . (This is the reason for the name cut-oﬀ regularization.9) = 2π 2 .

However. (15. 2 ˜ 32π M Λ2 (15. Λ are scalar parameters).12) . We add here for completeness that the cut-oﬀ can also be implemented through a weight factor in the following manner.11) ˜ Λ→∞ n γ = lim n→∞ k=1 1 − ln n k = 0. since there is no momentum vector in the ﬁnal result in (15.8) in a regularized manner as ∞ 0 I = − = = = where iλ 32π 2 dy 1 − ∞ 0 M2 y + M2 M2 y + M2 ˜ e− Λ2 y ˜ Λ→∞ lim − iλ 32π 2 dy 1 − ˜ Λ→∞ lim − lim − iλ ˜ 2 M2 1 Λ + M 2 γ + ln + O( ) 2 ˜2 ˜2 32π Λ Λ ˜ 1 Λ2 iλ ˜ 2 Λ + M 2 γ − ln 2 + O( ) .577215665.10) (and m.10) The result at the end should be rotated back to the Minkowski space.3 Cut-off regularization 625 iλ I = lim − Λ→∞ 32π 2 = lim − Λ2 0 dy − M 2 0 Λ2 dy y + M2 Λ2 0 Λ→∞ iλ Λ2 − M 2 ln y + M 2 32π 2 iλ Λ2 − M 2 ln 32π 2 Λ2 + M 2 M2 = = Λ→∞ lim − Λ→∞ lim − iλ Λ2 Λ2 − M 2 ln 2 + O 32π 2 M 1 Λ2 . (15.15. in this case this is also the result in the Minkowski space. We note that we can write (15.

Gradshteyn and Ryzhik 3.214. for example.8) can be evaluated without rotating the momentum into Euclidean space as follows iλ 2 i d4 k (2π)4 k2 − M 2 + iǫ d3 kdk0 I = − = = λ 2(2π)4 λ (−2πi) 2(2π)4 1 (k0 − (Ek − iǫ))(k0 + Ek − iǫ) 1 iλ =− 2Ek 32π 3 d3 k 1 .10) with the identiﬁcation ˜ Λ2 = Λ2 − M 2 γ.14) does not depend on the angular variables. Ek (15. the integral takes the form iλ (4π) 32π 3 ∞ 0 ¯ Λ I = − dy y2 y2 + M 2 y2 + M 2 − M2 y2 + M 2 y2 + M 2 ¯ Λ 0 iλ = lim − 2 ¯ 8π Λ→∞ = ¯ Λ→∞ dy 0 √ ¯ ¯ iλ ¯ ¯ 2 Λ + Λ2 + M 2 2 − M 2 ln = lim − Λ Λ +M ¯ 16π 2 M Λ→∞ 2 ¯ 1 4Λ iλ ¯ + O( ¯ 2 ) . (15.1).14) d3 k where we have identiﬁed (see.4 and 8.352. Furthermore. (5. deﬁning y = |k|. (15. This result coincides with (15. for example.626 15 Regularization of Feynman diagrams denotes the Euler’s constant (also known as the Euler-Mascheroni constant and sometimes denoted by C) and we have used the standard tables of integrals (see.47)) Ek = |k|2 + M 2 . we also note here that the diagram in (15.13) Furthermore. Since the integrand in (15. which gives (4π) in three dimensions.15) 2Λ2 + M 2 − M 2 ln = lim − 2 2 ¯ M Λ Λ→∞ 32π lim − iλ 1 y 8π 2 2 y 2 + M 2 − M 2 ln y + . the integration over the angles can be carried out.

10) if we identify 1 ¯ (15.17) dτ −τ M 2 .3 Cut-off regularization 627 Here we have used the standard table of integrals in the intermediate step (see.15. This result coincides with (15.271. but is divergent at the lower limit τ = 0.17) is well behaved for large values of τ . The integral in (15.8) can also be evaluated in an alternative manner which is sometimes useful and so let us discuss the alternate method as well.2 and 2.16) Λ2 = Λ2 − M 2 (1 − ln 2) . 2.4). for example. Gradshteyn and Ryzhik.10) has been transformed into an infrared divergence in the Schwinger parameter τ . We note that the integrand in (15. the ultraviolet divergence in (15.271. Thus. This integral can be regularized by cutting 1 oﬀ the integral at some lower limit Λ2 as I = lim − iλ 32π 2 ∞ 0 Λ→∞ dτ −τ M 2 − 1 Λ2 τ e τ2 1 2M 2 Λ −1 2 iλ ×2 = lim − Λ→∞ 32π 2 K−1 2M Λ . e τ2 Here Γ(2) denotes the gamma function and the parameter τ is known as the Schwinger parameter (or the proper time parameter). 2 This derivation makes it clear that the evaluation of the amplitude is much easier in the Euclidean space. We note that the integral in Euclidean space has the form iλ × 2π 2 2(2π)4 iλ 32π 2 iλ 32π 2 iλ 32π 2 iλ 32π 2 0 ∞ 0 ∞ 0 ∞ 0 ∞ ∞ 0 I = − = − = − = − = − k2 1 2 dkE 2 E 2 2 kE + M dy y y + M2 2 dydτ y e−τ (y+M ) dτ 1 2 Γ(2) e−τ M 2 τ (15.

It is also clear from this simple calculation that the nature of the divergence depends on the dimensionality of space-time (we are restricting ourselves to four space-time dimensions. However.18) Λ→∞ lim − iλ M2 Λ2 + M 2 ln 2 + O 32π 2 Λ iλ Λ2 Λ2 − M 2 ln 2 + O 2 32π M Λ→∞ lim − where Km (z) denotes the modiﬁed Bessel function of the second kind of order m and we have used the standard tables of integral (see.18) can be compared with (15. for example. Equation (15. For simplicity we would consider all the incoming momenta to vanish. regularizing the integral brings out the divergence structure of the Feynman graph.18)) that the self-energy diagram at one loop contains a term that diverges quadratically as well as a term which is logarithmically divergent.18) can also be obtained by simply cut1 ting oﬀ the τ integral at the lower limit Λ2 (without the regularizing exponential factor). we note that the result in (15.10) (or (15. We note that since the cutoﬀ Λ has to be actually taken to inﬁnity in the ﬁnal result.3. (15. This diagram is also divergent (in four dimensions) and with the cut-oﬀ regularization leads to (we assume p1 = p2 = p3 = p4 = 0 for simplicity so that each graph contributes the same amount and also because we are interested in looking at only the divergence structure of the graph) .628 = = = = lim − lim − 15 Regularization of Feynman diagrams Λ→∞ iλ × (2ΛM )K1 32π 2 iλ × (2ΛM ) 32π 2 2M Λ 2M Λ −1 Λ→∞ + M M ln + ··· Λ Λ 1 Λ2 1 Λ2 . namely. but quantum ﬁeld theories can be deﬁned in any number of space-time dimensions). 15.471. Without going into details. Gradshteyn and Ryzhik 3. we note from (15.10) and we see that both ways of evaluating the integral lead to the same result. the one loop diagram for the vertex correction shown in Fig. For example.9) to evaluate the integral over τ . this graph is divergent. Let us next look at another one loop graph.

(15.3 Cut-off regularization 629 p3 p2 + p1 p4 p1 p3 p2 p4 p4 p3 + p1 p2 Figure 15.19) Λ→∞ lim There are several things to note from this derivation. I = = = 3(−iλ)2 2 3λ2 2(2π)4 3iλ2 2(2π)4 i i d4 k 4 k2 − M 2 k2 − M 2 (2π) id4 kE d4 kE −k2 1 2 E − M 1 2 (kE + M 2 )2 ∞ 0 3 kE dkE 2 = = 3iλ2 (2π 2 ) 2(2π)4 3iλ2 32π 2 3iλ2 32π 2 3iλ2 32π 2 3iλ2 32π 2 lim ∞ 0 ∞ 0 1 (k2 E + M 2 )2 dy y (y + M 2 )2 y + M2 − M2 (y + M 2 )2 Λ2 = dy = Λ→∞ lim 0 dy − M2 y + M2 Λ2 2 ∞ 0 dy (y + M 2 )2 ∞ 0 = = = Λ→∞ lim ln (y + M ) 0 + M2 y + M2 Λ→∞ 3iλ2 Λ2 + M 2 −1 ln 32π 2 M2 Λ2 3iλ2 ln 2 − 1 + O 32π 2 M 1 Λ2 .3: One loop correction to the four point vertex function.15. First we have factored out an overall momentum conserving delta function along .

we have four powers of momentum in the numerator coming from the momentum integration while there are four powers of momentum in the denominator coming from two propagators. Thus. the cutoﬀ regularizes and deﬁnes . we are free to redeﬁne the value of the ground state energy to zero.19). we see that there is a net 4 − 2 = 2 powers of momentum in the numerator reﬂected in the fact that the highest degree of divergence of the diagram is quadratic. as we have seen earlier. can be removed by normal ordering the Hamiltonian of the theory. equivalent to saying that since we only measure diﬀerences in the energy levels. This is. 15. in general. each of these diagrams has an associated symmetry factor of 1 2 and we have used the value of the angular integral from (15. However. any given quantum ﬁeld theory. but the divergence in this case is logarithmic. First we introduce a regularization procedure (scheme) which gives a meaning to the divergent Feynman integrals by isolating the divergent parts of the diagram. Thus we see that this graph is also divergent (in four dimensions).9) in the intermediate step. The second factor of (2π)4 that comes 1 from the second vertex has cancelled with the factor of (2π)4 in the integration over the momentum of one of the internal propagators. Consequently.19) is ∂ given by − ∂M 2 of that in (15. In the examples we have studied in (15. Renormalization is the process of redeﬁning these inﬁnities that we encounter in perturbation theory. Finally.3. for the ﬁrst graph in (15.8) we have four powers of momentum in the numerator coming from the momentum integration while there are two powers of momentum in the propagator.) Let us note here that these divergences are there in the quantum ﬁeld theory besides the zero point energy or the ground state energy divergence which. (This can also be seen from the fact that the integral in (15. Very broadly it consists of two essential parts. of course. there is a net 4 − 4 = 0 power of momentum in the numerator reﬂected in the logarithmic divergence of the diagram. The divergence structure of any diagram can be determined simply by counting the powers of momentum in the numerator and in the denominator. In fact. is plagued with divergences at every (loop) order in perturbation theory (unless we are in lower dimensions). for the diagrams in Fig. for example. the present divergences are nontrivial and are present even after normal ordering of the theory.10) and (15.8) up to a multiplicative factor. Similarly.630 15 Regularization of Feynman diagrams with the factor of (2π)4 . As a result.

depends analytically on the external momenta. (15.3 Cut-off regularization 631 the integrals. the lack of dependence on the external momenta is simply due to our special choice of the external momenta to be all vanishing. namely. − M2 p = iGF (p) = p2 p p3 p2 p1 p4 p3 = iSF (p) = i(p + m) / i = 2 . p1 p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). 15. Let us note that any Feyn- man diagram (amplitude).19). But let us examine the dependence of diagrams on external momenta (and this will be important in studying renormalization) in another theory.3.15. (15. p−m / p − m2 = −(2π)4 igδ4 (p1 + p2 + p3 ). particularly in the second example (15.20) L = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 4! This theory describes the interaction between a fermion and a scalar ﬁeld and the Feynman rules for this theory are given by i .21) .1 Calculation in the Yukawa theory. In the previous examples. The second part of renormalization consists of removing the cutoﬀ dependence (or regularization dependence) of the theory which we will take up in the next chapter. in general. But in this process the amplitude (and the theory) becomes cutoﬀ dependent. the theory with a Yukawa coupling described by the Lagrangian density 1 λ M2 2 / φ − gψψφ − φ4 .

however.23) = 0 It is worth emphasizing here that this simple combination formula 1 1 holds if both the factors A and B have the same analyticity property (namely. diagrams necessarily involve more internal propagators. Before we proceed to calculate higher order diagrams in this theory. (15. it would be useful to ﬁnd a way to combine denominators and this is achieved by 1 Feynman’s formula as follows.9) which is why the cubic coupling involving fermions is called the Yukawa coupling. In quantum ﬁeld theory at zero temperature. then we note that 1 0 1 dx (xA + (1 − x)B)2 = 0 = − = − = 1 1 1 − A−B A B 1 . let us note that at higher orders in perturbation theory.632 15 Regularization of Feynman diagrams where the arrows in the vertices denote the fact that momenta are all incoming. the iǫ term with the same sign). On the other hand when the two factors have opposite analytic behavior (iǫ terms with opposite sign). we see that we can combine two denominators in a simple manner as 1 1 AB = 0 1 dx (xA + (1 − x)B)2 dx1 dx2 δ(1 − x1 − x2 ) .22) Therefore. (x1 A + x2 B)2 (15. then this formula needs to be generalized (which is important in studying ﬁnite temperature ﬁeld theories). Suppose we would like to write AB as a single factor. The lowest order graph involving the fermions in this theory gives rise to the Yukawa potential (as we have seen in section 8.23) holds and is quite . Therefore. AB 1 1 A − B x(A − B) + B dx (x(A − B) + B)2 1 0 (15.

24) which we have already used in (13. (15.3 Cut-off regularization 633 useful. Generalization of (15.26) Using the Feynman combination formula (15. ( n xi Ai )n i=1 (15. The fermion self-energy at one loop has the form k p k+p p = (−ig)2 i i d4 k 4 k 2 − M 2 (k + p) − m (2π) / / = g2 = g2 = − 1 (k + p) + m / / d4 k 4 k 2 − M 2 (k + p)2 − m2 (2π) ig2 (2π)4 id4 kE 1 −(kE + pE ) + m / / 4 −k 2 − M 2 (2π) − (kE + pE )2 − m2 E d4 kE 2 (kE + M 2 ) (kE + pE )2 + m2 (kE + pE ) − m / / .23).23) to include more denominators is straightforward and has the form n i=1 1 = Ai 1 n dxi 0 i=1 δ(1 − x1 − · · · − xn ) .15. / / (15. Equipped with the Feynman combination formula.25) can be written as (for simplicity we factor out the multiig 2 plicative factor − (2π)4 which we will put back at the end of the calculation) . let us next calculate the two simplest nontrivial graphs involving loops in this theory.121).25) Here we have rotated the momenta as well as the gamma matrices 0 to Euclidean space as γ 0 → iγE so that 0 0 k = γ 0 k0 − γ · k = −γE kE − γ · k = −kE . the integral in (15.

Furthermore.27) where we have deﬁned Q2 = x(1 − x)p2 + xm2 + (1 − x)M 2 .28) In (15. E (15. (15.27) we have shifted the variable of integration kE → kE − xpE in the intermediate step after which the term kE in the numerator / vanishes because of anti-symmetry.634 15 Regularization of Feynman diagrams = d4 kE 0 1 dx 1 (kE + pE ) − m / / 2 x (kE + pE )2 + m2 + (1 − x)(kE + M 2 ) 2 = = 0 d4 kE 0 1 dx (k2 (kE + pE ) − m / / 2 + xm2 + (1 − x)M 2 )2 E + 2xkE · pE + xpE (kE + pE ) − m / / 2 dx d4 kE (kE + xpE )2 + x(1 − x)p2 + xm2 + (1 − x)M 2 E (k2 E 1 = 0 1 dx d4 kE + x(1 − x)p2 + xm2 + (1 − x)M 2 )2 E ∞ 0 3 (2π 2 )kE dkE kE + (1 − x)pE − m / / = 0 dx((1 − x)pE − m) / 1 (k2 E + Q2 ) 2 . the kE integral can be carried out to give (we now ig 2 put back the overall multiplicative factor − (2π)4 ) = − = − = − ig2 16π 2 ig2 16π 2 ig2 16π 2 1 0 1 0 1 0 dx ((1 − x)pE − m) / ∞ 0 dy y [y + Q2 ]2 −1 dx ((1 − x)pE − m) ln / dx ((1 − x)pE − m) / Λ2 + Q2 Q2 × ln = ig2 16π 2 1 0 Λ2 + x(1 − x)p2 + xm2 + (1 − x)M 2 E −1 x(1 − x)p2 + xm2 + (1 − x)M 2 E dx ((1 − x)p + m) / .

/B the self energy graph in (15. g). where it is understood that the limit Λ → ∞ is to be taken.19) in the intermediate step and have rotated back to Minkowski space in the ﬁnal step. Λ. Similarly the scalar self-energy graph is given by k p k+p d4 k i i 4 k − m (k + p) − m (2π) / / / (15. m.30) p = −Tr (−ig)2 = −g2 Tr (k + m)((k + p) + m) / / / d4 k .31) Tr A = 0.15. Using the trace identities Tr ½ = 4. / Tr A / = 4A · B. We also note here that we have used the integral from (15. 4 (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Here the overall negative sign reﬂects the fact that we are evaluating a graph with a fermion loop. M. the trace is a consequence of the fact that the Dirac matrix indices coming from the two fermion propagators are being summed over (there are no free Dirac indices in the diagram).30) takes the form . (15.3 Cut-off regularization 635 × ln Λ2 − x(1 − x)p2 + xm2 + (1 − x)M 2 −1 −x(1 − x)p2 + xm2 + (1 − x)M 2 (15.29) = f (p. Furthermore.

636 15 Regularization of Feynman diagrams = − = − = = g2 (2π)4 4g2 (2π)4 d4 k 4(k · (k + p) + m2 ) (k2 − m2 )((k + p)2 − m2 ) (−kE · (kE + pE ) + m2 ) 2 (−kE − m2 )(−(kE + pE )2 − m2 ) id4 kE d4 kE d4 kE 4ig2 (2π)4 4ig2 (2π)4 1 (kE · (kE + pE ) − m2 ) 2 (kE + m2 )((kE + pE )2 + m2 ) × = = = dx 0 (kE · (kE + pE ) − m2 ) 2 (x((kE + pE )2 + m2 ) + (1 − x)(kE + m2 ))2 1 4ig2 (2π)4 4ig2 (2π)4 4ig2 (2π)4 4ig2 π 2 (2π)4 ig2 4π 2 ig2 4π 2 lim dx 0 1 d4 kE d4 kE (kE · (kE + pE ) − m2 ) ((kE + xpE )2 + x(1 − x)p2 + m2 )2 E 2 (kE − x(1 − x)p2 − m2 ) E 2 (kE + x(1 − x)p2 + m2 )2 E 3 kE dkE 2 (kE + Q2 )2 2 kE − Q2 dx 0 1 0 1 dx 2π 2 0 ∞ 0 ∞ 0 ∞ = dx 0 1 dy y y − Q2 (y + Q2 )2 = dx 0 1 dy y 1 2Q2 − y + Q2 (y + Q2 )2 Q2 2Q2 2Q4 − + y + Q2 y + Q2 (y + Q2 )2 Λ2 0 = dx 0 0 1 ∞ dy 1 − Λ2 = ig2 Λ→∞ 4π 2 +2Q4 0 ∞ dx 0 0 dy − 3Q2 dy y + Q2 dy 1 0 (y + Q2 )2 dx Λ2 − 3Q2 ln Λ2 + Q2 + 2Q2 Q2 = Λ→∞ lim ig2 4π 2 .

32) we can write . we can write = ig2 16π 2 1 0 dx ((1 − x)p + m) / Λ2 − x(1 − x)p2 + xm2 + (1 − x)M 2 −1 −x(1 − x)p2 + xm2 + (1 − x)M 2 dx m ln Λ2 + xm2 + (1 − x)M 2 −1 xm2 + (1 − x)M 2 × ln = ig2 16π 2 1 0 +(1 − x)p ln / Λ2 + xm2 + (1 − x)M 2 −1 xm2 + (1 − x)M 2 + ﬁnite terms as Λ → ∞ = ig2 16π 2 1 0 dx ((1 − x)p + m) ln / Λ2 xm2 + (1 − x)M 2 (15. Thus.32) where we have identiﬁed Q2 = x(1 − x)p2 + m2 while Q = −x(1 − E x)p2 + m2 represents the function rotated to Minkowski space. for the fermion self-energy in (15. a Feynman amplitude with n external lines is an analytic function of (n − 1) external momentum variables. m. Λ). In fact.15. These two simple calculations show that Feynman amplitudes are functions of external momenta.33) +ﬁnite terms as Λ → ∞ . g. for the scalar self-energy in (15. 2 (15. This is because the overall momentum conservation eliminates one of the momentum variables. In the examples worked out above it is clear that one can Taylor expand the amplitudes around zero momentum of the external lines. Similarly.3 Cut-off regularization 637 2 2 ig2 = lim Λ→∞ 4π 2 1 0 dx Λ − 3Q ln Λ2 + Q Q 2 2 + 2Q 2 = f (p. for example.29).

it is clear that such a regularization violates manifest Poincar´ invariance of the e quantum theory. a lattice regularization which regularizes a theory by deﬁning it on a discrete space-time lattice (we will not go into a detailed discussion of lattice regularization). it is useful to ﬁnd . a cut-oﬀ can lead to violation of gauge invariance in gauge theories by giving a mass to the gauge boson. Therefore. while quite useful. To avoid this problem. Furthermore.34) +ﬁnite terms as Λ → ∞ .638 15 Regularization of Feynman diagrams ig2 = lim Λ→∞ 4π 2 ig2 4π 2 1 0 1 0 dx Λ − 3Q ln 2 2 Λ2 + Q Q 2 2 + 2Q 2 = Λ→∞ lim dx Λ2 − 3m2 ln Λ2 + m2 m2 Λ2 + m2 + 2m2 m2 +3x(1 − x)p2 ln + ﬁnite terms as Λ → ∞ = ig2 4π 2 1 0 dx Λ2 − 3m2 ln Λ2 Λ2 + 3x(1 − x)p2 ln 2 m2 m (15. then expanding around zero external momenta would be disastrous. in massless theories we Taylor expand the amplitudes around a nonzero but ﬁnite value of the external momentum. leads to a lack of manifest rotational invariance (although in the continuum limit this symmetry is recovered). 15.4 Pauli-Villars regularization Although regularizing a Feynman diagram by cutting oﬀ contributions from large values of momentum seems natural. Let us note here that if the theory is massless. Similarly. Thus we see explicitly that any Feynman amplitude can be Taylor expanded so that the divergent parts can be separated out as local functions (independent of momenta). This is because of the infrared divergences of the theory which can be seen in the above examples by setting m = 0.

For example.35) ν = iGF . To discuss the Pauli-Villars regularization scheme.(15. 2 − m2 p r µ = −(2π)4 ieγ µ δ4 (p + q + r). we can now calculate various 1PI amplitudes in QED. let us consider QED (quantum electrodynamics) in the Feynman gauge (see (9.15.80) and (9.36) With these Feynman rules. 4 2 which leads to the Feynman rules µ p (15.125)) 1 1 /ψ L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 .µν (p) = − iηµν .4 Pauli-Villars regularization 639 a covariant regularization scheme which also respects gauge invariance. p2 p q p = iSF (p) = i(p + m) / . in this theory the fermion selfenergy at one loop takes the form (as usual we factor out the overall energy-momentum conserving delta function along with a factor of (2π)4 ) k p k+p p = (−ie)2 ηµν / / d4 k µ i(k + p + m) ν γ γ −i 2 4 2 − m2 (2π) (k + p) k − µ2 . The Pauli-Villars regularization provides such a regularization scheme.

µ) (15. 2 kE + Q2 (m.112) as well as (15.37) ie2 8π 2 dx((1 − x)pE + 2m) / where we have deﬁned Q2 (m.111) and (2. In the .9) for the angular integral.) The momentum integral is. µ) − 2 2 kE + Q2 (m. We have also used gamma matrix identities in (2. divergent (logarithmically by power counting) as is the case in the earlier calculation of the fermion self-energy with the Yukawa coupling (see (15.26).640 = − = − = − × e2 (2π)4 e2 (2π)4 ie2 (2π)4 d4 k 15 Regularization of Feynman diagrams −2(k + p) + 4m / / 2 − m2 ) (k 2 − µ2 ) ((k + p) 2(kE + pE ) + 4m / / 2 + m2 ) (k 2 + µ2 ) ((kE + pE ) E id4 kE dx d4 kE 2(kE + pE ) + 4m / / (kE + xpE ) + x(1 − x)p2 + xm2 + (1 − x)µ2 E ie2 (2π)4 ie2 16π 2 ie2 8π 2 ∞ 0 2 2 = − = − = − × = − dxd4 kE 2 dxdkE 2kE + 2(1 − x)pE + 4m / / 2 (kE + Q2 (m. µ))2 ∞ 0 2 dkE −1 .38) and have introduced a mass µ for the photon (to regulate infrared divergences) which can be taken to zero at the end of the calculations. E (15. µ))2 2 / kE (2(1 − x)pE + 4m) 2 (kE + Q2 (m. µ) = x(1 − x)p2 + xm2 + (1 − x)µ2 . µ) (kE + Q2 (m. of course.27)). (The Dirac gamma matrices have also been rotated to Euclidean space as discussed in (15. µ))2 dx ((1 − x)pE + 2m) / 2 dkE 1 Q2 (m.

(The covariant derivative in (15. The contribution from the heavy photon to the self-energy would be exactly the same as what we have already calculated in (15. Q2 (m. let us add to the Lagrangian density of QED (15. without doing any further calculation. Therefore.41) I (reg) = I + I ′ = − dx ((1 − x)pE + 2m) ln / . µ and ψ ′ . 2 kE + Q2 (m.37) coming from the diagram where a heavy photon is being exchanged. we deﬁne the theory with a minimal coupling of the photon to another fermion which we assume to be heavy as well as introduce a second massive photon which couples to the normal as well as the heavy fermions.37) except for a sign and µ → Λ. The idea is to take Λ → ∞ at the end and in that limit the heavy ﬁelds do not propagate and. µ) (15. Namely.4 Pauli-Villars regularization 641 Pauli-Villars regularization.39) where we assume that Λ ≫ m.15. Λ) .35) another gauge invariant term of the form 1 ′ ′µν Λ2 ′ ′µ ′ / Fµν F − Aµ A − iψ D (A + A′ )ψ ′ 4 2 +Λψ ψ ′ + eψA ′ ψ. However.39) is deﬁned with the gauge ﬁeld combination Aµ + A′ .) µ We note that the sign of the Lagrangian density for these ﬁctitious ﬁelds is opposite to that of the standard ﬁelds and hence they act as ghost ﬁelds and subtract out contributions. decouple.40) I′ = dx ((1 − x)pE + 2m) / so that the eﬀective regularized fermion self-energy can be written as ie2 8π 2 Q2 (m. / ′ L′ = (15. A′ are ﬁctitious heavy µ ﬁelds introduced to regularize diagrams. we can write the additional contribution to the fermion self-energy as ie2 8π 2 ∞ 0 2 dkE −1 . these additional interacting ﬁelds give an additional contribution to the fermion self-energy (15. Λ) (15. therefore.

We can also calculate the one loop photon self-energy in QED in a similar manner.642 15 Regularization of Feynman diagrams which is ﬁnite for any given value of Λ. µ) 2 2 I (reg) dx ((1 − x)p − 2m) ln / . (15. (15.36) lead to k p k+p i(k + m) i(k + p + m) / / / d4 k γ γ 4 µ k 2 − m2 ν (k + p)2 − m2 (2π) p = iΠµν (p) 2 = −(−ie)2 Tr =− =− kµ (k + p)ν + kν (k + p)µ − ηµν k · (k + p) + m2 ηµν 4e2 d4 k (2π)4 (k2 − m2 )((k + p)2 − m2 ) 4e2 (2π)4 d4 k dx × =− kµ (k + p)ν + kν (k + p)µ − ηµν (k · (k + p) − m2 ) ((k + xp)2 + x(1 − x)p2 − m2 )2 d4 k dx 4e2 (2π)4 × =− 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − (x(1 − x)p2 + m2 )) (k2 + Q (m))2 2 4e2 (2π)4 dx d4 k 2 × 2x(1 − x)(ηµν p2 − pµ pν ) + 2kµ kν − ηµν (k2 + Q (m)) (k2 + Q (m))2 2 . The ﬁnal result can now be rotated back to Minkowski space and has the form ie2 = 2 8π Q (m. Λ) Q (m. µ) = −x(1 − x)p2 + xm2 + (1 − x)µ2 .43) .42) where Q (m. The Feynman rules (15.

Π00 → (i) 00 0i 0i iΠE = − µν × 4ie2 (2π)4 d4 kE dx 2 2x(1 − x)(δµν p2 − pµE pν E ) + 2kµE kν E − δµν (kE + Q2 (m)) E . 2 (kE + Q2 (m))2 (15.4 Pauli-Villars regularization 643 where we have used the gamma matrix identities in (2. p2 ). p2 ) → −(kE . Π → (i)ΠE ) and we have δ00 . 2 (15. To evaluate the momentum integral.116) and have deﬁned Q (m) = x(1 − x)p2 − m2 . η00 → E 2 ΠE . For terms involving kµE kν E we can use 2 symmetric integration (kµE kν E → 1 δµν kE ) and in this way we have 4 d4 kE = π2 π2 δµν 2 π2 δµν 2 k2 2kµE kν E 2 (kE + Q2 (m))2 = d4 kE 1 δµν 2 1 2 2 δµν kE (k2 + Q2 (m))2 E dy y2 (y + Q2 (m))2 = dy (y + Q2 (m))2 − 2(y + Q2 (m))Q2 (m) + (Q2 (m))2 (y + Q2 (m))2 dy 1− 2Q2 (m) (Q2 (m))2 .45) where Q2 (m) = x(1 − x)p2 + m2 .43) to 2 Euclidean space ((k0 . We can now integrate each of the E terms in (15. (k2 .115) and (2. we can rotate (15.44) The overall negative sign in (15. p0E ).45) individually.43) reﬂects the fact that we are evaluating a fermion loop. p0 ) → i(k0E .15. + y + Q2 (m) (y + Q2 (m))2 = d4 kE E 1 (−δµν ) + Q2 (m) .

is quadratically divergent. .9) in the intermediate steps.644 = −π 2 δµν = −π 2 δµν so that we have 2kµE kν E dy dy 15 Regularization of Feynman diagrams y y + Q2 (m) 1− Q2 (m) . as we see. we obtain for the self-energy of the photon (in Euclidean space). This integral. Putting this back into (15.47) 1 + δµν − 2 dy + Q2 (m) .45).46) =− π2 δµν 2 dy 1− (y + Q2 (m))2 Here we have used (15. However. ie2 4π 2 ∞ 0 iΠE = − µν dx dy 2x(1 − x) δµν p2 − pµE pν E y E (y + Q2 (m))2 +δµν = − × ie2 4π 2 (Q2 (m))2 1 − + 2 2(y + Q2 (m))2 dx 0 ∞ dy 2x(1 − x) δµν p2 − pµE pν E E Q2 (m) 1 − y + Q2 (m) (y + Q2 (m))2 1 (Q2 (m))2 − + 2 2(y + Q2 (m))2 ∞ 0 +δµν = − ie2 dx 2x(1 − x)(δµν p2 − pµE pν E ) E 4π 2 ∞ 0 dy −1 y + Q2 (m) (15. (15. y + Q2 (m) d4 kE 2 (kE + Q2 (m)) 2 − k2 δµν 2 E + Q (m) (Q2 (m))2 . recalling that the photon has coupling to ﬁctitious heavy fermions.

Λ) µν µν = − ie2 4π 2 dx 2x(1 − x) δµν p2 − pµE pν E ln E 1 − δµν Q2 (Λ) − Q2 (m) 2 Q2 (Λ) Q2 (m) .48) This shows that the diagram is ﬁnite for any given value of Λ. for example.1) which would violate gauge invariance. Then the regularized photon amplitude with these two sets of heavy fermions will have the form iΠE (reg) (pE ) = iΠE (pE . section 14.4 Pauli-Villars regularization 645 we will have another contribution coming from the heavy fermion loop with exactly the same contribution except for a sign and m → Λ. Λ2 and charges c1 e. As a result. Let us assume that we have two sets of heavy ghost fermions with masses √ √ Λ1 . see. c2 e respectively. m) − iΠE (pE . Such a term can be cancelled by adding another set of heavy fermion ﬁelds. Λ2 ) µν µν µν µν = − × ie2 4π 2 ∞ 0 dx 2x(1 − x) δµν p2 − pµE pν E E dy c1 c2 1 − − y + Q2 (m) y + Q2 (Λ1 ) y + Q2 (Λ2 ) −(1 − c1 − c2 ) 1 − δµν (1 − c1 − c2 ) 2 ∞ 0 dy − Q2 (m) − c1 Q2 (Λ1 ) − c2 Q2 (Λ2 ) = − ie2 4π 2 dx 2x(1 − x) δµν p2 − pµE pν E E . the regularized photon self-energy will have the form iΠE (pE . this has the unpleasant feature that the photon self-energy has developed a mass term (term proportional to δµν . (15. However. m) − ic1 ΠE (pE .15. Λ1 ) − ic2 ΠE (pE .

2. the quadratic divergences as well as the logarithmic divergences in (15.43) does not involve the photon propagator. · · · . Pauli-Villars regularization involves introducing. in a gauge invariant manner. since the diagram for the photon self-energy in (15.51) which represents a gauge invariant photon self-energy (it is manifestly transverse). c1 Λ2 + c2 Λ2 = m2 . a set of heavy ﬁelds with masses Λi and √ charges ( ci e) with i = 1.646 × ∞ 0 15 Regularization of Feynman diagrams dy 1 c1 c2 − − 2 (m) 2 (Λ ) y+Q y+Q y + Q2 (Λ2 ) 1 −(1 − c1 − c2 ) 1 − δµν (1 − c1 − c2 ) 2 ∞ 0 dy −(1 − c1 − c2 )x(1 − x)p2 − (m2 − c1 Λ2 − c2 Λ2 ) .52) We note here that even though we are working in the Feynman gauge.50) then.49) would cancel leading to ie2 δµν p2 − pµE pν E E 2π 2 iΠE (reg) = − µν dx x(1 − x) × c1 ln Q2 (Λ1 ) + c2 ln Q2 (Λ2 ) − ln Q2 (m) . n depending on the nature of the divergence in the diagram such that the regularized amplitude .49) E 1 2 It is clear. In general. 1 2 (15.(15.(15. Rotating this back to Minkowski space we obtain ie2 ηµν p2 − pµ pν 2π 2 2 iΠ(reg) = − µν dx x(1 − x) 2 2 × c1 ln Q (Λ1 ) + c2 ln Q (Λ2 ) − ln Q (m) . the result (15.52) holds for any covariant gauge with an arbitrary value of the gauge ﬁxing parameter ξ. (15. that if the parameters satisfy c1 + c2 = 1. therefore.

Λi ). it is by construction a gauge invariant regularization scheme.53) with the conditions on the parameters given by ci = 1. Thus we see that a method of regularizing Feynman integrals can very well be to analytically continue the integral into n dimensions where it is well deﬁned (well behaved).5 Dimensional regularization 647 n I (reg) = I(p. i ··· . In non-Abelian theories. i i ci Λ2 = m2 . m) − ci Ii (p. (k2 − m2 ) ((k + p)2 − M 2 ) (15. Furthermore. however. in less than four dimensions it is ﬁnite. since gauge invariance is independent of the number of space time dimensions. For example. (Very roughly speaking the Pauli-Villars regularization is a gauge invariant regularization but not a gauge covariant regularization which is why it fails in nonAbelian theories.4 which cannot be regularized by the Pauli-Villars method.15. it is quite useful in studying non-Abelian gauge theories for which the Pauli-Villars regularization does not work.55) is logarithmically divergent in four dimensions. However.5 Dimensional regularization From the analysis of the divergence structure of amplitudes in the last two sections it is clear that the divergences are functions of the dimension of space time. 15. the integral d4 k . 15. This procedure is known as dimensional regularization.54) will be ﬁnite and gauge invariant. i=1 (15. It is worth pointing out here that the Pauli-Villars regularization is suﬃcient to regularize all the Feynman amplitudes in an Abelian gauge theory in a gauge invariant way. (15. In fact.) We should mention here that dimensional regularization has its own problems which we will discuss at the end of . there are graphs of the form shown in Fig.

2 2 4! L= (15. (15.4: One of the one loop graphs contributing to the gluon self-energy. Figure 15. 2 [M ] = 1.648 15 Regularization of Feynman diagrams this section.58) are given by .58) so that the coupling constant λ is rendered dimensionless. 2 2 4! L= (15. if we are looking at the φ4 theory described by the Lagrangian density M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ . The Feynman rules for the theory (15.57) Therefore.56) carries a dimension and introducing an arbitrary mass scale µ we can then write the Lagrangian density (15.56) for the theory in n dimensions as M 2 2 λµ4−n 4 1 ∂µ φ∂ µ φ − φ − φ . In dimensional regularization we analytically continue the theory to n dimensions. but it is worth pointing out here that manipulations with dimensional regularization are extremely simple which is why it is quite useful. For example.56) we can carry out the (canonical) dimensional analysis in n dimensions to determine n−2 . [φ] = [λ] = 4 − n. the coupling constant for the quartic interaction in (15.

E To evaluate this integral. let us note that the integrand is spherically symmetric (independent of angular coordinates) and. to calculate amplitudes in n dimensions. we can separate out the angular part of the integral . (15. we need to evaluate the basic integral dn k 1 n (k 2 + 2k · p − M 2 )α (2π) i (2π)n i(−1)α (2π)n i(−1)α (2π)n dn kE (−1)α 2 (kE + 2kE · pE + M 2 )α ((kE + pE )2 1 − p2 + M 2 )α E (15.61) I = = = = dn kE dn kE 1 . 2 (kE + Q2 )α where we have rotated to Euclidean space. p1 p2 (15. therefore.15. it follows that in n dimensions we have µ ηµ = n. since each vector index takes n values.60) Furthermore.5 Dimensional regularization 649 p = p2 i . shifted the variable of integration and have deﬁned Q2 = M 2 − p2 .59) We need to be careful about manipulating tensors in n dimensions. For example. − M2 p4 p3 = −(2π)4 iλµ4−n δn (p1 + p2 + p3 + p4 ).

We see that . Let us consider the basic n dimensional Gaussian integral whose value is given by dn kE e− 2 kE 2 n = (2π) 2 .62) The integral over the angles dΩ can be evaluated from the simple Gaussian integral as follows.63) in spherical coordinates as n 2 kE 2 (2π) 2 = = = = n dn kE e− dΩ 0 ∞ n−1 dkE kE e− 2 kE 2 dΩ 0 ∞ 2 1 2 2 n −1 − kE dkE (kE ) 2 e 2 2 dΩ 2 2 −1 0 n ∞ dy y 2 −1 e−y (15.64) n = 2 2 −1 Γ n 2 k2 dΩ. we can write dΩ = dθ1 sinn−2 θ1 dθ2 sinn−3 θ2 .65) Let us compare this with known results in lower dimensions that we are familiar with. (15. This determines the value of the angular integral to be dΩ = 2π 2 (2π) 2 = . . where.650 15 Regularization of Feynman diagrams n−1 dn kE = kE dkE dΩ. E where we have deﬁned y = 2 in the intermediate step and have used the deﬁnition of the gamma function. in n dimensions. we can evaluate the integral in (15. (15.63) On the other hand. n ≥ 2. . dθn−1 . n −1 n Γ n 22 Γ 2 2 n n (15.

68) Therefore. we see that with the identiﬁcation n . 2 n .67) Let us recall that the beta function is deﬁned as Γ(p)Γ(q) =2 Γ(p + q) ∞ 0 −p−q B(p. dθ1 sin2 θ1 dθ2 sin θ2 dθ3 = 2π 2 .66) which agree with what we already know from explicitly doing these integrals. Thus.69) the integral (15. for the basic integral (15. dθ1 sin θ1 dθ2 = 4π. q) = dt t2p−1 1 + t2 .15.61) we have 1 dn k n (k 2 + 2k · p − M 2 )α (2π) dΩ 0 n 2 I = = = = i(−1)α (2π)n ∞ n−1 dkE kE (k2 E 1 + Q2 )α i(−1)α 2π (2π)n Γ n 2 i(−1)α 2π (2π)n Γ n 2 n ∞ 0 ∞ 0 n−1 dkE kE (k2 kE Q ∞ 0 1 2 α E + Q ) n−1 n 2 d kE Q Qn 1 2 2α kE Q 1 + Q2 α 1 i(−1)α π 2 = n × 2 n Γ n 2 )α− 2 (2π) 2 (Q dt tn−1 (1 + t2 )−α . (15.67) can be written as .5 Dimensional regularization 651 n=2: n=3: n=4: dΩ = dΩ = dΩ = dθ = 2π. (15. 2 p= q =α−p=α− (15. (15.

any other formula can be obtained from (15. we see that this gives us our basic integral as dn k 1 n (k 2 + 2k · p − M 2 )α (2π) n I = = iπ 2 (−1)α Γ α − n 2 n . n Γ α − (2π)n Γ(α) (p2 + M 2 )α− 2 2 n (15. using (15.71) by diﬀerentiation.71) This basic integral generates all other integrals that we need for evaluating amplitudes in n dimensions and.652 15 Regularization of Feynman diagrams I = = Γ i(−1)α π 2 1 n n Γ n 2 )α− 2 (2π) 2 (Q iπ 2 (−1)α Γ α − n 2 n . in fact.71) we note that we can write (this result can also be obtained by shifting the variable of integration) kµ dn k n (k 2 + 2k · p − M 2 )α (2π) Iµ = = − = − 1 ∂ 2(α − 1) ∂pµ ∂ 1 2(α − 1) ∂pµ n 1 dn k (2π)n (k2 + 2k · p − M 2 )α−1 n iπ 2 (−1)α−1 Γ α − 1 − n 2 n (2π)n Γ(α − 1) (p2 + M 2 )α−1− 2 −2 α − 1 − n pµ (−1)α−1 iπ 2 2 = − n (2π)n 2(α − 1)Γ(α − 1) (p2 + M 2 )α− 2 n × Γ α−1− 2 = n iπ 2 (−1)α−1 pµ .70) Rotating back to Minkowski space. (2π)n Γ (α) (Q2 )α− 2 n n n 2 Γ α− Γ(α) n 2 (15. (2π)n Γ(α) (p2 + M 2 )α− 2 (15.72) Similarly we can obtain . For example.

First of all.74) becomes . With this. we are analytically continuing away from four dimensions (to a lower dimension where the integral is well deﬁned). i.8)) k p iλµ4−n 2 p i dn k n k2 − M 2 (2π) = − = λµ4−n iπ 2 (−1) Γ 1 − n 2 n 2 (2π)n Γ(1) (M 2 )1− 2 iλµ4−n π 2 Γ 1 − n 2 n . take the limit n → 4 which translates to ǫ → 0. the one loop scalar self-energy takes the form (see also (15. of course. the amplitude (15.e. 2 (2π)n (M 2 )1− 2 n n = − (15.73) n kµ kν dn k n (k 2 + 2k · p − M 2 )α (2π) and so on. (15.74) Let us next set n = 4 − ǫ.5 Dimensional regularization 653 Iµν = 1 iπ 2 (−1)α = n n Γ(α) (2π) (p2 + M 2 )α− 2 × pµ pν Γ α − 1 n n − ηµν p2 + M 2 Γ α − 1 − 2 2 2 .58). let us now calculate various amplitudes in the φ4 theory (15.15. With these basic integration formulae.. At the end of our calculations we should.

654 15 Regularization of Feynman diagrams k p p n ǫ iλµǫ π 2 Γ 1 − 2 + 2 = − ǫ 2 (2π)n (M 2 )1−2+ 2 ǫ iλµǫ π 2 Γ −1 + 2 = − ǫ . ǫ (15. we have n 2 n Γ 2− 2 Γ 3− ǫ ǫ ≃ 1 − γ. 16π 2 2 (15.76) as well as (15. Similarly. In the limit ǫ → 0. 2 2 2 ǫ ǫ = Γ = Γ 1+ 2 ǫ 2 ǫ 2 2 1 − γ) = − γ. the scalar self-energy (15.76) where γ denotes the Euler’s constant deﬁned in (15. we have π2 (2π)n n = ≃ 1 1 n = ǫ (4π) 2 (4π)2− 2 ǫ 1 1 + ln 4π .77) Using (15. ≃ ǫ 2 ǫ = Γ 1+ ǫ Γ 2 ǫ = Γ −1 + = 2 −1 + ǫ 2 n Γ 1− 2 ≃ − 1+ ǫ 2 2 −γ ǫ 2 ≃ − + (γ − 1). 2 (2π)n (M 2 )−1+ 2 n (15.12).77).75) Let us next work out some of the identities involving the gamma functions that will be useful to us.75) takes the form .

122).3 and. 15.5 Dimensional regularization 655 k p p n iλµǫ π 2 ǫ = − M 2 M −ǫ Γ −1 + 2 (2π)n 2 iλ π 2 ǫ = − M 2 Γ −1 + n 2 (2π) 2 ≃ − = − ≃ − = − iλ M 2 ǫ 1 + ln 4π 2 2 16π 2 iλ M 2 2 16π 2 iλM 2 32π 2 iλM 2 32π 2 n M2 µ2 ǫ −2 2 − + (γ − 1) ǫ 1− 1− ǫ M2 ln 2 2 µ 2 − + (γ − 1) − ln 4π ǫ ǫ M2 ln 2 2 µ M2 2 − + ln 2 + (γ − 1) − ln 4π ǫ µ M2 2 + (γ − 1) . we can calculate the one loop correction to the vertex function shown in Fig. we will set all the external momenta to vanish as we had done earlier in (15. In this case.78) We note here that for M 2 = 0. the amplitude in n dimensions takes the form = = 3 (−iλµǫ )2 2 3λ2 µ2ǫ 2 dn k (2π)n dn k 1 n 2 − M 2 )2 (2π) (k i 2 − M2 k 2 . this graph would be regularized to zero in dimensional regularization which is the type of argument used in (13. Similarly.19). − + ln ǫ 4πµ2 (15.15. for simplicity.

81) As a result. 2 2 2 ǫ .77).656 = 15 Regularization of Feynman diagrams i (−1)2 Γ 2 − n 3λ2 µ2ǫ 2 n n 2 Γ(2) (M 2 )2− 2 2 (4π) ǫ 3iλ2 µǫ 1 1 + ln 4π 2 2 16π 2 3iλ2 µǫ 32π 2 3iλ2 µǫ 32π 2 3iλ2 µǫ 32π 2 2 − γ + ln 4π ǫ 2 −γ ǫ 1− M2 µ2 ǫ −2 ≃ ≃ ≃ = ǫ M2 ln 2 2 µ 2 M2 − ln 2 − γ + ln 4π ǫ µ 2 M2 − ln −γ . ǫ (15. the amplitudes diverge. Let us next calculate one loop ampli- tudes in QED described by the Lagrangian density in the Feynman gauge (see (15. we let e → e µ2 .82) . as we take the limit ǫ → 0 (to go to four dimensions). the canonical dimensions of various ﬁelds can be easily determined to be (we are assuming n = 4 − ǫ) [Aµ ] = [ψ] = [e] = ǫ n−2 =1− . 2 2 n−1 3 ǫ ψ = = − .76) and (15.5.79) where we have used (15. to make the coupling constant dimensionless.1 Calculations in QED. We note that these amplitudes are well behaved (regularized) for any ﬁnite value of ǫ. /ψ 4 2 (15. 2 (15. 15. However.35)) 1 1 L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 . ǫ 4πµ2 (15.80) In n dimensions.

84) ǫ With these Feynman rules. n k 2 ((k + p)2 − m2 ) (2π) . p2 p = iSF (p) = i(p + m) / . the Feynman rules of the theory in n = 4 − ǫ dimensions are given by µ p ν = iGF .15.37) and the discussion following that equation) k p k+p ǫ p / / iηµν dn k µ i ((k + p) + m) ν γ γ − 2 (2π)n (k + p)2 − m2 k (15.µν (p) = − iηµν . 2 − m2 p r µ q p = −(2π)4 ieµ 2 γ µ δ4 (p + q + r). ǫ (15. we can calculate the fermion selfenergy at one loop which gives (see also (15. (15.85) = −ieµ 2 2 = −e2 µǫ / / dn k γ µ ((k + p) + m) γµ .5 Dimensional regularization 657 where µ is an arbitrary mass scale so that the covariant derivative in the theory is understood to have the form Dµ ψ = ∂µ + ieµ 2 Aµ ψ.83) As a result.

the one loop fermion self energy in (15.86) Using this.85) takes the form = −e2 µǫ = −e2 µǫ = −e2 µǫ = −e2 µǫ = −e2 µǫ ≃ ie2 8π 2 dn k (2 − n)(k + p) + nm / / n k 2 ((k + p)2 − m2 ) (2π) dx dx dx (15.658 15 Regularization of Feynman diagrams We note here that if we use the algebra of the gamma matrices (1. γµ kγ µ = (2 − n)k.87) (2 − n)(k + p) + nm / / dn k n 2 + x(1 − x)p2 − xm2 )2 (2π) ((k + xp) / / dn k (2 − n)(k + (1 − x)p) + nm 2 − Q2 ) 2 (2π)n (k dn k (2 − n)(1 − x)p + nm / 2 − Q2 )2 (2π)n (k (15. ǫ 4πµ2 (15.79) in n dimensions as well as (15. / / (15.60) we obtain γµ γ µ = n.89) .88) dx ((2 − n)(1 − x)p + nm) / dx ((1 − x)p − 2m) 1 − / × 2 −γ ǫ 1− ǫ 2 i (−1)2 Γ 2 − n 2 n n (4π) 2 Γ(2) (Q2 )2− 2 1+ ǫ ln 4π 2 ǫ Q2 ln 2 µ2 ǫ 2 2 − γ + ln 4π ǫ ≃ ie2 8π 2 dx ((1 − x)p − 2m) 1 − / × 1− ǫ Q2 ln 2 µ2 ≃ = ie2 8π 2 ie2 8π 2 × dx ((1 − x)p − 2m) / dx ((1 − x)p − 2m) / Q2 2 − ln 2 − γ − 1 + ln 4π ǫ µ xm2 − x(1 − x)p2 2 − ln −γ−1 .

Tr A = 0. /B / Tr γµ γν γλ γρ = n (ηµν ηλρ − ηµλ ηνρ + ηµρ ηνλ ) .115) and (2.42). Similarly the photon self energy at one loop can also be calculated to have the form (the overall negative sign is because of the fermion loop) k p k+p ǫ p = iΠµν (p) = − −ieµ 2 = −e2 µǫ 2 i(k + m) i(k + p + m) / / / dn k Tr γµ 2 γ n 2 ν (k + p)2 − m2 (2π) k −m (15.76) as well as (15.90) becomes (we factor out (−e2 µǫ ) for simplicity and will restore this later) dn k n(kµ (k + p)ν − ηµν k · (k + p) + kν (k + p)µ + m2 ηµν ) (2π)n (k2 − m2 )((k + p)2 − m2 ) dx dx dn k kµ (k + p)ν + kν (k + p)µ − ηµν (k · (k + p) − m2 ) (2π)n ((k + xp)2 + x(1 − x)p2 − m2 )2 dn k (2π)n = =n =n . This can be compared with (15.15. /B Tr A /C = 0.77). / Tr A / = nA · B.116)). (15.5 Dimensional regularization 659 where we have deﬁned Q2 = −x(1−x)p2 +xm2 and have used (15. n (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Using the n-dimensional identities (see (2.91) the photon self-energy (15.90) / / / dn k Tr γµ (k + m)γν (k + p + m) . Tr ½ = n.

660 × =n × 15 Regularization of Feynman diagrams 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − x(1 − x)p2 − m2 ) (k2 − Q2 )2 dx dn k (2π)n ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν + 2kµ kν − ηµν k2 . Let us next look at the terms with kµ kν in the numerator separately which leads to .76) as well as (15. (15.73).71)(15. Let us look at the terms without any k in the numerator and this takes the form (we now put back the factor (−e2 µǫ )) = −ne2 µǫ = −ne2 µǫ × ≃− dn k ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν (2π)n (k2 − Q2 )2 dx dx ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν (−1)2 Γ 2 − n i 2 n n (4π) 2 Γ(2) (Q2 )2− 2 dx ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν ǫ 4 1+ ǫ ln 4π 2 2 −γ ǫ 1− ǫ Q2 ln 2 µ2 ie2 4π 2 × 1− ≃− ie2 4π 2 dx ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν 1− ǫ Q2 ln 2 µ2 × ≃− 1 2 − γ − + ln 4π ǫ 2 dx ie2 4π 2 Q2 1 2 − ln −γ− ǫ 4πµ2 2 . The integral in (15.92) where we have deﬁned Q2 = −x(1 − x)p2 + m2 . (k2 − Q2 )2 (15.77).92) contains three diﬀerent kinds of terms which can be integrated using the standard formulae in (15.93) × ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν where we have used (15.

94) Q2 1 2 +γ− × − + ln 2 ǫ 4πµ 2 Similarly.5 Dimensional regularization 661 = −ne2 µǫ = −2ne2 µǫ ≃ ie2 ηµν 4π 2 dx 1 i (−1)2−1 Γ 1 − n 2 dx ηµν n n 2 (4π) 2 Γ(2) (Q2 )1− 2 dx Q2 1 − ǫ 4 1+ 1− ǫ ln 4π 2 2kµ kν dn k n 2 − Q2 ) 2 (2π) (k 2 × − + (γ − 1) ǫ ≃ ≃ = ie2 ηµν 4π 2 ie2 ηµν 4π 2 ie2 ηµν 4π 2 ǫ Q2 ln 2 µ2 1− ǫ Q2 ln 2 µ2 1 2 dx Q2 − + γ − − ln 4π ǫ 2 Q2 1 2 dx Q2 − + ln 2 + γ − − ln 4π ǫ µ 2 dx (−x(1 − x)p2 + m2 ) .15. the term with k2 in the numerator leads to = −ne2 µǫ = ne2 µǫ ηµν ≃− ie2 ηµν 4π 2 −ηµν k2 dn k (2π)n (k2 − Q2 )2 dx dx (−1)2−1 Γ 1 − n i n 2 n n 2 )1− 2 2 (4π) 2 Γ(2) (Q ǫ 4 2− ǫ 2 1+ ǫ ln 4π 2 dx Q2 1 − 2 × − +γ −1 ǫ ≃− ie2 ηµν 4π 2 1− ǫ Q2 ln 2 µ2 dx Q2 (2 + ǫ (−1 + ln 4π)) Q2 2 × − + ln 2 + γ − 1 ǫ µ . (15.

p q=0 µ k+p p =(−ieµ 2 )3 ǫ k+p k iηλρ / / / / dn k λ i(k + p + m) µ i(k + p + m)γ ρ γ γ − 2 n 2 − m2 2 − m2 (2π) (k + p) (k + p) k . For simplicity we will put the momentum of the external photon equal to zero.95).96) =− dxx(1 − x) This shows that the photon self energy graph is completely transverse which is consistent with gauge invariance (see also (15. the photon self energy in (15. Let us next calculate the one loop amplitude corresponding to vertex correction (charge renormalization). × − + 2 ln ǫ 4πµ2 Adding the contributions from (15. the gauge ﬁxing Lagrangian density does not receive any quantum correction which can also be seen from the BRST identities for QED.52)).93)-(15.95) Q2 4 + 2γ .92) takes the ﬁnal form ie2 (ηµν p2 − pµ pν ) 2π 2 2 Q2 1 − ln −γ− . Since there is no longitudinal term in the photon self-energy. The photon does not acquire a mass term and we note here that even though we have chosen to work with the Feynman gauge. this result holds in any covariant gauge with the gauge ﬁxing term ξ arbitrary. 2 ǫ 4πµ 2 (15.662 ≃− =− ie2 ηµν 4π 2 ie2 ηµν 4π 2 15 Regularization of Feynman diagrams Q2 4 dx Q2 − + 2 − 2 ln 4π + 2 ln 2 + 2γ − 2 ǫ µ dx −x(1 − x)p2 + m2 (15. since the photon self-energy does not involve the photon propagator.

97) = −e3 µ 2 3ǫ / / / / dn k γ λ (k + p + m)γ µ (k + p + m)γλ .113)) γ λ A µ A λ = (2 − n)(2Aµ A − γ µ A2 ). 2 ((k + p)2 − m2 )2 k (15. we can write the amplitude (15.98) γ λ (k + p + m)γ µ (k + p + m)γλ / / / / = γ λ (k + p)γ µ (k + p)γλ + mγ λ (γ µ (k + p) + (k + p)γ µ )γλ / / / / / / / / +m2 γ λ γ µ γλ = (2 − n)(2(k + p)(k + p)µ − (k + p)2 γ µ ) + 2nm(k + p)µ / / +(2 − n)m2 γ µ = −(2 − n)γ µ ((k + p)2 − m2 ) +2((2 − n)(k + p) + nm)(k + p)µ . n (2π) k2 ((k + p)2 − m2 )2 Using the n-dimensional identities (see also (2.99) = −e3 µ 2 × 3ǫ −(2 − n)γ µ ((k + p)2 − m2 ) + 2((2 − n)(k + p) + nm)(k + p)µ / / 3ǫ = −e3 µ 2 −(2 − n)γ µ dn k (2π)n k2 ((k + p)2 − m2 ) + 2 (2 − n)(k + p) + nm (k + p)µ / / .5 Dimensional regularization 663 (15.97) as dn k (2π)n k2 ((k + p)2 − m2 )2 (15.15. /γ /γ / as well as (15.100) Let us next use (15.86) we have (15. / / Therefore.23) as well as the identity .

100) which leads to 1 + p)2 − m2 ) 1 (k + xp)2 − Q2 2x (k + xp)2 − Q2 k2 ((k = = dx dx 2.101) to combine the denominators in (15. 1 k2 ((k + p)2 − m2 )2 3.664 15 Regularization of Feynman diagrams 1 AB 2 = − = ∂ 1 ∂ =− ∂B AB ∂B dx dx 2x . (15.102) where we have deﬁned Q2 = −x(1 − x)p2 + xm2 .100) and shifting the variable of integration. Using this in (15. we obtain for the vertex correction p q=0 µ k+p p = −e3 µ 2 × 3ǫ k+p k dx dn k −(2 − n)γ µ (2π)n (k2 − Q2 )2 / / dn k 4x (2 − n)kkµ + (1 − x) (2 − n)(1 − x)p + nm pµ 3 (2π)n (k2 − Q2 ) 3ǫ = −e3 µ 2 dx − (2 − n)γ µ i (−1)2 Γ 2 − n 2 n n (4π) 2 Γ(2) (Q2 )2− 2 . ((1 − x)A + xB)3 1 ((1 − x)A + xB)2 (15.

15.103) ie3 µ 2 16π 2 2 ln 4π − γ (−2 + ) × 1− + =− x(1 − x)(−2(1 − x)p + 4m)pµ / −x(1 − x)p2 + xm2 dx(1 − x) γ µ 2 − ln ie3 µ 2 8π 2 + Q2 −γ−1 4πµ2 (15.104) respectively.5 Dimensional regularization 665 − 1 2 Γ 2− n 2 n 2 )2− 2 (Q +4x(2 − n)γ µ (−1)3 i n (4π) 2 Γ(3) +4x(1 − x) (2 − n)(1 − x)p + nm pµ / = −ie3 µ 2 − = 3 i (−1)3 Γ 3 − n 2 n n (4π) 2 Γ(3) (Q2 )3− 2 dx Γ 2− (4π) n 2 n 2 − (2 − n)(1 − x)γ µ n (Q2 )2− 2 n (4 − n)x(1 − x) (2 − n)(1 − x)p + nm pµ / (Q2 )3− 2 dx (1 − x)γ µ 1 + 2 ln Q2 µ2 2 −γ 2 (15.87).104) 2 (1 − x)p − 2m pµ / .100). (15.109) and recall that the coupling constant in the present case is eµ 2 ) k ∂ ∂pµ p k+p p =− 1 eµ 2 k .103) and (15.89) with (15.105) . p k+p k+p p q=0 (15. −(1 − x)p2 + m2 It is worth noting from this derivation that at every step if we compare (15. we see that the fermion self-energy and the vertex function are related as (see also (9.88) and (15. (15.

1439 (1948). Physical Review 74. the higher derivative method. (15. we know that i 4! γ5 = iγ 0 γ 1 γ 2 γ 3 = − µνλρ γ µ ν λ ρ γ γ γ . However.666 15 Regularization of Feynman diagrams which we recognize as the Ward-Takahashi identity. We will discuss this further in section 16. . we will not go into the details of these other methods here. (Note that the divergent parts as well as the ﬁnite parts satisfy the identity which demonstrates that dimensional regularization preserves gauge invariance. 15. Analytic continuation of these to other dimensions is nontrivial. The lattice regularization is also quite useful.107) There are several ways to address this issue with the γ5 matrix. there exist several other important regularization schemes. (15. We simply note here that we select a particular regularization scheme depending on the theory (or the problem) that we are analyzing. Schwinger.106) is deﬁned only in four dimensions (both γ5 and the Levi-Civita tensor µνλρ are manifestly four dimensional quantities). the point splitting method.) The power of dimensional regularization is quite obvious from these calculations. In addition to the regularization methods that we have discussed in this chapter. Hence if the theory or the physical quantity of interest involves such objects. the ζ-function regularization etc. for example. It is not naively applicable if the calculation involves quantities that typically exist only in four dimensions. J. However. it has its own drawbacks. Such quantities often occur in physics like in the chiral anomaly which is related to the life time of the π meson decaying through π 0 → 2γ.6 where we discuss chiral anomaly. It is gauge invariant and extremely simple to manipulate with.6 References 1. naive dimensional regularization leads to incorrect answers. For example.

R. W. C. Reviews of Modern Physics 21.6 References 667 2. Introduction to the theory of Quantized Fields. Diagrammar. 434 (1949). Schwinger. . 8. Shirkov. 20 (1972). 189 (1972). 4. Giambiaggi. Nauka. J. G. Veltman. N. ’t Hooft and M. Physical Review 76. ’t Hooft and M. V. 6. 5. G. Nuovo Cimento 12B. Villars. J. N. 664 (1951). 7. Pauli and F. Physical Review 82. G. 769 (1949). Bollini and J. Nuclear Physics B44. 3. Bogoliubov and D. CERN preprint (1973). Feynman. Moscow (1984).15. Veltman.

This page intentionally left blank .

the action for the theory (16. The next step in understanding the process of renormalization is to devise a method for determining which Feynman diagrams in a theory will be divergent as well as the nature of the divergence (without actually evaluating the integrals) and this is achieved through the notion of the superﬁcial degree of divergence of a graph.Chapter 16 Renormalization theory 16. the Lagrangian density must have canonical dimension 4 in these units (since we are in four space time dimensions.1) i where L0 denotes the sum of the free Lagrangian densities for all the ﬁeld variables in the theory and each Li representing some interaction is a monomial in the basic ﬁeld variables and derivatives. In our units.1 Superﬁcial degree of divergence In the last chapter. (16.2) . we have seen that loop diagrams in various theories become divergent and need to be regularized. it is easy to check that [M ] = [L]−1 = 1. In units of = c = 1 which we have been using. To deﬁne this let us start with the Lagrangian density for a given theory L = L0 + Li . we have also seen that since Feynman amplitudes are analytic functions of external momenta.1) is dimensionless. in n dimensions it should be n). As a result. Furthermore. 669 (16. we can expand them around some reference momentum value (in massive theories conventionally chosen to be zero) so that the regularized divergent parts can be isolated as local terms.

(The mass term also has the correct dimension with this assignment.5) = 4.6) Namely. for example. and L0 we have [φ] = 1. 2 2 L0 = that since (φ) (16. (φ) (16.3) [∂µ ] = [L]−1 = 1. bosons and derivatives at an interaction vertex following from the interaction Lagrangian density Li . / 3 . for the Yukawa interaction LY = gψψφ. 2 (16.7) leads to [ψ] = [ψ] = (16. the canonical dimension of a scalar ﬁeld (in our units and in four dimensions) is 1.) Similarly the free fermion Lagrangian density (ψ) L0 = iψ∂ ψ − mψψ. (16.670 16 Renormalization theory Let us note from the form of the free Lagrangian density for a real scalar ﬁeld M2 2 1 ∂µ φ∂ µ φ − φ . we can show that boson ﬁelds in general have canonical dimension 1 as long as we ignore gravitation (even gravitation can be included in this category depending on what we consider as the basic ﬁeld variable) while that of the fermion ﬁelds is 3 in four space 2 time dimensions. (16. bi and di to denote respectively the number of fermions.9) .8) In fact.4) (16. Let us further introduce the notations fi . Thus.

On the other hand.1 Superficial degree of divergence 671 we have f = 2. d = 1. 4! (16. we can determine easily the superﬁcial degree of divergence of the following Feynman graphs as . b = 4.11) f = 0. d = 0. (16.14) and so on. (16. For example. d = 0.16. we have or.13) f = 2. This is deﬁned to be the diﬀerence between the number of momenta in the numerator arising from the loop integrations as well as derivative couplings and the number of momenta in the denominator arising from the propagators. b = 1. the φ4 interaction (16. With these notations we are now ready to introduce the notion of the superﬁcial degree of divergence of a Feynman graph. b = 1. κψσ µν ψFµν . for an interaction of the form (16.10) LI = leads to λ 4 φ .12) LI = hψγµ ψ∂ µ φ. Similarly.

D = 4 − 2 − 2 × 1 = 0. Similarly. diagrams with D = 1 or 2 are known . D = 8 − 3 × 2 = 2. D = 4 − 2 × 1 = 2. we say that the diagram is superﬁcially logarithmically divergent. D = 4 − 2 − 1 = 1. D = 8 − 5 × 2 = −2. (16.15) If D = 0 for a diagram. D = 4 − 2 × 2 = 0.672 16 Renormalization theory D = 4 − 2 = 2.

number of external fermion lines. as we will see it is the notion of superﬁcial degree of divergence which is useful in the study of renormalization theory. Similarly. (16. then the Feynman diagram is said to be superﬁcially convergent. The meaning of the adjective “superﬁcial” becomes clear once we look at the last graph in (16. number of internal boson lines. let us develop a general formula for the superﬁcial degree of divergence of any connected Feynman graph. it is actually divergent since it contains a subgraph which is divergent.18) The superﬁcial degree of divergence is easily seen to be given by .1 Superficial degree of divergence 673 respectively to have superﬁcial linear or quadratic divergences. If D < 0 for a graph.17) The factor 2IB reﬂects the fact that it takes two boson lines at two diﬀerent vertices to form a propagator (internal boson line). However. we have F + 2IF = i ni fi .15). (16. number of vertices of the ith type from Li . Let us consider a Feynman diagram with B = IB = F IF = = number of external bosons lines. Rather than calculating the superﬁcial degree of divergence for each graph.16. number of internal fermion lines. Since a vertex of Li has bi boson lines attached to it and since each of these lines can become either an external boson line or an internal boson line we must have B + 2IB = i n i bi . (16. Although this graph is superﬁcially convergent.16) ni = There exist topological relations between these numbers.

674

16 Renormalization theory

D=

i

ni di + 2IB + 3IF − 4

ni + 4.

i

(16.19)

The ﬁrst term in (16.19) simply says that each derivative at a vertex gives rise to a momentum in the numerator and if there are ni vertices of the ith type in a graph, this would lead to a power of momentum in the numerator ni di (which must be summed over all possible types of vertices). With each internal boson line is associated a momentum integration and a propagator. Thus each internal boson line eﬀectively leads to two powers of momentum in the numerator. Similarly each internal fermion line adds three powers of the momentum to the numerator. At each vertex, however, energy-momentum has to be conserved and since a four dimensional delta function (expressing conservation of energy-momentum) has dimension −4, each vertex takes away four powers of momentum except for an overall delta function which is necessary for the overall energy-momentum conservation. The last two terms reﬂect this. Using (16.17) and (16.18) and eliminating IB and IF from (16.19) we obtain 3 2 3 ni fi − F − 4 2

D =

i

ni di +

i

n i bi − B +

ni + 4

i

i

3 = 4−B− F + 2 3 = 4−B− F + 2 where we have deﬁned 3 δi = di + bi + fi − 4. 2

i

3 ni di + bi + fi − 4 2 ni δi , (16.20)

i

(16.21)

This is known as the index of divergence of the interaction Lagrangian density Li and can also be expressed as δi = dim Li − 4, (16.22)

16.1 Superficial degree of divergence

675

where dim Li is calculated only from the dimensions of the ﬁeld variables and derivatives and not from any dimensionful parameters. In all theories we consider, the interaction Lagrangian density has dimension 4 and hence

δi = 0. In such cases (16.20) reduces to 3 D = 4 − B − F. 2

(16.23)

(16.24)

Namely, in such cases the superﬁcial degree of divergence is completely determined by the number of external lines in the graph. Let us note here that the superﬁcial degree of divergence is a function of the number of space-time dimensions (we are working in four dimensions). We can check the relation (16.24) against our explicit power counting calculations in (16.15),

D = 4 − 2 − 0 = 2,

D = 4 − 4 − 0 = 0,

D = 4−0−

3 × 2 = 1, 2

D = 4 − 2 − 0 = 2,

676

16 Renormalization theory

D = 4 − 2 − 0 = 2,

D = 4−1−

3 × 2 = 0, 2

D = 4 − 6 − 0 = −2,

(16.25)

and they agree completely. From the form of the formula for the superﬁcial degree of divergence, it is clear that only a few graphs in a theory would have non-negative superﬁcial degree of divergence. Thus for example, if we are considering the φ4 theory, then only the following 1PI functions would be superﬁcially divergent. B 0 1 2 3 4 D =4−B 4 3 2 1 0

The zero point function contributes only to the zero point energy which can be eliminated by normal ordering the theory. The one point function is in principle divergent. However, such graphs do not exist in the φ4 theory since the theory has the discrete symmetry

φ → −φ,

(16.26)

16.1 Superficial degree of divergence

677

and the one point function violates this symmetry. Similarly, although the three point function can be superﬁcially divergent it does not exist because of this discrete symmetry. Thus, only two graphs are superﬁcially divergent, namely the 2-point and the 4-point functions. In the last chapter we have explicitly calculated the two point as well as the four point functions at one loop in the φ4 theory where we have seen that these graphs are indeed divergent. We have to develop a systematic way of making these graphs ﬁnite. Any higher point 1PI graph can, of course, contain these graphs as subgraphs and even though these graphs are superﬁcially convergent, they will in fact be divergent. However, corresponding to each such graph we can deﬁne a skeleton graph. Thus for example, in the φ4 theory, the graph for the six point function in Fig. 16.1

Figure 16.1: A superﬁcially convergent graph with a divergent subgraph. has the skeleton graph shown in Fig. 16.2.

Figure 16.2: The skeleton graph associated with Fig. 16.1. Namely, the skeleton graph of a superﬁcially convergent graph is a graph where no divergent subgraph can be found (divergent subgraphs shrunk to a point). The skeleton graph of a superﬁcially convergent graph is, therefore, by deﬁnition convergent. The full graph can be obtained from the skeleton graph by making insertion

678

16 Renormalization theory

of the two point and the four point functions at appropriate places. However, if we have a method of making these graphs, namely, Γ(2) , and Γ(4) , ﬁnite and regularization independent through some renormalization procedure, then the same procedure would make all the n point functions ﬁnite and regularization independent. Let us next look at QED described by the Lagrangian density

1 1 (∂µ Aµ )2 , /ψ L = − Fµν F µν + iψD − mψψ − 4 2ξ Dµ ψ = (∂µ + ieAµ ) ψ. (16.27)

The Feynman rules for the theory (in the Feynman gauge with ξ = 1) are

µ

p

ν = iGF ,µν (p) = −

iηµν , p2

p

= iSF (p) =

i(p + m) / , 2 − m2 p

r

µ

q p = −(2π)4 ieγ µ δ4 (p + q + r). (16.28)

Let us now analyze the divergence structure of graphs in this theory.

16.2 A brief history of renormalization

679

B 0 1 2 3 4 0 0 1

F 0 0 0 0 0 1 2 2

D = 4 − B − 3F 2 4 3 2 1 0 1 0

5 2

The zero point graphs are neglected because they can be taken care of by normal ordering. The 1 point boson graphs do not exist because they violate Lorentz invariance as well as gauge invariance. Similarly the 3 point boson graphs also do not exist. (Vanishing of photon graphs with an odd number of photons is a consequence of the symmetry of charge conjugation C, also known as the Furry’s theorem (see (11.188)).) The four photon graph, in this analysis, would appear to be superﬁcially divergent, but it is actually ﬁnite because of gauge invariance. In fact, in gauge theories the actual degree of divergence may be softer than the naive power counting because of constraints coming from gauge invariance. Fermion graphs with only an even number of fermion lines can be nonzero because of Lorentz invariance (as well as fermion number conservation). Therefore, there are only three possible superﬁcially divergent graphs, namely, the fermion and the photon self-energy as well as the fermion interaction vertex with the photon, and if we can somehow make these ﬁnite in a regularization independent manner, the theory will be well deﬁned. 16.2 A brief history of renormalization Dyson laid the foundations for the systematic study of renormalization in two papers in 1949 where he studied the renormalization of QED. He basically used the Schwinger-Dyson equations (to be discussed in the next section) and showed that most of the divergences can be absorbed into a redeﬁnition of parameters of the theory. The class of graphs which he could not incorporate into his study are known as overlapping divergent graphs of the type shown in Fig. 16.3. Salam showed how the overlapping divergences can be handled in any theory.

680

16 Renormalization theory

Figure 16.3: The overlapping divergent graph in the fermion selfenergy at two loops.

In QED things were a bit simpler because of the Ward-Takahashi identities (which relate various amplitudes, see for example, section 9.7) and renormalization of QED was thought to be straightforward. However, Yang and Mills noticed that at the 14th order (in the coupling), the photon self-energy graph shown in Fig. 16.4 does lead to overlapping divergence and needed further prescription to handle this graph as Ward-Takahashi identities do not restrict the photon selfenergy. (It was believed earlier that the photon self energy cannot have overlapping divergences.) Yang and Mills solved the problem of the photon self-energy and together with Salam’s work as well as the subsequent work of Weinberg, the question of overlapping divergence was considered to be solved. (This is within the framework of integral equations for Green’s functions and skeleton expansions.)

Figure 16.4: An overlapping divergent graph in the photon selfenergy at the 14th order. There are mainly two equivalent renormalization methods known as the multiplicative renormalization and the BPHZ renormalization. In multiplicative renormalization, we calculate 1PI Feynman amplitudes in perturbation theory (as we have done in the last chapter) until we encounter a divergent graph. The amplitude is then regularized with a momentum cut oﬀ (or dimensional regularization or any other regularization). The regularized amplitude is then Tay-

16.2 A brief history of renormalization

681

lor expanded about zero external momenta (for massive theories) so as to separate out the local divergent parts. We then add counter terms to the Lagrangian density to exactly cancel these divergences. We continue calculating with this modiﬁed Lagrangian density and add more counter terms whenever faced with new divergences. This procedure renders the theory ﬁnite. Let us consider a speciﬁc theory to see how this procedure works. We have seen that in the φ4 theory only the two point and the four point functions are divergent. Thus we add counter terms to cancel these divergences. For example, at one loop we obtain the counter terms from calculating Γ(2) and Γ(4) as shown in Fig. 16.5 so that the divergent contributions from the one loop graph and the counter terms cancel and render the amplitude ﬁnite as shown in Fig. 16.6. The counter terms are clearly of one loop order (this is counted by the power of in the coeﬃcient which we have set to unity, but should be understood).

Figure 16.5: One loop counter terms for the two point and the four point functions in the φ4 theory. Including these one loop counter terms, at two loops the self energy graphs would, therefore, have the form shown in Fig. 16.7. These would be divergent and hence we have to add two loop counter terms for the two point function to cancel the new divergences at this order. Similarly the four point function would also need counter terms at the two loop level. However, no other 1PI graph would be divergent at this order. As an example, let us look at the 6 point function at two loops in Fig. 16.1 whose superﬁcial degree of divergence is −2 but which is divergent because of a divergent subgraph. However, with the modiﬁed Lagrangian density (i.e., with one loop counter

682

16 Renormalization theory

+

= ﬁnite

+

= ﬁnite

Figure 16.6: The sum of the graphs and the counter terms make the amplitudes ﬁnite.

+

+

+

Figure 16.7: Two loop self-energy graphs including the one loop counter terms in the φ4 theory.

terms) there exist another two loop graph in this theory shown in Fig. 16.8 (this is of two loop order because the counter term is of one loop order).

Figure 16.8: Two loop 6-point function graph with the one loop vertex counter term in the φ4 theory. It is clear that since the counter term cancels the divergence of the one loop four point function, the sum of the two graphs in Fig. 16.1 and Fig. 16.8 is convergent. The point which this analysis brings out is that the only counter terms we really need to make a theory ﬁnite are for the graphs whose superﬁcial degree of divergence is non-negative (this is why the concept of superﬁcial divergence is important within the context of renormalization). The renormaliza-

16.2 A brief history of renormalization

683

tion procedure works because the structure of the counter terms is the same as the terms in the original Lagrangian density and hence the divergence analysis remains the same even after adding counter terms. Furthermore, since the counter terms have the same form as the terms in the original Lagrangian density, they can be simply absorbed into a redeﬁnition of the original parameters in the theory. For example, in the φ4 theory we have (C.T. stands for counter terms) M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ + C.T. 2 2 4! M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ 2 2 4! A B C + ∂µ φ∂ µ φ − φ2 − φ4 . 2 2 4! (16.29) The constants A, B and C receive contributions from various loops and are regularization dependent. We can now combine the counter terms with the original terms in the Lagrangian density as 1 1 1 (1 + A)∂µ φ∂ µ φ − M 2 + B φ2 − (λ + C)φ4 , (16.30) 2 2 4!

L = =

L=

and deﬁne

φ0 = (1 + A) 2 φ = Z 2 φ,

2 M0 =

1

1

M 2 + B (1 + A)−1 M 2 + B Z −1 = M 2 ZM Z −1 , (16.31)

=

λ0 = (λ + C)(1 + A)−2 = (λ + C)Z −2 = λZ1 Z −2 .

(The mass renormalization takes this form only if the regularization procedure does not introduce any mass parameter.) With this redefinition, therefore, the Lagrangian density becomes

684

16 Renormalization theory

L = =

1 1 1 (1 + A)∂µ φ∂ µ φ − M 2 + B φ2 − (λ + C)φ4 2 2 4! 1 M2 λ0 ∂µ φ0 ∂ µ φ0 − 0 φ2 − φ4 . 0 2 2 4! 0 (16.32)

We see that this Lagrangian density in (16.32) has the same form as the one we started out with. However, our new ﬁeld variable as well as the parameters m0 , λ0 have become regularization dependent. These are known as the bare ﬁeld and the bare parameters of the theory. (Sometimes they are also referred to as the unrenormalized ﬁeld and the unrenormalized parameters of the theory and are denoted respectively by φu , Mu , λu . We will use these notations interchangeably.) However, the renormalized ﬁelds and parameters are ﬁnite. Furthermore, if we calculate with the bare Lagrangian density and bare parameters, then the amplitudes will be ﬁnite in terms of the renormalized variables. Let us also note here that the renormalized parameters we are talking about are not the conventional renormalized parameters since our graphs are expanded around zero external momenta. However, these renormalized parameters are related to the usual renormalized parameters through renormalization group equations which we will study later. Since we have already calculated the one loop amplitudes in the scalar theory, let us indicate the one loop renormalization of the φ4 theory. The only divergences in one loop come from Γ(2) , Γ(4) which in the cut-oﬀ regularization have the forms (see (15.10) and (15.19))

= −

iλ Λ2 + M 2 Λ2 − M 2 ln , 32π 2 M2

=

3iλ2 Λ2 + M 2 −1 . ln 32π 2 M2

(16.33)

Thus, we see that to one loop the parameters of the counter terms in (16.29) are

16.2 A brief history of renormalization

685

A = 0, B = − C = Λ2 + M 2 λ Λ2 − M 2 ln , 32π 2 M2 (16.34)

3λ2 Λ2 + M 2 ln , 32π 2 M2

**which lead to the one loop deﬁnition of the bare ﬁelds and parameters as
**

1 1

φ0 = (1 + A) 2 φ = Z 2 φ = φ,

2 M0 =

M 2 + B (1 + A)−1 M2 1 + Λ2 + M 2 λ ln 32π 2 M2 − λΛ2 , 32π 2

=

λ0 = (λ + C)(1 + A)−2 = λZ1 = λ 1+ Λ2 + M 2 3λ ln . 32π 2 M2 (16.35)

On the other hand, we have also seen that in the dimensional regularization, we can write λ C M2 2 A B 1 ∂µ φ∂ µ φ − φ − µǫ φ4 + ∂µ φ∂ µ φ − φ2 − µǫ φ4 2 2 4! 2 2 4! 1 1 (λ + C) 4 = (1 + A)∂µ φ∂ µ φ − φ M 2 + B φ2 − µǫ 2 2 4! = M2 λ0 1 ∂µ φ0 ∂ µ φ0 − 0 φ2 − φ4 , 0 2 2 4! 0 (16.36)

L=

with (see (15.78) and (15.79)) λM 2 2 , 32π 2 ǫ 3λ2 2 , 32π 2 ǫ

A = 0,

B=

C=

(16.37)

so that we have

686

16 Renormalization theory

Z = 1,

ZM =

1+

λ 2 32π 2 ǫ

,

Z1 =

1+

3λ 2 32π 2 ǫ

, (16.38)

and in this case, we have deﬁned λ0 = µǫ λZ1 Z −2 . (16.39)

Our calculation so far has been at the one loop. However, at nloops, the divergence structure and, therefore, the counter terms in the dimensional regularization, in general, have the form am , ǫm m=−∞

m=n

(16.40)

where am represents constants. At higher loops, we calculate amplitudes using the counter terms already present at lower orders (namely, we also include diagrams coming from counter terms at lower order). Similarly, in QED in the covariant gauge in the dimensional regularization we can write the Lagrangian density of QED in the covariant gauge with counter terms as

ǫ 1 1 (∂µ Aµ )2 / /ψ L = − Fµν F µν + iψ∂ ψ − mψψ − eµ 2 ψA − 4 2ξ ǫ A Fµν F µν + iBψ∂ ψ − Cψψ − µ 2 DψA / /ψ 4 1 / = − (1 + A)Fµν F µν + i(1 + B)ψ∂ ψ − (m + C)ψψ 4 ǫ 1 /ψ − (e + D)µ 2 ψA − (∂µ Aµ )2 2ξ

−

1 (0) (0) (0) / = − Fµν F µν(0) + iψ ∂ ψ (0) − m0 ψ ψ (0) 4 2 1 (0) / − e0 ψ A(0) ψ (0) − ∂µ A(0)µ , 2ξ0 where the bare ﬁelds and the bare parameters are deﬁned as

**16.2 A brief history of renormalization
**

2 A(0) = (1 + A) 2 Aµ = Z3 Aµ , µ 2 ψ (0) = (1 + B) 2 ψ = Z2 ψ, 1 1 1 1

687

−1 −1 m0 = (m + C)(1 + B)−1 = (m + C)Z2 = mZm Z2 ,

e0 = µ 2 (e + D)(1 + B)−1 (1 + A)− 2

−1 = µ 2 (e + D)Z2 Z3

ǫ 1 −2

ǫ

1

−1 = µ 2 eZ1 Z2 Z3 2 ,

ǫ

−1

ξ0 = ξ(1 + A) = ξZ3 .

(16.41)

We have explicitly determined at one loop that (see (15.89), (15.96) and (15.104))

A = − B = − C = − D = −

e2 2 , 12π 2 ǫ e2 2 , 16π 2 ǫ e2 m 2 , 4π 2 ǫ e3 2 , 16π 2 ǫ

Z3 = Z2 = Zm = Z1 =

1− 1−

e2 2 12π 2 ǫ e2 2 16π 2 ǫ e2 2 4π 2 ǫ

, , , . (16.42)

1− 1−

e2 2 16π 2 ǫ

The Ward identities of the theory imply that Z1 = Z2 which is explicitly seen at one loop (see section 9.7 as well as the discussion after (15.104)). This, in turn, implies that e0 = eµ 2 Z3 2 . This is interesting because it says that the renormalization of charge depends only on the photon wave function renormalization. As a result, the charge (coupling) of any fermion is renormalized exactly in the same manner. This is commonly known as the universality of charge which is a consequence of the Ward-Takahashi identity of the theory. The theories which have the property that all the divergences can be absorbed into a redeﬁnition of the parameters of the theory are known as renormalizable theories. It is clear that only theories whose index of divergence δi ≤ 0 would be renormalizable. This is because since

ǫ

−1

688

16 Renormalization theory

3 D =4−B− F + 2

ni δi ,

i

(16.43)

if δi > 0 then when we go to higher and higher orders of interaction, we generate divergent graphs with more and more external lines. That would correspond to adding to the Lagrangian density counter terms which cannot be absorbed into a redeﬁnition of the existing parameters of the theory. Motivated by the notion of counter terms, Bogoliubov and Parasiuk developed a recursive subtraction scheme. However, one of their intermediate theorems was not true which was corrected by Hepp. Hence the method is known as the BPH method. Subsequently, Zimmermann provided a solution to their equation thereby extending it to what is known as the BPHZ method which represents the second renormalization method. The BPHZ method generalizes quite nicely to non-Abelian gauge theories and, therefore, we will discuss this in detail in section 16.4. Here we simply summarize the method brieﬂy. The BPHZ method corresponds to deﬁning forests associated with each graph in the following manner. Corresponding to each Feynman graph, if there are renormalization parts in the graph (subgraphs with superﬁcial degree of divergence non-negative), then we draw boxes around the renormalization parts in various ways such that no boxes ever overlap. A particular laying down of boxes is called a forest F . The elements or boxes in a forest F are denoted by γ. Associated with each graph there is a set of forests corresponding to all possible ways of laying down boxes around renormalization parts. A forest may be empty and it is all right to draw a box around the entire graph provided the graph is a renormalization part. As an example let us look at the complete set of forests shown in Fig. 16.9 for the two loop graph of Γ(4) in the φ4 theory. It should be emphasized that the boxes contain only the renormalization parts and not any propagator external to the renormalization part. Furthermore, the graphs are considered functions of the internal loop momenta as well as the external momenta. However, the loop momenta are not integrated yet (namely, we are working with the integrand of the amplitude). We now deﬁne a Taylor operator tγ which acts on a boxed subgraph γ and replaces it by its Taylor expansion in the external mo-

Then the assertion is that the renormalized Feynman graph is convergent. Note that the BPHZ method subtracts out the divergent parts in the integrand itself so that the renormalized integral is ﬁnite. mentum variables about zero four-momentum out to order D(γ) which denotes the superﬁcial degree of divergence of γ. F ∈Φγ ∈F (16. . p) p p 2! ∂pµ ∂pν tγ Γ(2) (k. 0) + . We would see in detail how this works in section 16. then D(γ) = 0 and tγ simply evaluates the graph γ at zero external momentum. we have tγ H = 0.4). For a convergent graph H.44) Here we have neglected the linear term in the Taylor expansion since it would vanish upon integration because of antisymmetry. With the notion of the Taylor operator. we can obtain a renormalized Feynman integrand R(G) for the graph G given by the expression R(G) = (1 − tG ) (−tγ ) I(G). if γ is a Γ(4) .45) where I(G) represents the integrand of the graph G and Φ represents the complete set of disjoint and nested forests associated with the graph (these concepts will be discussed in section 16. p) = Γ(2) (k. p=0 (16.4. We should only remember that when nested boxes exist then the tγ operation should be carried from inside out.9: Forest diagrams associated with the simple two loop vertex correction diagram in the φ4 theory. If γ denotes the two point function Γ(2) .2 A brief history of renormalization 689 Figure 16. For example.16. then (D(γ) = 2 and k denotes the generic internal momentum) 1 µ ν ∂ 2 Γ(2) (k.

the generating functional would be stationary leading to R δZ = 0 = N Dφ (F [φ] − J) ei(S[φ]+ Jφ) . under an arbitrary ﬁeld redeﬁnition φ → φ + δφ inside the path integral.690 16 Renormalization theory Weinberg’s theorem. This theorem is extremely important in the study of renormalization. δφ 2 − (16. 2 2 3! L= (16. when we include quantum corrections.48) We note that since the generating functional does not depend on the ﬁeld variable (we are integrating over all ﬁeld conﬁgurations). However. We recall that the generating functional for the theory in the presence of an external source is deﬁned as (see (12. (16. (16.47). 16.120)) R Z[J] = eiW [J] = N Dφ ei(S[φ]+ Jφ) .46) The Euler-Lagrange equation for the theory is given by δS g = F [φ] = ∂µ ∂ µ + m2 φ + φ2 = 0. this equation modiﬁes and the modiﬁed equation can be obtained from the generating functional for the theory as follows.49) leads to . (16.49) where the Euler-Lagrange operator F is deﬁned in (16. Working out in detail. The integral of a Feynman graph G is abso- lutely convergent if the superﬁcial degree of divergence DH is negative for every subgraph H of G including the case when H = G.47) This describes the dynamics of the system at the tree level.3 Schwinger-Dyson equation Let us consider the φ3 theory described by the Lagrangian density m2 2 g 1 ∂µ φ∂ µ φ − φ − φ3 .

78). δJ δJ δ − J = 0. 16. this can be written explicitly as g 2 δ δJ δ δJ ∂µ ∂ µ + m2 φc + or. φc − i φc − i − J = 0. For the φ3 theory described by (16.3 Schwinger-Dyson equation 691 eiW [J] F − i or. or.51) with respect to φc and set φc = 0. δJ δW δ −i − J = 0. F φc − i δJ (16.80).46). or. If we now take the functional derivative of (16. (16.10 where the line with the blob on the left-hand side represents the complete self-energy while the lines with blobs as well as the vertex with a blob on the right-hand side denote the full propagators and the full vertex of the theory including quantum corrections to all .16. Σ= ig 2 δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c . g ig δφc ∂µ ∂ µ + m2 φc + φ2 − i − J = 0.51) δφc where Γ[φc ] corresponds to the eﬀective action deﬁned in (13. Equation (16.52) can be represented graphically as in Fig. (16.50) Here φc denotes the classical ﬁeld which is deﬁned in (13. F δ − J eiW [J] = 0. c 2 2 δJ ig g ∂µ ∂ µ + m2 φc + φ2 + 2 c 2 δ2 Γ δφ2 c −1 + δΓ = 0.52) where the restriction “|” stands for setting φc = 0 and Σ denotes the complete self-energy the φ ﬁeld (the self-energy is deﬁned to be the complete two point function minus the tree level contribution). we obtain ig δ δ2 Γ (∂µ ∂ + m )½ + 2 = − δφc 2 δφc µ 2 δ2 Γ δφ2 c −1 −1 or.

= + + Figure 16. The equations in (16. (16. we obtain δ3 Γ = −g − ig δφ3 c ig + 2 δ2 Γ δφ2 c δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ4 Γ δφ4 c δ2 Γ δφ2 c δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c −1 −1 −1 .52) and (16.10: Schwinger-Dyson equation for the two point function. To see in some detail how the BPHZ method works. orders. by taking the second functional derivative of (16.4 BPHZ renormalization We have seen in detail how multiplicative renormalization works. Such equations (and for the higher point functions) are known as the Schwinger-Dyson equations.53) are integral equations (they are written in a compact notation here) for the two point and the three point functions. let us consider the φ3 theory in six dimensions described by the Lagrangian density . 16.51) with respect to φc and setting φc = 0. (We are using a very compact notation where integrations over intermediate coordinates/momenta are suppressed.53) which can be represented graphically as in Fig.11.) Similarly. 16. Figure 16.692 16 Renormalization theory = .11: Schwinger-Dyson equation for the three point function.

the 2-point and the 3-point amplitudes are superﬁcially divergent.56) . the superﬁcial degree of divergence for any graph is given by B 0 1 2 3 D = 6 − 2B 6 4 2 0 and we see that the one point. the one point amplitude vanishes (in the massless theory) as can be seen easily from k dn k 1 = lim (2π)n k2 M →0 n −1 2 p ∼ 1 dn k n k2 − M 2 (2π) ∼ M →0 lim M 2 → 0.55) Thus. 2 2 3! (16. for this theory.) Although the zero point amplitude is divergent. we can remove this divergence by normal ordering the theory. in dimensional regularization which we will be using.54) This is a simple quantum ﬁeld theory that is quite helpful in understanding various features of renormalization. (16. In six dimensions.16. since we are interested in the ultraviolet behavior of the theory. (Note that unlike the φ4 theory. here the Lagrangian density is not invariant under φ → −φ. Furthermore.4 BPHZ renormalization 693 L= 1 M2 2 g ∂µ φ∂ µ φ − φ − φ3 . let us set M = 0 for simplicity. Furthermore. the counting of the canonical dimension gives [φ] = 2. (16.

an arbitrary mass scale to make the coupling constant dimensionless. namely. Using dimensional regularization we analytically continue to n dimensions and deﬁne ǫ = 6 − n. In order to see how renormalization works in some detail and. let us calculate some amplitudes beyond the simple one loop. 2 3(n − 2) (6 − n) ǫ = = . Therefore.694 16 Renormalization theory when n → 6. the one loop self-energy can be evaluated as k p k+p = = ǫ 1 − igµ 2 2 ǫ (16. Thus. the naive dimensional analysis gives (n − 2) .59) .57) so that we can introduce. in particular. how the overlapping divergences are handled. In this case.58) p 2 (i)2 dn k (2π)n k2 (k + p)2 1 dn k n 2 + x(1 − x)p2 )2 (2π) ((k + xp) n g 2 µǫ 2 g 2 µǫ 2 dx = Γ 2− n (−1)2 iπ 2 2 dx n Γ(2) (2π)n (−x(1 − x)p2 )2− 2 dx (−x(1 − Γ −1 + ǫ 2 n x)p2 )2− 2 = ig2 µǫ 1 n 2 (4π) 2 = ǫ ig2 µǫ 1 Γ −1 + 2 n n 2 (4π) 2 (−p2 )2− 2 dx 1 (x(1 − x))2− 2 n . In this case. 2 2 2 [φ] = [g] = n − (16. g → gµ 2 . this theory should be renormalizable. only the 2-point and the 3-point amplitudes are divergent which can be made ﬁnite with conventional counter terms. (16. as before.

76)). n ≃ 3 (4π) 2 (4π) 2 1 ǫ ǫ 1 ≃ 1 + ln 4π n Γ −1 + 3 2 (4π) 2 (4π) 2 ≃ 1 2 − + (γ − 1) ǫ 1 (4π)3 1 2 − + (γ − 1) − ln 4π . (16.59) to be k p k+p ≃− 5ǫ ig2 µǫ p2 1+ 3 6 2(4π) 6 − 2 + (γ − 1) − ln 4π ǫ 1− ǫ ln(−p2 ) 2 p . 2 2 2 2 ǫ ≃ 2 2 − + (γ − 1) . 2− 6−ǫ ǫ ǫ n =2− = 2 − 3 + = −1 + . ǫ 1 2− n 2 dx 0 (x(1 − x)) 1 = 0 dx (x(1 − x))1− 2 ǫ ln x(1 − x) 2 ǫ ≃ = 0 1 0 dx x(1 − x) 1 − dx x(1 − x) (1 − ǫ ln x) 1+ 5ǫ 6 . However. ǫ Γ −1 + 1 1 ǫ 1 + ln 4π . let us simplify this.60) = 1 6 Using these. As we have discussed earlier (see (15. we obtain the one loop self energy of the theory in (16. to understand the structure of the self energy at one loop.16.4 BPHZ renormalization 695 This form is quite suitable to carry out two loop calculations which we will do shortly.

63) Here we have used the generalized Feynman combination formula p i=1 1 (Ai )ni = Γ( i i ni ) Γ(ni ) dx1 · · · dxp δ(1 − x1 − · · · − xp ) p ni −1 i=1 xi n1 +···+np . but since the BPHZ method subtracts out divergences in every graph.696 ≃ ≃ 16 Renormalization theory ig2 p2 2 (−p2 ) 5 + ln 4π − ln − γ −1− 3 ǫ 2 12(4π) µ 3 ig2 p2 2 (−p2 ) 8 − ln − γ− 3 ǫ 2 12(4π) 4πµ 3 . (16. 12(4π)3 ǫ =− (16.62) Similarly. i xi Ai ) × ( (16.64) . (16. we can calculate the three point amplitude at one loop as (there is a second graph with p1 ↔ p2 which contributes to the amplitude. we will concentrate on graphs) k p k+p ǫ p2 k + p + p1 p1 3 = −igµ 2 3ǫ dn k (i)3 (2π)n k2 (k + p)2 (k + p + p1 )2 dn k dx1 dx2 dx3 (2π)n = g3 µ 2 2! × x1 k2 + x2 (k + p + p1 )2 + x3 (k + p)2 δ (1 − x1 − x2 − x3 ) 3.61) This leads to the one loop counter term ig2 p2 2 .

by shifting the variable of integration we obtain the value of the integral in (16. Therefore. (16.66) (16.63) to be k p k+p 3ǫ p2 k + p + p1 p1 1 1−x1 = 2g3 µ 2 dx1 0 0 dx2 1 dn k n 2 + x (1 − x )Q2 )3 (2π) (k 1 1 .65) where we have used the constraint from the delta function in the intermediate steps.4 BPHZ renormalization 697 Let us next simplify the denominator of the integrand as x1 k2 + x2 (k + p + p1 )2 + x3 (k + p)2 = (x1 + x2 + x3 ) k2 + 2k · (x3 p + x2 (p + p1 )) + x3 p2 + x2 (p + p1 )2 = k2 + 2k · ((1 − x1 )p + x2 p1 ) + (1 − x1 )p2 + x2 p2 + 2x2 p · p1 1 = (k + (1 − x1 )p + x2 p1 )2 − (1 − x1 )2 p2 − x2 p2 2 1 − 2x2 (1 − x1 )p · p1 + (1 − x1 )p2 + x2 p2 + 2x2 p · p1 1 = (k + (1 − x1 )p + x2 p1 )2 + x1 (1 − x1 )p2 + x2 (1 − x2 )p2 1 + 2x1 x2 p · p1 = (k + (1 − x1 )p + x2 p1 )2 + x1 (1 − x1 )Q2 . where we have identiﬁed x1 (1 − x1 )Q2 = x1 (1 − x1 ) p2 + x2 (1 − x2 )p2 1 +2x1 x2 p · p1 .16.

ǫ 4πµ2 (16.70) Furthermore. (16.68) as ig3 µ 2 =− (4π)3 ǫ (16. we can rewrite (16. ⇒ x2 = 1 − x1 . ǫ 4πµ2 (16.698 = 2g3 µ 2 3ǫ 16 Renormalization theory 1 1−x1 dx1 0 0 1 3ǫ dx2 1−x1 Γ 3− n i(−1)3 2 n 3− n 2 (−x (1 − x )Q2 ) 2 Γ(3)(4π) 1 1 (−x1 (1 − x1 )Q2 )3− 2 2 −γ ǫ Γ 3− n 2 n ig3 µ 2 = − n (4π) 2 ig3 µ 2 ≃ − (4π)3 ǫ dx1 0 1 0 dx2 1−x1 (16. we have x3 = 1 − x1 − x2 and x3 = 1 x3 = 0 so that 1 1−x1 ⇒ x1 = x2 = 0.67) dx1 0 0 dx2 × 1− ≃ − ig3 µ 2 (4π)3 ǫ ǫ (−x1 (1 − x1 )Q2 ) ln 2 4πµ2 (−x1 (1 − x1 )Q2 ) 2 − ln −γ .72) . (16. deﬁning a new variable (this would be quite useful for the calculation of the two loop self-energy which we will do shortly) x2 = (1 − x1 )u.71) 1 1 dx1 0 0 du (1−x1 ) 2 (−x1 (1 − x1 )Q2 ) − ln −γ .68) 1 1−x1 dx1 0 0 dx2 In this derivation we have used the fact that because of the δ function involving the Feynman parameters.69) dx1 dx2 dx3 δ(1 − x1 − x2 − x3 ) = dx1 0 0 dx2 .

62).4 BPHZ renormalization 699 where in the new variables x1 (1 − x1 )Q2 = x1 (1 − x1 )(p + up1 )2 + u(1 − u)p2 . x2 ) ig3 µ 2 2 .75) separately.73) 1 2 Equation (16. = + −igµ 2 −igµ 2 ǫ ǫ 2 i i dn k i iΠ(1) (k2 ) 2 n k2 (2π) k (k + p)2 i dn k i (−ig2 k2 ) 2 i . Let us ﬁrst look at the graphs with non-overlapping divergence of the form shown in Fig. 16.74) Let us next calculate the self-energy at two loops. (16.75) n k 2 12(4π)3 ǫ k 2 (k + p)2 (2π) 2 where we have used the form of the one loop counter term in (16. the ﬁrst integral in (16. Using the form of the one loop self-energy in (16.68) leads to the one loop counter term (the factor of arises from doing the integrals over x1 .12 (there would also be diagrams with a one loop self-energy and counter term insertion on the other internal line) k k + p k+p p k+p k p p Figure 16.16.12: Non-overlapping two loop self-energy diagrams in the φ3 theory. 1 (16.75) has the form . = 2(4π)3 ǫ ǫ (16.59). Let us look at the two integrals in (16.

700 16 Renormalization theory I1 = ig2 µǫ dn k (2π)n dx ǫ Γ −1 + 2 1 1 ig2 µǫ n n 2 2 )2− 2 k 2 k 2(4π) 2 (−x(1 − x)k × = − ≃ − × = − × ≃ g 2 µǫ 2 n 1 (k + p)2 ǫ 2 2− n 2 2(4π) 2 g 2 µǫ 2(4π) dx (x(1 − x)) ǫ 1 2 6 Γ −1 + (−1) 2 −2 dn k n (2π)n (k2 )4− 2 (k + p)2 5ǫ 6 n n n 2 n 2 Γ −1 + 1+ Γ 5− n (−1) 2 −2 (1 − y)3− 2 dn k 2 dy n (2π)n Γ 4 − n Γ(1) ((k + yp)2 + y(1 − y)p2 )5− 2 2 g 2 µǫ 2 n Γ −1 + n 12(4π) 2 dy ǫ 2 1+ 5ǫ 6 n 2 Γ 5− Γ 4− n n 2 n 2 n i(−1)5− 2 Γ 5 − n − 2 n Γ 5− n (4π) 2 2 2 (−1) 2 −2 (1 − y)3− 2 (−y(1 − y)p2 )5−n 5ǫ (−p2 )n−5 6 ǫ i g2 µǫ Γ −1 + 2 12(4π)n Γ 4 − n 2 n Γ(5 − n) 1 + × ≃ − dy (1 − y)3− 2 (y(1 − y))5−n − 1 +γ−1 ǫ 2 i(g2 )2 p2 (1 + ǫ ln 4π) − + γ − 1 6 12(4π) ǫ ǫγ 2 1+ 5ǫ 6 1 − ǫ ln × 1+ ≃ − (−p2 ) 1 5ǫ 1+ 2 µ 6 4 1 + ǫ ln 4π 1+ 5ǫ 6 ig4 p2 γ 3 2 − (γ − 1) + 6 ǫ2 72(4π) ǫ ǫ × 1 − ǫ ln (−p2 ) 5ǫ + ﬁnite + µ2 4 .

76) Here we have used (16. let us look at the second integral in (16.4 BPHZ renormalization 701 2 1 2 (−p2 ) 5 − (2γ − 3) − ln + 2 2 ǫ ǫ 4πµ 2ǫ ǫ ≃ − ig4 p2 5ǫ 1+ 6 72(4π) 6 + ﬁnite = − 1 2 (−p2 ) 5 ig4 p2 2 − (2γ − 3) − ln + 6 ǫ2 2 72(4π) ǫ 4πµ 2ǫ ǫ + 5 + ﬁnite 3ǫ = (−p2 ) 43 2 2 ig4 p2 − ln − γ + ﬁnite . + − 2− 72(4π)6 ǫ ǫ 4πµ2 12 (16.64) as well as (15. (The y integration is related to the appropriate beta 2 function. they will require non-local counter terms to cancel them.85).) We note here that the 1 ln (−p 2) terms in the ǫ 4πµ above expression are potentially dangerous because if such divergent terms are present. see (16. Next.76) and (16.16.60) in the intermediate steps.75) which leads to I2 = −igµ 2 = = ǫ 2 dn k i (−ig2 k2 ) 2 i i n k 2 12(4π)3 ǫ k 2 (k + p)2 (2π) 1 dn k n k 2 (k + p)2 (2π) dn k 1 dx n (2π) ((k + xp)2 + x(1 − x)p2 )2 Γ 2− n iπ 2 2 dx n (2π)n (−x(1 − x)p2 )2− 2 n n g 4 µǫ 2 12(4π)3 ǫ g 4 µǫ 2 12(4π)3 ǫ g 4 µǫ 2 12(4π)3 ǫ = g4 p2 2 iπ 2 ǫ ≃ − Γ −1 + 3 ǫ (2π)n 12(4π) 2 × 1− ǫ (−p2 ) ln 2 µ2 dx x(1 − x) 1 − ǫ ln x(1 − x) 2 .

78) I1 + I2 = ig4 p2 2 11 + ﬁnite .75) takes the form (−p2 ) 43 2 2 ig4 p2 − ln + −γ − 2− 72(4π)6 ǫ ǫ 4πµ2 12 + = (−p2 ) 8 2 4 ln − − + γ + ﬁnite ǫ2 ǫ 4πµ2 3 (16.77) ig4 p2 4 (−p2 ) 8 2 − ln − − + γ + ﬁnite . adding (16. takes the form ℓ k p ℓ+p k+p p . − 6 ǫ2 72(4π) 6ǫ We see that the potentially dangerous non-local divergences have cancelled in the non-overlapping divergent graphs and the remaining divergences can be cancelled by local counter terms.702 ≃ − 16 Renormalization theory g 4 p2 2 i 2 − + (γ − 1) − ln 4π 3 ǫ (4π)3 12(4π) ǫ × ǫ (−p2 ) 5ǫ 1 1 − ln + 6 2 µ2 6 ≃ − = ig4 p2 2 2 (−p2 ) 5 − + ln − + (γ − 1) + O(ǫ) 72(4π)6 ǫ ǫ 4πµ2 3 (16.77).76) and (16.67) (with x2 expressed in terms of u as in (16. upon using the form of the one loop vertex function in (16.71)). Let us next look at the overlapping divergent graph in the self energy at two loops which. the sum of the two graphs in (16. 72(4π)6 ǫ2 ǫ 4πµ2 3 As a result.

Using (16. we can write (16.73) (with appropriate momenta).16. (2π)n ℓ2 (ℓ + p)2 (Q2 )3− 2 (16. x1 (16.81) .79) as g4 µ2ǫ (−1) 2 −3 n Γ 3− n 2 2 2(4π) Γ 5− × Γ 3− = n 2 n 2 n n n n dn ℓ −3 2 dx1 du x1 (1 − x1 ) 2 −2 (2π)n = dy1 dy2 dy3 y1 2− n 2 δ(1 − y1 − y2 − y3 ) D2 n 5− n 2 g4 µ2ǫ (−1) 2 −3 n Γ 5− n 2 2(4π) 2 dn ℓ (2π)n dy1 dy2 dy3 y1 2 dx1 du x1 −3 (1 − x1 ) 2 −2 .4 BPHZ renormalization 703 3ǫ = (−igµ 2 ) 2 × ǫ i i dn ℓ ig3 µ 2 dx1 du 2 − n (2π)n ℓ (ℓ + p)2 (4π) 2 n 2 n 2 3− 2 − x1 (1 − x1 )Q n (1 − x1 )Γ 3 − n g4 µ2ǫ (−1) 2 −3 Γ 3− = n 2 2(4π) 2 dn ℓ x 2 (1 − x1 ) 2 −2 dx1 du 1 n .79) n −3 n where Q2 is deﬁned in (16.64) to combine denominators. (16.80) n 2− n 2 × δ(1 − y1 − y2 − y3 ) D2 5− n 2 where we have identiﬁed D2 = y1 Q2 + y2 (ℓ + p)2 + y3 ℓ2 = y1 (ℓ + up)2 + = u(1 − u) 2 p + y2 (ℓ + p)2 + (1 − y1 − y2 )ℓ2 x1 2 ℓ + (uy1 + y2 )p + y1 (1 − y1 )u2 + y2 (1 − y2 ) − 2uy1 y2 + y1 u(1 − u) 2 p .

let us evaluate the integral in (16. y1 .80) in some detail. so that the (shifted) denominator in (16.81) has the form D2 = ℓ2 + (1 − y1 )(y1 (u − v)2 + v(1 − v)) + = ℓ2 + d(x1 . we can redeﬁne the variable y2 (as we have already done in (16.84) . Finally. This is a special feature of overlapping divergent graphs and to appreciate this. in fact. this is not right and. where d is deﬁned in (16.83). the divergence of the subdiagram (one loop vertex) has been transformed to the Feynman parametric integrals.83) (16. To see the divergence .80) would appear naively to be contained in the ℓ integral which can only have a one loop divergence structure (note that the singular multiplicative gamma function coming from the one loop vertex has been cancelled by those coming from the Feynman combination formula). First we note that we can shift the variable of integration ℓ → ℓ − (uy1 + y2 )p. we can do the integration over y3 using the delta function in (16.80). u. y1 u(1 − u) 2 p x1 (16.82) Substituting this into (16. However. The one loop pole structure is manifest in the overall multiplicative gamma function. Next. v)p2 .80) and carrying out the ℓ integral we obtain n g4 µ2ǫ (−1) 2 −3 Γ 5− n 2 2(4π) 2 dy1 dv (1 − y1 )y1 ig4 p2 (−p2 ) 2(4π)6 4πµ2 2− n 2 2− n 2 n n 2 dx1 du x1 n = −3 (1 − x1 ) 2 −2 n × = − i(−1)5− 2 n (4π) 2 Γ 5 − n 2 Γ(5 − n) (−dp2 )5−n −ǫ Γ(−1 + ǫ) n × dx1 dudy1 dv x1 (1 − x1 ) 2 −2 y1 2− n 2 (1 − y1 )(x1 d)1−ǫ (16.704 16 Renormalization theory The divergence of the two loop integral in (16.71)) as y2 = (1 − y1 )v.

substituting this result into (16. 6 1 Γ(α)Γ(β) = B(α. 6 (16.4 BPHZ renormalization 705 structure from the parametric integration.84) gives dx1 dudy1 dv x1 = 2− n 2 (1 − x1 ) 2 −2 y1 2− n 2 n 2− n 2 (1 − y1 )(x1 d) 2− n 2 dx1 dy1 dudv x1 (1 − x1 ) 2 −2 y1 n (1 − y1 ) × x1 (1 − y1 )(y1 (u − v)2 + v(1 − v)) + y1 u(1 − u) = 1 6 dx1 dy1 x1 2− n 2 (1 − x1 ) 2 −2 y1 n 2− n 2 (1 − y1 ) × x1 (y1 + 1)(1 − y1 ) + y1 = ≃ ǫ ǫ 1 Γ(2) Γ(2)Γ( 2 ) Γ(2)Γ( 2 ) + + 6 Γ(4) Γ(3) Γ(3) 1 1 2 .84) can be evaluated exactly using Gegenbauer polynomials and the only modiﬁcation from the leading order result in (16.86) This demonstrates how the subdivergence is hidden in the parametric integration. Using the standard results 1 0 1 0 dt tα−1 (1 − t)β−1 = 1 du u(1 − u) = .85) the leading term in the parametric integration in (16. β). −γ+ 6 ǫ 6 (16. Thus.86) is that the ﬁnite constant 1 inside the parenthesis changes from −γ + 6 to 4. Γ(α + β) 0 1 dudv (u − v)2 = .84) gives the value of the overlapping self-energy diagram to be . The parametric integration in (16. let us look at only the leading order term in (x1 d)1−ǫ .16.

87) ig4 p2 2 (−p2 ) 2 + − γ + 3 + ﬁnite . − ln 12(4π)6 ǫ2 ǫ 4πµ2 We see again that the dangerous nonlocal divergent terms of the form (−p2 ) 1 ǫ ln 4πµ2 are present in this diagram.706 16 Renormalization theory = − = 1 (−p2 ) 1 2 ig4 p2 − + γ − 1 1 − ǫ ln + 4 + O(ǫ) 6 2(4π) ǫ 4πµ2 6 ǫ (16. we also have a contribution from the graph k p k+p = (−igµ 2 ) 2 g 4 µǫ 2 4(4π)3 ǫ g 4 µǫ 2 4(4π)3 ǫ n ǫ p i dn k i ig3 µ 2 2 n k 2 2(4π)3 ǫ (k + p)2 (2π) 1 dn k dx n (2π) ((k + xp)2 + x(1 − x)p2 )2 dx Γ 2− n (−1)2 i 2 n n (4π) 2 Γ(2) (−x(1 − x)p2 )2− 2 dx x(1 − x) ǫ 2 Γ n 2 n −2 2 ǫ = − = − ig4 µǫ (−p2 ) 2 −2 2 ǫ = − Γ −1 + n 3 (4π) 2 ǫ 2 4(4π) ig4 p2 4(4π)6 2 ǫ (−p2 ) 4πµ2 1− − ǫ −2 = ≃ = Γ −1 + Γ(n − 2) 1 6 1+ 5ǫ 6 −1 2 ig4 p2 2 4(4π)6 ǫ ig4 p2 24(4π)6 ǫ (−p2 ) ln 2 4πµ2 ln 2 − + (γ − 1) ǫ 2 4 + 2 ǫ ǫ 8 (−p2 ) +γ− 2 4πµ 3 + ﬁnite.74)). (16.88) . with the one loop counter term for the three point function (see (16. On the other hand.

16. BPHZ is a renormalization procedure where divergences are subtracted in the graph itself so that it works with any regularization. ǫ2 3ǫ We see that all the potentially dangerous nonlocal divergent terms have disappeared from the self-energy and the remaining divergences can be removed by adding a local two loop counter term.4 BPHZ renormalization 707 Here we have used (16. = 2 2 ig4 p2 + 6 ǫ2 12(4π) ǫ − 4 2 + 2 ǫ ǫ − ln ln (−p2 ) −γ+3 4πµ2 + ﬁnite (16.16.89) p so that the sum of the three graphs in Fig. .85) in the intermediate step.90) 8 (−p2 ) +γ− 2 4πµ 3 = ig4 p2 12(4π)6 − 2 2 + + ﬁnite . the graph with the counter term at the left vertex also contributes an equal amount k p k+p = ig4 p2 24(4π)6 − 2 4 + 2 ǫ ǫ ln 8 (−p2 ) +γ− 2 4πµ 3 + ﬁnite. Similarly.13: Overlapping two loop self-energy diagrams in the φ3 theory. (16.13 gives + + Figure 16.

92) while γ1 and γ2 are called nested if one is contained inside the other. γ γ Figure 16. namely. if none of the following holds. 16. if γ1 ⊂ γ2 .91) namely. A couple of examples of proper subgraphs which correspond to renormalization parts are shown in Fig.15 shows examples of diﬀerent ways of drawing boxes around the renormalization parts. 16. if it is superﬁcially divergent. (16.708 16 Renormalization theory Consider an arbitrary Feynman graph G. (16. for example.93) Finally. Two renormalization parts γ1 and γ2 are called disjoint if γ1 ∩ γ2 = 0. Fig.14: Examples of proper renormalization parts in the two loop self-energy graph in the φ3 theory. for the three loop self energy graph in the φ3 theory. Thus. namely. It may contain one particle irreducible subgraphs that are superﬁcially divergent. Given a Feynman graph G let us draw boxes around all renormalization parts (including the graph G if it is superﬁcially divergent) in all possible ways. An 1PI subgraph γ is called proper if it is diﬀerent from the graph G itself and it is called a renormalization part if D(γ) ≥ 0.14. two renormalization parts γ1 and γ2 are called overlapping if they share lines and vertices. . or γ2 ⊂ γ1 . (16.

16. Thus. 16. A forest is called normal if there is no box around the complete graph G.17 shows the nested forests in the two loop self-energy graph in the φ3 theory.15: Examples of forests in the three loop self-energy graph in the φ3 theory. 16. γ1 ⊂ γ2 . γ2 ⊂ γ1 .16. the complete set of laying down of boxes (forests) containing disjoint renormalization parts is shown in Fig. (16. for the two loop self-energy graph in the φ3 theory. Similarly. A forest is called empty if there is no box around any subdiagram. a forest is called full if there is a box around G. Furthermore. . Fig.94) A set of such laying down of boxes around renormalization parts is called a forest. On the other hand. γ1 ∩ γ2 = 0. Let D denote the set of laying down of boxes which contain only disjoint renormalization parts. let us add to the set D the graph with no boxes.4 BPHZ renormalization 709 Figure 16. let N denote the set of laying down of boxes (forests) containing only nested renormalization parts.

Figure 16. we deﬁne O to denote the set of laying down of boxes (forests) containing overlapping renormalization parts. Figure 16.18 for the two loop self-energy graph in the φ3 theory. 16.17: The complete set of nested forests for the two loop self-energy graph in the φ3 theory.710 16 Renormalization theory Furthermore. we must subtract out the divergences from each of these sub-diagrams. it Taylor expands .16: The complete set of disjoint forests for the two loop self-energy graph in the φ3 theory. shown in Fig.18: The complete set of overlapping forests for the two loop self-energy graph in the φ3 theory. for example. In the conventional BPHZ method. we introduce an operator tγ (the Taylor operator or the Taylor expansion operator) for any graph γ such that acting on the integrand of the graph. Figure 16. The reason why we draw boxes around renormalization parts is because these sub-diagrams are superﬁcially divergent and to make the full diagram ﬁnite.

(2π)n D(γ) = 6 − 6 = 0. p. 0) + p = dn k I(k.16. p) 2! ∂pµ ∂pν . Note that.96) This is the conventional tγ operator in the BPHZ prescription and it is clear that using this operation. p). if D(γ) < 0. tγ I (k. acting on a graph.4 BPHZ renormalization 711 it in the external momenta (external to the graph) up to terms of order D(γ). tγ IG = 0. 0). Thus. p1 ) .95) In other words. (2π)n 1 µ ν ∂2 p p I(k. p.97) (16. (in the self-energy diagram the linear term in the Taylor expansion has been set to zero because of anti-symmetry) k γ: p k+p D(γ) = 6 − 4 = 2. we are simply able to throw . p) = I(k. p1 ) = I(k. (16. tγ I(k. by deﬁnition. p=0 k γ: p k+p p2 k + p + p1 p1 = dn k I (k. 0. the Taylor operator simply separates out the potentially divergent terms in the integrand so that the ﬁnite part of the graph can be identiﬁed with the integrand (1 − tγ ) IG = I G = ﬁnite. (16.

101) Given a renormalization part γ in a Feynman diagram. in fact.98) In such a case. ∞ m=0 (16. the tγ operation is deﬁned to be (it separates out the pole terms) −1 m=−n t FG = G am ǫm . (16. compatible with any regularization. (16. has the general form ∞ m=−n FG = am ǫm .712 16 Renormalization theory away potentially divergent parts inside the integral. (16. this makes it clear that for a superﬁcially convergent graph (with no pole term) tG FG = 0. (16. a graph (integral) at n-loops. if the integral has the form ∞ m=−n Iγ = am ǫm . it never uses any particular regularization and is. Therefore. In particular.100) Once again.102) then as we have seen the Taylor operation is deﬁned such that −1 m=−n tγ Iγ = (1 − tγ ) Iγ = am ǫm = divergent parts. am ǫm = ﬁnite parts. in the case of dimensional regularization.103) .99) so that the ﬁnite part of the graph can be identiﬁed with ∞ m=0 1−t G FG = am ǫm = F G = ﬁnite. (16.

γ2 . one deﬁnes s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . Let us consider an arbitrary Feynman diagram G with proper renormalization parts γ1 .107) but this is not true otherwise. .105) We note that if G is not superﬁcially divergent.16. However. . and (16. (16. (16. . Thus. Thus. the eﬀect of the term (−tγ Iγ ) can be thought of as that of adding counter terms which subtract the divergence. makes all the subdiagrams (proper renormalization parts) in G ﬁnite. The only understanding here is that for nested diagrams. one needs a ﬁnal subtraction to render the diagram ﬁnite. graph by graph this procedure can be applied and every graph can be made ﬁnite. from the deﬁnition of RG IG in (16. (16. we see that all the superﬁcially divergent subdiagrams as well as the full diagram have been made ﬁnite by the subtractions (counter terms) so that the superﬁcial degree of divergence of every subdiagram as well as the full diagram is negative and. the graph G itself may be superﬁcially divergent in which case. In any case. let us deﬁne s RG IG = i=1 (1 − tγi ) IG . by deﬁnition tG RG IG = 0. the graph would now be ﬁnite. by Weinberg’s theorem.4 BPHZ renormalization 713 Thus. this method is correspondingly more general. However. then. since in this method we do not explicitly use counter terms.104) This. . of course. Then. γs as subdiagrams.105).106) RG IG = RG IG . the subtraction must be carried out .

Its eﬀect is to give IG itself). That this is true will be seen shortly through examples. Thus. With this. Equation (16. requires that the overlapping divergences be taken care of by lower order counter terms. as we have seen.108) is a recursion relation which does not involve nested or overlapping divergences and where at every order the operator RG is determined by the lower order operators Rγk . Zimmermann’s solution of the BPH recursion relation (16.108) as well as .19 deﬁnes all the disjoint and nested proper renormalization parts of the three loop self-energy diagram. of course.714 16 Renormalization theory inside out. The BPH recursion relation (16. for example. can lead to non-local counter terms. 16.109) where Φi is the complete set of proper renormalization parts that are non-overlapping. The proof of renormalization by local counter terms. this does not address the issue of locality of counter terms since we are subtracting out overlapping divergences which.108) is obtained as follows.108) where Φi denotes the complete set of disjoint proper renormalization parts (subgraphs) of G (Φi may be empty and when Φi is empty there is no Taylor operation. let us look at the complete set of subgraphs that include only disjoint and nested proper renormalization parts. Fig. (16. the order of the subtraction is irrelevant. For disjoint or overlapping divergences. we can now write RG IG = = 1 − tG RG IG 1 − tG (−tγk ) IG . Instead of the set of disjoint proper renormalization parts (subgraphs) of a Feynman graph G. This is seen through Bogoliubov’s R-operator which deﬁnes RG IG = = 1 − tG RG IG 1 − tG Φi γk ∈Φi −tγk Rγk IG . Φi γk ∈Φi (16.104) associated with the superﬁcially divergent renormalization part γk . This is the BPH formula. Although this method of making a Feynman diagram ﬁnite is absolutely correct. Here Rγk is the operator (16.

γ Figure 16. we have .4 BPHZ renormalization 715 γ2 γ1 Φ0 = {0} γ3 Φ1 = {γ1 } γ4 γ1 Φ2 = {γ2 } γ2 Φ3 = {γ3 } γ3 Φ4 = {γ4 } Φ5 = {γ1 .105).19: The complete set of disjoint and nested proper renormalization parts for the three loop self-energy graph in the φ3 theory. Zimmermann’s solution (16. γ4 } Figure 16. let us look at some simple examples just to get acquainted with the notions of the BPH procedure.16.20. Therefore. Before we proceed to see this. γ3 } Φ7 = {γ2 . in fact. The set of disjoint proper renormalization parts for the two loop vertex correction is shown in Fig. from the BPH formula (16. γ2 } γ1 γ2 γ4 Φ6 = {γ1 .20: The complete set of disjoint proper renormalization parts of the two loop 3 point function graph in the φ3 theory. let us next show through examples that they are all.109) seem quite diﬀerent from the original relation (16. 16. equivalent. However.108).

716 16 Renormalization theory RG IG = IG + IG/γ −tγ Rγ Iγ = IG + IG/γ (−tγ Iγ ) . From the relation IG = IG/γ Iγ . 16.110) also as RG IG = IG + (−tγ IG ) = (1 − tγ ) IG .21 containing only proper disjoint renormalization parts. We note that. Let us next look at the two loop self-energy graphs shown in Fig.21: The complete set of disjoint proper renormalization parts of the two loop self-energy graph in the φ3 theory.108)) we can write γ1 γ2 Figure 16. this is how we are supposed to renormalize the theory.110) where IG/γ is the part of the diagram which does not contain the subdiagram Iγ .112) This is exactly the original formula (16. RG IG = (16.111) 1 − tG RG IG = 1 − tG (1 − tγ ) IG . Rγ Iγ = Iγ because there is no superﬁcially divergent proper subdiagram of γ.105) and. there is no subdiagram in this case. in fact. Here we have overlapping divergences and following BPH (see (16. (16. we can write (16. in fact. RG IG = IG + IG/γ1 −tγ1 Rγ1 Iγ1 + IG/γ2 −tγ2 Rγ2 Iγ2 = IG + IG/γ1 (−tγ1 Iγ1 ) + IG/γ2 (−tγ2 Iγ2 ) = IG + (−tγ1 IG ) + (−tγ2 IG ) = (1 − tγ1 − tγ2 ) IG .113) . (16. (16. by deﬁnition.

(16.117) tγ1 tγ2 = −tγ1 = − = = ǫ ig3 2 γ1 1 −igµ 2 t 3 ǫ 2(4π) 2 dn k (i)2 (2π)n k2 (k + p)2 − 2 ǫ (16.74).116) (16.4 BPHZ renormalization 717 so that RG IG = 1 − tG (1 − tγ1 − tγ2 ) IG . sets the factor of µ 2 = 1 in the graph at one loop) = so that ig3 2 = (−tγ2 Iγ2 ) .16. At one loop we had found that (see (16.105) RG IG = 1 − tG (1 − tγ1 ) (1 − tγ2 ) IG . 2(4π)3 ǫ (16.118) 2 i (−p2 ) g4 4(4π)3 ǫ (4π)3 6 ig4 p2 2 . It is such terms that complicate the proof of renormalizability. Let us recall some of the calculations we had done earlier.114) This does not exactly coincide with the original formula (16. tγ which is supposed to isolate ǫ pole terms in ǫ. (16.115) which we recognize to correspond to the counter term associated with an overlapping divergence if it did not vanish. The diﬀerence between the two is given by 1 − tG tγ1 tγ2 IG . 12(4π)6 ǫ2 .

then. (16.105) and Zimmermann’s solution (16.121) involving subtractions of only non-overlapping proper renormalization parts. (16. there is a theorem that says that if γ1 and γ2 denote two proper renormalization parts of G that are overlapping. the ǫ1 term precisely has the same coeﬃcient and it follows. The product of the factors in (16. If we recall our calculation for the two-loop overlapping divergence (see(16. but more importantly. then we can always ﬁnd a renormalization part γ12 which contains both γ1 and γ2 and for which (1 − tγ12 ) tγ1 tγ2 IG = 0. To make contact with the formula of BPH. (16.119) Physically.122) i where Φi ’s deﬁne forests containing only disjoint or nested proper renormalization parts. namely. Let us recall that the three loop diagram has eight . In general. it is best to study an example. that 1 − tG tγ1 tγ2 IG = 0.718 16 Renormalization theory where we have used the fact that n = 6 − ǫ and that tγ1 acts on ǫ the integral as well as on µ 2 to isolate only pole terms. we note that using this. this is the statement that we do not need subtractions associated with overlapping divergences. This makes the connection between (16.87)). γi ’s are only disjoint or nested proper renormalization parts. we can now write s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG .120) This result is very important.109). 2 therefore. physically because it says that we do not need to subtract overlapping divergences. Φi γk ∈Φi (16.121) can also be written as (1 − tγi ) ≡ (−tγk ) .

γ2 } (−tγ1 Iγ1 ) (−tγ2 Iγ2 ) + IG/γ3 −tγ3 Iγ3 /γ1 (−tγ1 Iγ1 ) + IG/γ4 −tγ4 Iγ4 /γ2 (−tγ2 Iγ2 ) . Thus.123) Let us next recall that since γ1 and γ2 do not contain any proper renormalization parts. 16.γ2 } (−tγ1 Rγ1 Iγ1 )(−tγ2 Rγ2 Iγ2 ) + IG/γ4 (−tγ4 ) Rγ4 Iγ4 = 1 − tG 1 − tG 1 − tγ1 Rγ1 − tγ2 Rγ2 − tγ3 Rγ3 − tγ4 Rγ4 −tγk Rγk IG . (16. + (tγ1 Rγ1 )(tγ2 Rγ2 ) IG = (16.19.123) as RG IG = (1 − tG ) IG + IG/γ1 (−tγ1 Rγ1 Iγ1 ) + IG/γ2 (−tγ2 Rγ2 Iγ2 ) + IG/γ3 (−tγ3 ) Iγ3 + Iγ3 /γ1 −tγ1 Rγ1 Iγ1 + IG/{γ1 . (16. = Iγ2 .16.124) Using this. we can write Rγ1 Iγ1 Rγ2 Iγ2 = Iγ1 .125) Φi γk ∈Φi . we can rewrite (16.4 BPHZ renormalization 719 forests consisting of disjoint and nested proper renormalization parts which are shown in Fig. writing out explicitly the Zimmermann solution we obtain RG IG = 1 − tG IG + IG/γ1 (−tγ1 Iγ1 ) + IG/γ2 (−tγ2 Iγ2 ) + IG/γ3 (−tγ3 Iγ3 ) + IG/γ4 (−tγ4 Iγ4 ) + IG/{γ1 .γ2 } −tγ1 Rγ1 Iγ1 1 − tG −tγ2 Rγ2 Iγ2 + IG/γ4 (−tγ4 ) Iγ4 + Iγ4 /γ2 −tγ2 Rγ2 Iγ2 = IG + IG/γ1 −tγ1 Rγ1 Iγ1 + IG/γ2 −tγ2 Rγ2 Iγ2 + IG/γ3 (−tγ3 ) Rγ3 Iγ3 + IG/{γ1 .

(16. therefore. In this case.127) then. it is equivalent to making any Feynman graph ﬁnite by local subtractions (counter terms). This is. the number of such counter terms we need will be inﬁnite and we cannot simply absorb them into a redeﬁnition of the ﬁnite number of existing parameters and ﬁelds of the theory. if the index of divergence were negative.21)) 2F + 3 δi = di + bi + fi − 4. namely. in this case.126) vanishes so that the superﬁcial degree of divergence of any graph is completely determined by the number of external lines in the Feynman diagram. δi < 0. 2 (16. as we have seen the theory can be renormalized by adding a ﬁnite number of local counter terms which simply redeﬁne the ﬁelds and the parameters of the original theory.20) and (16.720 16 Renormalization theory where Φi is the complete set of proper renormalization parts consisting of only disjoint graphs. Such theories are known as renormalizable theories. Such theories are called nonrenormalizable theories. On the other hand if δi > 0. However. but more importantly only a ﬁnite number of graphs (and that too only at low orders in perturbation theory since increasing the number of interaction vertices reduces the superﬁcial degree . we can also remove the divergences by adding local counter terms. we have talked about theories where the index of divergence of interactions in four dimensions (recall that D = 4 − B − 3 i ni δi . This analysis. So far.128) not only would we have a ﬁnite number of types of graphs that can be divergent. see (16. Of course. of course. (16.108). shows how the BPH formula leads to Zimmermann’s solution and how. We would have divergence structures with more and more numbers of external lines. the BPH formula (16. through Weinberg’s theorem. it is clear that the superﬁcial degree of divergence of a graph would increase with increasing number of interaction vertices. In contrast.

p ν. we can add counter terms to write (1 + A) ∂µ Aa (∂ µ Aν.a 4 2 (16.a ) + gf abc Ab Ac ∂ µ Aν. µ Naive power counting shows that the only graphs that are superﬁcially divergent are the two point. 16. the three point and the four point amplitudes involving gauge ﬁelds as well as the two point function for the ghosts and the three point ghost interaction vertex function.5 Renormalization of gauge theories The analysis in this chapter demonstrates renormalizability of standard ﬁeld theories involving scalar and fermion ﬁelds. Let us note here (as we have seen explicitly in the case of QED) that the true degree of divergence in a gauge amplitude is generally lower than its superﬁcial degree of divergence because of gauge invariance.a ν µ ν 2 − g2 abc apq b c µ.12)) ξ 1 a L = − Fµν F µν a + F a F a + (∂µ F a )Aµ. Such theories are conventionally known as super-renormalizable theories. some subtleties arise in the case of gauge theories. However.q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ.a )+(1+B)gf abc Ab Ac ∂ µ Aν.a − ∂ ν Aµ.p Aν.a − ∂ ν Aµ.a µ ν 4 2 L=− − .5 Renormalization of gauge theories 721 of divergence) will be divergent. we do not need counter terms for the gauge ﬁxing terms).q + F a F a + (∂ µ F a )Aµ. here also we can check that only the transverse part of the gauge self-energy diverges (we have seen this explicitly in the calculation of the photon self-energy) so that the longitudinal part is not renormalized (namely. As a result.129) + ∂µ ca ∂ µ ca − gf abc (∂ µ ca )Ab cc . The total Lagrangian density with the gauge ﬁxing and the ghost Lagrangian densities has the form (see (13.a + ∂ µ ca Dµ ca 4 2 1 = ∂µ Aa (∂ µ Aν. Let us discuss this brieﬂy with the example of the pure non-Abelian Yang-Mills theory. Like QED.a ν µ ν 2 g2 ξ (1 + C)f abc f apq Ab Ac Aµ.16.

all the divergences can be taken care of. This is.a 4 2 ˜ ˜ (16. as we have seen explicitly in the calculation in QED.91)).a ) + Z1 gf abc Ab Ac ∂ µ Aν. the longitudinal part of the photon self-energy is not renormalized by quantum corrections.722 16 Renormalization theory +(1 + D)∂µ ca ∂ µ ca − (1 + E)gf abc (∂ µ ca )Ab cc µ = − Z3 ∂µ Aa (∂ µ Aν.p ν. But the main question is whether renormalization would preserve the gauge invariance – actually the BRST invariance – present in the original theory. (16. Namely. ˜ Z3 = 1 + D.a − ∂ ν Aµ. −1 1 F a(u) = Z3 2 F a . ˜ Z1 = 1 + E.a ν µ ν 2 − Z4 g2 abc apq b c µ.130) + Z3 ∂µ ca ∂ µ ca − Z1 gf abc ∂ µ ca Ab cc .q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ. if we redeﬁne the ﬁelds and parameters as a(u) 2 Aµ = Z3 Aa .131) It is worth emphasizing here again that we have not added any counter term for the gauge ﬁxing terms because. Z4 = 1 + C. of course. With these counter terms. g(u) = Z1 Z3 2 g. in fact. Z1 = 1 + B. . ˜2 ca(u) = Z3 ca . µ where we have identiﬁed Z3 = 1 + A. µ 1 ˜2 ca(u) = Z3 ca . g1 (u)2 (u) −2 = Z4 Z3 g2 . −3 1 1 g2 ˜ − ˜ −1 = Z1 Z3 2 Z3 g. a general feature of gauge theories following from the BRST invariance of the theory (see (13.

˜3 Z3 Z1 Z (16. g(u) = g1 (u) (u) = g2 . the same relation as in (16. of course. −2 2 −3 = Z4 Z3 g2 = Z1 Z3 g2 .133) and with this identiﬁcation we can write (16. namely.130) as . (13.132) ξ (u) = Z3 ξ. But the real question is how can the renormalization process guarantee this. 2 −3 ˜ 2 −1 ˜ −2 = Z1 Z3 Z3 g2 = Z1 Z3 g2 . if we choose a regularization scheme which respects the BRST symmetry.134) and remove all the divergences from the theory. (u) 2 (u) 2 g2 (16.5 Renormalization of gauge theories 723 (16. The Slavnov-Taylor identities following from the BRST invariance (see. (16.136) This is. the counterterms will satisfy (16. for example.133) This would.91)) lead to relations between various amplitudes and we can show from these that they lead to relations among renormalization constants of the form (analogous to the relation Z1 = Z2 in QED following from the Ward-Takahashi identity) ˜ Z4 Z1 Z1 = = .134) Therefore. (16.16. we will have 2 g(u) g1 2 −3 = Z1 Z3 g2 .135) which would lead to g(u) 2 = g1 (u) 2 = g2 (u) 2 . In this case. in turn require that the corresponding counter terms have to be related. therefore. naively we would expect that the BRST invariance (or even the gauge invariance in the gauge unﬁxed theory) would require the diﬀerent (bare) unrenormalized couplings to be the same.

It is for these reasons that in the case of gauge theories. (16. 16. the regularization scheme may introduce unwanted spurious eﬀects into the theory. k . 4 (16. Otherwise. consequently. · · · .137) where the covariant derivative is deﬁned with g(u) . n2 − 1 and k = 1. µ δψ k = iǫa (x) ψT a δψk = −iǫa (x) (T a ψ)k . It is essential in identifying the physical states of the theory.6 Anomalous Ward identity Let us next consider a non-Abelian gauge theory (belonging to the group SU (n)) interacting with massless fermions (quarks) described by the Lagrangian density (see (12.46)) 1 a L = − Fµν F µν. As we have seen. n. in studying nonAbelian gauge theories it is natural to discuss renormalizability using dimensional regularization. We note that this theory. dimensional regularization is one of the regularization schemes (and it is quite simple) which respects gauge invariance and BRST symmetry and.a(u) L = − Fµν F µν a(u) + 4 2 +∂ µ ca(u) Dµ ca(u) . in establishing unitarity as well as in showing the gauge independence of the physical matrix elements.139) . in addition to having the inﬁnitesimal local gauge invariance δAa = Dµ ǫa (x). 2. Let us note here that the BRST symmetry is quite fundamental in the study of gauge theories (as we have tried to emphasize in chapter 13). The BRST invariance of this Lagrangian density is now manifest (with unrenormalized ﬁelds and parameters). the choice of a regularization which preserves this symmetry is quite crucial. 2.724 16 Renormalization theory ξ (u) a(u) a(u) 1 a(u) F F + (∂µ F a(u) )Aµ. (16. a + iψ k γ µ Dµ ψk .138) where a = 1. · · · .

δψ k = −iλψ k γ5 . (16. We can write down in a straightforward manner the Ward identities associated with these conserved currents which are simply expressions of the conservation of currents. is covariantly conserved. we can still deﬁne the axial vector current as in (16. As is the case with continuous symmetries. however. (16.e. therefore.6 Anomalous Ward identity 725 is also invariant under the global transformation δψk = −iλγ5 ψk . J µ. In this theory.142) then this would not be invariant under the chiral symmetry transformations (16. (16. Let us note that. µ ∂µ J5 = 0.141) which. therefore. Dµ J µ. there are two conserved currents. We can also derive identities from this relation that would relate various matrix elements of the . if the fermions are massive.140) describes the chiral symmetry transformations under which the massless theory is invariant. Nonetheless.16. will not be conserved. µ J5 = ψ k γ5 γ µ ψk . i.141) We note here that the vector current in (16.a = 0.143) where the variation of the Lagrangian density in (16.. the invariance of the theory leads to a conserved current.a = ψ j γ µ (T a )jk ψk . rather it would satisfy the equation µ ∂µ J5 = −δLm = −2imψ k γ5 ψk . (16.143) is without the parameter of chiral transformation.140) Here λ is the space time independent real inﬁnitesimal parameter of transformation and (16.140). if there is a term in the Lagrangian density of the form Lm = −mψ k ψk .141) transforms covariantly under a gauge transformation and.

Secondly. There are various ways to see and understand this phenomenon in quantum ﬁeld theories. The Schwinger model is described by the 1 + 1 dimensional Lagrangian density 1 L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ.145) Furthermore. we ﬁnd that the axial vector Ward identities are violated. the electromagnetic coupling (electric charge) in the Schwinger model carries dimensions unlike the four dimensional QED where the electric charge is dimensionless. we have 1 [ψ] = [ψ] = . ν = 0.144) This model can be exactly solved and has been studied from various points of view. we say that the current is anomalous or that the conservation law has anomalies. 2 [Aµ ] = 0. therefore. as we have mentioned earlier. (16. gauge invariance reduces the superﬁcial degree of divergence and if treated in a gauge invariant manner even this amplitude is ﬁnite. However. . 1. This is easily seen from the fact that in 1 + 1 dimensions the Lagrangian density has canonical dimension 2 and. [e] = 1. On the other hand. by naive power counting we see that only the photon two point function (among photon amplitudes) is superﬁcially divergent. when we calculate matrix elements regularized in a gauge invariant way.726 16 Renormalization theory theory (also known as broken axial vector Ward identities). a massless photon does not have any true dynamical degrees of freedom since there cannot be any transverse polarization. When classical identities involving a current do not hold quantum mechanically. Let us note that in 1 + 1 dimensions. Let us note that in two dimensions we can choose the two Dirac matrices as 0 1 1 0 γ 0 = (γ 0 )† = σ1 = . (16. Let us discuss this in the simple model of two dimensional massless QED known as the Schwinger model. 4 µ.

µ ∂µ J5 = 0. (16. 16. (16. (16. (16. with ǫ01 = 1. γ µ γ ν = η µν + γ5 ǫµν .6 Anomalous Ward identity 727 0 1 1 0 −1 0 0 −1 γ 1 = −(γ 1 )† = −iσ2 = so that we have † γ5 = γ5 = γ 0 γ 1 = σ3 = .146) . (16.22 and this is given by / / / d2 k Tr γ5 γ µ kγ ν (k + p) . (16. −). note that in addition to the diagonal metric tensor η µν = (+. 2 2 (k + p)2 (2π) k µν I5 (p) = −e2 (16. in two dimensions we have a second rank anti-symmetric tensor (Levi-Civita tensor) given by ǫµν = −ǫνµ .16.149) As a result of these identities.152) .148) There are various simple identities in 1 + 1 dimensions involving the Dirac matrices.151) Let us now calculate the one loop amplitude involving a photon and an axial vector current as shown in Fig.147) Furthermore. but the ones that are directly of interest to us are γ5 γ µ = ǫµν γν .150) and at the classical (tree) level both these currents are conserved (recall that the fermions are massless in the Schwinger model and the gauge group in the present case is U (1)) ∂µ J µ = 0. in 1 + 1 dimensions the vector and the axial vector currents are related as µ J5 = ψγ5 γ µ ψ = ǫµν ψγν ψ = ǫµν Jν .

71) and (15.36) in the limit m = 0.728 16 Renormalization theory γ5 γ µ k γν p k+p Figure 16. (16. Furthermore.152) using dimensional regularization which we know would preserve gauge invariance. since . Thus.22: The amplitude with an axial vector current and a photon.152) corresponds to the Fourier transform of the matrix element µ 0|T J5 (x)Aν (0) |0 .153) We can evaluate the integral in (16. we note that the amplitude in (16.154) where we have taken out the multiplicative factor e2 Tr(γ5 γ µ γ λ γ ν γ ρ ) (to be restored shortly) and we have used the basic formulae of dimensional regularization in (15.73). Here we have used the propagators for the fermions following from (15. Furthermore. generalizing the integral to n = 2 − ǫ dimensions would lead to µν I5 (p) = − dn k kλ (k + p)ρ (2π)n k2 (k + p)2 kλ (k + p)ρ dn k dx n (2π) (k + xp)2 − x(1 − x)p2 dx dn k kλ kρ − x(1 − x)pλ pρ (2π)n (k2 − x(1 − x)p2 )2 n = − = − = − 2 Γ(1 − n ) 1 ηλρ iπ 2 2 − dx n n (2π) Γ(2) 2 (−x(1 − x)p2 )1− 2 − Γ(2 − n ) x(1 − x)pλ pρ 2 n . Γ(2) (−x(1 − x)p2 )2− 2 (16.

the ﬁrst term is divergent and we have to be careful in evaluating the trace.155) where we have used the gamma matrix identity in n dimension in (15. (16.6 Anomalous Ward identity 729 γ5 is a two dimensional quantity we have not yet evaluated the trace over the Dirac matrices. π (16.158) = − .154) takes the form (we now restore the multiplicative factor) µν I5 (p) = − ie2 n (4π) 2 − dx − Tr(γ5 γ µ γ λ γ ν γ ρ )x(1 − x)pλ pρ (−x(1 − x)p2 )1+ 2 ǫ ǫ 1 Tr(γ5 γ µ γ ν )ǫΓ( 2 ) ǫ 2 (−x(1 − x)p2 ) 2 . It is clear now that the factor (2 − n) = ǫ makes even the ﬁrst term ﬁnite and. We note from (16.157) is quite interesting for it leads to µν pµ I5 (p) = − pµ pλ pρ ie2 λρ νµ η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 ie2 νµ ǫ pµ . η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 (16. (16.16. On the other hand. However.154) that the second term inside the bracket is ﬁnite and hence for this term the trace over the Dirac matrices can be carried out in two dimensions (because any diﬀerence in the trace from extending to n diemnsions would be proportional to ǫ and hence would vanish in the limit ǫ → 0).149) and the ﬁnal result is obtained to be (the x integration is trivial) µν I5 (p) = − ie2 λρ νµ pλ pρ .157) The result in (16.156) Since each of the terms is now completely ﬁnite. using this relation. we note that the trace in the ﬁrst term has the form Tr (γ5 γ µ γ λ γ ν γρ )ηλρ = Tr (γ5 γ µ γ λ γ ν γλ ) = (2 − n)Tr (γ5 γ µ γ ν ).86). the trace over the Dirac matrices can be carried out in two dimensions using the identities (16. (16. in fact.

µ ∂µ J5 (x) = γλ γ5 γ µ γν Figure 16. this leads to the operator relation e2 e2 ǫµν ∂ µ Aν = ǫµν F µν . The origin of this result can be traced to the fact that the dimensional regularization which we have used in our calculation violates chiral symmetry although it preserves gauge invariance. the same behavior is ob- . quantum corrections have made the conservation law anomalous and the anomaly in the divergence of the current is given by (16. In fact. we always choose a regularization scheme which would preserve gauge invariance in the quantum theory so that the chiral current becomes anomalous. On the other hand. As a result. since gauge invariance is so vital to the physical theory.159) π 2π This shows that although the axial vector current is divergence free (conserved) at the tree level. (16.730 16 Renormalization theory Recalling that this amplitude is the Fourier transform of the matrix element in (16. This is very important in building models of particle interactions.153).23: The triangle diagram in four dimensional QED contributing to the anomaly. one of the two currents (or a linear combination of them) would always become anomalous because of quantum corrections. it is possible to show that there is no regularization which will simultaneously preserve gauge invariance and chiral invariance. the particle multiplets are carefully chosen such that the anomalous contributions coming from various multiplets cancel completely making the gauge current anomaly free.159). Although our discussion here has been within the context of the two dimensional QED (Schwinger model). In some gauge theories.

as is clear from its structure.160) corresponds to the anomaly in four dimensional QED. it is possible to deﬁne a modiﬁed axial vector current as e2 µ ˜µ J5 = J5 − 2 ǫµνλρ Aν Fλρ . a consistent grand uniﬁed theory cannot be constructed unless we can cancel out the chiral anomaly and this leads to the study of groups and representations which are anomaly free. 8π 2 (16. they have precisely this kind of coupling.162) which would satisfy a nonanomalous divergence.6 Anomalous Ward identity 731 tained in any dimension. in this case. However. We note from (16. In grand uniﬁed theories where we treat both quarks and leptons as massless. the model cannot be renormalized unless the anomaly is cancelled. Therefore.161) and the second term in (16. such a current is not gauge invariant (although the associated charge is gauge invariant). As we have discussed in the last chapter.16. Furthermore. anomalies have observable eﬀect. Fµν = 2 (16.23 leads to an anomaly in the chiral current.160) µ ∂µ J5 = −2imψγ5 ψ + = −2imψγ5 ψ + where the dual ﬁeld strength tensor is deﬁned as 1 ˜ ǫµνλρ F λρ . if we have a theory where there are both vector and axial vector gauge couplings of the fermions. For example. 8π (16. 16. We note here that. In string theories similar anomaly free considerations ﬁx the . An explicit evaluation of this diagram (that we do not go into) using dimensional regularization leads to an anomaly in the axial vector current of the form (remember that electrons are massive unlike the fermions in the Schwinger model) e2 µνλρ ǫ Fµν Fλρ 16π 2 e2 ˜ Fµν F µν . the correct pion life time can be calculated only if the chiral anomaly is taken into account.159) that a similar statement can be made in the case of the Schwinger model as well. For example. in the four dimensional QED a graph of the form shown in Fig.

7 References 1. 486 (1949). these topics are beyond the scope of these lectures. S. 47 (1969). C. T. Physical Review 177. 84. 392 (1982). 75. Coleman. 1985). E. S. 838 (1960). Caswell and A. 7. S. 301 (1966). 6. Foundations of Quantum Chromodynamics. 5. 54 (1951). 11. J. J. K. Physical Review 75. 2426 (1969). However. F. Physical Review 128. Parasiuk. Nuovo Cimento 60A. S. Physical Review 25. Hepp. Kennedy. Proceedings of the International School of Mathematical Physics “Ettore Majorana”. Dyson. 9. Velo and A. Ward. N. 16. A. Wightman in “Renormalization Theory”. Bell and R. Erice (1976). Bardeen. Adler. 1848 (1969). 4. Communications in Mathematical Physics 2. S. Salam. Wightman. ed G. 12. Weinberg. J. 426 (1951). 2. 1736 (1949). . D. S. N. Proceedings of Royal Society A64. Cambridge University Press (Cambridge. 217 (1951). Bogoliubov and O. J. 13. W. 3. 227 (1957). A. World Scientiﬁc. Physical Review 184. Acta Mathematica 97. 10. W. Singapore (1987). Schwinger. Muta. 8. Aspects of symmetry.732 16 Renormalization theory gauge group uniquely to be SO(32) or E(8) × E(8). 14. 2425 (1962). Physical Review 82. Physical Review 118. Jackiw.

) It is clear from (17. namely. −→ V (r) = e2 αp . (Physical electric charge is determined from the scattering of on-shell electrons. We can. 1+ r 3π rme 733 (17. We have seen earlier at the end of section 8. In the same manner.2) . ep denote the ﬁne strucp c ture constant and the physical electric charge of the fermion respectively. the Coulomb potential can also be obtained from the Born amplitude for the scattering of two electrons exchanging a photon.1) 1 where in CGS units α2 = p ≃ 137 and αp .1) that at the lowest order the potential is independent of the mass of the fermion and it scales with distance as 1 . 17. r (17. calculate quantum corrections to r this potential and if we add the one loop correction to the potential shown in Fig.1 Gell-Mann-Low equation In the early 1950s. Gell-Mann and Low were interested in studying how the electric force and the potential behave at extremely short distances. of course.1. then the potential is obtained to have the form V (r) = 2αp αp 1 ln + ··· .Chapter 17 Renormalization group and equation 17.10 that the Yukawa potential can be related to the Born amplitude for the scattering of two electrons exchanging a (pseudo) scalar meson.

Furthermore.4) where we have identiﬁed Λ 2α ln αp = α 1 − 3π me . we note that the one loop potential calculated with a cut oﬀ Λ leads to the form (together with the lowest order term) 2α α 1− ln r 3π 1 + r 2 Λ2 + ··· . V (r) → 1 as r → 0 (at exr tremely short distances) as naive scaling would imply. namely r ≫ me ≫ Λ . (17. 1 + r 2 m2 e V (r) = (17. In fact.734 17 Renormalization group and equation Figure 17. the potential diverges which is surprising because there is no infrared divergence in the theory and it is not clear what gives rise to such a behavior of the potential. It is clear that the one loop corrected potential has a very diﬀerent character from the lowest order potential.5) This relation can be inverted to express the renormalized parameter in terms of the physical parameter as .2) that the potential no longer scales as 1 and furthermore.1: The one loop correction to the electromagnetic potential. When r is 1 1 large. we see from (17. it r depends on the mass of the electron in such a way that we cannot take the limit me → 0. r (17. For example. it also behaves as V (r) → α r 1− Λ 2α ln 3π me = αp . in this limit.3) where α denotes the renormalized ﬁne structure constant of the theory and we note from this result that the potential is well behaved in the limit me → 0. To understand the origin of this behavior.

8) Furthermore. 1 r= m e (17. We also see that the naive scaling of the potential breaks down and mass dependence comes in due to renormalization eﬀects. (17.8). For example.9) then.1 Gell-Mann-Low equation 735 α = αp 1 + Λ 2αp ln 3π me + ··· . in the limit Λ → ∞.3) in terms of this (physical) parameter. 1+ r 3π rme This expression coincides with (17.17.6) Expressing the potential (17. Let us next investigate what would happen if we deﬁne the electric charge at some arbitrary (intermediate) distance R and not as in (17. on dimensional grounds we see that the potential can be expressed as . rme ≪ 1 we obtain αp 1 + 2αp 3π Λ ln me V (r) = r αp 2αp 1+ r 3π ln 1− 2αp ln 3π 1 + r 2 Λ2 + ··· 1 + r 2 m2 e + ··· + ··· (17.7) = Λ − ln me Λ √ 1 + r 2 Λ2 1 + r 2 Λ2 1 + r 2 m2 e = ≃ 2αp αp ln 1+ r 3π 1 + r 2 m2 e me αp 1 2αp ln + ··· . this derivation clariﬁes that the singularity in the potential at me → 0 arises purely because of the renormalization of charge at large distances.2) and also shows that the physical ﬁne structure constant can be identiﬁed with αp = rV (r) . (17. if we deﬁne αR = RV (R).

.3) as follows. r R (17. Rme ≪ 1 (the second of these limits corresponds to me → 0).3) leads. R .12) Putting this back into the potential (17.11) Inverting this relation we obtain α = αR 1 + 2αR ln 3π 1 + R2 Λ2 + ··· . 1 + R 2 m2 e (17. the dependence of the potential on me can be neglected.13) × 1− = ≃ αR r αR r 1+ 1+ 2αR ln 3π 2αR R ln + · · · 3π r which is well behaved as we have already mentioned. Rme . We note that αR = RV (R) = α 1 − 2α ln 3π 1 + R2 Λ2 + ··· . Let us also note that if we deﬁne the electric charge (ﬁne structure constant) as . 1 + R 2 m2 e (17. (We note that rme = Rme × R so that it is not an independent quantity. which can be seen from (17. R ≪ me .736 17 Renormalization group and equation r 1 V (r) = F αR . in the limit Λ → ∞ and rme . to V (r) = αR r 1+ 2αR ln 3π 2αR ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· 1 + r 2 m2 e 1 + R2 Λ2 1 + r 2 Λ2 .10) where F is a dimensionless function (function of the independent r r dimensionless variables αR .) This function should 1 not be singular as me → 0 which implies that for r. Rme ). 1 + r 2 m2 e + ··· 1 + R 2 m2 e (17.

r R r r . r V (r) = (17.14)) αr .18) (see also (17. write (see (17. = F αR .1 Gell-Mann-Low equation 737 αr = rV (r).15) which. 1 + r 2 m2 e 1 me .13) (see also the deﬁnition (17.12) that we can identify (17. R ≪ 2αR ln 3π leads to the relation αr = αR 1 + = αR 1 + 1 + R2 Λ2 − ln 1 + R 2 m2 e .17.18) This shows that the charge (ﬁne structure constant) changes (runs) with distance and from the second of the equations in (17.16)) we see that the equation governing this change is given by . . in general. then we see from (17.16) 2αR R ln + · · · 3π r and this is compatible with (17.10)) αr r 1 = F αR . 1 + r 2 Λ2 1 + r 2 m2 e + ··· (17. = αr 1 + (17. R V (r) = αr (17.17) From this analysis we see that in the limit of me → 0 we can. in the limit Λ → ∞ and r.14) α = αR 1 + 2αR ln 3π 2αr ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· .

x) . Gell-Mann-Low equation basically relates the electric charge at one distance (scale) to another (namely. In this case. if we let r.24) .21) 1 me We see explicitly from (17.18) leads to (α0 = αr=0 ) α0 = F (α0 . for example.20) and this is known as the Gell-Mann-Low equation.18) αr = F αR . dr 3π 3π r (17. As R → 0. 2.738 17 Renormalization group and equation r where dαr dr = −β(αr ).16) that. x) ∂x (17. We see from this discussion that through proper renormalization the dependence of the potential on the electron mass me disappears. R → 0 with R = x ﬁxed. As R → 0. dαr 2α2 2α2 = − R = − r.22) There are two possibilities for the solution of the equation (17. R (17. then (17. x=1 (17. (17. in the present case.23) 1. (17. In this case. the electric charge at small distances can be obtained from its value at large distances and vice versa through the Gell-Mann-Low equation) and viewed in this manner. αR does have a limit which is nonzero and is inde1 pendent of the value 137 which arises only as an initial condir tion. we note that αp and me arise only as initial conditions in this evolution. the bare charge becomes inﬁnity. the theory develops ghosts and other problems. αr= = αp . namely. ∂F (αr . r . αR does not exist and consequently.19) β(αr ) = − .

such as the Euler’s constant etc. The second source of arbitrariness in the renormalized parameters arises from the fact that any renormalization prescription introduces a mass scale (or a length scale). It is important to note that in a quantum ﬁeld theory scale invariance is broken by the masses of particles. Even within a given regularization. as r. We will discuss these ideas in more detail in section 17. (In the example of QED that we discussed in the last section. R → 0. The renormalized parameters. However.19) (also where the beta function vanishes). therefore. There are two possible sources for this arbitrariness.5. We can keep track of this by using a running coupling constant. On the other hand. Even in dimensional regularization. First.16) (remember / r that R is ﬁxed as the limit is taken) and. for example. but some constants as well. the eﬀects of masses are negligible at high energies if we renormalize the theory in an appropriate way.2 Renormalization group We see from our discussion of QED in the last section that there is an arbitrariness in deﬁning the renormalized parameters of the theory. are diﬀerent in the two renormalization schemes even though the regularization is the same (dimensional regularization). in the MS scheme we not only subtract out the pole parts (in ǫ) of an amplitude. α0 → 0 which is obvious from (17. For example.2 Renormalization group 739 This is known as a ﬁxed point of the evolution equation (17. a particular renormalization scheme may be preferrable if it makes the higher order contributions in perturbation theory smaller. the MS scheme (minimal subtraction) corresponds to subtracting out only the pole parts (in ǫ) of an amplitude. the renormalization prescription makes a diﬀerence in the sense that how much of the ﬁnite part we subtract out along with the divergent parts changes the renormalized parameters of the theory. therefore. 17. we deﬁned the charge at some length . we saw that the coupling constants are deﬁned with an arbitrary mass scale µ. This introduces an additional arbitrariness into the renormalized parameters depending on the arbitrariness in the mass scale used. in dimensional regularization. In QED.17. The only remaining breaking of scale invariance is due to renormalization itself. it corresponds to the second possibility.

u −2 λu = Z1 Z3 λ. Namely. u ′ ′ λu = Z1 Z3−2 λ′ . µ′ = S(p.26) ′ where we are denoting Z3 = Z3 (µ′ ) and so on. m′ . ′ 1 (17. If we are calculating physical quantities such as the scattering amplitudes. ′ ′ m2 = Zm Z3−1 m′ 2 . they must also have the same value independent of which set of parameters are used to calculate them. (17. (17.) Let us consider a renormalizable theory. see the discussion after (16.740 17 Renormalization group and equation scale which illustrates the source of arbitrariness of the second kind. we must have (S and S ′ denote representative S-matrix elements calculated in the two prescriptions) S ′ p. On the other hand. Namely. we are using the same regularization but diﬀerent µ scales and the assumption here is that the unrenormalized ﬁeld and the unrenormalized parameters are independent of the renormalization prescription.25) deﬁne a set of renormalized parameters. λ. −1 m2 = Zm Z3 m2 . Calculations with either of these sets of parameters will lead to ﬁnite quantities.32)) 1 2 φu = Z3 φ.27) so that if we expand the S matrix elements in a perturbation series . λ′ . m. they should not only be ﬁnite. a different renormalization scale prescription would lead to φu = Z3 2 φ′ . µ). say the φ4 theory in 4dimensions and let us assume that (in this chapter we will use the terms unrenormalized ﬁelds and unrenormalized parameters denoted by a subscript“u” for the bare ﬁelds and bare parameters.

31) = Z3 (µ′ ) Z3 (µ) . We can invert the relations in (17. λ. the residual terms can be non-negligible. (17. λ′ . In such a case.25) and (17. µ) = O λn+1 .28) the coeﬃcients of expansion in each of the two series will be renormalization prescription dependent in such a way that the two series would have exactly the same value. µ′ φu . µ) = n an λn . (17. n n S ′ p. φ′ = z µ′ . where α ≃ 137 . u (17.32) Similarly. u −1 m′ 2 = Zm (µ′ )Z3 (µ′ )m2 . say the nth order. we can check this. Let us next see that this arbitrariness in the renormalized parameters is due to a ﬁnite renormalization. m′ . m.26) so that we can write the renormalized ﬁelds in the two prescriptions as φ = Z3 2 (λ. from (17. a′ λ′ n .17.25) and (17. m′ .2 Renormalization group 741 S(p.33) . (17. m. such as QED.29) 1 In a theory.30) which leads to a relation between the two (17. λ′ . in a theory like QCD (where the value of the coupling constant is not small).26) we can also write −1 m2 = Zm (µ)Z3 (µ)m2 . in perturbation theory.µ ′ −1 2 −1 φ′ = Z3 1 −2 λ′ . µ)φu . all we can say and check is that S ′ p. If we can calculate the exact series. we can only calculate up to a given order. µ φ. µ′ − S(p. However. the higher order corrections become negligible very quickly. λ. On the other hand. Here we have identiﬁed z µ . µ′ = (17.

µ) = ′ Zm (µ′ ) Zm (µ) −1 Z3 (µ′ ) Z3 (µ) . Let us next show that these ﬁnite renormalizations (ﬁnite transformations) deﬁne a group. µ) = Z1 (µ′ ) Z1 (µ) −1 Z3 (µ′ ) Z3 (µ) 2 . (17.37) zλ (µ′ . In fact. Z1 ) and (Z3 . with (17. Thus.26) imply −1 2 λ = Z1 Z3 λu . parameters corresponding to two distinct renormalization prescriptions are related by a ﬁnite renormalization (which can be thought of as a ﬁnite transformation much like the symmetry transformations we have talked about in earlier chapters).38) ′ ′ ′ Since (Z3 .32). (17. Zm .35) Finally.34) zm (µ .36) which leads to the relation λ′ = zλ λ. it follows that (z. zm . µ)m2 . Zm . (17.742 which implies that 17 Renormalization group and equation m′ 2 = zm (µ′ .38) we note that . ′ ′ λ′ = Z1−1 Z3 2 λu .35) and (17. (17. Z1 ) diﬀer only in ﬁnite parts (the divergent parts are the same since the “µ” dependence comes only in the ﬁnite parts). from the deﬁnitions in (17.25) and (17. with (17. zλ ) are completely ﬁnite. (17.

therefore. µ). µ . µ′ ) = z −1 (µ′ . −1 zm (µ. −1 zλ (µ. .17. Zm (µ′′ ) Zm (µ′ ) −1 −1 zm µ′′ . two successive ﬁnite renormalizations lead to an equivalent single ﬁnite renormalization. (17. µ .µ z µ .2 Renormalization group 743 z µ . µ = × = Zm (µ′ ) Zm (µ) −1 Z3 (µ′′ ) Z3 (µ′ ) Z3 (µ′ ) Z3 (µ) = zm µ′′ . µ′ zλ µ′ . µ′ zm µ′ . z(µ.µ = = Z3 (µ′′ ) Z3 (µ) −1 2 ′′ ′ ′ Z3 (µ′′ ) Z3 (µ′ ) −1 2 Z3 (µ′ ) Z3 (µ) −1 2 = z µ′′ .40) The set of ﬁnite renormalizations. µ). µ). deﬁne an Abelian group known as the renormalization group and all physical quantities must be invariant under the renormalization group transformations.39) Namely. µ . µ) = 1 = zm (µ. These transformations further satisfy z(µ. physical quantities cannot depend on the arbitrary scale introduced to carry out perturbation calculations in a theory. µ) = zλ (µ. µ′ ) = zm (µ′ . Namely. µ). µ′ ) = zλ (µ′ . −1 2 Z3 (µ′′ ) 2 Z3 (µ′ ) Zm (µ′′ ) Zm (µ) Z3 (µ′′ ) Z3 (µ) Z1 (µ′′ ) Z1 (µ′ ) −1 zλ µ′′ . µ = × = Z1 (µ′′ ) Z1 (µ) Z1 (µ′ ) Z1 (µ) −1 2 Z3 (µ′ ) 2 Z3 (µ) 2 Z3 (µ′′ ) = zλ µ′′ . 2 Z3 (µ) (17.

u Written out explicitly. λ(µ′ ). . m(µ). . . . . . . mu . this has the form Γ(n) (p. pn ) = Z32 Γ(n) (p1 . G(n) (p1 . λ(µ). µ)Γ(n) (p.42) where only (n−1) of the external momenta are independent (because of overall energy-momentum conservation). . µ) . . xn ) . . namely. . µ′ = Z32 (µ′ ) Γ(n) (p. . . . pn ) = Z3 2 G(n) (p1 .744 17 Renormalization group and equation 17. λu ) u = z 2 (µ′ . xn ) = T (φ (x1 ) . pn ) . m(µ).41) Such a relation between the renormalized and the unrenormalized Green’s functions of the theory holds true in momentum space as well. µ) = Z32 (µ)Γ(n) (p. The connected renormalized Green’s functions of the theory are deﬁned as G(n) (x1 . λu ) . mu . φu (xn )) = Z3 G(n) (x1 . u −n (17. . p2 . λ(µ). n n (17. The 1PI vertex functions are obtained from the connected Green’s functions by removing the external legs or the external propagators (two point function scales as Z3 ) so that we have Γ(n) (p1 .44) which implies that the renormalized 1PI functions in two renormalization prescriptions are related as Γ(n) p. . .3 Renormalization group equation To understand the consequences of the renormalization group of transformations described in the last section. . let us next look at the Green’s functions of the scalar ﬁeld theory. . u −n 2 (17.43) (17. . . u n n (17. . pn ) . φ (xn )) −n 2 c c = Z3 T (φu (x1 ) .45) . . . . m(µ′ ).

48) −2 where we have identiﬁed Zλ = Z1 Z3 .39)) −1 −1 2 µǫ λ = Z1 (µ)Z3 (µ)µǫ λu = µǫ Zλ (µ)λu .19)) µ dλ dµ = β(λ). Let us deﬁne the renormalized quartic coupling in the scalar ﬁeld theory as µǫ λ.50) . Here we have used p to denote collectively all the independent external momenta of the 1PI vertex function. we see that a change in the renormalization scale µ → µ′ gives rise to a ﬁnite multiplicative renormalization or a ﬁnite group of transformations acting on the vertex functions of the theory.49) where we see from (17. λu = µ µ0 ǫ λZλ (µ). it follows that (see also (17.32).46) while the unrenormalized coupling can be deﬁned as (in discussions in earlier chapters we had used µ0 = 1) µǫ λu . (17. (17. (17. Since λu is µ-independent (independent of the renormalization prescription). In dimensional regularization. Thus.47) Both λu and λ are dimensionless and are related as (compare with (16.17. 0 (17. the couplings are deﬁned with an arbitrary mass scale. As we have noted this deﬁnes the Abelian group of transformations known as the renormalization group. (17. Here µ0 is assumed to be ﬁxed and µ is the variable renormalization scale. 0 0 or.48) that β(λ) = − ǫ + µ dZλ Zλ dµ λ.3 Renormalization group equation 745 which follows from the deﬁnition of z in (17.

49) decouples from the mass equation (17.54) in the MS (MS) scheme. (17.55) where the constants a. functions of λ. This is because in these schemes the renormalization constants can be expressed in perturbative series of the forms a 2 λ + ǫ q 2 λ + ǫ b c + 2 ǫ ǫ s t + 2 ǫ ǫ Zλ = 1 + Zm = 1 + λ4 + · · · .52) γm = (17. . . their dependences are of the form (the µ dependence comes in only through the coupling) β = β(λ). q. the equation for the coupling constant (17. b. m. We note that this is a consequence of the fact that dimensional regularization is a mass independent regularization. By dimensional reasoning then they cannot depend on m either. λ4 + · · · .51) we obtain dm dµ µ where = −mγm . in general. c. in fact. 2Z m dµ (17. are independent of µ because the pole terms (in ǫ) are independent of µ as there are no overlapping divergences in a renormalizable theory. t. . They are. . µ and. µ dZ m . γm = γm (λ). u u −1 (17. from −1 m2 = Zm (µ)Z3 (µ)m2 = Z m (µ)m2 . (17. Similarly.53) Both β and γm are ﬁnite functions of µ as ǫ → 0 since the divergences in the numerator and denominator cancel out.52) and can be solved independently which we will analyze in the next section. s. mu held ﬁxed. As a result.746 17 Renormalization group and equation ∂Z ∂µ and the derivative on the right-hand side should be thought of as with λu .

59) Equation (17. λu ) = 0. dµ d −n Z3 2 (µ) Γ(n) (p. m(µ). see also (17. ∂µ µ ∂m .3 Renormalization group equation 747 Let us next look at the 1PI amplitudes (17. ∂µ ∂µ ∂λ(µ) ∂µ ∂m ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) = 0. namely. λ(µ).58) β(λ) = µ γm = − γ = µ ∂Z3 . dµ µ µ ∂ ∂λ(µ) ∂ ∂m(µ) ∂ +µ +µ − nγ Γ(n) = 0. where we have deﬁned (the derivatives are taken with λu . .49) and (17. 2Z3 ∂µ We can also argue in the same way (as we have done for β and γm ) that γ is also ﬁnite and is a function of λ alone (in the MS/MS scheme). mu . λ(µ). mu .52)) ∂λ .44) Γ(n) (p. u −n (17.57) (n) µ or. µ) . (17. γ = γ(λ).56) which leads to the relation (since the unrenormalized quantities are independent of the renormalization prescription) dΓu (p. ∂µ ∂λ ∂m (17. µ) = 0. or. λu ) = Z3 2 (µ) Γ(n) (p. mu ﬁxed.17. m(µ). m ∂µ (17. µ or.57) is known as the renormalization group equation for the 1PI vertex functions and expresses how amplitudes change when the renormalization scale is changed.

62) Let us note from our earlier discussions (see (16.4 Solving the renormalization group equation As we have seen. (17. the renormalization group equation (17. µ) = 0.57) describes how renormalized amplitudes (1PI vertex functions) change as the renormalization scale is changed ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) (pi . m. ∂µ ∂λ ∂m (17. mu ﬁxed) ∂λ . if we deﬁne the Mandelstam variable associated with the n-point function as s = p2 + p2 + · · · + p2 . let us set m = 0 in which case.60) µ where pi denotes the independent external momenta and as we have seen in (17. λ. ∂µ ∂λ µ (17. the equation takes the form ∂ ∂ + β(λ) − nγ Γ(n) (pi .64) . m ∂µ (17. µ) = 0. ∂µ µ ∂m .24)) that the npoint 1PI amplitude Γ(n) in the φ4 theory in four dimensions has the canonical dimension (the 4 simply represents the fact that the vertex is deﬁned without the energy-momentum conserving delta function) Γ(n) = 4 − n. 1 2 n (17. 2Z3 ∂µ For simplicity.748 17 Renormalization group and equation 17.58) (the derivatives are taken with λu .61) β(λ) = µ γm = − γ = µ ∂Z3 . λ.63) Consequently.

∂µ ∂µ ∂t ∂t µ (17.66) we have used ∂µ = − 1 ∂µ ln µ2 = µ ) so that the 2 renormalization group equation (17. or. It follows now that ∂t ∂ ∂ ∂ =µ = . changing the renormalization scale merely corresponds to changing t.67) ∂t ∂ s 1 (where from (17. t) + v(x) = L(x)ρ(x.66) so that for a ﬁxed s. µ2 s . s (17. consequently. it can be thought of as scaling the momentum) µ= √ set .62) can be written as ∂ ∂ + β(λ) − nγ Γ(n) = 0. t).17. λ. µ) = s 4−n 2 √ s.70) . scaling all momenta by a factor of Γ(n) (pi . λ. we can write f (n) pi · pj s . 2 µ (17. ∂t ∂λ and Γ(n) = s 4−n 2 (17. a change in t can be thought of as changing µ or for a ﬁxed µ.4 Solving the renormalization group equation 749 then.68) f (n) t.68) now becomes straightforward from the observation that a linear ﬁrst order diﬀerential equation of the form ∂ρ(x. Let us relate the renormalization scale µ to a new dimensionless variable t as (this has the advantage that for ﬁxed s. ∂t ∂x (17. t = − 1 s ln 2 . t) ∂ρ(x. f (n) is a dimensionless function of independent dimensionless variables which are Lorentz invariant.69) Solving the renormalization group equation (17. (17.65) where we have used the fact that the n-point vertex function of a scalar ﬁeld theory is Lorentz invariant and. pi · pj . λ.

76) . 0)) = ρ (x(x. dt which can be integrated to give ρ (x(x. −t)) e R0 −t R0 −t dt′ L(x(x. t)) = ρ (x(x. t = 0)) e Rt 0 (17. (17.72). Therefore. 0) = x. We can. are deﬁned by the equation dx(x. (17.72) (17.71) Using (17. λ ↔ x.70) as t ↔ t. ρ(x) = ρ (x(x. we see that solving the equation (17. (17.68) with the hydrodynamic problem (17. where we have used (17. of course.t′ )) .70) can be written as dρ(x) = L(x)ρ(x).73) dt′ L(x(x. dt with the boundary condition x(x. in this case. Γ(n) ↔ ρ. (17. directly identify the quantities in the renormalization group equation (17.750 17 Renormalization group and equation is of hydrodynamic type and is best solved by the method of characteristics.70) clearly depends on determining its characteristics x(x. β(λ) ↔ v(x). we then obtain ρ (x(x. t).t′ )) . The characteristics.t′ )) . nγ ↔ L.71). t) = v(x).74) Translating t by −t. we note that the diﬀerential equation (17.75) dt′ L(x(x. −t)) e or.

−t + ln α). 0 pi · pj n R−t dt′ γ(λ(λ.68) requires solving the characteristic equation dλ = β(λ).77) and once we know the solution to this equation. Let us next note that if λ∗ is a ﬁxed point of the coupling constant evolution equation (17. (17.17. This is known as Weinberg’s theorem (the same theorem which also made its appearance in renormalization theory). (17. we can obtain Γ(n) = s 4−n 2 f (n) (λ(λ.77).81) where D Γ(n) represents the superﬁcial degree of divergence of the n-point function and ‘a’ is a constant which can be determined from the explicit form of the amplitude. dt (17.t′ −ln α)) → α4−n (ln α)a · · · = (n) (α)D(Γ ) (ln α)a · · · .78) Note that if we scale all momenta (keeping µ ﬁxed) as s → α2 s. −t).79) so that (remember that −t = t → t − ln α. if . )e s (17. namely. (17. × en R ln α −t+ln α pi · pj ) s dt′ γ(λ(λ.4 Solving the renormalization group equation 751 so that the solution of the renormalization group equation (17. for large α (large momentum).t′ )) . 1 2 s ln µ2 ) pi → αpi . the leading behavior of the amplitude is given by 4−n 2 Γ(n) → α2 s f (n) (λ(λ.80) then.

84) As a result. the n-point amplitude will scale. under a scaling pi → αpi (17.752 17 Renormalization group and equation β (λ∗ ) = 0. namely.87) (17.24)).86) which follows from the fact that in perturbation theory γ(λ) is a power series in λ. as Γ(n) ≃ α4−n(1+γ(λ∗ )) Γ(n) . = s 2 e 2 ∗ s 4−n 2 f (n) λ∗ . s (17. near the ﬁxed point of the renormalization group equation. . then. at the ﬁxed point. In such a case the amplitudes will have almost naive scaling behavior at very high energies (“almost” because f (n) s also depends on t through the explicit dependence on µ2 which we have not displayed).79). the n-point amplitude (17. (17. if λ∗ = 0. λ = λ∗ = constant.82) (17. If the ﬁxed point of the beta function happens to be at the origin. (17.85) We note that this is almost like the naive canonical scaling behavior of the amplitude except for an extra anomalous scale dimension γ(λ∗ ) for the scalar ﬁelds (recall (16.83) Near the ﬁxed point. then γ(λ∗ ) = 0.78) behaves as Γ(n) = s pi · pj ntγ(λ∗ ) e s 4−n − n γ(λ∗ ) ln s pi · pj µ2 f (n) λ . (17. = s 4−n 2 s µ2 − n γ(λ∗ ) 2 f (n) λ∗ . pi · pj .

90) λ = 0.89) Before solving this equation explicitly. 0. 17. Therefore. if β(λ) is negative (β < 0. the ﬁrst thing that we notice is that (17. the behavior of β(λ) is shown in Fig.92) On the other hand. this leads to the behavior of the β function as shown in Fig.88) implies that (recall that λ(λ. t = 0) = λ) λ λ dλ′ β (λ′ ) t = 0 dt′ = t. Qualitatively.17. In perturbation theory the beta function has the form of a power series in the coupling constant β(λ) = β0 λn + O λn+1 . therefore. λ(t) → (17. In this case. t → ∞. If there are no other ﬁxed points of β(λ) and it is positive (namely if β0 > 0 and λ restricted to the positive half line).2.3 and in this case. that the solution of the coupling constant equation is quite crucial in understanding the behavior of the renormalized amplitudes at high energy as well as under a change of the renormalization scale. t → −∞. (17.4 Solving the renormalization group equation 753 It is clear. 17. and λ > 0). qualitatively we have . (17. let us analyze the qualitative behavior of the solution. the function dλ is monotonically increasing for positive t and we say that dt the origin in the coupling constant space is an infrared ﬁxed point. the behavior of the coupling constant is given by ∞. We note that dλ dt = β λ . (17.91) is a ﬁxed point of the beta function.

namely. ∞. . (17. we recover the two cases we have discussed earlier). Let us assume that the beta function has a nontrivial ﬁxed point at λ∗ on the positive axis.93) and we say that the origin in the coupling constant space is an ultraviolet ﬁxed point.2: The graph showing that origin in the coupling constant space is an infrared ﬁxed point of the β function.754 17 Renormalization group and equation β(λ) λ Figure 17. β(λ) λ Figure 17.3: The graph showing that the origin in the coupling constant space is an ultraviolet ﬁxed point of the β function. then expanding around this ﬁxed point. Let us next analyze the behavior of the solutions in some detail. t → ∞. we have two distinct cases to consider (when λ∗ = 0. t → −∞. λ(t) → 0.

In the diagram Fig. (17. ′−λ ) β0 (λ β0 λ − λ∗ ∗ (17. if β0 < 0. λ → λ∗ . (17. (17.4 (which shows the behavior of a general β function). λ − λ∗ 1 dλ′ = ln . λ → λ∗ .95)). if β0 > 0. µ → ∞ and when t → −∞. λ − λ∗ = (λ − λ∗ ) eβ0 t .95) In the ﬁrst case (17. then as t → ∞.4 Solving the renormalization group equation 755 1.96) which implies that 1. (17. (17. β(λ) = β0 (λ − λ∗ ) + O (λ − λ∗ )2 . β(λ) = β0 (λ − λ∗ )r + O (λ − λ∗ )r+1 . In the second case (see (17.98) The limits t → ∞ (t → −∞) are known respectively as the ultraviolet (infrared) limits and ﬁxed points which arise in these limits are correspondingly known as ultraviolet (infrared) ﬁxed points (note from (17.17.66) that when t → ∞.97) 2. 2. µ → 0 for ﬁxed s). 17. the arrows indicate the direction in which λ moves as t → ∞ (if we start near the ﬁxed point). then as t → −∞.89) leads to .89) leads to λ t= λ dλ′ = β(λ′ ) λ λ or.94) r > 1.

100) where β0 is a constant calculable in perturbation theory. Thus. 1 (r − 1)β0 r−1 − 1 (λ − λ∗ )r−1 . g = 0 is .756 17 Renormalization group and equation β(λ) λ∗ λ β(λ) λ∗ λ Figure 17. λ t= λ dλ′ = β(λ′ ) λ λ dλ′ β0 (λ′ − λ∗ )r 1 λ − λ∗ r−1 =− or. the beta function has the form (namely.4: Ultraviolet and Infrared ﬁxed points of the β function. (17. (17.22)). for example. The graph on the left shows an ultraviolet ﬁxed point at λ∗ while the one on the right corresponds to a nontrivial infrared ﬁxed point.99) Both in QED and QCD. = 1 − (r − 1)β0 t (λ − λ∗ )r−1 1 − (r − 1)β0 (λ − λ∗ )r−1 t . the series starts at the cubic order in the coupling) β(g) = β0 g3 + O g5 . while for QCD. 1 − (r − 1)β0 (λ − λ∗ )r−1 t (17. 1 λ − λ∗ = or. The diﬀerence between the two theories is that for QED β0 > 0 (see. (λ − λ∗ )r−1 r−1 λ − λ∗ = (λ − λ∗ )r−1 . β0 < 0. in both cases.

as we have seen. identifying the coupling with electric charge g = e and recalling β0 > 0).99) e2 = 1 − 2β0 e2 t 1 .101) As t increases.87)). at large momenta almost naive scaling sets in (see discussion after (17.99) g2 = 1 + 2|β0 |g2 t 1 . (g∗ = 0.104) Since the unrenormalized (bare) coupling does not depend on the renormalization scale. 2β0 e2 t→ (17. we see that the coupling grows and. Such a pole in the evolution of the coupling constant at a particular momentum scale is called the Landau pole. diverges when 1 . − 2β0 t e2 (t) = 1 e2 (17. (17. for QED we have from (17.46)-(17. β0 < 0) we have from (17. it follows that .102) and perturbation theory breaks down.38)) µ µ0 ǫ λu = λ 1+ 3λ 16π 2 ǫ . in particular. the coupling becomes weaker and weaker and ultimately vanishes. For QCD.48) as well as (16. So (setting g∗ = 0.17. the theory behaves like a free theory and such theories are called asymptotically free theories.103) As t → ∞. on the other hand.4 Solving the renormalization group equation 757 the only ﬁxed point – for QED it is the infrared ﬁxed point while for QCD it is the ultraviolet ﬁxed point. therefore. Let us now illustrate how the β function is calculated in the context of the φ4 theory in four dimensions which we have studied in some detail. + 2|β0 |t g 2 (t) = 1 g2 (17. At inﬁnite energy. In such theories. Let us recall that with the MS (minimal subtraction) scheme at one loop we have (see (17.

ǫλ 1 + 3λ 16π 2 ǫ +µ ∂λ ∂µ 1+ 3λ 8π 2 ǫ = 0.758 17 Renormalization group and equation µ ∂λu = 0.(17. 2 µ 16π µ or. at one loop in the φ4 theory in four dimensions we obtain ∂λ 3λ2 = .106) = 3λ2 . 2 λ(µ) λ(µ) 16π µ λ(µ) = 1 1 λ(µ) − 3 16π 2 µ ln µ . 2 λ(µ) λ(µ) 16π µ 1 1 3 µ = − ln . Therefore. or.107) so that we have λ(µ) λ(µ) µ µ dλ 3 = 2 λ 16π 2 dµ 3 µ = ln . (17. ∂µ 16π 2 β(λ) = µ (17.108) . − 1 1 3 µ + = ln . ∂µ µ µ0 ǫ or.105) This leads to −1 µ ∂λ ∂µ = lim −ǫλ 1 + ǫ→0 3λ 16π 2 ǫ 3λ 16π 2 ǫ 1+ 1− 3λ 8π 2 ǫ = lim −ǫλ 1 + ǫ→0 3λ + O(λ2 ) 8π 2 ǫ = lim −ǫλ 1 + ǫ→0 3λ 3λ − 2 + O(λ2 ) 2ǫ 16π 8π ǫ (17. 16π 2 which is ﬁnite as discussed earlier. or.

2 (17. (In fact. where we have identiﬁed λ 1 Linv = ∂µ φ∂ µ φ − φ4 . only theories involving non-Abelian gauge interactions may be asymptotically free. all other theories have the same character as QED. (17.5 Callan-Symanzik equation Let us consider the φ4 theory in four dimensions described by the Lagrangian density L = Linv + Lm .101)). physical quantities of the theory denoted by P (such as the physical masses of particles etc) satisfy the renormalization group equation ∂ ∂ + β(λ) ∂µ ∂λ µ P = 0.109) which simply emphasizes the fact that they are invariant under a change of the renormalization scale.112) where α denotes the ﬁnite global parameter of scale transformation. (17. For an inﬁnitesimal scale transformation (α = ǫ = inﬁnitesimal). We note that under a scale transformation the coordinates transform as xµ → eα xµ . we note that unlike the general renormalized amplitudes. 17. we have δxµ = ǫxµ .113) .17.5 Callan-Symanzik equation 759 This is precisely the same behavior as we ﬁnd in QED (see (17. therefore.) Finally.111) (17. (17.110) The rationale for dividing the Lagrangian density in this way will become clear shortly. 2 4! Lm = − m2 2 φ .

we note from (17.112).760 17 Renormalization group and equation Under the scale transformation (17.114) Here d represents the scale dimension of the ﬁeld which coincides with the canonical dimension at the tree level. which has the inﬁnitesimal form φ(x) → (1 + ǫd)φ(x + ǫx). δφ = ǫ (xµ ∂µ + d) φ(x). As we have discussed earlier.116) Therefore. or. (17. for a scalar ﬁeld in four dimensions d = 1 at the tree level. (17.111) that (here we will use d = 1 in (17.115) (which is the reason for the subscript in the Lagrangian density). Under an inﬁnitesimal scale transformation. the corresponding action would be invariant under the inﬁnitesimal scale transformations (17.115) (17. The N¨ther current for o . the scalar ﬁeld transforms as φ(x) → eαd φ (eα x) .115) for the scalar theory in four dimensions) λ 3 φ δφ 3! λ 3 φ ǫ (1 + xν ∂ν ) φ 3! δLinv = ∂µ φ∂ µ δφ − = ∂µ φ∂ µ (ǫ (1 + xν ∂ν ) φ) − = ǫ ∂µ φ∂ µ φ − λ 4 φ + ∂ µ φ∂µ φ 3! λ ν 3 x φ ∂ν φ 3! + xν ∂ν λ 1 ∂µ φ∂ µ φ − φ4 2 4! +xν ∂µ φ∂ µ ∂ν φ − = ǫ 4 1 λ ∂µ φ∂ µ φ − φ4 2 4! = ǫ (4Linv + xµ ∂µ Linv ) = ǫ∂µ (xµ Linv ) . but we have also seen that the scaling dimension of a ﬁeld is diﬀerent in diﬀerent dimensions.

5 Callan-Symanzik equation 761 µν scale transformations can be constructed as (see (6.13) and Timp corresponds to the improved energy-momentum tensor of the theory) µν S µ = xν Timp . we note here that the improved energymomentum tensor diﬀers from the conventional T µν by total divergence terms and has the explicit form 1 µν η 6 µν Timp = T µν + − ∂ µ ∂ ν φ2 . (17.120) .117) and the conservation of this current leads to µ µν µ ∂µ S µ = Timp µ + xν ∂µ Timp = Timp µ = 0. this can be checked explicitly to hold using the equations of motion.) For completeness. the conservation of the scale current requires that the energy-momentum tensor be traceless. under the inﬁnitesimal scale transformation in (17.118) where we have used the fact that the improved energy-momentum tensor is divergence free. (17. On the other hand. (17.17. the mass term in the Lagrangian density (17.111) behaves as δLm = −m2 φδφ = −m2 φǫ (1 + xµ ∂µ ) φ = −ǫ m2 φ2 + m2 xµ φ∂µ φ = −ǫ m2 φ2 + = −ǫ m2 φ2 + ∂µ = −ǫ∂µ m2 µ 2 x φ 2 m2 µ 2 x φ 2 + ǫm2 φ2 − 2m2 φ2 m2 µ x ∂µ φ2 2 = ǫ ∂µ (xµ Lm ) + m2 φ2 . (17. (When m = 0.115). Namely.119) This improved tensor is important in studying quantum ﬁeld theories in a curved (gravitational) background.

[S 0 .111)). φ (xn )) = T (∆(x)φ (x1 ) .124) where we have used the fact that the integral of S 0 (x) over all space corresponds to the charge associated with the scale current and generates inﬁnitesimal scale transformations of the scalar ﬁelds through . there is a second term that does not vanish. in this case. As a result.119) using the equations of motion) µ ∂µ S µ = Timp µ = ∆. (17. the Ward identity for the scale current. φ (xn ) + · · · + δ x0 − x0 n T φ (x1 ) . ∆ = m2 φ2 .123) When integrated over x the left-hand side in (17. . . .121) This is analogous to the case of fermion masses breaking the chiral invariance of the theory leading to an identity (in the conservation equation for the axial current) of the form (see (16. + ··· (17.143)) µ ∂µ J5 = −2imψγ5 ψ. φ (xn )) + δ x0 − x0 1 T [S 0 (x).122) Correspondingly. . the scale current will not be conserved in the full theory. This shows that the mass term breaks the invariance under scale transformations (which is the reason for the particular decomposition of the Lagrangian density in (17. φ (x1 )] . (17. .762 17 Renormalization group and equation The divergence term vanishes upon integration in the action. . . In fact. . (17. rather the divergence of the scale current will satisfy a relation of the type (this can be checked easily from the deﬁnition (17.123) vanishes and the integrated Ward identity takes the form i T (δφ(x1 ) · · · φ(xn )) + · · · + i T (φ(x1 ) · · · δφ(xn )) =− d4 x T (∆(x)φ(x1 · · · φ(xn )) . φ (xn )] . would have the form ∂µ T (S µ (x)φ (x1 ) . any term in the Lagrangian density with a dimensional coupling will break scale invariance. However.

5 Callan-Symanzik equation (n) 763 commutators (see. If G(n) (pi ) and G∆ (p. for example.17. pi ) denote the 1PI vertex with a ∆ insertion (which is deﬁned in (17. Then. we have used identities such as x∂x δ(x) = −δ(x)) and the factor (n(d − 4) + 4) denotes the naive canonical dimension of the n-point Green’s function (with d denoting the dimension of the scalar ﬁeld). pi ) denote the connected Green’s functions respectively for n scalar ﬁelds and n scalar ﬁelds as well as a ∆ composite operator.125) i × T ∆(x) as well as (17. (17.37)). however. pi ) i = d x i=1 d4 xi eipi ·xi φ(xi ) . Let Γ(n) (pi ) denote the 1PI vertex function with n scalar ﬁelds and (n) Γ∆ (p. We are. pi ) . pi ).121) and we are suppressing the dependence on other variables for simplicity). then using the deﬁnitions (we are not writing explicitly a subscript “c” to denote it is a connected Green’s function.127) . but this is to be understood) n (2π) δ 4 4 4 4 i pi G (n) (pi ) = i=1 4 d4 xi eipi ·xi T i n φ(xi ) .115). we can write (17. ∂pµ i (17.126) Note here that here the factor “4” arises from the overall energymomentum conserving delta function (in simplifying the action of the diﬀerential operator on the delta function. interested in the 1PI vertex functions.126) the identity satisﬁed by the n-point 1PI vertex functions which takes the form n−1 pµ i i=1 ∂ (n) + nd − 4 Γ(n) (pi ) = −iΓ∆ (0. (13. (n) (17.124) in the momentum space to have the form n−1 n(d − 4) + 4 − pµ i i=1 ∂ ∂pµ i G(n) (pi ) = iG∆ (0. we can obtain from (17. (2π) δ (n) pi G∆ (0.

m (17.764 17 Renormalization group and equation This relation is quite easy to understand. (17.129) which leads to ∂Γ(n) = ∂pµ i ∂ ∂m pµ i i 4−n−m Γ(n) . in this case (with d denoting the dimension of the scalar ﬁeld). In fact. which can also be written as (n) . Since at the tree level (where d = 1) the n-point function has a canonical dimension of (4 − n) we can write Γ(n) (pi . This leads to ∂Γ(n) (pi . is given by (2d − 4) so that the naive canonical dimension of the n-point 1PI vertex function is obtained to be (n(d − 4) + 4 − n(2d − 4) = 4 − nd which is reﬂected in (17.130) Using this in (17.128) where f (n) is a dimensionless function (of dimensionless variables).127) we obtain ∂ + nd − 4 Γ(n) ∂pµ i ∂ + nd − 4 Γ(n) ∂m pµ i i = 4−n−m = − m ∂ + n(1 − d) Γ(n) ∂m (17. The canonical dimension of the propagator. pi .131) = −iΓ∆ (0. (17. the 1PI vertex functions are obtained from the connected Green’s functions by removing external lines (propagators).λ . λ) = m4−n f (n) pi . λ). λ) = ∂m ∂ ∂pµ i m 4−n− pµ i i Γ(n) (pi . λ).127).

5 Callan-Symanzik equation 765 m ∂ (n) + n(1 − d) Γ(n) (pi . ∂m (17.132) has the form m ∂ + n(1 − d) Γ(n) = 0.133) → s so that the leading behavior of the relation (17.132) In the deep Euclidean region where s → ∞ (recall (17. pi . However. for example. λ) = iΓ∆ (0. there must be some violation of the naive scale Ward identity (17.19)). we have .10) or (15.17. (17.134) We note that if d = 1 as in the tree level. mu ∂Γu ∂mu (n) (n) n n (n) = iΓ∆u (p. (15. Thus. (n) (17. as we have seen through explicitly calculations. To understand this question better. Therefore.121) has two extra scalar ﬁelds) Γ(n) → s Γ∆ (n) 4−n 2 2−n 2 . λ). this relation would imply ∂ m ∂m Γ(n) = 0.132).65) and note that a vertex with a ∆ insertion (17. pi ) . . pi ) = ZZ32 Γ∆u (p. u Γ∆ (p. ∂m (17. this is not true for the renormalized amplitudes since they depend on mass through logarithm terms (see. let us note that we can relate the renormalized vertex functions to the unrenormalized ones as Γ(n) (pi ) = Z32 Γ(n) (pi ) . pi ) .121) (related to mass renormalization).135) where Z represents the renormalization of the composite operator (17.

Renormalization introduces a mass scale and thereby violates scale invariance. d = 1 + γ. (17.138) ∂ ∂ (n) + β(λ) − nγ Γ(n) = iΓ∆ . this shows that there is a violation of the naive scale Ward identity in a renormalized theory. pi ) = iZZ32 Γ∆u (0.132) with the identiﬁcation γ = −(1 − d). in general.136) which can be written as m with β(λ) = Zmu m = Zmu γ = ∂λ . ∂m ∂λ (17. It is impossible to ﬁnd any regularization which will respect scale invariance which is the essence of the Callan-Symanzik equation. (17.766 17 Renormalization group and equation iΓ∆ (0. pi ) = ZZ32 mu n (n) n (n) n ∂Γu ∂mu (n) = ZZ32 mu = Zmu = Zmu ∂ −n Z3 2 Γ(n) ∂mu ∂Γ(n) nZmu ∂Z3 (n) − Γ ∂mu 2Z3 ∂mu ∂λ ∂ ∂m ∂ + ∂mu ∂m ∂mu ∂λ − n 2 ∂ ln Z3 ∂mu Γ(n) .137) reduces to the earlier equation (17. This equation is known as the Callan-Symanzik equation. It simply reﬂects the fact that there is an anomaly in the scale Ward identity due to renormalization. or. .139) However. ∂mu (17. (17. ∂mu ∂m . 2 ∂mu When β(λ) = 0.137) Zmu ∂ ln Z3 .

Bogoliubov and D. Shirkov. Helvetica Physica Acta 26. J.17. 7.6 References 767 17. D. MIT Press (Cambridge. 3. Symanzik. S. North Holland Publishing (Amsterdam. 42 (1970). L. 1985). K. Communications in Mathematical Physics 18. 4. A. Nauka. Physical Review D2. 499 (1953). C. G. Physical Review 95. V. M. eds. Stuckelberg and A. Moscow (1984). G. eds. 1300 (1954). Peterman. Annals of Physics 59. Low. 227 (1970). 8. Aspects of symmetry. 5. 2. Coleman and R. C. Jackiw. G. Huang and R. Gell-Mann and F. S. 9. 1541 (1970). Guth. Cambridge University Press (Cambridge. R. Introduction to the theory of Quantized Fields. K. Low. H. 6. S. E. Jaﬀe. Weinberg in Asymptotic realms of physics: essays in honor of Francis E. N.6 References 1. Callan. Balian and J. 1983). Gross in Methods in ﬁeld theory. Coleman. C. . Zinn-Justin. Callan. 1981). N.

This page intentionally left blank .

457 Weyl fermions.Index Rξ gauge. 726 Anti-BRST transformation. 549 Anti-commutation relation. 5 Bogoliubov’s R operator. 480 Equality of life times. 480 Equality of masses. 130 Adiabatic hypothesis. 733 Born approximation. 594 ’t Hooft trick. 214 Charge conjugation. 263. 597 MS scheme. 229 Adjoint representation. 684 Baryon number. 479 ’t Hooft gauge. 752 Anomalous Ward identity. 267. 725 Bare ﬁeld. 313. 548. 578 Axial vector current. 353 Bianchi identity. 757 Axial gauge. 111 Anomalous scale dimension. 500 Adjoint spinor. 739 φ4 theory. 714 Born amplitude. 436 Dirac bilinears. 299 Bhabha scattering. 556 Callan-Symanzik equation. 759 Canonical dimension. 442 Eigenstates. 453 Spin zero ﬁeld. 147. 219 CPT theorem. 446 Dirac ﬁeld. 36 Anomalous current. 230 V − A theory. 151 Charge. 138. 266 Charge renormalization. 499 Casimir operator. 545 N¨ther current. 220. 437 Weak current. 724 Anomaly. 669 Canonical Poisson bracket. 20 Asymptotically free theory. 324 BRST charge. 285 Anti-commutator. 231 1PI vertex function. 517. 168 Casimir of representation. 662 769 . 533 Bjorken-Drell metric. 479 Electric charge. 726 Anomalous magnetic moment. 594 S-matrix. 552 o BRST transformation Nilpotency. 545 BRST symmetry. 167. 449 Charge operator. 521. 546. 229. 684 Bare parameter. 536 “In” and “out” states. 567 Abelian group. 553 BRST invariance.

300 Compton scattering. 328 Solution. 137 Covariance. 474 Maxwell ﬁeld theory. 303 Cliﬀord algebra.770 Index Chiral anomaly. 27 Einstein relation. 305 Constrained system. 379 Quantization. 379 Current. 107 Dirac ﬁeld Energy eigenfunction. 666 Left-handed