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

the dynamical equation (1. in this case. we start with the Hamiltonian description of the corresponding classical.1) takes the form 1 . 2m H= (1. has the generic form p2 + V (x). non-relativistic quantum mechanics. is given by the time dependent Schr¨dinger equation which has the form o ∂ψ = Hψ.1 Introduction As we know. in single particle.) This formalism is clearly non-relativistic (non-covariant) which can be easily seen by noting that. (Throughout the book we will use a bold symbol to represent a three dimensional quantity. 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. even for a free particle. H. non-relativistic physical system and promote each of the observables to a Hermitian operator.1) Here ψ(x. The time evolution of the quantum mechanical system. ∂t i (1.Chapter 1 Relativistic equations 1.2) with p denoting the momentum of the particle and V (x) representing the potential through which the particle moves.

even for a simple fundamental system such as the Hydrogen atom. A = (A1 . space and time are not treated on an equal footing in this case and. consequently. which we are all familiar with.5) ∂t 2m This equation is linear in the time derivative while it is quadratic in the space derivatives.) Thus. Consequently.6) . the equation cannot have the same form (covariant) in diﬀerent Lorentz frames. x2 . In this chapter. in this case. We note that in the three dimensional Euclidean space. (1. must treat space and time coordinates on an equal footing and remain form invariant in all inertial frames (Lorentz frames). there is a need to generalize the quantum mechanical description to relativistic systems. the momentum operator has the form i p → −i ∇. x = (x1 . A2 . 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.4) so that the time dependent Schr¨dinger equation. a vector is labelled uniquely by its three components. on the other hand. J3 ). the ground state electron is fairly relativistic ( v . for the ground state electron is c of the order of the ﬁne structure constant). x3 ).2 1 Relativistic equations p2 ∂ψ = ψ. (1.2 Notations Before proceeding any further.3) ∂t 2m In the coordinate basis. Therefore. A3 ). Let us also recall that. let us ﬁx our notations. takes o the form 2 ∂ψ =− ∇2 ψ. J2 . (We denote three dimensional vectors in boldface. (1. J = (J1 . i 1. A relativistic equation. (1.

(µ = 0. represents the metric of the Euclidean space and is trivial (in the sense that all the nonzero components are simply unity). 2. for any vector.8) The Kronecker delta. unlike the case of the Euclidean space. it is necessarily positive deﬁnite. 3 and we are being a little sloppy in representing the four vector by what may seem like its component) . (1. there are now two distinct four vectors that we can deﬁne on this manifold. these would now consist of four components. A2 ≥ 0. (1. namely.2 Notations 3 where x and J represent respectively the position and the angular momentum vectors while A stands for any arbitrary vector. However. consequently. namely.1. (1. as we know. we can again deﬁne vectors in this manifold. it does not matter whether we write the indices “up” or “down”. However.9) When we treat space and time on an equal footing and enlarge our three dimensional Euclidean manifold to the four dimensional spacetime manifold. any vector would also have four components. In such a space. A2 = 0. This also allows us to deﬁne the length of a vector simply as A2 = A · A = Ai Ai = δij Ai Aj = δij Ai Aj . any point in this manifold will be speciﬁed uniquely by four coordinates and. 1.7) where repeated indices are assumed to be summed. the scalar product of any two arbitrary vectors is deﬁned to be A · B = Ai Bi = δij Ai Bj = δij Ai Bj . in this case. Let us note from the deﬁnition of the length of a vector in Euclidean space that. Namely. if and only if A = 0. 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. δij . Consequently.

xµ = (ct. −. (1. we can immediately read out the metric tensors for the four dimensional Minkowski space which are diagonal with the signature (+. (1.11).4 1 Relativistic equations xµ = (ct.13) The contravariant metric tensor. 0 0 0 −1 η µν (1. On a more fundamental level.11) From the forms of the contravariant and the covariant vectors in (1. x). the two four vectors have distinct transformation properties under Lorentz transformations (in fact. commonly known as the Minkowski space. since they satisfy . 0 0 0 −1 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . xµ = η µν xν . −x). Namely. −). one transforms inversely with respect to the other) and are known respectively as contravariant and covariant vectors.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. are inverses of each other. and the covariant metric tensor.10) as well as using (1. we can write them in the matrix form as 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . namely. ηµν . −. xµ = ηµν xν . The contravariant and the covariant vectors are related to each other through the metric tensor of the four dimensional manifold. η µν .12) ηµν (1.

the nonuniqueness in the choice of the metric tensors reﬂects the nonuniqueness of the embedding of space and time components into a four vector. however.) . = η µν Aµ Bν = ηµν Aµ B ν (1. Diﬀerent authors. use diﬀerent metric conventions and you should be careful in reading the literature. however. (Two Lorentz indices are said to be contracted if a contravariant and a covariant index are summed over all possible values. (1. namely. are independent of the choice of a metric.) Given two arbitrary four vectors Aµ = (A0 . B). 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. B µ = (B 0 . (1. ηµν = ηνµ .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. (1. if there is no free Lorentz index.2 Notations 5 η µλ ηλν = δµν .1. η µν = η νµ .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. each is symmetric as they are expected to be. namely. such a product is easily seen to be invariant under Lorentz transformations. Physical results.17) Since the contravariant and the covariant vectors transform in an inverse manner.14) Furthermore. (As is clear from the above discussion. A). any product of Lorentz tensors deﬁnes a scalar if all the Lorentz indices are contracted. In fact.

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

u ˜ σ3 p3 u(p) + (E− + m)˜(p) = 0. (2. (2. 3 (2. v (p) = ˜ .18). (2.22) v (p) = − ˜ σ3 p 3 E+ + m 0 1 = 0 p3 E+ + m .9) becomes (E− − m)˜(p) + σ3 p3 v (p) = 0.) ˜ Similarly. Thus. ˜ E− − m (2.23) This determines the two positive energy solutions of the Dirac equation (remember that the energy eigenvalues are doubly degenerate). for the negative energy solutions we write p0 = E− = −E = − p 2 + m2 . we note that v must vanish in the rest frame while u remains ˜ ˜ arbitrary. for example.25) (2. Thus. ˜ v We can solve these as u(p) = − ˜ σ3 p 3 v (p). u can be thought of as the independent solution.24) and the set of equations (2.32 2 Solutions of the Dirac equation σ3 p 3 v (p) = − ˜ E+ + m and 1 0 = p3 −E + m + 0 . (The question of which components can be chosen independently follows from an examination of the dynamical equations.27) . from the second of the two two-component Dirac equations in (2.26) Choosing the independent solutions as 1 0 0 1 v (p) = ˜ .

32) . Thus.1 Plane wave solutions 33 we obtain respectively p3 −E − m − 0 σ3 p 3 u(p) = − ˜ E− − m and 1 0 = .29) and these determine the two negative energy solutions of the Dirac equation. 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 . (2.21) and (2. v (p) ˜ (2. (2. from the fact that we can write σ p ˜ − E 3−3m v (p) − . The independent two component wave functions in (2.27) are reminiscent of the spin up and spin down states of a two component spinor. (2.31) p3 −E − m − 0 ↑ u− (p) = . 0 1 (2.28) u(p) = − ˜ σ3 p 3 E− − m 0 1 = 0 p3 E− − m . 1 0 0 p3 ↓ u− = E− − m .2. − E+ + m 0 u↓ + = .30) u+ (p) = u(p) ˜ .

for the case of general motion. namely. u− (p) = β σ·p ˜ E + m u(p) σ· − E + p v (p) m ˜ v (p) ˜ u+ (p) = α . u− (p) = σ·p u(p) ˜ E+ + m σ·p ˜ E− − m v (p) .35) then. pi = −pi .) From the structure of the wave function.33) which can be explicitly veriﬁed. we can write the solutions for motion along a general direction as u(p) ˜ .) 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. 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− . where p1 = p2 = 0. it is also clear that. (2.21) and (2. the solutions take the forms (with p0 = E± = ± p2 + m2 ) u+ (p) = u(p) ˜ . (2. (The change in the sign in the dependent components comes from raising the index of the momentum. unlike the nonrelativistic counting (2s + 1).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. (We will determine the spin of 2 the Dirac particle shortly. (The correct counting for the number of components of the wave function for a massive.) 2. (2. The two component spinors u(p) and v (p) in (2.27) respectively are ˜ ˜ normalized as (for the same spin components) .36) Here α and β are normalization constants to be determined.34) v (p) ˜ (2.

Similarly. this product vanishes. Given this.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). ˜ v E +m . 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. ˜ 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 (2.2 Normalization of the wave function 35 u† (p)˜(p) = 1 = v † (p)˜(p). ˜ u ˜ v For diﬀerent spin components.2.38) where we have used the familiar identity satisﬁed by the Pauli matrices (σ · A)(σ · B) = A · B + iσ · (A × B). for the negative energy solutions we have σ·p v = β β −˜† (p) E + m ∗ (2.

Thus. another wave function (known as the adjoint spinor) that plays an important role is deﬁned to be u(p) = u† (p)γ 0 . 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. u† u = 1 = v † v .36 2 Solutions of the Dirac equation It is worth remarking here that although we have seen in (2. for the same spin components.42) −˜† (p) . (2.40) simply because we have not speciﬁed their 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. we have carried along ˜ ˜ ˜ ˜ these factors in (2. for example. v Thus. In dealing with the Dirac equation.38) and (2.37) that.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 . 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 .

40) (for the same spin components when (2. must be positive. the motivation for such a normalization condition comes from the fact that. this will reduce to u† u± = 1 which corresponds to ± the non-relativistic normalization. + − (2. u† u. as in the non-relativistic case.44) Our naive instinct will be to normalize the wave function. with this normalization condition.43) 2m |α|2 u† (p)˜(p).2 Normalization of the wave function 37 = |α|2 = 2m(E + m) † u (p)˜(p) ˜ u (E + m)2 (2.) The independent wave functions for a free particle. as we will see shortly.37) holds) . in the rest frame of the particle. we can show that 2m |β|2 v † (p)˜(p). In fact.47) With the requirement (2. − m u† (p)u+ (p) = + (2.38) and (2. by requiring (for the same spin components) u† (p)u+ (p) = 1 = u† (p)u− (p). would give (for the same spin components) E (2π)3 δ3 (p − p′ ). Thus. ˜ v E+m u− (p)u− (p) = − (2.45) However. a relativistically covariant normalization would be to require (for the same spin components) E = u† (p)u− (p). as we will see. we determine from (2. By the way.46). this is not a relativistic normalization. therefore. ˜ u E +m Similarly.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.2. ψp (x) = e−ik·x u(p) with p0 = ±E. m † d3 x ψp (x)ψp′ (x) = (2.

appears to be a Lorentz invariant (scalar) and we will see later that this is indeed true. we will obtain from (2.49) It is also clear that. Let us also note here that by construction the positive and the negative energy solutions are orthogonal. α = α∗ = 2E E |α|2 = . therefore. . E+m E + m 2m 2m 2m E + m |β|2 = − = −1.38 2 Solutions of the Dirac equation u† (p)u+ (p) = + or. 2m E 2E |β|2 = . v (p) ˜ (2.43) and (2. For example.48) or. E+m E + m 2m (2. u† (p)u− (p) = − E+m m β = β∗ = E+m . Therefore. σ·p ˜ E + m u(p) σ·p − E + m v (p) ˜ . with this normalization. 2m (2.50) u− (p)u− (p) = − This particular product. 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) ˜ .44) (for the same spin components) u+ (p)u+ (p) = 2m 2m E + m |α|2 = = 1. with this normalization. E+m m E+m .

˜ E+m σ·p u− (p)u+ (p) = −2β ∗ α v † (p) ˜ u(p) = 0. orthonormal solutions of the Dirac equation. let us note that it becomes meaningless for massless particles. ˜ E+m u+ (p)u− (p) = −2α∗ β u† (p) ˜ (2.2 Normalization of the wave function 39 u† (p)u− (p) + = α∗ β = 0. 2 (2. however. α = α∗ = 2E |α|2 = E. − + (2.) The probability density has to be well deﬁned.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 (2.54) or. Note. E+m E+m .40) (for the same spin components) u† (p)u+ (p) = + or. Correspondingly. we will obtain from (2. an alternative normalization which works well for both massive and massless particles is given by u† (p)u+ (p) = E = u† (p)u− (p). that σ·p v (p) = 0. u† (p)u− (p) = − E+m β = β∗ = E+m . .52) While we will be using this particular normalization for massive particles.2.51) Therefore.53) This still behaves like the time component of a four vector (m is a Lorentz scalar). In this case. the solutions we have constructed correspond to four linearly independent. (There is no rest frame for a massless particle. 2 2E |β|2 = E.

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 . The true normalization is really contained in the total probability.55) which vanishes for a massless particle. E +m 2m |β|2 = −m.3 Spin of the Dirac particle As we have argued repeatedly. ψ(x) = (2. (2. This product continues to be a scalar. u− (p)u− (p) = − E+m u+ (p)u+ (p) = (2. 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. d3 x ψ † (x)ψ(x) = 1.57) Namely. 2. That this is indeed true 2 can be seen explicitly as follows. Let us deﬁne a four dimensional generalization of the Pauli matrices as (in this section.40 2 Solutions of the Dirac equation Correspondingly. we obtain 2m |α|2 = m. we will use the notations of three dimensional . 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).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 note here parenthetically that. it can be understood in light of what we have already pointed out earlier as follows. in this case. while the arbitrariness in the normalization of u(p) may seem strange. Let us note once again that this is a particularly convenient normalization for massless particles.

3 Spin of the Dirac particle 41 Euclidean space since we will be dealing only with three dimensional vectors) σi 0 0 σi αi = ˜ . 2. 3. the generalized ˜ Pauli matrices.61) From the structures of the matrices αi and αi we conclude that ˜ [σi .60) We note that ρ2 = ½ so that we can invert the deﬁning relation (2. (Note. that this deﬁnes a reducible representation of spin generators since the matrices are block diagonal.2.58) It can. i = 1. α ˜ (2. σj ] 0 2iǫijk σk 0 0 [σi . (2. let us note that .59) and write αi = ραi = αi ρ.59) ρ= 0 ½ ½ 0 . αi . σj ] 0 2iǫijk σk = 2iǫijk αk .62) αi . of course.) Furthermore. be checked readily that this is related to the αi matrices deﬁned in (1. (2.101) through the relation 0 σi σi 0 αi = where = ρ˜ i = αi ρ. ˜ (2. αj ˜ ˜ = = This shows that 1 αi satisﬁes the angular momentum algebra (remem2˜ ber = 1) and this is why we call the matrices. ˜ (2. however.98) and (1.

In other words.100) (remember that we are using three dimensional Euclidean notations in this section) H = α · p + βm = αi pi + βm. σj ] 0 2iǫijk σk [σi . β ˜ = = 0. (2. ½] = 2iǫijk αk . σj ] = 2iǫijk σk 0 −[σi . Lorentz transformations deﬁne a symmetry of the Dirac Hamiltonian and.65) Calculating the commutator of this operator with the Dirac Hamiltonian. 2. the Dirac equation transforms covariantly under a Lorentz transformation. rotations which correspond to a subset of the Lorentz transformations must also be a symmetry of the Dirac Hamiltonian. Consequently. ½] 0 0 0 σi σj − σj σi 0 = [σi . we obtain . Let us recall that the orbital angular momentum operator is given by Li = ǫijk xj pk . i. (2. therefore.42 2 Solutions of the Dirac equation αi . αj ˜ = σi 0 0 σi 0 0 σj σj 0 − 0 σj σj 0 σi 0 0 σi = σi σj − σj σi 0 [σi .64) As we will see in the next chapter.63) With these relations at our disposal. (2. let us look at the free Dirac Hamiltonian in (1. αi . j. the angular momentum operators which generate rotations should commute with the Dirac Hamiltonian. k = 1. 3.

˜ 2 Ji = Li + (2.66) we obtain 1 αi . the total angular momentum which should commute with the Hamiltonian must contain a spin part as well. β m ˜ (2. H] + = iǫijk αj pk − iǫijk αj pk = 0.67) = 2iǫijk αk pj = −2iǫijk αj pk . we can identify the spin angular momentum operator with . αℓ pℓ ] = ǫijk αℓ [xj . H ˜ = = αi . H] = [ǫijk xj pk . αj pj + αi . αℓ pℓ + βm] = [ǫijk xj pk .68) In other words. αj pj + βm ˜ ˜ αi . if rotations are a symmetry of the system. Thus.69) In this case. (2. Consequently. we note that the orbital angular momentum operator does not commute with the Dirac Hamiltonian. therefore. To determine the spin angular momentum. H ˜ 2 Li + = [Li .2. the second term in the Hamiltonian drops out of the commutator. can be identiﬁed with 1 αi . the total angular momentum which should commute with the Hamiltonian.3 Spin of the Dirac particle 43 [Li . (2.66) Here we have used the fact that since β is a constant matrix and m is a constant. pℓ ]pk = ǫijk αℓ (iδjℓ ) pk = iǫijk αj pk . H ˜ 2 1 αi . we note that αi . so that combining this relation with (2.

4 Continuity equation The Dirac equation.72) with ψ † on the left and (2. we obtain ∂ψ † ∂ψ +i ψ = −i ψ † α · ∇ψ + (∇ψ † ) · αψ . i ∂ (ψ † ψ) = −i∇ · ψ † αψ . Therefore. that 1 1 α3 = ˜ 2 2 σ3 0 0 σ3 S3 = . (2. written in the Hamiltonian form. 2 2. ∂t ∂ (ψ † ψ) = −∇ · ψ † αψ .71) which has doubly degenerate eigenvalues ± 1 .74) This is the continuity equation for the probability current density associated with the Dirac equation and we note that we can identify . ∂t (2.72) Taking the Hermitian conjugate of this equation. ∂t −i (2. Multiplying (2. in particular.73) where the gradient is assumed to act on ψ † . ˜ 2 (2. ∂t ∂t iψ † or. is given by ∂ψ = Hψ = − iα · ∇ + βm ψ. we conclude 2 that the particle described by the Dirac equation corresponds to a spin 1 particle.73) with ψ on the right and subtracting the second from the ﬁrst.70) Note. or. we obtain ← − ∂ψ † = ψ † iα · ∇ + βm). ∂t i (2.44 2 Solutions of the Dirac equation Si = 1 αi .

must transform like the time coordinate under a Lorentz transformation. (2.75) (2. P = (2. ∂t This suggests that we can write the current four vector as J µ = (ρ. ρ. of course.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 = 0.78) This. J) = (ψ † ψ. in fact. the total probability d3 x ρ = d3 x ψ † ψ. ψ † αψ).79) is a constant independent of any particular Lorentz frame. therefore.76) so that the continuity equation can be written in the manifestly covariant form ∂µ J µ = 0. and note that the Hermitian conjugate of ψ satisﬁes (2.4 Continuity equation 45 ρ = ψ † ψ = probability density. It is worth recalling that we have already used this Lorentz transformation property of ρ in deﬁning the normalization of the wave function. shows that the probability density. to write the continuity equation as ∂ρ = −∇ · J. (We are.80) .77)) and. yet to show that j µ transforms like a four vector which we will do in the next chapter. (2.77) (2.) On the other hand. J = ψ † αψ = probability current density. is the time component of j µ (see (2.

82) where we have used the property of the gamma matrices that (for µ = 0. ← − ψ − iγ µ ∂µ − m = 0. (2. J = ψγψ = ψ † γ 0 γψ = ψ † αψ.86) (2. or. i∂µ ψγ µ ψ = 0. or. we obtain ← − ψ − iγ 0 (γ µ )† γ 0 ∂µ − m) = 0. we obtain ← − i ψγ µ ∂µ ψ + ψγ µ ∂µ ψ = 0.84) This is.46 2 Solutions of the Dirac equation ← − ψ † − i(γ µ )† ∂µ − m = 0. Note from the deﬁnition in (2.85) that J 0 = ψγ 0 ψ = ψ † γ 0 γ 0 ψ = ψ † ψ = ρ.82) with ψ on the right and subtracting the second from the ﬁrst. the covariant continuity equation and we can identify J µ = ψγ µ ψ.85) . which is what we had derived earlier in (2. 2.83) Multiplying (2.80) with ψ on the left and (2. ∂µ ψγ µ ψ = 0. 1. (2. γ 0 (γ µ )† γ 0 = γ µ . (2. in fact. (2. 3) γ 0 γ µ γ 0 = (γ µ )† . or.77).81) Multiplying this equation with γ 0 on the right and using the fact that (γ 0 )2 = ½. (2.

2. in such a case. 2. First of all.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). Even when the probability density is consistently deﬁned.1.87) Thus. We note from Fig. the presence of negative energy solutions leads to many conceptual diﬃculties. (2. This is rather diﬀerent from the case of the Klein-Gordon equation that we have studied in chapter 1. we note that the energy spectrum .1: Energy spectrum for a free Dirac particle.5 Dirac’s hole theory We have seen that Dirac’s equation leads to both positive and negative energy solutions. the probability density is independent of time derivative much like the Schr¨dinger equation. the probability density. because it is a ﬁrst order equation (particularly in the time derivative). In the free particle case.40). for example.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. even for this simple case of a free particle the energy spectrum has the form shown in Fig. 2. Consequently. m −m Figure 2. the energy eigenvalues are given by p0 = E± = ±E = ± p2 + m2 .38) and (2. as we o have seen explicitly in (2. can be deﬁned to be positive deﬁnite even in the presence of negative energy solutions. 2.

2 Since fermions obey Pauli exclusion principle. This simply means that one redeﬁnes the values of all these observables with respect to this ground state. unlike the conventional picture of the ground state as being the state without any particle (quantum). 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. Since physical systems have a tendency to go to the lowest energy state available. any given energy state can at any time accommodate at the most two electrons with opposite spin projections. In the case of Dirac particles. Classically. Taking advantage of this fact. therefore.) On the other hand. however. Namely. Furthermore. fermions. But as we have argued earlier within the context of the Klein-Gordon equation. For exam- . (Note that this would not work for a bosonic system such as particles described by the Klein-Gordon equation. It is only because fermions obey Pauli exclusion principle that this works for the Dirac equation. electromagnetic ﬁeld etc. Dirac postulated that the ground state in such a theory is the state where all the negative energy states are ﬁlled with electrons. quantum mechanically this is not acceptable. Dirac postulated that the physical ground state (vacuum) in such a theory should be redeﬁned for consistency. 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. even if we start out with a positive energy solution. Let us recall that the Dirac particles carry spin 2 and are. Thus. there is a way out of 1 this diﬃculty. it does predict some new physical phenomena which are experimentally observed. here the ground state.48 2 Solutions of the Dirac equation is unbounded from below. 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. destabilizing the physical system and leading to an ultimate collapse. For example.) This redeﬁnition of the vacuum automatically prevents the instability associated with matter. let us assume that the particles described by the Dirac equation are the spin 1 electrons. (Momentum and energy of these electrons are also assumed to be unobservable. contains an inﬁnite number of negative energy particles. any perturbation would cause the energy to lower. To be speciﬁc. we can restrict ourselves to the subspace of positive energy solutions. of course. Namely. in fact.

However. a negative energy electron can make a transition to a positive energy state and can appear as a positive energy electron. Furthermore. thereby. i = 1. were crucial in taking a matrix square root of the Einstein relation and. the Dirac theory predicts an anti-particle of equal mass for every Dirac particle. this is a general feature that combining quantum mechanics with relativity necessarily leads to a many particle theory. On the other hand. γ µ . 2. 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. This “hole” state is what we have come to recognize as the anti-particle – in this case. .6 Properties of the Dirac matrices The Dirac matrices.6 Properties of the Dirac matrices 49 ple. (γ i )† = −γ i . the four Dirac matrices satisfy (in addition to the Cliﬀord algebra) (γ 0 )† = γ 0 . 3. In this section. The amount of energy necessary to excite a negative energy electron to a positive energy state is E ≥ 2m. (2.88) Tr γ µ = 0.) This is Dirac’s theory of electrons and works quite well. if enough energy is provided to such a ground state. 1. 3. As we have seen. 2. µ = 0.) In spite of this. (This unconventional deﬁnition of a vacuum state can be avoided in a second quantized ﬁeld theory which we will study later. 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. the Dirac equation passes as a one particle equation primarily because of the Pauli exclusion principle. in deﬁning a ﬁrst order equation. we will study some of the useful properties of these matrices. 2.2. Thus. (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. a positron – and the process under discussion is commonly referred to as pair creation (production).

91) that we can identify this with the matrix ρ deﬁned earlier in (2. 4! (2.93) .92) We recognize from (2. 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 . Of course. Note that. (2. a complete set of Dirac matrices must consist of 16 such matrices. To obtain the other basis matrices.90) i ǫµνλρ γ µ γ ν γ λ γ ρ . (2.50 2 Solutions of the Dirac equation Since these are 4 × 4 matrices. † γ5 = γ5 . by deﬁnition.89) represents the four-dimensional generalization of the Levi-Civita tensor.91). Note that in our particular representation for the γ µ matrices given in (1. 2 γ5 = ½. let us deﬁne the following sets of matrices. the identity matrix will correspond to one of them. (2. Let γ5 = iγ 0 γ 1 γ 2 γ 3 = − where ǫ0123 = 1 = −ǫ0123 .91) where we have used the property of the Pauli matrices σ1 σ2 σ3 = i½.60). (2.

(2. since it is the product of all the four γ µ matrices.96) Finally. let us write out the forms of γ5 γ µ also in this representation. 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 . (2. we can also deﬁne six anti-symmetric matrices.97) = −i η µν − γ µ γ ν . it anticommutes with any one of them. Namely. 1.94) Given the matrix γ5 . 3) .2. we can deﬁne four new matrices as γ5 γ µ . σ µν . (2. γ ν = (γ µ γ ν − γ ν γ µ ) 2 2 (2. 2.95) Since we know the explicit forms of the matrices ½.6 Properties of the Dirac matrices 51 and that. = i η µν − γ ν γ µ whose explicit forms in our representation can be worked out to be (i. k = 1. 2. 3) σ µν = −σ νµ = i i γ µ . γ5 . γ µ + = 0. µ = 0. 3. ν = 0. γ µ and γ5 in our representation. ½ 0 = σi . as (µ. 1. 2. j.

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 . ǫijk ǫℓjk = 2 δiℓ . namely.100) Γ(T ) = Γ(P ) = = γ5 where the numbers on the right denote the number of matrices in each category and these. 4. (2. (This relation can be obtained from (2.98) 1 We have already seen in (2. 6. 1.98) we conclude. the notation is suggestive and stands for the fact that ψΓ(S) ψ .98) using the identity for products of Levi-Civita tensors. therefore. in fact.99) can be identiﬁed with the spin operators for the Dirac particle. γ5 . namely. provide a basis for all the 4 × 4 matrices.) We have thus constructed a set of sixteen Dirac matrices. γµ. Γ(S) Γ(V ) Γ(A) = = ½. 1. 4. σ µν . 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 . γµ.70) that the matrices 2 αi represent the ˜ spin operators for the Dirac particle. ¯ Here. that the matrices 1 1 αi = ǫijk σ jk . ˜ 2 4 (2. ˜ (2. From (2.

103) γ 0 σ µν γ 0 = The Dirac matrices satisfy nontrivial (anti) commutation relations.104) (2. has its own hermiticity property. (2.2. ¯ ¯ ¯ ¯ Similarly. We already know that . Finally. ψΓ(T ) ψ. V. axial-vector and a pseudo-scalar under a Lorentz and parity transformations as we will see in the next chapter. (2. A. Let us note here that each of the matrices. ψΓ(V ) ψ.6 Properties of the Dirac matrices 53 transforms like a scalar under Lorentz and parity transformations. tensor. 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 )† . In fact. which is deﬁned to be Hermitian. However. 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 )† . all other matrices satisfy γ 0 Γ(α) γ 0 = (Γ(α) )† . T. (2.101) where we have used the fact that γ5 is Hermitian and it anti-commutes with γ µ . γ 0 γ µ γ 0 = (γ µ )† . ψΓ(A) ψ and ψΓ(P ) ψ behave respectively like a vector. γ 0 γ5 γ µ γ 0 = −γ5 γ 0 γ µ γ 0 = −(γ5 )† (γ µ )† = (γ5 γ µ )† . it can be checked that except for γ5 .102) α = S. it follows easily that γ 0 ½γ 0 = (γ 0 )2 = ½ = (½)† . even within a given class.

γ ν γ λ = i γµ. γν + + = 2i η µν γ λ − η µλ γ ν . For example. C + . Similarly. −i η νλ − γ ν γ λ γλ − γν γµ. = 0. In this derivation.107) We note here parenthetically that the commutator in (2. we obtain . BC] = ABC − BCA = (AB + BA)C − B(AC + CA) = A. C] . γν γ5 . (2.105) We can also calculate various other commutation relations in a straightforward and representation independent manner. γ µ + + = 2η µν ½. since γ µ matrices satisfy simple anti-commutation relations. B] C + B [A. σ νλ = γ µ . B + C − B A. the form in (2. γ µ . BC] = [A.107) is more useful for our purpose.54 2 Solutions of the Dirac equation γµ. (2.106) [A. (2. γλ = i γ µ.107) can also be expressed in terms of commutators (instead of anti-commutators) as [A. we have used the fact that (2.108) However. for the commutator of two σ µν matrices.

This becomes particularly useful in calculating various amplitudes involving Dirac particles. γ µ = 2η νµ γµ − 4γ = 2γ ν − 4γ ν = −2γ ν . σ λρ = i γ µ γ ν . Similarly. + ν − γµγν (2. σ λρ = iγ µ γ ν . (2. We will see in the next chapter that they provide a representation for the Lorentz algebra. where we have used (γµ = ηµν γ ν ) γµ γ µ = 4 ½. 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.110) (2. The various commutation and anti-commutation relations also lead to many algebraic simpliﬁcations in dealing with such matrices.111) (2. (these relations are true only in 4-dimensions) γµ γ ν γ µ = γµ γ ν . and it follows now that. σ λρ + i γ µ . for example. 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 .112) . σ λρ = − i η µν − γ µ γ ν .2.6 Properties of the Dirac matrices 55 σ µν . γµ A µ = γµ Aν γ ν γ µ = Aν γµ γ ν γ µ = −2Aν γ ν = −2A /γ /. Thus.109) Thus.

it follows (in 4-dimensions) that 1 1 Tr γ µ γ ν + Tr γ ν γ µ = Tr γ µ . For example. For example.115) Even more complicated traces can be evaluated by using the basic relations we have developed so far. γ µ + − γµγλ = 4η λν ½. γ ρ or.114) Tr γ µ γ ν + Tr γ5 γ µ (2.113) = 2η λµ γµ γ ν + 2γ ν γ λ = 2γ λ γ ν + 2γ ν γ λ = 2 γ λ . γ ν + and so on.116) . (2. 2 Tr γ µ γ ν γ λ γ ρ = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ . we note that Tr γ µ γ ν γ λ γ ρ = Tr γµ. − γργµ Tr γ µ γ ν γ λ γ ρ = 4 η µν η λρ − η µλ η νρ + η νλ η µρ .56 2 Solutions of the Dirac equation γµ γ ν γ λ γ µ = γµ γ ν γ λ . γ ν 2 2 1 = Tr 2η µν ½ = η µν Tr ½ = 4η µν . 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. = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ − Tr γ ν γ λ γ ρ γ µ . Therefore. 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 . we know from the cyclicity of traces that Tr γ µ γ ν = Tr γ ν γ µ . or. γ λ + = 8η µν η λρ − 8η µλ η νρ + Tr γ ν γ λ γ µ . (2.

(2. 1 γM = . 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 = . there exists a representation for the Dirac matrices where γ µ are all purely imaginary. there are other equivalent representations possible which may be more useful for a particular system under study. Explicitly.119) σ µ = (½. −σ). For example. We would use all these properties in the next chapter to study the covariance of the Dirac equation under a Lorentz transformation. It can be checked . (2.2. there exists yet another representation for the γ µ matrices. 3 γM = . let us note that we have constructed a particular representation for the Dirac matrices commonly known as the Pauli-Dirac representation. σ). 0 σµ ˜ σµ 0 γW = where µ . ˜ (2. σ µ = (½.117) 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 . namely. (2.6 Properties of the Dirac matrices 57 and so on. To conclude this section. This is known as the Majorana representation and is quite useful in the study of Majorana fermions which are charge neutral fermions. 2 γM = .120) This is known as the Weyl representation for the Dirac matrices and is quite useful in the study of massless fermions. 1 S = √ γ0 2 ½ + γ2 .118) Similarly. However.

(2. it follows now that if .1 Fierz rearrangement.100). We have explicitly constructed the sixteen matrices to correspond to the set Γ(a) = {½. This is easily demonstrated by showing that they are linearly independent which is seen as follows. γµ . (2. For example. the linear independence of the set of matrices in (2. given this set of matrices. γ5 } . σ µν . (2. γ5 γ µ .121) 2. the sixteen Dirac matrices Γ(a) .125) Γ(a) = (2.126) With this.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 . 4 (b) (2. (2. As a result.124) Tr Γ(a) Γ(b) = δ(a) . it can be easily checked that Tr Γ(a) Γ(b) = 0. Explicitly. we can construct the inverse set of matrices as Γ(a) .122) From the properties of the γ µ matrices. −γ5 γµ .122) is straightforward. T.6.123) where “Tr” denotes trace over the matrix indices. a = b. Tr Γ(a) Γ(a) Γ(a) = such that a not summed. V. 1 S=√ 2 ½ + γ5 γ 0 . As we have pointed out in (2. a = S. A. γ µ . P deﬁne a complete basis for 4 × 4 matrices. γ5 } . we can write the inverse set of matrices as 1 {½. σµν .

127) with Γ(b) .127) implies that all the coeﬃcients of expansion must vanishing which shows that the set of sixteen matrices Γ(a) in (2. C(a) δ(b) = 0. V. (M ) (2. As a result they constitute a basis for 4 × 4 matrices. (2. M= (a) C(a) Γ(a) .6 Properties of the Dirac matrices 59 C(a) Γ(a) = 0.122) are linearly independent. where b is arbitrary. and taking trace over the matrix indices and using (2. A.130) into the expansion (2. (a) C(a) Tr Γ(b) Γ(a) = 0. T. we obtain . (2. (a) (a) or. multiplying (2. (a) Tr Γ(b) or.130) C(b) = Tr Γ(b) M .122) provide a basis for the 4 × 4 matrix space. any arbitrary 4 × 4 matrix M can be expanded as a linear superposition of these matrices.129) Multiplying this expression with Γ(b) and taking trace over the matrix indices. C(b) = 0.128) for any b = S. Since the set of matrices in (2. Substituting (2. namely.125) we obtain C(a) Γ(a) = 0. or.129). Therefore. (2. P .127) then. we obtain C(a) Tr Γ(b) Γ(a) = (a) (a) (M ) (M ) (a) Tr Γ(b) M = or. (a) (2.2. (M ) C(a) δ(b) .

For example. we note that if M and N denote two arbitrary 4 × 4 matrices. it can be used eﬀectively in many ways.132) describes a fundamental relation which expresses the completeness relation for the sixteen basis matrices.132) we can derive (for simplicity.132) Here α.60 2 Solutions of the Dirac equation M= (a) Tr Γ(a) M Γ(a) . 4 and correspond to the matrix indices of the 4 × 4 matrices and we are assuming that the repeated indices are being summed. 3. η = 1. (2. this leads to Mαβ = (a) Γ(a) (a) γη Mηγ Γαβ .131) Introducing the matrix indices explicitly.133). ¯ ¯ ¯ ¯ (2. 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 . (a) Γ(a) γη Γ(a) αβ = δαη δβγ . ¯ ¯ ¯ ¯ 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 . (2. then using (2. Equation (2. or. Just like any other completeness relation.133) Using the relations in (2. it is now straightforward to obtain . γ. 2. β.

we have assumed that the spinors are ordinary functions.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 .134) are known as the Fierz rearrangement identities which are very useful in calculating cross sections.124) respectively. On the other hand.135) Since this is true for any matrices M.134) pick up a negative sign which arises from commuting the fermionic ﬁelds past one another.135) equivalently as . Let us note that using the explicit forms for Γ(a) and Γ(a) in (2.6 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 . we can write the ﬁrst of the Fierz rearrangement identities in (2. N and any spinors.122) and (2. (2.2.134) The two relations in (2. if they correspond to anti-commuting fermion operators. (2. we can ˜ deﬁne a new spinor ψ4 = N ψ4 to write the identity in (2. the right-hand sides of the 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 . In deriving these identities.

137) then using various properties of the gamma matrices derived earlier as well as (2. (2. we obtain from (2. I. Quantum Mechanics.136) Thus. Schiﬀ. New York.113). Relativistic Quantum Mechanics. Drell. (2. if we which is often calculationally simpler. McGraw-Hill. 1968.62 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 . 2. J.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). N = (½ − γ5 ) γµ . McGraw-Hill. choose M = (½ − γ5 ) γ µ . Bjorken and S. for example.7 References 1.110) and (2. New York. D. L. 1964. . 2.136) ¯ ¯ ψ1 (½ − γ5 ) γ µ ψ2 ψ3 (½ − γ5 ) γµ ψ4 ¯ ¯ = −ψ1 (½ − γ5 ) γ µ ψ4 ψ3 (½ − γ5 ) γµ ψ2 . (2.

Okubo. McGrawHill. Lectures on Quantum Mechanics. A. . 1860 (1958). Das. (1957).7 References 63 3. C. C. G. Itzykson and J-B. Hindustan Publishing. 4. New Delhi. Sudarshan and R. Quantum Field Theory. E. Proceedings of PaduaVenice conference on mesons and newly discovered particles. S. Zuber. Real representations of ﬁnite Cliﬀord algebras. Classiﬁcation. New York. I. Marshak.2. 2003. 1657 (1991). Journal of Mathematical Physics 32. India. 1980. 5. E. Physical Review 109. 6.

This page intentionally left blank .

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

(3. 2 (3. with unit determinant. 1 x′ = x3 .7) We recognize that (3. We observe here that the matrix representing the inﬁnitesimal change under a rotation is anti-symmetric.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. x′ 1 = −γβx0 + γx1 . x′ 2 = x2 . 3 x′ = ǫx1 + x2 . such that (3. (3.4) where we have identiﬁed θ = ǫ = inﬁnitesimal. x′ 3 = x3 .6) can also be written in the matrix form as .6) where the Lorentz factor γ is deﬁned in terms of the boost velocity to be 1 1 − β2 γ= .2) that an inﬁnitesimal rotation around the 3-axis (z-axis) takes the form x′ = x1 − ǫx2 . Under a Lorentz boost along the x-axis. we also know that the coordinates transform as (boost velocity v = β since c = 1) xµ → x′ µ . R (RT R = ½).5) x′ 0 = γx0 − γβx1 . We also note from (3.

12) . . where cosh ω − sinh ω 0 0 cosh ω 0 0 (3.9) that −∞ ≤ ω ≤ ∞.11) 1 β2 − = 1. Thus.9) 1 − β2 cosh2 ω − sinh2 ω = Since the range of the boost velocity is given by −1 ≤ β ≤ 1. = 0 0 1 0 x2 0 0 0 1 x3 x′ 0 γ (3. 1 − β2 1 − β2 (3. (3.8) 1 1 − β2 β .3. we note that a Lorentz boost along the x-direction can be written as x′ µ = Λµν xν .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 .10) Λµν − sinh ω = 0 0 0 0 . we conclude from (3. 1 0 0 1 (3.

Lorentz boosts correspond to noncompact transformations unlike space rotations. the “angle” of rotation. (Incidentally. (or the parameter of boost) can take any real value and.14) sinh ω = 0 0 0 0 cosh ω 0 0 . = 0 0 1 0 0 0 0 1 1 Λµν (3.12) or (3. µ (3. is easily seen from (3. (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. we can obtain. as we have seen.13) The matrix representing the Lorentz transformation of the coordinates.3). namely.) Therefore. (3. ω. 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).15) and also has a unit determinant. 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). 0 1 0 0 0 1 sinh ω (3.) Furthermore. much like the rotation matrix R in (3.13) to be an ν orthogonal matrix in the sense that µ (ΛT )µν Λνλ = Λνµ Λνλ = δλ . x′ = Λµν xν .16) . then we can write (ω = ǫ = inﬁnitesimal) −ǫ 0 0 −ǫ 1 0 0 = δµν + ǫµν .68 From this. (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. as a result. Λµ (or Λµν ).

18) In other words. namely. which can be thought of as rotations in the four dimensional space-time. or. Λµρ Λµσ = ηρσ . = 0 0 0 0 0 0 0 0 0 ǫµν = η νλ ǫµλ (3. or. General Lorentz transformations are deﬁned as the transformations x′ µ = Λµν xν . ηµν Λµρ xρ Λνσ xσ = ηµν xµ xν . (3.1 Lorentz transformations 69 where.3. or. we note that we can combine rotations and Lorentz boosts into what are known as the homogeneous Lorentz transformations. ηµν Λµρ Λνσ xρ xσ = ηρσ xρ xσ .20) . ρ Λµρ Λµσ = δσ .19) ηµν x′ µ x′ ν = ηµν xµ xν . which leave the length of the vector invariant. the matrix representing the change under an inﬁnitesimal Lorentz boost is anti-symmetric just like the case of an inﬁnitesimal rotation. 0 −ǫ 0 0 0 0 0 ǫµν −ǫ = 0 0 0 0 . or. (3. 0 0 0 0 (3. In a general language. therefore.17) It follows from this that ǫ 0 0 −ǫ 0 0 0 = −ǫνµ .

such a Lorentz transformation does not change the direction of time.25) . therefore. (3. or.24) µ ν (3. orthogonal means the corner that is straight up (perpendicular). In the same spirit. Λ00 ≤ −1. what we have seen before in (3.70 3 Properties of the Dirac equation This is. (The Greek preﬁx “ortho” means straight up. (3. we note that ΛT as can be seen from (3. or.22) If Λ00 ≥ 1. we conclude that Λ00 ≥ 1. Λ00 Λ00 2 2 − Λi 0 2 = 1.20) explicitly as Λ00 Λ00 + Λi 0 Λi 0 = 1. of course. Thus. or. Incidentally.23) = Λνµ and ΛT ν µ = Λνµ The set of homogeneous Lorentz transformations satisfying Λ00 ≥ 1. orthochronous means straight up in time.) Note also that since ΛT Λ = ½.20)) (det Λ)2 = 1.21) Therefore. (3. 2 = 1 + Λi 0 ≥ 1. det Λ = ±1. Choosing ρ = σ = 0. Namely. we can write out the relation (3. Lorentz transformations deﬁne the maximal symmetry of the space-time manifold which leaves the origin invariant.15). and (3. or. an orthodontist is someone who can make your teeth straight. “gonia” in Greek means an angle or a corner and. we obtain (for clarity. then the transformation is called orthochronous. det Λ = 1.

then we obtain a Lorentz transformation −1 0 0 0 0 1 0 0 Λprop . = 0 0 −1 0 0 0 0 −1 Λ = Λspace Λprop (3. Finally. det Λ = ±1. det Λ = −1 (which is orthochronous but no longer proper). to a proper Lorentz tranformation. xµ → −xµ .27) This would correspond to having Λ00 ≥ 1. there are four kinds of Lorentz transformations.28) satisfying Λ00 ≤ −1 and det Λ = −1 (which is neither proper nor orthochronous). if we add both space and time reﬂections. orthochronous Lorentz transformations and constitute a set of continuous transformations that can be connected to the identity matrix.26) Λ00 ≤ −1. 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. = 0 0 1 0 0 0 0 1 Λ = Λtime Λprop (3.3. (Just to emphasize. we can obtain the other Lorentz transformations by simply appending space reﬂection or time reﬂection or both (which are discrete transformations). If we add time reversal. to a proper orthochronous Lorentz transformation. however. we obtain a Lorentz transformation . Λ00 ≥ 1. (3.1 Lorentz transformations 71 are known as the proper. we obtain a Lorentz transformation 1 0 0 0 0 −1 0 0 Λprop . namely. Thus. t → −t.) In general. then by adding space reﬂection. Given the proper orthochronous Lorentz transformations. if Λprop denotes a proper Lorentz transformation. det Λ = ±1. x → −x.

let us now consider the Dirac equation iγ µ ∂µ − m ψ(x) = 0.30) −1 0 0 0 −1 0 = 0 0 −1 0 0 0 0 Λprop . (3. These additional transformations.29) where L is a linear operator. In simple terms. we would refer to proper orthochronous Lorentz transformations as the Lorentz transformations. Under a Lorentz transformation xµ → x′ µ = Λµν xν .31) where ψ ′ represents the transformed wavefunction and L′ stands for the transformed operator (namely.33) (3. (3. the operator L. however. covariance implies that a given equation is form invariant under a particular transformation (has the same form in diﬀerent reference frames). 3. we say that it is covariant under a given transformation provided the transformed equation has the form L′ ψ ′ = 0. In these lectures.2 Covariance of the Dirac equation Given any dynamical equation of the form Lψ = 0.72 3 Properties of the Dirac equation Λ = Λspace−time Λprop with Λ00 ≤ −1 and det Λ = 1 (which is proper but not orthochronous). (3. With this general deﬁnition. 0 −1 0 (3. cannot be continuously connected to the identity matrix since they involve discrete reﬂections.32) . with the transformed variables).

are a set of four space-time independent matrices and.2 Covariance of the Dirac equation 73 if the transformed equation has the form ′ iγ µ ∂µ − m ψ ′ (x′ ) = 0. Parenthetically. (3. Thus. then the Dirac equation would be covariant under a Lorentz transformation. 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 .36) However.3.34) where ψ ′ (x′ ) is the Lorentz transformed wave function. do not change under a Lorentz transformation. γ µ . we can deﬁne the Lorentz transformation acting on the Dirac Hilbert space (Hilbert space of states describing a Dirac particle) as. therefore. (3. under a Lorentz transformation. (3. (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) . since ψ(x) is a four component spinor.35) where S(Λ) is a 4 × 4 matrix. the transformed wavefunction has the form ψ ′ (x′ ) = ψ ′ (Λx) = S(Λ)ψ(x). what this means is that we are ﬁnding a representation of the Lorentz transformation on the Hilbert space. In the notation of other symmetries that we know from studies in non-relativistic quantum mechanics. Let us assume that. Note that the Dirac matrices. the Lorentz transformation can also mix up the spinor components (much like angular momentum/rotation). since the Dirac wavefunction is a four component spinor. in addition to the change in the coordinates.

′ iΛµν Sγ ν S −1 ∂µ − m ψ ′ (x′ ) = 0. Let us also note from (3.37) we obtain (see (3. but get rotated by the S(Λ) matrix. Since the Lorentz transformations are invertible. (3.74 = 3 Properties of the Dirac equation dx |Λx S(Λ)ψ(x).38) Namely.41) ∂x′ ν ∂ − m S −1 (Λ)ψ ′ (x′ ) = 0.33) that ∂x′ µ = Λµν . (3.39) (3. the eﬀect of the Lorentz transformation.35) we can write ψ(x) = S −1 (Λ)ψ ′ (x′ ).42) . or. so that. iγ µ (3.35)) ψ ′ (x′ ) = x′ |ψ ′ = S(Λ)ψ(x). or. or. (3. or.37) where the wave function is recognized to be ψ(x) = x|ψ .40) (3. the matrix S(Λ) must possess an inverse so that from (3. Thus. on the wave function. ∂xν deﬁne a set of real quantities. ′ iΛµν γ ν ∂µ − m S −1 (Λ)ψ ′ (x′ ) = 0. from (3. can be represented by a matrix S(Λ) which depends only on the parameter of transformation Λ and not on the space-time coordinates. ∂xµ ∂x′ ν iγ µ Λνµ ∂ν′ − m S −1 (Λ)ψ ′ (x′ ) = 0. A more physical way to understand this is to note that the Dirac wave function simply consists of four functions which do not change. we can write iγ µ ∂µ − m ψ(x) = 0.

γ µ and γ ′µ . It now follows from (3.42) that the Dirac equation will be form invariant (covariant) under a Lorentz transformation provided there exists a matrix S(Λ).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. (3.45) where we have used the orthogonality of the Lorentz transformations (see (3. we see from (3. Λνσ γ σ (3. γ 0 (S −1 γ µ S)† γ 0 = S −1 γ µ S. or.15)). γ ′ µ. (Λ∗ )µν = Λµν ) γ 0 Λµν γ ν γ 0 = Λµν γ 0 (γ ν )† γ 0 = Λµν γ ν .3.44) + = + + µν = Λµρ Λνσ γ ρ . such that Λµν Sγ ν S −1 = γ µ . Λµν γ ν = S −1 γ µ S. γ σ = 2Λµρ Λνρ ½ = 2η = Λµρ Λνσ 2η ρσ ½ ½. generating Lorentz transformations (for the Dirac wavefunction). or. (3. then. by Pauli’s fundamental theorem. let us note that since the parameters of Lorentz transformation are real (namely. or.2 Covariance of the Dirac equation 75 where we have used (3. Next.43) Let us note that if we deﬁne γ ′ µ = Λµν γ ν . Sγ 0 S † γ 0 γ 0 γ µ γ 0 γ 0 S −1 γ 0 S −1 = γ µ . Therefore. γ ′ ν Λµρ γ ρ . or.46) † † † .40). there must exist a matrix connecting the two representations. (Sγ 0 S † γ 0 )γ µ (Sγ 0 S † γ 0 )−1 = γ µ . the matrices γ ′ µ satisfy the Cliﬀord algebra and. Therefore. (3.

or.51) In fact.100) and calculating the commutator with γ µ ).47) Taking the Hermitian conjugate of (3. b = b∗ . therefore.46) that the matrix Sγ 0 S † γ 0 commutes with all the γ µ matrices and. we can denote Sγ 0 S † γ 0 = b ½. (3. (3.50) The real roots of this equation are b = ±1.48) which. det S = 1. As a result. Sγ 0 S † γ 0 = b∗ ½ = b ½. determines that the parameter b is real. we can determine the unique value of b in the following way. b4 = 1. namely. 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 (2.76 3 Properties of the Dirac equation Here we have used (3. (3. we determine det (Sγ 0 S † γ 0 ) = det (b ½). therefore. or. (3. γ 0 Sγ 0 S † = b∗ ½.47).47). S † γ 0 = b γ 0 S −1 .43) and the relations (γ 0 )† = γ 0 = (γ 0 )−1 as well as γ 0 (γ µ )† γ 0 = γ µ . (3. γ 0 (γ 0 Sγ 0 S † )γ 0 = b∗ ½. or. It is clear from (3. or. .49) We also note that det γ 0 = 1 and since we are interested in proper Lorentz transformations. or. Using these in (3.

Thus. Tr S † S = 4bΛ00 0. 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 . . or. we are assuming Λ00 ≥ 1. we conclude from (3. or. S † γ 0 = γ 0 S −1 . b < 0.56) (3. (3.51) that b = 1. (3. The two solutions of this equation are obvious Λ00 ≥ 1.54) which implies (see (3.53) or.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. using (3. it follows from (3. Since we are dealing with proper Lorentz transformations. b > 0. (3.47) that Sγ 0 S † γ 0 = ½. (3.47).3.52) which follows since S † S represents a non-negative matrix.2 Covariance of the Dirac equation 77 Let us note. Λ00 ≤ −1.43) and (3.53)) that b > 0 and. therefore.

60) ½+ ½+ i Mµν ǫµν 4 ½− i Mστ ǫστ 4 i i Mµν ǫµν − Mµν ǫµν + O(ǫ2 ) 4 4 (3. We can also write S −1 (ǫ) = ½ + so that S −1 (ǫ)S(ǫ) = = = i Mµν ǫµν .57) From our earlier discussion in (3. Mµν = −Mνµ . let us consider an inﬁnitesimal Lorentz transformation of the form (ǫµν inﬁnitesimal) x′µ = Λµν xν = (δµ + ǫµ )xν = xµ + ǫµ xν .18).58) For an inﬁnitesimal transformation. 4 (3. we recall that the inﬁnitesimal transformation matrix is antisymmetric.78 3 Properties of the Dirac equation Next.59) where the matrices Mµν are assumed to be anti-symmetric in the Lorentz indices (for diﬀerent values of the Lorentz indices. namely. 4 (3.61) (3. we can expand the matrix S(Λ) as S(Λ) = S(ǫ) = ½ − i Mµν ǫµν . therefore. ǫµν = −ǫνµ .62) (3. ν ν ν (3. . (3.63) ½ + O(ǫ2 ). Mµν denote matrices in the Dirac space). since ǫµν = −ǫνµ .

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

69) with the generators of inﬁnitesimal Lorentz transformations for the 1 Dirac wave function. (The other factor of 2 is there to avoid double counting. we have shown that there exists a S(Λ) which satisies (3. at least for inﬁnitesimal Lorentz transformations.43) and generates Lorentz transformations and as a result. ˜ 2 2 2 (3. In such a case (see also (2. For completeness. the Dirac equation is form invariant (covariant) under such a Lorentz transformation. Thus.98)).70) with all other components of ǫµν vanishing. 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. of course. (3. 2 (3. A ﬁnite transformation can. let us note that inﬁnitesimal rotations around the 3-axis or in the 1-2 plane would correspond to choosing ǫ12 = ǫ = −ǫ21 .109) which the generators of the inﬁnitesimal transformations.80 3 Properties of the Dirac equation Let us note here from the form of S(ǫ) that we can identify 1 σµν .) We will see in the next chapter (when we study the representations of the Lorentz group) that the algebra (2. 1 σµν . satisfy can 2 be identiﬁed with the Lorentz algebra (which also explains why they are closed under multiplication). then.72) . i (3. be obtained from an inﬁnite sequence of inﬁnitesimal transformations resulting in an exponentiation of the inﬁnitesimal generators as ˜ S(θ) = e− 2 α3 θ . be constructed out of a series of inﬁnitesimal transformations and. S(ǫ) = ½ − i 1 i σ12 ǫ = ½ + γ1 γ2 ǫ = ½ − α3 ǫ.71) A ﬁnite rotation by angle θ in the 1-2 plane would. consequently.

3. (3. therefore. 2 ½− (3. α 2 2 (3.74) and. In this case. θ θ − i˜ 3 sin α 2 2 = −S(θ).75) That is.73) α3 ˜ = ½ 0 = ½. so that we can write (see also (2.78) . Furthermore. of course. Let us next consider an inﬁnitesimal rotation in the 0-1 plane. we can identify ǫ01 = ǫ = −ǫ10 . therefore.76) θ θ − i˜ 3 sin .98)) S(ǫ) = i 1 σ01 ǫ = ½ + γ0 γ1 ǫ 2 2 1 = ½ − α1 ǫ. (3. we are considering an inﬁnitesimal boost along the 1-axis. recalling that σ3 0 0 σ3 α3 = ˜ we have 2 . namely. is double valued and. the rotation operator. in this case. corresponds to a spinor representation. (3.2 Covariance of the Dirac equation 81 Note that since α† = αi . we have S † (θ) = S −1 (θ). namely. we can determine S(θ) = cos This shows that S(θ + 2π) = − cos S(θ + 4π) = S(θ). This is.77) with all other components of ǫµν vanishing. consistent with the fact that the Dirac equation describes spin 1 2 particles. rotations ˜i ˜ deﬁne unitary transformations. 0 ½ (3.

ω.80) and.82) We note here that since α† = α1 . 3. α2 = 1 ½ 0 = ½. 1 S † (ω) = S −1 (ω). recalling that 0 σ1 σ1 0 1 (3. can be obtained through exponentiation as S(ω) = e− 2 α1 ω . as 4 × 4 matrices). Let us note next that. since under a Lorentz transformation . 2 2 S(ω) = cosh (3. (3. All the unitary representations are necessarily inﬁnite dimensional.82 3 Properties of the Dirac equation In this case.79) α1 = . This is related to the fact that Lorentz boosts are non-compact transformations and for such transformations.81) we can determine ω ω − α1 sinh . Furthermore. (3.3 Transformation of bilinears In the last section. 0 ½ (3. the matrix for a ﬁnite boost. we have shown how to construct the matrix S(Λ) for ﬁnite Lorentz transformations (for both rotations and boosts). operators deﬁning boosts are not unitary.83) That is. there does not exist any ﬁnite dimensional unitary representation. therefore. in this four dimensional space (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). it would transform as a four vector under a proper Lorentz transformation. it follows that ψ ′ † (x′ ) = ψ † (x)S † (Λ). (3. ′ (3.86) ′ ψ(x)ψ(x) → = Namely.3 Transformation of bilinears 83 ψ ′ (x′ ) = S(Λ)ψ(x).50) and (2. namely. we see that the adjoint wave function ψ(x) transforms inversely compared to the wave function ψ(x) under a Lorentz transformation.56).88) .3. J µ (x) = ψ(x)γ µ ψ(x) → Λµν J ν (x). 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).55)). (3.84) (3. ψ (x′ ) = ψ ′ † (x′ )γ 0 = ψ † (x)S † (Λ)γ 0 = ψ † (x)γ 0 S −1 (Λ) = ψ(x)S −1 (Λ).87) ′ where we have used (3.43). (3. Thus. 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. we see that if we deﬁne a current of the form j µ (x) = ψ(x)γ µ ψ(x). Similarly. In other words.

(3.4 Projection operators. (3. 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) = .85)) transforms like a four vector so that the probability density transforms as the time component of a four vector.91). E+m 2m σ ·p E+m v ˜ v ˜ v(p) = u− (−p0 . −p) = u− (−p) = .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. Finally.49). p) = 0.93) . 3. / (3. (3. 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 . we note that in this way. (3. of course. / where p0 = ω = p2 + m2 .91) with the same value of p0 as in (3. p) = 0.90) (3.89) and (3.90).92) so that the equations satisﬁed by u(p) and v(p) (positive and negative energy solutions). It is customary to identify (see (2. the probability current density (see also (2. what we had observed earlier. can be written as (p − m)u(p) = 0.89) while the negative energy solutions satisfy − γ 0 p0 − γ · p − m u− (−p0 .

(3. 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. we can write down the Lorentz invariant conditions we had derived earlier from the normalization of a massive Dirac particle as (see (2.50)) 4 u r (p)us (p) = α=1 4 r uα (p)us (p) = δrs . completeness relation 85 and − γ 0 p0 + γ · p − m v(p) = 0. 2. / (p + m)v(p) = 0. or. / or. r = 1. can represent the spin projection of the two component spinors (in terms of which the four component solutions were obtained).3. u (p)v (p) = 0 = v r (p)us (p). Let us denote them by ur (p) and v r (p). u(p)(p − m) = 0. / (3. 4. / v † (p) (p)† + m γ 0 = 0.94) Given these equations.95) where we have used (γ µ )† γ 0 = γ 0 γ µ (see (2. Let us also note that each of the four solutions really represents a four component spinor. 2.83)). / (3.4 Projection operators. α v r (p)v s (p) = α=1 r s r s v α (p)vα (p) = −δrs . / or. or. Let us denote the spinor index by α = 1. as we had seen earlier. (3. As we have seen earlier there are two positive energy solutions and two negative energy solutions of the Dirac equation.97) . With these notations. 3.96) where r. (−p − m)v(p) = 0. v(p)(p + m) = 0.

4 × 4 matrices and their eﬀect on the Dirac spinors is quite clear. (3.94). it is clear that we can deﬁne projection operators for such solutions as p+m / . namely.86 3 Properties of the Dirac equation Although we had noted earlier that u+ (p)u− (p) = 0.52)). 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. 2m −p + m r / u (p) = 0.97) can be checked to be true simply because v(p) = u− (−p0 . (3.93) and (3. 2m −p + m / . From the form of the equations satisﬁed by the positive and the negative energy spinors.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). This also allows us to write 2 r=1 [u r (p)ur (p) − v r (p)v r (p)] = 4. the last relation in (3.100) Λ− (p) = 2m These are. 2m (3. −p). 2m p+m r / v (p) = 0. (3. (3.99) for any two spin components of the positive and the negative energy spinors. of course.101) .

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. 2m −p + m / 2m 2 = p2 − 2mp + m2 / 4m2 2m(−p + m) / m2 − 2mp + m2 / = 2 4m 4m2 −p + m / = Λ− (p).103) as it should be since all the solutions can be divided into either positive or negative energy ones. 2m / p + m −p + m / × 2m 2m 1 1 −p2 + m2 = −m2 + m2 2 4m 4m2 0 = Λ− (p)Λ+ (p). 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). let us also note that p + m −p + m / / + = ½. Let us next consider the outer product of the spinor solutions. we see that / Λ± (p) are indeed projection operators and they are orthogonal to each other.102) where we have used (p)2 = p2 and p2 = m2 . (3. Let us deﬁne a 4 × 4 matrix P with elements . while Λ− (p) projects only on to the space of negative energy ones.4 Projection operators. 2m 2m Λ+ (p) + Λ− (p) = (3.3. Thus. Furthermore.

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

(3.108) Namely. r=1 r r vα (p)v β (p) = p−m / 2m . 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) = 0.3. the matrix Q projects only on to the space of negative energy solutions with a phase (a negative sign). 2 or.4 Projection operators.103)) .107) then.109) The completeness relation for the solutions of the Dirac equation now follows from the observation that (see (3. completeness relation 89 2 Qαβ (p) = r=1 r r vα (p)v β (p). Hence we can identify Qαβ = −(Λ− (p))αβ . αβ (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). (3. 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 .

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

4 Projection operators.β.3. 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.λ=1 r. 2 (3.σ. think of an experiment with unpolarized initial electron states where the ﬁnal spin polarization is not measured).s=1 1 2 r.s=1 2 2 u r (p)M us (p′ ) † = 1 u r (p)M us (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).σ. 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.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.β.β=1 r=1 4 = = = Mαβ (Λ+ (p))βα = α.λ=1 2 s Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α.114) Similarly.β=1 1 Tr M Λ+ (p).s=1 σλ r ur (p)uα (p) λ σλ = ur (p) λ = .β=1 4 2 r Mαβ ur (p)uα (p) β α.s=1 1 2 4 2 4 r s uα (p)Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α.

118) As a consequence.117) satisﬁes the commutation relation (see (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. . we have already seen that the spin operator 1 ˜ α. 3. in the relativistic case this is not useful since spin is not a constant of motion. In fact. that. H] = (3. ˜ 2 [Si . On the other hand.λ=1 βσ γ 0M †γ 0 σλ (Λ+ (p))λα (3.5 Helicity As we have seen.β.92 1 2 4 3 Properties of the Dirac equation = Mαβ Λ+ (p′ α. 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 2.116) does not commute either with the orbital angular momentum or with spin (rather. (3. Thus. however. we note that since momentum commutes with the Dirac Hamiltonian. H = −iǫijk αj pk . by the eigenvalue of Sz ). Note.115) = 1 Tr M Λ+ (p′ )γ 0 M † γ 0 Λ+ (p) . namely.σ. for a particle at rest. (3. spin commutes with the Hamiltonian (since in this frame p = 0) and such solutions can be labelled by the spin projection.67)) 1 αi .6. it commutes with the total angular momentum). it can be easily checked that the plane wave solutions which we had derived earlier are not eigenstates of the spin operator. the Dirac Hamiltonian H = α · p + βm. 2 σi 0 0 σi S= αi = ˜ .

121) measures the longitudinal component of the spin of the particle or the projection of the spin along the direction of motion. 4 (3.120) Therefore. h′ ) = δhh′ = −v(p.123) Therefore. α · p + βm] = 0. Namely. the eigenvalues of the helicity operator. h)v(p. h′ ). h′ ). v(p. (3. can only be ± 2 and we can label the positive and the negative energy solutions also as u(p. h′ ) = 0 = v(p. this operator is a constant of motion. for a Dirac parti1 cle. Note that S·p |p | 2 h2 = = 1 ½. h)u(p. [Si pi . H] = [Si . h). This is known as the helicity operator and we note that since the Hamiltonian commutes with helicity. h)u(p. H]pi = −iǫijk αj pk pi = 0. therefore. (3. h) with h = ± 1 (the two helicity 2 eigenvalues). 4 4 (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 ½.124) . (3. H] = [pi .3.5 Helicity 93 [pi . the eigenstates of energy can also be labelled by the helicity eigenvalues.119) the operator S · p does also (momentum and spin commute and. |p | (3. The normalized operator h= S·p . The normalization relations in this case will take the forms u(p. the order of these operators in the product is not relevant). h)v(p. u(p.

or. γ 0 p0 − γ · p − m u(p) = 0. (3.126) where we are not assuming any relation between p0 and p as yet.110) or (3.129) p0 u2 (p) − σ · p u1 (p) = −mu2 (p). (3. (3.6 Massless Dirac particle Let us consider the free Dirac equation for a massive spin (γ µ pµ − m) u(p) = 0. Let us represent the four component spinor (as before) as u1 (p) u2 (p) u(p) = . which can also be written as p0 u1 (p) − σ · p u2 (p) = mu1 (p). this leads to the two (2-component) coupled equations p0 − m u1 (p) − σ · p u2 (p) = 0. the completeness relation. h)v(p. (3.125) 1 h=± 2 3.94 3 Properties of the Dirac equation Furthermore.127) where u1 (p) and u2 (p) are two component spinors.111). can now be written as [u(p. (3. In terms of u1 (p) and u2 (p). (3. the Dirac equation takes the form (p0 − m)½ σ·p −σ · p u1 u2 −(p0 + m)½ = 0.130) . h)u(p. (3. 1 2 particle.128) Explicitly. σ · p u1 (p) − p0 + m u2 (p) = 0. h)] = ½. h) − v(p.

2 1 (u1 (p) + u2 (p)) . are not invariant under parity or space reﬂection and are known as the Weyl equations. These equations. like the Dirac equation. 2 uL (p) = uR (p) = (3. p0 uL (p) = −σ · p uL (p). We note that if we deﬁne two new (2-component) spinors as 1 (u1 (p) − u2 (p)) .131) p0 + σ · p (u1 (p) − u2 (p)) = m (u1 (p) + u2 (p)) . (3.133) We note that it is the mass term which couples the two equations. however. the equations in (3.130).132) then. Let us note that in the limit m → 0.3. . we obtain p0 − σ · p (u1 (p) + u2 (p)) = m (u1 (p) − u2 (p)) .134) These two equations. The corresponding two component spinors uL and uR are also known as Weyl spinors.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). the two equations in (3. (3. p0 uL (p) + σ · p uL (p) = muR (p). since vanishing of the mass which is a Lorentz scalar should not change the behavior of the equation under proper Lorentz transformations). (3.6 Massless Dirac particle 95 Taking the sum and the diﬀerence of the two equations in (3.133) reduce to two (2-component) spinor equations which are decoupled and have the simpler forms p0 uR (p) = σ · p uR (p). can be shown to be covariant under proper Lorentz transformations (as they should be.

= (σ · p) (σ · p) uR (p) = p 2 uR (p).137) which is the Einstein relation for a massless particle. we note that p0 uR (p) = σ · p uR (p). for a nontrivial solution of these equations to exist. σ·p uL (p) = −uL (p). (p0 )2 uR (p) = σ · p p0 uR (p) (p0 )2 − p 2 uR (p) = 0. σ·p uR (p) = uR (p).138) For p0 = |p |. while (3. namely. Similarly. that for such solutions. we must have (p0 )2 − p 2 = 0.135) or.136) Thus. (3. (3. the two diﬀerent Weyl equations really describe particles with opposite helicity. |p | or. we can show that uL (p) also satisﬁes (p0 )2 − p 2 uL (p) = 0. (3. Recalling that 1 σ denotes the spin 2 . for the positive energy solutions. p0 uL (p) = −σ · p uL (p). |p | (3. (3. It is clear. we must have p0 = ±|p |. therefore.140) In other words.96 Let us note that 3 Properties of the Dirac equation or. p0 uR (p) = σ · p uR (p).139) or.

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

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

for example. see.147).61)).3. (2.53) and (2. γ5 vR (p) = vR (p). namely. Since 2 γ5 = ½. ˜ |p| v(p) = σ·p v (p) ˜ v (p) ˜ = σ ·p v (p) ˜ |p| v (p) ˜ .100)) H = α · p. in the Pauli-Dirac representation γ5 = ρ deﬁned in (2.148) From the structure of the massless Dirac equation (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.92)) u(p) ˜ σ·p u(p) ˜ E E (3.147) u(p) = E 2 E 2 = |p| 2 |p| 2 u(p) ˜ σ·p u(p) .7 Chirality 99 3. (3.54). / / can be written as (see also (3. we note that if u(p) (or v(p)) is a solution. Therefore. γ5 uR (p) = uR (p). (3.7 Chirality With the normalization for massless spinors discussed in (2. then γ5 u(p) (or γ5 v(p)) is also a solution. the solutions of the massless Dirac equation (m = 0) pu(p) = 0 = pv(p).60) and ρ commutes with α. (3.151) .149) commutes with γ5 (in fact. This can also be seen from the fact that the Hamiltonian for a massless Dirac fermion (see (1. (3.150) it follows that the eigenvalues of γ5 are ±1 and spinors with the eigenvalue +1.

then the Dirac Hamiltonian (1. γ5 vL (p) = −vL (p). 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. (3. Given a general spinor.153) where we have deﬁned . = 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) ˜ . γ5 uL (p) = −uL (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) .100 3 Properties of the Dirac equation are known as right-handed (positive chirality) spinors while those with the eigenvalue −1. We note that if the fermion is massive (m = 0). (3. vR (p) = PR v(p) = = |p| 2 1 2 1 2 1 (½ + γ5 )v(p) 2 ½ + σ ·p v(p) ˜ |p| .152) are called left-handed (negative chirality) spinors. 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| .

91)) 1 2 PR = ½ ½ 1 ½ −½ . PL = (½ − γ5 ). From the deﬁnition of the helicity operator in (3. hχ(±) (p) = σ·p 1 2|p| 2 ½± ½± 1 1 = ± × 2 2 σ·p χ ˜ |p| 1 σ·p χ = ± χ(±) (p). these projection operators have the explicit forms (see (2. which implies that any four component spinor can be uniquely decomposed into a right-handed and a left-handed component.156) We note from (3. we note that 2 spinors of the form χ(±) (p) = 1 σ·p 1± χ. The reducibility of the spinors is best seen in the Weyl representation for the Dirac matrices discussed in (2.6) that in the massless limit. the Dirac equation reduces to two decoupled two component Weyl equations (recall that it is the mass term which generally couples these two spinors).119). (In the Pauli-Dirac representation. (3. 2 −½ ½ ½ ½ (3.121) (for the two component spinors S = 1 σ).153) that in the massless Dirac theory.158) . ˜ |p| 2 (3. However. PL 2 = PL . namely.3.7 Chirality 101 1 1 PR = (½ + γ5 ). PL = .157) correspond to states with deﬁnite helicity. (3.155) PR PL = 0 = PL PR . ˜ 2 |p| (3. PR + PL = 1. 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 . we will continue our discussion in the Pauli-Dirac representation of the Dirac matrices which we have used throughout. the four component spinors can be eﬀectively described by two component spinors.

(3.160) can also be written in a covariant notation as P (+) = 1 µ σ pµ .162) and. |p| 2 P (−) = 1 µ σ pµ . Explicitly. uL (p) = .102 3 Properties of the Dirac equation so that the right-handed (four component) spinors in (3. vL (p) = −v (−) (p) v (−) (p) . (3. They can be easily generalized to a reducible representation of operators acting on the four component spinors and have the form (see (2.159) We note here that the operators P (±) = 1 2 ½± σ·p 1 ˆ = (½ ± σ · p).153) and (3. σ µ deﬁned in (2. P (+) P (−) = 0 = P (−) P (+) .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. ˆ 2 (3. It is straightforward ˜ ˆ p to check that they satisfy the relations (P (+) )2 = P (+) . deﬁne projection operators into the space of positive and negative helicity two component spinors.161) with σ µ . ˜ ˆ 2 (3.120) and pµ = (1.(3.70) or (3.157) 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) = .117)) P (±) 0 1 2( P4×4 (±) = 0 P (±) 0 1 2( = ˆ ½ ± σ · p) 0 ˆ ½ ± σ · p) . vR (p) = . P (+) + P (−) = 1. −ˆ ). we see from (3. therefore.163) . (P (−) )2 = P (−) .

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 −1.167) such that when p1 = p2 = 0.3.166) For example.167). from (3.165) 2 2 Each of these spinors is one dimensional and together they span the two dimensional spinor space. Furthermore.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. (3. (3.168) . We can choose u and v to be ˜ ˜ normalized so that we have u(+) (p) = u† u = 1 = v † v . let us derive some properties of these spinors.164) which is the reason the spinors can be simultaneous eigenstates of chirality and helicity (when mass vanishes).7 Chirality 103 and it is straightforward to check from (3. ˜ ˜ ˜ ˜ uu† + v v † = ½. For example. v(+) (p) = ½ + σ · p v.163) that PL. 1 (3. 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 . we can choose u= ˜ 1 . with the choice of the basis in (3. In fact. For completeness as well as for later use.R . ˜˜ ˜˜ (3.156) and (3. we can also deﬁne normalized spinors u(+) and v (+) . the helicity spinors simply reduce to eigenstates of σ3 .159) as well as (3. 0 v= ˜ 0 . (±) (3. P4×4 = 0. v (+) (p) = . We note from (3.

(3. In general. as well as (3. u(+) (p)u(+)† (p) = v (+) (p)v (+)† (p) = 1 2 ˆ ½ + σ · p .170) can be simpliﬁed by noting the following identity. u(+)† (p)v (+) (−p) = 0 = v (+)† (−p)u(+) (p).170) The completeness relation in (3. the positive helicity spinors satisfy u(+)† (p)u(+) (p) = v (+)† (p)v (+) (p) = 1. R † uR (p)u† (p) = vR (p)vR (p) = R |p| 2 P (+) P (+) P (+) 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 .171) = P (+) P (+) P (+) P (+) . Given the form of the right-handed spinors in (3. (3. (3.169) which can be checked from the explicit forms of the spinors in (3.104 3 Properties of the Dirac equation However.168). R † u† (p)vR (−p) = vR (−p)uR (p) = 0. We note that with p0 = |p|.169) it now follows in a straightforward manner that † u† (p)uR (p) = |p| = vR (p)vR (p). we do not need to use any particular representation for our discussions.159). 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).

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 = .176) . ˜ 2m σ·p E+m σ·p u= ˜ uL (p).8 Non-relativistic limit of the Dirac equation 105 so that we can write the completeness relation in (3. / 4 (3.4. (3. / 4 (3.49)) u ˜ σ·p ˜ E+mu uL (p) uS (p) u+ (p) = E+m 2m = .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. L † u† (p)vL (−p) = 0 = vL (−p)uL (p). L † uL (p)u† (p) = vL (p)vL (p) = L 1 0 pγ (½ − γ5 ). 2m E + m E+m (3.175) In (3.170) as † uR (p)u† (p) = vR (p)vR (p) = R 1 0 pγ (½ + γ5 ).173) 3.

σ · p uL (p) = (E + m)uS (p).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. vL (p) and vS (p) are also known as the large and the small components of the negative energy solution. In the non-relativistic limit.174). Similarly. uL (p) and uS (p) are known as the large and the small components of the positive energy Dirac solution.177) It is clear that in the non-relativistic limit. when |p | ≪ m. (3.179) We note that the second equation in (3. (3.175) we have denoted vL (p) = vS (p) = − E+m v. in (3. we expect the large components to give the dominant contribution to the wave function.179) gives the relation σ·p uL (p). 2m E + m E+m (3. or. which satisfy the equation Hu+ (p) = Eu+ (p). m½ σ · p −m½ σ·p uL (p) uS (p) =E uL (p) uS (p) . E+m uS (p) = (3. the component uS (p) is much smaller than (of the order of v ) uL (p) and c correspondingly.178) This would lead to the two (2-component) equations σ · p uS (p) = (E − m)uL (p). Similarly. Let us next look at the positive energy solutions in (3. ˜ 2m σ·p E +m σ·p v=− ˜ vL (p). or.180) . (α · p + βm) u+ (p) = Eu+ (p).

2 3. the Dirac equation in this case reduces to the Schr¨dinger o equation for a two component spinor which we are familiar with.179) takes the form (σ · p) or. Furthermore. equation (3.3. |p| ≪ m. with the substitution of this. what we know for a free non-relativistic electron (spin 1 particle).183) Namely. of course. the ﬁrst equation in (3.33) and remember that we are choosing = 1) . therefore. This preserves the gauge invariance associated with the Maxwell’s equations and corresponds to deﬁning pµ → pµ − eAµ . E+m (3. (3. 2m (3.181) has the form p2 uL (p) = ENR uL (p). σ·p uL (p) = (E − m)uL (p). and.182) (3.181) p2 uL (p) ≃ (E − m)uL (p). then. E ≈ m (recall that we have set c = 1). Since the coordinate representation of pµ is given by (see (1. 2m where we have used the fact that for a non-relativistic system. This is.184) where e denotes the charge of the particle and Aµ represents the four vector potential of the associated electromagnetic ﬁeld.9 Electron in an external magnetic field 107 while. if we identify the non-relativistic energy (without the rest mass term) as ENR = E − m.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.

we can write the two (2-component) equations as σ · (p − eA ) uS (p) = (E − m)uL (p). takes the form (α · (p − eA ) + βm) u(p) = Eu(p). E+m 2m (3. we have A0 = 0 = φ. we have used |p| ≪ m in the non-relativistic limit. the second equation in (3. (3.108 3 Properties of the Dirac equation pµ → i∂µ . we obtain .186) Let us next consider an electron interacting with a time independent external magnetic ﬁeld. The Dirac equation for the positive energy electrons. In this case. (3.189). (3. or. in this case.187) where we are assuming that A = A(x). Substituting this back into the ﬁrst equation in (3.185) the minimal coupling prescription also corresponds to deﬁning (in the coordinate representation) ∂µ → ∂µ + ieAµ . In this case.189) uS (p) = (3.189) leads to σ · (p − eA) σ · (p − eA) uL (p) ≃ uL (p). (3. σ · (p − eA ) uL (p) = (E + m)uS (p).188) Explicitly. B = (∇ × A ) .190) where in the last relation. m½ σ · (p − eA ) σ · (p − eA ) −m½ uL (p) uS (p) =E uL (p) uS (p) .

195) 2m 2m where we have identiﬁed (as before) . Ak ] = −iǫijk [∇j . We can use this in (3.194) Consequently. equation (3. 2m (3.3. (3. Ak ] = −i (∇ × A )i = −i(B)i .191) takes the form 1 (p − eA)2 − eσ · B uL (p) = (E − m)uL (p). 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) . 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 .191) Let us simplify the expression on the left hand side of (3.193) = (p − eA )2 − ieσ · (−iB) (3.9 Electron in an external magnetic field 109 (σ · (p − eA)) (σ · (p − eA)) uL (p) ≃ (E − m)uL (p). in the non-relativistic limit.192) Note that (here. 1 e (p − eA)2 − σ · B uL (p) ≃ ENR uL (p). (3.192) to write (σ · (p − eA)) (σ · (p − eA)) = (p − eA )2 − eσ · B. (3. when we can approximate the Dirac equation by that satisﬁed by the two component spinor uL (p).

however. 2m µ=g (3. In this case. a minimally coupled Dirac particle automatically leads.197) we obtain g = 2.110 3 Properties of the Dirac equation ENR = E − m. Thus.199) 1 Since S = 2 σ for a two component electron. (3. to a magnetic dipole interaction (recall that in the non-relativistic theory. this shows that a point Dirac particle has a magnetic moment corresponding to a gyro-magnetic ratio g = 2. 2m µ= (3. Quantum mechanical corrections (higher order corrections) in an interacting theory such as quantum electrodynamics. in the non-relativistic limit. we know that the magnetic moments are given by .197) Of course. however.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. (3. one says that there is an anomalous contribution to the magnetic moment. for the proton and the neutron. comparing with (3. for example. 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.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. Namely. can have g-factors quite diﬀerent from 2.196) We recognize (3. Particles with a nontrivial structure (that is particles which are not point like and have extended structures).

the Hamiltonian becomes non-Hermitian). (3.10 Foldy-Wouthuysen transformation 111 µN µP = −1. we have described how the non-relativistic limit of a Dirac theory can be taken in a simple manner. For example. This is commonly known as the Pauli coupling or the Pauli interaction. In the non-relativistic limit. The process of eliminating the “small” .200) where the nuclear magneton is deﬁned to be |e| . = 2. the electric dipole 1 moment becomes imaginary at order m2 (namely. 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 method runs into diﬃculty.79 µnm . 3. 2m (3. (3. the relevant expansion parameter is |p| and m the method works quite well in the lowest order of expansion. However.202) denotes the electromagnetic ﬁeld strength tensor and κ represents the anomalous magnetic moment of the particle.203) eκ µν σ Fµν .91 µnm .10 Foldy-Wouthuysen transformation In the last two sections.3.201) with mP denoting the mass of the proton. This puzzling feature can be understood in a simple manner as follows. at higher orders. 2mP µnm = (3. 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. as we have seen explicitly.

180)) σ·p . The matrix T that takes us to the two component “large” spinors in (3.207).100) as well as (1. Since the lack of unitarity in (3.206) It is clear from the form of the matrix in (3. for example. Thus. has the form (see (3. such a transformation would . in the case of the free Dirac equation.206) is the source of the problem in taking the non-relativistic limit consistently. 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. 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. E+m A= (3.205) for the positive energy spinors. let us look at the free Dirac theory where we know that the Hamiltonian has the form (see (1.112 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 .207) Let us next look for a unitary transformation that will diagonalize the Hamiltonian in (3. for example.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). (3. (3.204) has the form T = ½ −A 0 ½ .204) where the matrix A.101)) H = α · p + βm = γ 0 (γ · p + m) . (3. In this case.

1 (3.211) It follows now that U † (θ) = cos which leads to |p|θ 2m − γ ·p sin |p| |p|θ 2m .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. (3.210) and using this we can simplify and write U (θ) = ∞ n=0 1 (2n)! |p|θ 2m γ · pθ 2m + 2n + 1 (2n + 1)! . Let us consider a transformation of the type U (θ) = e 2m γ ·p θ .3.208) where the real scalar parameter of the transformation is a function of p and m. (3. (γ · p)2 = −p2 = −|p|2 .209) From the properties of the gamma matrices in (1. avoid the problem with “large” and “small” spinors). θ = θ(|p|. m).212) .91). (3. we note that (γ)† = −γ. γ · pθ 2m 2n+1 = cos γ ·p sin |p| |p|θ 2m (3.

208). Namely.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) is indeed unitary.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 free Dirac Hamiltonian (3. (3. the transformation (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 .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. Under the unitary transformation (3.

from (3. = 0. tan θ= − m sin |p| . sin = p2 + m2 .214). the two components of the transformed spinor u′ (p) 1 u′ (p) 2 u′ (p) = U (θ)u(p) = . m |p| m .217) that the Hamiltonian is now diagonalized in the positive and the negative energy spaces. we have |p|θ m m p2 + m2 |p|θ m |p| cos = . our discussion has been quite general and the parameter of the transformation. (3.218) 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).217) We see from (3.214) that we can choose the parameter of transformation to satisfy |p|θ m = |p|θ m |p| cos or. it is clear from (3. |p|θ m m arctan |p| (3. (3. has been arbitrary.3. However. or. 1 1 (3. θ. if we want the transformation to diagonalize the Hamiltonian. (3.216) which. leads to the diagonalized Hamiltonian m2 p2 + m2 p2 p2 + m2 H ′ = γ0 + = γ0 p2 + m2 .219) . As a result.215) In this case.10 Foldy-Wouthuysen transformation 115 So far.

There is a second limit of the Dirac equation. tan + |p| sin =− m . (3. θ=− m arctan |p| (3.215).211) is also quite useful.222) p2 + m2 As a result.220) UFW (θ) = cos + which has a natural non-relativistic expansion in powers of |p| . this leads to the non-relativistic equation in (3.116 3 Properties of the Dirac equation For |p| ≪ m. Let us note from (3.221) this would lead to |p|θ m |p|θ m |p| cos sin = = − p2 + m2 m . m (3. namely. = 0.214) that if we choose the parameter of transformation to satisfy |p|θ m |p|θ m |p|θ m m cos or. 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. In this case. for which the generalized Foldy-Wouthuysen transformation (3.211) takes the form 1 arctan 2 |p| m γ·p sin |p| 1 arctan 2 |p| . or. the transformation is known as the Cini-Touschek transformation and is obtained as follows. We also note that with the parameter θ determined in (3. in this case. .183) to 1 the lowest order and it can be expanded to any order in m without any problem. the transformed Hamiltonian (3. the ultrarelativistic limit |p| ≫ m. the unitary transformation in (3. |p| m |p| .214) will have the form .

Therefore. where operators carry time dependence and states are time independent. |p| m which has a natural expansion in powers of |p| .225) In the Heisenberg picture. In fact.211) has the form UCT (θ) = cos 1 arctan 2 m |p| − γ ·p sin |p| 1 arctan 2 m . |p| (3. 3.224) m which clearly has a natural expansion in powers of |p| (ultrarelativistic expansion). For example.3.207) while the Cini-Touschek transformation rotates away the mass term βm from the Hamiltonian (3. (3. the unitary transformation (3. we can think of the Foldy-Wouthuysen transformation (3. in this case.100) H = α · p + βm. let us consider the free Dirac Hamiltonian (1.11 Zitterbewegung The presence of negative energy solutions for the Dirac equation leads to various interesting consequences.223) p2 + m2 γ·p= |p| p2 + m2 α · p.11 Zitterbewegung 117 H ′ = γ0 = γ0 p2 p2 + m2 + m2 p2 + m2 γ·p |p| (3.207).220) as transforming away the α · p term in the Hamiltonian (3. the Heisenberg equations of motion take the forms ( = 1) .

α · p + βm] = 0. The ﬁrst equation. H] = [α. = i ˙ or. it follows that the . the velocity is not.118 3 Properties of the Dirac equation 1 i 1 [x . for example. Furthermore. using (1. H] = 0. The second equation in (3.228) (3. on the other hand. p = 0. xi = ˙ 1 1 i [p . Let us recall that. α(t) = eiHt α e−iHt .101)). As a result.101) we conclude that [α. (3. since the eigenvalues of α are ±1 (see. αj pj + βm] = 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. (1. identiﬁes α(t) with the velocity operator. H] = [xi . x = α. α p ] = αj [xi .226) shows that the momentum is a constant of motion as it should be for a free particle. pj ] = i i 1 j ij α (iδ ) = αi . Secondly. even though the momentum of a free particle is a constant of motion.229) [α(t). (3.230) In other words. it follows that (3. H] = [pi . by deﬁnition. pi = ˙ (3. i i ˙ or.226) Here a dot denotes diﬀerentiation with respect to time.

it follows that eiHt αe−iHt eiHt |ψ = λ eiHt |ψ . = λ eiHt |ψ .233) Equation (3. This is easily understood from the fact that the eigenvalues of an operator do not change under a unitary transformation. H] i + ˙ α(t) = = −i 2α(t)H − α(t). This would seem to imply that the velocity of an electron is equal to the speed of light which is unacceptable even classically. or. αj αi . (3. = 0.3. We note from Heisenberg’s equations of motion that the time derivative of the velocity operator is given by 1 [α(t). (3.231) where λ denotes the eigenvalue of the velocity operator α. (3. α(t) eiHt |ψ where we have identiﬁed |ψ ′ = eiHt |ψ . (3. More explicitly. (3.102)) αi . therefore. α(t)|ψ ′ = λ |ψ ′ . since the electron is a massive particle.235) + + .11 Zitterbewegung 119 eigenvalues of α(t) are ±1 as well. are ±1. we note that if α|ψ = λ|ψ . then.232) shows that the eigenvalues of α(t) are the same as those of α (only the eigenfunctions are transformed) and. H = −2iα(t)H + 2ip. β = 2δij ½.234) Here we have used the relations (see (1.232) or. These peculiarities of the relativistic theory can be understood as follows.

+ α(0) − α(t) = H H The ﬁrst term in (3.234) we have ˙ α(0) = −2iα(0)H + 2ip = −2i(α(0) − pH −1 )H.239) or. diﬀerentiating (3. we ﬁnally determine −2i(α(0) − p H −1 )He−2iHt = −2iα(t)H + 2ip. (c = 1) E = mγ.239).120 3 Properties of the Dirac equation as well as the fact that momentum commutes with the Hamiltonian. It represents an additional component to the .239) is quite expected. (3.236). α(t) = pH −1 + (α(0) − pH −1 )e−2iHt .234). It is the second term. p −2iHt p e . ˙ ˙ or. however. which is unexpected. We note that.241) p = mγv.238) Furthermore. for a relativistic particle. Therefore.237) (3. so that mγv p = = v. (3. For example. we obtain ¨ ˙ α(t) = −2iα(t)H. from (3.240) which is the ﬁrst term in (3. E mγ (3. we obtain ˙ α(t) = −2i(α(0) − p H −1 )He−2iHt . using this relation in (3. On the other hand. (3. α(t) = α(0)e−2iHt . or. in an p eigenstate of momentum it would have the form E which is the true relativistic expression for velocity. Substituting this back into (3. Let us note next that both p and H are constants of motion.236) (3.234) with respect to time.

the energy is ≈ . This quivering motion of the electron was ﬁrst studied by Schr¨dinger o and is known as Zitterbewegung (“jittery motion”). we obtain i p t+ H 2 p H −1 e−2iHt . for example. H x(t) = a + α(0) − (3.) and gives a time dependence to α(t).11 Zitterbewegung 121 velocity which is oscillating at a very high frequency (for an electron at rest.3. However.243) can be shown in the Schr¨dinger picture to arise from the presence of o negative energy solutions. the third term represents an additional contribution to the electron trajectory which is oscillatory with a very high frequency.226) that since ˙ x(t) = α(t) = p −2iHt p e . The ﬁrst two terms in (3. + α(0) − H H (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 . Its occurrence is quite surprising.5MeV corresponding to a frequency of the order of 1021 /sec.239) and (3. In fact. σ·p ˜ E+mu u(p) = we have (3. The unconventional operator relations in (3. Let us also note from (3.243) are again what we will expect classically for uniform motion. since there is no potential whatsoever in the problem.242) integrating this over time.243) where a is a constant. it is easy to check that for a positive energy electron state E+m 2m u ˜ .

1964. α(t). (3.245) 2E 2p p × + E E+m E+m u† u = 0. (This makes clear that neglecting the negative energy solutions of the Dirac equation would lead to inconsistencies. This shows that even though the eigenvalues of the operator α(t) are ±1 corresponding to motion with the speed of light. Relativistic Quantum Mechanics. J. Drell. 3.246) as we should expect from the Ehrenfest theorem. McGraw-Hill. in a positive energy electron state p H α(t) + = + . McGrawHill. . 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.12 References 1. 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. 1968. Quantum Mechanics. New York. Quantum Field Theory. Schiﬀ. This also shows that the eigenstates of the velocity operator. Zuber. I. Bjorken and S. D. New York. 1980.) 3. C.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. New York. ˜ ˜ This shows that even though the operator relations are unconventional. Itzykson and J-B. McGraw-Hill. L. 2.

5. S. L. 6. M. . 267 (1951). ibid. 422 (1958). S. 7. Progress of Theoretical Physics 12. Cini and B. 102 (1954). Progress of Theoretical Physics 6. Tani. Physical Review 78. Foldy and S. 603 (1954). 29 (1950). Okubo. L. Nuovo Cimento 7. 12. Wouthuysen.12 References 123 4.3. A. Touschek.

This page intentionally left blank .

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

ω µν in order to avoid confusion.126 4 Representations of Lorentz and Poincar´ groups e notation from (1. It follows now that i R(ω)xi = −ω kj xj ∂k xi = −ω kj xj δk = −ω ij xj = −ω i j xj .5) R(ω) = = ω ij xi ∂j = −ω ij xj ∂i .34) and (1. (4.) ω i j = ǫi jk αk . (4.35) here for clarity.6) In other words. we note that ω ij = −ω ji . ǫµν . ǫijk denotes the three dimensional Levi-Civita tensor with ǫ123 = 1. (4. the form of the rotation that we had discussed in the last chapter. then.3) (4. we can write the inﬁnitesimal rotations also as δxi = −R(ω)xi = ω i j xj . of course. but here we denote them by ω ij . (4. etc. ǫi jk = η iℓ ǫℓjk .) If we now identify (In the last chapter we had denoted the inﬁnitesimal transformation matrices by ǫi j .4) This is. 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 (4.2) 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) .

the vector operator. (4. rather. xk ] ∂ℓ + xk [xi .8) = R(ω). The Lie algebra of the group of rotations can be obtained from the algebra of the vector operators themselves. ω]ij . they give back a rotation. Thus.8). are known as the generators of inﬁnitesimal rotations. xk ∂ℓ ] = ω ij ω kℓ (xi [∂j .5). ∂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. Alternatively. we can calculate them directly as . Using the form of R(ω) in (4.4. R(ω) = 1 ij ω Mij . Mij . R(ω) = ω ij xi ∂j . we note that R(ω). two rotations do not commute.9) Namely. Such an algebra is called a non-Abelian (non-commutative) algebra.10) we can obtain the algebra satisﬁed by the generators of inﬁnitesimal rotations.1 Symmetry algebras 127 Namely. Mij . namely.5) generates inﬁnitesimal rotations and the operators. from the algebra of the vector operators in (4. R(ω) in (4. 2 (4. ω kℓ xk ∂ℓ = ω ij ω kℓ [xi ∂j . where we have identiﬁed ω ij = ω ik ω kj − ω ik ωkj = −ω ji = [ω.

then. xk ∂ℓ ] − [xi ∂j . To obtain the familiar algebra of the angular momentum operators. 2. xℓ ∂k ] − [xj ∂i . (4. Jj ] = = 1 pq rs ǫ ǫ [Mpq . we note that we can deﬁne (recall that in the four vector notation Ji = −(J)i ) Mij or. xk ∂ℓ − xℓ ∂k ] = [xi ∂j . 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.11) represents the Lie algebra of the group SO(3) or equivalently SU (2). then we may deﬁne the operators.12) 1 1 Ji = − ǫi jk Mjk = − ǫijk (xj ∂k − xk ∂j ). ǫjrs Mrs 2 2 [Ji . Using this. r.11) This is the Lie algebra for the group of rotations. with a factor of “i”. q. 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 .128 4 Representations of Lorentz and Poincar´ groups e [Mij . 2 2 which gives the familiar orbital angular momentum operators. (4. Mkℓ ] = [xi ∂j − xj ∂i . s = 1. Mij . xk ∂ℓ ] + [xj ∂i . 3) 1 pq 1 ǫi Mpq . = (xi ∂j − xj ∂i ) = ǫijk Jk . we obtain (p. But up to a rescaling. (4.

(4.17) (4. In the same spirit. let us note that a constant in- ﬁnitesimal space-time translation of the form δxµ = ǫµ .11) or (4.13) is. then. namely. the Lie algebra of SO(3) or SU (2) (or the familiar algebra of angular momentum operators) up to a rescaling. (4.16) (4.18) .12)) 1 q ǫri ǫjsr Mqs = − ǫijr ǫrsq Mqs = −ǫijr Jr .2 Translation.13) 1 + ǫis q ǫjrs Mqr + 4 1 1 = − ǫriq ǫjsr Mqs − 4 4 1 − ǫri q ǫjsr Mqs − 4 or.15) The algebra of the generators in (4.4. so that µ R(ǫ)xµ = −ǫν ∂ν xµ = −ǫν δν = −ǫµ . 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).14) which. ǫri q ǫjsr + ǫrsq ǫij r + ǫrjq ǫsi r = 0. = −ǫri q ǫjsr Mqs . leads to (see (4.1. 4. of course. Jj ] = ǫij r Jr . can be generated by the inﬁnitesimal vector operator R(ǫ) = −ǫµ Pµ = −ǫµ ∂µ .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. 2 (4. [Ji .

∂ν ] = 0. ν = 0.19) The Lie algebra associated with translations is then obtained from R(ǫ). the inﬁnitesimal. constant parameters of transformation satisfy ω µν = −ω νµ . as we have seen in (3. ǫ ν ∂ν ] = ǫµ ǫ ν [∂µ . (4.22) where. R(ω) = (4.3 Lorentz transformation. (4. (4. (4.21) Namely. (4. µ.20) In other words. we obtain . As we have seen in the last chapter. 4.18). 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 . translations form an Abelian (commuting) group while the three dimensional rotations form a non-Abelian group. two translations commute and the corresponding relation for the generators is [Pµ . ∂ν ] = 0.23) As in the case of rotations. 2. Pν ] = [∂µ . 3. R(ǫ) = [ǫµ ∂µ .1.24) then. 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 . 1.130 and we can write 4 Representations of Lorentz and Poincar´ groups e δxµ = −R(ǫ)xµ .

(4. Mµν = −Mνµ .26) M0i = −Mi0 = x0 ∂i − xi ∂0 = Ki . ω]µν . 1 µν 1 ω Mµν .1 Symmetry algebras 131 µ R(ω)xµ = −ω λν xν ∂λ xµ = −ω λν xν δλ = −ω µν xν = −ω µν xν = −δxµ . as the generators of the inﬁnitesimal transformations. as in the case of rotations. ω λρ Mλρ 2 2 R(ω). and the generators of inﬁnitesimal boosts with (4.27) As before. we can think of R(ω) as the vector operator generating inﬁnitesimal proper Lorentz transformations and the operators.29) .28) = = ω µν xµ ∂ν = R(ω). we have ω µν = −ω νµ = ω µλ ω λν − ω µ ω λν = [ω.12)) Mij = −Mji = xi ∂j − xj ∂i = ǫij k Jk .4. (4.25) Therefore. We also note that we can identify the inﬁnitesimal generators of spatial rotations with (see (4. R(ω) = = ω µν ω λρ [xµ ∂ν . 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. we can determine the group properties of the Lorentz transformations from the algebra of the vector operators generating the transformations. λ (4. where. Thus.

it is n(n−1) ) 2 1 × 4 × (4 − 1) = 2 × 3 = 6. In fact. the matrices σµν provide a representation for the generators of the Lorentz group. . This implies that. However. 1). xρ ∂λ ] − [xν ∂µ . in this case.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. xλ ∂ρ ] + [xν ∂µ . therefore.31) coincides exactly with the six generators we have (namely.30) This. (4. in four dimensions and. Lorentz transformations (boosts) are non-compact unlike rotations in Euclidean space. the Lie algebra of the generators is isomorphic to that of the group SO(4). we note that the number of generators for SO(4) which is (for SO(n). 2 (4. As we have seen these transformations correspond to rotations. therefore.109) coincides with (4. Mλρ ] = [xµ ∂ν − xν ∂µ .69) in connection with the discussion of covariance of the Dirac equation. The algebra of the generators can also be calculated directly and has the form [Mµν . 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. three rotations and three boosts). xλ ∂ρ ] − [xµ ∂ν .) We end this section by pointing out that the algebra in (2. xλ ∂ρ − xρ ∂λ ] = [xµ ∂ν . 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 . up to a scaling. This is what we had seen explicitly in (3. (Namely. gives the Lie algebra associated with Lorentz transformations.30) (up to a scaling).

(4. we also consider inﬁnitesimal translations. The transformations in (4. R(ǫ. µ µ R(ǫ. 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. ω µν denote respectively the parameters of inﬁnitesimal translation and Lorentz transformation. (4. ω) = ǫµ ∂µ + ω µν xν ∂µ . in this case.4. ω)xµ = − ǫν ∂ν + ω λν xν ∂λ xµ = −(ǫν δν + ω λν xν δλ ) = −(ǫµ + ω µν xν ) = −δxµ . in addition to inﬁnitesimal Lorentz e transformations.33) then.32) where ǫµ . 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 ) . the general transformation of the coordinates takes the form δxµ = ǫµ + ω µν xν . If. acting on the coordinates.34) The algebra of the vector operators for the Poincar´ transformae tions can also be easily calculated as R(ǫ. xρ ∂λ ] + ω µν ǫ λ [xν ∂µ . Namely.1 Symmetry algebras 133 4. Clearly.35) . if we deﬁne an inﬁnitesimal vector operator as R(ǫ.4 Poincar´ transformation. (4.32) are known as the (inﬁnitesimal) Poincar´ transformations or the inhomoe geneous Lorentz transformations.1. it generates inﬁnitesimal Poincar´ e transformations. ω) . ǫ λ ∂λ + ω λρ xρ ∂λ = ǫµ ω λρ [∂µ . (4. ∂λ ] + ω µν ω λρ [xν ∂µ .

30) and (4.38) [Mµν . However. Pµ behaves 1 like a covariant four vector. see.21)).22) and (3. for example. 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. (4. (4.21). Note that [Pµ . Mνλ ] = ηµν Pλ − ηµλ Pν . [Pµ . Therefore.) Thus. Mλρ ] = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ . ω]µν . We note that the algebra of translations deﬁnes an Abelian subalgebra of the Poincar´ algebra (4. e We already know the commutation relations [Mµν . (4. For change in the coordinate four vector under an inﬁnitesimal Lorentz transformation. (4. (This is seen by recalling that 2 ω µν Mµν generates inﬁnitesimal Lorentz transformations. (4. (4.134 4 Representations of Lorentz and Poincar´ groups e where we have identiﬁed ǫµ = (ω µν ǫ ν − ω µν ǫν ) . ω µν = ω µλ ω λν − ω µλ ωλ ν = [ω. we conclude that the Lie algebra associated with the Poincar´ e transformations (inhomogeneous Lorentz group) is given by [Pµ .36) We can also calculate the algebra of the generators of Poincar´ group.30). xν ∂λ − xλ ∂ν ] = ηµν ∂λ − ηµλ ∂ν = ηµν Pλ − ηµλ Pν .57).38). combining with our earlier results on the algebra of the translation group. since the generae tors of translations do not commute with the generators of Lorentz . Mνλ ] = [∂µ . Pν ] = 0. Pν ] (see (4.37) which simply shows that under a Lorentz transformation. the only relation that needs to be calculated is the commutator between the generators of translation and Lorentz transformations. as well as the homogeneous Lorentz group. Mλρ ] as well as [Pµ .

it is what is known as a semi-direct sum of the two algebras.4. Poincar´ algebra cannot be written as a direct sum e of those for translations and Lorentz transformations. Namely. 2 Ki = M0i .) 4. (The general convention is to denote groups by capital letters while the algebras are represented by lower case letters. Poincar´ algebra = t4 ⊕ so(3.40) We recall from (4. Kj ] . (4.12).39) Rather.30) [Mµν .42) .27) that we can identify the angular momentum and boost operators as 1 Ji = − ǫi jk Mjk . (4. − 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 . Kj ] = [M0i . 3.2 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. the Lorentz algebra takes the form [Ki . (4. Mλρ ] = (−ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ ). Jj ] = ǫij k Jk .26) and (4. Jj ] = M0i . [Ki .2 Representations of the Lorentz group 135 transformations.41) Written out in terms of these generators. i = 1. M0j ] = −Mij = −ǫij k Jk . [Ji . (4. 2. e (4. 1).

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

Bj ] = = 1 ([Ji . Kj ] + i [Ki . 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 . Bj ] = = = = [Ai . Jj ] − i [Ji .4. 2 ij 1 1 (Ji + iKi ) . Mathematically. Jj ] − i [Ji . (Jj + iKj ) 2 2 1 ([Ji . (4. 1) ≃ so(3) ⊕ so(3) ≃ su(2) ⊕ su(2). Jj ] + i [Ji . Jj ] − [Ki .49) Incidentally.43).42) takes the form 1 1 (Ji + iKi ) . (Jj − iKj ) 2 2 [Ai . one says that the Lorentz algebra is isomorphic to the direct sum of two angular momentum algebras. (Jj − iKj ) 2 2 1 ([Ji . Jj ] + [Ki . Kj ]) 4 1 ǫ k Jk − iǫij k Kk + iǫij k Kk − ǫij k Jk = 4 ij = 0. the Lorentz algebra (4.2 Representations of the Lorentz group 137 In the new basis (4. the algebra separates into two angular momentum algebras which are decoupled. Therefore. Aj ] = = = = [Bi . 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 .48) In other words. the Lorentz group is double valued (doubly connected). Kj ] + i [Ki . (4. Jj ] − [Ki . 2 ij 1 1 (Ji − iKi ) . as we have already seen in the last chapter. so(3. in this new basis. it is more . Kj ] − i [Ki .

1.138 4 Representations of Lorentz and Poincar´ groups e meaningful to consider the simply connected universal covering group of SO(3.jB ) (x′ ) = D(jA . therefore. The ﬁnite dimensional unitary representations of each of the angular momentum algebras is well known. .jB ) “ ” (j ) (j ) i i −i θA Ai A +θB Bi B “ ” (j . 1) which is SL(2. . much the same way we consider the universal covering group SU (2) of SO(3) to describe rotations.51) where Λ represents the ﬁnite Lorentz transformation parameter. Namely. 2 2 (4. .jB) (Λ)ψ (jA . . .53) . 2 A (4. . (4. 2 2 3 1 jB = 0. . can be speciﬁed uniquely once we know the values of jA and jB and is labelled as D(jA . ) to describe the Lorentz transformations. 2 A δi = i i i θ − θB . . we have 3 1 jA = 0. .j ) (j .jB ) (x) = D(jA ) (Λ)D(jB ) (Λ)ψ (jA . (4.jB ) (just as the representation of the rotation group is denoted by D(j) ).52) where the ﬁnite parameters of rotation and boost can be identiﬁed with θi = 1 i i θ + θB . this represents the operator implementing ﬁnite transformations on the Hilbert space of states or wave functions as ψ ′ (jA .50) An irreducible nonunitary representation of the homogeneous Lorentz group.) (jA .jB ) (x). Explicitly. we can write (This is the generalization of the S(Λ) matrix that we studied in (3.35) in connection with the covariance of the Dirac equation. 1.j ) −i θ i Ji A B +δi Ki A B D (Λ) = e = e . . Denoting by jA and jB the eigenvalues of the Casimir operators A2 and B2 respectively for the two algebras. . .

For jA = 1 . jB = 0.58) 1 corresponds to a two component spinor representation with spin 2 . dimensionality = 1. . 1 1 D(0. 2 (4. . (4. For jA = jB = 2 . 1 Similarly. 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. dimensionality = 2.57) corresponds to a scalar representation with zero spin (which acts on the wave function of a Klein-Gordon particle). .4. For jA = jB = 0. j = 0. j = . jB ) can lie between j = |jA − jB |. 2 (4.44)) J = A + B. j = .2 Representations of the Lorentz group 139 Such a representation labelled by (jA .55) Consequently. . jA + jB . (4. dimensionality (4. 2 1 1 D( 2 . dimensionality = 2. for jA = 0.0) . These two representations are inequivalent and.56) The ﬁrst few low lying representations of the Lorentz group are as follows.59) also corresponds to a two component spinor representation with spin 1 2 .0) . 2 ) . (4.54) and its spin content follows from the fact that (see (4. from our knowledge of the addition of angular momenta. in fact. we conclude that the values of the spin in a given representation characterized by (jA . jB ) will have the dimensionality D (jA . jB = 2 . . |jA − jB | + 1.jB ) = (2jA + 1) (2jB + 1) . D(0.

2 (1) ( 1 . 1 1 (4.44). It actually corresponds to a reducible representation of the Lorentz group of the form 1 1 1 D(0. Let us now con- struct explicitly a few of the low order representations for the generators of the Lorentz group.48). Bi by a factor of i. 1. Ki . (4.62) arises because ǫij = −ǫijk in (4.) Ai (0) = 0. 2 ) ⊗ D ( 2 . this leads to the ﬁrst two nontrivial representations for the angular momentum and boost operators of the forms 1 ( 1 . (Note that a four vector such as xµ has a spin zero component.62) Using (4. ··· .0) . we have (We note here that the negative sign in the spin 2 k representation in (4. dimensionality = 4.2.1 Similarity transformations and representations.60) is known as a four component vector representation and can be identiﬁed with a spin content of 0 and 1 for the components. Bi (0) 1 (1) Ai 2 = − σi . t and a spin 1 component x (under rotations) and the same is true for any other four vector.140 4 Representations of Lorentz and Poincar´ groups e D( 2 . we now consider Hermitian generators by letting Mµν → iMµν as in (4. (Namely. Ai . dimensionality = 2 × 2 = 4. To compare with the results that we had derived earlier.) It may be puzzling as to where the four component Dirac spinor ﬁts into this description. 2 (4.50).47).63) . 2 ··· . = 0. namely. 2 (4. 2 ) . we note that for the ﬁrst few low order represen1 tations.61) This discussion can similarly be carried over to higher dimensional representations.0) (1) (0) Ji 2 = Ai 2 + Bi = − σi . we scale all the generators Ji .) From (4. j = 0. 4. j = . 2 1 1 ( ) Bi 2 = − σi .0) (0) = −i Ai 2 − Bi Ki 2 = i σi .

71) and (3. .66) which is clearly impossible. 1 ) 2 = 1 − 2 σi Ji (Dirac) 0 1 − 2 σi 0 .98) (or (3.4. 1 ) ( 1 . if there exists a similarity transformation relating the two.69) and (2.63) and (4. 0 and 0. if we can ﬁnd a similarity transformation S leading to (0.64). (4. we note that these do not resemble the generators of the Lorentz algebra deﬁned in (3. Therefore. 1 ) (0) Ji 2 = Ai + Bi 2 = − σi . 2 (4. 0 and 0. 1 ) (1) (0) Ki 2 = −i Ai − Bi 2 i = − σi . from (4. 2 (0.2 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.0) Ji 2 = S −1 Ji 2 S. (4.67) However.0) 2 ⊕ Ki (0.63) and (4. (4. S −1 σi S = −σi . the two representations la1 belled by 1 . Two representations are said to be equivalent.63) and (4. (Dirac) Ki i 2 σi 0 0 i − 2 σi .64) give the two inequivalent representations of dimensionality 2 as we have noted earlier. we can obtain the representation of the Lorentz generators for the reducible four component Dirac spinors as (0. 1 ) Ki 2 = S −1 Ki 2 S.65) would imply the existence of an invertible matrix S such that S −1 σi S = σi . For example.64) Equations (4. 1 ) ( 1 . From (4.65) 1 then.0) = Ji 2 ⊕ Ji 2 = = Ki ( 1 .0) (0.78)). 2 are inequivalent representations. 1 are 2 equivalent. In fact. we would say that the two representations 2 .64) we see that the condition (4. ( 1 .

γW = ǫijk 2 W . On the other hand. if we would like the generators in the standard Pauli-Dirac representation (which is what we had used in our earlier discussions).70) = σi i − 2 σi 0 1 PD σ .142 4 Representations of Lorentz and Poincar´ groups e This puzzle can be understood as follows.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 . (4. We note that in the Weyl representation for the gamma matrices deﬁned in (2. ij σW = i i j γ .68) As a result. 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. 2 0i Therefore.67) are obtained from 1 W σ .γ = 2 W W −iσi 0 σk 0 0 iσi 0 σk 0i σW = .119). consequently.71) and (3.98) (see also (3. (4.69) 2 µν and. we can apply the inverse similarity transformation in (2. we note that the generators in (4. we note that the angular momentum and boost operators in (4. give a representation of the Lorentz generators in the Weyl representation. i 0 i γ .78)) are equivalent since they are connected by a similarity transformation that .69) and (2.

2 Representations of the Lorentz group 143 relates the Weyl representation of the Dirac matrices in (2. (Here we will use 3dimensional Euclidean notation without worrying about raising and lowering of the indices. considering inﬁnitesimal rotations of the coordinates around the x-axis and the y-axis respectively. the two representations for the generators constructed from two diﬀerent perspectives (for the same four vector representation) appear rather diﬀerent and. we can deduce the matrix form of the corresponding generators to be . there must be a similarity transformation relating the two representations. On the other hand. Let us consider a three dimensional inﬁnitesimal rotation of coordinates around the z-axis as described in (3. from the Lie algebra point of view the four vector representation corresponds to jA = jB = 1 (see (4. (4.4).71) we can immediately read out from (3. as we discussed earlier. from the inﬁnitesimal change in the coordinates under a Lorentz transformation (see. Let us illustrate this for the simpler case of rotations. for example. (3.4.4) the matrix structure of the generator J3 to be 0 0 i 0 0 0 J3 = −i 0 0 . therefore.119) to the standard Pauli-Dirac representation.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. For example. There is yet another interesting example which sheds light on similarity transformations between representation.18)). Surprisingly.) Representing the inﬁnitesimal change in the coordinates as δxi = iǫ (J3 )ij xj . (4. The case for Lorentz transformations follows in a parallel manner.72) Similarly. we can determine a representation for the generators of the Lorentz transformations belonging to the representation for the four vectors.

Das (Hindustan Book Agency. 0 0 −1 0 −1 0 i = √ 1 0 −1 . Quantum Mechanics: A Modern Introduction. k = 1.72) and (4. the representation in the space of three vectors which would correspond to j = 1. New Delhi). Das and A. Jj ] = iǫijk Jk .72) and (4. A. j. J2 = 0 0 0 i −i 0 0 0 0 . page 182 (note there is a typo in the sign of the 23 element for L2 in this reference) 1 .73) and. let us note that the generators obtained (LA) from the Lie algebra are constructed by choosing the generator J3 See.73) in spite of the fact that they belong to the same representation for j = 1. On the other hand. 3.73) that they satisfy [Ji . i.144 4 Representations of Lorentz and Poincar´ groups e It can be directly checked from the matrix structures in (4. therefore provide a representation for the generators of rotations. To construct the similarity transformation (which actually is a unitary transformation). 2. 2 0 1 0 1 0 0 = 0 0 0 . therefore. page 289 or Lectures on Quantum Mechanics. in fact. 2 0 1 0 (4. (4. A. C. they are equivalent.75) (LA) J1 (LA) J2 (LA) J3 which look really diﬀerent from the generators in (4. (4. 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 .74) J1 = 0 0 0 0 0 0 −i 0 i . This is. Melissinos (Gordon and Breach). for example. This puzzle can be resolved by noting that there is a similarity transformation that connects the two representations and.

72) that the three normalized eigenstates of J3 have the forms 1 √ 2 i − √ .4. 2 ψ (j3 =1) = 0 ψ (j3 =−1) 1 −√ i2 = − √2 . Let us note from (4. 0 ψ (j3 =0) = 0 . 0 0 (4. (4.2 Representations of the Lorentz group 145 to be diagonal.76) 0 Let us construct a unitary matrix from the three eigenstates in (4.78) then. −1 (4. 1 √ 2 i − √ 2 0 0 −1 1 − √2 U= 0 i − √2 . 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 .76) which will diagonalize the matrix J3 . S = U. 0 U† = 0 1 − √2 1 √ 2 i √ 2 0 i √ 2 −1 .77) If we now deﬁne a similarity transformation (unitary) S −1 = U † .

(4. in fact. 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 .72). therefore.75) which look rather diﬀerent are. 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 . are equivalent. . related by a similarity transformation and.146 4 Representations of Lorentz and Poincar´ groups e = S −1 J2 S 0 1 0 1 (LA) √ 1 0 1 = J1 . √ 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 (4.79) This shows explicitly that the two representations for J corresponding to j = 1 in (4.73) and (4.

81) The last relation in (4.4. namely. Pµ = 0. it is useful to study the representations of the Poincar´ group. which is supposed to characterize the inﬁnitesimal transformation of P 2 . (4. The operators Mµν generate inﬁnitesimal Lorentz transformations through commutation relations and the relation above. let us note that the operator P 2 = η µν Pµ Pν = P µ Pµ . Mµν = = = P λ [Pλ .38) e since it commutes with all the ten generators. Pµ P 2 . Mµν ] P λ = P λ (ηλµ Pν − ηλν Pµ ) + (ηλµ Pν − ηλν Pµ ) P λ = Pµ Pν − Pν Pµ + Pν Pµ − Pµ Pν = 0. Mµν P λ Pλ . Therefore. This e would help us in understanding the kinds of theories we can consider and the nature of the states they can have. P 2 . from the generators of the Poincar´ group as e . In order to determine the unitary representations. we seek to ﬁnd unitary representations in some inﬁnite dimensional Hilbert space where the generators Pµ .81) can be intuitively understood as follows. (4.80) deﬁnes a quadratic Casimir operator of the Poincar´ algebra (4. it is known that it has only inﬁnite dimensional unitary representations except for the trivial representation that is one dimensional. P λ Pλ . Mµν act as Hermitian operators.3 Unitary representations of the Poincar´ group e 147 4. Let us deﬁne a new vector 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. Mµν ] + [Pλ . known as the Pauli-Lubanski operator. Since Poincar´ e group is non-compact (like the Lorentz group). 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.

Let us deﬁne the dual of the generators of Lorentz transformation as M µν = Mµν 1 µνλρ ǫ Mλρ . therefore. the vector Wµ is orthogonal to Pµ .83) which follows from the fact that the generators of translation commute. 2 2 (4. Furthermore. in general. the order of Pµ and Mνλ are irrelevant in the deﬁnition of the Pauli-Lubanski operator. As a result. First.83) implies that the PauliLubanski operator has only three independent components (both in the massive and massless cases). we note that Pµ W µ = 1 µνλρ ǫ Pµ Pν Mλρ = 0. we can write (4.84) With this. (4. 2 1 = − ǫµνλρ M λρ .82) also as W µ = Pν M µν = M µν Pν .) In general. (4. It follows from (4.148 4 Representations of Lorentz and Poincar´ groups e Wµ = 1 1 µνλρ ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . Let us next calculate the commutators between W µ and the ten generators of the Poincar´ group. 2 (4. this is not true for massless theories as we will see shortly. (However.82) The commutator between Pµ and Mνλ introduces metric tensors (see (4.83) that.85) where the order of the operators is once again not important.38)) which vanish when contracted with the anti-symmetric LeviCivita tensor. 2 (4. we have e .

4. Mστ = = 1 λρ ǫ Mλρ . any function of Wµ and. in particular Wµ W µ . Pσ ] = = 1 µνλρ ǫ Pν Mλρ . 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 . will also commute with the generators of translation. Consequently. (4. Pσ 2 1 µνλρ ǫ Pν [Mλρ .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 (−ηλσ Pρ + ηρσ Pλ ) = 2 = 0. We also note that Mµν . Mστ 2 µν 1 λρ ǫ [Mλρ .

88) Equation (4. Mρσ ] Pλ 2 2 1 λρσ ǫ (ηµρ Wσ − ηµσ Wρ ) Pλ 2 ν (4. ǫµνλρ ǫστ ζρ = −ηµσ (ηντ ηλζ − ηνζ ηλτ ) − ηµτ (ηνζ ηλσ − ηνσ ηλζ ) −ηµζ (ηνσ ηλτ − ηντ ηλσ ) . Namely. . Namely. it transforms like a vector which we should expect since it has a free Lorentz index. Mστ ] = Pν M µν . Mστ = [Pν . (4.90) = −ǫµν λρ Wλ Pρ . then. we can now evaluate [W µ . Mστ ] M µν + Pν M µν . 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 . Wν ] = = Wµ . (4. 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. Let us note here.150 4 Representations of Lorentz and Poincar´ groups e = −ηµσ Mντ − ηντ Mµσ + ηµτ Mνσ + ηνσ Mµτ .87) simply says that under a Lorentz transformation. 1 1 λρσ ǫν Mρσ Pλ = ǫνλρσ [Wµ . the operator Mµν behaves exactly like the generators of Lorentz transformation (see (4. Using this. for completeness as well as for later use.30)).89) In other words.87) Here we have used the identity satisﬁed by the four dimensional LeviCivita tensors. it behaves like a second rank antisymmetric tensor under a Lorentz transformation. (4.

of course. pµ . we can choose the momentum operators. 4. cannot represent Casimir operators of the algebra. satisfying . Mστ ] W µ = W µ (ηµσ Wτ − ηµτ Wσ ) + (ηµσ Wτ − ηµτ Wσ ) W µ = Wσ Wτ − Wτ Wσ + Wτ Wσ − Wσ Wτ = 0. The eigenstates of the momentum operators |p are. this would also represent a Casimir operator of the Poincar´ e algebra since Wµ commutes with the generators of translation (see (4. Pµ . In fact. therefore.89) that [W µ Wµ . which we treat separately. the representations can be labelled by the eigenvalues of these operators.4. consequently. Namely. (4. The irreducible representations of the Poincar´ group can be clase siﬁed into two distinct categories. It can be shown that P 2 and W 2 represent the only Casimir operators of the algebra and.86)). To ﬁnd unitary irreducible representa- tions of the Poincar´ algebra. labelled by the momentum eigenvalues. (4. There are other Lorentz scalars that can be constructed from Pµ and Mµν such as Mµν M µν . to be diagonal (they satisfy an Abelian subalgebra). it is easy to check that these do not commute with the generators of translation and.93) However.3.1 Massive representation. (4. we choose the basis vectors of the repe resentation to be eigenstates of the momentum operators. we conclude that if we deﬁne an operator W 2 = W µ Wµ . Therefore.91) which is to be expected since Wµ W µ is a Lorentz scalar.3 Unitary representations of the Poincar´ group e 151 It follows now from (4. 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 ). L2 = Lµ Lµ with Lµ = P ν Mνµ . Mµν M µν . Mστ ] + [Wµ . without loss of generality. Mστ ] = W µ [Wµ .92) then.

using (4. the eigenvalues of W 2 are not so obvious. the eigenvalues of the operator P 2 = P µ Pµ are obvious.95) (4.152 4 Representations of Lorentz and Poincar´ groups e Pµ |p = 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). However. let us study this operator in some detail. 2 2 (4. (4.94) and in this basis. namely. Therefore.96) Here m denotes the rest mass of the single particle state and we assume the rest mass to be non-zero. where p 2 = p µ p µ = m2 . We recall that Wµ = 1 µνλρ 1 ǫ Pν Mλρ = ǫµνλρ Mλρ Pν .97) Therefore. P 2 |p = pµ pµ |p . (4.

2 (4. p 2 = p µ p µ = m2 .98) where we have simpliﬁed terms in the intermediate steps using the anti-symmetry of the Lorentz generators. 0.100) = − m2 M ij Mij .99). Let us note that in the rest frame (4.4. pµ = (m. 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. (4. 0) . 2 Recalling that (see (4.102) The result in (4.26)) Mij = ǫij k Jk .102) can also be derived in an alternative manner which is simpler and quite instructive.101) where Jk represents the total angular momentum of the particle. In this frame. To understand the meaning of this operator. let us go to the rest frame of the massive particle. (4.99) and the operator W 2 acting on such a state. the Pauli-Lubanski operator (4.3 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 . 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 . 0. (4.82) has the form .

therefore. 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. 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. Clearly. Namely. (m. The dimensionality of the representation can also be understood in an alternative manner as follows. we can think of W 2 as being proportional to J 2 and in the rest frame of the particle. all the degenerate states can be labelled not .103) where we have used (1. 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). For a state at rest with momentum of the form pµ = (m. we ﬁnd eigenvalues of W 2 : w2 = −m2 s(s + 1). 2 m 1 iνλρ ǫ Mλρ pν = − ǫijk Mjk = 2 2 m jk = − ǫi Mjk = mJi . It is not hard to see that all possible 3dimensional rotations would leave such a vector invariant.104) which is the result obtained in (4. this simply measures the spin of the particle. Therefore. 2 (4. these would deﬁne an invariant subgroup of the Lorentz group and will lead to the degeneracy of states.12). s). for a massive particle. 0). for a massive particle at rest. It follows now that W 2 = W µ Wµ = W i Wi = m2 J i Ji = −m2 J2 .34) as well as (4. 0. we can ask what Lorentz transformations would leave such a vector invariant. (4. Therefore.154 4 Representations of Lorentz and Poincar´ groups e W0 = Wi 1 0ijk ǫ Mjk pi = 0. would deﬁne the stability group of such vectors. Technically.105) Thus. 0. That is. we see that the representations with p2 = 0 can be labelled by the eigenvalues of the Casimir operators. (4.102).

Namely. a state at rest is an eigenstate of the P0 operator. · · · . −s. p = 0. 4. up to a normalization factor. commute with P0 . s = 0. (4. This can also be seen algebraically. we note that (see (4. (4. let us note from (4. which generate 3-dimensional rotations and are related to the angular momentum operators. we note that (see (4.83) we see that our states in the representation should also satisfy Pµ W µ = 0. −p).3 Unitary representations of the Poincar´ group e 155 just by the eigenvalue of the momentum.3. 0. 1. In contrast to the massive representa- tions of the Poincar´ group. acting on states in such a vector space. Mij ] = 0. · · · and 2 ms = s. 0. the operators Mij . the representations for a massless partie cle are slightly more involved.107) Consequently. (4. from (4. 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.108) However.109) .2 Massless representation. In this case. From the Lorentz algebra. namely.95)) pµ pµ = 0.106) Namely. 1 .30)) [P0 . The basic reason behind this is that the “little” group of a light-like vector is not so obvious. s − 1.103) that. pµ = (p. In closing. (4. −s + 1. but also by the eigenvalues of three dimensional rotations. the three nontrivial Pauli-Lubanski operators correspond to the generators of symmetry of the “little group” in the rest frame. This deﬁnes the 2s + 1 dimensional representation for a massive particle of spin s. Consequently.83)) P 2 = 0. we would have (see (4.4.

W 2 = W µ Wµ = 0. or. it would lead to (see (4. namely. we will not consider such representations. 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. the simultaneous eigenstates of P 2 and W 2 would . consequently. acting on states in such a space.110) In the ﬁrst case. p0 h= (4. Such representations do not correspond to physical particles and. we can easily show that the action of W µ in such a space is proportional to that of the momentum vector. namely.114) This is nothing other than the helicity operator (since L · p = 0) and.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.113) W0 = Comparing with (4. 2 (4. To determine h. (4. for a massless particle if W 2 = 0. therefore. 2 2 (4. in the second case where W 2 = 0 on the states of the representation. On the other hand. W 2 = W µ Wµ = 0.111) where h represents a proportionality factor (operator).112) from which it follows that acting on a general momentum basis state |p (not necessarily restricting to massless states).111) we conclude that in this space J·p . W µ has the form W µ = −hpµ . (4. let us recall that Wµ = 1 µνλρ ǫ Pν Mλρ .88)) 1 0ijk 1 ǫ pi Mjk = ǫijk pi ǫjkℓ Jℓ = −p · J = −J · p.

3 Unitary representations of the Poincar´ group e 157 correspond to the eigenstates of momentum and helicity.4. It follows now from (4. We recognize that in this case. let us note here that in the light-like frame (4.113). W 0 . 2 W0 = W1 = W2 = W3 = (4. (P0 − P3 ) |p = 2p|p . 0. namely. This can be seen intuitively from the fact that the motion of the particle is along the z axis.107). −p) and we would like to determine the “little” group of symmetries associated with such a vector.115) We see from both (4. 2 1 2νλρ ǫ Mλρ pν = −p (M01 − M13 ) . 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 Pauli-Lubanski operator indeed has only three independent components because of the transversality condition (4. (4. the set of Lorentz transformations which would leave this four vector invariant must include rotations around the z-axis. in the massless case.82) takes the form 1 0ijk 1 ǫ Mjk Pi = ǫijk Mjk pi = −pM12 . the Pauli-Lubanski operator (4. (4.116) Let us now determine the dimensionality of the massless representations algebraically. as we had pointed out earlier. Let us recall that we are considering a massless state with momentum of the form pµ = (p. For completeness.83). 2 2 1 1νλρ ǫ Mλρ pν = p (M02 − M23 ) .115) that.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 .103) and (4.117) . 0. We also note from (4. 2 1 3νλρ ǫ Mλρ pν = −pM12 = W 0 . indeed represents the helicity operator up to a normalization as we had noted in (4.

We note that the algebra of the symmetry generators takes the form [M12 . p W1 . M12 ] = 0. the three independent Pauli-Lubanski operators are. (4.90) that in the . Π2 ] = [M01 − M13 . M02 − M23 ] = 0. We note from (4.120) and. M02 − M23 ] |p = 2P2 |p = 0. Π1 ] = [M12 .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. as we had also seen in the massive case. M12 is the generator of rotations around the z axis or in the two dimensional plane and Π1 . namely. from the Poincar´ algebra in (4. (4. comparing with W i . [Π1 .115). To determine the other symmetries of a light-like vector. These represent all the symmetries of the light-like vector (state). Π1 ] |p [P0 − P3 . (4. we see that e [P0 − P3 . let us deﬁne two new operators as W2 . we say that the stability group or the “little” group of a light-like vector is E2 . Π2 ] |p = [P0 − P3 . it is isomorphic to the algebra of the Euclidean group in two dimensions. Clearly. but can be easily understood as follows. we see that up to a normalization. Π2 ] = [M12 . = [P0 − P3 . 3 in (4. therefore.158 4 Representations of Lorentz and Poincar´ groups e Furthermore. the generators of symmetry of the “little” group. This may seem puzzling. [M12 .38). Furthermore. (4. M02 − M23 ] = −Π1 . [P0 − P3 . in fact. M01 − M13 ] |p = 2P1 |p = 0. 2.121) Namely. Π2 have the same commutation relations as those of translations in this two dimensional space. i = 1. also deﬁne symmetries of light-like states. M01 − M13 ] = Π2 .118) so that rotations around the z-axis deﬁne a symmetry of the lightlike vector (state) that we are considering. Thus. E2 (which consists of translations and rotation).

116).111). Clearly.) This corresponds to the one dimensional representation of E2 known as the “degenerate” representation.122) Let us also note for completeness that the Casimir of the E2 algebra is given by Π2 = Π2 + Π2 = (Π1 ∓ iΠ2 )(Π1 ± iΠ2 ).4. As a result. they generate transformations which will leave the momentum basis states invariant. namely.3 Unitary representations of the Poincar´ group e 159 momentum basis states (where pρ is a number).124) (Alternatively. we can say that Π1 |p.86) as the transformations that leave pµ invariant. as we have already pointed out. massless theories with nontrivial spin that are parity invariant would have two dimensional representations corresponding to the highest and the . the space of physical states would also include the state with the opposite helicity (recall that helicity changes sign under space reﬂection. then it follows from (4.121) that Π1 ∓ iΠ2 correspond respectively to raising and lowering operators for M12 . This is. therefore. Let us note from (4. (4. then. the Pauli-Lubanski operators satisfy an algebra and. we see that in the space of light-like momentum states W 2 ∝ Π2 . follows from (4. Π2 correspond to generators of “translation”. Π1 ∓ iΠ2 ] = ±i (Π1 ∓ iΠ2 ) . Namely. On the other hand. can be thought of as generators of some transformations. their eigenvalues can take any value. Since Π1 .122) that spin can take an inﬁnite number of values which. h = 0.123) and comparing with (4. As a result. h = 0 = Π2 |p.123) that (h corresponds to the helicity quantum number) (Π1 ± iΠ2 ) |p. see (3. The meaning of the transformations. Furthermore. [M12 . (4. does not correspond to any physical system. we note from (4. 1 2 (4.146)). such a state would correspond to the highest or the lowest helicity state. if our theory is also invariant under parity (or space reﬂection). h and this is the reason for the assertion in (4. if W 2 = 0 in this space. of course. if W 2 = 0 in this space of states. what we have been investigating within the context of “little” groups.

as we have seen explicitly in the case of massless fermion theories describing neutrinos. On unitary representations of the inhomogeneous Lorentz group. New York (1959). 568 (1947). Group Theory and its Applications to the Quantum Mechanics of Atomic Spectra. Annals of Mathematics 40. if parity (space reﬂection) is a symmetry of the system. spin can only be either parallel or anti-parallel to the direction of motion leading to the one dimensional nature of the representation. V. In other words. E. Incidentally. Let us consider spin as arising from a circular motion. On the other hand. independent of the spin of the particle. Wigner. However. can be seen in a heuristic way as follows. Academic Press.4 References 1. . E. the only consistent circular motion a massless particle can have. 3. it is clear that since a massless particle moves at the speed of light. if the theory is not parity invariant. in general.160 4 Representations of Lorentz and Poincar´ groups e lowest helicity states. Irreducible unitary representations of the Lorentz group. the dimensionality of the representation will be one dimensional. Then. 149 (1939). then we must have states corresponding to both the circular motions leading to the two dimensional representation. Wigner. is in a plane perpendicular to the direction of motion (otherwise. Annals of Mathematics 48. some component of the velocity would exceed the speed of light). 2. the fact that the massless representations have to be one dimensional. Bargmann. in such a case. 4.

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

namely. lim φ(x) → 0. has the form + m2 φ(x) = ∂µ ∂ µ + m2 φ(x) = 0.162 5 Free Klein-Gordon field theory Figure 5. namely. the dynamical equation involving the ﬁeld variables must be covariant under Poincar´ transformations.40)). each ﬁeld must belong to a speciﬁc representation of the Lorentz group. (5. the classical free Klein-Gordon theory where we assume that the ﬁeld variable φ(x) ≡ φ(x. (5. φ(x) = φ∗ (x).2) where φ (x ) denotes the Lorentz transformed ﬁeld. Secondly. e the quantization of the ﬁelds as operators can then be carried out in the Hamiltonian formalism. t) is real and that it behaves like a scalar under a Lorentz transformation. it is generally preferrable to work in the Lagrangian formulation which is manifestly covariant. (5.4) . Given a relativistic dynamical equation. With this in mind. free Klein-Gordon equation with mass m. whenever any ambiguity arises in the quantization. We also assume that the ﬁeld falls oﬀ rapidly at spatial inﬁnity (this is necessary for the theory to be well deﬁned). However. On the other hand.3) |x|→∞ The derivatives of the ﬁeld are also assumed to fall oﬀ at inﬁnity. since the Hamiltonian is not a manifestly covariant concept.1: Oscillating string ﬁxed at both ends. The classical. let us look at the simplest of ﬁeld theories. therefore. We would like to have a relativistic. one reverts back to the Hamiltonian formalism for its resolution. φ (x ) = φ(x). as we have seen (see (1. so far. covariant theory for fundamental physical processes and.

However. 5. Namely. the Lagrangian density can depend at the . 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. L = L (φ(x). (Note that the operator in the parenthesis in the momentum space corresponds to (P 2 − m2 ).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.2 Lagrangian density 163 This equation is clearly invariant under Lorentz transformations. ∂µ φ(x)) . 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. (5. of course. namely.6) Furthermore.) 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.5) where we have written the Lagrangian in terms of a Lagrangian density as L= d3 x L. (5. a Lagrangian density can depend on higher order derivatives. we follow exactly what we normally do in point particle mechanics. (5. 0) and φ(x. However. To that end. P 2 denotes one of the Casimirs of the Poincar´ algebra and m is a constant.7) In general. 0) (a dot denotes a derivative with respect to t). As we have seen. we assume that the Lagrangian density depends only on the ﬁeld variables and their ﬁrst derivatives. 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. for equations which are at most second order in the derivatives.

3)) to simplify the surface term. ti (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. tf ) = 0. 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). We have . ti ) = δφ (x. Note that under an inﬁnitesimal change tf (5.10) Here we have used Gauss’ theorem (as well as the vanishing of ﬁeld variables asymptotically (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.

(5. tf ) . inﬁnitesimal change in the ﬁeld variable. ∂L ∂L − ∂µ = 0. (5. ∂φ(x) ∂∂µ φ(x) (5. As in point particle mechanics. ∂µ ∂ µ φ + m2 φ = 0. δφ (x. the change in the action is given by tf δS = ti d4 x ∂L ∂L − ∂µ ∂φ(x) ∂∂µ φ(x) δφ(x). the action will be stationary under an arbitrary change in the ﬁeld variable subject to the boundary conditions provided (namely. Thus.13) Therefore. (5. namely.12) Thus. ∂µ φ(x)).9). for the Klein-Gordon equation.10) – vanishes because of the boundary conditions (5.15) . we conclude that under an arbitrary.5. (5. ti ) = 0 = δφ (x. δS = 0 only if).14) This is known as the Euler-Lagrange equation associated with the action S or the Lagrangian density L(φ. we can think of the dynamical equations as arising from the minimum principle associated with a given action.11) We note that the “surface term” – the last term in the expression (5.2 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).

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

∂x ˙ p= (5. p) = px − L (x. {x. the basic variables are φ(x) and ∂µ φ(x) and we have from (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. (5. ˙ p = {p. the Lagrangian is a function of the coordinate and the velocity of the particle (a dot denotes a derivative with respect to t). L = L (x.20) The conjugate momentum associated with the dynamical variable x is deﬁned as ∂L . on the other hand. H} . p} . ˙ ˙ (5.22) 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.21) and this leads to the deﬁnition of the Hamiltonian in the form (this is an example of a Legendre transformation) H(x. ˙ (5. in turn. p} = 1.3 Quantization 167 5. H} . x} = 0 = {p. x) . ˙ (5. lead to the dynamical equations in the Hamiltonian form as x = {x.24) In the case of the classical scalar (Klein-Gordon) ﬁeld theory.23) These. In the case of a point particle.19) (xµ in this case merely labels space-time coordinates) . x) .5.

= δ3 (x − y). (5.21).27) H= (5. and a Hamiltonian d3 x H. Π(y)}x0 =y0 . (5. Π(y)}x0 =y0 = 0 = {Π(x). the KleinGordon equation in this case) can be written in the Hamiltonian form as ˙ φ(x) = {φ(x). in turn. H} .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. ∂µ φ(x)) .23)) {φ(x). (5. we deﬁne a momentum canonically conjugate to the ﬁeld variable φ(x) as ∂L . allow us to deﬁne a Hamiltonian density as ˙ H = Π(x)φ(x) − L.26) which will. (5. Note that the Lagrangian density in this case is given by .25) In an analogous manner to the point particle mechanics in (5. φ(y)}x0 =y0 {φ(x).168 5 Free Klein-Gordon field theory L = L (φ(x). ˙ Π(x) = {Π(x). ˙ ∂ φ(x) Π(x) = (5.30) Let us examine all of this in detail for the case of the free KleinGordon theory.29) then. we can show that the dynamical equation (namely. H} .

29)) {φ(x). 2 2 2 (5. = δ3 (x − y). 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) .5. we can identify the conjugate momentum with ∂L ˙ = φ(x). (5. 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 (5.33) 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x).31) Therefore.32) Consequently. Π(y)}x0 =y0 .35) .3 Quantization 169 L = = m2 2 1 ∂µ φ∂ µ φ − φ 2 2 1 ˙2 1 m2 2 φ .34) With the equal time canonical Poisson bracket relations deﬁned to be (see (5. Π(y)}x0 =y0 we obtain = 0 = {Π(x). ˙ ∂ φ(x) Π(x) = (5. φ − ∇φ · ∇φ − 2 2 2 (5. φ(y)}x0 =y0 {φ(x).

Π(y)} x0 =y 0 d3 y Π y. (5.38) . 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 (5.36) where we have used the fact that.170 5 Free Klein-Gordon field theory ˙ φ(x) = {φ(x). H} = Π(x). we also have ˙ Π(x) = {Π(x).29). ˙ Π(x) = ∇ · ∇φ(x) − m2 φ(x) = ∇ 2 φ(x) − m2 φ(x).73)). ∇y φ(y)} + m2 φ(y) {Π(x). Π2 (y) d3 y Π(y) {φ(x). φ(y)} x0 =y0 = d3 y ∇y φ y.37). since the Hamiltonian is time independent (this can be checked easily using the equations of motion and will also be shown in (5. −δ3 (x − y) (5. Furthermore. x0 · ∇y −δ3 (x − y) +m2 φ y.37) Thus. (5. 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).36) and (5. x0 = ∇ · ∇φ(x) − m2 φ(x). the Hamiltonian equations (5. lead to ˙ φ(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). H} = = = = φ(x). in this case. x0 δ3 (x − y) = Π(x).

¨ φ − ∇ 2 φ(x) + m2 φ(x) = 0. However. see (5. ∂µ ∂ µ + m2 φ(x) = 0. φ(y)]x0 =y0 [φ(x).2)) satisfying the equal time commutation relations (remember = 1) [φ(x). the free Klein-Gordon equation (1. This is conventionally referred to as second quantization.40) (which is a second order Euler-Lagrange equation) and this discussion.4 Field decomposition 171 These are ﬁrst order equations as Hamiltonian equations should be and from these. we can expand the ﬁeld operator φ(x) in such a basis. In the second quantization. Π(y)]x0 =y0 .41) of the KleinGordon equation deﬁne a complete basis and. (5. but not the wave function. here we are treating φ(x) as an operator which is quantized and not as a wavefunction as we had done earlier.40) Note that this is the same Klein-Gordon equation as we had studied earlier in chapter 1. the quantization of such a system is quite straightforward. we obtain the second order equation ¨ ˙ φ(x) = Π(x) = ∇ 2 φ(x) − m2 φ(x). In the quantum theory. = iδ3 (x − y). We already know that the plane wave solutions (1. of course. Π(y)]x0 =y0 = 0 = [Π(x). This is. we are supposed to treat φ(x) and Π(x) as Hermitian operators (because we are dealing with real ﬁelds. therefore. Once we have the classical Hamiltonian description of a physical system.39) or.) 5. therefore. or. (Namely. we quantize the “wavefunction”. brings out the classical Hamiltonian description of the classical KleinGordon ﬁeld theory.5.4 Field decomposition Given any complete set of solutions of the classical Klein-Gordon equation. in the ﬁrst quantization we quantize x’s and p’s. we can expand the ﬁeld operator in this basis as . (5.

) Since φ(x) satisﬁes the Klein-Gordon equation ∂µ ∂ µ + m2 φ(x) = 0.41) in the Klein-Gordon equation.172 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 d4 k e−ik·x φ(k).41).44) where the operator a(k) is no longer constrained by the equation of motion. Substituting this back into the expansion (5. diﬀerent. We also emphasize here that the hermiticity properties of a given ﬁeld and its Fourier transform are. 1 (2π) 3 2 3 2 ∂µ ∂ µ + m2 d4 k e−ik·x φ(k) = 0. (5.43) d4 k −k2 + m2 φ(k)e−ik·x = 0. 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). we obtain 1 (2π) or. in general. (5. As a result.46) . Namely. (5. we can denote φ(k) = δ k2 − m2 a(k). This shows that unless k2 = m2 . (5. (5. φ(k) = 0.45) (5. (Recall that 1 k · x = k0 x0 − k · x and the factor (2π)3/2 is introduced for later convenience.41) where we recognize that the operators φ(x) and φ(k) are Fourier transforms of each other (up to a multiplicative factor).42) using the ﬁeld expansion (5.

2: Mass shell on which φ(k) has support.4 Field decomposition k0 173 k2 = m2. (k0 )2 = k2 + m2 . (5.48) Using this relation in (5. or.5. 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 . or.47) Correspondingly. k0 < 0 Figure 5. k0 = ± k2 + m2 ≡ ±Ek .46). the ﬁeld expansion becomes . k0 > 0 k k 2 = m 2 . (5. We note that the argument of the delta function vanishes for k 2 = m2 .

(5. (5.2)). φ† (x) = φ(x). Therefore. 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 . so that we can write 1 (2π) 3 2 a† (−k) = a(k). k) 0 x0 +ik·x = 1 (2π) 3 2 d3 k 1 0 e−iEk x +ik·x a(Ek .52) φ(x) = d3 k −ik·x e a(k) + eik·x a† (k) .50) Let us note that we are dealing with a Hermitian ﬁeld (see (5.49) Changing k → −k in the second term in (5. or. 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) . k) . we obtain a† (k) = a(−k).53) .51). 2k0 (5. 2k0 (5. 2k0 (5.51) Comparing the left-hand side and the right-hand side of (5.174 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 .49) and identifying k0 = Ek > 0.

φ(x) = φ(+) (x) + φ(−) (x).53) as d3 k (2π)3 2k0 φ(x) = e−ik·x a(k) + eik·x a† (k) . conventionally one deﬁnes a(k) .58) 5. eik·x a† (k). In fact.55) and in terms of these operators we can write the ﬁeld operator (5.54) It is clear from (5. (5. (5.5.5 Creation and annihilation operators Given the ﬁeld expansion (5.56) 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. 2k0 (5.56) .5 Creation and annihilation operators 175 Here we are supposed to understand that k0 = Ek = k2 + m2 > 0.54) that a(k) and a† (k) are really functions of the three momentum k alone.57) so that we can write the ﬁeld operator as a sum of its positive and negative energy parts. (5. a(k) = √ 2k0 a† (k) a† (k) = √ . Thus. (5. 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).

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

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

(5. Π(y)]x0 =y0 = − ′ ′ x0 =y 0 (5. a(k′ ) = 0 = a† (k).66) This shows that the operators a(k) and a† (k). which are the coeﬃcients of expansion of the ﬁeld operator in a plane wave basis. Any other redeﬁnition will introduce a multiplicative factor into the commutation relations. a(k). we have a(k). a(k′ ) − e−ik·x+ik ·y a(k). a† (k′ ) = 0. have commutation relations analogous to the annihilation and creation operators of a harmonic oscillator. Thus.178 ′ 5 Free Klein-Gordon field theory − eik·x−ik ·y a† (k). a† (k′ ) = ′ d3 xd3 y √ eik·x−ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x). recalling that the integrals are ime independent.66) can also be obtained more directly from the inversion formulae in (5.55).) a(k). However. it appears that there is an inﬁnite number of such operators in the present case – one for every value of the momentum k.64) i (2π)3 d3 k d3 k ′ k′ 0 4k0 ′ × e−ik·x−ik ·y a(k). a(k′ ) + eik·x+ik ·y a† (k). for example. a† (k′ ) . a(k′ ) − eik·x+ik ·y a† (k). a† (k′ ) + eik·x−ik ·y a† (k). The commutation relations in (5. a† (k′ ) = iδ3 (x − y). 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′ ) = δ3 k − k′ .65) From these. k′ 0 φ(y) − iΠ(y) x0 =y 0 .61). [φ(x). ′ ′ x0 =y 0 (5.

k′ 0 φ(y) + iΠ(y) = d3 xd3 y ′ √ eik·x+ik ·y 3 4k 0 k ′ 0 (2π) × ik0 [φ(x). a† (k′ ) = ′ d3 xd3 y √ e−ik·x−ik ·y 3 4k 0 k ′ 0 (2π) .66) can also be derived in the same manner a(k).5 Creation and annihilation operators 179 = d3 xd3 y ′ √ eik·x−ik ·y (2π)3 4k0 k′ 0 × −ik0 [φ(x). Π(y)] + ik′ 0 [Π(x).5.67) 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′ ). Π(y)] + ik′ 0 [Π(x). we have k′ 0 = k0 . e 4k0 k′ 0 a† (k). a(k′ ) = d3 xd3 y ′ √ eik·x+ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x). e 0 k′ 0 4k (5. The other two commutation relations in (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.

e 4k0 k′ 0 (5. let us look at the Hamiltonian of the system. 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). To understand the meaning of these operators further. Here.69) − a† (k)a(k) + e2ik 0 x0 a† (k)a† (−k) . 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) (5.68) where we have used the fact that k′ 0 = k0 for k′ = −k. 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. Π(y)] − ik′ 0 [Π(x). we have used the fact that.180 5 Free Klein-Gordon field theory × k0 φ(x) − iΠ(x). independent of whether k′ = ±k. .

70) Similarly.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. (5.71) 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′ ) .

72) into the expression for the Hamiltonian (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) .34).72) Substituting (5.71) and (5. (5.69). (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) .

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

73) (namely. = a† (k)a† (k′ ) = a† (k′ )a† (k).O.O. Taking advantage of this.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. N. Thus. : a(k)a(k′ ) : = a(k)a(k′ ) = a(k′ )a(k). the ordering of the operators is ambiguous. N. = = = 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). Let us recall that in the passage from classical to a quantum theory. we deﬁne normal ordering as the ordering in which the creation operators stand to the left of the annihilation operators. a(k)a† (k′ ) a† (k)a(k′ ) a(k)a(k′ ) a† (k)a† (k′ ) = = = = : a(k)a† (k′ ) : = a† (k′ )a(k).O. If we normal order the Hamiltonian using (5.O. : a† (k)a(k′ ) : = a† (k)a(k′ ).77) H N.76) N. (5.78) and we can deﬁne the total number operator for all the oscillators (inﬁnite number of them) in the system as . 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). (for bosons) N. This is conveniently done by normal ordering of the operators. (5. then. : a† (k)a† (k′ ) : (5.76).O. we have from (5.

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. a† (k′ ) a(k′ ) = a(k′ )δ3 (k − k′ ). . d3 k′ N (k′ ) d3 k′ −a† (k′ )δ3 (k − k′ ) = −a† (k).80) [a(k).79) that a(k). a† (k′ )a(k′ ) a(k). N ] = = a† (k). a(k′ ) = −a† (k′ )δ3 (k − k′ ). N = = a(k). N (k′ ) = a† (k). d3 k′ N (k′ ) d3 k′ a(k′ )δ3 (k − k′ ) = a(k). a† (k). a† (k′ )a(k′ ) = a† (k′ ) a† (k). (5. a† (k).79) It is worth noting from the derivation in (5. It now follows from the deﬁnition of the number operators in (5.80) then leads to (5.5 Creation and annihilation operators 185 N= d3 k N (k) = d3 k a† (k)a(k). (5.77) 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). Equation (5. N (k′ ) = = a(k).78) and (5.5.

(5.75)) [a(k).82) Let us consider an energy eigenstate |E of this Hamiltonian satisfying H|E = E|E . a† (k).85) . consequently. (5.83) that E = = = ≥ 0.186 5. the inner product represents the norm of the state a(k)|E ) and. the integrand is positive semideﬁnite. E|H|E E| d3 k Ek a† (k)a(k)|E d3 k Ek E|a† (k)a(k)|E (5. Let us next recall that the operators a(k) and a† (k) satisfy the commutation relations (see (5. H] = Ek a(k). H = −Ek a† (k).77) and (5. which can be achieved by quantizing the system in a box. It is clear from (5.84) This is because both Ek as well as the inner product of states is positive deﬁnite (namely. This shows that the eigenvalues of the Hamiltonian have to be positive semideﬁnite and in the second quantized ﬁeld theory.77) (as describing the quantum free Klein-Gordon theory) of the form H= d3 k Ek a† (k)a(k). let us consider the normal ordered Hamiltonian (5.6 Energy eigenstates 5 Free Klein-Gordon field theory From now on. we do not have the problem of negative energy states that we had in the single particle theory. (5.83) where we are assuming that the state |E is normalized and that the eigenvalue E is discrete for simplicity.

(5. the energy eigenvalue is bounded from below (E ≥ 0). it would appear that by applying the annihilation operator successively. = Ek a(k)|E . Therefore. But as we have seen in (5.88) and we cannot lower the energy any further. or. |Emin .86) − Ha(k)|E H {a(k)|E } = (E − Ek ) {a(k)|E } . or.89) This is. we can lower the energy eigenvalue arbitrarily. = −Ek a† (k)|E . The creation operator a† (k).6 Energy eigenstates 187 As a result.5. a(k)H|E = Ek a(k)|E . H] |E or. we note that a† (k). raises the energy value. we had redeﬁned . Namely. the ground state of the system or the vacuum state (recall that by normal ordering the Hamiltonian. d3 k Ek Emin |a† (k)a(k)|Emin Emin = = 0. therefore. then a† (k) acting on it gives another energy eigenstate with the higher value of energy E+Ek . (5. The annihilation operator a(k) lowers the energy eigenvalue as it should.87) − Ha† (k)|E = (E + Ek ) a† (k)|E In other words. we have [a(k). (5. therefore. if |E is an energy eigenstate with energy eigenvalue E. a† (k)H|E H a† (k)|E = −Ek a† (k)|E . such that a(k)|Emin = 0. . if |E represents an energy eigenstate with eigenvalue E. then {a(k)|E } is also an eigenstate of energy with the lower energy value E − Ek . Similarly. In such a state. H |E or. there must exist a state (energy eigenstate) with a minimum energy. clearly.84). Let us next note that since a(k) acting on an energy eigenstate lowers the energy. (5.

n2 . n2 . kℓ .92) This will be an eigenstate of the total number operator satisfying N |n1 . · · · . · · · . = 0 = 0|a† (k). A general energy eigenstate with higher energy will have the form (up to normalizations) a† (k1 ) n1 a† (k2 ) n2 · · · a† (kℓ ) nℓ |0 . n2 . (5. kℓ = a† (k1 ) √ n1 ! n1 a† (k2 ) √ n2 ! n2 ··· a† (kℓ ) √ nℓ ! nℓ |0 . = 0. (5. = 0. nℓ . k1 .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 . the vacuum state contains no quantum of energy or no particle if energy is related to that of particles. it is clear now that the expectation value of any normal ordered operator in the vacuum state. k1 . 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 . (5. k1 . would vanish. n2 . · · · . (Incidentally. nℓ . kℓ (5. = 1. . nℓ . · · · . one can build up states of higher energy by simply applying the creation operator. conventionally called the vacuum expectation value or vev. k1 .188 5 Free Klein-Gordon field theory the energy of the ground state to be zero). kℓ = d3 k N (k)|n1 . k2 . nℓ .) Given the vacuum state. k2 .93) = (n1 + n2 + · · · + nℓ ) |n1 . k2 . k2 .90) In other words.

kℓ . 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. k2 . (5. k2 .6 Energy eigenstates 189 This relation can be easily shown using the identity (see also (5. · · · . n2 quanta with four momentum k2 and so on. kℓ (5. the state |n1 . the natural deﬁnition from considerations of covariance. nℓ . nℓ . k1 . k2 . a† (k1 ) = n1 a† (k1 ) which also leads to n1 n1 δ3 (k − k1 ) . (5. (5. · · · .) P= then d3 k k a† (k)a(k) = d3 k k N (k). k1 .97) H|n1 . n2 . nℓ . · · · . nℓ . Namely. k1 . kℓ contains n1 quanta with four µ µ momentum k1 . · · · . k1 . n2 . of course. · · · . and . kℓ = (5. kℓ = d3 k Ek N (k)|n1 .96) if we deﬁne the momentum operator as (This is. kℓ . k1 .5. Let us further note that since H= d3 k Ek a† (k)a(k) = d3 k Ek N (k).95) n1 δ3 (k − k1 ) + · · · + nℓ δ3 (k − kℓ ) |n1 . n2 . k1 . nℓ . (a† (k1 ))n1 = a† (k) a(k). However.98) = (n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ ) |n1 . · · · .80)) N (k).94) N (k)|n1 .

99) = (n1 k1 + n2 k2 + · · · + nℓ kℓ ) |n1 . let us analyze the state obtained from the vacuum by applying a single creation operator. = k|k .66) as well as the properties (5.100) We can show in a straightforward manner (using the fundamental commutation relations in (5. k1 . (5. k2 . |k = a† (k)|0 . · · · . n2 . (5. · · · . nℓ .7 Physical meaning of energy eigenstates To obtain the physical meaning of these energy eigenstates. Namely. · · · . p = n1 k1 + n2 k2 + · · · + nℓ kℓ . k1 . k2 .102) It follows from (5. (5. k2 . n2 .101) This state satisﬁes N |k H|k P|k = |k .103) .90)) that these states are orthonormal and deﬁne a complete basis of states for the Hilbert space of the Klein-Gordon theory. kℓ . k1 . nℓ . n2 . kℓ (5.190 5 Free Klein-Gordon field theory P|n1 . kℓ = d3 k k N (k)|n1 .102) that H 2 − P2 |k 2 = (Ek − k2 )|k = m2 |k . 5. = Ek |k . nℓ . (5. these states are eigenstates of H and P with total energy and momentum given by E = n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ .

n2 . k2 ) and so on. of course.105) Since φ(x) is linear in the creation and the annihilation operators. an ordinary function representing the projection of the state |φ(x) on to the one particle state |k . kℓ . This function satisﬁes ∂µ ∂ µ + m2 k|φ(x) = k| ∂µ ∂ µ + m2 φ(x)|0 = 0. we can show that the state |n1 . as a result. k1 . therefore. (5. (5. |φ(x) = φ(x)|0 = φ(−) (x)|0 .104) µ can be thought of as the state with n1 particles with k1 = (Ek1 . (5. it is clear that 0|φ(x)|0 = 0. therefore. Let us next consider the state that is produced by the ﬁeld operator acting on the vacuum. (5.5. · · · . Similarly. Note that the particles described by such states are necessarily identical (although with diﬀerent energy and momentum) and. nℓ .106) In fact. (5.7 Physical meaning of energy eigenstates 191 This state. ﬁeld theories naturally describe systems of many identical particles. k).108) where we have used (5. k1 ). the only nonzero matrix element of the ﬁeld operator involving the vacuum state is given by k|φ(x)|0 = k|φ(x) . k2 .107) This is. satisﬁes the single particle Klein-Gordon equation with positive energy (it provides a representation for the Poincar´ e group) and. we can think of this as the one particle state with four momentum kµ = (Ek . µ n2 particles with k2 = (Ek2 .105) as well as the fact that only the ﬁeld operator φ(x) depends on the space-time coordinates on which the .

If |ψ denotes a state of the system. 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). (5. 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. in the ﬁrst quantized description. |k represents an energy eigenstate of the system describing a single particle (namely. Let us note here parenthetically that the wave function for a single particle is really identiﬁed with φ(x)|k . Therefore.192 5 Free Klein-Gordon field theory Klein-Gordon operator ( + m2 ) can act. In the present case.109) to be the positive energy plane wave solutions of the single particle Klein-Gordon equation. . then.110) In the second quantized description. (5. |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 . 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. the wave function is given by ψ(x) = x|ψ . This brings out the connection between the second quantized theory and the ﬁrst quantized theory. We recognize the function in (5.109) √ with k0 = Ek = k 2 + m2 > 0.

109). Furthermore. Such probabilities are. k′ ) at the coordinates x1 and x2 at a given time. which is what we have seen explicitly in (5. φ (x2 ) = k. xn . · · · . (5. · · · . positive semi-deﬁnite and there is no problem of negative .7 Physical meaning of energy eigenstates 193 Thus. k) and k′ µ = (Ek′ . by construction. the absolute square of the amplitude (with x0 = x0 ) 2 1 k.5. φ (xn ) = φ(−) (x1 ) φ(−) (x2 ) · · · φ(−) (xn ) |0 . The state = φ(−) (x)|0 . such a state contains a superposition of all possible momentum states (which follows from the ﬁeld decomposition in (5. (5. k′ |φ(−) (x1 ) φ(−) (x2 ) |0 .113) which will describe a state with n particles at coordinates x1 . We can similarly construct multi-particle states in the conﬁguration space of the form |φ (x1 ) . As we have seen.111) |φ(x) (5.114) would give the probability for ﬁnding two Klein-Gordon particles with four momenta kµ = (Ek . such conﬁguran 1 2 tion states are automatically symmetric as we would expect for a system of identical Bose particles. (5.112) 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. These states naturally lead to quantum mechanical probabilities which are non-negative. φ (x2 ) . Such states are physically meaningful only if the time coordinates are equal (namely. k′ |φ (x1 ) . the wave functional for a single particle with a deﬁnite energymomentum has the form ψ(φ(x)) = φ(x)|ψ = φ(x)|k .56)). They can also be shown to deﬁne a complete basis and describe what is known as the Fock space for the system. Thus. for example. x0 = x0 = · · · = x0 ).

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

Thus. or. the Fourier transformation turns the partial diﬀerential equation into an algebraic equation which is trivial to solve. . 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). 1 1 −k2 + m2 G(k) = − .117). 4 (2π) (2π)4 G(k) = k2 1 . we obtain ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). The Green’s function is. (2π)4 and substituting them back into (5.119) = J(x).120) d4 k −ik·(x−y) e G(k). 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.8 Green’s functions 195 then the particular solution of the inhomogeneous equation (5.118) d4 y −δ4 (x − y) J(y) (5.5. therefore. or. (5. − m2 (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). (5.121) In other words. deﬁning the Fourier transforms δ4 (x − y) = G(x − y) = 1 (2π)4 d4 k e−ik·(x−y) .

Im k 0 −Ek Ek Re k 0 Figure 5. this is expressed by writing . If we choose a contour of the form shown in Fig. (2π)4 k2 − m2 (5. 5.4.120). 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. or. substituting (5. (5. e Let us note that the integrand of the Green’s function in (5.121) back into (5.122) G(x − y) = = This explicitly shows that the Green’s function is Poincar´ invariant.123) which lie on the real axis in the complex k0 -plane as shown in Fig. this is equivalent to moving the poles inﬁnitesimally into the upper half plane. Therefore.3.3: Poles of the integrand in the complex k0 -plane. 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) . (k0 )2 = k2 + m2 . then.122) has poles at k2 − m2 = 0. k0 = ±Ek . or.196 5 Free Klein-Gordon field theory Thus. 5. Mathematically.

124). or. if we enclose the contour in the lower half plane.125) where we have identiﬁed ǫ = η . iη k − 2 0 2 ≃ k 2 + m2 .124) In this case. We note that if x0 − y 0 > 0. GA (x − y) = lim η→0+ e−ik·(x−y) d4 k . or. then we must close the contour in the lower half plane for the damping of the exponential in (5. the integral will vanish. on the other hand. or.5. 2 Since in this case the poles in (5. (5. for x0 − y 0 > 0. we conclude that GA (x − y) = 0. (2π)4 k2 − m2 − ik0 η (5.126) For x0 − y 0 < 0. we have to close the contour in the upper half plane (for the damping of the exponential) in which . k0 − iǫ = ±Ek . k0 = ±Ek + iǫ. the poles of the integrand occur at k2 − m2 − ik0 η = (k0 )2 − k2 − m2 − ik0 η = 0.4: Choice of the contour in the complex k0 -plane for the advanced Green’s function.124) are in the upper half plane. (5.8 Green’s functions 197 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek + iǫ Ek + iǫ Re k 0 Figure 5. Therefore.

we can express this mathematically by deﬁning e−ik·x d4 k . 5.129) would appear at . GA (x) = θ −x0 d3 k eik·x sin Ek x0 . Explicitly. (5. the poles of the integrand in (5.128) or. For x0 − y 0 < 0. if we choose the contour in the complex k0 -plane as shown in Fig. 4 k 2 − m2 + ik 0 η (2π) GR (x) = lim η→0+ (5. d3 k eik·x (2π)3 Ek for x0 > 0.198 5 Free Klein-Gordon field theory case we can evaluate the integral by the method of residues. we can write (we denote the argument of the Green’s function by x for simplicity) G (x) = A 0.129) In this case. then. sin Ek x0 .127) d3 k eik·(x−y) sin Ek x0 − y 0 .5 or equivalently if we push both the poles to the lower half plane. (2π)3 Ek This is known as the advanced Green’s function and has support only in the past light cone. 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. for x0 < 0. (2π)3 Ek Similarly.

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. we have to close the contour in the upper half plane for the damping of the exponential in (5. (5. k0 = ±Ek − iǫ.130) where we have again identiﬁed ǫ = η . we conclude that GR (x) = 0. Note that if x0 < 0. x0 < 0. if we close the contour in the upper half plane.5. (5. or. or. k0 + iǫ ≃ ±Ek .131) For x0 > 0. since both the 2 poles are in the lower half plane. on the other hand. therefore. In this case.5: Choice of the contour in the complex k0 -plane for the retarded Green’s function. k0 + iη 2 2 2 ≃ k2 + m2 = Ek . 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 ) .129) and. k2 − m2 + ik0 η = (k0 )2 − k2 − m2 + ik0 η = 0. then the integral will vanish. or.

132) e−i(−Ek −iǫ)x + −2Ek 0 +ik·x The overall negative sign in (5. GR (x) = −θ x0 d3 k eik·x sin Ek x0 .128) and (5. Therefore.132) arises because the contour is clockwise in the lower half plane. for x0 > 0.133) that GR (x) = GA (−x). (2π)3 Ek (5. (The advanced and the retarded Green’s functions have similar form except for their support and the overall sign. (5. − d3 k eik·x (2π)3 Ek GR (x) = for x0 < 0. 2 (5.117) with a delta function source. Thus.135) would also deﬁne a Green’s function corresponding to the contour picking up the principal values at the poles. we see that we can write 0.200 = 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 . 1 GA (x) + GR (x) . (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. sin Ek x0 . On the other hand. namely. the average of the two.134) Both the retarded and the advanced Green’s functions satisfy the inhomogeneous equation (5. the diﬀerence between GA (x) and GR (x) would deﬁne a function which would satisfy the homogeneous (Klein-Gordon) equation and . (5.) It is clear from (5.

which corresponds to pushing one of the poles (on the left) to the upper half plane while moving the other (on the right) . it is easily seen from (5. which is more useful in the calculation of scattering matrix elements.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. (5.136) that the Schwinger function is real (G(x))∗ = G(x).138) (5. and is antisymmetric. Quantum mechanically. there is another Green’s function. (5. 5. however. G(x) = −G(−x). (2π)3 Ek This clearly satisﬁes the homogeneous Klein-Gordon equation ∂µ ∂ µ + m2 G(x) = 0. This is deﬁned with the contour in the complex k0 -plane as shown in Fig.8 Green’s functions 201 would not strictly correspond to a Green’s function of the theory.136) + θ x0 = = θ x0 + θ −x0 d3 k eik·x sin Ek x0 .137) Furthermore.5.6. 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. namely. known as the Feynman Green’s function.

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. to the lower half plane. Mathematically.140) occur at k2 − m2 + iη = 0. = ± (Ek − iǫ) .142) . 2 (k0 )2 − Ek + iη = 0. (5. or.140) In this case.6: Choice of the contour in the complex k0 -plane for the Feynman Green’s function.141) η where we have identiﬁed ǫ = 2Ek with η → 0+ . (5. 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. or. there will be a nontrivial contribution independent of whether we close the contour in the upper half or in the lower half plane. the poles of the integrand in (5. (k0 )2 − Ek − k0 = ± Ek − iη 2Ek iη 2Ek 2 ≃ 0. (2π)4 k2 − m2 + iη GF (x) = lim η→0+ (5. this can be implemented by deﬁning d4 k e−ik·x . in this case. or. Since there are poles in both halves of the complex k0 -plane.

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. 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. Therefore. for x0 > 0. if x0 > 0. On the other hand. 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. The Feynman Green’s function clearly has support in both the future as well as the past light cones and can be written as .8 Green’s functions 203 Therefore.145) = −G(+) (x). for x0 < 0.143) = G(−) (x). we have used k ↔ −k in the last step (and we understand that k0 = Ek ). we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 (5. where. Here the contour is clockwise leading to an overall negative sign.5.

151) G(+) (x) = i G(−) (x) = −i .147) Therefore. We note from the structure of G(−) (x) in (5.149) 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.136) as d3 k eik·x sin Ek x0 = G(+) (x) + G(−) (x).143) and (5. (2π)3 k0 (5.146) G(+) (−x) = (5. therefore. (2π)3 (5. we will not go into this in more detail here. (2π)4 k2 − m2 − iη G(x) = lim η→0+ (5. (2π)3 Ek G(x) = (5.143) and (5. namely.143) and G(+) (x) in (5. the Feynman Green’s function is an even function. However. (2π)3 d4 k θ(−k0 )δ(k2 − m2 ) e−ik·x . GF (−x) = GF (x). (5. 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). It is deﬁned as e−ik·x d4 k .148) There is yet another choice of the contour and.150) The functions G(+) (x) and G(−) (x) (see (5.145)) can be written in the manifestly covariant forms d4 k θ(k0 )δ(k2 − m2 ) e−ik·x .145) that we can write the Schwinger function (5.145) that i 2 d3 k eik·x = −G(−) (x). Note from (5.204 5 Free Klein-Gordon field theory GF (x) = −θ x0 G(+) (x) + θ −x0 G(−) (x).

For completeness. (5. Let us recall from (5.154) which will be useful in our later discussions. (5.58) that φ(x) = φ(+) (x) + φ(−) (x).57) that (we recall that k0 = Ek = 0) √ (5.9 Covariant commutation relations To make contact between the quantum ﬁeld theory and the various Green’s functions that we have constructed.9 Covariant commutation relations 205 and are correspondingly known as the positive and the negative energy (frequency) Green’s functions. (2π)3 (5. 5.150) that we can write GA (x) = θ(−x0 )G(x) = θ(−x0 ) G(+) (x) + G(−) (x) . We note from (5.151)) G(+) (x) ∗ = G(−) (x).152) and all the Green’s functions can be expressed as linear combinations of these two fundamental Green’s functions. we note from (5.128).5. GR (x) = −θ(x0 )G(x) = −θ(x0 ) G(+) (x) + G(−) (x) . They satisfy (see (5. (5. where we note from (5.133) and (5.155) k2 + m2 > . let us calculate some of the covariant commutation relations satisﬁed by the ﬁeld operators.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 .153) where ǫ(x) = (θ(x) − θ(−x)) is known as the sign function (or the alternating step function).

d3 k ′ (2π)3 2k′ 0 eik ·y a† (k′ ) ′ d3 k d3 k′ ik·x ik′ ·y † √ √ a (k). a(k).156) Since we know the commutation relations (5.66) between the creation and the annihilation operators. a(k′ ) = 0 = a† (k). a† (k′ ) . a(k). a(k′ ) 0 ′0 2k 2k = 0. a† (k′ ) = δ3 k − k′ . (5. eik·x a† (k).157) d3 k d3 k′ −ik·x −ik′ ·y √ √ e e a(k). φ(−) (y) = d3 k e−ik·x a(k). φ(−) (y) = = 1 (2π)3 d3 k (2π)3 2k0 eik·x a† (k). d3 k ′ (2π)3 2k′ 0 e−ik ·y a(k′ ) ′ (5.206 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(+) (x). φ(+) (y) = = 1 (2π)3 d3 k (2π)3 2k0 e−ik·x a(k). (2π)3 2k0 d3 k ′ ′ eik ·y a† (k′ ) (2π)3 2k′ 0 . it is straightforward to evaluate φ(+) (x). φ(−) (x). a† (k′ ) e e 2k0 2k′ 0 = 0.

9 Covariant commutation relations 207 = = = 1 (2π)3 1 (2π)3 1 (2π)3 d3 k d3 k′ −ik·x+ik′·y √ √ a(k). (5.158) and (5. using the results in (5. we can evaluate [φ(x). φ(y)] = φ(+) (x) + φ(−) (x). It is now clear that the Schwinger function satisﬁes the homogeneous equation µ ∂x µ ∂x + m2 G(x − y) = 0. (5. . the commutator of two ﬁeld operators. 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. φ(+) (y) = −iG(+) (x − y) − iG(−) (x − y) = −iG(x − y) = −i d3 k eik·(x−y) sin Ek x0 − y 0 .161) simply because the ﬁeld operator φ(x) satisﬁes the Klein-Gordon equation. φ(−) (y) + φ(−) (x).158) and (5. is proportional to the Schwinger Green’s function. at unequal times.158) = −iG(+) (x − y).5. furthermore. It follows now that φ(−) (x).159).159) are not restricted to have equal time arguments any more (unlike in (5. (2π)3 Ek (5. φ(+) (y) = iG(+) (y − x) = −iG(−) (x − y).159) Note that the ﬁeld operators in the commutators (5.160) Namely. φ(+) (y) + φ(−) (y) = φ(+) (x).40)) and.

163) where Λµν represents a Lorentz transformation. Namely.138) and (5. (5.165) x0 =y 0 We note that by a Lorentz transformation. for x0 = y 0 d3 k eik·(x−y) sin Ek x0 − y 0 (2π)3 Ek G(x − y)|x0 =y0 = = 0. x0 =y 0 (5. (5. therefore.166) In other words.160) also shows that two ﬁeld operators φ(x) and φ(y) do not commute for arbitrary values of the coordinates.) On the other hand. two measurements at space-like separations should not inﬂuence each other.40)) [φ(x). We also note that . φ(y)]x0 =y0 = −iG(x − y) = 0. for space-like separations.208 5 Free Klein-Gordon field theory ∂µ ∂ µ + m2 φ(x) = 0. This is consistent with our intuitive expectation.162) These commutation relations are known as covariant commutation relations simply because the (scalar) Green’s functions are invariant under Lorentz transformations. the commutator of two ﬁeld operators vanishes. This is known as the principle of microscopic causality. (5. as can be checked even in the simple classical Poisson bracket relations for a one dimensional free particle or a harmonic oscillator. for (x − y)2 < 0. G(Λx) = G(x). (The non-commutativity is a reﬂection of the Hamiltonian dynamics. this relation can also be seen to imply that [φ(x). Equation (5. (5.164) which is expected from the anti-symmetry of the Schwinger function (5. a light signal cannot connect two space-like points and. φ(y)] = 0.160) leads to the familiar relation (see (5. namely.

D. Drell. (5. 5. (5. let us note from (5.167) d3 k ik·(x−y) e = −δ3 (x − y).40). Π(y)]x0 =y0 = x0 =y 0 = −i ∂G(x − y) ∂y 0 x0 =y 0 = iδ3 (x − y). New York. φ(y)] = −iθ(y 0 − x0 ) G(+) (x − y) + G(−) (x − y) = −iθ(y 0 − x0 )G(x − y) = −iGA (x − y). φ(y)] = −iθ(x0 − y 0 ) G(+) (x − y) + G(−) (x − y) = −iθ(x0 − y 0 )G(x − y) = iGR (x − y). θ(y 0 − x0 ) [φ(x). namely. φ(y) [φ(x). (2π)3 This is consistent with our earlier ﬁeld quantization rule (5.10 References 1.150) and (5. which will be useful later. McGrawHill. 1964. ˙ φ(x). Bjorken and S.136). J.5.168) To conclude this section.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). (5.160) as well as the relations in (5. Relativistic Quantum Fields.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 (5.169) .

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

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

∂µ φ(x)) = 0. ∂µ φ′ x′ ). Thus. Symmetries have interesting consequences for continuous transformations. It is clear that in such a case.7) . ∂µ φ′ (x) − L (φ(x).3) includes a very interesting class of transformations where the space-time coordinates do not change. or. ∂µ φ(x)) = 0.3) would imply ′ d4 x′ L φ′ x′ . (6. (6. as indicated in (6.6) which holds for most global space-time symmetries as well as internal symmetries.212 6 Self-interacting scalar field theory S= d4 x L (φ(x). d4 x L (φ(x).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. In such a case.5) and only the dynamical variables of the theory transform. namely. ∂µ φ(x)) = 0. invariance of the action under the inﬁnitesimal forms of the transformations in (6.3). ∂x (6. d4 x L φ′ (x). let us consider an inﬁnitesimal transformation with ∂x′ = 1. x′ µ = xµ . It is worth emphasizing here that (6. − d4 x L (φ(x). ∂µ φ(x)) = ′ d4 x′ L(φ′ x′ . the transformations in (6. (6. ∂µ φ′ x′ δS = or. ∂µ φ′ (x) − d4 x L φ′ (x). 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.4) then. the Euler-Lagrange equations for the primed and the unprimed systems would remain form invariant.

8) which must hold independent of the use of equations of motion for the system under study. Clearly. In fact. ∂µ φ′ (x) − L (φ(x). ∂µ φ′ (x) − L (φ(x).9) we can calculate explicitly (we remember that δφ(x) is inﬁnitesimal and. (6.7) will hold if L φ′ (x). as we will see later. (6.10) (6. ∂µ φ(x)) = L (φ(x). ∂µ φ(x)) = ∂µ K µ .¨ 6. keep only linear terms in δφ(x)) L φ′ (x).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. so that δ(∂µ φ(x)) = ∂µ φ′ (x) − ∂µ φ(x) = ∂µ δφ(x). 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). (6. therefore. ∂µ φ(x)) + δφ(x) −L (φ(x). ∂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) . this is indeed the case for internal symmetry transformations.11) ∂L ∂L + δ(∂µ φ(x)) ∂φ(x) ∂∂µ φ(x) = δφ(x)∂µ = ∂µ δφ(x) . This is quite general and it is possible to have symmetry transformations for which K µ = 0. (6. On the other hand.

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

In the operator language. δxµ = x′µ − xµ = ǫµ . The converse is also true. Q and H can have simultaneous eigenstates. 6. Q corresponds to the generator of the inﬁnitesimal symmetry transformations of the theory. consequently. the time independence of Q also means that it commutes with the Hamiltonian of the theory [Q.17) and. with the usual assumptions on the asymptotic fall oﬀ of the ﬁeld operators (see (5. Namely. (6.1 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. . (6. there exists a conserved charge which is the generator of these inﬁnitesimal symmetry transformations. As an example of the consequences of N¨ther’s theorem. In this case. H] = 0. the integral of a total divergence (∇ · J) vanishes. let us consider the simple case of an inﬁnitesimal o global space-time translation deﬁned by xµ → x′µ = xµ + ǫµ .¨ 6.3)).18) where the parameter of translation ǫµ is assumed to be inﬁnitesimal and constant (global). or. if there exists a conserved charge in a given theory. if we have a continuous symmetry in the theory (transformations under which the action is invariant).1 Space-time translation. then it generates inﬁnitesimal transformations (through commutation relations) which deﬁne a symmetry of the theory. we are dealing with a scalar ﬁeld. This is known as N¨ther’s theorem and is quite important in the study of symmeo tries in dynamical systems. Since. (6.16) Here we have used the fact that. Thus.

∂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) (6.216 6 Self-interacting scalar field theory φ′ x′ = φ (x) .20).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). (6. We see that since the change in the Lagrangian density is a total divergence.20) where in the last step. = − φ′ x′ − φ′ (x) = −ǫµ ∂µ φ′ (x) (6. where we have used (5.21) = −ǫµ ∂µ φ(x) = −ǫµ ∂µ L = ∂µ K µ .10).21) that (ǫµ is a constant parameter) K µ = −ǫµ L. (6.8)) L φ′ (x). we note that under the transformation (6. We can now identify from (6.11) and (6.7) and (6. then.19) and in this case we obtain the change in the ﬁeld to correspond to (see (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. With this. we can now calculate explicitly the change in the Lagrangian density (see (6.22) . the action is invariant under inﬁnitesimal translations which deﬁne a symmetry of the system. ∂µ φ′ (x) − L (φ(x). Next.

Therefore. T µν = T νµ .13)) ∂L − Kµ ∂∂µ φ(x) ∂L + ǫµ L ∂∂µ φ(x) ∂L − η µν L ∂∂µ φ(x) = −ǫν T µν . depending on the parameter of transformation.24) µ Jǫ (x) = δφ = −ǫν (∂ν φ(x)) = −ǫν (∂ ν φ(x)) It is clear. (6.1 Nother’s theorem 217 δφ(x) ∂L ∂L = −ǫν (∂ν φ(x)) .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 . therefore. 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. ∂∂µ φ(x) T µν = (∂ ν φ(x)) (6.23) so that we obtain the current associated with the symmetry transformation. that the conserved current independent of the parameter can be identiﬁed with ∂L − η µν L. namely.27) . (6. (6.25) which can be easily checked to satisfy (using the equations of motion) ∂µ T µν = 0.¨ 6. to be (see (6.). even when the naive N¨ther o procedure does not lead to a symmetric stress tensor. ∂∂µ φ(x) ∂∂µ φ(x) (6.

would correspond to the components of a four vector which we identify with the energy-momentum operator as Pµ = d3 x T 0µ . the current is a tensor one rank higher than the parameter of transformation. in this case. P µ . in the present case. which. For the free Klein-Gordon theory. ∂∂µ φ(x) and this leads to the explicit form for the stress tensor (6.15)). should generate inﬁnitesimal space-time translations. is manifestly symmetric. the conserved charges associated with the symmetry transformation (see (6.218 6 Self-interacting scalar field theory The conserved current.30) T µν = ∂ ν φ(x) ∂L − η µν L ∂∂µ φ(x) (6. let us recall that the Lagrangian density has the form L= m2 2 1 ∂µ φ∂ µ φ − φ . .25) (6. for the free theory. (6.29) Therefore. we obtain ∂L = ∂ µ φ(x). in this case.) As a result. (In general.28) This is consistent with our physical intuition that the energy-momentum. We note that. is a second rank tensor simply because the parameter of transformation is a four vector. 2 2 (6.31) = ∂ ν φ(x)∂ µ φ(x) − η µν L = T νµ .

33) To include interactions. ˙ d3 x ∇φ(x)φ(x).32) We see that P 0 = H coincides with the form of the Hamiltonian we had derived earlier in (5. since the free Lae grangian density is invariant under the discrete transformation φ(x) ↔ −φ(x).97) and it can be checked that P in (6.32) corresponds exactly to the former expression when normal ordered and expressed in terms of creation and annihilation operators. we note that the Lagrangian density for the interaction must be invariant under Lorentz transformations as well as translations (Poincar´ invariant). (6.2 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 . 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).34) .6.34). 2 2 L0 = (6.2 Self-interacting φ4 theory 219 P0 = = = = H. (6. Furthermore. We had also given an expression for the momentum operator in terms of creation and annihilation operators in (5.

(6.220 6 Self-interacting scalar field theory we would like to preserve this symmetry in the interactions as well. therefore. would lead to a potential (and. a Hamiltonian) which is unbounded from below for any value of the coupling constant g. the action is dimensionless. for various other reasons we can show that (6. To explain this brieﬂy. is there so that the Hamiltonian will be positive deﬁnite.35) represents the only meaningful interaction term for such a quantum ﬁeld theory.35) The restriction on the coupling constant.) As a result. 4! LI = − λ > 0.37) and has. λ. the simplest interaction Lagrangian density involving only the scalar ﬁelds which we can think of has the form λ 4 φ (x). which does not respect the discrete symmetry (6. Thus.34). let us introduce the concept of canonical dimensions. the same canonical dimension as . in units of = c = 1 (which we have been using). then. we can show that the canonical dimension of any variable can be expressed in powers of an arbitrary mass dimension [M ]. . Let us also g note here that an interaction Lagrangian density of the form − 3! φ3 . Consequently. in fact. 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. (We will see this shortly. With these conditions. which in turn allows us to deﬁne a vacuum state of the theory. (6. the fully self-interacting theory of a real Klein-Gordon ﬁeld is described by a Lagrangian density L = L0 + LI . (6. In these units.

or. d4 x = [M ]0 .6. therefore.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.35) is dimensionless. we obtain λ 4 φ = [M ]0 . has a canonical dimension 1 and. (The dimensionality of the ﬁeld variable depends on the number of space-time dimensions which we will see later. (6. let us note that for a monomial interaction action of the form .40) Thus. d4 x [∂µ ] [φ] [∂ µ ] [φ] = [M ]0 . (6. 4! [SI ] = − or. 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. (6.35). From the canonical dimension of the interaction term in the Lagrangian density in (6. [∂µ ] = [M ].) The mass parameter. or. [λ] = [M ]0 . [φ] = [M ]. or. [M ]−4 [λ] [M ]4 = [M ]0 . or.2 Self-interacting φ4 theory 221 [L] = [T ] = [xµ ] = [M ]−1 . d4 x d4 x [λ] ([φ])4 = [M ]0 . ([φ])2 = [M ]2 . we conclude that the coupling constant or the interaction strength for the φ4 -self-interaction in (6. or. in these units. the Klein-Gordon ﬁeld variable in four dimensions has a canonical dimension 1. [M ]−4 [M ][φ][M ][φ] = [M ]0 . of course. In general.39) In other words.

(6.222 6 Self-interacting scalar field theory g ˜ SI = − n! d4 x φn . or. for example.32)) ∂L ˙ = φ(x). 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. [g] [M ]−4 [M ]n = [M ]0 . In such a case. Such theories are known as non-renormalizable theories.41) Therefore. Since the interaction Lagrangian density does not involve derivatives of ﬁelds.) This. ˜ SI = [g] d4 x [φn ] = [M ]0 . for n > 4. restricts n ≤ 4 and. consequently. the coupling constant of the interaction Lagrangian density will have inverse dimensions of mass. the coupling constants in the theory cannot have dimensions of inverse mass. the φ4 interaction is the only physically allowed interaction in this case.42) so that from (6.34)). [g] = [M ]4−n . consistent with our symmetry requirements (see. or. ˙ ∂ φ(x) Π(x) = (6. therefore. If we want to restrict to renormalizable theories which can give rise to meaningful physical predictions.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! . (6. (We will see this in a later chapter when we discuss renormalization of quantum ﬁeld theories. the canonical momentum conjugate to the ﬁeld variable of the theory continues to be (see (5.

the observables correspond to expectation values of Hermitian operators in quantum states. As a result. = which leads to (6. 6.43) H = = d3 x H d3 x m2 2 λ 4 1 2 1 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! (6. the ground state of the free Hamiltonian will not be stable under perturbations. the operators carry time dependence while the state vectors do not and the dynamical equations are given by the Heisenberg equations of motion. 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).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 . Consequently. neither the states in the Hilbert space nor the operators acting on state vectors are observables. also the . t) carries time dependence. therefore.44) is positive deﬁnite and. This is why we restrict to λ > 0 in the interacting theory. Our discussion so far has been within the context of the Heisenberg picture where the ﬁeld operator φ(x.44) = H0 + HI . Rather.3 Interaction picture and time evolution operator In quantum mechanics as well as in quantum ﬁeld theory. of course. In this picture.6. each term in the integrand in (6. This leads to diﬀerent possible pictures for describing the same quantum mechanical system through distinct time evolutions. 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. For λ < 0. There is. the Hamiltonian will be bounded from below leading to a meaningful vacuum state. It is now clear that for λ > 0. on the other hand. as we know. the Hamiltonian is indeﬁnite because of the interaction term.

for example.45) that (H) ψ|O(H) (t)|ψ (H) = (S) ψ(t)|O(S) |ψ(t) (S) .46) In writing (6.45)) and is. for example. of course.48) The total Hamiltonian operator in the Schr¨dinger picture coino cides with that in the Heisenberg picture (see. The two pictures are related by the unitary transformation = e−iHt |ψ |ψ(t) (S) (H) . t0 ) = e−iH(t−t0 ) . (6. namely.45). (6.49) .47) O(S) = O(H) (0). (6. t) eiHt = φ(H) (x. In the Schr¨dinger picture. in the Schr¨dinger picture. 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. Thus. 0). (6. time independent. It follows from (6.224 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.45) O(S) = e−iHt O(H) (t)eiHt . (6. we have assumed that at t = 0 both the pictures coincide. |ψ(0) (S) = |ψ (H) . the Klein-Gordon ﬁeld o operator will be given by φ(S) (x) = e−iHt φ(H) (x. (6. 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.

.3 Interaction picture and time evolution operator 225 However. o Here. intermediate between the Schr¨dinger and the Heisenberg pictures. (IP H0 ) (t) = eiH0 t (S) H0 e−iH0 (S) t (S) = H0 . therefore. in some sense. in relativistic interacting ﬁeld theories. once again. (IP ) (IP ) (6. (However. O(IP ) (0) = O(S) = O(H) (0). this picture is useful.50) Note. O(IP ) (t) = eiH0 = eiH0 (S) O(S) e−iH0 e t (S) (S) t −iHt O(H) (t)eiHt e−iH0 . both the operators as well as the state vectors carry time dependence. In this case.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. 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) . the more convenient description goes under the name of interaction picture which is. 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. (6. (6. that we can identify |ψ(0) (IP ) = |ψ(0) (S) (S) = |ψ (H) .6.) Although the Heisenberg picture has a manifestly covariant description. by deﬁnition.51) In the interaction picture. it is worth emphasizing here that for certain applications. we deﬁne |ψ(t) (IP ) = eiH0 (S) t t |ψ(t) (S) = eiH0 (S) (S) t −iHt e |ψ t (H) .

is governed by the Schr¨dinger equation with the interaction o Hamiltonian playing the role of the Hamiltonian. t0 ) = eiH0 (S) t −iH(t−t0 ) −iH0 e e (S) t0 . (6. This shows that once we deﬁne commutation relations for the operators in the free theory. In this picture.50) to have the explicit form U (t.54) In other words. t0 ) |ψ (t0 ) .53) Similarly. (6.56) then.51) that the free Hamiltonian in the interaction picture is time independent and.49) and (6.55) can be seen from (6. (6. the time evolution of the state vectors. H0 = 1 (IP ) (IP O (t). the ﬁeld operators can have a plane wave expansion. therefore. if we deﬁne the time evolution operator through the relation |ψ(t) = U (t. it is easy to show that the time evolution operator satisﬁes the properties. . while the dynamical evolution of the operators is governed by the free Hamiltonian through the Heisenberg equations of motion. in the interaction picture they continue to hold even in the presence of interactions. i (6. let us drop the superscript (IP ) with the understanding that we are nonetheless working in the interaction picture.55) where the time evolution operator in the interaction picture in (6. We note from (6. in this picture. From now on. H0 ) . 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).226 6 Self-interacting scalar field theory (IP HI ) (t) = eiH0 (S) t (S) HI e−iH0 (S) t .

it follows that the total Hamiltonian is time independent in the interaction picture. let us look at the second order term in (6. t0 ) = HI (t)U (t. t0 ) t0 t t1 dt1 HI (t1 ) + (−i)2 dt1 t0 t0 dt2 HI (t1 ) HI (t2 ) +··· + ··· +(−i)n +··· .52) we see that the time evolution operator in (6.58) iteratively subject to the initial condition in (6. t0 ) = U (t0 .59) and note that . t0 ) t0 ½−i ½−i t dt1 HI (t1 ) t0 t t0 ½−i t1 dt2 HI (t2 ) U (t2 . U † (t. U (t2 . t0 ) = U (t2 .3 Interaction picture and time evolution operator 227 U (t. (6.6. t t1 t0 tn−1 t0 dt1 t0 dt2 · · · dtn HI (t1 ) HI (t2 ) · · · HI (tn ) (6. t0 ) = U −1 (t. i ∂t i or. t0 ) . t) . (6. t) = ½.55) satisﬁes the equation (this also follows from (6. (Since H0 is time independent in the interaction picture (see (6. t0 ) = ½ − i = = dt1 HI (t1 ) U (t1 . to obtain t U (t.50)) ∂|ψ(t) = HI (t)|ψ(t) .51)).56) with the use of (6. from (6.) We can now solve (6.58) We note from (6.59) To bring this to a more convenient form. t1 ) U (t1 . ∂t ∂U (t.57) Furthermore. t0 ) .57).53) that the interaction Hamiltonian in the interaction picture is time dependent.

61) with the operator at later time standing to the left of the operator at earlier time.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 ). In a similar manner. t0 (6. we can show that the n-th term in the series in (6.59) can be written as . (6.228 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 )) .

62) Therefore. For example. t0 t0 dt1 · · · dtn HI (t1 ) · · · HI (tn ) U (t. (6. in the scattering of particles in relativistic quantum ﬁeld theory.63). t0 ) = T e t0 dt′ HI (t′ ) ..59) takes the form t U (t. we also assume that the incoming particles at t = −∞ as well as the outgoing particles at t = ∞ are described by free particle states.) 6.4 S-matrix 229 (−i)n n! t t0 t ··· t0 dt1 · · · dtn T (HI (t1 ) · · · HI (tn )) . −i Rt . (This is a formal deﬁnition which is to be understood in the sense of the expansion described in (6. we can implement this by modifying the interaction Hamiltonian as (η) HI (t) → lim HI (t) = lim e−η|t| HI (t).4 S-matrix In the non-relativistic quantum mechanical scattering problems. 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.63) In other words. the iterative solution for the time evolution operator in (6.6. Similarly. we assume that the initial and the ﬁnal states are plane wave states corresponding to free particles. 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. η→0+ η→0+ (6. (6.64) .

−∞)|i = S|i . (6. As a result.67) which has .230 6 Self-interacting scalar field theory so that in the inﬁnite past as well as in the inﬁnite future.67) Therefore. (6. the S-matrix (or the scattering matrix) of the theory can be identiﬁed with the time evolution operator (6. the interaction vanishes. (6. 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. 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 state into which this will evolve at t = ∞ is deﬁned from (6. −∞). which is the deﬁnition of the scattering amplitude.65) Then. −∞)|ψi (−∞) = U (∞.55) to be |ψi (∞) = U (∞. (6.66) where we have identiﬁed S = U (∞. The condition (6. is obtained from (6.68) Consequently. the probability amplitude for an initial state |i to be in the ﬁnal free particle state |f at t = ∞. Let us denote the initial state at inﬁnite past as the free particle state |ψi (−∞) = |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. −∞)|i = f |S|i .64)) leads to the same result for the rate of transition in the physical scattering of particles.

say.69) Furthermore.69) through the appropriate boundary conditions.63)) −i ∞ R S = U (∞. this state would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (in) = U (0. (6. the S-matrix (the scattering matrix) is unitary since the time evolution operator is (see (6. Then. Let us see this explicitly in the case of the linear term in the expansion of the time evolution operator in (6. −∞)|Ψ(−∞) (in) .64).57)). Let us suppose that at t → −∞ we have a free incoming state denoted as |Ψ(−∞) (in) = |i . These states can be constructed by noting that the three pictures (Heisenberg. they are uniquely given by the states (in any picture) at t = 0. These are asymptotic free states as t → −∞ and t → ∞ respectively. the linear contribution to the right-hand side of (6.51)) and o since the Heisenberg states are time independent.4 S-matrix 231 the explicit perturbative expansion of the form (see (6.63). −∞) is. in (6. In this case.64) provides a regularized meaning to the oscillatory terms in (6. This also leads to the notion of “in” and “out” states which are quite important in a formal description of scattering theory.70) will have the form . in general.6. not well deﬁned if we do not use an adiabatic interaction of the form.70) where the time evolution operator U (0. −∞) = 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+ +··· . Schr¨dinger and interaction) coincide at t = 0 (see (6. It is clear from the expansion above that the adiabatic switching of the interaction in (6. (6.

from their deﬁnitions in (6.72) we see that even though both the “in” and the “out” states are related to the same Heisenberg state.70) and (6. 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 ) .70) and (6. Rather. (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.72) and an analysis as in (6.72) we see that they are related to each other as . From (6. We also note that the integral in (6.71) would show that in this case.64) naturally provides a regularization for the formal deﬁnition of the time evolution operator U (0. then this would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (out) = U (0. In a parallel manner.71) is not deﬁned at the lower limit in the absence of the adiabatic factor eηt .71) (IP where in the intermediate steps. they are not identical because their deﬁnitions use diﬀerent regularizing factors. −∞). ∞) would not be well deﬁned at the upper limit. ∞)|Ψ(∞) (out) . (6. This clariﬁes how the boundary condition (6. Furthermore. if we assume that as t → ∞. we have a free outgoing state |Ψ(∞) (out) .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) . the “in” and the “out” states deﬁne a complete space of states at t → −∞ and t → ∞ respectively. the regularizing factor would have to be e−ηt without which U (0.

First.5 Normal ordered product and Wick’s theorem The normal ordered product. 0)U (0. such a calculation will involve time ordered products of normal ordered products and this is where the Wick’s theorem comes in handy.74) where the T-matrix represents the true nontrivial eﬀects of scattering. 6. therefore. since our (interaction) Hamiltonian is assumed to be normal ordered. by a phase since the S-matrix is unitary). we will simplify the time ordered product of factors of normal ordered terms. we need to evaluate the matrix elements of time ordered products of operators between free particle states. φ(+) . −∞)|Ψ(−∞) = U (∞. (6. This shows that the “in” and the “out” states are related by the S-matrix in (6. ∞)|Ψ (H) = U (∞. It is clear from this discussion that in calculating the S-matrix elements.5 Normal ordered product and Wick’s theorem 233 |Ψ(∞) (out) = U † (0. Furthermore.76). we will derive a simple relation for the product of factors each of which is normal ordered and then using this relation. The calculation can be carried out in two steps.73) where we have used the properties of the time evolution operator given in (6.69) corresponds to no scattering at all. we can deﬁne S − ½ = T. simply corresponds to arranging factors in the product so that the creation operators stand to the left of the annihilation operators. Later we will see that the Feynman rules are a wonderful and simple way of systematizing these results. Correspondingly. In terms of ﬁeld operators. (6.57). −∞)|Ψ(−∞) (in) = S|Ψ(−∞) (in) . 0)|Ψ (in) (H) = U (∞. The ﬁrst term in the expansion of the S-matrix in (6.67) (and.6. Sf i . contains the annihilation operator while . 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. as we have deﬁned earlier in (5.

N φ(−) (x)φ(+) (y) ≡ : φ(−) (x)φ(+) (y) : = φ(−) (x)φ(+) (y). using the relations in (6. as far as bosonic operators are concerned. Furthermore.77) It is clear.76)). N φ(+) (x)φ(−) (y) ≡ : φ(+) (x)φ(−) (y) : = φ(−) (y)φ(+) (x).76) Since the vacuum state is. that the vacuum expectation value of any normal ordered product vanishes.75). φ(−) . (6. by deﬁnition. N φ(−) (x)φ(−) (y) ≡ : φ(−) (x)φ(−) (y) : = φ(−) (x)φ(−) (y).234 6 Self-interacting scalar field theory the negative energy part. the order of the factors inside a normal ordered product is not important. for the bosonic scalar ﬁeld. annihilated by the annihilation operator. we note from (6. (6. 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).76) that .75) We note that. For example. therefore. (6. Thus. we have N (φ(x)) ≡ : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). N φ(+) (x)φ(+) (y) ≡ : φ(+) (x)φ(+) (y) : = φ(+) (x)φ(+) (y). we can equivalently say that φ(+) (x)|0 =0= 0|φ(−) (x). has the creation operator so that this prescription is equivalent to the deﬁnition of normal ordering in (5.

in the simple example of the product of two ﬁeld operators. does not carry any nontrivial quantum number.79) where we have used (5. To see how we can write a product of operators in normal ordered form. 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).80) with the pairing of the two ﬁeld operators deﬁned to be related to the invariant positive energy Green’s function in (5.5 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 = 0.78) Therefore. let us note that. (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.151) as φ(x)φ(y) = −iG(+) (x − y).81) . Let us denote (6. (6. φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) = : φ(x)φ(y) : −iG(+) (x − y).79) as φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y). Furthermore.158) and (6. (6. therefore.76). (6. This gives another reason to choose the prescription of normal ordering for physical observables. this also shows that if we can rewrite any product of operators (including normal ordered factors) in terms of normal ordered terms.6. the calculation of S-matrix elements will simplify enormously.

φ(−) (y) = = : φ(x)φ(−) (y) : + φ(+) (x).83) We recognize that the ﬁrst term in (6. : φ(x) : : φ(y) : = : φ(x) : φ(+) (y) + : φ(x) : φ(−) (y).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).84) (6. (6. Since we choose the Hamiltonian to be normal ordered. this will be necessary in developing a perturbation expansion for the quantum ﬁeld theory where the interaction Hamiltonian is normal ordered. We know that by deﬁnition : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). we note from (6.82) because the vacuum expectation value of the normal ordered product in (6.85) and (6.80) is zero. Clearly.84) simply corresponds to a normal ordered product : φ(x) : φ(+) (y) = φ(x)φ(+) (y) = : φ(x)φ(+) (y) : . but since it is a c-number function.86) we obtain .86) (6. the second term in (6. On the other hand. (6. This brings out yet another connection between Green’s functions and the quantum ﬁeld theory. Therefore. (6.84) gives : φ(x) : φ(−) (y) = φ(x)φ(−) (y) = φ(−) (y)φ(x) + φ(x). let us next see how we can combine a product of normal ordered factors into normal ordered terms.85) Thus adding the two terms in (6. φ(−) (y) : φ(x)φ(−) (y) : − iG(+) (x − y).80) that we can also identify φ(x)φ(y) = 0|φ(x)φ(y)|0 = −iG(+) (x − y).

let us note that . : φ(x)φ(y) : −iG(+) (x − y) : φ(x)φ(y) : + φ(x)φ(y).79) and (6. of course. (6.87) This is. φ(−) (z) = : φ(x)φ(y)φ(z) : + φ(+) (x).87) shows how a product of two simple normal ordered factors can be expressed as a normal ordered term and a pairing. what we had seen earlier in (6. (6.80). namely. 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) : .6.89) Finally. 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(−) (z) φ(y) +φ(x) φ(+) (y). Next. φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + : φ(x)φ(y)φ(−) (z) : + φ(x)φ(y) + iG(+) (x − y) . (6.88) 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).

φ(−) (w)]φ(+) (z) + φ(−) (z)φ(−) (w) : φ(x)φ(y) : +[: φ(x)φ(y) :. 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). 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) :. φ(−) (z)]φ(+) (w) + φ(−) (w) : φ(x)φ(y) : φ(+) (z) +[: φ(x)φ(y) :. φ(−) (z)]φ(w) + [: φ(x)φ(y) :.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) :. (6. φ(−) (w)]φ(+) (z) +φ(−) (z)[: 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 .

92) Similarly.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) : . that (6. (6. 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) : .6. It follows. 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. For example.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). we already know from (6.80) that φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y). 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. therefore.

93) In other words. we have already noted in (6. (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.89) that .) 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). For example. (6. In general. 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). 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 ) : + ··· .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. This construction can also be extended to the case where only some of the factors are normal ordered.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.

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 .

2 257 .2) can be inverted to give 1 √ φ(x) + φ† (x) . φ† (x) = φ(x). The relations in (7. 2 1 √ (φ1 (x) − iφ2 (x)) .3) φ1 (x) = i φ2 (x) = − √ φ(x) − φ† (x) . ∂µ ∂ µ + m2 φ† (x) = 0. In this case.Chapter 7 Complex scalar ﬁeld theory 7. 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)) . 2 (7. there are now two distinct equations. the dynamical equations take the forms ∂µ ∂ µ + m2 φ(x) = 0.1 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. spin zero (scalar) ﬁelds which are Hermitian. (7.1) Therefore. 2 (7.2) φ(x) = φ† (x) = where φ1 (x) and φ2 (x) are two distinct.

7) . independent spin zero Klein-Gordon ﬁelds (and.5) L = = = ∂µ φ† ∂ µ φ − m2 φ† φ.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 (7.258 7 Complex scalar field theory It follows now that we can express the dynamical equations in (7. The Lagrangian density which will give the dynamical equations in (7. the equal time canonical commutation relations follow to be [φi (x). ∂µ ∂ µ + m2 φ2 (x) = 0. Πj (y)]x0 =y0 = iδij δ3 (x − y). Therefore. ˙ ∂ φi (x) i = 1.6) Consequently. two distinct particles) with degenerate mass. such a system equivalently describes two free. [φi (x).5) because distinct degrees of freedom are expected to commute. φj (y)]x0 =y0 = 0 = [Πi (x). If we treat φ1 (x) and φ2 (x) as the independent variables. (7.4) Namely.1) or (7.) We can quantize this theory in one of the two equivalent ways. therefore. 2. Πj (y)]x0 =y0 .1) in terms of the real ﬁeld variables as ∂µ ∂ µ + m2 φ1 (x) = 0. (7. then we can deﬁne the momenta conjugate to the two ﬁeld operators as Πi (x) = ∂L ˙ = φi (x). (Note that we have not been careful about the order of the factors in rearranging terms in (7. 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. (7. and is manifestly Hermitian.

(7.9) Alternatively. then we can deﬁne the momenta conjugate to the ﬁeld operators from (7.7. 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.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)) . 2 2 . if we treat φ(x) and φ† (x) as the independent dynamical variables.6). 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 . 2 2 2 i (7.1 Quantization 259 In this case.

in this case. (7. take the forms (we do not write the equal time condition explicitly for simplicity.56)). φ(y)] = φ(x). in the present case.2 Field decomposition As in the case of the Hermitian (real) Klein-Gordon ﬁeld. (7. 7.11) Furthermore. φ(x). Π† (y) = Π† (x).10) 2 2 As a result. using (7. in this case. φ† (y) = φ† (x). Π(y) = iδ3 (x − y). we can expand the ﬁeld φ(x) in the basis of plane wave solutions of the Klein-Gordon equation (see (5.12) = Π† Π + ∇φ† · ∇φ + m2 φ† φ.8) and H= d3 x H. which can be compared with (7.13) Note that the Hamiltonian. Π† (y) = φ† (x). Π† (y) = 0. 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 (7.11)) [φ(x).10). is manifestly Hermitian as it should be even though the dynamical variables are not. of the form .5) and (7. φ† (y) = 0. (7.260 Π† (x) = ∂L ˙ = φ† (x) ˙ ∂ φ(x) 7 Complex scalar field theory 1 1 ˙ ˙ = √ φ1 (x) − iφ2 (x) = √ (Π1 (x) − iΠ2 (x)). the equal time canonical commutation relations. Π(y)] = Π(x). [Π(x). although it is to be understood that x0 = y 0 in all of the relations in (7.

14) with k0 = Ek = k2 + m2 > 0. As before. a† (k′ ) . e−ik·x b(k) + eik·x a† (k) . aj (k′ ) = 0 = a† (k).15) Given the decomposition in (7.66)) ai (k). with i i = 1. 2 1 √ (a1 (k) − ia2 (k)) . 2. i j ai (k). The commutation relations between these creation and annihilation operators can then be obtained from the canonical quantization relations to be (see (5. j (7. ai (k) and a† (k). it is clear that the ﬁeld operators φ(x) and φ† (x) can also be expanded in the plane wave basis as (see (7.2 Field decomposition 261 φi (x) = d3 k (2π)3 2k0 √ e−ik·x ai (k) + eik·x a† (k) . i i = 1. (7.16) where we have deﬁned a(k) = b(k) = 1 √ (a1 (k) + ia2 (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) .14).17) . 2 (7. 2. (7. a† (k′ ) = δij δ3 (k − k′ ). can then be interpreted as the annihilation and the creation operators for the two kinds of quanta associated with the ﬁelds φi (x).7.

b† (k′ ) = 0. b(k). b(k′ ) = a(k).15) and the deﬁnition in (7.19) with respect to the a(k) and b(k) operators. a(k). a(k′ ) = b(k). a† (k′ ) = b(k). in terms of the creation and the annihilation operators.16)) and take the form a(k).20) (We can equivalently normal order the Hamiltonian (7.8) or (7. a† (k′ ) = a† (k). a† (k) and † (k) can now be obtained from (7. we can write : H := d3 k Ek a† (k)a(k) + b† (k)b(k) .18) The Hamiltonian for the system (see (7. (7.19) with respect to the ai (k) operators and in that case the Hamiltonian would .12)) can be written. The commutation relations between the a(k). a† (k).17) b (or directly from (7. b(k′ ) = 0.11) and (7. (7.19) If we normal order the Hamiltonian in (7.262 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. 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) . (7. b† (k′ ) = δ3 (k − k′ ). b† (k′ ) = b(k). a† (k′ ) = 0. a(k). b† (k′ ) = b† (k).

3 Charge operator 263 be a sum of two Klein-Gordon Hamiltonians that we have already studied in chapter 5.) The vacuum state in such a case can be deﬁned to be the lowest energy state annihilated by both a(k) and b(k). (7. as we will see shortly. Namely. degeneracies arise when there is a symmetry present in the system under study. φ† (x) → φ′ † (x) = φ† (x)eiθ . as was also the case for a real Klein-Gordon theory. H|0 = 0.22) ˜ |k Similarly. as we know from our studies in non-relativistic quantum mechanics. The degeneracy can be understood as follows. Namely. However. The theory is. After all. (7. energy) given by |k = a† (k)|0 . 7.3 Charge operator To understand the physical meaning of the two distinct and degenerate one particle states. of course. a(k)|0 = 0 = b(k)|0 = 0|a† (k) = 0|b† (k). therefore.23) where θ is a real constant parameter of transformation (a global transformation).7. (7. 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. in the present case. the theory is invariant. = b† (k)|0 . manifestly Lorentz invariant and translation invariant. there is in addition an internal symmetry associated with the system. there will now be two distinct one particle states degenerate in mass (and.21) However. under the transformation φ(x) → φ′ (x) = e−iθ φ(x). Such a transformation is not a space-time symmetry transformation since .

φ′ † (x). the transformation in (7. We note that under the global phase transformation (7. in this case. ∂µ φ′ (x).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.23) takes the form (α = ǫ = inﬁnitesimal) δφ† (x) = φ′ † (x) − φ† (x) = iǫφ† (x). Inﬁnitesimally. Rather.13)) . for inﬁnitesimal transformations in (7. ∂µ φ′ † (x) −L φ(x).24)) showing that it is a symmetry of the system and in this case we have K µ = 0. δφ(x) = φ′ (x) − φ(x) = −iǫφ(x). (7. Thus. the change in the Lagrangian density (see (6. ∂µ φ(x). the Lagrangian density (7.24).23) (or the inﬁnitesimal form (7.264 7 Complex scalar field theory the space-time coordinates are not changed by this transformation.26) On the other hand.23) (or the inﬁnitesimal form of it in (7. we have (see (6. (7.8)) is given by L φ′ (x).24)).5) is invariant under the global phase transformation 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. (7. φ† (x).24) where ǫ denotes the inﬁnitesimal constant parameter of transformation.

29) This is a vector current since the parameter of transformation is a scalar. therefore. Even in a quantum theory.7. as we have seen.) Therefore. (7. we have freely rearranged the order of factors in (7.27). this is the probability current density (up to a normalization) which we had studied in the ﬁrst quantized theory. (7.13)) ← → ∂L ∂L + δφ† − K µ = iǫφ† (x) ∂ µ φ(x).28) and the parameter independent current follows to be ← → J µ = iφ† (x) ∂ µ φ(x). we can show that the associated conserved charge of the theory can be written in terms of the creation and annihilation operators as . the current can be easily checked to be conserved using the equations of motion in (7.27) (Note here that classically the order of the factors is not important. we can deﬁne the N¨ther current associated with this o symmetry to be (see (6. under normal ordering the order of factors does not matter (for a bosonic theory).30) Using the ﬁeld decomposition (7.1) ∂µ J µ (x) = 0.16). Of course. But here it is an operator. (7.3 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). we normal order all observables such as currents and charges and.) Furthermore. † ∂∂µ φ ∂∂µ φ µ Jǫ = δφ (7. (In fact.

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) . a U (1) phase transformation) we can identify this operator as the electric charge operator. Q|0 = d3 k a† (k)a(k) − b† (k)b(k) |0 = 0. (7.32) Since this charge operator is associated with a global phase transformation of the kind associated with electromagnetic interactions (namely.34) d3 k′ a† (k′ )δ3 (k′ − k)|0 = a† (k)|0 = |k . a† (k) + a† (k)a(k′ ) |0 (7. we are suppressing the normal ordering symbol for simplicity with the understanding that the charge operator has the normal ordered form in (7.32). Acting on the ﬁrst of the two one particle states in (7.32) annihilates the vacuum.33) so that the vacuum does not carry any charge.22). (7. (7. We note that the normal ordered charge operator (7. . the charge operator takes the form : Q := d3 k a† (k)a(k) − b† (k)b(k) . Here. 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′ ).31) If we normal order.

In contrast.22). the second one particle state of the theory |k is also an eigenstate of the charge operator but with the eigenvalue (−1). from the form of the Hamiltonian (7. in particular. contains no particle and this should be contrasted with the deﬁnition of the Dirac vacuum discussed in section 2.20) d3 k Ek a† (k)a(k) + b† (k)b(k) .5. This transformation is otherwise known as the charge conjugation symmetry which we will .37) That is.7. The meaning of the two one particle states is now clear.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.3 Charge operator 267 In other words. (7. H= (7. acting on the second one particle state in (7. of course. They describe two distinct particles with degenerate mass but with exactly opposite electric charge. 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′ ). These are. b† (k) + b† (k)b(k′ ) |0 = = − = − = − d3 k′ b† (k′ )δ3 (k′ − k)|0 = −b† (k)|0 (7. the second quantized description is symmetric under the interchange of particles and anti-particles. the particle and the anti-particle states – but both of them now correspond to positive energy states. Furthermore. the one particle state |k is an eigenstate of the charge operator with eigenvalue (+1). Therefore. ˜ Namely. The vacuum in this bosonic theory.35) ˜ = −|k .

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. to conclude 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. Therefore. in this section. 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. We note from (6. Thus. let us discuss the Feynman Green’s function in the basis of the complex φ(x) and φ† (x) ﬁelds. 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. we will construct only the Feynman Green’s function for the complex KleinGordon ﬁeld theory.146). 7.4 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. 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 ′ ′ ′ ′ .16) as well as the commutation relations in (7. Using the ﬁeld decomposition in (7.268 7 Complex scalar field theory study in a later chapter.

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 (5.7. 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. (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 = 0.143) and (5. Here we have used (7.4 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).38) = iG(−) (x − y).38).145).39) .

42) .40) Here we have used the deﬁnition of the Feynman Green’s function in (5.5 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).41) describes the most general (renormalizable) self-interacting theory for the complex Klein-Gordon ﬁeld. 2 (7.146). we see that in this case. Thus. 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). However. (7. Denoting φ1 = σ and φ2 = χ in the deﬁnition in (7. in this case we should look for interactions that would preserve the internal global symmetry in (7. It is easy to check that the Lagrangian density with the quartic interaction λ † 2 (φ φ) . (7. deﬁning 1 φ(x) = √ (σ(x) + iχ(x)) . 4 L = ∂µ φ† ∂ µ φ − m2 φ† φ − λ > 0.270 On the other hand. namely. 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). 7.23).

7.44) The Hamiltonian for the interacting theory (7.24). 16 L = (7.43) Namely. for constant ﬁeld conﬁgurations. (7. δχ(x) = −ǫσ(x).43) is invariant under the global phase transformation (7.47) . In the real basis of (7.23) or the inﬁnitesimal form in (7. the minimum of the potential is obtained by requiring ∂V ∂V =0= .5 Spontaneous symmetry breaking and the Goldstone theorem 271 we note that the Lagrangian density (7.42).45) and we note that. this describes the theory of two real Klein-Gordon ﬁelds with quartic self-interactions and which are also interacting with each other. ∂φ ∂φ† or. ∂σ ∂χ (7.46) In general. the other (kinetic energy) terms would only add positively to the energy. The Lagrangian density (7.41) has the form λ † 2 (φ φ) . the minimum of energy coincides with the minimum of the potential (We note that if the ﬁeld φ is not constant.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 . 4 H= d3 x Π† Π + ∇φ† · ∇φ + m2 φ† φ + (7. the inﬁnitesimal transformations (7. ∂V ∂V =0= . (7.) λ † 2 (φ φ) 4 2 V = m2 φ† φ + = λ m2 2 σ + χ2 + σ 2 + χ2 2 16 .41) or (7.24) can be shown to have the form δσ(x) = ǫχ(x).

50) In terms of the real ﬁelds σ and χ deﬁned in (7.51) Both the Lagrangian densities in (7. it is clear that the minimum occurs at (the subscript c is supposed to denote a classical ﬁeld. c χc = 0|χ(x)|0 = 0. In quantum theory.42). We note from our discussion of the real Klein-Gordon theory that the plane wave expansion for the ﬁeld operator in (5. a concept that we do not get into here) φc = φ† = 0. 16 L = (7.48) As we know from studies in classical mechanics. or. c or. (7.56) has exactly this property (see (5. equivalently σc = χc = 0. since the potential in (7. perturbation around the minimum of the potential is stable.46) is a sum of positive terms. σc = 0|σ(x)|0 = 0.49) φc = 0|φ(x)|0 = 0. 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 .50) and (7. Changing the sign of the mass term does not change the . 4 λ > 0.106)). (7.24) and (7. and perturbation theory developed around such a vacuum would be stable.23) (or (7.41) except that we change the sign of the mass term L = ∂µ φ† ∂ µ φ + m2 φ† φ − λ † 2 (φ φ) .51) are clearly invariant under the global phase transformation in (7. (7.44)).272 7 Complex scalar field theory In the present case. Let us next consider the same Lagrangian density for the complex scalar ﬁeld as in (7. this translates to the fact that we can deﬁne the ground state of the theory such that φ† = 0|φ† (x)|0 = 0.

(7. k0 = k2 − m2 .48) .7. (7. in general. To begin with. in this case. let us note that in the absence of interaction (λ = 0).40) or (5. 2 − m2 ∂|k| k (7.4)) − m2 φ(x) = 0. Such particles are known as tachyons and a theory with tachyons. the theory (7.5 Spontaneous symmetry breaking and the Goldstone theorem 273 symmetry behaviour.41) would lead to the dynamical equation (compare with (1.55) The minimum of this potential is obtained from (7. therefore.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 =√ > 1.52) k2 = −m2 .54) Namely. has many undesirable features. c = 1) leading to an acausal propagation (and. but it does change the Hilbert space structure of the theory in a very interesting way. Let us note that.51) has the form V = −m2 φ† φ + = − λ † 2 (φ φ) 4 2 m2 2 λ σ + χ2 + σ 2 + χ2 2 16 . the particle will travel faster than the speed of light (recall that in our units. corresponding to the Einstein relation (7. for constant ﬁeld conﬁgurations the potential of the theory (7. or. a violation of causality).50) or (7.

(7. (7.56) = χ −m2 + There now appear to be two sets of solutions to the equations in (7. λ χc = 0. 2 σc + χ2 = c 2m σc = ± √ . σc =0=χc = −m2 + = −m2 + = 0. (7. let us analyze the second derivatives of the potential at the two extrema in (7. we can choose σc = 0 = χc . = 0. = σ −m2 + ∂V ∂χ = −m2 χ + λ χ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.58) To determine the true minimum. 2m σc =± √ . for example.χc =0 λ (7.60) 2m σc =± √ . namely. 3λ 4 λ 4 4m2 λ 4m2 λ = 2m2 . (7.56).57) which lead to ∂2V ∂σ 2 ∂2V ∂σ∂χ while ∂2V ∂σ 2 ∂2V ∂χ2 ∂2V ∂σ∂χ 2m σc =± √ . 4m2 .χc =0 λ = σc =0=χc ∂2V ∂χ2 σc =0=χc = −m2 .57) λ For the second of these solutions.χc =0 λ .59) = 0.274 7 Complex scalar field theory ∂V ∂σ = −m2 σ + λ σ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.

λ (7. we note explicitly from the form of the potential in (7. λ λ λ (7.62) so that the second solution.1. However.7. On the other hand.1: Potential leading to spontaneous breaking of symmetry. First. leads to a lower energy. the second solution in (7. V 2 ((σc + χ2 ) c m2 4m2 ) = − = λ 2 = − 4m2 λ λ + 16 4m2 λ 2 2m4 m4 m4 + =− .61).55) that V (σc = χc = 0) = 0. Each point on this circle is related by the symmetry transformation (7.61) In fact. we note that there is an inﬁnity of minima (in the σ-χ ﬁeld space) of the potential given by the circle (7. 7. V (σ. indeed. . There are several interesting things to note from this analysis. All of this is more easily seen from the graph of the potential in Fig.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).5 Spontaneous symmetry breaking and the Goldstone theorem 275 In other words we note that out of the two solutions in (7. the ﬁrst corresponds to a local maximum. χ) V =0 χ 2m −√ λ 2m √ λ σ Figure 7.57).

the inﬁnitesimal transformations in (7. if we choose the minimum in (7. We recall that any choice for the minimum of the potential would translate into the properties of the vacuum state of the quantum theory. χ(x)] = −ǫσ(x). the action or the Hamiltonian of the theory is invariant under the symmetry transformation. λ (7.44) can be written in terms of the commutators with the charge operator as δσ(x) = −iǫ [Q. From (7. δχ(x) = −iǫ [Q.63) then.44) spontaneously.64). it now follows that = −iǫ 0| [Q.49)) 2m 0|σ(x)|0 = √ . for example. λ 0|χ(x)|0 = 0.61) to correspond to the point (in the σ-χ ﬁeld space) 2m σc = √ .65) 0|δχ(x)|0 2m = −ǫ 0|σ(x)|0 = − √ ǫ.64) We recall that conserved charges are the generators of inﬁnitesimal symmetry transformations of the theory and generate the transformations through commutators. such a breaking manifests in the following manner. λ χc = 0. χ(x)] |0 (7. In this case. In the quantum theory. σ(x)] = ǫχ(x). Thus. In this case.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 the quantum theory this would imply (see (7. (7. but the ground state is not. (7. This has to be contrasted with the case where an interaction term in the Hamiltonian can break the symmetry of a theory explicitly.66) which implies that .

H]|0 = 0.69) where we have denoted the energy of the vacuum |0 as E0 and this derivation shows that the state |χ deﬁned in (7. H|χ = E0 |χ . (7. [Q. t)Q|0 d3 x 0|eiP ·x J 0 (0)e−iP ·x Q|0 . (7.5 Spontaneous symmetry breaking and the Goldstone theorem 277 Q|0 = 0. (7. although the Hamiltonian of the theory is.71) . This can be easily seen from (7. namely.69) as follows. H] = 0. let us denote Q|0 = |χ It follows now from (7. HQ|0 = E0 Q|0 .31) and (7. (note that Q is Hermitian) χ|χ = = = 0|QQ|0 0| d3 x J 0 (x. or.67) This explicitly shows that the vacuum state of the theory is not invariant under the symmetry transformation in this case. Therefore. or. it is suggestive that we can think of this state as another vacuum state.70) (7. (QH − HQ)|0 = 0. (7. The problem with this interpretation is that the state is not normalizable.69) would appear to be degenerate with the vacuum state |0 in energy.69) that [Q.68) and (7. or.7.68) Since Q does not annihilate the vacuum of the present theory.

.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. therefore. This analysis also shows that the ﬁnite transformation operator U (θ) = e−iθQ .74) In other words. (7. the state |χ is not normalizeable and hence cannot be thought of as another vacuum. (7. 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. Q] = 0.73) where we have used the property of the ground state Pµ |0 = 0. (7. and as a consequence.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 −→ ∞. we have [Pµ .75) does not act unitarily on the Hilbert space.75) takes a state out of the Hilbert space. In fact. 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. Since Q does not depend on spatial coordinates. (7.72) leads to (7.66) that χ(x) corresponds to the Nambu-Goldstone boson of the theory. it follows that the momentum operator also commutes with Q and.

We can see this quantitatively in the quantum theory as follows.64) 2m σ = 0|σ(x)|0 = √ . (7. is given by σ(x) = σ ′ (x) + σ with the later identiﬁcation σ ′ (x) = σ(x) as the dynamical variable.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) + √ .54))).51) becomes . With the choice of the solutions for the minima as in (7. the Lagrangian density in (7.77).53) and (7. 7.7.60) shows that in this case of spontaneous breakdown of the symmetry.5 Spontaneous symmetry breaking and the Goldstone theorem 279 The second interesting observation that follows from the analysis in (7.77). We note from Fig. λ χ = 0|χ(x)|0 = 0. √ 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. (7. λ χ(x) → χ(x) + χ = χ(x). (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.) Under the shift (7.78) (Explicitly.77) so that the shifted ﬁelds have vanishing vacuum expectation values (vevs) 0|σ(x)|0 = 0 = 0|χ(x)|0 .60) is as follows.1 that the massless mode would correspond to oscillations along the valley of minima while the massive mode would represent the orthogonal oscillations. say for the ﬁrst term. the shift in (7.

We are.79) This shows explicitly that perturbed around the true minimum. 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. a continuous symmetry is spontaneously broken. of course. 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.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 . For example. then there must appear massless excitations. 2 16 (7. 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. aware of this phenomenon (known as the spontaneous symmetry breaking phenomenon) from studies in other areas of physics.66)). if we look at the Hamiltonian describing a .

∂µ → ∂µ + ieAµ . (7. when the spins are all pointing up or down.) However.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. Consequently the rotational symmetry is spontaneously broken. it is clear that once we have chosen a particular ground state (vacuum) where all the spins are pointing in a given direction. Unfortunately.53). there are two such vacuum conﬁgurations allowed in one dimension. (7.6 Electromagnetic coupling 281 magnet (for example. H = −g si · si+1 . The analogue of the massless Nambu-Goldstone bosons in this case correspond to the long range correlations associated with the spin waves. (For example. in the case of the Ising model).81) where “e” denotes the charge carried by the particle. Given the complex Klein-Gordon ﬁeld theory.) 7. 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 . we can ask how such a system can be coupled to a background electromagnetic ﬁeld. namely. The prescription is again through the minimal coupling discussed in (1.80) i where the negative sign in the Hamiltonian (g > 0) is for ferromagnets. we do not observe any massless spin 0 bosons in nature.7. 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. Thus. the Hamiltonian is invariant under rotations. (The π meson which is the lightest among the spin zero bosons has a mass mπ ∼ 140 MeV. The ground state of the theory is clearly the state (spin conﬁguration) where all the spins are aligned. we have eﬀectively chosen a deﬁnite orientation in the physical space.

85) . (7. As we can verify readily.82) Dµ φ = ∂µ φ + ieAµ φ.23)). In fact.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.84) where the parameter of transformation θ(x) is now a local function of space-time coordinates (unlike in (7.82) The modiﬁed derivative Dµ introduced in (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. e (7. (7. note that under this transformation. φ† (x) → φ′ † (x) = φ† (x)eiθ(x) . ′ 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).82) is invariant under a local (U (1)) phase transformation of the form φ(x) → φ′ (x) = e−iθ(x) φ(x). the Lagrangian density (7. → ((∂µ + ieAµ ) φ)† (∂ µ + ieAµ ) φ − m2 φ† φ = = (7. Aµ (x) → A′ (x) = Aµ (x) + µ 1 ∂µ θ(x).

84) L(φ(x). the local phase invariance (7. 345 (1961). unlike the global phase invariance in (7. Goldstone.7. Weinberg.) It follows from this observation that under the local transformation (7. which is obtained by minimally coupling the complex scalar ﬁeld to the electromagnetic ﬁeld.84) requires an additional ﬁeld (Aµ ). Aµ (x)) → L(φ′ (x). Schweber. 4. Y. note that ∂µ φ(x) does not transform covariantly. Salam and S. the covariant derivative acting on a ﬁeld transforms exactly like the ﬁeld itself (covariantly) under such a transformation (see (7. φ† (x). 965 (1962). Of course. Nuovo Cimento 19. the theory described by (7.7 References 283 In other words. Jona-Lasinio.84). 154 (1961). A. 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. namely.86) Consequently.84)). J. Goldstone. S. 7. invariance under a local transformation necessarily requires additional ﬁelds known as gauge ﬁelds. Aµ .7 References 1. φ′ † (x). is invariant under the local phase transformation (7. φ† . (In contrast. Physical Review 122. Row. (7. Peterson. This is a very general feature in a quantum ﬁeld theory. Physical Review 127.82). Nambu and G. Local phase transformations are commonly known as gauge transformations and we note that. J. 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. Introduction to Relativistic Quantum Field Theory. Evanston (1961). 3. 2.23). .

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

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

2) automatically lead to NF (NF − 1) = 0. (8. Namely.6) where ψα (x). F F F In other words. we obtain 2 NF = a† aF a† aF F F 8 Dirac field theory = a† F aF . if this system is analyzed in more detail. In fact. (8. It is clear.5) This is. what the Pauli exclusion principle would say. there can be at the most one quantum in a given state. the Dirac equation for a massive spin particle. is a four component complex spinor function. in such a theory. the anti-commutation relations in (8. 3. 2.2 Quantization of the Dirac ﬁeld As we have seen in (1. we will be naturally dealing with anti-commuting variables (these variables are commonly known as Grassmann variables) and anti-commutation relations. 1. therefore. of course. 8.3) = a† 1 − a† aF aF = a† aF = NF .96). This equation is manifestly Lorentz covariant and the adjoint equation has the form . α = 1. 4.4) which shows that the eigenvalues of the number operator. a† F + − a† aF aF F (8. that in dealing with fermionic ﬁeld operators. has the form (iγ µ ∂µ − m) ψ = 0. the wave function of the system will have the necessary antisymmetry property characteristic of fermions. 1 2 (8. can only be nF = 0.286 then. it leads to the fact that with the anti-commutation relations.

consequently.8) In trying to second quantize the Dirac equation (theory). as we have seen in chapter 3 as well as in chapter 4. in this case.10) . let us note that since ψ and ψ are anti-commuting (Grassmann) variables. (8.61)) describing fermionic particles and.6) and (8. however. In dealing with the Dirac ﬁeld theory. In deriving the Euler-Lagrange equations. the Lagrangian density of the form / L = ψ (iγ µ ∂µ − m) ψ = iψ∂ ψ − mψψ. we would like to decide on a Lorentz invariant Lagrangian density which would give rise to the two Dirac equations (8.8. we will always use the convention of taking fermionic (Grassmann) derivatives from the left. the Euler-Lagrange equations following from the Lagrangian density (8. we have to be careful about signs that may arise in taking derivative operators (with respect to these variables) past other Grassmann variables. With this. Clearly.41)) ψ(x) = ψ † (x)γ 0 .9) is manifestly Lorentz invariant (recall the transformation of the Dirac bilinears under a Lorentz transformation discussed in section 3.7) (8. 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. where the adjoint spinor is deﬁned to be (see also (2. However. need to be treated as anti-commuting operators.9) give ∂L = (iγ µ ∂µ − m) ψ = 0. the 1 Dirac ﬁeld operators are expected to belong to the (reducible) spin 2 representation of the Lorentz group (see (4.3). In general.7) as the Euler-Lagrange equations.2 Quantization of the Dirac field 287 ← − ψ iγ µ ∂µ + m = 0. ∂ψ ← − ∂L ∂L − ∂µ = mψ − ∂µ (−iψγ µ ) = ψ iγ µ ∂µ + m = 0. ∂ψ ∂∂µ ψ (8. (8.

It is worth noting here that even though the Lagragian density (8.9) only by a total divergence. (8. we note that the Lagrangian density in (8.12) diﬀers from L in (8. 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 (8.9) is Lorentz invariant and gives the two Dirac equations as the Euler-Lagrange equations.7) that we are interested in. 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.13) = L − ∂µ ψγ µ ψ . 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.6) and (8.9) is not Hermitian. (8.11) † where we have used (2.288 8 Dirac field theory which are exactly the two Dirac equations (8. 2 .12) However. In fact.83). it is not Hermitian. Even though the Lagrangian density (8. Namely. 2 2 L′ = (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. An alternate Lagrangian density which is Lorentz invariant.

˙ ∂ψ (8. = 0.9) to describe the dynamics of the system. (Πψ )β (y)]+ . (8. † † ψα (x). 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 † . 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. but we should understand that x0 = y 0 in all the relations in (8. If we work with the Lagrangian density (8.15) can also be written as (Πψ )α (x). (Πψ )β (y) or.14). ˙ ∂ψ ∂L = 0.15) is a reﬂection of our choice of left derivatives.15) The negative sign in the last relation in (8.12) which naturally leads to a symmetric stress tensor (see (6.8. ψβ (y)]+ = 0 = [(Πψ )α (x). −iψβ (y) + ψα (x). (8. + † = ψα (x). in the presence of gravitation. the second and the third relations in (8. . (Πψ )β (y) or. it is more appropriate to use the Lagrangian density in (8. we use L in (8. [ψα (x).16) = δαβ δ3 (x − y).14) Πψ = Πψ = Consequently.9) to describe the Dirac theory (equations). ψβ (y) + = −iδαβ δ3 (x − y). (Πψ )β (y)]+ = −iδαβ δ3 (x − y). ψβ (y) + + = 0.27)) for the theory. Using the ﬁrst relation in (8. † ψα (x). However.2 Quantization of the Dirac field 289 For this reason as well as for simplicity.16)) [ψα (x).15) and (8.

(8. H = d3 y † ψα (x).) ˙ H = −Πψ ψ − L ˙ ˙ = − −iψ † ψ − iψ † ψ − iψγ · ∇ψ + mψψ = −iψγ · ∇ψ + mψψ.19) . We will discuss the method of Dirac in more detail in chapter 10. ˙ iψα (x) = ψα (x).16).) The Hamiltonian density for the Dirac ﬁeld theory (8. (8.18) (8. it is now easy to check that this Hamiltonian leads to the two Dirac equations as the Hamiltonian equations.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. Such systems can be systematically studied through the method of Dirac brackets and such an analysis leads to the conclusion that the relations (8.15) and (8.17) Using the ﬁeld anti-commutation relations (8.15) and (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) .16) are all that we need for quantizing the Dirac theory.14) can be thought of as constraints on the dynamics of the system. leading to the Hamiltonian H= d3 x H = d3 x −iψγ · ∇ψ + mψψ .9) has the form (The negative sign in the ﬁrst term arises from our convention of choosing left derivatives for fermions. ψβ (y) + −iγ 0 γ · ∇y ψ(y) + β † + m ψα (x). For example.

ψβ (y) + † −mψ β (y) ψα (x). ← − ψ iγ µ ∂µ + m = 0. or.22) Namely. we obtain ˙† iψα (x) = = † ψα (x).18) indeed reproduce the two Dirac equations (8. or.6) and (8.8.21) and this can be written in the matrix form as ← − ˙ iψ † = −iψγ · ∇ − mψ. H d3 y i ψ(y)γ · ∇y β † ψα (x). (iγ µ ∂µ − m) ψ(x) = 0. the anti-commutation relations (8.20) Similarly. ˙ iγ 0 ψ(x) = −iγ · ∇ψ(x) + mψ(x). ← − ˙ iψγ 0 = −iψγ · ∇ − mψ. or. (8. or.7) as the Hamiltonian (Heisenberg) equations.16) as well as the Hamiltonian (8. (8.15) and (8. (8. . 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).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).

5).292 8. Taking advantage of our experience with the complex Klein-Gordon ﬁeld theory. s)uα (k′ . s) . s)uα (k. s)vα (k′ .23) has been chosen keeping in mind the massive normalization of the plane wave solutions in (2. s)ψα (x) α (2π)3 k0 d3 x d3 k ′ m ′ √ ei(k−k )·x u† (k. s) and vα (k. s′ ) α ′ . uα (k.93) and (3.3 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) can be inverted to give √ (k0 = Ek = k2 + m2 > 0) d3 x = s′ m eik·x u† (k. s′ ) α 3 (2π) k0 k′ 0 +ei(k+k )·x u† (k. let us note that the ﬁeld decomposition in (8. we write √ (k0 = Ek = k2 + m2 > 0) ψα (x) = 1 s=± 2 d3 k m (2π)3 k0 (8. we can expand the Dirac ﬁeld operator in the basis of these plane wave solutions. s) + eik·x d† (k. s)vα (k. where c(k. s) and d† (k.24) The speciﬁc form of the prefactor in (8.46). As in the case of the Klein-Gordon ﬁeld (see section 5.94) and the ﬁeld (8. Therefore. s) denote the positive and the negative energy spinors deﬁned in (3. s′ ) c(k′ . / (8. s′ ) d† (k′ . s) are anti-commuting operators.23) automatically satisﬁes the Dirac equation in this basis (i∂ − m) ψ(x) = 0.23) × e−ik·x c(k.4).

c† (k.8. s ) + . using the inversion relations (8. s ) α 0 x0 = s d3 k √ m k0 k 0 + δ3 (k + k ) e2ik = s u† (k. it can be shown that m † e−ik·x vα (k. s). s) = d(k. s ) = δss δ3 (k − k ) = d(k. d† (k . s)ψα (x) = d† (k.27). s). ψβ (y) † † ψα (x).28) Therefore. d† (k. s). (2π)3 k0 (8.46) (see also (3. s ) d† (k . s)vα (k . = δαβ δ3 (x − y). s ) c(k . s).26) It follows from (8.29) + .26) that m † e−ik·x ψα (x)uα (k. s)uα (k .x0 =y 0 +. s) and d(k.27) c† (k. = 0.25) m k0 δss c(k.25)-(8.25) and (8. s ) = c(k. s). (2π)3 k0 m † eik·x ψα (x)vα (k. s). k0 m where we have used the massive normalization for the spinors discussed in (2.3 Field decomposition 293 δ3 (k − k ) u† (k. In a similar manner. we know the equal time quantization conditions for the ﬁeld operators from (8. (8. s). s ) α (8.x0 =y 0 +. s) = d3 x d3 x Furthermore.92)). s) in a straightforward manner from the quantization conditions (8. (8. s).x0 =y 0 = 0. ψβ (y) +. we can now obtain the anti-commutation relations satisﬁed by the operators c(k. (2π)3 k0 d3 x (8. ψβ (y) † ψα (x). c† (k .28) and they correspond to c(k.15) and (8.16) to be ψα (x).

s′ =± 2 ′ d3 x ψ (−iγ · ∇ + m) ψ m k0 ′ m 3 d k k′ 0 × u(k′ . s) . / (8.294 8 Dirac field theory with all other anti-commutators vanishing.31) . The Hamiltonian (8. s) + eik·x d† (k. s) = 0. s) + eik·x d† (k. s) = 0. s′ )c† (k′ . s). s)(−γ · k + m)v(k. s) = (γ 0 k0 − γ · k + m)v(k. s′ c† (k′ s′ )eik ·x + v k′ . s′ )e−ik ·x × (−iγ · ∇ + m) × e−ik·x c(k. This shows that we can think of c(k. s′ )eik ·x + v(k′ . s) = (γ 0 k0 − γ · k − m)u(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′ d(k′ .18) of the theory.30) and write (k − m) u(k. 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. s)(γ · k + m)u(k. s) as well as d(k.93) and (3.94) (8. in this case.30) (k + m) v(k. c† (k. s). s′ )e−ik ·x × e−ik·x c(k. / we can simplify the Hamiltonian in (8. s) = 1 s. If we now use the relations (3. s′ )d(k′ . s)u(k. d† (k. s)v(k.s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m k0 ′ × u 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)c† (k. s′ γ 0 u(k. −k. s) m − v † k. s)d(−k. s′ v(k.46) (see also (3. s) v † k0 . 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′ )d† (k.s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m 0 k k0 ′ × u k′ .32) + e−2ik = 0 x0 m d3 k 1 s. s) − eik·x d† (k.92)). s′ )e−ik ·x ×γ 0 e−ik·x c(k. s′ )eik ·x + v k′ . s) − v k′ .s′ =± 2 = 1 s=± 2 d3 k Ek c† (k. s′ d(k′ . s)d(k′ . We are yet to normal order the Hamiltonian.8.s′ =± 2 ′ 0 +k 0 ′ 0 +k 0 )x0 u k′ . s) − d(k. s)c† (k′ . s)c(k. s′ )d† (k. s′ )d† (k. s)u(k. s) Ek m δss′ c† (k. s′ γ 0 v(k. −k. s′ γ 0 v(k. s) −e2ik 0 x0 u† k0 . s)d(k. s) . s)v(k. s′ )d† (k. s′ v(k. s)c† (k′ . s)c† (−k. s) d3 k u† k. s′ u(k. s) (8. s′ γ 0 u(k. s) − d(k.3 Field decomposition 295 H= 1 s. s)d† (k. s) −δ3 (k + k′ ) ei(k − e−i(k = 1 s. s′ )c(k. s) = 1 s. s′ c† (k′ . s′ )c(k. s′ u(k. s′ )c(k. s′ )c(k. s′ )d† (k. s) )x0 v k′ .

35) c(k. s)d(k′ .34) even though we do not put the normal ordering symbol explicitly. s′ ). s). we write products with the creation operators standing to the left of the annihilation operators. : d(k′ . s). = 0. s)d(k′ . Using (8.36) |k. s)d(k′ . s)d(k.21)). (8. namely. the normal ordered Hamiltonian for the free Dirac ﬁeld theory in (8.22)) = c† (k. (8. s)|0 d(k. s .34) As in the case of the complex (Klein-Gordon) scalar ﬁeld theory (see (7. we pick up a negative sign since they anti-commute (as opposed to bosonic variables). = d† (k.) = 0 = 0|c† (k. (8. s) and d(k.33) so that the order of factors inside a product that is normal ordered becomes important. s)|0 H|0 The one particle states of the theory can now be deﬁned as (see (7. s)|0 . s) : = −d† (k. s) . (8.296 8 Dirac field theory way as for the bosonic operators. s′ ). s′ ) : = d† (k. every time we move a fermionic operator past another in order to bring the product into its normal ordered form. s) (The Hamiltonian is understood to be normal ordered as in (8. s |k. = 0 = 0|d† (k. s) + d† (k. However. we now have : d† (k. we can deﬁne the vacuum state of the theory to correspond to the state which is annihilated by both c(k.33). s′ )d† (k. s)c(k.32) is obtained to have the form : H := s=± 1 2 d3 k Ek c† (k. s)|0 . For example.

′ (8.9) L = ψ (iγ µ ∂µ − m) ψ. ψ ) = ψ (iγ µ ∂µ − m) ψ ′ ′ ′ ′ (8. ψ(x) → ψ (x) = ψ(x)eiθ .9) (or (8. transforms as L(ψ.4 Charge operator 297 Both these states can be checked to be eigenstates of the Hamil√ tonian (8.39) = ψeiθ (iγ µ ∂µ − m) e−iθ ψ = ψ (iγ µ ∂µ − m) ψ = L(ψ.8. there are two distinct one particle states with degenerate mass (energy) and to understand the meaning of these states. as in the case of the complex scalar ﬁeld theory.34) with energy eigenvalue k0 = Ek = k2 + m2 > 0. the Lagrangian density (8. like the Lagrangian density for the complex Klein-Gordon ﬁeld theory can be seen to be invariant under a global U (1) phase transformation.12)). ψ). We note that under the global phase transformation (α is the constant parameter of transformation) ψ(x) → ψ ′ (x) = e−iθ ψ(x). we need to analyze the symmetry properties of the theory. Therefore. ψ) → L(ψ ′ .38) (8.37) or the inﬁnitesimal form of it (ǫ is the constant inﬁnitesimal parameter of transformation) δǫ ψ = ψ ′ (x) − ψ(x) = −iǫψ(x). (8. δǫ ψ = ψ (x) − ψ(x) = iǫψ(x).40) . 8.4 Charge operator The Dirac Lagrangian density (8.

45) The conserved charge associated with the symmetry transformation (8. we note that K µ = 0. the N¨ther current associated with the internal symmetry o transformation (8.38). (8. (8.9) is invariant under the global phase transformation (8.42) is consistent with the convention of taking left derivatives for the fermion ﬁelds) ∂L ∂L + δǫ ψα ∂∂µ ψα ∂∂µ ψ α (8.7) that this current is conserved.43) where J µ (x) = ψ(x)γ µ ψ(x).44) represents the current independent of the parameter of transformation. we obtain (see (6.42) δǫ ψ α = (−iǫψ)α (−iψγ µ )α = ǫψ(x)γ µ ψ(x).27) and we note that the form of the expression in (8.37) (or (8. Thus. where we have rearranged the fermionic operators keeping in mind their anti-commuting nature.38)). ∂µ J µ (x) = 0.13) and (7.6) and (8. using the inﬁnitesimal transformation (8.26).13) to be µ Jǫ (x) = ǫψ(x)γ µ ψ(x) = ǫJ µ (x). (8.29)) and it can be checked using the equations of motion (8.38)) is obtained from (6.41) On the other hand. the Lagrangian density (8. It is a vector current as in the case of the complex Klein-Gordon ﬁeld theory (see (7. as in the case of the complex scalar ﬁeld theory in (7. (8. In this case.298 8 Dirac field theory Namely.37) (or (8.37) of the system is given by (note that in the ﬁrst quantized theory. this would correspond to the total probability) .

s′ ) − d† (k′ . s)|0 = |k. lepton number. s′ ) c† (k. s′ ).37) (or (8. But. Therefore. s . s′ ) |0 = 1 s′ =± 2 d3 k′ c† (k′ . s)d(k. since we are dealing with a fermion system.33) and (7.34)) Q|0 Q|k.8. etc) which are associated with U (1) symmetry transformations as well.38)).46) We note here that the charge (8. s′ ) c(k′ . s) + − c† (k. s) − d† (k. s = 0. (8. . s′ )d(k′ . s)c(k′ .4 Charge operator 299 Q= d3 x J 0 (x) = d3 x ψ † (x)ψ(x). we can now show that the charge operator (normal ordered) satisﬁes (see (7. s′ )δss′ δ3 (k′ − k)|0 = c† (k. as in the case of the complex Klein-Gordon ﬁeld we can associate the charge operator with that of the electric charge.23). s′ )c(k′ .47) As in the case of the complex Klein-Gordon ﬁeld. we can express (8. s) . s)c(k.46) is associated with an Abelian symmetry transformation (8. (8.46) in terms of creation and annihilation operators as : Q := s=± 1 2 d3 k c† (k. Using the ﬁeld decomposition in (8. s)|0 = 1 s′ =± 2 d3 k′ c† (k′ . c† (k. = 1 s′ =± 2 d3 k′ c† (k′ . it is also possible to associate the charge operator with a fermion number (such as baryon number.

s′ ) 1 s′ =± 2 d(k′ . s . s) and d† (k. Namely. (8. s′ ). s′ )d(k′ . d† (k. This shows that. we can deﬁne the Green’s function for the Dirac ﬁeld theory as satisfying the Dirac equation with a delta function potential (source) as . which are degenerate in mass and are created by the operators c† (k. the vacuum state of the theory is charge neutral while the two distinct one particle states. s)d(k′ . s′ )c(k′ . s)|0 = −|k. we can identify them with the particle and the anti-particle states of the theory with positive energy. s′ ) d† (k. s′ ) − d† (k′ . carry opposite charge.48) 1 s′ =± 2 = −d† (k. s)|0 = − d3 k′ d† (k′ . the second quantized description of the Dirac theory is symmetric under the interchange of particles and anti-particles. the vacuum does not contain any particle as we would intuitively expect (compare this with the ﬁrst quantized description of the vacuum in section 2.5) and that the (normal ordered) Hamiltonian in (8. as we have also seen in the case of the complex Klein-Gordon ﬁeld theory.49) namely. in the second quantized description of the Dirac theory. s′ ) |0 = − d3 k′ d† (k′ .300 Q|k. s′ )δss′ δ3 (k − k′ )|0 (8. s = 1 s′ =± 2 8 Dirac field theory d3 k′ c† (k′ . s) respectively. 8. s). s) + − d† (k.37)) c(k. Thus.34) is symmetric under the discrete transformation (see also (7.5 Green’s functions As we have discussed within the context of the Klein-Gordon theory. s) ↔ d(k.

we would concentrate mainly on the Feynman Green’s function since that is the most useful .117) we know that the Green’s function for the Klein-Gordon equation satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y).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. advanced. Although the relation (8.53) (i∂ − m) S(x − y) = (i∂ − m) (i∂ + m) G(x − y) / / / = − ∂µ ∂ µ + m2 G(x − y) = δ4 (x − y). Furthermore.). but it is the trivial 4 × 4 identity matrix (which we do not write explicitly). we can identify (8.8.51) = − ∂µ ∂ µ + m2 . (8. etc. The right-hand side is also a matrix.52) S(x − y) = (i∂ + m) G(x − y). From these then. from (5.54) Therefore. / (8. S(x − y) is a 4 × 4 matrix function in the spinor space.5 Green’s functions 301 (i∂ − m) S(x − y) = δ4 (x − y). we do not have to calculate the Green’s function for the Dirac ﬁeld theory separately. 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. Feynman. / so that (8.50) Here.

53) we can also obtain the positive and the negative energy Green’s functions for the Dirac ﬁeld theory as (see (5. (see (5. We note that the relation (8. / so that we can write.143) and (5. k2 − m2 + iǫ (8. for example.55) SF (k) = (k + m) GF (k) = lim / which can also be written as 1 . 3 (2π) k0 (8.56) SF (k) = lim ǫ→0+ (8.57) From (8. k − m + iǫ / ǫ→0+ (8.58) We can express the Feynman Green’s function in terms of the positive and the negative energy Green’s functions as well (see (5.302 8 Dirac field theory in the calculation of S-matrix elements.140)) k+m / .146)) .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 .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 .

8.59) where we have used the fact that the terms involving derivatives of the step function cancel out (namely. (2π)3 k0 m eik·x d† (k. other Green’s functions for the Dirac ﬁeld theory can also be easily obtained from the corresponding functions for the KleinGordon ﬁeld theory.61) (2π)3 k0 (−) ψα (x) = 1 s=± 2 d3 k Similarly. s)v α (k. s). s). (2π)3 k0 m eik·x c† (k.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). 8.6 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.8) also as ψ α (x) = ψ α + ψ α (x). s)uα (k. where ψ α (x) = 1 s=± 2 (+) (−) (8. s). (8. s).60) where (+) ψα (x) = 1 s=± 2 d3 k m e−ik·x c(k.63) (2π)3 k0 ψ α (x) = 1 s=± 2 (−) d3 k . (8. s)uα (k. we can decompose the adjoint ﬁeld operator (8. (8. Similarly. s)vα (k. (8. δ(x0 )(G(+) (x) + G(−) (x)) = 0).

Similarly. s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 × e−ik·x+ik ·y uα (k.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. we have d3 kd3 k′ (2π)3 (−) ψα (x).s′ =± 2 ′ d† (k. c† (k. s′ δss′ δ3 (k − k′ ) = 1 s=± 2 d3 k m −ik·(x−y) e uα (k. s).s′ =± 1 2 m2 k0 k′ 0 + × eik·x−ik ·y vα (k. ψ β (y) (−) + ′ = s. s)v β k′ . s′ = 1 s. s′ = s. ψ β (y) (+) + = s.s′ =± 1 2 ′ c(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) and d† (k. (8. s).s′ =± 1 2 m2 k0 k′ 0 + × e−ik·x+ik ·y uα (k. s) in (8.64) where we have used the completeness relation in (3. d(k′ .304 8 Dirac field theory It is clear from the basic anti-commutation relations for the creation and the annihilation operators c(k.58). s). s)uβ k′ . c† (k′ . s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 . d(k.106) as well as the identiﬁcation in (8. s)uβ k′ .

9. As a result. 8. s)v β k′ .136) (see also (5. since fermionic ﬁeld operators anti-commute (Grassmann variables). s)v β (k.5 and 6.109).63) that the positive and the negative energy parts of the ﬁeld operators contain annihilation and creation operators respectively) . However. we can write (+) = ψα (x). in rearranging terms to bring them to a particular form. ψ β (y) (+) (−) (8. ψ β (y) + + (−) + ψα (x).65) ψα (x). where we have used (3.6.7 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 6.7 Normal ordered and time ordered products 305 × eik·x−ik ·y vα (k. Thus. by deﬁnition the normal ordered product of two ﬁeld operators yields (note from (8.160)) 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. αβ (8. s′ δss′ δ3 (k − k′ ) = 1 s=± 2 ′ d3 k m ik·(x−y) e vα (k. we need to be careful about negative signs that can arise from changing the order of any two such operators.66) which is the analogue of the Schwinger function (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).8.61) and (8. ψ β (y) (−) (+) + = −iSαβ (x − y) − iSαβ (x − y) = −i S (+) (x − y) + S (−) (x − y) = −iSαβ (x − y).

it is easy to show that ψ β (y)ψα (x) = : ψ β (y)ψα (x) : −iSαβ (x − y).67). we can write (8.306 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) :. In other words. unlike the bosonic case.71) . 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). ψ β (y) (−) + (+) (−) (+) (−) (+) (−) (+) (−) (+) (−) = ψα (x)ψ β (y) + iSαβ (x − y). (+) (−) (+) (8. As a result.68) as ψα (x)ψ β (y) = : ψα (x)ψ β (y) : −iSαβ (x − y). From the deﬁnition in (8.69) (8. using (8.70) (8.65). Similarly. (+) (8. we see that we can denote 0|ψα (x)ψ β (y)|0 = −iSαβ (x − y) = ψα (x)ψ β (y). (+) (+) (+) (+) (+) (8. here the order of factors inside normal ordering can lead to additional signs.64).67) and we note that.68) where in the last step we have made the identiﬁcation in (8.

59) and which can be compared with (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. 0|T ψα (x)ψ β (y) |0 = iSF .7 Normal ordered and time ordered products 307 which can be compared with (6.8. It follows now from (8.αβ (x − y).73) (+) (−) (−) (+) where we have used the deﬁnition in (8.αβ (x − y) = ψα (x)ψ β (y). (8.81). 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).5 which we will not repeat here.74) Once again. (8. we can develop Wick’s theorem for time ordered products of Dirac ﬁelds in a straightforward manner as discussed in section 6.6. . (8.αβ (x − y) : ψα (x)ψ β (y) : + iSF .104).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 .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).

54) which is appropriate for massless spinors.77) to obtain (here k0 = |k|) . The discussion for the left-handed ﬁeld can be carried out in an analogous manner.164)). it is most natural to decompose it into its chirality states. in the case of the standard model in 14.75) As a result. we can also decompose a Dirac ﬁeld into a left-handed and a right-handed component as 1 (½ + γ5 )ψ(x). Therefore. for example.23).7. Using the deﬁnition of the chiral spinors in section 3. we have identiﬁed k0 = Ek = |k|. Let us consider the ﬁeld quantization of a right-handed Dirac ﬁeld (Weyl ﬁeld). we can expand the ﬁeld as (see (8.3). (8.170) we can invert the decomposition in (8. 2 1 (½ − γ5 ).23) because we are using here the normalization (2. in this section we will discuss brieﬂy the quantization of such ﬁelds. 2 ψR (x) = ψL (x) = (8. the free Dirac Lagrangian density for a massless spinor ﬁeld naturally decomposes as / / / L = iψ∂ ψ = iψ R ∂ ψR + iψ L ∂ ψL . the discussion following (3. Furthermore. Correspondingly. Using the orthonormality relations for the right-handed spinors described in (3. (8.53) and (2. Unlike in (8. we note here that there is no sum over spin polarization because the chiral spinors are charaterized by their unique chirality or helicity (see.23)) d3 k (2π)3 |k| ψR (x) = e−ik·x c(k)uR (k) + eik·x d† (k)vR (k) .7.77) is diﬀerent from those in (8. the multiplicative factors in (8. when the Dirac particle is massless. for example.77) where.308 8. as before.76) Such massless left-handed and right-handed Dirac ﬁelds (Weyl ﬁelds) arise naturally in physical theories (as we will see.8 Massless Dirac ﬁelds 8 Dirac field theory As we have seen in section 3.

3 |k| (2π) (8.76)).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). d† (k′ ) + + .81) and the higher particle states created by applying the creation operators and other quantities of interest can be derived in the standard manner. .8.78) 0 Similarly. (8. (8. (8. from the equal-time anti-commutation relations for the ﬁeld variables following from the Lagrangian density for the righthanded spinor ﬁeld (see.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).80) The Hilbert space of the theory can now be built with the vacuum deﬁned as satisfying c(k)|0 = 0 = d(k)|0 . (8. it can also be shown that (it is assumed that k0 = |k|) d† (k) = d3 x † e−ik·x vR (k)ψR (x). c† (k′ ) = δ3 (k − k′ ) = d(k). for example.

80) in evaluating the vacuum expectation values. (8. We recall that the Feynman propagator is deﬁned as the time ordered product (see.81) we obtain (we assume k0 = |k| in the ﬁeld decomposition) iSF .83) Here we have used (8.172) this can also be written as d3 k 1 iSF .77) and using the properties of the vacuum in (8. (8.82) Putting in the ﬁeld decomposition in (8. Dropping the spinor indices and using (3.310 8 Dirac field theory As an example. for example.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. (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 .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 ). let us determine the propagator for a righthanded fermion.84) .74)) iSF .

2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 (8.84) separately. Using the integral representation for the theta function (see (6.85) Here we have used the fact that the integral has contributions only when k0 = |k| and correspondingly have rewritten some of the terms.8 Massless Dirac fields 311 Let us evaluate the two terms in (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 .8.100) 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.R (k) = iSF .85) and (8. The reason for this is that the adjoint ﬁeld ψ R has negative chirality.88) We note here that we have deﬁned the propagator in (8. However. adding the two results in (8.88) by simply multiplying a factor of γ 0 on the right which leads to iS F . For .312 = − / dk0 k e−ik (x −y ) .R (x − y) = / d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) e .R (k) = (8. 2(k2 + iǫ) k + iǫ iSF . for example. the conventional covariant propagator (deﬁned as the time ordered product of ψR ψ R ) can be obtained from (8.74)).89) We note that although our discussion has been within the context of right-handed fermion ﬁelds.9 Yukawa interaction Let us next consider the theory which describes the interaction between a Dirac ﬁeld and a charge neutral Klein-Gordon ﬁeld.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. everything carries over in a straightforward manner to left-handed fermion ﬁelds. 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 . (8. 2(k2 + iǫ) k + iǫ (8.86) and substituting into (8.87) iSF .82) as † the time ordered product of ψR ψR and not as the conventional time ordered product of ψR ψ R (see.R (k)γ 0 = ik(½ − γ5 ) / ikPL / = 2 .86) Therefore. 2πi 4k0 k0 + |k| − iǫ 0 0 0 8 Dirac field theory (8.

) In this case.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. has the form 1 λ M2 2 / φ − g ψψφ − φ4 . the π 0 meson) has mass M . (8. the correct parity invariant trilinear interaction for a pseudoscalar meson should be (−gψγ5 ψφ).90) into a free part and an interaction part. in this case. Namely. The trilinear coupling gψψφ is manifestly Lorentz invariant (as we have seen in (3.9 Yukawa interaction 313 example. we can separate the Lagrangian density (8.91) (8..36). The Lorentz invariant Lagrangian density. we can denote the interaction Lagrangian density as (remember everything is normal ordered) . L = L0 + LI . 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. ψψ is invariant under a Lorentz transformation and φ is a scalar ﬁeld) and is known as the Yukawa interaction. As in (6.8. 4! (8. However. since it leads to the Yukawa potential in the Born approximation and the coupling constant g represents the strength of the interaction. where 1 M2 2 / φ . we are going to ignore this aspect of the meson and also switch oﬀ the self-interactions of the meson ﬁeld for simplicity.g. (The self-interaction can be switched oﬀ at the lowest order by setting λ = 0.92) 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. since here we are only interested in developing calculational methods in a quantum ﬁeld theory.86). L0 = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 λ LI = −g ψψφ − φ4 .92).

the adiabatic switch-oﬀ is assumed) S = T e−i R∞ R −∞ LI = −g : ψψφ := −HI .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′ ) : (8.94) 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′ ) + ······ .6). Using Wick’s theorem (see sections 6.314 8 Dirac field theory As a result. in this theory the S-matrix will have the expansion (see (6. (8.5 and 6.95) .94) +··· . we can write (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.9 Yukawa interaction 315 To see how perturbation theory works in the interacting quantum ﬁeld theory.97) contains an annihilation operator and.95).98) where the initial state contains a proton with momentum k and spin s. anti-proton and π 0 meson than the ﬁnal state. therefore. let us look at the lowest order (order g) term in the expansion (8. q .96) S1 = −ig (1) (+) d4 x ψ α (x)ψα (x)φ(+) (x). s. For example. s′ .98) we substitute the ﬁeld decomposition for each of the positive energy ﬁeld operators and write . (+) (8. (1) (8.97) Each positive energy operator in (8. k′ . (−) (−) (+) (+) (8. let us look at the ﬁrst term in (8. (There can be other nontrivial matrix elements where the initial state has one more proton.8. 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) .96) Each term in this expression leads to a distinct physical process. a nontrivial matrix element would result for the case 0|S1 |k.) To evaluate the matrix element (8. just to get a feeling for what is involved.

we note here that this lowest order matrix element in (8.99) where we have used the appropriate (anti) commutation relations between the creation and the annihilation operators in evaluating the vacuum expectation value. In fact. s) ˜ ˜ × 0|d(p′ . 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′ .96). let us consider the second . We can similarly associate a physical process with every other term of S (1) in (8. s′ .95) also leads to a distinct physical process. s)a(˜ )c† (k. The matrix element (8.99) represents the process where a proton. q = −ig d4 x s. s)d† (k′ . our goal in this section has been to describe the calculational methods in quantum ﬁeld theory. s.99) can be satisﬁed only if the energy of each of the particles vanishes which is not possible since the particles are massive. s′ )a† (q)|0 ˜ ˜ p = −ig d 3 p d 3 p d 3 p′ ˜ 1 s.˜′ =± 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′ . s′ )uα (k. s′ )c(p.99) vanishes because energy-momentum conservation cannot be satisﬁed. For example. k′ 0 . The delta function merely ensures the overall conservation of energy and momentum in the process. (2π)3 k′ 0 (8. an anti-proton along with a π 0 meson are annihilated. 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′ . energy conservation δ(k0 + k′ 0 + q 0 ) in (8. q 0 > 0. note that since k0 . k′ .) Each term at order g2 in the expansion in (8. s).316 8 Dirac field theory 0|S1 |k. s′ )uα (p. (For completeness. However.

k3 . k2 . s1 . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x ) (−) + ψ α (x )ψα (x )ψ β (x)ψβ (x)|k1 . s4 )v β (k2 . 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. s2 = − ig2 2 d4 xd4 x GF (x − x ) (−) (+) (+) (2) (−) × k4 . s1 ). (8. s4 . s4 . In such a case. k2 . (2π)4 GF x − x = (8. s3 |S2 |k1 .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 ) : . Recalling that the Fourier transformation of the Green’s function is given by (see (5. s1 . s2 )uβ (k1 . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x )|k1 .120)) d4 q −iq·(x−x ) e GF (q).101) × uα (k3 . s3 )vα (k4 . k3 . the matrix element will have the form k4 .8.102) . s4 . s1 .9 Yukawa interaction 317 term in the expansion at order g2 in (8. k2 . s2 (−) (+) (+) = −ig2 × d4 xd4 x GF (x − x ) (−) (+) (+) (−) k4 .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.

s2 )uβ (k1 . s1 ). The delta function merely ensures the overall energy momentum conservation in the process.318 the matrix element (8. Feynman developed a graphical method for evaluating these matrix elements which is in one to one correspondence with the Wick expansion. s2 )uβ (k1 . s4 )iGF (k1 + k2 )v β (k2 . 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. It is clear from looking at the expansion of the S-matrix that it contains terms . 8. s4 . s3 |S2 |k1 . k2 . s3 )vα (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. s4 )v β (k2 . k3 .101) can be written as (2) 8 Dirac field theory k4 . s2 )uβ (k1 . s3 )vα (k4 . 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 . 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 . 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 (8.103) × uα (k3 . s4 ) v β (k2 . s1 . s3 )vα (k4 .

as we have seen. simply because we have a charge neutral scalar ﬁeld . namely.αβ (k) = ǫ→0+ k2 − m2 + iǫ k = lim ǫ→0+ i k − m + iǫ / . let us introduce a graphical representation for these two basic elements of Wick’s expansion. The initial and the ﬁnal states. (This is the same as saying that a fermionic particle moves in the direction shown or an anti-particle moves in the opposite direction. the line representing this propagator has no direction associated with it. Let us deﬁne the two kinds of Feynman propagators that are possible for our theory (the Yukawa theory). We note that the scalar (meson) Feynman propagator is an even function and corresponds to that of a charge neutral ﬁeld. going from from ψ β to ψα . simply give rise to normalization factors as well as spinor functions for the particles in the initial and the ﬁnal states. are represented by external lines.104) Several things are to be noted here. an external boson line can represent either the annihilation or the creation of a particle at the interaction point (vertex). the line representing the fermion propagator has a direction associated with it. occur as internal lines in a diagram. αβ (8.10 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. Therefore. − M 2 + iǫ β i (k + m)αβ / lim α = iSF .) It should also be remembered that the fermion propagator is a 4×4 matrix with α and β representing the (Dirac) matrix indices. the Feynman propagator for the fermions is not an even function and corresponds to that of a (Dirac) ﬁeld carrying charge. therefore. Hence. k = iGF (k) = lim ǫ→0 k 2 i . On the other hand. on the other hand. Thus.8. 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. For example.

s). (anti-particle creation). for the Yukawa interaction (8.105) On the other hand. s). 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). vα (k. carries a direction for charge ﬂow). s α = Note that it is the interaction between the diﬀerent ﬁelds which gives rise to nontrivial scattering. the normalization factor for the external line √ (state) is denoted by (k0 = k2 + M 2 ) k = 1 (2π)3 2k0 . (particle creation). (8. The interaction can be read out from the Lagrangian density LI and. Thus. (particle annihilation). 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. v α (k. s α = However. in a local quantum ﬁeld theory. since the fermion ﬁeld carries a charge (and.320 8 Dirac field theory and in such a case.107) k. is represented as a vertex (point of interaction) with no meaning attached to the external lines. we . 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.93). s).106) k. (anti-particle annihilation). (8. therefore.

With these rules. one has to have a factor of (−1) for every internal fermion loop.108) These are the basic elements out of which Feynman diagrams are constructed by attaching internal and external lines to vertices.10 Feynman diagrams 321 have with the convention that all momenta are incoming at a vertex (basically. calculations of S-matrix elements are simpler in momentum space. 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.8. s k ′ . with the normalization factor (2π)4 for every momentum integration. conventionally known as Feynman rules. (8. s′ Figure 8. We can also deﬁne analogous Feynman rules in coordinate space. we can show that every Feynman diagram corresponds to a unique physical process in one to one correspondence with Wick’s expansion.) q α β k. (There is one other rule. which we do not worry about at this point. . namely. However. We should also note for completeness that these are the Feynman rules in momentum space.1: Lowest order Feynman diagram. 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 . Furthermore. There are also some other subtleties like the symmetry factor associated with a diagram (graph).

s. The . the corresponding lowest order Feynman diagram will be given by Fig. s1 k4 . 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′ .2: A second order Feynman diagram. (2π)3 k′ 0 (8. q = −(2π)4 igδ4 (k + k′ + q)δαβ × 1 (2π)3 2q 0 m uα (k. s3 Figure 8.109) This is exactly the value of the ﬁrst order S-matrix element which we had evaluated in (8. 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. s2 k3 . s′ . k′ .99). s) (2π)3 k0 m v β (k′ . therefore. k1 . 8.1. s4 α β q δ γ k2 . where the external lines represent the respective physical particles that are annihilated. s).99)) will correspond in the lowest order to evaluating the Feynman diagram in Fig.322 8 Dirac field theory In this language. The value of this diagram can be obtained from the Feynman rules to be 0|S1 |k. s′ )uα (k. an anti-proton and a π 0 meson are annihilated (which we considered in (8. 8. the physical process where a proton.2. with the external lines representing physical particles.

8. it is always resolved by going back to Wick’s expansion (of importance are concepts like the symmetry factor associated with a graph). s2 )uβ (k1 . s2 )iGF (q)(−(2π)4 ig)δ4 (q − k3 − k4 )δγδ × m 0 (2π)3 k3 m 0 uγ (k3 . 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 )) .8. For example. let us observe that the same Feynman diagram can describe distinct physical processes depending on the external lines. However. it is worth emphasizing that whenever confusions are likely to arise in the Feynman method.10 Feynman diagrams 323 value of the diagram can now be calculated using the Feynman rules to be (2) k4 . To conclude this section. This shows the power of the Feynman diagram method. the Feynman diagram in Fig. k3 . 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 (8.2 can describe the scattering of two protons through the exchange of a π 0 meson. If we are interested in only the basic diagram (and not the external line factors since the diagram can represent various distinct processes).110) × uα (k3 . this is exactly the second order S-matrix element which we had calculated from the Wick expansion in (8. s4 )iGF (k1 + k2 )v β (k2 . s1 . s3 )vα (k4 . s1 ). s3 )vδ (k4 . s3 |S2 |k1 .103). s1 )v β (k2 . Once again. s4 . 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 . namely. 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 . k2 .

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| .112) then. q 2 − M 2 + iǫ q = k1 − k2 . if we are in the center of mass frame where k1 = −k4 . 4π |x| = (8. 0 0 k2 = k3 . k2 = −k3 . we see that V (q) = V (q) and taking the Fourier transform of (8.113). (8. 0 0 k1 = k4 . it follows from energy conservation that 0 0 k1 = k2 .113) Therefore.324 8 Dirac field theory −iV (q) = −ig2 GF (q) = ǫ→0+ lim − ig2 .111) where V (q) represents the scattering amplitude in the Born approximation. (8. q = (k1 − k2 ) = (0.114) . Furthermore. (8. k1 − k2 ).

Relativistic Quantum Mechanics and Field Theory. Introduction to Relativistic Quantum Field Theory. New York. Row. Introduction to the theory of Quantized Fields. V. Relativistic Quantum Fields. Nauka. Evanston (1961). Introduction to Quantum Field Theory. Quantum Field Theory. C. Itzykson and J-B. New York (1969). New York. 2. Peterson. 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. D. 4. J.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). P. Moscow (1984). Zuber. 5.8. S. N.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 Wliley.93) leads to the Yukawa potential in the Born approximation. John Wiley. 8.11 References 1. Bogoliubov and D. Gross. 3. Bjorken and S. N. McGrawHill. McGrawHill. we obtain from (8. Drell. Roman. Shirkov.114) the potential between the fermions to be g2 e−M |x| . . F. 6. Schweber. New York (1993). 4π |x| V (x) = − (8. 1964. 1980. therefore.

This page intentionally left blank .

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. 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. this is described by the vector potential Aµ (x) which belongs to the ( 1 . Thus. To that end.2). ∂t ∇×E=− ∇ · B = 0.1) in a manifestly Lorentz covariant form. Let us start with the classical Maxwell’s equations in vacuum which are given by (remember c = 1) ∇ · E = 0. ∂t ∂E ∇×B= .Chapter 9 Maxwell ﬁeld theory 9. It is more convenient from our point of view to rewrite Maxwell’s equations (9. 1 ) representation of the Lorentz group (see sec2 2 tion 4. the dynamical ﬁeld would naturally describe particles with spin 1. we note that we can solve the second and the third equations in (9. ∂B . (9. As we know.1) where E and B represent the electric and the magnetic ﬁelds respectively.1) to write 327 .

6) (9. 2. (9.2) where A(x) and Φ(x) are known as the vector and the scalar potentials respectively.8) denotes the magnetic ﬁeld vector. (9. We can.3) If we now deﬁne the four dimensional curl of the four vector potential as Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .5) µ. Fij = ∂i Aj − ∂j Ai = −ǫijk Bk .7) (9. B2 . In other words. (9.328 9 Maxwell field theory B = ∇ × A. the six independent components of the tensor Fµν are given by (three components each of) the electric and the magnetic ﬁelds. where E = (E1 . E3 ).4) where ǫijk denotes the three dimensional Levi-Civita tensor and B = (B1 . E2 . 1. 3. of course. denotes the electric ﬁeld vector. we note that F0i = ∂0 Ai − ∂i A0 = Ei . ∂t (9. E = − ∂A − ∇Φ. Similarly. Therefore. A). (9. the second . combine them into the four vector potential Aµ = (Φ. −A). ν = 0. Aµ = (Φ. B3 ).

9) leads to ∂µ F µj = 0. for ν = j. or.12) . ∂t (9. ∂0 F 0j + ∂i F ij = 0. or.9) For example. we see that the set of four Maxwell’s equations (9.1) are given by ∂µ F µν = 0. or.1).13) (9. (9.4) as Fµν = ∂µ Aν − ∂ν Aµ .9. Thus. equation (9.4)) ∂i F i0 = 0.10) which coincides with the ﬁrst of the equations in (9.1) can be written in the manifestly covariant form as ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0.1). (9. or. (9. or. ∂Ej + ∂ i (−ǫijk Bk ) = 0. for ν = 0. or.11) which is the last equation in (9.1 Maxwell’s equations 329 rank anti-symmetric tensor Fµν is also known as the ﬁeld strength tensor. equation (9. Similarly. It is now easy to check that the other two equations (the ﬁrst and the fourth) in (9. ∂ 0 F0j + ∂ i Fij = 0. ∂t ∂Ej = −ǫjik ∂ i Bk . with the ﬁeld strength tensor deﬁned in (9.9) takes the form (F 00 = 0 by deﬁnition (9. ∂t ∂E = ∇ × B. ∇ · E = 0.

330 9 Maxwell field theory Furthermore. But ﬁrst. (9. For example. local parameter of transformation.16) = − ∂µ Aν (∂ µ Aν − ∂ ν Aµ ). the gauge invariance of Maxwell’s equations is obvious in this formulation. µ or. the ﬁeld strength tensor is unchanged under the gauge transformation (9.15) In other words. 9. 2 which leads to . µ (9. Consequently. as an operator. which denotes the dynamical variable of the theory. let us examine whether there exists a Lagrangian density which will lead to Maxwell’s equations as its Euler-Lagrange equations.2 Canonical quantization In trying to second quantize the Maxwell ﬁeld theory. real. we note that under a gauge transformation Aµ (x) → A′ (x) = Aµ (x) + ∂µ θ(x). Let us consider the Lagrangian density 1 1 L = − Fµν F µν = − (∂µ Aν − ∂ν Aµ )(∂ µ Aν − ∂ ν Aµ ) 4 4 1 (9.12) are also invariant under these transformations. Fµν ′ → Fµν = ∂µ A′ − ∂ν A′ µ ν = = ∂µ (Aν + ∂ν θ) − ∂ν (Aµ + ∂µ θ) ∂µ Aν − ∂ν Aµ = Fµν . we have to treat the four vector potential Aµ (x).14) where α(x) is an arbitrary. Maxwell’s equations (9. δAµ (x) = A′ (x) − Aµ (x) = ∂µ θ(x).14) of the four vector potential Aµ (x).

19) with both Aµ and Fµν treated as independent dynamical variables would also lead to Maxwell’s equations as its Euler-Lagrange equations. . (9.16). Note that a Lagrangian density 1 1 Fµν F µν − Fµν (∂ µ Aν − ∂ ν Aµ ) . 4 2 L= (9.19) is equivalent to (9.20) ∂µ F µν = 0. ∂Aν ∂∂µ Aν or. ∂µ (∂ µ Aν − ∂ ν Aµ ) = ∂µ F µν = 0. It is manifestly Lorentz invariant. we will work with the simpler Lagrangian density in (9. gives Maxwell’s equations as its Euler-Lagrange equations. take the form ∂L ∂L − ∂µ = 0. Fµν = ∂µ Aν − ∂ν Aµ . The Lagrangian density (9. so that the Euler-Lagrange equations.2 Canonical quantization 331 ∂L ∂Aν ∂L ∂∂µ Aν = 0.20).9). ∂∂µ Aν (9.16) since the Lagrangian density (9. ∂L = 0. However.16) when the ﬁeld variable Fµν is eliminated using the (ﬁrst) auxiliary ﬁeld equation in (9.18) which coincides with the manifestly covariant Maxwell’s equations (9. ∂Fµν or. indeed. ∂L = 0.9.17) = − (∂ µ Aν − ∂ ν Aµ ) = −F µν . namely. ∂µ or. = − 1 1 1 (∂ µ Aν − ∂ ν Aµ ) − ∂ µ Aν + ∂ ν Aµ 2 2 2 (9. in this case.

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. since the ﬁeld strength tensor Fµν is gauge invariant (see (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).24) The naive canonical quantization conditions. Furthermore. would then appear to be [Ai (x). ˙ ∂ Aµ (x) (9.21). 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. This is a consequence of the gauge invariance of the theory.25) . If we treat Aµ (x) as the independent ﬁeld variable. x0 =y 0 .21) and the ﬁrst peculiarity of the Maxwell ﬁeld theory is now obvious from (9. for canonical quantization of such a system.15)). Π0 (x) = −F 00 = 0. we have to choose a gauge and this particular choice is known as the temporal (axial) gauge. Πj (y) j = iδi δ3 (x − y). Therefore. Πj (y) x0 =y 0 = 0 = Πi (x). namely.22) In other words. we can deﬁne the associated conjugate momentum as (see (9. in this gauge. As a result.332 9 Maxwell field theory Furthermore. Aj (y)]x0 =y0 Ai (x). now must commute with all the ﬁeld variables in the theory. we note that A0 (x) which would have a nontrivial commutation relation only with Π0 (x).14) of the vector potential. (9. (9. the momentum conjugate to the ﬁeld variable A0 (x) vanishes. (9. this Lagrangian density is also invariant under the gauge transformation (9.23) This is equivalent to saying that.17)) Πµ (x) = ∂L = −F 0µ . (9.

namely. this is not quite consistent which can be seen as follows.9. in this gauge. Aj (y) j = iδi δ3 (x − y). lead to (for ν = 0) ∂µ (∂ µ A0 − ∂ 0 Aµ ) = 0. ˙ Ai (x).27) (9.27).25) leads to Ai (x).) On the other hand. the theory can be checked to have a residual gauge invariance under a time independent gauge transformation which preserves (9. we note that even after imposing the gauge condition (9. For example. the dynamical ﬁeld variable A(x) must be transverse. The proper quantization condition consistent with the transversality condition (9. x0 =y 0 x0 =y 0 (9. or.23). or. (9.23).28) and the condition (9. ∂ 0 (∇ · A) = 0. or.2 Canonical quantization 333 However. the canonical quantization condition (9. ∇ · A = 0. in the gauge (9.18)) ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0. j = −iδi δ3 (x − y). Aj (y) j = −iδi TR (x − y).27) can be thought of as the gauge choice for this residual invariance.30) . (As a parenthetical remark.27) is given by ˙ Ai (x).26) Namely. Πj (y) or. x0 =y 0 (9.23).29) which does not satisfy the transversality condition (9. (9. ∂ 0 (∂µ Aµ ) = 0. ¯ A(x) → A′ (x) = A(x) + ∇θ(x). the dynamical equations (see (9.

known as the Dirac method. there exists a systematic procedure for deriving the correct quantization conditions for such systems. ∇2 (9.32) = ∂ j + ∂ i ∂i 1 j j j 2 ∂ = ∂ − ∂ = 0.27) is given by (9.22). As we have pointed out in connection with the discussions in the Dirac ﬁeld theory. which we will discuss in the next chapter.23) and (9.27). Here 1 2 stands for the inverse of the Laplacian operator and for∇ mally denotes the Green’s function for ∇2 .30). We note here that relations such as (9. the transverse delta function in (9. When quantized systematically through the Dirac method. Let us note here that in momentum space. the Hamiltonian density for the Maxwell ﬁeld theory can be obtained to be . ∇ and (9. the correct quantization condition (Dirac bracket) that results in the gauge (9. (2π)3 (9.33) In the gauge (9.27) constitute constraints on the dynamics of the theory and systems with constraints are known as constrained systems.31) so that (remember that ∂ i ∂i = −∇2 ) j j ∂ i δi TR = ∂ i δi + ∂i 1 j 2 ∂ ∇ (9.23).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).30) is consistent with the transversality condition in (9.334 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 .23) subject to (9. (9.

9. (9.37) .3 Field decomposition In the temporal gauge (9.18) take the form ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = Aν = 0.38) (9.35) H= Indeed this is what we know to be the Hamiltonian (energy) for the Maxwell theory even from studies in classical electrodynamics. ∇ · A = 0.36) which also lead to ∂ µ Aµ = 0.34) and leads to the Hamiltonian d3 x H = = 1 2 1 2 d3 x Π2 + B2 d3 x E2 + B2 . (9. 2 (9.27)) A0 = 0. Maxwell’s equations (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 . (9.23) (see also (9.

39) Thus.27) that.336 9 Maxwell field theory As we have seen. the three components of the vector potential. λ) = 0.40) (9. the polarization vector characterizing the vector potential must be transverse to the direction of propagation of the plane wave. We can choose the two independent polarization vectors to be orthonormal and.44) ǫ∗ (k. as shown in Fig. k · ǫ (k) = 0. (9. the wave equation (9. λ′ ) = δλλ′ . we must have ∇ · A(x) = 0. 2. On the other hand. for ν = i. complex satisfying k · ǫ (k. (9. in general. (9. or. for ν = 0. we have Ai = 0.1.41) 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.42) (9. in this case. satisfy the Klein-Gordon equation for a massless particle (wave equation) and. since the vector potential is transverse.43) Namely. therefore. There can be two such independent directions. will have plane wave solutions of the form A(x) ∝ ǫ(k) e∓ik·x . for example. 9. . We note from (9. λ = 1.38) vanishes identically (because of the choice of the gauge condition (9.23)) A0 ≡ 0. λ) · ǫ(k. (9. with k0 = Ek = √ k2 = |k |.

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

49) leads to d3 x (2π)3 2k0 ← → (e−ik·x ∂0 A(x)) · ǫ(k. a† (k. λ). λ) . λ)e 2 a(k′ . a(k. we can show that d3 x (2π)3 2 ˙ eik·x A(x) 0 k0 ǫ(k. (9. λ) = −i (9. Taking the Hermitian conjugate of (9. we conclude that i d3 x ← → 2k0 eik·x ∂0 A(x) = √ (2π)3 2k0 d3 x (2π)3 2k0 2 ǫ (k. λ)a(k. λ)a(k. λ) + ǫ∗ (−k.48). λ) = 1 0 √ ǫ(k. λ)a(k. (9.338 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′ . λ) 2k′ 0 0 +k ′ 0 )t +δ3 (k + k′ )ǫ∗ (k′ .48) 2 =−i λ=1 From (9.49) where we have used the normalization for the polarization vectors in (9. λ)ei(k +k)·x a† (k′ . λ)e2ik t a† (−k. λ) + ǫ∗ (k′ . λ). λ) . λ). λ) = i ← → (eik·x ∂0 A(x)) · ǫ∗ (k. λ)e2ik t a† (−k. λ)ei(k 2 a† (k′ . λ)−ǫ∗ (−k.50) . λ)a(k′ .44). λ=1 or.47) and (9. λ) = λ=1 d3 k ′ √ δ3 (k − k′ )ǫ(k′ . (9.47) 2k0 λ=1 Similarly.

λ) behave exactly like the annihilation and the creation operators for a harmonic oscillator system. (9. If we denote the vacuum state of the theory by |0 .61) in the case of the Klein-Gordon ﬁeld).50) denote the two inversion formulae for the Maxwell ﬁeld (analogous to (5. the operators a(k. then it satisﬁes .50). (9. [a(k. λ). (9. λ)a(k.9.3 Field decomposition 339 Equations (9. λ). Aj (y)]x0 =y0 (9. Aj (y)]x0 =y0 . λ′ )] = δλλ′ δ3 (k − k′ ). λ′ )].46)) that Ek = |k|.77) and (5.97) and we do not show the normal ordering symbol explicitly) 2 H = λ=1 2 d3 k Ek a† (k. a† (k′ .51) using (9. The normal ordered forms for these operators can be shown to correspond to (see (5. Aj (y)]x0 =y0 ˙ [Ai (x).49) and (9. λ)a(k. λ).30)). Given the quantization conditions (see (9. λ). λ) and a† (k.54) The Hilbert space for this theory can be built up in the standard manner. [Ai (x). λ).52) In other words.49) and (9. we can easily show that [a(k. P = λ=1 d3 k k a† (k. j = −iδi TR (x − y). The Hamiltonian and the momentum operators for the Maxwell theory can again be expressed in terms of the creation and the annihilation operators. a† (k′ . a(k′ . λ′ )] = 0 = [a† (k. ˙ ˙ = 0 = [Ai (x).53) where we recall from the Einstein relation (see (9.

There are several things to note here. Later. the particles of the theory.23) and (9. Furthermore. involving the true degrees of freedom.340 9 Maxwell field theory a(k.27)). the true physical degrees of freedom are two in number. the energy of the system is positive semideﬁnite and the Hilbert space is the standard harmonic oscillator space. Second. First of all. (9. therefore. we achieved quantization. the one particle state satisﬁes (H 2 − P2 )|k. since the A(x) ﬁeld is Hermitian like the charge neutral Klein-Gordon ﬁeld. consequently. λ)|0 H|0 = 0 = 0|a† (k.55) = 0 = P|0 . (9. We can. λ = Ek |k.56) which satisﬁes H|k. 2) |k. by eliminating the extra degrees of freedom through non-covariant gauge choices (see (9. (9. λ = a† (k. λ)|0 . However. in fact. do not carry any electric charge. The one particle state is then obtained to be (λ = 1. λ P|k. shows that even though the vector potential Aµ (x) has four independent ﬁeld degrees of freedom. This. (9. = k |k. λ 2 = (Ek − k2 )|k. λ = 0. λ). there are two distinct one photon states corresponding to the two possible independent transverse polarizations a photon can have. describes a one photon state and the index λ merely labels the polarization of the photon. λ .57) Therefore. 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. As a result. show explicitly that the polarization is directly related to the spin of photon. we will also . namely the photons.58) This state. λ .

. 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.62) describe respectively the left and the right circular polarization vectors.) 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 .59) then. 2 1 √ a† (k. 2)) 2 (9. 2) as creating a photon polarized along the y-axis. in this language.60) would then create photons which are left and right circularly polarized respectively. 2) . e 2 1 ˆ ǫR (k) = √ ex + iˆy . 1) as creating a photon polarized along the x-axis and a† (k. 1) + ia† (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)) e e 2 1 e e + (ˆx − iˆy )(a† (k. 1) − ia† (k. 1) − ia† (k.61) = ǫ∗ (k)a† (k) + ǫ∗ (k)a† (k). 2) λ=1 = 1 (ˆx + iˆy )(a† (k. 2) . λ)a† (k. 1) = ex . 2 (9. It is clear. λ) = ex a† (k. 1) + ey a† (k. ˆ ǫ(k. 2) = ey . e 2 (9. that the operators a† (k) = L a† (k) = R 1 √ a† (k. 1) + ia† (k.9. we can identify a† (k. (These one photon states are eigenstates of helicity with eigenvalue ±1. (9.

it follows that the only nontrivial covariant commutation relations have the form (these are not at equal time) 2 [A(+) (x).λ′ =1 1 (2π)3 d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ ×ǫµ (k. λ).65) A(−) (x) = µ λ=1 From the commutation relations for a(k.342 9. 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. ǫ(k. λ)(e−ik·x a(k. λ)ǫν (k′ . λ). for simplicity. if we deﬁne the two polarization vectors for λ = 1. 2 as four vectors ǫµ (k. 3 2k 0 (2π) d3 k (2π)3 2k0 ǫµ (k.45) as (recall (9. λ) in (9. (9.64) with k0 = |k|. λ)eik·x a† (k. λ)ǫν (k′ . λ) = (0. λ)). a† (k′ . A(−) (y)] µ ν = λ. (9. λ′ )e−ik·x+ik ·y δλλ′ δ3 (k − k′ ) . λ)).λ′ =1 1 (2π)3 ′ d3 k d3 k ′ √ √ 2k0 2k′ 0 ×ǫµ (k. λ) and a† (k.52). λ′ )] 2 = λ. λ)e−ik·x a(k.27)). (9. λ)+eik·x a† (k. if we choose the polarization vectors to be real as in (9.23) (see also (9. λ′ )e−ik·x+ik ·y [a(k.23)) 2 Aµ (x) = λ=1 d3 k (2π)3 2k0 ǫµ (k.59) then we can write the ﬁeld decomposition (9.4 Photon propagator 9 Maxwell field theory Let us note that in the temporal gauge (9. λ).63) and furthermore.

λ)ǫν (k. we obtain : Aµ (x)Aν (y) : +[A(+) (x).145)).68) From this. A(−) (x)] = iG(+) (y − x) µ ν ν µ νµ = − 1 2 d3 k eik·(x−y) (2π)3 k0 2 ǫµ (k. A(−) (y)] ν µ : Aµ (x)Aν (y) : −iG(+) (x − y) µν : Aµ (x)Aν (y) : + Aµ (x)Aν (y). namely. µν (9. 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.66) which are the analogs of the positive and the negative energy Green’s functions for the Maxwell ﬁeld (see (5.4 Photon propagator 2 343 d3 k ǫµ (k. we can deﬁne the Feynman propagator in the standard manner (see (6.143) and (5. A(+) (y)] = −[A(+) (y).107)). . (9. λ) λ=1 = −iG(+) (x − y). µν (9.67) Aµ (x)Aν (y) = 0|Aµ (x)Aν (y)|0 = −iG(+) (x − y).9. λ)ǫν (k. 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. λ)ǫν (k. µν [A(−) (x). λ) λ=1 = −iG(−) (x − y).

59)). 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. (9. therefore. (9.100) 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 .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. for example.344 9 Maxwell field theory 0|T (Aµ (x)Aν (y)) |0 = iGF .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)). 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 ) . λ)ǫν (k.69) µν By deﬁnition. the Feynman Green’s function has the form GF .(9. we use the integral representation for the step function given in (6. λ)ǫν (k.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. To simplify this expression.

2πi (k′ 0 )2 − k2 + iǫ ′0 0 0 0 1 1 − k′ 0 − k0 + iǫ k′ 0 + k0 − iǫ 0 ǫ→0+ lim − (9.74) .µν (x − y) = lim ǫ→0+ d4 k e−ik·(x−y) (2π)4 k2 + iǫ 2 ǫµ (k.46). we can identify the momentum space Feynman Green’s function for the Maxwell (photon) ﬁeld to be GF .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 ) . let us note that if we introduce two new four vectors η µ = (1. 0.73) where GF (k) denotes the Feynman Green’s function for a massless scalar ﬁeld (see (5. λ) λ=1 = GF (k) λ=1 ǫµ (k. |k| (9. ˆ kµ = kµ − (k · η)η µ (k · η)2 − k2 = 0.9. λ=1 (9. λ)ǫν (k.71) where we have used the Einstein relation (9.140)).72) Therefore.70) and denoting the variable of integration k′ 0 → k0 . λ)ǫν (k. k . Substituting this into (9. λ) .µν (k) = ǫ→0+ lim 1 k2 + iǫ 2 2 ǫµ (k. λ) . The Feynman Green’s function in (9. 0. (9.73) depends on the polarization sum and to evaluate the polarization sum. we obtain GF . 0). λ)ǫν (k.

physical calculations carried out with this propagator do lead to Lorentz covariant results. Substituting this back into (9.73). η µ ηµ = 1. they satisfy ǫµ (k.77) The Feynman propagator has a non-covariant look in this gauge. λ) = −ηµν + ηµ ην − kµ kν . ˆ kµ ǫµ (k. λ)ǫν (k.76) or. λ) + ηµ ην − kµ kν = ηµν .µν (k) = 0. λ)ǫσ (k. ˆ η µ ǫµ (k. However. as expected (since we are in the gauge A0 = 0).78) and. η µ . λ)ǫν (k. − 2 λ=1 ˆ ˆ ǫµ (k. ˆ ˆ (ǫσ (k. also satisﬁes . (9. λ)ǫν (k. kµ GF . is manifestly transverse (see (9.µν (k) = lim − ǫ→0+ k2 (9. λ) + σ + = ηµν . we can write the completeness relation for these basis vectors as ˆ ˆ ηµ ην kµ kν ǫµ (k. λ′ ) = −δλλ′ .75) Consequently. we obtain the Feynman propagator for the photon ﬁeld in the temporal gauge to be i ˆ ˆ (ηµν − ηµ ην + kµ kν ). in this gauge. 2 deﬁne an orthonormal basis in four dimensions. λ). Namely. kµ and ǫµ (k.346 9 Maxwell field theory ˆ then. λ=1 ˆ ˆ ǫµ (k. λ) = 0. λ) = 0 = η µ kµ . λ)) (η ησ ) (kσ kσ ) λ=1 2 2 or. + iǫ iGF . The Feynman propagator.75)). namely. (9. λ)ǫµ (k. (9. λ = 1. ˆ ˆ kµ kµ = −1.

83) .83)) and we have separated the Lagrangian density in (9.82)) can be checked to be invariant under the ﬁnite local gauge transformations ψ(x) → ψ ′ (x) = e−iθ(x) ψ(x). Aµ (x) → A′ (x) = Aµ (x) + µ ′ 1 ∂µ θ(x).5 Quantum electrodynamics 347 GF . (9.0µ (k) = 0. 4 = L0 + LI . 9.82) The total Lagrangian density (9. 4 LI = −eψγ µ ψAµ .74).80) into the free and the interaction parts as 1 / L0 = − Fµν F µν + iψ∂ ψ − mψψ. ψ(x) → ψ (x) = ψ(x)eiθ(x) . (9. e (9. 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.80) denotes the covariant derivative (see (7.81) (9.80) (or (9.5 Quantum electrodynamics (9. where Dµ ψ(x) = (∂µ + ieAµ (x))ψ(x). which follows from (9.79) Quantum electrodynamics (QED) is the theory describing the interaction of electrons and positrons with the electromagnetic ﬁeld.9.

the minimally coupled Dirac Lagrangian density transforms as ′ LDirac (ψ.85) It follows from this that under a (ﬁnite) gauge transformation. let us note that the covariant derivative in (9.80).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. ψ . First.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) where ǫ(x) denotes the inﬁnitesimal local parameter of transformation. In fact.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). Aµ ) → LDirac (ψ ′ . Therefore. ψ. δǫ Aµ (x) = A′ (x) − Aµ (x) = µ ′ 1 ∂µ ǫ(x).81) transforms covariantly under a gauge transformation (see (7. we need to examine the invariance of the minimally coupled Dirac Lagrangian density in (9.86) . (9. e (9. ψ.84)). Aµ ). A′ ) µ ′ = iψ γ µ Dµ ψ ′ − mψ ψ ′ ′ ′ = iψγ µ Dµ ψ − mψψ = LDirac(ψ. = iψeiθ(x) γ µ e−iθ(x) Dµ ψ − mψeiθ(x) e−iθ(x) ψ (9.83) (or (9. δǫ ψ(x) = ψ (x) − ψ(x) = iψ(x)ǫ(x). we have already seen that the Lagrangian density for the Maxwell theory is invariant under the gauge transformation in (9.

therefore.87) In momentum space the current conservation takes the form kµ J µ (k) = 0.90) (2π)3 2k0 where we have chosen a real polarization vector as in (9.64) so that the external line in (9.83) and.77)) correponds to the temporal gauge (9. (9. In this theory. In addition.µν (k) = k. µ k ν = iGF . the complete QED Lagrangian density (9.90) can represent both the annihilation as well as the creation of a photon.89) lim − 1 ˆ ˆ i(ηµν − ηµ ην + kµ kν ) . The form of the propagator in (9. λ). (9. (9.80) r µβ α q p = −(2π)4 ie (γ µ )βα δ4 (p+q +r). in QED we have the photon propagator as well as the photon external line factor which take the forms. ∂µ J µ (x) = 0. We note here that the photon propagator is a gauge dependent quantity. The interaction vertex for QED can also be read out from the Lagrangian density (9.23) (see also (9.104) and in (8. J µ (x) = ψγ µ ψ. We have already derived the fermion propagator as well as the fermion external line factors in (8.89) (or (9.107).27)). (9. λ µ = ǫ→0+ (9.5 Quantum electrodynamics 349 Namely. there is a conserved current.80) is gauge invariant. the minimally coupled Dirac Lagrangian density is also invariant under the gauge transformation (9.91) . k2 + iǫ ǫµ (k.106)-(8.88) The Feynman rules for QED can be derived in the standard manner.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. s1 ν + k1 − k4 k2 . 9.2. . λ′ ν µ k3 . s2 k4 . but not the vertices (points) where each of the two photons was produced. s2 k1 . λ µ k1 − k3 k1 .2: The two lowest order Feynman diagrams that contribute to the pair production of photons as well as to the Compton scattering. 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 ). Therefore.2. 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 .350 9. a) Photon pair production: For example. 9.6 Physical processes 9 Maxwell field theory With these rules. we only measure two photons in the ﬁnal state. λ Figure 9. the amplitude for this process would simply be the sum of the amplitudes (quantum superposition principle) given by the two diagrams in Fig. λ′ k4 . s1 k2 . We note here that experimentally.92) To lowest order there are two Feynman diagrams which will contribute to such a process and they are shown in Fig. (9.

λ). λ)u (k1 . (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 . 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′ uα (k1 . e+ (k2 ) = 1 1 4 (2π)6 (2π) 1 1 0 2k3 0 2k4 m 0 k1 m 0 k2 ×ǫµ (k3 . K = k1 − k3 .4). depending on whether we are interested in summing over all spin states of the fermions or not.94) and the limit ǫ → 0+ is understood.93) where. s2 ) ǫ (k3 . λ) / 1 ǫ (k3 . (9. the γ-algebra can be simpliﬁed and a simple formula for the diﬀerential cross-section can be obtained (as discussed in section 3.6 Physical processes 351 γ(k3 . λ′ ) +v (k2 .95) . / ˜ − m + iǫ K / (9. λ) ǫν k4 . γ(k4 . λ′ u (k1 . s2 ) × −(2π)4 ie (γ ν )βδ iSF . s1 ) / K − m + iǫ / 1 ǫ k4 .δγ (K)(−(2π)4 ie) (γ µ )γα ˜ −(2π)4 ie (γ µ )βδ iSF . s2 ) ǫ (k4 . s1 ) v β (k2 . The diﬀerential cross-section will be proportional to the absolute square of the matrix element. we have deﬁned ˜ K = k1 − k4 . for simplicity.9. s1 ) . λ′ )|S (2) |e− (k1 ). Furthermore.

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 ). λ)u(k1 . we will assume the ﬁnal state electron and photon (e− (k3 ). γ(k4 )) to be outgoing. Thus. instead of an electron and a positron being annihilated as was the case in (9. λ′ ) +u(k3 .93) to be e− (k3 ). s3 )ǫ (k4 . this amplitude will differ from (9. λ′ )u (k1 .3. (9.2 except for the labeling of the momenta. 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 (9.93) (or (9. 9.98) To the lowest order. Therefore. γ(k4 .96)) not only in the external line factors. (9.94) K = k1 − k4 and we have introduced Q = k1 + k2 . namely. s3 )ǫ (k2 . λ) = − ie2 (2π)2 1 0 2k2 1 0 2k4 m 0 k1 m 0 k3 / × u(k3 . Here. this process is described by the two Feynman diagrams shown in Fig. 9.96) / ˜ − m + iǫ K / ˜ where as deﬁned in (9. λ′ )|S (2) |e− (k1 ). λ) / 1 ǫ (k2 .90) is the same for the creation or the annihilation of a photon).93). but . s1 ) . γ(k2 . In this case. (9. The amplitude diﬀers from the case of the pair production of photons by only the external line factors.352 9 Maxwell field theory are given to the lowest order by the same diagrams as in Fig. we can obtain the S-matrix element for Compton scattering directly from the results in (9. s1 ) / Q − m + iǫ / 1 ǫ (k4 . the internal line represents a photon propagator signifying that the electrons scatter by exchanging (emitting and absorbing) a photon.

s4 )γ µ u(k2 . +u(k3 . This can be calculated in a straightforward manner as before.94). s3 )γ u(k2 . carrying out the momentum integration for the internal line and omitting the overall energy-momentum conserving delta function δ4 (k1 + k2 − k3 − k4 ). the Bhabha scattering can be thought of as the process . s2 ) ˜ 2 + iǫ K (9. s3 )γ ν u(k1 .99) ˜ where K and K are deﬁned in (9. Assuming that the ﬁnal state electrons are outgoing. also in the propagator of the diagram. s2 ) µ ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . the S-matrix element takes the form e− (k3 ). e− (k2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k4 .9. s1 ) . d) Bhabha scattering: Finally.6 Physical processes 353 k3 k4 k4 k3 + k1 − k3 k1 − k4 k1 k2 k1 k2 Figure 9. s4 )γ ν u(k1 . e− (k4 )|S (2) |e− (k1 ).3: The two lowest order Feynman diagrams for M¨ller scato tering. s1 ) 2 + iǫ K ˆ ˆ ˜ ˜ (ηµν − ηµ ην + Kµ Kν ) u(k4 .

9. the second diagram simply describes the scattering of an electron and a positron by exchanging a photon.4: The two lowest order Feynman diagrams for Bhabha scattering. The lowest order Feynman diagrams for this process are shown in Fig. On the other hand. s3 )γ ν u(k1 . k2 k4 k3 (9. 9. e+ (k4 . In this case. 9.100) k4 + k1 + k2 k1 k3 k1 k1 − k3 k2 Figure 9.4. s1 ). s2 )γ µ v(k4 . but in the present case we will treat the ﬁnal state electron and the positron as outgoing (e− (k3 ). s2 )γ ν u(k1 . the ﬁrst diagram in Fig. We see that topologically these diagrams are the same as in Fig. s3 )γ µ v(k4 .3. s2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k3 . s1 ) .354 9 Maxwell field theory e− (k1 ) + e+ (k2 ) → e− (k3 ) + e+ (k4 ). e+ (k2 . We can calculate this matrix element as before and omitting the overall delta function representing conservation of energy-momentum.4 represents the annihilation of an electron and a positron to create a photon which subsequently creates an electron and a positron. 2 + iǫ K (9. we have e− (k3 . s4 ) +v(k2 . s1 ) 2 + iǫ Q ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . s4 )|S (2) |e− (k1 .101) . s4 ) ˆ ˆ (ηµν − ηµ ην + Qµ Qν ) v(k2 . s3 ). e+ (k4 )).

This gives a ﬂavor of low order calculations in QED. 4 (9. e (9. 0) (iSF (p)) . 9.103) where the i factor in the denominator is understood. depends on the choice of gauge.80) for QED has the form 1 ¯ ¯ L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ. 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. However. It follows from (9.102) The structure of the photon propagator in the theory.94) and (9. (9.7 Ward-Takahashi identity in QED 355 where as deﬁned in (9. the Lagrangian density (9. Here we would only give a very simple derivation of this relation in QED.104) .9. of course. −p. independent of the choice of gauge. 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. let us discuss very brieﬂy the Ward-Takahashi identity in QED.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.7 Ward-Takahashi identity in QED Before we discuss more complicated symmetries. As we have seen.97) K = k1 − k3 and Q = k1 + k2 . let us note that the fermion propagator in QED can be represented as i = p−m / iSF (p) = p . which we will study later.

−p. by inverting (9. As a result the right-hand side of (9. µ F ∂p e (9. (9.108) . −p. a diagrammatic representation for the basic identity (9. −p. −p.107) ∂ (iSF (p)) (iSF (p))−1 ∂pµ = −i which can also be written as i 1 ∂ −1 S (p) = − Γµ (p.91) in QED with a zero momentum photon (without the delta function and the factor of (2π)4 ). k = 0). k = 0) = −(iSF (p))−1 e = ∂ (iSF (p))−1 ∂pµ ∂ −1 S (p).106) In this case. 0) denotes the three point vertex (9. ∂pµ F (9. e Consequently.104) in QED resulting from the structure of the fermion propagator has the form ∂ ∂pµ p =− 1 e p k=0 p .105) 1 = − (iSF (p))Γµ (p. k = 0)(iSF (p)).356 9 Maxwell field theory where Γµ (p.104) we can obtain a relation between the fermion two point function and the three point vertex function as 1 Γµ (p.104) can be diagrammatically represented as i i iγµ p−m / p−m / 1 e = − p k=0 p (9.

let us look at the fermion self-energy at one loop.5: Two loop corrections to the fermion self-energy in QED. we see that (9. such relations are called Slavnov-Taylor identities which we will discuss later). Once again. but we understand that the external momentum is channelled only through the fermion lines) + Figure 9.109) Here we are looking only at the proper vertex parts of the diagrams (namely.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. The diagram on the left-hand side in (9. (The trick in obtaining this relation in a simple manner lies in channeling the external momentum only through the fermion lines.108) holds for the one loop corrections to the amplitudes. 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. 9. Using the graphical identity in (9.5 (there are no external lines and we do not label the momenta for simplicity.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.) At two loop order the fermion self energy diagrams have the forms shown in Fig.9. without the external lines). . For example.106).

Thus. (9.106) at this order we obtain ∂ ∂pµ + =− 1 e + + + + + =− 1 e + + + + + .108) diagrammatically at every order in perturbation theory. 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 (9.110) 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 . A simple and powerful consequence of the Ward-Takahashi identity can be derived as follows.110) correspond precisely to the two loop corrections to the vertex function. using (9.106) we can easily verify the validity of the identity (9.

requires that the kinetic energy part of the fermion and the photon interaction term must combine into the form of a covariant derivative (see (9.108) we conclude that these constants must be related as Z1 = Z2 .108) is a consequence of gauge invariance can be seen heuristically in the following way.7 Ward-Takahashi identity in QED 359 −1 −1 SF (p.83). (9. we can identify a = 1 + B = 1 + C. C.115) (9.113). Z2 are constants.9.114) .112) This physically implies that if there are divergent parts in the fermion self-energy graphs at higher order. 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). on the other hand. then from (9. B. (9. D are constants and the dots denote other structures that may be induced due to quantum corrections. Γµ −→ Z1 Γµ .111) where Z1 . Invariance under the local gauge transformations (9. That the Ward-Takahashi identity (9. (9.113) −eCψA − Dψψ + · · · /ψ where A. m = 0). Thus comparing with (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 + · · · . m = 0) −→ Z2 SF (p. (9.85)) iaψD = iaψ(∂ + ieA /ψ / /)ψ.

21) and (9.360 which is another way of saying 9 Maxwell field theory Z2 = 1 + B = 1 + C = Z1 . the Lagrangian density (9. ˙ ∂ Aµ Πµ = which implies the constraint Π0 = −F 00 = 0. 9. 4 Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .118) As we have already seen in (9. then the equations of motion (see (9. since the ﬁeld strength tensor Fµν does not change under a shift of the vector potential by a gradient.15). Let us recapitulate brieﬂy how this arises. The canonical momenta conjugate to the ﬁeld variables Aµ can be calculated from the Lagrangian density and take the forms (see (9. (9. As a result. the present choice of gauge is known as the Coulomb gauge).8 Covariant quantization of the Maxwell theory As we have seen. (9.117) as well as the action for this theory are invariant under the gauge transformation δAµ = ∂µ α(x). canonical quantization of Maxwell’s theory results in a lack of manifest Lorentz invariance.117) where Fµν denotes the ﬁeld strength tensor.18)) . 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. (9. We recall that the Lagrangian density for Maxwell’s theory is given by 1 L = − Fµν F µν .116) We will discuss these topics in more detail when we study renormalization chapter 16.22)) ∂L = −F 0µ .

A0 = − 1 J0 .119) On the other hand. ∂i ∂ i A0 − ∂ 0 Ai = 0. lead to (for ν = 0) ∂µ F µ0 = ∂i F i0 = 0. or. But.9. the dynamical equations ∂µ F µν = J ν . ∇2 (9. We recognize that this is a consequence of our choice of gauge. . remains manifestly Lorentz invariant in the canonical formalism. (9. but we lose manifest Lorentz covariance in the process (because all components of Aµ are not being treated on a equal footing). The canonical quantization can now be carried out for the two physical degrees of freedom. or. or.8 Covariant quantization of the Maxwell theory 361 ∂µ F µν = 0. in the Coulomb gauge ∇ · A = 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.120) In either case. A0 = 0. or. ∇2 A0 = −J0 . however. or. Let us emphasize here that the ﬁnal result for the calculation of any physical amplitude. if sources (charges and currents) are present. we lose manifest Lorentz covariance in the intermediate steps (which is highly desirable in any calculation). would lead to ∂µ F µ0 = J 0 . ∇2 A0 = 0.

) Furthermore. we see that P µν can be thought of as µν a projection operator. − ∂µ∂λ (9. it projects on to the space of the components of any vector transverse (perpendicular) to the gradient operator ∂ µ . we note that ∂µ P µν = ∂µ (η µν = (∂ ν − − ∂µ∂ν ) ∂ ν ) = 0 = ∂ν P µν . (In fact.122) (9.362 9 Maxwell field theory The need for a choosing a gauge can be seen in an alternative manner as follows.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 .) 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. (9. Consequently. 4 2 where P µν = η µν − ∂µ∂ν . (This is relevant in the discussion within the context of a path integral quantization of the system which we will discuss brieﬂy in chapter 12.123) With a suitable normalization. This implies that even if we can quantize the theory (say. therefore.124) so that this is the transverse projection operator. the Feynman propagator for the theory cannot be deﬁned. in the naive path integral formalism). (9. the inverse of P µν does not exist (this also means that the determinant of P µν vanishes) and the Green’s function and. we cannot carry out calculations in perturbation theory. . P = 1 P µν deﬁnes the normalized projection operator. namely.

this gauge is known as the Feynman-Fermi gauge. We can take an alternative approach. (This formulation of the theory is due to Fermi and considered as a gauge choice. Therefore. (9.125) . the Cauchy initial value problem cannot be uniquely solved.126) Here we have generalized Maxwell’s theory to include a conserved 1 current.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.125) ∂µ J µ = 0. But to understand the issue better. the Lagrangian density 1 1 L = − Fµν F µν − (∂µ Aµ )2 + J µ Aµ . Let us consider. But more than that we have also added a term − 2 (∂µ Aµ )2 to Maxwell’s Lagrangian density. simply because the Green’s function does not exist. without any further input. for example. But clearly the theory (9. In such a case. However.) This additional term breaks gauge invariance and consequently leads to a nonsingular theory. Namely. at this point there is no justiﬁcation for adding this new term to the Lagrangian density. let us look at the equations of motion following from the action in (9. we can try to modify the theory so as to make it nonsingular.125) would appear to be diﬀerent from Maxwell’s theory. We can solve for the constraints and quantize only the true dynamical degrees of freedom. 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. 4 2 where J µ represents a conserved current (9.9. since we realize that the diﬃculties in quantization arise because of the singular nature of the Lagrangian density. the system is singular and contains constraints among the ﬁeld variables (as we will discuss in the next chapter). in this procedure we lose manifest Lorentz covariance since we single out the transverse degrees of freedom.

∂∂µ Aν ∂Aν −∂µ F µν − ∂ ν (∂ · A) − J ν = 0. ∂µ F µν + ∂ ν (∂ · A) = −J ν . (∂ · A) = ∂ν J ν .129) ∂ · A = −∂ · J = 0. (9. (9.127) takes the form ∂µ (∂ µ Aν − ∂ ν Aµ ) + ∂ ν (∂ · A) = −J ν .131) seems to modify the theory.128) (9. or.127). If we now write out the left-hand side explicitly. An alternate way to see this is to note that if we take the divergence of the equations of motion in (9. (9. ∂t χ = 0. or. we recognize that if we restrict classically to the initial value conditions ∂χ = 0. (9. ∂L ∂L − = 0.125) where χ = ∂ · A.364 9 Maxwell field theory ∂µ or.130) where we have used the anti-symmetry of the ﬁeld strength tensor as well as the conservation of J µ . Thus we see that although the presence of the term − 1 χ2 in the 2 Lagrangian density (9.127) Without the second term on the left-hand side in (9.132) . (∂ · A) = 0. therefore. χ is a free ﬁeld and. (9. Furthermore. t = 0.127). we have ∂ν ∂µ F µν + or. the presence of this additional term in the Lagrangian density would not change the physics of Maxwell’s theory. this is just the Maxwell’s equations in the presence of conserved sources. or. Aν = −J ν . (9.

˙ ∂ Aµ Πµ = (9.133) (9. We now see that this quantization relation is exactly like the quantization condition for four distinct scalar ﬁelds (see (5. ˙ 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. 2 (9. Πν (y)]x0 =y0 = iδµ δ3 (x − y). Thus. Aν (y) x0 =y 0 = −iηµν δ3 (x − y).136) with all other commutators vanishing. (9. classically we can think of Maxwell’s theory as described by the modiﬁed Lagrangian density (9.125) with the supplementary condition ∂ · A = 0. We can deﬁne the canonical momenta as ∂L ˙ = −Aµ . or.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).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).8 Covariant quantization of the Maxwell theory 365 then χ = 0 at all times and we recover the familiar Maxwell’s theory.134) It is now obvious that the coeﬃcient matrix of highest derivatives is nonsingular in this case. φ(y) = iδ3 (x − y). (9.137) x0 =y 0 .

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

even (9.142) is a very stringent condition and we have to relax this further.) We note that.141) x0 =x′0 ∂µ Aµ (x). we can think of the supplementary condition as selecting out the subspace of the physical Hilbert space of the theory.136) are consistent.140) in the quantum theory.) As we would see shortly. We . Πν (x′ ) x0 =x′0 x0 =x′0 (This relation is essentially correct. whereas the left-hand side of the above expression would be zero if the supplementary condition (9. not made use of the supplementary condition (9. We have. ∂µ Aµ |phys = 0. (9.9. = iη µν δ3 (x − x′ ). namely. A little bit of thinking tells us that although in the classical theory we can impose the condition ∂µ Aµ = 0. (9. the non vanishing right-hand side would lead to an inconsistency. Πν (x′ ) ∂µ Aµ (x). The supplementary condition (9. = i∂ ν δ3 (x − x′ ). Πν (x′ ) or.133) and say that this condition is true only on the space of physical states of the theory.133) may be thought of as picking out the smaller physical subspace from the larger total vector space. Thus. where Aµ ’s are operators.133) were to hold as an operator equation. we can weaken the supplementary condition (9.8 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. 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. (Note that since all ﬁeld components are dynamical in this modiﬁed theory. however. It is easy to see that the commutation relation (9.142) In other words.136) would lead to Aµ (x). = i∂ ν δ3 (x − x′ ).133) as yet. but the derivation needs to be done carefully in a limiting manner x0 → x′0 . or. (9. such a condition is hard to implement as an operator condition.

(9. Πν (x′ ) |ψ or. λ) as annihilation and creation operators satisfying a(k. here they are not required to be transverse. (Unlike the earlier discussion. λ). λ′ ) .144) where fk (x) = 1 (2π)3 2ω e−ik·x . 3. λ)’s are the components of the polarization vector for a photon travelling along k and polarization index λ = 0.74)). λ)fk (x)a† (k. a† (k′ .368 9 Maxwell field theory can see this already from the commutator relation (9.145) Here ǫµ (k.) They are normalized as 1. a† (k′ . 0 = i∂ ν δ3 (x − x′ ). λ′ ) = 0 = ′ a† (k. At this point we can quantize the coeﬃcients a(k. λ).146) ǫµ (k. 1. To proceed further. λ = λ′ = i = 1. λ) and a† (k. we can always choose a basis such that the polarization vector for λ = 0 is time-like (recall η µ ). k0 = ω(k) = |k|. λ′ ) = η λλ = ′ As we have seen earlier (see (9. 2.141). (9. then ψ| ∂ · A(x).143) which is inconsistent. (9. we note that if |ψ represents a physical state. a(k′ . let us make a plane wave expansion for the ﬁeld variables Aµ (x) = λ ∗ + ǫ∗ (k. (9. −1. Basically. λ)fk (x)a(k.147) = −η λλ δ3 (k − k′ ). . x0 =x′ 0 = ψ|i∂ ν δ3 (x − x′ )|ψ . λ)ǫ∗µ (k. µ d3 k ǫµ (k. λ′ ) 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. 2. λ) . 3. λ). λ = λ′ = 0. λ) (9.

ǫσ (k. λ) (9. ǫσ (k. µ = λ. λ). λ)ǫν (k. λ)ǫ∗ (k. λ′ )fk′ (y)a† (k′ . Aν (y) = λ. λ′ )fk′ (y)a(k′ . λ) + ǫ∗ (k.149) λ in (9. we do not write the equal-time arguments explicitly although they are assumed) ˙ Aµ (x). λ) k (9. λ)ǫ∗ (k′ . λ′ )fk (x)fk′ (y) ν µ = λ d3 k(−iω) ǫµ (k. λ)fk (x)a(k. λ) ν ∗ fk (x)fk (y) ǫσ (k. we have to check that these conditions are consistent with the equal-time quantization conditions for the ﬁeld variables (for simplicity. λ)ǫν (k′ . λ)ǫ∗σ (k. λ)fk (x)a† (k. λ)ǫ∗ (k′ .λ′ ∗ −iω ′ ǫν (k′ .λ′ d3 kd3 k′ iω ′ ∗ × ǫµ (k. λ′ )fk (x)fk′ (y) a† (k. λ) ∗ µ f (x)fk (y) . λ). λ)ǫν (k′ . 1. we obtain the equal-time commutator to be ˙ Aµ (x). λ′ ) µ = λ. λ). λ)ǫ∗ (k. (9.150) . λ)ǫ∗σ (k. λ)ǫ∗σ (k.λ′ d3 kd3 k′ (−iω ′ η λλ )δ3 (k − k′ ) ′ ∗ ∗ × ǫµ (k.148) Using the completeness relation for the polarization vectors ǫµ (k. λ) ν = ηµν . λ) + ǫ∗ (k. λ′ ) ν ∗ d3 kd3 k′ ǫµ (k. λ′ )fk (x)fk′ (y) a(k. λ′ ) − ǫ∗ (k′ . λ′ = 0. Aν (y) = ∗ ∗ d3 k(−iω)ηµν [fk (x)fk (y) + fk (x)fk (y)] = −iηµν δ3 (x − y).148). Of course. a(k′ .9.8 Covariant quantization of the Maxwell theory 369 where λ. a† (k′ . λ′ ) ν ∗ −ǫ∗ (k. λ′ )fk (x)fk′ (y) + ǫ∗ (k. 2. 3.

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. 3) .147) that although a(k. 3. 1. λ)|0 = 0. (9. 2. let us denote by |0 the vacuum state of the theory so that 0|0 = 1. λ = 0 = d3 k F (k)a† (k. For example. 3. (9. 2) + a† (k. (9. 2)a(k. The norm of this one photon state can now be calculated (9. λ = 0)|0 . it has the form :H: = d3 k ω(k) −a† (k. 1) +a† (k. 0)a(k. λ) and a† (k. 1)a(k.147). λ = 0.154) . If we calculate the normal ordered Hamiltonian of the theory (recall J µ = 0). 2. a(k.152) Let us next consider the state with one time-like photon |1. However. 0) + a† (k. it is strange to note that there are four independent photon degrees of freedom. We also note from (9.151) 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. λ) behave respectively like annihilation and creation operators for λ = 1.153) where F (k) is a suitable smearing function such that d3 k |F (k)|2 < ∞. for the time-like component there is a relative negative sign in the commutation relation (9.

(9. 1) +a† (k. 3) . 0)a(k. so that the one time-like photon state is given by (9.156) |k. λ = 0|1.9. λ = 0 = = = − d3 kd3 k′ F ∗ (k)F (k′ ) 0|a(k.157) In this case. 3)a(k. (9.158) and the energy of this one photon state is given by . 0) + a† (k. 0)a† (k′ . 1)a(k. However. 0 = a(k.147) for the annihilation and the creation operator for a time-like photon. (9. 0)|0 . 0) − δ3 (k − k′ )|0 d3 k |F (k)|2 < 0. 2) + a† (k. 2)a(k.8 Covariant quantization of the Maxwell theory 371 1. this also leads to trouble as we see below.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. 0)|0 = 0. Let us suppose that a† (k. 0)a† (k. the normal ordered Hamiltonian will have the form (compare with (9. 0)|0 d3 kd3 k′ F ∗ (k)F (k′ ) 0|a† (k′ .151)) :H: = d3 k ω(k) −a(k. We also realize that this negative norm is a consequence of the negative sign in the commutator (9. Thus the natural modiﬁcation that comes to mind is to interchange the roles of these operators for the time-like photon.

1. deﬁnite states of arbitrarily high negative energy values are possible and the Hamiltonian becomes unbounded from below. mλ denote number of photons for each of the polarizations λ = 0. the physical states must satisfy ∂µ Aµ (x)|phys = 0. 0 = Ha(k. a(k. n2 . 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 .160) where nλ . a(k. n1 . This not only demands that certain . In fact it is obvious that with this deﬁnition. (9. Thus states containing an odd number of time-like photons have negative norm. 0)|0 = −ω(k)|k. This is seen by using the supplementary condition (subsidiary condition) which we still have not imposed. as discussed earlier this is too stringent a condition to allow any state of the radiation ﬁeld. 0 . In a free theory we may impose suitable conditions to prohibit any unwanted state.161) However. 0)a† (k′ .159) = −ω(k)a(k. As we will see shortly in spite of the presence of negative norm states. The question that immediately comes to our mind is what happens to the probabilistic interpretation of the theory. Namely. 2.372 9 Maxwell field theory H|k. But in the presence of interactions such states may be excited. m3 = (−1)n0 δn0 m0 δn1 m1 δn2 m2 δn3 m3 . 0). 0)]|0 (9. 0)|0 = [H. Thus we see that if we exchange the roles of the creation and the annihilation operators for the time-like photon. physical results are well behaved. n3 |m0 . m1 . 3. (9. 0)]|0 = d3 k′ ω(k′ )[−a(k′ . where we have used the fact that H|0 = 0. m2 . the energy of the state with one time-like photon becomes negative.

165) implies that . λ) = 0 for λ = 1. |k| (9. (9.129) or (9.130)) ∂µ Aµ (x) = 0. 3)a(k.8 Covariant quantization of the Maxwell theory 373 kinds of photons are not present in the physical state. (9. as we have seen.9. the supplementary condition (9.74) as well as the fact that for positive energy photons k0 = ω = |k|. or. it can be decomposed into positive and negative frequency parts uniquely in a relativistically invariant manner and this decomposition is preserved under time evolution. but it also requires that those photons cannot be emitted. λ or. λ)a(k. Gupta and Bleuler weakened the supplementary condition to have the form ∂µ Aµ (+) (x)|phys = 0.164) the Gupta-Bleuler supplementary condition leads to kµ ǫµ (k. where we have used the deﬁnitions and the properties of the polarization vectors in (9. Therefore. kµ ǫµ (k. Recalling that kµ ǫµ (k.165) (a(k.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. 3))|phys = 0. λ = 0) − k2 a(k. (This is commonly known as the Gupta-Bleuler quantization. λ = 3) |phys = 0. 0) + ǫµ (k.163) ∂µ Aµ (x) is like a free scalar ﬁeld. λ)|phys = 0. 3) |phys = 0. 2. 0)a(k. Let Vphys = {P } denote the set of all states which satisfy the supplementary condition and let |ψ represent such a state. or. (9. Then.) We remark here that since (see (9. k0 a(k. 0) − a(k.

n1 . n2 . (9.n1 . 3)a(k. or. n2 .n3 in the expansion can be determined recursively from (9. n2 . 0)|ψ = a(k.n1 .n2 .374 9 Maxwell field theory a(k.168).n1 .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 .n1 .n2 . n1 .n3 = 0. n1 .n3 |n0 . n2 . n3 − 1 = 0. 0) − a(k. A general physical state is.n1 .0 = 1. ψ|a† (k.170) . All the coeﬃcients Cn0 .169) which leads to the relation in (9. (9. n3 .n1 .n3 (a(k. 0)a(k.n1 .168) This can be seen as follows or. (9.n3 −1 + n3 Cn0 −1. Cn0 .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 . n3 = 0. 3)|ψ .n1 .n1 . n1 .n2 .n3 −1 + n3 Cn0 −1.n2 . a superposition of diﬀerent states of the form |ψ = n0 .n2 .n3 − n0 |n0 − 1.168) with the condition C0. 3)) |n0 .n3 ) ×|n0 − 1.n2 . √ Cn0 . n1 .n2 . or. (9.n2 . of course. √ √ ( n0 Cn0 . 3)|ψ . 0)|ψ = ψ|a† (k.n3 Cn0 . n3 √ − n3 |n0 .n2 . n1 . (9.n2 . n3 − 1 = 0.

0 = 1 − 1 = 0. n1 . 0|2. n1 . n2 . n1 . n2 .175) . 1 − |1. (9. 1 + |2. . 0 = 1 − 2 + 1 = 0.n3 ’s determined recursively.n1 . n1 . n1 . = |0. n1 . In other words. . n2 . n1 . 1 + 2. n1 . 1|1. n2 . 0 . n1 .171) and so on where we have used the values of Cn0 . For example. n2 . 1 + 1. n2 .172) and so on. Thus although we can write a general physical state with a ﬁxed number n1 . n2 . n2 . 1|0. 2|0.173) (9. n0 = n3 = 0. = φ0 |φ0 . n1 . = 0. . n2 . n1 .8 Covariant quantization of the Maxwell theory 375 It follows now that for a ﬁxed number n1 . n2 . n2 . √ = |0. 2 − 2|1. the physical states allowed by the supplementary condition have the forms |φ0 |φ1 |φ3 = |0. 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. n2 .9. (9. (9. (9. n2 of transverse photons as a linear superposition of states of the form |φ because φ|φ = |φ0 + b1 |φ1 + b2 |φ2 + . n1 . n1 . n2 .n2 . Let us note that because of the negative norm of states containing an odd number of time-like photons. n2 of transverse photons. 0 . n2 . n1 . n1 . we have φ1 |φ1 φ2 |φ2 = 0. 0|1. n2 . n1 . n2 . all such states except for |φ0 have zero norm. 0 .

Vphys . + iǫ (9. The physical S-matrix elements. namely.µν (k) = lim − ǫ→0+ k2 iηµν . Since the zero norm states are orthogonal to all the states including themselves. If interactions are present time-like and longitudinal states may occur as intermediate states. if we further mod out the states by the zero norm states.177) (9. we have the true physical subspace of the theory where the norm of states is positive deﬁnite. However their eﬀect cancels out in the ﬁnal result. such states decouple.179) This is diﬀerent from (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. consequently.376 9 Maxwell field theory in such states. A diﬀerent way of saying this is to note that being orthogonal to every other state.77) or (9. are gauge .176) where |ψ represents a physical state. on the space of physical states since ψ|∂µ Aµ |ψ = 0.89) 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). 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. (9. V0 V phys = (9. The physical subspace of the theory selected by the supplementary condition (9.178) where V0 represents the set of states with zero norm. Without going into details we note here that the Feynman propagator for the photon in this theory has the simpler form iGF .165) contains states with positive semi-deﬁnite norm (negative norm states are eliminated by the supplementary condition or the physical state condition and. however.

9 References 377 independent and do not depend on the form of the photon propagator in a particular gauge. Drell. J. Relativistic Quantum Fields. . John Wiley. J. 7. Physical Review 78. 681 (1950). Bleuler. 371 (1957). Zuber. New York (1969). 3. Y. 2. Helvetica Physica Acta 23.9. 1980. Shirkov. 8. Evanston (1961). Quantum Field Theory. Proceedings of the Physical Society (London) A63. Row. Introduction to the theory of Quantized Fields. Schweber. McGrawHill. Itzykson and J-B. 6. N. S. K. S. C. New York. 5. We will develop these ideas further when we study the covariant quantization of non-Abelian gauge theories in chapter 13. Peterson. F. Introduction to Relativistic Quantum Field Theory. John Wliley. Bjorken and S. New York (1993). Nuovo Cimento 6. 10. Gross. Nauka. N. C.9 References 1. Introduction to Quantum Field Theory. McGrawHill. V. D. Moscow (1984). 182 (1950). Takahashi. 4. 9. Ward. P. 567 (1950). N. 9. Gupta. Bogoliubov and D. Roman. New York. 1964. Relativistic Quantum Mechanics and Field Theory.

This page intentionally left blank .

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.Chapter 10 Dirac method for constrained systems 10. i = 1. · · · . carry out the quantization of a theory) and in this chapter we will discuss the Dirac method in some detail. To appreciate the diﬃculties associated with constrained systems. ∂ qi ˙ (10. we can go from the conﬁguration space of the system to the phase space 379 . thereby. 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. let us consider a classical system of point particles described by the Lagrangian L(qi . In this case. the dynamical phase space variables of these theories are not all independent. N . velocities can be expressed in terms of momenta (and coordinates). we deﬁne the momenta canonically conjugate to the coordinates as pi = ∂L . qi ). In this case.1) Equation (10. relates momenta to velocities (and coordinates) of the theory and if this relation is invertible. Given the Lagrangian of the theory.1). rather some of these variables have to satisfy constraints following from the structure of the theory.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. 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.

˙ ˙ (10. ∂qi ∂pi ∂pi ∂qi {A1 .3) where summation over repeated indices is understood. pi ) in the phase space can be written in the Hamiltonian form . In fact.2) and uniquely deﬁne the Hamiltonian of the system through the Legendre transformation H(qi .5). (10. ˙ (10. ∂pi ∂H . j {qi . qi ) → (qi . pi ) as ∂A1 ∂A2 ∂A1 ∂A2 − . pi ). pj } = δi = −{pj .5) Using (10. A2 (qi .4) or (10. pi ) = pi qi − L(qi . ∂qi (10. it follows now that the 2N ﬁrst order dynamical equations (evolution equations) for the system can be written in the Hamiltonian form ∂H .6) qi = {qi . using the deﬁnition (10. qi ).3). qj } = 0 = {pi . pj }. H} = − ˙ which can be shown. the time evolution of any dynamical variable A(qi .380 10 Dirac method for constrained systems (qi . H} = ˙ pi = {pi . A2 } = (10. The phase space spanned by the independent coordinates and momenta is also 2N dimensional with a (equal time) canonical Poisson bracket structure {qi .4) which allows us to write the Poisson bracket between any two dynamical variables A1 (qi . to be equivalent to the N second order Euler-Lagrange equations following from the Lagrangian description. pi ). qi }.

α = 1.9) can be inverted to express velocities in terms of the momenta. if ˙ we have a classical system described by the Lagrangian L(θα .9) However. θα ).12).9) and as discussed in chapter 8 (see (8. pi ). unlike the bosonic case in (10. θα }.8) then. θα ) = −Πα θα − L(θα . β {θα . (10. we can deﬁne the momenta conjugate to the coordinates as Πα = ∂L .1 Constrained systems 381 ˙ A(qi . (10.5).7) This discussion can also be generalized to a classical system described by Grassmann (anti-commuting) variables. ˙ ∂ θα (10. θα ). (10. (10. For example. (10.10) through the Legendre transformation (this is consistent with the convention of left derivatives) ˙ ˙ ˙ ˙ H(θα . If the relations (10. we will choose the convention of taking derivatives from the left.10. Πα ) = θα Πα − L(θα . since the dynamical variables are now of Grassmann odd (fermionic) nature.14)). we can go from the conﬁguration space to the phase space ˙ (θα . · · · M with θα θβ = −θβ θα . H}. θβ } = 0 = {Πα .11) The 2M dimensional phase space is endowed with a canonical Poisson bracket structure (with the convention of left derivatives) {θα . Πβ }. 2. θα ) → (θα . we have to deﬁne the derivative in (10.12) The Poisson bracket between any two dynamical variables can now be obtained from (10. Πα ). However. . Πβ } = −δα = {Πβ . pi ) = {A(qi .

B} = − {B. Thus. H} = − α . In a similar manner. with the convention of left derivatives. . Πα = {Πα .14) the 2M ﬁrst order dynamical equations for the system can now be written in the Hamiltonian form as ∂H ˙ θα = {θα . B2 } = − = ∂B ∂B where we have used (10. B2 (θα . (10.10) is invertible so that all the velocities can be expressed uniquely in terms of independent momenta leading to the . for example. (10. {F1 .14) Using (10.13) {B1 .12) as well as the fact that ∂θα and ∂Πα are Grassmann odd (fermionic) variables. Πα ). H} = − ∂θα (10. 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. Πα ) takes the form ∂B2 ∂B1 α ∂B2 ∂B1 {θα . F2 } = − ∂F1 ∂F2 ∂F1 ∂F2 + α ∂θα ∂Π ∂Πα ∂θα . F } = ∂F ∂B ∂F ∂B + α ∂θα ∂Π ∂Πα ∂θα ∂B ∂F ∂B ∂F + ∂θα ∂Πα ∂Πα ∂θα .6).2) or (10.382 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). Πβ } β − {Π . the Poisson bracket between two bosonic dynamical variables B1 (θα .15) which can be compared with (10. θβ } α ∂θα ∂Π ∂Π ∂θβ ∂B1 ∂B2 ∂B1 ∂B2 + ∂θα ∂Πα ∂Πα ∂θα . ∂Π ∂H ˙ . The discussion so far assumes that the transformation to phase space in (10.

˜ denote N × N matrices.18) so that A−1 exists only if the matrix b is invertible. For conventional b theories described by Lagrangians with quadratic velocity terms. ∂ qj ˙ ∂ qj ∂ qi ˙ ˙ (10.20) Therefore. In this case.19) bij (q) = (10. ∂ qi ∂ qj ˙ ˙ det (10.1 Constrained systems 383 unique Hamiltonian of the system. (10. ˜=˜ b b(q).16) where the elements b. the transformation to phase space in (10.10. (10. it follows that b = b(q). However. it follows that the transformation is invertible (A−1 exists) if ∂2L = 0. let us concentrate on the bosonic theory for simplicity. b ˙ so that ∂pi ∂2L = . the diﬃculty in passage to a Hamiltonian description arises when this transformation is noninvertible.16) can be easily seen to have the form A−1 = ½ 0 −b−1˜ b−1 b . To understand the diﬃculty. (10. From (10.16) we note that pi = ˜ij (q)qj + bij (q)qj .21) .17) The inverse of the matrix A in (10.2) can be written in general in the matrix form as q p q q ˙ =A = ½ 0 ˜ b b q q ˙ .

it is not clear a priori how to deﬁne the Hamiltonian uniquely. It follows from (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. ∂ qi ∂ qj ˙ ˙ (10.4) need not satisfy the constraints of the theory and. there exist relations or constraints between various dynamical variables and such systems are known as constrained systems. In this case. R. it is also clear that the naive canonical Poisson brackets in (10. pb ). we can write .23) of the physical system is of rank R < N . In other words. the transformation (10. the passage to a Hamiltonian description is achieved systematically through the method of Dirac which we describe in this section. 10. we can determine only R of the N velocities in terms of coordinates and independent momenta as qa = fa (qi .22) then. if the Lagrangian describing the theory satisﬁes ∂2L = 0. b = 1. As a result. ∂ qi ∂ qj ˙ ˙ det (10. On the other hand.24) and the other N − R velocities cannot be determined. ˙ a.2) is not invertible. therefore. · · · . may not represent the true brackets necessary for a Hamiltonian description of the system. Let us assume that the N × N matrix ∂2L . therefore.2 Dirac method and Dirac bracket In a physical system with constraints. (10.384 10 Dirac method for constrained systems The naive passage to the Hamiltonian description can be carried out only in this case. In this case. 2. Furthermore.

R + 2. the Hamiltonian of the theory is no longer unique. pa ).28) that in spite of the fact that qα cannot be determined.25) It is clear that the second set of equations in (10. qi ) = pa qa + gα qα − L(qi .29) so that on the constrained hypersurface Γc .10. · · · . pa ).26) and we recognize that such constraints will reduce the dimensionality of the true phase space of the system. if (F − G) = 0. (10.28) where we have used (10. a.25) deﬁne N − R constraints among dynamical variables which we denote by ϕα = pα − gα (qi .2 Dirac method and Dirac bracket 385 pa = ga (qi . (10. G in Γ are said to be weakly equal. the dimensionality of Γc would be further reduced. b = 1. the canonical Hamiltonian does not depend on velocities as we would expect. 2. (10. On the other hand. We note from (10. N.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. ˙ ˙ ˙ ˙ ˙ (10. ˙ ∂L ∂Hcan = gα − = gα − pα ≈ 0. In fact. ∂ qα ˙ ∂ qα ˙ (10. qb ).27) Γc As usual. namely. α = R + 1. namely. we note that the constraints in (10.2) (presence of constraints). we can deﬁne the canonical Hamiltonian of the theory as the Legendre transform Hcan = pi qi − L(qi . from the canonical . qi ). Hcan = Hcan (qi .) Two dynamical variables F. (If there are further constraints in the theory.25). if they are equal on the constrained hypersurface Γc . R. F ≈ G. we also note that because of the non-invertibility of (10. · · · . pa ) = 0. ˙ pα = gα (qi . For example.

32) qi ≈ {qi .30) where λα denote undetermined Lagrange multipliers and it follows that Hp ≈ Hcan .26)) Hp = Hcan + λα ϕα . Hp } = − ˙ and using (10.35) . ˙ (10. Hp } = {F. (10. it follows that we can identify the Lagrange multipliers in (10. λβ }ϕβ ≈ {ϕα . (10. The time evolution of any dynamical variable F can now be written as ˙ F ≈ {F.29). (10. (10. ∂pi ∂pi ∂Hp ∂(Hcan + λα ϕα ) =− . Hp } = ˙ pi ≈ {pi . ϕβ } + {ϕα . ∂qi ∂qi (10.33) which remain undetermined. Hcan } + λβ {ϕα .31) Furthermore.34) this leads to ϕα ≈ {ϕα .386 10 Dirac method for constrained systems Hamiltonian of the theory. Hcan + λα ϕα }. ϕβ } ≈ 0.26) as (it is known as the primary Hamiltonian since it incorporates only the primary constraints in (10.34) The constraints of the theory should be invariant under time evolution and using (10. we can deﬁne the primary Hamiltonian associated with the system by incorporating the primary constraints of the theory (10.30) with velocities λα ≈ qα . with the canonical Poisson brackets in (10. Hcan + λβ ϕβ } ˙ = {ϕα .4). Hcan } + λβ {ϕα . we can write the Hamiltonian equations as ∂Hp ∂(Hcan + λα ϕα ) = .

35) (with Hp as the Hamiltonian) may determine some other Lagrange multipliers or generate newer constraints (tertiary etc). ) collectively as ϕα ≈ 0. · · · .2 Dirac method and Dirac bracket 387 where we have used (10. α = 1.37) 2. n1 . (10. .35) may determine some of the Lagrange multipliers or may lead to new constraints known as secondary constraints. On the other hand. n < 2N. It is clear that (10.36)-(10.38) that n1 + 2n2 = n < 2N . It is also worth noting that in this process of determining all the constraints that are time independent (evolution free). We continue with this process until all the constraints are determined to be evolution free. Let us denote all the constraints of the theory (primary. 2. all of the Lagrange multipliers may be completely determined or some of them may remain undetermined. tertiary. In general. α = 1. 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. The constraints which have weakly vanishing (see (10. · · · . . Requiring the secondary constraints to be invariant under time evolution as in (10. . 2. (10. · · · . then there remain undetermined Lagrange multipliers in the primary Hamiltonian after all the .36) which can be divided into two distinct classes. 1.38) 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. 2.27)) Poisson bracket with all the constraints (including themselves) are known as ﬁrst class constraints and are denoted as e ψ α ≈ 0. if there are primary constraints in the theory which are ﬁrst class. 2n2 .26) in the last step. (10. α = 1. secondary.10.

vanish {F. (10.40) so that the true phase space of the theory is 2(N − n1 − n2 ) dimensional. First. (The number of undetermined Lagrange multipliers equals the number of ﬁrst class constraints among the primary constraints. A = 1. 2. Namely.37) into second class constraints. let us note that we can construct the matrix of Poisson brackets of all the constraints in (10. α = 1. which are all second class.42) This issue of incompatibility of the Poisson brackets can be addressed through the use of the Dirac brackets which are constructed as follows. even though φA ≈ 0. (10. (10. φA } ≈ 0. · · · . n1 .) The ﬁrst class constraints have a special signiﬁcance in that they are associated with gauge invariances (local invariances) in the theory. 2. a consistent Hamiltonian description. after gauge ﬁxing we denote all the constraints of the theory (including the gauge ﬁxing conditions). collectively as φA ≈ 0. / (10. in general. · · · . in this case. we cannot yet impose them directly in the theory since the canonical Poisson brackets in (10.40) as . Although we have identiﬁed all the constraints of the theory.4) are not compatible with them. Thus. The gauge ﬁxing conditions are generally denoted as e χα ≈ 0. 2(n1 + n2 ) < 2N.388 10 Dirac method for constrained systems constraints have been determined.41) the Poisson bracket of the constraints with any dynamical variable F does not. requires gauge ﬁxing conditions and we must choose as many gauge ﬁxing conditions as there are ﬁrst class contraints. As we have seen in the case of Maxwell ﬁeld theory.39) and are chosen such that they convert the ﬁrst class constraints (10.

(10. φA } −1 ≈ {F. In fact. in the deﬁnitions of the Hamiltonian. φA } − {F. φA }CAB {φB . φA } − {F. φB }CBD C DA = {F.44) We note here parenthetically that in ﬁeld theories (where the variables depend on space coordinates as well). φA } − {F. (10.40). G}.46) Namely. A −1 −1 C AD CDB = δB = CBD C DA . (10. As a result. (10. 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. φB }CBD {φD . the matrix of Poisson brackets in (10. in contrast to the canonical Poisson brackets. G} − {F. Jacobi identity) and in addition satisﬁes −1 {F.10. φA } = 0. G}D = {F. 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. With this. through the use of Lagrange brackets.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. φA }D = {F.43) This is an even dimensional anti-symmetric matrix and Dirac had −1 shown that it is nonsingular so that its inverse CAB exists. we can show that the Dirac brackets are indeed the correct Poisson brackets of the theory subject to the constraints in (10. we can now deﬁne the Dirac bracket between any two dynamical variables as −1 {F.43) will be coordinate dependent and the product in (10.45) which can be shown to have all the properties of a Poisson bracket (anti-symmetry. φB } ≈ C AB . The Dirac bracket has the interesting property . namely.44) will involve integration over the intermediate space coordinate.

F into qa = (qi . 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. · · · . 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. If we denote the coordinate of the particle as qi . then the coordinates are constrained to satisfy (the metric in this simple theory is Euclidean and.47) where summation over repeated indices is understood. n + 1. Namely.390 10 Dirac method for constrained systems that it can be constructed in stages. (10.3 Particle moving on a sphere As a simple example of constrained systems. It is also worth emphasizing here that in dealing with ﬁeld theories where the matrix C AB is coordinate dependent. · · · . rather than dealing with a large matrix of Poisson brackets in (10. 10. a = 1.49) gives the constraint (10. 2. therefore.47) on the dynamics. We note that the dynamics of the system can be described by the Lagrangian 1 (qi qi − F (qi qi − 1)) . We note that if we combine the dynamical variables qi . i = 1.43) and its inverse. the inverse needs to be deﬁned carefully with the appropriate boundary condition relevant for the problem. when the number of constraints is large. let us consider the motion of a point particle constrained to move on the surface of a ndimensional sphere of radius unity. n. ∂F 2 (10. raising and lowering of indices can be carried out trivially) qi qi = 1.48) where F is a Lagrange multiplier ﬁeld (an auxiliary ﬁeld without independent dynamics) whose Euler-Lagrange equation 1 ∂L = − (qi qi − 1) = 0. then the matrix of highest derivatives has the form . 2. ˙ ˙ 2 L= (10. F ).

(10.51) Therefore. pj } = δi .48) and (10. pF } = 0. ˙ ∂F pi = pF = (10. 2 (10.53) Hp = Hcan + λ1 ϕ1 = (10. F } = {pF . To understand the passage to the Hamiltonian description in this simple case. pF } = {pi .52) The canonical Hamiltonian following from (10. pF } = 0. (10. qj } = {pi . {qi .51) has the form 1 i i ˙ Hcan = pi qi + pF F − L = ˙ p p + F (qi qi − 1) . (10.55) . {F. {qi .50) which is indeed singular (see (10. F } = {pi .54) The equal-time canonical Poisson brackets of the theory are given by j {qi . pj } = {F. ˙ ∂ qi ˙ ∂L = 0.22)) reﬂecting the fact that there are constraints in the theory and the naive Hamiltonian description would not hold.30)) 1 i i p p + F (qi qi − 1) + λ1 pF .10. the theory has a primary constraint given by ϕ1 = pF ≈ 0. F } = {qi . we note that the conjugate momenta of the theory are given by ∂L = qi . pF } = 1.3 Particle moving on a sphere 391 ∂2L = ∂ qa ∂ qb ˙ ˙ δij 0 0 0 . 2 leading to the primary Hamiltonian (see (10.

60) . pj } + qi F qj {pi .59) to be time independent. Hp } ˙ 1 = {qi pi . Hp } ˙ 1 1 = { (qi qi − 1).57) Requiring (10. qj } = pi pi − F qi qi (10. = pi pj {qi . requiring the primary constraint to be independent of time leads to ϕ1 ≈ {ϕ1 . we obtain ϕ2 ≈ {ϕ2 . 2 (10.392 10 Dirac method for constrained systems With these we can now determine the time evolution of any dynamical variable. pj } = qi pi ≈ 0. (10. 2 (10.56) Let us denote this secondary constraint as ϕ2 = 1 (qi qi − 1) ≈ 0. Hp } ˙ 1 = {pF .58) which generates a new constraint ϕ3 = qi pi ≈ 0. (pj pj + F (qj qj − 1) + λ1 pF )} 2 ≈ pi pi − F ≈ 0.59) Requiring the new constraint (10. (10.57) to be time independent. (pj pj + F (qj qj − 1) + λ1 pF )} 2 2 = qi pj {qi . we obtain ϕ3 ≈ {ϕ3 . (pi pi + F (qi qi − 1) + λ1 pF } 2 1 = − (qi qi − 1) ≈ 0. In particular.

It is easy to check that all the constraints of the theory (10.3 Particle moving on a sphere 393 so that we can identify ϕ4 = pi pi − F ≈ 0. ϕ3 }. ϕ2 . 4.61) 1 = {pi pi − F. φB } = C AB . (pj pj + F (qj qj − 1) + λ1 pF )} 2 (10. 3.64) = −{ϕ3 . (10. Collecting all the constraints into φA = (ϕ1 . pj } = 2pi pi = 2p2 = −{ϕ4 .52). ϕ4 ). (10. we can calculate the matrix of the (equal-time) Poisson brackets of constraints {φA .10.59) and (10. pj } = qi qi ≈ 1 2 {ϕ3 .61) to be time independent leads to ϕ4 ≈ {ϕ4 . Hp } ˙ (10. ϕ4 } = {pF .57). which determines the Lagrange multiplier and the chain of constraints terminates. which takes the form (10. ϕ2 }. (10. the nontrivial equal-time Poisson brackets between the constraints take the forms {ϕ1 . ϕ1 }. Requiring (10. ϕ3 } = { (qi qi − 1). = −{ϕ4 . ϕ3 . pi pi − F } = 1 1 {ϕ2 . In fact. ϕ4 } = {qi pi . A = 1.61) are second class. qj } − λ1 {F. qj pj } = qi qj {qi .62) = 2pi F qj {pi . . pj pj − F } = 2pi pj {qi .63) where we have identiﬁed pi pi = p2 . 2. pF } = −2F qi pi − λ1 ≈ −λ1 ≈ 0.

qk pk }C32 { (qℓ qℓ − 1). {F. 1 −1 {pi .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 . In particular. pj } 2 −1 1 {qi . pF }D = {pi . F } = 2pi . 2 {qi . (10. = 0 −1 0 2p2 2 −1 0 −2p 0 0 0 0 1 (10. F }D = {pi . pj pj − F }C41 {pF . F }D = {qi . −1 {qi . F }D = {pF . = −qi pj + pi qj . pj }D = {pi .45)) −1 {F. (qj qj − 1)}C21 {pF . G}. G} − {F. F } = −2qi p2 . F } − {pi . pj }D = {qi .67) −1 with CAB given in (10. F } − {qi . pF }D = 0.68) .66). qk pk }C32 { (qℓ qℓ − 1). qj }D = 0.66) (10. pj } 2 −1 1 −{pi .65) −1 CAB The Dirac brackets between any two dynamical variables can now be deﬁned as (see (10. {qi . 2 2p = 0 1 0 . pj } − {qi . G}D = {F. pF }D = {F. 0 0 (10. (qk qk − 1)}C23 {qℓ pℓ . φA }CAB {φB . pj } 2 j = δi − qi qj . pF }D = 0. pj } − {pi . the Dirac brackets between the fundamental dynamical variables take the forms 1 −1 {pi .

54)) 1 i i pp. we can set the constraints to zero so that the true independent dynamical variables are (qi .69) where we have used (10. Let us next study the Hamiltonian description of a free relativistic massive particle which is described by the equation d2 xµ = 0.70) 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.61)) and the Hamiltonian for the theory is given by (see (10. The constraints of this theory were all second class.4 Relativistic particle In the last section we studied a simple example of point particle dynamics which is constrained.4 Relativistic particle 395 With the use of the Dirac brackets.70) can also be written as dpµ = 0. We note that if we deﬁne the relativistic four velocity and momentum as uµ = dxµ .72) The dynamical equation (10.62). This is the starting point for the quantization of this theory. dτ pµ = muµ .73) where λ is a parameter labelling the trajectory and we have identiﬁed . dτ 2 m (10. 10.70) where τ denoting the proper time is assumed to label the trajectory of the particle and m is the rest mass of the particle. pi ) (we note that F = pi pi because of the constraint (10.71) then the dynamical equation (10. 2 Hp = (10.10. dτ (10. (10. (10.

(10. (10. (10.73) that ∂2L = ∂ xµ xν ˙ ˙ m xρ xρ ˙ ˙ xµ xν ˙ ˙ xσ xσ ˙ ˙ Pµν = ηµν − . However. If we choose a gauge λ = τ.76) Namely.77) where τ denotes the proper time associated with the particle. then (10. dλ xµ = ηµν xν .70) or (10. the action (10. We note from (10.396 10 Dirac method for constrained systems xµ = ˙ dxµ .75) which is an example of a local gauge transformation much like in the Maxwell ﬁeld theory. dλ dλ dξ λ → ξ = ξ(λ).72) as the Euler-Lagrange equations.74) We note that under a transformation xµ = ˙ dξ dxµ dxµ = . so that (10.73) will lead to the dynamical equation (10.78) which leads to .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.73) is invariant under the diﬀeomorphism (10. let us proceed with the Hamiltonian description without choosing a gauge at this point. ˙ ˙ (10.

73) in the form 1 2 √ ˙ ˙ dλ g g−1 xµ xµ + m2 .81) is manifestly diﬀeomorphism invariant and the massless limit can now be taken in a straightforward manner. ∂ xµ xν ˙ ˙ det Pµν = det (10. we can rewrite the action (10. the action (10. However. For example. S= dλ L = (10. m2 (10. We note that introducing an auxiliary variable g.81) are xµ and g. then we obtain (10.81).73).124)) and the naive passage to the Hamiltonian description fails (see (10.79) Therefore. Let us note that the action (10.4 Relativistic particle 397 xµ Pµν = ˙ m xρ xρ ˙ ˙ xν − ˙ xµ xµ xν ˙ ˙ ˙ xσ xσ ˙ ˙ = 0 = Pµν xν . Viewed in this manner.80) as in the case of the Maxwell theory (see (9. To see that the two actions (10. The two dynamical variables of the theory in (10.73) can also be written in an alternate form that is more useful.73) we can think of g as the metric on the one dimensional manifold of the trajectory of the particle.73) and (10.82). it is not clear how to take the m = 0 limit in (10. ˙ (10.22)). g= xµ xµ ˙ ˙ . We can obtain the momenta conjugate to these variables to be .81) where because of the diﬀeomorphism invariance of (10.82) If we eliminate g in (10. we note that the equation for the auxiliary ﬁeld g takes the form g−1/2 ∂L = −g−1 xµ xµ + m2 = 0. it is more convenient for us to work with the action (10.10. Pµν is a projection operator and ∂2L = 0.81) are equivalent.81) using (10.73) to describe the motion of a massless particle. ˙ ˙ ∂g 4 or.

84).85) Adding the primary constraint (10.84) Hcan (10. g} = {xµ . g} = {pµ . {g. The canonical Hamiltonian of the theory follows to be √ g µ = pµ x + pg g − L = ˙ ˙ p p µ − m2 . pg } = {pµ . ˙ ∂ xµ ˙ pg = ∂L = 0. Hp } = {pg .84) and requiring the constraint to be time independent. we obtain √ g µ p pµ − m2 + λ1 pg } 2 (10. pν } = {g.398 10 Dirac method for constrained systems pµ = ∂L = g−1/2 xµ .88) ϕ ˙ 1 ≈ {ϕ . pν } = δν . pg } = 0. 2 µ (10. xν } = {pµ . pg } = 0.86) The equal time canonical Poisson brackets for the theory are given by {xµ . g} = {pg . (10.83) so that the primary constraint of the theory follows to be ϕ1 = pg ≈ 0. we obtain the primary Hamiltonian for the theory to correspond to √ g µ p p µ − m 2 + λ1 p g . µ {xµ . 1 1 = − √ pµ pµ − m2 ≈ 0. 4 g which leads to the secondary constraint .87) With these we can now determine the evolution of the primary constraint (10. pg } = 1. Hp = Hcan + λ1 ϕ = 2 1 (10. {xµ . ∂g ˙ (10.

92) that the matrix of Poisson brackets of constraints has the form .91) χ1 = g − χ2 = λ − x0 ≈ 0. m2 (10.92) = −{χ2 . ϕ2 . The non-vanishing Poisson brackets between the constraints are given by 1 } = −1 = −{χ1 . χ1 } = {pg .89) is the familiar Einstein relation for a massive relativistic particle (the multiplicative factor is for later calculational simplicity). χ2 ). χ2 } = { 1 µ p pµ − m2 . (10. ϕ1 }. 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). 2 (10. which render all the constraints to be second class. m2 {ϕ1 . Let us choose the gauge ﬁxing conditions 1 ≈ 0. g − {ϕ2 . χ1 .4 Relativistic particle 399 ϕ2 = 1 µ p pµ − m2 ≈ 0.89) We recognize that the secondary constraint (10. λ − x0 } = p0 2 (10. ϕ2 }. Time evolution of the secondary constraint leads to √ g ν 1 µ 2 p p ν − m 2 + λ1 p g } ≈ {ϕ . As a result. if we combine all the constraints into φA = (ϕ1 .90) 2 ϕ ˙ 2 so that it is time independent and the chain of constraints terminates.10. then it follows from (10. 2 2 = 0. Hp } = { p pµ − m .

G}. −1 {g. φA }CAB {φB . pν }D = {g. pg } {xµ . 0 0 0 (10. pg }{g − 1 . φ1 }C13 {φ3 .95) and with (10. µ = δν − 1 λ 1 p pλ − m2 } − 0 {λ − x0 . = {g. g}D = {pµ . G}D = {F. pν } 2 p (10. pg } = 0. 0 = −1 0 1 0 − p0 . pg } − {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 }D = 0.94) (10. pν }D = {xµ .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 . pg }D = 0. φ2 }C24 {φ4 .94) we can calculate the Dirac bracket between the fundamental variables to be {xµ . 0 0 0 0 (10. (On . G} − {F. m2 −1 {xµ . pν } µ = δν − {xµ .96) pµ 0 δ . pν } − {xµ . g}D = {pg .93) 1 0 −1 CAB The equal time Dirac bracket between any two dynamical variables is given by (see (10. g}D = {xµ . xν }D = {pµ . pg } − {g.86) vanishes.45)) −1 {F. pg }D = {pµ . pg }D = {g.

89) in particular leads to H = p0 = p2 + m2 which can be taken as the Hamiltonian. if we set the constraints strongly to zero. ψα .100) † The dynamical variables of the theory can be chosen to be ψα . 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 401 the other hand.22)).91) would imply dxµ dxµ = 1. dλ dλ xµ xµ = ˙ ˙ (10. 10.) Let us also note from (10. 3. or. 2.9).99) (10.97) On the other hand. where the adjoint spinor is given by ψ = ψ† γ 0 .20)). from the deﬁnition of the inﬁnitesimal invariant length in the Minkowski space (see (1. where α = 1. we have (remember that c = 1) dτ 2 = dxµ dxµ .10. therefore. (10. (10. This manifests in the deﬁnition of the conjugate momenta as (we have chosen the convention of left derivatives) .5 Dirac ﬁeld theory As we have seen in (8. dxµ dxµ = 1. the free Dirac theory is described by the Lagrangian density / L = iψ∂ ψ − mψψ.98) It follows. 4 denote the Dirac indices and since the Lagrangian density is ﬁrst order in derivatives. dτ dτ (10. that the ﬁrst of the gauge ﬁxing conditions corresponds to choosing λ = τ .

The canonical Hamiltonian is now obtained to be Hcan = d3 x Hcan = d3 x −iψγ · ∇ψ + mψψ . (10. ˙ ∂ ψα ∂L = 0.101) (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. The canonical Hamiltonian density of the theory (consistent with the convention of left derivatives) is obtained to be (see (10.11)) ˙ ˙† Hcan = −Π† ψα + ψα Πα − L α = −iψγ · ∇ψ + mψψ.) As a result. ˙† ∂ ψα (10.103) where we have used (10.30)) Hp = Hcan + d3 x φ† ξα + λ† ρα . we obtain the primary Hamiltonian for the Dirac ﬁeld theory as (see (10. α α (10.104) Adding the primary constraints to the canonical Hamiltonian. † ˙ † ˙ = iψα ψα − iψα ψα − iψγ · ∇ψ + mψψ (10.101) in the intermediate steps.102) which in this case correspond to fermionic constraints. α α ρα = Πα ≈ 0.105) .402 10 Dirac method for constrained systems Π† = α Πα = ∂L † = −i(ψγ 0 )α = −iψα . Rather they correspond to the momenta conjugate to ψ and ψ † respectively.

φ† (y)} α β −λ† (y){φ† (x).107) α β Let us evaluate each of these terms separately. ψβ (y)} α d3 y iψ(y)γ · ∇y − mψ(y) i∇ψ(x) · γ + mψ(x) α β − δαβ δ3 (x − y) . −iψ(y)γ · ∇y ψ(y) + mψ(y)ψ(y)} α d3 y iψ(y)γ · ∇y − mψ(y) β {Π† (x). Using the fact that the primary Hamiltonian is independent of time. ρβ (y)} .13)(10. Hcan } α = = = = † d3 y {Π† (x) + iψα (x).106) with all other brackets vanishing. 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). Using (10. ˙ φ† (x) ≈ {φ† (x). Hp } α α ≈ {φ† (x). λ† in the present case denote α fermionic variables.14)). For example.12)) † {ψα (x). φ† (y)} α β = † † d3 y ξβ (y){Π† (x) + iψα (x).10. Π† (y) + iψβ (y)} α β . The equal-time canonical Poisson brackets for the ﬁeld variables take the forms (see (10.5 Dirac field theory 403 where the Lagrange multipliers ξα . Πβ (y)}. d3 y ξβ (y){φ† (x). (10. Π† (y)} = −δαβ δ3 (x − y) = {ψα (x). Hcan } + α d3 y ξβ (y){φ† (x). we can now calculate the time evolution of the primary constraints (10. β (10.102).106) (see also (10.

109) which determines the Lagrange multiplier λ† = − ∇ψ(x) · γ + imψ(x) α α . Hp } ˙ = = † d3 y {Πα (x).111) which determines the other Lagrange multiplier ξα = γ 0 (γ · ∇ + im)ψ(x) α . α α (10. α Substituting these results into (10.404 = 0. − 10 Dirac method for constrained systems d3 y λ† (y){φ† (x).112) . Hp } = {Πα (x).107) and requiring the constraint to be independent of time leads to ˙ φ† (x) ≈ i∇ψ(x) · γ + mψ(x) α + iλ† (x) = 0.110) Similarly. (10. (10.108) = iλ† (x). ρβ (y)} α β = − = − † d3 y λ† (y){Π† (x) + iψα (x). (10. requiring the second set of primary constraints to be time independent we obtain ρα (x) ≈ {ρα (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. Πβ (y)} α β d3 y λ† (y) − iδαβ δ3 (x − y) β (10.

Substituting (10. combining the constraints into ΦA = (φ† . Πβ (y)} = 0.115) Thus. deﬁne second class constraints (this is consistent with the fact that there is no undetermined Lagrange multiplier in the theory). we can write α the matrix of Poisson brackets as . (10.105).114) Let us note next that the two sets of primary constraints (10.110) and (10. (10. ρβ (y)} = {Πα (x).5 Dirac field theory 405 In this case. φ† (y)} = {Π† (x) + iψα (x). β {ρα (x).10. we have (at equal time) † † {φ† (x). (10. ρβ (y)} = {Π† (x) + iψα (x). we see that requiring the two primary constraints to be time independent does not introduce any new constraint. ρα ).102). φ† (y)}. α α β β † {φ† (x). we have deﬁned Π = Π† γ 0 . Π† (y) + iψβ (y)} = 0.112) into the primary Hamiltonian. rather it determines both the Lagrange multipliers in the primary Hamiltonian (10. 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 .113) where. which constitute all the constraints of the theory. for simplicity of notation. Πβ (y)} α α = −iδαβ δ3 (x − y) = {ρα (x). In fact.

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

we can set the constraints to zero in the theory. ψβ (y)}D = 0 = {ψα (x). Π† (z) + iψγ (z)}{Πγ (z). ψβ (y)}D . 4 µ. Maxwell’s equations can be obtained as the Euler-Lagrange equations from the Lagrangian density 1 L = − Fµν F µν . ψβ (y)} −i = −i † † d3 z{ψα (x). 3. (10. As we have seen in (9.18)).122) where the ﬁeld strength tensor (see (9. Therefore.4)) . ν = 0.16)). 1. Furthermore. ψβ (y)} γ d3 z(−δαγ δ3 (x − z))(−δγβ δ3 (z − y)) (10.6 Maxwell field theory 407 † † {ψα (x).6 Maxwell ﬁeld theory Let us next consider the Maxwell ﬁeld theory which as we know exhibits gauge invariance. we can show that † † {ψα (x).118) in evaluating this bracket because Πγ has a vanishing Poisson bracket with ψα .113) becomes Hp = d3 x −iψγ · ∇ψ + mψψ . 10. we recall that when using the Dirac brackets. (10.10. the Hamiltonian of the theory (10.121) which is the Hamiltonian we have used in the quantization of the Dirac ﬁeld theory (see (8. where we have neglected the last term in (10.16). ψβ (y)}D = {ψα (x). Similarly.119) = −iδαβ δ3 (x − y). (10.120) and these are the relations we have used in quantizing the Dirac theory (see (8. 2.

j. ˙ Π = E = −(A + ∇A0 ).128) We can now obtain the canonical Hamiltonian density of the theory to have the form . Fij = −ǫijk Bk .127) This determines that the theory has a primary constraint given by ϕ1 (x) = Π0 (x) ≈ 0. (10. (10. 2.408 10 Dirac method for constrained systems Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . 3.5) and (9. As a result of the gauge invariance of the theory.123) is the four dimensional curl of the four vector potential and contains the electric and the magnetic ﬁelds as its components (see (9. µ (10. k = 1. (10.7)) F0i = Ei . i. ˙ ∂ Aµ Πµ = (10.126) so that we have Π0 = 0. the matrix of quadratic derivatives becomes singular (see (9.124) We also know that Maxwell’s theory is invariant under gauge transformations Aµ (x) → A′ (x) = Aµ (x) + ∂µ α(x). (10.122)) and this manifests in the deﬁnition of the conjugate momenta as ∂L = −F 0µ .125) where α(x) denotes the local parameter of gauge transformation.

132) Using these we can calculate the time evolution of the primary constraint (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 . 2 (10. (10.127). ν {Aµ (x).10. Πν (y)} = δµ δ3 (x − y) = −{Πν (y). is given by Hcan = = d3 x Hcan d3 x 1 2 (Π + B2 ) − A0 ∇ · Π . Πν (y)}. 2 (10.128) Hp = Hcan + = d3 x d3 x λ1 ϕ1 1 2 (Π + B2 ) − A0 ∇ · Π + λ1 Π0 .131) where λ1 denotes the Lagrange multiplier of the theory. The canonical Hamiltonian.130) where we have neglected a surface term (total divergence) in integrating the last term by parts. Aν (y)} = 0 = {Πµ (x). = 2 (10.128) 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 ) . The equal-time canonical Poisson brackets of the theory are given by {Aµ (x). The primary Hamiltonian of the theory is now obtained by adding the primary constraint (10. Aµ (x)}. therefore.129) where we have used (10.

˙ so that the chain of constraints terminates. we have a secondary constraint in the theory given by ϕ2 (x) = ∇ · Π(x) ≈ 0.134) It can now be checked that the secondary constraint is time independent. According to the Dirac procedure. (Π2 (y) + B2 (y)) 2 − A0 (y)∇ · Π(y) + λ1 (y)Π0 (y)} ≈ − d3 y {Π0 (x). A0 (y)}∇ · Π(y) (10. (10. as we have seen. namely. Therefore. 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.410 10 Dirac method for constrained systems ϕ1 (x) ≈ {ϕ1 (x).136) and it is clear that both these constraints are ﬁrst class constraints. Let us choose . Hp } ≈ 0. We see that Maxwell’s theory has two constraints ϕ1 (x) = Π0 (x) ≈ 0. the Lagrange multiplier λ1 remains as yet undetermined (there is one primary constraint which is ﬁrst class). In this case. in the presence of ﬁrst class constraints we are supposed to add gauge ﬁxing conditions to convert them into second class constraints. ϕ2 (x) ≈ {ϕ2 (x).133) = ∇ · Π(x) ≈ 0. ϕ2 (x) = ∇ · Π(x) ≈ 0. Hp } = {∇ · Π(x).135) (10. Hp } ˙ = 1 d3 y {Π0 (x). Furthermore. 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.

139) The inverse of this matrix is easily calculated to be 0 0 1 0 −1 CAB (x.10. relations (10.138) which leads to (x0 = y 0 ) C AB (x. A0 (y)} {ϕ2 (x).136) into second class constraints. χ2 (x) = ∇ · A(x) ≈ 0.137) as the gauge ﬁxing conditions which clearly convert the constraints (10. y) −2 0 0 0 ∇x 3 δ (x − y).137) determine all the constraints of the theory.136) and (10.6 Maxwell field theory 411 χ1 (x) = A0 (x) ≈ 0.140) . ∇ · A(y)} = −δ3 (x − y) = −{χ1 (x). 2. χ1 (y)} = {Π0 (x). (10. ϕ2 (y)}. 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). φB (y)} 0 0 −1 0 2 0 0 0 −∇x 3 δ (x − y). y) = {φA (x). Combining the ﬁrst class constraints as well as the gauge ﬁxing conditions into φA = (ϕα . (10. = (∇x )i (∇y )j {(Π)i (x). χα ). = 1 0 0 0 2 0 ∇x 0 0 (10. = −1 0 0 0 −2 0 −∇x 0 0 (10. α = 1. Therefore. χ2 (y)} = {∇ · Π(x). ϕ1 (y)}. (Π)j (y)} = (∇x )i (∇y )j (δij δ3 (x − y)) 2 = −∇x δ3 (x − y) = −{χ2 (x).

d3 zd3 z {Aµ (x). Πν (y)} − −2 ¯ z +{Aµ (x). Π0 (z)}(δ3 (z − z )){A0 (¯). G(y)}. G(y)} − −1 ¯ z d3 zd3 z {F (x).141) and. Πν (y)}D . φA (z)}CAB (z.136) and (10. Πν (y)}D = {Aµ (x).144) j = δi TR (x − y). j {Ai (x). 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). G(y)}D = {F (x). Πj (y)}D = δi + (∇x )i 1 3 j 2 (∇x ) δ (x − y) ∇x (10.412 10 Dirac method for constrained systems where. ∇z · Π(z)}(∇z δ3 (z − z )){∇z · A(¯). it leads to {Aµ (x).142) ¯ 1 1 3 2 δ (x − y) = − 4π|x − y| . Aj (y)}D = 0 = {Πi (x). z ){φB (¯). ∇x (10.143) Since we can now set the constraints (10.137) to zero. . with the usual asymptotic condition that ﬁelds vanish at spatial inﬁnity. Πj (y)}D . in particular. for the dynamical ﬁeld variables. Aν (y)}D = 0 = {Πµ (x). (10.143) to be {Ai (x). the relevant Dirac brackets for the Maxwell ﬁeld theory follow from (10. Πν (y)} ¯ ¯ z {Aµ (x). ∇x (10. 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).

Constrained Hamiltonian Systems. 2 Hp = (10. K. A. Lectures on Quantum Mechanics. Roma (1976). L. 10. Dirac. A. Das and W. P. New York (1964). 129 (1950). J. Acta Physica Polonica B18. 269 (1987).7 References 413 which are the relations used in (9. Sundermeyer.25)) in quantizing Maxwell’s theory. Physical Review 83. Accademia Nazionale dei Lincei. Berlin (1982). Scherer. Regge and C. Dirac. M. T. P. 4. 6. Hanson. Furthermore. 2. 1 (1951). we obtain the Hamiltonian for the theory (10. 3.7 References 1. A. Anderson and P. 3. J. ibid. Springer-Verlag. Constrained Dynamics. Yeshiva University. A. Barcelos-Neto. 1018 (1951). 5.131) to be d3 x 1 Π2 + B2 . Bergmann. G. M.136) to zero. . Canadian Journal of Mathematics 2.30) (along with the ﬁrst two of (9.10.145) which is the Hamiltonian we have used in the study of Maxwell’s theory in chapter 9. setting the constraints (10. Teitelboim.

This page intentionally left blank .

we will describe three such important symmetries. consequently. these are not continuous symmetries and. it is not a continuous transformation. Classically. In addition to continuous symmetries. this represents a space-time transformation. P (11. there are a few discrete symmetries which play a fundamental role in physics. parity (P).1) while t remains unchanged. namely. we can group objects into diﬀerent categories depending on their behavior under parity or space inversion. It is clear that a vector would change sign in a left-handed coordinate system. We note that even classically.Chapter 11 Discrete symmetries So far we have discussed mainly continuous symmetries of dynamical systems. 11. therefore. Note that (in three space dimensions) space inversion cannot be obtained through any rotation and. however. 415 . we cannot talk of an inﬁnitesimal form for such transformations. 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. 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. In this chapter. Therefore. By deﬁnition. Therefore.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.

A × B −→ (−A) × (−B) = A × B. for example the length of a vector) is expected to remain unchanged under a parity transformation. t) −→ ρ(−x. t) = ρ(x. p −→ −p. A · (B × C) −→ (−A) · ((−B)×(−C)) = −A · (B × C).1) that applying the parity transformation twice. say. P (11. L·p |p| −→ P P P P P P P P P L · (−p) L·p =− . We note from (11. t) −→ −J(−x. (11. 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. we note from (11. Similarly. t) = J 0 (−x. t). A · B −→ (−A) · (−B) = A · B.3) In quantum mechanics. L = x × p −→ (−x) × (−p) = x × p = L. let us note the transformation properties of some of the well known classical variables under parity.416 11 Discrete symmetries A −→ −A. |p| |p| J(x. of course.2) that even though a vector is expected to change sign under a parity transformation. J 0 (x. P P (11. x −→ −x. t). be promoted to operators and they would satisfy the corresponding operator transformation properties.4) . while a scalar (from the point of view of rotations. We see from (11. namely. these variables would. In particular.2) and so on. x −→ −x −→ x.2) that the volume (which behaves like a scalar under rotations) changes sign and is known as a pseudoscalar. This characterization carries over to the quantum domain as well.

11.1) would result in X −→ − X .1 Parity in quantum mechanics. In other words. shows that even that does not present any diﬃculty. For simplicity. For a long time it was believed that the microscopic systems are also invariant under parity. let us assume that P represents the operator which implements the action of parity on the quantum mechanical states.6) from two equivalent points of view. The only diﬃculty we may apprehend is in the case of spinor states since they are double valued. P P (11. we know that dynamical laws of physics (such as Newton’s equation) do not depend on the handedness of the coordinate frame.1 Parity 417 so that that even classically parity operation deﬁnes a group with two elements (a group of order two). of course. under space reﬂection an arbitrary quantum mechanical state transforms as .6) holds. let us consider a one dimensional quantum mechanical system. Classically. But a systematic analysis. In this case.1. simply corresponds to asking whether this group can be carried over to quantum mechanics without any obstruction. but we know now that there exist processes in nature which do not respect parity. namely.6) where we have used the fact that classically vectors change sign under space reﬂection (see (11. we can analyze the parity transformation in (11. as we will see. the classical laws of physics are invariant under parity or space inversion.11. P −→ − P . quantum mechanical states change. the parity transformation (11. (11. In this case. First. Namely. but not the operators such that (11. ½ and P with P 2 = ½. let us assume that under the parity transformation.5) Deﬁning a symmetry in a quantum system. As is standard in quantum mechanics.2)) and that expectation values of quantum operators behave like classical objects (Ehrenfest theorem).

9) that xP |y P = x|P † P|y . It is now obvious from (11. (11. acting on the coordinate basis P must lead to |x → |xP = P|x = | − x .7) ψ P |X|ψ P = ψ|P † XP|ψ = − ψ|X|ψ . = P| − x = |x .418 11 Discrete symmetries |ψ → |ψ P = P|ψ . Since P has real eigenvalues we conclude that the parity operator is Hermitian and thus we have . P is an idempotent operator and the only eigenvalues of P are ±1.9) δ(x − y) = x|P † P|y . P (11. or.12) P 2 |x which shows that P2 = ½. or.13) Namely.10) which leads to P † P = ½.11) In other words. ψ P |P |ψ P = ψ|P † P P|ψ = − ψ|P |ψ .8) Note that parity inverts space coordinates and. such that ψ|X|ψ ψ|P |ψ → → P P (11. the parity operator is unitary. since P|x = |−x . therefore. (11. (11. (11. Furthermore. (11. −x| − y = x|P † P|y .

if |ψ is an eigenstate of parity. 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 . dx |x x|ψ dx ψ(x)P|x (11.16) Namely.11.18) (11. (11. P (11.17) Let us note further that since the eigenvalues of P are ±1. for such a state |ψ −→ |ψ P = P|ψ = ±|ψ .14) To see how a scalar (not a multi-component) wave function transforms under parity. ½. we note from (11. or.19) Namely.7) that P |ψ → |ψ P = P|ψ = P = = P dx ψ(x)|x = dx ψ(x)| − x . In other words. P P (11. (11. under a parity transformation the wave function transforms as ψ(x) −→ ψ P (x) = ψ(−x).1 Parity 419 P2 = P † = P = P −1 .15) so that we obtain x|ψ P = ψ P (x) = ψ(−x). ψ(x) −→ ψ P (x) = ψ(−x) = ±ψ(x).

does not have to be an eigenstate of the parity operator and. we can obtain that [P. 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. one can show that for any operator in this simple one dimensional theory. such that (11.22) where we have used (11. P ) −→ OP (X. (11. PX + XP = 0.23) In general. [P. P P (11.420 11 Discrete symmetries operator for that state. A general state of the system. (11. P ) = P † O(X. ψ|P P |ψ = − ψ|P |ψ . O(X. namely. In the other point of view.6) holds. we may assume that the states do not transform under the parity transformation. therefore. P = O P † XP. the corresponding wave function need not have any such symmetry behavior.20) ψ|X|ψ ψ|P |ψ → → P ψ|X P |ψ = − ψ|X|ψ . −P ).24) . however.21) It follows from this that P X −→ X P = P † XP = −X. P )P = O(−X. or. (11. P † P P (11. Similarly. X]+ = 0.14). P ]+ = 0. or.

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

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

Therefore.14) are more important.32) d |x dx = −P. The important thing to note here is that such a representation is always nonlocal and not very useful from a practical point of view. mix up the diﬀerent components of the object. 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. t) = ηψ Sα ψβ (−x. in principle. from the fact that two space inversions are equivalent to leaving the object unchanged. we are led to the requirement that . As we have seen. t) −→ ψα (x. Furthermore. We have already seen this in the case of Lorentz transformations (see (3. not only does x → −x. 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. such systems can be described consistently only in the language of quantized ﬁelds which are. the properties of the operator P such as in (11.33) β where the matrix Sα can. we need to generalize our discussion of parity transformation to multicomponent objects.35) for transformation of wavefunction). t). P (11. Namely. We note that under a parity transformation a multi-component object would transform as P β ψα (x.11. multi-component operators (recall the Dirac ﬁeld or the Maxwell ﬁeld). in general. On the other hand. Let us now go over to relativistic quantum systems. under a parity transformation. but the components of the object may mix with one another.

t) is a solution of the Klein-Gordon equation.35). t). then φP (x. the above relation represents an operator relation P β ψα (x.34) In the language of quantized ﬁelds. t). Let us begin with the spin zero ﬁeld which is described by the Klein-Gordon equation ∂2 − ∇2 φ(x) = 0. φ(x.2 Spin zero ﬁeld. if φ(x. t)P = ηψ Sα ψβ (−x.1. we conclude that in this case. t). (11. φ= (11.36) depending on whether the ﬁeld is massless or massive. 11. so is φ(−x. cannot have simultaneous eigenstates unless the eigenvalue of momentum vanishes. (11. (Remember that parity and the momentum operators do not commute and.37) If ηφ = 1 or the intrinsic parity of the spin zero ﬁeld is positive. t) = φ(−x. ( + m2 )φ(x) = 0. t) −→ ψα (x.38) P (11.35) 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. t) −→ φP (x. Consequently. In either of the cases. the equation is clearly invariant under space inversion. From our general discussion in (11.424 11 Discrete symmetries 2 ηψ S 2 = ½. ∂t2 or. therefore. t).34) and (11. (11. t) = P † ψα (x. with ηφ = ±1. (11.39) . t) = ηφ φ(−x.) 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.

a ﬁeld or a state may not have a well deﬁned parity value (consider.44) In such a case. π + ϕ) = (−1)ℓ Yℓ. t).40) and we say that such a ﬁeld is a pseudoscalar ﬁeld and that the associated particles are pseudoscalar particles (see. In general. from the fact that P x −→ −x. we can deﬁne the total parity for the state (with orbital angular momentum ℓ) as . (11. ϕ).43) so that φP (x. On the other hand. t) = Furthermore. t) = −φ(−x. t) = ηφ φℓ (−x. t) = ηφ (−1)ℓ φℓ (x. ϕ.m (θ.m (θ. ϕ −→ π + ϕ. if the intrinsic parity of the spin zero ﬁeld is negative or ηφ = −1. then φP (x. t). ϕ). (11. namely. θ.11. t). (11. for example. θ. we have P P P (11. we can expand such a ﬁeld (or a state) in the basis of the angular momentum eigenstates. since parity commutes with angular momentum. On the other hand. ℓ ℓ gℓ (r. ϕ. a one particle state not at rest). ϕ) −→ Yℓ. t) = φP (r. therefore. for example. (11.2)). θ −→ π − θ. ℓ P (11. however.m (π − θ.m (θ.1 Parity 425 and we say that such a ﬁeld is a scalar ﬁeld and that the associated particles are scalar particles. t) = ℓ φℓ (r. the spherical harmonics as φ(x.42) r −→ r.41) φP (x. t)Yℓ. Yℓ.

whereas processes such as (ℓ = 0) S → P + S.49) will be forbidden. (11.46) (11.426 11 Discrete symmetries ηTOT = ηφ (−1)ℓ . (11. Let us further note that for a complex spin zero ﬁeld. P P → S + S.48) S → P + P. It now follows that in a parity conserving theory.45) If parity is a symmetry of the theory. if we have . 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. (11. B and C respectively and ℓ is the orbital angular momentum of the B-C system. (11. one spin zero particle can decay into two spin zero particles in an s-state (ℓ = 0) only through the channels S → S + S. P → P + S. → P + P. we must have ηA = ηB ηC (−1)ℓ .47) where ηA . ηB and ηC are the intrinsic parities of the particles A.

3. t) = i|ηφ |2 φ† (−x. t).11. t)∂ µ φ(−x. since ηφ is assumed to take only real values ±1 (see (11. t) . In quantum ﬁeld theory φ and φ† are associated with particles and antiparticles.38)).29). quite general for the bosons and says that for bosons. t) = −J(x. in fact.(11. see discussion in sections 7. t) = i (φP )† (x. t) −→ J P µ (x. t)∂ µ φP (x.50) On the other hand. t).1 Parity 427 φP (x. t) − (∂ µ (φP )† (x. t). Let us next look at the electromagnetic current associated with a charged Klein-Gordon system. t) = J 0 (−x. t))φ(x. t) = ηφ φ(−x. (φ destroys a particle or creates an antiparticle whereas φ† destroys an antiparticle or creates a particle. the current density has the explicit form J µ (x. J P 0 (x. t) − ∂ µ φ† (x. the intrinsic parity of the ﬁeld φ† is the same as that of the ﬁeld φ. t) = ηφ φ† (−x. ∗ (φP )† (x.51) Under a parity transformation J µ (x. (11. 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. As we have seen in (7. (11. . t).2). t) = i φ† (x. t)∂ µ φ(x.2 and 7. t))φP (x. t) − (∂ µ φ† (−x. t) φ(x. (11.52) which leads explicitly to (recall that |ηφ |2 = 1) JP (x. This result is.53) P which is the correct transformation for the current four vector as we have seen in (11. t) .

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

t)AP (x. t) − JP (x. t) d3 x J 0 (−x. 11. t). In other words. t) = Hem . Let us recall that the Dirac wave function or the Dirac ﬁeld is described by a four component object. parity (quantum number) must be conserved in electromagnetic processes.1. (11. (11.33) or (11. t)Aµ (x. t)A0 (−x.59) from which we conclude that the intrinsic parity of the photon ﬁeld (or the one photon state) is negative. under a parity transformation . t)Aµ (x.1 Parity 429 Note that a physical photon has only transverse degrees of freedom (namely. t) = −A(−x. From the parity transformation of the dynamical components. t) d3 x J P µ (x. t) µ d3 x J P 0 (x.4 Dirac ﬁeld. t)) d3 x J µ (−x. t)Aµ (x. therefore. A0 is nondynamical).11. (11.61) where in the last step we have let x → −x under the integral. we see that AP (x.35). namely A. From the form of the electromagnetic interaction Hamiltonian Hem = d3 x J µ (x. Thus according to our general discussion in (11. This shows that the electromagnetic interaction is invariant under parity and. t)AP 0 (x. t). t) − (−J(x. the photon is a vector particle. t) · AP (x. t)) · (−A(x. t) d3 x J µ (x.60) we note that under a parity transformation Hem = → = = = = P d3 x J µ (x. t)Aµ (−x.

t). 3.66) Let us recall that associated with the Dirac system is a conserved current (probability current or electromagnetic current. (11. t)γ 0 S † γ 0 . t). t) = ηψ Sψ(−x.65) which will satisfy the above relation.67) 2 = ηψ ψ(−x. t)γ 0 S † γ 0 γ µ Sψ(−x. t) = (ψ P )† (x. (11. t). we can always choose ηψ = ±1. see (8. With this choice (which also ∗ implies ηψ = ηψ ) we note that ∗ ψ (x. t)S † γ 0 ∗ = ηψ ψ † (−x. and under a parity transformation this would transform as J µ (x.64) Since we have assumed the parity eigenvalues to be ±1.430 11 Discrete symmetries P β ψα (x. with (see (11. α = 1. t)γ 0 γ 0 S † γ 0 P = ηψ ψ(−x.62) In matrix notation. S 2 = ½. 2. t) −→ J P µ (x. (11. t)γ 0 = ηψ ψ † (−x. (11. t) P P (11. t). t) = ψ(x.34)) 2 ηψ S 2 = ½. 4. t) = ψ(−x. t) = ηψ Sα ψβ (−x. t) = ψ (x. t)γ µ ψ(x.63) (11. P (11.44)) J µ (x.68) . t)γ µ ψ P (x. t)γ 0 S † γ 0 γ µ Sψ(−x. we can write the transformation as ψ P (x. t) −→ ψα (x.

namely. t) = ψ(−x.68) and (11.65) we have S = S † = S −1 . or. or. t)γ 0 ψ(−x. t) also satisﬁes the Dirac equation in the new coordinates. t) = ηψ ψ(−x. Namely. t). t) −→ ψ P (x. On the other hand. t) = −ψ(−x. S = γ 0.53)). comparing (11. (11.1 Parity 431 2 where we have used the fact that ηψ = 1.73) (11. t) −→ ψ (x. we know the transformation properties of the current four vector under space inversion (see (11. ψ(x. t).69) for µ = i. t)γ 0 S † γ 0 γ i Sψ(−x. we obtain ψ(−x. (11.68) and (11. P P P (11. if we deﬁne . then the parity transformed function ψ P (x. t)γ 0 . S † S = ½. t)γ 0 S † Sψ(−x. J P 0 (x. t).69) for µ = 0. (11. (11. with ηψ = ±1.69) Comparing (11. t) = J 0 (−x. t). t). t) = −J i (−x. t) = ηψ γ 0 ψ(−x. t)γ i ψ(−x.71) Similarly.11. we obtain ψ(−x.70) so that using (11.72) Thus we determine the transformation of the Dirac ﬁeld (wave function) under a space reﬂection to be ψ(x. t) is a solution of the Dirac equation.74) It is now quite easy to see that if ψ(x. J P i (x.

−x). in fact. 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. then ∂ − m ψ P (y) = ∂y µ (11. ∂xµ (11. 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. the scalar transition amplitude will have the general form Mi→f = (a ǫ1 · ǫ2 + b k · (ǫ1 × ǫ2 )) ψpositronium . be measured and it . (11. the two photons move with equal and opposite momentum.9). Let ǫ1 and ǫ2 denote the polarization vectors for the two photons while k represents their relative momentum.8 eV) e− + e+ → γ + γ. However. as we have discussed in section 3.76) In other words. the two component Weyl fermions violate P.77) In the rest frame of the positronium. (11.432 11 Discrete symmetries y µ = (t.6.75) 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. space inversion is a symmetry of the Dirac equation. The polarization planes of the photons in this decay can. Since the electromagnetic interactions conserve parity. To determine the relative parity between Dirac particles and antiparticles.

11. Let us recall from our earlier discussions that antiparticles are related to negative energy states. in the rest frame of the positronium. Since the transition amplitude is a scalar under parity while P (11. (11. (11. Namely. (11. we must have ηTOT = ηe− ηe+ = −1. for this decay we have ǫ1 · ǫ2 = 0. the total parity of the state is given by ηTOT = ηe− ηe+ . On the other hand.1 Parity 433 is experimentally observed that the two are perpendicular. the transition amplitude for this decay takes the form Mi→f = (b k · (ǫ1 × ǫ2 )) ψpositronium . in fact. the electron and the positron must have relative negative intrinsic parity.82) and for this state to represent a pseudoscalar state.83) In other words. the relative intrinsic parity between Dirac particles and antiparticles is negative.81) we conclude that the positronium state must be a pseudoscalar under parity. Let us choose two representative solutions (of positive and negative energy) of the Dirac equation in the rest frame (see (2. This is. However. namely. This result can be more directly seen when we study charge conjugation in the next section.79) Consequently. a very general result for fermions.80) k · (ǫ1 × ǫ2 ) −→ −k · (ǫ1 × ǫ2 ) . (11. it can also be seen in an intuitive manner from the structure of the spinor functions as follows.49)) .

t)ψ(−x. we note that ψ(x. Let us next calculate the transformation properties of the sixteen Dirac bilinears ψΓ(α) ψ. On the other hand. t)γ5 ψ P (x. (11. 1 1 0 0 1 1 (11. t)ψ P (x.87) −ψ(−x. the only meaningful quantity as far as the fermions are concerned is the relative parity of a pair of fermions.3) under parity. t) −→ ψ (x. (11. t) (11. since fermions always appear in pairs in any process (because of conservation of fermion number. t). 0 0 0 0 0 0 0 0 = −η . For example. t)ψ(x. t)γ5 ψ(−x.434 11 Discrete symmetries ψp0 >0 ψp0 <0 This analysis also brings out another feature involving the fermions. t)γ5 ψ(x.100) (or in section 3. angular momentum).85) 0 P 0 = −→ ηγ 0 0 0 0 P −→ ηγ 0 = 1 0 1 0 0 = η .84) introduced in (2. t) −→ ψ (x. t)γ 0 γ5 γ 0 ψ(−x.86) which shows that this combination behaves like a scalar under parity. t) ψ(−x. . namely. t)γ 0 γ 0 ψ(−x. t). ψ(x. t) = = P P |η|2 ψ(−x. t) = = P P |η|2 ψ(−x.

t).88) (11. t)σ ij ψ(−x. t)γ i ψ(−x. namely ψψφ. namely. Therefore. t) −→ −ψ(−x. (11. t). the only allowed interaction in this case has the form (recall also the discussion in section 8. t). The transformation properties of the bilinears are important in constructing relativistic theories. Finally. We have already seen that the combination ψγ µ ψ behaves like a four vector. t)σ 0i ψ(x. t)γ 0 ψ(x. t). the spin zero meson can be identiﬁed with the π-mesons which are known to be pseudoscalar mesons. Thus.91) Since strong interactions conserve parity. it was thought sometime ago that the strong force between the nucleons was mediated through the Yukawa interaction of spin zero mesons. As it turns out. t) −→ −ψ(−x. t)γ5 ψ i ψ(x. ψ(x. ψ(x.92)) . t)γ5 γ i ψ(−x. ψ(x.11. t) −→ ψ(−x.1 Parity 435 so that this combination behaves like a pseudoscalar under parity. it is known as an axial vector (or a pseudovector). only one of the forms of the interaction can be allowed depending on the intrinsic parity of the spin zero meson. It can similarly be checked that ψ(x. t)γ5 γ 0 ψ(−x. or ψγ5 ψφ. However. t). t). there are two possible Lorentz invariant interactions we can write down involving fermions and spin zero particles. ψ(x. t) −→ ψ(−x. we can also derive that ψ(x. t) −→ −ψ(−x.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. P P P P (11.89) Namely. t)γ5 γ 0 ψ(x. t)σ ij ψ(x. the combination ψγ5 γ µ ψ behaves in an opposite way from the current four vector and consequently. for example.8 after (8. t)γ i ψ(x. t)γ 0 ψ(−x. t) −→ ψ(−x. t)σ 0i ψ(−x. P P (11.

∂t (11. if we know the interaction Lagrangian or Hamiltonian density for a given process. since ∂A − ∇A0 . ∂t ∂E + J. B → −B. 11. (11. ∇ · B = 0.94)) and is denoted by C.93) We note that this set of equations is invariant under the discrete transformations ρ → −ρ. 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. Furthermore. (11. J → −J. ∇×E = − ∇×B = ∂B . let us go back to Maxwell’s equations in the presence of charges and currents ∇ · E = ρ. ∂t B = ∇ × A. E = − classically we see that charge conjugation leads to (11. E → −E.92) Conversely.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 (11.95) .436 11 Discrete symmetries ψγ5 ψφ.2 Charge conjugation To understand the discrete symmetry known as charge conjugation.

charge conjugation really interchanges every distinct quantum number between particles and antiparticles. two charge conjugation operations (reﬂecting the charge twice) would bring back any variable to itself.97) is invariant under this transformation. C Jµ −→ −Jµ . in a quantum theory. therefore.1 Spin zero ﬁeld. µ C Jµ (x) −→ Jµ (x) = ηJ Jµ (x) = −Jµ (x).96) It is obvious that the electromagnetic interaction Hamiltonian Hem = d3 x J µ Aµ . (11. charge conjugation is trivial in this theory as we will discuss shortly. Therefore. C (11. we can generalize the classical result (11. Note that the transformation of charge conjugation does not change the space-time coordinates and. In the quantum theory.11. charge conjugation is sometimes also referred to as particleantiparticle conjugation.29) as .98) and we say that the charge conjugation parity (or simply the charge parity) of the photon as well as the current is −1. 11.96) by saying that C Aµ (x) −→ AC (x) = ηA Aµ (x) = −Aµ (x). Let us also note that in a quantum theory. a real Klein-Gordon theory describes charge neutral spin zero particles and. Let us now analyze the complex Klein-Gordon ﬁeld theory which describes charged spin zero particles. Let us also note here that like parity. opposite charges are assigned to particles and antiparticles. reﬂecting charges in a quantum theory is equivalent to saying that charge conjugation really interchanges particles and antiparticles. For this reason. (As we have seen. it is not a space-time transformation. In other words.2.2 Charge conjugation 437 Aµ −→ −Aµ .) We have seen that for such a theory we can deﬁne an electromagnetic current (7. therefore. C (11.

99) We note that if we treat φ(x) as a classical function (even though the functions are classical. ∗ φ† (x) −→ (φC )† (x) = ηφ φ(x). we have (|ηφ |2 = 1) φ(x) −→ ηφ φ† (x). we continue to use “†” instead of “∗” so that the passage to quantum theory will be straightforward). (11.100) with |ηφ |2 = 1. ∗ φ† (x) −→ ηφ φ(x). It follows now that under such a transformation.100) is the appropriate charge conjugation transformation for complex scalar ﬁeld.438 11 Discrete symmetries J µ = i φ† ∂ µ φ − (∂ µ φ† )φ . C (11.101) C Thus the ﬁeld transformations in (11.102) .100) would deﬁne the appropriate charge conjugation transformation for the current. Therefore. (11. C C C Aµ (x) −→ −Aµ (x). (11. 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). We have already discussed (see sections 7. under a charge conjugation. this further supports the fact that (11.2 and 7. then C φ(x) −→ φC (x) = ηφ φ† (x). As a result.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).

can be traced to the fact that in going from a classical theory to the quantum theory. All of the above discussion carries over to the quantum theory if we are careful about the operator ordering of φ and φ† .2 Charge conjugation 439 Fµν (x) −→ −Fµν (x).99). 4 (11.103) Therefore. spin zero particles interacting with photons described by 1 L = (Dµ φ)† (Dµ φ) − m2 φ† φ − Fµν F µν .100) do not lead to the correct transformation properties for the current operator. it would appear as if the ﬁeld transformations (11. C C (11. As is known. then under the transformation (11. we note that if we deﬁne the electromagnetic current in the quantum theory as in (11. C ∗ ∗ (Dµ φ(x))† = (∂µ − ieAµ )φ† (x) −→ ηφ (∂µ + ieAµ )φ = ηφ Dµ φ. As a result. Dµ φ(x) = (∂µ + ieAµ )φ(x) −→ ηφ (∂µ − ieAµ )φ† = ηφ (Dµ φ)† . we have to choose an operator ordering which we have not taken care of in the above discussion.11. 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 . 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 .104) is invariant under charge conjugation. of course. the Lagrangian density of charged. (11. For example.105) C since the operators φ and φ† do not commute with ∂µ φ† and ∂µ φ respectively.

In such a case.106) 2 µ Jsym (x) = and it is easy to check that this current will transform properly (namely. (11. under a charge conjugation (11.108) This is esthetically more pleasing and treats the vacuum in a more symmetrical way.440 11 Discrete symmetries normal ordering the product). it is described by a Hermitian ﬁeld operator φ† (x) = φ(x). If we symmetrize the products in the deﬁnition of the current. will change sign) under charge conjugation.110) .106) leads to (this should be compared with the normal ordered charge deﬁned in section 7. (11. Note that we can also write the symmetrized current as 1 µ (J (x) − J Cµ (x)) . we obtain i φ† ∂ µ φ + (∂ µ φ)φ† − (∂ µ φ† )φ − φ∂ µ φ† .107) µ the eﬀect of the inﬁnite constant in Jsym in (11. As we have discussed earlier. if a spin-zero ﬁeld theory describes charge neutral particles which are their own antiparticles.3) µ 0|Jsym (x)|0 = 0.109) which brings out the particle-antiparticle symmetry in a more direct way. 2 µ Jsym (x) = (11. But this has the interesting consequence that while (remember the Dirac vacuum picture) 0|J µ (x)|0 −→ ∞. 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). (11.

(11. C (11.2 Charge conjugation 441 φ(x) −→ φC (x) = ηφ φ† (x) = ηφ φ(x). If charge parity is to be conserved in this process (this corresponds to an electromagnetic process since photons are involved). we conclude that the π 0 meson must also have charge parity (+1) and similar arguments hold for other mesons.112) In this case. But by analyzing various physical processes where charge conjugation is a symmetry or by analyzing the invariance of various interaction terms under charge conjugation.113) φ(x) −→ φC (x) = φ(x). so that in these cases (11. ηφ = ±1.11. we expect 2 ηφ = 1. C (11.114) Such ﬁelds or particles are known as self-charge-conjugate and simply describe particles that are their own antiparticles. As an example. .111) From the fact that two charge conjugation operations should return the ﬁeld to itself. let us note that in the decay π 0 → 2γ. the total charge parity of the ﬁnal state is (+1). one can show that for all Hermitian spin zero ﬁelds that we know of (scalar or pseudoscalar) ηφ = 1.115) since the charge parity of the photon is (−1). (11. the sign of the charge parity cannot be ﬁxed from theory. or.

To see the relation between ψ C (x) and ψ(x). interacting with an electromagnetic ﬁeld. If the fermions are minimally coupled to the photons.120) (11. T T T (11. let us note from (11.118) we would have C i(γ µ )T (∂µ − ieAµ ) + m ψ = 0. from (11.116) +m = 0. iC(γ µ )T C −1 (∂µ − ieAµ ) + m Cψ = 0. it would satisfy the equation (iγ µ (∂µ − ieAµ ) − m) ψ C (x) = 0.118) 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.118) and (11. In fact.2 Dirac ﬁeld. or.119) that the two equations are very similar in structure. (iγ µ (∂µ − ieAµ ) − m) Cψ = 0. we note that if we can ﬁnd a matrix C satisfying C(γ µ )T C −1 = −γ µ . the equation of motion takes the form (iγ µ (∂µ + ieAµ ) − m) ψ(x) = 0. Let us next analyze a system of Dirac fermions in- teracting with an electromagnetic ﬁeld.117). T (11.117) Taking the transpose of the adjoint equation (11.2.121) . Thus. or. (11.119) where we have identiﬁed the charge conjugate function ψ C as describing the positron.442 11 Discrete symmetries 11. we obtain i(γ µ )T (∂µ − ieAµ ) + m ψ = 0. (11. The adjoint equation is easily obtained to correspond to ← ψ(x) iγ µ ∂µ −ieAµ (11.

124) and hence from the generalized Pauli theorem (see (1. (11. In 4-dimension C T = −C. let us assume that instead C is symmetric. namely.125) . −(γ µ )T . C T = C.11. we obtain γ µ C = −C(γ µ )T = −C T (γ µ )T = − (γ µ C)T .2 Charge conjugation 443 which can be compared with (11.127) (11. The important symmetry property that C must have is that it should be antisymmetric. from C −1 γ µ C = −(γ µ )T . C −1 γ µ C = −(γ µ )T . (11. Similarly. we can also show that σ µν C = − (σ µν C)T . we can identify (with |ηψ |2 = 1) ψ C (x) = ηψ Cψ . it follows that there must exist a similarity transformation which relates −(γ µ )T and γ µ . T (11.92)). (This property depends on the dimensionality of space-time.119).123) That such a matrix exists can be deduced from the fact that −(γ µ )T also satisﬁes the Cliﬀord algebra. namely. Therefore we conclude that if we can ﬁnd a matrix C with the property C(γ µ )T C −1 = −γ µ .128) (11.122) (11. or.) To see this.126) (11. −(γ µ )T + = 2η µν ½. In this case.

we have (with |ηψ |2 = 1. therefore. (11.129) Unlike the parity transformation.44)) . given the classical current (see (8. Thus. satisfy anticommutation relations to reﬂect the Fermi-Dirac statistics that the underlying particles obey. C † C = 1) ψ(x) −→ ψ C (x) = ηψ Cψ . (11. as quantum mechanical operators. there can be only six linearly independent antisymmetric matrices and C cannot be symmetric. we can construct 10 linearly independent antisymmetric matrices. C T = −C.444 11 Discrete symmetries so that if C T = C.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 . But these are 4 × 4 matrices and. it is easy to determine that C = γ 0γ 2. It now follows that ψ C = ηψ Cψ = ηψ γ 0 γ 2 (γ 0 )T ψ ∗ = −ηψ γ 2 ψ ∗ . it follows that C must be antisymmetric. Therefore.132) In general. the structure of the matrix C depends on the representation of the Dirac matrices and. if we do not restrict to any particular representation. in the PauliDirac representation that we have been using.133) The fermion ﬁelds. however. C C T (11. ∗ ψ(x) −→ ψ (x) = −ηψ ψ T C −1 .131) (11. T (11. the appropriate operator ordering for the current is antisymmetrization (this is equivalent to normal ordering in this case). In this case.

2 Charge conjugation 445 J µ (x) = ψ(x)γ µ ψ(x). (11.138) 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 .136) µ = −Janti (x). Therefore.122) in the intermediate steps.11. this current transforms properly under charge conjugation. 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 = 0.134) = 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) . (11.137) whereas 0|J µ (x)|0 −→ ∞. we can obtain the antisymmetrized current as µ Janti (x) = (11.135) 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)) .139) Let us next calculate how the Dirac bilinears transform under charge conjugation. . 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.142) 2 Therefore. if we deﬁne properly antisymmetrized Dirac bilinear combinations as = − (γ µ γ5 )T = (γ5 γ µ )T . 2 (11.140) − (γ 1 )T T − (γ 2 )T T − (γ 3 )T T = γ5 . = i γ 0γ 1γ 2γ 3 (11.122) that the charge conjugation matrix C satisﬁes C −1 γ µ C = −(γ µ )T . we obtain T C −1 γ5 γ µ C = C −1 γ5 CC −1 γ µ C = γ5 − (γ µ )T (11.141) 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 . We have already seen in (11. (11.

147) where the phase ǫα for the diﬀerent classes has the form 1.2 Charge conjugation 447 ψΓ(α) ψ anti ≡ 1 2 ψΓ(α) ψ − ψ T Γ(α) T ψ T . for α = S. T. α = V. (11.) We have already seen that ψγ µ ψ −→ − ψγ µ ψ C anti anti . . (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).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 . A. ǫα = −1.146) ψγ5 γ µ ψ anti −→ − ψσ µν ψ anti In general we can denote ψΓ(α) ψ −→ ǫα ψΓ(α) ψ C anti anti . P. for (11. (11. ψψ − ψ T ψ (11.11. (11. (11.143) 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 . .148) .144) 2 (If we think for a moment.

t) = ηψ Cψ (x. t) = ηψ C(ψ )T (x. (11. T we obtain .152) Of course. t) = ηψ γ 0 ψ(−x.151) Here we have used the fact that γ 0 is Hermitian in our metric. t) C P −ηψ γ 0 ηψ Cψ (−x. t)γ 0 . P ψ(x. t)γ 0 )T C P ηψ ηψ Cγ 0 C −1 C(γ 0 )∗ ψ ∗ (−x. t) −→ (ψ C )P (x. t) = ηψ ψ(−x. Thus P if ηψ represents the intrinsic parity of a Dirac particle. t).149) Since the antiparticle is described by C ψ C (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. T T (11. t) −→ ψ P (x. Next. t) P P = = = = = C P ηψ ηψ C(ψ(−x.448 11 Discrete symmetries Let us next recall from (11. t) P −ηψ γ 0 ψ C (−x.83).150) under parity.73) that under parity a Dirac particle transforms as P ψ(x. t) P C −ηψ ηψ γ 0 C(γ 0 )T ψ ∗ (−x. P P P (11. t). T (11. t) −→ ψ (x. we have already seen this from the analysis of the positronium decay in (11. the antiparticle will transform as C ψ C (x. t). let us note that since the charge conjugate wave function is deﬁned to be C ψ C = ηψ Cψ . but this derivation is simpler and more general.

already seen this in the study of neutrinos in section 3. (p − m)ψ C = 0. if γ5 ψ = −ψ.153) As a result.159) . then it follows that C γ5 ψ C = −ηψ C − ψ (11. of course.157) In other words. namely.156) T T C = ηψ Cψ = ψ C .119) we see that the antiparticle must also satisfy the same free equation. We have. (11. / where (11. Let us recall that if a free Dirac particle satisﬁes the equation (p − m)ψ = 0. / (11. the antiparticle of a particle with a given handedness has the opposite handedness. 11.155) On the other hand.3 Majorana fermions.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) . we have C γ5 ψ C = −ηψ Cψ = −ψ C .3. we see that if γ5 ψ = ψ. T (11.11.158) then from (11. (11.6 and have used this in the construction of the standard model in section 14.154) T (11.2.

162) Such particles are known as self charge conjugate fermions or Majorana fermions. and would also satisfy the free Dirac equation (p − m)ψ = (p − m)ψ C = 0. The answer to this question is. From the deﬁnition of charge conjugation in (11. Let us note from the deﬁnition of the antisymmetrized current in (11.165) so that the Majorana fermions are charge neutral and hence cannot have any electromagnetic interaction. (11.164) that for Majorana fermions (ψ = ψ C ) µ Janti (x) = 0. namely. quite obvious. whether there are also fermionic particles that are their own antiparticles and if so how does one describe them. in fact. 2 = (11. These are charge neutral particles described by real quantum ﬁelds φ† = φ. They can be massive and will have only two independent degrees of freedom (because of the constraint in (11.160) We have also seen that there are bosons in nature which are their own antiparticles. T 11 Discrete symmetries (11. / / (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) .123) we see that a Dirac particle which is its own antiparticle must satisfy ψ = ψ C = ηψ Cψ .450 ψ C = ηψ Cψ .162)).161) We can now ask the corresponding question in the case of the Dirac particles. (11.163) T (11. .

) We have also seen that if the intrinsic parity of a Dirac particle P P is ηψ .2 Charge conjugation 451 Secondly.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. however. To answer this. However. (This is. Thus we cannot talk of a Majorana particle with a given helicity. let us note that if we try to write a Lagrangian density for a massive Weyl fermion (say left-handed). In fact. we note that if we deﬁne χ = ψL + (ψL )C . In particular for d = 2 mod 8. not true if the space time dimensionality is not four. this would require P P ηψ = −(ηψ )∗ .167) (11.166) which can be satisﬁed only if the intrinsic parity of a Majorana fermion is imaginary.168) so that we can express a Majorana fermion in terms of Weyl fermions. the naive mass term would have the form / L = iψ L ∂ ψL − mψ L ψL . we can have Majorana-Weyl spinors. we can ﬁnd a simple relation between them. this implies that it cannot have a well deﬁned handedness.169) . 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. Let us also note here that since both the Weyl and the Majorana fermions have only two degrees of freedom. Since a Majorana fermion is its own antiparticle. (11. it can in principle be complex). then χC = (ψL )C + ψL = χ. as we have seen a particle with a deﬁnite handedness will lead to an antiparticle with the opposite handedness. the mass term would trivially vanish since (11.

. the mass term is not zero and we can give a mass to the Weyl fermion. the neutron and the neutrino are charge neutral and hence are prime candidates to be Majorana particles.172) then this would imply violation of baryon number which is not seen at present. 2 2 (11. the essence of the statement that a Weyl fermion is massless. / (11.170) This is. Obviously. In fact. of course.173) this would imply violation of lepton number which is also not seen yet. note that any majorana particle would necessarily imply some sort of violation of a fermion number which is not seen in nature. a Weyl fermion cannot have mass. However. But such an identiﬁcation is not compatible with the experimental results. note that if we go beyond the conventional Dirac mass term in the Lagrangian density. if (once again ν denotes the antineutrino) ν = ν. in this case. if (here n represents the antineutron) n = n. Obviously. namely. we can write ( we will choose C ηψ = 1) C / L = iψ L ∂ ψL + mψL ψL T = iψ L ∂ ψL − mψL C −1 ψL .452 11 Discrete symmetries 1 1 ψ L ψL = ψ (1 + γ5 ) × (1 − γ5 ) ψ = 0. if we require conservation of fermion number. Similarly. Let us next ask whether there are any particles in nature which we can think of as Majorana particles. 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.171) Clearly. 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. (11. (11.

its theoretical studies would require the study of the properties of such particles. It is. = QC|Q. . . then acting on an eigenstate it gives ˆ Q|Q. (11. Another reason for the study of Majorana particles is that they are fundamental in the study of supersymmetry which we will not go into. the direct experimental limits (mνe < 2ev. This. . . we have ˆ ˆ QC|Q. . . On the other hand. 11. . = −Q| − Q. . To understand the eigenstates of charge ˆ conjugation in general. Similarly. We have seen that the pho- ton has odd charge conjugation parity. mνµ < 190keV. . Independent of whether supersymmetry is a symmetry of nature. . . . It follows now that ˆ C Q|Q. means that a one photon state is an eigenstate of charge conjugation with eigenvalue −1. . mντ < 18. = Q|Q. .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.2. = Q| − Q. so that comparing the two relations we obtain (11.2 Charge conjugation 453 We can now ask why anyone would then like to study such particles or mass terms. .4 Eigenstates of charge conjugation. .11. we have also seen that a Hermitian spin zero ﬁeld has charge parity +1. of course. . up to the experiments to decide whether this is really what happens. . . . The answer to this lies in the fact that while no one has really measured the mass of the neutrino. = Q| − Q. . .176) . 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. refer to other quantum numbers the state can depend on. of course. . .175) (11.174) where . let us note that if Q denotes the operator for charge (electric). . . . . .

not all charge neutral states would be eigenstates of charge conjugation operator. they cannot have simultaneous eigenstates unless the charge of the state vanishes. Let us note. ℓ = (−1)ℓ |π − π + . however. ℓ = |π + π − . (11. s respectively. Q + = 0.178) which clariﬁes the point. It is easy to see that C|n C|π − p C|K 0 = |n . = |π + p . that while all eigenstates of C must be charge neutral. or. the charge operator does not commute with the charge conjugation operator and. . a state such as |π − π + can be an eigenstate of C. for the positronium with orbital angular momentum and spin values ℓ. Note that C|π − π + = |π + π − . Similarly. We have.454 11 Discrete symmetries ˆ ˆ C Q = −QC. 0 (11. (11.180) so that such a state would be an eigenstate of charge conjugation with eigenvalue (−1)ℓ . consequently. On the other hand. ℓ . ˆ C. 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.179) The ﬁnal state simply corresponds to the initial state with the two particles interchanged. Thus if the state has well deﬁned angular momentum quantum number. of course. we would have C|π − π + . (11.177) In other words. = |K .

(11. ℓ. Thus if we look at the decay of the positronium to n photons. / (11. then the charge parity must be conserved. would lead to the result that 1 = (−1)n . (11. ℓ. that upon requiring conservation of charge parity.1 which can.185) (11. that the positronium in the ℓ = 0 = s state decays into two photons. similarly.184) It would. |e− e+ .2 Charge conjugation 455 C|e− e+ . this is an eigenstate of charge conjugation with charge parity (−1)ℓ+s .183) This is. consistent with our earlier discussion. in particular.182) then the conservation of charge parity would require (recall that the photon has charge parity −1) (−1)ℓ+s = (−1)n . the decay π 0 → nγ. Note. 11. s → nγ. say that if charge conjugation is a symmetry. of course.181) = |e+ e− . s In other words. (11. s . for example. s = (−1)ℓ+s |e− e+ . namely.11.186) Let us also note here that if we have an amplitude involving N photons shown in Fig. ℓ. If charge conjugation is a symmetry of the theory. then π 0 −→ 3γ. s = −(−1)ℓ (−1)s+1 |e− e+ . describe the decay . ℓ. ℓ. (11. namely.

(−1)n = (−1)N −n . note that if a Hamiltonian is invariant under charge conjugation.189) where U denotes the time evolution operator. Similarly. and it follows from this that C † U C = U. (11. it is also known that charge conjugation symmetry holds true in strong interactions also. any amplitude involving an odd number of photons must vanish if charge conjugation is a symmetry. or. electromagnetic interactions are invariant under charge conjugation. if the strong interaction Hamiltonian respects charge conjugation symmetry.190) (11. (11.187) then conservation of charge parity would require (electromagnetic interactions are invariant under charge conjugation) Figure 11. (11. This result goes under the name of Furry’s theorem. Therefore.1: An amplitude with N photons. then C † HC = H.456 11 Discrete symmetries nγ → (N − n)γ. As we have seen.188) In other words. (−1)N = (−1)2n = 1. then we must have . In fact.

Namely.2 Charge conjugation 457 π − p|U |π − p = π − p|C † U C|π − p = π + p|U |π + p .11. On the other hand the product operation CP appears to be a good symmetry in most weak processes. (11. it is violated in weak processes just like parity. The simplest way to see this is to note that the weak current (see (14. the weak current does not have a well deﬁned transformation property under charge conjugation and hence will violate C-invariance.192) 2 2 where ηJ denotes a phase. Thus we see that P and C (P for parity) are conserved in strong and electromagnetic interactions but are violated in weak interactions. Let us note here without going into details that while charge conjugation is a good symmetry for both strong and electromagnetic interactions.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 . 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 .191) In other words. (11. the scattering cross-section for π − p elastic scattering would be identical to the cross-section for π + p scattering which is experimentally observed.

11. ∇ × E(x) = − ∇ × B(x) = ∂B . ∂t (11. T (11.3 Time reversal Classically. namely. ∂t ∂E(x) + J. Maxwell’s equations in the presence of sources ∇ · E(x) = ρ(x). 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 . (11.195) We note that the Newton’s equation is invariant under time reversal since it is second order in the time derivative. (11. x −→ x. T t −→ −t. time reversal or time reﬂection corresponds to the space time transformation where we reverse only the sign of time. ∇ · B(x) = 0. Similarly.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 .196) are also form invariant if we deﬁne .193) 2 2 where ηTOT denotes the total phase under the transformation.

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

203) ∂ (T |ψ ) = H(T |ψ ). ∂t (11. o However.460 11 Discrete symmetries |ψ −→ |ψ T T = T |ψ . Let us for the moment give up the assumption that the operator T is linear. (11. In fact.201) then it follows that under a time inversion. 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.200) If we assume that the time independent Hamiltonian is also invariant under time inversion. an antilinear operator T satisﬁes . [T . H] = 0. namely. the time dependent Schr¨dinger equation o i ∂|ψ = H|ψ . By deﬁnition. ∂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.202) would lead to iT or. −i ∂|ψ ∂t = T H|ψ . let us assume that T is an antilinear (antiunitary) operator. (11. (11.

the dynamical equation would transform as ∂|ψ ∂t i would lead to T or. (11. T ABT † = AT B T = ηA ηB AB. i −i = H|ψ . (We can also show that T † = T −1 . but we will not need this for our discussions. ∂|ψ ∂t ∂T |ψ ∂(−t) = T H|ψ . = HT |ψ . 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.205) or. o Let us next assume that we are looking at a process which goes from the state |ψi to the state |ψf . if the Hamiltonian is symmetric under time inversion. T AT † = AT = ηA A.3 Time reversal 461 T (α|ψ1 + β|ψ2 ) = α∗ T |ψ1 + β ∗ T |ψ2 .206) . (11.11. i ∂ (T |ψ ) = H(T |ψ ). (11. ∂t Therefore. T iT † = −i.) Under such an antilinear (antiunitary) transformation. = φ|ψ ∗ = ψ|φ . Thus the quantity we are interested in is the transition amplitude ψf |ψi . ηB are phases. with such a deﬁnition of T .204) where A and B are assumed to be Hermitian operators and ηA . φ|ψ −→ T φ|ψ T T |ψ −→ |ψ T T = T |ψ .

204) (see. (11. −t). Pj ] T † = T (Xi Pj − Pj Xi ) T † = Xi (−Pj ) − (−Pj )Xi = − [Xi .209) Let us derive the transformation properties of some of the operators under time inversion. T (11.204) that ψ(x. t) = x|ψ(−t) = ψ ∗ (x. T (11. Classically. T p −→ −p. .211). the role of the initial and the ﬁnal states are interchanged and the transition amplitude of interest would be ψi |ψf .204) as well as (11. In particular. let us note from (11. (11. we expect ψf |ψi −→ ψi |ψf . It follows now that T [Xi .212) where we have used (11. we know that x −→ x. Pj ] . Thus.210) Thus from Ehrenfest’s theorem we would expect the corresponding operators to transform as T XT † = ηX X = X.207) (11. It is reassuring to note that the canonical commutation relations are unchanged by the ηX = 1. ηP = −1.208) which is indeed consistent with the deﬁnition of time inversion transformations given in (11.211) T PT † = ηP P = −P. on the other hand. T ∗ (11. in particular the third relation).462 11 Discrete symmetries In the time reversed process. t) = x|ψ(t) −→ ψ T (x. under time inversion.

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

φ) = = (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. mj = T |j. where we have used the fact that the spherical harmonics for m < 0 are deﬁned with a phase +1. (11. mj = (i)2mj |j.221) It can be shown that the above relation holds true even for half integer angular momentum values.222) 11. T (11. which follows because under time inversion (say for m > 0) (11.m (θ. m (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ.m (cos θ)e−imφ 4π (ℓ + |m|)! (11.219) Yℓ.1 Spin zero ﬁeld and Maxwell’s theory.223) . t) −→ AµT (x. we can write T |j.m (cos θ)eimφ 4π ℓ + |m|)! ∗ −→ Yℓ.3. m = T |ℓ. Thus. φ) = = T θ. −t). namely.220) (−1)m Yℓ.m (θ. m = (i)2m |ℓ. (11. φ|ℓ.−m (θ. the constant (phase) is chosen to be Cm = (−1)m = i2m .218) (11. in general. t) = ηAµ Aµ (x. we can indeed write T |ℓ. φ). We have already seen that under time reversal Aµ (x.464 11 Discrete symmetries |Cm |2 = 1. By convention. −mj . −m .

we have seen that under time reversal. −t). T Fij (x.224) It follows. the current four vector transforms as J µ (x. ηJ = −1. t). of course. t) − ∂ν AT (x. that Fµν (x. −t) F µν (x.3 Time reversal 465 where ηA0 = 1. ηA = −1. 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.228) The Maxwell theory is. where T (11. t) − ∂ν Aµ (x. t) −→ F0i (x − t) = F0i (x. t) −→ J µT (x. −t) 4 1 = d4 x − Fµν (x)F µν (x) − J µ (x)Aµ (x) 4 = S. T T (11. (11. t) = −Fij (x.11. −t). ν µ so that T F0i (x. t) = T ∂µ Aν (x. −t). invariant under time reversal as we have already seen and as can be checked directly. −t) − J µ (x. (11.229) = d4 x S .226) Similarly.225) T T −→ ∂µ AT (x. (11. therefore. t) (11.227) ηJ 0 = 1. −t)Aµ (x. t) −→ Fij (x. t) = ηJ µ J µ (x.

−t).233) T (11. 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. we have φ(x) −→ φT (x) = ηφ φ∗ (x. −t) ∂t2 (11.230) where the matrix T mixes the components of the function and |ηψ |2 = 1. −t))φ(x. For example.234) ∂2 − ∇2 + m2 φ(x) = 0. 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. T (11. −t)). −t) − (∂ µ φ(x. (11. −t))φ∗ (x. −t) − (∂ µ φ∗ (x. −t). for a spin zero ﬁeld. the Klein-Gordon equation is invariant under time inversion. ∂t2 (11.235) . ∂2 − ∇2 + m2 φT (x) ∂t2 ∂2 − ∇2 + m2 ηφ φ∗ (x. Furthermore. ∂t2 ∂2 − ∇2 + m2 φ∗ (x) = 0.232) That is. −t)) −→ i (φT ∗ (x)∂ µ φT (x) − (∂ µ φT ∗ (x))φT (x)) T = −i (φ∗ (x. −t)∂ µ φ∗ (x.231) (11. 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) = = = 0.466 11 Discrete symmetries In general. −t)∂ µ φ(x.

−t)∂ i φ(x. or. t) −→ −i φ∗ (x. −t) − (∂ i φ∗ (x. −t). (11. −iγ 0∗ −iγ 0∗ iγ 0∗ ∂ ∂ − iγ i∗ i − m ψ ∗ (x. From our general discussion (see (11.3.) 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 . t) −→ −i φ∗ (x. −t).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) = 0. t) = ηψ T ψ ∗ (x.239) (11. ∂(−t) ∂x (11. −t) ∂t ∂t J 0 (x. −t) = 0. (Here ∂ we have used the fact that J 0 (x. 11.197). t) = 0. −t). −t) = which is. −t) = T T ∂ ∂ φ(x. −t))φ(x. t) −→ ψ T (x. −J i (x. or. −t) − ( φ∗ (x. −t) is deﬁned with − ∂t .2 Dirac ﬁelds. the correct behavior of the current four-vector under time inversion as we have already seen in (11. of course. −t))φ(x. or. T (11.11. we expect that under time reversal ψ(x.238) ∂ ∂ − iγ i∗ i − m ψ ∗ (x.230)). we obtain J 0 (x. leads to (upon complex conjugation) (−iγ µ∗ ∂µ − m) ψ ∗ (x) = 0. ∂t ∂x ∂ ∂ − iγ i∗ i − m ψ ∗ (x.236) J i (x.3 Time reversal 467 Therefore.

γ ν∗ } = 2η µν ½. if ψ T (x.244) (11. comparing the two equations we see that if we can ﬁnd a matrix T such that T γ 0∗ T −1 = γ 0 . (11. t).242) would satisfy the same Dirac equation as ψ(x. t) were to be a solution of the Dirac equation. −t) = 0. We note that the matrices γ µ∗ satisfy the Cliﬀord algebra {γ µ∗ . ∂t ∂x (11. or.468 11 Discrete symmetries On the other hand. or. −t). γ ν } = 2η µν ½. ∂t ∂x ∂ ∂ + iγ i i − m T ψ ∗ (x. ˜ also satisfy the Cliﬀord algebra {˜ µ . then the time reversed wave function (11.243) γ 0 = γ 0.245) . and the matrices γ µ deﬁned as (˜ µ = γ µ† ) ˜ γ (11. iγ 0 iγ 0 ∂ ∂ + iγ i i − m ψ T (x) = 0. T γ i∗ T −1 = −γ i . ˜ γ i = −γ i .240) Therefore. γ ˜ (11.241) ψ T (x) = ηψ T ψ ∗ (x. we should have (iγ µ ∂µ − m) ψ T (x) = 0.

Thus we can identify T γ µ∗ T −1 = γ µ . the matrix CT −1 anticommutes with all the γ µ ’s and hence must be a multiple of γ5 .246) (11. ˜ leading to T γ 0∗ T −1 = γ 0 .11. They must also be related to each other through a similarity transformation.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). (11.122) that Cγ µT C −1 = −γ µ . Combining the two results we obtain (11. let us invert these relations and write T T −1 γ 0 T T −1 γ i T = γ 0∗ = γ 0† = γ 0T . (11. T = −γ i∗ = − γ i† = γ iT .249) (11. CT −1 γ µ CT −1 −1 = −γ µ . (11.250) CT −1 γ µ T C −1 = Cγ µT C −1 = −γ µ .248) Thus we see that the inverse relation can be written as T −1 γ µ T = γ µT .247) To determine the explicit form of T . Choosing . T γ i∗ T −1 = −γ i .251) In other words. or. Let us also recall from (11.

252) (11. the T matrix satisﬁes T = T † = T −1 .255) Furthermore. t) −→ ψ (x. and using the deﬁnition (11.470 11 Discrete symmetries CT −1 = −γ5 = −iγ 0 γ 1 γ 2 γ 3 . Therefore we see that the time inversion transformations T ψ(x. −t) = T ηψ T γ 0T ψ (x. as in the case of charge conjugation. t) = ηψ ψ T (x. t) = ηψ T ψ ∗ (x. we can show that in four space-time dimensions.256) so that the time inversion matrix in the spinor space is antisymmetric. −t) = ηψ γ 0 T ψ (x.254) (11.257) with (|ηψ |2 = 1) deﬁnes a symmetry of the Dirac equation. independent of any representation (the superscript denotes transposition) (T )T = −T. (11.130) C = γ 0γ 2. −t). (11. t) −→ ψ T (x. Note that with this choice. we determine (in our representation) T = iγ 1 γ 3 . −t)T −1 γ 0 .253) (11. T T T ∗ ψ(x. (11. Let us next analyze the behavior of the current four-vector under time inversion .

Furthermore. −t)T −1 γ 0 γ µ γ 0 T ψ (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. −t)γ 0 ψ(x. −t). −t)T −1 γ 0 γ µ γ 0 T ψ (x. −t) ψ T (x. (11. −t)γ i ψ (x. we also see that a Dirac system interacting with photons would be invariant under such a transformation. let us note that since T −1 γ µ T = γ µT . t) −→ ψ T (x. −t) T T = −J i (x. Before we examine the transformation properties of Dirac bilinears under time inversion. we have . −t) ψ(x. (11.258) T J 0 (x) −→ ψ T (x. −t)γ i ψ(x. −t) = = T T T ψ T (x. −t) − ψ(x.122) except for the −1 sign). −t) T T T T = J 0 (x.259) which is the behavior of the current four vector under a time inversion and we can identify the transformations in (11. −t)T −1 γ 0 T ψ (x. −t). −t). −t) = = −ψ T (x. −t)T −1 γ 0 γ i γ 0 T ψ (x.257) to correspond to time inversion for a Dirac system.260) (this is very much like the charge conjugation in (11. (11.11. −t)γ 0 ψ (x. T J i (x.

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

j. the tensor bilinears have precisely the same behavior under time reversal as the electromagnetic ﬁeld strength tensor F µν .265) In other words.3. the chirality of a spinor does not change. t) −→ ψ(x. then the transition amplitude can be represented as (11. −t)σ 0i ψ(x. namely. t)σ ij ψ(x. (11.3 Consequences of T invariance. Let us recall that under time reversal.264) We can also show that ψ(x. t)σ 0i ψ(x. a state with given angular momentum transforms as (see (11. p. (11. −t)σ ij ψ(x. t) −→ −ψ(x. It is also easy to see now that under time reversal.3 Time reversal 473 the helicity does not change. We could have expected this from the deﬁnition (see (3.11. j.268) . T = ηψ γ5 T ψ ∗ = ηψ T γ5 ψ ∗ † = ηψ T γ5 ψ ∗ γ5 ψ T = ηψ T (γ5 ψ)∗ . 11. |p| |p| |p| h= (11.266) so that a left-handed spinor remains left-handed under time reversal and a right-handed spinor remains right-handed.267) where α stands for all the other quantum numbers of the state. −mj . Let us also note that if we are looking at a reaction of the form a + b −→ c + d.121)) s·p s · p T (−s) · (−p) −→ = = h. −p.222)) T |α. mj −→ (i)2mj |α. ψ(x. T T (11. −t). −t).

it does not follow a priori that | i|S|f |2 = | f |S|i |2 . (11.272) (11. if we are looking at the reciprocal reaction c + d −→ a + b. to obtain the spin of π-meson etc.273) On the other hand. namely. (11.274) (11. It is worth emphasizing here that the detailed balance does not say that the rates for the two reactions a + b −→ c + d. b . experimentally this is seen to hold in many processes. namely. |i = |a. S † = S.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. and c + d −→ a + b. This is known as the principle of detailed balance and has been used.271) (11.269) where |i stands for the initial state. then the transition amplitude can be represented as i|S|f .474 11 Discrete symmetries f |S|i . |f = |c. (11. where we are still using the same notation. On the other hand. for example.275) . d . (11.

pf . (11.278) In this case. 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 . In this case. (11. = |β. mi . mf . Let us next try to understand.276) where ρf denotes the density of states for the ﬁnal state which may not be the same for the two processes. T ST † = S. then we have | f |S|i | = 2π| f |Hint |i |δ (Ei − Ef ) = 2π| i|Hint |f |δ (Ei − Ef ) = | i|S|f |. theoretically. (11. when the principle of detailed balance may hold.11.3 Time reversal 475 have to be the same. (11. therefore. if the interaction Hamiltonian is Hermitian. That is. the principle of detailed balance will hold. Let us assume that we are dealing with a system where the interaction Hamiltonian can be treated as a perturbation. In fact. h (11.279) Let us choose the initial and the ﬁnal states to correspond to |i |f = |α.277) and furthermore. we can write (for distinct |i and |f states) f |S|i = −2πi f |Hint |i δ (Ei − Ef ) . Let us next assume that we are dealing with a system whose Hamiltonian is invariant under time reversal.280) . pi . It follows then (in the operator formulation) that the evolution operator or the S-matrix must also be invariant under time inversion.

mf |S|α. Since the transitions can only be from integer spin states to integer spin states or half integer spin states to half integer spin states. −mf |. mi | = | α. | β. −mf |. −pf . or. pi . −pi . if we square this relation and sum over all the spin projections mi . mi = ± α. pf . pi . −pf . (11.222) T |i T |f = (i)2mi |α. −mf . pf .283) This is known as the reciprocity relation and holds for any system where time reversal invariance is a symmetry. mf | = | α. −mi |S|β. −mi .281) so that in this case. pi . −mi |S|β. −pf . mf |T −1 T ST −1 T |α. mi = | α. −pi . mf |S|α. −pi . Under time reversal. −mf . −mf . But this is not the same as the detailed balance. −pi . Let us note that if. pi . (11. pf . mf |S|α. pi . mf |S|α. pi . the phase factor can only take values ±1 and we have β. −pf . pf . we have | β. −pf . = (i)2mf |β. pf . we obtain .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). mi |P −1 PSP −1 P|β. mi = = = (i)2mi −2mf α. the system is invariant under parity. (11. −mi |S|β. −mi |S|β.204). (11. as we have seen in (11. pf . in addition. time reversal invariance would imply that f |S|i β. −pi .284) Therefore. mi | β. mf .282) where we have used (the third) relation in (11.

pf . Note that in weak interactions. pi . the principle of detailed balance will hold. Time reversal invariance holds extremely well in strong and electromagnetic interactions. mf |S|α. pf . pi . m (11.286) Furthermore. pi . (11. m|S|α. pf . Note that if we are looking for a subclass of processes where mi = mf = m.3 Time reversal 477 mi . therefore. pi . pi . Therefore. Thus in this case we will have β. mi |S|β. then from P and T invariance we have β. it has . since rotations are assumed to be a symmetry of physical systems. m = α. −m|S|β. m|S|α. pf . pi . it is obvious that the principle of detailed balance or the semi-detailed balance must hold in practically all processes. pf . mi |2 | α. The stringent limits on T -invariance in electromagnetic interactions comes from the measurement of the electric dipole moment of the neutron.287) (11. mf |2 .4 Electric dipole moment of neutron. P is violated – but the strength of the interaction Hamiltonian is assumed to be weak. This relation is known as the principle of semi-detailed balance and holds true in every system for which P and T are symmetries. m = α. 11.11. On the other hand. we can make a rotation on the right-hand side to bring the state | − m → |m . pf .3.mf This is very similar to the principle of detailed balance but not quite. (11. −m . strong and electromagnetic interactions are assumed to be P and T symmetric.288) In this case. Let us recall that although the neutron is electrically neutral.mf | β. m|S|β.285) = mi .

so that we have σ · E −→ −σ · E. σ · B −→ σ · B. σ · B −→ σ · B.291) σ · E −→ −σ · E. Experiments have obtained extremely small limits on µe . the electric dipole moment would be parallel to its spin.63 × 10−25 e-cm.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. σ −→ −σ. (µe )exp < 0. B −→ B.292) In other words. P P P P P E −→ E. Consequently.290) We have already seen that under parity and time reversal E −→ −E. then the electric dipole moment of the neutron must vanish.293) This basically measures the extent to which P and T invariances hold in electromagnetic interactions. Since there is no natural axis for the neutron other than the spin axis. T T T (11. (11. 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. (11.478 11 Discrete symmetries a magnetic dipole moment. T T (11. Let us assume that it also has an electric dipole moment. 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. σ −→ σ. this is the Pauli coupling) Hint = µe σ · E + µm σ · B. if electromagnetic interactions are invariant under P and T .

4 CPT theorem 479 11. Let us consider the state of a particle at rest which satisﬁes H|α.296) If the Hamiltonian is CPT invariant. m. s . −s . m. s −→ η α. s α. −s|. Thus under the CPT transformation. Thus we see that m = = = α. The operation of CP. s|(CPT )−1 (CPT )H(CPT )−1 (CPT )|α. We can naturally ask whether there exists any combination of these symmetries which will be a symmetry of the weak Hamiltonian. CPT (11. Rather than going into the technical proof of this theorem. m. m. −s|H|α. C and T in any order even though the individual transformations may not be a symmetry of the system. m. 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. m. of course.297) . −s|(CPT )H(CPT )−1 |α. 11.1 Equality of mass for particles and antiparticles.4 CPT theorem We have seen that some of the discrete symmetries appear to be violated in weak interactions. (11.11. s = m|α. s α. (11. m. m.4. m. m.295) where m. then it follows that m = α. s its spin and α denotes the other quantum numbers. m. quantities with bars would refer to antiparticle parameters). let us discuss the obvious consequences of this theorem. α denote the parameters associated with the antiparticle (in this section. m. s|H|α. −s = m. takes a particle to the antiparticle state. |α.294) where m denotes the mass of the particle. (11.

for example.2 Electric charge for particles and antiparticles.301) .298) = −q.480 11 Discrete symmetries In other words. the equality of particle and antiparticle masses is a consequence of CPT invariance. q. so that ˆ CPT ˆ Q −→ −Q. ˆ 11. 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− . In fact. K 0 and K have the same mass. Let Q denote the operator which measures the charge of a state. it is CPT invariance which guarantees 0 that. (11. s ˆ α. s|(CPT )−1 (CPT )Q(CPT )−1 (CPT )|α. q.4. 11. −s|(−Q)|α. q. Since the weak interactions violate parity. In this way. q. one can show that all the quantum numbers of antiparticles are equal in magnitude but opposite in sign from those of the particles. q. the electric charge for particles is equal in magnitude but opposite in sign from that of the antiparticles. Even though C-invariance is violated in weak interactions.300) (11.3 Equality of lifetimes for particles and antiparticles. As we have seen ˆ P ˆ Q −→ Q. we have q = = = ˆ α. s|Q|α. Namely. (11. −s (11. s ˆ α.) Thus for a one particle state with charge q at rest. ˆ (Recall Q = d3 x ψγ 0 ψ for a Dirac particle. q. the ex0 perimental limit on the mass diﬀerence between the K 0 −K provides the best limit on CPT invariance.4. ˆ T ˆ Q −→ Q.299) ˆ C ˆ Q −→ −Q.

−mj |H± |β. j. Let us consider the decay channel of a particle |α. mj −→ |β. mj |(CPT )−1 (CPT )H± (CPT )−1 (CPT )|α. 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. β.304) = (i)2mj −2mj α. j. j. j.303) The cross terms would have a pseudoscalar character and would vanish since the reaction rate is a scalar quantity. mj −→ |β. Furthermore.307) . We know that β. −mj = α. we can rewrite this relation as β. j. j.302) In such a case. mj . −mj |H± |β. (11.306) where we have used the fact that H± would be individually invariant under CPT according to the CPT theorem. −mj . using the rotational invariance of physical Hamiltonians. (11. and the corresponding antiparticle decay channel |α. mj . j.4 CPT theorem 481 where PH± P −1 = ±H± . j. mj (11. mj |H± |α. j. j. mj |H± |β. mj |H± |α. j.11. j. (11. mj .305) (11. j. mj = (11. mj = 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. j. mj |H± |α. 984 (1958). we note here that the individual decay channels may have diﬀerent life times for particle and anti-particle. However. mj |H+ |α. j. (11. namely 2π 1 = τ = 2π β β | β. S. mj |H± |α. j. j. it follows that the weak interactions must violate T -invariance by a small amount as well so that CPT will hold. mj |2 . From CPT invariance. j. j. mj |2 + | β.482 11 Discrete symmetries Since the Hamiltonians H± are Hermitian we also have β. j. . Physical Review 109. j.310) = 1 . (11. and τ ) of particles and antiparticles. j. mj |2 ρβ (11. mj |2 + | β. mj |H+ |α.5 References 1.308) Here we have neglected an overall phase factor (depending on α. mj |H± |α. j. mj |H− |α. therefore. mj |H± |α. mj |2 = | β. j. 11. mj |2 ρβ | β. j. mj = β. 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. Okubo.309) which leads to a relation between the total life times (τ . mj |H− |α. j. β) which is not relevant for the subsequent discussion. This follows trivially from the equality of particle and antiparticle masses. we have obtained | β. j. There are many other interesting consequences of the CPT theorem including the connection between spin and statistics which we cannot go into. mj ∗ . j.

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

This page intentionally left blank .

We will now study gauge theories based on more complicated symmetries. (12. Let us recall that gauge ﬁeld theories necessarily arise when we try to promote a global symmetry to a local symmetry. we have talked about the simplest of gauge theories.1) where ψ provides a representation of U (1) (namely. (12. namely.9) / L = iψ∂ ψ − mψψ.3) .1 Non-Abelian gauge theories So far. this Lagrangian density has the global symmetry ψ → e−iθ ψ. Let us start with the Dirac Lagrangian density for a free fermion (8. As we have seen in (8. the Maxwell theory which is based on the Abelian gauge group U (1). For example.Chapter 12 Yang-Mills theory 12. let us quickly review how the gauge ﬁelds come into the theory in the case of quantum electrodynamics (QED).2) ψ → ψeiθ . 485 (12.38). iǫψ. we have −iǫψ. 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. it is a complex ﬁeld). δψ δψ = = where θ is a real constant parameter and inﬁnitesimally.37) and (8.

1) whose variation would cancel the ∂µ ǫ(x) term. If we now want to promote the symmetry to be local. δψ = iǫ(x)ψ. the phase of ψ remains arbitrary.7) . (12.4) = −ǫ(x)ψ∂ ψ + ψγ µ ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ / = (∂µ ǫ(x))ψγ µ ψ.486 12 Yang-Mills theory with θ = ǫ = inﬁnitesimal.5) (12. As we have seen earlier (see (9.6) 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) since ∂µ ψ does not transform covariantly and there is no term in the Lagrangian density (12. (12.1) nor the corresponding action is invariant under the local phase transformation in (12. we have δL = iδψ∂ ψ + iψγ µ ∂µ δψ − mδψψ − mψδψ / −imǫ(x)ψψ + imǫ(x)ψψ. Consequently. the way out of this diﬃculty is to deﬁne a covariant derivative Dµ ψ such that it transforms covariantly under the local transformation. Neither the Lagrangian density (12. 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. namely ǫ = ǫ(x) (or θ = θ(x)). Consequently. δ (Dµ ψ) = −iǫ(x)Dµ ψ(x). we note that δψ = −iǫ(x)ψ. namely. (12. δ (∂µ ψ) = ∂µ (δψ) = ∂µ (−iǫ(x)ψ) = −i ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ(x).

13) . is nondynamical in this theory.10). e δAµ (x) = (12. δψ = iǫ(x)ψ(x). (12.12. 4 (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 .86). δAµ = 1 ∂µ ǫ(x). e (12. the Lagrangian density L = iψγ µ Dµ ψ(x) − mψψ.10) However.1 Non-Abelian gauge theories 487 with the transformation property for the additional ﬁeld 1 ∂µ ǫ(x). (12. this would transform as δFµν = ∂µ δAν − ∂ν δAµ = 1 (∂µ ∂ν ǫ − ∂ν ∂µ ǫ) = 0. e (12. To introduce the kinetic energy part for the gauge ﬁeld (in order to give it dynamics) in a gauge invariant manner.8) We recognize (12. we note that since the gauge ﬁeld transforms longitudinally (by a gradient). if we deﬁne a ﬁeld strength tensor Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .7) as describing the minimal coupling of the charged fermions to the electromagnetic ﬁeld and as we have seen in (9. the ﬁeld Aµ known as the gauge ﬁeld.9) is invariant under the inﬁnitesimal local gauge transformations δψ = −iǫ(x)ψ(x). That is.11) under the gauge transformation (12. the curl of the ﬁeld would be unchanged by such a transformation.

· · · . . (12. aT a k = ψU −1 = ψeiθ k . / k = 1. dim G represents the dimensionality of the symmetry group. 2. Furthermore. .10). a = 1. Let us consider a free Dirac theory described by the Lagrangian density (containing several complex fermion ﬁelds) L = iψ k ∂ ψk − mψ k ψk . dim R. (12. (We note here parenthetically that dim G = n2 − 1 for the group SU (n). dim G. Namely. . 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. . . . (12. . Here the T a ’s represent the generators of the internal symmetry group G and are assumed to be Hermitian. . 4 (12. a δψk = −iǫa Tkℓ ψℓ . 2.15) where k is an internal symmetry index. 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.) The generators (matrices) T a satisfy the Lie algebra of the Lie group G (this is necessary for the transformations to form a group) .14) and has the local gauge invariance described in (12.16) where θ a denotes the real global parameter of transformation and inﬁnitesimally we have δψ k = iǫa ψ ℓ (T a )ℓk . k a = 1.488 12 Yang-Mills theory The total Lagrangian density for QED is then given by (see also (9. 2.17) with θ a = ǫa = inﬁnitesimal. dim G. Let us now generalize these ideas to theories with more complicated symmetries.

. T b = if abc T c . 2.17) (or (12. a δψ k = iǫa (x)ψ ℓ Tℓk . we already know from the study of angular momenta. . we note that the ordinary derivative acting on the fermion ﬁeld does not transform covariantly under this transformation. 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. b. 2. f 147 = f 246 = f 257 = f 345 = 2 . For example.19) Here we have used the fact that the symmetry generators are Hermitian. namely. (For semi-simple groups the structure constants can always be chosen to be completely antisymmetric. (12. Let us now try to make this symmetry a local symmetry of our theory. b. That is. f 458 = f 678 = 23 . (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. the structure constants of the group SU (2) can be identiﬁed with f abc = ǫabc . the structure constants are nontrivial). the symmetry group is known as a non-Abelian group.) When the generators of the symmetry group 2 do not commute (namely.18) − 1 .12. c = 1. namely. 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. a. . dim G. It is easy to see that the Lagrangian density is invariant under the transformations (12. a. (12. let us consider the inﬁnitesimal local phase transformations a δψk = −iǫa (x)Tkℓ ψℓ .16)). (12.20) Once again.1 Non-Abelian gauge theories 489 T 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. . c = 1.

23) Introducing a new ﬁeld (gauge ﬁeld). (12.20).22).22) As in the case of the U (1) theory.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 .24) (12. We note that the number of gauge ﬁelds (that are needed to deﬁne a covariant derivative) is the same as dim G. µ µ (12. / / δL = δψ k (i∂ − m)ψk + ψ k (i∂ − m)δψk = (∂µ ǫa (x))ψ k γ µ (T a )kℓ ψℓ . µ (12.21) so that under the local transformation (12. (12. (12.25) 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.24) explicitly as a a Dµ ψk = ∂µ ψk + igTkℓ Aa ψℓ = ∂µ δkℓ + igTkℓ Aa ψℓ . 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 . we write the covariant derivative (in the matrix notation) as Dµ ψ = (∂µ + igAµ ) ψ. where g denotes the appropriate coupling constant and Aµ = T a Aa . there is no other term in the Lagrangian density whose variation would cancel the ∂µ ǫa (x) term in (12. We can compare this with QED where the generator is the identity matrix ½ which commutes with every operator. We can also write the covariant derivative in (12.26) . the dimension of the Lie group.

28) We note that the left-hand side in (12.12.29) Substituting this into (12. µ µ (12. 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. a δ (Dµ ψ)k = −iǫa Tkℓ (Dµ ψ)ℓ . 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 . µ This can also be written as . 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 .27) we conclude from (12. (12.26) 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 .30) a = i (∂µ ǫa ) Tkℓ ψℓ + igǫa f abc (T c )km Ab ψm .28).

. we can write the Lagrangian density in (12.34) If we use the deﬁnition (see (12.33) with the covariant derivative deﬁned in (12. µ g (12.492 12 Yang-Mills theory a igTkℓ ψℓ δAa − µ 1 ∂µ ǫa + f abc Ab ǫc = 0.32) Thus the Lagrangian density L = iψ k γ µ (Dµ ψ)k − mψ k ψk . . µ µ g g (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).36) with normal product rules for matrices. µ g (12. dim G. µ a = 1. . . the Lagrangian density is a scalar) L = iψγ µ Dµ ψ − mψψ.33) in terms of matrices as (of course.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 . (12.25)) Aµ = T a Aa . (12. δAa = µ 1 ∂µ ǫa − f abc Ab ǫc . 2. (12.26) is invariant under the inﬁnitesimal local transformations a δψk = −iǫa (x)Tkℓ ψℓ (x). We can now deﬁne a unitary representation of the group as describing the ﬁnite symmetry transformation matrix (see (12. a δψ k = iǫa (x)ψ ℓ Tℓk .16)) .

39) .38).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 . (12. U † = U −1 .38) To construct the Lagrangian density for the dynamical part of the gauge ﬁeld in a gauge invariant manner. (12.12.1 Non-Abelian gauge theories 493 U = e−iθ a (x)T a . the tensor representing the Abelian ﬁeld strength (see (12. Aµ → U Aµ U −1 − 1 (∂µ U ) U −1 .37) so that we can write the ﬁnite transformations (corresponding to (12. ig (12. ψ → ψU −1 .34)) for the ﬁeld variables in the matrix form as ψ → U ψ. we note that 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 (12.39) and (12.38).41) (12. if we deﬁne the ﬁeld strength tensor as Fµν = ∂µ Aν − ∂ν Aµ + ig [Aµ .494 12 Yang-Mills theory Namely.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 . Aν ] U −1 − − ∂µ U Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 .40). then under the gauge transformation (12. (12. in the present case. Fµν → U Fµν U −1 . we see that unlike in the case of QED. It is now easy to construct the gauge invariant Lagrangian density for the dynamical part of the gauge ﬁeld as (quadratic in derivatives) . Aν ] . Let us also note that under the gauge transformation ig [Aµ .42) so that Fµν transforms covariantly.38). (12. it is clear that.

43) where a particular normalization for the trace of the generators is assumed (namely. the ﬁeld strength tensor (12.38) which have the inﬁnitesimal form (see (12. therefore. a Fµν T a = ∂µ Aa − ∂ν Aa T a + igAb Ac T b . In components. Aν . T a T b = 1 δab .59) and the comments fol2 lowing that equation).1 Non-Abelian gauge theories 495 1 1 a Lgauge = − Tr Fµν F µν = − Fµν F µν a . 2 4 (12. more generally for fermions interacting with a non-Abelian gauge ﬁeld.38).34)) . Fµν = ∂µ Aν − ∂ν Aµ + ig Aµ . or. (12. T c ν µ µ ν = ∂µ Aa − ∂ν Aa T a − gf abc Ab Ac T a . This Lagrangian density is invariant under the non-Abelian gauge symmetry transformations (12.41) is easily seen to take the form.46) where we can set m = 0 if the fermions (quarks) are massless.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 .12. see (12. From the cyclicity of trace this Lagrangian density is easily seen to be invariant under the gauge transformations (12. a a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac = −Fνµ . ν µ µ ν or.45) The complete Lagrangian density for QCD (quantum chromodynamics. µ ν ν µ (12. QCD corresponds to the particular case when G = SU (3)) is. 4 (12. given by 1 a LQCD = − Fµν F µν a + iψ k γ µ (Dµ ψ)k − mψ k ψk .

(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. then the Lie algebra breaks up into two commuting non-Abelian subalgebras.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. (12.49) imposes a restriction on the structure constants of the group of the form if abp T p . The algebra of the Hermitian generators. T a . T c + if cap T p . T a = 0. T b + if bcp T p . f abp f pcq T q + f cap f pbq T q + f bcpf paq T q = 0. T c + T c. 1 ∂µ ǫa − f abc Ab ǫc . T b = if abc T c . or. T a = 0. T b + T b . has the form T a . T b .37) as well) U (x) = eiθ a (x)T a . f abp f pcq + f cap f pbq + f bcp f paq = 0.49) Using (12.47) Let us discuss brieﬂy some of the properties of the Lie algebra of the symmetry group G. δAa = µ δψk = −iǫa (x)(T a )kℓ ψℓ . T c . or. µ g (12. (12. (12.18).48) it can be seen that the Jacobi identity (12. as we have noted in (12.48) and the Jacobi identity associated with this algebra is given by T a.496 12 Yang-Mills theory δψ k = iǫa (x)ψ ℓ (T a )ℓk . In this case .

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. For any representation of a simple Lie group we can write Tr T a T b = C2 δab . Furthermore.54) Tr Tr T a T b T c − Tr T b T a T c . A non-Abelian Lie group that cannot be so factorized is called a simple Lie group.1 Non-Abelian gauge theories 497 the group G is a direct product of two independent non-Abelian Lie groups.52) We note that C2 is a normalization constant which determines the values of the structure constants. T b T c (12.53) Using the cyclicity property of the trace. let us next note that the quantity habc = = T a.12. we will assume that the symmetry group G is simple. On the other hand. (12. Tr T a T b = (12.55) where the index c is ﬁxed (and not summed). using the commutation relation (12.18) we obtain habc = Tr if abp T p T c = if abp Tr T p T c = if abp Kp δpc = iKc f abc . a = b. In all our discussions. is completely antisymmetric in all its indices. A direct product of simple Lie groups is called semi-simple. (12. It depends on the representation but not on the indices a and b. we note that .

(12. we have hacb = −habc . This shows that we can write Tr (T a T b ) = C2 δab .56) we conclude that Kb = Kc = K. 2.61) . (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. m.60) then T (R) is known as the index of the representation R. T c T b = if acp Tr T p T b = if acp Kp δpb = iKb f acb = −iKb f abc .59) where the constant C2 depends only on the representation. dim R. n = 1. since habc is completely antisymmetric. (12.) We note that if we write Tr T a T b = T (R) δab .498 12 Yang-Mills theory hacb = Tr T a. Furthermore. (12. · · · . b = 1. dim G. we note here that it is (12.57) Using this and comparing (12. Similarly. However.43).55) and (12. a.58) (12. (12.56) with the index b ﬁxed. 2. we have (T a T a )mn = C(R) δmn . · · · . (12.59) which is used to show that the structure constants for a semi-simple group are completely antisymmetric in the indices.

T a† = = Ta ∗ ∗ bc cb − if acb = if acb = −if abc = T a bc . (12. we can easily check that this representation satisﬁes the Lie algebra.62) We can determine the generators of the group in various representations much like in the case of angular momentum. a particular representation that is very important as well as useful is given by Ta = −if abc . However. namely.63) bc This is consistent with the hermiticity requirement for the generators.64) Furthermore. 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. we conclude that the identiﬁcation .65) where we have used the anti-symmetry of the structure constants as well as (12. (12.1 Non-Abelian gauge theories 499 where C(R) is known as the Casimir of the representation R. As a result. (12. T a. 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 .50). (12.

µ (12. transforms according to the adjoint representation of the group.66) indeed deﬁnes a representation of the Lie algebra known as the adjoint representation.26)) δAa = µ 1 (adj) a D θ .500 12 Yang-Mills theory a T(adj) bc = −if abc . T b .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 . which inﬁnitesimally has the form Fµν → U Fµν U −1 so that = Fµν + iǫb Fµν . Let us next look at the transformation for the gauge ﬁelds in (12. g µ (12. (12. (12. (12. Aµ . This can also be seen from the transformation of the ﬁeld strength tensor in (12.42).69) .68) which shows that the gauge ﬁeld.67) so that we can equivalently write (see. for example.

(12.72) where Aµ = Aa T a and ǫ = ǫa T a (with T a in any representation). let us scale the ﬁeld variables as . (12.73) Let us note the following features of this Lagrangian density. ǫ]. the gauge ﬁeld must transform in the adjoint representation. T b . T b a = iǫb Fµν T a . let us note here that the covariant derivative in the adjoint representation (12. 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.12. or.67) (adj) Dµ ǫ a = ∂µ ǫa − gf abc Ab ǫc .34) we conclude that the ﬁeld strength Fµν as well as the gauge ﬁeld Aµ transform according to the adjoint representation of the group.71) can also be written in the matrix form as (adj) Dµ ǫ = ∂µ ǫ + ig[Aµ . a a δFµν T a = iǫb Fµν if abc T c .1 Non-Abelian gauge theories 501 δFµν = iǫb Fµν .) For completeness. µ (12. (It does not matter what representation the matter ﬁelds belong to. ab µν (12. 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 .70) Comparing this with the transformation of the fermions in (12. or. For example.

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. therefore. 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.43) (or (12.45)) 1 a 1 (12. 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 . ν µ µ ν = ∂µ Aν − ∂ν Aµ + ig[Aµ . The other feature to note is that unlike the photon ﬁeld. Since the gauge ﬁelds couple to any ﬁeld (particle) carrying charge of the symmetry group.75) LYM = − Tr Fµν F µν = − Fµν F µν a .502 12 Yang-Mills theory Aa → µ 1 a A .41) and (12. g µ It follows then that 1 a Fµν → ∂µ Aa − ∂ν Aa − f abc Ab Ac . Aν ]. 12. (12. Physically we understand this in the following way. in the case of non-Abelian symmetry they must couple to themselves and possess self interactions.74) Thus we note that in this theory. 2 4 where the ﬁeld strength tensor is given by (see (12. here the gauge ﬁelds have self interaction (they interact with themselves) and. the pure Yang-Mills theory is an interacting theory unlike the Maxwell theory. In the present case. We note that the Lagrangian density for the Yang-Mills theory is given by (12. (12.76) .44)) Fµν a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac . the coupling constant can be scaled out and written as an overall multiplicative factor in the Lagrangian density.

80) Let us note that the coeﬃcient matrix of highest derivatives in (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 (12.77) as 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a . i 0 0 i a Fij a = ∂i Aa − ∂j Aa − gf abc Ab Ac = −ǫijk Bk .2 Canonical quantization of Yang-Mills theory 503 As in the case of the Abelian theory (Maxwell theory). F µν ] = 0. ∂∂µ Aa ∂Aa ν ν ∂µ or. (12. we can write the Euler-Lagrange equations as Dµ F µν = ∂µ F µν + ig[Aµ .77) The Lagrangian density (12. µ (12.79) where the covariant derivative is deﬁned to be in the adjoint representation of the group (see (12. −∂µ F µν a − −gf abc Ab F µν c = 0.12.72)).67) and (12. Therefore. we can identify the non-Abelian electric and magnetic ﬁelds from (12.75) and have the form ∂L ∂L − = 0. In the matrix notation (see (12.76) as a a F0i = ∂0 Aa − ∂i Aa − gf abc Ab Ac = Ei .75) can also be written explicitly in terms of the non-Abelian electric and magnetic ﬁeld strength tensors (12. j i i j (12. or. 2 2 4 LYM = Tr Ei Ei − (12.75) is the transverse projection operator just like in the case of Maxwell’s theory.68) and we are suppressing the symbol “(adj)” to denote the covariant derivative in the adjoint representation for simplicity) .68)).78) The Euler-Lagrange equation for the theory are obtained from (12. µ (Dµ F µν )a = ∂µ F µν a − gf abc Ab F µν c = 0.

namely. as in the case of Maxwell’s theory and a Πi a (x) = −F 0i a = Ei (x).82) which is a constraint relation.84) (12. requiring that . Let us obtain the momenta canonically conjugate to the ﬁeld variables Aa from (12. (12.77). the constrained structure of the theory is already obvious in the Euler-Lagrange equation of motion (12.81) In fact. ˙ a (x) ∂ Aµ (12. ν.83) a Noting that the ﬁeld strength Fµν is antisymmetric in the indices µ.75) and this leads to µ Πµ a (x) = ∂L = −F 0µ a (x). Namely. (12.86) 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. Di F i0 a = 0. we choose Aa (x) = 0. A0 (x) = 0. This implies that 0 Aa is like a c-number quantity which commutes with every other 0 operator in the theory.79) for ν = 0. we see that we have only 3N canonical momenta where N = dim G. The momentum conjugate to Aa does not exist. we have Π0 a (x) = −F 00 a (x) = 0.504 12 Yang-Mills theory δAa = µ 1 (Dµ ǫ(x))a . 0 or. (12. For example. Thus. Thus we can choose a gauge condition and set it equal to zero. g (12.85) a where the non-Abelian electric ﬁeld Ei (x) is deﬁned in (12.

i or.77) takes the form a ˙ Ei = Aa .82) that Di Ei = ∂i Ei + ig[Ai . (12.89) From our discussion in the case of the Maxwell theory. let us also note that the potential term in (12.87) we can determine the parameter of transformation and we see that a gauge ﬁeld can always be transformed to satisfy (12. ig (12.86) the Lagrangian density (12. Ei ] = 0.2 Canonical quantization of Yang-Mills theory 505 A′ = U A0 U −1 − 0 or.12.86). (12.88) 2 2 4 LYM = Tr Ei Ei − where we have used the fact that in the gauge (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 . (12.91) can be inverted to give .88) (namely.91) ig ig so that there is a cyclic variable in the theory that needs to be separated. Let us note that with the gauge condition (12. However.86) the non-Abelian electric ﬁeld in (12.90) so that neither the gauge ﬁeld Ai nor the electric ﬁeld Ei is transverse ˙ (recall that Ei = Ai ). We note that the relation (12. 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 . 1 (∂0 U ) U −1 = 0. U −1 ∂0 U = igA0 . ˙ Ei = Ai . in the present case. we expect the physical dynamical degrees of freedom of a massless gauge ﬁeld to be transverse. we note from (12. (12. On the other hand.

then we can deﬁne a new ﬁeld variable Ai satisfying the transversality condition (12.72)) Di (A)G = ∂i G + ig[Ai . (12.92) in the intermediate step and the covariant derivative in the adjoint representation is deﬁned for a matrix function G as (see (12.93).94) where we have used (12. G].92) ig ig Keeping in mind that we would like the dynamical variables to be ¯ transverse.95) This shows that if we can ﬁnd a gauge transformation which satisﬁes ¯ (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. namely. let us require the new ﬁeld variable Ai to be transverse.94).91).93) 1 1 (∂i S) (∂i S −1 ) − ∂i (∂i S)S −1 ig ig 1 ∂i (∂i S)S −1 − [Ai . (∂i S)S −1 ] ig 1 Di (A) (∂i S)S −1 . Imposing this condition in (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 . ig (12. (12. ¯ ∂i Ai = 0. We note next from the ﬁeld redeﬁnition in (12.

98) (12.100) . 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 .99) Furthermore. where (12. decomposing E i into its transverse and longitudinal components as T E i = E i + ∂i φ.86) we can write 1 ˙ ˙ ¯ ¯ Ei = Ai = S Ai − Di (A)f S −1 = SE i S −1 .96) we note that in the gauge (12. ig where we can identify 1 ˙ ¯ ¯ E i = Ai − Di (A)f.12. (12.96) ˙ ¯ where we have used ∂0 S −1 = −S −1 SS −1 . ig (12.95)) with respect to ¯ the gauge ﬁeld Ai and we have identiﬁed ˙ f = S −1 S.97) From (12. Here Di (A) is the covariant derivative in the adjoint representation (see (12.2 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 . ig (12.

E i ]S −1 ig ¯ = S(∂i E i + ig[Ai . E i + ∂i φ ] = 0. We can invert this relation to determine .103) where we have used (12. ¯ ¯ ∂i ∂i φ + ig[Ai . E i ] = 0. or. As a result of this relation. or. E i ] = 0. Ei ] ¯ = ∂i SE i S −1 + ig[S(Ai − 1 −1 S (∂i S))S −1 . ¯ ∂i Di (A)φ = ρ.90) takes the form ¯ ¯ Di (A)E i = ∂i E i + ig[Ai . φ=− (12. (12. E i ])S −1 ¯ = S(Di (A)E i )S −1 .104) T ¯ where we have used ∂i Ai = 0 in the intermediate step. T ¯ ∂i ∂i φ + ig[Ai .102) Let us next note that under the ﬁeld redeﬁnition (12.91) and (12.91). (12. or. φ] + ig[Ai .508 12 Yang-Mills theory ∂i E i = 0. T ∂i E i = ∂i ∂i φ = ∇2 φ. the constraint equation (12.98). E i ] = ρ. 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). (12. ig ∇2 ˙ ¯T ¯ Ei = Ai . φ] = −ig[Ai . Di (A)Ei = ∂i Ei + ig[Ai .99) that 1 1 ¯ ∂i Di (A)f.101) we conclude from the deﬁnition in (12. or. T ¯ ¯ ∂i ∂i φ + ig∂i [Ai .

107) to be Π ia = ∂LYM ˙ ¯ = Aa .100) we obtain T ˙ ¯ E i = E i + ∂i φ = Ai + ∂i (12.91). ∂j Dj (A) (12. ∂i Di (A) 1 ¯ ρ.107) where we have neglected total divergence terms in the intermediate ¯ ¯ steps (note also that because of (12. ¯ 2 ∂j Dj (A) (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 . (12.12. ∂i Di (A) = Di (A)∂i which has been used together with integration by parts).108) . the Lagrangian density (12.105) so that from (12. i ˙ ¯a ∂A i (12.106) Using the deﬁnition in (12.93). ¯ We have isolated the dynamical variables of the theory to be Aa i which are transverse and the conjugate momenta can be determined from (12.2 Canonical quantization of Yang-Mills theory 509 φ = 1 ¯ ρ.98) and the decomposition in (12.106) as well as the cyclicity of trace.

i (12.107) has exactly the same form as in the Abelian theory except for the last term. A′ a = Aa + i i g such that ¯i ∂i A′ a = 0. 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. (12.510 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). Thus the long range behavior of this theory is still not well deﬁned in the Coulomb gauge. therefore.110) then we can make a time independent inﬁnitesimal gauge transformation (under which the theory is invariant) 1 ¯ǫ a ¯ ¯ Di (A)¯ . This is related to the question of gauge ﬁxing.109) with all other brackets vanishing. This can be seen from the ¯ fact that if we have a ﬁeld conﬁguration Aa which is transverse. ¯ Gribov has shown that the operator ∂i Di (A) does possess zero modes or eigenvectors with zero eigenvalue. We note that the Lagrangian density in (12. this last term has the form ρ 1 2 ρ which corresponds to ∇ the long range Coulomb interaction).31).111) (12.e. even in the absence of other matter ﬁelds there is a long range interaction. Here we see that since the gauge ﬁelds are self interacting. i. of the theory) ¯ is.112) . therefore. Π (y)} = δab δi j (x − y). namely the Coulomb gauge does not uniquely deﬁne the gauge ﬁelds. Let us note here that the long range behavior of the interaction (and. controlled by the eigenvalues of the operator ∂i Di (A). Here the transverse delta function corresponds to the one already deﬁned in the case of Maxwell’s theory in (9. i TR jb (12. i ¯ ∂i Aa = 0.

2 Canonical quantization of Yang-Mills theory 511 ¯ provided ǫa corresponds to a zero mode of ∂i Di (A).) g Hence such gauge ﬁeld conﬁgurations can only be reached nonperturbatively. ∂ · A behaves like a free ﬁeld and. Gribov ambiguity is a phenomenon associated with large gauge transformations and is an important issue since it leads to nonperturbative eﬀects. let us modify the theory (12.111) that this has a singularity at g = 0. known as the Gribov ambiguity. However.115) .125) with J µ = 0) 1 1 a L = − Fµν F µν a − (∂µ Aµa )2 . In other words. we would not worry about it within the context of perturbation theory. we can also try to quantize the non-Abelian gauge theory covariantly. consequently. There now exist prescriptions to ¯ take care of this problem. there are serious diﬀerences from Maxwell’s theory. This shows that ¯ the Coulomb gauge does not really specify the gauge ﬁeld conﬁguration uniquely. (12. Thus. As long as we are within perturbation theory we can neglect such conﬁgurations. (It is of the form O 1 . However. occurs in all gauges other than the axial gauge.113) The additional term clearly breaks gauge invariance and. classically we can impose the condition ∂ · A = 0. The canonical quantization of Yang-Mills theories does not have manifest covariance very much like the Maxwell theory.129)) (∂ · A) = 0. for example. For example. much like in the Maxwell theory. Therefore. makes the theory nonsingular. (12. 4 2 (12. therefore. we recall that in QED (see (9. namely. by modifying the theory. in the present case. But let us note from the form of A′ a in i (12. very much along the lines of the Abelian theory that we have discussed in chapter 9. On the other hand. The other way of saying this is that within the framework of perturbation theory we are interested in inﬁnitesimal gauge transformations.12. We can show that this nonuniqueness.114) namely.75) as (see (9. the hypersurface deﬁning the gauge choice intersects the gauge orbits more than once.

(12. it cannot be uniquely decomposed into a positive and a negative frequency part (in a time invariant manner). consequently. since it is not invariant under time evolution. In other words.117) = 0. (12. however.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 . We would see next how we can derive intuition on this important question from the path integral quantization of the theory. Therefore.113) has truly modiﬁed the theory.116) In the case of the Yang-Mills theory. Furthermore. just modifying the Lagrangian density as in (12. we obtain (∂ · Aa ) = gf abc ∂ν Ab F µν µ c (12. we need to analyze the question of modifying the theory in a more systematic and detailed manner. the additional term in (12.) Correspondingly the naive analog of the Gupta-Bleuler condition for non-Abelian gauge theories does not seem to exist. (Namely. 12. µ Contracting with ∂ν .118) Thus. even classically the equations of motion are given by ∂µ F µν a − gf abc Ab F µν c + ∂ ν (∂ · Aa ) = 0. the physical subspace would keep changing with time which is not desirable.512 12 Yang-Mills theory which then translates to the Gupta-Bleuler condition on the physical states ∂ · A(+) |phys = 0. nor can we think of the supplementary condition ∂ · Aa(+) |phys = 0.119) in a physically meaningful manner. we note that (∂ · Aa ) does not behave like a free ﬁeld and. (12.113) does not seem to be suﬃcient in contrast to the Abelian gauge theory. since (∂ · Aa ) is not a free ﬁeld. in contrast to the Abelian theory.

4 for the deﬁnitions of “in” and “out” states) in the presence of interactions. Like the derivative of a function. For example. we will not go into the systematic details of this description.120) where N is a normalization constant. we will only go over some of the essential concepts involved in such a description.3 Path integral quantization of gauge theories 513 gauge theories. for a real scalar ﬁeld interacting only with an external source. (12.12. .122) and various Green’s functions (n-point functions) are obtained from the generating functionals Z[J] and W [J] through functional diﬀerentiation.121) Furthermore. Since this is a complete topic in itself. 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. 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] . The functional dependence of a quantity on a variable is generally denoted by a square bracket. from the “in” vacuum to the “out” vacuum. 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 . Z[J] and W [J] are known as generating functionals for the Green’s functions of the theory. rather. 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. 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. see section 6. 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 . 2 (12. δφ(y) ǫ ǫ→0+ (12.

120) is known as a functional integration and is deﬁned as follows. namely. (12. the functional integration is deﬁned (up to a normalization constant) as Dφ = dφi . (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.514 12 Yang-Mills theory The integration in (12. Let us divide the entire space-time into inﬁnitesimal cells labelled by “i” of volume δVi and deﬁne 1 δVi d4 x φ(x). Basically.120) can be evaluated in a straightforward manner. i (12. 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) . G = (−∂µ ∂ µ − m2 )−1 . When we have an . (12.123) Clearly in the limit of vanishing volume we recover lim φi = φ(x).126) is independent of ﬁeld variables and is a constant (possibly divergent) and can be incorporated into the normalization constant N .125) With these basics a simple functional integral for a quadratic action such as in (12. δVi φi = (12.126) where the Green’s function G(x − y) is formally the inverse of the operator in the quadratic part of the action.) The determinant in (12.124) δVi →0 and in terms of these discretized ﬁeld variables.

The advantage of using the path integral description lies in the fact that there are no operators in (12.3 Path integral quantization of gauge theories 515 interacting theory (not quadratic in the ﬁeld variables) the functional integral.132) where we have used the notation in (12.128) where J µ represents a conserved current (source).P −1 J ν µν ).131). This is a Gaussian functional integral and we can evaluate this using our earlier result in (12. (12. 4 (12. everything is a classical function.120). However. Aµ ) = − ∂ µ ∂ ν ) δ4 (x − y).122)) we have identiﬁed P µν (x − y) = (η µν (J µ .126) Z [Jµ ] = eiW [Jµ ] = N (det (−P µν ))− 2 e− 2 (J 1 i µ . let us go back to the Maxwell theory 1 L(J) = − Fµν F µν + J µ Aµ . (12. cannot be obtained in a closed form. In this case.124)).131) d4 x J µ (x)Aµ (x).12.123) and (9.P 1 ] . In this case. as we have seen before (see (9.A µ) DAµ ei[ 2 (Aµ . in general. the generating functional in the path integral formalism is given by (J ) [A Z [Jµ ] = eiW [Jµ ] = N = N DAµ eiS µ] (12. With this brief introduction to path integral description. the operator P µν is a projection .130) where N is a normalization constant and (see (9.129) µν A ν )+(J µ . (12. 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.

the determinant of P µν vanishes. the generating functional. ie Aµ → A(θ) = U Aµ U −1 − µ where U (θ) = e−iθ(x) . Therefore. (12.133) in terms of U (θ) is quite general and holds for the non-Abelian theories as well. (As we have seen the form of transformation for the gauge ﬁelds (12. This implies that the generating functional does not exist in this case. on the other hand. The longitudinal vectors kµ (or ∂µ ) are its eigenvectors with zero eigenvalue.134) We can immediately see that for a U (1) gauge group. The Lagrangian density for Maxwell’s theory is invariant under the gauge transformation 1 (∂µ U )U −1 . Clearly therefore.136) .) The source of the diﬃculty is not hard to see. (The operator possesses zero modes and. The action S. x (12. e Aµ → A(θ) = Aµ + µ (12. the inverse of the matrix cannot be deﬁned either. we can write the transformation as 1 ∂µ θ(x).133) (12.135) which is the familiar gauge transformation for Maxwell’s theory. consequently. is proportional to the “volume” of the orbits denoted by dθ(x).516 12 Yang-Mills theory operator for transverse photons. is constant on such orbits. 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. even in the absence of any sources. This is a U (1) symmetry and the symmetry is noncompact since the parameter of transformation θ(x) can take any real value.) (θ) For a ﬁxed Aµ .

the Coulomb gauge. (12. A3 (x) = 0. gauge invariant and do not depend on the choice of the hypersurface (gauge). this should be replaced by the group invariant Haar measure x dU (x).139) is known as the gauge condition (or gauge ﬁxing condition). Thus. we pick up a representative gauge ﬁeld from each gauge orbit.3 Path integral quantization of gauge theories 517 (In the non-Abelian case. of course.) 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.138) has a unique solution for some θ(x). 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. for example. then even if Aµ does not satisfy the condition. Physical quantities are. we can ﬁnd a gauge trans(θ) formed Aµ which does and F [A(θ) (x)] = 0. (12.140) and so on. Rather we should integrate over each orbit only once. if F [Aµ (x)] = 0. In this way. We recognize that we should not integrate over all gauge ﬁeld conﬁgurations because they are not really distinct. A0 (x) = 0.137) deﬁnes the hypersurface which intersects the gauge orbits once. the Lorentz/Landau gauge. (12. ∇ · A(x) = 0. the axial gauge. The way this is carried out is by choosing a hypersurface which intersects each orbit only once. namely.12. This procedure is known as gauge ﬁxing and the condition F [Aµ (x)] = 0. µ (12. the temporal gauge. We can . here are a few of the familiar gauge ﬁxing conditions F [Aµ (x)] = ∂µ Aµ (x) = 0.

for example. gauge dependent (recall.142) which can. Thus we have to ﬁx a gauge without which even the Cauchy initial value problem cannot be solved.143) This follows from the fact that the measure in the group space is invariant under a gauge transformation. (12. (Only physical quantities need to be gauge independent. we know from ordinary perturbation theory that the Green’s functions are. On the other hand.) Note that the quantity ∆FP [Aµ ]. in general. is gauge invariant which can be (θ ′ ) seen as follows. Let us deﬁne ∆−1 [Aµ ] = FP x dθ(x) δ F [A(θ) (x)] . the photon propagator) although the physical S-matrix (the scattering matrix) elements are gauge independent.141).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). be thought of as a completeness relation. 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 ] . let us do the following trick due to Faddeev and Popov. Therefore. therefore. (12. µ (12.) To extract out the inﬁnite gauge volume factor. (The integration measure should be dU (x) which is essential in the case of non-Abelian theories. That is (for the Abelian . Let us make a gauge transformation Aµ → Aµ in (12.141) This can also be written as ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] = 1. known as the Faddeev-Popov determinant. it would lead only to gauge invariant Green’s functions.

144) In the non-Abelian case. let us make an inverse gauge transformation Aµ → A(−θ) .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] . (12. However. d(U U ′ ) = dU.12.147) 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 .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. µ under which the generating functional takes the form dθ(x) δ F [Aµ (x)] eiS x (J ) [A µ] (J ) [A µ] (12. this gives the correct functional form for the generating functional. dθ(x) = (12. To do this let us note from (12. we should have the Haar measure which is gauge invariant.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. Therefore. (12.146) Furthermore.141) that . namely.

) 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). therefore. Thus.151) θ=0 . (12. be thought of as the Jacobian that goes with a given gauge choice. we have θ(x) = 0.149)-(12.152) The Faddeev-Popov determinant can. for gauge ﬁelds which satisfy the condition F [Aµ (x)] = 0. On the other hand.153) . 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).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 (12.149) δθ δF F [Aµ (x)]=0 (θ) . (12.150) (12. (Here we can completely ignore the problem of Gribov ambiguity associated with large gauge transformations. 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.151) we determine δF [Aµ (x)] ∆FP [Aµ ] = det δθ(y) (θ) (12. using (12.

We note further that since δF [Aµ (x)] δθ(y) (θ) ∆FP [Aµ ] = det θ=0 . we can now do the following (also known as the ’t Hooft trick).157) and ξ is known as the gauge ﬁxing parameter. (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] . (12. (12.156) where we have deﬁned a gauge ﬁxing action as SGF = d4 x LGF = − d4 x (F [Aµ (x)])2 . Hence we can multiply the generating functional by a weight factor and integrate over all f (x). Then we can insert the identity ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] − f (x) = 1. The Faddeev-Popov determinant is unchanged by this modiﬁcation because f (x) does not depend on Aµ (x).12. Thus. 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 ). (12.155) Following ’t Hooft.158) . We note that physical quantities are independent of f (x). (12.3 Path integral quantization of gauge theories 521 where f (x) is independent of Aµ .154) into the functional integral.

to write the Faddeev-Popov determinant in the form of a ghost action (action involving ghost ﬁelds). c(y)]+ = 0. [c(x). known as ghost ﬁelds. graphs involving these ﬁctitious particles in closed loops must have an additional (−1) factor just like the fermions.. [c(x). (12. These ﬁelds are known as Faddeev-Popov ghosts and as is obvious from their anticommutation relations.162) . they obey anticommutation rules. 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.e.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. but opposite statistics. (12. δF [Aθ (x)] µ d xd y c(x) δθ(y) 4 4 (12. c(y)]+ = 0. c(y)]+ = 0.159) θ=0 c(y).161) Thus although these ﬁelds behave as scalar ﬁelds under Lorentz transformations. Thus the generating functional now takes the form Z[Jµ ] = eiW [Jµ] = N DAµ Dc Dc eiSeﬀ (J ) [Aµ .c] .( 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 .c. i. [c(x).).160) Here we have introduced two independent ﬁctitious ﬁelds c(x) and c(x).

c] . the gauge ﬁxing Lagrangian density (see (12. of course. the gauge ﬁxing Lagrangian density should make the quadratic part of the eﬀective Lagrangian density nonsingular. (J) (J) (12.164) In this case.12.157)) has the form LGF = − 1 1 (F [Aµ (x)])2 = − (∂µ Aµ (x))2 . for the moment let us emphasize that the purpose of a gauge ﬁxing Lagrangian density is to break gauge invariance. for example. the covariant condition (see (12. this modiﬁes the starting theory. our responsibility to show that the physical interpretation of the theory has not changed. (12. It is worth noting at this point that we have modiﬁed our starting Lagrangian density by a series of formal manipulations.153)) F [Aµ (x)] = ∂µ Aµ (x) = f (x).165) . 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. It is.3 Path integral quantization of gauge theories 523 where Seﬀ [Aµ . Otherwise it fails its purpose and will not correspond to an acceptable gauge ﬁxing condition. 2ξ 2ξ (12. However. c. To do covariant perturbation theory in the path integral formalism.142) that both these actions arose from inserting a factor of unity into the functional integral). Of course. Let us now look at a simple gauge ﬁxing condition in Maxwell’s theory.163) Thus we can summarize what we have done so far. c. We would show this in the next chapter. 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. namely. 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. c] = S (J) [Aµ ] + SGF + Sghost = d4 x Leﬀ [Aµ .

(12. 4 2ξ (12.524 12 Yang-Mills theory It is clear that this provides longitudinal components to the quadratic terms in ﬁelds and hence breaks gauge invariance. the ghost ﬁelds are interacting and have to be present.125).169) We note here that the ghost ﬁelds are noninteracting in the case of the Maxwell theory in ﬂat space-time and. we note that F [A(θ) (x)] = ∂µ A(θ)µ (x) = ∂µ Aµ (x) + µ so that (see (12. the ghost action (12. we may neglect them and then for ξ = 1 we recognize that our eﬀective Lagrangian density (12.152)) δF [Aµ (x)] δθ(y) (θ) 1 µ ∂ θ(x) . e (12. 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 . however.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 .167) Absorbing the factor 1 into the normalization factor (alternatively e scaling the ghost ﬁelds).168) d4 x ∂µ c(x)∂ µ c(x) = where we have neglected total divergence terms. This . therefore. To obtain the corresponding ghost Lagrangian density for this gauge choice.166) = θ=0 1 e xδ 4 (x − y). (12.169) is nothing other than Maxwell’s theory in the Feynman-Fermi gauge (9. In non-Abelian gauge theories.

171) . On the other hand. One way of looking at the ghost ﬁelds is as if they are there to subtract out the unphysical ﬁeld degrees of freedom. the ghost ﬁelds automatically couple to the geometry also.169) and ask how we can recover the Lorentz gauge from the Lorentz like covariant gauge condition in (12. we cannot neglect them even if they are noninteracting particularly when we are doing calculations at ﬁnite temperature. (The ghost degrees of freedom subtract because they have the unphysical statistics. (12.164). namely. 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. has only one ﬁeld degree of freedom (we will discuss the question of hermiticity of the ghost ﬁelds in the next chapter).) Let us now go back to the eﬀective Lagrangian density (12. Furthermore. To do this let us rewrite the eﬀective Lagrangian density (12.169).170) 4 2 Here we have introduced an auxiliary ﬁeld F which does not have any dynamics. This naive counting works pretty well as we will see later.169) as ξ 1 (J) Leﬀ = − Fµν F µν + F 2 −F (∂µ Aµ )+∂µ c∂ µ c+J µ Aµ . the Aµ ﬁeld has four ﬁeld degrees of freedom. It is also true that when Maxwell’s theory is coupled to a gravitational ﬁeld.3 Path integral quantization of gauge theories 525 explains why the naive modiﬁcation in (12. 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. the ghost Lagrangian density was missing). we see that the equation of motion for F is given by ξF (x) = ∂µ Aµ (x). On the other hand each of the ghost ﬁelds. (12. For example. they anticommute even though they are scalar ﬁelds.12.113) of the non-Abelian gauge theory failed to be suﬃcient unlike in the Maxwell theory (namely. Hence omitting the ghost Lagrangian density in such a case would lead to incorrect results. since the ghost ﬁelds and the ghost action are really necessary for the unitarity of the S-matrix. being a scalar.

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

In particular. The µ correct description for the generating functional. is given by (J ) [AT ] µ Z[Jµ ] = N = N DAT eiS µ DAµ ∆FP [Aµ ]δ(∂ · A(x)) eiS (J ) [A µ] . In the path integral formalism. However. 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 summary. their contributions cancel out and eﬀectively we are left with only two physical transverse degrees of freedom. non-invertible. as the Lagrangian density describing the . in this formalism there is no reference to the Hilbert space of the theory. The second method of quantizing the Abelian gauge theory is through the method of path integrals. 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. according to Faddeev and Popov. the ghost contributions cancel those from the unphysical gauge ﬁeld degrees of freedom). 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] .3 Path integral quantization of gauge theories 527 result. (12. As a result. 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. we note that gauge invariance puts a very strong constraint on the structure of the Lagrangian density for the gauge ﬁeld. the coeﬃcient matrix of the quadratic terms in the Lagrangian density is singular and.176) In this way. therefore.12. In this formalism the ﬁeld variables are treated as classical variables. (12.175) where AT denotes the transverse physical degrees of freedom. if we take Linv .

the Lagrangian density for the gauge ﬁelds is given by (see (12.75)) 1 a Linv = − Fµν F µνa . then we cannot deﬁne propagators and the entire philosophy of doing perturbative calculations with Feynman diagrams breaks down. in the non-Abelian gauge theory. g (ǫ)a Aa (x) → Aµ (x) = Aa (x) + µ µ (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. In order to circumvent this diﬃculty.76)) Dµ ǫa (x) = ∂µ ǫa (x) − gf abc Ab (x)ǫc (x). 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. The standard covariant gauge ﬁxing (12. Let us now apply these ideas to the study of the non-Abelian gauge theory. 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.178) where ǫa (x) is the inﬁnitesimal parameter of transformation.179) ν µ µ ν with g denoting the coupling constant of the theory.177) which is invariant under the inﬁnitesimal gauge transformation 1 Dµ ǫa (x). 4 (12. 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. consists of adding to the invariant Lagrangian density a gauge ﬁxing Lagrangian density of the form . in general.67) and (12. On the other hand. (12. µ a Fµν (x) = ∂µ Aa (x) − ∂ν Aa (x) − gf abc Ab (x)Ac (x).528 12 Yang-Mills theory dynamics of the gauge ﬁeld theory. As we have seen.164).

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).180) which corresponds to a gauge ﬁxing condition of the form F a [Aµ (x)] = ∂µ Aµa (x) = f a (x).122)) δF a [Aµ (x)] δǫb (y) (ǫ) (12. we can write 1 (ǫ)a F a [A(ǫ) (x)] = ∂ µ Aµ (x) = ∂ µ Aa (x) + Dµ ǫa (x) . 2ξ (12.181) Here ξ represents a real arbitrary constant parameter known as the gauge ﬁxing parameter. (12.184) µ 1 Consequently. δǫb (y) (ǫ) (12.3 Path integral quantization of gauge theories 529 LGF = − 1 (∂µ Aµa (x))2 . we can write the ghost action corresponding to this gauge choice as (see (12.160) and note that since ǫa is an inﬁnitesimal parameter.183) = 1 µ ab 4 ∂ D δ (x − y) g x xµ µ = ∂x ∂xµ δab − gf acb Ac (x) δ4 (x − y).181) that we are using. (12. Following the prescription of Faddeev and Popov. 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 .182) We note that for the covariant gauge choice (12.12. µ µ g so that we have (see (12.

With all these modiﬁcations. in the path integral formulation. In a deeper sense. let us consider the gauge theory of the antisymmetric tensor ﬁeld Aµν (x) . (12. the gauge ﬁxing and the ghost Lagrangian densities.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. merely correspond to a multiplicative factor of unity (see (12. cannot be neglected even in ﬂat space-time. (12.186) = − Fµν F µνa − 4 2ξ We note that the ghost ﬁelds in the present case are interacting unlike in Maxwell’s theory and. 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 . As we have mentioned earlier. in some cases we have to work through the details of this analysis rather carefully in order to obtain the correct result. However. 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 . As an example of how we should be careful in carrying out the Faddeev-Popov analysis in complicated gauge theories.185) where we have dropped total derivative terms (surface terms). 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. 12.142)) which does not change the physical content of the theory.

since Fµνλ is completely antisymmetric in all the indices. ˙ ∂ Aµν (x) Πµν (x) = (12. 2. However.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. we conclude that . the canonical momenta are antisymmetric (like the ﬁeld variables).188) where the ﬁeld strength tensor corresponds to the totally antisymmetric tensor Fµνλ = ∂[µ A νλ] = ∂µ Aνλ + ∂ν Aλµ + ∂λ Aµν .191) The independent ﬁeld variables of the theory are A0i . 1.12. (12. ν = 0.192) namely. Let us consider only the free Lagrangian density for this ﬁeld deﬁned by 1 L = − Fµνλ F µνλ . we also note that this Lagrangian density is invariant under the gauge transformation δAµν (x) = ∂µ θν (x) − ∂ν θµ (x). Aij and we note that Πµν = −Πνµ = −F 0µν .190) so that not all ﬁeld variables are independent. 3. This can be seen from the deﬁnition of the conjugate momenta ∂L = −F 0µν (x). (12.189) Naive counting shows that Aµν has six ﬁeld degrees of freedom. Furthermore. 6 (12. (12. µ. (12.4 Path integral quantization of tensor fields 531 Aµν (x) = −Aνµ (x).

197) so that we can impose a Coulomb gauge condition of the form ∂i Aij (x) = 0. 2. Π0µ = −F 00µ = 0. k = 1. 3. on closer inspection we notice that the Coulomb gauge condition (12. i = 1. (12.194) (12. With these conditions we have ˙ Πij = −F0ij = −Aij . and the Lagrangian density (12.532 12 Yang-Mills theory or.193) Since these momenta identically vanish.198) This would seem like three constraints and hence we would naively conclude that this theory has no dynamical degrees of freedom. We can see this by writing out the gauge condition explicitly . However. i. (12. Π0i = 0. 3. 2. the corresponding ﬁeld variables are like c-number quantities and we can choose the gauge condition A0i (x) = 0. 6 (12.195) (12. However.198) actually represents only two independent conditions.196) Thus it would seem that the theory has truly three degrees of freedom.188) takes the form 1 1 L = − Fµνλ F µνλ = − 3F0ij F 0ij + Fijk F ijk 6 6 1 = − 3Πij Πij + Fijk F ijk . j. (12. we note that the theory still possesses a residual (static) gauge invariance under the transformation δAij = ∂i θj (x) − ∂j θi (x).

massless real scalar ﬁeld describing a single degree of freedom.12. 2.200) It is now clear that any two of the three conditions lead to the third condition so that there are only two independent conditions. The fact that the theory has only one degree of freedom can also be seen in the following manner.202) where φ(x) represents a real scalar ﬁeld. Thus.199) ∂2 A21 + ∂3 A31 = 0. 3 are (12.203) We recognize this as the Lagrangian density for a free.189)). . the three conditions corresponding to j = 1.4 Path integral quantization of tensor fields 533 ∂1 A1j + ∂2 A2j + ∂3 A3j = 0. ∂1 A13 + ∂2 A23 = 0.201) 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. In this case. the Lagrangiandensity (12. 2 (12. (12. ∂1 A12 + ∂3 A32 = 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. 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). 12 2 (12. ∂ρ Fµνλ − ∂λ Fρµν + ∂ν Fλρµ − ∂µ Fνλρ = 0. Since Fµνλ represent a totally antisymmetric third rank tensor in four space-time dimensions satisfying the Bianchi identity (see (12.

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).534 12 Yang-Mills theory Let us now look at the path integral quantization of this theory. Hence to write the determinant as an action we need ghost ﬁelds which carry a vector index and we have .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)). (12.208) This is a matrix with two vector indices. so that we have (12. 2ξ LGF = − (12.205) corresponding to the choice of gauge condition F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). 6 L=− (12. the starting gauge invariant Lagrangian density has the form 1 Fµνλ F µνλ . (12.204) and we add to it the covariant gauge ﬁxing Lagrangian density (ξ is the gauge ﬁxing parameter) 1 (∂µ Aµν )2 . According to our earlier discussions.

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.209) Therefore. 2 (12.4 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. Thus the eﬀective Lagrangian density appears to have 6 − 2 × 4 = 6 − 8 = −2. The ﬁeld variable Aµν has six degrees of freedom.212) . Let us start with the generating functional (J ) [A Z [Jµν ] = N DAµν eiS µν ] . (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 ) . they subtract out ﬁeld degrees of freedom.211) eﬀective ﬁeld degrees of freedom. 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.12. (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)] . (12.209) 1 2 ). we introduce the identity element as (see (12. 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. F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x).216) Sghost = d4 x (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) .217) In the naive application of the quantization procedure. (12.218) . (θ) ∆FP [Aµν ] = det = det ( (12.152)). as we have seen (see (12. (12.213) Therefore. δF ν [Aµλ (x)] δθ σ (y) ν x δσ θ=0 ν − ∂x ∂xσ ) δ4 (x − y) .154)) (θ) ∆FP [Aµν ] x dθ σ (x)δ F ν [Aµλ (x)] − f ν (x) = 1.214) where.536 12 Yang-Mills theory The gauge symmetry allows us to choose a gauge condition and we have chosen a Lorentz like gauge condition. (12.215) Thus inserting this identity element into the functional integral.

221) where P (x − y) is the normalized transverse projection operator deﬁned earlier (see.224) so that we can write . (12. we note that the gauge condition ∂µ Aµν (x) = f ν (x). implies that ∂ν ∂µ Aµν = ∂ν f ν (x) = 0. 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) . (12.12.223) 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). (12.220) (12.221) has a gauge invariance (just like Maxwell’s theory) of the form δfµ (x) = ∂µ θ(x).123)) as µν µ ν ∂x ∂x x P (x − y) = η µν − δ4 (x − y). for example. (9. (12. the action for fµ in (12.4 Path integral quantization of tensor fields 537 However.219) That is.222) We see that because of the transverse projection operator. 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. (12.

remembering that (see (12. (12.f µ )− 2 (c. c) i µλ ]. c) = − 2 2 d4 x c(x) c(x).228) det xδ 4 (x − y) 1 2 Dc(x) eiS ghost . (12.F (fµ )) i µν fν ) = Df µδ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ . (12. (12.225) where δF [fµ (x)] δθ(y) (θ) 1 2 ∆FP [fµ ] = det θ=0 = det xδ 4 (x−y) . c) .f i µ) ∆FP [fµ ].f ) Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .226)) ∆FP [fµ ] = = we determine 1 1 S ghost = − (c.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 .229) .538 12 Yang-Mills theory ∆FP [fµ ] x (θ) dθ(x)δ F [fµ (x)] − f (x) = 1.P × = i µν fν ) Df ∆FP [fµ ] δ F [fµ (x)] − f (x) e i i − 2ξ (f. Substituting this into (12.Fν [Aµλ ])− 2 (c.227) Furthermore.P ×∆FP [fµ ] e− 2ξ (F (fµ ). (12.226) Thus using the ’t Hooft trick we can write Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .

4 Path integral quantization of tensor fields 539 Note here that the ﬁeld c(x) is a real anticommuting scalar ﬁeld. The ghosts of the present form are known as Nielsen ghosts. .F µ[Aλσ ]) J +S GF +Sghost +S ghost ).233) This is again not right and the agreement with canonical quantization seems to be worse than before. may be neglected. (12. 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. Thus using this modiﬁed ’t Hooft weighting factor.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. (12. (12. but in the presence of gravitation it cannot be written as a total divergence and is quite relevant.12. 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.230) SGF = − = − Therefore. (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 .231) DAµν Dcµ Dcν Dc ei(S DAµν Dcµ Dcν Dc ei(S J +S ghost +S ghost i ) e− 2ξ (Fµ [Aλσ ].

237) We are now ready to calculate the Faddeev-Popov determinants associated with these gauge ﬁxing conditions.236) Let us choose for simplicity Lorentz like gauge conditions F and F = ∂µ cµ (x) =f (x). (12. (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. µ (12.540 12 Yang-Mills theory We notice at this point that although we have ﬁxed up the gauge invariance of the gauge ﬁxing Lagrangian density. and δcµ (x) = ∂µ λ(x).238) = ∂µ cµ (x) = f (x).234) where λ(x) and λ(x) are anticommuting scalar parameters. namely. λ=0 . (12.235) and ∆FP [cµ ] x dλ(x) δ F [c(λ) (x)] −f (x) = 1. the ghost Lagrangian density also possesses a gauge invariance.239) (λ) = det δF [cµ (x)] δλ(y) . under δcµ (x) = ∂µ λ(x). the ghost Lagrangian density Lghost is invariant. (12. We can again use the FaddeevPopov trick and write ∆FP [cµ ] x (λ) dλ(x) δ F [cµ (x)] − f (x) = 1.

243) since the exponents vanish (because of the fermionic nature of the variables). (12.241) We note here that since we are writing an inverse determinant as an action. . the ghost ﬁelds c and c behave like commuting scalars. e .238) correspond to fermionic conditions.240) where we have used the fact that for anticommuting variables. in the present case we can write 4 −1 ∆FP [cµ ] = = det xδ (x − y) c e) c Dc Dc e−i(e. namely. the Faddeev-Popov term is an inverse determinant and. f and f are Grassmann (fermionic) functions. the Jacobian for a change of variables is the inverse of the determinant. the appropriate weight factor in this case would correspond to . consequently. e i − 2χ R e e e e d4 x f (x) f (x) .237) and (12. (12.4 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) . the ’t Hooft trick needs to be carried out rather carefully. (12. Namely. we cannot use weight factors of the forms i − 2χ e R e e d4 x f (x)f (x) . Therefore. We note that contrary to the usual case.12. Similarly we can show that 4 −1 ∆FP [cµ ] = = det xδ (x − y) Dc Dc e ” “ e c c e −i e. (12. Rather.242) The gauge conditions (12.

cµ : −4 degrees of freedom. cµ : −4 degrees of freedom.542 12 Yang-Mills theory e i − 2χ R e e e d4 x f (x) f (x) . c : 1 degrees of freedom.247) We are now ready to count the number of eﬀective ﬁeld degrees of freedom in the theory. c. c : 1 degrees of freedom.248) The ghosts c.245) d4 x Leﬀ . c and c being commuting scalars contribute positively to the counting of the number of degrees of freedom. anticommuting. The other . (12.244) the generating functional takes the form Z [Jµν ] = N where Seﬀ and Leﬀ = DAµν Dcµ Dcν DcDc Dc D c D c eiSeﬀ .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. c : −1 degrees of freedom. (12. c : 1 degrees of freedom. Aµν : 6 degrees of freedom. (12. 2χ c (12. commuting. (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.

2. Siegel. 1 (1970). R. World Scientiﬁc. Physics Letters 25B. Feynman. Quantum Field Theory. (12. Physical Review 96. McGrawHill. Qualks. L. K.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 . A. Huang.5 References 1. C. Mills. Faddeev. N. Quantum Mechanics and Path Integrals. Theoretical and Mathematical Physics 1. ’t Hooft and M. N. 4. 8. Hibbs and R. D. 1 (1978).249) degree of freedom. McGraw-Hill. Yang and R. Leptons and Gauge Fields. 9. Diagrammar. 29 (1967). Nuclear Physics B139. P. R. Thus counting the degrees of freedom. 12. 3. we see that eﬀectively the theory has 6 − 2 × 4 − 1 + 4 × 1 = 6 − 8 − 1 + 4 = 1. D. 191 (1954). Popov. 5. 7. Gribov. Physics Letters 93B. 11. Utiyama. 170 (1980). Nuclear Physics B44. N. ’t Hooft and M.12. 189 (1972). W. Veltman. Itzykson and J-B. . 10. Veltman. Physical Review 101. 378 (1971). N. Faddeev and V. Zuber. 1980. 12. R. G. 6. V. L. C. Mohapatra. G. CERN preprint (1973). L. Physical Review D4. New York (1965). New York. Singapore (1982). 1597 (1956). This matches exactly with the counting of the degrees of freedom from the canonical quantization.

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

does not have the gauge invariance (12.34) of the original theory.1) 2 + ∂ µ ca (Dµ c)a . µ (13. therefore.181) is given by 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ (13.67)) (Dµ c)a = ∂µ ca − gf abc Ab cc .186).3) where g denotes the coupling constant of the theory.2) is invariant under the global transformations 545 .Chapter 13 BRST invariance and its consequences 13. However.1 BRST symmetry As we have seen in (12. develops a global fermionic symmetry which. It is easy to check that the total Lagrangian density (13. The total Lagrangian density has been gauge ﬁxed and. 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. with gauge ﬁxing and ghost terms. (13. the total Lagrangian density for the non-Abelian gauge theory (Yang-Mills theory) has the form LTOT = Linv + LGF + Lghost . the total Lagrangian density. in some sense. remembers the gauge invariance of the original theory.

2 respectively. only when the ghost equation of motion is used). (13.7) for the ﬁelds φa = Aa . We note that the nilpotency of the transformations holds oﬀ-shell only for the ﬁelds Aa . g (13. This shows that under two successive transformations of the kind (13.546 13 BRST invariance and its consequences δAa = µ δca δca ω (Dµ c)a . 2 ω ∂ µ Aa . g ω abc b c = f cc.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 = 0.6) when the ghost equation of motion is used. Similarly.4) we have δ2 δ1 φa = 0.50)). ca . ca independent of the parameters of transµ formations where δ1.5) Here we have used the Jacobi identity for the symmetry algebra (see (12. we obtain δ ∂ µ Aa = µ ω µ ∂ Dµ ca = 0.2 correspond to transformations with the parameters ω1. while for ca it is µ true only on-shell (namely. 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. = 6 (13. The invariance of the Lagrangian density (13.4) where ω is an arbitrary anti-commuting constant parameter of the global transformations. ca . . = − µ gξ (13.

4) can now be easily checked. we have used the fact that δ (Dµ c)a = 0 which we have seen in (13. we note that the transformation for Aa can really be thought of as an inﬁnitesiµ mal gauge transformation (see (12. however. The present formulation of the BRST symmetry. First.1 BRST symmetry 547 The invariance of the Lagrangian density (13. Generally. (13.170) we can write the gauge ﬁxing Lagrangian density by introducing an auxiliary ﬁeld.2) is invariant under the global transformations (13.2) under the transformations (13. the invariant Lagrangian density is trivially invariant under these transformations.9) so that the action is invariant. namely.8) Consequently. is slightly unpleasant in the sense that the nilpotency of the anti-ghost ﬁeld transformation holds only on-shell.4) with an anti-commuting constant parameter.2) can also be written equivalently as . therefore. let us recall that the gauge ﬁxing Lagrangian density in (13. As we have seen earlier within the context of the Abelian gauge theory in (12.13. Since this will also be quite useful for our latter discussions. This shows that the action for the total Lagrangian density (13.5). In this derivation.34)) with the parameter ǫa (x) = ωca (x) and. δLinv = 0. ν gξ (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 . 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). 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.

δ2 δ1 φa = 0. (13. 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 . 2 ω = − F a. (13.548 13 BRST invariance and its consequences LGF = ξ a a F F + (∂ µ F a )Aa . µ (13. and it is straightforward to check that these transformations are nilpotent oﬀ-shell. but it is this form that is very useful as we will see.10) that the equation of motion for the auxiliary ﬁeld takes the form ξF a = ∂ µ Aa . we recover the original gauge ﬁxing Lagrangian density (up to a total divergence term).4) take the form δAa = µ δca δca ω (Dµ c)a .10) where F a is an auxiliary ﬁeld.14) . 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 = 0.11) and when we eliminate F a from the Lagrangian density using this equation. µ 2 (13. g ω abc b c = f cc.) It is clear from the form of LGF in (13.12) µ 4 2 In this case.10) diﬀers from that in (12. namely. (The form of the gauge ﬁxing Lagrangian density in (13. g (13. the BRST transformations in (13.170) by a total divergence.13) δF a = 0.

F a represents µ the missing auxiliary ﬁeld that we had alluded to earlier.13. We note that Linv is invariant under the BRST transformation as we had argued earlier in (13. replace the original gauge invariance of the theory and play a fundamental role in the study of non-Abelian gauge theories. These are known as the anti-BRST transformations. However.12) invariant. F a .16) δAa = µ δca = δca = δF a = −ωf abc F b cc . There is also a second set of fermionic transformations involving the anti-ghost ﬁelds of the form ω (Dµ c)a .9)). since these do not lead to any new constraint on the structure of the theory beyond what the BRST invariance provides. ca . ca .1 BRST symmetry 549 for all the ﬁeld variables φa = Aa . 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 . 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 (13.15) Unlike in the formulation of BRST variations without the auxiliary ﬁeld in (13. 2 (13.8) 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. g g (13. In some sense the BRST transformations. which deﬁne a residual global symmetry of the full theory.4). we will not pursue this symmetry further in our discussions. Therefore. g ω F a + gf abc cb cc . g ω abc b c f cc. which can also be easily checked to leave the total Lagrangian density (13.

In the covariant gauge in Maxwell’s theory. This is known as the ghost scaling symmetry of the theory.12) is also invariant under the inﬁnitesimal bosonic global symmetry transformations δca = ǫca . the physical Hilbert space needs to be properly selected for a discussion of physical questions associated with the non-Abelian gauge theory. Here ǫ represents a constant. as we have emphasized several times by now. contains many more states than the physical states alone. we note here that these fermionic symmetries arise naturally in a superspace formulation of gauge theories. it allows us to carry out covariant quantization of the non-Abelian gauge theory. Without going into details. δca = −ǫca .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. we . namely.17) with all other ﬁelds remaining inert. Therefore. the total Lagrangian density (13. We have already seen that the naive Gupta-Bleuler quantization does not work in the nonAbelian case.) 13. (The fact that these transformations are like scale transformations and not like phase transformations. In addition to these two anti-commuting symmetries. We recall that the physical space must be selected in such a way that it remains invariant under the time evolution of the system. it corresponds to the operator that counts the ghost number of the ﬁelds. commuting inﬁnitesimal parameter and the generator of this symmetry transformation merely corresponds to the ghost number operator. (13. which is normally associated with the number operator. for example. 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. For example. Let us recall that the vector space of the full theory in the covariant gauge (12.12).181).550 13 BRST invariance and its consequences reﬂection of the asymmetric manner in which these ﬁelds occur in the ghost Lagrangian density in (13.2) or (13.

(13. On the other hand.164) and hence the physical space so selected remains invariant under time evolution. does not satisfy a free equation and hence it is not a suitable operator for identifying the physical subspace in a time invariant manner. = 0. the Gupta-Bleuler condition works because ∂ µ Aµ satisﬁes the free Klein-Gordon equation (9. whether they can reduce the vector space suﬃciently enough to coincide with the physical space).130) in the covariant gauge (12.) o Thus.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. 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 = 0.19) are suﬃcient to reproduce even the GuptaBleuler condition in the case of the Abelian theory (namely. In the Abelian theory. in reality it is an inﬁnite number of conditions since it has to hold for every value of the coordinates.19) Note that even in the case of Maxwell’s theory.118). (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.18). QBRST . and the ghost scaling symmetry. are conserved and hence can be used to deﬁne a physical Hilbert space which would remain invariant under the time evolution of the system.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. µ (13. stands for the positive frequency part of the ﬁeld. the generators of the BRST symmetry.162)) ∂ µ A(+) (x)|phys = 0. We recognize that even though this looks like one condition. therefore. Qc . the conditions in (13. as we have seen in (12.18) where the superscript. The corresponding operator in a non-Abelian theory. . “(+)”. not clear a priori if the conditions (13.13. It is.

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 . 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 ) . 2 (13. g abc µ a b c f (∂ c ) c c .20) where we have used the fact that the auxiliary ﬁeld does not transform under the BRST transformations and the ﬁelds Aa .552 13 BRST invariance and its consequences To see that these conditions indeed lead to the Gupta-Bleuler condition in Maxwell’s theory. F a are inert µ under the scaling of ghost ﬁelds. 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 . 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 . From this. .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 . (13.

12).26) implies that the physical states must have (net) zero ghost number.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).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. n . In principle. Thus. (13. this allows for states containing an equal number of ghost and anti-ghost particles. (13.27) . we note that for the Abelian theory where f abc = 0 and there is no internal index “a”.13. In particular. the BRST and the ghost scaling charge operators can be obtained from (13.23) following from (13. ← → d3 x c ∂ 0 c. (13.19) Qc |phys = 0.25) 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) . denoting the physical states of the theory as |phys = |Aµ ⊗ |n. (13.22) to correspond to ← → d3 x F ∂ 0 c.

for every momentum mode k). Thus.30) (13. we feel conﬁdent that the physical state conditions in (13. 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. 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). and must further satisfy (with k0 = |k|) F (+) (k)|phys = 0 = kµ Aµ(+) (k) |phys . n = 0. n = 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. To investigate systematically whether the physical state conditions in (13.554 13 BRST invariance and its consequences where n.31) where we have used the equation of motion for the auxiliary ﬁeld (see (13. and F (+) (k)|Aµ = 0.12) that we can obtain the canonical momenta conjugate to various ﬁeld variables of the theory to correspond to . 0 = |Aµ .28) c(+) (k)|n. (13. (13.29) Namely. then this implies that (13.19) really select out the subspace of physical states. we note from the form of the Lagrangian density (13.11)).19) are the right ones even for the non-Abelian theory. |phys = |Aµ ⊗ |0. n denote the (equal) number of ghost and anti-ghost particles.

= iδab δ3 (x − y).33) + The hermiticity conditions for the ghost ﬁelds which arise out of various consistency conditions (for example. t) c + j = iδab δi δ3 (x − y). = iδab δ3 (x − y). t) ca (x. Πb (y. 0 ˙ ∂F a ∂LTOT ˙ = −ca . ∂ ca ˙ ∂LTOT = (D0 c)a . t) i F a (x. the (anti) commutation relations (13.35) .33) have to satisfy the proper hermiticity properties etc. the BRST parameter ω is seen to be anti-Hermitian ω † = −ω.13. the Lagrangian density and the conserved charges have to be Hermitian. (13. t). ˙ ∂ Aa i ∂LTOT = Aa . Πb (y. t). The equal-time 0 canonical (anti) commutation relations for the theory can now be written as ( = 1) Aa (x. = iδab δ3 (x − y). Πb (y. t). (13. (13. ˙ ∂ ca (13.) are given by (ca )† = ca . t). t) c ca (x. Πjb (y. Aa plays the role of the momentum conjugate to F a . (ca )† = −ca . In particular.32) Here we have used left derivatives for the anti-commuting ghost ﬁelds.2 Covariant quantization of Yang-Mills theory 555 Πia = Πa = Πa = c Πa = c ∂LTOT = −F 0ia . we see that in this formulation with the auxiliary ﬁeld.34) With this choice.

The algebra of the conserved charges also follows in a straightforward manner from (13. Qc ] = ica (x).33) [QBRST .22) can now be expressed in terms of the ﬁelds and the conjugate momenta as (before integrating the ﬁrst term by parts in QBRST . see (13.36) and we can calculate the algebra of charges as well as various other relations of interest using the (anti) commutation relations (13. c QBRST = − Qc = − (13. that (this basically describes the behavior of the ghost ﬁelds under a scaling (13. An immediate consequence of the algebra in (13. for example. The ﬁrst equation in (13.38) This algebra can also be derived directly from the transformation laws for the ﬁeld variables in (13. Qc ] = 0.38) is that . QBRST ]+ = 2Q2 BRST = 0. [QBRST .13) and (13.17). Qc ] = −ica (x). (13.38) simply reiterates in an operator language the fact that the BRST transformations are nilpotent (see (13.37) [ca (x). We note. The second equation implies that the ghost scaling symmetry is Abelian. c c 2 a d3 x (Πa ca + ca Πc ) .33) for the ﬁeld variables.34) for the ghost ﬁelds. the conserved charges (13. we see that we can think of iQc as the ghost number operator (with the ghost number for ca being negative).17)) [ca (x). The conserved charges (13.22)) g d3 x Πia (Di c)a − F a Πa + f abc Πa cb cc .556 13 BRST invariance and its consequences and we note that the currents (13. therefore. (13. The third is a statement of the fact that the BRST charge carries a ghost number of unity. [Qc .22) are Hermitian with the assigned hermiticity conditions (13.14)). Qc ] = −iQBRST .21) and. Therefore.36) and (13.

Namely. it is truly a BRST singlet (invariant) state. if a state |Ψ cannot be written as ˜ |Ψ = QBRST |Ψ . The ﬁeld operator which creates such a state must necessarily commute with QBRST and. for every |Ψ ∈ Vphys .40). eπQc + = 0. or. As is clear from our earlier discussions. we have QBRST |Ψ = 0. [QBRST . As we have discussed earlier. then. (13.12) contains various unphysical states as well as states with negative norm in addition to the physical states. we can select out a subspace Vphys by requiring that states in this space are annihilated by QBRST .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.2 Covariant quantization of Yang-Mills theory 557 QBRST eπQc = eπ(Qc −i) QBRST = −eπQc QBRST . (13.41) We note that there are two possible kinds of states which will satisfy the physical state condition (13. the total vector space V of the complete gauge theory (13. H] = 0.39) which is quite useful (particularly) in dealing with gauge theories at ﬁnite temperature. (we assume that the vacuum state is a BRST singlet (invariant) state) .42) ˜ for some |Ψ and still satisﬁes the physical state condition. 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. (13. Consequently. (13. First. QBRST . must represent a truly gauge invariant ﬁeld variable.13. therefore. namely.

13).40) can be written in the form ˜ |Ψ = QBRST |Ψ . namely. has a diﬀerent character. as we have seen in (13.) The second class of states which would satisfy the physical state condition (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 . Ψ] |0 = 0. Ψ|0 = |Ψ . it can be written as the BRST variation of the anti-ghost ﬁeld and. (Note that the auxiliary ﬁeld does not transform under a BRST transformation and. This implies that (13.44) [QBRST . such states would have positive norm. (13.43) where we have assumed that the operator Ψ creates the state |Ψ from vacuum. (13. therefore.46) We see that the physical state condition is trivially satisﬁed in this case because of the nilpotency of QBRST (see (13.558 13 BRST invariance and its consequences QBRST |Ψ = QBRST Ψ|0 = [QBRST .) ˜ ˜ Ψ|Ψ = Ψ|QBRST QBRST |Ψ = 0. The nilpotency of QBRST also implies that all such states would have zero norm because (QBRST is Hermitian with our choice of hermiticity conditions (13.38)). (13. therefore. also satisﬁes the above relation. Ψ] = 0. (13. Such states would. However. correspond to truly physical states of the theory and since gauge invariant degrees of freedom have physical (non-negative) commutation relations.34) for the ghost ﬁelds.45) and transverse ﬁelds would represent such operators which are gauge invariant asymptotically. therefore.

2 Covariant quantization of Yang-Mills theory 559 (13. with no unphysical particles) and from the completeness of the projection operators in (13. V0 Every state in Vphys can. therefore.49). (13. ca ]+ .50) It is clear from the deﬁnition in (13. by unphysical. the longitudinal components of Aa and the auxiliary ﬁelds. this would contain all the zero norm states which would be orthogonal to Vphys itself.13) through (anti) commutation with the appropriate ﬁeld variables) F a ∼ [QBRST . Thus. The true physical states.13. we mean states containing quanta of the ghost ﬁelds. then. of course. we note that we can write P (0) = ½ − P ′ .40) because if |Ψ represents a physical state (of either kind) while |Ψ′ is a state of the second kind.48) which follows from the physical state condition. Here.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. be decomposed into states containing ﬁxed numbers of unphysical particles. we have ∞ n=0 P (n) = ½.49) that P (0) projects onto the truly physical states of the theory (namely. then ˜ Ψ′ |Ψ = Ψ′ |QBRST |Ψ = 0. (13.51) . µ We recall from (13. deﬁning P (n) as the projection operator onto the n-unphysical particle sector. If we denote all the states of Vphys of the second kind by V0 . (13. be identiﬁed as belonging to the V quotient space V phys = phys and will have positive deﬁnite norm. which satisfy the physical state condition and are of the ﬁrst kind can. (13.49) P (n) P (m) = δnm P (n) .

is the fact that P (0) projects onto the true physical states and. of course. P ′ = 0.56) + P′ It is now easy to show that P (n) . must be gauge invariant and will commute with QBRST . = 0. we can even show that for every P (n) . the diﬀerence between the two lies in the fact that P ′ must necessarily involve unphysical ﬁelds since it projects onto unphysical states and. cannot be truly gauge invariant. Namely. then. [QBRST . it must have the form (for some fermionic operator R. R]+ . therefore. whose explicit form is not important for our discussions) P ′ = [QBRST .54) (13. What is interesting. but this is not very illuminating.52) We can. R(n) . therefore.53) However. (13. so that QBRST . construct the actual forms of all the projection operators explicitly. n ≥ 1. (13. therefore. project onto the zero norm space V0 when acting on states |Ψ .55) which follows from the nilpotency of QBRST . we have QBRST . that P ′ must also commute with QBRST . R]+ = 0. however. P (0) QBRST . (13. It follows. for n ≥ 1. P ′ = QBRST . Consequently.560 where ∞ n=1 13 BRST invariance and its consequences P = ′ P (n) . (13. and. In fact. we can write P (n) = QBRST .

therefore.13. let |Ψ .58) (13. (13. Any vector |Ψ ∈ Vphys can now be written as |Ψ = P (0) |Ψ + P ′ |Ψ . This also makes it clear that the V norm of a state |Ψ ∈ V phys = phys would be positive deﬁnite. This V0 would correspond to the true physical subspace of the total vector space.40). R(n) |Ψ = 0. then. which follows from the physical state condition (13. 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.3 Unitarity The BRST invariance of a gauge theory is quite important from yet another consideration. and the norm of any such state is.19). We will only outline the proof of unitarity in this section. |Ψ′ ∈ Vphys .59) This. Ψ′ |P (n) |Ψ Ψ′ |P ′ |Ψ = = Ψ′ | QBRST . 13.34). For example. It leads to a formal proof of unitarity of the theory when restricted to the subspace of the physical Hilbert space. then. The question of the unitarity of the S-matrix in a gauge theory can be formulated in the following manner.40) (as well as the hermiticity of QBRST ). (13. obtained to be Ψ|Ψ = Ψ|P (0) |Ψ ≥ 0. 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. with the hermiticity assignments for the ghost ﬁelds in (13. the total Lagrangian . deﬁned by the physical state condition.57) + Ψ′ | [QBRST . We note that. R]+ |Ψ = 0. shows that Vphys .3 Unitarity 561 which satisfy the physical state condition (13.

we note that given any state |Ψ ∈ Vphys .e. we can deﬁne a unique state |Ψ ∈ V phys as |Ψ = P (0) |Ψ . namely. namely. This question can be systematically analyzed as follows. the S-matrix is BRST invariant since the full theory is.64) The two states.562 13 BRST invariance and its consequences density (13. (13. (13. (13. (13.12) is Hermitian so that the S-matrix of the theory is formally unitary.60) Furthermore.57). |Φ ∈ Vphys and any two operators A and B acting on Vphys (i. P (0) |Ψ = P (0) |Ψ = P (0) |Ψ = |Ψ . any two operators which do not take us out of Vphys ) Ψ|P (0) |Φ Ψ|(½ − P ′ )|Φ = Ψ|Φ . |Ψ and |Ψ . 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 . Let us note here from the discussions of the last section that for any |Ψ . = = Ψ|AP (0) B|Φ Ψ|A(½ − P ′ )B|Φ = Ψ|AB|Φ . Furthermore.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. 2 (13. S † S = SS † = ½.62) which follow from (13. S] = 0. P (0) acts as the identity operator. so that [QBRST .63) where we note that in the space V phys ..

3 Unitarity 563 Ψ|Φ = Ψ| P (0) |Φ = Ψ|P (0) |Φ = Ψ|Φ . 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) .65) where the last line follows from the ﬁrst of the conditions in (13. Given this. the physical state condition (13. (13.63).68) that † Sphys Sphys = ½. also automatically leads to a formal proof of unitarity of the S-matrix in the subspace of the truly physical states of the theory. In a similar manner.62). Furthermore. (13.70) . |Ψ = P (0) |Ψ ∈ V phys as deﬁned in (13. we can also show that † Sphys Sphys = ½. 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 . (13. Sphys |Ψ = Sphys P (0) |Ψ = P (0) S|Ψ = S|Ψ .62) as well as the formal unitarity of S (13.67) It is clear that since S is BRST invariant. (13.60).66) and (13.66) such that for |Ψ ∈ Vphys .13. (13. it will leave the space Vphys invariant and hence Sphys will not take a state out of V phys .68) where we have used (13.40) which naturally follows from the BRST invariance of the theory. It follows from (13.69) In other words. 2 (13.

= 0. 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) . − ca F a − ∂ µ ca Aa µ 2 + (13. be written as a BRST variation (with the parameter of transformation taken out).40) that phys| (LGF + Lghost ) |phys′ = −g phys| QBRST . in fact. namely. ξ a a c F + ∂ µ ca Aa µ 2 + |phys′ (13.564 13 BRST invariance and its consequences As another consequence of the BRST invariance of the theory. 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 . 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. = g QBRST . the gauge ﬁxing and the ghost Lagrangian densities) can. we note that these extra terms in the Lagrangian density (namely. It now follows from the physical state condition (13. we can show that the gauge ﬁxing and the ghost Lagrangian densities lead to no physical consequences. To show this.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. 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. For simplicity. ca F a ]+ |0 J = 0. (13.74) where S[φ] denotes the dynamical action of the self-interacting scalar ﬁeld. (13.7.4 Slavnov-Taylor identity The BRST invariance of the full theory leads to many relations between various scattering amplitudes of the theory. the vacuum functional is given by 0|0 J = Z[J] = eiW [J] = N Dφ eiS J [φ] .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.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 . (13. as we have seen in (12. 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. 13. In this case. We have already given a simple diagrammatic derivation of such identities in the case of QED in section 9. let us recapitulate brieﬂy some of the essential concepts from path integrals (a detailed discussion is beyond the scope of these lectures).13.75) . before we go into the actual derivation.120). However. 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).

In particular we note that 0|φ(x)|0 J δn W [J] δJ(x1 ) · · · δJ(xn ) J=0 . To clarify this distinction further.77) in the path integral formalism corresponds to a unique Feynman diagram in perturbation theory. This is more fundamental since the general Green’s functions can be constructed in terms of these.566 13 BRST invariance and its consequences As we have already mentioned in the last chapter.77) involve only Feynman diagrams where all the parts of the diagram are connected. On the other hand. For example.76) J=0 deﬁnes the n-point time ordered Green’s function of the theory.76)) includes contributions of all diagrams (including the ones that are not connected) to a Green’s function. (13.78) is known as the classical ﬁeld (see (7. 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. (13.49)) and is a functional of J. 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 ) . the Green’s functions calculated from W [J] (the logarithm of Z[J]) through (13. the n-point Green’s functions would involve Feynman diagrams which consist of parts that are not connected and Z[J] (through (13. Z[J] and W [J] are known as the generating functionals for the Green’s functions of the theory.79) .77) = φc (x) = (−i) δZ[J] δW [J] = . In general. let us note here that each term in the calculation of the Green’s function in (13.76) and (13.77) 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 . (13. On the other hand.

These are quite fundamental in the study of quantum ﬁeld theory and the generating functional which generates such vertex functions is constructed as follows.82) . With these basic ideas from path integrals. (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.4 Slavnov-Taylor identity 567 The connected n-point time ordered Green’s functions consist of connected Feynman diagrams with external lines (propagators). let us consider the eﬀective Lagrangian density (at the lowest order or tree level) which consists of LTOT for the Yang-Mills theory (13.80) It can now be checked using (13. 2 (13. the more fundamental concept in perturbation theory is the diagram without the external propagators known as the vertex function.81) = −J(y). 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.13.78) 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) (13.12) 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 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).

in the presence of sources.84) Here. δη a 0| (Dµ c)a |0 0| = (Dµ c)a J = δW [J] . we obtain the change . but we have also added sources (K µa . for the variations in (13. we have assumed the convention of left derivatives for the anticommuting ﬁelds. The eﬀective Lagrangian density is no longer invariant under the BRST transformations (13. The ﬁelds A(c) are known as the classical ﬁelds (see. ca which are nonlinear in µ the ﬁeld variables). δJ µa δW [J] .49)) and in what follows we will ignore the superscript (c) for notational simplicity. recalling that LTOT in (13. In fact.13) when the external sources are held ﬁxed. Denoting all the ﬁelds and the sources generically by A and J respectively. (13.82) is BRST invariant and that the BRST transformations are nilpotent.568 13 BRST invariance and its consequences Here. K a ) for the composite variations under the BRST transformation (namely. for example. δK a (13. can now be written as δW [J] . we have not only introduced sources for all the ﬁeld variables in the theory. It is worth noting here that the source K µa is of fermionic nature (in addition to the fermionic sources η a . (7. we can write the generating functional for the theory as R d4 x Leﬀ 0|0 J = Z[J] = eiW [J] = N DA ei .13) of the ﬁelds Aa .83) The vacuum expectation values of operators. η a ). δK µa J g abc b c f c c |0 2 = g abc b c f cc 2 = δW [J] . The usefulness of this will become clear shortly. 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] . 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) .

85) g 2 We note that the generating functional is deﬁned by integrating over all possible ﬁeld conﬁgurations. δJ a (x) (13. (13. if we redeﬁne the ﬁelds under the path integral as.13.87) +iη a (x) The functional integration measure can be easily checked to be invariant under such a fermionic transformation and using (13. it should not change under any ﬁeld redeﬁnition).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 . It is here that the . Therefore.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.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 .87) can also be written as δW δW δW = 0. (13.86) the generating functional should be invariant (namely.75) we note that (13. − iη a (x) + iη a (x) a µa (x) a (x) δK δK δJ (x) (13. since it does not depend on the ﬁeld variables. A → A + δBRST A.85) to (see also (13. This immediately leads from (13.

From the deﬁnition of the generating functional for the proper vertices (see (13. µ (13. we have Γ[A. Most often.81)). = = δW .90) Using these deﬁnitions. 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. K] − d4 x J µa Aa + J a F a + i (η a ca − ca η a ) . however. = iη a .89) where K stands generically for the sources for the composite BRST variations.570 13 BRST invariance and its consequences usefulness of the sources for the composite BRST variations becomes evident. a δKµ δW .80) and the ﬁeld variables are really the classical ﬁelds and we are dropping the superscript (c) for simplicity).88) in terms of the generating functional for the proper vertices as . it is clear that δΓ δAa µ δΓ δF a δΓ δca δΓ δca δΓ a δKµ δΓ δK a = −J µa . = −J a . δK a (13. we are interested in the identities satisﬁed by the proper (1PI) vertices of the theory. = iη a . we see that we can rewrite the master equation (13. K] = W [J.

follow from the BRST invariance of the theory. (We would discuss renormalization of ﬁeld theories in a later chapter. in this way. 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. like the Slavnov-Taylor identities.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.13. Thus. This is essential in proving the renormalizability of gauge theories. The BRST invariance.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.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. (13.4 Slavnov-Taylor identity 571 d4 x δΓ δΓ δΓ δΓ δΓ + a − F a (x) a = 0. a (x) δK µa (x) a (x) δAµ δc (x) δK δc (x) (13. δ2 δF b (p)δcc (−p) d4 k (13. consequently. is very fundamental in the study of gauge theories as far as renormalizability and gauge independence of physical observables are concerned. For example. the counter terms (quantum corrections) should satisfy such a relation.92) and setting all Taking derivative of this with respect to ﬁeld variables to zero gives δ2 Γ δ2 Γ δ2 Γ = 0. let us note that we can write the master identity (13. for example.) .

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.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).95) LQ = − = = 1 a µν A η 2 µ 1 a µν ab b A O Aν + ca M ab cb . ξ M ab = −δab . These are the essential elements in carrying out any perturbative calculation in a quantum ﬁeld theory. Let us recall that the total Lagrangian density after gauge ﬁxing (in the covariant gauge) has the form (13. (They are the Green’s functions of the free theory.) The inverses of the two point functions are easily calculated in the momentum space where the quadratic operators in (13.5 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.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. (13.572 13. − 1− (13. We note that since gauge ﬁxing breaks gauge invariance. 2 µ where we have neglected total divergence terms and have identiﬁed Oµν ab = δab η µν 1 µ ν ∂ ∂ .96) take the forms .2) 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ 2 + ∂ µ ca Dµ ca .

λ ξ ξ (13. or. λ or.99) where α. let us derive this inverse systematically.100) Substituting the parameterization (13.5 Feynman rules 573 M ab (p) = δab p2 . η µν p2 − 1 − 1 µ ν −1ac p p O νλ (p) = −δac δµ . λ αp2 δµ − α 1 − λ 1 1 µ p pλ + β pµ pλ − β 1 − pµ pλ = −δµ .97) is quite simple (it is like the propagator of a massless scalar ﬁeld) M −1ab (p) = δab . ξ (13. Oµν ab (p) = −δab η µν p2 − 1 − 1 µ ν p p . ηµν ).101) 1 β µ (αp2 + 1)δµ − α(1 − ) − p pλ = 0. 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 . Therefore.98) while the derivation of the inverse of Oµν ab (p) is a bit more involved. we note that the inverse is deﬁned to satisfy (repeated indices are summed) −1bc Oµν ab (p)Oνλ (p) = δac δµ . With this parameterization.97) The inverse of M ab (p) in (13.13. p2 (13. λ ξ ξ . 1 µ ν p p ξ αηνλ + β pν pλ p2 = −δµ . β are arbitrary parameters to be determined (they are not necessarily constants and can be Lorentz invariant functions of the momentum). λ ξ (13.99) for the inverse and carrying out the multiplication explicitly we have η µν p2 − 1 − or. p2 (13. we can parameterize the most general form of the inverse as O−1ab (p) = δab α ηµν + β µν pµ pν .

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 . β = α(ξ − 1) = − so that we can write the inverse in (13. 2 p p −1ab Oµν (p) = − (13. ξ ξ 1 (ξ − 1).574 13 BRST invariance and its consequences Setting the coeﬃcients of each distinct Lorentz structure to zero.Oµν (p)J νb (p) +i η a (−p).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.126)) h −1ab i − 2 J µa (−p).M −1ab (p)ηb (p) Z0 = N e i . (13.102) α=− α 1− or.79).103) With these. ) represents the integral over p and we deduce from this that (see (13. 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. 2 p p (13. see (12. it follows that 1 .104) where ( . . p2 1 β − = 0. p2 (13.99) as pµ pν δab ηµν + (ξ − 1) 2 .

110) −gf abc Aa ∂ µ cc cb .13.5 Feynman rules 575 When ξ = 0. µ . b pµ pν iδab ηµν + (ξ − 1) 2 . i. p2 (13.106) which deﬁnes the ghost propagator (which is sometimes also denoted by Dab (p)).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. when we are in the Landau gauge this propagator is transverse.107) µ. the gauge is known as the Feynman gauge where the propagator has a much simpler form which is more suitable for perturbative calculations. while for ξ = 1. a p ν. (13. The propagators can be diagrammatically represented as = iGab (p) µν = − a p b (13.e.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]. p2 (13. 2 p p iδab .. 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 .

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. p4 ) = i (13. V µνλρ abcd (p1 . 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 .111) × (−ipµ ) Aa (p1 ) cc (p3 ) cb (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 . p2 .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).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 . p2 . p3 .

p3 ) = i δ3 S δAa (p1 )δcb (p2 )δcc (p3 ) µ = i δ3 Sint δAa (p1 )δcb (p2 )δcc (p3 ) µ (13. 3 Therefore. a ρ. b c = V µ abc (p1 . with (13. p2 . (13. p2 . p2 . p3 . p2 ν.13.114) we can represent all the interaction vertices of the theory as λ. (13. p2 . p4 ).114) = −(2π)4 gf abc pµ δ4 (p1 + p2 + p3 ). b = V µνλ abc (p1 .112)-(13. a p2 b p2 ν. c = V µνλρ abcd (p1 . p3 ).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 .115) . p4 p3 p1 µ.113) V µ abc (p1 . a p3 p1 µ. d λ. c p3 p1 µ. p3 ).

from two distinct points of view. If possible. 13. LTOT = − Fµν F µνa − 4 2ξ (13. 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.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. this would.6 Ghost free gauges In quantizing a non-Abelian gauge theory. such a gauge is known as the light-cone gauge. the diagrams with ghosts will be absent). Such a class of gauge choices is conventionally known as ghost free gauges.160)) that the total Lagrangian density for a non-Abelian gauge theory takes the form 1 1 a (n · Aa )2 − ca n · Dca . We will now discuss.116) where nµ is an arbitrary vector does indeed achieve this. or (13.578 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. simplify the perturbative calculations enormously primarily because the number of diagrams to evaluate will be much smaller (namely.114)). In a general axial-like gauge (13. When n2 = 0. we recall (see (12. A class of gauge conditions of the form n · Aa (x) = f a (x). These represent all the Feynman rules for the Yang-Mills theory and perturbative calculations can now be carried out using these Feynman rules. 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. Let us ﬁrst discuss this from the diagrammatic point of view.117) . for n2 = 1 (normalized) it is called the temporal gauge while for n2 = −1 it is known as the axial gauge. see for example (13. of course. the cc ﬁeld.153) and (12.

. (n · p)2 (13. the ghost propagator in the axial-like gauge (13.116) in this theory has the form δab .13.1: Examples of diagrams with open and closed ghost lines. n·p iGab = (13. the integrand of any diagram involving ghosts. either open lines or closed loops. (n · p) nµ Gab = −ξδab µν nµ nν Gab = −ξδab .118) so that we have (independent of whether n2 = 0 or n2 = 0) pν .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 .119) Similarly. Therefore. µν (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 13.120) and the antighost-ghost-gauge vertex involves a multiplicative factor of nµ .

580 13 BRST invariance and its consequences dD k = 1 (n · k) (n · (k + p1 )) .123) Here G[f a ] is an arbitrary functional of f a which we normally choose to correspond to the ’t Hooft weight factor . (k2 )(m+1)/2 (13. . 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. An alternative way to see this is to note that the generating functional in the axial-like gauge (13. (13. namely. Using the fact that the integral is Lorentz invariant. .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)). we can rewrite this also as 1 m dxi 0 i=0 δ(1 − x0 − x1 − · · · − xm ) (n2 )(m+1)/2 dD k = 0. 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)m+1 δ(1 − x0 − x1 − · · · − xm ) m+1 = 0 i=0 dxi (13. (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 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 . Note that such a decoupling works only because nµ (in n · Aa inside the determinant) is a multiplicative operator. A. correspond to ghost free gauges very much like in the Abelian gauge theories. Rouet and R.13. since the generating functional does not depend on the functional form of G [f a ] (up to irrelevant multiplicative constants).125) In this case. Stora. J. A. Taylor. 3. (13. . 99 (1972). 13. Slavnov. Nuclear Physics B33. let us choose G[f a ] = det n · ∂δab + gf abc f c −1 − i R d4 x f a f a 2ξ e . Theoretical and Mathematical Physics 10. Becchi. (13. A. C.7 References 1.124) However. Physics Letters 52B.7 References 581 G[f a ] = e i − 2ξ R d4 x f a f a . C. (13. 436 (1971). the covariant gauge. The axial-like gauge choices (13.126) This shows that in this gauge the ghosts decouple completely and can be absorbed into the normalization factor. 2.116) in a non-Abelian gauge theory. 344 (1974). It would not work in say. therefore.

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

14) or (12. in this chapter we would discuss this important question of massive gauge ﬁelds concluding with the standard model of the electroweak interactions. 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.1) is not invariant under the gauge transformation (9. L = − Fµν F µν + 4 2 583 (14. As we have mentioned earlier gauge ﬁelds can be thought of as the carriers of physical forces.Chapter 14 Higgs phenomenon and the standard model In chapters 9.2) . 2 M 2 a µa Aµ A . 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 . 14. a massless gauge ﬁeld is completely consistent with the observed fact that electromagnetic forces are long ranged. However. 12 and 13 we have studied gauge theories (both Abelian and non-Abelian) in great detail. Therefore.34) respectively. Therefore. 2 or (14.1 St¨ckelberg formalism u Let us consider the Lagrangian density for a charge neutral spin 1 ﬁeld Aµ given by M2 1 Aµ Aµ .

∂µ F µν + M 2 Aν = 0. Note that the ﬁeld variable Aµ has four ﬁeld degrees of freedom while the constraint (14.2) does not have a gauge invariance (because of the mass term) unlike the Maxwell theory. ∂∂µ Aν ∂Aν ∂µ or.6) Equation (14. + M 2 Aν = 0.5) in a simple manner and in momentum space has the form . (14.6) together with the constraint (14.2) has the form ∂L ∂L − = 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) yields ∂µ (∂ µ Aν − ∂ ν Aµ ) + M 2 Aν = 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.4) Contracting (14. (14. The Euler-Lagrange equation following from (14. As has been mentioned earlier. we also note here that the theory (14. The propagator for this theory can be worked out (see. (14.4) with ∂ν and using the anti-symmetry of the ﬁeld strength tensor.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). −∂µ F µν − M 2 Aν = 0. section 13. we obtain ∂ · A = 0. or.2) describes a massive photon which can be seen as follows.5) deﬁnes the Proca equation (and (14.5) and substituting this back into the Euler-Lagrange equation in (14. (14. for example.3) The Lagrangian density (14. or.

The limit M → 0 of the Proca theory is clearly quite subtle.9) Aµ + M 1 1 1 ∂µ χ − ∂µ χ → Aµ + ∂µ θ + ∂µ θ M e M e 1 = Aµ + ∂µ χ.8) was constructed to be invariant under a Maxwell-like gauge transformation Aµ (x) → Aµ (x) + 1 ∂µ θ(x). 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 − . To see the invariance of the full theory. This leads to diﬃculties in establishing renormalizability of the (interacting) Proca theory for massive photons. invariant under the gauge transformation (14. In contrast to the Proca theory (14. χ(x) → χ(x) − e (14.8) = − Fµν F µν + 4 2 2 which. the St¨ckelberg theory u (14. e M θ(x). in addition to the spin 1 ﬁeld. 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.10) . As a result of these diﬃculties. (14.2). M (14. 2 −M M2 (14.7) M2 1 1 Aµ Aµ + ∂µ χ∂ µ χ + M Aµ ∂µ χ. of course. it is suﬃcient to note that under the transformation (14. say in (9.9) where e denotes the coupling constant of QED (this can be set to unity. also contains a dynamical charge neutral spin zero ﬁeld (real scalar ﬁeld) which mixes with Aµ . The ﬁeld strength tensor is. but we have kept it for consistency with earlier discussion.¨ 14.9).1 Stuckelberg formalism 585 iGµν (p) = − p2 Thus.83) as well as for discussions in connection with the Higgs phenomenon in the next section).

11) which can be achieved by the choice of the gauge transformation parameter e χ(x). This is typical of the behavior of theories in a unitary gauge.15) this would induce a ghost action. On the other hand.14) has the unpleasant features described earlier. . As a result.12) then the Lagrangian density (14. section 12. However. the gauge propu agator (see also (14. in this unitary gauge.8) takes the form M2 1 Aµ Aµ . for example. 2ξ − (14. the there is no diﬃculty in establishing renormalizability in such a gauge. M θ(x) = (14.586 14 Higgs phenomenon and the standard model Because of the gauge invariance of the theory. but would lead to propagators that are well behaved at high momentum.13) We recognize this to be the Proca theory (14.2) and this makes it clear that the Proca theory can be thought of as the gauge ﬁxed St¨ckelberg theory.7)) i pµ pν ηµν − . L = − Fµν F µν + 4 2 (14. the massless limit M → 0 is now straightforward. (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 . 2 −M M2 iGµν (p) = − p2 (14. we can choose a gauge and if we choose the unitary gauge χ(x) = 0. Furthermore. if we choose a covariant gauge condition which leads to a gauge ﬁxing Lagrangian density of the form (see.

¨ 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. We will now brieﬂy describe the St¨ckelberg formalism for non-Abelian gauge u theories.41)) Aµ → U Aµ U −1 − V → U V. the Lagrangian density for the gauge ﬁeld is invariant under the transformation (14. (12.16) ig ig where the trace is over the fundamental representation of the group. Aν ] . This Lagrangian density can be easily checked to be invariant under the gauge transformations (see. 1 (∂µ U )U −1 .17) where χ denotes a matrix valued scalar ﬁeld.38) and (12. it is suﬃcient to note that . say. for example.18). = e M χ. ig (14.18) As we have already seen in chapter 12. ig (14. Here the ﬁeld strength tensor and the operator V are deﬁned as Fµν V = ∂µ Aν − ∂ν Aµ + ig [Aµ .1). (14. To see that the complete Lagrangian density is also invariant. SU (n) (as we have discussed earlier in section 12. 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 .

In this gauge. then the Lagrangian density (14.20) which can be achieved by the choice of the gauge transformation matrix U (x) = V −1 (x).23) .21) (14. 2 (14.16) becomes 1 L = − Tr Fµν F µν + M 2 Tr Aµ Aµ . as before.2) to the non-Abelian case and describes a massive spin 1 ﬁeld belonging to SU (n).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 . (14.18). 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 − . 2 − M2 p M2 iGab (p) = − µν (14. If we choose the unitary gauge condition V = ½.22) This can be thought of as a generalization of the Proca theory (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. ig (14.

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). (14. (14.25) The Lagrangian density (14.5 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.26) .84) and (7.2 Higgs phenomenon In section 7.85)) to be invariant under the inﬁnitesimal local transformations δφ† = iǫ(x)φ† (x). 14. (14. if we choose a covariant gauge ﬁxing condition (a generalization of (14. but now interacting with an Abelian gauge ﬁeld as well (scalar QED).5).24) 4 4 where the theory has a “wrong” sign for the mass term and the covariant derivative is deﬁned as in (7. see also section 7. the propagator will have the correct asymptotic behavior for large values of the momenta.15)). The Lagrangian density for such a theory is given by (with λ > 0. e δφ = −iǫ(x)φ(x). On the other hand.24) can be checked easily (see (7. In that case. Let us next ask what happens if there is a spontaneous breakdown of a local symmetry. This is known as the Nambu-Goldstone phenomena and the massless scalar ﬁeld is known as the Nambu-Goldstone boson.83) Dµ φ = ∂µ φ + ieAµ φ. For simplicity we consider again the self-interacting theory of a complex scalar ﬁeld (as in section 7.6) λ 1 2 L = − Fµν F µν + (Dµ φ)† (D µ φ) + m2 φ† φ − φ† φ . δAµ = 1 ∂µ ǫ(x).2 Higgs phenomenon 589 with the unpleasant features discussed earlier that are characteristics of the unitary gauge choice.14.

since the mass term for the scalar ﬁeld has the “wrong” sign. this is just one of them) 2m σc = σ = √ .27) does not correspond to the true vacuum. λ where (see also (7. there is actually an inﬁnity of minima of the potential.) Rather the true vacuum of the theory satisﬁes (see (7. (As we have seen in section 7. 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 . On the other hand.77)) .24) in terms of the σ.30) 2 16 Let us now rewrite the theory around the true vacuum by shifting the ﬁeld variables as (see (7. as we have discussed in section 7. (14.64).29) For further analysis.590 14 Higgs phenomenon and the standard model The gauge boson in this theory is massless because the gauge symmetry (14.26) does not allow a mass term for the gauge ﬁeld. 2 χc = χ = 0.5. let us rewrite the Lagrangian density (14. (14.28) (14. this actually corresponds to the local maximum of the potential.5. (14. the normal vacuum for which φ = φ† = 0.42)) 1 φ = √ (σ + iχ).

31) where we have identiﬁed the vacuum expectation value (vev) of the ﬁeld with 2m σ =v= √ . (14. but it now mixes with the gauge ﬁeld Aµ so that we have .2 Higgs phenomenon 591 2m σ → σ + σ = σ + √ = σ + v. (14.33) where we have used the form of v in (14. Let us look at only the terms in (14.33) 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 . the Lagrangian density (14.14. λ In this case.34) 2 2 We note here that the χ ﬁeld appears to be massless in this theory as in (7.30) 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 (14.32) to simplify some of the terms.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 ). λ χ → χ + χ = χ. 16 (14.79).

34).37) 2 We see from (14. 1 1 ∂µ χ Aµ + ∂ µ χ ev ev 2 1 = Aµ Aµ + Aµ ∂µ χ + 2 2 ∂µ χ∂ µ χ.35) the quadratic Lagrangian density (14. σ (14.35) ∂µ Bν − ∂ν Bµ = ∂µ Aν + 1 1 ∂ν χ − ∂ν Aµ + ∂µ χ ev ev = ∂µ Aν − ∂ν Aµ .36) Thus in terms of this new ﬁeld variable (14.38) and a massive gauge ﬁeld Bµ with mass (this is the analog of the unitary gauge in (14.39) . λ m2 = M 2 = e2 v 2 = B (14.) To achieve this. As in (7.34) 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 . ev Bµ = Aµ + then we have (14.592 14 Higgs phenomenon and the standard model to diagonalize the theory in order to determine the true spectrum of the theory.37) that the quadratic part of the shifted Lagrangian density is completely diagonalized.8).79). let us deﬁne 1 ∂µ χ.11)) 4e2 m2 . (14. it describes a massive scalar ﬁeld σ with mass m2 = 2m2 . ev e v Aµ + Bµ B µ = (14. ignoring the σ ﬁeld should be compared with the St¨ckelberg Lagrangian density u (14. (The Lagrangian density in (14.

8) the metric of the Hilbert space becomes indeﬁnite. .the NambuGoldstone boson . 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. we can see this happening in the following way. (We can think of the physical degrees of freedom as the two transverse components. namely. Brout and Englert. we have a mechanism for giving masses to gauge bosons without violating renormalizability and unitarity of the theory. The physical states. The answer to this question lies in the fact that if a gauge theory is written in a manifestly Lorentz invariant way. of course. however. At this point we may ask why the Goldstone theorem fails in this case.has completely disappeared from the spectrum of the theory. 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. it is the incompatibility of these two conditions that avoids the Goldstone theorem. as we have seen (recall Gupta-Bleuler quantization in 9. the massless mode has combined with the massless gauge ﬁeld to give it an additional degree of freedom. Guralnik. 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. Intuitively. Thus the other two states must have cancelling eﬀect with each other as we have seen earlier.2 Higgs phenomenon 593 The other real component χ of the spin 0 boson .14. The scalar ﬁeld responsible for this phenomenon is conventionally known as the Higgs ﬁeld. why is there no massless particle even though we have a spontaneous breakdown of symmetry. namely.) However. the Goldstone mode of the complex scalar ﬁeld. Thus some states must have negative norm in such a description. Hagen and Kibble. In fact. this second aspect that is important in developing a gauge theory of the weak interactions. It is. This is consistent with the unitarity of the theory (from the point of view of ﬁeld degrees of freedom). 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. consist only of transverse polarization. When we are dealing with a gauge theory. Hence the gauge ﬁeld has become massive with three degrees of freedom without violating unitarity.

e 1 ∂µ ǫ(x). 2ξ 2 (14. 1 ∂µ ǫ(x). To introduce the gauge choice.34) (namely. δχ(x) = −ǫ(x)(v + σ(x)) = −ǫ(x) δAµ (x) = M + σ(x) . that is more useful for this purpose let us proceed as follows.41) e where M is deﬁned in (14.40) e The shifted theory (14. δχ(x) = −ǫ(x)σ(x). as pointed out in the last section. (14. First we note that the unshifted theory (14. it is not the right gauge to study renormalizability of the theory. commonly known as the ’t Hooft gauge or the Rξ gauge.33) which reﬂects a spontaneous breakdown of the local symmetry (14.84)) δσ(x) = ǫ(x)χ(x). eliminate the mixing term between the Aµ and the χ ﬁelds).39).42) . 2ξ 2ξ ξM 2 2 1 µ (∂ Aµ )2 + M (∂ µ Aµ )χ − χ .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 (14.30) is invariant under the inﬁnitesimal gauge transformations (see (7. (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. The presence of the local symmetry (14. 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 .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).

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 ’t Hooft gauge choice in (14.164). The gauge ﬁxing Lagrangian density (14. thereby. As a result. − ξM 2 pµ pν i ηµν − (1 − ξ) 2 .170)) LGF = ξ 2 F − F (∂ · A − ξM χ). (14.42) combine into a total divergence and.42) (or (14. the propagators for the Aµ and the χ ﬁelds in this gauge are given respectively by iGµν (p) = − iGχ (p) = p2 p2 i .15).34) together with the cross terms in (14.42) can also be written with an auxiliary ﬁeld as (see (12. The complete Lagrangian density in this gauge has the form LTOT = L + LGF + Lghost . (14. disappear from the action.45) We see that unlike in the conventional covariant gauge ﬁxing (12.43) and we note that in this gauge the propagators fall oﬀ for large values of the momenta unlike in the unitary gauge.46) and can be easily checked to be invariant under the BRST transformations . 2 (14.44)) leads to an interacting ghost Lagrangian density. It is clear that the mixing terms in (14. 2 −M p − ξM 2 (14.14.2 Higgs phenomenon 595 which can be compared with (14.

let us review some of these essential earlier results before constructing the standard model (known as the Weinberg-SalamGlashow theory). Therefore. 14. (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. weak interactions (for example. for example.48) where Pσ and Pχ denote respectively the sources for the composite BRST variations of the ﬁelds σ and χ in (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. 2 (14.596 14 Higgs phenomenon and the standard model δAµ = 1 ω∂µ c. δF = 0.4 (see. At low energies. the β decay of the neutron etc. 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 .47) Following the discussions in section 13. δc = −ωF.49) . (13. e δσ = ωχc.47). δAµ (x) δσ(x) δPσ (x) δχ(x) δPχ (x) δc(x) (14. M + σ c. This is the starting point for establishing the renormalizability of the gauge theory with a spontaneously broken symmetry (Higgs theory).

proton. 2 Q = I3L + (14. ψν denoting the fermion ﬁelds associated with neutron. Furthermore.52) where Q denotes the electric charge of the particle. Experimentally . (14.3 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 (3.140) for a deﬁnition of right and left-handed particles/ﬁelds). 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 . the currents in (14.139) and (3.51) The weak interactions are. (14.14. (The V − A structure is connected with the fact that parity is violated maximally in weak interactions. electron and neutrino respectively. 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.) In addition to the weak isospin quantum number. it is clear from the V − A (vector minus axial vector) structure of the currents in (14.50) with ψn . ψe . I3L the quantum number associated with the 3-component of the weak isospin generator and Y its weak hypercharge quantum number. ψp . 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).17 × 10−5 (GeV)−2 .50) have a weak isospin structure and are also known as the weak isospin currents. As a result. therefore. quite weak and were also known to be extremely short ranged.

the V − A theory cannot be the fundamental theory underlying weak interactions (forces). 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.598 14 Higgs phenomenon and the standard model it was known that weak interactions violate both weak isospin and weak hypercharge quantum numbers. we would like the gauge theory to describe the symmetry behavior (“SSB” denotes spontaneous symmetry breakdown) SSB SUL (2) × UY (1) −→ UEM (1). since weak interactions are short ranged. Although the V − A theory with four fermion interactions works quite well at low energy. On the other hand.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. (14. . (14.54) With these basics. As a result. but in such a way that the electric charge is conserved in any weak process.51) is dimensional with inverse dimensions of mass (energy). (The nonrenormalizable character is already manifest in the fact that the coupling constant of the theory (14. it is not renormalizable. 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). Furthermore.) Therefore. 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. Namely. let us construct the standard model in several steps. As we have discussed weak interactions violate both the weak isospin and the weak hypercharge symmetries and. it is logical to associate the Higgs mechanism to the spontaneous breakdown of these symmetries. it is clear that we should look for a gauge theory of weak interactions based on the local symmetry group SUL (2) × UY (1). therefore. a gauge theoretic description would naturally involve massive gauge bosons.

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

the opposite of that for the left-handed lepton doublet. the minimal multiplet of Higgs ﬁelds that can achieve the spontaneous breakdown of the symmetry (as in (14.600 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. a Wµ . . a = 1.53) to describe the local (gauge) symmetry of the theory.52). in fact. then the corresponding gauge ﬁelds are determined to be SUL (2) : 3 gauge ﬁelds. As we will see shortly. 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).54)) is given by a doublet of charged ﬁelds.) The gauge ﬁelds which are expected to correspond to the carriers of forces are determined from the local symmetry structure of the theory. UY (1) : 1 gauge ﬁeld. Let us now write down the quantum numbers for the matter and the Higgs ﬁeld and verify explicitly that (14. it carries a nontrivial weak hypercharge quantum number given by Y = 1 which is. (14. 3. (φ0 )† = φ . (φ+ )† = φ− . Furthermore.58) 0 Since the Higgs multiplet is charged. so that the theory should contain four gauge ﬁelds in total. if we would like the gauge bosons to be massive. If we assume (14. (14. 2.57) Yµ . minimal Higgs ﬁeld : φ = φ+ φ0 .52) holds.

(14. Note that ǫabc denotes the structure constant of SUL (2) as we know from the study of angular momentum.16) and (12.4) and (12. 2.2 Lagrangian density. 14.14. (The Abelian gauge ﬁelds are not self-interacting which is why g′ does not appear in the Lagrangian density for the gauge ﬁelds. 4 4 a = 1. but it will be present in the matter Lagrangian density.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. a a b c = ∂µ Wν − ∂ν Wµ − gǫabc Wµ Wν .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. (14.60) . 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.3.60) where the ﬁeld strength tensors for the Abelian as well as the nonAbelian gauge ﬁelds are deﬁned as (see (9.) The Lagrangian density (14. let us note that the gauge invariant Lagrangian density for the gauge ﬁelds follows from our earlier discussions (see (9. First. 3.44)) Fµν a Fµν = ∂µ Yν − ∂ν Yµ .45)) to correspond to 1 1 a LG = − Fµν F µν a − Fµν F µν .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 (14.52).

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. Since the left-handed fermions belong to an isospin doublet while the right-handed fermions are isospin singlets. g g 1 ∂µ ǫ(x). 2. The gauge invariant Lagrangian density for the matter ﬁelds (leptons) can be obtained through minimal coupling. (We note here that within the context of isospin. the Lagrangian density for the leptons takes the form ig′ σ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R. Since the two symmetry groups are commuta ing (direct products). they are inert under the corresponding transformations. We have also used a vector notation here to represent the three components of the isospin vectors which is often used interchangeably. a = 1. the gauge ﬁeld Wµ for the group SUL (2) does not transform under the group UY (1) and vice versa.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) . 3 and ǫ(x) denote the local inﬁnitesimal parameters of symmetry transformations associated with the groups SUL (2) and UY (1) respectively. sometimes it is conventional to denote the generators as τ by making the identiﬁcation 2 σ = τ . we simply have to remember that there are two local symmetries present in the theory.62) a δWµ (x) = δYµ (x) = where ǫa (x). therefore.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.) The other thing to note from the structure of (14. 2 2 (14. g′ (14.63) is that the right-handed fermions couple twice as strongly to the Yµ .

3 The standard model 603 ﬁeld as the left-handed ones simply reﬂecting the Y quantum numbers for the two ﬁelds. δL(x) = −i (14.62). 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. The Lagrangian density (14. a = 1. we also need a Lagrangian density for the scalar Higgs ﬁelds which are expected to lead to a spontaneous breakdown of the local symmetries. Note also that the SUL (2) invariance forbids a mass term for the lepton ﬁelds (the electron and the neutrino are. 2 2 (14. massless in this theory to begin with).64).14.66) together with the gauge ﬁeld transformations in (14. We note that we have combined the three components of ǫa (x).62). 2.63) can be easily checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). Besides the gauge ﬁelds and the leptons. 3 into a vector ǫ(x) in (14. The transformations (14. In fact. 2 2 δR(x) = −iǫ(x)R(x).64) show explicitly that the right-handed fermions do not change under the SUL (2) transformations.65) can be checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)φ(x) + ǫ(x)φ(x). since we .64) together with the transformations in (14. the coupling to the UY (1) gauge ﬁeld is correspondingly with the opposite sign. A mass term as well as a potential for the Higgs ﬁeld can be introduced consistent with the SUL (2) symmetry. therefore. The minimally coupled fermion Lagrangian density (14. 2 2 δφ(x) = −i (14.65) We note that since the hypercharge of the Higgs doublet is just the opposite of that of the left-handed lepton doublet.

68) where h denotes the Yukawa coupling and the Lagrangian density (14. (14. 2 2 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) .67) and (14.24). (14. φ† . (14.604 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. In addition. (14. (14.68) can be checked to be invariant under the transformations (14.70) δYµ (x) = .68).67) which is invariant under the local transformations in (14. 4 λ > 0.60). it is worth noting that in spite of the similarity in their forms for the potential.63). δL(x) = −i δφ(x) = −i a δWµ (x) = i σ · ǫ(x)φ(x) + ǫ(x)φ(x).66). 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. Therefore. φ† = −m2 φ† φ + λ † 2 φφ . collecting all the terms in (14. (14. here the Higgs ﬁeld is represented by an isospin doublet) V φ. g g 1 ∂µ ǫ(x).66).65).64) and (14.69) which is invariant under the inﬁnitesimal local gauge transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). 2 2 δR(x) = −iǫ(x)R(x). g′ (14. the Yukawa interaction leads to masses for fermions after the spontaneous breakdown of the symmetry. the fermions can have Yukawa interactions with the scalar ﬁelds which can be described by the Lagrangian density LY = −h Rφ† L + LφR . of course. As we will see shortly.

72) λ † φφ 2 As we have seen in the last section.74) . the factor of √2 below is a consequence of the deﬁnition in (14. namely. φ = 0. 4 V φ. φ† = 0. We have chosen the Higgs potential (14.5 (and also in section 14.71) so that the minimum of the potential occurs at ∂V ∂φ ∂V ∂φ† λ † φφ 2 = φ† − m2 + = − m2 + = 0. 2m2 .3 Spontaneous symmetry breaking.14.72).29)) 0 v √ 2 φ = .3.2) that the ﬁrst solution in (14.73) represents a local maximum while the second solution corresponds to the true minimum of the potential. we have noted earlier in section 7. 2m v=√ . furthermore. the weak interactions are short ranged so that we would like the gauge bosons to be massive.73) φ φ† φ = = Furthermore. λ (14. (14. there are two solutions to (14. 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. λ (14. Furthermore.67) in the form λ † 2 φφ . 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. 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). (14.3 The standard model 605 14. φ† = −m2 φ† φ + λ > 0.

(14.77) σ · Wµ = 2 = 1 2 √ + 2Wµ φ+ φ0 . (14. the other parts will not change under the shift.) 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 .76) where we have deﬁned 1 ± 1 2 Wµ = √ (Wµ ∓ iWµ ).65). √ − 3 2Wµ −Wµ (14.75) However. before shifting the ﬁelds let us rewrite the Lagrangian densities for the Higgs and the Yukawa sector of the theory for simplicity (see (14.68)).67) and (14. this is the sector of the theory that involves the scalar ﬁeld and will be aﬀected by the shift.606 14 Higgs phenomenon and the standard model Therefore. as we have discussed earlier. 2 As a result. (14.58) we obtain . (Basically. 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+ . it follows that 3 Wµ √ − 2Wµ (14.78) 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.

3 The standard model 607 LH + LY − V φ. 2 (14. 4 (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 . 2 v 0 0 φ → φ +√ . φ† = ∂µ + ig − 0 ig 3 ig′ v Yµ − Wµ φ− − √ Wµ φ + √ 2 2 2 2 .79) The expansion around (14.14.74) can be done equivalently by letting v φ0 → φ0 + √ .80) under which the Lagrangian density (14.79) in the Higgs and the Yukawa sector becomes LH + LY − V φ.

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. the complete quadratic Lagrangian density is obtained to be . The spectrum of the theory can be obtained from the quadratic part of the Lagrangian density and the quadratic part of (14.63) respectively.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 (14.82) hv λv 2 0 0 2 φ +φ . Adding the quadratic terms from the gauge and the fermion Lagrangian densities in (14. − √ (eR eL + eL eR ) − 8 2 0 3 We note that the ﬁeld combinations (φ0 + φ ). we cannot truly determine the spectrum of the theory until we diagonalize the quadratic part of the Lagrangian density. (g′ Yµ + gWµ ) as well ± appear to have become massive. However. since as the ﬁelds Wµ there is mixing between various ﬁelds.60) and (14.81) .

8 2 It is now a straightforward matter to diagonalize the Lagrangian density (14.83).84) where θW is known as the Weinberg angle (also called the weak mixing angle). (14. We can invert the relations in (14.85) . (14. Yµ = Zµ sin θW + Aµ cos θW . Aµ = g g2 + g′2 Yµ − g′ g2 + g′2 3 Wµ 3 = cos θW Yµ − sin θW Wµ .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 (14.83) + λv 2 0 hv 0 2 − √ (eR eL + eL eR ) − φ +φ .14.84) to write 3 Wµ = Zµ cos θW − Aµ sin θW . Let us deﬁne g′ g2 + g′2 g g2 + g′2 Zµ = Yµ + 3 Wµ 3 = sin θW Yµ + cos θW Wµ .

if we deﬁne the parameters gv . we can deﬁne (this is equivalent to deﬁning φ0 = iχ).29)) σ = χ = 1 0 √ φ + φ0 . 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 ) .86) 1 √ (σ 2 ∂µW ν 3 − ∂ν W µ 3 Similarly. we can write the quadratic part of the Lagrangian density (14. (14. 2 MW g2 + g′2 v = .89) MW MZ = = me = 2 MH = then together with (14. 2 λ 4m2 λv 2 = 2 2 λ = 2m2 . 2 2 (14. see.87) so that we can write 0 ∂µ φ ∂ µ φ0 = 1 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ. (14. φ = 0 1 √ (σ 2 + − iχ). (14. for example.89). 2 i 0 √ φ − φ0 .84)-(14.610 14 Higgs phenomenon and the standard model With this. 2 (14. 2 cos θW hv √ .88) Furthermore.83) as .

eR to deﬁne the massive electron ﬁeld as (see (14.(14.56)) e = eL + eR .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 .91) .90) 2 2 where we have combined the two chirality states eL . If we now redeﬁne the (massive) gauge ﬁelds as (14.14.

94) while the neutrino remains massless as we would like. namely.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 .38)). The Higgs √ particle has the usual mass MH = 2m (see (14.95) One of the two neutral vector bosons. 2 (14. 2 (14. Zµ is also massive with mass . we see that after spontaneous breakdown of symmetry the electron has become massive with mass hv me = √ . MW 1 ∂µ χ → Zµ . Although the starting theory had a massless electron. MZ (14. The charged ± vector (spin 1) bosons Wµ are massive with mass MW = gv .93) The quadratic Lagrangian density (14.93) is now completely diagonalized and shows some very desirable features.612 14 Higgs phenomenon and the standard model ± Wµ ∓ i ± ∂µ φ± → Wµ .92) Zµ − then the quadratic Lagrangian density (14. 1 2 2 M σ 2 H (14.

.97) There is another neutral vector boson in the theory (14.84)) g′ g2 + g′2 g g2 + g′2 sin θW = cos θW = . 2 cos θW (14. That this is indeed true can be seen as follows.93). 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).96) where the Weinberg angle θW is deﬁned by (see (14.14. namely. Aµ which is massless. (14.63) 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 . Let us look at the minimally coupled fermion Lagrangian density (14.3 The standard model 613 MZ = MW g2 + g′2 v = .

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.98) that the standard model has. in addition to the correct electromagnetic current. (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.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 .99) Thus. 2 2 2 g g + Jµ = − √ eL γµ νeL = − √ eL γµ (1 − γ5 ) νeL .98) We note from (14. 2 2 2 . both charged as well as neutral weak currents given by g g − Jµ = − √ ν eL γµ eL = − √ ν eL γµ (1 − γ5 ) eL . (14. We also note from (14.

100) We recall that the V − A theory describes only charged current interactions. If we look at the charged current interaction terms in (14. 14.103) Experimental measurements lead to a value of the Weinberg angle as .14.3 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 ) .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 . they would lead to a low energy eﬀective current-current interation Hamiltonian density in the lowest order of the form (basically.98).102) which leads to 2 MW = (14. 8MW (14. W Figure 14.49) (for just one family of leptons) and comparing the two we determine G √F 2 g2 . (14.101) This has exactly the form of the V − A four-fermion interaction in (14.1: Charged current-current interaction in the lowest order. 2 8MW g2 e2 √ = √ . 4 2GF 4 2GF sin2 θW = (14. two charged currents interact through the charged W gauge boson propagator (exchange) as shown in Fig.

Such processes have also been experimentally observed. 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. the standard model diﬀers from the conventional Fermi theory in (14.4 References 1. 347 (1936). A. (14. Proca. (14. e2 √ 4 2GF sin2 θW MW ≃ 90 GeV. de Phys. the WeinbergSalam-Glashow theory does not predict a unique mass for the Higgs particle since its mass depends on the quartic coupling λ. Search for Higgs particles is a top priority in many experiments at the upcoming LHC. However. cos θW 1 2 ≃ 80 GeV. J. The standard model compares very well with the experimental results.104) and using the values of other constants (see.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.23. et le Radium 7.51) and recall that we have set = c = 1) e2 1 ≃ . for example. 14. .105) These gauge bosons have been discovered and their masses are determined very close to the theoretically predicted values. 4π 137 we can now determine MW MZ = = GF ≃ 10−5 (GeV)−2 . Once again these can be accomodated into extensions of the standard model that we will not go into here. although bounds on the value of its mass can be put from various other arguments. Furthermore.616 14 Higgs phenomenon and the standard model sin2 θW ≃ 0.

u 3. 10. Physical Review 145. S. 167 (1971). . Physical Review 109. 193 (1958). J. S. Feynman and M. St¨ckelberg. Ruegg and M. Proceedings of PaduaVenice conference on mesons and newly discovered particles. 7. Marshak. 6. 321 (1964). Englert and R. E. Higgs. Reviews of Modern Physics 46. Physical Review 109. 1156 (1966). 5. Physical Review Letters 19. R. Helvetica Physica Acta 30. Hagen and T. G. C. C. G. 585 (1964). Nuclear Physics B35. Bernstein. B. 9. 11. C. G. International Journal of Modern Physics A19. 1860 (1958). ’t Hooft. 4.14. 209 (1957). W. 7 (1974). Ruiz-Altaba. E. P. Guralnik. E. W. Weinberg. P. Brout. Kibble. (1957). Sudarshan and R. Physical Review Letters 13. G. F. 8. H. Gell-Mann. 3265 (2004). 1264 (1967). Physical Review Letters 13.4 References 617 2. R.

This page intentionally left blank .

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

620 15 Regularization of Feynman diagrams theories. the scalar φ4 theory described by the Lagrangian density (see (6. namely. is not.1: A few examples of 1PI diagrams in the φ4 theory. therefore. given these Feynman rules. Clearly the 1PI diagrams are fundamental since any other diagram can be built using them as basic elements. . for example. Therefore. 15.2.1) where the coupling constant λ is assumed to be positive. (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. − M2 p p4 p3 = iGF (p) = p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). we can calculate any perturbative amplitude in the φ4 theory.1 (which topologically involve loops and. The Feynman rules for this theory are given by i . we can construct the graphs in Fig. With these Feynman rules. are called loop diagrams) Figure 15. which are 1PI graphs whereas the graph in Fig. 2 2 4! L= (15. it is suﬃcient to concentrate on the 1PI graphs. in studying quantum ﬁeld theories at higher orders. 15. In fact.36)) M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ .

problems with divergences arise in a quantum ﬁeld theory at the loop level. Therefore. 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. 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.2 Loop expansion As we have mentioned repeatedly.1 and Fig. 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. ∂µ φ) . Since the 1PI graphs are fundamental. this discussion applies to any quantum ﬁeld theory) L (φ. it is suﬃcient to study the divergence behavior of only the 1PI graphs. 15. in evaluating scattering amplitudes particularly at the loop level (where the topology of the diagram involves loops as in Fig. in general. and P is the power of “a” associated with this graph.15. (15. let us discuss the notion of loop expansion in a quantum ﬁeld theory in this section. As we have mentioned. Thus if we are considering a proper vertex graph (1PI graph). to study the divergence structure of any amplitude (graph).2 Loop expansion 621 Figure 15. a) = a−1 L (φ. then it is clear that each interaction vertex in the theory will have a factor a−1 multiplied to it. 15. we invariably run into divergences. the integrals over the arbitrary loop momenta may not. we are forced to give a meaning to the Feynman amplitudes in some manner and this process is known as regularization of the diagrams.2: An example of a diagram in the φ4 theory which is not 1PI. 15. be convergent.2). then it follows easily that .3) where a is a constant parameter. ∂µ φ. Namely.

3). Let us recall that in the path integral description of a quantum ﬁeld theory. (15. . (15.622 15 Regularization of Feynman diagrams P = I − V.7) where represents the Planck’s constant and we see that it can be thought of as an overall multiplicative parameter like “a” in (15.4) where I denotes the number of internal lines and V the number of interaction vertices in the graph. therefore. therefore. 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 . the generating functional is deﬁned as (see. 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). (12.4) and (15. the number of loops (independent momenta) in such a graph.5) where the term (+1) reﬂects the fact that every 1PI vertex function has an overall energy-momentum conserving delta function associated with it.5).120).6) In other words. the power of “a” in a diagram in such a theory also determines the number of loops in the diagram and can. we can eliminate I − V to obtain L = P + 1. (15. for example. 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. (15. is given by L = I − V + 1. we have set = 1 throughout our discussion. be thought of as the parameter of expansion in the number of loops. From the two relations in (15.

This is quite important.3 Cut-off regularization 623 As a result. Loop expansion is a diﬀerent perturbative expansion. 15.3 Cut-oﬀ regularization To get an idea of how divergences arise in a quantum ﬁeld theory at the loop level. We are usually accustomed to the idea of perturbation in powers of the coupling constant in the theory.5 and 14.6). when we perturb around the true vacuum (by shifting the ﬁeld variable by its vacuum expectation value). the entire perturbation expansion has to be rearranged which is nontrivial. In addition the small value of makes it a legitimate expansion (perturbation) parameter. 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=− . Therefore. let us calculate the simplest nontrivial graph in the φ4 theory (15. for example.15. expanding in powers of in a quantum ﬁeld theory also corresponds to expanding in the number of loops as described in (15. The reason why such an expansion is advantageous can be seen from the fact that since this parameter multiplies the whole action. when we study theories where there is spontaneous breakdown of a symmetry.2). as we have seen (see. the expansion is unaﬀected by how one divides the total action into a free part and an interaction part. In such a case. On the other hand. the scalar ﬁeld picks up a vacuum expectation value which depends on the coupling constant of the theory. sections 7. for example. 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.1) which gives the one loop correction to the two point function.

(This is the reason for the name cut-oﬀ regularization. the two point function in (15.8) that this integral is divergent and. y + M2 (15.8) dy 1 − Here we have used the fact that the Feynman diagram in (15.8) takes the form .9) = 2π 2 . 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. 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. We see from (15.8) 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. Furthermore. therefore. such divergences are also known as ultraviolet divergences. With this regularization.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 .) Since large values of momentum correspond to small values of coordinate separation.

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 . 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. Λ are scalar parameters).577215665. since there is no momentum vector in the ﬁnal result in (15. 2 ˜ 32π M Λ2 (15. in this case this is also the result in the Minkowski space. However.10) (and m.15. We note that we can write (15. (15.10) The result at the end should be rotated back to the Minkowski space. (15.12) .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( ) .

Gradshteyn and Ryzhik 3.214. This result coincides with (15. Ek (15. deﬁning y = |k|.1).14) does not depend on the angular variables. for example. Since the integrand in (15.8) 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 . we also note here that the diagram in (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 ) .10) with the identiﬁcation ˜ Λ2 = Λ2 − M 2 γ.4 and 8.352. (5. which gives (4π) in three dimensions. (15. the integration over the angles can be carried out. Furthermore.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.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 + .14) d3 k where we have identiﬁed (see. (15.13) Furthermore.47)) Ek = |k|2 + M 2 . for example.

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 (15.10) has been transformed into an infrared divergence in the Schwinger parameter τ . 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 .15. Gradshteyn and Ryzhik.8) can also be evaluated in an alternative manner which is sometimes useful and so let us discuss the alternate method as well. The integral in (15.4). the ultraviolet divergence in (15.17) dτ −τ M 2 .17) is well behaved for large values of τ . This result coincides with (15. but is divergent at the lower limit τ = 0.271. We note that the integrand in (15. for example.2 and 2. e τ2 Here Γ(2) denotes the gamma function and the parameter τ is known as the Schwinger parameter (or the proper time parameter). Thus.10) if we identify 1 ¯ (15.3 Cut-off regularization 627 Here we have used the standard table of integrals in the intermediate step (see. 2. 2 This derivation makes it clear that the evaluation of the amplitude is much easier in the Euclidean space.271.16) Λ2 = Λ2 − M 2 (1 − ln 2) .

Gradshteyn and Ryzhik 3.471.18) 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.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 . 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. For simplicity we would consider all the incoming momenta to vanish.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 be compared with (15. the one loop diagram for the vertex correction shown in Fig. but quantum ﬁeld theories can be deﬁned in any number of space-time dimensions).10) and we see that both ways of evaluating the integral lead to the same result. for example. 15. namely.9) to evaluate the integral over τ . However. Without going into details. For example. Equation (15. (15. We note that since the cutoﬀ Λ has to be actually taken to inﬁnity in the ﬁnal result. Let us next look at another one loop graph. this graph is divergent.10) (or (15. we note from (15. regularizing the integral brings out the divergence structure of the Feynman graph.18) can also be obtained by simply cut1 ting oﬀ the τ integral at the lower limit Λ2 (without the regularizing exponential factor). 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) .3. we note that the result in (15.

19) Λ→∞ lim There are several things to note from this derivation. First we have factored out an overall momentum conserving delta function along .3 Cut-off regularization 629 p3 p2 + p1 p4 p1 p3 p2 p4 p4 p3 + p1 p2 Figure 15. 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 . (15.15.3: One loop correction to the four point vertex function.

) 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. in general. As a result. for the diagrams in Fig. but the divergence in this case is logarithmic.19). Thus we see that this graph is also divergent (in four dimensions). Similarly. Consequently. as we have seen earlier. However. This is. the cutoﬀ regularizes and deﬁnes .630 15 Regularization of Feynman diagrams with the factor of (2π)4 . can be removed by normal ordering the Hamiltonian of the theory. the present divergences are nontrivial and are present even after normal ordering of the theory. of course.3. we are free to redeﬁne the value of the ground state energy to zero. In the examples we have studied in (15. The divergence structure of any diagram can be determined simply by counting the powers of momentum in the numerator and in the denominator.9) in the intermediate step. for the ﬁrst graph in (15. any given quantum ﬁeld theory. First we introduce a regularization procedure (scheme) which gives a meaning to the divergent Feynman integrals by isolating the divergent parts of the diagram. In fact. Thus. for example.8) up to a multiplicative factor.19) is ∂ given by − ∂M 2 of that in (15. each of these diagrams has an associated symmetry factor of 1 2 and we have used the value of the angular integral from (15. 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. Very broadly it consists of two essential parts. Renormalization is the process of redeﬁning these inﬁnities that we encounter in perturbation theory. 15. 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.10) and (15. there is a net 4 − 4 = 0 power of momentum in the numerator reﬂected in the logarithmic divergence of the diagram. Finally. is plagued with divergences at every (loop) order in perturbation theory (unless we are in lower dimensions). equivalent to saying that since we only measure diﬀerences in the energy levels. 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. (This can also be seen from the fact that the integral 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.

15. p−m / p − m2 = −(2π)4 igδ4 (p1 + p2 + p3 ). − M2 p = iGF (p) = p2 p p3 p2 p1 p4 p3 = iSF (p) = i(p + m) / i = 2 . in general. Let us note that any Feyn- man diagram (amplitude). But let us examine the dependence of diagrams on external momenta (and this will be important in studying renormalization) in another theory.19). (15.15.3 Cut-off regularization 631 the integrals.3.1 Calculation in the Yukawa theory. depends analytically on the external momenta. the lack of dependence on the external momenta is simply due to our special choice of the external momenta to be all vanishing.21) . In the previous examples. particularly in the second example (15. (15. namely.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 . 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. But in this process the amplitude (and the theory) becomes cutoﬀ dependent. p1 p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). the theory with a Yukawa coupling described by the Lagrangian density 1 λ M2 2 / φ − gψψφ − φ4 .

In quantum ﬁeld theory at zero temperature. however. Before we proceed to calculate higher order diagrams in this theory. AB 1 1 A − B x(A − B) + B dx (x(A − B) + B)2 1 0 (15. Suppose we would like to write AB as a single factor.22) Therefore. Therefore. diagrams necessarily involve more internal propagators. (15. 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 ) . The lowest order graph involving the fermions in this theory gives rise to the Yukawa potential (as we have seen in section 8.632 15 Regularization of Feynman diagrams where the arrows in the vertices denote the fact that momenta are all incoming. then this formula needs to be generalized (which is important in studying ﬁnite temperature ﬁeld theories). it would be useful to ﬁnd a way to combine denominators and this is achieved by 1 Feynman’s formula as follows. let us note that at higher orders in perturbation theory.9) which is why the cubic coupling involving fermions is called the Yukawa coupling. the iǫ term with the same sign). On the other hand when the two factors have opposite analytic behavior (iǫ terms with opposite sign). then we note that 1 0 1 dx (xA + (1 − x)B)2 = 0 = − = − = 1 1 1 − A−B A B 1 . (x1 A + x2 B)2 (15.23) = 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.23) holds and is quite .

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 .23).26) Using the Feynman combination formula (15.23) 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. 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 / / .3 Cut-off regularization 633 useful.121). let us next calculate the two simplest nontrivial graphs involving loops in this theory. / / (15. the integral in (15.24) which we have already used in (13.15. ( n xi Ai )n i=1 (15. Equipped with the Feynman combination formula. Generalization of (15.25) 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) .

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.28) In (15. E (15. 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) / . Furthermore.27) where we have deﬁned Q2 = x(1 − x)p2 + xm2 + (1 − x)M 2 .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 . (15.

where it is understood that the limit Λ → ∞ is to be taken. 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.30) takes the form . (15.30) p = −Tr (−ig)2 = −g2 Tr (k + m)((k + p) + m) / / / d4 k . Λ. M.31) Tr A = 0.15. We also note here that we have used the integral from (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. / Tr A / = 4A · B. Using the trace identities Tr ½ = 4. g).29) = f (p. Similarly the scalar self-energy graph is given by k p k+p d4 k i i 4 k − m (k + p) − m (2π) / / / (15.19) in the intermediate step and have rotated back to Minkowski space in the ﬁnal step. 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). Furthermore. /B the self energy graph in (15.

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 .

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. Λ). This is because the overall momentum conservation eliminates one of the momentum variables. These two simple calculations show that Feynman amplitudes are functions of external momenta.33) +ﬁnite terms as Λ → ∞ . 2 (15. Similarly.29). 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. for the fermion self-energy in (15. for the scalar self-energy in (15. In the examples worked out above it is clear that one can Taylor expand the amplitudes around zero momentum of the external lines.32) we can write . Thus. for example.15. In fact. m. g. a Feynman amplitude with n external lines is an analytic function of (n − 1) external momentum variables.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.

Similarly. a cut-oﬀ can lead to violation of gauge invariance in gauge theories by giving a mass to the gauge boson. it is clear that such a regularization violates manifest Poincar´ invariance of the e quantum theory. To avoid this problem.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. 15. in massless theories we Taylor expand the amplitudes around a nonzero but ﬁnite value of the external momentum. Therefore. it is useful to ﬁnd . 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). Let us note here that if the theory is massless. then expanding around zero external momenta would be disastrous.4 Pauli-Villars regularization Although regularizing a Feynman diagram by cutting oﬀ contributions from large values of momentum seems natural.34) +ﬁnite terms as Λ → ∞ . Furthermore. while quite useful. leads to a lack of manifest rotational invariance (although in the continuum limit this symmetry is recovered). 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.

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 . For example.80) and (9. we can now calculate various 1PI amplitudes in QED. p2 p q p = iSF (p) = i(p + m) / .35) ν = iGF .µν (p) = − iηµν .36) With these Feynman rules.(15.125)) 1 1 /ψ L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 . 2 − m2 p r µ = −(2π)4 ieγ µ δ4 (p + q + r).15. The Pauli-Villars regularization provides such a regularization scheme. let us consider QED (quantum electrodynamics) in the Feynman gauge (see (9.4 Pauli-Villars regularization 639 a covariant regularization scheme which also respects gauge invariance. 4 2 which leads to the Feynman rules µ p (15. To discuss the Pauli-Villars regularization scheme.

µ))2 dx ((1 − x)pE + 2m) / 2 dkE 1 Q2 (m. µ) − 2 2 kE + Q2 (m.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. 2 kE + Q2 (m. µ) = x(1 − x)p2 + xm2 + (1 − x)µ2 .27)).26).112) as well as (15.37) ie2 8π 2 dx((1 − x)pE + 2m) / where we have deﬁned Q2 (m. (The Dirac gamma matrices have also been rotated to Euclidean space as discussed in (15.38) 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.) The momentum integral is. µ) (15. µ))2 2 / kE (2(1 − x)pE + 4m) 2 (kE + Q2 (m. In the . E (15. of course.9) for the angular integral. µ))2 ∞ 0 2 dkE −1 . µ) (kE + Q2 (m.111) and (2. 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.

37) except for a sign and µ → Λ.15. 2 kE + Q2 (m. Q2 (m. let us add to the Lagrangian density of QED (15. 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.) 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. / ′ L′ = (15. we can write the additional contribution to the fermion self-energy as ie2 8π 2 ∞ 0 2 dkE −1 . Namely. without doing any further calculation. decouple. µ) (15. µ and ψ ′ . Λ) . (The covariant derivative in (15. these additional interacting ﬁelds give an additional contribution to the fermion self-energy (15. A′ are ﬁctitious heavy µ ﬁelds introduced to regularize diagrams.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. therefore. Λ) (15. The idea is to take Λ → ∞ at the end and in that limit the heavy ﬁelds do not propagate and. The contribution from the heavy photon to the self-energy would be exactly the same as what we have already calculated in (15.41) I (reg) = I + I ′ = − dx ((1 − x)pE + 2m) ln / .4 Pauli-Villars regularization 641 Pauli-Villars regularization.40) I′ = dx ((1 − x)pE + 2m) / so that the eﬀective regularized fermion self-energy can be written as ie2 8π 2 Q2 (m.37) coming from the diagram where a heavy photon is being exchanged. Therefore.39) is deﬁned with the gauge ﬁeld combination Aµ + A′ . However.39) where we assume that Λ ≫ m.

The ﬁnal result can now be rotated back to Minkowski space and has the form ie2 = 2 8π Q (m.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 .642 15 Regularization of Feynman diagrams which is ﬁnite for any given value of Λ.43) . (15. The Feynman rules (15. We can also calculate the one loop photon self-energy in QED in a similar manner. µ) 2 2 I (reg) dx ((1 − x)p − 2m) ln / . Λ) Q (m. µ) = −x(1 − x)p2 + xm2 + (1 − x)µ2 .42) where Q (m. (15.

(k2 .45) individually.44) The overall negative sign in (15. η00 → E 2 ΠE .43) reﬂects the fact that we are evaluating a fermion loop. p0E ). To evaluate the momentum integral.4 Pauli-Villars regularization 643 where we have used the gamma matrix identities in (2. Π → (i)ΠE ) and we have δ00 .15.45) where Q2 (m) = x(1 − x)p2 + m2 .43) to 2 Euclidean space ((k0 .115) and (2. p2 ). 2 (15. 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 . we can rotate (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 . p2 ) → −(kE . p0 ) → i(k0E . + y + Q2 (m) (y + Q2 (m))2 = d4 kE E 1 (−δµν ) + Q2 (m) . We can now integrate each of the E terms in (15. 2 (kE + Q2 (m))2 (15.116) and have deﬁned Q (m) = x(1 − x)p2 − m2 .

This integral. . 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. is quadratically divergent. as we see.47) 1 + δµν − 2 dy + Q2 (m) . Putting this back into (15. recalling that the photon has coupling to ﬁctitious heavy fermions. (15. However. y + Q2 (m) d4 kE 2 (kE + Q2 (m)) 2 − k2 δµν 2 E + Q (m) (Q2 (m))2 .9) in the intermediate steps. we obtain for the self-energy of the photon (in Euclidean space).46) =− π2 δµν 2 dy 1− (y + Q2 (m))2 Here we have used (15.45).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) .

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) . (15. Such a term can be cancelled by adding another set of heavy fermion ﬁelds. Λ2 and charges c1 e. see. Then the regularized photon amplitude with these two sets of heavy fermions will have the form iΠE (reg) (pE ) = iΠE (pE .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. for example. the regularized photon self-energy will have the form iΠE (pE . m) − iΠE (pE . Λ1 ) − ic2 ΠE (pE . m) − ic1 ΠE (pE .48) This shows that the diagram is ﬁnite for any given value of Λ. However.1) which would violate gauge invariance. As a result. section 14. 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 . this has the unpleasant feature that the photon self-energy has developed a mass term (term proportional to δµν . Let us assume that we have two sets of heavy ghost fermions with masses √ √ Λ1 . c2 e respectively.15.

· · · . 2.49) 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) . c1 Λ2 + c2 Λ2 = m2 .52) holds for any covariant gauge with an arbitrary value of the gauge ﬁxing parameter ξ.50) then.(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) .52) We note here that even though we are working in the Feynman gauge. since the diagram for the photon self-energy in (15.646 × ∞ 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 ) . the quadratic divergences as well as the logarithmic divergences in (15. in a gauge invariant manner. Pauli-Villars regularization involves introducing.49) E 1 2 It is clear. therefore. n depending on the nature of the divergence in the diagram such that the regularized amplitude . 1 2 (15.43) does not involve the photon propagator. the result (15. (15. a set of heavy ﬁelds with masses Λi and √ charges ( ci e) with i = 1. that if the parameters satisfy c1 + c2 = 1.(15.51) which represents a gauge invariant photon self-energy (it is manifestly transverse). In general.

For example.54) will be ﬁnite and gauge invariant. i i ci Λ2 = m2 . 15. In fact. (15.53) with the conditions on the parameters given by ci = 1.15. it is quite useful in studying non-Abelian gauge theories for which the Pauli-Villars regularization does not work. (k2 − m2 ) ((k + p)2 − M 2 ) (15. 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).) We should mention here that dimensional regularization has its own problems which we will discuss at the end of . it is by construction a gauge invariant regularization scheme. However.5 Dimensional regularization 647 n I (reg) = I(p. 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 less than four dimensions it is ﬁnite. Furthermore.4 which cannot be regularized by the Pauli-Villars method. the integral d4 k .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. i ··· .55) is logarithmically divergent in four dimensions. i=1 (15. since gauge invariance is independent of the number of space time dimensions. there are graphs of the form shown in Fig. m) − ci Ii (p. (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. In non-Abelian theories. Λi ). This procedure is known as dimensional regularization. however.

2 [M ] = 1.56) carries a dimension and introducing an arbitrary mass scale µ we can then write the Lagrangian density (15. (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 . In dimensional regularization we analytically continue the theory to n dimensions.58) are given by .57) Therefore. [φ] = [λ] = 4 − n.4: One of the one loop graphs contributing to the gluon self-energy. the coupling constant for the quartic interaction in (15. 2 2 4! L= (15. The Feynman rules for the theory (15.56) for the theory in n dimensions as M 2 2 λµ4−n 4 1 ∂µ φ∂ µ φ − φ − φ .58) so that the coupling constant λ is rendered dimensionless. For example.648 15 Regularization of Feynman diagrams this section.56) we can carry out the (canonical) dimensional analysis in n dimensions to determine n−2 . Figure 15. 2 2 4! L= (15. but it is worth pointing out here that manipulations with dimensional regularization are extremely simple which is why it is quite useful.

59) We need to be careful about manipulating tensors in n dimensions.5 Dimensional regularization 649 p = p2 i . we can separate out the angular part of the integral . p1 p2 (15. therefore. For example.15. − M2 p4 p3 = −(2π)4 iλµ4−n δn (p1 + p2 + p3 + p4 ). 2 (kE + Q2 )α where we have rotated to Euclidean space. (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. E To evaluate this integral.61) I = = = = dn kE dn kE 1 . it follows that in n dimensions we have µ ηµ = n. let us note that the integrand is spherically symmetric (independent of angular coordinates) and. shifted the variable of integration and have deﬁned Q2 = M 2 − p2 . since each vector index takes n values.60) Furthermore. to calculate amplitudes in n dimensions.

where. in n dimensions. E where we have deﬁned y = 2 in the intermediate step and have used the deﬁnition of the gamma function. . we can write dΩ = dθ1 sinn−2 θ1 dθ2 sinn−3 θ2 .64) n = 2 2 −1 Γ n 2 k2 dΩ.65) Let us compare this with known results in lower dimensions that we are familiar with. . We see that .63) On the other hand.62) The integral over the angles dΩ can be evaluated from the simple Gaussian integral as follows.650 15 Regularization of Feynman diagrams n−1 dn kE = kE dkE dΩ. dθn−1 . n ≥ 2. we can evaluate the integral in (15. Let us consider the basic n dimensional Gaussian integral whose value is given by dn kE e− 2 kE 2 n = (2π) 2 . (15. (15. This determines the value of the angular integral to be dΩ = 2π 2 (2π) 2 = .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. n −1 n Γ n 22 Γ 2 2 n n (15.

68) Therefore. for the basic integral (15.5 Dimensional regularization 651 n=2: n=3: n=4: dΩ = dΩ = dΩ = dθ = 2π. (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 .66) which agree with what we already know from explicitly doing these integrals.15. 2 n . dθ1 sin2 θ1 dθ2 sin θ2 dθ3 = 2π 2 . we see that with the identiﬁcation n . dθ1 sin θ1 dθ2 = 4π.67) Let us recall that the beta function is deﬁned as Γ(p)Γ(q) =2 Γ(p + q) ∞ 0 −p−q B(p. 2 p= q =α−p=α− (15. q) = dt t2p−1 1 + t2 . Thus. (15. (15.67) can be written as .69) the integral (15.

For example. n Γ α − (2π)n Γ(α) (p2 + M 2 )α− 2 2 n (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 .71) by diﬀerentiation. using (15. (2π)n Γ(α) (p2 + M 2 )α− 2 (15. (2π)n Γ (α) (Q2 )α− 2 n n n 2 Γ α− Γ(α) n 2 (15. in fact. any other formula can be obtained from (15.72) Similarly we can obtain .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 .71) This basic integral generates all other integrals that we need for evaluating amplitudes in n dimensions and.70) Rotating back to Minkowski space.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 .

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 . First of all.73) n kµ kν dn k n (k 2 + 2k · p − M 2 )α (2π) and so on. the amplitude (15.8)) 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 .15. the one loop scalar self-energy takes the form (see also (15. At the end of our calculations we should.58). let us now calculate various amplitudes in the φ4 theory (15.74) Let us next set n = 4 − ǫ. 2 (2π)n (M 2 )1− 2 n n = − (15. take the limit n → 4 which translates to ǫ → 0.. we are analytically continuing away from four dimensions (to a lower dimension where the integral is well deﬁned). (15. With these basic integration formulae.e.74) becomes . of course. With this. i.

we have π2 (2π)n n = ≃ 1 1 n = ǫ (4π) 2 (4π)2− 2 ǫ 1 1 + ln 4π . ǫ (15. the scalar self-energy (15.75) Let us next work out some of the identities involving the gamma functions that will be useful to us. Similarly. ≃ 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).75) takes the form .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 . 2 (2π)n (M 2 )−1+ 2 n (15. 16π 2 2 (15. In the limit ǫ → 0. we have n 2 n Γ 2− 2 Γ 3− ǫ ǫ ≃ 1 − γ.12).76) where γ denotes the Euler’s constant deﬁned in (15.77).77) Using (15.76) as well as (15. 2 2 2 ǫ ǫ = Γ = Γ 1+ 2 ǫ 2 ǫ 2 2 1 − γ) = − γ.

19). we can calculate the one loop correction to the vertex function shown in Fig. for simplicity. 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 . 15. Similarly.122). we will set all the external momenta to vanish as we had done earlier in (15.15.78) We note here that for M 2 = 0.3 and.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) . − + ln ǫ 4πµ2 (15. this graph would be regularized to zero in dimensional regularization which is the type of argument used in (13. In this case.

ǫ 4πµ2 (15. the amplitudes diverge. we let e → e µ2 . 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− .77). 2 (15. /ψ 4 2 (15. 2 2 2 ǫ .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 .79) where we have used (15. to make the coupling constant dimensionless. We note that these amplitudes are well behaved (regularized) for any ﬁnite value of ǫ.80) In n dimensions. 2 2 n−1 3 ǫ ψ = = − .76) and (15. ǫ (15.82) .81) As a result. However.5. Let us next calculate one loop ampli- tudes in QED described by the Lagrangian density in the Feynman gauge (see (15. as we take the limit ǫ → 0 (to go to four dimensions).35)) 1 1 L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 .1 Calculations in QED. 15.

the Feynman rules of the theory in n = 4 − ǫ dimensions are given by µ p ν = iGF . 2 − m2 p r µ q p = −(2π)4 ieµ 2 γ µ δ4 (p + q + r). ǫ (15. (15.15.85) = −ieµ 2 2 = −e2 µǫ / / dn k γ µ ((k + p) + m) γµ . p2 p = iSF (p) = i(p + m) / .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 (15.83) As a result. we can calculate the fermion selfenergy at one loop which gives (see also (15.µν (p) = − iηµν .84) ǫ With these Feynman rules. n k 2 ((k + p)2 − m2 ) (2π) .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.

658 15 Regularization of Feynman diagrams We note here that if we use the algebra of the gamma matrices (1. ǫ 4πµ2 (15. / / (15. γµ kγ µ = (2 − n)k.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 (15.60) we obtain γµ γ µ = n.86) Using this. the one loop fermion self energy in (15.87) (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 (15.88) 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 .79) in n dimensions as well as (15.89) .

42). Tr A = 0.116)).115) and (2. n (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Using the n-dimensional identities (see (2.15. /B Tr A /C = 0. /B / Tr γµ γν γλ γρ = n (ηµν ηλρ − ηµλ ηνρ + ηµρ ηνλ ) . 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 (15.90) 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 .5 Dimensional regularization 659 where we have deﬁned Q2 = −x(1−x)p2 +xm2 and have used (15. Tr ½ = n.76) as well as (15.77). (15.90) / / / dn k Tr γµ (k + m)γν (k + p + m) . / Tr A / = nA · B. This can be compared with (15.91) the photon self-energy (15.

92) where we have deﬁned Q2 = −x(1 − x)p2 + m2 .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 . (k2 − Q2 )2 (15. (15.93) × ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν where we have used (15.77). Let us next look at the terms with kµ kν in the numerator separately which leads to .73).76) as well as (15.92) contains three diﬀerent kinds of terms which can be integrated using the standard formulae in (15. The integral in (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 .71)(15.

94) Q2 1 2 +γ− × − + ln 2 ǫ 4πµ 2 Similarly. 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 . (15.5 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.

For simplicity we will put the momentum of the external photon equal to zero.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. the gauge ﬁxing Lagrangian density does not receive any quantum correction which can also be seen from the BRST identities for QED. since the photon self-energy does not involve the photon propagator. Let us next calculate the one loop amplitude corresponding to vertex correction (charge renormalization). × − + 2 ln ǫ 4πµ2 Adding the contributions from (15. this result holds in any covariant gauge with the gauge ﬁxing term ξ arbitrary.93)-(15. the photon self energy in (15. 2 ǫ 4πµ 2 (15.96) =− dxx(1 − x) This shows that the photon self energy graph is completely transverse which is consistent with gauge invariance (see also (15.52)). The photon does not acquire a mass term and we note here that even though we have chosen to work with the Feynman gauge.92) takes the ﬁnal form ie2 (ηµν p2 − pµ pν ) 2π 2 2 Q2 1 − ln −γ− .95) Q2 4 + 2γ .95). 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 . Since there is no longitudinal term in the photon self-energy.

86) we have (15. / / Therefore. 2 ((k + p)2 − m2 )2 k (15.99) = −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 / / .23) as well as the identity .15.5 Dimensional regularization 663 (15.97) as dn k (2π)n k2 ((k + p)2 − m2 )2 (15. we can write the amplitude (15. /γ /γ / as well as (15.97) = −e3 µ 2 3ǫ / / / / dn k γ λ (k + p + m)γ µ (k + p + m)γλ .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 (2.100) Let us next use (15.113)) γ λ A µ A λ = (2 − n)(2Aµ A − γ µ A2 ).

1 k2 ((k + p)2 − m2 )2 3. 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 .664 15 Regularization of Feynman diagrams 1 AB 2 = − = ∂ 1 ∂ =− ∂B AB ∂B dx dx 2x .100) and shifting the variable of integration.102) where we have deﬁned Q2 = −x(1 − x)p2 + xm2 .101) to combine the denominators in (15. Using this in (15.100) which leads to 1 + p)2 − m2 ) 1 (k + xp)2 − Q2 2x (k + xp)2 − Q2 k2 ((k = = dx dx 2. (15. ((1 − x)A + xB)3 1 ((1 − x)A + xB)2 (15.

105) .87).104) 2 (1 − x)p − 2m pµ / . (15. (15.89) with (15.15.109) 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 (15.88) and (15. we see that the fermion self-energy and the vertex function are related as (see also (9.100). 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 (15.104) 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 (15.103) 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.

666 15 Regularization of Feynman diagrams which we recognize as the Ward-Takahashi identity. Analytic continuation of these to other dimensions is nontrivial. the point splitting method. we will not go into the details of these other methods here. we know that i 4! γ5 = iγ 0 γ 1 γ 2 γ 3 = − µνλρ γ µ ν λ ρ γ γ γ . the ζ-function regularization etc. Physical Review 74. 1439 (1948). (Note that the divergent parts as well as the ﬁnite parts satisfy the identity which demonstrates that dimensional regularization preserves gauge invariance. there exist several other important regularization schemes. Hence if the theory or the physical quantity of interest involves such objects. We simply note here that we select a particular regularization scheme depending on the theory (or the problem) that we are analyzing. for example. It is gauge invariant and extremely simple to manipulate with. However. 15. naive dimensional regularization leads to incorrect answers. J. the higher derivative method. . The lattice regularization is also quite useful.) The power of dimensional regularization is quite obvious from these calculations. Schwinger.6 where we discuss chiral anomaly. it has its own drawbacks. (15. (15. For example. In addition to the regularization methods that we have discussed in this chapter. 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. However.6 References 1. It is not naively applicable if the calculation involves quantities that typically exist only in four dimensions.107) There are several ways to address this issue with the γ5 matrix. We will discuss this further in section 16.106) is deﬁned only in four dimensions (both γ5 and the Levi-Civita tensor µνλρ are manifestly four dimensional quantities).

5. 664 (1951). Introduction to the theory of Quantized Fields. 4. Veltman. Physical Review 82. 8. 20 (1972). N. 434 (1949). Schwinger. 189 (1972). Pauli and F. Giambiaggi. Shirkov. V. CERN preprint (1973). J. . N. Veltman.6 References 667 2. Moscow (1984). R. Feynman. 3. C. G. Nauka.15. Nuovo Cimento 12B. Reviews of Modern Physics 21. ’t Hooft and M. W. Physical Review 76. 769 (1949). Nuclear Physics B44. Villars. 6. 7. Bogoliubov and D. Diagrammar. G. J. ’t Hooft and M. Bollini and J. G.

This page intentionally left blank .

Chapter 16 Renormalization theory 16.1) 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. 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. To deﬁne this let us start with the Lagrangian density for a given theory L = L0 + Li .1) is dimensionless. the Lagrangian density must have canonical dimension 4 in these units (since we are in four space time dimensions. we have also seen that since Feynman amplitudes are analytic functions of external momenta.1 Superﬁcial degree of divergence In the last chapter. (16. the action for the theory (16. In units of = c = 1 which we have been using. Furthermore. in n dimensions it should be n). In our units. it is easy to check that [M ] = [L]−1 = 1. we have seen that loop diagrams in various theories become divergent and need to be regularized. As a result.2) . 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.

7) leads to [ψ] = [ψ] = (16. (The mass term also has the correct dimension with this assignment. / 3 .9) .6) Namely. (φ) (16. the canonical dimension of a scalar ﬁeld (in our units and in four dimensions) is 1. for the Yukawa interaction LY = gψψφ. for example. 2 2 L0 = that since (φ) (16. bi and di to denote respectively the number of fermions. 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.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 . and L0 we have [φ] = 1.8) In fact. (16.3) [∂µ ] = [L]−1 = 1. Let us further introduce the notations fi . bosons and derivatives at an interaction vertex following from the interaction Lagrangian density Li . Thus.) Similarly the free fermion Lagrangian density (ψ) L0 = iψ∂ ψ − mψψ. 2 (16. (16.5) = 4.4) (16.

b = 4.12) LI = hψγµ ψ∂ µ φ. we have or. With these notations we are now ready to introduce the notion of the superﬁcial degree of divergence of a Feynman graph. (16. For example. b = 1. d = 0. we can determine easily the superﬁcial degree of divergence of the following Feynman graphs as . d = 0.14) and so on. 4! (16.10) LI = leads to λ 4 φ .1 Superficial degree of divergence 671 we have f = 2.11) f = 0. Similarly. κψσ µν ψFµν . for an interaction of the form (16. b = 1. the φ4 interaction (16. d = 1. On the other hand.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.16. (16.

D = 4 − 2 × 2 = 0. D = 4 − 2 − 1 = 1.15) If D = 0 for a diagram. diagrams with D = 1 or 2 are known . we say that the diagram is superﬁcially logarithmically divergent. D = 8 − 3 × 2 = 2. D = 8 − 5 × 2 = −2. (16. D = 4 − 2 − 2 × 1 = 0. D = 4 − 2 × 1 = 2. Similarly.672 16 Renormalization theory D = 4 − 2 = 2.

17) The factor 2IB reﬂects the fact that it takes two boson lines at two diﬀerent vertices to form a propagator (internal boson line). Although this graph is superﬁcially convergent. number of internal boson lines. (16.16) ni = There exist topological relations between these numbers. number of external fermion lines. Let us consider a Feynman diagram with B = IB = F IF = = number of external bosons lines. then the Feynman diagram is said to be superﬁcially convergent. 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. number of vertices of the ith type from Li . as we will see it is the notion of superﬁcial degree of divergence which is useful in the study of renormalization theory.18) The superﬁcial degree of divergence is easily seen to be given by . However. (16.16. Rather than calculating the superﬁcial degree of divergence for each graph.1 Superficial degree of divergence 673 respectively to have superﬁcial linear or quadratic divergences.15). 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 . If D < 0 for a graph. Similarly. we have F + 2IF = i ni fi . number of internal fermion lines. (16. The meaning of the adjective “superﬁcial” becomes clear once we look at the last graph in (16.

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-

With the notion of the Taylor operator.4). then D(γ) = 0 and tγ simply evaluates the graph γ at zero external momentum. p=0 (16. p) p p 2! ∂pµ ∂pν tγ Γ(2) (k.4. Note that the BPHZ method subtracts out the divergent parts in the integrand itself so that the renormalized integral is ﬁnite.9: Forest diagrams associated with the simple two loop vertex correction diagram in the φ4 theory. For example.44) Here we have neglected the linear term in the Taylor expansion since it would vanish upon integration because of antisymmetry. We should only remember that when nested boxes exist then the tγ operation should be carried from inside out. if γ is a Γ(4) . We would see in detail how this works in section 16.16.45) 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. If γ denotes the two point function Γ(2) .2 A brief history of renormalization 689 Figure 16. . mentum variables about zero four-momentum out to order D(γ) which denotes the superﬁcial degree of divergence of γ. we have tγ H = 0. then (D(γ) = 2 and k denotes the generic internal momentum) 1 µ ν ∂ 2 Γ(2) (k. Then the assertion is that the renormalized Feynman graph is convergent. 0) + . we can obtain a renormalized Feynman integrand R(G) for the graph G given by the expression R(G) = (1 − tG ) (−tγ ) I(G). F ∈Φγ ∈F (16. p) = Γ(2) (k. For a convergent graph H.

δφ 2 − (16.46) 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. the generating functional would be stationary leading to R δZ = 0 = N Dφ (F [φ] − J) ei(S[φ]+ Jφ) . Working out in detail. when we include quantum corrections. 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.120)) R Z[J] = eiW [J] = N Dφ ei(S[φ]+ Jφ) .48) We note that since the generating functional does not depend on the ﬁeld variable (we are integrating over all ﬁeld conﬁgurations). 2 2 3! L= (16.49) leads to . We recall that the generating functional for the theory in the presence of an external source is deﬁned as (see (12.3 Schwinger-Dyson equation Let us consider the φ3 theory described by the Lagrangian density m2 2 g 1 ∂µ φ∂ µ φ − φ − φ3 . (16. 16.690 16 Renormalization theory Weinberg’s theorem.49) where the Euler-Lagrange operator F is deﬁned in (16. under an arbitrary ﬁeld redeﬁnition φ → φ + δφ inside the path integral. However. (16.47) This describes the dynamics of the system at the tree level. (16. This theorem is extremely important in the study of renormalization.47).

this can be written explicitly as g 2 δ δJ δ δJ ∂µ ∂ µ + m2 φc + or. F δ − J eiW [J] = 0. φc − i φc − i − J = 0.3 Schwinger-Dyson equation 691 eiW [J] F − i or. or.46). δJ δJ δ − J = 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.80). Σ= ig 2 δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c . Equation (16.51) with respect to φc and set φc = 0. (16. δJ δW δ −i − J = 0.16. If we now take the functional derivative of (16. For the φ3 theory described by (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 . g ig δφc ∂µ ∂ µ + m2 φc + φ2 − i − J = 0. F φc − i δJ (16. c 2 2 δJ ig g ∂µ ∂ µ + m2 φc + φ2 + 2 c 2 δ2 Γ δφ2 c −1 + δΓ = 0.52) can be represented graphically as in Fig. or.51) δφc where Γ[φc ] corresponds to the eﬀective action deﬁned in (13.78).50) Here φc denotes the classical ﬁeld which is deﬁned in (13. (16.

The equations in (16.52) and (16.) Similarly. 16. by taking the second functional derivative of (16.53) are integral equations (they are written in a compact notation here) for the two point and the three point functions. 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 . To see in some detail how the BPHZ method works.4 BPHZ renormalization We have seen in detail how multiplicative renormalization works. 16. = + + Figure 16.53) which can be represented graphically as in Fig. Figure 16. let us consider the φ3 theory in six dimensions described by the Lagrangian density .11: Schwinger-Dyson equation for the three point function. (16.51) with respect to φc and setting φc = 0.10: Schwinger-Dyson equation for the two point function. Such equations (and for the higher point functions) are known as the Schwinger-Dyson equations. (We are using a very compact notation where integrations over intermediate coordinates/momenta are suppressed. orders.692 16 Renormalization theory = .11.

16.56) . in dimensional regularization which we will be using. Furthermore. since we are interested in the ultraviolet behavior of the theory. the 2-point and the 3-point amplitudes are superﬁcially divergent. 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.55) Thus. for this theory. (Note that unlike the φ4 theory. (16. 2 2 3! (16. (16.) Although the zero point amplitude is divergent.4 BPHZ renormalization 693 L= 1 M2 2 g ∂µ φ∂ µ φ − φ − φ3 . 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. here the Lagrangian density is not invariant under φ → −φ.54) This is a simple quantum ﬁeld theory that is quite helpful in understanding various features of renormalization. the counting of the canonical dimension gives [φ] = 2. Furthermore. let us set M = 0 for simplicity. we can remove this divergence by normal ordering the theory. In six dimensions.

only the 2-point and the 3-point amplitudes are divergent which can be made ﬁnite with conventional counter terms.57) so that we can introduce. the one loop self-energy can be evaluated as k p k+p = = ǫ 1 − igµ 2 2 ǫ (16.58) 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 3(n − 2) (6 − n) ǫ = = . (16. this theory should be renormalizable. Therefore. an arbitrary mass scale to make the coupling constant dimensionless. 2 2 2 [φ] = [g] = n − (16. in particular. namely.59) . In this case. how the overlapping divergences are handled. let us calculate some amplitudes beyond the simple one loop. In order to see how renormalization works in some detail and. g → gµ 2 .694 16 Renormalization theory when n → 6. as before. Using dimensional regularization we analytically continue to n dimensions and deﬁne ǫ = 6 − n. the naive dimensional analysis gives (n − 2) . Thus.

4 BPHZ renormalization 695 This form is quite suitable to carry out two loop calculations which we will do shortly. (16. we obtain the one loop self energy of the theory in (16. let us simplify this.16. 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 .60) = 1 6 Using these.76)). However. 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 . 2− 6−ǫ ǫ ǫ n =2− = 2 − 3 + = −1 + .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 . to understand the structure of the self energy at one loop. ǫ Γ −1 + 1 1 ǫ 1 + ln 4π . As we have discussed earlier (see (15. 2 2 2 2 ǫ ≃ 2 2 − + (γ − 1) .

but since the BPHZ method subtracts out divergences in every graph.62) Similarly.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 .64) .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 . 12(4π)3 ǫ =− (16. 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. i xi Ai ) × ( (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. (16.61) This leads to the one loop counter term ig2 p2 2 . (16.

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 .16. (16.66) (16. where we have identiﬁed x1 (1 − x1 )Q2 = x1 (1 − x1 ) p2 + x2 (1 − x2 )p2 1 +2x1 x2 p · p1 . Therefore.65) where we have used the constraint from the delta function in the intermediate steps.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 . by shifting the variable of integration we obtain the value of the integral in (16.

71) 1 1 dx1 0 0 du (1−x1 ) 2 (−x1 (1 − x1 )Q2 ) − ln −γ . ǫ 4πµ2 (16.67) 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 . we can rewrite (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 (16.69) dx1 dx2 dx3 δ(1 − x1 − x2 − x3 ) = dx1 0 0 dx2 .70) Furthermore.72) . we have x3 = 1 − x1 − x2 and x3 = 1 x3 = 0 so that 1 1−x1 ⇒ x1 = x2 = 0. (16.68) as ig3 µ 2 =− (4π)3 ǫ (16. ⇒ x2 = 1 − x1 . 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. ǫ 4πµ2 (16.68) 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. (16.

59). 1 (16.73) 1 2 Equation (16.12 (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. x2 ) ig3 µ 2 2 .62).74) Let us next calculate the self-energy at two loops. (16. the ﬁrst integral in (16.75) separately.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. Using the form of the one loop self-energy in (16.75) has the form . = + −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 . = 2(4π)3 ǫ ǫ (16. Let us look at the two integrals in (16.4 BPHZ renormalization 699 where in the new variables x1 (1 − x1 )Q2 = x1 (1 − x1 )(p + up1 )2 + u(1 − u)p2 .16. Let us ﬁrst look at the graphs with non-overlapping divergence of the form shown in Fig. 16.12: Non-overlapping two loop self-energy diagrams in the φ3 theory.68) leads to the one loop counter term (the factor of arises from doing the integrals over x1 .

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 .

Next.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 . let us look at the second integral in (16.85).16. (The y integration is related to the appropriate beta 2 function. see (16.60) in the intermediate steps. + − 2− 72(4π)6 ǫ ǫ 4πµ2 12 (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 . they will require non-local counter terms to cancel them.64) as well as (15.76) Here we have used (16.) 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.76) and (16.

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 . upon using the form of the one loop vertex function in (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.67) (with x2 expressed in terms of u as in (16.78) I1 + I2 = ig4 p2 2 11 + ﬁnite .77).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. 72(4π)6 ǫ2 ǫ 4πµ2 3 As a result. Let us next look at the overlapping divergent graph in the self energy at two loops which.76) and (16. adding (16.71)). the sum of the two graphs in (16.

81) .16. (2π)n ℓ2 (ℓ + p)2 (Q2 )3− 2 (16. x1 (16. we can write (16.79) 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 . Using (16. (16.64) to combine denominators.73) (with appropriate momenta).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 .79) n −3 n where Q2 is deﬁned in (16.4 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 .

83) (16. Next. y1 u(1 − u) 2 p x1 (16.80) 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. we can do the integration over y3 using the delta function in (16. However. let us evaluate the integral in (16. The one loop pole structure is manifest in the overall multiplicative gamma function. where d is deﬁned in (16.80) in some detail. this is not right and. we can redeﬁne the variable y2 (as we have already done in (16.80).83). so that the (shifted) denominator in (16.704 16 Renormalization theory The divergence of the two loop integral in (16. y1 .81) has the form D2 = ℓ2 + (1 − y1 )(y1 (u − v)2 + v(1 − v)) + = ℓ2 + d(x1 . This is a special feature of overlapping divergent graphs and to appreciate this. Finally. the divergence of the subdiagram (one loop vertex) has been transformed to the Feynman parametric integrals.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.82) Substituting this into (16. in fact.71)) as y2 = (1 − y1 )v. To see the divergence .84) . u. v)p2 .

16.86) This demonstrates how the subdivergence is hidden in the parametric integration.85) the leading term in the parametric integration in (16. The parametric integration in (16. 6 1 Γ(α)Γ(β) = B(α. 6 (16. −γ+ 6 ǫ 6 (16. let us look at only the leading order term in (x1 d)1−ǫ .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 . substituting this result into (16.84) gives the value of the overlapping self-energy diagram to be . Using the standard results 1 0 1 0 dt tα−1 (1 − t)β−1 = 1 du u(1 − u) = . Thus.86) is that the ﬁnite constant 1 inside the parenthesis changes from −γ + 6 to 4. β).84) can be evaluated exactly using Gegenbauer polynomials and the only modiﬁcation from the leading order result in (16. Γ(α + β) 0 1 dudv (u − v)2 = .

87) ig4 p2 2 (−p2 ) 2 + − γ + 3 + ﬁnite .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 (16.74)). 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. with the one loop counter term for the three point function (see (16. On the other hand. (16. − 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.88) .

4 BPHZ renormalization 707 Here we have used (16. Similarly. 16. 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. . 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. (16.90) 8 (−p2 ) +γ− 2 4πµ 3 = ig4 p2 12(4π)6 − 2 2 + + ﬁnite .13 gives + + Figure 16.16.89) p so that the sum of the three graphs in Fig. BPHZ is a renormalization procedure where divergences are subtracted in the graph itself so that it works with any regularization.13: Overlapping two loop self-energy diagrams in the φ3 theory.85) in the intermediate step. = 2 2 ig4 p2 + 6 ǫ2 12(4π) ǫ − 4 2 + 2 ǫ ǫ − ln ln (−p2 ) −γ+3 4πµ2 + ﬁnite (16.

It may contain one particle irreducible subgraphs that are superﬁcially divergent.91) namely. two renormalization parts γ1 and γ2 are called overlapping if they share lines and vertices. namely. . (16. for example. γ γ Figure 16. Fig.708 16 Renormalization theory Consider an arbitrary Feynman graph G. 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.14: Examples of proper renormalization parts in the two loop self-energy graph in the φ3 theory. 16. Two renormalization parts γ1 and γ2 are called disjoint if γ1 ∩ γ2 = 0.15 shows examples of diﬀerent ways of drawing boxes around the renormalization parts.14. namely. if none of the following holds. 16. or γ2 ⊂ γ1 . (16. 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. if γ1 ⊂ γ2 . Thus. (16. for the three loop self energy graph in the φ3 theory.92) while γ1 and γ2 are called nested if one is contained inside the other. A couple of examples of proper subgraphs which correspond to renormalization parts are shown in Fig. if it is superﬁcially divergent.93) Finally.

16. Let D denote the set of laying down of boxes which contain only disjoint renormalization parts.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.4 BPHZ renormalization 709 Figure 16. for the two loop self-energy graph in the φ3 theory. let N denote the set of laying down of boxes (forests) containing only nested renormalization parts. (16. γ2 ⊂ γ1 . 16. let us add to the set D the graph with no boxes.16. A forest is called normal if there is no box around the complete graph G. a forest is called full if there is a box around G. A forest is called empty if there is no box around any subdiagram. On the other hand. Thus. γ1 ⊂ γ2 . Similarly.94) A set of such laying down of boxes around renormalization parts is called a forest. Furthermore. . the complete set of laying down of boxes (forests) containing disjoint renormalization parts is shown in Fig. Fig. 16. γ1 ∩ γ2 = 0.

it Taylor expands .16: The complete set of disjoint forests for the two loop self-energy graph in the φ3 theory. shown in Fig. for example. Figure 16.18: The complete set of overlapping forests for the two loop self-energy graph in the φ3 theory. we deﬁne O to denote the set of laying down of boxes (forests) containing overlapping renormalization parts. 16.18 for the two loop self-energy graph in the φ3 theory. In the conventional BPHZ method. Figure 16. Figure 16. we must subtract out the divergences from each of these sub-diagrams. The reason why we draw boxes around renormalization parts is because these sub-diagrams are superﬁcially divergent and to make the full diagram ﬁnite. 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.17: The complete set of nested forests for the two loop self-energy graph in the φ3 theory.710 16 Renormalization theory Furthermore.

96) This is the conventional tγ operator in the BPHZ prescription and it is clear that using this operation. tγ I(k. 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. if D(γ) < 0. tγ IG = 0. by deﬁnition.97) (16.4 BPHZ renormalization 711 it in the external momenta (external to the graph) up to terms of order D(γ). (16. p) = I(k. (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. 0) + p = dn k I(k. acting on a graph. p). p) 2! ∂pµ ∂pν . p. 0.16. (16. tγ I (k. p1 ) . (2π)n D(γ) = 6 − 6 = 0.95) In other words. 0). p. Note that. p1 ) = I(k. we are simply able to throw . (2π)n 1 µ ν ∂2 p p I(k. Thus. p=0 k γ: p k+p p2 k + p + p1 p1 = dn k I (k.

compatible with any regularization.99) so that the ﬁnite part of the graph can be identiﬁed with ∞ m=0 1−t G FG = am ǫm = F G = ﬁnite. (16.103) .712 16 Renormalization theory away potentially divergent parts inside the integral. Therefore. has the general form ∞ m=−n FG = am ǫm . (16. in the case of dimensional regularization.98) In such a case.101) Given a renormalization part γ in a Feynman diagram. in fact. the tγ operation is deﬁned to be (it separates out the pole terms) −1 m=−n t FG = G am ǫm . am ǫm = ﬁnite parts. (16. ∞ m=0 (16.102) then as we have seen the Taylor operation is deﬁned such that −1 m=−n tγ Iγ = (1 − tγ ) Iγ = am ǫm = divergent parts. this makes it clear that for a superﬁcially convergent graph (with no pole term) tG FG = 0. (16.100) Once again. it never uses any particular regularization and is. In particular. if the integral has the form ∞ m=−n Iγ = am ǫm . (16. a graph (integral) at n-loops.

Thus.106) RG IG = RG IG .104) This. Then. 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. Let us consider an arbitrary Feynman diagram G with proper renormalization parts γ1 . makes all the subdiagrams (proper renormalization parts) in G ﬁnite. the eﬀect of the term (−tγ Iγ ) can be thought of as that of adding counter terms which subtract the divergence. . γs as subdiagrams.16. let us deﬁne s RG IG = i=1 (1 − tγi ) IG . (16. one needs a ﬁnal subtraction to render the diagram ﬁnite. one deﬁnes s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG .105) We note that if G is not superﬁcially divergent. . the subtraction must be carried out . Thus. However. γ2 . (16. by deﬁnition tG RG IG = 0. since in this method we do not explicitly use counter terms. from the deﬁnition of RG IG in (16.4 BPHZ renormalization 713 Thus. then. . graph by graph this procedure can be applied and every graph can be made ﬁnite. In any case. this method is correspondingly more general.105). . However. The only understanding here is that for nested diagrams. the graph would now be ﬁnite. (16. by Weinberg’s theorem. of course. the graph G itself may be superﬁcially divergent in which case. and (16.107) but this is not true otherwise.

109) where Φi is the complete set of proper renormalization parts that are non-overlapping. can lead to non-local counter terms. With this. for example. Its eﬀect is to give IG itself). Instead of the set of disjoint proper renormalization parts (subgraphs) of a Feynman graph G. That this is true will be seen shortly through examples.108) is obtained as follows. let us look at the complete set of subgraphs that include only disjoint and nested proper renormalization parts. Although this method of making a Feynman diagram ﬁnite is absolutely correct. Φi γk ∈Φi (16. Here Rγk is the operator (16. of course. requires that the overlapping divergences be taken care of by lower order counter terms. The BPH recursion relation (16. For disjoint or overlapping divergences. we can now write RG IG = = 1 − tG RG IG 1 − tG (−tγk ) IG . 16. This is seen through Bogoliubov’s R-operator which deﬁnes RG IG = = 1 − tG RG IG 1 − tG Φi γk ∈Φi −tγk Rγk IG .714 16 Renormalization theory inside out.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. Equation (16. this does not address the issue of locality of counter terms since we are subtracting out overlapping divergences which. Fig. 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 . (16.108) as well as . The proof of renormalization by local counter terms.104) associated with the superﬁcially divergent renormalization part γk . the order of the subtraction is irrelevant. Zimmermann’s solution of the BPH recursion relation (16.19 deﬁnes all the disjoint and nested proper renormalization parts of the three loop self-energy diagram. This is the BPH formula. Thus.

Zimmermann’s solution (16.4 BPHZ renormalization 715 γ2 γ1 Φ0 = {0} γ3 Φ1 = {γ1 } γ4 γ1 Φ2 = {γ2 } γ2 Φ3 = {γ3 } γ3 Φ4 = {γ4 } Φ5 = {γ1 .109) seem quite diﬀerent from the original relation (16. γ2 } γ1 γ2 γ4 Φ6 = {γ1 .20. γ Figure 16. However. γ3 } Φ7 = {γ2 . let us next show through examples that they are all. Before we proceed to see this. let us look at some simple examples just to get acquainted with the notions of the BPH procedure. from the BPH formula (16. we have . The set of disjoint proper renormalization parts for the two loop vertex correction is shown in Fig.105). γ4 } Figure 16.16. equivalent. in fact.19: The complete set of disjoint and nested proper renormalization parts for the three loop self-energy graph in the φ3 theory. 16.20: The complete set of disjoint proper renormalization parts of the two loop 3 point function graph in the φ3 theory. Therefore.108).

113) . 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 .21 containing only proper disjoint renormalization parts. we can write (16. there is no subdiagram in this case. (16. RG IG = (16. in fact.105) and.110) also as RG IG = IG + (−tγ IG ) = (1 − tγ ) IG . From the relation IG = IG/γ Iγ . 16.110) where IG/γ is the part of the diagram which does not contain the subdiagram Iγ . (16. by deﬁnition.112) This is exactly the original formula (16. this is how we are supposed to renormalize the theory. Here we have overlapping divergences and following BPH (see (16. in fact.108)) we can write γ1 γ2 Figure 16.111) 1 − tG RG IG = 1 − tG (1 − tγ ) IG . We note that. Let us next look at the two loop self-energy graphs shown in Fig.716 16 Renormalization theory RG IG = IG + IG/γ −tγ Rγ Iγ = IG + IG/γ (−tγ Iγ ) . (16. Rγ Iγ = Iγ because there is no superﬁcially divergent proper subdiagram of γ.21: The complete set of disjoint proper renormalization parts of the two loop self-energy graph in the φ3 theory.

(16.118) 2 i (−p2 ) g4 4(4π)3 ǫ (4π)3 6 ig4 p2 2 .4 BPHZ renormalization 717 so that RG IG = 1 − tG (1 − tγ1 − tγ2 ) IG . Let us recall some of the calculations we had done earlier.74). (16. tγ which is supposed to isolate ǫ pole terms in ǫ. 2(4π)3 ǫ (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. At one loop we had found that (see (16. The diﬀerence between the two is given by 1 − tG tγ1 tγ2 IG .114) This does not exactly coincide with the original formula (16. It is such terms that complicate the proof of renormalizability. sets the factor of µ 2 = 1 in the graph at one loop) = so that ig3 2 = (−tγ2 Iγ2 ) .105) RG IG = 1 − tG (1 − tγ1 ) (1 − tγ2 ) IG .116) (16. 12(4π)6 ǫ2 .115) which we recognize to correspond to the counter term associated with an overlapping divergence if it did not vanish.16.

we can now write s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG .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. If we recall our calculation for the two-loop overlapping divergence (see(16. (16. we note that using this. 2 therefore. it is best to study an example.122) i where Φi ’s deﬁne forests containing only disjoint or nested proper renormalization parts. the ǫ1 term precisely has the same coeﬃcient and it follows.119) Physically.87)). there is a theorem that says that if γ1 and γ2 denote two proper renormalization parts of G that are overlapping. this is the statement that we do not need subtractions associated with overlapping divergences. that 1 − tG tγ1 tγ2 IG = 0. This makes the connection between (16. (16. then. γi ’s are only disjoint or nested proper renormalization parts. Φi γk ∈Φi (16. Let us recall that the three loop diagram has eight . 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. In general.121) involving subtractions of only non-overlapping proper renormalization parts. physically because it says that we do not need to subtract overlapping divergences.121) can also be written as (1 − tγi ) ≡ (−tγk ) .105) and Zimmermann’s solution (16.120) This result is very important. namely. To make contact with the formula of BPH. (16.109). The product of the factors in (16.

123) Let us next recall that since γ1 and γ2 do not contain any proper renormalization parts. + (tγ1 Rγ1 )(tγ2 Rγ2 ) IG = (16.16.123) 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.γ2 } (−tγ1 Iγ1 ) (−tγ2 Iγ2 ) + IG/γ3 −tγ3 Iγ3 /γ1 (−tγ1 Iγ1 ) + IG/γ4 −tγ4 Iγ4 /γ2 (−tγ2 Iγ2 ) .124) Using this.4 BPHZ renormalization 719 forests consisting of disjoint and nested proper renormalization parts which are shown in Fig. 16. we can rewrite (16.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 .125) Φi γk ∈Φi . Thus. 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 .19. (16. = Iγ2 . we can write Rγ1 Iγ1 Rγ2 Iγ2 = Iγ1 .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 .

This is.108). if the index of divergence were negative.127) then. 2 (16. We would have divergence structures with more and more numbers of external lines. (16.720 16 Renormalization theory where Φi is the complete set of proper renormalization parts consisting of only disjoint graphs. In this case. 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. it is equivalent to making any Feynman graph ﬁnite by local subtractions (counter terms). (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.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. therefore.128) not only would we have a ﬁnite number of types of graphs that can be divergent. namely. 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. the BPH formula (16. So far. shows how the BPH formula leads to Zimmermann’s solution and how.20) and (16. δi < 0. Such theories are known as renormalizable theories. we have talked about theories where the index of divergence of interactions in four dimensions (recall that D = 4 − B − 3 i ni δi . see (16. However. On the other hand if δi > 0. of course.21)) 2F + 3 δi = di + bi + fi − 4. Such theories are called nonrenormalizable theories. it is clear that the superﬁcial degree of divergence of a graph would increase with increasing number of interaction vertices. in this case. This analysis. In contrast. Of course. through Weinberg’s theorem.

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.a µ ν 4 2 L=− − .5 Renormalization of gauge theories The analysis in this chapter demonstrates renormalizability of standard ﬁeld theories involving scalar and fermion ﬁelds. we do not need counter terms for the gauge ﬁxing terms).129) + ∂µ ca ∂ µ ca − gf abc (∂ µ ca )Ab cc .a − ∂ ν Aµ.a )+(1+B)gf abc Ab Ac ∂ µ Aν. As a result. Let us discuss this brieﬂy with the example of the pure non-Abelian Yang-Mills theory.q + F a F a + (∂ µ F a )Aµ.p Aν.p ν. 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. some subtleties arise in the case of gauge theories.q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ. 16. Like QED.a ν µ ν 2 g2 ξ (1 + C)f abc f apq Ab Ac Aµ.16. µ Naive power counting shows that the only graphs that are superﬁcially divergent are the two point. Such theories are conventionally known as super-renormalizable theories.12)) ξ 1 a L = − Fµν F µν a + F a F a + (∂µ F a )Aµ. we can add counter terms to write (1 + A) ∂µ Aa (∂ µ Aν.5 Renormalization of gauge theories 721 of divergence) will be divergent.a ν µ ν 2 − g2 abc apq b c µ.a ) + gf abc Ab Ac ∂ µ Aν.a − ∂ ν Aµ. The total Lagrangian density with the gauge ﬁxing and the ghost Lagrangian densities has the form (see (13.a 4 2 (16.a + ∂ µ ca Dµ ca 4 2 1 = ∂µ Aa (∂ µ Aν. 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. However.

˜ Z3 = 1 + D.a − ∂ ν Aµ.130) + Z3 ∂µ ca ∂ µ ca − Z1 gf abc ∂ µ ca Ab cc . of course. as we have seen explicitly in the calculation in QED.131) It is worth emphasizing here again that we have not added any counter term for the gauge ﬁxing terms because. all the divergences can be taken care of. g(u) = Z1 Z3 2 g. ˜2 ca(u) = Z3 ca . −3 1 1 g2 ˜ − ˜ −1 = Z1 Z3 2 Z3 g. g1 (u)2 (u) −2 = Z4 Z3 g2 .a ν µ ν 2 − Z4 g2 abc apq b c µ. . With these counter terms. But the main question is whether renormalization would preserve the gauge invariance – actually the BRST invariance – present in the original theory. µ where we have identiﬁed Z3 = 1 + A. µ 1 ˜2 ca(u) = Z3 ca . a general feature of gauge theories following from the BRST invariance of the theory (see (13.q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ. in fact. if we redeﬁne the ﬁelds and parameters as a(u) 2 Aµ = Z3 Aa . Z4 = 1 + C. (16. −1 1 F a(u) = Z3 2 F a . Namely.a ) + Z1 gf abc Ab Ac ∂ µ Aν. ˜ Z1 = 1 + E. This is.a 4 2 ˜ ˜ (16.722 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. the longitudinal part of the photon self-energy is not renormalized by quantum corrections.91)). Z1 = 1 + B.

136) This is.135) which would lead to g(u) 2 = g1 (u) 2 = g2 (u) 2 . for example. −2 2 −3 = Z4 Z3 g2 = Z1 Z3 g2 .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) and remove all the divergences from the theory. 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. (16. But the real question is how can the renormalization process guarantee this. In this case. (16. in turn require that the corresponding counter terms have to be related.132) ξ (u) = Z3 ξ. the same relation as in (16. 2 −3 ˜ 2 −1 ˜ −2 = Z1 Z3 Z3 g2 = Z1 Z3 g2 . of course.133) This would. therefore. the counterterms will satisfy (16. The Slavnov-Taylor identities following from the BRST invariance (see.133) and with this identiﬁcation we can write (16. g(u) = g1 (u) (u) = g2 .16. we will have 2 g(u) g1 2 −3 = Z1 Z3 g2 . (u) 2 (u) 2 g2 (16. namely. ˜3 Z3 Z1 Z (16. if we choose a regularization scheme which respects the BRST symmetry. (13.134) Therefore.5 Renormalization of gauge theories 723 (16.130) as .

the choice of a regularization which preserves this symmetry is quite crucial. a + iψ k γ µ Dµ ψk . 4 (16. (16. (16. Otherwise. · · · . 2. It is for these reasons that in the case of gauge theories. in studying nonAbelian gauge theories it is natural to discuss renormalizability using dimensional regularization. 2. in addition to having the inﬁnitesimal local gauge invariance δAa = Dµ ǫa (x). n2 − 1 and k = 1.724 16 Renormalization theory ξ (u) a(u) a(u) 1 a(u) F F + (∂µ F a(u) )Aµ. dimensional regularization is one of the regularization schemes (and it is quite simple) which respects gauge invariance and BRST symmetry and. 16. n. µ δψ k = iǫa (x) ψT a δψk = −iǫa (x) (T a ψ)k . k . the regularization scheme may introduce unwanted spurious eﬀects into the theory. consequently.46)) 1 a L = − Fµν F µν.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 (12.138) where a = 1. · · · . As we have seen. We note that this theory. 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).a(u) L = − Fµν F µν a(u) + 4 2 +∂ µ ca(u) Dµ ca(u) . The BRST invariance of this Lagrangian density is now manifest (with unrenormalized ﬁelds and parameters).139) . It is essential in identifying the physical states of the theory. in establishing unitarity as well as in showing the gauge independence of the physical matrix elements.137) where the covariant derivative is deﬁned with g(u) .

e. Dµ J µ. i. we can still deﬁne the axial vector current as in (16. In this theory. δψ k = −iλψ k γ5 .143) where the variation of the Lagrangian density in (16. (16.140) describes the chiral symmetry transformations under which the massless theory is invariant. will not be conserved. Let us note that.141) We note here that the vector current in (16.140) Here λ is the space time independent real inﬁnitesimal parameter of transformation and (16. however. (16.143) is without the parameter of chiral transformation. therefore. µ J5 = ψ k γ5 γ µ ψk . We can also derive identities from this relation that would relate various matrix elements of the . µ ∂µ J5 = 0. is covariantly conserved. (16.16.141) transforms covariantly under a gauge transformation and. 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. Nonetheless. there are two conserved currents.141) which.6 Anomalous Ward identity 725 is also invariant under the global transformation δψk = −iλγ5 ψk .a = 0.a = ψ j γ µ (T a )jk ψk . rather it would satisfy the equation µ ∂µ J5 = −δLm = −2imψ k γ5 ψk .140). if the fermions are massive. J µ. therefore.. if there is a term in the Lagrangian density of the form Lm = −mψ k ψk . (16. the invariance of the theory leads to a conserved current. As is the case with continuous symmetries.142) then this would not be invariant under the chiral symmetry transformations (16.

[e] = 1. On the other hand. when we calculate matrix elements regularized in a gauge invariant way. 2 [Aµ ] = 0. . This is easily seen from the fact that in 1 + 1 dimensions the Lagrangian density has canonical dimension 2 and.726 16 Renormalization theory theory (also known as broken axial vector Ward identities). The Schwinger model is described by the 1 + 1 dimensional Lagrangian density 1 L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ. a massless photon does not have any true dynamical degrees of freedom since there cannot be any transverse polarization. we say that the current is anomalous or that the conservation law has anomalies. Let us discuss this in the simple model of two dimensional massless QED known as the Schwinger model. (16. gauge invariance reduces the superﬁcial degree of divergence and if treated in a gauge invariant manner even this amplitude is ﬁnite. Secondly. we have 1 [ψ] = [ψ] = . the electromagnetic coupling (electric charge) in the Schwinger model carries dimensions unlike the four dimensional QED where the electric charge is dimensionless. therefore. ν = 0. as we have mentioned earlier. Let us note that in two dimensions we can choose the two Dirac matrices as 0 1 1 0 γ 0 = (γ 0 )† = σ1 = . When classical identities involving a current do not hold quantum mechanically. Let us note that in 1 + 1 dimensions.144) This model can be exactly solved and has been studied from various points of view.145) Furthermore. 1. by naive power counting we see that only the photon two point function (among photon amplitudes) is superﬁcially divergent. we ﬁnd that the axial vector Ward identities are violated. (16. However. There are various ways to see and understand this phenomenon in quantum ﬁeld theories. 4 µ.

(16. 2 2 (k + p)2 (2π) k µν I5 (p) = −e2 (16.6 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 = . µ ∂µ J5 = 0. note that in addition to the diagonal metric tensor η µν = (+. (16.150) 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. (16. −).149) As a result of these identities. with ǫ01 = 1.148) There are various simple identities in 1 + 1 dimensions involving the Dirac matrices. (16. (16. but the ones that are directly of interest to us are γ5 γ µ = ǫµν γν .16. 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 .22 and this is given by / / / d2 k Tr γ5 γ µ kγ ν (k + p) . 16.152) . in two dimensions we have a second rank anti-symmetric tensor (Levi-Civita tensor) given by ǫµν = −ǫνµ . γ µ γ ν = η µν + γ5 ǫµν . (16.147) Furthermore.151) Let us now calculate the one loop amplitude involving a photon and an axial vector current as shown in Fig.146) .

152) using dimensional regularization which we know would preserve gauge invariance. Here we have used the propagators for the fermions following from (15. Γ(2) (−x(1 − x)p2 )2− 2 (16.728 16 Renormalization theory γ5 γ µ k γν p k+p Figure 16.153) We can evaluate the integral in (16.73).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. we note that the amplitude in (16.22: The amplitude with an axial vector current and a photon. Furthermore.71) and (15. 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 .36) in the limit m = 0. Furthermore. (16.152) corresponds to the Fourier transform of the matrix element µ 0|T J5 (x)Aν (0) |0 . since . Thus.

We note from (16.158) = − .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).157) The result in (16. On the other hand.156) Since each of the terms is now completely ﬁnite.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 .86). the trace over the Dirac matrices can be carried out in two dimensions using the identities (16. However. π (16. using this relation.6 Anomalous Ward identity 729 γ5 is a two dimensional quantity we have not yet evaluated the trace over the Dirac matrices. (16. in fact.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 .16.149) and the ﬁnal result is obtained to be (the x integration is trivial) µν I5 (p) = − ie2 λρ νµ pλ pρ . (16. the ﬁrst term is divergent and we have to be careful in evaluating the trace. (16. 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 ).155) where we have used the gamma matrix identity in n dimension in (15. It is clear now that the factor (2 − n) = ǫ makes even the ﬁrst term ﬁnite and. η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 (16.

This is very important in building models of particle interactions. one of the two currents (or a linear combination of them) would always become anomalous because of quantum corrections. In fact. (16.23: The triangle diagram in four dimensional QED contributing to the anomaly. the particle multiplets are carefully chosen such that the anomalous contributions coming from various multiplets cancel completely making the gauge current anomaly free.730 16 Renormalization theory Recalling that this amplitude is the Fourier transform of the matrix element in (16.159) π 2π This shows that although the axial vector current is divergence free (conserved) at the tree level.153). Although our discussion here has been within the context of the two dimensional QED (Schwinger model). In some gauge theories. we always choose a regularization scheme which would preserve gauge invariance in the quantum theory so that the chiral current becomes anomalous. µ ∂µ J5 (x) = γλ γ5 γ µ γν Figure 16. 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. As a result.159). the same behavior is ob- . On the other hand. this leads to the operator relation e2 e2 ǫµν ∂ µ Aν = ǫµν F µν . since gauge invariance is so vital to the physical theory. quantum corrections have made the conservation law anomalous and the anomaly in the divergence of the current is given by (16. it is possible to show that there is no regularization which will simultaneously preserve gauge invariance and chiral invariance.

For example. In grand uniﬁed theories where we treat both quarks and leptons as massless. Furthermore.23 leads to an anomaly in the chiral current.6 Anomalous Ward identity 731 tained in any dimension.162) which would satisfy a nonanomalous divergence.160) µ ∂µ J5 = −2imψγ5 ψ + = −2imψγ5 ψ + where the dual ﬁeld strength tensor is deﬁned as 1 ˜ ǫµνλρ F λρ . the correct pion life time can be calculated only if the chiral anomaly is taken into account.160) corresponds to the anomaly in four dimensional QED. 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. However.161) and the second term in (16. As we have discussed in the last chapter. in this case. 16. 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 . if we have a theory where there are both vector and axial vector gauge couplings of the fermions. the model cannot be renormalized unless the anomaly is cancelled. such a current is not gauge invariant (although the associated charge is gauge invariant). in the four dimensional QED a graph of the form shown in Fig.16. In string theories similar anomaly free considerations ﬁx the .159) that a similar statement can be made in the case of the Schwinger model as well. Therefore. 8π 2 (16. anomalies have observable eﬀect. For example. 8π (16. they have precisely this kind of coupling. We note from (16. We note here that. Fµν = 2 (16. as is clear from its structure. it is possible to deﬁne a modiﬁed axial vector current as e2 µ ˜µ J5 = J5 − 2 ǫµνλρ Aν Fλρ .

J. Physical Review 75. 838 (1960). S. Bell and R. C. 47 (1969). Velo and A. . 3. Proceedings of the International School of Mathematical Physics “Ettore Majorana”. Muta. 11. 4. these topics are beyond the scope of these lectures. Physical Review 128. J. Coleman. World Scientiﬁc. W. W. Physical Review 177. S.7 References 1. Proceedings of Royal Society A64. 10.732 16 Renormalization theory gauge group uniquely to be SO(32) or E(8) × E(8). J. Wightman. 8. S. 2425 (1962). Physical Review 118. 14. Singapore (1987). Ward. N. 426 (1951). D. Kennedy. Aspects of symmetry. 16. S. Caswell and A. 392 (1982). Hepp. Salam. N. J. 75. Communications in Mathematical Physics 2. 486 (1949). S. 5. 1736 (1949). Foundations of Quantum Chromodynamics. Jackiw. Physical Review 184. A. 12. 1985). Dyson. However. Schwinger. T. Weinberg. 6. 9. Bogoliubov and O. 301 (1966). Cambridge University Press (Cambridge. 227 (1957). Wightman in “Renormalization Theory”. Physical Review 25. Bardeen. 13. Erice (1976). Adler. 2426 (1969). K. A. Physical Review 82. 2. Parasiuk. ed G. 217 (1951). F. 1848 (1969). Acta Mathematica 97. 7. S. E. 84. Nuovo Cimento 60A. 54 (1951).

then the potential is obtained to have the form V (r) = 2αp αp 1 ln + ··· . We have seen earlier at the end of section 8. 17.10 that the Yukawa potential can be related to the Born amplitude for the scattering of two electrons exchanging a (pseudo) scalar meson. r (17.1.1) 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 .Chapter 17 Renormalization group and equation 17. −→ V (r) = e2 αp . Gell-Mann and Low were interested in studying how the electric force and the potential behave at extremely short distances.1 Gell-Mann-Low equation In the early 1950s. 1+ r 3π rme 733 (17. In the same manner.2) . We can.) It is clear from (17. (Physical electric charge is determined from the scattering of on-shell electrons. 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. calculate quantum corrections to r this potential and if we add the one loop correction to the potential shown in Fig. of course. namely.

4) where we have identiﬁed Λ 2α ln αp = α 1 − 3π me . 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. V (r) → 1 as r → 0 (at exr tremely short distances) as naive scaling would imply. 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 + ··· . it also behaves as V (r) → α r 1− Λ 2α ln 3π me = αp .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. When r is 1 1 large. To understand the origin of this behavior. r (17.3) 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. namely r ≫ me ≫ Λ . (17.1: The one loop correction to the electromagnetic potential. we see from (17. In fact. For example. 1 + r 2 m2 e V (r) = (17. in this limit. Furthermore.734 17 Renormalization group and equation Figure 17. it r depends on the mass of the electron in such a way that we cannot take the limit me → 0. It is clear that the one loop corrected potential has a very diﬀerent character from the lowest order potential.

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. 1+ r 3π rme This expression coincides with (17.17.3) in terms of this (physical) parameter. We also see that the naive scaling of the potential breaks down and mass dependence comes in due to renormalization eﬀects.1 Gell-Mann-Low equation 735 α = αp 1 + Λ 2αp ln 3π me + ··· .9) then. on dimensional grounds we see that the potential can be expressed as . (17.6) Expressing the potential (17. For example.8) Furthermore. Let us next investigate what would happen if we deﬁne the electric charge at some arbitrary (intermediate) distance R and not as in (17. in the limit Λ → ∞. (17.8).2) and also shows that the physical ﬁne structure constant can be identiﬁed with αp = rV (r) . 1 r= m e (17. this derivation clariﬁes that the singularity in the potential at me → 0 arises purely because of the renormalization of charge at large distances. if we deﬁne αR = RV (R).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 + ··· .

3) leads. which can be seen from (17. Rme ). R ≪ me .11) Inverting this relation we obtain α = αR 1 + 2αR ln 3π 1 + R2 Λ2 + ··· . R . in the limit Λ → ∞ and rme . (We note that rme = Rme × R so that it is not an independent quantity. We note that αR = RV (R) = α 1 − 2α ln 3π 1 + R2 Λ2 + ··· . 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 . Let us also note that if we deﬁne the electric charge (ﬁne structure constant) as .12) Putting this back into the potential (17.736 17 Renormalization group and equation r 1 V (r) = F αR . Rme . r R (17. 1 + R 2 m2 e (17.3) as follows. the dependence of the potential on me can be neglected.10) where F is a dimensionless function (function of the independent r r dimensionless variables αR .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. .) This function should 1 not be singular as me → 0 which implies that for r. 1 + R 2 m2 e (17. 1 + r 2 m2 e + ··· 1 + R 2 m2 e (17. Rme ≪ 1 (the second of these limits corresponds to me → 0).

R ≪ 2αR ln 3π leads to the relation αr = αR 1 + = αR 1 + 1 + R2 Λ2 − ln 1 + R 2 m2 e . write (see (17.16)) we see that the equation governing this change is given by .18) This shows that the charge (ﬁne structure constant) changes (runs) with distance and from the second of the equations in (17. in general.13) (see also the deﬁnition (17. .14)) αr .1 Gell-Mann-Low equation 737 αr = rV (r). in the limit Λ → ∞ and r. = F αR .12) that we can identify (17.10)) αr r 1 = F αR . = αr 1 + (17.17) From this analysis we see that in the limit of me → 0 we can.17.14) α = αR 1 + 2αR ln 3π 2αr ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· . R V (r) = αr (17.16) 2αR R ln + · · · 3π r and this is compatible with (17. 1 + r 2 m2 e 1 me .15) which. 1 + r 2 Λ2 1 + r 2 m2 e + ··· (17. then we see from (17. r R r r .18) (see also (17. r V (r) = (17.

αR does not exist and consequently. the theory develops ghosts and other problems.16) that.18) leads to (α0 = αr=0 ) α0 = F (α0 .19) β(αr ) = − . R → 0 with R = x ﬁxed. ∂F (αr . 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.18) αr = F αR . x) ∂x (17. In this case. αr= = αp . αR does have a limit which is nonzero and is inde1 pendent of the value 137 which arises only as an initial condir tion. R (17. namely. in the present case. As R → 0. for example. As R → 0.24) . (17. r . dαr 2α2 2α2 = − R = − r. (17. dr 3π 3π r (17. In this case. then (17.738 17 Renormalization group and equation r where dαr dr = −β(αr ).22) There are two possibilities for the solution of the equation (17.21) 1 me We see explicitly from (17. we note that αp and me arise only as initial conditions in this evolution. x) . Gell-Mann-Low equation basically relates the electric charge at one distance (scale) to another (namely.20) and this is known as the Gell-Mann-Low equation. 2. the bare charge becomes inﬁnity. We see from this discussion that through proper renormalization the dependence of the potential on the electron mass me disappears. if we let r. x=1 (17.23) 1.

19) (also where the beta function vanishes). 17. such as the Euler’s constant etc. the eﬀects of masses are negligible at high energies if we renormalize the theory in an appropriate way. as r. in the MS scheme we not only subtract out the pole parts (in ǫ) of an amplitude. we deﬁned the charge at some length . in dimensional regularization. However. (In the example of QED that we discussed in the last section. 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. a particular renormalization scheme may be preferrable if it makes the higher order contributions in perturbation theory smaller. The only remaining breaking of scale invariance is due to renormalization itself. In QED.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. For example. but some constants as well. the MS scheme (minimal subtraction) corresponds to subtracting out only the pole parts (in ǫ) of an amplitude. therefore. It is important to note that in a quantum ﬁeld theory scale invariance is broken by the masses of particles. 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). First. α0 → 0 which is obvious from (17. The renormalized parameters. This introduces an additional arbitrariness into the renormalized parameters depending on the arbitrariness in the mass scale used. we saw that the coupling constants are deﬁned with an arbitrary mass scale µ.16) (remember / r that R is ﬁxed as the limit is taken) and. it corresponds to the second possibility. R → 0.5. Even in dimensional regularization. are diﬀerent in the two renormalization schemes even though the regularization is the same (dimensional regularization). for example.17. therefore. There are two possible sources for this arbitrariness. We will discuss these ideas in more detail in section 17. On the other hand. We can keep track of this by using a running coupling constant.2 Renormalization group 739 This is known as a ﬁxed point of the evolution equation (17. Even within a given regularization.

λ. µ). u −2 λu = Z1 Z3 λ. 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. µ′ = S(p. see the discussion after (16. Namely. Namely. u ′ ′ λu = Z1 Z3−2 λ′ . λ′ . On the other hand. Calculations with either of these sets of parameters will lead to ﬁnite quantities.25) deﬁne a set of renormalized parameters. (17.32)) 1 2 φu = Z3 φ. m′ .27) so that if we expand the S matrix elements in a perturbation series . a different renormalization scale prescription would lead to φu = Z3 2 φ′ . (17. If we are calculating physical quantities such as the scattering amplitudes.740 17 Renormalization group and equation scale which illustrates the source of arbitrariness of the second kind. m. ′ ′ m2 = Zm Z3−1 m′ 2 . they should not only be ﬁnite. −1 m2 = Zm Z3 m2 .) Let us consider a renormalizable theory. we must have (S and S ′ denote representative S-matrix elements calculated in the two prescriptions) S ′ p. ′ 1 (17. they must also have the same value independent of which set of parameters are used to calculate them.26) ′ where we are denoting Z3 = Z3 (µ′ ) and so on. 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.

m′ . If we can calculate the exact series.µ ′ −1 2 −1 φ′ = Z3 1 −2 λ′ . in perturbation theory.32) Similarly. We can invert the relations in (17. the residual terms can be non-negligible. µ′ = (17.30) which leads to a relation between the two (17. On the other hand. m. However. µ′ φu . all we can say and check is that S ′ p.17. µ) = O λn+1 . say the nth order. φ′ = z µ′ . we can check this.25) and (17.26) we can also write −1 m2 = Zm (µ)Z3 (µ)m2 . u −1 m′ 2 = Zm (µ′ )Z3 (µ′ )m2 . u (17. µ φ. the higher order corrections become negligible very quickly. In such a case. λ. m′ . λ′ . such as QED. we can only calculate up to a given order.29) 1 In a theory. µ′ − S(p. n n S ′ p.25) and (17. in a theory like QCD (where the value of the coupling constant is not small). (17. µ) = n an λn . where α ≃ 137 .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. (17. from (17. λ. Let us next see that this arbitrariness in the renormalized parameters is due to a ﬁnite renormalization.33) . a′ λ′ n . m. λ′ .2 Renormalization group 741 S(p.26) so that we can write the renormalized ﬁelds in the two prescriptions as φ = Z3 2 (λ. µ)φu . (17.31) = Z3 (µ′ ) Z3 (µ) . Here we have identiﬁed z µ .

Zm .26) imply −1 2 λ = Z1 Z3 λu .34) zm (µ .32). ′ ′ λ′ = Z1−1 Z3 2 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). µ) = ′ Zm (µ′ ) Zm (µ) −1 Z3 (µ′ ) Z3 (µ) . Zm . with (17.35) Finally. (17. (17. zλ ) are completely ﬁnite. (17.38) we note that . (17.35) and (17.742 which implies that 17 Renormalization group and equation m′ 2 = zm (µ′ .38) ′ ′ ′ Since (Z3 . Let us next show that these ﬁnite renormalizations (ﬁnite transformations) deﬁne a group. µ)m2 . it follows that (z. zm .25) and (17. with (17.37) zλ (µ′ . Thus. (17. In fact.36) which leads to the relation λ′ = zλ λ. from the deﬁnitions in (17. Z1 ) diﬀer only in ﬁnite parts (the divergent parts are the same since the “µ” dependence comes only in the ﬁnite parts). µ) = Z1 (µ′ ) Z1 (µ) −1 Z3 (µ′ ) Z3 (µ) 2 . Z1 ) and (Z3 .

−1 zm (µ.40) The set of ﬁnite renormalizations.µ = = Z3 (µ′′ ) Z3 (µ) −1 2 ′′ ′ ′ Z3 (µ′′ ) Z3 (µ′ ) −1 2 Z3 (µ′ ) Z3 (µ) −1 2 = z µ′′ . µ . Zm (µ′′ ) Zm (µ′ ) −1 −1 zm µ′′ . µ′ zλ µ′ . µ). µ).39) Namely. µ) = 1 = zm (µ. µ′ zm µ′ . (17. deﬁne an Abelian group known as the renormalization group and all physical quantities must be invariant under the renormalization group transformations. µ′ ) = zm (µ′ .µ z µ . µ) = zλ (µ. µ′ ) = zλ (µ′ . µ). µ . µ = × = Zm (µ′ ) Zm (µ) −1 Z3 (µ′′ ) Z3 (µ′ ) Z3 (µ′ ) Z3 (µ) = zm µ′′ . physical quantities cannot depend on the arbitrary scale introduced to carry out perturbation calculations in a theory. two successive ﬁnite renormalizations lead to an equivalent single ﬁnite renormalization. −1 zλ (µ. z(µ.2 Renormalization group 743 z µ . µ . . Namely. These transformations further satisfy z(µ.17. µ′ ) = z −1 (µ′ . therefore. −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. µ).

. u −n 2 (17. λ(µ). . xn ) . . n n (17. . pn ) = Z32 Γ(n) (p1 . u n n (17. µ) . . λ(µ). xn ) = T (φ (x1 ) . The connected renormalized Green’s functions of the theory are deﬁned as G(n) (x1 .45) . µ)Γ(n) (p. m(µ′ ). λ(µ′ ). . . . u Written out explicitly. . λu ) u = z 2 (µ′ . µ) = Z32 (µ)Γ(n) (p. m(µ). . . λu ) . pn ) . pn ) = Z3 2 G(n) (p1 . φ (xn )) −n 2 c c = Z3 T (φu (x1 ) .41) Such a relation between the renormalized and the unrenormalized Green’s functions of the theory holds true in momentum space as well. 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 . µ′ = Z32 (µ′ ) Γ(n) (p. namely. G(n) (p1 . . . mu . . u −n (17. p2 . . .43) (17.3 Renormalization group equation To understand the consequences of the renormalization group of transformations described in the last section.44) which implies that the renormalized 1PI functions in two renormalization prescriptions are related as Γ(n) p. . . . pn ) . mu . φu (xn )) = Z3 G(n) (x1 .42) where only (n−1) of the external momenta are independent (because of overall energy-momentum conservation). . this has the form Γ(n) (p. let us next look at the Green’s functions of the scalar ﬁeld theory.744 17 Renormalization group and equation 17. . . . . m(µ). .

In dimensional regularization. Let us deﬁne the renormalized quartic coupling in the scalar ﬁeld theory as µǫ λ.48) that β(λ) = − ǫ + µ dZλ Zλ dµ λ. 0 (17. Thus. As we have noted this deﬁnes the Abelian group of transformations known as the renormalization group. 0 0 or.32).47) Both λu and λ are dimensionless and are related as (compare with (16.19)) µ dλ dµ = β(λ).39)) −1 −1 2 µǫ λ = Z1 (µ)Z3 (µ)µǫ λu = µǫ Zλ (µ)λu . it follows that (see also (17.49) where we see from (17.46) while the unrenormalized coupling can be deﬁned as (in discussions in earlier chapters we had used µ0 = 1) µǫ λu .3 Renormalization group equation 745 which follows from the deﬁnition of z in (17. Here µ0 is assumed to be ﬁxed and µ is the variable renormalization scale.50) . Since λu is µ-independent (independent of the renormalization prescription). (17. the couplings are deﬁned with an arbitrary mass scale. 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.17. (17. (17. Here we have used p to denote collectively all the independent external momenta of the 1PI vertex function. λu = µ µ0 ǫ λZλ (µ). (17.48) −2 where we have identiﬁed Zλ = Z1 Z3 .

By dimensional reasoning then they cannot depend on m either. the equation for the coupling constant (17.55) where the constants a. q. s. 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 + · · · . . in fact. Similarly. u u −1 (17. We note that this is a consequence of the fact that dimensional regularization is a mass independent regularization. µ dZ m . c. 2Z m dµ (17. t. from −1 m2 = Zm (µ)Z3 (µ)m2 = Z m (µ)m2 .53) Both β and γm are ﬁnite functions of µ as ǫ → 0 since the divergences in the numerator and denominator cancel out. As a result.52) γm = (17.49) decouples from the mass equation (17. (17. mu held ﬁxed. their dependences are of the form (the µ dependence comes in only through the coupling) β = β(λ). in general.746 17 Renormalization group and equation ∂Z ∂µ and the derivative on the right-hand side should be thought of as with λu .52) and can be solved independently which we will analyze in the next section. .54) in the MS (MS) scheme. functions of λ. b. . They are. m. γm = γm (λ). are independent of µ because the pole terms (in ǫ) are independent of µ as there are no overlapping divergences in a renormalizable theory.51) we obtain dm dµ µ where = −mγm . . λ4 + · · · . µ and. (17.

∂µ ∂µ ∂λ(µ) ∂µ ∂m ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) = 0. 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). λ(µ). γ = γ(λ). u −n (17. mu ﬁxed. λu ) = Z3 2 (µ) Γ(n) (p. dµ µ µ ∂ ∂λ(µ) ∂ ∂m(µ) ∂ +µ +µ − nγ Γ(n) = 0. namely.52)) ∂λ . m ∂µ (17. µ or.57) (n) µ or.58) β(λ) = µ γm = − γ = µ ∂Z3 . or.49) and (17. µ) = 0. ∂µ ∂λ ∂m (17.17. . m(µ). λu ) = 0. m(µ). (17.44) Γ(n) (p.59) Equation (17. see also (17.57) is known as the renormalization group equation for the 1PI vertex functions and expresses how amplitudes change when the renormalization scale is changed. dµ d −n Z3 2 (µ) Γ(n) (p. mu . mu . λ(µ). µ) .56) which leads to the relation (since the unrenormalized quantities are independent of the renormalization prescription) dΓu (p. ∂µ µ ∂m . where we have deﬁned (the derivatives are taken with λu .3 Renormalization group equation 747 Let us next look at the 1PI amplitudes (17.

∂µ ∂λ ∂m (17. µ) = 0. the renormalization group equation (17.63) Consequently.57) describes how renormalized amplitudes (1PI vertex functions) change as the renormalization scale is changed ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) (pi . m.58) (the derivatives are taken with λu . 1 2 n (17.4 Solving the renormalization group equation As we have seen. mu ﬁxed) ∂λ . ∂µ µ ∂m . λ.64) . ∂µ ∂λ µ (17. (17. if we deﬁne the Mandelstam variable associated with the n-point function as s = p2 + p2 + · · · + p2 .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. µ) = 0.62) Let us note from our earlier discussions (see (16. m ∂µ (17.60) µ where pi denotes the independent external momenta and as we have seen in (17.748 17 Renormalization group and equation 17. λ. the equation takes the form ∂ ∂ + β(λ) − nγ Γ(n) (pi . let us set m = 0 in which case.61) β(λ) = µ γm = − γ = µ ∂Z3 . 2Z3 ∂µ For simplicity.

s (17. a change in t can be thought of as changing µ or for a ﬁxed µ. 2 µ (17. consequently.62) can be written as ∂ ∂ + β(λ) − nγ Γ(n) = 0. f (n) is a dimensionless function of independent dimensionless variables which are Lorentz invariant. ∂t ∂λ and Γ(n) = s 4−n 2 (17.70) . ∂µ ∂µ ∂t ∂t µ (17.65) where we have used the fact that the n-point vertex function of a scalar ﬁeld theory is Lorentz invariant and. µ2 s . ∂t ∂x (17.67) ∂t ∂ s 1 (where from (17. λ. or. pi · pj . It follows now that ∂t ∂ ∂ ∂ =µ = . λ. Let us relate the renormalization scale µ to a new dimensionless variable t as (this has the advantage that for ﬁxed s.66) so that for a ﬁxed s. t) + v(x) = L(x)ρ(x. λ. t) ∂ρ(x. changing the renormalization scale merely corresponds to changing t. t = − 1 s ln 2 .68) f (n) t. we can write f (n) pi · pj s . (17.4 Solving the renormalization group equation 749 then.66) we have used ∂µ = − 1 ∂µ ln µ2 = µ ) so that the 2 renormalization group equation (17.69) Solving the renormalization group equation (17. t). scaling all momenta by a factor of Γ(n) (pi .68) now becomes straightforward from the observation that a linear ﬁrst order diﬀerential equation of the form ∂ρ(x. it can be thought of as scaling the momentum) µ= √ set . µ) = s 4−n 2 √ s.17.

t′ )) . t). (17. We can. (17. Γ(n) ↔ ρ. we note that the diﬀerential equation (17.70) can be written as dρ(x) = L(x)ρ(x). −t)) e R0 −t R0 −t dt′ L(x(x.71).74) Translating t by −t. are deﬁned by the equation dx(x.75) dt′ L(x(x. of course. β(λ) ↔ v(x). 0) = x.71) Using (17. t) = v(x). t)) = ρ (x(x. dt which can be integrated to give ρ (x(x.68) with the hydrodynamic problem (17. Therefore. dt with the boundary condition x(x. ρ(x) = ρ (x(x.73) dt′ L(x(x. nγ ↔ L. (17.t′ )) . (17.76) .70) clearly depends on determining its characteristics x(x. 0)) = ρ (x(x.750 17 Renormalization group and equation is of hydrodynamic type and is best solved by the method of characteristics. directly identify the quantities in the renormalization group equation (17. −t)) e or.70) as t ↔ t.72) (17. where we have used (17.t′ )) .72). t = 0)) e Rt 0 (17. we see that solving the equation (17. in this case. λ ↔ x. we then obtain ρ (x(x. The characteristics.

for large α (large momentum). )e s (17. This is known as Weinberg’s theorem (the same theorem which also made its appearance in renormalization theory). namely. dt (17.t′ )) . −t + ln α). (17.79) so that (remember that −t = t → t − ln α.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.17.78) Note that if we scale all momenta (keeping µ ﬁxed) as s → α2 s.68) requires solving the characteristic equation dλ = β(λ). 1 2 s ln µ2 ) pi → αpi . (17. the leading behavior of the amplitude is given by 4−n 2 Γ(n) → α2 s f (n) (λ(λ.80) then. (17. if . × en R ln α −t+ln α pi · pj ) s dt′ γ(λ(λ. −t).77).t′ −ln α)) → α4−n (ln α)a · · · = (n) (α)D(Γ ) (ln α)a · · · . Let us next note that if λ∗ is a ﬁxed point of the coupling constant evolution equation (17.4 Solving the renormalization group equation 751 so that the solution of the renormalization group equation (17. 0 pi · pj n R−t dt′ γ(λ(λ. we can obtain Γ(n) = s 4−n 2 f (n) (λ(λ.77) and once we know the solution to this equation.

the n-point amplitude will scale. the n-point amplitude (17. .86) which follows from the fact that in perturbation theory γ(λ) is a power series in λ.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 (16.87) (17. (17. If the ﬁxed point of the beta function happens to be at the origin. λ = λ∗ = constant. 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). pi · pj . s (17. then. at the ﬁxed point.79). under a scaling pi → αpi (17.82) (17.83) Near the ﬁxed point.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 .752 17 Renormalization group and equation β (λ∗ ) = 0. (17.84) As a result. then γ(λ∗ ) = 0. namely. (17. = s 4−n 2 s µ2 − n γ(λ∗ ) 2 f (n) λ∗ . near the ﬁxed point of the renormalization group equation. if λ∗ = 0.24)). as Γ(n) ≃ α4−n(1+γ(λ∗ )) Γ(n) . = s 2 e 2 ∗ s 4−n 2 f (n) λ∗ .

In this case. 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). (17.4 Solving the renormalization group equation 753 It is clear.92) On the other hand. 17. let us analyze the qualitative behavior of the solution.2.88) implies that (recall that λ(λ.91) is a ﬁxed point of the beta function.90) λ = 0. 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.17. t → −∞. if β(λ) is negative (β < 0. 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. 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. the behavior of the coupling constant is given by ∞. this leads to the behavior of the β function as shown in Fig. t = 0) = λ) λ λ dλ′ β (λ′ ) t = 0 dt′ = t. the behavior of β(λ) is shown in Fig. the ﬁrst thing that we notice is that (17. λ(t) → (17. Qualitatively. (17.3 and in this case. 17. 0. t → ∞.89) Before solving this equation explicitly. Therefore. We note that dλ dt = β λ . qualitatively we have . and λ > 0). (17.

2: The graph showing that origin in the coupling constant space is an infrared ﬁxed point of the β function. β(λ) λ Figure 17. we have two distinct cases to consider (when λ∗ = 0. then expanding around this ﬁxed point. .754 17 Renormalization group and equation 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. t → −∞. namely. Let us assume that the beta function has a nontrivial ﬁxed point at λ∗ on the positive axis. λ(t) → 0. we recover the two cases we have discussed earlier). ∞. (17.93) and we say that the origin in the coupling constant space is an ultraviolet ﬁxed point. t → ∞. Let us next analyze the behavior of the solutions in some detail.

17.97) 2. (17. λ → λ∗ .95) In the ﬁrst case (17. if β0 > 0. λ − λ∗ = (λ − λ∗ ) eβ0 t . (17.4 (which shows the behavior of a general β function). λ − λ∗ 1 dλ′ = ln . β(λ) = β0 (λ − λ∗ ) + O (λ − λ∗ )2 .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. µ → ∞ and when t → −∞. λ → λ∗ . then as t → −∞.89) leads to λ t= λ dλ′ = β(λ′ ) λ λ or. ′−λ ) β0 (λ β0 λ − λ∗ ∗ (17. In the diagram Fig. then as t → ∞. (17.94) r > 1. 2.17.66) that when t → ∞.89) leads to . if β0 < 0. β(λ) = β0 (λ − λ∗ )r + O (λ − λ∗ )r+1 . (17. µ → 0 for ﬁxed s).4 Solving the renormalization group equation 755 1.96) which implies that 1. (17. the arrows indicate the direction in which λ moves as t → ∞ (if we start near the ﬁxed point).95)). In the second case (see (17.

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.4: Ultraviolet and Infrared ﬁxed points of the β function. β0 < 0. λ t= λ dλ′ = β(λ′ ) λ λ dλ′ β0 (λ′ − λ∗ )r 1 λ − λ∗ r−1 =− or. (λ − λ∗ )r−1 r−1 λ − λ∗ = (λ − λ∗ )r−1 .100) where β0 is a constant calculable in perturbation theory. the series starts at the cubic order in the coupling) β(g) = β0 g3 + O g5 . the beta function has the form (namely. 1 (r − 1)β0 r−1 − 1 (λ − λ∗ )r−1 . (17. = 1 − (r − 1)β0 t (λ − λ∗ )r−1 1 − (r − 1)β0 (λ − λ∗ )r−1 t . 1 − (r − 1)β0 (λ − λ∗ )r−1 t (17.22)). Thus. for example. in both cases.99) Both in QED and QCD. g = 0 is . (17. while for QCD. 1 λ − λ∗ = or.756 17 Renormalization group and equation β(λ) λ∗ λ β(λ) λ∗ λ Figure 17. The diﬀerence between the two theories is that for QED β0 > 0 (see.

+ 2|β0 |t g 2 (t) = 1 g2 (17. β0 < 0) we have from (17. For QCD. 2β0 e2 t→ (17.103) As t → ∞. Such a pole in the evolution of the coupling constant at a particular momentum scale is called the Landau pole.99) g2 = 1 + 2|β0 |g2 t 1 . at large momenta almost naive scaling sets in (see discussion after (17.99) e2 = 1 − 2β0 e2 t 1 . − 2β0 t e2 (t) = 1 e2 (17.104) Since the unrenormalized (bare) coupling does not depend on the renormalization scale. in particular. So (setting g∗ = 0. diverges when 1 . At inﬁnite energy.101) As t increases. we see that the coupling grows and.46)-(17. Let us recall that with the MS (minimal subtraction) scheme at one loop we have (see (17. 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. In such theories. the coupling becomes weaker and weaker and ultimately vanishes.87)). the theory behaves like a free theory and such theories are called asymptotically free theories. for QED we have from (17.102) and perturbation theory breaks down.38)) µ µ0 ǫ λu = λ 1+ 3λ 16π 2 ǫ .48) as well as (16. on the other hand. (17. therefore. it follows that .17.4 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. (g∗ = 0. identifying the coupling with electric charge g = e and recalling β0 > 0). as we have seen.

(17. 16π 2 which is ﬁnite as discussed earlier. ∂µ 16π 2 β(λ) = µ (17. 2 λ(µ) λ(µ) 16π µ 1 1 3 µ = − ln . 2 µ 16π µ or.758 17 Renormalization group and equation µ ∂λu = 0. − 1 1 3 µ + = ln . ǫλ 1