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

1 Introduction As we know.) This formalism is clearly non-relativistic (non-covariant) which can be easily seen by noting that. has the generic form p2 + V (x).2) with p denoting the momentum of the particle and V (x) representing the potential through which the particle moves. ∂t i (1. the dynamical equation (1.1) Here ψ(x.1) takes the form 1 . is given by the time dependent Schr¨dinger equation which has the form o ∂ψ = Hψ. non-relativistic physical system and promote each of the observables to a Hermitian operator. non-relativistic quantum mechanics. even for a free particle. 2m H= (1. in this case. (Throughout the book we will use a bold symbol to represent a three dimensional quantity. The time evolution of the quantum mechanical system. H. in single particle.Chapter 1 Relativistic equations 1. t) represents the wave function of the system which corresponds to the probability amplitude for ﬁnding the particle at the coordinate x at a given time t and the Hamiltonian. we start with the Hamiltonian description of the corresponding classical.

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

Consequently. represents the metric of the Euclidean space and is trivial (in the sense that all the nonzero components are simply unity). these would now consist of four components. any vector would also have four components. if and only if A = 0. as we know. there are now two distinct four vectors that we can deﬁne on this manifold. it is necessarily positive deﬁnite. Namely. (1. δij .2 Notations 3 where x and J represent respectively the position and the angular momentum vectors while A stands for any arbitrary vector. Let us note from the deﬁnition of the length of a vector in Euclidean space that.1. However.7) where repeated indices are assumed to be summed. namely. 3 and we are being a little sloppy in representing the four vector by what may seem like its component) . namely. in this case. (1. (1.9) When we treat space and time on an equal footing and enlarge our three dimensional Euclidean manifold to the four dimensional spacetime manifold. A2 ≥ 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. 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 . 2. consequently. the scalar product of any two arbitrary vectors is deﬁned to be A · B = Ai Bi = δij Ai Bj = δij Ai Bj .8) The Kronecker delta. (µ = 0. we can again deﬁne vectors in this manifold. In such a space. 1. However. any point in this manifold will be speciﬁed uniquely by four coordinates and. for any vector. it does not matter whether we write the indices “up” or “down”. unlike the case of the Euclidean space.

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

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

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

(2.23) This determines the two positive energy solutions of the Dirac equation (remember that the energy eigenvalues are doubly degenerate).25) (2.22) v (p) = − ˜ σ3 p 3 E+ + m 0 1 = 0 p3 E+ + m . ˜ v We can solve these as u(p) = − ˜ σ3 p 3 v (p). v (p) = ˜ .32 2 Solutions of the Dirac equation σ3 p 3 v (p) = − ˜ E+ + m and 1 0 = p3 −E + m + 0 .18).27) . from the second of the two two-component Dirac equations in (2.24) and the set of equations (2. we note that v must vanish in the rest frame while u remains ˜ ˜ arbitrary. Thus. u can be thought of as the independent solution.) ˜ Similarly. (2. for the negative energy solutions we write p0 = E− = −E = − p 2 + m2 .9) becomes (E− − m)˜(p) + σ3 p3 v (p) = 0. ˜ E− − m (2. Thus. for example. (The question of which components can be chosen independently follows from an examination of the dynamical equations.26) Choosing the independent solutions as 1 0 0 1 v (p) = ˜ . u ˜ σ3 p3 u(p) + (E− + m)˜(p) = 0. (2. 3 (2.

(2.27) are reminiscent of the spin up and spin down states of a two component spinor. 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 .28) u(p) = − ˜ σ3 p 3 E− − m 0 1 = 0 p3 E− − m .30) u+ (p) = u(p) ˜ . Thus.32) . The independent two component wave functions in (2.21) and (2. from the fact that we can write σ p ˜ − E 3−3m v (p) − . 0 1 (2. (2. − E+ + m 0 u↓ + = . v (p) ˜ (2.29) and these determine the two negative energy solutions of the Dirac equation.2.1 Plane wave solutions 33 we obtain respectively p3 −E − m − 0 σ3 p 3 u(p) = − ˜ E− − m and 1 0 = .31) p3 −E − m − 0 ↑ u− (p) = . (2. 1 0 0 p3 ↓ u− = E− − m .

unlike the nonrelativistic counting (2s + 1). (2.33) which can be explicitly veriﬁed.27) respectively are ˜ ˜ normalized as (for the same spin components) .36) Here α and β are normalization constants to be determined. for the case of general motion. we can write the solutions for motion along a general direction as u(p) ˜ . pi = −pi .34) v (p) ˜ (2. u− (p) = σ·p u(p) ˜ E+ + m σ·p ˜ E− − m v (p) .21) and (2. (We will determine the spin of 2 the Dirac particle shortly.35) then. 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.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.) 2. namely.) From the structure of the wave function. relativistic particle of spin s in the presence of both positive and negative energies follows to be 2(2s + 1).) 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. (The correct counting for the number of components of the wave function for a massive. (2. u− (p) = β σ·p ˜ E + m u(p) σ· − E + p v (p) m ˜ v (p) ˜ u+ (p) = α . it is also clear that. The two component spinors u(p) and v (p) in (2.2 Normalization of the wave function Let us note that if we deﬁne E = E+ = p2 + m2 = −E− . where p1 = p2 = 0.

˜ v E +m . for the negative energy solutions we have σ·p v = β β −˜† (p) E + m ∗ (2.2 Normalization of the wave function 35 u† (p)˜(p) = 1 = v † (p)˜(p).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).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. ˜ u E+m (2.38) where we have used the familiar identity satisﬁed by the Pauli matrices (σ · A)(σ · B) = A · B + iσ · (A × B). ˜ u ˜ v For diﬀerent spin components. we can now calculate σ·p ˜ ˜ u† (p)u+ (p) = α∗ α u† (p) u† (p) E + m + = |α|2 u† (p)˜(p) + u† (p) ˜ u ˜ = |α|2 1 + = |α|2 = |α|2 = p2 (E + m)2 (2. Given this.2. Similarly.40) p2 +1 (E + m)2 2E |β|2 v † (p)˜(p). this product vanishes.

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 . 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.36 2 Solutions of the Dirac equation It is worth remarking here that although we have seen in (2. v Thus. another wave function (known as the adjoint spinor) that plays an important role is deﬁned to be u(p) = u† (p)γ 0 . Thus. we have carried along ˜ ˜ ˜ ˜ these factors in (2.42) −˜† (p) . In dealing with the Dirac equation. for example.40) simply because we have not speciﬁed their spin components. (2. u† u = 1 = v † v . σ·p v u− (p) = β ∗ −˜† (p) E + m σ·p v = β ∗ −˜† (p) E + m v † (p) ˜ ½ 0 0 −½ (2. for the same spin components.38) and (2.41) σ·p ˜ ˜ u+ (p) = α∗ u† (p) u† (p) E + m ½ 0 0 −½ σ·p ˜ u = α∗ u† (p) −˜† (p) E + m .37) that.

with this normalization condition.2 Normalization of the wave function 37 = |α|2 = 2m(E + m) † u (p)˜(p) ˜ u (E + m)2 (2. in the rest frame of the particle. we determine from (2. is related to the probability density which transforms like the time component of a four vector. this is not a relativistic normalization. the motivation for such a normalization condition comes from the fact that.43) 2m |α|2 u† (p)˜(p). In fact.37) holds) . Thus. we can show that 2m |β|2 v † (p)˜(p).47) With the requirement (2.2. as we will see. By the way. by requiring (for the same spin components) u† (p)u+ (p) = 1 = u† (p)u− (p). − m u† (p)u+ (p) = + (2.) The independent wave functions for a free particle. ψp (x) = e−ik·x u(p) with p0 = ±E.40) (for the same spin components when (2. as in the non-relativistic case. ˜ v E+m u− (p)u− (p) = − (2. would give (for the same spin components) E (2π)3 δ3 (p − p′ ). m † d3 x ψp (x)ψp′ (x) = (2. as we will see shortly.45) However. u† u.38) and (2.44) Our naive instinct will be to normalize the wave function.46). a relativistically covariant normalization would be to require (for the same spin components) E = u† (p)u− (p). therefore.46) (Remember that this will correspond to the probability density and. this will reduce to u† u± = 1 which corresponds to ± the non-relativistic normalization. must be positive. ˜ u E +m Similarly. + − (2.

σ·p ˜ E + m u(p) σ·p − E + m v (p) ˜ .48) or.38 2 Solutions of the Dirac equation u† (p)u+ (p) = + or. with this normalization. E+m E + m 2m (2. 2m (2. with this normalization. . 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) ˜ .50) u− (p)u− (p) = − This particular product.44) (for the same spin components) u+ (p)u+ (p) = 2m 2m E + m |α|2 = = 1. we will obtain from (2. therefore.43) and (2. For example. 2m E 2E |β|2 = . v (p) ˜ (2. α = α∗ = 2E E |α|2 = . Therefore. E+m E + m 2m 2m 2m E + m |β|2 = − = −1. u† (p)u− (p) = − E+m m β = β∗ = E+m . Let us also note here that by construction the positive and the negative energy solutions are orthogonal. E+m m E+m . appears to be a Lorentz invariant (scalar) and we will see later that this is indeed true.49) It is also clear that.

(There is no rest frame for a massless particle. In this case. an alternative normalization which works well for both massive and massless particles is given by u† (p)u+ (p) = E = u† (p)u− (p). E+m E+m .2. u† (p)u− (p) = − E+m β = β∗ = E+m . we will obtain from (2. the solutions we have constructed correspond to four linearly independent. . orthonormal solutions of the Dirac equation. ˜ E+m σ·p u− (p)u+ (p) = −2β ∗ α v † (p) ˜ u(p) = 0.2 Normalization of the wave function 39 u† (p)u− (p) + = α∗ β = 0.53) This still behaves like the time component of a four vector (m is a Lorentz scalar). Note.54) or.51) Therefore. − + (2. 2 2E |β|2 = E.52) While we will be using this particular normalization for massive particles.) The probability density has to be well deﬁned. that σ·p v (p) = 0. =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. α = α∗ = 2E |α|2 = E. 2 (2. let us note that it becomes meaningless for massless particles. Correspondingly. ˜ E+m u+ (p)u− (p) = −2α∗ β u† (p) ˜ (2.38) and (2.40) (for the same spin components) u† (p)u+ (p) = + or. however.

E +m 2m |β|2 = −m. 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 . d3 x ψ † (x)ψ(x) = 1. it can be understood in light of what we have already pointed out earlier as follows. (2. 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). The true normalization is really contained in the total probability. in this case.3 Spin of the Dirac particle As we have argued repeatedly. Let us deﬁne a four dimensional generalization of the Pauli matrices as (in this section. This product continues to be a scalar.40 2 Solutions of the Dirac equation Correspondingly. while the arbitrariness in the normalization of u(p) may seem strange. we obtain 2m |α|2 = m. Let us note here parenthetically that. we will use the notations of three dimensional . ψ(x) = (2. That this is indeed true 2 can be seen explicitly as follows.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 once again that this is a particularly convenient normalization for massless particles. u− (p)u− (p) = − E+m u+ (p)u+ (p) = (2.57) Namely. 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.55) which vanishes for a massless particle.

αj ˜ ˜ = = This shows that 1 αi satisﬁes the angular momentum algebra (remem2˜ ber = 1) and this is why we call the matrices. i = 1. σj ] 0 2iǫijk σk = 2iǫijk αk .58) It can. the generalized ˜ Pauli matrices.101) through the relation 0 σi σi 0 αi = where = ρ˜ i = αi ρ.) Furthermore. 2. (Note. of course. let us note that . σj ] 0 2iǫijk σk 0 0 [σi .61) From the structures of the matrices αi and αi we conclude that ˜ [σi .2.62) αi . ˜ (2. (2.59) and write αi = ραi = αi ρ. αi . however. that this deﬁnes a reducible representation of spin generators since the matrices are block diagonal. (2. 3. ˜ (2.59) ρ= 0 ½ ½ 0 .98) and (1.3 Spin of the Dirac particle 41 Euclidean space since we will be dealing only with three dimensional vectors) σi 0 0 σi αi = ˜ . be checked readily that this is related to the αi matrices deﬁned in (1.60) We note that ρ2 = ½ so that we can invert the deﬁning relation (2. α ˜ (2.

Let us recall that the orbital angular momentum operator is given by Li = ǫijk xj pk . the angular momentum operators which generate rotations should commute with the Dirac Hamiltonian. σj ] = 2iǫijk σk 0 −[σi . (2.42 2 Solutions of the Dirac equation αi . 3. the Dirac equation transforms covariantly under a Lorentz transformation. β ˜ = = 0. let us look at the free Dirac Hamiltonian in (1. αj ˜ = σi 0 0 σi 0 0 σj σj 0 − 0 σj σj 0 σi 0 0 σi = σi σj − σj σi 0 [σi . ½] = 2iǫijk αk . 2. j. (2. i. (2.64) As we will see in the next chapter.100) (remember that we are using three dimensional Euclidean notations in this section) H = α · p + βm = αi pi + βm. Consequently. In other words.65) Calculating the commutator of this operator with the Dirac Hamiltonian. we obtain . therefore. σj ] 0 2iǫijk σk [σi . ½] 0 0 0 σi σj − σj σi 0 = [σi . Lorentz transformations deﬁne a symmetry of the Dirac Hamiltonian and. rotations which correspond to a subset of the Lorentz transformations must also be a symmetry of the Dirac Hamiltonian.63) With these relations at our disposal. k = 1. αi .

H] = [ǫijk xj pk . we note that αi .66) we obtain 1 αi . ˜ 2 Ji = Li + (2. the total angular momentum which should commute with the Hamiltonian. we note that the orbital angular momentum operator does not commute with the Dirac Hamiltonian. αj pj + βm ˜ ˜ αi . can be identiﬁed with 1 αi . αℓ pℓ ] = ǫijk αℓ [xj . the total angular momentum which should commute with the Hamiltonian must contain a spin part as well.68) In other words. Thus. H ˜ = = αi . Consequently.66) Here we have used the fact that since β is a constant matrix and m is a constant. so that combining this relation with (2.67) = 2iǫijk αk pj = −2iǫijk αj pk . (2. β m ˜ (2. (2. αℓ pℓ + βm] = [ǫijk xj pk . αj pj + αi .69) In this case. To determine the spin angular momentum. H ˜ 2 1 αi .3 Spin of the Dirac particle 43 [Li . H] + = iǫijk αj pk − iǫijk αj pk = 0. H ˜ 2 Li + = [Li . therefore. we can identify the spin angular momentum operator with . pℓ ]pk = ǫijk αℓ (iδjℓ ) pk = iǫijk αj pk . if rotations are a symmetry of the system. the second term in the Hamiltonian drops out of the commutator.2.

that 1 1 α3 = ˜ 2 2 σ3 0 0 σ3 S3 = . written in the Hamiltonian form.74) This is the continuity equation for the probability current density associated with the Dirac equation and we note that we can identify . in particular.70) Note. ∂t i (2. (2.72) Taking the Hermitian conjugate of this equation. ˜ 2 (2. is given by ∂ψ = Hψ = − iα · ∇ + βm ψ. or. Multiplying (2.44 2 Solutions of the Dirac equation Si = 1 αi . we conclude 2 that the particle described by the Dirac equation corresponds to a spin 1 particle.4 Continuity equation The Dirac equation. ∂t (2.71) which has doubly degenerate eigenvalues ± 1 .72) with ψ † on the left and (2. Therefore. ∂t ∂t iψ † or. ∂t ∂ (ψ † ψ) = −∇ · ψ † αψ . 2 2. ∂t −i (2.73) with ψ on the right and subtracting the second from the ﬁrst.73) where the gradient is assumed to act on ψ † . we obtain ∂ψ † ∂ψ +i ψ = −i ψ † α · ∇ψ + (∇ψ † ) · αψ . i ∂ (ψ † ψ) = −i∇ · ψ † αψ . we obtain ← − ∂ψ † = ψ † iα · ∇ + βm).

must transform like the time coordinate under a Lorentz transformation. J) = (ψ † ψ. (We are.80) . the total probability d3 x ρ = d3 x ψ † ψ. (2.78) This.) On the other hand. shows that the probability density.77)) and. ρ. to write the continuity equation as ∂ρ = −∇ · J. P = (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. (2. is the time component of j µ (see (2. It is worth recalling that we have already used this Lorentz transformation property of ρ in deﬁning the normalization of the wave function. ∂t This suggests that we can write the current four vector as J µ = (ρ.75) (2.4 Continuity equation 45 ρ = ψ † ψ = probability density. of course. ψ † αψ). yet to show that j µ transforms like a four vector which we will do in the next chapter. therefore.77) (2. in fact.79) is a constant independent of any particular Lorentz frame. J = ψ † αψ = probability current density.76) so that the continuity equation can be written in the manifestly covariant form ∂µ J µ = 0.2. and note that the Hermitian conjugate of ψ satisﬁes (2.

(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. ← − ψ − iγ µ ∂µ − m = 0. 3) γ 0 γ µ γ 0 = (γ µ )† . in fact.86) (2. 1.46 2 Solutions of the Dirac equation ← − ψ † − i(γ µ )† ∂µ − m = 0.84) This is. J = ψγψ = ψ † γ 0 γψ = ψ † αψ.81) Multiplying this equation with γ 0 on the right and using the fact that (γ 0 )2 = ½.82) where we have used the property of the gamma matrices that (for µ = 0. which is what we had derived earlier in (2.80) with ψ on the left and (2. we obtain ← − i ψγ µ ∂µ ψ + ψγ µ ∂µ ψ = 0. Note from the deﬁnition in (2. or. we obtain ← − ψ − iγ 0 (γ µ )† γ 0 ∂µ − m) = 0.85) . the covariant continuity equation and we can identify J µ = ψγ µ ψ.83) Multiplying (2. (2. (2. i∂µ ψγ µ ψ = 0. or. γ 0 (γ µ )† γ 0 = γ µ . or. (2. (2.77). 2. ∂µ ψγ µ ψ = 0.

1: Energy spectrum for a free Dirac particle. Even when the probability density is consistently deﬁned. the probability density is independent of time derivative much like the Schr¨dinger equation.2. 2.1.5 Dirac’s hole theory We have seen that Dirac’s equation leads to both positive and negative energy solutions. the probability density.38) and (2. the energy eigenvalues are given by p0 = E± = ±E = ± p2 + m2 .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. m −m Figure 2. In the free particle case. for example. in such a case. 2. the presence of negative energy solutions leads to many conceptual diﬃculties. can be deﬁned to be positive deﬁnite even in the presence of negative energy solutions. because it is a ﬁrst order equation (particularly in the time derivative).40). Consequently. 2. (2. we note that the energy spectrum .87) Thus.1 (as well as from the equation above) that the positive and the negative energy solutions are separated by a gap of magnitude 2m (remember that we are using c = 1). First of all. This is rather diﬀerent from the case of the Klein-Gordon equation that we have studied in chapter 1. as we o have seen explicitly in (2. We note from Fig. even for this simple case of a free particle the energy spectrum has the form shown in Fig.

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

a negative energy electron can make a transition to a positive energy state and can appear as a positive energy electron.6 Properties of the Dirac matrices The Dirac matrices. (This unconventional deﬁnition of a vacuum state can be avoided in a second quantized ﬁeld theory which we will study later. In this section. 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. were crucial in taking a matrix square root of the Einstein relation and. 3. However. the four Dirac matrices satisfy (in addition to the Cliﬀord algebra) (γ 0 )† = γ 0 .2. (2. the Dirac theory predicts an anti-particle of equal mass for every Dirac particle.6 Properties of the Dirac matrices 49 ple. 2. 2. (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. γ µ .) This is Dirac’s theory of electrons and works quite well.88) Tr γ µ = 0. in deﬁning a ﬁrst order equation. (γ i )† = −γ i . thereby. the Dirac equation passes as a one particle equation primarily because of the Pauli exclusion principle. a positron – and the process under discussion is commonly referred to as pair creation (production).) In spite of this. Thus. 1. Furthermore. 2. 3. . we will study some of the useful properties of these matrices. As we have seen. this is a general feature that combining quantum mechanics with relativity necessarily leads to a many particle theory. i = 1. µ = 0. 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. if enough energy is provided to such a ground state. The amount of energy necessary to excite a negative energy electron to a positive energy state is E ≥ 2m. On the other hand. This “hole” state is what we have come to recognize as the anti-particle – in this case.

To obtain the other basis matrices. Note that.91) that we can identify this with the matrix ρ deﬁned earlier in (2. (2. a complete set of Dirac matrices must consist of 16 such matrices.50 2 Solutions of the Dirac equation Since these are 4 × 4 matrices. Note that in our particular representation for the γ µ matrices given in (1. let us deﬁne the following sets of matrices.90) i ǫµνλρ γ µ γ ν γ λ γ ρ .93) . (2. (2. the identity matrix will correspond to one of them. (2. Of course. 2 γ5 = ½.91).92) We recognize from (2.89) represents the four-dimensional generalization of the Levi-Civita tensor. 4! (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 . by deﬁnition. † γ5 = γ5 .60).91) where we have used the property of the Pauli matrices σ1 σ2 σ3 = i½. Let γ5 = iγ 0 γ 1 γ 2 γ 3 = − where ǫ0123 = 1 = −ǫ0123 .

6 Properties of the Dirac matrices 51 and that. k = 1.95) Since we know the explicit forms of the matrices ½. = i η µν − γ ν γ µ whose explicit forms in our representation can be worked out to be (i. 1. 3) . γ µ + = 0. µ = 0. we can deﬁne four new matrices as γ5 γ µ . 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 .94) Given the matrix γ5 . 3. ½ 0 = σi . 2. 2. 1. γ5 . γ ν = (γ µ γ ν − γ ν γ µ ) 2 2 (2. it anticommutes with any one of them. since it is the product of all the four γ µ matrices. (2. Namely. (2. 3) σ µν = −σ νµ = i i γ µ . we can also deﬁne six anti-symmetric matrices.97) = −i η µν − γ µ γ ν . σ µν . (2. as (µ.2. ν = 0. 2. j.96) Finally. γ µ and γ5 in our representation. let us write out the forms of γ5 γ µ also in this representation.

namely. in fact. γµ. From (2.98) 1 We have already seen in (2.52 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 . γµ. namely. 4. that the matrices 1 1 αi = ǫijk σ jk . 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 . the notation is suggestive and stands for the fact that ψΓ(S) ψ .99) can be identiﬁed with the spin operators for the Dirac particle.98) using the identity for products of Levi-Civita tensors.) We have thus constructed a set of sixteen Dirac matrices. therefore. 1. γ5 . ǫijk ǫℓjk = 2 δiℓ . ˜ (2. ¯ Here. Γ(S) Γ(V ) Γ(A) = = ½. (This relation can be obtained from (2. 4. ˜ 2 4 (2. (2. σ µν . provide a basis for all the 4 × 4 matrices.70) that the matrices 2 αi represent the ˜ spin operators for the Dirac particle. 6. 1.98) we conclude.100) Γ(T ) = Γ(P ) = = γ5 where the numbers on the right denote the number of matrices in each category and these.

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

105) We can also calculate various other commutation relations in a straightforward and representation independent manner. B + C − B A. −i η νλ − γ ν γ λ γλ − γν γµ. BC] = ABC − BCA = (AB + BA)C − B(AC + CA) = A. (2. C] .107) We note here parenthetically that the commutator in (2. we have used the fact that (2. since γ µ matrices satisfy simple anti-commutation relations. we obtain .106) [A. γν + + = 2i η µν γ λ − η µλ γ ν . σ νλ = γ µ . γν γ5 . γ ν γ λ = i γµ. γ µ . For example. γλ = i γ µ. (2. In this derivation. γ µ + + = 2η µν ½.108) However.54 2 Solutions of the Dirac equation γµ. C + . BC] = [A. B] C + B [A. Similarly.107) is more useful for our purpose. the form in (2. for the commutator of two σ µν matrices.107) can also be expressed in terms of commutators (instead of anti-commutators) as [A. = 0. (2.

γ µ = 2η νµ γµ − 4γ = 2γ ν − 4γ ν = −2γ ν .112) .111) (2. σ λρ = iγ µ γ ν . σ λρ γ ν = iγ µ 2iη νλ γ ρ − 2iη νρ γ λ + i 2iη µλ γ ρ − 2iη µρ γ λ γ ν = −2i η µλ i(η νρ − γ ρ γ ν ) + η νρ − i(η µλ − γ µ γ λ ) −η µρ i(η νλ − γ λ γ ν ) − η νλ (−i(η µρ − γ µ γ ρ ) = −2i η µλ σ νρ + η νρ σ µλ − η µρ σ νλ − η νλ σ µρ . Thus. σ λρ = − i η µν − γ µ γ ν . 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. This becomes particularly useful in calculating various amplitudes involving Dirac particles.6 Properties of the Dirac matrices 55 σ µν . for example. 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. The various commutation and anti-commutation relations also lead to many algebraic simpliﬁcations in dealing with such matrices. γµ A µ = γµ Aν γ ν γ µ = Aν γµ γ ν γ µ = −2Aν γ ν = −2A /γ /.110) (2. σ λρ = i γ µ γ ν . and it follows now that. + ν − γµγν (2.109) Thus. (these relations are true only in 4-dimensions) γµ γ ν γ µ = γµ γ ν . σ λρ + i γ µ .2. Similarly. (2.

(2. 2 Tr γ µ γ ν γ λ γ ρ = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ . Therefore. γ ν + and so on.116) .113) = 2η λµ γµ γ ν + 2γ ν γ λ = 2γ λ γ ν + 2γ ν γ λ = 2 γ λ .115) Even more complicated traces can be evaluated by using the basic relations we have developed so far. γ µ + − γµγλ = 4η λν ½.56 2 Solutions of the Dirac equation γµ γ ν γ λ γ µ = γµ γ ν γ λ . we note that Tr γ µ γ ν γ λ γ ρ = Tr γµ. γ ν 2 2 1 = Tr 2η µν ½ = η µν Tr ½ = 4η µν . or. (2. γ ρ or. − γργµ Tr γ µ γ ν γ λ γ ρ = 4 η µν η λρ − η µλ η νρ + η νλ η µρ . 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 .114) Tr γ µ γ ν + Tr γ5 γ µ (2. we know from the cyclicity of traces that Tr γ µ γ ν = Tr γ ν γ µ . The commutation and anticommutation relations also come in handy when we are evaluating traces of products of such matrices. For example. 2 = Tr γ µ γ5 = −Tr γ5 γ µ = 0. γ λ + = 8η µν η λρ − 8η µλ η νρ + Tr γ ν γ λ γ µ . = (2. = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ − Tr γ ν γ λ γ ρ γ µ . it follows (in 4-dimensions) that 1 1 Tr γ µ γ ν + Tr γ ν γ µ = Tr γ µ . For example.

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

−γ5 γµ . a = S. P deﬁne a complete basis for 4 × 4 matrices. γµ . we can construct the inverse set of matrices as Γ(a) . A.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 . Explicitly.122) is straightforward.126) With this.124) Tr Γ(a) Γ(b) = δ(a) .125) Γ(a) = (2.122) From the properties of the γ µ matrices. γ5 γ µ . For example. Tr Γ(a) Γ(a) Γ(a) = such that a not summed. γ5 } . the linear independence of the set of matrices in (2. it can be easily checked that Tr Γ(a) Γ(b) = 0. σµν . 1 S=√ 2 ½ + γ5 γ 0 . the sixteen Dirac matrices Γ(a) . V. 4 (b) (2. (2.100). As a result. γ5 } .1 Fierz rearrangement. we can write the inverse set of matrices as 1 {½. given this set of matrices.6. We have explicitly constructed the sixteen matrices to correspond to the set Γ(a) = {½. (2. (2.123) where “Tr” denotes trace over the matrix indices. T. a = b.121) 2. σ µν . (2. γ µ . it follows now that if . As we have pointed out in (2. This is easily demonstrated by showing that they are linearly independent which is seen as follows.

As a result they constitute a basis for 4 × 4 matrices. (2. C(a) δ(b) = 0. or.129). C(b) = 0. A. (M ) (2. (2. namely. (2. (M ) C(a) δ(b) . where b is arbitrary.130) C(b) = Tr Γ(b) M .125) we obtain C(a) Γ(a) = 0.122) provide a basis for the 4 × 4 matrix space.2. T. we obtain . M= (a) C(a) Γ(a) . V. we obtain C(a) Tr Γ(b) Γ(a) = (a) (a) (M ) (M ) (a) Tr Γ(b) M = or.6 Properties of the Dirac matrices 59 C(a) Γ(a) = 0. (a) (a) or.127) then.127) with Γ(b) . and taking trace over the matrix indices and using (2. (a) (2. (a) C(a) Tr Γ(b) Γ(a) = 0. any arbitrary 4 × 4 matrix M can be expanded as a linear superposition of these matrices. Since the set of matrices in (2. (a) Tr Γ(b) or. multiplying (2.122) are linearly independent.129) Multiplying this expression with Γ(b) and taking trace over the matrix indices. Substituting (2.130) into the expansion (2.127) implies that all the coeﬃcients of expansion must vanishing which shows that the set of sixteen matrices Γ(a) in (2. P .128) for any b = S. Therefore.

then using (2. this leads to Mαβ = (a) Γ(a) (a) γη Mηγ Γαβ . 3. ¯ ¯ ¯ ¯ (2.132) we can derive (for simplicity. Equation (2. For example. it is now straightforward to obtain . we note that if M and N denote two arbitrary 4 × 4 matrices. Just like any other completeness relation. β. ¯ ¯ ¯ ¯ 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 .133). (a) Γ(a) γη Γ(a) αβ = δαη δβγ .132) Here α. it can be used eﬀectively in many ways.60 2 Solutions of the Dirac equation M= (a) Tr Γ(a) M Γ(a) .131) Introducing the matrix indices explicitly. 4 and correspond to the matrix indices of the 4 × 4 matrices and we are assuming that the repeated indices are being summed. 2. or. 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 . (2.133) Using the relations in (2. (2.132) describes a fundamental relation which expresses the completeness relation for the sixteen basis matrices. γ. η = 1.

N and any spinors. In deriving these identities.122) and (2.2. we can ˜ deﬁne a new spinor ψ4 = N ψ4 to write the identity in (2. On the other hand. we can write the ﬁrst of the Fierz rearrangement identities in (2. we have assumed that the spinors are ordinary functions.134) are known as the Fierz rearrangement identities which are very useful in calculating cross sections.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) pick up a negative sign which arises from commuting the fermionic ﬁelds past one another.135) equivalently as . if they correspond to anti-commuting fermion operators. (2. the right-hand sides of the identities in (2. (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 . ¯ = 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 .134) The two relations in (2. Let us note that using the explicit forms for Γ(a) and Γ(a) in (2.135) Since this is true for any matrices M.124) respectively.

L. Bjorken and S. .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). (2. Schiﬀ. Quantum Mechanics.136) ¯ ¯ ψ1 (½ − γ5 ) γ µ ψ2 ψ3 (½ − γ5 ) γµ ψ4 ¯ ¯ = −ψ1 (½ − γ5 ) γ µ ψ4 ψ3 (½ − γ5 ) γµ ψ2 . choose M = (½ − γ5 ) γ µ . 1964. 2. McGraw-Hill. D. (2.62 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 .7 References 1. (2. New York. 1968.110) and (2. McGraw-Hill. Drell. J.136) Thus.113). if we which is often calculationally simpler. I. N = (½ − γ5 ) γµ . for example.137) then using various properties of the gamma matrices derived earlier as well as (2. 2. Relativistic Quantum Mechanics. we obtain from (2. New York.

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

This page intentionally left blank .

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. x′ 1 x′ 3 0 0 1 x3 (3. where R represents the rotation matrix such that x′ = cos θ x1 − sin θ x2 .2) can also be written in matrix form as x1 cos θ − sin θ 0 ′ x2 = sin θ cos θ 0 x2 . we are well acquainted with rotations.3) that a 65 .3) so that the coeﬃcient matrix on the right hand side can be identiﬁed with the rotation matrix in (3. 1 (3. 2 x′ = x3 . For example. we see from (3.2) (3. Thus.1 Lorentz transformations In three dimensions. but this can also be written in terms of the four vector notation we have developed. 3 Here we are using a three dimensional notation. The rotation around the z-axis in (3.1).Chapter 3 Properties of the Dirac equation 3.1) x′ = sin θ x1 + cos θ x2 .

x′ 2 = x2 . (3. we also know that the coordinates transform as (boost velocity v = β since c = 1) xµ → x′ µ .66 3 Properties of the Dirac equation rotation around the 3-axis (z-axis) or in the 1-2 plane is denoted by an orthogonal matrix. R (RT R = ½).4) where we have identiﬁed θ = ǫ = inﬁnitesimal. 2 (3. (3. 3 x′ = ǫx1 + x2 .6) where the Lorentz factor γ is deﬁned in terms of the boost velocity to be 1 1 − β2 γ= . x′ 1 = −γβx0 + γx1 .5) x′ 0 = γx0 − γβx1 . Under a Lorentz boost along the x-axis. x′ 3 = x3 .6) can also be written in the matrix form as . We observe here that the matrix representing the inﬁnitesimal change under a rotation is anti-symmetric.7) We recognize that (3. with unit determinant. such that (3. We also note from (3.2) that an inﬁnitesimal rotation around the 3-axis (z-axis) takes the form x′ = x1 − ǫx2 . 1 x′ = x3 .

where cosh ω − sinh ω 0 0 cosh ω 0 0 (3. (3.1 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. .11) 1 β2 − = 1.3.8) 1 1 − β2 β .12) . = 0 0 1 0 x2 0 0 0 1 x3 x′ 0 γ (3. we note that a Lorentz boost along the x-direction can be written as x′ µ = Λµν xν . 1 0 0 1 (3. Thus.9) that −∞ ≤ ω ≤ ∞. 1 − β2 1 − β2 (3.9) 1 − β2 cosh2 ω − sinh2 ω = Since the range of the boost velocity is given by −1 ≤ β ≤ 1.

15) also shows that the covariant vector transforms in an inverse manner compared with the contravariant vector.) Furthermore. (That these rotations become complex is related to the fact that the metric has opposite signature for time and space components.14) sinh ω = 0 0 0 0 cosh ω 0 0 . ω. the “angle” of rotation. Lorentz boosts correspond to noncompact transformations unlike space rotations.16) .13) to be an ν orthogonal matrix in the sense that µ (ΛT )µν Λνλ = Λνµ Λνλ = δλ . 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). 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. as we have seen. as a result. 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). µ (3.) Therefore. we can obtain. (or the parameter of boost) can take any real value and.12) or (3.13) The matrix representing the Lorentz transformation of the coordinates.15) and also has a unit determinant. namely.68 From this. (3. = 0 0 1 0 0 0 0 1 1 Λµν (3. much like the rotation matrix R in (3. x′ = Λµν xν .3). Λµ (or Λµν ). 0 1 0 0 0 1 sinh ω (3. is easily seen from (3. (Incidentally. then we can write (ω = ǫ = inﬁnitesimal) −ǫ 0 0 −ǫ 1 0 0 = δµν + ǫµν . (3.

(3.18) In other words.3.20) . = 0 0 0 0 0 0 0 0 0 ǫµν = η νλ ǫµλ (3. namely. which can be thought of as rotations in the four dimensional space-time. or. which leave the length of the vector invariant. 0 −ǫ 0 0 0 0 0 ǫµν −ǫ = 0 0 0 0 . ρ Λµρ Λµσ = δσ . Λµρ Λµσ = ηρσ . ηµν Λµρ xρ Λνσ xσ = ηµν xµ xν . or. ηµν Λµρ Λνσ xρ xσ = ηρσ xρ xσ . In a general language. (3. we note that we can combine rotations and Lorentz boosts into what are known as the homogeneous Lorentz transformations.1 Lorentz transformations 69 where. the matrix representing the change under an inﬁnitesimal Lorentz boost is anti-symmetric just like the case of an inﬁnitesimal rotation. therefore. or. or. 0 0 0 0 (3. General Lorentz transformations are deﬁned as the transformations x′ µ = Λµν xν .17) It follows from this that ǫ 0 0 −ǫ 0 0 0 = −ǫνµ .19) ηµν x′ µ x′ ν = ηµν xµ xν .

25) .20)) (det Λ)2 = 1. or. Λ00 ≤ −1. and (3. of course. (3. “gonia” in Greek means an angle or a corner and. Λ00 Λ00 2 2 − Λi 0 2 = 1.24) µ ν (3. we conclude that Λ00 ≥ 1. Namely. orthochronous means straight up in time. an orthodontist is someone who can make your teeth straight.23) = Λνµ and ΛT ν µ = Λνµ The set of homogeneous Lorentz transformations satisfying Λ00 ≥ 1. Thus.22) If Λ00 ≥ 1. Choosing ρ = σ = 0.21) Therefore.70 3 Properties of the Dirac equation This is. 2 = 1 + Λi 0 ≥ 1. what we have seen before in (3. therefore.) Note also that since ΛT Λ = ½. then the transformation is called orthochronous. Incidentally. (The Greek preﬁx “ortho” means straight up. Lorentz transformations deﬁne the maximal symmetry of the space-time manifold which leaves the origin invariant. we can write out the relation (3. (3.15). we obtain (for clarity. det Λ = 1. det Λ = ±1. or. or. or. In the same spirit. we note that ΛT as can be seen from (3. orthogonal means the corner that is straight up (perpendicular).20) explicitly as Λ00 Λ00 + Λi 0 Λi 0 = 1. such a Lorentz transformation does not change the direction of time. (3.

If we add time reversal. to a proper orthochronous Lorentz transformation. we can obtain the other Lorentz transformations by simply appending space reﬂection or time reﬂection or both (which are discrete transformations). det Λ = −1 (which is orthochronous but no longer proper). x → −x.1 Lorentz transformations 71 are known as the proper. det Λ = ±1.28) satisfying Λ00 ≤ −1 and det Λ = −1 (which is neither proper nor orthochronous). namely. 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. xµ → −xµ .27) This would correspond to having Λ00 ≥ 1. (Just to emphasize. Λ00 ≥ 1. Finally.26) Λ00 ≤ −1. there are four kinds of Lorentz transformations. t → −t.) In general. then by adding space reﬂection. Given the proper orthochronous Lorentz transformations. we obtain a Lorentz transformation . however. orthochronous Lorentz transformations and constitute a set of continuous transformations that can be connected to the identity matrix. (3. if we add both space and time reﬂections. Thus.3. = 0 0 1 0 0 0 0 1 Λ = Λtime Λprop (3. = 0 0 −1 0 0 0 0 −1 Λ = Λspace Λprop (3. if Λprop denotes a proper Lorentz transformation. we obtain a Lorentz transformation 1 0 0 0 0 −1 0 0 Λprop . then we obtain a Lorentz transformation −1 0 0 0 0 1 0 0 Λprop . det Λ = ±1. to a proper Lorentz tranformation.

Under a Lorentz transformation xµ → x′ µ = Λµν xν . With this general deﬁnition. 3.31) where ψ ′ represents the transformed wavefunction and L′ stands for the transformed operator (namely. with the transformed variables).30) −1 0 0 0 −1 0 = 0 0 −1 0 0 0 0 Λprop .72 3 Properties of the Dirac equation Λ = Λspace−time Λprop with Λ00 ≤ −1 and det Λ = 1 (which is proper but not orthochronous). covariance implies that a given equation is form invariant under a particular transformation (has the same form in diﬀerent reference frames). (3. (3.32) . In these lectures. 0 −1 0 (3. In simple terms. however.29) where L is a linear operator. These additional transformations. let us now consider the Dirac equation iγ µ ∂µ − m ψ(x) = 0. we would refer to proper orthochronous Lorentz transformations as the Lorentz transformations.2 Covariance of the Dirac equation Given any dynamical equation of the form Lψ = 0. the operator L. cannot be continuously connected to the identity matrix since they involve discrete reﬂections. we say that it is covariant under a given transformation provided the transformed equation has the form L′ ψ ′ = 0. (3.33) (3.

since the Dirac wavefunction is a four component spinor.36) However. Let us assume that. since ψ(x) is a four component spinor. are a set of four space-time independent matrices and. γ µ . In the notation of other symmetries that we know from studies in non-relativistic quantum mechanics.2 Covariance of the Dirac equation 73 if the transformed equation has the form ′ iγ µ ∂µ − m ψ ′ (x′ ) = 0. under a Lorentz transformation. the Lorentz transformation can also mix up the spinor components (much like angular momentum/rotation). we can deﬁne an operator L(Λ) to represent the Lorentz transformation on the coordinate states as (with indices suppressed) |x → |x′ = |Λx = L(Λ)|x .34) where ψ ′ (x′ ) is the Lorentz transformed wave function. (3. then the Dirac equation would be covariant under a Lorentz transformation. (3. Thus. (3. we can deﬁne the Lorentz transformation acting on the Dirac Hilbert space (Hilbert space of states describing a Dirac particle) as. 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. therefore. in addition to the change in the coordinates. do not change under a Lorentz transformation. (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) .3. Note that the Dirac matrices.35) where S(Λ) is a 4 × 4 matrix. Parenthetically.

from (3.39) (3.35)) ψ ′ (x′ ) = x′ |ψ ′ = S(Λ)ψ(x). Let us also note from (3.38) Namely.74 = 3 Properties of the Dirac equation dx |Λx S(Λ)ψ(x).40) (3. A more physical way to understand this is to note that the Dirac wave function simply consists of four functions which do not change. or. ′ iΛµν γ ν ∂µ − m S −1 (Λ)ψ ′ (x′ ) = 0. the eﬀect of the Lorentz transformation. but get rotated by the S(Λ) matrix.33) that ∂x′ µ = Λµν . Thus. ′ iΛµν Sγ ν S −1 ∂µ − m ψ ′ (x′ ) = 0.42) .37) we obtain (see (3. we can write iγ µ ∂µ − m ψ(x) = 0. ∂xµ ∂x′ ν iγ µ Λνµ ∂ν′ − m S −1 (Λ)ψ ′ (x′ ) = 0. iγ µ (3. or. (3. on the wave function.35) we can write ψ(x) = S −1 (Λ)ψ ′ (x′ ). can be represented by a matrix S(Λ) which depends only on the parameter of transformation Λ and not on the space-time coordinates. ∂xν deﬁne a set of real quantities.37) where the wave function is recognized to be ψ(x) = x|ψ . the matrix S(Λ) must possess an inverse so that from (3. so that. or. Since the Lorentz transformations are invertible. (3. (3. or.41) ∂x′ ν ∂ − m S −1 (Λ)ψ ′ (x′ ) = 0.

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. we see from (3. or. generating Lorentz transformations (for the Dirac wavefunction). Next. Λνσ γ σ (3. (3. γ µ and γ ′µ .2 Covariance of the Dirac equation 75 where we have used (3.40). or. or.43) Let us note that if we deﬁne γ ′ µ = Λµν γ ν . γ σ = 2Λµρ Λνρ ½ = 2η = Λµρ Λνσ 2η ρσ ½ ½. (Λ∗ )µν = Λµν ) γ 0 Λµν γ ν γ 0 = Λµν γ 0 (γ ν )† γ 0 = Λµν γ ν .15)). then.45) where we have used the orthogonality of the Lorentz transformations (see (3. Sγ 0 S † γ 0 γ 0 γ µ γ 0 γ 0 S −1 γ 0 S −1 = γ µ . let us note that since the parameters of Lorentz transformation are real (namely. Λµν γ ν = S −1 γ µ S. Therefore.42) that the Dirac equation will be form invariant (covariant) under a Lorentz transformation provided there exists a matrix S(Λ). γ 0 (S −1 γ µ S)† γ 0 = S −1 γ µ S. by Pauli’s fundamental theorem. γ ′ µ. the matrices γ ′ µ satisfy the Cliﬀord algebra and. or. (3.3. γ ′ ν Λµρ γ ρ . It now follows from (3.44) + = + + µν = Λµρ Λνσ γ ρ . Therefore. (3. there must exist a matrix connecting the two representations.46) † † † . (Sγ 0 S † γ 0 )γ µ (Sγ 0 S † γ 0 )−1 = γ µ . such that Λµν Sγ ν S −1 = γ µ .

100) and calculating the commutator with γ µ ). S † γ 0 = b γ 0 S −1 . b = b∗ .76 3 Properties of the Dirac equation Here we have used (3.50) The real roots of this equation are b = ±1. Using these in (3. As a result. or.47). Sγ 0 S † γ 0 = b∗ ½ = b ½. or. (3. (3. γ 0 (γ 0 Sγ 0 S † )γ 0 = b∗ ½. . therefore. namely. or. or. (3. or. (3.43) and the relations (γ 0 )† = γ 0 = (γ 0 )−1 as well as γ 0 (γ µ )† γ 0 = γ µ .47). γ 0 Sγ 0 S † = b∗ ½. therefore. b4 = 1.47) Taking the Hermitian conjugate of (3. det S = 1.48) which. must be proportional to the identity matrix (this can be easily checked by taking a linear combination of the sixteen basis matrices in (2. determines that the parameter b is real.46) that the matrix Sγ 0 S † γ 0 commutes with all the γ µ matrices and. we determine det (Sγ 0 S † γ 0 ) = det (b ½).49) We also note that det γ 0 = 1 and since we are interested in proper Lorentz transformations. we obtain (Sγ 0 S † γ 0 )† = b∗ ½. we can determine the unique value of b in the following way.51) In fact. It is clear from (3. we can denote Sγ 0 S † γ 0 = b ½. (3.

S † γ 0 = γ 0 S −1 .53) or. or. (3.52) which follows since S † S represents a non-negative matrix. b > 0.53)) that b > 0 and.51) that b = 1.3.55) 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. The two solutions of this equation are obvious Λ00 ≥ 1. or. b < 0.47) that Sγ 0 S † γ 0 = ½. .2 Covariance of the Dirac equation 77 Let us note. (3.47). 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 . therefore.56) (3.54) which implies (see (3. Since we are dealing with proper Lorentz transformations. (3. Λ00 ≤ −1. Thus. we conclude from (3. Tr S † S = 4bΛ00 0. (3. it follows from (3. we are assuming Λ00 ≥ 1.43) and (3. using (3.

61) (3. therefore. .58) For an inﬁnitesimal transformation.60) ½+ ½+ i Mµν ǫµν 4 ½− i Mστ ǫστ 4 i i Mµν ǫµν − Mµν ǫµν + O(ǫ2 ) 4 4 (3.18). let us consider an inﬁnitesimal Lorentz transformation of the form (ǫµν inﬁnitesimal) x′µ = Λµν xν = (δµ + ǫµ )xν = xµ + ǫµ xν . Mµν = −Mνµ . we recall that the inﬁnitesimal transformation matrix is antisymmetric.63) ½ + O(ǫ2 ). (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. ν ν ν (3.62) (3. 4 (3. Mµν denote matrices in the Dirac space). we can expand the matrix S(Λ) as S(Λ) = S(ǫ) = ½ − i Mµν ǫµν .59) where the matrices Mµν are assumed to be anti-symmetric in the Lorentz indices (for diﬀerent values of the Lorentz indices. ǫµν = −ǫνµ . since ǫµν = −ǫνµ .78 3 Properties of the Dirac equation Next. 4 (3. namely.

Therefore. λ 2 ρ = − (3. σ λρ 4 i = − ǫλρ × 2i η µλ γ ρ − η µρ γ λ 4 1 µ ρ = ǫ γ + ǫµ γ λ = ǫµν γ ν . i ǫλρ γ µ . or.64) i ǫλρ γ µ .66) (3. i i λρ ǫ Mλρ γ µ − ǫλρ γ µ Mλρ + 0(ǫ2 ) = γ µ + ǫµν γ ν . or.67) which coincides with the right hand side of (3. M λρ = ǫµν γ ν .68) . we see that for inﬁnitesimal transformations.2 Covariance of the Dirac equation 79 To the leading order. or. 4 4 (3.106) γ µ .65) − (3. M λρ 4 i ǫλρ γ µ .64) that if we identify M λρ = σ λρ .64). let us recall the commutation relation (2.3.43) now takes the form S −1 (ǫ)γ µ S(ǫ) = Λµν γ ν = δµν + ǫµν γ ν . σ νλ = 2i η µν γ λ − η µλ γ ν . then. and note from (3. 4 4 (3. The deﬁning relation for the matrix S(Λ) in (3. therefore. we have determined the form of S(ǫ) to be S(ǫ) = ½ − i i Mµν ǫµν = ½ − σµν ǫµν . 4 At this point. S −1 (ǫ) indeed represents the inverse of the matrix S(ǫ). ½+ γµ + − i Mλρ ǫλρ γ µ 4 ½− i Mστ ǫστ 4 = γ µ + ǫµν γ ν .

For completeness. (3.69) with the generators of inﬁnitesimal Lorentz transformations for the 1 Dirac wave function.72) .98)). of course. ˜ 2 2 2 (3.71) A ﬁnite rotation by angle θ in the 1-2 plane would.80 3 Properties of the Dirac equation Let us note here from the form of S(ǫ) that we can identify 1 σµν . satisfy can 2 be identiﬁed with the Lorentz algebra (which also explains why they are closed under multiplication).109) which the generators of the inﬁnitesimal transformations.43) and generates Lorentz transformations and as a result. 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. i (3. let us note that inﬁnitesimal rotations around the 3-axis or in the 1-2 plane would correspond to choosing ǫ12 = ǫ = −ǫ21 . A ﬁnite transformation can. 2 (3. Thus. S(ǫ) = ½ − i 1 i σ12 ǫ = ½ + γ1 γ2 ǫ = ½ − α3 ǫ. (The other factor of 2 is there to avoid double counting. then. we have shown that there exists a S(Λ) which satisies (3.70) with all other components of ǫµν vanishing. at least for inﬁnitesimal Lorentz transformations. 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.) We will see in the next chapter (when we study the representations of the Lorentz group) that the algebra (2. consequently. the Dirac equation is form invariant (covariant) under such a Lorentz transformation. 1 σµν . In such a case (see also (2.

therefore. namely. therefore. we can identify ǫ01 = ǫ = −ǫ10 .98)) S(ǫ) = i 1 σ01 ǫ = ½ + γ0 γ1 ǫ 2 2 1 = ½ − α1 ǫ. θ θ − i˜ 3 sin α 2 2 = −S(θ). 0 ½ (3. corresponds to a spinor representation. rotations ˜i ˜ deﬁne unitary transformations.73) α3 ˜ = ½ 0 = ½. (3. (3.77) with all other components of ǫµν vanishing. recalling that σ3 0 0 σ3 α3 = ˜ we have 2 .76) θ θ − i˜ 3 sin . in this case. 2 ½− (3. we can determine S(θ) = cos This shows that S(θ + 2π) = − cos S(θ + 4π) = S(θ). consistent with the fact that the Dirac equation describes spin 1 2 particles. Let us next consider an inﬁnitesimal rotation in the 0-1 plane. namely. so that we can write (see also (2. we have S † (θ) = S −1 (θ).2 Covariance of the Dirac equation 81 Note that since α† = αi . This is. (3.3. α 2 2 (3.75) That is.78) .74) and. Furthermore. is double valued and. In this case. the rotation operator. we are considering an inﬁnitesimal boost along the 1-axis. of course.

α2 = 1 ½ 0 = ½. operators deﬁning boosts are not unitary. 0 ½ (3.79) α1 = . the matrix for a ﬁnite boost. since under a Lorentz transformation . in this four dimensional space (namely. can be obtained through exponentiation as S(ω) = e− 2 α1 ω .82) We note here that since α† = α1 . recalling that 0 σ1 σ1 0 1 (3. as 4 × 4 matrices). there does not exist any ﬁnite dimensional unitary representation. 3.83) That is. Furthermore. Let us note next that.80) and. ω. therefore. (3. 1 S † (ω) = S −1 (ω).82 3 Properties of the Dirac equation In this case. 2 2 S(ω) = cosh (3. This is related to the fact that Lorentz boosts are non-compact transformations and for such transformations. All the unitary representations are necessarily inﬁnite dimensional. (3.3 Transformation of bilinears In the last section.81) we can determine ω ω − α1 sinh . we have shown how to construct the matrix S(Λ) for ﬁnite Lorentz transformations (for both rotations and boosts).

Similarly. 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. (3.43). In other words. ′ (3.50) and (2. we see that if we deﬁne a current of the form j µ (x) = ψ(x)γ µ ψ(x).3.85) where we have used the relation (3. namely.87) ′ where we have used (3.3 Transformation of bilinears 83 ψ ′ (x′ ) = S(Λ)ψ(x). it would transform as a four vector under a proper Lorentz transformation. 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).84) (3.86) ′ ψ(x)ψ(x) → = Namely. under a Lorentz transformation ψ(x)γ µ ψ(x) → = = ψ (x′ )γ µ ψ ′ (x′ ) ψ(x)S −1 (Λ)γ µ S(Λ)ψ(x) ψ(x)Λµν γ ν ψ(x) = Λµν ψ(x)γ ν ψ(x). Thus.88) . J µ (x) = ψ(x)γ µ ψ(x) → Λµν J ν (x). we see that the adjoint wave function ψ(x) transforms inversely compared to the wave function ψ(x) under a Lorentz transformation. it follows that ψ ′ † (x′ ) = ψ † (x)S † (Λ). ψ (x′ ) = ψ ′ † (x′ )γ 0 = ψ † (x)S † (Λ)γ 0 = ψ † (x)γ 0 S −1 (Λ) = ψ(x)S −1 (Λ). (3.55)). (3.56).

89) while the negative energy solutions satisfy − γ 0 p0 − γ · p − m u− (−p0 . / where p0 = ω = p2 + m2 . p) = 0.90) (3. (3. / (3. (3. −p) = u− (−p) = . we note that in this way.49).91). Namely. (3. Finally.89) and (3. of course.93) . E+m 2m σ ·p E+m v ˜ v ˜ v(p) = u− (−p0 .90). can be written as (p − m)u(p) = 0. p) = 0.85)) transforms like a four vector so that the probability density transforms as the time component of a four vector. 3.92) so that the equations satisﬁed by u(p) and v(p) (positive and negative energy solutions).84 3 Properties of the Dirac equation This is. 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) = . the probability current density (see also (2. we can determine the transformation properties of the other bilinears under a Lorentz transformation in a straightforward manner.4 Projection operators.91) with the same value of p0 as in (3. It is customary to identify (see (2. what we had observed earlier. (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 .

4. r = 1. completeness relation 85 and − γ 0 p0 + γ · p − m v(p) = 0. / (3. Let us also note that each of the four solutions really represents a four component spinor. 2. u (p)v (p) = 0 = v r (p)us (p). 3. / or.94) Given these equations. can represent the spin projection of the two component spinors (in terms of which the four component solutions were obtained). 2.83)). / (p + m)v(p) = 0. / v † (p) (p)† + m γ 0 = 0. (3. we can write down the Lorentz invariant conditions we had derived earlier from the normalization of a massive Dirac particle as (see (2. With these notations. / (3. / or. Let us denote the spinor index by α = 1.3. α v r (p)v s (p) = α=1 r s r s v α (p)vα (p) = −δrs .96) where r. (3. as we had seen earlier. or. u(p)(p − m) = 0.50)) 4 u r (p)us (p) = α=1 4 r uα (p)us (p) = δrs . Let us denote them by ur (p) and v r (p).97) . As we have seen earlier there are two positive energy solutions and two negative energy solutions of the Dirac equation. (−p − m)v(p) = 0. v(p)(p + m) = 0.4 Projection operators. 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. or.95) where we have used (γ µ )† γ 0 = γ 0 γ µ (see (2.

94).99) for any two spin components of the positive and the negative energy spinors. the last relation in (3.93) and (3. (3. From the form of the equations satisﬁed by the positive and the negative energy spinors. 4 × 4 matrices and their eﬀect on the Dirac spinors is quite clear.101) . of course. 2m −p + m / . namely. 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. it is clear that we can deﬁne projection operators for such solutions as p+m / . 2m p+m r / v (p) = 0.52)).97) can be checked to be true simply because v(p) = u− (−p0 . This also allows us to write 2 r=1 [u r (p)ur (p) − v r (p)v r (p)] = 4. 2m −p + m r / v (p) 2m −p − m + 2m r / v (p) = v r (p).100) Λ− (p) = 2m These are. 2m −p + m r / u (p) = 0. 2m (3.98) For completeness we note here that it is easy to check u† (p)v(−p) = 0 = v † (−p)u(p).86 3 Properties of the Dirac equation Although we had noted earlier that u+ (p)u− (p) = 0. (3. −p). (3. (3.

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

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

103)) . r=1 r r vα (p)v β (p) = p−m / 2m . Hence we can identify Qαβ = −(Λ− (p))αβ .108) Namely. (3.3. 2 or. the matrix Q projects only on to the space of negative energy solutions with a phase (a negative sign). (3. β β=1 r=1 4 4 s Qαβ (p)vβ (p) = β=1 2 β=1 r=1 r s vα (p)(−δrs ) = −vα (p). 4 s v β (p)Qβα (p) = β=1 2 β=1 r=1 2 s r r v β (p)vβ (p)v α (p) 2 r s r vα (p)v β (p)vβ (p) = (Q(p)v s (p))α = = r=1 4 (v s (p)Q(p))α = = r=1 r s (−δrs )v α = −v α . completeness relation 89 2 Qαβ (p) = r=1 r r vα (p)v β (p). αβ (3. 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) = 0.4 Projection operators.107) then.109) The completeness relation for the solutions of the Dirac equation now follows from the observation that (see (3.

but rather wish to obtain an average over the two possible . For example. (3.90 3 Properties of the Dirac equation Pαβ − Qαβ = (Λ+ (p))αβ + (Λ− (p))αβ = δαβ . the latter being negative while the former is positive. Hence. (3.110) or in (3.112) where M stands for a 4 × 4 matrix. r=1 r r r ur (p)uβ (p) − vα (p)v β (p) = δαβ . α (3. rather it involves a sum with the metric structure of the space built in. As we have seen. uu and vv have opposite sign.111) can be understood as follows.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. let us suppose that we are interested in a transition amplitude which has the form u r (p)M us (p′ ).111) We note here that the relative negative sign between the two terms in (3. M (3. we can think of the space of solutions of the Dirac equation as an indeﬁnite metric space. the completeness relation does not involve a sum of terms with positive deﬁnite sign. 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).110) In a matrix notation.113) If we are not interested in the expectation value in a particular electron state. the completeness relation (3. 2 or. In such a space. These relations are particularly useful in simplifying the evaluations of transition amplitudes and probabilities.

114) Similarly. think of an experiment with unpolarized initial electron states where the ﬁnal spin polarization is not measured).λ=1 2 s Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α.3.4 Projection operators.β=1 1 Tr M Λ+ (p). the probability for such a transition will be determined from 1 u r (p)M us (p′ ) 2 r.s=1 1 2 r.β.β.s=1 1 2 4 2 4 r s uα (p)Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α.s=1 σλ r ur (p)uα (p) λ σλ = ur (p) λ = . 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. completeness relation 91 electron states (in experiments we may want to average over the spin polarization states).λ=1 r.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=1 4 2 r Mαβ ur (p)uα (p) β α.s=1 2 2 u r (p)M us (p′ ) † = 1 u r (p)M us (p′ ) 2 r.β=1 r=1 4 = = = Mαβ (Λ+ (p))βα = α.σ. then we will have 1 2 1 2 1 2 1 2 2 M = u r (p)M ur (p) r=1 2 4 r uα (p)Mαβ ur (p) β r=1 α. 2 (3.σ.

67)) 1 αi .117) satisﬁes the commutation relation (see (2. H = −iǫijk αj pk .6. Note.118) As a consequence. in the relativistic case this is not useful since spin is not a constant of motion. 3. ˜ 2 [Si . 2 σi 0 0 σi S= αi = ˜ . (3. unlike the case of non-relativistic systems where we specify a given energy state by the projection of spin along the z-axis (namely. that.β.λ=1 βσ γ 0M †γ 0 σλ (Λ+ (p))λα (3. In fact. Thus. On the other hand.115) = 1 Tr M Λ+ (p′ )γ 0 M † γ 0 Λ+ (p) . we have already seen that the spin operator 1 ˜ α. spin commutes with the Hamiltonian (since in this frame p = 0) and such solutions can be labelled by the spin projection. (3. the Dirac Hamiltonian H = α · p + βm. namely. we note that since momentum commutes with the Dirac Hamiltonian. 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 2.5 Helicity As we have seen.116) does not commute either with the orbital angular momentum or with spin (rather.92 1 2 4 3 Properties of the Dirac equation = Mαβ Λ+ (p′ α. . by the eigenvalue of Sz ).σ. H] = (3. however. for a particle at rest. it can be easily checked that the plane wave solutions which we had derived earlier are not eigenstates of the spin operator. it commutes with the total angular momentum).

h) with h = ± 1 (the two helicity 2 eigenvalues).122) where we have used (this is the generalization of the identity satisﬁed by the Pauli matrices) (S · p )(S · p ) = 1 1 ˜ ˜ (α · p) (α · p) = p2 ½. h)v(p. H] = [pi . H] = [Si . h′ ) = δhh′ = −v(p. Namely. (3.3.120) Therefore. v(p. the eigenvalues of the helicity operator. The normalized operator h= S·p . the eigenstates of energy can also be labelled by the helicity eigenvalues. h′ ). [Si pi . h)u(p. can only be ± 2 and we can label the positive and the negative energy solutions also as u(p. h′ ). (3. for a Dirac parti1 cle. H]pi = −iǫijk αj pk pi = 0. Note that S·p |p | 2 h2 = = 1 ½. h′ ) = 0 = v(p. h)u(p. (3. 4 4 (3.5 Helicity 93 [pi .119) the operator S · p does also (momentum and spin commute and. u(p. h)v(p. h). The normalization relations in this case will take the forms u(p. 4 (3.123) Therefore. α · p + βm] = 0. therefore.121) measures the longitudinal component of the spin of the particle or the projection of the spin along the direction of motion. the order of these operators in the product is not relevant). this operator is a constant of motion. This is known as the helicity operator and we note that since the Hamiltonian commutes with helicity. |p | (3.124) .

h) − v(p.110) or (3. or.126) where we are not assuming any relation between p0 and p as yet.111). (3. which can also be written as p0 u1 (p) − σ · p u2 (p) = mu1 (p). γ 0 p0 − γ · p − m u(p) = 0.6 Massless Dirac particle Let us consider the free Dirac equation for a massive spin (γ µ pµ − m) u(p) = 0. h)v(p.128) Explicitly. (3.125) 1 h=± 2 3. (3. (3. h)] = ½. h)u(p. can now be written as [u(p.94 3 Properties of the Dirac equation Furthermore. (3.127) where u1 (p) and u2 (p) are two component spinors. (3. σ · p u1 (p) − p0 + m u2 (p) = 0.130) .129) p0 u2 (p) − σ · p u1 (p) = −mu2 (p). Let us represent the four component spinor (as before) as u1 (p) u2 (p) u(p) = . In terms of u1 (p) and u2 (p). the Dirac equation takes the form (p0 − m)½ σ·p −σ · p u1 u2 −(p0 + m)½ = 0. the completeness relation. this leads to the two (2-component) coupled equations p0 − m u1 (p) − σ · p u2 (p) = 0. 1 2 particle. (3.

131) p0 + σ · p (u1 (p) − u2 (p)) = m (u1 (p) + u2 (p)) . We note that if we deﬁne two new (2-component) spinors as 1 (u1 (p) − u2 (p)) .6 Massless Dirac particle 95 Taking the sum and the diﬀerence of the two equations in (3. we obtain p0 − σ · p (u1 (p) + u2 (p)) = m (u1 (p) − u2 (p)) . . 2 uL (p) = uR (p) = (3.133) We note that it is the mass term which couples the two equations.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). p0 uL (p) + σ · p uL (p) = muR (p). (3. since vanishing of the mass which is a Lorentz scalar should not change the behavior of the equation under proper Lorentz transformations).3.130). (3. The corresponding two component spinors uL and uR are also known as Weyl spinors. p0 uL (p) = −σ · p uL (p). like the Dirac equation. 2 1 (u1 (p) + u2 (p)) . 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).132) then.134) These two equations. Let us note that in the limit m → 0. are not invariant under parity or space reﬂection and are known as the Weyl equations. (3. however. can be shown to be covariant under proper Lorentz transformations (as they should be. the equations in (3. These equations.

(3.138) For p0 = |p |. we must have p0 = ±|p |.136) Thus.96 Let us note that 3 Properties of the Dirac equation or. = (σ · p) (σ · p) uR (p) = p 2 uR (p). namely. that for such solutions. |p | or.135) or. p0 uR (p) = σ · p uR (p). therefore. (3. we can show that uL (p) also satisﬁes (p0 )2 − p 2 uL (p) = 0. Similarly. for a nontrivial solution of these equations to exist. the two diﬀerent Weyl equations really describe particles with opposite helicity. Recalling that 1 σ denotes the spin 2 . It is clear. we must have (p0 )2 − p 2 = 0. (p0 )2 uR (p) = σ · p p0 uR (p) (p0 )2 − p 2 uR (p) = 0. p0 uL (p) = −σ · p uL (p).137) which is the Einstein relation for a massless particle. (3. while (3. we note that p0 uR (p) = σ · p uR (p).140) In other words. for the positive energy solutions. σ·p uL (p) = −uL (p). (3.139) or. σ·p uR (p) = uR (p). |p | (3.

Such a particle is known as a right-handed particle since its spin motion would correspond to that of a right-handed screw.1: Right-handed and left-handed particles with spins parallel and anti-parallel to the direction of motion. As we know.141) is massless (present experiments suggest they are almost massless) and. can be described by a two component Weyl equation.) p p Left-handed Right-handed Figure 3. the electron neutrino emitted in a beta decay A X Z+1 → A Y Z + e+ + νe . therefore. that νe is left-handed. . Correspondingly. uR (p) describes a particle with helicity + 1 or a particle with spin parallel to its direction of 2 motion. we note that uL (p) describes a particle with helicity − 1 or a particle with spin anti-parallel to its 2 direction of motion.3. This is shown in Fig. then we conclude that for such a particle. (3. namely.1 and we note here that this nomenclature is opposite of what is commonly used in optics. (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. 3.6 Massless Dirac particle 97 operator for a two component spinor. We also know. the circular motion would correspond to that of a left-handed screw. experimentally. On the other hand. such a particle is called a left-handed particle (which is the reason for the subscript L). If we think of spin as arising from a circular motion.

x → −x. so that under a space reﬂection σ · (−p) σ·p σ·p → =− . the absence of a negative energy neutrino would appear as a “hole” with the momentum reversed. |p | (3. observed in experiments such as n → p + e− + ν e . would violate parity. In the hole theoretic language. |p | → |p |. the absence of a negative energy neutrino would appear as a “hole” with opposite helicity. Alternatively.142) and has negative helicity.144) Under parity or space reﬂection. (3. (3. This has been experimentally veriﬁed in a number of processes. then. The neutrino is described by the equation σ·p uL (p) = −uL (p). Therefore. the neutrino equation is not invariant under parity. of course. . p → −p. the anti-neutrino. we conclude that it must transform under parity like L.143) A very heuristic way to conclude that parity is violated in processes involving neutrinos is as follows. It is helicity which is the conserved quantum number and.145) Since σ represents an angular momentum. therefore. (3.146) Consequently. L = x × p → (−x ) × (−p ) = x × p = L. will have opposite helicity or will be right-handed. That the antineutrino is right-handed is. the neutrino is left-handed and hence satisﬁes the equation p0 uL = −σ · p uL . |p | |p | |p | (3. in this description. hence. and processes involving neutrinos.98 3 Properties of the Dirac equation 1 its helicity is − 2 .

(3. (3. Since 2 γ5 = ½. in the Pauli-Dirac representation γ5 = ρ deﬁned in (2. Therefore.60) and ρ commutes with α.100)) H = α · p.151) . (2. then γ5 u(p) (or γ5 v(p)) is also a solution. ˜ |p| v(p) = σ·p v (p) ˜ v (p) ˜ = σ ·p v (p) ˜ |p| v (p) ˜ .147). / / can be written as (see also (3. namely.148) From the structure of the massless Dirac equation (3.53) and (2. for example.150) it follows that the eigenvalues of γ5 are ±1 and spinors with the eigenvalue +1. the solutions of the massless Dirac equation (m = 0) pu(p) = 0 = pv(p).61)). This can also be seen from the fact that the Hamiltonian for a massless Dirac fermion (see (1. (3.147) u(p) = E 2 E 2 = |p| 2 |p| 2 u(p) ˜ σ·p u(p) . we note that if u(p) (or v(p)) is a solution.54). the solutions of the massless Dirac equation can be classiﬁed according to the eigenvalues of γ5 also known as the chirality or the handedness. γ5 uR (p) = uR (p).92)) u(p) ˜ σ·p u(p) ˜ E E (3. (3. see.7 Chirality 99 3.3.7 Chirality With the normalization for massless spinors discussed in (2. γ5 vR (p) = vR (p).149) commutes with γ5 (in fact.

153) where we have deﬁned . (3. Given a general spinor. γ5 uL (p) = −uL (p). We note that if the fermion is massive (m = 0). (3. = 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) ˜ . 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) . 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| .100) would no longer commute with γ5 and in this case chirality would not be a good quantum number to label the states with. γ5 vL (p) = −vL (p). namely. then the Dirac Hamiltonian (1.152) are called left-handed (negative chirality) spinors.100 3 Properties of the Dirac equation are known as right-handed (positive chirality) spinors while those with the eigenvalue −1. vR (p) = PR v(p) = = |p| 2 1 2 1 2 1 (½ + γ5 )v(p) 2 ½ + σ ·p v(p) ˜ |p| .

hχ(±) (p) = σ·p 1 2|p| 2 ½± ½± 1 1 = ± × 2 2 σ·p χ ˜ |p| 1 σ·p χ = ± χ(±) (p). (In the Pauli-Dirac representation.6) that in the massless limit. these projection operators have the explicit forms (see (2.155) PR PL = 0 = PL PR .7 Chirality 101 1 1 PR = (½ + γ5 ). This is connected with our earlier observation (see section 3. PL = . the four component spinors can be eﬀectively described by two component spinors.3. (3. PR + PL = 1. PL 2 = PL . The reducibility of the spinors is best seen in the Weyl representation for the Dirac matrices discussed in (2. ˜ |p| 2 (3. we note that 2 spinors of the form χ(±) (p) = 1 σ·p 1± χ.91)) 1 2 PR = ½ ½ 1 ½ −½ .157) correspond to states with deﬁnite helicity.153) that in the massless Dirac theory. the Dirac equation reduces to two decoupled two component Weyl equations (recall that it is the mass term which generally couples these two spinors).121) (for the two component spinors S = 1 σ).154) 2 2 We note that by deﬁnition these projection operators satisfy PR 2 = PR . However.158) .119). 2 −½ ½ ½ ½ (3. which implies that any four component spinor can be uniquely decomposed into a right-handed and a left-handed component. PL = (½ − γ5 ).156) We note from (3. we will continue our discussion in the Pauli-Dirac representation of the Dirac matrices which we have used throughout. namely. ˜ 2 |p| (3. From the deﬁnition of the helicity operator in (3. (3.

(P (−) )2 = P (−) .163) .159) We note here that the operators P (±) = 1 2 ½± σ·p 1 ˆ = (½ ± σ · p).102 3 Properties of the Dirac equation so that the right-handed (four component) spinors in (3. vR (p) = .120) and pµ = (1.70) or (3. Explicitly. deﬁne projection operators into the space of positive and negative helicity two component spinors.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) = . vL (p) = −v (−) (p) v (−) (p) . uL (p) = .161) with σ µ . P (+) P (−) = 0 = P (−) P (+) . They can be easily generalized to a reducible representation of operators acting on the four component spinors and have the form (see (2. P (+) + P (−) = 1. ˆ 2 (3. |p| 2 P (−) = 1 µ σ pµ .153) and (3. (3.117)) P (±) 0 1 2( P4×4 (±) = 0 P (±) 0 1 2( = ˆ ½ ± σ · p) 0 ˆ ½ ± σ · p) . −ˆ ). It is straightforward ˜ ˆ p to check that they satisfy the relations (P (+) )2 = P (+) . ˜ ˆ 2 (3.153) are described by two component spinors with positive helicity while the lefthanded (four component) spinors are described in terms of two component spinors of negative helicity.160) can also be written in a covariant notation as P (+) = 1 µ σ pµ .162) and. σ µ deﬁned in (2. we see from (3. (3.(3. therefore.

with the choice of the basis in (3.168) . ˜˜ ˜˜ (3.3. For completeness as well as for later use.163) we see that the right-handed spinors with chirality +1 are characterized by helicity +1 while the left-handed spinors with chirality −1 have helicity −1. the helicity spinors simply reduce to eigenstates of σ3 . ˜ ˜ ˜ ˜ uu† + v v † = ½.164) which is the reason the spinors can be simultaneous eigenstates of chirality and helicity (when mass vanishes). v (+) (p) = .7 Chirality 103 and it is straightforward to check from (3. (3. In fact.R . v(+) (p) = ½ + σ · p v. We can choose u and v to be ˜ ˜ normalized so that we have u(+) (p) = u† u = 1 = v † v . P4×4 = 0. 0 v= ˜ 0 . (3.159) as well as (3.167). we can also deﬁne normalized spinors u(+) and v (+) . We note from (3.163) that PL.156) and (3. we can choose u= ˜ 1 . For example. from (3.166) For example. (±) (3.157) 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. Furthermore. 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 .165) 2 2 Each of these spinors is one dimensional and together they span the two dimensional spinor space. 1 (3. let us derive some properties of these spinors.167) such that when p1 = p2 = 0.

R † u† (p)vR (−p) = vR (−p)uR (p) = 0. (3. u(+) (p)u(+)† (p) = v (+) (p)v (+)† (p) = 1 2 ˆ ½ + σ · 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. In general.171) = P (+) P (+) P (+) P (+) .170) can be simpliﬁed by noting the following identity.170) The completeness relation in (3.168). R † uR (p)u† (p) = vR (p)vR (p) = R |p| 2 P (+) P (+) P (+) P (+) . as well as (3.169) which can be checked from the explicit forms of the spinors in (3. We note that with p0 = |p|.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). Given the form of the right-handed spinors in (3. (3.104 3 Properties of the Dirac equation However. (3. 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 . the positive helicity spinors satisfy u(+)† (p)u(+) (p) = v (+)† (p)v (+) (p) = 1. u(+)† (p)v (+) (−p) = 0 = v (+)† (−p)u(+) (p).

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) .173) 3. (3. L † uL (p)u† (p) = vL (p)vL (p) = L 1 0 pγ (½ − γ5 ). (3. / 4 (3. 2m E + m E+m (3.3.4.175) In (3.8 Non-relativistic limit of the Dirac equation 105 so that we can write the completeness relation in (3.172) which can also be derived using the methods in section 3.8 Non-relativistic limit of the Dirac equation Let us recall that the positive energy solutions of the Dirac equation have the form (see (2.49)) u ˜ σ·p ˜ E+mu uL (p) uS (p) u+ (p) = E+m 2m = .174) we have deﬁned uL (p) = uS (p) = E+m u. 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). L † u† (p)vL (−p) = 0 = vL (−p)uL (p). ˜ 2m σ·p E+m σ·p u= ˜ uL (p).170) as † uR (p)u† (p) = vR (p)vR (p) = R 1 0 pγ (½ + γ5 ). / 4 (3.

m½ σ · p −m½ σ·p uL (p) uS (p) =E uL (p) uS (p) . Let us next look at the positive energy solutions in (3. 2m E + m E+m (3. or. uL (p) and uS (p) are known as the large and the small components of the positive energy Dirac solution. when |p | ≪ m. the component uS (p) is much smaller than (of the order of v ) uL (p) and c correspondingly. in (3.106 3 Properties of the Dirac equation and we emphasize that the subscript “L” here does not stand for the left-handed particles introduced in the last section. (3. (α · p + βm) u+ (p) = Eu+ (p).175) we have denoted vL (p) = vS (p) = − E+m v. (3.174). vL (p) and vS (p) are also known as the large and the small components of the negative energy solution.179) gives the relation σ·p uL (p). Similarly.180) . or.179) We note that the second equation in (3. In the non-relativistic limit. Similarly. which satisfy the equation Hu+ (p) = Eu+ (p). E+m uS (p) = (3.178) This would lead to the two (2-component) equations σ · p uS (p) = (E − m)uL (p). we expect the large components to give the dominant contribution to the wave function. σ · p uL (p) = (E + m)uS (p). ˜ 2m σ·p E +m σ·p v=− ˜ vL (p).177) It is clear that in the non-relativistic limit.

This is.33) and remember that we are choosing = 1) . if we identify the non-relativistic energy (without the rest mass term) as ENR = E − m. σ·p uL (p) = (E − m)uL (p). 2 3.3. This preserves the gauge invariance associated with the Maxwell’s equations and corresponds to deﬁning pµ → pµ − eAµ . E ≈ m (recall that we have set c = 1). what we know for a free non-relativistic electron (spin 1 particle). equation (3. E+m (3. the Dirac equation in this case reduces to the Schr¨dinger o equation for a two component spinor which we are familiar with. (3. and. Since the coordinate representation of pµ is given by (see (1. of course. 2m (3.183) Namely.179) takes the form (σ · p) or.181) p2 uL (p) ≃ (E − m)uL (p). Furthermore. then. 2m where we have used the fact that for a non-relativistic system.181) has the form p2 uL (p) = ENR uL (p).182) (3. therefore. the ﬁrst equation in (3.9 Electron in an external magnetic field 107 while. with the substitution of this.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. |p| ≪ m.184) where e denotes the charge of the particle and Aµ represents the four vector potential of the associated electromagnetic ﬁeld.

in this case.186) Let us next consider an electron interacting with a time independent external magnetic ﬁeld. (3. The Dirac equation for the positive energy electrons. E+m 2m (3.188) Explicitly.189). we obtain . (3. σ · (p − eA ) uL (p) = (E + m)uS (p). B = (∇ × A ) . m½ σ · (p − eA ) σ · (p − eA ) −m½ uL (p) uS (p) =E uL (p) uS (p) .189) leads to σ · (p − eA) σ · (p − eA) uL (p) ≃ uL (p). Substituting this back into the ﬁrst equation in (3. or. the second equation in (3.190) where in the last relation. takes the form (α · (p − eA ) + βm) u(p) = Eu(p). (3. we can write the two (2-component) equations as σ · (p − eA ) uS (p) = (E − m)uL (p).108 3 Properties of the Dirac equation pµ → i∂µ . we have used |p| ≪ m in the non-relativistic limit. (3. In this case. In this case.189) uS (p) = (3.185) the minimal coupling prescription also corresponds to deﬁning (in the coordinate representation) ∂µ → ∂µ + ieAµ .187) where we are assuming that A = A(x). we have A0 = 0 = φ.

1 e (p − eA)2 − σ · B uL (p) ≃ ENR uL (p). 2m or. Ak ] = −iǫijk [∇j . 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 .194) Consequently.191) Let us simplify the expression on the left hand side of (3.192) Note that (here.195) 2m 2m where we have identiﬁed (as before) .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) .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). when we can approximate the Dirac equation by that satisﬁed by the two component spinor uL (p). equation (3. We can use this in (3. Ak ] = −i (∇ × A )i = −i(B)i . in the non-relativistic limit. 2m (3. (3.3.191) takes the form 1 (p − eA)2 − eσ · B uL (p) = (E − m)uL (p). (3. (3.192) to write (σ · (p − eA)) (σ · (p − eA)) = (p − eA )2 − eσ · B.

this shows that a point Dirac particle has a magnetic moment corresponding to a gyro-magnetic ratio g = 2.196) We recognize (3. Namely. comparing with (3. for the proton and the neutron. Particles with a nontrivial structure (that is particles which are not point like and have extended structures). however. for example.197) Of course. Quantum mechanical corrections (higher order corrections) in an interacting theory such as quantum electrodynamics. can have g-factors quite diﬀerent from 2. 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. 2m µ= (3. a minimally coupled Dirac particle automatically leads. In this case.110 3 Properties of the Dirac equation ENR = E − m. however. to a magnetic dipole interaction (recall that in the non-relativistic theory. Thus.198) Let us recall that the magnetic moment of a particle is deﬁned to be (c = 1) e S. 2m µ=g (3. (3. one says that there is an anomalous contribution to the magnetic moment.199) 1 Since S = 2 σ for a two component electron. we know that the magnetic moments are given by .197) we obtain g = 2. (3. 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. in the non-relativistic limit.195) to be the Schr¨dinger equation for a charged o electron with a minimal coupling to an external vector ﬁeld along with a magnetic dipole interaction with the external magnetic ﬁeld.

2mP µnm = (3.10 Foldy-Wouthuysen transformation In the last two sections. For example.91 µnm . the relevant expansion parameter is |p| and m the method works quite well in the lowest order of expansion.201) with mP denoting the mass of the proton. 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.202) denotes the electromagnetic ﬁeld strength tensor and κ represents the anomalous magnetic moment of the particle. the electric dipole 1 moment becomes imaginary at order m2 (namely. This is commonly known as the Pauli coupling or the Pauli interaction. the Hamiltonian becomes non-Hermitian). at higher orders. This puzzling feature can be understood in a simple manner as follows. The process of eliminating the “small” . we have described how the non-relativistic limit of a Dirac theory can be taken in a simple manner.3. 3. as we have seen explicitly. However.10 Foldy-Wouthuysen transformation 111 µN µP = −1.203) eκ µν σ Fµν . (3.79 µnm .200) where the nuclear magneton is deﬁned to be |e| . (3. = 2. this method runs into diﬃculty. 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 . 2m (3. In the non-relativistic limit.

(3.206) is the source of the problem in taking the non-relativistic limit consistently.101)) H = α · p + βm = γ 0 (γ · p + m) .207) Let us next look for a unitary transformation that will diagonalize the Hamiltonian in (3.204) has the form T = ½ −A 0 ½ .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). Since the lack of unitarity in (3.100) as well as (1. 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. has the form (see (3.180)) σ·p .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. in the case of the free Dirac equation. for example. Thus.207). for example. such a transformation would .206) It is clear from the form of the matrix in (3. (3.112 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 . The matrix T that takes us to the two component “large” spinors in (3.204) where the matrix A. In this case. (3. E+m A= (3. This diﬃculty in taking a consistent non-relativistic limit 1 to any order in the expansion in m was successfully solved by Foldy and Wouthuysen and also independently by Tani which we describe below.

(3.83) or (1.212) .210) and using this we can simplify and write U (θ) = ∞ n=0 1 (2n)! |p|θ 2m γ · pθ 2m + 2n + 1 (2n + 1)! .211) It follows now that U † (θ) = cos which leads to |p|θ 2m − γ ·p sin |p| |p|θ 2m .209) From the properties of the gamma matrices in (1. (γ · p)2 = −p2 = −|p|2 . (3. 1 (3. (3. θ = θ(|p|. γ · pθ 2m 2n+1 = cos γ ·p sin |p| |p|θ 2m (3.10 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. m). avoid the problem with “large” and “small” spinors). we note that (γ)† = −γ. 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.91).

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 . Namely. the transformation (3. the free Dirac Hamiltonian (3.208) is indeed unitary.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. (3.208).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 . Under the unitary transformation (3.213) × cos = cos 2 γ·p sin |p| 2 |p|θ 2m |p|θ 2m γ ·p |p| sin2 = cos2 + sin2 |p|θ 2m = ½.

217) We see from (3.217) that the Hamiltonian is now diagonalized in the positive and the negative energy spaces. it is clear from (3. = 0. (3. θ. the two components of the transformed spinor u′ (p) 1 u′ (p) 2 u′ (p) = U (θ)u(p) = . we have |p|θ m m p2 + m2 |p|θ m |p| cos = . if we want the transformation to diagonalize the Hamiltonian. leads to the diagonalized Hamiltonian m2 p2 + m2 p2 p2 + m2 H ′ = γ0 + = γ0 p2 + m2 . |p|θ m m arctan |p| (3. (3. our discussion has been quite general and the parameter of the transformation. from (3. m |p| m . has been arbitrary. However. tan θ= − m sin |p| .216) which. or.219) .10 Foldy-Wouthuysen transformation 115 So far.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). As a result.214). (3. 1 1 (3.214) that we can choose the parameter of transformation to satisfy |p|θ m = |p|θ m |p| cos or.3. sin = p2 + m2 .215) In this case.

the transformation is known as the Cini-Touschek transformation and is obtained as follows. = 0. tan + |p| sin =− m . 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. θ=− m arctan |p| (3. for which the generalized Foldy-Wouthuysen transformation (3.214) will have the form . in this case.211) takes the form 1 arctan 2 |p| m γ·p sin |p| 1 arctan 2 |p| . namely. the ultrarelativistic limit |p| ≫ m. . or. In this case. the unitary transformation in (3.214) that if we choose the parameter of transformation to satisfy |p|θ m |p|θ m |p|θ m m cos or. We also note that with the parameter θ determined in (3. (3.222) p2 + m2 As a result. There is a second limit of the Dirac equation.116 3 Properties of the Dirac equation For |p| ≪ m.183) to 1 the lowest order and it can be expanded to any order in m without any problem. |p| m |p| . Let us note from (3.215).211) is also quite useful. m (3. the transformed Hamiltonian (3.220) UFW (θ) = cos + which has a natural non-relativistic expansion in powers of |p| .221) this would lead to |p|θ m |p|θ m |p| cos sin = = − p2 + m2 m . this leads to the non-relativistic equation in (3.

223) p2 + m2 γ·p= |p| p2 + m2 α · p.224) m which clearly has a natural expansion in powers of |p| (ultrarelativistic expansion). For example. the unitary transformation (3.11 Zitterbewegung The presence of negative energy solutions for the Dirac equation leads to various interesting consequences.211) has the form UCT (θ) = cos 1 arctan 2 m |p| − γ ·p sin |p| 1 arctan 2 m .207). where operators carry time dependence and states are time independent. (3. let us consider the free Dirac Hamiltonian (1.100) H = α · p + βm. the Heisenberg equations of motion take the forms ( = 1) . Therefore. |p| m which has a natural expansion in powers of |p| . 3. in this case.3.220) as transforming away the α · p term in the Hamiltonian (3. In fact. |p| (3.11 Zitterbewegung 117 H ′ = γ0 = γ0 p2 p2 + m2 + m2 p2 + m2 γ·p |p| (3.207) while the Cini-Touschek transformation rotates away the mass term βm from the Hamiltonian (3. we can think of the Foldy-Wouthuysen transformation (3.225) In the Heisenberg picture.

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

These peculiarities of the relativistic theory can be understood as follows. or. β = 2δij ½. (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).232) shows that the eigenvalues of α(t) are the same as those of α (only the eigenfunctions are transformed) and.102)) αi . we note that if α|ψ = λ|ψ . then.234) Here we have used the relations (see (1. α(t) eiHt |ψ where we have identiﬁed |ψ ′ = eiHt |ψ . = λ eiHt |ψ . This would seem to imply that the velocity of an electron is equal to the speed of light which is unacceptable even classically. = 0. α(t)|ψ ′ = λ |ψ ′ .235) + + . it follows that eiHt αe−iHt eiHt |ψ = λ eiHt |ψ .3.231) where λ denotes the eigenvalue of the velocity operator α. (3. are ±1. (3.233) Equation (3. H = −2iα(t)H + 2ip. We note from Heisenberg’s equations of motion that the time derivative of the velocity operator is given by 1 [α(t). therefore. (3.11 Zitterbewegung 119 eigenvalues of α(t) are ±1 as well. (3. More explicitly.232) or. since the electron is a massive particle. αj αi .

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

11 Zitterbewegung 121 velocity which is oscillating at a very high frequency (for an electron at rest. the energy is ≈ . In fact. + α(0) − H H (3.242) integrating this over time. it is easy to check that for a positive energy electron state E+m 2m u ˜ .) and gives a time dependence to α(t). This quivering motion of the electron was ﬁrst studied by Schr¨dinger o and is known as Zitterbewegung (“jittery motion”).243) are again what we will expect classically for uniform motion. Its occurrence is quite surprising.226) that since ˙ x(t) = α(t) = p −2iHt p e . the third term represents an additional contribution to the electron trajectory which is oscillatory with a very high frequency. H x(t) = a + α(0) − (3.243) where a is a constant.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 . for example. The ﬁrst two terms in (3. Let us also note from (3.3. since there is no potential whatsoever in the problem. However.239) and (3.5MeV corresponding to a frequency of the order of 1021 /sec. we obtain i p t+ H 2 p H −1 e−2iHt . σ·p ˜ E+mu u(p) = we have (3. The unconventional operator relations in (3.243) can be shown in the Schr¨dinger picture to arise from the presence of o negative energy solutions.

Schiﬀ. This also shows that the eigenstates of the velocity operator.) 3. 1980. 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. Relativistic Quantum Mechanics. C. New York. Drell.12 References 1. in a positive energy electron state p H α(t) + = + . McGrawHill. 2. Bjorken and S. (This makes clear that neglecting the negative energy solutions of the Dirac equation would lead to inconsistencies. .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. Zuber. This shows that even though the eigenvalues of the operator α(t) are ±1 corresponding to motion with the speed of light. 1968. New York. Itzykson and J-B. I. 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. 3.246) as we should expect from the Ehrenfest theorem.245) 2E 2p p × + E E+m E+m u† u = 0. New York. D. McGraw-Hill. Quantum Field Theory. ˜ ˜ This shows that even though the operator relations are unconventional. McGraw-Hill. Quantum Mechanics. J. α(t). L. (3. 1964.

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

This page intentionally left blank .

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

etc.5) R(ω) = = ω ij xi ∂j = −ω ij xj ∂i . (4. of course.35) here for clarity.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 .3) (4. ǫµν . but here we denote them by ω ij . we can write the inﬁnitesimal rotations also as δxi = −R(ω)xi = ω i j xj . (4. It follows now that i R(ω)xi = −ω kj xj ∂k xi = −ω kj xj δk = −ω ij xj = −ω i j xj . the form of the rotation that we had discussed in the last chapter. (4.) If we now identify (In the last chapter we had denoted the inﬁnitesimal transformation matrices by ǫi j . (4. ω µν in order to avoid confusion.) ω i j = ǫi jk αk . 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 (4.126 4 Representations of Lorentz and Poincar´ groups e notation from (1.34) and (1. ǫi jk = η iℓ ǫℓjk .7) .4) This is. ǫijk denotes the three dimensional Levi-Civita tensor with ǫ123 = 1.6) In other words. we note that ω ij = −ω ji . then.

R(ω) = ω ij xi ∂j . Such an algebra is called a non-Abelian (non-commutative) algebra.5) generates inﬁnitesimal rotations and the operators. Mij . namely. 2 (4. we can calculate them directly as . from the algebra of the vector operators in (4.8) = R(ω). rather. ω kℓ xk ∂ℓ = ω ij ω kℓ [xi ∂j . The Lie algebra of the group of rotations can be obtained from the algebra of the vector operators themselves.9) Namely. the vector operator. ∂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 (4.1 Symmetry algebras 127 Namely. Mij .8). R(ω) = 1 ij ω Mij . xk ∂ℓ ] = ω ij ω kℓ (xi [∂j . are known as the generators of inﬁnitesimal rotations. Alternatively. they give back a rotation.5). we note that R(ω). where we have identiﬁed ω ij = ω ik ω kj − ω ik ωkj = −ω ji = [ω. xk ] ∂ℓ + xk [xi . ω]ij .10) we can obtain the algebra satisﬁed by the generators of inﬁnitesimal rotations. (4.4. two rotations do not commute. R(ω) in (4. Thus. Using the form of R(ω) in (4.

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

2 (4.1 Symmetry algebras 129 1 p rs ǫ ǫ Mps 4 i r j ǫi q r ǫjsr Mqs 1 q sr ǫ ǫ Mqs 4 i r j (4.13) is. can be generated by the inﬁnitesimal vector operator R(ǫ) = −ǫµ Pµ = −ǫµ ∂µ .15) The algebra of the generators in (4.18) .4.14) which.1. then. Jj ] = ǫij r Jr .11) or (4. (4. In the same spirit. so that µ R(ǫ)xµ = −ǫν ∂ν xµ = −ǫν δν = −ǫµ . (4. of course.12)) 1 q ǫri ǫjsr Mqs = − ǫijr ǫrsq Mqs = −ǫijr Jr . namely.2 Translation. 4. 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). [Ji . = −ǫri q ǫjsr Mqs . let us note that a constant in- ﬁnitesimal space-time translation of the form δxµ = ǫµ .13) 1 + ǫis q ǫjrs Mqr + 4 1 1 = − ǫriq ǫjsr Mqs − 4 4 1 − ǫri q ǫjsr Mqs − 4 or. the Lie algebra of SO(3) or SU (2) (or the familiar algebra of angular momentum operators) up to a rescaling.17) (4. leads to (see (4.16) (4. ǫri q ǫjsr + ǫrsq ǫij r + ǫrjq ǫsi r = 0.

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 .24) then.18). ∂ν ] = 0. R(ǫ) = [ǫµ ∂µ . ν = 0. (4.130 and we can write 4 Representations of Lorentz and Poincar´ groups e δxµ = −R(ǫ)xµ . two translations commute and the corresponding relation for the generators is [Pµ . R(ω) = (4. as we have seen in (3. the inﬁnitesimal. translations form an Abelian (commuting) group while the three dimensional rotations form a non-Abelian group.20) In other words.22) where.1. (4. 1. ǫ ν ∂ν ] = ǫµ ǫ ν [∂µ .21) Namely.23) As in the case of rotations. 3. Pν ] = [∂µ . let us note that if we deﬁne an inﬁnitesimal vector operator 1 µν 1 ω Mµν = ω µν (xµ ∂ν − xν ∂µ ) 2 2 µν = ω xµ ∂ν = −ω µν xν ∂µ . (4. constant parameters of transformation satisfy ω µν = −ω νµ . As we have seen in the last chapter. µ.3 Lorentz transformation.19) The Lie algebra associated with translations is then obtained from R(ǫ). (4. ∂ν ] = 0. (4. 4. 2. we obtain .

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. ω]µν . Thus. and the generators of inﬁnitesimal boosts with (4. as the generators of the inﬁnitesimal transformations. 1 µν 1 ω Mµν . ω λρ Mλρ 2 2 R(ω).28) = = ω µν xµ ∂ν = R(ω).26) M0i = −Mi0 = x0 ∂i − xi ∂0 = Ki . Mµν = −Mνµ . We also note that we can identify the inﬁnitesimal generators of spatial rotations with (see (4. we have ω µν = −ω νµ = ω µλ ω λν − ω µ ω λν = [ω.4.29) . (4.25) Therefore.27) As before.12)) Mij = −Mji = xi ∂j − xj ∂i = ǫij k Jk . we can think of R(ω) as the vector operator generating inﬁnitesimal proper Lorentz transformations and the operators. where. λ (4. as in the case of rotations. R(ω) = = ω µν ω λρ [xµ ∂ν . (4. we can determine the group properties of the Lorentz transformations from the algebra of the vector operators generating the transformations.1 Symmetry algebras 131 µ R(ω)xµ = −ω λν xν ∂λ xµ = −ω λν xν δλ = −ω µν xν = −ω µν xν = −δxµ .

in this case. xλ ∂ρ ] + [xν ∂µ . therefore. 1). xλ ∂ρ ] − [xµ ∂ν . xλ ∂ρ − xρ ∂λ ] = [xµ ∂ν . This is what we had seen explicitly in (3.69) in connection with the discussion of covariance of the Dirac equation. (Namely.30) (up to a scaling). 2 (4. up to a scaling. Lorentz transformations (boosts) are non-compact unlike rotations in Euclidean space.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. Mλρ ] = [xµ ∂ν − xν ∂µ . xρ ∂λ ] − [xν ∂µ . three rotations and three boosts).109) coincides with (4. As we have seen these transformations correspond to rotations. (4. .) We end this section by pointing out that the algebra in (2. 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. the matrices σµν provide a representation for the generators of the Lorentz group. gives the Lie algebra associated with Lorentz transformations. 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 . it is n(n−1) ) 2 1 × 4 × (4 − 1) = 2 × 3 = 6. the Lie algebra of the generators is isomorphic to that of the group SO(4).30) This. However.31) coincides exactly with the six generators we have (namely. This implies that. therefore. In fact. in four dimensions and. we note that the number of generators for SO(4) which is (for SO(n). The algebra of the generators can also be calculated directly and has the form [Mµν .

if we deﬁne an inﬁnitesimal vector operator as R(ǫ. the general transformation of the coordinates takes the form δxµ = ǫµ + ω µν xν . (4. ω) . R(ǫ. in this case. The transformations in (4. we also consider inﬁnitesimal translations. ω µν denote respectively the parameters of inﬁnitesimal translation and Lorentz transformation.1 Symmetry algebras 133 4. (4. ω)xµ = − ǫν ∂ν + ω λν xν ∂λ xµ = −(ǫν δν + ω λν xν δλ ) = −(ǫµ + ω µν xν ) = −δxµ .34) The algebra of the vector operators for the Poincar´ transformae tions can also be easily calculated as R(ǫ. it generates inﬁnitesimal Poincar´ e transformations.32) where ǫµ . µ µ R(ǫ. ω).1.32) are known as the (inﬁnitesimal) Poincar´ transformations or the inhomoe geneous Lorentz transformations. (4. xρ ∂λ ] + ω µν ǫ λ [xν ∂µ . 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 ) . acting on the coordinates. Clearly. ∂λ ] + ω µν ω λρ [xν ∂µ . ǫ λ ∂λ + ω λρ xρ ∂λ = ǫµ ω λρ [∂µ .4 Poincar´ transformation. Namely.33) then. xρ ∂λ ] = ǫµ ω λρ ηµρ ∂λ + ω µν ǫ λ (−ηνλ ) ∂µ + ω µν ω λρ (ηµρ xν ∂λ − ηνλ xρ ∂µ ) = − (ω µν ǫ ν − ω µν ǫν ) ∂µ + ω µλ ωλν − ω µλ ωλ ν xµ ∂ν = − (ǫµ ∂µ + ω µν xν ∂µ ) = R (ǫ.4. in addition to inﬁnitesimal Lorentz e transformations. (4.35) . If. ω) = ǫµ ∂µ + ω µν xν ∂µ .

22) and (3. 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. Mλρ ] as well as [Pµ . see.36) We can also calculate the algebra of the generators of Poincar´ group. xν ∂λ − xλ ∂ν ] = ηµν ∂λ − ηµλ ∂ν = ηµν Pλ − ηµλ Pν . combining with our earlier results on the algebra of the translation group. (4. ω]µν .30). For change in the coordinate four vector under an inﬁnitesimal Lorentz transformation. (4. Mνλ ] = ηµν Pλ − ηµλ Pν .) Thus. for example.134 4 Representations of Lorentz and Poincar´ groups e where we have identiﬁed ǫµ = (ω µν ǫ ν − ω µν ǫν ) . e We already know the commutation relations [Mµν . Mλρ ] = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ .38) [Mµν . since the generae tors of translations do not commute with the generators of Lorentz . (4. We note that the algebra of translations deﬁnes an Abelian subalgebra of the Poincar´ algebra (4.21). (4.30) and (4. the only relation that needs to be calculated is the commutator between the generators of translation and Lorentz transformations. Pν ] (see (4. (This is seen by recalling that 2 ω µν Mµν generates inﬁnitesimal Lorentz transformations. Pν ] = 0. Pµ behaves 1 like a covariant four vector. (4. Therefore. we conclude that the Lie algebra associated with the Poincar´ e transformations (inhomogeneous Lorentz group) is given by [Pµ .21)). However.37) which simply shows that under a Lorentz transformation. Mνλ ] = [∂µ . ω µν = ω µλ ω λν − ω µλ ωλ ν = [ω. [Pµ .57).38). as well as the homogeneous Lorentz group. (4. Note that [Pµ .

Namely. 1). e (4. (4. Jj ] = M0i . [Ki .40) We recall from (4.4. − 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 . 3.2 Representations of the Lorentz group 135 transformations. (4. (4.39) Rather. the Lorentz algebra takes the form [Ki . Mλρ ] = (−ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ ).12).27) that we can identify the angular momentum and boost operators as 1 Ji = − ǫi jk Mjk . M0j ] = −Mij = −ǫij k Jk . i = 1. (The general convention is to denote groups by capital letters while the algebras are represented by lower case letters.41) Written out in terms of these generators. 2 Ki = M0i .30) [Mµν .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. Kj ] = [M0i .) 4. Jj ] = ǫij k Jk . (4. Poincar´ algebra cannot be written as a direct sum e of those for translations and Lorentz transformations. [Ji .26) and (4.42) . Kj ] . it is what is known as a semi-direct sum of the two algebras. 2. Poincar´ algebra = t4 ⊕ so(3.

if we deﬁne the generators with a factor of “i”.42) that we can assign the following hermiticity properties to the generators. the generators in the new basis are all anti-Hermitian.46) Namely. i † Bi = −Bi . as can be seen from the hermiticity of the generators in the coordinate basis.47) .45) This unconventional hermiticity for Ji arises because. in choosing the coordinate representation for the generators. namely. namely. Ki† = Ki . This is a set of coupled commutation relations.44) Parenthetically. (Ji − iKi ) . However. As a consequence of (4. Ji† = −Ji . the ﬁnite dimensional representations of boosts are non-unitary (hence the opposite Hermiticity of Ki ).13) in the last relation. (4. Ki . (compared to Ji ) is connected with the fact that such transformations are non-compact and. we have not been particularly careful about choosing Hermitian operators. (4. (4. (4. (4. Ki = −i (Ai − Bi ) .43) which also leads to the inverse relations Ji = (Ai + Bi ) . † Mµν = i(xµ ∂ν − xν ∂µ ) = Mµν . inﬁnite dimensional representations are unitary. The opposite hermiticity property of the generators of boosts.45). consequently. 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 ) . we have A† = −Ai . let us note from the form of the algebra in (4.136 4 Representations of Lorentz and Poincar´ groups e where we have used (4.

Kj ]) 4 1 ǫ k Jk + iǫij k Kk + iǫij k Kk + ǫij k Jk 4 ij 1 k ǫ (Jk + iKk ) = ǫij k Ak . Jj ] + i [Ji . (Jj − iKj ) 2 2 1 ([Ji . one says that the Lorentz algebra is isomorphic to the direct sum of two angular momentum algebras. Kj ]) 4 1 ǫ k Jk − iǫij k Kk + iǫij k Kk − ǫij k Jk = 4 ij = 0. so(3.42) takes the form 1 1 (Ji + iKi ) . as we have already seen in the last chapter.4. (Jj − iKj ) 2 2 [Ai . the algebra separates into two angular momentum algebras which are decoupled. Kj ] + i [Ki . in this new basis. Aj ] = = = = [Bi . the Lorentz algebra (4. Therefore. 2 ij 1 1 (Ji − iKi ) . (Jj + iKj ) 2 2 1 ([Ji .49) Incidentally. 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 . Mathematically. Jj ] − [Ki . Jj ] − i [Ji . Bj ] = = 1 ([Ji . the Lorentz group is double valued (doubly connected).48) In other words. Kj ] − i [Ki .43). (4. Jj ] − i [Ji . Jj ] + [Ki . Jj ] − [Ki . (4. Bj ] = = = = [Ai . Kj ] + i [Ki .2 Representations of the Lorentz group 137 In the new basis (4. 2 ij 1 1 (Ji + iKi ) . it is more . 1) ≃ so(3) ⊕ so(3) ≃ su(2) ⊕ su(2).

1.52) where the ﬁnite parameters of rotation and boost can be identiﬁed with θi = 1 i i θ + θB .jB ) (x). . (4. . much the same way we consider the universal covering group SU (2) of SO(3) to describe rotations. .35) in connection with the covariance of the Dirac equation. . can be speciﬁed uniquely once we know the values of jA and jB and is labelled as D(jA . 2 A (4. this represents the operator implementing ﬁnite transformations on the Hilbert space of states or wave functions as ψ ′ (jA . ) to describe the Lorentz transformations.jB) (Λ)ψ (jA . .138 4 Representations of Lorentz and Poincar´ groups e meaningful to consider the simply connected universal covering group of SO(3. . 2 2 (4. (4.) (jA . . Explicitly.j ) (j .50) An irreducible nonunitary representation of the homogeneous Lorentz group. 1.j ) −i θ i Ji A B +δi Ki A B D (Λ) = e = e . . 2 2 3 1 jB = 0. The ﬁnite dimensional unitary representations of each of the angular momentum algebras is well known. Denoting by jA and jB the eigenvalues of the Casimir operators A2 and B2 respectively for the two algebras. we have 3 1 jA = 0. Namely. . . . . we can write (This is the generalization of the S(Λ) matrix that we studied in (3.53) . 2 A δi = i i i θ − θB .jB ) (x′ ) = D(jA .51) where Λ represents the ﬁnite Lorentz transformation parameter.jB ) (just as the representation of the rotation group is denoted by D(j) ).jB ) “ ” (j ) (j ) i i −i θA Ai A +θB Bi B “ ” (j .jB ) (x) = D(jA ) (Λ)D(jB ) (Λ)ψ (jA . therefore. 1) which is SL(2.

These two representations are inequivalent and. |jA − jB | + 1. j = .54) and its spin content follows from the fact that (see (4. 2 ) . For jA = jB = 0.58) 1 corresponds to a two component spinor representation with spin 2 . (4. in fact. . jA + jB . we conclude that the values of the spin in a given representation characterized by (jA . For jA = jB = 2 . 2 1 1 D( 2 . D(0. . 1 Similarly.59) also corresponds to a two component spinor representation with spin 1 2 . (4. dimensionality (4. dimensionality = 2. jB = 2 . . For jA = 1 . from our knowledge of the addition of angular momenta. are complex conjugates of each other and can be identiﬁed to act on the wave functions of the two kinds of massless Dirac particles (Weyl 1 fermions) we had discussed in the last chapter.jB ) = (2jA + 1) (2jB + 1) . 2 (4.0) . . 2 (4. jB ) will have the dimensionality D (jA . dimensionality = 1. dimensionality = 2.4. . jB = 0.44)) J = A + B. jB ) can lie between j = |jA − jB |.0) .55) Consequently. j = 0. j = .2 Representations of the Lorentz group 139 Such a representation labelled by (jA . 1 1 D(0.57) corresponds to a scalar representation with zero spin (which acts on the wave function of a Klein-Gordon particle). (4. for jA = 0.56) The ﬁrst few low lying representations of the Lorentz group are as follows.

4. To compare with the results that we had derived earlier.48). 2 (4.140 4 Representations of Lorentz and Poincar´ groups e D( 2 . Let us now con- struct explicitly a few of the low order representations for the generators of the Lorentz group. 2 (1) ( 1 . 1 1 (4.0) . = 0. It actually corresponds to a reducible representation of the Lorentz group of the form 1 1 1 D(0.62) arises because ǫij = −ǫijk in (4.0) (0) = −i Ai 2 − Bi Ki 2 = i σi . dimensionality = 2 × 2 = 4. Ai . ··· . (Namely.61) This discussion can similarly be carried over to higher dimensional representations. Ki . 2 ) . j = 0. Bi by a factor of i. 2 (4. we scale all the generators Ji .2. 2 1 1 ( ) Bi 2 = − σi . 2 ) ⊗ D ( 2 .) From (4. 1.) Ai (0) = 0.50).47).44). we have (We note here that the negative sign in the spin 2 k representation in (4. namely.63) .60) is known as a four component vector representation and can be identiﬁed with a spin content of 0 and 1 for the components. we now consider Hermitian generators by letting Mµν → iMµν as in (4. (Note that a four vector such as xµ has a spin zero component.) It may be puzzling as to where the four component Dirac spinor ﬁts into this description. dimensionality = 4. 2 ··· . (4. Bi (0) 1 (1) Ai 2 = − σi .62) Using (4. t and a spin 1 component x (under rotations) and the same is true for any other four vector. this leads to the ﬁrst two nontrivial representations for the angular momentum and boost operators of the forms 1 ( 1 . we note that for the ﬁrst few low order represen1 tations.1 Similarity transformations and representations. j = .0) (1) (0) Ji 2 = Ai 2 + Bi = − σi .

from (4. Therefore. 2 (0. 1 ) (0) Ji 2 = Ai + Bi 2 = − σi . (4.67) However. (4. we note that these do not resemble the generators of the Lorentz algebra deﬁned in (3. 1 ) (1) (0) Ki 2 = −i Ai − Bi 2 i = − σi .65) would imply the existence of an invertible matrix S such that S −1 σi S = σi .64) we see that the condition (4. 2 (4.0) 2 ⊕ Ki (0.2 Representations of the Lorentz group 141 and 1 (1) (0. ( 1 .63) and (4.66) which is clearly impossible. we would say that the two representations 2 . 1 ) Ki 2 = S −1 Ki 2 S. 0 and 0.78)). .0) (0.64) give the two inequivalent representations of dimensionality 2 as we have noted earlier. From (4.63) and (4. For example. Two representations are said to be equivalent.0) = Ji 2 ⊕ Ji 2 = = Ki ( 1 . 2 are inequivalent representations.64).98) (or (3. 1 ) 2 = 1 − 2 σi Ji (Dirac) 0 1 − 2 σi 0 .71) and (3. 1 ) ( 1 . 0 and 0.4.63) and (4.69) and (2. (Dirac) Ki i 2 σi 0 0 i − 2 σi . if there exists a similarity transformation relating the two. S −1 σi S = −σi . (4. 1 ) ( 1 . 1 are 2 equivalent.0) Ji 2 = S −1 Ji 2 S.64) Equations (4.65) 1 then. They 2 provide the representations of angular momentum and boost for the left-handed and the right-handed Weyl particles. we can obtain the representation of the Lorentz generators for the reducible four component Dirac spinors as (0. the two representations la1 belled by 1 . if we can ﬁnd a similarity transformation S leading to (0. In fact.

67) and in our earlier discussion in (3. consequently.69) 2 µν and.142 4 Representations of Lorentz and Poincar´ groups e This puzzle can be understood as follows.69) and (2.78)) are equivalent since they are connected by a similarity transformation that . (4. we note that the angular momentum and boost operators in (4. 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). γW = ǫijk 2 W . we note that the generators in (4.70) = σi i − 2 σi 0 1 PD σ . we can apply the inverse similarity transformation in (2. 2 0i Therefore. ij σW = i i j γ .γ = 2 W W −iσi 0 σk 0 0 iσi 0 σk 0i σW = .68) As a result. i 0 i γ . give a representation of the Lorentz generators in the Weyl representation.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 .98) (see also (3. We note that in the Weyl representation for the gamma matrices deﬁned in (2. (4.119).67) are obtained from 1 W σ . 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 (4.71) and (3.

2 Representations of the Lorentz group 143 relates the Weyl representation of the Dirac matrices in (2. There is yet another interesting example which sheds light on similarity transformations between representation.4.4) the matrix structure of the generator J3 to be 0 0 i 0 0 0 J3 = −i 0 0 .72) Similarly. (3.4). we can determine a representation for the generators of the Lorentz transformations belonging to the representation for the four vectors. from the Lie algebra point of view the four vector representation corresponds to jA = jB = 1 (see (4. for example.71) we can immediately read out from (3. Let us consider a three dimensional inﬁnitesimal rotation of coordinates around the z-axis as described in (3. as we discussed earlier. we can deduce the matrix form of the corresponding generators to be . (Here we will use 3dimensional Euclidean notation without worrying about raising and lowering of the indices. Let us illustrate this for the simpler case of rotations. For example.119) to the standard Pauli-Dirac representation. 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.) Representing the inﬁnitesimal change in the coordinates as δxi = iǫ (J3 )ij xj . there must be a similarity transformation relating the two representations. from the inﬁnitesimal change in the coordinates under a Lorentz transformation (see.60)) and we can 2 construct the representations for J and K in this case as well from a knowledge of the addition of angular momenta. (4. The case for Lorentz transformations follows in a parallel manner. Surprisingly. On the other hand.18)). (4. therefore.

in fact. Das and A. let us note that the generators obtained (LA) from the Lie algebra are constructed by choosing the generator J3 See. (4. k = 1. To construct the similarity transformation (which actually is a unitary transformation).72) and (4. 2 0 1 0 1 0 0 = 0 0 0 . (4.73) and. 2 0 1 0 (4. This puzzle can be resolved by noting that there is a similarity transformation that connects the two representations and.73) in spite of the fact that they belong to the same representation for j = 1. Melissinos (Gordon and Breach). 3. Jj ] = iǫijk Jk . 2. therefore provide a representation for the generators of rotations. i. Quantum Mechanics: A Modern Introduction.75) (LA) J1 (LA) J2 (LA) J3 which look really diﬀerent from the generators in (4.73) that they satisfy [Ji . J2 = 0 0 0 i −i 0 0 0 0 . 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 . page 289 or Lectures on Quantum Mechanics. Das (Hindustan Book Agency.144 4 Representations of Lorentz and Poincar´ groups e It can be directly checked from the matrix structures in (4. therefore.74) J1 = 0 0 0 0 0 0 −i 0 i . page 182 (note there is a typo in the sign of the 23 element for L2 in this reference) 1 . This is. 0 0 −1 0 −1 0 i = √ 1 0 −1 . they are equivalent. C. the representation in the space of three vectors which would correspond to j = 1. New Delhi). j.72) and (4. On the other hand. A. A. for example.

2 ψ (j3 =1) = 0 ψ (j3 =−1) 1 −√ i2 = − √2 . (4.2 Representations of the Lorentz group 145 to be diagonal.76) which will diagonalize the matrix J3 .77) If we now deﬁne a similarity transformation (unitary) S −1 = U † . 1 √ 2 i − √ 2 0 0 −1 1 − √2 U= 0 i − √2 . 0 0 (4. Let us note from (4. 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 . S = U. −1 (4.4.78) then. 0 U† = 0 1 − √2 1 √ 2 i √ 2 0 i √ 2 −1 . 0 ψ (j3 =0) = 0 .76) 0 Let us construct a unitary matrix from the three eigenstates in (4.72) that the three normalized eigenstates of J3 have the forms 1 √ 2 i − √ .

. therefore. (4.146 4 Representations of Lorentz and Poincar´ groups e = S −1 J2 S 0 1 0 1 (LA) √ 1 0 1 = J1 . related by a similarity transformation and. √ 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 (4.72).73) and (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 .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 .79) This shows explicitly that the two representations for J corresponding to j = 1 in (4. are equivalent.

Mµν ] P λ = P λ (ηλµ Pν − ηλν Pµ ) + (ηλµ Pν − ηλν Pµ ) P λ = Pµ Pν − Pν Pµ + Pν Pµ − Pµ Pν = 0. P 2 . it is useful to study the representations of the Poincar´ group. Mµν = = = P λ [Pλ .80) deﬁnes a quadratic Casimir operator of the Poincar´ algebra (4.3 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. Pµ = 0. In order to determine the unitary representations.38) e since it commutes with all the ten generators. Mµν P λ Pλ . Pµ P 2 . Mµν ] + [Pλ . namely. Since Poincar´ e group is non-compact (like the Lorentz group). it is known that it has only inﬁnite dimensional unitary representations except for the trivial representation that is one dimensional. we seek to ﬁnd unitary representations in some inﬁnite dimensional Hilbert space where the generators Pµ . The operators Mµν generate inﬁnitesimal Lorentz transformations through commutation relations and the relation above.81) can be intuitively understood as follows. known as the Pauli-Lubanski operator. Mµν act as Hermitian operators. This e would help us in understanding the kinds of theories we can consider and the nature of the states they can have. Let us deﬁne a new vector operator. (4. which is supposed to characterize the inﬁnitesimal transformation of P 2 . Therefore. P λ Pλ . (4.4. 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.81) The last relation in (4. from the generators of the Poincar´ group as e .3 Unitary representations of the Poincar´ group e 147 4. let us note that the operator P 2 = η µν Pµ Pν = P µ Pµ .

148 4 Representations of Lorentz and Poincar´ groups e Wµ = 1 1 µνλρ ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . 2 2 (4. in general. 2 (4. we note that Pµ W µ = 1 µνλρ ǫ Pµ Pν Mλρ = 0.82) The commutator between Pµ and Mνλ introduces metric tensors (see (4. therefore.82) also as W µ = Pν M µν = M µν Pν . (4. It follows from (4. the vector Wµ is orthogonal to Pµ . Let us deﬁne the dual of the generators of Lorentz transformation as M µν = Mµν 1 µνλρ ǫ Mλρ .38)) which vanish when contracted with the anti-symmetric LeviCivita tensor.85) where the order of the operators is once again not important. we can write (4.83) implies that the PauliLubanski operator has only three independent components (both in the massive and massless cases). First. Let us next calculate the commutators between W µ and the ten generators of the Poincar´ group. this is not true for massless theories as we will see shortly.83) that. As a result. 2 (4. 2 1 = − ǫµνλρ M λρ . we have e .83) which follows from the fact that the generators of translation commute. the order of Pµ and Mνλ are irrelevant in the deﬁnition of the Pauli-Lubanski operator. (4. Furthermore. (However.84) With this.) In general.

Consequently. in particular Wµ W µ . We also note that Mµν . Pσ ] = = 1 µνλρ ǫ Pν Mλρ . Mστ = = 1 λρ ǫ Mλρ . (4.86) which follows from the the fact that momenta commute. Pσ 2 1 µνλρ ǫ Pν [Mλρ .3 Unitary representations of the Poincar´ group e 149 [W µ . any function of Wµ and. Pσ ] 2 1 µνλρ ǫ Pν (−ηλσ Pρ + ηρσ Pλ ) = 2 = 0. will also commute with the generators of translation. 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 .4. Mστ 2 µν 1 λρ ǫ [Mλρ .

88) Equation (4. Using this. Mστ ] = Pν M µν . 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.90) = −ǫµν λρ Wλ Pρ . Namely. Namely.150 4 Representations of Lorentz and Poincar´ groups e = −ηµσ Mντ − ηντ Mµσ + ηµτ Mνσ + ηνσ Mµτ .30)). 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 . (4. 1 1 λρσ ǫν Mρσ Pλ = ǫνλρσ [Wµ .87) simply says that under a Lorentz transformation. Mστ ] M µν + Pν M µν . ǫµνλρ ǫστ ζρ = −ηµσ (ηντ ηλζ − ηνζ ηλτ ) − ηµτ (ηνζ ηλσ − ηνσ ηλζ ) −ηµζ (ηνσ ηλτ − ηντ ηλσ ) . the operator Mµν behaves exactly like the generators of Lorentz transformation (see (4. Wν ] = = Wµ . we can now evaluate [W µ .89) In other words. it behaves like a second rank antisymmetric tensor under a Lorentz transformation. Mρσ ] Pλ 2 2 1 λρσ ǫ (ηµρ Wσ − ηµσ Wρ ) Pλ 2 ν (4. Mστ = [Pν . then. (4.87) Here we have used the identity satisﬁed by the four dimensional LeviCivita tensors. that [Wµ . for completeness as well as for later use. . (4. it transforms like a vector which we should expect since it has a free Lorentz index.

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

let us study this operator in some detail. (4. namely. P 2 |p = pµ pµ |p .96) Here m denotes the rest mass of the single particle state and we assume the rest mass to be non-zero. However. using (4. 2 2 (4. where p 2 = p µ p µ = m2 .95) (4. We recall that Wµ = 1 µνλρ 1 ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . the eigenvalues of W 2 are not so obvious. the eigenvalues of the operator P 2 = P µ Pµ are obvious.152 4 Representations of Lorentz and Poincar´ groups e Pµ |p = pµ |p .94) and in this basis. 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 . (4. Therefore.97) Therefore.88).

0) . let us go to the rest frame of the massive particle. 0.102) can also be derived in an alternative manner which is simpler and quite instructive.100) = − m2 M ij Mij . pµ = (m.4.98) where we have simpliﬁed terms in the intermediate steps using the anti-symmetry of the Lorentz generators.101) where Jk represents the total angular momentum of the particle.82) has the form . To understand the meaning of this operator. 0. 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. 2 Recalling that (see (4. p 2 = p µ p µ = m2 .99). the Pauli-Lubanski operator (4. 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 .99) and the operator W 2 acting on such a state.26)) Mij = ǫij k Jk . 2 (4. (4. (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 .102) The result in (4. Let us note that in the rest frame (4. In this frame.

It is not hard to see that all possible 3dimensional rotations would leave such a vector invariant. 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).34) as well as (4.102). Namely.12).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. (m. this simply measures the spin of the particle. we can think of W 2 as being proportional to J 2 and in the rest frame of the particle. (4.103) where we have used (1. for a massive particle at rest. It follows now that W 2 = W µ Wµ = W i Wi = m2 J i Ji = −m2 J2 . 0). we see that the representations with p2 = 0 can be labelled by the eigenvalues of the Casimir operators. 2 (4.104) which is the result obtained in (4. therefore. we can ask what Lorentz transformations would leave such a vector invariant. 0. Clearly. That is. Technically. we ﬁnd eigenvalues of W 2 : w2 = −m2 s(s + 1). 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. all the degenerate states can be labelled not . these would deﬁne an invariant subgroup of the Lorentz group and will lead to the degeneracy of states. 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. Therefore. For a state at rest with momentum of the form pµ = (m. for a massive particle.105) Thus. Therefore. 2 m 1 iνλρ ǫ Mλρ pν = − ǫijk Mjk = 2 2 m jk = − ǫi Mjk = mJi . (4. 0. The dimensionality of the representation can also be understood in an alternative manner as follows. s).

This deﬁnes the 2s + 1 dimensional representation for a massive particle of spin s. 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.103) that.107) Consequently. (4. pµ = (p. Mij ] = 0. Consequently. The basic reason behind this is that the “little” group of a light-like vector is not so obvious. which generate 3-dimensional rotations and are related to the angular momentum operators. a state at rest is an eigenstate of the P0 operator. let us note from (4. 4. Namely.83)) P 2 = 0.30)) [P0 . −s + 1. (4. · · · . up to a normalization factor. the three nontrivial Pauli-Lubanski operators correspond to the generators of symmetry of the “little group” in the rest frame. from (4. −s. we would have (see (4. From the Lorentz algebra.2 Massless representation. In contrast to the massive representa- tions of the Poincar´ group. we note that (see (4. This can also be seen algebraically.95)) pµ pµ = 0. commute with P0 .106) Namely. In closing.3. s = 0. In this case. acting on states in such a vector space. the representations for a massless partie cle are slightly more involved.83) we see that our states in the representation should also satisfy Pµ W µ = 0. s − 1. (4. 1 . namely.4.3 Unitary representations of the Poincar´ group e 155 just by the eigenvalue of the momentum. the operators Mij .108) However. −p). 1. we note that (see (4. · · · and 2 ms = s. 0.109) . p = 0. but also by the eigenvalues of three dimensional rotations. 0. (4.

for a massless particle if W 2 = 0. the simultaneous eigenstates of P 2 and W 2 would . 2 (4. 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. we can easily show that the action of W µ in such a space is proportional to that of the momentum vector. (4. Such representations do not correspond to physical particles and. in the second case where W 2 = 0 on the states of the representation.110) In the ﬁrst case.113) W0 = Comparing with (4. let us recall that Wµ = 1 µνλρ ǫ Pν Mλρ . consequently. p0 h= (4.114) This is nothing other than the helicity operator (since L · p = 0) and. W 2 = W µ Wµ = 0. namely. On the other hand. W µ has the form W µ = −hpµ . To determine h.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. therefore. it would lead to (see (4. namely.111) we conclude that in this space J·p . 2 2 (4.88)) 1 0ijk 1 ǫ pi Mjk = ǫijk pi ǫjkℓ Jℓ = −p · J = −J · p. W 2 = W µ Wµ = 0.112) from which it follows that acting on a general momentum basis state |p (not necessarily restricting to massless states). or. (4.111) where h represents a proportionality factor (operator). acting on states in such a space. namely.

2 1 3νλρ ǫ Mλρ pν = −pM12 = W 0 .82) takes the form 1 0ijk 1 ǫ Mjk Pi = ǫijk Mjk pi = −pM12 . indeed represents the helicity operator up to a normalization as we had noted in (4. 0. −p) and we would like to determine the “little” group of symmetries associated with such a vector. W 0 . the set of Lorentz transformations which would leave this four vector invariant must include rotations around the z-axis. It follows now from (4.115) that the Pauli-Lubanski operator indeed has only three independent components because of the transversality condition (4.4.3 Unitary representations of the Poincar´ group e 157 correspond to the eigenstates of momentum and helicity. (4. 0.115) 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 . We recognize that in this case.83). as we had pointed out earlier. the Pauli-Lubanski operator (4. (P0 − P3 ) |p = 2p|p .115) that. 2 2 1 1νλρ ǫ Mλρ pν = p (M02 − M23 ) . namely. but also algebraically by recognizing that a light-like vector of the form being considered is an eigenstate of the operator P0 − P3 . in the massless case.103) and (4.116) Let us now determine the dimensionality of the massless representations algebraically.113). For completeness. 2 1 2νλρ ǫ Mλρ pν = −p (M01 − M13 ) .115) We see from both (4. We also note from (4.107). This can be seen intuitively from the fact that the motion of the particle is along the z axis.117) . Let us recall that we are considering a massless state with momentum of the form pµ = (p. (4. let us note here that in the light-like frame (4. 2 W0 = W1 = W2 = W3 = (4.

Π2 have the same commutation relations as those of translations in this two dimensional space.120) and.115). but can be easily understood as follows. This may seem puzzling. the generators of symmetry of the “little” group. we see that e [P0 − P3 . it is isomorphic to the algebra of the Euclidean group in two dimensions. We note from (4. Π1 ] = [M12 . Furthermore. the three independent Pauli-Lubanski operators are. Π1 ] |p [P0 − P3 . (4. in fact. [M12 . 3 in (4.158 4 Representations of Lorentz and Poincar´ groups e Furthermore. M01 − M13 ] |p = 2P1 |p = 0. as we had also seen in the massive case. (4. M12 is the generator of rotations around the z axis or in the two dimensional plane and Π1 . M12 ] = 0. M02 − M23 ] |p = 2P2 |p = 0. We note that the algebra of the symmetry generators takes the form [M12 . To determine the other symmetries of a light-like vector.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. therefore. [P0 − P3 . = [P0 − P3 .118) so that rotations around the z-axis deﬁne a symmetry of the lightlike vector (state) that we are considering. comparing with W i . M02 − M23 ] = 0. i = 1. M01 − M13 ] = Π2 .121) Namely. namely. Π2 ] = [M01 − M13 .90) that in the . we say that the stability group or the “little” group of a light-like vector is E2 . p W1 . (4. Clearly. (4. let us deﬁne two new operators as W2 .38). 2. Π2 ] |p = [P0 − P3 . from the Poincar´ algebra in (4. [Π1 . we see that up to a normalization. M02 − M23 ] = −Π1 . also deﬁne symmetries of light-like states. Thus. Π2 ] = [M12 . These represent all the symmetries of the light-like vector (state). E2 (which consists of translations and rotation).

123) that (h corresponds to the helicity quantum number) (Π1 ± iΠ2 ) |p. As a result. massless theories with nontrivial spin that are parity invariant would have two dimensional representations corresponding to the highest and the . what we have been investigating within the context of “little” groups.124) (Alternatively. we see that in the space of light-like momentum states W 2 ∝ Π2 . follows from (4. such a state would correspond to the highest or the lowest helicity state. if W 2 = 0 in this space of states. As a result.116). we note from (4.) This corresponds to the one dimensional representation of E2 known as the “degenerate” representation. 1 2 (4. (4. Let us note from (4. the Pauli-Lubanski operators satisfy an algebra and.146)). h = 0 = Π2 |p. Furthermore. if our theory is also invariant under parity (or space reﬂection). as we have already pointed out. This is. therefore.111). they generate transformations which will leave the momentum basis states invariant. their eigenvalues can take any value.86) as the transformations that leave pµ invariant. Clearly. Π2 correspond to generators of “translation”. (4. The meaning of the transformations. h and this is the reason for the assertion in (4. see (3. [M12 .123) and comparing with (4.3 Unitary representations of the Poincar´ group e 159 momentum basis states (where pρ is a number). if W 2 = 0 in this space. Π1 ∓ iΠ2 ] = ±i (Π1 ∓ iΠ2 ) . On the other hand. Namely.122) Let us also note for completeness that the Casimir of the E2 algebra is given by Π2 = Π2 + Π2 = (Π1 ∓ iΠ2 )(Π1 ± iΠ2 ). of course. h = 0. can be thought of as generators of some transformations. namely. the space of physical states would also include the state with the opposite helicity (recall that helicity changes sign under space reﬂection.122) that spin can take an inﬁnite number of values which. Since Π1 . then. does not correspond to any physical system. then it follows from (4. we can say that Π1 |p.4.121) that Π1 ∓ iΠ2 correspond respectively to raising and lowering operators for M12 .

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

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

We also assume that the ﬁeld falls oﬀ rapidly at spatial inﬁnity (this is necessary for the theory to be well deﬁned). The classical. let us look at the simplest of ﬁeld theories.40)). On the other hand. one reverts back to the Hamiltonian formalism for its resolution.1: Oscillating string ﬁxed at both ends. lim φ(x) → 0. namely. therefore. namely. (5. t) is real and that it behaves like a scalar under a Lorentz transformation. However. has the form + m2 φ(x) = ∂µ ∂ µ + m2 φ(x) = 0.3) |x|→∞ The derivatives of the ﬁeld are also assumed to fall oﬀ at inﬁnity. e the quantization of the ﬁelds as operators can then be carried out in the Hamiltonian formalism. so far. φ (x ) = φ(x).162 5 Free Klein-Gordon field theory Figure 5. the dynamical equation involving the ﬁeld variables must be covariant under Poincar´ transformations. free Klein-Gordon equation with mass m. Given a relativistic dynamical equation. (5. whenever any ambiguity arises in the quantization. each ﬁeld must belong to a speciﬁc representation of the Lorentz group. covariant theory for fundamental physical processes and. it is generally preferrable to work in the Lagrangian formulation which is manifestly covariant. Secondly.4) . With this in mind. (5. φ(x) = φ∗ (x). We would like to have a relativistic. as we have seen (see (1.2) where φ (x ) denotes the Lorentz transformed ﬁeld. since the Hamiltonian is not a manifestly covariant concept. the classical free Klein-Gordon theory where we assume that the ﬁeld variable φ(x) ≡ φ(x.

P 2 denotes one of the Casimirs of the Poincar´ algebra and m is a constant. However. 0) and φ(x. To that end. of course.6) Furthermore.5.5) where we have written the Lagrangian in terms of a Lagrangian density as L= d3 x L. (5. (5. namely. Namely. for equations which are at most second order in the derivatives. ∂µ φ(x)) .7) In general.2 Lagrangian density 163 This equation is clearly invariant under Lorentz transformations. we assume that the Lagrangian density depends only on the ﬁeld variables and their ﬁrst derivatives. a Lagrangian density can depend on higher order derivatives. we follow exactly what we normally do in point particle mechanics. However. As we have seen. 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. (Note that the operator in the parenthesis in the momentum space corresponds to (P 2 − m2 ). The speciﬁcation of the initial values would appear to single out a preferred hyper-surface which can possibly destroy covariance under a Lorentz transformation.) 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. the Lagrangian density can depend at the . (5. 0) (a dot denotes a derivative with respect to t). 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). 5.2 Lagrangian density The next question that we would like to ask is whether there exists a Lagrangian or an action which would lead to the Klein-Gordon equation under a minimum action principle.

tf ) = 0. We have .3)) to simplify the surface term. These are the kinds of equations we will be interested in and correspondingly we will assume this dependence of the Lagrangian density on the ﬁeld variables through out. We 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).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) . ti ) = δφ (x. ti (5.10) Here we have used Gauss’ theorem (as well as the vanishing of ﬁeld variables asymptotically (5. Note that under an inﬁnitesimal change tf (5. subject to the boundary conditions (5.164 5 Free Klein-Gordon field theory most on the ﬁrst order derivatives of the ﬁeld variables.

δφ (x. (5. (5.10) – vanishes because of the boundary conditions (5. (5.15) .11) We note that the “surface term” – the last term in the expression (5. ti ) = 0 = δφ (x.13) Therefore.14) This is known as the Euler-Lagrange equation associated with the action S or the Lagrangian density L(φ.5. ∂µ φ(x)). Thus. for the Klein-Gordon equation. inﬁnitesimal change in the ﬁeld variable. we conclude that under an arbitrary. tf ) . As in point particle mechanics.9). ∂L ∂L − ∂µ = 0. δS = 0 only if). 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. namely. (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).12) Thus. we can think of the dynamical equations as arising from the minimum principle associated with a given action. ∂φ(x) ∂∂µ φ(x) (5. ∂µ ∂ µ φ + m2 φ = 0.

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. real scalar ﬁeld as its Euler Lagrange equation. 2 2 L= then (5. . Of course. 2 2 L= (5. ∂φ(x) ∂∂µ φ(x) or. Note that the Lagrangian density L in (5.17) ∂L ∂L − ∂µ = −m2 φ(x) − ∂µ ∂ µ φ(x) = 0. in this case. ∂∂µ φ(x) Therefore. ∂µ ∂ µ + m2 φ(x) = 0. Note that if we choose 1 m2 2 ∂µ φ∂ µ φ − φ .19) gives the Klein-Gordon equation for a free. This would simply correspond to adding a zero point energy. we see that the Lagrangian density 1 m2 2 ∂µ φ∂ µ φ − φ . (5. the action does not change either.18) In other words. the Euler-Lagrange equation. adding a total divergence to L also does not change the equations of motion. ∂φ(x) ∂L = ∂ µ φ(x).16) ∂L = −m2 φ(x). It is worth noting here that one can add a constant to L without changing the equation of motion. gives (5.3)).19) is manifestly Lorentz invariant. Similarly. in this case. if we assume the ﬁelds to fall oﬀ rapidly at inﬁnite separation (see (5.

in turn. p) = px − L (x. on the other hand. lead to the dynamical equations in the Hamiltonian form as x = {x. ˙ ˙ (5.21) and this leads to the deﬁnition of the Hamiltonian in the form (this is an example of a Legendre transformation) H(x.20) The conjugate momentum associated with the dynamical variable x is deﬁned as ∂L . ˙ (5. p} = 1. x) .23) These. H} . the basic variables are φ(x) and ∂µ φ(x) and we have from (5. {x. (5. ˙ p = {p. In the case of a point particle. L = L (x.5. x} = 0 = {p.19) (xµ in this case merely labels space-time coordinates) . ∂x ˙ p= (5.3 Quantization 167 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. the Lagrangian is a function of the coordinate and the velocity of the particle (a dot denotes a derivative with respect to t).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. x) . p} . H} .24) In the case of the classical scalar (Klein-Gordon) ﬁeld theory. ˙ (5.

H} . and a Hamiltonian d3 x H.25) In an analogous manner to the point particle mechanics in (5. we deﬁne a momentum canonically conjugate to the ﬁeld variable φ(x) as ∂L . the KleinGordon equation in this case) can be written in the Hamiltonian form as ˙ φ(x) = {φ(x). in turn.23)) {φ(x).26) which will. Π(y)}x0 =y0 = 0 = {Π(x).27) H= (5. ∂µ φ(x)) . ˙ ∂ φ(x) Π(x) = (5. (5. we can show that the dynamical equation (namely. φ(y)}x0 =y0 {φ(x). = δ3 (x − y). ˙ Π(x) = {Π(x). Π(y)}x0 =y0 .21).168 5 Free Klein-Gordon field theory L = L (φ(x). H} . Note that the Lagrangian density in this case is given by . (5.29) then. (5.28) If we further assume the equal time canonical Poisson bracket relations between the ﬁeld variable φ(x) and the conjugate momentum Π(x) to be (see (5. allow us to deﬁne a Hamiltonian density as ˙ H = Π(x)φ(x) − L. (5.30) Let us examine all of this in detail for the case of the free KleinGordon theory.

φ − ∇φ · ∇φ − 2 2 2 (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). Π(y)}x0 =y0 we obtain = 0 = {Π(x).34) With the equal time canonical Poisson bracket relations deﬁned to be (see (5. 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) .3 Quantization 169 L = = m2 2 1 ∂µ φ∂ µ φ − φ 2 2 1 ˙2 1 m2 2 φ . (5. = δ3 (x − y). we can identify the conjugate momentum with ∂L ˙ = φ(x).31) Therefore. Π(y)}x0 =y0 .32) Consequently. 2 2 2 (5.35) . φ(y)}x0 =y0 {φ(x).29)) {φ(x). ˙ ∂ φ(x) Π(x) = (5.5.

Furthermore. x0 · ∇y −δ3 (x − y) +m2 φ y. H} = Π(x). ˙ Π(x) = ∇ · ∇φ(x) − m2 φ(x) = ∇ 2 φ(x) − m2 φ(x). the Hamiltonian equations (5. Π2 (y) d3 y Π(y) {φ(x).36) where we have used the fact that. lead to ˙ φ(x) = Π(x). (5. H} = = = = φ(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). x0 = ∇ · ∇φ(x) − m2 φ(x).38) . we also have ˙ Π(x) = {Π(x).29). −δ3 (x − y) (5. since the Hamiltonian is time independent (this can be checked easily using the equations of motion and will also be shown in (5. Π(y)} x0 =y 0 d3 y Π y.170 5 Free Klein-Gordon field theory ˙ φ(x) = {φ(x).37).73)).37) Thus. ∇y φ(y)} + m2 φ(y) {Π(x). d3 y 1 m2 2 1 2 Π (y) + ∇y φ · ∇y φ + φ (y) 2 2 2 x0 =y 0 = d3 y ∇y φ(y) ·{Π(x). (5.36) and (5. 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. in this case. x0 δ3 (x − y) = Π(x). φ(y)} x0 =y0 = d3 y ∇y φ y.

we are supposed to treat φ(x) and Π(x) as Hermitian operators (because we are dealing with real ﬁelds. we can expand the ﬁeld operator φ(x) in such a basis. the quantization of such a system is quite straightforward.39) or. brings out the classical Hamiltonian description of the classical KleinGordon ﬁeld theory. ¨ φ − ∇ 2 φ(x) + m2 φ(x) = 0. However.) 5. (Namely. = iδ3 (x − y).5. φ(y)]x0 =y0 [φ(x). we quantize the “wavefunction”.4 Field decomposition 171 These are ﬁrst order equations as Hamiltonian equations should be and from these.2)) satisfying the equal time commutation relations (remember = 1) [φ(x). Π(y)]x0 =y0 . (5. or. In the quantum theory. therefore.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. the free Klein-Gordon equation (1. This is conventionally referred to as second quantization. This is. ∂µ ∂ µ + m2 φ(x) = 0. but not the wave function. we obtain the second order equation ¨ ˙ φ(x) = Π(x) = ∇ 2 φ(x) − m2 φ(x). we can expand the ﬁeld operator in this basis as . In the second quantization.40) Note that this is the same Klein-Gordon equation as we had studied earlier in chapter 1. see (5. therefore. Π(y)]x0 =y0 = 0 = [Π(x). We already know that the plane wave solutions (1.40) (which is a second order Euler-Lagrange equation) and this discussion. (5.41) of the KleinGordon equation deﬁne a complete basis and. Once we have the classical Hamiltonian description of a physical system. here we are treating φ(x) as an operator which is quantized and not as a wavefunction as we had done earlier. of course.

41) where we recognize that the operators φ(x) and φ(k) are Fourier transforms of each other (up to a multiplicative factor). We also emphasize here that the hermiticity properties of a given ﬁeld and its Fourier transform are.42) using the ﬁeld expansion (5.172 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 d4 k e−ik·x φ(k). 1 (2π) 3 2 3 2 ∂µ ∂ µ + m2 d4 k e−ik·x φ(k) = 0. (5. we obtain 1 (2π) or. Namely.45) (5. This shows that unless k2 = m2 .44) where the operator a(k) is no longer constrained by the equation of motion. we obtain 1 (2π) 3 2 φ(x) = d4 k δ k2 − m2 e−ik·x a(k). As a result.46) .) Since φ(x) satisﬁes the Klein-Gordon equation ∂µ ∂ µ + m2 φ(x) = 0. (5. (5.43) d4 k −k2 + m2 φ(k)e−ik·x = 0. in general. the Fourier transform has non-vanishing contribution (support) only on the massshell of the particle deﬁned by k2 = (k0 )2 − k2 = m2 . (Recall that 1 k · x = k0 x0 − k · x and the factor (2π)3/2 is introduced for later convenience. (5. φ(k) = 0.41). diﬀerent. Substituting this back into the expansion (5.41) in the Klein-Gordon equation. (5. we can denote φ(k) = δ k2 − m2 a(k).

2: Mass shell on which φ(k) has support.48) Using this relation in (5.47) Correspondingly. or. (k0 )2 = k2 + m2 .4 Field decomposition k0 173 k2 = m2. or. We note that the argument of the delta function vanishes for k 2 = m2 . k0 < 0 Figure 5. k0 > 0 k k 2 = m 2 . we can write δ k 2 − m2 2 = δ (k0 )2 − Ek = = 1 δ k0 − Ek + δ k0 + Ek 2|k0 | 1 δ k0 − Ek + δ k0 + Ek 2Ek . k0 = ± k2 + m2 ≡ ±Ek . the ﬁeld expansion becomes . (5.46).5. (5.

52) φ(x) = d3 k −ik·x e a(k) + eik·x a† (k) . k) 2Ek 0 +ik·x + eiEk x a(−Ek .49) and identifying k0 = Ek > 0. 2k0 (5. or. Therefore.51) Comparing the left-hand side and the right-hand side of (5.50) Let us note that we are dealing with a Hermitian ﬁeld (see (5. so that we can write 1 (2π) 3 2 a† (−k) = a(k). (5. 2k0 (5.53) . we obtain a† (k) = a(−k).51).2)). k) 0 x0 +ik·x = 1 (2π) 3 2 d3 k 1 0 e−iEk x +ik·x a(Ek . 2k0 (5. 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) . φ† (x) = φ(x). k) .49) Changing k → −k in the second term in (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 . we obtain 1 (2π) 2 3 φ(x) = d3 k −ik·x e a(k) + eik·x a(−k) . (5.

In fact. 2k0 (5.5 Creation and annihilation operators Given the ﬁeld expansion (5.56) .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.58) 5.53) as d3 k (2π)3 2k0 φ(x) = e−ik·x a(k) + eik·x a† (k) . a(k) = √ 2k0 a† (k) a† (k) = √ . Thus. (5. φ(x) = φ(+) (x) + φ(−) (x). eik·x a† (k).5 Creation and annihilation operators 175 Here we are supposed to understand that k0 = Ek = k2 + m2 > 0. (5.55) and in terms of these operators we can write the ﬁeld operator (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).57) so that we can write the ﬁeld operator as a sum of its positive and negative energy parts. (5.54) It is clear from (5. conventionally one deﬁnes a(k) .5. (5.54) that a(k) and a† (k) are really functions of the three momentum k alone.

61) ← → where ∂t is deﬁned in (1.176 5 Free Klein-Gordon field theory φ(x) = d3 k (2π)3 2k0 e−ik·x a(k) + eik·x a† (k) .59) and (5. k0 is not an independent variable).60) 2(2π)3 The two deﬁning relations in (5.61). Namely. d3 x e−ik·x k0 φ(x) − iΠ(x) ← → d3 x e−ik·x ∂t φ(x).59) we obtain the expansion for the conjugate momentum in (5. the expressions on the righthand side must also be time independent. It is important to emphasize here that in the expressions in (5. For example.61) we obtain (neglecting the overall multiplicative factors) . it is assumed that k0 = Ek (namely. taking the time derivative of the ﬁrst relation in (5. (5. (5. It is clear from (5.61) that since the left-hand sides are independent of time. we note from these relations that we can write the operators a(k) and ˙ a† (k) in terms of φ(x) and Π(x) as (recall that Π(x) = φ(x)) a(k) = = a† (k) = = − 1 (2π)3 2k0 i (2π)3 2k0 1 (2π)3 2k0 i (2π)3 2k0 d3 x eik·x k0 φ(x) + iΠ(x) ← → d3 x eik·x ∂t φ(x). (5.60) are invertible. This can also be checked explicitly.56).32) to be ˙ Π(x) = φ(x) = −i d3 k k0 e−ik·x a(k) − eik·x a† (k) .

40)) we obtain d3 k (2π)3 2k0 d3 k ′ (2π)3 2k′ 0 ′ [φ(x).62) Here. Imposing now the quantization relations between φ(x) and Π(x) (see (5. integrated the space derivatives by parts assuming that the surface terms vanish (see (5. Similarly. Π(y)]x0 =y0 = − ′ ′ ′ x0 =y 0 (5. [Π(x). φ(y)]x0 =y0 = ′ × e−ik·x−ik ·y a(k). This shows explicitly that the integrals in (5. the time independence of the second expression in (5. a† (k′ ) . a(k′ ) + e−ik·x+ik ·y a(k). a(k′ ) − e−ik·x+ik ·y a(k). in the intermediate steps.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) = 0.63) d kd k 3 3 ′ √ k0 k′ 0 2(2π)3 ′ × e−ik·x−ik ·y a(k). a† (k′ ) + eik·x−ik ·y a† (k). (5. a(k′ ) + eik·x+ik ·y a† (k).3)) and have √ made the identiﬁcation k0 = Ek = k2 + m2 . a† (k′ ) = 0.5. we have used the fact that φ(x) satisﬁes the Klein-Gordon equation. therefore. must be time independent).61) are indeed time independent.61) can also be derived in a simple manner (or note that the second relation is the Hermitian conjugate of the ﬁrst and.

65) From these. a† (k′ ) . k′ 0 φ(y) − iΠ(y) x0 =y 0 . it appears that there is an inﬁnite number of such operators in the present case – one for every value of the momentum k. which are the coeﬃcients of expansion of the ﬁeld operator in a plane wave basis. Thus. a(k). a† (k′ ) = δ3 k − k′ .178 ′ 5 Free Klein-Gordon field theory − eik·x−ik ·y a† (k). we can deduce the fundamental commutation relations between the coeﬃcients of expansion to be (This nice form arises because of the redeﬁnition (5. The commutation relations in (5. a† (k′ ) + eik·x−ik ·y a† (k). a† (k′ ) = 0. we have a(k). have commutation relations analogous to the annihilation and creation operators of a harmonic oscillator. Any other redeﬁnition will introduce a multiplicative factor into the commutation relations. recalling that the integrals are ime independent. a† (k′ ) = ′ d3 xd3 y √ eik·x−ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x). for example. a(k′ ) − e−ik·x+ik ·y a(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). ′ ′ x0 =y 0 (5. [φ(x). a† (k′ ) = iδ3 (x − y). (5.66) This shows that the operators a(k) and a† (k). a(k′ ) = 0 = a† (k). a(k′ ) − eik·x+ik ·y a† (k).55). However.66) can also be obtained more directly from the inversion formulae in (5.) a(k).61). Π(y)]x0 =y0 = − ′ ′ x0 =y 0 (5.

a(k′ ) = d3 xd3 y ′ √ eik·x+ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x). we have k′ 0 = k0 .5 Creation and annihilation operators 179 = d3 xd3 y ′ √ eik·x−ik ·y (2π)3 4k0 k′ 0 × −ik0 [φ(x).67) where we have used the fact that for k′ = k. a† (k′ ) = ′ d3 xd3 y √ e−ik·x−ik ·y 3 4k 0 k ′ 0 (2π) . k′ 0 φ(y) + iΠ(y) = d3 xd3 y ′ √ eik·x+ik ·y 3 4k 0 k ′ 0 (2π) × ik0 [φ(x). Π(y)] + ik′ 0 [Π(x). 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 0 k′ 0 4k (5.5.66) can also be derived in the same manner a(k). Π(y)] + ik′ 0 [Π(x). e 4k0 k′ 0 a† (k). The other two commutation relations in (5. 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′ ).

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

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 Creation and annihilation operators 181 k′ 0 = (k′ )2 + m2 = k2 + m2 = k0 . (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′ ) .70) Similarly. d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ (5.

71) and (5.72) into the expression for the Hamiltonian (5. we obtain H = = 1 2 d3 x H d3 x Π2 (x) + ∇φ(x) · ∇φ(x) + m2 φ2 (x) 1 4 d3 k k0 e−2ik 0 x0 = − a(k)a(−k) − a(k)a† (k) − a† (k)a(k) + e2ik + 0 x0 a† (k)a† (−k) k2 0 0 −e−2ik x a(k)a(−k) − a(k)a† (k) 0 k 0 x0 − a† (k)a(k) − e2ik − a† (k)a† (−k) m2 −2ik0 x0 a(k)a(−k) + a(k)a† (k) e k0 0 x0 + a† (k)a(k) + e2ik = − 1 4 d3 k k0 0 x0 a† (k)a† (−k) (k0 )2 − k 2 − m2 a(k)a(−k) + e2ik 0 x0 × e−2ik a† (k)a† (−k) − (k0 )2 + k 2 + m2 = 1 2 a(k)a† (k) + a† (k)a(k) d3 k k0 a(k)a† (k) + a† (k)a(k) . (5.34). (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) .72) Substituting (5.69).

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

O. we deﬁne normal ordering as the ordering in which the creation operators stand to the left of the annihilation operators.O. (5. we have from (5.78) and we can deﬁne the total number operator for all the oscillators (inﬁnite number of them) in the system as . Taking advantage of this.76) N.73) (namely.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. : a† (k)a† (k′ ) : (5. = a† (k)a† (k′ ) = a† (k′ )a† (k). = = = 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). (5. Let us recall that in the passage from classical to a quantum theory. (for bosons) N. we assume that the quantum theory is deﬁned by the normal ordered Hamiltonian) Ek : a(k)a† (k) + a† (k)a(k) : 2 Ek † a (k)a(k) + a† (k)a(k) 2 d3 k Ek N (k).O.O. then. : a(k)a(k′ ) : = a(k)a(k′ ) = a(k′ )a(k). the ordering of the operators is ambiguous. a(k)a† (k′ ) a† (k)a(k′ ) a(k)a(k′ ) a† (k)a† (k′ ) = = = = : a(k)a† (k′ ) : = a† (k′ )a(k). Thus. This is conveniently done by normal ordering of the operators. If we normal order the Hamiltonian using (5. N. : a† (k)a(k′ ) : = a† (k)a(k′ ).76).O.77) H N. N.

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

the integrand is positive semideﬁnite. H] = Ek a(k). consequently. (5. which can be achieved by quantizing the system in a box. let us consider the normal ordered Hamiltonian (5.83) that E = = = ≥ 0. It is clear from (5.82) Let us consider an energy eigenstate |E of this Hamiltonian satisfying H|E = E|E . (5.6 Energy eigenstates 5 Free Klein-Gordon field theory From now on.75)) [a(k). Let us next recall that the operators a(k) and a† (k) satisfy the commutation relations (see (5.77) and (5. (5. 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).84) This is because both Ek as well as the inner product of states is positive deﬁnite (namely. the inner product represents the norm of the state a(k)|E ) and. E|H|E E| d3 k Ek a† (k)a(k)|E d3 k Ek E|a† (k)a(k)|E (5.186 5. H = −Ek a† (k).83) where we are assuming that the state |E is normalized and that the eigenvalue E is discrete for simplicity. a† (k).85) . we do not have the problem of negative energy states that we had in the single particle theory.

a(k)H|E = Ek a(k)|E . = Ek a(k)|E . the energy eigenvalue is bounded from below (E ≥ 0). Similarly. The annihilation operator a(k) lowers the energy eigenvalue as it should. if |E is an energy eigenstate with energy eigenvalue E. |Emin . (5.89) This is. (5. then a† (k) acting on it gives another energy eigenstate with the higher value of energy E+Ek . or. we note that a† (k).6 Energy eigenstates 187 As a result. H] |E or. clearly. a† (k)H|E H a† (k)|E = −Ek a† (k)|E . or. = −Ek a† (k)|E . therefore. Let us next note that since a(k) acting on an energy eigenstate lowers the energy. there must exist a state (energy eigenstate) with a minimum energy.88) and we cannot lower the energy any further. we have [a(k).84). Namely. therefore.87) − Ha† (k)|E = (E + Ek ) a† (k)|E In other words. (5. In such a state.5. The creation operator a† (k). it would appear that by applying the annihilation operator successively. we can lower the energy eigenvalue arbitrarily. Therefore. such that a(k)|Emin = 0. if |E represents an energy eigenstate with eigenvalue E. d3 k Ek Emin |a† (k)a(k)|Emin Emin = = 0. (5. But as we have seen in (5. . the ground state of the system or the vacuum state (recall that by normal ordering the Hamiltonian.86) − Ha(k)|E H {a(k)|E } = (E − Ek ) {a(k)|E } . raises the energy value. we had redeﬁned . H |E or. then {a(k)|E } is also an eigenstate of energy with the lower energy value E − Ek .

kℓ (5. k2 . nℓ . kℓ . n2 . nℓ . = 1. = 0 = 0|a† (k). (5. A general energy eigenstate with higher energy will have the form (up to normalizations) a† (k1 ) n1 a† (k2 ) n2 · · · a† (kℓ ) nℓ |0 . (5.92) This will be an eigenstate of the total number operator satisfying N |n1 . conventionally called the vacuum expectation value or vev. would vanish. kℓ = a† (k1 ) √ n1 ! n1 a† (k2 ) √ n2 ! n2 ··· a† (kℓ ) √ nℓ ! nℓ |0 . k2 . k1 . · · · . k1 . nℓ . it is clear now that the expectation value of any normal ordered operator in the vacuum state.93) = (n1 + n2 + · · · + nℓ ) |n1 . k2 . = 0. one can build up states of higher energy by simply applying the creation operator. n2 . k1 .90) In other words. nℓ . . (5.188 5 Free Klein-Gordon field theory the energy of the ground state to be zero). · · · . = 0. the vacuum state contains no quantum of energy or no particle if energy is related to that of particles. kℓ = d3 k N (k)|n1 . 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 . n2 . (Incidentally. k2 . · · · . · · · .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 .) Given the vacuum state. n2 .

Namely.94) N (k)|n1 . nℓ . nℓ . k1 . (a† (k1 ))n1 = a† (k) a(k). · · · .97) H|n1 . the state |n1 . the natural deﬁnition from considerations of covariance. k1 .) P= then d3 k k a† (k)a(k) = d3 k k N (k). of course. nℓ . · · · . n2 . 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. k1 .95) n1 δ3 (k − k1 ) + · · · + nℓ δ3 (k − kℓ ) |n1 . · · · . n2 . a† (k1 ) = n1 a† (k1 ) which also leads to n1 n1 δ3 (k − k1 ) . · · · . k1 . (5.6 Energy eigenstates 189 This relation can be easily shown using the identity (see also (5. n2 . nℓ . and . kℓ . However. Let us further note that since H= d3 k Ek a† (k)a(k) = d3 k Ek N (k). n2 quanta with four momentum k2 and so on.5. k2 . kℓ (5. · · · . kℓ = (5. kℓ contains n1 quanta with four µ µ momentum k1 . nℓ . kℓ . · · · . k1 . nℓ . kℓ = d3 k Ek N (k)|n1 . k2 . (5. k1 . (5.80)) N (k).96) if we deﬁne the momentum operator as (This is. k2 .98) = (n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ ) |n1 .

90)) that these states are orthonormal and deﬁne a complete basis of states for the Hilbert space of the Klein-Gordon theory. 5. Namely. k2 .102) It follows from (5. kℓ .100) We can show in a straightforward manner (using the fundamental commutation relations in (5. = k|k . k2 . nℓ .7 Physical meaning of energy eigenstates To obtain the physical meaning of these energy eigenstates. · · · . · · · .190 5 Free Klein-Gordon field theory P|n1 . these states are eigenstates of H and P with total energy and momentum given by E = n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ . let us analyze the state obtained from the vacuum by applying a single creation operator. = Ek |k . k1 . |k = a† (k)|0 . p = n1 k1 + n2 k2 + · · · + nℓ kℓ . · · · .66) as well as the properties (5. k2 . (5. (5.102) that H 2 − P2 |k 2 = (Ek − k2 )|k = m2 |k .101) This state satisﬁes N |k H|k P|k = |k . (5. n2 .103) . nℓ . nℓ . kℓ (5. n2 . kℓ = d3 k k N (k)|n1 .99) = (n1 k1 + n2 k2 + · · · + nℓ kℓ ) |n1 . (5. k1 . n2 . k1 .

107) This is.105) as well as the fact that only the ﬁeld operator φ(x) depends on the space-time coordinates on which the . k2 . This function satisﬁes ∂µ ∂ µ + m2 k|φ(x) = k| ∂µ ∂ µ + m2 φ(x)|0 = 0.106) In fact. of course.5.108) where we have used (5. · · · . nℓ . (5. we can think of this as the one particle state with four momentum kµ = (Ek . satisﬁes the single particle Klein-Gordon equation with positive energy (it provides a representation for the Poincar´ e group) and. (5. the only nonzero matrix element of the ﬁeld operator involving the vacuum state is given by k|φ(x)|0 = k|φ(x) . ﬁeld theories naturally describe systems of many identical particles.7 Physical meaning of energy eigenstates 191 This state. Note that the particles described by such states are necessarily identical (although with diﬀerent energy and momentum) and. therefore. therefore. Similarly. (5. Let us next consider the state that is produced by the ﬁeld operator acting on the vacuum. k2 ) and so on. n2 . k). (5. we can show that the state |n1 . |φ(x) = φ(x)|0 = φ(−) (x)|0 . µ n2 particles with k2 = (Ek2 .105) Since φ(x) is linear in the creation and the annihilation operators. k1 ). an ordinary function representing the projection of the state |φ(x) on to the one particle state |k . it is clear that 0|φ(x)|0 = 0. (5. kℓ . k1 .104) µ can be thought of as the state with n1 particles with k1 = (Ek1 . as a result.

110) In the second quantized description. This brings out the connection between the second quantized theory and the ﬁrst quantized theory. 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). then. Therefore. the wave function is given by ψ(x) = x|ψ .109) to be the positive energy plane wave solutions of the single particle Klein-Gordon equation. 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. 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. |ψ = |k ). (5.192 5 Free Klein-Gordon field theory Klein-Gordon operator ( + m2 ) can act. If |ψ denotes a state of the system. In the present case.109) √ with k0 = Ek = k 2 + m2 > 0. Let us note here parenthetically that the wave function for a single particle is really identiﬁed with φ(x)|k . |k represents an energy eigenstate of the system describing a single particle (namely. in the ﬁrst quantized description. (5. We recognize the function in (5. .

(5.56)). positive semi-deﬁnite and there is no problem of negative . k′ |φ(−) (x1 ) φ(−) (x2 ) |0 . φ (x2 ) = k. Such probabilities are. They can also be shown to deﬁne a complete basis and describe what is known as the Fock space for the system.109). We can similarly construct multi-particle states in the conﬁguration space of the form |φ (x1 ) .113) which will describe a state with n particles at coordinates x1 . the absolute square of the amplitude (with x0 = x0 ) 2 1 k. k) and k′ µ = (Ek′ . for example. such a state contains a superposition of all possible momentum states (which follows from the ﬁeld decomposition in (5. x0 = x0 = · · · = x0 ). (5. The state = φ(−) (x)|0 . These states naturally lead to quantum mechanical probabilities which are non-negative.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. k′ ) at the coordinates x1 and x2 at a given time. k′ |φ (x1 ) . Thus. (5.5. Such states are physically meaningful only if the time coordinates are equal (namely. · · · . xn . φ (x2 ) .114) would give the probability for ﬁnding two Klein-Gordon particles with four momenta kµ = (Ek . As we have seen. by construction. Furthermore.7 Physical meaning of energy eigenstates 193 Thus.111) |φ(x) (5. · · · . such conﬁguran 1 2 tion states are automatically symmetric as we would expect for a system of identical Bose particles. the wave functional for a single particle with a deﬁnite energymomentum has the form ψ(φ(x)) = φ(x)|ψ = φ(x)|k . which is what we have seen explicitly in (5. φ (xn ) = φ(−) (x1 ) φ(−) (x2 ) · · · φ(−) (xn ) |0 .

the second quantized noninteracting Klein-Gordon theory is free from the problems of the ﬁrst quantized theory that we had discussed earlier. Even classically we know that we can solve an inhomogeneous equation of the kind (5. eventually we would like to study interactions of the system. it is invariant under a Lorentz transformation) leading to the fact that the Green’s function is Poincar´ invariant.115) provided we know the Green’s function of the system. (5. namely. Furthermore. all the physical states have positive semideﬁnite energy and that the probabilities are non-negative as they should be. For example.8 Green’s functions Even though we have been examining the free Klein-Gordon ﬁeld theory so far. If we know the Green’s function in (5.117) Here. 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). 2 2 (5.117). e . the equation will take the form ∂µ ∂ µ + m2 φ(x) = J(x). for the Klein-Gordon equation. 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.115) L= (5. The Green’s function for a given inhomogeneous equation is deﬁned as the solution of the equation with a delta source.116) as the Euler-Lagrange equation (5.14). namely. the Green’s function G(x−y) satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). the translation invariance of the Green’s function arises because the right-hand side of the equation is invariant under translations. Thus. 5. which can be obtained from the Lagrangian density m2 2 1 ∂µ φ∂ µ φ − φ + Jφ. which we will study in the following chapters. When there are interactions present. the Klein-Gordon equation will modify.194 5 Free Klein-Gordon field theory probabilities in the second quantized description.

we obtain ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). the Fourier transformation turns the partial diﬀerential equation into an algebraic equation which is trivial to solve.120) d4 k −ik·(x−y) e G(k). 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).8 Green’s functions 195 then the particular solution of the inhomogeneous equation (5. . 1 1 −k2 + m2 G(k) = − .119) = J(x). therefore. 4 (2π) (2π)4 G(k) = k2 1 .117). or.121) In other words. The Green’s function is. Thus. or.5. − m2 (5.118) d4 y −δ4 (x − y) J(y) (5.115) can be written as φ(x) = − so that ∂µ ∂ µ + m2 φ(x) = − = − µ d4 y ∂xµ ∂x + m2 G(x − y)J(y) d4 y G(x − y)J(y). (5. deﬁning the Fourier transforms δ4 (x − y) = G(x − y) = 1 (2π)4 d4 k e−ik·(x−y) . 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. (2π)4 and substituting them back into (5. (5.

substituting (5. then.4.3.120). 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) . Im k 0 −Ek Ek Re k 0 Figure 5.3: Poles of the integrand in the complex k0 -plane.122) has poles at k2 − m2 = 0.121) back into (5.123) which lie on the real axis in the complex k0 -plane as shown in Fig. Therefore.196 5 Free Klein-Gordon field theory Thus. or. If we choose a contour of the form shown in Fig. this is equivalent to moving the poles inﬁnitesimally into the upper half plane. (5. Mathematically. 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. 5. k0 = ±Ek . 5. this is expressed by writing . or.122) G(x − y) = = This explicitly shows that the Green’s function is Poincar´ invariant. (k0 )2 = k2 + m2 . e Let us note that the integrand of the Green’s function in (5. (2π)4 k2 − m2 (5.

4: Choice of the contour in the complex k0 -plane for the advanced Green’s function. we conclude that GA (x − y) = 0.125) where we have identiﬁed ǫ = η . then we must close the contour in the lower half plane for the damping of the exponential in (5. or. or.5. we have to close the contour in the upper half plane (for the damping of the exponential) in which . or. k0 − iǫ = ±Ek .124) are in the upper half plane. (5.124) In this case. if we enclose the contour in the lower half plane. the integral will vanish. 2 Since in this case the poles in (5. the poles of the integrand occur at k2 − m2 − ik0 η = (k0 )2 − k2 − m2 − ik0 η = 0. k0 = ±Ek + iǫ.126) For x0 − y 0 < 0. We note that if x0 − y 0 > 0. GA (x − y) = lim η→0+ e−ik·(x−y) d4 k .124). Therefore. for x0 − y 0 > 0.8 Green’s functions 197 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek + iǫ Ek + iǫ Re k 0 Figure 5. on the other hand. (5. (2π)4 k2 − m2 − ik0 η (5. iη k − 2 0 2 ≃ k 2 + m2 .

sin Ek x0 . For x0 − y 0 < 0.5 or equivalently if we push both the poles to the lower half plane.198 5 Free Klein-Gordon field theory case we can evaluate the integral by the method of residues.127) d3 k eik·(x−y) sin Ek x0 − y 0 . (2π)3 Ek This is known as the advanced Green’s function and has support only in the past light cone.129) would appear at . GA (x) = θ −x0 d3 k eik·x sin Ek x0 . (5. 4 k 2 − m2 + ik 0 η (2π) GR (x) = lim η→0+ (5. 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. if we choose the contour in the complex k0 -plane as shown in Fig. Explicitly. then. 5. the poles of the integrand in (5.128) or. (2π)3 Ek Similarly. we can write (we denote the argument of the Green’s function by x for simplicity) G (x) = A 0. we can express this mathematically by deﬁning e−ik·x d4 k . for x0 < 0. d3 k eik·x (2π)3 Ek for x0 > 0.129) In this case.

8 Green’s functions 199 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek − iǫ Ek − iǫ Re k 0 Figure 5.130) where we have again identiﬁed ǫ = η . x0 < 0. k0 = ±Ek − iǫ. or. k0 + iǫ ≃ ±Ek .129) and. or. we have to close the contour in the upper half plane for the damping of the exponential in (5. then the integral will vanish. if we close the contour in the upper half plane. therefore. (5.5: Choice of the contour in the complex k0 -plane for the retarded Green’s function. we conclude that GR (x) = 0. since both the 2 poles are in the lower half plane. (5. k2 − m2 + ik0 η = (k0 )2 − k2 − m2 + ik0 η = 0. In this case. or. 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 ) . k0 + iη 2 2 2 ≃ k2 + m2 = Ek .131) For x0 > 0.5. Note that if x0 < 0. on the other hand.

134) Both the retarded and the advanced Green’s functions satisfy the inhomogeneous equation (5.133) or. GR (x) = −θ x0 d3 k eik·x sin Ek x0 . we see that we can write 0.135) would also deﬁne a Green’s function corresponding to the contour picking up the principal values at the poles. (5.132) arises because the contour is clockwise in the lower half plane. for x0 > 0. (The advanced and the retarded Green’s functions have similar form except for their support and the overall sign.133) that GR (x) = GA (−x).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 .132) e−i(−Ek −iǫ)x + −2Ek 0 +ik·x The overall negative sign in (5.117) with a delta function source. Therefore.) It is clear from (5. (2π)3 Ek (5. − d3 k eik·x (2π)3 Ek GR (x) = for x0 < 0.128) and (5. the average of the two. sin Ek x0 . (5. (2π)3 Ek Note that this Green’s function has support only in the future light cone and is known as the retarded Green’s function. the diﬀerence between GA (x) and GR (x) would deﬁne a function which would satisfy the homogeneous (Klein-Gordon) equation and . 1 GA (x) + GR (x) . 2 (5. Thus. namely. On the other hand.

known as the Feynman Green’s function. 5. and is antisymmetric. G(x) = −G(−x).8 Green’s functions 201 would not strictly correspond to a Green’s function of the theory. namely. (5. it is easily seen from (5. Quantum mechanically.6. which corresponds to pushing one of the poles (on the left) to the upper half plane while moving the other (on the right) . there is another Green’s function. 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 (5.138) (5.136) that the Schwinger function is real (G(x))∗ = G(x).5. which is more useful in the calculation of scattering matrix elements.137) Furthermore. This is deﬁned with the contour in the complex k0 -plane as shown in Fig. (5. (2π)3 Ek This clearly satisﬁes the homogeneous Klein-Gordon equation ∂µ ∂ µ + m2 G(x) = 0.139) The retarded and the advanced Green’s functions play an important role in many classical calculations as well as in many calculations in statistical mechanics. however.136) + θ x0 = = θ x0 + θ −x0 d3 k eik·x sin Ek x0 .

2 (k0 )2 − Ek + iη = 0.6: Choice of the contour in the complex k0 -plane for the Feynman Green’s function. or.142) . (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. in this case.140) In this case.140) occur at k2 − m2 + iη = 0. or. Mathematically. (k0 )2 − Ek − k0 = ± Ek − iη 2Ek iη 2Ek 2 ≃ 0. there will be a nontrivial contribution independent of whether we close the contour in the upper half or in the lower half plane.141) η where we have identiﬁed ǫ = 2Ek with η → 0+ . (5. (2π)4 k2 − m2 + iη GF (x) = lim η→0+ (5. Since there are poles in both halves of the complex k0 -plane. this can be implemented by deﬁning d4 k e−ik·x .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. the poles of the integrand in (5. = ± (Ek − iǫ) . to the lower half plane.

for x0 < 0. we have used k ↔ −k in the last step (and we understand that k0 = Ek ). for x0 > 0. The Feynman Green’s function clearly has support in both the future as well as the past light cones and can be written as . 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.145) = −G(+) (x). 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 (5.5. if x0 > 0.143) = G(−) (x). where.8 Green’s functions 203 Therefore. On the other hand. Here the contour is clockwise leading to an overall negative sign. we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 (5. Therefore.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.

(2π)3 k0 (5. (2π)4 k2 − m2 − iη G(x) = lim η→0+ (5. (2π)3 Ek G(x) = (5.146) G(+) (−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). We note from the structure of G(−) (x) in (5. we will not go into this in more detail here.148) There is yet another choice of the contour and. (2π)3 d4 k θ(−k0 )δ(k2 − m2 ) e−ik·x .151) G(+) (x) = i G(−) (x) = −i . However.143) and (5.145) that i 2 d3 k eik·x = −G(−) (x).143) and (5.145)) can be written in the manifestly covariant forms d4 k θ(k0 )δ(k2 − m2 ) e−ik·x .150) The functions G(+) (x) and G(−) (x) (see (5. (2π)3 (5.147) Therefore.136) as d3 k eik·x sin Ek x0 = G(+) (x) + G(−) (x). GF (−x) = GF (x). therefore. the Feynman Green’s function is an even function.204 5 Free Klein-Gordon field theory GF (x) = −θ x0 G(+) (x) + θ −x0 G(−) (x). Note from (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.145) that we can write the Schwinger function (5. It is deﬁned as e−ik·x d4 k . namely.143) and G(+) (x) in (5. (5.

we note from (5.154) which will be useful in our later discussions.128). For completeness. 5. GR (x) = −θ(x0 )G(x) = −θ(x0 ) G(+) (x) + G(−) (x) . (5.151)) G(+) (x) ∗ = G(−) (x).9 Covariant commutation relations 205 and are correspondingly known as the positive and the negative energy (frequency) Green’s functions. (2π)3 (5.9 Covariant commutation relations To make contact between the quantum ﬁeld theory and the various Green’s functions that we have constructed.150) that we can write GA (x) = θ(−x0 )G(x) = θ(−x0 ) G(+) (x) + G(−) (x) .133) and (5.5. (5. let us calculate some of the covariant commutation relations satisﬁed by the ﬁeld operators. Let us recall from (5.57) that (we recall that k0 = Ek = 0) √ (5.153) where ǫ(x) = (θ(x) − θ(−x)) is known as the sign function (or the alternating step function).155) k2 + m2 > .58) that φ(x) = φ(+) (x) + φ(−) (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. They satisfy (see (5. (5.151) that the Schwinger function has the manifestly covariant representation G(x) = G(+) (x) + G(−) (x) = i d4 k ǫ(k0 )δ(k2 − m2 ) e−ik·x . where we note from (5.

φ(+) (x). (5. it is straightforward to evaluate φ(+) (x). φ(−) (y) = d3 k e−ik·x a(k). a(k). (2π)3 2k0 d3 k ′ ′ eik ·y a† (k′ ) (2π)3 2k′ 0 . a(k). φ(−) (x). a(k′ ) 0 ′0 2k 2k = 0. φ(−) (y) = = 1 (2π)3 d3 k (2π)3 2k0 eik·x a† (k). φ(+) (y) = = 1 (2π)3 d3 k (2π)3 2k0 e−ik·x a(k).157) d3 k d3 k′ −ik·x −ik′ ·y √ √ e e a(k).206 5 Free Klein-Gordon field theory φ(+) (x) = φ(−) (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k). eik·x a† (k). a† (k′ ) = δ3 k − k′ . a† (k′ ) .66) between the creation and the annihilation operators. d3 k ′ (2π)3 2k′ 0 eik ·y a† (k′ ) ′ d3 k d3 k′ ik·x ik′ ·y † √ √ a (k). a† (k′ ) e e 2k0 2k′ 0 = 0.156) Since we know the commutation relations (5. d3 k ′ (2π)3 2k′ 0 e−ik ·y a(k′ ) ′ (5. a(k′ ) = 0 = a† (k).

40)) and. . φ(y)] = φ(+) (x) + φ(−) (x). 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. we can evaluate [φ(x). (2π)3 Ek (5. φ(+) (y) = iG(+) (y − x) = −iG(−) (x − y). at unequal times. φ(+) (y) = −iG(+) (x − y) − iG(−) (x − y) = −iG(x − y) = −i d3 k eik·(x−y) sin Ek x0 − y 0 .159). the commutator of two ﬁeld operators.158) = −iG(+) (x − y).9 Covariant commutation relations 207 = = = 1 (2π)3 1 (2π)3 1 (2π)3 d3 k d3 k′ −ik·x+ik′·y √ √ a(k).5.159) are not restricted to have equal time arguments any more (unlike in (5.159) Note that the ﬁeld operators in the commutators (5. (5. It follows now that φ(−) (x). φ(+) (y) + φ(−) (y) = φ(+) (x).158) and (5. It is now clear that the Schwinger function satisﬁes the homogeneous equation µ ∂x µ ∂x + m2 G(x − y) = 0. φ(−) (y) + φ(−) (x). using the results in (5. is proportional to the Schwinger Green’s function.161) simply because the ﬁeld operator φ(x) satisﬁes the Klein-Gordon equation.158) and (5.160) Namely. (5. furthermore.

(5.160) leads to the familiar relation (see (5. φ(y)] = 0.40)) [φ(x). Namely. x0 =y 0 (5. for space-like separations. φ(y)]x0 =y0 = −iG(x − y) = 0. (5. G(Λx) = G(x). a light signal cannot connect two space-like points and. for x0 = y 0 d3 k eik·(x−y) sin Ek x0 − y 0 (2π)3 Ek G(x − y)|x0 =y0 = = 0. (The non-commutativity is a reﬂection of the Hamiltonian dynamics. Equation (5. two measurements at space-like separations should not inﬂuence each other. (5. We also note that .160) also shows that two ﬁeld operators φ(x) and φ(y) do not commute for arbitrary values of the coordinates. therefore. this relation can also be seen to imply that [φ(x). 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. (5.166) In other words. This is consistent with our intuitive expectation. namely.138) and (5. the commutator of two ﬁeld operators vanishes.) On the other hand.162) These commutation relations are known as covariant commutation relations simply because the (scalar) Green’s functions are invariant under Lorentz transformations.164) which is expected from the anti-symmetry of the Schwinger function (5.208 5 Free Klein-Gordon field theory ∂µ ∂ µ + m2 φ(x) = 0.163) where Λµν represents a Lorentz transformation. This is known as the principle of microscopic causality.165) x0 =y 0 We note that by a Lorentz transformation.

J. Π(y)]x0 =y0 = x0 =y 0 = −i ∂G(x − y) ∂y 0 x0 =y 0 = iδ3 (x − y). Drell.40). McGrawHill.169) . (5. (5.10 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. Relativistic Quantum Fields. θ(y 0 − x0 ) [φ(x). ˙ φ(x). let us note from (5. which will be useful later. (5. New York.10 References 1. φ(y)] = −iθ(y 0 − x0 ) G(+) (x − y) + G(−) (x − y) = −iθ(y 0 − x0 )G(x − y) = −iGA (x − y).160) as well as the relations in (5. D. 5. Bjorken and S.150) and (5. φ(y)] = −iθ(x0 − y 0 ) G(+) (x − y) + G(−) (x − y) = −iθ(x0 − y 0 )G(x − y) = iGR (x − y). 1964.168) To conclude this section. namely.167) d3 k ik·(x−y) e = −δ3 (x − y).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).136). φ(y) [φ(x).5. (2π)3 This is consistent with our earlier ﬁeld quantization rule (5.

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

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

In such a case. the Euler-Lagrange equations for the primed and the unprimed systems would remain form invariant.5) and only the dynamical variables of the theory transform. It is worth emphasizing here that (6. let us consider an inﬁnitesimal transformation with ∂x′ = 1. x′ µ = xµ . or. 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.3) includes a very interesting class of transformations where the space-time coordinates do not change. ∂µ φ(x)) = 0. (6. d4 x L (φ(x). namely.3) deﬁne a symmetry of the system. ∂µ φ′ x′ δS = or. ∂µ φ(x)) = ′ d4 x′ L(φ′ x′ . Symmetries have interesting consequences for continuous transformations. ∂µ φ(x)) = 0. ∂x (6.3). 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. invariance of the action under the inﬁnitesimal forms of the transformations in (6. Thus. ∂µ φ′ x′ ). (6. the transformations in (6.4) then.3) would imply ′ d4 x′ L φ′ x′ .7) .212 6 Self-interacting scalar field theory S= d4 x L (φ(x). ∂µ φ(x)) = 0. It is clear that in such a case. − d4 x L (φ(x). d4 x L φ′ (x). ∂µ φ′ (x) − d4 x L φ′ (x). ∂µ φ′ (x) − L (φ(x).6) which holds for most global space-time symmetries as well as internal symmetries. as indicated in (6. (6.

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). In fact. (6.¨ 6. ∂µ φ(x)) + δφ(x) −L (φ(x). (6. as we will see later. (6.1 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.7) will hold if L φ′ (x).11) ∂L ∂L + δ(∂µ φ(x)) ∂φ(x) ∂∂µ φ(x) = δφ(x)∂µ = ∂µ δφ(x) . ∂µ φ(x)) = L (φ(x).8) which must hold independent of the use of equations of motion for the system under study. therefore. ∂µ φ(x)) = δφ(x) ∂L ∂L + (∂µ δφ(x)) ∂φ(x) ∂∂µ φ(x) ∂L ∂L + (∂µ δφ(x)) ∂∂µ φ(x) ∂∂µ φ(x) ∂L ∂∂µ φ(x) . (6. Clearly. this is indeed the case for internal symmetry transformations. ∂µ φ′ (x) − L (φ(x). so that δ(∂µ φ(x)) = ∂µ φ′ (x) − ∂µ φ(x) = ∂µ δφ(x). ∂µ φ(x)) = ∂µ K µ .9) we can calculate explicitly (we remember that δφ(x) is inﬁnitesimal and. ∂µ φ′ (x) − L (φ(x). On the other hand. This is quite general and it is possible to have symmetry transformations for which K µ = 0. keep only linear terms in δφ(x)) L φ′ (x).10) (6.

we note that . (6. there is no restriction on δφ and. not all such variations will deﬁne a symmetry.214 6 Self-interacting scalar field theory Here we have used the Euler-Lagrange equation in the intermediate steps. ∂µ J µ (x) = 0.14) Several comments are in order here. However. (6.11). First. we obtain ∂µ δφ(x) or. this is always true when the equations of motion are used. given a conserved current. for any inﬁnitesimal variation. This is just the principle of least action. we can deﬁne a current ∂L − K µ. ∂∂µ φ(x) J µ (x) = δφ(x) (6. therefore. the conserved current independent of the parameter of transformation is not always a vector. x) . For a symmetry.) Comparing (6. we can deﬁne a charge Q= d3 x J 0 (t. In fact. ∂µ δφ(x) ∂L ∂∂µ φ(x) = ∂µ K µ = 0. namely.8) and (6.15) which will be a constant (in time) and will generate the symmetry transformations. Second. Namely. on the other hand. Its tensor structure is determined by the tensor structure of the inﬁnitesimal parameter of transformation. (Note that. φ′ must also satisfy the same equation of motion in the transformed frame as the original ﬁeld. (6. in general.13) which is conserved.12) ∂L − Kµ ∂∂µ φ(x) This shows that whenever there is a continuous symmetry associated with a system. on φ′ .

(6. Q and H can have simultaneous eigenstates. δxµ = x′µ − xµ = ǫµ . In the operator language. The converse is also true. Q corresponds to the generator of the inﬁnitesimal symmetry transformations of the theory. (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. . or. H] = 0. As an example of the consequences of N¨ther’s theorem. This is known as N¨ther’s theorem and is quite important in the study of symmeo tries in dynamical systems.18) where the parameter of translation ǫµ is assumed to be inﬁnitesimal and constant (global). Namely. 6.16) Here we have used the fact that. Since. then it generates inﬁnitesimal transformations (through commutation relations) which deﬁne a symmetry of the theory.3)). the integral of a total divergence (∇ · J) vanishes.1. the time independence of Q also means that it commutes with the Hamiltonian of the theory [Q. if there exists a conserved charge in a given theory. we are dealing with a scalar ﬁeld. Thus. with the usual assumptions on the asymptotic fall oﬀ of the ﬁeld operators (see (5.17) and. (6. there exists a conserved charge which is the generator of these inﬁnitesimal symmetry transformations. In this case.1 Space-time translation.¨ 6. if we have a continuous symmetry in the theory (transformations under which the action is invariant). let us consider the simple case of an inﬁnitesimal o global space-time translation deﬁned by xµ → x′µ = xµ + ǫµ . consequently.

20) where in the last step.10). (6. We see that since the change in the Lagrangian density is a total divergence.22) . where we have used (5.8)) L φ′ (x). then. (6.216 6 Self-interacting scalar field theory φ′ x′ = φ (x) . 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.7) and (6.11) and (6. = − φ′ x′ − φ′ (x) = −ǫµ ∂µ φ′ (x) (6.9) and this is how it corresponds to the negative of the Lie derivative) δφ(x) = φ′ (x) − φ(x) = φ′ (x) − φ′ (x′ ) = −ǫµ ∂µ φ(x). ∂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. we note that under the transformation (6.21) that (ǫµ is a constant parameter) K µ = −ǫµ L. we can now calculate explicitly the change in the Lagrangian density (see (6.20). ∂µ φ′ (x) − L (φ(x).21) = −ǫµ ∂µ φ(x) = −ǫµ ∂µ L = ∂µ K µ .19) and in this case we obtain the change in the ﬁeld to correspond to (see (6. the action is invariant under inﬁnitesimal translations which deﬁne a symmetry of the system. Next. We can now identify from (6.

namely.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 . to be (see (6. Therefore. therefore. (6. depending on the parameter of transformation.13)) ∂L − Kµ ∂∂µ φ(x) ∂L + ǫµ L ∂∂µ φ(x) ∂L − η µν L ∂∂µ φ(x) = −ǫν T µν . that the conserved current independent of the parameter can be identiﬁed with ∂L − η µν L. ∂∂µ φ(x) ∂∂µ φ(x) (6.27) . (6.24) µ Jǫ (x) = δφ = −ǫν (∂ν φ(x)) = −ǫν (∂ ν φ(x)) It is clear.). T µν = T νµ .25) which can be easily checked to satisfy (using the equations of motion) ∂µ T µν = 0.23) so that we obtain the current associated with the symmetry transformation. ∂∂µ φ(x) T µν = (∂ ν φ(x)) (6.1 Nother’s theorem 217 δφ(x) ∂L ∂L = −ǫν (∂ν φ(x)) . (6.¨ 6. even when the naive N¨ther o procedure does not lead to a symmetric stress tensor. 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.

in this case.28) This is consistent with our physical intuition that the energy-momentum. for the free theory. we obtain ∂L = ∂ µ φ(x). would correspond to the components of a four vector which we identify with the energy-momentum operator as Pµ = d3 x T 0µ .) As a result.25) (6. 2 2 (6. the current is a tensor one rank higher than the parameter of transformation. We note that. in this case.29) Therefore. the conserved charges associated with the symmetry transformation (see (6. P µ . which.15)). ∂∂µ φ(x) and this leads to the explicit form for the stress tensor (6. should generate inﬁnitesimal space-time translations. For the free Klein-Gordon theory.218 6 Self-interacting scalar field theory The conserved current. is manifestly symmetric. (In general. is a second rank tensor simply because the parameter of transformation is a four vector.31) = ∂ ν φ(x)∂ µ φ(x) − η µν L = T νµ . in the present case.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 . . (6.

2 2 L0 = (6.34) . Furthermore. we note that the Lagrangian density for the interaction must be invariant under Lorentz transformations as well as translations (Poincar´ invariant).97) and it can be checked that P in (6.6.2 Self-interacting φ4 theory 219 P0 = = = = H. (6. Pi = = P = − d3 x T 00 d3 x d3 x ˙ φ(x) 2 − 2 1 ˙ φ(x) 2 2 m2 2 1 φ + ∇φ · ∇φ + 2 2 1 ˙ φ(x) 2 1 m2 2 + ∇φ · ∇φ + φ 2 2 d3 x T 0i ˙ d3 x ∂ i φ(x)φ(x). (6.34). We had also given an expression for the momentum operator in terms of creation and annihilation operators in (5.33) To include interactions. since the free Lae grangian density is invariant under the discrete transformation φ(x) ↔ −φ(x). 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 . ˙ 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 (5.32) corresponds exactly to the former expression when normal ordered and expressed in terms of creation and annihilation operators.

In these units. λ. the action is dimensionless.35) represents the only meaningful interaction term for such a quantum ﬁeld theory.220 6 Self-interacting scalar field theory we would like to preserve this symmetry in the interactions as well. for various other reasons we can show that (6. which does not respect the discrete symmetry (6.36) While there are other interaction Lagrangian densities that we can construct consistent with our symmetry requirements.) As a result. therefore. (6.35) The restriction on the coupling constant. (We will see this shortly. Thus. . let us introduce the concept of canonical dimensions. is there so that the Hamiltonian will be positive deﬁnite. which in turn allows us to deﬁne a vacuum state of the theory. in fact. With these conditions. To explain this brieﬂy. the same canonical dimension as . the fully self-interacting theory of a real Klein-Gordon ﬁeld is described by a Lagrangian density L = L0 + LI . (6. Let us also g note here that an interaction Lagrangian density of the form − 3! φ3 . would lead to a potential (and. in units of = c = 1 (which we have been using).34). a Hamiltonian) which is unbounded from below for any value of the coupling constant g. then.37) and has. Let us recall that the action for a quantum mechanical particle has the generic form S = = dt (p q + · · · ) ˙ (p dq + · · · ). (6. the simplest interaction Lagrangian density involving only the scalar ﬁelds which we can think of has the form λ 4 φ (x). we can show that the canonical dimension of any variable can be expressed in powers of an arbitrary mass dimension [M ]. Consequently. 4! LI = − λ > 0.

[φ] = [M ]. [M ]−4 [λ] [M ]4 = [M ]0 . [M ]−4 [M ][φ][M ][φ] = [M ]0 .40) Thus. in these units. (6. In general. From the canonical dimension of the interaction term in the Lagrangian density in (6. of course. (6. ([φ])2 = [M ]2 . or. d4 x = [M ]0 . d4 x d4 x [λ] ([φ])4 = [M ]0 . therefore.35). 4! [SI ] = − or.39) In other words. the mass term in the Lagrangian density (the second term) automatically leads to a dimensionless action in these units with the canonical dimension of the ﬁeld already determined. d4 x [∂µ ] [φ] [∂ µ ] [φ] = [M ]0 . [λ] = [M ]0 . let us note that for a monomial interaction action of the form . (6. (The dimensionality of the ﬁeld variable depends on the number of space-time dimensions which we will see later.2 Self-interacting φ4 theory 221 [L] = [T ] = [xµ ] = [M ]−1 .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. the Klein-Gordon ﬁeld variable in four dimensions has a canonical dimension 1.35) is dimensionless.6.) The mass parameter. or. [∂µ ] = [M ]. or. or. has a canonical dimension 1 and. we conclude that the coupling constant or the interaction strength for the φ4 -self-interaction in (6. we obtain λ 4 φ = [M ]0 . or.

the coupling constants in the theory cannot have dimensions of inverse mass. therefore. for example.222 6 Self-interacting scalar field theory g ˜ SI = − n! d4 x φn .34)). In such a case. [g] = [M ]4−n . Such theories are known as non-renormalizable theories. If we want to restrict to renormalizable theories which can give rise to meaningful physical predictions.32)) ∂L ˙ = φ(x). or. for n > 4. (We will see this in a later chapter when we discuss renormalization of quantum ﬁeld theories. the coupling constant of the interaction Lagrangian density will have inverse dimensions of mass. ˜ SI = [g] d4 x [φn ] = [M ]0 .42) so that from (6. restricts n ≤ 4 and. ˙ ∂ φ(x) Π(x) = (6.41) Therefore. the φ4 interaction is the only physically allowed interaction in this case. 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. (6. consistent with our symmetry requirements (see. the canonical momentum conjugate to the ﬁeld variable of the theory continues to be (see (5. Since the interaction Lagrangian density does not involve derivatives of ﬁelds. [g] [M ]−4 [M ]n = [M ]0 . consequently.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! .) This. (6. or.

of course. This leads to diﬀerent possible pictures for describing the same quantum mechanical system through distinct time evolutions. the Hamiltonian is indeﬁnite because of the interaction term. on the other hand. 6. t) carries time dependence. 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. also the . Consequently. the operators carry time dependence while the state vectors do not and the dynamical equations are given by the Heisenberg equations of motion. For λ < 0. the ground state of the free Hamiltonian will not be stable under perturbations. neither the states in the Hilbert space nor the operators acting on state vectors are observables.43) H = = d3 x H d3 x m2 2 λ 4 1 2 1 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! (6.44) = H0 + HI . as we know. the observables correspond to expectation values of Hermitian operators in quantum states. There is. = which leads to (6. each term in the integrand in (6. It is now clear that for λ > 0. 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). This is why we restrict to λ > 0 in the interacting theory.3 Interaction picture and time evolution operator In quantum mechanics as well as in quantum ﬁeld theory.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 .44) is positive deﬁnite and. the Hamiltonian will be bounded from below leading to a meaningful vacuum state. Rather. In this picture.6. Our discussion so far has been within the context of the Heisenberg picture where the ﬁeld operator φ(x. As a result. therefore.

(6. (6. for example. 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. namely. we have assumed that at t = 0 both the pictures coincide. 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.49) . Thus. t) eiHt = φ(H) (x. (6. In the Schr¨dinger picture. It follows from (6. of course.48) The total Hamiltonian operator in the Schr¨dinger picture coino cides with that in the Heisenberg picture (see.45)) and is.46) In writing (6.47) O(S) = O(H) (0). The two pictures are related by the unitary transformation = e−iHt |ψ |ψ(t) (S) (H) .45) that (H) ψ|O(H) (t)|ψ (H) = (S) ψ(t)|O(S) |ψ(t) (S) . for example. (6. 0).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 . |ψ(0) (S) = |ψ (H) . time independent. (6. (6. in the Schr¨dinger picture. t0 ) = e−iH(t−t0 ) .45). the Klein-Gordon ﬁeld o operator will be given by φ(S) (x) = e−iHt φ(H) (x.

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

(6. H0 ) .226 6 Self-interacting scalar field theory (IP HI ) (t) = eiH0 (S) t (S) HI e−iH0 (S) t . therefore.51) that the free Hamiltonian in the interaction picture is time independent and. In this picture. is governed by the Schr¨dinger equation with the interaction o Hamiltonian playing the role of the Hamiltonian. in this picture. From now on. .56) then. if we deﬁne the time evolution operator through the relation |ψ(t) = U (t. t0 ) = eiH0 (S) t −iH(t−t0 ) −iH0 e e (S) t0 . let us drop the superscript (IP ) with the understanding that we are nonetheless working in the interaction picture. t0 ) |ψ (t0 ) . H0 = 1 (IP ) (IP O (t). the time evolution of the state vectors.55) can be seen from (6. it is easy to show that the time evolution operator satisﬁes the properties.54) In other words. (6. the ﬁeld operators can have a plane wave expansion. We note from (6. This shows that once we deﬁne commutation relations for the operators in the free theory.55) where the time evolution operator in the interaction picture in (6. i (6. in the interaction picture they continue to hold even in the presence of interactions.49) and (6. 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).53) Similarly. (6. while the dynamical evolution of the operators is governed by the free Hamiltonian through the Heisenberg equations of motion.50) to have the explicit form U (t.

t0 ) = U (t0 .50)) ∂|ψ(t) = HI (t)|ψ(t) . let us look at the second order term in (6.51)).6. t0 ) = U −1 (t.56) with the use of (6.55) satisﬁes the equation (this also follows from (6. it follows that the total Hamiltonian is time independent in the interaction picture. (6. ∂t ∂U (t.52) we see that the time evolution operator in (6.) We can now solve (6.57) Furthermore. t0 ) = U (t2 . to obtain t U (t. t) = ½.53) that the interaction Hamiltonian in the interaction picture is time dependent. t0 ) t0 t t1 dt1 HI (t1 ) + (−i)2 dt1 t0 t0 dt2 HI (t1 ) HI (t2 ) +··· + ··· +(−i)n +··· . from (6. i ∂t i or. (6. t1 ) U (t1 . t t1 t0 tn−1 t0 dt1 t0 dt2 · · · dtn HI (t1 ) HI (t2 ) · · · HI (tn ) (6. t0 ) . t0 ) t0 ½−i ½−i t dt1 HI (t1 ) t0 t t0 ½−i t1 dt2 HI (t2 ) U (t2 .58) We note from (6. U (t2 .59) and note that . t) .58) iteratively subject to the initial condition in (6. t0 ) . t0 ) = ½ − i = = dt1 HI (t1 ) U (t1 .59) To bring this to a more convenient form. (Since H0 is time independent in the interaction picture (see (6. U † (t.3 Interaction picture and time evolution operator 227 U (t.57). t0 ) = HI (t)U (t.

59) can be written as .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 )) . In a similar manner. (6.60) where T denotes the time ordering operator and is conventionally deﬁned for two bosonic operators as T (A(t1 )B(t2 )) = θ(t1 − t2 )A(t1 )B(t2 ) + θ(t2 − t1 )B(t2 )A(t1 ). t0 (6. we can show that the n-th term in the series in (6.61) with the operator at later time standing to the left of the operator at earlier time.

we also assume that the incoming particles at t = −∞ as well as the outgoing particles at t = ∞ are described by free particle states. 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.. Similarly. (6.62) Therefore. η→0+ η→0+ (6.) 6. in the scattering of particles in relativistic quantum ﬁeld theory. the iterative solution for the time evolution operator in (6.59) takes the form t U (t.4 S-matrix In the non-relativistic quantum mechanical scattering problems.4 S-matrix 229 (−i)n n! t t0 t ··· t0 dt1 · · · dtn T (HI (t1 ) · · · HI (tn )) . −i Rt . in the interaction picture the time evolution operator is given by the time ordered exponential involving the integral of only the interaction Hamiltonian. we can implement this by modifying the interaction Hamiltonian as (η) HI (t) → lim HI (t) = lim e−η|t| HI (t). 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). (6.63) In other words. t0 ) = T e t0 dt′ HI (t′ ) .6.64) . t0 t0 dt1 · · · dtn HI (t1 ) · · · HI (tn ) U (t..63). (This is a formal deﬁnition which is to be understood in the sense of the expansion described in (6. For example.

66) to be Sf i = f |ψi (∞) = f |U (∞.55) to be |ψi (∞) = U (∞. (6. −∞)|i = S|i . (6. As a result. (6. the S-matrix (or the scattering matrix) of the theory can be identiﬁed with the time evolution operator (6.230 6 Self-interacting scalar field theory so that in the inﬁnite past as well as in the inﬁnite future. is obtained from (6. −∞).68) Consequently. Let us denote the initial state at inﬁnite past as the free particle state |ψi (−∞) = |i . (6.64)) leads to the same result for the rate of transition in the physical scattering of particles.64) can be thought of as the relevant boundary condition for the scattering problem under study. which is the deﬁnition of the scattering amplitude. the probability amplitude for an initial state |i to be in the ﬁnal free particle state |f at t = ∞. The condition (6. −∞)|i = f |S|i . the state into which this will evolve at t = ∞ is deﬁned from (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 (6.65) Then. −∞)|ψi (−∞) = U (∞.66) where we have identiﬁed S = U (∞. the interaction vanishes.67) Therefore.67) which has . 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.

63)) −i ∞ R S = U (∞. the linear contribution to the right-hand side of (6. Then. in (6. in general.64) provides a regularized meaning to the oscillatory terms in (6. −∞) = 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+ +··· . This also leads to the notion of “in” and “out” states which are quite important in a formal description of scattering theory.4 S-matrix 231 the explicit perturbative expansion of the form (see (6.70) will have the form . These states can be constructed by noting that the three pictures (Heisenberg. (6.51)) and o since the Heisenberg states are time independent.70) where the time evolution operator U (0. −∞)|Ψ(−∞) (in) . they are uniquely given by the states (in any picture) at t = 0. Let us suppose that at t → −∞ we have a free incoming state denoted as |Ψ(−∞) (in) = |i . say. the S-matrix (the scattering matrix) is unitary since the time evolution operator is (see (6. In this case. not well deﬁned if we do not use an adiabatic interaction of the form.57)).64).6. −∞) is. These are asymptotic free states as t → −∞ and t → ∞ respectively.63). (6.69) through the appropriate boundary conditions. this state would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (in) = U (0. Schr¨dinger and interaction) coincide at t = 0 (see (6.69) Furthermore. It is clear from the expansion above that the adiabatic switching of the interaction in (6. Let us see this explicitly in the case of the linear term in the expansion of the time evolution operator in (6.

64) naturally provides a regularization for the formal deﬁnition of the time evolution operator U (0. they are not identical because their deﬁnitions use diﬀerent regularizing factors. the “in” and the “out” states deﬁne a complete space of states at t → −∞ and t → ∞ respectively.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) . 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 ) . ∞) would not be well deﬁned at the upper limit. the regularizing factor would have to be e−ηt without which U (0. −∞). We also note that the integral in (6.72) and an analysis as in (6.72) we see that even though both the “in” and the “out” states are related to the same Heisenberg state.71) would show that in this case.71) is not deﬁned at the lower limit in the absence of the adiabatic factor eηt . ∞)|Ψ(∞) (out) .71) (IP where in the intermediate steps. Rather.72) we see that they are related to each other as . then this would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (out) = U (0. (6.70) and (6. Furthermore. From (6. (6. In a parallel manner. if we assume that as t → ∞. from their deﬁnitions in (6. This clariﬁes how the boundary condition (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. we have a free outgoing state |Ψ(∞) (out) .70) and (6.

(6. −∞)|Ψ(−∞) = U (∞. 0)|Ψ (in) (H) = U (∞.5 Normal ordered product and Wick’s theorem 233 |Ψ(∞) (out) = U † (0. we need to evaluate the matrix elements of time ordered products of operators between free particle states. Correspondingly. by a phase since the S-matrix is unitary). In terms of ﬁeld operators. (6. such a calculation will involve time ordered products of normal ordered products and this is where the Wick’s theorem comes in handy. we will simplify the time ordered product of factors of normal ordered terms.69) corresponds to no scattering at all. Furthermore. −∞)|Ψ(−∞) (in) = S|Ψ(−∞) (in) . It is clear from this discussion that in calculating the S-matrix elements.5 Normal ordered product and Wick’s theorem The normal ordered product. since our (interaction) Hamiltonian is assumed to be normal ordered.6. contains the annihilation operator while . Later we will see that the Feynman rules are a wonderful and simple way of systematizing these results.73) where we have used the properties of the time evolution operator given in (6. φ(+) . The calculation can be carried out in two steps. simply corresponds to arranging factors in the product so that the creation operators stand to the left of the annihilation operators.57). therefore. as we have deﬁned earlier in (5. we will derive a simple relation for the product of factors each of which is normal ordered and then using this relation. This shows that the “in” and the “out” states are related by the S-matrix in (6. ∞)|Ψ (H) = U (∞. 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.74) where the T-matrix represents the true nontrivial eﬀects of scattering.76). we can deﬁne S − ½ = T. 0)U (0. 6.67) (and. Sf i . The ﬁrst term in the expansion of the S-matrix in (6. First.

For example.75). the order of the factors inside a normal ordered product is not important. annihilated by the annihilation operator. 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). Thus. (6. φ(−) .75) We note that. N φ(−) (x)φ(+) (y) ≡ : φ(−) (x)φ(+) (y) : = φ(−) (x)φ(+) (y). has the creation operator so that this prescription is equivalent to the deﬁnition of normal ordering in (5. we have N (φ(x)) ≡ : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). we note from (6.76)). Furthermore. N φ(+) (x)φ(+) (y) ≡ : φ(+) (x)φ(+) (y) : = φ(+) (x)φ(+) (y). (6.76) that .234 6 Self-interacting scalar field theory the negative energy part.76) Since the vacuum state is. we can equivalently say that φ(+) (x)|0 =0= 0|φ(−) (x). by deﬁnition. N φ(+) (x)φ(−) (y) ≡ : φ(+) (x)φ(−) (y) : = φ(−) (y)φ(+) (x). for the bosonic scalar ﬁeld. N φ(−) (x)φ(−) (y) ≡ : φ(−) (x)φ(−) (y) : = φ(−) (x)φ(−) (y). that the vacuum expectation value of any normal ordered product vanishes. using the relations in (6. (6. therefore. as far as bosonic operators are concerned.77) It is clear.

this also shows that if we can rewrite any product of operators (including normal ordered factors) in terms of normal ordered terms. (6. φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) = : φ(x)φ(y) : −iG(+) (x − y).151) as φ(x)φ(y) = −iG(+) (x − y). in the simple example of the product of two ﬁeld operators.158) and (6. (6. (6.79) where we have used (5. (6. let us note that.81) . therefore. 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. To see how we can write a product of operators in normal ordered form. the calculation of S-matrix elements will simplify enormously.80) with the pairing of the two ﬁeld operators deﬁned to be related to the invariant positive energy Green’s function in (5.78) Therefore.79) as φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y). Furthermore. Let us denote (6. This gives another reason to choose the prescription of normal ordering for physical observables. does not carry any nontrivial quantum number.76). 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).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 = 0.6.

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

80).89) Finally. Next. φ(−) (z) φ(y) +φ(x) φ(+) (y). φ(−) (z) = : φ(x)φ(y)φ(z) : + φ(+) (x). 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) : . 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) : . of course.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(x)φ(y) : −iG(+) (x − y) : φ(x)φ(y) : + φ(x)φ(y).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. (6.87) This is. let us note that . (6.79) and (6. φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + : φ(x)φ(y)φ(−) (z) : + φ(x)φ(y) + iG(+) (x − y) . (6.6. namely.

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

(6.91) 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) : .92) Similarly. Let us note that any product of operators (and not just a product of normal ordered factors) can be written in terms of normal ordered products. 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) : . For example. This now leads us to an important theorem known as the Wick’s theorem (this is one form of it) which says that a product of factors of normal ordered ﬁeld operators can be written as a sum of terms with normal ordered terms and all possible pairings between ﬁeld operators except for pairings between operators within the same normal ordered factor.80) that φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y).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). that (6.6. It follows. we already know from (6. therefore.

In general.93) In other words.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. 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 ) : + ··· . (We are interested in vacuum expectation values because any matrix element can be written as a vacuum expectation value. For example. 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).89) that . (6. This construction can also be extended to the case where only some of the factors are normal ordered. 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. (6.240 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 have already noted in (6.) An immediate consequence of this result is that the vacuum expectation value of an odd number of ﬁeld operators must vanish (since not all ﬁeld operators can be paired).

6.6 Time ordered products and Wick’s theorem

241

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

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

: φ(x)φ(y) : φ(z)φ(w) =: φ(x)φ(y) :

: φ(z)φ(w) : + φ(z)φ(w)

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

T (φ(x)φ(y)) = T (φ(y)φ(x)) =

(6.97)

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

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

(6.98)

242

6 Self-interacting scalar field theory

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

θ(x) =

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

(6.99)

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

θ(x) = lim −

→0+

dk e−ikx . 2πi k + i

(6.100)

Of course, by deﬁnition

T (φ(x)) = φ(x) =: φ(x) : .

(6.101)

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

6.6 Time ordered products and Wick’s theorem

243

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

+θ x0 − z 0 θ z 0 − y 0 φ(x)φ(z)φ(y)

+θ y 0 − z 0 θ z 0 − x0 φ(y)φ(z)φ(x)

+θ y 0 − x0 θ x0 − z 0 φ(y)φ(x)φ(z)

+θ z 0 − x0 θ x0 − y 0 φ(z)φ(x)φ(y)

+θ z 0 − y 0 θ y 0 − x0 φ(z)φ(y)φ(x),

(6.102)

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

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

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

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

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

244

6 Self-interacting scalar field theory

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

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

(6.105)

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

φ(x)φ(y) = iGF (x − y).

(6.106)

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

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

(6.107)

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

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

(6.108)

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

6.6 Time ordered products and Wick’s theorem

245

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

φ(x)φ(y) = iGF (x − y) = φ(y)φ(x),

(6.110)

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

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

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

246

6 Self-interacting scalar field theory

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

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

6.7 Spectral representation and dispersion relation

247

0| [φ(x), φ(y)] |0

= −iG(x − y) =

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

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

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

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

n

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

= i

n

−eikn ·(x−y) 0|B(0)|n n|A(0)|0

,

(6.116)

248

6 Self-interacting scalar field theory

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

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

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

n

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

=

i (2π)3

n

= where

i (2π)3

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

(6.118)

ρ(q) = (2π)3

n

δ4 (q − kn )| 0|A(0)|0 |2 ,

(6.119)

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

6.7 Spectral representation and dispersion relation

249

GAA (x − y) = = =

i (2π)3

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

d4 q ρ(q 2 )

∞ 0

dσ 2 ρ(σ 2 ) i

=

0

∞

dσ 2 ρ(σ 2 ) G(x − y, σ 2 ),

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

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

∞ −∞

f (t) =

dk0 −ik0 t e f (k0 ). 2π

(6.123)

Using the integral representation of the step function

250

6 Self-interacting scalar field theory

θ(t) = lim −

ǫ→0+

1 2πi

∞ −∞

dk0

e−ik0 t , k0 + iǫ

(6.124)

**from (6.122) we obtain 1 2πi
**

∞ −∞

f (t) =

ǫ→0+

lim

−

dk0

e−ik0 t k0 + iǫ

′

∞ −∞

′ dk0 −ik0 t ′ ′ e f (k0 ) 2π

∞

=

ǫ→0+

lim −

−∞ ∞

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

=

ǫ→0+

lim − lim

−∞ ∞

=

ǫ→0+

−∞

dk0 −ik0 t e 2π

∞ −∞

′ ′ f (k0 ) dk0 ′ − k − iǫ . 2πi k0 0

(6.125)

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

∞ −∞ ′ dk0 ′ f (k0 ) , − k0 − iǫ

f (k0 ) = lim

ǫ→0+

′ k0

(6.126)

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

ǫ→0+

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

∞ −∞ ′ dk0 ′ f (k0 ) , ′ k0 − k0

f (k0 ) =

(6.128)

6.7 Spectral representation and dispersion relation

251

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

∞ −∞ ′ Im f (k0 ) , ′ k0 − k0

Re f (k0 ) =

′ dk0

Im f (k0 ) = −

∞

−∞

′ dk0

′ Re f (k0 ) . ′ k0 − k0

(6.129)

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

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

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

−∞

′ Im f (k0 ) , ′ k0 − k0 − iǫ

(6.130)

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

252

6 Self-interacting scalar field theory

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

|k0 |→∞

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

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

ǫ→0+

ǫ→0+

lim

−∞ ∞ −∞

′ dk0

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

ǫ→0+

lim

′ dk0

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

ǫ→0+

lim

1 ¯ π(k0 − k0 )

′ k0

×

1 1 . − ′ ¯ − k0 − iǫ k0 − k0 − iǫ

(6.133)

6.7 Spectral representation and dispersion relation

253

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

′ k0 ∞ −∞

¯ f (k0 ) − f (k0 ) =

ǫ→0+

lim ×

′ ′ ¯ dk0 Im (f (k0 ) − f (k0 ))

1 1 − ′ ¯ . − k0 − iǫ k0 − k0 − iǫ

(6.134)

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

Im f (k0 ) = −Im (−k0 ).

(6.135)

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

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

f (k0 ) − f (0) = lim

ǫ→0+

′ dk0

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

η→0+

(6.137)

This has the imaginary part given by

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

(6.138)

254

6 Self-interacting scalar field theory

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

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

ǫ→0+

lim = = = =

1 π

′ dk0

ǫ→0+

lim − lim −

π 2πEk

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

ǫ→0+

ǫ→0+

lim

1 2 (k0 + iǫ)2 − Ek 1

2 k0

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

η→0+

lim

−

2 Ek

+ ik0 η

,

(6.139)

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

6.8 References

255

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

This page intentionally left blank .

In this case. spin zero (scalar) ﬁelds which are Hermitian.1) Therefore. φ† (x) = φ(x). The relations in (7. there are now two distinct equations. 2 257 .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. 2 (7. 2 1 √ (φ1 (x) − iφ2 (x)) . (7. 2 (7.Chapter 7 Complex scalar ﬁeld theory 7.2) can be inverted to give 1 √ φ(x) + φ† (x) . the dynamical equations take the forms ∂µ ∂ µ + m2 φ(x) = 0.2) φ(x) = φ† (x) = where φ1 (x) and φ2 (x) are two distinct. 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)) . ∂µ ∂ µ + m2 φ† (x) = 0.3) φ1 (x) = i φ2 (x) = − √ φ(x) − φ† (x) .

two distinct particles) with degenerate mass. (Note that we have not been careful about the order of the factors in rearranging terms in (7.5) because distinct degrees of freedom are expected to commute. therefore.) We can quantize this theory in one of the two equivalent ways. Πj (y)]x0 =y0 .1) in terms of the real ﬁeld variables as ∂µ ∂ µ + m2 φ1 (x) = 0. Πj (y)]x0 =y0 = iδij δ3 (x − y). such a system equivalently describes two free. 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. (7.7) . φj (y)]x0 =y0 = 0 = [Πi (x). and is manifestly Hermitian. The Lagrangian density which will give the dynamical equations in (7. If we treat φ1 (x) and φ2 (x) as the independent variables. then we can deﬁne the momenta conjugate to the two ﬁeld operators as Πi (x) = ∂L ˙ = φi (x). independent spin zero Klein-Gordon ﬁelds (and.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. (7. ˙ ∂ φi (x) i = 1.258 7 Complex scalar field theory It follows now that we can express the dynamical equations in (7.5) L = = = ∂µ φ† ∂ µ φ − m2 φ† φ. 2. [φi (x).1) or (7. Therefore.6) Consequently.4) Namely. the equal time canonical commutation relations follow to be [φi (x). ∂µ ∂ µ + m2 φ2 (x) = 0. (7.

6). 2 2 2 i (7. (7. if we treat φ(x) and φ† (x) as the independent dynamical variables.7.8) where we have used the identiﬁcation in (7.1 Quantization 259 In this case. 2 2 . 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 . then we can deﬁne the momenta conjugate to the ﬁeld operators from (7. We see that the Hamiltonian density of the free complex scalar ﬁeld theory is simply the sum of the Hamiltonian densities for two non-interacting real Klein-Gordon ﬁeld theories with degenerate mass and H= d3 x H.9) Alternatively.5) to be Π(x) = = ∂L ˙ = φ(x) ˙ ∂ φ† (x) 1 1 ˙ ˙ √ φ1 (x) + iφ2 (x) = √ (Π1 (x) + iΠ2 (x)) .

Π† (y) = Π† (x).260 Π† (x) = ∂L ˙ = φ† (x) ˙ ∂ φ(x) 7 Complex scalar field theory 1 1 ˙ ˙ = √ φ1 (x) − iφ2 (x) = √ (Π1 (x) − iΠ2 (x)).12) = Π† Π + ∇φ† · ∇φ + m2 φ† φ. (7. φ† (y) = φ† (x).56)).13) Note that the Hamiltonian. in this case.8) and H= d3 x H.11)) [φ(x). take the forms (we do not write the equal time condition explicitly for simplicity. (7.10). of the form . in this case.5) and (7. Π(y) = iδ3 (x − y). in the present case. Π(y)] = Π(x).10) 2 2 As a result. Π† (y) = φ† (x). Π† (y) = 0.2 Field decomposition As in the case of the Hermitian (real) Klein-Gordon ﬁeld. φ† (y) = 0. [Π(x). which can be compared with (7. the equal time canonical commutation relations. φ(x).11) Furthermore. 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. although it is to be understood that x0 = y 0 in all of the relations in (7. φ(y)] = φ(x). is manifestly Hermitian as it should be even though the dynamical variables are not. 7. (7. we can expand the ﬁeld φ(x) in the basis of plane wave solutions of the Klein-Gordon equation (see (5. using (7.

2 1 √ (a1 (k) − ia2 (k)) . 2 (7.2 Field decomposition 261 φi (x) = d3 k (2π)3 2k0 √ e−ik·x ai (k) + eik·x a† (k) . can then be interpreted as the annihilation and the creation operators for the two kinds of quanta associated with the ﬁelds φi (x). it is clear that the ﬁeld operators φ(x) and φ† (x) can also be expanded in the plane wave basis as (see (7. e−ik·x b(k) + eik·x a† (k) . As before.7. i i = 1. 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. with i i = 1. (7. (7. a† (k′ ) = δij δ3 (k − k′ ).14).15) Given the decomposition in (7.17) . j (7. ai (k) and a† (k).16) where we have deﬁned a(k) = b(k) = 1 √ (a1 (k) + ia2 (k)) . a† (k′ ) .14) with k0 = Ek = k2 + m2 > 0.66)) ai (k). aj (k′ ) = 0 = a† (k). 2.2)) φ(x) = = 1 √ (φ1 (x) + iφ2 (x)) 2 1 √ 2 d3 k (2π)3 2k0 e−ik·x (a1 (k) + ia2 (k)) +eik·x a† (k) + ia† (k) 1 2 = φ† (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k) + eik·x b† (k) .

17) b (or directly from (7. a† (k) and † (k) can now be obtained from (7.20) (We can equivalently normal order the Hamiltonian (7. we can write : H := d3 k Ek a† (k)a(k) + b† (k)b(k) . b† (k′ ) = δ3 (k − k′ ). a(k′ ) = b(k). b(k). in two equivalent ways as H= = d3 x H d3 k Ek † a1 (k)a1 (k) + a1 (k)a† (k) 1 2 + a† (k)a2 (k) + a2 (k)a† (k) 2 2 = d3 k Ek † a (k)a(k) + a(k)a† (k) 2 + b† (k)b(k) + b(k)b† (k) .15) and the deﬁnition in (7. a† (k′ ) = a† (k). b† (k′ ) = 0. b† (k′ ) = b(k). b† (k′ ) = b† (k). b(k′ ) = 0.262 7 Complex scalar field theory which correspond to the annihilation operators associated with the quanta for the two ﬁelds φ(x) and φ† (x) respectively. a† (k′ ) = 0.19) If we normal order the Hamiltonian in (7. a† (k).12)) can be written. a(k).19) with respect to the a(k) and b(k) operators. The commutation relations between the a(k). a(k).18) The Hamiltonian for the system (see (7. b(k′ ) = a(k).19) with respect to the ai (k) operators and in that case the Hamiltonian would . (7.16)) and take the form a(k). (7. in terms of the creation and the annihilation operators. (7.8) or (7.11) and (7. a† (k′ ) = b(k).

as was also the case for a real Klein-Gordon theory. (7.23) where θ is a real constant parameter of transformation (a global transformation). under the transformation φ(x) → φ′ (x) = e−iθ φ(x). therefore. there is in addition an internal symmetry associated with the system. After all. degeneracies arise when there is a symmetry present in the system under study. as we will see shortly. H|0 = 0. The degeneracy can be understood as follows.21) However. energy) given by |k = a† (k)|0 . the theory is invariant. The theory is.3 Charge operator To understand the physical meaning of the two distinct and degenerate one particle states. (7. let us analyze the symmetry properties of this system. (7. 7. manifestly Lorentz invariant and translation invariant. Namely. there will now be two distinct one particle states degenerate in mass (and. Such a transformation is not a space-time symmetry transformation since . However. φ† (x) → φ′ † (x) = φ† (x)eiθ . in the present case. 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. = b† (k)|0 .3 Charge operator 263 be a sum of two Klein-Gordon Hamiltonians that we have already studied in chapter 5. Namely. of course. as we know from our studies in non-relativistic quantum mechanics.7.22) ˜ |k Similarly.) The vacuum state in such a case can be deﬁned to be the lowest energy state annihilated by both a(k) and b(k). a(k)|0 = 0 = b(k)|0 = 0|a† (k) = 0|b† (k).

8)) is given by L φ′ (x). Inﬁnitesimally. (7. (7. (7.13)) . ∂µ φ′ (x).5) is invariant under the global phase transformation in (7. in this case. Rather. we have (see (6.24). the transformation in (7. Thus.24)).24) where ǫ denotes the inﬁnitesimal constant parameter of transformation. φ′ † (x). ∂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. the Lagrangian density (7.23) (or the inﬁnitesimal form (7. ∂µ φ(x).26) On the other hand. such a transformation is known as an internal symmetry transformation. φ† (x).23) takes the form (α = ǫ = inﬁnitesimal) δφ† (x) = φ′ † (x) − φ† (x) = iǫφ† (x).23) (or the inﬁnitesimal form of it in (7. the change in the Lagrangian density (see (6. for inﬁnitesimal transformations in (7. ∂µ φ′ † (x) −L φ(x). We note that under the global phase transformation (7. δφ(x) = φ′ (x) − φ(x) = −iǫφ(x).25) The derivatives do not act on the exponentials since θ is a constant parameter (independent of space-time coordinates).264 7 Complex scalar field theory the space-time coordinates are not changed by this transformation.24)) showing that it is a symmetry of the system and in this case we have K µ = 0.

(In fact.1) ∂µ J µ (x) = 0. (7.) Therefore. we normal order all observables such as currents and charges and.29) This is a vector current since the parameter of transformation is a scalar. therefore. But here it is an operator.27).30) Using the ﬁeld decomposition (7.13)) ← → ∂L ∂L + δφ† − K µ = iǫφ† (x) ∂ µ φ(x). Of course. this is the probability current density (up to a normalization) which we had studied in the ﬁrst quantized theory.28) and the parameter independent current follows to be ← → J µ = iφ† (x) ∂ µ φ(x).27) (Note here that classically the order of the factors is not important. we can show that the associated conserved charge of the theory can be written in terms of the creation and annihilation operators as . † ∂∂µ φ ∂∂µ φ µ Jǫ = δφ (7. (7. under normal ordering the order of factors does not matter (for a bosonic theory). we have freely rearranged the order of factors in (7. we can deﬁne the N¨ther current associated with this o symmetry to be (see (6.3 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).16). as we have seen. Even in a quantum theory. (7.) Furthermore. the current can be easily checked to be conserved using the equations of motion in (7.7.

(7. (7. We note that the normal ordered charge operator (7.32). (7. we are suppressing the normal ordering symbol for simplicity with the understanding that the charge operator has the normal ordered form in (7. Acting on the ﬁrst of the two one particle states in (7. a† (k) + a† (k)a(k′ ) |0 (7. the charge operator gives Q|k = = = = d3 k′ a† (k′ )a(k′ ) − b† (k′ )b(k′ ) a† (k)|0 d3 k′ a† (k′ )a(k′ )a† (k)|0 d3 k′ a† (k′ ) a(k′ ).34) d3 k′ a† (k′ )δ3 (k′ − k)|0 = a† (k)|0 = |k . the charge operator takes the form : Q := d3 k a† (k)a(k) − b† (k)b(k) .32) annihilates the vacuum. Q|0 = d3 k a† (k)a(k) − b† (k)b(k) |0 = 0. a U (1) phase transformation) we can identify this operator as the electric charge operator. .22).32) Since this charge operator is associated with a global phase transformation of the kind associated with electromagnetic interactions (namely.31) If we normal order. Here.33) so that the vacuum does not carry any charge.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) .

7.3 Charge operator 267 In other words. the second quantized theory leads to a much more satisfactory description of particles and anti-particles. acting on the second one particle state in (7.35) ˜ = −|k . ˜ Namely. the second quantized description is symmetric under the interchange of particles and anti-particles.20) d3 k Ek a† (k)a(k) + b† (k)b(k) . The vacuum in this bosonic theory. 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′ ). Furthermore. contains no particle and this should be contrasted with the deﬁnition of the Dirac vacuum discussed in section 2. from the form of the Hamiltonian (7.36) we note that it is invariant under the discrete transformation a(k) ↔ b(k).22).37) That is. H= (7. in particular. the particle and the anti-particle states – but both of them now correspond to positive energy states. (7.5. of course. Therefore. These are. In contrast. the second one particle state of the theory |k is also an eigenstate of the charge operator but with the eigenvalue (−1). This transformation is otherwise known as the charge conjugation symmetry which we will . the one particle state |k is an eigenstate of the charge operator with eigenvalue (+1). b† (k) + b† (k)b(k′ ) |0 = = − = − = − d3 k′ b† (k′ )δ3 (k′ − k)|0 = −b† (k)|0 (7. The meaning of the two one particle states is now clear. They describe two distinct particles with degenerate mass but with exactly opposite electric charge.

Therefore. to conclude this section. we will construct only the Feynman Green’s function for the complex KleinGordon ﬁeld theory. let us discuss the Feynman Green’s function in the basis of the complex φ(x) and φ† (x) ﬁelds.146).107) that the Feynman Green’s function can be related to the vacuum expectation value of the time ordered product of two ﬁeld operators.268 7 Complex scalar field theory study in a later chapter. 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.16) as well as the commutation relations in (7. 7. Using the ﬁeld decomposition in (7. Since we have already constructed the Feynman Green’s function for the real Klein-Gordon ﬁeld theory in (5. We note from (6. Thus.18) we now obtain 0|φ(x)φ(y)|0 = d3 kd3 k′ √ 0| e−ik·x a(k) + eik·x b† (k) 3 k0 k′ 0 2(2π) × e−ik ·y a(k′ ) + eik ·y b† (k′ ) |0 = 0. in this section.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 ′ ′ ′ ′ . 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.

38) = iG(−) (x − y).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 = 0.145). 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). Given the relations in (7. 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 (7.38).21) as well as the deﬁnitions of the positive and negative energy Green’s functions in (5. (7.143) and (5. 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 = 0.39) .

namely. 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). (7.270 On the other hand. (7. 4 L = ∂µ φ† ∂ µ φ − m2 φ† φ − λ > 0. 0|T (φ† (x)φ(y))|0 = 0|T (φ(y)φ† (x))|0 = iGF (y − x) = iGF (x − y). However.146). Denoting φ1 = σ and φ2 = χ in the deﬁnition in (7.41) describes the most general (renormalizable) self-interacting theory for the complex Klein-Gordon ﬁeld. Thus.2) (in order to be consistent with the standard notation used in the discussion of spontaneous symmetry breaking in the literature).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. 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. we see that in this case.42) . in this case we should look for interactions that would preserve the internal global symmetry in (7. deﬁning 1 φ(x) = √ (σ(x) + iχ(x)) .40) Here we have used the deﬁnition of the Feynman Green’s function in (5. 2 (7. It is easy to check that the Lagrangian density with the quartic interaction λ † 2 (φ φ) .23).

) λ † 2 (φ φ) 4 2 V = m2 φ† φ + = λ m2 2 σ + χ2 + σ 2 + χ2 2 16 .24) can be shown to have the form δσ(x) = ǫχ(x).47) .41) or (7.44) The Hamiltonian for the interacting theory (7. 16 L = (7. the minimum of energy coincides with the minimum of the potential (We note that if the ﬁeld φ is not constant.5 Spontaneous symmetry breaking and the Goldstone theorem 271 we note that the Lagrangian density (7. the other (kinetic energy) terms would only add positively to the energy. for constant ﬁeld conﬁgurations. 4 H= d3 x Π† Π + ∇φ† · ∇φ + m2 φ† φ + (7.7.43) Namely.43) is invariant under the global phase transformation (7.23) or the inﬁnitesimal form in (7. (7. The Lagrangian density (7. the minimum of the potential is obtained by requiring ∂V ∂V =0= . this describes the theory of two real Klein-Gordon ﬁelds with quartic self-interactions and which are also interacting with each other. ∂V ∂V =0= .42). ∂σ ∂χ (7.46) In general.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 .41) has the form λ † 2 (φ φ) .24). δχ(x) = −ǫσ(x). In the real basis of (7. ∂φ ∂φ† or. the inﬁnitesimal transformations (7. (7.45) and we note that.

(7. a concept that we do not get into here) φc = φ† = 0. (7. Changing the sign of the mass term does not change the . it is clear that the minimum occurs at (the subscript c is supposed to denote a classical ﬁeld. perturbation around the minimum of the potential is stable. (7. this Lagrangian density takes the form 1 m2 2 (∂µ σ∂ µ σ + ∂µ χ∂ µ χ) + σ + χ2 2 2 λ 2 − σ 2 + χ2 .44)).106)). c or. 4 λ > 0.56) has exactly this property (see (5. this translates to the fact that we can deﬁne the ground state of the theory such that φ† = 0|φ† (x)|0 = 0. 16 L = (7.50) and (7. or.50) In terms of the real ﬁelds σ and χ deﬁned in (7.41) except that we change the sign of the mass term L = ∂µ φ† ∂ µ φ + m2 φ† φ − λ † 2 (φ φ) .51) Both the Lagrangian densities in (7. equivalently σc = χc = 0.49) φc = 0|φ(x)|0 = 0. σc = 0|σ(x)|0 = 0.272 7 Complex scalar field theory In the present case.51) are clearly invariant under the global phase transformation in (7. We note from our discussion of the real Klein-Gordon theory that the plane wave expansion for the ﬁeld operator in (5.24) and (7. since the potential in (7. In quantum theory. Let us next consider the same Lagrangian density for the complex scalar ﬁeld as in (7.46) is a sum of positive terms.48) As we know from studies in classical mechanics. and perturbation theory developed around such a vacuum would be stable.42).23) (or (7. c χc = 0|χ(x)|0 = 0.

To begin with. let us note that in the absence of interaction (λ = 0). therefore.55) The minimum of this potential is obtained from (7. (7.40) or (5.4)) − m2 φ(x) = 0. in this case. c = 1) leading to an acausal propagation (and. has many undesirable features.7. k0 = k2 − m2 . 2 − m2 ∂|k| k (7.48) .52) k2 = −m2 . the particle will travel faster than the speed of light (recall that in our units.50) or (7. a violation of causality). or.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 theory (7. corresponding to the Einstein relation (7.54) Namely. but it does change the Hilbert space structure of the theory in a very interesting way. in general.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. for constant ﬁeld conﬁgurations the potential of the theory (7.5 Spontaneous symmetry breaking and the Goldstone theorem 273 symmetry behaviour.41) would lead to the dynamical equation (compare with (1. (7. Let us note that. Such particles are known as tachyons and a theory with tachyons.

χc =0 λ = σc =0=χc ∂2V ∂χ2 σc =0=χc = −m2 . (7.56). = 0.56) = χ −m2 + There now appear to be two sets of solutions to the equations in (7. for example. (7. = σ −m2 + ∂V ∂χ = −m2 χ + λ χ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.59) = 0. 2 σc + χ2 = c 2m σc = ± √ .χc =0 λ (7. namely. 2m σc =± √ .χc =0 λ .57) λ For the second of these solutions.57) which lead to ∂2V ∂σ 2 ∂2V ∂σ∂χ while ∂2V ∂σ 2 ∂2V ∂χ2 ∂2V ∂σ∂χ 2m σc =± √ . 4m2 .58) To determine the true minimum. λ χc = 0. we can choose σc = 0 = χc .60) 2m σc =± √ . 3λ 4 λ 4 4m2 λ 4m2 λ = 2m2 . σc =0=χc = −m2 + = −m2 + = 0.274 7 Complex scalar field theory ∂V ∂σ = −m2 σ + λ σ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0. (7. let us analyze the second derivatives of the potential at the two extrema in (7. (7.

1.1: Potential leading to spontaneous breaking of symmetry. indeed. we note that there is an inﬁnity of minima (in the σ-χ ﬁeld space) of the potential given by the circle (7. we note explicitly from the form of the potential in (7. V (σ. All of this is more easily seen from the graph of the potential in Fig. the second solution in (7.57) is a true minimum and we conclude that the true minimum of the potential occurs for 2 2φ† φc = σc + χ2 = c c 4m2 . There are several interesting things to note from this analysis. the ﬁrst corresponds to a local maximum. First.44). However. Each point on this circle is related by the symmetry transformation (7.62) so that the second solution.55) that V (σc = χc = 0) = 0. leads to a lower energy.5 Spontaneous symmetry breaking and the Goldstone theorem 275 In other words we note that out of the two solutions in (7. λ (7. V 2 ((σc + χ2 ) c m2 4m2 ) = − = λ 2 = − 4m2 λ λ + 16 4m2 λ 2 2m4 m4 m4 + =− . .61).7. χ) V =0 χ 2m −√ λ 2m √ λ σ Figure 7.57). λ λ λ (7.61) In fact. On the other hand. 7.

in the quantum theory this would imply (see (7.49)) 2m 0|σ(x)|0 = √ . the action or the Hamiltonian of the theory is invariant under the symmetry transformation.64).65) 0|δχ(x)|0 2m = −ǫ 0|σ(x)|0 = − √ ǫ. In this case. if we choose the minimum in (7. (7. (7. χ(x)] |0 (7. χ(x)] = −ǫσ(x). δχ(x) = −iǫ [Q.61) to correspond to the point (in the σ-χ ﬁeld space) 2m σc = √ .44) spontaneously. We recall that any choice for the minimum of the potential would translate into the properties of the vacuum state of the quantum theory.44) can be written in terms of the commutators with the charge operator as δσ(x) = −iǫ [Q. but the ground state is not. the inﬁnitesimal transformations in (7. From (7. Thus. such a breaking manifests in the following manner. λ (7.64) We recall that conserved charges are the generators of inﬁnitesimal symmetry transformations of the theory and generate the transformations through commutators. λ χc = 0. In this case. for example.66) which implies that . it now follows that = −iǫ 0| [Q. In the quantum theory. σ(x)] = ǫχ(x).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. λ 0|χ(x)|0 = 0. This has to be contrasted with the case where an interaction term in the Hamiltonian can break the symmetry of a theory explicitly.63) then.

68) and (7. The problem with this interpretation is that the state is not normalizable. (7. or. (note that Q is Hermitian) χ|χ = = = 0|QQ|0 0| d3 x J 0 (x. H] = 0. or. HQ|0 = E0 Q|0 .31) and (7.69) as follows. namely.70) (7.68) Since Q does not annihilate the vacuum of the present theory. (7. let us denote Q|0 = |χ It follows now from (7. (7. or. it is suggestive that we can think of this state as another vacuum state.71) .5 Spontaneous symmetry breaking and the Goldstone theorem 277 Q|0 = 0.7. t)Q|0 d3 x 0|eiP ·x J 0 (0)e−iP ·x Q|0 . (QH − HQ)|0 = 0. 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. Therefore. This can be easily seen from (7. [Q. H|χ = E0 |χ .69) would appear to be degenerate with the vacuum state |0 in energy. (7.67) This explicitly shows that the vacuum state of the theory is not invariant under the symmetry transformation in this case.69) that [Q. although the Hamiltonian of the theory is.

we have [Pµ . it follows that the momentum operator also commutes with Q and. 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 (7.75) does not act unitarily on the Hilbert space. (7.73) where we have used the property of the ground state Pµ |0 = 0.74) In other words. . Since Q does not depend on spatial coordinates. it is straightforward to show that the charge Q does not exist when there is spontaneous breakdown of the symmetry and the formally unitary operator in (7. Q] = 0.72) χ|χ = = = d3 x 0|eiP ·x J 0 (0)Qe−iP ·x |0 d3 x 0|J 0 (0)Q|0 0|J 0 (0)Q|0 d3 x −→ ∞.66) that χ(x) corresponds to the Nambu-Goldstone boson of the theory.72) leads to (7. and as a consequence. This analysis also shows that the ﬁnite transformation operator U (θ) = e−iθQ .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. (7. (7. the state |χ is not normalizeable and hence cannot be thought of as another vacuum. In fact. (7.75) takes a state out of the Hilbert space.

With the choice of the solutions for the minima as in (7.51) becomes . (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 (7.64) 2m σ = 0|σ(x)|0 = √ .77) so that the shifted ﬁelds have vanishing vacuum expectation values (vevs) 0|σ(x)|0 = 0 = 0|χ(x)|0 .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) + √ .1 that the massless mode would correspond to oscillations along the valley of minima while the massive mode would represent the orthogonal oscillations. We note from Fig. √ 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. λ χ = 0|χ(x)|0 = 0.60) 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 (7.78) (Explicitly. (7. say for the ﬁrst term.60) is as follows. is given by σ(x) = σ ′ (x) + σ with the later identiﬁcation σ ′ (x) = σ(x) as the dynamical variable.7. 7. We can see this quantitatively in the quantum theory as follows. λ χ(x) → χ(x) + χ = χ(x). (7.53) and (7. the shift in (7. the Lagrangian density in (7.) Under the shift (7.54))).77).77).

66)).79) This shows explicitly that perturbed around the true minimum. if we look at the Hamiltonian describing a . a continuous symmetry is spontaneously broken. aware of this phenomenon (known as the spontaneous symmetry breaking phenomenon) from studies in other areas of physics. We are. 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 (7.280 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 . 2 16 (7. then there must appear massless excitations. 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. For example. 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.

(7. Given the complex Klein-Gordon ﬁeld theory. it is clear that once we have chosen a particular ground state (vacuum) where all the spins are pointing in a given direction.81) where “e” denotes the charge carried by the particle. The analogue of the massless Nambu-Goldstone bosons in this case correspond to the long range correlations associated with the spin waves. The prescription is again through the minimal coupling discussed in (1. (7. ∂µ → ∂µ + ieAµ .6 Electromagnetic coupling It is clear from our discussions so far that the complex Klein-Gordon ﬁeld theory describes spin zero charged particles whereas the real Klein-Gordon ﬁeld theory describes spin zero charge neutral particles. The ground state of the theory is clearly the state (spin conﬁguration) where all the spins are aligned. namely.7.) However. we do not observe any massless spin 0 bosons in nature. Unfortunately. there are two such vacuum conﬁgurations allowed in one dimension.80) i where the negative sign in the Hamiltonian (g > 0) is for ferromagnets. (For example. we have eﬀectively chosen a deﬁnite orientation in the physical space. H = −g si · si+1 . when the spins are all pointing up or down.6 Electromagnetic coupling 281 magnet (for example.) 7.53). (The π meson which is the lightest among the spin zero bosons has a mass mπ ∼ 140 MeV. the Hamiltonian is invariant under rotations. 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. Consequently the rotational symmetry is spontaneously broken. we can ask how such a system can be coupled to a background electromagnetic ﬁeld. 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 . in the case of the Ising model).

→ ((∂µ + ieAµ ) φ)† (∂ µ + ieAµ ) φ − m2 φ† φ = = (7.83) is known as the covariant derivative since it transforms covariantly under a local phase transformation (or equivalently under a gauge transformation) as we will see shortly.82) is invariant under a local (U (1)) phase transformation of the form φ(x) → φ′ (x) = e−iθ(x) φ(x). As we can verify readily. 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) Dµ φ = ∂µ φ + ieAµ φ. note that under this transformation. e (7. In fact. Aµ (x) → A′ (x) = Aµ (x) + µ 1 ∂µ θ(x).23)). (7.84) where the parameter of transformation θ(x) is now a local function of space-time coordinates (unlike in (7. the Lagrangian density (7.282 7 Complex scalar field theory L = ∂µ φ† ∂ µ φ − m2 φ† φ (∂µ − ieAµ ) φ† (∂ µ + ieAµ ) φ − m2 φ† φ (Dµ φ)† (D µ φ) − m2 φ† φ.82) The modiﬁed derivative Dµ introduced in (7.85) . (7.

7 References 1.84). 4. the local phase invariance (7. Of course. .7. Salam and S. Goldstone. is invariant under the local phase transformation (7. J.86) Consequently. (In contrast. 345 (1961). unlike the global phase invariance in (7. Aµ (x)) → L(φ′ (x). Local phase transformations are commonly known as gauge transformations and we note that.84) L(φ(x).84) requires an additional ﬁeld (Aµ ). Physical Review 127. Schweber. Peterson. invariance under a local transformation necessarily requires additional ﬁelds known as gauge ﬁelds. Aµ . the theory described by (7. Weinberg.23).) It follows from this observation that under the local transformation (7. 2. 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 9.7 References 283 In other words. S. φ† . Jona-Lasinio. Nambu and G. Nuovo Cimento 19. 965 (1962). This is a very general feature in a quantum ﬁeld theory. (7. Row. φ′ † (x). the covariant derivative acting on a ﬁeld transforms exactly like the ﬁeld itself (covariantly) under such a transformation (see (7. note that ∂µ φ(x) does not transform covariantly. J. Introduction to Relativistic Quantum Field Theory. 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. which is obtained by minimally coupling the complex scalar ﬁeld to the electromagnetic ﬁeld. 3. Goldstone. Y. namely. Evanston (1961). A. Physical Review 122. φ† (x).84)). 154 (1961). 7.82).

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

Chapter 8 Dirac ﬁeld theory 8. (There can be more fermions only if they are non-identical. we have to ﬁnd a mechanism for incorporating the Pauli principle into our theory. fermions obey the Pauli exclusion principle which simply says that there can at the most be one fermion in a given state. therefore. a† F F aF . Let us consider an oscillator described by the annihilation and the creation operators aF and a† respectively. In trying to go beyond and study the theory of spin 1 ﬁelds. Unlike 2 Bose particles. aF ]+ = a2 + a2 = 0. 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.) Thus.2) = aF a† + a† aF = 1.1) It is easy to show that we can assign Fermi-Dirac statistics to such an oscillator (and. in dealing with such systems. F F 285 . 2 we note that spin 1 particles such as electrons are fermions. F F a† . a† F + + = a† F 2 + a† F 2 = 0.1 Pauli exclusion principle So far we have discussed ﬁeld theories which describe spin zero bosonic particles. For example. if we require [aF . F (8. The number operator F for the system is given as usual by NF = a† aF . (8.

therefore. 3.4) which shows that the eigenvalues of the number operator.6) where ψα (x). 4. 8. 1 2 (8. It is clear. if this system is analyzed in more detail. has the form (iγ µ ∂µ − m) ψ = 0.286 then. can only be nF = 0. the anti-commutation relations in (8. that in dealing with fermionic ﬁeld operators.3) = a† 1 − a† aF aF = a† aF = NF . what the Pauli exclusion principle would say. the wave function of the system will have the necessary antisymmetry property characteristic of fermions. α = 1. we obtain 2 NF = a† aF a† aF F F 8 Dirac field theory = a† F aF .5) This is. a† F + − a† aF aF F (8. there can be at the most one quantum in a given state. This equation is manifestly Lorentz covariant and the adjoint equation has the form . 2. (8. In fact. (8. Namely. the Dirac equation for a massive spin particle.2 Quantization of the Dirac ﬁeld As we have seen in (1. it leads to the fact that with the anti-commutation relations. of course. is a four component complex spinor function. 1.96). in such a theory. F F F In other words.2) automatically lead to NF (NF − 1) = 0. we will be naturally dealing with anti-commuting variables (these variables are commonly known as Grassmann variables) and anti-commutation relations.

9) give ∂L = (iγ µ ∂µ − m) ψ = 0. as we have seen in chapter 3 as well as in chapter 4. let us note that since ψ and ψ are anti-commuting (Grassmann) variables.8.8) In trying to second quantize the Dirac equation (theory). we have to be careful about signs that may arise in taking derivative operators (with respect to these variables) past other Grassmann variables.41)) ψ(x) = ψ † (x)γ 0 . however.6) and (8. With this. ∂ψ ← − ∂L ∂L − ∂µ = mψ − ∂µ (−iψγ µ ) = ψ iγ µ ∂µ + m = 0. consequently. in this case. need to be treated as anti-commuting operators. the Lagrangian density of the form / L = ψ (iγ µ ∂µ − m) ψ = iψ∂ ψ − mψψ. ∂ψ ∂∂µ ψ (8.9) is manifestly Lorentz invariant (recall the transformation of the Dirac bilinears under a Lorentz transformation discussed in section 3. the 1 Dirac ﬁeld operators are expected to belong to the (reducible) spin 2 representation of the Lorentz group (see (4. the Euler-Lagrange equations following from the Lagrangian density (8. (8.2 Quantization of the Dirac field 287 ← − ψ iγ µ ∂µ + m = 0. 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. In dealing with the Dirac ﬁeld theory.10) . where the adjoint spinor is deﬁned to be (see also (2.7) as the Euler-Lagrange equations.61)) describing fermionic particles and. Clearly. we would like to decide on a Lorentz invariant Lagrangian density which would give rise to the two Dirac equations (8. In general. In deriving the Euler-Lagrange equations.3).7) (8. (8. we will always use the convention of taking fermionic (Grassmann) derivatives from the left. However.

It is worth noting here that even though the Lagragian density (8. 2 2 L′ = (8. (8.83). (8. the action associated with it is since a total divergence does not contribute to the action. 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. 2 . An alternate Lagrangian density which is Lorentz invariant. Namely.8) as well as the fact that γ 0 is Hermitian.6) and (8.288 8 Dirac field theory which are exactly the two Dirac equations (8.7) that we are interested in.12) However. In fact. it is not Hermitian. we note that the Lagrangian density in (8. 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.9) is not Hermitian.9) only by a total divergence.11) † where we have used (2. Even though the Lagrangian density (8. note that L† = ψ (iγ µ ∂µ − m) ψ ← − † = ψ † − iγ µ † ∂µ − m ψ ← − = ψ † − iγ µ † ∂µ − m γ 0 ψ ← − = ψ − iγ µ ∂µ − m ψ = ψ (iγ µ ∂µ − m) ψ − i∂µ ψγ µ ψ = L.12) diﬀers from L in (8.9) is Lorentz invariant and gives the two Dirac equations as the Euler-Lagrange equations.13) = L − ∂µ ψγ µ ψ .

. ˙ ∂ψ (8.16) = δαβ δ3 (x − y).14) Πψ = Πψ = Consequently. (Πψ )β (y)]+ = −iδαβ δ3 (x − y). Using the ﬁrst relation in (8. it is more appropriate to use the Lagrangian density in (8. If we work with the Lagrangian density (8. + † = ψα (x).14).15) can also be written as (Πψ )α (x). ψβ (y) + = −iδαβ δ3 (x − y). ˙ ∂ψ ∂L = 0. (Πψ )β (y) or. = 0.12) which naturally leads to a symmetric stress tensor (see (6.8. ψβ (y)]+ = 0 = [(Πψ )α (x). † ψα (x).9) to describe the Dirac theory (equations). (8. (Πψ )β (y)]+ . † † ψα (x). [ψα (x). we use L in (8.9) to describe the dynamics of the system.27)) for the theory.2 Quantization of the Dirac field 289 For this reason as well as for simplicity. ψβ (y) + + = 0.15) The negative sign in the last relation in (8.16)) [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 † .15) and (8. the second and the third relations in (8. (8. in the presence of gravitation. However.15) is a reﬂection of our choice of left derivatives. but we should understand that x0 = y 0 in all the relations in (8. (Πψ )β (y) or. −iψβ (y) + ψα (x). 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.

18) (8.17) Using the ﬁeld anti-commutation relations (8. We will discuss the method of Dirac in more detail in chapter 10. Such systems can be systematically studied through the method of Dirac brackets and such an analysis leads to the conclusion that the relations (8. For example.15) and (8. (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.9) has the form (The negative sign in the ﬁrst term arises from our convention of choosing left derivatives for fermions.19) . ψβ (y) + −iγ 0 γ · ∇y ψ(y) + β † + m ψα (x). H = d3 y † ψα (x). (8.290 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 (8.15) and (8.) ˙ H = −Πψ ψ − L ˙ ˙ = − −iψ † ψ − iψ † ψ − iψγ · ∇ψ + mψψ = −iψγ · ∇ψ + mψψ.16). it is now easy to check that this Hamiltonian leads to the two Dirac equations as the Hamiltonian equations.14) can be thought of as constraints on the dynamics of the system.) The Hamiltonian density for the Dirac ﬁeld theory (8. ˙ iψα (x) = ψα (x). leading to the Hamiltonian H= d3 x H = d3 x −iψγ · ∇ψ + mψψ .

15) and (8. or.18) indeed reproduce the two Dirac equations (8. (8. (iγ µ ∂µ − m) ψ(x) = 0. ← − ψ iγ µ ∂µ + m = 0. ψβ (y) + † −mψ β (y) ψα (x).6) and (8. the anti-commutation relations (8.21) and this can be written in the matrix form as ← − ˙ iψ † = −iψγ · ∇ − mψ. or. ψβ (y) + x0 =y 0 = d3 y i ψ(y)γ · ∇y β δαβ δ3 (x − y) x0 =y 0 −mψ β (y)δαβ δ3 (x − y) ← − = −i ψ(x)γ · ∇ α − mψ α (x).20) Similarly.8. or.2 Quantization of the Dirac field 291 where we have used the fact that the Hamiltonian is independent of time to identify x0 = y 0 and this equation can be written in the matrix form as ˙ iψ(x) = −iγ 0 γ · ∇ψ(x) + mγ 0 ψ(x). ˙ iγ 0 ψ(x) = −iγ · ∇ψ(x) + mψ(x).22) Namely. (8.16) as well as the Hamiltonian (8. ← − ˙ iψγ 0 = −iψγ · ∇ − mψ. H d3 y i ψ(y)γ · ∇y β † ψα (x). .7) as the Hamiltonian (Heisenberg) equations. we obtain ˙† iψα (x) = = † ψα (x). or. (8.

/ (8.23) can be inverted to give √ (k0 = Ek = k2 + m2 > 0) d3 x = s′ m eik·x u† (k. s)vα (k′ .23) automatically satisﬁes the Dirac equation in this basis (i∂ − m) ψ(x) = 0.5).94) and the ﬁeld (8. s) denote the positive and the negative energy spinors deﬁned in (3. let us note that the ﬁeld decomposition in (8. s) and vα (k. s′ ) α ′ . where c(k. s) + eik·x d† (k.93) and (3. s) . s) and d† (k. s′ ) α 3 (2π) k0 k′ 0 +ei(k+k )·x u† (k. As in the case of the Klein-Gordon ﬁeld (see section 5.23) × e−ik·x c(k. s)uα (k′ .46).4). s)uα (k. we can expand the Dirac ﬁeld operator in the basis of these plane wave solutions.292 8.23) has been chosen keeping in mind the massive normalization of the plane wave solutions in (2. Therefore. Taking advantage of our experience with the complex Klein-Gordon ﬁeld theory. uα (k.24) The speciﬁc form of the prefactor in (8. s′ ) d† (k′ . s)ψα (x) α (2π)3 k0 d3 x d3 k ′ m ′ √ ei(k−k )·x u† (k.3 Field decomposition 8 Dirac field theory We have already seen that the positive and the negative energy plane wave spinor solutions of the Dirac equation deﬁne a complete basis for the space of four component spinor functions (see the discussion in section 3. s) are anti-commuting operators. s′ ) c(k′ . we write √ (k0 = Ek = k2 + m2 > 0) ψα (x) = 1 s=± 2 d3 k m (2π)3 k0 (8. s)vα (k.

x0 =y 0 +. s)ψα (x) = d† (k. s) in a straightforward manner from the quantization conditions (8.16) to be ψα (x).26) It follows from (8. ψβ (y) +. d† (k . s ) d† (k . s).3 Field decomposition 293 δ3 (k − k ) u† (k. In a similar manner.29) + .8. (2π)3 k0 m † eik·x ψα (x)vα (k. (8. s ) = δss δ3 (k − k ) = d(k.25) and (8. using the inversion relations (8.46) (see also (3.27) c† (k.92)). = 0. s ) c(k .25)-(8. s) = d3 x d3 x Furthermore. s).28) and they correspond to c(k. (2π)3 k0 d3 x (8. ψβ (y) † ψα (x). c† (k . s). s ) α 0 x0 = s d3 k √ m k0 k 0 + δ3 (k + k ) e2ik = s u† (k. s). s).15) and (8. (2π)3 k0 (8. we know the equal time quantization conditions for the ﬁeld operators from (8. s). d† (k.28) Therefore. it can be shown that m † e−ik·x vα (k. s) = d(k.x0 =y 0 +. c† (k. (8. s)uα (k . s ) + . s). s ) α (8.26) that m † e−ik·x ψα (x)uα (k. s)vα (k . s ) = c(k.25) m k0 δss c(k. we can now obtain the anti-commutation relations satisﬁed by the operators c(k. = δαβ δ3 (x − y). ψβ (y) † † ψα (x). s).x0 =y 0 = 0. k0 m where we have used the massive normalization for the spinors discussed in (2. s) and d(k.27).

s′ c† (k′ s′ )eik ·x + v k′ . s) + eik·x d† (k. s′ )e−ik ·x × e−ik·x c(k.18) of the theory.93) and (3. s) = (γ 0 k0 − γ · k − m)u(k.s′ =± 2 ′ d3 x ψ (−iγ · ∇ + m) ψ m k0 ′ m 3 d k k′ 0 × u(k′ .30) and write (k − m) u(k. s) + eik·x d† (k. s). s)u(k. s) = (γ 0 k0 − γ · k + m)v(k. / we can simplify the Hamiltonian in (8. This shows that we can think of c(k. / (8.294 8 Dirac field theory with all other anti-commutators vanishing. s) = 0. c† (k. s′ d(k′ . s) = 0. s) .94) (8. s)(−γ · k + m)v(k. If we now use the relations (3. s) as annihilation and creation operators for the two kinds of fermionic quanta in this theory (much like the complex Klein-Gordon ﬁeld theory). The Hamiltonian (8.30) (k + m) v(k.31) . 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′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m k0 ′ × u k′ . s′ )e−ik ·x × (−iγ · ∇ + m) × e−ik·x c(k. in this case. s′ )d(k′ . s). s′ )c† (k′ . s) = 1 s. s′ )eik ·x + v(k′ . s)(γ · k + m)u(k. d† (k. s) as well as d(k. s)v(k.

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

we pick up a negative sign since they anti-commute (as opposed to bosonic variables). s)|0 .21)). s)d(k. every time we move a fermionic operator past another in order to bring the product into its normal ordered form.32) is obtained to have the form : H := s=± 1 2 d3 k Ek c† (k.33). s′ ). = d† (k.22)) = c† (k. For example. s). the normal ordered Hamiltonian for the free Dirac ﬁeld theory in (8. = 0. Using (8. (8. namely.33) so that the order of factors inside a product that is normal ordered becomes important. s)d(k′ . s) and d(k. s)|0 H|0 The one particle states of the theory can now be deﬁned as (see (7.34) even though we do not put the normal ordering symbol explicitly. we can deﬁne the vacuum state of the theory to correspond to the state which is annihilated by both c(k. we now have : d† (k. s) + d† (k. s)|0 d(k. s)c(k. s)d(k′ .35) c(k. s |k. s′ ).) = 0 = 0|c† (k. s) (The Hamiltonian is understood to be normal ordered as in (8.34) As in the case of the complex (Klein-Gordon) scalar ﬁeld theory (see (7. : d(k′ . s)|0 . (8. s′ )d† (k. s). s′ ) : = d† (k. (8. s) : = −d† (k. However. s)d(k′ . (8. s) .36) |k. we write products with the creation operators standing to the left of the annihilation operators. s . = 0 = 0|d† (k.296 8 Dirac field theory way as for the bosonic operators.

we need to analyze the symmetry properties of the theory.9) L = ψ (iγ µ ∂µ − m) ψ. (8. ψ(x) → ψ (x) = ψ(x)eiθ . 8. transforms as L(ψ.39) = ψeiθ (iγ µ ∂µ − m) e−iθ ψ = ψ (iγ µ ∂µ − m) ψ = L(ψ.40) . there are two distinct one particle states with degenerate mass (energy) and to understand the meaning of these states.4 Charge operator The Dirac Lagrangian density (8. We note that under the global phase transformation (α is the constant parameter of transformation) ψ(x) → ψ ′ (x) = e−iθ ψ(x). ′ (8. as in the case of the complex scalar ﬁeld theory. δǫ ψ = ψ (x) − ψ(x) = iǫψ(x).38) (8. Therefore. ψ) → L(ψ ′ .12)).34) with energy eigenvalue k0 = Ek = k2 + m2 > 0.37) or the inﬁnitesimal form of it (ǫ is the constant inﬁnitesimal parameter of transformation) δǫ ψ = ψ ′ (x) − ψ(x) = −iǫψ(x).8. ψ ) = ψ (iγ µ ∂µ − m) ψ ′ ′ ′ ′ (8.9) (or (8. ψ).4 Charge operator 297 Both these states can be checked to be eigenstates of the Hamil√ tonian (8. 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.

the N¨ther current associated with the internal symmetry o transformation (8. where we have rearranged the fermionic operators keeping in mind their anti-commuting nature. we note that K µ = 0.37) (or (8.38). as in the case of the complex scalar ﬁeld theory in (7.13) to be µ Jǫ (x) = ǫψ(x)γ µ ψ(x) = ǫJ µ (x).26). (8. using the inﬁnitesimal transformation (8.42) δǫ ψ α = (−iǫψ)α (−iψγ µ )α = ǫψ(x)γ µ ψ(x).9) is invariant under the global phase transformation (8.6) and (8.42) is consistent with the convention of taking left derivatives for the fermion ﬁelds) ∂L ∂L + δǫ ψα ∂∂µ ψα ∂∂µ ψ α (8.37) (or (8.7) that this current is conserved.298 8 Dirac field theory Namely. the Lagrangian density (8.29)) and it can be checked using the equations of motion (8.43) where J µ (x) = ψ(x)γ µ ψ(x). (8. this would correspond to the total probability) .37) of the system is given by (note that in the ﬁrst quantized theory. (8.13) and (7.44) represents the current independent of the parameter of transformation. (8. It is a vector current as in the case of the complex Klein-Gordon ﬁeld theory (see (7.38)).41) On the other hand.38)) is obtained from (6. Thus. ∂µ J µ (x) = 0.27) and we note that the form of the expression in (8. In this case.45) The conserved charge associated with the symmetry transformation (8. we obtain (see (6.

s)c(k′ .33) and (7. s′ )d(k′ . But. .38)). lepton number.34)) Q|0 Q|k. = 1 s′ =± 2 d3 k′ c† (k′ . s′ ) c(k′ . we can express (8.46) We note here that the charge (8.23). etc) which are associated with U (1) symmetry transformations as well.47) As in the case of the complex Klein-Gordon ﬁeld. s′ ). s)d(k. (8. s)c(k. s′ ) c† (k. (8.46) is associated with an Abelian symmetry transformation (8. s) − d† (k. s = 0. s′ ) |0 = 1 s′ =± 2 d3 k′ c† (k′ . s′ )δss′ δ3 (k′ − k)|0 = c† (k. s)|0 = 1 s′ =± 2 d3 k′ c† (k′ .8.4 Charge operator 299 Q= d3 x J 0 (x) = d3 x ψ † (x)ψ(x). s . s′ )c(k′ . it is also possible to associate the charge operator with a fermion number (such as baryon number.46) in terms of creation and annihilation operators as : Q := s=± 1 2 d3 k c† (k. s) + − c† (k.37) (or (8. s′ ) − d† (k′ . Using the ﬁeld decomposition in (8. s)|0 = |k. we can now show that the charge operator (normal ordered) satisﬁes (see (7. c† (k. Therefore. since we are dealing with a fermion system. s) . as in the case of the complex Klein-Gordon ﬁeld we can associate the charge operator with that of the electric charge.

s′ ) 1 s′ =± 2 d(k′ . d† (k.300 Q|k. s′ ) d† (k. s) ↔ d(k. s′ ). s)d(k′ .48) 1 s′ =± 2 = −d† (k. as we have also seen in the case of the complex Klein-Gordon ﬁeld theory. which are degenerate in mass and are created by the operators c† (k. Namely. s′ ) |0 = − d3 k′ d† (k′ . s′ )d(k′ . the vacuum state of the theory is charge neutral while the two distinct one particle states. carry opposite charge. 8. s)|0 = −|k. s) respectively. s′ )δss′ δ3 (k − k′ )|0 (8. s) and d† (k. the vacuum does not contain any particle as we would intuitively expect (compare this with the ﬁrst quantized description of the vacuum in section 2.5 Green’s functions As we have discussed within the context of the Klein-Gordon theory. s′ )c(k′ . s .37)) c(k. s = 1 s′ =± 2 8 Dirac field theory d3 k′ c† (k′ . s)|0 = − d3 k′ d† (k′ . Thus. in the second quantized description of the Dirac theory. This shows that. s) + − d† (k. we can deﬁne the Green’s function for the Dirac ﬁeld theory as satisfying the Dirac equation with a delta function potential (source) as . we can identify them with the particle and the anti-particle states of the theory with positive energy. the second quantized description of the Dirac theory is symmetric under the interchange of particles and anti-particles. (8.5) and that the (normal ordered) Hamiltonian in (8.34) is symmetric under the discrete transformation (see also (7.49) namely. s). s′ ) − d† (k′ .

/ so that (8.117) we know that the Green’s function for the Klein-Gordon equation satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). / (8.). but it is the trivial 4 × 4 identity matrix (which we do not write explicitly). 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). Let us recall the identity (i∂ − m) (i∂ + m) = (i∂ )2 − m2 / / / = − 1 µ ν γ .51) = − ∂µ ∂ µ + m2 . The right-hand side is also a matrix.8.52) S(x − y) = (i∂ + m) G(x − y). etc. Feynman. From these then. we can identify (8. we do not have to calculate the Green’s function for the Dirac ﬁeld theory separately.54) Therefore.50) Here. (8. from (5. S(x − y) is a 4 × 4 matrix function in the spinor space. advanced. Furthermore.5 Green’s functions 301 (i∂ − m) S(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. we would concentrate mainly on the Feynman Green’s function since that is the most useful .53) holds for any Green’s function of the theory (retarded.

55) SF (k) = (k + m) GF (k) = lim / which can also be written as 1 .140)) k+m / . / so that we can write. 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 . (see (5.143) and (5. 3 (2π) k0 (8.146)) .53) we can also obtain the positive and the negative energy Green’s functions for the Dirac ﬁeld theory as (see (5. We note that the relation (8.53) can be written in the momentum space as S(k) = (k + m) G(k).58) We can express the Feynman Green’s function in terms of the positive and the negative energy Green’s functions as well (see (5.145) √ where k0 = Ek = k2 + m2 > 0) S (+) (x) = (i∂ + m) G(+) (x) / = (i∂ + m) / = i 2 i 2 d3 k 1 −ik·x e (2π)3 k0 d3 k k + m −ik·x / e .56) SF (k) = lim ǫ→0+ (8.57) From (8.302 8 Dirac field theory in the calculation of S-matrix elements. for example. k2 − m2 + iǫ (8. k − m + iǫ / ǫ→0+ (8.

62) (+) d3 k m e−ik·x d(k. 8. δ(x0 )(G(+) (x) + G(−) (x)) = 0). s). s).8.61) (2π)3 k0 (−) ψα (x) = 1 s=± 2 d3 k Similarly.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). (2π)3 k0 m eik·x d† (k. s)uα (k.60) where (+) ψα (x) = 1 s=± 2 d3 k m e−ik·x c(k. s)uα (k.8) also as ψ α (x) = ψ α + ψ α (x). other Green’s functions for the Dirac ﬁeld theory can also be easily obtained from the corresponding functions for the KleinGordon ﬁeld theory. s).59) where we have used the fact that the terms involving derivatives of the step function cancel out (namely.63) (2π)3 k0 ψ α (x) = 1 s=± 2 (−) d3 k . Similarly. we can decompose the adjoint ﬁeld operator (8. where ψ α (x) = 1 s=± 2 (+) (−) (8. s). (2π)3 k0 m eik·x c† (k. s)v α (k. (8. (8. (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). s)vα (k. (8.

s′ = 1 s. s) in (8. s)uβ k′ . s′ δss′ δ3 (k − k′ ) = 1 s=± 2 d3 k m −ik·(x−y) e uα (k. we have d3 kd3 k′ (2π)3 (−) ψα (x). s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 . s′ = s. Similarly.58). (8.106) as well as the identiﬁcation in (8.s′ =± 2 ′ d† (k.s′ =± 1 2 ′ c(k.s′ =± 1 2 m2 k0 k′ 0 + × e−ik·x+ik ·y uα (k. 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). c† (k′ .304 8 Dirac field theory It is clear from the basic anti-commutation relations for the creation and the annihilation operators c(k. c† (k. s)v β k′ . s)uβ k′ . s). ψ β (y) (+) + = s. s)uβ (k. s). s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 × e−ik·x+ik ·y uα (k. d(k. ψ β (y) (−) + ′ = s.s′ =± 1 2 m2 k0 k′ 0 + × eik·x−ik ·y vα (k.64) where we have used the completeness relation in (3. s) and d† (k. s). s). d(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).

ψ β (y) (−) (+) + = −iSαβ (x − y) − iSαβ (x − y) = −i S (+) (x − y) + S (−) (x − y) = −iSαβ (x − y).7 Normal ordered and time ordered products 305 × eik·x−ik ·y vα (k.63) that the positive and the negative energy parts of the ﬁeld operators contain annihilation and creation operators respectively) .136) (see also (5.6.65) ψα (x). since fermionic ﬁeld operators anti-commute (Grassmann variables).109). s′ δss′ δ3 (k − k′ ) = 1 s=± 2 ′ d3 k m ik·(x−y) e vα (k. s)v β (k. by deﬁnition the normal ordered product of two ﬁeld operators yields (note from (8. we can write (+) = ψα (x).8. ψ β (y) + + (−) + ψα (x). αβ (8. Thus. in rearranging terms to bring them to a particular form.9.66) 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). As a result.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 6.5 and 6. where we have used (3.160)) 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. However. s)v β k′ .61) and (8. ψ β (y) (+) (−) (8. 8. we need to be careful about negative signs that can arise from changing the order of any two such operators.

69) (8. Similarly.70) (8.65). unlike the bosonic case. it is easy to show that ψ β (y)ψα (x) = : ψ β (y)ψα (x) : −iSαβ (x − y).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) :. (+) (+) (+) (+) (+) (8.68) as ψα (x)ψ β (y) = : ψα (x)ψ β (y) : −iSαβ (x − y).64). In other words. we can write (8.68) where in the last step we have made the identiﬁcation in (8.67). (+) (8.67) and we note that.71) . As a result. 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). using (8. (+) (−) (+) (8. ψ β (y) (−) + (+) (−) (+) (−) (+) (−) (+) (−) (+) (−) = ψα (x)ψ β (y) + iSαβ (x − y). From the deﬁnition in (8. here the order of factors inside normal ordering can lead to additional signs. we see that we can denote 0|ψα (x)ψ β (y)|0 = −iSαβ (x − y) = ψα (x)ψ β (y).

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 6.59) and which can be compared with (6.74) Once again.αβ (x − y).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 .8. 0|T ψα (x)ψ β (y) |0 = iSF .αβ (x − y) : ψα (x)ψ β (y) : + iSF .αβ (x − y) = ψα (x)ψ β (y).7 Normal ordered and time ordered products 307 which can be compared with (6.5 which we will not repeat here.73) (+) (−) (−) (+) where we have used the deﬁnition in (8. .6. The time ordered product of Dirac ﬁeld operators is again deﬁned with time ordering of the operators from left to right (largest time on the left) keeping in mind the appropriate negative signs associated with rearranging fermionic operators as T (ψα (x)ψ β (y)) = θ(x0 − y 0 )ψα (x)ψ β (y) − θ(y 0 − x0 )ψ β (y)ψα (x).104). (8.73) that the Feynman Green’s function is related to the vacuum expectation value of the time ordered product of ψα (x) and ψ β (y). (8. (8.81). It follows now from (8. we can develop Wick’s theorem for time ordered products of Dirac ﬁelds in a straightforward manner as discussed in section 6.

77) to obtain (here k0 = |k|) . the free Dirac Lagrangian density for a massless spinor ﬁeld naturally decomposes as / / / L = iψ∂ ψ = iψ R ∂ ψR + iψ L ∂ ψL . in this section we will discuss brieﬂy the quantization of such ﬁelds. for example.75) As a result. we can expand the ﬁeld as (see (8.77) where.8 Massless Dirac ﬁelds 8 Dirac field theory As we have seen in section 3. it is most natural to decompose it into its chirality states. in the case of the standard model in 14.7.308 8. we can also decompose a Dirac ﬁeld into a left-handed and a right-handed component as 1 (½ + γ5 )ψ(x). (8. when the Dirac particle is massless.77) is diﬀerent from those in (8.3). Unlike in (8.76) Such massless left-handed and right-handed Dirac ﬁelds (Weyl ﬁelds) arise naturally in physical theories (as we will see. The discussion for the left-handed ﬁeld can be carried out in an analogous manner.170) we can invert the decomposition in (8. Using the deﬁnition of the chiral spinors in section 3.53) and (2.23) because we are using here the normalization (2. Let us consider the ﬁeld quantization of a right-handed Dirac ﬁeld (Weyl ﬁeld). Using the orthonormality relations for the right-handed spinors described in (3.164)). we note here that there is no sum over spin polarization because the chiral spinors are charaterized by their unique chirality or helicity (see. for example. the discussion following (3. Therefore. Correspondingly. 2 1 (½ − γ5 ). (8.7. the multiplicative factors in (8. 2 ψR (x) = ψL (x) = (8. we have identiﬁed k0 = Ek = |k|. Furthermore.23).54) which is appropriate for massless spinors.23)) d3 k (2π)3 |k| ψR (x) = e−ik·x c(k)uR (k) + eik·x d† (k)vR (k) . as before.

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

83) Here we have used (8.77) and using the properties of the vacuum in (8.81) we obtain (we assume k0 = |k| in the ﬁeld decomposition) iSF . (8. We recall that the Feynman propagator is deﬁned as the time ordered product (see. Dropping the spinor indices and using (3.82) Putting in the ﬁeld decomposition in (8.84) . (8. for example.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 .172) this can also be written as d3 k 1 iSF .74)) iSF .310 8 Dirac field theory As an example. (8.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 ).80) in evaluating the vacuum expectation values.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) . let us determine the propagator for a righthanded fermion. (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 .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 ) / . 2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 (8.8 Massless Dirac fields 311 Let us evaluate the two terms in (8.84) separately.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. Using the integral representation for the theta function (see (6.

R (k) = iSF .84) we obtain d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) / e 4 (2π) 4k0 × = k0 1 1 + 0 − |k| + iǫ k + |k| − iǫ (8. 2(k2 + iǫ) k + iǫ (8. 2(k2 + iǫ) k + iǫ iSF .86) Therefore.87) iSF .R (k) = (8. adding the two results in (8.R (x − y) = / d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) e .312 = − / dk0 k e−ik (x −y ) . 2πi 4k0 k0 + |k| − iǫ 0 0 0 8 Dirac field theory (8. (8.88) by simply multiplying a factor of γ 0 on the right which leads to iS F . For .82) as † the time ordered product of ψR ψR and not as the conventional time ordered product of ψR ψ R (see.9 Yukawa interaction Let us next consider the theory which describes the interaction between a Dirac ﬁeld and a charge neutral Klein-Gordon ﬁeld.85) and (8. However.88) We note here that we have deﬁned the propagator in (8. everything carries over in a straightforward manner to left-handed fermion ﬁelds.86) and substituting into (8.74)). 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 .R (k)γ 0 = ik(½ − γ5 ) / ikPL / = 2 . for example. 8.89) We note that although our discussion has been within the context of right-handed fermion ﬁelds. the conventional covariant propagator (deﬁned as the time ordered product of ψR ψ R ) can be obtained from (8. The reason for this is that the adjoint ﬁeld ψ R has negative chirality.

90) into a free part and an interaction part. However.) 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. L0 = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 λ LI = −g ψψφ − φ4 .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. The Lorentz invariant Lagrangian density. (8. the correct parity invariant trilinear interaction for a pseudoscalar meson should be (−gψγ5 ψφ). the π 0 meson) has mass M . we are going to ignore this aspect of the meson and also switch oﬀ the self-interactions of the meson ﬁeld for simplicity.8.92). we can denote the interaction Lagrangian density as (remember everything is normal ordered) . As in (6. we can separate the Lagrangian density (8. ψψ is invariant under a Lorentz transformation and φ is a scalar ﬁeld) and is known as the Yukawa interaction. The trilinear coupling gψψφ is manifestly Lorentz invariant (as we have seen in (3.9 Yukawa interaction 313 example. where 1 M2 2 / φ .. such a theory can represent the interaction between protons (if assumed to be fundamental) and the π 0 -meson as well as the self-interaction of the mesons. has the form 1 λ M2 2 / φ − g ψψφ − φ4 . Namely.g. since it leads to the Yukawa potential in the Born approximation and the coupling constant g represents the strength of the interaction. L = L0 + LI . 4! (8.36). since here we are only interested in developing calculational methods in a quantum ﬁeld theory.91) (8. in this case. (The self-interaction can be switched oﬀ at the lowest order by setting λ = 0.86).

6). Using Wick’s theorem (see sections 6.93) 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. we can write (8.95) .314 8 Dirac field theory As a result. (8.5 and 6.94) +··· .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′ ) + ······ . (8. the adiabatic switch-oﬀ is assumed) S = T e−i R∞ R −∞ LI = −g : ψψφ := −HI .69). in this theory the S-matrix will have the expansion (see (6.

a nontrivial matrix element would result for the case 0|S1 |k.96) Each term in this expression leads to a distinct physical process. This can be written out in detail as S (1) = −ig (−) d4 x (+) (−) ψ α (x)ψα (x) − ψα (x)ψ α (x) (−) (+) (+) (+) (−) + ψ α (x)ψα (x) + ψ α (x)ψα (x) φ(+) (x) (+) (−) + φ(−) (x) ψ α (x)ψα (x) − ψα (x)ψ α (x) (+) (−) + ψ α (x)ψα (x) + ψ α (x)ψα (x) . let us look at the ﬁrst term in (8. s. just to get a feeling for what is involved.98) we substitute the ﬁeld decomposition for each of the positive energy ﬁeld operators and write . (+) (8.98) where the initial state contains a proton with momentum k and spin s. (There can be other nontrivial matrix elements where the initial state has one more proton. (1) (8. (−) (−) (+) (+) (8.97) contains an annihilation operator and. k′ . let us look at the lowest order (order g) term in the expansion (8.8.) To evaluate the matrix element (8. s′ .9 Yukawa interaction 315 To see how perturbation theory works in the interacting quantum ﬁeld theory. an anti-proton with momentum k′ and spin s′ and a π 0 meson with momentum q and the ﬁnal state contains no particles. q .97) Each positive energy operator in (8.96) S1 = −ig (1) (+) d4 x ψ α (x)ψα (x)φ(+) (x). For example.95). anti-proton and π 0 meson than the ﬁnal state. therefore.

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

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. k3 . (2π)4 GF x − x = (8. s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x )|k1 . s4 )v β (k2 . the matrix element will have the form k4 . s4 . k2 . s1 . k3 .9 Yukawa interaction 317 term in the expansion at order g2 in (8. s4 . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x ) (−) + ψ α (x )ψα (x )ψ β (x)ψβ (x)|k1 . s4 . In such a case.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 ) : . s1 . s1 ).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.102) . s1 .120)) d4 q −iq·(x−x ) e GF (q). s2 )uβ (k1 . (8. s2 (−) (+) (+) = −ig2 × d4 xd4 x GF (x − x ) (−) (+) (+) (−) k4 . k2 .8.101) × uα (k3 . Recalling that the Fourier transformation of the Green’s function is given by (see (5. s3 |S2 |k1 . k2 . s3 )vα (k4 . k3 . s2 = − ig2 2 d4 xd4 x GF (x − x ) (−) (+) (+) (2) (−) × k4 .

s3 )vα (k4 . s4 ) v β (k2 .101) can be written as (2) 8 Dirac field theory k4 . s1 . s3 )vα (k4 . 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 . s2 )uβ (k1 . The delta function merely ensures the overall energy momentum conservation in the process. k2 . s4 )iGF (k1 + k2 )v β (k2 . s2 )uβ (k1 . s3 )vα (k4 . s1 ).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. 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. k3 .103) × uα (k3 . 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. s4 )v β (k2 . Instead. 8. It is clear from looking at the expansion of the S-matrix that it contains terms . Feynman developed a graphical method for evaluating these matrix elements which is in one to one correspondence with the Wick expansion. s4 .318 the matrix element (8. s3 |S2 |k1 . s2 )uβ (k1 . 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 .

namely.αβ (k) = ǫ→0+ k2 − m2 + iǫ k = lim ǫ→0+ i k − m + iǫ / . The propagators are completely independent of the initial and the ﬁnal states and. − M 2 + iǫ β i (k + m)αβ / lim α = iSF . (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. on the other hand. Thus.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. On the other hand. Therefore. k = iGF (k) = lim ǫ→0 k 2 i . The normal ordered products have to give nonvanishing matrix elements between the initial and the ﬁnal state and. Let us deﬁne the two kinds of Feynman propagators that are possible for our theory (the Yukawa theory).) It should also be remembered that the fermion propagator is a 4×4 matrix with α and β representing the (Dirac) matrix indices.104) Several things are to be noted here. simply because we have a charge neutral scalar ﬁeld . an external boson line can represent either the annihilation or the creation of a particle at the interaction point (vertex). therefore. αβ (8. We note that the scalar (meson) Feynman propagator is an even function and corresponds to that of a charge neutral ﬁeld. the Feynman propagator for the fermions is not an even function and corresponds to that of a (Dirac) ﬁeld carrying charge. the line representing the fermion propagator has a direction associated with it. let us introduce a graphical representation for these two basic elements of Wick’s expansion. For example. as we have seen. Hence. going from from ψ β to ψα .8. are represented by external lines. The initial and the ﬁnal states. occur as internal lines in a diagram. simply give rise to normalization factors as well as spinor functions for the particles in the initial and the ﬁnal states.

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

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

s2 k3 . the corresponding lowest order Feynman diagram will be given by Fig. with the external lines representing physical particles. s4 α β q δ γ k2 . the physical process where a proton. 8.1.109) This is exactly the value of the ﬁrst order S-matrix element which we had evaluated in (8. 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′ .99)) will correspond in the lowest order to evaluating the Feynman diagram in Fig. 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. The value of this diagram can be obtained from the Feynman rules to be 0|S1 |k. s1 k4 . where the external lines represent the respective physical particles that are annihilated. The . s).99). s3 Figure 8. k′ . 8.2.322 8 Dirac field theory In this language. k1 . therefore. an anti-proton and a π 0 meson are annihilated (which we considered in (8. (2π)3 k′ 0 (8. s) (2π)3 k0 m v β (k′ . s′ .2: A second order Feynman diagram. s. s′ )uα (k. q = −(2π)4 igδ4 (k + k′ + q)δαβ × 1 (2π)3 2q 0 m uα (k.

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

114) . (8. (8.113) Therefore. 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| . it follows from energy conservation that 0 0 k1 = k2 . if we are in the center of mass frame where k1 = −k4 . Furthermore. (8. q = (k1 − k2 ) = (0. k2 = −k3 . q 2 − M 2 + iǫ q = k1 − k2 . 0 0 k1 = k4 . 0 0 k2 = k3 . we see that V (q) = V (q) and taking the Fourier transform of (8. 4π |x| = (8.112) then.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. k1 − k2 ).113).

3. 1964. we obtain from (8. New York (1993).11 References 1. Bogoliubov and D. Moscow (1984). New York. Peterson. Row. Roman. Introduction to Relativistic Quantum Field Theory. Introduction to Quantum Field Theory. New York. John Wliley. . Zuber. Schweber. 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. John Wiley. Relativistic Quantum Fields.114) the potential between the fermions to be g2 e−M |x| .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. Itzykson and J-B. 1980. Quantum Field Theory. D. Nauka. 5. N. 4π |x| V (x) = − (8. Relativistic Quantum Mechanics and Field Theory. 2.93) leads to the Yukawa potential in the Born approximation. McGrawHill. Drell. N. 6. McGrawHill. 4. Shirkov. V. C. New York (1969). 8. Introduction to the theory of Quantized Fields. therefore. F.8. Gross. S. P. J. Bjorken and S.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). Evanston (1961).

This page intentionally left blank .

Let us start with the classical Maxwell’s equations in vacuum which are given by (remember c = 1) ∇ · E = 0. To that end. 1 ) representation of the Lorentz group (see sec2 2 tion 4.1) to write 327 . ∂B . (9.1) in a manifestly Lorentz covariant form. we note that we can solve the second and the third equations in (9. Thus. 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. ∂t ∇×E=− ∇ · B = 0. ∂t ∂E ∇×B= .Chapter 9 Maxwell ﬁeld theory 9.1 Maxwell’s equations The next ﬁeld theory to consider logically is that of an Abelian gauge theory and the Maxwell ﬁeld theory is a prototype of such a theory. the dynamical ﬁeld would naturally describe particles with spin 1.2). It is more convenient from our point of view to rewrite Maxwell’s equations (9.1) where E and B represent the electric and the magnetic ﬁelds respectively. this is described by the vector potential Aµ (x) which belongs to the ( 1 . As we know.

B2 . 1. ν = 0. In other words. −A). Similarly. the six independent components of the tensor Fµν are given by (three components each of) the electric and the magnetic ﬁelds. (9. 3. We can. where E = (E1 .2) where A(x) and Φ(x) are known as the vector and the scalar potentials respectively. (9. Therefore.5) µ. of course.3) If we now deﬁne the four dimensional curl of the four vector potential as Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .4) where ǫijk denotes the three dimensional Levi-Civita tensor and B = (B1 . B3 ). Aµ = (Φ. A). (9.6) (9. we note that F0i = ∂0 Ai − ∂i A0 = Ei . combine them into the four vector potential Aµ = (Φ. E = − ∂A − ∇Φ.7) (9. 2. the second .8) denotes the magnetic ﬁeld vector. ∂t (9. Fij = ∂i Aj − ∂j Ai = −ǫijk Bk . denotes the electric ﬁeld vector. E3 ). (9.328 9 Maxwell field theory B = ∇ × A. E2 .

equation (9. Thus.9) leads to ∂µ F µj = 0.1) are given by ∂µ F µν = 0. ∂0 F 0j + ∂i F ij = 0. or. (9.1).1 Maxwell’s equations 329 rank anti-symmetric tensor Fµν is also known as the ﬁeld strength tensor. ∂t ∂Ej = −ǫjik ∂ i Bk . for ν = 0. we see that the set of four Maxwell’s equations (9. ∂Ej + ∂ i (−ǫijk Bk ) = 0. (9. ∂t (9. equation (9. Similarly. ∂t ∂E = ∇ × B. or. or. or. It is now easy to check that the other two equations (the ﬁrst and the fourth) in (9. or. or.9) For example.10) which coincides with the ﬁrst of the equations in (9. ∂ 0 F0j + ∂ i Fij = 0.1).4)) ∂i F i0 = 0.4) as Fµν = ∂µ Aν − ∂ν Aµ .9) takes the form (F 00 = 0 by deﬁnition (9.11) which is the last equation in (9. with the ﬁeld strength tensor deﬁned in (9. (9.9. ∇ · E = 0. for ν = j.12) .13) (9.1) can be written in the manifestly covariant form as ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0.

16) = − ∂µ Aν (∂ µ Aν − ∂ ν Aµ ). (9.14) of the four vector potential Aµ (x). Maxwell’s equations (9.15) In other words. Consequently. as an operator. we note that under a gauge transformation Aµ (x) → A′ (x) = Aµ (x) + ∂µ θ(x). For example. real. the ﬁeld strength tensor is unchanged under the gauge transformation (9.2 Canonical quantization In trying to second quantize the Maxwell ﬁeld theory.14) where α(x) is an arbitrary. local parameter of transformation. 9. But ﬁrst. µ (9. 2 which leads to . δAµ (x) = A′ (x) − Aµ (x) = ∂µ θ(x). let us examine whether there exists a Lagrangian density which will lead to Maxwell’s equations as its Euler-Lagrange equations. Fµν ′ → Fµν = ∂µ A′ − ∂ν A′ µ ν = = ∂µ (Aν + ∂ν θ) − ∂ν (Aµ + ∂µ θ) ∂µ Aν − ∂ν Aµ = Fµν . we have to treat the four vector potential Aµ (x).12) are also invariant under these transformations. the gauge invariance of Maxwell’s equations is obvious in this formulation. µ or.330 9 Maxwell field theory Furthermore. Let us consider the Lagrangian density 1 1 L = − Fµν F µν = − (∂µ Aν − ∂ν Aµ )(∂ µ Aν − ∂ ν Aµ ) 4 4 1 (9. which denotes the dynamical variable of the theory.

9). The Lagrangian density (9. Note that a Lagrangian density 1 1 Fµν F µν − Fµν (∂ µ Aν − ∂ ν Aµ ) . (9. we will work with the simpler Lagrangian density in (9. ∂∂µ Aν (9.19) is equivalent to (9. It is manifestly Lorentz invariant. . = − 1 1 1 (∂ µ Aν − ∂ ν Aµ ) − ∂ µ Aν + ∂ ν Aµ 2 2 2 (9. so that the Euler-Lagrange equations. Fµν = ∂µ Aν − ∂ν Aµ . namely. gives Maxwell’s equations as its Euler-Lagrange equations. ∂L = 0.9.17) = − (∂ µ Aν − ∂ ν Aµ ) = −F µν .19) with both Aµ and Fµν treated as independent dynamical variables would also lead to Maxwell’s equations as its Euler-Lagrange equations. However.20) ∂µ F µν = 0. indeed. ∂µ or. ∂Aν ∂∂µ Aν or.16) when the ﬁeld variable Fµν is eliminated using the (ﬁrst) auxiliary ﬁeld equation in (9. 4 2 L= (9. take the form ∂L ∂L − ∂µ = 0. in this case.16) since the Lagrangian density (9. ∂µ (∂ µ Aν − ∂ ν Aµ ) = ∂µ F µν = 0.20).18) which coincides with the manifestly covariant Maxwell’s equations (9. ∂L = 0.2 Canonical quantization 331 ∂L ∂Aν ∂L ∂∂µ Aν = 0. ∂Fµν or.16).

25) . (9. Πj (y) j = iδi δ3 (x − y). x0 =y 0 . since the ﬁeld strength tensor Fµν is gauge invariant (see (9. Therefore. in this gauge the nontrivial components of the conjugate momentum are given by ˙ Πi (x) = −F 0i = Ei = −(∂ 0 Ai − ∂ i A0 ) = −Ai (x). the momentum conjugate to the ﬁeld variable A0 (x) vanishes. Aj (y)]x0 =y0 Ai (x). for canonical quantization of such a system.24) The naive canonical quantization conditions.21) and the ﬁrst peculiarity of the Maxwell ﬁeld theory is now obvious from (9. Π0 (x) = −F 00 = 0. now must commute with all the ﬁeld variables in the theory. If we treat Aµ (x) as the independent ﬁeld variable. (9. we have to choose a gauge and this particular choice is known as the temporal (axial) gauge. 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. (9.332 9 Maxwell field theory Furthermore. This is a consequence of the gauge invariance of the theory. Πj (y) x0 =y 0 = 0 = Πi (x).15)).23) This is equivalent to saying that. As a result. we note that A0 (x) which would have a nontrivial commutation relation only with Π0 (x). in this gauge. (9.21).17)) Πµ (x) = ∂L = −F 0µ .14) of the vector potential. this Lagrangian density is also invariant under the gauge transformation (9. Furthermore. 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. would then appear to be [Ai (x). ˙ ∂ Aµ (x) (9.22) In other words. namely. we can deﬁne the associated conjugate momentum as (see (9.

in the gauge (9.26) Namely. Πj (y) or. or. in this gauge.29) which does not satisfy the transversality condition (9.23). (As a parenthetical remark. lead to (for ν = 0) ∂µ (∂ µ A0 − ∂ 0 Aµ ) = 0. ∂ 0 (∇ · A) = 0. ∂ 0 (∂µ Aµ ) = 0. Aj (y) j = iδi δ3 (x − y).23). For example. x0 =y 0 (9.27) can be thought of as the gauge choice for this residual invariance. ˙ Ai (x). x0 =y 0 x0 =y 0 (9. The proper quantization condition consistent with the transversality condition (9.27). Aj (y) j = −iδi TR (x − y). ¯ A(x) → A′ (x) = A(x) + ∇θ(x).2 Canonical quantization 333 However. namely. ∇ · A = 0.27) (9. the theory can be checked to have a residual gauge invariance under a time independent gauge transformation which preserves (9. (9. (9. j = −iδi δ3 (x − y). the dynamical equations (see (9.28) and the condition (9. the canonical quantization condition (9.25) leads to Ai (x). the dynamical ﬁeld variable A(x) must be transverse.18)) ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0.9. or.) On the other hand. or.30) . this is not quite consistent which can be seen as follows. we note that even after imposing the gauge condition (9.23).27) is given by ˙ Ai (x).

∇2 (9. there exists a systematic procedure for deriving the correct quantization conditions for such systems. ∇ and (9.31) so that (remember that ∂ i ∂i = −∇2 ) j j ∂ i δi TR = ∂ i δi + ∂i 1 j 2 ∂ ∇ (9. Let us note here that in momentum space.27) constitute constraints on the dynamics of the theory and systems with constraints are known as constrained systems. which we will discuss in the next chapter.33) In the gauge (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 .27). the Hamiltonian density for the Maxwell ﬁeld theory can be obtained to be . the transverse delta function in (9.32) = ∂ j + ∂ i ∂i 1 j j j 2 ∂ = ∂ − ∂ = 0.30) 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).23).23) subject to (9. As we have pointed out in connection with the discussions in the Dirac ﬁeld theory. the correct quantization condition (Dirac bracket) that results in the gauge (9. Here 1 2 stands for the inverse of the Laplacian operator and for∇ mally denotes the Green’s function for ∇2 .22).27) is given by (9.23) and (9. known as the Dirac method. (2π)3 (9. (9. We note here that relations such as (9.30) is consistent with the transversality condition in (9. When quantized systematically through the Dirac method.30).

35) H= Indeed this is what we know to be the Hamiltonian (energy) for the Maxwell theory even from studies in classical electrodynamics.18) take the form ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = Aν = 0. 2 (9. ∇ · A = 0. (9.27)) A0 = 0.9. (9. 9.36) which also lead to ∂ µ Aµ = 0.38) (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 .34) and leads to the Hamiltonian d3 x H = = 1 2 1 2 d3 x Π2 + B2 d3 x E2 + B2 .3 Field decomposition In the temporal gauge (9. Maxwell’s equations (9. (9.23) (see also (9.37) .

(9.336 9 Maxwell field theory As we have seen. the three components of the vector potential. .38) vanishes identically (because of the choice of the gauge condition (9. with k0 = Ek = √ k2 = |k |.43) Namely. λ′ ) = δλλ′ . On the other hand. (9. λ) · ǫ(k.23)) A0 ≡ 0. or. 2. the wave equation (9. k · ǫ (k) = 0. for ν = 0. we must have ∇ · A(x) = 0. complex satisfying k · ǫ (k. There can be two such independent directions. (9. therefore. in this case. (9. 9. for ν = i. since the vector potential is transverse. We note from (9. as shown in Fig.42) (9. λ) = 0. for example. We can choose the two independent polarization vectors to be orthonormal and.44) ǫ∗ (k.39) Thus. in general. λ = 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. will have plane wave solutions of the form A(x) ∝ ǫ(k) e∓ik·x .40) (9. satisfy the Klein-Gordon equation for a massless particle (wave equation) and.27) that. the polarization vector characterizing the vector potential must be transverse to the direction of propagation of the plane wave.1. we have Ai = 0.

3 Field decomposition 337 k ǫ(k. λ) . λ) (9. 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. the ﬁeld operator A(x) is Hermitian as it should be. 2) ǫ(k. 1) Figure 9. λ) eik·x a† (k.9.44) (since they deﬁne a basis in the plane transverse to the direction of propagation). 2) transverse to the direction of propagation k. The vector nature of the ﬁeld variable is now contained completely in the polarization vectors.46) Note that a(k. therefore. This discussion is. λ) e−ik·x a(k. Given this.1: Two independent and orthogonal polarization vectors ǫ(k. λ) and a† (k. λ) are operators and as we have deﬁned. (9. 1) and ǫ(k.45) +ǫ∗ (k. 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. where k0 = Ek = |k |. quite analogous to that of the neutral Klein-Gordon ﬁeld except for the presence of the polarization vectors. Let us next note that .

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) .50) . a† (k. λ)a(k. λ)e2ik t a† (−k. λ) = 1 0 √ ǫ(k. λ) = λ=1 d3 k ′ √ δ3 (k − k′ )ǫ(k′ . λ). λ) = i ← → (eik·x ∂0 A(x)) · ǫ∗ (k.48) 2 =−i λ=1 From (9.47) 2k0 λ=1 Similarly. λ)a(k. λ) + ǫ∗ (−k.49) where we have used the normalization for the polarization vectors in (9. λ) + ǫ∗ (k′ . λ). λ) . (9.49) leads to d3 x (2π)3 2k0 ← → (e−ik·x ∂0 A(x)) · ǫ(k. λ). λ) = −i (9. λ)a(k′ . λ) 2k′ 0 0 +k ′ 0 )t +δ3 (k + k′ )ǫ∗ (k′ . λ)e2ik t a† (−k. λ)e 2 a(k′ . λ=1 or. 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. (9.48). a(k. λ)ei(k 2 a† (k′ . Taking the Hermitian conjugate of (9. we can show that d3 x (2π)3 2 ˙ eik·x A(x) 0 k0 ǫ(k. λ)ei(k +k)·x a† (k′ .47) and (9. (9.44). λ)−ǫ∗ (−k.

λ′ )]. (9.52) In other words.49) and (9.54) The Hilbert space for this theory can be built up in the standard manner. ˙ ˙ = 0 = [Ai (x).61) in the case of the Klein-Gordon ﬁeld). P = λ=1 d3 k k a† (k. The Hamiltonian and the momentum operators for the Maxwell theory can again be expressed in terms of the creation and the annihilation operators. λ′ )] = δλλ′ δ3 (k − k′ ).3 Field decomposition 339 Equations (9. a† (k′ . λ) behave exactly like the annihilation and the creation operators for a harmonic oscillator system. If we denote the vacuum state of the theory by |0 . Aj (y)]x0 =y0 (9. λ).53) where we recall from the Einstein relation (see (9. λ). we can easily show that [a(k.30)).49) and (9. Given the quantization conditions (see (9.9. [Ai (x).50) denote the two inversion formulae for the Maxwell ﬁeld (analogous to (5. (9.50). [a(k. λ)a(k. j = −iδi TR (x − y).51) using (9. Aj (y)]x0 =y0 . The normal ordered forms for these operators can be shown to correspond to (see (5. λ). λ). λ).97) and we do not show the normal ordering symbol explicitly) 2 H = λ=1 2 d3 k Ek a† (k. the operators a(k. λ) and a† (k. (9.46)) that Ek = |k|. λ)a(k. Aj (y)]x0 =y0 ˙ [Ai (x). a† (k′ . then it satisﬁes . λ′ )] = 0 = [a† (k. a(k′ .77) and (5.

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

R L R L where we have used the familiar identiﬁcation that 1 ˆ ǫL (k) = √ ex − iˆy . λ)a† (k. 1) + ia† (k. 1) as creating a photon polarized along the x-axis and a† (k. 1) − ia† (k. 1) = ex .9. 1) + ey a† (k. 2) = ey .61) = ǫ∗ (k)a† (k) + ǫ∗ (k)a† (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. e 2 (9. 1) + ia† (k.) A simple way to see that these would represent the creation operators for left and right circularly polarized photons is to note that 2 ˆ ˆ ǫ∗ (k. λ) = ex a† (k. . e 2 1 ˆ ǫR (k) = √ ex + iˆy . 2)) 2 (9. 2) . It is clear. in this language. 2) as creating a photon polarized along the y-axis. 2 1 √ a† (k. that the operators a† (k) = L a† (k) = R 1 √ a† (k.62) describe respectively the left and the right circular polarization vectors. 2) λ=1 = 1 (ˆx + iˆy )(a† (k. 2) . 2)) e e 2 1 e e + (ˆx − iˆy )(a† (k. (9. ˆ ǫ(k.59) then. 2 (9.60) would then create photons which are left and right circularly polarized respectively. 1) − ia† (k. we can identify a† (k. (These one photon states are eigenstates of helicity with eigenvalue ±1. 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.

2 as four vectors ǫµ (k. a† (k′ . ǫ(k. λ). λ) = (0. λ)ǫν (k′ . λ)e−ik·x a(k.52). λ) in (9.342 9.23)) 2 Aµ (x) = λ=1 d3 k (2π)3 2k0 ǫµ (k. λ) and a† (k. 3 2k 0 (2π) d3 k (2π)3 2k0 ǫµ (k. it follows that the only nontrivial covariant commutation relations have the form (these are not at equal time) 2 [A(+) (x). (9. (9.63) and furthermore. λ)ǫν (k′ . λ)(e−ik·x a(k. for simplicity. λ′ )e−ik·x+ik ·y δλλ′ δ3 (k − k′ ) . λ′ )] 2 = λ.64) with k0 = |k|. if we choose the polarization vectors to be real as in (9. λ)eik·x a† (k. λ′ )e−ik·x+ik ·y [a(k.27)). if we deﬁne the two polarization vectors for λ = 1. λ)). λ)). λ)+eik·x a† (k. A(−) (y)] µ ν = λ. λ). λ).65) A(−) (x) = µ λ=1 From the commutation relations for a(k. 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.59) then we can write the ﬁeld decomposition (9.45) as (recall (9. (9.λ′ =1 1 (2π)3 ′ d3 k d3 k ′ √ √ 2k0 2k′ 0 ×ǫµ (k.4 Photon propagator 9 Maxwell field theory Let us note that in the temporal gauge (9.23) (see also (9.λ′ =1 1 (2π)3 d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ ×ǫµ (k.

we obtain : Aµ (x)Aν (y) : +[A(+) (x). λ)ǫν (k. we can deﬁne the Feynman propagator in the standard manner (see (6. µν [A(−) (x). (9. namely. 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). µν (9.9.145)).4 Photon propagator 2 343 d3 k ǫµ (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. µν (9. .67) Aµ (x)Aν (y) = 0|Aµ (x)Aν (y)|0 = −iG(+) (x − y). λ)ǫν (k. It follows now that Aµ (x)Aν (y) = (A(+) (x) + A(−) (x))(A(+) (y) + A(−) (y)) µ µ ν ν = A(+) (x)A(+) (y) + A(+) (x)A(−) (y) µ ν µ ν (−) +A(−) (x)A(+) (y) + A(−) (x)Aν (y) µ ν µ = = = Therefore.68) From this. λ) λ=1 = −iG(−) (x − y). A(+) (y)] = −[A(+) (y).143) and (5. λ)ǫν (k. λ) λ=1 = −iG(+) (x − y).107)).66) which are the analogs of the positive and the negative energy Green’s functions for the Maxwell ﬁeld (see (5.

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 ) .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)). (9. for example. (9. To simplify this expression. λ)ǫν (k.69) µν By deﬁnition.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 .(9. therefore.344 9 Maxwell field theory 0|T (Aµ (x)Aν (y)) |0 = iGF .70) 0 2k where we have changed k → −k in the second term and have used the fact that the polarization vectors are unaﬀected by this change of variables (see. λ)ǫν (k.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. we use the integral representation for the step function given in (6. the Feynman Green’s function has the form GF . 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.59)).

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 ) . λ) . Substituting this into (9. λ) λ=1 = GF (k) λ=1 ǫµ (k.µν (x − y) = lim ǫ→0+ d4 k e−ik·(x−y) (2π)4 k2 + iǫ 2 ǫµ (k. λ) . 0. |k| (9.72) Therefore.74) .9. let us note that if we introduce two new four vectors η µ = (1. λ)ǫν (k. λ)ǫν (k.73) depends on the polarization sum and to evaluate the polarization sum.73) where GF (k) denotes the Feynman Green’s function for a massless scalar ﬁeld (see (5. λ=1 (9.70) and denoting the variable of integration k′ 0 → k0 . ˆ kµ = kµ − (k · η)η µ (k · η)2 − k2 = 0.71) where we have used the Einstein relation (9. 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.µν (k) = ǫ→0+ lim 1 k2 + iǫ 2 2 ǫµ (k. (9.140)). The Feynman Green’s function in (9. 0. k .46). we can identify the momentum space Feynman Green’s function for the Maxwell (photon) ﬁeld to be GF . we obtain GF . 0). λ)ǫν (k.

λ)ǫµ (k. λ)) (η ησ ) (kσ kσ ) λ=1 2 2 or.µν (k) = lim − ǫ→0+ k2 (9. is manifestly transverse (see (9. ˆ ˆ kµ kµ = −1. 2 deﬁne an orthonormal basis in four dimensions. λ)ǫν (k.75)).78) and. (9. λ) = 0. we obtain the Feynman propagator for the photon ﬁeld in the temporal gauge to be i ˆ ˆ (ηµν − ηµ ην + kµ kν ). Substituting this back into (9. λ). in this gauge. + iǫ iGF .µν (k) = 0.75) Consequently. physical calculations carried out with this propagator do lead to Lorentz covariant results.346 9 Maxwell field theory ˆ then. namely. − 2 λ=1 ˆ ˆ ǫµ (k. also satisﬁes . η µ ηµ = 1. λ) = 0 = η µ kµ . However. we can write the completeness relation for these basis vectors as ˆ ˆ ηµ ην kµ kν ǫµ (k. kµ GF . they satisfy ǫµ (k. (9.76) or. λ) = −ηµν + ηµ ην − kµ kν . ˆ ˆ (ǫσ (k. λ)ǫν (k. λ)ǫσ (k. kµ and ǫµ (k. η µ . ˆ kµ ǫµ (k. λ) + ηµ ην − kµ kν = ηµν .77) The Feynman propagator has a non-covariant look in this gauge. λ)ǫν (k. ˆ η µ ǫµ (k. λ) + σ + = ηµν . as expected (since we are in the gauge A0 = 0). The Feynman propagator. Namely. λ′ ) = −δλλ′ . λ = 1. λ=1 ˆ ˆ ǫµ (k.73). (9.

81) (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 (7.0µ (k) = 0. 9.83)) and we have separated the Lagrangian density in (9.80) (or (9.83) . 4 = L0 + LI .5 Quantum electrodynamics 347 GF .80) into the free and the interaction parts as 1 / L0 = − Fµν F µν + iψ∂ ψ − mψψ.79) Quantum electrodynamics (QED) is the theory describing the interaction of electrons and positrons with the electromagnetic ﬁeld. 4 LI = −eψγ µ ψAµ .82)) can be checked to be invariant under the ﬁnite local gauge transformations ψ(x) → ψ ′ (x) = e−iθ(x) ψ(x). (9. (9. where Dµ ψ(x) = (∂µ + ieAµ (x))ψ(x). Aµ (x) → A′ (x) = Aµ (x) + µ ′ 1 ∂µ θ(x). which follows from (9. ψ(x) → ψ (x) = ψ(x)eiθ(x) .9.74).5 Quantum electrodynamics (9.82) The total Lagrangian density (9. e (9.

(9. δǫ ψ(x) = ψ (x) − ψ(x) = iψ(x)ǫ(x). Therefore. let us note that the covariant derivative in (9. e (9.84) where ǫ(x) denotes the inﬁnitesimal local parameter of transformation. the minimally coupled Dirac Lagrangian density transforms as ′ LDirac (ψ.83) (or (9.85)) ′ Dµ ψ(x) → Dµ ψ ′ (x) = = ∂µ + ieA′ (x) ψ ′ (x) µ ∂µ + ie Aµ (x) + 1 (∂µ θ(x)) e e−iθ(x) ψ(x) = e−iθ(x) (−i(∂µ θ(x)) + ∂µ + ieAµ (x) + i(∂µ θ(x))) ψ(x) = e−iθ(x) (∂µ + ieAµ (x)) ψ(x) = e−iθ(x) Dµ ψ(x). In fact.84)).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).80).85) It follows from this that under a (ﬁnite) gauge transformation.81) transforms covariantly under a gauge transformation (see (7. δǫ Aµ (x) = A′ (x) − Aµ (x) = µ ′ 1 ∂µ ǫ(x). Aµ ).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. Aµ ) → LDirac (ψ ′ . ψ. we need to examine the invariance of the minimally coupled Dirac Lagrangian density in (9. = iψeiθ(x) γ µ e−iθ(x) Dµ ψ − mψeiθ(x) e−iθ(x) ψ (9. First. A′ ) µ ′ = iψ γ µ Dµ ψ ′ − mψ ψ ′ ′ ′ = iψγ µ Dµ ψ − mψψ = LDirac(ψ. ψ. we have already seen that the Lagrangian density for the Maxwell theory is invariant under the gauge transformation in (9.86) . ψ .

27)).80) r µβ α q p = −(2π)4 ie (γ µ )βα δ4 (p+q +r).104) and in (8.83) and. there is a conserved current. in QED we have the photon propagator as well as the photon external line factor which take the forms. We have already derived the fermion propagator as well as the fermion external line factors in (8.88) The Feynman rules for QED can be derived in the standard manner.77)) correponds to the temporal gauge (9.106)-(8.5 Quantum electrodynamics 349 Namely. In addition.80) is gauge invariant. λ µ = ǫ→0+ (9. k2 + iǫ ǫµ (k. The interaction vertex for QED can also be read out from the Lagrangian density (9.90) (2π)3 2k0 where we have chosen a real polarization vector as in (9. (9.µν (k) = k.23) (see also (9.89) lim − 1 ˆ ˆ i(ηµν − ηµ ην + kµ kν ) . In this theory.91) . ∂µ J µ (x) = 0. (9. therefore.89) (or (9. (9. the minimally coupled Dirac Lagrangian density is also invariant under the gauge transformation (9. J µ (x) = ψγ µ ψ. µ k ν = iGF . (9. We note here that the photon propagator is a gauge dependent quantity. The form of the propagator in (9.9.64) so that the external line in (9. the complete QED Lagrangian density (9.90) can represent both the annihilation as well as the creation of a photon.107). λ).87) In momentum space the current conservation takes the form kµ J µ (k) = 0.

92) To lowest order there are two Feynman diagrams which will contribute to such a process and they are shown in Fig. the S-matrix element for this process can be worked out to be (we are assuming that the ﬁnal state photons are outgoing and have carried out the momentum integration for the internal line and are omitting the overall energy-momentum conserving delta function δ4 (k1 + k2 − k3 − k4 )) k3 . s1 k2 . a) Photon pair production: For example. .6 Physical processes 9 Maxwell field theory With these rules. Therefore. λ′ k4 . λ′ ν µ k3 . 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.350 9.2. As a result. 9.2. 9.2: The two lowest order Feynman diagrams that contribute to the pair production of photons as well as to the Compton scattering. 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. s1 ν + k1 − k4 k2 . λ µ k1 − k3 k1 . We note here that experimentally. 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 ). we only measure two photons in the ﬁnal state. s2 k1 . but not the vertices (points) where each of the two photons was produced. s2 k4 . (9.

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

λ) = − ie2 (2π)2 1 0 2k2 1 0 2k4 m 0 k1 m 0 k3 / × u(k3 . this process is described by the two Feynman diagrams shown in Fig. namely. In this case.94) K = k1 − k4 and we have introduced Q = k1 + k2 . Here. but . s1 ) . λ′ )|S (2) |e− (k1 ). γ(k4 )) to be outgoing. the internal line represents a photon propagator signifying that the electrons scatter by exchanging (emitting and absorbing) a photon.3. Thus. 9. (9. we can obtain the S-matrix element for Compton scattering directly from the results in (9. s1 ) / Q − m + iǫ / 1 ǫ (k4 . 9.93) (or (9. this amplitude will differ from (9.93) to be e− (k3 ). (9.98) To the lowest order. λ) / 1 ǫ (k2 . (9. s3 )ǫ (k2 . The amplitude diﬀers from the case of the pair production of photons by only the external line factors.93). s3 )ǫ (k4 .90) is the same for the creation or the annihilation of a photon).96)) not only in the external line factors.96) / ˜ − m + iǫ K / ˜ where as deﬁned in (9. λ′ )u (k1 . instead of an electron and a positron being annihilated as was the case in (9. Therefore. γ(k4 . we will assume the ﬁnal state electron and photon (e− (k3 ). 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 (9.97) 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(k3 .2 except for the labeling of the momenta.352 9 Maxwell field theory are given to the lowest order by the same diagrams as in Fig. γ(k2 . λ)u(k1 .

s2 ) µ ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . Assuming that the ﬁnal state electrons are outgoing. s3 )γ ν u(k1 .6 Physical processes 353 k3 k4 k4 k3 + k1 − k3 k1 − k4 k1 k2 k1 k2 Figure 9. e− (k2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k4 . s1 ) . s4 )γ ν u(k1 . also in the propagator of the diagram.99) ˜ where K and K are deﬁned in (9. +u(k3 . e− (k4 )|S (2) |e− (k1 ). s1 ) 2 + iǫ K ˆ ˆ ˜ ˜ (ηµν − ηµ ην + Kµ Kν ) u(k4 . s4 )γ µ u(k2 . d) Bhabha scattering: Finally. the S-matrix element takes the form e− (k3 ). carrying out the momentum integration for the internal line and omitting the overall energy-momentum conserving delta function δ4 (k1 + k2 − k3 − k4 ). s2 ) ˜ 2 + iǫ K (9.9. the Bhabha scattering can be thought of as the process . This can be calculated in a straightforward manner as before. s3 )γ u(k2 .94).3: The two lowest order Feynman diagrams for M¨ller scato tering.

The lowest order Feynman diagrams for this process are shown in Fig.354 9 Maxwell field theory e− (k1 ) + e+ (k2 ) → e− (k3 ) + e+ (k4 ). e+ (k2 . e+ (k4 . the second diagram simply describes the scattering of an electron and a positron by exchanging a photon. s2 )γ ν u(k1 . s3 )γ µ v(k4 . We see that topologically these diagrams are the same as in Fig.100) k4 + k1 + k2 k1 k3 k1 k1 − k3 k2 Figure 9. s2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k3 .4: The two lowest order Feynman diagrams for Bhabha scattering. s4 ) +v(k2 .3. s3 ). On the other hand. we have e− (k3 . s4 )|S (2) |e− (k1 . e+ (k4 )). s1 ) 2 + iǫ Q ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . s3 )γ ν u(k1 . s2 )γ µ v(k4 .4. s1 ) . 9. but in the present case we will treat the ﬁnal state electron and the positron as outgoing (e− (k3 ). 9.101) .4 represents the annihilation of an electron and a positron to create a photon which subsequently creates an electron and a positron. the ﬁrst diagram in Fig. 2 + iǫ K (9. s4 ) ˆ ˆ (ηµν − ηµ ην + Qµ Qν ) v(k2 . We can calculate this matrix element as before and omitting the overall delta function representing conservation of energy-momentum. k2 k4 k3 (9. 9. In this case. s1 ).

It follows from (9. This gives a ﬂavor of low order calculations in QED. 9. let us discuss very brieﬂy the Ward-Takahashi identity in QED. As we have seen. independent of the choice of gauge. However. 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.80) for QED has the form 1 ¯ ¯ L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ.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 . −p. (9. of course. 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. which we will study later.7 Ward-Takahashi identity in QED 355 where as deﬁned in (9.102) The structure of the photon propagator in the theory. Here we would only give a very simple derivation of this relation in QED. depends on the choice of gauge.103) where the i factor in the denominator is understood.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. the Lagrangian density (9. 4 (9.7 Ward-Takahashi identity in QED Before we discuss more complicated symmetries.9. 0) (iSF (p)) .94) and (9.104) . e (9.

k = 0) = −(iSF (p))−1 e = ∂ (iSF (p))−1 ∂pµ ∂ −1 S (p). −p. by inverting (9.105) 1 = − (iSF (p))Γµ (p.104) can be diagrammatically represented as i i iγµ p−m / p−m / 1 e = − p k=0 p (9. a diagrammatic representation for the basic identity (9. k = 0). 0) denotes the three point vertex (9.106) In this case.104) in QED resulting from the structure of the fermion propagator has the form ∂ ∂pµ p =− 1 e p k=0 p . −p. (9. e Consequently.107) ∂ (iSF (p)) (iSF (p))−1 ∂pµ = −i which can also be written as i 1 ∂ −1 S (p) = − Γµ (p. −p.104) we can obtain a relation between the fermion two point function and the three point vertex function as 1 Γµ (p. As a result the right-hand side of (9.108) . −p.91) in QED with a zero momentum photon (without the delta function and the factor of (2π)4 ). ∂pµ F (9. µ F ∂p e (9. k = 0)(iSF (p)).356 9 Maxwell field theory where Γµ (p.

we can write k ∂ ∂pµ p k+p p =− 1 e p k p k+p k+p q=0 (9. (The trick in obtaining this relation in a simple manner lies in channeling the external momentum only through the fermion lines.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.108) holds for the one loop corrections to the amplitudes. without the external lines).5: Two loop corrections to the fermion self-energy in QED.106). Once again. Using the graphical identity in (9. . Such a relation can be seen to hold at any higher order in perturbation theory as well. but we understand that the external momentum is channelled only through the fermion lines) + Figure 9.7 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. The diagram on the left-hand side in (9. such relations are called Slavnov-Taylor identities which we will discuss later).) At two loop order the fermion self energy diagrams have the forms shown in Fig.9. 9. let us look at the fermion self-energy at one loop.109) Here we are looking only at the proper vertex parts of the diagrams (namely.5 (there are no external lines and we do not label the momenta for simplicity. we see that (9. For example.

if the fermion self-energy and the vertex function change as .110) We recognize that the graphs in the bracket on the right-hand side of (9. using (9.108) diagrammatically at every order in perturbation theory. (9.110) correspond precisely to the two loop corrections to the vertex function. We note that including quantum corrections (we will study renormalization in a later chapter). A simple and powerful consequence of the Ward-Takahashi identity can be derived as follows.106) at this order we obtain ∂ ∂pµ + =− 1 e + + + + + =− 1 e + + + + + . Thus.358 9 Maxwell field theory Using the basic graphical identity in (9.106) we can easily verify the validity of the identity (9.

m = 0) −→ Z2 SF (p. 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 + · · · . 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. we can identify a = 1 + B = 1 + C. then they must equal those present in the vertex correction graphs at the same order (we will study these issues in more detail within the context of renormalization later). Z2 are constants.114) . then from (9.83). C.7 Ward-Takahashi identity in QED 359 −1 −1 SF (p. (9.108) is a consequence of gauge invariance can be seen heuristically in the following way. m = 0). That the Ward-Takahashi identity (9.112) This physically implies that if there are divergent parts in the fermion self-energy graphs at higher order.111) where Z1 . on the other hand.108) we conclude that these constants must be related as Z1 = Z2 . Γµ −→ Z1 Γµ . (9.85)) iaψD = iaψ(∂ + ieA /ψ / /)ψ.115) (9.9. Thus comparing with (9. (9. Invariance under the local gauge transformations (9. D are constants and the dots denote other structures that may be induced due to quantum corrections.113) −eCψA − Dψψ + · · · /ψ where A.113). (9. B.

9.21) and (9.118) As we have already seen in (9. ˙ ∂ Aµ Πµ = which implies the constraint Π0 = −F 00 = 0. 4 Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .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. (9.117) as well as the action for this theory are invariant under the gauge transformation δAµ = ∂µ α(x). (9.15).116) We will discuss these topics in more detail when we study renormalization chapter 16. As a result. canonical quantization of Maxwell’s theory results in a lack of manifest Lorentz invariance.117) where Fµν denotes the ﬁeld strength tensor. since the ﬁeld strength tensor Fµν does not change under a shift of the vector potential by a gradient. The canonical momenta conjugate to the ﬁeld variables Aµ can be calculated from the Lagrangian density and take the forms (see (9.8 Covariant quantization of the Maxwell theory As we have seen. then the equations of motion (see (9.360 which is another way of saying 9 Maxwell field theory Z2 = 1 + B = 1 + C = Z1 . Let us recapitulate brieﬂy how this arises. We recall that the Lagrangian density for Maxwell’s theory is given by 1 L = − Fµν F µν .22)) ∂L = −F 0µ . the Lagrangian density (9. the present choice of gauge is known as the Coulomb gauge). (9.

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

(9. P = 1 P µν deﬁnes the normalized projection operator.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 .) Furthermore. the inverse of P µν does not exist (this also means that the determinant of P µν vanishes) and the Green’s function and. namely. we see that P µν can be thought of as µν a projection operator. therefore. .122) (9. (9. we cannot carry out calculations in perturbation theory.) We note that we can write the Lagrangian density (9. This implies that even if we can quantize the theory (say. it projects on to the space of the components of any vector transverse (perpendicular) to the gradient operator ∂ µ . in the naive path integral formalism).124) so that this is the transverse projection operator.123) With a suitable normalization. (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. we note that ∂µ P µν = ∂µ (η µν = (∂ ν − − ∂µ∂ν ) ∂ ν ) = 0 = ∂ν P µν .117) for the Maxwell theory also as 1 1 L = − Fµν F µν = Aµ P µν Aν + total derivatives. the Feynman propagator for the theory cannot be deﬁned. (In fact. − ∂µ∂λ (9. 4 2 where P µν = η µν − ∂µ∂ν . Consequently.

126) Here we have generalized Maxwell’s theory to include a conserved 1 current. We can take an alternative approach. the Lagrangian density 1 1 L = − Fµν F µν − (∂µ Aµ )2 + J µ Aµ .125) . Namely. But to understand the issue better. (9. (This formulation of the theory is due to Fermi and considered as a gauge choice. without any further input. since we realize that the diﬃculties in quantization arise because of the singular nature of the Lagrangian density.125) ∂µ J µ = 0. 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. let us look at the equations of motion following from the action in (9.125) would appear to be diﬀerent from Maxwell’s theory. the system is singular and contains constraints among the ﬁeld variables (as we will discuss in the next chapter). Therefore. However. But clearly the theory (9. the Cauchy initial value problem cannot be uniquely solved. this gauge is known as the Feynman-Fermi gauge.) This additional term breaks gauge invariance and consequently leads to a nonsingular theory. in this procedure we lose manifest Lorentz covariance since we single out the transverse degrees of freedom. we can try to modify the theory so as to make it nonsingular. 4 2 where J µ represents a conserved current (9.8 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. Let us consider.9. But more than that we have also added a term − 2 (∂µ Aµ )2 to Maxwell’s Lagrangian density. In such a case. We can solve for the constraints and quantize only the true dynamical degrees of freedom. for example. at this point there is no justiﬁcation for adding this new term to the Lagrangian density. simply because the Green’s function does not exist.

If we now write out the left-hand side explicitly. (9. χ is a free ﬁeld and. this is just the Maxwell’s equations in the presence of conserved sources. the presence of this additional term in the Lagrangian density would not change the physics of Maxwell’s theory. (9. An alternate way to see this is to note that if we take the divergence of the equations of motion in (9.132) .127). ∂µ F µν + ∂ ν (∂ · A) = −J ν . Thus we see that although the presence of the term − 1 χ2 in the 2 Lagrangian density (9. (∂ · A) = ∂ν J ν . Furthermore. or. (9. therefore. or.364 9 Maxwell field theory ∂µ or. we have ∂ν ∂µ F µν + or. t = 0.127) Without the second term on the left-hand side in (9. (9. ∂∂µ Aν ∂Aν −∂µ F µν − ∂ ν (∂ · A) − J ν = 0.129) ∂ · A = −∂ · J = 0.130) where we have used the anti-symmetry of the ﬁeld strength tensor as well as the conservation of J µ . ∂t χ = 0.127) takes the form ∂µ (∂ µ Aν − ∂ ν Aµ ) + ∂ ν (∂ · A) = −J ν . or.125) where χ = ∂ · A. Aν = −J ν .131) seems to modify the theory. (9.128) (9.127). we recognize that if we restrict classically to the initial value conditions ∂χ = 0. ∂L ∂L − = 0. (∂ · A) = 0. (9.

134) It is now obvious that the coeﬃcient matrix of highest derivatives is nonsingular in this case.136) with all other commutators vanishing.137) x0 =y 0 . (9. ˙ ∂ Aµ Πµ = (9.133) (9. ˙ Aµ (x). Aν (y) x0 =y 0 = −iηµν δ3 (x − y).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). (9. 2 (9. We can deﬁne the canonical momenta as ∂L ˙ = −Aµ .8 Covariant quantization of the Maxwell theory 365 then χ = 0 at all times and we recover the familiar Maxwell’s theory. classically we can think of Maxwell’s theory as described by the modiﬁed Lagrangian density (9.125) with the supplementary condition ∂ · A = 0. We now see that this quantization relation is exactly like the quantization condition for four distinct scalar ﬁelds (see (5. φ(y) = iδ3 (x − y).40) and note that we are being slightly sloppy here in the sense that the commutation relation should really involve ﬁeld variables and their conjugate momenta) ˙ φ(x).9. Thus. Πν (y)]x0 =y0 = iδµ δ3 (x − y). 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. or.

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

142) is a very stringent condition and we have to relax this further.133) and say that this condition is true only on the space of physical states of the theory. Thus.142) In other words.133) were to hold as an operator equation. (Note that since all ﬁeld components are dynamical in this modiﬁed theory.140) in the quantum theory.9. we can think of the supplementary condition as selecting out the subspace of the physical Hilbert space of the theory. or. = i∂ ν δ3 (x − x′ ). even (9. It is easy to see that the commutation relation (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. (9.136) are consistent.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.) We note that. We have. the non vanishing right-hand side would lead to an inconsistency.133) may be thought of as picking out the smaller physical subspace from the larger total vector space. such a condition is hard to implement as an operator condition. where Aµ ’s are operators.) As we would see shortly. (9.141) x0 =x′0 ∂µ Aµ (x). namely. Πν (x′ ) x0 =x′0 x0 =x′0 (This relation is essentially correct. (9. whereas the left-hand side of the above expression would be zero if the supplementary condition (9. ∂µ Aµ |phys = 0. however. = i∂ ν δ3 (x − x′ ). not made use of the supplementary condition (9. A little bit of thinking tells us that although in the classical theory we can impose the condition ∂µ Aµ = 0. = iη µν δ3 (x − x′ ). but the derivation needs to be done carefully in a limiting manner x0 → x′0 .133) as yet. Πν (x′ ) ∂µ Aµ (x). The supplementary condition (9. Πν (x′ ) or. we can weaken the supplementary condition (9. We .136) would lead to Aµ (x).

(Unlike the earlier discussion. a† (k′ .146) ǫµ (k. At this point we can quantize the coeﬃcients a(k. µ d3 k ǫµ (k. λ′ ) = η λλ = ′ As we have seen earlier (see (9. 3. λ)ǫ∗µ (k. 2. (9. (9. a(k′ . λ) and a† (k. 1. . λ = λ′ = i = 1.368 9 Maxwell field theory can see this already from the commutator relation (9. Πν (x′ ) |ψ or. Basically. a† (k′ .147) = −η λλ δ3 (k − k′ ). λ)’s are the components of the polarization vector for a photon travelling along k and polarization index λ = 0. λ′ ) .74)). we note that if |ψ represents a physical state. let us make a plane wave expansion for the ﬁeld variables Aµ (x) = λ ∗ + ǫ∗ (k. λ).144) where fk (x) = 1 (2π)3 2ω e−ik·x . here they are not required to be transverse. then ψ| ∂ · A(x). λ = λ′ = 0. λ) . λ′ ) = 0 = ′ a† (k. λ).143) which is inconsistent. ˆ while that for λ = 3 is longitudinal (recall kµ ) and the other two polarization vectors are transverse to the momentum four vector. (9. λ). 0 = i∂ ν δ3 (x − x′ ). we can always choose a basis such that the polarization vector for λ = 0 is time-like (recall η µ ). (9. λ)fk (x)a† (k.) They are normalized as 1. λ) as annihilation and creation operators satisfying a(k. k0 = ω(k) = |k|.141). λ′ ) a(k. To proceed further. 3. λ)fk (x)a(k. λ) (9. x0 =x′ 0 = ψ|i∂ ν δ3 (x − x′ )|ψ . 2.145) Here ǫµ (k. −1.

λ′ )fk′ (y)a† (k′ . 1. we have to check that these conditions are consistent with the equal-time quantization conditions for the ﬁeld variables (for simplicity. a(k′ . λ) (9. λ)ǫ∗σ (k. µ = λ. λ′ ) ν ∗ −ǫ∗ (k. λ)ǫ∗σ (k. λ′ )fk (x)fk′ (y) ν µ = λ d3 k(−iω) ǫµ (k.148). λ)ǫ∗ (k. λ′ ) − ǫ∗ (k′ . λ′ ) ν ∗ d3 kd3 k′ ǫµ (k.λ′ ∗ −iω ′ ǫν (k′ . λ)ǫν (k′ . λ) ν ∗ fk (x)fk (y) ǫσ (k. λ′ )fk (x)fk′ (y) a(k. λ)ǫ∗ (k′ . λ) k (9.λ′ d3 kd3 k′ (−iω ′ η λλ )δ3 (k − k′ ) ′ ∗ ∗ × ǫµ (k. λ′ )fk (x)fk′ (y) + ǫ∗ (k.149) λ in (9.148) Using the completeness relation for the polarization vectors ǫµ (k. λ)fk (x)a(k. (9. λ). λ) ∗ µ f (x)fk (y) .λ′ d3 kd3 k′ iω ′ ∗ × ǫµ (k. 3.8 Covariant quantization of the Maxwell theory 369 where λ. λ′ )fk (x)fk′ (y) a† (k. λ′ = 0. Aν (y) = ∗ ∗ d3 k(−iω)ηµν [fk (x)fk (y) + fk (x)fk (y)] = −iηµν δ3 (x − y). λ).150) . a† (k′ . ǫσ (k. λ)ǫν (k′ . λ) + ǫ∗ (k. λ)ǫ∗σ (k. λ). 2. ǫσ (k. λ)ǫ∗ (k. we do not write the equal-time arguments explicitly although they are assumed) ˙ Aµ (x). λ)ǫ∗ (k′ . Aν (y) = λ.9. λ) + ǫ∗ (k. λ′ ) µ = λ. Of course. λ)fk (x)a† (k. we obtain the equal-time commutator to be ˙ Aµ (x). λ′ )fk′ (y)a(k′ . λ) ν = ηµν . λ)ǫν (k.

let us denote by |0 the vacuum state of the theory so that 0|0 = 1. λ) and a† (k. 3.153) where F (k) is a suitable smearing function such that d3 k |F (k)|2 < ∞. If we calculate the normal ordered Hamiltonian of the theory (recall J µ = 0). a(k. for the time-like component there is a relative negative sign in the commutation relation (9. 2. However. (9. 0) + a† (k. 3)a(k. The norm of this one photon state can now be calculated (9. 0)a(k.147). it is strange to note that there are four independent photon degrees of freedom. (9. λ = 0)|0 .152) Let us next consider the state with one time-like photon |1.370 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. λ = 0 = d3 k F (k)a† (k. 3) . 3. 1. it has the form :H: = d3 k ω(k) −a† (k. 2)a(k.154) . For example. 1)a(k.147) that although a(k. λ)|0 = 0. We also note from (9. 2. λ = 0. 1) +a† (k. 2) + a† (k. (9. λ) behave respectively like annihilation and creation operators for λ = 1.151) We can now develop the Hilbert space description for the photons in this theory which leads to some unusual features in the present case.

157) In this case. 0)|0 = 0. so that the one time-like photon state is given by (9. this also leads to trouble as we see below. 3) . Thus the natural modiﬁcation that comes to mind is to interchange the roles of these operators for the time-like photon. (9. 0 = a(k.147) for the annihilation and the creation operator for a time-like photon. λ = 0 = = = − d3 kd3 k′ F ∗ (k)F (k′ ) 0|a(k. 2) + a† (k.8 Covariant quantization of the Maxwell theory 371 1. 3)a(k. λ = 0|1. the normal ordered Hamiltonian will have the form (compare with (9.151)) :H: = d3 k ω(k) −a(k. Let us suppose that a† (k. 2)a(k. 0)a† (k′ . 1)a(k. 0)a(k.9. We also realize that this negative norm is a consequence of the negative sign in the commutator (9.155) 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.158) and the energy of this one photon state is given by . However. 1) +a† (k. 0) − δ3 (k − k′ )|0 d3 k |F (k)|2 < 0. 0) + a† (k. 0)|0 .156) |k. 0)a† (k. 0)|0 d3 kd3 k′ F ∗ (k)F (k′ ) 0|a† (k′ . (9. (9.

0 . The question that immediately comes to our mind is what happens to the probabilistic interpretation of the theory. Thus we see that if we exchange the roles of the creation and the annihilation operators for the time-like photon. This is seen by using the supplementary condition (subsidiary condition) which we still have not imposed. m2 . This not only demands that certain . mλ denote number of photons for each of the polarizations λ = 0. as discussed earlier this is too stringent a condition to allow any state of the radiation ﬁeld. (9. a(k. 0)|0 = −ω(k)|k. deﬁnite states of arbitrarily high negative energy values are possible and the Hamiltonian becomes unbounded from below.372 9 Maxwell field theory H|k. where we have used the fact that H|0 = 0. 3. n3 |m0 . (9. 0)]|0 = d3 k′ ω(k′ )[−a(k′ . physical results are well behaved. As we will see shortly in spite of the presence of negative norm states. 0).161) However. 2. m3 = (−1)n0 δn0 m0 δn1 m1 δn2 m2 δn3 m3 . In fact it is obvious that with this deﬁnition. 0 = Ha(k. a(k. 0)a† (k′ . Namely. the physical states must satisfy ∂µ Aµ (x)|phys = 0. the energy of the state with one time-like photon becomes negative. m1 . In a free theory we may impose suitable conditions to prohibit any unwanted state. n1 . But in the presence of interactions such states may be excited. 1. 0)]|0 (9. Thus states containing an odd number of time-like photons have negative norm.159) = −ω(k)a(k. 0)|0 = [H. 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 .

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. Recalling that kµ ǫµ (k. λ)|phys = 0. 3) |phys = 0. (9. (9.163) ∂µ Aµ (x) is like a free scalar ﬁeld.) We remark here that since (see (9.165) (a(k.9. or. 0) − a(k. Then. k0 a(k. λ)a(k. kµ ǫµ (k. Gupta and Bleuler weakened the supplementary condition to have the form ∂µ Aµ (+) (x)|phys = 0. λ or.165) implies that . λ = 3) |phys = 0. as we have seen. Let Vphys = {P } denote the set of all states which satisfy the supplementary condition and let |ψ represent such a state. (9. λ = 0) − k2 a(k. 3))|phys = 0. |k| (9.129) or (9.130)) ∂µ Aµ (x) = 0. 0) + ǫµ (k. Therefore.74) as well as the fact that for positive energy photons k0 = ω = |k|. 3)a(k. but it also requires that those photons cannot be emitted.8 Covariant quantization of the Maxwell theory 373 kinds of photons are not present in the physical state. (This is commonly known as the Gupta-Bleuler quantization. or.164) the Gupta-Bleuler supplementary condition leads to kµ ǫµ (k. the supplementary condition (9. where we have used the deﬁnitions and the properties of the polarization vectors in (9. 0)a(k. λ) = 0 for λ = 1. 2. it can be decomposed into positive and negative frequency parts uniquely in a relativistically invariant manner and this decomposition is preserved under time evolution.

n1 . n1 . 0) − a(k. n2 . n1 . Cn0 .n2 . 0)|ψ = a(k. (9.n2 . n2 .170) .n3 − n0 |n0 − 1.167) 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 .n1 . 3)a(k.169) which leads to the relation in (9. n2 .n1 .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 . or.168) with the condition C0.n2 . a superposition of diﬀerent states of the form |ψ = n0 .168). 0)|ψ = ψ|a† (k.n3 = 0. n1 .n2 .n3 −1 + n3 Cn0 −1.n2 . n2 . n3 √ − n3 |n0 . or. A general physical state is.n3 (a(k.n2 .n3 in the expansion can be determined recursively from (9. n1 .n1 .n1 .168) This can be seen as follows or. n3 − 1 = 0.0 = 1. ψ|a† (k. 3)|ψ . All the coeﬃcients Cn0 . n3 − 1 = 0. n1 . √ √ ( n0 Cn0 . 3)|ψ . n3 = 0. of course.n1 .n1 . (9.n1 . n2 . (9.n2 .n3 |n0 .374 9 Maxwell field theory a(k. (9. 3)) |n0 . 0)a(k.n2 .n3 Cn0 .n2 . (9.n3 −1 + n3 Cn0 −1.n1 . √ Cn0 . n3 .n3 ) ×|n0 − 1.

175) . = |0. 2 + 2 1. 1|0. n1 . n2 . n2 . n2 .9. n2 . Thus although we can write a general physical state with a ﬁxed number n1 .174) we can assume that in a truly physical state of the radiation ﬁeld there are no time-like and longitudinal photons present. n0 = n3 = 0. n1 . n2 . For example.n1 . the physical states allowed by the supplementary condition have the forms |φ0 |φ1 |φ3 = |0. all such states except for |φ0 have zero norm. (9.172) and so on. 1 + 1. 0 . 0 = 1 − 2 + 1 = 0. 1 − |1. (9. (9. n2 of transverse photons. n1 . n2 . 1|1. n1 . 1 + |2. n1 . n2 .n2 . n2 . . n1 .8 Covariant quantization of the Maxwell theory 375 It follows now that for a ﬁxed number n1 . n1 . (9. n1 . = 0. 0 . n2 . n1 . . √ = |0. n2 . n2 . n1 . n1 . n1 . = φ0 |φ0 .171) and so on where we have used the values of Cn0 . n2 . n2 . n2 of transverse photons as a linear superposition of states of the form |φ because φ|φ = |φ0 + b1 |φ1 + b2 |φ2 + . . 0 . In other words. n1 . n2 .173) (9. we have φ1 |φ1 φ2 |φ2 = 0. 1 + 2. 0|1. 0 = 1 − 1 = 0. Let us note that because of the negative norm of states containing an odd number of time-like photons.n3 ’s determined recursively. 2 − 2|1. n2 . n1 . n1 . 2|0. n2 . n1 . 0|2.

are gauge . However their eﬀect cancels out in the ﬁnal result. The physical S-matrix elements.89) simply because this can be thought of as quantizing the theory in a diﬀerent gauge (FeynmanFermi gauge). Since the zero norm states are orthogonal to all the states including themselves.177) (9. such states decouple. The physical subspace of the theory selected by the supplementary condition (9.179) This is diﬀerent from (9.376 9 Maxwell field theory in such states. on the space of physical states since ψ|∂µ Aµ |ψ = 0. 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. + iǫ (9. consequently. A diﬀerent way of saying this is to note that being orthogonal to every other state. there is no problem with a probabilistic interpretation).178) where V0 represents the set of states with zero norm. V0 V phys = (9. Without going into details we note here that the Feynman propagator for the photon in this theory has the simpler form iGF .77) or (9. 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. If interactions are present time-like and longitudinal states may occur as intermediate states. (9. however.165) contains states with positive semi-deﬁnite norm (negative norm states are eliminated by the supplementary condition or the physical state condition and.176) where |ψ represents a physical state. if we further mod out the states by the zero norm states.µν (k) = lim − ǫ→0+ k2 iηµν . namely. Vphys . we have the true physical subspace of the theory where the norm of states is positive deﬁnite.

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

This page intentionally left blank .

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

The phase space spanned by the independent coordinates and momenta is also 2N dimensional with a (equal time) canonical Poisson bracket structure {qi . ∂pi ∂H . In fact. it follows now that the 2N ﬁrst order dynamical equations (evolution equations) for the system can be written in the Hamiltonian form ∂H . pi ). pi ) = pi qi − L(qi .6) qi = {qi . ˙ (10. (10.3).5). qi }. the time evolution of any dynamical variable A(qi . j {qi . qj } = 0 = {pi .5) Using (10. ˙ ˙ (10. to be equivalent to the N second order Euler-Lagrange equations following from the Lagrangian description. ∂qi (10.3) where summation over repeated indices is understood. pi ) in the phase space can be written in the Hamiltonian form . pi ) as ∂A1 ∂A2 ∂A1 ∂A2 − .380 10 Dirac method for constrained systems (qi .4) or (10. H} = ˙ pi = {pi . ∂qi ∂pi ∂pi ∂qi {A1 . qi ) → (qi .2) and uniquely deﬁne the Hamiltonian of the system through the Legendre transformation H(qi . pj }. qi ). using the deﬁnition (10. A2 (qi . A2 } = (10. pi ).4) which allows us to write the Poisson bracket between any two dynamical variables A1 (qi . pj } = δi = −{pj . H} = − ˙ which can be shown.

θα }. · · · M with θα θβ = −θβ θα .9) and as discussed in chapter 8 (see (8.11) The 2M dimensional phase space is endowed with a canonical Poisson bracket structure (with the convention of left derivatives) {θα . Πβ }. (10. ˙ ∂ θα (10. (10. For example. pi ). θα ) → (θα . we have to deﬁne the derivative in (10.10. we will choose the convention of taking derivatives from the left. β {θα .7) This discussion can also be generalized to a classical system described by Grassmann (anti-commuting) variables.5). Πα ) = θα Πα − L(θα . we can deﬁne the momenta conjugate to the coordinates as Πα = ∂L . θβ } = 0 = {Πα . Πβ } = −δα = {Πβ . unlike the bosonic case in (10.12) The Poisson bracket between any two dynamical variables can now be obtained from (10.9) can be inverted to express velocities in terms of the momenta. Πα ). since the dynamical variables are now of Grassmann odd (fermionic) nature. H}. 2. θα ).8) then. we can go from the conﬁguration space to the phase space ˙ (θα . . θα ).14)). (10. θα ) = −Πα θα − L(θα . However. if ˙ we have a classical system described by the Lagrangian L(θα . pi ) = {A(qi . (10. α = 1. If the relations (10.10) through the Legendre transformation (this is consistent with the convention of left derivatives) ˙ ˙ ˙ ˙ H(θα .9) However.1 Constrained systems 381 ˙ A(qi .12). (10.

2) or (10. . Πα ) takes the form ∂B2 ∂B1 α ∂B2 ∂B1 {θα . Πβ } β − {Π .13) {B1 . the Poisson bracket between two bosonic dynamical variables B1 (θα . H} = − ∂θα (10. θβ } α ∂θα ∂Π ∂Π ∂θβ ∂B1 ∂B2 ∂B1 ∂B2 + ∂θα ∂Πα ∂Πα ∂θα . {F1 .12) as well as the fact that ∂θα and ∂Πα are Grassmann odd (fermionic) variables. with the convention of left derivatives.6). ∂Π ∂H ˙ .14) the 2M ﬁrst order dynamical equations for the system can now be written in the Hamiltonian form as ∂H ˙ θα = {θα . F2 } = − ∂F1 ∂F2 ∂F1 ∂F2 + α ∂θα ∂Π ∂Πα ∂θα .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).10) is invertible so that all the velocities can be expressed uniquely in terms of independent momenta leading to the . B} = − {B. H} = − α . F } = ∂F ∂B ∂F ∂B + α ∂θα ∂Π ∂Πα ∂θα ∂B ∂F ∂B ∂F + ∂θα ∂Πα ∂Πα ∂θα . The discussion so far assumes that the transformation to phase space in (10. B2 } = − = ∂B ∂B where we have used (10. In a similar manner. (10.14) Using (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. for example. B2 (θα .15) which can be compared with (10. Thus. Πα ). Πα = {Πα . (10.

16) we note that pi = ˜ij (q)qj + bij (q)qj . (10.21) .20) Therefore. it follows that the transformation is invertible (A−1 exists) if ∂2L = 0. ∂ qj ˙ ∂ qj ∂ qi ˙ ˙ (10. For conventional b theories described by Lagrangians with quadratic velocity terms.1 Constrained systems 383 unique Hamiltonian of the system. (10.16) where the elements b. it follows that b = b(q). let us concentrate on the bosonic theory for simplicity. ∂ qi ∂ qj ˙ ˙ det (10. In this case. To understand the diﬃculty. However. From (10.16) can be easily seen to have the form A−1 = ½ 0 −b−1˜ b−1 b . (10.18) so that A−1 exists only if the matrix b is invertible.2) can be written in general in the matrix form as q p q q ˙ =A = ½ 0 ˜ b b q q ˙ . b ˙ so that ∂pi ∂2L = .10. ˜=˜ b b(q). the transformation to phase space in (10.17) The inverse of the matrix A in (10. the diﬃculty in passage to a Hamiltonian description arises when this transformation is noninvertible. ˜ denote N × N matrices.19) bij (q) = (10.

may not represent the true brackets necessary for a Hamiltonian description of the system. there exist relations or constraints between various dynamical variables and such systems are known as constrained systems.24) and the other N − R velocities cannot be determined. we can determine only R of the N velocities in terms of coordinates and independent momenta as qa = fa (qi . 2. therefore. if the Lagrangian describing the theory satisﬁes ∂2L = 0. ∂ qi ∂ qj ˙ ˙ (10. Let us assume that the N × N matrix ∂2L .4) need not satisfy the constraints of the theory and. we can write .2 Dirac method and Dirac bracket In a physical system with constraints. 10. In this case. As a result.22) then. therefore.23) of the physical system is of rank R < N . pb ).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. the transformation (10. It follows from (10. it is not clear a priori how to deﬁne the Hamiltonian uniquely. ∂ qi ∂ qj ˙ ˙ det (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. b = 1. it is also clear that the naive canonical Poisson brackets in (10. the passage to a Hamiltonian description is achieved systematically through the method of Dirac which we describe in this section. (10.2) is not invertible. On the other hand. Furthermore. ˙ a. In other words. · · · . R.

˙ ∂L ∂Hcan = gα − = gα − pα ≈ 0.10.25) deﬁne N − R constraints among dynamical variables which we denote by ϕα = pα − gα (qi . ∂ qα ˙ ∂ qα ˙ (10. 2. a. In fact. G in Γ are said to be weakly equal. the dimensionality of Γc would be further reduced.28) where we have used (10.29) so that on the constrained hypersurface Γc . For example. the canonical Hamiltonian does not depend on velocities as we would expect.) Two dynamical variables F. qi ). N. ˙ ˙ ˙ ˙ ˙ (10. (If there are further constraints in the theory. qi ) = pa qa + gα qα − L(qi . · · · .25) It is clear that the second set of equations in (10. (10. the Hamiltonian of the theory is no longer unique.2 Dirac method and Dirac bracket 385 pa = ga (qi . R + 2.27) Γc As usual. pa ) = 0. Hcan = Hcan (qi . (10. · · · . F ≈ G. (10.25). On the other hand.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. We note from (10.2) (presence of constraints). pa ). b = 1. namely. pa ). ˙ pα = gα (qi .28) that in spite of the fact that qα cannot be determined. namely. if (F − G) = 0. qb ). if they are equal on the constrained hypersurface Γc . we can deﬁne the canonical Hamiltonian of the theory as the Legendre transform Hcan = pi qi − L(qi .26) and we recognize that such constraints will reduce the dimensionality of the true phase space of the system. we also note that because of the non-invertibility of (10. from the canonical . we note that the constraints in (10. α = R + 1. R.

Hcan } + λβ {ϕα . Hcan + λβ ϕβ } ˙ = {ϕα . we can write the Hamiltonian equations as ∂Hp ∂(Hcan + λα ϕα ) = . Hp } = {F. with the canonical Poisson brackets in (10. Hcan } + λβ {ϕα .31) Furthermore. ∂pi ∂pi ∂Hp ∂(Hcan + λα ϕα ) =− . ϕβ } + {ϕα .34) this leads to ϕα ≈ {ϕα . Hp } = − ˙ and using (10.34) The constraints of the theory should be invariant under time evolution and using (10. ϕβ } ≈ 0.30) where λα denote undetermined Lagrange multipliers and it follows that Hp ≈ Hcan . we can deﬁne the primary Hamiltonian associated with the system by incorporating the primary constraints of the theory (10. ∂qi ∂qi (10. ˙ (10.4). (10. λβ }ϕβ ≈ {ϕα .35) . Hp } = ˙ pi ≈ {pi . it follows that we can identify the Lagrange multipliers in (10. (10.26) as (it is known as the primary Hamiltonian since it incorporates only the primary constraints in (10. Hcan + λα ϕα }.386 10 Dirac method for constrained systems Hamiltonian of the theory.26)) Hp = Hcan + λα ϕα .32) qi ≈ {qi .29).33) which remain undetermined. The time evolution of any dynamical variable F can now be written as ˙ F ≈ {F. (10. (10.30) with velocities λα ≈ qα .

We continue with this process until all the constraints are determined to be evolution free. · · · .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 Dirac method and Dirac bracket 387 where we have used (10. 2.36) which can be divided into two distinct classes. In general. 1. . · · · . n < 2N. all of the Lagrange multipliers may be completely determined or some of them may remain undetermined. α = 1. 2n2 . It is clear that (10. n1 .37) 2. if there are primary constraints in the theory which are ﬁrst class. .26) in the last step. then there remain undetermined Lagrange multipliers in the primary Hamiltonian after all the . secondary. (10. It is also worth noting that in this process of determining all the constraints that are time independent (evolution free).27)) Poisson bracket with all the constraints (including themselves) are known as ﬁrst class constraints and are denoted as e ψ α ≈ 0. 2. · · · .35) (with Hp as the Hamiltonian) may determine some other Lagrange multipliers or generate newer constraints (tertiary etc). Let us denote all the constraints of the theory (primary. . 2.36)-(10.10. The constraints which have weakly vanishing (see (10.38) that n1 + 2n2 = n < 2N . 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. α = 1.35) may determine some of the Lagrange multipliers or may lead to new constraints known as secondary constraints. On the other hand. ) collectively as ϕα ≈ 0. (10. tertiary. (10. Requiring the secondary constraints to be invariant under time evolution as in (10. α = 1.

/ (10. in general. Although we have identiﬁed all the constraints of the theory. even though φA ≈ 0.4) are not compatible with them. 2. φA } ≈ 0. Thus. (10. we cannot yet impose them directly in the theory since the canonical Poisson brackets in (10. (10.42) This issue of incompatibility of the Poisson brackets can be addressed through the use of the Dirac brackets which are constructed as follows. 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. let us note that we can construct the matrix of Poisson brackets of all the constraints in (10. a consistent Hamiltonian description. collectively as φA ≈ 0.41) the Poisson bracket of the constraints with any dynamical variable F does not. · · · .388 10 Dirac method for constrained systems constraints have been determined.39) and are chosen such that they convert the ﬁrst class constraints (10.37) into second class constraints. n1 . (10. Namely. α = 1. after gauge ﬁxing we denote all the constraints of the theory (including the gauge ﬁxing conditions). The gauge ﬁxing conditions are generally denoted as e χα ≈ 0. First.) The ﬁrst class constraints have a special signiﬁcance in that they are associated with gauge invariances (local invariances) in the theory. 2(n1 + n2 ) < 2N. · · · . vanish {F.40) as . A = 1.40) so that the true phase space of the theory is 2(N − n1 − n2 ) dimensional. in this case. (The number of undetermined Lagrange multipliers equals the number of ﬁrst class constraints among the primary constraints. 2. which are all second class.

(10. Jacobi identity) and in addition satisﬁes −1 {F. φA } − {F. φB } ≈ C AB . in the deﬁnitions of the Hamiltonian. (10. 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. φA } − {F. φA } − {F. A −1 −1 C AD CDB = δB = CBD C DA .45) which can be shown to have all the properties of a Poisson bracket (anti-symmetry.43) will be coordinate dependent and the product in (10. in contrast to the canonical Poisson brackets. The Dirac bracket has the interesting property . (10.44) We note here parenthetically that in ﬁeld theories (where the variables depend on space coordinates as well). 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. the matrix of Poisson brackets in (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.46) Namely. we can now deﬁne the Dirac bracket between any two dynamical variables as −1 {F. G} − {F. With this. namely. φA }CAB {φB . φB }CBD {φD . we can work with the Dirac brackets and set the constraints to zero in the theory (namely. (10. φA } = 0. we can show that the Dirac brackets are indeed the correct Poisson brackets of the theory subject to the constraints in (10.40). G}D = {F. As a result. φA } −1 ≈ {F. G}. φB }CBD C DA = {F.44) will involve integration over the intermediate space coordinate. through the use of Lagrange brackets.2 Dirac method and Dirac bracket 389 {φA .10. In fact.

48) where F is a Lagrange multiplier ﬁeld (an auxiliary ﬁeld without independent dynamics) whose Euler-Lagrange equation 1 ∂L = − (qi qi − 1) = 0.3 Particle moving on a sphere As a simple example of constrained systems. 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.390 10 Dirac method for constrained systems that it can be constructed in stages. the inverse needs to be deﬁned carefully with the appropriate boundary condition relevant for the problem.47) where summation over repeated indices is understood. then the matrix of highest derivatives has the form . 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. It is also worth emphasizing here that in dealing with ﬁeld theories where the matrix C AB is coordinate dependent. · · · . when the number of constraints is large. i = 1. 2. therefore. ˙ ˙ 2 L= (10. let us consider the motion of a point particle constrained to move on the surface of a ndimensional sphere of radius unity.49) gives the constraint (10. F into qa = (qi . · · · . raising and lowering of indices can be carried out trivially) qi qi = 1. We note that if we combine the dynamical variables qi . ∂F 2 (10. (10. 10. F ).43) and its inverse. n + 1.47) on the dynamics. 2. Namely. rather than dealing with a large matrix of Poisson brackets in (10. a = 1. We note that the dynamics of the system can be described by the Lagrangian 1 (qi qi − F (qi qi − 1)) . n.

˙ ∂ qi ˙ ∂L = 0. To understand the passage to the Hamiltonian description in this simple case.50) which is indeed singular (see (10. {qi . {qi .30)) 1 i i p p + F (qi qi − 1) + λ1 pF . pF } = 1. pj } = {F. F } = {pi . 2 leading to the primary Hamiltonian (see (10. the theory has a primary constraint given by ϕ1 = pF ≈ 0. (10.51) has the form 1 i i ˙ Hcan = pi qi + pF F − L = ˙ p p + F (qi qi − 1) .52) The canonical Hamiltonian following from (10. F } = {pF . F } = {qi . pF } = 0.10. pF } = 0. {F.54) The equal-time canonical Poisson brackets of the theory are given by j {qi .55) . we note that the conjugate momenta of the theory are given by ∂L = qi . (10.51) Therefore. qj } = {pi .53) Hp = Hcan + λ1 ϕ1 = (10.48) and (10. (10.3 Particle moving on a sphere 391 ∂2L = ∂ qa ∂ qb ˙ ˙ δij 0 0 0 . 2 (10. pF } = {pi . pj } = δi .22)) reﬂecting the fact that there are constraints in the theory and the naive Hamiltonian description would not hold. ˙ ∂F pi = pF = (10.

(pj pj + F (qj qj − 1) + λ1 pF )} 2 2 = qi pj {qi .58) which generates a new constraint ϕ3 = qi pi ≈ 0. (pi pi + F (qi qi − 1) + λ1 pF } 2 1 = − (qi qi − 1) ≈ 0. Hp } ˙ 1 1 = { (qi qi − 1). = pi pj {qi .60) . 2 (10. we obtain ϕ2 ≈ {ϕ2 .56) Let us denote this secondary constraint as ϕ2 = 1 (qi qi − 1) ≈ 0.57) Requiring (10. Hp } ˙ 1 = {pF . we obtain ϕ3 ≈ {ϕ3 .59) to be time independent. qj } = pi pi − F qi qi (10. 2 (10. Hp } ˙ 1 = {qi pi . pj } + qi F qj {pi .57) to be time independent.392 10 Dirac method for constrained systems With these we can now determine the time evolution of any dynamical variable. In particular.59) Requiring the new constraint (10. (10. (pj pj + F (qj qj − 1) + λ1 pF )} 2 ≈ pi pi − F ≈ 0. requiring the primary constraint to be independent of time leads to ϕ1 ≈ {ϕ1 . pj } = qi pi ≈ 0. (10.

ϕ4 } = {qi pi . ϕ4 } = {pF . which determines the Lagrange multiplier and the chain of constraints terminates. ϕ3 } = { (qi qi − 1).61) to be time independent leads to ϕ4 ≈ {ϕ4 . Collecting all the constraints into φA = (ϕ1 .64) = −{ϕ3 .57). It is easy to check that all the constraints of the theory (10. Hp } ˙ (10. ϕ2 . pi pi − F } = 1 1 {ϕ2 .61) are second class.62) = 2pi F qj {pi .3 Particle moving on a sphere 393 so that we can identify ϕ4 = pi pi − F ≈ 0. A = 1. the nontrivial equal-time Poisson brackets between the constraints take the forms {ϕ1 . Requiring (10. ϕ2 }. ϕ1 }. 2.59) and (10.61) 1 = {pi pi − F. (pj pj + F (qj qj − 1) + λ1 pF )} 2 (10. (10.52). . qj pj } = qi qj {qi . we can calculate the matrix of the (equal-time) Poisson brackets of constraints {φA . (10. 3. which takes the form (10.10. qj } − λ1 {F. pj } = 2pi pi = 2p2 = −{ϕ4 . pj } = qi qi ≈ 1 2 {ϕ3 .63) where we have identiﬁed pi pi = p2 . (10. pj pj − F } = 2pi pj {qi . ϕ3 }. ϕ4 ). φB } = C AB . = −{ϕ4 . 4. pF } = −2F qi pi − λ1 ≈ −λ1 ≈ 0. ϕ3 . In fact.

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

This is the starting point for the quantization of this theory.62).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.71) then the dynamical equation (10.69) where we have used (10.72) The dynamical equation (10. dτ (10. 10. we can set the constraints to zero so that the true independent dynamical variables are (qi . The constraints of this theory were all second class.4 Relativistic particle 395 With the use of the Dirac brackets. We note that if we deﬁne the relativistic four velocity and momentum as uµ = dxµ .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.70) can also be written as dpµ = 0. pi ) (we note that F = pi pi because of the constraint (10.61)) and the Hamiltonian for the theory is given by (see (10. Let us next study the Hamiltonian description of a free relativistic massive particle which is described by the equation d2 xµ = 0. dτ pµ = muµ . (10.54)) 1 i i pp. (10. 2 Hp = (10.73) where λ is a parameter labelling the trajectory and we have identiﬁed .4 Relativistic particle In the last section we studied a simple example of point particle dynamics which is constrained. dτ 2 m (10.

then (10. ˙ ˙ (10. (10.72) as the Euler-Lagrange equations.75) which is an example of a local gauge transformation much like in the Maxwell ﬁeld theory. (10. We note from (10. (10. However. If we choose a gauge λ = τ.74) We note that under a transformation xµ = ˙ dξ dxµ dxµ = .78) which leads to . dλ xµ = ηµν xν . so that (10.75) 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) will lead to the dynamical equation (10.77) where τ denotes the proper time associated with the particle.73) that ∂2L = ∂ xµ xν ˙ ˙ m xρ xρ ˙ ˙ xµ xν ˙ ˙ xσ xσ ˙ ˙ Pµν = ηµν − .73) is invariant under the diﬀeomorphism (10. the action (10. dλ dλ dξ λ → ξ = ξ(λ).76) Namely. let us proceed with the Hamiltonian description without choosing a gauge at this point.396 10 Dirac method for constrained systems xµ = ˙ dxµ .70) or (10.

We note that introducing an auxiliary variable g. However. we note that the equation for the auxiliary ﬁeld g takes the form g−1/2 ∂L = −g−1 xµ xµ + m2 = 0.10. Let us note that the action (10.22)). We can obtain the momenta conjugate to these variables to be .82). we can rewrite the action (10.73) can also be written in an alternate form that is more useful. ˙ ˙ ∂g 4 or. For example. g= xµ xµ ˙ ˙ .81) where because of the diﬀeomorphism invariance of (10. it is more convenient for us to work with the action (10.80) as in the case of the Maxwell theory (see (9. the action (10. The two dynamical variables of the theory in (10.73) to describe the motion of a massless particle. m2 (10.124)) and the naive passage to the Hamiltonian description fails (see (10.81) using (10. ∂ xµ xν ˙ ˙ det Pµν = det (10.82) If we eliminate g in (10. ˙ (10.81). S= dλ L = (10. To see that the two actions (10.81) are xµ and g. Pµν is a projection operator and ∂2L = 0.81) are equivalent.73) we can think of g as the metric on the one dimensional manifold of the trajectory of the particle.79) Therefore.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. Viewed in this manner. it is not clear how to take the m = 0 limit in (10.73) and (10. then we obtain (10.4 Relativistic particle 397 xµ Pµν = ˙ m xρ xρ ˙ ˙ xν − ˙ xµ xµ xν ˙ ˙ ˙ xσ xσ ˙ ˙ = 0 = Pµν xν .73).

87) With these we can now determine the evolution of the primary constraint (10. pg } = 1.85) Adding the primary constraint (10.88) ϕ ˙ 1 ≈ {ϕ . g} = {xµ . Hp = Hcan + λ1 ϕ = 2 1 (10. pg } = {pµ . {g. 4 g which leads to the secondary constraint .83) so that the primary constraint of the theory follows to be ϕ1 = pg ≈ 0. pν } = {g. {xµ . we obtain the primary Hamiltonian for the theory to correspond to √ g µ p p µ − m 2 + λ1 p g .84) Hcan (10. (10. pν } = δν . we obtain √ g µ p pµ − m2 + λ1 pg } 2 (10. 1 1 = − √ pµ pµ − m2 ≈ 0. ∂g ˙ (10. pg } = 0. Hp } = {pg . g} = {pµ .398 10 Dirac method for constrained systems pµ = ∂L = g−1/2 xµ . ˙ ∂ xµ ˙ pg = ∂L = 0. The canonical Hamiltonian of the theory follows to be √ g µ = pµ x + pg g − L = ˙ ˙ p p µ − m2 .86) The equal time canonical Poisson brackets for the theory are given by {xµ . 2 µ (10. g} = {pg . pg } = 0. xν } = {pµ .84). µ {xµ .84) and requiring the constraint to be time independent.

Hp } = { p pµ − m . 2 (10. χ2 } = { 1 µ p pµ − m2 .10. then it follows from (10.89) is the familiar Einstein relation for a massive relativistic particle (the multiplicative factor is for later calculational simplicity). λ − x0 } = p0 2 (10. Let us choose the gauge ﬁxing conditions 1 ≈ 0. which render all the constraints to be second class.92) that the matrix of Poisson brackets of constraints has the form . m2 {ϕ1 . χ1 . 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). As a result. Time evolution of the secondary constraint leads to √ g ν 1 µ 2 p p ν − m 2 + λ1 p g } ≈ {ϕ . g − {ϕ2 .91) χ1 = g − χ2 = λ − x0 ≈ 0. if we combine all the constraints into φA = (ϕ1 .92) = −{χ2 . χ2 ). 2 2 = 0.89) We recognize that the secondary constraint (10. ϕ2 }. ϕ2 . The non-vanishing Poisson brackets between the constraints are given by 1 } = −1 = −{χ1 .4 Relativistic particle 399 ϕ2 = 1 µ p pµ − m2 ≈ 0. m2 (10. (10. ϕ1 }.90) 2 ϕ ˙ 2 so that it is time independent and the chain of constraints terminates. χ1 } = {pg .

φ2 }C24 {φ4 . = {g. g}D = {xµ . G}. pν }D = {xµ . 0 = −1 0 1 0 − p0 . pg } = 0. pν } − {xµ .94) we can calculate the Dirac bracket between the fundamental variables to be {xµ . −1 {g. pg }D = 0.93) 1 0 −1 CAB The equal time Dirac bracket between any two dynamical variables is given by (see (10.45)) −1 {F. φA }CAB {φB .400 10 Dirac method for constrained systems C AB whose inverse is given by 0 0 0 0 1 p0 0 0 = 1 0 0 −p0 0 0 −1 0 0 0 p0 . pν } 2 p (10. (On . pg }{g − 1 . p0 ν With the Dirac brackets we can set the constraints to zero in which case we see that the primary Hamiltonian (10. 0 0 0 (10. pg } {xµ . φ1 }C13 {φ3 . pg } − {g.95) and with (10. pν }D = {g. pν } µ = δν − {xµ . pg } − {g. pg }D = {pµ . m2 −1 {xµ .86) vanishes. 0 0 0 0 (10.94) (10. xν }D = {pµ .96) pµ 0 δ . g}D = {pg . µ = δν − 1 λ 1 p pλ − m2 } − 0 {λ − x0 . pg }D = {g. pg }D = 0. g}D = {pµ . G}D = {F. G} − {F.

the free Dirac theory is described by the Lagrangian density / L = iψ∂ ψ − mψψ. therefore.) Let us also note from (10. we have (remember that c = 1) dτ 2 = dxµ dxµ . ψα .97) On the other hand. dλ dλ xµ xµ = ˙ ˙ (10. (10. dτ dτ (10.10.5 Dirac ﬁeld theory As we have seen in (8. where α = 1.9).20)). 2. that the ﬁrst of the gauge ﬁxing conditions corresponds to choosing λ = τ .5 Dirac field theory 401 the other hand.99) (10. where the adjoint spinor is given by ψ = ψ† γ 0 . This manifests in the deﬁnition of the conjugate momenta as (we have chosen the convention of left derivatives) .100) † The dynamical variables of the theory can be chosen to be ψα . if we set the constraints strongly to zero. dxµ dxµ = 1. (10.98) It follows.89) in particular leads to H = p0 = p2 + m2 which can be taken as the Hamiltonian. 3. or.91) would imply dxµ dxµ = 1. 10. from the deﬁnition of the inﬁnitesimal invariant length in the Minkowski space (see (1. it is obvious that the matrix of second order dervatives vanishes and the system is singular (see (10.22)). 4 denote the Dirac indices and since the Lagrangian density is ﬁrst order in derivatives.82) that the ﬁrst of the gauge ﬁxing conditions in (10.

104) Adding the primary constraints to the canonical Hamiltonian. (10. ˙† ∂ ψα (10.) As a result.11)) ˙ ˙† Hcan = −Π† ψα + ψα Πα − L α = −iψγ · ∇ψ + mψψ. ˙ ∂ ψα ∂L = 0.101) in the intermediate steps. we obtain the two sets of primary constraints for the theory to be † ϕ† = Π† + iψα ≈ 0. we obtain the primary Hamiltonian for the Dirac ﬁeld theory as (see (10. The canonical Hamiltonian density of the theory (consistent with the convention of left derivatives) is obtained to be (see (10. Rather they correspond to the momenta conjugate to ψ and ψ † respectively.105) . (10.402 10 Dirac method for constrained systems Π† = α Πα = ∂L † = −i(ψγ 0 )α = −iψα .103) where we have used (10.30)) Hp = Hcan + d3 x φ† ξα + λ† ρα . The canonical Hamiltonian is now obtained to be Hcan = d3 x Hcan = d3 x −iψγ · ∇ψ + mψψ . α α ρα = Πα ≈ 0.102) which in this case correspond to fermionic constraints. α α (10.101) (Note that Π† is not the Hermitian conjugate of Π. † ˙ † ˙ = iψα ψα − iψα ψα − iψγ · ∇ψ + mψψ (10.

Using the fact that the primary Hamiltonian is independent of time. Π† (y) + iψβ (y)} α β . λ† in the present case denote α fermionic variables. Π† (y)} = −δαβ δ3 (x − y) = {ψα (x). Hp } α α ≈ {φ† (x).10.12)) † {ψα (x). Hcan } α = = = = † d3 y {Π† (x) + iψα (x). Hcan } + α d3 y ξβ (y){φ† (x). φ† (y)} α β = † † d3 y ξβ (y){Π† (x) + iψα (x).5 Dirac field theory 403 where the Lagrange multipliers ξα . Using (10. (10. d3 y ξβ (y){φ† (x).14)).102). For example. 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). Πβ (y)}. ˙ φ† (x) ≈ {φ† (x). φ† (y)} α β −λ† (y){φ† (x).106) (see also (10.107) α β Let us evaluate each of these terms separately.106) with all other brackets vanishing.13)(10. we can now calculate the time evolution of the primary constraints (10. 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). ρβ (y)} . β (10. The equal-time canonical Poisson brackets for the ﬁeld variables take the forms (see (10.

α Substituting these results into (10. (10.111) which determines the other Lagrange multiplier ξα = γ 0 (γ · ∇ + im)ψ(x) α .110) Similarly. α α (10.404 = 0.109) which determines the Lagrange multiplier λ† = − ∇ψ(x) · γ + imψ(x) α α . Hp } ˙ = = † d3 y {Πα (x).108) = iλ† (x). requiring the second set of primary constraints to be time independent we obtain ρα (x) ≈ {ρα (x). ρβ (y)} α β = − = − † d3 y λ† (y){Π† (x) + iψα (x). Πβ (y)} α β d3 y λ† (y) − iδαβ δ3 (x − y) β (10. Hp } = {Πα (x). ψβ (y)} γ 0 (−iγ · ∇y + m)ψ(y) + iξ(y) β d3 y − δαβ δ3 (x − y) γ 0 (−iγ · ∇y + m)ψ(y) + iξ(y) α β = − γ 0 (−iγ · ∇ + m)ψ(x) + iξ(x) = 0. − 10 Dirac method for constrained systems d3 y λ† (y){φ† (x). (10.112) . (10.107) and requiring the constraint to be independent of time leads to ˙ φ† (x) ≈ i∇ψ(x) · γ + mψ(x) α + iλ† (x) = 0.

ρβ (y)} = {Π† (x) + iψα (x). for simplicity of notation. which constitute all the constraints of the theory. (10.114) Let us note next that the two sets of primary constraints (10. Πβ (y)} α α = −iδαβ δ3 (x − y) = {ρα (x). In fact.10. Πβ (y)} = 0. rather it determines both the Lagrange multipliers in the primary Hamiltonian (10. Π† (y) + iψβ (y)} = 0. we have (at equal time) † † {φ† (x). β {ρα (x). Substituting (10. we have deﬁned Π = Π† γ 0 . combining the constraints into ΦA = (φ† . φ† (y)} = {Π† (x) + iψα (x). deﬁne second class constraints (this is consistent with the fact that there is no undetermined Lagrange multiplier in the theory). φ† (y)}. ρα ).115) Thus.102).5 Dirac field theory 405 In this case.112) into the primary Hamiltonian. ρβ (y)} = {Πα (x). we see that requiring the two primary constraints to be time independent does not introduce any new constraint. we can write α the matrix of Poisson brackets as . 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 .105). (10.110) and (10.113) where. (10. α α β β † {φ† (x).

G(y)} ¯ = {F (x). γ (10. (10. y) = i 0 ½ ½ 0 δ3 (x − y). G(y)} −i † d3 z {F (x). G(y)}D = {F (x). Recalling that constraints can be set to zero inside the Dirac bracket. Π† (z) + iψγ (z)}{Πγ (z). G(y)} − d3 zd3 z {F (x). we obtain . φ† (y)} {φ† (x).406 10 Dirac method for constrained systems C(x. G(y)} γ † +{F (x). G(y)} ¯ ¯ z γ +{F (x). G(y)} . φ† (z)}(iδγδ δ3 (z − z )){ρδ (¯).117) The equal-time Dirac bracket of any two dynamical variables can now be easily deﬁned (see (10. y) = {φ† (x). ρβ (y)} α α β {ρα (x). G(y)} ¯ δ z = {F (x). ΦA (z)}CAB (z.45)) {F (x). ρβ (y)} β 0 = −i ½ ½ 0 δ3 (x − y).118) we can calculate the Dirac bracket between the ﬁeld variables. z ){ΦB (¯). (10.118) Using (10.116) The inverse of the matrix is easily seen to be C −1 (x. ργ (z)}(iδγδ δ3 (z − z )){φ† (¯). Πγ (z)}{Π† (z) + iψγ (z). φ† (y)} {ρα (x). G(y)} − −1 ¯ z d3 zd3 z {F (x).

113) becomes Hp = d3 x −iψγ · ∇ψ + mψψ . Maxwell’s equations can be obtained as the Euler-Lagrange equations from the Lagrangian density 1 L = − Fµν F µν .16)). Π† (z) + iψγ (z)}{Πγ (z).18)). (10. ν = 0. where we have neglected the last term in (10. (10.6 Maxwell ﬁeld theory Let us next consider the Maxwell ﬁeld theory which as we know exhibits gauge invariance. 4 µ.6 Maxwell field theory 407 † † {ψα (x).121) which is the Hamiltonian we have used in the quantization of the Dirac ﬁeld theory (see (8.10.120) and these are the relations we have used in quantizing the Dirac theory (see (8. ψβ (y)}D . As we have seen in (9. ψβ (y)} γ d3 z(−δαγ δ3 (x − z))(−δγβ δ3 (z − y)) (10. we can show that † † {ψα (x). 3.16). 10.118) in evaluating this bracket because Πγ has a vanishing Poisson bracket with ψα . Therefore. ψβ (y)}D = 0 = {ψα (x). ψβ (y)}D = {ψα (x). the Hamiltonian of the theory (10. we recall that when using the Dirac brackets.4)) .119) = −iδαβ δ3 (x − y).122) where the ﬁeld strength tensor (see (9. ψβ (y)} −i = −i † † d3 z{ψα (x). 2. 1. Similarly. Furthermore. we can set the constraints to zero in the theory. (10.

k = 1.5) and (9. 3.123) is the four dimensional curl of the four vector potential and contains the electric and the magnetic ﬁelds as its components (see (9.122)) and this manifests in the deﬁnition of the conjugate momenta as ∂L = −F 0µ . (10. µ (10. 2. ˙ ∂ Aµ Πµ = (10.128) We can now obtain the canonical Hamiltonian density of the theory to have the form .124) We also know that Maxwell’s theory is invariant under gauge transformations Aµ (x) → A′ (x) = Aµ (x) + ∂µ α(x). i. ˙ Π = E = −(A + ∇A0 ).127) This determines that the theory has a primary constraint given by ϕ1 (x) = Π0 (x) ≈ 0.125) where α(x) denotes the local parameter of gauge transformation.408 10 Dirac method for constrained systems Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .126) so that we have Π0 = 0. j. As a result of the gauge invariance of the theory. (10. (10. (10. Fij = −ǫijk Bk .7)) F0i = Ei . the matrix of quadratic derivatives becomes singular (see (9.

Aµ (x)}.10. (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 ) .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 .131) where λ1 denotes the Lagrange multiplier of the theory. 2 (10.130) where we have neglected a surface term (total divergence) in integrating the last term by parts. The canonical Hamiltonian. therefore. = 2 (10.128) Hp = Hcan + = d3 x d3 x λ1 ϕ1 1 2 (Π + B2 ) − A0 ∇ · Π + λ1 Π0 .129) where we have used (10. The equal-time canonical Poisson brackets of the theory are given by {Aµ (x). is given by Hcan = = d3 x Hcan d3 x 1 2 (Π + B2 ) − A0 ∇ · Π . Πν (y)} = δµ δ3 (x − y) = −{Πν (y).127).132) Using these we can calculate the time evolution of the primary constraint (10. 2 (10. The primary Hamiltonian of the theory is now obtained by adding the primary constraint (10. Πν (y)}. Aν (y)} = 0 = {Πµ (x). ν {Aµ (x).

as we have seen. Hp } ≈ 0.135) (10.133) = ∇ · Π(x) ≈ 0. (10. Therefore. Hp } = {∇ · Π(x). We see that Maxwell’s theory has two constraints ϕ1 (x) = Π0 (x) ≈ 0. According to the Dirac procedure. the Lagrange multiplier λ1 remains as yet undetermined (there is one primary constraint which is ﬁrst class). in the presence of ﬁrst class constraints we are supposed to add gauge ﬁxing conditions to convert them into second class constraints. In this case.136) and it is clear that both these constraints are ﬁrst class constraints. 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. ˙ so that the chain of constraints terminates. namely. (10. A0 (y)}∇ · Π(y) (10. (Π2 (y) + B2 (y)) 2 − A0 (y)∇ · Π(y) + λ1 (y)Π0 (y)} ≈ − d3 y {Π0 (x). ϕ2 (x) = ∇ · Π(x) ≈ 0. Hp } ˙ = 1 d3 y {Π0 (x). Furthermore.134) It can now be checked that the secondary constraint is time independent. Let us choose .410 10 Dirac method for constrained systems ϕ1 (x) ≈ {ϕ1 (x). ϕ2 (x) ≈ {ϕ2 (x). 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. we have a secondary constraint in the theory given by ϕ2 (x) = ∇ · Π(x) ≈ 0.

χ1 (y)} = {Π0 (x). χα ). A0 (y)} {ϕ2 (x). φB (y)} 0 0 −1 0 2 0 0 0 −∇x 3 δ (x − y). = (∇x )i (∇y )j {(Π)i (x).136) and (10. ϕ2 (y)}. = −1 0 0 0 −2 0 −∇x 0 0 (10.6 Maxwell field theory 411 χ1 (x) = A0 (x) ≈ 0. α = 1.139) The inverse of this matrix is easily calculated to be 0 0 1 0 −1 CAB (x. ∇ · A(y)} = −δ3 (x − y) = −{χ1 (x).136) into second class constraints. (Π)j (y)} = (∇x )i (∇y )j (δij δ3 (x − y)) 2 = −∇x δ3 (x − y) = −{χ2 (x). (10. = 1 0 0 0 2 0 ∇x 0 0 (10. y) −2 0 0 0 ∇x 3 δ (x − y). relations (10.140) .137) determine all the constraints of the theory. ϕ1 (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).138) which leads to (x0 = y 0 ) C AB (x. Combining the ﬁrst class constraints as well as the gauge ﬁxing conditions into φA = (ϕα . χ2 (y)} = {∇ · Π(x). χ2 (x) = ∇ · A(x) ≈ 0.10. y) = {φA (x).137) as the gauge ﬁxing conditions which clearly convert the constraints (10. 2. (10. Therefore.

G(y)} − −1 ¯ z d3 zd3 z {F (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). it leads to {Aµ (x). Aj (y)}D = 0 = {Πi (x).144) j = δi TR (x − y).142) ¯ 1 1 3 2 δ (x − y) = − 4π|x − y| . G(y)}. Π0 (z)}(δ3 (z − z )){A0 (¯).136) and (10. d3 zd3 z {Aµ (x). Aν (y)}D = 0 = {Πµ (x). Πν (y)} ¯ ν 0 ν = δµ δ3 (x − y) − δµ δ0 δ3 (x − y) − d3 zd3 z δµ δj ¯ i ν −2 ×((∇z )i δ3 (x − z))(∇z δ3 (z − z ))((∇z )j δ3 (¯ − y)) ¯ z ¯ ν 0 ν i ν = (δµ − δµ δ0 ) + δµ δj (∇x )i 1 3 j 2 (∇x ) δ (x − y).141) and. ∇x (10. z ){φB (¯). j {Ai (x). in particular. ∇z · Π(z)}(∇z δ3 (z − z )){∇z · A(¯).412 10 Dirac method for constrained systems where.143) Since we can now set the constraints (10. ∇x (10. Πj (y)}D = δi + (∇x )i 1 3 j 2 (∇x ) δ (x − y) ∇x (10. Πν (y)}D . . for the dynamical ﬁeld variables. the relevant Dirac brackets for the Maxwell ﬁeld theory follow from (10.137) to zero. Πν (y)} − −2 ¯ z +{Aµ (x).143) to be {Ai (x). Πν (y)} ¯ ¯ z {Aµ (x). (10. Πν (y)}D = {Aµ (x). with the usual asymptotic condition that ﬁelds vanish at spatial inﬁnity. φA (z)}CAB (z. Πj (y)}D .

Accademia Nazionale dei Lincei. 2 Hp = (10. Bergmann. Hanson. G. Anderson and P. M. Das and W. T. J. New York (1964). Barcelos-Neto. A. we obtain the Hamiltonian for the theory (10. L.145) which is the Hamiltonian we have used in the study of Maxwell’s theory in chapter 9.136) to zero. Physical Review 83. Scherer. ibid.10. Acta Physica Polonica B18. 10. setting the constraints (10. Constrained Dynamics. 269 (1987). Constrained Hamiltonian Systems.7 References 1. . 3. M.7 References 413 which are the relations used in (9.131) to be d3 x 1 Π2 + B2 . Teitelboim. Lectures on Quantum Mechanics.30) (along with the ﬁrst two of (9. J. 1 (1951). Roma (1976). A. Berlin (1982). A. Regge and C. Canadian Journal of Mathematics 2. A. 5. Springer-Verlag. P. Dirac. Sundermeyer. P. K. 129 (1950). Yeshiva University. 4. 6. 1018 (1951). 2.25)) in quantizing Maxwell’s theory. Furthermore. Dirac. 3.

This page intentionally left blank .

parity (P). it is not a continuous transformation.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. 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. we will describe three such important symmetries. these are not continuous symmetries and. we cannot talk of an inﬁnitesimal form for such transformations. Therefore. Therefore. 415 . there are a few discrete symmetries which play a fundamental role in physics. Note that (in three space dimensions) space inversion cannot be obtained through any rotation and. In addition to continuous symmetries. this represents a space-time transformation. P (11.1) while t remains unchanged. however. we can group objects into diﬀerent categories depending on their behavior under parity or space inversion. By deﬁnition. therefore. 11. In this chapter. namely. It is clear that a vector would change sign in a left-handed coordinate system.Chapter 11 Discrete symmetries So far we have discussed mainly continuous symmetries of dynamical systems. Classically. We note that even classically. consequently.

x −→ −x −→ x. A · (B × C) −→ (−A) · ((−B)×(−C)) = −A · (B × C). say. J 0 (x. P (11. namely. L = x × p −→ (−x) × (−p) = x × p = L. 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. t). t). In particular. p −→ −p.4) .2) that the volume (which behaves like a scalar under rotations) changes sign and is known as a pseudoscalar. t) −→ ρ(−x.2) that even though a vector is expected to change sign under a parity transformation. these variables would. t) = ρ(x.1) that applying the parity transformation twice. L·p |p| −→ P P P P P P P P P L · (−p) L·p =− . of course.3) In quantum mechanics. We see from (11. |p| |p| J(x. while a scalar (from the point of view of rotations. (11. Similarly. for example the length of a vector) is expected to remain unchanged under a parity transformation. t) −→ −J(−x. be promoted to operators and they would satisfy the corresponding operator transformation properties. A × B −→ (−A) × (−B) = A × B. This characterization carries over to the quantum domain as well. let us note the transformation properties of some of the well known classical variables under parity. We note from (11.416 11 Discrete symmetries A −→ −A. x −→ −x. P P (11. t) = J 0 (−x. A · B −→ (−A) · (−B) = A · B.2) and so on. we note from (11.

namely.5) Deﬁning a symmetry in a quantum system. of course. For simplicity. ½ and P with P 2 = ½. but not the operators such that (11. we can analyze the parity transformation in (11. In this case.6) holds.1) would result in X −→ − X . let us assume that P represents the operator which implements the action of parity on the quantum mechanical states. but we know now that there exist processes in nature which do not respect parity. shows that even that does not present any diﬃculty. the parity transformation (11. simply corresponds to asking whether this group can be carried over to quantum mechanics without any obstruction.2)) and that expectation values of quantum operators behave like classical objects (Ehrenfest theorem). let us consider a one dimensional quantum mechanical system. we know that dynamical laws of physics (such as Newton’s equation) do not depend on the handedness of the coordinate frame. But a systematic analysis. let us assume that under the parity transformation.6) where we have used the fact that classically vectors change sign under space reﬂection (see (11. First. Classically. Namely. The only diﬃculty we may apprehend is in the case of spinor states since they are double valued. In other words. P −→ − P .6) from two equivalent points of view. as we will see. under space reﬂection an arbitrary quantum mechanical state transforms as . For a long time it was believed that the microscopic systems are also invariant under parity. 11. In this case.1 Parity in quantum mechanics. P P (11.11. (11.1. the classical laws of physics are invariant under parity or space inversion. quantum mechanical states change. As is standard in quantum mechanics.1 Parity 417 so that that even classically parity operation deﬁnes a group with two elements (a group of order two).

9) that xP |y P = x|P † P|y . (11.10) which leads to P † P = ½.8) Note that parity inverts space coordinates and. P (11.11) In other words. or. (11. = P| − x = |x . (11. the parity operator is unitary. (11. P is an idempotent operator and the only eigenvalues of P are ±1. since P|x = |−x . therefore. (11. Furthermore. −x| − y = x|P † P|y . such that ψ|X|ψ ψ|P |ψ → → P P (11.7) ψ P |X|ψ P = ψ|P † XP|ψ = − ψ|X|ψ .13) Namely.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 .9) δ(x − y) = x|P † P|y . It is now obvious from (11. acting on the coordinate basis P must lead to |x → |xP = P|x = | − x .12) P 2 |x which shows that P2 = ½. or. ψ P |P |ψ P = ψ|P † P P|ψ = − ψ|P |ψ .

7) that P |ψ → |ψ P = P|ψ = P = = P dx ψ(x)|x = dx ψ(x)| − x . under a parity transformation the wave function transforms as ψ(x) −→ ψ P (x) = ψ(−x). an eigenstate of parity has an associated wave function which is either even or odd depending on the eigenvalue of the parity . P P (11. In other words. (11. P (11. we note from (11.11.14) To see how a scalar (not a multi-component) wave function transforms under parity.18) (11. ½.16) Namely. dx |x x|ψ dx ψ(x)P|x (11. for such a state |ψ −→ |ψ P = P|ψ = ±|ψ .19) Namely.17) Let us note further that since the eigenvalues of P are ±1. (11. or.15) so that we obtain x|ψ P = ψ P (x) = ψ(−x). if |ψ is an eigenstate of parity.1 Parity 419 P2 = P † = P = P −1 . ψ(x) −→ ψ P (x) = ψ(−x) = ±ψ(x). then P|ψ = ±|ψ .

Similarly.23) In general. (11.20) ψ|X|ψ ψ|P |ψ → → P ψ|X P |ψ = − ψ|X|ψ . we may assume that the states do not transform under the parity transformation. P ]+ = 0. ψ|P P |ψ = − ψ|P |ψ . we can obtain that [P. (11. however. or.24) .6) holds. does not have to be an eigenstate of the parity operator and. the corresponding wave function need not have any such symmetry behavior. therefore. [P. P )P = O(−X. P † P P (11. (11. −P ). P ) −→ OP (X. P P (11. 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.420 11 Discrete symmetries operator for that state.22) where we have used (11. namely.14). In the other point of view. or. P ) = P † O(X. such that (11. PX + XP = 0. one can show that for any operator in this simple one dimensional theory. X]+ = 0.21) It follows from this that P X −→ X P = P † XP = −X. A general state of the system. P = O P † XP. O(X.

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

ψ P (x) = P † XP = x|ψ P = ψ(−x).) However. = P|ψ = or.422 As a result. in this one dimensional theory. dxdy | − y y|X| − x x| .31) It is obvious from this deﬁnition that P† = P2 = = = |ψ P dx |x −x| = dx | − x x| = P. (11. dx | − x x|ψ = dx ψ(x)| − x .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. let us note that in this simple one dimensional quantum theory. 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. takes the form P= dx | − x x|. (11. (Recall that N¨ther’s theorem holds for o continuous symmetries. we cannot construct any function of X and P which would represent the parity operator. This is another way to say that parity is conserved in such a theory. we conclude that 11 Discrete symmetries P|ψ(t) = PU (t)|ψ(0) = U (t)P|ψ(0) . a nontrivial representation of the parity operator is still possible and. dxdy | − x x| − y y| dxdy δ(x + y)| − x y| dy |y y| = ½.

so that it truly gives a representation of the parity operator. under a parity transformation. the properties of the operator P such as in (11. we need to generalize our discussion of parity transformation to multicomponent objects. t) = ηψ Sα ψβ (−x. in principle. in general. 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| (11.32) d |x dx = −P. Furthermore. not only does x → −x. P (11. multi-component operators (recall the Dirac ﬁeld or the Maxwell ﬁeld). such systems can be described consistently only in the language of quantized ﬁelds which are.11. Therefore. we are led to the requirement that . but the components of the object may mix with one another. As we have seen. On the other hand.14) are more important.35) for transformation of wavefunction). Namely. We note that under a parity transformation a multi-component object would transform as P β ψα (x. Let us now go over to relativistic quantum systems. t). We have already seen this in the case of Lorentz transformations (see (3.1 Parity 423 = = − = −X dxdy (−xδ(x + y))| − y x| dx x|x x| = − dx X|x x| dx |x x| = −X.33) β where the matrix Sα can. mix up the diﬀerent components of the object. The important thing to note here is that such a representation is always nonlocal and not very useful from a practical point of view. from the fact that two space inversions are equivalent to leaving the object unchanged. t) −→ ψα (x.

35). therefore. if φ(x. so is φ(−x. the equation is clearly invariant under space inversion. In either of the cases. we conclude that in this case.37) If ηφ = 1 or the intrinsic parity of the spin zero ﬁeld is positive. ( + m2 )φ(x) = 0. t) −→ φP (x. From our general discussion in (11. t). (Remember that parity and the momentum operators do not commute and.2 Spin zero ﬁeld.39) .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. (11. (11. (11.) 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.34) In the language of quantized ﬁelds. t) = P † ψα (x. t). t)P = ηψ Sα ψβ (−x. t) = φ(−x. t) −→ ψα (x. cannot have simultaneous eigenstates unless the eigenvalue of momentum vanishes. φ(x. then φP (x. ∂t2 or. Consequently.38) P (11.424 11 Discrete symmetries 2 ηψ S 2 = ½. the above relation represents an operator relation P β ψα (x. t). (11. t) is a solution of the Klein-Gordon equation. with ηφ = ±1. t).1. t) = ηφ φ(−x. φ= (11. Let us begin with the spin zero ﬁeld which is described by the Klein-Gordon equation ∂2 − ∇2 φ(x) = 0. 11.34) and (11.36) depending on whether the ﬁeld is massless or massive.

On the other hand. ϕ.44) In such a case. t)Yℓ. ϕ. a one particle state not at rest).2)). (11. (11. t). (11. if the intrinsic parity of the spin zero ﬁeld is negative or ηφ = −1. t) = Furthermore. a ﬁeld or a state may not have a well deﬁned parity value (consider.42) r −→ r. (11.m (θ. since parity commutes with angular momentum.m (θ. we have P P P (11. ϕ −→ π + ϕ. we can expand such a ﬁeld (or a state) in the basis of the angular momentum eigenstates.41) φP (x.m (θ. ϕ) −→ Yℓ.m (π − θ. t) = ηφ (−1)ℓ φℓ (x. In general.11. θ. t) = ηφ φℓ (−x. t).40) and we say that such a ﬁeld is a pseudoscalar ﬁeld and that the associated particles are pseudoscalar particles (see. t) = −φ(−x. t) = ℓ φℓ (r.43) so that φP (x. namely. θ. ϕ). therefore. ℓ P (11. t) = φP (r. t). θ −→ π − θ. ℓ ℓ gℓ (r. π + ϕ) = (−1)ℓ Yℓ.1 Parity 425 and we say that such a ﬁeld is a scalar ﬁeld and that the associated particles are scalar particles. then φP (x. On the other hand. ϕ). the spherical harmonics as φ(x. Yℓ. from the fact that P x −→ −x. for example. for example. however. we can deﬁne the total parity for the state (with orbital angular momentum ℓ) as .

45) If parity is a symmetry of the theory.426 11 Discrete symmetries ηTOT = ηφ (−1)ℓ .46) (11. B and C respectively and ℓ is the orbital angular momentum of the B-C system. (11.48) S → P + P.47) where ηA . → P + P. P P → S + S. (11. It now follows that in a parity conserving theory. whereas processes such as (ℓ = 0) S → P + S. the total parity quantum number must be conserved in a physical process and this leads to the fact that for a decay (in the rest frame of A) of a spin zero particle into two spin zero particles A → B + C. P → P + S. one spin zero particle can decay into two spin zero particles in an s-state (ℓ = 0) only through the channels S → S + S. if we have . Let us further note that for a complex spin zero ﬁeld. we must have ηA = ηB ηC (−1)ℓ .49) will be forbidden. ηB and ηC are the intrinsic parities of the particles A. (11. (11.

t) = i|ηφ |2 φ† (−x. t) . t) = ηφ φ(−x. in fact.2). quite general for the bosons and says that for bosons. the intrinsic parity of the ﬁeld φ† is the same as that of the ﬁeld φ. t) = i (φP )† (x. the current density has the explicit form J µ (x. t) = −J(x. particles and antiparticles would have the same intrinsic parity.50) On the other hand. t) = ηφ φ† (−x.38)). (11. t) = J 0 (−x. t).1 Parity 427 φP (x. Let us next look at the electromagnetic current associated with a charged Klein-Gordon system. t)∂ µ φP (x.51) Under a parity transformation J µ (x. t) = i φ† (x. t). t)∂ µ φ(−x. t)∂ µ φ(x. see discussion in sections 7. since ηφ is assumed to take only real values ±1 (see (11. t) −→ J P µ (x. t) − (∂ µ (φP )† (x. t))φ(x. t). t) φ(x. t) − ∂ µ φ† (x. (φ destroys a particle or creates an antiparticle whereas φ† destroys an antiparticle or creates a particle.53) P which is the correct transformation for the current four vector as we have seen in (11. t))φP (x. t). J P 0 (x. As we have seen in (7.52) which leads explicitly to (recall that |ηφ |2 = 1) JP (x. This result is. ∗ (φP )† (x.29). t) .2 and 7. (11. .) Thus we conclude that a charged spin zero particle and its antiparticle have the same intrinsic parity. In quantum ﬁeld theory φ and φ† are associated with particles and antiparticles. t) − (∂ µ φ† (−x. (11.3.(11.11.

t). ∂A ∂A − ∇φ = − − ∇A0 . t) −→ AP 0 (x. t) = ηA0 A0 (−x. B(x. P P (11. we would then generalize these as operator transformations A(x.54) we note that invariance of (11.54) under space reﬂection requires that under a parity transformation. from the deﬁnitions of the electric and the magnetic ﬁelds in (9. t) −→ −E(−x. In the quantum theory. t) = A0 (−x. ∇×B = ∂E + J. (11. A0 (x. t).55) Namely.58) P P . t) −→ A0 (−x.56) we conclude that under a parity transformation the components of the four vector potential would transform as A(x.428 11 Discrete symmetries 11. t). t) −→ −A(−x. t) = ηA A(−x. From Maxwell’s equations in the presence of sources (charges and currents) ∇ · E = ρ = J 0. namely. t).1. A0 (x. t). t) −→ AP (x.2).3 Photon ﬁeld. ∂t (11.57) This is very much like the transformations (11. t) −→ B(−x. we must have E(x. Furthermore. t). E = − (11. P P (11. the magnetic ﬁeld would behave like an axial vector if parity is a symmetry of the system. ∂t ∂t B = ∇ × A.53) of the components of the current four vector under a space reﬂection. t) = −A(−x. we see that whereas the electric ﬁeld would transform like a vector.

t)Aµ (x. t)AP (x.1 Parity 429 Note that a physical photon has only transverse degrees of freedom (namely. This shows that the electromagnetic interaction is invariant under parity and. t)A0 (−x. t)Aµ (−x. t)Aµ (x.35).1. From the parity transformation of the dynamical components. we see that AP (x. Let us recall that the Dirac wave function or the Dirac ﬁeld is described by a four component object. t) · AP (x. 11.60) we note that under a parity transformation Hem = → = = = = P d3 x J µ (x. parity (quantum number) must be conserved in electromagnetic processes. t) = Hem .33) or (11. t) = −A(−x. the photon is a vector particle. namely A. From the form of the electromagnetic interaction Hamiltonian Hem = d3 x J µ (x. In other words. A0 is nondynamical).4 Dirac ﬁeld. t) d3 x J 0 (−x. t)AP 0 (x. therefore. t) d3 x J P µ (x. t) − (−J(x. (11. t) µ d3 x J P 0 (x. (11. t)Aµ (x.61) where in the last step we have let x → −x under the integral. t)) · (−A(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 .11. Thus according to our general discussion in (11. t) d3 x J µ (x. t) − JP (x. t). t)) d3 x J µ (−x. (11. t).

44)) J µ (x. t).64) Since we have assumed the parity eigenvalues to be ±1. t)γ µ ψ(x. S 2 = ½.62) In matrix notation.67) 2 = ηψ ψ(−x. and under a parity transformation this would transform as J µ (x. With this choice (which also ∗ implies ηψ = ηψ ) we note that ∗ ψ (x. with (see (11. t) = ηψ Sψ(−x. we can always choose ηψ = ±1. t) = ηψ Sα ψβ (−x. t)γ 0 S † γ 0 γ µ Sψ(−x. P (11. 4. t). we can write the transformation as ψ P (x. t)γ 0 = ηψ ψ † (−x. α = 1. see (8. t)γ µ ψ P (x.34)) 2 ηψ S 2 = ½. (11. t)γ 0 γ 0 S † γ 0 P = ηψ ψ(−x.430 11 Discrete symmetries P β ψα (x. t)S † γ 0 ∗ = ηψ ψ † (−x. (11. (11. t) = (ψ P )† (x.63) (11. 3. t). t) −→ J P µ (x. 2.66) Let us recall that associated with the Dirac system is a conserved current (probability current or electromagnetic current. t) P P (11. t) = ψ(−x. (11. t) = ψ(x.65) which will satisfy the above relation. t). t)γ 0 S † γ 0 γ µ Sψ(−x. t) −→ ψα (x.68) . t)γ 0 S † γ 0 . t) = ψ (x.

1 Parity 431 2 where we have used the fact that ηψ = 1. t) is a solution of the Dirac equation. t)γ 0 S † Sψ(−x. (11. t) = −ψ(−x. namely.69) for µ = i. t).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) also satisﬁes the Dirac equation in the new coordinates. t). t)γ 0 . or.68) and (11. t) = J 0 (−x. (11. we obtain ψ(−x.11. (11. t)γ 0 ψ(−x.74) It is now quite easy to see that if ψ(x. t) = ηψ ψ(−x. Namely. (11. t)γ 0 S † γ 0 γ i Sψ(−x. ψ(x. t). t) −→ ψ P (x. P P P (11. then the parity transformed function ψ P (x.53)). S † S = ½. if we deﬁne . J P i (x. t) −→ ψ (x.68) and (11. J P 0 (x. t) = −J i (−x. or. we know the transformation properties of the current four vector under space inversion (see (11.69) Comparing (11. comparing (11. On the other hand.65) we have S = S † = S −1 . t) = ψ(−x. we obtain ψ(−x.73) (11.71) Similarly. with ηψ = ±1. t) = ηψ γ 0 ψ(−x. S = γ 0. t)γ i ψ(−x.69) for µ = 0. t). t).

−x).6. the two component Weyl fermions violate P. 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. the two photons move with equal and opposite momentum.9). To determine the relative parity between Dirac particles and antiparticles.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. However. This can also be seen by noting that parity is a symmetry of the free Dirac Lagrangian density (8. Let ǫ1 and ǫ2 denote the polarization vectors for the two photons while k represents their relative momentum.78) where ‘a’ and ‘b’ are scalar functions of momentum and ψpositronium represents the wave function for the positronium. be measured and it . in fact. space inversion is a symmetry of the Dirac equation. ∂xµ (11. as we have discussed in section 3.8 eV) e− + e+ → γ + γ. (11. then ∂ − m ψ P (y) = ∂y µ (11. The polarization planes of the photons in this decay can.432 11 Discrete symmetries y µ = (t.76) In other words. Since the electromagnetic interactions conserve parity. the scalar transition amplitude will have the general form Mi→f = (a ǫ1 · ǫ2 + b k · (ǫ1 × ǫ2 )) ψpositronium .77) In the rest frame of the positronium. (11.

83) In other words. On the other hand. (11. the relative intrinsic parity between Dirac particles and antiparticles is negative. the electron and the positron must have relative negative intrinsic parity. This is. in the rest frame of the positronium.11.81) we conclude that the positronium state must be a pseudoscalar under parity. we must have ηTOT = ηe− ηe+ = −1. it can also be seen in an intuitive manner from the structure of the spinor functions as follows. Since the transition amplitude is a scalar under parity while P (11. the transition amplitude for this decay takes the form Mi→f = (b k · (ǫ1 × ǫ2 )) ψpositronium . Namely. Let us recall from our earlier discussions that antiparticles are related to negative energy states. for this decay we have ǫ1 · ǫ2 = 0. (11.79) Consequently. However. This result can be more directly seen when we study charge conjugation in the next section.1 Parity 433 is experimentally observed that the two are perpendicular. (11. (11.49)) . a very general result for fermions. in fact. the total parity of the state is given by ηTOT = ηe− ηe+ .82) and for this state to represent a pseudoscalar state. namely.80) k · (ǫ1 × ǫ2 ) −→ −k · (ǫ1 × ǫ2 ) . Let us choose two representative solutions (of positive and negative energy) of the Dirac equation in the rest frame (see (2.

3) under parity. . t). On the other hand.100) (or in section 3. t) −→ ψ (x. t).84) introduced in (2. (11.86) which shows that this combination behaves like a scalar under parity. angular momentum). 0 0 0 0 0 0 0 0 = −η . namely.87) −ψ(−x. For example. t)γ5 ψ(−x. (11. t)ψ(−x. since fermions always appear in pairs in any process (because of conservation of fermion number. t)γ 0 γ5 γ 0 ψ(−x.434 11 Discrete symmetries ψp0 >0 ψp0 <0 This analysis also brings out another feature involving the fermions. t) ψ(−x. t)ψ(x. t) = = P P |η|2 ψ(−x. Let us next calculate the transformation properties of the sixteen Dirac bilinears ψΓ(α) ψ. 1 1 0 0 1 1 (11. t)γ 0 γ 0 ψ(−x.85) 0 P 0 = −→ ηγ 0 0 0 0 P −→ ηγ 0 = 1 0 1 0 0 = η . t) −→ ψ (x. t)ψ P (x. we note that ψ(x. t)γ5 ψ P (x. t)γ5 ψ(x. t) = = P P |η|2 ψ(−x. ψ(x. t) (11. the only meaningful quantity as far as the fermions are concerned is the relative parity of a pair of fermions.

for example. t)γ5 γ 0 ψ(−x. we can also derive that ψ(x. t)γ 0 ψ(−x.11. namely. t)σ 0i ψ(x. t)γ5 γ 0 ψ(x. t). t). t). t)γ5 ψ i ψ(x. ψ(x.1 Parity 435 so that this combination behaves like a pseudoscalar under parity. t)γ i ψ(−x. the combination ψγ5 γ µ ψ behaves in an opposite way from the current four vector and consequently. t). t)γ i ψ(x. t) −→ −ψ(−x. the spin zero meson can be identiﬁed with the π-mesons which are known to be pseudoscalar mesons. namely ψψφ. only one of the forms of the interaction can be allowed depending on the intrinsic parity of the spin zero meson. Thus. t)γ 0 ψ(x. Finally. ψ(x. t). t)γ5 γ i ψ(−x. However. t)σ ij ψ(−x. P P P P (11. t)σ 0i ψ(−x.8 after (8. t). ψ(x. Therefore.92)) . It can similarly be checked that ψ(x. The transformation properties of the bilinears are important in constructing relativistic theories. there are two possible Lorentz invariant interactions we can write down involving fermions and spin zero particles. t)σ ij ψ(x. t) −→ −ψ(−x. t) −→ ψ(−x.89) Namely. it is known as an axial vector (or a pseudovector). t) −→ −ψ(−x. or ψγ5 ψφ. ψ(x. As it turns out.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.88) (11. it was thought sometime ago that the strong force between the nucleons was mediated through the Yukawa interaction of spin zero mesons. We have already seen that the combination ψγ µ ψ behaves like a four vector. (11. t) −→ ψ(−x. P P (11. the only allowed interaction in this case has the form (recall also the discussion in section 8.91) Since strong interactions conserve parity. t) −→ ψ(−x.

E = − classically we see that charge conjugation leads to (11.95) . J → −J. ∇×E = − ∇×B = ∂B . ∂t (11. (11.436 11 Discrete symmetries ψγ5 ψφ. 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. E → −E. since ∂A − ∇A0 . ∂t B = ∇ × A. ∇ · B = 0.92) 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. (11. 11. let us go back to Maxwell’s equations in the presence of charges and currents ∇ · E = ρ. if we know the interaction Lagrangian or Hamiltonian density for a given process.2 Charge conjugation To understand the discrete symmetry known as charge conjugation.94)) and is denoted by C. B → −B. ∂t ∂E + J.93) We note that this set of equations is invariant under the discrete transformations ρ → −ρ.

a real Klein-Gordon theory describes charge neutral spin zero particles and. opposite charges are assigned to particles and antiparticles. C (11. (As we have seen. reﬂecting charges in a quantum theory is equivalent to saying that charge conjugation really interchanges particles and antiparticles.96) by saying that C Aµ (x) −→ AC (x) = ηA Aµ (x) = −Aµ (x). it is not a space-time transformation. 11. Note that the transformation of charge conjugation does not change the space-time coordinates and.96) It is obvious that the electromagnetic interaction Hamiltonian Hem = d3 x J µ Aµ . µ C Jµ (x) −→ Jµ (x) = ηJ Jµ (x) = −Jµ (x). charge conjugation is sometimes also referred to as particleantiparticle conjugation.) We have seen that for such a theory we can deﬁne an electromagnetic current (7. charge conjugation really interchanges every distinct quantum number between particles and antiparticles.2. charge conjugation is trivial in this theory as we will discuss shortly. C Jµ −→ −Jµ . we can generalize the classical result (11.11. Let us now analyze the complex Klein-Gordon ﬁeld theory which describes charged spin zero particles. therefore. For this reason. therefore. In other words.29) as . two charge conjugation operations (reﬂecting the charge twice) would bring back any variable to itself.98) and we say that the charge conjugation parity (or simply the charge parity) of the photon as well as the current is −1. In the quantum theory.2 Charge conjugation 437 Aµ −→ −Aµ . Therefore.97) is invariant under this transformation.1 Spin zero ﬁeld. (11. Let us also note here that like parity. in a quantum theory. Let us also note that in a quantum theory. C (11.

this further supports the fact that (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). (11. then C φ(x) −→ φC (x) = ηφ φ† (x). As a result.100) would deﬁne the appropriate charge conjugation transformation for the current.100) is the appropriate charge conjugation transformation for complex scalar ﬁeld.438 11 Discrete symmetries J µ = i φ† ∂ µ φ − (∂ µ φ† )φ . ∗ φ† (x) −→ (φC )† (x) = ηφ φ(x). It follows now that under such a transformation. we have (|ηφ |2 = 1) φ(x) −→ ηφ φ† (x). We have already discussed (see sections 7. we continue to use “†” instead of “∗” so that the passage to quantum theory will be straightforward). C (11.100) with |ηφ |2 = 1.2 and 7. under a charge conjugation. C C C Aµ (x) −→ −Aµ (x). (11. (11. ∗ φ† (x) −→ ηφ φ(x). Therefore.101) C Thus the ﬁeld transformations in (11.99) We note that if we treat φ(x) as a classical function (even though the functions are classical.102) .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).

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

(11. If we symmetrize the products in the deﬁnition of the current.440 11 Discrete symmetries normal ordering the product). (11. will change sign) under charge conjugation. under a charge conjugation (11.106) leads to (this should be compared with the normal ordered charge deﬁned in section 7.110) .107) µ the eﬀect of the inﬁnite constant in Jsym in (11.109) which brings out the particle-antiparticle symmetry in a more direct way. if a spin-zero ﬁeld theory describes charge neutral particles which are their own antiparticles.106) 2 µ Jsym (x) = and it is easy to check that this current will transform properly (namely. it is described by a Hermitian ﬁeld operator φ† (x) = φ(x). 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).108) This is esthetically more pleasing and treats the vacuum in a more symmetrical way. But this has the interesting consequence that while (remember the Dirac vacuum picture) 0|J µ (x)|0 −→ ∞. (11. As we have discussed earlier. In such a case. 2 µ Jsym (x) = (11. Note that we can also write the symmetrized current as 1 µ (J (x) − J Cµ (x)) .3) µ 0|Jsym (x)|0 = 0. we obtain i φ† ∂ µ φ + (∂ µ φ)φ† − (∂ µ φ† )φ − φ∂ µ φ† .

or. C (11. we conclude that the π 0 meson must also have charge parity (+1) and similar arguments hold for other mesons. 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. If charge parity is to be conserved in this process (this corresponds to an electromagnetic process since photons are involved).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. . As an example. (11. (11.2 Charge conjugation 441 φ(x) −→ φC (x) = ηφ φ† (x) = ηφ φ(x). C (11.115) since the charge parity of the photon is (−1). we expect 2 ηφ = 1. ηφ = ±1.114) Such ﬁelds or particles are known as self-charge-conjugate and simply describe particles that are their own antiparticles.11. But by analyzing various physical processes where charge conjugation is a symmetry or by analyzing the invariance of various interaction terms under charge conjugation. the sign of the charge parity cannot be ﬁxed from theory. so that in these cases (11.113) φ(x) −→ φC (x) = φ(x).112) In this case.

442 11 Discrete symmetries 11.2 Dirac ﬁeld. (iγ µ (∂µ − ieAµ ) − m) Cψ = 0.121) . let us note from (11. T T T (11. we obtain i(γ µ )T (∂µ − ieAµ ) + m ψ = 0. Thus. it would satisfy the equation (iγ µ (∂µ − ieAµ ) − m) ψ C (x) = 0. Let us next analyze a system of Dirac fermions in- teracting with an electromagnetic ﬁeld. iC(γ µ )T C −1 (∂µ − ieAµ ) + m Cψ = 0.117). we note that if we can ﬁnd a matrix C satisfying C(γ µ )T C −1 = −γ µ .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.117) Taking the transpose of the adjoint equation (11.2. (11.120) (11. interacting with an electromagnetic ﬁeld. In fact.116) +m = 0. from (11. The adjoint equation is easily obtained to correspond to ← ψ(x) iγ µ ∂µ −ieAµ (11.118) we would have C i(γ µ )T (∂µ − ieAµ ) + m ψ = 0. If the fermions are minimally coupled to the photons.118) and (11. (11. T (11. To see the relation between ψ C (x) and ψ(x). or. or. the equation of motion takes the form (iγ µ (∂µ + ieAµ ) − m) ψ(x) = 0.119) where we have identiﬁed the charge conjugate function ψ C as describing the positron.119) that the two equations are very similar in structure.

125) .119). C T = C. from C −1 γ µ C = −(γ µ )T . C −1 γ µ C = −(γ µ )T .127) (11. Similarly. it follows that there must exist a similarity transformation which relates −(γ µ )T and γ µ .123) That such a matrix exists can be deduced from the fact that −(γ µ )T also satisﬁes the Cliﬀord algebra.92)). In 4-dimension C T = −C.122) (11. (11.2 Charge conjugation 443 which can be compared with (11. or. Therefore we conclude that if we can ﬁnd a matrix C with the property C(γ µ )T C −1 = −γ µ . (This property depends on the dimensionality of space-time. namely. let us assume that instead C is symmetric. we can identify (with |ηψ |2 = 1) ψ C (x) = ηψ Cψ . we obtain γ µ C = −C(γ µ )T = −C T (γ µ )T = − (γ µ C)T .128) (11.124) and hence from the generalized Pauli theorem (see (1.126) (11. namely.11. (11. T (11. In this case. The important symmetry property that C must have is that it should be antisymmetric. we can also show that σ µν C = − (σ µν C)T . −(γ µ )T . −(γ µ )T + = 2η µν ½.) To see this.

133) The fermion ﬁelds. we can construct 10 linearly independent antisymmetric matrices. however.129) Unlike the parity transformation. (11. in the PauliDirac representation that we have been using. it follows that C must be antisymmetric. if we do not restrict to any particular representation. the appropriate operator ordering for the current is antisymmetrization (this is equivalent to normal ordering in this case).44)) .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 . given the classical current (see (8. satisfy anticommutation relations to reﬂect the Fermi-Dirac statistics that the underlying particles obey. C T = −C. It now follows that ψ C = ηψ Cψ = ηψ γ 0 γ 2 (γ 0 )T ψ ∗ = −ηψ γ 2 ψ ∗ .131) (11. as quantum mechanical operators. C C T (11. Therefore. it is easy to determine that C = γ 0γ 2. T (11. Thus. In this case. there can be only six linearly independent antisymmetric matrices and C cannot be symmetric. But these are 4 × 4 matrices and. the structure of the matrix C depends on the representation of the Dirac matrices and.132) In general. C † C = 1) ψ(x) −→ ψ C (x) = ηψ Cψ . we have (with |ηψ |2 = 1. (11. therefore. ∗ ψ(x) −→ ψ (x) = −ηψ ψ T C −1 .444 11 Discrete symmetries so that if C T = C.

Therefore.122) in the intermediate steps. where we have used (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 .137) whereas 0|J µ (x)|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) . this current transforms properly under charge conjugation.136) µ = −Janti (x).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. (11. we can obtain the antisymmetrized current as µ Janti (x) = (11.11.2 Charge conjugation 445 J µ (x) = ψ(x)γ µ ψ(x). 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. (11. (11.

if we deﬁne properly antisymmetrized Dirac bilinear combinations as = − (γ µ γ5 )T = (γ5 γ µ )T .446 11 Discrete symmetries µ Janti (x) = 1 µ (J (x) − J Cµ (x)) . We have already seen in (11. = i γ 0γ 1γ 2γ 3 (11. (11.139) Let us next calculate how the Dirac bilinears transform under charge conjugation.141) i C −1 σ µν C = C −1 (γ µ γ ν − γ ν γ µ ) C 2 i = C −1 γ µ CC −1 γ ν C − C −1 γ ν CC −1 γ µ C 2 i − (γ µ )T − (γ ν )T − − (γ ν )T − (γ µ )T = 2 i (γ µ )T (γ ν )T − (γ ν )T (γ µ )T = 2 i ν µ = (γ γ − γ µ γ ν )T = −(σ µν )T . 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. we obtain T C −1 γ5 γ µ C = C −1 γ5 CC −1 γ µ C = γ5 − (γ µ )T (11.122) that the charge conjugation matrix C satisﬁes C −1 γ µ C = −(γ µ )T . .142) 2 Therefore.140) − (γ 1 )T T − (γ 2 )T T − (γ 3 )T T = γ5 . 2 (11.

A. (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). . (11. . ψψ − ψ T ψ (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 .144) 2 (If we think for a moment. P. T.) We have already seen that ψγ µ ψ −→ − ψγ µ ψ C anti anti . α = V.2 Charge conjugation 447 ψΓ(α) ψ anti ≡ 1 2 ψΓ(α) ψ − ψ T Γ(α) T ψ T .147) where the phase ǫα for the diﬀerent classes has the form 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 . (11. ǫα = −1. for (11.145) Similarly we can also show that ψγ5 ψ ψγ5 γ µ ψ ψσ µν ψ −→ −→ C C C anti anti anti ψγ5 ψ anti .148) .11. (11. (11. for α = S.

151) Here we have used the fact that γ 0 is Hermitian in our metric.150) under parity. t) P −ηψ γ 0 ψ C (−x.152) Of course. t) = ηψ γ 0 ψ(−x. Thus P if ηψ represents the intrinsic parity of a Dirac particle. t). t). 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 we obtain . t) = ηψ C(ψ )T (x.83). P ψ(x. we have already seen this from the analysis of the positronium decay in (11. P P P (11.149) Since the antiparticle is described by C ψ C (x. t) = ηψ ψ(−x.73) that under parity a Dirac particle transforms as P ψ(x. t)γ 0 )T C P ηψ ηψ Cγ 0 C −1 C(γ 0 )∗ ψ ∗ (−x. (11. t) P C −ηψ ηψ γ 0 C(γ 0 )T ψ ∗ (−x. t) = ηψ Cψ (x. t) −→ ψ (x. t). Next. t) −→ (ψ C )P (x. let us note that since the charge conjugate wave function is deﬁned to be C ψ C = ηψ Cψ . t) P P = = = = = C P ηψ ηψ C(ψ(−x. but this derivation is simpler and more general. T (11. the antiparticle will transform as C ψ C (x. t) −→ ψ P (x. t)γ 0 .448 11 Discrete symmetries Let us next recall from (11. T T (11. t) C P −ηψ γ 0 ηψ Cψ (−x.

159) . Let us recall that if a free Dirac particle satisﬁes the equation (p − m)ψ = 0. the antiparticle of a particle with a given handedness has the opposite handedness.3.153) As a result. T (11.119) we see that the antiparticle must also satisfy the same free equation.2. of course. (11. (p − m)ψ C = 0.11.6 and have used this in the construction of the standard model in section 14. we see that if γ5 ψ = ψ. already seen this in the study of neutrinos in section 3.3 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 .157) In other words. 11. (11. if γ5 ψ = −ψ. namely. then it follows that C γ5 ψ C = −ηψ C − ψ (11.156) T T C = ηψ Cψ = ψ C . / (11.154) T (11. We have. / where (11.155) On the other hand.158) then from (11.

2 = (11. (11.123) we see that a Dirac particle which is its own antiparticle must satisfy ψ = ψ C = ηψ Cψ . namely. From the deﬁnition of charge conjugation in (11.162) Such particles are known as self charge conjugate fermions or Majorana fermions. in fact. quite obvious.164) that for Majorana fermions (ψ = ψ C ) µ Janti (x) = 0.165) so that the Majorana fermions are charge neutral and hence cannot have any electromagnetic interaction. They can be massive and will have only two independent degrees of freedom (because of the constraint in (11. .163) T (11. The answer to this question is. T 11 Discrete symmetries (11.450 ψ C = ηψ Cψ .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) . and would also satisfy the free Dirac equation (p − m)ψ = (p − m)ψ C = 0. whether there are also fermionic particles that are their own antiparticles and if so how does one describe them. (11. These are charge neutral particles described by real quantum ﬁelds φ† = φ. / / (11.160) We have also seen that there are bosons in nature which are their own antiparticles.162)).161) We can now ask the corresponding question in the case of the Dirac particles. Let us note from the deﬁnition of the antisymmetrized current in (11.

However. To answer this.166) which can be satisﬁed only if the intrinsic parity of a Majorana fermion is imaginary. we can ﬁnd a simple relation between them. this implies that it cannot have a well deﬁned handedness. then χC = (ψL )C + ψL = χ. (This is. Since a Majorana fermion is its own antiparticle. this would require P P ηψ = −(ηψ )∗ . Let us also note here that since both the Weyl and the Majorana fermions have only two degrees of freedom. we note that if we deﬁne χ = ψL + (ψL )C . not true if the space time dimensionality is not four. it can in principle be complex).2 Charge conjugation 451 Secondly. we can have Majorana-Weyl spinors. the mass term would trivially vanish since (11. as we have seen a particle with a deﬁnite handedness will lead to an antiparticle with the opposite handedness. however.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 have also seen that if the intrinsic parity of a Dirac particle P P is ηψ .169) . let us note that if we try to write a Lagrangian density for a massive Weyl fermion (say left-handed).11. In particular for d = 2 mod 8. For a Majorana fermion then. 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. In fact. 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. Thus we cannot talk of a Majorana particle with a given helicity. (11. (11.167) (11.

172) then this would imply violation of baryon number which is not seen at present. note that any majorana particle would necessarily imply some sort of violation of a fermion number which is not seen in nature.170) This is.452 11 Discrete symmetries 1 1 ψ L ψL = ψ (1 + γ5 ) × (1 − γ5 ) ψ = 0. of course. if (here n represents the antineutron) n = n. 2 2 (11. namely. But such an identiﬁcation is not compatible with the experimental results. Obviously. In fact. the neutron and the neutrino are charge neutral and hence are prime candidates to be Majorana particles. / (11. (11. a Weyl fermion cannot have mass. 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. the essence of the statement that a Weyl fermion is massless. if (once again ν denotes the antineutrino) ν = ν. we can write ( we will choose C ηψ = 1) C / L = iψ L ∂ ψL + mψL ψL T = iψ L ∂ ψL − mψL C −1 ψL . Obviously. However. Similarly. the mass term is not zero and we can give a mass to the Weyl fermion. . note that if we go beyond the conventional Dirac mass term in the Lagrangian density.173) this would imply violation of lepton number which is also not seen yet.171) Clearly. 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. (11. if we require conservation of fermion number. in this case. Let us next ask whether there are any particles in nature which we can think of as Majorana particles.

. we have ˆ ˆ QC|Q. the direct experimental limits (mνe < 2ev.176) . . We have seen that the pho- ton has odd charge conjugation parity. . .2 Charge conjugation 453 We can now ask why anyone would then like to study such particles or mass terms.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. . . mνµ < 190keV. 11. 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. . = −Q| − Q. It is. .11. To understand the eigenstates of charge ˆ conjugation in general. . then acting on an eigenstate it gives ˆ Q|Q. = Q| − Q. of course. . . its theoretical studies would require the study of the properties of such particles. . = Q|Q. . . . . . . we have also seen that a Hermitian spin zero ﬁeld has charge parity +1. mντ < 18. This. . . of course. = QC|Q. On the other hand. Similarly. . . Independent of whether supersymmetry is a symmetry of nature. (11. . . .4 Eigenstates of charge conjugation. . .174) where . The answer to this lies in the fact that while no one has really measured the mass of the neutrino. means that a one photon state is an eigenstate of charge conjugation with eigenvalue −1.175) (11. refer to other quantum numbers the state can depend on. = Q| − Q. Another reason for the study of Majorana particles is that they are fundamental in the study of supersymmetry which we will not go into. . . let us note that if Q denotes the operator for charge (electric). up to the experiments to decide whether this is really what happens.2. It follows now that ˆ C Q|Q. so that comparing the two relations we obtain (11.

We have. of course. ℓ . = |π + p . or. the charge operator does not commute with the charge conjugation operator and. ℓ = (−1)ℓ |π − π + . Let us note. = |K . consequently. we would have C|π − π + . for the positronium with orbital angular momentum and spin values ℓ. Note that C|π − π + = |π + π − . 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. It is easy to see that C|n C|π − p C|K 0 = |n . Q + = 0.178) which clariﬁes the point. Thus if the state has well deﬁned angular momentum quantum number. (11. On the other hand.454 11 Discrete symmetries ˆ ˆ C Q = −QC. ˆ C. that while all eigenstates of C must be charge neutral. Similarly. ℓ = |π + π − .179) The ﬁnal state simply corresponds to the initial state with the two particles interchanged. (11. . not all charge neutral states would be eigenstates of charge conjugation operator. (11. s respectively. they cannot have simultaneous eigenstates unless the charge of the state vanishes.180) so that such a state would be an eigenstate of charge conjugation with eigenvalue (−1)ℓ . a state such as |π − π + can be an eigenstate of C. 0 (11.177) In other words. however.

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

190) (11. and it follows from this that C † U C = U. (11.456 11 Discrete symmetries nγ → (N − n)γ. note that if a Hamiltonian is invariant under charge conjugation. then we must have . Therefore. 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. In fact. As we have seen. (11. or. This result goes under the name of Furry’s theorem.189) where U denotes the time evolution operator. (11. (−1)n = (−1)N −n .1: An amplitude with N photons.188) In other words. Similarly. electromagnetic interactions are invariant under charge conjugation.187) then conservation of charge parity would require (electromagnetic interactions are invariant under charge conjugation) Figure 11. if the strong interaction Hamiltonian respects charge conjugation symmetry. (−1)N = (−1)2n = 1. then C † HC = H.

(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. the weak current does not have a well deﬁned transformation property under charge conjugation and hence will violate C-invariance. (11. On the other hand the product operation CP appears to be a good symmetry in most weak processes.2 Charge conjugation 457 π − p|U |π − p = π − p|C † U C|π − p = π + p|U |π + p . Let us note here without going into details that while charge conjugation is a good symmetry for both strong and electromagnetic interactions.192) 2 2 where ηJ denotes a phase. 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 . Thus we see that P and C (P for parity) are conserved in strong and electromagnetic interactions but are violated in weak 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 . The simplest way to see this is to note that the weak current (see (14.11. it is violated in weak processes just like parity. Namely.191) In other words.

194) W 2 2 This has well deﬁned transformation properties under CP transformations and leads to CP invariance. 11. time reversal or time reﬂection corresponds to the space time transformation where we reverse only the sign of time. ∇ · B(x) = 0.195) We note that the Newton’s equation is invariant under time reversal since it is second order in the time derivative. (11.193) 2 2 where ηTOT denotes the total phase under the transformation. T t −→ −t. ∂t ∂E(x) + J.458 = = = 11 Discrete symmetries g C C P P √ (ηe )∗ ην (ην )∗ ηe ν eL γ 0 γµ (1 − γ5 ) γ 0 eL 2 2 g C C P P √ (ηe )∗ ην (ην )∗ ηe ν eL γ 0 γµ γ 0 (1 − γ5 ) eL 2 2 g ηTOT √ ν eL γ 0 γµ γ 0 (1 − γ5 ) eL . Similarly. 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 . ∇ × E(x) = − ∇ × B(x) = ∂B . (11. namely. Maxwell’s equations in the presence of sources ∇ · E(x) = ρ(x). T (11. x −→ x.196) are also form invariant if we deﬁne . ∂t (11.3 Time reversal Classically.

t) = ρ(x. however.199) The transformation of the Aµ (x) potentials under time reversal. symmetry transformations normally are implemented by linear operators which are also unitary. A(x. we determine that under time reversal A0 (x. t) −→ JT (x. t) = E(x. we will have . −t). t) = −J(x. Thus we would like to investigate whether time reversal can be a symmetry of quantum mechanical systems. B(x. −t).198) (11. E(x.11. T T T T T T (11.57). t) −→ A0T (x. we know that macroscopic physics is not. t) −→ ET (x. However. therefore. Thus assuming that time inversion is also implemented by such an operator T . −t). t) = −A(x.3 Time reversal 459 ρ(x. 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) = A0 (x. The ﬁrst problem which we face.197) ∂A . The diﬀusion equation which is ﬁrst order in the time derivative is not symmetric under t → −t. From the deﬁnition of the electric and the magnetic ﬁelds E = −∇A0 − B = ∇ × A. t) −→ AT (x. J(x. t) = −B(x. the entropy or disorder increases with time. −t). −t). −t). is opposite from their behavior under parity in (11. in statistical mechanics we know that microscopic reversibility does hold. t) −→ ρT (x. While some of the macroscopic equations are invariant under time reversal. Note that in quantum mechanics. namely. ∂t (11. t) −→ BT (x. we know that there is a sense of direction for time. Macroscopically.

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

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

462 11 Discrete symmetries In the time reversed process.211) T PT † = ηP P = −P. under time inversion.204) (see.207) (11. the role of the initial and the ﬁnal states are interchanged and the transition amplitude of interest would be ψi |ψf . ηP = −1. we know that x −→ x. . t) = x|ψ(t) −→ ψ T (x. let us note from (11.204) as well as (11. Thus. (11.208) which is indeed consistent with the deﬁnition of time inversion transformations given in (11.209) Let us derive the transformation properties of some of the operators under time inversion. −t). It follows now that T [Xi .212) where we have used (11. It is reassuring to note that the canonical commutation relations are unchanged by the ηX = 1.211). T ∗ (11. Classically. T (11. In particular.204) that ψ(x. in particular the third relation). T p −→ −p. (11. on the other hand. Pj ] T † = T (Xi Pj − Pj Xi ) T † = Xi (−Pj ) − (−Pj )Xi = − [Xi . Pj ] . T (11. we expect ψf |ψi −→ ψi |ψf . t) = x|ψ(−t) = ψ ∗ (x.210) Thus from Ehrenfest’s theorem we would expect the corresponding operators to transform as T XT † = ηX X = X.

It follows now that for the angular momentum eigenstates satisfying L2 |ℓ. m L3 T |ℓ. m is an eigenstate of L2 and L3 with eigenvalues 2 ℓ(ℓ + 1) and − m respectively so that we can identify T |ℓ. (11. Therefore. −m . (11. m we have L2 T |ℓ. (11. Since the angular momentum operators can be written as Li = ǫijk Xj Pk . (11. ℓ(ℓ + 1)(T |ℓ.216) = −T L3 |ℓ. Let us note here that the naive extension of the time inversion transformation with a linear operator would have led to an inconsistency in the commutation relations. we obtain T Li T † = ǫijk T Xj Pk T † = −ǫijk Xj Pk = −Li . m . under time inversion each component of the angular momentum operator would change sign which is consistent with the classical result. m L3 |ℓ. m with T = Cm |ℓ.204) that i → −i under the anti-linear time reversal transformation). m ).213) Namely. it follows that T L2 T † = T Li Li T † = (−Li ) (−Li ) = Li Li = L2 . m = − m(T |ℓ.11.214) so that the L2 operator remains unchanged under time inversion.217) .3 Time reversal 463 time inversion transformation (note from (11. We see that the state T |ℓ. (11. m ). m = 2 2 = = ℓ(ℓ + 1)|ℓ. m = |ℓ.215) m|ℓ. m . m = T L2 |ℓ.

3.220) (−1)m Yℓ. We have already seen that under time reversal Aµ (x.m (cos θ)e−imφ 4π (ℓ + |m|)! (11. φ). Thus. in general. the constant (phase) is chosen to be Cm = (−1)m = i2m . (11. m = (i)2m |ℓ. we can write T |j. t) −→ AµT (x. T (11.m (θ. we can indeed write T |ℓ. By convention.−m (θ. −mj . (11. φ) = = T θ.m (cos θ)eimφ 4π ℓ + |m|)! ∗ −→ Yℓ.1 Spin zero ﬁeld and Maxwell’s theory. −t). mj = (i)2mj |j. φ) = = (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. namely. −m . which follows because under time inversion (say for m > 0) (11. m (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. mj = T |j.221) It can be shown that the above relation holds true even for half integer angular momentum values.223) . where we have used the fact that the spherical harmonics for m < 0 are deﬁned with a phase +1.218) (11.464 11 Discrete symmetries |Cm |2 = 1. t) = ηAµ Aµ (x. φ|ℓ.222) 11. m = T |ℓ.219) Yℓ.m (θ.

invariant under time reversal as we have already seen and as can be checked directly. (11.11. −t) − J µ (x. ηA = −1.225) T T −→ ∂µ AT (x. therefore. t). −t). t) −→ F0i (x − t) = F0i (x. that Fµν (x. (11.226) Similarly. (11.227) ηJ 0 = 1. we have seen that under time reversal. t) − ∂ν Aµ (x. −t). t) = ηJ µ J µ (x. T Fij (x. where T (11. 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. t) = −Fij (x. t) −→ Fij (x. −t) F µν (x.229) = d4 x S .228) The Maxwell theory is. T T (11. −t) 4 1 = d4 x − Fµν (x)F µν (x) − J µ (x)Aµ (x) 4 = S. t) −→ J µT (x.3 Time reversal 465 where ηA0 = 1. ηJ = −1. t) = T ∂µ Aν (x. −t). of course. ν µ so that T F0i (x. t) (11.224) It follows. t) − ∂ν AT (x. the current four vector transforms as J µ (x. −t)Aµ (x.

−t). −t) − (∂ µ φ(x. −t)) −→ i (φT ∗ (x)∂ µ φT (x) − (∂ µ φT ∗ (x))φT (x)) T = −i (φ∗ (x.234) ∂2 − ∇2 + m2 φ(x) = 0.231) (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. −t) ∂t2 (11. (11.233) T (11. −t).466 11 Discrete symmetries In general. −t))φ(x. For example. Note that since the complex spin zero ﬁeld satisﬁes + m2 φ(x) = + m2 φ∗ (x) = it is clear that in this case + m2 φT (x) = = = 0. the Klein-Gordon equation is invariant under time inversion. Furthermore. we have φ(x) −→ φT (x) = ηφ φ∗ (x. ∂t2 ∂2 − ∇2 + m2 φ∗ (x) = 0. T (11. for a spin zero ﬁeld. ∂2 − ∇2 + m2 φT (x) ∂t2 ∂2 − ∇2 + m2 ηφ φ∗ (x. for a multicomponent function or ﬁeld we expect the behavior under time reversal to be T β ∗ ψα (x) −→ ψα (x) = ηψ Tα ψβ (x.235) . −t))φ∗ (x. −t)).232) That is.230) where the matrix T mixes the components of the function and |ηψ |2 = 1. −t) − (∂ µ φ∗ (x. −t)∂ µ φ∗ (x. −t)∂ µ φ(x. ∂t2 (11.

−t). −J i (x. of course. t) = 0. 11. we obtain J 0 (x.) Thus we conclude that the spin zero charged particles interacting with photons can be described by a theory which will be time reversal invariant.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) is deﬁned with − ∂t . t) −→ −i φ∗ (x. (Here ∂ we have used the fact that J 0 (x.197). −t).2 Dirac ﬁelds. −t)∂ i φ(x. −t) = 0. t) −→ −i φ∗ (x. (11. the correct behavior of the current four-vector under time inversion as we have already seen in (11. −t))φ(x. ∂t ∂x ∂ ∂ − iγ i∗ i − m ψ ∗ (x. −t).238) ∂ ∂ − iγ i∗ i − m ψ ∗ (x.11.3 Time reversal 467 Therefore. or. −t) = 0. ∂(−t) ∂x (11. t) −→ ψ T (x. −t))φ(x.239) (11. −t) − ( φ∗ (x. t) = ηψ T ψ ∗ (x. −iγ 0∗ −iγ 0∗ iγ 0∗ ∂ ∂ − iγ i∗ i − m ψ ∗ (x. T (11.3. we expect that under time reversal ψ(x. ∂t ∂x . −t) = which is. From our general discussion (see (11. or.230)). −t) = T T ∂ ∂ φ(x. −t) ∂t ∂t J 0 (x. −t) − (∂ i φ∗ (x.236) J i (x. or. leads to (upon complex conjugation) (−iγ µ∗ ∂µ − m) ψ ∗ (x) = 0.

∂t ∂x ∂ ∂ + iγ i i − m T ψ ∗ (x.245) .241) ψ T (x) = ηψ T ψ ∗ (x. ∂t ∂x (11. −t). −t) = 0. or. and the matrices γ µ deﬁned as (˜ µ = γ µ† ) ˜ γ (11. ˜ γ i = −γ i . iγ 0 iγ 0 ∂ ∂ + iγ i i − m ψ T (x) = 0. t) were to be a solution of the Dirac equation.242) would satisfy the same Dirac equation as ψ(x. we should have (iγ µ ∂µ − m) ψ T (x) = 0. ˜ also satisfy the Cliﬀord algebra {˜ µ . T γ i∗ T −1 = −γ i . We note that the matrices γ µ∗ satisfy the Cliﬀord algebra {γ µ∗ . comparing the two equations we see that if we can ﬁnd a matrix T such that T γ 0∗ T −1 = γ 0 . t). or.468 11 Discrete symmetries On the other hand. then the time reversed wave function (11.243) γ 0 = γ 0.240) Therefore. (11.244) (11. γ ν∗ } = 2η µν ½. if ψ T (x. γ ˜ (11. γ ν } = 2η µν ½.

They must also be related to each other through a similarity transformation. Let us also recall from (11.246) (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 .248) Thus we see that the inverse relation can be written as T −1 γ µ T = γ µT .249) (11. or. (11. (11.250) CT −1 γ µ T C −1 = Cγ µT C −1 = −γ µ . CT −1 γ µ CT −1 −1 = −γ µ .247) To determine the explicit form of T . T = −γ i∗ = − γ i† = γ iT . Combining the two results we obtain (11.11. the matrix CT −1 anticommutes with all the γ µ ’s and hence must be a multiple of γ5 .251) In other words.122) that Cγ µT C −1 = −γ µ . (11. Thus we can identify T γ µ∗ T −1 = γ µ . Choosing .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). T γ i∗ T −1 = −γ i . ˜ leading to T γ 0∗ T −1 = γ 0 .

we can show that in four space-time dimensions.254) (11.252) (11.256) so that the time inversion matrix in the spinor space is antisymmetric. −t)T −1 γ 0 . (11. the T matrix satisﬁes T = T † = T −1 . −t).253) (11.130) C = γ 0γ 2. (11.470 11 Discrete symmetries CT −1 = −γ5 = −iγ 0 γ 1 γ 2 γ 3 . Let us next analyze the behavior of the current four-vector under time inversion . T T T ∗ ψ(x. we determine (in our representation) T = iγ 1 γ 3 . −t) = T ηψ T γ 0T ψ (x. as in the case of charge conjugation.255) Furthermore. −t) = ηψ γ 0 T ψ (x. t) = ηψ ψ T (x. Note that with this choice. t) −→ ψ (x. and using the deﬁnition (11. t) −→ ψ T (x. (11.257) with (|ηψ |2 = 1) deﬁnes a symmetry of the Dirac equation. t) = ηψ T ψ ∗ (x. Therefore we see that the time inversion transformations T ψ(x. independent of any representation (the superscript denotes transposition) (T )T = −T.

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

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

t) −→ −ψ(x. namely. (11. Let us also note that if we are looking at a reaction of the form a + b −→ c + d. T = ηψ γ5 T ψ ∗ = ηψ T γ5 ψ ∗ † = ηψ T γ5 ψ ∗ γ5 ψ T = ηψ T (γ5 ψ)∗ .267) where α stands for all the other quantum numbers of the state. j. −t).265) In other words. p. the tensor bilinears have precisely the same behavior under time reversal as the electromagnetic ﬁeld strength tensor F µν . −t)σ ij ψ(x. −p. Let us recall that under time reversal. It is also easy to see now that under time reversal. a state with given angular momentum transforms as (see (11.264) We can also show that ψ(x. the chirality of a spinor does not change. We could have expected this from the deﬁnition (see (3. mj −→ (i)2mj |α.222)) T |α. −t).11. −t)σ 0i ψ(x. T T (11. t)σ ij ψ(x. −mj .3 Consequences of T invariance. then the transition amplitude can be represented as (11. j. |p| |p| |p| h= (11. t) −→ ψ(x.121)) s·p s · p T (−s) · (−p) −→ = = h. t)σ 0i ψ(x.3. ψ(x.266) so that a left-handed spinor remains left-handed under time reversal and a right-handed spinor remains right-handed.268) . (11. 11.3 Time reversal 473 the helicity does not change.

b . (11. namely. d . |f = |c. |f for the ﬁnal state and S denotes the S-matrix or the time evolution matrix. if we are looking at the reciprocal reaction c + d −→ a + b. On the other hand. namely.275) . It is worth emphasizing here that the detailed balance does not say that the rates for the two reactions a + b −→ c + d.474 11 Discrete symmetries f |S|i .270) Since the time evolution operator is not Hermitian. (11. for example. (11.274) (11. and c + d −→ a + b. experimentally this is seen to hold in many processes.269) where |i stands for the initial state. it does not follow a priori that | i|S|f |2 = | f |S|i |2 .271) (11. This is known as the principle of detailed balance and has been used. |i = |a. where we are still using the same notation. (11. to obtain the spin of π-meson etc.272) (11. S † = S.273) On the other hand. then the transition amplitude can be represented as i|S|f .

280) . 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 . mi . mf . T ST † = S. if the interaction Hamiltonian is Hermitian. when the principle of detailed balance may hold. the principle of detailed balance will hold.278) In this case. Let us next assume that we are dealing with a system whose Hamiltonian is invariant under time reversal. h (11. (11. = |β. It follows then (in the operator formulation) that the evolution operator or the S-matrix must also be invariant under time inversion. then we have | f |S|i | = 2π| f |Hint |i |δ (Ei − Ef ) = 2π| i|Hint |f |δ (Ei − Ef ) = | i|S|f |.11. In fact. Let us next try to understand. (11. (11. That is.277) and furthermore. Let us assume that we are dealing with a system where the interaction Hamiltonian can be treated as a perturbation.279) Let us choose the initial and the ﬁnal states to correspond to |i |f = |α. we can write (for distinct |i and |f states) f |S|i = −2πi f |Hint |i δ (Ei − Ef ) . pf . therefore. pi . theoretically.3 Time reversal 475 have to be the same. (11. In this case.276) where ρf denotes the density of states for the ﬁnal state which may not be the same for the two processes.

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

pi . Let us recall that although the neutron is electrically neutral. Note that if we are looking for a subclass of processes where mi = mf = m. the principle of detailed balance will hold.286) Furthermore. pi .mf | β. Thus in this case we will have β.285) = mi . −m|S|β.11. This relation is known as the principle of semi-detailed balance and holds true in every system for which P and T are symmetries.3 Time reversal 477 mi . m|S|β. P is violated – but the strength of the interaction Hamiltonian is assumed to be weak. since rotations are assumed to be a symmetry of physical systems. pf .mf This is very similar to the principle of detailed balance but not quite. 11. it has . then from P and T invariance we have β. m|S|α. mi |S|β. it is obvious that the principle of detailed balance or the semi-detailed balance must hold in practically all processes. Therefore. (11. pi . Time reversal invariance holds extremely well in strong and electromagnetic interactions. pi . mf |2 . pf .287) (11. pf . (11. m|S|α. pi .288) In this case.4 Electric dipole moment of neutron. pi . On the other hand. strong and electromagnetic interactions are assumed to be P and T symmetric. mi |2 | α. pf . The stringent limits on T -invariance in electromagnetic interactions comes from the measurement of the electric dipole moment of the neutron. m = α. therefore. m (11. −m . Note that in weak interactions. pf . mf |S|α. m = α. pf .3. we can make a rotation on the right-hand side to bring the state | − m → |m .

(11. σ −→ σ.292) In other words. P P P P P E −→ E.63 × 10−25 e-cm. whereas the interaction involving the magnetic dipole moment is invariant under P and T . then the electric dipole moment of the neutron must vanish.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 · B −→ σ · B. Consequently. T T T (11. . (11. Experiments have obtained extremely small limits on µe . this is the Pauli coupling) Hint = µe σ · E + µm σ · B. the electric dipole moment would be parallel to its spin.293) This basically measures the extent to which P and T invariances hold in electromagnetic interactions. Let us assume that it also has an electric dipole moment. (11. so that we have σ · E −→ −σ · E. σ −→ −σ. if electromagnetic interactions are invariant under P and T . T T (11. Thus the neutron would interact with an electromagnetic ﬁeld through the interaction Hamiltonian (there is no minimal coupling.478 11 Discrete symmetries a magnetic dipole moment. 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.290) We have already seen that under parity and time reversal E −→ −E. B −→ −B.291) σ · E −→ −σ · E. (µe )exp < 0. σ · B −→ σ · B. Since there is no natural axis for the neutron other than the spin axis. B −→ B.

4. m. −s|(CPT )H(CPT )−1 |α. m. The operation of CP. (11. m. (11.11. Thus we see that m = = = α. 11. m. Let us consider the state of a particle at rest which satisﬁes H|α. m. s α. −s|. (11.294) where m denotes the mass of the particle. s|H|α. Thus under the CPT transformation. m.1 Equality of mass for particles and antiparticles. of course. |α.295) where m. s . m. We can naturally ask whether there exists any combination of these symmetries which will be a symmetry of the weak Hamiltonian. α denote the parameters associated with the antiparticle (in this section. Rather than going into the technical proof of this theorem. quantities with bars would refer to antiparticle parameters). −s . then it follows that m = α. −s = m. m. s|(CPT )−1 (CPT )H(CPT )−1 (CPT )|α. s α.297) . s its spin and α denotes the other quantum numbers. m. −s|H|α.296) If the Hamiltonian is CPT invariant.4 CPT theorem 479 11. m. s = m|α. C and T in any order even though the individual transformations may not be a symmetry of the system. takes a particle to the antiparticle state. let us discuss the obvious consequences of this theorem. m. s −→ η α. 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.4 CPT theorem We have seen that some of the discrete symmetries appear to be violated in weak interactions. CPT (11.

4. (11. ˆ 11. q. s|(CPT )−1 (CPT )Q(CPT )−1 (CPT )|α. s ˆ α. the electric charge for particles is equal in magnitude but opposite in sign from that of the antiparticles.300) (11.301) . Namely.2 Electric charge for particles and antiparticles. q. 11. it is CPT invariance which guarantees 0 that. q. In fact. for example. one can show that all the quantum numbers of antiparticles are equal in magnitude but opposite in sign from those of the particles. ˆ T ˆ Q −→ Q. (11.4. q. As we have seen ˆ P ˆ Q −→ Q. Even though C-invariance is violated in weak interactions.299) ˆ C ˆ Q −→ −Q. so that ˆ CPT ˆ Q −→ −Q. −s (11. ˆ (Recall Q = d3 x ψγ 0 ψ for a Dirac particle. K 0 and K have the same mass. the ex0 perimental limit on the mass diﬀerence between the K 0 −K provides the best limit on CPT invariance. we have q = = = ˆ α. Since the weak interactions violate parity. Let Q denote the operator which measures the charge of a state. q. 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− .3 Equality of lifetimes for particles and antiparticles.298) = −q. s ˆ α.480 11 Discrete symmetries In other words. the equality of particle and antiparticle masses is a consequence of CPT invariance. In this way. s|Q|α.) Thus for a one particle state with charge q at rest. q. −s|(−Q)|α.

j. mj (11.303) The cross terms would have a pseudoscalar character and would vanish since the reaction rate is a scalar quantity. mj |(CPT )−1 (CPT )H± (CPT )−1 (CPT )|α. mj |H± |α.302) In such a case. and the corresponding antiparticle decay channel |α. mj −→ |β. Let us consider the decay channel of a particle |α. mj . −mj . we can rewrite this relation as β.11. using the rotational invariance of physical Hamiltonians.4 CPT theorem 481 where PH± P −1 = ±H± . j. j. j. j. mj −→ |β. j. (11. Furthermore.304) = (i)2mj −2mj α. (11. −mj |H± |β. −mj = 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 .306) where we have used the fact that H± would be individually invariant under CPT according to the CPT theorem.305) (11. j. mj = α. mj |H± |α. j. j. j. mj . j. β.307) . We know that β. j. j. mj |H± |β. (11. j. −mj |H± |β. j. mj . mj = (11. j.

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

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

This page intentionally left blank .

We will now study gauge theories based on more complicated symmetries.1 Non-Abelian gauge theories So far. (12. Such theories belonging to non-Abelian (non-commutative) gauge groups are commonly known as Yang-Mills theories and are fundamental building blocks in the construction of physical theories. (12. let us quickly review how the gauge ﬁelds come into the theory in the case of quantum electrodynamics (QED). Let us recall that gauge ﬁeld theories necessarily arise when we try to promote a global symmetry to a local symmetry. it is a complex ﬁeld). the Maxwell theory which is based on the Abelian gauge group U (1). we have −iǫψ. namely.37) and (8.Chapter 12 Yang-Mills theory 12. As we have seen in (8.9) / L = iψ∂ ψ − mψψ. δψ δψ = = where θ is a real constant parameter and inﬁnitesimally. Let us start with the Dirac Lagrangian density for a free fermion (8. we have talked about the simplest of gauge theories.3) .2) ψ → ψeiθ . iǫψ. 485 (12. this Lagrangian density has the global symmetry ψ → e−iθ ψ.38). For example.1) where ψ provides a representation of U (1) (namely.

Consequently.1) whose variation would cancel the ∂µ ǫ(x) term. the phase of ψ remains arbitrary. Neither the Lagrangian density (12. As we have seen earlier (see (9. δψ = iǫ(x)ψ. The internal symmetry in this case arises from the fact that ψ is a complex ﬁeld variable while the Lagrangian density which is a functional of ψ is Hermitian.4) since ∂µ ψ does not transform covariantly and there is no term in the Lagrangian density (12. δ (∂µ ψ) = ∂µ (δψ) = ∂µ (−iǫ(x)ψ) = −i ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ(x). namely ǫ = ǫ(x) (or θ = θ(x)).81)). δ (Dµ ψ) = −iǫ(x)Dµ ψ(x). the way out of this diﬃculty is to deﬁne a covariant derivative Dµ ψ such that it transforms covariantly under the local transformation. namely. (12. (12. we have δL = iδψ∂ ψ + iψγ µ ∂µ δψ − mδψψ − mψδψ / −imǫ(x)ψψ + imǫ(x)ψψ. we note that δψ = −iǫ(x)ψ.5) (12.7) .4) = −ǫ(x)ψ∂ ψ + ψγ µ ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ / = (∂µ ǫ(x))ψγ µ ψ. (12.486 12 Yang-Mills theory with θ = ǫ = inﬁnitesimal. Consequently. If we now want to promote the symmetry to be local.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).1) nor the corresponding action is invariant under the local phase transformation in (12.

δψ = iǫ(x)ψ(x). e (12.13) .86). That is. if we deﬁne a ﬁeld strength tensor Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .11) under the gauge transformation (12.8) We recognize (12.10) However. the Lagrangian density L = iψγ µ Dµ ψ(x) − mψψ.12. is nondynamical in this theory.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 . the curl of the ﬁeld would be unchanged by such a transformation.9) is invariant under the inﬁnitesimal local gauge transformations δψ = −iǫ(x)ψ(x).10). 4 (12. the ﬁeld Aµ known as the gauge ﬁeld. this would transform as δFµν = ∂µ δAν − ∂ν δAµ = 1 (∂µ ∂ν ǫ − ∂ν ∂µ ǫ) = 0. To introduce the kinetic energy part for the gauge ﬁeld (in order to give it dynamics) in a gauge invariant manner.1 Non-Abelian gauge theories 487 with the transformation property for the additional ﬁeld 1 ∂µ ǫ(x). we note that since the gauge ﬁeld transforms longitudinally (by a gradient). e δAµ (x) = (12.7) as describing the minimal coupling of the charged fermions to the electromagnetic ﬁeld and as we have seen in (9. (12. (12. e (12. δAµ = 1 ∂µ ǫ(x).

k a = 1.14) and has the local gauge invariance described in (12. · · · . dim R. (We note here parenthetically that dim G = n2 − 1 for the group SU (n). . dim G. . Let us consider a free Dirac theory described by the Lagrangian density (containing several complex fermion ﬁelds) L = iψ k ∂ ψk − mψ k ψk . . 2. a = 1. Here the T a ’s represent the generators of the internal symmetry group G and are assumed to be Hermitian. 4 (12. . .) The generators (matrices) T a satisfy the Lie algebra of the Lie group G (this is necessary for the transformations to form a group) . Let us now generalize these ideas to theories with more complicated symmetries. 2.10).16) where θ a denotes the real global parameter of transformation and inﬁnitesimally we have δψ k = iǫa ψ ℓ (T a )ℓk .80)) 1 LQED = − Fµν F µν + iψγ µ Dµ ψ − mψψ. Furthermore. (12. (12. dim G. Namely. .17) with θ a = ǫa = inﬁnitesimal. 2. This Lagrangian density is invariant under the global phase transformations aT a ψk → ψk → Uψ ψU † k = e−iθ ψ k.488 12 Yang-Mills theory The total Lagrangian density for QED is then given by (see also (9. . a δψk = −iǫa Tkℓ ψℓ . / k = 1. 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.15) where k is an internal symmetry index. aT a k = ψU −1 = ψeiθ k . (12. . dim G represents the dimensionality of the symmetry group.

(12. a. the symmetry group is known as a non-Abelian group. b. namely. b. namely. 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. 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. . For example. . (12. (For semi-simple groups the structure constants can always be chosen to be completely antisymmetric. the structure constants of the group SU (2) can be identiﬁed with f abc = ǫabc . the structure constants are nontrivial). It is easy to see that the Lagrangian density is invariant under the transformations (12. c = 1. we note that the ordinary derivative acting on the fermion ﬁeld does not transform covariantly under this transformation. 2. let us consider the inﬁnitesimal local phase transformations a δψk = −iǫa (x)Tkℓ ψℓ .12.16)). (12. a. T b = if abc T c . f 458 = f 678 = 23 .1 Non-Abelian gauge theories 489 T a . That is.18) − 1 . c = 1. . we already know from the study of angular momenta.17) (or (12.20) Once again. a δψ k = iǫa (x)ψ ℓ Tℓk . (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. f 147 = f 246 = f 257 = f 345 = 2 .) When the generators of the symmetry group 2 do not commute (namely. dim G. . Let us now try to make this symmetry a local symmetry of our theory. 3 while for the group SU (3) the structure constants are more complicated and the nontrivial ones 1 are given by f 123 = 1.19) Here we have used the fact that the symmetry generators are Hermitian. 2. .

22) As in the case of the U (1) theory. / / δL = δψ k (i∂ − m)ψk + ψ k (i∂ − m)δψk = (∂µ ǫa (x))ψ k γ µ (T a )kℓ ψℓ . we write the covariant derivative (in the matrix notation) as Dµ ψ = (∂µ + igAµ ) ψ. µ (12. µ µ (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. We can compare this with QED where the generator is the identity matrix ½ which commutes with every operator.24) (12. We can also write the covariant derivative in (12.22).23) Introducing a new ﬁeld (gauge ﬁeld). (12.21) so that under the local transformation (12.24) explicitly as a a Dµ ψk = ∂µ ψk + igTkℓ Aa ψℓ = ∂µ δkℓ + igTkℓ Aa ψℓ .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 . We note that the number of gauge ﬁelds (that are needed to deﬁne a covariant derivative) is the same as dim G.20). the dimension of the Lie group. there is no other term in the Lagrangian density whose variation would cancel the ∂µ ǫa (x) term in (12. where g denotes the appropriate coupling constant and Aµ = T a Aa . (12. (12.26) . 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 .

(12. µ This can also be written as . a δ (Dµ ψ)k = −iǫa Tkℓ (Dµ ψ)ℓ .1 Non-Abelian gauge theories 491 Noting that we would like the covariant derivative to transform covariantly under an inﬁnitesimal local transformation. µ (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 .27) we conclude from (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 . T b Ab ψm µ (12.12.28) We note that the left-hand side in (12.30) a = i (∂µ ǫa ) Tkℓ ψℓ + igǫa f abc (T c )km Ab ψm . µ µ (12.29) Substituting this into (12.28). 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 .

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. . dim G. µ µ g g (12.33) in terms of matrices as (of course. we can write the Lagrangian density in (12.31) which determines the transformation for the gauge ﬁeld to be δAa = µ 1 1 ∂µ ǫa − gf abc Ab ǫc = ∂µ ǫa − f abc Ab ǫc . .36) with normal product rules for matrices. µ g (12. 2. . (12.33) with the covariant derivative deﬁned in (12. δAa = µ 1 ∂µ ǫa − f abc Ab ǫc . (12. µ a = 1. the Lagrangian density is a scalar) L = iψγ µ Dµ ψ − mψψ.32) Thus the Lagrangian density L = iψ k γ µ (Dµ ψ)k − mψ k ψk . (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).25)) Aµ = T a Aa . . µ g (12.492 12 Yang-Mills theory a igTkℓ ψℓ δAa − µ 1 ∂µ ǫa + f abc Ab ǫc = 0.16)) .34) If we use the deﬁnition (see (12. a δψ k = iǫa (x)ψ ℓ Tℓk .

we note that under the gauge transformation (12.34)) for the ﬁeld variables in the matrix form as ψ → U ψ. ig (12.11)) would transform as fµν = ∂µ Aν − ∂ν Aµ 1 (∂ν U ) U −1 ig 1 (∂µ U ) U −1 ig 1 1 (∂µ ∂ν U ) U −1 − (∂ν U ) ∂µ U −1 ig ig 1 1 (∂ν ∂µ U ) U −1 + (∂µ U ) ∂ν U −1 ig ig → ∂µ U Aν U −1 − −∂ν U Aµ U −1 − = ∂µ U Aν U −1 − −∂ν U Aµ U −1 + = 1 1 (∂µ U ) ∂ν U −1 − (∂ν U ) ∂µ U −1 + ∂µ U Aν U −1 ig ig +U ∂µ Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 −U ∂ν Aµ U −1 − U Aµ ∂ν U −1 1 ∂µ U ∂ν U −1 − ∂ν U ∂µ U −1 + ∂µ U Aν U −1 ig = +U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 +U (∂µ Aν − ∂ν Aµ ) U −1 .38).39) . Aµ → U Aµ U −1 − 1 (∂µ U ) U −1 . ψ → ψU −1 . (12. (12.37) so that we can write the ﬁnite transformations (corresponding to (12.12.38) To construct the Lagrangian density for the dynamical part of the gauge ﬁeld in a gauge invariant manner. the tensor representing the Abelian ﬁeld strength (see (12.1 Non-Abelian gauge theories 493 U = e−iθ a (x)T a . U † = U −1 .

40). Aν ] U −1 − − ∂µ U Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 .39) and (12. Aν ] .42) so that Fµν transforms covariantly. in the present case.38). it is clear that. Fµν → U Fµν U −1 . It is now easy to construct the gauge invariant Lagrangian density for the dynamical part of the gauge ﬁeld as (quadratic in derivatives) . we see that unlike in the case of QED.38). here fµν = ∂µ Aν − ∂ν Aµ is neither invariant nor does it have a simple transformation under the gauge transformation (12. then under the gauge transformation (12. (12. Let us also note that under the gauge transformation ig [Aµ . if we deﬁne the ﬁeld strength tensor as Fµν = ∂µ Aν − ∂ν Aµ + ig [Aµ .40) Thus comparing (12.494 12 Yang-Mills theory Namely. 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 . (12.41) (12.

µ ν ν µ (12. therefore.38).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 . ν µ µ ν or.43) where a particular normalization for the trace of the generators is assumed (namely. From the cyclicity of trace this Lagrangian density is easily seen to be invariant under the gauge transformations (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. 2 4 (12. given by 1 a LQCD = − Fµν F µν a + iψ k γ µ (Dµ ψ)k − mψ k ψk . a a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac = −Fνµ .45) The complete Lagrangian density for QCD (quantum chromodynamics.34)) . 4 (12. (12. Aν .59) and the comments fol2 lowing that equation). or.12.1 Non-Abelian gauge theories 495 1 1 a Lgauge = − Tr Fµν F µν = − Fµν F µν a . the ﬁeld strength tensor (12. QCD corresponds to the particular case when G = SU (3)) is. Fµν = ∂µ Aν − ∂ν Aµ + ig Aµ .38) which have the inﬁnitesimal form (see (12. T a T b = 1 δab .41) is easily seen to take the form.46) where we can set m = 0 if the fermions (quarks) are massless. In components. more generally for fermions interacting with a non-Abelian gauge ﬁeld. a Fµν T a = ∂µ Aa − ∂ν Aa T a + igAb Ac T b . see (12.

T c . T a . f abp f pcq T q + f cap f pbq T q + f bcpf paq T q = 0. as we have noted in (12. T a = 0. T b + if bcp T p . f abp f pcq + f cap f pbq + f bcp f paq = 0. T c + T c.496 12 Yang-Mills theory δψ k = iǫa (x)ψ ℓ (T a )ℓk . (12. The algebra of the Hermitian generators. T b .48) it can be seen that the Jacobi identity (12. (12. T b + T b .49) Using (12. µ g (12. 1 ∂µ ǫa − f abc Ab ǫc .49) imposes a restriction on the structure constants of the group of the form if abp T p . has the form T a .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. (12. T b = if abc T c .18).48) and the Jacobi identity associated with this algebra is given by T a.50) From the structure of the Lie algebra we know that we can write the ﬁnite group elements in a unitary representation as (recall ﬁnite rotations and the angular momentum operators and see (12. In this case .37) as well) U (x) = eiθ a (x)T a . (12.47) Let us discuss brieﬂy some of the properties of the Lie algebra of the symmetry group G. or. T c + if cap T p . T a = 0. δAa = µ δψk = −iǫa (x)(T a )kℓ ψℓ . then the Lie algebra breaks up into two commuting non-Abelian subalgebras. or.

1 Non-Abelian gauge theories 497 the group G is a direct product of two independent non-Abelian Lie groups. we will assume that the symmetry group G is simple. T b T c (12. (12.54) Tr Tr T a T b T c − Tr T b T a T c .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 . let us next note that the quantity habc = = T a. using the commutation relation (12. For any representation of a simple Lie group we can write Tr T a T b = C2 δab .53) Using the cyclicity property of the trace. a = b.55) where the index c is ﬁxed (and not summed). On the other hand. we note that . (12. It depends on the representation but not on the indices a and b.52) We note that C2 is a normalization constant which determines the values of the structure constants. A non-Abelian Lie group that cannot be so factorized is called a simple Lie group.12. is completely antisymmetric in all its indices. Furthermore. In all our discussions. Tr T a T b = (12. To prove this let us note that we can always diagonalize the tensor Tr T a T b such that (this is a symmetric real matrix in the “ab” space) 0 Ka if if a = b. A direct product of simple Lie groups is called semi-simple.

(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. (12. (12. since habc is completely antisymmetric. 2. we have hacb = −habc . we note here that it is (12.60) then T (R) is known as the index of the representation R. a.59) which is used to show that the structure constants for a semi-simple group are completely antisymmetric in the indices.) We note that if we write Tr T a T b = T (R) δab .61) . · · · .57) Using this and comparing (12. However. Furthermore. n = 1.56) with the index b ﬁxed.55) and (12. (12. 2. T c T b = if acp Tr T p T b = if acp Kp δpb = iKb f acb = −iKb f abc .56) we conclude that Kb = Kc = K.59) where the constant C2 depends only on the representation. (12. b = 1.43).498 12 Yang-Mills theory hacb = Tr T a. m. Similarly.58) (12. dim R. · · · . we have (T a T a )mn = C(R) δmn . This shows that we can write Tr (T a T b ) = C2 δab . (12. dim G.

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

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

µ (12. or. (12.12.72) where Aµ = Aa T a and ǫ = ǫa T a (with T a in any representation). the gauge ﬁeld must transform in the adjoint representation. T b . 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 . T b a = iǫb Fµν T a .67) (adj) Dµ ǫ a = ∂µ ǫa − gf abc Ab ǫc .1 Non-Abelian gauge theories 501 δFµν = iǫb Fµν . ab µν (12. or. a a δFµν T a = iǫb Fµν if abc T c . For example. let us scale the ﬁeld variables as .71) can also be written in the matrix form as (adj) Dµ ǫ = ∂µ ǫ + ig[Aµ . ǫ]. (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.70) Comparing this with the transformation of the fermions in (12.34) we conclude that the ﬁeld strength Fµν as well as the gauge ﬁeld Aµ transform according to the adjoint representation of the group.73) Let us note the following features of this Lagrangian density. let us note here that the covariant derivative in the adjoint representation (12. (It does not matter what representation the matter ﬁelds belong to.) For completeness.

74) Thus we note that in this theory.45)) 1 a 1 (12.43) (or (12.76) . ν µ µ ν = ∂µ Aν − ∂ν Aµ + ig[Aµ . (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 . in the case of non-Abelian symmetry they must couple to themselves and possess self interactions. the pure Yang-Mills theory is an interacting theory unlike the Maxwell theory. The other feature to note is that unlike the photon ﬁeld. g µ It follows then that 1 a Fµν → ∂µ Aa − ∂ν Aa − f abc Ab Ac .502 12 Yang-Mills theory Aa → µ 1 a A . the coupling constant can be scaled out and written as an overall multiplicative factor in the Lagrangian density. here the gauge ﬁelds have self interaction (they interact with themselves) and. Since the gauge ﬁelds couple to any ﬁeld (particle) carrying charge of the symmetry group. 2 4 where the ﬁeld strength tensor is given by (see (12. 12. We note that the Lagrangian density for the Yang-Mills theory is given by (12. therefore.44)) Fµν a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac .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 9.75) LYM = − Tr Fµν F µν = − Fµν F µν a . Aν ]. (12. In the present case. Physically we understand this in the following way.41) and (12. 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.

72)). −∂µ F µν a − −gf abc Ab F µν c = 0. ∂∂µ Aa ∂Aa ν ν ∂µ or. we can write the Euler-Lagrange equations as Dµ F µν = ∂µ F µν + ig[Aµ .77) The Lagrangian density (12.76) as a a F0i = ∂0 Aa − ∂i Aa − gf abc Ab Ac = Ei . or. we can identify the non-Abelian electric and magnetic ﬁelds from (12.68)).12. µ (12. In the matrix notation (see (12.79) where the covariant derivative is deﬁned to be in the adjoint representation of the group (see (12.2 Canonical quantization of Yang-Mills theory 503 As in the case of the Abelian theory (Maxwell theory). 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. F µν ] = 0.67) and (12.75) and have the form ∂L ∂L − = 0.78) The Euler-Lagrange equation for the theory are obtained from (12.75) is the transverse projection operator just like in the case of Maxwell’s theory. i 0 0 i a Fij a = ∂i Aa − ∂j Aa − gf abc Ab Ac = −ǫijk Bk .80) Let us note that the coeﬃcient matrix of highest derivatives in (12. 2 2 4 LYM = Tr Ei Ei − (12. j i i j (12. Therefore.77) as 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a .75) can also be written explicitly in terms of the non-Abelian electric and magnetic ﬁeld strength tensors (12. (12. µ (Dµ F µν )a = ∂µ F µν a − gf abc Ab F µν c = 0.68) and we are suppressing the symbol “(adj)” to denote the covariant derivative in the adjoint representation for simplicity) .

ν.84) (12.85) a where the non-Abelian electric ﬁeld Ei (x) is deﬁned in (12. (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. (12. we have Π0 a (x) = −F 00 a (x) = 0. Let us obtain the momenta canonically conjugate to the ﬁeld variables Aa from (12. This implies that 0 Aa is like a c-number quantity which commutes with every other 0 operator in the theory. 0 or. Thus we can choose a gauge condition and set it equal to zero.82) which is a constraint relation. g (12. the constrained structure of the theory is already obvious in the Euler-Lagrange equation of motion (12. we choose Aa (x) = 0.83) 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). Thus. we see that we have only 3N canonical momenta where N = dim G.77). as in the case of Maxwell’s theory and a Πi a (x) = −F 0i a = Ei (x). For example.79) for ν = 0. The momentum conjugate to Aa does not exist. Di F i0 a = 0. ˙ a (x) ∂ Aµ (12. requiring that . Namely.504 12 Yang-Mills theory δAa = µ 1 (Dµ ǫ(x))a .81) In fact. namely. A0 (x) = 0. (12.

U −1 ∂0 U = igA0 . we note from (12.12.77) takes the form a ˙ Ei = Aa .88) (namely. (12. let us also note that the potential term in (12. We note that the relation (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 . Let us note that with the gauge condition (12.91) ig ig so that there is a cyclic variable in the theory that needs to be separated. (12.91) can be inverted to give .86) the non-Abelian electric ﬁeld in (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 .86) the Lagrangian density (12. ig (12. in the present case.88) 2 2 4 LYM = Tr Ei Ei − where we have used the fact that in the gauge (12.82) that Di Ei = ∂i Ei + ig[Ai .87) we can determine the parameter of transformation and we see that a gauge ﬁeld can always be transformed to satisfy (12. we expect the physical dynamical degrees of freedom of a massless gauge ﬁeld to be transverse. However.2 Canonical quantization of Yang-Mills theory 505 A′ = U A0 U −1 − 0 or.89) From our discussion in the case of the Maxwell theory. i or. Ei ] = 0. (12. On the other hand. ˙ Ei = Ai .90) so that neither the gauge ﬁeld Ai nor the electric ﬁeld Ei is transverse ˙ (recall that Ei = Ai ). (12.86). 1 (∂0 U ) U −1 = 0.

93) 1 1 (∂i S) (∂i S −1 ) − ∂i (∂i S)S −1 ig ig 1 ∂i (∂i S)S −1 − [Ai .92) in the intermediate step and the covariant derivative in the adjoint representation is deﬁned for a matrix function G as (see (12. (12.91).93). let us require the new ﬁeld variable Ai to be transverse. ¯ ∂i Ai = 0. Imposing this condition in (12.91) that . (∂i S)S −1 ] ig 1 Di (A) (∂i S)S −1 . ig (12.72)) Di (A)G = ∂i G + ig[Ai . 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. G]. (12.92) ig ig Keeping in mind that we would like the dynamical variables to be ¯ transverse. namely.94). then we can deﬁne a new ﬁeld variable Ai satisfying the transversality condition (12. We note next from the ﬁeld redeﬁnition in (12.94) where we have used (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.95) This shows that if we can ﬁnd a gauge transformation which satisﬁes ¯ (12.

(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 . Here Di (A) is the covariant derivative in the adjoint representation (see (12.12.96) ˙ ¯ where we have used ∂0 S −1 = −S −1 SS −1 . ig where we can identify 1 ˙ ¯ ¯ E i = Ai − Di (A)f. ig (12. ig (12. S −1 S] − ∂i (S −1 S) S −1 ig 1 ˙ ¯ ¯ ˙ = S Ai − Di (A)(S −1 S) S −1 ig 1 ˙ ¯ ¯ = S Ai − Di (A)f S −1 .95)) with respect to ¯ the gauge ﬁeld Ai and we have identiﬁed ˙ f = S −1 S.86) we can write 1 ˙ ˙ ¯ ¯ Ei = Ai = S Ai − Di (A)f S −1 = SE i S −1 . decomposing E i into its transverse and longitudinal components as T E i = E i + ∂i φ.98) (12.96) we note that in the gauge (12. where (12.100) .99) Furthermore.97) From (12.

T ¯ ¯ ∂i ∂i φ + ig∂i [Ai . φ] + ig[Ai . or. E i ]S −1 ig ¯ = S(∂i E i + ig[Ai . φ] = −ig[Ai .508 12 Yang-Mills theory ∂i E i = 0. ig ∇2 ˙ ¯T ¯ Ei = Ai .91) and (12.104) T ¯ where we have used ∂i Ai = 0 in the intermediate step. (12. E i ] = 0.99) that 1 1 ¯ ∂i Di (A)f.101) we conclude from the deﬁnition in (12. E i ] = ρ. E i ] = 0. ¯ ¯ ∂i ∂i φ + ig[Ai . (12.98). As a result of this relation. or. Ei ] ¯ = ∂i SE i S −1 + ig[S(Ai − 1 −1 S (∂i S))S −1 . T ∂i E i = ∂i ∂i φ = ∇2 φ. E i ])S −1 ¯ = S(Di (A)E i )S −1 . We can invert this relation to determine .91). ¯ ∂i Di (A)φ = ρ.103) where we have used (12. Di (A)Ei = ∂i Ei + ig[Ai . SE i S −1 ] ig = (∂i S)E i S −1 + S(∂i E i )S −1 − SE i S −1 (∂i S)S −1 ¯ +igS[Ai − 1 −1 S (∂i S). (12. E i + ∂i φ ] = 0.90) takes the form ¯ ¯ Di (A)E i = ∂i E i + ig[Ai . or. or. T ¯ ∂i ∂i φ + ig[Ai .102) Let us next note that under the ﬁeld redeﬁnition (12. the constraint equation (12. φ=− (12.

93).100) we obtain T ˙ ¯ E i = E i + ∂i φ = Ai + ∂i (12.106) Using the deﬁnition in (12. ∂i Di (A) = Di (A)∂i which has been used together with integration by parts).98) and the decomposition in (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 . ¯ We have isolated the dynamical variables of the theory to be Aa i which are transverse and the conjugate momenta can be determined from (12.107) where we have neglected total divergence terms in the intermediate ¯ ¯ steps (note also that because of (12. (12.108) . i ˙ ¯a ∂A i (12. the Lagrangian density (12.106) as well as the cyclicity of trace.12.91). ¯ 2 ∂j Dj (A) (12. ∂i Di (A) 1 ¯ ρ.105) so that from (12. ∂j Dj (A) (12.2 Canonical quantization of Yang-Mills theory 509 φ = 1 ¯ ρ.107) to be Π ia = ∂LYM ˙ ¯ = Aa .

109) with all other brackets vanishing. Π (y)} = δab δi j (x − y). controlled by the eigenvalues of the operator ∂i Di (A). (12. A′ a = Aa + i i g such that ¯i ∂i A′ a = 0. this last term has the form ρ 1 2 ρ which corresponds to ∇ the long range Coulomb interaction).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). This can be seen from the ¯ fact that if we have a ﬁeld conﬁguration Aa which is transverse. Thus the long range behavior of this theory is still not well deﬁned in the Coulomb gauge.31).111) (12. i ¯ ∂i Aa = 0.107) has exactly the same form as in the Abelian theory except for the last term. Let us note here that the long range behavior of the interaction (and. namely the Coulomb gauge does not uniquely deﬁne the gauge ﬁelds. Here we see that since the gauge ﬁelds are self interacting. therefore. ¯ Gribov has shown that the operator ∂i Di (A) does possess zero modes or eigenvectors with zero eigenvalue.110) then we can make a time independent inﬁnitesimal gauge transformation (under which the theory is invariant) 1 ¯ǫ a ¯ ¯ Di (A)¯ .e. Here the transverse delta function corresponds to the one already deﬁned in the case of Maxwell’s theory in (9.112) . 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. We note that the Lagrangian density in (12. i TR jb (12. i (12. therefore. This is related to the question of gauge ﬁxing. i. even in the absence of other matter ﬁelds there is a long range interaction. of the theory) ¯ is.

114) namely. for example. let us modify the theory (12. There now exist prescriptions to ¯ take care of this problem.111) that this has a singularity at g = 0. we recall that in QED (see (9.2 Canonical quantization of Yang-Mills theory 511 ¯ provided ǫa corresponds to a zero mode of ∂i Di (A). However. The other way of saying this is that within the framework of perturbation theory we are interested in inﬁnitesimal gauge transformations. therefore. But let us note from the form of A′ a in i (12.129)) (∂ · A) = 0.) g Hence such gauge ﬁeld conﬁgurations can only be reached nonperturbatively. much like in the Maxwell theory. This shows that ¯ the Coulomb gauge does not really specify the gauge ﬁeld conﬁguration uniquely. known as the Gribov ambiguity. classically we can impose the condition ∂ · A = 0. namely.113) The additional term clearly breaks gauge invariance and. by modifying the theory. very much along the lines of the Abelian theory that we have discussed in chapter 9. ∂ · A behaves like a free ﬁeld and.12. consequently. there are serious diﬀerences from Maxwell’s theory. Therefore. the hypersurface deﬁning the gauge choice intersects the gauge orbits more than once. (It is of the form O 1 . For example. we would not worry about it within the context of perturbation theory. (12. As long as we are within perturbation theory we can neglect such conﬁgurations. 4 2 (12. we can also try to quantize the non-Abelian gauge theory covariantly.75) as (see (9. The canonical quantization of Yang-Mills theories does not have manifest covariance very much like the Maxwell theory. Gribov ambiguity is a phenomenon associated with large gauge transformations and is an important issue since it leads to nonperturbative eﬀects.125) with J µ = 0) 1 1 a L = − Fµν F µν a − (∂µ Aµa )2 .115) . In other words. Thus. (12. However. in the present case. On the other hand. makes the theory nonsingular. We can show that this nonuniqueness. occurs in all gauges other than the axial gauge.

we need to analyze the question of modifying the theory in a more systematic and detailed manner. 12.) Correspondingly the naive analog of the Gupta-Bleuler condition for non-Abelian gauge theories does not seem to exist. in contrast to the Abelian theory.116) In the case of the Yang-Mills theory.117) = 0. since it is not invariant under time evolution. even classically the equations of motion are given by ∂µ F µν a − gf abc Ab F µν c + ∂ ν (∂ · Aa ) = 0. we note that (∂ · Aa ) does not behave like a free ﬁeld and. just modifying the Lagrangian density as in (12. the additional term in (12. We would see next how we can derive intuition on this important question from the path integral quantization of the theory. (12. (Namely.512 12 Yang-Mills theory which then translates to the Gupta-Bleuler condition on the physical states ∂ · A(+) |phys = 0.113) does not seem to be suﬃcient in contrast to the Abelian gauge theory. nor can we think of the supplementary condition ∂ · Aa(+) |phys = 0. µ Contracting with ∂ν .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 . (12.119) in a physically meaningful manner.113) has truly modiﬁed the theory. we obtain (∂ · Aa ) = gf abc ∂ν Ab F µν µ c (12. the physical subspace would keep changing with time which is not desirable. consequently. In other words. it cannot be uniquely decomposed into a positive and a negative frequency part (in a time invariant manner). however. Therefore. Furthermore. (12.118) Thus. since (∂ · Aa ) is not a free ﬁeld.

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

125) With these basics a simple functional integral for a quadratic action such as in (12. G = (−∂µ ∂ µ − m2 )−1 . namely. δVi φi = (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) .123) Clearly in the limit of vanishing volume we recover lim φi = φ(x).126) where the Green’s function G(x − y) is formally the inverse of the operator in the quadratic part of the action. i (12.514 12 Yang-Mills theory The integration in (12. (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. (12. (12. Basically. When we have an .126) is independent of ﬁeld variables and is a constant (possibly divergent) and can be incorporated into the normalization constant N .) The determinant in (12.124) δVi →0 and in terms of these discretized ﬁeld variables. Let us divide the entire space-time into inﬁnitesimal cells labelled by “i” of volume δVi and deﬁne 1 δVi d4 x φ(x).120) can be evaluated in a straightforward manner. the functional integration is deﬁned (up to a normalization constant) as Dφ = dφi .120) is known as a functional integration and is deﬁned as follows.

129) µν A ν )+(J µ . the operator P µν is a projection . everything is a classical function. cannot be obtained in a closed form. (12.120). The advantage of using the path integral description lies in the fact that there are no operators in (12. However. Aµ ) = − ∂ µ ∂ ν ) δ4 (x − y). the generating functional in the path integral formalism is given by (J ) [A Z [Jµ ] = eiW [Jµ ] = N = N DAµ eiS µ] (12.128) where J µ represents a conserved current (source).131) d4 x J µ (x)Aµ (x). in general.P −1 J ν µν ).132) where we have used the notation in (12. as we have seen before (see (9. In this case.131).122)) we have identiﬁed P µν (x − y) = (η µν (J µ . let us go back to the Maxwell theory 1 L(J) = − Fµν F µν + J µ Aµ .3 Path integral quantization of gauge theories 515 interacting theory (not quadratic in the ﬁeld variables) the functional integral. In this case. With this brief introduction to path integral description. (12.124)). (12. 4 (12.A µ) DAµ ei[ 2 (Aµ .130) where N is a normalization constant and (see (9.P 1 ] . This is a Gaussian functional integral and we can evaluate this using our earlier result in (12.126) Z [Jµ ] = eiW [Jµ ] = N (det (−P µν ))− 2 e− 2 (J 1 i 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.123) and (9.

This is a U (1) symmetry and the symmetry is noncompact since the parameter of transformation θ(x) can take any real value. is constant on such orbits. the generating functional. (As we have seen the form of transformation for the gauge ﬁelds (12. Clearly therefore. ie Aµ → A(θ) = U Aµ U −1 − µ where U (θ) = e−iθ(x) . (The operator possesses zero modes and.133) (12. e Aµ → A(θ) = Aµ + µ (12. the determinant of P µν vanishes.) The source of the diﬃculty is not hard to see. 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. consequently. even in the absence of any sources. on the other hand.133) in terms of U (θ) is quite general and holds for the non-Abelian theories as well.) (θ) For a ﬁxed Aµ . This implies that the generating functional does not exist in this case. (12. we can write the transformation as 1 ∂µ θ(x). The Lagrangian density for Maxwell’s theory is invariant under the gauge transformation 1 (∂µ U )U −1 . x (12.516 12 Yang-Mills theory operator for transverse photons.135) which is the familiar gauge transformation for Maxwell’s theory. Therefore.136) .134) We can immediately see that for a U (1) gauge group. The action S. the inverse of the matrix cannot be deﬁned either. is proportional to the “volume” of the orbits denoted by dθ(x). The longitudinal vectors kµ (or ∂µ ) are its eigenvectors with zero eigenvalue.

) 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. (12. In this way. (12. (12. Physical quantities are. the Coulomb gauge. if F [Aµ (x)] = 0. This procedure is known as gauge ﬁxing and the condition F [Aµ (x)] = 0. we pick up a representative gauge ﬁeld from each gauge orbit. A3 (x) = 0. the temporal gauge. A0 (x) = 0.3 Path integral quantization of gauge theories 517 (In the non-Abelian case. gauge invariant and do not depend on the choice of the hypersurface (gauge). µ (12. Thus.138) has a unique solution for some θ(x).140) and so on. namely. we can ﬁnd a gauge trans(θ) formed Aµ which does and F [A(θ) (x)] = 0. of course. 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. Rather we should integrate over each orbit only once. then even if Aµ does not satisfy the condition. the Lorentz/Landau gauge. for example. We recognize that we should not integrate over all gauge ﬁeld conﬁgurations because they are not really distinct. this should be replaced by the group invariant Haar measure x dU (x). the axial gauge. here are a few of the familiar gauge ﬁxing conditions F [Aµ (x)] = ∂µ Aµ (x) = 0. ∇ · A(x) = 0. The way this is carried out is by choosing a hypersurface which intersects each orbit only once. We can .12.139) is known as the gauge condition (or gauge ﬁxing condition).137) deﬁnes the hypersurface which intersects the gauge orbits once.

On the other hand. known as the Faddeev-Popov determinant. That is (for the Abelian . in general. is gauge invariant which can be (θ ′ ) seen as follows. Then ∆−1 [A(θ ) ] = FP µ x ′ dθ(x) δ F [A(θ+θ ) (x)] µ (θ) dθ(x) δ F [Aµ (x)] x −1 ′ = = ∆FP [Aµ ] . Let us make a gauge transformation Aµ → Aµ in (12. Let us deﬁne ∆−1 [Aµ ] = FP x dθ(x) δ F [A(θ) (x)] . be thought of as a completeness relation.142) which can.141). the photon propagator) although the physical S-matrix (the scattering matrix) elements are gauge independent. (12. (The integration measure should be dU (x) which is essential in the case of non-Abelian theories. (12.143) This follows from the fact that the measure in the group space is invariant under a gauge transformation. we know from ordinary perturbation theory that the Green’s functions are. Therefore.141) This can also be written as ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] = 1. for example.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). Thus we have to ﬁx a gauge without which even the Cauchy initial value problem cannot be solved. therefore.) Note that the quantity ∆FP [Aµ ].) To extract out the inﬁnite gauge volume factor. it would lead only to gauge invariant Green’s functions. let us do the following trick due to Faddeev and Popov. (Only physical quantities need to be gauge independent. µ (12. gauge dependent (recall.

12. Therefore.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 .144) In the non-Abelian case. µ under which the generating functional takes the form dθ(x) δ F [Aµ (x)] eiS x (J ) [A µ] (J ) [A µ] (12. (12. However.141) that . let us make an inverse gauge transformation Aµ → A(−θ) . d(U U ′ ) = dU. dθ(x) = (12. we still have to determine what ∆FP [Aµ ] is. (12.145) Remembering that ∆FP [Aµ ] is gauge invariant we can now insert this identity factor into the generating functional to write (J ) [A Z[Jµ ] = N DAµ ∆FP [Aµ ] (θ) dθ(x) δ F [Aµ (x)] eiS x µ] . (12.148) where the (inﬁnite) gauge volume has been factored out and absorbed into the normalization constant N of the path integral. namely. we should have the Haar measure which is gauge invariant.146) Furthermore. To do this let us note from (12. this gives the correct functional form for the generating functional.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) .

using (12. be thought of as the Jacobian that goes with a given gauge choice. On the other hand. 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). therefore.152) The Faddeev-Popov determinant can. (12.153) .149)-(12.520 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. Thus. (Here we can completely ignore the problem of Gribov ambiguity associated with large gauge transformations. we have θ(x) = 0.) 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). for gauge ﬁelds which satisfy the condition F [Aµ (x)] = 0.149) δθ δF F [Aµ (x)]=0 (θ) . (12.151) θ=0 .151) we determine δF [Aµ (x)] ∆FP [Aµ ] = det δθ(y) (θ) (12.150) (12. 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.

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 ).157) and ξ is known as the gauge ﬁxing parameter. (12.3 Path integral quantization of gauge theories 521 where f (x) is independent of Aµ .155) Following ’t Hooft.154) into the functional integral. we can now do the following (also known as the ’t Hooft trick).12. (12. Thus the generating functional in this case is given by Z[Jµ ] = N DAµ ∆FP [Aµ ]δ F [Aµ (x)] − f (x) eiS (J ) [A µ] . We note further that since δF [Aµ (x)] δθ(y) (θ) ∆FP [Aµ ] = det θ=0 . (12. We note that physical quantities are independent of f (x).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). (12. The Faddeev-Popov determinant is unchanged by this modiﬁcation because f (x) does not depend on Aµ (x). Then we can insert the identity ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] − f (x) = 1.158) . (12. Thus.

Thus the generating functional now takes the form Z[Jµ ] = eiW [Jµ] = N DAµ Dc Dc eiSeﬀ (J ) [Aµ .e.162) . These ﬁelds are known as Faddeev-Popov ghosts and as is obvious from their anticommutation relations. c(y)]+ = 0. (12. 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. but opposite statistics.160) Here we have introduced two independent ﬁctitious ﬁelds c(x) and c(x).c.). [c(x). [c(x). c(y)]+ = 0. graphs involving these ﬁctitious particles in closed loops must have an additional (−1) factor just like the fermions. they obey anticommutation rules. i. known as ghost ﬁelds. δF [Aθ (x)] µ d xd y c(x) δθ(y) 4 4 (12.c] . (12.( 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 . to write the Faddeev-Popov determinant in the form of a ghost action (action involving ghost ﬁelds). [c(x). (12. c(y)]+ = 0..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.161) Thus although these ﬁelds behave as scalar ﬁelds under Lorentz transformations.159) θ=0 c(y).

for example. 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. our responsibility to show that the physical interpretation of the theory has not changed.153)) F [Aµ (x)] = ∂µ Aµ (x) = f (x). Let us now look at a simple gauge ﬁxing condition in Maxwell’s theory. (J) (J) (12. Otherwise it fails its purpose and will not correspond to an acceptable gauge ﬁxing condition. for the moment let us emphasize that the purpose of a gauge ﬁxing Lagrangian density is to break gauge invariance.165) . this modiﬁes the starting theory. Of course.3 Path integral quantization of gauge theories 523 where Seﬀ [Aµ . c] = S (J) [Aµ ] + SGF + Sghost = d4 x Leﬀ [Aµ .163) Thus we can summarize what we have done so far. the gauge ﬁxing Lagrangian density should make the quadratic part of the eﬀective Lagrangian density nonsingular.142) that both these actions arose from inserting a factor of unity into the functional integral). (12. we start with a gauge invariant Lagrangian density and add to it a gauge ﬁxing Lagrangian density determined by the gauge ﬁxing condition that we want to work with. c] .12. It is. c. To do covariant perturbation theory in the path integral formalism. the gauge ﬁxing Lagrangian density (see (12. namely. of course.157)) has the form LGF = − 1 1 (F [Aµ (x)])2 = − (∂µ Aµ (x))2 . However.164) In this case. It is worth noting at this point that we have modiﬁed our starting Lagrangian density by a series of formal manipulations. 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. 2ξ 2ξ (12. the covariant condition (see (12. We would show this in the next chapter.

the ghost ﬁelds are interacting and have to be present.524 12 Yang-Mills theory It is clear that this provides longitudinal components to the quadratic terms in ﬁelds and hence breaks gauge invariance. (12. 4 2ξ (12.125).166) = θ=0 1 e xδ 4 (x − y). (12. however.152)) δF [Aµ (x)] δθ(y) (θ) 1 µ ∂ θ(x) .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 .169) is nothing other than Maxwell’s theory in the Feynman-Fermi gauge (9. therefore.169) We note here that the ghost ﬁelds are noninteracting in the case of the Maxwell theory in ﬂat space-time and. To obtain the corresponding ghost Lagrangian density for this gauge choice. e (12. This . the ghost action (12. we may neglect them and then for ξ = 1 we recognize that our eﬀective Lagrangian density (12. In non-Abelian gauge theories. 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 .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. we note that F [A(θ) (x)] = ∂µ A(θ)µ (x) = ∂µ Aµ (x) + µ so that (see (12.

113) of the non-Abelian gauge theory failed to be suﬃcient unlike in the Maxwell theory (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. This naive counting works pretty well as we will see later.171) . being a scalar.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. Furthermore. we cannot neglect them even if they are noninteracting particularly when we are doing calculations at ﬁnite temperature. has only one ﬁeld degree of freedom (we will discuss the question of hermiticity of the ghost ﬁelds in the next chapter). we see that the equation of motion for F is given by ξF (x) = ∂µ Aµ (x).) Let us now go back to the eﬀective Lagrangian density (12. namely. For example. the ghost Lagrangian density was missing). To do this let us rewrite the eﬀective Lagrangian density (12. they anticommute even though they are scalar ﬁelds. (The ghost degrees of freedom subtract because they have the unphysical statistics. since the ghost ﬁelds and the ghost action are really necessary for the unitarity of the S-matrix. One way of looking at the ghost ﬁelds is as if they are there to subtract out the unphysical ﬁeld degrees of freedom.12.170) 4 2 Here we have introduced an auxiliary ﬁeld F which does not have any dynamics.169). the Aµ ﬁeld has four ﬁeld degrees of freedom.164). On the other hand each of the ghost ﬁelds.169) as ξ 1 (J) Leﬀ = − Fµν F µν + F 2 −F (∂µ Aµ )+∂µ c∂ µ c+J µ Aµ . It is also true that when Maxwell’s theory is coupled to a gravitational ﬁeld.169) and ask how we can recover the Lorentz gauge from the Lorentz like covariant gauge condition in (12. (12. On the other hand. the ghost ﬁelds automatically couple to the geometry also. (12. Hence omitting the ghost Lagrangian density in such a case would lead to incorrect results.

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

according to Faddeev and Popov. The µ correct description for the generating functional. In particular. therefore. As a result.176) In this way. In this formalism the ﬁeld variables are treated as classical variables. non-invertible. if we take Linv . in this formalism there is no reference to the Hilbert space of the theory. 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. their contributions cancel out and eﬀectively we are left with only two physical transverse degrees of freedom. 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] . However. 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. In summary. The second method of quantizing the Abelian gauge theory is through the method of path integrals. we note that gauge invariance puts a very strong constraint on the structure of the Lagrangian density for the gauge ﬁeld.3 Path integral quantization of gauge theories 527 result. (12. 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. (12. In the path integral formalism.175) where AT denotes the transverse physical degrees of freedom.12. the coeﬃcient matrix of the quadratic terms in the Lagrangian density is singular and. is given by (J ) [AT ] µ Z[Jµ ] = N = N DAT eiS µ DAµ ∆FP [Aµ ]δ(∂ · A(x)) eiS (J ) [A µ] . as the Lagrangian density describing the .

µ a Fµν (x) = ∂µ Aa (x) − ∂ν Aa (x) − gf abc Ab (x)Ac (x).164). In order to circumvent this diﬃculty.528 12 Yang-Mills theory dynamics of the gauge ﬁeld theory.67) and (12. On the other hand.179) ν µ µ ν with g denoting the coupling constant of the theory. and to compensate for that we have to add a corresponding Lagrangian density for the ghost ﬁelds following the prescription of Faddeev and Popov. 4 (12. (12. then we cannot deﬁne propagators and the entire philosophy of doing perturbative calculations with Feynman diagrams breaks down. in general. adding a gauge ﬁxing Lagrangian density changes the theory.75)) 1 a Linv = − Fµν F µνa . consists of adding to the invariant Lagrangian density a gauge ﬁxing Lagrangian density of the form . g (ǫ)a Aa (x) → Aµ (x) = Aa (x) + µ µ (12.178) where ǫa (x) is the inﬁnitesimal parameter of transformation. The standard covariant gauge ﬁxing (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. 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. As we have seen. Let us now apply these ideas to the study of the non-Abelian gauge theory. the Lagrangian density for the gauge ﬁelds is given by (see (12. in 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 (12.76)) Dµ ǫa (x) = ∂µ ǫa (x) − gf abc Ab (x)ǫc (x).177) which is invariant under the inﬁnitesimal gauge transformation 1 Dµ ǫa (x).

182) We note that for the covariant gauge choice (12.183) = 1 µ ab 4 ∂ D δ (x − y) g x xµ µ = ∂x ∂xµ δab − gf acb Ac (x) δ4 (x − y). we can write the ghost action corresponding to this gauge choice as (see (12.12.184) µ 1 Consequently. 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).160) and note that since ǫa is an inﬁnitesimal parameter. (12. δǫb (y) (ǫ) (12. we can write 1 (ǫ)a F a [A(ǫ) (x)] = ∂ µ Aµ (x) = ∂ µ Aa (x) + Dµ ǫa (x) .181) Here ξ represents a real arbitrary constant parameter known as the gauge ﬁxing parameter.122)) δF a [Aµ (x)] δǫb (y) (ǫ) (12. (12. Following the prescription of Faddeev and Popov.180) which corresponds to a gauge ﬁxing condition of the form F a [Aµ (x)] = ∂µ Aµa (x) = f a (x). µ µ g so that we have (see (12. 2ξ (12.3 Path integral quantization of gauge theories 529 LGF = − 1 (∂µ Aµa (x))2 . 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 .181) that we are using.

let us consider the gauge theory of the antisymmetric tensor ﬁeld Aµν (x) . merely correspond to a multiplicative factor of unity (see (12. in the path integral formulation. 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. With all these modiﬁcations. In a deeper sense. 12. therefore. (12.4 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. cannot be neglected even in ﬂat space-time. However.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 . in some cases we have to work through the details of this analysis rather carefully in order to obtain the correct result.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. As we have mentioned earlier.142)) which does not change the physical content of the theory. the gauge ﬁxing and the ghost Lagrangian densities.186) = − Fµν F µνa − 4 2ξ We note that the ghost ﬁelds in the present case are interacting unlike in Maxwell’s theory and. (12.

µ.12. since Fµνλ is completely antisymmetric in all the indices. Let us consider only the free Lagrangian density for this ﬁeld deﬁned by 1 L = − Fµνλ F µνλ . (12. 1. 2. ν = 0. (12. the canonical momenta are antisymmetric (like the ﬁeld variables). This can be seen from the deﬁnition of the conjugate momenta ∂L = −F 0µν (x).190) so that not all ﬁeld variables are independent.191) The independent ﬁeld variables of the theory are A0i . (12.188) where the ﬁeld strength tensor corresponds to the totally antisymmetric tensor Fµνλ = ∂[µ A νλ] = ∂µ Aνλ + ∂ν Aλµ + ∂λ Aµν . 6 (12. we conclude that .189) Naive counting shows that Aµν has six ﬁeld degrees of freedom.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. ˙ ∂ Aµν (x) Πµν (x) = (12.4 Path integral quantization of tensor fields 531 Aµν (x) = −Aνµ (x). 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 . However.192) namely. 3. Furthermore. (12.

we note that the theory still possesses a residual (static) gauge invariance under the transformation δAij = ∂i θj (x) − ∂j θi (x). However.198) actually represents only two independent conditions. 3.195) (12. However.198) This would seem like three constraints and hence we would naively conclude that this theory has no dynamical degrees of freedom. With these conditions we have ˙ Πij = −F0ij = −Aij . the corresponding ﬁeld variables are like c-number quantities and we can choose the gauge condition A0i (x) = 0. on closer inspection we notice that the Coulomb gauge condition (12.532 12 Yang-Mills theory or.193) Since these momenta identically vanish.196) Thus it would seem that the theory has truly three degrees of freedom. i.194) (12.197) so that we can impose a Coulomb gauge condition of the form ∂i Aij (x) = 0.188) takes the form 1 1 L = − Fµνλ F µνλ = − 3F0ij F 0ij + Fijk F ijk 6 6 1 = − 3Πij Πij + Fijk F ijk . Π0µ = −F 00µ = 0. 3. We can see this by writing out the gauge condition explicitly . (12. k = 1. (12. 2. j. (12. and the Lagrangian density (12. Π0i = 0. 2. i = 1. 6 (12.

.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. Thus. ∂ρ Fµνλ − ∂λ Fρµν + ∂ν Fλρµ − ∂µ Fνλρ = 0.202) where φ(x) represents a real scalar ﬁeld. massless real scalar ﬁeld describing a single degree of freedom. (12. 3 are (12. Since Fµνλ represent a totally antisymmetric third rank tensor in four space-time dimensions satisfying the Bianchi identity (see (12.189)).203) We recognize this as the Lagrangian density for a free.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). 2 (12. (12. The fact that the theory has only one degree of freedom can also be seen in the following manner. 2. the three conditions corresponding to j = 1. 12 2 (12. ∂1 A13 + ∂2 A23 = 0. the Lagrangiandensity (12. In this case.199) ∂2 A21 + ∂3 A31 = 0.200) It is now clear that any two of the three conditions lead to the third condition so that there are only two independent conditions. 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). ∂1 A12 + ∂3 A32 = 0.4 Path integral quantization of tensor fields 533 ∂1 A1j + ∂2 A2j + ∂3 A3j = 0.12.

According to our earlier discussions. 6 L=− (12. so that we have (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). 2ξ LGF = − (12.205) corresponding to the choice of gauge condition F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). the starting gauge invariant Lagrangian density has the form 1 Fµνλ F µνλ . Hence to write the determinant as an action we need ghost ﬁelds which carry a vector index and we have . (12.204) and we add to it the covariant gauge ﬁxing Lagrangian density (ξ is the gauge ﬁxing parameter) 1 (∂µ Aµν )2 .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)). (12.534 12 Yang-Mills theory Let us now look at the path integral quantization of this theory.208) This is a matrix with two vector indices.

Thus the eﬀective Lagrangian density appears to have 6 − 2 × 4 = 6 − 8 = −2.12. (12. The ghost ﬁelds being vectors have four degrees of freedom each and since they satisfy anticommutation relations.212) . 2 (12. The ﬁeld variable Aµν has six degrees of freedom.4 Path integral quantization of tensor fields 535 Sghost = − = − = = 1 2 d4 xd4 y cν (x) δF ν [Aµσ (x)] λ c (y) δθ λ (y) x ν − ∂xλ ∂x ) δ4 (x − y) cλ (y) (θ) ν d4 xd4 y cν (x) (δλ d4 x (∂µ cν (x) − ∂ν cµ (x)) (∂ µ cν (x) − ∂ ν cµ (x)) d4 x Lghost . 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. they subtract out ﬁeld degrees of freedom. (12. (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 ) .209) Therefore. Let us start with the generating functional (J ) [A Z [Jµν ] = N DAµν eiS µν ] .210) Let us now try our naive counting of degrees of freedom of the theory.211) eﬀective ﬁeld degrees of freedom.

154)) (θ) ∆FP [Aµν ] x dθ σ (x)δ F ν [Aµλ (x)] − f ν (x) = 1. (12.209) 1 2 ). (12.536 12 Yang-Mills theory The gauge symmetry allows us to choose a gauge condition and we have chosen a Lorentz like gauge condition.215) Thus inserting this identity element into the functional integral. we introduce the identity element as (see (12.218) . (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)] . as we have seen (see (12.216) Sghost = d4 x (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) .152)).213) Therefore. (θ) ∆FP [Aµν ] = det = det ( (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. (12.217) In the naive application of the quantization procedure. F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). δF ν [Aµλ (x)] δθ σ (y) ν x δσ θ=0 ν − ∂x ∂xσ ) δ4 (x − y) . (12.214) where.

(9. (12. for example.219) That is.220) (12.4 Path integral quantization of tensor fields 537 However.222) We see that because of the transverse projection operator. 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) . the action for fµ in (12.221) has a gauge invariance (just like Maxwell’s theory) of the form δfµ (x) = ∂µ θ(x). implies that ∂ν ∂µ Aµν = ∂ν f ν (x) = 0.12. (12. (12. (12.221) where P (x − y) is the normalized transverse projection operator deﬁned earlier (see.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). (12.123)) as µν µ ν ∂x ∂x x P (x − y) = η µν − δ4 (x − y). we note that the gauge condition ∂µ Aµν (x) = f ν (x).224) so that we can write . 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.

remembering that (see (12.P ×∆FP [fµ ] e− 2ξ (F (fµ ). (12.f i µ) ∆FP [fµ ].f µ )− 2 (c. c) .226)) ∆FP [fµ ] = = we determine 1 1 S ghost = − (c.Fν [Aµλ ])− 2 (c. (12.227) Furthermore.225) where δF [fµ (x)] δθ(y) (θ) 1 2 ∆FP [fµ ] = det θ=0 = det xδ 4 (x−y) . Substituting this into (12.P × = i µν fν ) Df ∆FP [fµ ] δ F [fµ (x)] − f (x) e i i − 2ξ (f. (12.f ) Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .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) . (12.538 12 Yang-Mills theory ∆FP [fµ ] x (θ) dθ(x)δ F [fµ (x)] − f (x) = 1. (12. c) = − 2 2 d4 x c(x) c(x).F (fµ )) i µν fν ) = Df µδ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .226) Thus using the ’t Hooft trick we can write Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .228) det xδ 4 (x − y) 1 2 Dc(x) eiS ghost . c) i µλ ].

F µ[Aλσ ]) J +S GF +Sghost +S ghost ). (12.231) DAµν Dcµ Dcν Dc ei(S DAµν Dcµ Dcν Dc ei(S J +S ghost +S ghost i ) e− 2ξ (Fµ [Aλσ ]. . 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 .233) This is again not right and the agreement with canonical quantization seems to be worse than before.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. This is diﬀerent from the usual Faddeev-Popov ghosts in the sense that the Faddeev-Popov ghosts seem to come in pairs.4 Path integral quantization of tensor fields 539 Note here that the ﬁeld c(x) is a real anticommuting scalar ﬁeld.12. Thus using this modiﬁed ’t Hooft weighting factor.230) SGF = − = − Therefore. The ghosts of the present form are known as Nielsen ghosts. may be neglected. (12. (12. therefore. In ﬂat space-time this ghost Lagrangian density can be seen to give a total divergence and. (12. but in the presence of gravitation it cannot be written as a total divergence and is quite relevant. 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.

and δcµ (x) = ∂µ λ(x). ∆−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) (12.234) where λ(x) and λ(x) are anticommuting scalar parameters. µ (12. the ghost Lagrangian density also possesses a gauge invariance. (12. (12. under δcµ (x) = ∂µ λ(x).239) (λ) = det δF [cµ (x)] δλ(y) . (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. We can again use the FaddeevPopov trick and write ∆FP [cµ ] x (λ) dλ(x) δ F [cµ (x)] − f (x) = 1. (12.236) Let us choose for simplicity Lorentz like gauge conditions F and F = ∂µ cµ (x) =f (x). the ghost Lagrangian density Lghost is invariant.237) We are now ready to calculate the Faddeev-Popov determinants associated with these gauge ﬁxing conditions.238) = ∂µ cµ (x) = f (x). λ=0 . namely.235) and ∆FP [cµ ] x dλ(x) δ F [c(λ) (x)] −f (x) = 1.

the ’t Hooft trick needs to be carried out rather carefully. consequently. (12. Namely. the Faddeev-Popov term is an inverse determinant and. (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) . .242) The gauge conditions (12. (12.237) and (12. we cannot use weight factors of the forms i − 2χ e R e e d4 x f (x)f (x) . Therefore. namely.12. e i − 2χ R e e e e d4 x f (x) f (x) . f and f are Grassmann (fermionic) functions. the Jacobian for a change of variables is the inverse of the determinant. We note that contrary to the usual case. Rather. the appropriate weight factor in this case would correspond to . Similarly we can show that 4 −1 ∆FP [cµ ] = = det xδ (x − y) Dc Dc e ” “ e c c e −i e.243) since the exponents vanish (because of the fermionic nature of the variables).241) We note here that since we are writing an inverse determinant as an action. (12. the ghost ﬁelds c and c behave like commuting scalars. in the present case we can write 4 −1 ∆FP [cµ ] = = det xδ (x − y) c e) c Dc Dc e−i(e.238) correspond to fermionic conditions. e .240) where we have used the fact that for anticommuting variables.

c : 1 degrees of freedom. (12. (12. (12. c : 1 degrees of freedom. c : 1 degrees of freedom. 2χ c (12. commuting. c : −1 degrees of freedom.542 12 Yang-Mills theory e i − 2χ R e e e d4 x f (x) f (x) .247) We are now ready to count the number of eﬀective ﬁeld degrees of freedom in the theory. (12.244) Thus putting in these identity elements into the functional integral for the generating functional and using the ’t Hooft trick with the weight factor (12. c : 1 degrees of freedom. c.244) the generating functional takes the form Z [Jµν ] = N where Seﬀ and Leﬀ = DAµν Dcµ Dcν DcDc Dc D c D c eiSeﬀ .245) d4 x Leﬀ . anticommuting. Aµν : 6 degrees of freedom. cµ : −4 degrees of freedom.248) The ghosts c. The other . cµ : −4 degrees of freedom.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 and c being commuting scalars contribute positively to the counting of the number of degrees of freedom.

Veltman. C. 3. W. R. L. (12. Nuclear Physics B44.249) degree of freedom. CERN preprint (1973). 4. 11. Itzykson and J-B. New York (1965). Mohapatra. 189 (1972). L. N. Utiyama. 6. . Zuber. 378 (1971). G. 170 (1980). Physical Review D4. Hibbs and R. Mills. McGrawHill. 2. 5. Thus counting the degrees of freedom. L. G. Huang. Gribov. Physical Review 96. 1 (1970). Faddeev. D. N. A. Faddeev and V. 12.12. Popov. R. we see that eﬀectively the theory has 6 − 2 × 4 − 1 + 4 × 1 = 6 − 8 − 1 + 4 = 1. McGraw-Hill.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 . Yang and R. World Scientiﬁc. 191 (1954). This matches exactly with the counting of the degrees of freedom from the canonical quantization. Feynman. P. Veltman. Physics Letters 25B. Physical Review 101. New York. Diagrammar. 10. R. 9. D. K. V. Quantum Mechanics and Path Integrals. N. 12. ’t Hooft and M. 1597 (1956). ’t Hooft and M. 7. 1980. Physics Letters 93B. Quantum Field Theory. 29 (1967).5 References 1. Singapore (1982). N. Siegel. Leptons and Gauge Fields. 1 (1978). C. Nuclear Physics B139. 8. Theoretical and Mathematical Physics 1. Qualks.

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

67)) (Dµ c)a = ∂µ ca − gf abc Ab cc . The total Lagrangian density has been gauge ﬁxed and.2) is invariant under the global transformations 545 .2) Here ξ represents the arbitrary gauge ﬁxing parameter and the covariant derivative in the adjoint representation is deﬁned as (see (12.181) is given by 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ (13. therefore. develops a global fermionic symmetry which. with gauge ﬁxing and ghost terms. the total Lagrangian density.Chapter 13 BRST invariance and its consequences 13.1 BRST symmetry As we have seen in (12. the total Lagrangian density for the non-Abelian gauge theory (Yang-Mills theory) has the form LTOT = Linv + LGF + Lghost .186).3) where g denotes the coupling constant of the theory. in some sense. µ (13. (13.1) 2 + ∂ µ ca (Dµ c)a .34) of the original theory. does not have the gauge invariance (12. remembers the gauge invariance of the original theory. However. It is easy to check that the total Lagrangian density (13. which for the covariant gauge in (12.

Similarly. g (13.546 13 BRST invariance and its consequences δAa = µ δca δca ω (Dµ c)a . ca . ca independent of the parameters of transµ formations where δ1. This shows that under two successive transformations of the kind (13. .5) Here we have used the Jacobi identity for the symmetry algebra (see (12.4) we have δ2 δ1 φa = 0. 2 ω ∂ µ Aa .2 correspond to transformations with the parameters ω1. The invariance of the Lagrangian density (13.2 respectively.7) for the ﬁelds φa = Aa . = 6 (13. = − µ gξ (13.4) where ω is an arbitrary anti-commuting constant parameter of the global transformations. We note that the nilpotency of the transformations holds oﬀ-shell only for the ﬁelds Aa .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. g ω abc b c = f cc.50)).6) when the ghost equation of motion is used. ca . only when the ghost equation of motion is used). (13. 2 1 ω δ f abc cb cc = f abc δcb cc = f abc f bpq cp cq cc 2 2 ω abc bpq f f + f abp f bqc + f abq f bcp cp cq cc = 0. while for ca it is µ true only on-shell (namely. we obtain δ ∂ µ Aa = µ ω µ ∂ Dµ ca = 0.

5).2) can also be written equivalently as . Generally. ν gξ (13.13.4) with an anti-commuting constant parameter. Since this will also be quite useful for our latter discussions.9) so that the action is invariant. this is a reﬂection of the fact that the theory is lacking in some auxiliary ﬁeld variables and once the correct auxiliary ﬁelds are incorporated the symmetry algebra will close oﬀ-shell (without the use of equations of motion).2) is invariant under the global transformations (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 . In this derivation. δLinv = 0. namely. however. we note that the transformation for Aa can really be thought of as an inﬁnitesiµ mal gauge transformation (see (12.8) Consequently.2) under the transformations (13. we have used the fact that δ (Dµ c)a = 0 which we have seen in (13. (13. let us recall that the gauge ﬁxing Lagrangian density in (13. 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.4) can now be easily checked. therefore. As we have seen earlier within the context of the Abelian gauge theory in (12. This shows that the action for the total Lagrangian density (13. First. The present formulation of the BRST symmetry.1 BRST symmetry 547 The invariance of the Lagrangian density (13. is slightly unpleasant in the sense that the nilpotency of the anti-ghost ﬁeld transformation holds only on-shell. the invariant Lagrangian density is trivially invariant under these transformations.34)) with the parameter ǫa (x) = ωca (x) and.170) we can write the gauge ﬁxing Lagrangian density by introducing an auxiliary ﬁeld.

g (13.11) and when we eliminate F a from the Lagrangian density using this equation.4) take the form δAa = µ δca δca ω (Dµ c)a .) It is clear from the form of LGF in (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 .548 13 BRST invariance and its consequences LGF = ξ a a F F + (∂ µ F a )Aa .14) .12) µ 4 2 In this case. µ (13.13) δF a = 0. 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. namely.170) by a total divergence.10) that the equation of motion for the auxiliary ﬁeld takes the form ξF a = ∂ µ Aa . 2 ω = − F a. (13. (The form of the gauge ﬁxing Lagrangian density in (13. but it is this form that is very useful as we will see.10) diﬀers from that in (12. δ2 δ1 φa = 0. g ω abc b c = f cc.10) where F a is an auxiliary ﬁeld. (13. and it is straightforward to check that these transformations are nilpotent oﬀ-shell. the BRST transformations in (13. µ 2 (13. we recover the original gauge ﬁxing Lagrangian density (up to a total divergence term).

which deﬁne a residual global symmetry of the full theory.4). We note that Linv is invariant under the BRST transformation as we had argued earlier in (13. we will not pursue this symmetry further in our discussions.1 BRST symmetry 549 for all the ﬁeld variables φa = Aa . There is also a second set of fermionic transformations involving the anti-ghost ﬁelds of the form ω (Dµ c)a . g g (13. In some sense the BRST transformations. 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 .15) Unlike in the formulation of BRST variations without the auxiliary ﬁeld in (13.12) invariant.16) δAa = µ δca = δca = δF a = −ωf abc F b cc . ca . 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. However. g ω abc b c f cc. 2 (13. which can also be easily checked to leave the total Lagrangian density (13. replace the original gauge invariance of the theory and play a fundamental role in the study of non-Abelian gauge theories. ca .9)).13. Therefore. These are known as the anti-BRST transformations. F a represents µ the missing auxiliary ﬁeld that we had alluded to earlier. g ω F a + gf abc cb cc .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. since these do not lead to any new constraint on the structure of the theory beyond what the BRST invariance provides. F a .

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

whether they can reduce the vector space suﬃciently enough to coincide with the physical space).130) in the covariant gauge (12. “(+)”.18) where the superscript. On the other hand. . (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. (13.118). in reality it is an inﬁnite number of conditions since it has to hold for every value of the coordinates. Qc . In the Abelian theory. The corresponding operator in a non-Abelian theory.164) and hence the physical space so selected remains invariant under time evolution.18).) o Thus.19) are suﬃcient to reproduce even the GuptaBleuler condition in the case of the Abelian theory (namely. 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. stands for the positive frequency part of the ﬁeld. and the ghost scaling symmetry. It is.162)) ∂ µ A(+) (x)|phys = 0.19) Note that even in the case of Maxwell’s theory. not clear a priori if the conditions (13.2 Covariant quantization of Yang-Mills theory 551 have seen that the states in the physical space are selected as the ones satisfying the Gupta-Bleuler condition (see (9. the conditions in (13. the generators of the BRST symmetry.13. the Gupta-Bleuler condition works because ∂ µ Aµ satisﬁes the free Klein-Gordon equation (9. QBRST . 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. We recognize that even though this looks like one condition.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. are conserved and hence can be used to deﬁne a physical Hilbert space which would remain invariant under the time evolution of the system. therefore. as we have seen in (12.

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 . F a are inert µ under the scaling of ghost ﬁelds. 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 (13.21) 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 . From this. 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 .552 13 BRST invariance and its consequences To see that these conditions indeed lead to the Gupta-Bleuler condition in Maxwell’s theory. 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 . (13. .

26) implies that the physical states must have (net) zero ghost number. (13.13. Thus. In principle. (13.19) Qc |phys = 0. n .12). ← → d3 x c ∂ 0 c.23) following from (13. this allows for states containing an equal number of ghost and anti-ghost particles.27) . In particular. denoting the physical states of the theory as |phys = |Aµ ⊗ |n. 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) .22) to correspond to ← → d3 x F ∂ 0 c.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). (13. (13. we note that for the Abelian theory where f abc = 0 and there is no internal index “a”. (13.2 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 . the BRST and the ghost scaling charge operators can be obtained from (13.25) Let us note that the condition (13.

554 13 BRST invariance and its consequences where n. 0 = |Aµ .28) c(+) (k)|n. Thus. |phys = |Aµ ⊗ |0. This is precisely the Gupta-Bleuler condition in momentum space and this derivation shows how a single condition can give rise to an inﬁnite number of conditions (in this case. n denote the (equal) number of ghost and anti-ghost particles. we note from the form of the Lagrangian density (13. (13.31) where we have used the equation of motion for the auxiliary ﬁeld (see (13. n = 0. for every momentum mode k). 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).29) Namely. and F (+) (k)|Aµ = 0. To investigate systematically whether the physical state conditions in (13.30) (13. 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. we feel conﬁdent that the physical state conditions in (13.11)).12) that we can obtain the canonical momenta conjugate to various ﬁeld variables of the theory to correspond to . n = 0. and must further satisfy (with k0 = |k|) F (+) (k)|phys = 0 = kµ Aµ(+) (k) |phys .19) are the right ones even for the non-Abelian theory.19) really select out the subspace of physical states. then this implies that (13. (13.

t) i F a (x.2 Covariant quantization of Yang-Mills theory 555 Πia = Πa = Πa = c Πa = c ∂LTOT = −F 0ia . (13. = iδab δ3 (x − y). t).35) . The equal-time 0 canonical (anti) commutation relations for the theory can now be written as ( = 1) Aa (x. 0 ˙ ∂F a ∂LTOT ˙ = −ca . t) ca (x. = iδab δ3 (x − y). the BRST parameter ω is seen to be anti-Hermitian ω † = −ω.33) + The hermiticity conditions for the ghost ﬁelds which arise out of various consistency conditions (for example.34) With this choice.32) Here we have used left derivatives for the anti-commuting ghost ﬁelds. Πb (y. In particular. ˙ ∂ ca (13. Πjb (y. (ca )† = −ca .33) have to satisfy the proper hermiticity properties etc. ∂ ca ˙ ∂LTOT = (D0 c)a . the (anti) commutation relations (13.13. Πb (y. (13. t). t). Aa plays the role of the momentum conjugate to F a . (13. ˙ ∂ Aa i ∂LTOT = Aa .) are given by (ca )† = ca . t) c + j = iδab δi δ3 (x − y). we see that in this formulation with the auxiliary ﬁeld. t) c ca (x. = iδab δ3 (x − y). Πb (y. the Lagrangian density and the conserved charges have to be Hermitian. t).

the conserved charges (13. The third is a statement of the fact that the BRST charge carries a ghost number of unity.21) and. see (13.14)). Therefore. for example. Qc ] = 0.17).33) [QBRST . Qc ] = ica (x). that (this basically describes the behavior of the ghost ﬁelds under a scaling (13. An immediate consequence of the algebra in (13. The ﬁrst equation in (13.38) simply reiterates in an operator language the fact that the BRST transformations are nilpotent (see (13.33) for the ﬁeld variables. therefore. (13. The second equation implies that the ghost scaling symmetry is Abelian.22)) g d3 x Πia (Di c)a − F a Πa + f abc Πa cb cc . c QBRST = − Qc = − (13. The algebra of the conserved charges also follows in a straightforward manner from (13. [Qc .22) are Hermitian with the assigned hermiticity conditions (13. We note.37) [ca (x). Qc ] = −iQBRST .22) can now be expressed in terms of the ﬁelds and the conjugate momenta as (before integrating the ﬁrst term by parts in QBRST . [QBRST .17)) [ca (x). c c 2 a d3 x (Πa ca + ca Πc ) .34) for the ghost ﬁelds. Qc ] = −ica (x). The conserved charges (13.556 13 BRST invariance and its consequences and we note that the currents (13. (13.13) and (13.38) This algebra can also be derived directly from the transformation laws for the ﬁeld variables in (13.38) is that .36) and (13.36) and we can calculate the algebra of charges as well as various other relations of interest using the (anti) commutation relations (13. we see that we can think of iQc as the ghost number operator (with the ghost number for ca being negative). QBRST ]+ = 2Q2 BRST = 0.

39) which is quite useful (particularly) in dealing with gauge theories at ﬁnite temperature. [QBRST . for every |Ψ ∈ Vphys . H] = 0.40) 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. First.13. must represent a truly gauge invariant ﬁeld variable. we can select out a subspace Vphys by requiring that states in this space are annihilated by QBRST . therefore. (13. it is truly a BRST singlet (invariant) state. (we assume that the vacuum state is a BRST singlet (invariant) state) . As we have discussed earlier. 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.2 Covariant quantization of Yang-Mills theory 557 QBRST eπQc = eπ(Qc −i) QBRST = −eπQc QBRST . or. the total vector space V of the complete gauge theory (13. (13. QBRST . we have QBRST |Ψ = 0.42) ˜ for some |Ψ and still satisﬁes the physical state condition. eπQc + = 0. namely. then. (13. The ﬁeld operator which creates such a state must necessarily commute with QBRST and.12) contains various unphysical states as well as states with negative norm in addition to the physical states. (13.40). As is clear from our earlier discussions. if a state |Ψ cannot be written as ˜ |Ψ = QBRST |Ψ . Namely. Consequently.41) We note that there are two possible kinds of states which will satisfy the physical state condition (13.

therefore.13). (Note that the auxiliary ﬁeld does not transform under a BRST transformation and. This implies that (13. as we have seen in (13.558 13 BRST invariance and its consequences QBRST |Ψ = QBRST Ψ|0 = [QBRST . correspond to truly physical states of the theory and since gauge invariant degrees of freedom have physical (non-negative) commutation relations. (13.43) where we have assumed that the operator Ψ creates the state |Ψ from vacuum. (13. However. has a diﬀerent character.40) can be written in the form ˜ |Ψ = QBRST |Ψ . Ψ] |0 = 0.47) Such a state would also be orthogonal to any physical state (either of the ﬁrst or the second kind) satisfying the physical state condition .46) We see that the physical state condition is trivially satisﬁed in this case because of the nilpotency of QBRST (see (13.34) for the ghost ﬁelds. (13. such states would have positive norm.38)). it can be written as the BRST variation of the anti-ghost ﬁeld and. therefore.44) [QBRST .45) and transverse ﬁelds would represent such operators which are gauge invariant asymptotically. therefore. Ψ] = 0. (13. namely.) ˜ ˜ Ψ|Ψ = Ψ|QBRST QBRST |Ψ = 0. also satisﬁes the above relation. Such states would. Ψ|0 = |Ψ .) The second class of states which would satisfy the physical state condition (13. 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.

be identiﬁed as belonging to the V quotient space V phys = phys and will have positive deﬁnite norm. then. (13. deﬁning P (n) as the projection operator onto the n-unphysical particle sector. of course. (13. Here.48) which follows from the physical state condition. ca ]+ .13) through (anti) commutation with the appropriate ﬁeld variables) F a ∼ [QBRST . (13. Thus. then ˜ Ψ′ |Ψ = Ψ′ |QBRST |Ψ = 0. we note that we can write P (0) = ½ − P ′ . If we denote all the states of Vphys of the second kind by V0 .49) P (n) P (m) = δnm P (n) . which satisfy the physical state condition and are of the ﬁrst kind can. by unphysical. (13. V0 Every state in Vphys can.40) because if |Ψ represents a physical state (of either kind) while |Ψ′ is a state of the second kind. this would contain all the zero norm states which would be orthogonal to Vphys itself.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. we have ∞ n=0 P (n) = ½.2 Covariant quantization of Yang-Mills theory 559 (13.51) .49). therefore.13.49) that P (0) projects onto the truly physical states of the theory (namely. the longitudinal components of Aa and the auxiliary ﬁelds. with no unphysical particles) and from the completeness of the projection operators in (13.50) It is clear from the deﬁnition in (13. The true physical states. µ We recall from (13. we mean states containing quanta of the ghost ﬁelds. be decomposed into states containing ﬁxed numbers of unphysical particles.

that P ′ must also commute with QBRST . (13. of course. P ′ = 0. so that QBRST .55) which follows from the nilpotency of QBRST . Namely. we can write P (n) = QBRST . (13. for n ≥ 1.53) However. cannot be truly gauge invariant. R]+ . Consequently. n ≥ 1. P (0) QBRST . R]+ = 0. and. therefore. whose explicit form is not important for our discussions) P ′ = [QBRST . = 0. It follows. then.56) + P′ It is now easy to show that P (n) . (13. it must have the form (for some fermionic operator R. R(n) .52) We can. therefore. What is interesting. however. construct the actual forms of all the projection operators explicitly. the diﬀerence between the two lies in the fact that P ′ must necessarily involve unphysical ﬁelds since it projects onto unphysical states and. is the fact that P (0) projects onto the true physical states and. we have QBRST . (13.54) (13. P ′ = QBRST . therefore. we can even show that for every P (n) . project onto the zero norm space V0 when acting on states |Ψ . but this is not very illuminating.560 where ∞ n=1 13 BRST invariance and its consequences P = ′ P (n) . must be gauge invariant and will commute with QBRST . In fact. [QBRST .

R]+ |Ψ = 0. It leads to a formal proof of unitarity of the theory when restricted to the subspace of the physical Hilbert space. (13. For example. the total Lagrangian .19).3 Unitarity 561 which satisfy the physical state condition (13. This also makes it clear that the V norm of a state |Ψ ∈ V phys = phys would be positive deﬁnite. deﬁned by the physical state condition. let |Ψ .3 Unitarity The BRST invariance of a gauge theory is quite important from yet another consideration. then. with the hermiticity assignments for the ghost ﬁelds in (13. This V0 would correspond to the true physical subspace of the total vector space.40). |Ψ′ ∈ Vphys . which follows from the physical state condition (13. therefore.58) (13. We will only outline the proof of unitarity in this section. 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. Any vector |Ψ ∈ Vphys can now be written as |Ψ = P (0) |Ψ + P ′ |Ψ .59) This. 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.40) (as well as the hermiticity of QBRST ). 13. Ψ′ |P (n) |Ψ Ψ′ |P ′ |Ψ = = Ψ′ | QBRST . R(n) |Ψ = 0. (13. then.57) + Ψ′ | [QBRST . shows that Vphys .13. and the norm of any such state is. The question of the unitarity of the S-matrix in a gauge theory can be formulated in the following manner. We note that.34). obtained to be Ψ|Ψ = Ψ|P (0) |Ψ ≥ 0.

we can deﬁne a unique state |Ψ ∈ V phys as |Ψ = P (0) |Ψ . = = Ψ|AP (0) B|Φ Ψ|A(½ − P ′ )B|Φ = Ψ|AB|Φ . |Ψ and |Ψ . namely.62) which follow from (13. |Φ ∈ Vphys and any two operators A and B acting on Vphys (i. the S-matrix is BRST invariant since the full theory is.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.12) is Hermitian so that the S-matrix of the theory is formally unitary. S † S = SS † = ½. P (0) acts as the identity operator. This question can be systematically analyzed as follows. 2 (13.562 13 BRST invariance and its consequences density (13. (13.63) where we note that in the space V phys . (13. (13. P (0) |Ψ = P (0) |Ψ = P (0) |Ψ = |Ψ . namely.57). S] = 0.e. 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 . so that [QBRST . we note that given any state |Ψ ∈ Vphys . (13. any two operators which do not take us out of Vphys ) Ψ|P (0) |Φ Ψ|(½ − P ′ )|Φ = Ψ|Φ .60) Furthermore.64) The two states. Furthermore..

It follows from (13. Given this.60). (13.68) 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. 2 (13.67) It is clear that since S is BRST invariant. (13.66) and (13.65) where the last line follows from the ﬁrst of the conditions in (13. the physical state condition (13. 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) .62). (13.70) .40) which naturally follows from the BRST invariance of the theory. In a similar manner. (13. 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 .68) where we have used (13.66) such that for |Ψ ∈ Vphys .63). Furthermore. |Ψ = P (0) |Ψ ∈ V phys as deﬁned in (13.69) In other words.13.62) as well as the formal unitarity of S (13. it will leave the space Vphys invariant and hence Sphys will not take a state out of V phys .3 Unitarity 563 Ψ|Φ = Ψ| P (0) |Φ = Ψ|P (0) |Φ = Ψ|Φ . we can also show that † Sphys Sphys = ½. Sphys |Ψ = Sphys P (0) |Ψ = P (0) S|Ψ = S|Ψ . (13.

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. ξ a a c F + ∂ µ ca Aa µ 2 + |phys′ (13. namely. 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.72) This shows that the terms added to modify the original Lagrangian density have no contribution to the physical matrix elements of the theory.40) that phys| (LGF + Lghost ) |phys′ = −g phys| QBRST . = g QBRST . It now follows from the physical state condition (13. − ca F a − ∂ µ ca Aa µ 2 + (13. To show this. 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) . the gauge ﬁxing and the ghost Lagrangian densities) can. be written as a BRST variation (with the parameter of transformation taken out).564 13 BRST invariance and its consequences As another consequence of the BRST invariance of the theory. we note that these extra terms in the Lagrangian density (namely. 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 . = 0. we can show that the gauge ﬁxing and the ghost Lagrangian densities lead to no physical consequences. in fact.

13. before we go into the actual derivation.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 . We have already given a simple diagrammatic derivation of such identities in the case of QED in section 9. However. 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. as we have seen in (12.75) .4 Slavnov-Taylor identity The BRST invariance of the full theory leads to many relations between various scattering amplitudes of the theory.74) where S[φ] denotes the dynamical action of the self-interacting scalar ﬁeld. However.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. ca F a ]+ |0 J = 0. In this case. the vacuum functional is given by 0|0 J = Z[J] = eiW [J] = N Dφ eiS J [φ] . 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. (13. For simplicity. (13.120).7. let us recapitulate brieﬂy some of the essential concepts from path integrals (a detailed discussion is beyond the scope of these lectures). 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). (13.13.

In particular we note that 0|φ(x)|0 J δn W [J] δJ(x1 ) · · · δJ(xn ) J=0 .76) and (13.77) involve only Feynman diagrams where all the parts of the diagram are connected. (13. the n-point Green’s functions would involve Feynman diagrams which consist of parts that are not connected and Z[J] (through (13. let us note here that each term in the calculation of the Green’s function in (13. 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.79) . the Green’s functions calculated from W [J] (the logarithm of Z[J]) through (13. In general. 0|T φ(x1 ) · · · φ(xn ) |0 = δn Z[J] (−i)n Z[J] δJ(x1 ) · · · δJ(xn ) . This is more fundamental since the general Green’s functions can be constructed in terms of these. For example. To clarify this distinction further. On the other hand.76) J=0 deﬁnes the n-point time ordered Green’s function of the theory.78) is known as the classical ﬁeld (see (7.77) in the path integral formalism corresponds to a unique Feynman diagram in perturbation theory.49)) and is a functional of J. Z[J] and W [J] are known as the generating functionals for the Green’s functions of the theory. On the other hand. (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 .76)) includes contributions of all diagrams (including the ones that are not connected) to a Green’s function. δJ(x) Z[J] δJ(x) (13. (13. The connected two point time ordered Green’s function which also corresponds to the Feynman propagator of the theory is similarly obtained from (13.566 13 BRST invariance and its consequences As we have already mentioned in the last chapter.77) = φc (x) = (−i) δZ[J] δW [J] = .

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. 2 (13. 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). 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. (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 . These are quite fundamental in the study of quantum ﬁeld theory and the generating functional which generates such vertex functions is constructed as follows.81) = −J(y).4 Slavnov-Taylor identity 567 The connected n-point time ordered Green’s functions consist of connected Feynman diagrams with external lines (propagators). However. let us consider the eﬀective Lagrangian density (at the lowest order or tree level) which consists of LTOT for the Yang-Mills theory (13.82) .80) It can now be checked using (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. the more fundamental concept in perturbation theory is the diagram without the external propagators known as the vertex function.

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

85) g 2 We note that the generating functional is deﬁned by integrating over all possible ﬁeld conﬁgurations.87) can also be written as δW δW δW = 0.88) 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. it should not change under any ﬁeld redeﬁnition).87) +iη a (x) The functional integration measure can be easily checked to be invariant under such a fermionic transformation and using (13. δJ a (x) (13.13. since it does not depend on the ﬁeld variables.85) to (see also (13. (13.4 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 . − iη a (x) + iη a (x) a µa (x) a (x) δK δK δJ (x) (13.86) the generating functional should be invariant (namely. This immediately leads from (13. It is here that the .75) we note that (13. if we redeﬁne the ﬁelds under the path integral as. (13. A → A + δBRST A. Therefore.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 .

a δKµ δW . Most often. µ (13. we are interested in the identities satisﬁed by the proper (1PI) vertices of the theory. δK a (13.88) in terms of the generating functional for the proper vertices as . = −J a .89) where K stands generically for the sources for the composite BRST variations. K] − d4 x J µa Aa + J a F a + i (η a ca − ca η a ) . we have Γ[A.80) and the ﬁeld variables are really the classical ﬁelds and we are dropping the superscript (c) for simplicity). K] = W [J. however.90) Using these deﬁnitions.570 13 BRST invariance and its consequences usefulness of the sources for the composite BRST variations becomes evident. = iη a . we see that we can rewrite the master equation (13. it is clear that δΓ δAa µ δΓ δF a δΓ δca δΓ δca δΓ a δKµ δΓ δK a = −J µa .81)). 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. = iη a . From the deﬁnition of the generating functional for the proper vertices (see (13. = = δW .

(We would discuss renormalization of ﬁeld theories in a later chapter.13. For example. let us note that we can write the master identity (13. consequently. 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. Thus. (13. δ2 δF b (p)δcc (−p) d4 k (13. follow from the BRST invariance of the theory.4 Slavnov-Taylor identity 571 d4 x δΓ δΓ δΓ δΓ δΓ + a − F a (x) a = 0. The BRST invariance. a (x) δK µa (x) a (x) δAµ δc (x) δK δc (x) (13. like the Slavnov-Taylor identities. 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 = 0.93) − 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. This is essential in proving the renormalizability of gauge theories. is very fundamental in the study of gauge theories as far as renormalizability and gauge independence of physical observables are concerned. in this way.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) = 0.91) 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. for example.) .

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. − 1− (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. 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. (They are the Green’s functions of the free theory. ξ M ab = −δab .2) 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ 2 + ∂ µ ca Dµ ca . 2 µ where we have neglected total divergence terms and have identiﬁed Oµν ab = δab η µν 1 µ ν ∂ ∂ . These are the essential elements in carrying out any perturbative calculation in a quantum ﬁeld theory.95) LQ = − = = 1 a µν A η 2 µ 1 a µν ab b A O Aν + ca M ab cb . We note that since gauge ﬁxing breaks gauge invariance. Let us recall that the total Lagrangian density after gauge ﬁxing (in the covariant gauge) has the form (13.96) take the forms .572 13.) The inverses of the two point functions are easily calculated in the momentum space where the quadratic operators in (13.96) 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).

Oµν ab (p) = −δab η µν p2 − 1 − 1 µ ν p p . λ ξ (13. λ ξ ξ .5 Feynman rules 573 M ab (p) = δab p2 . let us derive this inverse systematically. λ ξ ξ (13. ηµν ).13.99) for the inverse and carrying out the multiplication explicitly we have η µν p2 − 1 − or. η µν p2 − 1 − 1 µ ν −1ac p p O νλ (p) = −δac δµ . λ or. we note that the inverse is deﬁned to satisfy (repeated indices are summed) −1bc Oµν ab (p)Oνλ (p) = δac δµ . p2 (13.100) Substituting the parameterization (13. β are arbitrary parameters to be determined (they are not necessarily constants and can be Lorentz invariant functions of the momentum).101) 1 β µ (αp2 + 1)δµ − α(1 − ) − p pλ = 0. With this parameterization. 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 .99) where α. or.97) The inverse of M ab (p) in (13.97) is quite simple (it is like the propagator of a massless scalar ﬁeld) M −1ab (p) = δab . p2 (13. we can parameterize the most general form of the inverse as O−1ab (p) = δab α ηµν + β µν pµ pν .98) while the derivation of the inverse of Oµν ab (p) is a bit more involved. λ αp2 δµ − α 1 − λ 1 1 µ p pλ + β pµ pλ − β 1 − pµ pλ = −δµ . ξ (13. Therefore. 1 µ ν p p ξ αηνλ + β pν pλ p2 = −δµ .

79).102) α=− α 1− or.104) where ( . . 2 p p −1ab Oµν (p) = − (13.Oµν (p)J νb (p) +i η a (−p). 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. β = α(ξ − 1) = − so that we can write the inverse in (13.105) 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.99) as pµ pν δab ηµν + (ξ − 1) 2 .126)) h −1ab i − 2 J µa (−p). ξ ξ 1 (ξ − 1). p2 (13. (13.M −1ab (p)ηb (p) Z0 = N e i . see (12. 2 p p (13.574 13 BRST invariance and its consequences Setting the coeﬃcients of each distinct Lorentz structure to zero.103) With these. 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 . ) represents the integral over p and we deduce from this that (see (13. it follows that 1 . p2 1 β − = 0.

109) where we have used J generically for all the sources and the interaction Lagrangian density can now be identiﬁed with Lint = LTOT − LQ = gf abc ∂µ Aa Aµb Aνc − ν g2 abp cdp a b µc νd f f Aµ Aν A A 4 (13. the gauge is known as the Feynman gauge where the propagator has a much simpler form which is more suitable for perturbative calculations.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]. i. (13. when we are in the Landau gauge this propagator is transverse. 2 p p iδab . p2 (13. b pµ pν iδab ηµν + (ξ − 1) 2 .e. 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 .13. while for ξ = 1. p2 (13. µ . a p ν.106) which deﬁnes the ghost propagator (which is sometimes also denoted by Dab (p)). The propagators can be diagrammatically represented as = iGab (p) µν = − a p b (13.5 Feynman rules 575 When ξ = 0.107) µ..110) −gf abc Aa ∂ µ cc cb .

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). p3 .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 . V µνλρ abcd (p1 . 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. 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 .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 . p2 .

a p3 p1 µ.113) V µ abc (p1 . p3 . b c = V µ abc (p1 . with (13. p3 ).115) . a p2 b p2 ν. p3 ). p3 ) = i δ3 S δAa (p1 )δcb (p2 )δcc (p3 ) µ = i δ3 Sint δAa (p1 )δcb (p2 )δcc (p3 ) µ (13.13. p4 p3 p1 µ. (13. p2 .114) = −(2π)4 gf abc pµ δ4 (p1 + p2 + p3 ). (13. 3 Therefore. p2 .114) we can represent all the interaction vertices of the theory as λ. p2 . p4 ). c = V µνλρ abcd (p1 . p2 . b = V µνλ abc (p1 .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 . d λ.112)-(13. c p3 p1 µ. a ρ. p2 ν.

These represent all the Feynman rules for the Yang-Mills theory and perturbative calculations can now be carried out using these Feynman rules. We will now discuss. 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. If possible. from two distinct points of view.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. A class of gauge conditions of the form n · Aa (x) = f a (x).160)) that the total Lagrangian density for a non-Abelian gauge theory takes the form 1 1 a (n · Aa )2 − ca n · Dca . The dot over the ghost line denotes where the derivative acts in the interaction term in the Lagrangian density (namely.114)). n2 = 0 n2 = 0. simplify the perturbative calculations enormously primarily because the number of diagrams to evaluate will be much smaller (namely. the cc ﬁeld.116) where nµ is an arbitrary vector does indeed achieve this.6 Ghost free gauges In quantizing a non-Abelian gauge theory. the diagrams with ghosts will be absent). 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. In a general axial-like gauge (13. such a gauge is known as the light-cone gauge. in what sense these gauges become ghost free in a non-Abelian gauge theory.117) . or (13.116). we recall (see (12. of course. Such a class of gauge choices is conventionally known as ghost free gauges. Let us ﬁrst discuss this from the diagrammatic point of view. LTOT = − Fµν F µνa − 4 2ξ (13.153) and (12. When n2 = 0. 13. see for example (13. this would. for n2 = 1 (normalized) it is called the temporal gauge while for n2 = −1 it is known as the axial gauge.

the ghost propagator in the axial-like gauge (13. either open lines or closed loops. (n · p) nµ Gab = −ξδab µν nµ nν Gab = −ξδab . 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. µν (13.1: Examples of diagrams with open and closed ghost lines.116) in this theory has the form δab .120) and the antighost-ghost-gauge vertex involves a multiplicative factor of nµ . . n·p iGab = (13.13.119) Similarly.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 . the integrand of any diagram involving ghosts. (n · p)2 (13. Therefore.118) so that we have (independent of whether n2 = 0 or n2 = 0) pν .

Using the fact that the integral is Lorentz invariant. (k2 )(m+1)/2 (13. we can rewrite this also as 1 m dxi 0 i=0 δ(1 − x0 − x1 − · · · − xm ) (n2 )(m+1)/2 dD k = 0. 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.122) and the vanishing of this integral (in the last step) is easily seen using dimensional regularization (that we will discuss in chapter 15. An alternative way to see this is to note that the generating functional in the axial-like gauge (13. . (n · k)m+1 δ(1 − x0 − x1 − · · · − xm ) m+1 = 0 i=0 dxi (13. (13.580 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 . (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 ) . namely.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 . 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.10)).

J. Note that such a decoupling works only because nµ (in n · Aa inside the determinant) is a multiplicative operator.126) This shows that in this gauge the ghosts decouple completely and can be absorbed into the normalization factor. C. the covariant gauge. (13.7 References 1.7 References 581 G[f a ] = e i − 2ξ R d4 x f a f a . therefore. 99 (1972). 436 (1971). It would not work in say. The axial-like gauge choices (13. (13. Nuclear Physics B33. Theoretical and Mathematical Physics 10. (13. A.124) However. correspond to ghost free gauges very much like in the Abelian gauge theories. Rouet and R. let us choose G[f a ] = det n · ∂δab + gf abc f c −1 − i R d4 x f a f a 2ξ e . since the generating functional does not depend on the functional form of G [f a ] (up to irrelevant multiplicative constants). 2.125) In this case. 13.116) in a non-Abelian gauge theory. 344 (1974).13. 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 . Physics Letters 52B. Slavnov. . Stora. A. A. Becchi. Taylor. 3. C.

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

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

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. (14.2) describes a massive photon which can be seen as follows. for example.6) together with the constraint (14. (14. The Euler-Lagrange equation following from (14. + M 2 Aν = 0. −∂µ F µν − M 2 Aν = 0. or. ∂µ F µν + M 2 Aν = 0. or. (14.2) denotes the Proca Lagrangian density) and it is clear that this system of equations deﬁnes a massive spin 1 ﬁeld theory (a massive photon). ∂∂µ Aν ∂Aν ∂µ or.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 .3) The Lagrangian density (14. we also note here that the theory (14.5) deﬁnes the Proca equation (and (14. The propagator for this theory can be worked out (see.4) yields ∂µ (∂ µ Aν − ∂ ν Aµ ) + M 2 Aν = 0.6) Equation (14.2) does not have a gauge invariance (because of the mass term) unlike the Maxwell theory. we obtain ∂ · A = 0.4) with ∂ν and using the anti-symmetry of the ﬁeld strength tensor. As has been mentioned earlier. section 13.2) has the form ∂L ∂L − = 0.5) and substituting this back into the Euler-Lagrange equation in (14.4) Contracting (14.5) in a simple manner and in momentum space has the form . (14. Note that the ﬁeld variable Aµ has four ﬁeld degrees of freedom while the constraint (14.

7) M2 1 1 Aµ Aµ + ∂µ χ∂ µ χ + M Aµ ∂µ χ. the St¨ckelberg theory u (14.1 Stuckelberg formalism 585 iGµν (p) = − p2 Thus. 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 − . it is suﬃcient to note that under the transformation (14.¨ 14. also contains a dynamical charge neutral spin zero ﬁeld (real scalar ﬁeld) which mixes with Aµ . To see the invariance of the full theory. The ﬁeld strength tensor is. e M θ(x). In contrast to the Proca theory (14. invariant under the gauge transformation (14.9) where e denotes the coupling constant of QED (this can be set to unity. of course.8) = − Fµν F µν + 4 2 2 which. 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. 2 −M M2 (14.8) was constructed to be invariant under a Maxwell-like gauge transformation Aµ (x) → Aµ (x) + 1 ∂µ θ(x). This leads to diﬃculties in establishing renormalizability of the (interacting) Proca theory for massive photons. but we have kept it for consistency with earlier discussion. As a result of these diﬃculties.9) Aµ + M 1 1 1 ∂µ χ − ∂µ χ → Aµ + ∂µ θ + ∂µ θ M e M e 1 = Aµ + ∂µ χ. (14. say in (9. The limit M → 0 of the Proca theory is clearly quite subtle. in addition to the spin 1 ﬁeld.10) .83) as well as for discussions in connection with the Higgs phenomenon in the next section). χ(x) → χ(x) − e (14. M (14.2).9).

the massless limit M → 0 is now straightforward.12) then the Lagrangian density (14.13) We recognize this to be the Proca theory (14. the there is no diﬃculty in establishing renormalizability in such a gauge.7)) i pµ pν ηµν − . This is typical of the behavior of theories in a unitary gauge.11) which can be achieved by the choice of the gauge transformation parameter e χ(x). . the gauge propu agator (see also (14.2) and this makes it clear that the Proca theory can be thought of as the gauge ﬁxed St¨ckelberg theory.586 14 Higgs phenomenon and the standard model Because of the gauge invariance of the theory. (14.8) takes the form M2 1 Aµ Aµ . for example. Furthermore. if we choose a covariant gauge condition which leads to a gauge ﬁxing Lagrangian density of the form (see. but would lead to propagators that are well behaved at high momentum. 2ξ − (14.15) this would induce a ghost action. As a result. M θ(x) = (14. we can choose a gauge and if we choose the unitary gauge χ(x) = 0. 2 −M M2 iGµν (p) = − p2 (14.14) has the unpleasant features described earlier. in this unitary gauge. L = − Fµν F µν + 4 2 (14. However. section 12.3 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 . On the other hand.

To see that the complete Lagrangian density is also invariant. SU (n) (as we have discussed earlier in section 12.1). say.¨ 14. the Lagrangian density for the gauge ﬁeld is invariant under the transformation (14.16) ig ig where the trace is over the fundamental representation of the group. (12.17) where χ denotes a matrix valued scalar ﬁeld. This Lagrangian density can be easily checked to be invariant under the gauge transformations (see. Aν ] .38) and (12. Here the ﬁeld strength tensor and the operator V are deﬁned as Fµν V = ∂µ Aν − ∂ν Aµ + ig [Aµ . ig (14.18).41)) Aµ → U Aµ U −1 − V → U V. ig (14. We will now brieﬂy describe the St¨ckelberg formalism for non-Abelian gauge u theories. it is suﬃcient to note that . for example. 1 (∂µ U )U −1 . = e M χ. (14.18) As we have already seen in chapter 12.1 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. 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 .

as before. 2 (14.2) to the non-Abelian case and describes a massive spin 1 ﬁeld belonging to SU (n).19) Using the cyclicity under trace it is now straightforward to see that the full theory is invariant under the gauge transformation in (14. 2 − M2 p M2 iGab (p) = − µν (14.588 14 Higgs phenomenon and the standard model Aµ + 1 (∂µ V )V −1 ig 1 1 (∂µ U )U −1 + (∂µ U V )(U V )−1 ig ig 1 (∂µ U )U −1 ig → U Aµ U −1 − = U Aµ U −1 − + = U Aµ + 1 (∂µ U )U −1 + U (∂µ V )V −1 U −1 ig 1 (∂µ V )V −1 U −1 . 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 − .21) (14. (14.22) This can be thought of as a generalization of the Proca theory (14.23) . In this gauge.20) which can be achieved by the choice of the gauge transformation matrix U (x) = V −1 (x).18).16) becomes 1 L = − Tr Fµν F µν + M 2 Tr Aµ Aµ . then the Lagrangian density (14. ig (14. If we choose the unitary gauge condition V = ½.

84) and (7. This is known as the Nambu-Goldstone phenomena and the massless scalar ﬁeld is known as the Nambu-Goldstone boson. 14. but now interacting with an Abelian gauge ﬁeld as well (scalar QED). For simplicity we consider again the self-interacting theory of a complex scalar ﬁeld (as in section 7.24) 4 4 where the theory has a “wrong” sign for the mass term and the covariant derivative is deﬁned as in (7.83) Dµ φ = ∂µ φ + ieAµ φ. (14. The Lagrangian density for such a theory is given by (with λ > 0.26) . (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). On the other hand.2 Higgs phenomenon In section 7. Let us next ask what happens if there is a spontaneous breakdown of a local symmetry. the propagator will have the correct asymptotic behavior for large values of the momenta.85)) to be invariant under the inﬁnitesimal local transformations δφ† = iǫ(x)φ† (x). if we choose a covariant gauge ﬁxing condition (a generalization of (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. δAµ = 1 ∂µ ǫ(x).6) λ 1 2 L = − Fµν F µν + (Dµ φ)† (D µ φ) + m2 φ† φ − φ† φ .14.2 Higgs phenomenon 589 with the unpleasant features discussed earlier that are characteristics of the unitary gauge choice.25) The Lagrangian density (14.5). see also section 7.24) can be checked easily (see (7. e δφ = −iǫ(x)φ(x).15)). (14. In that case.

590 14 Higgs phenomenon and the standard model The gauge boson in this theory is massless because the gauge symmetry (14.42)) 1 φ = √ (σ + iχ).5.24) in terms of the σ. (14. this actually corresponds to the local maximum of the potential.30) 2 16 Let us now rewrite the theory around the true vacuum by shifting the ﬁeld variables as (see (7.28) (14. On the other hand. (14.29) For further analysis. the normal vacuum for which φ = φ† = 0. since the mass term for the scalar ﬁeld has the “wrong” sign. let us rewrite the Lagrangian density (14. as we have discussed in section 7. (As we have seen in section 7.27) does not correspond to the true vacuum. this is just one of them) 2m σc = σ = √ .77)) . 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 . (14. there is actually an inﬁnity of minima of the potential.64).5.) Rather the true vacuum of the theory satisﬁes (see (7. λ where (see also (7.26) does not allow a mass term for the gauge ﬁeld. 2 χc = χ = 0.

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 . Let us look at only the terms in (14. λ In this case.33) where we have used the form of v in (14.30) 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. (14.14.31) where we have identiﬁed the vacuum expectation value (vev) of the ﬁeld with 2m σ =v= √ .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. but it now mixes with the gauge ﬁeld Aµ so that we have .32) to simplify some of the terms.79).2 Higgs phenomenon 591 2m σ → σ + σ = σ + √ = σ + v.34) 2 2 We note here that the χ ﬁeld appears to be massless in this theory as in (7. the Lagrangian density (14. (14. 16 (14.

11)) 4e2 m2 . ignoring the σ ﬁeld should be compared with the St¨ckelberg Lagrangian density u (14.34). λ m2 = M 2 = e2 v 2 = B (14.592 14 Higgs phenomenon and the standard model to diagonalize the theory in order to determine the true spectrum of the theory.35) ∂µ Bν − ∂ν Bµ = ∂µ Aν + 1 1 ∂ν χ − ∂ν Aµ + ∂µ χ ev ev = ∂µ Aν − ∂ν Aµ . 1 1 ∂µ χ Aµ + ∂ µ χ ev ev 2 1 = Aµ Aµ + Aµ ∂µ χ + 2 2 ∂µ χ∂ µ χ.) To achieve this.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 . it describes a massive scalar ﬁeld σ with mass m2 = 2m2 .79).38) and a massive gauge ﬁeld Bµ with mass (this is the analog of the unitary gauge in (14. let us deﬁne 1 ∂µ χ. ev Bµ = Aµ + then we have (14. As in (7.35) the quadratic Lagrangian density (14. (14.39) . (The Lagrangian density in (14. ev e v Aµ + Bµ B µ = (14.37) that the quadratic part of the shifted Lagrangian density is completely diagonalized. σ (14.37) 2 We see from (14.36) Thus in terms of this new ﬁeld variable (14.8).

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

as pointed out in the last section.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).84)) δσ(x) = ǫ(x)χ(x). that is more useful for this purpose let us proceed as follows. The presence of the local symmetry (14.41) 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. δχ(x) = −ǫ(x)(v + σ(x)) = −ǫ(x) δAµ (x) = M + σ(x) . To introduce the gauge choice. 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 . 1 ∂µ ǫ(x).34) (namely.39). it is not the right gauge to study renormalizability of the theory. 2ξ 2 (14. eliminate the mixing term between the Aµ and the χ ﬁelds). commonly known as the ’t Hooft gauge or the Rξ gauge.42) .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.41) e where M is deﬁned in (14. 2ξ 2ξ ξM 2 2 1 µ (∂ Aµ )2 + M (∂ µ Aµ )χ − χ . First we note that the unshifted theory (14. (14. δχ(x) = −ǫ(x)σ(x).40) e The shifted theory (14. e 1 ∂µ ǫ(x). (14.33) which reﬂects a spontaneous breakdown of the local symmetry (14.

42) can also be written with an auxiliary ﬁeld as (see (12. (14. the ’t Hooft gauge choice in (14. It is clear that the mixing terms in (14.42) (or (14.170)) LGF = ξ 2 F − F (∂ · A − ξM χ). (14. disappear from the action.45) We see that unlike in the conventional covariant gauge ﬁxing (12. The gauge ﬁxing Lagrangian density (14.44)) leads to an interacting ghost Lagrangian density. the propagators for the Aµ and the χ ﬁelds in this gauge are given respectively by iGµν (p) = − iGχ (p) = p2 p2 i . 2 −M p − ξM 2 (14.43) and we note that in this gauge the propagators fall oﬀ for large values of the momenta unlike in the unitary gauge.2 Higgs phenomenon 595 which can be compared with (14.44) and the choice of this gauge ﬁxing leads to a ghost Lagrangian density of the form (scaling out a factor of 1 ) e Lghost = ∂ µ c∂µ c − ξM 2 cc − ξM eσcc. 2 (14.15).164).34) together with the cross terms in (14.42) combine into a total divergence and. − ξM 2 pµ pν i ηµν − (1 − ξ) 2 . thereby. As a result. The complete Lagrangian density in this gauge has the form LTOT = L + LGF + Lghost .46) and can be easily checked to be invariant under the BRST transformations .14.

M + σ c. the β decay of the neutron etc.4 (see. 2 (14. e δσ = ωχc. (13. 14. let us review some of these essential earlier results before constructing the standard model (known as the Weinberg-SalamGlashow theory). for example. weak interactions (for example.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. Therefore. e δχ = −ω δc = 0. At low energies. This is the starting point for establishing the renormalizability of the gauge theory with a spontaneously broken symmetry (Higgs theory).596 14 Higgs phenomenon and the standard model δAµ = 1 ω∂µ c.47) Following the discussions in section 13. δF = 0.49) .48) where Pσ and Pχ denote respectively the sources for the composite BRST variations of the ﬁelds σ and χ in (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. (14.) 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 . δc = −ωF.47). δAµ (x) δσ(x) δPσ (x) δχ(x) δPχ (x) δc(x) (14.

ψe . proton. therefore. 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. it is clear from the V − A (vector minus axial vector) structure of the currents in (14. I3L the quantum number associated with the 3-component of the weak isospin generator and Y its weak hypercharge quantum number. 2 Q = I3L + (14. (The V − A structure is connected with the fact that parity is violated maximally in weak interactions.17 × 10−5 (GeV)−2 . Furthermore.139) and (3. ψp .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 + · · · . (14.50) with ψn . 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 .) In addition to the weak isospin quantum number. Experimentally .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 (3.51) The weak interactions are.50) have a weak isospin structure and are also known as the weak isospin currents. As a result.14. ψν denoting the fermion ﬁelds associated with neutron. electron and neutrino respectively. 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).52) where Q denotes the electric charge of the particle. the currents in (14. (14. quite weak and were also known to be extremely short ranged.140) for a deﬁnition of right and left-handed particles/ﬁelds).

therefore. .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.) Therefore. On the other hand. we would like the gauge theory to describe the symmetry behavior (“SSB” denotes spontaneous symmetry breakdown) SSB SUL (2) × UY (1) −→ UEM (1). 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). (14. since weak interactions are short ranged. As a result. let us construct the standard model in several steps. (The nonrenormalizable character is already manifest in the fact that the coupling constant of the theory (14. 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.51) is dimensional with inverse dimensions of mass (energy). Furthermore.54) With these basics. a gauge theoretic description would naturally involve massive gauge bosons. it is clear that we should look for a gauge theory of weak interactions based on the local symmetry group SUL (2) × UY (1). 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. Although the V − A theory with four fermion interactions works quite well at low energy. As we have discussed weak interactions violate both the weak isospin and the weak hypercharge symmetries and. but in such a way that the electric charge is conserved in any weak process. it is not renormalizable. Namely. it is logical to associate the Higgs mechanism to the spontaneous breakdown of these symmetries. (14. the V − A theory cannot be the fundamental theory underlying weak interactions (forces).

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

54)) is given by a doublet of charged ﬁelds. minimal Higgs ﬁeld : φ = φ+ φ0 . 2. Let us now write down the quantum numbers for the matter and the Higgs ﬁeld and verify explicitly that (14.52) holds. the minimal multiplet of Higgs ﬁelds that can achieve the spontaneous breakdown of the symmetry (as in (14. 3. so that the theory should contain four gauge ﬁelds in total. then the corresponding gauge ﬁelds are determined to be SUL (2) : 3 gauge ﬁelds. a Wµ . As we will see shortly.52). the opposite of that for the left-handed lepton doublet.) The gauge ﬁelds which are expected to correspond to the carriers of forces are determined from the local symmetry structure of the theory. Furthermore.53) to describe the local (gauge) symmetry of the theory. If we assume (14.57) Yµ . (14.58) 0 Since the Higgs multiplet is charged. 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). (φ0 )† = φ . UY (1) : 1 gauge ﬁeld. in fact. a = 1.600 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. it carries a nontrivial weak hypercharge quantum number given by Y = 1 which is. . (14. (φ+ )† = φ− . if we would like the gauge bosons to be massive.

14.3 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. (14.2 Lagrangian density. let us note that the gauge invariant Lagrangian density for the gauge ﬁelds follows from our earlier discussions (see (9.3.4) and (12. (14. Note that ǫabc denotes the structure constant of SUL (2) as we know from the study of angular momentum. First. 4 4 a = 1. (The Abelian gauge ﬁelds are not self-interacting which is why g′ does not appear in the Lagrangian density for the gauge ﬁelds.52).60) where the ﬁeld strength tensors for the Abelian as well as the nonAbelian gauge ﬁelds are deﬁned as (see (9.16) and (12. but it will be present in the matter Lagrangian density.45)) to correspond to 1 1 a LG = − Fµν F µν a − Fµν F µν .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.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 (14.60) . a a b c = ∂µ Wν − ∂ν Wµ − gǫabc Wµ Wν . 2. 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. 3.44)) Fµν a Fµν = ∂µ Yν − ∂ν Yµ . 14.) The Lagrangian density (14.

3 and ǫ(x) denote the local inﬁnitesimal parameters of symmetry transformations associated with the groups SUL (2) and UY (1) respectively. g′ (14. Since the left-handed fermions belong to an isospin doublet while the right-handed fermions are isospin singlets. 2.63) is that the right-handed fermions couple twice as strongly to the Yµ . the gauge ﬁeld Wµ for the group SUL (2) does not transform under the group UY (1) and vice versa. (We note here that within the context of isospin.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) . 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. We have also used a vector notation here to represent the three components of the isospin vectors which is often used interchangeably. 2 2 (14.) The other thing to note from the structure of (14. The gauge invariant Lagrangian density for the matter ﬁelds (leptons) can be obtained through minimal coupling. we simply have to remember that there are two local symmetries present in the theory.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. therefore.62) a δWµ (x) = δYµ (x) = where ǫa (x). they are inert under the corresponding transformations. sometimes it is conventional to denote the generators as τ by making the identiﬁcation 2 σ = τ . a = 1. 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). g g 1 ∂µ ǫ(x).

2 2 δR(x) = −iǫ(x)R(x).63) can be easily checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)L(x) − ǫ(x)L(x).14.66) together with the gauge ﬁeld transformations in (14. The transformations (14. We note that we have combined the three components of ǫa (x). Note also that the SUL (2) invariance forbids a mass term for the lepton ﬁelds (the electron and the neutrino are.3 The standard model 603 ﬁeld as the left-handed ones simply reﬂecting the Y quantum numbers for the two ﬁelds.65) can be checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)φ(x) + ǫ(x)φ(x). a = 1.62).62).65) We note that since the hypercharge of the Higgs doublet is just the opposite of that of the left-handed lepton doublet. δL(x) = −i (14. 2. 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. 3 into a vector ǫ(x) in (14. The Lagrangian density (14. A mass term as well as a potential for the Higgs ﬁeld can be introduced consistent with the SUL (2) symmetry.64) show explicitly that the right-handed fermions do not change under the SUL (2) transformations. In fact. 2 2 (14. we also need a Lagrangian density for the scalar Higgs ﬁelds which are expected to lead to a spontaneous breakdown of the local symmetries. since we . Besides the gauge ﬁelds and the leptons. The minimally coupled fermion Lagrangian density (14. 2 2 δφ(x) = −i (14.64).64) together with the transformations in (14. massless in this theory to begin with). the coupling to the UY (1) gauge ﬁeld is correspondingly with the opposite sign. therefore.

of course. (14. 2 2 δR(x) = −iǫ(x)R(x).65).64) and (14.69) which is invariant under the inﬁnitesimal local gauge transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). 4 λ > 0. the Yukawa interaction leads to masses for fermions after the spontaneous breakdown of the symmetry.68). (14. φ† = −m2 φ† φ + λ † 2 φφ . 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. (14. δL(x) = −i δφ(x) = −i a δWµ (x) = i σ · ǫ(x)φ(x) + ǫ(x)φ(x). here the Higgs ﬁeld is represented by an isospin doublet) V φ. it is worth noting that in spite of the similarity in their forms for the potential.66).63). 2 2 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) .70) δYµ (x) = .24). As we will see shortly. (14.60). φ† . collecting all the terms in (14. (14. g′ (14.68) can be checked to be invariant under the transformations (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 (14.67) which is invariant under the local transformations in (14.66).68) where h denotes the Yukawa coupling and the Lagrangian density (14. the fermions can have Yukawa interactions with the scalar ﬁelds which can be described by the Lagrangian density LY = −h Rφ† L + LφR . (14.67) and (14. g g 1 ∂µ ǫ(x). Therefore. In addition.

φ† = 0. (14.71) so that the minimum of the potential occurs at ∂V ∂φ ∂V ∂φ† λ † φφ 2 = φ† − m2 + = − m2 + = 0. (14.3 The standard model 605 14. 2m2 .73) φ φ† φ = = Furthermore.3.29)) 0 v √ 2 φ = . λ (14. We know that both SUL (2) and UY (1) symmetries are violated in weak interactions (the associated quantum numbers are not conserved in weak processes). the factor of √2 below is a consequence of the deﬁnition in (14. 2m v=√ .73) represents a local maximum while the second solution corresponds to the true minimum of the potential.14.72) λ † φφ 2 As we have seen in the last section. λ (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.67) in the form λ † 2 φφ .72). there are two solutions to (14.3 Spontaneous symmetry breaking. the weak interactions are short ranged so that we would like the gauge bosons to be massive. we have noted earlier in section 7. φ† = −m2 φ† φ + λ > 0.5 (and also in section 14. 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. furthermore. namely. We have chosen the Higgs potential (14.2) that the ﬁrst solution in (14. φ = 0. 4 V φ. Furthermore.74) .

68)). this is the sector of the theory that involves the scalar ﬁeld and will be aﬀected by the shift. (14.58) we obtain . 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+ .67) and (14. (14. the other parts will not change under the shift.65). 2 As a result. it follows that 3 Wµ √ − 2Wµ (14. before shifting the ﬁelds let us rewrite the Lagrangian densities for the Higgs and the Yukawa sector of the theory for simplicity (see (14.77) σ · Wµ = 2 = 1 2 √ + 2Wµ φ+ φ0 .75) However. √ − 3 2Wµ −Wµ (14.606 14 Higgs phenomenon and the standard model Therefore. (Basically. (14.) 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 ). as we have discussed earlier.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.

14. 4 (14. 2 v 0 0 φ → φ +√ .3 The standard model 607 LH + LY − V φ. 2 (14.80) under which the Lagrangian density (14.79) The expansion around (14.74) can be done equivalently by letting v φ0 → φ0 + √ .79) in the Higgs and the Yukawa sector becomes LH + LY − V φ. φ† = ∂µ + ig − 0 ig 3 ig′ v Yµ − Wµ φ− − √ Wµ φ + √ 2 2 2 2 . 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 .

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. (g′ Yµ + gWµ ) as well ± appear to have become massive. Adding the quadratic terms from the gauge and the fermion Lagrangian densities in (14.60) and (14. we cannot truly determine the spectrum of the theory until we diagonalize the quadratic part of the Lagrangian density.63) respectively.81) . since as the ﬁelds Wµ there is mixing between various ﬁelds. − √ (eR eL + eL eR ) − 8 2 0 3 We note that the ﬁeld combinations (φ0 + φ ). However.82) hv λv 2 0 0 2 φ +φ . the complete quadratic Lagrangian density is obtained to be .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. The spectrum of the theory can be obtained from the quadratic part of the Lagrangian density and the quadratic part of (14.

84) where θW is known as the Weinberg angle (also called the weak mixing angle). Yµ = Zµ sin θW + Aµ cos θW . 8 2 It is now a straightforward matter to diagonalize the Lagrangian density (14.83) + λv 2 0 hv 0 2 − √ (eR eL + eL eR ) − φ +φ .85) .14.83). Let us deﬁne g′ g2 + g′2 g g2 + g′2 Zµ = Yµ + 3 Wµ 3 = sin θW Yµ + cos θW Wµ . (14. (14.84) to write 3 Wµ = Zµ cos θW − Aµ sin θW .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. Aµ = g g2 + g′2 Yµ − g′ g2 + g′2 3 Wµ 3 = cos θW Yµ − sin θW Wµ . We can invert the relations in (14.

2 λ 4m2 λv 2 = 2 2 λ = 2m2 . 2 i 0 √ φ − φ0 . 2 cos θW hv √ . (14. we can write the quadratic part of the Lagrangian density (14. (14. φ = 0 1 √ (σ 2 + − iχ).86) 1 √ (σ 2 ∂µW ν 3 − ∂ν W µ 3 Similarly.89).29)) σ = χ = 1 0 √ φ + φ0 . (14.89) MW MZ = = me = 2 MH = then together with (14. 2 2 (14. see. 2 MW g2 + g′2 v = . 2 (14.83) as . for example.84)-(14. we can deﬁne (this is equivalent to deﬁning φ0 = iχ).610 14 Higgs phenomenon and the standard model With this.87) so that we can write 0 ∂µ φ ∂ µ φ0 = 1 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ.88) Furthermore. it is easy to check that 3 3 ∂µ Wν − ∂ν Wµ = (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) + (∂µ Yν − ∂ν Yν ) (∂ µ Y ν − ∂ ν Y µ ) + (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) . if we deﬁne the parameters gv .

eR to deﬁne the massive electron ﬁeld as (see (14.90) 2 2 where we have combined the two chirality states eL .(14.56)) e = eL + eR .91) .14.3 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 . If we now redeﬁne the (massive) gauge ﬁelds as (14.

1 2 2 M σ 2 H (14. Zµ is also massive with mass . namely.38)).612 14 Higgs phenomenon and the standard model ± Wµ ∓ i ± ∂µ φ± → Wµ . we see that after spontaneous breakdown of symmetry the electron has become massive with mass hv me = √ . MW 1 ∂µ χ → Zµ . 2 (14.93) The quadratic Lagrangian density (14.92) Zµ − then the quadratic Lagrangian density (14. 2 (14. MZ (14.90) 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 .94) while the neutrino remains massless as we would like. The Higgs √ particle has the usual mass MH = 2m (see (14. Although the starting theory had a massless electron.95) One of the two neutral vector bosons.93) is now completely diagonalized and shows some very desirable features. The charged ± vector (spin 1) bosons Wµ are massive with mass MW = gv .

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

(14. 2 2 2 .99) Thus.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) We note from (14.98) that the massless neutral vector boson Aµ couples only to the charged fermions as the photon should and we can identify the electric charge as gg′ g2 + g′2 e= = g sin θW = g′ cos θW . 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. (14. in addition to the correct electromagnetic current. 2 2 2 g g + Jµ = − √ eL γµ νeL = − √ eL γµ (1 − γ5 ) νeL .98) that the standard model has. both charged as well as neutral weak currents given by g g − Jµ = − √ ν eL γµ eL = − √ ν eL γµ (1 − γ5 ) eL . We also note from (14.

102) which leads to 2 MW = (14.100) We recall that the V − A theory describes only charged current interactions. 8MW (14.98).14.101) This has exactly the form of the V − A four-fermion interaction in (14. (14. two charged currents interact through the charged W gauge boson propagator (exchange) as shown in Fig. they would lead to a low energy eﬀective current-current interation Hamiltonian density in the lowest order of the form (basically.1: Charged current-current interaction in the lowest order.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 ) . 4 2GF 4 2GF sin2 θW = (14. If we look at the charged current interaction terms in (14. 14.103) Experimental measurements lead to a value of the Weinberg angle as .49) (for just one family of leptons) and comparing the two we determine G √F 2 g2 . 2 8MW g2 e2 √ = √ . W Figure 14.1 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 .

51) and recall that we have set = c = 1) e2 1 ≃ . 14. (14. Search for Higgs particles is a top priority in many experiments at the upcoming LHC.104) and using the values of other constants (see. (14. (14. . 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). 4π 137 we can now determine MW MZ = = GF ≃ 10−5 (GeV)−2 . et le Radium 7. the WeinbergSalam-Glashow theory does not predict a unique mass for the Higgs particle since its mass depends on the quartic coupling λ. Proca. Such processes have also been experimentally observed. 347 (1936). Once again these can be accomodated into extensions of the standard model that we will not go into here. the standard model diﬀers from the conventional Fermi theory in (14.105) These gauge bosons have been discovered and their masses are determined very close to the theoretically predicted values. e2 √ 4 2GF sin2 θW MW ≃ 90 GeV.616 14 Higgs phenomenon and the standard model sin2 θW ≃ 0. de Phys. Furthermore. J.4 References 1. However. The standard model compares very well with the experimental results. for example. cos θW 1 2 ≃ 80 GeV. although bounds on the value of its mass can be put from various other arguments.49) in that it predicts weak neutral currents (in addition to the conventional charged weak currents) and hence processes where a neutral heavy vector boson (Zµ ) is exchanged. A.23.

J. B. G. Feynman and M. ’t Hooft. P. Englert and R. E. 3265 (2004). Brout. Hagen and T. Kibble. 585 (1964). 7 (1974). St¨ckelberg. 8. Proceedings of PaduaVenice conference on mesons and newly discovered particles. F. Physical Review 109. Ruegg and M. C. 9. International Journal of Modern Physics A19. Nuclear Physics B35. 1860 (1958). 4. W. Physical Review 145. Gell-Mann. 167 (1971).4 References 617 2.14. Higgs. Physical Review Letters 19. G. . Bernstein. E. Physical Review Letters 13. 193 (1958). 321 (1964). 7. H. 5. Marshak. Helvetica Physica Acta 30. 6. G. 209 (1957). Reviews of Modern Physics 46. 1264 (1967). C. R. R. C. W. 1156 (1966). Physical Review 109. u 3. P. (1957). Physical Review Letters 13. Ruiz-Altaba. Sudarshan and R. E. 11. S. 10. Guralnik. Weinberg. G. S.

This page intentionally left blank .

we have to ﬁnd a way of extracting meaningful results from such a quantum ﬁeld theory. let us look at one of the simplest interacting ﬁeld 619 . As we have discussed earlier (see section 13. In such a case. in general. As a result. the Feynman diagrams become topologically more complicated and may contain internal propagators whose momenta are not uniquely determined in terms of the external momenta. as we go to higher orders in perturbation. This procedure is known.4). any scattering matrix element (S matrix element) in a given quantum ﬁeld theory can be built out of 1PI (one particle irreducible) graphs. For example.6.6 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 9.1 Introduction So far we have discussed the basic structures of various quantum ﬁeld theories. We will study this question in a systematic manner developing the necessary ideas in this chapter as well as in the next. However. The simple diagrams in 9. as renormalization of a quantum ﬁeld theory and involves several steps. Instead some of the internal propagators may involve additional momenta (loop momenta) that are integrated over all possible values.6 there was only one internal propagator). 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.Chapter 15 Regularization of Feynman diagrams 15. We have also carried out a few calculations of simple Feynman diagrams describing physical processes in section 9. the evaluation of Feynman diagrams may lead to divergences depending on the structure of the integral that is being evaluated.

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

∂µ φ. Since the 1PI graphs are fundamental. it is suﬃcient to study the divergence behavior of only the 1PI graphs. and P is the power of “a” associated with this graph. we invariably run into divergences. 15. As we have mentioned.2).2 Loop expansion As we have mentioned repeatedly. to study the divergence structure of any amplitude (graph). 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.3) where a is a constant parameter. ∂µ φ) . this discussion applies to any quantum ﬁeld theory) L (φ. (15.15. Therefore. be convergent. Thus if we are considering a proper vertex graph (1PI graph). 15. in evaluating scattering amplitudes particularly at the loop level (where the topology of the diagram involves loops as in Fig. let us discuss the notion of loop expansion in a quantum ﬁeld theory in this section. In such a case.1 and Fig. a) = a−1 L (φ. we are forced to give a meaning to the Feynman amplitudes in some manner and this process is known as regularization of the diagrams. the integrals over the arbitrary loop momenta may not. then it follows easily that . 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 621 Figure 15.2: An example of a diagram in the φ4 theory which is not 1PI. In the discussions below we will introduce a number of regularization schemes that are commonly used in studying divergent Feynman amplitudes. Namely. then it is clear that each interaction vertex in the theory will have a factor a−1 multiplied to it. in general. 15. problems with divergences arise in a quantum ﬁeld theory at the loop level.

we have set = 1 throughout our discussion. 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.4) where I denotes the number of internal lines and V the number of interaction vertices in the graph. (15. be thought of as the parameter of expansion in the number of loops.5) where the term (+1) reﬂects the fact that every 1PI vertex function has an overall energy-momentum conserving delta function associated with it. Let us recall that in the path integral description of a quantum ﬁeld theory. the power of “a” in a diagram in such a theory also determines the number of loops in the diagram and can.3).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. 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). the generating functional is deﬁned as (see. 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 . From the two relations in (15.120).4) and (15. is given by L = I − V + 1. for example. therefore. (15.622 15 Regularization of Feynman diagrams P = I − V. .5). therefore. we can eliminate I − V to obtain L = P + 1. the number of loops (independent momenta) in such a graph. (12. (15.6) In other words. (15.

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

) Since large values of momentum correspond to small values of coordinate separation. Furthermore.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 . We see from (15.8) takes the form .8) that this integral is divergent and. y + M2 (15. therefore.8) dy 1 − Here we have used the fact that the Feynman diagram in (15. With this regularization.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. such divergences are also known as ultraviolet divergences. the two point function in (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. (This is the reason for the name cut-oﬀ regularization.

Λ are scalar parameters).12) . (15.8) in a regularized manner as ∞ 0 I = − = = = where iλ 32π 2 dy 1 − ∞ 0 M2 y + M2 M2 y + M2 ˜ e− Λ2 y ˜ Λ→∞ lim − iλ 32π 2 dy 1 − ˜ Λ→∞ lim − lim − iλ ˜ 2 M2 1 Λ + M 2 γ + ln + O( ) 2 ˜2 ˜2 32π Λ Λ ˜ 1 Λ2 iλ ˜ 2 Λ + M 2 γ − ln 2 + O( ) .10) (and m.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 . However.11) ˜ Λ→∞ n γ = lim n→∞ k=1 1 − ln n k = 0. in this case this is also the result in the Minkowski space. since there is no momentum vector in the ﬁnal result in (15. We note that we can write (15. 2 ˜ 32π M Λ2 (15.577215665. We add here for completeness that the cut-oﬀ can also be implemented through a weight factor in the following manner.15.10) The result at the end should be rotated back to the Minkowski space. (15.

1). (15. we also note here that the diagram in (15. (5. for example.13) Furthermore. 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 ) .14) d3 k where we have identiﬁed (see. the integration over the angles can be carried out. (15.15) 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 + . 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 .47)) Ek = |k|2 + M 2 . deﬁning y = |k|.10) with the identiﬁcation ˜ Λ2 = Λ2 − M 2 γ.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. Gradshteyn and Ryzhik 3. which gives (4π) in three dimensions.214.14) does not depend on the angular variables.4 and 8. Furthermore. Ek (15.352. for example. This result coincides with (15.

but is divergent at the lower limit τ = 0.2 and 2.17) is well behaved for large values of τ .271. for example. e τ2 Here Γ(2) denotes the gamma function and the parameter τ is known as the Schwinger parameter (or the proper time parameter).3 Cut-off regularization 627 Here we have used the standard table of integrals in the intermediate step (see. This result coincides with (15. Gradshteyn and Ryzhik.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 .271.15. 2.8) can also be evaluated in an alternative manner which is sometimes useful and so let us discuss the alternate method as well.10) if we identify 1 ¯ (15.16) Λ2 = Λ2 − M 2 (1 − ln 2) .4). Thus. 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 (15.17) dτ −τ M 2 . The integral in (15. We note that the integrand in (15. 2 This derivation makes it clear that the evaluation of the amplitude is much easier in the Euclidean space. the ultraviolet divergence in (15.

Gradshteyn and Ryzhik 3. for example.471. we note that the result in (15.18)) that the self-energy diagram at one loop contains a term that diverges quadratically as well as a term which is logarithmically divergent.10) (or (15.18) can be compared with (15. It is also clear from this simple calculation that the nature of the divergence depends on the dimensionality of space-time (we are restricting ourselves to four space-time dimensions. For example. Let us next look at another one loop graph.18) can also be obtained by simply cut1 ting oﬀ the τ integral at the lower limit Λ2 (without the regularizing exponential factor).3.18) 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. Equation (15. For simplicity we would consider all the incoming momenta to vanish. the one loop diagram for the vertex correction shown in Fig. Without going into details.9) to evaluate the integral over τ .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 . However. We note that since the cutoﬀ Λ has to be actually taken to inﬁnity in the ﬁnal result. but quantum ﬁeld theories can be deﬁned in any number of space-time dimensions). we note from (15. This diagram is also divergent (in four dimensions) and with the cut-oﬀ regularization leads to (we assume p1 = p2 = p3 = p4 = 0 for simplicity so that each graph contributes the same amount and also because we are interested in looking at only the divergence structure of the graph) . namely. this graph is divergent. regularizing the integral brings out the divergence structure of the Feynman graph. (15.10) and we see that both ways of evaluating the integral lead to the same result. 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 .3 Cut-off regularization 629 p3 p2 + p1 p4 p1 p3 p2 p4 p4 p3 + p1 p2 Figure 15.15. First we have factored out an overall momentum conserving delta function along .3: One loop correction to the four point vertex function.19) Λ→∞ lim There are several things to note from this derivation. (15.

Finally.) 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 the examples we have studied in (15. the present divergences are nontrivial and are present even after normal ordering of the theory.19). However.630 15 Regularization of Feynman diagrams with the factor of (2π)4 . Renormalization is the process of redeﬁning these inﬁnities that we encounter in perturbation theory. As a result. of course. 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.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. there is a net 4 − 4 = 0 power of momentum in the numerator reﬂected in the logarithmic divergence of the diagram.8) up to a multiplicative factor. Similarly. for the ﬁrst graph in (15. for example. Thus we see that this graph is also divergent (in four dimensions).3. (This can also be seen from the fact that the integral in (15. as we have seen earlier. 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. 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. Consequently. In fact. can be removed by normal ordering the Hamiltonian of the 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. is plagued with divergences at every (loop) order in perturbation theory (unless we are in lower dimensions).9) in the intermediate step. for the diagrams in Fig. The divergence structure of any diagram can be determined simply by counting the powers of momentum in the numerator and in the denominator. each of these diagrams has an associated symmetry factor of 1 2 and we have used the value of the angular integral from (15.19) is ∂ given by − ∂M 2 of that in (15. the cutoﬀ regularizes and deﬁnes . 15. any given quantum ﬁeld theory. but the divergence in this case is logarithmic. Thus. This is. Very broadly it consists of two essential parts. in general. equivalent to saying that since we only measure diﬀerences in the energy levels. we are free to redeﬁne the value of the ground state energy to zero.10) and (15.

21) .15. 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.1 Calculation in the Yukawa theory.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 lack of dependence on the external momenta is simply due to our special choice of the external momenta to be all vanishing.19). In the previous examples. (15. p−m / p − m2 = −(2π)4 igδ4 (p1 + p2 + p3 ). depends analytically on the external momenta. in general. But in this process the amplitude (and the theory) becomes cutoﬀ dependent. (15. − M2 p = iGF (p) = p2 p p3 p2 p1 p4 p3 = iSF (p) = i(p + m) / i = 2 .3. 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. namely. particularly in the second example (15. 15. the theory with a Yukawa coupling described by the Lagrangian density 1 λ M2 2 / φ − gψψφ − φ4 .3 Cut-off regularization 631 the integrals. p1 p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ).

(15. Suppose we would like to write AB as a single factor. Therefore. AB 1 1 A − B x(A − B) + B dx (x(A − B) + B)2 1 0 (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. Before we proceed to calculate higher order diagrams in this theory. the iǫ term with the same sign). then we note that 1 0 1 dx (xA + (1 − x)B)2 = 0 = − = − = 1 1 1 − A−B A B 1 .23) holds and is quite . On the other hand when the two factors have opposite analytic behavior (iǫ terms with opposite sign).9) which is why the cubic coupling involving fermions is called the Yukawa coupling. let us note that at higher orders in perturbation theory.632 15 Regularization of Feynman diagrams where the arrows in the vertices denote the fact that momenta are all incoming. however. 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 ) . then this formula needs to be generalized (which is important in studying ﬁnite temperature ﬁeld theories).22) Therefore. it would be useful to ﬁnd a way to combine denominators and this is achieved by 1 Feynman’s formula as follows. The lowest order graph involving the fermions in this theory gives rise to the Yukawa potential (as we have seen in section 8. (x1 A + x2 B)2 (15. In quantum ﬁeld theory at zero temperature. diagrams necessarily involve more internal propagators.

let us next calculate the two simplest nontrivial graphs involving loops in this theory.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 ) .23). 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 / / .26) Using the Feynman combination formula (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) .15.24) which we have already used in (13.25) 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 .3 Cut-off regularization 633 useful. Generalization of (15. / / (15. (15. Equipped with the Feynman combination formula. the integral in (15.121). ( n xi Ai )n i=1 (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) / .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 . (15.27) 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. Furthermore.27) where we have deﬁned Q2 = x(1 − x)p2 + xm2 + (1 − x)M 2 . E (15.

M.30) takes the form .31) Tr A = 0.3 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. (15. the trace is a consequence of the fact that the Dirac matrix indices coming from the two fermion propagators are being summed over (there are no free Dirac indices in the diagram).30) p = −Tr (−ig)2 = −g2 Tr (k + m)((k + p) + m) / / / d4 k .15. / Tr A / = 4A · B. /B the self energy graph in (15. Furthermore. Λ. m. We also note here that we have used the integral from (15. 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. Using the trace identities Tr ½ = 4. where it is understood that the limit Λ → ∞ is to be taken.19) in the intermediate step and have rotated back to Minkowski space in the ﬁnal step.29) = f (p. g). 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.

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 .

for example.15. These two simple calculations show that Feynman amplitudes are functions of external momenta. Λ). In fact.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.32) we can write . we can write = ig2 16π 2 1 0 dx ((1 − x)p + m) / Λ2 − x(1 − x)p2 + xm2 + (1 − x)M 2 −1 −x(1 − x)p2 + xm2 + (1 − x)M 2 dx m ln Λ2 + xm2 + (1 − x)M 2 −1 xm2 + (1 − x)M 2 × ln = ig2 16π 2 1 0 +(1 − x)p ln / Λ2 + xm2 + (1 − x)M 2 −1 xm2 + (1 − x)M 2 + ﬁnite terms as Λ → ∞ = ig2 16π 2 1 0 dx ((1 − x)p + m) ln / Λ2 xm2 + (1 − x)M 2 (15.3 Cut-off regularization 637 2 2 ig2 = lim Λ→∞ 4π 2 1 0 dx Λ − 3Q ln Λ2 + Q Q 2 2 + 2Q 2 = f (p. for the scalar self-energy in (15. g. a Feynman amplitude with n external lines is an analytic function of (n − 1) external momentum variables.33) +ﬁnite terms as Λ → ∞ . m. This is because the overall momentum conservation eliminates one of the momentum variables. Similarly. In the examples worked out above it is clear that one can Taylor expand the amplitudes around zero momentum of the external lines. for the fermion self-energy in (15.29). Thus. 2 (15.

Therefore. 15. This is because of the infrared divergences of the theory which can be seen in the above examples by setting m = 0. 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. leads to a lack of manifest rotational invariance (although in the continuum limit this symmetry is recovered).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 (15.4 Pauli-Villars regularization Although regularizing a Feynman diagram by cutting oﬀ contributions from large values of momentum seems natural. To avoid this problem. Furthermore. in massless theories we Taylor expand the amplitudes around a nonzero but ﬁnite value of the external momentum. then expanding around zero external momenta would be disastrous. a lattice regularization which regularizes a theory by deﬁning it on a discrete space-time lattice (we will not go into a detailed discussion of lattice regularization). it is useful to ﬁnd . while quite useful. Similarly. Let us note here that if the theory is massless. 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).34) +ﬁnite terms as Λ → ∞ .

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. let us consider QED (quantum electrodynamics) in the Feynman gauge (see (9. we can now calculate various 1PI amplitudes in QED.35) ν = iGF .125)) 1 1 /ψ L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 .µν (p) = − iηµν .15. p2 p q p = iSF (p) = i(p + m) / .36) With these Feynman rules. 4 2 which leads to the Feynman rules µ p (15. The Pauli-Villars regularization provides such a regularization scheme. 2 − m2 p r µ = −(2π)4 ieγ µ δ4 (p + q + r).(15. To discuss the Pauli-Villars regularization scheme.4 Pauli-Villars regularization 639 a covariant regularization scheme which also respects gauge invariance.

µ))2 ∞ 0 2 dkE −1 . 2 kE + Q2 (m. µ) (kE + Q2 (m.111) and (2.38) 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. We have also used gamma matrix identities in (2.112) as well as (15. µ))2 dx ((1 − x)pE + 2m) / 2 dkE 1 Q2 (m.27)).26). µ) − 2 2 kE + Q2 (m. E (15. of course.) The momentum integral is. µ) = x(1 − x)p2 + xm2 + (1 − x)µ2 . µ) (15.640 = − = − = − × 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. (The Dirac gamma matrices have also been rotated to Euclidean space as discussed in (15. In the . µ))2 2 / kE (2(1 − x)pE + 4m) 2 (kE + Q2 (m. divergent (logarithmically by power counting) as is the case in the earlier calculation of the fermion self-energy with the Yukawa coupling (see (15.9) for the angular integral.37) ie2 8π 2 dx((1 − x)pE + 2m) / where we have deﬁned Q2 (m.

A′ are ﬁctitious heavy µ ﬁelds introduced to regularize diagrams. The contribution from the heavy photon to the self-energy would be exactly the same as what we have already calculated in (15. (The covariant derivative in (15. 2 kE + Q2 (m.37) coming from the diagram where a heavy photon is being exchanged.40) I′ = dx ((1 − x)pE + 2m) / so that the eﬀective regularized fermion self-energy can be written as ie2 8π 2 Q2 (m. let us add to the Lagrangian density of QED (15. decouple. Therefore.39) is deﬁned with the gauge ﬁeld combination Aµ + A′ .) µ We note that the sign of the Lagrangian density for these ﬁctitious ﬁelds is opposite to that of the standard ﬁelds and hence they act as ghost ﬁelds and subtract out contributions. without doing any further calculation. / ′ L′ = (15. Λ) (15. µ and ψ ′ . Namely.41) I (reg) = I + I ′ = − dx ((1 − x)pE + 2m) ln / .15. Λ) .35) another gauge invariant term of the form 1 ′ ′µν Λ2 ′ ′µ ′ / Fµν F − Aµ A − iψ D (A + A′ )ψ ′ 4 2 +Λψ ψ ′ + eψA ′ ψ. The idea is to take Λ → ∞ at the end and in that limit the heavy ﬁelds do not propagate and. 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.4 Pauli-Villars regularization 641 Pauli-Villars regularization. Q2 (m.39) where we assume that Λ ≫ m. we can write the additional contribution to the fermion self-energy as ie2 8π 2 ∞ 0 2 dkE −1 . µ) (15. However. therefore. these additional interacting ﬁelds give an additional contribution to the fermion self-energy (15.37) except for a sign and µ → Λ.

642 15 Regularization of Feynman diagrams which is ﬁnite for any given value of Λ. Λ) Q (m. The Feynman rules (15. The ﬁnal result can now be rotated back to Minkowski space and has the form ie2 = 2 8π Q (m.42) where Q (m.43) . µ) = −x(1 − x)p2 + xm2 + (1 − x)µ2 . µ) 2 2 I (reg) dx ((1 − x)p − 2m) ln / . (15. (15.36) lead to k p k+p i(k + m) i(k + p + m) / / / d4 k γ γ 4 µ k 2 − m2 ν (k + p)2 − m2 (2π) p = iΠµν (p) 2 = −(−ie)2 Tr =− =− kµ (k + p)ν + kν (k + p)µ − ηµν k · (k + p) + m2 ηµν 4e2 d4 k (2π)4 (k2 − m2 )((k + p)2 − m2 ) 4e2 (2π)4 d4 k dx × =− kµ (k + p)ν + kν (k + p)µ − ηµν (k · (k + p) − m2 ) ((k + xp)2 + x(1 − x)p2 − m2 )2 d4 k dx 4e2 (2π)4 × =− 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − (x(1 − x)p2 + m2 )) (k2 + Q (m))2 2 4e2 (2π)4 dx d4 k 2 × 2x(1 − x)(ηµν p2 − pµ pν ) + 2kµ kν − ηµν (k2 + Q (m)) (k2 + Q (m))2 2 . We can also calculate the one loop photon self-energy in QED in a similar manner.

15. η00 → E 2 ΠE . 2 (15. + y + Q2 (m) (y + Q2 (m))2 = d4 kE E 1 (−δµν ) + Q2 (m) . Π → (i)ΠE ) and we have δ00 . 2 (kE + Q2 (m))2 (15.45) individually.43) to 2 Euclidean space ((k0 . We can now integrate each of the E terms in (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 (15.44) The overall negative sign in (15.116) and have deﬁned Q (m) = x(1 − x)p2 − m2 . p2 ) → −(kE .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. p2 ). 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 .45) where Q2 (m) = x(1 − x)p2 + m2 . (k2 . p0 ) → i(k0E .115) and (2.

. However. This integral.45).47) 1 + δµν − 2 dy + Q2 (m) . recalling that the photon has coupling to ﬁctitious heavy fermions. as we see.46) =− π2 δµν 2 dy 1− (y + Q2 (m))2 Here we have used (15. 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. y + Q2 (m) d4 kE 2 (kE + Q2 (m)) 2 − k2 δµν 2 E + Q (m) (Q2 (m))2 . Putting this back into (15.644 = −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) . is quadratically divergent. we obtain for the self-energy of the photon (in Euclidean space).9) in the intermediate steps. (15.

Then the regularized photon amplitude with these two sets of heavy fermions will have the form iΠE (reg) (pE ) = iΠE (pE . see. 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) . the regularized photon self-energy will have the form iΠE (pE . for example.1) which would violate gauge invariance. (15.15. Λ1 ) − ic2 ΠE (pE . this has the unpleasant feature that the photon self-energy has developed a mass term (term proportional to δµν . Λ2 and charges c1 e.48) This shows that the diagram is ﬁnite for any given value of Λ. 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 . However. section 14. c2 e respectively. m) − iΠE (pE . Let us assume that we have two sets of heavy ghost fermions with masses √ √ Λ1 . Such a term can be cancelled by adding another set of heavy fermion ﬁelds. As a result. m) − ic1 ΠE (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.

the result (15. (15. the quadratic divergences as well as the logarithmic divergences in (15.50) then.49) would cancel leading to ie2 δµν p2 − pµE pν E E 2π 2 iΠE (reg) = − µν dx x(1 − x) × c1 ln Q2 (Λ1 ) + c2 ln Q2 (Λ2 ) − ln Q2 (m) . Pauli-Villars regularization involves introducing. n depending on the nature of the divergence in the diagram such that the regularized amplitude .(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 ) .43) does not involve the photon propagator. c1 Λ2 + c2 Λ2 = m2 . 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) . in a gauge invariant manner. 1 2 (15. that if the parameters satisfy c1 + c2 = 1. In general. therefore. since the diagram for the photon self-energy in (15. 2. · · · . a set of heavy ﬁelds with masses Λi and √ charges ( ci e) with i = 1.51) which represents a gauge invariant photon self-energy (it is manifestly transverse).52) holds for any covariant gauge with an arbitrary value of the gauge ﬁxing parameter ξ.49) E 1 2 It is clear.52) We note here that even though we are working in the Feynman gauge.(15.

(15. In fact. however. (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. the integral d4 k . i ··· . 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). Furthermore. However.4 which cannot be regularized by the Pauli-Villars method.15. (k2 − m2 ) ((k + p)2 − M 2 ) (15. i i ci Λ2 = m2 . 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.54) will be ﬁnite and gauge invariant. For example.) We should mention here that dimensional regularization has its own problems which we will discuss at the end of .55) is logarithmically divergent in four dimensions.5 Dimensional regularization 647 n I (reg) = I(p. This procedure is known as dimensional regularization. since gauge invariance is independent of the number of space time dimensions. it is quite useful in studying non-Abelian gauge theories for which the Pauli-Villars regularization does not work. in less than four dimensions it is ﬁnite. 15. i=1 (15.53) with the conditions on the parameters given by ci = 1. it is by construction a gauge invariant regularization scheme. there are graphs of the form shown in Fig. Λi ). m) − ci Ii (p. In non-Abelian theories. 15.5 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.

56) carries a dimension and introducing an arbitrary mass scale µ we can then write the Lagrangian density (15. Figure 15. but it is worth pointing out here that manipulations with dimensional regularization are extremely simple which is why it is quite useful.58) are given by .4: One of the one loop graphs contributing to the gluon self-energy.56) we can carry out the (canonical) dimensional analysis in n dimensions to determine n−2 . [φ] = [λ] = 4 − n. For example.57) Therefore.648 15 Regularization of Feynman diagrams this section. the coupling constant for the quartic interaction in (15. In dimensional regularization we analytically continue the theory to n dimensions. 2 2 4! L= (15.56) for the theory in n dimensions as M 2 2 λµ4−n 4 1 ∂µ φ∂ µ φ − φ − φ . if we are looking at the φ4 theory described by the Lagrangian density M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ . 2 [M ] = 1. (15. 2 2 4! L= (15.58) so that the coupling constant λ is rendered dimensionless. The Feynman rules for the theory (15.

since each vector index takes n values.15. it follows that in n dimensions we have µ ηµ = n. 2 (kE + Q2 )α where we have rotated to Euclidean space. we can separate out the angular part of the integral . − M2 p4 p3 = −(2π)4 iλµ4−n δn (p1 + p2 + p3 + p4 ).5 Dimensional regularization 649 p = p2 i . For example.60) Furthermore. 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 . E To evaluate this integral. p1 p2 (15. therefore. 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 (15.59) We need to be careful about manipulating tensors in n dimensions.61) I = = = = dn kE dn kE 1 . to calculate amplitudes in n dimensions. (15.

63) On the other hand. (15. 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 . (15. Let us consider the basic n dimensional Gaussian integral whose value is given by dn kE e− 2 kE 2 n = (2π) 2 . n ≥ 2. we can evaluate the integral in (15.62) The integral over the angles dΩ can be evaluated from the simple Gaussian integral as follows.64) n = 2 2 −1 Γ n 2 k2 dΩ. dθn−1 . We see that . n −1 n Γ n 22 Γ 2 2 n n (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. where. in n dimensions.65) Let us compare this with known results in lower dimensions that we are familiar with.650 15 Regularization of Feynman diagrams n−1 dn kE = kE dkE dΩ.

Thus. dθ1 sin θ1 dθ2 = 4π. (15. for the basic integral (15. 2 n .66) which agree with what we already know from explicitly doing these integrals. 2 p= q =α−p=α− (15. (15.67) Let us recall that the beta function is deﬁned as Γ(p)Γ(q) =2 Γ(p + q) ∞ 0 −p−q B(p. (15.69) the integral (15.67) can be written as . we see that with the identiﬁcation n .5 Dimensional regularization 651 n=2: n=3: n=4: dΩ = dΩ = dΩ = dθ = 2π.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 .68) Therefore.15. dθ1 sin2 θ1 dθ2 sin θ2 dθ3 = 2π 2 . q) = dt t2p−1 1 + t2 .

71) by diﬀerentiation. n Γ α − (2π)n Γ(α) (p2 + M 2 )α− 2 2 n (15. in fact.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 .72) Similarly we can obtain . (2π)n Γ (α) (Q2 )α− 2 n n n 2 Γ α− Γ(α) n 2 (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 . (2π)n Γ(α) (p2 + M 2 )α− 2 (15. For example.70) Rotating back to Minkowski space. any other formula can be obtained from (15.71) This basic integral generates all other integrals that we need for evaluating amplitudes in n dimensions and.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 . using (15.

At the end of our calculations we should.58). 2 (2π)n (M 2 )1− 2 n n = − (15. we are analytically continuing away from four dimensions (to a lower dimension where the integral is well deﬁned). First of all.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 . i..15. let us now calculate various amplitudes in the φ4 theory (15.e.74) becomes . (15.73) n kµ kν dn k n (k 2 + 2k · p − M 2 )α (2π) and so on. With these basic integration formulae.74) Let us next set n = 4 − ǫ. the amplitude (15. the one loop scalar self-energy takes the form (see also (15. With this. of course. take the limit n → 4 which translates to ǫ → 0.8)) 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 .

76) as well as (15. we have n 2 n Γ 2− 2 Γ 3− ǫ ǫ ≃ 1 − γ. 16π 2 2 (15. the scalar self-energy (15.75) takes the form . 2 2 2 ǫ ǫ = Γ = Γ 1+ 2 ǫ 2 ǫ 2 2 1 − γ) = − γ. Similarly.75) Let us next work out some of the identities involving the gamma functions that will be useful to us.654 15 Regularization of Feynman diagrams k p p n ǫ iλµǫ π 2 Γ 1 − 2 + 2 = − ǫ 2 (2π)n (M 2 )1−2+ 2 ǫ iλµǫ π 2 Γ −1 + 2 = − ǫ . ǫ (15.77). In the limit ǫ → 0. ≃ ǫ 2 ǫ = Γ 1+ ǫ Γ 2 ǫ = Γ −1 + = 2 −1 + ǫ 2 n Γ 1− 2 ≃ − 1+ ǫ 2 2 −γ ǫ 2 ≃ − + (γ − 1). 2 (2π)n (M 2 )−1+ 2 n (15. we have π2 (2π)n n = ≃ 1 1 n = ǫ (4π) 2 (4π)2− 2 ǫ 1 1 + ln 4π .77) Using (15.12).76) where γ denotes the Euler’s constant deﬁned in (15.

the amplitude in n dimensions takes the form = = 3 (−iλµǫ )2 2 3λ2 µ2ǫ 2 dn k (2π)n dn k 1 n 2 − M 2 )2 (2π) (k i 2 − M2 k 2 . this graph would be regularized to zero in dimensional regularization which is the type of argument used in (13.15. for simplicity.78) We note here that for M 2 = 0. In this case.3 and. we will set all the external momenta to vanish as we had done earlier in (15. 15. − + ln ǫ 4πµ2 (15. Similarly.122).19). we can calculate the one loop correction to the vertex function shown in Fig.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) .

the amplitudes diverge. Let us next calculate one loop ampli- tudes in QED described by the Lagrangian density in the Feynman gauge (see (15.656 = 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 .35)) 1 1 L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 .82) . 2 2 2 ǫ . as we take the limit ǫ → 0 (to go to four dimensions). to make the coupling constant dimensionless. 15.1 Calculations in QED.81) As a result. However. ǫ (15. 2 (15. /ψ 4 2 (15.5.79) where we have used (15. 2 2 n−1 3 ǫ ψ = = − . We note that these amplitudes are well behaved (regularized) for any ﬁnite value of ǫ. ǫ 4πµ2 (15.80) In n dimensions.76) and (15. 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). we let e → e µ2 .

p2 p = iSF (p) = i(p + m) / .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.µν (p) = − iηµν . ǫ (15.83) As a result.15.37) and the discussion following that equation) k p k+p ǫ p / / iηµν dn k µ i ((k + p) + m) ν γ γ − 2 (2π)n (k + p)2 − m2 k (15. we can calculate the fermion selfenergy at one loop which gives (see also (15.85) = −ieµ 2 2 = −e2 µǫ / / dn k γ µ ((k + p) + m) γµ . (15.84) ǫ With these Feynman rules. n k 2 ((k + p)2 − m2 ) (2π) . 2 − m2 p r µ q p = −(2π)4 ieµ 2 γ µ δ4 (p + q + r). the Feynman rules of the theory in n = 4 − ǫ dimensions are given by µ p ν = iGF .

86) Using this. ǫ 4πµ2 (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 . 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. the one loop fermion self energy in (15.89) . / / (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 (15.79) in n dimensions as well as (15.658 15 Regularization of Feynman diagrams We note here that if we use the algebra of the gamma matrices (1.

115) and (2. This can be compared with (15.15. /B / Tr γµ γν γλ γρ = n (ηµν ηλρ − ηµλ ηνρ + ηµρ ηνλ ) .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 .90) / / / dn k Tr γµ (k + m)γν (k + p + m) . /B Tr A /C = 0.42). Similarly the photon self energy at one loop can also be calculated to have the form (the overall negative sign is because of the fermion loop) k p k+p ǫ p = iΠµν (p) = − −ieµ 2 = −e2 µǫ 2 i(k + m) i(k + p + m) / / / dn k Tr γµ 2 γ n 2 ν (k + p)2 − m2 (2π) k −m (15.91) the photon self-energy (15.116)).77).5 Dimensional regularization 659 where we have deﬁned Q2 = −x(1−x)p2 +xm2 and have used (15. (15. / Tr A / = nA · B. n (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Using the n-dimensional identities (see (2. Tr ½ = n.76) as well as (15. Tr A = 0.

73). (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 .77).92) where we have deﬁned Q2 = −x(1 − x)p2 + m2 . Let us next look at the terms with kµ kν in the numerator separately which leads to . (k2 − Q2 )2 (15.76) as well as (15.92) contains three diﬀerent kinds of terms which can be integrated using the standard formulae in (15.71)(15.93) × ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν where we have used (15.660 × =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 .

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.

92) takes the ﬁnal form ie2 (ηµν p2 − pµ pν ) 2π 2 2 Q2 1 − ln −γ− .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.93)-(15. the gauge ﬁxing Lagrangian density does not receive any quantum correction which can also be seen from the BRST identities for QED. 2 ǫ 4πµ 2 (15.95) Q2 4 + 2γ .95). × − + 2 ln ǫ 4πµ2 Adding the contributions from (15. the photon self energy in (15. p q=0 µ k+p p =(−ieµ 2 )3 ǫ k+p k iηλρ / / / / dn k λ i(k + p + m) µ i(k + p + m)γ ρ γ γ − 2 n 2 − m2 2 − m2 (2π) (k + p) (k + p) k . For simplicity we will put the momentum of the external photon equal to zero. Since there is no longitudinal term in the photon self-energy. since the photon self-energy does not involve the photon propagator. this result holds in any covariant gauge with the gauge ﬁxing term ξ arbitrary. Let us next calculate the one loop amplitude corresponding to vertex correction (charge renormalization).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.

23) as well as the identity .97) as dn k (2π)n k2 ((k + p)2 − m2 )2 (15. n (2π) k2 ((k + p)2 − m2 )2 Using the n-dimensional identities (see also (2. we can write the amplitude (15.113)) γ λ A µ A λ = (2 − n)(2Aµ A − γ µ A2 ). / / Therefore. /γ /γ / as well as (15.15.86) we have (15.100) Let us next use (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 .5 Dimensional regularization 663 (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 / / . 2 ((k + p)2 − m2 )2 k (15.

664 15 Regularization of Feynman diagrams 1 AB 2 = − = ∂ 1 ∂ =− ∂B AB ∂B dx dx 2x .101) to combine the denominators in (15. we obtain for the vertex correction p q=0 µ k+p p = −e3 µ 2 × 3ǫ k+p k dx dn k −(2 − n)γ µ (2π)n (k2 − Q2 )2 / / dn k 4x (2 − n)kkµ + (1 − x) (2 − n)(1 − x)p + nm pµ 3 (2π)n (k2 − Q2 ) 3ǫ = −e3 µ 2 dx − (2 − n)γ µ i (−1)2 Γ 2 − n 2 n n (4π) 2 Γ(2) (Q2 )2− 2 .102) where we have deﬁned Q2 = −x(1 − x)p2 + xm2 . Using this in (15. 1 k2 ((k + p)2 − m2 )2 3. ((1 − x)A + xB)3 1 ((1 − x)A + xB)2 (15.100) and shifting the variable of integration.100) which leads to 1 + p)2 − m2 ) 1 (k + xp)2 − Q2 2x (k + xp)2 − Q2 k2 ((k = = dx dx 2. (15.

100).15.103) and (15.87). we see that the fermion self-energy and the vertex function are related as (see also (9. p k+p k+p p q=0 (15.5 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 (15.89) with (15.104) 2 (1 − x)p − 2m pµ / . (15.88) and (15. (15.105) . −(1 − x)p2 + m2 It is worth noting from this derivation that at every step if we compare (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) ie3 µ 2 16π 2 2 ln 4π − γ (−2 + ) × 1− + =− x(1 − x)(−2(1 − x)p + 4m)pµ / −x(1 − x)p2 + xm2 dx(1 − x) γ µ 2 − ln ie3 µ 2 8π 2 + Q2 −γ−1 4πµ2 (15.104) respectively.

We simply note here that we select a particular regularization scheme depending on the theory (or the problem) that we are analyzing. In addition to the regularization methods that we have discussed in this chapter. It is not naively applicable if the calculation involves quantities that typically exist only in four dimensions. the point splitting method. naive dimensional regularization leads to incorrect answers. (15. 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. we will not go into the details of these other methods here. the higher derivative method. However. (Note that the divergent parts as well as the ﬁnite parts satisfy the identity which demonstrates that dimensional regularization preserves gauge invariance. the ζ-function regularization etc. Physical Review 74. For example. for example. However. (15.) The power of dimensional regularization is quite obvious from these calculations. it has its own drawbacks. Schwinger. J.6 where we discuss chiral anomaly. we know that i 4! γ5 = iγ 0 γ 1 γ 2 γ 3 = − µνλρ γ µ ν λ ρ γ γ γ . there exist several other important regularization schemes. . We will discuss this further in section 16.666 15 Regularization of Feynman diagrams which we recognize as the Ward-Takahashi identity. Hence if the theory or the physical quantity of interest involves such objects. It is gauge invariant and extremely simple to manipulate with. 15.107) There are several ways to address this issue with the γ5 matrix. Analytic continuation of these to other dimensions is nontrivial. 1439 (1948). The lattice regularization is also quite useful.6 References 1.106) is deﬁned only in four dimensions (both γ5 and the Levi-Civita tensor µνλρ are manifestly four dimensional quantities).

N. J. ’t Hooft and M. Physical Review 82. 664 (1951). Villars. Feynman. J. Bollini and J. 769 (1949). 4. Pauli and F. Bogoliubov and D. Reviews of Modern Physics 21. 20 (1972). CERN preprint (1973). Diagrammar. Nuovo Cimento 12B. ’t Hooft and M. 8.6 References 667 2. Schwinger. N. 5. W. 3. G. Physical Review 76. 189 (1972). 7. Nauka. Shirkov. Introduction to the theory of Quantized Fields.15. C. Veltman. 434 (1949). Moscow (1984). . G. Giambiaggi. G. V. R. Veltman. 6. Nuclear Physics B44.

This page intentionally left blank .

Furthermore. we can expand them around some reference momentum value (in massive theories conventionally chosen to be zero) so that the regularized divergent parts can be isolated as local terms. the action for the theory (16.1 Superﬁcial degree of divergence In the last chapter. the Lagrangian density must have canonical dimension 4 in these units (since we are in four space time dimensions. In units of = c = 1 which we have been using. 669 (16. we have also seen that since Feynman amplitudes are analytic functions of external momenta.2) . To deﬁne this let us start with the Lagrangian density for a given theory L = L0 + Li . it is easy to check that [M ] = [L]−1 = 1. 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. we have seen that loop diagrams in various theories become divergent and need to be regularized.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.Chapter 16 Renormalization theory 16. (16. in n dimensions it should be n). In our units.1) is dimensionless. As a result.

Let us further introduce the notations fi .) Similarly the free fermion Lagrangian density (ψ) L0 = iψ∂ ψ − mψψ. for the Yukawa interaction LY = gψψφ. the canonical dimension of a scalar ﬁeld (in our units and in four dimensions) is 1. / 3 . and L0 we have [φ] = 1.9) .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 .3) [∂µ ] = [L]−1 = 1. 2 (16.4) (16. (The mass term also has the correct dimension with this assignment. Thus. (φ) (16.7) leads to [ψ] = [ψ] = (16.8) In fact. bi and di to denote respectively the number of fermions.6) Namely. (16. for example.5) = 4. bosons and derivatives at an interaction vertex following from the interaction Lagrangian density Li . 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. 2 2 L0 = that since (φ) (16. (16.

For example. b = 1. the φ4 interaction (16. κψσ µν ψFµν . d = 0. b = 4. 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. 4! (16. d = 1. With these notations we are now ready to introduce the notion of the superﬁcial degree of divergence of a Feynman graph. d = 0. we have or.1 Superficial degree of divergence 671 we have f = 2.10) LI = leads to λ 4 φ .12) LI = hψγµ ψ∂ µ φ. b = 1.14) and so on. (16. for an interaction of the form (16.11) f = 0. (16.13) f = 2. we can determine easily the superﬁcial degree of divergence of the following Feynman graphs as . Similarly.16. On the other hand.

D = 4 − 2 × 2 = 0.15) If D = 0 for a diagram. D = 4 − 2 × 1 = 2.672 16 Renormalization theory D = 4 − 2 = 2. D = 4 − 2 − 2 × 1 = 0. Similarly. D = 8 − 3 × 2 = 2. diagrams with D = 1 or 2 are known . we say that the diagram is superﬁcially logarithmically divergent. D = 4 − 2 − 1 = 1. (16. D = 8 − 5 × 2 = −2.

Rather than calculating the superﬁcial degree of divergence for each graph.16. If D < 0 for a graph. number of internal fermion lines. number of vertices of the ith type from Li . number of external fermion lines. Although this graph is superﬁcially convergent. Let us consider a Feynman diagram with B = IB = F IF = = number of external bosons lines. (16.15). 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 . However. Similarly. as we will see it is the notion of superﬁcial degree of divergence which is useful in the study of renormalization theory. it is actually divergent since it contains a subgraph which is divergent. then the Feynman diagram is said to be superﬁcially convergent.18) The superﬁcial degree of divergence is easily seen to be given by . (16. The meaning of the adjective “superﬁcial” becomes clear once we look at the last graph in (16.17) The factor 2IB reﬂects the fact that it takes two boson lines at two diﬀerent vertices to form a propagator (internal boson line). (16. let us develop a general formula for the superﬁcial degree of divergence of any connected Feynman graph. we have F + 2IF = i ni fi . number of internal boson lines.1 Superficial degree of divergence 673 respectively to have superﬁcial linear or quadratic divergences.16) ni = There exist topological relations between these numbers.

674

16 Renormalization theory

D=

i

ni di + 2IB + 3IF − 4

ni + 4.

i

(16.19)

The ﬁrst term in (16.19) simply says that each derivative at a vertex gives rise to a momentum in the numerator and if there are ni vertices of the ith type in a graph, this would lead to a power of momentum in the numerator ni di (which must be summed over all possible types of vertices). With each internal boson line is associated a momentum integration and a propagator. Thus each internal boson line eﬀectively leads to two powers of momentum in the numerator. Similarly each internal fermion line adds three powers of the momentum to the numerator. At each vertex, however, energy-momentum has to be conserved and since a four dimensional delta function (expressing conservation of energy-momentum) has dimension −4, each vertex takes away four powers of momentum except for an overall delta function which is necessary for the overall energy-momentum conservation. The last two terms reﬂect this. Using (16.17) and (16.18) and eliminating IB and IF from (16.19) we obtain 3 2 3 ni fi − F − 4 2

D =

i

ni di +

i

n i bi − B +

ni + 4

i

i

3 = 4−B− F + 2 3 = 4−B− F + 2 where we have deﬁned 3 δi = di + bi + fi − 4. 2

i

3 ni di + bi + fi − 4 2 ni δi , (16.20)

i

(16.21)

This is known as the index of divergence of the interaction Lagrangian density Li and can also be expressed as δi = dim Li − 4, (16.22)

16.1 Superficial degree of divergence

675

where dim Li is calculated only from the dimensions of the ﬁeld variables and derivatives and not from any dimensionful parameters. In all theories we consider, the interaction Lagrangian density has dimension 4 and hence

δi = 0. In such cases (16.20) reduces to 3 D = 4 − B − F. 2

(16.23)

(16.24)

Namely, in such cases the superﬁcial degree of divergence is completely determined by the number of external lines in the graph. Let us note here that the superﬁcial degree of divergence is a function of the number of space-time dimensions (we are working in four dimensions). We can check the relation (16.24) against our explicit power counting calculations in (16.15),

D = 4 − 2 − 0 = 2,

D = 4 − 4 − 0 = 0,

D = 4−0−

3 × 2 = 1, 2

D = 4 − 2 − 0 = 2,

676

16 Renormalization theory

D = 4 − 2 − 0 = 2,

D = 4−1−

3 × 2 = 0, 2

D = 4 − 6 − 0 = −2,

(16.25)

and they agree completely. From the form of the formula for the superﬁcial degree of divergence, it is clear that only a few graphs in a theory would have non-negative superﬁcial degree of divergence. Thus for example, if we are considering the φ4 theory, then only the following 1PI functions would be superﬁcially divergent. B 0 1 2 3 4 D =4−B 4 3 2 1 0

The zero point function contributes only to the zero point energy which can be eliminated by normal ordering the theory. The one point function is in principle divergent. However, such graphs do not exist in the φ4 theory since the theory has the discrete symmetry

φ → −φ,

(16.26)

16.1 Superficial degree of divergence

677

and the one point function violates this symmetry. Similarly, although the three point function can be superﬁcially divergent it does not exist because of this discrete symmetry. Thus, only two graphs are superﬁcially divergent, namely the 2-point and the 4-point functions. In the last chapter we have explicitly calculated the two point as well as the four point functions at one loop in the φ4 theory where we have seen that these graphs are indeed divergent. We have to develop a systematic way of making these graphs ﬁnite. Any higher point 1PI graph can, of course, contain these graphs as subgraphs and even though these graphs are superﬁcially convergent, they will in fact be divergent. However, corresponding to each such graph we can deﬁne a skeleton graph. Thus for example, in the φ4 theory, the graph for the six point function in Fig. 16.1

Figure 16.1: A superﬁcially convergent graph with a divergent subgraph. has the skeleton graph shown in Fig. 16.2.

Figure 16.2: The skeleton graph associated with Fig. 16.1. Namely, the skeleton graph of a superﬁcially convergent graph is a graph where no divergent subgraph can be found (divergent subgraphs shrunk to a point). The skeleton graph of a superﬁcially convergent graph is, therefore, by deﬁnition convergent. The full graph can be obtained from the skeleton graph by making insertion

678

16 Renormalization theory

of the two point and the four point functions at appropriate places. However, if we have a method of making these graphs, namely, Γ(2) , and Γ(4) , ﬁnite and regularization independent through some renormalization procedure, then the same procedure would make all the n point functions ﬁnite and regularization independent. Let us next look at QED described by the Lagrangian density

1 1 (∂µ Aµ )2 , /ψ L = − Fµν F µν + iψD − mψψ − 4 2ξ Dµ ψ = (∂µ + ieAµ ) ψ. (16.27)

The Feynman rules for the theory (in the Feynman gauge with ξ = 1) are

µ

p

ν = iGF ,µν (p) = −

iηµν , p2

p

= iSF (p) =

i(p + m) / , 2 − m2 p

r

µ

q p = −(2π)4 ieγ µ δ4 (p + q + r). (16.28)

Let us now analyze the divergence structure of graphs in this theory.

16.2 A brief history of renormalization

679

B 0 1 2 3 4 0 0 1

F 0 0 0 0 0 1 2 2

D = 4 − B − 3F 2 4 3 2 1 0 1 0

5 2

The zero point graphs are neglected because they can be taken care of by normal ordering. The 1 point boson graphs do not exist because they violate Lorentz invariance as well as gauge invariance. Similarly the 3 point boson graphs also do not exist. (Vanishing of photon graphs with an odd number of photons is a consequence of the symmetry of charge conjugation C, also known as the Furry’s theorem (see (11.188)).) The four photon graph, in this analysis, would appear to be superﬁcially divergent, but it is actually ﬁnite because of gauge invariance. In fact, in gauge theories the actual degree of divergence may be softer than the naive power counting because of constraints coming from gauge invariance. Fermion graphs with only an even number of fermion lines can be nonzero because of Lorentz invariance (as well as fermion number conservation). Therefore, there are only three possible superﬁcially divergent graphs, namely, the fermion and the photon self-energy as well as the fermion interaction vertex with the photon, and if we can somehow make these ﬁnite in a regularization independent manner, the theory will be well deﬁned. 16.2 A brief history of renormalization Dyson laid the foundations for the systematic study of renormalization in two papers in 1949 where he studied the renormalization of QED. He basically used the Schwinger-Dyson equations (to be discussed in the next section) and showed that most of the divergences can be absorbed into a redeﬁnition of parameters of the theory. The class of graphs which he could not incorporate into his study are known as overlapping divergent graphs of the type shown in Fig. 16.3. Salam showed how the overlapping divergences can be handled in any theory.

680

16 Renormalization theory

Figure 16.3: The overlapping divergent graph in the fermion selfenergy at two loops.

In QED things were a bit simpler because of the Ward-Takahashi identities (which relate various amplitudes, see for example, section 9.7) and renormalization of QED was thought to be straightforward. However, Yang and Mills noticed that at the 14th order (in the coupling), the photon self-energy graph shown in Fig. 16.4 does lead to overlapping divergence and needed further prescription to handle this graph as Ward-Takahashi identities do not restrict the photon selfenergy. (It was believed earlier that the photon self energy cannot have overlapping divergences.) Yang and Mills solved the problem of the photon self-energy and together with Salam’s work as well as the subsequent work of Weinberg, the question of overlapping divergence was considered to be solved. (This is within the framework of integral equations for Green’s functions and skeleton expansions.)

Figure 16.4: An overlapping divergent graph in the photon selfenergy at the 14th order. There are mainly two equivalent renormalization methods known as the multiplicative renormalization and the BPHZ renormalization. In multiplicative renormalization, we calculate 1PI Feynman amplitudes in perturbation theory (as we have done in the last chapter) until we encounter a divergent graph. The amplitude is then regularized with a momentum cut oﬀ (or dimensional regularization or any other regularization). The regularized amplitude is then Tay-

16.2 A brief history of renormalization

681

lor expanded about zero external momenta (for massive theories) so as to separate out the local divergent parts. We then add counter terms to the Lagrangian density to exactly cancel these divergences. We continue calculating with this modiﬁed Lagrangian density and add more counter terms whenever faced with new divergences. This procedure renders the theory ﬁnite. Let us consider a speciﬁc theory to see how this procedure works. We have seen that in the φ4 theory only the two point and the four point functions are divergent. Thus we add counter terms to cancel these divergences. For example, at one loop we obtain the counter terms from calculating Γ(2) and Γ(4) as shown in Fig. 16.5 so that the divergent contributions from the one loop graph and the counter terms cancel and render the amplitude ﬁnite as shown in Fig. 16.6. The counter terms are clearly of one loop order (this is counted by the power of in the coeﬃcient which we have set to unity, but should be understood).

Figure 16.5: One loop counter terms for the two point and the four point functions in the φ4 theory. Including these one loop counter terms, at two loops the self energy graphs would, therefore, have the form shown in Fig. 16.7. These would be divergent and hence we have to add two loop counter terms for the two point function to cancel the new divergences at this order. Similarly the four point function would also need counter terms at the two loop level. However, no other 1PI graph would be divergent at this order. As an example, let us look at the 6 point function at two loops in Fig. 16.1 whose superﬁcial degree of divergence is −2 but which is divergent because of a divergent subgraph. However, with the modiﬁed Lagrangian density (i.e., with one loop counter

682

16 Renormalization theory

+

= ﬁnite

+

= ﬁnite

Figure 16.6: The sum of the graphs and the counter terms make the amplitudes ﬁnite.

+

+

+

Figure 16.7: Two loop self-energy graphs including the one loop counter terms in the φ4 theory.

terms) there exist another two loop graph in this theory shown in Fig. 16.8 (this is of two loop order because the counter term is of one loop order).

Figure 16.8: Two loop 6-point function graph with the one loop vertex counter term in the φ4 theory. It is clear that since the counter term cancels the divergence of the one loop four point function, the sum of the two graphs in Fig. 16.1 and Fig. 16.8 is convergent. The point which this analysis brings out is that the only counter terms we really need to make a theory ﬁnite are for the graphs whose superﬁcial degree of divergence is non-negative (this is why the concept of superﬁcial divergence is important within the context of renormalization). The renormaliza-

16.2 A brief history of renormalization

683

tion procedure works because the structure of the counter terms is the same as the terms in the original Lagrangian density and hence the divergence analysis remains the same even after adding counter terms. Furthermore, since the counter terms have the same form as the terms in the original Lagrangian density, they can be simply absorbed into a redeﬁnition of the original parameters in the theory. For example, in the φ4 theory we have (C.T. stands for counter terms) M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ + C.T. 2 2 4! M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ 2 2 4! A B C + ∂µ φ∂ µ φ − φ2 − φ4 . 2 2 4! (16.29) The constants A, B and C receive contributions from various loops and are regularization dependent. We can now combine the counter terms with the original terms in the Lagrangian density as 1 1 1 (1 + A)∂µ φ∂ µ φ − M 2 + B φ2 − (λ + C)φ4 , (16.30) 2 2 4!

L = =

L=

and deﬁne

φ0 = (1 + A) 2 φ = Z 2 φ,

2 M0 =

1

1

M 2 + B (1 + A)−1 M 2 + B Z −1 = M 2 ZM Z −1 , (16.31)

=

λ0 = (λ + C)(1 + A)−2 = (λ + C)Z −2 = λZ1 Z −2 .

(The mass renormalization takes this form only if the regularization procedure does not introduce any mass parameter.) With this redefinition, therefore, the Lagrangian density becomes

684

16 Renormalization theory

L = =

1 1 1 (1 + A)∂µ φ∂ µ φ − M 2 + B φ2 − (λ + C)φ4 2 2 4! 1 M2 λ0 ∂µ φ0 ∂ µ φ0 − 0 φ2 − φ4 . 0 2 2 4! 0 (16.32)

We see that this Lagrangian density in (16.32) has the same form as the one we started out with. However, our new ﬁeld variable as well as the parameters m0 , λ0 have become regularization dependent. These are known as the bare ﬁeld and the bare parameters of the theory. (Sometimes they are also referred to as the unrenormalized ﬁeld and the unrenormalized parameters of the theory and are denoted respectively by φu , Mu , λu . We will use these notations interchangeably.) However, the renormalized ﬁelds and parameters are ﬁnite. Furthermore, if we calculate with the bare Lagrangian density and bare parameters, then the amplitudes will be ﬁnite in terms of the renormalized variables. Let us also note here that the renormalized parameters we are talking about are not the conventional renormalized parameters since our graphs are expanded around zero external momenta. However, these renormalized parameters are related to the usual renormalized parameters through renormalization group equations which we will study later. Since we have already calculated the one loop amplitudes in the scalar theory, let us indicate the one loop renormalization of the φ4 theory. The only divergences in one loop come from Γ(2) , Γ(4) which in the cut-oﬀ regularization have the forms (see (15.10) and (15.19))

= −

iλ Λ2 + M 2 Λ2 − M 2 ln , 32π 2 M2

=

3iλ2 Λ2 + M 2 −1 . ln 32π 2 M2

(16.33)

Thus, we see that to one loop the parameters of the counter terms in (16.29) are

16.2 A brief history of renormalization

685

A = 0, B = − C = Λ2 + M 2 λ Λ2 − M 2 ln , 32π 2 M2 (16.34)

3λ2 Λ2 + M 2 ln , 32π 2 M2

**which lead to the one loop deﬁnition of the bare ﬁelds and parameters as
**

1 1

φ0 = (1 + A) 2 φ = Z 2 φ = φ,

2 M0 =

M 2 + B (1 + A)−1 M2 1 + Λ2 + M 2 λ ln 32π 2 M2 − λΛ2 , 32π 2

=

λ0 = (λ + C)(1 + A)−2 = λZ1 = λ 1+ Λ2 + M 2 3λ ln . 32π 2 M2 (16.35)

On the other hand, we have also seen that in the dimensional regularization, we can write λ C M2 2 A B 1 ∂µ φ∂ µ φ − φ − µǫ φ4 + ∂µ φ∂ µ φ − φ2 − µǫ φ4 2 2 4! 2 2 4! 1 1 (λ + C) 4 = (1 + A)∂µ φ∂ µ φ − φ M 2 + B φ2 − µǫ 2 2 4! = M2 λ0 1 ∂µ φ0 ∂ µ φ0 − 0 φ2 − φ4 , 0 2 2 4! 0 (16.36)

L=

with (see (15.78) and (15.79)) λM 2 2 , 32π 2 ǫ 3λ2 2 , 32π 2 ǫ

A = 0,

B=

C=

(16.37)

so that we have

686

16 Renormalization theory

Z = 1,

ZM =

1+

λ 2 32π 2 ǫ

,

Z1 =

1+

3λ 2 32π 2 ǫ

, (16.38)

and in this case, we have deﬁned λ0 = µǫ λZ1 Z −2 . (16.39)

Our calculation so far has been at the one loop. However, at nloops, the divergence structure and, therefore, the counter terms in the dimensional regularization, in general, have the form am , ǫm m=−∞

m=n

(16.40)

where am represents constants. At higher loops, we calculate amplitudes using the counter terms already present at lower orders (namely, we also include diagrams coming from counter terms at lower order). Similarly, in QED in the covariant gauge in the dimensional regularization we can write the Lagrangian density of QED in the covariant gauge with counter terms as

ǫ 1 1 (∂µ Aµ )2 / /ψ L = − Fµν F µν + iψ∂ ψ − mψψ − eµ 2 ψA − 4 2ξ ǫ A Fµν F µν + iBψ∂ ψ − Cψψ − µ 2 DψA / /ψ 4 1 / = − (1 + A)Fµν F µν + i(1 + B)ψ∂ ψ − (m + C)ψψ 4 ǫ 1 /ψ − (e + D)µ 2 ψA − (∂µ Aµ )2 2ξ

−

1 (0) (0) (0) / = − Fµν F µν(0) + iψ ∂ ψ (0) − m0 ψ ψ (0) 4 2 1 (0) / − e0 ψ A(0) ψ (0) − ∂µ A(0)µ , 2ξ0 where the bare ﬁelds and the bare parameters are deﬁned as

**16.2 A brief history of renormalization
**

2 A(0) = (1 + A) 2 Aµ = Z3 Aµ , µ 2 ψ (0) = (1 + B) 2 ψ = Z2 ψ, 1 1 1 1

687

−1 −1 m0 = (m + C)(1 + B)−1 = (m + C)Z2 = mZm Z2 ,

e0 = µ 2 (e + D)(1 + B)−1 (1 + A)− 2

−1 = µ 2 (e + D)Z2 Z3

ǫ 1 −2

ǫ

1

−1 = µ 2 eZ1 Z2 Z3 2 ,

ǫ

−1

ξ0 = ξ(1 + A) = ξZ3 .

(16.41)

We have explicitly determined at one loop that (see (15.89), (15.96) and (15.104))

A = − B = − C = − D = −

e2 2 , 12π 2 ǫ e2 2 , 16π 2 ǫ e2 m 2 , 4π 2 ǫ e3 2 , 16π 2 ǫ

Z3 = Z2 = Zm = Z1 =

1− 1−

e2 2 12π 2 ǫ e2 2 16π 2 ǫ e2 2 4π 2 ǫ

, , , . (16.42)

1− 1−

e2 2 16π 2 ǫ

The Ward identities of the theory imply that Z1 = Z2 which is explicitly seen at one loop (see section 9.7 as well as the discussion after (15.104)). This, in turn, implies that e0 = eµ 2 Z3 2 . This is interesting because it says that the renormalization of charge depends only on the photon wave function renormalization. As a result, the charge (coupling) of any fermion is renormalized exactly in the same manner. This is commonly known as the universality of charge which is a consequence of the Ward-Takahashi identity of the theory. The theories which have the property that all the divergences can be absorbed into a redeﬁnition of the parameters of the theory are known as renormalizable theories. It is clear that only theories whose index of divergence δi ≤ 0 would be renormalizable. This is because since

ǫ

−1

688

16 Renormalization theory

3 D =4−B− F + 2

ni δi ,

i

(16.43)

if δi > 0 then when we go to higher and higher orders of interaction, we generate divergent graphs with more and more external lines. That would correspond to adding to the Lagrangian density counter terms which cannot be absorbed into a redeﬁnition of the existing parameters of the theory. Motivated by the notion of counter terms, Bogoliubov and Parasiuk developed a recursive subtraction scheme. However, one of their intermediate theorems was not true which was corrected by Hepp. Hence the method is known as the BPH method. Subsequently, Zimmermann provided a solution to their equation thereby extending it to what is known as the BPHZ method which represents the second renormalization method. The BPHZ method generalizes quite nicely to non-Abelian gauge theories and, therefore, we will discuss this in detail in section 16.4. Here we simply summarize the method brieﬂy. The BPHZ method corresponds to deﬁning forests associated with each graph in the following manner. Corresponding to each Feynman graph, if there are renormalization parts in the graph (subgraphs with superﬁcial degree of divergence non-negative), then we draw boxes around the renormalization parts in various ways such that no boxes ever overlap. A particular laying down of boxes is called a forest F . The elements or boxes in a forest F are denoted by γ. Associated with each graph there is a set of forests corresponding to all possible ways of laying down boxes around renormalization parts. A forest may be empty and it is all right to draw a box around the entire graph provided the graph is a renormalization part. As an example let us look at the complete set of forests shown in Fig. 16.9 for the two loop graph of Γ(4) in the φ4 theory. It should be emphasized that the boxes contain only the renormalization parts and not any propagator external to the renormalization part. Furthermore, the graphs are considered functions of the internal loop momenta as well as the external momenta. However, the loop momenta are not integrated yet (namely, we are working with the integrand of the amplitude). We now deﬁne a Taylor operator tγ which acts on a boxed subgraph γ and replaces it by its Taylor expansion in the external mo-

p=0 (16. We would see in detail how this works in section 16. p) p p 2! ∂pµ ∂pν tγ Γ(2) (k. F ∈Φγ ∈F (16. With the notion of the Taylor operator. We should only remember that when nested boxes exist then the tγ operation should be carried from inside out.2 A brief history of renormalization 689 Figure 16. 0) + . For example.9: Forest diagrams associated with the simple two loop vertex correction diagram in the φ4 theory. we can obtain a renormalized Feynman integrand R(G) for the graph G given by the expression R(G) = (1 − tG ) (−tγ ) I(G).4). then D(γ) = 0 and tγ simply evaluates the graph γ at zero external momentum.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 16.44) Here we have neglected the linear term in the Taylor expansion since it would vanish upon integration because of antisymmetry. p) = Γ(2) (k.16. if γ is a Γ(4) . we have tγ H = 0. then (D(γ) = 2 and k denotes the generic internal momentum) 1 µ ν ∂ 2 Γ(2) (k.4. mentum variables about zero four-momentum out to order D(γ) which denotes the superﬁcial degree of divergence of γ. Note that the BPHZ method subtracts out the divergent parts in the integrand itself so that the renormalized integral is ﬁnite. . Then the assertion is that the renormalized Feynman graph is convergent. For a convergent graph H. If γ denotes the two point function Γ(2) .

49) leads to . We recall that the generating functional for the theory in the presence of an external source is deﬁned as (see (12. δφ 2 − (16. 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.46) The Euler-Lagrange equation for the theory is given by δS g = F [φ] = ∂µ ∂ µ + m2 φ + φ2 = 0. under an arbitrary ﬁeld redeﬁnition φ → φ + δφ inside the path integral.120)) R Z[J] = eiW [J] = N Dφ ei(S[φ]+ Jφ) . (16. the generating functional would be stationary leading to R δZ = 0 = N Dφ (F [φ] − J) ei(S[φ]+ Jφ) . (16. this equation modiﬁes and the modiﬁed equation can be obtained from the generating functional for the theory as follows. 2 2 3! L= (16.49) where the Euler-Lagrange operator F is deﬁned in (16.47) This describes the dynamics of the system at the tree level. Working out in detail. 16.47).48) We note that since the generating functional does not depend on the ﬁeld variable (we are integrating over all ﬁeld conﬁgurations). (16. when we include quantum corrections. This theorem is extremely important in the study of renormalization.690 16 Renormalization theory Weinberg’s theorem.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 . However.

16.50) Here φc denotes the classical ﬁeld which is deﬁned in (13. φc − i φc − i − J = 0. 16.3 Schwinger-Dyson equation 691 eiW [J] F − i or. or. this can be written explicitly as g 2 δ δJ δ δJ ∂µ ∂ µ + m2 φc + or.78).80). Σ= ig 2 δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c . For the φ3 theory described by (16. we obtain ig δ δ2 Γ (∂µ ∂ + m )½ + 2 = − δφc 2 δφc µ 2 δ2 Γ δφ2 c −1 −1 or.46). δJ δJ δ − J = 0. Equation (16. F φc − i δJ (16.52) can be represented graphically as in Fig. c 2 2 δJ ig g ∂µ ∂ µ + m2 φc + φ2 + 2 c 2 δ2 Γ δφ2 c −1 + δΓ = 0. If we now take the functional derivative of (16.51) δφc where Γ[φc ] corresponds to the eﬀective action deﬁned in (13. (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 = 0.52) 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). δJ δW δ −i − J = 0. F δ − J eiW [J] = 0.51) with respect to φc and set φc = 0. or. (16.

To see in some detail how the BPHZ method works.53) which can be represented graphically as in Fig. let us consider the φ3 theory in six dimensions described by the Lagrangian density .53) are integral equations (they are written in a compact notation here) for the two point and the three point functions. The equations in (16.4 BPHZ renormalization We have seen in detail how multiplicative renormalization works.) Similarly. = + + Figure 16.11. 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 .11: Schwinger-Dyson equation for the three point function.51) with respect to φc and setting φc = 0. Figure 16.52) and (16. Such equations (and for the higher point functions) are known as the Schwinger-Dyson equations.692 16 Renormalization theory = .10: Schwinger-Dyson equation for the two point function. 16. (16. 16. orders. (We are using a very compact notation where integrations over intermediate coordinates/momenta are suppressed. by taking the second functional derivative of (16.

let us set M = 0 for simplicity. here the Lagrangian density is not invariant under φ → −φ. 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. (Note that unlike the φ4 theory.56) . in dimensional regularization which we will be using.) Although the zero point amplitude is divergent.4 BPHZ renormalization 693 L= 1 M2 2 g ∂µ φ∂ µ φ − φ − φ3 . we can remove this divergence by normal ordering the theory.54) This is a simple quantum ﬁeld theory that is quite helpful in understanding various features of renormalization. for this theory. the 2-point and the 3-point amplitudes are superﬁcially divergent. In six dimensions. 2 2 3! (16.16. (16. Furthermore. since we are interested in the ultraviolet behavior of the theory.55) Thus. the counting of the canonical dimension gives [φ] = 2. Furthermore. (16. 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.

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

As we have discussed earlier (see (15.59) to be k p k+p ≃− 5ǫ ig2 µǫ p2 1+ 3 6 2(4π) 6 − 2 + (γ − 1) − ln 4π ǫ 1− ǫ ln(−p2 ) 2 p . 2 2 2 2 ǫ ≃ 2 2 − + (γ − 1) . ǫ 1 2− n 2 dx 0 (x(1 − x)) 1 = 0 dx (x(1 − x))1− 2 ǫ ln x(1 − x) 2 ǫ ≃ = 0 1 0 dx x(1 − x) 1 − dx x(1 − x) (1 − ǫ ln x) 1+ 5ǫ 6 . However. to understand the structure of the self energy at one loop. 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 .76)).4 BPHZ renormalization 695 This form is quite suitable to carry out two loop calculations which we will do shortly. let us simplify this.16. (16. ǫ Γ −1 + 1 1 ǫ 1 + ln 4π . we obtain the one loop self energy of the theory in (16.60) = 1 6 Using these. 2− 6−ǫ ǫ ǫ n =2− = 2 − 3 + = −1 + .

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. but since the BPHZ method subtracts out divergences in every graph. 12(4π)3 ǫ =− (16.696 ≃ ≃ 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 .61) This leads to the one loop counter term ig2 p2 2 . 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. (16.64) .62) Similarly. i xi Ai ) × ( (16.63) 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 .

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

ǫ 4πµ2 (16. (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 .71) 1 1 dx1 0 0 du (1−x1 ) 2 (−x1 (1 − x1 )Q2 ) − ln −γ . 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. we have x3 = 1 − x1 − x2 and x3 = 1 x3 = 0 so that 1 1−x1 ⇒ x1 = x2 = 0.68) 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. we can rewrite (16.70) Furthermore.72) .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 . ǫ 4πµ2 (16. (16. ⇒ x2 = 1 − x1 .68) as ig3 µ 2 =− (4π)3 ǫ (16.

16.4 BPHZ renormalization 699 where in the new variables x1 (1 − x1 )Q2 = x1 (1 − x1 )(p + up1 )2 + u(1 − u)p2 . Let us look at the two integrals in (16.73) 1 2 Equation (16. the ﬁrst integral in (16.75) n k 2 12(4π)3 ǫ k 2 (k + p)2 (2π) 2 where we have used the form of the one loop counter term in (16. 16. = + −igµ 2 −igµ 2 ǫ ǫ 2 i i dn k i iΠ(1) (k2 ) 2 n k2 (2π) k (k + p)2 i dn k i (−ig2 k2 ) 2 i .59).74) Let us next calculate the self-energy at two loops. x2 ) ig3 µ 2 2 .75) separately. 1 (16.62).75) has the form . (16. Using the form of the one loop self-energy in (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. Let us ﬁrst look at the graphs with non-overlapping divergence of the form shown in Fig.68) leads to the one loop counter term (the factor of arises from doing the integrals over x1 . = 2(4π)3 ǫ ǫ (16.12: Non-overlapping two loop self-energy diagrams in the φ3 theory.

700 16 Renormalization theory I1 = ig2 µǫ dn k (2π)n dx ǫ Γ −1 + 2 1 1 ig2 µǫ n n 2 2 )2− 2 k 2 k 2(4π) 2 (−x(1 − x)k × = − ≃ − × = − × ≃ g 2 µǫ 2 n 1 (k + p)2 ǫ 2 2− n 2 2(4π) 2 g 2 µǫ 2(4π) dx (x(1 − x)) ǫ 1 2 6 Γ −1 + (−1) 2 −2 dn k n (2π)n (k2 )4− 2 (k + p)2 5ǫ 6 n n n 2 n 2 Γ −1 + 1+ Γ 5− n (−1) 2 −2 (1 − y)3− 2 dn k 2 dy n (2π)n Γ 4 − n Γ(1) ((k + yp)2 + y(1 − y)p2 )5− 2 2 g 2 µǫ 2 n Γ −1 + n 12(4π) 2 dy ǫ 2 1+ 5ǫ 6 n 2 Γ 5− Γ 4− n n 2 n 2 n i(−1)5− 2 Γ 5 − n − 2 n Γ 5− n (4π) 2 2 2 (−1) 2 −2 (1 − y)3− 2 (−y(1 − y)p2 )5−n 5ǫ (−p2 )n−5 6 ǫ i g2 µǫ Γ −1 + 2 12(4π)n Γ 4 − n 2 n Γ(5 − n) 1 + × ≃ − dy (1 − y)3− 2 (y(1 − y))5−n − 1 +γ−1 ǫ 2 i(g2 )2 p2 (1 + ǫ ln 4π) − + γ − 1 6 12(4π) ǫ ǫγ 2 1+ 5ǫ 6 1 − ǫ ln × 1+ ≃ − (−p2 ) 1 5ǫ 1+ 2 µ 6 4 1 + ǫ ln 4π 1+ 5ǫ 6 ig4 p2 γ 3 2 − (γ − 1) + 6 ǫ2 72(4π) ǫ ǫ × 1 − ǫ ln (−p2 ) 5ǫ + ﬁnite + µ2 4 .

76) and (16.64) as well as (15. (The y integration is related to the appropriate beta 2 function.76) Here we have used (16. let us look at the second integral in (16.60) in the intermediate steps. they will require non-local counter terms to cancel them.85). Next. + − 2− 72(4π)6 ǫ ǫ 4πµ2 12 (16.75) 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 .) 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.16. see (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 .

72(4π)6 ǫ2 ǫ 4πµ2 3 As a result.71)).77) ig4 p2 4 (−p2 ) 8 2 − ln − − + γ + ﬁnite .78) I1 + I2 = ig4 p2 2 11 + ﬁnite . − 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 (16.76) and (16. takes the form ℓ k p ℓ+p k+p p .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. Let us next look at the overlapping divergent graph in the self energy at two loops which. the sum of the two graphs in (16.75) 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). adding (16. upon using the form of the one loop vertex function in (16.

81) . (16.73) (with appropriate momenta). we can write (16. (2π)n ℓ2 (ℓ + p)2 (Q2 )3− 2 (16. x1 (16.80) n 2− n 2 × δ(1 − y1 − y2 − y3 ) D2 5− n 2 where we have identiﬁed D2 = y1 Q2 + y2 (ℓ + p)2 + y3 ℓ2 = y1 (ℓ + up)2 + = u(1 − u) 2 p + y2 (ℓ + p)2 + (1 − y1 − y2 )ℓ2 x1 2 ℓ + (uy1 + y2 )p + y1 (1 − y1 )u2 + y2 (1 − y2 ) − 2uy1 y2 + y1 u(1 − u) 2 p .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 .16.64) to combine denominators. Using (16.79) 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 .

However. The one loop pole structure is manifest in the overall multiplicative gamma function. Next. v)p2 . First we note that we can shift the variable of integration ℓ → ℓ − (uy1 + y2 )p.81) has the form D2 = ℓ2 + (1 − y1 )(y1 (u − v)2 + v(1 − v)) + = ℓ2 + d(x1 . so that the (shifted) denominator in (16.80). This is a special feature of overlapping divergent graphs and to appreciate this.704 16 Renormalization theory The divergence of the two loop integral in (16.80) 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). this is not right and.83). y1 . let us evaluate the integral in (16.82) Substituting this into (16.83) (16.71)) as y2 = (1 − y1 )v. where d is deﬁned in (16. u. in fact. To see the divergence .80) in some detail. Finally.84) . we can do the integration over y3 using the delta function in (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. the divergence of the subdiagram (one loop vertex) has been transformed to the Feynman parametric integrals. we can redeﬁne the variable y2 (as we have already done in (16. y1 u(1 − u) 2 p x1 (16.

16. β). −γ+ 6 ǫ 6 (16. Using the standard results 1 0 1 0 dt tα−1 (1 − t)β−1 = 1 du u(1 − u) = . The parametric integration in (16.84) gives the value of the overlapping self-energy diagram to be .85) the leading term in the parametric integration in (16. 6 1 Γ(α)Γ(β) = B(α.86) This demonstrates how the subdivergence is hidden in the parametric integration.84) can be evaluated exactly using Gegenbauer polynomials and the only modiﬁcation from the leading order result in (16.4 BPHZ renormalization 705 structure from the parametric integration.86) is that the ﬁnite constant 1 inside the parenthesis changes from −γ + 6 to 4. let us look at only the leading order term in (x1 d)1−ǫ . Γ(α + β) 0 1 dudv (u − v)2 = .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 . 6 (16. Thus. substituting this result into (16.

87) ig4 p2 2 (−p2 ) 2 + − γ + 3 + ﬁnite . − ln 12(4π)6 ǫ2 ǫ 4πµ2 We see again that the dangerous nonlocal divergent terms of the form (−p2 ) 1 ǫ ln 4πµ2 are present in this diagram. On the other hand.74)). (16. we also have a contribution from the graph k p k+p = (−igµ 2 ) 2 g 4 µǫ 2 4(4π)3 ǫ g 4 µǫ 2 4(4π)3 ǫ n ǫ p i dn k i ig3 µ 2 2 n k 2 2(4π)3 ǫ (k + p)2 (2π) 1 dn k dx n (2π) ((k + xp)2 + x(1 − x)p2 )2 dx Γ 2− n (−1)2 i 2 n n (4π) 2 Γ(2) (−x(1 − x)p2 )2− 2 dx x(1 − x) ǫ 2 Γ n 2 n −2 2 ǫ = − = − ig4 µǫ (−p2 ) 2 −2 2 ǫ = − Γ −1 + n 3 (4π) 2 ǫ 2 4(4π) ig4 p2 4(4π)6 2 ǫ (−p2 ) 4πµ2 1− − ǫ −2 = ≃ = Γ −1 + Γ(n − 2) 1 6 1+ 5ǫ 6 −1 2 ig4 p2 2 4(4π)6 ǫ ig4 p2 24(4π)6 ǫ (−p2 ) ln 2 4πµ2 ln 2 − + (γ − 1) ǫ 2 4 + 2 ǫ ǫ 8 (−p2 ) +γ− 2 4πµ 3 + ﬁnite.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 (16.88) . with the one loop counter term for the three point function (see (16.

(16. 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.4 BPHZ renormalization 707 Here we have used (16.89) p so that the sum of the three graphs in Fig. . 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.16.13 gives + + Figure 16.90) 8 (−p2 ) +γ− 2 4πµ 3 = ig4 p2 12(4π)6 − 2 2 + + ﬁnite . 16. BPHZ is a renormalization procedure where divergences are subtracted in the graph itself so that it works with any regularization. Similarly. = 2 2 ig4 p2 + 6 ǫ2 12(4π) ǫ − 4 2 + 2 ǫ ǫ − ln ln (−p2 ) −γ+3 4πµ2 + ﬁnite (16.85) in the intermediate step.13: Overlapping two loop self-energy diagrams in the φ3 theory.

(16. 16. for the three loop self energy graph in the φ3 theory. if it is superﬁcially divergent. Two renormalization parts γ1 and γ2 are called disjoint if γ1 ∩ γ2 = 0. 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. It may contain one particle irreducible subgraphs that are superﬁcially divergent. Fig. if none of the following holds.91) namely. .92) while γ1 and γ2 are called nested if one is contained inside the other. 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. namely.15 shows examples of diﬀerent ways of drawing boxes around the renormalization parts. (16. Thus. γ γ Figure 16. (16. two renormalization parts γ1 and γ2 are called overlapping if they share lines and vertices. namely.14. for example.708 16 Renormalization theory Consider an arbitrary Feynman graph G. A couple of examples of proper subgraphs which correspond to renormalization parts are shown in Fig. 16.93) Finally. or γ2 ⊂ γ1 . if γ1 ⊂ γ2 .14: Examples of proper renormalization parts in the two loop self-energy graph in the φ3 theory.

(16. the complete set of laying down of boxes (forests) containing disjoint renormalization parts is shown in Fig. Similarly.94) A set of such laying down of boxes around renormalization parts is called a forest. . Furthermore. for the two loop self-energy graph in the φ3 theory.4 BPHZ renormalization 709 Figure 16. Fig. 16.16. On the other hand. a forest is called full if there is a box around G. γ1 ⊂ γ2 . 16.16.15: Examples of forests in the three loop self-energy graph in the φ3 theory. A forest is called normal if there is no box around the complete graph G. γ1 ∩ γ2 = 0. let N denote the set of laying down of boxes (forests) containing only nested renormalization parts.17 shows the nested forests in the two loop self-energy graph in the φ3 theory. let us add to the set D the graph with no boxes. γ2 ⊂ γ1 . A forest is called empty if there is no box around any subdiagram. Let D denote the set of laying down of boxes which contain only disjoint renormalization parts. Thus.

it Taylor expands . we must subtract out the divergences from each of these sub-diagrams. Figure 16. for example. 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.16: The complete set of disjoint forests for the two loop self-energy graph in the φ3 theory. The reason why we draw boxes around renormalization parts is because these sub-diagrams are superﬁcially divergent and to make the full diagram ﬁnite.18 for the two loop self-energy graph in the φ3 theory.710 16 Renormalization theory Furthermore. In the conventional BPHZ method. 16. Figure 16.17: The complete set of nested forests for the two loop self-energy graph in the φ3 theory. Figure 16. shown in Fig.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.

p) 2! ∂pµ ∂pν . (2π)n 1 µ ν ∂2 p p I(k.4 BPHZ renormalization 711 it in the external momenta (external to the graph) up to terms of order D(γ). (16. we are simply able to throw . 0) + p = dn k I(k. Thus. p). tγ I (k. p1 ) . if D(γ) < 0. tγ I(k. Note that.97) (16. 0. p1 ) = I(k. p) = I(k. p.96) This is the conventional tγ operator in the BPHZ prescription and it is clear that using this operation. acting on a graph. by deﬁnition. tγ IG = 0. (2π)n D(γ) = 6 − 6 = 0.16. (in the self-energy diagram the linear term in the Taylor expansion has been set to zero because of anti-symmetry) k γ: p k+p D(γ) = 6 − 4 = 2.95) In other words. 0). the Taylor operator simply separates out the potentially divergent terms in the integrand so that the ﬁnite part of the graph can be identiﬁed with the integrand (1 − tγ ) IG = I G = ﬁnite. p. p=0 k γ: p k+p p2 k + p + p1 p1 = dn k I (k. (16.

(16. (16. in fact. if the integral has the form ∞ m=−n Iγ = am ǫm . has the general form ∞ m=−n FG = am ǫm . ∞ m=0 (16.99) so that the ﬁnite part of the graph can be identiﬁed with ∞ m=0 1−t G FG = am ǫm = F G = ﬁnite.100) Once again. (16. In particular. compatible with any regularization. am ǫm = ﬁnite parts. it never uses any particular regularization and is. (16.103) .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. the tγ operation is deﬁned to be (it separates out the pole terms) −1 m=−n t FG = G am ǫm . a graph (integral) at n-loops. this makes it clear that for a superﬁcially convergent graph (with no pole term) tG FG = 0. in the case of dimensional regularization. (16. Therefore.712 16 Renormalization theory away potentially divergent parts inside the integral.101) Given a renormalization part γ in a Feynman diagram.98) In such a case.

. In any case. the subtraction must be carried out . However. by deﬁnition tG RG IG = 0. this method is correspondingly more general.107) but this is not true otherwise. . one deﬁnes s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . The only understanding here is that for nested diagrams. (16. graph by graph this procedure can be applied and every graph can be made ﬁnite. (16.104) This. since in this method we do not explicitly use counter terms. of course. Thus. the graph G itself may be superﬁcially divergent in which case.105) We note that if G is not superﬁcially divergent.16.106) RG IG = RG IG . one needs a ﬁnal subtraction to render the diagram ﬁnite. the graph would now be ﬁnite. However. let us deﬁne s RG IG = i=1 (1 − tγi ) IG . by Weinberg’s theorem. 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 . (16. γs as subdiagrams. Then. the eﬀect of the term (−tγ Iγ ) can be thought of as that of adding counter terms which subtract the divergence. . from the deﬁnition of RG IG in (16. Thus.105).4 BPHZ renormalization 713 Thus. . and (16. γ2 . then. makes all the subdiagrams (proper renormalization parts) in G ﬁnite.

This is seen through Bogoliubov’s R-operator which deﬁnes RG IG = = 1 − tG RG IG 1 − tG Φi γk ∈Φi −tγk Rγk IG .109) where Φi is the complete set of proper renormalization parts that are non-overlapping.104) associated with the superﬁcially divergent renormalization part γk . of course.714 16 Renormalization theory inside out. for example. the order of the subtraction is irrelevant. this does not address the issue of locality of counter terms since we are subtracting out overlapping divergences which. (16. For disjoint or overlapping divergences. Here Rγk is the operator (16. 16. With this.108) is obtained as follows. That this is true will be seen shortly through examples. Thus. The BPH recursion relation (16. Equation (16. Instead of the set of disjoint proper renormalization parts (subgraphs) of a Feynman graph G.19 deﬁnes all the disjoint and nested proper renormalization parts of the three loop self-energy diagram. as we have seen. The proof of renormalization by local counter terms. can lead to non-local counter terms. Zimmermann’s solution of the BPH recursion relation (16. Although this method of making a Feynman diagram ﬁnite is absolutely correct.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 .108) as well as . This is the BPH formula. Φi γk ∈Φi (16. we can now write RG IG = = 1 − tG RG IG 1 − tG (−tγk ) IG . requires that the overlapping divergences be taken care of by lower order counter terms. Its eﬀect is to give IG itself).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. Fig. let us look at the complete set of subgraphs that include only disjoint and nested proper renormalization parts.

Therefore.108). 16.109) seem quite diﬀerent from the original relation (16.16. equivalent. γ2 } γ1 γ2 γ4 Φ6 = {γ1 . The set of disjoint proper renormalization parts for the two loop vertex correction is shown in Fig. γ4 } Figure 16.20. let us next show through examples that they are all. in fact. γ Figure 16.19: The complete set of disjoint and nested proper renormalization parts for the three loop self-energy graph in the φ3 theory.105). However. Before we proceed to see this.4 BPHZ renormalization 715 γ2 γ1 Φ0 = {0} γ3 Φ1 = {γ1 } γ4 γ1 Φ2 = {γ2 } γ2 Φ3 = {γ3 } γ3 Φ4 = {γ4 } Φ5 = {γ1 . we have . from the BPH formula (16. Zimmermann’s solution (16. γ3 } Φ7 = {γ2 .20: The complete set of disjoint proper renormalization parts of the two loop 3 point function graph in the φ3 theory. let us look at some simple examples just to get acquainted with the notions of the BPH procedure.

this is how we are supposed to renormalize the theory.110) also as RG IG = IG + (−tγ IG ) = (1 − tγ ) IG .112) This is exactly the original formula (16.113) . From the relation IG = IG/γ Iγ . Here we have overlapping divergences and following BPH (see (16.21: The complete set of disjoint proper renormalization parts of the two loop self-energy graph in the φ3 theory. (16. 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 .108)) we can write γ1 γ2 Figure 16. by deﬁnition. 16. (16.716 16 Renormalization theory RG IG = IG + IG/γ −tγ Rγ Iγ = IG + IG/γ (−tγ Iγ ) . there is no subdiagram in this case.21 containing only proper disjoint renormalization parts. (16.111) 1 − tG RG IG = 1 − tG (1 − tγ ) IG . in fact. Rγ Iγ = Iγ because there is no superﬁcially divergent proper subdiagram of γ. RG IG = (16.105) and. we can write (16. We note that. in fact. Let us next look at the two loop self-energy graphs shown in Fig.110) where IG/γ is the part of the diagram which does not contain the subdiagram Iγ .

tγ which is supposed to isolate ǫ pole terms in ǫ. (16.115) which we recognize to correspond to the counter term associated with an overlapping divergence if it did not vanish.105) RG IG = 1 − tG (1 − tγ1 ) (1 − tγ2 ) IG .4 BPHZ renormalization 717 so that RG IG = 1 − tG (1 − tγ1 − tγ2 ) IG . The diﬀerence between the two is given by 1 − tG tγ1 tγ2 IG . (16.118) 2 i (−p2 ) g4 4(4π)3 ǫ (4π)3 6 ig4 p2 2 . 2(4π)3 ǫ (16. It is such terms that complicate the proof of renormalizability.16.114) This does not exactly coincide with the original formula (16. Let us recall some of the calculations we had done earlier. 12(4π)6 ǫ2 . sets the factor of µ 2 = 1 in the graph at one loop) = so that ig3 2 = (−tγ2 Iγ2 ) .74).116) (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.

This makes the connection between (16.718 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. the ǫ1 term precisely has the same coeﬃcient and it follows. physically because it says that we do not need to subtract overlapping divergences.121) can also be written as (1 − tγi ) ≡ (−tγk ) . but more importantly. Let us recall that the three loop diagram has eight . it is best to study an example. Φi γk ∈Φi (16.87)). γi ’s are only disjoint or nested proper renormalization parts.120) This result is very important.121) involving subtractions of only non-overlapping proper renormalization parts. that 1 − tG tγ1 tγ2 IG = 0.119) Physically. To make contact with the formula of BPH. (16.122) i where Φi ’s deﬁne forests containing only disjoint or nested proper renormalization parts. this is the statement that we do not need subtractions associated with overlapping divergences. then we can always ﬁnd a renormalization part γ12 which contains both γ1 and γ2 and for which (1 − tγ12 ) tγ1 tγ2 IG = 0.109). In general. there is a theorem that says that if γ1 and γ2 denote two proper renormalization parts of G that are overlapping. The product of the factors in (16. If we recall our calculation for the two-loop overlapping divergence (see(16.105) and Zimmermann’s solution (16. (16. namely. 2 therefore. we note that using this. (16. we can now write s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . then.

16.124) Using this.19.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 .4 BPHZ renormalization 719 forests consisting of disjoint and nested proper renormalization parts which are shown in Fig. we can write Rγ1 Iγ1 Rγ2 Iγ2 = Iγ1 .123) Let us next recall that since γ1 and γ2 do not contain any proper renormalization parts. (16. we can rewrite (16. Thus.125) Φi γk ∈Φi .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 . + (tγ1 Rγ1 )(tγ2 Rγ2 ) IG = (16. 16. = Iγ2 . (16. writing out explicitly the Zimmermann solution we obtain RG IG = 1 − tG IG + IG/γ1 (−tγ1 Iγ1 ) + IG/γ2 (−tγ2 Iγ2 ) + IG/γ3 (−tγ3 Iγ3 ) + IG/γ4 (−tγ4 Iγ4 ) + IG/{γ1 .γ2 } (−tγ1 Iγ1 ) (−tγ2 Iγ2 ) + IG/γ3 −tγ3 Iγ3 /γ1 (−tγ1 Iγ1 ) + IG/γ4 −tγ4 Iγ4 /γ2 (−tγ2 Iγ2 ) .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 .

we can also remove the divergences by adding local counter terms. 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. namely. it is clear that the superﬁcial degree of divergence of a graph would increase with increasing number of interaction vertices. see (16. Such theories are called nonrenormalizable theories. of course. the BPH formula (16.127) then. (16. through Weinberg’s theorem. Such theories are known as renormalizable theories. So far. in this case.126) vanishes so that the superﬁcial degree of divergence of any graph is completely determined by the number of external lines in the Feynman diagram. 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 . However. In contrast.21)) 2F + 3 δi = di + bi + fi − 4. We would have divergence structures with more and more numbers of external lines. This analysis. This is.20) and (16. δi < 0. In this case. if the index of divergence were negative.720 16 Renormalization theory where Φi is the complete set of proper renormalization parts consisting of only disjoint graphs.128) not only would we have a ﬁnite number of types of graphs that can be divergent. Of course. 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). therefore. (16. shows how the BPH formula leads to Zimmermann’s solution and how. 2 (16. On the other hand if δi > 0. we have talked about theories where the index of divergence of interactions in four dimensions (recall that D = 4 − B − 3 i ni δi .108).

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

µ 1 ˜2 ca(u) = Z3 ca . . of course.p ν. ˜ Z3 = 1 + D. the longitudinal part of the photon self-energy is not renormalized by quantum corrections. Namely. in fact.a ν µ ν 2 − Z4 g2 abc apq b c µ. (16.a 4 2 ˜ ˜ (16. if we redeﬁne the ﬁelds and parameters as a(u) 2 Aµ = Z3 Aa .91)). But the main question is whether renormalization would preserve the gauge invariance – actually the BRST invariance – present in the original theory. all the divergences can be taken care of.a ) + Z1 gf abc Ab Ac ∂ µ Aν. ˜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. g(u) = Z1 Z3 2 g.722 16 Renormalization theory +(1 + D)∂µ ca ∂ µ ca − (1 + E)gf abc (∂ µ ca )Ab cc µ = − Z3 ∂µ Aa (∂ µ Aν. 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.130) + Z3 ∂µ ca ∂ µ ca − Z1 gf abc ∂ µ ca Ab cc . This is. With these counter terms. ˜ Z1 = 1 + E.a − ∂ ν Aµ. µ where we have identiﬁed Z3 = 1 + A. −1 1 F a(u) = Z3 2 F a . Z4 = 1 + C. −3 1 1 g2 ˜ − ˜ −1 = Z1 Z3 2 Z3 g. g1 (u)2 (u) −2 = Z4 Z3 g2 . Z1 = 1 + B.

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. namely. for example. the counterterms will satisfy (16.135) which would lead to g(u) 2 = g1 (u) 2 = g2 (u) 2 . in turn require that the corresponding counter terms have to be related. therefore.5 Renormalization of gauge theories 723 (16.130) as . g(u) = g1 (u) (u) = g2 . (u) 2 (u) 2 g2 (16. (16. In this case. 2 −3 ˜ 2 −1 ˜ −2 = Z1 Z3 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 = = .132) ξ (u) = Z3 ξ.133) This would. (16.134) and remove all the divergences from the theory.136) This is. the same relation as in (16. The Slavnov-Taylor identities following from the BRST invariance (see. But the real question is how can the renormalization process guarantee this. of course. ˜3 Z3 Z1 Z (16. we will have 2 g(u) g1 2 −3 = Z1 Z3 g2 .16.133) and with this identiﬁcation we can write (16. if we choose a regularization scheme which respects the BRST symmetry. −2 2 −3 = Z4 Z3 g2 = Z1 Z3 g2 .134) Therefore. (13.

(16. 16.a(u) L = − Fµν F µν a(u) + 4 2 +∂ µ ca(u) Dµ ca(u) . k .46)) 1 a L = − Fµν F µν. the choice of a regularization which preserves this symmetry is quite crucial. 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).139) . a + iψ k γ µ Dµ ψk . As we have seen. n2 − 1 and k = 1. in addition to having the inﬁnitesimal local gauge invariance δAa = Dµ ǫa (x). (16. The BRST invariance of this Lagrangian density is now manifest (with unrenormalized ﬁelds and parameters). µ δψ k = iǫa (x) ψT a δψk = −iǫa (x) (T a ψ)k . It is essential in identifying the physical states of the theory. in establishing unitarity as well as in showing the gauge independence of the physical matrix elements. 2. Otherwise. in studying nonAbelian gauge theories it is natural to discuss renormalizability using dimensional regularization.137) where the covariant derivative is deﬁned with g(u) .724 16 Renormalization theory ξ (u) a(u) a(u) 1 a(u) F F + (∂µ F a(u) )Aµ. n.138) where a = 1.6 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. We note that this theory. · · · . It is for these reasons that in the case of gauge theories. · · · . consequently. dimensional regularization is one of the regularization schemes (and it is quite simple) which respects gauge invariance and BRST symmetry and. 2. the regularization scheme may introduce unwanted spurious eﬀects into the theory. 4 (16.

(16. there are two conserved currents. Let us note that.a = ψ j γ µ (T a )jk ψk . Dµ J µ.140).6 Anomalous Ward identity 725 is also invariant under the global transformation δψk = −iλγ5 ψk . (16.142) then this would not be invariant under the chiral symmetry transformations (16.a = 0. µ J5 = ψ k γ5 γ µ ψk . δψ k = −iλψ k γ5 . we can still deﬁne the axial vector current as in (16. will not be conserved. J µ. i.141) which. is covariantly conserved.143) is without the parameter of chiral transformation.140) describes the chiral symmetry transformations under which the massless theory is invariant. (16. µ ∂µ J5 = 0.e. rather it would satisfy the equation µ ∂µ J5 = −δLm = −2imψ k γ5 ψk . if the fermions are massive..141) transforms covariantly under a gauge transformation and. As is the case with continuous symmetries. (16.141) We note here that the vector current in (16. if there is a term in the Lagrangian density of the form Lm = −mψ k ψk . therefore.16. We can write down in a straightforward manner the Ward identities associated with these conserved currents which are simply expressions of the conservation of currents. Nonetheless. the invariance of the theory leads to a conserved current. We can also derive identities from this relation that would relate various matrix elements of the .143) where the variation of the Lagrangian density in (16. however. In this theory.140) Here λ is the space time independent real inﬁnitesimal parameter of transformation and (16. therefore.

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

but the ones that are directly of interest to us are γ5 γ µ = ǫµν γν . in two dimensions we have a second rank anti-symmetric tensor (Levi-Civita tensor) given by ǫµν = −ǫνµ . −).147) Furthermore. (16. note that in addition to the diagonal metric tensor η µν = (+.16. with ǫ01 = 1. (16. 16. 2 2 (k + p)2 (2π) k µν I5 (p) = −e2 (16.22 and this is given by / / / d2 k Tr γ5 γ µ kγ ν (k + p) . (16.146) .151) Let us now calculate the one loop amplitude involving a photon and an axial vector current as shown in Fig. γ µ γ ν = η µν + γ5 ǫµν .149) As a result of these identities. (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. µ ∂µ J5 = 0. in 1 + 1 dimensions the vector and the axial vector currents are related as µ J5 = ψγ5 γ µ ψ = ǫµν ψγν ψ = ǫµν Jν .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 = . (16.148) There are various simple identities in 1 + 1 dimensions involving the Dirac matrices.152) .

22: The amplitude with an axial vector current and a photon. Furthermore. Γ(2) (−x(1 − x)p2 )2− 2 (16.152) using dimensional regularization which we know would preserve gauge invariance.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 (15.728 16 Renormalization theory γ5 γ µ k γν p k+p Figure 16.153) We can evaluate the integral in (16.152) corresponds to the Fourier transform of the matrix element µ 0|T J5 (x)Aν (0) |0 . 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 . (16.71) and (15.36) in the limit m = 0. Thus. we note that the amplitude in (16. since . Here we have used the propagators for the fermions following from (15. Furthermore.73).

On the other hand.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 ). η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 (16.154) 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. (16. π (16.154) 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 . the trace over the Dirac matrices can be carried out in two dimensions using the identities (16. (16.155) where we have used the gamma matrix identity in n dimension in (15. using this relation. (16. It is clear now that the factor (2 − n) = ǫ makes even the ﬁrst term ﬁnite and.86). However. We note from (16.157) is quite interesting for it leads to µν pµ I5 (p) = − pµ pλ pρ ie2 λρ νµ η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 ie2 νµ ǫ pµ .149) and the ﬁnal result is obtained to be (the x integration is trivial) µν I5 (p) = − ie2 λρ νµ pλ pρ . in fact.6 Anomalous Ward identity 729 γ5 is a two dimensional quantity we have not yet evaluated the trace over the Dirac matrices.158) = − .156) Since each of the terms is now completely ﬁnite. the ﬁrst term is divergent and we have to be careful in evaluating the trace.

(16. the particle multiplets are carefully chosen such that the anomalous contributions coming from various multiplets cancel completely making the gauge current anomaly free. In fact.730 16 Renormalization theory Recalling that this amplitude is the Fourier transform of the matrix element in (16. µ ∂µ J5 (x) = γλ γ5 γ µ γν Figure 16. On the other hand. since gauge invariance is so vital to the physical theory. this leads to the operator relation e2 e2 ǫµν ∂ µ Aν = ǫµν F µν .153). Although our discussion here has been within the context of the two dimensional QED (Schwinger model). it is possible to show that there is no regularization which will simultaneously preserve gauge invariance and chiral invariance. The origin of this result can be traced to the fact that the dimensional regularization which we have used in our calculation violates chiral symmetry although it preserves gauge invariance. the same behavior is ob- . quantum corrections have made the conservation law anomalous and the anomaly in the divergence of the current is given by (16. This is very important in building models of particle interactions.159) π 2π This shows that although the axial vector current is divergence free (conserved) at the tree level. In some gauge theories.23: The triangle diagram in four dimensional QED contributing to the anomaly.159). As a result. one of the two currents (or a linear combination of them) would always become anomalous because of quantum corrections. we always choose a regularization scheme which would preserve gauge invariance in the quantum theory so that the chiral current becomes anomalous.

In grand uniﬁed theories where we treat both quarks and leptons as massless. Furthermore. they have precisely this kind of coupling. However.159) that a similar statement can be made in the case of the Schwinger model as well. 8π 2 (16. it is possible to deﬁne a modiﬁed axial vector current as e2 µ ˜µ J5 = J5 − 2 ǫµνλρ Aν Fλρ .162) which would satisfy a nonanomalous divergence.23 leads to an anomaly in the chiral current. Therefore.160) µ ∂µ J5 = −2imψγ5 ψ + = −2imψγ5 ψ + where the dual ﬁeld strength tensor is deﬁned as 1 ˜ ǫµνλρ F λρ . 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 . For example.160) corresponds to the anomaly in four dimensional QED. We note from (16. in the four dimensional QED a graph of the form shown in Fig. 16.16. Fµν = 2 (16. We note here that. if we have a theory where there are both vector and axial vector gauge couplings of the fermions. As we have discussed in the last chapter. the model cannot be renormalized unless the anomaly is cancelled. the correct pion life time can be calculated only if the chiral anomaly is taken into account. as is clear from its structure. For example. 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. such a current is not gauge invariant (although the associated charge is gauge invariant). in this case. In string theories similar anomaly free considerations ﬁx the .6 Anomalous Ward identity 731 tained in any dimension. anomalies have observable eﬀect. 8π (16.161) and the second term in (16.

14. Erice (1976).732 16 Renormalization theory gauge group uniquely to be SO(32) or E(8) × E(8). E. Physical Review 184. 486 (1949). 84. Jackiw. Cambridge University Press (Cambridge. Hepp. C. J. 12. Physical Review 177. 1985). Physical Review 128. J. 1736 (1949). 4. W. Singapore (1987). 426 (1951). Salam. 13. W. 2426 (1969). Physical Review 118. Velo and A. Adler. S. 8. Wightman. N. 301 (1966). Foundations of Quantum Chromodynamics. Bardeen. S. J. Nuovo Cimento 60A. Muta. Coleman. F. A. World Scientiﬁc. 2. 10. Physical Review 75. 227 (1957). 54 (1951). 392 (1982). S. 5. Communications in Mathematical Physics 2. 7. 11. Proceedings of the International School of Mathematical Physics “Ettore Majorana”. Ward. A. Aspects of symmetry. Caswell and A. 6. Wightman in “Renormalization Theory”. N. 3. T. 2425 (1962). Acta Mathematica 97. 1848 (1969). 9.7 References 1. Dyson. these topics are beyond the scope of these lectures. 75. Bogoliubov and O. J. Schwinger. Weinberg. S. 217 (1951). Proceedings of Royal Society A64. 47 (1969). S. Bell and R. ed G. 16. Parasiuk. Kennedy. . 838 (1960). K. Physical Review 82. D. Physical Review 25. S. However.

Gell-Mann and Low were interested in studying how the electric force and the potential behave at extremely short distances.Chapter 17 Renormalization group and equation 17. We have seen earlier at the end of section 8.1.2) .1) that at the lowest order the potential is independent of the mass of the fermion and it scales with distance as 1 . namely. ep denote the ﬁne strucp c ture constant and the physical electric charge of the fermion respectively. 17. (Physical electric charge is determined from the scattering of on-shell electrons. We can. of course. −→ V (r) = e2 αp .1) 1 where in CGS units α2 = p ≃ 137 and αp . In the same manner.1 Gell-Mann-Low equation In the early 1950s. the Coulomb potential can also be obtained from the Born amplitude for the scattering of two electrons exchanging a photon. 1+ r 3π rme 733 (17. calculate quantum corrections to r this potential and if we add the one loop correction to the potential shown in Fig. r (17. then the potential is obtained to have the form V (r) = 2αp αp 1 ln + ··· .10 that the Yukawa potential can be related to the Born amplitude for the scattering of two electrons exchanging a (pseudo) scalar meson.) It is clear from (17.

it also behaves as V (r) → α r 1− Λ 2α ln 3π me = αp . in this limit. 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. we see from (17. Furthermore. (17.4) where we have identiﬁed Λ 2α ln αp = α 1 − 3π me . 1 + r 2 m2 e V (r) = (17. 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 + ··· . namely r ≫ me ≫ Λ . V (r) → 1 as r → 0 (at exr tremely short distances) as naive scaling would imply.5) This relation can be inverted to express the renormalized parameter in terms of the physical parameter as . To understand the origin of this behavior.2) that the potential no longer scales as 1 and furthermore. For example. 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 → 0.1: The one loop correction to the electromagnetic potential.734 17 Renormalization group and equation Figure 17. In fact. It is clear that the one loop corrected potential has a very diﬀerent character from the lowest order potential. it r depends on the mass of the electron in such a way that we cannot take the limit me → 0. When r is 1 1 large.

this derivation clariﬁes that the singularity in the potential at me → 0 arises purely because of the renormalization of charge at large distances.8) Furthermore.2) and also shows that the physical ﬁne structure constant can be identiﬁed with αp = rV (r) .9) then. (17. in the limit Λ → ∞. Let us next investigate what would happen if we deﬁne the electric charge at some arbitrary (intermediate) distance R and not as in (17. 1 r= m e (17. We also see that the naive scaling of the potential breaks down and mass dependence comes in due to renormalization eﬀects. if we deﬁne αR = RV (R).3) in terms of this (physical) parameter.1 Gell-Mann-Low equation 735 α = αp 1 + Λ 2αp ln 3π me + ··· . For example.17. 1+ r 3π rme This expression coincides with (17. (17.8).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 + ··· .6) Expressing the potential (17. 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. on dimensional grounds we see that the potential can be expressed as .

3) as follows.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. R .10) where F is a dimensionless function (function of the independent r r dimensionless variables αR . Rme . .11) Inverting this relation we obtain α = αR 1 + 2αR ln 3π 1 + R2 Λ2 + ··· . We note that αR = RV (R) = α 1 − 2α ln 3π 1 + R2 Λ2 + ··· .736 17 Renormalization group and equation r 1 V (r) = F αR .3) leads. which can be seen from (17. R ≪ me . 1 + R 2 m2 e (17. r R (17.12) Putting this back into the potential (17.) This function should 1 not be singular as me → 0 which implies that for r. (We note that rme = Rme × R so that it is not an independent quantity. the dependence of the potential on me can be neglected. 1 + R 2 m2 e (17. in the limit Λ → ∞ and rme . to V (r) = αR r 1+ 2αR ln 3π 2αR ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· 1 + r 2 m2 e 1 + R2 Λ2 1 + r 2 Λ2 . 1 + r 2 m2 e + ··· 1 + R 2 m2 e (17. Let us also note that if we deﬁne the electric charge (ﬁne structure constant) as . Rme ). Rme ≪ 1 (the second of these limits corresponds to me → 0).

16)) we see that the equation governing this change is given by . 1 + r 2 Λ2 1 + r 2 m2 e + ··· (17. = F αR . r R r r .13) (see also the deﬁnition (17. R ≪ 2αR ln 3π leads to the relation αr = αR 1 + = αR 1 + 1 + R2 Λ2 − ln 1 + R 2 m2 e .1 Gell-Mann-Low equation 737 αr = rV (r).18) (see also (17. 1 + r 2 m2 e 1 me .10)) αr r 1 = F αR .15) which.17.18) This shows that the charge (ﬁne structure constant) changes (runs) with distance and from the second of the equations in (17.16) 2αR R ln + · · · 3π r and this is compatible with (17. in general. = αr 1 + (17.14)) αr .12) that we can identify (17.14) α = αR 1 + 2αR ln 3π 2αr ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· . . write (see (17. then we see from (17. in the limit Λ → ∞ and r. r V (r) = (17.17) From this analysis we see that in the limit of me → 0 we can. R V (r) = αr (17.

(17. R (17. R → 0 with R = x ﬁxed. αr= = αp .23) 1. x) ∂x (17. namely. for example.22) There are two possibilities for the solution of the equation (17.21) 1 me We see explicitly from (17.18) leads to (α0 = αr=0 ) α0 = F (α0 . In this case. As R → 0. αR does have a limit which is nonzero and is inde1 pendent of the value 137 which arises only as an initial condir tion.19) β(αr ) = − . ∂F (αr . dαr 2α2 2α2 = − R = − r. the bare charge becomes inﬁnity. dr 3π 3π r (17.16) that. x=1 (17. the theory develops ghosts and other problems. As R → 0. we note that αp and me arise only as initial conditions in this evolution.18) αr = F αR . αR does not exist and consequently. Gell-Mann-Low equation basically relates the electric charge at one distance (scale) to another (namely. 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. (17. x) .20) and this is known as the Gell-Mann-Low equation. We see from this discussion that through proper renormalization the dependence of the potential on the electron mass me disappears. in the present case. then (17. In this case. if we let r.738 17 Renormalization group and equation r where dαr dr = −β(αr ). 2.24) . r .

It is important to note that in a quantum ﬁeld theory scale invariance is broken by the masses of particles. This introduces an additional arbitrariness into the renormalized parameters depending on the arbitrariness in the mass scale used. in the MS scheme we not only subtract out the pole parts (in ǫ) of an amplitude. (In the example of QED that we discussed in the last section. Even in dimensional regularization. Even within a given regularization. In QED.5.17. the MS scheme (minimal subtraction) corresponds to subtracting out only the pole parts (in ǫ) of an amplitude. therefore. such as the Euler’s constant etc. we deﬁned the charge at some length .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.2 Renormalization group 739 This is known as a ﬁxed point of the evolution equation (17. as r. 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. The only remaining breaking of scale invariance is due to renormalization itself.19) (also where the beta function vanishes). For example. therefore. The renormalized parameters. 17. We can keep track of this by using a running coupling constant. it corresponds to the second possibility. First. On the other hand. in dimensional regularization. There are two possible sources for this arbitrariness. for example. a particular renormalization scheme may be preferrable if it makes the higher order contributions in perturbation theory smaller. are diﬀerent in the two renormalization schemes even though the regularization is the same (dimensional regularization). the eﬀects of masses are negligible at high energies if we renormalize the theory in an appropriate way. 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). but some constants as well.16) (remember / r that R is ﬁxed as the limit is taken) and. We will discuss these ideas in more detail in section 17. R → 0. we saw that the coupling constants are deﬁned with an arbitrary mass scale µ. However. α0 → 0 which is obvious from (17.

Namely. λ. −1 m2 = Zm Z3 m2 . they must also have the same value independent of which set of parameters are used to calculate them. ′ ′ m2 = Zm Z3−1 m′ 2 . we must have (S and S ′ denote representative S-matrix elements calculated in the two prescriptions) S ′ p. On the other hand. Calculations with either of these sets of parameters will lead to ﬁnite quantities.) Let us consider a renormalizable theory.32)) 1 2 φu = Z3 φ. they should not only be ﬁnite.27) so that if we expand the S matrix elements in a perturbation series . see the discussion after (16. λ′ .740 17 Renormalization group and equation scale which illustrates the source of arbitrariness of the second kind. m′ . 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. (17. ′ 1 (17. µ). If we are calculating physical quantities such as the scattering amplitudes. (17.25) deﬁne a set of renormalized parameters.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. u −2 λu = Z1 Z3 λ. µ′ = S(p. m. u ′ ′ λu = Z1 Z3−2 λ′ . a different renormalization scale prescription would lead to φu = Z3 2 φ′ . Namely.

in a theory like QCD (where the value of the coupling constant is not small). u (17. µ φ.µ ′ −1 2 −1 φ′ = Z3 1 −2 λ′ .17. µ′ = (17.25) and (17. all we can say and check is that S ′ p. we can only calculate up to a given order.29) 1 In a theory. in perturbation theory. n n S ′ p. where α ≃ 137 . Here we have identiﬁed z µ .2 Renormalization group 741 S(p. (17. (17. we can check this. µ) = n an λn . m. λ′ .25) and (17.32) Similarly. µ′ φu .26) so that we can write the renormalized ﬁelds in the two prescriptions as φ = Z3 2 (λ. φ′ = z µ′ . On the other hand.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.26) we can also write −1 m2 = Zm (µ)Z3 (µ)m2 . m′ . µ′ − S(p. We can invert the relations in (17. u −1 m′ 2 = Zm (µ′ )Z3 (µ′ )m2 . the higher order corrections become negligible very quickly. λ′ . m′ . such as QED. say the nth order. λ.31) = Z3 (µ′ ) Z3 (µ) . m.33) . from (17. If we can calculate the exact series. a′ λ′ n . In such a case. Let us next see that this arbitrariness in the renormalized parameters is due to a ﬁnite renormalization. (17. µ)φu . the residual terms can be non-negligible. However. λ.30) which leads to a relation between the two (17. µ) = O λn+1 .

38) we note that . ′ ′ λ′ = Z1−1 Z3 2 λu . zλ ) are completely ﬁnite.36) which leads to the relation λ′ = zλ λ. Let us next show that these ﬁnite renormalizations (ﬁnite transformations) deﬁne a group. µ) = Z1 (µ′ ) Z1 (µ) −1 Z3 (µ′ ) Z3 (µ) 2 . Z1 ) and (Z3 . with (17. (17. In fact. µ)m2 .742 which implies that 17 Renormalization group and equation m′ 2 = zm (µ′ . (17. 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).34) zm (µ .25) and (17. it follows that (z. (17.35) Finally. Z1 ) diﬀer only in ﬁnite parts (the divergent parts are the same since the “µ” dependence comes only in the ﬁnite parts). zm .37) zλ (µ′ . Zm . µ) = ′ Zm (µ′ ) Zm (µ) −1 Z3 (µ′ ) Z3 (µ) . (17. Thus.38) ′ ′ ′ Since (Z3 . from the deﬁnitions in (17. (17. with (17. Zm .35) and (17.32).26) imply −1 2 λ = Z1 Z3 λu .

µ). µ . µ).39) Namely. Zm (µ′′ ) Zm (µ′ ) −1 −1 zm µ′′ .µ z µ . z(µ. (17. µ′ zλ µ′ . physical quantities cannot depend on the arbitrary scale introduced to carry out perturbation calculations in a theory. µ′ ) = z −1 (µ′ . Namely.17. −1 zm (µ.2 Renormalization group 743 z µ . 2 Z3 (µ) (17.µ = = Z3 (µ′′ ) Z3 (µ) −1 2 ′′ ′ ′ Z3 (µ′′ ) Z3 (µ′ ) −1 2 Z3 (µ′ ) Z3 (µ) −1 2 = z µ′′ . therefore. µ) = 1 = zm (µ. deﬁne an Abelian group known as the renormalization group and all physical quantities must be invariant under the renormalization group transformations. . µ). µ . two successive ﬁnite renormalizations lead to an equivalent single ﬁnite renormalization. µ . µ = × = Z1 (µ′′ ) Z1 (µ) Z1 (µ′ ) Z1 (µ) −1 2 Z3 (µ′ ) 2 Z3 (µ) 2 Z3 (µ′′ ) = zλ µ′′ . −1 zλ (µ. µ). µ = × = Zm (µ′ ) Zm (µ) −1 Z3 (µ′′ ) Z3 (µ′ ) Z3 (µ′ ) Z3 (µ) = zm µ′′ . µ′ ) = zm (µ′ .40) The set of ﬁnite renormalizations. These transformations further satisfy z(µ. µ) = zλ (µ. µ′ ) = zλ (µ′ . −1 2 Z3 (µ′′ ) 2 Z3 (µ′ ) Zm (µ′′ ) Zm (µ) Z3 (µ′′ ) Z3 (µ) Z1 (µ′′ ) Z1 (µ′ ) −1 zλ µ′′ . µ′ zm µ′ .

φu (xn )) = Z3 G(n) (x1 . u n n (17. . λ(µ′ ).3 Renormalization group equation To understand the consequences of the renormalization group of transformations described in the last section. λu ) . m(µ). . . pn ) . The connected renormalized Green’s functions of the theory are deﬁned as G(n) (x1 . . . . . λ(µ). pn ) = Z3 2 G(n) (p1 . µ) . . . xn ) = T (φ (x1 ) . . . u −n 2 (17. m(µ′ ). .45) . . . . p2 . µ)Γ(n) (p.41) Such a relation between the renormalized and the unrenormalized Green’s functions of the theory holds true in momentum space as well. . u −n (17. . n n (17. mu .44) which implies that the renormalized 1PI functions in two renormalization prescriptions are related as Γ(n) p. mu . . pn ) = Z32 Γ(n) (p1 . xn ) . let us next look at the Green’s functions of the scalar ﬁeld theory. m(µ). φ (xn )) −n 2 c c = Z3 T (φu (x1 ) . λ(µ). G(n) (p1 . . . λu ) u = z 2 (µ′ . µ′ = Z32 (µ′ ) Γ(n) (p. u Written out explicitly.42) where only (n−1) of the external momenta are independent (because of overall energy-momentum conservation). . namely.43) (17. The 1PI vertex functions are obtained from the connected Green’s functions by removing the external legs or the external propagators (two point function scales as Z3 ) so that we have Γ(n) (p1 . . pn ) . this has the form Γ(n) (p. . . . µ) = Z32 (µ)Γ(n) (p.744 17 Renormalization group and equation 17.

As we have noted this deﬁnes the Abelian group of transformations known as the renormalization group.19)) µ dλ dµ = β(λ). Thus. 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.48) that β(λ) = − ǫ + µ dZλ Zλ dµ λ.39)) −1 −1 2 µǫ λ = Z1 (µ)Z3 (µ)µǫ λu = µǫ Zλ (µ)λu . Here µ0 is assumed to be ﬁxed and µ is the variable renormalization scale.46) while the unrenormalized coupling can be deﬁned as (in discussions in earlier chapters we had used µ0 = 1) µǫ λu . 0 0 or.32). λu = µ µ0 ǫ λZλ (µ). Let us deﬁne the renormalized quartic coupling in the scalar ﬁeld theory as µǫ λ.17. Since λu is µ-independent (independent of the renormalization prescription). the couplings are deﬁned with an arbitrary mass scale. Here we have used p to denote collectively all the independent external momenta of the 1PI vertex function. In dimensional regularization. (17.49) where we see from (17.3 Renormalization group equation 745 which follows from the deﬁnition of z in (17. (17. 0 (17. it follows that (see also (17.48) −2 where we have identiﬁed Zλ = Z1 Z3 . (17. (17.50) .47) Both λu and λ are dimensionless and are related as (compare with (16.

from −1 m2 = Zm (µ)Z3 (µ)m2 = Z m (µ)m2 . . m. . λ4 + · · · . (17. c. in fact. Similarly. . b.49) decouples from the mass equation (17. 2Z m dµ (17. µ and.55) where the constants a. q.746 17 Renormalization group and equation ∂Z ∂µ and the derivative on the right-hand side should be thought of as with λu .51) we obtain dm dµ µ where = −mγm . s. mu held ﬁxed. (17. γm = γm (λ). This is because in these schemes the renormalization constants can be expressed in perturbative series of the forms a 2 λ + ǫ q 2 λ + ǫ b c + 2 ǫ ǫ s t + 2 ǫ ǫ Zλ = 1 + Zm = 1 + λ4 + · · · . They are. their dependences are of the form (the µ dependence comes in only through the coupling) β = β(λ). t. the equation for the coupling constant (17.52) and can be solved independently which we will analyze in the next section. are independent of µ because the pole terms (in ǫ) are independent of µ as there are no overlapping divergences in a renormalizable theory. . in general. u u −1 (17. By dimensional reasoning then they cannot depend on m either. As a result.53) Both β and γm are ﬁnite functions of µ as ǫ → 0 since the divergences in the numerator and denominator cancel out.52) γm = (17. µ dZ m . We note that this is a consequence of the fact that dimensional regularization is a mass independent regularization. functions of λ.54) in the MS (MS) scheme.

µ) = 0. λ(µ). µ or. γ = γ(λ).56) which leads to the relation (since the unrenormalized quantities are independent of the renormalization prescription) dΓu (p. 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). see also (17. ∂µ µ ∂m . u −n (17.52)) ∂λ . λ(µ). .57) is known as the renormalization group equation for the 1PI vertex functions and expresses how amplitudes change when the renormalization scale is changed.44) Γ(n) (p.58) β(λ) = µ γm = − γ = µ ∂Z3 . or.3 Renormalization group equation 747 Let us next look at the 1PI amplitudes (17. mu ﬁxed. λu ) = 0. dµ d −n Z3 2 (µ) Γ(n) (p. mu . ∂µ ∂λ ∂m (17. µ) . mu .17. namely. where we have deﬁned (the derivatives are taken with λu . m(µ).59) Equation (17. ∂µ ∂µ ∂λ(µ) ∂µ ∂m ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) = 0. λu ) = Z3 2 (µ) Γ(n) (p.57) (n) µ or. (17. dµ µ µ ∂ ∂λ(µ) ∂ ∂m(µ) ∂ +µ +µ − nγ Γ(n) = 0.49) and (17. m ∂µ (17. m(µ).

57) describes how renormalized amplitudes (1PI vertex functions) change as the renormalization scale is changed ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) (pi .62) Let us note from our earlier discussions (see (16.61) β(λ) = µ γm = − γ = µ ∂Z3 .64) .748 17 Renormalization group and equation 17. λ. let us set m = 0 in which case.4 Solving the renormalization group equation As we have seen. ∂µ ∂λ µ (17. µ) = 0. 2Z3 ∂µ For simplicity.63) Consequently. the equation takes the form ∂ ∂ + β(λ) − nγ Γ(n) (pi . (17.60) µ where pi denotes the independent external momenta and as we have seen in (17. µ) = 0.24)) 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.58) (the derivatives are taken with λu . 1 2 n (17. ∂µ ∂λ ∂m (17. m. ∂µ µ ∂m . m ∂µ (17. mu ﬁxed) ∂λ . if we deﬁne the Mandelstam variable associated with the n-point function as s = p2 + p2 + · · · + p2 . the renormalization group equation (17. λ.

changing the renormalization scale merely corresponds to changing t. λ. µ2 s . a change in t can be thought of as changing µ or for a ﬁxed µ. µ) = s 4−n 2 √ s.17. consequently. ∂t ∂x (17.70) . ∂t ∂λ and Γ(n) = s 4−n 2 (17. λ. It follows now that ∂t ∂ ∂ ∂ =µ = . we can write f (n) pi · pj s .68) f (n) t.68) now becomes straightforward from the observation that a linear ﬁrst order diﬀerential equation of the form ∂ρ(x.4 Solving the renormalization group equation 749 then. scaling all momenta by a factor of Γ(n) (pi . t) + v(x) = L(x)ρ(x.66) so that for a ﬁxed s. it can be thought of as scaling the momentum) µ= √ set . t = − 1 s ln 2 .69) Solving the renormalization group equation (17. pi · pj .65) where we have used the fact that the n-point vertex function of a scalar ﬁeld theory is Lorentz invariant and. t).67) ∂t ∂ s 1 (where from (17. ∂µ ∂µ ∂t ∂t µ (17. s (17. λ. 2 µ (17.66) we have used ∂µ = − 1 ∂µ ln µ2 = µ ) so that the 2 renormalization group equation (17. or. t) ∂ρ(x. (17.62) can be written as ∂ ∂ + β(λ) − nγ Γ(n) = 0. Let us relate the renormalization scale µ to a new dimensionless variable t as (this has the advantage that for ﬁxed s. f (n) is a dimensionless function of independent dimensionless variables which are Lorentz invariant.

74) Translating t by −t.t′ )) . nγ ↔ L. (17. (17.t′ )) . t)) = ρ (x(x. −t)) e or. Therefore. dt with the boundary condition x(x.70) clearly depends on determining its characteristics x(x. where we have used (17.75) dt′ L(x(x. of course. we see that solving the equation (17.68) with the hydrodynamic problem (17. We can.70) can be written as dρ(x) = L(x)ρ(x). are deﬁned by the equation dx(x. β(λ) ↔ v(x).70) as t ↔ t.71).t′ )) . ρ(x) = ρ (x(x. The characteristics. dt which can be integrated to give ρ (x(x. in this case.73) dt′ L(x(x.72) (17. −t)) e R0 −t R0 −t dt′ L(x(x. t) = v(x).71) Using (17. we then obtain ρ (x(x. t). λ ↔ x. t = 0)) e Rt 0 (17. we note that the diﬀerential equation (17. 0) = x. (17. Γ(n) ↔ ρ. directly identify the quantities in the renormalization group equation (17.76) . (17.72). 0)) = ρ (x(x.750 17 Renormalization group and equation is of hydrodynamic type and is best solved by the method of characteristics.

This is known as Weinberg’s theorem (the same theorem which also made its appearance in renormalization theory). (17. 1 2 s ln µ2 ) pi → αpi .79) so that (remember that −t = t → t − ln α.80) then. Let us next note that if λ∗ is a ﬁxed point of the coupling constant evolution equation (17. )e s (17. (17.77) and once we know the solution to this equation. −t).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.77).4 Solving the renormalization group equation 751 so that the solution of the renormalization group equation (17. dt (17. if .17. for large α (large momentum). namely.t′ −ln α)) → α4−n (ln α)a · · · = (n) (α)D(Γ ) (ln α)a · · · .68) requires solving the characteristic equation dλ = β(λ).78) Note that if we scale all momenta (keeping µ ﬁxed) as s → α2 s. (17. the leading behavior of the amplitude is given by 4−n 2 Γ(n) → α2 s f (n) (λ(λ. we can obtain Γ(n) = s 4−n 2 f (n) (λ(λ. × en R ln α −t+ln α pi · pj ) s dt′ γ(λ(λ.t′ )) . −t + ln α). 0 pi · pj n R−t dt′ γ(λ(λ.

if λ∗ = 0.24)). = s 4−n 2 s µ2 − n γ(λ∗ ) 2 f (n) λ∗ . (17.87) (17.84) As a result.79). pi · pj . near the ﬁxed point of the renormalization group equation. then γ(λ∗ ) = 0. λ = λ∗ = constant.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 . as Γ(n) ≃ α4−n(1+γ(λ∗ )) Γ(n) . the n-point amplitude will scale. at the ﬁxed point.82) (17. 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).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 (16.752 17 Renormalization group and equation β (λ∗ ) = 0.86) which follows from the fact that in perturbation theory γ(λ) is a power series in λ.83) Near the ﬁxed point. (17. (17. the n-point amplitude (17. under a scaling pi → αpi (17. = s 2 e 2 ∗ s 4−n 2 f (n) λ∗ . namely. . If the ﬁxed point of the beta function happens to be at the origin. then. s (17.

and λ > 0).92) On the other hand.90) λ = 0.3 and in this case.2. the behavior of the coupling constant is given by ∞. t → −∞. 17. In this case.17.89) Before solving this equation explicitly. t → ∞.4 Solving the renormalization group equation 753 It is clear. In perturbation theory the beta function has the form of a power series in the coupling constant β(λ) = β0 λn + O λn+1 . We note that dλ dt = β λ . therefore.88) implies that (recall that λ(λ. (17. the behavior of β(λ) is shown in Fig.91) is a ﬁxed point of the beta function. Qualitatively. the ﬁrst thing that we notice is that (17. qualitatively we have . 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. 17. (17. 0. λ(t) → (17. this leads to the behavior of the β function as shown in Fig. if β(λ) is negative (β < 0. Therefore. 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. 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). t = 0) = λ) λ λ dλ′ β (λ′ ) t = 0 dt′ = t. let us analyze the qualitative behavior of the solution. (17.

(17. namely. then expanding around this ﬁxed point. β(λ) λ Figure 17. Let us assume that the beta function has a nontrivial ﬁxed point at λ∗ on the positive axis. λ(t) → 0.93) and we say that the origin in the coupling constant space is an ultraviolet ﬁxed point.3: The graph showing that the origin in the coupling constant space is an ultraviolet ﬁxed point of the β function. Let us next analyze the behavior of the solutions in some detail. ∞. we have two distinct cases to consider (when λ∗ = 0. t → −∞. .2: The graph showing that origin in the coupling constant space is an infrared ﬁxed point of the β function. t → ∞. we recover the two cases we have discussed earlier).754 17 Renormalization group and equation β(λ) λ Figure 17.

λ − λ∗ 1 dλ′ = ln . λ − λ∗ = (λ − λ∗ ) eβ0 t .89) leads to . β(λ) = β0 (λ − λ∗ )r + O (λ − λ∗ )r+1 .89) leads to λ t= λ dλ′ = β(λ′ ) λ λ or. λ → λ∗ . 17. then as t → −∞.95) In the ﬁrst case (17.94) r > 1. λ → λ∗ . 2. the arrows indicate the direction in which λ moves as t → ∞ (if we start near the ﬁxed point). µ → 0 for ﬁxed s).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 (17.97) 2. (17. (17.4 (which shows the behavior of a general β function). (17. In the second case (see (17. (17. if β0 < 0. ′−λ ) β0 (λ β0 λ − λ∗ ∗ (17. In the diagram Fig.95)). β(λ) = β0 (λ − λ∗ ) + O (λ − λ∗ )2 . then as t → ∞.17.4 Solving the renormalization group equation 755 1. (17.66) that when t → ∞.96) which implies that 1. if β0 > 0. µ → ∞ and when t → −∞.

for example. (λ − λ∗ )r−1 r−1 λ − λ∗ = (λ − λ∗ )r−1 .4: Ultraviolet and Infrared ﬁxed points of the β function. (17.22)). 1 (r − 1)β0 r−1 − 1 (λ − λ∗ )r−1 . 1 λ − λ∗ = or. 1 − (r − 1)β0 (λ − λ∗ )r−1 t (17.100) where β0 is a constant calculable in perturbation theory. g = 0 is . in both cases. The diﬀerence between the two theories is that for QED β0 > 0 (see. while for QCD. β0 < 0.99) Both in QED and QCD. = 1 − (r − 1)β0 t (λ − λ∗ )r−1 1 − (r − 1)β0 (λ − λ∗ )r−1 t . Thus. 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. (17. the series starts at the cubic order in the coupling) β(g) = β0 g3 + O g5 . the beta function has the form (namely. λ t= λ dλ′ = β(λ′ ) λ λ dλ′ β0 (λ′ − λ∗ )r 1 λ − λ∗ r−1 =− or.756 17 Renormalization group and equation β(λ) λ∗ λ β(λ) λ∗ λ Figure 17.

99) g2 = 1 + 2|β0 |g2 t 1 . Let us recall that with the MS (minimal subtraction) scheme at one loop we have (see (17.87)).103) As t → ∞.102) and perturbation theory breaks down. − 2β0 t e2 (t) = 1 e2 (17. (g∗ = 0. on the other hand. Such a pole in the evolution of the coupling constant at a particular momentum scale is called the Landau pole. diverges when 1 .46)-(17. we see that the coupling grows and. In such theories. + 2|β0 |t g 2 (t) = 1 g2 (17. β0 < 0) we have from (17.104) Since the unrenormalized (bare) coupling does not depend on the renormalization scale. 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.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. identifying the coupling with electric charge g = e and recalling β0 > 0). the theory behaves like a free theory and such theories are called asymptotically free theories. (17. therefore.101) As t increases. At inﬁnite energy. So (setting g∗ = 0.48) as well as (16. 2β0 e2 t→ (17.38)) µ µ0 ǫ λu = λ 1+ 3λ 16π 2 ǫ . the coupling becomes weaker and weaker and ultimately vanishes. it follows that . for QED we have from (17.99) e2 = 1 − 2β0 e2 t 1 . For QCD. in particular. at large momenta almost naive scaling sets in (see discussion after (17. as we have seen.

16π 2 which is ﬁnite as discussed earlier.107) so that we have λ(µ) λ(µ) µ µ dλ 3 = 2 λ 16