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

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

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

(1. it does not matter whether we write the indices “up” or “down”. (µ = 0. in this case.8) The Kronecker delta. However. namely. 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. namely. for any vector. it is necessarily positive deﬁnite. these would now consist of four components. 3 and we are being a little sloppy in representing the four vector by what may seem like its component) . However. the scalar product of any two arbitrary vectors is deﬁned to be A · B = Ai Bi = δij Ai Bj = δij Ai Bj .9) When we treat space and time on an equal footing and enlarge our three dimensional Euclidean manifold to the four dimensional spacetime manifold. Namely.2 Notations 3 where x and J represent respectively the position and the angular momentum vectors while A stands for any arbitrary vector. 2. Let us note from the deﬁnition of the length of a vector in Euclidean space that. 1. any vector would also have four components. unlike the case of the Euclidean space. (1. This also allows us to deﬁne the length of a vector simply as A2 = A · A = Ai Ai = δij Ai Aj = δij Ai Aj .1. In such a space. A2 ≥ 0. there are now two distinct four vectors that we can deﬁne on this manifold. we can again deﬁne vectors in this manifold. δij . as we know. (1. if and only if A = 0. any point in this manifold will be speciﬁed uniquely by four coordinates and. Consequently.7) where repeated indices are assumed to be summed. represents the metric of the Euclidean space and is trivial (in the sense that all the nonzero components are simply unity). consequently.

11) From the forms of the contravariant and the covariant vectors in (1. since they satisfy .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. the two four vectors have distinct transformation properties under Lorentz transformations (in fact. −x). and the covariant metric tensor.4 1 Relativistic equations xµ = (ct. −. On a more fundamental level.11). xµ = (ct. −.13) The contravariant metric tensor. The contravariant and the covariant vectors are related to each other through the metric tensor of the four dimensional manifold. x). Namely. −). 0 0 0 −1 η µν (1.12) ηµν (1. namely. ηµν . (1. (1. are inverses of each other. 0 0 0 −1 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . we can immediately read out the metric tensors for the four dimensional Minkowski space which are diagonal with the signature (+. commonly known as the Minkowski space. xµ = ηµν xν . η µν . one transforms inversely with respect to the other) and are known respectively as contravariant and covariant vectors.10) as well as using (1. we can write them in the matrix form as 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . xµ = η µν xν .

) Given two arbitrary four vectors Aµ = (A0 . A). B µ = (B 0 . namely.1.) . such a product is easily seen to be invariant under Lorentz transformations. η µν = η νµ . Physical results. use diﬀerent metric conventions and you should be careful in reading the literature. (As is clear from the above discussion. (Two Lorentz indices are said to be contracted if a contravariant and a covariant index are summed over all possible values. In fact. (1. B). Diﬀerent authors.2 Notations 5 η µλ ηλν = δµν . ηµν = ηνµ .15) This particular choice of the metric is conventionally known as the Bjorken-Drell metric and this is what we will be using throughout these lectures. however.14) Furthermore. = η µν Aµ Bν = ηµν Aµ B ν (1. the nonuniqueness in the choice of the metric tensors reﬂects the nonuniqueness of the embedding of space and time components into a four vector. however. any product of Lorentz tensors deﬁnes a scalar if all the Lorentz indices are contracted. are independent of the choice of a metric. each is symmetric as they are expected to be.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. 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. namely. if there is no free Lorentz index. (1. (1.

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

Thus. we note that v must vanish in the rest frame while u remains ˜ ˜ arbitrary. (2. (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) = ˜ . ˜ E− − m (2. v (p) = ˜ .25) (2. (2. for example. for the negative energy solutions we write p0 = E− = −E = − p 2 + m2 . ˜ v We can solve these as u(p) = − ˜ σ3 p 3 v (p).27) . from the second of the two two-component Dirac equations in (2.24) and the set of equations (2. Thus. u ˜ σ3 p3 u(p) + (E− + m)˜(p) = 0.23) This determines the two positive energy solutions of the Dirac equation (remember that the energy eigenvalues are doubly degenerate). (2. u can be thought of as the independent solution.18).22) v (p) = − ˜ σ3 p 3 E+ + m 0 1 = 0 p3 E+ + m .) ˜ Similarly. 3 (2.32 2 Solutions of the Dirac equation σ3 p 3 v (p) = − ˜ E+ + m and 1 0 = p3 −E + m + 0 .9) becomes (E− − m)˜(p) + σ3 p3 v (p) = 0.

Thus.2. v (p) ˜ (2. (2. (2. 0 1 (2.32) .30) u+ (p) = u(p) ˜ .1 Plane wave solutions 33 we obtain respectively p3 −E − m − 0 σ3 p 3 u(p) = − ˜ E− − m and 1 0 = . from the fact that we can write σ p ˜ − E 3−3m v (p) − . σ3 p 3 ˜ − E + m u(p) + u− (p) = the positive and the negative energy solutions have the explicit forms 1 0 1 0 p3 E+ + m u↑ (p) + 0 = p3 . 1 0 0 p3 ↓ u− = E− − m .31) p3 −E − m − 0 ↑ u− (p) = .21) and (2.27) are reminiscent of the spin up and spin down states of a two component spinor.28) u(p) = − ˜ σ3 p 3 E− − m 0 1 = 0 p3 E− − m . The independent two component wave functions in (2.29) and these determine the two negative energy solutions of the Dirac equation. − E+ + m 0 u↓ + = . (2.

21) and (2.) From the structure of the wave function.35) then. (The change in the sign in the dependent components comes from raising the index of the momentum. unlike the nonrelativistic counting (2s + 1).34 2 Solutions of the Dirac equation The notation is suggestive and implies that the wave function corresponds to that of a spin 1 particle. The two component spinors u(p) and v (p) in (2. the solutions take the forms (with p0 = E± = ± p2 + m2 ) u+ (p) = u(p) ˜ .34) v (p) ˜ (2.36) Here α and β are normalization constants to be determined.) 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. u− (p) = σ·p u(p) ˜ E+ + m σ·p ˜ E− − m v (p) . where p1 = p2 = 0. pi = −pi . (2. (We will determine the spin of 2 the Dirac particle shortly. u− (p) = β σ·p ˜ E + m u(p) σ· − E + p v (p) m ˜ v (p) ˜ u+ (p) = α .2 Normalization of the wave function Let us note that if we deﬁne E = E+ = p2 + m2 = −E− . (2.27) respectively are ˜ ˜ normalized as (for the same spin components) . it is also clear that. relativistic particle of spin s in the presence of both positive and negative energies follows to be 2(2s + 1).33) which can be explicitly veriﬁed. for the case of general motion. we can write the solutions for motion along a general direction as u(p) ˜ . namely. (The correct counting for the number of components of the wave function for a massive.) 2. (2.

˜ v E +m .2.40) p2 +1 (E + m)2 2E |β|2 v † (p)˜(p).38) where we have used the familiar identity satisﬁed by the Pauli matrices (σ · A)(σ · B) = A · B + iσ · (A × B). ˜ u E+m (2. Similarly. Given this. for the negative energy solutions we have σ·p v = β β −˜† (p) E + m ∗ (2. ˜ u ˜ v For diﬀerent spin components. this product vanishes.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 (2.37) 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). 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 (2.2 Normalization of the wave function 35 u† (p)˜(p) = 1 = v † (p)˜(p).

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.37) that. 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 . u† u = 1 = v † v .38) and (2. (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 . Thus. 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 . for example. 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 (2.40) simply because we have not speciﬁed their spin components.42) −˜† (p) . In dealing with the Dirac equation. we have carried along ˜ ˜ ˜ ˜ these factors in (2.

in the rest frame of the particle. Thus. ˜ u E +m Similarly.40) (for the same spin components when (2. by requiring (for the same spin components) u† (p)u+ (p) = 1 = u† (p)u− (p).46).45) However. as in the non-relativistic case. − m u† (p)u+ (p) = + (2.2 Normalization of the wave function 37 = |α|2 = 2m(E + m) † u (p)˜(p) ˜ u (E + m)2 (2. as we will see.44) Our naive instinct will be to normalize the wave function. must be positive. ψp (x) = e−ik·x u(p) with p0 = ±E.47) With the requirement (2.2. we can show that 2m |β|2 v † (p)˜(p). therefore.37) holds) . + − (2. this will reduce to u† u± = 1 which corresponds to ± the non-relativistic normalization. m † d3 x ψp (x)ψp′ (x) = (2. as we will see shortly. would give (for the same spin components) E (2π)3 δ3 (p − p′ ). this is not a relativistic normalization. u† u.43) 2m |α|2 u† (p)˜(p). a relativistically covariant normalization would be to require (for the same spin components) E = u† (p)u− (p).38) and (2. with this normalization condition. By the way. In fact. is related to the probability density which transforms like the time component of a four vector. ˜ v E+m u− (p)u− (p) = − (2. we determine from (2.46) (Remember that this will correspond to the probability density and. the motivation for such a normalization condition comes from the fact that.) The independent wave functions for a free particle.

44) (for the same spin components) u+ (p)u+ (p) = 2m 2m E + m |α|2 = = 1. v (p) ˜ (2. . Let us also note here that by construction the positive and the negative energy solutions are orthogonal.49) It is also clear that. appears to be a Lorentz invariant (scalar) and we will see later that this is indeed true. α = α∗ = 2E E |α|2 = .38 2 Solutions of the Dirac equation u† (p)u+ (p) = + or. For example. E+m E + m 2m 2m 2m E + m |β|2 = − = −1. with this normalization.50) u− (p)u− (p) = − This particular product. with this normalization. σ·p ˜ E + m u(p) σ·p − E + m v (p) ˜ . we can write the normalized positive and negative energy solutions of the Dirac equation to be u+ (p) = E+m 2m E+m 2m u− (p) = u(p) ˜ . we will obtain from (2. E+m m E+m . 2m (2.48) or. u† (p)u− (p) = − E+m m β = β∗ = E+m . Therefore. therefore. 2m E 2E |β|2 = .43) and (2. E+m E + m 2m (2.

51) Therefore. E+m E+m .2 Normalization of the wave function 39 u† (p)u− (p) + = α∗ β = 0. α = α∗ = 2E |α|2 = E.53) This still behaves like the time component of a four vector (m is a Lorentz scalar). orthonormal solutions of the Dirac equation. (There is no rest frame for a massless particle. In this case. let us note that it becomes meaningless for massless particles.38) and (2. u† (p)u− (p) = − E+m β = β∗ = E+m . that σ·p v (p) = 0.54) or. 2 2E |β|2 = E. we will obtain from (2. . however.52) While we will be using this particular normalization for massive particles. Correspondingly. an alternative normalization which works well for both massive and massless particles is given by u† (p)u+ (p) = E = u† (p)u− (p). Note. ˜ E+m σ·p u− (p)u+ (p) = −2β ∗ α v † (p) ˜ u(p) = 0. − + (2.) The probability density has to be well deﬁned. the solutions we have constructed correspond to four linearly independent. 2 (2. =α β ∗ u† (p) ˜ u† (p) ˜ σ·p E+m σ·p − E + m v (p) ˜ v (p) ˜ − u† (p) ˜ σ·p σ·p v (p) + u† (p) ˜ ˜ v (p) ˜ E+m E+m (2. ˜ E+m u+ (p)u− (p) = −2α∗ β u† (p) ˜ (2.40) (for the same spin components) u† (p)u+ (p) = + or.2.

while the arbitrariness in the normalization of u(p) may seem strange. Let us note once again that this is a particularly convenient normalization for massless particles. ψ(x) = (2. u− (p)u− (p) = − E+m u+ (p)u+ (p) = (2.3 Spin of the Dirac particle As we have argued repeatedly. d3 x ψ † (x)ψ(x) = 1.55) which vanishes for a massless particle. That this is indeed true 2 can be seen explicitly as follows. Let us note here parenthetically that. (2. E +m 2m |β|2 = −m. 2. it can be understood in light of what we have already pointed out earlier as follows.40 2 Solutions of the Dirac equation Correspondingly. we will use the notations of three dimensional . Let us deﬁne a four dimensional generalization of the Pauli matrices as (in this section.57) Namely. This product continues to be a scalar.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. 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. 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 . in this case. The true normalization is really contained in the total probability. 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). we obtain 2m |α|2 = m.

2. α ˜ (2.59) ρ= 0 ½ ½ 0 .60) We note that ρ2 = ½ so that we can invert the deﬁning relation (2. that this deﬁnes a reducible representation of spin generators since the matrices are block diagonal. (2. αi . i = 1.61) From the structures of the matrices αi and αi we conclude that ˜ [σi . (2. αj ˜ ˜ = = This shows that 1 αi satisﬁes the angular momentum algebra (remem2˜ ber = 1) and this is why we call the matrices.58) It can. the generalized ˜ Pauli matrices.101) through the relation 0 σi σi 0 αi = where = ρ˜ i = αi ρ. σj ] 0 2iǫijk σk 0 0 [σi . 2.3 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. ˜ (2.) Furthermore. σj ] 0 2iǫijk σk = 2iǫijk αk .98) and (1. however. (Note.59) and write αi = ραi = αi ρ.62) αi . let us note that . 3. ˜ (2. of course.

Let us recall that the orbital angular momentum operator is given by Li = ǫijk xj pk . αi . let us look at the free Dirac Hamiltonian in (1. Consequently. i.100) (remember that we are using three dimensional Euclidean notations in this section) H = α · p + βm = αi pi + βm. the angular momentum operators which generate rotations should commute with the Dirac Hamiltonian. (2. j.42 2 Solutions of the Dirac equation αi . 2. rotations which correspond to a subset of the Lorentz transformations must also be a symmetry of the Dirac Hamiltonian. (2.63) With these relations at our disposal. we obtain .64) As we will see in the next chapter. σj ] = 2iǫijk σk 0 −[σi . (2. αj ˜ = σi 0 0 σi 0 0 σj σj 0 − 0 σj σj 0 σi 0 0 σi = σi σj − σj σi 0 [σi . k = 1. 3. therefore. the Dirac equation transforms covariantly under a Lorentz transformation.65) Calculating the commutator of this operator with the Dirac Hamiltonian. β ˜ = = 0. Lorentz transformations deﬁne a symmetry of the Dirac Hamiltonian and. In other words. ½] = 2iǫijk αk . ½] 0 0 0 σi σj − σj σi 0 = [σi . σj ] 0 2iǫijk σk [σi .

H] + = iǫijk αj pk − iǫijk αj pk = 0. αj pj + αi . β m ˜ (2. αj pj + βm ˜ ˜ αi . (2.68) In other words. Thus. H ˜ 2 1 αi . H ˜ = = αi . To determine the spin angular momentum.66) we obtain 1 αi . αℓ pℓ ] = ǫijk αℓ [xj . ˜ 2 Ji = Li + (2. pℓ ]pk = ǫijk αℓ (iδjℓ ) pk = iǫijk αj pk . we can identify the spin angular momentum operator with .3 Spin of the Dirac particle 43 [Li .2. if rotations are a symmetry of the system. the total angular momentum which should commute with the Hamiltonian. can be identiﬁed with 1 αi . we note that αi .66) Here we have used the fact that since β is a constant matrix and m is a constant. we note that the orbital angular momentum operator does not commute with the Dirac Hamiltonian. Consequently. therefore. H] = [ǫijk xj pk . the second term in the Hamiltonian drops out of the commutator. αℓ pℓ + βm] = [ǫijk xj pk .67) = 2iǫijk αk pj = −2iǫijk αj pk .69) In this case. H ˜ 2 Li + = [Li . so that combining this relation with (2. the total angular momentum which should commute with the Hamiltonian must contain a spin part as well. (2.

74) This is the continuity equation for the probability current density associated with the Dirac equation and we note that we can identify .70) Note. in particular.72) with ψ † on the left and (2.73) where the gradient is assumed to act on ψ † .71) which has doubly degenerate eigenvalues ± 1 . Therefore. (2. or.4 Continuity equation The Dirac equation.44 2 Solutions of the Dirac equation Si = 1 αi .72) Taking the Hermitian conjugate of this equation. 2 2. ∂t −i (2. written in the Hamiltonian form. ˜ 2 (2. that 1 1 α3 = ˜ 2 2 σ3 0 0 σ3 S3 = . is given by ∂ψ = Hψ = − iα · ∇ + βm ψ. we obtain ← − ∂ψ † = ψ † iα · ∇ + βm). Multiplying (2. ∂t i (2. ∂t (2. i ∂ (ψ † ψ) = −i∇ · ψ † αψ . ∂t ∂t iψ † or. we conclude 2 that the particle described by the Dirac equation corresponds to a spin 1 particle. ∂t ∂ (ψ † ψ) = −∇ · ψ † αψ .73) with ψ on the right and subtracting the second from the ﬁrst. we obtain ∂ψ † ∂ψ +i ψ = −i ψ † α · ∇ψ + (∇ψ † ) · αψ .

P = (2. J) = (ψ † ψ. of course. to write the continuity equation as ∂ρ = −∇ · J. It is worth recalling that we have already used this Lorentz transformation property of ρ in deﬁning the normalization of the wave function. in fact.76) so that the continuity equation can be written in the manifestly covariant form ∂µ J µ = 0. 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. (We are.77)) and. is the time component of j µ (see (2.4 Continuity equation 45 ρ = ψ † ψ = probability density.2. therefore. ψ † αψ). J = ψ † αψ = probability current density.) On the other hand. yet to show that j µ transforms like a four vector which we will do in the next chapter.80) .78) This.77) (2. ∂t This suggests that we can write the current four vector as J µ = (ρ.75) (2. ρ. (2. must transform like the time coordinate under a Lorentz transformation.79) is a constant independent of any particular Lorentz frame. (2. shows that the probability density. the total probability d3 x ρ = d3 x ψ † ψ. and note that the Hermitian conjugate of ψ satisﬁes (2.

86) (2. we obtain ← − ψ − iγ 0 (γ µ )† γ 0 ∂µ − m) = 0.81) Multiplying this equation with γ 0 on the right and using the fact that (γ 0 )2 = ½. J = ψγψ = ψ † γ 0 γψ = ψ † αψ. ∂µ ψγ µ ψ = 0. 2.85) that J 0 = ψγ 0 ψ = ψ † γ 0 γ 0 ψ = ψ † ψ = ρ. or.77). (2.83) Multiplying (2. γ 0 (γ µ )† γ 0 = γ µ .82) with ψ on the right and subtracting the second from the ﬁrst.84) This is. (2.80) with ψ on the left and (2. or.82) where we have used the property of the gamma matrices that (for µ = 0.46 2 Solutions of the Dirac equation ← − ψ † − i(γ µ )† ∂µ − m = 0. i∂µ ψγ µ ψ = 0. 3) γ 0 γ µ γ 0 = (γ µ )† . (2. ← − ψ − iγ µ ∂µ − m = 0. 1. in fact. or. Note from the deﬁnition in (2. (2.85) . (2. we obtain ← − i ψγ µ ∂µ ψ + ψγ µ ∂µ ψ = 0. the covariant continuity equation and we can identify J µ = ψγ µ ψ. which is what we had derived earlier in (2.

1 (as well as from the equation above) that the positive and the negative energy solutions are separated by a gap of magnitude 2m (remember that we are using c = 1). In the free particle case.87) Thus. m −m Figure 2. We note from Fig. for example. 2. we note that the energy spectrum . can be deﬁned to be positive deﬁnite even in the presence of negative energy solutions. This is rather diﬀerent from the case of the Klein-Gordon equation that we have studied in chapter 1. Even when the probability density is consistently deﬁned.1: Energy spectrum for a free Dirac particle. (2.40). even for this simple case of a free particle the energy spectrum has the form shown in Fig. in such a case.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. Consequently.2.5 Dirac’s hole theory We have seen that Dirac’s equation leads to both positive and negative energy solutions. First of all.1. 2. because it is a ﬁrst order equation (particularly in the time derivative). the presence of negative energy solutions leads to many conceptual diﬃculties. 2.38) and (2. as we o have seen explicitly in (2. the energy eigenvalues are given by p0 = E± = ±E = ± p2 + m2 . the probability density. the probability density is independent of time derivative much like the Schr¨dinger equation.

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

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. 2. the Dirac equation passes as a one particle equation primarily because of the Pauli exclusion principle. 2. we will study some of the useful properties of these matrices. 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.2. .) This is Dirac’s theory of electrons and works quite well. (This unconventional deﬁnition of a vacuum state can be avoided in a second quantized ﬁeld theory which we will study later. On the other hand. the Dirac theory predicts an anti-particle of equal mass for every Dirac particle. Furthermore. In this section. the four Dirac matrices satisfy (in addition to the Cliﬀord algebra) (γ 0 )† = γ 0 . 1. if enough energy is provided to such a ground state. γ µ . (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. This “hole” state is what we have come to recognize as the anti-particle – in this case.) In spite of this. 2. 3. The amount of energy necessary to excite a negative energy electron to a positive energy state is E ≥ 2m. As we have seen. a positron – and the process under discussion is commonly referred to as pair creation (production).6 Properties of the Dirac matrices The Dirac matrices.6 Properties of the Dirac matrices 49 ple. 3. µ = 0. Thus. (γ i )† = −γ i . this is a general feature that combining quantum mechanics with relativity necessarily leads to a many particle theory. thereby. However. (2. a negative energy electron can make a transition to a positive energy state and can appear as a positive energy electron. i = 1.88) Tr γ µ = 0. in deﬁning a ﬁrst order equation.

91) where we have used the property of the Pauli matrices σ1 σ2 σ3 = i½. by deﬁnition. 2 γ5 = ½.50 2 Solutions of the Dirac equation Since these are 4 × 4 matrices. let us deﬁne the following sets of matrices. Note that.91) that we can identify this with the matrix ρ deﬁned earlier in (2. Let γ5 = iγ 0 γ 1 γ 2 γ 3 = − where ǫ0123 = 1 = −ǫ0123 . Note that in our particular representation for the γ µ matrices given in (1.89) represents the four-dimensional generalization of the Levi-Civita tensor.93) . (2. 4! (2. (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 .92) We recognize from (2.91). (2. (2. the identity matrix will correspond to one of them.60). a complete set of Dirac matrices must consist of 16 such matrices. Of course. † γ5 = γ5 .90) i ǫµνλρ γ µ γ ν γ λ γ ρ . To obtain the other basis matrices.

as (µ.96) Finally.95) Since we know the explicit forms of the matrices ½. we can also deﬁne six anti-symmetric matrices. 3) σ µν = −σ νµ = i i γ µ . γ5 . γ µ + = 0. we can deﬁne four new matrices as γ5 γ µ . (2. ν = 0. 3.6 Properties of the Dirac matrices 51 and that. since it is the product of all the four γ µ matrices. 1.97) = −i η µν − γ µ γ ν . γ ν = (γ µ γ ν − γ ν γ µ ) 2 2 (2. let us write out the forms of γ5 γ µ also in this representation. µ = 0. 1.2. it anticommutes with any one of them. 2. (2. σ µν . 2. Namely. = i η µν − γ ν γ µ whose explicit forms in our representation can be worked out to be (i. γ5 γ 0 = γ5 γ i = 0 ½ ½ ½ 0 −σi 0 ½ 0 0 0 −½ σi 0 = ½ 0 −½ 0 −σi 0 0 . γ µ and γ5 in our representation. 2.94) Given the matrix γ5 . (2. k = 1. ½ 0 = σi . j. 3) .

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 . namely. 1. γµ. therefore. 4.70) that the matrices 2 αi represent the ˜ spin operators for the Dirac particle. γµ.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 . ¯ Here. 4.98) 1 We have already seen in (2.100) Γ(T ) = Γ(P ) = = γ5 where the numbers on the right denote the number of matrices in each category and these. From (2. provide a basis for all the 4 × 4 matrices. γ5 . (This relation can be obtained from (2. ˜ 2 4 (2.98) using the identity for products of Levi-Civita tensors. 1. namely. that the matrices 1 1 αi = ǫijk σ jk . σ µν .) We have thus constructed a set of sixteen Dirac matrices.98) we conclude. ˜ (2.99) can be identiﬁed with the spin operators for the Dirac particle. ǫijk ǫℓjk = 2 δiℓ . Γ(S) Γ(V ) Γ(A) = = ½. 6. (2. in fact. the notation is suggestive and stands for the fact that ψΓ(S) ψ .

However. ¯ ¯ ¯ ¯ Similarly. which is deﬁned to be Hermitian. even within a given class.103) γ 0 σ µν γ 0 = The Dirac matrices satisfy nontrivial (anti) commutation relations. V.104) (2.2. it can be checked that except for γ5 . ψΓ(A) ψ and ψΓ(P ) ψ behave respectively like a vector. γ 0 γ5 γ µ γ 0 = −γ5 γ 0 γ µ γ 0 = −(γ5 )† (γ µ )† = (γ5 γ µ )† . it 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 )† . (2. Finally.102) α = S. In fact. We already know that . it follows easily that γ 0 ½γ 0 = (γ 0 )2 = ½ = (½)† . all other matrices satisfy γ 0 Γ(α) γ 0 = (Γ(α) )† . (2. has its own hermiticity property. A. T.101) where we have used the fact that γ5 is Hermitian and it anti-commutes with γ µ .6 Properties of the Dirac matrices 53 transforms like a scalar under Lorentz and parity transformations. Let us note here that each of the matrices. γ 0 γ µ γ 0 = (γ µ )† . tensor. axial-vector and a pseudo-scalar under a Lorentz and parity transformations as we will see in the next chapter. ψΓ(V ) ψ. 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 )† . (2. ψΓ(T ) ψ.

Similarly. we obtain . C + .105) We can also calculate various other commutation relations in a straightforward and representation independent manner. γ µ + + = 2η µν ½. for the commutator of two σ µν matrices.107) can also be expressed in terms of commutators (instead of anti-commutators) as [A. since γ µ matrices satisfy simple anti-commutation relations.107) We note here parenthetically that the commutator in (2. −i η νλ − γ ν γ λ γλ − γν γµ. (2. (2. γ µ . = 0.108) However. In this derivation. B + C − B A.54 2 Solutions of the Dirac equation γµ. we have used the fact that (2. γ ν γ λ = i γµ. γν + + = 2i η µν γ λ − η µλ γ ν .107) is more useful for our purpose. BC] = ABC − BCA = (AB + BA)C − B(AC + CA) = A. the form in (2. C] . σ νλ = γ µ . For example.106) [A. γν γ5 . B] C + B [A. γλ = i γ µ. BC] = [A. (2.

for example. σ λρ = i γ µ γ ν . γµ A µ = γµ Aν γ ν γ µ = Aν γµ γ ν γ µ = −2Aν γ ν = −2A /γ /.109) Thus.112) . σ λρ = − i η µν − γ µ γ ν .2. (these relations are true only in 4-dimensions) γµ γ ν γ µ = γµ γ ν . + ν − γµγν (2. σ λρ + i γ µ . This becomes particularly useful in calculating various amplitudes involving Dirac particles. (2. 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. We will see in the next chapter that they provide a representation for the Lorentz algebra. Thus. and it follows now that.110) (2. γ µ = 2η νµ γµ − 4γ = 2γ ν − 4γ ν = −2γ ν .111) (2. 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 . σ λρ = iγ µ γ ν . The various commutation and anti-commutation relations also lead to many algebraic simpliﬁcations in dealing with such matrices.6 Properties of the Dirac matrices 55 σ µν . where we have used (γµ = ηµν γ ν ) γµ γ µ = 4 ½. Similarly.

56 2 Solutions of the Dirac equation γµ γ ν γ λ γ µ = γµ γ ν γ λ .113) = 2η λµ γµ γ ν + 2γ ν γ λ = 2γ λ γ ν + 2γ ν γ λ = 2 γ λ . 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 .115) Even more complicated traces can be evaluated by using the basic relations we have developed so far. or. γ ρ or.114) Tr γ µ γ ν + Tr γ5 γ µ (2. γ λ + = 8η µν η λρ − 8η µλ η νρ + Tr γ ν γ λ γ µ . 2 Tr γ µ γ ν γ λ γ ρ = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ . 2 = Tr γ µ γ5 = −Tr γ5 γ µ = 0. − γργµ Tr γ µ γ ν γ λ γ ρ = 4 η µν η λρ − η µλ η νρ + η νλ η µρ . For example.116) . we know from the cyclicity of traces that Tr γ µ γ ν = Tr γ ν γ µ . γ ν 2 2 1 = Tr 2η µν ½ = η µν Tr ½ = 4η µν . γ ν + and so on. it follows (in 4-dimensions) that 1 1 Tr γ µ γ ν + Tr γ ν γ µ = Tr γ µ . = (2. The commutation and anticommutation relations also come in handy when we are evaluating traces of products of such matrices. γ µ + − γµγλ = 4η λν ½. Therefore. (2. For example. we note that Tr γ µ γ ν γ λ γ ρ = Tr γµ. = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ − Tr γ ν γ λ γ ρ γ µ . (2.

120) This is known as the Weyl representation for the Dirac matrices and is quite useful in the study of massless fermions. 1 S = √ γ0 2 ½ + γ2 . 1 γM = . (2. 3 γM = . the µ γM matrices have the forms 0 σ2 0 σ2 σ2 0 −σ2 0 iσ3 0 −iσ1 0 0 iσ3 0 −iσ1 0 γM = .2. 0 σµ ˜ σµ 0 γW = where µ . However.119) σ µ = (½. let us note that we have constructed a particular representation for the Dirac matrices commonly known as the Pauli-Dirac representation. This is known as the Majorana representation and is quite useful in the study of Majorana fermions which are charge neutral fermions. σ µ = (½. namely. σ). It can be checked . −σ). (2. 2 γM = .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 . For example.6 Properties of the Dirac matrices 57 and so on. there exists a representation for the Dirac matrices where γ µ are all purely imaginary.118) Similarly. there exists yet another representation for the γ µ matrices. there are other equivalent representations possible which may be more useful for a particular system under study. ˜ (2. We would use all these properties in the next chapter to study the covariance of the Dirac equation under a Lorentz transformation. Explicitly. To conclude this section. (2.

As we have pointed out in (2.100). it can be easily checked that Tr Γ(a) Γ(b) = 0. γµ . This is easily demonstrated by showing that they are linearly independent which is seen as follows.124) Tr Γ(a) Γ(b) = δ(a) . γ µ . γ5 } .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 . T. As a result. V. we can write the inverse set of matrices as 1 {½. γ5 γ µ . (2. σµν .123) where “Tr” denotes trace over the matrix indices. (2. (2. For example. the linear independence of the set of matrices in (2.121) 2. given this set of matrices. σ µν . Tr Γ(a) Γ(a) Γ(a) = such that a not summed. −γ5 γµ . the sixteen Dirac matrices Γ(a) . A. 1 S=√ 2 ½ + γ5 γ 0 . γ5 } . we can construct the inverse set of matrices as Γ(a) . We have explicitly constructed the sixteen matrices to correspond to the set Γ(a) = {½. a = S.125) Γ(a) = (2. (2. a = b.1 Fierz rearrangement.122) is straightforward.126) With this.122) From the properties of the γ µ matrices. Explicitly. it follows now that if .6. P deﬁne a complete basis for 4 × 4 matrices. 4 (b) (2.

M= (a) C(a) Γ(a) . where b is arbitrary. C(b) = 0. Since the set of matrices in (2. (a) Tr Γ(b) or. Therefore. any arbitrary 4 × 4 matrix M can be expanded as a linear superposition of these matrices. V.127) then. and taking trace over the matrix indices and using (2.128) for any b = S.129) Multiplying this expression with Γ(b) and taking trace over the matrix indices.122) are linearly independent. (M ) (2. C(a) δ(b) = 0. A.127) with Γ(b) . or.2. (2. we obtain . namely. (M ) C(a) δ(b) . (a) (a) or. As a result they constitute a basis for 4 × 4 matrices.130) C(b) = Tr Γ(b) M . we obtain C(a) Tr Γ(b) Γ(a) = (a) (a) (M ) (M ) (a) Tr Γ(b) M = or.127) implies that all the coeﬃcients of expansion must vanishing which shows that the set of sixteen matrices Γ(a) in (2. Substituting (2. T.130) into the expansion (2.6 Properties of the Dirac matrices 59 C(a) Γ(a) = 0. multiplying (2.125) we obtain C(a) Γ(a) = 0. (a) C(a) Tr Γ(b) Γ(a) = 0. P .129). (2.122) provide a basis for the 4 × 4 matrix space. (2. (a) (2.

2. 4 and correspond to the matrix indices of the 4 × 4 matrices and we are assuming that the repeated indices are being summed.133).131) Introducing the matrix indices explicitly. it is now straightforward to obtain . γ. For example.133) Using the relations in (2. η = 1. then using (2. or. (2.132) Here α. this leads to Mαβ = (a) Γ(a) (a) γη Mηγ Γαβ . (2.132) we can derive (for simplicity. we note that if M and N denote two arbitrary 4 × 4 matrices. Just like any other completeness relation. ¯ ¯ ¯ ¯ 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 . Equation (2.132) describes a fundamental relation which expresses the completeness relation for the sixteen basis matrices. (a) Γ(a) γη Γ(a) αβ = δαη δβγ . 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 . 3.60 2 Solutions of the Dirac equation M= (a) Tr Γ(a) M Γ(a) . β. it can be used eﬀectively in many ways. ¯ ¯ ¯ ¯ (2.

In deriving these identities.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 .134) pick up a negative sign which arises from commuting the fermionic ﬁelds past one another.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 . On the other hand.135) equivalently as . Let us note that using the explicit forms for Γ(a) and Γ(a) in (2. we can ˜ deﬁne a new spinor ψ4 = N ψ4 to write the identity in (2. N and any spinors.134) are known as the Fierz rearrangement identities which are very useful in calculating cross sections. (2.135) Since this is true for any matrices M. ¯ = 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 . (2. we can write the ﬁrst of the Fierz rearrangement identities in (2.134) The two relations in (2. the right-hand sides of the identities in (2. we have assumed that the spinors are ordinary functions.124) respectively. if they correspond to anti-commuting fermion operators.122) and (2.2.

J. we obtain from (2. (2. Relativistic Quantum Mechanics. L. D. New York. (2.136) Thus. Bjorken and S.110) and (2.113). for example. 1968. choose M = (½ − γ5 ) γ µ .136) ¯ ¯ ψ1 (½ − γ5 ) γ µ ψ2 ψ3 (½ − γ5 ) γµ ψ4 ¯ ¯ = −ψ1 (½ − γ5 ) γ µ ψ4 ψ3 (½ − γ5 ) γµ ψ2 . (2. Drell. N = (½ − γ5 ) γµ . if we which is often calculationally simpler. Quantum Mechanics. Schiﬀ. McGraw-Hill.137) then using various properties of the gamma matrices derived earlier as well as (2. . 2.138) 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). McGraw-Hill.7 References 1. New York.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 . 1964. 2. I.

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

This page intentionally left blank .

1) x′ = sin θ x1 + cos θ x2 .2) (3. we see from (3. Thus. but this can also be written in terms of the four vector notation we have developed. The rotation around the z-axis in (3.3) that a 65 . we know that a rotation of coordinates around the z-axis by an angle θ can be represented as the transformation x → x′ = R x.Chapter 3 Properties of the Dirac equation 3. where R represents the rotation matrix such that x′ = cos θ x1 − sin θ x2 . 1 (3. For example. we are well acquainted with rotations.1 Lorentz transformations In three dimensions. 3 Here we are using a three dimensional notation.2) can also be written in matrix form as x1 cos θ − sin θ 0 ′ x2 = sin θ cos θ 0 x2 .3) so that the coeﬃcient matrix on the right hand side can be identiﬁed with the rotation matrix in (3. x′ 1 x′ 3 0 0 1 x3 (3.1). 2 x′ = x3 .

x′ 3 = x3 .4) where we have identiﬁed θ = ǫ = inﬁnitesimal. We observe here that the matrix representing the inﬁnitesimal change under a rotation is anti-symmetric.6) can also be written in the matrix form as . 3 x′ = ǫx1 + x2 . Under a Lorentz boost along the x-axis. such that (3.5) x′ 0 = γx0 − γβx1 . 2 (3.66 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. (3. we also know that the coordinates transform as (boost velocity v = β since c = 1) xµ → x′ µ . with unit determinant. x′ 2 = x2 . We also note from (3. R (RT R = ½).6) where the Lorentz factor γ is deﬁned in terms of the boost velocity to be 1 1 − β2 γ= . x′ 1 = −γβx0 + γx1 . 1 x′ = x3 .2) that an inﬁnitesimal rotation around the 3-axis (z-axis) takes the form x′ = x1 − ǫx2 . (3.7) We recognize that (3.

where cosh ω − sinh ω 0 0 cosh ω 0 0 (3.11) 1 β2 − = 1. 1 0 0 1 (3. we note that a Lorentz boost along the x-direction can be written as x′ µ = Λµν xν . Thus. 1 − β2 1 − β2 (3.3.9) 1 − β2 cosh2 ω − sinh2 ω = Since the range of the boost velocity is given by −1 ≤ β ≤ 1.10) Λµν − sinh ω = 0 0 0 0 .9) that −∞ ≤ ω ≤ ∞.8) 1 1 − β2 β . (3. = 0 0 1 0 x2 0 0 0 1 x3 x′ 0 γ (3. we conclude from (3.12) . .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 .

the “angle” of rotation. (That these rotations become complex is related to the fact that the metric has opposite signature for time and space components.) Furthermore. x′ = Λµν xν . Lorentz boosts correspond to noncompact transformations unlike space rotations. µ (3.13) The matrix representing the Lorentz transformation of the coordinates.16) .15) and also has a unit determinant. then we can write (ω = ǫ = inﬁnitesimal) −ǫ 0 0 −ǫ 1 0 0 = δµν + ǫµν . is easily seen from (3. = 0 0 1 0 0 0 0 1 1 Λµν (3.13) to be an ν orthogonal matrix in the sense that µ (ΛT )µν Λνλ = Λνµ Λνλ = δλ .15) also shows that the covariant vector transforms in an inverse manner compared with the contravariant vector. as we have seen. 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). Λµ (or Λµν ). ω. much like the rotation matrix R in (3.3). cosh ω 3 Properties of the Dirac equation ν Λµ = ηµλ η νρ Λλρ which would lead to the transformation of the covariant coordinate vector. (3. 0 1 0 0 0 1 sinh ω (3.) Therefore. (or the parameter of boost) can take any real value and. 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). we can obtain. namely. (3. as a result.12) or (3. (Incidentally.14) sinh ω = 0 0 0 0 cosh ω 0 0 .68 From this.

20) . or. therefore. = 0 0 0 0 0 0 0 0 0 ǫµν = η νλ ǫµλ (3. the matrix representing the change under an inﬁnitesimal Lorentz boost is anti-symmetric just like the case of an inﬁnitesimal rotation. 0 0 0 0 (3. ρ Λµρ Λµσ = δσ . ηµν Λµρ Λνσ xρ xσ = ηρσ xρ xσ . Λµρ Λµσ = ηρσ .3.18) In other words.17) It follows from this that ǫ 0 0 −ǫ 0 0 0 = −ǫνµ .19) ηµν x′ µ x′ ν = ηµν xµ xν . (3. which can be thought of as rotations in the four dimensional space-time. (3.1 Lorentz transformations 69 where. General Lorentz transformations are deﬁned as the transformations x′ µ = Λµν xν . namely. or. which leave the length of the vector invariant. we note that we can combine rotations and Lorentz boosts into what are known as the homogeneous Lorentz transformations. ηµν Λµρ xρ Λνσ xσ = ηµν xµ xν . 0 −ǫ 0 0 0 0 0 ǫµν −ǫ = 0 0 0 0 . or. or. In a general language.

det Λ = 1.25) . or. Lorentz transformations deﬁne the maximal symmetry of the space-time manifold which leaves the origin invariant. we conclude that Λ00 ≥ 1. Incidentally.24) µ ν (3. “gonia” in Greek means an angle or a corner and. then the transformation is called orthochronous. of course.23) = Λνµ and ΛT ν µ = Λνµ The set of homogeneous Lorentz transformations satisfying Λ00 ≥ 1. or. (3.15).21) Therefore. therefore. Λ00 Λ00 2 2 − Λi 0 2 = 1. or. we can write out the relation (3. orthochronous means straight up in time. what we have seen before in (3. Choosing ρ = σ = 0. (3. det Λ = ±1. Namely.22) If Λ00 ≥ 1. orthogonal means the corner that is straight up (perpendicular). (3.70 3 Properties of the Dirac equation This is. (The Greek preﬁx “ortho” means straight up. and (3.) Note also that since ΛT Λ = ½. In the same spirit. or. such a Lorentz transformation does not change the direction of time. 2 = 1 + Λi 0 ≥ 1. we note that ΛT as can be seen from (3. we obtain (for clarity.20) explicitly as Λ00 Λ00 + Λi 0 Λi 0 = 1.20)) (det Λ)2 = 1. Thus. an orthodontist is someone who can make your teeth straight. Λ00 ≤ −1.

we can obtain the other Lorentz transformations by simply appending space reﬂection or time reﬂection or both (which are discrete transformations). then we obtain a Lorentz transformation −1 0 0 0 0 1 0 0 Λprop . if we add both space and time reﬂections. Given the proper orthochronous Lorentz transformations. If we add time reversal. = 0 0 −1 0 0 0 0 −1 Λ = Λspace Λprop (3. (Just to emphasize.) In general.27) This would correspond to having Λ00 ≥ 1. Thus.26) Λ00 ≤ −1. det Λ = −1 (which is orthochronous but no longer proper). det Λ = ±1.28) satisfying Λ00 ≤ −1 and det Λ = −1 (which is neither proper nor orthochronous). Λ00 ≥ 1. xµ → −xµ . we obtain a Lorentz transformation . we obtain a Lorentz transformation 1 0 0 0 0 −1 0 0 Λprop . orthochronous Lorentz transformations and constitute a set of continuous transformations that can be connected to the identity matrix. to a proper Lorentz tranformation. to a proper orthochronous Lorentz transformation. there are four kinds of Lorentz transformations. namely. Finally. (3. t → −t.1 Lorentz transformations 71 are known as the proper. if Λprop denotes a proper Lorentz transformation. however. 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. then by adding space reﬂection. x → −x. = 0 0 1 0 0 0 0 1 Λ = Λtime Λprop (3. det Λ = ±1.3.

(3.72 3 Properties of the Dirac equation Λ = Λspace−time Λprop with Λ00 ≤ −1 and det Λ = 1 (which is proper but not orthochronous). In simple terms. let us now consider the Dirac equation iγ µ ∂µ − m ψ(x) = 0. cannot be continuously connected to the identity matrix since they involve discrete reﬂections.2 Covariance of the Dirac equation Given any dynamical equation of the form Lψ = 0. we would refer to proper orthochronous Lorentz transformations as the Lorentz transformations. With this general deﬁnition. (3. however. Under a Lorentz transformation xµ → x′ µ = Λµν xν . 0 −1 0 (3. we say that it is covariant under a given transformation provided the transformed equation has the form L′ ψ ′ = 0. the operator L. with the transformed variables).33) (3.32) . (3. These additional transformations. In these lectures.31) where ψ ′ represents the transformed wavefunction and L′ stands for the transformed operator (namely. covariance implies that a given equation is form invariant under a particular transformation (has the same form in diﬀerent reference frames).29) where L is a linear operator.30) −1 0 0 0 −1 0 = 0 0 −1 0 0 0 0 Λprop . 3.

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

but get rotated by the S(Λ) matrix.74 = 3 Properties of the Dirac equation dx |Λx S(Λ)ψ(x). A more physical way to understand this is to note that the Dirac wave function simply consists of four functions which do not change.40) (3.38) Namely. Let us also note from (3. iγ µ (3.35)) ψ ′ (x′ ) = x′ |ψ ′ = S(Λ)ψ(x). the eﬀect of the Lorentz transformation. ∂xµ ∂x′ ν iγ µ Λνµ ∂ν′ − m S −1 (Λ)ψ ′ (x′ ) = 0.42) .37) we obtain (see (3.39) (3. on the wave function. (3.41) ∂x′ ν ∂ − m S −1 (Λ)ψ ′ (x′ ) = 0. Since the Lorentz transformations are invertible. (3. so that. from (3. ′ iΛµν γ ν ∂µ − m S −1 (Λ)ψ ′ (x′ ) = 0. or.33) that ∂x′ µ = Λµν . or. the matrix S(Λ) must possess an inverse so that from (3. ′ iΛµν Sγ ν S −1 ∂µ − m ψ ′ (x′ ) = 0.35) we can write ψ(x) = S −1 (Λ)ψ ′ (x′ ). or. we can write iγ µ ∂µ − m ψ(x) = 0. Thus. can be represented by a matrix S(Λ) which depends only on the parameter of transformation Λ and not on the space-time coordinates.37) where the wave function is recognized to be ψ(x) = x|ψ . ∂xν deﬁne a set of real quantities. (3. or.

the matrices γ ′ µ satisfy the Cliﬀord algebra and. It now follows from (3. γ ′ µ. we see from (3. Λνσ γ σ (3.45) where we have used the orthogonality of the Lorentz transformations (see (3. γ ′ ν Λµρ γ ρ . there must exist a matrix connecting the two representations. generating Lorentz transformations (for the Dirac wavefunction). (3. (3. γ 0 (S −1 γ µ S)† γ 0 = S −1 γ µ S.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. Sγ 0 S † γ 0 γ 0 γ µ γ 0 γ 0 S −1 γ 0 S −1 = γ µ .43) Let us note that if we deﬁne γ ′ µ = Λµν γ ν . Therefore. (3. Next. Λµν γ ν = S −1 γ µ S. (Sγ 0 S † γ 0 )γ µ (Sγ 0 S † γ 0 )−1 = γ µ . or. such that Λµν Sγ ν S −1 = γ µ .42) that the Dirac equation will be form invariant (covariant) under a Lorentz transformation provided there exists a matrix S(Λ). Therefore. or.15)).44) + = + + µν = Λµρ Λνσ γ ρ . by Pauli’s fundamental theorem. or. then. let us note that since the parameters of Lorentz transformation are real (namely.46) † † † .3. γ µ and γ ′µ . γ σ = 2Λµρ Λνρ ½ = 2η = Λµρ Λνσ 2η ρσ ½ ½.2 Covariance of the Dirac equation 75 where we have used (3. (Λ∗ )µν = Λµν ) γ 0 Λµν γ ν γ 0 = Λµν γ 0 (γ ν )† γ 0 = Λµν γ ν .40). or.

47). must be proportional to the identity matrix (this can be easily checked by taking a linear combination of the sixteen basis matrices in (2. or. γ 0 (γ 0 Sγ 0 S † )γ 0 = b∗ ½.47) Taking the Hermitian conjugate of (3. we obtain (Sγ 0 S † γ 0 )† = b∗ ½. As a result. determines that the parameter b is real. therefore. or. (3. . we can determine the unique value of b in the following way. (3. we can denote Sγ 0 S † γ 0 = b ½. Using these in (3. therefore.43) and the relations (γ 0 )† = γ 0 = (γ 0 )−1 as well as γ 0 (γ µ )† γ 0 = γ µ . or.50) The real roots of this equation are b = ±1. b = b∗ .46) that the matrix Sγ 0 S † γ 0 commutes with all the γ µ matrices and. or. S † γ 0 = b γ 0 S −1 .48) which. (3. γ 0 Sγ 0 S † = b∗ ½. namely. or. Sγ 0 S † γ 0 = b∗ ½ = b ½. It is clear from (3. (3. b4 = 1. we determine det (Sγ 0 S † γ 0 ) = det (b ½).51) In fact. (3.76 3 Properties of the Dirac equation Here we have used (3. det S = 1.49) We also note that det γ 0 = 1 and since we are interested in proper Lorentz transformations.47).100) and calculating the commutator with γ µ ).

52) which follows since S † S represents a non-negative matrix. 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 .2 Covariance of the Dirac equation 77 Let us note. Thus.43) and (3.54) which implies (see (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. we are assuming Λ00 ≥ 1.47). (3. or. it follows from (3. Since we are dealing with proper Lorentz transformations. S † γ 0 = γ 0 S −1 .56) (3. therefore. using (3. (3.3. (3. b > 0. Tr S † S = 4bΛ00 0. Λ00 ≤ −1. or.53) or.53)) that b > 0 and.47) that Sγ 0 S † γ 0 = ½. b < 0. (3.51) that b = 1. . we conclude from (3. The two solutions of this equation are obvious Λ00 ≥ 1.

61) (3.18). namely. .60) ½+ ½+ i Mµν ǫµν 4 ½− i Mστ ǫστ 4 i i Mµν ǫµν − Mµν ǫµν + O(ǫ2 ) 4 4 (3. 4 (3. ν ν ν (3.63) ½ + O(ǫ2 ). we can expand the matrix S(Λ) as S(Λ) = S(ǫ) = ½ − i Mµν ǫµν . ǫµν = −ǫνµ . 4 (3. Mµν = −Mνµ .78 3 Properties of the Dirac equation Next.57) From our earlier discussion in (3. let us consider an inﬁnitesimal Lorentz transformation of the form (ǫµν inﬁnitesimal) x′µ = Λµν xν = (δµ + ǫµ )xν = xµ + ǫµ xν . (3.62) (3.59) where the matrices Mµν are assumed to be anti-symmetric in the Lorentz indices (for diﬀerent values of the Lorentz indices. Mµν denote matrices in the Dirac space). therefore.58) For an inﬁnitesimal transformation. we recall that the inﬁnitesimal transformation matrix is antisymmetric. since ǫµν = −ǫνµ . We can also write S −1 (ǫ) = ½ + so that S −1 (ǫ)S(ǫ) = = = i Mµν ǫµν .

Therefore.43) now takes the form S −1 (ǫ)γ µ S(ǫ) = Λµν γ ν = δµν + ǫµν γ ν .64) that if we identify M λρ = σ λρ . then. we have determined the form of S(ǫ) to be S(ǫ) = ½ − i i Mµν ǫµν = ½ − σµν ǫµν . ½+ γµ + − i Mλρ ǫλρ γ µ 4 ½− i Mστ ǫστ 4 = γ µ + ǫµν γ ν . M λρ 4 i ǫλρ γ µ . we see that for inﬁnitesimal transformations. or. and note from (3.66) (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 .65) − (3. 4 4 (3. let us recall the commutation relation (2. or. or.2 Covariance of the Dirac equation 79 To the leading order. σ νλ = 2i η µν γ λ − η µλ γ ν .3. 4 At this point. 4 4 (3. therefore.67) which coincides with the right hand side of (3.64). S −1 (ǫ) indeed represents the inverse of the matrix S(ǫ).64) i ǫλρ γ µ . i i λρ ǫ Mλρ γ µ − ǫλρ γ µ Mλρ + 0(ǫ2 ) = γ µ + ǫµν γ ν . The deﬁning relation for the matrix S(Λ) in (3.68) . i ǫλρ γ µ . M λρ = ǫµν γ ν . λ 2 ρ = − (3.106) γ µ .

A ﬁnite transformation can. of course. (3. the matrix S(Λ) for a ﬁnite Lorentz transformation will be the product of a series of such inﬁnitesimal matrices which leads to an exponentiation of the inﬁnitesimal generators with the appropriate parameters of transformation. satisfy can 2 be identiﬁed with the Lorentz algebra (which also explains why they are closed under multiplication). ˜ 2 2 2 (3. Thus. For completeness. i (3.72) . 2 (3.80 3 Properties of the Dirac equation Let us note here from the form of S(ǫ) that we can identify 1 σµν . then. consequently. In such a case (see also (2.98)). we have shown that there exists a S(Λ) which satisies (3. be constructed out of a series of inﬁnitesimal transformations and.69) with the generators of inﬁnitesimal Lorentz transformations for the 1 Dirac wave function. let us note that inﬁnitesimal rotations around the 3-axis or in the 1-2 plane would correspond to choosing ǫ12 = ǫ = −ǫ21 .109) which the generators of the inﬁnitesimal transformations. 1 σµν .71) A ﬁnite rotation by angle θ in the 1-2 plane would.) We will see in the next chapter (when we study the representations of the Lorentz group) that the algebra (2. at least for inﬁnitesimal Lorentz transformations. (The other factor of 2 is there to avoid double counting. the Dirac equation is form invariant (covariant) under such a Lorentz transformation. S(ǫ) = ½ − i 1 i σ12 ǫ = ½ + γ1 γ2 ǫ = ½ − α3 ǫ.43) and generates Lorentz transformations and as a result.70) with all other components of ǫµν vanishing. be obtained from an inﬁnite sequence of inﬁnitesimal transformations resulting in an exponentiation of the inﬁnitesimal generators as ˜ S(θ) = e− 2 α3 θ .

therefore. so that we can write (see also (2.75) That is.2 Covariance of the Dirac equation 81 Note that since α† = αi . is double valued and. α 2 2 (3. we have S † (θ) = S −1 (θ).77) with all other components of ǫµν vanishing.78) . consistent with the fact that the Dirac equation describes spin 1 2 particles. namely. of course. In this case. Let us next consider an inﬁnitesimal rotation in the 0-1 plane. the rotation operator. This is. in this case. θ θ − i˜ 3 sin α 2 2 = −S(θ). Furthermore. we can determine S(θ) = cos This shows that S(θ + 2π) = − cos S(θ + 4π) = S(θ). 0 ½ (3. (3. (3. corresponds to a spinor representation. (3. rotations ˜i ˜ deﬁne unitary transformations. 2 ½− (3. therefore.76) θ θ − i˜ 3 sin .73) α3 ˜ = ½ 0 = ½.74) and. namely.98)) S(ǫ) = i 1 σ01 ǫ = ½ + γ0 γ1 ǫ 2 2 1 = ½ − α1 ǫ. we are considering an inﬁnitesimal boost along the 1-axis.3. recalling that σ3 0 0 σ3 α3 = ˜ we have 2 . we can identify ǫ01 = ǫ = −ǫ10 .

This is related to the fact that Lorentz boosts are non-compact transformations and for such transformations. since under a Lorentz transformation .80) and. (3.3 Transformation of bilinears In the last section. Furthermore. therefore. 2 2 S(ω) = cosh (3.83) That is. we have shown how to construct the matrix S(Λ) for ﬁnite Lorentz transformations (for both rotations and boosts). 1 S † (ω) = S −1 (ω). as 4 × 4 matrices). 0 ½ (3.81) we can determine ω ω − α1 sinh . ω.82) We note here that since α† = α1 . Let us note next that. All the unitary representations are necessarily inﬁnite dimensional.79) α1 = . in this four dimensional space (namely. (3. 3. α2 = 1 ½ 0 = ½.82 3 Properties of the Dirac equation In this case. recalling that 0 σ1 σ1 0 1 (3. operators deﬁning boosts are not unitary. can be obtained through exponentiation as S(ω) = e− 2 α1 ω . the matrix for a ﬁnite boost. there does not exist any ﬁnite dimensional unitary representation.

we see that the adjoint wave function ψ(x) transforms inversely compared to the wave function ψ(x) under a Lorentz transformation. it would transform as a four vector under a proper Lorentz transformation. (3. it follows that ψ ′ † (x′ ) = ψ † (x)S † (Λ). 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).43).84) (3. In other words. (3.85) where we have used the relation (3.86) ′ ψ(x)ψ(x) → = Namely. namely.3. 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). 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 (2.56). Similarly.50) and (2.3 Transformation of bilinears 83 ψ ′ (x′ ) = S(Λ)ψ(x). J µ (x) = ψ(x)γ µ ψ(x) → Λµν J ν (x). ψ (x′ ) = ψ ′ † (x′ )γ 0 = ψ † (x)S † (Λ)γ 0 = ψ † (x)γ 0 S −1 (Λ) = ψ(x)S −1 (Λ).55)). Thus.87) ′ where we have used (3.88) . (3. ′ (3. we see that if we deﬁne a current of the form j µ (x) = ψ(x)γ µ ψ(x).

Namely. It is customary to identify (see (2. Finally. we note that in this way.90).84 3 Properties of the Dirac equation This is.91). can be written as (p − m)u(p) = 0.89) and (3.93) . what we had observed earlier. −p) = u− (−p) = . E+m 2m σ ·p E+m v ˜ v ˜ v(p) = u− (−p0 . 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 . 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) = . of course. 3.49). / where p0 = ω = p2 + m2 . (3. (3.92) so that the equations satisﬁed by u(p) and v(p) (positive and negative energy solutions). / (3. p) = 0.91) with the same value of p0 as in (3. we can determine the transformation properties of the other bilinears under a Lorentz transformation in a straightforward manner. (3. (3. the probability current density (see also (2.4 Projection operators. p) = 0.90) (3.89) while the negative energy solutions satisfy − γ 0 p0 − γ · p − m u− (−p0 .85)) transforms like a four vector so that the probability density transforms as the time component of a four vector.

95) where we have used (γ µ )† γ 0 = γ 0 γ µ (see (2. can represent the spin projection of the two component spinors (in terms of which the four component solutions were obtained).96) where r. / or. r = 1. 3. Let us also note that each of the four solutions really represents a four component spinor. / v † (p) (p)† + m γ 0 = 0. Let us denote the spinor index by α = 1. 4. / or. u(p)(p − m) = 0. v(p)(p + m) = 0. completeness relation 85 and − γ 0 p0 + γ · p − m v(p) = 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. With these notations. / (3. As we have seen earlier there are two positive energy solutions and two negative energy solutions of the Dirac equation. as we had seen earlier.97) . (3.3. (3.94) Given these equations. or. / (3. u (p)v (p) = 0 = v r (p)us (p). (−p − m)v(p) = 0. we can write down the Lorentz invariant conditions we had derived earlier from the normalization of a massive Dirac particle as (see (2. or.50)) 4 u r (p)us (p) = α=1 4 r uα (p)us (p) = δrs .83)). 2. Let us denote them by ur (p) and v r (p). 2. / (p + m)v(p) = 0. α v r (p)v s (p) = α=1 r s r s v α (p)vα (p) = −δrs .

4 × 4 matrices and their eﬀect on the Dirac spinors is quite clear. (3.100) Λ− (p) = 2m These are. (3.97) can be checked to be true simply because v(p) = u− (−p0 . (3. From the form of the equations satisﬁed by the positive and the negative energy spinors.99) for any two spin components of the positive and the negative energy spinors. 2m −p + m r / v (p) 2m −p − m + 2m r / v (p) = v r (p). −p). because the direction of momentum changes for v(p) (see the derivation in (2.98) For completeness we note here that it is easy to check u† (p)v(−p) = 0 = v † (−p)u(p). 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). it is clear that we can deﬁne projection operators for such solutions as p+m / . This also allows us to write 2 r=1 [u r (p)ur (p) − v r (p)v r (p)] = 4. 2m −p + m / .101) . 2m (3. namely. 2m p+m r / v (p) = 0. (3.52)).86 3 Properties of the Dirac equation Although we had noted earlier that u+ (p)u− (p) = 0.93) and (3. the last relation in (3. of course.94). 2m −p + m r / u (p) = 0.

(3. 2m 2m Λ+ (p) + Λ− (p) = (3. we see that / Λ± (p) are indeed projection operators and they are orthogonal to each other. completeness relation 87 Similar relations also hold for the adjoint spinors and it is clear that Λ+ (p) projects only on to the space of positive energy solutions. 2m / p + m −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). while Λ− (p) projects only on to the space of negative energy ones. 2m −p + m / 2m 2 = p2 − 2mp + m2 / 4m2 2m(−p + m) / m2 − 2mp + m2 / = 2 4m 4m2 −p + m / = Λ− (p).3. let us also note that p + m −p + m / / + = ½.4 Projection operators. Thus. Furthermore.103) as it should be since all the solutions can be divided into either positive or negative energy ones.102) where we have used (p)2 = p2 and p2 = m2 . Let us deﬁne a 4 × 4 matrix P with elements . Let us next consider the outer product of the spinor solutions.

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. (3. 2 or. r=1 r ur (p)uβ (p) = α p+m / 2m . α β=1 r=1 = (3. 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).88 3 Properties of the Dirac equation 2 Pαβ (p) = r=1 r ur (p)uβ (p). 3. β = 1.105) Thus. αβ (3.104) This matrix has the property that acting on a positive energy spinor it gives back the same spinor. 2. α α 4 s uβ (p)Pβα (p) β=1 2 2 s r uβ (p)ur (p)uα (p) β β=1 r=1 r s δrs uα (p) = uα (p). we can identify Pαβ (p) = (Λ+ (p))αβ . Namely. we see that the matrix P projects only on to the space of positive energy solutions and. α α. if we deﬁne . therefore.106) Similarly. 4.

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 . 2 or. the matrix Q projects only on to the space of negative energy solutions with a phase (a negative sign). (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) = 0.103)) . 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).107) then. Hence we can identify Qαβ = −(Λ− (p))αβ . r=1 r r vα (p)v β (p) = p−m / 2m . (3. αβ (3. completeness relation 89 2 Qαβ (p) = r=1 r r vα (p)v β (p).3.4 Projection operators.108) Namely.109) The completeness relation for the solutions of the Dirac equation now follows from the observation that (see (3.

M (3. If the initial and the ﬁnal states are the same. the completeness relation (3.111) can be understood as follows. In such a space. 2 or. (3.110) or in (3. rather it involves a sum with the metric structure of the space built in. 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. let us suppose that we are interested in a transition amplitude which has the form u r (p)M us (p′ ). the completeness relation does not involve a sum of terms with positive deﬁnite sign.113) If we are not interested in the expectation value in a particular electron state. α (3. Hence.110) can also be written as 2 r=1 (ur (p)u r (p) − v r (p)v r (p)) = ½. the latter being negative while the former is positive.90 3 Properties of the Dirac equation Pαβ − Qαβ = (Λ+ (p))αβ + (Λ− (p))αβ = δαβ . we can think of the space of solutions of the Dirac equation as an indeﬁnite metric space. r=1 r r r ur (p)uβ (p) − vα (p)v β (p) = δαβ . (3. For example. These relations are particularly useful in simplifying the evaluations of transition amplitudes and probabilities.111) We note here that the relative negative sign between the two terms in (3.112) where M stands for a 4 × 4 matrix. uu and vv have opposite sign. As we have seen. but rather wish to obtain an average over the two possible .

β.4 Projection operators.σ.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.s=1 1 2 4 2 4 r s uα (p)Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α. if we have a transition from a given electron state to another and if we are interested in a process where we average over the initial electron states and sum over the ﬁnal electron states (for example. 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.s=1 σλ r ur (p)uα (p) λ σλ = ur (p) λ = .3. 2 (3.114) Similarly. the probability for such a transition will be determined from 1 u r (p)M us (p′ ) 2 r.s=1 2 2 u r (p)M us (p′ ) † = 1 u r (p)M us (p′ ) 2 r. think of an experiment with unpolarized initial electron states where the ﬁnal spin polarization is not measured).s=1 1 2 r.λ=1 r.β=1 4 2 r Mαβ ur (p)uα (p) β α.σ.β=1 r=1 4 = = = Mαβ (Λ+ (p))βα = α.β=1 1 Tr M Λ+ (p).β.λ=1 2 s Mαβ us (p′ )uσ (p′ ) γ 0 M † γ 0 β α. completeness relation 91 electron states (in experiments we may want to average over the spin polarization states).

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

120) Therefore. u(p. H] = [Si . Namely. 4 4 (3. This is known as the helicity operator and we note that since the Hamiltonian commutes with helicity. can only be ± 2 and we can label the positive and the negative energy solutions also as u(p. H]pi = −iǫijk αj pk pi = 0. h)v(p. for a Dirac parti1 cle. h) with h = ± 1 (the two helicity 2 eigenvalues). h′ ). H] = [pi . α · p + βm] = 0. v(p. Note that S·p |p | 2 h2 = = 1 ½. (3.5 Helicity 93 [pi .123) Therefore.121) measures the longitudinal component of the spin of the particle or the projection of the spin along the direction of motion. h′ ). The normalized operator h= S·p . h)v(p. (3. the eigenvalues of the helicity operator.124) . |p | (3. (3. the eigenstates of energy can also be labelled by the helicity eigenvalues. The normalization relations in this case will take the forms u(p.3. h)u(p. therefore. h). the order of these operators in the product is not relevant).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)u(p. 4 (3. h′ ) = δhh′ = −v(p. this operator is a constant of motion. h′ ) = 0 = v(p. [Si pi .119) the operator S · p does also (momentum and spin commute and.

h)v(p. (3.129) p0 u2 (p) − σ · p u1 (p) = −mu2 (p). γ 0 p0 − γ · p − m u(p) = 0. which can also be written as p0 u1 (p) − σ · p u2 (p) = mu1 (p).126) where we are not assuming any relation between p0 and p as yet. h)u(p. this leads to the two (2-component) coupled equations p0 − m u1 (p) − σ · p u2 (p) = 0. (3. h)] = ½. (3.110) or (3. In terms of u1 (p) and u2 (p).111). Let us represent the four component spinor (as before) as u1 (p) u2 (p) u(p) = .94 3 Properties of the Dirac equation Furthermore. 1 2 particle. h) − v(p.127) where u1 (p) and u2 (p) are two component spinors. or.6 Massless Dirac particle Let us consider the free Dirac equation for a massive spin (γ µ pµ − m) u(p) = 0. can now be written as [u(p.125) 1 h=± 2 3. the Dirac equation takes the form (p0 − m)½ σ·p −σ · p u1 u2 −(p0 + m)½ = 0. (3. (3. (3. σ · p u1 (p) − p0 + m u2 (p) = 0. the completeness relation.130) . (3.128) Explicitly.

The corresponding two component spinors uL and uR are also known as Weyl spinors. We note that if we deﬁne two new (2-component) spinors as 1 (u1 (p) − u2 (p)) . we obtain p0 − σ · p (u1 (p) + u2 (p)) = m (u1 (p) − u2 (p)) .132) then.133) reduce to two (2-component) spinor equations which are decoupled and have the simpler forms p0 uR (p) = σ · p uR (p). These equations. are not invariant under parity or space reﬂection and are known as the Weyl equations. 2 1 (u1 (p) + u2 (p)) . like the Dirac equation. the equations in (3. 2 uL (p) = uR (p) = (3.131) can be rewritten as a set of two coupled (2-component) spinor equations of the form p0 uR (p) − σ · p uR (p) = muL (p). Let us note that in the limit m → 0. can be shown to be covariant under proper Lorentz transformations (as they should be. (3. since vanishing of the mass which is a Lorentz scalar should not change the behavior of the equation under proper Lorentz transformations). .6 Massless Dirac particle 95 Taking the sum and the diﬀerence of the two equations in (3. however. p0 uL (p) = −σ · p uL (p). (3.134) These two equations.130). p0 uL (p) + σ · p uL (p) = muR (p). (3.131) p0 + σ · p (u1 (p) − u2 (p)) = m (u1 (p) + u2 (p)) .3.133) We note that it is the mass term which couples the two equations. the two equations in (3.

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

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

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

see. (3. This can also be seen from the fact that the Hamiltonian for a massless Dirac fermion (see (1.149) commutes with γ5 (in fact. (2.151) . γ5 uR (p) = uR (p).53) and (2.147) u(p) = E 2 E 2 = |p| 2 |p| 2 u(p) ˜ σ·p u(p) .3. (3. (3. in the Pauli-Dirac representation γ5 = ρ deﬁned in (2.7 Chirality With the normalization for massless spinors discussed in (2.60) and ρ commutes with α. then γ5 u(p) (or γ5 v(p)) is also a solution.150) it follows that the eigenvalues of γ5 are ±1 and spinors with the eigenvalue +1. the solutions of the massless Dirac equation can be classiﬁed according to the eigenvalues of γ5 also known as the chirality or the handedness.148) From the structure of the massless Dirac equation (3. Since 2 γ5 = ½. namely.100)) H = α · p.61)).92)) u(p) ˜ σ·p u(p) ˜ E E (3. ˜ |p| v(p) = σ·p v (p) ˜ v (p) ˜ = σ ·p v (p) ˜ |p| v (p) ˜ .147). we note that if u(p) (or v(p)) is a solution. / / can be written as (see also (3. for example. γ5 vR (p) = vR (p). (3. the solutions of the massless Dirac equation (m = 0) pu(p) = 0 = pv(p). Therefore.54).7 Chirality 99 3.

namely. then the Dirac Hamiltonian (1. 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) . We note that if the fermion is massive (m = 0).152) are called left-handed (negative chirality) spinors. (3.153) where we have deﬁned . vR (p) = PR v(p) = = |p| 2 1 2 1 2 1 (½ + γ5 )v(p) 2 ½ + σ ·p v(p) ˜ |p| . 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| . γ5 vL (p) = −vL (p). Given a general spinor. γ5 uL (p) = −uL (p).100 3 Properties of the Dirac equation are known as right-handed (positive chirality) spinors while those with the eigenvalue −1. = 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) ˜ .100) would no longer commute with γ5 and in this case chirality would not be a good quantum number to label the states with. (3.

these projection operators have the explicit forms (see (2. (3.156) We note from (3. namely. However. the four component spinors can be eﬀectively described by two component spinors. ˜ |p| 2 (3. 2 −½ ½ ½ ½ (3. we note that 2 spinors of the form χ(±) (p) = 1 σ·p 1± χ.121) (for the two component spinors S = 1 σ). which implies that any four component spinor can be uniquely decomposed into a right-handed and a left-handed component.119).7 Chirality 101 1 1 PR = (½ + γ5 ). PL = (½ − γ5 ). (3.153) that in the massless Dirac theory.6) that in the massless limit. This is connected with our earlier observation (see section 3. we will continue our discussion in the Pauli-Dirac representation of the Dirac matrices which we have used throughout. ˜ 2 |p| (3.155) PR PL = 0 = PL PR .157) correspond to states with deﬁnite helicity. PL 2 = PL .158) . PR + PL = 1. From the deﬁnition of the helicity operator in (3. The reducibility of the spinors is best seen in the Weyl representation for the Dirac matrices discussed in (2.154) 2 2 We note that by deﬁnition these projection operators satisfy PR 2 = PR . the Dirac equation reduces to two decoupled two component Weyl equations (recall that it is the mass term which generally couples these two spinors).91)) 1 2 PR = ½ ½ 1 ½ −½ . PL = . (In the Pauli-Dirac representation.3. hχ(±) (p) = σ·p 1 2|p| 2 ½± ½± 1 1 = ± × 2 2 σ·p χ ˜ |p| 1 σ·p χ = ± χ(±) (p).

70) or (3.153) and (3. vR (p) = .102 3 Properties of the Dirac equation so that the right-handed (four component) spinors in (3. therefore. ˜ ˆ 2 (3. uL (p) = . σ µ deﬁned in (2.160) can also be written in a covariant notation as P (+) = 1 µ σ pµ . Explicitly. (3.120) and pµ = (1. P (+) + P (−) = 1. deﬁne projection operators into the space of positive and negative helicity two component spinors. (P (−) )2 = P (−) . −ˆ ). (3.(3. |p| 2 P (−) = 1 µ σ pµ .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) = . 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 (−) = 0 = P (−) P (+) .153) are described by two component spinors with positive helicity while the lefthanded (four component) spinors are described in terms of two component spinors of negative helicity.161) with σ µ .162) and.159) We note here that the operators P (±) = 1 2 ½± σ·p 1 ˆ = (½ ± σ · p).117)) P (±) 0 1 2( P4×4 (±) = 0 P (±) 0 1 2( = ˆ ½ ± σ · p) 0 ˆ ½ ± σ · p) . we see from (3.163) . It is straightforward ˜ ˆ p to check that they satisfy the relations (P (+) )2 = P (+) . vL (p) = −v (−) (p) v (−) (p) . ˆ 2 (3.

For example.167). ˜˜ ˜˜ (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. 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 .164) which is the reason the spinors can be simultaneous eigenstates of chirality and helicity (when mass vanishes). For completeness as well as for later use. 0 v= ˜ 0 .165) 2 2 Each of these spinors is one dimensional and together they span the two dimensional spinor space. from (3.R .163) that PL. we can also deﬁne normalized spinors u(+) and v (+) . We note from (3. (3. we can choose u= ˜ 1 . let us derive some properties of these spinors.163) we see that the right-handed spinors with chirality +1 are characterized by helicity +1 while the left-handed spinors with chirality −1 have helicity −1. the normalized spinors take the forms (here we are using the three dimensional notation so that pi = (p)i ) u(+) (p) = 1 2|p|(|p| + p3 ) 1 2|p|(|p| − p3 ) |p| + p3 p1 + ip2 p1 − ip2 |p| − p3 . 1 (3. with the choice of the basis in (3.156) and (3. the helicity spinors simply reduce to eigenstates of σ3 .159) as well as (3. (3.167) such that when p1 = p2 = 0.166) For example. v (+) (p) = . Furthermore.3.168) . In fact.7 Chirality 103 and it is straightforward to check from (3. ˜ ˜ ˜ ˜ uu† + v v † = ½. P4×4 = 0. (±) (3.

In general. 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 . u(+) (p)u(+)† (p) = v (+) (p)v (+)† (p) = 1 2 ˆ ½ + σ · p .168). (3.169) which can be checked from the explicit forms of the spinors in (3. We note that with p0 = |p|. Given the form of the right-handed spinors in (3. 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). u(+)† (p)v (+) (−p) = 0 = v (+)† (−p)u(+) (p). the positive helicity spinors satisfy u(+)† (p)u(+) (p) = v (+)† (p)v (+) (p) = 1.171) = P (+) P (+) P (+) P (+) . R † uR (p)u† (p) = vR (p)vR (p) = R |p| 2 P (+) P (+) P (+) P (+) .170) The completeness relation in (3.159). we do not need to use any particular representation for our discussions. as well as (3.169) it now follows in a straightforward manner that † u† (p)uR (p) = |p| = vR (p)vR (p). (3. (3.170) can be simpliﬁed by noting the following identity. R † u† (p)vR (−p) = vR (−p)uR (p) = 0.104 3 Properties of the Dirac equation However.

8 Non-relativistic limit of the Dirac equation 105 so that we can write the completeness relation in (3.49)) u ˜ σ·p ˜ E+mu uL (p) uS (p) u+ (p) = E+m 2m = .8 Non-relativistic limit of the Dirac equation Let us recall that the positive energy solutions of the Dirac equation have the form (see (2.3. / 4 (3. L † uL (p)u† (p) = vL (p)vL (p) = L 1 0 pγ (½ − γ5 ). ˜ 2m σ·p E+m σ·p u= ˜ uL (p).172) which can also be derived using the methods in section 3. / 4 (3.174) we have deﬁned uL (p) = uS (p) = E+m u.176) .175) In (3. (3. 2m E + m E+m (3.4. L † u† (p)vL (−p) = 0 = vL (−p)uL (p).170) as † uR (p)u† (p) = vR (p)vR (p) = R 1 0 pγ (½ + γ5 ). 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).174) while the negative energy solutions have the form σ·p ˜ −E + m v v ˜ vS (p) vL (p) u− (p) = E+m 2m = . (3.173) 3.

In the non-relativistic limit. σ · p uL (p) = (E + m)uS (p). when |p | ≪ m. Let us next look at the positive energy solutions in (3. or. in (3. vL (p) and vS (p) are also known as the large and the small components of the negative energy solution. Similarly.179) We note that the second equation in (3. Similarly. (α · p + βm) u+ (p) = Eu+ (p). (3.175) we have denoted vL (p) = vS (p) = − E+m v. ˜ 2m σ·p E +m σ·p v=− ˜ vL (p).106 3 Properties of the Dirac equation and we emphasize that the subscript “L” here does not stand for the left-handed particles introduced in the last section.178) This would lead to the two (2-component) equations σ · p uS (p) = (E − m)uL (p). m½ σ · p −m½ σ·p uL (p) uS (p) =E uL (p) uS (p) .180) . or. the component uS (p) is much smaller than (of the order of v ) uL (p) and c correspondingly.174). (3. E+m uS (p) = (3. 2m E + m E+m (3. uL (p) and uS (p) are known as the large and the small components of the positive energy Dirac solution. which satisfy the equation Hu+ (p) = Eu+ (p).177) It is clear that in the non-relativistic limit.179) gives the relation σ·p uL (p). we expect the large components to give the dominant contribution to the wave function.

2m where we have used the fact that for a non-relativistic system. the ﬁrst equation in (3. equation (3.33) and remember that we are choosing = 1) .184) where e denotes the charge of the particle and Aµ represents the four vector potential of the associated electromagnetic ﬁeld. E ≈ m (recall that we have set c = 1). what we know for a free non-relativistic electron (spin 1 particle). This is. then. |p| ≪ m.9 Electron in an external magnetic ﬁeld The coupling of a charged particle to an external electromagnetic ﬁeld can be achieved through what is conventionally known as the minimal coupling.181) has the form p2 uL (p) = ENR uL (p). Furthermore. with the substitution of this.182) (3.9 Electron in an external magnetic field 107 while. (3. the Dirac equation in this case reduces to the Schr¨dinger o equation for a two component spinor which we are familiar with. σ·p uL (p) = (E − m)uL (p). and. 2m (3. E+m (3. therefore.3. This preserves the gauge invariance associated with the Maxwell’s equations and corresponds to deﬁning pµ → pµ − eAµ . 2 3.179) takes the form (σ · p) or. of course.181) p2 uL (p) ≃ (E − m)uL (p).183) Namely. if we identify the non-relativistic energy (without the rest mass term) as ENR = E − m. Since the coordinate representation of pµ is given by (see (1.

takes the form (α · (p − eA ) + βm) u(p) = Eu(p). we obtain .189) leads to σ · (p − eA) σ · (p − eA) uL (p) ≃ uL (p).187) where we are assuming that A = A(x). In this case. (3.189) uS (p) = (3.190) where in the last relation.189).186) Let us next consider an electron interacting with a time independent external magnetic ﬁeld.188) Explicitly. or. σ · (p − eA ) uL (p) = (E + m)uS (p). The Dirac equation for the positive energy electrons. the second equation in (3. m½ σ · (p − eA ) σ · (p − eA ) −m½ uL (p) uS (p) =E uL (p) uS (p) . we have used |p| ≪ m in the non-relativistic limit. (3.185) the minimal coupling prescription also corresponds to deﬁning (in the coordinate representation) ∂µ → ∂µ + ieAµ . we have A0 = 0 = φ. E+m 2m (3.108 3 Properties of the Dirac equation pµ → i∂µ . (3. (3. Substituting this back into the ﬁrst equation in (3. In this case. we can write the two (2-component) equations as σ · (p − eA ) uS (p) = (E − m)uL (p). in this case. B = (∇ × A ) .

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

In this case.196) We recognize (3.197) we obtain g = 2. in the non-relativistic limit. we know that the magnetic moments are given by . can have g-factors quite diﬀerent from 2. for the proton and the neutron.197) Of course.110 3 Properties of the Dirac equation ENR = E − m. a minimally coupled Dirac particle automatically leads.199) 1 Since S = 2 σ for a two component electron. (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. Particles with a nontrivial structure (that is particles which are not point like and have extended structures). Namely. 2m µ= (3. 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. comparing with (3.198) Let us recall that the magnetic moment of a particle is deﬁned to be (c = 1) e S. however. Thus. for example. one says that there is an anomalous contribution to the magnetic moment. to a magnetic dipole interaction (recall that in the non-relativistic theory. this shows that a point Dirac particle has a magnetic moment corresponding to a gyro-magnetic ratio g = 2. however. Quantum mechanical corrections (higher order corrections) in an interacting theory such as quantum electrodynamics.195) to be the Schr¨dinger equation for a charged o electron with a minimal coupling to an external vector ﬁeld along with a magnetic dipole interaction with the external magnetic ﬁeld. 2m µ=g (3. (3.

we have described how the non-relativistic limit of a Dirac theory can be taken in a simple manner.202) denotes the electromagnetic ﬁeld strength tensor and κ represents the anomalous magnetic moment of the particle. the Hamiltonian becomes non-Hermitian).200) where the nuclear magneton is deﬁned to be |e| . if we were to calculate the electric dipole interaction of an electron in a background electromagnetic ﬁeld using the method described in the earlier sections. the relevant expansion parameter is |p| and m the method works quite well in the lowest order of expansion. For example. at higher orders. This is commonly known as the Pauli coupling or the Pauli interaction.203) eκ µν σ Fµν . the electric dipole 1 moment becomes imaginary at order m2 (namely.10 Foldy-Wouthuysen transformation 111 µN µP = −1.201) with mP denoting the mass of the proton. The process of eliminating the “small” . 2mP µnm = (3. as we have seen explicitly. 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 .91 µnm .79 µnm . In the non-relativistic limit. However. (3. this method runs into diﬃculty. = 2. 2m (3.3. This puzzling feature can be understood in a simple manner as follows.10 Foldy-Wouthuysen transformation In the last two sections. (3. 3.

has the form (see (3.101)) H = α · p + βm = γ 0 (γ · p + m) . (3.204) where the matrix A.100) as well as (1.180)) σ·p . in the case of the free Dirac equation. 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.205) for the positive energy spinors.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).207). 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. for example. (3.207) Let us next look for a unitary transformation that will diagonalize the Hamiltonian in (3. for example. The matrix T that takes us to the two component “large” spinors in (3. such a transformation would .204) has the form T = ½ −A 0 ½ . E+m A= (3. In this case. let us look at the free Dirac theory where we know that the Hamiltonian has the form (see (1.112 3 Properties of the Dirac equation components from the Dirac equation described in the earlier sections can be understood mathematically as uL (p) uS (p) uL (p) AuL (p) uL (p) 0 u(p) = = −→ T .206) It is clear from the form of the matrix in (3. (3. Since the lack of unitarity in (3.206) is the source of the problem in taking the non-relativistic limit consistently. Thus.

212) . we note that (γ)† = −γ. (3. (γ · p)2 = −p2 = −|p|2 .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)! .91).3. avoid the problem with “large” and “small” spinors). 1 (3. (3. Let us consider a transformation of the type U (θ) = e 2m γ ·p θ .209) From the properties of the gamma matrices in (1.211) It follows now that U † (θ) = cos which leads to |p|θ 2m − γ ·p sin |p| |p|θ 2m . (3.208) where the real scalar parameter of the transformation is a function of p and m. θ = θ(|p|.83) or (1. m).10 Foldy-Wouthuysen transformation 113 also transform the spinor into two 2-component spinors that will be decoupled and we do not have to eliminate one in favor of the other (namely. γ · pθ 2m 2n+1 = cos γ ·p sin |p| |p|θ 2m (3.

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 .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 .213) × cos = cos 2 γ·p sin |p| 2 |p|θ 2m |p|θ 2m γ ·p |p| sin2 = cos2 + sin2 |p|θ 2m = ½.208) is indeed unitary. the free Dirac Hamiltonian (3. (3. the transformation (3.114 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. Namely.208). Under the unitary transformation (3.

θ.216) which. sin = p2 + m2 .214) that we can choose the parameter of transformation to satisfy |p|θ m = |p|θ m |p| cos or. we have |p|θ m m p2 + m2 |p|θ m |p| cos = .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). (3. However. |p|θ m m arctan |p| (3. from (3. has been arbitrary. (3. leads to the diagonalized Hamiltonian m2 p2 + m2 p2 p2 + m2 H ′ = γ0 + = γ0 p2 + m2 . tan θ= − m sin |p| . (3. = 0.217) that the Hamiltonian is now diagonalized in the positive and the negative energy spaces. it is clear from (3.215) In this case. our discussion has been quite general and the parameter of the transformation. 1 1 (3.10 Foldy-Wouthuysen transformation 115 So far. m |p| m . As a result.3. or.217) We see from (3. if we want the transformation to diagonalize the Hamiltonian.214). the two components of the transformed spinor u′ (p) 1 u′ (p) 2 u′ (p) = U (θ)u(p) = .219) .

214) that if we choose the parameter of transformation to satisfy |p|θ m |p|θ m |p|θ m m cos or. or.211) is also quite useful. We also note that with the parameter θ determined in (3.215).116 3 Properties of the Dirac equation For |p| ≪ m.221) this would lead to |p|θ m |p|θ m |p| cos sin = = − p2 + m2 m .183) to 1 the lowest order and it can be expanded to any order in m without any problem. . There is a second limit of the Dirac equation. tan + |p| sin =− m . the transformed Hamiltonian (3.222) p2 + m2 As a result. = 0. m (3. In this case. in this case. this leads to the non-relativistic equation in (3. the unitary transformation in (3. namely.214) will have the form . θ=− m arctan |p| (3. Let us note from (3. 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. the transformation is known as the Cini-Touschek transformation and is obtained as follows. (3. the ultrarelativistic limit |p| ≫ m. |p| m |p| . for which the generalized Foldy-Wouthuysen transformation (3.220) UFW (θ) = cos + which has a natural non-relativistic expansion in powers of |p| .211) takes the form 1 arctan 2 |p| m γ·p sin |p| 1 arctan 2 |p| .

220) as transforming away the α · p term in the Hamiltonian (3.11 Zitterbewegung 117 H ′ = γ0 = γ0 p2 p2 + m2 + m2 p2 + m2 γ·p |p| (3.11 Zitterbewegung The presence of negative energy solutions for the Dirac equation leads to various interesting consequences. we can think of the Foldy-Wouthuysen transformation (3. the unitary transformation (3. |p| m which has a natural expansion in powers of |p| . Therefore. where operators carry time dependence and states are time independent. in this case. the Heisenberg equations of motion take the forms ( = 1) . 3.3.224) m which clearly has a natural expansion in powers of |p| (ultrarelativistic expansion). In fact. let us consider the free Dirac Hamiltonian (1.223) p2 + m2 γ·p= |p| p2 + m2 α · p.207) while the Cini-Touschek transformation rotates away the mass term βm from the Hamiltonian (3.211) has the form UCT (θ) = cos 1 arctan 2 m |p| − γ ·p sin |p| 1 arctan 2 m . For example.207).225) In the Heisenberg picture.100) H = α · p + βm. (3. |p| (3.

101) we conclude that [α. As a result. it follows that (3. (3. αj pj + βm] i i 1 1 i j j [x .226) shows that the momentum is a constant of motion as it should be for a free particle. for example. by deﬁnition. The second equation in (3. it follows that the . Let us recall that. αj pj + βm] = 0.226) Here a dot denotes diﬀerentiation with respect to time.230) In other words.229) [α(t). α p ] = αj [xi . Furthermore.118 3 Properties of the Dirac equation 1 i 1 [x . i i ˙ or.227) where we have denoted the operator in the Schr¨dinger picture by o α(0) = α. The ﬁrst equation. H] = 0.101)). α(t) = eiHt α e−iHt . x = α. Secondly. H] = [xi . (1. xi = ˙ 1 1 i [p . using (1. even though the momentum of a free particle is a constant of motion. (3. α · p + βm] = 0. pi = ˙ (3. the velocity is not. = i ˙ or. identiﬁes α(t) with the velocity operator. p = 0. since the eigenvalues of α are ±1 (see. pj ] = i i 1 j ij α (iδ ) = αi . on the other hand. H] = [pi . H] = [α.228) (3.

(3. = 0.235) + + . we note that if α|ψ = λ|ψ . α(t) eiHt |ψ where we have identiﬁed |ψ ′ = eiHt |ψ . More explicitly. (3. (3.231) where λ denotes the eigenvalue of the velocity operator α. We note from Heisenberg’s equations of motion that the time derivative of the velocity operator is given by 1 [α(t). (3. = λ eiHt |ψ . αj αi . α(t)|ψ ′ = λ |ψ ′ . it follows that eiHt αe−iHt eiHt |ψ = λ eiHt |ψ .102)) αi .233) Equation (3.232) or. β = 2δij ½. These peculiarities of the relativistic theory can be understood as follows. since the electron is a massive particle. (3.11 Zitterbewegung 119 eigenvalues of α(t) are ±1 as well. then. This would seem to imply that the velocity of an electron is equal to the speed of light which is unacceptable even classically.234) Here we have used the relations (see (1. H] i + ˙ α(t) = = −i 2α(t)H − α(t). This is easily understood from the fact that the eigenvalues of an operator do not change under a unitary transformation.3. or. H = −2iα(t)H + 2ip. are ±1.232) shows that the eigenvalues of α(t) are the same as those of α (only the eigenfunctions are transformed) and. therefore.

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

In fact.242) integrating this over time. Let us also note from (3.239) and (3.11 Zitterbewegung 121 velocity which is oscillating at a very high frequency (for an electron at rest.226) that since ˙ x(t) = α(t) = p −2iHt p e .243) can be shown in the Schr¨dinger picture to arise from the presence of o negative energy solutions. The unconventional operator relations in (3. H x(t) = a + α(0) − (3. σ·p ˜ E+mu u(p) = we have (3.243) are again what we will expect classically for uniform motion.3.244) u† (p) α(0) − E+m = 2m = p u(p) H u† ˜ u ˜ σ·p ˜ σ E+mu p σ·p ˜ ˜ −E u + σ E + m u σ·p u† E + m ˜ p σ·p σu − E E + m u ˜ ˜ u† ˜ σ·p E+m σ p −E p −E E+m † u ˜ 2m . it is easy to check that for a positive energy electron state E+m 2m u ˜ . since there is no potential whatsoever in the problem. we obtain i p t+ H 2 p H −1 e−2iHt . + α(0) − H H (3. for example. the energy is ≈ . The ﬁrst two terms in (3.243) where a is a constant. However. This quivering motion of the electron was ﬁrst studied by Schr¨dinger o and is known as Zitterbewegung (“jittery motion”). Its occurrence is quite surprising.) and gives a time dependence to α(t). the third term represents an additional contribution to the electron trajectory which is oscillatory with a very high frequency.5MeV corresponding to a frequency of the order of 1021 /sec.

1964. McGraw-Hill. New York.246) as we should expect from the Ehrenfest theorem. This also shows that the eigenstates of the velocity operator. Drell. 1980. Zuber. 1968. J. Schiﬀ. Quantum Mechanics.12 References 1. 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. McGrawHill.) 3. α(t). 2.122 = 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. C. New York. Bjorken and S. I. McGraw-Hill. . (3. Relativistic Quantum Mechanics.245) 2E 2p p × + E E+m E+m u† u = 0. ˜ ˜ This shows that even though the operator relations are unconventional. 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. Itzykson and J-B. L. This shows that even though the eigenvalues of the operator α(t) are ±1 corresponding to motion with the speed of light. D. (This makes clear that neglecting the negative energy solutions of the Dirac equation would lead to inconsistencies. New York. Quantum Field Theory. in a positive energy electron state p H α(t) + = + .

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

This page intentionally left blank .

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

35) here for clarity. (4.) ω i j = ǫi jk αk .7) . of course. then. (4.3) (4. ǫijk denotes the three dimensional Levi-Civita tensor with ǫ123 = 1. ω µν in order to avoid confusion. etc.4) This is.34) and (1.126 4 Representations of Lorentz and Poincar´ groups e notation from (1. we note that ω ij = −ω ji . the form of the rotation that we had discussed in the last chapter.5) R(ω) = = ω ij xi ∂j = −ω ij xj ∂i . we can write the inﬁnitesimal rotations also as δxi = −R(ω)xi = ω i j xj . (4. ǫµν . (4.6) In other words. 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. ǫi jk = η iℓ ǫℓjk . but here we denote them by ω ij . It follows now that i R(ω)xi = −ω kj xj ∂k xi = −ω kj xj δk = −ω ij xj = −ω i j xj .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 .) If we now identify (In the last chapter we had denoted the inﬁnitesimal transformation matrices by ǫi j .

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

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

Jj ] = ǫij r Jr . then. ǫri q ǫjsr + ǫrsq ǫij r + ǫrjq ǫsi r = 0.13) is. so that µ R(ǫ)xµ = −ǫν ∂ν xµ = −ǫν δν = −ǫµ .13) 1 + ǫis q ǫjrs Mqr + 4 1 1 = − ǫriq ǫjsr Mqs − 4 4 1 − ǫri q ǫjsr Mqs − 4 or.1.14) which. namely. leads to (see (4.16) (4.17) (4.15) The algebra of the generators in (4. let us note that a constant in- ﬁnitesimal space-time translation of the form δxµ = ǫµ . = −ǫri q ǫjsr Mqs . the Lie algebra of SO(3) or SU (2) (or the familiar algebra of angular momentum operators) up to a rescaling.18) . where in the last step we have used the Jacobi identity for the structure constants of SO(3) or SU (2) (or the identity satisﬁed by the Levi-Civita tensors). (4. 2 (4. [Ji . of course.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. In the same spirit.11) or (4.12)) 1 q ǫri ǫjsr Mqs = − ǫijr ǫrsq Mqs = −ǫijr Jr . 4. (4. can be generated by the inﬁnitesimal vector operator R(ǫ) = −ǫµ Pµ = −ǫµ ∂µ .2 Translation.4.

∂ν ] = 0. (4. the inﬁnitesimal. (4. two translations commute and the corresponding relation for the generators is [Pµ . 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 . ν = 0.19) The Lie algebra associated with translations is then obtained from R(ǫ). (4. 3.18).24) then. (4. 2.130 and we can write 4 Representations of Lorentz and Poincar´ groups e δxµ = −R(ǫ)xµ .21) Namely.3 Lorentz transformation. 1.1. Pν ] = [∂µ . (4. R(ω) = (4. µ. we obtain . ∂ν ] = 0. As we have seen in the last chapter. constant parameters of transformation satisfy ω µν = −ω νµ .20) In other words. ǫ ν ∂ν ] = ǫµ ǫ ν [∂µ .23) As in the case of rotations. R(ǫ) = [ǫµ ∂µ . 4. 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 . as we have seen in (3. translations form an Abelian (commuting) group while the three dimensional rotations form a non-Abelian group.22) where.

as the generators of the inﬁnitesimal transformations. (4.4.27) As before. Mµν = −Mνµ .25) Therefore.1 Symmetry algebras 131 µ R(ω)xµ = −ω λν xν ∂λ xµ = −ω λν xν δλ = −ω µν xν = −ω µν xν = −δxµ . λ (4. we can think of R(ω) as the vector operator generating inﬁnitesimal proper Lorentz transformations and the operators. ω]µν . 1 µν 1 ω Mµν .28) = = ω µν xµ ∂ν = R(ω). R(ω) = = ω µν ω λρ [xµ ∂ν .26) M0i = −Mi0 = x0 ∂i − xi ∂0 = Ki .12)) Mij = −Mji = xi ∂j − xj ∂i = ǫij k Jk . as in the case of rotations. where. Thus. we have ω µν = −ω νµ = ω µλ ω λν − ω µ ω λν = [ω. we can determine the group properties of the Lorentz transformations from the algebra of the vector operators generating the transformations. 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 λρ Mλρ 2 2 R(ω). We also note that we can identify the inﬁnitesimal generators of spatial rotations with (see (4.29) . and the generators of inﬁnitesimal boosts with (4. (4.

gives the Lie algebra associated with Lorentz transformations. (Namely. In fact.30) This. xλ ∂ρ ] + [xν ∂µ .) We end this section by pointing out that the algebra in (2. we note that the number of generators for SO(4) which is (for SO(n). 2 (4.31) coincides exactly with the six generators we have (namely. 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 .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. xρ ∂λ ] − [xν ∂µ . therefore. xλ ∂ρ − xρ ∂λ ] = [xµ ∂ν .30) (up to a scaling). (4. This implies that. However. the matrices σµν provide a representation for the generators of the Lorentz group. it is n(n−1) ) 2 1 × 4 × (4 − 1) = 2 × 3 = 6. The algebra of the generators can also be calculated directly and has the form [Mµν . up to a scaling. . in this case.109) coincides with (4. therefore. the Lie algebra of the generators is isomorphic to that of the group SO(4). As we have seen these transformations correspond to rotations. 1). Lorentz transformations (boosts) are non-compact unlike rotations in Euclidean space. xλ ∂ρ ] − [xµ ∂ν . in four dimensions and.69) in connection with the discussion of covariance of the Dirac equation. This is what we had seen explicitly in (3. Mλρ ] = [xµ ∂ν − xν ∂µ . three rotations and three boosts). 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.

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

38). However.22) and (3. for example.30) and (4.57).) Thus. (4. Pµ behaves 1 like a covariant four vector. (4. 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. Pν ] = 0. Mλρ ] = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ . combining with our earlier results on the algebra of the translation group.30). ω µν = ω µλ ω λν − ω µλ ωλ ν = [ω. (4. Therefore. see.37) which simply shows that under a Lorentz transformation. We note that the algebra of translations deﬁnes an Abelian subalgebra of the Poincar´ algebra (4. xν ∂λ − xλ ∂ν ] = ηµν ∂λ − ηµλ ∂ν = ηµν Pλ − ηµλ Pν . For change in the coordinate four vector under an inﬁnitesimal Lorentz transformation. the only relation that needs to be calculated is the commutator between the generators of translation and Lorentz transformations. Mνλ ] = [∂µ . (4. we conclude that the Lie algebra associated with the Poincar´ e transformations (inhomogeneous Lorentz group) is given by [Pµ . (4. since the generae tors of translations do not commute with the generators of Lorentz . [Pµ . Note that [Pµ . ω]µν . Mνλ ] = ηµν Pλ − ηµλ Pν .21).36) We can also calculate the algebra of the generators of Poincar´ group. (4. (This is seen by recalling that 2 ω µν Mµν generates inﬁnitesimal Lorentz transformations. e We already know the commutation relations [Mµν . as well as the homogeneous Lorentz group. Mλρ ] as well as [Pµ . Pν ] (see (4.134 4 Representations of Lorentz and Poincar´ groups e where we have identiﬁed ǫµ = (ω µν ǫ ν − ω µν ǫν ) .38) [Mµν .21)).

42) .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. Jj ] = M0i . e (4.27) that we can identify the angular momentum and boost operators as 1 Ji = − ǫi jk Mjk . (4. it is what is known as a semi-direct sum of the two algebras. (The general convention is to denote groups by capital letters while the algebras are represented by lower case letters. Kj ] . i = 1. 2. Namely. − 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 . [Ki .12). Jj ] = ǫij k Jk .30) [Mµν . 3. 1). Kj ] = [M0i .40) We recall from (4.39) Rather. 2 Ki = M0i .41) Written out in terms of these generators. (4. the Lorentz algebra takes the form [Ki . M0j ] = −Mij = −ǫij k Jk . Mλρ ] = (−ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ ).2 Representations of the Lorentz group 135 transformations. Poincar´ algebra cannot be written as a direct sum e of those for translations and Lorentz transformations. (4.) 4. Poincar´ algebra = t4 ⊕ so(3.4.26) and (4. [Ji . (4.

44) Parenthetically. inﬁnite dimensional representations are unitary. namely.136 4 Representations of Lorentz and Poincar´ groups e where we have used (4.47) . (compared to Ji ) is connected with the fact that such transformations are non-compact and. (4. However. as can be seen from the hermiticity of the generators in the coordinate basis. Ki . let us note from the form of the algebra in (4. (Ji − iKi ) . the generators in the new basis are all anti-Hermitian. Ki† = Ki . (4. namely. i † Bi = −Bi . we have not been particularly careful about choosing Hermitian operators. † Mµν = i(xµ ∂ν − xν ∂µ ) = Mµν . This is a set of coupled commutation relations.45) This unconventional hermiticity for Ji arises because. Ki = −i (Ai − Bi ) . we have A† = −Ai . (4.13) in the last relation. The opposite hermiticity property of the generators of boosts. if we deﬁne the generators with a factor of “i”. (4.42) that we can assign the following hermiticity properties to the generators. As a consequence of (4. consequently.45).46) Namely. (4. in choosing the coordinate representation for the generators. Ji† = −Ji . 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 ) .43) which also leads to the inverse relations Ji = (Ai + Bi ) . the ﬁnite dimensional representations of boosts are non-unitary (hence the opposite Hermiticity of Ki ).

43). Jj ] − i [Ji . Jj ] + [Ki . Bj ] = = = = [Ai . Jj ] − [Ki . Jj ] − i [Ji . Kj ] + i [Ki . in this new basis. Jj ] + i [Ji . 2 ij 1 1 (Ji + iKi ) . 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 . Bj ] = = 1 ([Ji . 2 ij 1 1 (Ji − iKi ) . Jj ] − [Ki . the Lorentz group is double valued (doubly connected). Mathematically. Kj ] − i [Ki . (Jj − iKj ) 2 2 [Ai .49) Incidentally. the Lorentz algebra (4. one says that the Lorentz algebra is isomorphic to the direct sum of two angular momentum algebras. Therefore. (4. so(3. Kj ] + i [Ki . Kj ]) 4 1 ǫ k Jk − iǫij k Kk + iǫij k Kk − ǫij k Jk = 4 ij = 0.42) takes the form 1 1 (Ji + iKi ) . (Jj − iKj ) 2 2 1 ([Ji . Aj ] = = = = [Bi . (Jj + iKj ) 2 2 1 ([Ji . 1) ≃ so(3) ⊕ so(3) ≃ su(2) ⊕ su(2).2 Representations of the Lorentz group 137 In the new basis (4.48) In other words. the algebra separates into two angular momentum algebras which are decoupled. it is more .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 . (4. as we have already seen in the last chapter.

2 A δi = i i i θ − θB . .) (jA .138 4 Representations of Lorentz and Poincar´ groups e meaningful to consider the simply connected universal covering group of SO(3. can be speciﬁed uniquely once we know the values of jA and jB and is labelled as D(jA .53) . 2 A (4. The ﬁnite dimensional unitary representations of each of the angular momentum algebras is well known.jB) (Λ)ψ (jA . Denoting by jA and jB the eigenvalues of the Casimir operators A2 and B2 respectively for the two algebras. 1.35) in connection with the covariance of the Dirac equation.51) where Λ represents the ﬁnite Lorentz transformation parameter.50) An irreducible nonunitary representation of the homogeneous Lorentz group. we have 3 1 jA = 0. this represents the operator implementing ﬁnite transformations on the Hilbert space of states or wave functions as ψ ′ (jA . 2 2 3 1 jB = 0.jB ) (x) = D(jA ) (Λ)D(jB ) (Λ)ψ (jA . . 1) which is SL(2. we can write (This is the generalization of the S(Λ) matrix that we studied in (3. ) to describe the Lorentz transformations. Namely. .52) where the ﬁnite parameters of rotation and boost can be identiﬁed with θi = 1 i i θ + θB . 2 2 (4. . . .jB ) (just as the representation of the rotation group is denoted by D(j) ).j ) (j .jB ) “ ” (j ) (j ) i i −i θA Ai A +θB Bi B “ ” (j .jB ) (x). therefore. . 1. . . . Explicitly. . (4.j ) −i θ i Ji A B +δi Ki A B D (Λ) = e = e . much the same way we consider the universal covering group SU (2) of SO(3) to describe rotations. (4.jB ) (x′ ) = D(jA . .

(4. j = 0. dimensionality = 2. jB = 0. For jA = jB = 0. dimensionality (4. dimensionality = 1.54) and its spin content follows from the fact that (see (4. 1 Similarly.jB ) = (2jA + 1) (2jB + 1) . jB ) will have the dimensionality D (jA . .55) Consequently.2 Representations of the Lorentz group 139 Such a representation labelled by (jA .59) also corresponds to a two component spinor representation with spin 1 2 .0) . D(0. jB ) can lie between j = |jA − jB |. j = . jA + jB . For jA = 1 . 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.4. jB = 2 . (4. in fact. for jA = 0. . dimensionality = 2.56) The ﬁrst few low lying representations of the Lorentz group are as follows. we conclude that the values of the spin in a given representation characterized by (jA . (4. .57) corresponds to a scalar representation with zero spin (which acts on the wave function of a Klein-Gordon particle). 2 1 1 D( 2 . 1 1 D(0. These two representations are inequivalent and. from our knowledge of the addition of angular momenta. For jA = jB = 2 . 2 (4. . 2 ) .58) 1 corresponds to a two component spinor representation with spin 2 . . 2 (4. j = .44)) J = A + B.0) . |jA − jB | + 1.

we note that for the ﬁrst few low order represen1 tations.1 Similarity transformations and representations. j = 0.) From (4. j = . 2 (4. 4.60) is known as a four component vector representation and can be identiﬁed with a spin content of 0 and 1 for the components. To compare with the results that we had derived earlier. = 0.50).47). 2 ) ⊗ D ( 2 . (4. Let us now con- struct explicitly a few of the low order representations for the generators of the Lorentz group.48). 2 ) .61) This discussion can similarly be carried over to higher dimensional representations. t and a spin 1 component x (under rotations) and the same is true for any other four vector. It actually corresponds to a reducible representation of the Lorentz group of the form 1 1 1 D(0.62) Using (4.) It may be puzzling as to where the four component Dirac spinor ﬁts into this description. 2 ··· . dimensionality = 4. Ki . (Note that a four vector such as xµ has a spin zero component.0) (0) = −i Ai 2 − Bi Ki 2 = i σi . 1. Bi by a factor of i. Ai .0) (1) (0) Ji 2 = Ai 2 + Bi = − σi . we have (We note here that the negative sign in the spin 2 k representation in (4. ··· . namely. (Namely. this leads to the ﬁrst two nontrivial representations for the angular momentum and boost operators of the forms 1 ( 1 . 2 (4.44). 1 1 (4. we scale all the generators Ji .) Ai (0) = 0. 2 1 1 ( ) Bi 2 = − σi . Bi (0) 1 (1) Ai 2 = − σi . 2 (1) ( 1 .0) . dimensionality = 2 × 2 = 4.140 4 Representations of Lorentz and Poincar´ groups e D( 2 .62) arises because ǫij = −ǫijk in (4.63) . we now consider Hermitian generators by letting Mµν → iMµν as in (4.2.

1 ) 2 = 1 − 2 σi Ji (Dirac) 0 1 − 2 σi 0 . In fact. 1 ) ( 1 . 2 are inequivalent representations.71) and (3. 1 ) (0) Ji 2 = Ai + Bi 2 = − σi .63) and (4. (4.63) and (4. if there exists a similarity transformation relating the two. .64) we see that the condition (4.63) and (4.64) Equations (4.69) and (2. from (4. ( 1 . Two representations are said to be equivalent. if we can ﬁnd a similarity transformation S leading to (0.98) (or (3. 0 and 0.2 Representations of the Lorentz group 141 and 1 (1) (0.66) which is clearly impossible.0) (0.65) 1 then.64) give the two inequivalent representations of dimensionality 2 as we have noted earlier. They 2 provide the representations of angular momentum and boost for the left-handed and the right-handed Weyl particles. S −1 σi S = −σi . we would say that the two representations 2 . 2 (0. the two representations la1 belled by 1 . 1 ) ( 1 .67) However. (4. (Dirac) Ki i 2 σi 0 0 i − 2 σi .0) = Ji 2 ⊕ Ji 2 = = Ki ( 1 . 1 ) (1) (0) Ki 2 = −i Ai − Bi 2 i = − σi . 2 (4. 1 are 2 equivalent. From (4. we note that these do not resemble the generators of the Lorentz algebra deﬁned in (3. For example.78)). (4.65) would imply the existence of an invertible matrix S such that S −1 σi S = σi . we can obtain the representation of the Lorentz generators for the reducible four component Dirac spinors as (0. 1 ) Ki 2 = S −1 Ki 2 S.4.0) 2 ⊕ Ki (0. 0 and 0.0) Ji 2 = S −1 Ji 2 S. Therefore.64).

67) are obtained from 1 W σ . (4. 2 0i Therefore. we note that the generators in (4.119). 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 (4.121) 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 .70) = σi i − 2 σi 0 1 PD σ . γW = ǫijk 2 W . we can apply the inverse similarity transformation in (2.69) and (2. ij σW = i i j γ . we note that the angular momentum and boost operators in (4. give a representation of the Lorentz generators in the Weyl representation.γ = 2 W W −iσi 0 σk 0 0 iσi 0 σk 0i σW = .69) 2 µν and. (4. On the other hand.68) As a result. i 0 i γ .78)) are equivalent since they are connected by a similarity transformation that .98) (see also (3.71) and (3.142 4 Representations of Lorentz and Poincar´ groups e This puzzle can be understood as follows. if we would like the generators in the standard Pauli-Dirac representation (which is what we had used in our earlier discussions).67) and in our earlier discussion in (3. consequently. We note that in the Weyl representation for the gamma matrices deﬁned in (2.

as we discussed earlier. (Here we will use 3dimensional Euclidean notation without worrying about raising and lowering of the indices. (4. For example. Surprisingly.18)).4. there must be a similarity transformation relating the two representations.4). from the inﬁnitesimal change in the coordinates under a Lorentz transformation (see.) Representing the inﬁnitesimal change in the coordinates as δxi = iǫ (J3 )ij xj . (4. There is yet another interesting example which sheds light on similarity transformations between representation.119) to the standard Pauli-Dirac representation. The case for Lorentz transformations follows in a parallel manner.4) the matrix structure of the generator J3 to be 0 0 i 0 0 0 J3 = −i 0 0 .60)) and we can 2 construct the representations for J and K in this case as well from a knowledge of the addition of angular momenta. for example. Let us consider a three dimensional inﬁnitesimal rotation of coordinates around the z-axis as described in (3.2 Representations of the Lorentz group 143 relates the Weyl representation of the Dirac matrices in (2. Let us illustrate this for the simpler case of rotations.71) we can immediately read out from (3. from the Lie algebra point of view the four vector representation corresponds to jA = jB = 1 (see (4. the two representations for the generators constructed from two diﬀerent perspectives (for the same four vector representation) appear rather diﬀerent and. therefore. we can determine a representation for the generators of the Lorentz transformations belonging to the representation for the four vectors. (3. considering inﬁnitesimal rotations of the coordinates around the x-axis and the y-axis respectively.72) Similarly. we can deduce the matrix form of the corresponding generators to be . On the other hand.

let us note that the generators obtained (LA) from the Lie algebra are constructed by choosing the generator J3 See. A. the representation in the space of three vectors which would correspond to j = 1. Das (Hindustan Book Agency. Melissinos (Gordon and Breach).75) (LA) J1 (LA) J2 (LA) J3 which look really diﬀerent from the generators in (4. Quantum Mechanics: A Modern Introduction. 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 182 (note there is a typo in the sign of the 23 element for L2 in this reference) 1 . 2.72) and (4. j. On the other hand.144 4 Representations of Lorentz and Poincar´ groups e It can be directly checked from the matrix structures in (4. New Delhi). they are equivalent. 2 0 1 0 1 0 0 = 0 0 0 . therefore. This is. A. (4. Jj ] = iǫijk Jk . 3.74) J1 = 0 0 0 0 0 0 −i 0 i . (4. Das and A. C. 0 0 −1 0 −1 0 i = √ 1 0 −1 . therefore provide a representation for the generators of rotations. for example. This puzzle can be resolved by noting that there is a similarity transformation that connects the two representations and. page 289 or Lectures on Quantum Mechanics.73) and. k = 1.73) in spite of the fact that they belong to the same representation for j = 1.73) that they satisfy [Ji . in fact.72) and (4. i. 2 0 1 0 (4. To construct the similarity transformation (which actually is a unitary transformation).

Let us note from (4. −1 (4. 0 ψ (j3 =0) = 0 . 0 U† = 0 1 − √2 1 √ 2 i √ 2 0 i √ 2 −1 . it is straightforward to check S −1 J1 S = 0 1 − √2 1 2 = 0 1 − √2 √ 1 √ 2 i √ 2 0 i √ 2 i √ 2 0 i √ 2 √ 1 1 0 − √2 2 i i −1 0 0 i − √2 0 − √2 0 −i 0 0 0 −1 0 0 0 0 0 −i 0 −1 0 1 1 − √2 0 − √2 0 0 0 0 0 .4. 1 √ 2 i − √ 2 0 0 −1 1 − √2 U= 0 i − √2 .76) which will diagonalize the matrix J3 .2 Representations of the Lorentz group 145 to be diagonal. S = U.78) then.77) If we now deﬁne a similarity transformation (unitary) S −1 = U † . 2 ψ (j3 =1) = 0 ψ (j3 =−1) 1 −√ i2 = − √2 . 0 0 (4.76) 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 − √ . (4.

therefore.72). √ 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. related by a similarity transformation and. are equivalent. . 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 . (4.75) which look rather diﬀerent are.79) This shows explicitly that the two representations for J corresponding to j = 1 in (4. in fact.146 4 Representations of Lorentz and Poincar´ groups e = S −1 J2 S 0 1 0 1 (LA) √ 1 0 1 = J1 . 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 .73) and (4.

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

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

We also note that Mµν . (4. Pσ ] 2 1 µνλρ ǫ Pν (−ηλσ Pρ + ηρσ Pλ ) = 2 = 0. will also commute with the generators of translation.86) which follows from the the fact that momenta commute.3 Unitary representations of the Poincar´ group e 149 [W µ . Mστ 2 µν 1 λρ ǫ [Mλρ . Pσ ] = = 1 µνλρ ǫ Pν Mλρ . any function of Wµ and. Consequently. Mστ = = 1 λρ ǫ Mλρ . Pσ 2 1 µνλρ ǫ Pν [Mλρ . Mστ ] 2 µν 1 λρ ǫ (−ηλσ Mρτ − ηρτ Mλσ + ηλτ Mρσ + ηρσ Mλτ ) = 2 µν 1 1 1 1 = − ǫµνσρ Mρτ + ǫµντ ρ Mρσ + ǫµντ ρ Mρσ − ǫµνσρ Mρτ 2 2 2 2 ρ ρ = −ǫµνσ Mρτ + ǫµντ Mρσ = −ǫµνσρ − 1 ǫ δζ Mδζ 2 ρτ 1 + ǫµντ ρ − ǫρσδζ Mδζ 2 1 1 = − ǫµνσρ ǫτδζ ρ Mδζ + ǫµντ ρ ǫσδζ ρ Mδζ 2 2 1 ζ ζ δ δ ζ δ ζ ηµτ δν δσ − δσ δν + δµ δν ητ σ − δσ ητ ν = 2 δ δ ζ +δµ ηντ δσ − ηστ δν − σ ↔ τ Mδζ = 1 ηµτ Mνσ − Mσν + Mµν ητ σ − Mµσ ητ ν 2 +ηντ Mσµ − ηστ Mνµ − σ ↔ τ = 1 2ηµτ Mνσ − 2ηντ Mµσ − 2ηµσ Mντ + 2ηνσ Mµτ 2 . in particular Wµ W µ .4.

Mρσ ] Pλ 2 2 1 λρσ ǫ (ηµρ Wσ − ηµσ Wρ ) Pλ 2 ν (4. (4.89) In other words.87) simply says that under a Lorentz transformation. Let us note here. Mστ ] = Pν M µν . ǫµνλρ ǫστ ζρ = −ηµσ (ηντ ηλζ − ηνζ ηλτ ) − ηµτ (ηνζ ηλσ − ηνσ ηλζ ) −ηµζ (ηνσ ηλτ − ηντ ηλσ ) . Wν ] = = Wµ . 1 1 λρσ ǫν Mρσ Pλ = ǫνλρσ [Wµ . the operator Mµν behaves exactly like the generators of Lorentz transformation (see (4. Mστ ] M µν + Pν M µν . Using this. Namely. we can now evaluate [W µ . Namely. that [Wµ .30)).90) = −ǫµν λρ Wλ Pρ . then.150 4 Representations of Lorentz and Poincar´ groups e = −ηµσ Mντ − ηντ Mµσ + ηµτ Mνσ + ηνσ Mµτ . it behaves like a second rank antisymmetric tensor under a Lorentz transformation.88) Equation (4. it transforms like a vector which we should expect since it has a free Lorentz index. for completeness as well as for later use. 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. (4. Mστ = [Pν .87) Here we have used the identity satisﬁed by the four dimensional LeviCivita tensors. 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.

we choose the basis vectors of the repe resentation to be eigenstates of the momentum operators. the representations can be labelled by the eigenvalues of these operators. Namely. consequently.89) that [W µ Wµ . Pµ . Therefore. 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 ). labelled by the momentum eigenvalues. 4.3. of course.91) which is to be expected since Wµ W µ is a Lorentz scalar.93) However. satisfying . (4. without loss of generality.86)). it is easy to check that these do not commute with the generators of translation and. therefore. to be diagonal (they satisfy an Abelian subalgebra). Mµν M µν . L2 = Lµ Lµ with Lµ = P ν Mνµ . The irreducible representations of the Poincar´ group can be clase siﬁed into two distinct categories. Mστ ] W µ = W µ (ηµσ Wτ − ηµτ Wσ ) + (ηµσ Wτ − ηµτ Wσ ) W µ = Wσ Wτ − Wτ Wσ + Wτ Wσ − Wσ Wτ = 0.4. which we treat separately. Mστ ] = W µ [Wµ . The eigenstates of the momentum operators |p are.92) then. we conclude that if we deﬁne an operator W 2 = W µ Wµ . this would also represent a Casimir operator of the Poincar´ e algebra since Wµ commutes with the generators of translation (see (4. we can choose the momentum operators. In fact. (4.3 Unitary representations of the Poincar´ group e 151 It follows now from (4. There are other Lorentz scalars that can be constructed from Pµ and Mµν such as Mµν M µν . It can be shown that P 2 and W 2 represent the only Casimir operators of the algebra and.1 Massive representation. cannot represent Casimir operators of the algebra. Mστ ] + [Wµ . pµ . (4. To ﬁnd unitary irreducible representa- tions of the Poincar´ algebra.

where p 2 = p µ p µ = m2 .95) (4. (4. 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 . Therefore. P 2 |p = pµ pµ |p .97) Therefore. (4.152 4 Representations of Lorentz and Poincar´ groups e Pµ |p = pµ |p . namely.88).96) Here m denotes the rest mass of the single particle state and we assume the rest mass to be non-zero. We recall that Wµ = 1 µνλρ 1 ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . using (4. the eigenvalues of W 2 are not so obvious. 2 2 (4. let us study this operator in some detail.94) and in this basis. However. the eigenvalues of the operator P 2 = P µ Pµ are obvious.

98) where we have simpliﬁed terms in the intermediate steps using the anti-symmetry of the Lorentz generators.4. (4. takes the form 1 W 2 = − m2 M λρ Mλρ − m2 Mλ0 M 0λ 2 1 = − m2 2M 0λ M0λ + M ij Mij + m2 M 0λ M0λ 2 1 (4.100) = − m2 M ij Mij . To understand the meaning of this operator.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) can also be derived in an alternative manner which is simpler and quite instructive.102) The result in (4. 2 Recalling that (see (4. let us go to the rest frame of the massive particle. (4. 0) .99).26)) Mij = ǫij k Jk . 0. 0. In this frame. p 2 = p µ p µ = m2 . Let us note that in the rest frame (4. (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 .101) where Jk represents the total angular momentum of the particle. 2 (4. pµ = (m. the Pauli-Lubanski operator (4.99) and the operator W 2 acting on such a state.82) has the form .

Technically. all the degenerate states can be labelled not . we see that the representations with p2 = 0 can be labelled by the eigenvalues of the Casimir operators. That is. 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.105) Thus. 0. Therefore. s).103) where we have used (1.154 4 Representations of Lorentz and Poincar´ groups e W0 = Wi 1 0ijk ǫ Mjk pi = 0. therefore.102). Namely. for a massive particle. The dimensionality of the representation can also be understood in an alternative manner as follows. 0. we ﬁnd eigenvalues of W 2 : w2 = −m2 s(s + 1). It follows now that W 2 = W µ Wµ = W i Wi = m2 J i Ji = −m2 J2 . these would deﬁne an invariant subgroup of the Lorentz group and will lead to the degeneracy of states. we can ask what Lorentz transformations would leave such a vector invariant. rotations around the x or the y or the z axis will not change the time component of a four vector (recall that it is the spin 0 component of a four vector) and. for a massive particle at rest. 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). (m. It is not hard to see that all possible 3dimensional rotations would leave such a vector invariant.12). we can think of W 2 as being proportional to J 2 and in the rest frame of the particle. For a state at rest with momentum of the form pµ = (m. 0). (4. this simply measures the spin of the particle. Clearly. would deﬁne the stability group of such vectors. Therefore.104) which is the result obtained in (4. (4.34) as well as (4. 2 m 1 iνλρ ǫ Mλρ pν = − ǫijk Mjk = 2 2 m jk = − ǫi Mjk = mJi . 2 (4.

p = 0.83) we see that our states in the representation should also satisfy Pµ W µ = 0.3. · · · and 2 ms = s. Mij ] = 0. s − 1. −p). the representations for a massless partie cle are slightly more involved. commute with P0 . we would have (see (4.3 Unitary representations of the Poincar´ group e 155 just by the eigenvalue of the momentum. (4. In closing. we note that (see (4.30)) [P0 . but also by the eigenvalues of three dimensional rotations. Namely. The basic reason behind this is that the “little” group of a light-like vector is not so obvious. In contrast to the massive representa- tions of the Poincar´ group. acting on states in such a vector space.106) Namely.83)) P 2 = 0. namely. we note that (see (4. from (4. This can also be seen algebraically.4. a state at rest is an eigenstate of the P0 operator. let us note from (4. (4. 0. the three nontrivial Pauli-Lubanski operators correspond to the generators of symmetry of the “little group” in the rest frame.103) that. 4. · · · . up to a normalization factor. pµ = (p. This deﬁnes the 2s + 1 dimensional representation for a massive particle of spin s.107) Consequently. the eigenstates of P0 are invariant under three dimensional rotations and are simultaneous eigenstates of the angular momentum operators as well and such spaces are (2s+1) dimensional.108) However.95)) pµ pµ = 0. 1. 0. 1 . From the Lorentz algebra. (4.2 Massless representation. Consequently. −s.109) . −s + 1. which generate 3-dimensional rotations and are related to the angular momentum operators. (4. the operators Mij . In this case. s = 0.

113) W0 = Comparing with (4. for a massless particle if W 2 = 0. W µ has the form W µ = −hpµ . therefore. namely. 2 (4. we will not consider such representations. or. Such representations do not correspond to physical particles and.111) where h represents a proportionality factor (operator). namely.156 4 Representations of Lorentz and Poincar´ groups e There now appear two distinct possibilities for the action of the Casimir W 2 on the states of the representation.111) we conclude that in this space J·p .88)) 1 0ijk 1 ǫ pi Mjk = ǫijk pi ǫjkℓ Jℓ = −p · J = −J · p. (4. we can easily show that the action of W µ in such a space is proportional to that of the momentum vector. let us recall that Wµ = 1 µνλρ ǫ Pν Mλρ . namely. 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. On the other hand. it would lead to (see (4. W 2 = W µ Wµ = 0.110) In the ﬁrst case. W 2 = W µ Wµ = 0. in the second case where W 2 = 0 on the states of the representation.114) This is nothing other than the helicity operator (since L · p = 0) and. the simultaneous eigenstates of P 2 and W 2 would . (4. consequently. p0 h= (4.112) from which it follows that acting on a general momentum basis state |p (not necessarily restricting to massless states). 2 2 (4. To determine h. acting on states in such a space.

It follows now from (4. 2 1 3νλρ ǫ Mλρ pν = −pM12 = W 0 .115) that.3 Unitary representations of the Poincar´ group e 157 correspond to the eigenstates of momentum and helicity. namely.107). 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 . (4.83). We recognize that in this case. indeed represents the helicity operator up to a normalization as we had noted in (4.82) takes the form 1 0ijk 1 ǫ Mjk Pi = ǫijk Mjk pi = −pM12 . We also note from (4. (P0 − P3 ) |p = 2p|p . Let us recall that we are considering a massless state with momentum of the form pµ = (p. W 0 .113).115) We see from both (4.116) Let us now determine the dimensionality of the massless representations algebraically. 2 1 2νλρ ǫ Mλρ pν = −p (M01 − M13 ) . This can be seen intuitively from the fact that the motion of the particle is along the z axis. 2 2 1 1νλρ ǫ Mλρ pν = p (M02 − M23 ) . the Pauli-Lubanski operator (4. in the massless case.117) .4. (4. the set of Lorentz transformations which would leave this four vector invariant must include rotations around the z-axis. let us note here that in the light-like frame (4. but also algebraically by recognizing that a light-like vector of the form being considered is an eigenstate of the operator P0 − P3 . 0. as we had pointed out earlier.103) and (4. 2 W0 = W1 = W2 = W3 = (4. For completeness. −p) and we would like to determine the “little” group of symmetries associated with such a vector.115) that the Pauli-Lubanski operator indeed has only three independent components because of the transversality condition (4.

let us deﬁne two new operators as W2 . Π1 ] = [M12 . Π2 have the same commutation relations as those of translations in this two dimensional space. (4. [M12 . as we had also seen in the massive case. Π2 ] = [M12 . (4. the three independent Pauli-Lubanski operators are. the generators of symmetry of the “little” group. namely.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. M02 − M23 ] = 0. but can be easily understood as follows. M12 ] = 0. [Π1 . Π2 ] |p = [P0 − P3 . also deﬁne symmetries of light-like states. = [P0 − P3 . 2. [P0 − P3 . M02 − M23 ] |p = 2P2 |p = 0. M02 − M23 ] = −Π1 . Π2 ] = [M01 − M13 . therefore.38).121) Namely. M12 is the generator of rotations around the z axis or in the two dimensional plane and Π1 . This may seem puzzling. (4. M01 − M13 ] = Π2 . we say that the stability group or the “little” group of a light-like vector is E2 . 3 in (4. We note that the algebra of the symmetry generators takes the form [M12 . p W1 . We note from (4. These represent all the symmetries of the light-like vector (state). we see that e [P0 − P3 . (4. Π1 ] |p [P0 − P3 .158 4 Representations of Lorentz and Poincar´ groups e Furthermore. E2 (which consists of translations and rotation).118) so that rotations around the z-axis deﬁne a symmetry of the lightlike vector (state) that we are considering. from the Poincar´ algebra in (4. comparing with W i . it is isomorphic to the algebra of the Euclidean group in two dimensions. we see that up to a normalization. To determine the other symmetries of a light-like vector.90) that in the . M01 − M13 ] |p = 2P1 |p = 0. in fact. Thus. Clearly. Furthermore. i = 1.120) and.115).

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

spin can only be either parallel or anti-parallel to the direction of motion leading to the one dimensional nature of the representation. Academic Press. However. independent of the spin of the particle. Wigner. 2. Group Theory and its Applications to the Quantum Mechanics of Atomic Spectra. Incidentally. On the other hand. Wigner. 149 (1939). Annals of Mathematics 40. if the theory is not parity invariant. Then. Irreducible unitary representations of the Lorentz group.4 References 1. 568 (1947). if parity (space reﬂection) is a symmetry of the system. can be seen in a heuristic way as follows. Annals of Mathematics 48. New York (1959). the dimensionality of the representation will be one dimensional. 3.160 4 Representations of Lorentz and Poincar´ groups e lowest helicity states. E. Bargmann. On unitary representations of the inhomogeneous Lorentz group. it is clear that since a massless particle moves at the speed of light. the fact that the massless representations have to be one dimensional. V. E. the only consistent circular motion a massless particle can have. In other words. 4. . in general. is in a plane perpendicular to the direction of motion (otherwise. in such a case. Let us consider spin as arising from a circular motion. then we must have states corresponding to both the circular motions leading to the two dimensional representation. some component of the velocity would exceed the speed of light). as we have seen explicitly in the case of massless fermion theories describing neutrinos.

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

On the other hand. With this in mind. covariant theory for fundamental physical processes and. since the Hamiltonian is not a manifestly covariant concept. Given a relativistic dynamical equation. lim φ(x) → 0. therefore. However.1: Oscillating string ﬁxed at both ends. e the quantization of the ﬁelds as operators can then be carried out in the Hamiltonian formalism. free Klein-Gordon equation with mass m. We would like to have a relativistic. the classical free Klein-Gordon theory where we assume that the ﬁeld variable φ(x) ≡ φ(x. φ (x ) = φ(x). (5. it is generally preferrable to work in the Lagrangian formulation which is manifestly covariant. φ(x) = φ∗ (x). t) is real and that it behaves like a scalar under a Lorentz transformation. as we have seen (see (1. let us look at the simplest of ﬁeld theories.2) where φ (x ) denotes the Lorentz transformed ﬁeld.3) |x|→∞ The derivatives of the ﬁeld are also assumed to fall oﬀ at inﬁnity.4) .40)). whenever any ambiguity arises in the quantization. (5. We also assume that the ﬁeld falls oﬀ rapidly at spatial inﬁnity (this is necessary for the theory to be well deﬁned). (5.162 5 Free Klein-Gordon field theory Figure 5. has the form + m2 φ(x) = ∂µ ∂ µ + m2 φ(x) = 0. The classical. each ﬁeld must belong to a speciﬁc representation of the Lorentz group. namely. so far. one reverts back to the Hamiltonian formalism for its resolution. Secondly. the dynamical equation involving the ﬁeld variables must be covariant under Poincar´ transformations. namely.

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

10) Here we have used Gauss’ theorem (as well as the vanishing of ﬁeld variables asymptotically (5.9) 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) . subject to the boundary conditions (5. ti (5. 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.164 5 Free Klein-Gordon field theory most on the ﬁrst order derivatives of the ﬁeld variables. Note that under an inﬁnitesimal change tf (5. tf ) = 0. ti ) = δφ (x.

(5. tf ) .5. namely. the action will be stationary under an arbitrary change in the ﬁeld variable subject to the boundary conditions provided (namely.9).10) – vanishes because of the boundary conditions (5. ti ) = 0 = δφ (x. ∂µ φ(x)).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). δS = 0 only if).12) Thus. ∂φ(x) ∂∂µ φ(x) (5.15) . ∂L ∂L − ∂µ = 0. for the Klein-Gordon equation. As in point particle mechanics. inﬁnitesimal change in the ﬁeld variable. ∂µ ∂ µ φ + m2 φ = 0. (5.14) This is known as the Euler-Lagrange equation associated with the action S or the Lagrangian density L(φ. we can think of the dynamical equations as arising from the minimum principle associated with a given action. (5.13) Therefore. Thus. δφ (x.11) We note that the “surface term” – the last term in the expression (5. we conclude that under an arbitrary. the change in the action is given by tf δS = ti d4 x ∂L ∂L − ∂µ ∂φ(x) ∂∂µ φ(x) δφ(x). (5.

19) is manifestly Lorentz invariant. the Euler-Lagrange equation. Of course. ∂φ(x) ∂∂µ φ(x) or. in this case. 2 2 L= (5. ∂∂µ φ(x) Therefore. adding a total divergence to L also does not change the equations of motion. (5.19) gives the Klein-Gordon equation for a free. 2 2 L= then (5. This would simply correspond to adding a zero point energy.166 5 Free Klein-Gordon field theory we can ask whether there exists a Lagrangian density whose EulerLagrange equation will lead to this dynamical equation. ∂µ ∂ µ + m2 φ(x) = 0. . Note that if we choose 1 m2 2 ∂µ φ∂ µ φ − φ . Similarly. Note that the Lagrangian density L in (5. in this case.18) In other words.16) ∂L = −m2 φ(x). real scalar ﬁeld as its Euler Lagrange equation. if we assume the ﬁelds to fall oﬀ rapidly at inﬁnite separation (see (5.3)). the action does not change either.17) ∂L ∂L − ∂µ = −m2 φ(x) − ∂µ ∂ µ φ(x) = 0. gives (5. ∂φ(x) ∂L = ∂ µ φ(x). we see that the Lagrangian density 1 m2 2 ∂µ φ∂ µ φ − φ . It is worth noting here that one can add a constant to L without changing the equation of motion.

x} = 0 = {p. L = L (x. x) . (5.3 Quantization Since we have a Lagrangian description for the classical Klein-Gordon ﬁeld theory. 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. H} . In the case of a point particle. p} = 1. p) = px − L (x. ˙ (5.24) In the case of the classical scalar (Klein-Gordon) ﬁeld theory. lead to the dynamical equations in the Hamiltonian form as x = {x. ∂x ˙ p= (5. ˙ p = {p.20) The conjugate momentum associated with the dynamical variable x is deﬁned as ∂L . the basic variables are φ(x) and ∂µ φ(x) and we have from (5. {x.21) and this leads to the deﬁnition of the Hamiltonian in the form (this is an example of a Legendre transformation) H(x. in turn.3 Quantization 167 5.23) These. x) . p} .22) The classical canonical Poisson bracket relations between the coordinate and the conjugate momentum have the familiar form {x. ˙ (5. ˙ ˙ (5. H} . the Lagrangian is a function of the coordinate and the velocity of the particle (a dot denotes a derivative with respect to t).5. on the other hand.19) (xµ in this case merely labels space-time coordinates) .

(5.25) In an analogous manner to the point particle mechanics in (5.26) which will. ∂µ φ(x)) .27) H= (5. Π(y)}x0 =y0 . allow us to deﬁne a Hamiltonian density as ˙ H = Π(x)φ(x) − L.23)) {φ(x).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 (5.21). H} . we deﬁne a momentum canonically conjugate to the ﬁeld variable φ(x) as ∂L . Π(y)}x0 =y0 = 0 = {Π(x).30) Let us examine all of this in detail for the case of the free KleinGordon theory. (5.29) then. (5. (5. in turn. Note that the Lagrangian density in this case is given by . = δ3 (x − y). and a Hamiltonian d3 x H. H} . ˙ ∂ φ(x) Π(x) = (5. φ(y)}x0 =y0 {φ(x). the KleinGordon equation in this case) can be written in the Hamiltonian form as ˙ φ(x) = {φ(x). ˙ Π(x) = {Π(x).168 5 Free Klein-Gordon field theory L = L (φ(x). we can show that the dynamical equation (namely.

= δ3 (x − y).5.33) 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x). we can identify the conjugate momentum with ∂L ˙ = φ(x). (5.34) With the equal time canonical Poisson bracket relations deﬁned to be (see (5. φ(y)}x0 =y0 {φ(x). ˙ ∂ φ(x) Π(x) = (5.3 Quantization 169 L = = m2 2 1 ∂µ φ∂ µ φ − φ 2 2 1 ˙2 1 m2 2 φ .32) Consequently.35) . 2 2 2 which gives the Hamiltonian for the theory H = = d3 x H d3 x 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x) . Π(y)}x0 =y0 we obtain = 0 = {Π(x). Π(y)}x0 =y0 .29)) {φ(x).31) Therefore. this leads to the Hamiltonian density ˙ H = Π(x)φ(x) − L ˙ = Π(x)φ(x) − = Π(x)Π(x) − = 1 ˙2 1 m2 2 φ (x) φ + ∇φ · ∇φ + 2 2 2 1 m2 2 1 2 Π (x) + ∇φ · ∇φ + φ (x) 2 2 2 (5. φ − ∇φ · ∇φ − 2 2 2 (5. 2 2 2 (5.

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

Once we have the classical Hamiltonian description of a physical system. (Namely. but not the wave function. This is. we are supposed to treat φ(x) and Π(x) as Hermitian operators (because we are dealing with real ﬁelds.39) or. in the ﬁrst quantization we quantize x’s and p’s. ∂µ ∂ µ + m2 φ(x) = 0.5. we can expand the ﬁeld operator in this basis as .40) Note that this is the same Klein-Gordon equation as we had studied earlier in chapter 1. see (5.2)) satisfying the equal time commutation relations (remember = 1) [φ(x). here we are treating φ(x) as an operator which is quantized and not as a wavefunction as we had done earlier. the quantization of such a system is quite straightforward. Π(y)]x0 =y0 . This is conventionally referred to as second quantization. therefore. brings out the classical Hamiltonian description of the classical KleinGordon ﬁeld theory.40) (which is a second order Euler-Lagrange equation) and this discussion. In the quantum theory. (5. we can expand the ﬁeld operator φ(x) in such a basis. In the second quantization. we quantize the “wavefunction”. Π(y)]x0 =y0 = 0 = [Π(x). However. of course. φ(y)]x0 =y0 [φ(x). the free Klein-Gordon equation (1. therefore. (5. = iδ3 (x − y). or.4 Field decomposition Given any complete set of solutions of the classical Klein-Gordon equation.) 5.4 Field decomposition 171 These are ﬁrst order equations as Hamiltonian equations should be and from these. ¨ φ − ∇ 2 φ(x) + m2 φ(x) = 0. we obtain the second order equation ¨ ˙ φ(x) = Π(x) = ∇ 2 φ(x) − m2 φ(x). We already know that the plane wave solutions (1.41) of the KleinGordon equation deﬁne a complete basis and.

1 (2π) 3 2 3 2 ∂µ ∂ µ + m2 d4 k e−ik·x φ(k) = 0.41). We also emphasize here that the hermiticity properties of a given ﬁeld and its Fourier transform are.41) where we recognize that the operators φ(x) and φ(k) are Fourier transforms of each other (up to a multiplicative factor).42) using the ﬁeld expansion (5. in general. This shows that unless k2 = m2 .172 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 d4 k e−ik·x φ(k). φ(k) = 0. we can denote φ(k) = δ k2 − m2 a(k). Substituting this back into the expansion (5. we obtain 1 (2π) 3 2 φ(x) = d4 k δ k2 − m2 e−ik·x a(k).41) in the Klein-Gordon equation. we obtain 1 (2π) or.45) (5. (5. (5.43) d4 k −k2 + m2 φ(k)e−ik·x = 0. the Fourier transform has non-vanishing contribution (support) only on the massshell of the particle deﬁned by k2 = (k0 )2 − k2 = m2 . As a result. (5. diﬀerent. Namely. (5. (Recall that 1 k · x = k0 x0 − k · x and the factor (2π)3/2 is introduced for later convenience.44) where the operator a(k) is no longer constrained by the equation of motion.46) . (5.) Since φ(x) satisﬁes the Klein-Gordon equation ∂µ ∂ µ + m2 φ(x) = 0.

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

(5. we obtain 1 (2π) 2 3 φ(x) = d3 k −ik·x e a(k) + eik·x a(−k) . or.49) Changing k → −k in the second term in (5.50) Let us note that we are dealing with a Hermitian ﬁeld (see (5.51) Comparing the left-hand side and the right-hand side of (5. (5. 2k0 (5.49) and identifying k0 = Ek > 0.52) φ(x) = d3 k −ik·x e a(k) + eik·x a† (k) .53) . so that we can write 1 (2π) 3 2 a† (−k) = a(k). k) 0 x0 +ik·x = 1 (2π) 3 2 d3 k 1 0 e−iEk x +ik·x a(Ek . 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) . 2k0 (5. k) . 2k0 (5.51). k) 2Ek 0 +ik·x + eiEk x a(−Ek .2)).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 . φ† (x) = φ(x). we obtain a† (k) = a(−k). Therefore.

54) that a(k) and a† (k) are really functions of the three momentum k alone.5.55) and in terms of these operators we can write the ﬁeld operator (5.58) 5. (5.56) This unique decomposition of the ﬁeld operator φ(x) into positive and negative energy (frequency) parts is quite signiﬁcant as we will see in the course of our discussions. Thus.53) as d3 k (2π)3 2k0 φ(x) = e−ik·x a(k) + eik·x a† (k) . 2k0 (5. a(k) = √ 2k0 a† (k) a† (k) = √ . (5.54) It is clear from (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). conventionally one deﬁnes a(k) . (5.5 Creation and annihilation operators Given the ﬁeld expansion (5. eik·x a† (k).56) . φ(x) = φ(+) (x) + φ(−) (x). In fact.5 Creation and annihilation operators 175 Here we are supposed to understand that k0 = Ek = k2 + m2 > 0.57) so that we can write the ﬁeld operator as a sum of its positive and negative energy parts. (5.

60) 2(2π)3 The two deﬁning relations in (5. k0 is not an independent variable).59) and (5. (5. we note from these relations that we can write the operators a(k) and ˙ a† (k) in terms of φ(x) and Π(x) as (recall that Π(x) = φ(x)) a(k) = = a† (k) = = − 1 (2π)3 2k0 i (2π)3 2k0 1 (2π)3 2k0 i (2π)3 2k0 d3 x eik·x k0 φ(x) + iΠ(x) ← → d3 x eik·x ∂t φ(x). it is assumed that k0 = Ek (namely.61) that since the left-hand sides are independent of time. (5.60) are invertible. This can also be checked explicitly.61) ← → where ∂t is deﬁned in (1. the expressions on the righthand side must also be time independent.61) we obtain (neglecting the overall multiplicative factors) .61).176 5 Free Klein-Gordon field theory φ(x) = d3 k (2π)3 2k0 e−ik·x a(k) + eik·x a† (k) . d3 x e−ik·x k0 φ(x) − iΠ(x) ← → d3 x e−ik·x ∂t φ(x). (5. Namely.32) to be ˙ Π(x) = φ(x) = −i d3 k k0 e−ik·x a(k) − eik·x a† (k) .59) we obtain the expansion for the conjugate momentum in (5. It is important to emphasize here that in the expressions in (5. It is clear from (5. For example. taking the time derivative of the ﬁrst relation in (5.56).

we have used the fact that φ(x) satisﬁes the Klein-Gordon equation. a(k′ ) + eik·x+ik ·y a† (k). the time independence of the second expression in (5.61) are indeed time independent. φ(y)]x0 =y0 = ′ × e−ik·x−ik ·y a(k). Similarly.63) d kd k 3 3 ′ √ k0 k′ 0 2(2π)3 ′ × e−ik·x−ik ·y a(k). (5.40)) we obtain d3 k (2π)3 2k0 d3 k ′ (2π)3 2k′ 0 ′ [φ(x). must be time independent). a† (k′ ) = 0.3)) and have √ made the identiﬁcation k0 = Ek = k2 + m2 .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. in the intermediate steps. integrated the space derivatives by parts assuming that the surface terms vanish (see (5. Π(y)]x0 =y0 = − ′ ′ ′ x0 =y 0 (5. This shows explicitly that the integrals in (5.5. a(k′ ) + e−ik·x+ik ·y a(k). therefore. [Π(x). Imposing now the quantization relations between φ(x) and Π(x) (see (5. a† (k′ ) + eik·x−ik ·y a† (k).62) Here.61) can also be derived in a simple manner (or note that the second relation is the Hermitian conjugate of the ﬁrst and. a(k′ ) − e−ik·x+ik ·y a(k). a† (k′ ) .

′ ′ x0 =y 0 (5. we can deduce the fundamental commutation relations between the coeﬃcients of expansion to be (This nice form arises because of the redeﬁnition (5.65) From these. a† (k′ ) + eik·x−ik ·y a† (k).) a(k).61). a(k′ ) + eik·x+ik ·y a† (k). Π(y)]x0 =y0 = − ′ ′ x0 =y 0 (5. a† (k′ ) . Any other redeﬁnition will introduce a multiplicative factor into the commutation relations. However. Thus.178 ′ 5 Free Klein-Gordon field theory − eik·x−ik ·y a† (k). it appears that there is an inﬁnite number of such operators in the present case – one for every value of the momentum k. a† (k′ ) = 0. [φ(x). a† (k′ ) = δ3 k − k′ . a(k′ ) = 0 = a† (k). a(k′ ) − eik·x+ik ·y a† (k). The commutation relations in (5. a(k).66) can also be obtained more directly from the inversion formulae in (5. a† (k′ ) = ′ d3 xd3 y √ eik·x−ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x). a(k′ ) − e−ik·x+ik ·y a(k). (5. for example. we have a(k). which are the coeﬃcients of expansion of the ﬁeld operator in a plane wave basis. a† (k′ ) = iδ3 (x − y).55). recalling that the integrals are ime independent. have commutation relations analogous to the annihilation and creation operators of a harmonic oscillator.66) This shows that the operators a(k) and a† (k). k′ 0 φ(y) − iΠ(y) x0 =y 0 .64) i (2π)3 d3 k d3 k ′ k′ 0 4k0 ′ × e−ik·x−ik ·y a(k).

a(k′ ) = d3 xd3 y ′ √ eik·x+ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(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′ ) = 0.5. k′ 0 φ(y) + iΠ(y) = d3 xd3 y ′ √ eik·x+ik ·y 3 4k 0 k ′ 0 (2π) × ik0 [φ(x). φ(y)] x0 =y 0 = = = (k0 + k′ 0 ) √ 4k0 k′ 0 (k0 + k′ 0 ) √ 4k0 k′ 0 d3 xd3 y ik·x−ik′ ·y 3 e δ (x − y) (2π)3 d3 x i(k−k′ )·x e (2π)3 x0 =y 0 (k0 + k′ 0 ) i(k0 −k′ 0 )x0 3 √ δ (k − k′ ) = δ3 (k − k′ ). e 4k0 k′ 0 a† (k).67) where we have used the fact that for k′ = k.66) can also be derived in the same manner a(k).5 Creation and annihilation operators 179 = d3 xd3 y ′ √ eik·x−ik ·y (2π)3 4k0 k′ 0 × −ik0 [φ(x). a† (k′ ) = ′ d3 xd3 y √ e−ik·x−ik ·y 3 4k 0 k ′ 0 (2π) . e 0 k′ 0 4k (5. Π(y)] + ik′ 0 [Π(x). The other two commutation relations in (5. Π(y)] + ik′ 0 [Π(x). we have k′ 0 = k0 .

68) where we have used the fact that k′ 0 = k0 for 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). we have used the fact that. Here. To understand the meaning of these operators further.69) − a† (k)a(k) + e2ik 0 x0 a† (k)a† (−k) . φ(y)] = = = (k0 − k′ 0 ) √ 4k0 k′ 0 (k0 − k′ 0 ) √ 4k0 k′ 0 d3 xd3 y −ik·x−ik′ ·y 3 e δ (x − y) (2π)3 d3 x −i(k+k′ )·x e (2π)3 x0 =y 0 x0 =y 0 (k0 − k′ 0 ) −i(k0 +k′ 0 )x0 3 √ δ (k + k′ ) = 0.180 5 Free Klein-Gordon field theory × k0 φ(x) − iΠ(x). Π(y)] − ik′ 0 [Π(x). let us look at the Hamiltonian of the system. e 4k0 k′ 0 (5. . 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. independent of whether k′ = ±k.

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 .70) Similarly.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′ ) . d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ (5. (5.

69).71) and (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. (5.34).72) Substituting (5.72) into the expression for the Hamiltonian (5.182 +ei(k 0 −k ′ 0 )x0 5 Free Klein-Gordon field theory δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) + ei(k = 1 2 0 +k ′ 0 )x0 d3 k 0 0 1 e−2ik x a(k)a(−k) + a(k)a† (k) 0 k 0 x0 + a† (k)a(k) + e2ik a† (k)a† (−k) .

5. It follows now that Ek′ a(k′ )a† (k′ ) + a† (k′ )a(k′ ) 2 [a(k). a† (k′ ) + a(k).73). Thus. the zero point energy of such a system would be inﬁnite. we see from (5. (5.5 Creation and annihilation operators 183 (5. we then conclude that we can think of this as the Hamiltonian for an inﬁnite collection of decoupled harmonic oscillators. a(k′ ) a† (k′ ) + a† (k′ ) a† (k). H] = a(k).75) = −Ek a† (k).74) Thus. = = d3 k ′ d3 k ′ d3 k ′ Ek′ a(k′ ) a(k). But since this is an additive constant . H = 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 . From the form of the Hamiltonian for the Klein-Gordon ﬁeld in (5.73) = d3 k Ek a(k)a† (k) + a† (k)a(k) . Clearly. 2 where we have used the relation (5. a(k′ ) d3 k ′ Ek′ −δ3 k − k′ a† (k′ ) − a† (k′ )δ3 k − k′ 2 (5. we can indeed think of them as the annihilation and creation operators of a harmonic oscillator system. a† (k′ ) a(k′ ) 2 Ek′ a(k′ )δ3 k − k′ + δ3 k − k′ a(k′ ) 2 = Ek a(k). The two relations in (5.54) k0 = k2 + m2 = Ek . a† (k).75) show that the operators a(k) and a† (k) annihilate and create respectively a quantum of energy Ek . = = d3 k ′ Ek′ 2 d3 k ′ Ek′ a(k′ )a† (k′ ) + a† (k′ )a(k′ ) 2 a† (k).

O. (5. (5. = = = d3 k d3 k d3 k Ek a† (k)a(k) = where we have deﬁned the number operator for an oscillator with momentum k as in the case of a simple harmonic oscillator as N (k) = a† (k)a(k).73) (namely. N.76) N. we have from (5.O. : a† (k)a† (k′ ) : (5. (for bosons) N.O.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. Taking advantage of this.77) H N. the ordering of the operators is ambiguous. then.O. Thus. This is conveniently done by normal ordering of the operators.78) and we can deﬁne the total number operator for all the oscillators (inﬁnite number of them) in the system as . Let us recall that in the passage from classical to a quantum theory. we deﬁne normal ordering as the ordering in which the creation operators stand to the left of the annihilation operators. : a(k)a(k′ ) : = a(k)a(k′ ) = a(k′ )a(k). a(k)a† (k′ ) a† (k)a(k′ ) a(k)a(k′ ) a† (k)a† (k′ ) = = = = : a(k)a† (k′ ) : = a† (k′ )a(k). 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).76). If we normal order the Hamiltonian using (5. : a† (k)a(k′ ) : = a† (k)a(k′ ). = a† (k)a† (k′ ) = a† (k′ )a† (k).O.

78) and (5. d3 k′ N (k′ ) d3 k′ a(k′ )δ3 (k − k′ ) = a(k).81) which reﬂects again the fact that a(k) and a† (k) merely lower or raise the number of quanta of momentum k by one unit. N (k′ ) = = a(k).79) that a(k). d3 k′ N (k′ ) d3 k′ −a† (k′ )δ3 (k − k′ ) = −a† (k). N ] = = a† (k). (5. (5.5 Creation and annihilation operators 185 N= d3 k N (k) = d3 k a† (k)a(k).79) It is worth noting from the derivation in (5. a(k′ ) = −a† (k′ )δ3 (k − k′ ). Equation (5. It now follows from the deﬁnition of the number operators in (5. a† (k′ )a(k′ ) = a† (k′ ) a† (k). .80) then leads to (5.80) [a(k). N (k′ ) = a† (k). a† (k).5.77) how the normal ordering has redeﬁned away an inﬁnite zero point energy (which basically would arise from commuting a(k)a† (k) to bring it to the normal ordered form). a† (k′ ) a(k′ ) = a(k′ )δ3 (k − k′ ). a† (k′ )a(k′ ) a(k). a† (k). N = = a(k).

77) (as describing the quantum free Klein-Gordon theory) of the form H= d3 k Ek a† (k)a(k).82) Let us consider an energy eigenstate |E of this Hamiltonian satisfying H|E = E|E .75)) [a(k). let us consider the normal ordered Hamiltonian (5.186 5.85) .83) where we are assuming that the state |E is normalized and that the eigenvalue E is discrete for simplicity. (5.6 Energy eigenstates 5 Free Klein-Gordon field theory From now on. H] = Ek a(k). (5.83) that E = = = ≥ 0. It is clear from (5. Let us next recall that the operators a(k) and a† (k) satisfy the commutation relations (see (5. the inner product represents the norm of the state a(k)|E ) and. which can be achieved by quantizing the system in a box. the integrand is positive semideﬁnite. a† (k).77) and (5. (5.84) This is because both Ek as well as the inner product of states is positive deﬁnite (namely. This shows that the eigenvalues of the Hamiltonian have to be positive semideﬁnite and in the second quantized ﬁeld theory. we do not have the problem of negative energy states that we had in the single particle theory. E|H|E E| d3 k Ek a† (k)a(k)|E d3 k Ek E|a† (k)a(k)|E (5. consequently. H = −Ek a† (k).

we had redeﬁned . = Ek a(k)|E . = −Ek a† (k)|E . we have [a(k). Namely. or. (5. Therefore. such that a(k)|Emin = 0. .86) − Ha(k)|E H {a(k)|E } = (E − Ek ) {a(k)|E } .87) − Ha† (k)|E = (E + Ek ) a† (k)|E In other words. the ground state of the system or the vacuum state (recall that by normal ordering the Hamiltonian. the energy eigenvalue is bounded from below (E ≥ 0). we can lower the energy eigenvalue arbitrarily. (5. if |E represents an energy eigenstate with eigenvalue E.5. (5. raises the energy value. if |E is an energy eigenstate with energy eigenvalue E.6 Energy eigenstates 187 As a result.84). it would appear that by applying the annihilation operator successively. then a† (k) acting on it gives another energy eigenstate with the higher value of energy E+Ek . we note that a† (k). In such a state. The annihilation operator a(k) lowers the energy eigenvalue as it should. d3 k Ek Emin |a† (k)a(k)|Emin Emin = = 0. But as we have seen in (5. therefore. H] |E or. Let us next note that since a(k) acting on an energy eigenstate lowers the energy. then {a(k)|E } is also an eigenstate of energy with the lower energy value E − Ek .89) This is. a† (k)H|E H a† (k)|E = −Ek a† (k)|E . (5. there must exist a state (energy eigenstate) with a minimum energy. |Emin . clearly. or. a(k)H|E = Ek a(k)|E . The creation operator a† (k). therefore. H |E or. Similarly.88) and we cannot lower the energy any further.

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

k1 .5.96) if we deﬁne the momentum operator as (This is. nℓ . kℓ contains n1 quanta with four µ µ momentum k1 . k2 . kℓ .94) N (k)|n1 .95) n1 δ3 (k − k1 ) + · · · + nℓ δ3 (k − kℓ ) |n1 . and . k1 . the state |n1 . (5. However.97) H|n1 . · · · . k2 . k1 . nℓ . (5. k1 . kℓ = d3 k Ek N (k)|n1 . nℓ . · · · . kℓ . n2 . n2 . nℓ . · · · . Namely. k1 . 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.98) = (n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ ) |n1 . k1 .80)) N (k). of course. the natural deﬁnition from considerations of covariance. n2 . · · · . k2 .6 Energy eigenstates 189 This relation can be easily shown using the identity (see also (5. · · · . nℓ . nℓ . n2 quanta with four momentum k2 and so on. kℓ (5. · · · .) P= then d3 k k a† (k)a(k) = d3 k k N (k). a† (k1 ) = n1 a† (k1 ) which also leads to n1 n1 δ3 (k − k1 ) . kℓ = (5. (a† (k1 ))n1 = a† (k) a(k). Let us further note that since H= d3 k Ek a† (k)a(k) = d3 k Ek N (k). (5.

k1 .66) as well as the properties (5. · · · . n2 . nℓ . nℓ . = Ek |k .100) We can show in a straightforward manner (using the fundamental commutation relations in (5. k2 .102) that H 2 − P2 |k 2 = (Ek − k2 )|k = m2 |k . k2 . = k|k . p = n1 k1 + n2 k2 + · · · + nℓ kℓ . (5. 5. k2 .101) This state satisﬁes N |k H|k P|k = |k . kℓ . let us analyze the state obtained from the vacuum by applying a single creation operator. k1 . n2 . (5.103) .90)) that these states are orthonormal and deﬁne a complete basis of states for the Hilbert space of the Klein-Gordon theory. · · · . (5. k1 . nℓ . n2 . these states are eigenstates of H and P with total energy and momentum given by E = n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ . kℓ (5.190 5 Free Klein-Gordon field theory P|n1 . · · · . (5.7 Physical meaning of energy eigenstates To obtain the physical meaning of these energy eigenstates.102) It follows from (5.99) = (n1 k1 + n2 k2 + · · · + nℓ kℓ ) |n1 . |k = a† (k)|0 . Namely. kℓ = d3 k k N (k)|n1 .

107) This is. k2 ) and so on. we can think of this as the one particle state with four momentum kµ = (Ek .105) Since φ(x) is linear in the creation and the annihilation operators. µ n2 particles with k2 = (Ek2 . (5. k2 . (5. satisﬁes the single particle Klein-Gordon equation with positive energy (it provides a representation for the Poincar´ e group) and. k1 .105) as well as the fact that only the ﬁeld operator φ(x) depends on the space-time coordinates on which the . n2 . k). the only nonzero matrix element of the ﬁeld operator involving the vacuum state is given by k|φ(x)|0 = k|φ(x) . nℓ . therefore.106) In fact.5. This function satisﬁes ∂µ ∂ µ + m2 k|φ(x) = k| ∂µ ∂ µ + m2 φ(x)|0 = 0. · · · . (5. therefore.108) where we have used (5. k1 ). |φ(x) = φ(x)|0 = φ(−) (x)|0 . ﬁeld theories naturally describe systems of many identical particles. 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. Similarly. as a result. (5. of course.7 Physical meaning of energy eigenstates 191 This state. Let us next consider the state that is produced by the ﬁeld operator acting on the vacuum.104) µ can be thought of as the state with n1 particles with k1 = (Ek1 . Note that the particles described by such states are necessarily identical (although with diﬀerent energy and momentum) and. (5. we can show that the state |n1 . kℓ .

. 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 represents an energy eigenstate of the system describing a single particle (namely. in the ﬁrst quantized description. then. If |ψ denotes a state of the system. we note explicitly that φ(x)|k = = k|φ(x) ∗ = 0|a(k)φ(−) (x)|0 d3 k ′ (2π)3 2k′ 0 e ik ′ ·x † ′ ∗ ik ′ ·x ∗ 0 a(k) d3 k ′ a (k )e 0 ∗ = = (2π)3 2k′ 0 d3 k ′ (2π)3 2k′ 0 0|a(k)a (k )|0 † ′ × eik ·x 0 a(k). 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. 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.109) √ with k0 = Ek = k 2 + m2 > 0. In the present case. the wave function is given by ψ(x) = x|ψ . (5.110) In the second quantized description. This brings out the connection between the second quantized theory and the ﬁrst quantized theory. Let us note here parenthetically that the wave function for a single particle is really identiﬁed with φ(x)|k .192 5 Free Klein-Gordon field theory Klein-Gordon operator ( + m2 ) can act.109) to be the positive energy plane wave solutions of the single particle Klein-Gordon equation. Therefore. (5. We recognize the function in (5. |ψ = |k ).

k′ |φ(−) (x1 ) φ(−) (x2 ) |0 . by construction. k′ |φ (x1 ) . These states naturally lead to quantum mechanical probabilities which are non-negative.56)).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. (5. (5. φ (x2 ) .114) would give the probability for ﬁnding two Klein-Gordon particles with four momenta kµ = (Ek . As we have seen.5.113) which will describe a state with n particles at coordinates x1 . k) and k′ µ = (Ek′ . positive semi-deﬁnite and there is no problem of negative . Furthermore. We can similarly construct multi-particle states in the conﬁguration space of the form |φ (x1 ) . Such probabilities are. k′ ) at the coordinates x1 and x2 at a given time. φ (xn ) = φ(−) (x1 ) φ(−) (x2 ) · · · φ(−) (xn ) |0 . φ (x2 ) = k. · · · . such conﬁguran 1 2 tion states are automatically symmetric as we would expect for a system of identical Bose particles. Thus. such a state contains a superposition of all possible momentum states (which follows from the ﬁeld decomposition in (5. Such states are physically meaningful only if the time coordinates are equal (namely. They can also be shown to deﬁne a complete basis and describe what is known as the Fock space for the system. the absolute square of the amplitude (with x0 = x0 ) 2 1 k. The state = φ(−) (x)|0 .109). · · · . (5. xn . which is what we have seen explicitly in (5. x0 = x0 = · · · = x0 ).111) |φ(x) (5. for example.7 Physical meaning of energy eigenstates 193 Thus. the wave functional for a single particle with a deﬁnite energymomentum has the form ψ(φ(x)) = φ(x)|ψ = φ(x)|k .

5.117) Here. for the Klein-Gordon equation. the equation will take the form ∂µ ∂ µ + m2 φ(x) = J(x). the Green’s function G(x−y) satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). Furthermore. the Klein-Gordon equation will modify. which we will study in the following chapters. this equation is covariant (in fact. (5. When there are interactions present. Thus. If we know the Green’s function in (5.115) provided we know the Green’s function of the system.115) L= (5. the translation invariance of the Green’s function arises because the right-hand side of the equation is invariant under translations. all the physical states have positive semideﬁnite energy and that the probabilities are non-negative as they should be. namely. it is invariant under a Lorentz transformation) leading to the fact that the Green’s function is Poincar´ invariant. The Green’s function for a given inhomogeneous equation is deﬁned as the solution of the equation with a delta source. eventually we would like to study interactions of the system.117). which can be obtained from the Lagrangian density m2 2 1 ∂µ φ∂ µ φ − φ + Jφ. for the simple case of the Klein-Gordon ﬁeld interacting with an external source J(x) (which is a c-number function or a classical function). In a realistic theory the ﬁeld can have self-interactions or can interact with other dynamical ﬁelds of the system. 2 2 (5. the second quantized noninteracting Klein-Gordon theory is free from the problems of the ﬁrst quantized theory that we had discussed earlier.194 5 Free Klein-Gordon field theory probabilities in the second quantized description. Even classically we know that we can solve an inhomogeneous equation of the kind (5. namely. e . For example.8 Green’s functions Even though we have been examining the free Klein-Gordon ﬁeld theory so far.116) as the Euler-Lagrange equation (5.14).

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

Mathematically. substituting (5.196 5 Free Klein-Gordon field theory Thus. or.121) back into (5.3. this is equivalent to moving the poles inﬁnitesimally into the upper half plane.3: Poles of the integrand in the complex k0 -plane. k0 = ±Ek .122) G(x − y) = = This explicitly shows that the Green’s function is Poincar´ invariant. Im k 0 −Ek Ek Re k 0 Figure 5.122) has poles at k2 − m2 = 0.4.123) which lie on the real axis in the complex k0 -plane as shown in Fig. 5. (2π)4 k2 − m2 (5. e Let us note that the integrand of the Green’s function in (5. If we choose a contour of the form shown in Fig. (k0 )2 = k2 + m2 . (5. Therefore. this is expressed by writing . 5. the Green’s function is not uniquely deﬁned until we specify a contour of integration in the complex k0 -plane and specifying a contour simply corresponds to specifying a boundary condition for the Green’s function. or. 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) .120). then.

or.124).5. then we must close the contour in the lower half plane for the damping of the exponential in (5.124) In this case. the poles of the integrand occur at k2 − m2 − ik0 η = (k0 )2 − k2 − m2 − ik0 η = 0. k0 = ±Ek + iǫ. (5. we conclude that GA (x − y) = 0. for x0 − y 0 > 0.124) are in the upper half plane. 2 Since in this case the poles in (5. GA (x − y) = lim η→0+ e−ik·(x−y) d4 k . if we enclose the contour in the lower half plane.126) For x0 − y 0 < 0. (2π)4 k2 − m2 − ik0 η (5.8 Green’s functions 197 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek + iǫ Ek + iǫ Re k 0 Figure 5.4: Choice of the contour in the complex k0 -plane for the advanced Green’s function. we have to close the contour in the upper half plane (for the damping of the exponential) in which . or.125) where we have identiﬁed ǫ = η . Therefore. or. on the other hand. (5. We note that if x0 − y 0 > 0. iη k − 2 0 2 ≃ k 2 + m2 . the integral will vanish. k0 − iǫ = ±Ek .

sin Ek x0 . Explicitly. (2π)3 Ek Similarly. d3 k eik·x (2π)3 Ek for x0 > 0. GA (x) = θ −x0 d3 k eik·x sin Ek x0 . then. we can express this mathematically by deﬁning e−ik·x d4 k . 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 ) (5.127) d3 k eik·(x−y) sin Ek x0 − y 0 . 4 k 2 − m2 + ik 0 η (2π) GR (x) = lim η→0+ (5.129) In this case. For x0 − y 0 < 0. the poles of the integrand in (5. for x0 < 0. if we choose the contour in the complex k0 -plane as shown in Fig.198 5 Free Klein-Gordon field theory case we can evaluate the integral by the method of residues. (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 .5 or equivalently if we push both the poles to the lower half plane.128) or. we can write (we denote the argument of the Green’s function by x for simplicity) G (x) = A 0. 5. (5.

we have to close the contour in the upper half plane for the damping of the exponential in (5.8 Green’s functions 199 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek − iǫ Ek − iǫ Re k 0 Figure 5. k0 = ±Ek − iǫ. therefore. (5. or. we conclude that GR (x) = 0. on the other hand. In this case.5.131) For x0 > 0. k0 + iǫ ≃ ±Ek . (5. k0 + iη 2 2 2 ≃ k2 + m2 = Ek .5: Choice of the contour in the complex k0 -plane for the retarded Green’s function. if we close the contour in the upper half plane. since both the 2 poles are in the lower half plane. then the integral will vanish. 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 ) . Note that if x0 < 0. k2 − m2 + ik0 η = (k0 )2 − k2 − m2 + ik0 η = 0.130) where we have again identiﬁed ǫ = η . or. or. x0 < 0.129) and.

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 . (5. the average of the two.) It is clear from (5. the diﬀerence between GA (x) and GR (x) would deﬁne a function which would satisfy the homogeneous (Klein-Gordon) equation and . 2 (5.132) e−i(−Ek −iǫ)x + −2Ek 0 +ik·x The overall negative sign in (5. we see that we can write 0. (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. Therefore. On the other hand. (2π)3 Ek (5. for x0 > 0.117) with a delta function source. sin Ek x0 . 1 GA (x) + GR (x) . Thus. GR (x) = −θ x0 d3 k eik·x sin Ek x0 . (The advanced and the retarded Green’s functions have similar form except for their support and the overall sign. − d3 k eik·x (2π)3 Ek GR (x) = for x0 < 0.128) and (5. (5.133) that GR (x) = GA (−x).132) arises because the contour is clockwise in the lower half plane.134) Both the retarded and the advanced Green’s functions satisfy the inhomogeneous equation (5. namely.133) or.135) would also deﬁne a Green’s function corresponding to the contour picking up the principal values at the poles.

5.6. namely.136) + θ x0 = = θ x0 + θ −x0 d3 k eik·x sin Ek x0 .8 Green’s functions 201 would not strictly correspond to a Green’s function of the theory. known as the Feynman Green’s function. however. G(x) = −G(−x). This is deﬁned with the contour in the complex k0 -plane as shown in Fig.136) that the Schwinger function is real (G(x))∗ = G(x).137) Furthermore. Quantum mechanically. it is easily seen from (5. 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. 5. there is another Green’s function.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. and is antisymmetric. (2π)3 Ek This clearly satisﬁes the homogeneous Klein-Gordon equation ∂µ ∂ µ + m2 G(x) = 0. (5. which corresponds to pushing one of the poles (on the left) to the upper half plane while moving the other (on the right) . (5.138) (5. which is more useful in the calculation of scattering matrix elements.

Since there are poles in both halves of the complex k0 -plane. 2 (k0 )2 − Ek + iη = 0. the poles of the integrand in (5. in this case. or.142) . this can be implemented by deﬁning d4 k e−ik·x . (5. (5.140) occur at k2 − m2 + iη = 0. or. to the lower half plane. or. = ± (Ek − iǫ) .6: Choice of the contour in the complex k0 -plane for the Feynman Green’s function.140) In this case.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 5.141) η where we have identiﬁed ǫ = 2Ek with η → 0+ . 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. 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. (2π)4 k2 − m2 + iη GF (x) = lim η→0+ (5.

if x0 > 0. On the other hand. Here the contour is clockwise leading to an overall negative sign. we have used k ↔ −k in the last step (and we understand that k0 = Ek ). we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 (5.144) GF (x) = = ǫ→0+ lim ǫ→0+ lim (−2πi) i 2 i 2 d3 k e−i(Ek −iǫ)x +ik·x (2π)4 2 (Ek − iǫ) 0 = − = − d3 k e−iEk x +ik·x (2π)3 Ek d3 k e−ik·x (2π)3 k0 (5. then we have to close the contour in the lower half plane which will pick up the residue of the pole at k0 = Ek − iǫ. The Feynman Green’s function clearly has support in both the future as well as the past light cones and can be written as .8 Green’s functions 203 Therefore. for x0 < 0.5. we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 GF (x) = = ǫ→0+ lim ǫ→0+ lim 2πi d3 k e−i(−Ek +iǫ)x +ik·x (2π)4 2 (−Ek + iǫ) 0 i = − 2 = − i 2 d3 k eiEk x +ik·x (2π)3 Ek d3 k eik·x (2π)3 k0 (5. where.143) = G(−) (x).145) = −G(+) (x). for x0 > 0. Therefore.

136) as d3 k eik·x sin Ek x0 = G(+) (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 .145) that we can write the Schwinger function (5. the Feynman Green’s function is an even function. However.143) and G(+) (x) in (5. It is deﬁned as e−ik·x d4 k . GF (−x) = GF (x).148) There is yet another choice of the contour and. Note from (5. (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).147) Therefore. (2π)3 k0 (5.146) G(+) (−x) = (5.150) The functions G(+) (x) and G(−) (x) (see (5. namely. (2π)3 Ek G(x) = (5.204 5 Free Klein-Gordon field theory GF (x) = −θ x0 G(+) (x) + θ −x0 G(−) (x). (2π)3 (5.151) G(+) (x) = i G(−) (x) = −i .143) and (5. therefore.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. We note from the structure of G(−) (x) in (5. (2π)4 k2 − m2 − iη G(x) = lim η→0+ (5. (2π)3 d4 k θ(−k0 )δ(k2 − m2 ) e−ik·x .145) that i 2 d3 k eik·x = −G(−) (x). we will not go into this in more detail here.

(5.9 Covariant commutation relations 205 and are correspondingly known as the positive and the negative energy (frequency) Green’s functions.152) and all the Green’s functions can be expressed as linear combinations of these two fundamental Green’s functions.9 Covariant commutation relations To make contact between the quantum ﬁeld theory and the various Green’s functions that we have constructed. (5. (5. We note from (5.58) that φ(x) = φ(+) (x) + φ(−) (x).57) that (we recall that k0 = Ek = 0) √ (5. (2π)3 (5.151)) G(+) (x) ∗ = G(−) (x). 5.155) k2 + m2 > . Let us recall from (5. let us calculate some of the covariant commutation relations satisﬁed by the ﬁeld operators.154) which will be useful in our later discussions.128).153) where ǫ(x) = (θ(x) − θ(−x)) is known as the sign function (or the alternating step function).150) that we can write GA (x) = θ(−x0 )G(x) = θ(−x0 ) G(+) (x) + G(−) (x) .133) and (5. GR (x) = −θ(x0 )G(x) = −θ(x0 ) G(+) (x) + G(−) (x) . For completeness.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 . we note from (5.5. where we note from (5. They satisfy (see (5.

156) Since we know the commutation relations (5. φ(−) (y) = d3 k e−ik·x a(k). it is straightforward to evaluate φ(+) (x). a(k). a† (k′ ) e e 2k0 2k′ 0 = 0. d3 k ′ (2π)3 2k′ 0 e−ik ·y a(k′ ) ′ (5. d3 k ′ (2π)3 2k′ 0 eik ·y a† (k′ ) ′ d3 k d3 k′ ik·x ik′ ·y † √ √ a (k). φ(−) (y) = = 1 (2π)3 d3 k (2π)3 2k0 eik·x a† (k).157) d3 k d3 k′ −ik·x −ik′ ·y √ √ e e a(k). eik·x a† (k). φ(−) (x). (2π)3 2k0 d3 k ′ ′ eik ·y a† (k′ ) (2π)3 2k′ 0 .206 5 Free Klein-Gordon field theory φ(+) (x) = φ(−) (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k). φ(+) (y) = = 1 (2π)3 d3 k (2π)3 2k0 e−ik·x a(k). (5. a(k′ ) = 0 = a† (k). a(k). a† (k′ ) = δ3 k − k′ . a(k′ ) 0 ′0 2k 2k = 0.66) between the creation and the annihilation operators. a† (k′ ) . φ(+) (x).

(5. φ(−) (y) + φ(−) (x). the commutator of two ﬁeld operators.40)) and. φ(+) (y) = −iG(+) (x − y) − iG(−) (x − y) = −iG(x − y) = −i d3 k eik·(x−y) sin Ek x0 − y 0 . It follows now that φ(−) (x).159) are not restricted to have equal time arguments any more (unlike in (5. (5. (2π)3 Ek (5. is proportional to the Schwinger Green’s function.159). φ(+) (y) = iG(+) (y − x) = −iG(−) (x − y).5.160) Namely. at unequal times. we can evaluate [φ(x).9 Covariant commutation relations 207 = = = 1 (2π)3 1 (2π)3 1 (2π)3 d3 k d3 k′ −ik·x+ik′·y √ √ a(k). a† (k′ ) e 2k0 2k′ 0 d3 k d3 k′ −ik·x+ik′·y 3 √ √ e δ k − k′ 2k0 2k′ 0 d3 k −ik·(x−y) e 2k0 (5.158) and (5.161) simply because the ﬁeld operator φ(x) satisﬁes the Klein-Gordon equation. φ(+) (y) + φ(−) (y) = φ(+) (x). using the results in (5.158) and (5. furthermore.159) Note that the ﬁeld operators in the commutators (5. φ(y)] = φ(+) (x) + φ(−) (x). It is now clear that the Schwinger function satisﬁes the homogeneous equation µ ∂x µ ∂x + m2 G(x − y) = 0.158) = −iG(+) (x − y). .

x0 =y 0 (5. φ(y)]x0 =y0 = −iG(x − y) = 0. Equation (5. (5. this relation can also be seen to imply that [φ(x). for space-like separations. therefore.208 5 Free Klein-Gordon field theory ∂µ ∂ µ + m2 φ(x) = 0.166) In other words.165) x0 =y 0 We note that by a Lorentz transformation.160) also shows that two ﬁeld operators φ(x) and φ(y) do not commute for arbitrary values of the coordinates.164) which is expected from the anti-symmetry of the Schwinger function (5.160) leads to the familiar relation (see (5. a light signal cannot connect two space-like points and. This is known as the principle of microscopic causality. Namely. 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.162) These commutation relations are known as covariant commutation relations simply because the (scalar) Green’s functions are invariant under Lorentz transformations. namely.) On the other hand. G(Λx) = G(x).40)) [φ(x). for (x − y)2 < 0.163) where Λµν represents a Lorentz transformation. (5. as can be checked even in the simple classical Poisson bracket relations for a one dimensional free particle or a harmonic oscillator.138) and (5. This is consistent with our intuitive expectation. We also note that . two measurements at space-like separations should not inﬂuence each other. φ(y)] = 0. (5. the commutator of two ﬁeld operators vanishes. (5.

(5. Relativistic Quantum Fields. McGrawHill.154) that the retarded and the advanced Green’s function can also be expressed in terms of covariant commutation relations as θ(x0 − y 0 ) [φ(x).5. φ(y)] = −iθ(x0 − y 0 ) G(+) (x − y) + G(−) (x − y) = −iθ(x0 − y 0 )G(x − y) = iGR (x − y). (5. New York. Drell. θ(y 0 − x0 ) [φ(x). let us note from (5. (2π)3 This is consistent with our earlier ﬁeld quantization rule (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 (5.136). which will be useful later.40). J. (5. φ(y)] = −iθ(y 0 − x0 ) G(+) (x − y) + G(−) (x − y) = −iθ(y 0 − x0 )G(x − y) = −iGA (x − y). D.169) .10 References 1.167) d3 k ik·(x−y) e = −δ3 (x − y). 1964. Bjorken and S. φ(y) [φ(x).160) as well as the relations in (5. ˙ φ(x).150) and (5. Π(y)]x0 =y0 = x0 =y 0 = −i ∂G(x − y) ∂y 0 x0 =y 0 = iδ3 (x − y).168) To conclude this section. namely. 5.

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

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

∂µ φ′ x′ δS = or. In such a case. ∂x (6. ∂µ φ′ x′ ). Symmetries have interesting consequences for continuous transformations. (6. Thus. − d4 x L (φ(x). ∂µ φ(x)) = 0.4) then. invariance of the action under the inﬁnitesimal forms of the transformations in (6. ∂µ φ′ (x) − L (φ(x). the transformations in (6. Such transformations are known as internal symmetry transformations to be contrasted with space-time transformations where space-time coordinates transform along with the dynamical variables. let us consider an inﬁnitesimal transformation with ∂x′ = 1. as indicated in (6. d4 x L (φ(x). ∂µ φ(x)) = ′ d4 x′ L(φ′ x′ . It is clear that in such a case.6) which holds for most global space-time symmetries as well as internal symmetries.3) deﬁne a symmetry of the system.3) includes a very interesting class of transformations where the space-time coordinates do not change. (6.7) . the Euler-Lagrange equations for the primed and the unprimed systems would remain form invariant. namely. ∂µ φ(x)) = 0. or. It is worth emphasizing here that (6. 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. ∂µ φ(x)) = 0.5) and only the dynamical variables of the theory transform. x′ µ = xµ .212 6 Self-interacting scalar field theory S= d4 x L (φ(x).3). d4 x L φ′ (x). (6.3) would imply ′ d4 x′ L φ′ x′ . ∂µ φ′ (x) − d4 x L φ′ (x).

∂µ φ(x)) = L (φ(x).8) which must hold independent of the use of equations of motion for the system under study. ∂µ φ′ (x) − L (φ(x). as we will see later. Clearly. This is quite general and it is possible to have symmetry transformations for which K µ = 0. ∂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) . keep only linear terms in δφ(x)) L φ′ (x). this is indeed the case for internal symmetry transformations. ∂µ φ(x)) = ∂µ K µ . On the other hand. (6.7) will hold if L φ′ (x).1 Nother’s theorem 213 where we have identiﬁed the integration variable in the ﬁrst term on the right-hand side to be x (instead of x′ ) in the intermediate step. (6. ∂µ φ(x)) + δφ(x) −L (φ(x). (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. therefore.10) (6. (6.11) ∂L ∂L + δ(∂µ φ(x)) ∂φ(x) ∂∂µ φ(x) = δφ(x)∂µ = ∂µ δφ(x) . ∂µ φ′ (x) − L (φ(x). so that δ(∂µ φ(x)) = ∂µ φ′ (x) − ∂µ φ(x) = ∂µ δφ(x).¨ 6.9) we can calculate explicitly (we remember that δφ(x) is inﬁnitesimal and.

we note that . However. the conserved current independent of the parameter of transformation is not always a vector. This is just the principle of least action. φ′ must also satisfy the same equation of motion in the transformed frame as the original ﬁeld. In fact. (Note that. (6. namely.13) which is conserved. (6. x) . ∂µ J µ (x) = 0. ∂µ δφ(x) ∂L ∂∂µ φ(x) = ∂µ K µ = 0. this is always true when the equations of motion are used.11). First. given a conserved current. in general. for any inﬁnitesimal variation. Namely. For a symmetry. not all such variations will deﬁne a symmetry.12) ∂L − Kµ ∂∂µ φ(x) This shows that whenever there is a continuous symmetry associated with a system. on φ′ . Its tensor structure is determined by the tensor structure of the inﬁnitesimal parameter of transformation. we can deﬁne a current ∂L − K µ. there is no restriction on δφ and.214 6 Self-interacting scalar field theory Here we have used the Euler-Lagrange equation in the intermediate steps. ∂∂µ φ(x) J µ (x) = δφ(x) (6. therefore.14) Several comments are in order here. Second. (6. we can deﬁne a charge Q= d3 x J 0 (t. we obtain ∂µ δφ(x) or.8) and (6.15) which will be a constant (in time) and will generate the symmetry transformations.) Comparing (6. on the other hand.

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 = 0.17) and. (6. the time independence of Q also means that it commutes with the Hamiltonian of the theory [Q. with the usual assumptions on the asymptotic fall oﬀ of the ﬁeld operators (see (5. Since. let us consider the simple case of an inﬁnitesimal o global space-time translation deﬁned by xµ → x′µ = xµ + ǫµ .¨ 6. if we have a continuous symmetry in the theory (transformations under which the action is invariant). the integral of a total divergence (∇ · J) vanishes. This is known as N¨ther’s theorem and is quite important in the study of symmeo tries in dynamical systems. (6. Q corresponds to the generator of the inﬁnitesimal symmetry transformations of the theory.18) where the parameter of translation ǫµ is assumed to be inﬁnitesimal and constant (global). if there exists a conserved charge in a given theory.1. we are dealing with a scalar ﬁeld. Q and H can have simultaneous eigenstates. (6.16) Here we have used the fact that. then it generates inﬁnitesimal transformations (through commutation relations) which deﬁne a symmetry of the theory. . Namely. consequently. or. there exists a conserved charge which is the generator of these inﬁnitesimal symmetry transformations. H] = 0. δxµ = x′µ − xµ = ǫµ . 6. Thus. As an example of the consequences of N¨ther’s theorem.1 Space-time translation. In the operator language.3)). The converse is also true. In this case.

∂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) (6.10).21) that (ǫµ is a constant parameter) K µ = −ǫµ L. = − φ′ x′ − φ′ (x) = −ǫµ ∂µ φ′ (x) (6. ∂µ φ′ (x) − L (φ(x).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).8)) L φ′ (x).216 6 Self-interacting scalar field theory φ′ x′ = φ (x) .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. We can now identify from (6. where we have used (5. then. We see that since the change in the Lagrangian density is a total divergence.21) = −ǫµ ∂µ φ(x) = −ǫµ ∂µ L = ∂µ K µ . (6. we note that under the transformation (6. we have identiﬁed φ′ (x) = φ(x) simply because the parameter ǫµ multiplying on the right-hand side is already inﬁnitesimal and any further correction coming from φ′ (x) will only be of higher order. we can now calculate explicitly the change in the Lagrangian density (see (6.22) . (6.20) where in the last step.20). With this. Next.7) and (6.11) and (6.

depending on the parameter of transformation.27) . ∂∂µ φ(x) ∂∂µ φ(x) (6. Therefore.25) which can be easily checked to satisfy (using the equations of motion) ∂µ T µν = 0. namely. therefore. we can always deﬁne an improved symmetric stress tensor by coupling the theory to a gravitational background and taking variation with respect to the gravitational background.24) µ Jǫ (x) = δφ = −ǫν (∂ν φ(x)) = −ǫν (∂ ν φ(x)) It is clear. (6. even when the naive N¨ther o procedure does not lead to a symmetric stress tensor.¨ 6. ∂∂µ φ(x) T µν = (∂ ν φ(x)) (6.1 Nother’s theorem 217 δφ(x) ∂L ∂L = −ǫν (∂ν φ(x)) . to be (see (6. that the conserved current independent of the parameter can be identiﬁed with ∂L − η µν L.23) so that we obtain the current associated with the symmetry transformation.13)) ∂L − Kµ ∂∂µ φ(x) ∂L + ǫµ L ∂∂µ φ(x) ∂L − η µν L ∂∂µ φ(x) = −ǫν T µν . (6. (6.26) T µν is known as the stress tensor of the theory and can always be deﬁned to be symmetric (It is the source for the gravitational ﬁeld gµν just as the electromagnetic current j µ is the source for the electromagnetic potential Aµ .). T µν = T νµ .

∂∂µ φ(x) and this leads to the explicit form for the stress tensor (6.) As a result. in the present case.218 6 Self-interacting scalar field theory The conserved current.15)).30) T µν = ∂ ν φ(x) ∂L − η µν L ∂∂µ φ(x) (6. in this case. the conserved charges associated with the symmetry transformation (see (6. we obtain ∂L = ∂ µ φ(x). in this case.25) (6.28) This is consistent with our physical intuition that the energy-momentum. 2 2 (6. For the free Klein-Gordon theory. . which. P µ . is a second rank tensor simply because the parameter of transformation is a four vector.29) Therefore. We note that. let us recall that the Lagrangian density has the form L= m2 2 1 ∂µ φ∂ µ φ − φ . would correspond to the components of a four vector which we identify with the energy-momentum operator as Pµ = d3 x T 0µ .31) = ∂ ν φ(x)∂ µ φ(x) − η µν L = T νµ . is manifestly symmetric. the current is a tensor one rank higher than the parameter of transformation. (6. for the free theory. should generate inﬁnitesimal space-time translations. (In general.

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

which in turn allows us to deﬁne a vacuum state of the theory.35) The restriction on the coupling constant. λ. for various other reasons we can show that (6. then. a Hamiltonian) which is unbounded from below for any value of the coupling constant g. therefore. which does not respect the discrete symmetry (6. the simplest interaction Lagrangian density involving only the scalar ﬁelds which we can think of has the form λ 4 φ (x). (We will see this shortly. in fact. the same canonical dimension as . (6. Let us also g note here that an interaction Lagrangian density of the form − 3! φ3 . Consequently.37) and has. in units of = c = 1 (which we have been using).36) While there are other interaction Lagrangian densities that we can construct consistent with our symmetry requirements. . is there so that the Hamiltonian will be positive deﬁnite. With these conditions. In these units. Let us recall that the action for a quantum mechanical particle has the generic form S = = dt (p q + · · · ) ˙ (p dq + · · · ). (6. To explain this brieﬂy. the fully self-interacting theory of a real Klein-Gordon ﬁeld is described by a Lagrangian density L = L0 + LI . we can show that the canonical dimension of any variable can be expressed in powers of an arbitrary mass dimension [M ]. let us introduce the concept of canonical dimensions. would lead to a potential (and. Thus. the action is dimensionless.) As a result.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. (6. 4! LI = − λ > 0.34).

d4 x d4 x [λ] ([φ])4 = [M ]0 . we obtain λ 4 φ = [M ]0 . [φ] = [M ]. ([φ])2 = [M ]2 . let us note that for a monomial interaction action of the form .6.38) With this we can now study the canonical dimension of the free part of the Klein-Gordon action and this leads to 1 m2 2 ∂µ φ∂ µ φ − φ 2 2 [S0 ] = or. (6. d4 x = [M ]0 . 4! [SI ] = − or. in these units. In general. [λ] = [M ]0 . [M ]−4 [M ][φ][M ][φ] = [M ]0 . or. d4 x [∂µ ] [φ] [∂ µ ] [φ] = [M ]0 . [∂µ ] = [M ]. or. (6. [M ]−4 [λ] [M ]4 = [M ]0 .2 Self-interacting φ4 theory 221 [L] = [T ] = [xµ ] = [M ]−1 . the Klein-Gordon ﬁeld variable in four dimensions has a canonical dimension 1. has a canonical dimension 1 and.40) Thus. From the canonical dimension of the interaction term in the Lagrangian density in (6.) The mass parameter. (The dimensionality of the ﬁeld variable depends on the number of space-time dimensions which we will see later. we conclude that the coupling constant or the interaction strength for the φ4 -self-interaction in (6.35).35) is dimensionless. or. or. of course. (6.39) 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. therefore. or.

(We will see this in a later chapter when we discuss renormalization of quantum ﬁeld theories. restricts n ≤ 4 and. (6.32)) ∂L ˙ = φ(x). or. Such theories are known as non-renormalizable theories.) This.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! . consequently. 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. the coupling constant of the interaction Lagrangian density will have inverse dimensions of mass. ˜ SI = [g] d4 x [φn ] = [M ]0 .34)). for n > 4. the coupling constants in the theory cannot have dimensions of inverse mass.41) Therefore. therefore. the canonical momentum conjugate to the ﬁeld variable of the theory continues to be (see (5. the φ4 interaction is the only physically allowed interaction in this case. ˙ ∂ φ(x) Π(x) = (6. [g] = [M ]4−n .42) so that from (6. Since the interaction Lagrangian density does not involve derivatives of ﬁelds. for example. (6. If we want to restrict to renormalizable theories which can give rise to meaningful physical predictions. consistent with our symmetry requirements (see. [g] [M ]−4 [M ]n = [M ]0 . or. In such a case.222 6 Self-interacting scalar field theory g ˜ SI = − n! d4 x φn .

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

(6. t) eiHt = φ(H) (x. the Klein-Gordon ﬁeld o operator will be given by φ(S) (x) = e−iHt φ(H) (x. (6. for example. (6. In the Schr¨dinger picture. 0).47) O(S) = O(H) (0). Thus. (6. for example. (6.45) that (H) ψ|O(H) (t)|ψ (H) = (S) ψ(t)|O(S) |ψ(t) (S) .46) In writing (6. the 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. (6. t0 ) = e−iH(t−t0 ) . namely. time independent. in the Schr¨dinger picture. It follows from (6. The two pictures are related by the unitary transformation = e−iHt |ψ |ψ(t) (S) (H) . we have assumed that at t = 0 both the pictures coincide. 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.45) O(S) = e−iHt O(H) (t)eiHt . |ψ(0) (S) = |ψ (H) .49) .45)) and is. of course.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.48) The total Hamiltonian operator in the Schr¨dinger picture coino cides with that in the Heisenberg picture (see.45).

(6.51) In the interaction picture. 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. that we can identify |ψ(0) (IP ) = |ψ(0) (S) (S) = |ψ (H) .50) Note. both the operators as well as the state vectors carry time dependence.6. therefore. In this case. once again. by deﬁnition. it is worth emphasizing here that for certain applications.52) where we have used the fact that the total Hamiltonian H is the same in both the Schr¨dinger as well as the Heisenberg pictures and o that. we deﬁne |ψ(t) (IP ) = eiH0 (S) t t |ψ(t) (S) = eiH0 (S) (S) t −iHt e |ψ t (H) . (However. o Here. O(IP ) (t) = eiH0 = eiH0 (S) O(S) e−iH0 e t (S) (S) t −iHt O(H) (t)eiHt e−iH0 . the more convenient description goes under the name of interaction picture which is. this picture is useful. intermediate between the Schr¨dinger and the Heisenberg pictures. (IP H0 ) (t) = eiH0 t (S) H0 e−iH0 (S) t (S) = H0 . (6. (IP ) (IP ) (6.3 Interaction picture and time evolution operator 225 However. O(IP ) (0) = O(S) = O(H) (0).) Although the Heisenberg picture has a manifestly covariant description. in relativistic interacting ﬁeld theories. in some sense. 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) . .

it is easy to show that the time evolution operator satisﬁes the properties.50) to have the explicit form U (t.56) then.51) that the free Hamiltonian in the interaction picture is time independent and. (6. in this picture. (6.226 6 Self-interacting scalar field theory (IP HI ) (t) = eiH0 (S) t (S) HI e−iH0 (S) t .55) where the time evolution operator in the interaction picture in (6.53) Similarly. . if we deﬁne the time evolution operator through the relation |ψ(t) = U (t. the ﬁeld operators can have a plane wave expansion. while the dynamical evolution of the operators is governed by the free Hamiltonian through the Heisenberg equations of motion.55) can be seen from (6. In this picture. in the interaction picture they continue to hold even in the presence of interactions. therefore. We note from (6. t0 ) = eiH0 (S) t −iH(t−t0 ) −iH0 e e (S) t0 .54) In other words. we can derive the time evolution for the operators to be ∂O(IP ) ∂t = (S) (S) ∂ eiH0 t O(S) e−iH0 t ∂t (S) (S) = iH0 O(IP ) (t) − iO(IP ) (t)H0 (S) = −i O(IP ) (t). t0 ) |ψ (t0 ) . (6.49) and (6. H0 ) . From now on. This shows that once we deﬁne commutation relations for the operators in the free theory. let us drop the superscript (IP ) with the understanding that we are nonetheless working in the interaction picture. i (6. is governed by the Schr¨dinger equation with the interaction o Hamiltonian playing the role of the Hamiltonian. the time evolution of the state vectors. H0 = 1 (IP ) (IP O (t).

(6. t0 ) = U (t0 . ∂t ∂U (t.6.57). t0 ) .59) To bring this to a more convenient form.3 Interaction picture and time evolution operator 227 U (t. i ∂t i or. (6.50)) ∂|ψ(t) = HI (t)|ψ(t) . t1 ) U (t1 . t t1 t0 tn−1 t0 dt1 t0 dt2 · · · dtn HI (t1 ) HI (t2 ) · · · HI (tn ) (6. (Since H0 is time independent in the interaction picture (see (6. t0 ) t0 ½−i ½−i t dt1 HI (t1 ) t0 t t0 ½−i t1 dt2 HI (t2 ) U (t2 .52) we see that the time evolution operator in (6. t0 ) = U −1 (t. it follows that the total Hamiltonian is time independent in the interaction picture. t0 ) . U † (t. t) = ½. let us look at the second order term in (6.53) that the interaction Hamiltonian in the interaction picture is time dependent. to obtain t U (t.55) satisﬁes the equation (this also follows from (6.59) and note that . t0 ) = ½ − i = = dt1 HI (t1 ) U (t1 .57) Furthermore.56) with the use of (6.51)). from (6. U (t2 .58) iteratively subject to the initial condition in (6.58) We note from (6. t0 ) = HI (t)U (t. t0 ) t0 t t1 dt1 HI (t1 ) + (−i)2 dt1 t0 t0 dt2 HI (t1 ) HI (t2 ) +··· + ··· +(−i)n +··· . t0 ) = U (t2 . t) .) We can now solve (6.

we can show that the n-th term in the series in (6.228 6 Self-interacting scalar field theory (−i)2 t t1 dt1 t0 t0 t dt2 HI (t1 ) HI (t2 ) t2 = (−i)2 = (−i)2 2 t dt2 t0 t t0 dt1 HI (t2 ) HI (t1 ) t1 dt1 t0 t2 t0 dt2 HI (t1 ) HI (t2 ) dt1 HI (t2 ) HI (t1 ) + t0 dt2 t0 t = (−i)2 2 t t1 dt1 t0 t2 t0 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + t0 dt2 t0 t dt1 θ (t2 − t1 ) HI (t2 ) HI (t1 ) t = (−i)2 2 t dt1 t0 t t0 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + t0 dt2 t0 t dt1 θ (t2 − t1 ) HI (t2 ) HI (t1 ) = (−i)2 2 t0 dt1 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + θ (t2 − t1 ) HI (t2 ) HI (t1 ) = (−i)2 2! t t0 t dt1 dt2 T (HI (t1 ) HI (t2 )) .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 (6.59) can be written as . In a similar manner.61) with the operator at later time standing to the left of the operator at earlier time. (6.

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

the state into which this will evolve at t = ∞ is deﬁned from (6.65) Then. As a result.230 6 Self-interacting scalar field theory so that in the inﬁnite past as well as in the inﬁnite future. (6. −∞).64) can be thought of as the relevant boundary condition for the scattering problem under study. (6.68) Consequently. The condition (6. −∞)|i = f |S|i .67) which has . −∞)|i = S|i . the S-matrix (or the scattering matrix) of the theory can be identiﬁed with the time evolution operator (6. the probability amplitude for an initial state |i to be in the ﬁnal free particle state |f at t = ∞.67) Therefore. 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. which is the deﬁnition of the scattering amplitude. 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.66) to be Sf i = f |ψi (∞) = f |U (∞.55) to be |ψi (∞) = U (∞. (6. (6.66) where we have identiﬁed S = U (∞. Let us denote the initial state at inﬁnite past as the free particle state |ψi (−∞) = |i .64)) leads to the same result for the rate of transition in the physical scattering of particles. is obtained from (6. the interaction vanishes. −∞)|ψi (−∞) = U (∞.

These states can be constructed by noting that the three pictures (Heisenberg.63). this state would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (in) = U (0.69) Furthermore. Let us see this explicitly in the case of the linear term in the expansion of the time evolution operator in (6.6.70) will have the form . (6. in (6. −∞)|Ψ(−∞) (in) . It is clear from the expansion above that the adiabatic switching of the interaction in (6.69) through the appropriate boundary conditions.70) where the time evolution operator U (0. Schr¨dinger and interaction) coincide at t = 0 (see (6. Then. These are asymptotic free states as t → −∞ and t → ∞ respectively.57)). In this case.64). say.64) provides a regularized meaning to the oscillatory terms in (6. not well deﬁned if we do not use an adiabatic interaction of the form.4 S-matrix 231 the explicit perturbative expansion of the form (see (6. (6.63)) −i ∞ R S = U (∞. This also leads to the notion of “in” and “out” states which are quite important in a formal description of scattering theory. the S-matrix (the scattering matrix) is unitary since the time evolution operator is (see (6. the linear contribution to the right-hand side of (6.51)) and o since the Heisenberg states are time independent. they are uniquely given by the states (in any picture) at t = 0. in general. Let us suppose that at t → −∞ we have a free incoming state denoted as |Ψ(−∞) (in) = |i . −∞) = 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+ +··· . −∞) is.

the “in” and the “out” states deﬁne a complete space of states at t → −∞ and t → ∞ respectively. from their deﬁnitions in (6. ∞) would not be well deﬁned at the upper limit.70) and (6. This clariﬁes how the boundary condition (6. we have used the fact that H0 ) = (S) H0 and that |Ψ(−∞) (in) = |i is a free state with Ei the energy (IP eigenvalue of H0 ) .71) would show that in this case. if we assume that as t → ∞. (6.72) we see that they are related to each other as . we have a free outgoing state |Ψ(∞) (out) .72) we see that even though both the “in” and the “out” states are related to the same Heisenberg state. In a parallel manner. This shows that the asymptotically free “in” and “out” states can be deﬁned in a unique manner by relating them to the Heisenberg state. ∞)|Ψ(∞) (out) . the regularizing factor would have to be e−ηt without which U (0.232 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) . From (6. then this would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (out) = U (0. We also note that the integral in (6.70) and (6. they are not identical because their deﬁnitions use diﬀerent regularizing factors. Rather.71) (IP where in the intermediate steps. (6.71) is not deﬁned at the lower limit in the absence of the adiabatic factor eηt .64) naturally provides a regularization for the formal deﬁnition of the time evolution operator U (0. Furthermore.72) and an analysis as in (6. −∞).

by a phase since the S-matrix is unitary).76). as we have deﬁned earlier in (5. 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. therefore. The ﬁrst term in the expansion of the S-matrix in (6. (6.5 Normal ordered product and Wick’s theorem 233 |Ψ(∞) (out) = U † (0. Later we will see that the Feynman rules are a wonderful and simple way of systematizing these results. This shows that the “in” and the “out” states are related by the S-matrix in (6. 6. The calculation can be carried out in two steps.69) corresponds to no scattering at all.5 Normal ordered product and Wick’s theorem The normal ordered product. φ(+) . In terms of ﬁeld operators.67) (and. contains the annihilation operator while . Correspondingly. First. It is clear from this discussion that in calculating the S-matrix elements. we will simplify the time ordered product of factors of normal ordered terms. ∞)|Ψ (H) = U (∞. we can deﬁne S − ½ = T. −∞)|Ψ(−∞) = U (∞. since our (interaction) Hamiltonian is assumed to be normal ordered. we will derive a simple relation for the product of factors each of which is normal ordered and then using this relation.74) where the T-matrix represents the true nontrivial eﬀects of scattering. 0)U (0. we need to evaluate the matrix elements of time ordered products of operators between free particle states.57). (6. −∞)|Ψ(−∞) (in) = S|Ψ(−∞) (in) .73) where we have used the properties of the time evolution operator given in (6. Furthermore. 0)|Ψ (in) (H) = U (∞. such a calculation will involve time ordered products of normal ordered products and this is where the Wick’s theorem comes in handy. Sf i . simply corresponds to arranging factors in the product so that the creation operators stand to the left of the annihilation operators.6.

(6. annihilated by the annihilation operator.234 6 Self-interacting scalar field theory the negative energy part.76) Since the vacuum state is. 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). we have N (φ(x)) ≡ : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). Thus.75). N φ(−) (x)φ(+) (y) ≡ : φ(−) (x)φ(+) (y) : = φ(−) (x)φ(+) (y). as far as bosonic operators are concerned. by deﬁnition. we note from (6. has the creation operator so that this prescription is equivalent to the deﬁnition of normal ordering in (5. N φ(−) (x)φ(−) (y) ≡ : φ(−) (x)φ(−) (y) : = φ(−) (x)φ(−) (y). φ(−) . using the relations in (6.77) It is clear. for the bosonic scalar ﬁeld. (6. For example.76) that . Furthermore. (6. the order of the factors inside a normal ordered product is not important. N φ(+) (x)φ(−) (y) ≡ : φ(+) (x)φ(−) (y) : = φ(−) (y)φ(+) (x).75) We note that. therefore. N φ(+) (x)φ(+) (y) ≡ : φ(+) (x)φ(+) (y) : = φ(+) (x)φ(+) (y). we can equivalently say that φ(+) (x)|0 =0= 0|φ(−) (x).76)). that the vacuum expectation value of any normal ordered product vanishes.

(6. this also shows that if we can rewrite any product of operators (including normal ordered factors) in terms of normal ordered terms.158) and (6.79) where we have used (5.80) with the pairing of the two ﬁeld operators deﬁned to be related to the invariant positive energy Green’s function in (5.78) Therefore. in the simple example of the product of two ﬁeld operators. does not carry any nontrivial quantum number. (6.81) .6.151) as φ(x)φ(y) = −iG(+) (x − y). 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). let us note that. 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. the calculation of S-matrix elements will simplify enormously. (6. This gives another reason to choose the prescription of normal ordering for physical observables. φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) = : φ(x)φ(y) : −iG(+) (x − y). (6.76). Furthermore. To see how we can write a product of operators in normal ordered form. therefore.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. Let us denote (6.79) as φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y).

82) because the vacuum expectation value of the normal ordered product in (6. but since it is a c-number function.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. 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. On the other hand.80) that we can also identify φ(x)φ(y) = 0|φ(x)φ(y)|0 = −iG(+) (x − y). We know that by deﬁnition : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). the second term in (6.84) (6. Clearly.84) simply corresponds to a normal ordered product : φ(x) : φ(+) (y) = φ(x)φ(+) (y) = : φ(x)φ(+) (y) : . φ(−) (y) = = : φ(x)φ(−) (y) : + φ(+) (x). (6.84) gives : φ(x) : φ(−) (y) = φ(x)φ(−) (y) = φ(−) (y)φ(x) + φ(x). : φ(x) : : φ(y) : = : φ(x) : φ(+) (y) + : φ(x) : φ(−) (y). This brings out yet another connection between Green’s functions and the quantum ﬁeld theory.85) and (6. we note from (6.86) we obtain .83) We recognize that the ﬁrst term in (6. φ(−) (y) : φ(x)φ(−) (y) : − iG(+) (x − y). Therefore.80) is zero. let us next see how we can combine a product of normal ordered factors into normal ordered terms. (6.86) (6. Since we choose the Hamiltonian to be normal ordered.

87) This is. 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) : .87) shows how a product of two simple normal ordered factors can be expressed as a normal ordered term and a pairing. of course. (6.79) and (6. φ(−) (z) = : φ(x)φ(y)φ(z) : − iG(+) (x − z)φ(y) − iG(+) (y − z)φ(x) = : φ(x)φ(y)φ(z) : + φ(x)φ(z) : φ(y) : + φ(y)φ(z) : φ(x) : . φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + : φ(x)φ(y)φ(−) (z) : + φ(x)φ(y) + iG(+) (x − y) .89) Finally. (6. Next. (6. φ(−) (z) φ(y) +φ(x) φ(+) (y).88) φ(x)φ(y) = = but (6.6.5 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).80). namely. φ(−) (z) = : φ(x)φ(y)φ(z) : + φ(+) (x). : φ(x)φ(y) : −iG(+) (x − y) : φ(x)φ(y) : + φ(x)φ(y). let us note that . what we had seen earlier in (6.

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

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) : .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) : . 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.92) Similarly. 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. For example. It follows.80) that φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y). we already know from (6. (6. therefore. that (6.5 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).6.

For example. (6. (We are interested in vacuum expectation values because any matrix element can be written as a vacuum expectation value. This construction can also be extended to the case where only some of the factors are normal ordered.93) In other words. (6. In general.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).89) that .) An immediate consequence of this result is that the vacuum expectation value of an odd number of ﬁeld operators must vanish (since not all ﬁeld operators can be paired). we have already noted in (6. This has to be contrasted with the products of normal ordered terms which also had a similar expansion except that there were no pairings between terms within the same factor. 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).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 ) : + ··· .

6.6 Time ordered products and Wick’s theorem

241

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

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

: φ(x)φ(y) : φ(z)φ(w) =: φ(x)φ(y) :

: φ(z)φ(w) : + φ(z)φ(w)

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

T (φ(x)φ(y)) = T (φ(y)φ(x)) =

(6.97)

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

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

(6.98)

242

6 Self-interacting scalar field theory

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

θ(x) =

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

(6.99)

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

θ(x) = lim −

→0+

dk e−ikx . 2πi k + i

(6.100)

Of course, by deﬁnition

T (φ(x)) = φ(x) =: φ(x) : .

(6.101)

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

6.6 Time ordered products and Wick’s theorem

243

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

+θ x0 − z 0 θ z 0 − y 0 φ(x)φ(z)φ(y)

+θ y 0 − z 0 θ z 0 − x0 φ(y)φ(z)φ(x)

+θ y 0 − x0 θ x0 − z 0 φ(y)φ(x)φ(z)

+θ z 0 − x0 θ x0 − y 0 φ(z)φ(x)φ(y)

+θ z 0 − y 0 θ y 0 − x0 φ(z)φ(y)φ(x),

(6.102)

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

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

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

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

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

244

6 Self-interacting scalar field theory

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

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

(6.105)

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

φ(x)φ(y) = iGF (x − y).

(6.106)

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

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

(6.107)

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

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

(6.108)

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

6.6 Time ordered products and Wick’s theorem

245

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

φ(x)φ(y) = iGF (x − y) = φ(y)φ(x),

(6.110)

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

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

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

246

6 Self-interacting scalar field theory

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

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

6.7 Spectral representation and dispersion relation

247

0| [φ(x), φ(y)] |0

= −iG(x − y) =

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

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

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

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

n

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

= i

n

−eikn ·(x−y) 0|B(0)|n n|A(0)|0

,

(6.116)

248

6 Self-interacting scalar field theory

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

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

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

n

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

=

i (2π)3

n

= where

i (2π)3

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

(6.118)

ρ(q) = (2π)3

n

δ4 (q − kn )| 0|A(0)|0 |2 ,

(6.119)

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

6.7 Spectral representation and dispersion relation

249

GAA (x − y) = = =

i (2π)3

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

d4 q ρ(q 2 )

∞ 0

dσ 2 ρ(σ 2 ) i

=

0

∞

dσ 2 ρ(σ 2 ) G(x − y, σ 2 ),

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

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

∞ −∞

f (t) =

dk0 −ik0 t e f (k0 ). 2π

(6.123)

Using the integral representation of the step function

250

6 Self-interacting scalar field theory

θ(t) = lim −

ǫ→0+

1 2πi

∞ −∞

dk0

e−ik0 t , k0 + iǫ

(6.124)

**from (6.122) we obtain 1 2πi
**

∞ −∞

f (t) =

ǫ→0+

lim

−

dk0

e−ik0 t k0 + iǫ

′

∞ −∞

′ dk0 −ik0 t ′ ′ e f (k0 ) 2π

∞

=

ǫ→0+

lim −

−∞ ∞

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

=

ǫ→0+

lim − lim

−∞ ∞

=

ǫ→0+

−∞

dk0 −ik0 t e 2π

∞ −∞

′ ′ f (k0 ) dk0 ′ − k − iǫ . 2πi k0 0

(6.125)

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

∞ −∞ ′ dk0 ′ f (k0 ) , − k0 − iǫ

f (k0 ) = lim

ǫ→0+

′ k0

(6.126)

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

ǫ→0+

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

∞ −∞ ′ dk0 ′ f (k0 ) , ′ k0 − k0

f (k0 ) =

(6.128)

6.7 Spectral representation and dispersion relation

251

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

∞ −∞ ′ Im f (k0 ) , ′ k0 − k0

Re f (k0 ) =

′ dk0

Im f (k0 ) = −

∞

−∞

′ dk0

′ Re f (k0 ) . ′ k0 − k0

(6.129)

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

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

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

−∞

′ Im f (k0 ) , ′ k0 − k0 − iǫ

(6.130)

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

252

6 Self-interacting scalar field theory

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

|k0 |→∞

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

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

ǫ→0+

ǫ→0+

lim

−∞ ∞ −∞

′ dk0

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

ǫ→0+

lim

′ dk0

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

ǫ→0+

lim

1 ¯ π(k0 − k0 )

′ k0

×

1 1 . − ′ ¯ − k0 − iǫ k0 − k0 − iǫ

(6.133)

6.7 Spectral representation and dispersion relation

253

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

′ k0 ∞ −∞

¯ f (k0 ) − f (k0 ) =

ǫ→0+

lim ×

′ ′ ¯ dk0 Im (f (k0 ) − f (k0 ))

1 1 − ′ ¯ . − k0 − iǫ k0 − k0 − iǫ

(6.134)

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

Im f (k0 ) = −Im (−k0 ).

(6.135)

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

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

f (k0 ) − f (0) = lim

ǫ→0+

′ dk0

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

η→0+

(6.137)

This has the imaginary part given by

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

(6.138)

254

6 Self-interacting scalar field theory

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

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

ǫ→0+

lim = = = =

1 π

′ dk0

ǫ→0+

lim − lim −

π 2πEk

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

ǫ→0+

ǫ→0+

lim

1 2 (k0 + iǫ)2 − Ek 1

2 k0

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

η→0+

lim

−

2 Ek

+ ik0 η

,

(6.139)

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

6.8 References

255

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

This page intentionally left blank .

2) φ(x) = φ† (x) = where φ1 (x) and φ2 (x) are two distinct. 2 257 . 2 (7. 2 (7.2) can be inverted to give 1 √ φ(x) + φ† (x) . ∂µ ∂ µ + m2 φ† (x) = 0.1 Quantization The natural ﬁeld theoretic example to consider next is that of a free Klein-Gordon ﬁeld theory where the ﬁeld operator is not Hermitian. The relations in (7. there are now two distinct equations. 2 1 √ (φ1 (x) − iφ2 (x)) . φ† (x) = φ(x).3) φ1 (x) = i φ2 (x) = − √ φ(x) − φ† (x) . In this case. spin zero (scalar) ﬁelds which are Hermitian. (7. the dynamical equations take the forms ∂µ ∂ µ + m2 φ(x) = 0.1) Therefore.Chapter 7 Complex scalar ﬁeld theory 7. The physical meaning of this system can be understood by noting that we can express the complex ﬁeld operator in terms of two real (Hermitian) ﬁeld operators as 1 √ (φ1 (x) + iφ2 (x)) .

4) as the Euler-Lagrange equations is easily determined to be m2 2 1 m2 2 1 ∂µ φ1 ∂ µ φ1 − φ1 + ∂µ φ2 ∂ µ φ2 − φ 2 2 2 2 2 m2 1 ∂µ (φ1 − iφ2 ) ∂ µ (φ1 + iφ2 ) − (φ1 − iφ2 ) (φ1 + iφ2 ) 2 2 (7.258 7 Complex scalar field theory It follows now that we can express the dynamical equations in (7. The Lagrangian density which will give the dynamical equations in (7.4) Namely. ˙ ∂ φi (x) i = 1.5) L = = = ∂µ φ† ∂ µ φ − m2 φ† φ.) We can quantize this theory in one of the two equivalent ways. 2.1) in terms of the real ﬁeld variables as ∂µ ∂ µ + m2 φ1 (x) = 0. (7. and is manifestly Hermitian. Πj (y)]x0 =y0 .1) or (7. such a system equivalently describes two free. (7. ∂µ ∂ µ + m2 φ2 (x) = 0. Therefore. (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. (7. the equal time canonical commutation relations follow to be [φi (x).6) Consequently. [φi (x). independent spin zero Klein-Gordon ﬁelds (and. then we can deﬁne the momenta conjugate to the two ﬁeld operators as Πi (x) = ∂L ˙ = φi (x). φj (y)]x0 =y0 = 0 = [Πi (x).7) . we can think of the complex Klein-Gordon ﬁeld theory as describing two independent ﬁeld degrees of freedom – either in terms of φ1 (x) and φ2 (x) or through φ(x) and φ† (x) – each of which is treated as an independent variable. Πj (y)]x0 =y0 = iδij δ3 (x − y). If we treat φ1 (x) and φ2 (x) as the independent variables. two distinct particles) with degenerate mass. therefore.

7.8) where we have used the identiﬁcation in (7.5) to be Π(x) = = ∂L ˙ = φ(x) ˙ ∂ φ† (x) 1 1 ˙ ˙ √ φ1 (x) + iφ2 (x) = √ (Π1 (x) + iΠ2 (x)) . if we treat φ(x) and φ† (x) as the independent dynamical variables. 2 2 . then we can deﬁne the momenta conjugate to the ﬁeld operators from (7. the Hamiltonian density is obtained in a straightforward manner to be of the form 2 H = i=1 ˙ Πi φi − L 1 1 m2 2 = Π2 + Π2 − Π2 + ∇φ1 · ∇φ1 + φ 1 2 1 2 2 2 1 1 m2 2 1 φ − Π2 + ∇φ2 · ∇φ2 + 2 2 2 2 2 = 1 2 1 m2 2 Π1 + ∇φ1 · ∇φ1 + φ 2 2 2 1 1 1 m2 2 + Π2 + ∇φ2 · ∇φ2 + φ 2 2 2 2 2 2 = i=1 1 2 1 m2 2 Πi + ∇φi · ∇φi + φ . 2 2 2 i (7. (7.1 Quantization 259 In this case. 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.6).

Π† (y) = Π† (x).10).5) and (7. [Π(x). (7. we can expand the ﬁeld φ(x) in the basis of plane wave solutions of the Klein-Gordon equation (see (5. in the present case.8) and H= d3 x H. (7. (7. take the forms (we do not write the equal time condition explicitly for simplicity. φ† (y) = φ† (x). is manifestly Hermitian as it should be even though the dynamical variables are not. using (7. Π(y) = iδ3 (x − y).12) = Π† Π + ∇φ† · ∇φ + m2 φ† φ. φ† (y) = 0. Π† (y) = 0. of the form . φ(y)] = φ(x). although it is to be understood that x0 = y 0 in all of the relations in (7. which can be compared with (7. Π† (y) = φ† (x).13) Note that the Hamiltonian. 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 (7.2 Field decomposition As in the case of the Hermitian (real) Klein-Gordon ﬁeld. in this case.260 Π† (x) = ∂L ˙ = φ† (x) ˙ ∂ φ(x) 7 Complex scalar field theory 1 1 ˙ ˙ = √ φ1 (x) − iφ2 (x) = √ (Π1 (x) − iΠ2 (x)).10) 2 2 As a result.56)). in this case. φ(x).11) Furthermore. Π(y)] = Π(x).11)) [φ(x). 7. the equal time canonical commutation relations.

2 Field decomposition 261 φi (x) = d3 k (2π)3 2k0 √ e−ik·x ai (k) + eik·x a† (k) .2)) φ(x) = = 1 √ (φ1 (x) + iφ2 (x)) 2 1 √ 2 d3 k (2π)3 2k0 e−ik·x (a1 (k) + ia2 (k)) +eik·x a† (k) + ia† (k) 1 2 = φ† (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k) + eik·x b† (k) . As before. with i i = 1. j (7.14) with k0 = Ek = k2 + m2 > 0. e−ik·x b(k) + eik·x a† (k) . 2 (7. a† (k′ ) . 2. it is clear that the ﬁeld operators φ(x) and φ† (x) can also be expanded in the plane wave basis as (see (7. a† (k′ ) = δij δ3 (k − k′ ). (7. i j ai (k).16) where we have deﬁned a(k) = b(k) = 1 √ (a1 (k) + ia2 (k)) .66)) ai (k).17) .14). 2 1 √ (a1 (k) − ia2 (k)) .15) Given the decomposition in (7. ai (k) and a† (k). The commutation relations between these creation and annihilation operators can then be obtained from the canonical quantization relations to be (see (5. i i = 1. 2. (7. can then be interpreted as the annihilation and the creation operators for the two kinds of quanta associated with the ﬁelds φi (x). aj (k′ ) = 0 = a† (k).7.

a† (k). b† (k′ ) = b(k). a(k).19) with respect to the ai (k) operators and in that case the Hamiltonian would . (7. a(k′ ) = b(k). b(k′ ) = 0. (7.18) The Hamiltonian for the system (see (7. 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 (7.20) (We can equivalently normal order the Hamiltonian (7. The commutation relations between the a(k). b† (k′ ) = δ3 (k − k′ ).16)) and take the form a(k). b(k′ ) = a(k). a† (k′ ) = a† (k).19) If we normal order the Hamiltonian in (7. b† (k′ ) = 0.8) or (7. a(k). we can write : H := d3 k Ek a† (k)a(k) + b† (k)b(k) . in terms of the creation and the annihilation operators. (7.19) with respect to the a(k) and b(k) operators. b† (k′ ) = b† (k).17) b (or directly from (7.262 7 Complex scalar field theory which correspond to the annihilation operators associated with the quanta for the two ﬁelds φ(x) and φ† (x) respectively.11) and (7. a† (k′ ) = 0.12)) can be written. a† (k) and † (k) can now be obtained from (7.

in the present case. (7.22) ˜ |k Similarly. degeneracies arise when there is a symmetry present in the system under study.7. (7.21) However.3 Charge operator To understand the physical meaning of the two distinct and degenerate one particle states. therefore. Namely. energy) given by |k = a† (k)|0 . After all.3 Charge operator 263 be a sum of two Klein-Gordon Hamiltonians that we have already studied in chapter 5. The theory is. The degeneracy can be understood as follows. as we will see shortly. of course. Such a transformation is not a space-time symmetry transformation since . 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.23) where θ is a real constant parameter of transformation (a global transformation). manifestly Lorentz invariant and translation invariant. as we know from our studies in non-relativistic quantum mechanics. = b† (k)|0 . (7.) The vacuum state in such a case can be deﬁned to be the lowest energy state annihilated by both a(k) and b(k). φ† (x) → φ′ † (x) = φ† (x)eiθ . 7. However. there will now be two distinct one particle states degenerate in mass (and. the theory is invariant. under the transformation φ(x) → φ′ (x) = e−iθ φ(x). let us analyze the symmetry properties of this system. Namely. there is in addition an internal symmetry associated with the system. H|0 = 0. a(k)|0 = 0 = b(k)|0 = 0|a† (k) = 0|b† (k). as was also the case for a real Klein-Gordon theory.

∂µ φ′ † (x) −L φ(x). We note that under the global phase transformation (7.23) (or the inﬁnitesimal form (7.264 7 Complex scalar field theory the space-time coordinates are not changed by this transformation.13)) . ∂µ φ′ (x). for inﬁnitesimal transformations in (7. in this case. Inﬁnitesimally.24) where ǫ denotes the inﬁnitesimal constant parameter of transformation.5) is invariant under the global phase transformation in (7. (7. φ† (x). (7. (7.24)).25) The derivatives do not act on the exponentials since θ is a constant parameter (independent of space-time coordinates).26) On the other hand. we have (see (6. such a transformation is known as an internal symmetry transformation. Rather. the Lagrangian density (7.24)) showing that it is a symmetry of the system and in this case we have K µ = 0. ∂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 = 0.24).23) (or the inﬁnitesimal form of it in (7. φ′ † (x). Thus. δφ(x) = φ′ (x) − φ(x) = −iǫφ(x). the transformation in (7.23) takes the form (α = ǫ = inﬁnitesimal) δφ† (x) = φ′ † (x) − φ† (x) = iǫφ† (x). ∂µ φ(x). the change in the Lagrangian density (see (6.8)) is given by L φ′ (x).

as we have seen.28) and the parameter independent current follows to be ← → J µ = iφ† (x) ∂ µ φ(x). we normal order all observables such as currents and charges and. Of course. † ∂∂µ φ ∂∂µ φ µ Jǫ = δφ (7. (7.) Therefore. But here it is an operator. we have freely rearranged the order of factors in (7. (In fact.7. (7. (7. Even in a quantum theory.) Furthermore. the current can be easily checked to be conserved using the equations of motion in (7.29) This is a vector current since the parameter of transformation is a scalar.27). this is the probability current density (up to a normalization) which we had studied in the ﬁrst quantized theory.27) (Note here that classically the order of the factors is not important.16).30) Using the ﬁeld decomposition (7.13)) ← → ∂L ∂L + δφ† − K µ = iǫφ† (x) ∂ µ φ(x). we can deﬁne the N¨ther current associated with this o symmetry to be (see (6. under normal ordering the order of factors does not matter (for a bosonic theory).1) ∂µ J µ (x) = 0. therefore. we can show that the associated conserved charge of the theory can be written in terms of the creation and annihilation operators as .3 Charge operator 265 δφ ∂L ∂L + δφ† ∂∂µ φ ∂∂µ φ† = −iǫφ(x)∂ µ φ† (x) + iǫφ† (x)∂ µ φ(x) = iǫ φ† (x)∂ µ φ(x) − φ(x)∂ µ φ† (x) = iǫ φ† (x)∂ µ φ(x) − (∂ µ φ† (x))φ(x) ← → = iǫφ† (x) ∂ µ φ(x).

We note that the normal ordered charge operator (7.32) Since this charge operator is associated with a global phase transformation of the kind associated with electromagnetic interactions (namely. the charge operator takes the form : Q := d3 k a† (k)a(k) − b† (k)b(k) .34) d3 k′ a† (k′ )δ3 (k′ − k)|0 = a† (k)|0 = |k . Q|0 = d3 k a† (k)a(k) − b† (k)b(k) |0 = 0. a† (k) + a† (k)a(k′ ) |0 (7.32). a U (1) phase transformation) we can identify this operator as the electric charge operator. (7.22). 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′ ). . (7.31) If we normal order. Acting on the ﬁrst of the two one particle states in (7.33) so that the vacuum does not carry any charge. we are suppressing the normal ordering symbol for simplicity with the understanding that the charge operator has the normal ordered form in (7. Here. (7.266 7 Complex scalar field theory Q = = i = d3 x J 0 (x) = i ← → d3 x φ† (x) ∂ 0 φ(x) ˙ ˙ d3 x φ† (x)φ(x) − φ† (x)φ(x) d3 k a† (k)a(k) − b(k)b† (k) .32) annihilates the vacuum.

37) That is. Furthermore. the second one particle state of the theory |k is also an eigenstate of the charge operator but with the eigenvalue (−1). acting on the second one particle state in (7.35) ˜ = −|k . in particular. In contrast. This transformation is otherwise known as the charge conjugation symmetry which we will . from the form of the Hamiltonian (7.20) d3 k Ek a† (k)a(k) + b† (k)b(k) . the charge operator leads to ˜ Q|k d3 k′ a† (k′ )a(k′ ) − b† (k′ )b(k′ ) b† (k)|0 d3 k′ b† (k′ )b(k′ )b† (k)|0 d3 k′ b† (k′ ) b(k′ ). These are.3 Charge operator 267 In other words. Therefore. b† (k) + b† (k)b(k′ ) |0 = = − = − = − d3 k′ b† (k′ )δ3 (k′ − k)|0 = −b† (k)|0 (7. the second quantized theory leads to a much more satisfactory description of particles and anti-particles. (7.7. contains no particle and this should be contrasted with the deﬁnition of the Dirac vacuum discussed in section 2. the one particle state |k is an eigenstate of the charge operator with eigenvalue (+1). the second quantized description is symmetric under the interchange of particles and anti-particles. the particle and the anti-particle states – but both of them now correspond to positive energy states. of course.36) we note that it is invariant under the discrete transformation a(k) ↔ b(k). The vacuum in this bosonic theory. H= (7. They describe two distinct particles with degenerate mass but with exactly opposite electric charge.5.22). The meaning of the two one particle states is now clear. ˜ Namely.

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. to conclude this section. 7. Thus. 0|φ† (x)φ† (y)|0 = d3 kd3 k′ √ 0| e−ik·x b(k) + eik·x a† (k) 2(2π)3 k0 k′ 0 × e−ik ·y b(k′ ) + eik ·y a† (k′ ) |0 = 0. We note from (6.18) 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. Using the ﬁeld decomposition in (7. let us discuss the Feynman Green’s function in the basis of the complex φ(x) and φ† (x) ﬁelds.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. Since we have already constructed the Feynman Green’s function for the real Klein-Gordon ﬁeld theory in (5.146). Therefore.16) as well as the commutation relations in (7. 0|φ(x)φ† (y)|0 = d3 kd3 k′ √ 0| e−ik·x a(k) + eik·x b† (k) 2(2π)3 k0 k′ 0 ′ ′ ′ ′ . in this section.107) that the Feynman Green’s function can be related to the vacuum expectation value of the time ordered product of two ﬁeld operators. we will construct only the Feynman Green’s function for the complex KleinGordon ﬁeld theory.

Here we have used (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 (7.21) as well as the deﬁnitions of the positive and negative energy Green’s functions in (5.145). 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 = 0.38) = iG(−) (x − y). (7.39) .4 Green’s functions 269 × e−ik ·y b(k ) + eik ·y a† (k ) |0 = = = d3 kd3 k √ e−ik·x+ik ·y 0|a(k)a† (k )|0 2(2π)3 k0 k 0 d3 kd3 k √ 2(2π)3 k0 k 0 e−ik·x+ik ·y δ3 (k − k ) d3 k e−ik·(x−y) (2π)3 2k0 = −iG(+) (x − y).38). Given the relations in (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 = 0.7.

namely.146). 0|T (φ† (x)φ(y))|0 = 0|T (φ(y)φ† (x))|0 = iGF (y − x) = iGF (x − y). Thus. 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).2) (in order to be consistent with the standard notation used in the discussion of spontaneous symmetry breaking in the literature). 2 (7. (7. 4 L = ∂µ φ† ∂ µ φ − m2 φ† φ − λ > 0.5 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. deﬁning 1 φ(x) = √ (σ(x) + iχ(x)) . we see that in this case. Denoting φ1 = σ and φ2 = χ in the deﬁnition in (7. in this case we should look for interactions that would preserve the internal global symmetry in (7.270 On the other hand.40) Here we have used the deﬁnition of the Feynman Green’s function in (5. However. It is easy to check that the Lagrangian density with the quartic interaction λ † 2 (φ φ) . (7.41) describes the most general (renormalizable) self-interacting theory for the complex Klein-Gordon ﬁeld.42) .23). 7. there is only one independent Feynman Green’s function that is related to the vacuum expectation value of the time ordered product of φ(x)φ† (y).

7. ∂σ ∂χ (7.46) In general.41) has the form λ † 2 (φ φ) . this describes the theory of two real Klein-Gordon ﬁelds with quartic self-interactions and which are also interacting with each other. 16 L = (7. the inﬁnitesimal transformations (7. 4 H= d3 x Π† Π + ∇φ† · ∇φ + m2 φ† φ + (7. ∂V ∂V =0= .45) and we note that.42). (7. for constant ﬁeld conﬁgurations.24) can be shown to have the form δσ(x) = ǫχ(x). In the real basis of (7.43) Namely. The Lagrangian density (7.44) The Hamiltonian for the interacting theory (7.23) or the inﬁnitesimal form in (7.47) .43) is invariant under the global phase transformation (7.) λ † 2 (φ φ) 4 2 V = m2 φ† φ + = λ m2 2 σ + χ2 + σ 2 + χ2 2 16 . the minimum of energy coincides with the minimum of the potential (We note that if the ﬁeld φ is not constant. ∂φ ∂φ† or.24).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 . the minimum of the potential is obtained by requiring ∂V ∂V =0= . the other (kinetic energy) terms would only add positively to the energy. δχ(x) = −ǫσ(x). (7.41) or (7.5 Spontaneous symmetry breaking and the Goldstone theorem 271 we note that the Lagrangian density (7.

51) are clearly invariant under the global phase transformation in (7. c χc = 0|χ(x)|0 = 0. 16 L = (7. Let us next consider the same Lagrangian density for the complex scalar ﬁeld as in (7. c or.106)). and perturbation theory developed around such a vacuum would be stable.48) As we know from studies in classical mechanics. (7.272 7 Complex scalar field theory In the present case. Changing the sign of the mass term does not change the .51) Both the Lagrangian densities in (7. or. since the potential in (7. In quantum theory. 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 .42). 4 λ > 0. this translates to the fact that we can deﬁne the ground state of the theory such that φ† = 0|φ† (x)|0 = 0.56) has exactly this property (see (5. perturbation around the minimum of the potential is stable.50) and (7.41) except that we change the sign of the mass term L = ∂µ φ† ∂ µ φ + m2 φ† φ − λ † 2 (φ φ) .49) φc = 0|φ(x)|0 = 0. We note from our discussion of the real Klein-Gordon theory that the plane wave expansion for the ﬁeld operator in (5. equivalently σc = χc = 0.46) is a sum of positive terms. σc = 0|σ(x)|0 = 0. (7. it is clear that the minimum occurs at (the subscript c is supposed to denote a classical ﬁeld.24) and (7. (7.50) In terms of the real ﬁelds σ and χ deﬁned in (7. a concept that we do not get into here) φc = φ† = 0.23) (or (7.44)).

5 Spontaneous symmetry breaking and the Goldstone theorem 273 symmetry behaviour.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. corresponding to the Einstein relation (7. therefore.4)) − m2 φ(x) = 0. has many undesirable features. but it does change the Hilbert space structure of the theory in a very interesting way. 2 − m2 ∂|k| k (7.48) .51) has the form V = −m2 φ† φ + = − λ † 2 (φ φ) 4 2 m2 2 λ σ + χ2 + σ 2 + χ2 2 16 .52) k2 = −m2 . Such particles are known as tachyons and a theory with tachyons.50) or (7. for constant ﬁeld conﬁgurations the potential of the theory (7.7. in general. (7. in this case.40) or (5. (7.55) The minimum of this potential is obtained from (7. or.54) Namely.41) would lead to the dynamical equation (compare with (1. c = 1) leading to an acausal propagation (and. Let us note that. the particle will travel faster than the speed of light (recall that in our units. a violation of causality). To begin with. the theory (7. let us note that in the absence of interaction (λ = 0). k0 = k2 − m2 .

60) 2m σc =± √ .χc =0 λ = σc =0=χc ∂2V ∂χ2 σc =0=χc = −m2 . for example. = 0.59) = 0. 3λ 4 λ 4 4m2 λ 4m2 λ = 2m2 .56). (7.56) = χ −m2 + There now appear to be two sets of solutions to the equations in (7. we can choose σc = 0 = χc .χc =0 λ . let us analyze the second derivatives of the potential at the two extrema in (7.χc =0 λ (7. λ χc = 0. σc =0=χc = −m2 + = −m2 + = 0. (7.57) which lead to ∂2V ∂σ 2 ∂2V ∂σ∂χ while ∂2V ∂σ 2 ∂2V ∂χ2 ∂2V ∂σ∂χ 2m σc =± √ . (7.274 7 Complex scalar field theory ∂V ∂σ = −m2 σ + λ σ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.57) λ For the second of these solutions. namely. 2 σc + χ2 = c 2m σc = ± √ . 2m σc =± √ .58) To determine the true minimum. (7. = σ −m2 + ∂V ∂χ = −m2 χ + λ χ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0. 4m2 .

61). . There are several interesting things to note from this analysis.1: Potential leading to spontaneous breaking of symmetry. the ﬁrst corresponds to a local maximum.44). χ) V =0 χ 2m −√ λ 2m √ λ σ Figure 7. leads to a lower energy. However. the second solution in (7. On the other hand.1.57) is a true minimum and we conclude that the true minimum of the potential occurs for 2 2φ† φc = σc + χ2 = c c 4m2 . V 2 ((σc + χ2 ) c m2 4m2 ) = − = λ 2 = − 4m2 λ λ + 16 4m2 λ 2 2m4 m4 m4 + =− . we note explicitly from the form of the potential in (7. All of this is more easily seen from the graph of the potential in Fig.55) that V (σc = χc = 0) = 0. 7.61) In fact.57). indeed. V (σ. Each point on this circle is related by the symmetry transformation (7. First. we note that there is an inﬁnity of minima (in the σ-χ ﬁeld space) of the potential given by the circle (7.5 Spontaneous symmetry breaking and the Goldstone theorem 275 In other words we note that out of the two solutions in (7. λ (7.62) so that the second solution. λ λ λ (7.7.

We recall that any choice for the minimum of the potential would translate into the properties of the vacuum state of the quantum theory. From (7.64) We recall that conserved charges are the generators of inﬁnitesimal symmetry transformations of the theory and generate the transformations through commutators. χ(x)] = −ǫσ(x). λ χc = 0. This has to be contrasted with the case where an interaction term in the Hamiltonian can break the symmetry of a theory explicitly. σ(x)] = ǫχ(x). λ 0|χ(x)|0 = 0. in the quantum theory this would imply (see (7.64).66) which implies that . In this case. the action or the Hamiltonian of the theory is invariant under the symmetry transformation. In the quantum theory. χ(x)] |0 (7.63) then. for example.44) can be written in terms of the commutators with the charge operator as δσ(x) = −iǫ [Q. δχ(x) = −iǫ [Q. it now follows that = −iǫ 0| [Q. (7. but the ground state is not.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. the inﬁnitesimal transformations in (7.65) 0|δχ(x)|0 2m = −ǫ 0|σ(x)|0 = − √ ǫ.61) to correspond to the point (in the σ-χ ﬁeld space) 2m σc = √ . λ (7.49)) 2m 0|σ(x)|0 = √ . (7. such a breaking manifests in the following manner.44) spontaneously. In this case. if we choose the minimum in (7. Thus.

(note that Q is Hermitian) χ|χ = = = 0|QQ|0 0| d3 x J 0 (x. [Q. or.70) (7.68) Since Q does not annihilate the vacuum of the present theory. H] = 0. or. H|χ = E0 |χ . HQ|0 = E0 Q|0 .69) would appear to be degenerate with the vacuum state |0 in energy. This can be easily seen from (7. (7.71) . (7. namely. although the Hamiltonian of the theory is. it is suggestive that we can think of this state as another vacuum state.68) and (7. Therefore. H]|0 = 0.5 Spontaneous symmetry breaking and the Goldstone theorem 277 Q|0 = 0.69) that [Q. The problem with this interpretation is that the state is not normalizable.67) This explicitly shows that the vacuum state of the theory is not invariant under the symmetry transformation in this case.69) where we have denoted the energy of the vacuum |0 as E0 and this derivation shows that the state |χ deﬁned in (7. let us denote Q|0 = |χ It follows now from (7. t)Q|0 d3 x 0|eiP ·x J 0 (0)e−iP ·x Q|0 .69) as follows. (7. (7.31) and (7. or. (QH − HQ)|0 = 0.7.

(7. (7. therefore. This analysis also shows that the ﬁnite transformation operator U (θ) = e−iθQ .75) does not act unitarily on the Hilbert space.278 7 Complex scalar field theory We have already seen that the Hamiltonian commutes with Q expressing the fact that it is independent of time. Since Q does not depend on spatial coordinates. we have [Pµ .74) In other words.66) that χ(x) corresponds to the Nambu-Goldstone boson of the theory. Q] = 0. and as a consequence.72) leads to (7. 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) takes a state out of the Hilbert space. .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 −→ ∞. In fact. (7. it follows that the momentum operator also commutes with Q and. 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. the state |χ is not normalizeable and hence cannot be thought of as another vacuum. (7.73) where we have used the property of the ground state Pµ |0 = 0.

is given by σ(x) = σ ′ (x) + σ with the later identiﬁcation σ ′ (x) = σ(x) as the dynamical variable.77). λ χ = 0|χ(x)|0 = 0. 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. With the choice of the solutions for the minima as in (7. We note from Fig.77) so that the shifted ﬁelds have vanishing vacuum expectation values (vevs) 0|σ(x)|0 = 0 = 0|χ(x)|0 . (7. λ χ(x) → χ(x) + χ = χ(x).5 Spontaneous symmetry breaking and the Goldstone theorem 279 The second interesting observation that follows from the analysis in (7.64) 2m σ = 0|σ(x)|0 = √ .60) is as follows. the shift in (7.51) becomes . the Lagrangian density in (7.7.76) we can shift the ﬁeld variables as (in classical mechanics this corresponds to analyzing small oscillations around the minimum of a potential) 2m σ(x) → σ(x) + σ = σ(x) + √ .54))). (7. 7.60) shows that in this case of spontaneous breakdown of the symmetry. We can see this quantitatively in the quantum theory as follows. √ 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 (7.53) and (7.77). (7.) Under the shift (7.1 that the massless mode would correspond to oscillations along the valley of minima while the massive mode would represent the orthogonal oscillations. say for the ﬁrst term.78) (Explicitly.

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. of course.79) This shows explicitly that perturbed around the true minimum. then there must appear massless excitations.280 7 Complex scalar field theory L = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 λ − 16 2m σ+ √ λ 2 2 2m σ+ √ λ 2 + χ2 +χ 2 = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 λ − 16 2 2 4m 4m2 σ 2 + χ2 + √ σ + λ λ 2 4m2 4m σ +χ + √ σ+ λ λ = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 − + λ 16 (σ 2 + χ2 )2 + 4m2 4m σ 2 + χ2 + √ σ + λ λ 8m 16m2 2 16m4 σ + + √ σ(σ 2 + χ2 ) 2 λ λ λ 8m2 2 32m3 (σ + χ2 ) + √ σ λ λ3 = 1 m4 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ − m2 σ 2 + 2 2 λ √ m λ λ − σ(σ 2 + χ2 ) − (σ 2 + χ2 )2 . For example. aware of this phenomenon (known as the spontaneous symmetry breaking phenomenon) from studies in other areas of physics.66)). if we look at the Hamiltonian describing a . 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. 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. 2 16 (7. a continuous symmetry is spontaneously broken. We are.

(7. it is clear that once we have chosen a particular ground state (vacuum) where all the spins are pointing in a given direction. (The π meson which is the lightest among the spin zero bosons has a mass mπ ∼ 140 MeV. we can ask how such a system can be coupled to a background electromagnetic ﬁeld.53). the Hamiltonian is invariant under rotations.) 7. in the case of the Ising model). The prescription is again through the minimal coupling discussed in (1. Thus.7. (7.81) where “e” denotes the charge carried by the particle.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. H = −g si · si+1 . Given the complex Klein-Gordon ﬁeld theory. we do not observe any massless spin 0 bosons in nature.) However. Unfortunately. The analogue of the massless Nambu-Goldstone bosons in this case correspond to the long range correlations associated with the spin waves.80) i where the negative sign in the Hamiltonian (g > 0) is for ferromagnets. 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 . when the spins are all pointing up or down.6 Electromagnetic coupling 281 magnet (for example. there are two such vacuum conﬁgurations allowed in one dimension. namely. The ground state of the theory is clearly the state (spin conﬁguration) where all the spins are aligned. ∂µ → ∂µ + ieAµ . we have eﬀectively chosen a deﬁnite orientation in the physical space. (For example. Consequently the rotational symmetry is spontaneously broken. 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.

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

Physical Review 122.86) Consequently. (In contrast. Local phase transformations are commonly known as gauge transformations and we note that. the local phase invariance (7. Salam and S. Row. J.82). Aµ (x)) → L(φ′ (x).84) requires an additional ﬁeld (Aµ ). Weinberg. unlike the global phase invariance in (7. 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. 345 (1961). φ′ † (x). Jona-Lasinio. Schweber. (7.84).7.84) L(φ(x). which is obtained by minimally coupling the complex scalar ﬁeld to the electromagnetic ﬁeld. 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. Of course. Y. the theory described by (7. Aµ . Evanston (1961). invariance under a local transformation necessarily requires additional ﬁelds known as gauge ﬁelds. Peterson. φ† . 154 (1961). 3.84)).7 References 1. Physical Review 127. S. J. Introduction to Relativistic Quantum Field Theory. note that ∂µ φ(x) does not transform covariantly. 7. the covariant derivative acting on a ﬁeld transforms exactly like the ﬁeld itself (covariantly) under such a transformation (see (7.7 References 283 In other words. Nuovo Cimento 19. Nambu and G. is invariant under the local phase transformation (7. namely. Goldstone. 4.) It follows from this observation that under the local transformation (7.23). 2. φ† (x). 965 (1962). Goldstone. This is a very general feature in a quantum ﬁeld theory. . A.

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

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

there can be at the most one quantum in a given state. the wave function of the system will have the necessary antisymmetry property characteristic of fermions.5) This is. 1. that in dealing with fermionic ﬁeld operators. 4. a† F + − a† aF aF F (8. (8. This equation is manifestly Lorentz covariant and the adjoint equation has the form . of course. we will be naturally dealing with anti-commuting variables (these variables are commonly known as Grassmann variables) and anti-commutation relations. F F F In other words.2) automatically lead to NF (NF − 1) = 0.4) which shows that the eigenvalues of the number operator.286 then. in such a theory.6) where ψα (x). It is clear. what the Pauli exclusion principle would say. has the form (iγ µ ∂µ − m) ψ = 0. the anti-commutation relations in (8. therefore. it leads to the fact that with the anti-commutation relations. the Dirac equation for a massive spin particle. 1 2 (8. is a four component complex spinor function. (8. α = 1. In fact. we obtain 2 NF = a† aF a† aF F F 8 Dirac field theory = a† F aF . can only be nF = 0.3) = a† 1 − a† aF aF = a† aF = NF .2 Quantization of the Dirac ﬁeld As we have seen in (1. 2. if this system is analyzed in more detail. Namely. 8. 3.96).

10) . (8. the Lagrangian density of the form / L = ψ (iγ µ ∂µ − m) ψ = iψ∂ ψ − mψψ.2 Quantization of the Dirac field 287 ← − ψ iγ µ ∂µ + m = 0. the 1 Dirac ﬁeld operators are expected to belong to the (reducible) spin 2 representation of the Lorentz group (see (4.8) In trying to second quantize the Dirac equation (theory).6) and (8. However. where the adjoint spinor is deﬁned to be (see also (2.9) is manifestly Lorentz invariant (recall the transformation of the Dirac bilinears under a Lorentz transformation discussed in section 3. In dealing with the Dirac ﬁeld theory. we would like to decide on a Lorentz invariant Lagrangian density which would give rise to the two Dirac equations (8.9) give ∂L = (iγ µ ∂µ − m) ψ = 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. Clearly. in this case.61)) describing fermionic particles and. (8. In deriving the Euler-Lagrange equations.3). In general. we have to be careful about signs that may arise in taking derivative operators (with respect to these variables) past other Grassmann variables. ∂ψ ∂∂µ ψ (8. With this. need to be treated as anti-commuting operators.7) (8. the Euler-Lagrange equations following from the Lagrangian density (8. however. ∂ψ ← − ∂L ∂L − ∂µ = mψ − ∂µ (−iψγ µ ) = ψ iγ µ ∂µ + m = 0. let us note that since ψ and ψ are anti-commuting (Grassmann) variables.7) as the Euler-Lagrange equations.41)) ψ(x) = ψ † (x)γ 0 . consequently.8. as we have seen in chapter 3 as well as in chapter 4. we will always use the convention of taking fermionic (Grassmann) derivatives from the left.

we note that the Lagrangian density in (8.12) diﬀers from L in (8. 2 2 L′ = (8. (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. the action associated with it is since a total divergence does not contribute to the action.13) = L − ∂µ ψγ µ ψ .8) as well as the fact that γ 0 is Hermitian. 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.9) is Lorentz invariant and gives the two Dirac equations as the Euler-Lagrange equations.6) and (8. It is worth noting here that even though the Lagragian density (8. it is not Hermitian. An alternate Lagrangian density which is Lorentz invariant. In fact. (8.288 8 Dirac field theory which are exactly the two Dirac equations (8.83). 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. Namely. Even though the Lagrangian density (8.12) However. 2 .9) only by a total divergence.7) that we are interested in.11) † where we have used (2.9) is not Hermitian.

we use L in (8. (Πψ )β (y)]+ = −iδαβ δ3 (x − y).14) Πψ = Πψ = Consequently.15) and (8. ψβ (y)]+ = 0 = [(Πψ )α (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 † . ψβ (y) + + = 0. 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.8. (Πψ )β (y) or. but we should understand that x0 = y 0 in all the relations in (8. ˙ ∂ψ (8. ψβ (y) + = −iδαβ δ3 (x − y).15) The negative sign in the last relation in (8. However.16)) [ψα (x). = 0. the second and the third relations in (8. † † ψα (x).14).12) which naturally leads to a symmetric stress tensor (see (6. (8.27)) for the theory.15) is a reﬂection of our choice of left derivatives. it is more appropriate to use the Lagrangian density in (8. (8. [ψα (x).16) = δαβ δ3 (x − y). + † = ψα (x).9) to describe the Dirac theory (equations). If we work with the Lagrangian density (8.15) can also be written as (Πψ )α (x). ˙ ∂ψ ∂L = 0. † ψα (x). Using the ﬁrst relation in (8. (Πψ )β (y) or.9) to describe the dynamics of the system.2 Quantization of the Dirac field 289 For this reason as well as for simplicity. in the presence of gravitation. (Πψ )β (y)]+ . . −iψβ (y) + ψα (x).

17) Using the ﬁeld anti-commutation relations (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.) The Hamiltonian density for the Dirac ﬁeld theory (8.19) .15) and (8.16). ψβ (y) + −iγ 0 γ · ∇y ψ(y) + β † + m ψα (x). H = d3 y † ψα (x). it is now easy to check that this Hamiltonian leads to the two Dirac equations as the Hamiltonian equations.) ˙ H = −Πψ ψ − L ˙ ˙ = − −iψ † ψ − iψ † ψ − iψγ · ∇ψ + mψψ = −iψγ · ∇ψ + mψψ. For example.9) has the form (The negative sign in the ﬁrst term arises from our convention of choosing left derivatives for fermions. (8. ˙ iψα (x) = ψα (x).14) can be thought of as constraints on the dynamics of the system.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. leading to the Hamiltonian H= d3 x H = d3 x −iψγ · ∇ψ + mψψ . (8.18) (8.15) and (8. Such systems can be systematically studied through the method of Dirac brackets and such an analysis leads to the conclusion that the relations (8. We will discuss the method of Dirac in more detail in chapter 10.

or. H d3 y i ψ(y)γ · ∇y β † ψα (x).16) as well as the Hamiltonian (8.7) as the Hamiltonian (Heisenberg) equations. or. (8. or. ψβ (y) + x0 =y 0 = d3 y i ψ(y)γ · ∇y β δαβ δ3 (x − y) x0 =y 0 −mψ β (y)δαβ δ3 (x − y) ← − = −i ψ(x)γ · ∇ α − mψ α (x). or.6) and (8. (8. (iγ µ ∂µ − m) ψ(x) = 0.20) Similarly.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).22) Namely. ψβ (y) + † −mψ β (y) ψα (x). (8.21) and this can be written in the matrix form as ← − ˙ iψ † = −iψγ · ∇ − mψ.15) and (8. .18) indeed reproduce the two Dirac equations (8. ˙ iγ 0 ψ(x) = −iγ · ∇ψ(x) + mψ(x). ← − ˙ iψγ 0 = −iψγ · ∇ − mψ. ← − ψ iγ µ ∂µ + m = 0. the anti-commutation relations (8.8. we obtain ˙† iψα (x) = = † ψα (x).

where c(k.46). s) and d† (k. s)vα (k. let us note that the ﬁeld decomposition in (8. s)vα (k′ . s′ ) d† (k′ . s) and vα (k. Therefore.4). s) are anti-commuting operators. s′ ) c(k′ .23) automatically satisﬁes the Dirac equation in this basis (i∂ − m) ψ(x) = 0. As in the case of the Klein-Gordon ﬁeld (see section 5.292 8. s)ψα (x) α (2π)3 k0 d3 x d3 k ′ m ′ √ ei(k−k )·x u† (k.94) and the ﬁeld (8. s) + eik·x d† (k. s)uα (k. s) denote the positive and the negative energy spinors deﬁned in (3. / (8. we write √ (k0 = Ek = k2 + m2 > 0) ψα (x) = 1 s=± 2 d3 k m (2π)3 k0 (8.5).23) has been chosen keeping in mind the massive normalization of the plane wave solutions in (2.23) can be inverted to give √ (k0 = Ek = k2 + m2 > 0) d3 x = s′ m eik·x u† (k. Taking advantage of our experience with the complex Klein-Gordon ﬁeld theory. s′ ) α ′ . s)uα (k′ .24) The speciﬁc form of the prefactor in (8. uα (k.93) and (3. s′ ) α 3 (2π) k0 k′ 0 +ei(k+k )·x u† (k. we can expand the Dirac ﬁeld operator in the basis of these plane wave solutions. s) .3 Field decomposition 8 Dirac field theory We have already seen that the positive and the negative energy plane wave spinor solutions of the Dirac equation deﬁne a complete basis for the space of four component spinor functions (see the discussion in section 3.23) × e−ik·x c(k.

27) c† (k. c† (k. (8. s) in a straightforward manner from the quantization conditions (8. s).29) + . = δαβ δ3 (x − y).92)). s).3 Field decomposition 293 δ3 (k − k ) u† (k. d† (k . s). (8. s ) d† (k .28) Therefore. d† (k.28) and they correspond to c(k. s).16) to be ψα (x).27). s) = d(k. s ) + . s) and d(k. s)vα (k . s). (2π)3 k0 m † eik·x ψα (x)vα (k. (2π)3 k0 (8.25)-(8. ψβ (y) † ψα (x). s ) c(k . s) = d3 x d3 x Furthermore. s ) α 0 x0 = s d3 k √ m k0 k 0 + δ3 (k + k ) e2ik = s u† (k. ψβ (y) +.x0 =y 0 +. s). ψβ (y) † † ψα (x).8. s ) = c(k.46) (see also (3. s ) α (8. using the inversion relations (8. s)ψα (x) = d† (k. (2π)3 k0 d3 x (8.x0 =y 0 = 0. = 0.25) m k0 δss c(k. s). it can be shown that m † e−ik·x vα (k.25) and (8. s). k0 m where we have used the massive normalization for the spinors discussed in (2.26) that m † e−ik·x ψα (x)uα (k.15) and (8. s)uα (k . In a similar manner. c† (k .26) It follows from (8. s ) = δss δ3 (k − k ) = d(k. we can now obtain the anti-commutation relations satisﬁed by the operators c(k.x0 =y 0 +. we know the equal time quantization conditions for the ﬁeld operators from (8.

30) (k + m) v(k. If we now use the relations (3. s) = 1 s. s) = (γ 0 k0 − γ · k + m)v(k.s′ =± 2 ′ d3 x ψ (−iγ · ∇ + m) ψ m k0 ′ m 3 d k k′ 0 × u(k′ . s). s) = (γ 0 k0 − γ · k − m)u(k.30) and write (k − m) u(k. can be written out in terms of creation and annihilation operators as H= = d3 x −iψγ · ∇ψ + mψψ = d3 x (2π)3 d3 k ′ 1 s. s′ )eik ·x + v(k′ . / (8.94) (8. s) = 0. d† (k. c† (k.18) of the theory. s) + eik·x d† (k. s)u(k. s) . s) = 0. s′ )c† (k′ . s) + eik·x d† (k. s′ )e−ik ·x × e−ik·x c(k. s)(γ · k + m)u(k. s′ c† (k′ s′ )eik ·x + v k′ . in this case. s′ d(k′ . This shows that we can think of c(k.93) and (3.294 8 Dirac field theory with all other anti-commutators vanishing. s)v(k. The Hamiltonian (8.31) . s) as well as d(k. s′ )d(k′ . / we can simplify the Hamiltonian in (8. s) as annihilation and creation operators for the two kinds of fermionic quanta in this theory (much like the complex Klein-Gordon ﬁeld theory). s′ )e−ik ·x × (−iγ · ∇ + m) × e−ik·x c(k. s).s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m k0 ′ × u k′ . s)(−γ · k + m)v(k.

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

(8. s)|0 d(k. s)d(k. every time we move a fermionic operator past another in order to bring the product into its normal ordered form. = 0.36) |k. s). = d† (k. we can deﬁne the vacuum state of the theory to correspond to the state which is annihilated by both c(k. s) and d(k. s)d(k′ . s |k.296 8 Dirac field theory way as for the bosonic operators. However.34) even though we do not put the normal ordering symbol explicitly. s). (8. s′ ) : = d† (k. we pick up a negative sign since they anti-commute (as opposed to bosonic variables). s)|0 . s′ ). namely.) = 0 = 0|c† (k. s) (The Hamiltonian is understood to be normal ordered as in (8. we now have : d† (k. s) : = −d† (k. s) .35) c(k. = 0 = 0|d† (k.21)).33). s′ ).33) so that the order of factors inside a product that is normal ordered becomes important. s)c(k. s)|0 H|0 The one particle states of the theory can now be deﬁned as (see (7. s . we write products with the creation operators standing to the left of the annihilation operators.32) is obtained to have the form : H := s=± 1 2 d3 k Ek c† (k. (8. : d(k′ . s)d(k′ . s)|0 .34) As in the case of the complex (Klein-Gordon) scalar ﬁeld theory (see (7. Using (8. s)d(k′ . (8. s) + d† (k.22)) = c† (k. the normal ordered Hamiltonian for the free Dirac ﬁeld theory in (8. s′ )d† (k. For example.

ψ(x) → ψ (x) = ψ(x)eiθ . ψ ) = ψ (iγ µ ∂µ − m) ψ ′ ′ ′ ′ (8. the Lagrangian density (8. ′ (8.39) = ψeiθ (iγ µ ∂µ − m) e−iθ ψ = ψ (iγ µ ∂µ − m) ψ = L(ψ. we need to analyze the symmetry properties of the theory.38) (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. δǫ ψ = ψ (x) − ψ(x) = iǫψ(x). 8. ψ).40) .4 Charge operator The Dirac Lagrangian density (8.8.37) or the inﬁnitesimal form of it (ǫ is the constant inﬁnitesimal parameter of transformation) δǫ ψ = ψ ′ (x) − ψ(x) = −iǫψ(x).9) (or (8.34) with energy eigenvalue k0 = Ek = k2 + m2 > 0.12)).4 Charge operator 297 Both these states can be checked to be eigenstates of the Hamil√ tonian (8. ψ) → L(ψ ′ . transforms as L(ψ.9) L = ψ (iγ µ ∂µ − m) ψ. (8. there are two distinct one particle states with degenerate mass (energy) and to understand the meaning of these states. We note that under the global phase transformation (α is the constant parameter of transformation) ψ(x) → ψ ′ (x) = e−iθ ψ(x). Therefore. as in the case of the complex scalar ﬁeld theory.

the Lagrangian density (8. where we have rearranged the fermionic operators keeping in mind their anti-commuting nature.7) that this current is conserved.26). as in the case of the complex scalar ﬁeld theory in (7.42) is consistent with the convention of taking left derivatives for the fermion ﬁelds) ∂L ∂L + δǫ ψα ∂∂µ ψα ∂∂µ ψ α (8.9) is invariant under the global phase transformation (8.38)) is obtained from (6.43) where J µ (x) = ψ(x)γ µ ψ(x). (8.37) (or (8.41) On the other hand. ∂µ J µ (x) = 0. (8. It is a vector current as in the case of the complex Klein-Gordon ﬁeld theory (see (7.298 8 Dirac field theory Namely. this would correspond to the total probability) . we note that K µ = 0.45) The conserved charge associated with the symmetry transformation (8.38)).44) represents the current independent of the parameter of transformation. we obtain (see (6.38).27) and we note that the form of the expression in (8.37) (or (8.42) δǫ ψ α = (−iǫψ)α (−iψγ µ )α = ǫψ(x)γ µ ψ(x). the N¨ther current associated with the internal symmetry o transformation (8.6) and (8.37) of the system is given by (note that in the ﬁrst quantized theory. (8.13) and (7. Thus.13) to be µ Jǫ (x) = ǫψ(x)γ µ ψ(x) = ǫJ µ (x). In this case.29)) and it can be checked using the equations of motion (8. using the inﬁnitesimal transformation (8. (8.

s)|0 = 1 s′ =± 2 d3 k′ c† (k′ . = 1 s′ =± 2 d3 k′ c† (k′ . s)|0 = |k.47) As in the case of the complex Klein-Gordon ﬁeld. s′ ) c† (k. s′ )δss′ δ3 (k′ − k)|0 = c† (k. s) − d† (k.38)). Therefore. Using the ﬁeld decomposition in (8. lepton number.23).34)) Q|0 Q|k. s) .37) (or (8.4 Charge operator 299 Q= d3 x J 0 (x) = d3 x ψ † (x)ψ(x).33) and (7. s′ ) − d† (k′ . s)c(k. s′ ). it is also possible to associate the charge operator with a fermion number (such as baryon number. s = 0. s′ ) |0 = 1 s′ =± 2 d3 k′ c† (k′ . s . But. (8.46) is associated with an Abelian symmetry transformation (8. we can now show that the charge operator (normal ordered) satisﬁes (see (7. s) + − c† (k. we can express (8. as in the case of the complex Klein-Gordon ﬁeld we can associate the charge operator with that of the electric charge. s′ )d(k′ . .46) in terms of creation and annihilation operators as : Q := s=± 1 2 d3 k c† (k. s′ ) c(k′ . since we are dealing with a fermion system. s′ )c(k′ . s)d(k. c† (k. etc) which are associated with U (1) symmetry transformations as well. s)c(k′ .8.46) We note here that the charge (8. (8.

s = 1 s′ =± 2 8 Dirac field theory d3 k′ c† (k′ .34) is symmetric under the discrete transformation (see also (7. 8. s′ ) 1 s′ =± 2 d(k′ .300 Q|k.5 Green’s functions As we have discussed within the context of the Klein-Gordon theory. s) ↔ d(k. s)|0 = −|k. Thus. the vacuum does not contain any particle as we would intuitively expect (compare this with the ﬁrst quantized description of the vacuum in section 2. s) + − d† (k. s . s′ )d(k′ .37)) c(k. in the second quantized description of the Dirac theory. we can identify them with the particle and the anti-particle states of the theory with positive energy. s′ ). s′ ) |0 = − d3 k′ d† (k′ . s)d(k′ . s)|0 = − d3 k′ d† (k′ . s) and d† (k. which are degenerate in mass and are created by the operators c† (k. d† (k.48) 1 s′ =± 2 = −d† (k. s′ )δss′ δ3 (k − k′ )|0 (8. Namely.49) namely. as we have also seen in the case of the complex Klein-Gordon ﬁeld theory. s). the vacuum state of the theory is charge neutral while the two distinct one particle states. carry opposite charge. s′ ) − d† (k′ . s′ ) d† (k. (8. we can deﬁne the Green’s function for the Dirac ﬁeld theory as satisfying the Dirac equation with a delta function potential (source) as . the second quantized description of the Dirac theory is symmetric under the interchange of particles and anti-particles. s′ )c(k′ . s) respectively.5) and that the (normal ordered) Hamiltonian in (8. This shows that.

Let us recall the identity (i∂ − m) (i∂ + m) = (i∂ )2 − m2 / / / = − 1 µ ν γ . From these then.117) we know that the Green’s function for the Klein-Gordon equation satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). / (8. advanced. we can identify (8.52) S(x − y) = (i∂ + m) G(x − y).51) = − ∂µ ∂ µ + m2 . S(x − y) is a 4 × 4 matrix function in the spinor space.).54) Therefore.5 Green’s functions 301 (i∂ − m) S(x − y) = δ4 (x − y). Feynman. we would concentrate mainly on the Feynman Green’s function since that is the most useful . from (5. we do not have to calculate the Green’s function for the Dirac ﬁeld theory separately. etc. / so that (8.50) Here. (8. but it is the trivial 4 × 4 identity matrix (which we do not write explicitly). Although the relation (8.γ 2 = −γ µ ∂µ γ ν ∂ν − m2 = −γ µ γ ν ∂µ ∂ν − m2 ∂ ∂ + µ ν − m2 = −η µν ∂µ ∂ν − m2 (8.53) holds for any Green’s function of the theory (retarded.53) (i∂ − m) S(x − y) = (i∂ − m) (i∂ + m) G(x − y) / / / = − ∂µ ∂ µ + m2 G(x − y) = δ4 (x − y).8. The right-hand side is also a matrix. Furthermore.

We note that the relation (8.145) √ where k0 = Ek = k2 + m2 > 0) S (+) (x) = (i∂ + m) G(+) (x) / = (i∂ + m) / = i 2 i 2 d3 k 1 −ik·x e (2π)3 k0 d3 k k + m −ik·x / e .53) we can also obtain the positive and the negative energy Green’s functions for the Dirac ﬁeld theory as (see (5. k2 − m2 + iǫ (8.53) can be written in the momentum space as S(k) = (k + m) G(k).55) SF (k) = (k + m) GF (k) = lim / which can also be written as 1 .56) SF (k) = lim ǫ→0+ (8. k − m + iǫ / ǫ→0+ (8.58) We can express the Feynman Green’s function in terms of the positive and the negative energy Green’s functions as well (see (5. for example.146)) .57) From (8. / so that we can write.143) and (5. (see (5. 3 (2π) k0 (8.302 8 Dirac field theory in the calculation of S-matrix elements.140)) k+m / . 3 (2π) k0 S (−) (x) = (i∂ + m) G(−) (x) / = (i∂ + m) − / = − = i 2 i 2 i 2 d3 k 1 ik·x e (2π)3 k0 / d3 k (−k + m) ik·x e 3 (2π) k0 / d3 k (k − m) ik·x e .

(8.6 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).60) where (+) ψα (x) = 1 s=± 2 d3 k m e−ik·x c(k. we can decompose the adjoint ﬁeld operator (8. (8. s). Similarly. δ(x0 )(G(+) (x) + G(−) (x)) = 0). s). s)uα (k. s).59) where we have used the fact that the terms involving derivatives of the step function cancel out (namely.8) also as ψ α (x) = ψ α + ψ α (x). s)v α (k. (8.8.62) (+) d3 k m e−ik·x d(k. s)vα (k.61) (2π)3 k0 (−) ψα (x) = 1 s=± 2 d3 k Similarly. (8. other Green’s functions for the Dirac ﬁeld theory can also be easily obtained from the corresponding functions for the KleinGordon ﬁeld theory.63) (2π)3 k0 ψ α (x) = 1 s=± 2 (−) d3 k . 8. (2π)3 k0 m eik·x c† (k.6 Covariant anti-commutation relations Let us decompose the ﬁeld operator into its positive and negative energy (frequency) parts as (+) (−) ψα (x) = ψα (x) + ψα (x). s)uα (k. where ψ α (x) = 1 s=± 2 (+) (−) (8. (2π)3 k0 m eik·x d† (k. s).

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

where we have used (3.136) (see also (5.65) ψα (x).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. Thus. ψ β (y) (−) (+) + = −iSαβ (x − y) − iSαβ (x − y) = −i S (+) (x − y) + S (−) (x − y) = −iSαβ (x − y). in rearranging terms to bring them to a particular form.5 and 6. by deﬁnition the normal ordered product of two ﬁeld operators yields (note from (8.7 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. ψ β (y) + + (−) + ψα (x). s)v β (k.6.8. However.9.109). ψ β (y) (+) (−) (8. 8. since fermionic ﬁeld operators anti-commute (Grassmann variables).63) that the positive and the negative energy parts of the ﬁeld operators contain annihilation and creation operators respectively) . αβ (8.61) and (8.7 Normal ordered and time ordered products 305 × eik·x−ik ·y vα (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 (2π)3 k0 (−) = −iSαβ (x − y). s′ δss′ δ3 (k − k′ ) = 1 s=± 2 ′ d3 k m ik·(x−y) e vα (k. we can write (+) = ψα (x). we need to be careful about negative signs that can arise from changing the order of any two such operators.66) which is the analogue of the Schwinger function (5. s)v β k′ . As a result.

Similarly. ψ β (y) (−) + (+) (−) (+) (−) (+) (−) (+) (−) (+) (−) = ψα (x)ψ β (y) + iSαβ (x − y).69) (8.67). (+) (−) (+) (8. using (8. (+) (+) (+) (+) (+) (8.67) and we note that. (+) (8. As a result. here the order of factors inside normal ordering can lead to additional signs.68) where in the last step we have made the identiﬁcation in (8.71) . it is easy to show that ψ β (y)ψα (x) = : ψ β (y)ψα (x) : −iSαβ (x − y).68) as ψα (x)ψ β (y) = : ψα (x)ψ β (y) : −iSαβ (x − y). we see that we can denote 0|ψα (x)ψ β (y)|0 = −iSαβ (x − y) = ψα (x)ψ β (y). 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).70) (8.64). unlike the bosonic case. In other words.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) :.65). we can write (8. From the deﬁnition in (8.

(8. . 0|T ψα (x)ψ β (y) |0 = iSF .αβ (x − y).81).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 .αβ (x − y) : ψα (x)ψ β (y) : + iSF . The time ordered product of Dirac ﬁeld operators is again deﬁned with time ordering of the operators from left to right (largest time on the left) keeping in mind the appropriate negative signs associated with rearranging fermionic operators as T (ψα (x)ψ β (y)) = θ(x0 − y 0 )ψα (x)ψ β (y) − θ(y 0 − x0 )ψ β (y)ψα (x).5 which we will not repeat here.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).αβ (x − y) = ψα (x)ψ β (y). 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 6.8.6.7 Normal ordered and time ordered products 307 which can be compared with (6.104). (8.73) (+) (−) (−) (+) where we have used the deﬁnition in (8.59) and which can be compared with (6. (8. 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 6.74) Once again.

2 ψR (x) = ψL (x) = (8. we have identiﬁed k0 = Ek = |k|. Therefore.23)) d3 k (2π)3 |k| ψR (x) = e−ik·x c(k)uR (k) + eik·x d† (k)vR (k) .308 8.3). Unlike in (8.8 Massless Dirac ﬁelds 8 Dirac field theory As we have seen in section 3. when the Dirac particle is massless.75) As a result.54) which is appropriate for massless spinors.7.77) to obtain (here k0 = |k|) . we note here that there is no sum over spin polarization because the chiral spinors are charaterized by their unique chirality or helicity (see. as before. the discussion following (3. we can expand the ﬁeld as (see (8. 2 1 (½ − γ5 ). the free Dirac Lagrangian density for a massless spinor ﬁeld naturally decomposes as / / / L = iψ∂ ψ = iψ R ∂ ψR + iψ L ∂ ψL . for example. (8. we can also decompose a Dirac ﬁeld into a left-handed and a right-handed component as 1 (½ + γ5 )ψ(x).53) and (2.23) because we are using here the normalization (2.7. in this section we will discuss brieﬂy the quantization of such ﬁelds.23).76) Such massless left-handed and right-handed Dirac ﬁelds (Weyl ﬁelds) arise naturally in physical theories (as we will see. in the case of the standard model in 14. (8. Furthermore. the multiplicative factors in (8. The discussion for the left-handed ﬁeld can be carried out in an analogous manner.77) where. for example. Using the deﬁnition of the chiral spinors in section 3. 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)).77) is diﬀerent from those in (8. it is most natural to decompose it into its chirality states. Correspondingly.170) we can invert the decomposition in (8.

(8. 3 |k| (2π) (8. it can be shown that the nontrivial anti-commutation relations between the creation and the annihilation operators take the form c(k). d† (k′ ) + + . (8. for example. .78) 0 Similarly.80) The Hilbert space of the theory can now be built with the vacuum deﬁned as satisfying c(k)|0 = 0 = d(k)|0 . c† (k′ ) = δ3 (k − k′ ) = d(k). (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. (8.76)).81) and the higher particle states created by applying the creation operators and other quantities of interest can be derived in the standard manner.79) As a result.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. We recall that the Feynman propagator is deﬁned as the time ordered product (see.172) this can also be written as d3 k 1 iSF .74)) iSF . (8.84) .77) and using the properties of the vacuum in (8.310 8 Dirac field theory As an example.83) Here we have used (8.82) Putting in the ﬁeld decomposition in (8.Rαβ (x − y) = d3 kd3 k (2π)3 |k||k | × θ(x0 − y 0 )e−ik·x+ik ·y uRα (k)u† β (k ) 0|c(k)c† (k )|0 R † −θ(y 0 − x0 )eik·x−ik ·y vRα (k)vRβ (k ) 0|d(k )d† (k)|0 = d3 k e−ik·(x−y) θ(x0 − y 0 )uRα (k)u† β (k) R (2π)3 |k| † −eik·(x−y) θ(y 0 − x0 )vRα (k)vRβ (k) .80) in evaluating the vacuum expectation values. let us determine the propagator for a righthanded fermion. Dropping the spinor indices and using (3.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 ). (8.Rαβ (x − y) = † 0|T ψRα (x)ψRβ (y) |0 † = θ(x0 − y 0 ) 0|ψRα (x)ψRβ (y)|0 † −θ(y 0 − x0 ) 0|ψRβ (y)ψRα (x)|0 . (8. for example.81) we obtain (we assume k0 = |k| in the ﬁeld decomposition) iSF .

Using the integral representation for the theta function (see (6.85) Here we have used the fact that the integral has contributions only when k0 = |k| and correspondingly have rewritten some of the terms.100) and the limit → 0 is to be understood) we can write the ﬁrst term in the integrand as 1 0 0 θ(x0 − y 0 )(|k|γ 0 − γ · k)e−i|k|(x −y ) 4|k| = − = − = − = − dk0 (|k|γ 0 − γ · k) e−i(k +|k|)(x 2πi 4|k| k0 + i 0 0 0 0 0 −y 0 ) dk0 (|k|γ 0 − γ · k) e−ik (x −y ) 2πi 4|k| k0 − |k| + i dk0 (k0 γ 0 − γ · k) e−ik (x −y ) 2πi 4k0 k0 − |k| + i dk0 k e−ik (x −y ) / .8 Massless Dirac fields 311 Let us evaluate the two terms in (8.84) separately. 2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 (8.8. 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 .

312 = − / dk0 k e−ik (x −y ) . For .89) We note that although our discussion has been within the context of right-handed fermion ﬁelds.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 (8.82) as † the time ordered product of ψR ψR and not as the conventional time ordered product of ψR ψ R (see. 2πi 4k0 k0 + |k| − iǫ 0 0 0 8 Dirac field theory (8. adding the two results in (8. However. (8.85) and (8. 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 .9 Yukawa interaction Let us next consider the theory which describes the interaction between a Dirac ﬁeld and a charge neutral Klein-Gordon ﬁeld.R (x − y) = / d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) e . 2(k2 + iǫ) k + iǫ iSF .R (k) = (8.87) iSF .88) We note here that we have deﬁned the propagator in (8. 2(k2 + iǫ) k + iǫ (8.86) and substituting into (8.74)). everything carries over in a straightforward manner to left-handed fermion ﬁelds. 8. the conventional covariant propagator (deﬁned as the time ordered product of ψR ψ R ) can be obtained from (8.R (k)γ 0 = ik(½ − γ5 ) / ikPL / = 2 .86) Therefore. The reason for this is that the adjoint ﬁeld ψ R has negative chirality.R (k) = iSF .88) by simply multiplying a factor of γ 0 on the right which leads to iS F . for example.

g. the π 0 meson) has mass M . The trilinear coupling gψψφ is manifestly Lorentz invariant (as we have seen in (3.86). 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.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. (8.36). in this case.) In this case.90) into a free part and an interaction part. Namely. L = L0 + LI .. where 1 M2 2 / φ . As in (6. we can denote the interaction Lagrangian density as (remember everything is normal ordered) . we are going to ignore this aspect of the meson and also switch oﬀ the self-interactions of the meson ﬁeld for simplicity. L0 = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 λ LI = −g ψψφ − φ4 . The Lorentz invariant Lagrangian density.92). since it leads to the Yukawa potential in the Born approximation and the coupling constant g represents the strength of the interaction. we can separate the Lagrangian density (8. (The self-interaction can be switched oﬀ at the lowest order by setting λ = 0. has the form 1 λ M2 2 / φ − g ψψφ − φ4 .9 Yukawa interaction 313 example. ψψ is invariant under a Lorentz transformation and φ is a scalar ﬁeld) and is known as the Yukawa interaction.91) (8.90) L = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 4! where we are assuming that the fermionic particle (say the proton) has mass m while the spin zero meson (e. However. the correct parity invariant trilinear interaction for a pseudoscalar meson should be (−gψγ5 ψφ). since here we are only interested in developing calculational methods in a quantum ﬁeld theory.8. 4! (8.

94) +··· . the adiabatic switch-oﬀ is assumed) S = T e−i R∞ R −∞ LI = −g : ψψφ := −HI . in this theory the S-matrix will have the expansion (see (6.314 8 Dirac field theory As a result.6). (8. we can write (8.69).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′ ) + ······ . (8.5 and 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′ ) : (8.95) . Using Wick’s theorem (see sections 6.

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

s). For example. s′ )uα (p. s) ˜ ˜ ˜ p × δss δ3 (p − k) δs′ s′ δ3 (p′ − k′ ) δ3 (˜ − q) ˜ ˜ = −(2π)4 ig δ4 (k + k′ + q) × 1 (2π)3 2q 0 m (2π)3 k0 m v α (k′ .99) represents the process where a proton.96). (For completeness.99) where we have used the appropriate (anti) commutation relations between the creation and the annihilation operators in evaluating the vacuum expectation value. we note here that this lowest order matrix element in (8. k′ .˜′ =± 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′ .99) vanishes because energy-momentum conservation cannot be satisﬁed. 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. s′ )c(p. s′ .99) can be satisﬁed only if the energy of each of the particles vanishes which is not possible since the particles are massive. s)a(˜ )c† (k. The delta function merely ensures the overall conservation of energy and momentum in the process. k′ 0 . The matrix element (8. s)d† (k′ . note that since k0 . However. s′ )uα (k. energy conservation δ(k0 + k′ 0 + q 0 ) in (8.˜′ =± 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′ . (2π)3 k′ 0 (8. s′ )uα (p.95) also leads to a distinct physical process. let us consider the second .316 8 Dirac field theory 0|S1 |k.) Each term at order g2 in the expansion in (8. q 0 > 0. our goal in this section has been to describe the calculational methods in quantum ﬁeld theory. s. In fact. s) ˜ ˜ × 0|d(p′ . an anti-proton along with a π 0 meson are annihilated.

s2 = −ig2 × d4 xd4 x GF (x − x ) ei(k3 +k4 )·x e−i(k1 +k2 )·x m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 (8. k2 . k2 . k3 . s4 . k3 . (2π)4 GF x − x = (8.102) . k2 . s1 . s4 . Recalling that the Fourier transformation of the Green’s function is given by (see (5. (8. s3 )vα (k4 . k3 .101) × uα (k3 . the matrix element will have the form k4 .8.95) (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 ) : . In such a case. s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x ) (−) + ψ α (x )ψα (x )ψ β (x)ψβ (x)|k1 . s2 = − ig2 2 d4 xd4 x GF (x − x ) (−) (+) (+) (2) (−) × k4 . s3 |S2 |k1 . s2 (−) (+) (+) = −ig2 × d4 xd4 x GF (x − x ) (−) (+) (+) (−) k4 . s1 ). s1 .120)) d4 q −iq·(x−x ) e GF (q). s1 . s2 )uβ (k1 .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. s4 . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x )|k1 . s4 )v β (k2 .9 Yukawa interaction 317 term in the expansion at order g2 in (8.

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

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

s). s).106) k. is represented as a vertex (point of interaction) with no meaning attached to the external lines. for the Yukawa interaction (8. The interaction can be read out from the Lagrangian density LI and. (particle annihilation). (anti-particle creation). in such a case is given by (the simplest way to visualize this is to note that the end of a fermion line along the direction of the arrow corresponds to a ψ ﬁeld which can either annihilate a particle or create an anti-particle) = m (2π)3 k 0 m (2π)3 k 0 uα (k. (8. carries a direction for charge ﬂow). the normalization factor for the external line √ (state) is denoted by (k0 = k2 + M 2 ) k = 1 (2π)3 2k0 . (anti-particle annihilation). (8.93). vα (k. s). a fermion external line with an arrow ﬂowing into an interaction point (vertex) can represent either the annihilation of a particle or the creation of an anti-particle at that point and the external line factor.320 8 Dirac field theory and in such a case. Thus. in a local quantum ﬁeld theory. since the fermion ﬁeld carries a charge (and. therefore. s α = However. v α (k. s α = Note that it is the interaction between the diﬀerent ﬁelds which gives rise to nontrivial scattering. (8. (particle creation).107) k.105) On the other hand. 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. s). we .

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

an anti-proton and a π 0 meson are annihilated (which we considered in (8. s). (2π)3 k′ 0 (8.2: A second order Feynman diagram. therefore. k′ .2. 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′ . s) (2π)3 k0 m v β (k′ .1. with the external lines representing physical particles. s′ .99)) will correspond in the lowest order to evaluating the Feynman diagram in Fig.322 8 Dirac field theory In this language. s. the physical process where a proton. s4 α β q δ γ k2 . 8. The . 8. s3 Figure 8. s2 k3 . k1 .99). s′ )uα (k. s1 k4 .109) This is exactly the value of the ﬁrst order S-matrix element which we had evaluated in (8. The value of this diagram can be obtained from the Feynman rules to be 0|S1 |k. the corresponding lowest order Feynman diagram will be given by 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. where the external lines represent the respective physical particles that are annihilated. q = −(2π)4 igδ4 (k + k′ + q)δαβ × 1 (2π)3 2q 0 m uα (k.

s2 )uβ (k1 . Once again. For example. it is worth emphasizing that whenever confusions are likely to arise in the Feynman method. let us observe that the same Feynman diagram can describe distinct physical processes depending on the external lines. s3 )vδ (k4 . s4 . namely. 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). 8. To conclude this section. s3 |S2 |k1 . s4 )iGF (k1 + k2 )v β (k2 . 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 . the Feynman diagram in Fig. it is always resolved by going back to Wick’s expansion (of importance are concepts like the symmetry factor associated with a graph). this is exactly the second order S-matrix element which we had calculated from the Wick expansion in (8. k2 .103). This shows the power of the Feynman diagram method. from the Feynman rules we see that it will be given by (we are ignoring the overall energymomentum conserving delta function (2π)4 δ4 (k1 + k2 − k3 − k4 )) . s3 )vα (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. k3 . 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 ).8.110) × uα (k3 .2 can describe the scattering of two protons through the exchange of a π 0 meson. s1 . s2 )iGF (q)(−(2π)4 ig)δ4 (q − k3 − k4 )δγδ × m 0 (2π)3 k3 m 0 uγ (k3 . However.10 Feynman diagrams 323 value of the diagram can now be calculated using the Feynman rules to be (2) k4 .

Furthermore. it follows from energy conservation that 0 0 k1 = k2 . if we are in the center of mass frame where k1 = −k4 . (8. 0 0 k2 = k3 .113). (8. 0 0 k1 = k4 . q 2 − M 2 + iǫ q = k1 − k2 . k1 − k2 ).324 8 Dirac field theory −iV (q) = −ig2 GF (q) = ǫ→0+ lim − ig2 . 4π |x| = (8.112) then.114) . (8. k2 = −k3 . q = (k1 − k2 ) = (0. we obtain d3 q iq·x e V (q) (2π)3 eiq·x d3 q (2π)3 −(q2 + M 2 ) + iǫ d|q||q|2 0 ∞ 0 ∞ −∞ π 2π −iV (x) = −ig2 = = = = = ǫ→0+ lim −ig2 ∞ 0 ig2 (2π)3 ig2 (2π)2 g2 dθ sin θ 0 dφ d|q| i|q||x|(|q|2 + M 2 ) zeiz|x| z2 + M 2 iM e−M |x| 2iM |q|2 ei|q||x| cos θ |q|2 + M 2 ei|q||x| − e−i|q||x| (2π)2 |x| dz g2 (2πi) (2π)2 |x| ig2 e−M |x| . we see that V (q) = V (q) and taking the Fourier transform of (8.113) Therefore.111) where V (q) represents the scattering amplitude in the Born approximation.

Drell. 1964.11 References 1. Bjorken and S.114) the potential between the fermions to be g2 e−M |x| . 4. New York (1969). we obtain from (8. P. Schweber. Relativistic Quantum Mechanics and Field Theory. Itzykson and J-B. Introduction to Relativistic Quantum Field Theory. N. 4π |x| V (x) = − (8. Introduction to Quantum Field Theory. 6. D. New York. John Wliley.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. S. 1980. . Gross. therefore. C. 2. Zuber. 8. Introduction to the theory of Quantized Fields. New York (1993). 3. Relativistic Quantum Fields. John Wiley. 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. Row. Evanston (1961). Peterson. Moscow (1984). Roman. Shirkov. McGrawHill. F. Quantum Field Theory. V. Bogoliubov and D.8.11 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). 5.93) leads to the Yukawa potential in the Born approximation. N. New York. Nauka. McGrawHill. J.

This page intentionally left blank .

Let us start with the classical Maxwell’s equations in vacuum which are given by (remember c = 1) ∇ · E = 0.1) where E and B represent the electric and the magnetic ﬁelds respectively. the dynamical ﬁeld would naturally describe particles with spin 1. ∂t ∂E ∇×B= . 1 ) representation of the Lorentz group (see sec2 2 tion 4.2). ∂B . It is more convenient from our point of view to rewrite Maxwell’s equations (9. this is described by the vector potential Aµ (x) which belongs to the ( 1 . we note that we can solve the second and the third equations in (9.Chapter 9 Maxwell ﬁeld theory 9. As we know.1) in a manifestly Lorentz covariant form. (9. ∂t ∇×E=− ∇ · B = 0. 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.1) to write 327 . To that end.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.

A). we note that F0i = ∂0 Ai − ∂i A0 = Ei . 3.7) (9. E2 .2) where A(x) and Φ(x) are known as the vector and the scalar potentials respectively. Fij = ∂i Aj − ∂j Ai = −ǫijk Bk .8) denotes the magnetic ﬁeld vector. the second . 1. where E = (E1 .328 9 Maxwell field theory B = ∇ × A. ∂t (9. of course. ν = 0. (9.4) where ǫijk denotes the three dimensional Levi-Civita tensor and B = (B1 . In other words.3) If we now deﬁne the four dimensional curl of the four vector potential as Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . Similarly. (9.6) (9. the six independent components of the tensor Fµν are given by (three components each of) the electric and the magnetic ﬁelds. We can. −A). (9. E3 ). Aµ = (Φ. 2. Therefore. E = − ∂A − ∇Φ. B2 . denotes the electric ﬁeld vector.5) µ. (9. combine them into the four vector potential Aµ = (Φ. B3 ).

(9.1).1).1) are given by ∂µ F µν = 0.1 Maxwell’s equations 329 rank anti-symmetric tensor Fµν is also known as the ﬁeld strength tensor. ∂t ∂Ej = −ǫjik ∂ i Bk . for ν = 0.12) .11) which is the last equation in (9.9. or. or. ∂ 0 F0j + ∂ i Fij = 0. It is now easy to check that the other two equations (the ﬁrst and the fourth) in (9. with the ﬁeld strength tensor deﬁned in (9. for ν = j.9) takes the form (F 00 = 0 by deﬁnition (9.1) can be written in the manifestly covariant form as ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0.4) as Fµν = ∂µ Aν − ∂ν Aµ .10) which coincides with the ﬁrst of the equations in (9. equation (9. (9. Thus. or. (9. ∂0 F 0j + ∂i F ij = 0. or. equation (9. Similarly.13) (9. ∇ · E = 0. ∂t (9.4)) ∂i F i0 = 0.9) leads to ∂µ F µj = 0. or. or. ∂Ej + ∂ i (−ǫijk Bk ) = 0. we see that the set of four Maxwell’s equations (9.9) For example. ∂t ∂E = ∇ × B.

as an operator. For example. we note that under a gauge transformation Aµ (x) → A′ (x) = Aµ (x) + ∂µ θ(x).15) In other words.14) of the four vector potential Aµ (x). the ﬁeld strength tensor is unchanged under the gauge transformation (9.2 Canonical quantization In trying to second quantize the Maxwell ﬁeld theory. local parameter of transformation. µ (9.330 9 Maxwell field theory Furthermore. (9.16) = − ∂µ Aν (∂ µ Aν − ∂ ν Aµ ). which denotes the dynamical variable of the theory. δAµ (x) = A′ (x) − Aµ (x) = ∂µ θ(x). Fµν ′ → Fµν = ∂µ A′ − ∂ν A′ µ ν = = ∂µ (Aν + ∂ν θ) − ∂ν (Aµ + ∂µ θ) ∂µ Aν − ∂ν Aµ = Fµν . µ or. 9. the gauge invariance of Maxwell’s equations is obvious in this formulation. Maxwell’s equations (9.12) are also invariant under these transformations. 2 which leads to . real. let us examine whether there exists a Lagrangian density which will lead to Maxwell’s equations as its Euler-Lagrange equations.14) where α(x) is an arbitrary. Let us consider the Lagrangian density 1 1 L = − Fµν F µν = − (∂µ Aν − ∂ν Aµ )(∂ µ Aν − ∂ ν Aµ ) 4 4 1 (9. we have to treat the four vector potential Aµ (x). Consequently. But ﬁrst.

It is manifestly Lorentz invariant.18) which coincides with the manifestly covariant Maxwell’s equations (9. namely.16) since the Lagrangian density (9. take the form ∂L ∂L − ∂µ = 0.9). Note that a Lagrangian density 1 1 Fµν F µν − Fµν (∂ µ Aν − ∂ ν Aµ ) . ∂Aν ∂∂µ Aν or. = − 1 1 1 (∂ µ Aν − ∂ ν Aµ ) − ∂ µ Aν + ∂ ν Aµ 2 2 2 (9. The Lagrangian density (9. 4 2 L= (9. indeed.20).9.19) is equivalent to (9. ∂L = 0. ∂µ (∂ µ Aν − ∂ ν Aµ ) = ∂µ F µν = 0.16). in this case.20) ∂µ F µν = 0.16) when the ﬁeld variable Fµν is eliminated using the (ﬁrst) auxiliary ﬁeld equation in (9. However. ∂L = 0.17) = − (∂ µ Aν − ∂ ν Aµ ) = −F µν . ∂Fµν or. we will work with the simpler Lagrangian density in (9. . Fµν = ∂µ Aν − ∂ν Aµ .2 Canonical quantization 331 ∂L ∂Aν ∂L ∂∂µ Aν = 0. ∂µ or. so that the Euler-Lagrange equations. ∂∂µ Aν (9.19) with both Aµ and Fµν treated as independent dynamical variables would also lead to Maxwell’s equations as its Euler-Lagrange equations. (9. gives Maxwell’s equations as its Euler-Lagrange equations.

the momentum conjugate to the ﬁeld variable A0 (x) vanishes. namely. for canonical quantization of such a system.23) This is equivalent to saying that. As a result. we can deﬁne the associated conjugate momentum as (see (9.25) . (9. in this gauge the nontrivial components of the conjugate momentum are given by ˙ Πi (x) = −F 0i = Ei = −(∂ 0 Ai − ∂ i A0 ) = −Ai (x). Furthermore. (9. Πj (y) x0 =y 0 = 0 = Πi (x). we can think of A0 (x) as a classical function (c-number function) – not as an operator – and without loss of generality we can set it equal to zero A0 (x) = 0. we have to choose a gauge and this particular choice is known as the temporal (axial) gauge. This is a consequence of the gauge invariance of the theory. since the ﬁeld strength tensor Fµν is gauge invariant (see (9. this Lagrangian density is also invariant under the gauge transformation (9. Πj (y) j = iδi δ3 (x − y). Therefore. 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. ˙ ∂ Aµ (x) (9.24) The naive canonical quantization conditions.17)) Πµ (x) = ∂L = −F 0µ .14) of the vector potential. (9. x0 =y 0 . we note that A0 (x) which would have a nontrivial commutation relation only with Π0 (x).332 9 Maxwell field theory Furthermore. (9.21). in this gauge.22) In other words. If we treat Aµ (x) as the independent ﬁeld variable.21) and the ﬁrst peculiarity of the Maxwell ﬁeld theory is now obvious from (9. would then appear to be [Ai (x). Π0 (x) = −F 00 = 0.15)). Aj (y)]x0 =y0 Ai (x). now must commute with all the ﬁeld variables in the theory.

Aj (y) j = −iδi TR (x − y).27).30) . this is not quite consistent which can be seen as follows.2 Canonical quantization 333 However. Aj (y) j = iδi δ3 (x − y).9. namely.28) and the condition (9.25) leads to Ai (x).23). the dynamical ﬁeld variable A(x) must be transverse.27) can be thought of as the gauge choice for this residual invariance.27) is given by ˙ Ai (x).26) Namely. For example.) On the other hand.18)) ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0. the theory can be checked to have a residual gauge invariance under a time independent gauge transformation which preserves (9. ¯ A(x) → A′ (x) = A(x) + ∇θ(x). x0 =y 0 (9. in the gauge (9.29) which does not satisfy the transversality condition (9. (9. x0 =y 0 x0 =y 0 (9. ∂ 0 (∂µ Aµ ) = 0. ∂ 0 (∇ · A) = 0. or. lead to (for ν = 0) ∂µ (∂ µ A0 − ∂ 0 Aµ ) = 0.23). or. (As a parenthetical remark. (9. in this gauge. the canonical quantization condition (9. the dynamical equations (see (9. The proper quantization condition consistent with the transversality condition (9.27) (9.23). j = −iδi δ3 (x − y). or. ∇ · A = 0. ˙ Ai (x). we note that even after imposing the gauge condition (9. Πj (y) or.

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).27) constitute constraints on the dynamics of the theory and systems with constraints are known as constrained systems. the Hamiltonian density for the Maxwell ﬁeld theory can be obtained to be .30) is consistent with the transversality condition in (9.30). (2π)3 (9. known as the Dirac method.23) and (9.23). Here 1 2 stands for the inverse of the Laplacian operator and for∇ mally denotes the Green’s function for ∇2 .31) so that (remember that ∂ i ∂i = −∇2 ) j j ∂ i δi TR = ∂ i δi + ∂i 1 j 2 ∂ ∇ (9.23) subject to (9. We note here that relations such as (9.27).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 . there exists a systematic procedure for deriving the correct quantization conditions for such systems. ∇ and (9. the transverse delta function in (9. the correct quantization condition (Dirac bracket) that results in the gauge (9.32) = ∂ j + ∂ i ∂i 1 j j j 2 ∂ = ∂ − ∂ = 0. As we have pointed out in connection with the discussions in the Dirac ﬁeld theory. Let us note here that in momentum space.33) In the gauge (9. (9. ∇2 (9. which we will discuss in the next chapter.22). When quantized systematically through the Dirac method.27) is given by (9.

18) take the form ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = Aν = 0.34) and leads to the Hamiltonian d3 x H = = 1 2 1 2 d3 x Π2 + B2 d3 x E2 + B2 . Maxwell’s equations (9. (9.37) . (9.38) (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 .27)) A0 = 0.35) H= Indeed this is what we know to be the Hamiltonian (energy) for the Maxwell theory even from studies in classical electrodynamics. 9.36) which also lead to ∂ µ Aµ = 0. (9.9. ∇ · A = 0.3 Field decomposition In the temporal gauge (9. 2 (9.23) (see also (9.

40) (9. will have plane wave solutions of the form A(x) ∝ ǫ(k) e∓ik·x . in general.42) (9.1. since the vector potential is transverse.38) vanishes identically (because of the choice of the gauge condition (9. We note from (9.336 9 Maxwell field theory As we have seen. satisfy the Klein-Gordon equation for a massless particle (wave equation) and. or. λ) = 0. with k0 = Ek = √ k2 = |k |. for ν = i. On the other hand.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.44) ǫ∗ (k. we must have ∇ · A(x) = 0. 2. . (9. complex satisfying k · ǫ (k. 9. (9.43) Namely. as shown in Fig. There can be two such independent directions. (9. for ν = 0. the three components of the vector potential.39) Thus. in this case. for example. We can choose the two independent polarization vectors to be orthonormal and. the polarization vector characterizing the vector potential must be transverse to the direction of propagation of the plane wave. (9. the wave equation (9. k · ǫ (k) = 0. therefore. λ = 1. λ′ ) = δλλ′ . λ) · ǫ(k.23)) A0 ≡ 0. we have Ai = 0.27) that.

1) and ǫ(k. λ) (9. λ) . 2) transverse to the direction of propagation k. λ) eik·x a† (k. λ) and a† (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.45) +ǫ∗ (k. where k0 = Ek = |k |. quite analogous to that of the neutral Klein-Gordon ﬁeld except for the presence of the polarization vectors.46) Note that a(k. therefore. This discussion is. λ) e−ik·x a(k.44) (since they deﬁne a basis in the plane transverse to the direction of propagation). (9. 2) ǫ(k.3 Field decomposition 337 k ǫ(k. Given this. 1) Figure 9.1: Two independent and orthogonal polarization vectors ǫ(k.9. λ) are operators and as we have deﬁned. the ﬁeld operator A(x) is Hermitian as it should be. The vector nature of the ﬁeld variable is now contained completely in the polarization vectors. Let us next note that . 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.

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

j = −iδi TR (x − y). P = λ=1 d3 k k a† (k. Aj (y)]x0 =y0 .49) and (9. λ) and a† (k.50) denote the two inversion formulae for the Maxwell ﬁeld (analogous to (5. ˙ ˙ = 0 = [Ai (x). λ).54) The Hilbert space for this theory can be built up in the standard manner. a† (k′ . λ)a(k. [a(k.52) In other words. [Ai (x). Aj (y)]x0 =y0 ˙ [Ai (x). The normal ordered forms for these operators can be shown to correspond to (see (5. λ). (9. λ′ )] = 0 = [a† (k. (9.46)) that Ek = |k|. a(k′ . a† (k′ . λ) behave exactly like the annihilation and the creation operators for a harmonic oscillator system.77) and (5. λ)a(k. then it satisﬁes . Given the quantization conditions (see (9. The Hamiltonian and the momentum operators for the Maxwell theory can again be expressed in terms of the creation and the annihilation operators.30)). If we denote the vacuum state of the theory by |0 . the operators a(k.61) in the case of the Klein-Gordon ﬁeld). λ).49) and (9.53) where we recall from the Einstein relation (see (9. λ′ )] = δλλ′ δ3 (k − k′ ). (9.97) and we do not show the normal ordering symbol explicitly) 2 H = λ=1 2 d3 k Ek a† (k.51) using (9. λ). we can easily show that [a(k. Aj (y)]x0 =y0 (9.50).9. λ′ )].3 Field decomposition 339 Equations (9. λ).

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

2) λ=1 = 1 (ˆx + iˆy )(a† (k. R L R L where we have used the familiar identiﬁcation that 1 ˆ ǫL (k) = √ ex − iˆy . 1) − ia† (k. It is clear. 2) . 2) . 1) + ia† (k. 2) = ey . Let us note that if we choose the direction of propagation of the plane wave solution to be along the z-axis (k = kˆz ) and e ˆ ǫ(k.61) = ǫ∗ (k)a† (k) + ǫ∗ (k)a† (k). (These one photon states are eigenstates of helicity with eigenvalue ±1. 2)) e e 2 1 e e + (ˆx − iˆy )(a† (k.) A simple way to see that these would represent the creation operators for left and right circularly polarized photons is to note that 2 ˆ ˆ ǫ∗ (k. 1) + ia† (k. λ)a† (k. 1) as creating a photon polarized along the x-axis and a† (k. that the operators a† (k) = L a† (k) = R 1 √ a† (k. 2)) 2 (9. 2) as creating a photon polarized along the y-axis.9. e 2 (9.62) describe respectively the left and the right circular polarization vectors.60) would then create photons which are left and right circularly polarized respectively. 2 1 √ a† (k. λ) = ex a† (k. e 2 1 ˆ ǫR (k) = √ ex + iˆy . we can identify a† (k.59) then. 1) − ia† (k. 1) = ex . in this language. (9.3 Field decomposition 341 quantize the Maxwell ﬁeld theory in a manifestly covariant manner which will bring out some other interesting aspects of such a theory. 2 (9. . 1) + ey a† (k. ˆ ǫ(k.

λ′ )] 2 = λ. a† (k′ . λ)+eik·x a† (k. λ) = (0. λ) and a† (k. if we deﬁne the two polarization vectors for λ = 1. λ′ )e−ik·x+ik ·y [a(k. (9. λ)(e−ik·x a(k. λ)ǫν (k′ . (9. 2 as four vectors ǫµ (k. 3 2k 0 (2π) d3 k (2π)3 2k0 ǫµ (k. (9.λ′ =1 1 (2π)3 ′ d3 k d3 k ′ √ √ 2k0 2k′ 0 ×ǫµ (k.45) as (recall (9.λ′ =1 1 (2π)3 d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ ×ǫµ (k. λ′ )e−ik·x+ik ·y δλλ′ δ3 (k − k′ ) . λ)eik·x a† (k.65) A(−) (x) = µ λ=1 From the commutation relations for a(k.4 Photon propagator 9 Maxwell field theory Let us note that in the temporal gauge (9.27)). if we choose the polarization vectors to be real as in (9.52). λ). λ)). A(−) (y)] µ ν = λ.59) then we can write the ﬁeld decomposition (9. λ).23)) 2 Aµ (x) = λ=1 d3 k (2π)3 2k0 ǫµ (k. ǫ(k.23) (see also (9. λ) in (9. λ). λ)e−ik·x a(k.63) and furthermore. it follows that the only nontrivial covariant commutation relations have the form (these are not at equal time) 2 [A(+) (x). λ)ǫν (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.342 9.64) with k0 = |k|. λ)). for simplicity.

67) Aµ (x)Aν (y) = 0|Aµ (x)Aν (y)|0 = −iG(+) (x − y).4 Photon propagator 2 343 d3 k ǫµ (k. .68) From this. A(−) (y)] ν µ : Aµ (x)Aν (y) : −iG(+) (x − y) µν : Aµ (x)Aν (y) : + Aµ (x)Aν (y). (9. λ)ǫν (k. we can deﬁne the Feynman propagator in the standard manner (see (6.145)).107)). A(+) (y)] = −[A(+) (y). A(−) (x)] = iG(+) (y − x) µ ν ν µ νµ = − 1 2 d3 k eik·(x−y) (2π)3 k0 2 ǫµ (k. µν (9.9. 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 [A(−) (x).143) and (5. λ) λ=1 = −iG(+) (x − y). namely. λ) λ=1 = −iG(−) (x − y). µν (9.66) which are the analogs of the positive and the negative energy Green’s functions for the Maxwell ﬁeld (see (5. 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. λ)ǫν (k. λ)ǫν (k. we obtain : Aµ (x)Aν (y) : +[A(+) (x).

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

9. λ)ǫν (k. |k| (9.74) . we can identify the momentum space Feynman Green’s function for the Maxwell (photon) ﬁeld to be GF . λ) .µν (k) = ǫ→0+ lim 1 k2 + iǫ 2 2 ǫµ (k. λ)ǫν (k. λ)ǫν (k. λ=1 (9. k . 0). (9.70) and denoting the variable of integration k′ 0 → k0 .72) Therefore. 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 (x − y) = lim ǫ→0+ d4 k e−ik·(x−y) (2π)4 k2 + iǫ 2 ǫµ (k.73) where GF (k) denotes the Feynman Green’s function for a massless scalar ﬁeld (see (5. The Feynman Green’s function in (9.4 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) .71) where we have used the Einstein relation (9. ˆ kµ = kµ − (k · η)η µ (k · η)2 − k2 = 0.140)). 0. Substituting this into (9. let us note that if we introduce two new four vectors η µ = (1. we obtain GF .46).73) depends on the polarization sum and to evaluate the polarization sum. 0. λ) λ=1 = GF (k) λ=1 ǫµ (k.

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

4 LI = −eψγ µ ψAµ . 9.82)) can be checked to be invariant under the ﬁnite local gauge transformations ψ(x) → ψ ′ (x) = e−iθ(x) ψ(x). where Dµ ψ(x) = (∂µ + ieAµ (x))ψ(x).82) The total Lagrangian density (9. ψ(x) → ψ (x) = ψ(x)eiθ(x) .80) denotes the covariant derivative (see (7.5 Quantum electrodynamics 347 GF . 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. Aµ (x) → A′ (x) = Aµ (x) + µ ′ 1 ∂µ θ(x).83) . (9.5 Quantum electrodynamics (9.9.80) (or (9.80) 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. (9.81) (9.83)) and we have separated the Lagrangian density in (9. 4 = L0 + LI .74). e (9. which follows from (9.0µ (k) = 0.

85) It follows from this that under a (ﬁnite) gauge transformation.348 9 Maxwell field theory where θ(x) is the real and ﬁnite local parameter of transformation or under the inﬁnitesimal form of the transformation (9. ψ .83) (or (9.81) transforms covariantly under a gauge transformation (see (7.83) (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). e (9. Aµ ). let us note that the covariant derivative in (9.84)).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). we need to examine the invariance of the minimally coupled Dirac Lagrangian density in (9.84) where ǫ(x) denotes the inﬁnitesimal local parameter of transformation. ψ. A′ ) µ ′ = iψ γ µ Dµ ψ ′ − mψ ψ ′ ′ ′ = iψγ µ Dµ ψ − mψψ = LDirac(ψ. First. In fact. δǫ ψ(x) = ψ (x) − ψ(x) = iψ(x)ǫ(x). Therefore. the minimally coupled Dirac Lagrangian density transforms as ′ LDirac (ψ. we have already seen that the Lagrangian density for the Maxwell theory is invariant under the gauge transformation in (9. (9.86) . = iψeiθ(x) γ µ e−iθ(x) Dµ ψ − mψeiθ(x) e−iθ(x) ψ (9. ψ. δǫ Aµ (x) = A′ (x) − Aµ (x) = µ ′ 1 ∂µ ǫ(x). Aµ ) → LDirac (ψ ′ .

27)).89) lim − 1 ˆ ˆ i(ηµν − ηµ ην + kµ kν ) .23) (see also (9.106)-(8. The interaction vertex for QED can also be read out from the Lagrangian density (9. the minimally coupled Dirac Lagrangian density is also invariant under the gauge transformation (9. λ).µν (k) = k. (9.80) r µβ α q p = −(2π)4 ie (γ µ )βα δ4 (p+q +r). (9. In addition. We have already derived the fermion propagator as well as the fermion external line factors in (8.64) so that the external line in (9. there is a conserved current.5 Quantum electrodynamics 349 Namely. in QED we have the photon propagator as well as the photon external line factor which take the forms. k2 + iǫ ǫµ (k. λ µ = ǫ→0+ (9. The form of the propagator in (9. therefore. the complete QED Lagrangian density (9.91) . (9. ∂µ J µ (x) = 0.88) The Feynman rules for QED can be derived in the standard manner.107). J µ (x) = ψγ µ ψ.77)) correponds to the temporal gauge (9.89) (or (9. (9.83) and.90) (2π)3 2k0 where we have chosen a real polarization vector as in (9.80) is gauge invariant.9. We note here that the photon propagator is a gauge dependent quantity.90) can represent both the annihilation as well as the creation of a photon. µ k ν = iGF .104) and in (8. In this theory.87) In momentum space the current conservation takes the form kµ J µ (k) = 0.

9. We note here that experimentally. λ Figure 9. we can now calculate the S-matrix elements for various physical processes in QED and let us see how some of these calculations are carried out at low orders.2. but not the vertices (points) where each of the two photons was produced. s2 k1 . As a result. 9. λ µ k1 − k3 k1 . λ′ ν µ k3 . . 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. s1 k2 .350 9. a) Photon pair production: For example. s1 ν + k1 − k4 k2 .6 Physical processes 9 Maxwell field theory With these rules.2. s2 k4 . Therefore. the amplitude for this process would simply be the sum of the amplitudes (quantum superposition principle) given by the two diagrams in Fig. (9.92) To lowest order there are two Feynman diagrams which will contribute to such a process and they are shown in Fig. λ′ k4 .2: The two lowest order Feynman diagrams that contribute to the pair production of photons as well as to the Compton scattering. 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 .

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

3.2 except for the labeling of the momenta. λ)u(k1 .96) / ˜ − m + iǫ K / ˜ where as deﬁned in (9.93). instead of an electron and a positron being annihilated as was the case in (9. namely. this process is described by the two Feynman diagrams shown in Fig. (9. s3 )ǫ (k2 . here we have the annihilation of an electron with momentum k1 and the creation of an electron with momentum k3 (the photon line factor in (9. λ′ )u (k1 . λ′ )|S (2) |e− (k1 ). λ) = − ie2 (2π)2 1 0 2k2 1 0 2k4 m 0 k1 m 0 k3 / × u(k3 . the internal line represents a photon propagator signifying that the electrons scatter by exchanging (emitting and absorbing) a photon. we can obtain the S-matrix element for Compton scattering directly from the results in (9. γ(k4 .93) (or (9. (9. λ) / 1 ǫ (k2 . Therefore.90) is the same for the creation or the annihilation of a photon). The amplitude diﬀers from the case of the pair production of photons by only the external line factors. In this case.94) K = k1 − k4 and we have introduced Q = k1 + k2 . s1 ) / Q − m + iǫ / 1 ǫ (k4 .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 ). γ(k4 )) to be outgoing.93) to be e− (k3 ).98) To the lowest order. s1 ) .96)) not only in the external line factors. Here. λ′ ) +u(k3 . 9. we will assume the ﬁnal state electron and photon (e− (k3 ). γ(k2 . s3 )ǫ (k4 . but . Thus.352 9 Maxwell field theory are given to the lowest order by the same diagrams as in Fig. (9. 9. this amplitude will differ from (9.

9. e− (k2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k4 . the S-matrix element takes the form e− (k3 ).3: The two lowest order Feynman diagrams for M¨ller scato tering. d) Bhabha scattering: Finally. also in the propagator of the diagram.94). s4 )γ µ u(k2 . the Bhabha scattering can be thought of as the process . s1 ) 2 + iǫ K ˆ ˆ ˜ ˜ (ηµν − ηµ ην + Kµ Kν ) u(k4 . s2 ) µ ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . This can be calculated in a straightforward manner as before. e− (k4 )|S (2) |e− (k1 ). s3 )γ u(k2 . s3 )γ ν u(k1 .6 Physical processes 353 k3 k4 k4 k3 + k1 − k3 k1 − k4 k1 k2 k1 k2 Figure 9. s2 ) ˜ 2 + iǫ K (9.99) ˜ where K and K are deﬁned in (9. s1 ) . s4 )γ ν u(k1 . Assuming that the ﬁnal state electrons are outgoing. carrying out the momentum integration for the internal line and omitting the overall energy-momentum conserving delta function δ4 (k1 + k2 − k3 − k4 ). +u(k3 .

s1 ) . the ﬁrst diagram in Fig. e+ (k4 )). 2 + iǫ K (9. s2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k3 . s3 ).4.101) . s2 )γ µ v(k4 . In this case. e+ (k4 . s3 )γ µ v(k4 .4: The two lowest order Feynman diagrams for Bhabha scattering. 9. 9.354 9 Maxwell field theory e− (k1 ) + e+ (k2 ) → e− (k3 ) + e+ (k4 ). s2 )γ ν u(k1 . s4 ) +v(k2 . but in the present case we will treat the ﬁnal state electron and the positron as outgoing (e− (k3 ). s3 )γ ν u(k1 . e+ (k2 . s1 ) 2 + iǫ Q ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . On the other hand. We see that topologically these diagrams are the same as in Fig. We can calculate this matrix element as before and omitting the overall delta function representing conservation of energy-momentum. s1 ). the second diagram simply describes the scattering of an electron and a positron by exchanging a photon. we have e− (k3 .3. s4 ) ˆ ˆ (ηµν − ηµ ην + Qµ Qν ) v(k2 . 9. The lowest order Feynman diagrams for this process are shown in Fig.100) k4 + k1 + k2 k1 k3 k1 k1 − k3 k2 Figure 9.4 represents the annihilation of an electron and a positron to create a photon which subsequently creates an electron and a positron. s4 )|S (2) |e− (k1 . k2 k4 k3 (9.

let us discuss very brieﬂy the Ward-Takahashi identity in QED. 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.9. which we will study later. the Lagrangian density (9.80) for QED has the form 1 ¯ ¯ L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ.97) K = k1 − k3 and Q = k1 + k2 . of course. 4 (9. let us note that the fermion propagator in QED can be represented as i = p−m / iSF (p) = p . However.103) that ∂ iSF (p) = ∂pµ = = ∂ ∂pµ i p−m / i 1 (−γµ ) p−m / p−m / i i (iγµ ) p−m / p−m / 1 = − (iSF (p)) Γµ (p. 9. 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. This gives a ﬂavor of low order calculations in QED.7 Ward-Takahashi identity in QED 355 where as deﬁned in (9. −p.104) . (9. As we have seen. independent of the choice of gauge. depends on the choice of gauge.103) where the i factor in the denominator is understood.94) and (9. Here we would only give a very simple derivation of this relation in QED. 0) (iSF (p)) . e (9.7 Ward-Takahashi identity in QED Before we discuss more complicated symmetries.102) The structure of the photon propagator in the theory. It follows from (9.

k = 0). −p. a diagrammatic representation for the basic identity (9.104) in QED resulting from the structure of the fermion propagator has the form ∂ ∂pµ p =− 1 e p k=0 p .108) .107) ∂ (iSF (p)) (iSF (p))−1 ∂pµ = −i which can also be written as i 1 ∂ −1 S (p) = − Γµ (p.91) in QED with a zero momentum photon (without the delta function and the factor of (2π)4 ). µ F ∂p e (9. −p. As a result the right-hand side of (9. −p. 0) denotes the three point vertex (9. k = 0)(iSF (p)). by inverting (9.105) 1 = − (iSF (p))Γµ (p.104) we can obtain a relation between the fermion two point function and the three point vertex function as 1 Γµ (p.356 9 Maxwell field theory where Γµ (p. −p.106) In this case. (9. k = 0) = −(iSF (p))−1 e = ∂ (iSF (p))−1 ∂pµ ∂ −1 S (p). e Consequently.104) can be diagrammatically represented as i i iγµ p−m / p−m / 1 e = − p k=0 p (9. ∂pµ F (9.

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

108) diagrammatically at every order in perturbation theory.110) We recognize that the graphs in the bracket on the right-hand side of (9. We note that including quantum corrections (we will study renormalization in a later chapter).110) correspond precisely to the two loop corrections to the vertex function. if the fermion self-energy and the vertex function change as . Thus. A simple and powerful consequence of the Ward-Takahashi identity can be derived as follows. using (9. (9.358 9 Maxwell field theory Using the basic graphical identity in (9.106) we can easily verify the validity of the identity (9.106) at this order we obtain ∂ ∂pµ + =− 1 e + + + + + =− 1 e + + + + + .

That the Ward-Takahashi identity (9. Thus comparing with (9. (9.85)) iaψD = iaψ(∂ + ieA /ψ / /)ψ. m = 0). (9.83).115) (9. then from (9.108) is a consequence of gauge invariance can be seen heuristically in the following way. 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. D are constants and the dots denote other structures that may be induced due to quantum corrections.114) .113) −eCψA − Dψψ + · · · /ψ where A. we can identify a = 1 + B = 1 + C. Γµ −→ Z1 Γµ . (9.113).111) where Z1 . Z2 are constants. Invariance under the local gauge transformations (9. C.108) we conclude that these constants must be related as Z1 = Z2 . m = 0) −→ Z2 SF (p.112) This physically implies that if there are divergent parts in the fermion self-energy graphs at higher order. (9.7 Ward-Takahashi identity in QED 359 −1 −1 SF (p. on the other hand. B. 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).9. The Lagrangian density for QED including quantum corrections can be written as A 1 /ψ / L = − Fµν F µν + iψD − mψψ − Fµν F µν + iBψ∂ ψ 4 4 1 = − (1 + A)Fµν F µν + i(1 + B)ψ∂ ψ − e(1 + C)ψA / /ψ 4 −(m + D)ψψ + · · · .

since the ﬁeld strength tensor Fµν does not change under a shift of the vector potential by a gradient.21) and (9. the Lagrangian density (9. As a result.116) We will discuss these topics in more detail when we study renormalization chapter 16. 4 Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . Let us recapitulate brieﬂy how this arises. 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. We recall that the Lagrangian density for Maxwell’s theory is given by 1 L = − Fµν F µν .15). (9. (9. then the equations of motion (see (9.360 which is another way of saying 9 Maxwell field theory Z2 = 1 + B = 1 + C = Z1 .117) where Fµν denotes the ﬁeld strength tensor. ˙ ∂ Aµ Πµ = which implies the constraint Π0 = −F 00 = 0.118) As we have already seen in (9.22)) ∂L = −F 0µ . 9.117) as well as the action for this theory are invariant under the gauge transformation δAµ = ∂µ α(x).8 Covariant quantization of the Maxwell theory As we have seen. (9.18)) . canonical quantization of Maxwell’s theory results in a lack of manifest Lorentz invariance. the present choice of gauge is known as the Coulomb gauge). The canonical momenta conjugate to the ﬁeld variables Aµ can be calculated from the Lagrangian density and take the forms (see (9.

we lose manifest Lorentz covariance in the intermediate steps (which is highly desirable in any calculation). if sources (charges and currents) are present. ∂i ∂ i A0 − ∂ 0 Ai = 0. ∇2 A0 = 0.119) On the other hand. . the dynamical equations ∂µ F µν = J ν . or.120) In either case. A0 = 0. ∇2 (9. remains manifestly Lorentz invariant in the canonical formalism. ∇2 A0 = −J0 . or. (9. or. in the Coulomb gauge ∇ · A = 0. however. Let us emphasize here that the ﬁnal result for the calculation of any physical amplitude. would lead to ∂µ F µ0 = J 0 . the two transverse physical degrees of freedom describe the true dynamics of the theory while the time-like degree of freedom is merely related to the charge density. A0 = − 1 J0 .8 Covariant quantization of the Maxwell theory 361 ∂µ F µν = 0. We recognize that this is a consequence of our choice of gauge. or. But.9. lead to (for ν = 0) ∂µ F µ0 = ∂i F i0 = 0. but we lose manifest Lorentz covariance in the process (because all components of Aµ are not being treated on a equal footing). The canonical quantization can now be carried out for the two physical degrees of freedom. or.

namely.122) (9. (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. This implies that even if we can quantize the theory (say. . therefore. the inverse of P µν does not exist (this also means that the determinant of P µν vanishes) and the Green’s function and.117) for the Maxwell theory also as 1 1 L = − Fµν F µν = Aµ P µν Aν + total derivatives.) We note that we can write the Lagrangian density (9. it projects on to the space of the components of any vector transverse (perpendicular) to the gradient operator ∂ µ .362 9 Maxwell field theory The need for a choosing a gauge can be seen in an alternative manner as follows. in the naive path integral formalism). the Feynman propagator for the theory cannot be deﬁned.) Furthermore. we note that ∂µ P µν = ∂µ (η µν = (∂ ν − − ∂µ∂ν ) ∂ ν ) = 0 = ∂ν P µν . we see that P µν can be thought of as µν a projection operator. Consequently.123) With a suitable normalization. (9. 4 2 where P µν = η µν − ∂µ∂ν . P = 1 P µν deﬁnes the normalized projection operator.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 . (9. (In fact.124) so that this is the transverse projection operator. − ∂µ∂λ (9. we cannot carry out calculations in perturbation theory.

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

∂t χ = 0.364 9 Maxwell field theory ∂µ or. ∂µ F µν + ∂ ν (∂ · A) = −J ν . therefore. or.132) . (∂ · A) = 0. Thus we see that although the presence of the term − 1 χ2 in the 2 Lagrangian density (9. ∂∂µ Aν ∂Aν −∂µ F µν − ∂ ν (∂ · A) − J ν = 0. we have ∂ν ∂µ F µν + or.130) where we have used the anti-symmetry of the ﬁeld strength tensor as well as the conservation of J µ . (9. (∂ · A) = ∂ν J ν . (9.127).125) where χ = ∂ · A.127) Without the second term on the left-hand side in (9. If we now write out the left-hand side explicitly. or. Furthermore. or. Aν = −J ν .127). ∂L ∂L − = 0. (9.131) seems to modify the theory. (9. t = 0.128) (9.129) ∂ · A = −∂ · J = 0. this is just the Maxwell’s equations in the presence of conserved sources. we recognize that if we restrict classically to the initial value conditions ∂χ = 0. An alternate way to see this is to note that if we take the divergence of the equations of motion in (9. the presence of this additional term in the Lagrangian density would not change the physics of Maxwell’s theory. χ is a free ﬁeld and. (9. (9.127) takes the form ∂µ (∂ µ Aν − ∂ ν Aµ ) + ∂ ν (∂ · A) = −J ν .

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.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. φ(y) = iδ3 (x − y). ˙ Aµ (x). classically we can think of Maxwell’s theory as described by the modiﬁed Lagrangian density (9.136) with all other commutators vanishing. 2 (9. Πν (y)]x0 =y0 = iδµ δ3 (x − y). Thus.9.133) (9. ˙ ∂ Aµ Πµ = (9. We can deﬁne the canonical momenta as ∂L ˙ = −Aµ . Aν (y) x0 =y 0 = −iηµν δ3 (x − y). We now see that this quantization relation is exactly like the quantization condition for four distinct scalar ﬁelds (see (5.137) x0 =y 0 .125) with the supplementary condition ∂ · A = 0.8 Covariant quantization of the Maxwell theory 365 then χ = 0 at all times and we recover the familiar Maxwell’s theory. (9.134) It is now obvious that the coeﬃcient matrix of highest derivatives is nonsingular in this case. or.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).

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

we can weaken the supplementary condition (9. (9. not made use of the supplementary condition (9.) We note that.8 Covariant quantization of the Maxwell theory 367 which is the correct Heisenberg equation of motion for the ﬁeld variables Aµ and hence the quantization relations (9. where Aµ ’s are operators.) As we would see shortly.140) in the quantum theory.133) were to hold as an operator equation. namely.9. We have.142) In other words.133) as yet. 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. we can think of the supplementary condition as selecting out the subspace of the physical Hilbert space of the theory. or. It is easy to see that the commutation relation (9.133) may be thought of as picking out the smaller physical subspace from the larger total vector space.142) is a very stringent condition and we have to relax this further. even (9. (Note that since all ﬁeld components are dynamical in this modiﬁed theory. = i∂ ν δ3 (x − x′ ). = i∂ ν δ3 (x − x′ ). A little bit of thinking tells us that although in the classical theory we can impose the condition ∂µ Aµ = 0. whereas the left-hand side of the above expression would be zero if the supplementary condition (9. however. (9. We .141) x0 =x′0 ∂µ Aµ (x).136) would lead to Aµ (x). Thus. Πν (x′ ) or. ∂µ Aµ |phys = 0.133) and say that this condition is true only on the space of physical states of the theory. Πν (x′ ) ∂µ Aµ (x). but the derivation needs to be done carefully in a limiting manner x0 → x′0 . the non vanishing right-hand side would lead to an inconsistency. The supplementary condition (9. Πν (x′ ) x0 =x′0 x0 =x′0 (This relation is essentially correct.136) are consistent. such a condition is hard to implement as an operator condition. = iη µν δ3 (x − x′ ).

λ) (9. λ) as annihilation and creation operators satisfying a(k. λ). λ). At this point we can quantize the coeﬃcients a(k. 3.147) = −η λλ δ3 (k − k′ ).) They are normalized as 1. we note that if |ψ represents a physical state. here they are not required to be transverse. x0 =x′ 0 = ψ|i∂ ν δ3 (x − x′ )|ψ . 2.74)). λ)’s are the components of the polarization vector for a photon travelling along k and polarization index λ = 0. let us make a plane wave expansion for the ﬁeld variables Aµ (x) = λ ∗ + ǫ∗ (k. then ψ| ∂ · A(x).368 9 Maxwell field theory can see this already from the commutator relation (9. Basically. (Unlike the earlier discussion. ˆ while that for λ = 3 is longitudinal (recall kµ ) and the other two polarization vectors are transverse to the momentum four vector.141). k0 = ω(k) = |k|. (9.145) Here ǫµ (k. λ = λ′ = 0. 0 = i∂ ν δ3 (x − x′ ). To proceed further. 1. 2. . λ). a† (k′ . (9. λ′ ) = η λλ = ′ As we have seen earlier (see (9. a(k′ . λ) . −1.146) ǫµ (k.143) which is inconsistent. λ = λ′ = i = 1. λ) and a† (k. λ′ ) = 0 = ′ a† (k. λ)ǫ∗µ (k. λ)fk (x)a(k. we can always choose a basis such that the polarization vector for λ = 0 is time-like (recall η µ ). (9. µ d3 k ǫµ (k. (9. λ′ ) a(k. Πν (x′ ) |ψ or.144) where fk (x) = 1 (2π)3 2ω e−ik·x . λ′ ) . a† (k′ . λ)fk (x)a† (k. 3.

(9. we obtain the equal-time commutator to be ˙ Aµ (x). λ′ ) − ǫ∗ (k′ . 2. λ′ = 0. λ′ ) µ = λ.8 Covariant quantization of the Maxwell theory 369 where λ. λ′ )fk (x)fk′ (y) a(k. 3. λ)ǫ∗ (k. λ) ∗ µ f (x)fk (y) . λ′ ) ν ∗ −ǫ∗ (k. λ′ )fk (x)fk′ (y) + ǫ∗ (k.λ′ d3 kd3 k′ (−iω ′ η λλ )δ3 (k − k′ ) ′ ∗ ∗ × ǫµ (k. a(k′ . ǫσ (k. λ)ǫ∗ (k′ . λ) (9.148) Using the completeness relation for the polarization vectors ǫµ (k. λ)ǫν (k′ .150) . λ)ǫν (k. ǫσ (k.148). λ)ǫ∗σ (k. λ) k (9. λ′ ) ν ∗ d3 kd3 k′ ǫµ (k. a† (k′ . λ) + ǫ∗ (k.149) λ in (9. Aν (y) = λ. λ) + ǫ∗ (k. λ′ )fk′ (y)a† (k′ .9. λ) ν ∗ fk (x)fk (y) ǫσ (k. λ).λ′ d3 kd3 k′ iω ′ ∗ × ǫµ (k. µ = λ. λ)ǫν (k′ . we do not write the equal-time arguments explicitly although they are assumed) ˙ Aµ (x). λ)ǫ∗σ (k. λ′ )fk′ (y)a(k′ . λ′ )fk (x)fk′ (y) a† (k. λ)ǫ∗ (k′ .λ′ ∗ −iω ′ ǫν (k′ . Aν (y) = ∗ ∗ d3 k(−iω)ηµν [fk (x)fk (y) + fk (x)fk (y)] = −iηµν δ3 (x − y). λ)fk (x)a(k. we have to check that these conditions are consistent with the equal-time quantization conditions for the ﬁeld variables (for simplicity. λ) ν = ηµν . λ). λ′ )fk (x)fk′ (y) ν µ = λ d3 k(−iω) ǫµ (k. Of course. λ)ǫ∗ (k. 1. λ)ǫ∗σ (k. λ). λ)fk (x)a† (k.

We also note from (9. let us denote by |0 the vacuum state of the theory so that 0|0 = 1. it is strange to note that there are four independent photon degrees of freedom. λ = 0 = d3 k F (k)a† (k. 2) + a† (k. 3. However.152) Let us next consider the state with one time-like photon |1. 3) . 3. (9. 1. 2)a(k. 1) +a† (k. λ = 0. The norm of this one photon state can now be calculated (9. For example. λ) behave respectively like annihilation and creation operators for λ = 1. λ = 0)|0 .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. a(k. 3)a(k. λ)|0 = 0. If we calculate the normal ordered Hamiltonian of the theory (recall J µ = 0). 2.147). 2.147) that although a(k. λ) and a† (k. (9. it has the form :H: = d3 k ω(k) −a† (k.154) . 0) + a† (k. (9. 0)a(k.370 9 Maxwell field theory Thus the commutation relations for the creation and the annihilation operators are indeed consistent with the quantization condition for the ﬁeld variables.153) where F (k) is a suitable smearing function such that d3 k |F (k)|2 < ∞. 1)a(k. for the time-like component there is a relative negative sign in the commutation relation (9.

Let us suppose that a† (k.8 Covariant quantization of the Maxwell theory 371 1.157) In this case.9. (9. 1)a(k. 0)|0 . 0) − δ3 (k − k′ )|0 d3 k |F (k)|2 < 0. However.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.151)) :H: = d3 k ω(k) −a(k. We also realize that this negative norm is a consequence of the negative sign in the commutator (9. Thus the natural modiﬁcation that comes to mind is to interchange the roles of these operators for the time-like photon. 1) +a† (k. 0)a† (k. 0)|0 = 0. so that the one time-like photon state is given by (9. λ = 0|1. (9. this also leads to trouble as we see below. 0) + a† (k. the normal ordered Hamiltonian will have the form (compare with (9.156) |k. λ = 0 = = = − d3 kd3 k′ F ∗ (k)F (k′ ) 0|a(k.158) and the energy of this one photon state is given by . 0)a(k. 0)|0 d3 kd3 k′ F ∗ (k)F (k′ ) 0|a† (k′ . 2) + a† (k. 3)a(k. 0 = a(k. 3) .147) for the annihilation and the creation operator for a time-like photon. 2)a(k. (9. 0)a† (k′ .

1. 0 = Ha(k. 0)|0 = [H. But in the presence of interactions such states may be excited. Namely. where we have used the fact that H|0 = 0. (9. In fact it is obvious that with this deﬁnition. As we will see shortly in spite of the presence of negative norm states. This is seen by using the supplementary condition (subsidiary condition) which we still have not imposed. the energy of the state with one time-like photon becomes negative. 0)a† (k′ . deﬁnite states of arbitrarily high negative energy values are possible and the Hamiltonian becomes unbounded from below. (9.160) where nλ . In a free theory we may impose suitable conditions to prohibit any unwanted state. 0 . The question that immediately comes to our mind is what happens to the probabilistic interpretation of the theory. 0)]|0 = d3 k′ ω(k′ )[−a(k′ . 0)]|0 (9.159) = −ω(k)a(k. n1 . 2. 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 . the physical states must satisfy ∂µ Aµ (x)|phys = 0. physical results are well behaved. Thus we see that if we exchange the roles of the creation and the annihilation operators for the time-like photon. mλ denote number of photons for each of the polarizations λ = 0. 0)|0 = −ω(k)|k. 3. Thus states containing an odd number of time-like photons have negative norm. m3 = (−1)n0 δn0 m0 δn1 m1 δn2 m2 δn3 m3 . as discussed earlier this is too stringent a condition to allow any state of the radiation ﬁeld. This not only demands that certain . n3 |m0 . a(k.372 9 Maxwell field theory H|k. a(k.161) However. m2 . n2 . 0). m1 .

74) as well as the fact that for positive energy photons k0 = ω = |k|. k0 a(k. Let Vphys = {P } denote the set of all states which satisfy the supplementary condition and let |ψ represent such a state.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. as we have seen. the supplementary condition (9.9. kµ ǫµ (k. 3)a(k. but it also requires that those photons cannot be emitted.163) ∂µ Aµ (x) is like a free scalar ﬁeld.130)) ∂µ Aµ (x) = 0. 0) + ǫµ (k.) We remark here that since (see (9. or. Therefore. (9. (This is commonly known as the Gupta-Bleuler quantization. where we have used the deﬁnitions and the properties of the polarization vectors in (9. 2.165) implies that .129) or (9. 3) |phys = 0. 0)a(k. Then. (9.8 Covariant quantization of the Maxwell theory 373 kinds of photons are not present in the physical state. λ) = 0 for λ = 1. it can be decomposed into positive and negative frequency parts uniquely in a relativistically invariant manner and this decomposition is preserved under time evolution.165) (a(k. 0) − a(k. λ = 0) − k2 a(k. |k| (9. λ)a(k. λ = 3) |phys = 0. (9. λ or. 3))|phys = 0. or.164) the Gupta-Bleuler supplementary condition leads to kµ ǫµ (k. Recalling that kµ ǫµ (k. Gupta and Bleuler weakened the supplementary condition to have the form ∂µ Aµ (+) (x)|phys = 0. λ)|phys = 0.

166) This leads to the fact that the physical states must contain superpositions of states with an equal number of time-like and longitudinal photons.n2 .n1 .n2 . All the coeﬃcients Cn0 .n1 . A general physical state is.n3 Cn0 .n3 −1 + n3 Cn0 −1. a superposition of diﬀerent states of the form |ψ = n0 . or. 0) − a(k. n2 .n1 .168) This can be seen as follows or. n2 .n3 in the expansion can be determined recursively from (9.n2 . n3 − 1 = 0. n3 − 1 = 0. n2 . n1 . √ Cn0 . (9.170) .n1 . n2 . 3)|ψ . 3)|ψ .n3 ) ×|n0 − 1.n1 .n2 .n2 .0 = 1. n3 . of course. (9. (9.n2 .n1 . n1 . 0)|ψ = a(k. √ √ ( n0 Cn0 .374 9 Maxwell field theory a(k. Cn0 .n1 . n1 .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 . ψ|a† (k. 0)|ψ = ψ|a† (k. 0)a(k.n1 .n2 .n3 (a(k.168) with the condition C0. n1 . n3 = 0. n1 . or.n3 = 0. n2 .n1 .169) which leads to the relation in (9.168).n2 . (9. 3)a(k.n2 . (9.n3 − n0 |n0 − 1. 3)) |n0 .n3 |n0 .n3 −1 + n3 Cn0 −1.n2 . n3 √ − n3 |n0 .

n1 . 2|0. n2 . 1 + 2. n1 . (9. 1 + |2. n2 . (9. .n1 . all such states except for |φ0 have zero norm. n1 . n1 . 2 + 2 1. 1|0. . n2 .172) and so on. n1 . 1|1. n1 . n1 . 1 − |1. = |0. 0|2. = 0. n0 = n3 = 0. n1 .9. (9.174) we can assume that in a truly physical state of the radiation ﬁeld there are no time-like and longitudinal photons present. n1 . n2 . n2 of transverse photons as a linear superposition of states of the form |φ because φ|φ = |φ0 + b1 |φ1 + b2 |φ2 + . In other words. n2 of transverse photons.173) (9. √ = |0. n1 . 2 − 2|1. 0 = 1 − 2 + 1 = 0. n1 . the physical states allowed by the supplementary condition have the forms |φ0 |φ1 |φ3 = |0. 0 . 0|1. 0 . 0 . n2 . 0 = 1 − 1 = 0.171) and so on where we have used the values of Cn0 . 1 + 1. n2 . For example. = φ0 |φ0 . (9.175) . n1 . n2 .8 Covariant quantization of the Maxwell theory 375 It follows now that for a ﬁxed number n1 . n2 . n1 . n1 . n1 . Thus although we can write a general physical state with a ﬁxed number n1 .n2 . n1 . we have φ1 |φ1 φ2 |φ2 = 0. Let us note that because of the negative norm of states containing an odd number of time-like photons. n2 . . n2 . n2 . n2 . n2 . n2 .n3 ’s determined recursively. n2 . n2 .

namely. Since the zero norm states are orthogonal to all the states including themselves. + iǫ (9. The physical S-matrix elements. on the space of physical states since ψ|∂µ Aµ |ψ = 0.µν (k) = lim − ǫ→0+ k2 iηµν . 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. If interactions are present time-like and longitudinal states may occur as intermediate states. Without going into details we note here that the Feynman propagator for the photon in this theory has the simpler form iGF . A diﬀerent way of saying this is to note that being orthogonal to every other state.165) contains states with positive semi-deﬁnite norm (negative norm states are eliminated by the supplementary condition or the physical state condition and. are gauge . we have the true physical subspace of the theory where the norm of states is positive deﬁnite. (9. such states decouple. Vphys .77) or (9. there is no problem with a probabilistic interpretation).376 9 Maxwell field theory in such states. consequently. V0 V phys = (9. however.177) (9.89) simply because this can be thought of as quantizing the theory in a diﬀerent gauge (FeynmanFermi gauge).178) where V0 represents the set of states with zero norm.176) where |ψ represents a physical state. 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.179) This is diﬀerent from (9. However their eﬀect cancels out in the ﬁnal result. The physical subspace of the theory selected by the supplementary condition (9. if we further mod out the states by the zero norm states.

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

This page intentionally left blank .

The Lagrangian is a func˙ tion of N coordinates qi as well as N velocities qi which are assumed ˙ to be independent so that the conﬁguration space of the theory is 2N dimensional. In 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. we can go from the conﬁguration space of the system to the phase space 379 . thereby. 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. let us consider a classical system of point particles described by the Lagrangian L(qi . 2. ∂ qi ˙ (10.Chapter 10 Dirac method for constrained systems 10.1) Equation (10. rather some of these variables have to satisfy constraints following from the structure of the theory.1 Constrained systems As we have seen in the study of the Dirac ﬁeld theory in chapter 8 as well as the Maxwell ﬁeld theory in chapter 9. To appreciate the diﬃculties associated with constrained systems.1). qi ). Given the Lagrangian of the theory. the dynamical phase space variables of these theories are not all independent. In this case. i = 1. N . we deﬁne the momenta canonically conjugate to the coordinates as pi = ∂L . in general. 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. · · · . velocities can be expressed in terms of momenta (and coordinates).

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

pi ) = {A(qi .10. pi ).7) This discussion can also be generalized to a classical system described by Grassmann (anti-commuting) variables.12) The Poisson bracket between any two dynamical variables can now be obtained from (10. θα ) → (θα . (10.9) and as discussed in chapter 8 (see (8. θα ). · · · M with θα θβ = −θβ θα . (10. θα ). 2.14)).1 Constrained systems 381 ˙ A(qi . (10. α = 1. unlike the bosonic case in (10. (10. . θβ } = 0 = {Πα .9) However. we will choose the convention of taking derivatives from the left. H}. Πβ }. Πα ) = θα Πα − L(θα . we can deﬁne the momenta conjugate to the coordinates as Πα = ∂L .9) can be inverted to express velocities in terms of the momenta. Πα ). θα }.11) The 2M dimensional phase space is endowed with a canonical Poisson bracket structure (with the convention of left derivatives) {θα . since the dynamical variables are now of Grassmann odd (fermionic) nature. ˙ ∂ θα (10. we have to deﬁne the derivative in (10.8) then. Πβ } = −δα = {Πβ . For example. If the relations (10.5).12). θα ) = −Πα θα − L(θα . (10. β {θα .10) through the Legendre transformation (this is consistent with the convention of left derivatives) ˙ ˙ ˙ ˙ H(θα . we can go from the conﬁguration space to the phase space ˙ (θα . if ˙ we have a classical system described by the Lagrangian L(θα . However.

θβ } α ∂θα ∂Π ∂Π ∂θβ ∂B1 ∂B2 ∂B1 ∂B2 + ∂θα ∂Πα ∂Πα ∂θα .14) the 2M ﬁrst order dynamical equations for the system can now be written in the Hamiltonian form as ∂H ˙ θα = {θα . H} = − α . Πα ) takes the form ∂B2 ∂B1 α ∂B2 ∂B1 {θα . Thus. B2 (θα . B2 } = − = ∂B ∂B where we have used (10. (10. Πα = {Πα .14) Using (10. F2 } = − ∂F1 ∂F2 ∂F1 ∂F2 + α ∂θα ∂Π ∂Πα ∂θα . In a similar manner.6). Πα ). .2) or (10. {F1 . F } = ∂F ∂B ∂F ∂B + α ∂θα ∂Π ∂Πα ∂θα ∂B ∂F ∂B ∂F + ∂θα ∂Πα ∂Πα ∂θα .12) as well as the fact that ∂θα and ∂Πα are Grassmann odd (fermionic) variables. H} = − ∂θα (10.13) {B1 . 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. with the convention of left derivatives. B} = − {B.10) is invertible so that all the velocities can be expressed uniquely in terms of independent momenta leading to the . The discussion so far assumes that the transformation to phase space in (10. Πβ } β − {Π . (10.15) which can be compared with (10. for example. ∂Π ∂H ˙ . the Poisson bracket between two bosonic dynamical variables B1 (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).

16) can be easily seen to have the form A−1 = ½ 0 −b−1˜ b−1 b . To understand the diﬃculty. From (10.16) we note that pi = ˜ij (q)qj + bij (q)qj . ˜=˜ b b(q). it follows that the transformation is invertible (A−1 exists) if ∂2L = 0.2) can be written in general in the matrix form as q p q q ˙ =A = ½ 0 ˜ b b q q ˙ . ˜ denote N × N matrices. it follows that b = b(q).16) where the elements b.17) The inverse of the matrix A in (10. let us concentrate on the bosonic theory for simplicity.1 Constrained systems 383 unique Hamiltonian of the system. In this case. (10. (10. b ˙ so that ∂pi ∂2L = .18) so that A−1 exists only if the matrix b is invertible.10.20) Therefore. For conventional b theories described by Lagrangians with quadratic velocity terms. ∂ qj ˙ ∂ qj ∂ qi ˙ ˙ (10. (10.21) . the transformation to phase space in (10. However.19) bij (q) = (10. the diﬃculty in passage to a Hamiltonian description arises when this transformation is noninvertible. ∂ qi ∂ qj ˙ ˙ det (10.

∂ qi ∂ qj ˙ ˙ (10. pb ). It follows from (10. may not represent the true brackets necessary for a Hamiltonian description of the system. the passage to a Hamiltonian description is achieved systematically through the method of Dirac which we describe in this section.2) is not invertible. On the other hand. it is not clear a priori how to deﬁne the Hamiltonian uniquely.22) then. In this case.4) need not satisfy the constraints of the theory and.23) of the physical system is of rank R < N . As a result. ˙ a.384 10 Dirac method for constrained systems The naive passage to the Hamiltonian description can be carried out only in this case. In other words. ∂ qi ∂ qj ˙ ˙ det (10. R. if the Lagrangian describing the theory satisﬁes ∂2L = 0. we can write . In this case.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. it is also clear that the naive canonical Poisson brackets in (10. the transformation (10. 2. Let us assume that the N × N matrix ∂2L . (10.2 Dirac method and Dirac bracket In a physical system with constraints. therefore. b = 1. 10. · · · . Furthermore. we can determine only R of the N velocities in terms of coordinates and independent momenta as qa = fa (qi . therefore. 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 note from (10. we can deﬁne the canonical Hamiltonian of the theory as the Legendre transform Hcan = pi qi − L(qi . Hcan = Hcan (qi . 2. ˙ ∂L ∂Hcan = gα − = gα − pα ≈ 0. namely. α = R + 1. we note that the constraints in (10.28) that in spite of the fact that qα cannot be determined. the canonical Hamiltonian does not depend on velocities as we would expect. On the other hand.28) where we have used (10. · · · .25) deﬁne N − R constraints among dynamical variables which we denote by ϕα = pα − gα (qi . qi ).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 also note that because of the non-invertibility of (10. b = 1. pa ). from the canonical .25).29) so that on the constrained hypersurface Γc . · · · . ˙ ˙ ˙ ˙ ˙ (10. the dimensionality of Γc would be further reduced. qb ).27) Γc As usual. (If there are further constraints in the theory. if (F − G) = 0. pa ). R + 2.) Two dynamical variables F. (10. if they are equal on the constrained hypersurface Γc . R. ∂ qα ˙ ∂ qα ˙ (10.25) It is clear that the second set of equations in (10.2 Dirac method and Dirac bracket 385 pa = ga (qi .26) and we recognize that such constraints will reduce the dimensionality of the true phase space of the system.10. In fact. For example. ˙ pα = gα (qi . the Hamiltonian of the theory is no longer unique. N. G in Γ are said to be weakly equal. (10.2) (presence of constraints). pa ) = 0. namely. (10. qi ) = pa qa + gα qα − L(qi . F ≈ G. a.

26) as (it is known as the primary Hamiltonian since it incorporates only the primary constraints in (10. Hcan } + λβ {ϕα . The time evolution of any dynamical variable F can now be written as ˙ F ≈ {F.34) The constraints of the theory should be invariant under time evolution and using (10.35) . (10.31) Furthermore. Hp } = {F. Hp } = ˙ pi ≈ {pi .33) which remain undetermined.30) where λα denote undetermined Lagrange multipliers and it follows that Hp ≈ Hcan . λβ }ϕβ ≈ {ϕα . Hcan + λβ ϕβ } ˙ = {ϕα .4). we can deﬁne the primary Hamiltonian associated with the system by incorporating the primary constraints of the theory (10.26)) Hp = Hcan + λα ϕα .30) with velocities λα ≈ qα . (10.32) qi ≈ {qi . ˙ (10. ∂pi ∂pi ∂Hp ∂(Hcan + λα ϕα ) =− . (10. (10. it follows that we can identify the Lagrange multipliers in (10. ϕβ } + {ϕα .34) this leads to ϕα ≈ {ϕα .386 10 Dirac method for constrained systems Hamiltonian of the theory. Hcan } + λβ {ϕα .29). Hp } = − ˙ and using (10. ∂qi ∂qi (10. Hcan + λα ϕα }. we can write the Hamiltonian equations as ∂Hp ∂(Hcan + λα ϕα ) = . with the canonical Poisson brackets in (10. ϕβ } ≈ 0.

α = 1. We continue with this process until all the constraints are determined to be evolution free. α = 1. (10.38) that n1 + 2n2 = n < 2N .36) which can be divided into two distinct classes. · · · . secondary. tertiary. 2. n < 2N.27)) Poisson bracket with all the constraints (including themselves) are known as ﬁrst class constraints and are denoted as e ψ α ≈ 0. · · · .35) may determine some of the Lagrange multipliers or may lead to new constraints known as secondary constraints.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. · · · . 1.36)-(10.37) 2. n1 .35) (with Hp as the Hamiltonian) may determine some other Lagrange multipliers or generate newer constraints (tertiary etc). if there are primary constraints in the theory which are ﬁrst class. In general. α = 1.10. 2n2 . all of the Lagrange multipliers may be completely determined or some of them may remain undetermined. (10.2 Dirac method and Dirac bracket 387 where we have used (10. . . It is clear that (10. then there remain undetermined Lagrange multipliers in the primary Hamiltonian after all the . ) collectively as ϕα ≈ 0. Let us denote all the constraints of the theory (primary. It is also worth noting that in this process of determining all the constraints that are time independent (evolution free). On the other hand.26) in the last step. 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. (10. . 2. Requiring the secondary constraints to be invariant under time evolution as in (10. The constraints which have weakly vanishing (see (10.

requires gauge ﬁxing conditions and we must choose as many gauge ﬁxing conditions as there are ﬁrst class contraints. 2. · · · . Although we have identiﬁed all the constraints of the theory. in this case.42) This issue of incompatibility of the Poisson brackets can be addressed through the use of the Dirac brackets which are constructed as follows. The gauge ﬁxing conditions are generally denoted as e χα ≈ 0. 2. a consistent Hamiltonian description. Namely. (10.) The ﬁrst class constraints have a special signiﬁcance in that they are associated with gauge invariances (local invariances) in the theory. let us note that we can construct the matrix of Poisson brackets of all the constraints in (10. which are all second class. α = 1. even though φA ≈ 0. we cannot yet impose them directly in the theory since the canonical Poisson brackets in (10.4) are not compatible with them. vanish {F. φA } ≈ 0.40) as . 2(n1 + n2 ) < 2N. (10. n1 . First. collectively as φA ≈ 0. · · · . (The number of undetermined Lagrange multipliers equals the number of ﬁrst class constraints among the primary constraints. in general.37) into second class constraints.39) and are chosen such that they convert the ﬁrst class constraints (10.388 10 Dirac method for constrained systems constraints have been determined. As we have seen in the case of Maxwell ﬁeld theory. Thus. after gauge ﬁxing we denote all the constraints of the theory (including the gauge ﬁxing conditions). / (10.41) the Poisson bracket of the constraints with any dynamical variable F does not. A = 1.40) so that the true phase space of the theory is 2(N − n1 − n2 ) dimensional. (10.

In fact. As a result. G}. we can now deﬁne the Dirac bracket between any two dynamical variables as −1 {F. φB } ≈ C AB . (10. G} − {F. the matrix of Poisson brackets in (10. Jacobi identity) and in addition satisﬁes −1 {F. φB }CBD C DA = {F.40). (10. φA } − {F. (10. G}D = {F. φA }CAB {φB .45) which can be shown to have all the properties of a Poisson bracket (anti-symmetry. 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.43) This is an even dimensional anti-symmetric matrix and Dirac had −1 shown that it is nonsingular so that its inverse CAB exists. we can show that the Dirac brackets are indeed the correct Poisson brackets of the theory subject to the constraints in (10. through the use of Lagrange brackets. we can work with the Dirac brackets and set the constraints to zero in the theory (namely.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. in the deﬁnitions of the Hamiltonian.2 Dirac method and Dirac bracket 389 {φA . φA } − {F.43) will be coordinate dependent and the product in (10. The Dirac bracket has the interesting property . (10. in contrast to the canonical Poisson brackets.46) Namely. φA } − {F. φA } −1 ≈ {F. φA }D = {F. φA } = 0. φB }CBD {φD .44) We note here parenthetically that in ﬁeld theories (where the variables depend on space coordinates as well). With this. namely.44) will involve integration over the intermediate space coordinate. A −1 −1 C AD CDB = δB = CBD C DA .

rather than dealing with a large matrix of Poisson brackets in (10.49) gives the constraint (10. let us consider the motion of a point particle constrained to move on the surface of a ndimensional sphere of radius unity. Namely. therefore.390 10 Dirac method for constrained systems that it can be constructed in stages. · · · . then the matrix of highest derivatives has the form .47) on the dynamics. ˙ ˙ 2 L= (10. 2. a = 1. when the number of constraints is large. 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. n. 2. It is also worth emphasizing here that in dealing with ﬁeld theories where the matrix C AB is coordinate dependent. ∂F 2 (10. 10. F into qa = (qi .48) where F is a Lagrange multiplier ﬁeld (an auxiliary ﬁeld without independent dynamics) whose Euler-Lagrange equation 1 ∂L = − (qi qi − 1) = 0. F ). n + 1. the inverse needs to be deﬁned carefully with the appropriate boundary condition relevant for the problem. We note that if we combine the dynamical variables qi .43) and its inverse. raising and lowering of indices can be carried out trivially) qi qi = 1.3 Particle moving on a sphere As a simple example of constrained systems. If we denote the coordinate of the particle as qi . We note that the dynamics of the system can be described by the Lagrangian 1 (qi qi − F (qi qi − 1)) . This simple system is an interesting example in the study of constrained quantization and is a prototype of the ﬁeld theoretic model known as the nonlinear sigma model.47) where summation over repeated indices is understood. (10. then the coordinates are constrained to satisfy (the metric in this simple theory is Euclidean and. · · · . i = 1.

3 Particle moving on a sphere 391 ∂2L = ∂ qa ∂ qb ˙ ˙ δij 0 0 0 . ˙ ∂ qi ˙ ∂L = 0.51) has the form 1 i i ˙ Hcan = pi qi + pF F − L = ˙ p p + F (qi qi − 1) .10. we note that the conjugate momenta of the theory are given by ∂L = qi . pF } = 0. (10.54) The equal-time canonical Poisson brackets of the theory are given by j {qi .30)) 1 i i p p + F (qi qi − 1) + λ1 pF . pF } = 0. 2 (10. pF } = {pi . pj } = δi .55) .52) The canonical Hamiltonian following from (10. {F.53) Hp = Hcan + λ1 ϕ1 = (10.48) and (10. F } = {pi . the theory has a primary constraint given by ϕ1 = pF ≈ 0. 2 leading to the primary Hamiltonian (see (10. F } = {qi .22)) reﬂecting the fact that there are constraints in the theory and the naive Hamiltonian description would not hold. (10.51) Therefore. To understand the passage to the Hamiltonian description in this simple case. pj } = {F. F } = {pF . pF } = 1. qj } = {pi . {qi . {qi .50) which is indeed singular (see (10. (10. ˙ ∂F pi = pF = (10.

59) Requiring the new constraint (10.59) to be time independent. (10. Hp } ˙ 1 1 = { (qi qi − 1).57) to be time independent. pj } = qi pi ≈ 0.58) which generates a new constraint ϕ3 = qi pi ≈ 0. (pj pj + F (qj qj − 1) + λ1 pF )} 2 ≈ pi pi − F ≈ 0. (10. In particular.56) Let us denote this secondary constraint as ϕ2 = 1 (qi qi − 1) ≈ 0. 2 (10. qj } = pi pi − F qi qi (10. Hp } ˙ 1 = {pF . requiring the primary constraint to be independent of time leads to ϕ1 ≈ {ϕ1 . we obtain ϕ2 ≈ {ϕ2 . pj } + qi F qj {pi . we obtain ϕ3 ≈ {ϕ3 .57) Requiring (10. (pj pj + F (qj qj − 1) + λ1 pF )} 2 2 = qi pj {qi . (pi pi + F (qi qi − 1) + λ1 pF } 2 1 = − (qi qi − 1) ≈ 0. Hp } ˙ 1 = {qi pi . = pi pj {qi . 2 (10.392 10 Dirac method for constrained systems With these we can now determine the time evolution of any dynamical variable.60) .

ϕ4 } = {pF .10.59) and (10. pi pi − F } = 1 1 {ϕ2 . we can calculate the matrix of the (equal-time) Poisson brackets of constraints {φA . (pj pj + F (qj qj − 1) + λ1 pF )} 2 (10. ϕ2 }. . It is easy to check that all the constraints of the theory (10. pj } = 2pi pi = 2p2 = −{ϕ4 . pF } = −2F qi pi − λ1 ≈ −λ1 ≈ 0.62) = 2pi F qj {pi . 3. the nontrivial equal-time Poisson brackets between the constraints take the forms {ϕ1 . 2. Requiring (10. (10.57). (10.61) 1 = {pi pi − F.61) to be time independent leads to ϕ4 ≈ {ϕ4 .61) are second class. ϕ3 }. = −{ϕ4 .52). Collecting all the constraints into φA = (ϕ1 . φB } = C AB . ϕ2 . which determines the Lagrange multiplier and the chain of constraints terminates. ϕ3 } = { (qi qi − 1). ϕ4 } = {qi pi . qj pj } = qi qj {qi .63) where we have identiﬁed pi pi = p2 . Hp } ˙ (10. In fact. ϕ3 . (10. 4. pj } = qi qi ≈ 1 2 {ϕ3 . A = 1. qj } − λ1 {F.3 Particle moving on a sphere 393 so that we can identify ϕ4 = pi pi − F ≈ 0. ϕ1 }. ϕ4 ). pj pj − F } = 2pi pj {qi . which takes the form (10.64) = −{ϕ3 .

pj } 2 −1 1 {qi . F } = 2pi .45)) −1 {F. pF }D = 0.65) −1 CAB The Dirac brackets between any two dynamical variables can now be deﬁned as (see (10. pj } − {pi .66) (10. {qi .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 . G}. pj } 2 −1 1 −{pi . pj } − {qi . qk pk }C32 { (qℓ qℓ − 1). G}D = {F. qj }D = 0. pF }D = {pi . (qj qj − 1)}C21 {pF . F } − {qi . the Dirac brackets between the fundamental dynamical variables take the forms 1 −1 {pi . (10. (qk qk − 1)}C23 {qℓ pℓ . F }D = {pF . pj }D = {pi . F }D = {qi . {F.66). 2 2p = 0 1 0 . = 0 −1 0 2p2 2 −1 0 −2p 0 0 0 0 1 (10. pj } 2 j = δi − qi qj . G} − {F. F } − {pi . 0 0 (10. 1 −1 {pi . φA }CAB {φB . pF }D = 0. = −qi pj + pi qj . pF }D = {F. pj }D = {qi . F } = −2qi p2 . 2 {qi . In particular.67) −1 with CAB given in (10. −1 {qi .68) . qk pk }C32 { (qℓ qℓ − 1). pj pj − F }C41 {pF . F }D = {pi .

pi ) (we note that F = pi pi because of the constraint (10.10.69) where we have used (10.62). Let us next study the Hamiltonian description of a free relativistic massive particle which is described by the equation d2 xµ = 0.70) can also be written as dpµ = 0.61)) and the Hamiltonian for the theory is given by (see (10.71) then the dynamical equation (10. dτ (10. This is the starting point for the quantization of this theory. 2 Hp = (10. We note that if we deﬁne the relativistic four velocity and momentum as uµ = dxµ .54)) 1 i i pp.73) where λ is a parameter labelling the trajectory and we have identiﬁed .72) The dynamical equation (10.4 Relativistic particle In the last section we studied a simple example of point particle dynamics which is constrained.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. (10. dτ pµ = muµ . The constraints of this theory were all second class. dτ 2 m (10. we can set the constraints to zero so that the true independent dynamical variables are (qi . 10.4 Relativistic particle 395 With the use of the Dirac brackets.70) where τ denoting the proper time is assumed to label the trajectory of the particle and m is the rest mass of the particle. (10.

dλ dλ dξ λ → ξ = ξ(λ).73) will lead to the dynamical equation (10.396 10 Dirac method for constrained systems xµ = ˙ dxµ .72) as the Euler-Lagrange equations.78) which leads to . then (10. so that (10.77) where τ denotes the proper time associated with the particle. However. (10.73) that ∂2L = ∂ xµ xν ˙ ˙ m xρ xρ ˙ ˙ xµ xν ˙ ˙ xσ xσ ˙ ˙ Pµν = ηµν − .75) which is an example of a local gauge transformation much like in the Maxwell ﬁeld theory. let us proceed with the Hamiltonian description without choosing a gauge at this point. We note from (10. (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) is invariant under the diﬀeomorphism (10.70) or (10. the action (10.74) We note that under a transformation xµ = ˙ dξ dxµ dxµ = . ˙ ˙ (10. dλ xµ = ηµν xν . If we choose a gauge λ = τ. (10.76) Namely.

73). We note that introducing an auxiliary variable g. g= xµ xµ ˙ ˙ . we note that the equation for the auxiliary ﬁeld g takes the form g−1/2 ∂L = −g−1 xµ xµ + m2 = 0.73) we can think of g as the metric on the one dimensional manifold of the trajectory of the particle. Pµν is a projection operator and ∂2L = 0. m2 (10. Viewed in this manner.73) in the form 1 2 √ ˙ ˙ dλ g g−1 xµ xµ + m2 . ∂ xµ xν ˙ ˙ det Pµν = det (10.81) using (10. then we obtain (10.81) is manifestly diﬀeomorphism invariant and the massless limit can now be taken in a straightforward manner.81).81) are equivalent.81) are xµ and g. it is more convenient for us to work with the action (10. The two dynamical variables of the theory in (10.80) as in the case of the Maxwell theory (see (9.22)). However.73) can also be written in an alternate form that is more useful. it is not clear how to take the m = 0 limit in (10. Let us note that the action (10.4 Relativistic particle 397 xµ Pµν = ˙ m xρ xρ ˙ ˙ xν − ˙ xµ xµ xν ˙ ˙ ˙ xσ xσ ˙ ˙ = 0 = Pµν xν . the action (10. To see that the two actions (10.73) to describe the motion of a massless particle.81) where because of the diﬀeomorphism invariance of (10. For example. we can rewrite the action (10. ˙ ˙ ∂g 4 or.79) Therefore.73) and (10. We can obtain the momenta conjugate to these variables to be .82).124)) and the naive passage to the Hamiltonian description fails (see (10.10. ˙ (10. S= dλ L = (10.82) If we eliminate g in (10.

398 10 Dirac method for constrained systems pµ = ∂L = g−1/2 xµ . ˙ ∂ xµ ˙ pg = ∂L = 0. pν } = {g.84) Hcan (10.84). µ {xµ . we obtain √ g µ p pµ − m2 + λ1 pg } 2 (10. Hp } = {pg .87) With these we can now determine the evolution of the primary constraint (10. we obtain the primary Hamiltonian for the theory to correspond to √ g µ p p µ − m 2 + λ1 p g .86) The equal time canonical Poisson brackets for the theory are given by {xµ . {xµ . g} = {pg . g} = {xµ .85) Adding the primary constraint (10. pg } = 1. Hp = Hcan + λ1 ϕ = 2 1 (10. pg } = 0. 4 g which leads to the secondary constraint . g} = {pµ . xν } = {pµ .88) ϕ ˙ 1 ≈ {ϕ . ∂g ˙ (10.84) and requiring the constraint to be time independent. 1 1 = − √ pµ pµ − m2 ≈ 0. pg } = 0. 2 µ (10. The canonical Hamiltonian of the theory follows to be √ g µ = pµ x + pg g − L = ˙ ˙ p p µ − m2 . pν } = δν . (10. {g.83) so that the primary constraint of the theory follows to be ϕ1 = pg ≈ 0. pg } = {pµ .

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

93) 1 0 −1 CAB The equal time Dirac bracket between any two dynamical variables is given by (see (10.86) vanishes. φ2 }C24 {φ4 . pν } − {xµ . −1 {g.95) and with (10. pg } {xµ . pg } = 0. pg } − {g.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 .45)) −1 {F. G}.96) pµ 0 δ .94) (10. xν }D = {pµ . m2 −1 {xµ . pg }{g − 1 . pν }D = {xµ .94) we can calculate the Dirac bracket between the fundamental variables to be {xµ . G} − {F. g}D = {xµ . µ = δν − 1 λ 1 p pλ − m2 } − 0 {λ − x0 . pν }D = {g. φ1 }C13 {φ3 . φA }CAB {φB . g}D = {pµ . pg }D = {g. p0 ν With the Dirac brackets we can set the constraints to zero in which case we see that the primary Hamiltonian (10. (On . pg }D = {pµ . pg }D = 0. = {g. g}D = {pg . 0 0 0 (10. G}D = {F. pg }D = 0. pg } − {g. pν } µ = δν − {xµ . 0 = −1 0 1 0 − p0 . pν } 2 p (10. 0 0 0 0 (10.

or.91) would imply dxµ dxµ = 1. it is obvious that the matrix of second order dervatives vanishes and the system is singular (see (10. dτ dτ (10. dxµ dxµ = 1.10. This manifests in the deﬁnition of the conjugate momenta as (we have chosen the convention of left derivatives) .97) On the other hand. (10. we have (remember that c = 1) dτ 2 = dxµ dxµ . dλ dλ xµ xµ = ˙ ˙ (10. (10.9). where the adjoint spinor is given by ψ = ψ† γ 0 .20)).98) It follows.99) (10. the free Dirac theory is described by the Lagrangian density / L = iψ∂ ψ − mψψ. 4 denote the Dirac indices and since the Lagrangian density is ﬁrst order in derivatives. if we set the constraints strongly to zero.5 Dirac field theory 401 the other hand. ψα .82) that the ﬁrst of the gauge ﬁxing conditions in (10.5 Dirac ﬁeld theory As we have seen in (8. 10. therefore.22)).100) † The dynamical variables of the theory can be chosen to be ψα . 2. from the deﬁnition of the inﬁnitesimal invariant length in the Minkowski space (see (1. that the ﬁrst of the gauge ﬁxing conditions corresponds to choosing λ = τ .89) in particular leads to H = p0 = p2 + m2 which can be taken as the Hamiltonian. where α = 1. 3.) Let us also note from (10.

α α ρα = Πα ≈ 0.) As a result. we obtain the two sets of primary constraints for the theory to be † ϕ† = Π† + iψα ≈ 0. α α (10. The canonical Hamiltonian is now obtained to be Hcan = d3 x Hcan = d3 x −iψγ · ∇ψ + mψψ .104) Adding the primary constraints to the canonical Hamiltonian.402 10 Dirac method for constrained systems Π† = α Πα = ∂L † = −i(ψγ 0 )α = −iψα .101) in the intermediate steps. (10.101) (Note that Π† is not the Hermitian conjugate of Π.102) which in this case correspond to fermionic constraints.11)) ˙ ˙† Hcan = −Π† ψα + ψα Πα − L α = −iψγ · ∇ψ + mψψ. † ˙ † ˙ = iψα ψα − iψα ψα − iψγ · ∇ψ + mψψ (10.103) where we have used (10. (10. ˙† ∂ ψα (10. 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 (10.30)) Hp = Hcan + d3 x φ† ξα + λ† ρα .105) . ˙ ∂ ψα ∂L = 0. Rather they correspond to the momenta conjugate to ψ and ψ † respectively.

Using the fact that the primary Hamiltonian is independent of time.102).13)(10. ρβ (y)} . ψβ (y)} α d3 y iψ(y)γ · ∇y − mψ(y) i∇ψ(x) · γ + mψ(x) α β − δαβ δ3 (x − y) . For example.107) α β Let us evaluate each of these terms separately. β (10. φ† (y)} α β = † † d3 y ξβ (y){Π† (x) + iψα (x). Hcan } + α d3 y ξβ (y){φ† (x).5 Dirac field theory 403 where the Lagrange multipliers ξα . −iψ(y)γ · ∇y ψ(y) + mψ(y)ψ(y)} α d3 y iψ(y)γ · ∇y − mψ(y) β {Π† (x). Using (10.14)).106) with all other brackets vanishing.10. Hcan } α = = = = † d3 y {Π† (x) + iψα (x). we can now calculate the time evolution of the primary constraints (10.12)) † {ψα (x). The equal-time canonical Poisson brackets for the ﬁeld variables take the forms (see (10. Hp } α α ≈ {φ† (x). d3 y ξβ (y){φ† (x). 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). Π† (y) + iψβ (y)} α β . Π† (y)} = −δαβ δ3 (x − y) = {ψα (x). φ† (y)} α β −λ† (y){φ† (x). λ† in the present case denote α fermionic variables.106) (see also (10. (10. Πβ (y)}. ˙ φ† (x) ≈ {φ† (x).

112) . α Substituting these results into (10. (10. ρβ (y)} α β = − = − † d3 y λ† (y){Π† (x) + iψα (x). − 10 Dirac method for constrained systems d3 y λ† (y){φ† (x).111) which determines the other Lagrange multiplier ξα = γ 0 (γ · ∇ + im)ψ(x) α . α α (10. Hp } ˙ = = † d3 y {Πα (x). (10. requiring the second set of primary constraints to be time independent we obtain ρα (x) ≈ {ρα (x).108) = iλ† (x).404 = 0. (10.110) Similarly. Πβ (y)} α β d3 y λ† (y) − iδαβ δ3 (x − y) β (10.109) which determines the Lagrange multiplier λ† = − ∇ψ(x) · γ + imψ(x) α α . 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) = 0.107) and requiring the constraint to be independent of time leads to ˙ φ† (x) ≈ i∇ψ(x) · γ + mψ(x) α + iλ† (x) = 0. Hp } = {Πα (x).

5 Dirac field theory 405 In this case. we have (at equal time) † † {φ† (x). (10. we have deﬁned Π = Π† γ 0 . Πβ (y)} = 0. α α β β † {φ† (x). (10. Substituting (10. deﬁne second class constraints (this is consistent with the fact that there is no undetermined Lagrange multiplier in the theory). β {ρα (x).114) Let us note next that the two sets of primary constraints (10. ρα ). ρβ (y)} = {Π† (x) + iψα (x).110) and (10. Π† (y) + iψβ (y)} = 0. for simplicity of notation. ρβ (y)} = {Πα (x). rather it determines both the Lagrange multipliers in the primary Hamiltonian (10.112) into the primary Hamiltonian. Πβ (y)} α α = −iδαβ δ3 (x − y) = {ρα (x).102).105).115) Thus. which constitute all the constraints of the theory. φ† (y)}. we obtain Hp = = d3 x −iψγ · ∇ψ + mψψ + φ† ξα + λ† ρα α α d3 x −iψγ · ∇ψ + mψψ † + (Π† + iψα ) γ 0 (γ · ∇ + im)ψ α α + = − ∇ψ · γ + imψ α Πα d3 x Πγ · ∇ψ + imΠψ + (−∇ψ · γ + imψ)Π . (10.10. φ† (y)} = {Π† (x) + iψα (x). In fact. combining the constraints into ΦA = (φ† .113) where. we can write α the matrix of Poisson brackets as . we see that requiring the two primary constraints to be time independent does not introduce any new constraint.

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

122) where the ﬁeld strength tensor (see (9. we can show that † † {ψα (x). Maxwell’s equations can be obtained as the Euler-Lagrange equations from the Lagrangian density 1 L = − Fµν F µν . ψβ (y)} γ d3 z(−δαγ δ3 (x − z))(−δγβ δ3 (z − y)) (10.4)) . Furthermore.113) becomes Hp = d3 x −iψγ · ∇ψ + mψψ .121) which is the Hamiltonian we have used in the quantization of the Dirac ﬁeld theory (see (8. 2.10. we recall that when using the Dirac brackets. 4 µ. As we have seen in (9.6 Maxwell ﬁeld theory Let us next consider the Maxwell ﬁeld theory which as we know exhibits gauge invariance.118) in evaluating this bracket because Πγ has a vanishing Poisson bracket with ψα . where we have neglected the last term in (10. ψβ (y)}D = {ψα (x).16). ν = 0. 1. ψβ (y)}D .16)).119) = −iδαβ δ3 (x − y). ψβ (y)}D = 0 = {ψα (x). Π† (z) + iψγ (z)}{Πγ (z).6 Maxwell field theory 407 † † {ψα (x). Similarly. the Hamiltonian of the theory (10. Therefore. ψβ (y)} −i = −i † † d3 z{ψα (x). we can set the constraints to zero in the theory. 10.18)). (10. (10.120) and these are the relations we have used in quantizing the Dirac theory (see (8. 3. (10.

(10.122)) and this manifests in the deﬁnition of the conjugate momenta as ∂L = −F 0µ . As a result of the gauge invariance of the theory. i.7)) F0i = Ei .126) so that we have Π0 = 0.5) and (9.125) where α(x) denotes the local parameter of gauge transformation. the matrix of quadratic derivatives becomes singular (see (9. 3. (10.408 10 Dirac method for constrained systems Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . (10. (10.124) We also know that Maxwell’s theory is invariant under gauge transformations Aµ (x) → A′ (x) = Aµ (x) + ∂µ α(x). k = 1.123) is the four dimensional curl of the four vector potential and contains the electric and the magnetic ﬁelds as its components (see (9.128) We can now obtain the canonical Hamiltonian density of the theory to have the form . µ (10.127) This determines that the theory has a primary constraint given by ϕ1 (x) = Π0 (x) ≈ 0. ˙ ∂ Aµ Πµ = (10. j. ˙ Π = E = −(A + ∇A0 ). 2. Fij = −ǫijk Bk .

128) Hp = Hcan + = d3 x d3 x λ1 ϕ1 1 2 (Π + B2 ) − A0 ∇ · Π + λ1 Π0 . = 2 (10.130) where we have neglected a surface term (total divergence) in integrating the last term by parts.10. Πν (y)}.129) where we have used (10.131) where λ1 denotes the Lagrange multiplier of the theory.127). is given by Hcan = = d3 x Hcan d3 x 1 2 (Π + B2 ) − A0 ∇ · Π . 2 (10. therefore.132) Using these we can calculate the time evolution of the primary constraint (10. Aν (y)} = 0 = {Πµ (x). Aµ (x)}. Πν (y)} = δµ δ3 (x − y) = −{Πν (y). ν {Aµ (x).6 Maxwell field theory 409 1 ˙ ˙ Hcan = Πµ Aµ − L = −Π · A + Fµν F µν 4 1 = −Π · (−Π − ∇A0 ) + (−E2 + B2 ) 2 1 2 (Π + B2 ) + Π · ∇A0 . 2 (10. The equal-time canonical Poisson brackets of the theory are given by {Aµ (x). (10.128) and requiring the constraint to be time independent we obtain (the Poisson brackets are at equal-time which can be achieved using the time independence of Hp ) . The canonical Hamiltonian. The primary Hamiltonian of the theory is now obtained by adding the primary constraint (10.

ϕ2 (x) ≈ {ϕ2 (x).136) and it is clear that both these constraints are ﬁrst class constraints. in the presence of ﬁrst class constraints we are supposed to add gauge ﬁxing conditions to convert them into second class constraints.410 10 Dirac method for constrained systems ϕ1 (x) ≈ {ϕ1 (x). In this case. as we have seen. Hp } ˙ = 1 d3 y {Π0 (x).133) = ∇ · Π(x) ≈ 0. (10. (10. A0 (y)}∇ · Π(y) (10. Hp } ≈ 0. the Lagrange multiplier λ1 remains as yet undetermined (there is one primary constraint which is ﬁrst class). Therefore.134) It can now be checked that the secondary constraint is time independent. ˙ so that the chain of constraints terminates. Let us choose .135) (10. 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. According to the Dirac procedure. We see that Maxwell’s theory has two constraints ϕ1 (x) = Π0 (x) ≈ 0. (Π2 (y) + B2 (y)) 2 − A0 (y)∇ · Π(y) + λ1 (y)Π0 (y)} ≈ − d3 y {Π0 (x). namely. 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. Hp } = {∇ · Π(x). we have a secondary constraint in the theory given by ϕ2 (x) = ∇ · Π(x) ≈ 0. ϕ2 (x) = ∇ · Π(x) ≈ 0. Furthermore.

139) The inverse of this matrix is easily calculated to be 0 0 1 0 −1 CAB (x. 2. = 1 0 0 0 2 0 ∇x 0 0 (10. y) −2 0 0 0 ∇x 3 δ (x − y). y) = {φA (x). relations (10. A0 (y)} {ϕ2 (x). Therefore. χ2 (x) = ∇ · A(x) ≈ 0.136) and (10. (10.6 Maxwell field theory 411 χ1 (x) = A0 (x) ≈ 0.137) as the gauge ﬁxing conditions which clearly convert the constraints (10. 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).10. χ2 (y)} = {∇ · Π(x).140) . χ1 (y)} = {Π0 (x). = (∇x )i (∇y )j {(Π)i (x). ∇ · A(y)} = −δ3 (x − y) = −{χ1 (x). Combining the ﬁrst class constraints as well as the gauge ﬁxing conditions into φA = (ϕα . (Π)j (y)} = (∇x )i (∇y )j (δij δ3 (x − y)) 2 = −∇x δ3 (x − y) = −{χ2 (x).138) which leads to (x0 = y 0 ) C AB (x.136) into second class constraints. ϕ2 (y)}. (10.137) determine all the constraints of the theory. χα ). α = 1. = −1 0 0 0 −2 0 −∇x 0 0 (10. φB (y)} 0 0 −1 0 2 0 0 0 −∇x 3 δ (x − y). ϕ1 (y)}.

Aj (y)}D = 0 = {Πi (x). ∇z · Π(z)}(∇z δ3 (z − z )){∇z · A(¯). Πν (y)} ¯ ¯ z {Aµ (x). Aν (y)}D = 0 = {Πµ (x). G(y)} − −1 ¯ z d3 zd3 z {F (x).143) to be {Ai (x). d3 zd3 z {Aµ (x).136) and (10. it leads to {Aµ (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).143) Since we can now set the constraints (10. Π0 (z)}(δ3 (z − z )){A0 (¯). (10.137) to zero. φA (z)}CAB (z. Πν (y)}D . for the dynamical ﬁeld variables. z ){φB (¯). with the usual asymptotic condition that ﬁelds vanish at spatial inﬁnity.142) ¯ 1 1 3 2 δ (x − y) = − 4π|x − y| . G(y)}. Πν (y)} − −2 ¯ z +{Aµ (x). Πj (y)}D = δi + (∇x )i 1 3 j 2 (∇x ) δ (x − y) ∇x (10.141) and. ∇x (10. G(y)}D = {F (x). Πj (y)}D .412 10 Dirac method for constrained systems where. in particular. 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). ∇x (10. the relevant Dirac brackets for the Maxwell ﬁeld theory follow from (10. Πν (y)}D = {Aµ (x).144) j = δi TR (x − y). j {Ai (x). .

6. setting the constraints (10. Berlin (1982). Bergmann. 3. Constrained Hamiltonian Systems. ibid. Anderson and P. 1018 (1951). 5. Furthermore. G. 1 (1951).7 References 413 which are the relations used in (9.30) (along with the ﬁrst two of (9. K. 129 (1950). Dirac. Roma (1976). Physical Review 83. A. P. Constrained Dynamics.7 References 1.10. M. A. we obtain the Hamiltonian for the theory (10. New York (1964).145) which is the Hamiltonian we have used in the study of Maxwell’s theory in chapter 9. Das and W. 2. A. A. Sundermeyer.136) to zero. Canadian Journal of Mathematics 2. 2 Hp = (10. Springer-Verlag. Regge and C. 3. Accademia Nazionale dei Lincei. P. T. Acta Physica Polonica B18. 4. Lectures on Quantum Mechanics. 10. Teitelboim. 269 (1987).25)) in quantizing Maxwell’s theory. M.131) to be d3 x 1 Π2 + B2 . J. . Hanson. Barcelos-Neto. Yeshiva University. J. Dirac. L. Scherer.

This page intentionally left blank .

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

1) that applying the parity transformation twice. returns the coordinates to their original value. be promoted to operators and they would satisfy the corresponding operator transformation properties. say.2) that even though a vector is expected to change sign under a parity transformation. (11. t) −→ ρ(−x. L·p |p| −→ P P P P P P P P P L · (−p) L·p =− . p −→ −p. t) −→ −J(−x. t) = ρ(x.416 11 Discrete symmetries A −→ −A. while a scalar (from the point of view of rotations. t) = J 0 (−x. L = x × p −→ (−x) × (−p) = x × p = L. A × B −→ (−A) × (−B) = A × B. we note from (11.4) . these variables would.2) and so on. let us note the transformation properties of some of the well known classical variables under parity. namely.2) that the volume (which behaves like a scalar under rotations) changes sign and is known as a pseudoscalar. A · B −→ (−A) · (−B) = A · B. t). This characterization carries over to the quantum domain as well. We see from (11. In particular. P (11. P P (11. x −→ −x −→ x. for example the length of a vector) is expected to remain unchanged under a parity transformation. t). x −→ −x. J 0 (x. We note from (11. of course. A · (B × C) −→ (−A) · ((−B)×(−C)) = −A · (B × C).3) In quantum mechanics. |p| |p| J(x. Similarly. the cross product of two vectors remains invariant and is correspondingly known as a pseudovector or an axial vector.

First. 11. P P (11. In this case. Classically. as we will see.1) would result in X −→ − X .11.6) holds.1 Parity in quantum mechanics.5) Deﬁning a symmetry in a quantum system. under space reﬂection an arbitrary quantum mechanical state transforms as . (11. let us assume that under the parity transformation. In this case. let us consider a one dimensional quantum mechanical system. Namely.6) from two equivalent points of view.1 Parity 417 so that that even classically parity operation deﬁnes a group with two elements (a group of order two). namely. let us assume that P represents the operator which implements the action of parity on the quantum mechanical states. As is standard in quantum mechanics. For a long time it was believed that the microscopic systems are also invariant under parity. but not the operators such that (11. The only diﬃculty we may apprehend is in the case of spinor states since they are double valued. But a systematic analysis. but we know now that there exist processes in nature which do not respect parity.6) where we have used the fact that classically vectors change sign under space reﬂection (see (11. we can analyze the parity transformation in (11. For simplicity. of course.1. simply corresponds to asking whether this group can be carried over to quantum mechanics without any obstruction. we know that dynamical laws of physics (such as Newton’s equation) do not depend on the handedness of the coordinate frame. the parity transformation (11. quantum mechanical states change. ½ and P with P 2 = ½.2)) and that expectation values of quantum operators behave like classical objects (Ehrenfest theorem). P −→ − P . the classical laws of physics are invariant under parity or space inversion. shows that even that does not present any diﬃculty. In other words.

8) Note that parity inverts space coordinates and. or. acting on the coordinate basis P must lead to |x → |xP = P|x = | − x . (11. (11.11) In other words.7) ψ P |X|ψ P = ψ|P † XP|ψ = − ψ|X|ψ . (11. Since P has real eigenvalues we conclude that the parity operator is Hermitian and thus we have . or. Furthermore.13) Namely. such that ψ|X|ψ ψ|P |ψ → → P P (11.10) which leads to P † P = ½. the parity operator is unitary. P is an idempotent operator and the only eigenvalues of P are ±1.9) δ(x − y) = x|P † P|y . (11. therefore. ψ P |P |ψ P = ψ|P † P P|ψ = − ψ|P |ψ .12) P 2 |x which shows that P2 = ½. = P| − x = |x .418 11 Discrete symmetries |ψ → |ψ P = P|ψ . (11.9) that xP |y P = x|P † P|y . It is now obvious from (11. −x| − y = x|P † P|y . since P|x = |−x . P (11.

1 Parity 419 P2 = P † = P = P −1 .11. then P|ψ = ±|ψ .7) that P |ψ → |ψ P = P|ψ = P = = P dx ψ(x)|x = dx ψ(x)| − x . or. ½.17) Let us note further that since the eigenvalues of P are ±1. we note from (11.16) Namely.19) Namely.15) so that we obtain x|ψ P = ψ P (x) = ψ(−x). ψ(x) −→ ψ P (x) = ψ(−x) = ±ψ(x). for such a state |ψ −→ |ψ P = P|ψ = ±|ψ .18) (11. an eigenstate of parity has an associated wave function which is either even or odd depending on the eigenvalue of the parity . if |ψ is an eigenstate of parity. P (11. (11. (11. dx |x x|ψ dx ψ(x)P|x (11. In other words.14) To see how a scalar (not a multi-component) wave function transforms under parity. under a parity transformation the wave function transforms as ψ(x) −→ ψ P (x) = ψ(−x). P P (11.

does not have to be an eigenstate of the parity operator and.22) where we have used (11. or. the corresponding wave function need not have any such symmetry behavior. P ) −→ OP (X. Similarly. ψ|P P |ψ = − ψ|P |ψ . (11. P )P = O(−X. we can obtain that [P. P ) = P † O(X. or. namely.24) . P ]+ = 0. [P. (11. In the other point of view.21) It follows from this that P X −→ X P = P † XP = −X. we may assume that the states do not transform under the parity transformation. however.420 11 Discrete symmetries operator for that state. such that (11. therefore. X]+ = 0. 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. −P ). (11.14). O(X. one can show that for any operator in this simple one dimensional theory.20) ψ|X|ψ ψ|P |ψ → → P ψ|X P |ψ = − ψ|X|ψ .23) In general. A general state of the system. P P (11. P = O P † XP. P † P P (11. PX + XP = 0.6) holds.

namely. (11. H(X. indeed. and it follows from (11. Therefore. 2m 2 P H(X.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. we see that if parity is a symmetry of a quantum mechanical system. P and H can be simultaneously diagonalized and the eigenstates of the Hamiltonian would carry speciﬁc parity quantum numbers. the eigenstates of the Hamiltonian would naturally decompose into even and odd states. P )] = 0. [P.27) that [P. P )P = H(−X. (11. in such a case.26) that this is. the solution of the time dependent Schr¨dinger equation o can be written as |ψ(t) = e−iHt |ψ(0) = U (t)|ψ(0) . H] = 0.11. P ) = (11.29) (11.25) As a result. P ) −→ P † H(X. It is obvious from simple systems like the one dimensional harmonic oscillator 1 P2 + mω 2 X 2 = H(−X. U (t)] = 0. (11. H(X. If parity is a symmetry of the theory. (Recall that the Hermite polynomials.) If the Hamiltonian H of a quantum mechanical system is time independent. Hn (x).27) where U (t) is know as the time evolution operator. −P ) = H(X. what happens. P ). or. then as we have seen [P. −P ). which are eigenfunctions of the Hamiltonian are even or odd functions of x depending on the value of the index n.28) .

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

so that it truly gives a representation of the parity operator. t) = ηψ Sα ψβ (−x.1 Parity 423 = = − = −X dxdy (−xδ(x + y))| − y x| dx x|x x| = − dx X|x x| dx |x x| = −X. t). under a parity transformation. we are led to the requirement that . P (11. 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| (11.11. from the fact that two space inversions are equivalent to leaving the object unchanged. but the components of the object may mix with one another. We have already seen this in the case of Lorentz transformations (see (3.35) for transformation of wavefunction). not only does x → −x. Let us now go over to relativistic quantum systems.14) are more important. As we have seen. Furthermore. On the other hand. such systems can be described consistently only in the language of quantized ﬁelds which are. we need to generalize our discussion of parity transformation to multicomponent objects. The important thing to note here is that such a representation is always nonlocal and not very useful from a practical point of view.33) β where the matrix Sα can. Namely. mix up the diﬀerent components of the object. Therefore. multi-component operators (recall the Dirac ﬁeld or the Maxwell ﬁeld). in general. the properties of the operator P such as in (11. t) −→ ψα (x. We note that under a parity transformation a multi-component object would transform as P β ψα (x.32) d |x dx = −P. in principle.

t) −→ φP (x. t) = ηφ φ(−x. t). (Remember that parity and the momentum operators do not commute and. t). (11. if φ(x. (11.39) . we conclude that in this case. so is φ(−x.) Thus we see that ηψ really measures the intrinsic behavior (and not the space part) of the wave function of a particle under space inversion. therefore. φ= (11. t) = φ(−x. ∂t2 or. then φP (x. the equation is clearly invariant under space inversion.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.38) P (11. t).2 Spin zero ﬁeld.34) and (11.1. In either of the cases. Consequently.37) If ηφ = 1 or the intrinsic parity of the spin zero ﬁeld is positive. (11. t) is a solution of the Klein-Gordon equation. From our general discussion in (11. t) = P † ψα (x.35). t) −→ ψα (x. t). t)P = ηψ Sα ψβ (−x. φ(x.34) In the language of quantized ﬁelds. (11.424 11 Discrete symmetries 2 ηψ S 2 = ½.36) depending on whether the ﬁeld is massless or massive. cannot have simultaneous eigenstates unless the eigenvalue of momentum vanishes. with ηφ = ±1. ( + m2 )φ(x) = 0. Let us begin with the spin zero ﬁeld which is described by the Klein-Gordon equation ∂2 − ∇2 φ(x) = 0. 11. the above relation represents an operator relation P β ψα (x.

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

P P → S + S.48) S → P + P. P → P + S. It now follows that in a parity conserving theory. (11.49) will be forbidden.45) If parity is a symmetry of the theory.426 11 Discrete symmetries ηTOT = ηφ (−1)ℓ . Let us further note that for a complex spin zero ﬁeld. (11. we must have ηA = ηB ηC (−1)ℓ . if we have .46) (11. → P + P. (11. one spin zero particle can decay into two spin zero particles in an s-state (ℓ = 0) only through the channels S → S + S. whereas processes such as (ℓ = 0) S → P + S. B and C respectively and ℓ is the orbital angular momentum of the B-C system.47) where ηA . ηB and ηC are the intrinsic parities of the particles A. the total parity quantum number must be conserved in a physical process and this leads to the fact that for a decay (in the rest frame of A) of a spin zero particle into two spin zero particles A → B + C. (11.

the intrinsic parity of the ﬁeld φ† is the same as that of the ﬁeld φ. t) . t) = −J(x. t)∂ µ φP (x. t) = i φ† (x. In quantum ﬁeld theory φ and φ† are associated with particles and antiparticles. Let us next look at the electromagnetic current associated with a charged Klein-Gordon system. (11. t). the current density has the explicit form J µ (x. t) − (∂ µ (φP )† (x. t)∂ µ φ(x.3. t) = i|ηφ |2 φ† (−x. t) = ηφ φ(−x. t))φ(x. t) . (11. ∗ (φP )† (x. t). since ηφ is assumed to take only real values ±1 (see (11. t) φ(x. t) = J 0 (−x. in fact. t) = ηφ φ† (−x.50) On the other hand. J P 0 (x.53) P which is the correct transformation for the current four vector as we have seen in (11. (11. t) − (∂ µ φ† (−x. (φ destroys a particle or creates an antiparticle whereas φ† destroys an antiparticle or creates a particle. particles and antiparticles would have the same intrinsic parity. t).2 and 7. t))φP (x.29). This result is. t) = i (φP )† (x.2).(11. .52) which leads explicitly to (recall that |ηφ |2 = 1) JP (x. quite general for the bosons and says that for bosons.1 Parity 427 φP (x.51) Under a parity transformation J µ (x.11. As we have seen in (7. t)∂ µ φ(−x.) Thus we conclude that a charged spin zero particle and its antiparticle have the same intrinsic parity. t). see discussion in sections 7.38)). t) −→ J P µ (x. t) − ∂ µ φ† (x.

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

60) we note that under a parity transformation Hem = → = = = = P d3 x J µ (x. t)Aµ (−x. t)AP 0 (x. t) − (−J(x. under a parity transformation . t)) d3 x J µ (−x. t)) · (−A(x. the photon is a vector particle. t) − JP (x. parity (quantum number) must be conserved in electromagnetic processes. 11. t) µ d3 x J P 0 (x. t).1 Parity 429 Note that a physical photon has only transverse degrees of freedom (namely. we see that AP (x. t) = −A(−x. t) d3 x J P µ (x. t)Aµ (x. (11. A0 is nondynamical). (11. (11. In other words. t)Aµ (x.33) or (11.11. namely A.35). This shows that the electromagnetic interaction is invariant under parity and. t) · AP (x.59) from which we conclude that the intrinsic parity of the photon ﬁeld (or the one photon state) is negative. t)A0 (−x. t)AP (x. t)Aµ (x.1.61) where in the last step we have let x → −x under the integral.4 Dirac ﬁeld. t). From the parity transformation of the dynamical components. therefore. Let us recall that the Dirac wave function or the Dirac ﬁeld is described by a four component object. Thus according to our general discussion in (11. t) d3 x J 0 (−x. t) = Hem . t) d3 x J µ (x. From the form of the electromagnetic interaction Hamiltonian Hem = d3 x J µ (x.

t).34)) 2 ηψ S 2 = ½. (11.44)) J µ (x. 2. t)γ 0 S † γ 0 . t)γ 0 S † γ 0 γ µ Sψ(−x. P (11.65) which will satisfy the above relation. t)γ 0 = ηψ ψ † (−x. With this choice (which also ∗ implies ηψ = ηψ ) we note that ∗ ψ (x. with (see (11. t) = (ψ P )† (x. t) = ηψ Sψ(−x. t). and under a parity transformation this would transform as J µ (x. t) = ψ(−x. (11. we can always choose ηψ = ±1. t)γ 0 S † γ 0 γ µ Sψ(−x. (11. t)γ µ ψ P (x. t)γ µ ψ(x. (11. t)γ 0 γ 0 S † γ 0 P = ηψ ψ(−x. t) = ηψ Sα ψβ (−x. see (8.64) Since we have assumed the parity eigenvalues to be ±1.62) In matrix notation.430 11 Discrete symmetries P β ψα (x. t) P P (11. 3. t).68) . t) = ψ (x. t) −→ J P µ (x. t)S † γ 0 ∗ = ηψ ψ † (−x. t) −→ ψα (x.63) (11. 4. we can write the transformation as ψ P (x. α = 1. t) = ψ(x. S 2 = ½. t).67) 2 = ηψ ψ(−x.66) Let us recall that associated with the Dirac system is a conserved current (probability current or electromagnetic current.

S = γ 0. t) = ηψ ψ(−x. t).73) (11. t)γ 0 S † Sψ(−x. t). t) = J 0 (−x.68) and (11. (11.70) so that using (11. t). with ηψ = ±1. t) = ηψ γ 0 ψ(−x. if we deﬁne . t) = ψ(−x. S † S = ½. comparing (11. we know the transformation properties of the current four vector under space inversion (see (11.11.71) Similarly. t)γ 0 ψ(−x. t) = −J i (−x. P P P (11.72) Thus we determine the transformation of the Dirac ﬁeld (wave function) under a space reﬂection to be ψ(x. then the parity transformed function ψ P (x. t)γ 0 .1 Parity 431 2 where we have used the fact that ηψ = 1. J P 0 (x. t). t) also satisﬁes the Dirac equation in the new coordinates.69) for µ = 0. t)γ 0 S † γ 0 γ i Sψ(−x. t) is a solution of the Dirac equation. t) −→ ψ (x.65) we have S = S † = S −1 . t).74) It is now quite easy to see that if ψ(x. (11. On the other hand.69) for µ = i. t) = −ψ(−x. t) −→ ψ P (x. Namely. ψ(x. t)γ i ψ(−x. (11. namely. we obtain ψ(−x.68) and (11. (11. or. we obtain ψ(−x.53)). or.69) Comparing (11. J P i (x.

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. −x). in fact. (11.75) 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) = 0.432 11 Discrete symmetries y µ = (t. ∂xµ (11. the two component Weyl fermions violate P.78) where ‘a’ and ‘b’ are scalar functions of momentum and ψpositronium represents the wave function for the positronium. To determine the relative parity between Dirac particles and antiparticles. (11.9). the scalar transition amplitude will have the general form Mi→f = (a ǫ1 · ǫ2 + b k · (ǫ1 × ǫ2 )) ψpositronium .6. Since the electromagnetic interactions conserve parity. as we have discussed in section 3. the two photons move with equal and opposite momentum. The polarization planes of the photons in this decay can. Let ǫ1 and ǫ2 denote the polarization vectors for the two photons while k represents their relative momentum.8 eV) e− + e+ → γ + γ. This can also be seen by noting that parity is a symmetry of the free Dirac Lagrangian density (8. be measured and it . space inversion is a symmetry of the Dirac equation.76) In other words.77) In the rest frame of the positronium. However. then ∂ − m ψ P (y) = ∂y µ (11.

the transition amplitude for this decay takes the form Mi→f = (b k · (ǫ1 × ǫ2 )) ψpositronium . Let us recall from our earlier discussions that antiparticles are related to negative energy states. the electron and the positron must have relative negative intrinsic parity.80) k · (ǫ1 × ǫ2 ) −→ −k · (ǫ1 × ǫ2 ) .79) Consequently.49)) . the relative intrinsic parity between Dirac particles and antiparticles is negative. Namely.1 Parity 433 is experimentally observed that the two are perpendicular.11.82) and for this state to represent a pseudoscalar state. Since the transition amplitude is a scalar under parity while P (11. we must have ηTOT = ηe− ηe+ = −1. Let us choose two representative solutions (of positive and negative energy) of the Dirac equation in the rest frame (see (2. in fact. for this decay we have ǫ1 · ǫ2 = 0. This is.83) In other words. (11. the total parity of the state is given by ηTOT = ηe− ηe+ . (11.81) we conclude that the positronium state must be a pseudoscalar under parity. However. in the rest frame of the positronium. namely. On the other hand. a very general result for fermions. it can also be seen in an intuitive manner from the structure of the spinor functions as follows. (11. This result can be more directly seen when we study charge conjugation in the next section. (11.

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

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

∂t B = ∇ × A. ∇×E = − ∇×B = ∂B . since ∂A − ∇A0 . ∂t ∂E + J. 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. ∇ · B = 0.94) This discrete transformation can be thought of as charge reﬂection or charge conjugation (simply because reﬂecting the sign of the charge generates the transformations in (11.436 11 Discrete symmetries ψγ5 ψφ. E → −E. if we know the interaction Lagrangian or Hamiltonian density for a given process. J → −J.95) . B → −B. let us go back to Maxwell’s equations in the presence of charges and currents ∇ · E = ρ.2 Charge conjugation To understand the discrete symmetry known as charge conjugation.93) We note that this set of equations is invariant under the discrete transformations ρ → −ρ. (11. 11. E = − classically we see that charge conjugation leads to (11.92) Conversely. ∂t (11. (11.94)) and is denoted by C.

therefore. charge conjugation really interchanges every distinct quantum number between particles and antiparticles. we can generalize the classical result (11.98) and we say that the charge conjugation parity (or simply the charge parity) of the photon as well as the current is −1. a real Klein-Gordon theory describes charge neutral spin zero particles and. charge conjugation is trivial in this theory as we will discuss shortly. µ C Jµ (x) −→ Jµ (x) = ηJ Jµ (x) = −Jµ (x). in a quantum theory.2 Charge conjugation 437 Aµ −→ −Aµ . Let us also note here that like parity. two charge conjugation operations (reﬂecting the charge twice) would bring back any variable to itself.29) as . charge conjugation is sometimes also referred to as particleantiparticle conjugation.96) It is obvious that the electromagnetic interaction Hamiltonian Hem = d3 x J µ Aµ . Let us also note that in a quantum theory.97) is invariant under this transformation. Let us now analyze the complex Klein-Gordon ﬁeld theory which describes charged spin zero particles. In the quantum theory. reﬂecting charges in a quantum theory is equivalent to saying that charge conjugation really interchanges particles and antiparticles.11. opposite charges are assigned to particles and antiparticles. (As we have seen. therefore. C (11. 11.) We have seen that for such a theory we can deﬁne an electromagnetic current (7. C (11.96) by saying that C Aµ (x) −→ AC (x) = ηA Aµ (x) = −Aµ (x). In other words.1 Spin zero ﬁeld. (11. For this reason. Note that the transformation of charge conjugation does not change the space-time coordinates and. C Jµ −→ −Jµ . it is not a space-time transformation. Therefore.2.

Therefore. We have already discussed (see sections 7.100) with |ηφ |2 = 1. (11.101) C Thus the ﬁeld transformations in (11.2 and 7.102) .100) would deﬁne the appropriate charge conjugation transformation for the current. ∗ φ† (x) −→ (φC )† (x) = ηφ φ(x). ∗ φ† (x) −→ ηφ φ(x). under a charge conjugation. C (11. then C φ(x) −→ φC (x) = ηφ φ† (x).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). It follows now that under such a transformation. this further supports the fact that (11. C C C Aµ (x) −→ −Aµ (x). 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).100) is the appropriate charge conjugation transformation for complex scalar ﬁeld. we have (|ηφ |2 = 1) φ(x) −→ ηφ φ† (x). (11. we continue to use “†” instead of “∗” so that the passage to quantum theory will be straightforward). (11. As a result.438 11 Discrete symmetries J µ = i φ† ∂ µ φ − (∂ µ φ† )φ .99) We note that if we treat φ(x) as a classical function (even though the functions are classical.

11. The problem. the Lagrangian density of charged. it would appear as if the ﬁeld transformations (11.100) 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 . Dµ φ(x) = (∂µ + ieAµ )φ(x) −→ ηφ (∂µ − ieAµ )φ† = ηφ (Dµ φ)† .104) is invariant under charge conjugation. All of the above discussion carries over to the quantum theory if we are careful about the operator ordering of φ and φ† .100) do not lead to the correct transformation properties for the current operator. then under the transformation (11. For example. spin zero particles interacting with photons described by 1 L = (Dµ φ)† (Dµ φ) − m2 φ† φ − Fµν F µν . can be traced to the fact that in going from a classical theory to the quantum theory. As is known.2 Charge conjugation 439 Fµν (x) −→ −Fµν (x). we have to choose an operator ordering which we have not taken care of in the above discussion. 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 . of course.103) Therefore.105) C since the operators φ and φ† do not commute with ∂µ φ† and ∂µ φ respectively.99). 4 (11. As a result. we note that if we deﬁne the electromagnetic current in the quantum theory as in (11. C C (11. (11. C ∗ ∗ (Dµ φ(x))† = (∂µ − ieAµ )φ† (x) −→ ηφ (∂µ + ieAµ )φ = ηφ Dµ φ.

Note that we can also write the symmetrized current as 1 µ (J (x) − J Cµ (x)) . (11. If we symmetrize the products in the deﬁnition of the current. As we have discussed earlier. In such a case. under a charge conjugation (11.106) leads to (this should be compared with the normal ordered charge deﬁned in section 7. (11. if a spin-zero ﬁeld theory describes charge neutral particles which are their own antiparticles.109) which brings out the particle-antiparticle symmetry in a more direct way.110) . 2 µ Jsym (x) = (11. it is described by a Hermitian ﬁeld operator φ† (x) = φ(x). we obtain i φ† ∂ µ φ + (∂ µ φ)φ† − (∂ µ φ† )φ − φ∂ µ φ† . (11. But this has the interesting consequence that while (remember the Dirac vacuum picture) 0|J µ (x)|0 −→ ∞. will change sign) under charge conjugation.107) µ the eﬀect of the inﬁnite constant in Jsym in (11.108) This is esthetically more pleasing and treats the vacuum in a more symmetrical way.106) 2 µ Jsym (x) = and it is easy to check that this current will transform properly (namely. 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).3) µ 0|Jsym (x)|0 = 0.440 11 Discrete symmetries normal ordering the product).

we expect 2 ηφ = 1.114) Such ﬁelds or particles are known as self-charge-conjugate and simply describe particles that are their own antiparticles.2 Charge conjugation 441 φ(x) −→ φC (x) = ηφ φ† (x) = ηφ φ(x). C (11. the total charge parity of the ﬁnal state is (+1). If charge parity is to be conserved in this process (this corresponds to an electromagnetic process since photons are involved).113) φ(x) −→ φC (x) = φ(x).111) From the fact that two charge conjugation operations should return the ﬁeld to itself. (11. ηφ = ±1. we conclude that the π 0 meson must also have charge parity (+1) and similar arguments hold for other mesons.112) In this case. so that in these cases (11.115) since the charge parity of the photon is (−1).11. As an example. one can show that for all Hermitian spin zero ﬁelds that we know of (scalar or pseudoscalar) ηφ = 1. (11. let us note that in the decay π 0 → 2γ. or. 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. C (11. .

119) that the two equations are very similar in structure.119) where we have identiﬁed the charge conjugate function ψ C as describing the positron. (iγ µ (∂µ − ieAµ ) − m) Cψ = 0. To see the relation between ψ C (x) and ψ(x).120) (11. (11.118) we would have C i(γ µ )T (∂µ − ieAµ ) + m ψ = 0. or. In fact.2 Dirac ﬁeld.118) and (11.2.116) +m = 0. we obtain i(γ µ )T (∂µ − ieAµ ) + m ψ = 0.118) Let us recall that a positron which is the antiparticle of the electron is also a Dirac particle with the same mass but carries an opposite electric charge. T (11. (11. iC(γ µ )T C −1 (∂µ − ieAµ ) + m Cψ = 0. or. The adjoint equation is easily obtained to correspond to ← ψ(x) iγ µ ∂µ −ieAµ (11. interacting with an electromagnetic ﬁeld. let us note from (11. it would satisfy the equation (iγ µ (∂µ − ieAµ ) − m) ψ C (x) = 0. from (11.442 11 Discrete symmetries 11. the equation of motion takes the form (iγ µ (∂µ + ieAµ ) − m) ψ(x) = 0.121) .117). Thus. Let us next analyze a system of Dirac fermions in- teracting with an electromagnetic ﬁeld. T T T (11.117) Taking the transpose of the adjoint equation (11. If the fermions are minimally coupled to the photons. we note that if we can ﬁnd a matrix C satisfying C(γ µ )T C −1 = −γ µ .

92)).123) That such a matrix exists can be deduced from the fact that −(γ µ )T also satisﬁes the Cliﬀord algebra. we obtain γ µ C = −C(γ µ )T = −C T (γ µ )T = − (γ µ C)T .2 Charge conjugation 443 which can be compared with (11. In 4-dimension C T = −C.119). Therefore we conclude that if we can ﬁnd a matrix C with the property C(γ µ )T C −1 = −γ µ . T (11.127) (11.126) (11. it follows that there must exist a similarity transformation which relates −(γ µ )T and γ µ . The important symmetry property that C must have is that it should be antisymmetric.125) .11. namely. (This property depends on the dimensionality of space-time.) To see this. Similarly. (11. In this case. we can also show that σ µν C = − (σ µν C)T .128) (11. we can identify (with |ηψ |2 = 1) ψ C (x) = ηψ Cψ . from C −1 γ µ C = −(γ µ )T .124) and hence from the generalized Pauli theorem (see (1. (11. C T = C. namely. C −1 γ µ C = −(γ µ )T . let us assume that instead C is symmetric. or.122) (11. −(γ µ )T . −(γ µ )T + = 2η µν ½.

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

we can obtain the antisymmetrized current as µ Janti (x) = (11.137) whereas 0|J µ (x)|0 −→ ∞. (11.11.136) µ = −Janti (x). (11. where we have used (11. this current transforms properly under charge conjugation.134) = 1 ψ α (x) (γ µ )αβ ψβ (x) − ψβ (x) (γ µ )αβ ψ α (x) 2 1 T ψ(x)γ µ ψ(x) − ψ T (x)(γ µ )T ψ (x) . (11.138) Thus this current treats the vacuum more symmetrically which can also be seen from the fact that we can write the current in the form .135) 2 It can now be easily checked that under charge conjugation the antisymmetrized current transforms as µ Janti (x) −→ C 1 C C ψ (x)γ µ ψ C (x) − (ψ C )T (x)(γ µ )T (ψ )T (x) 2 = = = 1 T |ηψ |2 −ψ T (x)C −1 γ µ Cψ (x) + ψ C T (γ µ )T (C −1 )T ψ 2 1 T T ψ T (x)(γ µ )T ψ (x) + ψ(x) C −1 γ µ C ψ(x) 2 1 T T ψ T (x)γ µ ψ (x) − ψ(x)γ µ ψ(x) 2 (11. 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 = 0.2 Charge conjugation 445 J µ (x) = ψ(x)γ µ ψ(x). Therefore.122) in the intermediate steps.

We have already seen in (11.122) that the charge conjugation matrix C satisﬁes C −1 γ µ C = −(γ µ )T . 2 (11.142) 2 Therefore. if we deﬁne properly antisymmetrized Dirac bilinear combinations as = − (γ µ γ5 )T = (γ5 γ µ )T .139) Let us next calculate how the Dirac bilinears transform under charge conjugation. = i γ 0γ 1γ 2γ 3 (11. 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.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 . (11. .140) − (γ 1 )T T − (γ 2 )T T − (γ 3 )T T = γ5 . we obtain T C −1 γ5 γ µ C = C −1 γ5 CC −1 γ µ C = γ5 − (γ µ )T (11.446 11 Discrete symmetries µ Janti (x) = 1 µ (J (x) − J Cµ (x)) .

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 . for α = S. for (11.) We have already seen that ψγ µ ψ −→ − ψγ µ ψ C anti anti . (11. (11.146) 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. (11. ψψ − ψ T ψ (11. . (11.148) . .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 1.145) Similarly we can also show that ψγ5 ψ ψγ5 γ µ ψ ψσ µν ψ −→ −→ C C C anti anti anti ψγ5 ψ anti .144) 2 (If we think for a moment. (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). T. α = V. P. ǫα = −1. A.

t)γ 0 . t) = ηψ γ 0 ψ(−x. P ψ(x. let us note that since the charge conjugate wave function is deﬁned to be C ψ C = ηψ Cψ . t). t). t) P C −ηψ ηψ γ 0 C(γ 0 )T ψ ∗ (−x. t) C P −ηψ γ 0 ηψ Cψ (−x. t) = ηψ Cψ (x. t) −→ ψ P (x. P P P (11. t) −→ (ψ C )P (x.151) Here we have used the fact that γ 0 is Hermitian in our metric. we have already seen this from the analysis of the positronium decay in (11. t) = ηψ C(ψ )T (x.149) Since the antiparticle is described by C ψ C (x. t) P P = = = = = C P ηψ ηψ C(ψ(−x. t) P −ηψ γ 0 ψ C (−x. the antiparticle will transform as C ψ C (x. T (11.448 11 Discrete symmetries Let us next recall from (11. t) = ηψ ψ(−x. t) −→ ψ (x. t). T T (11.73) that under parity a Dirac particle transforms as P ψ(x.83). Thus P if ηψ represents the intrinsic parity of a Dirac particle. 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 = −1.150) under parity. t)γ 0 )T C P ηψ ηψ Cγ 0 C −1 C(γ 0 )∗ ψ ∗ (−x. Next.152) Of course. T we obtain . (11. but this derivation is simpler and more general.

the antiparticle of a particle with a given handedness has the opposite handedness. We have. we have C γ5 ψ C = −ηψ Cψ = −ψ C .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) .11.156) T T C = ηψ Cψ = ψ C . then it follows that C γ5 ψ C = −ηψ C − ψ (11. 11. / (11.155) On the other hand.157) In other words.6 and have used this in the construction of the standard model in section 14. (11. already seen this in the study of neutrinos in section 3. we see that if γ5 ψ = ψ.3 Majorana fermions. (p − m)ψ C = 0. Let us recall that if a free Dirac particle satisﬁes the equation (p − m)ψ = 0.119) we see that the antiparticle must also satisfy the same free equation. namely.153) As a result.154) T (11. if γ5 ψ = −ψ. of course.159) .158) then from (11.2. / where (11.3. (11. T (11.

in fact.450 ψ C = ηψ Cψ .165) so that the Majorana fermions are charge neutral and hence cannot have any electromagnetic interaction. These are charge neutral particles described by real quantum ﬁelds φ† = φ. Let us note from the deﬁnition of the antisymmetrized current in (11.135) µ Janti (x) = T 1 T ψ(x)γ µ ψ(x) − ψ T (x)γ µ ψ (x) 2 1 C ψ(x)γ µ ψ(x) − ψ (x)γ µ ψ C (x) .123) we see that a Dirac particle which is its own antiparticle must satisfy ψ = ψ C = ηψ Cψ . 2 = (11.161) We can now ask the corresponding question in the case of the Dirac particles. whether there are also fermionic particles that are their own antiparticles and if so how does one describe them.162)). / / (11. (11. . and would also satisfy the free Dirac equation (p − m)ψ = (p − m)ψ C = 0. They can be massive and will have only two independent degrees of freedom (because of the constraint in (11.162) Such particles are known as self charge conjugate fermions or Majorana fermions.160) We have also seen that there are bosons in nature which are their own antiparticles. quite obvious. (11. The answer to this question is. T 11 Discrete symmetries (11.163) T (11. From the deﬁnition of charge conjugation in (11.164) that for Majorana fermions (ψ = ψ C ) µ Janti (x) = 0. namely.

Let us also note here that since both the Weyl and the Majorana fermions have only two degrees of freedom. In particular for d = 2 mod 8. we can have Majorana-Weyl spinors. then χC = (ψL )C + ψL = χ. 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. this would require P P ηψ = −(ηψ )∗ . However. the mass term would trivially vanish since (11. 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. (This is. (11. In fact.168) so that we can express a Majorana fermion in terms of Weyl fermions.166) which can be satisﬁed only if the intrinsic parity of a Majorana fermion is imaginary. let us note that if we try to write a Lagrangian density for a massive Weyl fermion (say left-handed).169) . we note that if we deﬁne χ = ψL + (ψL )C . Since a Majorana fermion is its own antiparticle.) We have also seen that if the intrinsic parity of a Dirac particle P P is ηψ . not true if the space time dimensionality is not four.167) (11.2 Charge conjugation 451 Secondly. as we have seen a particle with a deﬁnite handedness will lead to an antiparticle with the opposite handedness. the naive mass term would have the form / L = iψ L ∂ ψL − mψ L ψL . however. we can ﬁnd a simple relation between them. (11. Thus we cannot talk of a Majorana particle with a given helicity. To answer this. it can in principle be complex).11. this implies that it cannot have a well deﬁned handedness.

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

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

= |π + p . however. the charge operator does not commute with the charge conjugation operator and. ℓ = (−1)ℓ |π − π + . Note that C|π − π + = |π + π − . they cannot have simultaneous eigenstates unless the charge of the state vanishes. that while all eigenstates of C must be charge neutral. for the positronium with orbital angular momentum and spin values ℓ.179) The ﬁnal state simply corresponds to the initial state with the two particles interchanged. = |K .180) so that such a state would be an eigenstate of charge conjugation with eigenvalue (−1)ℓ . of course.454 11 Discrete symmetries ˆ ˆ C Q = −QC. ℓ . Thus if the state has well deﬁned angular momentum quantum number. 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. (11.178) which clariﬁes the point. (11. Q + = 0.177) In other words. Similarly. Let us note. s respectively. ℓ = |π + π − . or. We have. (11. 0 (11. On the other hand. It is easy to see that C|n C|π − p C|K 0 = |n . consequently. not all charge neutral states would be eigenstates of charge conjugation operator. we would have C|π − π + . . ˆ C. a state such as |π − π + can be an eigenstate of C.

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

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

Let us note here without going into details that while charge conjugation is a good symmetry for both strong and electromagnetic interactions.2 Charge conjugation 457 π − p|U |π − p = π − p|C † U C|π − p = π + p|U |π + p . The simplest way to see this is to note that the weak current (see (14. We can again check with the charged weak current that + JµW K C −→ −→ P g C C √ (ηe )∗ ην ν eL γµ (1 + γ5 ) eL 2 2 g C C √ (ηe )∗ ην ν PL γµ (1 + γ5 ) eP e L 2 2 . Thus we see that P and C (P for parity) are conserved in strong and electromagnetic interactions but are violated in weak interactions.191) In other words. On the other hand the product operation CP appears to be a good symmetry in most weak processes. it is violated in weak processes just like parity. (11. (11.192) 2 2 where ηJ denotes a phase. 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. Namely.11.100)) 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 .

T t −→ −t. ∂t (11. 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 . namely. ∇ × E(x) = − ∇ × B(x) = ∂B . T (11.193) 2 2 where ηTOT denotes the total phase under the transformation.196) are also form invariant if we deﬁne .194) W 2 2 This has well deﬁned transformation properties under CP transformations and leads to CP invariance. (11.3 Time reversal Classically. ∇ · B(x) = 0. 11. time reversal or time reﬂection corresponds to the space time transformation where we reverse only the sign of time. x −→ x. (11. Maxwell’s equations in the presence of sources ∇ · E(x) = ρ(x).195) We note that the Newton’s equation is invariant under time reversal since it is second order in the time derivative.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. ∂t ∂E(x) + J.

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

−i ∂|ψ ∂t = T H|ψ .202) would lead to iT or.200) If we assume that the time independent Hamiltonian is also invariant under time inversion.203) ∂ (T |ψ ) = H(T |ψ ).460 11 Discrete symmetries |ψ −→ |ψ T T = T |ψ . By deﬁnition. [T . ∂t Thus we see that in such a case |ψ and T |ψ satisfy diﬀerent equations and hence this operation cannot deﬁne a symmetry of the time dependent Schr¨dinger equation. the time dependent Schr¨dinger equation o i ∂|ψ = H|ψ . namely. an antilinear operator T satisﬁes . 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. Let us for the moment give up the assumption that the operator T is linear. let us assume that T is an antilinear (antiunitary) operator. (11. o However. ∂t (11. (11. In fact. H] = 0.201) then it follows that under a time inversion. (11.

o Let us next assume that we are looking at a process which goes from the state |ψi to the state |ψf . φ|ψ −→ T φ|ψ T T |ψ −→ |ψ T T = T |ψ .) Under such an antilinear (antiunitary) transformation. ∂t Therefore. = HT |ψ . i ∂ (T |ψ ) = H(T |ψ ).206) . 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. if the Hamiltonian is symmetric under time inversion. but we will not need this for our discussions. i −i = H|ψ . with such a deﬁnition of T .11.205) or. (11.204) where A and B are assumed to be Hermitian operators and ηA . Thus the quantity we are interested in is the transition amplitude ψf |ψi . (We can also show that T † = T −1 . ∂|ψ ∂t ∂T |ψ ∂(−t) = T H|ψ . the dynamical equation would transform as ∂|ψ ∂t i would lead to T or. T AT † = AT = ηA A. (11. (11.3 Time reversal 461 T (α|ψ1 + β|ψ2 ) = α∗ T |ψ1 + β ∗ T |ψ2 . T iT † = −i. = φ|ψ ∗ = ψ|φ . T ABT † = AT B T = ηA ηB AB. ηB are phases.

the role of the initial and the ﬁnal states are interchanged and the transition amplitude of interest would be ψi |ψf . we know that x −→ x. −t).207) (11. T (11. under time inversion. .208) which is indeed consistent with the deﬁnition of time inversion transformations given in (11. T p −→ −p. Classically. let us note from (11. Pj ] .204) that ψ(x.211) T PT † = ηP P = −P.204) as well as (11. (11. It follows now that T [Xi . T ∗ (11. ηP = −1. t) = x|ψ(t) −→ ψ T (x. we expect ψf |ψi −→ ψi |ψf . Thus.211). In particular. on the other hand.204) (see.209) Let us derive the transformation properties of some of the operators under time inversion. (11.462 11 Discrete symmetries In the time reversed process.210) Thus from Ehrenfest’s theorem we would expect the corresponding operators to transform as T XT † = ηX X = X. T (11. It is reassuring to note that the canonical commutation relations are unchanged by the ηX = 1.212) where we have used (11. in particular the third relation). t) = x|ψ(−t) = ψ ∗ (x. Pj ] T † = T (Xi Pj − Pj Xi ) T † = Xi (−Pj ) − (−Pj )Xi = − [Xi .

m we have L2 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.204) that i → −i under the anti-linear time reversal transformation). (11. m . m = T L2 |ℓ. It follows now that for the angular momentum eigenstates satisfying L2 |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. (11. (11. Therefore. (11. m ). m with T = Cm |ℓ. under time inversion each component of the angular momentum operator would change sign which is consistent with the classical result.3 Time reversal 463 time inversion transformation (note from (11. m . m L3 T |ℓ. ℓ(ℓ + 1)(T |ℓ. −m . m ). (11.214) so that the L2 operator remains unchanged under time inversion.11.215) m|ℓ.216) = −T L3 |ℓ. m = − m(T |ℓ. m = 2 2 = = ℓ(ℓ + 1)|ℓ.213) Namely. we obtain T Li T † = ǫijk T Xj Pk T † = −ǫijk Xj Pk = −Li . m L3 |ℓ.217) . Since the angular momentum operators can be written as Li = ǫijk Xj Pk . it follows that T L2 T † = T Li Li T † = (−Li ) (−Li ) = Li Li = L2 . m = |ℓ. We see that the state T |ℓ.

where we have used the fact that the spherical harmonics for m < 0 are deﬁned with a phase +1. which follows because under time inversion (say for m > 0) (11. φ|ℓ.3. (11. m = T |ℓ. φ) = = T θ. φ) = = (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ.m (θ. −mj .219) Yℓ.218) (11. T (11. in general.222) 11. By convention.1 Spin zero ﬁeld and Maxwell’s theory.m (θ.221) It can be shown that the above relation holds true even for half integer angular momentum values. m = (i)2m |ℓ.223) . We have already seen that under time reversal Aµ (x. mj = T |j. we can write T |j. t) −→ AµT (x.m (cos θ)e−imφ 4π (ℓ + |m|)! (11. (11.464 11 Discrete symmetries |Cm |2 = 1. m (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. we can indeed write T |ℓ. Thus.−m (θ. −t). −m . φ).220) (−1)m Yℓ. mj = (i)2mj |j. the constant (phase) is chosen to be Cm = (−1)m = i2m . namely. t) = ηAµ Aµ (x.m (cos θ)eimφ 4π ℓ + |m|)! ∗ −→ Yℓ.

t) − ∂ν AT (x. −t) F µν (x.225) T T −→ ∂µ AT (x.227) ηJ 0 = 1. ηA = −1. that Fµν (x. t) = T ∂µ Aν (x. the current four vector transforms as J µ (x. of course. T Fij (x. t) − ∂ν Aµ (x. where T (11.224) It follows. −t). T T (11. −t)Aµ (x. t).226) Similarly. −t). invariant under time reversal as we have already seen and as can be checked directly. ηJ = −1.228) The Maxwell theory is. t) (11. t) = −Fij (x. t) = ηJ µ J µ (x. (11. −t). (11. ν µ so that T F0i (x. −t) 4 1 = d4 x − Fµν (x)F µν (x) − J µ (x)Aµ (x) 4 = S. therefore.11.3 Time reversal 465 where ηA0 = 1. t) −→ F0i (x − t) = F0i (x. t) −→ Fij (x. −t) − J µ (x. t) −→ J µT (x. (11. we have seen that under time reversal. 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.229) = d4 x S .

−t))φ(x. −t) ∂t2 (11. −t)∂ µ φ(x. −t)).230) where the matrix T mixes the components of the function and |ηψ |2 = 1. the Klein-Gordon equation is invariant under time inversion.233) T (11. we have φ(x) −→ φT (x) = ηφ φ∗ (x. (11.234) ∂2 − ∇2 + m2 φ(x) = 0. −t) − (∂ µ φ∗ (x. −t). ∂t2 ∂2 − ∇2 + m2 φ∗ (x) = 0.231) (11. −t)∂ µ φ∗ (x. ∂t2 (11. Note that since the complex spin zero ﬁeld satisﬁes + m2 φ(x) = + m2 φ∗ (x) = it is clear that in this case + m2 φT (x) = = = 0.232) That is. 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.466 11 Discrete symmetries In general. ∂2 − ∇2 + m2 φT (x) ∂t2 ∂2 − ∇2 + m2 ηφ φ∗ (x.235) . for a spin zero ﬁeld. −t). Furthermore. −t)) −→ i (φT ∗ (x)∂ µ φT (x) − (∂ µ φT ∗ (x))φT (x)) T = −i (φ∗ (x. T (11. −t) − (∂ µ φ(x. −t))φ∗ (x. for a multicomponent function or ﬁeld we expect the behavior under time reversal to be T β ∗ ψα (x) −→ ψα (x) = ηψ Tα ψβ (x. For example.

−t). or. t) = 0. From our general discussion (see (11. −t) = 0. (Here ∂ we have used the fact that J 0 (x. −t). −iγ 0∗ −iγ 0∗ iγ 0∗ ∂ ∂ − iγ i∗ i − m ψ ∗ (x.197).230)).236) J i (x. t) −→ ψ T (x. ∂t ∂x ∂ ∂ − iγ i∗ i − m ψ ∗ (x. we obtain J 0 (x.238) ∂ ∂ − iγ i∗ i − m ψ ∗ (x. we expect that under time reversal ψ(x. ∂(−t) ∂x (11. T (11.11. the correct behavior of the current four-vector under time inversion as we have already seen in (11.2 Dirac ﬁelds. or. ∂t ∂x . −t). −t) = 0.) Thus we conclude that the spin zero charged particles interacting with photons can be described by a theory which will be time reversal invariant.3 Time reversal 467 Therefore. −t))φ(x. −t) − ( φ∗ (x. −t) is deﬁned with − ∂t . −t))φ(x. leads to (upon complex conjugation) (−iγ µ∗ ∂µ − m) ψ ∗ (x) = 0. −t) ∂t ∂t J 0 (x. −t)∂ i φ(x. or. t) = ηψ T ψ ∗ (x. 11. t) −→ −i φ∗ (x. of course. −t) = T T ∂ ∂ φ(x. −t) − (∂ i φ∗ (x. (11. −t) = which is. −J i (x.237) To determine when this function will be a solution of the Dirac equation (so that time reversal can be a symmetry of the system) we note that the Dirac equation (iγ µ ∂µ − m) ψ(x) = 0. t) −→ −i φ∗ (x.239) (11.3.

242) would satisfy the same Dirac equation as ψ(x. t). γ ˜ (11. if ψ T (x.244) (11.245) . we should have (iγ µ ∂µ − m) ψ T (x) = 0. iγ 0 iγ 0 ∂ ∂ + iγ i i − m ψ T (x) = 0. and the matrices γ µ deﬁned as (˜ µ = γ µ† ) ˜ γ (11. ˜ also satisfy the Cliﬀord algebra {˜ µ .468 11 Discrete symmetries On the other hand.243) γ 0 = γ 0.241) ψ T (x) = ηψ T ψ ∗ (x. T γ i∗ T −1 = −γ i . −t). γ ν } = 2η µν ½. comparing the two equations we see that if we can ﬁnd a matrix T such that T γ 0∗ T −1 = γ 0 . γ ν∗ } = 2η µν ½. or.240) Therefore. then the time reversed wave function (11. t) were to be a solution of the Dirac equation. We note that the matrices γ µ∗ satisfy the Cliﬀord algebra {γ µ∗ . −t) = 0. or. (11. ˜ γ i = −γ i . ∂t ∂x (11. ∂t ∂x ∂ ∂ + iγ i i − m T ψ ∗ (x.

11. T = −γ i∗ = − γ i† = γ iT . Let us also recall from (11.3 Time reversal 469 Thus both γ µ∗ and γ µ must be related to γ µ through similarity trans˜ formations (by Pauli’s fundamental theorem). They must also be related to each other through a similarity transformation.249) (11. Thus we can identify T γ µ∗ T −1 = γ µ .251) In other words. (11.246) (11. CT −1 γ µ CT −1 −1 = −γ µ .250) CT −1 γ µ T C −1 = Cγ µT C −1 = −γ µ . Combining the two results we obtain (11.248) Thus we see that the inverse relation can be written as T −1 γ µ T = γ µT . (11. or. ˜ leading to T γ 0∗ T −1 = γ 0 . the matrix CT −1 anticommutes with all the γ µ ’s and hence must be a multiple of γ5 . (11.247) To determine the explicit form of T . T γ i∗ T −1 = −γ i . Choosing .122) that Cγ µT C −1 = −γ µ . let us invert these relations and write T T −1 γ 0 T T −1 γ i T = γ 0∗ = γ 0† = γ 0T .

Note that with this choice.255) Furthermore. as in the case of charge conjugation. the T matrix satisﬁes T = T † = T −1 . −t) = T ηψ T γ 0T ψ (x. and using the deﬁnition (11.256) so that the time inversion matrix in the spinor space is antisymmetric. t) = ηψ T ψ ∗ (x. −t)T −1 γ 0 . −t) = ηψ γ 0 T ψ (x. independent of any representation (the superscript denotes transposition) (T )T = −T.130) C = γ 0γ 2. we determine (in our representation) T = iγ 1 γ 3 .252) (11. t) −→ ψ T (x. (11. Therefore we see that the time inversion transformations T ψ(x. T T T ∗ ψ(x.254) (11. Let us next analyze the behavior of the current four-vector under time inversion .257) with (|ηψ |2 = 1) deﬁnes a symmetry of the Dirac equation. t) = ηψ ψ T (x. we can show that in four space-time dimensions.470 11 Discrete symmetries CT −1 = −γ5 = −iγ 0 γ 1 γ 2 γ 3 . −t).253) (11. t) −→ ψ (x. (11. (11.

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

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

then the transition amplitude can be represented as (11. t)σ 0i ψ(x.265) In other words. T T (11.3 Consequences of T invariance. −t). the tensor bilinears have precisely the same behavior under time reversal as the electromagnetic ﬁeld strength tensor F µν . We could have expected this from the deﬁnition (see (3. j. Let us recall that under time reversal.222)) T |α.267) where α stands for all the other quantum numbers of the state. −t). the chirality of a spinor does not change. Let us also note that if we are looking at a reaction of the form a + b −→ c + d.264) We can also show that ψ(x.268) . ψ(x. (11. |p| |p| |p| h= (11. 11. t) −→ −ψ(x.3 Time reversal 473 the helicity does not change. p. It is also easy to see now that under time reversal. −mj . namely. a state with given angular momentum transforms as (see (11. −p.11. (11. t) −→ ψ(x. t)σ ij ψ(x.121)) s·p s · p T (−s) · (−p) −→ = = h. j. −t)σ ij ψ(x. −t)σ 0i ψ(x. T = ηψ γ5 T ψ ∗ = ηψ T γ5 ψ ∗ † = ηψ T γ5 ψ ∗ γ5 ψ T = ηψ T (γ5 ψ)∗ . mj −→ (i)2mj |α.3.266) so that a left-handed spinor remains left-handed under time reversal and a right-handed spinor remains right-handed.

270) Since the time evolution operator is not Hermitian. namely.271) (11. experimentally this is seen to hold in many processes. if we are looking at the reciprocal reaction c + d −→ a + b. On the other hand. (11. and c + d −→ a + b.275) . (11. then the transition amplitude can be represented as i|S|f . (11. to obtain the spin of π-meson etc. namely. |i = |a. S † = S.269) where |i stands for the initial state. d .272) (11. |f = |c. b . for example.474 11 Discrete symmetries f |S|i . it does not follow a priori that | i|S|f |2 = | f |S|i |2 . |f for the ﬁnal state and S denotes the S-matrix or the time evolution matrix.274) (11.273) On the other hand. It is worth emphasizing here that the detailed balance does not say that the rates for the two reactions a + b −→ c + d. (11. This is known as the principle of detailed balance and has been used. where we are still using the same notation.

mf . therefore.276) where ρf denotes the density of states for the ﬁnal state which may not be the same for the two processes. (11. h (11. theoretically. pi . the principle of detailed balance will hold. Let us next try to understand. = |β. In fact. Let us assume that we are dealing with a system where the interaction Hamiltonian can be treated as a perturbation.3 Time reversal 475 have to be the same. Let us next assume that we are dealing with a system whose Hamiltonian is invariant under time reversal. (11. then we have | f |S|i | = 2π| f |Hint |i |δ (Ei − Ef ) = 2π| i|Hint |f |δ (Ei − Ef ) = | i|S|f |. (11. It follows then (in the operator formulation) that the evolution operator or the S-matrix must also be invariant under time inversion.278) In this case. from Fermi’s Golden rule we know that the rate for a process |i → |f is given by Wi→f ∝ 1 | f |S|i |2 ρf .279) Let us choose the initial and the ﬁnal states to correspond to |i |f = |α. pf .277) and furthermore. In this case. when the principle of detailed balance may hold. T ST † = S. we can write (for distinct |i and |f states) f |S|i = −2πi f |Hint |i δ (Ei − Ef ) . if the interaction Hamiltonian is Hermitian. mi .280) . That is.11. (11.

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

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

σ −→ −σ. B −→ B. then the electric dipole moment of the neutron must vanish. if electromagnetic interactions are invariant under P and T . σ −→ σ.292) In other words. (µe )exp < 0. so that we have σ · E −→ −σ · E. this is the Pauli coupling) Hint = µe σ · E + µm σ · B. the electric dipole moment would be parallel to its spin.290) We have already seen that under parity and time reversal E −→ −E. (11. T T (11. Experiments have obtained extremely small limits on µe . Thus the magnitude of any dipole moment (electric) will be of the order of µe ≃ ed ≃ e 10−13 cm = 10−13 e-cm. whereas the interaction involving the magnetic dipole moment is invariant under P and T . the electric dipole interaction changes sign under each one of the transformations. .478 11 Discrete symmetries a magnetic dipole moment. P P P P P E −→ E.291) σ · E −→ −σ · E. T T T (11.293) 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.63 × 10−25 e-cm. σ · B −→ σ · B. Consequently. (11. Since there is no natural axis for the neutron other than the spin axis. σ · B −→ σ · B. Thus the neutron would interact with an electromagnetic ﬁeld through the interaction Hamiltonian (there is no minimal coupling. B −→ −B.289) Let us also note here that the only natural dimension associated with the neutron is its size which is of the order of 1F = 10−13 cm. (11.

m. 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. quantities with bars would refer to antiparticle parameters). C and T in any order even though the individual transformations may not be a symmetry of the system.297) . s −→ η α. s α. We can naturally ask whether there exists any combination of these symmetries which will be a symmetry of the weak Hamiltonian. m.4. takes a particle to the antiparticle state. s|(CPT )−1 (CPT )H(CPT )−1 (CPT )|α. 11.295) where m. Thus under the CPT transformation. let us discuss the obvious consequences of this theorem. |α. (11. −s . m. −s|. (11. then it follows that m = α. −s|H|α. s α. of course. m. α denote the parameters associated with the antiparticle (in this section. m. m. m. s . CPT (11. Let us consider the state of a particle at rest which satisﬁes H|α. (11. −s|(CPT )H(CPT )−1 |α.11. Rather than going into the technical proof of this theorem. m.4 CPT theorem 479 11.296) If the Hamiltonian is CPT invariant. m. s = m|α. Thus we see that m = = = α. The operation of CP.1 Equality of mass for particles and antiparticles. −s = m.294) where m denotes the mass of the particle. m. m. s|H|α. s its spin and α denotes the other quantum numbers.4 CPT theorem We have seen that some of the discrete symmetries appear to be violated in weak interactions.

for example. ˆ 11. Namely.300) (11. Since the weak interactions violate parity. In fact. the equality of particle and antiparticle masses is a consequence of CPT invariance. q.480 11 Discrete symmetries In other words. K 0 and K have the same mass.4. one can show that all the quantum numbers of antiparticles are equal in magnitude but opposite in sign from those of the particles. 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− .) Thus for a one particle state with charge q at rest. we have q = = = ˆ α. Even though C-invariance is violated in weak interactions. −s (11. it is CPT invariance which guarantees 0 that. (11. ˆ (Recall Q = d3 x ψγ 0 ψ for a Dirac particle. (11.2 Electric charge for particles and antiparticles. q. q. s ˆ α. 11.3 Equality of lifetimes for particles and antiparticles. q.4. In this way.298) = −q. s|Q|α. s ˆ α. ˆ T ˆ Q −→ Q. Let Q denote the operator which measures the charge of a state. q. so that ˆ CPT ˆ Q −→ −Q. −s|(−Q)|α. As we have seen ˆ P ˆ Q −→ Q. q.299) ˆ C ˆ Q −→ −Q. the electric charge for particles is equal in magnitude but opposite in sign from that of the antiparticles.301) . s|(CPT )−1 (CPT )Q(CPT )−1 (CPT )|α. the ex0 perimental limit on the mass diﬀerence between the K 0 −K provides the best limit on CPT invariance.

j. j. (11.4 CPT theorem 481 where PH± P −1 = ±H± . j. −mj |H± |β.306) where we have used the fact that H± would be individually invariant under CPT according to the CPT theorem. j. mj −→ |β. Furthermore.302) In such a case. mj |(CPT )−1 (CPT )H± (CPT )−1 (CPT )|α. mj . β. j. j. Let us consider the decay channel of a particle |α.303) The cross terms would have a pseudoscalar character and would vanish since the reaction rate is a scalar quantity. mj |H± |β. −mj . j. mj . mj |H± |α. −mj |H± |β. and the corresponding antiparticle decay channel |α. using the rotational invariance of physical Hamiltonians. j. j. we can rewrite this relation as β. mj |H± |α. j. −mj = α. (11.11. mj = (11. We know that β.305) (11. mj −→ |β. 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 . mj (11. j. j. (11.304) = (i)2mj −2mj α.307) . j. j. j. mj = α. j. mj .

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

. F. New York (1964). New York. New York (1966). 3. Benjamin. New York (1969). Evanston (1961). J. Elementary Particle Physics. R. Spin and Statistics and All That. 5. Zuber. S. D. F. Invariance Principles and Elementary Particles. Itzykson and J-B. Schweber. John Wliley. Bjorken and S. McGrawHill. Row. Roman. J. Gasiorowicz. Relativistic Quantum Fields.5 References 483 2.11. Gross. S. P CT . Quantum Field Theory. Sakurai. Relativistic Quantum Mechanics and Field Theory. 1964. 9. C. 6. Introduction to Relativistic Quantum Field Theory. Princeton University Press. S. 4. J. 7. 8. Introduction to Quantum Field Theory. Wightman. John Wiley. Princeton (1964). John Wiley. P. New York (1993). Drell. Peterson. McGrawHill. 1980. Streater and A. New York.

This page intentionally left blank .

(12.Chapter 12 Yang-Mills theory 12.1 Non-Abelian gauge theories So far. we have talked about the simplest of gauge theories. namely. let us quickly review how the gauge ﬁelds come into the theory in the case of quantum electrodynamics (QED). the Maxwell theory which is based on the Abelian gauge group U (1). 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. δψ δψ = = where θ is a real constant parameter and inﬁnitesimally.3) .38).37) and (8.2) ψ → ψeiθ . Let us start with the Dirac Lagrangian density for a free fermion (8. We will now study gauge theories based on more complicated symmetries. Let us recall that gauge ﬁeld theories necessarily arise when we try to promote a global symmetry to a local symmetry.9) / L = iψ∂ ψ − mψψ. we have −iǫψ. 485 (12. As we have seen in (8. For example. it is a complex ﬁeld). (12.1) where ψ provides a representation of U (1) (namely. this Lagrangian density has the global symmetry ψ → e−iθ ψ. iǫψ.

we have δL = iδψ∂ ψ + iψγ µ ∂µ δψ − mδψψ − mψδψ / −imǫ(x)ψψ + imǫ(x)ψψ.1) whose variation would cancel the ∂µ ǫ(x) term.1) nor the corresponding action is invariant under the local phase transformation in (12.7) . (12. δψ = iǫ(x)ψ.4) since ∂µ ψ does not transform covariantly and there is no term in the Lagrangian density (12. the phase of ψ remains arbitrary.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).5) (12. Consequently. (12.4) = −ǫ(x)ψ∂ ψ + ψγ µ ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ / = (∂µ ǫ(x))ψγ µ ψ. the way out of this diﬃculty is to deﬁne a covariant derivative Dµ ψ such that it transforms covariantly under the local transformation.81)). (12. Neither the Lagrangian density (12. If we now want to promote the symmetry to be local. δ (Dµ ψ) = −iǫ(x)Dµ ψ(x). namely ǫ = ǫ(x) (or θ = θ(x)). As we have seen earlier (see (9. Consequently. δ (∂µ ψ) = ∂µ (δψ) = ∂µ (−iǫ(x)ψ) = −i ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ(x). 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. we note that δψ = −iǫ(x)ψ.486 12 Yang-Mills theory with θ = ǫ = inﬁnitesimal. namely.

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 . this would transform as δFµν = ∂µ δAν − ∂ν δAµ = 1 (∂µ ∂ν ǫ − ∂ν ∂µ ǫ) = 0. the ﬁeld Aµ known as the gauge ﬁeld.86). e (12.9) is invariant under the inﬁnitesimal local gauge transformations δψ = −iǫ(x)ψ(x).10) However.8) We recognize (12. the Lagrangian density L = iψγ µ Dµ ψ(x) − mψψ. is nondynamical in this theory.11) under the gauge transformation (12. δAµ = 1 ∂µ ǫ(x). δψ = iǫ(x)ψ(x). To introduce the kinetic energy part for the gauge ﬁeld (in order to give it dynamics) in a gauge invariant manner. That is. 4 (12.7) as describing the minimal coupling of the charged fermions to the electromagnetic ﬁeld and as we have seen in (9.10). the curl of the ﬁeld would be unchanged by such a transformation. e δAµ (x) = (12.12.13) . if we deﬁne a ﬁeld strength tensor Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . (12. e (12. (12.1 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).

15) where k is an internal symmetry index. k a = 1. (12. Namely. a δψk = −iǫa Tkℓ ψℓ .10). Furthermore. 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. . . Let us consider a free Dirac theory described by the Lagrangian density (containing several complex fermion ﬁelds) L = iψ k ∂ ψk − mψ k ψk .80)) 1 LQED = − Fµν F µν + iψγ µ Dµ ψ − mψψ. 2.14) and has the local gauge invariance described in (12. . .488 12 Yang-Mills theory The total Lagrangian density for QED is then given by (see also (9. dim R. (12. / k = 1. Let us now generalize these ideas to theories with more complicated symmetries. .) The generators (matrices) T a satisfy the Lie algebra of the Lie group G (this is necessary for the transformations to form a group) . . . This Lagrangian density is invariant under the global phase transformations aT a ψk → ψk → Uψ ψU † k = e−iθ ψ k. (12. dim G. dim G represents the dimensionality of the symmetry group. · · · .16) where θ a denotes the real global parameter of transformation and inﬁnitesimally we have δψ k = iǫa ψ ℓ (T a )ℓk . 4 (12.17) with θ a = ǫa = inﬁnitesimal. (We note here parenthetically that dim G = n2 − 1 for the group SU (n). 2. aT a k = ψU −1 = ψeiθ k . dim G. 2. Here the T a ’s represent the generators of the internal symmetry group G and are assumed to be Hermitian. a = 1. .

Let us now try to make this symmetry a local symmetry of our theory. That is. T b = if abc T c . (12. the structure constants are nontrivial). . namely. (12. b. . a. 2. 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 = 0.1 Non-Abelian gauge theories 489 T a . . the symmetry group is known as a non-Abelian group. f 147 = f 246 = f 257 = f 345 = 2 . namely.16)).18) − 1 . we already know from the study of angular momenta. f 458 = f 678 = 23 . For example. (12.) When the generators of the symmetry group 2 do not commute (namely.12. c = 1. .19) Here we have used the fact that the symmetry generators are Hermitian.17) (or (12. (For semi-simple groups the structure constants can always be chosen to be completely antisymmetric. let us consider the inﬁnitesimal local phase transformations a δψk = −iǫa (x)Tkℓ ψℓ . the structure constants of the group SU (2) can be identiﬁed with f abc = ǫabc .20) Once again. 3 while for the group SU (3) the structure constants are more complicated and the nontrivial ones 1 are given by f 123 = 1. a. It is easy to see that the Lagrangian density is invariant under the transformations (12. (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. we note that the ordinary derivative acting on the fermion ﬁeld does not transform covariantly under this transformation. 2. a δψ k = iǫa (x)ψ ℓ Tℓk . dim G. . b. 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. c = 1.

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

a δ (Dµ ψ)k = −iǫa Tkℓ (Dµ ψ)ℓ . µ (12. µ This can also be written as . we obtain a igTkℓ δAa ψℓ µ a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a T b km Ab ψm µ −gǫa T b T a km Ab ψm µ km a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a .12.30) a = i (∂µ ǫa ) Tkℓ ψℓ + igǫa f abc (T c )km Ab ψm .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 (12.26) 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 . (12. µ µ (12.28) We note that the left-hand side in (12.1 Non-Abelian gauge theories 491 Noting that we would like the covariant derivative to transform covariantly under an inﬁnitesimal local transformation.27) we conclude from (12.29) Substituting this into (12.28).

. 2.33) with the covariant derivative deﬁned in (12. dim G.36) with normal product rules for matrices. a δψ k = iǫa (x)ψ ℓ Tℓk .492 12 Yang-Mills theory a igTkℓ ψℓ δAa − µ 1 ∂µ ǫa + f abc Ab ǫc = 0.31) which determines the transformation for the gauge ﬁeld to be δAa = µ 1 1 ∂µ ǫa − gf abc Ab ǫc = ∂µ ǫa − f abc Ab ǫc . (12. µ g (12.35) where the gauge ﬁelds Aµ are matrices belonging to the same representation of the group as the fermions (since the generators T a belong to this representation). µ µ g g (12. (12. . we can write the Lagrangian density in (12. µ a = 1.33) in terms of matrices as (of course. the Lagrangian density is a scalar) L = iψγ µ Dµ ψ − mψψ.34) If we use the deﬁnition (see (12. We can now deﬁne a unitary representation of the group as describing the ﬁnite symmetry transformation matrix (see (12.32) Thus the Lagrangian density L = iψ k γ µ (Dµ ψ)k − mψ k ψk .16)) . δAa = µ 1 ∂µ ǫa − f abc Ab ǫc .25)) Aµ = T a Aa . µ g (12. (12. .26) is invariant under the inﬁnitesimal local transformations a δψk = −iǫa (x)Tkℓ ψℓ (x). .

Aµ → U Aµ U −1 − 1 (∂µ U ) U −1 . U † = U −1 .39) . ψ → ψU −1 . (12.12.34)) for the ﬁeld variables in the matrix form as ψ → U ψ. the tensor representing the Abelian ﬁeld strength (see (12.1 Non-Abelian gauge theories 493 U = e−iθ a (x)T a . we note that under the gauge transformation (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 .37) so that we can write the ﬁnite transformations (corresponding to (12. (12.38). ig (12.38) To construct the Lagrangian density for the dynamical part of the gauge ﬁeld in a gauge invariant manner.

(12. if we deﬁne the ﬁeld strength tensor as Fµν = ∂µ Aν − ∂ν Aµ + ig [Aµ . here fµν = ∂µ Aν − ∂ν Aµ is neither invariant nor does it have a simple transformation under the gauge transformation (12.40).40) Thus comparing (12.38). then under the gauge transformation (12.41) (12. we see that unlike in the case of QED. in the present case. 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 .494 12 Yang-Mills theory Namely. It is now easy to construct the gauge invariant Lagrangian density for the dynamical part of the gauge ﬁeld as (quadratic in derivatives) . Fµν → U Fµν U −1 . Let us also note that under the gauge transformation ig [Aµ .42) so that Fµν transforms covariantly. (12.38). it is clear that. Aν ] U −1 − − ∂µ U Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 . Aν ] .39) and (12.

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 .1 Non-Abelian gauge theories 495 1 1 a Lgauge = − Tr Fµν F µν = − Fµν F µν a . more generally for fermions interacting with a non-Abelian gauge ﬁeld. ν µ µ ν or. QCD corresponds to the particular case when G = SU (3)) is.41) is easily seen to take the form. a Fµν T a = ∂µ Aa − ∂ν Aa T a + igAb Ac T b . given by 1 a LQCD = − Fµν F µν a + iψ k γ µ (Dµ ψ)k − mψ k ψk .46) where we can set m = 0 if the fermions (quarks) are massless. the ﬁeld strength tensor (12. or.43) where a particular normalization for the trace of the generators is assumed (namely. a a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac = −Fνµ .45) The complete Lagrangian density for QCD (quantum chromodynamics.59) and the comments fol2 lowing that equation). 4 (12. Aν . From the cyclicity of trace this Lagrangian density is easily seen to be invariant under the gauge transformations (12.38) which have the inﬁnitesimal form (see (12. T a T b = 1 δab . therefore.12.38).34)) . (12. T c ν µ µ ν = ∂µ Aa − ∂ν Aa T a − gf abc Ab Ac T a . µ ν ν µ (12. In components. 2 4 (12. Fµν = ∂µ Aν − ∂ν Aµ + ig Aµ . This Lagrangian density is invariant under the non-Abelian gauge symmetry transformations (12. see (12.

51) If we can divide the generators of the non-Abelian Lie algebra into two non-Abelian subsets such that f abc = 0 when one index is in one set and another index is in the second set. f abp f pcq + f cap f pbq + f bcp f paq = 0.49) imposes a restriction on the structure constants of the group of the form if abp T p . T c + T c. T b + T b . T b = if abc T c .48) it can be seen that the Jacobi identity (12. or. T c + if cap T p .48) and the Jacobi identity associated with this algebra is given by T a. µ g (12. (12. (12.50) 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. δAa = µ δψk = −iǫa (x)(T a )kℓ ψℓ . The algebra of the Hermitian generators.47) Let us discuss brieﬂy some of the properties of the Lie algebra of the symmetry group G.18). T a . as we have noted in (12. T a = 0. (12. In this case .496 12 Yang-Mills theory δψ k = iǫa (x)ψ ℓ (T a )ℓk . T c . has the form T a . T b + if bcp T p . T b . (12. T a = 0. then the Lie algebra breaks up into two commuting non-Abelian subalgebras. f abp f pcq T q + f cap f pbq T q + f bcpf paq T q = 0. 1 ∂µ ǫa − f abc Ab ǫc .37) as well) U (x) = eiθ a (x)T a . or.49) Using (12.

Furthermore. In all our discussions.12. For any representation of a simple Lie group we can write Tr T a T b = C2 δab .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 .55) where the index c is ﬁxed (and not summed). A direct product of simple Lie groups is called semi-simple. (12. a = b. (12. we will assume that the symmetry group G is simple. Tr T a T b = (12. using the commutation relation (12. A non-Abelian Lie group that cannot be so factorized is called a simple Lie group.54) Tr Tr T a T b T c − Tr T b T a T c . we note that .53) Using the cyclicity property of the trace. On the other hand. T b T c (12. let us next note that the quantity habc = = T a. It depends on the representation but not on the indices a and b. 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. is completely antisymmetric in all its indices.52) We note that C2 is a normalization constant which determines the values of the structure constants.1 Non-Abelian gauge theories 497 the group G is a direct product of two independent non-Abelian Lie groups.

) We note that if we write Tr T a T b = T (R) δab . T c T b = if acp Tr T p T b = if acp Kp δpb = iKb f acb = −iKb f abc .58) (12. · · · . 2. n = 1. we note here that it is (12. since habc is completely antisymmetric.59) which is used to show that the structure constants for a semi-simple group are completely antisymmetric in the indices. (12.56) with the index b ﬁxed. Similarly.56) we conclude that Kb = Kc = K. 2. However.57) Using this and comparing (12.498 12 Yang-Mills theory hacb = Tr T a.60) then T (R) is known as the index of the representation R. (12. dim G. m. (12. (12.61) . b = 1. (12. a. we have hacb = −habc . dim R. · · · . Furthermore.55) and (12.43). we have (T a T a )mn = C(R) δmn .59) where the constant C2 depends only on the representation. (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. This shows that we can write Tr (T a T b ) = C2 δab .

64) Furthermore. (12. a particular representation that is very important as well as useful is given by Ta = −if abc .65) where we have used the anti-symmetry of the structure constants as well as (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 .1 Non-Abelian gauge theories 499 where C(R) is known as the Casimir of the representation R. As a result.50). However. (12. 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. namely.63) bc This is consistent with the hermiticity requirement for the generators. T a. we conclude that the identiﬁcation . (12. we can easily check that this representation satisﬁes the Lie algebra. T a† = = Ta ∗ ∗ bc cb − if acb = if acb = −if abc = T a bc .62) We can determine the generators of the group in various representations much like in the case of angular momentum.12.

Let us next look at the transformation for the gauge ﬁelds in (12. g µ (12.68) which shows that the gauge ﬁeld.69) .26)) δAa = µ 1 (adj) a D θ . transforms according to the adjoint representation of the group.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 .66) indeed deﬁnes a representation of the Lie algebra known as the adjoint representation.500 12 Yang-Mills theory a T(adj) bc = −if abc . (12. (12. T b . This can also be seen from the transformation of the ﬁeld strength tensor in (12. µ (12. for example. (12. which inﬁnitesimally has the form Fµν → U Fµν U −1 so that = Fµν + iǫb Fµν .42). Aµ .67) so that we can equivalently write (see.

For example. or. ab µν (12. µ Let us now concentrate only on the gauge ﬁeld (Yang-Mills) part of the Lagrangian density 1 a LYM = − Fµν F µν a 4 1 = − ∂µ Aa − ∂ν Aa − gf abc Ab Ac ν µ µ ν 4 × ∂ µ Aνa − ∂ ν Aµa − gf apq Aµp Aνq . (12. (It does not matter what representation the matter ﬁelds belong to.67) (adj) Dµ ǫ a = ∂µ ǫa − gf abc Ab ǫc .) For completeness.1 Non-Abelian gauge theories 501 δFµν = iǫb Fµν . let us scale the ﬁeld variables as .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. let us note here that the covariant derivative in the adjoint representation (12. T b .34) we conclude that the ﬁeld strength Fµν as well as the gauge ﬁeld Aµ transform according to the adjoint representation of the group. or. 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.73) Let us note the following features of this Lagrangian density.71) can also be written in the matrix form as (adj) Dµ ǫ = ∂µ ǫ + ig[Aµ . µ (12. ǫ]. T b a = iǫb Fµν T a . (12.70) Comparing this with the transformation of the fermions in (12. a a δFµν T a = iǫb Fµν if abc T c .

ν µ µ ν = ∂µ Aν − ∂ν Aµ + ig[Aµ . The other feature to note is that unlike the photon ﬁeld. 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.45)) 1 a 1 (12. 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. Aν ]. (12.76) . ν µ µ ν g which leads to LYM → − 1 ∂µ Aa − ∂ν Aa − f abc Ab Ac ν µ µ ν 4g2 × ∂ µ Aνa − ∂ ν Aµa − f apq Aµp Aνq . the pure Yang-Mills theory is an interacting theory unlike the Maxwell theory. We note that the Lagrangian density for the Yang-Mills theory is given by (12.74) Thus we note that in this theory. in the case of non-Abelian symmetry they must couple to themselves and possess self interactions. Physically we understand this in the following way. the coupling constant can be scaled out and written as an overall multiplicative factor in the Lagrangian density.41) and (12. (12. here the gauge ﬁelds have self interaction (they interact with themselves) and. 12. g µ It follows then that 1 a Fµν → ∂µ Aa − ∂ν Aa − f abc Ab Ac .44)) Fµν a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac . therefore.502 12 Yang-Mills theory Aa → µ 1 a A . In the present case.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 9.43) (or (12.75) LYM = − Tr Fµν F µν = − Fµν F µν a .

72)).75) and have the form ∂L ∂L − = 0. 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) . (12. i 0 0 i a Fij a = ∂i Aa − ∂j Aa − gf abc Ab Ac = −ǫijk Bk . F µν ] = 0.75) can also be written explicitly in terms of the non-Abelian electric and magnetic ﬁeld strength tensors (12.80) Let us note that the coeﬃcient matrix of highest derivatives in (12. −∂µ F µν a − −gf abc Ab F µν c = 0.75) is the transverse projection operator just like in the case of Maxwell’s theory. we can identify the non-Abelian electric and magnetic ﬁelds from (12.68)).2 Canonical quantization of Yang-Mills theory 503 As in the case of the Abelian theory (Maxwell theory).79) where the covariant derivative is deﬁned to be in the adjoint representation of the group (see (12.77) The Lagrangian density (12.77) as 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a . In the matrix notation (see (12.67) and (12. ∂∂µ Aa ∂Aa ν ν ∂µ or.78) The Euler-Lagrange equation for the theory are obtained from (12. j i i j (12. the theory is singular implying that there are constraints which is evident from the fact that the YangMills Lagrangian density is invariant under the inﬁnitesimal gauge transformation (see (12. 2 2 4 LYM = Tr Ei Ei − (12. or. µ (12. Therefore. we can write the Euler-Lagrange equations as Dµ F µν = ∂µ F µν + ig[Aµ .12.76) as a a F0i = ∂0 Aa − ∂i Aa − gf abc Ab Ac = Ei .

The momentum conjugate to Aa does not exist. we have Π0 a (x) = −F 00 a (x) = 0. This implies that 0 Aa is like a c-number quantity which commutes with every other 0 operator in the theory.79) for ν = 0. ˙ a (x) ∂ Aµ (12. Thus. namely. A0 (x) = 0. (12. Thus we can choose a gauge condition and set it equal to zero. requiring that . 0 or.75) and this leads to µ Πµ a (x) = ∂L = −F 0µ a (x). g (12. the constrained structure of the theory is already obvious in the Euler-Lagrange equation of motion (12. Di F i0 a = 0. we choose Aa (x) = 0.77).81) In fact.504 12 Yang-Mills theory δAa = µ 1 (Dµ ǫ(x))a . For example. (12. ν. Let us obtain the momenta canonically conjugate to the ﬁeld variables Aa from (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.84) (12.82) which is a constraint relation.83) a Noting that the ﬁeld strength Fµν is antisymmetric in the indices µ. Namely. (12. we see that we have only 3N canonical momenta where N = dim G.85) a where the non-Abelian electric ﬁeld Ei (x) is deﬁned in (12. as in the case of Maxwell’s theory and a Πi a (x) = −F 0i a = Ei (x).

88) 2 2 4 LYM = Tr Ei Ei − where we have used the fact that in the gauge (12. let us also note that the potential term in (12. we expect the physical dynamical degrees of freedom of a massless gauge ﬁeld to be transverse.86) the Lagrangian density (12.86) the non-Abelian electric ﬁeld in (12.88) (namely. in the present case. (12. On the other hand.90) so that neither the gauge ﬁeld Ai nor the electric ﬁeld Ei is transverse ˙ (recall that Ei = Ai ). we note from (12. Ei ] = 0. 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 .89) From our discussion in the case of the Maxwell theory. ˙ Ei = Ai .86).12.2 Canonical quantization of Yang-Mills theory 505 A′ = U A0 U −1 − 0 or.77) takes the form a ˙ Ei = Aa .87) we can determine the parameter of transformation and we see that a gauge ﬁeld can always be transformed to satisfy (12.91) ig ig so that there is a cyclic variable in the theory that needs to be separated. ig (12.91) can be inverted to give . However. 1 (∂0 U ) U −1 = 0. Let us note that with the gauge condition (12. We note that the relation (12. (12. (12. (12. i or.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 . U −1 ∂0 U = igA0 .82) that Di Ei = ∂i Ei + ig[Ai .

let us require the new ﬁeld variable Ai to be transverse. then we can deﬁne a new ﬁeld variable Ai satisfying the transversality condition (12.93) 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 (12.94) where we have used (12. We note next from the ﬁeld redeﬁnition in (12. we obtain 1 ¯ ∂i Ai = ∂i S Ai S −1 − ∂i (∂i S)S −1 ig 1 ¯ ¯ = (∂i S)Ai S −1 + S Ai (∂i S −1 ) − ∂i (∂i S)S −1 ig = (∂i S) S −1 Ai − + Ai S + = − = − 1 (∂i S −1 ) ig (12. Imposing this condition in (12.91) that .91).95) This shows that if we can ﬁnd a gauge transformation which satisﬁes ¯ (12. ig (12. G]. (∂i S)S −1 ] ig 1 Di (A) (∂i S)S −1 . (12.506 12 Yang-Mills theory 1 1 ¯ Ai = S −1 Ai S + S −1 ∂i S = S −1 Ai + (∂i S)S −1 S. namely. ¯ ∂i Ai = 0.92) ig ig Keeping in mind that we would like the dynamical variables to be ¯ transverse. (12.93).72)) Di (A)G = ∂i G + ig[Ai .94).

99) Furthermore. Here Di (A) is the covariant derivative in the adjoint representation (see (12.2 Canonical quantization of Yang-Mills theory 507 1 ˙ ¯ Ai = ∂0 S Ai S −1 − (∂i S)S −1 ig ˙ ˙¯ ¯ ¯ ˙ = S Ai S −1 + S Ai S −1 − S Ai S −1 SS −1 − 1 ˙ ˙ (∂i S)S −1 − (∂i S)S −1 SS −1 ig 1 ˙ ¯ ¯ ˙ ˙ = S Ai − [Ai .100) . (12. where (12. ig where we can identify 1 ˙ ¯ ¯ E i = Ai − Di (A)f.98) (12. ig (12.86) we can write 1 ˙ ˙ ¯ ¯ Ei = Ai = S Ai − Di (A)f S −1 = SE i S −1 .95)) with respect to ¯ the gauge ﬁeld Ai and we have identiﬁed ˙ f = S −1 S.97) From (12.96) ˙ ¯ where we have used ∂0 S −1 = −S −1 SS −1 .96) we note that in the gauge (12. decomposing E i into its transverse and longitudinal components as T E i = E i + ∂i φ.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 .

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 ] = 0.90) takes the form ¯ ¯ Di (A)E i = ∂i E i + ig[Ai . or. E i + ∂i φ ] = 0. ¯ ∂i Di (A)φ = ρ.101) we conclude from the deﬁnition in (12.91). As a result of this relation. E i ]S −1 ig ¯ = S(∂i E i + ig[Ai . ¯ ¯ ∂i ∂i φ + ig[Ai . We can invert this relation to determine .102) Let us next note that under the ﬁeld redeﬁnition (12.103) where we have used (12. Di (A)Ei = ∂i Ei + ig[Ai . φ] = −ig[Ai .91) and (12. φ=− (12. (12. φ] + ig[Ai .104) T ¯ where we have used ∂i Ai = 0 in the intermediate step. E i ] = 0.99) that 1 1 ¯ ∂i Di (A)f. or. or. the constraint equation (12. or. (12. T ¯ ∂i ∂i φ + ig[Ai . ig ∇2 ˙ ¯T ¯ Ei = 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).508 12 Yang-Mills theory ∂i E i = 0. (12.98). T ¯ ¯ ∂i ∂i φ + ig∂i [Ai . E i ])S −1 ¯ = S(Di (A)E i )S −1 . E i ] = ρ.

106) Using the deﬁnition in (12. ¯ We have isolated the dynamical variables of the theory to be Aa i which are transverse and the conjugate momenta can be determined from (12. ¯ 2 ∂j Dj (A) (12. ∂j Dj (A) (12.108) .105) so that from (12. (12.2 Canonical quantization of Yang-Mills theory 509 φ = 1 ¯ ρ. the Lagrangian density (12.93).107) to be Π ia = ∂LYM ˙ ¯ = Aa .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 . ∂i Di (A) = Di (A)∂i which has been used together with integration by parts). ∂i Di (A) 1 ¯ ρ.100) we obtain T ˙ ¯ E i = E i + ∂i φ = Ai + ∂i (12.107) where we have neglected total divergence terms in the intermediate ¯ ¯ steps (note also that because of (12.98) and the decomposition in (12. i ˙ ¯a ∂A i (12.12.91).106) as well as the cyclicity of trace.

Π (y)} = δab δi j (x − y). therefore.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). Here we see that since the gauge ﬁelds are self interacting. 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.112) . This is related to the question of gauge ﬁxing. (12. Let us note here that the long range behavior of the interaction (and. namely the Coulomb gauge does not uniquely deﬁne the gauge ﬁelds. i. i (12. of the theory) ¯ is. this last term has the form ρ 1 2 ρ which corresponds to ∇ the long range Coulomb interaction). Thus the long range behavior of this theory is still not well deﬁned in the Coulomb gauge. even in the absence of other matter ﬁelds there is a long range interaction.111) (12. i ¯ ∂i Aa = 0. This can be seen from the ¯ fact that if we have a ﬁeld conﬁguration Aa which is transverse.31). Here the transverse delta function corresponds to the one already deﬁned in the case of Maxwell’s theory in (9. controlled by the eigenvalues of the operator ∂i Di (A). i TR jb (12.109) with all other brackets vanishing. A′ a = Aa + i i g such that ¯i ∂i A′ a = 0. therefore.110) then we can make a time independent inﬁnitesimal gauge transformation (under which the theory is invariant) 1 ¯ǫ a ¯ ¯ Di (A)¯ .107) has exactly the same form as in the Abelian theory except for the last term. We note that the Lagrangian density in (12. ¯ Gribov has shown that the operator ∂i Di (A) does possess zero modes or eigenvectors with zero eigenvalue.e.

(12. by modifying the theory. However. The canonical quantization of Yang-Mills theories does not have manifest covariance very much like the Maxwell theory.129)) (∂ · A) = 0. the hypersurface deﬁning the gauge choice intersects the gauge orbits more than once. known as the Gribov ambiguity. The other way of saying this is that within the framework of perturbation theory we are interested in inﬁnitesimal gauge transformations. Gribov ambiguity is a phenomenon associated with large gauge transformations and is an important issue since it leads to nonperturbative eﬀects.111) that this has a singularity at g = 0. As long as we are within perturbation theory we can neglect such conﬁgurations. Therefore. there are serious diﬀerences from Maxwell’s theory. classically we can impose the condition ∂ · A = 0. There now exist prescriptions to ¯ take care of this problem. occurs in all gauges other than the axial gauge. 4 2 (12. namely. However.12. we recall that in QED (see (9. let us modify the theory (12. ∂ · A behaves like a free ﬁeld and. for example. we can also try to quantize the non-Abelian gauge theory covariantly. But let us note from the form of A′ a in i (12.115) . consequently.114) namely. therefore. This shows that ¯ the Coulomb gauge does not really specify the gauge ﬁeld conﬁguration uniquely. makes the theory nonsingular. In other words. Thus. very much along the lines of the Abelian theory that we have discussed in chapter 9. much like in the Maxwell theory.2 Canonical quantization of Yang-Mills theory 511 ¯ provided ǫa corresponds to a zero mode of ∂i Di (A). We can show that this nonuniqueness. (12. in the present case.75) as (see (9. For example. (It is of the form O 1 .125) with J µ = 0) 1 1 a L = − Fµν F µν a − (∂µ Aµa )2 .113) The additional term clearly breaks gauge invariance and. On the other hand. we would not worry about it within the context of perturbation theory.) g Hence such gauge ﬁeld conﬁgurations can only be reached nonperturbatively.

We would see next how we can derive intuition on this important question from the path integral quantization of the theory.) Correspondingly the naive analog of the Gupta-Bleuler condition for non-Abelian gauge theories does not seem to exist.118) Thus. just modifying the Lagrangian density as in (12. nor can we think of the supplementary condition ∂ · Aa(+) |phys = 0. (12.3 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 . (12.512 12 Yang-Mills theory which then translates to the Gupta-Bleuler condition on the physical states ∂ · A(+) |phys = 0. since (∂ · Aa ) is not a free ﬁeld.116) In the case of the Yang-Mills theory. the physical subspace would keep changing with time which is not desirable. the additional term in (12. Furthermore. Therefore.113) does not seem to be suﬃcient in contrast to the Abelian gauge theory. (12. consequently. we obtain (∂ · Aa ) = gf abc ∂ν Ab F µν µ c (12. 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.119) in a physically meaningful manner. In other words. (Namely. we need to analyze the question of modifying the theory in a more systematic and detailed manner. µ Contracting with ∂ν .113) has truly modiﬁed the theory. it cannot be uniquely decomposed into a positive and a negative frequency part (in a time invariant manner). since it is not invariant under time evolution. in contrast to the Abelian theory.117) = 0. however. 12.

The functional dependence of a quantity on a variable is generally denoted by a square bracket. Z[J] and W [J] are known as generating functionals for the Green’s functions of the theory. we will not go into the systematic details of this description.120) where N is a normalization constant. For example. from the “in” vacuum to the “out” vacuum.4 for the deﬁnitions of “in” and “out” states) in the presence of interactions. see section 6. 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. (12. Since this is a complete topic in itself. the path integral description is given by (recall = 1) J 0|0 = Z[J] = eiW [J] = N = N Dφ ei(S[φ]+ Dφ eiS R d4 x J(x)φ(x)) (J ) [φ] . 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.122) and various Green’s functions (n-point functions) are obtained from the generating functionals Z[J] and W [J] through functional diﬀerentiation.121) Furthermore. J(x) is the external source to which the scalar ﬁeld is coupled and S[φ] = d4 x 1 ∂µ φ∂ µ φ − m2 φ2 . we 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 . rather.12. for a real scalar ﬁeld interacting only with an external source. 2 (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 only go over some of the essential concepts involved in such a description. . δφ(y) ǫ ǫ→0+ (12. Like the derivative of a function.

514 12 Yang-Mills theory The integration in (12. When we have an . namely.123) Clearly in the limit of vanishing volume we recover lim φi = φ(x).126) is independent of ﬁeld variables and is a constant (possibly divergent) and can be incorporated into the normalization constant N .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.125) With these basics a simple functional integral for a quadratic action such as in (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) . the functional integration is deﬁned (up to a normalization constant) as Dφ = dφi .120) can be evaluated in a straightforward manner.124) δVi →0 and in terms of these discretized ﬁeld variables.120) is known as a functional integration and is deﬁned as follows. δVi φi = (12. (12. (12. i (12. G = (−∂µ ∂ µ − m2 )−1 . Basically. (12.) The determinant in (12.126) where the Green’s function G(x − y) is formally the inverse of the operator in the quadratic part of the action. Let us divide the entire space-time into inﬁnitesimal cells labelled by “i” of volume δVi and deﬁne 1 δVi d4 x φ(x).

With this brief introduction to path integral description. (12.P −1 J ν µν ).A µ) DAµ ei[ 2 (Aµ .126) Z [Jµ ] = eiW [Jµ ] = N (det (−P µν ))− 2 e− 2 (J 1 i µ . 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. (12. This is a Gaussian functional integral and we can evaluate this using our earlier result in (12. In this case. the generating functional in the path integral formalism is given by (J ) [A Z [Jµ ] = eiW [Jµ ] = N = N DAµ eiS µ] (12. (12.122)) we have identiﬁed P µν (x − y) = (η µν (J µ . Aµ ) = − ∂ µ ∂ ν ) δ4 (x − y). in general. However.130) where N is a normalization constant and (see (9. cannot be obtained in a closed form.128) where J µ represents a conserved current (source).120).124)). The advantage of using the path integral description lies in the fact that there are no operators in (12. everything is a classical function.129) µν A ν )+(J µ .3 Path integral quantization of gauge theories 515 interacting theory (not quadratic in the ﬁeld variables) the functional integral. 4 (12. In this case.131).131) d4 x J µ (x)Aµ (x). as we have seen before (see (9. the operator P µν is a projection . let us go back to the Maxwell theory 1 L(J) = − Fµν F µν + J µ Aµ .P 1 ] .12.132) where we have used the notation in (12.123) and (9.

even in the absence of any sources.133) in terms of U (θ) is quite general and holds for the non-Abelian theories as well. consequently. x (12.136) . Therefore. e Aµ → A(θ) = Aµ + µ (12. we can write the transformation as 1 ∂µ θ(x). is constant on such orbits. Clearly therefore.134) We can immediately see that for a U (1) gauge group.) The source of the diﬃculty is not hard to see. The Lagrangian density for Maxwell’s theory is invariant under the gauge transformation 1 (∂µ U )U −1 . (The operator possesses zero modes and. The longitudinal vectors kµ (or ∂µ ) are its eigenvectors with zero eigenvalue. (12.516 12 Yang-Mills theory operator for transverse photons. the determinant of P µν vanishes. (As we have seen the form of transformation for the gauge ﬁelds (12.135) which is the familiar gauge transformation for Maxwell’s theory. on the other hand.133) (12. The action S. ie Aµ → A(θ) = U Aµ U −1 − µ where U (θ) = e−iθ(x) . This is a U (1) symmetry and the symmetry is noncompact since the parameter of transformation θ(x) can take any real value. is proportional to the “volume” of the orbits denoted by dθ(x). the inverse of the matrix cannot be deﬁned either. the generating functional.) (θ) For a ﬁxed Aµ . 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. This implies that the generating functional does not exist in this case.

(12.12. A0 (x) = 0. gauge invariant and do not depend on the choice of the hypersurface (gauge). then even if Aµ does not satisfy the condition. namely.139) is known as the gauge condition (or gauge ﬁxing condition). the axial gauge.) This is an inﬁnite factor (which is one of the reasons for the divergence in the naive evaluation of the functional integral) and must be extracted out before doing any calculations.138) has a unique solution for some θ(x). for example. In this way. (12. of course. we can ﬁnd a gauge trans(θ) formed Aµ which does and F [A(θ) (x)] = 0. µ (12. This procedure is known as gauge ﬁxing and the condition F [Aµ (x)] = 0. Rather we should integrate over each orbit only once. here are a few of the familiar gauge ﬁxing conditions F [Aµ (x)] = ∂µ Aµ (x) = 0. The way this is carried out is by choosing a hypersurface which intersects each orbit only once. 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).3 Path integral quantization of gauge theories 517 (In the non-Abelian case. the Lorentz/Landau gauge. if F [Aµ (x)] = 0. we pick up a representative gauge ﬁeld from each gauge orbit. the temporal gauge. Physical quantities are. We can . ∇ · A(x) = 0. 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. A3 (x) = 0. the Coulomb gauge.137) deﬁnes the hypersurface which intersects the gauge orbits once.140) and so on. Thus. (12.

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

12. dθ(x) = (12. µ under which the generating functional takes the form dθ(x) δ F [Aµ (x)] eiS x (J ) [A µ] (J ) [A µ] (12. d(U U ′ ) = dU.141) that . namely. Therefore. let us make an inverse gauge transformation Aµ → A(−θ) .147) Z[Jµ ] = N = N DAµ ∆FP [Aµ ] dθ(x) x DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS (J ) [A µ] = N DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS .3 Path integral quantization of gauge theories 519 group a translation of the transformation parameter corresponds to a gauge transformation) d θ(x) + θ ′ (x) . (12. (12. To do this let us note from (12. However. this gives the correct functional form for the generating functional. (12.144) In the non-Abelian case.148) where the (inﬁnite) gauge volume has been factored out and absorbed into the normalization constant N of the path integral. we should have the Haar measure which is gauge invariant.146) Furthermore.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] . we still have to determine what ∆FP [Aµ ] is.

we have θ(x) = 0.149) δθ δF F [Aµ (x)]=0 (θ) .) 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). 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.150) (12.153) . using (12. On the other hand. (12.152) The Faddeev-Popov determinant can.151) θ=0 . (12. 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). for gauge ﬁelds which satisfy the condition F [Aµ (x)] = 0.520 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. (Here we can completely ignore the problem of Gribov ambiguity associated with large gauge transformations.149)-(12. be thought of as the Jacobian that goes with a given gauge choice.151) we determine δF [Aµ (x)] ∆FP [Aµ ] = det δθ(y) (θ) (12. therefore. Thus.

Thus. the generating functional becomes Z[Jµ ] = N × = N = N DAµ ∆FP [Aµ ] Df δ F [Aµ (x)] − f (x) e DAµ ∆FP [Aµ ]e i − 2ξ R d4 x(f (x))2 iS (J ) [Aµ ] e h i R 1 i S (J ) [Aµ ]− 2ξ d4 x(F [Aµ (x)])2 (J ) +S DAµ ∆FP [Aµ ] ei(S 1 2ξ GF ). (12. We note that physical quantities are independent of f (x). Thus the generating functional in this case is given by Z[Jµ ] = N DAµ ∆FP [Aµ ]δ F [Aµ (x)] − f (x) eiS (J ) [A µ] . Hence we can multiply the generating functional by a weight factor and integrate over all f (x). (12. Then we can insert the identity ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] − f (x) = 1.12.158) . we can now do the following (also known as the ’t Hooft trick). (12.155) Following ’t Hooft.157) and ξ is known as the gauge ﬁxing parameter. We note further that since δF [Aµ (x)] δθ(y) (θ) ∆FP [Aµ ] = det θ=0 .154) into the functional integral.3 Path integral quantization of gauge theories 521 where f (x) is independent of Aµ . (12. (12.156) where we have deﬁned a gauge ﬁxing action as SGF = d4 x LGF = − d4 x (F [Aµ (x)])2 . The Faddeev-Popov determinant is unchanged by this modiﬁcation because f (x) does not depend on Aµ (x).

to write the Faddeev-Popov determinant in the form of a ghost action (action involving ghost ﬁelds). known as ghost ﬁelds. (12. [c(x). 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.c..161) Thus although these ﬁelds behave as scalar ﬁelds under Lorentz transformations.162) .( δθ )θ=0 c) Dc Dc e −i R d4 xd4 y c(x) δF [Aθ (x)] µ δθ(y) θ=0 c(y) Dc Dc eiSghost .c] . [c(x).159) θ=0 c(y). c(y)]+ = 0. (12.160) Here we have introduced two independent ﬁctitious ﬁelds c(x) and c(x). δF [Aθ (x)] µ d xd y c(x) δθ(y) 4 4 (12. Thus the generating functional now takes the form Z[Jµ ] = eiW [Jµ] = N DAµ Dc Dc eiSeﬀ (J ) [Aµ . (12. they obey anticommutation rules. [c(x). graphs involving these ﬁctitious particles in closed loops must have an additional (−1) factor just like the fermions. but opposite statistics.e. These ﬁelds are known as Faddeev-Popov ghosts and as is obvious from their anticommutation relations.). c(y)]+ = 0. i. 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.

142) that both these actions arose from inserting a factor of unity into the functional integral). 2ξ 2ξ (12. 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. the gauge ﬁxing Lagrangian density (see (12.12. Let us now look at a simple gauge ﬁxing condition in Maxwell’s theory. Otherwise it fails its purpose and will not correspond to an acceptable gauge ﬁxing condition. Of course. c] = S (J) [Aµ ] + SGF + Sghost = d4 x Leﬀ [Aµ .157)) has the form LGF = − 1 1 (F [Aµ (x)])2 = − (∂µ Aµ (x))2 . our responsibility to show that the physical interpretation of the theory has not changed. To do covariant perturbation theory in the path integral formalism. the gauge ﬁxing Lagrangian density should make the quadratic part of the eﬀective Lagrangian density nonsingular. However. c.164) In this case. (12. of course. c] . for the moment let us emphasize that the purpose of a gauge ﬁxing Lagrangian density is to break gauge invariance. the covariant condition (see (12.153)) F [Aµ (x)] = ∂µ Aµ (x) = f (x).3 Path integral quantization of gauge theories 523 where Seﬀ [Aµ . (J) (J) (12. this modiﬁes the starting theory.165) . It is. 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. that the S-matrix we obtain is independent of the gauge choice and is unitary and that this formulation leads naturally to the Gupta-Bleuler states as physical states. It is worth noting at this point that we have modiﬁed our starting Lagrangian density by a series of formal manipulations.163) Thus we can summarize what we have done so far. We would show this in the next chapter. namely. for example.

the ghost action (12.169) is nothing other than Maxwell’s theory in the Feynman-Fermi gauge (9. we note that F [A(θ) (x)] = ∂µ A(θ)µ (x) = ∂µ Aµ (x) + µ so that (see (12. (12.160) for this gauge choice is obtained to be δF [Aθ (x)] µ δθ(y) xδ 4 Sghost = − = − = d4 xd4 y c(x) d4 xd4 y c(x) c(y) θ=0 (x − y) c(y) d4 x Lghost .166) = θ=0 1 e xδ 4 (x − y). (12.169) We note here that the ghost ﬁelds are noninteracting in the case of the Maxwell theory in ﬂat space-time and.125).524 12 Yang-Mills theory It is clear that this provides longitudinal components to the quadratic terms in ﬁelds and hence breaks gauge invariance. e (12. 4 2ξ (12. therefore. however. 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 . we may neglect them and then for ξ = 1 we recognize that our eﬀective Lagrangian density (12.168) d4 x ∂µ c(x)∂ µ c(x) = where we have neglected total divergence terms.167) Absorbing the factor 1 into the normalization factor (alternatively e scaling the ghost ﬁelds). In non-Abelian gauge theories. the ghost ﬁelds are interacting and have to be present.152)) δF [Aµ (x)] δθ(y) (θ) 1 µ ∂ θ(x) . This . To obtain the corresponding ghost Lagrangian density for this gauge choice.

(12. The equation of motion for this ﬁeld leads to the gauge ﬁxing condition and elimination of this ﬁeld through its equations of motion leads to the familiar gauge ﬁxing Lagrangian density in (12. the ghost Lagrangian density was missing). has only one ﬁeld degree of freedom (we will discuss the question of hermiticity of the ghost ﬁelds in the next chapter).169).3 Path integral quantization of gauge theories 525 explains why the naive modiﬁcation in (12. being a scalar.170) 4 2 Here we have introduced an auxiliary ﬁeld F which does not have any dynamics. It is also true that when Maxwell’s theory is coupled to a gravitational ﬁeld.113) of the non-Abelian gauge theory failed to be suﬃcient unlike in the Maxwell theory (namely. On the other hand. we see that the equation of motion for F is given by ξF (x) = ∂µ Aµ (x). For example. This naive counting works pretty well as we will see later.169) and ask how we can recover the Lorentz gauge from the Lorentz like covariant gauge condition in (12. the ghost ﬁelds automatically couple to the geometry also. namely.171) . they anticommute even though they are scalar ﬁelds. To do this let us rewrite the eﬀective Lagrangian density (12. 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. Hence omitting the ghost Lagrangian density in such a case would lead to incorrect results. we cannot neglect them even if they are noninteracting particularly when we are doing calculations at ﬁnite temperature. (12.12.164). On the other hand each of the ghost ﬁelds. Furthermore. (The ghost degrees of freedom subtract because they have the unphysical statistics. the Aµ ﬁeld has four ﬁeld degrees of freedom.169) as ξ 1 (J) Leﬀ = − Fµν F µν + F 2 −F (∂µ Aµ )+∂µ c∂ µ c+J µ Aµ .) Let us now go back to the eﬀective Lagrangian density (12. 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.

First. In this case. in the process of eliminating the dependent ﬁeld degrees of freedom we lose manifest Lorentz covariance. As a . Here we modify the theory so that all the ﬁeld degrees of freedom are independent.172) Therefore. Namely. The Hilbert space contains only photon states of physical polarization. We maintain manifest Lorentz covariance and quantize all components of the ﬁeld as independent variables. we have the operator quantization and again there appear to be two distinct possibilities in this case.174) and describes Maxwell’s theory in the Lorentz gauge. 4 (12. in the limit ξ → 0. Thus the vector space that we work with in this case is much larger than the physical Hilbert space.173) ξ→0 4 2ξ which can also be written equivalently as (see (12.170) with ξ = 0) 1 (J) Leﬀ = − Fµν F µν − F (∂µ Aµ ) + ∂µ c∂ µ c + J µ Aµ . in the intermediate states we ﬁnd time-like photon states which contribute negatively whereas the longitudinal states contribute an equal positive amount. (12. the larger vector space of the theory contains states of indeﬁnite norm and when we do perturbation theory in this formalism. Here we explicitly eliminate the unphysical (dependent) ﬁeld degrees of freedom and then quantize the physical (independent) ﬁeld degrees of freedom. therefore. we see that there are two distinct quantization procedures. (12. we can quantize the Abelian gauge theory canonically. The second possibility is to use the GuptaBleuler quantization method to quantize the theory in a manifestly covariant manner. However. this equation leads to the Lorentz condition (or the Landau gauge) ∂µ Aµ (x) = 0. We select out the physical Hilbert space by imposing supplementary conditions on the state vectors in a Lorentz covariant way. From our discussions of the Abelian gauge theories thus far. the eﬀective Lagrangian density in the Lorentz gauge has the form 1 1 (J) Leﬀ = lim − Fµν F µν − (∂µ Aµ )2 + ∂µ c∂ µ c + J µ Aµ .526 12 Yang-Mills theory and.

as the Lagrangian density describing the . therefore. In the path integral formalism. we note that gauge invariance puts a very strong constraint on the structure of the Lagrangian density for the gauge ﬁeld. if we take Linv . their contributions cancel out and eﬀectively we are left with only two physical transverse degrees of freedom. 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. we also modify the theory (as in the Gupta-Bleuler method) so that there are no dependent variables and all components of the ﬁeld contribute to any Green’s function of the theory. The µ correct description for the generating functional. In summary. in this formalism there is no reference to the Hilbert space of the theory. the coeﬃcient matrix of the quadratic terms in the Lagrangian density is singular and.175) where AT denotes the transverse physical 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] .12. according to Faddeev and Popov. Rather the extra contributions (from the unphysical degrees of freedom) are cancelled by the Faddeev-Popov determinant which we can write as a ghost action (namely. In particular. (12. non-invertible. However. In this formalism the ﬁeld variables are treated as classical variables. is given by (J ) [AT ] µ Z[Jµ ] = N = N DAT eiS µ DAµ ∆FP [Aµ ]δ(∂ · A(x)) eiS (J ) [A µ] . The second method of quantizing the Abelian gauge theory is through the method of path integrals. (12.176) In this way.3 Path integral quantization of gauge theories 527 result. As a result. the ghost contributions cancel those from the unphysical gauge ﬁeld degrees of freedom).

As we have seen. 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. the Lagrangian density for the gauge ﬁelds is given by (see (12.177) which is invariant under the inﬁnitesimal gauge transformation 1 Dµ ǫa (x).67) and (12. consists of adding to the invariant Lagrangian density a gauge ﬁxing Lagrangian density of the form .164). In order to circumvent this diﬃculty.75)) 1 a Linv = − Fµν F µνa . On the other hand. in the non-Abelian gauge theory. Let us now apply these ideas to the study of the non-Abelian gauge theory. then we cannot deﬁne propagators and the entire philosophy of doing perturbative calculations with Feynman diagrams breaks down. g (ǫ)a Aa (x) → Aµ (x) = Aa (x) + µ µ (12. µ a Fµν (x) = ∂µ Aa (x) − ∂ν Aa (x) − gf abc Ab (x)Ac (x).528 12 Yang-Mills theory dynamics of the gauge ﬁeld theory. (12. adding a gauge ﬁxing Lagrangian density changes the theory. in general.178) where ǫa (x) is the inﬁnitesimal parameter of transformation. The standard covariant gauge ﬁxing (12. 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). 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 (12.179) ν µ µ ν with g denoting the coupling constant of the theory. 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.

12. µ µ g so that we have (see (12. we can write the ghost action corresponding to this gauge choice as (see (12. (12. 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).184) µ 1 Consequently.3 Path integral quantization of gauge theories 529 LGF = − 1 (∂µ Aµa (x))2 .183) = 1 µ ab 4 ∂ D δ (x − y) g x xµ µ = ∂x ∂xµ δab − gf acb Ac (x) δ4 (x − y). δǫb (y) (ǫ) (12. (12.181) that we are using.122)) δF a [Aµ (x)] δǫb (y) (ǫ) (12. we can write 1 (ǫ)a F a [A(ǫ) (x)] = ∂ µ Aµ (x) = ∂ µ Aa (x) + Dµ ǫa (x) . Following the prescription of Faddeev and Popov.160) and note that since ǫa is an inﬁnitesimal parameter. 2ξ (12.180) which corresponds to a gauge ﬁxing condition of the form F a [Aµ (x)] = ∂µ Aµa (x) = f a (x). rescaling the ghost ﬁelds (to absorb the factor of g ) we can write the ghost Lagrangian density for this choice of gauge ﬁxing to be .182) We note that for the covariant gauge choice (12.181) Here ξ represents a real arbitrary constant parameter known as the gauge ﬁxing parameter.

let us consider the gauge theory of the antisymmetric tensor ﬁeld Aµν (x) . In a deeper sense. As we have mentioned earlier.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 .185) where we have dropped total derivative terms (surface terms). 12. (12.186) = − Fµν F µνa − 4 2ξ We note that the ghost ﬁelds in the present case are interacting unlike in Maxwell’s theory and. 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 . However. therefore. 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 an example of how we should be careful in carrying out the Faddeev-Popov analysis in complicated gauge theories. With all these modiﬁcations. merely correspond to a multiplicative factor of unity (see (12. (12. cannot be neglected even in ﬂat space-time. in some cases we have to work through the details of this analysis rather carefully in order to obtain the correct result. in the path integral formulation. the gauge ﬁxing and the ghost Lagrangian densities.142)) which does not change the physical content of the theory.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.

2. µ.12. (12. ν = 0.4 Path integral quantization of tensor fields 531 Aµν (x) = −Aνµ (x). (12. we also note that this Lagrangian density is invariant under the gauge transformation δAµν (x) = ∂µ θν (x) − ∂ν θµ (x).192) namely. 1. This can be seen from the deﬁnition of the conjugate momenta ∂L = −F 0µν (x). Let us consider only the free Lagrangian density for this ﬁeld deﬁned by 1 L = − Fµνλ F µνλ .191) The independent ﬁeld variables of the theory are A0i .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.188) where the ﬁeld strength tensor corresponds to the totally antisymmetric tensor Fµνλ = ∂[µ A νλ] = ∂µ Aνλ + ∂ν Aλµ + ∂λ Aµν . since Fµνλ is completely antisymmetric in all the indices. 6 (12. 3.189) Naive counting shows that Aµν has six ﬁeld degrees of freedom. (12. Furthermore. ˙ ∂ Aµν (x) Πµν (x) = (12.190) so that not all ﬁeld variables are independent. Aij and we note that Πµν = −Πνµ = −F 0µν . the canonical momenta are antisymmetric (like the ﬁeld variables). However. (12. we conclude that .

However.198) This would seem like three constraints and hence we would naively conclude that this theory has no dynamical degrees of freedom. 2.193) Since these momenta identically vanish. j.197) so that we can impose a Coulomb gauge condition of the form ∂i Aij (x) = 0. 3.194) (12.196) Thus it would seem that the theory has truly three degrees of freedom. and the Lagrangian density (12.188) takes the form 1 1 L = − Fµνλ F µνλ = − 3F0ij F 0ij + Fijk F ijk 6 6 1 = − 3Πij Πij + Fijk F ijk .195) (12. we note that the theory still possesses a residual (static) gauge invariance under the transformation δAij = ∂i θj (x) − ∂j θi (x). 2. 6 (12. the corresponding ﬁeld variables are like c-number quantities and we can choose the gauge condition A0i (x) = 0. (12. 3.532 12 Yang-Mills theory or.198) actually represents only two independent conditions. on closer inspection we notice that the Coulomb gauge condition (12. Π0i = 0. However. (12. With these conditions we have ˙ Πij = −F0ij = −Aij . (12. i. i = 1. Π0µ = −F 00µ = 0. We can see this by writing out the gauge condition explicitly . k = 1.

202) where φ(x) represents a real scalar ﬁeld. 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). Since Fµνλ represent a totally antisymmetric third rank tensor in four space-time dimensions satisfying the Bianchi identity (see (12.200) It is now clear that any two of the three conditions lead to the third condition so that there are only two independent conditions. 12 2 (12. In this case. the three conditions corresponding to j = 1.12. The fact that the theory has only one degree of freedom can also be seen in the following manner. 3 are (12. ∂1 A12 + ∂3 A32 = 0.203) We recognize this as the Lagrangian density for a free. 2. 2 (12. (12. the Lagrangiandensity (12.4 Path integral quantization of tensor fields 533 ∂1 A1j + ∂2 A2j + ∂3 A3j = 0.189)). . ∂1 A13 + ∂2 A23 = 0. massless real scalar ﬁeld describing a single degree of freedom. ∂ρ Fµνλ − ∂λ Fρµν + ∂ν Fλρµ − ∂µ Fνλρ = 0. (12. Thus.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.199) ∂2 A21 + ∂3 A31 = 0.201) we can represent them also as (this is the dual of the ﬁeld strength tensor in four dimensions) 1 Fµνλ (x) = √ ǫµνλρ ∂ρ φ(x).

Hence to write the determinant as an action we need ghost ﬁelds which carry a vector index and we have .534 12 Yang-Mills theory Let us now look at the path integral quantization of this theory.207) (θ) δF ν [Aµσ (x)] δθ λ (y) (θ) θ=0 µ ν ν µ = ∂xµ δλ ∂x − δλ ∂x δ4 (x − y) ν = (δλ x ν − ∂xλ ∂x ) δ4 (x − y). 2ξ LGF = − (12.206) 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)). 6 L=− (12. According to our earlier discussions. so that we have (12. the starting gauge invariant Lagrangian density has the form 1 Fµνλ F µνλ .205) corresponding to the choice of gauge condition F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). (12. (12.208) This is a matrix with two vector indices.204) and we add to it the covariant gauge ﬁxing Lagrangian density (ξ is the gauge ﬁxing parameter) 1 (∂µ Aµν )2 .

(12. (12.211) eﬀective ﬁeld degrees of freedom.210) Let us now try our naive counting of degrees of freedom of the theory. they subtract out ﬁeld degrees of freedom. The ghost ﬁelds being vectors have four degrees of freedom each and since they satisfy anticommutation relations.212) .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 .209) Therefore. The ﬁeld variable Aµν has six degrees of freedom. Thus the eﬀective Lagrangian density appears to have 6 − 2 × 4 = 6 − 8 = −2. 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. 2 (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 ) . Let us start with the generating functional (J ) [A Z [Jµν ] = N DAµν eiS µν ] . (12.12.

as we have seen (see (12. (12.215) Thus inserting this identity element into the functional integral.154)) (θ) ∆FP [Aµν ] x dθ σ (x)δ F ν [Aµλ (x)] − f ν (x) = 1.536 12 Yang-Mills theory The gauge symmetry allows us to choose a gauge condition and we have chosen a Lorentz like gauge condition.209) 1 2 ).216) Sghost = d4 x (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) . we introduce the identity element as (see (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)] .214) where. (12. (12. (12. (θ) ∆FP [Aµν ] = det = det ( (12.213) Therefore. F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). 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 (12.152)). δF ν [Aµλ (x)] δθ σ (y) ν x δσ θ=0 ν − ∂x ∂xσ ) δ4 (x − y) .217) In the naive application of the quantization procedure. (12.218) .

(9. (12.222) We see that because of the transverse projection operator. (12. we note that the gauge condition ∂µ Aµν (x) = f ν (x).224) so that we can write . (12.223) Thus one has to use a gauge ﬁxing condition and we choose again a covariant gauge condition 1 1 2 F [fµ (x)] = ∂µ f µ (x) = f (x). (12. implies that ∂ν ∂µ Aµν = ∂ν f ν (x) = 0.4 Path integral quantization of tensor fields 537 However. (12. We take into account the transverse nature of f µ (x) exactly like the Maxwell ﬁeld and we apply the ’t Hooft trick by integrating over a transverse weight factor i − 2ξ Df µ δ (F ν [Aµλ (x)] − f ν (x)) e µν R d4 xd4 y fµ (x)P µν (x−y)fν (y) .221) has a gauge invariance (just like Maxwell’s theory) of the form δfµ (x) = ∂µ θ(x). for example.219) That is.220) (12.123)) as µν µ ν ∂x ∂x x P (x − y) = η µν − δ4 (x − y).12. 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. the action for fµ in (12.221) where P (x − y) is the normalized transverse projection operator deﬁned earlier (see.

228) det xδ 4 (x − y) 1 2 Dc(x) eiS ghost .Fν [Aµλ ])− 2 (c. (12.f µ )− 2 (c.P ×∆FP [fµ ] e− 2ξ (F (fµ ).f ) Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ . c) .226) Thus using the ’t Hooft trick we can write Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .227) Furthermore.538 12 Yang-Mills theory ∆FP [fµ ] x (θ) dθ(x)δ F [fµ (x)] − f (x) = 1. Substituting this into (12.229) .225) where δF [fµ (x)] δθ(y) (θ) 1 2 ∆FP [fµ ] = det θ=0 = det xδ 4 (x−y) .F (fµ )) i µν fν ) = Df µδ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ . remembering that (see (12.P × = i µν fν ) Df ∆FP [fµ ] δ F [fµ (x)] − f (x) e i i − 2ξ (f. c) = − 2 2 d4 x c(x) c(x). (12. (12.227) 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 .f i µ) ∆FP [fµ ].226)) ∆FP [fµ ] = = we determine 1 1 S ghost = − (c. (12. (12. c) i µλ ].

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. In ﬂat space-time this ghost Lagrangian density can be seen to give a total divergence and.F µ[Aλσ ]) J +S GF +Sghost +S ghost ).231) DAµν Dcµ Dcν Dc ei(S DAµν Dcµ Dcν Dc ei(S J +S ghost +S ghost i ) e− 2ξ (Fµ [Aλσ ].232) Counting the ﬁeld degrees of freedom we see that the eﬀective number of degrees of freedom seems to be 6 − 2 × 4 − 1 = −3. .12. Thus using this modiﬁed ’t Hooft weighting factor. but in the presence of gravitation it cannot be written as a total divergence and is quite relevant. (12. (12.230) SGF = − = − Therefore. This is diﬀerent from the usual Faddeev-Popov ghosts in the sense that the Faddeev-Popov ghosts seem to come in pairs.233) This is again not right and the agreement with canonical quantization seems to be worse than before.4 Path integral quantization of tensor fields 539 Note here that the ﬁeld c(x) is a real anticommuting scalar ﬁeld. (12. therefore. (12. The ghosts of the present form are known as Nielsen ghosts. 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 . may be neglected.

235) and ∆FP [cµ ] x dλ(x) δ F [c(λ) (x)] −f (x) = 1.237) We are now ready to calculate the Faddeev-Popov determinants associated with these gauge ﬁxing conditions. namely.238) = ∂µ cµ (x) = f (x). µ (12. (12. the ghost Lagrangian density also possesses a gauge invariance. ∆−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. the ghost Lagrangian density Lghost is invariant.234) where λ(x) and λ(x) are anticommuting scalar parameters. (12.236) Let us choose for simplicity Lorentz like gauge conditions F and F = ∂µ cµ (x) =f (x). λ=0 . and δcµ (x) = ∂µ λ(x). We can again use the FaddeevPopov trick and write ∆FP [cµ ] x (λ) dλ(x) δ F [cµ (x)] − f (x) = 1. (12. under δcµ (x) = ∂µ λ(x).239) (λ) = det δF [cµ (x)] δλ(y) .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. (12.

e .238) correspond to fermionic conditions. (12. (12. in the present case we can write 4 −1 ∆FP [cµ ] = = det xδ (x − y) c e) c Dc Dc e−i(e.240) where we have used the fact that for anticommuting variables. namely. we cannot use weight factors of the forms i − 2χ e R e e d4 x f (x)f (x) . Rather. We note that contrary to the usual case. the Jacobian for a change of variables is the inverse of the determinant.12. f and f are Grassmann (fermionic) functions. . Therefore. the Faddeev-Popov term is an inverse determinant and.241) We note here that since we are writing an inverse determinant as an action. (12.4 Path integral quantization of tensor fields 541 so that (λ) −1 λ=0 −1 ∆FP [cµ ] = = δF [cµ (x)] det δλ(y) det xδ 4 (x − y) . the appropriate weight factor in this case would correspond to .242) The gauge conditions (12. Namely.243) since the exponents vanish (because of the fermionic nature of the variables).237) and (12. Similarly we can show that 4 −1 ∆FP [cµ ] = = det xδ (x − y) Dc Dc e ” “ e c c e −i e. the ghost ﬁelds c and c behave like commuting scalars. (12. e i − 2χ R e e e e d4 x f (x) f (x) . consequently. the ’t Hooft trick needs to be carried out rather carefully.

cµ : −4 degrees of freedom. c : 1 degrees of freedom. (12.542 12 Yang-Mills theory e i − 2χ R e e e d4 x f (x) f (x) . commuting. 2χ c (12. c : 1 degrees of freedom.248) The ghosts c. The other . (12. (12. c. c : 1 degrees of freedom.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 and c being commuting scalars contribute positively to the counting of the number of degrees of freedom. cµ : −4 degrees of freedom.245) d4 x Leﬀ . c : −1 degrees of freedom. c : 1 degrees of freedom.244) the generating functional takes the form Z [Jµν ] = N where Seﬀ and Leﬀ = DAµν Dcµ Dcν DcDc Dc D c D c eiSeﬀ . Aµν : 6 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.247) We are now ready to count the number of eﬀective ﬁeld degrees of freedom in the theory. anticommuting. (12.

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

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

the total Lagrangian density for the non-Abelian gauge theory (Yang-Mills theory) has the form LTOT = Linv + LGF + Lghost .2) is invariant under the global transformations 545 . remembers the gauge invariance of the original theory.1) 2 + ∂ µ ca (Dµ c)a . The total Lagrangian density has been gauge ﬁxed and.186). in some sense.3) where g denotes the coupling constant of the theory.67)) (Dµ c)a = ∂µ ca − gf abc Ab cc . µ (13. the total Lagrangian density. It is easy to check that the total Lagrangian density (13.Chapter 13 BRST invariance and its consequences 13.2) Here ξ represents the arbitrary gauge ﬁxing parameter and the covariant derivative in the adjoint representation is deﬁned as (see (12. with gauge ﬁxing and ghost terms.1 BRST symmetry As we have seen in (12.181) is given by 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ (13. which for the covariant gauge in (12. develops a global fermionic symmetry which. does not have the gauge invariance (12.34) of the original theory. However. (13. therefore.

= 6 (13.2 respectively. we obtain δ ∂ µ Aa = µ ω µ ∂ Dµ ca = 0. (13. 2 ω ∂ µ Aa . while for ca it is µ true only on-shell (namely.4) where ω is an arbitrary anti-commuting constant parameter of the global transformations. .4) we have δ2 δ1 φa = 0.546 13 BRST invariance and its consequences δAa = µ δca δca ω (Dµ c)a . ca . We note that the nilpotency of the transformations holds oﬀ-shell only for the ﬁelds Aa . g ω abc b c = f cc.2) can be seen by ﬁrst noting that (recall that the ghost ﬁelds are Grassmann variables) δ (Dµ ca ) = Dµ δca − gf abc δAb cc µ = ω Dµ f abc cb cc − ωf abc Dµ cb cc = 0. g (13.50)). ca independent of the parameters of transµ formations where δ1.5) Here we have used the Jacobi identity for the symmetry algebra (see (12.6) when the ghost equation of motion is used. only when the ghost equation of motion is used). 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 = 0.7) for the ﬁelds φa = Aa . Similarly. = − µ gξ (13. The invariance of the Lagrangian density (13. This shows that under two successive transformations of the kind (13. ca .2 correspond to transformations with the parameters ω1.

Since this will also be quite useful for our latter discussions.2) can also be written equivalently as .34)) with the parameter ǫa (x) = ωca (x) and. ν gξ (13.9) so that the action is invariant. let us recall that the gauge ﬁxing Lagrangian density in (13. This shows that the action for the total Lagrangian density (13.1 BRST symmetry 547 The invariance of the Lagrangian density (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. therefore. δLinv = 0. we need to worry only about the changes in the gauge ﬁxing and the ghost Lagrangian densities which lead to 1 δ (LGF + Lghost ) = − (∂ ν Aa ) ∂ µ δAa + (∂ µ δca ) Dµ ca ν µ ξ ω µ ν a ω ∂ (∂ Aν ) Dµ ca = − (∂ ν Aa ) ∂ µ Dµ ca − ν gξ gξ ω = −∂ µ ∂ ν Aa Dµ ca . this is a reﬂection of the fact that the theory is lacking in some auxiliary ﬁeld variables and once the correct auxiliary ﬁelds are incorporated the symmetry algebra will close oﬀ-shell (without the use of equations of motion).13. Generally. we have used the fact that δ (Dµ c)a = 0 which we have seen in (13.170) we can write the gauge ﬁxing Lagrangian density by introducing an auxiliary ﬁeld. In this derivation.4) with an anti-commuting constant parameter. namely. The present formulation of the BRST symmetry.2) under the transformations (13. however.5). the invariant Lagrangian density is trivially invariant under these transformations.4) can now be easily checked. First. (13.2) is invariant under the global transformations (13. As we have seen earlier within the context of the Abelian gauge theory in (12. is slightly unpleasant in the sense that the nilpotency of the anti-ghost ﬁeld transformation holds only on-shell. we note that the transformation for Aa can really be thought of as an inﬁnitesiµ mal gauge transformation (see (12.8) Consequently.

g ω abc b c = f cc.12) µ 4 2 In this case. µ 2 (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 . δ2 δ1 φa = 0. but it is this form that is very useful as we will see. (The form of the gauge ﬁxing Lagrangian density in (13.170) by a total divergence. µ (13. namely.10) diﬀers from that in (12. 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.) It is clear from the form of LGF in (13. we recover the original gauge ﬁxing Lagrangian density (up to a total divergence term). the BRST transformations in (13.548 13 BRST invariance and its consequences LGF = ξ a a F F + (∂ µ F a )Aa .10) that the equation of motion for the auxiliary ﬁeld takes the form ξF a = ∂ µ Aa .11) and when we eliminate F a from the Lagrangian density using this equation.13) δF a = 0. (13. g (13. (13.4) take the form δAa = µ δca δca ω (Dµ c)a .14) .10) where F a is an auxiliary ﬁeld. 2 ω = − F a. and it is straightforward to check that these transformations are nilpotent oﬀ-shell.

g ω F a + gf abc cb cc . replace the original gauge invariance of the theory and play a fundamental role in the study of non-Abelian gauge theories. F a . which can also be easily checked to leave the total Lagrangian density (13. g g (13. F a represents µ the missing auxiliary ﬁeld that we had alluded to earlier. which deﬁne a residual global symmetry of the full theory. ca . We note that Linv is invariant under the BRST transformation as we had argued earlier in (13. There is also a second set of fermionic transformations involving the anti-ghost ﬁelds of the form ω (Dµ c)a . we will not pursue this symmetry further in our discussions.9)).4). These are known as the anti-BRST transformations.8) and the auxiliary ﬁeld F a does not transform at all which leads to δLTOT = δ (LGF + Lghost ) = = ∂ µ F a δAa + ∂ µ δca (Dµ c)a µ ω ω µ a ∂ F (Dµ c)a − ∂ µ F a (Dµ c)a = 0. 2 (13.1 BRST symmetry 549 for all the ﬁeld variables φa = Aa . In some sense the BRST transformations.13.12) invariant. since these do not lead to any new constraint on the structure of the theory beyond what the BRST invariance provides. Therefore. 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. g ω abc b c f cc. ca .16) δAa = µ δca = δca = δF a = −ωf abc F b cc . However.15) Unlike in the formulation of BRST variations without the auxiliary ﬁeld in (13. We note here that the BRST and the anti-BRST transformations are not quite symmetric in the ghost and the anti-ghost ﬁelds which is a .

We have already seen that the naive Gupta-Bleuler quantization does not work in the nonAbelian case. as we have emphasized several times by now. For example. we note here that these fermionic symmetries arise naturally in a superspace formulation of gauge theories. In addition to these two anti-commuting symmetries. the total Lagrangian density (13.12). We recall that the physical space must be selected in such a way that it remains invariant under the time evolution of the system. Therefore. the physical Hilbert space needs to be properly selected for a discussion of physical questions associated with the non-Abelian gauge theory. δca = −ǫca . (The fact that these transformations are like scale transformations and not like phase transformations. commuting inﬁnitesimal parameter and the generator of this symmetry transformation merely corresponds to the ghost number operator. Here ǫ represents a constant. for example. In the covariant gauge in Maxwell’s theory. (13.17) with all other ﬁelds remaining inert.181).550 13 BRST invariance and its consequences reﬂection of the asymmetric manner in which these ﬁelds occur in the ghost Lagrangian density in (13. it allows us to carry out covariant quantization of the non-Abelian gauge theory. contains many more states than the physical states alone. This is known as the ghost scaling symmetry of the theory. namely.12) is also invariant under the inﬁnitesimal bosonic global symmetry transformations δca = ǫca .2 Covariant quantization of Yang-Mills theory The presence of the BRST symmetry and the ghost scaling symmetry in the gauge ﬁxed Yang-Mills theory leads to many interesting consequences. it corresponds to the operator that counts the ghost number of the ﬁelds.2) or (13. Let us recall that the vector space of the full theory in the covariant gauge (12. Without going into details. 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.) 13. which is normally associated with the number operator. we .

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 (9.18). the generators of the BRST symmetry. On the other hand. Qc . = 0. not clear a priori if the conditions (13. “(+)”. the Gupta-Bleuler condition works because ∂ µ Aµ satisﬁes the free Klein-Gordon equation (9. are conserved and hence can be used to deﬁne a physical Hilbert space which would remain invariant under the time evolution of the system.18) where the superscript. QBRST . stands for the positive frequency part of the ﬁeld.130) in the covariant gauge (12.118). in reality it is an inﬁnite number of conditions since it has to hold for every value of the coordinates. We recognize that even though this looks like one condition. µ (13. (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. as we have seen in (12. It is. the conditions in (13. we can identify the physical space of states of the gauge theory as satisfying (we note that at this point this only deﬁnes a subspace of the total vector space and we still have to show that this subspace indeed coincides with the physical Hilbert space) QBRST |phys Qc |phys = 0. and the ghost scaling symmetry.162)) ∂ µ A(+) (x)|phys = 0.19) 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. (13.19) Note that even in the case of Maxwell’s theory. .) o Thus. does not satisfy a free equation and hence it is not a suitable operator for identifying the physical subspace in a time invariant manner. The corresponding operator in a non-Abelian theory.164) and hence the physical space so selected remains invariant under time evolution. whether they can reduce the vector space suﬃciently enough to coincide with the physical space).19) are suﬃcient to reproduce even the GuptaBleuler condition in the case of the Abelian theory (namely. In the Abelian theory. therefore.

20) where we have used the fact that the auxiliary ﬁeld does not transform under the BRST transformations and the ﬁelds Aa .22) d3 x − (∂ 0 F a )ca + F a (D 0 c)a + 0 d3 x Jc = ˙ d3 x ca ca − ca (D 0 c)a . From this.552 13 BRST invariance and its consequences To see that these conditions indeed lead to the Gupta-Bleuler condition in Maxwell’s theory. g abc µ a b c f (∂ c ) c c . 2 (13. 2 (13. g 2 µ(ǫ) Jc = δca ∂LTOT ∂LTOT + δca ∂∂µ ca ∂∂µ ca = −ǫca (D µ c)a − ǫca ∂ µ ca = ǫ ((∂ µ ca )ca − ca (Dµ c)a ) . we can obtain the BRST as well as the ghost scaling current densities without the parameters of transformation to correspond to µ JBRST = F µνa (Dν c)a + F a (D µ c)a + µ Jc = (∂ µ ca ) ca − ca (Dµ c)a . let us note that the N¨ther current o densities associated with the BRST transformation as well as the ghost scaling transformation have the forms (recall that we use left derivatives for Grassmann variables and that ω is an anticommuting parameter) JBRST (x) = (ω)µ ∂LTOT ∂LTOT a ∂LTOT a a ∂LTOT δAν + δF + δca a + δc ∂∂ ca a a ∂∂µ Aν ∂∂µ F ∂∂µ c µ = −F µνa δAa + δca (D µ c)a − δca (∂ µ ca ) ν ω g = − F µνa (Dν c)a + F a (Dµ c)a + f abc (∂ µ ca ) cb cc .21) The corresponding conserved charges can also be obtained from these current densities and take the forms QBRST = = = Qc = 0 d3 x JBRST d3 x F 0i a (Di c)a + F a (D 0 c)a + g abc ˙ a b c f c cc 2 g abc ˙ a b c f c cc . F a are inert µ under the scaling of ghost ﬁelds. (13. .

26) implies that the physical states must have (net) zero ghost number.27) . (13.25) Let us note that the condition (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 . denoting the physical states of the theory as |phys = |Aµ ⊗ |n. we note that for the Abelian theory where f abc = 0 and there is no internal index “a”. ← → d3 x c ∂ 0 c. (13. In principle.22) to correspond to ← → d3 x F ∂ 0 c. Thus. (13. (13.24) QBRST = Qc = ˙ d3 x − F c + F c = ˙ ˙ ˙ d3 x cc − cc = − If we use the ﬁeld decomposition for the ﬁelds and normal order the charges (so that the annihilation/positive frequency operators are to the right of the creation/negative frequency operators).12). n . this allows for states containing an equal number of ghost and anti-ghost particles. (13.13. the BRST and the ghost scaling charge operators can be obtained from (13. In particular.23) following from (13. the BRST charge has the explicit form (we do not show the normal ordering explicitly) QBRST = i d3 k c(−) (−k)F (+) (k) − F (−) (−k)c(+) (k) .19) Qc |phys = 0.

Thus. (13.11)). we feel conﬁdent that the physical state conditions in (13.19) really select out the subspace of physical states. we note that if the physical states have to further satisfy the condition QBRST |phys = i d3 k c(−) (−k)F (+) (k) − F (−) (−k)c(+) (k) |Aµ ⊗ |n.31) where we have used the equation of motion for the auxiliary ﬁeld (see (13. and F (+) (k)|Aµ = 0.30) (13. 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).554 13 BRST invariance and its consequences where n.12) that we can obtain the canonical momenta conjugate to various ﬁeld variables of the theory to correspond to . (13. n denote the (equal) number of ghost and anti-ghost particles. then this implies that (13. n = 0. |phys = |Aµ ⊗ |0. 0 = |Aµ . n = 0.28) c(+) (k)|n.29) Namely. 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. To investigate systematically whether the physical state conditions in (13. and must further satisfy (with k0 = |k|) F (+) (k)|phys = 0 = kµ Aµ(+) (k) |phys . we note from the form of the Lagrangian density (13.19) are the right ones even for the non-Abelian theory.

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

the conserved charges (13. The algebra of the conserved charges also follows in a straightforward manner from (13. Qc ] = −ica (x).22) are Hermitian with the assigned hermiticity conditions (13. The third is a statement of the fact that the BRST charge carries a ghost number of unity. c c 2 a d3 x (Πa ca + ca Πc ) .556 13 BRST invariance and its consequences and we note that the currents (13. The second equation implies that the ghost scaling symmetry is Abelian.38) simply reiterates in an operator language the fact that the BRST transformations are nilpotent (see (13. We note.34) for the ghost ﬁelds. The conserved charges (13.33) [QBRST . Therefore.36) and we can calculate the algebra of charges as well as various other relations of interest using the (anti) commutation relations (13.36) and (13.17)) [ca (x).13) and (13. for example. that (this basically describes the behavior of the ghost ﬁelds under a scaling (13.38) is that . An immediate consequence of the algebra in (13.33) for the ﬁeld variables. [Qc .14)). therefore.38) This algebra can also be derived directly from the transformation laws for the ﬁeld variables in (13. Qc ] = 0. Qc ] = −iQBRST . QBRST ]+ = 2Q2 BRST = 0.22) can now be expressed in terms of the ﬁelds and the conjugate momenta as (before integrating the ﬁrst term by parts in QBRST . Qc ] = ica (x).37) [ca (x).17). [QBRST . The ﬁrst equation in (13. we see that we can think of iQc as the ghost number operator (with the ghost number for ca being negative).21) and.22)) g d3 x Πia (Di c)a − F a Πa + f abc Πa cb cc . (13. (13. c QBRST = − Qc = − (13. see (13.

The ﬁeld operator which creates such a state must necessarily commute with QBRST and.39) which is quite useful (particularly) in dealing with gauge theories at ﬁnite temperature. then. (13.12) contains various unphysical states as well as states with negative norm in addition to the physical states. if a state |Ψ cannot be written as ˜ |Ψ = QBRST |Ψ . for every |Ψ ∈ Vphys . [QBRST . therefore.2 Covariant quantization of Yang-Mills theory 557 QBRST eπQc = eπ(Qc −i) QBRST = −eπQc QBRST . H] = 0. (13. Namely.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. namely. eπQc + = 0. (we assume that the vacuum state is a BRST singlet (invariant) state) . or.40).42) ˜ for some |Ψ and still satisﬁes the physical state condition. QBRST . we can select out a subspace Vphys by requiring that states in this space are annihilated by QBRST . must represent a truly gauge invariant ﬁeld variable. As we have discussed earlier. (13. it is truly a BRST singlet (invariant) state. the total vector space V of the complete gauge theory (13. As is clear from our earlier discussions. the metric of this space and the inner product become indeﬁnite and a probabilistic description of the quantum theory is lost unless we can restrict to a suitable subspace with a positive deﬁnite inner product. (13.13. First. we have QBRST |Ψ = 0.41) We note that there are two possible kinds of states which will satisfy the physical state condition (13. Consequently.

46) We see that the physical state condition is trivially satisﬁed in this case because of the nilpotency of QBRST (see (13. therefore.558 13 BRST invariance and its consequences QBRST |Ψ = QBRST Ψ|0 = [QBRST . (13.40) can be written in the form ˜ |Ψ = QBRST |Ψ .38)).) ˜ ˜ Ψ|Ψ = Ψ|QBRST QBRST |Ψ = 0. correspond to truly physical states of the theory and since gauge invariant degrees of freedom have physical (non-negative) commutation relations. 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. it can be written as the BRST variation of the anti-ghost ﬁeld and. also satisﬁes the above relation. Ψ] |0 = 0.) The second class of states which would satisfy the physical state condition (13. (13. namely. therefore. Ψ] = 0. (Note that the auxiliary ﬁeld does not transform under a BRST transformation and. has a diﬀerent character. (13. Ψ|0 = |Ψ .43) where we have assumed that the operator Ψ creates the state |Ψ from vacuum.45) and transverse ﬁelds would represent such operators which are gauge invariant asymptotically. as we have seen in (13. However. such states would have positive norm. This implies that (13. (13.34) for the ghost ﬁelds.13).44) [QBRST . therefore.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 . Such states would.

which satisfy the physical state condition and are of the ﬁrst kind can. (13. be identiﬁed as belonging to the V quotient space V phys = phys and will have positive deﬁnite norm. by unphysical. then ˜ Ψ′ |Ψ = Ψ′ |QBRST |Ψ = 0. we mean states containing quanta of the ghost ﬁelds. be decomposed into states containing ﬁxed numbers of unphysical particles. If we denote all the states of Vphys of the second kind by V0 . the longitudinal components of Aa and the auxiliary ﬁelds. this would contain all the zero norm states which would be orthogonal to Vphys itself.13) 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 (13.40) because if |Ψ represents a physical state (of either kind) while |Ψ′ is a state of the second kind. we note that we can write P (0) = ½ − P ′ .13) through (anti) commutation with the appropriate ﬁeld variables) F a ∼ [QBRST .2 Covariant quantization of Yang-Mills theory 559 (13. (13. (13. ca ]+ . therefore.49) that P (0) projects onto the truly physical states of the theory (namely. we have ∞ n=0 P (n) = ½.51) . µ We recall from (13.50) It is clear from the deﬁnition in (13. The true physical states. (13. deﬁning P (n) as the projection operator onto the n-unphysical particle sector.48) which follows from the physical state condition.13.49). then.49) P (n) P (m) = δnm P (n) . V0 Every state in Vphys can. Thus. of course. with no unphysical particles) and from the completeness of the projection operators in (13. Here.

R]+ . R(n) . What is interesting.53) However. but this is not very illuminating. whose explicit form is not important for our discussions) P ′ = [QBRST . of course. for n ≥ 1.52) We can.54) (13. then. we can even show that for every P (n) . R]+ = 0. In fact. and. P ′ = 0. therefore. (13. P ′ = QBRST .55) which follows from the nilpotency of QBRST . n ≥ 1. however. it must have the form (for some fermionic operator R. P (0) QBRST . that P ′ must also commute with QBRST . therefore. project onto the zero norm space V0 when acting on states |Ψ . It follows. must be gauge invariant and will commute with QBRST . (13. cannot be truly gauge invariant. Namely. Consequently. (13. 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 can write P (n) = QBRST . we have QBRST . = 0. [QBRST . therefore. so that QBRST .56) + P′ It is now easy to show that P (n) . construct the actual forms of all the projection operators explicitly.560 where ∞ n=1 13 BRST invariance and its consequences P = ′ P (n) . (13.

58) (13. shows that Vphys .40). We note that. and the norm of any such state is. (13.34). therefore. R]+ |Ψ = 0. 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. which follows from the physical state condition (13. We will only outline the proof of unitarity in this section. (13. This also makes it clear that the V norm of a state |Ψ ∈ V phys = phys would be positive deﬁnite. then. obtained to be Ψ|Ψ = Ψ|P (0) |Ψ ≥ 0. then. 13.13. deﬁned by the physical state condition. The question of the unitarity of the S-matrix in a gauge theory can be formulated in the following manner. R(n) |Ψ = 0.40) (as well as the hermiticity of QBRST ). This V0 would correspond to the true physical subspace of the total vector space. 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 (13.59) This. It leads to a formal proof of unitarity of the theory when restricted to the subspace of the physical Hilbert space. |Ψ′ ∈ Vphys . Any vector |Ψ ∈ Vphys can now be written as |Ψ = P (0) |Ψ + P ′ |Ψ . For example.19). with the hermiticity assignments for the ghost ﬁelds in (13. let |Ψ . the total Lagrangian .57) + Ψ′ | [QBRST .3 Unitarity 561 which satisfy the physical state condition (13.3 Unitarity The BRST invariance of a gauge theory is quite important from yet another consideration. Ψ′ |P (n) |Ψ Ψ′ |P ′ |Ψ = = Ψ′ | QBRST .

(13. we can deﬁne a unique state |Ψ ∈ V phys as |Ψ = P (0) |Ψ . so that [QBRST . 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 . This question can be systematically analyzed as follows. (13.63) where we note that in the space V phys .62) which follow from (13. |Φ ∈ Vphys and any two operators A and B acting on Vphys (i.562 13 BRST invariance and its consequences density (13. (13. namely. 2 (13. = = Ψ|AP (0) B|Φ Ψ|A(½ − P ′ )B|Φ = Ψ|AB|Φ . namely. Furthermore. the S-matrix is BRST invariant since the full theory is. (13. S † S = SS † = ½. P (0) acts as the identity operator.60) Furthermore. Let us note here from the discussions of the last section that for any |Ψ .57). any two operators which do not take us out of Vphys ) Ψ|P (0) |Φ Ψ|(½ − P ′ )|Φ = Ψ|Φ .64) The two states.12) is Hermitian so that the S-matrix of the theory is formally unitary. P (0) |Ψ = P (0) |Ψ = P (0) |Ψ = |Ψ .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. |Ψ and |Ψ .. we note that given any state |Ψ ∈ Vphys .

70) . (13. it will leave the space Vphys invariant and hence Sphys will not take a state out of V phys . 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) . 2 (13.62) as well as the formal unitarity of S (13. we can also show that † Sphys Sphys = ½.65) where the last line follows from the ﬁrst of the conditions in (13. In a similar manner. Given this. also automatically leads to a formal proof of unitarity of the S-matrix in the subspace of the truly physical states of the theory. (13. (13.60).69) In other words. It follows from (13. (13. the physical state condition (13.66) such that for |Ψ ∈ Vphys . |Ψ = P (0) |Ψ ∈ V phys as deﬁned in (13.62).68) that † Sphys Sphys = ½.63).3 Unitarity 563 Ψ|Φ = Ψ| P (0) |Φ = Ψ|P (0) |Φ = Ψ|Φ . we now have † Ψ|Sphys Sphys |Φ † Ψ|P (0) Sphys Sphys P (0) |Φ = = = Ψ|S † P (0) S|Φ Ψ|S † S|Φ = Ψ|Φ = Ψ|Φ .13. Furthermore. (13. Sphys |Ψ = Sphys P (0) |Ψ = P (0) S|Ψ = S|Ψ .40) which naturally follows from the BRST invariance of the theory.66) and (13.68) where we have used (13.67) It is clear that since S is BRST invariant.

namely. It now follows from the physical state condition (13.40) that phys| (LGF + Lghost ) |phys′ = −g phys| QBRST . we can show that the gauge ﬁxing and the ghost Lagrangian densities lead to no physical consequences. ξ a a c F + ∂ µ ca Aa µ 2 + |phys′ (13. we note that these extra terms in the Lagrangian density (namely. the gauge ﬁxing and the ghost Lagrangian densities) can. = g QBRST . 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. − ca F a − ∂ µ ca Aa µ 2 + (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. in fact. = 0.72) This shows that the terms added to modify the original Lagrangian density have no contribution to the physical matrix elements of the theory.564 13 BRST invariance and its consequences As another consequence of the BRST invariance of the theory. To show this. be written as a BRST variation (with the parameter of transformation taken out). 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 . 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) .

4 Slavnov-Taylor identity The BRST invariance of the full theory leads to many relations between various scattering amplitudes of the theory. 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.75) . the vacuum functional is given by 0|0 J = Z[J] = eiW [J] = N Dφ eiS J [φ] . However. 13. 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. let us recapitulate brieﬂy some of the essential concepts from path integrals (a detailed discussion is beyond the scope of these lectures).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 . ca F a ]+ |0 J = 0. (13. as we have seen in (12.74) where S[φ] denotes the dynamical action of the self-interacting scalar ﬁeld.7. 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). (13.13.120). (13.73) 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. We have already given a simple diagrammatic derivation of such identities in the case of QED in section 9. before we go into the actual derivation. For simplicity. However. In this case.

77) in the path integral formalism corresponds to a unique Feynman diagram in perturbation theory. δJ(x) Z[J] δJ(x) (13. Z[J] and W [J] are known as the generating functionals for the Green’s functions of the theory.79) .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 .77) involve only Feynman diagrams where all the parts of the diagram are connected. This is more fundamental since the general Green’s functions can be constructed in terms of these. 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.77) = φc (x) = (−i) δZ[J] δW [J] = . the Green’s functions calculated from W [J] (the logarithm of Z[J]) through (13. let us note here that each term in the calculation of the Green’s function in (13. For example. In particular we note that 0|φ(x)|0 J δn W [J] δJ(x1 ) · · · δJ(xn ) J=0 .49)) and is a functional of J. (13. (13.76) and (13. The connected two point time ordered Green’s function which also corresponds to the Feynman propagator of the theory is similarly obtained from (13.76)) includes contributions of all diagrams (including the ones that are not connected) to a Green’s function. On the other hand. On the other hand.566 13 BRST invariance and its consequences As we have already mentioned in the last chapter.78) is known as the classical ﬁeld (see (7. (13. To clarify this distinction further. 0|T φ(x1 ) · · · φ(xn ) |0 = δn Z[J] (−i)n Z[J] δJ(x1 ) · · · δJ(xn ) . In general.76) J=0 deﬁnes the n-point time ordered Green’s function of the theory. the n-point Green’s functions would involve Feynman diagrams which consist of parts that are not connected and Z[J] (through (13.

let us consider the eﬀective Lagrangian density (at the lowest order or tree level) which consists of LTOT for the Yang-Mills theory (13. With these basic ideas from path integrals. the more fundamental concept in perturbation theory is the diagram without the external propagators known as the vertex function.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 .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. 2 (13. However.80) It can now be checked using (13. (13. The generating functional Γ[φc ] is known as the eﬀective action of the theory (including all quantum corrections) and generates the proper (1PI) vertex functions of the theory. 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. These are quite fundamental in the study of quantum ﬁeld theory and the generating functional which generates such vertex functions is constructed as follows.4 Slavnov-Taylor identity 567 The connected n-point time ordered Green’s functions consist of connected Feynman diagrams with external lines (propagators).82) . Let us Legendre transform W [J] as (this is like going from the Lagrangian to the Hamiltonian) Γ[φc ] = W [J] − d4 x J(x)φc (x).13.81) = −J(y).

The eﬀective Lagrangian density is no longer invariant under the BRST transformations (13. δK a (13.84) Here. recalling that LTOT in (13.13) when the external sources are held ﬁxed. for example. δK µa J g abc b c f c c |0 2 = g abc b c f cc 2 = δW [J] . we have assumed the convention of left derivatives for the anticommuting ﬁelds. The usefulness of this will become clear shortly. for the variations in (13.13) of the ﬁelds Aa . Denoting all the ﬁelds and the sources generically by A and J respectively.82) is BRST invariant and that the BRST transformations are nilpotent. we have not only introduced sources for all the ﬁeld variables in the theory. δη a 0| (Dµ c)a |0 0| = (Dµ c)a J = δW [J] . It is worth noting here that the source K µa is of fermionic nature (in addition to the fermionic sources η a .49)) and in what follows we will ignore the superscript (c) for notational simplicity. K a ) for the composite variations under the BRST transformation (namely. (7. in the presence of sources. δJ µa δW [J] . ca which are nonlinear in µ the ﬁeld variables). we can write the generating functional for the theory as R d4 x Leﬀ 0|0 J = Z[J] = eiW [J] = N DA ei . In fact.83) The vacuum expectation values of operators. (13. we obtain the change . The ﬁelds A(c) are known as the classical ﬁelds (see. η a ). but we have also added sources (K µa . 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] . can now be written as δW [J] . 0|F a |0 J = F a J = F (c)a = δJ a δW [J] 0|ca |0 J = ca J = c(c)a = (−i) .568 13 BRST invariance and its consequences Here.

87) can also be written as δW δW δW = 0. − iη a (x) + iη a (x) a µa (x) a (x) δK δK δJ (x) (13. (13. This immediately leads from (13.87) +iη a (x) The functional integration measure can be easily checked to be invariant under such a fermionic transformation and using (13.86) the generating functional should be invariant (namely.13. since it does not depend on the ﬁeld variables.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 .75) we note that (13. Therefore. It is here that the .85) to (see also (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 .85) g 2 We note that the generating functional is deﬁned by integrating over all possible ﬁeld conﬁgurations. if we redeﬁne the ﬁelds under the path integral as. it should not change under any ﬁeld redeﬁnition). (13. δJ a (x) (13. A → A + δBRST A.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 is clear that δΓ δAa µ δΓ δF a δΓ δca δΓ δca δΓ a δKµ δΓ δK a = −J µa . K] − d4 x J µa Aa + J a F a + i (η a ca − ca η a ) . we see that we can rewrite the master equation (13. we are interested in the identities satisﬁed by the proper (1PI) vertices of the theory. Most often. = = δW . From the deﬁnition of the generating functional for the proper vertices (see (13. we have Γ[A. µ (13.90) Using these deﬁnitions.81)). = iη a . K] = W [J.89) where K stands generically for the sources for the composite BRST variations.570 13 BRST invariance and its consequences usefulness of the sources for the composite BRST variations becomes evident. 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 . however.88) in terms of the generating functional for the proper vertices as . a δKµ δW .80) and the ﬁeld variables are really the classical ﬁelds and we are dropping the superscript (c) for simplicity). = −J a . δK a (13.

92) and setting all Taking derivative of this with respect to ﬁeld variables to zero gives δ2 Γ δ2 Γ δ2 Γ = 0.13.) . The BRST invariance. (We would discuss renormalization of ﬁeld theories in a later chapter. is very fundamental in the study of gauge theories as far as renormalizability and gauge independence of physical observables are concerned. let us note that we can write the master identity (13. (13. This is essential in proving the renormalizability of gauge theories. a (x) δK µa (x) a (x) δAµ δc (x) δK δc (x) (13. δ2 δF b (p)δcc (−p) d4 k (13.93) − 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. follow from the BRST invariance of the theory. 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.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. Thus.4 Slavnov-Taylor identity 571 d4 x δΓ δΓ δΓ δΓ δΓ + a − F a (x) a = 0. For example. the counter terms (quantum corrections) should satisfy such a relation. consequently. in this way. like the Slavnov-Taylor identities. 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. These are the essential elements in carrying out any perturbative calculation in a quantum ﬁeld theory.96) take the forms .) The inverses of the two point functions are easily calculated in the momentum space where the quadratic operators in (13. We note that since gauge ﬁxing breaks gauge invariance.95) LQ = − = = 1 a µν A η 2 µ 1 a µν ab b A O Aν + ca M ab cb . − 1− (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).2) 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ 2 + ∂ µ ca Dµ ca . (13. (They are the Green’s functions of the free theory. Let us recall that the total Lagrangian density after gauge ﬁxing (in the covariant gauge) has the form (13. 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. 2 µ where we have neglected total divergence terms and have identiﬁed Oµν ab = δab η µν 1 µ ν ∂ ∂ .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.572 13. ξ M ab = −δab .

λ ξ (13.99) where α.5 Feynman rules 573 M ab (p) = δab p2 . Therefore. or. With this parameterization. λ or. p2 (13.97) The inverse of M ab (p) in (13.101) 1 β µ (αp2 + 1)δµ − α(1 − ) − p pλ = 0.98) while the derivation of the inverse of Oµν ab (p) is a bit more involved. we can parameterize the most general form of the inverse as O−1ab (p) = δab α ηµν + β µν pµ pν .100) Substituting the parameterization (13. let us derive this inverse systematically. we note that the inverse is deﬁned to satisfy (repeated indices are summed) −1bc Oµν ab (p)Oνλ (p) = δac δµ . λ ξ ξ . ηµν ). p2 (13. λ αp2 δµ − α 1 − λ 1 1 µ p pλ + β pµ pλ − β 1 − pµ pλ = −δµ . β are arbitrary parameters to be determined (they are not necessarily constants and can be Lorentz invariant functions of the momentum).13. Oµν ab (p) = −δab η µν p2 − 1 − 1 µ ν p p . 1 µ ν p p ξ αηνλ + β pν pλ p2 = −δµ . ξ (13.99) for the inverse and carrying out the multiplication explicitly we have η µν p2 − 1 − or. λ ξ ξ (13. Let us note that −1ab the Green’s function Oµν (p) is a symmetric second rank tensor (of rank 2) and with all the Lorentz structures available (pµ . η µν p2 − 1 − 1 µ ν −1ac p p O νλ (p) = −δac δµ .97) is quite simple (it is like the propagator of a massless scalar ﬁeld) M −1ab (p) = δab .

(13. . ) represents the integral over p and we deduce from this that (see (13. p2 (13. ξ ξ 1 (ξ − 1).126)) h −1ab i − 2 J µa (−p). 2 p p −1ab Oµν (p) = − (13.104) where ( . 2 p p (13.102) α=− α 1− or. p2 1 β − = 0.Oµν (p)J νb (p) +i η a (−p).99) as pµ pν δab ηµν + (ξ − 1) 2 . sometimes the gauge propagator is also denoted by ab Dµν (p)) δ2 Z0 (−i)2 Z0 δJ µa (−p)δJ νb (p) iGab (p) = µν J µa =η a =η a =0 −1ab −1ab = (−1) − iOµν (p) = iOµν (p) = − iδab pµ pν ηµν + (ξ − 1) 2 .574 13 BRST invariance and its consequences Setting the coeﬃcients of each distinct Lorentz structure to zero. it follows that 1 .103) With these.79). see (12.105) 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.M −1ab (p)ηb (p) Z0 = N e i . β = α(ξ − 1) = − so that we can write the inverse in (13. we can write the generating functional for the free theory as (involving only the quadratic terms of the action as well as sources for the ﬁelds.

108) = iGab (p) = We can now write the complete generating functional as Z[J] = exp i d4 x Lint 1 δ i δJ Z0 [J].5 Feynman rules 575 When ξ = 0. p2 (13. µ . while for ξ = 1.109) 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 propagators can be diagrammatically represented as = iGab (p) µν = − a p b (13.13. a p ν.110) −gf abc Aa ∂ µ cc cb ..107) µ. p2 (13. the gauge is known as the Feynman gauge where the propagator has a much simpler form which is more suitable for perturbative calculations.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 . b pµ pν iδab ηµν + (ξ − 1) 2 . (13. 2 p p iδab . when we are in the Landau gauge this propagator is transverse.106) which deﬁnes the ghost propagator (which is sometimes also denoted by Dab (p)). i.

p3 ) = i δ3 S δAa (p1 )δAb (p2 )δAc (p3 ) µ ν λ = i δ3 Sint δAa (p1 )δAb (p2 )δAc (p3 ) µ ν λ = (2π)4 gf abc δ4 (p1 + p2 + p3 ) × η µν (p1 − p2 )λ + η νλ (p2 − p3 )µ + η λµ (p3 − p1 )ν . p2 .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 . 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 (13.576 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). V µνλρ abcd (p1 . 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 .

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

114)).6 Ghost free gauges In quantizing a non-Abelian gauge theory. n2 = 0 n2 = 0.578 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. for n2 = 1 (normalized) it is called the temporal gauge while for n2 = −1 it is known as the axial gauge. In a general axial-like gauge (13. see for example (13. we recall (see (12.116) where nµ is an arbitrary vector does indeed achieve this. of course.116). These represent all the Feynman rules for the Yang-Mills theory and perturbative calculations can now be carried out using these Feynman rules. or (13.117) . We will now discuss. such a gauge is known as the light-cone gauge. A class of gauge conditions of the form n · Aa (x) = f a (x). in what sense these gauges become ghost free in a non-Abelian gauge theory. Let us ﬁrst discuss this from the diagrammatic point of view. simplify the perturbative calculations enormously primarily because the number of diagrams to evaluate will be much smaller (namely. LTOT = − Fµν F µνa − 4 2ξ (13. this would.153) and (12. 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. When n2 = 0. the diagrams with ghosts will be absent). from two distinct points of view.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 cc ﬁeld. 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. Such a class of gauge choices is conventionally known as ghost free gauges. 13. If possible. The dot over the ghost line denotes where the derivative acts in the interaction term in the Lagrangian density (namely.

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 13.1: Examples of diagrams with open and closed ghost lines. µν (13.120) and the antighost-ghost-gauge vertex involves a multiplicative factor of nµ . (n · p)2 (13. n·p iGab = (13.118) so that we have (independent of whether n2 = 0 or n2 = 0) pν . the integrand of any diagram involving ghosts.116) in this theory has the form δab . either open lines or closed loops. . the ghost propagator in the axial-like gauge (13.13.119) Similarly. (n · p) nµ Gab = −ξδab µν nµ nν Gab = −ξδab . Therefore.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 .

(n · k)m+1 δ(1 − x0 − x1 − · · · − xm ) m+1 = 0 i=0 dxi (13.121) 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. namely. .122) and the vanishing of this integral (in the last step) is easily seen using dimensional regularization (that we will discuss in chapter 15. . (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 ) .580 13 BRST invariance and its consequences dD k = 1 (n · k) (n · (k + p1 )) . we can rewrite this also as 1 m dxi 0 i=0 δ(1 − x0 − x1 − · · · − xm ) (n2 )(m+1)/2 dD k = 0.10)). (13. in (15. An alternative way to see this is to note that the generating functional in the axial-like gauge (13.123) Here G[f a ] is an arbitrary functional of f a which we normally choose to correspond to the ’t Hooft weight factor . (k2 )(m+1)/2 (13. Using the fact that the integral is Lorentz invariant. 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.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 .

(13. Slavnov. Rouet and R. C. A. 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 .7 References 1. J. Note that such a decoupling works only because nµ (in n · Aa inside the determinant) is a multiplicative operator.116) in a non-Abelian gauge theory. C. Theoretical and Mathematical Physics 10. 99 (1972). Taylor. 2. A. (13. correspond to ghost free gauges very much like in the Abelian gauge theories. 344 (1974). It would not work in say. 3.124) However.125) In this case. 436 (1971). Becchi. (13. since the generating functional does not depend on the functional form of G [f a ] (up to irrelevant multiplicative constants). therefore. let us choose G[f a ] = det n · ∂δab + gf abc f c −1 − i R d4 x f a f a 2ξ e .7 References 581 G[f a ] = e i − 2ξ R d4 x f a f a . The axial-like gauge choices (13. 13. the covariant gauge. Physics Letters 52B. A.126) This shows that in this gauge the ghosts decouple completely and can be absorbed into the normalization factor.13. Nuclear Physics B33. Stora. .

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

As we have mentioned earlier gauge ﬁelds can be thought of as the carriers of physical forces. 2 M 2 a µa Aµ A . One of the striking features of these theories is that the gauge ﬁelds are massless simply because the action for a mass term for the gauge ﬁeld in the Lagrangian density M2 Aµ Aµ . in this chapter we would discuss this important question of massive gauge ﬁelds concluding with the standard model of the electroweak interactions.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 .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. Therefore. 14. a massless gauge ﬁeld is completely consistent with the observed fact that electromagnetic forces are long ranged. L = − Fµν F µν + 4 2 583 (14.1) is not invariant under the gauge transformation (9.Chapter 14 Higgs phenomenon and the standard model In chapters 9. However.14) or (12.2) . 2 or (14. 12 and 13 we have studied gauge theories (both Abelian and non-Abelian) in great detail. Therefore.

2) does not have a gauge invariance (because of the mass term) unlike the Maxwell theory.4) with ∂ν and using the anti-symmetry of the ﬁeld strength tensor. ∂∂µ Aν ∂Aν ∂µ or. + M 2 Aν = 0. or. we obtain ∂ · A = 0.5) eliminates one degree of freedom leaving us with three ﬁeld degrees of freedom which is the correct number of dynamical ﬁeld degrees of freedom for a massive spin 1 ﬁeld. or. (14.4) Contracting (14.6) Equation (14. Note that the ﬁeld variable Aµ has four ﬁeld degrees of freedom while the constraint (14.6) together with the constraint (14.584 14 Higgs phenomenon and the standard model where the ﬁeld strength tensor is deﬁned as in the Maxwell theory Fµν = ∂µ Aν − ∂ν Aµ .5) and substituting this back into the Euler-Lagrange equation in (14.4) yields ∂µ (∂ µ Aν − ∂ ν Aµ ) + M 2 Aν = 0. −∂µ F µν − M 2 Aν = 0.3) The Lagrangian density (14. (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).2) has the form ∂L ∂L − = 0.5) deﬁnes the Proca equation (and (14. ∂µ F µν + M 2 Aν = 0. As has been mentioned earlier. (14. The Euler-Lagrange equation following from (14. we also note here that the theory (14. for example. (14. The propagator for this theory can be worked out (see. section 13.5) in a simple manner and in momentum space has the form .2) describes a massive photon which can be seen as follows.

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

Furthermore. section 12.11) which can be achieved by the choice of the gauge transformation parameter e χ(x).13) We recognize this to be the Proca theory (14. in this unitary gauge. but would lead to propagators that are well behaved at high momentum. M θ(x) = (14. if we choose a covariant gauge condition which leads to a gauge ﬁxing Lagrangian density of the form (see. the gauge propu agator (see also (14.14) has the unpleasant features described earlier. we can choose a gauge and if we choose the unitary gauge χ(x) = 0.12) then the Lagrangian density (14.3 and we note that this is known as the ’t Hooft gauge which we will discuss in the next section) 1 (∂µ Aµ − ξM χ)2 . This is typical of the behavior of theories in a unitary gauge. the massless limit M → 0 is now straightforward.15) this would induce a ghost action. On the other hand. (14. .7)) i pµ pν ηµν − . 2 −M M2 iGµν (p) = − p2 (14.8) takes the form M2 1 Aµ Aµ . for example. 2ξ − (14. the there is no diﬃculty in establishing renormalizability in such a gauge.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. L = − Fµν F µν + 4 2 (14. However. As a result.

1).¨ 14. SU (n) (as we have discussed earlier in section 12. (12. Aν ] . say.18). = e M χ. To see that the complete Lagrangian density is also invariant. We will now brieﬂy describe the St¨ckelberg formalism for non-Abelian gauge u theories.18) As we have already seen in chapter 12. Here the ﬁeld strength tensor and the operator V are deﬁned as Fµν V = ∂µ Aν − ∂ν Aµ + ig [Aµ . the Lagrangian density for the gauge ﬁeld is invariant under the transformation (14. Let us consider the Lagrangian density L = Tr − 1 Fµν F µν 2 1 1 +M 2 Aµ + (∂µ V )V −1 Aµ + (∂ µ V )V −1 .16) ig ig where the trace is over the fundamental representation of the group. This Lagrangian density can be easily checked to be invariant under the gauge transformations (see. ig (14. (14.38) and (12. for example. ig (14.41)) Aµ → U Aµ U −1 − V → U V.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. it is suﬃcient to note that . 1 (∂µ U )U −1 .17) where χ denotes a matrix valued scalar ﬁeld.

20) which can be achieved by the choice of the gauge transformation matrix U (x) = V −1 (x). 2 − M2 p M2 iGab (p) = − µν (14. 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 − . (14. ig (14.2) to the non-Abelian case and describes a massive spin 1 ﬁeld belonging to SU (n).21) (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 .22) This can be thought of as a generalization of the Proca theory (14.16) becomes 1 L = − Tr Fµν F µν + M 2 Tr Aµ Aµ . If we choose the unitary gauge condition V = ½. as before.23) . 2 (14.19) Using the cyclicity under trace it is now straightforward to see that the full theory is invariant under the gauge transformation in (14. then the Lagrangian density (14. In this gauge.18).

On the other hand. (14.24) can be checked easily (see (7. the propagator will have the correct asymptotic behavior for large values of the momenta.15)). For simplicity we consider again the self-interacting theory of a complex scalar ﬁeld (as in section 7.84) and (7.25) The Lagrangian density (14. we saw that one of the real components of the scalar ﬁeld becomes massless while the other picks up a mass (with the right sign). 14.2 Higgs phenomenon 589 with the unpleasant features discussed earlier that are characteristics of the unitary gauge choice.26) . (14. (14. but now interacting with an Abelian gauge ﬁeld as well (scalar QED).14. Let us next ask what happens if there is a spontaneous breakdown of a local symmetry. if we choose a covariant gauge ﬁxing condition (a generalization of (14.5). The Lagrangian density for such a theory is given by (with λ > 0. see also section 7. This is known as the Nambu-Goldstone phenomena and the massless scalar ﬁeld is known as the Nambu-Goldstone boson. e δφ = −iǫ(x)φ(x).24) 4 4 where the theory has a “wrong” sign for the mass term and the covariant derivative is deﬁned as in (7.5 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. In that case.6) λ 1 2 L = − Fµν F µν + (Dµ φ)† (D µ φ) + m2 φ† φ − φ† φ . δAµ = 1 ∂µ ǫ(x).2 Higgs phenomenon In section 7.83) Dµ φ = ∂µ φ + ieAµ φ.85)) to be invariant under the inﬁnitesimal local transformations δφ† = iǫ(x)φ† (x).

there is actually an inﬁnity of minima of the potential.42)) 1 φ = √ (σ + iχ).77)) . let us rewrite the Lagrangian density (14. since the mass term for the scalar ﬁeld has the “wrong” sign.30) 2 16 Let us now rewrite the theory around the true vacuum by shifting the ﬁeld variables as (see (7. this actually corresponds to the local maximum of the potential.) Rather the true vacuum of the theory satisﬁes (see (7. 2 χc = χ = 0. On the other hand. as we have discussed in section 7. (14. (14.28) (14.64). the normal vacuum for which φ = φ† = 0.5.590 14 Higgs phenomenon and the standard model The gauge boson in this theory is massless because the gauge symmetry (14. (14. λ where (see also (7.29) For further analysis. this is just one of them) 2m σc = σ = √ .26) does not allow a mass term for the gauge ﬁeld.24) in terms of the σ.27) does not correspond to the true vacuum. (As we have seen in section 7. χ ﬁelds 1 λ † L = − Fµν F µν + (Dµ φ)† (Dµ φ) + m2 φ† φ − φφ 4 4 ← → 1 = − Fµν F µν + ∂µ φ† ∂ µ φ − ieAµ φ† ∂µ φ 4 λ † 2 φφ + e2 Aµ Aµ φ† φ + m2 φ† φ − 4 2 1 1 1 m2 2 = − Fµν F µν + ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + σ + χ2 4 2 2 2 ← → e2 λ 2 − eAµ χ ∂µ σ + Aµ Aµ σ 2 + χ2 − σ 2 + χ2 .5.

16 (14. (14. (14.32) 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 ). Let us look at only the terms in (14.14. λ In this case. the Lagrangian density (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. λ χ → χ + χ = χ.33) where we have used the form of v in (14.79).32) to simplify some of the terms.31) where we have identiﬁed the vacuum expectation value (vev) of the ﬁeld with 2m σ =v= √ .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 . but it now mixes with the gauge ﬁeld Aµ so that we have .34) 2 2 We note here that the χ ﬁeld appears to be massless in this theory as in (7.2 Higgs phenomenon 591 2m σ → σ + σ = σ + √ = σ + v.

1 1 ∂µ χ Aµ + ∂ µ χ ev ev 2 1 = Aµ Aµ + Aµ ∂µ χ + 2 2 ∂µ χ∂ µ χ. ev Bµ = Aµ + then we have (14.79). let us deﬁne 1 ∂µ χ. As in (7.11)) 4e2 m2 . ev e v Aµ + Bµ B µ = (14. (The Lagrangian density in (14.35) the quadratic Lagrangian density (14.) To achieve this.37) 2 We see from (14. λ m2 = M 2 = e2 v 2 = B (14. ignoring the σ ﬁeld should be compared with the St¨ckelberg Lagrangian density u (14.35) ∂µ Bν − ∂ν Bµ = ∂µ Aν + 1 1 ∂ν χ − ∂ν Aµ + ∂µ χ ev ev = ∂µ Aν − ∂ν Aµ .592 14 Higgs phenomenon and the standard model to diagonalize the theory in order to determine the true spectrum of the theory.37) that the quadratic part of the shifted Lagrangian density is completely diagonalized.8).39) . σ (14.36) Thus in terms of this new ﬁeld variable (14. it describes a massive scalar ﬁeld σ with mass m2 = 2m2 .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 .34).38) and a massive gauge ﬁeld Bµ with mass (this is the analog of the unitary gauge in (14. (14.

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

eliminate the mixing term between the Aµ and the χ ﬁelds).33) which reﬂects a spontaneous breakdown of the local symmetry (14.39). (14.40) e The shifted theory (14. To introduce the gauge choice. The presence of the local symmetry (14. that is more useful for this purpose let us proceed as follows. 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 . δχ(x) = −ǫ(x)(v + σ(x)) = −ǫ(x) δAµ (x) = M + σ(x) . as pointed out in the last section.42) .34) (namely. 2ξ 2 (14. commonly known as the ’t Hooft gauge or the Rξ gauge.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. (14.40) nonetheless is invariant the shifted local transformations (recall that it is the vacuum of the theory that breaks the symmetry) δAµ (x) = δσ(x) = ǫ(x)χ(x). First we note that the unshifted theory (14. 2ξ 2ξ ξM 2 2 1 µ (∂ Aµ )2 + M (∂ µ Aµ )χ − χ . e 1 ∂µ ǫ(x).41) e where M is deﬁned in (14. 1 ∂µ ǫ(x).30) is invariant under the inﬁnitesimal gauge transformations (see (7. δχ(x) = −ǫ(x)σ(x).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.84)) δσ(x) = ǫ(x)χ(x). it is not the right gauge to study renormalizability of the theory.

(14. The complete Lagrangian density in this gauge has the form LTOT = L + LGF + Lghost .42) (or (14. (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.34) together with the cross terms in (14. thereby.42) combine into a total divergence and.2 Higgs phenomenon 595 which can be compared with (14. 2 (14.42) can also be written with an auxiliary ﬁeld as (see (12.14.44)) 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 .46) and can be easily checked to be invariant under the BRST transformations . The gauge ﬁxing Lagrangian density (14. the ’t Hooft gauge choice in (14. − ξM 2 pµ pν i ηµν − (1 − ξ) 2 . As a result.164). disappear from the action. It is clear that the mixing terms in (14.15).170)) LGF = ξ 2 F − F (∂ · A − ξM χ).45) We see that unlike in the conventional covariant gauge ﬁxing (12. 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.

(13. let us review some of these essential earlier results before constructing the standard model (known as the Weinberg-SalamGlashow theory). the β decay of the neutron etc. At low energies.3 The standard model The standard model or the gauge theory of electroweak interactions was built on the results of years of earlier work on weak interactions. e δσ = ωχc. M + σ c.47). 2 (14. e δχ = −ω δc = 0.596 14 Higgs phenomenon and the standard model δAµ = 1 ω∂µ c. for example. δAµ (x) δσ(x) δPσ (x) δχ(x) δPχ (x) δc(x) (14.47) Following the discussions in section 13.4 (see.49) . Therefore.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) = 0.48) where Pσ and Pχ denote respectively the sources for the composite BRST variations of the ﬁelds σ and χ in (14. δc = −ωF.) 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 . 14. δF = 0. weak interactions (for example. (14. This is the starting point for establishing the renormalizability of the gauge theory with a spontaneously broken symmetry (Higgs theory).

quite weak and were also known to be extremely short ranged.50) have a weak isospin structure and are also known as the weak isospin currents. As a result.52) where Q denotes the electric charge of the particle. Experimentally .50) with ψn . ψe .51) The weak interactions are. ψν denoting the fermion ﬁelds associated with neutron. (14. I3L the quantum number associated with the 3-component of the weak isospin generator and Y its weak hypercharge quantum number.14.140) for a deﬁnition of right and left-handed particles/ﬁelds).17 × 10−5 (GeV)−2 .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 + · · · . proton. 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 . it is clear from the V − A (vector minus axial vector) structure of the currents in (14. (The V − A structure is connected with the fact that parity is violated maximally in weak interactions.139) and (3. therefore. 2 Q = I3L + (14. the currents in (14. electron and neutrino respectively. ψp . (14. 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). Furthermore. The coupling constant (interaction strength) GF is known as the Fermi coupling and is determined from low energy experiments to have the value GF = 1.50) 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.) In addition to the weak isospin quantum number.

(14. Furthermore. 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.54) With these basics. Although the V − A theory with four fermion interactions works quite well at low energy.51) is dimensional with inverse dimensions of mass (energy).) Therefore. As a result. it is logical to associate the Higgs mechanism to the spontaneous breakdown of these symmetries. (14. it is not renormalizable. On the other hand. As we have discussed weak interactions violate both the weak isospin and the weak hypercharge symmetries and. Namely. the V − A theory cannot be the fundamental theory underlying weak interactions (forces). since weak interactions are short ranged. we would like the gauge theory to describe the symmetry behavior (“SSB” denotes spontaneous symmetry breakdown) SSB SUL (2) × UY (1) −→ UEM (1). 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. it is clear that we should look for a gauge theory of weak interactions based on the local symmetry group SUL (2) × UY (1). a gauge theoretic description would naturally involve massive gauge bosons. 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).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. 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. therefore. . but in such a way that the electric charge is conserved in any weak process.598 14 Higgs phenomenon and the standard model it was known that weak interactions violate both weak isospin and weak hypercharge quantum numbers.

and the right-handed components of a spinor ﬁeld are deﬁned by (see also section 3.3 The standard model 599 14.56) 2 ψ = ψL + ψR . therefore.55) where the left. 2 1 ψR = (½ + γ5 )ψ. Thus. We note that the electron is massive and. The only diﬀerence is that since all the quark ﬁelds are massive. unlike the V − A theory (see (14. The standard model is expected to describe the weak and electromagnetic interactions of all fundamental particles that participate in the weak processes.7) 1 ψL = (½ − γ5 )ψ. (14. (14. the neutrino is experimentally known to be essentially left-handed (see section 3. eR = R.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. there will be right-handed singlet components for every quark ﬁeld (as opposed to the absence of νR in the leptonic sector). the electron and its neutrino. we now know that the constituents of hadronic matter are the quarks. although the leptons are believed to be fundamental particles. The matter ﬁelds associated with the electron family can be grouped as multiplets of SUL (2) as νe e L L= . The quarks as well as the other families can be introduced in a straightforward manner. for simplicity we will consider the matter to consist of only one family of leptons.6). The quark families can also be described by such structures. These include both hadrons and leptons as we have already mentioned. namely.14. In contrast. 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. its ﬁeld can be decomposed into a left-handed part and a right-handed part. . the left-handed doublets have Y = −1 while for the righthanded singlets carry a hypercharge of Y = −2.1 Field content. The other lepton families (µ and the τ families) can be introduced through an obvious generalization. However.3. The left-handed multiplets of the standard model are doublets of SUL (2) while the right-handed particles correspond to singlets. Let us ﬁrst discuss the ﬁeld content of the stan- dard model.

if we would like the gauge bosons to be massive. UY (1) : 1 gauge ﬁeld.53) to describe the local (gauge) symmetry of the theory.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 (14.52). so that the theory should contain four gauge ﬁelds in total. it carries a nontrivial weak hypercharge quantum number given by Y = 1 which is.58) 0 Since the Higgs multiplet is charged.52) holds. minimal Higgs ﬁeld : φ = φ+ φ0 . 2. 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). a Wµ .54)) is given by a doublet of charged ﬁelds. the minimal multiplet of Higgs ﬁelds that can achieve the spontaneous breakdown of the symmetry (as in (14. then the corresponding gauge ﬁelds are determined to be SUL (2) : 3 gauge ﬁelds. Let us now write down the quantum numbers for the matter and the Higgs ﬁeld and verify explicitly that (14.) The gauge ﬁelds which are expected to correspond to the carriers of forces are determined from the local symmetry structure of the theory. (14. .57) Yµ . 3. If we assume (14. a = 1. (14. the opposite of that for the left-handed lepton doublet. (φ+ )† = φ− . (φ0 )† = φ . Furthermore. As we will see shortly. in fact.

4) and (12. (14. 4 4 a = 1.61) 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. First.16) and (12. let us note that the gauge invariant Lagrangian density for the gauge ﬁelds follows from our earlier discussions (see (9.52).45)) to correspond to 1 1 a LG = − Fµν F µν a − Fµν F µν . but it will be present in the matter Lagrangian density.60) .) The Lagrangian density (14. a a b c = ∂µ Wν − ∂ν Wµ − gǫabc Wµ Wν . 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. (14. Note that ǫabc denotes the structure constant of SUL (2) as we know from the study of angular momentum.60) where the ﬁeld strength tensors for the Abelian as well as the nonAbelian gauge ﬁelds are deﬁned as (see (9.3 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 (14.44)) Fµν a Fµν = ∂µ Yν − ∂ν Yµ . (The Abelian gauge ﬁelds are not self-interacting which is why g′ does not appear in the Lagrangian density for the gauge ﬁelds.14.2 Lagrangian density. 3.3. 2. 14.

g g 1 ∂µ ǫ(x). they are inert under the corresponding transformations. Since the two symmetry groups are commuta ing (direct products). 2 2 (14. sometimes it is conventional to denote the generators as τ by making the identiﬁcation 2 σ = τ .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. We have also used a vector notation here to represent the three components of the isospin vectors which is often used interchangeably. we simply have to remember that there are two local symmetries present in the theory. the gauge ﬁeld Wµ for the group SUL (2) does not transform under the group UY (1) and vice versa. 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.) The other thing to note from the structure of (14. therefore. a = 1. 2. (We note here that within the context of isospin. g′ (14. Since the left-handed fermions belong to an isospin doublet while the right-handed fermions are isospin singlets.62) a δWµ (x) = δYµ (x) = where ǫa (x). The gauge invariant Lagrangian density for the matter ﬁelds (leptons) can be obtained through minimal coupling.63) is that the right-handed fermions couple twice as strongly to the Yµ . the Lagrangian density for the leptons takes the form ig′ σ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R.602 14 Higgs phenomenon and the standard model is clearly invariant under the inﬁnitesimal gauge transformations of SUL (2) and UY (1) deﬁned by 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) . 3 and ǫ(x) denote the local inﬁnitesimal parameters of symmetry transformations associated with the groups SUL (2) and UY (1) respectively.

δL(x) = −i (14. since we . 2 2 δR(x) = −iǫ(x)R(x).64) show explicitly that the right-handed fermions do not change under the SUL (2) transformations.65) We note that since the hypercharge of the Higgs doublet is just the opposite of that of the left-handed lepton doublet. We note that we have combined the three components of ǫa (x). therefore.3 The standard model 603 ﬁeld as the left-handed ones simply reﬂecting the Y quantum numbers for the two ﬁelds.62). we also need a Lagrangian density for the scalar Higgs ﬁelds which are expected to lead to a spontaneous breakdown of the local symmetries. 2 2 (14. A mass term as well as a potential for the Higgs ﬁeld can be introduced consistent with the SUL (2) symmetry. a = 1.64) together with the transformations in (14. massless in this theory to begin with).62).64).14. In fact.63) can be easily checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). The Lagrangian density (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. Besides the gauge ﬁelds and the leptons.66) together with the gauge ﬁeld transformations in (14. The minimally coupled fermion Lagrangian density (14. the coupling to the UY (1) gauge ﬁeld is correspondingly with the opposite sign. The transformations (14.65) can be checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)φ(x) + ǫ(x)φ(x). 2 2 δφ(x) = −i (14. Note also that the SUL (2) invariance forbids a mass term for the lepton ﬁelds (the electron and the neutrino are. 3 into a vector ǫ(x) in (14.

(14.68) can be checked to be invariant under the transformations (14.65).60). of course. As we will see shortly. 2 2 δR(x) = −iǫ(x)R(x). φ† . φ† = −m2 φ† φ + λ † 2 φφ . collecting all the terms in (14.24).69) which is invariant under the inﬁnitesimal local gauge transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). here the Higgs ﬁeld is represented by an isospin doublet) V φ. the fermions can have Yukawa interactions with the scalar ﬁelds which can be described by the Lagrangian density LY = −h Rφ† L + LφR .64) and (14. it is worth noting that in spite of the similarity in their forms for the potential. (14.63). (14. 2 2 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) . (14.66). In addition.70) δYµ (x) = .68). (14.67) and (14. (14.68) where h denotes the Yukawa coupling and the Lagrangian density (14. g′ (14. Therefore.67) which is invariant under the local transformations in (14. g g 1 ∂µ ǫ(x). the complete weak interaction Lagrangian density describing just one family of leptons (electron family) can be written as L = LG + Lf + LH + LY − V φ. the Yukawa interaction leads to masses for fermions after the spontaneous breakdown of the symmetry. 4 λ > 0.604 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. δL(x) = −i δφ(x) = −i a δWµ (x) = i σ · ǫ(x)φ(x) + ǫ(x)φ(x).66).

29)) 0 v √ 2 φ = . we have noted earlier in section 7. λ (14.74) . (14.2) that the ﬁrst solution in (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. 4 V φ. φ† = −m2 φ† φ + λ > 0.3. We have chosen the Higgs potential (14. Furthermore. namely. φ = 0. λ (14.72).5 (and also in section 14.73) represents a local maximum while the second solution corresponds to the true minimum of the potential.3 Spontaneous symmetry breaking.73) φ φ† φ = = Furthermore.67) in the form λ † 2 φφ . the weak interactions are short ranged so that we would like the gauge bosons to be massive. there are two solutions to (14. 2m2 .72) λ † φφ 2 As we have seen in the last section. 2m v=√ . Both these issues can be addressed simultaneously if the local symmetries are spontaneously broken by the Higgs phenomenon and this is what we discuss next. We know that both SUL (2) and UY (1) symmetries are violated in weak interactions (the associated quantum numbers are not conserved in weak processes).14. the factor of √2 below is a consequence of the deﬁnition in (14.71) so that the minimum of the potential occurs at ∂V ∂φ ∂V ∂φ† λ † φφ 2 = φ† − m2 + = − m2 + = 0. φ† = 0.3 The standard model 605 14. (14. furthermore.

(14.76) where we have deﬁned 1 ± 1 2 Wµ = √ (Wµ ∓ iWµ ). the other parts will not change under the shift.77) σ · Wµ = 2 = 1 2 √ + 2Wµ φ+ φ0 . √ − 3 2Wµ −Wµ (14. (Basically. (14.58) we obtain .606 14 Higgs phenomenon and the standard model Therefore.65).68)). 2 As a result. to determine the spectrum of particles we need to expand the theory around the true ground state (vacuum) by shifting the scalar ﬁeld as 0 v √ 2 φ →φ+ φ =φ+ . it follows that 3 Wµ √ − 2Wµ (14.78) 3 −Wµ 1 1 3 + + 0 √ 2 Wµ φ + 2 Wµ φ 1 √ W − φ+ − 1 W 3 φ0 µ µ 2 2 Using these as well as the explicit doublet representation of the Higgs ﬁeld in (14. before shifting the ﬁelds let us rewrite the Lagrangian densities for the Higgs and the Yukawa sector of the theory for simplicity (see (14. (14.67) and (14. as we have discussed earlier.) 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 .75) However. this is the sector of the theory that involves the scalar ﬁeld and will be aﬀected by the shift.

3 The standard model 607 LH + LY − V φ. 2 (14. φ† = ∂µ + ig − 0 ig 3 ig′ v Yµ − Wµ φ− − √ Wµ φ + √ 2 2 2 2 . 4 (14.79) The expansion around (14. 2 v 0 0 φ → φ +√ . 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 .80) under which the Lagrangian density (14.14.79) in the Higgs and the Yukawa sector becomes LH + LY − V φ.74) can be done equivalently by letting v φ0 → φ0 + √ .

− √ (eR eL + eL eR ) − 8 2 0 3 We note that the ﬁeld combinations (φ0 + φ ).81) gives ∂µ φ− ∂ µ φ+ + ∂µ φ ∂ µ φ0 + + + 0 igv + µ − igv − µ + Wµ ∂ φ − Wµ ∂ φ 2 2 iv g2 v 2 + µ− 0 3 Wµ W − √ (g′ Yµ + gWµ )∂ µ (φ − φ0 ) 4 2 2 v2 ′ 3 g Yµ + gWµ 8 g′ Y µ + gW µ3 (14.608 × + × ∂µ − ∂µ + ∂µ − 14 Higgs phenomenon and the standard model ig v ig′ µ ig µ3 + φ + √ W µ+ φ0 + √ Y + W 2 2 2 2 ig 3 ig′ Yµ + Wµ 2 2 ig′ µ ig µ3 Y − W 2 2 ig + v 0 φ + √ − √ Wµ φ− 2 2 ig v φ0 + √ + √ W µ− φ+ 2 2 v 0 − heR νeL φ− − heR eL φ + √ − hν eL eR φ+ 2 v v 0 −heL eR φ0 + √ + m2 φ− φ+ + φ + √ 2 2 − v λ − + 0 φ φ + φ +√ 4 2 v φ0 + √ 2 2 v φ0 + √ 2 (14.82) hv λv 2 0 0 2 φ +φ .63) respectively. (g′ Yµ + gWµ ) as well ± appear to have become massive. we cannot truly determine the spectrum of the theory until we diagonalize the quadratic part of the Lagrangian density.60) and (14. the complete quadratic Lagrangian density is obtained to be . However. Adding the quadratic terms from the gauge and the fermion Lagrangian densities in (14.81) . since as the ﬁelds Wµ there is mixing between various ﬁelds. 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).83).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 (14.14. Let us deﬁne g′ g2 + g′2 g g2 + g′2 Zµ = Yµ + 3 Wµ 3 = sin θW Yµ + cos θW Wµ . We can invert the relations in (14. (14.84) to write 3 Wµ = Zµ cos θW − Aµ sin θW .83) + λv 2 0 hv 0 2 − √ (eR eL + eL eR ) − φ +φ .85) . Aµ = g g2 + g′2 Yµ − g′ g2 + g′2 3 Wµ 3 = cos θW Yµ − sin θW Wµ . Yµ = Zµ sin θW + Aµ cos θW . 8 2 It is now a straightforward matter to diagonalize the Lagrangian density (14. (14.

φ = 0 1 √ (σ 2 + − iχ).87) so that we can write 0 ∂µ φ ∂ µ φ0 = 1 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ.86) 1 √ (σ 2 ∂µW ν 3 − ∂ν W µ 3 Similarly.88) Furthermore. 2 i 0 √ φ − φ0 . it is easy to check that 3 3 ∂µ Wν − ∂ν Wµ = (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) + (∂µ Yν − ∂ν Yν ) (∂ µ Y ν − ∂ ν Y µ ) + (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) . for example. we can deﬁne (this is equivalent to deﬁning φ0 = iχ). if we deﬁne the parameters gv .610 14 Higgs phenomenon and the standard model With this. see.84)-(14. 2 2 (14. 2 MW g2 + g′2 v = . 2 (14. (14. 2 cos θW hv √ . 2 λ 4m2 λv 2 = 2 2 λ = 2m2 .89) MW MZ = = me = 2 MH = then together with (14. (14.89). (14.29)) σ = χ = 1 0 √ φ + φ0 . we can write the quadratic part of the Lagrangian density (14.83) as .

eR to deﬁne the massive electron ﬁeld as (see (14.14.90) 2 2 where we have combined the two chirality states eL .56)) e = eL + eR . If we now redeﬁne the (massive) gauge ﬁelds as (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 .91) .(14.

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

This suggests that there is still a residual unbroken gauge symmetry in the theory which can possibly be identiﬁed with the low energy electromagnetic gauge symmetry UEM (1).96) where the Weinberg angle θW is deﬁned by (see (14. namely. .97) There is another neutral vector boson in the theory (14.14. 2 cos θW (14.84)) g′ g2 + g′2 g g2 + g′2 sin θW = cos θW = . (14.93). That this is indeed true can be seen as follows.63) Lf = iLγ µ ∂µ + ig = i ν eL e L σ ig′ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R 2 2 ′ ig + 3 ∂µ + ig Yµ + ig Wµ νeL + √2 Wµ eL 2 2 γµ ig ′ ig ig 3 − ∂µ + 2 Yµ − 2 Wµ eL + √2 Wµ νeL ig + ig 3 ig′ Yµ + Wµ νeL + √ Wµ eL 2 2 2 ig′ ig − ig 3 Yµ − Wµ eL + √ Wµ νeL 2 2 2 + ieR γ µ ∂µ + ig′ Yµ eR = iν eL γ µ + ieL γ µ ∂µ + ∂µ + + ieR γ µ ∂µ + ig′ Yµ eR / / / = iν eL ∂ νeL + ieL ∂ eL + ieR ∂ eR g g + − − √ Wµ ν eL γ µ eL − √ Wµ eL γ µ νeL 2 2 g 3 − Wµ (ν eL γ µ νeL − eL γ µ eL ) 2 . Let us look at the minimally coupled fermion Lagrangian density (14.3 The standard model 613 MZ = MW g2 + g′2 v = . Aµ which is massless.

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

101) This has exactly the form of the V − A four-fermion interaction in (14.49) (for just one family of leptons) and comparing the two we determine G √F 2 g2 . 8MW (14. (14. 2 8MW g2 e2 √ = √ . 14. they would lead to a low energy eﬀective current-current interation Hamiltonian density in the lowest order of the form (basically.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 ) .98). If we look at the charged current interaction terms in (14.100) We recall that the V − A theory describes only charged current interactions. W Figure 14. two charged currents interact through the charged W gauge boson propagator (exchange) as shown in Fig.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 .103) Experimental measurements lead to a value of the Weinberg angle as .14. 4 2GF 4 2GF sin2 θW = (14.102) which leads to 2 MW = (14.1: Charged current-current interaction in the lowest order.

for example. J.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. the WeinbergSalam-Glashow theory does not predict a unique mass for the Higgs particle since its mass depends on the quartic coupling λ. (14. Once again these can be accomodated into extensions of the standard model that we will not go into here. However. 347 (1936). the standard model diﬀers from the conventional Fermi theory in (14. A.616 14 Higgs phenomenon and the standard model sin2 θW ≃ 0. de Phys.4 References 1. Furthermore. e2 √ 4 2GF sin2 θW MW ≃ 90 GeV. (14. et le Radium 7. Proca. 4π 137 we can now determine MW MZ = = GF ≃ 10−5 (GeV)−2 . 14. although bounds on the value of its mass can be put from various other arguments. Such processes have also been experimentally observed. cos θW 1 2 ≃ 80 GeV. .105) These gauge bosons have been discovered and their masses are determined very close to the theoretically predicted values. Search for Higgs particles is a top priority in many experiments at the upcoming LHC.104) and using the values of other constants (see. The standard model compares very well with the experimental results.51) and recall that we have set = c = 1) e2 1 ≃ .23. The standard model also suﬀers from the fact that the neutrino is massless in this theory while we now believe that the neutrino may have a small mass (from neutrino oscillation experiments). (14.

St¨ckelberg. 1860 (1958). Marshak. 7 (1974). Ruiz-Altaba. Helvetica Physica Acta 30. P. 3265 (2004). C. 1264 (1967). Physical Review 109. (1957). B. u 3. International Journal of Modern Physics A19. Kibble. Bernstein. C. Weinberg. Brout. 6. 167 (1971). G. H. Feynman and M. S. 5.4 References 617 2. C. Physical Review 145. 8. Gell-Mann. G. S. . E. Sudarshan and R. Physical Review Letters 19. Guralnik. Higgs. R. 209 (1957). 321 (1964). 4. 7. R. Physical Review Letters 13. W. J. Proceedings of PaduaVenice conference on mesons and newly discovered particles. Nuclear Physics B35. Hagen and T. E. 193 (1958). 11. Physical Review Letters 13. Englert and R. P. W. 10. G. Reviews of Modern Physics 46. Physical Review 109.14. E. 9. Ruegg and M. G. ’t Hooft. 1156 (1966). F. 585 (1964).

This page intentionally left blank .

as renormalization of a quantum ﬁeld theory and involves several steps.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. We will study this question in a systematic manner developing the necessary ideas in this chapter as well as in the next. For example. We have also carried out a few calculations of simple Feynman diagrams describing physical processes in section 9. let us look at one of the simplest interacting ﬁeld 619 . any scattering matrix element (S matrix element) in a given quantum ﬁeld theory can be built out of 1PI (one particle irreducible) graphs.Chapter 15 Regularization of Feynman diagrams 15.4).6 there was only one internal propagator). However. 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. we have to ﬁnd a way of extracting meaningful results from such a quantum ﬁeld theory. the Feynman diagrams become topologically more complicated and may contain internal propagators whose momenta are not uniquely determined in terms of the external momenta. in general.1 Introduction So far we have discussed the basic structures of various quantum ﬁeld theories. the evaluation of Feynman diagrams may lead to divergences depending on the structure of the integral that is being evaluated. as we go to higher orders in perturbation. This procedure is known.6. As a result. Instead some of the internal propagators may involve additional momenta (loop momenta) that are integrated over all possible values. The simple diagrams in 9. As we have discussed earlier (see section 13. In such a case.

it is suﬃcient to concentrate on the 1PI graphs. we can calculate any perturbative amplitude in the φ4 theory.620 15 Regularization of Feynman diagrams theories. 15. is not. 2 2 4! L= (15. (15. − M2 p p4 p3 = iGF (p) = p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). which are 1PI graphs whereas the graph in Fig. In fact. The Feynman rules for this theory are given by i . With these Feynman rules. we can construct the graphs in Fig. therefore. Clearly the 1PI diagrams are fundamental since any other diagram can be built using them as basic elements. in studying quantum ﬁeld theories at higher orders. 15. given these Feynman rules. for example. .1 (which topologically involve loops and.36)) M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ . are called loop diagrams) Figure 15.1) where the coupling constant λ is assumed to be positive. the scalar φ4 theory described by the Lagrangian density (see (6.2.2) 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. namely. Therefore.1: A few examples of 1PI diagrams in the φ4 theory.

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

we have set = 1 throughout our discussion.3).622 15 Regularization of Feynman diagrams P = I − V. From the two relations in (15. is given by L = I − V + 1. (15. . therefore.120).4) and (15. but for a clear understanding of the loop expansion we are going to restore in this section only) i Z[J] = N Dφ e SJ . Let us recall that in the path integral description of a quantum ﬁeld theory. for example. be thought of as the parameter of expansion in the number of loops. 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 number of loops (independent momenta) in such a graph. (15.6) In other words.5). (12. (15.4) where I denotes the number of internal lines and V the number of interaction vertices in the graph.7) where represents the Planck’s constant and we see that it can be thought of as an overall multiplicative parameter like “a” in (15.5) where the term (+1) reﬂects the fact that every 1PI vertex function has an overall energy-momentum conserving delta function associated with it. we can eliminate I − V to obtain L = P + 1. (15. the power of “a” in a diagram in such a theory also determines the number of loops in the diagram and can. the generating functional is deﬁned as (see. 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. therefore.

Loop expansion is a diﬀerent perturbative expansion. On the other hand.5 and 14.3 Cut-off regularization 623 As a result. Factoring out an overall momentum conserving delta function (2π)4 δ4 (p1 −p2 ) and identifying p1 = p2 = p we have (we will always factor out the overall momentum conserving delta function along with the factor of (2π)4 ) k iλ 2 d4 k i (2π)4 k2 − M 2 p p =I=− . the expansion is unaﬀected by how one divides the total action into a free part and an interaction part.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). when we study theories where there is spontaneous breakdown of a symmetry. for example. Therefore. the entire perturbation expansion has to be rearranged which is nontrivial. 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. let us calculate the simplest nontrivial graph in the φ4 theory (15. as we have seen (see. In addition the small value of makes it a legitimate expansion (perturbation) parameter. expanding in powers of in a quantum ﬁeld theory also corresponds to expanding in the number of loops as described in (15.6). when we perturb around the true vacuum (by shifting the ﬁeld variable by its vacuum expectation value). The reason why such an expansion is advantageous can be seen from the fact that since this parameter multiplies the whole action. This is quite important. the scalar ﬁeld picks up a vacuum expectation value which depends on the coupling constant of the theory. for example.15. We are usually accustomed to the idea of perturbation in powers of the coupling constant in the theory. 15. sections 7. In such a case.

(This is the reason for the name cut-oﬀ regularization. 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. y + M2 (15. We see from (15.8) takes the form . therefore.8) dy 1 − Here we have used the fact that the Feynman diagram in (15. the two point function in (15.8) that this integral is divergent and.9) = 2π 2 . 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.) Since large values of momentum correspond to small values of coordinate separation. such divergences are also known as ultraviolet divergences.8) has 1 an associated symmetry factor of 2 and have rotated the momentum to Euclidean space by letting k0 → ik0E in the intermediate step to facilitate the evaluation of the integral. Furthermore. With this regularization.624 = − = − = − = − = − = − iλ 2 15 Regularization of Feynman diagrams id4 kE i 4 −k 2 − M 2 (2π) E d4 kE dΩ 0 ∞ 0 iλ 2(2π)4 iλ 2(2π)4 k2 ∞ 1 2 E + M 3 kE dkE k2 1 2 E + M iλ (2π 2 ) 2(2π)4 iλ 32π 2 iλ 32π 2 ∞ 0 ∞ 0 1 2 k2 dkE 2 E 2 2 kE + M dy y + M2 − M2 y + M2 M2 .

10) (and m.577215665. 2 ˜ 32π M Λ2 (15.12) . in this case this is also the result in the Minkowski space.15.10) The result at the end should be rotated back to the Minkowski space. However.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( ) . (15. We add here for completeness that the cut-oﬀ can also be implemented through a weight factor in the following manner. since there is no momentum vector in the ﬁnal result in (15. We note that we can write (15. Λ are scalar parameters).3 Cut-off regularization 625 iλ I = lim − Λ→∞ 32π 2 = lim − Λ2 0 dy − M 2 0 Λ2 dy y + M2 Λ2 0 Λ→∞ iλ Λ2 − M 2 ln y + M 2 32π 2 iλ Λ2 − M 2 ln 32π 2 Λ2 + M 2 M2 = = Λ→∞ lim − Λ→∞ lim − iλ Λ2 Λ2 − M 2 ln 2 + O 32π 2 M 1 Λ2 . (15.11) ˜ Λ→∞ n γ = lim n→∞ k=1 1 − ln n k = 0.

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 . Since the integrand in (15. (15. This result coincides with (15.4 and 8. (5.214.1). for example.352. 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 ) .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. Furthermore. Ek (15. the integration over the angles can be carried out.47)) Ek = |k|2 + M 2 . we also note here that the diagram in (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 + .14) d3 k where we have identiﬁed (see. for example.14) does not depend on the angular variables. (15. which gives (4π) in three dimensions.13) Furthermore. deﬁning y = |k|.10) with the identiﬁcation ˜ Λ2 = Λ2 − M 2 γ.

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 .10) has been transformed into an infrared divergence in the Schwinger parameter τ . for example. the ultraviolet divergence in (15. Gradshteyn and Ryzhik.3 Cut-off regularization 627 Here we have used the standard table of integrals in the intermediate step (see. This result coincides with (15. We note that the integral in Euclidean space has the form iλ × 2π 2 2(2π)4 iλ 32π 2 iλ 32π 2 iλ 32π 2 iλ 32π 2 0 ∞ 0 ∞ 0 ∞ 0 ∞ ∞ 0 I = − = − = − = − = − k2 1 2 dkE 2 E 2 2 kE + M dy y y + M2 2 dydτ y e−τ (y+M ) dτ 1 2 Γ(2) e−τ M 2 τ (15.10) if we identify 1 ¯ (15. 2.16) Λ2 = Λ2 − M 2 (1 − ln 2) .8) can also be evaluated in an alternative manner which is sometimes useful and so let us discuss the alternate method as well.271. Thus.17) dτ −τ M 2 . 2 This derivation makes it clear that the evaluation of the amplitude is much easier in the Euclidean space. e τ2 Here Γ(2) denotes the gamma function and the parameter τ is known as the Schwinger parameter (or the proper time parameter).271. We note that the integrand in (15.4).15. The integral in (15.17) is well behaved for large values of τ .2 and 2. but is divergent at the lower limit τ = 0.

we note from (15. namely.18) can be compared with (15. Let us next look at another one loop graph. we note that the result in (15.471.628 = = = = 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 . (15. regularizing the integral brings out the divergence structure of the Feynman graph.10) (or (15. 15. For example. Gradshteyn and Ryzhik 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.3. the one loop diagram for the vertex correction shown in Fig. Without going into details.18) can also be obtained by simply cut1 ting oﬀ the τ integral at the lower limit Λ2 (without the regularizing exponential factor). for example.10) and we see that both ways of evaluating the integral lead to the same result. We note that since the cutoﬀ Λ has to be actually taken to inﬁnity in the ﬁnal result.9) to evaluate the integral over τ . Equation (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. but quantum ﬁeld theories can be deﬁned in any number of space-time dimensions). However. For simplicity we would consider all the incoming momenta to vanish. this graph is divergent.18)) that the self-energy diagram at one loop contains a term that diverges quadratically as well as a term which is logarithmically divergent. 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) .

(15.3: One loop correction to the four point vertex function.15.3 Cut-off regularization 629 p3 p2 + p1 p4 p1 p3 p2 p4 p4 p3 + p1 p2 Figure 15. First we have factored out an overall momentum conserving delta function along .19) Λ→∞ lim There are several things to note from this derivation. I = = = 3(−iλ)2 2 3λ2 2(2π)4 3iλ2 2(2π)4 i i d4 k 4 k2 − M 2 k2 − M 2 (2π) id4 kE d4 kE −k2 1 2 E − M 1 2 (kE + M 2 )2 ∞ 0 3 kE dkE 2 = = 3iλ2 (2π 2 ) 2(2π)4 3iλ2 32π 2 3iλ2 32π 2 3iλ2 32π 2 3iλ2 32π 2 lim ∞ 0 ∞ 0 1 (k2 E + M 2 )2 dy y (y + M 2 )2 y + M2 − M2 (y + M 2 )2 Λ2 = dy = Λ→∞ lim 0 dy − M2 y + M2 Λ2 2 ∞ 0 dy (y + M 2 )2 ∞ 0 = = = Λ→∞ lim ln (y + M ) 0 + M2 y + M2 Λ→∞ 3iλ2 Λ2 + M 2 −1 ln 32π 2 M2 Λ2 3iλ2 ln 2 − 1 + O 32π 2 M 1 Λ2 .

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

(15.20) L = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 4! This theory describes the interaction between a fermion and a scalar ﬁeld and the Feynman rules for this theory are given by i . the theory with a Yukawa coupling described by the Lagrangian density 1 λ M2 2 / φ − gψψφ − φ4 . Let us note that any Feyn- man diagram (amplitude). 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. 15. particularly in the second example (15.15. p−m / p − m2 = −(2π)4 igδ4 (p1 + p2 + p3 ). In the previous examples. But let us examine the dependence of diagrams on external momenta (and this will be important in studying renormalization) in another theory. depends analytically on the external momenta. − M2 p = iGF (p) = p2 p p3 p2 p1 p4 p3 = iSF (p) = i(p + m) / i = 2 . namely. in general.3 Cut-off regularization 631 the integrals.21) .1 Calculation in the Yukawa theory.3. But in this process the amplitude (and the theory) becomes cutoﬀ dependent.19). (15. p1 p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). the lack of dependence on the external momenta is simply due to our special choice of the external momenta to be all vanishing.

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 ) . however. Suppose we would like to write AB as a single factor. it would be useful to ﬁnd a way to combine denominators and this is achieved by 1 Feynman’s formula as follows. then we note that 1 0 1 dx (xA + (1 − x)B)2 = 0 = − = − = 1 1 1 − A−B A B 1 .9) which is why the cubic coupling involving fermions is called the Yukawa coupling. In quantum ﬁeld theory at zero temperature. Before we proceed to calculate higher order diagrams in this theory. The lowest order graph involving the fermions in this theory gives rise to the Yukawa potential (as we have seen in section 8. then this formula needs to be generalized (which is important in studying ﬁnite temperature ﬁeld theories). let us note that at higher orders in perturbation theory. On the other hand when the two factors have opposite analytic behavior (iǫ terms with opposite sign). the iǫ term with the same sign).23) holds and is quite .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.22) Therefore. Therefore. (x1 A + x2 B)2 (15. AB 1 1 A − B x(A − B) + B dx (x(A − B) + B)2 1 0 (15. (15.632 15 Regularization of Feynman diagrams where the arrows in the vertices denote the fact that momenta are all incoming. diagrams necessarily involve more internal propagators.

26) Using the Feynman combination formula (15.23) to include more denominators is straightforward and has the form n i=1 1 = Ai 1 n dxi 0 i=1 δ(1 − x1 − · · · − xn ) .3 Cut-off regularization 633 useful. / / (15. (15. Equipped with the Feynman combination formula. let us next calculate the two simplest nontrivial graphs involving loops in this theory. the integral in (15. Generalization of (15.25) can be written as (for simplicity we factor out the multiig 2 plicative factor − (2π)4 which we will put back at the end of the calculation) .121). ( n xi Ai )n i=1 (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 . 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 / / .15.23).

27) where we have deﬁned Q2 = x(1 − x)p2 + xm2 + (1 − x)M 2 . the kE integral can be carried out to give (we now ig 2 put back the overall multiplicative factor − (2π)4 ) = − = − = − ig2 16π 2 ig2 16π 2 ig2 16π 2 1 0 1 0 1 0 dx ((1 − x)pE − m) / ∞ 0 dy y [y + Q2 ]2 −1 dx ((1 − x)pE − m) ln / dx ((1 − x)pE − m) / Λ2 + Q2 Q2 × ln = ig2 16π 2 1 0 Λ2 + x(1 − x)p2 + xm2 + (1 − x)M 2 E −1 x(1 − x)p2 + xm2 + (1 − x)M 2 E dx ((1 − x)p + m) / . E (15. (15.634 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 .28) In (15. Furthermore.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.

4 (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Here the overall negative sign reﬂects the fact that we are evaluating a graph with a fermion loop.30) p = −Tr (−ig)2 = −g2 Tr (k + m)((k + p) + m) / / / d4 k . / Tr A / = 4A · B. Using the trace identities Tr ½ = 4. 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). where it is understood that the limit Λ → ∞ is to be taken. /B the self energy graph in (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. Furthermore.15. m. (15. g).29) = f (p. M.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 (15.19) in the intermediate step and have rotated back to Minkowski space in the ﬁnal step. Λ.30) takes the form . We also note here that we have used the integral from (15.

636 15 Regularization of Feynman diagrams = − = − = = g2 (2π)4 4g2 (2π)4 d4 k 4(k · (k + p) + m2 ) (k2 − m2 )((k + p)2 − m2 ) (−kE · (kE + pE ) + m2 ) 2 (−kE − m2 )(−(kE + pE )2 − m2 ) id4 kE d4 kE d4 kE 4ig2 (2π)4 4ig2 (2π)4 1 (kE · (kE + pE ) − m2 ) 2 (kE + m2 )((kE + pE )2 + m2 ) × = = = dx 0 (kE · (kE + pE ) − m2 ) 2 (x((kE + pE )2 + m2 ) + (1 − x)(kE + m2 ))2 1 4ig2 (2π)4 4ig2 (2π)4 4ig2 (2π)4 4ig2 π 2 (2π)4 ig2 4π 2 ig2 4π 2 lim dx 0 1 d4 kE d4 kE (kE · (kE + pE ) − m2 ) ((kE + xpE )2 + x(1 − x)p2 + m2 )2 E 2 (kE − x(1 − x)p2 − m2 ) E 2 (kE + x(1 − x)p2 + m2 )2 E 3 kE dkE 2 (kE + Q2 )2 2 kE − Q2 dx 0 1 0 1 dx 2π 2 0 ∞ 0 ∞ 0 ∞ = dx 0 1 dy y y − Q2 (y + Q2 )2 = dx 0 1 dy y 1 2Q2 − y + Q2 (y + Q2 )2 Q2 2Q2 2Q4 − + y + Q2 y + Q2 (y + Q2 )2 Λ2 0 = dx 0 0 1 ∞ dy 1 − Λ2 = ig2 Λ→∞ 4π 2 +2Q4 0 ∞ dx 0 0 dy − 3Q2 dy y + Q2 dy 1 0 (y + Q2 )2 dx Λ2 − 3Q2 ln Λ2 + Q2 + 2Q2 Q2 = Λ→∞ lim ig2 4π 2 .

15. Thus. g. This is because the overall momentum conservation eliminates one of the momentum variables. In the examples worked out above it is clear that one can Taylor expand the amplitudes around zero momentum of the external lines. These two simple calculations show that Feynman amplitudes are functions of external momenta. for the fermion self-energy in (15. In fact. Similarly. for the scalar self-energy in (15.32) we can write . m.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. 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. a Feynman amplitude with n external lines is an analytic function of (n − 1) external momentum variables. for example. Λ). 2 (15.29).33) +ﬁnite terms as Λ → ∞ .

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. leads to a lack of manifest rotational invariance (although in the continuum limit this symmetry is recovered). a cut-oﬀ can lead to violation of gauge invariance in gauge theories by giving a mass to the gauge boson. To avoid this problem. 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). 15. it is clear that such a regularization violates manifest Poincar´ invariance of the e quantum theory. This is because of the infrared divergences of the theory which can be seen in the above examples by setting m = 0. 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). while quite useful. it is useful to ﬁnd . Therefore.34) +ﬁnite terms as Λ → ∞ . Furthermore.4 Pauli-Villars regularization Although regularizing a Feynman diagram by cutting oﬀ contributions from large values of momentum seems natural. Similarly.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. Let us note here that if the theory is massless.

36) With these Feynman rules.125)) 1 1 /ψ L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 . To discuss the Pauli-Villars regularization scheme. For example.80) and (9. we can now calculate various 1PI amplitudes in QED.µν (p) = − iηµν . 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 .4 Pauli-Villars regularization 639 a covariant regularization scheme which also respects gauge invariance.15.35) ν = iGF . 2 − m2 p r µ = −(2π)4 ieγ µ δ4 (p + q + r). 4 2 which leads to the Feynman rules µ p (15. The Pauli-Villars regularization provides such a regularization scheme. let us consider QED (quantum electrodynamics) in the Feynman gauge (see (9.(15. p2 p q p = iSF (p) = i(p + m) / .

µ))2 ∞ 0 2 dkE −1 . µ) = x(1 − x)p2 + xm2 + (1 − x)µ2 .) The momentum integral is.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. µ) (kE + Q2 (m. In the .9) for the angular integral.37) ie2 8π 2 dx((1 − x)pE + 2m) / where we have deﬁned Q2 (m. µ))2 2 / kE (2(1 − x)pE + 4m) 2 (kE + Q2 (m. 2 kE + Q2 (m. µ) − 2 2 kE + Q2 (m.112) as well as (15. E (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.26). We have also used gamma matrix identities in (2. (The Dirac gamma matrices have also been rotated to Euclidean space as discussed in (15.27)). µ) (15. of course.111) and (2. µ))2 dx ((1 − x)pE + 2m) / 2 dkE 1 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.

Namely. Q2 (m.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.4 Pauli-Villars regularization 641 Pauli-Villars regularization. Λ) . Λ) (15. let us add to the Lagrangian density of QED (15.39) is deﬁned with the gauge ﬁeld combination Aµ + A′ .) µ We note that the sign of the Lagrangian density for these ﬁctitious ﬁelds is opposite to that of the standard ﬁelds and hence they act as ghost ﬁelds and subtract out contributions. decouple. µ) (15. / ′ L′ = (15.39) where we assume that Λ ≫ m. 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. The idea is to take Λ → ∞ at the end and in that limit the heavy ﬁelds do not propagate and. (The covariant derivative in (15.37) except for a sign and µ → Λ. without doing any further calculation. these additional interacting ﬁelds give an additional contribution to the fermion self-energy (15. therefore.37) coming from the diagram where a heavy photon is being exchanged. Therefore. we can write the additional contribution to the fermion self-energy as ie2 8π 2 ∞ 0 2 dkE −1 .15. 2 kE + Q2 (m. A′ are ﬁctitious heavy µ ﬁelds introduced to regularize diagrams.41) I (reg) = I + I ′ = − dx ((1 − x)pE + 2m) ln / . However.40) I′ = dx ((1 − x)pE + 2m) / so that the eﬀective regularized fermion self-energy can be written as ie2 8π 2 Q2 (m. µ and ψ ′ . The contribution from the heavy photon to the self-energy would be exactly the same as what we have already calculated in (15.

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

2 (15. + y + Q2 (m) (y + Q2 (m))2 = d4 kE E 1 (−δµν ) + Q2 (m) . 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 . η00 → E 2 ΠE . (k2 . p2 ). 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 .4 Pauli-Villars regularization 643 where we have used the gamma matrix identities in (2. To evaluate the momentum integral. Π → (i)ΠE ) and we have δ00 . p2 ) → −(kE .15.43) to 2 Euclidean space ((k0 . We can now integrate each of the E terms in (15.43) reﬂects the fact that we are evaluating a fermion loop.116) and have deﬁned Q (m) = x(1 − x)p2 − m2 .115) and (2.44) The overall negative sign in (15. p0E ). we can rotate (15. 2 (kE + Q2 (m))2 (15.45) individually. p0 ) → i(k0E .45) where Q2 (m) = x(1 − x)p2 + m2 .

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

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. Λ1 ) − ic2 ΠE (pE . the regularized photon self-energy will have the form iΠE (pE . m) − iΠE (pE . Let us assume that we have two sets of heavy ghost fermions with masses √ √ Λ1 . Then the regularized photon amplitude with these two sets of heavy fermions will have the form iΠE (reg) (pE ) = iΠE (pE . Λ2 and charges c1 e. Such a term can be cancelled by adding another set of heavy fermion ﬁelds.48) This shows that the diagram is ﬁnite for any given value of Λ. However. for example. c2 e respectively. Λ) µν µν = − ie2 4π 2 dx 2x(1 − x) δµν p2 − pµE pν E ln E 1 − δµν Q2 (Λ) − Q2 (m) 2 Q2 (Λ) Q2 (m) . (15. As a result. see. this has the unpleasant feature that the photon self-energy has developed a mass term (term proportional to δµν . section 14. Λ2 ) µν µν µν µν = − × ie2 4π 2 ∞ 0 dx 2x(1 − x) δµν p2 − pµE pν E E dy c1 c2 1 − − y + Q2 (m) y + Q2 (Λ1 ) y + Q2 (Λ2 ) −(1 − c1 − c2 ) 1 − δµν (1 − c1 − c2 ) 2 ∞ 0 dy − Q2 (m) − c1 Q2 (Λ1 ) − c2 Q2 (Λ2 ) = − ie2 4π 2 dx 2x(1 − x) δµν p2 − pµE pν E E .15.1) which would violate gauge invariance.

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) .(15. Pauli-Villars regularization involves introducing. therefore. 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) .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. 2. the result (15. c1 Λ2 + c2 Λ2 = m2 . 1 2 (15. n depending on the nature of the divergence in the diagram such that the regularized amplitude .49) E 1 2 It is clear.43) does not involve the photon propagator. in a gauge invariant manner. (15.52) We note here that even though we are working in the Feynman gauge. since the diagram for the photon self-energy in (15.646 × ∞ 0 15 Regularization of Feynman diagrams dy 1 c1 c2 − − 2 (m) 2 (Λ ) y+Q y+Q y + Q2 (Λ2 ) 1 −(1 − c1 − c2 ) 1 − δµν (1 − c1 − c2 ) 2 ∞ 0 dy −(1 − c1 − c2 )x(1 − x)p2 − (m2 − c1 Λ2 − c2 Λ2 ) . · · · .(15.50) then. that if the parameters satisfy c1 + c2 = 1. the quadratic divergences as well as the logarithmic divergences in (15. a set of heavy ﬁelds with masses Λi and √ charges ( ci e) with i = 1. In general.

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). 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. However. i=1 (15. Furthermore.) We should mention here that dimensional regularization has its own problems which we will discuss at the end of . it is by construction a gauge invariant regularization scheme.53) with the conditions on the parameters given by ci = 1. 15. however. Λi ). m) − ci Ii (p. In non-Abelian theories.54) will be ﬁnite and gauge invariant. (k2 − m2 ) ((k + p)2 − M 2 ) (15.4 which cannot be regularized by the Pauli-Villars method.5 Dimensional regularization 647 n I (reg) = I(p. it is quite useful in studying non-Abelian gauge theories for which the Pauli-Villars regularization does not work. the integral d4 k . This procedure is known as dimensional regularization. since gauge invariance is independent of the number of space time dimensions.5 Dimensional regularization From the analysis of the divergence structure of amplitudes in the last two sections it is clear that the divergences are functions of the dimension of space time. 15.55) is logarithmically divergent in four dimensions. For example. i ··· . in less than four dimensions it is ﬁnite. there are graphs of the form shown in Fig. In fact. (15. i i ci Λ2 = m2 . (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.15.

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

let us note that the integrand is spherically symmetric (independent of angular coordinates) and. (15. to calculate amplitudes in n dimensions. p1 p2 (15.60) Furthermore. For example.5 Dimensional regularization 649 p = p2 i . it follows that in n dimensions we have µ ηµ = n. shifted the variable of integration and have deﬁned Q2 = M 2 − p2 . 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. therefore. we can separate out the angular part of the integral . 2 (kE + Q2 )α where we have rotated to Euclidean space.59) We need to be careful about manipulating tensors in n dimensions.61) I = = = = dn kE dn kE 1 . − M2 p4 p3 = −(2π)4 iλµ4−n δn (p1 + p2 + p3 + p4 ).15. since each vector index takes n values. E To evaluate this integral.

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

dθ1 sin θ1 dθ2 = 4π. dθ1 sin2 θ1 dθ2 sin θ2 dθ3 = 2π 2 .67) can be written as .67) Let us recall that the beta function is deﬁned as Γ(p)Γ(q) =2 Γ(p + q) ∞ 0 −p−q B(p. (15.68) Therefore.66) which agree with what we already know from explicitly doing these integrals. for the basic integral (15.5 Dimensional regularization 651 n=2: n=3: n=4: dΩ = dΩ = dΩ = dθ = 2π. q) = dt t2p−1 1 + t2 . 2 n .61) we have 1 dn k n (k 2 + 2k · p − M 2 )α (2π) dΩ 0 n 2 I = = = = i(−1)α (2π)n ∞ n−1 dkE kE (k2 E 1 + Q2 )α i(−1)α 2π (2π)n Γ n 2 i(−1)α 2π (2π)n Γ n 2 n ∞ 0 ∞ 0 n−1 dkE kE (k2 kE Q ∞ 0 1 2 α E + Q ) n−1 n 2 d kE Q Qn 1 2 2α kE Q 1 + Q2 α 1 i(−1)α π 2 = n × 2 n Γ n 2 )α− 2 (2π) 2 (Q dt tn−1 (1 + t2 )−α . (15.15. 2 p= q =α−p=α− (15.69) the integral (15. Thus. we see that with the identiﬁcation n . (15.

any other formula can be obtained from (15.70) Rotating back to Minkowski space. (2π)n Γ(α) (p2 + M 2 )α− 2 (15. using (15. in fact. we see that this gives us our basic integral as dn k 1 n (k 2 + 2k · p − M 2 )α (2π) n I = = iπ 2 (−1)α Γ α − n 2 n . n Γ α − (2π)n Γ(α) (p2 + M 2 )α− 2 2 n (15.71) by diﬀerentiation.71) This basic integral generates all other integrals that we need for evaluating amplitudes in n dimensions and.72) Similarly we can obtain . (2π)n Γ (α) (Q2 )α− 2 n n n 2 Γ α− Γ(α) n 2 (15. For example.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 .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 .

74) becomes . take the limit n → 4 which translates to ǫ → 0. With these basic integration formulae.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 . (15. the one loop scalar self-energy takes the form (see also (15.74) Let us next set n = 4 − ǫ.73) n kµ kν dn k n (k 2 + 2k · p − M 2 )α (2π) and so on. let us now calculate various amplitudes in the φ4 theory (15. of course. 2 (2π)n (M 2 )1− 2 n n = − (15. First of all. At the end of our calculations we should.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 . we are analytically continuing away from four dimensions (to a lower dimension where the integral is well deﬁned). i..e. With this.58). the amplitude (15.15.

2 2 2 ǫ ǫ = Γ = Γ 1+ 2 ǫ 2 ǫ 2 2 1 − γ) = − γ. ≃ ǫ 2 ǫ = Γ 1+ ǫ Γ 2 ǫ = Γ −1 + = 2 −1 + ǫ 2 n Γ 1− 2 ≃ − 1+ ǫ 2 2 −γ ǫ 2 ≃ − + (γ − 1).75) Let us next work out some of the identities involving the gamma functions that will be useful to us. In the limit ǫ → 0.76) where γ denotes the Euler’s constant deﬁned in (15.77).75) takes the form . 2 (2π)n (M 2 )−1+ 2 n (15. Similarly.77) Using (15. we have n 2 n Γ 2− 2 Γ 3− ǫ ǫ ≃ 1 − γ. 16π 2 2 (15. ǫ (15.12). we have π2 (2π)n n = ≃ 1 1 n = ǫ (4π) 2 (4π)2− 2 ǫ 1 1 + ln 4π .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 .76) as well as (15. the scalar self-energy (15.

we can calculate the one loop correction to the vertex function shown in Fig. 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 .122).78) We note here that for M 2 = 0.15. for simplicity. this graph would be regularized to zero in dimensional regularization which is the type of argument used in (13. In this case. Similarly.19).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) .3 and. we will set all the external momenta to vanish as we had done earlier in (15. − + ln ǫ 4πµ2 (15.

ǫ 4πµ2 (15.35)) 1 1 L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 . 2 (15. Let us next calculate one loop ampli- tudes in QED described by the Lagrangian density in the Feynman gauge (see (15.81) As a result.1 Calculations in QED. to make the coupling constant dimensionless. the amplitudes diverge. ǫ (15. /ψ 4 2 (15.80) In n dimensions.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 .5. We note that these amplitudes are well behaved (regularized) for any ﬁnite value of ǫ.79) where we have used (15. 2 2 n−1 3 ǫ ψ = = − . 2 2 2 ǫ . we let e → e µ2 . 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). as we take the limit ǫ → 0 (to go to four dimensions).82) . However.76) and (15.

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 . 2 − m2 p r µ q p = −(2π)4 ieµ 2 γ µ δ4 (p + q + r).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. (15. n k 2 ((k + p)2 − m2 ) (2π) .83) As a result.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 .84) ǫ With these Feynman rules. ǫ (15. the Feynman rules of the theory in n = 4 − ǫ dimensions are given by µ p ν = iGF . p2 p = iSF (p) = i(p + m) / .

79) in n dimensions as well as (15.85) 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.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 .89) .86) Using this.658 15 Regularization of Feynman diagrams We note here that if we use the algebra of the gamma matrices (1.87) (2 − n)(k + p) + nm / / dn k n 2 + x(1 − x)p2 − xm2 )2 (2π) ((k + xp) / / dn k (2 − n)(k + (1 − x)p) + nm 2 − Q2 ) 2 (2π)n (k dn k (2 − n)(1 − x)p + nm / 2 − Q2 )2 (2π)n (k (15. ǫ 4πµ2 (15. the one loop fermion self energy in (15. γµ kγ µ = (2 − n)k. / / (15.

5 Dimensional regularization 659 where we have deﬁned Q2 = −x(1−x)p2 +xm2 and have used (15. n (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Using the n-dimensional identities (see (2. /B / Tr γµ γν γλ γρ = n (ηµν ηλρ − ηµλ ηνρ + ηµρ ηνλ ) .91) the photon self-energy (15. / Tr A / = nA · B.115) and (2.42). /B Tr A /C = 0. (15. Tr ½ = n.116)).77). Similarly the photon self energy at one loop can also be calculated to have the form (the overall negative sign is because of the fermion loop) k p k+p ǫ p = iΠµν (p) = − −ieµ 2 = −e2 µǫ 2 i(k + m) i(k + p + m) / / / dn k Tr γµ 2 γ n 2 ν (k + p)2 − m2 (2π) k −m (15.90) becomes (we factor out (−e2 µǫ ) for simplicity and will restore this later) dn k n(kµ (k + p)ν − ηµν k · (k + p) + kν (k + p)µ + m2 ηµν ) (2π)n (k2 − m2 )((k + p)2 − m2 ) dx dx dn k kµ (k + p)ν + kν (k + p)µ − ηµν (k · (k + p) − m2 ) (2π)n ((k + xp)2 + x(1 − x)p2 − m2 )2 dn k (2π)n = =n =n . This can be compared with (15. Tr A = 0.76) as well as (15.15.90) / / / dn k Tr γµ (k + m)γν (k + p + m) .

660 × =n × 15 Regularization of Feynman diagrams 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − x(1 − x)p2 − m2 ) (k2 − Q2 )2 dx dn k (2π)n ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν + 2kµ kν − ηµν k2 . Let us next look at the terms with kµ kν in the numerator separately which leads to . 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 .76) as well as (15.73). (15.93) × ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν where we have used (15.92) where we have deﬁned Q2 = −x(1 − x)p2 + m2 .92) contains three diﬀerent kinds of terms which can be integrated using the standard formulae in (15. (k2 − Q2 )2 (15.71)(15. The integral in (15.77).

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

this result holds in any covariant gauge with the gauge ﬁxing term ξ arbitrary.96) =− dxx(1 − x) This shows that the photon self energy graph is completely transverse which is consistent with gauge invariance (see also (15. Let us next calculate the one loop amplitude corresponding to vertex correction (charge renormalization). Since there is no longitudinal term in the photon self-energy. For simplicity we will put the momentum of the external photon equal to zero.95).52)).92) takes the ﬁnal form ie2 (ηµν p2 − pµ pν ) 2π 2 2 Q2 1 − ln −γ− . the gauge ﬁxing Lagrangian density does not receive any quantum correction which can also be seen from the BRST identities for QED. since the photon self-energy does not involve the photon propagator. 2 ǫ 4πµ 2 (15.95) Q2 4 + 2γ . The photon does not acquire a mass term and we note here that even though we have chosen to work with the Feynman gauge. 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 . × − + 2 ln ǫ 4πµ2 Adding the contributions from (15.93)-(15. the photon self energy in (15.662 ≃− =− ie2 ηµν 4π 2 ie2 ηµν 4π 2 15 Regularization of Feynman diagrams Q2 4 dx Q2 − + 2 − 2 ln 4π + 2 ln 2 + 2γ − 2 ǫ µ dx −x(1 − x)p2 + m2 (15.

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 .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 / / . / / Therefore. n (2π) k2 ((k + p)2 − m2 )2 Using the n-dimensional identities (see also (2.5 Dimensional regularization 663 (15. 2 ((k + p)2 − m2 )2 k (15.86) we have (15.23) as well as the identity .15.100) Let us next use (15.97) as dn k (2π)n k2 ((k + p)2 − m2 )2 (15.97) = −e3 µ 2 3ǫ / / / / dn k γ λ (k + p + m)γ µ (k + p + m)γλ . we can write the amplitude (15. /γ /γ / as well as (15.113)) γ λ A µ A λ = (2 − n)(2Aµ A − γ µ A2 ).

((1 − x)A + xB)3 1 ((1 − x)A + xB)2 (15.100) which leads to 1 + p)2 − m2 ) 1 (k + xp)2 − Q2 2x (k + xp)2 − Q2 k2 ((k = = dx dx 2. (15. Using this in (15. 1 k2 ((k + p)2 − m2 )2 3.100) and shifting the variable of integration. 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 .101) to combine the denominators in (15.664 15 Regularization of Feynman diagrams 1 AB 2 = − = ∂ 1 ∂ =− ∂B AB ∂B dx dx 2x .102) where we have deﬁned Q2 = −x(1 − x)p2 + xm2 .

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 (15.104) respectively.89) with (15. we see that the fermion self-energy and the vertex function are related as (see also (9.100).105) . (15.104) 2 (1 − x)p − 2m pµ / . (15.103) and (15.103) ie3 µ 2 16π 2 2 ln 4π − γ (−2 + ) × 1− + =− x(1 − x)(−2(1 − x)p + 4m)pµ / −x(1 − x)p2 + xm2 dx(1 − x) γ µ 2 − ln ie3 µ 2 8π 2 + Q2 −γ−1 4πµ2 (15.109) and recall that the coupling constant in the present case is eµ 2 ) k ∂ ∂pµ p k+p p =− 1 eµ 2 k .15.87).88) and (15. −(1 − x)p2 + m2 It is worth noting from this derivation that at every step if we compare (15. p k+p k+p p q=0 (15.

naive dimensional regularization leads to incorrect answers. In addition to the regularization methods that we have discussed in this chapter.6 References 1. 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. Schwinger.106) is deﬁned only in four dimensions (both γ5 and the Levi-Civita tensor µνλρ are manifestly four dimensional quantities).) The power of dimensional regularization is quite obvious from these calculations. Analytic continuation of these to other dimensions is nontrivial. the point splitting method. the ζ-function regularization etc.107) There are several ways to address this issue with the γ5 matrix.666 15 Regularization of Feynman diagrams which we recognize as the Ward-Takahashi identity. The lattice regularization is also quite useful. (15. It is not naively applicable if the calculation involves quantities that typically exist only in four dimensions. 15. we will not go into the details of these other methods here. . we know that i 4! γ5 = iγ 0 γ 1 γ 2 γ 3 = − µνλρ γ µ ν λ ρ γ γ γ . We will discuss this further in section 16. (15. it has its own drawbacks. It is gauge invariant and extremely simple to manipulate with. Physical Review 74.6 where we discuss chiral anomaly. However. 1439 (1948). We simply note here that we select a particular regularization scheme depending on the theory (or the problem) that we are analyzing. However. there exist several other important regularization schemes. For example. (Note that the divergent parts as well as the ﬁnite parts satisfy the identity which demonstrates that dimensional regularization preserves gauge invariance. J. for example. Hence if the theory or the physical quantity of interest involves such objects. the higher derivative method.

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

This page intentionally left blank .

1) i where L0 denotes the sum of the free Lagrangian densities for all the ﬁeld variables in the theory and each Li representing some interaction is a monomial in the basic ﬁeld variables and derivatives. in n dimensions it should be n). we have also seen that since Feynman amplitudes are analytic functions of external momenta. (16. Furthermore. To deﬁne this let us start with the Lagrangian density for a given theory L = L0 + Li .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 our units.1) is dimensionless. the action for the theory (16.2) . In units of = c = 1 which we have been using. As a result. The next step in understanding the process of renormalization is to devise a method for determining which Feynman diagrams in a theory will be divergent as well as the nature of the divergence (without actually evaluating the integrals) and this is achieved through the notion of the superﬁcial degree of divergence of a graph.Chapter 16 Renormalization theory 16. it is easy to check that [M ] = [L]−1 = 1. we have seen that loop diagrams in various theories become divergent and need to be regularized. 669 (16. we can expand them around some reference momentum value (in massive theories conventionally chosen to be zero) so that the regularized divergent parts can be isolated as local terms.

for example. the canonical dimension of a scalar ﬁeld (in our units and in four dimensions) is 1.4) (16.6) Namely.9) .) Similarly the free fermion Lagrangian density (ψ) L0 = iψ∂ ψ − mψψ. Thus. (The mass term also has the correct dimension with this assignment. bi and di to denote respectively the number of fermions.3) [∂µ ] = [L]−1 = 1. 2 2 L0 = that since (φ) (16. 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.7) leads to [ψ] = [ψ] = (16. (16. 2 (16. bosons and derivatives at an interaction vertex following from the interaction Lagrangian density Li . (φ) (16.5) = 4. / 3 . and L0 we have [φ] = 1. for the Yukawa interaction LY = gψψφ.8) In fact. Let us further introduce the notations fi . (16.670 16 Renormalization theory Let us note from the form of the free Lagrangian density for a real scalar ﬁeld M2 2 1 ∂µ φ∂ µ φ − φ .

(16. d = 0. κψσ µν ψFµν . For example.11) f = 0. for an interaction of the form (16. d = 1.13) f = 2. the φ4 interaction (16. 4! (16.1 Superficial degree of divergence 671 we have f = 2. we can determine easily the superﬁcial degree of divergence of the following Feynman graphs as . b = 1.10) LI = leads to λ 4 φ . With these notations we are now ready to introduce the notion of the superﬁcial degree of divergence of a Feynman graph. b = 1. b = 4. d = 0.14) and so on. we have or.16. (16.12) LI = hψγµ ψ∂ µ φ. 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. Similarly. On the other hand.

D = 4 − 2 − 2 × 1 = 0. (16. D = 8 − 5 × 2 = −2. 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 = 2. D = 4 − 2 × 2 = 0.672 16 Renormalization theory D = 4 − 2 = 2.15) If D = 0 for a diagram. Similarly. D = 4 − 2 − 1 = 1.

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 . let us develop a general formula for the superﬁcial degree of divergence of any connected Feynman graph. Rather than calculating the superﬁcial degree of divergence for each graph. then the Feynman diagram is said to be superﬁcially convergent. we have F + 2IF = i ni fi .1 Superficial degree of divergence 673 respectively to have superﬁcial linear or quadratic divergences. Similarly. Although this graph is superﬁcially convergent. number of external fermion lines. number of vertices of the ith type from Li . Let us consider a Feynman diagram with B = IB = F IF = = number of external bosons lines. (16. If D < 0 for a graph.16. as we will see it is the notion of superﬁcial degree of divergence which is useful in the study of renormalization theory. number of internal boson lines. it is actually divergent since it contains a subgraph which is divergent. However.18) The superﬁcial degree of divergence is easily seen to be given by . (16.15).17) The factor 2IB reﬂects the fact that it takes two boson lines at two diﬀerent vertices to form a propagator (internal boson line). The meaning of the adjective “superﬁcial” becomes clear once we look at the last graph in (16. (16. number of internal fermion lines.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-

For example. With the notion of the Taylor operator. We would see in detail how this works in section 16.9: Forest diagrams associated with the simple two loop vertex correction diagram in the φ4 theory. then (D(γ) = 2 and k denotes the generic internal momentum) 1 µ ν ∂ 2 Γ(2) (k. if γ is a Γ(4) .4). p) = Γ(2) (k. For a convergent graph H. . Then the assertion is that the renormalized Feynman graph is convergent. p) p p 2! ∂pµ ∂pν tγ Γ(2) (k. Note that the BPHZ method subtracts out the divergent parts in the integrand itself so that the renormalized integral is ﬁnite. 0) + . mentum variables about zero four-momentum out to order D(γ) which denotes the superﬁcial degree of divergence of γ. 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. we have tγ H = 0.16.44) Here we have neglected the linear term in the Taylor expansion since it would vanish upon integration because of antisymmetry. We should only remember that when nested boxes exist then the tγ operation should be carried from inside out. we can obtain a renormalized Feynman integrand R(G) for the graph G given by the expression R(G) = (1 − tG ) (−tγ ) I(G). If γ denotes the two point function Γ(2) .4.2 A brief history of renormalization 689 Figure 16. F ∈Φγ ∈F (16. p=0 (16.

This theorem is extremely important in the study of renormalization. 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. the generating functional would be stationary leading to R δZ = 0 = N Dφ (F [φ] − J) ei(S[φ]+ Jφ) . Working out in detail.120)) R Z[J] = eiW [J] = N Dφ ei(S[φ]+ Jφ) . this equation modiﬁes and the modiﬁed equation can be obtained from the generating functional for the theory as follows.47) This describes the dynamics of the system at the tree level.690 16 Renormalization theory Weinberg’s theorem. 2 2 3! L= (16.46) 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. δφ 2 − (16. (16. when we include quantum corrections.3 Schwinger-Dyson equation Let us consider the φ3 theory described by the Lagrangian density m2 2 g 1 ∂µ φ∂ µ φ − φ − φ3 . (16.47).49) leads to .48) We note that since the generating functional does not depend on the ﬁeld variable (we are integrating over all ﬁeld conﬁgurations). We recall that the generating functional for the theory in the presence of an external source is deﬁned as (see (12. 16. However. (16.49) where the Euler-Lagrange operator F is deﬁned in (16.

δJ δJ δ − J = 0. g ig δφc ∂µ ∂ µ + m2 φc + φ2 − i − J = 0. F δ − J eiW [J] = 0. δJ δW δ −i − J = 0. c 2 2 δJ ig g ∂µ ∂ µ + m2 φc + φ2 + 2 c 2 δ2 Γ δφ2 c −1 + δΓ = 0.52) can be represented graphically as in Fig.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 .51) with respect to φc and set φc = 0. (16. or. Equation (16.50) Here φc denotes the classical ﬁeld which is deﬁned in (13.16. this can be written explicitly as g 2 δ δJ δ δJ ∂µ ∂ µ + m2 φc + or.51) δφc where Γ[φc ] corresponds to the eﬀective action deﬁned in (13. If we now take the functional derivative of (16. (16.80). we obtain ig δ δ2 Γ (∂µ ∂ + m )½ + 2 = − δφc 2 δφc µ 2 δ2 Γ δφ2 c −1 −1 or. 16.52) 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).3 Schwinger-Dyson equation 691 eiW [J] F − i or.78). F φc − i δJ (16.46). Σ= ig 2 δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c . φc − i φc − i − J = 0. or. For the φ3 theory described by (16.

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. by taking the second functional derivative of (16. 16.692 16 Renormalization theory = . 16.10: Schwinger-Dyson equation for the two point function.) Similarly.11: Schwinger-Dyson equation for the three point function. (16.4 BPHZ renormalization We have seen in detail how multiplicative renormalization works. orders.52) and (16. let us consider the φ3 theory in six dimensions described by the Lagrangian density . Such equations (and for the higher point functions) are known as the Schwinger-Dyson equations. Figure 16.51) with respect to φc and setting φc = 0. (We are using a very compact notation where integrations over intermediate coordinates/momenta are suppressed. = + + Figure 16.53) which can be represented graphically as in Fig. To see in some detail how the BPHZ method works.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 .

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

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

2 2 2 2 ǫ ≃ 2 2 − + (γ − 1) . 2− 6−ǫ ǫ ǫ n =2− = 2 − 3 + = −1 + .16. let us simplify this. (16. we obtain the one loop self energy of the theory in (16.59) to be k p k+p ≃− 5ǫ ig2 µǫ p2 1+ 3 6 2(4π) 6 − 2 + (γ − 1) − ln 4π ǫ 1− ǫ ln(−p2 ) 2 p . As we have discussed earlier (see (15.76)). n ≃ 3 (4π) 2 (4π) 2 1 ǫ ǫ 1 ≃ 1 + ln 4π n Γ −1 + 3 2 (4π) 2 (4π) 2 ≃ 1 2 − + (γ − 1) ǫ 1 (4π)3 1 2 − + (γ − 1) − ln 4π .4 BPHZ renormalization 695 This form is quite suitable to carry out two loop calculations which we will do shortly. ǫ Γ −1 + 1 1 ǫ 1 + ln 4π . However.60) = 1 6 Using these. 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 . to understand the structure of the self energy at one loop.

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 . i xi Ai ) × ( (16.61) 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.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 . 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 =− (16.62) Similarly. (16. (16.64) .

by shifting the variable of integration we obtain the value of the integral in (16. (16. Therefore.65) where we have used the constraint from the delta function in the intermediate steps.4 BPHZ renormalization 697 Let us next simplify the denominator of the integrand as x1 k2 + x2 (k + p + p1 )2 + x3 (k + p)2 = (x1 + x2 + x3 ) k2 + 2k · (x3 p + x2 (p + p1 )) + x3 p2 + x2 (p + p1 )2 = k2 + 2k · ((1 − x1 )p + x2 p1 ) + (1 − x1 )p2 + x2 p2 + 2x2 p · p1 1 = (k + (1 − x1 )p + x2 p1 )2 − (1 − x1 )2 p2 − x2 p2 2 1 − 2x2 (1 − x1 )p · p1 + (1 − x1 )p2 + x2 p2 + 2x2 p · p1 1 = (k + (1 − x1 )p + x2 p1 )2 + x1 (1 − x1 )p2 + x2 (1 − x2 )p2 1 + 2x1 x2 p · p1 = (k + (1 − x1 )p + x2 p1 )2 + x1 (1 − x1 )Q2 . where we have identiﬁed x1 (1 − x1 )Q2 = x1 (1 − x1 ) p2 + x2 (1 − x2 )p2 1 +2x1 x2 p · p1 .16.63) 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 .66) (16.

we can rewrite (16. ǫ 4πµ2 (16.71) 1 1 dx1 0 0 du (1−x1 ) 2 (−x1 (1 − x1 )Q2 ) − ln −γ . we have x3 = 1 − x1 − x2 and x3 = 1 x3 = 0 so that 1 1−x1 ⇒ x1 = x2 = 0. 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.70) Furthermore.68) as ig3 µ 2 =− (4π)3 ǫ (16. ⇒ x2 = 1 − x1 .72) .69) dx1 dx2 dx3 δ(1 − x1 − x2 − x3 ) = dx1 0 0 dx2 . (16.68) 1 1−x1 dx1 0 0 dx2 In this derivation we have used the fact that because of the δ function involving the Feynman parameters.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. 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 .

= + −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 . (16.73) 1 2 Equation (16. 1 (16. Using the form of the one loop self-energy in (16.75) separately.75) has the form .74) Let us next calculate the self-energy at two loops. = 2(4π)3 ǫ ǫ (16.62).12: Non-overlapping two loop self-energy diagrams in the φ3 theory.4 BPHZ renormalization 699 where in the new variables x1 (1 − x1 )Q2 = x1 (1 − x1 )(p + up1 )2 + u(1 − u)p2 .59). Let us ﬁrst look at the graphs with non-overlapping divergence of the form shown in Fig. Let us look at the two integrals in (16. the ﬁrst integral in (16. 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 (16.16.68) leads to the one loop counter term (the factor of arises from doing the integrals over x1 . x2 ) ig3 µ 2 2 .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.

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 .

see (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 .85). + − 2− 72(4π)6 ǫ ǫ 4πµ2 12 (16.16.4 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 .64) as well as (15.) We note here that the 1 ln (−p 2) terms in the ǫ 4πµ above expression are potentially dangerous because if such divergent terms are present.76) Here we have used (16. Next. let us look at the second integral in (16. they will require non-local counter terms to cancel them.60) in the intermediate steps.76) and (16. (The y integration is related to the appropriate beta 2 function.

takes the form ℓ k p ℓ+p k+p p . the sum of the two graphs in (16. Let us next look at the overlapping divergent graph in the self energy at two loops which.71)).76) and (16.75) takes the form (−p2 ) 43 2 2 ig4 p2 − ln + −γ − 2− 72(4π)6 ǫ ǫ 4πµ2 12 + = (−p2 ) 8 2 4 ln − − + γ + ﬁnite ǫ2 ǫ 4πµ2 3 (16. 72(4π)6 ǫ2 ǫ 4πµ2 3 As a result.77).702 ≃ − 16 Renormalization theory g 4 p2 2 i 2 − + (γ − 1) − ln 4π 3 ǫ (4π)3 12(4π) ǫ × ǫ (−p2 ) 5ǫ 1 1 − ln + 6 2 µ2 6 ≃ − = ig4 p2 2 2 (−p2 ) 5 − + ln − + (γ − 1) + O(ǫ) 72(4π)6 ǫ ǫ 4πµ2 3 (16. adding (16. − 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.77) ig4 p2 4 (−p2 ) 8 2 − ln − − + γ + ﬁnite .67) (with x2 expressed in terms of u as in (16. upon using the form of the one loop vertex function in (16.78) I1 + I2 = ig4 p2 2 11 + ﬁnite .

16. Using (16. (2π)n ℓ2 (ℓ + p)2 (Q2 )3− 2 (16.81) . x1 (16.79) as g4 µ2ǫ (−1) 2 −3 n Γ 3− n 2 2 2(4π) Γ 5− × Γ 3− = n 2 n 2 n n n n dn ℓ −3 2 dx1 du x1 (1 − x1 ) 2 −2 (2π)n = dy1 dy2 dy3 y1 2− n 2 δ(1 − y1 − y2 − y3 ) D2 n 5− n 2 g4 µ2ǫ (−1) 2 −3 n Γ 5− n 2 2(4π) 2 dn ℓ (2π)n dy1 dy2 dy3 y1 2 dx1 du x1 −3 (1 − x1 ) 2 −2 .64) to combine denominators.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 . (16.73) (with appropriate momenta).4 BPHZ renormalization 703 3ǫ = (−igµ 2 ) 2 × ǫ i i dn ℓ ig3 µ 2 dx1 du 2 − n (2π)n ℓ (ℓ + p)2 (4π) 2 n 2 n 2 3− 2 − x1 (1 − x1 )Q n (1 − x1 )Γ 3 − n g4 µ2ǫ (−1) 2 −3 Γ 3− = n 2 2(4π) 2 dn ℓ x 2 (1 − x1 ) 2 −2 dx1 du 1 n .79) n −3 n where Q2 is deﬁned in (16. we can write (16.

y1 . To see the divergence .80). v)p2 . the divergence of the subdiagram (one loop vertex) has been transformed to the Feynman parametric integrals.84) .704 16 Renormalization theory The divergence of the two loop integral in (16. we can redeﬁne the variable y2 (as we have already done in (16.83).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. Next. in fact. so that the (shifted) denominator in (16. y1 u(1 − u) 2 p x1 (16.71)) as y2 = (1 − y1 )v. this is not right and.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). where d is deﬁned in (16. First we note that we can shift the variable of integration ℓ → ℓ − (uy1 + y2 )p. u. This is a special feature of overlapping divergent graphs and to appreciate this.83) (16. let us evaluate the integral in (16. we can do the integration over y3 using the delta function in (16.80) in some detail. The one loop pole structure is manifest in the overall multiplicative gamma function.81) has the form D2 = ℓ2 + (1 − y1 )(y1 (u − v)2 + v(1 − v)) + = ℓ2 + d(x1 .82) Substituting this into (16. However. Finally.

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

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) .87) ig4 p2 2 (−p2 ) 2 + − γ + 3 + ﬁnite . (16. − ln 12(4π)6 ǫ2 ǫ 4πµ2 We see again that the dangerous nonlocal divergent terms of the form (−p2 ) 1 ǫ ln 4πµ2 are present in this diagram. with the one loop counter term for the three point function (see (16.74)). we also have a contribution from the graph k p k+p = (−igµ 2 ) 2 g 4 µǫ 2 4(4π)3 ǫ g 4 µǫ 2 4(4π)3 ǫ n ǫ p i dn k i ig3 µ 2 2 n k 2 2(4π)3 ǫ (k + p)2 (2π) 1 dn k dx n (2π) ((k + xp)2 + x(1 − x)p2 )2 dx Γ 2− n (−1)2 i 2 n n (4π) 2 Γ(2) (−x(1 − x)p2 )2− 2 dx x(1 − x) ǫ 2 Γ n 2 n −2 2 ǫ = − = − ig4 µǫ (−p2 ) 2 −2 2 ǫ = − Γ −1 + n 3 (4π) 2 ǫ 2 4(4π) ig4 p2 4(4π)6 2 ǫ (−p2 ) 4πµ2 1− − ǫ −2 = ≃ = Γ −1 + Γ(n − 2) 1 6 1+ 5ǫ 6 −1 2 ig4 p2 2 4(4π)6 ǫ ig4 p2 24(4π)6 ǫ (−p2 ) ln 2 4πµ2 ln 2 − + (γ − 1) ǫ 2 4 + 2 ǫ ǫ 8 (−p2 ) +γ− 2 4πµ 3 + ﬁnite. On the other hand.

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.85) in the intermediate step. (16.89) p so that the sum of the three graphs in Fig. .13 gives + + Figure 16.90) 8 (−p2 ) +γ− 2 4πµ 3 = ig4 p2 12(4π)6 − 2 2 + + ﬁnite .16. ǫ2 3ǫ We see that all the potentially dangerous nonlocal divergent terms have disappeared from the self-energy and the remaining divergences can be removed by adding a local two loop counter term. Similarly.13: Overlapping two loop self-energy diagrams in the φ3 theory.4 BPHZ renormalization 707 Here we have used (16. = 2 2 ig4 p2 + 6 ǫ2 12(4π) ǫ − 4 2 + 2 ǫ ǫ − ln ln (−p2 ) −γ+3 4πµ2 + ﬁnite (16. 16. BPHZ is a renormalization procedure where divergences are subtracted in the graph itself so that it works with any regularization.

708 16 Renormalization theory Consider an arbitrary Feynman graph G. two renormalization parts γ1 and γ2 are called overlapping if they share lines and vertices. It may contain one particle irreducible subgraphs that are superﬁcially divergent. Two renormalization parts γ1 and γ2 are called disjoint if γ1 ∩ γ2 = 0.93) Finally. or γ2 ⊂ γ1 .14. namely. 16. if it is superﬁcially divergent. 16. (16. Thus. for example. 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. if none of the following holds. Fig. γ γ Figure 16. 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. for the three loop self energy graph in the φ3 theory. (16. . A couple of examples of proper subgraphs which correspond to renormalization parts are shown in Fig.91) namely.92) while γ1 and γ2 are called nested if one is contained inside the other. if γ1 ⊂ γ2 .14: Examples of proper renormalization parts in the two loop self-energy graph in the φ3 theory. (16.

let N denote the set of laying down of boxes (forests) containing only nested renormalization parts.15: Examples of forests in the three loop self-energy graph in the φ3 theory.16. Similarly. Thus. the complete set of laying down of boxes (forests) containing disjoint renormalization parts is shown in Fig. Let D denote the set of laying down of boxes which contain only disjoint renormalization parts.17 shows the nested forests in the two loop self-energy graph in the φ3 theory. A forest is called empty if there is no box around any subdiagram. a forest is called full if there is a box around G. γ1 ⊂ γ2 . 16. On the other hand. A forest is called normal if there is no box around the complete graph G. (16.4 BPHZ renormalization 709 Figure 16. 16. Furthermore. let us add to the set D the graph with no boxes.16.94) A set of such laying down of boxes around renormalization parts is called a forest. γ1 ∩ γ2 = 0. . for the two loop self-energy graph in the φ3 theory. Fig. γ2 ⊂ γ1 .

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

(16. tγ I(k. by deﬁnition.95) In other words. acting on a graph. (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. tγ I (k.4 BPHZ renormalization 711 it in the external momenta (external to the graph) up to terms of order D(γ). we are simply able to throw . p. p1 ) = I(k. tγ IG = 0. 0).16. p=0 k γ: p k+p p2 k + p + p1 p1 = dn k I (k. the Taylor operator simply separates out the potentially divergent terms in the integrand so that the ﬁnite part of the graph can be identiﬁed with the integrand (1 − tγ ) IG = I G = ﬁnite.96) This is the conventional tγ operator in the BPHZ prescription and it is clear that using this operation. p. p1 ) . 0. Note that. p) = I(k. if D(γ) < 0. (2π)n 1 µ ν ∂2 p p I(k. Thus. p) 2! ∂pµ ∂pν .97) (16. (2π)n D(γ) = 6 − 6 = 0. (16. p). 0) + p = dn k I(k.

compatible with any regularization. it never uses any particular regularization and is. (16. this makes it clear that for a superﬁcially convergent graph (with no pole term) tG FG = 0.100) Once again.103) . in the case of dimensional regularization. am ǫm = ﬁnite parts. In particular. (16. a graph (integral) at n-loops. ∞ m=0 (16. Therefore. (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.98) In such a case. has the general form ∞ m=−n FG = am ǫm .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.712 16 Renormalization theory away potentially divergent parts inside the integral. if the integral has the form ∞ m=−n Iγ = am ǫm . (16.101) Given a renormalization part γ in a Feynman diagram. in fact. (16. the tγ operation is deﬁned to be (it separates out the pole terms) −1 m=−n t FG = G am ǫm .

γ2 . and (16. However.4 BPHZ renormalization 713 Thus.104) This. . In any case. by Weinberg’s theorem. . graph by graph this procedure can be applied and every graph can be made ﬁnite. since in this method we do not explicitly use counter terms. (16. of course. . the graph G itself may be superﬁcially divergent in which case.105).106) RG IG = RG IG . this method is correspondingly more general. one deﬁnes s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . Then. γs as subdiagrams.107) but this is not true otherwise. . (16. However. the graph would now be ﬁnite. Thus. one needs a ﬁnal subtraction to render the diagram ﬁnite. The only understanding here is that for nested diagrams. by deﬁnition tG RG IG = 0. Let us consider an arbitrary Feynman diagram G with proper renormalization parts γ1 . 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. Thus. the subtraction must be carried out .16. (16. makes all the subdiagrams (proper renormalization parts) in G ﬁnite. the eﬀect of the term (−tγ Iγ ) can be thought of as that of adding counter terms which subtract the divergence. from the deﬁnition of RG IG in (16. then.105) We note that if G is not superﬁcially divergent. let us deﬁne s RG IG = i=1 (1 − tγi ) IG .

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 . For disjoint or overlapping divergences. 16.714 16 Renormalization theory inside out. for example.108) as well as . the order of the subtraction is irrelevant.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. The proof of renormalization by local counter terms. This is the BPH formula. With this. as we have seen. Although this method of making a Feynman diagram ﬁnite is absolutely correct. Thus.108) is obtained as follows. requires that the overlapping divergences be taken care of by lower order counter terms. Zimmermann’s solution of the BPH recursion relation (16. let us look at the complete set of subgraphs that include only disjoint and nested proper renormalization parts. of course. Here Rγk is the operator (16. this does not address the issue of locality of counter terms since we are subtracting out overlapping divergences which. we can now write RG IG = = 1 − tG RG IG 1 − tG (−tγk ) IG . Its eﬀect is to give IG itself). This is seen through Bogoliubov’s R-operator which deﬁnes RG IG = = 1 − tG RG IG 1 − tG Φi γk ∈Φi −tγk Rγk IG . Φi γk ∈Φi (16.104) associated with the superﬁcially divergent renormalization part γk . Equation (16.109) where Φi is the complete set of proper renormalization parts that are non-overlapping. can lead to non-local counter terms. That this is true will be seen shortly through examples. (16.19 deﬁnes all the disjoint and nested proper renormalization parts of the three loop self-energy diagram. The BPH recursion relation (16. Fig. Instead of the set of disjoint proper renormalization parts (subgraphs) of a Feynman graph G.

from the BPH formula (16. Before we proceed to see this. γ3 } Φ7 = {γ2 . The set of disjoint proper renormalization parts for the two loop vertex correction is shown in Fig. Zimmermann’s solution (16. γ2 } γ1 γ2 γ4 Φ6 = {γ1 .109) seem quite diﬀerent from the original relation (16. γ Figure 16. let us look at some simple examples just to get acquainted with the notions of the BPH procedure. γ4 } Figure 16. equivalent. in fact.108).105).4 BPHZ renormalization 715 γ2 γ1 Φ0 = {0} γ3 Φ1 = {γ1 } γ4 γ1 Φ2 = {γ2 } γ2 Φ3 = {γ3 } γ3 Φ4 = {γ4 } Φ5 = {γ1 .19: The complete set of disjoint and nested proper renormalization parts for the three loop self-energy graph in the φ3 theory.16. Therefore.20: The complete set of disjoint proper renormalization parts of the two loop 3 point function graph in the φ3 theory. we have . However. let us next show through examples that they are all. 16.20.

there is no subdiagram in this case. 16. in fact. (16.716 16 Renormalization theory RG IG = IG + IG/γ −tγ Rγ Iγ = IG + IG/γ (−tγ Iγ ) .105) and.111) 1 − tG RG IG = 1 − tG (1 − tγ ) IG .112) This is exactly the original formula (16. We note that. (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 . RG IG = (16. From the relation IG = IG/γ Iγ . Let us next look at the two loop self-energy graphs shown in Fig.21: The complete set of disjoint proper renormalization parts of the two loop self-energy graph in the φ3 theory. Rγ Iγ = Iγ because there is no superﬁcially divergent proper subdiagram of γ. this is how we are supposed to renormalize the theory.21 containing only proper disjoint renormalization parts.110) where IG/γ is the part of the diagram which does not contain the subdiagram Iγ . (16. by deﬁnition. we can write (16.108)) we can write γ1 γ2 Figure 16. Here we have overlapping divergences and following BPH (see (16.110) also as RG IG = IG + (−tγ IG ) = (1 − tγ ) IG .113) . in fact.

4 BPHZ renormalization 717 so that RG IG = 1 − tG (1 − tγ1 − tγ2 ) IG .114) This does not exactly coincide with the original formula (16.115) which we recognize to correspond to the counter term associated with an overlapping divergence if it did not vanish.16. 12(4π)6 ǫ2 . At one loop we had found that (see (16. sets the factor of µ 2 = 1 in the graph at one loop) = so that ig3 2 = (−tγ2 Iγ2 ) .117) tγ1 tγ2 = −tγ1 = − = = ǫ ig3 2 γ1 1 −igµ 2 t 3 ǫ 2(4π) 2 dn k (i)2 (2π)n k2 (k + p)2 − 2 ǫ (16.74). It is such terms that complicate the proof of renormalizability.118) 2 i (−p2 ) g4 4(4π)3 ǫ (4π)3 6 ig4 p2 2 . The diﬀerence between the two is given by 1 − tG tγ1 tγ2 IG . (16. (16.116) (16. 2(4π)3 ǫ (16. Let us recall some of the calculations we had done earlier.105) RG IG = 1 − tG (1 − tγ1 ) (1 − tγ2 ) IG . tγ which is supposed to isolate ǫ pole terms in ǫ.

This makes the connection between (16. Let us recall that the three loop diagram has eight . it is best to study an example. The product of the factors in (16. the ǫ1 term precisely has the same coeﬃcient and it follows. namely.121) involving subtractions of only non-overlapping proper renormalization parts. we note that using this.119) Physically. In general. then we can always ﬁnd a renormalization part γ12 which contains both γ1 and γ2 and for which (1 − tγ12 ) tγ1 tγ2 IG = 0. but more importantly. 2 therefore. (16. (16.120) This result is very important. To make contact with the formula of BPH. (16. there is a theorem that says that if γ1 and γ2 denote two proper renormalization parts of G that are overlapping.87)). that 1 − tG tγ1 tγ2 IG = 0.105) and Zimmermann’s solution (16. If we recall our calculation for the two-loop overlapping divergence (see(16.109).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. then.122) i where Φi ’s deﬁne forests containing only disjoint or nested proper renormalization parts. we can now write s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . physically because it says that we do not need to subtract overlapping divergences. this is the statement that we do not need subtractions associated with overlapping divergences. Φi γk ∈Φi (16.121) can also be written as (1 − tγi ) ≡ (−tγk ) . γi ’s are only disjoint or nested proper renormalization parts.

(16.γ2 } −tγ1 Rγ1 Iγ1 1 − tG −tγ2 Rγ2 Iγ2 + IG/γ4 (−tγ4 ) Iγ4 + Iγ4 /γ2 −tγ2 Rγ2 Iγ2 = IG + IG/γ1 −tγ1 Rγ1 Iγ1 + IG/γ2 −tγ2 Rγ2 Iγ2 + IG/γ3 (−tγ3 ) Rγ3 Iγ3 + IG/{γ1 .16.125) Φi γk ∈Φi . 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 .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 .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 . + (tγ1 Rγ1 )(tγ2 Rγ2 ) IG = (16. we can rewrite (16. = Iγ2 .124) Using this.γ2 } (−tγ1 Iγ1 ) (−tγ2 Iγ2 ) + IG/γ3 −tγ3 Iγ3 /γ1 (−tγ1 Iγ1 ) + IG/γ4 −tγ4 Iγ4 /γ2 (−tγ2 Iγ2 ) .4 BPHZ renormalization 719 forests consisting of disjoint and nested proper renormalization parts which are shown in Fig. Thus. (16.123) Let us next recall that since γ1 and γ2 do not contain any proper renormalization parts.19. we can write Rγ1 Iγ1 Rγ2 Iγ2 = Iγ1 . 16.

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 .20) and (16. if the index of divergence were negative. In this case. This analysis. see (16. we can also remove the divergences by adding local counter terms. This is. the number of such counter terms we need will be inﬁnite and we cannot simply absorb them into a redeﬁnition of the ﬁnite number of existing parameters and ﬁelds of the theory. it is equivalent to making any Feynman graph ﬁnite by local subtractions (counter terms). (16.108). (16. Such theories are called nonrenormalizable theories. namely. We would have divergence structures with more and more numbers of external lines.21)) 2F + 3 δi = di + bi + fi − 4. δi < 0.127) then. Of course. So far. shows how the BPH formula leads to Zimmermann’s solution and how.720 16 Renormalization theory where Φi is the complete set of proper renormalization parts consisting of only disjoint graphs. However. through Weinberg’s theorem. 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.128) not only would we have a ﬁnite number of types of graphs that can be divergent. of course. it is clear that the superﬁcial degree of divergence of a graph would increase with increasing number of interaction vertices. On the other hand if δi > 0. 2 (16. we have talked about theories where the index of divergence of interactions in four dimensions (recall that D = 4 − B − 3 i ni δi . in this case. the BPH formula (16. Such theories are known as renormalizable theories. therefore.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. In contrast.

The total Lagrangian density with the gauge ﬁxing and the ghost Lagrangian densities has the form (see (13. However. 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. we do not need counter terms for the gauge ﬁxing terms). Such theories are conventionally known as super-renormalizable theories.a ν µ ν 2 g2 ξ (1 + C)f abc f apq Ab Ac Aµ. 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. As a result.a ) + gf abc Ab Ac ∂ µ Aν. Let us discuss this brieﬂy with the example of the pure non-Abelian Yang-Mills theory.q + F a F a + (∂ µ F a )Aµ.12)) ξ 1 a L = − Fµν F µν a + F a F a + (∂µ F a )Aµ.p ν.p Aν.a + ∂ µ ca Dµ ca 4 2 1 = ∂µ Aa (∂ µ Aν. 16.a µ ν 4 2 L=− − .129) + ∂µ ca ∂ µ ca − gf abc (∂ µ ca )Ab cc . the three point and the four point amplitudes involving gauge ﬁelds as well as the two point function for the ghosts and the three point ghost interaction vertex function.a 4 2 (16.5 Renormalization of gauge theories 721 of divergence) will be divergent.16. some subtleties arise in the case of gauge theories.q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ. Like QED. µ Naive power counting shows that the only graphs that are superﬁcially divergent are the two point. we can add counter terms to write (1 + A) ∂µ Aa (∂ µ Aν.a − ∂ ν Aµ.a ν µ ν 2 − g2 abc apq b c µ.a − ∂ ν Aµ.a )+(1+B)gf abc Ab Ac ∂ µ Aν.5 Renormalization of gauge theories The analysis in this chapter demonstrates renormalizability of standard ﬁeld theories involving scalar and fermion ﬁelds.

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

for example. we will have 2 g(u) g1 2 −3 = Z1 Z3 g2 . the same relation as in (16.16.134) and remove all the divergences from the theory. But the real question is how can the renormalization process guarantee this. (16. the counterterms will satisfy (16. −2 2 −3 = Z4 Z3 g2 = Z1 Z3 g2 . namely. in turn require that the corresponding counter terms have to be related.134) Therefore.133) and with this identiﬁcation we can write (16. if we choose a regularization scheme which respects the BRST symmetry. therefore.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 = = . The Slavnov-Taylor identities following from the BRST invariance (see.132) ξ (u) = Z3 ξ. (13. g(u) = g1 (u) (u) = g2 .130) as .133) This would. 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. (u) 2 (u) 2 g2 (16.5 Renormalization of gauge theories 723 (16. 2 −3 ˜ 2 −1 ˜ −2 = Z1 Z3 Z3 g2 = Z1 Z3 g2 .135) which would lead to g(u) 2 = g1 (u) 2 = g2 (u) 2 . ˜3 Z3 Z1 Z (16.136) This is. of course. In this case. (16.

n. 2. in studying nonAbelian gauge theories it is natural to discuss renormalizability using dimensional regularization. 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). As we have seen. We note that this theory.138) where a = 1. 2.46)) 1 a L = − Fµν F µν. in addition to having the inﬁnitesimal local gauge invariance δAa = Dµ ǫa (x). a + iψ k γ µ Dµ ψk . in establishing unitarity as well as in showing the gauge independence of the physical matrix elements. 16. · · · .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. (16. Otherwise. · · · . It is essential in identifying the physical states of the theory. (16.139) .137) where the covariant derivative is deﬁned with g(u) . the choice of a regularization which preserves this symmetry is quite crucial. dimensional regularization is one of the regularization schemes (and it is quite simple) which respects gauge invariance and BRST symmetry and. the regularization scheme may introduce unwanted spurious eﬀects into the theory. n2 − 1 and k = 1. It is for these reasons that in the case of gauge theories. µ δψ k = iǫa (x) ψT a δψk = −iǫa (x) (T a ψ)k . 4 (16. The BRST invariance of this Lagrangian density is now manifest (with unrenormalized ﬁelds and parameters). consequently.a(u) L = − Fµν F µν a(u) + 4 2 +∂ µ ca(u) Dµ ca(u) .724 16 Renormalization theory ξ (u) a(u) a(u) 1 a(u) F F + (∂µ F a(u) )Aµ. k .

16. (16. is covariantly conserved.141) We note here that the vector current in (16. (16. therefore. In this theory. will not be conserved.141) which.143) is without the parameter of chiral transformation. Dµ J µ.140). 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. if there is a term in the Lagrangian density of the form Lm = −mψ k ψk .140) Here λ is the space time independent real inﬁnitesimal parameter of transformation and (16. µ J5 = ψ k γ5 γ µ ψk . (16. we can still deﬁne the axial vector current as in (16. if the fermions are massive. i.143) where the variation of the Lagrangian density in (16. there are two conserved currents.6 Anomalous Ward identity 725 is also invariant under the global transformation δψk = −iλγ5 ψk . the invariance of the theory leads to a conserved current. however. We can also derive identities from this relation that would relate various matrix elements of the .141) transforms covariantly under a gauge transformation and.142) then this would not be invariant under the chiral symmetry transformations (16. J µ. therefore. As is the case with continuous symmetries.a = ψ j γ µ (T a )jk ψk . δψ k = −iλψ k γ5 . (16. Nonetheless.e.. Let us note that. µ ∂µ J5 = 0.a = 0.140) describes the chiral symmetry transformations under which the massless theory is invariant. rather it would satisfy the equation µ ∂µ J5 = −δLm = −2imψ k γ5 ψk .

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

but the ones that are directly of interest to us are γ5 γ µ = ǫµν γν .152) .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 = 0.149) As a result of these identities. γ µ γ ν = η µν + γ5 ǫµν . 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 . 16. in two dimensions we have a second rank anti-symmetric tensor (Levi-Civita tensor) given by ǫµν = −ǫνµ .22 and this is given by / / / d2 k Tr γ5 γ µ kγ ν (k + p) .151) Let us now calculate the one loop amplitude involving a photon and an axial vector current as shown in Fig. (16. with ǫ01 = 1. µ ∂µ J5 = 0. note that in addition to the diagonal metric tensor η µν = (+. (16. (16. (16.146) . (16.16.6 Anomalous Ward identity 727 0 1 1 0 −1 0 0 −1 γ 1 = −(γ 1 )† = −iσ2 = so that we have † γ5 = γ5 = γ 0 γ 1 = σ3 = .147) Furthermore. (16. −).148) There are various simple identities in 1 + 1 dimensions involving the Dirac matrices. 2 2 (k + p)2 (2π) k µν I5 (p) = −e2 (16.

154) where we have taken out the multiplicative factor e2 Tr(γ5 γ µ γ λ γ ν γ ρ ) (to be restored shortly) and we have used the basic formulae of dimensional regularization in (15.153) We can evaluate the integral in (16. Γ(2) (−x(1 − x)p2 )2− 2 (16.36) in the limit m = 0. (16. Thus.728 16 Renormalization theory γ5 γ µ k γν p k+p Figure 16. we note that the amplitude in (16.71) and (15. generalizing the integral to n = 2 − ǫ dimensions would lead to µν I5 (p) = − dn k kλ (k + p)ρ (2π)n k2 (k + p)2 kλ (k + p)ρ dn k dx n (2π) (k + xp)2 − x(1 − x)p2 dx dn k kλ kρ − x(1 − x)pλ pρ (2π)n (k2 − x(1 − x)p2 )2 n = − = − = − 2 Γ(1 − n ) 1 ηλρ iπ 2 2 − dx n n (2π) Γ(2) 2 (−x(1 − x)p2 )1− 2 − Γ(2 − n ) x(1 − x)pλ pρ 2 n . Here we have used the propagators for the fermions following from (15. Furthermore.22: The amplitude with an axial vector current and a photon.152) corresponds to the Fourier transform of the matrix element µ 0|T J5 (x)Aν (0) |0 .152) using dimensional regularization which we know would preserve gauge invariance. since . Furthermore.73).

86).157) The result in (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 . (16. using this relation. It is clear now that the factor (2 − n) = ǫ makes even the ﬁrst term ﬁnite and.6 Anomalous Ward identity 729 γ5 is a two dimensional quantity we have not yet evaluated the trace over the Dirac matrices. (16.158) = − .154) that the second term inside the bracket is ﬁnite and hence for this term the trace over the Dirac matrices can be carried out in two dimensions (because any diﬀerence in the trace from extending to n diemnsions would be proportional to ǫ and hence would vanish in the limit ǫ → 0). the trace over the Dirac matrices can be carried out in two dimensions using the identities (16. However. η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 (16. the ﬁrst term is divergent and we have to be careful in evaluating the trace.149) and the ﬁnal result is obtained to be (the x integration is trivial) µν I5 (p) = − ie2 λρ νµ pλ pρ . On the other hand. We note from (16.156) Since each of the terms is now completely ﬁnite. 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 ).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 .16.155) where we have used the gamma matrix identity in n dimension in (15. (16. π (16. in fact.

we always choose a regularization scheme which would preserve gauge invariance in the quantum theory so that the chiral current becomes anomalous. quantum corrections have made the conservation law anomalous and the anomaly in the divergence of the current is given by (16. one of the two currents (or a linear combination of them) would always become anomalous because of quantum corrections. 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. Although our discussion here has been within the context of the two dimensional QED (Schwinger model). In fact.159). This is very important in building models of particle interactions. the same behavior is ob- . it is possible to show that there is no regularization which will simultaneously preserve gauge invariance and chiral invariance. (16.153).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. As a result.159) π 2π This shows that although the axial vector current is divergence free (conserved) at the tree level. the particle multiplets are carefully chosen such that the anomalous contributions coming from various multiplets cancel completely making the gauge current anomaly free. In some gauge theories. On the other hand. this leads to the operator relation e2 e2 ǫµν ∂ µ Aν = ǫµν F µν . since gauge invariance is so vital to the physical theory.23: The triangle diagram in four dimensional QED contributing to the anomaly.

8π 2 (16.161) and the second term in (16.162) which would satisfy a nonanomalous divergence. 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 .23 leads to an anomaly in the chiral current. 8π (16. In grand uniﬁed theories where we treat both quarks and leptons as massless.16. the model cannot be renormalized unless the anomaly is cancelled.159) that a similar statement can be made in the case of the Schwinger model as well. in the four dimensional QED a graph of the form shown in Fig. As we have discussed in the last chapter.6 Anomalous Ward identity 731 tained in any dimension. We note here that. In string theories similar anomaly free considerations ﬁx the . Furthermore. such a current is not gauge invariant (although the associated charge is gauge invariant). it is possible to deﬁne a modiﬁed axial vector current as e2 µ ˜µ J5 = J5 − 2 ǫµνλρ Aν Fλρ .160) corresponds to the anomaly in four dimensional QED. in this case. Therefore. as is clear from its structure. if we have a theory where there are both vector and axial vector gauge couplings of the fermions. We note from (16. 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. However. anomalies have observable eﬀect. Fµν = 2 (16. For example. they have precisely this kind of coupling.160) µ ∂µ J5 = −2imψγ5 ψ + = −2imψγ5 ψ + where the dual ﬁeld strength tensor is deﬁned as 1 ˜ ǫµνλρ F λρ . the correct pion life time can be calculated only if the chiral anomaly is taken into account. 16.

3. C. Erice (1976). Bardeen. Coleman. ed G. 2. Aspects of symmetry. 7. W. S. J. Foundations of Quantum Chromodynamics. 392 (1982). Adler. 11. 54 (1951). Velo and A. Physical Review 75. 6. 486 (1949). 4. N. Proceedings of the International School of Mathematical Physics “Ettore Majorana”. Physical Review 184. 12. Physical Review 177. 2425 (1962). Physical Review 25. 84. Bell and R. 2426 (1969). 16. J. World Scientiﬁc. 13. K. A. 75. Physical Review 128. Dyson. Muta. Ward. S. Bogoliubov and O. 5. 1736 (1949). 47 (1969). N. 426 (1951). Proceedings of Royal Society A64. 9. Kennedy. S. S. Wightman. Weinberg. J. S. Salam. F. these topics are beyond the scope of these lectures. 10. Physical Review 118. 1848 (1969). Cambridge University Press (Cambridge. T. 1985). 301 (1966). A. 838 (1960). Hepp. Nuovo Cimento 60A. S.732 16 Renormalization theory gauge group uniquely to be SO(32) or E(8) × E(8). E. However. Parasiuk. . Singapore (1987). W. 227 (1957). Wightman in “Renormalization Theory”. Acta Mathematica 97. Physical Review 82.7 References 1. Caswell and A. Schwinger. Communications in Mathematical Physics 2. 14. 217 (1951). D. J. Jackiw. 8.

In the same manner.) It is clear from (17. r (17. of course.Chapter 17 Renormalization group and equation 17.2) . ep denote the ﬁne strucp c ture constant and the physical electric charge of the fermion respectively.1 Gell-Mann-Low equation In the early 1950s. We can. 17.10 that the Yukawa potential can be related to the Born amplitude for the scattering of two electrons exchanging a (pseudo) scalar meson. Gell-Mann and Low were interested in studying how the electric force and the potential behave at extremely short distances. the Coulomb potential can also be obtained from the Born amplitude for the scattering of two electrons exchanging a photon.1) 1 where in CGS units α2 = p ≃ 137 and αp . then the potential is obtained to have the form V (r) = 2αp αp 1 ln + ··· . calculate quantum corrections to r this potential and if we add the one loop correction to the potential shown in Fig. namely. We have seen earlier at the end of section 8.1. −→ V (r) = e2 αp . (Physical electric charge is determined from the scattering of on-shell electrons. 1+ r 3π rme 733 (17.1) that at the lowest order the potential is independent of the mass of the fermion and it scales with distance as 1 .

For example. r (17.734 17 Renormalization group and equation Figure 17. it also behaves as V (r) → α r 1− Λ 2α ln 3π me = αp . it r depends on the mass of the electron in such a way that we cannot take the limit me → 0. namely r ≫ me ≫ Λ . 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 + ··· .1: The one loop correction to the electromagnetic potential. we see from (17. Furthermore.2) that the potential no longer scales as 1 and furthermore. (17. When r is 1 1 large. To understand the origin of this behavior. 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. 1 + r 2 m2 e V (r) = (17. V (r) → 1 as r → 0 (at exr tremely short distances) as naive scaling would imply.4) where we have identiﬁed Λ 2α ln αp = α 1 − 3π me . in this limit.3) where α denotes the renormalized ﬁne structure constant of the theory and we note from this result that the potential is well behaved in the limit me → 0. In fact.5) This relation can be inverted to express the renormalized parameter in terms of the physical parameter as . It is clear that the one loop corrected potential has a very diﬀerent character from the lowest order potential.

in the limit Λ → ∞.2) and also shows that the physical ﬁne structure constant can be identiﬁed with αp = rV (r) . (17.9) then.17.8). rme ≪ 1 we obtain αp 1 + 2αp 3π Λ ln me V (r) = r αp 2αp 1+ r 3π ln 1− 2αp ln 3π 1 + r 2 Λ2 + ··· 1 + r 2 m2 e + ··· + ··· (17. 1 r= m e (17.7) = Λ − ln me Λ √ 1 + r 2 Λ2 1 + r 2 Λ2 1 + r 2 m2 e = ≃ 2αp αp ln 1+ r 3π 1 + r 2 m2 e me αp 1 2αp ln + ··· . on dimensional grounds we see that the potential can be expressed as . this derivation clariﬁes that the singularity in the potential at me → 0 arises purely because of the renormalization of charge at large distances. We also see that the naive scaling of the potential breaks down and mass dependence comes in due to renormalization eﬀects. Let us next investigate what would happen if we deﬁne the electric charge at some arbitrary (intermediate) distance R and not as in (17. (17.1 Gell-Mann-Low equation 735 α = αp 1 + Λ 2αp ln 3π me + ··· . For example. 1+ r 3π rme This expression coincides with (17.6) Expressing the potential (17.8) Furthermore. if we deﬁne αR = RV (R).3) in terms of this (physical) parameter.

11) Inverting this relation we obtain α = αR 1 + 2αR ln 3π 1 + R2 Λ2 + ··· . (We note that rme = Rme × R so that it is not an independent quantity.) This function should 1 not be singular as me → 0 which implies that for r.3) leads. Rme ). .12) Putting this back into the potential (17. Rme ≪ 1 (the second of these limits corresponds to me → 0).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. Let us also note that if we deﬁne the electric charge (ﬁne structure constant) as .736 17 Renormalization group and equation r 1 V (r) = F αR . R . 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 . R ≪ me . We note that αR = RV (R) = α 1 − 2α ln 3π 1 + R2 Λ2 + ··· .10) where F is a dimensionless function (function of the independent r r dimensionless variables αR . 1 + R 2 m2 e (17. which can be seen from (17. Rme . 1 + r 2 m2 e + ··· 1 + R 2 m2 e (17. r R (17. in the limit Λ → ∞ and rme . the dependence of the potential on me can be neglected.3) as follows. 1 + R 2 m2 e (17.

13) (see also the deﬁnition (17.14)) αr .12) that we can identify (17.1 Gell-Mann-Low equation 737 αr = rV (r).18) This shows that the charge (ﬁne structure constant) changes (runs) with distance and from the second of the equations in (17. r R r r .14) α = αR 1 + 2αR ln 3π 2αr ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· . = αr 1 + (17.16)) we see that the equation governing this change is given by .17) From this analysis we see that in the limit of me → 0 we can.10)) αr r 1 = F αR . in general.15) which. write (see (17. 1 + r 2 Λ2 1 + r 2 m2 e + ··· (17. . r V (r) = (17.18) (see also (17. 1 + r 2 m2 e 1 me . then we see from (17. = F αR .16) 2αR R ln + · · · 3π r and this is compatible with (17. R ≪ 2αR ln 3π leads to the relation αr = αR 1 + = αR 1 + 1 + R2 Λ2 − ln 1 + R 2 m2 e . R V (r) = αr (17. in the limit Λ → ∞ and r.17.

dr 3π 3π r (17. x=1 (17. in the present case. We see from this discussion that through proper renormalization the dependence of the potential on the electron mass me disappears. As R → 0. namely.22) There are two possibilities for the solution of the equation (17. (17.20) and this is known as the Gell-Mann-Low equation. αr= = αp . In this case. if we let r. x) .21) 1 me We see explicitly from (17.24) . then (17. 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.23) 1.18) αr = F αR . ∂F (αr .19) β(αr ) = − .18) leads to (α0 = αr=0 ) α0 = F (α0 . the bare charge becomes inﬁnity. dαr 2α2 2α2 = − R = − r.738 17 Renormalization group and equation r where dαr dr = −β(αr ). R (17. for example. As R → 0. αR does not exist and consequently. r . In this case. (17. 2. we note that αp and me arise only as initial conditions in this evolution. R → 0 with R = x ﬁxed. the theory develops ghosts and other problems. x) ∂x (17. αR does have a limit which is nonzero and is inde1 pendent of the value 137 which arises only as an initial condir tion.16) that.

it corresponds to the second possibility. we saw that the coupling constants are deﬁned with an arbitrary mass scale µ. 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.5. for example.16) (remember / r that R is ﬁxed as the limit is taken) and. We can keep track of this by using a running coupling constant. We will discuss these ideas in more detail in section 17. a particular renormalization scheme may be preferrable if it makes the higher order contributions in perturbation theory smaller. the eﬀects of masses are negligible at high energies if we renormalize the theory in an appropriate way. we deﬁned the charge at some length . are diﬀerent in the two renormalization schemes even though the regularization is the same (dimensional regularization). R → 0. in dimensional regularization. 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). therefore. Even in dimensional regularization. The renormalized parameters. 17. In QED.2 Renormalization group 739 This is known as a ﬁxed point of the evolution equation (17. It is important to note that in a quantum ﬁeld theory scale invariance is broken by the masses of particles. as r. such as the Euler’s constant etc. α0 → 0 which is obvious from (17. There are two possible sources for this arbitrariness. For example. (In the example of QED that we discussed in the last section. The only remaining breaking of scale invariance is due to renormalization itself. This introduces an additional arbitrariness into the renormalized parameters depending on the arbitrariness in the mass scale used. First. On the other hand. Even within a given regularization. therefore. the MS scheme (minimal subtraction) corresponds to subtracting out only the pole parts (in ǫ) of an amplitude.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.17. However. in the MS scheme we not only subtract out the pole parts (in ǫ) of an amplitude. but some constants as well.19) (also where the beta function vanishes).

) Let us consider a renormalizable theory. u ′ ′ λu = Z1 Z3−2 λ′ . (17.27) so that if we expand the S matrix elements in a perturbation series . m′ .25) deﬁne a set of renormalized parameters. λ′ . see the discussion after (16. Namely. ′ 1 (17. If we are calculating physical quantities such as the scattering amplitudes. ′ ′ m2 = Zm Z3−1 m′ 2 . −1 m2 = Zm Z3 m2 . say the φ4 theory in 4dimensions and let us assume that (in this chapter we will use the terms unrenormalized ﬁelds and unrenormalized parameters denoted by a subscript“u” for the bare ﬁelds and bare parameters. m. we must have (S and S ′ denote representative S-matrix elements calculated in the two prescriptions) S ′ p.740 17 Renormalization group and equation scale which illustrates the source of arbitrariness of the second kind. λ. u −2 λu = Z1 Z3 λ. µ).32)) 1 2 φu = Z3 φ. On the other hand. µ′ = S(p. 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. Calculations with either of these sets of parameters will lead to ﬁnite quantities. (17. they should not only be ﬁnite.26) ′ where we are denoting Z3 = Z3 (µ′ ) and so on. Namely. a different renormalization scale prescription would lead to φu = Z3 2 φ′ . they must also have the same value independent of which set of parameters are used to calculate them.

25) and (17.26) so that we can write the renormalized ﬁelds in the two prescriptions as φ = Z3 2 (λ. from (17. we can check this. If we can calculate the exact series.17.29) 1 In a theory. n n S ′ p. λ′ .25) and (17. We can invert the relations in (17. u (17. However. λ. On the other hand. where α ≃ 137 . (17.26) we can also write −1 m2 = Zm (µ)Z3 (µ)m2 . m′ .33) .µ ′ −1 2 −1 φ′ = Z3 1 −2 λ′ . µ) = O λn+1 . µ′ = (17. µ φ. Here we have identiﬁed z µ . (17. φ′ = z µ′ . all we can say and check is that S ′ p. the residual terms can be non-negligible. Let us next see that this arbitrariness in the renormalized parameters is due to a ﬁnite renormalization.30) which leads to a relation between the two (17. u −1 m′ 2 = Zm (µ′ )Z3 (µ′ )m2 . µ)φu .28) the coeﬃcients of expansion in each of the two series will be renormalization prescription dependent in such a way that the two series would have exactly the same value. µ′ φu . µ′ − S(p. µ) = n an λn .2 Renormalization group 741 S(p. in perturbation theory. In such a case. m. the higher order corrections become negligible very quickly.31) = Z3 (µ′ ) Z3 (µ) .32) Similarly. a′ λ′ n . say the nth order. λ′ . λ. such as QED. m′ . m. we can only calculate up to a given order. (17. in a theory like QCD (where the value of the coupling constant is not small).

Thus. Zm .26) imply −1 2 λ = Z1 Z3 λu .35) Finally.742 which implies that 17 Renormalization group and equation m′ 2 = zm (µ′ .35) and (17. µ) = Z1 (µ′ ) Z1 (µ) −1 Z3 (µ′ ) Z3 (µ) 2 .25) and (17. (17. it follows that (z. Let us next show that these ﬁnite renormalizations (ﬁnite transformations) deﬁne a group. Z1 ) and (Z3 . µ) = ′ Zm (µ′ ) Zm (µ) −1 Z3 (µ′ ) Z3 (µ) . (17. (17. with (17. zm . (17. zλ ) are completely ﬁnite.32). Z1 ) diﬀer only in ﬁnite parts (the divergent parts are the same since the “µ” dependence comes only in the ﬁnite parts). µ)m2 .34) zm (µ . from the deﬁnitions in (17. (17. In fact. Zm . 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). ′ ′ λ′ = Z1−1 Z3 2 λu .37) zλ (µ′ .38) we note that .38) ′ ′ ′ Since (Z3 .36) which leads to the relation λ′ = zλ λ. with (17.

µ = = Z3 (µ′′ ) Z3 (µ) −1 2 ′′ ′ ′ Z3 (µ′′ ) Z3 (µ′ ) −1 2 Z3 (µ′ ) Z3 (µ) −1 2 = z µ′′ . two successive ﬁnite renormalizations lead to an equivalent single ﬁnite renormalization. These transformations further satisfy z(µ.µ z µ . . −1 zλ (µ. µ) = 1 = zm (µ. µ). −1 2 Z3 (µ′′ ) 2 Z3 (µ′ ) Zm (µ′′ ) Zm (µ) Z3 (µ′′ ) Z3 (µ) Z1 (µ′′ ) Z1 (µ′ ) −1 zλ µ′′ .2 Renormalization group 743 z µ . µ′ ) = zλ (µ′ .40) The set of ﬁnite renormalizations. µ = × = Zm (µ′ ) Zm (µ) −1 Z3 (µ′′ ) Z3 (µ′ ) Z3 (µ′ ) Z3 (µ) = zm µ′′ .17. µ). −1 zm (µ. µ′ ) = z −1 (µ′ . µ′ ) = zm (µ′ . Zm (µ′′ ) Zm (µ′ ) −1 −1 zm µ′′ . µ . µ) = zλ (µ. µ′ zλ µ′ . Namely. µ . (17.39) Namely. 2 Z3 (µ) (17. µ). µ = × = Z1 (µ′′ ) Z1 (µ) Z1 (µ′ ) Z1 (µ) −1 2 Z3 (µ′ ) 2 Z3 (µ) 2 Z3 (µ′′ ) = zλ µ′′ . µ). µ′ zm µ′ . µ . physical quantities cannot depend on the arbitrary scale introduced to carry out perturbation calculations in a theory. deﬁne an Abelian group known as the renormalization group and all physical quantities must be invariant under the renormalization group transformations. z(µ. therefore.

. m(µ′ ). . . . m(µ). u −n 2 (17. . p2 . . pn ) . pn ) = Z32 Γ(n) (p1 . . . . pn ) .41) Such a relation between the renormalized and the unrenormalized Green’s functions of the theory holds true in momentum space as well. . . µ) . φu (xn )) = Z3 G(n) (x1 . . mu . xn ) = T (φ (x1 ) . . 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 .43) (17. . n n (17.42) where only (n−1) of the external momenta are independent (because of overall energy-momentum conservation). m(µ). u −n (17.44) which implies that the renormalized 1PI functions in two renormalization prescriptions are related as Γ(n) p. namely. . λu ) u = z 2 (µ′ . λ(µ). xn ) . u n n (17. λ(µ). this has the form Γ(n) (p. let us next look at the Green’s functions of the scalar ﬁeld theory. . µ)Γ(n) (p. λu ) . .45) . . . G(n) (p1 . . . φ (xn )) −n 2 c c = Z3 T (φu (x1 ) . µ′ = Z32 (µ′ ) Γ(n) (p. . µ) = Z32 (µ)Γ(n) (p.744 17 Renormalization group and equation 17.3 Renormalization group equation To understand the consequences of the renormalization group of transformations described in the last section. mu . . . The connected renormalized Green’s functions of the theory are deﬁned as G(n) (x1 . pn ) = Z3 2 G(n) (p1 . . λ(µ′ ). u Written out explicitly.

(17. (17.46) while the unrenormalized coupling can be deﬁned as (in discussions in earlier chapters we had used µ0 = 1) µǫ λu . 0 (17. Let us deﬁne the renormalized quartic coupling in the scalar ﬁeld theory as µǫ λ. it follows that (see also (17.47) Both λu and λ are dimensionless and are related as (compare with (16. Here µ0 is assumed to be ﬁxed and µ is the variable renormalization scale. Since λu is µ-independent (independent of the renormalization prescription).3 Renormalization group equation 745 which follows from the deﬁnition of z in (17.39)) −1 −1 2 µǫ λ = Z1 (µ)Z3 (µ)µǫ λu = µǫ Zλ (µ)λu .48) −2 where we have identiﬁed Zλ = Z1 Z3 . Here we have used p to denote collectively all the independent external momenta of the 1PI vertex function.50) . λu = µ µ0 ǫ λZλ (µ). 0 0 or. the couplings are deﬁned with an arbitrary mass scale. (17. (17.19)) µ dλ dµ = β(λ). Thus.32).17.48) that β(λ) = − ǫ + µ dZλ Zλ dµ λ. As we have noted this deﬁnes the Abelian group of transformations known as the renormalization group.49) where we see from (17. In dimensional regularization. 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.

in general.54) in the MS (MS) scheme.53) Both β and γm are ﬁnite functions of µ as ǫ → 0 since the divergences in the numerator and denominator cancel out. µ and.55) where the constants a. γm = γm (λ).49) decouples from the mass equation (17.746 17 Renormalization group and equation ∂Z ∂µ and the derivative on the right-hand side should be thought of as with λu . .52) and can be solved independently which we will analyze in the next section. c. By dimensional reasoning then they cannot depend on m either. mu held ﬁxed. t. λ4 + · · · . 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. Similarly. As a result. .52) γm = (17. s. m. are independent of µ because the pole terms (in ǫ) are independent of µ as there are no overlapping divergences in a renormalizable theory. . q.51) we obtain dm dµ µ where = −mγm . from −1 m2 = Zm (µ)Z3 (µ)m2 = Z m (µ)m2 . functions of λ. We note that this is a consequence of the fact that dimensional regularization is a mass independent regularization. 2Z m dµ (17. in fact. their dependences are of the form (the µ dependence comes in only through the coupling) β = β(λ). the equation for the coupling constant (17. u u −1 (17. . (17. b. µ dZ m . (17.

mu .57) is known as the renormalization group equation for the 1PI vertex functions and expresses how amplitudes change when the renormalization scale is changed. m(µ).52)) ∂λ . m ∂µ (17.57) (n) µ or.44) Γ(n) (p. λ(µ).58) β(λ) = µ γm = − γ = µ ∂Z3 .56) which leads to the relation (since the unrenormalized quantities are independent of the renormalization prescription) dΓu (p.59) Equation (17. see also (17. where we have deﬁned (the derivatives are taken with λu .3 Renormalization group equation 747 Let us next look at the 1PI amplitudes (17.49) and (17. u −n (17. m(µ). µ) . mu .17. or. ∂µ ∂λ ∂m (17. γ = γ(λ). λ(µ). (17. namely. µ or. . dµ µ µ ∂ ∂λ(µ) ∂ ∂m(µ) ∂ +µ +µ − nγ Γ(n) = 0. λu ) = 0. 2Z3 ∂µ We can also argue in the same way (as we have done for β and γm ) that γ is also ﬁnite and is a function of λ alone (in the MS/MS scheme). ∂µ µ ∂m . dµ d −n Z3 2 (µ) Γ(n) (p. µ) = 0. ∂µ ∂µ ∂λ(µ) ∂µ ∂m ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) = 0. λu ) = Z3 2 (µ) Γ(n) (p. mu ﬁxed.

57) describes how renormalized amplitudes (1PI vertex functions) change as the renormalization scale is changed ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) (pi . the equation takes the form ∂ ∂ + β(λ) − nγ Γ(n) (pi .24)) that the npoint 1PI amplitude Γ(n) in the φ4 theory in four dimensions has the canonical dimension (the 4 simply represents the fact that the vertex is deﬁned without the energy-momentum conserving delta function) Γ(n) = 4 − n. µ) = 0. λ. if we deﬁne the Mandelstam variable associated with the n-point function as s = p2 + p2 + · · · + p2 . ∂µ µ ∂m .60) µ where pi denotes the independent external momenta and as we have seen in (17.62) Let us note from our earlier discussions (see (16. ∂µ ∂λ µ (17.4 Solving the renormalization group equation As we have seen.58) (the derivatives are taken with λu .748 17 Renormalization group and equation 17. the renormalization group equation (17. (17. let us set m = 0 in which case.63) Consequently.64) . mu ﬁxed) ∂λ . 2Z3 ∂µ For simplicity. m ∂µ (17. m. µ) = 0. λ. ∂µ ∂λ ∂m (17. 1 2 n (17.61) β(λ) = µ γm = − γ = µ ∂Z3 .

it can be thought of as scaling the momentum) µ= √ set . ∂µ ∂µ ∂t ∂t µ (17. changing the renormalization scale merely corresponds to changing t.66) so that for a ﬁxed s.17. λ. It follows now that ∂t ∂ ∂ ∂ =µ = . we can write f (n) pi · pj s . t) + v(x) = L(x)ρ(x. f (n) is a dimensionless function of independent dimensionless variables which are Lorentz invariant. s (17. a change in t can be thought of as changing µ or for a ﬁxed µ. Let us relate the renormalization scale µ to a new dimensionless variable t as (this has the advantage that for ﬁxed s. λ.65) where we have used the fact that the n-point vertex function of a scalar ﬁeld theory is Lorentz invariant and. (17.69) Solving the renormalization group equation (17.67) ∂t ∂ s 1 (where from (17. t). t) ∂ρ(x. µ) = s 4−n 2 √ s. ∂t ∂λ and Γ(n) = s 4−n 2 (17.62) can be written as ∂ ∂ + β(λ) − nγ Γ(n) = 0. λ. scaling all momenta by a factor of Γ(n) (pi .68) f (n) t.70) . µ2 s . consequently. 2 µ (17.4 Solving the renormalization group equation 749 then. t = − 1 s ln 2 . or.66) we have used ∂µ = − 1 ∂µ ln µ2 = µ ) so that the 2 renormalization group equation (17.68) now becomes straightforward from the observation that a linear ﬁrst order diﬀerential equation of the form ∂ρ(x. ∂t ∂x (17. pi · pj .

0)) = ρ (x(x. t)) = ρ (x(x. 0) = x.t′ )) .72) (17. t). Therefore. we note that the diﬀerential equation (17.70) clearly depends on determining its characteristics x(x. β(λ) ↔ v(x). −t)) e R0 −t R0 −t dt′ L(x(x. λ ↔ x. Γ(n) ↔ ρ. −t)) e or.72).70) can be written as dρ(x) = L(x)ρ(x). ρ(x) = ρ (x(x. in this case.68) with the hydrodynamic problem (17. dt which can be integrated to give ρ (x(x. t = 0)) e Rt 0 (17.73) dt′ L(x(x. (17.t′ )) . (17. t) = v(x). of course. are deﬁned by the equation dx(x.71). (17. dt with the boundary condition x(x. we see that solving the equation (17.70) as t ↔ t. we then obtain ρ (x(x.750 17 Renormalization group and equation is of hydrodynamic type and is best solved by the method of characteristics.74) Translating t by −t. directly identify the quantities in the renormalization group equation (17. We can. The characteristics.t′ )) .71) Using (17.76) . where we have used (17. (17.75) dt′ L(x(x. nγ ↔ L.

68) requires solving the characteristic equation dλ = β(λ). (17. 0 pi · pj n R−t dt′ γ(λ(λ.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.4 Solving the renormalization group equation 751 so that the solution of the renormalization group equation (17. the leading behavior of the amplitude is given by 4−n 2 Γ(n) → α2 s f (n) (λ(λ.80) then. dt (17.79) so that (remember that −t = t → t − ln α. )e s (17. × en R ln α −t+ln α pi · pj ) s dt′ γ(λ(λ. 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 . namely.t′ )) . (17.77). −t).t′ −ln α)) → α4−n (ln α)a · · · = (n) (α)D(Γ ) (ln α)a · · · . (17. if .77) and once we know the solution to this equation. −t + ln α). Let us next note that if λ∗ is a ﬁxed point of the coupling constant evolution equation (17. we can obtain Γ(n) = s 4−n 2 f (n) (λ(λ. for large α (large momentum).78) Note that if we scale all momenta (keeping µ ﬁxed) as s → α2 s.

then γ(λ∗ ) = 0. (17. (17. λ = λ∗ = constant.84) As a result. s (17.78) behaves as Γ(n) = s pi · pj ntγ(λ∗ ) e s 4−n − n γ(λ∗ ) ln s pi · pj µ2 f (n) λ . pi · pj . under a scaling pi → αpi (17. as Γ(n) ≃ α4−n(1+γ(λ∗ )) Γ(n) . = s 2 e 2 ∗ s 4−n 2 f (n) λ∗ .87) (17. then. if λ∗ = 0. If the ﬁxed point of the beta function happens to be at the origin.86) which follows from the fact that in perturbation theory γ(λ) is a power series in λ.83) Near the ﬁxed point.24)). (17.82) (17. namely.79). the n-point amplitude will scale. at the ﬁxed point. In such a case the amplitudes will have almost naive scaling behavior at very high energies (“almost” because f (n) s also depends on t through the explicit dependence on µ2 which we have not displayed). near the ﬁxed point of the renormalization group equation. the n-point amplitude (17. = s 4−n 2 s µ2 − n γ(λ∗ ) 2 f (n) λ∗ .85) We note that this is almost like the naive canonical scaling behavior of the amplitude except for an extra anomalous scale dimension γ(λ∗ ) for the scalar ﬁelds (recall (16.752 17 Renormalization group and equation β (λ∗ ) = 0. .

therefore. let us analyze the qualitative behavior of the solution. In this case. In perturbation theory the beta function has the form of a power series in the coupling constant β(λ) = β0 λn + O λn+1 .91) is a ﬁxed point of the beta function. t → −∞. that the solution of the coupling constant equation is quite crucial in understanding the behavior of the renormalized amplitudes at high energy as well as under a change of the renormalization scale. the behavior of β(λ) is shown in Fig.90) λ = 0. We note that dλ dt = β λ .89) Before solving this equation explicitly. (17. this leads to the behavior of the β function as shown in Fig. (17. 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. Therefore.2. λ(t) → (17. Qualitatively.88) implies that (recall that λ(λ. 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).4 Solving the renormalization group equation 753 It is clear. the behavior of the coupling constant is given by ∞. 0. 17. t → ∞.92) On the other hand. 17.17.3 and in this case. the ﬁrst thing that we notice is that (17. and λ > 0). t = 0) = λ) λ λ dλ′ β (λ′ ) t = 0 dt′ = t. if β(λ) is negative (β < 0. qualitatively we have . (17.

∞.93) and we say that the origin in the coupling constant space is an ultraviolet ﬁxed point.2: The graph showing that origin in the coupling constant space is an infrared ﬁxed point of the β function. λ(t) → 0. β(λ) λ Figure 17. namely. Let us assume that the beta function has a nontrivial ﬁxed point at λ∗ on the positive axis.754 17 Renormalization group and equation β(λ) λ Figure 17. t → ∞. (17.3: 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. then expanding around this ﬁxed point. we have two distinct cases to consider (when λ∗ = 0. . t → −∞. we recover the two cases we have discussed earlier).

In the diagram Fig.89) leads to . µ → 0 for ﬁxed s). the arrows indicate the direction in which λ moves as t → ∞ (if we start near the ﬁxed point). (17. (17.17. (17. 17.94) r > 1.4 (which shows the behavior of a general β function). if β0 > 0. ′−λ ) β0 (λ β0 λ − λ∗ ∗ (17. In the second case (see (17. then as t → −∞. if β0 < 0. (17. β(λ) = β0 (λ − λ∗ ) + O (λ − λ∗ )2 .95)). then as t → ∞.96) which implies that 1.89) leads to λ t= λ dλ′ = β(λ′ ) λ λ or.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 (17.95) In the ﬁrst case (17.4 Solving the renormalization group equation 755 1. λ → λ∗ .97) 2. λ − λ∗ 1 dλ′ = ln . λ → λ∗ . λ − λ∗ = (λ − λ∗ ) eβ0 t .66) that when t → ∞. µ → ∞ and when t → −∞. β(λ) = β0 (λ − λ∗ )r + O (λ − λ∗ )r+1 . 2. (17.

(λ − λ∗ )r−1 r−1 λ − λ∗ = (λ − λ∗ )r−1 .99) Both in QED and QCD. = 1 − (r − 1)β0 t (λ − λ∗ )r−1 1 − (r − 1)β0 (λ − λ∗ )r−1 t . The diﬀerence between the two theories is that for QED β0 > 0 (see.4: Ultraviolet and Infrared ﬁxed points of the β function. g = 0 is . 1 λ − λ∗ = or. (17. the series starts at the cubic order in the coupling) β(g) = β0 g3 + O g5 . in both cases. 1 − (r − 1)β0 (λ − λ∗ )r−1 t (17. (17. λ t= λ dλ′ = β(λ′ ) λ λ dλ′ β0 (λ′ − λ∗ )r 1 λ − λ∗ r−1 =− or.22)). while for QCD. Thus.756 17 Renormalization group and equation β(λ) λ∗ λ β(λ) λ∗ λ Figure 17. for example.100) where β0 is a constant calculable in perturbation theory. 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. the beta function has the form (namely. β0 < 0. 1 (r − 1)β0 r−1 − 1 (λ − λ∗ )r−1 .

103) As t → ∞. 2β0 e2 t→ (17. Such a pole in the evolution of the coupling constant at a particular momentum scale is called the Landau pole. on the other hand. At inﬁnite energy. So (setting g∗ = 0. β0 < 0) we have from (17.87)).104) Since the unrenormalized (bare) coupling does not depend on the renormalization scale. identifying the coupling with electric charge g = e and recalling β0 > 0). (17. therefore.99) e2 = 1 − 2β0 e2 t 1 . we see that the coupling grows and. for QED we have from (17. in particular. it follows that . − 2β0 t e2 (t) = 1 e2 (17. as we have seen. at large momenta almost naive scaling sets in (see discussion after (17.48) as well as (16. In such theories. 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.101) As t increases. Let us recall that with the MS (minimal subtraction) scheme at one loop we have (see (17.99) g2 = 1 + 2|β0 |g2 t 1 .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.38)) µ µ0 ǫ λu = λ 1+ 3λ 16π 2 ǫ .46)-(17. For QCD.102) and perturbation theory breaks down. diverges when 1 . the theory behaves like a free theory and such theories are called asymptotically free theories. the coupling becomes weaker and weaker and ultimately vanishes. + 2|β0 |t g 2 (t) = 1 g2 (17. (g∗ = 0.

106) = 3λ2 . 2 λ(µ) λ(µ) 16π µ 1 1 3 µ = − ln . ǫλ 1 + 3λ 16π 2 ǫ +µ ∂λ ∂µ 1+ 3λ 8π 2 ǫ = 0.(17. 2 λ(µ) λ(µ) 16π µ λ(µ) = 1 1 λ(µ) − 3 16π 2 µ ln µ . ∂µ µ µ0 ǫ or.758 17 Renormalization group and equation µ ∂λu = 0.107) so that we have λ(µ) λ(µ) µ µ dλ 3 = 2 λ 16π 2 dµ 3 µ = ln . − 1 1 3 µ + = ln . or. 16π 2 which is ﬁnite as discussed earlier. 2 µ 16π µ or. ∂µ 16π 2 β(λ) = µ (17. Therefore.108) . (17.105) This leads to −1 µ ∂λ ∂µ = lim −ǫλ 1 + ǫ→0 3λ 16π 2 ǫ 3λ 16π 2 ǫ 1+ 1− 3λ 8π 2 ǫ = lim −ǫλ 1 + ǫ→0 3λ + O(λ2 ) 8π 2 ǫ = lim −ǫλ 1 + ǫ→0 3λ 3λ − 2 + O(λ2 ) 2ǫ 16π 8π ǫ (17. at one loop in the φ4 theory in four dimensions we obtain ∂λ 3λ2 = . or.

(17.111) (17. therefore. where we have identiﬁed λ 1 Linv = ∂µ φ∂ µ φ − φ4 .) Finally.17.5 Callan-Symanzik equation 759 This is precisely the same behavior as we ﬁnd in QED (see (17.109) which simply emphasizes the fact that they are invariant under a change of the renormalization scale. 2 4! Lm = − m2 2 φ . all other theories have the same character as QED. we note that unlike the general renormalized amplitudes. For an inﬁnitesimal scale transformation (α = ǫ = inﬁnitesimal).5 Callan-Symanzik equation Let us consider the φ4 theory in four dimensions described by the Lagrangian density L = Linv + Lm . physical quantities of the theory denoted by P (such as the physical masses of particles etc) satisfy the renormalization group equation ∂ ∂ + β(λ) ∂µ ∂λ µ P = 0. (17.113) . we have δxµ = ǫxµ . only theories involving non-Abelian gauge interactions may be asymptotically free.112) where α denotes the ﬁnite global parameter of scale transformation. 2 (17. We note that under a scale transformation the coordinates transform as xµ → eα xµ . (17. 17.110) The rationale for dividing the Lagrangian density in this way will become clear shortly. (In fact.101)).

but we have also seen that the scaling dimension of a ﬁeld is diﬀerent in diﬀerent dimensions. δφ = ǫ (xµ ∂µ + d) φ(x). we note from (17.115) (17. (17. As we have discussed earlier. or.111) that (here we will use d = 1 in (17. Under an inﬁnitesimal scale transformation.760 17 Renormalization group and equation Under the scale transformation (17. for a scalar ﬁeld in four dimensions d = 1 at the tree level. The N¨ther current for o .112). the scalar ﬁeld transforms as φ(x) → eαd φ (eα x) .115) for the scalar theory in four dimensions) λ 3 φ δφ 3! λ 3 φ ǫ (1 + xν ∂ν ) φ 3! δLinv = ∂µ φ∂ µ δφ − = ∂µ φ∂ µ (ǫ (1 + xν ∂ν ) φ) − = ǫ ∂µ φ∂ µ φ − λ 4 φ + ∂ µ φ∂µ φ 3! λ ν 3 x φ ∂ν φ 3! + xν ∂ν λ 1 ∂µ φ∂ µ φ − φ4 2 4! +xν ∂µ φ∂ µ ∂ν φ − = ǫ 4 1 λ ∂µ φ∂ µ φ − φ4 2 4! = ǫ (4Linv + xµ ∂µ Linv ) = ǫ∂µ (xµ Linv ) . which has the inﬁnitesimal form φ(x) → (1 + ǫd)φ(x + ǫx). the corresponding action would be invariant under the inﬁnitesimal scale transformations (17.115) (which is the reason for the subscript in the Lagrangian density).116) Therefore.114) Here d represents the scale dimension of the ﬁeld which coincides with the canonical dimension at the tree level. (17.

this can be checked explicitly to hold using the equations of motion. (17. (When m = 0. (17.118) where we have used the fact that the improved energy-momentum tensor is divergence free. Namely. we note here that the improved energymomentum tensor diﬀers from the conventional T µν by total divergence terms and has the explicit form 1 µν η 6 µν Timp = T µν + − ∂ µ ∂ ν φ2 .111) behaves as δLm = −m2 φδφ = −m2 φǫ (1 + xµ ∂µ ) φ = −ǫ m2 φ2 + m2 xµ φ∂µ φ = −ǫ m2 φ2 + = −ǫ m2 φ2 + ∂µ = −ǫ∂µ m2 µ 2 x φ 2 m2 µ 2 x φ 2 + ǫm2 φ2 − 2m2 φ2 m2 µ x ∂µ φ2 2 = ǫ ∂µ (xµ Lm ) + m2 φ2 . On the other hand.17.117) and the conservation of this current leads to µ µν µ ∂µ S µ = Timp µ + xν ∂µ Timp = Timp µ = 0. under the inﬁnitesimal scale transformation in (17.119) This improved tensor is important in studying quantum ﬁeld theories in a curved (gravitational) background. (17. (17.5 Callan-Symanzik equation 761 µν scale transformations can be constructed as (see (6.) For completeness.13) and Timp corresponds to the improved energy-momentum tensor of the theory) µν S µ = xν Timp . the conservation of the scale current requires that the energy-momentum tensor be traceless.120) . the mass term in the Lagrangian density (17.115).

would have the form ∂µ T (S µ (x)φ (x1 ) . the scale current will not be conserved in the full theory. (17. [S 0 . . φ (xn )) + δ x0 − x0 1 T [S 0 (x). (17. . φ (xn )) = T (∆(x)φ (x1 ) . However. φ (xn ) + · · · + δ x0 − x0 n T φ (x1 ) . the Ward identity for the scale current. in this case. . φ (x1 )] .119) using the equations of motion) µ ∂µ S µ = Timp µ = ∆. ∆ = m2 φ2 . .762 17 Renormalization group and equation The divergence term vanishes upon integration in the action. there is a second term that does not vanish. .143)) µ ∂µ J5 = −2imψγ5 ψ.124) where we have used the fact that the integral of S 0 (x) over all space corresponds to the charge associated with the scale current and generates inﬁnitesimal scale transformations of the scalar ﬁelds through .123) vanishes and the integrated Ward identity takes the form i T (δφ(x1 ) · · · φ(xn )) + · · · + i T (φ(x1 ) · · · δφ(xn )) =− d4 x T (∆(x)φ(x1 · · · φ(xn )) .122) Correspondingly.111)). φ (xn )] . In fact. + ··· (17.121) This is analogous to the case of fermion masses breaking the chiral invariance of the theory leading to an identity (in the conservation equation for the axial current) of the form (see (16. As a result. .123) When integrated over x the left-hand side in (17. (17. rather the divergence of the scale current will satisfy a relation of the type (this can be checked easily from the deﬁnition (17. . any term in the Lagrangian density with a dimensional coupling will break scale invariance. This shows that the mass term breaks the invariance under scale transformations (which is the reason for the particular decomposition of the Lagrangian density in (17. .

Then.126) Note here that here the factor “4” arises from the overall energymomentum conserving delta function (in simplifying the action of the diﬀerential operator on the delta function. we can obtain from (17. pi ) . (13.121) and we are suppressing the dependence on other variables for simplicity). we can write (17. If G(n) (pi ) and G∆ (p. ∂pµ i (17. for example.127) .17.124) in the momentum space to have the form n−1 n(d − 4) + 4 − pµ i i=1 ∂ ∂pµ i G(n) (pi ) = iG∆ (0.5 Callan-Symanzik equation (n) 763 commutators (see.125) i × T ∆(x) as well as (17. Let Γ(n) (pi ) denote the 1PI vertex function with n scalar ﬁelds and (n) Γ∆ (p. however. (n) (17. We are. but this is to be understood) n (2π) δ 4 4 4 4 i pi G (n) (pi ) = i=1 4 d4 xi eipi ·xi T i n φ(xi ) . pi ) denote the connected Green’s functions respectively for n scalar ﬁelds and n scalar ﬁelds as well as a ∆ composite operator. pi ) i = d x i=1 d4 xi eipi ·xi φ(xi ) .126) the identity satisﬁed by the n-point 1PI vertex functions which takes the form n−1 pµ i i=1 ∂ (n) + nd − 4 Γ(n) (pi ) = −iΓ∆ (0. interested in the 1PI vertex functions. we have used identities such as x∂x δ(x) = −δ(x)) and the factor (n(d − 4) + 4) denotes the naive canonical dimension of the n-point Green’s function (with d denoting the dimension of the scalar ﬁeld). pi ) denote the 1PI vertex with a ∆ insertion (which is deﬁned in (17. then using the deﬁnitions (we are not writing explicitly a subscript “c” to denote it is a connected Green’s function. (17. (2π) δ (n) pi G∆ (0.115).37)). pi ).

127). Since at the tree level (where d = 1) the n-point function has a canonical dimension of (4 − n) we can write Γ(n) (pi .764 17 Renormalization group and equation This relation is quite easy to understand. In fact.127) we obtain ∂ + nd − 4 Γ(n) ∂pµ i ∂ + nd − 4 Γ(n) ∂m pµ i i = 4−n−m = − m ∂ + n(1 − d) Γ(n) ∂m (17. λ). is given by (2d − 4) so that the naive canonical dimension of the n-point 1PI vertex function is obtained to be (n(d − 4) + 4 − n(2d − 4) = 4 − nd which is reﬂected in (17. m (17. in this case (with d denoting the dimension of the scalar ﬁeld). λ) = ∂m ∂ ∂pµ i m 4−n− pµ i i Γ(n) (pi . (17.131) = −iΓ∆ (0. which can also be written as (n) . This leads to ∂Γ(n) (pi . λ).λ . (17. The canonical dimension of the propagator. the 1PI vertex functions are obtained from the connected Green’s functions by removing external lines (propagators). λ) = m4−n f (n) pi .130) Using this in (17. pi .129) which leads to ∂Γ(n) = ∂pµ i ∂ ∂m pµ i i 4−n−m Γ(n) .128) where f (n) is a dimensionless function (of dimensionless variables).

∂m (17.121) (related to mass renormalization). However. (n) (17. this is not true for the renormalized amplitudes since they depend on mass through logarithm terms (see. λ) = iΓ∆ (0. To understand this question better. as we have seen through explicitly calculations.17. λ).135) where Z represents the renormalization of the composite operator (17. pi ) .132) has the form m ∂ + n(1 − d) Γ(n) = 0. .134) We note that if d = 1 as in the tree level.121) has two extra scalar ﬁelds) Γ(n) → s Γ∆ (n) 4−n 2 2−n 2 . pi .65) and note that a vertex with a ∆ insertion (17. mu ∂Γu ∂mu (n) (n) n n (n) = iΓ∆u (p. (17. pi ) = ZZ32 Γ∆u (p. u Γ∆ (p.132).5 Callan-Symanzik equation 765 m ∂ (n) + n(1 − d) Γ(n) (pi . for example.19)).133) → s so that the leading behavior of the relation (17. pi ) . Thus. (15. let us note that we can relate the renormalized vertex functions to the unrenormalized ones as Γ(n) (pi ) = Z32 Γ(n) (pi ) . Therefore. ∂m (17. we have .132) In the deep Euclidean region where s → ∞ (recall (17.10) or (15. this relation would imply ∂ m ∂m Γ(n) = 0. there must be some violation of the naive scale Ward identity (17.

d = 1 + γ. this shows that there is a violation of the naive scale Ward identity in a renormalized theory. (17. (17. or. in general. Renormalization introduces a mass scale and thereby violates scale invariance. (17. This equation is known as the Callan-Symanzik equation. pi ) = iZZ32 Γ∆u (0.766 17 Renormalization group and equation iΓ∆ (0.137) reduces to the earlier equation (17. pi ) = ZZ32 mu n (n) n (n) n ∂Γu ∂mu (n) = ZZ32 mu = Zmu = Zmu ∂ −n Z3 2 Γ(n) ∂mu ∂Γ(n) nZmu ∂Z3 (n) − Γ ∂mu 2Z3 ∂mu ∂λ ∂ ∂m ∂ + ∂mu ∂m ∂mu ∂λ − n 2 ∂ ln Z3 ∂mu Γ(n) .136) which can be written as m with β(λ) = Zmu m = Zmu γ = ∂λ .139) However. It simply reﬂects the fact that there is an anomaly in the scale Ward identity due to renormalization. ∂mu ∂m . ∂mu (17.132) with the identiﬁcation γ = −(1 − d). .137) Zmu ∂ ln Z3 . 2 ∂mu When β(λ) = 0.138) ∂ ∂ (n) + β(λ) − nγ Γ(n) = iΓ∆ . It is impossible to ﬁnd any regularization which will respect scale invariance which is the essence of the Callan-Symanzik equation. ∂m ∂λ (17.

1981). C. Coleman. S. R. Callan.6 References 1. H. Shirkov. Gell-Mann and F. C. K. Moscow (1984). eds. 4. Introduction to the theory of Quantized Fields. 1300 (1954). Helvetica Physica Acta 26. eds. 3. 6. Stuckelberg and A. G. Guth. Symanzik. 9. S. S. L. 42 (1970). Jaﬀe. Zinn-Justin. V. 1983). MIT Press (Cambridge. Jackiw. M. 499 (1953). . Cambridge University Press (Cambridge. 7. 1541 (1970). Callan. Nauka. 5. 8. E. J. G. Low. Annals of Physics 59. N.6 References 767 17. G. Coleman and R. Physical Review 95. Weinberg in Asymptotic realms of physics: essays in honor of Francis E. N. 2. Bogoliubov and D. 1985). K. 227 (1970).17. Peterman. Aspects of symmetry. C. Physical Review D2. Huang and R. Balian and J. Low. Gross in Methods in ﬁeld theory. D. North Holland Publishing (Amsterdam. A. Communications in Mathematical Physics 18.

This page intentionally left blank .

231 1PI vertex function. 20 Asymptotically free theory. 229. 230 V − A theory. 759 Canonical dimension. 479 Electric charge. 726 Anomalous magnetic moment. 229 Adjoint representation. 299 Bhabha scattering. 449 Charge operator. 521. 536 “In” and “out” states. 725 Bare ﬁeld. 597 MS scheme. 324 BRST charge. 479 ’t Hooft gauge. 669 Canonical Poisson bracket. 552 o BRST transformation Nilpotency. 480 Equality of life times. 662 769 . 726 Anti-BRST transformation. 147. 684 Bare parameter. 546. 739 φ4 theory. 36 Anomalous current. 151 Charge. 353 Bianchi identity. 285 Anti-commutator. 263. 757 Axial gauge. 267. 130 Adiabatic hypothesis. 220. 168 Casimir of representation. 500 Adjoint spinor. 556 Callan-Symanzik equation. 724 Anomaly. 684 Baryon number. 457 Weyl fermions. 446 Dirac ﬁeld. 436 Dirac bilinears. 437 Weak current. 533 Bjorken-Drell metric. 553 BRST invariance. 517. 548. 545 N¨ther current. 111 Anomalous scale dimension. 594 ’t Hooft trick. 313. 549 Anti-commutation relation. 442 Eigenstates. 545 BRST symmetry. 138. 752 Anomalous Ward identity. 167. 714 Born amplitude. 219 CPT theorem. 480 Equality of masses. 5 Bogoliubov’s R operator. 594 S-matrix. 578 Axial vector current. 214 Charge conjugation. 453 Spin zero ﬁeld.Index Rξ gauge. 733 Born approximation. 499 Casimir operator. 567 Abelian group. 266 Charge renormalization.

305 Contragradient. 567 tion. 167. 379 Quantization. 27 Einstein relation. 92 Dispersion relation. 23.Eﬀective action. 503 Spin. 40 Electron in magnetic ﬁeld. 395 Continuity equation. 72 Discrete symmetry. 285 Cini-Touschek transformation. 84 Electric dipole interaction. 297 Classical ﬁeld. 20 Field decomposition. 733 Pauli-Dirac representation. 308 Quantization. 566 Covariant anti-commutation relation. 328 Solution. 308 Negative. 308 Chiral transformation. 116 Charge operator. 7 Green’s function. 407 Dirac bracket. 725 Massless. 168 Normal ordered product. 99. 84 Relativistic particle. 351 Conjugate momentum. 107 Dirac ﬁeld Energy eigenfunction. 137 Covariance. 100 Right-handed. 386 Completeness relation. 34 Ehrenfest theorem. 11 . 57 Covariant quantization. 666 Left-handed. 7 Dirac matrices. 681. 474 Maxwell ﬁeld theory. 385 D’Alembertian. 389 Particle on a sphere. 360. 415 Helicity. 683 Properties. 49 Coulomb gauge. 214 Canonical Hamiltonian. 303 Cliﬀord algebra. 44 Direct sum. 49 Covariant derivative. 282 Weyl representation. 122 Plane wave solution. 57 Counter term. 379 Current. 308 Chirality. 57 Coulomb potential. 19 Primary Hamiltonian. 105 Normalization of wave func. 300 Compton scattering. 111 Properties. 8 Dirac ﬁeld theory. 305 Constrained system. 308 Positive. 252 Non-relativistic limit. 23. 244 Time ordered product. 384.770 Index Chiral anomaly. 100 Dirac ﬁeld theory. 65 Electric ﬁeld. 246 Hole theory. 286 Contraction. 8 Projection operators. 550 Dirac quantization. 517 Majorana representation. 390 Dirac equation. 27 Non-Abelian. 292 Cogradient. 47 Subtracted. 401 Detailed balance.

522 Fermi’s Golden rule. 280. 517 Foldy-Wouthuysen Handedness. 363 Gribov ambiguity. 308 Four vector. 200 Feynman-Fermi gauge. 516 Faddeev-Popov determinant. 194 Feynman parametric integral. 521 Gauge independence. 111 Heisenberg picture. 521 Gauge ﬁxing condition. 321 Negative energy. 522 Fermi-Dirac statistics. 387 Gyro-magnetic ratio. 201 Feynman rule. 165 External line. 733 Faddeev-Popov ghost. 751 Fock space. 580. 218 Euclidean metric. 593 Four vector potential. 626 Euler-Lagrange equation. 299 Ghost scaling charge. 285 Ghost number. 550. 513 Fermi theory. 2 Euler’s constant. 319 Feynman. 330 Gauge orbit. 328 Higgs mechanism. 278 Feynman Green’s function. 268 Goldstone theorem. 511. 101. Grassmann variable. 475 Ghost ﬁelds. 739. 318 Ghost scaling symmetry. 3 Higgs ﬁeld. 593 Feynman parameterization. 553 Feynman diagram. 605 . 283. 286.Index 771 Functional integration. 319 External line factor. 381 632 Green’s function. 593 Functional diﬀerentiation. 517 Gauge ﬁxing parameter. 704 Advanced. 190 Energy-momentum operator. 575 Goldstone ﬁeld. 193 Haar measure. 330 Gell-Mann-Low equation. 518 Gauge transformation. 373 First class constraint. 589. 550 Feynman gauge. 566 Fermi coupling. 223 Forest. 205 QED. 597 Green’s function. 520 Field strength tensor. 110 Fixed point. 349 Retarded. 514 Furry’s theorem. 596 Ghost action. 320 Gauge ﬁxing action. 688 Helicity. 3 Euclidean space. 522 Generating functional. 600 Four velocity. 395 Higgs phenomenon. 205 Feynman rules Positive energy. 99 transformation. 456. 679 Future light-cone. 513 Higgs potential. 198