This page intentionally left blank

University of Rochester, USA

Ashok Das

World Scientific
NEW JERSEY

LONDON

SINGAPORE

BEIJING

SHANGHAI

HONG KONG

TA I P E I

CHENNAI

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

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

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

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

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

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

Printed in Singapore.

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

This page intentionally left blank

Preface

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

viii

Preface

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

Contents

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

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

2

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

3

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

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

x

Contents

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

5

6

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

Contents

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

7

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

8

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

9

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

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

10 Dirac 10.1 10.2 10.3 10.4 10.5 10.6

xii

Contents

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

Contents

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

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

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

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

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

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

This page intentionally left blank

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

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

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

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

A). (Two Lorentz indices are said to be contracted if a contravariant and a covariant index are summed over all possible values. (1.16) we can define an invariant scalar product of the two vectors as A · B = Aµ Bµ = Aµ B µ = A0 B 0 − A · B.) . any product of Lorentz tensors defines a scalar if all the Lorentz indices are contracted. (As is clear from the above discussion. B µ = (B 0 . however. Physical results.1. B). such a product is easily seen to be invariant under Lorentz transformations. namely. (1. In fact. namely. 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. = η µν Aµ Bν = ηµν Aµ B ν (1. use different metric conventions and you should be careful in reading the literature. if there is no free Lorentz index. the nonuniqueness in the choice of the metric tensors reflects the nonuniqueness of the embedding of space and time components into a four vector. (1. ηµν = ηνµ .) Given two arbitrary four vectors Aµ = (A0 .14) Furthermore. each is symmetric as they are expected to be.2 Notations 5 η µλ ηλν = δµν . are independent of the choice of a metric. η µν = η νµ . Different authors. however.17) Since the contravariant and the covariant vectors transform in an inverse manner.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.

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

for the negative energy solutions we write p0 = E− = −E = − p 2 + m2 .22) v (p) = − ˜ σ3 p 3 E+ + m 0 1 = 0 p3 E+ + m .32 2 Solutions of the Dirac equation σ3 p 3 v (p) = − ˜ E+ + m and 1 0 = p3 −E + m + 0 .) ˜ Similarly. for example.18).23) This determines the two positive energy solutions of the Dirac equation (remember that the energy eigenvalues are doubly degenerate). v (p) = ˜ . from the second of the two two-component Dirac equations in (2. (2.24) and the set of equations (2. u ˜ σ3 p3 u(p) + (E− + m)˜(p) = 0.26) Choosing the independent solutions as 1 0 0 1 v (p) = ˜ . Thus.27) .25) (2. ˜ v We can solve these as u(p) = − ˜ σ3 p 3 v (p). u can be thought of as the independent solution. (2. (The question of which components can be chosen independently follows from an examination of the dynamical equations. (2.9) becomes (E− − m)˜(p) + σ3 p3 v (p) = 0. Thus. ˜ E− − m (2. we note that v must vanish in the rest frame while u remains ˜ ˜ arbitrary. 3 (2.

−   E+ + m  0  u↓ +    =       . v (p) ˜ (2.1 Plane wave solutions 33 we obtain respectively p3 −E − m − 0 σ3 p 3 u(p) = − ˜ E− − m and 1 0 = .2.21) and (2.28) u(p) = − ˜ σ3 p 3 E− − m 0 1 = 0 p3 E− − m . Thus.30) u+ (p) = u(p) ˜ .   1   0  0  p3      ↓ u− =  E− − m  .   (2.27) are reminiscent of the spin up and spin down states of a two component spinor. σ3 p 3 ˜ − E + m u(p) + u− (p) = the positive and the negative energy solutions have the explicit forms  1   0 1 0 p3 E+ + m  u↑ (p) +     0   = p3  . from the fact that we can write σ p ˜ − E 3−3m v (p) − . The independent two component wave functions in (2.31) p3  −E − m −     0   ↑ u− (p) =  .   0   1 (2. (2.29) and these determine the two negative energy solutions of the Dirac equation.32) . (2.

) From the structure of the wave function. the solutions take the forms (with p0 = E± = ± p2 + m2 ) u+ (p) = u(p) ˜ .35) then. for the case of general motion. unlike the nonrelativistic counting (2s + 1). we can write the solutions for motion along a general direction as u(p) ˜ . namely.) 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.34) v (p) ˜ (2. (2.) 2. u− (p) = β σ·p ˜ E + m u(p) σ· − E + p v (p) m ˜ v (p) ˜ u+ (p) = α . (2. pi = −pi . u− (p) = σ·p u(p) ˜ E+ + m σ·p ˜ E− − m v (p) .36) Here α and β are normalization constants to be determined. (We will determine the spin of 2 the Dirac particle shortly. (2. where p1 = p2 = 0.34 2 Solutions of the Dirac equation The notation is suggestive and implies that the wave function corresponds to that of a spin 1 particle.2 Normalization of the wave function Let us note that if we define E = E+ = p2 + m2 = −E− . relativistic particle of spin s in the presence of both positive and negative energies follows to be 2(2s + 1).27) respectively are ˜ ˜ normalized as (for the same spin components) .21) and (2. (The correct counting for the number of components of the wave function for a massive.33) which can be explicitly verified. (The change in the sign in the dependent components comes from raising the index of the momentum. it is also clear that. The two component spinors u(p) and v (p) in (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).40) p2 +1 (E + m)2 2E |β|2 v † (p)˜(p). for the negative energy solutions we have σ·p v = β β −˜† (p) E + m ∗ (2. ˜ v E +m . ˜ u ˜ v For different spin components.2. Similarly. we can now calculate σ·p ˜ ˜ u† (p)u+ (p) = α∗ α u† (p) u† (p) E + m + = |α|2 u† (p)˜(p) + u† (p) ˜ u ˜ = |α|2 1 + = |α|2 = |α|2 = p2 (E + m)2 (2. ˜ u E+m (2.39) 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. this product vanishes. Given this.2 Normalization of the wave function 35 u† (p)˜(p) = 1 = 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).

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

47) With the requirement (2. + − (2.2. ˜ u E +m Similarly. m † d3 x ψp (x)ψp′ (x) = (2.43) 2m |α|2 u† (p)˜(p).40) (for the same spin components when (2. this is not a relativistic normalization. − m u† (p)u+ (p) = + (2. we can show that 2m |β|2 v † (p)˜(p). In fact.) The independent wave functions for a free particle. in the rest frame of the particle.38) and (2. we determine from (2.46). as we will see shortly.45) However. a relativistically covariant normalization would be to require (for the same spin components) E = u† (p)u− (p). therefore.44) Our naive instinct will be to normalize the wave function. must be positive. By the way. is related to the probability density which transforms like the time component of a four vector. by requiring (for the same spin components) u† (p)u+ (p) = 1 = u† (p)u− (p). with this normalization condition. u† u. as we will see. this will reduce to u† u± = 1 which corresponds to ± the non-relativistic normalization.46) (Remember that this will correspond to the probability density and. the motivation for such a normalization condition comes from the fact that.37) holds) . would give (for the same spin components) E (2π)3 δ3 (p − p′ ). Thus. as in the non-relativistic case. ψp (x) = e−ik·x u(p) with p0 = ±E. ˜ v E+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.

E+m E + m 2m 2m 2m E + m |β|2 = − = −1. appears to be a Lorentz invariant (scalar) and we will see later that this is indeed true. u† (p)u− (p) = − E+m m β = β∗ = E+m .48) or. For example. therefore.44) (for the same spin components) u+ (p)u+ (p) = 2m 2m E + m |α|2 = = 1. E+m E + m 2m (2. E+m m E+m . we will obtain from (2.38 2 Solutions of the Dirac equation u† (p)u+ (p) = + or.50) u− (p)u− (p) = − This particular product. α = α∗ = 2E E |α|2 = . v (p) ˜ (2. 2m E 2E |β|2 = . Let us also note here that by construction the positive and the negative energy solutions are orthogonal. . with this normalization. we can write the normalized positive and negative energy solutions of the Dirac equation to be u+ (p) = E+m 2m E+m 2m u− (p) = u(p) ˜ . with this normalization. Therefore.43) and (2. σ·p ˜ E + m u(p) σ·p − E + m v (p) ˜ . 2m (2.49) It is also clear that.

54) or. α = α∗ = 2E |α|2 = E. 2 (2. (There is no rest frame for a massless particle. however.40) (for the same spin components) u† (p)u+ (p) = + or. we will obtain from (2. .52) While we will be using this particular normalization for massive particles.) The probability density has to be well defined.2. an alternative normalization which works well for both massive and massless particles is given by u† (p)u+ (p) = E = u† (p)u− (p).51) Therefore.53) This still behaves like the time component of a four vector (m is a Lorentz scalar). u† (p)u− (p) = − E+m β = β∗ = E+m .2 Normalization of the wave function 39 u† (p)u− (p) + = α∗ β = 0. 2 2E |β|2 = E. In this case. that σ·p v (p) = 0. =α β ∗ u† (p) ˜ u† (p) ˜ σ·p E+m σ·p − E + m v (p) ˜ v (p) ˜ − u† (p) ˜ σ·p σ·p v (p) + u† (p) ˜ ˜ v (p) ˜ E+m E+m (2. Correspondingly. let us note that it becomes meaningless for massless particles. − + (2. ˜ E+m σ·p u− (p)u+ (p) = −2β ∗ α v † (p) ˜ u(p) = 0. ˜ E+m u+ (p)u− (p) = −2α∗ β u† (p) ˜ (2.38) and (2. the solutions we have constructed correspond to four linearly independent. Note. E+m E+m . orthonormal solutions of the Dirac equation.

E +m 2m |β|2 = −m. This product continues to be a scalar. d3 x ψ † (x)ψ(x) = 1. 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. Let us note here parenthetically that. we will use the notations of three dimensional .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. in this case. Let us define a four dimensional generalization of the Pauli matrices as (in this section. while the arbitrariness in the normalization of u(p) may seem strange. ψ(x) = (2. we obtain 2m |α|2 = m. the structure of the plane wave solutions of the Dirac equation is suggestive of the fact that the particle described by the Dirac equation has spin 1 . The true normalization is really contained in the total probability. (2.57) Namely.40 2 Solutions of the Dirac equation Correspondingly. Let us note once again that this is a particularly convenient normalization for massless particles. 2. it can be understood in light of what we have already pointed out earlier as follows.3 Spin of the Dirac particle As we have argued repeatedly.55) which vanishes for a massless particle. That this is indeed true 2 can be seen explicitly as follows. u− (p)u− (p) = − E+m u+ (p)u+ (p) = (2. We can write the solution of the Dirac equation for a general motion (along an arbitrary direction) as d4 p a(p)δ(p2 − m2 ) e−ip·x u(p).

60) We note that ρ2 = ½ so that we can invert the defining relation (2. the generalized ˜ Pauli matrices. (2. σj ] 0 2iǫijk σk 0 0 [σi . i = 1. let us note that . however.) Furthermore. of course. (Note. αi .2. 2. that this defines a reducible representation of spin generators since the matrices are block diagonal. be checked readily that this is related to the αi matrices defined in (1.3 Spin of the Dirac particle 41 Euclidean space since we will be dealing only with three dimensional vectors) σi 0 0 σi αi = ˜ .101) through the relation 0 σi σi 0 αi = where = ρ˜ i = αi ρ.98) and (1. 3. ˜ (2.59) ρ= 0 ½ ½ 0 . σj ] 0 2iǫijk σk = 2iǫijk αk . ˜ (2. (2.61) From the structures of the matrices αi and αi we conclude that ˜ [σi . αj ˜ ˜ = = This shows that 1 αi satisfies the angular momentum algebra (remem2˜ ber = 1) and this is why we call the matrices.62) αi . α ˜ (2.58) It can.59) and write αi = ραi = αi ρ.

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

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

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

(We are. and note that the Hermitian conjugate of ψ satisfies (2.77) (2.80) . J = ψ † αψ = probability current density. It is worth recalling that we have already used this Lorentz transformation property of ρ in defining the normalization of the wave function.77)) and. is the time component of j µ (see (2.4 Continuity equation 45 ρ = ψ † ψ = probability density. ρ. of course.76) so that the continuity equation can be written in the manifestly covariant form ∂µ J µ = 0. J) = (ψ † ψ. ψ † αψ). 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. the total probability d3 x ρ = d3 x ψ † ψ. must transform like the time coordinate under a Lorentz transformation.78) This. yet to show that j µ transforms like a four vector which we will do in the next chapter. (2. shows that the probability density.75) (2.) On the other hand. (2. in fact. P = (2.79) is a constant independent of any particular Lorentz frame. to write the continuity equation as ∂ρ = −∇ · J. therefore. ∂t This suggests that we can write the current four vector as J µ = (ρ.2.

the covariant continuity equation and we can identify J µ = ψγ µ ψ.84) This is.85) . i∂µ ψγ µ ψ = 0. 3) γ 0 γ µ γ 0 = (γ µ )† . Note from the definition in (2. (2.86) (2.77).81) Multiplying this equation with γ 0 on the right and using the fact that (γ 0 )2 = ½. (2.83) Multiplying (2. (2. which is what we had derived earlier in (2.85) that J 0 = ψγ 0 ψ = ψ † γ 0 γ 0 ψ = ψ † ψ = ρ. (2. J = ψγψ = ψ † γ 0 γψ = ψ † αψ. 2. we obtain ← − i ψγ µ ∂µ ψ + ψγ µ ∂µ ψ = 0. γ 0 (γ µ )† γ 0 = γ µ . in fact. or.80) with ψ on the left and (2. or. ← − ψ − iγ µ ∂µ − m = 0. 1. or. ∂µ ψγ µ ψ = 0.46 2 Solutions of the Dirac equation ← − ψ † − i(γ µ )† ∂µ − m = 0. we obtain ← − ψ − iγ 0 (γ µ )† γ 0 ∂µ − m) = 0.82) with ψ on the right and subtracting the second from the first.82) where we have used the property of the gamma matrices that (for µ = 0. (2.

we note that the energy spectrum .87) Thus. even for this simple case of a free particle the energy spectrum has the form shown in Fig.2. the energy eigenvalues are given by p0 = E± = ±E = ± p2 + m2 . the probability density is independent of time derivative much like the Schr¨dinger equation.5 Dirac’s hole theory We have seen that Dirac’s equation leads to both positive and negative energy solutions. in such a case.38) and (2. for example. can be defined to be positive definite even in the presence of negative energy solutions. (2. Even when the probability density is consistently defined. First of all.1.1: Energy spectrum for a free Dirac particle. as we o have seen explicitly in (2.5 Dirac’s hole theory 47 Let me conclude this discussion by noting that although the Dirac equation has both positive and negative energy solutions. 2. We note from Fig. Consequently. m −m Figure 2. 2. This is rather different from the case of the Klein-Gordon equation that we have studied in chapter 1. 2.40).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). the presence of negative energy solutions leads to many conceptual difficulties. In the free particle case. the probability density. because it is a first order equation (particularly in the time derivative).

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

1.) This is Dirac’s theory of electrons and works quite well.) In spite of this. The amount of energy necessary to excite a negative energy electron to a positive energy state is E ≥ 2m.2. Thus. .88) Tr γ µ = 0. the four Dirac matrices satisfy (in addition to the Clifford algebra) (γ 0 )† = γ 0 . Furthermore. As we have seen. γ µ .6 Properties of the Dirac matrices 49 ple. in defining a first order equation. this is a general feature that combining quantum mechanics with relativity necessarily leads to a many particle theory. the Dirac equation passes as a one particle equation primarily because of the Pauli exclusion principle. if enough energy is provided to such a ground state. However. thereby. (γ i )† = −γ i . i = 1. This “hole” state is what we have come to recognize as the anti-particle – in this case. 2. 2. (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. 3.6 Properties of the Dirac matrices The Dirac matrices. µ = 0. 3. a negative energy electron can make a transition to a positive energy state and can appear as a positive energy electron. we will study some of the useful properties of these matrices. 2. the Dirac theory predicts an anti-particle of equal mass for every Dirac particle. 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. a positron – and the process under discussion is commonly referred to as pair creation (production). 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. In this section. (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. were crucial in taking a matrix square root of the Einstein relation and. (2.

To obtain the other basis matrices.91) where we have used the property of the Pauli matrices σ1 σ2 σ3 = i½. the identity matrix will correspond to one of them.89) represents the four-dimensional generalization of the Levi-Civita tensor.60). Note that. Let γ5 = iγ 0 γ 1 γ 2 γ 3 = − where ǫ0123 = 1 = −ǫ0123 .90) i ǫµνλρ γ µ γ ν γ λ γ ρ .91). † γ5 = γ5 .91) that we can identify this with the matrix ρ defined earlier in (2. Note that in our particular representation for the γ µ matrices given in (1. by definition. (2. 2 γ5 = ½. Of course. (2.92) We recognize from (2. let us define the following sets of matrices. 4! (2. a complete set of Dirac matrices must consist of 16 such matrices. 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 .50 2 Solutions of the Dirac equation Since these are 4 × 4 matrices. (2. (2.93) .

as (µ.6 Properties of the Dirac matrices 51 and that. ν = 0. γ µ and γ5 in our representation. it anticommutes with any one of them. 3) σ µν = −σ νµ = i i γ µ . Namely. we can define four new matrices as γ5 γ µ . j. 1. 1. σ µν .94) Given the matrix γ5 . since it is the product of all the four γ µ matrices. γ ν = (γ µ γ ν − γ ν γ µ ) 2 2 (2. (2. γ µ + = 0.2. = i η µν − γ ν γ µ whose explicit forms in our representation can be worked out to be (i. k = 1. 2. ½ 0 = σi .97) = −i η µν − γ µ γ ν . let us write out the forms of γ5 γ µ also in this representation. µ = 0.95) Since we know the explicit forms of the matrices ½. 3. (2. 2. γ5 . we can also define six anti-symmetric matrices.96) Finally. (2. 2. γ5 γ 0 = γ5 γ i = 0 ½ ½ ½ 0 −σi 0 ½ 0 0 0 −½ σi 0 = ½ 0 −½ 0 −σi 0 0 . 3) .

(This relation can be obtained from (2. ¯ Here. 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 . therefore. γµ. the notation is suggestive and stands for the fact that ψΓ(S) ψ . γ5 . 6. namely. 0 −σi 0 −σi σj 0 −iǫijk σk σi 0 = σ ij = iγ i γ j = i −σi σj 0 0 −σj σj 0 = i = i −iǫijk σk 0 = ǫijk αk . 1. σ µν . ˜ 2 4 (2. that the matrices 1 1 αi = ǫijk σ jk . in fact.98) we conclude. Γ(S) Γ(V ) Γ(A) = = ½.98) 1 We have already seen in (2. From (2.) We have thus constructed a set of sixteen Dirac matrices. ˜ (2. 4. provide a basis for all the 4 × 4 matrices.100) Γ(T ) = Γ(P ) = = γ5 where the numbers on the right denote the number of matrices in each category and these. namely.70) that the matrices 2 αi represent the ˜ spin operators for the Dirac particle. ǫijk ǫℓjk = 2 δiℓ .99) can be identified with the spin operators for the Dirac particle. 4. 1.98) using the identity for products of Levi-Civita tensors. (2.

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

C] . For example. the form in (2. γν γ5 . since γ µ matrices satisfy simple anti-commutation relations. (2. −i η νλ − γ ν γ λ γλ − γν γµ.54 2 Solutions of the Dirac equation γµ. In this derivation. γ ν γ λ = i γµ.108) However. γ µ . B] C + B [A.107) can also be expressed in terms of commutators (instead of anti-commutators) as [A. for the commutator of two σ µν matrices. σ νλ = γ µ . BC] = ABC − BCA = (AB + BA)C − B(AC + CA) = A. (2. (2.105) We can also calculate various other commutation relations in a straightforward and representation independent manner. B + C − B A.107) We note here parenthetically that the commutator in (2. we have used the fact that (2. γ µ + + = 2η µν ½. we obtain . BC] = [A.107) is more useful for our purpose. C + .106) [A. Similarly. = 0. γν + + = 2i η µν γ λ − η µλ γ ν . γλ = i γ µ.

Thus. σ λρ = i γ µ γ ν .110) (2.6 Properties of the Dirac matrices 55 σ µν . + ν − γµγν (2.112) . where we have used (γµ = ηµν γ ν ) γµ γ µ = 4 ½. γµ A µ = γµ Aν γ ν γ µ = Aν γµ γ ν γ µ = −2Aν γ ν = −2A /γ /.111) (2. and it follows now that. for example. The various commutation and anti-commutation relations also lead to many algebraic simplifications in dealing with such matrices. σ λρ = − i η µν − γ µ γ ν . we see that the σ µν matrices satisfy an algebra in the sense that the commutator of any two of them gives back a σ µν matrix. This becomes particularly useful in calculating various amplitudes involving Dirac particles. Similarly.2. (these relations are true only in 4-dimensions) γµ γ ν γ µ = γµ γ ν . (2. σ λρ + i γ µ .109) Thus. 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 . γ µ = 2η νµ γµ − 4γ = 2γ ν − 4γ ν = −2γ ν . We will see in the next chapter that they provide a representation for the Lorentz algebra.

For example. = (2. γ ν 2 2 1 = Tr 2η µν ½ = η µν Tr ½ = 4η µν . we note that Tr γ µ γ ν γ λ γ ρ = Tr γµ. − γργµ Tr γ µ γ ν γ λ γ ρ = 4 η µν η λρ − η µλ η νρ + η νλ η µρ . The commutation and anticommutation relations also come in handy when we are evaluating traces of products of such matrices. 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 . it follows (in 4-dimensions) that 1 1 Tr γ µ γ ν + Tr γ ν γ µ = Tr γ µ . or.116) . γ µ + − γµγλ = 4η λν ½. 2 = Tr γ µ γ5 = −Tr γ5 γ µ = 0.56 2 Solutions of the Dirac equation γµ γ ν γ λ γ µ = γµ γ ν γ λ . (2. Therefore. we know from the cyclicity of traces that Tr γ µ γ ν = Tr γ ν γ µ .113) = 2η λµ γµ γ ν + 2γ ν γ λ = 2γ λ γ ν + 2γ ν γ λ = 2 γ λ . 2 Tr γ µ γ ν γ λ γ ρ = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ . = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ − Tr γ ν γ λ γ ρ γ µ .114) Tr γ µ γ ν + Tr γ5 γ µ (2.115) Even more complicated traces can be evaluated by using the basic relations we have developed so far. γ λ + = 8η µν η λρ − 8η µλ η νρ + Tr γ ν γ λ γ µ . For example. γ ρ or. (2. γ ν + and so on.

1 S = √ γ0 2 ½ + γ2 . 0 σµ ˜ σµ 0 γW = where µ . 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 = . To conclude this section.120) This is known as the Weyl representation for the Dirac matrices and is quite useful in the study of massless fermions. (2. namely.117) It can be checked that the Dirac matrices in the Pauli-Dirac representation and the Majorana representation are related by the similarity (unitary) transformation µ γM = Sγ µ S −1 . It can be checked . However. there exists a representation for the Dirac matrices where γ µ are all purely imaginary. there exists yet another representation for the γ µ matrices. 1 γM = . −σ). We would use all these properties in the next chapter to study the covariance of the Dirac equation under a Lorentz transformation. This is known as the Majorana representation and is quite useful in the study of Majorana fermions which are charge neutral fermions.2.118) Similarly.119) σ µ = (½. σ µ = (½. (2. there are other equivalent representations possible which may be more useful for a particular system under study. ˜ (2. (2. For example. let us note that we have constructed a particular representation for the Dirac matrices commonly known as the Pauli-Dirac representation. σ).6 Properties of the Dirac matrices 57 and so on. 3 γM = . 2 γM = . Explicitly.

V. This is easily demonstrated by showing that they are linearly independent which is seen as follows. A. σµν . the linear independence of the set of matrices in (2. we can construct the inverse set of matrices as Γ(a) . a = S. P define a complete basis for 4 × 4 matrices. σ µν .123) where “Tr” denotes trace over the matrix indices.121) 2. a = b.126) With this. (2. we can write the inverse set of matrices as 1 {½. (2. γµ . As a result. γ5 } .122) From the properties of the γ µ matrices. it follows now that if . T. γ5 } . 4 (b) (2. it can be easily checked that Tr Γ(a) Γ(b) = 0. γ µ .124) Tr Γ(a) Γ(b) = δ(a) . We have explicitly constructed the sixteen matrices to correspond to the set Γ(a) = {½. For example. given this set of matrices.100). As we have pointed out in (2.125) Γ(a) = (2.6. Tr Γ(a) Γ(a) Γ(a) = such that a not summed. (2.58 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 . −γ5 γµ . Explicitly. 1 S=√ 2 ½ + γ5 γ 0 .122) is straightforward. (2.1 Fierz rearrangement. the sixteen Dirac matrices Γ(a) . γ5 γ µ .

V.129). (2. we obtain C(a) Tr Γ(b) Γ(a) = (a) (a) (M ) (M ) (a) Tr Γ(b) M = or.122) provide a basis for the 4 × 4 matrix space.129) Multiplying this expression with Γ(b) and taking trace over the matrix indices.127) implies that all the coefficients of expansion must vanishing which shows that the set of sixteen matrices Γ(a) in (2. (a) (2. multiplying (2. or. (2. (a) (a) or.2.125) we obtain C(a) Γ(a) = 0. any arbitrary 4 × 4 matrix M can be expanded as a linear superposition of these matrices. Substituting (2. (a) C(a) Tr Γ(b) Γ(a) = 0.6 Properties of the Dirac matrices 59 C(a) Γ(a) = 0. we obtain .127) then. M= (a) C(a) Γ(a) . C(a) δ(b) = 0. As a result they constitute a basis for 4 × 4 matrices. T.122) are linearly independent. A.130) into the expansion (2. P . and taking trace over the matrix indices and using (2. (2. Therefore. namely. where b is arbitrary. Since the set of matrices in (2.127) with Γ(b) .130) C(b) = Tr Γ(b) M .128) for any b = S. C(b) = 0. (a) Tr Γ(b) or. (M ) (2. (M ) C(a) δ(b) .

(2. 2.60 2 Solutions of the Dirac equation M= (a) Tr Γ(a) M Γ(a) . β. (a) Γ(a) γη Γ(a) αβ = δαη δβγ . Just like any other completeness relation. η = 1. γ. 3. it is now straightforward to obtain . this leads to Mαβ = (a) Γ(a) (a) γη Mηγ Γαβ . ¯ ¯ ¯ ¯ (2.131) Introducing the matrix indices explicitly. Equation (2.132) describes a fundamental relation which expresses the completeness relation for the sixteen basis matrices. (2. we note that if M and N denote two arbitrary 4 × 4 matrices. For example.133). ¯ ¯ ¯ ¯ M Γ(a) αδ N Γ(a) ¯ δδ γβ = Mαδ Γ(a) ¯ Nγ β Γ(a) ¯ ¯ ββ = Mαδ Nγ β Γ(a) ¯ ¯ ¯ δδ Γ(a) ¯ ββ = Mαδ Nγ β δδβ δδβ = Mαβ Nγδ .133) Using the relations in (2.132) Here α. 4 and correspond to the matrix indices of the 4 × 4 matrices and we are assuming that the repeated indices are being summed. then using (2. or. we will use the standard convention that the repeated index (a) as well as the matrix indices are being summed) Γ(a) M = γβ Γ(a) N ¯ γβ αδ ¯ αδ Γ(a) Mββ Γ(a) ¯ Nδδ = Mββ Nδδ Γ(a) ¯ ¯ ¯ ¯ γβ Γ(a) ¯ αδ = Mββ Nδδ δγ δ δβα = Mαβ Nγδ . it can be used effectively in many ways.132) we can derive (for simplicity.

135) equivalently as . In deriving these identities. (2. the right-hand sides of the identities in (2.134) as (assuming the spinors are ordinary functions) ¯ ¯ ψ1 M ψ2 ψ3 N ψ4 = 1 ¯ ¯ ¯ ¯ ψ1 M ψ4 ψ3 N ψ2 + ψ1 M γ µ ψ4 ψ3 N γµ ψ2 4 ¯ ¯ ¯ ¯ +ψ1 M σ µν ψ4 ψ3 N σµν ψ2 − ψ1 M γ5 γ µ ψ4 ψ3 N γ5 γµ ψ2 ¯ ¯ +ψ1 M γ5 ψ4 ψ3 N γ5 ψ2 . we have assumed that the spinors are ordinary functions. N and any spinors. Let us note that using the explicit forms for Γ(a) and Γ(a) in (2.124) respectively.134) The two relations in (2. (2.6 Properties of the Dirac matrices 61 ¯ ψ1 M Γ(a) ψ4 ¯ ψ3 N Γ(a) ψ2 αδ ¯ = ψ1α M Γ(a) ¯ ψ4δ ψ3γ N Γ(a) αδ γβ ψ2β γβ ¯ ¯ = ψ1α ψ4δ ψ3γ ψ2β M Γ(a) N Γ(a) ¯ ¯ ¯ = ψ1α ψ4δ ψ3γ ψ2β Mαβ Nγδ = ψ1 M ψ2 ¯ ψ1 Γ(a) M ψ4 ¯ ψ3 Γ(a) N ψ2 γβ ¯ ψ3 N ψ4 .134) pick up a negative sign which arises from commuting the fermionic fields past one another.134) are known as the Fierz rearrangement identities which are very useful in calculating cross sections. On the other hand.122) and (2.2.135) Since this is true for any matrices M. we can ˜ define a new spinor ψ4 = N ψ4 to write the identity in (2. if they correspond to anti-commuting fermion operators. ¯ = 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 . we can write the first of the Fierz rearrangement identities in (2.

Drell. J.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 . D. Relativistic Quantum Mechanics. New York.136) ¯ ¯ ψ1 (½ − γ5 ) γ µ ψ2 ψ3 (½ − γ5 ) γµ ψ4 ¯ ¯ = −ψ1 (½ − γ5 ) γ µ ψ4 ψ3 (½ − γ5 ) γµ ψ2 .7 References 1. L. I. we obtain from (2. 2.110) and (2. (2. choose M = (½ − γ5 ) γ µ . if we which is often calculationally simpler. for example. McGraw-Hill. New York. Bjorken and S. N = (½ − γ5 ) γµ . 1964. Quantum Mechanics. 2.113).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). 1968. .137) then using various properties of the gamma matrices derived earlier as well as (2.136) Thus. McGraw-Hill. Schiff. (2. (2.

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

This page intentionally left blank .

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

3 x′ = ǫx1 + x2 . Under a Lorentz boost along the x-axis. 2 (3. such that (3. We observe here that the matrix representing the infinitesimal change under a rotation is anti-symmetric.2) that an infinitesimal rotation around the 3-axis (z-axis) takes the form x′ = x1 − ǫx2 . R (RT R = ½).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. we also know that the coordinates transform as (boost velocity v = β since c = 1) xµ → x′ µ . x′ 1 = −γβx0 + γx1 .6) where the Lorentz factor γ is defined in terms of the boost velocity to be 1 1 − β2 γ= .6) can also be written in the matrix form as . We also note from (3.7) We recognize that (3.4) where we have identified θ = ǫ = infinitesimal. with unit determinant.5) x′ 0 = γx0 − γβx1 . x′ 3 = x3 . 1 x′ = x3 . x′ 2 = x2 . (3. (3.

where  cosh ω − sinh ω 0 0 cosh ω 0 0  (3. 1 − β2 1 − β2 (3.8) 1 1 − β2 β .11) 1 β2 − = 1.9) that −∞ ≤ ω ≤ ∞. Thus.9) 1 − β2 cosh2 ω − sinh2 ω = Since the range of the boost velocity is given by −1 ≤ β ≤ 1. (3. we conclude from (3. we note that a Lorentz boost along the x-direction can be written as x′ µ = Λµν xν . . =   0 0 1 0 x2     0 0 0 1 x3 x′ 0   γ (3. 1 0  0 1 (3.1 Lorentz transformations 67  where we have defined cosh ω = γ = sinh ω = γβ = so that   0 x −γβ 0 0  ′ 1   1  x  −γβ γ 0 0 x    =    x′ 2   0 0 1 0 x2       ′3 x 0 0 0 1 x3   0  x cosh ω − sinh ω 0 0   1  − sinh ω cosh ω 0 0 x   .3.12) .10) Λµν  − sinh ω =  0  0  0 0 .

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

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

In the same spirit. orthogonal means the corner that is straight up (perpendicular). and (3. Choosing ρ = σ = 0. an orthodontist is someone who can make your teeth straight. or. det Λ = ±1.70 3 Properties of the Dirac equation This is. we obtain (for clarity.20) explicitly as Λ00 Λ00 + Λi 0 Λi 0 = 1.) Note also that since ΛT Λ = ½. we note that ΛT as can be seen from (3. (3. Namely.22) If Λ00 ≥ 1. therefore. “gonia” in Greek means an angle or a corner and. Lorentz transformations define the maximal symmetry of the space-time manifold which leaves the origin invariant. 2 = 1 + Λi 0 ≥ 1. Λ00 Λ00 2 2 − Λi 0 2 = 1.15). such a Lorentz transformation does not change the direction of time.20)) (det Λ)2 = 1.23) = Λνµ and ΛT ν µ = Λνµ The set of homogeneous Lorentz transformations satisfying Λ00 ≥ 1. Λ00 ≤ −1. Incidentally. Thus. we conclude that Λ00 ≥ 1. what we have seen before in (3. then the transformation is called orthochronous. (3. of course. (The Greek prefix “ortho” means straight up. (3.21) Therefore. orthochronous means straight up in time. we can write out the relation (3. det Λ = 1. or. or.24) µ ν (3.25) . or.

we obtain a Lorentz transformation   1 0 0 0   0 −1 0 0  Λprop . to a proper orthochronous Lorentz transformation. we obtain a Lorentz transformation . det Λ = ±1.3.26) Λ00 ≤ −1. det Λ = −1 (which is orthochronous but no longer proper). then by adding space reflection.  =   0 0 1 0 0 0 0 1 Λ = Λtime Λprop (3. there are four kinds of Lorentz transformations.  = 0 0 −1 0    0 0 0 −1 Λ = Λspace Λprop (3. Finally.27) This would correspond to having Λ00 ≥ 1. t → −t. then we obtain a Lorentz transformation   −1 0 0 0    0 1 0 0  Λprop . we note that the set of transformations with det Λ = 1 are known as proper transformations and the set for which Λ00 ≥ 1 are called orthochronous. xµ → −xµ .1 Lorentz transformations 71 are known as the proper. x → −x. to a proper Lorentz tranformation.28) satisfying Λ00 ≤ −1 and det Λ = −1 (which is neither proper nor orthochronous). Λ00 ≥ 1. we can obtain the other Lorentz transformations by simply appending space reflection or time reflection or both (which are discrete transformations). (Just to emphasize. Given the proper orthochronous Lorentz transformations. orthochronous Lorentz transformations and constitute a set of continuous transformations that can be connected to the identity matrix. if we add both space and time reflections. if Λprop denotes a proper Lorentz transformation.) In general. namely. Thus. however. If we add time reversal. (3. det Λ = ±1.

33) (3. These additional transformations. With this general definition. covariance implies that a given equation is form invariant under a particular transformation (has the same form in different reference frames). however.29) where L is a linear operator. let us now consider the Dirac equation iγ µ ∂µ − m ψ(x) = 0. 0  −1 0  (3. with the transformed variables). Under a Lorentz transformation xµ → x′ µ = Λµν xν . (3. we would refer to proper orthochronous Lorentz transformations as the Lorentz transformations.31) where ψ ′ represents the transformed wavefunction and L′ stands for the transformed operator (namely. the operator L. (3.32) .72 3 Properties of the Dirac equation Λ = Λspace−time Λprop with Λ00 ≤ −1 and det Λ = 1 (which is proper but not orthochronous). (3.30)  −1 0 0   0 −1 0 = 0 0 −1  0 0 0  0  Λprop . In simple terms. cannot be continuously connected to the identity matrix since they involve discrete reflections. 3. we say that it is covariant under a given transformation provided the transformed equation has the form L′ ψ ′ = 0. In these lectures.2 Covariance of the Dirac equation Given any dynamical equation of the form Lψ = 0.

under a Lorentz transformation. (with S(Λ) representing the 4 × 4 matrix which rotates the matrix components of the wave function) |ψ → |ψ ′ = L(Λ)S(Λ)|ψ = = L(Λ)S(Λ) dx |x x|ψ dx L(Λ)|x S(Λ)ψ(x) . are a set of four space-time independent matrices and. (3.36) However. the transformed wavefunction has the form ψ ′ (x′ ) = ψ ′ (Λx) = S(Λ)ψ(x). (3.35) where S(Λ) is a 4 × 4 matrix. In the notation of other symmetries that we know from studies in non-relativistic quantum mechanics.3. 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. Note that the Dirac matrices.34) where ψ ′ (x′ ) is the Lorentz transformed wave function. what this means is that we are finding a representation of the Lorentz transformation on the Hilbert space.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. Parenthetically. do not change under a Lorentz transformation. since the Dirac wavefunction is a four component spinor. in addition to the change in the coordinates. Thus. we can define the Lorentz transformation acting on the Dirac Hilbert space (Hilbert space of states describing a Dirac particle) as. the Lorentz transformation can also mix up the spinor components (much like angular momentum/rotation). then the Dirac equation would be covariant under a Lorentz transformation. γ µ . therefore. (3.

38) Namely.37) where the wave function is recognized to be ψ(x) = x|ψ . ∂xµ ∂x′ ν iγ µ Λνµ ∂ν′ − m S −1 (Λ)ψ ′ (x′ ) = 0. or. ′ iΛµν γ ν ∂µ − m S −1 (Λ)ψ ′ (x′ ) = 0. from (3. we can write iγ µ ∂µ − m ψ(x) = 0. but get rotated by the S(Λ) matrix. the effect of the Lorentz transformation. (3. Thus.39) (3.33) that ∂x′ µ = Λµν . 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) we obtain (see (3. (3. the matrix S(Λ) must possess an inverse so that from (3. Let us also note from (3. or. iγ µ (3. A more physical way to understand this is to note that the Dirac wave function simply consists of four functions which do not change.40) (3.35)) ψ ′ (x′ ) = x′ |ψ ′ = S(Λ)ψ(x).74 = 3 Properties of the Dirac equation dx |Λx S(Λ)ψ(x). (3.41) ∂x′ ν ∂ − m S −1 (Λ)ψ ′ (x′ ) = 0. or.42) . so that.35) we can write ψ(x) = S −1 (Λ)ψ ′ (x′ ). Since the Lorentz transformations are invertible. ∂xν define a set of real quantities. or. ′ iΛµν Sγ ν S −1 ∂µ − m ψ ′ (x′ ) = 0. on the wave function.

(3. (Sγ 0 S † γ 0 )γ µ (Sγ 0 S † γ 0 )−1 = γ µ .3. Λνσ γ σ (3.40).15)). or. there must exist a matrix connecting the two representations. γ µ and γ ′µ . by Pauli’s fundamental theorem. then. γ ′ ν Λµρ γ ρ . Therefore. let us note that since the parameters of Lorentz transformation are real (namely. Λµν γ ν = S −1 γ µ S. (3. It now follows from (3. such that Λµν Sγ ν S −1 = γ µ . or.46) † † † . or. the matrices γ ′ µ satisfy the Clifford algebra and. generating Lorentz transformations (for the Dirac wavefunction). (3.44) + = + + µν = Λµρ Λνσ γ ρ .43) Let us note that if we define γ ′ µ = Λµν γ ν . (Λ∗ )µν = Λµν ) γ 0 Λµν γ ν γ 0 = Λµν γ 0 (γ ν )† γ 0 = Λµν γ ν . Sγ 0 S † γ 0 γ 0 γ µ γ 0 γ 0 S −1 γ 0 S −1 = γ µ . γ 0 (S −1 γ µ S)† γ 0 = S −1 γ µ S. Next.2 Covariance of the Dirac equation 75 where we have used (3. γ σ = 2Λµρ Λνρ ½ = 2η = Λµρ Λνσ 2η ρσ ½ ½. or. Therefore. we see from (3. γ ′ µ.45) where we have used the orthogonality of the Lorentz transformations (see (3.43) 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.42) that the Dirac equation will be form invariant (covariant) under a Lorentz transformation provided there exists a matrix S(Λ).

76 3 Properties of the Dirac equation Here we have used (3. As a result.46) that the matrix Sγ 0 S † γ 0 commutes with all the γ µ matrices and. or. determines that the parameter b is real. Sγ 0 S † γ 0 = b∗ ½ = b ½. namely. Using these in (3. (3. det S = 1. b4 = 1. . or.100) and calculating the commutator with γ µ ).50) The real roots of this equation are b = ±1.49) We also note that det γ 0 = 1 and since we are interested in proper Lorentz transformations. we can determine the unique value of b in the following way.47). It is clear from (3.48) which. (3. we can denote Sγ 0 S † γ 0 = b ½. γ 0 Sγ 0 S † = b∗ ½. must be proportional to the identity matrix (this can be easily checked by taking a linear combination of the sixteen basis matrices in (2. S † γ 0 = b γ 0 S −1 . we obtain (Sγ 0 S † γ 0 )† = b∗ ½. we determine det (Sγ 0 S † γ 0 ) = det (b ½). therefore. (3. or.43) and the relations (γ 0 )† = γ 0 = (γ 0 )−1 as well as γ 0 (γ µ )† γ 0 = γ µ .51) In fact. (3. therefore.47). or. γ 0 (γ 0 Sγ 0 S † )γ 0 = b∗ ½. (3.47) Taking the Hermitian conjugate of (3. b = b∗ . or.

or. .54) which implies (see (3. therefore. (3. Λ00 ≤ −1. Thus.43) and (3. b > 0. The two solutions of this equation are obvious Λ00 ≥ 1.53) or. b < 0.55) These are some of the properties satisfied by the matrix S which will be useful in showing that it provides a representation for the Lorentz transformations.52) which follows since S † S represents a non-negative matrix.3. we are assuming Λ00 ≥ 1.47).53)) that b > 0 and. using (3. it follows from (3. Tr S † S = 4bΛ00 0. Since we are dealing with proper Lorentz transformations.2 Covariance of the Dirac equation 77 Let us note. (3. we conclude from (3.47) that Sγ 0 S † γ 0 = ½.51) that b = 1.56) (3. (3. S † γ 0 = γ 0 S −1 . (3. that S †S = S †γ 0γ 0S = bγ 0 S −1 γ 0 S = bγ 0 Λ0ν γ ν = bγ 0 Λ00 γ 0 + Λ0i γ i = b Λ00 + Λ0i γ 0 γ i . or.

57) From our earlier discussion in (3.18). .61) (3. Mµν denote matrices in the Dirac space). 4 (3. (3.60) ½+ ½+ i Mµν ǫµν 4 ½− i Mστ ǫστ 4 i i Mµν ǫµν − Mµν ǫµν + O(ǫ2 ) 4 4 (3.62) (3. Mµν = −Mνµ . We can also write S −1 (ǫ) = ½ + so that S −1 (ǫ)S(ǫ) = = = i Mµν ǫµν . namely.63) ½ + O(ǫ2 ).78 3 Properties of the Dirac equation Next. we can expand the matrix S(Λ) as S(Λ) = S(ǫ) = ½ − i Mµν ǫµν . let us consider an infinitesimal Lorentz transformation of the form (ǫµν infinitesimal) x′µ = Λµν xν = (δµ + ǫµ )xν = xµ + ǫµ xν .58) For an infinitesimal transformation. we recall that the infinitesimal transformation matrix is antisymmetric. therefore. ǫµν = −ǫνµ . since ǫµν = −ǫνµ . ν ν ν (3.59) where the matrices Mµν are assumed to be anti-symmetric in the Lorentz indices (for different values of the Lorentz indices. 4 (3.

or.64). λ 2 ρ = − (3.64) that if we identify M λρ = σ λρ .106) γ µ . i ǫλρ γ µ . 4 At this point.67) which coincides with the right hand side of (3. 4 4 (3.66) (3.2 Covariance of the Dirac equation 79 To the leading order. Therefore.65) − (3. i i λρ ǫ Mλρ γ µ − ǫλρ γ µ Mλρ + 0(ǫ2 ) = γ µ + ǫµν γ ν .68) .64) i ǫλρ γ µ . 4 4 (3. let us recall the commutation relation (2.3. The defining relation for the matrix S(Λ) in (3. we see that for infinitesimal transformations. M λρ = ǫµν γ ν .43) now takes the form S −1 (ǫ)γ µ S(ǫ) = Λµν γ ν = δµν + ǫµν γ ν . or. then. and note from (3. therefore. M λρ 4 i ǫλρ γ µ . or. σ νλ = 2i η µν γ λ − η µλ γ ν . ½+ γµ + − i Mλρ ǫλρ γ µ 4 ½− i Mστ ǫστ 4 = γ µ + ǫµν γ ν . S −1 (ǫ) indeed represents the inverse of the matrix S(ǫ). σ λρ 4 i = − ǫλρ × 2i η µλ γ ρ − η µρ γ λ 4 1 µ ρ = ǫ γ + ǫµ γ λ = ǫµν γ ν . we have determined the form of S(ǫ) to be S(ǫ) = ½ − i i Mµν ǫµν = ½ − σµν ǫµν .

then. 2 (3. be constructed out of a series of infinitesimal transformations and. For completeness.70) with all other components of ǫµν vanishing. In such a case (see also (2. 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. (The other factor of 2 is there to avoid double counting. (3. consequently. i (3.98)). of course.72) . ˜ 2 2 2 (3. satisfy can 2 be identified with the Lorentz algebra (which also explains why they are closed under multiplication).) We will see in the next chapter (when we study the representations of the Lorentz group) that the algebra (2. the Dirac equation is form invariant (covariant) under such a Lorentz transformation. A finite transformation can. 1 σµν .109) which the generators of the infinitesimal transformations.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 . be obtained from an infinite sequence of infinitesimal transformations resulting in an exponentiation of the infinitesimal generators as ˜ S(θ) = e− 2 α3 θ .80 3 Properties of the Dirac equation Let us note here from the form of S(ǫ) that we can identify 1 σµν .43) and generates Lorentz transformations and as a result. S(ǫ) = ½ − i 1 i σ12 ǫ = ½ + γ1 γ2 ǫ = ½ − α3 ǫ. Thus. we have shown that there exists a S(Λ) which satisies (3. at least for infinitesimal Lorentz transformations.71) A finite rotation by angle θ in the 1-2 plane would.

(3. recalling that σ3 0 0 σ3 α3 = ˜ we have 2 . we can determine S(θ) = cos This shows that S(θ + 2π) = − cos S(θ + 4π) = S(θ). we are considering an infinitesimal boost along the 1-axis. namely. This is. θ θ − i˜ 3 sin α 2 2 = −S(θ). in this case. α 2 2 (3. we have S † (θ) = S −1 (θ). the rotation operator.76) θ θ − i˜ 3 sin .75) That is.98)) S(ǫ) = i 1 σ01 ǫ = ½ + γ0 γ1 ǫ 2 2 1 = ½ − α1 ǫ. of course. namely. rotations ˜i ˜ define unitary transformations. consistent with the fact that the Dirac equation describes spin 1 2 particles. corresponds to a spinor representation.77) with all other components of ǫµν vanishing. (3. therefore. In this case. Let us next consider an infinitesimal rotation in the 0-1 plane. therefore.2 Covariance of the Dirac equation 81 Note that since α† = αi .74) and. (3. 0 ½ (3. we can identify ǫ01 = ǫ = −ǫ10 .3.78) .73) α3 ˜ = ½ 0 = ½. is double valued and. so that we can write (see also (2. Furthermore. 2 ½− (3.

ω. This is related to the fact that Lorentz boosts are non-compact transformations and for such transformations. operators defining boosts are not unitary. 0 ½ (3. All the unitary representations are necessarily infinite dimensional.80) and. therefore. we have shown how to construct the matrix S(Λ) for finite Lorentz transformations (for both rotations and boosts). since under a Lorentz transformation . (3. can be obtained through exponentiation as S(ω) = e− 2 α1 ω .82 3 Properties of the Dirac equation In this case.83) That is. 3.79) α1 = .82) We note here that since α† = α1 . (3. α2 = 1 ½ 0 = ½. there does not exist any finite dimensional unitary representation. Furthermore. Let us note next that. the matrix for a finite boost. as 4 × 4 matrices).81) we can determine ω ω − α1 sinh . 1 S † (ω) = S −1 (ω). recalling that 0 σ1 σ1 0 1 (3. 2 2 S(ω) = cosh (3.3 Transformation of bilinears In the last section. in this four dimensional space (namely.

3.50) and (2. ′ (3. namely.3 Transformation of bilinears 83 ψ ′ (x′ ) = S(Λ)ψ(x). This implies that a bilinear product such as ψψ would transform under a Lorentz transformation as ψ (x′ )ψ ′ (x′ ) ψ(x)S −1 (Λ)S(Λ)ψ(x) = ψ(x)ψ(x). Similarly. (3.88) . In other words.85) where we have used the relation (3.56). such a product will not change under a Lorentz transformation – would behave like a scalar – which is what we had discussed earlier in connection with the normalization of the Dirac wavefunction (see (2. we see that the adjoint wave function ψ(x) transforms inversely compared to the wave function ψ(x) under a Lorentz transformation. 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). (3.86) ′ ψ(x)ψ(x) → = Namely. ψ (x′ ) = ψ ′ † (x′ )γ 0 = ψ † (x)S † (Λ)γ 0 = ψ † (x)γ 0 S −1 (Λ) = ψ(x)S −1 (Λ).43).55)). it would transform as a four vector under a proper Lorentz transformation. J µ (x) = ψ(x)γ µ ψ(x) → Λµν J ν (x). Thus. we see that if we define a current of the form j µ (x) = ψ(x)γ µ ψ(x). it follows that ψ ′ † (x′ ) = ψ † (x)S † (Λ). (3.84) (3.87) ′ where we have used (3.

(3. (3.84 3 Properties of the Dirac equation This is. the probability current density (see also (2. of course. It is customary to identify (see (2.91).92) so that the equations satisfied by u(p) and v(p) (positive and negative energy solutions). what we had observed earlier.90). can be written as (p − m)u(p) = 0.89) and (3. p) = 0. / (3.90) (3. / where p0 = ω = p2 + m2 . (3. the reason for this will become clear when we discuss the quantization of Dirac field theory later) E+m 2m u ˜ σ·p E+m u ˜ u(p) = u+ (p) = . we note that in this way. Namely.49). −p) = u− (−p) = . (3.4 Projection operators. 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 . 3. we can determine the transformation properties of the other bilinears under a Lorentz transformation in a straightforward manner. p) = 0.89) while the negative energy solutions satisfy − γ 0 p0 − γ · p − m u− (−p0 .93) .85)) transforms like a four vector so that the probability density transforms as the time component of a four vector.91) with the same value of p0 as in (3. Finally. E+m 2m σ ·p E+m v ˜ v ˜ v(p) = u− (−p0 .

Let us denote them by ur (p) and v r (p). u(p)(p − m) = 0.83)). 2. / or. or. α v r (p)v s (p) = α=1 r s r s v α (p)vα (p) = −δrs .4 Projection operators. 4. / (3.96) where r. v(p)(p + m) = 0. / or. (−p − m)v(p) = 0. (3. u (p)v (p) = 0 = v r (p)us (p). completeness relation 85 and − γ 0 p0 + γ · p − m v(p) = 0. Let us denote the spinor index by α = 1. 3. r = 1. 2. / v † (p) (p)† + m γ 0 = 0.97) .50)) 4 u r (p)us (p) = α=1 4 r uα (p)us (p) = δrs . With these notations. (3. / (p + m)v(p) = 0.3. the adjoint equations are easily obtained to be (taking the Hermitian conjugate and multiplying γ 0 on the right) u† (p) (p)† − m γ 0 = 0. or. As we have seen earlier there are two positive energy solutions and two negative energy solutions of the Dirac equation. Let us also note that each of the four solutions really represents a four component spinor. / (3. we can write down the Lorentz invariant conditions we had derived earlier from the normalization of a massive Dirac particle as (see (2.94) Given these equations. can represent the spin projection of the two component spinors (in terms of which the four component solutions were obtained). as we had seen earlier.95) where we have used (γ µ )† γ 0 = γ 0 γ µ (see (2.

86 3 Properties of the Dirac equation Although we had noted earlier that u+ (p)u− (p) = 0.94). 4 × 4 matrices and their effect on the Dirac spinors is quite clear. 2m −p + m r / v (p) 2m −p − m + 2m r / v (p) = v r (p). namely. −p). (3. From the form of the equations satisfied by the positive and the negative energy spinors. 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). of course.93) and (3. 2m (3.99) for any two spin components of the positive and the negative energy spinors. 2m −p + m / . 2m p+m r / v (p) = 0. This also allows us to write 2 r=1 [u r (p)ur (p) − v r (p)v r (p)] = 4.98) For completeness we note here that it is easy to check u† (p)v(−p) = 0 = v † (−p)u(p). the last relation in (3.97) can be checked to be true simply because v(p) = u− (−p0 . 2m −p + m r / u (p) = 0.101) .52)).100) Λ− (p) = 2m These are. (3. it is clear that we can define projection operators for such solutions as p+m / . because the direction of momentum changes for v(p) (see the derivation in (2. (3. (3.

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

3. 2 or. α α. 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). αβ (3.88 3 Properties of the Dirac equation 2 Pαβ (p) = r=1 r ur (p)uβ (p).105) Thus.106) Similarly. 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. we can identify Pαβ (p) = (Λ+ (p))αβ .104) This matrix has the property that acting on a positive energy spinor it gives back the same spinor. α β=1 r=1 = (3. Namely. if we define . 2. we see that the matrix P projects only on to the space of positive energy solutions and. (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). r=1 r ur (p)uβ (p) = α p+m / 2m . β = 1. 4. therefore.

completeness relation 89 2 Qαβ (p) = r=1 r r vα (p)v β (p). 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. r=1 r r vα (p)v β (p) = p−m / 2m . the matrix Q projects only on to the space of negative energy solutions with a phase (a negative sign). Hence we can identify Qαβ = −(Λ− (p))αβ . β β=1 r=1 4 4 s Qαβ (p)vβ (p) = β=1 2 β=1 r=1 r s vα (p)(−δrs ) = −vα (p). 4 s v β (p)Qβα (p) = β=1 2 β=1 r=1 2 s r r v β (p)vβ (p)v α (p) 2 r s r vα (p)v β (p)vβ (p) = (Q(p)v s (p))α = = r=1 4 (v s (p)Q(p))α = = r=1 r s (−δrs )v α = −v α . αβ (3.108) Namely.4 Projection operators. (3. 2 or.107) then.3. (3.103)) .109) The completeness relation for the solutions of the Dirac equation now follows from the observation that (see (3.

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

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

6.λ=1 βσ γ 0M †γ 0 σλ (Λ+ (p))λα (3. (3. 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 2.117) satisfies the commutation relation (see (2. spin commutes with the Hamiltonian (since in this frame p = 0) and such solutions can be labelled by the spin projection. in the relativistic case this is not useful since spin is not a constant of motion.116) does not commute either with the orbital angular momentum or with spin (rather.118) As a consequence.σ.92 1 2 4 3 Properties of the Dirac equation = Mαβ Λ+ (p′ α. unlike the case of non-relativistic systems where we specify a given energy state by the projection of spin along the z-axis (namely. H] = (3. ˜ 2 [Si . H = −iǫijk αj pk .β. that. it commutes with the total angular momentum). On the other hand. for a particle at rest.67)) 1 αi . by the eigenvalue of Sz ).115) = 1 Tr M Λ+ (p′ )γ 0 M † γ 0 Λ+ (p) . it can be easily checked that the plane wave solutions which we had derived earlier are not eigenstates of the spin operator. . we note that since momentum commutes with the Dirac Hamiltonian.5 Helicity As we have seen. the Dirac Hamiltonian H = α · p + βm. namely. 3. (3. Thus. In fact. we have already seen that the spin operator 1 ˜ α. however. 2 σi 0 0 σi S= αi = ˜ . Note.

119) the operator S · p does also (momentum and spin commute and. can only be ± 2 and we can label the positive and the negative energy solutions also as u(p.120) Therefore. therefore. h) with h = ± 1 (the two helicity 2 eigenvalues). Note that S·p |p | 2 h2 = = 1 ½. |p | (3. h′ ) = δhh′ = −v(p. Namely. h)u(p. (3.3. This is known as the helicity operator and we note that since the Hamiltonian commutes with helicity. H]pi = −iǫijk αj pk pi = 0. [Si pi . the eigenvalues of the helicity operator.123) Therefore. h). h)u(p.124) . The normalized operator h= S·p . for a Dirac parti1 cle. u(p. 4 4 (3. 4 (3. The normalization relations in this case will take the forms u(p. this operator is a constant of motion. h′ ).5 Helicity 93 [pi . h)v(p. H] = [Si . the order of these operators in the product is not relevant). (3.121) measures the longitudinal component of the spin of the particle or the projection of the spin along the direction of motion. h′ ). H] = [pi .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. v(p. the eigenstates of energy can also be labelled by the helicity eigenvalues. α · p + βm] = 0. (3. h′ ) = 0 = v(p. h)v(p.

111).125) 1 h=± 2 3. (3.127) where u1 (p) and u2 (p) are two component spinors. Let us represent the four component spinor (as before) as u1 (p) u2 (p) u(p) = .128) Explicitly. or. this leads to the two (2-component) coupled equations p0 − m u1 (p) − σ · p u2 (p) = 0. In terms of u1 (p) and u2 (p). which can also be written as p0 u1 (p) − σ · p u2 (p) = mu1 (p). γ 0 p0 − γ · p − m u(p) = 0.129) p0 u2 (p) − σ · p u1 (p) = −mu2 (p). (3.130) .6 Massless Dirac particle Let us consider the free Dirac equation for a massive spin (γ µ pµ − m) u(p) = 0. (3. h)] = ½. 1 2 particle. (3. (3.126) where we are not assuming any relation between p0 and p as yet.110) or (3. h)v(p. h)u(p.94 3 Properties of the Dirac equation Furthermore. h) − v(p. can now be written as [u(p. σ · p u1 (p) − p0 + m u2 (p) = 0. (3. the completeness relation. the Dirac equation takes the form (p0 − m)½ σ·p −σ · p u1 u2 −(p0 + m)½ = 0. (3.

(3. are not invariant under parity or space reflection and are known as the Weyl equations.134) These two equations. we obtain p0 − σ · p (u1 (p) + u2 (p)) = m (u1 (p) − u2 (p)) . 2 1 (u1 (p) + u2 (p)) . like the Dirac equation.133) reduce to two (2-component) spinor equations which are decoupled and have the simpler forms p0 uR (p) = σ · p uR (p). p0 uL (p) + σ · p uL (p) = muR (p).131) p0 + σ · p (u1 (p) − u2 (p)) = m (u1 (p) + u2 (p)) . however. We note that if we define two new (2-component) spinors as 1 (u1 (p) − u2 (p)) . . Let us note that in the limit m → 0.133) We note that it is the mass term which couples the two equations.6 Massless Dirac particle 95 Taking the sum and the difference of the two equations in (3. can be shown to be covariant under proper Lorentz transformations (as they should be. p0 uL (p) = −σ · p uL (p).132) then. the equations in (3. The corresponding two component spinors uL and uR are also known as Weyl spinors. (3. since vanishing of the mass which is a Lorentz scalar should not change the behavior of the equation under proper Lorentz transformations). the two 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). These equations. (3.3.130).

(p0 )2 uR (p) = σ · p p0 uR (p) (p0 )2 − p 2 uR (p) = 0.135) or. we must have p0 = ±|p |. p0 uL (p) = −σ · p uL (p). we note that p0 uR (p) = σ · p uR (p). σ·p uR (p) = uR (p). for a nontrivial solution of these equations to exist. Similarly. It is clear. = (σ · p) (σ · p) uR (p) = p 2 uR (p). we can show that uL (p) also satisfies (p0 )2 − p 2 uL (p) = 0.138) For p0 = |p |.96 Let us note that 3 Properties of the Dirac equation or. Recalling that 1 σ denotes the spin 2 .140) In other words. the two different Weyl equations really describe particles with opposite helicity. namely. (3. (3. for the positive energy solutions. we must have (p0 )2 − p 2 = 0. |p | or.139) or. that for such solutions. p0 uR (p) = σ · p uR (p). therefore. (3. σ·p uL (p) = −uL (p).137) which is the Einstein relation for a massless particle.136) Thus. (3. |p | (3. while (3.

Correspondingly. uR (p) describes a particle with helicity + 1 or a particle with spin parallel to its direction of 2 motion. we note that uL (p) describes a particle with helicity − 1 or a particle with spin anti-parallel to its 2 direction of motion. experimentally. This is shown in Fig. namely. then we conclude that for such a particle.6 Massless Dirac particle 97 operator for a two component spinor. therefore. 3.141) is massless (present experiments suggest they are almost massless) and. As we know. On the other hand. If we think of spin as arising from a circular motion.1: Right-handed and left-handed particles with spins parallel and anti-parallel to the direction of motion. We also know.) p p Left-handed Right-handed Figure 3. (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. .1 and we note here that this nomenclature is opposite of what is commonly used in optics. the circular motion would correspond to that of a left-handed screw. the electron neutrino emitted in a beta decay A X Z+1 → A Y Z + e+ + νe . that νe is left-handed. such a particle is called a left-handed particle (which is the reason for the subscript L). 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. (3.3.

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

then γ5 u(p) (or γ5 v(p)) is also a solution. the solutions of the massless Dirac equation (m = 0) pu(p) = 0 = pv(p). This can also be seen from the fact that the Hamiltonian for a massless Dirac fermion (see (1.149) commutes with γ5 (in fact. Since 2 γ5 = ½.150) it follows that the eigenvalues of γ5 are ±1 and spinors with the eigenvalue +1.147). / / can be written as (see also (3. we note that if u(p) (or v(p)) is a solution. ˜ |p| v(p) = σ·p v (p) ˜ v (p) ˜ = σ ·p v (p) ˜ |p| v (p) ˜ . namely. (3.54). the solutions of the massless Dirac equation can be classified according to the eigenvalues of γ5 also known as the chirality or the handedness. for example.7 Chirality With the normalization for massless spinors discussed in (2. γ5 uR (p) = uR (p).53) and (2. γ5 vR (p) = vR (p). (3. (2.61)).148) From the structure of the massless Dirac equation (3.60) and ρ commutes with α. in the Pauli-Dirac representation γ5 = ρ defined in (2. Therefore.100)) H = α · p. (3.7 Chirality 99 3.3.147) u(p) = E 2 E 2 = |p| 2 |p| 2 u(p) ˜ σ·p u(p) . (3.92)) u(p) ˜ σ·p u(p) ˜ E E (3.151) . see.

γ5 uL (p) = −uL (p). Given a general spinor. namely. We note that if the fermion is massive (m = 0). (3.152) are called left-handed (negative chirality) spinors.153) where we have defined . 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) . vR (p) = PR v(p) = = |p| 2 1 2 1 2 1 (½ + γ5 )v(p) 2 ½ + σ ·p v(p) ˜ |p| .100) would no longer commute with γ5 and in this case chirality would not be a good quantum number to label the states with. γ5 vL (p) = −vL (p).100 3 Properties of the Dirac equation are known as right-handed (positive chirality) spinors while those with the eigenvalue −1. then the Dirac Hamiltonian (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) ˜ . (3. 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| .

hχ(±) (p) = σ·p 1 2|p| 2 ½± ½± 1 1 = ± × 2 2 σ·p χ ˜ |p| 1 σ·p χ = ± χ(±) (p). PL = . 2 −½ ½ ½ ½ (3. (3.158) . these projection operators have the explicit forms (see (2.121) (for the two component spinors S = 1 σ). From the definition of the helicity operator in (3.7 Chirality 101 1 1 PR = (½ + γ5 ).156) We note from (3. PL 2 = PL .154) 2 2 We note that by definition these projection operators satisfy PR 2 = PR . the four component spinors can be effectively described by two component spinors. ˜ |p| 2 (3. we will continue our discussion in the Pauli-Dirac representation of the Dirac matrices which we have used throughout. This is connected with our earlier observation (see section 3. the Dirac equation reduces to two decoupled two component Weyl equations (recall that it is the mass term which generally couples these two spinors).6) that in the massless limit.91)) 1 2 PR = ½ ½ 1 ½ −½ . (3. we note that 2 spinors of the form χ(±) (p) = 1 σ·p 1± χ. ˜ 2 |p| (3.157) correspond to states with definite helicity. namely.3. PL = (½ − γ5 ). (In the Pauli-Dirac representation. PR + PL = 1.155) PR PL = 0 = PL PR . which implies that any four component spinor can be uniquely decomposed into a right-handed and a left-handed component.153) that in the massless Dirac theory.119). However. The reducibility of the spinors is best seen in the Weyl representation for the Dirac matrices discussed in (2.

P (+) + P (−) = 1. They can be easily generalized to a reducible representation of operators acting on the four component spinors and have the form (see (2. (P (−) )2 = P (−) . Explicitly. σ µ defined in (2. (3.159) We note here that the operators P (±) = 1 2 ½± σ·p 1 ˆ = (½ ± σ · p).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) = .163) .117)) P (±) 0 1 2( P4×4 (±) = 0 P (±) 0 1 2( = ˆ ½ ± σ · p) 0 ˆ ½ ± σ · p) .(3.70) or (3.162) and.161) with σ µ . It is straightforward ˜ ˆ p to check that they satisfy the relations (P (+) )2 = P (+) . ˆ 2 (3. define projection operators into the space of positive and negative helicity two component spinors. P (+) P (−) = 0 = P (−) P (+) . vL (p) = −v (−) (p) v (−) (p) .102 3 Properties of the Dirac equation so that the right-handed (four component) spinors in (3. ˜ ˆ 2 (3.153) and (3. |p| 2 P (−) = 1 µ σ pµ .120) and pµ = (1. therefore. −ˆ ). we see from (3.153) are described by two component spinors with positive helicity while the lefthanded (four component) spinors are described in terms of two component spinors of negative helicity. uL (p) = .160) can also be written in a covariant notation as P (+) = 1 µ σ pµ . (3. vR (p) = .

Furthermore. For completeness as well as for later use. (3.165) 2 2 Each of these spinors is one dimensional and together they span the two dimensional spinor space.167). 1 (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. For example. We can choose u and v to be ˜ ˜ normalized so that we have u(+) (p) = u† u = 1 = v † v . P4×4 = 0.168) . from (3. v (+) (p) = .3.163) that PL.163) we see that the right-handed spinors with chirality +1 are characterized by helicity +1 while the left-handed spinors with chirality −1 have helicity −1.167) such that when p1 = p2 = 0. 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 .159) as well as (3.164) which is the reason the spinors can be simultaneous eigenstates of chirality and helicity (when mass vanishes). we can choose u= ˜ 1 . 0 v= ˜ 0 .156) and (3. with the choice of the basis in (3.7 Chirality 103 and it is straightforward to check from (3. ˜˜ ˜˜ (3. (±) (3. let us derive some properties of these spinors. the helicity spinors simply reduce to eigenstates of σ3 . v(+) (p) = ½ + σ · p v.R . In fact. We note from (3. (3. we can also define normalized spinors u(+) and v (+) .166) For example. ˜ ˜ ˜ ˜ uu† + v v † = ½.

the positive helicity spinors satisfy u(+)† (p)u(+) (p) = v (+)† (p)v (+) (p) = 1. u(+)† (p)v (+) (−p) = 0 = v (+)† (−p)u(+) (p). (3. In general. Given the form of the right-handed spinors in (3.170) The completeness relation in (3.170) can be simplified by noting the following identity.159). as well as (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)u(+)† (p) = v (+) (p)v (+)† (p) = 1 2 ˆ ½ + σ · p .168). we do not need to use any particular representation for our discussions.169) it now follows in a straightforward manner that † u† (p)uR (p) = |p| = vR (p)vR (p). We note that with p0 = |p|. R † u† (p)vR (−p) = vR (−p)uR (p) = 0. (3.169) which can be checked from the explicit forms of the spinors in (3.171) = P (+) P (+) P (+) P (+) . R † uR (p)u† (p) = vR (p)vR (p) = R |p| 2 P (+) P (+) P (+) P (+) . we can write 1 0 pγ (½ + γ5 ) / 4 = 1 4 |p| 4 |p| 2 σ·p |p| −σ · p −|p| ˆ σ·p ½ 0 0 −½ ½ ½ ½ ½ = ½ ˆ σ·p ½ ½ ½ ½ ½ . (3.104 3 Properties of the Dirac equation However.

/ 4 (3.173) 3.170) as † uR (p)u† (p) = vR (p)vR (p) = R 1 0 pγ (½ + γ5 ).3. (3. 2m E + m E+m (3.4.49)) u ˜ σ·p ˜ E+mu uL (p) uS (p) u+ (p) = E+m 2m = . (3.175) In (3.176) .174) while the negative energy solutions have the form σ·p ˜ −E + m v v ˜ vS (p) vL (p) u− (p) = E+m 2m = . / 4 (3.8 Non-relativistic limit of the Dirac equation Let us recall that the positive energy solutions of the Dirac equation have the form (see (2. L † u† (p)vL (−p) = 0 = vL (−p)uL (p).8 Non-relativistic limit of the Dirac equation 105 so that we can write the completeness relation in (3. 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) we have defined uL (p) = uS (p) = E+m u. L † uL (p)u† (p) = vL (p)vL (p) = L 1 0 pγ (½ − γ5 ).172) which can also be derived using the methods in section 3. ˜ 2m σ·p E+m σ·p u= ˜ uL (p).

179) We note that the second equation in (3.175) we have denoted vL (p) = vS (p) = − E+m v. σ · p uL (p) = (E + m)uS (p). uL (p) and uS (p) are known as the large and the small components of the positive energy Dirac solution. 2m E + m E+m (3. Let us next look at the positive energy solutions in (3. which satisfy the equation Hu+ (p) = Eu+ (p). In the non-relativistic limit. the component uS (p) is much smaller than (of the order of v ) uL (p) and c correspondingly. m½ σ · p −m½ σ·p uL (p) uS (p) =E uL (p) uS (p) .106 3 Properties of the Dirac equation and we emphasize that the subscript “L” here does not stand for the left-handed particles introduced in the last section.180) . Similarly.177) It is clear that in the non-relativistic limit. vL (p) and vS (p) are also known as the large and the small components of the negative energy solution. (3. Similarly. or.174). (3. or. ˜ 2m σ·p E +m σ·p v=− ˜ vL (p). (α · p + βm) u+ (p) = Eu+ (p). when |p | ≪ m. we expect the large components to give the dominant contribution to the wave function. in (3.179) gives the relation σ·p uL (p). E+m uS (p) = (3.178) This would lead to the two (2-component) equations σ · p uS (p) = (E − m)uL (p).

if we identify the non-relativistic energy (without the rest mass term) as ENR = E − m. the first equation in (3. Since the coordinate representation of pµ is given by (see (1. the Dirac equation in this case reduces to the Schr¨dinger o equation for a two component spinor which we are familiar with.33) and remember that we are choosing = 1) . This preserves the gauge invariance associated with the Maxwell’s equations and corresponds to defining pµ → pµ − eAµ . with the substitution of this.9 Electron in an external magnetic field The coupling of a charged particle to an external electromagnetic field can be achieved through what is conventionally known as the minimal coupling.9 Electron in an external magnetic field 107 while.184) where e denotes the charge of the particle and Aµ represents the four vector potential of the associated electromagnetic field. E+m (3. E ≈ m (recall that we have set c = 1).182) (3. equation (3.181) p2 uL (p) ≃ (E − m)uL (p). (3. therefore.183) Namely. σ·p uL (p) = (E − m)uL (p).179) takes the form (σ · p) or. what we know for a free non-relativistic electron (spin 1 particle). 2m where we have used the fact that for a non-relativistic system. and.3. This is. |p| ≪ m. 2 3. of course. 2m (3.181) has the form p2 uL (p) = ENR uL (p). Furthermore. then.

(3. In this case. (3. we can write the two (2-component) equations as σ · (p − eA ) uS (p) = (E − m)uL (p). E+m 2m (3. we obtain .108 3 Properties of the Dirac equation pµ → i∂µ . in this case.187) where we are assuming that A = A(x). takes the form (α · (p − eA ) + βm) u(p) = Eu(p).189) uS (p) = (3.188) Explicitly. σ · (p − eA ) uL (p) = (E + m)uS (p). we have used |p| ≪ m in the non-relativistic limit.186) Let us next consider an electron interacting with a time independent external magnetic field. (3. B = (∇ × A ) . Substituting this back into the first equation in (3. (3. we have A0 = 0 = φ. the second equation in (3. The Dirac equation for the positive energy electrons.190) where in the last relation.185) the minimal coupling prescription also corresponds to defining (in the coordinate representation) ∂µ → ∂µ + ieAµ .189) leads to σ · (p − eA) σ · (p − eA) uL (p) ≃ uL (p). or. m½ σ · (p − eA ) σ · (p − eA ) −m½ uL (p) uS (p) =E uL (p) uS (p) .189). In this case.

we are going to use purely three dimensional notation for simplicity) (p × A + A × p )i = ǫijk (pj Ak + Aj pk ) = ǫijk (pj Ak − Ak pj ) = ǫijk [pj . (3. in the non-relativistic limit. (3.3.193) = (p − eA )2 − ieσ · (−iB) (3. 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. Ak ] = −i (∇ × A )i = −i(B)i .9 Electron in an external magnetic field 109 (σ · (p − eA)) (σ · (p − eA)) uL (p) ≃ (E − m)uL (p). We can use this in (3. 2m or.192) Note that (here.192) to write (σ · (p − eA)) (σ · (p − eA)) = (p − eA )2 − eσ · B. equation (3. 1 e (p − eA)2 − σ · B uL (p) ≃ ENR uL (p). 2m (3.191) using the following identity for the Pauli matrices (σ · (p − eA)) (σ · (p − eA)) = (p − eA) · (p − eA ) + iσ · ((p − eA) × (p − eA)) = (p − eA)2 − ieσ · (p × A + A × p) . Ak ] = −iǫijk [∇j .191) takes the form 1 (p − eA)2 − eσ · B uL (p) = (E − m)uL (p).195) 2m 2m where we have identified (as before) . (3.194) Consequently.

(3.196) We recognize (3. Particles with a nontrivial structure (that is particles which are not point like and have extended structures).197) we obtain g = 2. however.197) Of course. 2m µ=g (3. for the proton and the neutron. Quantum mechanical corrections (higher order corrections) in an interacting theory such as quantum electrodynamics. 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. however. comparing with (3. to a magnetic dipole interaction (recall that in the non-relativistic theory.110 3 Properties of the Dirac equation ENR = E − m. this shows that a point Dirac particle has a magnetic moment corresponding to a gyro-magnetic ratio g = 2. a minimally coupled Dirac particle automatically leads.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. can have g-factors quite different from 2.198) Let us recall that the magnetic moment of a particle is defined to be (c = 1) e S.199) 1 Since S = 2 σ for a two component electron. (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. for example. 2m µ= (3. In this case. we know that the magnetic moments are given by . Namely. Thus. in the non-relativistic limit. one says that there is an anomalous contribution to the magnetic moment.

79 µnm . This is commonly known as the Pauli coupling or the Pauli interaction. This puzzling feature can be understood in a simple manner as follows. the relevant expansion parameter is |p| and m the method works quite well in the lowest order of expansion. 2mP µnm = (3.201) with mP denoting the mass of the proton. if we were to calculate the electric dipole interaction of an electron in a background electromagnetic field using the method described in the earlier sections.10 Foldy-Wouthuysen transformation In the last two sections. However.3. at higher orders.91 µnm . (3. (3. 2m (3. the electric dipole 1 moment becomes imaginary at order m2 (namely. = 2.202) denotes the electromagnetic field strength tensor and κ represents the anomalous magnetic moment of the particle. the Hamiltonian becomes non-Hermitian).10 Foldy-Wouthuysen transformation 111 µN µP = −1.203) eκ µν σ Fµν . For example.200) where the nuclear magneton is defined to be |e| . this method runs into difficulty. 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 . In the non-relativistic limit. 3. we have described how the non-relativistic limit of a Dirac theory can be taken in a simple manner. The process of eliminating the “small” .

180)) σ·p . Since the lack of unitarity in (3. the main idea in the works of Foldy-Wouthuysen as well as Tani is to ensure that the relevant transformation used in going to the non-relativistic limit is manifestly unitary. (3. for example. In this case. has the form (see (3. for example.207).100) as well as (1. (3.204) has the form T = ½ −A 0 ½ .207) Let us next look for a unitary transformation that will diagonalize the Hamiltonian in (3. such a transformation would . in the case of the free Dirac equation.205) for the positive energy spinors.101)) H = α · p + βm = γ 0 (γ · p + m) . The matrix T that takes us to the two component “large” spinors in (3.204) where the matrix A.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). let us look at the free Dirac theory where we know that the Hamiltonian has the form (see (1. E+m A= (3.112 3 Properties of the Dirac equation components from the Dirac equation described in the earlier sections can be understood mathematically as uL (p) uS (p) uL (p) AuL (p) uL (p) 0 u(p) = = −→ T .206) is the source of the problem in taking the non-relativistic limit consistently. 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.206) It is clear from the form of the matrix in (3. Thus. (3.

209) From the properties of the gamma matrices in (1.10 Foldy-Wouthuysen transformation 113 also transform the spinor into two 2-component spinors that will be decoupled and we do not have to eliminate one in favor of the other (namely.212) .210) and using this we can simplify and write U (θ) = ∞ n=0 1 (2n)! |p|θ 2m γ · pθ 2m + 2n + 1 (2n + 1)! .208) where the real scalar parameter of the transformation is a function of p and m. (γ · p)2 = −p2 = −|p|2 .83) or (1. Let us consider a transformation of the type U (θ) = e 2m γ ·p θ . we note that (γ)† = −γ. (3. θ = θ(|p|. (3. 1 (3. avoid the problem with “large” and “small” spinors).91). (3. γ · pθ 2m 2n+1 = cos γ ·p sin |p| |p|θ 2m (3.211) It follows now that U † (θ) = cos which leads to |p|θ 2m − γ ·p sin |p| |p|θ 2m . m).3.

208). the free Dirac Hamiltonian (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 . Under the unitary transformation (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 (3.207) 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 .208) is indeed unitary.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. (3. Namely.

if we want the transformation to diagonalize the Hamiltonian.217) that the Hamiltonian is now diagonalized in the positive and the negative energy spaces. sin = p2 + m2 .219) . leads to the diagonalized Hamiltonian m2 p2 + m2 p2 p2 + m2 H ′ = γ0 + = γ0 p2 + m2 .10 Foldy-Wouthuysen transformation 115 So far. (3.217) We see from (3. m |p| m . or. the two components of the transformed spinor u′ (p) 1 u′ (p) 2 u′ (p) = U (θ)u(p) = .214) that we can choose the parameter of transformation to satisfy |p|θ m = |p|θ m |p| cos or. has been arbitrary. (3.218) would be decoupled in the energy eigenvalue equation and we can without any difficulty restrict ourselves to the positive energy sector where the energy eigenvalue equation takes the form p2 + m2 u′ (p) = Eu′ (p). 1 1 (3. As a result. (3. our discussion has been quite general and the parameter of the transformation. = 0. from (3.216) which. However. |p|θ m m arctan |p| (3.215) In this case.3. θ. it is clear from (3.214). we have |p|θ m m p2 + m2 |p|θ m |p| cos = . tan θ= − m sin |p| .

214) that if we choose the parameter of transformation to satisfy |p|θ m |p|θ m |p|θ m m cos or. = 0. In this case.215). We also note that with the parameter θ determined in (3. in this case. Let us note from (3. (3. |p| m |p| . the transformation is known as the Cini-Touschek transformation and is obtained as follows.116 3 Properties of the Dirac equation For |p| ≪ m. or.211) is also quite useful. This m analysis can be generalized even in the presence of interactions and the higher order terms in the interaction Hamiltonian are all well behaved without any problem of non-hermiticity. θ=− m arctan |p| (3. m (3.183) to 1 the lowest order and it can be expanded to any order in m without any problem.220) UFW (θ) = cos + which has a natural non-relativistic expansion in powers of |p| . There is a second limit of the Dirac equation. the ultrarelativistic limit |p| ≫ m. namely. the unitary transformation in (3. for which the generalized Foldy-Wouthuysen transformation (3.221) this would lead to |p|θ m |p|θ m |p| cos sin = = − p2 + m2 m . the transformed Hamiltonian (3.222) p2 + m2 As a result.214) will have the form . tan + |p| sin =− m . . this leads to the non-relativistic equation in (3.211) takes the form 1 arctan 2 |p| m γ·p sin |p| 1 arctan 2 |p| .

223) p2 + m2 γ·p= |p| p2 + m2 α · p. where operators carry time dependence and states are time independent.220) as transforming away the α · p term in the Hamiltonian (3. the Heisenberg equations of motion take the forms ( = 1) .224) m which clearly has a natural expansion in powers of |p| (ultrarelativistic expansion).225) In the Heisenberg picture. 3.3.100) H = α · p + βm. Therefore. |p| m which has a natural expansion in powers of |p| . In fact. the unitary transformation (3.11 Zitterbewegung The presence of negative energy solutions for the Dirac equation leads to various interesting consequences.211) has the form UCT (θ) = cos 1 arctan 2 m |p| − γ ·p sin |p| 1 arctan 2 m . let us consider the free Dirac Hamiltonian (1.207). in this case. |p| (3. (3.11 Zitterbewegung 117 H ′ = γ0 = γ0 p2 p2 + m2 + m2 p2 + m2 γ·p |p| (3. For example. we can think of the Foldy-Wouthuysen transformation (3.207) while the Cini-Touschek transformation rotates away the mass term βm from the Hamiltonian (3.

118 3 Properties of the Dirac equation 1 i 1 [x . Secondly. for example. H] = [α.226) Here a dot denotes differentiation with respect to time.101)). on the other hand. The first equation. α p ] = αj [xi . H] = 0. the velocity is not.228) (3. α · p + βm] = 0. (3.227) where we have denoted the operator in the Schr¨dinger picture by o α(0) = α. Furthermore. = i ˙ or. identifies α(t) with the velocity operator. it follows that (3. p = 0. even though the momentum of a free particle is a constant of motion. x = α. it follows that the .226) shows that the momentum is a constant of motion as it should be for a free particle. H] = [pi . xi = ˙ 1 1 i [p . H] = [xi .101) we conclude that [α. pj ] = i i 1 j ij α (iδ ) = αi . (3. using (1. αj pj + βm] i i 1 1 i j j [x . The second equation in (3. αj pj + βm] = 0. α(t) = eiHt α e−iHt .229) [α(t). As a result.230) In other words. since the eigenvalues of α are ±1 (see. Let us recall that. pi = ˙ (3. (1. i i ˙ or. by definition.

These peculiarities of the relativistic theory can be understood as follows. β = 2δij ½.235) + + . (3. = λ eiHt |ψ . then. H = −2iα(t)H + 2ip.3. (3.231) where λ denotes the eigenvalue of the velocity operator α. This is easily understood from the fact that the eigenvalues of an operator do not change under a unitary transformation.233) Equation (3. since the electron is a massive particle. H] i + ˙ α(t) = = −i 2α(t)H − α(t). We note from Heisenberg’s equations of motion that the time derivative of the velocity operator is given by 1 [α(t).232) shows that the eigenvalues of α(t) are the same as those of α (only the eigenfunctions are transformed) and. we note that if α|ψ = λ|ψ . This would seem to imply that the velocity of an electron is equal to the speed of light which is unacceptable even classically. (3. therefore. (3. α(t)|ψ ′ = λ |ψ ′ .234) Here we have used the relations (see (1. = 0. αj αi .232) or. are ±1. it follows that eiHt αe−iHt eiHt |ψ = λ eiHt |ψ .11 Zitterbewegung 119 eigenvalues of α(t) are ±1 as well. More explicitly. (3. α(t) eiHt |ψ where we have identified |ψ ′ = eiHt |ψ . or.102)) αi .

which is unexpected. (3.240) which is the first term in (3. α(t) = pH −1 + (α(0) − pH −1 )e−2iHt . from (3.234). we obtain ¨ ˙ α(t) = −2iα(t)H. It represents an additional component to the .234) with respect to time. (3.236) (3. p −2iHt p e .241) p = mγv. Substituting this back into (3. so that mγv p = = v.239).239) or.238) Furthermore. E mγ (3. For example.234) we have ˙ α(0) = −2iα(0)H + 2ip = −2i(α(0) − pH −1 )H. we obtain ˙ α(t) = −2i(α(0) − p H −1 )He−2iHt .120 3 Properties of the Dirac equation as well as the fact that momentum commutes with the Hamiltonian.236). + α(0) − α(t) = H H The first term in (3. ˙ ˙ or. (3. differentiating (3. On the other hand. using this relation in (3. It is the second term. Let us note next that both p and H are constants of motion. (c = 1) E = mγ. however. Therefore. α(t) = α(0)e−2iHt . in an p eigenstate of momentum it would have the form E which is the true relativistic expression for velocity. We note that. we finally determine −2i(α(0) − p H −1 )He−2iHt = −2iα(t)H + 2ip. or.239) is quite expected. for a relativistic particle.237) (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 .) and gives a time dependence to α(t). The first two terms in (3. Let us also note from (3.243) can be shown in the Schr¨dinger picture to arise from the presence of o negative energy solutions. Its occurrence is quite surprising. This quivering motion of the electron was first studied by Schr¨dinger o and is known as Zitterbewegung (“jittery motion”).243) are again what we will expect classically for uniform motion. for example. we obtain i p t+ H 2 p H −1 e−2iHt .3.226) that since ˙ x(t) = α(t) = p −2iHt p e . In fact.243) where a is a constant. the third term represents an additional contribution to the electron trajectory which is oscillatory with a very high frequency. The unconventional operator relations in (3. since there is no potential whatsoever in the problem.242) integrating this over time. However.11 Zitterbewegung 121 velocity which is oscillating at a very high frequency (for an electron at rest. σ·p ˜ E+mu u(p) = we have (3. it is easy to check that for a positive energy electron state E+m 2m u ˜ . the energy is ≈ .239) and (3.5MeV corresponding to a frequency of the order of 1021 /sec. H x(t) = a + α(0) − (3. + α(0) − H H (3.

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

Progress of Theoretical Physics 12. Physical Review 78. Progress of Theoretical Physics 6. 603 (1954). S. Tani. Okubo. 12.12 References 123 4. 5. Cini and B. 422 (1958). 267 (1951). M. Foldy and S. 7. . ibid. Wouthuysen.3. S. A. 6. L. 29 (1950). Touschek. L. 102 (1954). Nuovo Cimento 7.

This page intentionally left blank .

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

) ω i j = ǫi jk αk .7) . of course. ǫµν .34) and (1.6) In other words. etc.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 . ǫijk denotes the three dimensional Levi-Civita tensor with ǫ123 = 1.) If we now identify (In the last chapter we had denoted the infinitesimal transformation matrices by ǫi j .126 4 Representations of Lorentz and Poincar´ groups e notation from (1. the form of the rotation that we had discussed in the last chapter. then. It follows now that i R(ω)xi = −ω kj xj ∂k xi = −ω kj xj δk = −ω ij xj = −ω i j xj .4) This is. Let us next define an infinitesimal vector operator (also known as the tangent vector field operator) for rotations (an operator in the coordinate basis) of the form 1 1 ij ω Mij = ω ij (xi ∂j − xj ∂i ) 2 2 (4. (4. (4. ω µν in order to avoid confusion.3) (4.35) here for clarity. ǫi jk = η iℓ ǫℓjk . (4. we note that ω ij = −ω ji . but here we denote them by ω ij .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.

R(ω) = ω ij xi ∂j .8). R(ω) = 1 ij ω Mij . Such an algebra is called a non-Abelian (non-commutative) algebra. Mij . where we have identified ω ij = ω ik ω kj − ω ik ωkj = −ω ji = [ω. xk ∂ℓ ] = ω ij ω kℓ (xi [∂j . we can calculate them directly as . 2 (4. from the algebra of the vector operators in (4. namely. xk ] ∂ℓ + xk [xi . two rotations do not commute. Alternatively. R(ω) in (4.9) Namely.5) generates infinitesimal rotations and the operators. Thus. ω]ij . are known as the generators of infinitesimal rotations. the vector operator.8) = R(ω). they give back a rotation. ω kℓ xk ∂ℓ = ω ij ω kℓ [xi ∂j . rather. Using the form of R(ω) in (4. Mij . we note that R(ω). The Lie algebra of the group of rotations can be obtained from the algebra of the vector operators themselves. ∂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 (4.5).1 Symmetry algebras 127 Namely. (4.4.10) we can obtain the algebra satisfied by the generators of infinitesimal rotations.

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

13) 1 + ǫis q ǫjrs Mqr + 4 1 1 = − ǫriq ǫjsr Mqs − 4 4 1 − ǫri q ǫjsr Mqs − 4 or. 2 (4.18) .4. the Lie algebra of SO(3) or SU (2) (or the familiar algebra of angular momentum operators) up to a rescaling. can be generated by the infinitesimal vector operator R(ǫ) = −ǫµ Pµ = −ǫµ ∂µ . = −ǫri q ǫjsr Mqs . (4. namely.15) The algebra of the generators in (4. then.1. (4.13) is. [Ji . 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). In the same spirit. 4. Jj ] = ǫij r Jr .14) which. ǫri q ǫjsr + ǫrsq ǫij r + ǫrjq ǫsi r = 0.16) (4. let us note that a constant in- finitesimal space-time translation of the form δxµ = ǫµ .2 Translation.11) or (4.1 Symmetry algebras 129 1 p rs ǫ ǫ Mps 4 i r j ǫi q r ǫjsr Mqs 1 q sr ǫ ǫ Mqs 4 i r j (4. leads to (see (4.12)) 1 q ǫri ǫjsr Mqs = − ǫijr ǫrsq Mqs = −ǫijr Jr .17) (4. so that µ R(ǫ)xµ = −ǫν ∂ν xµ = −ǫν δν = −ǫµ . of course.

23) As in the case of rotations.21) Namely.22) where. (4. µ.18). ǫ ν ∂ν ] = ǫµ ǫ ν [∂µ . As we have seen in the last chapter.19) The Lie algebra associated with translations is then obtained from R(ǫ). 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. ∂ν ] = 0.1.24) then. translations form an Abelian (commuting) group while the three dimensional rotations form a non-Abelian group.130 and we can write 4 Representations of Lorentz and Poincar´ groups e δxµ = −R(ǫ)xµ . we obtain . the infinitesimal. R(ǫ) = [ǫµ ∂µ . (4.20) In other words. Pν ] = [∂µ . 4. (4. constant parameters of transformation satisfy ω µν = −ω νµ . (4. two translations commute and the corresponding relation for the generators is [Pµ .3 Lorentz transformation. 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. (4. R(ω) = (4. 3. 1. ∂ν ] = 0. 2.

29) . ω]µν . (4.28) = = ω µν xµ ∂ν = R(ω). where. R(ω) = = ω µν ω λρ [xµ ∂ν . we can think of R(ω) as the vector operator generating infinitesimal proper Lorentz transformations and the operators. we can determine the group properties of the Lorentz transformations from the algebra of the vector operators generating the transformations.1 Symmetry algebras 131 µ R(ω)xµ = −ω λν xν ∂λ xµ = −ω λν xν δλ = −ω µν xν = −ω µν xν = −δxµ . xλ ∂ρ ] = ω µν ω λρ (ηνλ xµ ∂ρ − ηµρ xλ ∂ν ) = ω µλ ω λρ xµ ∂ρ − ω µν ω λµ xλ ∂ν ω µλ ω λν − ωµλ ω λν xµ ∂ν (4. We also note that we can identify the infinitesimal generators of spatial rotations with (see (4. and the generators of infinitesimal boosts with (4.26) M0i = −Mi0 = x0 ∂i − xi ∂0 = Ki .4.27) As before. we have ω µν = −ω νµ = ω µλ ω λν − ω µ ω λν = [ω.25) Therefore. ω λρ Mλρ 2 2 R(ω). as in the case of rotations. 1 µν 1 ω Mµν . Mµν = −Mνµ .12)) Mij = −Mji = xi ∂j − xj ∂i = ǫij k Jk . (4. as the generators of the infinitesimal transformations. λ (4. Thus.

30) (up to a scaling). xλ ∂ρ − xρ ∂λ ] = [xµ ∂ν .69) in connection with the discussion of covariance of the Dirac equation. the Lie algebra of the generators is isomorphic to that of the group SO(4). In fact. in four dimensions and. we note that the number of generators for SO(4) which is (for SO(n).31) coincides exactly with the six generators we have (namely. xλ ∂ρ ] + [xν ∂µ . 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. gives the Lie algebra associated with Lorentz transformations. (Namely. 2 (4. 1). xλ ∂ρ ] − [xµ ∂ν . As we have seen these transformations correspond to rotations. This is what we had seen explicitly in (3. three rotations and three boosts). the matrices σµν provide a representation for the generators of the Lorentz group. therefore. The algebra of the generators can also be calculated directly and has the form [Mµν .) We end this section by pointing out that the algebra in (2.132 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. . (4. in this case. However. Lorentz transformations (boosts) are non-compact unlike rotations in Euclidean space. This implies that. xρ ∂λ ] − [xν ∂µ . xρ ∂λ ] = (ηνλ xµ ∂ρ − ηµρ xλ ∂ν ) − (ηνρ xµ ∂λ − ηµλ xρ ∂ν ) − (ηµλ xν ∂ρ − ηνρ xλ ∂µ ) + (ηµρ xν ∂λ − ηνλ xρ ∂µ ) = −ηµλ (xν ∂ρ − xρ ∂ν ) − ηνρ (xµ ∂λ − xλ ∂µ ) +ηµρ (xν ∂λ − xλ ∂ν ) + ηνλ (xµ ∂ρ − xρ ∂µ ) = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ . up to a scaling. it is n(n−1) ) 2 1 × 4 × (4 − 1) = 2 × 3 = 6.30) This.109) coincides with (4. therefore. Mλρ ] = [xµ ∂ν − xν ∂µ .

32) are known as the (infinitesimal) Poincar´ transformations or the inhomoe geneous Lorentz transformations.32) where ǫµ . ω) . ω). Namely. xρ ∂λ ] + ω µν ǫ λ [xν ∂µ . ∂λ ] + ω µν ω λρ [xν ∂µ .35) . ω) = ǫµ ∂µ + ω µν xν ∂µ . The transformations in (4.1 Symmetry algebras 133 4. (4.33) then.4 Poincar´ transformation. ω µν denote respectively the parameters of infinitesimal translation and Lorentz transformation. acting on the coordinates.1. if we define an infinitesimal vector operator as R(ǫ. ω)xµ = − ǫν ∂ν + ω λν xν ∂λ xµ = −(ǫν δν + ω λν xν δλ ) = −(ǫµ + ω µν xν ) = −δxµ .34) The algebra of the vector operators for the Poincar´ transformae tions can also be easily calculated as R(ǫ. we also consider infinitesimal translations. ǫ λ ∂λ + ω λρ xρ ∂λ = ǫµ ω λρ [∂µ . in this case. it generates infinitesimal Poincar´ e transformations. in addition to infinitesimal Lorentz e transformations. If. (4. 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. (4. (4.4. Clearly. µ µ R(ǫ. R(ǫ. 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 ) .

Therefore. ω µν = ω µλ ω λν − ω µλ ωλ ν = [ω. (This is seen by recalling that 2 ω µν Mµν generates infinitesimal Lorentz transformations. Pµ behaves 1 like a covariant four vector. Mνλ ] = ηµν Pλ − ηµλ Pν . 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. for example.38) [Mµν . (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.) Thus.37) which simply shows that under a Lorentz transformation. see. Pν ] (see (4.21)). e We already know the commutation relations [Mµν . Mνλ ] = [∂µ .36) We can also calculate the algebra of the generators of Poincar´ group. Note that [Pµ . However.30). (4.22) and (3. (4. Pν ] = 0.30) and (4. (4.134 4 Representations of Lorentz and Poincar´ groups e where we have identified ǫµ = (ω µν ǫ ν − ω µν ǫν ) . (4. as well as the homogeneous Lorentz group.21). since the generae tors of translations do not commute with the generators of Lorentz . ω]µν . (4.57). We note that the algebra of translations defines an Abelian subalgebra of the Poincar´ algebra (4.38). we conclude that the Lie algebra associated with the Poincar´ e transformations (inhomogeneous Lorentz group) is given by [Pµ . Mλρ ] = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ . combining with our earlier results on the algebra of the translation group. Mλρ ] as well as [Pµ . the only relation that needs to be calculated is the commutator between the generators of translation and Lorentz transformations. [Pµ .

the Lorentz algebra takes the form [Ki .12). (4.26) and (4. Kj ] . Kj ] = [M0i . (The general convention is to denote groups by capital letters while the algebras are represented by lower case letters. Poincar´ algebra cannot be written as a direct sum e of those for translations and Lorentz transformations. Namely.42) .41) Written out in terms of these generators. (4. (4. (4. [Ji . Mλρ ] = (−ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ ).) 4.39) Rather. [Ki .27) that we can identify the angular momentum and boost operators as 1 Ji = − ǫi jk Mjk . e (4. Jj ] = M0i . i = 1.40) We recall from (4.4. M0j ] = −Mij = −ǫij k Jk . Poincar´ algebra = t4 ⊕ so(3. Jj ] = ǫij k Jk .30) [Mµν . 3.2 Representations of the Lorentz group Let us next come back to the homogeneous Lorentz group and note that the Lie algebra in this case is given by (4.2 Representations of the Lorentz group 135 transformations. − 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 . it is what is known as a semi-direct sum of the two algebras. 2. 1). 2 Ki = M0i .

the generators in the new basis are all anti-Hermitian. † Mµν = i(xµ ∂ν − xν ∂µ ) = Mµν . (4.43) which also leads to the inverse relations Ji = (Ai + Bi ) . Ki = −i (Ai − Bi ) . namely. However.136 4 Representations of Lorentz and Poincar´ groups e where we have used (4. (4. This is a set of coupled commutation relations. i † Bi = −Bi .42) that we can assign the following hermiticity properties to the generators. consequently. in choosing the coordinate representation for the generators. The opposite hermiticity property of the generators of boosts. (compared to Ji ) is connected with the fact that such transformations are non-compact and. Ki . the finite dimensional representations of boosts are non-unitary (hence the opposite Hermiticity of Ki ). (4. namely. (Ji − iKi ) . Ki† = Ki . Ji† = −Ji . as can be seen from the hermiticity of the generators in the coordinate basis.13) in the last relation.44) Parenthetically. we have not been particularly careful about choosing Hermitian operators.47) . As a consequence of (4. if we define the generators with a factor of “i”. let us note from the form of the algebra in (4. infinite dimensional representations are unitary. (4.46) Namely. (4. 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 ) .45) This unconventional hermiticity for Ji arises because.45). we have A† = −Ai .

Bj ] = = = = [Ai .43). Kj ] + i [Ki . Aj ] = = = = [Bi . Jj ] − i [Ji . 2 ij 1 1 (Ji + iKi ) .49) Incidentally. Kj ] − i [Ki . Therefore. so(3.4.48) In other words. the algebra separates into two angular momentum algebras which are decoupled. Mathematically. the Lorentz group is double valued (doubly connected). Kj ] + i [Ki . it is more . the Lorentz algebra (4. (Jj − iKj ) 2 2 [Ai . Kj ]) 4 1 ǫ k Jk − iǫij k Kk − iǫij k Kk + ǫij k Jk 4 ij 1 k ǫ (Jk − iKk ) = ǫij k Bk . 2 ij 1 1 (Ji − iKi ) . Kj ]) 4 1 ǫ k Jk − iǫij k Kk + iǫij k Kk − ǫij k Jk = 4 ij = 0.2 Representations of the Lorentz group 137 In the new basis (4. Jj ] + [Ki . Jj ] − [Ki . (Jj + iKj ) 2 2 1 ([Ji . Jj ] − [Ki . as we have already seen in the last chapter. Bj ] = = 1 ([Ji . Jj ] + i [Ji . Jj ] − i [Ji . one says that the Lorentz algebra is isomorphic to the direct sum of two angular momentum algebras.42) takes the form 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 Ak . in this new basis. (4. (Jj − iKj ) 2 2 1 ([Ji . 1) ≃ so(3) ⊕ so(3) ≃ su(2) ⊕ su(2). (4.

. Denoting by jA and jB the eigenvalues of the Casimir operators A2 and B2 respectively for the two algebras. The finite dimensional unitary representations of each of the angular momentum algebras is well known. . ) to describe the Lorentz transformations. 2 2 (4. . .jB) (Λ)ψ (jA .jB ) (x′ ) = D(jA . 2 2 3 1 jB = 0. 1. 2 A (4. this represents the operator implementing finite transformations on the Hilbert space of states or wave functions as ψ ′ (jA .35) in connection with the covariance of the Dirac equation. Explicitly. . .jB ) “ ” (j ) (j ) i i −i θA Ai A +θB Bi B “ ” (j .52) where the finite parameters of rotation and boost can be identified with θi = 1 i i θ + θB . (4.51) where Λ represents the finite Lorentz transformation parameter. .138 4 Representations of Lorentz and Poincar´ groups e meaningful to consider the simply connected universal covering group of SO(3.jB ) (x) = D(jA ) (Λ)D(jB ) (Λ)ψ (jA .j ) (j .jB ) (just as the representation of the rotation group is denoted by D(j) ). (4.50) An irreducible nonunitary representation of the homogeneous Lorentz group.jB ) (x). Namely. . . .j ) −i θ i Ji A B +δi Ki A B D (Λ) = e = e . .53) .) (jA . much the same way we consider the universal covering group SU (2) of SO(3) to describe rotations. 2 A δi = i i i θ − θB . 1) which is SL(2. can be specified uniquely once we know the values of jA and jB and is labelled as D(jA . therefore. 1. we have 3 1 jA = 0. . we can write (This is the generalization of the S(Λ) matrix that we studied in (3.

dimensionality = 2. D(0.58) 1 corresponds to a two component spinor representation with spin 2 . dimensionality (4. in fact. 2 ) .4. 2 (4. . dimensionality = 2. j = 0.0) .56) The first few low lying representations of the Lorentz group are as follows.57) corresponds to a scalar representation with zero spin (which acts on the wave function of a Klein-Gordon particle). 1 1 D(0. j = . |jA − jB | + 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. .44)) J = A + B.0) . For jA = jB = 2 . For jA = jB = 0. 2 1 1 D( 2 . (4. jB = 0. .2 Representations of the Lorentz group 139 Such a representation labelled by (jA . . (4. we conclude that the values of the spin in a given representation characterized by (jA . dimensionality = 1.jB ) = (2jA + 1) (2jB + 1) . for jA = 0. jB ) will have the dimensionality D (jA . For jA = 1 . from our knowledge of the addition of angular momenta. 2 (4.54) and its spin content follows from the fact that (see (4. j = . These two representations are inequivalent and. 1 Similarly.59) also corresponds to a two component spinor representation with spin 1 2 . (4. jA + jB . .55) Consequently. jB ) can lie between j = |jA − jB |. jB = 2 .

(4.63) .48). 2 (4. 2 ) ⊗ D ( 2 . Let us now con- struct explicitly a few of the low order representations for the generators of the Lorentz group. 2 (1) ( 1 . we note that for the first few low order represen1 tations. 2 1 1 ( ) Bi 2 = − σi . dimensionality = 2 × 2 = 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. we scale all the generators Ji .140 4 Representations of Lorentz and Poincar´ groups e D( 2 . 2 ··· . Bi by a factor of i.62) Using (4. j = .61) This discussion can similarly be carried over to higher dimensional representations. (Namely. t and a spin 1 component x (under rotations) and the same is true for any other four vector.50). 2 (4. 4.0) .47). (Note that a four vector such as xµ has a spin zero component. 1.44). It actually corresponds to a reducible representation of the Lorentz group of the form 1 1 1 D(0. j = 0. Ai . Bi (0) 1 (1) Ai 2 = − σi .62) arises because ǫij = −ǫijk in (4. we have (We note here that the negative sign in the spin 2 k representation in (4.0) (1) (0) Ji 2 = Ai 2 + Bi = − σi .) It may be puzzling as to where the four component Dirac spinor fits into this description.0) (0) = −i Ai 2 − Bi Ki 2 = i σi .) Ai (0) = 0. = 0.1 Similarity transformations and representations.2. To compare with the results that we had derived earlier. 1 1 (4. this leads to the first two nontrivial representations for the angular momentum and boost operators of the forms 1 ( 1 . ··· . namely. dimensionality = 4. 2 ) . we now consider Hermitian generators by letting Mµν → iMµν as in (4. Ki .) From (4.

65) 1 then. 2 (0. They 2 provide the representations of angular momentum and boost for the left-handed and the right-handed Weyl particles.69) and (2.67) However.64) Equations (4.64) we see that the condition (4.4.0) Ji 2 = S −1 Ji 2 S. from (4.98) (or (3. if we can find a similarity transformation S leading to (0.78)). Therefore. 2 (4. (4. 0 and 0. (4. 1 ) 2 = 1 − 2 σi Ji (Dirac) 0 1 − 2 σi 0 .64). Two representations are said to be equivalent. For example. 1 ) (1) (0) Ki 2 = −i Ai − Bi 2 i = − σi . we can obtain the representation of the Lorentz generators for the reducible four component Dirac spinors as (0. (4. the two representations la1 belled by 1 . ( 1 .0) (0. 1 ) ( 1 . 2 are inequivalent representations. (Dirac) Ki i 2 σi 0 0 i − 2 σi .0) 2 ⊕ Ki (0. 1 ) (0) Ji 2 = Ai + Bi 2 = − σi . From (4.0) = Ji 2 ⊕ Ji 2 = = Ki ( 1 .65) would imply the existence of an invertible matrix S such that S −1 σi S = σi .64) give the two inequivalent representations of dimensionality 2 as we have noted earlier.63) and (4. S −1 σi S = −σi . 1 ) Ki 2 = S −1 Ki 2 S. if there exists a similarity transformation relating the two.63) and (4. 0 and 0. we note that these do not resemble the generators of the Lorentz algebra defined in (3.71) and (3. we would say that the two representations 2 .66) which is clearly impossible.2 Representations of the Lorentz group 141 and 1 (1) (0. 1 are 2 equivalent. In fact. .63) and (4. 1 ) ( 1 .

we can apply the inverse similarity transformation in (2.71) and (3. i 0 i γ .69) and (2. we note that the angular momentum and boost operators in (4.121) to obtain Mµν = Ji (Dirac) PD = S −1 Ji = 1 2 (Dirac) S − 1 σi 0 2 1 0 − 2 σi 0 σi ½ ½ −½ ½ − 1 σi 2 0 −1 2 ½ −½ ½ ½ = Ki (Dirac) PD 1 PD = − ǫi jk σjk .67) and in our earlier discussion in (3. We note that in the Weyl representation for the gamma matrices defined in (2. (4.78)) are equivalent since they are connected by a similarity transformation that . consequently.98) (see also (3. if we would like the generators in the standard Pauli-Dirac representation (which is what we had used in our earlier discussions).68) As a result. γW = ǫijk 2 W .142 4 Representations of Lorentz and Poincar´ groups e This puzzle can be understood as follows. (4. 2 0i Therefore. we note that the generators in (4.70) = σi i − 2 σi 0 1 PD σ .119). On the other hand.γ = 2 W W −iσi 0 σk 0 0 iσi 0 σk 0i σW = .69) 2 µν and. ij σW = i i j γ .67) are obtained from 1 W σ . give a representation of the Lorentz generators in the Weyl representation. 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.

2 Representations of the Lorentz group 143 relates the Weyl representation of the Dirac matrices in (2. as we discussed earlier.4). There is yet another interesting example which sheds light on similarity transformations between representation. On the other hand.4. For example.18)). (3. the two representations for the generators constructed from two different perspectives (for the same four vector representation) appear rather different and. Let us illustrate this for the simpler case of rotations. (4.71) we can immediately read out from (3. (Here we will use 3dimensional Euclidean notation without worrying about raising and lowering of the indices. we can deduce the matrix form of the corresponding generators to be . therefore.72) Similarly. from the infinitesimal change in the coordinates under a Lorentz transformation (see. Surprisingly.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. Let us consider a three dimensional infinitesimal rotation of coordinates around the z-axis as described in (3. (4. considering infinitesimal rotations of the coordinates around the x-axis and the y-axis respectively. The case for Lorentz transformations follows in a parallel manner. for example.) Representing the infinitesimal change in the coordinates as δxi = iǫ (J3 )ij xj . we can determine a representation for the generators of the Lorentz transformations belonging to the representation for the four vectors.119) to the standard Pauli-Dirac representation. there must be a similarity transformation relating the two representations. from the Lie algebra point of view the four vector representation corresponds to jA = jB = 1 (see (4.

Das (Hindustan Book Agency.  (4.73) in spite of the fact that they belong to the same representation for j = 1. To construct the similarity transformation (which actually is a unitary transformation). they are equivalent. 2 0 1 0 (4. A. page 289 or Lectures on Quantum Mechanics. On the other hand. therefore. k = 1. (4. in fact. New Delhi). Quantum Mechanics: A Modern Introduction.72) and (4. page 182 (note there is a typo in the sign of the 23 element for L2 in this reference) 1 .   J2 =  0  0 0 i −i 0 0  0 0 . Das and A. 0 0 −1   0 −1 0 i   = √ 1 0 −1 . This is. let us note that the generators obtained (LA) from the Lie algebra are constructed by choosing the generator J3 See. j. 3. Jj ] = iǫijk Jk .75) (LA) J1 (LA) J2 (LA) J3 which look really different from the generators in (4. 2 0 1 0   1 0 0   = 0 0 0  . therefore provide a representation for the generators of rotations. 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 .73) and. for example.74)  J1 = 0  0 0 0 0 0 −i 0  i .72) and (4.144 4 Representations of Lorentz and Poincar´ groups e It can be directly checked from the matrix structures in (4. A. the representation in the space of three vectors which would correspond to j = 1. Melissinos (Gordon and Breach).73) that they satisfy [Ji . 2. i. This puzzle can be resolved by noting that there is a similarity transformation that connects the two representations and. C.

  −1 (4.  0   U† =  0  1 − √2  1 √ 2 i √ 2 0 i √ 2  −1 .2 Representations of the Lorentz group 145 to be diagonal. (4.4.77) If we now define a similarity transformation (unitary) S −1 = U † . S = U.78) then.72) that the three normalized eigenstates of J3 have the forms  1 √  2  i − √  . 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  .  0 0  (4.  1 √  2 i − √  2 0 0 −1 1 − √2 U= 0  i − √2  .76)  0  Let us construct a unitary matrix from the three eigenstates in (4.   0   ψ (j3 =0) =  0  .76) which will diagonalize the matrix J3 .  2 ψ (j3 =1) =  0 ψ (j3 =−1)  1  −√  i2  = − √2  . Let us note from (4.

therefore. 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 . (4.146  4 Representations of Lorentz and Poincar´ groups e = S −1 J2 S   0 1 0 1   (LA) √ 1 0 1 = J1 .72).79) This shows explicitly that the two representations for J corresponding to j = 1 in (4.75) which look rather different are. . related by a similarity transformation and.73) and (4. in fact.   √ 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. 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 .

Therefore.4. Pµ = 0.3 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. The operators Mµν generate infinitesimal Lorentz transformations through commutation relations and the relation above. Let us define a new vector operator. Mµν ] + [Pλ . known as the Pauli-Lubanski operator. Mµν ] P λ = P λ (ηλµ Pν − ηλν Pµ ) + (ηλµ Pν − ηλν Pµ ) P λ = Pµ Pν − Pν Pµ + Pν Pµ − Pµ Pν = 0. 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. P 2 . namely. Mµν = = = P λ [Pλ .3 Unitary representations of the Poincar´ group e 147 4. from the generators of the Poincar´ group as e . it is known that it has only infinite dimensional unitary representations except for the trivial representation that is one dimensional. Mµν P λ Pλ . Since Poincar´ e group is non-compact (like the Lorentz group).81) can be intuitively understood as follows.81) The last relation in (4. P λ Pλ . let us note that the operator P 2 = η µν Pµ Pν = P µ Pµ . Mµν act as Hermitian operators. In order to determine the unitary representations.38) e since it commutes with all the ten generators. (4. Pµ P 2 . we seek to find unitary representations in some infinite dimensional Hilbert space where the generators Pµ . which is supposed to characterize the infinitesimal transformation of P 2 .80) defines a quadratic Casimir operator of the Poincar´ algebra (4. it is useful to study the representations of the Poincar´ group. (4. This e would help us in understanding the kinds of theories we can consider and the nature of the states they can have.

the vector Wµ is orthogonal to Pµ . the order of Pµ and Mνλ are irrelevant in the definition of the Pauli-Lubanski operator.) In general. 2 (4.148 4 Representations of Lorentz and Poincar´ groups e Wµ = 1 1 µνλρ ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . 2 (4. First. in general.38)) which vanish when contracted with the anti-symmetric LeviCivita tensor. (4. Furthermore. therefore. Let us define the dual of the generators of Lorentz transformation as M µν = Mµν 1 µνλρ ǫ Mλρ .82) also as W µ = Pν M µν = M µν Pν . Let us next calculate the commutators between W µ and the ten generators of the Poincar´ group. this is not true for massless theories as we will see shortly.84) With this.83) that. (4. (However.83) implies that the PauliLubanski operator has only three independent components (both in the massive and massless cases).82) The commutator between Pµ and Mνλ introduces metric tensors (see (4. we note that Pµ W µ = 1 µνλρ ǫ Pµ Pν Mλρ = 0. It follows from (4. we have e . 2 2 (4.83) which follows from the fact that the generators of translation commute. we can write (4.85) where the order of the operators is once again not important. 2 1 = − ǫµνλρ M λρ . As a result.

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

it behaves like a second rank antisymmetric tensor under a Lorentz transformation. Mστ ] = Pν M µν . Namely. ǫµνλρ ǫστ ζρ = −ηµσ (ηντ ηλζ − ηνζ ηλτ ) − ηµτ (ηνζ ηλσ − ηνσ ηλζ ) −ηµζ (ηνσ ηλτ − ηντ ηλσ ) . .30)). then. that [Wµ . Let us note here.150 4 Representations of Lorentz and Poincar´ groups e = −ηµσ Mντ − ηντ Mµσ + ηµτ Mνσ + ηνσ Mµτ . Wν ] = = Wµ . (4.88) Equation (4.90) = −ǫµν λρ Wλ Pρ . 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. 1 1 λρσ ǫν Mρσ Pλ = ǫνλρσ [Wµ . Using this. Mστ ] M µν + Pν M µν . the operator Mµν behaves exactly like the generators of Lorentz transformation (see (4.87) simply says that under a Lorentz transformation. for completeness as well as for later use. Namely. 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 . Mστ = [Pν . we can now evaluate [W µ .89) In other words. (4. Mρσ ] Pλ 2 2 1 λρσ ǫ (ηµρ Wσ − ηµσ Wρ ) Pλ 2 ν (4. it transforms like a vector which we should expect since it has a free Lorentz index.87) Here we have used the identity satisfied by the four dimensional LeviCivita tensors. (4.

therefore. 4. L2 = Lµ Lµ with Lµ = P ν Mνµ . Mµν M µν . labelled by the momentum eigenvalues. 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 ).92) then.93) However.91) which is to be expected since Wµ W µ is a Lorentz scalar.1 Massive representation. this would also represent a Casimir operator of the Poincar´ e algebra since Wµ commutes with the generators of translation (see (4.86)). consequently. the representations can be labelled by the eigenvalues of these operators. to be diagonal (they satisfy an Abelian subalgebra). (4. of course. pµ . The eigenstates of the momentum operators |p are. which we treat separately. In fact. To find unitary irreducible representa- tions of the Poincar´ algebra.3. It can be shown that P 2 and W 2 represent the only Casimir operators of the algebra and. Mστ ] = W µ [Wµ . cannot represent Casimir operators of the algebra. we can choose the momentum operators. satisfying . (4. Therefore. Mστ ] W µ = W µ (ηµσ Wτ − ηµτ Wσ ) + (ηµσ Wτ − ηµτ Wσ ) W µ = Wσ Wτ − Wτ Wσ + Wτ Wσ − Wσ Wτ = 0. Mστ ] + [Wµ . There are other Lorentz scalars that can be constructed from Pµ and Mµν such as Mµν M µν . without loss of generality. The irreducible representations of the Poincar´ group can be clase sified into two distinct categories. we conclude that if we define an operator W 2 = W µ Wµ .3 Unitary representations of the Poincar´ group e 151 It follows now from (4. we choose the basis vectors of the repe resentation to be eigenstates of the momentum operators. it is easy to check that these do not commute with the generators of translation and. (4.89) that [W µ Wµ .4. Pµ . Namely.

96) Here m denotes the rest mass of the single particle state and we assume the rest mass to be non-zero. 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 . 2 2 (4. (4. let us study this operator in some detail. P 2 |p = pµ pµ |p .88).95) (4.94) and in this basis. (4. Therefore. using (4.152 4 Representations of Lorentz and Poincar´ groups e Pµ |p = pµ |p . where p 2 = p µ p µ = m2 .97) Therefore. However. the eigenvalues of W 2 are not so obvious. namely. We recall that Wµ = 1 µνλρ 1 ǫ Pν Mλρ = ǫµνλρ Mλρ Pν . the eigenvalues of the operator P 2 = P µ Pµ are obvious.

Let us note that in the rest frame (4. 2 (4.100) = − m2 M ij Mij .99). pµ = (m.101) where Jk represents the total angular momentum of the particle. To understand the meaning of this operator. 2 Recalling that (see (4.102) can also be derived in an alternative manner which is simpler and quite instructive. p 2 = p µ p µ = m2 .26)) Mij = ǫij k Jk .4. takes the form 1 W 2 = − m2 M λρ Mλρ − m2 Mλ0 M 0λ 2 1 = − m2 2M 0λ M0λ + M ij Mij + m2 M 0λ M0λ 2 1 (4.98) where we have simplified terms in the intermediate steps using the anti-symmetry of the Lorentz generators. 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 .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 .99) and the operator W 2 acting on such a state. (4. In this frame. 0. (4.102) The result in (4. (4. the Pauli-Lubanski operator (4.82) has the form . 0. 0) . let us go to the rest frame of the massive particle.

2 m 1 iνλρ ǫ Mλρ pν = − ǫijk Mjk = 2 2 m jk = − ǫi Mjk = mJi . That is.104) which is the result obtained in (4. (4. 0.102). Technically. we can ask what Lorentz transformations would leave such a vector invariant. The dimensionality of the representation can also be understood in an alternative manner as follows. 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. for a massive particle at rest. Clearly. 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). Therefore. therefore. for a massive particle. It is not hard to see that all possible 3dimensional rotations would leave such a vector invariant. this simply measures the spin of the particle. Therefore.105) Thus. we see that the representations with p2 = 0 can be labelled by the eigenvalues of the Casimir operators. we can think of W 2 as being proportional to J 2 and in the rest frame of the particle.34) as well as (4. s).12). Namely.154 4 Representations of Lorentz and Poincar´ groups e W0 = Wi 1 0ijk ǫ Mjk pi = 0. 0). 0. rotations around the x or the y or the z axis will not change the time component of a four vector (recall that it is the spin 0 component of a four vector) and. all the degenerate states can be labelled not . (m. 2 (4. For a state at rest with momentum of the form pµ = (m. It follows now that W 2 = W µ Wµ = W i Wi = m2 J i Ji = −m2 J2 . we find eigenvalues of W 2 : w2 = −m2 s(s + 1).103) where we have used (1. would define the stability group of such vectors. these would define an invariant subgroup of the Lorentz group and will lead to the degeneracy of states. (4.

let us note from (4. s = 0. In contrast to the massive representa- tions of the Poincar´ group. a state at rest is an eigenstate of the P0 operator. p = 0.83) we see that our states in the representation should also satisfy Pµ W µ = 0. −s. Mij ] = 0. namely. from (4. acting on states in such a vector space. we note that (see (4. up to a normalization factor.109) . the representations for a massless partie cle are slightly more involved.30)) [P0 . (4.95)) pµ pµ = 0. s − 1.3. 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. · · · and 2 ms = s. −s + 1. but also by the eigenvalues of three dimensional rotations. In this case. pµ = (p. the three nontrivial Pauli-Lubanski operators correspond to the generators of symmetry of the “little group” in the rest frame. The basic reason behind this is that the “little” group of a light-like vector is not so obvious. · · · . −p). In closing. 4. (4. (4. This defines the 2s + 1 dimensional representation for a massive particle of spin s. commute with P0 . From the Lorentz algebra. we note that (see (4. 1 . 0. we would have (see (4.2 Massless representation.103) that.4.83)) P 2 = 0. (4.106) Namely. 0. 1.107) Consequently. Namely. the operators Mij .108) However. Consequently. This can also be seen algebraically.3 Unitary representations of the Poincar´ group e 155 just by the eigenvalue of the momentum. which generate 3-dimensional rotations and are related to the angular momentum operators.

namely. p0 h= (4. 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. To determine h. Such representations do not correspond to physical particles and. the simultaneous eigenstates of P 2 and W 2 would .111) where h represents a proportionality factor (operator). 2 2 (4. 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. therefore. consequently. On the other hand. acting on states in such a space. we can easily show that the action of W µ in such a space is proportional to that of the momentum vector. we will not consider such representations. W 2 = W µ Wµ = 0.114) This is nothing other than the helicity operator (since L · p = 0) and. (4. or.110) In the first case. W µ has the form W µ = −hpµ . W 2 = W µ Wµ = 0. in the second case where W 2 = 0 on the states of the representation. let us recall that Wµ = 1 µνλρ ǫ Pν Mλρ . (4.111) we conclude that in this space J·p . for a massless particle if W 2 = 0.113) W0 = Comparing with (4.112) from which it follows that acting on a general momentum basis state |p (not necessarily restricting to massless states). 2 (4. namely.88)) 1 0ijk 1 ǫ pi Mjk = ǫijk pi ǫjkℓ Jℓ = −p · J = −J · p. it would lead to (see (4.

W 0 . but also algebraically by recognizing that a light-like vector of the form being considered is an eigenstate of the operator P0 − P3 . 0.116) Let us now determine the dimensionality of the massless representations algebraically. 2 W0 = W1 = W2 = W3 = (4. the Pauli-Lubanski operator (4. indeed represents the helicity operator up to a normalization as we had noted in (4.113).3 Unitary representations of the Poincar´ group e 157 correspond to the eigenstates of momentum and helicity. in the massless case.103) and (4.83).82) takes the form 1 0ijk 1 ǫ Mjk Pi = ǫijk Mjk pi = −pM12 .117) . let us note here that in the light-like frame (4. For completeness.115) We see from both (4. Let us recall that we are considering a massless state with momentum of the form pµ = (p. (4.4. This can be seen intuitively from the fact that the motion of the particle is along the z axis.115) that. the set of Lorentz transformations which would leave this four vector invariant must include rotations around the z-axis. −p) and we would like to determine the “little” group of symmetries associated with such a vector. namely.115) that the Pauli-Lubanski operator indeed has only three independent components because of the transversality condition (4. We recognize that in this case. 2 1 2νλρ ǫ Mλρ pν = −p (M01 − M13 ) . (P0 − P3 ) |p = 2p|p . It follows now from (4. 0. We also note from (4. as we had pointed out earlier.107). 2 1 3νλρ ǫ Mλρ pν = −pM12 = W 0 . 2 2 1 1νλρ ǫ Mλρ pν = p (M02 − M23 ) .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.

2.90) that in the . M02 − M23 ] = −Π1 . We note that the algebra of the symmetry generators takes the form [M12 .38). we say that the stability group or the “little” group of a light-like vector is E2 .158 4 Representations of Lorentz and Poincar´ groups e Furthermore. [M12 . M12 is the generator of rotations around the z axis or in the two dimensional plane and Π1 . Π2 ] |p = [P0 − P3 . E2 (which consists of translations and rotation). (4. (4. M12 ] = 0. We note from (4. therefore. we see that up to a normalization. Furthermore. These represent all the symmetries of the light-like vector (state).120) and. Π2 ] = [M01 − M13 . [Π1 . we see that e [P0 − P3 . (4. Π2 ] = [M12 . but can be easily understood as follows. 3 in (4. M01 − M13 ] = Π2 . Π1 ] = [M12 . [P0 − P3 . = [P0 − P3 . To determine the other symmetries of a light-like vector. p W1 . i = 1. Π2 have the same commutation relations as those of translations in this two dimensional space. also define symmetries of light-like states. namely. Thus. from the Poincar´ algebra in (4. in fact.118) so that rotations around the z-axis define a symmetry of the lightlike vector (state) that we are considering. the generators of symmetry of the “little” group. M02 − M23 ] = 0. Clearly. This may seem puzzling.121) Namely. the three independent Pauli-Lubanski operators are. it is isomorphic to the algebra of the Euclidean group in two dimensions.119) p Π1 = M01 − M13 = Π2 = M02 − M23 = − It follows now that these operators commute with P0 − P3 in the space of light-like states. Π1 ] |p [P0 − P3 . (4. M01 − M13 ] |p = 2P1 |p = 0. M02 − M23 ] |p = 2P2 |p = 0. comparing with W i .115). let us define two new operators as W2 . as we had also seen in the massive case.

123) that (h corresponds to the helicity quantum number) (Π1 ± iΠ2 ) |p. [M12 . Furthermore. then. h = 0 = Π2 |p. their eigenvalues can take any value. 1 2 (4. as we have already pointed out.4. the space of physical states would also include the state with the opposite helicity (recall that helicity changes sign under space reflection. (4. namely. Clearly. does not correspond to any physical system. we see that in the space of light-like momentum states W 2 ∝ Π2 .121) that Π1 ∓ iΠ2 correspond respectively to raising and lowering operators for M12 .123) and comparing with (4. (4. if W 2 = 0 in this space. Let us note from (4. massless theories with nontrivial spin that are parity invariant would have two dimensional representations corresponding to the highest and the . As a result. they generate transformations which will leave the momentum basis states invariant. what we have been investigating within the context of “little” groups.122) Let us also note for completeness that the Casimir of the E2 algebra is given by Π2 = Π2 + Π2 = (Π1 ∓ iΠ2 )(Π1 ± iΠ2 ). Since Π1 .86) as the transformations that leave pµ invariant. then it follows from (4.122) that spin can take an infinite number of values which. of course. As a result.3 Unitary representations of the Poincar´ group e 159 momentum basis states (where pρ is a number).146)). Π2 correspond to generators of “translation”. we note from (4.) This corresponds to the one dimensional representation of E2 known as the “degenerate” representation. we can say that Π1 |p. can be thought of as generators of some transformations. h and this is the reason for the assertion in (4. Namely. follows from (4. therefore. if our theory is also invariant under parity (or space reflection). Π1 ∓ iΠ2 ] = ±i (Π1 ∓ iΠ2 ) . see (3. The meaning of the transformations. such a state would correspond to the highest or the lowest helicity state. h = 0.124) (Alternatively. if W 2 = 0 in this space of states. On the other hand.116).111). This is. the Pauli-Lubanski operators satisfy an algebra and.

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

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

3) |x|→∞ The derivatives of the field are also assumed to fall off at infinity. has the form + m2 φ(x) = ∂µ ∂ µ + m2 φ(x) = 0. each field must belong to a specific representation of the Lorentz group. We would like to have a relativistic. φ(x) = φ∗ (x). lim φ(x) → 0. On the other hand. namely. free Klein-Gordon equation with mass m. We also assume that the field falls off rapidly at spatial infinity (this is necessary for the theory to be well defined). Secondly. (5. e the quantization of the fields as operators can then be carried out in the Hamiltonian formalism. Given a relativistic dynamical equation. the classical free Klein-Gordon theory where we assume that the field variable φ(x) ≡ φ(x.2) where φ (x ) denotes the Lorentz transformed field. However. covariant theory for fundamental physical processes and. the dynamical equation involving the field variables must be covariant under Poincar´ transformations.1: Oscillating string fixed at both ends.162 5 Free Klein-Gordon field theory Figure 5. therefore. whenever any ambiguity arises in the quantization. so far. The classical. one reverts back to the Hamiltonian formalism for its resolution. namely. let us look at the simplest of field theories. it is generally preferrable to work in the Lagrangian formulation which is manifestly covariant. since the Hamiltonian is not a manifestly covariant concept. as we have seen (see (1. t) is real and that it behaves like a scalar under a Lorentz transformation. φ (x ) = φ(x). With this in mind.40)).4) . (5. (5.

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

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) .10) Here we have used Gauss’ theorem (as well as the vanishing of field variables asymptotically (5. We have .3)) to simplify the surface term.164 5 Free Klein-Gordon field theory most on the first order derivatives of the field variables. ti ) = δφ (x. tf ) = 0. subject to the boundary conditions (5. ti (5. 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). 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. Note that under an infinitesimal change tf (5.8) δφ (x.

13) Therefore. (5.12) Thus.15) . ∂L ∂L − ∂µ = 0. the change in the action is given by tf δS = ti d4 x ∂L ∂L − ∂µ ∂φ(x) ∂∂µ φ(x) δφ(x). Thus. infinitesimal change in the field variable. we conclude that under an arbitrary.9). (5. we can think of the dynamical equations as arising from the minimum principle associated with a given action.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). (5. for the Klein-Gordon equation.10) – vanishes because of the boundary conditions (5. δφ (x. namely. As in point particle mechanics. (5. ∂µ ∂ µ φ + m2 φ = 0. ∂µ φ(x)).11) We note that the “surface term” – the last term in the expression (5. δS = 0 only if). the action will be stationary under an arbitrary change in the field variable subject to the boundary conditions provided (namely.5.14) This is known as the Euler-Lagrange equation associated with the action S or the Lagrangian density L(φ. tf ) . ti ) = 0 = δφ (x. ∂φ(x) ∂∂µ φ(x) (5.

2 2 L= (5.166 5 Free Klein-Gordon field theory we can ask whether there exists a Lagrangian density whose EulerLagrange equation will lead to this dynamical equation. ∂φ(x) ∂∂µ φ(x) or. in this case. the Euler-Lagrange equation. we see that the Lagrangian density 1 m2 2 ∂µ φ∂ µ φ − φ . (5.3)). Note that the Lagrangian density L in (5. the action does not change either. ∂φ(x) ∂L = ∂ µ φ(x).17) ∂L ∂L − ∂µ = −m2 φ(x) − ∂µ ∂ µ φ(x) = 0.19) is manifestly Lorentz invariant.16) ∂L = −m2 φ(x). This would simply correspond to adding a zero point energy.19) gives the Klein-Gordon equation for a free. It is worth noting here that one can add a constant to L without changing the equation of motion. ∂∂µ φ(x) Therefore. Similarly. if we assume the fields to fall off rapidly at infinite separation (see (5. gives (5. . 2 2 L= then (5. ∂µ ∂ µ + m2 φ(x) = 0. Note that if we choose 1 m2 2 ∂µ φ∂ µ φ − φ . adding a total divergence to L also does not change the equations of motion.18) In other words. Of course. in this case. real scalar field as its Euler Lagrange equation.

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.19) (xµ in this case merely labels space-time coordinates) . lead to the dynamical equations in the Hamiltonian form as x = {x.22) The classical canonical Poisson bracket relations between the coordinate and the conjugate momentum have the familiar form {x. x} = 0 = {p. (5. x) . x) . p} = 1. ˙ ˙ (5. ∂x ˙ p= (5. the Lagrangian is a function of the coordinate and the velocity of the particle (a dot denotes a derivative with respect to t). in turn. p) = px − L (x. p} . ˙ p = {p.3 Quantization Since we have a Lagrangian description for the classical Klein-Gordon field theory. {x. H} . ˙ (5.5. H} .20) The conjugate momentum associated with the dynamical variable x is defined as ∂L .3 Quantization 167 5. on the other hand.21) and this leads to the definition of the Hamiltonian in the form (this is an example of a Legendre transformation) H(x. L = L (x.23) These. the basic variables are φ(x) and ∂µ φ(x) and we have from (5. In the case of a point particle. ˙ (5.24) In the case of the classical scalar (Klein-Gordon) field theory.

the KleinGordon equation in this case) can be written in the Hamiltonian form as ˙ φ(x) = {φ(x). in turn. and a Hamiltonian d3 x H. = δ3 (x − y). φ(y)}x0 =y0 {φ(x). allow us to define a Hamiltonian density as ˙ H = Π(x)φ(x) − L.29) then.30) Let us examine all of this in detail for the case of the free KleinGordon theory. Note that the Lagrangian density in this case is given by .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 (5.25) In an analogous manner to the point particle mechanics in (5.21). we define a momentum canonically conjugate to the field variable φ(x) as ∂L . we can show that the dynamical equation (namely.26) which will.27) H= (5. Π(y)}x0 =y0 . H} . ∂µ φ(x)) . (5. ˙ Π(x) = {Π(x). (5.168 5 Free Klein-Gordon field theory L = L (φ(x). (5. (5.23)) {φ(x). ˙ ∂ φ(x) Π(x) = (5. Π(y)}x0 =y0 = 0 = {Π(x). H} .

34) With the equal time canonical Poisson bracket relations defined to be (see (5. φ − ∇φ · ∇φ − 2 2 2 (5.35) . Π(y)}x0 =y0 . we can identify the conjugate momentum with ∂L ˙ = φ(x). = δ3 (x − y). (5.33) 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x). 2 2 2 (5. 2 2 2 which gives the Hamiltonian for the theory H = = d3 x H d3 x 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x) . ˙ ∂ φ(x) Π(x) = (5.29)) {φ(x). Π(y)}x0 =y0 we obtain = 0 = {Π(x).5.3 Quantization 169 L = = m2 2 1 ∂µ φ∂ µ φ − φ 2 2 1 ˙2 1 m2 2 φ .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 (5.32) Consequently. φ(y)}x0 =y0 {φ(x).

lead to ˙ φ(x) = Π(x). the Hamiltonian equations (5.170 5 Free Klein-Gordon field theory ˙ φ(x) = {φ(x). Π2 (y) d3 y Π(y) {φ(x). (5. Π(y)} x0 =y 0 d3 y Π y. x0 = ∇ · ∇φ(x) − m2 φ(x). φ(y)} x0 =y0 = d3 y ∇y φ y.37) Thus. since the Hamiltonian is time independent (this can be checked easily using the equations of motion and will also be shown in (5. x0 δ3 (x − y) = Π(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 (5.73)). H} = Π(x).38) .29). −δ3 (x − y) (5. we also have ˙ Π(x) = {Π(x). 1 2 d3 y 1 m2 2 1 2 Π (y) + ∇y φ · ∇y φ + φ (y) 2 2 2 x0 =y 0 x0 =y 0 d3 y φ(x). (5. H} = = = = 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).36) and (5.37). Furthermore. x0 · ∇y −δ3 (x − y) +m2 φ y. in this case. ∇y φ(y)} + m2 φ(y) {Π(x).36) where we have used the fact that. ˙ Π(x) = ∇ · ∇φ(x) − m2 φ(x) = ∇ 2 φ(x) − m2 φ(x).

therefore. in the first quantization we quantize x’s and p’s. therefore.5. but not the wave function. This is.2)) satisfying the equal time commutation relations (remember = 1) [φ(x). brings out the classical Hamiltonian description of the classical KleinGordon field theory. ∂µ ∂ µ + m2 φ(x) = 0.39) or. (Namely. Π(y)]x0 =y0 = 0 = [Π(x).40) Note that this is the same Klein-Gordon equation as we had studied earlier in chapter 1. In the second quantization. or. We already know that the plane wave solutions (1. of course.4 Field decomposition Given any complete set of solutions of the classical Klein-Gordon equation. In the quantum theory. However. we can expand the field operator φ(x) in such a basis. (5. here we are treating φ(x) as an operator which is quantized and not as a wavefunction as we had done earlier. Once we have the classical Hamiltonian description of a physical system. the quantization of such a system is quite straightforward. This is conventionally referred to as second quantization. we are supposed to treat φ(x) and Π(x) as Hermitian operators (because we are dealing with real fields.) 5. = iδ3 (x − y).4 Field decomposition 171 These are first order equations as Hamiltonian equations should be and from these. we quantize the “wavefunction”.40) (which is a second order Euler-Lagrange equation) and this discussion. ¨ φ − ∇ 2 φ(x) + m2 φ(x) = 0. see (5.41) of the KleinGordon equation define a complete basis and. the free Klein-Gordon equation (1. φ(y)]x0 =y0 [φ(x). we obtain the second order equation ¨ ˙ φ(x) = Π(x) = ∇ 2 φ(x) − m2 φ(x). we can expand the field operator in this basis as . Π(y)]x0 =y0 . (5.

42) using the field expansion (5.41) in the Klein-Gordon equation. (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. we obtain 1 (2π) or. different. we can denote φ(k) = δ k2 − m2 a(k).41). 1 (2π) 3 2 3 2 ∂µ ∂ µ + m2 d4 k e−ik·x φ(k) = 0. As a result. Namely.41) where we recognize that the operators φ(x) and φ(k) are Fourier transforms of each other (up to a multiplicative factor). φ(k) = 0. (5.172 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 d4 k e−ik·x φ(k).46) . we obtain 1 (2π) 3 2 φ(x) = d4 k δ k2 − m2 e−ik·x a(k).43) d4 k −k2 + m2 φ(k)e−ik·x = 0. (5. (5. the Fourier transform has non-vanishing contribution (support) only on the massshell of the particle defined by k2 = (k0 )2 − k2 = m2 . (5. in general. Substituting this back into the expansion (5.45) (5. We also emphasize here that the hermiticity properties of a given field and its Fourier transform are.) Since φ(x) satisfies the Klein-Gordon equation ∂µ ∂ µ + m2 φ(x) = 0. This shows that unless k2 = m2 .

(5.2: Mass shell on which φ(k) has support. the field expansion becomes . 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 .46). or. We note that the argument of the delta function vanishes for k 2 = m2 . k0 = ± k2 + m2 ≡ ±Ek .48) Using this relation in (5.5. (5. k0 < 0 Figure 5. or.4 Field decomposition k0 173 k2 = m2. k0 > 0 k k 2 = m 2 .47) Correspondingly. (k0 )2 = k2 + m2 .

51). Therefore. (5. we obtain a† (k) = a(−k). or. 2k0 (5.52) φ(x) = d3 k −ik·x e a(k) + eik·x a† (k) . we obtain 1 (2π) 2 3 φ(x) = d3 k −ik·x e a(k) + eik·x a(−k) .53) .49) and identifying k0 = Ek > 0.49) Changing k → −k in the second term in (5. so that we can write 1 (2π) 3 2 a† (−k) = a(k). 1 (2π) = 3 2 d3 k ik·x † e a (k) + e−ik·x a† (−k) 2k0 1 3 (2π) 2 d3 k −ik·x e a(k) + eik·x a(−k) . k) 0 x0 +ik·x = 1 (2π) 3 2 d3 k 1 0 e−iEk x +ik·x a(Ek . (5. φ† (x) = φ(x). k) .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. 2k0 (5.174 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 dk0 d3 k × e−ik 1 δ k0 − Ek + δ k0 + Ek 2Ek a(k0 . k) 2Ek 0 +ik·x + eiEk x a(−Ek . 2k0 (5.2)).

(5. eik·x a† (k). conventionally one defines a(k) .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.56) . a(k) = √ 2k0 a† (k) a† (k) = √ . (5. (5.5 Creation and annihilation operators Given the field expansion (5.57) so that we can write the field operator as a sum of its positive and negative energy parts. 2k0 (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). In fact.5. φ(x) = φ(+) (x) + φ(−) (x).5 Creation and annihilation operators 175 Here we are supposed to understand that k0 = Ek = k2 + m2 > 0.53) as d3 k (2π)3 2k0 φ(x) = e−ik·x a(k) + eik·x a† (k) .55) and in terms of these operators we can write the field operator (5. Thus.58) 5.54) that a(k) and a† (k) are really functions of the three momentum k alone. (5.54) It is clear from (5.

61) ← → where ∂t is defined in (1. Namely.176 5 Free Klein-Gordon field theory φ(x) = d3 k (2π)3 2k0 e−ik·x a(k) + eik·x a† (k) .59) we obtain the expansion for the conjugate momentum in (5. It is clear from (5. (5.56). we note from these relations that we can write the operators a(k) and ˙ a† (k) in terms of φ(x) and Π(x) as (recall that Π(x) = φ(x)) a(k) = = a† (k) = = − 1 (2π)3 2k0 i (2π)3 2k0 1 (2π)3 2k0 i (2π)3 2k0 d3 x eik·x k0 φ(x) + iΠ(x) ← → d3 x eik·x ∂t φ(x). This can also be checked explicitly.32) to be ˙ Π(x) = φ(x) = −i d3 k k0 e−ik·x a(k) − eik·x a† (k) . (5. (5.60) 2(2π)3 The two defining relations in (5.60) are invertible. the expressions on the righthand side must also be time independent. taking the time derivative of the first relation in (5. d3 x e−ik·x k0 φ(x) − iΠ(x) ← → d3 x e−ik·x ∂t φ(x). For example. It is important to emphasize here that in the expressions in (5.61) we obtain (neglecting the overall multiplicative factors) . it is assumed that k0 = Ek (namely.61).59) and (5. k0 is not an independent variable).61) that since the left-hand sides are independent of time.

a(k′ ) + eik·x+ik ·y a† (k). [Π(x).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. integrated the space derivatives by parts assuming that the surface terms vanish (see (5.61) can also be derived in a simple manner (or note that the second relation is the Hermitian conjugate of the first and. a† (k′ ) . therefore. a(k′ ) − e−ik·x+ik ·y a(k). φ(y)]x0 =y0 = ′ × e−ik·x−ik ·y a(k). Imposing now the quantization relations between φ(x) and Π(x) (see (5.61) are indeed time independent.3)) and have √ made the identification k0 = Ek = k2 + m2 .40)) we obtain d3 k (2π)3 2k0 d3 k ′ (2π)3 2k′ 0 ′ [φ(x).62) Here. a(k′ ) + e−ik·x+ik ·y a(k). Π(y)]x0 =y0 = − ′ ′ ′ x0 =y 0 (5. in the intermediate steps. must be time independent). (5. This shows explicitly that the integrals in (5. we have used the fact that φ(x) satisfies the Klein-Gordon equation. the time independence of the second expression in (5. Similarly.5.63) d kd k 3 3 ′ √ k0 k′ 0 2(2π)3 ′ × e−ik·x−ik ·y a(k). a† (k′ ) + eik·x−ik ·y a† (k). a† (k′ ) = 0.

a† (k′ ) = δ3 k − k′ .64) i (2π)3 d3 k d3 k ′ k′ 0 4k0 ′ × e−ik·x−ik ·y a(k). However. a† (k′ ) = ′ d3 xd3 y √ eik·x−ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x). (5. a(k). we have a(k). Any other redefinition will introduce a multiplicative factor into the commutation relations. 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′ ) = 0.65) From these. have commutation relations analogous to the annihilation and creation operators of a harmonic oscillator. a(k′ ) + eik·x+ik ·y a† (k). a† (k′ ) + eik·x−ik ·y a† (k).61). Thus. which are the coefficients of expansion of the field operator in a plane wave basis. a† (k′ ) = iδ3 (x − y). Π(y)]x0 =y0 = − ′ ′ x0 =y 0 (5.178 ′ 5 Free Klein-Gordon field theory − eik·x−ik ·y a† (k). k′ 0 φ(y) − iΠ(y) x0 =y 0 . recalling that the integrals are ime independent. a(k′ ) − e−ik·x+ik ·y a(k).) a(k). a(k′ ) = 0 = a† (k). [φ(x).66) can also be obtained more directly from the inversion formulae in (5. ′ ′ 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. The commutation relations in (5. for example. a† (k′ ) . a(k′ ) − eik·x+ik ·y a† (k).66) This shows that the operators a(k) and a† (k).55).

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

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

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

(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) .69).72) into the expression for the Hamiltonian (5.182 +ei(k 0 −k ′ 0 )x0 5 Free Klein-Gordon field theory δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) + ei(k = 1 2 0 +k ′ 0 )x0 d3 k 0 0 1 e−2ik x a(k)a(−k) + a(k)a† (k) 0 k 0 x0 + a† (k)a(k) + e2ik a† (k)a† (−k) .34). (5.71) and (5.72) Substituting (5.

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

we assume that the quantum theory is defined by the normal ordered Hamiltonian) Ek : a(k)a† (k) + a† (k)a(k) : 2 Ek † a (k)a(k) + a† (k)a(k) 2 d3 k Ek N (k).O. (5. then. we define normal ordering as the ordering in which the creation operators stand to the left of the annihilation operators. (5.184 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. Let us recall that in the passage from classical to a quantum theory. : 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′ ) : (5. (for bosons) N. Thus.O.73) (namely.76) N. = = = d3 k d3 k d3 k Ek a† (k)a(k) = where we have defined the number operator for an oscillator with momentum k as in the case of a simple harmonic oscillator as N (k) = a† (k)a(k). N.O. This is conveniently done by normal ordering of the operators. = a† (k)a† (k′ ) = a† (k′ )a† (k). we have from (5. : a(k)a(k′ ) : = a(k)a(k′ ) = a(k′ )a(k).77) H N.78) and we can define the total number operator for all the oscillators (infinite number of them) in the system as .O.76). Taking advantage of this.O. the ordering of the operators is ambiguous. If we normal order the Hamiltonian using (5. N.

(5. a† (k). a† (k′ ) a(k′ ) = a(k′ )δ3 (k − k′ ). a† (k′ )a(k′ ) = a† (k′ ) a† (k).77) how the normal ordering has redefined away an infinite zero point energy (which basically would arise from commuting a(k)a† (k) to bring it to the normal ordered form). (5. N (k′ ) = = a(k). N ] = = a† (k). a(k′ ) = −a† (k′ )δ3 (k − k′ ). a† (k).5. d3 k′ N (k′ ) d3 k′ −a† (k′ )δ3 (k − k′ ) = −a† (k). It now follows from the definition of the number operators in (5.78) and (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′ ) a(k). Equation (5. d3 k′ N (k′ ) d3 k′ a(k′ )δ3 (k − k′ ) = a(k).80) [a(k).79) that a(k).80) then leads to (5. N (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 = = a(k).

(5. H = −Ek a† (k). (5. H] = Ek a(k).84) This is because both Ek as well as the inner product of states is positive definite (namely.75)) [a(k). the integrand is positive semidefinite.186 5. the inner product represents the norm of the state a(k)|E ) and.85) . which can be achieved by quantizing the system in a box. consequently.82) Let us consider an energy eigenstate |E of this Hamiltonian satisfying H|E = E|E .83) that E = = = ≥ 0. Let us next recall that the operators a(k) and a† (k) satisfy the commutation relations (see (5.77) and (5.77) (as describing the quantum free Klein-Gordon theory) of the form H= d3 k Ek a† (k)a(k). a† (k). 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. It is clear from (5.83) where we are assuming that the state |E is normalized and that the eigenvalue E is discrete for simplicity.6 Energy eigenstates 5 Free Klein-Gordon field theory From now on. (5. let us consider the normal ordered Hamiltonian (5. E|H|E E| d3 k Ek a† (k)a(k)|E d3 k Ek E|a† (k)a(k)|E (5.

89) This is. such that a(k)|Emin = 0. (5. (5. a(k)H|E = Ek a(k)|E . Let us next note that since a(k) acting on an energy eigenstate lowers the energy. therefore. or.84). we note that a† (k). |Emin . Similarly. (5. it would appear that by applying the annihilation operator successively. clearly. H |E or. But as we have seen in (5. Namely. raises the energy value. . The annihilation operator a(k) lowers the energy eigenvalue as it should. H] |E or. = Ek a(k)|E . there must exist a state (energy eigenstate) with a minimum energy. if |E represents an energy eigenstate with eigenvalue E. we had redefined . d3 k Ek Emin |a† (k)a(k)|Emin Emin = = 0.87) − Ha† (k)|E = (E + Ek ) a† (k)|E In other words. we can lower the energy eigenvalue arbitrarily. if |E is an energy eigenstate with energy eigenvalue E. a† (k)H|E H a† (k)|E = −Ek a† (k)|E . the energy eigenvalue is bounded from below (E ≥ 0).6 Energy eigenstates 187 As a result.86) − Ha(k)|E H {a(k)|E } = (E − Ek ) {a(k)|E } . or. then a† (k) acting on it gives another energy eigenstate with the higher value of energy E+Ek . Therefore.5. therefore. the ground state of the system or the vacuum state (recall that by normal ordering the Hamiltonian.88) and we cannot lower the energy any further. we have [a(k). = −Ek a† (k)|E . In such a state. then {a(k)|E } is also an eigenstate of energy with the lower energy value E − Ek . (5. The creation operator a† (k).

(5.188 5 Free Klein-Gordon field theory the energy of the ground state to be zero). k2 .92) This will be an eigenstate of the total number operator satisfying N |n1 . kℓ (5. nℓ . k1 . · · · . = 0 = 0|a† (k).93) = (n1 + n2 + · · · + nℓ ) |n1 . = 0. k2 . n2 . . (5. · · · . kℓ . · · · . k2 . it is clear now that the expectation value of any normal ordered operator in the vacuum state. n2 . conventionally called the vacuum expectation value or vev.) Given the vacuum state. the vacuum state contains no quantum of energy or no particle if energy is related to that of particles. kℓ = a† (k1 ) √ n1 ! n1 a† (k2 ) √ n2 ! n2 ··· a† (kℓ ) √ nℓ ! nℓ |0 . 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 .90) In other words. · · · . (Incidentally. nℓ . nℓ . k1 .91) From our study of the harmonic oscillator we know that such states will be eigenstates of the number operator and we can denote such a state (normalized) as |n1 . = 1. (5. would vanish. kℓ = d3 k N (k)|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 . k1 . k1 . one can build up states of higher energy by simply applying the creation operator. = 0. n2 . k2 . nℓ .

k2 . · · · . Namely. k1 . of course. the state |n1 .6 Energy eigenstates 189 This relation can be easily shown using the identity (see also (5. kℓ (5. kℓ = d3 k Ek N (k)|n1 .) P= then d3 k k a† (k)a(k) = d3 k k N (k). kℓ = (5.5. nℓ . · · · .95) n1 δ3 (k − k1 ) + · · · + nℓ δ3 (k − kℓ ) |n1 . and . (a† (k1 ))n1 = a† (k) a(k). nℓ . k1 .96) if we define the momentum operator as (This is. · · · . k1 . (5. Let us further note that since H= d3 k Ek a† (k)a(k) = d3 k Ek N (k). kℓ . nℓ . (5.97) H|n1 . nℓ . · · · . k2 . nℓ . nℓ . n2 .94) N (k)|n1 . the natural definition from considerations of covariance. n2 . k1 . a† (k1 ) = n1 a† (k1 ) which also leads to n1 n1 δ3 (k − k1 ) . k2 . k1 .80)) N (k). k1 . However. n2 quanta with four momentum k2 and so on. · · · .98) = (n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ ) |n1 . kℓ contains n1 quanta with four µ µ momentum k1 . kℓ . the momentum operator can also be derived from N¨ether’s theorem and o coincides exactly with this in the normal ordered form as we will see later. (5. · · · . n2 .

|k = a† (k)|0 .101) This state satisfies N |k H|k P|k = |k .90)) that these states are orthonormal and define a complete basis of states for the Hilbert space of the Klein-Gordon theory. Namely. k2 .102) It follows from (5. k1 . k2 . kℓ . k1 . = k|k .102) that H 2 − P2 |k 2 = (Ek − k2 )|k = m2 |k . n2 . (5.7 Physical meaning of energy eigenstates To obtain the physical meaning of these energy eigenstates. · · · . these states are eigenstates of H and P with total energy and momentum given by E = n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ . let us analyze the state obtained from the vacuum by applying a single creation operator. (5. kℓ = d3 k k N (k)|n1 .66) as well as the properties (5. 5.99) = (n1 k1 + n2 k2 + · · · + nℓ kℓ ) |n1 . nℓ . (5. k1 .103) . · · · . · · · . nℓ . nℓ . = Ek |k . n2 .100) We can show in a straightforward manner (using the fundamental commutation relations in (5.190 5 Free Klein-Gordon field theory P|n1 . kℓ (5. k2 . p = n1 k1 + n2 k2 + · · · + nℓ kℓ . n2 . (5.

105) Since φ(x) is linear in the creation and the annihilation operators. k1 .5.7 Physical meaning of energy eigenstates 191 This state. field theories naturally describe systems of many identical particles. Note that the particles described by such states are necessarily identical (although with different energy and momentum) and. it is clear that 0|φ(x)|0 = 0.105) as well as the fact that only the field operator φ(x) depends on the space-time coordinates on which the . (5. an ordinary function representing the projection of the state |φ(x) on to the one particle state |k . satisfies the single particle Klein-Gordon equation with positive energy (it provides a representation for the Poincar´ e group) and. (5.106) In fact.108) where we have used (5. n2 . therefore. k2 . as a result. |φ(x) = φ(x)|0 = φ(−) (x)|0 . µ n2 particles with k2 = (Ek2 . kℓ . · · · . Similarly. (5.107) This is. (5. Let us next consider the state that is produced by the field operator acting on the vacuum. we can show that the state |n1 . k). of course.104) µ can be thought of as the state with n1 particles with k1 = (Ek1 . This function satisfies ∂µ ∂ µ + m2 k|φ(x) = k| ∂µ ∂ µ + m2 φ(x)|0 = 0. (5. nℓ . therefore. k2 ) and so on. k1 ). we can think of this as the one particle state with four momentum kµ = (Ek . the only nonzero matrix element of the field operator involving the vacuum state is given by k|φ(x)|0 = k|φ(x) .

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

xn .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.113) which will describe a state with n particles at coordinates x1 .111) |φ(x) (5. · · · . · · · . the absolute square of the amplitude (with x0 = x0 ) 2 1 k. such a state contains a superposition of all possible momentum states (which follows from the field decomposition in (5. These states naturally lead to quantum mechanical probabilities which are non-negative. (5. k′ ) at the coordinates x1 and x2 at a given time. φ (x2 ) . such configuran 1 2 tion states are automatically symmetric as we would expect for a system of identical Bose particles. k) and k′ µ = (Ek′ .109). for example. k′ |φ (x1 ) .56)). which is what we have seen explicitly in (5. (5. x0 = x0 = · · · = x0 ). by construction. As we have seen. The state = φ(−) (x)|0 . k′ |φ(−) (x1 ) φ(−) (x2 ) |0 .114) would give the probability for finding two Klein-Gordon particles with four momenta kµ = (Ek . φ (x2 ) = k.5. positive semi-definite and there is no problem of negative .7 Physical meaning of energy eigenstates 193 Thus. They can also be shown to define a complete basis and describe what is known as the Fock space for the system. φ (xn ) = φ(−) (x1 ) φ(−) (x2 ) · · · φ(−) (xn ) |0 . We can similarly construct multi-particle states in the configuration space of the form |φ (x1 ) . Such probabilities are. Furthermore. Thus. Such states are physically meaningful only if the time coordinates are equal (namely. (5. the wave functional for a single particle with a definite energymomentum has the form ψ(φ(x)) = φ(x)|ψ = φ(x)|k .

117). namely. Even classically we know that we can solve an inhomogeneous equation of the kind (5. Thus. it is invariant under a Lorentz transformation) leading to the fact that the Green’s function is Poincar´ invariant. e . namely. Furthermore. In a realistic theory the field can have self-interactions or can interact with other dynamical fields of the system.117) Here. the Klein-Gordon equation will modify. all the physical states have positive semidefinite energy and that the probabilities are non-negative as they should be. When there are interactions present. For example.14). If we know the Green’s function in (5. the second quantized noninteracting Klein-Gordon theory is free from the problems of the first quantized theory that we had discussed earlier. for the simple case of the Klein-Gordon field interacting with an external source J(x) (which is a c-number function or a classical function). the equation will take the form ∂µ ∂ µ + m2 φ(x) = J(x).115) provided we know the Green’s function of the system.8 Green’s functions Even though we have been examining the free Klein-Gordon field theory so far. which can be obtained from the Lagrangian density m2 2 1 ∂µ φ∂ µ φ − φ + Jφ. the Green’s function G(x−y) satisfies ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). which we will study in the following chapters.116) as the Euler-Lagrange equation (5. this equation is covariant (in fact. eventually we would like to study interactions of the system. the translation invariance of the Green’s function arises because the right-hand side of the equation is invariant under translations. The Green’s function for a given inhomogeneous equation is defined as the solution of the equation with a delta source. for the Klein-Gordon equation. 5.194 5 Free Klein-Gordon field theory probabilities in the second quantized description.115) L= (5. (5. 2 2 (5.

115) can be written as φ(x) = − so that ∂µ ∂ µ + m2 φ(x) = − = − µ d4 y ∂xµ ∂x + m2 G(x − y)J(y) d4 y G(x − y)J(y). 1 1 −k2 + m2 G(k) = − . therefore. (2π)4 and substituting them back into (5. Thus.5.120) d4 k −ik·(x−y) e G(k). . We note here that a homogeneous solution can always be added to the particular solution depending on the system under study (and sometimes in order to implement appropriate boundary conditions).117). or.119) = J(x). 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. or. (5.121) In other words. the Fourier transformation turns the partial differential equation into an algebraic equation which is trivial to solve.8 Green’s functions 195 then the particular solution of the inhomogeneous equation (5.118) d4 y −δ4 (x − y) J(y) (5. − m2 (5. The Green’s function is. 4 (2π) (2π)4 G(k) = k2 1 . defining the Fourier transforms δ4 (x − y) = G(x − y) = 1 (2π)4 d4 k e−ik·(x−y) . (5. we obtain ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y).

this is equivalent to moving the poles infinitesimally into the upper half plane.4.123) which lie on the real axis in the complex k0 -plane as shown in Fig. e Let us note that the integrand of the Green’s function in (5. 5. If we choose a contour of the form shown in Fig.3: Poles of the integrand in the complex k0 -plane. we determine the Green’s function for the Klein-Gordon equation to be d4 k −ik·(x−y) e G(k) (2π)4 d4 k e−ik·(x−y) . (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. Therefore. then.121) back into (5. (k0 )2 = k2 + m2 . this is expressed by writing .122) G(x − y) = = This explicitly shows that the Green’s function is Poincar´ invariant.196 5 Free Klein-Gordon field theory Thus. (2π)4 k2 − m2 (5.122) has poles at k2 − m2 = 0. k0 = ±Ek . or.120). substituting (5. Mathematically. or.3. Im k 0 −Ek Ek Re k 0 Figure 5. 5.

GA (x − y) = lim η→0+ e−ik·(x−y) d4 k . or. for x0 − y 0 > 0.126) For x0 − y 0 < 0. 2 Since in this case the poles in (5. the poles of the integrand occur at k2 − m2 − ik0 η = (k0 )2 − k2 − m2 − ik0 η = 0.124) are in the upper half plane. (5. the integral will vanish. on the other hand.4: Choice of the contour in the complex k0 -plane for the advanced Green’s function. or. iη k − 2 0 2 ≃ k 2 + m2 .124). or. (2π)4 k2 − m2 − ik0 η (5. we have to close the contour in the upper half plane (for the damping of the exponential) in which .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. then we must close the contour in the lower half plane for the damping of the exponential in (5. k0 = ±Ek + iǫ.124) In this case. if we enclose the contour in the lower half plane.125) where we have identified ǫ = η . We note that if x0 − y 0 > 0.5. we conclude that GA (x − y) = 0. Therefore. k0 − iǫ = ±Ek . (5.

we can write (we denote the argument of the Green’s function by x for simplicity) G (x) = A 0. we have d3 k (2π)4 d3 k (2π)4 e−ik·(x−y) (k0 − iǫ + Ek ) (k0 − iǫ − Ek ) GA (x − y) = lim = lim 2πi ǫ→0+ dk0 ǫ→0+ 0 0 e−i(Ek +iǫ)(x −y )+ik·(x−y) 2Ek 0 0 e−i(−Ek +iǫ)(x −y )+ik·(x−y) + −2Ek = = i 2 d3 k eik·(x−y) (2π)3 Ek e−iEk (x 0 −y 0 ) − eiEk (x0 −y0 ) (5. (2π)3 Ek This is known as the advanced Green’s function and has support only in the past light cone. for x0 < 0.5 or equivalently if we push both the poles to the lower half plane. the poles of the integrand in (5.129) In this case. d3 k eik·x (2π)3 Ek for x0 > 0.128) or.129) would appear at . Explicitly. sin Ek x0 . 5. (5. (2π)3 Ek Similarly. then. if we choose the contour in the complex k0 -plane as shown in Fig. GA (x) = θ −x0 d3 k eik·x sin Ek x0 . For x0 − y 0 < 0.198 5 Free Klein-Gordon field theory case we can evaluate the integral by the method of residues.127) d3 k eik·(x−y) sin Ek x0 − y 0 . we can express this mathematically by defining e−ik·x d4 k . 4 k 2 − m2 + ik 0 η (2π) GR (x) = lim η→0+ (5.

129) and. (5. Note that if x0 < 0.131) For x0 > 0. x0 < 0. In this case. (5. k0 = ±Ek − iǫ. on the other hand. 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. since both the 2 poles are in the lower half plane.5: Choice of the contour in the complex k0 -plane for the retarded Green’s function.5. k0 + iǫ ≃ ±Ek . we conclude that GR (x) = 0. k0 + iη 2 2 2 ≃ k2 + m2 = Ek . then the integral will vanish. or.130) where we have again identified ǫ = η . or. k2 − m2 + ik0 η = (k0 )2 − k2 − m2 + ik0 η = 0. therefore. or. if we close the contour in the upper half plane. we have to close the contour in the lower half plane (for the damping of the exponential) and the method of residues gives GR (x) = ǫ→0+ lim d3 k (2π)4 dk0 e−ik·x (k0 + iǫ + Ek ) (k0 + iǫ − Ek ) .

(5.200 = 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 .134) Both the retarded and the advanced Green’s functions satisfy the inhomogeneous equation (5. sin Ek x0 . 2 (5.133) or. for x0 > 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.132) e−i(−Ek −iǫ)x + −2Ek 0 +ik·x The overall negative sign in (5. the difference between GA (x) and GR (x) would define a function which would satisfy the homogeneous (Klein-Gordon) equation and . (The advanced and the retarded Green’s functions have similar form except for their support and the overall sign.135) would also define a Green’s function corresponding to the contour picking up the principal values at the poles. the average of the two.117) with a delta function source.) It is clear from (5. Therefore. we see that we can write 0. GR (x) = −θ x0 d3 k eik·x sin Ek x0 .128) and (5. Thus.133) that GR (x) = GA (−x). 1 GA (x) + GR (x) . − d3 k eik·x (2π)3 Ek GR (x) = for x0 < 0. (2π)3 Ek (5.132) arises because the contour is clockwise in the lower half plane. On the other hand. (5. namely.

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 (5.136) that the Schwinger function is real (G(x))∗ = G(x). This is defined with the contour in the complex k0 -plane as shown in Fig. Quantum mechanically.137) Furthermore.6. namely.8 Green’s functions 201 would not strictly correspond to a Green’s function of the theory.138) (5. which corresponds to pushing one of the poles (on the left) to the upper half plane while moving the other (on the right) .139) The retarded and the advanced Green’s functions play an important role in many classical calculations as well as in many calculations in statistical mechanics. (5. (2π)3 Ek This clearly satisfies the homogeneous Klein-Gordon equation ∂µ ∂ µ + m2 G(x) = 0. G(x) = −G(−x). there is another Green’s function. which is more useful in the calculation of scattering matrix elements.136) + θ x0 = = θ x0 + θ −x0 d3 k eik·x sin Ek x0 . 5. and is antisymmetric. however. known as the Feynman Green’s function.5. (5.

or. this can be implemented by defining d4 k e−ik·x . = ± (Ek − iǫ) . or. there will be a nontrivial contribution independent of whether we close the contour in the upper half or in the lower half plane.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. (5. to the lower half plane. the poles of the integrand in (5.141) η where we have identified ǫ = 2Ek with η → 0+ . Mathematically. (k0 )2 − Ek − k0 = ± Ek − iη 2Ek iη 2Ek 2 ≃ 0. Since there are poles in both halves of the complex k0 -plane. 2 (k0 )2 − Ek + iη = 0.140) occur at k2 − m2 + iη = 0. (2π)4 k2 − m2 + iη GF (x) = lim η→0+ (5.142) . 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.140) In this case. (5. or. in this case.6: Choice of the contour in the complex k0 -plane for the Feynman Green’s function.

where. for x0 > 0.8 Green’s functions 203 Therefore. if x0 > 0.143) = G(−) (x). 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.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. we have used k ↔ −k in the last step (and we understand that k0 = Ek ). Here the contour is clockwise leading to an overall negative sign.145) = −G(+) (x). Therefore. The Feynman Green’s function clearly has support in both the future as well as the past light cones and can be written as . for x0 < 0. On the other hand. we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 (5.5. 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.

145) that i 2 d3 k eik·x = −G(−) (x). However.146) G(+) (−x) = (5.145)) can be written in the manifestly covariant forms d4 k θ(k0 )δ(k2 − m2 ) e−ik·x . (2π)3 k0 (5.149) for which the pole at k0 = −Ek is pushed down to the lower half plane while the pole at k0 = Ek is moved up to the upper half plane. (5.136) as d3 k eik·x sin Ek x0 = G(+) (x) + G(−) (x). We note from the structure of G(−) (x) in (5.150) The functions G(+) (x) and G(−) (x) (see (5. we will not go into this in more detail here. Note from (5. therefore. GF (−x) = GF (x).151) G(+) (x) = i G(−) (x) = −i . namely.148) There is yet another choice of the contour and.143) and G(+) (x) in (5.143) and (5. It is defined as e−ik·x d4 k . (2π)4 k2 − m2 − iη G(x) = lim η→0+ (5. (2π)3 Ek G(x) = (5. (2π)3 (5.204 5 Free Klein-Gordon field theory GF (x) = −θ x0 G(+) (x) + θ −x0 G(−) (x). (2π)3 d4 k θ(−k0 )δ(k2 − m2 ) e−ik·x .143) and (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. the Feynman Green’s function is an even function.145) that we can write the Schwinger function (5.

133) and (5. For completeness.128).58) that φ(x) = φ(+) (x) + φ(−) (x). let us calculate some of the covariant commutation relations satisfied by the field operators.150) that we can write GA (x) = θ(−x0 )G(x) = θ(−x0 ) G(+) (x) + G(−) (x) . (5.153) where ǫ(x) = (θ(x) − θ(−x)) is known as the sign function (or the alternating step function). where we note from (5.151) that the Schwinger function has the manifestly covariant representation G(x) = G(+) (x) + G(−) (x) = i d4 k ǫ(k0 )δ(k2 − m2 ) e−ik·x .155) k2 + m2 > . Let us recall from (5. (5. 5. (5.154) which will be useful in our later discussions.152) and all the Green’s functions can be expressed as linear combinations of these two fundamental Green’s functions.9 Covariant commutation relations 205 and are correspondingly known as the positive and the negative energy (frequency) Green’s functions. we note from (5. (2π)3 (5.9 Covariant commutation relations To make contact between the quantum field theory and the various Green’s functions that we have constructed.5.151)) G(+) (x) ∗ = G(−) (x).57) that (we recall that k0 = Ek = 0) √ (5. We note from (5. GR (x) = −θ(x0 )G(x) = −θ(x0 ) G(+) (x) + G(−) (x) . They satisfy (see (5.

d3 k ′ (2π)3 2k′ 0 e−ik ·y a(k′ ) ′ (5. φ(−) (y) = = 1 (2π)3 d3 k (2π)3 2k0 eik·x a† (k). a† (k′ ) = δ3 k − k′ .156) Since we know the commutation relations (5. d3 k ′ (2π)3 2k′ 0 eik ·y a† (k′ ) ′ d3 k d3 k′ ik·x ik′ ·y † √ √ a (k). φ(−) (y) = d3 k e−ik·x a(k). a† (k′ ) . eik·x a† (k).206 5 Free Klein-Gordon field theory φ(+) (x) = φ(−) (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k). (2π)3 2k0 d3 k ′ ′ eik ·y a† (k′ ) (2π)3 2k′ 0 .157) d3 k d3 k′ −ik·x −ik′ ·y √ √ e e a(k). φ(−) (x). a(k′ ) = 0 = a† (k). it is straightforward to evaluate φ(+) (x). a(k). a† (k′ ) e e 2k0 2k′ 0 = 0. φ(+) (x). a(k). φ(+) (y) = = 1 (2π)3 d3 k (2π)3 2k0 e−ik·x a(k).66) between the creation and the annihilation operators. a(k′ ) 0 ′0 2k 2k = 0. (5.

a† (k′ ) e 2k0 2k′ 0 d3 k d3 k′ −ik·x+ik′·y 3 √ √ e δ k − k′ 2k0 2k′ 0 d3 k −ik·(x−y) e 2k0 (5. φ(y)] = φ(+) (x) + φ(−) (x). furthermore.158) and (5.158) and (5.159).5. (2π)3 Ek (5. φ(+) (y) = −iG(+) (x − y) − iG(−) (x − y) = −iG(x − y) = −i d3 k eik·(x−y) sin Ek x0 − y 0 .161) simply because the field operator φ(x) satisfies the Klein-Gordon equation.159) Note that the field operators in the commutators (5. φ(+) (y) = iG(+) (y − x) = −iG(−) (x − y). using the results in (5.9 Covariant commutation relations 207 = = = 1 (2π)3 1 (2π)3 1 (2π)3 d3 k d3 k′ −ik·x+ik′·y √ √ a(k). we can evaluate [φ(x).160) Namely.40)) and. φ(−) (y) + φ(−) (x). the commutator of two field operators. (5. at unequal times. It is now clear that the Schwinger function satisfies the homogeneous equation µ ∂x µ ∂x + m2 G(x − y) = 0. is proportional to the Schwinger Green’s function. φ(+) (y) + φ(−) (y) = φ(+) (x).158) = −iG(+) (x − y).159) are not restricted to have equal time arguments any more (unlike in (5. . It follows now that φ(−) (x). (5.

163) where Λµν represents a Lorentz transformation.162) These commutation relations are known as covariant commutation relations simply because the (scalar) Green’s functions are invariant under Lorentz transformations.) On the other hand. x0 =y 0 (5.138) and (5. namely. (5. therefore. a light signal cannot connect two space-like points and.160) also shows that two field operators φ(x) and φ(y) do not commute for arbitrary values of the coordinates. Namely. as can be checked even in the simple classical Poisson bracket relations for a one dimensional free particle or a harmonic oscillator.164) which is expected from the anti-symmetry of the Schwinger function (5. for (x − y)2 < 0. (5. Equation (5. 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.40)) [φ(x). G(Λx) = G(x). φ(y)]x0 =y0 = −iG(x − y) = 0. This is known as the principle of microscopic causality. We also note that . two measurements at space-like separations should not influence each other. this relation can also be seen to imply that [φ(x).208 5 Free Klein-Gordon field theory ∂µ ∂ µ + m2 φ(x) = 0. for space-like separations.165) x0 =y 0 We note that by a Lorentz transformation. This is consistent with our intuitive expectation. (5. (5. φ(y)] = 0. the commutator of two field operators vanishes.166) In other words.160) leads to the familiar relation (see (5.

136). (5.10 References 209 ∂G(x − y) ∂y 0 = − x0 =y 0 =− d3 k eik·(x−y) Ek cos Ek x0 − y 0 |x0 =y0 (2π)3 Ek (5. Relativistic Quantum Fields.168) To conclude this section.167) d3 k ik·(x−y) e = −δ3 (x − y). θ(y 0 − x0 ) [φ(x). φ(y) [φ(x). which will be useful later. Drell. (5. let us note from (5. D. Π(y)]x0 =y0 = x0 =y 0 = −i ∂G(x − y) ∂y 0 x0 =y 0 = iδ3 (x − y). (2π)3 This is consistent with our earlier field quantization rule (5.169) .5. φ(y)] = −iθ(y 0 − x0 ) G(+) (x − y) + G(−) (x − y) = −iθ(y 0 − x0 )G(x − y) = −iGA (x − y).10 References 1.154) 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).150) and (5. φ(y)] = −iθ(x0 − y 0 ) G(+) (x − y) + G(−) (x − y) = −iθ(x0 − y 0 )G(x − y) = iGR (x − y). 5. ˙ φ(x). namely. 1964.40). Bjorken and S.160) as well as the relations in (5. McGrawHill. New York. (5. J.

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

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

x′ µ = xµ . (6. as indicated in (6.3). (6. (6.7) . In such a case.4) then. 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. Thus.3) define a symmetry of the system. d4 x L φ′ (x). the Euler-Lagrange equations for the primed and the unprimed systems would remain form invariant. It is clear that in such a case. ∂µ φ(x)) = 0. or. ∂µ φ(x)) = 0. namely. It is worth emphasizing here that (6. invariance of the action under the infinitesimal forms of the transformations in (6. ∂µ φ(x)) = 0. ∂µ φ′ x′ δS = or.5) and only the dynamical variables of the theory transform. Symmetries have interesting consequences for continuous transformations.212 6 Self-interacting scalar field theory S= d4 x L (φ(x).3) includes a very interesting class of transformations where the space-time coordinates do not change. ∂x (6.6) which holds for most global space-time symmetries as well as internal symmetries. − d4 x L (φ(x). 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)) = ′ d4 x′ L(φ′ x′ . the transformations in (6.3) would imply ′ d4 x′ L φ′ x′ . ∂µ φ′ (x) − d4 x L φ′ (x). d4 x L (φ(x). ∂µ φ′ x′ ). let us consider an infinitesimal transformation with ∂x′ = 1. ∂µ φ′ (x) − L (φ(x).

defining the infinitesimal change in the field variable as (this is known as the Lie derivative of the field variable up to a sign) φ′ (x) − φ(x) = δφ(x). as we will see later. ∂µ φ′ (x) − L (φ(x). On the other hand. therefore.8) which must hold independent of the use of equations of motion for the system under study. (6. ∂µ φ(x)) = δφ(x) ∂L ∂L + (∂µ δφ(x)) ∂φ(x) ∂∂µ φ(x) ∂L ∂L + (∂µ δφ(x)) ∂∂µ φ(x) ∂∂µ φ(x) ∂L ∂∂µ φ(x) .¨ 6. ∂µ φ(x)) + δφ(x) −L (φ(x).9) we can calculate explicitly (we remember that δφ(x) is infinitesimal and. This is quite general and it is possible to have symmetry transformations for which K µ = 0.1 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. Clearly. (6. (6. this is indeed the case for internal symmetry transformations.11) ∂L ∂L + δ(∂µ φ(x)) ∂φ(x) ∂∂µ φ(x) = δφ(x)∂µ = ∂µ δφ(x) . ∂µ φ′ (x) − L (φ(x). ∂µ φ(x)) = L (φ(x). so that δ(∂µ φ(x)) = ∂µ φ′ (x) − ∂µ φ(x) = ∂µ δφ(x).10) (6. ∂µ φ(x)) = ∂µ K µ . keep only linear terms in δφ(x)) L φ′ (x).7) will hold if L φ′ (x). (6. In fact.

214 6 Self-interacting scalar field theory Here we have used the Euler-Lagrange equation in the intermediate steps. not all such variations will define a symmetry. on φ′ . However. we obtain ∂µ δφ(x) or. (6. therefore. we note that . (Note that. Second. ∂µ J µ (x) = 0. there is no restriction on δφ and. the conserved current independent of the parameter of transformation is not always a vector.) Comparing (6.11). namely.12) ∂L − Kµ ∂∂µ φ(x) This shows that whenever there is a continuous symmetry associated with a system.13) which is conserved.14) Several comments are in order here. on the other hand.8) and (6. x) . this is always true when the equations of motion are used. ∂∂µ φ(x) J µ (x) = δφ(x) (6. Its tensor structure is determined by the tensor structure of the infinitesimal parameter of transformation. (6. Namely. we can define a charge Q= d3 x J 0 (t.15) which will be a constant (in time) and will generate the symmetry transformations. we can define a current ∂L − K µ. given a conserved current. φ′ must also satisfy the same equation of motion in the transformed frame as the original field. (6. ∂µ δφ(x) ∂L ∂∂µ φ(x) = ∂µ K µ = 0. in general. In fact. For a symmetry. for any infinitesimal variation. First. This is just the principle of least action.

Q corresponds to the generator of the infinitesimal symmetry transformations of the theory.16) Here we have used the fact that.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. In this case. Thus. In the operator language. with the usual assumptions on the asymptotic fall off of the field operators (see (5. . we are dealing with a scalar field. This is known as N¨ther’s theorem and is quite important in the study of symmeo tries in dynamical systems. (6. the integral of a total divergence (∇ · J) vanishes.3)). H] = 0.18) where the parameter of translation ǫµ is assumed to be infinitesimal and constant (global).¨ 6. Since. (6. The converse is also true. Q and H can have simultaneous eigenstates. Namely. 6. As an example of the consequences of N¨ther’s theorem. there exists a conserved charge which is the generator of these infinitesimal symmetry transformations. the time independence of Q also means that it commutes with the Hamiltonian of the theory [Q. (6.1 Space-time translation. consequently. δxµ = x′µ − xµ = ǫµ . let us consider the simple case of an infinitesimal o global space-time translation defined by xµ → x′µ = xµ + ǫµ . then it generates infinitesimal transformations (through commutation relations) which define a symmetry of the theory. if there exists a conserved charge in a given theory.1. or. if we have a continuous symmetry in the theory (transformations under which the action is invariant).

21) = −ǫµ ∂µ φ(x) = −ǫµ ∂µ L = ∂µ K µ . we can now calculate explicitly the change in the Lagrangian density (see (6.19) and in this case we obtain the change in the field to correspond to (see (6. we note that under the transformation (6. where we have used (5.21) that (ǫµ is a constant parameter) K µ = −ǫµ L.22) . then.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). We see that since the change in the Lagrangian density is a total divergence. (6. ∂µ φ′ (x) − L (φ(x).20).216 6 Self-interacting scalar field theory φ′ x′ = φ (x) . We can now identify from (6. the action is invariant under infinitesimal translations which define a symmetry of the system.11) and (6.8)) L φ′ (x). (6.7) and (6. = − φ′ x′ − φ′ (x) = −ǫµ ∂µ φ′ (x) (6.20) where in the last step. Next. ∂µ φ(x)) = δφ(x) ∂L ∂L + (∂ν δφ(x)) ∂φ(x) ∂∂ν φ(x) ∂L ∂L − ǫµ (∂µ ∂ν φ(x)) ∂φ(x) ∂∂ν φ(x) (6. we 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.10). With this.

27) .24) µ Jǫ (x) = δφ = −ǫν (∂ν φ(x)) = −ǫν (∂ ν φ(x)) It is clear. therefore.¨ 6. T µν = T νµ . (6. ∂∂µ φ(x) T µν = (∂ ν φ(x)) (6. ∂∂µ φ(x) ∂∂µ φ(x) (6. that the conserved current independent of the parameter can be identified with ∂L − η µν L.13)) ∂L − Kµ ∂∂µ φ(x) ∂L + ǫµ L ∂∂µ φ(x) ∂L − η µν L ∂∂µ φ(x) = −ǫν T µν . 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.25) which can be easily checked to satisfy (using the equations of motion) ∂µ T µν = 0. even when the naive N¨ther o procedure does not lead to a symmetric stress tensor. depending on the parameter of transformation.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 . (6. (6. namely. to be (see (6.).1 Nother’s theorem 217 δφ(x) ∂L ∂L = −ǫν (∂ν φ(x)) . Therefore.23) so that we obtain the current associated with the symmetry transformation.

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

since the free Lae grangian density is invariant under the discrete transformation φ(x) ↔ −φ(x). Furthermore. ˙ d3 x ∇φ(x)φ(x). (6. 2 2 L0 = (6. 6.97) and it can be checked that P in (6.6. Pi = = P = − d3 x T 00 d3 x d3 x ˙ φ(x) 2 − 2 1 ˙ φ(x) 2 2 m2 2 1 φ + ∇φ · ∇φ + 2 2 1 ˙ φ(x) 2 1 m2 2 + ∇φ · ∇φ + φ 2 2 d3 x T 0i ˙ d3 x ∂ i φ(x)φ(x).34) .34). we note that the Lagrangian density for the interaction must be invariant under Lorentz transformations as well as translations (Poincar´ invariant). We had also given an expression for the momentum operator in terms of creation and annihilation operators in (5.2 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) We see that P 0 = H coincides with the form of the Hamiltonian we had derived earlier in (5.33) To include interactions. (6.32) corresponds exactly to the former expression when normal ordered and expressed in terms of creation and annihilation operators.2 Self-interacting φ4 theory 219 P0 = = = = H.

would lead to a potential (and. (6. for various other reasons we can show that (6.35) The restriction on the coupling constant. λ. the fully self-interacting theory of a real Klein-Gordon field is described by a Lagrangian density L = L0 + LI . the same canonical dimension as . To explain this briefly.35) represents the only meaningful interaction term for such a quantum field theory. (6. let us introduce the concept of canonical dimensions. Let us also g note here that an interaction Lagrangian density of the form − 3! φ3 . in units of = c = 1 (which we have been using).) As a result. is there so that the Hamiltonian will be positive definite. which does not respect the discrete symmetry (6.220 6 Self-interacting scalar field theory we would like to preserve this symmetry in the interactions as well.37) and has. then. In these units. Consequently.34). Let us recall that the action for a quantum mechanical particle has the generic form S = = dt (p q + · · · ) ˙ (p dq + · · · ). a Hamiltonian) which is unbounded from below for any value of the coupling constant g.36) While there are other interaction Lagrangian densities that we can construct consistent with our symmetry requirements. therefore. . Thus. (We will see this shortly. the action is dimensionless. we can show that the canonical dimension of any variable can be expressed in powers of an arbitrary mass dimension [M ]. the simplest interaction Lagrangian density involving only the scalar fields which we can think of has the form λ 4 φ (x). (6. which in turn allows us to define a vacuum state of the theory. in fact. With these conditions. 4! LI = − λ > 0.

the Klein-Gordon field variable in four dimensions has a canonical dimension 1.40) Thus. From the canonical dimension of the interaction term in the Lagrangian density in (6. (6.39) In other words. d4 x = [M ]0 . [∂µ ] = [M ]. we conclude that the coupling constant or the interaction strength for the φ4 -self-interaction in (6. in these units.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.2 Self-interacting φ4 theory 221 [L] = [T ] = [xµ ] = [M ]−1 . 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. (The dimensionality of the field variable depends on the number of space-time dimensions which we will see later. d4 x d4 x [λ] ([φ])4 = [M ]0 . (6. [λ] = [M ]0 . or. has a canonical dimension 1 and. ([φ])2 = [M ]2 . (6. [φ] = [M ].) The mass parameter. d4 x [∂µ ] [φ] [∂ µ ] [φ] = [M ]0 . therefore.35). or.35) is dimensionless. let us note that for a monomial interaction action of the form .6. or. In general. [M ]−4 [λ] [M ]4 = [M ]0 . 4! [SI ] = − or. of course. or. [M ]−4 [M ][φ][M ][φ] = [M ]0 . or. we obtain λ 4 φ = [M ]0 .

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

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 ground state of the free Hamiltonian will not be stable under perturbations. There is. As a result. of course. This is why we restrict to λ > 0 in the interacting theory. 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. therefore. Rather. Our discussion so far has been within the context of the Heisenberg picture where the field operator φ(x. t) carries time dependence. This leads to different possible pictures for describing the same quantum mechanical system through distinct time evolutions.43) H = = d3 x H d3 x m2 2 λ 4 1 2 1 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! (6. 6.44) is positive definite and. It is now clear that for λ > 0.44) = H0 + HI . as we know. the Hamiltonian will be bounded from below leading to a meaningful vacuum state. neither the states in the Hilbert space nor the operators acting on state vectors are observables. In this picture. the observables correspond to expectation values of Hermitian operators in quantum states. 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.3 Interaction picture and time evolution operator In quantum mechanics as well as in quantum field theory. the Hamiltonian is indefinite because of the interaction term. = which leads to (6. For λ < 0. also the . on the other hand.6. Consequently.3 Interaction picture and time evolution operator 223 1 2 1 m2 2 λ 4 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! = H0 + HI .

45) that (H) ψ|O(H) (t)|ψ (H) = (S) ψ(t)|O(S) |ψ(t) (S) . 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. 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. Thus. t) eiHt = φ(H) (x. (6.45). It follows from (6. In the Schr¨dinger picture. the Klein-Gordon field o operator will be given by φ(S) (x) = e−iHt φ(H) (x. 0).48) The total Hamiltonian operator in the Schr¨dinger picture coino cides with that in the Heisenberg picture (see. (6. (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. t0 ) = e−iH(t−t0 ) . namely.46) In writing (6. (6.47) O(S) = O(H) (0). (6.49) . time independent. we have assumed that at t = 0 both the pictures coincide. (6. for example. in the Schr¨dinger picture. |ψ(0) (S) = |ψ (H) . The two pictures are related by the unitary transformation = e−iHt |ψ |ψ(t) (S) (H) .45) O(S) = e−iHt O(H) (t)eiHt . for example.45)) and is.

the Schr¨dinger picture is not very desirable for relativiso tic theories since the field operators in the Schr¨dinger picture are o not manifestly Lorentz covariant. O(IP ) (0) = O(S) = O(H) (0). (6. 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) .51) In the interaction picture. in some sense. O(IP ) (t) = eiH0 = eiH0 (S) O(S) e−iH0 e t (S) (S) t −iHt O(H) (t)eiHt e−iH0 . we define |ψ(t) (IP ) = eiH0 (S) t t |ψ(t) (S) = eiH0 (S) (S) t −iHt e |ψ t (H) . the more convenient description goes under the name of interaction picture which is.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. both the operators as well as the state vectors carry time dependence. therefore. (However.) Although the Heisenberg picture has a manifestly covariant description.3 Interaction picture and time evolution operator 225 However.50) Note. this picture is useful. in relativistic interacting field theories. o Here. (IP H0 ) (t) = eiH0 t (S) H0 e−iH0 (S) t (S) = H0 . it is worth emphasizing here that for certain applications. (6. by definition. once again. In this case. that we can identify |ψ(0) (IP ) = |ψ(0) (S) (S) = |ψ (H) .6. (IP ) (IP ) (6. intermediate between the Schr¨dinger and the Heisenberg pictures. .

In this picture.226 6 Self-interacting scalar field theory (IP HI ) (t) = eiH0 (S) t (S) HI e−iH0 (S) t . let us drop the superscript (IP ) with the understanding that we are nonetheless working in the interaction picture.54) In other words. the time evolution of the state vectors. From now on. t0 ) |ψ (t0 ) .53) Similarly. 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). (6.50) to have the explicit form U (t. This shows that once we define commutation relations for the operators in the free theory.49) and (6.55) where the time evolution operator in the interaction picture in (6. it is easy to show that the time evolution operator satisfies the properties. is governed by the Schr¨dinger equation with the interaction o Hamiltonian playing the role of the Hamiltonian.51) that the free Hamiltonian in the interaction picture is time independent and. t0 ) = eiH0 (S) t −iH(t−t0 ) −iH0 e e (S) t0 .55) can be seen from (6. We note from (6.56) then. if we define the time evolution operator through the relation |ψ(t) = U (t. therefore. i (6. H0 ) . the field operators can have a plane wave expansion. (6. H0 = 1 (IP ) (IP O (t). (6. . while the dynamical evolution of the operators is governed by the free Hamiltonian through the Heisenberg equations of motion. in this picture. in the interaction picture they continue to hold even in the presence of interactions.

) We can now solve (6. t0 ) t0 ½−i ½−i t dt1 HI (t1 ) t0 t t0 ½−i t1 dt2 HI (t2 ) U (t2 . t0 ) = U (t2 . it follows that the total Hamiltonian is time independent in the interaction picture. to obtain t U (t.59) and note that .57) Furthermore.55) satisfies the equation (this also follows from (6. ∂t ∂U (t. U † (t.59) To bring this to a more convenient form. t0 ) t0 t t1 dt1 HI (t1 ) + (−i)2 dt1 t0 t0 dt2 HI (t1 ) HI (t2 ) +··· + ··· +(−i)n +··· .52) we see that the time evolution operator in (6. t0 ) . from (6. let us look at the second order term in (6. t) = ½.3 Interaction picture and time evolution operator 227 U (t. t) . t1 ) U (t1 .6. U (t2 .51)). t0 ) = HI (t)U (t. t0 ) . t0 ) = U −1 (t. (6. (6.53) that the interaction Hamiltonian in the interaction picture is time dependent.56) with the use of (6.57). t0 ) = U (t0 . t0 ) = ½ − i = = dt1 HI (t1 ) U (t1 .58) iteratively subject to the initial condition in (6. 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 (6.58) We note from (6.50)) ∂|ψ(t) = HI (t)|ψ(t) . i ∂t i or.

we can show that the n-th term in the series in (6. (6. In a similar manner.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 )) .61) with the operator at later time standing to the left of the operator at earlier time.60) where T denotes the time ordering operator and is conventionally defined for two bosonic operators as T (A(t1 )B(t2 )) = θ(t1 − t2 )A(t1 )B(t2 ) + θ(t2 − t1 )B(t2 )A(t1 ).59) can be written as . t0 (6.

the iterative solution for the time evolution operator in (6. (6.59) takes the form t U (t.64) ..63). (6. This can clearly be implemented by assuming that the interaction switches off adiabatically at t = ±∞ (adibatic hypothesis). η→0+ η→0+ (6. (This is a formal definition which is to be understood in the sense of the expansion described in (6. t0 ) = ½ − i + (−i)2 T 2! dt1 HI (t1 ) t0 t t0 t dt1 dt2 HI (t1 ) HI (t2 ) t0 t t (−i)n +··· + T n! +··· or. in the scattering of particles in relativistic quantum field theory. For example. we can implement this by modifying the interaction Hamiltonian as (η) HI (t) → lim HI (t) = lim e−η|t| HI (t). t0 ) = T e t0 dt′ HI (t′ ) .4 S-matrix In the non-relativistic quantum mechanical scattering problems. Similarly. we also assume that the incoming particles at t = −∞ as well as the outgoing particles at t = ∞ are described by free particle states.63) In other words.6.4 S-matrix 229 (−i)n n! t t0 t ··· t0 dt1 · · · dtn T (HI (t1 ) · · · HI (tn )) . −i Rt . we assume that the initial and the final states are plane wave states corresponding to free particles.. in the interaction picture the time evolution operator is given by the time ordered exponential involving the integral of only the interaction Hamiltonian.) 6. t0 t0 dt1 · · · dtn HI (t1 ) · · · HI (tn ) U (t.62) Therefore.

the S-matrix (or the scattering matrix) of the theory can be identified with the time evolution operator (6.55) to be |ψi (∞) = U (∞. (6. (6.68) Consequently. (6.65) Then.64) can be thought of as the relevant boundary condition for the scattering problem under study. the interaction vanishes.64)) leads to the same result for the rate of transition in the physical scattering of particles.66) where we have identified S = U (∞. the state into which this will evolve at t = ∞ is defined from (6. is obtained from (6. −∞)|i = S|i . the probability amplitude for an initial state |i to be in the final free particle state |f at t = ∞. The condition (6. we can take the initial and the final states to be the eigenstates of the free Hamiltonian which we know to be free particle energy-momentum states.67) Therefore. (6. which is the definition of the scattering amplitude. −∞).66) to be Sf i = f |ψi (∞) = f |U (∞. We are essentially assuming here that any smooth function describing the adiabatic switching off of the interaction (and not just the specific form in (6.230 6 Self-interacting scalar field theory so that in the infinite past as well as in the infinite future. −∞)|ψi (−∞) = U (∞. Let us denote the initial state at infinite past as the free particle state |ψi (−∞) = |i .67) which has . −∞)|i = f |S|i . As a result.

These states can be constructed by noting that the three pictures (Heisenberg.69) Furthermore.69) through the appropriate boundary conditions. in general. say.6.4 S-matrix 231 the explicit perturbative expansion of the form (see (6. This also leads to the notion of “in” and “out” states which are quite important in a formal description of scattering theory. (6.64) provides a regularized meaning to the oscillatory terms in (6. −∞) = lim T e η→0+ −∞ dt HI (t) (η)   = ½ − lim+ i η→0 ∞ (η) dt HI (t) −∞ ∞ −∞ (η) (η) dtdt′ T HI (t)HI t′ (−i)2 + lim 2! η→0+ +··· . Let us see this explicitly in the case of the linear term in the expansion of the time evolution operator in (6. not well defined if we do not use an adiabatic interaction of the form.70) where the time evolution operator U (0. (6. In this case.63). they are uniquely given by the states (in any picture) at t = 0.63))  −i ∞ R S = U (∞.70) will have the form . the S-matrix (the scattering matrix) is unitary since the time evolution operator is (see (6.64).51)) and o since the Heisenberg states are time independent. Schr¨dinger and interaction) coincide at t = 0 (see (6.57)). −∞) is. this state would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (in) = U (0. Let us suppose that at t → −∞ we have a free incoming state denoted as |Ψ(−∞) (in) = |i . It is clear from the expansion above that the adiabatic switching of the interaction in (6. Then. the linear contribution to the right-hand side of (6. in (6. These are asymptotic free states as t → −∞ and t → ∞ respectively. −∞)|Ψ(−∞) (in) .

71) is not defined at the lower limit in the absence of the adiabatic factor eηt . we have used the fact that H0 ) = (S) H0 and that |Ψ(−∞) (in) = |i is a free state with Ei the energy (IP eigenvalue of H0 ) .70) and (6. the “in” and the “out” states define a complete space of states at t → −∞ and t → ∞ respectively.232 6 Self-interacting scalar field theory 0 η→0+ lim −i = = = (η) dt HI (t)|Ψ(−∞) (in) −∞ 0 η→0+ lim −i lim −i lim dt eηt eiH0 (S) t (S) HI e−iH0 (S) t −∞ 0 |i η→0+ dt e−i(Ei −H0 (S) +iη)t −∞ (S) HI |i η→0+ 1 (S) HI |Ψ(−∞) (S) Ei − H0 + iη (in) .64) naturally provides a regularization for the formal definition of the time evolution operator U (0. Furthermore. they are not identical because their definitions use different regularizing factors.70) and (6. from their definitions in (6. ∞)|Ψ(∞) (out) .71) would show that in this case. we have a free outgoing state |Ψ(∞) (out) . From (6. Rather. then this would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (out) = U (0. (6. This clarifies how the boundary condition (6.72) we see that even though both the “in” and the “out” states are related to the same Heisenberg state. if we assume that as t → ∞. ∞) would not be well defined at the upper limit.71) (IP where in the intermediate steps. This shows that the asymptotically free “in” and “out” states can be defined in a unique manner by relating them to the Heisenberg state. (6. −∞). In a parallel manner.72) and an analysis as in (6. We also note that the integral in (6. the regularizing factor would have to be e−ηt without which U (0.72) we see that they are related to each other as .

This shows that the “in” and the “out” states are related by the S-matrix in (6. such a calculation will involve time ordered products of normal ordered products and this is where the Wick’s theorem comes in handy. simply corresponds to arranging factors in the product so that the creation operators stand to the left of the annihilation operators. this is equivalent to saying that the negative energy (frequency) parts of the field operator stand to the left of the positive energy (frequency) parts (recall that the positive energy part of the field operator. ∞)|Ψ (H) = U (∞. we will simplify the time ordered product of factors of normal ordered terms. Furthermore. Correspondingly. First. It is clear from this discussion that in calculating the S-matrix elements. as we have defined earlier in (5. (6. Sf i .69) corresponds to no scattering at all. we need to evaluate the matrix elements of time ordered products of operators between free particle states.57). In terms of field operators. The first term in the expansion of the S-matrix in (6. therefore. −∞)|Ψ(−∞) (in) = S|Ψ(−∞) (in) .67) (and.74) where the T-matrix represents the true nontrivial effects of scattering. 0)U (0. −∞)|Ψ(−∞) = U (∞. we can define S − ½ = T. by a phase since the S-matrix is unitary).6. 0)|Ψ (in) (H) = U (∞. contains the annihilation operator while . we will derive a simple relation for the product of factors each of which is normal ordered and then using this relation. 6. φ(+) . since our (interaction) Hamiltonian is assumed to be normal ordered. Later we will see that the Feynman rules are a wonderful and simple way of systematizing these results.76).5 Normal ordered product and Wick’s theorem 233 |Ψ(∞) (out) = U † (0. The calculation can be carried out in two steps. (6.73) where we have used the properties of the time evolution operator given in (6.5 Normal ordered product and Wick’s theorem The normal ordered product.

we have N (φ(x)) ≡ : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x).75) We note that. Furthermore. we note from (6.234 6 Self-interacting scalar field theory the negative energy part.75).76) that . For example. (6. we can write N (φ(x)φ(y)) ≡ : φ(x)φ(y) : = = : φ(+) (x) + φ(−) (x) φ(+) (y) + φ(−) (y) : : φ(+) (x)φ(+) (y) + φ(+) (x)φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) : = φ(+) (x)φ(+) (y) + φ(−) (y)φ(+) (x) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y). Thus. that the vacuum expectation value of any normal ordered product vanishes. as far as bosonic operators are concerned.76)). φ(−) .77) It is clear. 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). N φ(+) (x)φ(+) (y) ≡ : φ(+) (x)φ(+) (y) : = φ(+) (x)φ(+) (y). therefore. N φ(+) (x)φ(−) (y) ≡ : φ(+) (x)φ(−) (y) : = φ(−) (y)φ(+) (x). by definition. using the relations in (6. the order of the factors inside a normal ordered product is not important. (6.76) Since the vacuum state is. N φ(−) (x)φ(+) (y) ≡ : φ(−) (x)φ(+) (y) : = φ(−) (x)φ(+) (y). for the bosonic scalar field. has the creation operator so that this prescription is equivalent to the definition of normal ordering in (5. annihilated by the annihilation operator. (6.

the calculation of S-matrix elements will simplify enormously. (6.79) where we have used (5. (6. This gives another reason to choose the prescription of normal ordering for physical observables. does not carry any nontrivial quantum number.6. this also shows that if we can rewrite any product of operators (including normal ordered factors) in terms of normal ordered terms. therefore. To see how we can write a product of operators in normal ordered form.80) with the pairing of the two field operators defined to be related to the invariant positive energy Green’s function in (5. 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).158) and (6. Let us denote (6. normal ordering is quite useful since we would like the vacuum expectation value of any observable to vanish simply because the vacuum contains no particles – it is empty and.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. φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) = : φ(x)φ(y) : −iG(+) (x − y).79) as φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y).81) . (6.151) as φ(x)φ(y) = −iG(+) (x − y). in the simple example of the product of two field operators. Furthermore. let us note that.78) Therefore. (6.76).

86) (6.85) Thus adding the two terms in (6. φ(−) (y) = = : φ(x)φ(−) (y) : + φ(+) (x). but since it is a c-number function. Clearly. : 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. (6. Therefore. the second term in (6. On the other hand.84) (6.85) and (6. (6.80) that we can also identify φ(x)φ(y) = 0|φ(x)φ(y)|0 = −iG(+) (x − y).80) is zero.84) gives : φ(x) : φ(−) (y) = φ(x)φ(−) (y) = φ(−) (y)φ(x) + φ(x). Since we choose the Hamiltonian to be normal ordered.82) because the vacuum expectation value of the normal ordered product in (6. this will be necessary in developing a perturbation expansion for the quantum field theory where the interaction Hamiltonian is normal ordered.83) We recognize that the first term in (6. (6. let us next see how we can combine a product of normal ordered factors into normal ordered terms. φ(−) (y) : φ(x)φ(−) (y) : − iG(+) (x − y).236 6 Self-interacting scalar field theory We note that G(+) (x − y) is not a symmetric function (so that the pairing should be read carefully as from left to right). we note from (6.86) we obtain . We know that by definition : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x).84) simply corresponds to a normal ordered product : φ(x) : φ(+) (y) = φ(x)φ(+) (y) = : φ(x)φ(+) (y) : .

6. φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + : φ(x)φ(y)φ(−) (z) : + φ(x)φ(y) + iG(+) (x − y) . namely. (6.87) 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) : .88) φ(x)φ(y) = = but (6. (6. : φ(x)φ(y) : −iG(+) (x − y) : φ(x)φ(y) : + φ(x)φ(y). of course. (6. φ(−) (z) = : φ(x)φ(y)φ(z) : + φ(+) (x). let us note that . what we had seen earlier in (6. φ(−) (z) φ(y) +φ(x) φ(+) (y). Next.89) Finally. φ(−) (z) = : φ(x)φ(y)φ(z) : − iG(+) (x − z)φ(y) − iG(+) (y − z)φ(x) = : φ(x)φ(y)φ(z) : + φ(x)φ(z) : φ(y) : + φ(y)φ(z) : φ(x) : .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).79) and (6.87) shows how a product of two simple normal ordered factors can be expressed as a normal ordered term and a pairing.80).

φ(−) (z)]φ(+) (w) + φ(−) (w) : φ(x)φ(y) : φ(+) (z) +[: φ(x)φ(y) :. φ(−) (z)]φ(w) + [: φ(x)φ(y) :.238 6 Self-interacting scalar field theory : φ(x)φ(y) : : φ(z)φ(w) : = : φ(x)φ(y) : φ(+) (z)φ(+) (w)+ : φ(x)φ(y) : φ(−) (z)φ(+) (w) + : φ(x)φ(y) : φ(−) (w)φ(+) (z)+ : φ(x)φ(y) : φ(−) (z)φ(−) (w) = : φ(x)φ(y)φ(+) (z)φ(+) (w) : +φ(−) (z) : φ(x)φ(y) : φ(+) (w) +[: φ(x)φ(y) :. φ(−) (w)]φ(+) (z) + φ(−) (z)φ(−) (w) : φ(x)φ(y) : +[: 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 . (6. φ(−) (z)φ(−) (w)] = : φ(x)φ(y)φ(+) (z)φ(+) (w) : + : φ(x)φ(y)φ(−) (z)φ(+) (w) : + : φ(x)φ(y)φ(+) (z)φ(−) (w) : + : φ(x)φ(y)φ(−) (z)φ(−) (w) : +[: φ(x)φ(y) :. φ(−) (w)]φ(+) (z) +φ(−) (z)[: φ(x)φ(y) :. φ(−) (w)] = : φ(x)φ(y)φ(z)φ(w) : −iG(+) (x − z)φ(y)φ(w) −iG(+) (y − w)φ(x)φ(+) (z) − iG(+) (x − w)φ(−) (z)φ(y) −iG(+) (y − w)φ(−) (z)φ(x) = : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(y)φ(w) + φ(y)φ(z) : φ(x)φ(w) : + φ(x)φ(w) + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(w) : φ(x)φ(z) : = : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(z) : φ(x)φ(w) : + φ(y)φ(w) : φ(x)φ(z) : + φ(x)φ(z) φ(y)φ(w) + φ(x)φ(w) φ(y)φ(z).

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.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). 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) : . we already know from (6. This now leads us to an important theorem known as the Wick’s theorem (this is one form of it) which says that a product of factors of normal ordered field operators can be written as a sum of terms with normal ordered terms and all possible pairings between field operators except for pairings between operators within the same normal ordered factor. therefore.80) that φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y).6. It follows. that (6. For example. (6.92) Similarly.

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.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.93) In other words.89) that . For example. (6.240 6 Self-interacting scalar field theory + φ(y)φ(w) : φ(x)φ(z) : + φ(x)φ(y) : φ(z)φ(w) : + φ(z)φ(w) : φ(x)φ(y) : + φ(x)φ(y) φ(z)φ(w) + φ(x)φ(z) φ(y)φ(w) + φ(x)φ(w) φ(y)φ(z). we have already noted in (6. In general. (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. 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 ) : + ··· . 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).) An immediate consequence of this result is that the vacuum expectation value of an odd number of field operators must vanish (since not all field operators can be paired).

6.6 Time ordered products and Wick’s theorem

241

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

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

: φ(x)φ(y) : φ(z)φ(w) =: φ(x)φ(y) :

: φ(z)φ(w) : + φ(z)φ(w)

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

T (φ(x)φ(y)) = T (φ(y)φ(x)) =

(6.97)

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

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

(6.98)

242

6 Self-interacting scalar field theory

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

θ(x) =

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

(6.99)

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

θ(x) = lim −
→0+

dk e−ikx . 2πi k + i

(6.100)

Of course, by definition

T (φ(x)) = φ(x) =: φ(x) : .

(6.101)

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

6.6 Time ordered products and Wick’s theorem

243

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

+θ x0 − z 0 θ z 0 − y 0 φ(x)φ(z)φ(y)

+θ y 0 − z 0 θ z 0 − x0 φ(y)φ(z)φ(x)

+θ y 0 − x0 θ x0 − z 0 φ(y)φ(x)φ(z)

+θ z 0 − x0 θ x0 − y 0 φ(z)φ(x)φ(y)

+θ z 0 − y 0 θ y 0 − x0 φ(z)φ(y)φ(x),

(6.102)

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

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

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

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

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

244

6 Self-interacting scalar field theory

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

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

(6.105)

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

φ(x)φ(y) = iGF (x − y).

(6.106)

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

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

(6.107)

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

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

(6.108)

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

6.6 Time ordered products and Wick’s theorem

245

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

φ(x)φ(y) = iGF (x − y) = φ(y)φ(x),

(6.110)

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

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

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

246

6 Self-interacting scalar field theory

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

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

6.7 Spectral representation and dispersion relation

247

0| [φ(x), φ(y)] |0

= −iG(x − y) =

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

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

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

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

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

= i
n

−eikn ·(x−y) 0|B(0)|n n|A(0)|0

,

(6.116)

248

6 Self-interacting scalar field theory

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

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

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

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

=

i (2π)3

n

= where

i (2π)3

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

(6.118)

ρ(q) = (2π)3
n

δ4 (q − kn )| 0|A(0)|0 |2 ,

(6.119)

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

6.7 Spectral representation and dispersion relation

249

GAA (x − y) = = =

i (2π)3

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

d4 q ρ(q 2 )
∞ 0

dσ 2 ρ(σ 2 ) i

=
0

dσ 2 ρ(σ 2 ) G(x − y, σ 2 ),

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

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

f (t) =

dk0 −ik0 t e f (k0 ). 2π

(6.123)

Using the integral representation of the step function

250

6 Self-interacting scalar field theory

θ(t) = lim −
ǫ→0+

1 2πi

∞ −∞

dk0

e−ik0 t , k0 + iǫ

(6.124)

from (6.122) we obtain 1 2πi
∞ −∞

f (t) =

ǫ→0+

lim

dk0

e−ik0 t k0 + iǫ

∞ −∞

′ dk0 −ik0 t ′ ′ e f (k0 ) 2π

=

ǫ→0+

lim −

−∞ ∞

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

=

ǫ→0+

lim − lim

−∞ ∞

=

ǫ→0+

−∞

dk0 −ik0 t e 2π

∞ −∞

′ ′ f (k0 ) dk0 ′ − k − iǫ . 2πi k0 0

(6.125)

Comparing (6.125)with (6.123), we conclude that the Fourier transform of a retarded function must satisfy a relation 1 2πi
∞ −∞ ′ dk0 ′ f (k0 ) , − k0 − iǫ

f (k0 ) = lim

ǫ→0+

′ k0

(6.126)

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

ǫ→0+

where the first term on the right-hand side represents the principal part, the dispersion relation (6.126) can also be written as 1 πi
∞ −∞ ′ dk0 ′ f (k0 ) , ′ k0 − k0

f (k0 ) =

(6.128)

6.7 Spectral representation and dispersion relation

251

where the principal value for the pole is understood. Comparing the real and the imaginary parts in (6.128) we obtain 1 π 1 π
∞ −∞ ′ Im f (k0 ) , ′ k0 − k0

Re f (k0 ) =

′ dk0

Im f (k0 ) = −

−∞

′ dk0

′ Re f (k0 ) . ′ k0 − k0

(6.129)

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

f (k0 ) = Re f (k0 ) + iIm f (k0 ) = = 1 π 1 π
∞ −∞ ∞ ′ dk0 ′ dk0 ′ Im f (k0 ) ′ ′ + iπδ(k0 − k0 )Im f (k0 ) ′ k0 − k0

−∞

′ Im f (k0 ) , ′ k0 − k0 − iǫ

(6.130)

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

252

6 Self-interacting scalar field theory

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

|k0 |→∞

This is quite interesting for it says that, as long as only the imaginary part (or the real part in an alternate formulation) of the Fourier transform of a retarded function vanishes for large values of the argument, the dispersion relation (6.130) would hold. This can be contrasted with the vanishing of an analytic function for large values of its argument (which is what we use in complex analysis) which would imply that both the real and the imaginary parts of the function have to vanish in this regime. If Im f (k0 ) does not fall off fast enough, then we can define a subtracted dispersion relation that will ¯ be true and this is done as follows. Let k0 denote a fixed point in the complex k0 plane and let us define a new function ¯ f (k0 ) − f (k0 ) , (6.132) ¯ k0 − k0 Clearly, the imaginary part of this function will vanish for large values of its argument provided Im f (k0 ) does not grow for large values of its argument. Therefore, for this function we can write a dispersion relation of the form (6.130) F (k0 ) = F (k0 ) = = = = lim 1 π 1 π 1 π
∞ −∞ ∞ ′ dk0 ′ Im F (k0 ) ′ k0 − k0 − iǫ

ǫ→0+

ǫ→0+

lim

−∞ ∞ −∞

′ dk0

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

ǫ→0+

lim

′ dk0

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

ǫ→0+

lim

1 ¯ π(k0 − k0 )
′ k0

×

1 1 . − ′ ¯ − k0 − iǫ k0 − k0 − iǫ

(6.133)

6.7 Spectral representation and dispersion relation

253

Using the definition in (6.132), we can also write this relation as 1 π
′ k0 ∞ −∞

¯ f (k0 ) − f (k0 ) =

ǫ→0+

lim ×

′ ′ ¯ dk0 Im (f (k0 ) − f (k0 ))

1 1 − ′ ¯ . − k0 − iǫ k0 − k0 − iǫ

(6.134)

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

Im f (k0 ) = −Im (−k0 ).

(6.135)

¯ If this is true and if we choose the point of subtraction k0 = 0, then the relation (6.134) takes the form 1 π
∞ −∞ ′ ′ Im f (k0 ) Im f (k0 ) − ′ . ′ k0 − k0 − iǫ k0 − iǫ (6.136)

f (k0 ) − f (0) = lim

ǫ→0+

′ dk0

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

η→0+

(6.137)

This has the imaginary part given by

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

(6.138)

254

6 Self-interacting scalar field theory

We note that the imaginary part, in this case, satisfies (6.135). Furthermore, it vanishes for large values of k0 and, therefore, we expect (6.130) to hold in this case. Explicitly we note that
∞ −∞ ′ Im GR (k0 , k) ′ k0 − k0 − iǫ ′ dk0

ǫ→0+

lim = = = =

1 π

′ dk0

ǫ→0+

lim − lim −

π 2πEk

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

ǫ→0+

ǫ→0+

lim

1 2 (k0 + iǫ)2 − Ek 1
2 k0

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

η→0+

lim

2 Ek

+ ik0 η

,

(6.139)

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

6.8 References

255

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

This page intentionally left blank .

Chapter 7 Complex scalar field theory 7. 2 (7. the dynamical equations take the forms ∂µ ∂ µ + m2 φ(x) = 0. 2 1 √ (φ1 (x) − iφ2 (x)) . 2 (7. there are now two distinct equations. The relations in (7.1 Quantization The natural field theoretic example to consider next is that of a free Klein-Gordon field theory where the field operator is not Hermitian. 2 257 . φ† (x) = φ(x). ∂µ ∂ µ + m2 φ† (x) = 0. The physical meaning of this system can be understood by noting that we can express the complex field operator in terms of two real (Hermitian) field operators as 1 √ (φ1 (x) + iφ2 (x)) . In this case. (7.3) φ1 (x) = i φ2 (x) = − √ φ(x) − φ† (x) . spin zero (scalar) fields which are Hermitian.2) φ(x) = φ† (x) = where φ1 (x) and φ2 (x) are two distinct.1) Therefore.2) can be inverted to give 1 √ φ(x) + φ† (x) .

5) L = = = ∂µ φ† ∂ µ φ − m2 φ† φ. (7. ˙ ∂ φi (x) i = 1. then we can define the momenta conjugate to the two field operators as Πi (x) = ∂L ˙ = φi (x). Πj (y)]x0 =y0 . such a system equivalently describes two free. [φi (x). independent spin zero Klein-Gordon fields (and. If we treat φ1 (x) and φ2 (x) as the independent variables. (7. ∂µ ∂ µ + m2 φ2 (x) = 0. φj (y)]x0 =y0 = 0 = [Πi (x).6) Consequently.258 7 Complex scalar field theory It follows now that we can express the dynamical equations in (7.5) because distinct degrees of freedom are expected to commute. and is manifestly Hermitian. 2. the equal time canonical commutation relations follow to be [φi (x). therefore. two distinct particles) with degenerate mass.1) or (7. (7.7) . Πj (y)]x0 =y0 = iδij δ3 (x − y). Therefore.) We can quantize this theory in one of the two equivalent ways.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. The Lagrangian density which will give the dynamical equations in (7.4) Namely.1) in terms of the real field variables as ∂µ ∂ µ + m2 φ1 (x) = 0. 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. (Note that we have not been careful about the order of the factors in rearranging terms in (7.

7. (7.9) Alternatively. 2 2 2 i (7. then we can define the momenta conjugate to the field operators from (7. if we treat φ(x) and φ† (x) as the independent dynamical variables.5) to be Π(x) = = ∂L ˙ = φ(x) ˙ ∂ φ† (x) 1 1 ˙ ˙ √ φ1 (x) + iφ2 (x) = √ (Π1 (x) + iΠ2 (x)) . 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 . We see that the Hamiltonian density of the free complex scalar field theory is simply the sum of the Hamiltonian densities for two non-interacting real Klein-Gordon field theories with degenerate mass and H= d3 x H.8) where we have used the identification in (7.6).1 Quantization 259 In this case. 2 2 .

we can expand the field φ(x) in the basis of plane wave solutions of the Klein-Gordon equation (see (5. Π† (y) = φ† (x). we can derive the Hamiltonian density in this formulation to have the form ˙ ˙ H = Πφ† + Π† φ − L ˙ ˙ = ΠΠ† + Π† Π − φ† φ + ∇φ† · ∇φ + m2 φ† φ = 2Π† Π − Π† Π + ∇φ† · ∇φ + m2 φ† φ (7.12) = Π† Π + ∇φ† · ∇φ + m2 φ† φ. in this case.8) and H= d3 x H. is manifestly Hermitian as it should be even though the dynamical variables are not. of the form .11) Furthermore.5) and (7.11)) [φ(x). using (7. which can be compared with (7. [Π(x). Π† (y) = Π† (x). 7. Π† (y) = 0. the equal time canonical commutation relations. Π(y) = iδ3 (x − y). (7. φ(x). in the present case.13) Note that the Hamiltonian. φ(y)] = φ(x).10) 2 2 As a result. although it is to be understood that x0 = y 0 in all of the relations in (7. Π(y)] = Π(x). φ† (y) = φ† (x). φ† (y) = 0. in this case. (7.10). (7.56)).260 Π† (x) = ∂L ˙ = φ† (x) ˙ ∂ φ(x) 7 Complex scalar field theory 1 1 ˙ ˙ = √ φ1 (x) − iφ2 (x) = √ (Π1 (x) − iΠ2 (x)). take the forms (we do not write the equal time condition explicitly for simplicity.2 Field decomposition As in the case of the Hermitian (real) Klein-Gordon field.

As before. with i i = 1. e−ik·x b(k) + eik·x a† (k) . a† (k′ ) = δij δ3 (k − k′ ). 2 1 √ (a1 (k) − ia2 (k)) . The commutation relations between these creation and annihilation operators can then be obtained from the canonical quantization relations to be (see (5. aj (k′ ) = 0 = a† (k). can then be interpreted as the annihilation and the creation operators for the two kinds of quanta associated with the fields φi (x). j (7.15) Given the decomposition in (7.2 Field decomposition 261 φi (x) = d3 k (2π)3 2k0 √ e−ik·x ai (k) + eik·x a† (k) . 2.2)) φ(x) = = 1 √ (φ1 (x) + iφ2 (x)) 2 1 √ 2 d3 k (2π)3 2k0 e−ik·x (a1 (k) + ia2 (k)) +eik·x a† (k) + ia† (k) 1 2 = φ† (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k) + eik·x b† (k) . (7. 2.14) with k0 = Ek = k2 + m2 > 0.16) where we have defined a(k) = b(k) = 1 √ (a1 (k) + ia2 (k)) .7. i j ai (k). i i = 1. ai (k) and a† (k).14). a† (k′ ) . 2 (7. it is clear that the field operators φ(x) and φ† (x) can also be expanded in the plane wave basis as (see (7.17) .66)) ai (k). (7.

18) The Hamiltonian for the system (see (7. a† (k′ ) = b(k). b† (k′ ) = b(k). (7. (7.19) If we normal order the Hamiltonian in (7. a† (k′ ) = 0. a(k).19) with respect to the ai (k) operators and in that case the Hamiltonian would . (7.12)) can be written. a(k′ ) = b(k). b† (k′ ) = 0. b(k′ ) = 0. a† (k). in two equivalent ways as H= = d3 x H d3 k Ek † a1 (k)a1 (k) + a1 (k)a† (k) 1 2 + a† (k)a2 (k) + a2 (k)a† (k) 2 2 = d3 k Ek † a (k)a(k) + a(k)a† (k) 2 + b† (k)b(k) + b(k)b† (k) .17) b (or directly from (7. a(k). a† (k) and † (k) can now be obtained from (7. a† (k′ ) = a† (k). b(k′ ) = a(k).11) and (7. in terms of the creation and the annihilation operators.19) with respect to the a(k) and b(k) operators.8) or (7. b(k). b† (k′ ) = b† (k).20) (We can equivalently normal order the Hamiltonian (7. we can write : H := d3 k Ek a† (k)a(k) + b† (k)b(k) .15) and the definition in (7.16)) and take the form a(k).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. b† (k′ ) = δ3 (k − k′ ). The commutation relations between the a(k).

3 Charge operator To understand the physical meaning of the two distinct and degenerate one particle states. 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. under the transformation φ(x) → φ′ (x) = e−iθ φ(x). degeneracies arise when there is a symmetry present in the system under study. = b† (k)|0 . H|0 = 0.3 Charge operator 263 be a sum of two Klein-Gordon Hamiltonians that we have already studied in chapter 5. energy) given by |k = a† (k)|0 . the theory is invariant. manifestly Lorentz invariant and translation invariant.23) where θ is a real constant parameter of transformation (a global transformation). as was also the case for a real Klein-Gordon theory. as we will see shortly. in the present case. there will now be two distinct one particle states degenerate in mass (and. (7. Namely.22) ˜ |k Similarly. therefore. The degeneracy can be understood as follows. as we know from our studies in non-relativistic quantum mechanics. there is in addition an internal symmetry associated with the system. let us analyze the symmetry properties of this system. of course. The theory is. After all.) The vacuum state in such a case can be defined to be the lowest energy state annihilated by both a(k) and b(k). (7. φ† (x) → φ′ † (x) = φ† (x)eiθ . However.21) However. a(k)|0 = 0 = b(k)|0 = 0|a† (k) = 0|b† (k). 7.7. Namely. (7. Such a transformation is not a space-time symmetry transformation since .

∂µ φ(x). we have (see (6.26) On the other hand. (7.24) where ǫ denotes the infinitesimal constant parameter of transformation. for infinitesimal transformations in (7.5) is invariant under the global phase transformation in (7. δφ(x) = φ′ (x) − φ(x) = −iǫφ(x).23) takes the form (α = ǫ = infinitesimal) δφ† (x) = φ′ † (x) − φ† (x) = iǫφ† (x).8)) is given by L φ′ (x). (7. φ′ † (x).13)) .23) (or the infinitesimal form of it in (7.24). the Lagrangian density (7. in this case.24)) showing that it is a symmetry of the system and in this case we have K µ = 0.264 7 Complex scalar field theory the space-time coordinates are not changed by this transformation. the change in the Lagrangian density (see (6. Rather. Infinitesimally. We note that under the global phase transformation (7.24)). Thus.23) (or the infinitesimal form (7. ∂µ φ′ † (x) −L φ(x). ∂µ φ† (x) = ∂µ φ′ † ∂ µ φ′ − m2 φ′ † φ′ − ∂µ φ† ∂ µ φ + m2 φ† φ = ∂µ φ† eiθ e−iθ ∂ µ φ − m2 φ† eiθ e−iθ φ − ∂µ φ† ∂ µ φ + m2 φ† φ = 0. such a transformation is known as an internal symmetry transformation.25) The derivatives do not act on the exponentials since θ is a constant parameter (independent of space-time coordinates). φ† (x). the transformation in (7. ∂µ φ′ (x). (7.

1) ∂µ J µ (x) = 0. this is the probability current density (up to a normalization) which we had studied in the first quantized theory. as we have seen. (7. we normal order all observables such as currents and charges and. we have freely rearranged the order of factors in (7. we can show that the associated conserved charge of the theory can be written in terms of the creation and annihilation operators as .13)) ← → ∂L ∂L + δφ† − K µ = iǫφ† (x) ∂ µ φ(x).27) (Note here that classically the order of the factors is not important. therefore.29) This is a vector current since the parameter of transformation is a scalar. But here it is an operator.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). under normal ordering the order of factors does not matter (for a bosonic theory).) Furthermore. (7.27).16).28) and the parameter independent current follows to be ← → J µ = iφ† (x) ∂ µ φ(x).7. (7. Of course. Even in a quantum theory. we can define the N¨ther current associated with this o symmetry to be (see (6. the current can be easily checked to be conserved using the equations of motion in (7. (In fact. † ∂∂µ φ ∂∂µ φ µ Jǫ = δφ (7.) Therefore.30) Using the field decomposition (7.

(7. . we are suppressing the normal ordering symbol for simplicity with the understanding that the charge operator has the normal ordered form in (7. 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 . We note that the normal ordered charge operator (7.33) so that the vacuum does not carry any charge. Acting on the first of the two one particle states in (7.32).31) If we normal order.22). a† (k) + a† (k)a(k′ ) |0 (7. Here. (7.32) annihilates the vacuum. a U (1) phase transformation) we can identify this operator as the electric charge operator. (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) Since this charge operator is associated with a global phase transformation of the kind associated with electromagnetic interactions (namely. Q|0 = d3 k a† (k)a(k) − b† (k)b(k) |0 = 0. 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′ ).

22). contains no particle and this should be contrasted with the definition of the Dirac vacuum discussed in section 2. of course.35) ˜ = −|k . in particular. the second quantized theory leads to a much more satisfactory description of particles and anti-particles. ˜ Namely. b† (k) + b† (k)b(k′ ) |0 = = − = − = − d3 k′ b† (k′ )δ3 (k′ − k)|0 = −b† (k)|0 (7. the second one particle state of the theory |k is also an eigenstate of the charge operator but with the eigenvalue (−1). Furthermore. the particle and the anti-particle states – but both of them now correspond to positive energy states.20) d3 k Ek a† (k)a(k) + b† (k)b(k) . (7. the one particle state |k is an eigenstate of the charge operator with eigenvalue (+1).3 Charge operator 267 In other words. the second quantized description is symmetric under the interchange of particles and anti-particles.36) we note that it is invariant under the discrete transformation a(k) ↔ b(k).37) That is. Therefore. These are. They describe two distinct particles with degenerate mass but with exactly opposite electric charge.7. The vacuum in this bosonic theory. This transformation is otherwise known as the charge conjugation symmetry which we will . 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′ ).5. from the form of the Hamiltonian (7. H= (7. acting on the second one particle state in (7. The meaning of the two one particle states is now clear. In contrast.

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 ′ ′ ′ ′ . Since we have already constructed the Feynman Green’s function for the real Klein-Gordon field theory in (5. Thus. Using the field decomposition in (7.16) as well as the commutation relations in (7.146). let us discuss the Feynman Green’s function in the basis of the complex φ(x) and φ† (x) fields.107) that the Feynman Green’s function can be related to the vacuum expectation value of the time ordered product of two field operators. in this section.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. we will construct only the Feynman Green’s function for the complex KleinGordon field theory. We note from (6. to conclude this section.4 Green’s functions In chapter 6 we have already seen that the Feynman Green’s functions play the most important role in the calculations of the S-matrix in a quantum field theory. 0|φ† (x)φ† (y)|0 = d3 kd3 k′ √ 0| e−ik·x b(k) + eik·x a† (k) 2(2π)3 k0 k′ 0 × e−ik ·y b(k′ ) + eik ·y a† (k′ ) |0 = 0.268 7 Complex scalar field theory study in a later chapter. 7. 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. Therefore.

39) .4 Green’s functions 269 × e−ik ·y b(k ) + eik ·y a† (k ) |0 = = = d3 kd3 k √ e−ik·x+ik ·y 0|a(k)a† (k )|0 2(2π)3 k0 k 0 d3 kd3 k √ 2(2π)3 k0 k 0 e−ik·x+ik ·y δ3 (k − k ) d3 k e−ik·(x−y) (2π)3 2k0 = −iG(+) (x − y). (7. Here we have used (7.145). 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).143) and (5. it now follows that 0|T (φ(x)φ(y))|0 = θ(x0 − y 0 ) 0|φ(x)φ(y)|0 + θ(y 0 − x0 ) 0|φ(y)φ(x)|0 = 0. 0|φ† (y)φ(x)|0 = d3 kd3 k √ 2(2π)3 k0 k 0 0| e−ik·y b(k) + eik·y a† (k) × e−ik ·x a(k ) + eik ·x b† (k ) |0 = = = d3 kd3 k √ e−ik·y+ik ·x 0|b(k)b† (k )|0 2(2π)3 k0 k 0 d3 kd3 k √ 2(2π)3 k0 k 0 eik ·x−ik·y δ3 (k − k ) d3 k eik·(x−y) (2π)3 2k0 (7.38) = iG(−) (x − y). Given the relations in (7.21) as well as the definitions of the positive and negative energy Green’s functions in (5.7.

146). we see that in this case. It is easy to check that the Lagrangian density with the quartic interaction λ † 2 (φ φ) .41) describes the most general (renormalizable) self-interacting theory for the complex Klein-Gordon field. 0|T (φ† (x)φ(y))|0 = 0|T (φ(y)φ† (x))|0 = iGF (y − x) = iGF (x − y). namely. However.270 On the other hand. (7. 7.42) . (7. 4 L = ∂µ φ† ∂ µ φ − m2 φ† φ − λ > 0.23). in this case we should look for interactions that would preserve the internal global symmetry in (7. 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).40) Here we have used the definition of the Feynman Green’s function in (5.2) (in order to be consistent with the standard notation used in the discussion of spontaneous symmetry breaking in the literature).5 Spontaneous symmetry breaking and the Goldstone theorem We can introduce interactions into the complex Klein-Gordon field theory much the same way we did for the real Klein-Gordon field theory. 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). Thus. Denoting φ1 = σ and φ2 = χ in the definition in (7. 2 (7. defining 1 φ(x) = √ (σ(x) + iχ(x)) .

42).24).44) The Hamiltonian for the interacting theory (7.41) or (7. 16 L = (7. the minimum of energy coincides with the minimum of the potential (We note that if the field φ is not constant.43) is invariant under the global phase transformation (7. the infinitesimal transformations (7. the minimum of the potential is obtained by requiring ∂V ∂V =0= . ∂V ∂V =0= .24) can be shown to have the form δσ(x) = ǫχ(x). The Lagrangian density (7. this describes the theory of two real Klein-Gordon fields with quartic self-interactions and which are also interacting with each other. δχ(x) = −ǫσ(x). ∂φ ∂φ† or.5 Spontaneous symmetry breaking and the Goldstone theorem 271 we note that the Lagrangian density (7.47) .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 other (kinetic energy) terms would only add positively to the energy. (7.41) has the form λ † 2 (φ φ) . 4 H= d3 x Π† Π + ∇φ† · ∇φ + m2 φ† φ + (7. (7.46) In general. In the real basis of (7.45) and we note that. for constant field configurations.7.43) Namely.23) or the infinitesimal form in (7. ∂σ ∂χ (7.) λ † 2 (φ φ) 4 2 V = m2 φ† φ + = λ m2 2 σ + χ2 + σ 2 + χ2 2 16 .

We note from our discussion of the real Klein-Gordon theory that the plane wave expansion for the field operator in (5.56) has exactly this property (see (5.24) and (7. or. it is clear that the minimum occurs at (the subscript c is supposed to denote a classical field. (7.41) except that we change the sign of the mass term L = ∂µ φ† ∂ µ φ + m2 φ† φ − λ † 2 (φ φ) .51) are clearly invariant under the global phase transformation in (7. c χc = 0|χ(x)|0 = 0. equivalently σc = χc = 0.50) In terms of the real fields σ and χ defined in (7. this translates to the fact that we can define the ground state of the theory such that φ† = 0|φ† (x)|0 = 0.48) As we know from studies in classical mechanics. a concept that we do not get into here) φc = φ† = 0.50) and (7.23) (or (7. this Lagrangian density takes the form 1 m2 2 (∂µ σ∂ µ σ + ∂µ χ∂ µ χ) + σ + χ2 2 2 λ 2 − σ 2 + χ2 . 16 L = (7. (7. Changing the sign of the mass term does not change the . perturbation around the minimum of the potential is stable.42). (7.106)). since the potential in (7. σc = 0|σ(x)|0 = 0.46) is a sum of positive terms.44)). and perturbation theory developed around such a vacuum would be stable.51) Both the Lagrangian densities in (7. 4 λ > 0.272 7 Complex scalar field theory In the present case. Let us next consider the same Lagrangian density for the complex scalar field as in (7. In quantum theory.49) φc = 0|φ(x)|0 = 0. c or.

5 Spontaneous symmetry breaking and the Goldstone theorem 273 symmetry behaviour.52) k2 = −m2 . To begin with.48) . (7. 2 − m2 ∂|k| k (7.51) has the form V = −m2 φ† φ + = − λ † 2 (φ φ) 4 2 m2 2 λ σ + χ2 + σ 2 + χ2 2 16 .7. the particle will travel faster than the speed of light (recall that in our units. but it does change the Hilbert space structure of the theory in a very interesting way.4)) − m2 φ(x) = 0.54) Namely. k0 = k2 − m2 . or. therefore. for constant field configurations the potential of the theory (7.55) The minimum of this potential is obtained from (7. in general.53) This corresponds to a particle with an imaginary mass and has the immediate consequence that the group velocity for the particle motion (this is also the same as the phase velocity in the present case) exceeds unity |k| ∂k0 =√ > 1. the theory (7. (7.41) would lead to the dynamical equation (compare with (1.50) or (7. corresponding to the Einstein relation (7. Such particles are known as tachyons and a theory with tachyons. in this case. a violation of causality).40) or (5. has many undesirable features. c = 1) leading to an acausal propagation (and. Let us note that. let us note that in the absence of interaction (λ = 0).

let us analyze the second derivatives of the potential at the two extrema in (7.57) λ For the second of these solutions. 2m σc =± √ . = σ −m2 + ∂V ∂χ = −m2 χ + λ χ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.59) = 0.57) which lead to ∂2V ∂σ 2 ∂2V ∂σ∂χ while ∂2V ∂σ 2 ∂2V ∂χ2 ∂2V ∂σ∂χ 2m σc =± √ .60) 2m σc =± √ .χc =0 λ . (7. we can choose σc = 0 = χc .56). σc =0=χc = −m2 + = −m2 + = 0. 2 σc + χ2 = c 2m σc = ± √ . 3λ 4 λ 4 4m2 λ 4m2 λ = 2m2 . λ χc = 0. = 0. namely. for example.274 7 Complex scalar field theory ∂V ∂σ = −m2 σ + λ σ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0. (7.56) = χ −m2 + There now appear to be two sets of solutions to the equations in (7.58) To determine the true minimum. 4m2 . (7.χc =0 λ = σc =0=χc ∂2V ∂χ2 σc =0=χc = −m2 .χc =0 λ (7. (7.

the second solution in (7.61) In fact. we note explicitly from the form of the potential in (7.44). indeed.5 Spontaneous symmetry breaking and the Goldstone theorem 275 In other words we note that out of the two solutions in (7. leads to a lower energy. we note that there is an infinity of minima (in the σ-χ field space) of the potential given by the circle (7. All of this is more easily seen from the graph of the potential in Fig.61). There are several interesting things to note from this analysis. the first corresponds to a local maximum.1. First.55) that V (σc = χc = 0) = 0.7. On the other hand.1: Potential leading to spontaneous breaking of symmetry. V (σ. λ λ λ (7. V 2 ((σc + χ2 ) c m2 4m2 ) = − = λ 2 = − 4m2 λ λ + 16 4m2 λ 2 2m4 m4 m4 + =− .57).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 . However.62) so that the second solution. χ) V =0 χ 2m −√ λ 2m √ λ σ Figure 7. Each point on this circle is related by the symmetry transformation (7. . 7. λ (7.

χ(x)] |0 (7. the action or the Hamiltonian of the theory is invariant under the symmetry transformation. if we choose the minimum in (7.64) We recall that conserved charges are the generators of infinitesimal symmetry transformations of the theory and generate the transformations through commutators. λ χc = 0. it now follows that = −iǫ 0| [Q. (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 = √ .44) spontaneously. χ(x)] = −ǫσ(x).44) can be written in terms of the commutators with the charge operator as δσ(x) = −iǫ [Q. in the quantum theory this would imply (see (7. but the ground state is not. Thus. (7. In this case. This has to be contrasted with the case where an interaction term in the Hamiltonian can break the symmetry of a theory explicitly. for example.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. δχ(x) = −iǫ [Q. λ 0|χ(x)|0 = 0. In the quantum theory. λ (7.63) then. such a breaking manifests in the following manner. In this case. the infinitesimal transformations in (7.64). We recall that any choice for the minimum of the potential would translate into the properties of the vacuum state of the quantum theory.66) which implies that .49)) 2m 0|σ(x)|0 = √ . σ(x)] = ǫχ(x). From (7.

This can be easily seen from (7. Therefore.68) Since Q does not annihilate the vacuum of the present theory. HQ|0 = E0 Q|0 .68) and (7. let us denote Q|0 = |χ It follows now from (7. H] = 0. t)Q|0 d3 x 0|eiP ·x J 0 (0)e−iP ·x Q|0 .69) that [Q.69) where we have denoted the energy of the vacuum |0 as E0 and this derivation shows that the state |χ defined in (7.5 Spontaneous symmetry breaking and the Goldstone theorem 277 Q|0 = 0. (7. H|χ = E0 |χ .69) as follows. H]|0 = 0.7. or.70) (7. (7. it is suggestive that we can think of this state as another vacuum state.67) This explicitly shows that the vacuum state of the theory is not invariant under the symmetry transformation in this case. (7. (QH − HQ)|0 = 0.71) .31) and (7. (7. namely.69) would appear to be degenerate with the vacuum state |0 in energy. (note that Q is Hermitian) χ|χ = = = 0|QQ|0 0| d3 x J 0 (x. The problem with this interpretation is that the state is not normalizable. or. although the Hamiltonian of the theory is. [Q. or.

66) that χ(x) corresponds to the Nambu-Goldstone boson of the theory. (7.74) In other words. (7. (7. In fact.75) does not act unitarily on the Hilbert space.75) takes a state out of the Hilbert space. This analysis also shows that the finite transformation operator U (θ) = e−iθQ . (7. .73) where we have used the property of the ground state Pµ |0 = 0.72) χ|χ = = = d3 x 0|eiP ·x J 0 (0)Qe−iP ·x |0 d3 x 0|J 0 (0)Q|0 0|J 0 (0)Q|0 d3 x −→ ∞. 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. it follows that the momentum operator also commutes with Q and. and as a consequence.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. Q] = 0. Since Q does not depend on spatial coordinates.72) leads to (7. the state |χ is not normalizeable and hence cannot be thought of as another vacuum. we have [Pµ . therefore. 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.

77) so that the shifted fields have vanishing vacuum expectation values (vevs) 0|σ(x)|0 = 0 = 0|χ(x)|0 .64) 2m σ = 0|σ(x)|0 = √ . is given by σ(x) = σ ′ (x) + σ with the later identification σ ′ (x) = σ(x) as the dynamical variable. λ χ = 0|χ(x)|0 = 0. say for the first term.1 that the massless mode would correspond to oscillations along the valley of minima while the massive mode would represent the orthogonal oscillations.60) is as follows. We note from Fig. the Lagrangian density in (7.5 Spontaneous symmetry breaking and the Goldstone theorem 279 The second interesting observation that follows from the analysis in (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))).60) shows that in this case of spontaneous breakdown of the symmetry. With the choice of the solutions for the minima as in (7.) Under the shift (7. λ χ(x) → χ(x) + χ = χ(x). We can see this quantitatively in the quantum theory as follows. (7.77).77).78) (Explicitly. √ one of the fields has become massive with a mass 2m while the other is massless (in spite of the fact that the free theory had tachyons (see (7. We note that the second derivatives of the potential (with respect to field variables) give the mass (squared) parameters for the field operators (particles) in a theory and (7. (7. 7.7. (7.51) becomes . the shift in (7.53) and (7.

280 7 Complex scalar field theory L = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 λ − 16 2m σ+ √ λ 2 2 2m σ+ √ λ 2 + χ2 +χ 2 = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 λ − 16 2 2 4m 4m2 σ 2 + χ2 + √ σ + λ λ 2 4m2 4m σ +χ + √ σ+ λ λ = 1 1 m2 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ + 2 2 2 − + λ 16 (σ 2 + χ2 )2 + 4m2 4m σ 2 + χ2 + √ σ + λ λ 8m 16m2 2 16m4 σ + + √ σ(σ 2 + χ2 ) 2 λ λ λ 8m2 2 32m3 (σ + χ2 ) + √ σ λ λ3 = 1 m4 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ − m2 σ 2 + 2 2 λ √ m λ λ − σ(σ 2 + χ2 ) − (σ 2 + χ2 )2 . 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.66)). a continuous symmetry is spontaneously broken. if we look at the Hamiltonian describing a . 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. aware of this phenomenon (known as the spontaneous symmetry breaking phenomenon) from studies in other areas of physics. For example. one of the scalar fields (in this case χ(x)) becomes massless while the other √ scalar field (in this case σ(x)) is massive with a nontachyonic mass 2m. We are. then there must appear massless excitations.

the Hamiltonian is invariant under rotations. we do not observe any massless spin 0 bosons in nature. (For example. in the case of the Ising model). Given the complex Klein-Gordon field theory. Thus. namely. ∂µ → ∂µ + ieAµ . there are two such vacuum configurations allowed in one dimension. H = −g si · si+1 .) 7. The Goldstone theorem is interesting in the sense that it can provide a possible reason for the existence of massless spin zero bosons if they are found in nature. when the spins are all pointing up or down. Consequently the rotational symmetry is spontaneously broken.6 Electromagnetic coupling 281 magnet (for example.6 Electromagnetic coupling It is clear from our discussions so far that the complex Klein-Gordon field theory describes spin zero charged particles whereas the real Klein-Gordon field theory describes spin zero charge neutral particles. The ground state of the theory is clearly the state (spin configuration) where all the spins are aligned. it is clear that once we have chosen a particular ground state (vacuum) where all the spins are pointing in a given direction. Unfortunately. the Lagrangian density describing the interaction of charged spin zero fields with a background electromagnetic field (without the quartic self-interaction term) is given by .7.81) where “e” denotes the charge carried by the particle. The prescription is again through the minimal coupling discussed in (1. The analogue of the massless Nambu-Goldstone bosons in this case correspond to the long range correlations associated with the spin waves.) However. we can ask how such a system can be coupled to a background electromagnetic field. we have effectively chosen a definite orientation in the physical space. (The π meson which is the lightest among the spin zero bosons has a mass mπ ∼ 140 MeV. (7. (7.80) i where the negative sign in the Hamiltonian (g > 0) is for ferromagnets.53).

Aµ (x) → A′ (x) = Aµ (x) + µ 1 ∂µ θ(x). As we can verify readily.82) The modified derivative Dµ introduced in (7.85) .82) is invariant under a local (U (1)) phase transformation of the form φ(x) → φ′ (x) = e−iθ(x) φ(x).84) where the parameter of transformation θ(x) is now a local function of space-time coordinates (unlike in (7. φ† (x) → φ′ † (x) = φ† (x)eiθ(x) . the Lagrangian density (7.282 7 Complex scalar field theory L = ∂µ φ† ∂ µ φ − m2 φ† φ (∂µ − ieAµ ) φ† (∂ µ + ieAµ ) φ − m2 φ† φ (Dµ φ)† (D µ φ) − m2 φ† φ. In fact. note that under this transformation. e (7. (7.23)). ′ Dµ φ(x) → Dµ φ′ (x) = = ∂µ + ieA′ (x) φ′ (x) µ ∂µ + ie Aµ (x) + 1 (∂µ θ(x)) e e−iθ(x) φ(x) = e−iθ(x) (−i(∂µ θ(x)) + ∂µ + ieAµ (x) + i(∂µ θ(x))) φ(x) = e−iθ(x) (∂µ + ieAµ (x)) φ(x) = e−iθ(x) Dµ φ(x). (7.83) 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. → ((∂µ + ieAµ ) φ)† (∂ µ + ieAµ ) φ − m2 φ† φ = = (7.82) Dµ φ = ∂µ φ + ieAµ φ.

86) Consequently. 2. Peterson. which is obtained by minimally coupling the complex scalar field to the electromagnetic field. the covariant derivative acting on a field transforms exactly like the field itself (covariantly) under such a transformation (see (7. 3.84) L(φ(x). J. Physical Review 127. A. (7. 154 (1961). 345 (1961). (In contrast. Salam and S. 965 (1962).7 References 283 In other words.) It follows from this observation that under the local transformation (7.7 References 1.84)). the local phase invariance (7. 4. Jona-Lasinio. Row. Aµ (x)) → L(φ′ (x). Nambu and G. φ† . Nuovo Cimento 19. S. 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. Y. Local phase transformations are commonly known as gauge transformations and we note that. note that ∂µ φ(x) does not transform covariantly.84). Physical Review 122. Goldstone. Weinberg. the theory described by (7. φ′ † (x). Introduction to Relativistic Quantum Field Theory. φ† (x). Aµ . is invariant under the local phase transformation (7.82).7. 7. unlike the global phase invariance in (7. This is a very general feature in a quantum field theory. invariance under a local transformation necessarily requires additional fields known as gauge fields. Goldstone. Schweber. . J. namely.23). Of course. Evanston (1961).84) requires an additional field (Aµ ). 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.

1980. Reviews of Modern Physics 46.284 7 Complex scalar field theory 5. 1964. . J. Drell. 7 (1974). Bjorken and S. J. New York. C. Zuber. 7. Introduction to Quantum Field Theory. Roman. Bernstein. Relativistic Quantum Fields. P. John Wliley. New York (1969). 8. New York. Itzykson and J-B. McGrawHill. 6. McGrawHill. D. Quantum Field Theory.

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

Namely. if this system is analyzed in more detail. can only be nF = 0. 1. the anti-commutation relations in (8. 8.2 Quantization of the Dirac field As we have seen in (1.5) This is. the Dirac equation for a massive spin particle. α = 1. in such a theory. 2. it leads to the fact that with the anti-commutation relations. therefore. a† F + − a† aF aF F (8. there can be at the most one quantum in a given state. 4. what the Pauli exclusion principle would say. (8.286 then.96). This equation is manifestly Lorentz covariant and the adjoint equation has the form . 1 2 (8. (8. is a four component complex spinor function.4) which shows that the eigenvalues of the number operator. has the form (iγ µ ∂µ − m) ψ = 0.3) = a† 1 − a† aF aF = a† aF = NF . It is clear.6) where ψα (x). the wave function of the system will have the necessary antisymmetry property characteristic of fermions. we will be naturally dealing with anti-commuting variables (these variables are commonly known as Grassmann variables) and anti-commutation relations.2) automatically lead to NF (NF − 1) = 0. of course. In fact. F F F In other words. that in dealing with fermionic field operators. we obtain 2 NF = a† aF a† aF F F 8 Dirac field theory = a† F aF . 3.

3). the 1 Dirac field operators are expected to belong to the (reducible) spin 2 representation of the Lorentz group (see (4. However. where the adjoint spinor is defined to be (see also (2. consequently. In deriving the Euler-Lagrange equations. however. With this. In general. in this case.61)) describing fermionic particles and. need to be treated as anti-commuting operators. In dealing with the Dirac field theory.7) (8. ∂ψ ← − ∂L ∂L − ∂µ = mψ − ∂µ (−iψγ µ ) = ψ iγ µ ∂µ + m = 0.9) is manifestly Lorentz invariant (recall the transformation of the Dirac bilinears under a Lorentz transformation discussed in section 3.2 Quantization of the Dirac field 287 ← − ψ iγ µ ∂µ + m = 0.41)) ψ(x) = ψ † (x)γ 0 . ∂ψ ∂∂µ ψ (8.6) and (8. the Lagrangian density of the form / L = ψ (iγ µ ∂µ − m) ψ = iψ∂ ψ − mψψ. we will always use the convention of taking fermionic (Grassmann) derivatives from the left.8. the Euler-Lagrange equations following from the Lagrangian density (8. we would like to decide on a Lorentz invariant Lagrangian density which would give rise to the two Dirac equations (8. as we have seen in chapter 3 as well as in chapter 4. (8. (8.8) In trying to second quantize the Dirac equation (theory).9) give ∂L = (iγ µ ∂µ − m) ψ = 0. Clearly. let us note that since ψ and ψ are anti-commuting (Grassmann) variables.10) .7) as the Euler-Lagrange equations. we have to be careful about signs that may arise in taking derivative operators (with respect to these variables) past other Grassmann variables. 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.

12) However. (8.11) † where we have used (2. Even though the Lagrangian density (8. In fact. 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.288 8 Dirac field theory which are exactly the two Dirac equations (8.83).7) that we are interested in.9) is Lorentz invariant and gives the two Dirac equations as the Euler-Lagrange equations. 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.12) differs from L in (8.9) is not Hermitian.9) only by a total divergence.6) and (8. 2 . note that L† = ψ (iγ µ ∂µ − m) ψ ← − † = ψ † − iγ µ † ∂µ − m ψ ← − = ψ † − iγ µ † ∂µ − m γ 0 ψ ← − = ψ − iγ µ ∂µ − m ψ = ψ (iγ µ ∂µ − m) ψ − i∂µ ψγ µ ψ = L.13) = L − ∂µ ψγ µ ψ .8) as well as the fact that γ 0 is Hermitian. the action associated with it is since a total divergence does not contribute to the action. (8. An alternate Lagrangian density which is Lorentz invariant. it is not Hermitian. we note that the Lagrangian density in (8. It is worth noting here that even though the Lagragian density (8. 2 2 L′ = (8.

27)) for the theory. = 0.15) is a reflection of our choice of left derivatives.16) = δαβ δ3 (x − y).9) to describe the dynamics of the system.8.15) and (8. (Πψ )β (y)]+ . ˙ ∂ψ (8. If we work with the Lagrangian density (8. (Πψ )β (y) or. [ψα (x).16)) [ψα (x). However.2 Quantization of the Dirac field 289 For this reason as well as for simplicity. + † = ψα (x). † † ψα (x).14). −iψβ (y) + ψα (x). (Πψ )β (y) or.9) to describe the Dirac theory (equations). Using the first relation in (8. the second and the third relations in (8.15) can also be written as (Πψ )α (x). in the presence of gravitation. we use L in (8. (Πψ )β (y)]+ = −iδαβ δ3 (x − y). (8.15) The negative sign in the last relation in (8. but we should understand that x0 = y 0 in all the relations in (8. .12) which naturally leads to a symmetric stress tensor (see (6. 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 † . (8. ˙ ∂ψ ∂L = 0.14) Πψ = Πψ = Consequently. 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. ψβ (y) + + = 0. ψβ (y)]+ = 0 = [(Πψ )α (x). ψβ (y) + = −iδαβ δ3 (x − y). it is more appropriate to use the Lagrangian density in (8. † ψα (x).

15) and (8. (8.9) has the form (The negative sign in the first term arises from our convention of choosing left derivatives for fermions. H = d3 y † ψα (x).18) (8.16) are all that we need for quantizing the Dirac theory.19) . ˙ iψα (x) = ψα (x). We will discuss the method of Dirac in more detail in chapter 10.290 8 Dirac field theory (We note that we have not written any anti-commutation relation for ψ α (x) and (Πψ )α (x) simply because the relations in (8.14) can be thought of as constraints on the dynamics of the system. 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) .15) and (8. it is now easy to check that this Hamiltonian leads to the two Dirac equations as the Hamiltonian equations. leading to the Hamiltonian H= d3 x H = d3 x −iψγ · ∇ψ + mψψ . Such systems can be systematically studied through the method of Dirac brackets and such an analysis leads to the conclusion that the relations (8.17) Using the field anti-commutation relations (8.) The Hamiltonian density for the Dirac field theory (8. (8. For example.16).) ˙ H = −Πψ ψ − L ˙ ˙ = − −iψ † ψ − iψ † ψ − iψγ · ∇ψ + mψψ = −iψγ · ∇ψ + mψψ. ψβ (y) + −iγ 0 γ · ∇y ψ(y) + β † + m ψα (x).

15) and (8.8. ψβ (y) + † −mψ β (y) ψα (x).7) as the Hamiltonian (Heisenberg) equations. ← − ψ iγ µ ∂µ + m = 0.22) Namely. ˙ iγ 0 ψ(x) = −iγ · ∇ψ(x) + mψ(x). the anti-commutation relations (8. (iγ µ ∂µ − m) ψ(x) = 0.18) indeed reproduce the two Dirac equations (8. or. or. H d3 y i ψ(y)γ · ∇y β † ψα (x).21) and this can be written in the matrix form as ← − ˙ iψ † = −iψγ · ∇ − mψ. (8. . ψβ (y) + x0 =y 0 = d3 y i ψ(y)γ · ∇y β δαβ δ3 (x − y) x0 =y 0 −mψ β (y)δαβ δ3 (x − y) ← − = −i ψ(x)γ · ∇ α − mψ α (x). or. (8. or. we obtain ˙† iψα (x) = = † ψα (x). ← − ˙ iψγ 0 = −iψγ · ∇ − mψ.6) and (8.2 Quantization of the Dirac field 291 where we have used the fact that the Hamiltonian is independent of time to identify x0 = y 0 and this equation can be written in the matrix form as ˙ iψ(x) = −iγ 0 γ · ∇ψ(x) + mγ 0 ψ(x).16) as well as the Hamiltonian (8. (8.20) Similarly.

s′ ) α 3 (2π) k0 k′ 0 +ei(k+k )·x u† (k. s)uα (k′ .46). / (8. s)ψα (x) α (2π)3 k0 d3 x d3 k ′ m ′ √ ei(k−k )·x u† (k. s) and d† (k. we can expand the Dirac field operator in the basis of these plane wave solutions. s′ ) c(k′ .3 Field decomposition 8 Dirac field theory We have already seen that the positive and the negative energy plane wave spinor solutions of the Dirac equation define a complete basis for the space of four component spinor functions (see the discussion in section 3.23) × e−ik·x c(k. let us note that the field decomposition in (8. s)vα (k. s) and vα (k. Therefore. s) are anti-commuting operators. s) denote the positive and the negative energy spinors defined in (3. s) . As in the case of the Klein-Gordon field (see section 5. s′ ) α ′ .93) and (3.23) can be inverted to give √ (k0 = Ek = k2 + m2 > 0) d3 x = s′ m eik·x u† (k. s)vα (k′ .23) has been chosen keeping in mind the massive normalization of the plane wave solutions in (2.24) The specific form of the prefactor in (8. uα (k. s′ ) d† (k′ . s) + eik·x d† (k. where c(k. we write √ (k0 = Ek = k2 + m2 > 0) ψα (x) = 1 s=± 2 d3 k m (2π)3 k0 (8.292 8. Taking advantage of our experience with the complex Klein-Gordon field theory.94) and the field (8.5).4).23) automatically satisfies the Dirac equation in this basis (i∂ − m) ψ(x) = 0. s)uα (k.

s ) = δss δ3 (k − k ) = d(k. s).28) and they correspond to c(k. we know the equal time quantization conditions for the field operators from (8.25)-(8. s)vα (k .46) (see also (3. s)uα (k .29) + . s ) d† (k . s) and d(k.15) and (8.3 Field decomposition 293 δ3 (k − k ) u† (k.26) that m † e−ik·x ψα (x)uα (k.16) to be ψα (x). k0 m where we have used the massive normalization for the spinors discussed in (2.92)). we can now obtain the anti-commutation relations satisfied by the operators c(k. using the inversion relations (8. (2π)3 k0 (8. s). ψβ (y) † ψα (x). s).27).8. s ) α 0 x0 = s d3 k √ m k0 k 0 + δ3 (k + k ) e2ik = s u† (k. In a similar manner. (2π)3 k0 d3 x (8. d† (k. s) in a straightforward manner from the quantization conditions (8.28) Therefore.25) and (8. s ) + .x0 =y 0 +. ψβ (y) +. s). s ) = c(k. (8. s ) c(k . it can be shown that m † e−ik·x vα (k.x0 =y 0 +. s ) α (8. c† (k.27) c† (k. d† (k .26) It follows from (8. ψβ (y) † † ψα (x). (2π)3 k0 m † eik·x ψα (x)vα (k. s) = d(k. s).25) m k0 δss c(k. s)ψα (x) = d† (k. (8. = δαβ δ3 (x − y). s) = d3 x d3 x Furthermore. = 0.x0 =y 0 = 0. s). s). c† (k . s).

s′ =± 2 ′ d3 x ψ (−iγ · ∇ + m) ψ m k0 ′ m 3 d k k′ 0 × u(k′ . s′ )e−ik ·x × (−iγ · ∇ + m) × e−ik·x c(k. s)(γ · k + m)u(k.294 8 Dirac field theory with all other anti-commutators vanishing. in this case. / we can simplify the Hamiltonian in (8. If we now use the relations (3.94) (8. s′ )eik ·x + v(k′ . s) = 0. s) = (γ 0 k0 − γ · k + m)v(k. s)(−γ · k + m)v(k. s) = 0. s)u(k.93) and (3.18) of the theory. s′ c† (k′ s′ )eik ·x + v k′ . s) = 1 s. s) as well as d(k. s) . s) + eik·x d† (k. / (8. s′ )c† (k′ . s).30) (k + m) v(k. c† (k. s)v(k. This shows that we can think of c(k. s′ d(k′ . s) = (γ 0 k0 − γ · k − m)u(k. s) + eik·x d† (k. s′ )e−ik ·x × e−ik·x c(k. s).31) .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. The Hamiltonian (8. d† (k. s) as annihilation and creation operators for the two kinds of fermionic quanta in this theory (much like the complex Klein-Gordon field theory). s′ )d(k′ .s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m k0 ′ × u k′ .

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

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

δǫ ψ = ψ (x) − ψ(x) = iǫψ(x).12)). ψ(x) → ψ (x) = ψ(x)eiθ . Therefore. the Lagrangian density (8. like the Lagrangian density for the complex Klein-Gordon field theory can be seen to be invariant under a global U (1) phase transformation.39) = ψeiθ (iγ µ ∂µ − m) e−iθ ψ = ψ (iγ µ ∂µ − m) ψ = L(ψ. ψ) → L(ψ ′ . transforms as L(ψ.40) .9) L = ψ (iγ µ ∂µ − m) ψ.4 Charge operator The Dirac Lagrangian density (8. we need to analyze the symmetry properties of the theory. ψ). (8.38) (8.34) with energy eigenvalue k0 = Ek = k2 + m2 > 0. ψ ) = ψ (iγ µ ∂µ − m) ψ ′ ′ ′ ′ (8. We note that under the global phase transformation (α is the constant parameter of transformation) ψ(x) → ψ ′ (x) = e−iθ ψ(x).37) or the infinitesimal form of it (ǫ is the constant infinitesimal parameter of transformation) δǫ ψ = ψ ′ (x) − ψ(x) = −iǫψ(x). ′ (8.4 Charge operator 297 Both these states can be checked to be eigenstates of the Hamil√ tonian (8. as in the case of the complex scalar field theory. there are two distinct one particle states with degenerate mass (energy) and to understand the meaning of these states.9) (or (8.8. 8.

37) (or (8. Thus.41) On the other hand.9) is invariant under the global phase transformation (8.38)) is obtained from (6.29)) and it can be checked using the equations of motion (8.6) and (8.43) where J µ (x) = ψ(x)γ µ ψ(x).37) of the system is given by (note that in the first quantized theory. (8.38).44) represents the current independent of the parameter of transformation.26).45) The conserved charge associated with the symmetry transformation (8. (8. where we have rearranged the fermionic operators keeping in mind their anti-commuting nature.42) δǫ ψ α = (−iǫψ)α (−iψγ µ )α = ǫψ(x)γ µ ψ(x). we note that K µ = 0.7) that this current is conserved. the Lagrangian density (8. as in the case of the complex scalar field theory in (7. the N¨ther current associated with the internal symmetry o transformation (8. In this case.42) is consistent with the convention of taking left derivatives for the fermion fields) ∂L ∂L + δǫ ψα ∂∂µ ψα ∂∂µ ψ α (8.27) and we note that the form of the expression in (8.298 8 Dirac field theory Namely.13) and (7.13) to be µ Jǫ (x) = ǫψ(x)γ µ ψ(x) = ǫJ µ (x).37) (or (8. this would correspond to the total probability) . we obtain (see (6. (8. using the infinitesimal transformation (8. (8.38)). ∂µ J µ (x) = 0. It is a vector current as in the case of the complex Klein-Gordon field theory (see (7.

s′ )d(k′ .4 Charge operator 299 Q= d3 x J 0 (x) = d3 x ψ † (x)ψ(x).38)).46) is associated with an Abelian symmetry transformation (8. as in the case of the complex Klein-Gordon field we can associate the charge operator with that of the electric charge. s′ ). s) . s) + − c† (k. Therefore. s)|0 = 1 s′ =± 2 d3 k′ c† (k′ . since we are dealing with a fermion system.8. s)|0 = |k. it is also possible to associate the charge operator with a fermion number (such as baryon number. = 1 s′ =± 2 d3 k′ c† (k′ . s . s′ ) |0 = 1 s′ =± 2 d3 k′ c† (k′ . s′ ) c† (k. (8. (8. we can express (8. s′ ) − d† (k′ . s′ ) c(k′ .46) We note here that the charge (8. we can now show that the charge operator (normal ordered) satisfies (see (7.46) in terms of creation and annihilation operators as : Q := s=± 1 2 d3 k c† (k. s′ )c(k′ . lepton number. s)d(k. Using the field decomposition in (8. etc) which are associated with U (1) symmetry transformations as well.37) (or (8. s = 0. s)c(k.47) As in the case of the complex Klein-Gordon field. But. s) − d† (k. s′ )δss′ δ3 (k′ − k)|0 = c† (k. . c† (k.23).34)) Q|0 Q|k.33) and (7. s)c(k′ .

the vacuum does not contain any particle as we would intuitively expect (compare this with the first quantized description of the vacuum in section 2. the vacuum state of the theory is charge neutral while the two distinct one particle states. which are degenerate in mass and are created by the operators c† (k.49) namely. as we have also seen in the case of the complex Klein-Gordon field theory.37)) c(k. the second quantized description of the Dirac theory is symmetric under the interchange of particles and anti-particles. in the second quantized description of the Dirac theory. s′ )d(k′ . d† (k. Namely. This shows that. s′ )δss′ δ3 (k − k′ )|0 (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 .300 Q|k. s′ ) 1 s′ =± 2 d(k′ . 8. s .5 Green’s functions As we have discussed within the context of the Klein-Gordon theory. s) ↔ d(k. s′ )c(k′ .34) is symmetric under the discrete transformation (see also (7.48) 1 s′ =± 2 = −d† (k. (8. s). s = 1 s′ =± 2 8 Dirac field theory d3 k′ c† (k′ . carry opposite charge. s)d(k′ . s) + − d† (k. s)|0 = − d3 k′ d† (k′ .5) and that the (normal ordered) Hamiltonian in (8. Thus. we can identify them with the particle and the anti-particle states of the theory with positive energy. s′ ) − d† (k′ . s′ ) d† (k. s′ ) |0 = − d3 k′ d† (k′ . s) respectively. s) and d† (k. s′ ). s)|0 = −|k.

Furthermore.52) S(x − y) = (i∂ + m) G(x − y).5 Green’s functions 301 (i∂ − m) S(x − y) = δ4 (x − y). From these then.53) holds for any Green’s function of the theory (retarded.51) = − ∂µ ∂ µ + m2 . but it is the trivial 4 × 4 identity matrix (which we do not write explicitly).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. / so that (8. Although the relation (8.117) we know that the Green’s function for the Klein-Gordon equation satisfies ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y).50) Here. (8. The right-hand side is also a matrix. S(x − y) is a 4 × 4 matrix function in the spinor space. from (5.8. / (8. we would concentrate mainly on the Feynman Green’s function since that is the most useful .54) Therefore. advanced. Feynman. we do not have to calculate the Green’s function for the Dirac field theory separately. etc. Let us recall the identity (i∂ − m) (i∂ + m) = (i∂ )2 − m2 / / / = − 1 µ ν γ .53) (i∂ − m) S(x − y) = (i∂ − m) (i∂ + m) G(x − y) / / / = − ∂µ ∂ µ + m2 G(x − y) = δ4 (x − y). we can identify (8.).

146)) .302 8 Dirac field theory in the calculation of S-matrix elements.55) SF (k) = (k + m) GF (k) = lim / which can also be written as 1 . 3 (2π) k0 S (−) (x) = (i∂ + m) G(−) (x) / = (i∂ + m) − / = − = i 2 i 2 i 2 d3 k 1 ik·x e (2π)3 k0 / d3 k (−k + m) ik·x e 3 (2π) k0 / d3 k (k − m) ik·x e . (see (5. We note that the relation (8. / so that we can write.57) From (8.140)) k+m / . k2 − m2 + iǫ (8. k − m + iǫ / ǫ→0+ (8.53) can be written in the momentum space as S(k) = (k + m) G(k). 3 (2π) k0 (8.56) SF (k) = lim ǫ→0+ (8. for example.53) we can also obtain the positive and the negative energy Green’s functions for the Dirac field theory as (see (5.145) √ where k0 = Ek = k2 + m2 > 0) S (+) (x) = (i∂ + m) G(+) (x) / = (i∂ + m) / = i 2 i 2 d3 k 1 −ik·x e (2π)3 k0 d3 k k + m −ik·x / e .58) We can express the Feynman Green’s function in terms of the positive and the negative energy Green’s functions as well (see (5.143) and (5.

s).8) also as ψ α (x) = ψ α + ψ α (x). we can decompose the adjoint field operator (8.60) where (+) ψα (x) = 1 s=± 2 d3 k m e−ik·x c(k.6 Covariant anti-commutation relations 303 SF (x) = (i∂ + m) GF (x) / = (i∂ + m) −θ x0 G(+) (x) + θ −x0 G(−) (x) / = −θ x0 S (+) (x) + θ −x0 S (−) (x).62) (+) d3 k m e−ik·x d(k. (2π)3 k0 m eik·x d† (k.6 Covariant anti-commutation relations Let us decompose the field operator into its positive and negative energy (frequency) parts as (+) (−) ψα (x) = ψα (x) + ψα (x). (8. s). (8. other Green’s functions for the Dirac field theory can also be easily obtained from the corresponding functions for the KleinGordon field theory. s)uα (k. s). s)vα (k. s). Similarly. δ(x0 )(G(+) (x) + G(−) (x)) = 0). (8.61) (2π)3 k0 (−) ψα (x) = 1 s=± 2 d3 k Similarly. where ψ α (x) = 1 s=± 2 (+) (−) (8. (2π)3 k0 m eik·x c† (k. (8.63) (2π)3 k0 ψ α (x) = 1 s=± 2 (−) d3 k . s)uα (k. s)v α (k. 8.8.59) where we have used the fact that the terms involving derivatives of the step function cancel out (namely.

s′ = 1 s.s′ =± 1 2 ′ c(k. s) (2π)3 k0 = = 1 2 / d3 k m (k + m)αβ −ik·(x−y) e 3 k0 (2π) 2m / d3 k (k + m)αβ −ik·(x−y) e 3 (2π) k0 (+) = −iSαβ (x − y). ψ β (y) (+) + = s. s)uβ k′ . s). s)v β k′ . s′ δss′ δ3 (k − k′ ) = 1 s=± 2 d3 k m −ik·(x−y) e uα (k. s) in (8. d(k′ .s′ =± 1 2 m2 k0 k′ 0 + × e−ik·x+ik ·y uα (k. c† (k.106) as well as the identification in (8.64) where we have used the completeness relation in (3. s)uβ (k. s) and d† (k.29) that the only nontrivial anti-commutation relations among these four field components will have the form (these are not at equal times) d3 kd3 k′ (2π)3 (+) ψα (x).58). Similarly. d(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. c† (k′ . s). (8.s′ =± 1 2 m2 k0 k′ 0 + × eik·x−ik ·y vα (k.s′ =± 2 ′ d† (k. s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 × e−ik·x+ik ·y uα (k. ψ β (y) (−) + ′ = s. s). we have d3 kd3 k′ (2π)3 (−) ψα (x). s′ = s. s). s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 .

5 and 6. in rearranging terms to bring them to a particular form.7 Normal ordered and time ordered products 305 × eik·x−ik ·y vα (k. As a result. ψ β (y) + + (−) + ψα (x).6. ψ β (y) (+) (−) (8.61) and (8. αβ (8. by definition the normal ordered product of two field operators yields (note from (8. since fermionic field operators anti-commute (Grassmann variables). we need to be careful about negative signs that can arise from changing the order of any two such operators. where we have used (3.136) (see also (5.65) ψα (x).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. we can write (+) = ψα (x). s)v β k′ . Thus. s′ δss′ δ3 (k − k′ ) = 1 s=± 2 ′ d3 k m ik·(x−y) e vα (k.66) which is the analogue of the Schwinger function (5.109).63) that the positive and the negative energy parts of the field operators contain annihilation and creation operators respectively) . However. 8.8.160)) for the Dirac field theory and these relations are analogous to the covariant commutation relations for the Klein-Gordon field operators discussed in section 5.9. 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). ψ β (y) (−) (+) + = −iSαβ (x − y) − iSαβ (x − y) = −i S (+) (x − y) + S (−) (x − y) = −iSαβ (x − y). s)v β (k.

using (8.306 8 Dirac field theory (−) (−) : ψα (x)ψβ (y) : = ψα (x)ψβ (y) (−) (−) : ψβ (y)ψα (x) : = −ψα (x)ψβ (y) (−) = − : ψα (x)ψ β (y) :. here the order of factors inside normal ordering can lead to additional signs. In other words.64). As a result. we can write (8.69) (8. ψ β (y) (−) + (+) (−) (+) (−) (+) (−) (+) (−) (+) (−) = ψα (x)ψ β (y) + iSαβ (x − y).65).71) . we see that we can denote 0|ψα (x)ψ β (y)|0 = −iSαβ (x − y) = ψα (x)ψ β (y).67). it is easy to show that ψ β (y)ψα (x) = : ψ β (y)ψα (x) : −iSαβ (x − y). unlike the bosonic case. Similarly. it is clear that (+) (−) : ψα (x)ψ β (y) : = : ψα (x) + ψα (x) ψ β (y) + ψ β (y) : (+) (+) = : ψα (x)ψ β (y) + ψα (x)ψ β (y) (−) (−) + ψα (x)ψ β (y) + ψα (x)ψ β (y) : (+) (+) = ψα (x)ψ β (y) − ψ β (y)ψα (x) (−) (−) + ψα (x)ψ β (y) + ψα (x)ψ β (y) (+) = ψα (x)ψ β (y) − ψα (x).68) where in the last step we have made the identification in (8. (+) (−) (+) (8.70) (8.67) and we note that. (+) (+) (+) (+) (+) (8. (+) (8.68) as ψα (x)ψ β (y) = : ψα (x)ψ β (y) : −iSαβ (x − y). From the definition in (8.

we can develop Wick’s theorem for time ordered products of Dirac fields in a straightforward manner as discussed in section 6. .8.7 Normal ordered and time ordered products 307 which can be compared with (6.74) Once again.81). 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 6.73) (+) (−) (−) (+) where we have used the definition in (8. (8.6.59) and which can be compared with (6.104).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). It follows now from (8. (8. (8. 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.αβ (x − y).αβ (x − y) = ψα (x)ψ β (y).72) This can be expressed in terms of the normal ordered products as T ψα (x)ψ β (y) = θ(x0 − y 0 ) : ψα (x)ψ β (y) : −iSαβ (x − y) −θ(y 0 − x0 ) : ψ β (y)ψα (x) : −iSαβ (x − y) = θ(x0 − y 0 ) : ψα (x)ψ β (y) : + θ(y 0 − x0 ) : ψα (x)ψ β (y) : +i − θ(x0 − y 0 )Sαβ (x − y) + θ(y 0 − x0 )Sαβ (x − y) = = θ(x0 − y 0 ) + θ(y 0 − x0 ) : ψα (x)ψ β (y) : + iSF .αβ (x − y) : ψα (x)ψ β (y) : + iSF . 0|T ψα (x)ψ β (y) |0 = iSF .

7. Furthermore.23) because we are using here the normalization (2. Let us consider the field quantization of a right-handed Dirac field (Weyl field).170) we can invert the decomposition in (8.77) to obtain (here k0 = |k|) .3). for example.308 8. The discussion for the left-handed field can be carried out in an analogous manner.164)).77) is different from those in (8. we note here that there is no sum over spin polarization because the chiral spinors are charaterized by their unique chirality or helicity (see.53) and (2. Using the orthonormality relations for the right-handed spinors described in (3. it is most natural to decompose it into its chirality states. (8.75) As a result.23). Therefore. the free Dirac Lagrangian density for a massless spinor field naturally decomposes as / / / L = iψ∂ ψ = iψ R ∂ ψR + iψ L ∂ ψL .54) which is appropriate for massless spinors. in the case of the standard model in 14. the multiplicative factors in (8. when the Dirac particle is massless. we can also decompose a Dirac field into a left-handed and a right-handed component as 1 (½ + γ5 )ψ(x).76) Such massless left-handed and right-handed Dirac fields (Weyl fields) arise naturally in physical theories (as we will see. we can expand the field as (see (8.77) where. 2 ψR (x) = ψL (x) = (8. as before.7. the discussion following (3. 2 1 (½ − γ5 ). for example. (8. Using the definition of the chiral spinors in section 3. we have identified k0 = Ek = |k|.23)) d3 k (2π)3 |k| ψR (x) = e−ik·x c(k)uR (k) + eik·x d† (k)vR (k) .8 Massless Dirac fields 8 Dirac field theory As we have seen in section 3. in this section we will discuss briefly the quantization of such fields. Unlike in (8. Correspondingly.

76)). (8. (8. . from the equal-time anti-commutation relations for the field variables following from the Lagrangian density for the righthanded spinor field (see.8.78) 0 Similarly.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). (8.81) and the higher particle states created by applying the creation operators and other quantities of interest can be derived in the standard manner. d† (k′ ) + + .79) As a result. c† (k′ ) = δ3 (k − k′ ) = d(k). for example. 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).80) The Hilbert space of the theory can now be built with the vacuum defined as satisfying c(k)|0 = 0 = d(k)|0 . it can also be shown that (it is assumed that k0 = |k|) d† (k) = d3 x † e−ik·x vR (k)ψR (x). (8.

82) Putting in the field decomposition in (8.R (x − y) = (2π)3 4|k| × (|k|γ 0 − γ · k)θ(x0 − y 0 )e−i|k|(x −(|k|γ 0 − γ · k)θ(y 0 − x0 )ei|k|(x = 0 −y 0 )+ik·(x−y) 0 −y 0 )−ik·(x−y) γ 0 (½ + γ5 ) d3 k eik·(x−y) 0 0 (|k|γ 0 − γ · k)θ(x0 − y 0 )e−i|k|(x −y ) 3 (2π) 4|k| −(|k|γ 0 + γ · k)θ(y 0 − x0 )ei|k|(x 0 −y 0 ) γ 0 (½ + γ5 ).172) this can also be written as d3 k 1 iSF . let us determine the propagator for a righthanded fermion.84) .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) . We recall that the Feynman propagator is defined as the time ordered product (see. (8. (8.77) and using the properties of the vacuum in (8.Rαβ (x − y) = † 0|T ψRα (x)ψRβ (y) |0 † = θ(x0 − y 0 ) 0|ψRα (x)ψRβ (y)|0 † −θ(y 0 − x0 ) 0|ψRβ (y)ψRα (x)|0 .83) Here we have used (8.81) we obtain (we assume k0 = |k| in the field decomposition) iSF .74)) iSF .80) in evaluating the vacuum expectation values.310 8 Dirac field theory As an example. for example. (8. (8. Dropping the spinor indices and using (3.

2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 (8.8 Massless Dirac fields 311 Let us evaluate the two terms in (8.85) Here we have used the fact that the integral has contributions only when k0 = |k| and correspondingly have rewritten some of the terms.84) separately.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 ) / . The second term in the integrand can also be simplified in a similar manner − 0 0 1 θ(y 0 − x0 )(|k|γ 0 + γ · k)ei|k|(x −y ) 4|k| = = = = dk0 (|k|γ 0 + γ · k) ei(k +|k|)(x 2πi 4|k| k0 + i 0 0 0 0 −y 0 ) dk0 (|k|γ 0 + γ · k) eik (x −y ) 2πi 4|k| k0 − |k| + i dk0 (k0 γ 0 + γ · k) eik (x −y ) 2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 dk0 (−k0 γ 0 + γ · k) e−ik (x −y ) 2πi 4k0 k0 + |k| − i 0 .8. Using the integral representation for the theta function (see (6.

R (k) = (8. (8.86) Therefore. everything carries over in a straightforward manner to left-handed fermion fields.88) by simply multiplying a factor of γ 0 on the right which leads to iS F .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. However.312 = − / dk0 k e−ik (x −y ) . 2(k2 + iǫ) k + iǫ (8.9 Yukawa interaction Let us next consider the theory which describes the interaction between a Dirac field and a charge neutral Klein-Gordon field. The reason for this is that the adjoint field ψ R has negative chirality.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. the conventional covariant propagator (defined as the time ordered product of ψR ψ R ) can be obtained from (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 . 2(k2 + iǫ) k + iǫ iSF .85) and (8.87) iSF . adding the two results in (8.86) and substituting into (8. For .R (x − y) = / d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) e .88) We note here that we have defined the propagator in (8.R (k) = iSF . 8.74)).R (k)γ 0 = ik(½ − γ5 ) / ikPL / = 2 . for example.89) We note that although our discussion has been within the context of right-handed fermion fields.

91) (8.9 Yukawa interaction 313 example. ψψ is invariant under a Lorentz transformation and φ is a scalar field) and is known as the Yukawa interaction. the correct parity invariant trilinear interaction for a pseudoscalar meson should be (−gψγ5 ψφ). L = L0 + LI . we are going to ignore this aspect of the meson and also switch off the self-interactions of the meson field for simplicity. we can denote the interaction Lagrangian density as (remember everything is normal ordered) .8. (8. since here we are only interested in developing calculational methods in a quantum field theory. The trilinear coupling gψψφ is manifestly Lorentz invariant (as we have seen in (3. The Lorentz invariant Lagrangian density.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. 4! (8.92).86). L0 = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 λ LI = −g ψψφ − φ4 .g. we can separate the Lagrangian density (8..36). has the form 1 λ M2 2 / φ − g ψψφ − φ4 . since it leads to the Yukawa potential in the Born approximation and the coupling constant g represents the strength of the interaction. Namely. However. 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. where 1 M2 2 / φ . the π 0 meson) has mass M . in this case. (The self-interaction can be switched off at the lowest order by setting λ = 0.92) We note here that the π 0 meson is experimentally known to be a pseudoscalar meson (changes sign under a space reflection) which is not reflected in our trilinear interaction in (8.90) into a free part and an interaction part. As in (6.) In this case.

69).5 and 6.94) +··· . Using Wick’s theorem (see sections 6. (8. in this theory the S-matrix will have the expansion (see (6. we can write (8. the adiabatic switch-off is assumed) S = T e−i R∞ R −∞ LI = −g : ψψφ := −HI .95) . (8.94) in terms of normal ordered products as S = ½ − ig − g2 2 d4 x : ψ(x)ψ(x)φ(x) : d4 xd4 x′ : ψ(x)ψ(x)φ(x)ψ(x′ )ψ(x′ )φ(x′ ) : + : ψ(x)ψ(x)ψ(x′ )ψ(x′ ) : φ(x)φ(x′ ) + ψα (x)ψ β (x′ ) : ψ α (x)ψβ (x′ )φ(x)φ(x′ ) : − ψα (x′ )ψ β (x) : ψβ (x)ψ α (x′ )φ(x)φ(x′ ) : + φ(x)φ(x′ ) ψα (x)ψ β (x′ ) : ψ α (x)ψβ (x′ ) : − φ(x)φ(x′ ) ψα (x′ )ψ β (x) : ψβ (x)ψ α (x′ ) : − ψα (x)ψ β (x′ ) ψβ (x′ )ψ α (x) : φ(x)φ(x′ ) : − ψα (x)ψ β (x′ ) ψβ (x′ )ψ α (x) φ(x)φ(x′ ) + ······ .314 8 Dirac field theory As a result.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′ ) : (8.6).

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

The delta function merely ensures the overall conservation of energy and momentum in the process. In fact. q 0 > 0. (For completeness. note that since k0 . We can similarly associate a physical process with every other term of S (1) in (8.99) can be satisfied only if the energy of each of the particles vanishes which is not possible since the particles are massive. q = −ig d4 x s. k′ 0 . s) ˜ ˜ × 0|d(p′ . s. let us consider the second .˜′ =± 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′ .99) represents the process where a proton. s′ )uα (p. s)d† (k′ . s) ˜ ˜ ˜ p × δss δ3 (p − k) δs′ s′ δ3 (p′ − k′ ) δ3 (˜ − q) ˜ ˜ = −(2π)4 ig δ4 (k + k′ + q) × 1 (2π)3 2q 0 m (2π)3 k0 m v α (k′ . energy conservation δ(k0 + k′ 0 + q 0 ) in (8. s). However. our goal in this section has been to describe the calculational methods in quantum field theory. s′ )a† (q)|0 ˜ ˜ p = −ig d 3 p d 3 p d 3 p′ ˜ 1 s.95) also leads to a distinct physical process. s)a(˜ )c† (k. (2π)3 k′ 0 (8. we note here that this lowest order matrix element in (8. For example. s′ )c(p.˜′ =± 1 ˜s 2 ′ (1) d 3 p d 3 p d 3 p′ ˜ (2π)3 2˜ 0 p m (2π)3 p0 m (2π)3 p′ 0 p × e−i(p+p +˜)·x v α (p′ .316 8 Dirac field theory 0|S1 |k. k′ .99) where we have used the appropriate (anti) commutation relations between the creation and the annihilation operators in evaluating the vacuum expectation value. s′ . s′ )uα (p.96).99) vanishes because energy-momentum conservation cannot be satisfied. an anti-proton along with a π 0 meson are annihilated. s′ )uα (k. The matrix element (8.) Each term at order g2 in the expansion in (8.

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. s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x ) (−) + ψ α (x )ψα (x )ψ β (x)ψβ (x)|k1 . the matrix element will have the form k4 . k2 . k3 . s3 |S2 |k1 . s2 = − ig2 2 d4 xd4 x GF (x − x ) (−) (+) (+) (2) (−) × k4 . s2 )uβ (k1 . s4 . k2 . (2π)4 GF x − x = (8. (8. In such a case. s3 )vα (k4 . s1 ). s1 . s4 )v β (k2 . k3 .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 ) : . s4 . Recalling that the Fourier transformation of the Green’s function is given by (see (5. s1 . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x )|k1 . s1 .9 Yukawa interaction 317 term in the expansion at order g2 in (8.120)) d4 q −iq·(x−x ) e GF (q). s2 (−) (+) (+) = −ig2 × d4 xd4 x GF (x − x ) (−) (+) (+) (−) k4 .8. k2 . 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 (8.102) .101) × uα (k3 . s4 . k3 .

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. s4 . The delta function merely ensures the overall energy momentum conservation in the process. s3 |S2 |k1 . s3 )vα (k4 . s3 )vα (k4 . s4 ) v β (k2 .318 the matrix element (8.101) can be written as (2) 8 Dirac field theory k4 . s2 = −ig2 × d4 q 4 4 ′ ′ d xd x GF (q)e−i(q−k3 −k4 )·x ei(q−k1 −k2 )·x (2π)4 m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 × uα (k3 . s1 ) = −ig2 × d4 q (2π)4 δ4 (q − k3 − k4 )δ4 (q − k1 − k2 ) GF (q) m 0 (2π)3 k1 m 0 (2π)3 k2 m 0 (2π)3 k3 m 0 (2π)3 k4 × uα (k3 .10 Feynman diagrams It is quite clear from the two simple examples that we have worked out in the last section that the evaluation of the S-matrix elements using the Wick expansion is quite tedious. s1 . s2 )uβ (k1 . s2 )uβ (k1 . s3 )vα (k4 . 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. k2 . It is clear from looking at the expansion of the S-matrix that it contains terms . s4 )iGF (k1 + k2 )v β (k2 . Physically this matrix element corresponds to a proton and an antiproton annihilating to create a π 0 meson which subsequently pair produces a proton and an anti-proton. s1 ). Instead. k3 . s2 )uβ (k1 .103) × uα (k3 .

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

(anti-particle creation). since the fermion field carries a charge (and. in a local quantum field theory. Thus. v α (k. for the Yukawa interaction (8. (particle creation). s). s). (particle annihilation). a fermion external line with an arrow away from the interaction point (vertex) can represent the creation of a particle or the annihilation of an anti-particle (since it corresponds to the field ψ) with an external line factor  =  m (2π)3 k 0 m (2π)3 k 0 uα (k. s α  = Note that it is the interaction between the different fields which gives rise to nontrivial scattering. The interaction can be read out from the Lagrangian density LI and. (8. the normalization factor for the external line √ (state) is denoted by (k0 = k2 + M 2 ) k = 1 (2π)3 2k0 .93). carries a direction for charge flow). s). 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. 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. (8.105) On the other hand. we . vα (k.320 8 Dirac field theory and in such a case. therefore. is represented as a vertex (point of interaction) with no meaning attached to the external lines.107) k.106) k. (8. (anti-particle annihilation). s). s α  = However.

which we do not worry about at this point.8. with the normalization factor (2π)4 for every momentum integration. Furthermore. the rule for connecting vertices and lines is that the four momentum of every internal line (propagator) must be integrated 1 over all possible values. There are also some other subtleties like the symmetry factor associated with a diagram (graph). namely. s′ Figure 8. we can show that every Feynman diagram corresponds to a unique physical process in one to one correspondence with Wick’s expansion. With these rules. calculations of S-matrix elements are simpler in momentum space. (There is one other rule. (8.108) 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.) q α β k. conventionally known as Feynman rules. . However.10 Feynman diagrams 321 have with the convention that all momenta are incoming at a vertex (basically. one has to have a factor of (−1) for every internal fermion loop. 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 . s k ′ . We should also note for completeness that these are the Feynman rules in momentum space.1: Lowest order Feynman diagram.

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′ . The value of this diagram can be obtained from the Feynman rules to be 0|S1 |k. s′ )uα (k.99). therefore.322 8 Dirac field theory In this language. an anti-proton and a π 0 meson are annihilated (which we considered in (8. q = −(2π)4 igδ4 (k + k′ + q)δαβ × 1 (2π)3 2q 0 m uα (k. s). k1 . s) (2π)3 k0 m v β (k′ . s′ .109) This is exactly the value of the first order S-matrix element which we had evaluated in (8. the physical process where a proton. The . k′ . 8. s. s1 k4 . the corresponding lowest order Feynman diagram will be given by Fig.2.2: A second order Feynman diagram. where the external lines represent the respective physical particles that are annihilated.99)) will correspond in the lowest order to evaluating the Feynman diagram in Fig. s4 α β q δ γ k2 . (2π)3 k′ 0 (8. with the external lines representing physical particles. 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. 8.1. s3 Figure 8. s2 k3 .

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

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

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. N. McGrawHill. Evanston (1961). therefore. we obtain from (8. Peterson. N. J. 5. Moscow (1984). Shirkov. 4π |x| V (x) = − (8.114) the potential between the fermions to be g2 e−M |x| . John Wliley. Relativistic Quantum Mechanics and Field Theory. 1980. Quantum Field Theory.11 References 1. Introduction to the theory of Quantized Fields. F. 3. Introduction to Relativistic Quantum Field Theory. Relativistic Quantum Fields. S. Introduction to Quantum Field Theory. Bjorken and S.11 References 325 where we have used the method of residues to evaluate the integral (the poles of the integrand are along the imaginary axis and closing the contour in the upper half or in the lower half of the complex zplane leads to the same result). 2.8. New York (1969). Nauka. Row. 8. 6. New York.93) leads to the Yukawa potential in the Born approximation. D. McGrawHill. New York. Schweber. C. V. Roman. . 1964. P. Zuber. 4. John Wiley. Bogoliubov and D.115) We recognize this as the Yukawa potential (for the exchange of a particle with mass M ) and this shows that the Yukawa interaction (8. Gross. Drell. Itzykson and J-B. New York (1993).

This page intentionally left blank .

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

5) µ.328 9 Maxwell field theory B = ∇ × A. ∂t (9. (9. A). E = − ∂A − ∇Φ. E2 . 3. In other words. 2. of course. denotes the electric field vector. −A). (9.2) where A(x) and Φ(x) are known as the vector and the scalar potentials respectively. combine them into the four vector potential Aµ = (Φ. ν = 0. B2 . 1.8) denotes the magnetic field vector. where E = (E1 .4) where ǫijk denotes the three dimensional Levi-Civita tensor and B = (B1 . the second .6) (9. (9. the six independent components of the tensor Fµν are given by (three components each of) the electric and the magnetic fields. We can. Similarly. Therefore. we note that F0i = ∂0 Ai − ∂i A0 = Ei .3) If we now define the four dimensional curl of the four vector potential as Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . Aµ = (Φ. E3 ).7) (9. B3 ). Fij = ∂i Aj − ∂j Ai = −ǫijk Bk . (9.

It is now easy to check that the other two equations (the first and the fourth) in (9. Thus. for ν = j.9) leads to ∂µ F µj = 0. equation (9. or. or.11) which is the last equation in (9.9. ∂t ∂Ej = −ǫjik ∂ i Bk .1).9) takes the form (F 00 = 0 by definition (9.12) . or. equation (9.4) as Fµν = ∂µ Aν − ∂ν Aµ . (9. we see that the set of four Maxwell’s equations (9. or. (9. ∂0 F 0j + ∂i F ij = 0. ∂Ej + ∂ i (−ǫijk Bk ) = 0. or.13) (9.10) which coincides with the first of the equations in (9. for ν = 0. ∂t (9. ∂t ∂E = ∇ × B.1 Maxwell’s equations 329 rank anti-symmetric tensor Fµν is also known as the field strength tensor.1) can be written in the manifestly covariant form as ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0.1) are given by ∂µ F µν = 0.1).9) For example. (9. with the field strength tensor defined in (9.4)) ∂i F i0 = 0. Similarly. ∂ 0 F0j + ∂ i Fij = 0. ∇ · E = 0. or.

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

indeed. (9.9). namely. . = − 1 1 1 (∂ µ Aν − ∂ ν Aµ ) − ∂ µ Aν + ∂ ν Aµ 2 2 2 (9. Fµν = ∂µ Aν − ∂ν Aµ .16).20). ∂µ or. Note that a Lagrangian density 1 1 Fµν F µν − Fµν (∂ µ Aν − ∂ ν Aµ ) . ∂µ (∂ µ Aν − ∂ ν Aµ ) = ∂µ F µν = 0. ∂L = 0. ∂Aν ∂∂µ Aν or. ∂∂µ Aν (9. gives Maxwell’s equations as its Euler-Lagrange equations. we will work with the simpler Lagrangian density in (9. The Lagrangian density (9.18) which coincides with the manifestly covariant Maxwell’s equations (9.19) is equivalent to (9.16) when the field variable Fµν is eliminated using the (first) auxiliary field equation in (9. ∂L = 0. so that the Euler-Lagrange equations. take the form ∂L ∂L − ∂µ = 0. 4 2 L= (9. However.17) = − (∂ µ Aν − ∂ ν Aµ ) = −F µν . ∂Fµν or. It is manifestly Lorentz invariant.19) with both Aµ and Fµν treated as independent dynamical variables would also lead to Maxwell’s equations as its Euler-Lagrange equations. in this case.2 Canonical quantization 331 ∂L ∂Aν ∂L ∂∂µ Aν = 0.16) since the Lagrangian density (9.9.20) ∂µ F µν = 0.

(9.332 9 Maxwell field theory Furthermore. x0 =y 0 . 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. we note that A0 (x) which would have a nontrivial commutation relation only with Π0 (x). Πj (y) x0 =y 0 = 0 = Πi (x).25) .15)). ˙ ∂ Aµ (x) (9. (9.21). in this gauge the nontrivial components of the conjugate momentum are given by ˙ Πi (x) = −F 0i = Ei = −(∂ 0 Ai − ∂ i A0 ) = −Ai (x). this Lagrangian density is also invariant under the gauge transformation (9. If we treat Aµ (x) as the independent field variable. As a result.24) The naive canonical quantization conditions. (9. since the field strength tensor Fµν is gauge invariant (see (9. the momentum conjugate to the field variable A0 (x) vanishes. (9. now must commute with all the field variables in the theory.21) and the first peculiarity of the Maxwell field theory is now obvious from (9. we have to choose a gauge and this particular choice is known as the temporal (axial) gauge. namely. Aj (y)]x0 =y0 Ai (x).17)) Πµ (x) = ∂L = −F 0µ . in this gauge.14) of the vector potential. we can define the associated conjugate momentum as (see (9.23) This is equivalent to saying that. for canonical quantization of such a system. Π0 (x) = −F 00 = 0. Πj (y) j = iδi δ3 (x − y). Furthermore. Therefore. This is a consequence of the gauge invariance of the theory. we can think of A0 (x) as a classical function (c-number function) – not as an operator – and without loss of generality we can set it equal to zero A0 (x) = 0. would then appear to be [Ai (x).22) In other words.

in this gauge. (9. in the gauge (9. or.27) (9. this is not quite consistent which can be seen as follows.27) can be thought of as the gauge choice for this residual invariance.27) is given by ˙ Ai (x). ∂ 0 (∇ · A) = 0.23). For example.26) Namely. or. the dynamical field variable A(x) must be transverse. (As a parenthetical remark.23). the dynamical equations (see (9. ˙ Ai (x).28) and the condition (9.30) . ¯ A(x) → A′ (x) = A(x) + ∇θ(x). ∂ 0 (∂µ Aµ ) = 0. Πj (y) or.29) which does not satisfy the transversality condition (9. j = −iδi δ3 (x − y). x0 =y 0 (9.2 Canonical quantization 333 However. ∇ · A = 0. lead to (for ν = 0) ∂µ (∂ µ A0 − ∂ 0 Aµ ) = 0. the canonical quantization condition (9. (9. Aj (y) j = iδi δ3 (x − y).) On the other hand. the theory can be checked to have a residual gauge invariance under a time independent gauge transformation which preserves (9.23). namely.18)) ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0. Aj (y) j = −iδi TR (x − y).9.25) leads to Ai (x). we note that even after imposing the gauge condition (9.27). or. The proper quantization condition consistent with the transversality condition (9. x0 =y 0 x0 =y 0 (9.

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). As we have pointed out in connection with the discussions in the Dirac field theory. When quantized systematically through the Dirac method.23).32) = ∂ j + ∂ i ∂i 1 j j j 2 ∂ = ∂ − ∂ = 0.30) is consistent with the transversality condition in (9.23) and (9.27) is given by (9. (2π)3 (9.27).30). ∇ and (9. the correct quantization condition (Dirac bracket) that results in the gauge (9.23) subject to (9. (9. ∇2 (9. which we will discuss in the next chapter.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 . known as the Dirac method.27) constitute constraints on the dynamics of the theory and systems with constraints are known as constrained systems. Here 1 2 stands for the inverse of the Laplacian operator and for∇ mally denotes the Green’s function for ∇2 . Let us note here that in momentum space.22). We note here that relations such as (9.31) so that (remember that ∂ i ∂i = −∇2 ) j j ∂ i δi TR = ∂ i δi + ∂i 1 j 2 ∂ ∇ (9. there exists a systematic procedure for deriving the correct quantization conditions for such systems.33) In the gauge (9. the Hamiltonian density for the Maxwell field theory can be obtained to be . the transverse delta function in (9.

27)) A0 = 0.3 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 .18) take the form ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = Aν = 0. 2 (9.23) (see also (9.9.37) . Maxwell’s equations (9. (9.36) which also lead to ∂ µ Aµ = 0. ∇ · A = 0. 9.38) (9. (9.34) and leads to the Hamiltonian d3 x H = = 1 2 1 2 d3 x Π2 + B2 d3 x E2 + B2 .3 Field decomposition In the temporal gauge (9.35) H= Indeed this is what we know to be the Hamiltonian (energy) for the Maxwell theory even from studies in classical electrodynamics. (9.

40) (9. the wave equation (9. . we must have ∇ · A(x) = 0. We can choose the two independent polarization vectors to be orthonormal and. satisfy the Klein-Gordon equation for a massless particle (wave equation) and. On the other hand. we have Ai = 0.336 9 Maxwell field theory As we have seen.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. (9. for ν = i.42) (9. in general. therefore.39) Thus. There can be two such independent directions. We note from (9. the polarization vector characterizing the vector potential must be transverse to the direction of propagation of the plane wave. the three components of the vector potential. for ν = 0.23)) A0 ≡ 0. with k0 = Ek = √ k2 = |k |. or.43) Namely.44) ǫ∗ (k. for example. λ) = 0. (9.1. (9. 2. λ) · ǫ(k. λ′ ) = δλλ′ . as shown in Fig. will have plane wave solutions of the form A(x) ∝ ǫ(k) e∓ik·x .27) that. (9. in this case.38) vanishes identically (because of the choice of the gauge condition (9. complex satisfying k · ǫ (k. 9. λ = 1. since the vector potential is transverse. k · ǫ (k) = 0.

2) transverse to the direction of propagation k. This discussion is. Let us next note that . λ) .3 Field decomposition 337 k ǫ(k. therefore. λ) and a† (k. the field operator A(x) is Hermitian as it should be.44) (since they define a basis in the plane transverse to the direction of propagation). The vector nature of the field variable is now contained completely in the polarization vectors.46) Note that a(k. λ) (9.1: Two independent and orthogonal polarization vectors ǫ(k. 1) Figure 9. It is also clear that any transverse vector ˆ (to the direction k) can now be expressed as a linear superposition of the two independent polarization vectors (9. 1) and ǫ(k. we can decompose the vector field A(x) in the basis of the plane wave solutions as 2 A(x) = λ=1 d3 k (2π)3 2k0 ǫ(k. (9.9. λ) eik·x a† (k. Given this. 2) ǫ(k. where k0 = Ek = |k |. λ) are operators and as we have defined.45) +ǫ∗ (k. λ) e−ik·x a(k. quite analogous to that of the neutral Klein-Gordon field except for the presence of the polarization vectors.

λ) = −i (9. λ) 2k′ 0 0 +k ′ 0 )t +δ3 (k + k′ )ǫ∗ (k′ . λ)a(k′ . λ).49) leads to d3 x (2π)3 2k0 ← → (e−ik·x ∂0 A(x)) · ǫ(k. λ) = 1 0 √ ǫ(k. λ)ei(k +k)·x a† (k′ . λ) .48) 2 =−i λ=1 From (9.47) 2k0 λ=1 Similarly. λ=1 or. a(k. λ)e 2 a(k′ . λ)e2ik t a† (−k. (9. we can show that d3 x (2π)3 2 ˙ eik·x A(x) 0 k0 ǫ(k. (9.48). we conclude that i d3 x ← → 2k0 eik·x ∂0 A(x) = √ (2π)3 2k0 d3 x (2π)3 2k0 2 ǫ (k. Taking the Hermitian conjugate of (9. (9. λ)a(k. λ) = i ← → (eik·x ∂0 A(x)) · ǫ∗ (k. λ)a(k. λ)−ǫ∗ (−k. λ).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) + ǫ∗ (−k.49) where we have used the normalization for the polarization vectors in (9. λ)a(k. λ) .47) and (9. a† (k. λ). λ)e2ik t a† (−k. λ)ei(k 2 a† (k′ . λ) = λ=1 d3 k ′ √ δ3 (k − k′ )ǫ(k′ .50) .44). λ) + ǫ∗ (k′ .

P = λ=1 d3 k k a† (k.50) denote the two inversion formulae for the Maxwell field (analogous to (5.54) The Hilbert space for this theory can be built up in the standard manner.77) and (5. If we denote the vacuum state of the theory by |0 . λ)a(k. λ) and a† (k. Given the quantization conditions (see (9.3 Field decomposition 339 Equations (9. λ)a(k. [a(k. a† (k′ . j = −iδi TR (x − y). the operators a(k. (9.46)) that Ek = |k|.30)). The normal ordered forms for these operators can be shown to correspond to (see (5. The Hamiltonian and the momentum operators for the Maxwell theory can again be expressed in terms of the creation and the annihilation operators. λ). ˙ ˙ = 0 = [Ai (x). [Ai (x). λ′ )]. Aj (y)]x0 =y0 . λ). λ).49) and (9. a(k′ .51) using (9. Aj (y)]x0 =y0 (9.53) where we recall from the Einstein relation (see (9.50). then it satisfies .97) and we do not show the normal ordering symbol explicitly) 2 H = λ=1 2 d3 k Ek a† (k. λ) behave exactly like the annihilation and the creation operators for a harmonic oscillator system. λ′ )] = 0 = [a† (k.49) and (9.61) in the case of the Klein-Gordon field). λ). Aj (y)]x0 =y0 ˙ [Ai (x). λ′ )] = δλλ′ δ3 (k − k′ ). a† (k′ . λ). (9.9.52) In other words. (9. we can easily show that [a(k.

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

that the operators a† (k) = L a† (k) = R 1 √ 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.60) would then create photons which are left and right circularly polarized respectively. e 2 (9. It is clear. (These one photon states are eigenstates of helicity with eigenvalue ±1. 2) = ey . 1) − ia† (k. 2) . 1) − ia† (k. 2) λ=1 = 1 (ˆx + iˆy )(a† (k. 2) .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. we can identify a† (k. e 2 1 ˆ ǫR (k) = √ ex + iˆy . (9. 1) + ia† (k.61) = ǫ∗ (k)a† (k) + ǫ∗ (k)a† (k). 1) as creating a photon polarized along the x-axis and a† (k. ˆ ǫ(k. 1) + ia† (k. λ)a† (k. in this language.9.59) then. 2 1 √ a† (k. 2)) 2 (9. 1) = ex . 2 (9. 1) + ey a† (k. R L R L where we have used the familiar identification that 1 ˆ ǫL (k) = √ ex − iˆy . 2)) e e 2 1 e e + (ˆx − iˆy )(a† (k.62) describe respectively the left and the right circular polarization vectors. 2) as creating a photon polarized along the y-axis. 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. . λ) = ex a† (k.

λ)). λ). λ)ǫν (k′ .λ′ =1 1 (2π)3 d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ ×ǫµ (k.342 9.65) A(−) (x) = µ λ=1 From the commutation relations for a(k. if we define the two polarization vectors for λ = 1. λ).27)).λ′ =1 1 (2π)3 ′ d3 k d3 k ′ √ √ 2k0 2k′ 0 ×ǫµ (k. 3 2k 0 (2π) d3 k (2π)3 2k0 ǫµ (k. for simplicity. λ) in (9.63) and furthermore. λ). λ) and a† (k. This allows us to decompose the field into its positive and negative energy parts as 2 A(+) (x) µ = λ=1 2 d3 k ǫµ (k.45) as (recall (9. λ)+eik·x a† (k. ǫ(k. A(−) (y)] µ ν = λ.52). λ)ǫν (k′ . a† (k′ . λ)). λ) = (0.23)) 2 Aµ (x) = λ=1 d3 k (2π)3 2k0 ǫµ (k. λ′ )] 2 = λ. it follows that the only nontrivial covariant commutation relations have the form (these are not at equal time) 2 [A(+) (x). (9. λ′ )e−ik·x+ik ·y δλλ′ δ3 (k − k′ ) .59) then we can write the field decomposition (9.4 Photon propagator 9 Maxwell field theory Let us note that in the temporal gauge (9.23) (see also (9. if we choose the polarization vectors to be real as in (9. λ′ )e−ik·x+ik ·y [a(k. 2 as four vectors ǫµ (k.64) with k0 = |k|. (9. λ)eik·x a† (k. λ)e−ik·x a(k. (9. λ)(e−ik·x a(k.

A(−) (x)] = iG(+) (y − x) µ ν ν µ νµ = − 1 2 d3 k eik·(x−y) (2π)3 k0 2 ǫµ (k. λ)e−ik·(x−y) 2k0 2 = λ=1 1 (2π)3 = 1 2 d3 k e−ik·(x−y) (2π)3 k0 ǫµ (k. µν (9.145)). λ) λ=1 = −iG(+) (x − y). λ)ǫν (k. A(+) (y)] = −[A(+) (y). we obtain : Aµ (x)Aν (y) : +[A(+) (x). λ) λ=1 = −iG(−) (x − y). µν (9.67) Aµ (x)Aν (y) = 0|Aµ (x)Aν (y)|0 = −iG(+) (x − y). namely.68) From this. (9. µν [A(−) (x). . 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.9.66) which are the analogs of the positive and the negative energy Green’s functions for the Maxwell field (see (5.4 Photon propagator 2 343 d3 k ǫµ (k. λ)ǫν (k. λ)ǫν (k. A(−) (y)] ν µ : Aµ (x)Aν (y) : −iG(+) (x − y) µν : Aµ (x)Aν (y) : + Aµ (x)Aν (y).143) and (5.107)). we can define the Feynman propagator in the standard manner (see (6.

344 9 Maxwell field theory 0|T (Aµ (x)Aν (y)) |0 = iGF . for example. λ)ǫν (k. (9. λ) λ=1 0 0 0 0 0 0 1 θ(x0 − y 0 )e−ik (x −y ) + θ(y 0 − x0 )eik (x −y ) .70) 0 2k where we have changed k → −k in the second term and have used the fact that the polarization vectors are unaffected by this change of variables (see.µν (x − y) = 0|θ(x0 − y 0 )Aµ (x)Aν (y) + θ(y 0 − x0 )Aν (y)Aµ (x)|0 = −iθ(x0 − y 0 )G(+) (x − y) − iθ(y 0 − x0 )G(+) (y − x) µν νµ = −iθ(x0 − y 0 )G(+) (x − y) + iθ(y 0 − x0 )G(−) (x − y) µν µν (−) = i(−θ(x0 − y 0 )G(+) (x − y) + θ(y 0 − x0 )Gµν (x − y)).69) µν By definition.100) and write 1 0 0 0 0 0 0 θ(x0 − y 0 )e−ik (x −y ) + θ(y 0 − x0 )eik (x −y ) 0 2k = ǫ→0+ lim − dk′ 0 1 e−i(k +k )(x −y 2πi 2k0 k′ 0 + iǫ 0 ′0 0 0) − ei(k 0 −k ′ 0 )(x0 −y 0 ) k′ 0 − iǫ .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. 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. To simplify this expression. therefore. we use the integral representation for the step function given in (6. λ)ǫν (k.(9. (9. the Feynman Green’s function has the form GF .59)).

we obtain GF .71) where we have used the Einstein relation (9. Substituting this into (9.46).µν (k) = ǫ→0+ lim 1 k2 + iǫ 2 2 ǫµ (k. λ) . (9. λ)ǫν (k. we can identify the momentum space Feynman Green’s function for the Maxwell (photon) field to be GF .µν (x − y) = lim ǫ→0+ d4 k e−ik·(x−y) (2π)4 k2 + iǫ 2 ǫµ (k. λ) .73) where GF (k) denotes the Feynman Green’s function for a massless scalar field (see (5.70) and denoting the variable of integration k′ 0 → k0 . λ)ǫν (k. ˆ kµ = kµ − (k · η)η µ (k · η)2 − k2 = 0.9. 0).72) Therefore.73) depends on the polarization sum and to evaluate the polarization sum. 0. The Feynman Green’s function in (9. let us note that if we introduce two new four vectors η µ = (1. λ) λ=1 = GF (k) λ=1 ǫµ (k. λ)ǫν (k. λ=1 (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 ) .140)). 0. 2πi (k′ 0 )2 − k2 + iǫ ′0 0 0 0 1 1 − k′ 0 − k0 + iǫ k′ 0 + k0 − iǫ 0 ǫ→0+ lim − (9. k . |k| (9.74) .

physical calculations carried out with this propagator do lead to Lorentz covariant results. + iǫ iGF . λ)ǫν (k. η µ . as expected (since we are in the gauge A0 = 0). λ). λ) = −ηµν + ηµ ην − kµ kν . 2 define an orthonormal basis in four dimensions. we obtain the Feynman propagator for the photon field in the temporal gauge to be i ˆ ˆ (ηµν − ηµ ην + kµ kν ). (9. ˆ ˆ (ǫσ (k.76) or.346 9 Maxwell field theory ˆ then. λ) + σ + = ηµν .µν (k) = lim − ǫ→0+ k2 (9.75) Consequently. Substituting this back into (9.77) The Feynman propagator has a non-covariant look in this gauge. η µ ηµ = 1.µν (k) = 0. λ)ǫν (k. (9. λ) = 0. ˆ η µ ǫµ (k. we can write the completeness relation for these basis vectors as ˆ ˆ ηµ ην kµ kν ǫµ (k. Namely. λ=1 ˆ ˆ ǫµ (k. is manifestly transverse (see (9. ˆ ˆ kµ kµ = −1. kµ and ǫµ (k. λ) + ηµ ην − kµ kν = ηµν .73). namely. they satisfy ǫµ (k. λ) = 0 = η µ kµ . λ = 1. λ)ǫµ (k. λ)) (η ησ ) (kσ kσ ) λ=1 2 2 or. λ)ǫν (k. λ′ ) = −δλλ′ .75)). ˆ kµ ǫµ (k. (9.78) and. in this gauge. The Feynman propagator. kµ GF . also satisfies . λ)ǫσ (k. − 2 λ=1 ˆ ˆ ǫµ (k. However.

5 Quantum electrodynamics 347 GF .83)) and we have separated the Lagrangian density in (9. Aµ (x) → A′ (x) = Aµ (x) + µ ′ 1 ∂µ θ(x).9. The Lagrangian density for this theory is given by 1 L = − Fµν F µν + iψγ µ Dµ ψ − mψψ 4 1 = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ.74).82) The total Lagrangian density (9.5 Quantum electrodynamics (9.0µ (k) = 0. (9. 4 LI = −eψγ µ ψAµ .80) denotes the covariant derivative (see (7. which follows from (9. ψ(x) → ψ (x) = ψ(x)eiθ(x) .80) into the free and the interaction parts as 1 / L0 = − Fµν F µν + iψ∂ ψ − mψψ. 9.79) Quantum electrodynamics (QED) is the theory describing the interaction of electrons and positrons with the electromagnetic field. where Dµ ψ(x) = (∂µ + ieAµ (x))ψ(x).83) . e (9. 4 = L0 + LI .81) (9. (9.82)) can be checked to be invariant under the finite local gauge transformations ψ(x) → ψ ′ (x) = e−iθ(x) ψ(x).80) (or (9.

83) (with θ(x) = ǫ(x) =infinitesimal and keeping terms to linear order in ǫ) δǫ ψ(x) = ψ ′ (x) − ψ(x) = −iǫ(x)ψ(x). ψ.85) It follows from this that under a (finite) gauge transformation. Therefore. e (9. ψ . Aµ ) → LDirac (ψ ′ . Aµ ). we need to examine the invariance of the minimally coupled Dirac Lagrangian density in (9. ψ. In fact.84) where ǫ(x) denotes the infinitesimal local parameter of transformation. First. A′ ) µ ′ = iψ γ µ Dµ ψ ′ − mψ ψ ′ ′ ′ = iψγ µ Dµ ψ − mψψ = LDirac(ψ.80). we have already seen that the Lagrangian density for the Maxwell theory is invariant under the gauge transformation 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).83) (or (9.81) transforms covariantly under a gauge transformation (see (7. δǫ ψ(x) = ψ (x) − ψ(x) = iψ(x)ǫ(x).86) . let us note that the covariant derivative in (9. (9. the minimally coupled Dirac Lagrangian density transforms as ′ LDirac (ψ. = iψeiθ(x) γ µ e−iθ(x) Dµ ψ − mψeiθ(x) e−iθ(x) ψ (9.348 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. δǫ Aµ (x) = A′ (x) − Aµ (x) = µ ′ 1 ∂µ ǫ(x).

90) (2π)3 2k0 where we have chosen a real polarization vector as in (9.80) r µβ α q p = −(2π)4 ie (γ µ )βα δ4 (p+q +r).5 Quantum electrodynamics 349 Namely.89) lim − 1 ˆ ˆ i(ηµν − ηµ ην + kµ kν ) .23) (see also (9. k2 + iǫ ǫµ (k. µ k ν = iGF .64) so that the external line in (9.90) can represent both the annihilation as well as the creation of a photon. (9. ∂µ J µ (x) = 0. in QED we have the photon propagator as well as the photon external line factor which take the forms. λ µ = ǫ→0+ (9.106)-(8. The interaction vertex for QED can also be read out from the Lagrangian density (9.77)) correponds to the temporal gauge (9.91) . therefore.104) and in (8. J µ (x) = ψγ µ ψ.107). In addition. We have already derived the fermion propagator as well as the fermion external line factors in (8. the complete QED Lagrangian density (9.µν (k) = k. In this theory.83) and.88) The Feynman rules for QED can be derived in the standard manner.80) is gauge invariant. We note here that the photon propagator is a gauge dependent quantity.9.89) (or (9. (9. there is a conserved current.87) In momentum space the current conservation takes the form kµ J µ (k) = 0. (9. The form of the propagator in (9.27)). λ). the minimally coupled Dirac Lagrangian density is also invariant under the gauge transformation (9. (9.

λ′ ν µ k3 . λ µ k1 − k3 k1 .2. 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. λ Figure 9. As a result. 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 ).6 Physical processes 9 Maxwell field theory With these rules. . a) Photon pair production: For example. s2 k1 . (9. the amplitude for this process would simply be the sum of the amplitudes (quantum superposition principle) given by the two diagrams in Fig.350 9.2: The two lowest order Feynman diagrams that contribute to the pair production of photons as well as to the Compton scattering.2. the S-matrix element for this process can be worked out to be (we are assuming that the final state photons are outgoing and have carried out the momentum integration for the internal line and are omitting the overall energy-momentum conserving delta function δ4 (k1 + k2 − k3 − k4 )) k3 . λ′ k4 . s2 k4 . s1 ν + k1 − k4 k2 .92) To lowest order there are two Feynman diagrams which will contribute to such a process and they are shown in Fig. 9. We note here that experimentally. s1 k2 . but not the vertices (points) where each of the two photons was produced. we only measure two photons in the final state. Therefore. 9.

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

(9.90) is the same for the creation or the annihilation of a photon). this process is described by the two Feynman diagrams shown in Fig. we can obtain the S-matrix element for Compton scattering directly from the results in (9.93) to be e− (k3 ). (9. but .2 except for the labeling of the momenta. The amplitude differs from the case of the pair production of photons by only the external line factors. λ′ )|S (2) |e− (k1 ). γ(k4 )) to be outgoing. s3 )ǫ (k2 . λ′ )u (k1 . s1 ) / Q − m + iǫ / 1 ǫ (k4 .94) K = k1 − k4 and we have introduced Q = k1 + k2 . λ′ ) +u(k3 .93) (or (9. s1 ) . the internal line represents a photon propagator signifying that the electrons scatter by exchanging (emitting and absorbing) a photon. we will assume the final state electron and photon (e− (k3 ).352 9 Maxwell field theory are given to the lowest order by the same diagrams as in Fig. s3 )ǫ (k4 . γ(k4 . λ) / 1 ǫ (k2 . In this case. 9. Therefore.93). γ(k2 . Here.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 ). namely.98) To the lowest order.3. λ)u(k1 . Thus. instead of an electron and a positron being annihilated as was the case in (9.96) / ˜ − m + iǫ K / ˜ where as defined in (9. 9. this amplitude will differ from (9. λ) = − ie2 (2π)2 1 0 2k2 1 0 2k4 m 0 k1 m 0 k3 / × u(k3 . (9.96)) not only in the external line factors. 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.

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

s1 ) 2 + iǫ Q ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . e+ (k2 . e+ (k4 . We can calculate this matrix element as before and omitting the overall delta function representing conservation of energy-momentum. but in the present case we will treat the final state electron and the positron as outgoing (e− (k3 ). s1 ). the first diagram in Fig. s2 )γ µ v(k4 . s3 ). s1 ) . s4 )|S (2) |e− (k1 . k2 k4 k3 (9. s4 ) ˆ ˆ (ηµν − ηµ ην + Qµ Qν ) v(k2 .101) . the second diagram simply describes the scattering of an electron and a positron by exchanging a photon.354 9 Maxwell field theory e− (k1 ) + e+ (k2 ) → e− (k3 ) + e+ (k4 ).4: The two lowest order Feynman diagrams for Bhabha scattering. e+ (k4 )). s3 )γ µ v(k4 . s4 ) +v(k2 . s3 )γ ν u(k1 . The lowest order Feynman diagrams for this process are shown in Fig.3.4. we have e− (k3 . s2 )γ ν u(k1 . We see that topologically these diagrams are the same as in Fig.4 represents the annihilation of an electron and a positron to create a photon which subsequently creates an electron and a positron.100) k4 + k1 + k2 k1 k3 k1 k1 − k3 k2 Figure 9. 9. 9. In this case. On the other hand. 9. 2 + iǫ K (9. s2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k3 .

Here we would only give a very simple derivation of this relation in QED.9.7 Ward-Takahashi identity in QED 355 where as defined in (9. the Lagrangian density (9. 4 (9.103) 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. let us discuss very briefly the Ward-Takahashi identity in QED. let us note that the fermion propagator in QED can be represented as i = p−m / iSF (p) = p . This gives a flavor of low order calculations in QED. (9. 0) (iSF (p)) .80) for QED has the form 1 ¯ ¯ L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ. As we have seen. which we will study later. 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. of course. It follows from (9. independent of the choice of gauge. depends on the choice of gauge.103) where the i factor in the denominator is understood.104) . However. 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. e (9. 9. −p.97) K = k1 − k3 and Q = k1 + k2 .7 Ward-Takahashi identity in QED Before we discuss more complicated symmetries.102) The structure of the photon propagator in the theory.94) and (9.

∂pµ F (9. 0) denotes the three point vertex (9. e Consequently.91) in QED with a zero momentum photon (without the delta function and the factor of (2π)4 ). k = 0)(iSF (p)). −p. k = 0).107) ∂ (iSF (p)) (iSF (p))−1 ∂pµ = −i which can also be written as i 1 ∂ −1 S (p) = − Γµ (p.104) in QED resulting from the structure of the fermion propagator has the form ∂ ∂pµ p =− 1 e p k=0 p .105) 1 = − (iSF (p))Γµ (p.108) .104) can be diagrammatically represented as i i iγµ p−m / p−m / 1 e = − p k=0 p (9. −p. As a result the right-hand side of (9. (9. k = 0) = −(iSF (p))−1 e = ∂ (iSF (p))−1 ∂pµ ∂ −1 S (p). −p.104) we can obtain a relation between the fermion two point function and the three point vertex function as 1 Γµ (p. µ F ∂p e (9.106) In this case.356 9 Maxwell field theory where Γµ (p. a diagrammatic representation for the basic identity (9. −p. by inverting (9.

but we understand that the external momentum is channelled only through the fermion lines) + Figure 9. let us look at the fermion self-energy at one loop.5: Two loop corrections to the fermion self-energy in QED. (The trick in obtaining this relation in a simple manner lies in channeling the external momentum only through the fermion lines. Using the graphical identity in (9.) At two loop order the fermion self energy diagrams have the forms shown in Fig.106).5 (there are no external lines and we do not label the momenta for simplicity.108) holds for the one loop corrections to the amplitudes. we can write k ∂ ∂pµ p k+p p =− 1 e p k p k+p k+p q=0 (9.9. we see that (9.109) 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. For example. without the external lines). 9. Once again.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. .109) Here we are looking only at the proper vertex parts of the diagrams (namely. such relations are called Slavnov-Taylor identities which we will discuss later). The diagram on the left-hand side in (9. Such a relation can be seen to hold at any higher order in perturbation theory as well.

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

9.108) we conclude that these constants must be related as Z1 = Z2 .108) is a consequence of gauge invariance can be seen heuristically in the following way. (9. (9. Γµ −→ Z1 Γµ . D are constants and the dots denote other structures that may be induced due to quantum corrections. then from (9. Z2 are constants.7 Ward-Takahashi identity in QED 359 −1 −1 SF (p.113). That the Ward-Takahashi identity (9. Thus comparing with (9. then they must equal those present in the vertex correction graphs at the same order (we will study these issues in more detail within the context of renormalization later). (9. C.113) −eCψA − Dψψ + · · · /ψ where A.114) . m = 0). B. we can identify a = 1 + B = 1 + C. requires that the kinetic energy part of the fermion and the photon interaction term must combine into the form of a covariant derivative (see (9.111) where Z1 . (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 + · · · .85)) iaψD = iaψ(∂ + ieA /ψ / /)ψ.83).115) (9.112) This physically implies that if there are divergent parts in the fermion self-energy graphs at higher order. m = 0) −→ Z2 SF (p. on the other hand. Invariance under the local gauge transformations (9.

The canonical momenta conjugate to the field variables Aµ can be calculated from the Lagrangian density and take the forms (see (9.15). We recall that the Lagrangian density for Maxwell’s theory is given by 1 L = − Fµν F µν . we can choose a gauge condition and if we choose the gauge ∇·A = 0 (in our earlier discussion we had chosen the temporal gauge A0 = 0. (9.22)) ∂L = −F 0µ .117) as well as the action for this theory are invariant under the gauge transformation δAµ = ∂µ α(x).18)) . ˙ ∂ Aµ Πµ = which implies the constraint Π0 = −F 00 = 0.8 Covariant quantization of the Maxwell theory As we have seen.118) As we have already seen in (9. then the equations of motion (see (9. since the field strength tensor Fµν does not change under a shift of the vector potential by a gradient.116) We will discuss these topics in more detail when we study renormalization chapter 16. the present choice of gauge is known as the Coulomb gauge). the Lagrangian density (9. Let us recapitulate briefly how this arises. (9.21) and (9. 9.360 which is another way of saying 9 Maxwell field theory Z2 = 1 + B = 1 + C = Z1 . canonical quantization of Maxwell’s theory results in a lack of manifest Lorentz invariance. 4 Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .117) where Fµν denotes the field strength tensor. (9. As a result.

But. ∇2 A0 = 0. (9. but we lose manifest Lorentz covariance in the process (because all components of Aµ are not being treated on a equal footing). ∇2 A0 = −J0 . or. A0 = − 1 J0 . or. 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. We recognize that this is a consequence of our choice of gauge. remains manifestly Lorentz invariant in the canonical formalism. or. lead to (for ν = 0) ∂µ F µ0 = ∂i F i0 = 0.8 Covariant quantization of the Maxwell theory 361 ∂µ F µν = 0. or. ∇2 (9. we lose manifest Lorentz covariance in the intermediate steps (which is highly desirable in any calculation). in the Coulomb gauge ∇ · A = 0.120) In either case. The canonical quantization can now be carried out for the two physical degrees of freedom. if sources (charges and currents) are present. A0 = 0. or.119) On the other hand. would lead to ∂µ F µ0 = J 0 . however.9. Let us emphasize here that the final result for the calculation of any physical amplitude. ∂i ∂ i A0 − ∂ 0 Ai = 0. the dynamical equations ∂µ F µν = J ν . .

it projects on to the space of the components of any vector transverse (perpendicular) to the gradient operator ∂ µ .117) for the Maxwell theory also as 1 1 L = − Fµν F µν = Aµ P µν Aν + total derivatives.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 .362 9 Maxwell field theory The need for a choosing a gauge can be seen in an alternative manner as follows.122) (9. (9. This implies that even if we can quantize the theory (say. in the naive path integral formalism). 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. the inverse of P µν does not exist (this also means that the determinant of P µν vanishes) and the Green’s function and.124) so that this is the transverse projection operator. therefore.) Furthermore. (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. 4 2 where P µν = η µν − ∂µ∂ν . we cannot carry out calculations in perturbation theory. − ∂µ∂λ (9.123) With a suitable normalization. . (9. (In fact. the Feynman propagator for the theory cannot be defined. Consequently. namely.) We note that we can write the Lagrangian density (9. P = 1 P µν defines the normalized projection operator.

Therefore. 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) would appear to be different from Maxwell’s theory. for example. We can take an alternative approach. We can solve for the constraints and quantize only the true dynamical degrees of freedom. simply because the Green’s function does not exist.8 Covariant quantization of the Maxwell theory 363 We note here that whenever the determinant of the matrix of highest derivatives of the Lagrangian density vanishes.) This additional term breaks gauge invariance and consequently leads to a nonsingular theory. the system is singular and contains constraints among the field variables (as we will discuss in the next chapter). But more than that we have also added a term − 2 (∂µ Aµ )2 to Maxwell’s Lagrangian density. 4 2 where J µ represents a conserved current (9.125) ∂µ J µ = 0. in this procedure we lose manifest Lorentz covariance since we single out the transverse degrees of freedom. Let us consider. In such a case. without any further input. the Lagrangian density 1 1 L = − Fµν F µν − (∂µ Aµ )2 + J µ Aµ . But to understand the issue better. let us look at the equations of motion following from the action in (9. since we realize that the difficulties in quantization arise because of the singular nature of the Lagrangian density.126) Here we have generalized Maxwell’s theory to include a conserved 1 current. However. this gauge is known as the Feynman-Fermi gauge. we can try to modify the theory so as to make it nonsingular. (9. the Cauchy initial value problem cannot be uniquely solved.9. at this point there is no justification for adding this new term to the Lagrangian density. (This formulation of the theory is due to Fermi and considered as a gauge choice. But clearly the theory (9. Namely.125) .

we have ∂ν ∂µ F µν + or. ∂µ F µν + ∂ ν (∂ · A) = −J ν . Furthermore. An alternate way to see this is to note that if we take the divergence of the equations of motion in (9. (9.127) takes the form ∂µ (∂ µ Aν − ∂ ν Aµ ) + ∂ ν (∂ · A) = −J ν .127).364 9 Maxwell field theory ∂µ or.125) where χ = ∂ · A. ∂L ∂L − = 0. or. (9.132) . (9.127) Without the second term on the left-hand side in (9. or.127). ∂∂µ Aν ∂Aν −∂µ F µν − ∂ ν (∂ · A) − J ν = 0.129) ∂ · A = −∂ · J = 0. we recognize that if we restrict classically to the initial value conditions ∂χ = 0. or. (9. Aν = −J ν . (9. (∂ · A) = 0. Thus we see that although the presence of the term − 1 χ2 in the 2 Lagrangian density (9. (9. ∂t χ = 0. (∂ · A) = ∂ν J ν . If we now write out the left-hand side explicitly. therefore.131) seems to modify the theory. t = 0. χ is a free field and. the presence of this additional term in the Lagrangian density would not change the physics of Maxwell’s theory.128) (9.130) where we have used the anti-symmetry of the field strength tensor as well as the conservation of J µ . this is just the Maxwell’s equations in the presence of conserved sources.

125) with the supplementary condition ∂ · A = 0. 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.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). We now see that this quantization relation is exactly like the quantization condition for four distinct scalar fields (see (5.136) with all other commutators vanishing. Aν (y) x0 =y 0 = −iηµν δ3 (x − y). (9.8 Covariant quantization of the Maxwell theory 365 then χ = 0 at all times and we recover the familiar Maxwell’s theory. Thus. (9. φ(y) = iδ3 (x − y). classically we can think of Maxwell’s theory as described by the modified Lagrangian density (9. Πν (y)]x0 =y0 = iδµ δ3 (x − y).9.137) x0 =y 0 . or.134) It is now obvious that the coefficient matrix of highest derivatives is nonsingular in this case.133) (9.40) 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). ˙ ∂ Aµ Πµ = (9. ˙ Aµ (x). 2 (9. We can define the canonical momenta as ∂L ˙ = −Aµ .

the Hamiltonian) for this theory does not seem to be positive semi-definite since the time components (µ = 0) contribute a negative amount. However. the commutation relation between A0 ˙ and A0 has a relative negative sign. Let us construct the Hamiltonian density for this theory and see whether the above quantization relations are consistent (assume.366 9 Maxwell field theory except for one thing. be simply fixed by postulating that for the time component. But more serious than that is the fact that if we adopt the above convention. + 1 ′ ∂ Aν (x′ )∂ ′i Aν (x′ ) 2 i x0 =x′0 x0 =x′0 = − = − 1 2 d3 x′ Aµ (x). Πν (x′ )Πν (x′ ) d3 x′ iηµν δ3 (x − x′ )Πν (x′ ) x0 =x′0 ˙ = −iΠµ (x) = iAµ (x). of course. Namely. that would be against the spirit of Lorentz covariance since we no longer treat all the components of Aµ on the same footing. for simplicity.138) It is worth pointing out at this point that the Hamiltonian density (and. (9.139) . 2 2 (9. we note that if we calculate the commutator of the field variables with the Hamiltonian. the coordinate and the momenta exchange roles. H = Aµ (x). However. This problem can. the Hamiltonian for Maxwell’s theory would become unbounded from below. therefore. we find 1 d3 x′ − Πν (x′ )Πν (x′ ) 2 Aµ (x). J µ = 0 and we will neglect the total divergence terms) ˙ H = Πµ Aµ − L 1 = −Πµ Πµ + ∂µ Aν ∂ µ Aν 2 1 1 = −Πµ Πµ + Πµ Πµ + ∂i Aµ ∂ i Aµ 2 2 1 1 = − Πµ Πµ + ∂i Aµ ∂ i Aµ .

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. Thus.133) as yet. It is easy to see that the commutation relation (9. but the derivation needs to be done carefully in a limiting manner x0 → x′0 . The supplementary condition (9. namely.) As we would see shortly. or. even (9.140) in the quantum theory. not made use of the supplementary condition (9. where Aµ ’s are operators. we can weaken the supplementary condition (9.142) In other words.142) is a very stringent condition and we have to relax this further. however. the non vanishing right-hand side would lead to an inconsistency. We . = i∂ ν δ3 (x − x′ ). We have.141) x0 =x′0 ∂µ Aµ (x). = iη µν δ3 (x − x′ ). Πν (x′ ) or.136) are consistent. Πν (x′ ) ∂µ Aµ (x). (Note that since all field components are dynamical in this modified theory. whereas the left-hand side of the above expression would be zero if the supplementary condition (9. (9. we can think of the supplementary condition as selecting out the subspace of the physical Hilbert space of the theory. = i∂ ν δ3 (x − x′ ).) We note that. A little bit of thinking tells us that although in the classical theory we can impose the condition ∂µ Aµ = 0.133) and say that this condition is true only on the space of physical states of the theory. such a condition is hard to implement as an operator condition. (9. (9.136) would lead to Aµ (x). 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.133) were to hold as an operator equation. Πν (x′ ) x0 =x′0 x0 =x′0 (This relation is essentially correct.9. ∂µ Aµ |phys = 0.133) may be thought of as picking out the smaller physical subspace from the larger total vector space.

λ). λ) . Πν (x′ ) |ψ or. k0 = ω(k) = |k|. λ)fk (x)a(k. λ′ ) . let us make a plane wave expansion for the field variables Aµ (x) = λ ∗ + ǫ∗ (k. 3. (9.) They are normalized as 1. µ d3 k ǫµ (k. 3.141). a† (k′ .143) which is inconsistent.368 9 Maxwell field theory can see this already from the commutator relation (9.74)). a† (k′ . λ) as annihilation and creation operators satisfying a(k. λ)fk (x)a† (k. ˆ while that for λ = 3 is longitudinal (recall kµ ) and the other two polarization vectors are transverse to the momentum four vector. x0 =x′ 0 = ψ|i∂ ν δ3 (x − x′ )|ψ . we note that if |ψ represents a physical state. To proceed further. (9. 2. a(k′ . (Unlike the earlier discussion. λ). (9. λ = λ′ = 0. 2. λ′ ) = 0 = ′ a† (k.145) Here ǫµ (k.146) ǫµ (k. λ = λ′ = i = 1. λ) and a† (k. then ψ| ∂ · A(x). Basically. λ) (9. 0 = i∂ ν δ3 (x − x′ ). −1. λ′ ) a(k.147) = −η λλ δ3 (k − k′ ). λ)ǫ∗µ (k.144) where fk (x) = 1 (2π)3 2ω e−ik·x . 1. λ)’s are the components of the polarization vector for a photon travelling along k and polarization index λ = 0. here they are not required to be transverse. we can always choose a basis such that the polarization vector for λ = 0 is time-like (recall η µ ). At this point we can quantize the coefficients a(k. λ′ ) = η λλ = ′ As we have seen earlier (see (9. λ). . (9.

λ)ǫν (k′ . λ′ ) µ = λ. λ′ = 0. a† (k′ . λ)ǫ∗σ (k. λ′ )fk (x)fk′ (y) a(k. λ′ )fk′ (y)a(k′ . λ)ǫ∗ (k. 2. λ)ǫ∗ (k′ . Aν (y) = ∗ ∗ d3 k(−iω)ηµν [fk (x)fk (y) + fk (x)fk (y)] = −iηµν δ3 (x − y). Aν (y) = λ. λ)ǫ∗ (k′ .148) Using the completeness relation for the polarization vectors ǫµ (k. we do not write the equal-time arguments explicitly although they are assumed) ˙ Aµ (x). λ)ǫ∗ (k. λ′ )fk (x)fk′ (y) a† (k. a(k′ .λ′ d3 kd3 k′ iω ′ ∗ × ǫµ (k.148). λ) ∗ µ f (x)fk (y) . we obtain the equal-time commutator to be ˙ Aµ (x). λ′ )fk′ (y)a† (k′ . λ′ ) ν ∗ −ǫ∗ (k. λ) k (9. ǫσ (k.λ′ d3 kd3 k′ (−iω ′ η λλ )δ3 (k − k′ ) ′ ∗ ∗ × ǫµ (k. λ′ ) − ǫ∗ (k′ . we have to check that these conditions are consistent with the equal-time quantization conditions for the field variables (for simplicity. λ)ǫ∗σ (k. (9. λ)fk (x)a† (k. ǫσ (k. λ) + ǫ∗ (k. λ)ǫν (k′ .150) . 1. λ)ǫ∗σ (k. Of course. λ) ν = ηµν . λ) (9. λ). λ) + ǫ∗ (k. λ) ν ∗ fk (x)fk (y) ǫσ (k. λ). µ = λ.λ′ ∗ −iω ′ ǫν (k′ . λ′ )fk (x)fk′ (y) + ǫ∗ (k. λ)fk (x)a(k.9. λ′ )fk (x)fk′ (y) ν µ = λ d3 k(−iω) ǫµ (k. λ)ǫν (k. λ′ ) ν ∗ d3 kd3 k′ ǫµ (k.149) λ in (9.8 Covariant quantization of the Maxwell theory 369 where λ. λ). 3.

a(k. 2. 3. for the time-like component there is a relative negative sign in the commutation relation (9. If we calculate the normal ordered Hamiltonian of the theory (recall J µ = 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.154) . 3) . it is strange to note that there are four independent photon degrees of freedom.152) Let us next consider the state with one time-like photon |1. 1) +a† (k.147) that although a(k. We also note from (9. λ = 0)|0 . it has the form :H: = d3 k ω(k) −a† (k. For example. 2)a(k. 2. 1)a(k. (9. 0)a(k. 3)a(k. λ) behave respectively like annihilation and creation operators for λ = 1. 3. The norm of this one photon state can now be calculated (9. λ) and a† (k. λ = 0. λ)|0 = 0. 0) + a† (k. let us denote by |0 the vacuum state of the theory so that 0|0 = 1.370 9 Maxwell field theory Thus the commutation relations for the creation and the annihilation operators are indeed consistent with the quantization condition for the field variables. However. 1.147). λ = 0 = d3 k F (k)a† (k.153) where F (k) is a suitable smearing function such that d3 k |F (k)|2 < ∞. (9. (9. 2) + a† (k.

1)a(k. 0)|0 . Thus the natural modification that comes to mind is to interchange the roles of these operators for the time-like photon. However. λ = 0 = = = − d3 kd3 k′ F ∗ (k)F (k′ ) 0|a(k. 0) − δ3 (k − k′ )|0 d3 k |F (k)|2 < 0. (9. 0) + a† (k. (9. 3) .157) In this case. Let us suppose that a† (k. 0 = a(k. 3)a(k. 2)a(k. 2) + a† (k.9.158) and the energy of this one photon state is given by . the normal ordered Hamiltonian will have the form (compare with (9. 1) +a† (k. 0)|0 d3 kd3 k′ F ∗ (k)F (k′ ) 0|a† (k′ . We also realize that this negative norm is a consequence of the negative sign in the commutator (9. 0)a† (k.8 Covariant quantization of the Maxwell theory 371 1.156) |k. so that the one time-like photon state is given by (9. 0)a(k. 0)a† (k′ .147) for the annihilation and the creation operator for a time-like photon.155) This shows that the vector space of the theory has an indefinite metric – the norm for the one time-like photon state is negative although states containing only space-like photons have positive norm. (9. this also leads to trouble as we see below. 0)|0 = 0.151)) :H: = d3 k ω(k) −a(k. λ = 0|1.

as discussed earlier this is too stringent a condition to allow any state of the radiation field. 0)]|0 = d3 k′ ω(k′ )[−a(k′ . n3 |m0 . This is seen by using the supplementary condition (subsidiary condition) which we still have not imposed. 0)a† (k′ . n1 . In a free theory we may impose suitable conditions to prohibit any unwanted state. 0 = Ha(k. 0 . where we have used the fact that H|0 = 0. m3 = (−1)n0 δn0 m0 δn1 m1 δn2 m2 δn3 m3 .372 9 Maxwell field theory H|k. Thus states containing an odd number of time-like photons have negative norm.159) = −ω(k)a(k. m2 . So we are stuck with an indefinite metric space (with the standard interpretation for annihilation and creation operators for the timelike photon) and the normalization of states in this space is given by (we suppress the momentum label for simplicity) n0 . 0)|0 = −ω(k)|k. In fact it is obvious that with this definition. the physical states must satisfy ∂µ Aµ (x)|phys = 0. a(k. 1. definite states of arbitrarily high negative energy values are possible and the Hamiltonian becomes unbounded from below.161) However. (9. Thus we see that if we exchange the roles of the creation and the annihilation operators for the time-like photon.160) where nλ . As we will see shortly in spite of the presence of negative norm states. n2 . 2. This not only demands that certain . The question that immediately comes to our mind is what happens to the probabilistic interpretation of the theory. physical results are well behaved. 0)]|0 (9. the energy of the state with one time-like photon becomes negative. m1 . 0). Namely. a(k. But in the presence of interactions such states may be excited. (9. 0)|0 = [H. 3. mλ denote number of photons for each of the polarizations λ = 0.

but it also requires that those photons cannot be emitted. λ or. (This is commonly known as the Gupta-Bleuler quantization.) We remark here that since (see (9. Recalling that kµ ǫµ (k. 3) |phys = 0. or.74) as well as the fact that for positive energy photons k0 = ω = |k|. λ)|phys = 0.129) or (9. it can be decomposed into positive and negative frequency parts uniquely in a relativistically invariant manner and this decomposition is preserved under time evolution.8 Covariant quantization of the Maxwell theory 373 kinds of photons are not present in the physical state. Therefore.130)) ∂µ Aµ (x) = 0. (9. 0) + ǫµ (k.165) (a(k.9. Let Vphys = {P } denote the set of all states which satisfy the supplementary condition and let |ψ represent such a state. 3)a(k. λ = 0) − k2 a(k. or. 0)a(k. as we have seen.163) ∂µ Aµ (x) is like a free scalar field.164) the Gupta-Bleuler supplementary condition leads to kµ ǫµ (k. λ = 3) |phys = 0. where we have used the definitions and the properties of the polarization vectors in (9. |k| (9. the supplementary condition (9. 3))|phys = 0. 0) − a(k. 2. Gupta and Bleuler weakened the supplementary condition to have the form ∂µ Aµ (+) (x)|phys = 0. k0 a(k. (9. λ) = 0 for λ = 1. Then. λ)a(k.162) where ∂µ Aµ (+) (x) is the positive frequency part of the divergence of the vector potential (Maxwell’s field) and contains only the destruction operator or the annihilation operator. kµ ǫµ (k. (9.165) implies that .

n2 . n2 .168) This can be seen as follows or. All the coefficients Cn0 .n1 . 3)|ψ .n1 .168) with the condition C0. of course.n3 −1 + n3 Cn0 −1. (9. 3)a(k.n2 .n1 .168). 0) − a(k. Cn0 .n3 −1 + n3 Cn0 −1. n1 .n2 .n1 . a superposition of different states of the form |ψ = n0 . n3 . n2 . 0)|ψ = ψ|a† (k.n1 .n3 in the expansion can be determined recursively from (9. n1 . n2 .n2 .n3 − n0 |n0 − 1.169) which leads to the relation in (9. n1 . n3 − 1 = 0. √ √ ( n0 Cn0 . n3 √ − n3 |n0 .n1 . or.n2 . (9. n2 .n2 . √ Cn0 .167) where nλ denotes the number of photons with polarization λ and the supplementary condition relates the number of time-like and longitudinal photon states as √ √ n0 Cn0 .n1 .n2 .n1 . n3 = 0.n1 .n3 (a(k.n2 .0 = 1.n3 = 0.n3 |n0 .170) . 0)a(k. (9. A general physical state is.n2 . n1 . (9. 3)) |n0 .n2 . 0)|ψ = a(k. n2 .166) This leads to the fact that the physical states must contain superpositions of states with an equal number of time-like and longitudinal photons. ψ|a† (k.374 9 Maxwell field theory a(k.n3 ) ×|n0 − 1. or. n1 .n3 Cn0 . (9.n1 . n3 − 1 = 0. 3)|ψ .

n1 . n1 . n2 . n1 . For example.n1 . (9. 0 .9. we have φ1 |φ1 φ2 |φ2 = 0. n1 . 2 − 2|1. n2 of transverse photons as a linear superposition of states of the form |φ because φ|φ = |φ0 + b1 |φ1 + b2 |φ2 + . √ = |0. n2 . 1 + 1. 0|1. n2 . . = 0. 0 = 1 − 1 = 0.174) we can assume that in a truly physical state of the radiation field there are no time-like and longitudinal photons present. n2 . Let us note that because of the negative norm of states containing an odd number of time-like photons. n1 . (9. n1 . 0 . 1 + |2. n2 . n2 . n2 .173) (9. n1 . n2 . = φ0 |φ0 .8 Covariant quantization of the Maxwell theory 375 It follows now that for a fixed number n1 . n1 . In other words. 1|1. n2 . 1 + 2. 2 + 2 1. 0 . n1 . (9. n2 of transverse photons. . n1 . n2 . n2 . (9. n1 . Thus although we can write a general physical state with a fixed number n1 . n1 . 0 = 1 − 2 + 1 = 0.171) and so on where we have used the values of Cn0 .n2 . n2 . n2 . n1 . the physical states allowed by the supplementary condition have the forms |φ0 |φ1 |φ3 = |0. n2 . 2|0. n1 . = |0. n0 = n3 = 0.172) and so on. n1 . 1|0. 1 − |1.175) . 0|2. n2 . all such states except for |φ0 have zero norm. n1 . . n2 .n3 ’s determined recursively.

Vphys .165) contains states with positive semi-definite norm (negative norm states are eliminated by the supplementary condition or the physical state condition and. Since the zero norm states are orthogonal to all the states including themselves.176) where |ψ represents a physical state.µν (k) = lim − ǫ→0+ k2 iηµν .77) or (9.89) simply because this can be thought of as quantizing the theory in a different gauge (FeynmanFermi gauge).376 9 Maxwell field theory in such states. + iǫ (9. (9. such states decouple. The physical subspace of the theory selected by the supplementary condition (9. are gauge . on the space of physical states since ψ|∂µ Aµ |ψ = 0.178) where V0 represents the set of states with zero norm. we have the true physical subspace of the theory where the norm of states is positive definite. 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. namely. V0 V phys = (9. A different way of saying this is to note that being orthogonal to every other state. consequently. if we further mod out the states by the zero norm states. If interactions are present time-like and longitudinal states may occur as intermediate states. there is no problem with a probabilistic interpretation).179) This is different from (9. however. The physical S-matrix elements.177) (9. However their effect cancels out in the final result. 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. Without going into details we note here that the Feynman propagator for the photon in this theory has the simpler form iGF .

2. New York. Bjorken and S. Relativistic Quantum Fields. Nauka. We will develop these ideas further when we study the covariant quantization of non-Abelian gauge theories in chapter 13. Gross. Physical Review 78. 8. N. K. 1964. Helvetica Physica Acta 23. Itzykson and J-B. 371 (1957). Row. Schweber. 1980. Introduction to the theory of Quantized Fields. C. John Wiley. Bleuler. 3. 182 (1950). Nuovo Cimento 6. New York (1993). Proceedings of the Physical Society (London) A63. Quantum Field Theory.9 References 1. Roman. S. D. N. Y. Gupta. Introduction to Relativistic Quantum Field Theory. 681 (1950). 4. 7. John Wliley. Shirkov. J. Relativistic Quantum Mechanics and Field Theory. 6. Introduction to Quantum Field Theory. Evanston (1961). 10. Bogoliubov and D. Moscow (1984). . Peterson. McGrawHill. New York. S. Zuber. 5.9 References 377 independent and do not depend on the form of the photon propagator in a particular gauge. 9.9. Ward. Drell. V. C. 567 (1950). F. New York (1969). McGrawHill. P. Takahashi. N. 9. J.

This page intentionally left blank .

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

qi ) → (qi .2) and uniquely define the Hamiltonian of the system through the Legendre transformation H(qi . the time evolution of any dynamical variable A(qi . The phase space spanned by the independent coordinates and momenta is also 2N dimensional with a (equal time) canonical Poisson bracket structure {qi .380 10 Dirac method for constrained systems (qi . ˙ (10. using the definition (10. qj } = 0 = {pi . ˙ ˙ (10.5) Using (10. pi ) as ∂A1 ∂A2 ∂A1 ∂A2 − . H} = − ˙ which can be shown. pi ). it follows now that the 2N first order dynamical equations (evolution equations) for the system can be written in the Hamiltonian form ∂H . pi ). ∂pi ∂H .4) which allows us to write the Poisson bracket between any two dynamical variables A1 (qi . (10. to be equivalent to the N second order Euler-Lagrange equations following from the Lagrangian description. pi ) = pi qi − L(qi . pj } = δi = −{pj . pi ) in the phase space can be written in the Hamiltonian form .3). qi }.3) where summation over repeated indices is understood. j {qi . pj }. ∂qi ∂pi ∂pi ∂qi {A1 .5).4) or (10. A2 } = (10. H} = ˙ pi = {pi . ∂qi (10. qi ). In fact.6) qi = {qi . A2 (qi .

θα ). If the relations (10.11) The 2M dimensional phase space is endowed with a canonical Poisson bracket structure (with the convention of left derivatives) {θα . θα }. pi ) = {A(qi .10) through the Legendre transformation (this is consistent with the convention of left derivatives) ˙ ˙ ˙ ˙ H(θα . θα ) = −Πα θα − L(θα . Πβ }. (10.5). Πα ) = θα Πα − L(θα . pi ). α = 1. For example. 2. (10.9) can be inverted to express velocities in terms of the momenta.7) This discussion can also be generalized to a classical system described by Grassmann (anti-commuting) variables. unlike the bosonic case in (10. if ˙ we have a classical system described by the Lagrangian L(θα . (10. Πα ). (10. . we have to define the derivative in (10. since the dynamical variables are now of Grassmann odd (fermionic) nature. β {θα .9) and as discussed in chapter 8 (see (8. However.9) However.12) The Poisson bracket between any two dynamical variables can now be obtained from (10.10. H}. ˙ ∂ θα (10.14)).1 Constrained systems 381 ˙ A(qi . θα ) → (θα . θα ).8) then. · · · M with θα θβ = −θβ θα . we can define the momenta conjugate to the coordinates as Πα = ∂L . we can go from the configuration space to the phase space ˙ (θα . (10. Πβ } = −δα = {Πβ . θβ } = 0 = {Πα .12). we will choose the convention of taking derivatives from the left.

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

it follows that the transformation is invertible (A−1 exists) if ∂2L = 0. (10.20) Therefore.16) where the elements b.1 Constrained systems 383 unique Hamiltonian of the system. However.21) . it follows that b = b(q). In this case. ∂ qj ˙ ∂ qj ∂ qi ˙ ˙ (10. the transformation to phase space in (10.10. (10.18) so that A−1 exists only if the matrix b is invertible. b ˙ so that ∂pi ∂2L = .16) can be easily seen to have the form A−1 = ½ 0 −b−1˜ b−1 b . ∂ qi ∂ qj ˙ ˙ det (10.2) can be written in general in the matrix form as q p q q ˙ =A = ½ 0 ˜ b b q q ˙ .16) we note that pi = ˜ij (q)qj + bij (q)qj . the difficulty in passage to a Hamiltonian description arises when this transformation is noninvertible. ˜ denote N × N matrices. (10.19) bij (q) = (10. For conventional b theories described by Lagrangians with quadratic velocity terms. ˜=˜ b b(q). To understand the difficulty. let us concentrate on the bosonic theory for simplicity. From (10.17) The inverse of the matrix A in (10.

∂ qi ∂ qj ˙ ˙ (10. it is not clear a priori how to define the Hamiltonian uniquely.22) then. ∂ qi ∂ qj ˙ ˙ det (10.384 10 Dirac method for constrained systems The naive passage to the Hamiltonian description can be carried out only in this case. b = 1. 10. therefore. pb ). therefore.2 Dirac method and Dirac bracket In a physical system with constraints. ˙ a.24) and the other N − R velocities cannot be determined. we can write . On the other hand. we can determine only R of the N velocities in terms of coordinates and independent momenta as qa = fa (qi . It follows from (10. (10. In this case.2) is not invertible. if the Lagrangian describing the theory satisfies ∂2L = 0. 2. there exist relations or constraints between various dynamical variables and such systems are known as constrained systems. the passage to a Hamiltonian description is achieved systematically through the method of Dirac which we describe in this section. R. 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. may not represent the true brackets necessary for a Hamiltonian description of the system.4) need not satisfy the constraints of the theory and. In other words. Let us assume that the N × N matrix ∂2L . As a result. Furthermore.23) of the physical system is of rank R < N . the transformation (10. · · · . it is also clear that the naive canonical Poisson brackets in (10.

if (F − G) = 0. a.2) (presence of constraints). the Hamiltonian of the theory is no longer unique.28) that in spite of the fact that qα cannot be determined. (10. namely. F ≈ G.2 Dirac method and Dirac bracket 385 pa = ga (qi . R. On the other hand. ˙ ∂L ∂Hcan = gα − = gα − pα ≈ 0. (10. pa ). Hcan = Hcan (qi . (If there are further constraints in the theory.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. qi ). · · · . we note that the constraints in (10. ˙ pα = gα (qi . G in Γ are said to be weakly equal.25) It is clear that the second set of equations in (10. namely. the dimensionality of Γc would be further reduced. from the canonical . b = 1. ∂ qα ˙ ∂ qα ˙ (10.29) so that on the constrained hypersurface Γc . qi ) = pa qa + gα qα − L(qi . 2. the canonical Hamiltonian does not depend on velocities as we would expect.28) where we have used (10. qb ). if they are equal on the constrained hypersurface Γc . α = R + 1. For example. R + 2. pa ) = 0. ˙ ˙ ˙ ˙ ˙ (10. In fact.27) Γc As usual. We note from (10.25). · · · . we can define the canonical Hamiltonian of the theory as the Legendre transform Hcan = pi qi − L(qi . (10.25) define N − R constraints among dynamical variables which we denote by ϕα = pα − gα (qi . we also note that because of the non-invertibility of (10. pa ). N.26) and we recognize that such constraints will reduce the dimensionality of the true phase space of the system.) Two dynamical variables F.10.

The time evolution of any dynamical variable F can now be written as ˙ F ≈ {F. ϕβ } ≈ 0. Hp } = ˙ pi ≈ {pi .32) qi ≈ {qi . Hcan } + λβ {ϕα .4).33) which remain undetermined. (10.35) . with the canonical Poisson brackets in (10.30) with velocities λα ≈ qα . ˙ (10. (10. Hcan + λβ ϕβ } ˙ = {ϕα . Hp } = {F. ∂pi ∂pi ∂Hp ∂(Hcan + λα ϕα ) =− . (10.34) this leads to ϕα ≈ {ϕα . (10. we can write the Hamiltonian equations as ∂Hp ∂(Hcan + λα ϕα ) = . λβ }ϕβ ≈ {ϕα . ∂qi ∂qi (10. Hp } = − ˙ and using (10. it follows that we can identify the Lagrange multipliers in (10.34) The constraints of the theory should be invariant under time evolution and using (10. Hcan + λα ϕα }. Hcan } + λβ {ϕα .31) Furthermore.30) where λα denote undetermined Lagrange multipliers and it follows that Hp ≈ Hcan .26)) Hp = Hcan + λα ϕα . ϕβ } + {ϕα .26) as (it is known as the primary Hamiltonian since it incorporates only the primary constraints in (10.29). we can define the primary Hamiltonian associated with the system by incorporating the primary constraints of the theory (10.386 10 Dirac method for constrained systems Hamiltonian of the theory.

if there are primary constraints in the theory which are first class. n1 . 2. then there remain undetermined Lagrange multipliers in the primary Hamiltonian after all the . .38) Here we have used the observation due to Dirac that there are an even number of second class constraints in a theory and we note from (10.2 Dirac method and Dirac bracket 387 where we have used (10. 1.10.36)-(10. We continue with this process until all the constraints are determined to be evolution free. α = 1. α = 1. (10. secondary. In general. (10. the constraints which have at least one nonvanishing Poisson bracket with the constraints are known as second class constraints and are denoted as b φα ≈ 0.35) may determine some of the Lagrange multipliers or may lead to new constraints known as secondary constraints. · · · . all of the Lagrange multipliers may be completely determined or some of them may remain undetermined. The constraints which have weakly vanishing (see (10. α = 1. (10. 2n2 . Let us denote all the constraints of the theory (primary. .38) that n1 + 2n2 = n < 2N .35) (with Hp as the Hamiltonian) may determine some other Lagrange multipliers or generate newer constraints (tertiary etc). Requiring the secondary constraints to be invariant under time evolution as in (10. tertiary. 2.36) which can be divided into two distinct classes. n < 2N. 2.26) in the last step. · · · . It is also worth noting that in this process of determining all the constraints that are time independent (evolution free). · · · .37) 2. ) collectively as ϕα ≈ 0. It is clear that (10.27)) Poisson bracket with all the constraints (including themselves) are known as first class constraints and are denoted as e ψ α ≈ 0. On the other hand. .

2(n1 + n2 ) < 2N. vanish {F. First. (The number of undetermined Lagrange multipliers equals the number of first class constraints among the primary constraints. (10. Namely. collectively as φA ≈ 0. · · · . in this case. requires gauge fixing conditions and we must choose as many gauge fixing conditions as there are first class contraints. φA } ≈ 0. in general. a consistent Hamiltonian description. / (10. which are all second class. 2. even though φA ≈ 0. let us note that we can construct the matrix of Poisson brackets of all the constraints in (10.4) are not compatible with them.37) into second class constraints. 2.388 10 Dirac method for constrained systems constraints have been determined. n1 .) The first class constraints have a special significance in that they are associated with gauge invariances (local invariances) in the theory. (10.40) as . we cannot yet impose them directly in the theory since the canonical Poisson brackets in (10.39) and are chosen such that they convert the first class constraints (10. Thus.42) This issue of incompatibility of the Poisson brackets can be addressed through the use of the Dirac brackets which are constructed as follows. As we have seen in the case of Maxwell field theory. (10.40) so that the true phase space of the theory is 2(N − n1 − n2 ) dimensional. A = 1. Although we have identified all the constraints of the theory.41) the Poisson bracket of the constraints with any dynamical variable F does not. α = 1. The gauge fixing conditions are generally denoted as e χα ≈ 0. after gauge fixing we denote all the constraints of the theory (including the gauge fixing conditions). · · · .

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

therefore. 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. 10. 2. the inverse needs to be defined carefully with the appropriate boundary condition relevant for the problem. If we denote the coordinate of the particle as qi . · · · . ∂F 2 (10. It is also worth emphasizing here that in dealing with field theories where the matrix C AB is coordinate dependent. i = 1.47) where summation over repeated indices is understood.43) and its inverse. · · · . F into qa = (qi . 2. then the matrix of highest derivatives has the form . let us consider the motion of a point particle constrained to move on the surface of a ndimensional sphere of radius unity. 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. Namely. We note that if we combine the dynamical variables qi . ˙ ˙ 2 L= (10. n + 1.48) where F is a Lagrange multiplier field (an auxiliary field without independent dynamics) whose Euler-Lagrange equation 1 ∂L = − (qi qi − 1) = 0. F ). rather than dealing with a large matrix of Poisson brackets in (10. then the coordinates are constrained to satisfy (the metric in this simple theory is Euclidean and.3 Particle moving on a sphere As a simple example of constrained systems. (10. We note that the dynamics of the system can be described by the Lagrangian 1 (qi qi − F (qi qi − 1)) .47) on the dynamics.49) gives the constraint (10. raising and lowering of indices can be carried out trivially) qi qi = 1.390 10 Dirac method for constrained systems that it can be constructed in stages. n. when the number of constraints is large. a = 1.

To understand the passage to the Hamiltonian description in this simple case.51) has the form 1 i i ˙ Hcan = pi qi + pF F − L = ˙ p p + F (qi qi − 1) . pF } = 0.3 Particle moving on a sphere 391 ∂2L = ∂ qa ∂ qb ˙ ˙ δij 0 0 0 . F } = {pi . we note that the conjugate momenta of the theory are given by ∂L = qi . (10. 2 (10.54) The equal-time canonical Poisson brackets of the theory are given by j {qi .53) Hp = Hcan + λ1 ϕ1 = (10. the theory has a primary constraint given by ϕ1 = pF ≈ 0.48) and (10. pF } = {pi .55) . (10. pj } = δi . 2 leading to the primary Hamiltonian (see (10. pj } = {F.22)) reflecting the fact that there are constraints in the theory and the naive Hamiltonian description would not hold. {F. qj } = {pi . ˙ ∂ qi ˙ ∂L = 0.52) The canonical Hamiltonian following from (10.30)) 1 i i p p + F (qi qi − 1) + λ1 pF . pF } = 0. ˙ ∂F pi = pF = (10. {qi .50) which is indeed singular (see (10.51) Therefore. pF } = 1. F } = {qi . F } = {pF . (10. {qi .10.

requiring the primary constraint to be independent of time leads to ϕ1 ≈ {ϕ1 . (pj pj + F (qj qj − 1) + λ1 pF )} 2 ≈ pi pi − F ≈ 0.59) Requiring the new constraint (10. 2 (10. Hp } ˙ 1 = {qi pi . pj } = qi pi ≈ 0. qj } = pi pi − F qi qi (10. In particular.392 10 Dirac method for constrained systems With these we can now determine the time evolution of any dynamical variable. 2 (10. Hp } ˙ 1 = {pF . = pi pj {qi . we obtain ϕ3 ≈ {ϕ3 . Hp } ˙ 1 1 = { (qi qi − 1). (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) ≈ 0.58) which generates a new constraint ϕ3 = qi pi ≈ 0.57) Requiring (10. pj } + qi F qj {pi .56) Let us denote this secondary constraint as ϕ2 = 1 (qi qi − 1) ≈ 0. (10.60) . (10.57) to be time independent.59) to be time independent. we obtain ϕ2 ≈ {ϕ2 .

In fact. It is easy to check that all the constraints of the theory (10.57).61) to be time independent leads to ϕ4 ≈ {ϕ4 .52). 2. pF } = −2F qi pi − λ1 ≈ −λ1 ≈ 0. pj pj − F } = 2pi pj {qi . ϕ4 } = {qi pi . ϕ2 . 3.62) = 2pi F qj {pi . ϕ2 }. which determines the Lagrange multiplier and the chain of constraints terminates. qj pj } = qi qj {qi .61) are second class. (10. Hp } ˙ (10. Collecting all the constraints into φA = (ϕ1 . φB } = C AB .10. A = 1. .3 Particle moving on a sphere 393 so that we can identify ϕ4 = pi pi − F ≈ 0. pj } = 2pi pi = 2p2 = −{ϕ4 . (pj pj + F (qj qj − 1) + λ1 pF )} 2 (10.64) = −{ϕ3 .63) where we have identified pi pi = p2 . the nontrivial equal-time Poisson brackets between the constraints take the forms {ϕ1 . ϕ4 ). = −{ϕ4 . ϕ3 . (10. 4. Requiring (10. pj } = qi qi ≈ 1 2 {ϕ3 . which takes the form (10. ϕ1 }. ϕ3 } = { (qi qi − 1).59) and (10. we can calculate the matrix of the (equal-time) Poisson brackets of constraints {φA . pi pi − F } = 1 1 {ϕ2 . qj } − λ1 {F. ϕ4 } = {pF .61) 1 = {pi pi − F. (10. ϕ3 }.

(10.66).65) −1 CAB The Dirac brackets between any two dynamical variables can now be defined as (see (10. −1 {qi . qk pk }C32 { (qℓ qℓ − 1). {F. 1 −1 {pi . {qi . pj }D = {qi . (qk qk − 1)}C23 {qℓ pℓ . F } = −2qi p2 .66) (10. G} − {F. the Dirac brackets between the fundamental dynamical variables take the forms 1 −1 {pi . 0  0 (10. In particular. (qj qj − 1)}C21 {pF .  = 0 −1 0 2p2    2 −1 0 −2p 0 0 0 0 1  (10. pj } 2 j = δi − qi qj . qk pk }C32 { (qℓ qℓ − 1). pF }D = {F.45)) −1 {F. 2 {qi . pj } − {pi . F } = 2pi . pF }D = 0. F }D = {qi .68) .  2 2p =  0  1  0 . F } − {qi . φA }CAB {φB . pF }D = 0. pF }D = {pi . pj } 2 −1 1 −{pi . pj pj − F }C41 {pF . F }D = {pF .67) −1 with CAB given in (10. G}. = −qi pj + pi qj . pj } − {qi . pj } 2 −1 1 {qi . F } − {pi . G}D = {F. F }D = {pi . qj }D = 0. pj }D = {pi .394  10 Dirac method for constrained systems C AB The inverse of this matrix has the form  0 −2p2 0 1 0 0 −1 0 0 −1    0 0 1 0  .

61)) and the Hamiltonian for the theory is given by (see (10. we can set the constraints to zero so that the true independent dynamical variables are (qi . 10.69) where we have used (10.10.73) where λ is a parameter labelling the trajectory and we have identified . (10. We note that if we define the relativistic four velocity and momentum as uµ = dxµ . The constraints of this theory were all second class. Let us next study the Hamiltonian description of a free relativistic massive particle which is described by the equation d2 xµ = 0.72) The dynamical equation (10. pi ) (we note that F = pi pi because of the constraint (10.54)) 1 i i pp. dτ 2 m (10.70) can also be written as dpµ = 0.70) where τ denoting the proper time is assumed to label the trajectory of the particle and m is the rest mass of the particle. dτ (10. (10.62). This is the starting point for the quantization of this theory.4 Relativistic particle 395 With the use of the Dirac brackets.70) can be obtained as the EulerLagrange equation following from the action (recall that c = 1) dλ (xµ xµ ) 2 = ˙ ˙ 1 S=m ds = m dλ L.71) then the dynamical equation (10. dτ pµ = muµ .4 Relativistic particle In the last section we studied a simple example of point particle dynamics which is constrained. 2 Hp = (10.

73) will lead to the dynamical equation (10. (10. (10.70) or (10. then (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. However. dλ xµ = ηµν xν . (10. If we choose a gauge λ = τ. We note from (10. the action (10. let us proceed with the Hamiltonian description without choosing a gauge at this point.76) Namely.77) where τ denotes the proper time associated with the particle.73) is invariant under the diffeomorphism (10.396 10 Dirac method for constrained systems xµ = ˙ dxµ . dλ dλ dξ λ → ξ = ξ(λ).78) which leads to .73) that ∂2L = ∂ xµ xν ˙ ˙ m xρ xρ ˙ ˙ xµ xν ˙ ˙ xσ xσ ˙ ˙ Pµν = ηµν − . so that (10.72) as the Euler-Lagrange equations.74) We note that under a transformation xµ = ˙ dξ dxµ dxµ = .75) which is an example of a local gauge transformation much like in the Maxwell field theory. ˙ ˙ (10.

Pµν is a projection operator and ∂2L = 0. ˙ (10.81) where because of the diffeomorphism invariance of (10. Let us note that the action (10.73).73) in the form 1 2 √ ˙ ˙ dλ g g−1 xµ xµ + m2 . we can rewrite the action (10.73) can also be written in an alternate form that is more useful.73) and (10. m2 (10.80) as in the case of the Maxwell theory (see (9.81) are equivalent.81) using (10.82).4 Relativistic particle 397 xµ Pµν = ˙ m xρ xρ ˙ ˙ xν − ˙ xµ xµ xν ˙ ˙ ˙ xσ xσ ˙ ˙ = 0 = Pµν xν . the action (10. We can obtain the momenta conjugate to these variables to be . For example. it is more convenient for us to work with the action (10.22)).81) are xµ and g. However.81). we note that the equation for the auxiliary field g takes the form g−1/2 ∂L = −g−1 xµ xµ + m2 = 0. To see that the two actions (10. S= dλ L = (10.73) we can think of g as the metric on the one dimensional manifold of the trajectory of the particle. g= xµ xµ ˙ ˙ . ∂ xµ xν ˙ ˙ det Pµν = det (10.10.73) to describe the motion of a massless particle.79) Therefore. it is not clear how to take the m = 0 limit in (10. Viewed in this manner. The two dynamical variables of the theory in (10.81) is manifestly diffeomorphism invariant and the massless limit can now be taken in a straightforward manner. ˙ ˙ ∂g 4 or. then we obtain (10. We note that introducing an auxiliary variable g.124)) and the naive passage to the Hamiltonian description fails (see (10.82) If we eliminate g in (10.

g} = {pµ . µ {xµ . pg } = 0.85) Adding 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 . 4 g which leads to the secondary constraint . pg } = 0.84).83) so that the primary constraint of the theory follows to be ϕ1 = pg ≈ 0. pg } = {pµ . g} = {pg . pν } = δν . we obtain √ g µ p pµ − m2 + λ1 pg } 2 (10.84) Hcan (10.87) With these we can now determine the evolution of the primary constraint (10. The canonical Hamiltonian of the theory follows to be √ g µ = pµ x + pg g − L = ˙ ˙ p p µ − m2 . pg } = 1. 1 1 = − √ pµ pµ − m2 ≈ 0. ˙ ∂ xµ ˙ pg = ∂L = 0. {xµ . (10.398 10 Dirac method for constrained systems pµ = ∂L = g−1/2 xµ . {g.84) and requiring the constraint to be time independent. Hp } = {pg .86) The equal time canonical Poisson brackets for the theory are given by {xµ . ∂g ˙ (10. Hp = Hcan + λ1 ϕ = 2 1 (10.88) ϕ ˙ 1 ≈ {ϕ . 2 µ (10. pν } = {g. g} = {xµ . xν } = {pµ .

91) χ1 = g − χ2 = λ − x0 ≈ 0.89) is the familiar Einstein relation for a massive relativistic particle (the multiplicative factor is for later calculational simplicity). which render all the constraints to be second class.92) that the matrix of Poisson brackets of constraints has the form .10. χ2 ). then it follows from (10.92) = −{χ2 . Let us choose the gauge fixing conditions 1 ≈ 0. g − {ϕ2 . χ2 } = { 1 µ p pµ − m2 .4 Relativistic particle 399 ϕ2 = 1 µ p pµ − m2 ≈ 0. 2 2 = 0. Time evolution of the secondary constraint leads to √ g ν 1 µ 2 p p ν − m 2 + λ1 p g } ≈ {ϕ .89) We recognize that the secondary constraint (10.90) 2 ϕ ˙ 2 so that it is time independent and the chain of constraints terminates. As a result. m2 (10. χ1 } = {pg . λ − x0 } = p0 2 (10. The non-vanishing Poisson brackets between the constraints are given by 1 } = −1 = −{χ1 . ϕ2 . ϕ2 }. χ1 . if we combine all the constraints into φA = (ϕ1 . ϕ1 }. The two constraints of the theory can be easily checked to be first class which is associated with the fact that the theory has a local gauge invariance and is also reflected in the fact that the Lagrange multiplier λ1 remains arbitrary (the primary constraint is first class). Hp } = { p pµ − m . 2 (10. m2 {ϕ1 . (10.

pν } − {xµ .94) (10. pg } = 0. pg } {xµ . pν } µ = δν − {xµ . pg } − {g. pν }D = {g. G}D = {F. −1 {g.45)) −1 {F. G}. φ2 }C24 {φ4 . G} − {F. g}D = {pg . µ = δν − 1 λ 1 p pλ − m2 } − 0 {λ − x0 . pg }D = {pµ . pg }D = {g.93) 1 0 −1 CAB The equal time Dirac bracket between any two dynamical variables is given by (see (10. pg } − {g. = {g. p0 ν With the Dirac brackets we can set the constraints to zero in which case we see that the primary Hamiltonian (10.  0 = −1  0 1  0 − p0  .95) and with (10. m2 −1 {xµ . xν }D = {pµ . φA }CAB {φB . g}D = {pµ . φ1 }C13 {φ3 . (On . pg }D = 0. 0 0   0 0 (10. pν } 2 p (10. pν }D = {xµ .94) we can calculate the Dirac bracket between the fundamental variables to be {xµ . pg }{g − 1 .86) vanishes.96) pµ 0 δ .400 10 Dirac method for constrained systems C AB whose inverse is given by  0 0 0 0 1 p0  0 0 = 1 0  0 −p0  0 0 −1 0 0 0  p0  . pg }D = 0. 0  0  0  (10. g}D = {xµ .

2. where α = 1.100) † The dynamical variables of the theory can be chosen to be ψα . 10.20)). (10. that the first of the gauge fixing conditions corresponds to choosing λ = τ . dλ dλ xµ xµ = ˙ ˙ (10. ψα . or. therefore. the free Dirac theory is described by the Lagrangian density / L = iψ∂ ψ − mψψ.5 Dirac field theory 401 the other hand.10. it is obvious that the matrix of second order dervatives vanishes and the system is singular (see (10. dτ dτ (10.98) It follows. (10. from the definition of the infinitesimal invariant length in the Minkowski space (see (1. 4 denote the Dirac indices and since the Lagrangian density is first order in derivatives.89) in particular leads to H = p0 = p2 + m2 which can be taken as the Hamiltonian. This manifests in the definition of the conjugate momenta as (we have chosen the convention of left derivatives) . where the adjoint spinor is given by ψ = ψ† γ 0 .5 Dirac field theory As we have seen in (8.) Let us also note from (10.99) (10.22)). if we set the constraints strongly to zero.9).97) On the other hand. we have (remember that c = 1) dτ 2 = dxµ dxµ . dxµ dxµ = 1.82) that the first of the gauge fixing conditions in (10. 3.91) would imply dxµ dxµ = 1.

The canonical Hamiltonian is now obtained to be Hcan = d3 x Hcan = d3 x −iψγ · ∇ψ + mψψ . we obtain the two sets of primary constraints for the theory to be † ϕ† = Π† + iψα ≈ 0. α α (10.101) in the intermediate steps.402 10 Dirac method for constrained systems Π† = α Πα = ∂L † = −i(ψγ 0 )α = −iψα .11)) ˙ ˙† Hcan = −Π† ψα + ψα Πα − L α = −iψγ · ∇ψ + mψψ. The canonical Hamiltonian density of the theory (consistent with the convention of left derivatives) is obtained to be (see (10.105) . Rather they correspond to the momenta conjugate to ψ and ψ † respectively. ˙ ∂ ψα ∂L = 0.104) Adding the primary constraints to the canonical Hamiltonian. (10.) As a result. † ˙ † ˙ = iψα ψα − iψα ψα − iψγ · ∇ψ + mψψ (10.30)) Hp = Hcan + d3 x φ† ξα + λ† ρα .103) where we have used (10. we obtain the primary Hamiltonian for the Dirac field theory as (see (10. α α ρα = Πα ≈ 0.102) which in this case correspond to fermionic constraints. ˙† ∂ ψα (10. (10.101) (Note that Π† is not the Hermitian conjugate of Π.

−iψ(y)γ · ∇y ψ(y) + mψ(y)ψ(y)} α d3 y iψ(y)γ · ∇y − mψ(y) β {Π† (x). λ† in the present case denote α fermionic variables. φ† (y)} α β −λ† (y){φ† (x).106) with all other brackets vanishing. d3 y ξβ (y){φ† (x).5 Dirac field theory 403 where the Lagrange multipliers ξα . ψβ (y)} α d3 y iψ(y)γ · ∇y − mψ(y) i∇ψ(x) · γ + mψ(x) α β − δαβ δ3 (x − y) . Π† (y)} = −δαβ δ3 (x − y) = {ψα (x). Hp } α α ≈ {φ† (x). ρβ (y)} . The equal-time canonical Poisson brackets for the field variables take the forms (see (10. β (10. For example.107) α β Let us evaluate each of these terms separately. Π† (y) + iψβ (y)} α β . we can now calculate the time evolution of the primary constraints (10. φ† (y)} α β = † † d3 y ξβ (y){Π† (x) + iψα (x). Hcan } + α d3 y ξβ (y){φ† (x). Hcan } α = = = = † d3 y {Π† (x) + iψα (x). Using (10.10. Using the fact that the primary Hamiltonian is independent of time.13)(10. (10.12)) † {ψα (x).14)).102). Πβ (y)}. 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).106) (see also (10. ˙ φ† (x) ≈ {φ† (x).

α α (10.107) and requiring the constraint to be independent of time leads to ˙ φ† (x) ≈ i∇ψ(x) · γ + mψ(x) α + iλ† (x) = 0. ψβ (y)} γ 0 (−iγ · ∇y + m)ψ(y) + iξ(y) β d3 y − δαβ δ3 (x − y) γ 0 (−iγ · ∇y + m)ψ(y) + iξ(y) α β = − γ 0 (−iγ · ∇ + m)ψ(x) + iξ(x) = 0. − 10 Dirac method for constrained systems d3 y λ† (y){φ† (x).112) . Hp } = {Πα (x). Hp } ˙ = = † d3 y {Πα (x). (10. requiring the second set of primary constraints to be time independent we obtain ρα (x) ≈ {ρα (x).110) Similarly. (10. α Substituting these results into (10. ρβ (y)} α β = − = − † d3 y λ† (y){Π† (x) + iψα (x). (10. Πβ (y)} α β d3 y λ† (y) − iδαβ δ3 (x − y) β (10.111) which determines the other Lagrange multiplier ξα = γ 0 (γ · ∇ + im)ψ(x) α .404 = 0.108) = iλ† (x).109) which determines the Lagrange multiplier λ† = − ∇ψ(x) · γ + imψ(x) α α .

for simplicity of notation.113) where. combining the constraints into ΦA = (φ† . 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 . In fact. Πβ (y)} = 0. ρβ (y)} = {Π† (x) + iψα (x).114) Let us note next that the two sets of primary constraints (10.112) into the primary Hamiltonian. we can write α the matrix of Poisson brackets as .10. rather it determines both the Lagrange multipliers in the primary Hamiltonian (10. Π† (y) + iψβ (y)} = 0. we have (at equal time) † † {φ† (x). (10.5 Dirac field theory 405 In this case. which constitute all the constraints of the theory.102). Substituting (10. define second class constraints (this is consistent with the fact that there is no undetermined Lagrange multiplier in the theory). φ† (y)} = {Π† (x) + iψα (x).115) Thus. β {ρα (x). φ† (y)}. we see that requiring the two primary constraints to be time independent does not introduce any new constraint. α α β β † {φ† (x). ρβ (y)} = {Πα (x). (10. we have defined Π = Π† γ 0 . (10. Πβ (y)} α α = −iδαβ δ3 (x − y) = {ρα (x).110) and (10.105). ρα ).

116) The inverse of the matrix is easily seen to be C −1 (x. G(y)} γ † +{F (x). we obtain .117) The equal-time Dirac bracket of any two dynamical variables can now be easily defined (see (10. G(y)} ¯ ¯ z γ +{F (x). G(y)} ¯ = {F (x). G(y)} − −1 ¯ z d3 zd3 z {F (x). ΦA (z)}CAB (z. φ† (y)} {φ† (x). y) = {φ† (x). ρβ (y)} β 0 = −i ½ ½ 0 δ3 (x − y). φ† (y)} {ρα (x).118) we can calculate the Dirac bracket between the field variables. (10. G(y)} −i † d3 z {F (x).406 10 Dirac method for constrained systems C(x. G(y)} ¯ δ z = {F (x). Recalling that constraints can be set to zero inside the Dirac bracket. ρβ (y)} α α β {ρα (x). (10.45)) {F (x). Π† (z) + iψγ (z)}{Πγ (z). ργ (z)}(iδγδ δ3 (z − z )){φ† (¯). z ){ΦB (¯). G(y)}D = {F (x). y) = i 0 ½ ½ 0 δ3 (x − y).118) Using (10. γ (10. G(y)} . Πγ (z)}{Π† (z) + iψγ (z). φ† (z)}(iδγδ δ3 (z − z )){ρδ (¯). G(y)} − d3 zd3 z {F (x).

2.120) and these are the relations we have used in quantizing the Dirac theory (see (8. we recall that when using the Dirac brackets. Maxwell’s equations can be obtained as the Euler-Lagrange equations from the Lagrangian density 1 L = − Fµν F µν .113) becomes Hp = d3 x −iψγ · ∇ψ + mψψ . ψβ (y)}D . Similarly. (10. the Hamiltonian of the theory (10. Therefore. 1. 10. As we have seen in (9. we can show that † † {ψα (x). ψβ (y)} γ d3 z(−δαγ δ3 (x − z))(−δγβ δ3 (z − y)) (10. ν = 0.118) in evaluating this bracket because Πγ has a vanishing Poisson bracket with ψα . where we have neglected the last term in (10.122) where the field strength tensor (see (9.6 Maxwell field theory Let us next consider the Maxwell field theory which as we know exhibits gauge invariance. (10. 3.16).119) = −iδαβ δ3 (x − y). we can set the constraints to zero in the theory.18)). 4 µ.10. ψβ (y)}D = {ψα (x).6 Maxwell field theory 407 † † {ψα (x). Furthermore.16)). (10.4)) . Π† (z) + iψγ (z)}{Πγ (z). ψβ (y)} −i = −i † † d3 z{ψα (x). ψβ (y)}D = 0 = {ψα (x).121) which is the Hamiltonian we have used in the quantization of the Dirac field theory (see (8.

˙ Π = E = −(A + ∇A0 ). Fij = −ǫijk Bk .127) This determines that the theory has a primary constraint given by ϕ1 (x) = Π0 (x) ≈ 0. As a result of the gauge invariance of the theory.7)) F0i = Ei . j. i. 2. (10. 3.5) and (9.126) so that we have Π0 = 0.125) where α(x) denotes the local parameter of gauge transformation.408 10 Dirac method for constrained systems Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ .123) is the four dimensional curl of the four vector potential and contains the electric and the magnetic fields as its components (see (9. (10. ˙ ∂ Aµ Πµ = (10.128) We can now obtain the canonical Hamiltonian density of the theory to have the form . µ (10. (10. (10.124) We also know that Maxwell’s theory is invariant under gauge transformations Aµ (x) → A′ (x) = Aµ (x) + ∂µ α(x). the matrix of quadratic derivatives becomes singular (see (9. k = 1.122)) and this manifests in the definition of the conjugate momenta as ∂L = −F 0µ .

129) where we have used (10. Aν (y)} = 0 = {Πµ (x). ν {Aµ (x). = 2 (10.6 Maxwell field theory 409 1 ˙ ˙ Hcan = Πµ Aµ − L = −Π · A + Fµν F µν 4 1 = −Π · (−Π − ∇A0 ) + (−E2 + B2 ) 2 1 2 (Π + B2 ) + Π · ∇A0 .127). 2 (10. The equal-time canonical Poisson brackets of the theory are given by {Aµ (x). Aµ (x)}.131) where λ1 denotes the Lagrange multiplier of the theory.10. Πν (y)} = δµ δ3 (x − y) = −{Πν (y). is given by Hcan = = d3 x Hcan d3 x 1 2 (Π + B2 ) − A0 ∇ · Π .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 ) . 2 (10.130) where we have neglected a surface term (total divergence) in integrating the last term by parts.128) Hp = Hcan + = d3 x d3 x λ1 ϕ1 1 2 (Π + B2 ) − A0 ∇ · Π + λ1 Π0 . therefore. The primary Hamiltonian of the theory is now obtained by adding the primary constraint (10. (10.132) Using these we can calculate the time evolution of the primary constraint (10. Πν (y)}. The canonical Hamiltonian.

ϕ2 (x) = ∇ · Π(x) ≈ 0. as we have seen. Therefore.134) It can now be checked that the secondary constraint is time independent. in the presence of first class constraints we are supposed to add gauge fixing conditions to convert them into second class constraints. Hp } ˙ = 1 d3 y {Π0 (x). we have a secondary constraint in the theory given by ϕ2 (x) = ∇ · Π(x) ≈ 0. 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.135) (10. According to the Dirac procedure. We see that Maxwell’s theory has two constraints ϕ1 (x) = Π0 (x) ≈ 0. 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. Let us choose . ϕ2 (x) ≈ {ϕ2 (x). (10. the Lagrange multiplier λ1 remains as yet undetermined (there is one primary constraint which is first class). namely.410 10 Dirac method for constrained systems ϕ1 (x) ≈ {ϕ1 (x). ˙ so that the chain of constraints terminates. Furthermore. (10.136) and it is clear that both these constraints are first class constraints.133) = ∇ · Π(x) ≈ 0. Hp } = {∇ · Π(x). Hp } ≈ 0. (Π2 (y) + B2 (y)) 2 − A0 (y)∇ · Π(y) + λ1 (y)Π0 (y)} ≈ − d3 y {Π0 (x). A0 (y)}∇ · Π(y) (10. In this case.

136) into second class constraints. ∇ · A(y)} = −δ3 (x − y) = −{χ1 (x).  = −1 0 0 0    −2 0 −∇x 0 0 (10.140) .136) and (10. y)  −2  0 0 0 ∇x  3  δ (x − y). 2. 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). α = 1. ϕ1 (y)}.139) The inverse of this matrix is easily calculated to be 0 0 1 0 −1 CAB (x.137) as the gauge fixing conditions which clearly convert the constraints (10. = (∇x )i (∇y )j {(Π)i (x). A0 (y)} {ϕ2 (x).138) which leads to (x0 = y 0 ) C AB (x.10. χ2 (y)} = {∇ · Π(x). (10.6 Maxwell field theory 411 χ1 (x) = A0 (x) ≈ 0. χ1 (y)} = {Π0 (x). Therefore. φB (y)}   0 0 −1 0  2 0 0 0 −∇x  3   δ (x − y). ϕ2 (y)}. we can calculate the matrix of the second class constraints from the fact that the only nonvanishing equal-time Poisson brackets between the constraints are given by {ϕ1 (x). =  1 0 0 0    2 0 ∇x 0 0   (10. χ2 (x) = ∇ · A(x) ≈ 0. relations (10.137) determine all the constraints of the theory. χα ). (10. y) = {φA (x).

142) ¯ 1 1 3 2 δ (x − y) = − 4π|x − y| . the relevant Dirac brackets for the Maxwell field theory follow from (10. G(y)}. 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). in particular. G(y)}D = {F (x). Πj (y)}D . Π0 (z)}(δ3 (z − z )){A0 (¯). G(y)} − −1 ¯ z d3 zd3 z {F (x).141) and.143) Since we can now set the constraints (10. for the dynamical field variables.136) and (10. (10. Aj (y)}D = 0 = {Πi (x). ∇z · Π(z)}(∇z δ3 (z − z )){∇z · A(¯). Πν (y)} − −2 ¯ z +{Aµ (x). ∇x (10.144) j = δi TR (x − y). Πν (y)} ¯ ¯ z {Aµ (x).143) to be {Ai (x). Πν (y)}D . ∇x (10. Aν (y)}D = 0 = {Πµ (x). d3 zd3 z {Aµ (x). j {Ai (x).412 10 Dirac method for constrained systems where. .137) to zero. φA (z)}CAB (z. 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). it leads to {Aµ (x). Πν (y)}D = {Aµ (x). z ){φB (¯). with the usual asymptotic condition that fields vanish at spatial infinity. Πj (y)}D = δi + (∇x )i 1 3 j 2 (∇x ) δ (x − y) ∇x (10.

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

This page intentionally left blank .

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

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

shows that even that does not present any difficulty. the parity transformation (11. In this case.2)) and that expectation values of quantum operators behave like classical objects (Ehrenfest theorem). namely. 11. as we will see. But a systematic analysis. we know that dynamical laws of physics (such as Newton’s equation) do not depend on the handedness of the coordinate frame. let us assume that under the parity transformation. (11.6) where we have used the fact that classically vectors change sign under space reflection (see (11. of course.5) Defining a symmetry in a quantum system.6) holds.1. P −→ − P . we can analyze the parity transformation in (11. ½ and P with P 2 = ½. As is standard in quantum mechanics.6) from two equivalent points of view. let us consider a one dimensional quantum mechanical system.11. For a long time it was believed that the microscopic systems are also invariant under parity. In this case. In other words. but we know now that there exist processes in nature which do not respect parity. The only difficulty we may apprehend is in the case of spinor states since they are double valued. quantum mechanical states change.1) would result in X −→ − X . P P (11. Namely. First.1 Parity 417 so that that even classically parity operation defines a group with two elements (a group of order two). the classical laws of physics are invariant under parity or space inversion. but not the operators such that (11.1 Parity in quantum mechanics. Classically. under space reflection an arbitrary quantum mechanical state transforms as . simply corresponds to asking whether this group can be carried over to quantum mechanics without any obstruction. let us assume that P represents the operator which implements the action of parity on the quantum mechanical states. For simplicity.

11) In other words. or.9) δ(x − y) = x|P † P|y . Since P has real eigenvalues we conclude that the parity operator is Hermitian and thus we have . acting on the coordinate basis P must lead to |x → |xP = P|x = | − x . such that ψ|X|ψ ψ|P |ψ → → P P (11. (11.10) which leads to P † P = ½. = P| − x = |x .12) P 2 |x which shows that P2 = ½. −x| − y = x|P † P|y .7) ψ P |X|ψ P = ψ|P † XP|ψ = − ψ|X|ψ .8) Note that parity inverts space coordinates and. (11. P is an idempotent operator and the only eigenvalues of P are ±1. since P|x = |−x .13) Namely. It is now obvious from (11. the parity operator is unitary. P (11. therefore. (11.9) that xP |y P = x|P † P|y .418 11 Discrete symmetries |ψ → |ψ P = P|ψ . ψ P |P |ψ P = ψ|P † P P|ψ = − ψ|P |ψ . (11. (11. or. Furthermore.

(11.1 Parity 419 P2 = P † = P = P −1 .18) (11. In other words. if |ψ is an eigenstate of parity. P P (11. then P|ψ = ±|ψ .15) so that we obtain x|ψ P = ψ P (x) = ψ(−x). P (11.11.14) To see how a scalar (not a multi-component) wave function transforms under parity.19) Namely. ½.17) Let us note further that since the eigenvalues of P are ±1. under a parity transformation the wave function transforms as ψ(x) −→ ψ P (x) = ψ(−x). or. for such a state |ψ −→ |ψ P = P|ψ = ±|ψ .16) Namely. an eigenstate of parity has an associated wave function which is either even or odd depending on the eigenvalue of the parity .7) that P |ψ → |ψ P = P|ψ = P = = P dx ψ(x)|x = dx ψ(x)| − x . ψ(x) −→ ψ P (x) = ψ(−x) = ±ψ(x). (11. dx |x x|ψ dx ψ(x)P|x (11. we note from (11.

14). namely. we may assume that the states do not transform under the parity transformation.24) . −P ). In the other point of view.20) ψ|X|ψ ψ|P |ψ → → P ψ|X P |ψ = − ψ|X|ψ .6) holds.22) where we have used (11. X]+ = 0. PX + XP = 0. (11. rather under a transformation only the operators change as (this point of view may be more relevant within the context of field theories) O −→ OP = P † OP.420 11 Discrete symmetries operator for that state. P ) = P † O(X. (11. however. P † P P (11. therefore. does not have to be an eigenstate of the parity operator and. [P. one can show that for any operator in this simple one dimensional theory. such that (11. P ) −→ OP (X. P )P = O(−X. ψ|P P |ψ = − ψ|P |ψ . or. we can obtain that [P. P ]+ = 0.23) In general. P P (11. Similarly.21) It follows from this that P X −→ X P = P † XP = −X. O(X. P = O P † XP. (11. or. A general state of the system. the corresponding wave function need not have any such symmetry behavior.

namely. Hn (x). H(X. the eigenstates of the Hamiltonian would naturally decompose into even and odd states.11. Therefore. H] = 0. then as we have seen [P. or. If parity is a symmetry of the theory. P )] = 0.27) that [P. P ) −→ P † H(X. P )P = H(−X. indeed. U (t)] = 0.28) . we see that if parity is a symmetry of a quantum mechanical system. [P. (11. what happens. H(X.26) that this is. (Recall that the Hermite polynomials.27) where U (t) is know as the time evolution operator. P ). 2m 2 P H(X.29) (11. (11. P ) = (11. It is obvious from simple systems like the one dimensional harmonic oscillator 1 P2 + mω 2 X 2 = H(−X. in such a case. (11. −P ).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. the solution of the time dependent Schr¨dinger equation o can be written as |ψ(t) = e−iHt |ψ(0) = U (t)|ψ(0) .25) As a result. which are eigenfunctions of the Hamiltonian are even or odd functions of x depending on the value of the index n. −P ) = H(X. P and H can be simultaneously diagonalized and the eigenstates of the Hamiltonian would carry specific parity quantum numbers.) If the Hamiltonian H of a quantum mechanical system is time independent. and it follows from (11.

we conclude that 11 Discrete symmetries P|ψ(t) = PU (t)|ψ(0) = U (t)P|ψ(0) . (11. dxdy | − y y|X| − x x| . we cannot construct any function of X and P which would represent the parity operator. in this one dimensional theory. let us note that in this simple one dimensional quantum theory. = P|ψ = or. Before we go on to the discussion of parity in quantum field theories.) However. This is another way to say that parity is conserved in such a theory.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. This merely corresponds to the fact that classically there is no dynamical conserved quantity associated with this discrete symmetry.31) It is obvious from this definition that P† = P2 = = = |ψ P dx |x −x| = dx | − x x| = P. dxdy | − x x| − y y| dxdy δ(x + y)| − x y| dy |y y| = ½. ψ P (x) = P † XP = x|ψ P = ψ(−x). (Recall that N¨ther’s theorem holds for o continuous symmetries. dx | − x x|ψ = dx ψ(x)| − x . takes the form P= dx | − x x|. (11.422 As a result. a nontrivial representation of the parity operator is still possible and.

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

37) If ηφ = 1 or the intrinsic parity of the spin zero field is positive. t). (11. (Remember that parity and the momentum operators do not commute and.1. (11. 11. From our general discussion in (11. then φP (x. Let us begin with the spin zero field which is described by the Klein-Gordon equation ∂2 − ∇2 φ(x) = 0. if φ(x. (11. t) = P † ψα (x. (11. t) −→ ψα (x. φ= (11. ∂t2 or.36) depending on whether the field is massless or massive. t)P = ηψ Sα ψβ (−x. t) = ηφ φ(−x. ( + m2 )φ(x) = 0. Consequently. t) −→ φP (x.38) P (11. In either of the cases.2 Spin zero field. so is φ(−x.34) and (11.34) In the language of quantized fields. t). t).424 11 Discrete symmetries 2 ηψ S 2 = ½. we conclude that in this case. with ηφ = ±1.35). cannot have simultaneous eigenstates unless the eigenvalue of momentum vanishes. t) = φ(−x. therefore.) 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. t). φ(x.39) . t) is a solution of the Klein-Gordon equation. the equation is clearly invariant under space inversion. the above relation represents an operator relation P β ψα (x.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.

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

ηB and ηC are the intrinsic parities of the particles A.45) If parity is a symmetry of the theory. 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.47) where ηA . if we have . P → P + S.48) S → P + P. one spin zero particle can decay into two spin zero particles in an s-state (ℓ = 0) only through the channels S → S + S. Let us further note that for a complex spin zero field. (11. (11. B and C respectively and ℓ is the orbital angular momentum of the B-C system.49) will be forbidden. It now follows that in a parity conserving theory. P P → S + S.46) (11. we must have ηA = ηB ηC (−1)ℓ . (11. (11. → P + P. whereas processes such as (ℓ = 0) S → P + S.426 11 Discrete symmetries ηTOT = ηφ (−1)ℓ .

11. t) − ∂ µ φ† (x.2).) Thus we conclude that a charged spin zero particle and its antiparticle have the same intrinsic parity. the current density has the explicit form J µ (x. t). J P 0 (x. quite general for the bosons and says that for bosons. t) φ(x. t) − (∂ µ (φP )† (x. t) − (∂ µ φ† (−x. (11.2 and 7. t) .52) which leads explicitly to (recall that |ηφ |2 = 1) JP (x. t). Let us next look at the electromagnetic current associated with a charged Klein-Gordon system. t) = i|ηφ |2 φ† (−x. t) = i (φP )† (x. t)∂ µ φP (x. t)∂ µ φ(x. t))φ(x. the intrinsic parity of the field φ† is the same as that of the field φ. t) = ηφ φ(−x.51) Under a parity transformation J µ (x.3.50) On the other hand. t) = ηφ φ† (−x. As we have seen in (7. (11.(11.1 Parity 427 φP (x. t). ∗ (φP )† (x. t))φP (x. t) . t) = i φ† (x. particles and antiparticles would have the same intrinsic parity.38)). This result is. t) −→ J P µ (x. . In quantum field theory φ and φ† are associated with particles and antiparticles. t)∂ µ φ(−x.29). t) = −J(x. t) = J 0 (−x. (φ destroys a particle or creates an antiparticle whereas φ† destroys an antiparticle or creates a particle. t). see discussion in sections 7. (11. since ηφ is assumed to take only real values ±1 (see (11. in fact.53) P which is the correct transformation for the current four vector as we have seen in (11.

A0 (x. t). A0 (x.55) Namely. P P (11. From Maxwell’s equations in the presence of sources (charges and currents) ∇ · E = ρ = J 0. from the definitions of the electric and the magnetic fields in (9. t).56) we conclude that under a parity transformation the components of the four vector potential would transform as A(x. t) = ηA A(−x. B(x. t) −→ AP 0 (x. ∂t ∂t B = ∇ × A.1. we must have E(x. t). ∂t (11.3 Photon field. t) −→ AP (x. In the quantum theory. t) −→ −E(−x. Furthermore. (11.2). t) = A0 (−x. t).53) of the components of the current four vector under a space reflection.54) under space reflection requires that under a parity transformation.57) This is very much like the transformations (11. P P (11. we see that whereas the electric field would transform like a vector. we would then generalize these as operator transformations A(x. ∂A ∂A − ∇φ = − − ∇A0 .54) we note that invariance of (11.428 11 Discrete symmetries 11.58) P P . namely. t). t) −→ B(−x. t). the magnetic field would behave like an axial vector if parity is a symmetry of the system. t) −→ A0 (−x. ∇×B = ∂E + J. t) = −A(−x. t) = ηA0 A0 (−x. t) −→ −A(−x. E = − (11.

11.61) where in the last step we have let x → −x under the integral. t)AP 0 (x. t) − (−J(x. From the parity transformation of the dynamical components. A0 is nondynamical). t)Aµ (−x. t)AP (x.35). t)Aµ (x. (11. under a parity transformation . t) d3 x J 0 (−x. t)) · (−A(x. This shows that the electromagnetic interaction is invariant under parity and. Let us recall that the Dirac wave function or the Dirac field is described by a four component object.1. namely A. t)Aµ (x.1 Parity 429 Note that a physical photon has only transverse degrees of freedom (namely. (11. Thus according to our general discussion in (11. (11.33) or (11.11. therefore. t). In other words. t)) d3 x J µ (−x.4 Dirac field.60) we note that under a parity transformation Hem = → = = = = P d3 x J µ (x. t)Aµ (x. t) = Hem . t) d3 x J P µ (x. t). parity (quantum number) must be conserved in electromagnetic processes. t) − JP (x. t) = −A(−x.59) from which we conclude that the intrinsic parity of the photon field (or the one photon state) is negative. From the form of the electromagnetic interaction Hamiltonian Hem = d3 x J µ (x. t) d3 x J µ (x. t) · AP (x. we see that AP (x. the photon is a vector particle. t)A0 (−x. t) µ d3 x J P 0 (x.

65) which will satisfy the above relation. t) = (ψ P )† (x. 2. t). t) = ψ(−x.62) In matrix notation. t). t). we can always choose ηψ = ±1. t) −→ J P µ (x.67) 2 = ηψ ψ(−x. P (11. we can write the transformation as ψ P (x. t)S † γ 0 ∗ = ηψ ψ † (−x.34)) 2 ηψ S 2 = ½. with (see (11.430 11 Discrete symmetries P β ψα (x. S 2 = ½.64) Since we have assumed the parity eigenvalues to be ±1. t)γ 0 S † γ 0 . t) = ψ (x.66) Let us recall that associated with the Dirac system is a conserved current (probability current or electromagnetic current. (11. 4. α = 1. t)γ 0 γ 0 S † γ 0 P = ηψ ψ(−x. With this choice (which also ∗ implies ηψ = ηψ ) we note that ∗ ψ (x.44)) J µ (x. 3. t) = ηψ Sψ(−x. t) P P (11.68) . (11. t)γ µ ψ(x. t) = ψ(x.63) (11. t) = ηψ Sα ψβ (−x. (11. t)γ µ ψ P (x. t) −→ ψα (x. and under a parity transformation this would transform as J µ (x. t)γ 0 = ηψ ψ † (−x. (11. t)γ 0 S † γ 0 γ µ Sψ(−x. t)γ 0 S † γ 0 γ µ Sψ(−x. see (8. t).

73) (11. Namely.11. t)γ i ψ(−x. t). t)γ 0 . t) = ψ(−x.65) we have S = S † = S −1 .72) Thus we determine the transformation of the Dirac field (wave function) under a space reflection to be ψ(x. (11. (11. then the parity transformed function ψ P (x. we obtain ψ(−x. namely. we obtain ψ(−x. t) = −J i (−x. with ηψ = ±1.69) for µ = i. t). t) = J 0 (−x. t) = ηψ ψ(−x. J P 0 (x. P P P (11.71) Similarly.69) Comparing (11. comparing (11. t)γ 0 ψ(−x.74) It is now quite easy to see that if ψ(x. or. t).1 Parity 431 2 where we have used the fact that ηψ = 1. t) −→ ψ (x.68) and (11.70) so that using (11. S † S = ½. or. t) = ηψ γ 0 ψ(−x.53)). ψ(x. t) also satisfies the Dirac equation in the new coordinates. t)γ 0 S † γ 0 γ i Sψ(−x. if we define . t). t)γ 0 S † Sψ(−x. t) −→ ψ P (x. t) is a solution of the Dirac equation. t) = −ψ(−x. J P i (x. On the other hand.68) and (11.69) for µ = 0. S = γ 0. t). (11. we know the transformation properties of the current four vector under space inversion (see (11. (11.

77) In the rest frame of the positronium. in fact. Let ǫ1 and ǫ2 denote the polarization vectors for the two photons while k represents their relative momentum. The polarization planes of the photons in this decay can. However. as we have discussed in section 3. This can also be seen by noting that parity is a symmetry of the free Dirac Lagrangian density (8.6. ∂xµ (11. To determine the relative parity between Dirac particles and antiparticles. −x). Since the electromagnetic interactions conserve parity.8 eV) e− + e+ → γ + γ.76) In other words. the two component Weyl fermions violate P. (11. the scalar transition amplitude will have the general form Mi→f = (a ǫ1 · ǫ2 + b k · (ǫ1 × ǫ2 )) ψpositronium .432 11 Discrete symmetries y µ = (t.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. be measured and it .78) where ‘a’ and ‘b’ are scalar functions of momentum and ψpositronium represents the wave function for the positronium. (11.9). space inversion is a symmetry of the Dirac equation. then ∂ − m ψ P (y) = ∂y µ (11. let us next analyze the decay of the positronium in the 1 S state where we use the standard spectroscopic notation 2S+1 L 0 J for the states (the positronium is a bound state of an electron and a positron much like the Hydrogen with a ground state energy equal to −6. the two photons move with equal and opposite momentum.

namely. However. Namely. the transition amplitude for this decay takes the form Mi→f = (b k · (ǫ1 × ǫ2 )) ψpositronium . This result can be more directly seen when we study charge conjugation in the next section. (11. Since the transition amplitude is a scalar under parity while P (11. the electron and the positron must have relative negative intrinsic parity.81) we conclude that the positronium state must be a pseudoscalar under parity.11.80) k · (ǫ1 × ǫ2 ) −→ −k · (ǫ1 × ǫ2 ) . (11. in fact. in the rest frame of the positronium. Let us recall from our earlier discussions that antiparticles are related to negative energy states.79) Consequently.83) In other words. This is.1 Parity 433 is experimentally observed that the two are perpendicular. for this decay we have ǫ1 · ǫ2 = 0. the relative intrinsic parity between Dirac particles and antiparticles is negative. (11. the total parity of the state is given by ηTOT = ηe− ηe+ . On the other hand. a very general result for fermions.49)) . Let us choose two representative solutions (of positive and negative energy) of the Dirac equation in the rest frame (see (2. it can also be seen in an intuitive manner from the structure of the spinor functions as follows. we must have ηTOT = ηe− ηe+ = −1.82) and for this state to represent a pseudoscalar state. (11.

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

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

Furthermore. J → −J. ∂t B = ∇ × A.436 11 Discrete symmetries ψγ5 ψφ. B → −B. if we know the interaction Lagrangian or Hamiltonian density for a given process. E → −E. E = − classically we see that charge conjugation leads to (11. ∂t ∂E + J. since ∂A − ∇A0 .2 Charge conjugation To understand the discrete symmetry known as charge conjugation. ∇×E = − ∇×B = ∂B . (11. 11. let us go back to Maxwell’s equations in the presence of charges and currents ∇ · E = ρ. (11.92) Conversely.93) We note that this set of equations is invariant under the discrete transformations ρ → −ρ.95) . ∇ · B = 0.94)) and is denoted by C. ∂t (11.94) 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. 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.

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

we have (|ηφ |2 = 1) φ(x) −→ ηφ φ† (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). (11.101) C Thus the field transformations in (11. We have already discussed (see sections 7. It follows now that under such a transformation. ∗ φ† (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).100) is the appropriate charge conjugation transformation for complex scalar field. As a result. (11.102) . we continue to use “†” instead of “∗” so that the passage to quantum theory will be straightforward). C (11.100) would define the appropriate charge conjugation transformation for the current. then C φ(x) −→ φC (x) = ηφ φ† (x). (11. under a charge conjugation. Therefore. ∗ φ† (x) −→ (φC )† (x) = ηφ φ(x).438 11 Discrete symmetries J µ = i φ† ∂ µ φ − (∂ µ φ† )φ .100) with |ηφ |2 = 1. this further supports the fact that (11.99) We note that if we treat φ(x) as a classical function (even though the functions are classical. C C C Aµ (x) −→ −Aµ (x).2 and 7.

(11.2 Charge conjugation 439 Fµν (x) −→ −Fµν (x). the operator ordering which works for the current involving a product of bosonic operators is to symmetrize the product (this can also be checked to be equivalent to . As is known.105) C since the operators φ and φ† do not commute with ∂µ φ† and ∂µ φ respectively.11.104) 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† . 4 (11. we note that if we define the electromagnetic current in the quantum theory as in (11. As a result. we have to choose an operator ordering which we have not taken care of in the above discussion. For example.100) do not lead to the correct transformation properties for the current operator.103) Therefore. C ∗ ∗ (Dµ φ(x))† = (∂µ − ieAµ )φ† (x) −→ ηφ (∂µ + ieAµ )φ = ηφ Dµ φ. C C (11. The problem. can be traced to the fact that in going from a classical theory to the quantum theory. of course.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)† . it would appear as if the field transformations (11.99). spin zero particles interacting with photons described by 1 L = (Dµ φ)† (Dµ φ) − m2 φ† φ − Fµν F µν . then under the transformation (11. the Lagrangian density of charged.

(11. As we have discussed earlier. 2 µ Jsym (x) = (11.3) µ 0|Jsym (x)|0 = 0.440 11 Discrete symmetries normal ordering the product). In such a case. But this has the interesting consequence that while (remember the Dirac vacuum picture) 0|J µ (x)|0 −→ ∞.108) This is esthetically more pleasing and treats the vacuum in a more symmetrical way.110) .106) leads to (this should be compared with the normal ordered charge defined in section 7.109) which brings out the particle-antiparticle symmetry in a more direct way.106) 2 µ Jsym (x) = and it is easy to check that this current will transform properly (namely.107) µ the effect of the infinite constant in Jsym in (11. 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). under a charge conjugation (11. will change sign) under charge conjugation. If we symmetrize the products in the definition of the current. it is described by a Hermitian field operator φ† (x) = φ(x). (11. we obtain i φ† ∂ µ φ + (∂ µ φ)φ† − (∂ µ φ† )φ − φ∂ µ φ† . Note that we can also write the symmetrized current as 1 µ (J (x) − J Cµ (x)) . (11. if a spin-zero field theory describes charge neutral particles which are their own antiparticles.

As an example. one can show that for all Hermitian spin zero fields that we know of (scalar or pseudoscalar) ηφ = 1.112) In this case. or. C (11.115) since the charge parity of the photon is (−1). . we expect 2 ηφ = 1. (11. so that in these cases (11. the sign of the charge parity cannot be fixed from theory. ηφ = ±1.111) From the fact that two charge conjugation operations should return the field to itself.113) φ(x) −→ φC (x) = φ(x). the total charge parity of the final state is (+1). (11. But by analyzing various physical processes where charge conjugation is a symmetry or by analyzing the invariance of various interaction terms under charge conjugation.11. let us note that in the decay π 0 → 2γ. If charge parity is to be conserved in this process (this corresponds to an electromagnetic process since photons are involved).114) Such fields or particles are known as self-charge-conjugate and simply describe particles that are their own antiparticles. we conclude that the π 0 meson must also have charge parity (+1) and similar arguments hold for other mesons.2 Charge conjugation 441 φ(x) −→ φC (x) = ηφ φ† (x) = ηφ φ(x). C (11.

117).118) we would have C i(γ µ )T (∂µ − ieAµ ) + m ψ = 0. In fact. The adjoint equation is easily obtained to correspond to ← ψ(x) iγ µ ∂µ −ieAµ (11. Let us next analyze a system of Dirac fermions in- teracting with an electromagnetic field. from (11.2 Dirac field. (11.121) . interacting with an electromagnetic field. or.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 (11.120) (11.116) +m = 0. (11. or. we obtain i(γ µ )T (∂µ − ieAµ ) + m ψ = 0. If the fermions are minimally coupled to the photons.119) where we have identified the charge conjugate function ψ C as describing the positron. T T T (11. To see the relation between ψ C (x) and ψ(x).117) Taking the transpose of the adjoint equation (11.119) that the two equations are very similar in structure.442 11 Discrete symmetries 11. Thus.2. iC(γ µ )T C −1 (∂µ − ieAµ ) + m Cψ = 0.118) and (11. (iγ µ (∂µ − ieAµ ) − m) Cψ = 0. it would satisfy the equation (iγ µ (∂µ − ieAµ ) − m) ψ C (x) = 0. the equation of motion takes the form (iγ µ (∂µ + ieAµ ) − m) ψ(x) = 0. let us note from (11. we note that if we can find a matrix C satisfying C(γ µ )T C −1 = −γ µ .

2 Charge conjugation 443 which can be compared with (11. (11.122) (11. namely.125) . T (11.124) and hence from the generalized Pauli theorem (see (1.11. from C −1 γ µ C = −(γ µ )T .128) (11. (11. C −1 γ µ C = −(γ µ )T . we can also show that σ µν C = − (σ µν C)T . −(γ µ )T + = 2η µν ½. we can identify (with |ηψ |2 = 1) ψ C (x) = ηψ Cψ .127) (11. we obtain γ µ C = −C(γ µ )T = −C T (γ µ )T = − (γ µ C)T . it follows that there must exist a similarity transformation which relates −(γ µ )T and γ µ . namely. Therefore we conclude that if we can find a matrix C with the property C(γ µ )T C −1 = −γ µ . let us assume that instead C is symmetric. The important symmetry property that C must have is that it should be antisymmetric. C T = C. Similarly. (This property depends on the dimensionality of space-time. −(γ µ )T .) To see this. In this case.126) (11.119). or. In 4-dimension C T = −C.92)).123) That such a matrix exists can be deduced from the fact that −(γ µ )T also satisfies the Clifford algebra.

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

(11.136) µ = −Janti (x).134) = 1 ψ α (x) (γ µ )αβ ψβ (x) − ψβ (x) (γ µ )αβ ψ α (x) 2 1 T ψ(x)γ µ ψ(x) − ψ T (x)(γ µ )T ψ (x) . 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. where we have used (11. we can obtain the antisymmetrized current as µ Janti (x) = (11. (11.122) in the intermediate steps. this current transforms properly under charge conjugation. (11.137) whereas 0|J µ (x)|0 −→ ∞.2 Charge conjugation 445 J µ (x) = ψ(x)γ µ ψ(x).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 .11. Therefore.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.

we obtain T C −1 γ5 γ µ C = C −1 γ5 CC −1 γ µ C = γ5 − (γ µ )T (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. 2 (11. = i γ 0γ 1γ 2γ 3 (11. (11.142) 2 Therefore.140) − (γ 1 )T T − (γ 2 )T T − (γ 3 )T T = γ5 . . if we define properly antisymmetrized Dirac bilinear combinations as = − (γ µ γ5 )T = (γ5 γ µ )T .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 .139) Let us next calculate how the Dirac bilinears transform under charge conjugation. We have already seen in (11.122) that the charge conjugation matrix C satisfies C −1 γ µ C = −(γ µ )T .446 11 Discrete symmetries µ Janti (x) = 1 µ (J (x) − J Cµ (x)) .

148) .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 . P. A.144) 2 (If we think for a moment. T. . (11. (11.147) where the phase ǫα for the different classes has the form 1. (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). for (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 . . ψψ − ψ T ψ (11.143) then it is clear that under charge conjugation the scalar combination would transform as C ψψ anti −→ = = 1 C C ψ ψ C − (ψ C )T (ψ )T 2 1 T |ηψ |2 ψ T C −1 Cψ − ψC T − (C −1 )T ψ 2 1 T = ψψ anti . (11.11.2 Charge conjugation 447 ψΓ(α) ψ anti ≡ 1 2 ψΓ(α) ψ − ψ T Γ(α) T ψ T . α = V. for α = S. ǫα = −1.) We have already seen that ψγ µ ψ −→ − ψγ µ ψ C anti anti .

t) = ηψ C(ψ )T (x. Next. t) P C −ηψ ηψ γ 0 C(γ 0 )T ψ ∗ (−x. we have already seen this from the analysis of the positronium decay in (11.83). t)γ 0 )T C P ηψ ηψ Cγ 0 C −1 C(γ 0 )∗ ψ ∗ (−x. t) C P −ηψ γ 0 ηψ Cψ (−x. t) −→ ψ (x. but this derivation is simpler and more general. (11. t) −→ (ψ C )P (x. the antiparticle will transform as C ψ C (x.448 11 Discrete symmetries Let us next recall from (11. T (11. P ψ(x.73) that under parity a Dirac particle transforms as P ψ(x. t) P −ηψ γ 0 ψ C (−x.151) Here we have used the fact that γ 0 is Hermitian in our metric. T T (11.150) under parity.149) Since the antiparticle is described by C ψ C (x.152) Of course. t) −→ ψ P (x. the antiparP ticle will have the intrinsic parity −ηψ so that the relative particleantiparticle parity will be given by P P P ηψ −ηψ = −(ηψ )2 = −1. t) P P = = = = = C P ηψ ηψ C(ψ(−x. P P P (11. Thus P if ηψ represents the intrinsic parity of a Dirac particle. t). t) = ηψ γ 0 ψ(−x. t). let us note that since the charge conjugate wave function is defined to be C ψ C = ηψ Cψ . t) = ηψ Cψ (x. t) = ηψ ψ(−x. T we obtain . t). t)γ 0 .

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) . / where (11. (11.155) On the other hand.153) As a result.156) T T C = ηψ Cψ = ψ C . we see that if γ5 ψ = ψ.157) In other words.154) T (11.158) then from (11. (p − m)ψ C = 0.159) . namely.6 and have used this in the construction of the standard model in section 14. then it follows that C γ5 ψ C = −ηψ C − ψ (11.11. / (11.3. if γ5 ψ = −ψ. T (11. the antiparticle of a particle with a given handedness has the opposite handedness.2. of course. Let us recall that if a free Dirac particle satisfies the equation (p − m)ψ = 0. (11. 11.119) we see that the antiparticle must also satisfy the same free equation. already seen this in the study of neutrinos in section 3. We have. we have C γ5 ψ C = −ηψ Cψ = −ψ C .3 Majorana fermions.

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. From the definition of charge conjugation 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) .450 ψ C = ηψ Cψ . and would also satisfy the free Dirac equation (p − m)ψ = (p − m)ψ C = 0.123) we see that a Dirac particle which is its own antiparticle must satisfy ψ = ψ C = ηψ Cψ . They can be massive and will have only two independent degrees of freedom (because of the constraint in (11. namely.163) T (11. 2 = (11.164) that for Majorana fermions (ψ = ψ C ) µ Janti (x) = 0. These are charge neutral particles described by real quantum fields φ† = φ. Let us note from the definition of the antisymmetrized current in (11. T 11 Discrete symmetries (11. The answer to this question is. (11. in fact.165) so that the Majorana fermions are charge neutral and hence cannot have any electromagnetic interaction.160) We have also seen that there are bosons in nature which are their own antiparticles. quite obvious.162) Such particles are known as self charge conjugate fermions or Majorana fermions. (11.162)). / / (11. .

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

a Weyl fermion cannot have mass. (11. we can write ( we will choose C ηψ = 1) C / L = iψ L ∂ ψL + mψL ψL T = iψ L ∂ ψL − mψL C −1 ψL . Obviously. note that if we go beyond the conventional Dirac mass term in the Lagrangian density. note that any majorana particle would necessarily imply some sort of violation of a fermion number which is not seen in nature. Such a mass therm is known as a Majorana mass term and in writing this term we have obviously given up something to which we will return in a moment.171) Clearly. if we require conservation of fermion number. . Similarly. of course. the neutron and the neutrino are charge neutral and hence are prime candidates to be Majorana particles. the mass term is not zero and we can give a mass to the Weyl fermion. Let us next ask whether there are any particles in nature which we can think of as Majorana particles. if (once again ν denotes the antineutrino) ν = ν. / (11. However. In fact. in this case. Obviously. if (here n represents the antineutron) n = n. namely.172) then this would imply violation of baryon number which is not seen at present. It is now clear that the Majorana mass term would violate conservation of fermion number and this is what we give up in trying to write a mass term for the Weyl fermion. (11.173) this would imply violation of lepton number which is also not seen yet. But such an identification is not compatible with the experimental results. 2 2 (11. the essence of the statement that a Weyl fermion is massless.170) This is.452 11 Discrete symmetries 1 1 ψ L ψL = ψ (1 + γ5 ) × (1 − γ5 ) ψ = 0.

It is.4 Eigenstates of charge conjugation. . It follows now that ˆ C Q|Q. . . . . so that comparing the two relations we obtain (11. . of course.174) where . . of course. = −Q| − Q. means that a one photon state is an eigenstate of charge conjugation with eigenvalue −1. On the other hand. . .2 Charge conjugation 453 We can now ask why anyone would then like to study such particles or mass terms. . This. we have ˆ ˆ QC|Q. . .2MeV) (the present trend is to quote the lowest mass bound along with the differences of mass squares for the neutrinos determined from the neutrino oscillation experiments) suggest that the neutrinos may indeed have small masses. The answer to this lies in the fact that while no one has really measured the mass of the neutrino. mντ < 18.11. . . = QC|Q. . Independent of whether supersymmetry is a symmetry of nature. To understand the eigenstates of charge ˆ conjugation in general.2. . . . . . . mνµ < 190keV.176) . its theoretical studies would require the study of the properties of such particles. . . Similarly. let us note that if Q denotes the operator for charge (electric). . We have seen that the pho- ton has odd charge conjugation parity. . (11. . the direct experimental limits (mνe < 2ev. Another reason for the study of Majorana particles is that they are fundamental in the study of supersymmetry which we will not go into. = Q|Q.175) (11. up to the experiments to decide whether this is really what happens. . = Q| − Q. . then acting on an eigenstate it gives ˆ Q|Q. . The Majorana mass is one way to understand such small masses and can arise naturally in grand unified theories where both baryon and lepton numbers can be violated in small amounts. refer to other quantum numbers the state can depend on. 11. = Q| − Q. we have also seen that a Hermitian spin zero field has charge parity +1.

or. 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. . (11. ℓ = |π + π − . ℓ . ℓ = (−1)ℓ |π − π + . we would have C|π − π + . Let us note. a state such as |π − π + can be an eigenstate of C. On the other hand. Note that C|π − π + = |π + π − .178) which clarifies the point. = |K . ˆ C.454 11 Discrete symmetries ˆ ˆ C Q = −QC. the charge operator does not commute with the charge conjugation operator and. = |π + p . (11. Thus if the state has well defined angular momentum quantum number. they cannot have simultaneous eigenstates unless the charge of the state vanishes. consequently. not all charge neutral states would be eigenstates of charge conjugation operator.180) so that such a state would be an eigenstate of charge conjugation with eigenvalue (−1)ℓ . 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.177) In other words. Similarly. that while all eigenstates of C must be charge neutral. We have. (11. It is easy to see that C|n C|π − p C|K 0 = |n . 0 (11. Q + = 0. however. s respectively. of course.

ℓ. |e− e+ . Note. namely.1 which can. ℓ. / (11.186) Let us also note here that if we have an amplitude involving N photons shown in Fig. ℓ. that upon requiring conservation of charge parity.2 Charge conjugation 455 C|e− e+ . consistent with our earlier discussion. say that if charge conjugation is a symmetry. 11. would lead to the result that 1 = (−1)n . that the positronium in the ℓ = 0 = s state decays into two photons.185) (11.181) = |e+ e− . (11. the decay π 0 → nγ. in particular. this is an eigenstate of charge conjugation with charge parity (−1)ℓ+s . s → nγ. s In other words. then the charge parity must be conserved. (11. for example.11. Thus if we look at the decay of the positronium to n photons. s . (11. If charge conjugation is a symmetry of the theory. s = −(−1)ℓ (−1)s+1 |e− e+ . describe the decay . ℓ. namely. of course. ℓ. then π 0 −→ 3γ.184) It would.183) This is.182) then the conservation of charge parity would require (recall that the photon has charge parity −1) (−1)ℓ+s = (−1)n . (11. s = (−1)ℓ+s |e− e+ . similarly.

188) In other words. it is also known that charge conjugation symmetry holds true in strong interactions also. As we have seen. then C † HC = H. This result goes under the name of Furry’s theorem.189) where U denotes the time evolution operator. electromagnetic interactions are invariant under charge conjugation. (11.187) then conservation of charge parity would require (electromagnetic interactions are invariant under charge conjugation) Figure 11.456 11 Discrete symmetries nγ → (N − n)γ. if the strong interaction Hamiltonian respects charge conjugation symmetry. (−1)N = (−1)2n = 1. and it follows from this that C † U C = U. (11. Therefore. any amplitude involving an odd number of photons must vanish if charge conjugation is a symmetry. or. In fact.1: An amplitude with N photons. then we must have . note that if a Hamiltonian is invariant under charge conjugation.190) (11. Similarly. (11. (−1)n = (−1)N −n .

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

195) We note that the Newton’s equation is invariant under time reversal since it is second order in the time derivative. time reversal or time reflection corresponds to the space time transformation where we reverse only the sign of time. T (11. ∇ · B(x) = 0. Maxwell’s equations in the presence of sources ∇ · E(x) = ρ(x). 11.193) 2 2 where ηTOT denotes the total phase under the transformation.194) W 2 2 This has well defined transformation properties under CP transformations and leads to CP invariance. T t −→ −t.3 Time reversal Classically. ∂t (11. (11. (11.196) are also form invariant if we define . x −→ x. ∇ × E(x) = − ∇ × B(x) = ∂B . Similarly. namely. 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 .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 . ∂t ∂E(x) + J.

−t).3 Time reversal 459 ρ(x. −t). Thus assuming that time inversion is also implemented by such an operator T . therefore. A(x. t) −→ A0T (x. While some of the macroscopic equations are invariant under time reversal. From the definition of the electric and the magnetic fields E = −∇A0 − B = ∇ × A. The first problem which we face. t) = −A(x. the entropy or disorder increases with time. in statistical mechanics we know that microscopic reversibility does hold. t) = E(x. t) −→ AT (x. Thus we would like to investigate whether time reversal can be a symmetry of quantum mechanical systems. Macroscopically. t) = −B(x. t) = −J(x. −t).11.197) ∂A . symmetry transformations normally are implemented by linear operators which are also unitary. t) −→ BT (x. t) = ρ(x. t) −→ JT (x. we determine that under time reversal A0 (x. J(x.57). is that the time dependent Schr¨dinger equation is first order in the time derivative just o like the diffusion equation and hence cannot be invariant under a naive extension of the time inversion operation to the quantum theory. B(x. t) = A0 (x. Note that in quantum mechanics. we know that there is a sense of direction for time. E(x. T T T T T T (11. we know that macroscopic physics is not. However. ∂t (11. −t). The diffusion equation which is first order in the time derivative is not symmetric under t → −t. −t). we will have . t) −→ ET (x. t) −→ ρT (x. is opposite from their behavior under parity in (11. however. namely.199) The transformation of the Aµ (x) potentials under time reversal. −t).198) (11.

(11. ∂t Thus we see that in such a case |ψ and T |ψ satisfy different equations and hence this operation cannot define a symmetry of the time dependent Schr¨dinger equation. (11. o However. [T . In fact. 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.460 11 Discrete symmetries |ψ −→ |ψ T T = T |ψ .200) If we assume that the time independent Hamiltonian is also invariant under time inversion. an antilinear operator T satisfies .202) would lead to iT or. ∂t (11. −i ∂|ψ ∂t = T H|ψ . H] = 0.203) ∂ (T |ψ ) = H(T |ψ ). By definition.201) then it follows that under a time inversion. namely. the time dependent Schr¨dinger equation o i ∂|ψ = H|ψ . Let us for the moment give up the assumption that the operator T is linear. (11. let us assume that T is an antilinear (antiunitary) operator.

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

209) Let us derive the transformation properties of some of the operators under time inversion. in particular the third relation). the role of the initial and the final states are interchanged and the transition amplitude of interest would be ψi |ψf . on the other hand. (11. t) = x|ψ(t) −→ ψ T (x. let us note from (11.211).462 11 Discrete symmetries In the time reversed process. It is reassuring to note that the canonical commutation relations are unchanged by the ηX = 1.208) which is indeed consistent with the definition of time inversion transformations given in (11. we know that x −→ x.211) T PT † = ηP P = −P.210) Thus from Ehrenfest’s theorem we would expect the corresponding operators to transform as T XT † = ηX X = X. Pj ] T † = T (Xi Pj − Pj Xi ) T † = Xi (−Pj ) − (−Pj )Xi = − [Xi .204) as well as (11.204) that ψ(x. t) = x|ψ(−t) = ψ ∗ (x. Pj ] .212) where we have used (11.207) (11. under time inversion. T (11. In particular. . It follows now that T [Xi . (11. Classically. T (11. T p −→ −p. T ∗ (11.204) (see. ηP = −1. −t). Thus. we expect ψf |ψi −→ ψi |ψf .

(11. m L3 T |ℓ. It follows now that for the angular momentum eigenstates satisfying L2 |ℓ. under time inversion each component of the angular momentum operator would change sign which is consistent with the classical result. Since the angular momentum operators can be written as Li = ǫijk Xj Pk . m ). (11. m we have L2 T |ℓ. m L3 |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.215) m|ℓ. −m . ℓ(ℓ + 1)(T |ℓ.3 Time reversal 463 time inversion transformation (note from (11. m .11. m = 2 2 = = ℓ(ℓ + 1)|ℓ.204) that i → −i under the anti-linear time reversal transformation).217) . it follows that T L2 T † = T Li Li T † = (−Li ) (−Li ) = Li Li = L2 . (11. m with T = Cm |ℓ. m = T L2 |ℓ. m = − m(T |ℓ. (11. (11.213) Namely.214) so that the L2 operator remains unchanged under time inversion.216) = −T L3 |ℓ. m . Let us note here that the naive extension of the time inversion transformation with a linear operator would have led to an inconsistency in the commutation relations. m ). We see that the state T |ℓ. Therefore. we obtain T Li T † = ǫijk T Xj Pk T † = −ǫijk Xj Pk = −Li . m = |ℓ.

−t).218) (11.223) . By convention. φ).220) (−1)m Yℓ. in general. t) −→ AµT (x. (11. mj = T |j.464 11 Discrete symmetries |Cm |2 = 1. (11. m = T |ℓ.221) It can be shown that the above relation holds true even for half integer angular momentum values.m (cos θ)e−imφ 4π (ℓ + |m|)! (11.m (θ. where we have used the fact that the spherical harmonics for m < 0 are defined with a phase +1.3. we can write T |j. φ) = = (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. We have already seen that under time reversal Aµ (x. T (11. Thus. −mj . t) = ηAµ Aµ (x. φ|ℓ. −m . the constant (phase) is chosen to be Cm = (−1)m = i2m .m (θ. mj = (i)2mj |j.1 Spin zero field and Maxwell’s theory.222) 11. m (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. we can indeed write T |ℓ. φ) = = T θ. namely.m (cos θ)eimφ 4π ℓ + |m|)! ∗ −→ Yℓ. which follows because under time inversion (say for m > 0) (11. m = (i)2m |ℓ.−m (θ.219) Yℓ.

of course. T T (11.224) It follows.226) Similarly. invariant under time reversal as we have already seen and as can be checked directly. t). where T (11. −t) F µν (x.11. t) (11. −t). we have seen that under time reversal. t) −→ J µT (x. t) −→ Fij (x. t) = ηJ µ J µ (x. therefore.228) The Maxwell theory is. T Fij (x.3 Time reversal 465 where ηA0 = 1. that Fµν (x. ν µ so that T F0i (x. t) − ∂ν Aµ (x. ηA = −1. t) = −Fij (x.225) T T −→ ∂µ AT (x.229) = d4 x S . (11.227) ηJ 0 = 1. t) −→ F0i (x − t) = F0i (x. −t). (11. −t)Aµ (x. 1 − Fµν F µν − J µ Aµ 4 1 T µν T T − J µT AT −→ d4 x − Fµν F µ 4 1 = d4 x − Fµν (x. ηJ = −1. (11. −t) 4 1 = d4 x − Fµν (x)F µν (x) − J µ (x)Aµ (x) 4 = S. t) = T ∂µ Aν (x. −t). the current four vector transforms as J µ (x. t) − ∂ν AT (x. −t) − J µ (x.

T (11. for a spin zero field. we have φ(x) −→ φT (x) = ηφ φ∗ (x. −t) ∂t2 (11.230) where the matrix T mixes the components of the function and |ηψ |2 = 1. the Klein-Gordon equation is invariant under time inversion. 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.234) ∂2 − ∇2 + m2 φ(x) = 0. −t)) −→ i (φT ∗ (x)∂ µ φT (x) − (∂ µ φT ∗ (x))φT (x)) T = −i (φ∗ (x.231) (11. −t)). −t) − (∂ µ φ∗ (x. −t)∂ µ φ(x. ∂t2 (11. (11. For example. −t))φ∗ (x. 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. 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. ∂2 − ∇2 + m2 φT (x) ∂t2 ∂2 − ∇2 + m2 ηφ φ∗ (x. −t). ∂t2 ∂2 − ∇2 + m2 φ∗ (x) = 0. −t)∂ µ φ∗ (x.232) That is. −t))φ(x. −t).235) .233) T (11. −t) − (∂ µ φ(x.466 11 Discrete symmetries In general. Furthermore.

−iγ 0∗ −iγ 0∗ iγ 0∗ ∂ ∂ − iγ i∗ i − m ψ ∗ (x. or. ∂t ∂x ∂ ∂ − iγ i∗ i − m ψ ∗ (x. −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) = 0.238) ∂ ∂ − iγ i∗ i − m ψ ∗ (x. (11. −t))φ(x. −t) − ( φ∗ (x. we expect that under time reversal ψ(x. t) −→ −i φ∗ (x. −t). −t)∂ i φ(x. leads to (upon complex conjugation) (−iγ µ∗ ∂µ − m) ψ ∗ (x) = 0. −t).230)).239) (11. ∂t ∂x . −t))φ(x. t) = 0.197). of course.3 Time reversal 467 Therefore.2 Dirac fields. the correct behavior of the current four-vector under time inversion as we have already seen in (11. −t) = which is.3. −t) = 0. −t) = 0. 11. ∂(−t) ∂x (11. we obtain J 0 (x.11. −t) − (∂ i φ∗ (x. −t) ∂t ∂t J 0 (x. t) −→ −i φ∗ (x. or. (Here ∂ we have used the fact that J 0 (x. T (11. From our general discussion (see (11.) Thus we conclude that the spin zero charged particles interacting with photons can be described by a theory which will be time reversal invariant. −t). or. −t) is defined with − ∂t .236) J i (x. t) −→ ψ T (x. −t) = T T ∂ ∂ φ(x. t) = ηψ T ψ ∗ (x.

γ ˜ (11.240) Therefore. comparing the two equations we see that if we can find a matrix T such that T γ 0∗ T −1 = γ 0 . t). t) were to be a solution of the Dirac equation. iγ 0 iγ 0 ∂ ∂ + iγ i i − m ψ T (x) = 0. ˜ γ i = −γ i .243) γ 0 = γ 0. γ ν∗ } = 2η µν ½. or. then the time reversed wave function (11.241) ψ T (x) = ηψ T ψ ∗ (x.245) . or. ∂t ∂x ∂ ∂ + iγ i i − m T ψ ∗ (x. −t). and the matrices γ µ defined as (˜ µ = γ µ† ) ˜ γ (11. −t) = 0. We note that the matrices γ µ∗ satisfy the Clifford algebra {γ µ∗ . ˜ also satisfy the Clifford algebra {˜ µ . γ ν } = 2η µν ½.242) would satisfy the same Dirac equation as ψ(x.468 11 Discrete symmetries On the other hand. if ψ T (x.244) (11. we should have (iγ µ ∂µ − m) ψ T (x) = 0. T γ i∗ T −1 = −γ i . ∂t ∂x (11. (11.

Thus we can identify T γ µ∗ T −1 = γ µ .251) In other words.246) (11.122) that Cγ µT C −1 = −γ µ . or.248) Thus we see that the inverse relation can be written as T −1 γ µ T = γ µT . the matrix CT −1 anticommutes with all the γ µ ’s and hence must be a multiple of γ5 .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).247) To determine the explicit form of T . (11. Combining the two results we obtain (11. T = −γ i∗ = − γ i† = γ iT . CT −1 γ µ CT −1 −1 = −γ µ . (11.249) (11. ˜ leading to T γ 0∗ T −1 = γ 0 . Choosing . Let us also recall from (11. (11. They must also be related to each other through a similarity transformation.11. T γ i∗ T −1 = −γ i . let us invert these relations and write T T −1 γ 0 T T −1 γ i T = γ 0∗ = γ 0† = γ 0T .250) CT −1 γ µ T C −1 = Cγ µT C −1 = −γ µ .

255) Furthermore. t) = ηψ T ψ ∗ (x.252) (11.254) (11. and using the definition (11. t) −→ ψ (x. Let us next analyze the behavior of the current four-vector under time inversion . (11. Therefore we see that the time inversion transformations T ψ(x. (11.130) C = γ 0γ 2. we can show that in four space-time dimensions. as in the case of charge conjugation. Note that with this choice. t) −→ ψ T (x.253) (11. independent of any representation (the superscript denotes transposition) (T )T = −T. −t)T −1 γ 0 . t) = ηψ ψ T (x. we determine (in our representation) T = iγ 1 γ 3 . the T matrix satisfies T = T † = T −1 . −t) = ηψ γ 0 T ψ (x. −t) = T ηψ T γ 0T ψ (x.257) with (|ηψ |2 = 1) defines a symmetry of the Dirac equation. (11. −t).470 11 Discrete symmetries CT −1 = −γ5 = −iγ 0 γ 1 γ 2 γ 3 .256) so that the time inversion matrix in the spinor space is antisymmetric. T T T ∗ ψ(x.

−t)T −1 γ 0 T ψ (x. −t) T T = −J i (x.3 Time reversal 471 J µ (x) = T ψ(x)γ µ ψ(x) T −→ ψ (x)γ µ ψ T (x) = = so that we have T |ηψ |2 ψ T (x. −t) − ψ(x. −t)γ i ψ(x. (11. (11. we have . −t) = = T T T ψ T (x. Furthermore. −t). −t) T T T T = J 0 (x.259) which is the behavior of the current four vector under a time inversion and we can identify the transformations in (11.257) to correspond to time inversion for a Dirac system. we also see that a Dirac system interacting with photons would be invariant under such a transformation. −t). −t)T −1 γ 0 γ µ γ 0 T ψ (x. −t)γ i ψ (x. −t) ψ(x.11. (11. −t)T −1 γ 0 γ i γ 0 T ψ (x.260) (this is very much like the charge conjugation in (11. −t) = = −ψ T (x. t) −→ ψ T (x.122) except for the −1 sign). −t)γ 0 ψ(x. T J i (x. −t). Before we examine the transformation properties of Dirac bilinears under time inversion. let us note that since T −1 γ µ T = γ µT .258) T J 0 (x) −→ ψ T (x. −t)T −1 γ 0 γ µ γ 0 T ψ (x. −t)γ 0 ψ (x. −t) ψ T (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 . In particular. 2 (11. t)ψ(x.263) Namely. −t)T −1 γ 0 γ 0 T ψ (x.262) ψ(x. T T ψ(x. t) −→ ψ (x. −t)T −1 γ 0 γ5 γ 0 T ψ (x.261) The behavior of the Dirac bilinears under time reversal can now be obtained in a simple manner. −t) − ψ(x. ψ(x. t)γ5 ψ(x. the 0 transformation of J5 = −ψ † γ5 ψ shows that under time inversion . −t)ψ(x. t) −→ ψ (x. t)ψ T (x. t) −→ −ψ(x. −t)γ5 γ i ψ(x. the axial vector (current) behaves exactly the same way as the vector (current) under a time inversion. we can show that T ψ T (x. −t)ψ(x. t) −→ ψ(x. t)γ5 ψ T (x. t) = = = Similarly.472 11 Discrete symmetries T −1 γ5 T T −1 γ5 γ µ T T −1 σ µν T T = γ5 . −t). T T T ψ(x. −t) ψ(x. −t) ψ(x. T T (11. t)γ5 ψ i ψ(x. −t). T = γ5 γ µ = (γ µ γ5 )T = − (γ5 γ µ )T . −t) −ψ(x. (11. −t)γ5 ψ(x. t)γ5 ψ 0 ψ(x. −t)γ5 γ 0 ψ(x. −t). −t)γ5 ψ(x. −t). t) = = = T ψ T (x.

266) so that a left-handed spinor remains left-handed under time reversal and a right-handed spinor remains right-handed. −t). then the transition amplitude can be represented as (11.3 Consequences of T invariance. the chirality of a spinor does not change. −t). Let us also note that if we are looking at a reaction of the form a + b −→ c + d. Let us recall that under time reversal. We could have expected this from the definition (see (3. t)σ 0i ψ(x. j.11. |p| |p| |p| h= (11. It is also easy to see now that under time reversal. t) −→ ψ(x. T T (11. −t)σ ij ψ(x. t) −→ −ψ(x. the tensor bilinears have precisely the same behavior under time reversal as the electromagnetic field strength tensor F µν .3 Time reversal 473 the helicity does not change.3.121)) s·p s · p T (−s) · (−p) −→ = = h. namely. −t)σ 0i ψ(x.264) We can also show that ψ(x.265) In other words. mj −→ (i)2mj |α.222)) T |α. T = ηψ γ5 T ψ ∗ = ηψ T γ5 ψ ∗ † = ηψ T γ5 ψ ∗ γ5 ψ T = ηψ T (γ5 ψ)∗ . −mj . p. j. (11. 11. ψ(x.267) where α stands for all the other quantum numbers of the state.268) . (11. −p. a state with given angular momentum transforms as (see (11. t)σ ij ψ(x.

(11.474 11 Discrete symmetries f |S|i . d .274) (11. namely. then the transition amplitude can be represented as i|S|f . and c + d −→ a + b. (11. it does not follow a priori that | i|S|f |2 = | f |S|i |2 . experimentally this is seen to hold in many processes.275) .270) Since the time evolution operator is not Hermitian. S † = S.273) On the other hand. (11.272) (11. b . to obtain the spin of π-meson etc. |f for the final state and S denotes the S-matrix or the time evolution matrix. if we are looking at the reciprocal reaction c + d −→ a + b. |f = |c. On the other hand. (11. for example.269) where |i stands for the initial state. |i = |a.271) (11. namely. where we are still using the same notation. It is worth emphasizing here that the detailed balance does not say that the rates for the two reactions a + b −→ c + d. This is known as the principle of detailed balance and has been used.

(11.280) . when the principle of detailed balance may hold. Let us next try to understand.3 Time reversal 475 have to be the same. = |β. therefore. pf . pi . Let us next assume that we are dealing with a system whose Hamiltonian is invariant under time reversal. In this case. (11. mf . In fact.277) and furthermore. (11.276) where ρf denotes the density of states for the final state which may not be the same for the two processes. T ST † = S.11. 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 . theoretically. That is. the principle of detailed balance will hold. (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. Let us assume that we are dealing with a system where the interaction Hamiltonian can be treated as a perturbation. h (11. then we have | f |S|i | = 2π| f |Hint |i |δ (Ei − Ef ) = 2π| i|Hint |f |δ (Ei − Ef ) = | i|S|f |. 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.279) Let us choose the initial and the final states to correspond to |i |f = |α. mi .

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

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

if electromagnetic interactions are invariant under P and T . whereas the interaction involving the magnetic dipole moment is invariant under P and T . . this is the Pauli coupling) Hint = µe σ · E + µm σ · B. the electric dipole interaction changes sign under each one of the transformations. σ −→ −σ. σ · B −→ σ · B. B −→ −B. T T (11.63 × 10−25 e-cm.290) We have already seen that under parity and time reversal E −→ −E. σ · B −→ σ · B.292) In other words. Thus the magnitude of any dipole moment (electric) will be of the order of µe ≃ ed ≃ e 10−13 cm = 10−13 e-cm.478 11 Discrete symmetries a magnetic dipole moment. Consequently.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. T T T (11. Let us assume that it also has an electric dipole moment. P P P P P E −→ E. Experiments have obtained extremely small limits on µe . σ −→ σ. (11. (11. (11. B −→ B. so that we have σ · E −→ −σ · E.293) This basically measures the extent to which P and T invariances hold in electromagnetic interactions. Thus the neutron would interact with an electromagnetic field through the interaction Hamiltonian (there is no minimal coupling. Since there is no natural axis for the neutron other than the spin axis.291) σ · E −→ −σ · E. the electric dipole moment would be parallel to its spin. then the electric dipole moment of the neutron must vanish. (µe )exp < 0.

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

the electric charge for particles is equal in magnitude but opposite in sign from that of the antiparticles. In this way. q. q. Let Q denote the operator which measures the charge of a state.) Thus for a one particle state with charge q at rest.480 11 Discrete symmetries In other words. ˆ 11. Namely. ˆ (Recall Q = d3 x ψγ 0 ψ for a Dirac particle. s ˆ α. Since the weak interactions violate parity.4. −s (11. s|Q|α. q. 11. q.300) (11. the equality of particle and antiparticle masses is a consequence of CPT invariance. s ˆ α. Even though C-invariance is violated in weak interactions. 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. (11.298) = −q. In fact. (11. one can show that all the quantum numbers of antiparticles are equal in magnitude but opposite in sign from those of the particles. we have q = = = ˆ α. K 0 and K have the same mass. q. so that ˆ CPT ˆ Q −→ −Q. −s|(−Q)|α. 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− . As we have seen ˆ P ˆ Q −→ Q. q.299) ˆ C ˆ Q −→ −Q. it is CPT invariance which guarantees 0 that.2 Electric charge for particles and antiparticles.3 Equality of lifetimes for particles and antiparticles.301) . for example. ˆ T ˆ Q −→ Q.4.

j. −mj . j. j.302) In such a case. mj = (11. j.4 CPT theorem 481 where PH± P −1 = ±H± . j. j.305) (11. j. (11.304) = (i)2mj −2mj α. mj |H± |α. we can rewrite this relation as β. Furthermore. (11. mj = α. j.11. mj (11. j. using the rotational invariance of physical Hamiltonians. j. mj . mj −→ |β. and the corresponding antiparticle decay channel |α. mj −→ |β.303) The cross terms would have a pseudoscalar character and would vanish since the reaction rate is a scalar quantity. j. mj |H± |β. −mj = α. β. j. 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 |H± |β. mj |H± |α. Let us consider the decay channel of a particle |α. j. j. mj . mj |(CPT )−1 (CPT )H± (CPT )−1 (CPT )|α. We know that β.307) . −mj |H± |β. j. mj . j. (11.306) where we have used the fact that H± would be individually invariant under CPT according to the CPT theorem.

mj |2 ρβ | β. mj |H− |α. There are many other interesting consequences of the CPT theorem including the connection between spin and statistics which we cannot go into. Let us simply note here that since CPT must be conserved in any physical theory and we know that CP is violated in weak interactions by a small amount. mj |2 + | β. mj ∗ . Okubo. However. mj |2 = | β. Physical Review 109. j. mj |H− |α. j. mj |2 ρβ (11. mj |H± |α. mj |H± |α. This follows trivially from the equality of particle and antiparticle masses. τ Here we have assumed that the phase space for the decay products of the antiparticle is the same as that for the decay products of the particle. j. it follows that the weak interactions must violate T -invariance by a small amount as well so that CPT will hold. S. j. . mj |H± |α. (11. therefore. mj |2 . β) which is not relevant for the subsequent discussion. we note here that the individual decay channels may have different life times for particle and anti-particle. namely 2π 1 = τ = 2π β β | β. mj |H+ |α. 984 (1958). j. j. mj |2 + | β. From CPT invariance. j. j. mj |H± |α. j. j. j.310) = 1 . 11. j. mj |H+ |α. j. and τ ) of particles and antiparticles. we have obtained | β.5 References 1. j. (11.482 11 Discrete symmetries Since the Hamiltonians H± are Hermitian we also have β. j.309) which leads to a relation between the total life times (τ .308) Here we have neglected an overall phase factor (depending on α. j. mj = β.

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

This page intentionally left blank .

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

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. (12. namely. As we have seen earlier (see (9. we note that δψ = −iǫ(x)ψ.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). If we now want to promote the symmetry to be local. the phase of ψ remains arbitrary.4) since ∂µ ψ does not transform covariantly and there is no term in the Lagrangian density (12. δψ = iǫ(x)ψ. δ (∂µ ψ) = ∂µ (δψ) = ∂µ (−iǫ(x)ψ) = −i ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ(x). (12.1) nor the corresponding action is invariant under the local phase transformation in (12.486 12 Yang-Mills theory with θ = ǫ = infinitesimal. Consequently.81)). we have δL = iδψ∂ ψ + iψγ µ ∂µ δψ − mδψψ − mψδψ / −imǫ(x)ψψ + imǫ(x)ψψ. namely ǫ = ǫ(x) (or θ = θ(x)). Consequently. the way out of this difficulty is to define a covariant derivative Dµ ψ such that it transforms covariantly under the local transformation. δ (Dµ ψ) = −iǫ(x)Dµ ψ(x).4) = −ǫ(x)ψ∂ ψ + ψγ µ ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ / = (∂µ ǫ(x))ψγ µ ψ. Neither the Lagrangian density (12.1) whose variation would cancel the ∂µ ǫ(x) term.5) (12. (12.7) .

(12. is nondynamical in this theory. e (12. this would transform as δFµν = ∂µ δAν − ∂ν δAµ = 1 (∂µ ∂ν ǫ − ∂ν ∂µ ǫ) = 0.10). That is.7) as describing the minimal coupling of the charged fermions to the electromagnetic field and as we have seen in (9. we note that since the gauge field transforms longitudinally (by a gradient). e (12.10) However.12) Hence the field strength tensor is gauge invariant and the gauge invariant Lagrangian density for the dynamical part of the gauge field can be written as (quadratic in derivatives) 1 Lgauge = − Fµν F µν .86).8) We recognize (12. the field Aµ known as the gauge field. δψ = iǫ(x)ψ(x). 4 (12. δAµ = 1 ∂µ ǫ(x). the Lagrangian density L = iψγ µ Dµ ψ(x) − mψψ. e δAµ (x) = (12.9) is invariant under the infinitesimal local gauge transformations δψ = −iǫ(x)ψ(x).12.1 Non-Abelian gauge theories 487 with the transformation property for the additional field 1 ∂µ ǫ(x). To introduce the kinetic energy part for the gauge field (in order to give it dynamics) in a gauge invariant manner.11) under the gauge transformation (12. (12. if we define a field strength tensor Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . the curl of the field would be unchanged by such a transformation.13) .

2. .80)) 1 LQED = − Fµν F µν + iψγ µ Dµ ψ − mψψ. . Let us now generalize these ideas to theories with more complicated symmetries.16) where θ a denotes the real global parameter of transformation and infinitesimally we have δψ k = iǫa ψ ℓ (T a )ℓk . · · · . 2. (12. Namely.488 12 Yang-Mills theory The total Lagrangian density for QED is then given by (see also (9. 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. Here the T a ’s represent the generators of the internal symmetry group G and are assumed to be Hermitian. (12. / k = 1. Furthermore. dim G. 4 (12. dim G. a δψk = −iǫa Tkℓ ψℓ .10).14) and has the local gauge invariance described in (12. .) The generators (matrices) T a satisfy the Lie algebra of the Lie group G (this is necessary for the transformations to form a group) . . . aT a k = ψU −1 = ψeiθ k . . dim G represents the dimensionality of the symmetry group. (12. (We note here parenthetically that dim G = n2 − 1 for the group SU (n). .15) where k is an internal symmetry index. k a = 1. a = 1.17) with θ a = ǫa = infinitesimal. 2. dim R. . This Lagrangian density is invariant under the global phase transformations aT a ψk → ψk → Uψ ψU † k = e−iθ ψ k. Let us consider a free Dirac theory described by the Lagrangian density (containing several complex fermion fields) L = iψ k ∂ ψk − mψ k ψk .

b. a. c = 1. . the structure constants are nontrivial).18) − 1 . (For semi-simple groups the structure constants can always be chosen to be completely antisymmetric. (12. b. 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 δψ k = iǫa (x)ψ ℓ Tℓk . It is easy to see that the Lagrangian density is invariant under the transformations (12. f 147 = f 246 = f 257 = f 345 = 2 .19) Here we have used the fact that the symmetry generators are Hermitian. (12.) When the generators of the symmetry group 2 do not commute (namely. the structure constants of the group SU (2) can be identified with f abc = ǫabc . . That is. (12.16)).12. f 458 = f 678 = 23 . For example.20) Once again. we already know from the study of angular momenta. 2. . we note that the ordinary derivative acting on the fermion field does not transform covariantly under this transformation. let us consider the infinitesimal local phase transformations a δψk = −iǫa (x)Tkℓ ψℓ . 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. dim G. a. c = 1.1 Non-Abelian gauge theories 489 T a . Let us now try to make this symmetry a local symmetry of our theory.17) (or (12. T b = if abc T c . namely. the symmetry group is known as a non-Abelian group. (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. namely. . 2. 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. .

24) explicitly as a a Dµ ψk = ∂µ ψk + igTkℓ Aa ψℓ = ∂µ δkℓ + igTkℓ Aa ψℓ .26) .22) As in the case of the U (1) theory.490 12 Yang-Mills theory a δ (∂µ ψk ) = ∂µ (−iǫa (x)Tkℓ ψℓ ) a a = −i (∂µ ǫa (x)) Tkℓ ψℓ − iǫa (x)Tkℓ ∂µ ψℓ . / / δL = δψ k (i∂ − m)ψk + ψ k (i∂ − m)δψk = (∂µ ǫa (x))ψ k γ µ (T a )kℓ ψℓ . (12.23) Introducing a new field (gauge field).22). the dimension of the Lie group. We can also write the covariant derivative in (12. there is no other term in the Lagrangian density whose variation would cancel the ∂µ ǫa (x) term in (12.25) This is the generalization of the minimal coupling (in QED) to the present case and the T a ’s are the generators of the symmetry group belonging to the same representation as the fermions.24) (12. 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 . where g denotes the appropriate coupling constant and Aµ = T a Aa . µ µ (12. (12. (12. we write the covariant derivative (in the matrix notation) as Dµ ψ = (∂µ + igAµ ) ψ. We can compare this with QED where the generator is the identity matrix ½ which commutes with every operator. µ (12.20). We note that the number of gauge fields (that are needed to define a covariant derivative) is the same as dim G.21) so that under the local transformation (12.

µ This can also be written as .28).30) a = i (∂µ ǫa ) Tkℓ ψℓ + igǫa f abc (T c )km Ab ψm .27) we conclude from (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 .28) can be written in the form a a a igTkℓ δ Aa ψℓ = igTkℓ δAa ψℓ + igTkℓ Aa δψℓ µ µ µ a a b = igTkℓ δAa ψℓ + igTkℓ Aa − iǫb Tℓm ψm µ µ a b a = igTkℓ δAa ψℓ + gǫa Tkℓ Ab Tℓm ψm . (12.1 Non-Abelian gauge theories 491 Noting that we would like the covariant derivative to transform covariantly under an infinitesimal local transformation. a δ (Dµ ψ)k = −iǫa Tkℓ (Dµ ψ)ℓ . we obtain a igTkℓ δAa ψℓ µ a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a T b km Ab ψm µ −gǫa T b T a km Ab ψm µ km a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a .29) Substituting this into (12.12. T b Ab ψm µ (12.28) We note that the left-hand side in (12. µ µ (12. µ (12.

We can now define a unitary representation of the group as describing the finite symmetry transformation matrix (see (12. (12. µ g (12. . (12.25)) Aµ = T a Aa . µ g (12.16)) . dim G.33) with the covariant derivative defined in (12. the Lagrangian density is a scalar) L = iψγ µ Dµ ψ − mψψ.34) If we use the definition (see (12. δAa = µ 1 ∂µ ǫa − f abc Ab ǫc . (12.33) in terms of matrices as (of course. µ µ g g (12. we can write the Lagrangian density in (12. . a δψ k = iǫa (x)ψ ℓ Tℓk .26) is invariant under the infinitesimal local transformations a δψk = −iǫa (x)Tkℓ ψℓ (x).492 12 Yang-Mills theory a igTkℓ ψℓ δAa − µ 1 ∂µ ǫa + f abc Ab ǫc = 0. . .32) Thus the Lagrangian density L = iψ k γ µ (Dµ ψ)k − mψ k ψk .36) with normal product rules for matrices. µ a = 1.35) 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). 2.31) 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 .

38) To construct the Lagrangian density for the dynamical part of the gauge field in a gauge invariant manner. the tensor representing the Abelian field strength (see (12.37) so that we can write the finite transformations (corresponding to (12.38).39) . (12.12. U † = U −1 . (12.34)) for the field variables in the matrix form as ψ → U ψ. ψ → ψU −1 . ig (12.11)) would transform as fµν = ∂µ Aν − ∂ν Aµ 1 (∂ν U ) U −1 ig 1 (∂µ U ) U −1 ig 1 1 (∂µ ∂ν U ) U −1 − (∂ν U ) ∂µ U −1 ig ig 1 1 (∂ν ∂µ U ) U −1 + (∂µ U ) ∂ν U −1 ig ig → ∂µ U Aν U −1 − −∂ν U Aµ U −1 − = ∂µ U Aν U −1 − −∂ν U Aµ U −1 + = 1 1 (∂µ U ) ∂ν U −1 − (∂ν U ) ∂µ U −1 + ∂µ U Aν U −1 ig ig +U ∂µ Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 −U ∂ν Aµ U −1 − U Aµ ∂ν U −1 1 ∂µ U ∂ν U −1 − ∂ν U ∂µ U −1 + ∂µ U Aν U −1 ig = +U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 +U (∂µ Aν − ∂ν Aµ ) U −1 . we note that under the gauge transformation (12.1 Non-Abelian gauge theories 493 U = e−iθ a (x)T a . Aµ → U Aµ U −1 − 1 (∂µ U ) U −1 .

we see that unlike in the case of QED.41) (12.40) Thus comparing (12. Aν ] = ig (Aµ Aν − Aν Aµ ) → ig U Aµ U −1 − 1 1 (∂µ U ) U −1 U Aν U −1 − (∂ν U )U −1 ig ig − U Aν U −1 − 1 1 (∂ν U )U −1 U Aµ U −1 − (∂µ U )U −1 ig ig 1 ∂µ U Aν U −1 − U Aµ ∂ν U −1 ig = ig U Aµ Aν U −1 − + + = 1 ∂µ U ∂ν U −1 − U Aν Aµ U −1 g2 1 1 ∂ν U Aµ U −1 − U Aν ∂µ U −1 − 2 ∂ν U ∂µ U −1 ig g 1 ∂µ U ∂ν U −1 − ∂ν U ∂µ U −1 ig igU [Aµ . Aν ] . (12. here fµν = ∂µ Aν − ∂ν Aµ is neither invariant nor does it have a simple transformation under the gauge transformation (12.40).42) so that Fµν transforms covariantly. Aν ] U −1 − − ∂µ U Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 .494 12 Yang-Mills theory Namely.39) and (12. (12.38). Fµν → U Fµν U −1 . It is now easy to construct the gauge invariant Lagrangian density for the dynamical part of the gauge field as (quadratic in derivatives) . Let us also note that under the gauge transformation ig [Aµ . it is clear that. in the present case. then under the gauge transformation (12. if we define the field strength tensor as Fµν = ∂µ Aν − ∂ν Aµ + ig [Aµ .38).

This Lagrangian density is invariant under the non-Abelian gauge symmetry transformations (12. given by 1 a LQCD = − Fµν F µν a + iψ k γ µ (Dµ ψ)k − mψ k ψk .1 Non-Abelian gauge theories 495 1 1 a Lgauge = − Tr Fµν F µν = − Fµν F µν a . T a T b = 1 δab . From the cyclicity of trace this Lagrangian density is easily seen to be invariant under the gauge transformations (12. 4 (12. more generally for fermions interacting with a non-Abelian gauge field. a Fµν T a = ∂µ Aa − ∂ν Aa T a + igAb Ac T b .45) The complete Lagrangian density for QCD (quantum chromodynamics.41) is easily seen to take the form.12. see (12. ν µ µ ν or. a a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac = −Fνµ . or.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 . Fµν = ∂µ Aν − ∂ν Aµ + ig Aµ .38). the field strength tensor (12. therefore. QCD corresponds to the particular case when G = SU (3)) is. T c ν µ µ ν = ∂µ Aa − ∂ν Aa T a − gf abc Ab Ac T a .43) where a particular normalization for the trace of the generators is assumed (namely.34)) . Aν . µ ν ν µ (12. In components. (12.46) where we can set m = 0 if the fermions (quarks) are massless. 2 4 (12.59) and the comments fol2 lowing that equation).38) which have the infinitesimal form (see (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 (12.48) and the Jacobi identity associated with this algebra is given by T a.48) it can be seen that the Jacobi identity (12. T a . then the Lie algebra breaks up into two commuting non-Abelian subalgebras. T b + if bcp T p .47) Let us discuss briefly some of the properties of the Lie algebra of the symmetry group G. T b . T c + if cap T p .49) imposes a restriction on the structure constants of the group of the form if abp T p . T a = 0. or. 1 ∂µ ǫa − f abc Ab ǫc . T b + T b . T c + T c.49) Using (12.18). or. The algebra of the Hermitian generators. In this case . T c . (12. f abp f pcq + f cap f pbq + f bcp f paq = 0.37) as well) U (x) = eiθ a (x)T a . δAa = µ δψk = −iǫa (x)(T a )kℓ ψℓ .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. has the form T a . (12. as we have noted in (12.496 12 Yang-Mills theory δψ k = iǫa (x)ψ ℓ (T a )ℓk . µ g (12. T b = if abc T c . T a = 0. f abp f pcq T q + f cap f pbq T q + f bcpf paq T q = 0. (12. (12.

For any representation of a simple Lie group we can write Tr T a T b = C2 δab . T b T c (12.53) Using the cyclicity property of the trace. A direct product of simple Lie groups is called semi-simple. On the other hand. let us next note that the quantity habc = = T a. A non-Abelian Lie group that cannot be so factorized is called a simple Lie group. It depends on the representation but not on the indices a and b.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 . is completely antisymmetric in all its indices. (12.52) We note that C2 is a normalization constant which determines the values of the structure constants. In all our discussions. (12. we note that . Tr T a T b = (12. Furthermore. a = 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.54) Tr Tr T a T b T c − Tr T b T a T c .55) where the index c is fixed (and not summed). using the commutation relation (12.12.1 Non-Abelian gauge theories 497 the group G is a direct product of two independent non-Abelian Lie groups. we will assume that the symmetry group G is simple.

498 12 Yang-Mills theory hacb = Tr T a. · · · . dim R. we note here that it is (12. · · · . since habc is completely antisymmetric.55) and (12. a.) We note that if we write Tr T a T b = T (R) δab . Similarly.58) (12. (12. we have (T a T a )mn = C(R) δmn . (12.56) with the index b fixed. (12. (12. T c T b = if acp Tr T p T b = if acp Kp δpb = iKb f acb = −iKb f abc . dim G.43). we have hacb = −habc . However. Furthermore.57) Using this and comparing (12. This shows that we can write Tr (T a T b ) = C2 δab . 2.60) then T (R) is known as the index of the representation R.56) we conclude that Kb = Kc = K.61) .59) which is used to show that the structure constants for a semi-simple group are completely antisymmetric in the indices. n = 1. m. (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. 2. b = 1.59) where the constant C2 depends only on the representation. (12.

(12.63) bc This is consistent with the hermiticity requirement for the generators. (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.50). T a† = = Ta ∗ ∗ bc cb − if acb = if acb = −if abc = T a bc .1 Non-Abelian gauge theories 499 where C(R) is known as the Casimir of the representation R. T a. T b = = = T aT b − T bT a Ta cp cq cq cp Tb pq − Tb Ta pq − if acp − if bpq − − if bcp − if apq = −f acp f bpq + f bcp f apq = f abp f pcq = −f cap f pbq − f bcp f paq = if abp (−if pcq ) = if abp (T p )cq . However.65) where we have used the anti-symmetry of the structure constants as well as (12.62) We can determine the generators of the group in various representations much like in the case of angular momentum. we conclude that the identification .12. (12. namely. a particular representation that is very important as well as useful is given by Ta = −if abc . (12. As a result.64) Furthermore. we can easily check that this representation satisfies the Lie algebra.

69) .67) so that we can equivalently write (see. (12. T b . g µ (12.26)) δAa = µ 1 (adj) a D θ . Aµ .500 12 Yang-Mills theory a T(adj) bc = −if abc .66) indeed defines a representation of the Lie algebra known as the adjoint representation. Let us next look at the transformation for the gauge fields in (12. transforms according to the adjoint representation of the group.42). µ (12. This can also be seen from the transformation of the field strength tensor in (12.68) which shows that the gauge field.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 . (12. (12. which infinitesimally has the form Fµν → U Fµν U −1 so that = Fµν + iǫb Fµν . for example.

71) can also be written in the matrix form as (adj) Dµ ǫ = ∂µ ǫ + ig[Aµ . µ (12. For example. (12. the gauge field must transform in the adjoint representation. (It does not matter what representation the matter fields belong to. ab µν (12. let us scale the field variables as .72) where Aµ = Aa T a and ǫ = ǫa T a (with T a in any representation). 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 . ǫ].) For completeness.12.70) Comparing this with the transformation of the fermions in (12.73) Let us note the following features of this Lagrangian density. or. 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.1 Non-Abelian gauge theories 501 δFµν = iǫb Fµν . let us note here that the covariant derivative in the adjoint representation (12. a b δFµν = −f abc Fµν ǫc b = −i − if cab ǫc Fµν c = −i T(adj) ab b ǫc Fµν c = −iǫc T(adj) Fb. a a δFµν T a = iǫb Fµν if abc T c . T b a = iǫb Fµν T a .67) (adj) Dµ ǫ a = ∂µ ǫa − gf abc Ab ǫc . or. (12.

Physically we understand this in the following way.502 12 Yang-Mills theory Aa → µ 1 a A .76) .44)) Fµν a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac . In the present case. ν µ µ ν = ∂µ Aν − ∂ν Aµ + ig[Aµ . 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 .75) LYM = − Tr Fµν F µν = − Fµν F µν a . (12. therefore. in the case of non-Abelian symmetry they must couple to themselves and possess self interactions.41) and (12. the coupling constant can be scaled out and written as an overall multiplicative factor in the Lagrangian density. (12. g µ It follows then that 1 a Fµν → ∂µ Aa − ∂ν Aa − f abc Ab Ac . the pure Yang-Mills theory is an interacting theory unlike the Maxwell theory. 12. Aν ].74) Thus we note that in this theory.45)) 1 a 1 (12.2 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. The other feature to note is that unlike the photon field. Since the gauge fields couple to any field (particle) carrying charge of the symmetry group. here the gauge fields have self interaction (they interact with themselves) and. the gauge fields carry the charge of the non-Abelian symmetry group (they have a nontrivial symmetry index) in contrast to the photon field which is chargeless. 2 4 where the field strength tensor is given by (see (12.43) (or (12. We note that the Lagrangian density for the Yang-Mills theory is given by (12.

75) and have the form ∂L ∂L − = 0.79) where the covariant derivative is defined to be in the adjoint representation of the group (see (12.78) The Euler-Lagrange equation for the theory are obtained from (12. we can write the Euler-Lagrange equations as Dµ F µν = ∂µ F µν + ig[Aµ .12. F µν ] = 0. j i i j (12.77) as 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a . or.75) is the transverse projection operator just like in the case of Maxwell’s theory. i 0 0 i a Fij a = ∂i Aa − ∂j Aa − gf abc Ab Ac = −ǫijk Bk . 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 (12.67) and (12.76) as a a F0i = ∂0 Aa − ∂i Aa − gf abc Ab Ac = Ei .68) and we are suppressing the symbol “(adj)” to denote the covariant derivative in the adjoint representation for simplicity) .2 Canonical quantization of Yang-Mills theory 503 As in the case of the Abelian theory (Maxwell theory). µ (12. 2 2 4 LYM = Tr Ei Ei − (12. µ (Dµ F µν )a = ∂µ F µν a − gf abc Ab F µν c = 0.77) The Lagrangian density (12.75) can also be written explicitly in terms of the non-Abelian electric and magnetic field strength tensors (12. In the matrix notation (see (12.68)). we can identify the non-Abelian electric and magnetic fields from (12. (12. Therefore.80) Let us note that the coefficient matrix of highest derivatives in (12. ∂∂µ Aa ∂Aa ν ν ∂µ or.72)). −∂µ F µν a − −gf abc Ab F µν c = 0.

Namely. ˙ a (x) ∂ Aµ (12.75) and this leads to µ Πµ a (x) = ∂L = −F 0µ a (x).77).81) In fact.79) for ν = 0.83) a Noting that the field strength Fµν is antisymmetric in the indices µ. we choose Aa (x) = 0. requiring that . Thus we can choose a gauge condition and set it equal to zero. g (12. The momentum conjugate to Aa does not exist. the constrained structure of the theory is already obvious in the Euler-Lagrange equation of motion (12. Let us obtain the momenta canonically conjugate to the field variables Aa from (12. Di F i0 a = 0.85) a where the non-Abelian electric field Ei (x) is defined in (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. 0 or.84) (12. A0 (x) = 0. as in the case of Maxwell’s theory and a Πi a (x) = −F 0i a = Ei (x). (12. For example. we have Π0 a (x) = −F 00 a (x) = 0.82) which is a constraint relation. (12. namely. (12. Thus. we see that we have only 3N canonical momenta where N = dim G.504 12 Yang-Mills theory δAa = µ 1 (Dµ ǫ(x))a . ν. This implies that 0 Aa is like a c-number quantity which commutes with every other 0 operator in the theory.

˙ Ei = Ai . ig (12. (12. We note that the relation (12.91) can be inverted to give .88) (namely. (12. let us also note that the potential term in (12.86). 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 .87) we can determine the parameter of transformation and we see that a gauge field can always be transformed to satisfy (12. in the present case.12. Ei ] = 0.75) 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 .86) the Lagrangian density (12. i or.91) ig ig so that there is a cyclic variable in the theory that needs to be separated. we note from (12. On the other hand.86) the non-Abelian electric field in (12. (12.2 Canonical quantization of Yang-Mills theory 505 A′ = U A0 U −1 − 0 or.77) takes the form a ˙ Ei = Aa . (12. 1 (∂0 U ) U −1 = 0. Let us note that with the gauge condition (12. U −1 ∂0 U = igA0 .90) so that neither the gauge field Ai nor the electric field Ei is transverse ˙ (recall that Ei = Ai ).82) that Di Ei = ∂i Ei + ig[Ai . we expect the physical dynamical degrees of freedom of a massless gauge field to be transverse.88) 2 2 4 LYM = Tr Ei Ei − where we have used the fact that in the gauge (12.89) From our discussion in the case of the Maxwell theory. However.

506 12 Yang-Mills theory 1 1 ¯ Ai = S −1 Ai S + S −1 ∂i S = S −1 Ai + (∂i S)S −1 S.72)) Di (A)G = ∂i G + ig[Ai . ¯ ∂i Ai = 0.93).93) 1 1 (∂i S) (∂i S −1 ) − ∂i (∂i S)S −1 ig ig 1 ∂i (∂i S)S −1 − [Ai . we obtain 1 ¯ ∂i Ai = ∂i S Ai S −1 − ∂i (∂i S)S −1 ig 1 ¯ ¯ = (∂i S)Ai S −1 + S Ai (∂i S −1 ) − ∂i (∂i S)S −1 ig = (∂i S) S −1 Ai − + Ai S + = − = − 1 (∂i S −1 ) ig (12.94). let us require the new field variable Ai to be transverse.92) ig ig Keeping in mind that we would like the dynamical variables to be ¯ transverse. ig (12. namely. Imposing this condition in (12.91) that . (12.92) in the intermediate step and the covariant derivative in the adjoint representation is defined for a matrix function G as (see (12. (12.91). We note next from the field redefinition in (12.94) where we have used (12. then we can define a new field variable Ai satisfying the transversality condition (12. (∂i S)S −1 ] ig 1 Di (A) (∂i S)S −1 .95) This shows that if we can find a gauge transformation which satisfies ¯ (12. G].

ig (12. decomposing E i into its transverse and longitudinal components as T E i = E i + ∂i φ. 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 .100) . Here Di (A) is the covariant derivative in the adjoint representation (see (12. ig where we can identify 1 ˙ ¯ ¯ E i = Ai − Di (A)f. ig (12.96) we note that in the gauge (12.96) ˙ ¯ where we have used ∂0 S −1 = −S −1 SS −1 .98) (12.12. where (12. (12.86) we can write 1 ˙ ˙ ¯ ¯ Ei = Ai = S Ai − Di (A)f S −1 = SE i S −1 .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 .95)) with respect to ¯ the gauge field Ai and we have identified ˙ f = S −1 S.97) From (12.99) Furthermore.

E i ])S −1 ¯ = S(Di (A)E i )S −1 . T ¯ ∂i ∂i φ + ig[Ai .103) where we have used (12. or. E i ] = ρ. E i + ∂i φ ] = 0. E i ] = 0. φ] = −ig[Ai . T ¯ ¯ ∂i ∂i φ + ig∂i [Ai . We can invert this relation to determine .102) Let us next note that under the field redefinition (12.104) T ¯ where we have used ∂i Ai = 0 in the intermediate step. T ∂i E i = ∂i ∂i φ = ∇2 φ. As a result of this relation. ig ∇2 ˙ ¯T ¯ Ei = Ai .98). Ei ] ¯ = ∂i SE i S −1 + ig[S(Ai − 1 −1 S (∂i S))S −1 . (12. φ=− (12.91) and (12. (12. ¯ ¯ ∂i ∂i φ + ig[Ai . the constraint equation (12.101) we conclude from the definition in (12. ¯ ∂i Di (A)φ = ρ. E i ] = 0. φ] + ig[Ai . or.91).99) that 1 1 ¯ ∂i Di (A)f. SE i S −1 ] ig = (∂i S)E i S −1 + S(∂i E i )S −1 − SE i S −1 (∂i S)S −1 ¯ +igS[Ai − 1 −1 S (∂i S). (12.90) takes the form ¯ ¯ Di (A)E i = ∂i E i + ig[Ai . E i ]S −1 ig ¯ = S(∂i E i + ig[Ai .508 12 Yang-Mills theory ∂i E i = 0. or. Di (A)Ei = ∂i Ei + ig[Ai . or.

∂i Di (A) 1 ¯ ρ. ¯ We have isolated the dynamical variables of the theory to be Aa i which are transverse and the conjugate momenta can be determined from (12.93).106) Using the definition in (12.88) takes the form LYM = Tr Ei Ei − 1 Fij (A)F ij (A) 2 1 ¯ ¯ = Tr E i E i − Fij (A)F ij (A) 2 1 1 ˙ ˙ ¯ ¯ = Tr (Ai + ∂i ¯ ρ)(Ai + ∂i ∂k Dk (A) ρ) ¯ ∂j Dj (A) 1 ¯ ¯ − Fij (A)F ij (A) 2 1 ˙ ˙ ¯ ¯ ¯ ¯ = Tr Ai Ai − Fij (A)F ij (A) 2 1 1 2 −ρ ¯ ∇ ∂k Dk (A) ρ ¯ ∂j Dj (A) 1 ¯a ¯a 1 ˙ ˙ ¯ ¯ = A A − Fij (A)F ij (A) 2 i i 4 1 1 1 a 2 − ρa ¯ ∇ ∂k Dk (A) ρ . ∂i Di (A) = Di (A)∂i which has been used together with integration by parts).107) to be Π ia = ∂LYM ˙ ¯ = Aa . ∂j Dj (A) (12.91).98) and the decomposition in (12.2 Canonical quantization of Yang-Mills theory 509 φ = 1 ¯ ρ. i ˙ ¯a ∂A i (12. ¯ 2 ∂j Dj (A) (12.108) .100) we obtain T ˙ ¯ E i = E i + ∂i φ = Ai + ∂i (12.105) so that from (12.12.106) as well as the cyclicity of trace. (12.107) where we have neglected total divergence terms in the intermediate ¯ ¯ steps (note also that because of (12. the Lagrangian density (12.

i (12.110) then we can make a time independent infinitesimal gauge transformation (under which the theory is invariant) 1 ¯ǫ a ¯ ¯ Di (A)¯ . Here the transverse delta function corresponds to the one already defined in the case of Maxwell’s theory in (9. even in the absence of other matter fields there is a long range interaction. This is related to the question of gauge fixing. We note that the Lagrangian density in (12. Thus the long range behavior of this theory is still not well defined in the Coulomb gauge.111) (12. therefore. 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. of the theory) ¯ is. namely the Coulomb gauge does not uniquely define the gauge fields. i ¯ ∂i Aa = 0. Let us note here that the long range behavior of the interaction (and. controlled by the eigenvalues of the operator ∂i Di (A).107) has exactly the same form as in the Abelian theory except for the last term.112) . A′ a = Aa + i i g such that ¯i ∂i A′ a = 0. This can be seen from the ¯ fact that if we have a field configuration Aa which is transverse. Π (y)} = δab δi j (x − y).31). i. i TR jb (12.510 12 Yang-Mills theory The equal-time canonical Poisson brackets for the theory can now be obtained as in the case of the Maxwell theory ¯ {Aa (x).109) with all other brackets vanishing.e. this last term has the form ρ 1 2 ρ which corresponds to ∇ the long range Coulomb interaction). ¯ Gribov has shown that the operator ∂i Di (A) does possess zero modes or eigenvectors with zero eigenvalue. Here we see that since the gauge fields are self interacting. therefore. (12.

129)) (∂ · A) = 0. occurs in all gauges other than the axial gauge. However. (12. However. (12. makes the theory nonsingular. classically we can impose the condition ∂ · A = 0.113) The additional term clearly breaks gauge invariance and. let us modify the theory (12. (It is of the form O 1 . In other words.12. consequently.111) that this has a singularity at g = 0. For example. in the present case.125) with J µ = 0) 1 1 a L = − Fµν F µν a − (∂µ Aµa )2 . On the other hand. Therefore. The canonical quantization of Yang-Mills theories does not have manifest covariance very much like the Maxwell theory. There now exist prescriptions to ¯ take care of this problem.) g Hence such gauge field configurations can only be reached nonperturbatively. 4 2 (12.115) . by modifying the theory. But let us note from the form of A′ a in i (12. This shows that ¯ the Coulomb gauge does not really specify the gauge field configuration uniquely.114) namely. much like in the Maxwell theory. Thus. we can also try to quantize the non-Abelian gauge theory covariantly. As long as we are within perturbation theory we can neglect such configurations. ∂ · A behaves like a free field and. we recall that in QED (see (9. We can show that this nonuniqueness.75) as (see (9. Gribov ambiguity is a phenomenon associated with large gauge transformations and is an important issue since it leads to nonperturbative effects. namely. very much along the lines of the Abelian theory that we have discussed in chapter 9. for example. there are serious differences from Maxwell’s theory. The other way of saying this is that within the framework of perturbation theory we are interested in infinitesimal gauge transformations. the hypersurface defining the gauge choice intersects the gauge orbits more than once. therefore. known as the Gribov ambiguity. we would not worry about it within the context of perturbation theory.2 Canonical quantization of Yang-Mills theory 511 ¯ provided ǫa corresponds to a zero mode of ∂i Di (A).

the physical subspace would keep changing with time which is not desirable. we need to analyze the question of modifying the theory in a more systematic and detailed manner. just modifying the Lagrangian density as in (12. the additional term in (12. We would see next how we can derive intuition on this important question from the path integral quantization of the theory.118) Thus. we obtain (∂ · Aa ) = gf abc ∂ν Ab F µν µ c (12.116) In the case of the Yang-Mills theory. (12. (12.) Correspondingly the naive analog of the Gupta-Bleuler condition for non-Abelian gauge theories does not seem to exist. even classically the equations of motion are given by ∂µ F µν a − gf abc Ab F µν c + ∂ ν (∂ · Aa ) = 0. nor can we think of the supplementary condition ∂ · Aa(+) |phys = 0. (12. it cannot be uniquely decomposed into a positive and a negative frequency part (in a time invariant manner). 12. µ Contracting with ∂ν . In other words.512 12 Yang-Mills theory which then translates to the Gupta-Bleuler condition on the physical states ∂ · A(+) |phys = 0. (Namely. consequently. since it is not invariant under time evolution. we note that (∂ · Aa ) does not behave like a free field and.113) does not seem to be sufficient in contrast to the Abelian gauge theory. in contrast to the Abelian theory. since (∂ · Aa ) is not a free field. Therefore.117) = 0.113) has truly modified the theory. however.119) in a physically meaningful manner.3 Path integral quantization of gauge theories Path integrals provide an alternative to the canonical quantization of field theories and are particularly useful in studying complicated . Furthermore.

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 .12. 2 (12. δφ(y) ǫ ǫ→0+ (12. J(x) is the external source to which the scalar field is coupled and S[φ] = d4 x 1 ∂µ φ∂ µ φ − m2 φ2 . Z[J] and W [J] are known as generating functionals for the Green’s functions of the theory.121) Furthermore. Since this is a complete topic in itself. (12. see section 6. for a real scalar field interacting only with an external source. The functional dependence of a quantity on a variable is generally denoted by a square bracket.120) where N is a normalization constant. from the “in” vacuum to the “out” vacuum.3 Path integral quantization of gauge theories 513 gauge theories. 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.4 for the definitions of “in” and “out” states) in the presence of interactions. 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] . Like the derivative of a function. rather. . we will not go into the systematic details of this description. 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.122) and various Green’s functions (n-point functions) are obtained from the generating functionals Z[J] and W [J] through functional differentiation. 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. we will only go over some of the essential concepts involved in such a description. For example.

namely.514 12 Yang-Mills theory The integration in (12.120) is known as a functional integration and is defined as follows.126) where the Green’s function G(x − y) is formally the inverse of the operator in the quadratic part of the action.124) δVi →0 and in terms of these discretized field variables. the functional integration is defined (up to a normalization constant) as Dφ = dφi . (12. Basically.123) Clearly in the limit of vanishing volume we recover lim φi = φ(x). δVi φi = (12. (12. (12.120) can be evaluated in a straightforward manner. i (12. this is a generalization of the Gaussian integral to the functional space and leads to Z[J] = eiW [J] = N det ∂µ ∂ µ + m2 −1 2 e− 2 i R d4 xd4 y J(x)G(x−y)J(y) .125) With these basics a simple functional integral for a quadratic action such as in (12. Let us divide the entire space-time into infinitesimal cells labelled by “i” of volume δVi and define 1 δVi d4 x φ(x).126) is independent of field variables and is a constant (possibly divergent) and can be incorporated into the normalization constant N . G = (−∂µ ∂ µ − m2 )−1 . When we have an .) The determinant in (12.127) (We note here that a Gaussian integral for fermions leads to positive powers of determinants in contrast to the negative powers for bosons which is related to the fact that fermion loops have a negative sign associated with them.

128) where J µ represents a conserved current (source). The advantage of using the path integral description lies in the fact that there are no operators in (12.A µ) DAµ ei[ 2 (Aµ . 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. In this case. the operator P µν is a projection . (12.129) µν A ν )+(J µ . (12.131).123) and (9. in general.P −1 J ν µν ). the generating functional in the path integral formalism is given by (J ) [A Z [Jµ ] = eiW [Jµ ] = N = N DAµ eiS µ] (12.130) where N is a normalization constant and (see (9. With this brief introduction to path integral description.126) Z [Jµ ] = eiW [Jµ ] = N (det (−P µν ))− 2 e− 2 (J 1 i µ .132) where we have used the notation in (12. This is a Gaussian functional integral and we can evaluate this using our earlier result in (12.124)).P 1 ] . everything is a classical function. cannot be obtained in a closed form. as we have seen before (see (9.122)) we have identified P µν (x − y) = (η µν (J µ .131) d4 x J µ (x)Aµ (x).12. let us go back to the Maxwell theory 1 L(J) = − Fµν F µν + J µ Aµ . (12. In this case. 4 (12. Aµ ) = − ∂ µ ∂ ν ) δ4 (x − y).120).3 Path integral quantization of gauge theories 515 interacting theory (not quadratic in the field variables) the functional integral. However.

(12. the determinant of P µν vanishes. (As we have seen the form of transformation for the gauge fields (12. The Lagrangian density for Maxwell’s theory is invariant under the gauge transformation 1 (∂µ U )U −1 . ie Aµ → A(θ) = U Aµ U −1 − µ where U (θ) = e−iθ(x) . e Aµ → A(θ) = Aµ + µ (12. we can write the transformation as 1 ∂µ θ(x). consequently. Therefore.) (θ) For a fixed Aµ . is proportional to the “volume” of the orbits denoted by dθ(x).134) We can immediately see that for a U (1) gauge group. (The operator possesses zero modes and.136) . This is a U (1) symmetry and the symmetry is noncompact since the parameter of transformation θ(x) can take any real value. 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. on the other hand. x (12.133) in terms of U (θ) is quite general and holds for the non-Abelian theories as well.516 12 Yang-Mills theory operator for transverse photons. This implies that the generating functional does not exist in this case.135) which is the familiar gauge transformation for Maxwell’s theory. is constant on such orbits. Clearly therefore. the generating functional. the inverse of the matrix cannot be defined either. even in the absence of any sources. The longitudinal vectors kµ (or ∂µ ) are its eigenvectors with zero eigenvalue.133) (12. The action S.) The source of the difficulty is not hard to see.

the Coulomb gauge. (12. A3 (x) = 0. (12. ∇ · A(x) = 0.138) has a unique solution for some θ(x). of course. here are a few of the familiar gauge fixing conditions F [Aµ (x)] = ∂µ Aµ (x) = 0.) This is an infinite factor (which is one of the reasons for the divergence in the naive evaluation of the functional integral) and must be extracted out before doing any calculations. The way this is carried out is by choosing a hypersurface which intersects each orbit only once. the axial gauge. In this way. then even if Aµ does not satisfy the condition. We can . 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. gauge invariant and do not depend on the choice of the hypersurface (gauge). (12.139) is known as the gauge condition (or gauge fixing condition). Rather we should integrate over each orbit only once. the Lorentz/Landau gauge. if F [Aµ (x)] = 0. namely.12. the temporal gauge. Thus.3 Path integral quantization of gauge theories 517 (In the non-Abelian case. we can find a gauge trans(θ) formed Aµ which does and F [A(θ) (x)] = 0. This procedure is known as gauge fixing and the condition F [Aµ (x)] = 0.140) and so on. A0 (x) = 0. µ (12. for example. We recognize that we should not integrate over all gauge field configurations because they are not really distinct.137) defines the hypersurface which intersects the gauge orbits once. Physical quantities are. we pick up a representative gauge field from each gauge orbit. this should be replaced by the group invariant Haar measure x dU (x).

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

this gives the correct functional form for the generating functional.146) Furthermore. µ under which the generating functional takes the form dθ(x) δ F [Aµ (x)] eiS x (J ) [A µ] (J ) [A µ] (12. dθ(x) = (12.145) Remembering that ∆FP [Aµ ] is gauge invariant we can now insert this identity factor into the generating functional to write (J ) [A Z[Jµ ] = N DAµ ∆FP [Aµ ] (θ) dθ(x) δ F [Aµ (x)] eiS x µ] . namely. (12.144) In the non-Abelian case. let us make an inverse gauge transformation Aµ → A(−θ) .12.148) 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. (12. However. To do this let us note from (12. Therefore.147) Z[Jµ ] = N = N DAµ ∆FP [Aµ ] dθ(x) x DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS (J ) [A µ] = N DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS .141) that . d(U U ′ ) = dU. (12.3 Path integral quantization of gauge theories 519 group a translation of the transformation parameter corresponds to a gauge transformation) d θ(x) + θ ′ (x) . we still have to determine what ∆FP [Aµ ] is.

therefore.151) we determine δF [Aµ (x)] ∆FP [Aµ ] = det δθ(y) (θ) (12. Thus. be thought of as the Jacobian that goes with a given gauge choice.150) (12. for gauge fields which satisfy the condition F [Aµ (x)] = 0. (12.149) δθ δF F [Aµ (x)]=0 (θ) . (12.149)-(12. On the other hand.152) The Faddeev-Popov determinant can.153) .151) θ=0 . We see that the Faddeev-Popov determinant can be calculated simply by restricting to infinitesimal gauge transformations (since we take θ = 0 at the end). we have θ(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. 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.) 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). using (12. (Here we can completely ignore the problem of Gribov ambiguity associated with large gauge transformations.

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

c] . Thus the generating functional now takes the form Z[Jµ ] = eiW [Jµ] = N DAµ Dc Dc eiSeff (J ) [Aµ .( 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(y)]+ = 0. (12. δF [Aθ (x)] µ d xd y c(x) δθ(y) 4 4 (12.159) θ=0 c(y).). but opposite statistics. These fields are known as Faddeev-Popov ghosts and as is obvious from their anticommutation relations. c(y)]+ = 0..161) Thus although these fields behave as scalar fields under Lorentz transformations. c(y)]+ = 0.c. to write the Faddeev-Popov determinant in the form of a ghost action (action involving ghost fields). known as ghost fields. [c(x). graphs involving these fictitious particles in closed loops must have an additional (−1) factor just like the fermions.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.162) . (12.e.160) Here we have introduced two independent fictitious fields c(x) and c(x). they obey anticommutation rules. We note here that this is possible (since we have a determinant with a nonnegative power) only if the ghost fields c(x) and c(x) anticommute (ghost fields have the same Lorentz structure as the parameters of transformation. i. [c(x). [c(x). (12.

2ξ 2ξ (12.153)) F [Aµ (x)] = ∂µ Aµ (x) = f (x). 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. Otherwise it fails its purpose and will not correspond to an acceptable gauge fixing condition. Of course.12.157)) has the form LGF = − 1 1 (F [Aµ (x)])2 = − (∂µ Aµ (x))2 . c] = S (J) [Aµ ] + SGF + Sghost = d4 x Leff [Aµ . c. we start with a gauge invariant Lagrangian density and add to it a gauge fixing Lagrangian density determined by the gauge fixing condition that we want to work with. 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. of course. Let us now look at a simple gauge fixing condition in Maxwell’s theory. the gauge fixing Lagrangian density (see (12. However.163) Thus we can summarize what we have done so far. for example. To do covariant perturbation theory in the path integral formalism.142) that both these actions arose from inserting a factor of unity into the functional integral).3 Path integral quantization of gauge theories 523 where Seff [Aµ . (12. It is worth noting at this point that we have modified our starting Lagrangian density by a series of formal manipulations. this modifies the starting theory. the covariant condition (see (12. c. It is. namely. c] . for the moment let us emphasize that the purpose of a gauge fixing Lagrangian density is to break gauge invariance. our responsibility to show that the physical interpretation of the theory has not changed. (J) (J) (12. the gauge fixing Lagrangian density should make the quadratic part of the effective Lagrangian density nonsingular.165) .164) In this case. We would show this in the next chapter.

168) d4 x ∂µ c(x)∂ µ c(x) = where we have neglected total divergence terms. we note that F [A(θ) (x)] = ∂µ A(θ)µ (x) = ∂µ Aµ (x) + µ so that (see (12. the ghost fields are interacting and have to be present. Thus our effective Lagrangian density for the Maxwell theory with the choice of a covariant gauge fixing condition becomes 1 1 (J) Leff = − Fµν F µν − (∂µ Aµ )2 + ∂µ c∂ µ c + J µ Aµ . however. To obtain the corresponding ghost Lagrangian density for this gauge choice. (12.152)) δF [Aµ (x)] δθ(y) (θ) 1 µ ∂ θ(x) . therefore.167) Absorbing the factor 1 into the normalization factor (alternatively e scaling the ghost fields). the ghost action (12. we may neglect them and then for ξ = 1 we recognize that our effective Lagrangian density (12.524 12 Yang-Mills theory It is clear that this provides longitudinal components to the quadratic terms in fields and hence breaks gauge invariance. e (12.125).160) for this gauge choice is obtained to be δF [Aθ (x)] µ δθ(y) xδ 4 Sghost = − = − = d4 xd4 y c(x) d4 xd4 y c(x) c(y) θ=0 (x − y) c(y) d4 x Lghost .169) is nothing other than Maxwell’s theory in the Feynman-Fermi gauge (9. In non-Abelian gauge theories. (12. 4 2ξ (12.169) We note here that the ghost fields are noninteracting in the case of the Maxwell theory in flat space-time and.166) = θ=0 1 e xδ 4 (x − y). This .

171) . One way of looking at the ghost fields is as if they are there to subtract out the unphysical field degrees of freedom. the ghost fields automatically couple to the geometry also.) Let us now go back to the effective Lagrangian density (12.113) of the non-Abelian gauge theory failed to be sufficient unlike in the Maxwell theory (namely.169). It is also true that when Maxwell’s theory is coupled to a gravitational field.164). namely. since the ghost fields and the ghost action are really necessary for the unitarity of the S-matrix. 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. To do this let us rewrite the effective Lagrangian density (12. the Aµ field has four field degrees of freedom.12. On the other hand. Hence omitting the ghost Lagrangian density in such a case would lead to incorrect results. (12. they anticommute even though they are scalar fields. Furthermore. the ghost Lagrangian density was missing).169) as ξ 1 (J) Leff = − Fµν F µν + F 2 −F (∂µ Aµ )+∂µ c∂ µ c+J µ Aµ . Hence we can think of the effective Lagrangian density as having two (4−2×1 = 2) effective field degrees of freedom which is the correct number of physical dynamical components as we have already seen within the context of canonical quantization of Maxwell’s theory. On the other hand each of the ghost fields. This naive counting works pretty well as we will see later. we see that the equation of motion for F is given by ξF (x) = ∂µ Aµ (x).169) and ask how we can recover the Lorentz gauge from the Lorentz like covariant gauge condition in (12. For example. has only one field degree of freedom (we will discuss the question of hermiticity of the ghost fields in the next chapter).170) 4 2 Here we have introduced an auxiliary field F which does not have any dynamics. we cannot neglect them even if they are noninteracting particularly when we are doing calculations at finite temperature. being a scalar. (The ghost degrees of freedom subtract because they have the unphysical statistics.3 Path integral quantization of gauge theories 525 explains why the naive modification in (12. (12.

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

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

177) which is invariant under the infinitesimal gauge transformation 1 Dµ ǫa (x). (12. The standard covariant gauge fixing (12.179) ν µ µ ν with g denoting the coupling constant of the theory.67) and (12. As we have seen. 4 (12. g (ǫ)a Aa (x) → Aµ (x) = Aa (x) + µ µ (12. In order to circumvent this difficulty. in the non-Abelian gauge theory.528 12 Yang-Mills theory dynamics of the gauge field theory. the Lagrangian density for the gauge fields is given by (see (12. Let us now apply these ideas to the study of the non-Abelian gauge theory.75)) 1 a Linv = − Fµν F µνa . adding a gauge fixing Lagrangian density changes 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.178) where ǫa (x) is the infinitesimal parameter of transformation. and to compensate for that we have to add a corresponding Lagrangian density for the ghost fields following the prescription of Faddeev and Popov. µ a Fµν (x) = ∂µ Aa (x) − ∂ν Aa (x) − gf abc Ab (x)Ac (x). 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.164). Here the covariant derivative in the adjoint representation as well as the non-Abelian field strength tensors are defined as (see (12. then we cannot define propagators and the entire philosophy of doing perturbative calculations with Feynman diagrams breaks down. On the other hand. in general. consists of adding to the invariant Lagrangian density a gauge fixing Lagrangian density of the form .76)) Dµ ǫa (x) = ∂µ ǫa (x) − gf abc Ab (x)ǫc (x).

181) that we are using.122)) δF a [Aµ (x)] δǫb (y) (ǫ) (12.3 Path integral quantization of gauge theories 529 LGF = − 1 (∂µ Aµa (x))2 .12. δǫb (y) (ǫ) (12. Following the prescription of Faddeev and Popov. we can write the ghost action corresponding to this gauge choice as (see (12.182) We note that for the covariant gauge choice (12.184) µ 1 Consequently. 2ξ (12. µ µ g so that we have (see (12. we can write 1 (ǫ)a F a [A(ǫ) (x)] = ∂ µ Aµ (x) = ∂ µ Aa (x) + Dµ ǫa (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 . (12.160) and note that since ǫa is an infinitesimal parameter. 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).183) = 1 µ ab 4 ∂ D δ (x − y) g x xµ µ = ∂x ∂xµ δab − gf acb Ac (x) δ4 (x − y). (12.180) which corresponds to a gauge fixing condition of the form F a [Aµ (x)] = ∂µ Aµa (x) = f a (x).181) Here ξ represents a real arbitrary constant parameter known as the gauge fixing parameter.

in the path integral formulation. 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. the gauge fixing and the ghost Lagrangian densities. As an example of how we should be careful in carrying out the Faddeev-Popov analysis in complicated gauge theories. (12. merely correspond to a multiplicative factor of unity (see (12.185) where we have dropped total derivative terms (surface terms). As we have mentioned earlier.186) = − Fµν F µνa − 4 2ξ We note that the ghost fields in the present case are interacting unlike in Maxwell’s theory and. cannot be neglected even in flat space-time.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.142)) which does not change the physical content of the theory. in some cases we have to work through the details of this analysis rather carefully in order to obtain the correct result. 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 . In a deeper sense. (12. let us consider the gauge theory of the antisymmetric tensor field Aµν (x) . 12.530 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 . However. therefore. With all these modifications.

µ. 3. ˙ ∂ Aµν (x) Πµν (x) = (12.188) where the field strength tensor corresponds to the totally antisymmetric tensor Fµνλ = ∂[µ A νλ] = ∂µ Aνλ + ∂ν Aλµ + ∂λ Aµν . Furthermore.189) Naive counting shows that Aµν has six field degrees of freedom.12.190) so that not all field variables are independent. the canonical momenta are antisymmetric (like the field variables).4 Path integral quantization of tensor fields 531 Aµν (x) = −Aνµ (x). we conclude that . (12. (12. Let us consider only the free Lagrangian density for this field defined by 1 L = − Fµνλ F µνλ . (12. 2. This can be seen from the definition of the conjugate momenta ∂L = −F 0µν (x). However. Aij and we note that Πµν = −Πνµ = −F 0µν . ν = 0. 1.192) namely.191) The independent field variables of the theory are A0i . (12. since Fµνλ is completely antisymmetric in all the indices.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. 6 (12. we also note that this Lagrangian density is invariant under the gauge transformation δAµν (x) = ∂µ θν (x) − ∂ν θµ (x).

j.532 12 Yang-Mills theory or.194) (12. (12. 3.188) takes the form 1 1 L = − Fµνλ F µνλ = − 3F0ij F 0ij + Fijk F ijk 6 6 1 = − 3Πij Πij + Fijk F ijk . 2. we note that the theory still possesses a residual (static) gauge invariance under the transformation δAij = ∂i θj (x) − ∂j θi (x). However. on closer inspection we notice that the Coulomb gauge condition (12. i = 1.193) Since these momenta identically vanish. k = 1.195) (12. 2. (12. and the Lagrangian density (12. i. With these conditions we have ˙ Πij = −F0ij = −Aij .198) This would seem like three constraints and hence we would naively conclude that this theory has no dynamical degrees of freedom.197) so that we can impose a Coulomb gauge condition of the form ∂i Aij (x) = 0. Π0µ = −F 00µ = 0. However.196) Thus it would seem that the theory has truly three degrees of freedom. 6 (12. the corresponding field variables are like c-number quantities and we can choose the gauge condition A0i (x) = 0. 3. Π0i = 0. We can see this by writing out the gauge condition explicitly .198) actually represents only two independent conditions. (12.

Hence the theory has truly one physical degree of freedom and the antisymmetric tensor field describes a scalar field (it is a gauge theory describing a scalar degree of freedom). (12. ∂1 A13 + ∂2 A23 = 0. ∂ρ Fµνλ − ∂λ Fρµν + ∂ν Fλρµ − ∂µ Fνλρ = 0. . the Lagrangiandensity (12.188) takes the form 1 1 1 L = − Fµνλ F µνλ = − × ǫµνλρ ǫµνλσ ∂ρ φ∂ σ φ 6 6 2 1 1 ρ = − × (−6δσ )∂ρ φ∂ σ φ = ∂ρ φ∂ ρ φ.199) ∂2 A21 + ∂3 A31 = 0. massless real scalar field describing a single degree of freedom.4 Path integral quantization of tensor fields 533 ∂1 A1j + ∂2 A2j + ∂3 A3j = 0.189)).202) where φ(x) represents a real scalar field. The fact that the theory has only one degree of freedom can also be seen in the following manner. 12 2 (12.12.201) we can represent them also as (this is the dual of the field strength tensor in four dimensions) 1 Fµνλ (x) = √ ǫµνλρ ∂ρ φ(x). 2. 3 are (12. (12. In this case. Since Fµνλ represent a totally antisymmetric third rank tensor in four space-time dimensions satisfying the Bianchi identity (see (12. ∂1 A12 + ∂3 A32 = 0.203) We recognize this as the Lagrangian density for a free. the three conditions corresponding to j = 1. 2 (12. Thus.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.

2ξ LGF = − (12. so that we have (12. 6 L=− (12. According to our earlier discussions. (12.208) This is a matrix with two vector indices. the starting gauge invariant Lagrangian density has the form 1 Fµνλ F µνλ .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).205) corresponding to the choice of gauge condition F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). (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)).204) and we add to it the covariant gauge fixing Lagrangian density (ξ is the gauge fixing parameter) 1 (∂µ Aµν )2 . Hence to write the determinant as an action we need ghost fields which carry a vector index and we have .

This does not seem right (does not agree with the analysis from canonical quantization) and we will show now that this is a consequence of our careless application of the Faddeev-Popov procedure.212) .12. 2 (12. The field variable Aµν has six degrees of freedom.209) Therefore. 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 ) . (12. Thus the effective Lagrangian density appears to have 6 − 2 × 4 = 6 − 8 = −2. (12. (12. they subtract out field degrees of freedom.211) effective field degrees of freedom. The ghost fields being vectors have four degrees of freedom each and since they satisfy anticommutation relations.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 .210) Let us now try our naive counting of degrees of freedom of the theory. Let us start with the generating functional (J ) [A Z [Jµν ] = N DAµν eiS µν ] .

209) 1 2 ).216) Sghost = d4 x (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) . (12.217) In the naive application of the quantization procedure. 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 (12.536 12 Yang-Mills theory The gauge symmetry allows us to choose a gauge condition and we have chosen a Lorentz like gauge condition. we would use the ’t Hooft trick to write i − 2ξ Df µ δ (F ν [Aµλ (x)] − f ν (x)) e = e i − 2ξ R d4 x fν (x)f ν (x) R d4 x F ν [Aµλ (x)]Fν [Aµλ (x)] . (12.218) .215) Thus inserting this identity element into the functional integral. (θ) ∆FP [Aµν ] = det = det ( (12. as we have seen (see (12. (12.214) where. (12.213) Therefore. (12. F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). δF ν [Aµλ (x)] δθ σ (y) ν x δσ θ=0 ν − ∂x ∂xσ ) δ4 (x − y) . we introduce the identity element as (see (12.152)).154)) (θ) ∆FP [Aµν ] x dθ σ (x)δ F ν [Aµλ (x)] − f ν (x) = 1.

(12. (9. (12. for example.12. (12. implies that ∂ν ∂µ Aµν = ∂ν f ν (x) = 0.220) (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) where P (x − y) is the normalized transverse projection operator defined earlier (see. 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.224) so that we can write . the action for fµ in (12.222) We see that because of the transverse projection operator.223) Thus one has to use a gauge fixing condition and we choose again a covariant gauge condition 1 1 2 F [fµ (x)] = ∂µ f µ (x) = f (x).4 Path integral quantization of tensor fields 537 However.219) That is.221) has a gauge invariance (just like Maxwell’s theory) of the form δfµ (x) = ∂µ θ(x).123)) as µν µ ν ∂x ∂x x P (x − y) = η µν − δ4 (x − y). we note that the gauge condition ∂µ Aµν (x) = f ν (x). (12. (12.

(12. (12.P ×∆FP [fµ ] e− 2ξ (F (fµ ).f i µ) ∆FP [fµ ].Fν [Aµλ ])− 2 (c. c) .226)) ∆FP [fµ ] = = we determine 1 1 S ghost = − (c.F (fµ )) i µν fν ) = Df µδ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .538 12 Yang-Mills theory ∆FP [fµ ] x (θ) dθ(x)δ F [fµ (x)] − f (x) = 1.227) 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 .225) where δF [fµ (x)] δθ(y) (θ) 1 2 ∆FP [fµ ] = det θ=0 = det xδ 4 (x−y) . (12. c) = − 2 2 d4 x c(x) c(x).f µ )− 2 (c. (12.P × = i µν fν ) Df ∆FP [fµ ] δ F [fµ (x)] − f (x) e i i − 2ξ (f. c) i µλ ].f ) Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .227) Furthermore. remembering that (see (12. (12.229) . Substituting this into (12.228) det xδ 4 (x − y) 1 2 Dc(x) eiS ghost .226) Thus using the ’t Hooft trick we can write Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .

This is different from the usual Faddeev-Popov ghosts in the sense that the Faddeev-Popov ghosts seem to come in pairs.232) Counting the field degrees of freedom we see that the effective number of degrees of freedom seems to be 6 − 2 × 4 − 1 = −3.230) SGF = − = − Therefore. (12.F µ[Aλσ ]) J +S GF +Sghost +S ghost ). (12. In flat space-time this ghost Lagrangian density can be seen to give a total divergence and. . (12. (12.12. the generating functional takes the form Z [Jµν ] =N =N where 1 2ξ 1 2ξ d4 x Fµ [Aλσ (x)]F µ [Aλσ (x)] d4 x (∂µ Aµν (x)) ∂ λ Aλν . but in the presence of gravitation it cannot be written as a total divergence and is quite relevant. we can write the effective Lagrangian density as (J) Leff 1 1 (∂µ Aµν ) ∂ λ Aλν = − Fµνλ F µνλ + J µν Aµν − 6 2ξ + 1 1 (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) − c 2 2 c.233) This is again not right and the agreement with canonical quantization seems to be worse than before. Thus using this modified ’t Hooft weighting factor. The ghosts of the present form are known as Nielsen ghosts.231) DAµν Dcµ Dcν Dc ei(S DAµν Dcµ Dcν Dc ei(S J +S ghost +S ghost i ) e− 2ξ (Fµ [Aλσ ]. may be neglected. therefore.4 Path integral quantization of tensor fields 539 Note here that the field c(x) is a real anticommuting scalar field.

235) and ∆FP [cµ ] x dλ(x) δ F [c(λ) (x)] −f (x) = 1.540 12 Yang-Mills theory We notice at this point that although we have fixed up the gauge invariance of the gauge fixing Lagrangian density. µ (12. λ=0 . (12. ∆−1 [cµ ] = FP x (λ) dλ(x)δ F [cµ (x)] − f (x) (λ) dF δ F [cµ (x)] − f (x) det (λ) = x δF [cµ (x)] δλ(y) (12.238) = ∂µ cµ (x) = f (x). (12. We can again use the FaddeevPopov trick and write ∆FP [cµ ] x (λ) dλ(x) δ F [cµ (x)] − f (x) = 1. and δcµ (x) = ∂µ λ(x).234) where λ(x) and λ(x) are anticommuting scalar parameters. the ghost Lagrangian density Lghost is invariant.239) (λ) = det δF [cµ (x)] δλ(y) . namely. (12. under δcµ (x) = ∂µ λ(x). (12. the ghost Lagrangian density also possesses a gauge invariance.237) We are now ready to calculate the Faddeev-Popov determinants associated with these gauge fixing conditions.236) Let us choose for simplicity Lorentz like gauge conditions F and F = ∂µ cµ (x) =f (x).

f and f are Grassmann (fermionic) functions. Similarly we can show that 4 −1 ∆FP [cµ ] = = det xδ (x − y) Dc Dc e ” “ e c c e −i e.240) where we have used the fact that for anticommuting variables.241) We note here that since we are writing an inverse determinant as an action. the ’t Hooft trick needs to be carried out rather carefully.243) since the exponents vanish (because of the fermionic nature of the variables). (12. the ghost fields c and c behave like commuting scalars.242) The gauge conditions (12. (12. the Faddeev-Popov term is an inverse determinant and. Namely.237) and (12.238) correspond to fermionic conditions. . e i − 2χ R e e e e d4 x f (x) f (x) . we cannot use weight factors of the forms i − 2χ e R e e d4 x f (x)f (x) . We note that contrary to the usual case. Therefore. in the present case we can write 4 −1 ∆FP [cµ ] = = det xδ (x − y) c e) c Dc Dc e−i(e. Rather. consequently.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) . e . (12. namely. the Jacobian for a change of variables is the inverse of the determinant.12. (12. the appropriate weight factor in this case would correspond to .

2χ c (12.     c : 1 degrees of freedom. c : 1 degrees of freedom.  cµ : −4 degrees of freedom.  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 .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 (12.248) The ghosts c.542 12 Yang-Mills theory e i − 2χ R e e e d4 x f (x) f (x) . (12. (12. The other .  commuting.  cµ : −4 degrees of freedom. c and c being commuting scalars contribute positively to the counting of the number of degrees of freedom. anticommuting. Aµν : 6 degrees of freedom.245) d4 x Leff .  c : −1 degrees of freedom. (12. c.247) We are now ready to count the number of effective field degrees of freedom in the theory.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. (12.

McGrawHill. V. This matches exactly with the counting of the degrees of freedom from the canonical quantization. W. Feynman. Physical Review 96. 9. Physical Review 101. 3. C. 12. Faddeev and V. Itzykson and J-B. Mohapatra. 7. N. 11.5 References 1. G. Huang. N. Hibbs and R. Physics Letters 25B. Singapore (1982). McGraw-Hill. Popov. L. Nuclear Physics B44. ’t Hooft and M. 4. 191 (1954). G. World Scientific. 6. New York (1965). R.249) degree of freedom. ’t Hooft and M. Siegel. A. Utiyama.12. 170 (1980). Nuclear Physics B139. N. 2. 5. Faddeev. Diagrammar. R. Zuber. Physics Letters 93B. L. 1 (1978). 12.5 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 . Physical Review D4. Yang and R. 378 (1971). 1 (1970). Thus counting the degrees of freedom. Mills. Leptons and Gauge Fields. 8. 29 (1967). 189 (1972). New York. Gribov. D. Theoretical and Mathematical Physics 1. 1980. L. P. (12. R. Veltman. CERN preprint (1973). 10. Veltman. Qualks. N. . 1597 (1956). we see that effectively the theory has 6 − 2 × 4 − 1 + 4 × 1 = 6 − 8 − 1 + 4 = 1. K. Quantum Field Theory. D. Quantum Mechanics and Path Integrals. C.

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

(13. the total Lagrangian density. remembers the gauge invariance of the original theory. the total Lagrangian density for the non-Abelian gauge theory (Yang-Mills theory) has the form LTOT = Linv + LGF + Lghost .186).3) where g denotes the coupling constant of the theory. It is easy to check that the total Lagrangian density (13.34) of the original theory. in some sense. However.181) is given by 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ (13. µ (13. with gauge fixing and ghost terms. The total Lagrangian density has been gauge fixed and. develops a global fermionic symmetry which. which for the covariant gauge in (12.1) 2 + ∂ µ ca (Dµ c)a .67)) (Dµ c)a = ∂µ ca − gf abc Ab cc .1 BRST symmetry As we have seen in (12. therefore. does not have the gauge invariance (12.2) Here ξ represents the arbitrary gauge fixing parameter and the covariant derivative in the adjoint representation is defined as (see (12.2) is invariant under the global transformations 545 .Chapter 13 BRST invariance and its consequences 13.

g (13. only when the ghost equation of motion is used). while for ca it is µ true only on-shell (namely. g ω abc b c = f cc. We note that the nilpotency of the transformations holds off-shell only for the fields Aa .4) where ω is an arbitrary anti-commuting constant parameter of the global transformations. Similarly. ca . .50)).2 respectively.546 13 BRST invariance and its consequences δAa = µ δca δca ω (Dµ c)a . we obtain δ ∂ µ Aa = µ ω µ ∂ Dµ ca = 0. = − µ gξ (13. = 6 (13.2) 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 = 0.2 correspond to transformations with the parameters ω1. 2 ω ∂ µ Aa . The invariance of the Lagrangian density (13.4) we have δ2 δ1 φa = 0.5) Here we have used the Jacobi identity for the symmetry algebra (see (12. ca independent of the parameters of transµ formations where δ1.6) 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 . (13. This shows that under two successive transformations of the kind (13. ca .

(13. Generally. is slightly unpleasant in the sense that the nilpotency of the anti-ghost field transformation holds only on-shell. namely. therefore.9) so that the action is invariant.2) can also be written equivalently as . This shows that the action for the total 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. 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).34)) with the parameter ǫa (x) = ωca (x) and. 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 . As we have seen earlier within the context of the Abelian gauge theory in (12. δLinv = 0.170) we can write the gauge fixing Lagrangian density by introducing an auxiliary field. the invariant Lagrangian density is trivially invariant under these transformations.4) with an anti-commuting constant parameter.8) Consequently. In this derivation. we have used the fact that δ (Dµ c)a = 0 which we have seen in (13. First.5). ν gξ (13. let us recall that the gauge fixing Lagrangian density in (13.1 BRST symmetry 547 The invariance of the Lagrangian density (13.2) is invariant under the global transformations (13.4) can now be easily checked. The present formulation of the BRST symmetry.2) under the transformations (13. we note that the transformation for Aa can really be thought of as an infinitesiµ mal gauge transformation (see (12.13. Since this will also be quite useful for our latter discussions. however.

(13. µ (13.13) δF a = 0. we recover the original gauge fixing Lagrangian density (up to a total divergence term). the BRST transformations in (13. g ω abc b c = f cc. namely.548 13 BRST invariance and its consequences LGF = ξ a a F F + (∂ µ F a )Aa . g (13.170) by a total divergence.10) differs from that in (12.10) that the equation of motion for the auxiliary field takes the form ξF a = ∂ µ Aa .14) . and it is straightforward to check that these transformations are nilpotent off-shell. (13. 2 ω = − F a. The total Lagrangian density can now be written as LTOT = Linv + LGF + Lghost ξ 1 a = − Fµν F µνa + F a F a + ∂ µ F a Aa + ∂ µ ca (Dµ c)a .) It is clear from the form of LGF in (13.10) where F a is an auxiliary field.11) and when we eliminate F a from the Lagrangian density using this equation. (The form of the gauge fixing Lagrangian density in (13.12) µ 4 2 In this case. µ 2 (13.4) take the form δAa = µ δca δca ω (Dµ c)a . 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. but it is this form that is very useful as we will see. δ2 δ1 φa = 0.

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

(The fact that these transformations are like scale transformations and not like phase transformations. In the covariant gauge in Maxwell’s theory. the total Lagrangian density (13.12) is also invariant under the infinitesimal bosonic global symmetry transformations δca = ǫca . Here ǫ represents a constant. commuting infinitesimal parameter and the generator of this symmetry transformation merely corresponds to the ghost number operator. it allows us to carry out covariant quantization of the non-Abelian gauge theory.550 13 BRST invariance and its consequences reflection of the asymmetric manner in which these fields occur in the ghost Lagrangian density in (13. for example. it corresponds to the operator that counts the ghost number of the fields. We have already seen that the naive Gupta-Bleuler quantization does not work in the nonAbelian case. We recall that the physical space must be selected in such a way that it remains invariant under the time evolution of the system. (13. the physical Hilbert space needs to be properly selected for a discussion of physical questions associated with the non-Abelian gauge theory. we note here that these fermionic symmetries arise naturally in a superspace formulation of gauge theories. 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.181). Therefore. as we have emphasized several times by now. which is normally associated with the number operator.2) or (13. 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.12). we . Let us recall that the vector space of the full theory in the covariant gauge (12. Without going into details. contains many more states than the physical states alone. In addition to these two anti-commuting symmetries.17) with all other fields remaining inert.) 13. For example. This is known as the ghost scaling symmetry of the theory. namely.

162)) ∂ µ A(+) (x)|phys = 0.) o Thus. µ (13.19) 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. as we have seen in (12. and the ghost scaling symmetry. the conditions in (13. QBRST .19) are sufficient to reproduce even the GuptaBleuler condition in the case of the Abelian theory (namely. The corresponding operator in a non-Abelian theory.19) Note that even in the case of Maxwell’s theory. in reality it is an infinite number of conditions since it has to hold for every value of the coordinates.13. does not satisfy a free equation and hence it is not a suitable operator for identifying the physical subspace in a time invariant manner.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). On the other hand.118). We recognize that even though this looks like one condition. = 0. “(+)”. the generators of the BRST symmetry.130) in the covariant gauge (12. It is. .2 Covariant quantization of Yang-Mills theory 551 have seen that the states in the physical space are selected as the ones satisfying the Gupta-Bleuler condition (see (9. the Gupta-Bleuler condition works because ∂ µ Aµ satisfies the free Klein-Gordon equation (9. not clear a priori if the conditions (13. we can identify the physical space of states of the gauge theory as satisfying (we note that at this point this only defines a subspace of the total vector space and we still have to show that this subspace indeed coincides with the physical Hilbert space) QBRST |phys Qc |phys = 0. In the Abelian theory. (QBRST and Qc are the charges constructed from the N¨ther o current for the respective transformations whose explicit forms can be obtained from the N¨ther procedure and will be derived below.18). are conserved and hence can be used to define a physical Hilbert space which would remain invariant under the time evolution of the system. Qc . (13.18) where the superscript. therefore. stands for the positive frequency part of the field.

2 (13. F a are inert µ under the scaling of ghost fields. (13.552 13 BRST invariance and its consequences To see that these conditions indeed lead to the Gupta-Bleuler condition in Maxwell’s theory. 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 . 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 .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 .20) where we have used the fact that the auxiliary field does not transform under the BRST transformations and the fields Aa . 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 ) . From this. g abc µ a b c f (∂ c ) c c . 2 (13. .21) The corresponding conserved charges can also be obtained from these current densities and take the forms QBRST = = = Qc = 0 d3 x JBRST d3 x F 0i a (Di c)a + F a (D 0 c)a + g abc ˙ a b c f c cc 2 g abc ˙ a b c f c cc .

n .26) implies that the physical states must have (net) zero ghost number. (13. ← → d3 x c ∂ 0 c. (13.19) Qc |phys = 0.23) following from (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 . (13.13. the BRST and the ghost scaling charge operators can be obtained from (13. Thus. we note that for the Abelian theory where f abc = 0 and there is no internal index “a”. In principle.24) QBRST = Qc = ˙ d3 x − F c + F c = ˙ ˙ ˙ d3 x cc − cc = − If we use the field decomposition for the fields and normal order the charges (so that the annihilation/positive frequency operators are to the right of the creation/negative frequency operators). (13. 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) .27) . this allows for states containing an equal number of ghost and anti-ghost particles.22) to correspond to ← → d3 x F ∂ 0 c. denoting the physical states of the theory as |phys = |Aµ ⊗ |n. In particular.12).25) Let us note that the condition (13. (13.

and must further satisfy (with k0 = |k|) F (+) (k)|phys = 0 = kµ Aµ(+) (k) |phys .30) (13. Thus. To investigate systematically whether the physical state conditions in (13. |phys = |Aµ ⊗ |0. we note from the form of the Lagrangian density (13.11)). 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.12) that we can obtain the canonical momenta conjugate to various field variables of the theory to correspond to . (13.31) where we have used the equation of motion for the auxiliary field (see (13. (13. for every momentum mode k). and F (+) (k)|Aµ = 0. n = 0. n denote the (equal) number of ghost and anti-ghost particles. then this implies that (13.19) are the right ones even for the non-Abelian theory. n = 0. This is precisely the Gupta-Bleuler condition in momentum space and this derivation shows how a single condition can give rise to an infinite number of conditions (in this case. 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). 0 = |Aµ . we feel confident that the physical state conditions in (13.554 13 BRST invariance and its consequences where n.28) c(+) (k)|n.19) really select out the subspace of physical states.29) Namely.

Aa plays the role of the momentum conjugate to F a . Πb (y. In particular. = iδab δ3 (x − y). t). t) ca (x. t) c + j = iδab δi δ3 (x − y). t). (13. the (anti) commutation relations (13. we see that in this formulation with the auxiliary field. Πjb (y.34) With this choice. The equal-time 0 canonical (anti) commutation relations for the theory can now be written as ( = 1) Aa (x.33) + The hermiticity conditions for the ghost fields which arise out of various consistency conditions (for example. Πb (y.35) .13. = iδab δ3 (x − y). ∂ ca ˙ ∂LTOT = (D0 c)a . (ca )† = −ca . t).2 Covariant quantization of Yang-Mills theory 555 Πia = Πa = Πa = c Πa = c ∂LTOT = −F 0ia . (13.32) Here we have used left derivatives for the anti-commuting ghost fields. t) c ca (x. the BRST parameter ω is seen to be anti-Hermitian ω † = −ω. (13. = iδab δ3 (x − y). ˙ ∂ ca (13.33) have to satisfy the proper hermiticity properties etc. 0 ˙ ∂F a ∂LTOT ˙ = −ca . the Lagrangian density and the conserved charges have to be Hermitian. ˙ ∂ Aa i ∂LTOT = Aa . t). Πb (y.) are given by (ca )† = ca . t) i F a (x.

the conserved charges (13.21) and.33) [QBRST . for example.38) simply reiterates in an operator language the fact that the BRST transformations are nilpotent (see (13. c QBRST = − Qc = − (13. we see that we can think of iQc as the ghost number operator (with the ghost number for ca being negative).36) and we can calculate the algebra of charges as well as various other relations of interest using the (anti) commutation relations (13.38) This algebra can also be derived directly from the transformation laws for the field variables in (13.34) for the ghost fields. The second equation implies that the ghost scaling symmetry is Abelian. An immediate consequence of the algebra in (13.22) are Hermitian with the assigned hermiticity conditions (13. Qc ] = 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 . (13. The third is a statement of the fact that the BRST charge carries a ghost number of unity.13) and (13.14)). Therefore. (13. Qc ] = −ica (x).17)) [ca (x). [Qc . that (this basically describes the behavior of the ghost fields under a scaling (13.556 13 BRST invariance and its consequences and we note that the currents (13. therefore. We note. [QBRST .22)) g d3 x Πia (Di c)a − F a Πa + f abc Πa cb cc . Qc ] = −iQBRST .36) and (13. c c 2 a d3 x (Πa ca + ca Πc ) .17).37) [ca (x). The algebra of the conserved charges also follows in a straightforward manner from (13. QBRST ]+ = 2Q2 BRST = 0.33) for the field variables. see (13.38) is that . Qc ] = ica (x). The first equation in (13. The conserved charges (13.

if a state |Ψ cannot be written as ˜ |Ψ = QBRST |Ψ . (13. (13. the total vector space V of the complete gauge theory (13. First. As is clear from our earlier discussions. QBRST .40). we can select out a subspace Vphys by requiring that states in this space are annihilated by QBRST . Consequently. [QBRST . (we assume that the vacuum state is a BRST singlet (invariant) state) . The field operator which creates such a state must necessarily commute with QBRST and.2 Covariant quantization of Yang-Mills theory 557 QBRST eπQc = eπ(Qc −i) QBRST = −eπQc QBRST . (13.39) which is quite useful (particularly) in dealing with gauge theories at finite temperature. As we have discussed earlier. therefore.41) We note that there are two possible kinds of states which will satisfy the physical state condition (13. Namely.13. for every |Ψ ∈ Vphys .42) ˜ for some |Ψ and still satisfies the physical state condition. H] = 0. (13. eπQc + = 0. then. must represent a truly gauge invariant field variable. namely. it is truly a BRST singlet (invariant) state.12) contains various unphysical states as well as states with negative norm in addition to the physical states.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. we have QBRST |Ψ = 0. or. 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.

40) can be written in the form ˜ |Ψ = QBRST |Ψ .44) [QBRST . This implies that (13. as we have seen in (13. (Note that the auxiliary field does not transform under a BRST transformation and. also satisfies the above relation.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 . correspond to truly physical states of the theory and since gauge invariant degrees of freedom have physical (non-negative) commutation relations. Ψ] |0 = 0. (13. therefore. The nilpotency of QBRST also implies that all such states would have zero norm because (QBRST is Hermitian with our choice of hermiticity conditions (13.46) We see that the physical state condition is trivially satisfied in this case because of the nilpotency of QBRST (see (13. such states would have positive norm. However.13).34) for the ghost fields. therefore.38)).) The second class of states which would satisfy the physical state condition (13.43) where we have assumed that the operator Ψ creates the state |Ψ from vacuum. (13. namely. it can be written as the BRST variation of the anti-ghost field and. therefore. (13.) ˜ ˜ Ψ|Ψ = Ψ|QBRST QBRST |Ψ = 0. Such states would. has a different character. Ψ|0 = |Ψ .45) and transverse fields would represent such operators which are gauge invariant asymptotically.558 13 BRST invariance and its consequences QBRST |Ψ = QBRST Ψ|0 = [QBRST . (13. Ψ] = 0.

(13. we note that we can write P (0) = ½ − P ′ . which satisfy the physical state condition and are of the first kind can. be decomposed into states containing fixed numbers of unphysical particles.40) because if |Ψ represents a physical state (of either kind) while |Ψ′ is a state of the second kind. by unphysical. then ˜ Ψ′ |Ψ = Ψ′ |QBRST |Ψ = 0.48) which follows from the physical state condition. ca ]+ .51) . The true physical states. Thus. be identified as belonging to the V quotient space V phys = phys and will have positive definite norm. µ We recall from (13. of course. we have ∞ n=0 P (n) = ½. (13.13) through (anti) commutation with the appropriate field variables) F a ∼ [QBRST .2 Covariant quantization of Yang-Mills theory 559 (13.49) that P (0) projects onto the truly physical states of the theory (namely. Here.13) that the auxiliary field can be written as a BRST variation of the anti-ghost field as (note that QBRST is the generator of the BRST transformations and generates the transformations in (13.49). then. (13. (13. with no unphysical particles) and from the completeness of the projection operators in (13. the longitudinal components of Aa and the auxiliary fields. V0 Every state in Vphys can.13. defining P (n) as the projection operator onto the n-unphysical particle sector.49) P (n) P (m) = δnm P (n) . we mean states containing quanta of the ghost fields.50) It is clear from the definition in (13. therefore. this would contain all the zero norm states which would be orthogonal to Vphys itself. If we denote all the states of Vphys of the second kind by V0 .

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

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

P (0) acts as the identity operator. Let us note here from the discussions of the last section that for any |Ψ . |Φ ∈ Vphys and any two operators A and B acting on Vphys (i.e. (13. P (0) |Ψ = P (0) |Ψ = P (0) |Ψ = |Ψ .63) where we note that in the space V phys . any two operators which do not take us out of Vphys ) Ψ|P (0) |Φ Ψ|(½ − P ′ )|Φ = Ψ|Φ .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. This question can be systematically analyzed as follows. (13. we note that given any state |Ψ ∈ Vphys .60) Furthermore. we can define a unique state |Ψ ∈ V phys as |Ψ = P (0) |Ψ .. namely. (13. S] = 0.562 13 BRST invariance and its consequences density (13. = = Ψ|AP (0) B|Φ Ψ|A(½ − P ′ )B|Φ = Ψ|AB|Φ .64) The two states.57).12) is Hermitian so that the S-matrix of the theory is formally unitary. the S-matrix is BRST invariant since the full theory is. 2 (13. |Ψ and |Ψ . namely. Furthermore. so that [QBRST .62) which follow from (13. (13. S † S = SS † = ½. 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 .

67) It is clear that since S is BRST invariant.65) where the last line follows from the first of the conditions in (13.63).69) In other words.66) such that for |Ψ ∈ Vphys . it will leave the space Vphys invariant and hence Sphys will not take a state out of V phys . It follows from (13. In a similar manner.66) and (13. Given this.60). we can also show that † Sphys Sphys = ½. (13.13. the physical state condition (13. Sphys |Ψ = Sphys P (0) |Ψ = P (0) S|Ψ = S|Ψ . (13. it is clear that we can define the S-matrix which acts on the physical space V phys as satisfying P (0) S = Sphys P (0) . 2 (13.70) .40) which naturally follows from the BRST invariance of the theory. (13.62) as well as the formal unitarity of S (13. (13. Furthermore. |Ψ = P (0) |Ψ ∈ V phys as defined in (13.3 Unitarity 563 Ψ|Φ = Ψ| P (0) |Φ = Ψ|P (0) |Φ = Ψ|Φ .62).68) where we have used (13. we now have † Ψ|Sphys Sphys |Φ † Ψ|P (0) Sphys Sphys P (0) |Φ = = = Ψ|S † P (0) S|Φ Ψ|S † S|Φ = Ψ|Φ = Ψ|Φ . also automatically leads to a formal proof of unitarity of the S-matrix in the subspace of the truly physical states of the theory.68) that † Sphys Sphys = ½. (13.

40) that phys| (LGF + Lghost ) |phys′ = −g phys| QBRST . = 0. = g QBRST .71) Here we have used the fact that the BRST charge is the generator of the BRST transformations so that the transformations for any fermionic operator is generated through the anti-commutator of the operator with the generator. in fact. We can also show that all the physical matrix elements of the theory are independent of the choice of the gauge fixing parameter ξ in the following manner (the BRST variation denoted is with the parameter of transformation taken out) . the gauge fixing and the ghost Lagrangian densities) can. we note that these extra terms in the Lagrangian density (namely. be written as a BRST variation (with the parameter of transformation taken out). It now follows from the physical state condition (13. namely. − ca F a − ∂ µ ca Aa µ 2 + (13.72) This shows that the terms added to modify the original Lagrangian density have no contribution to the physical matrix elements of the theory. ξ a a c F + ∂ µ ca Aa µ 2 + |phys′ (13. 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. LGF + Lghost = ξ a a F F + (∂ µ F a )Aa + ∂ µ ca (Dµ c)a µ 2 ξ = gδ − ca F a − ∂ µ ca Aa µ 2 ξ . To show this. we can show that the gauge fixing and the ghost Lagrangian densities lead to no physical consequences.564 13 BRST invariance and its consequences As another consequence of the BRST invariance of the theory.

let us recapitulate briefly some of the essential concepts from path integrals (a detailed discussion is beyond the scope of these lectures).120). However.7. We have already given a simple diagrammatic derivation of such identities in the case of QED in section 9. These are known as the Ward-Takahashi identities (in the Abelian case) or the SlavnovTaylor identities (in the non-Abelian case) of the theory and are quite essential in establishing the renormalizability of gauge theories. For simplicity.4 Slavnov-Taylor identity The BRST invariance of the full theory leads to many relations between various scattering amplitudes of the theory. before we go into the actual derivation.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. (13. as we have seen in (12.13. 13.75) . In this case. However. such a simple diagrammatic derivation does not carry over to non-Abelian gauge theories and they are best described systematically within the context of path integrals which we will do next. the vacuum functional is given by 0|0 J = Z[J] = eiW [J] = N Dφ eiS J [φ] . let us consider the self-interacting field theory of a real scalar field coupled to an external source described by the action S J [φ] = S[φ] + d4 x J(x)φ(x).4 Slavnov-Taylor identity 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 = 0.74) where S[φ] denotes the dynamical action of the self-interacting scalar field. (13. (13.

the Green’s functions calculated from W [J] (the logarithm of Z[J]) through (13. (13.76)) includes contributions of all diagrams (including the ones that are not connected) to a Green’s function.76) J=0 defines the n-point time ordered Green’s function of the theory. δJ(x) Z[J] δJ(x) (13. In particular we note that 0|φ(x)|0 J δn W [J] δJ(x1 ) · · · δJ(xn ) J=0 .77) = φc (x) = (−i) δZ[J] δW [J] = .566 13 BRST invariance and its consequences As we have already mentioned in the last chapter. This is more fundamental since the general Green’s functions can be constructed in terms of these. On the other hand. 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) involve only Feynman diagrams where all the parts of the diagram are connected. In general.49)) and is a functional of J. 0|T φ(x1 ) · · · φ(xn ) |0 = δn Z[J] (−i)n Z[J] δJ(x1 ) · · · δJ(xn ) . For example. (13. On the other hand.77) in the path integral formalism corresponds to a unique Feynman diagram in perturbation theory. To clarify this distinction further.78) is known as the classical field (see (7.77) 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 .79) . the n-point Green’s functions would involve Feynman diagrams which consist of parts that are not connected and Z[J] (through (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. (13. let us note here that each term in the calculation of the Green’s function in (13. Z[J] and W [J] are known as the generating functionals for the Green’s functions of the theory.76) and (13.

82) . the more fundamental concept in perturbation theory is the diagram without the external propagators known as the vertex function. However.80) It can now be checked using (13. The diagrams that contribute to the n-point vertex function and which cannot be separated into two disconnected diagrams by cutting a single internal line (propagator) of the diagram lead to what are known as the 1PI (one particle irreducible) vertex functions or the proper vertex functions of the theory. let us consider the effective Lagrangian density (at the lowest order or tree level) which consists of LTOT for the Yang-Mills theory (13.78) that δΓ[φc ] δφc (x) = d4 y δW [J] δJ(y) δJ(y) − φc (y) − δ4 (x − y)J(y) δJ(y) δφc (x) δφc (x) (13. These are quite fundamental in the study of quantum field theory and the generating functional which generates such vertex functions is constructed as follows. 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). 2 (13.13. (13.12) as well as source terms as follows Leff = LTOT + J µa Aa + J a F a + i (η a ca − ca η a ) µ +K µa (Dµ c)a + K a g abc b c f cc . With these basic ideas from path integrals. 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.4 Slavnov-Taylor identity 567 The connected n-point time ordered Green’s functions consist of connected Feynman diagrams with external lines (propagators).81) = −J(y).

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

It is here that the .75) we note that (13. A → A + δBRST A. δJ a (x) (13.13.88) d4 x J µa (x) This is the master equation from which we can derive all the identities relating the connected Green’s functions of the theory by taking functional derivatives with respect to sources. (13. if we redefine the fields under the path integral as.85) g 2 We note that the generating functional is defined by integrating over all possible field configurations. − iη a (x) + iη a (x) a µa (x) a (x) δK δK δJ (x) (13.78)) 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 . Therefore. since it does not depend on the field variables.85) to (see also (13. (13.87) +iη a (x) The functional integration measure can be easily checked to be invariant under such a fermionic transformation and using (13.87) can also be written as δW δW δW = 0. it should not change under any field redefinition).4 Slavnov-Taylor identity 569 in Leff to be (remember that the parameter of the BRST transformation is anti-commuting) = J µa δAa + J a δF a + i (η a δca − δca η a ) µ = δLeff g ω µa J (Dµ c)a + i − f abc η a cb cc + F a η a . This immediately leads from (13.86) the generating functional should be invariant (namely.

it is clear that δΓ δAa µ δΓ δF a δΓ δca δΓ δca δΓ a δKµ δΓ δK a = −J µa .89) where K stands generically for the sources for the composite BRST variations. 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. = = δW . we are interested in the identities satisfied by the proper (1PI) vertices of the theory.81)). K] − d4 x J µa Aa + J a F a + i (η a ca − ca η a ) . Most often. K] = W [J. = iη a . a δKµ δW .88) in terms of the generating functional for the proper vertices as .80) and the field variables are really the classical fields and we are dropping the superscript (c) for simplicity). = −J a . = iη a . δK a (13. however. µ (13. we have Γ[A.570 13 BRST invariance and its consequences usefulness of the sources for the composite BRST variations becomes evident. we see that we can rewrite the master equation (13. From the definition of the generating functional for the proper vertices (see (13.90) Using these definitions.

(13.) . the counter terms (quantum corrections) should satisfy such a relation. The BRST invariance. like the Slavnov-Taylor identities. (We would discuss renormalization of field theories in a later chapter.4 Slavnov-Taylor identity 571 d4 x δΓ δΓ δΓ δΓ δΓ + a − F a (x) a = 0.92) and setting all Taking derivative of this with respect to field variables to zero gives δ2 Γ δ2 Γ δ2 Γ = 0. for example. This is essential in proving the renormalizability of gauge theories. is very fundamental in the study of gauge theories as far as renormalizability and gauge independence of physical observables are concerned. For example. a (x) δK µa (x) a (x) δAµ δc (x) δK δc (x) (13. δ2 δF b (p)δcc (−p) d4 k (13.13.91) This is the master equation from which we can derive all the relations between various (1PI) proper vertices resulting from the BRST invariance of the theory by taking functional derivatives with respect to (classical) fields. the gauge dependence of the effective potential in a gauge theory (with scalar fields) as well as the gauge independence of the physical poles of the propagator can be obtained systematically from the Nielsen identities which.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. let us note that we can write the master identity (13. in this way. Thus. follow from the BRST invariance of the theory.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. consequently.

Let us recall that the total Lagrangian density after gauge fixing (in the covariant gauge) has the form (13.95) LQ = − = = 1 a µν A η 2 µ 1 a µν ab b A O Aν + ca M ab cb . ξ M ab = −δab .2) 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ 2 + ∂ µ ca Dµ ca . (They are the Green’s functions of the free theory. 2 µ where we have neglected total divergence terms and have identified Oµν ab = δab η µν 1 µ ν ∂ ∂ .96) take the forms . 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. (13. These are the essential elements in carrying out any perturbative calculation in a quantum field theory. We note that since gauge fixing breaks gauge invariance.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. − 1− (13.94) The propagators of the theory can be derived from the free part of the Lagrangian density which is quadratic in the field variables 1 ∂µ Aa − ∂ν Aa (∂ µ Aνa − ∂ ν Aµa ) ν µ 4 1 2 ∂ µ Aa + ∂ µ ca ∂µ ca − µ 2ξ − 1− 1 µ ν a ∂ ∂ Aν − ca ca ξ (13.96) 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).572 13.) The inverses of the two point functions are easily calculated in the momentum space where the quadratic operators in (13.

13. Therefore. p2 (13.101) 1 β µ (αp2 + 1)δµ − α(1 − ) − p pλ = 0. λ or. or.5 Feynman rules 573 M ab (p) = δab p2 . 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 . Oµν ab (p) = −δab η µν p2 − 1 − 1 µ ν p p . With this parameterization. we note that the inverse is defined to satisfy (repeated indices are summed) −1bc Oµν ab (p)Oνλ (p) = δac δµ .97) The inverse of M ab (p) in (13.99) for the inverse and carrying out the multiplication explicitly we have η µν p2 − 1 − or. β are arbitrary parameters to be determined (they are not necessarily constants and can be Lorentz invariant functions of the momentum). p2 (13.97) is quite simple (it is like the propagator of a massless scalar field) M −1ab (p) = δab . λ ξ ξ .99) where α. λ ξ (13. ξ (13.100) Substituting the parameterization (13. η µν p2 − 1 − 1 µ ν −1ac p p O νλ (p) = −δac δµ . ηµν ). let us derive this inverse systematically.98) while the derivation of the inverse of Oµν ab (p) is a bit more involved. 1 µ ν p p ξ αηνλ + β pν pλ p2 = −δµ . λ ξ ξ (13. we can parameterize the most general form of the inverse as O−1ab (p) = δab α ηµν + β µν pµ pν . λ αp2 δµ − α 1 − λ 1 1 µ p pλ + β pµ pλ − β 1 − pµ pλ = −δµ .

) represents the integral over p and we deduce from this that (see (13. it follows that 1 . β = α(ξ − 1) = − so that we can write the inverse in (13.79). . 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.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.102) α=− α 1− or. ξ ξ 1 (ξ − 1). p2 (13.Oµν (p)J νb (p) +i η a (−p). 2 p p −1ab Oµν (p) = − (13. p2 1 β − = 0.99) as pµ pν δab ηµν + (ξ − 1) 2 .M −1ab (p)ηb (p) Z0 = N e i .126)) h −1ab i − 2 J µa (−p). 2 p p (13.104) where ( .103) With these.574 13 BRST invariance and its consequences Setting the coefficients of each distinct Lorentz structure to zero. see (12. 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 . (13.

p2 (13.108) = iGab (p) = We can now write the complete generating functional as Z[J] = exp i d4 x Lint 1 δ i δJ Z0 [J]. (13. while for ξ = 1. b pµ pν iδab ηµν + (ξ − 1) 2 . a p ν. 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 . The propagators can be diagrammatically represented as = iGab (p) µν = − a p b (13. p2 (13. µ . when we are in the Landau gauge this propagator is transverse. the gauge is known as the Feynman gauge where the propagator has a much simpler form which is more suitable for perturbative calculations.e.106) which defines the ghost propagator (which is sometimes also denoted by Dab (p)). 2 p p iδab . i.5 Feynman rules 575 When ξ = 0..109) where we have used J generically for all the sources and the interaction Lagrangian density can now be identified with Lint = LTOT − LQ = gf abc ∂µ Aa Aµb Aνc − ν g2 abp cdp a b µc νd f f Aµ Aν A A 4 (13.13.107) µ.110) −gf abc Aa ∂ µ cc cb .

p4 ) = i (13. p2 . p3 .576 13 BRST invariance and its consequences Let us now derive the interaction vertices for the theory which are more useful in the momentum space (where Feynman diagram calculations are primarily carried out). 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 . 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. V µνλρ abcd (p1 .112) δ4 S δAa (p1 )δAb (p2 )δAc (p3 )δAd (p4 ) µ ν ρ λ = i δ4 Sint δAa (p1 )δAb (p2 )δAc (p3 )δAd (p4 ) µ ν ρ λ .111) × (−ipµ ) Aa (p1 ) cc (p3 ) cb (p2 ) . µ 3 The tree level 3-point and 4-point interaction vertices of the theory follow from this to correspond to (the restriction | denotes setting all field variables to zero) V µνλ abc (p1 . p2 .

115) . p3 . (13. p2 ν. (13. b = V µνλ abc (p1 . p2 . d λ. p3 ). a p2 b p2 ν.114) = −(2π)4 gf abc pµ δ4 (p1 + p2 + p3 ). p3 ). p4 ).112)-(13. b c = V µ abc (p1 .5 Feynman rules 577 = −(2π)4 ig2 δ4 (p1 + p2 + p3 + p4 ) f abp f cdp η µλ η νρ − η µρ η νλ +f acpf dbp η µρ η νλ − η µν η λρ +f adp f bcp η µν η λρ − η µλ η νρ . with (13. a p3 p1 µ. p2 . 3 Therefore. p2 .113) V µ abc (p1 . c p3 p1 µ. c = V µνλρ abcd (p1 . p4 p3 p1 µ. p2 .13. p3 ) = i δ3 S δAa (p1 )δcb (p2 )δcc (p3 ) µ = i δ3 Sint δAa (p1 )δcb (p2 )δcc (p3 ) µ (13.114) we can represent all the interaction vertices of the theory as λ. a ρ.

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

n·p iGab = (13.120) and the antighost-ghost-gauge vertex involves a multiplicative factor of nµ . either open lines or closed loops.118) so that we have (independent of whether n2 = 0 or n2 = 0) pν . µν (13. (n · p) nµ Gab = −ξδab µν nµ nν Gab = −ξδab . will have the structure (we are not writing the possible multiplicative Lorentz factors involving nµ which do not effect the integral and we are evaluating the integral in D dimensions) Figure 13. .13.1: Examples of diagrams with open and closed ghost lines.6 Ghost free gauges 579 With such a choice of gauge the gauge propagator can be calculated to have the form iGab µν = iδab n µ pν + n ν pµ n 2 pµ pν 1 − − ηµν + p2 (n · p) (n · p)2 pµ pν −ξ . the ghost propagator in the axial-like gauge (13.116) in this theory has the form δab .119) Similarly. Therefore. (n · p)2 (13. the integrand of any diagram involving ghosts.

Using the fact that the integral is Lorentz invariant.10)). we can rewrite this also as 1 m dxi 0 i=0 δ(1 − x0 − x1 − · · · − xm ) (n2 )(m+1)/2 dD k = 0. An alternative way to see this is to note that the generating functional in the axial-like gauge (13. (13. 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. in (15. (n · k)m+1 δ(1 − x0 − x1 − · · · − xm ) m+1 = 0 i=0 dxi (13. This shows that all the diagrams involving ghost interactions can be regularized to zero so that the ghost degrees of freedom are irrelevant with such a choice of gauge.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. (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 ) .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.580 13 BRST invariance and its consequences dD k = 1 (n · k) (n · (k + p1 )) .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 .

since the generating functional does not depend on the functional form of G [f a ] (up to irrelevant multiplicative constants). (13. Becchi.7 References 581 G[f a ] = e i − 2ξ R d4 x f a f a . .126) This shows that in this gauge the ghosts decouple completely and can be absorbed into the normalization factor. Theoretical and Mathematical Physics 10. Physics Letters 52B. The axial-like gauge choices (13. (13. therefore. C. 13.116) in a non-Abelian gauge theory. Taylor. A. Note that such a decoupling works only because nµ (in n · Aa inside the determinant) is a multiplicative operator. 3. the covariant gauge. 436 (1971). 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 . (13. A.124) However. Rouet and R. Slavnov.125) In this case. correspond to ghost free gauges very much like in the Abelian gauge theories. J. Stora. A.7 References 1. let us choose G[f a ] = det n · ∂δab + gf abc f c −1 − i R d4 x f a f a 2ξ e . 99 (1972). C. 2. It would not work in say.13. Nuclear Physics B33. 344 (1974).

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

in this chapter we would discuss this important question of massive gauge fields concluding with the standard model of the electroweak interactions. a massless gauge field is completely consistent with the observed fact that electromagnetic forces are long ranged. Therefore.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 .Chapter 14 Higgs phenomenon and the standard model In chapters 9. 14.14) or (12. L = − Fµν F µν + 4 2 583 (14. 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. As we have mentioned earlier gauge fields can be thought of as the carriers of physical forces. Therefore.2) . 12 and 13 we have studied gauge theories (both Abelian and non-Abelian) in great detail.1) is not invariant under the gauge transformation (9. However. 2 M 2 a µa Aµ A .34) respectively. 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 . 2 or (14.

5) in a simple manner and in momentum space has the form .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).3) The Lagrangian density (14. ∂µ F µν + M 2 Aν = 0. we also note here that the theory (14. + M 2 Aν = 0.6) Equation (14. for example. we obtain ∂ · A = 0.4) Contracting (14.2) has the form ∂L ∂L − = 0. (14.5) and substituting this back into the Euler-Lagrange equation in (14. ∂∂µ Aν ∂Aν ∂µ or.6) together with the constraint (14.2) describes a massive photon which can be seen as follows.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. or. (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 . (14. −∂µ F µν − M 2 Aν = 0.4) with ∂ν and using the anti-symmetry of the field strength tensor. The propagator for this theory can be worked out (see. The Euler-Lagrange equation following from (14. section 13. (14. Note that the field variable Aµ has four field degrees of freedom while the constraint (14. As has been mentioned earlier.2) does not have a gauge invariance (because of the mass term) unlike the Maxwell theory.4) yields ∂µ (∂ µ Aν − ∂ ν Aµ ) + M 2 Aν = 0.5) defines the Proca equation (and (14.

it is sufficient to note that under the transformation (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.9) Aµ + M 1 1 1 ∂µ χ − ∂µ χ → Aµ + ∂µ θ + ∂µ θ M e M e 1 = Aµ + ∂µ χ.8) was constructed to be invariant under a Maxwell-like gauge transformation Aµ (x) → Aµ (x) + 1 ∂µ θ(x). As a result of these difficulties. e M θ(x). The limit M → 0 of the Proca theory is clearly quite subtle. M (14.2). The field strength tensor is.9) where e denotes the coupling constant of QED (this can be set to unity. in addition to the spin 1 field.8) = − Fµν F µν + 4 2 2 which. χ(x) → χ(x) − e (14. (14. but we have kept it for consistency with earlier discussion.¨ 14. This leads to difficulties in establishing renormalizability of the (interacting) Proca theory for massive photons. of course. In contrast to the Proca theory (14.1 Stuckelberg formalism 585 iGµν (p) = − p2 Thus. invariant under the gauge transformation (14.10) . 2 −M M2 (14.83) as well as for discussions in connection with the Higgs phenomenon in the next section). also contains a dynamical charge neutral spin zero field (real scalar field) which mixes with Aµ .9).7) M2 1 1 Aµ Aµ + ∂µ χ∂ µ χ + M Aµ ∂µ χ. the St¨ckelberg theory u (14. 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 − . say in (9. To see the invariance of the full theory.

On the other hand. the there is no difficulty in establishing renormalizability in such a gauge. This is typical of the behavior of theories in a unitary gauge.15) this would induce a ghost action.12) then the Lagrangian density (14. for example.11) which can be achieved by the choice of the gauge transformation parameter e χ(x).2) and this makes it clear that the Proca theory can be thought of as the gauge fixed St¨ckelberg theory. the massless limit M → 0 is now straightforward.7)) i pµ pν ηµν − . but would lead to propagators that are well behaved at high momentum.586 14 Higgs phenomenon and the standard model Because of the gauge invariance of the theory. Furthermore. if we choose a covariant gauge condition which leads to a gauge fixing Lagrangian density of the form (see. the gauge propu agator (see also (14. M θ(x) = (14. L = − Fµν F µν + 4 2 (14. in this unitary gauge. 2ξ − (14.3 and we note that this is known as the ’t Hooft gauge which we will discuss in the next section) 1 (∂µ Aµ − ξM χ)2 . we can choose a gauge and if we choose the unitary gauge χ(x) = 0.14) has the unpleasant features described earlier. section 12. . (14.8) takes the form M2 1 Aµ Aµ .13) We recognize this to be the Proca theory (14. 2 −M M2 iGµν (p) = − p2 (14. As a result. However.

(14. Here the field strength tensor and the operator V are defined as Fµν V = ∂µ Aν − ∂ν Aµ + ig [Aµ . Aν ] . for example. it is sufficient to note that . This Lagrangian density can be easily checked to be invariant under the gauge transformations (see. (12.1 Stuckelberg formalism 587 All of this analysis can be carried over to non-Abelian gauge theories as well as to the Einstein theory of gravitation.41)) Aµ → U Aµ U −1 − V → U V. ig (14. 1 (∂µ U )U −1 .18).17) where χ denotes a matrix valued scalar field. 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 . = e M χ.1). To see that the complete Lagrangian density is also invariant. say. SU (n) (as we have discussed earlier in section 12. ig (14.38) and (12.18) As we have already seen in chapter 12.¨ 14.16) ig ig where the trace is over the fundamental representation of the group. We will now briefly describe the St¨ckelberg formalism for non-Abelian gauge u theories.

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 (14.588 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. as before.23) . ig (14.21) (14. 2 − M2 p M2 iGab (p) = − µν (14. (14.20) which can be achieved by the choice of the gauge transformation matrix U (x) = V −1 (x).18).16) becomes 1 L = − Tr Fµν F µν + M 2 Tr Aµ Aµ . In this gauge. 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 − . If we choose the unitary gauge condition V = ½.2) to the non-Abelian case and describes a massive spin 1 field belonging to SU (n). 2 (14.

5). but now interacting with an Abelian gauge field as well (scalar QED).24) 4 4 where the theory has a “wrong” sign for the mass term and the covariant derivative is defined as in (7. For simplicity we consider again the self-interacting theory of a complex scalar field (as in section 7.24) can be checked easily (see (7. (14. e δφ = −iǫ(x)φ(x).2 Higgs phenomenon 589 with the unpleasant features discussed earlier that are characteristics of the unitary gauge choice.14.25) 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. In that case. The Lagrangian density for such a theory is given by (with λ > 0. δAµ = 1 ∂µ ǫ(x). see also section 7. if we choose a covariant gauge fixing condition (a generalization of (14.5 we studied the phenomena of spontaneous break down of a global symmetry in the self-interacting complex scalar field theory where the mass term for the scalar field had the “wrong” sign.2 Higgs phenomenon In section 7. This is known as the Nambu-Goldstone phenomena and the massless scalar field is known as the Nambu-Goldstone boson.85)) to be invariant under the infinitesimal local transformations δφ† = iǫ(x)φ† (x).15)).83) Dµ φ = ∂µ φ + ieAµ φ.26) . (14.84) and (7. 14. On the other hand.6) λ 1 2 L = − Fµν F µν + (Dµ φ)† (D µ φ) + m2 φ† φ − φ† φ . Let us next ask what happens if there is a spontaneous breakdown of a local symmetry. the propagator will have the correct asymptotic behavior for large values of the momenta.

the normal vacuum for which φ = φ† = 0.5. this actually corresponds to the local maximum of the potential.27) does not correspond to the true vacuum. since the mass term for the scalar field has the “wrong” sign.5.29) For further analysis.26) does not allow a mass term for the gauge field. (14. (14. 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 . as we have discussed in section 7.77)) .30) 2 16 Let us now rewrite the theory around the true vacuum by shifting the field variables as (see (7.590 14 Higgs phenomenon and the standard model The gauge boson in this theory is massless because the gauge symmetry (14.64).42)) 1 φ = √ (σ + iχ). (14. let us rewrite the Lagrangian density (14. λ where (see also (7. this is just one of them) 2m σc = σ = √ . On the other hand. (As we have seen in section 7.) Rather the true vacuum of the theory satisfies (see (7.24) in terms of the σ. 2 χc = χ = 0. there is actually an infinity of minima of the potential.28) (14.

34) 2 2 We note here that the χ field appears to be massless in this theory as in (7. the Lagrangian density (14.33) 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 . λ In this case.2 Higgs phenomenon 591 2m σ → σ + σ = σ + √ = σ + v.32) to simplify some of the terms. Let us look at only the terms in (14. (14.33) where we have used the form of v in (14.31) where we have identified the vacuum expectation value (vev) of the field with 2m σ =v= √ .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. 16 (14. λ χ → χ + χ = χ. but it now mixes with the gauge field Aµ so that we have .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 ).79). (14.

8).79).38) and a massive gauge field Bµ with mass (this is the analog of the unitary gauge in (14.37) 2 We see from (14. (The Lagrangian density in (14. σ (14.39) .35) ∂µ Bν − ∂ν Bµ = ∂µ Aν + 1 1 ∂ν χ − ∂ν Aµ + ∂µ χ ev ev = ∂µ Aν − ∂ν Aµ . 1 1 ∂µ χ Aµ + ∂ µ χ ev ev 2 1 = Aµ Aµ + Aµ ∂µ χ + 2 2 ∂µ χ∂ µ χ.34).11)) 4e2 m2 .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 . (14.37) that the quadratic part of the shifted Lagrangian density is completely diagonalized. let us define 1 ∂µ χ. it describes a massive scalar field σ with mass m2 = 2m2 . ev e v Aµ + Bµ B µ = (14. λ m2 = M 2 = e2 v 2 = B (14. ev Bµ = Aµ + then we have (14.35) the quadratic Lagrangian density (14. As in (7.36) Thus in terms of this new field variable (14. ignoring the σ field should be compared with the St¨ckelberg Lagrangian density u (14.) To achieve this.592 14 Higgs phenomenon and the standard model to diagonalize the theory in order to determine the true spectrum of the theory.

The answer to this question lies in the fact that if a gauge theory is written in a manifestly Lorentz invariant way. namely. Brout and Englert. Hagen and Kibble.2 Higgs phenomenon 593 The other real component χ of the spin 0 boson . we can see this happening in the following way. however. It is. of course. This is consistent with the unitarity of the theory (from the point of view of field degrees of freedom).) However. That is there can arise states in the Hilbert space with negative norm. Guralnik. we have a mechanism for giving masses to gauge bosons without violating renormalizability and unitarity of the theory. the Goldstone mode of the complex scalar field. Hence the gauge field has become massive with three degrees of freedom without violating unitarity.the NambuGoldstone boson . Intuitively. why is there no massless particle even though we have a spontaneous breakdown of symmetry. A crucial assumption in the Goldstone theorem is that the theory should be manifestly Lorentz invariant as well as the metric of the Hilbert space should be positive semi-definite. This 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. it is the incompatibility of these two conditions that avoids the Goldstone theorem. Thus the other two states must have cancelling effect with each other as we have seen earlier. The field Aµ is a four vector and hence must have four independent polarization states. In fact. 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. (We can think of the physical degrees of freedom as the two transverse components. what has happened is that the original gauge field which was massless had only two physical degrees of freedom just like the photon. consist only of transverse polarization. The physical states. Thus some states must have negative norm in such a description. the massless mode has combined with the massless gauge field to give it an additional degree of freedom.has completely disappeared from the spectrum of the theory. . The scalar field responsible for this phenomenon is conventionally known as the Higgs field. When we are dealing with a gauge theory.14. namely. At this point we may ask why the Goldstone theorem fails in this case. this second aspect that is important in developing a gauge theory of the weak interactions. as we have seen (recall Gupta-Bleuler quantization in 9.8) the metric of the Hilbert space becomes indefinite.

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 (14.84)) δσ(x) = ǫ(x)χ(x). it is not the right gauge to study renormalizability of the theory. To introduce the gauge choice. 2ξ 2 (14. as pointed out in the last section.40) e The shifted theory (14. This is achieved by the ’t Hooft gauge fixing which is described by the gauge fixing Lagrangian density LGF = − = − 1 1 µ (∂ Aµ − ξevχ)2 = − (∂ µ Aµ − ξM χ)2 .39).34) (namely. The presence of the local symmetry (14. eliminate the mixing term between the Aµ and the χ fields). that is more useful for this purpose let us proceed as follows. (14.41) e where M is defined in (14. δχ(x) = −ǫ(x)σ(x).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). 2ξ 2ξ ξM 2 2 1 µ (∂ Aµ )2 + M (∂ µ Aµ )χ − χ . e 1 ∂µ ǫ(x).42) . δχ(x) = −ǫ(x)(v + σ(x)) = −ǫ(x) δAµ (x) = M + σ(x) .33) which reflects a spontaneous breakdown of the local symmetry (14. First we note that the unshifted theory (14. commonly known as the ’t Hooft gauge or the Rξ gauge.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. 1 ∂µ ǫ(x). (14.30) is invariant under the infinitesimal gauge transformations (see (7.

42) can also be written with an auxiliary field as (see (12.164). 2 (14. the ’t Hooft gauge choice in (14. the propagators for the Aµ and the χ fields in this gauge are given respectively by iGµν (p) = − iGχ (p) = p2 p2 i .34) together with the cross terms in (14.14. thereby.46) and can be easily checked to be invariant under the BRST transformations . − ξM 2 pµ pν i ηµν − (1 − ξ) 2 .2 Higgs phenomenon 595 which can be compared with (14.42) (or (14. The gauge fixing Lagrangian density (14.44)) leads to an interacting ghost Lagrangian density. disappear from the action. The complete Lagrangian density in this gauge has the form LTOT = L + LGF + Lghost . (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.45) We see that unlike in the conventional covariant gauge fixing (12. (14.43) and we note that in this gauge the propagators fall off for large values of the momenta unlike in the unitary gauge.42) combine into a total divergence and. As a result. It is clear that the mixing terms in (14.15). 2 −M p − ξM 2 (14.170)) LGF = ξ 2 F − F (∂ · A − ξM χ).

2 (14. Therefore. 14. weak interactions (for example. e δσ = ωχc. (13.49) .48) where Pσ and Pχ denote respectively the sources for the composite BRST variations of the fields σ and χ in (14.4 (see. for example. e δχ = −ω δc = 0. δAµ (x) δσ(x) δPσ (x) δχ(x) δPχ (x) δc(x) (14. (14. let us review some of these essential earlier results before constructing the standard model (known as the Weinberg-SalamGlashow theory).596 14 Higgs phenomenon and the standard model δAµ = 1 ω∂µ c. This is the starting point for establishing the renormalizability of the gauge theory with a spontaneously broken symmetry (Higgs theory).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. M + σ c.47) Following the discussions in section 13. At low energies.47). δF = 0.3 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.) are described quite well by a specific form of the Fermi theory which involves four fermion interactions in the form of a current-current interaction Hamiltonian density GF † HI = √ Jµ J µ . δc = −ωF. the β decay of the neutron etc.

Furthermore. 2 Q = I3L + (14.52) where Q denotes the electric charge of the particle. ψν denoting the fermion fields associated with neutron. 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 .51) The weak interactions are.50) with ψn .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. therefore. 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). proton.14. ψe . ψp . (14. Experimentally .17 × 10−5 (GeV)−2 . electron and neutrino respectively. it is clear from the V − A (vector minus axial vector) structure of the currents in (14. I3L the quantum number associated with the 3-component of the weak isospin generator and Y its weak hypercharge quantum number. quite weak and were also known to be extremely short ranged. (The V − A structure is connected with the fact that parity is violated maximally in weak interactions.50) have a weak isospin structure and are also known as the weak isospin currents.139) and (3. (14.) In addition to the weak isospin quantum number. the currents in (14.3 The standard model 597 where the current is a sum of the hadronic and leptonic currents of the form (had) (lept) Jµ = Jµ + Jµ = ψ p γµ (1 − γ5 )ψn + ψ ν γµ (1 − γ5 )ψe + · · · . As a result.140) for a definition of right and left-handed particles/fields). 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.

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

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

53) to describe the local (gauge) symmetry of the theory. Let us now write down the quantum numbers for the matter and the Higgs field and verify explicitly that (14.52). a Wµ . so that the theory should contain four gauge fields in total. the minimal multiplet of Higgs fields that can achieve the spontaneous breakdown of the symmetry (as in (14.600 14 Higgs phenomenon and the standard model (The hypercharge quantum numbers for the quarks are also different from those for leptons in order to be compatible with (14. If we assume (14.54)) is given by a doublet of charged fields. a = 1. if we would like the gauge bosons to be massive. minimal Higgs field : φ = φ+ φ0 . (φ+ )† = φ− . it carries a nontrivial weak hypercharge quantum number given by Y = 1 which is. Furthermore.57) Yµ . the opposite of that for the left-handed lepton doublet.52) holds. (14. the local symmetry of the electroweak group must be spontaneously broken through the Higgs mechanism and this requires that the theory should also contain scalar Higgs fields which would be described by multiplets of SUL (2). UY (1) : 1 gauge field. . in fact. (14. 3. (φ0 )† = φ . 2.) The gauge fields which are expected to correspond to the carriers of forces are determined from the local symmetry structure of the theory.58) 0 Since the Higgs multiplet is charged. then the corresponding gauge fields are determined to be SUL (2) : 3 gauge fields. As we will see shortly.

) The Lagrangian density (14. let us note that the gauge invariant Lagrangian density for the gauge fields follows from our earlier discussions (see (9. a a b c = ∂µ Wν − ∂ν Wµ − gǫabc Wµ Wν . First. 3.60) where the field strength tensors for the Abelian as well as the nonAbelian gauge fields are defined as (see (9. It is now a straightforward matter to write down the complete Lagrangian density for the theory which has the local gauge invariance SUL (2) × UY (1) and we will do this in several steps. 2. (14.44)) Fµν a Fµν = ∂µ Yν − ∂ν Yµ .60) .14.45)) to correspond to 1 1 a LG = − Fµν F µν a − Fµν F µν .52). Note that ǫabc denotes the structure constant of SUL (2) as we know from the study of angular momentum.3.16) and (12. but it will be present in the matter Lagrangian density.59) =0 = −1 −1 2 0 1 2 −1 2 −1 1 2 1 2 0 − 1 = −1 1 2 =1 1 −2 + 1 = 0 2 1 2 + −1 2 This shows that all the quantum numbers assigned to various fields are consistent with the relation (14.61) We are assuming that g and g′ are the coupling constants for the gauge interactions associated with the groups SUL (2) and UY (1) respectively. 14.3 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 (14.2 Lagrangian density. 4 4 a = 1. (14. (The Abelian gauge fields are not self-interacting which is why g′ does not appear in the Lagrangian density for the gauge fields.4) and (12.

g′ (14. 2 2 (14. we simply have to remember that there are two local symmetries present in the theory. (We note here that within the context of isospin. g g 1 ∂µ ǫ(x).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) . therefore. the gauge field Wµ for the group SUL (2) does not transform under the group UY (1) and vice versa.63) is that the right-handed fermions couple twice as strongly to the Yµ . sometimes it is conventional to denote the generators as τ by making the identification 2 σ = τ . Since the two symmetry groups are commuta ing (direct products). 3 and ǫ(x) denote the local infinitesimal parameters of symmetry transformations associated with the groups SUL (2) and UY (1) respectively. We have also used a vector notation here to represent the three components of the isospin vectors which is often used interchangeably. 2. 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. they are inert under the corresponding transformations.62) a δWµ (x) = δYµ (x) = where ǫa (x).) The other thing to note from the structure of (14.63) 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. a = 1. The gauge invariant Lagrangian density for the matter fields (leptons) can be obtained through minimal coupling. Since the left-handed fermions belong to an isospin doublet while the right-handed fermions are isospin singlets. the Lagrangian density for the leptons takes the form ig′ σ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R.

We note that we have combined the three components of ǫa (x).64) show explicitly that the right-handed fermions do not change under the SUL (2) transformations.14. The minimally coupled fermion Lagrangian density (14. a = 1. The Lagrangian density (14.65) We note that since the hypercharge of the Higgs doublet is just the opposite of that of the left-handed lepton doublet. δL(x) = −i (14.3 The standard model 603 field as the left-handed ones simply reflecting the Y quantum numbers for the two fields.64) together with the transformations in (14.62).62). 2 2 δφ(x) = −i (14. therefore.64). 2. A mass term as well as a potential for the Higgs field can be introduced consistent with the SUL (2) symmetry. we also need a Lagrangian density for the scalar Higgs fields which are expected to lead to a spontaneous breakdown of the local symmetries. since we . The transformations (14. 3 into a vector ǫ(x) in (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). massless in this theory to begin with). the coupling to the UY (1) gauge field is correspondingly with the opposite sign. 2 2 (14. 2 2 δR(x) = −iǫ(x)R(x). Note also that the SUL (2) invariance forbids a mass term for the lepton fields (the electron and the neutrino are.65) can be checked to be invariant under the infinitesimal local transformations i σ · ǫ(x)φ(x) + ǫ(x)φ(x). Besides the gauge fields and the leptons.66) together with the gauge field transformations in (14. 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.

(14. Therefore. 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. collecting all the terms in (14. 2 2 δR(x) = −iǫ(x)R(x).69) which is invariant under the infinitesimal local gauge transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). (14.60).68) can be checked to be invariant under the transformations (14. 2 2 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) . 4 λ > 0. φ† .24). g g 1 ∂µ ǫ(x). the fermions can have Yukawa interactions with the scalar fields which can be described by the Lagrangian density LY = −h Rφ† L + LφR . (14.67) and (14. φ† = −m2 φ† φ + λ † 2 φφ . (14. As we will see shortly. of course. g′ (14.63).604 14 Higgs phenomenon and the standard model would like the potential for the Higgs field to lead to the spontaneous breakdown of the local symmetries we choose it to have the form (see (14. (14.68). here the Higgs field is represented by an isospin doublet) V φ.65). (14. the Yukawa interaction leads to masses for fermions after the spontaneous breakdown of the symmetry. In addition.66). it is worth noting that in spite of the similarity in their forms for the potential.68) where h denotes the Yukawa coupling and the Lagrangian density (14.66).70) δYµ (x) = .67) which is invariant under the local transformations in (14.64) and (14. δL(x) = −i δφ(x) = −i a δWµ (x) = i σ · ǫ(x)φ(x) + ǫ(x)φ(x).

71) so that the minimum of the potential occurs at ∂V ∂φ ∂V ∂φ† λ † φφ 2 = φ† − m2 + = − m2 + = 0.73) φ φ† φ = = Furthermore. Both these issues can be addressed simultaneously if the local symmetries are spontaneously broken by the Higgs phenomenon and this is what we discuss next.14. there are two solutions to (14. φ† = −m2 φ† φ + λ > 0.72) λ † φφ 2 As we have seen in the last section. 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.3 The standard model 605 14. Furthermore.29)) 0 v √ 2 φ = .2) that the first solution in (14.3 Spontaneous symmetry breaking.67) in the form λ † 2 φφ . 2m2 . 4 V φ. furthermore. λ (14. the weak interactions are short ranged so that we would like the gauge bosons to be massive. λ (14. 2m v=√ . φ† = 0. we have noted earlier in section 7. (14.72). 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).73) represents a local maximum while the second solution corresponds to the true minimum of the potential. φ = 0. We have chosen the Higgs potential (14.74) .5 (and also in section 14. the factor of √2 below is a consequence of the definition in (14. (14.3. namely.

) 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 . before shifting the fields let us rewrite the Lagrangian densities for the Higgs and the Yukawa sector of the theory for simplicity (see (14.65). √ − 3 2Wµ −Wµ (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. the other parts will not change under the shift. (14. (14.77) σ · Wµ = 2 = 1 2 √ + 2Wµ φ+ φ0 . 2 As a result.67) and (14. this is the sector of the theory that involves the scalar field and will be affected by the shift. (Basically. (14.606 14 Higgs phenomenon and the standard model Therefore.58) we obtain . to determine the spectrum of particles we need to expand the theory around the true ground state (vacuum) by shifting the scalar field as 0 v √ 2 φ →φ+ φ =φ+ . it follows that 3 Wµ √ − 2Wµ (14.76) where we have defined 1 ± 1 2 Wµ = √ (Wµ ∓ iWµ ).68)). as we have discussed earlier.75) However.

79) The expansion around (14. φ† = ∂µ + ig σ ig′ · Wµ − Yµ φ 2 2 † ∂ µ + ig σ ig′ µ · Wµ − Y φ 2 2 2 −h Rφ† L + LφR + m2 φ† φ − = ∂µ + × + λ † φφ 4 ig − 0 ig′ ig 3 Yµ − Wµ φ− − √ Wµ φ 2 2 2 ∂µ − ig′ µ ig µ3 + ig φ + √ W µ+ φ0 Y + W 2 2 2 ∂µ + × ig 3 0 ig′ ig + Yµ + Wµ φ − √ Wµ φ− 2 2 2 ig ig′ µ ig µ3 0 φ + √ W µ− φ+ Y − W 2 2 2 0 ∂µ − −h eR φ− νeL + φ eL + ν eL φ+ + eL φ0 eR +m2 φ− φ+ + φ φ0 − 0 λ − + 0 2 φ φ + φ φ0 .3 The standard model 607 LH + LY − V φ. 2 v 0 0 φ → φ +√ .80) under which the Lagrangian density (14. 2 (14.79) in the Higgs and the Yukawa sector becomes LH + LY − V φ. φ† = ∂µ + ig − 0 ig 3 ig′ v Yµ − Wµ φ− − √ Wµ φ + √ 2 2 2 2 .74) can be done equivalently by letting v φ0 → φ0 + √ . 4 (14.14.

60) and (14. The spectrum of the theory can be obtained from the quadratic part of the Lagrangian density and the quadratic part of (14. the complete quadratic Lagrangian density is obtained to be . − √ (eR eL + eL eR ) − 8 2 0 3 We note that the field combinations (φ0 + φ ). we cannot truly determine the spectrum of the theory until we diagonalize the quadratic part of the Lagrangian density.81) . since as the fields Wµ there is mixing between various fields.63) respectively. (g′ Yµ + gWµ ) as well ± appear to have become massive.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. Adding the quadratic terms from the gauge and the fermion Lagrangian densities in (14.81) 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 (14.82) hv λv 2 0 0 2 φ +φ . However.

84) to write 3 Wµ = Zµ cos θW − Aµ sin θW . Let us define g′ g2 + g′2 g g2 + g′2 Zµ = Yµ + 3 Wµ 3 = sin θW Yµ + cos θW Wµ . 8 2 It is now a straightforward matter to diagonalize the Lagrangian density (14. (14.85) . Yµ = Zµ sin θW + Aµ cos θW .83) + λv 2 0 hv 0 2 − √ (eR eL + eL eR ) − φ +φ .84) where θW is known as the Weinberg angle (also called the weak mixing angle).83). We can invert the relations in (14.14. (14. Aµ = g g2 + g′2 Yµ − g′ g2 + g′2 3 Wµ 3 = cos θW Yµ − sin θW Wµ .3 The standard model 609 LQ = − 1 + + ∂µ Wν − ∂ν Wµ ∂ µ W ν − − ∂ ν W µ − 2 1 3 3 − ∂µ Wν − ∂ν Wµ ∂ µ W ν 3 − ∂ ν W µ 3 4 1 / / − (∂µ Yν − ∂ν Yµ ) (∂ µ Y ν − ∂ ν Y µ ) + ieL ∂ eL + iν eL ∂ νeL 4 +ieR ∂ eR + ∂µ φ− ∂ µ φ+ + ∂µ φ ∂ µ φ0 + / 0 g2 v 2 + µ− Wµ W 4 + igv + − (Wµ ∂ µ φ− − Wµ ∂ µ φ+ ) 2 iv 0 3 − √ g′ Yµ + gWµ ∂ µ φ − φ0 2 2 v2 ′ 3 g Yµ + gWµ 8 g′ Y µ + gW µ3 (14.

2 (14. 2 2 (14.610 14 Higgs phenomenon and the standard model With this. (14. if we define the parameters gv . 2 i 0 √ φ − φ0 . (14. 2 λ 4m2 λv 2 = 2 2 λ = 2m2 . we can define (this is equivalent to defining φ0 = iχ).88) Furthermore. for example. 2 cos θW hv √ . we can write the quadratic part of the Lagrangian density (14. φ = 0 1 √ (σ 2 + − iχ). 2 MW g2 + g′2 v = .87) so that we can write 0 ∂µ φ ∂ µ φ0 = 1 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ. (14.29)) σ = χ = 1 0 √ φ + φ0 . see.83) as . it is easy to check that 3 3 ∂µ Wν − ∂ν Wµ = (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) + (∂µ Yν − ∂ν Yν ) (∂ µ Y ν − ∂ ν Y µ ) + (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) .86) 1 √ (σ 2 ∂µW ν 3 − ∂ν W µ 3 Similarly.89).89) MW MZ = = me = 2 MH = then together with (14.84)-(14.

90) 2 2 where we have combined the two chirality states eL . eR to define the massive electron field as (see (14. 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 .14.91) .56)) e = eL + eR .(14.

Zµ is also massive with mass . MW 1 ∂µ χ → Zµ . namely.94) while the neutrino remains massless as we would like. Although the starting theory had a massless electron.92) Zµ − then the quadratic Lagrangian density (14. 2 (14.90) becomes completely diagonal with the form LQ = − 1 + + ∂µ Wν − ∂ν Wµ 2 2 + ∂ µ W ν − − ∂ ν W µ − + MW Wµ W µ − 1 M2 − (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) + Z Zµ Z µ 4 2 1 1 − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) + ∂µ σ∂ µ σ − 4 2 / / +e(i∂ − me )e + iν eL ∂ νeL . MZ (14. 1 2 2 M σ 2 H (14. The charged ± vector (spin 1) bosons Wµ are massive with mass MW = gv .93) The quadratic Lagrangian density (14.93) is now completely diagonalized and shows some very desirable features.38)). 2 (14.95) One of the two neutral vector bosons. we see that after spontaneous breakdown of symmetry the electron has become massive with mass hv me = √ .612 14 Higgs phenomenon and the standard model ± Wµ ∓ i ± ∂µ φ± → Wµ . The Higgs √ particle has the usual mass MH = 2m (see (14.

namely.84)) g′ g2 + g′2 g g2 + g′2 sin θW = cos θW = .97) There is another neutral vector boson in the theory (14. Aµ which is massless.93).14.3 The standard model 613 MZ = MW g2 + g′2 v = . 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). That this is indeed true can be seen as follows. . (14. 2 cos θW (14.63) Lf = iLγ µ ∂µ + ig = i ν eL e L σ ig′ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R 2 2   ′ ig + 3 ∂µ + ig Yµ + ig Wµ νeL + √2 Wµ eL 2 2  γµ  ig ′ ig ig 3 − ∂µ + 2 Yµ − 2 Wµ eL + √2 Wµ νeL ig + ig 3 ig′ Yµ + Wµ νeL + √ Wµ eL 2 2 2 ig′ ig − ig 3 Yµ − Wµ eL + √ Wµ νeL 2 2 2 + ieR γ µ ∂µ + ig′ Yµ eR = iν eL γ µ + ieL γ µ ∂µ + ∂µ + + ieR γ µ ∂µ + ig′ Yµ eR / / / = iν eL ∂ νeL + ieL ∂ eL + ieR ∂ eR g g + − − √ Wµ ν eL γ µ eL − √ Wµ eL γ µ νeL 2 2 g 3 − Wµ (ν eL γ µ νeL − eL γ µ eL ) 2 .96) where the Weinberg angle θW is defined by (see (14. Let us look at the minimally coupled fermion Lagrangian density (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. (14. both charged as well as neutral weak currents given by g g − Jµ = − √ ν eL γµ eL = − √ ν eL γµ (1 − γ5 ) eL . We also note from (14.98) that the standard model has. (14. in addition to the correct electromagnetic current.98) We note from (14. 2 2 2 g g + Jµ = − √ eL γµ νeL = − √ eL γµ (1 − γ5 ) νeL . 2 2 2 .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.614 − 14 Higgs phenomenon and the standard model g′ Yµ (ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) 2 / / / = iν eL ∂ νeL + ieL ∂ eL + ieR ∂ eR g − + − √ Wµ ν eL γ µ eL + Wµ eL γ µ νeL 2 g − (Zµ cos θW − Aµ sin θW ) (ν eL γ µ νeL − eL γ µ eL ) 2 g′ − (Zµ sin θW + Aµ cos θW )(ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) 2 / / = iν eL ∂ νeL + ie∂ e g − + − √ Wµ ν eL γ µ eL + Wµ eL γ µ νeL 2 − Zµ 2 g2 + g′2 gg′ g2 + g′2 g2 (ν eL γ µ νeL − eL γ µ eL ) + g′2 (ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) − Aµ eγ µ e.

98).102) which leads to 2 MW = (14. W Figure 14.49) (for just one family of leptons) and comparing the two we determine G √F 2 g2 . 8MW (14. 4 2GF 4 2GF sin2 θW = (14.3 The standard model 0 Jµ = − 615 g2 (ν eL γ µ νeL − eL γ µ eL ) 1 2 g2 + g′2 +g′2 (ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) .103) Experimental measurements lead to a value of the Weinberg angle as . they would lead to a low energy effective current-current interation Hamiltonian density in the lowest order of the form (basically.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 .101) This has exactly the form of the V − A four-fermion interaction in (14. If we look at the charged current interaction terms in (14. two charged currents interact through the charged W gauge boson propagator (exchange) as shown in Fig.100) We recall that the V − A theory describes only charged current interactions. (14.1: Charged current-current interaction in the lowest order.14. 14. 2 8MW g2 e2 √ = √ .

for example.105) These gauge bosons have been discovered and their masses are determined very close to the theoretically predicted values. Proca. 4π 137 we can now determine MW MZ = = GF ≃ 10−5 (GeV)−2 . (14.4 References 1. (14.23. the WeinbergSalam-Glashow theory does not predict a unique mass for the Higgs particle since its mass depends on the quartic coupling λ. et le Radium 7. 347 (1936). cos θW 1 2 ≃ 80 GeV.616 14 Higgs phenomenon and the standard model sin2 θW ≃ 0. (14. Search for Higgs particles is a top priority in many experiments at the upcoming LHC. 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).104) and using the values of other constants (see. A. the standard model differs from the conventional Fermi theory in (14. The standard model compares very well with the experimental results. Furthermore. Once again these can be accomodated into extensions of the standard model that we will not go into here. . Such processes have also been experimentally observed. However.51) and recall that we have set = c = 1) e2 1 ≃ . although bounds on the value of its mass can be put from various other arguments. J. 14. de Phys. e2 √ 4 2GF sin2 θW MW ≃ 90 GeV.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.

P. C. G. C. 5. C. 8. . Physical Review 109. Gell-Mann. 1860 (1958). 6. 585 (1964). G. 1156 (1966). Marshak. International Journal of Modern Physics A19. Kibble. G. P. E. 7. Weinberg. R. Guralnik. 193 (1958). E. 167 (1971). Nuclear Physics B35. Proceedings of PaduaVenice conference on mesons and newly discovered particles. F. ’t Hooft. J. Reviews of Modern Physics 46. St¨ckelberg. Bernstein. 3265 (2004). 9. Physical Review 145. Physical Review 109. H.14. S. G. Helvetica Physica Acta 30. Higgs. 4. 1264 (1967). 7 (1974). u 3. 11. W. 209 (1957). Sudarshan and R. R. Physical Review Letters 13. Feynman and M. (1957). S. Ruegg and M. B. E. Physical Review Letters 13. Ruiz-Altaba. Englert and R. 10. Hagen and T. 321 (1964). Physical Review Letters 19.4 References 617 2. W. Brout.

This page intentionally left blank .

as renormalization of a quantum field theory and involves several steps. let us look at one of the simplest interacting field 619 .1 Introduction So far we have discussed the basic structures of various quantum field theories. 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. any scattering matrix element (S matrix element) in a given quantum field theory can be built out of 1PI (one particle irreducible) graphs. the evaluation of Feynman diagrams may lead to divergences depending on the structure of the integral that is being evaluated. We will study this question in a systematic manner developing the necessary ideas in this chapter as well as in the next. For example.Chapter 15 Regularization of Feynman diagrams 15. As a result.6 there was only one internal propagator). 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 such a case. However.6.4). This procedure is known. we have to find a way of extracting meaningful results from such a quantum field theory. The simple diagrams in 9. As we have discussed earlier (see section 13.6 are known as tree diagrams where the momenta of all the internal propagators are uniquely determined (through energy-momentum conserving delta functions) in terms of the external momenta (in the diagrams in 9. as we go to higher orders in perturbation. We have also carried out a few calculations of simple Feynman diagrams describing physical processes in section 9. in general. Instead some of the internal propagators may involve additional momenta (loop momenta) that are integrated over all possible values.

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

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

Realizing that with every internal line there is a momentum which is integrated and that each vertex has a delta function enforcing the conservation of energy and momentum at that vertex. (15.3). Let us recall that in the path integral description of a quantum field theory. . 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 . we have set = 1 throughout our discussion. be thought of as the parameter of expansion in the number of loops. therefore. the number of loops (independent momenta) in such a graph.4) where I denotes the number of internal lines and V the number of interaction vertices in the graph.4) and (15. the generating functional is defined as (see. (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. the power of “a” in a diagram in such a theory also determines the number of loops in the diagram and can.622 15 Regularization of Feynman diagrams P = I − V.5).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. (15. for example.6) In other words. is given by L = I − V + 1. we can eliminate I − V to obtain L = P + 1. therefore. 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). From the two relations in (15. (15.120). (12.

when we perturb around the true vacuum (by shifting the field variable by its vacuum expectation value). for example. for example. 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=− . let us calculate the simplest nontrivial graph in the φ4 theory (15.5 and 14.2). Therefore.1) which gives the one loop correction to the two point function.15. the expansion is unaffected by how one divides the total action into a free part and an interaction part. This is quite important. We are usually accustomed to the idea of perturbation in powers of the coupling constant in the theory. On the other hand. the scalar field picks up a vacuum expectation value which depends on the coupling constant of the theory. 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. when we study theories where there is spontaneous breakdown of a symmetry. the entire perturbation expansion has to be rearranged which is nontrivial. as we have seen (see. In such a case.6). sections 7. 15.3 Cut-off regularization 623 As a result. Loop expansion is a different perturbative expansion. In addition the small value of makes it a legitimate expansion (perturbation) parameter.3 Cut-off regularization To get an idea of how divergences arise in a quantum field theory at the loop level. expanding in powers of in a quantum field theory also corresponds to expanding in the number of loops as described in (15. The reason why such an expansion is advantageous can be seen from the fact that since this parameter multiplies the whole action.

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

10) (and m.577215665. 2 ˜ 32π M Λ2 (15. Λ are scalar parameters).15. We note that we can write (15.12) . in this case this is also the result in the Minkowski space.3 Cut-off regularization 625 iλ I = lim − Λ→∞ 32π 2 = lim − Λ2 0 dy − M 2 0 Λ2 dy y + M2 Λ2 0 Λ→∞ iλ Λ2 − M 2 ln y + M 2 32π 2 iλ Λ2 − M 2 ln 32π 2 Λ2 + M 2 M2 = = Λ→∞ lim − Λ→∞ lim − iλ Λ2 Λ2 − M 2 ln 2 + O 32π 2 M 1 Λ2 .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( ) . (15.11) ˜ Λ→∞ n γ = lim n→∞ k=1 1 − ln n k = 0. 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. (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 + . we also note here that the diagram in (15.352.47)) Ek = |k|2 + M 2 .10) with the identification ˜ Λ2 = Λ2 − M 2 γ. Gradshteyn and Ryzhik 3. This result coincides with (15.8) can be evaluated without rotating the momentum into Euclidean space as follows iλ 2 i d4 k (2π)4 k2 − M 2 + iǫ d3 kdk0 I = − = = λ 2(2π)4 λ (−2πi) 2(2π)4 1 (k0 − (Ek − iǫ))(k0 + Ek − iǫ) 1 iλ =− 2Ek 32π 3 d3 k 1 .14) d3 k where we have identified (see. Ek (15.214. the integration over the angles can be carried out.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. Furthermore.14) does not depend on the angular variables. Since the integrand in (15. (15. (15. the integral takes the form iλ (4π) 32π 3 ∞ 0 ¯ Λ I = − dy y2 y2 + M 2 y2 + M 2 − M2 y2 + M 2 y2 + M 2 ¯ Λ 0 iλ = lim − 2 ¯ 8π Λ→∞ = ¯ Λ→∞ dy 0 √ ¯ ¯ iλ ¯ ¯ 2 Λ + Λ2 + M 2 2 − M 2 ln = lim − Λ Λ +M ¯ 16π 2 M Λ→∞ 2 ¯ 1 4Λ iλ ¯ + O( ¯ 2 ) .1).4 and 8.13) Furthermore. for example. defining y = |k|. (5. for example. which gives (4π) in three dimensions.

10) has been transformed into an infrared divergence in the Schwinger parameter τ . 2 This derivation makes it clear that the evaluation of the amplitude is much easier in the Euclidean space. 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 .17) is well behaved for large values of τ . Thus. Gradshteyn and Ryzhik. 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. but is divergent at the lower limit τ = 0.8) can also be evaluated in an alternative manner which is sometimes useful and so let us discuss the alternate method as well.17) dτ −τ M 2 . We note that the integrand in (15.10) if we identify 1 ¯ (15. the ultraviolet divergence in (15.3 Cut-off regularization 627 Here we have used the standard table of integrals in the intermediate step (see.271.2 and 2.15. This result coincides with (15. 2. for example. The integral in (15.16) Λ2 = Λ2 − M 2 (1 − ln 2) .271. e τ2 Here Γ(2) denotes the gamma function and the parameter τ is known as the Schwinger parameter (or the proper time parameter).4).

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. Equation (15. we note that the result in (15. This diagram is also divergent (in four dimensions) and with the cut-off regularization leads to (we assume p1 = p2 = p3 = p4 = 0 for simplicity so that each graph contributes the same amount and also because we are interested in looking at only the divergence structure of the graph) . 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. We note that since the cutoff Λ has to be actually taken to infinity in the final result. for example. However. (15. the one loop diagram for the vertex correction shown in Fig.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. but quantum field theories can be defined in any number of space-time dimensions). Gradshteyn and Ryzhik 3.628 = = = = 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) and we see that both ways of evaluating the integral lead to the same result.18) can be compared with (15. For simplicity we would consider all the incoming momenta to vanish.18) can also be obtained by simply cut1 ting off the τ integral at the lower limit Λ2 (without the regularizing exponential factor).3. Without going into details.471. For example.9) to evaluate the integral over τ . namely.10) (or (15. we note from (15. Let us next look at another one loop graph.

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 . First we have factored out an overall momentum conserving delta function along .3 Cut-off regularization 629 p3 p2 + p1 p4 p1 p3 p2 p4 p4 p3 + p1 p2 Figure 15.3: One loop correction to the four point vertex function.15.19) Λ→∞ lim There are several things to note from this derivation. (15.

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

15.3. In the previous examples.1 Calculation in the Yukawa theory. in general. namely. (15. p−m / p − m2 = −(2π)4 igδ4 (p1 + p2 + p3 ). the theory with a Yukawa coupling described by the Lagrangian density 1 λ M2 2 / φ − gψψφ − φ4 . the lack of dependence on the external momenta is simply due to our special choice of the external momenta to be all vanishing.21) .19).20) L = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 4! This theory describes the interaction between a fermion and a scalar field and the Feynman rules for this theory are given by i . The second part of renormalization consists of removing the cutoff dependence (or regularization dependence) of the theory which we will take up in the next chapter. But in this process the amplitude (and the theory) becomes cutoff dependent.15. (15.3 Cut-off regularization 631 the integrals. particularly in the second example (15. − M2 p = iGF (p) = p2 p p3 p2 p1 p4 p3 = iSF (p) = i(p + m) / i = 2 . Let us note that any Feyn- man diagram (amplitude). But let us examine the dependence of diagrams on external momenta (and this will be important in studying renormalization) in another theory. p1 p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). depends analytically on the external momenta.

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

the integral in (15. let us next calculate the two simplest nontrivial graphs involving loops in this theory.23).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) .24) which we have already used in (13.23) to include more denominators is straightforward and has the form n i=1 1 = Ai 1 n dxi 0 i=1 δ(1 − x1 − · · · − xn ) . (15.3 Cut-off regularization 633 useful.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 . Generalization of (15. Equipped with the Feynman combination formula. / / (15. ( n xi Ai )n i=1 (15. The fermion self-energy at one loop has the form k p k+p p = (−ig)2 i i d4 k 4 k 2 − M 2 (k + p) − m (2π) / / = g2 = g2 = − 1 (k + p) + m / / d4 k 4 k 2 − M 2 (k + p)2 − m2 (2π) ig2 (2π)4 id4 kE 1 −(kE + pE ) + m / / 4 −k 2 − M 2 (2π) − (kE + pE )2 − m2 E d4 kE 2 (kE + M 2 ) (kE + pE )2 + m2 (kE + pE ) − m / / .121).26) Using the Feynman combination formula (15.

28) In (15.27) where we have defined Q2 = x(1 − x)p2 + xm2 + (1 − x)M 2 . Furthermore.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 . E (15.27) we have shifted the variable of integration kE → kE − xpE in the intermediate step after which the term kE in the numerator / vanishes because of anti-symmetry. 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) / . (15.

4 (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Here the overall negative sign reflects the fact that we are evaluating a graph with a fermion loop. /B the self energy graph in (15.29) = f (p. 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). g). 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. / Tr A / = 4A · B.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. Using the trace identities Tr ½ = 4.19) in the intermediate step and have rotated back to Minkowski space in the final step. Furthermore. m. where it is understood that the limit Λ → ∞ is to be taken.31) Tr A = 0. (15.30) p = −Tr (−ig)2 = −g2 Tr (k + m)((k + p) + m) / / / d4 k . M. We also note here that we have used the integral from (15.30) takes the form .15. Λ.

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 .

In the examples worked out above it is clear that one can Taylor expand the amplitudes around zero momentum of the external lines.32) where we have identified Q2 = x(1 − x)p2 + m2 while Q = −x(1 − E x)p2 + m2 represents the function rotated to Minkowski space. Similarly.33) +finite terms as Λ → ∞ . In fact. a Feynman amplitude with n external lines is an analytic function of (n − 1) external momentum variables.29).32) we can write . These two simple calculations show that Feynman amplitudes are functions of external momenta. for example.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. 2 (15. Λ). for the fermion self-energy in (15. This is because the overall momentum conservation eliminates one of the momentum variables. g. for the scalar self-energy in (15. m. Thus.15. 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.

To avoid this problem. 15. Therefore. Similarly. leads to a lack of manifest rotational invariance (although in the continuum limit this symmetry is recovered). while quite useful.34) +finite terms as Λ → ∞ . then expanding around zero external momenta would be disastrous. 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). it is clear that such a regularization violates manifest Poincar´ invariance of the e quantum theory. Let us note here that if the theory is massless.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. Furthermore. This is because of the infrared divergences of the theory which can be seen in the above examples by setting m = 0. in massless theories we Taylor expand the amplitudes around a nonzero but finite value of the external momentum.4 Pauli-Villars regularization Although regularizing a Feynman diagram by cutting off contributions from large values of momentum seems natural. a cut-off can lead to violation of gauge invariance in gauge theories by giving a mass to the gauge boson. a lattice regularization which regularizes a theory by defining it on a discrete space-time lattice (we will not go into a detailed discussion of lattice regularization). it is useful to find .

The Pauli-Villars regularization provides such a regularization scheme. we can now calculate various 1PI amplitudes in QED.125)) 1 1 /ψ L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 . let us consider QED (quantum electrodynamics) in the Feynman gauge (see (9.(15.4 Pauli-Villars regularization 639 a covariant regularization scheme which also respects gauge invariance. For example. 4 2 which leads to the Feynman rules µ p (15.80) and (9. p2 p q p = iSF (p) = i(p + m) / .35) ν = iGF . To discuss the Pauli-Villars regularization scheme. 2 − m2 p r µ = −(2π)4 ieγ µ δ4 (p + q + r).15.µν (p) = − iηµν .36) With these Feynman rules. 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 .

We have also used gamma matrix identities in (2. In the .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.640 = − = − = − × e2 (2π)4 e2 (2π)4 ie2 (2π)4 d4 k 15 Regularization of Feynman diagrams −2(k + p) + 4m / / 2 − m2 ) (k 2 − µ2 ) ((k + p) 2(kE + pE ) + 4m / / 2 + m2 ) (k 2 + µ2 ) ((kE + pE ) E id4 kE dx d4 kE 2(kE + pE ) + 4m / / (kE + xpE ) + x(1 − x)p2 + xm2 + (1 − x)µ2 E ie2 (2π)4 ie2 16π 2 ie2 8π 2 ∞ 0 2 2 = − = − = − × = − dxd4 kE 2 dxdkE 2kE + 2(1 − x)pE + 4m / / 2 (kE + Q2 (m. µ) = x(1 − x)p2 + xm2 + (1 − x)µ2 .111) and (2. divergent (logarithmically by power counting) as is the case in the earlier calculation of the fermion self-energy with the Yukawa coupling (see (15. (The Dirac gamma matrices have also been rotated to Euclidean space as discussed in (15.) The momentum integral is.26). of course. µ) − 2 2 kE + Q2 (m.27)). 2 kE + Q2 (m. µ))2 dx ((1 − x)pE + 2m) / 2 dkE 1 Q2 (m.37) ie2 8π 2 dx((1 − x)pE + 2m) / where we have defined Q2 (m.112) as well as (15. µ))2 2 / kE (2(1 − x)pE + 4m) 2 (kE + Q2 (m. µ))2 ∞ 0 2 dkE −1 . µ) (15. µ) (kE + Q2 (m. E (15.9) for the angular integral.

Λ) (15. Namely. without doing any further calculation.37) except for a sign and µ → Λ.37) coming from the diagram where a heavy photon is being exchanged.) 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.39) where we assume that Λ ≫ m. we can write the additional contribution to the fermion self-energy as ie2 8π 2 ∞ 0 2 dkE −1 . 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. 2 kE + Q2 (m.39) is defined with the gauge field combination Aµ + A′ . Therefore. decouple.40) I′ = dx ((1 − x)pE + 2m) / so that the effective regularized fermion self-energy can be written as ie2 8π 2 Q2 (m. Q2 (m. The idea is to take Λ → ∞ at the end and in that limit the heavy fields do not propagate and. these additional interacting fields give an additional contribution to the fermion self-energy (15. therefore. let us add to the Lagrangian density of QED (15. A′ are fictitious heavy µ fields introduced to regularize diagrams. (The covariant derivative in (15. Λ) . µ and ψ ′ .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.41) I (reg) = I + I ′ = − dx ((1 − x)pE + 2m) ln / . 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) (15. However. / ′ L′ = (15.

(15. µ) 2 2 I (reg) dx ((1 − x)p − 2m) ln / .43) . We can also calculate the one loop photon self-energy in QED in a similar manner.36) lead to k p k+p i(k + m) i(k + p + m) / / / d4 k γ γ 4 µ k 2 − m2 ν (k + p)2 − m2 (2π) p = iΠµν (p) 2 = −(−ie)2 Tr =− =− kµ (k + p)ν + kν (k + p)µ − ηµν k · (k + p) + m2 ηµν 4e2 d4 k (2π)4 (k2 − m2 )((k + p)2 − m2 ) 4e2 (2π)4 d4 k dx × =− kµ (k + p)ν + kν (k + p)µ − ηµν (k · (k + p) − m2 ) ((k + xp)2 + x(1 − x)p2 − m2 )2 d4 k dx 4e2 (2π)4 × =− 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − (x(1 − x)p2 + m2 )) (k2 + Q (m))2 2 4e2 (2π)4 dx d4 k 2 × 2x(1 − x)(ηµν p2 − pµ pν ) + 2kµ kν − ηµν (k2 + Q (m)) (k2 + Q (m))2 2 .642 15 Regularization of Feynman diagrams which is finite for any given value of Λ. µ) = −x(1 − x)p2 + xm2 + (1 − x)µ2 . Λ) Q (m. The final result can now be rotated back to Minkowski space and has the form ie2 = 2 8π Q (m. The Feynman rules (15. (15.42) where Q (m.

4 Pauli-Villars regularization 643 where we have used the gamma matrix identities in (2. we can rotate (15.44) The overall negative sign in (15. Π → (i)ΠE ) and we have δ00 .45) individually.45) where Q2 (m) = x(1 − x)p2 + m2 . + y + Q2 (m) (y + Q2 (m))2 = d4 kE E 1 (−δµν ) + Q2 (m) . (k2 . 2 (kE + Q2 (m))2 (15.15. To evaluate the momentum integral. Π00 → (i) 00 0i 0i iΠE = − µν × 4ie2 (2π)4 d4 kE dx 2 2x(1 − x)(δµν p2 − pµE pν E ) + 2kµE kν E − δµν (kE + Q2 (m)) E . p2 ) → −(kE . η00 → E 2 ΠE . p0E ). 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 . p0 ) → i(k0E .43) to 2 Euclidean space ((k0 .43) reflects the fact that we are evaluating a fermion loop. 2 (15. p2 ). We can now integrate each of the E terms in (15.115) and (2.116) and have defined Q (m) = x(1 − x)p2 − m2 .

we obtain for the self-energy of the photon (in Euclidean space).46) =− π2 δµν 2 dy 1− (y + Q2 (m))2 Here we have used (15.45). However. recalling that the photon has coupling to fictitious heavy fermions. is quadratically divergent.47) 1 + δµν − 2 dy + Q2 (m) .9) in the intermediate steps. This integral. y + Q2 (m) d4 kE 2 (kE + Q2 (m)) 2 − k2 δµν 2 E + Q (m) (Q2 (m))2 . Putting this back into (15. . (15. ie2 4π 2 ∞ 0 iΠE = − µν dx dy 2x(1 − x) δµν p2 − pµE pν E y E (y + Q2 (m))2 +δµν = − × ie2 4π 2 (Q2 (m))2 1 − + 2 2(y + Q2 (m))2 dx 0 ∞ dy 2x(1 − x) δµν p2 − pµE pν E E Q2 (m) 1 − y + Q2 (m) (y + Q2 (m))2 1 (Q2 (m))2 − + 2 2(y + Q2 (m))2 ∞ 0 +δµν = − ie2 dx 2x(1 − x)(δµν p2 − pµE pν E ) E 4π 2 ∞ 0 dy −1 y + Q2 (m) (15. as we see.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) .

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

50) then. c1 Λ2 + c2 Λ2 = m2 . Pauli-Villars regularization involves introducing.(15. 1 2 (15.49) E 1 2 It is clear.(15.51) which represents a gauge invariant photon self-energy (it is manifestly transverse).646 × ∞ 0 15 Regularization of Feynman diagrams dy 1 c1 c2 − − 2 (m) 2 (Λ ) y+Q y+Q y + Q2 (Λ2 ) 1 −(1 − c1 − c2 ) 1 − δµν (1 − c1 − c2 ) 2 ∞ 0 dy −(1 − c1 − c2 )x(1 − x)p2 − (m2 − c1 Λ2 − c2 Λ2 ) .52) holds for any covariant gauge with an arbitrary value of the gauge fixing parameter ξ. · · · . the quadratic divergences as well as the logarithmic divergences in (15.43) does not involve the photon propagator.52) We note here that even though we are working in the Feynman gauge. therefore. the result (15. in a gauge invariant manner. n depending on the nature of the divergence in the diagram such that the regularized amplitude . 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) . a set of heavy fields with masses Λi and √ charges ( ci e) with i = 1. (15. In general.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) . since the diagram for the photon self-energy in (15. 2. that if the parameters satisfy c1 + c2 = 1.

in less than four dimensions it is finite. m) − ci Ii (p.53) with the conditions on the parameters given by ci = 1. For example. i ··· . it is by construction a gauge invariant regularization scheme. (k2 − m2 ) ((k + p)2 − M 2 ) (15.5 Dimensional regularization 647 n I (reg) = I(p. Λi ).55) is logarithmically divergent in four dimensions. 15. (15. It is worth pointing out here that the Pauli-Villars regularization is sufficient to regularize all the Feynman amplitudes in an Abelian gauge theory in a gauge invariant way. In non-Abelian theories. i i ci Λ2 = m2 .54) will be finite and gauge invariant. (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. however. 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). there are graphs of the form shown in Fig. the integral d4 k .4 which cannot be regularized by the Pauli-Villars method. However. In fact.15.5 Dimensional regularization From the analysis of the divergence structure of amplitudes in the last two sections it is clear that the divergences are functions of the dimension of space time. This procedure is known as dimensional regularization. it is quite useful in studying non-Abelian gauge theories for which the Pauli-Villars regularization does not work. 15. since gauge invariance is independent of the number of space time dimensions. Furthermore.) We should mention here that dimensional regularization has its own problems which we will discuss at the end of . i=1 (15.

648 15 Regularization of Feynman diagrams this section. The Feynman rules for the theory (15. Figure 15.58) are given by . 2 [M ] = 1.56) carries a dimension and introducing an arbitrary mass scale µ we can then write the Lagrangian density (15.58) so that the coupling constant λ is rendered dimensionless. [φ] = [λ] = 4 − n. the coupling constant for the quartic interaction in (15. 2 2 4! L= (15. if we are looking at the φ4 theory described by the Lagrangian density M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ .56) we can carry out the (canonical) dimensional analysis in n dimensions to determine n−2 . In dimensional regularization we analytically continue the theory to n dimensions. 2 2 4! L= (15.4: One of the one loop graphs contributing to the gluon self-energy.57) Therefore. For example.56) for the theory in n dimensions as M 2 2 λµ4−n 4 1 ∂µ φ∂ µ φ − φ − φ . (15. but it is worth pointing out here that manipulations with dimensional regularization are extremely simple which is why it is quite useful.

5 Dimensional regularization 649 p = p2 i . to calculate amplitudes in n dimensions. since each vector index takes n values. shifted the variable of integration and have defined Q2 = M 2 − p2 . therefore. E To evaluate this integral. it follows that in n dimensions we have µ ηµ = n. For example.61) I = = = = dn kE dn kE 1 . let us note that the integrand is spherically symmetric (independent of angular coordinates) and. (15. we can separate out the angular part of the integral . 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 (15.15.60) Furthermore. 2 (kE + Q2 )α where we have rotated to Euclidean space.59) We need to be careful about manipulating tensors in n dimensions. p1 p2 (15. − M2 p4 p3 = −(2π)4 iλµ4−n δn (p1 + p2 + p3 + p4 ).

(15. . . This determines the value of the angular integral to be dΩ = 2π 2 (2π) 2 = . Let us consider the basic n dimensional Gaussian integral whose value is given by dn kE e− 2 kE 2 n = (2π) 2 . n −1 n Γ n 22 Γ 2 2 n n (15.63) 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. (15.650 15 Regularization of Feynman diagrams n−1 dn kE = kE dkE dΩ. where. dθn−1 .62) The integral over the angles dΩ can be evaluated from the simple Gaussian integral as follows.64) n = 2 2 −1 Γ n 2 k2 dΩ.63) On the other hand. in n dimensions. we can evaluate the integral in (15. We see that . n ≥ 2.65) Let us compare this with known results in lower dimensions that we are familiar with. E where we have defined y = 2 in the intermediate step and have used the definition of the gamma function. we can write dΩ = dθ1 sinn−2 θ1 dθ2 sinn−3 θ2 .

(15. (15.5 Dimensional regularization 651 n=2: n=3: n=4: dΩ = dΩ = dΩ = dθ = 2π.69) the integral (15.68) Therefore. dθ1 sin2 θ1 dθ2 sin θ2 dθ3 = 2π 2 . for the basic integral (15.67) 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.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 . 2 n . 2 p= q =α−p=α− (15. Thus. we see that with the identification n .66) which agree with what we already know from explicitly doing these integrals.15. dθ1 sin θ1 dθ2 = 4π. q) = dt t2p−1 1 + t2 . (15.

For example. using (15.72) Similarly we can obtain . any other formula can be obtained from (15.70) Rotating back to Minkowski space.652 15 Regularization of Feynman diagrams I = = Γ i(−1)α π 2 1 n n Γ n 2 )α− 2 (2π) 2 (Q iπ 2 (−1)α Γ α − n 2 n .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 . 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. we see that this gives us our basic integral as dn k 1 n (k 2 + 2k · p − M 2 )α (2π) n I = = iπ 2 (−1)α Γ α − n 2 n . (2π)n Γ (α) (Q2 )α− 2 n n n 2 Γ α− Γ(α) n 2 (15. (2π)n Γ(α) (p2 + M 2 )α− 2 (15. in fact.

take the limit n → 4 which translates to ǫ → 0.74) Let us next set n = 4 − ǫ. First of all. the one loop scalar self-energy takes the form (see also (15.5 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 . let us now calculate various amplitudes in the φ4 theory (15. 2 (2π)n (M 2 )1− 2 n n = − (15.8)) k p iλµ4−n 2 p i dn k n k2 − M 2 (2π) = − = λµ4−n iπ 2 (−1) Γ 1 − n 2 n 2 (2π)n Γ(1) (M 2 )1− 2 iλµ4−n π 2 Γ 1 − n 2 n . we are analytically continuing away from four dimensions (to a lower dimension where the integral is well defined). With these basic integration formulae..e. At the end of our calculations we should. i.73) n kµ kν dn k n (k 2 + 2k · p − M 2 )α (2π) and so on.15. With this. the amplitude (15. (15.58). of course.74) becomes .

77). ǫ (15.76) as well as (15.77) Using (15.75) takes the form . we have n 2 n Γ 2− 2 Γ 3− ǫ ǫ ≃ 1 − γ. 16π 2 2 (15. 2 (2π)n (M 2 )−1+ 2 n (15. In the limit ǫ → 0.654 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 .12). the scalar self-energy (15.76) where γ denotes the Euler’s constant defined in (15. we have π2 (2π)n n = ≃ 1 1 n = ǫ (4π) 2 (4π)2− 2 ǫ 1 1 + ln 4π . ≃ 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). Similarly. 2 2 2 ǫ ǫ = Γ = Γ 1+ 2 ǫ 2 ǫ 2 2 1 − γ) = − γ.75) Let us next work out some of the identities involving the gamma functions that will be useful to us.

In this case.78) We note here that for M 2 = 0.3 and. Similarly.15. − + ln ǫ 4πµ2 (15.19). we can calculate the one loop correction to the vertex function shown in Fig.122). this graph would be regularized to zero in dimensional regularization which is the type of argument used in (13. 15.5 Dimensional regularization 655 k p p n iλµǫ π 2 ǫ = − M 2 M −ǫ Γ −1 + 2 (2π)n 2 iλ π 2 ǫ = − M 2 Γ −1 + n 2 (2π) 2 ≃ − = − ≃ − = − iλ M 2 ǫ 1 + ln 4π 2 2 16π 2 iλ M 2 2 16π 2 iλM 2 32π 2 iλM 2 32π 2 n M2 µ2 ǫ −2 2 − + (γ − 1) ǫ 1− 1− ǫ M2 ln 2 2 µ 2 − + (γ − 1) − ln 4π ǫ ǫ M2 ln 2 2 µ M2 2 − + ln 2 + (γ − 1) − ln 4π ǫ µ M2 2 + (γ − 1) . we will set all the external momenta to vanish as we had done earlier in (15. for simplicity. the amplitude in n dimensions takes the form = = 3 (−iλµǫ )2 2 3λ2 µ2ǫ 2 dn k (2π)n dn k 1 n 2 − M 2 )2 (2π) (k i 2 − M2 k 2 .

we let e → e µ2 . 2 2 2 ǫ . ǫ (15.82) .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 . Let us next calculate one loop ampli- tudes in QED described by the Lagrangian density in the Feynman gauge (see (15. the amplitudes diverge.1 Calculations in QED. 15.79) where we have used (15. to make the coupling constant dimensionless.76) and (15. /ψ 4 2 (15.5. 2 2 n−1 3 ǫ ψ = = − .35)) 1 1 L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 .81) As a result. However.80) In n dimensions. as we take the limit ǫ → 0 (to go to four dimensions). We note that these amplitudes are well behaved (regularized) for any finite value of ǫ. ǫ 4πµ2 (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− . 2 (15.77).

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 .85) = −ieµ 2 2 = −e2 µǫ / / dn k γ µ ((k + p) + m) γµ .83) As a result. p2 p = iSF (p) = i(p + m) / . we can calculate the fermion selfenergy at one loop which gives (see also (15. n k 2 ((k + p)2 − m2 ) (2π) .84) ǫ With these Feynman rules.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. the Feynman rules of the theory in n = 4 − ǫ dimensions are given by µ p ν = iGF .15. (15. ǫ (15. 2 − m2 p r µ q p = −(2π)4 ieµ 2 γ µ δ4 (p + q + r).

658 15 Regularization of Feynman diagrams We note here that if we use the algebra of the gamma matrices (1. the one loop fermion self energy in (15. γµ kγ µ = (2 − n)k. / / (15.60) we obtain γµ γ µ = n.86) Using this.85) takes the form = −e2 µǫ = −e2 µǫ = −e2 µǫ = −e2 µǫ = −e2 µǫ ≃ ie2 8π 2 dn k (2 − n)(k + p) + nm / / n k 2 ((k + p)2 − m2 ) (2π) dx dx dx (15.88) 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) .87) (2 − n)(k + p) + nm / / dn k n 2 + x(1 − x)p2 − xm2 )2 (2π) ((k + xp) / / dn k (2 − n)(k + (1 − x)p) + nm 2 − Q2 ) 2 (2π)n (k dn k (2 − n)(1 − x)p + nm / 2 − Q2 )2 (2π)n (k (15. ǫ 4πµ2 (15.79) in n dimensions as well as (15.

91) the photon self-energy (15.90) / / / dn k Tr γµ (k + m)γν (k + p + m) . /B / Tr γµ γν γλ γρ = n (ηµν ηλρ − ηµλ ηνρ + ηµρ ηνλ ) .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 (15.115) and (2. / Tr A / = nA · B. Tr A = 0.116)).15. n (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Using the n-dimensional identities (see (2.90) 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 . /B Tr A /C = 0. This can be compared with (15.5 Dimensional regularization 659 where we have defined Q2 = −x(1−x)p2 +xm2 and have used (15.42). (15. Tr ½ = n.76) as well as (15.

93) × ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν where we have used (15. The integral in (15. Let us look at the terms without any k in the numerator and this takes the form (we now put back the factor (−e2 µǫ )) = −ne2 µǫ = −ne2 µǫ × ≃− dn k ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν (2π)n (k2 − Q2 )2 dx dx ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν (−1)2 Γ 2 − n i 2 n n (4π) 2 Γ(2) (Q2 )2− 2 dx ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν ǫ 4 1+ ǫ ln 4π 2 2 −γ ǫ 1− ǫ Q2 ln 2 µ2 ie2 4π 2 × 1− ≃− ie2 4π 2 dx ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν 1− ǫ Q2 ln 2 µ2 × ≃− 1 2 − γ − + ln 4π ǫ 2 dx ie2 4π 2 Q2 1 2 − ln −γ− ǫ 4πµ2 2 . (15. Let us next look at the terms with kµ kν in the numerator separately which leads to . (k2 − Q2 )2 (15.77).92) where we have defined Q2 = −x(1 − x)p2 + m2 .76) as well as (15.71)(15.660 × =n × 15 Regularization of Feynman diagrams 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − x(1 − x)p2 − m2 ) (k2 − Q2 )2 dx dn k (2π)n ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν + 2kµ kν − ηµν k2 .92) contains three different kinds of terms which can be integrated using the standard formulae in (15.73).

(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 ) . 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.

× − + 2 ln ǫ 4πµ2 Adding the contributions from (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. For simplicity we will put the momentum of the external photon equal to zero.93)-(15. Let us next calculate the one loop amplitude corresponding to vertex correction (charge renormalization).96) =− dxx(1 − x) This shows that the photon self energy graph is completely transverse which is consistent with gauge invariance (see also (15. the gauge fixing Lagrangian density does not receive any quantum correction which can also be seen from the BRST identities for QED. the photon self energy in (15. 2 ǫ 4πµ 2 (15. The photon does not acquire a mass term and we note here that even though we have chosen to work with the Feynman gauge. since the photon self-energy does not involve the photon propagator. Since there is no longitudinal term in the photon self-energy. this result holds in any covariant gauge with the gauge fixing term ξ arbitrary.92) takes the final form ie2 (ηµν p2 − pµ pν ) 2π 2 2 Q2 1 − ln −γ− .52)).95) Q2 4 + 2γ . 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 .95).

100) Let us next use (15.86) we have (15. / / Therefore.97) = −e3 µ 2 3ǫ / / / / dn k γ λ (k + p + m)γ µ (k + p + m)γλ .23) as well as the identity .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 / / .113)) γ λ A µ A λ = (2 − n)(2Aµ A − γ µ A2 ).98) γ λ (k + p + m)γ µ (k + p + m)γλ / / / / = γ λ (k + p)γ µ (k + p)γλ + mγ λ (γ µ (k + p) + (k + p)γ µ )γλ / / / / / / / / +m2 γ λ γ µ γλ = (2 − n)(2(k + p)(k + p)µ − (k + p)2 γ µ ) + 2nm(k + p)µ / / +(2 − n)m2 γ µ = −(2 − n)γ µ ((k + p)2 − m2 ) +2((2 − n)(k + p) + nm)(k + p)µ .5 Dimensional regularization 663 (15.97) as dn k (2π)n k2 ((k + p)2 − m2 )2 (15. we can write the amplitude (15. n (2π) k2 ((k + p)2 − m2 )2 Using the n-dimensional identities (see also (2.15. 2 ((k + p)2 − m2 )2 k (15. /γ /γ / as well as (15.

((1 − x)A + xB)3 1 ((1 − x)A + xB)2 (15. (15.100) which leads to 1 + p)2 − m2 ) 1 (k + xp)2 − Q2 2x (k + xp)2 − Q2 k2 ((k = = dx dx 2.102) where we have defined Q2 = −x(1 − x)p2 + xm2 . we obtain for the vertex correction p q=0 µ k+p p = −e3 µ 2 × 3ǫ k+p k dx dn k −(2 − n)γ µ (2π)n (k2 − Q2 )2 / / dn k 4x (2 − n)kkµ + (1 − x) (2 − n)(1 − x)p + nm pµ 3 (2π)n (k2 − Q2 ) 3ǫ = −e3 µ 2 dx − (2 − n)γ µ i (−1)2 Γ 2 − n 2 n n (4π) 2 Γ(2) (Q2 )2− 2 . 1 k2 ((k + p)2 − m2 )2 3.664 15 Regularization of Feynman diagrams 1 AB 2 = − = ∂ 1 ∂ =− ∂B AB ∂B dx dx 2x .100) and shifting the variable of integration. Using this in (15.101) to combine the denominators in (15.

5 Dimensional regularization 665 − 1 2 Γ 2− n 2 n 2 )2− 2 (Q +4x(2 − n)γ µ (−1)3 i n (4π) 2 Γ(3) +4x(1 − x) (2 − n)(1 − x)p + nm pµ / = −ie3 µ 2 − = 3 i (−1)3 Γ 3 − n 2 n n (4π) 2 Γ(3) (Q2 )3− 2 dx Γ 2− (4π) n 2 n 2 − (2 − n)(1 − x)γ µ n (Q2 )2− 2 n (4 − n)x(1 − x) (2 − n)(1 − x)p + nm pµ / (Q2 )3− 2 dx (1 − x)γ µ 1 + 2 ln Q2 µ2 2 −γ 2 (15.105) .88) and (15. −(1 − x)p2 + m2 It is worth noting from this derivation that at every step if we compare (15. (15.104) respectively.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 (15.87).89) with (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. we see that the fermion self-energy and the vertex function are related as (see also (9.103) and (15.100).15. p k+p k+p p q=0 (15.104) 2 (1 − x)p − 2m pµ / .

In addition to the regularization methods that we have discussed in this chapter.6 where we discuss chiral anomaly. we know that i 4! γ5 = iγ 0 γ 1 γ 2 γ 3 = − µνλρ γ µ ν λ ρ γ γ γ . the point splitting method. 1439 (1948). It is not naively applicable if the calculation involves quantities that typically exist only in four dimensions. it has its own drawbacks. for example. We simply note here that we select a particular regularization scheme depending on the theory (or the problem) that we are analyzing. the higher derivative method.) The power of dimensional regularization is quite obvious from these calculations. (15. J. It is gauge invariant and extremely simple to manipulate with.106) is defined only in four dimensions (both γ5 and the Levi-Civita tensor µνλρ are manifestly four dimensional quantities). the ζ-function regularization etc. For example. Analytic continuation of these to other dimensions is nontrivial. there exist several other important regularization schemes. Hence if the theory or the physical quantity of interest involves such objects.666 15 Regularization of Feynman diagrams which we recognize as the Ward-Takahashi identity. (15. However.6 References 1. 15.107) There are several ways to address this issue with the γ5 matrix. . naive dimensional regularization leads to incorrect answers. (Note that the divergent parts as well as the finite parts satisfy the identity which demonstrates that dimensional regularization preserves gauge invariance. Physical Review 74. We will discuss this further in section 16. we will not go into the details of these other methods here. However. The lattice regularization is also quite useful. Schwinger. 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.

664 (1951). G. N. Physical Review 82. Veltman. CERN preprint (1973). Schwinger.15. . Feynman. 769 (1949). J. Introduction to the theory of Quantized Fields. Bogoliubov and D. Giambiaggi. Nuclear Physics B44. 7. W. Diagrammar. 6. ’t Hooft and M. Reviews of Modern Physics 21. 8. G. 3. 5. Pauli and F. Nauka. Shirkov. ’t Hooft and M. 20 (1972). Moscow (1984). V. G. 189 (1972). J. 4. Nuovo Cimento 12B. 434 (1949). Physical Review 76.6 References 667 2. Villars. Veltman. N. R. C. Bollini and J.

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.Chapter 16 Renormalization theory 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.1 Superficial degree of divergence In the last chapter. As a result. In our units.1) is dimensionless. we have seen that loop diagrams in various theories become divergent and need to be regularized.2) . To define this let us start with the Lagrangian density for a given theory L = L0 + Li . the action for the theory (16. we have also seen that since Feynman amplitudes are analytic functions of external momenta. 669 (16. The next step in understanding the process of renormalization is to devise a method for determining which Feynman diagrams in a theory will be divergent as well as the nature of the divergence (without actually evaluating the integrals) and this is achieved through the notion of the superficial degree of divergence of a graph. In units of = c = 1 which we have been using. in n dimensions it should be n). it is easy to check that [M ] = [L]−1 = 1. the Lagrangian density must have canonical dimension 4 in these units (since we are in four space time dimensions. Furthermore. (16.

bosons and derivatives at an interaction vertex following from the interaction Lagrangian density Li . (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.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 .6) Namely.7) leads to [ψ] = [ψ] = (16. and L0 we have [φ] = 1. 2 (16.) Similarly the free fermion Lagrangian density (ψ) L0 = iψ∂ ψ − mψψ. (φ) (16. the canonical dimension of a scalar field (in our units and in four dimensions) is 1. 2 2 L0 = that since (φ) (16.5) = 4. (16. for example. (The mass term also has the correct dimension with this assignment.8) In fact. / 3 . Thus. for the Yukawa interaction LY = gψψφ.4) (16.3) [∂µ ] = [L]−1 = 1.9) . Let us further introduce the notations fi . bi and di to denote respectively the number of fermions.

This is defined to be the difference between the number of momenta in the numerator arising from the loop integrations as well as derivative couplings and the number of momenta in the denominator arising from the propagators. b = 1.11) f = 0. Similarly. (16.1 Superficial degree of divergence 671 we have f = 2. With these notations we are now ready to introduce the notion of the superficial degree of divergence of a Feynman graph. 4! (16. we can determine easily the superficial degree of divergence of the following Feynman graphs as . κψσ µν ψFµν . d = 0.13) f = 2.12) LI = hψγµ ψ∂ µ φ. d = 1. b = 4. b = 1. we have or.16. the φ4 interaction (16. On the other hand.10) LI = leads to λ 4 φ . d = 0. for an interaction of the form (16.14) and so on. For example. (16.

Similarly. D = 8 − 5 × 2 = −2. D = 4 − 2 − 1 = 1. we say that the diagram is superficially logarithmically divergent. D = 4 − 2 × 2 = 0. D = 4 − 2 × 1 = 2. D = 8 − 3 × 2 = 2. (16.672 16 Renormalization theory D = 4 − 2 = 2.15) If D = 0 for a diagram. diagrams with D = 1 or 2 are known . D = 4 − 2 − 2 × 1 = 0.

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

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-

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) . Note that the BPHZ method subtracts out the divergent parts in the integrand itself so that the renormalized integral is finite. . then D(γ) = 0 and tγ simply evaluates the graph γ at zero external momentum.2 A brief history of renormalization 689 Figure 16. F ∈Φγ ∈F (16. if γ is a Γ(4) .16. With the notion of the Taylor operator. Then the assertion is that the renormalized Feynman graph is convergent. 0) + . mentum variables about zero four-momentum out to order D(γ) which denotes the superficial degree of divergence of γ.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. For example. We would see in detail how this works in section 16. p=0 (16. then (D(γ) = 2 and k denotes the generic internal momentum) 1 µ ν ∂ 2 Γ(2) (k. we have tγ H = 0. For a convergent graph H.4.44) 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.9: Forest diagrams associated with the simple two loop vertex correction diagram in the φ4 theory. p) p p 2! ∂pµ ∂pν tγ Γ(2) (k.4). p) = Γ(2) (k.

49) leads to . (16. under an arbitrary field redefinition φ → φ + δφ inside the path integral.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 .47) This describes the dynamics of the system at the tree level.48) We note that since the generating functional does not depend on the field variable (we are integrating over all field configurations).46) The Euler-Lagrange equation for the theory is given by δS g = F [φ] = ∂µ ∂ µ + m2 φ + φ2 = 0. the generating functional would be stationary leading to R δZ = 0 = N Dφ (F [φ] − J) ei(S[φ]+ Jφ) . 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. this equation modifies and the modified equation can be obtained from the generating functional for the theory as follows. (16.690 16 Renormalization theory Weinberg’s theorem. 16. δφ 2 − (16. (16. when we include quantum corrections. Working out in detail. However. 2 2 3! L= (16.49) where the Euler-Lagrange operator F is defined in (16. This theorem is extremely important in the study of renormalization.120)) R Z[J] = eiW [J] = N Dφ ei(S[φ]+ Jφ) .47). We recall that the generating functional for the theory in the presence of an external source is defined as (see (12.

Equation (16. this can be written explicitly as g 2 δ δJ δ δJ ∂µ ∂ µ + m2 φc + or. F φc − i δJ (16. For the φ3 theory described by (16.80). g ig δφc ∂µ ∂ µ + m2 φc + φ2 − i − J = 0. or. or.51) with respect to φc and set φc = 0. Σ= ig 2 δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c . (16.3 Schwinger-Dyson equation 691 eiW [J] F − i or.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 . 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. 16. δJ δW δ −i − J = 0. If we now take the functional derivative of (16. we obtain ig δ δ2 Γ (∂µ ∂ + m )½ + 2 = − δφc 2 δφc µ 2 δ2 Γ δφ2 c −1 −1 or.51) δφc where Γ[φc ] corresponds to the effective action defined in (13. F δ − J eiW [J] = 0. (16.78).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).16.50) Here φc denotes the classical field which is defined in (13.46). φc − i φc − i − J = 0. δJ δJ δ − J = 0.

by taking the second functional derivative of (16.11. The equations in (16.4 BPHZ renormalization We have seen in detail how multiplicative renormalization works. 16. To see in some detail how the BPHZ method works. 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 .692 16 Renormalization theory = .53) are integral equations (they are written in a compact notation here) for the two point and the three point functions. Figure 16.10: Schwinger-Dyson equation for the two point function. 16. (We are using a very compact notation where integrations over intermediate coordinates/momenta are suppressed. (16.) Similarly. = + + Figure 16.53) which can be represented graphically as in Fig. Such equations (and for the higher point functions) are known as the Schwinger-Dyson equations. orders.52) and (16.51) with respect to φc and setting φc = 0.11: Schwinger-Dyson equation for the three point function. let us consider the φ3 theory in six dimensions described by the Lagrangian density .

the 2-point and the 3-point amplitudes are superficially divergent. in dimensional regularization which we will be using.) Although the zero point amplitude is divergent. 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. let us set M = 0 for simplicity. In six dimensions. 2 2 3! (16. we can remove this divergence by normal ordering the theory.56) . Furthermore. for this theory. 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.16. (Note that unlike the φ4 theory.55) Thus. (16.54) This is a simple quantum field theory that is quite helpful in understanding various features of renormalization. (16. since we are interested in the ultraviolet behavior of the theory. the counting of the canonical dimension gives [φ] = 2.4 BPHZ renormalization 693 L= 1 M2 2 g ∂µ φ∂ µ φ − φ − φ3 . Furthermore. here the Lagrangian density is not invariant under φ → −φ.

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

As we have discussed earlier (see (15. let us simplify this. 2− 6−ǫ ǫ ǫ n =2− = 2 − 3 + = −1 + .76)).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 . we obtain the one loop self energy of the theory in (16.16. (16. to understand the structure of the self energy at one loop. ǫ Γ −1 + 1 1 ǫ 1 + ln 4π .60) = 1 6 Using these. 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. However. 2 2 2 2 ǫ ≃ 2 2 − + (γ − 1) . ǫ 1 2− n 2 dx 0 (x(1 − x)) 1 = 0 dx (x(1 − x))1− 2 ǫ ln x(1 − x) 2 ǫ ≃ = 0 1 0 dx x(1 − x) 1 − dx x(1 − x) (1 − ǫ ln x) 1+ 5ǫ 6 .

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 .62) Similarly. 12(4π)3 ǫ =− (16. i xi Ai ) × ( (16.64) . we can calculate the three point amplitude at one loop as (there is a second graph with p1 ↔ p2 which contributes to the amplitude. we will concentrate on graphs) k p k+p ǫ p2 k + p + p1 p1 3 = −igµ 2 3ǫ dn k (i)3 (2π)n k2 (k + p)2 (k + p + p1 )2 dn k dx1 dx2 dx3 (2π)n = g3 µ 2 2! × x1 k2 + x2 (k + p + p1 )2 + x3 (k + p)2 δ (1 − x1 − x2 − x3 ) 3. (16. (16.61) This leads to the one loop counter term ig2 p2 2 .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 . but since the BPHZ method subtracts out divergences in every graph.

(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 . where we have identified x1 (1 − x1 )Q2 = x1 (1 − x1 ) p2 + x2 (1 − x2 )p2 1 +2x1 x2 p · p1 .66) (16. 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 .

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. (16. ǫ 4πµ2 (16.68) as ig3 µ 2 =− (4π)3 ǫ (16.71) 1 1 dx1 0 0 du (1−x1 ) 2 (−x1 (1 − x1 )Q2 ) − ln −γ . ǫ 4πµ2 (16.72) .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.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 .698 = 2g3 µ 2 3ǫ 16 Renormalization theory 1 1−x1 dx1 0 0 1 3ǫ dx2 1−x1 Γ 3− n i(−1)3 2 n 3− n 2 (−x (1 − x )Q2 ) 2 Γ(3)(4π) 1 1 (−x1 (1 − x1 )Q2 )3− 2 2 −γ ǫ Γ 3− n 2 n ig3 µ 2 = − n (4π) 2 ig3 µ 2 ≃ − (4π)3 ǫ dx1 0 1 0 dx2 1−x1 (16. we can rewrite (16. ⇒ x2 = 1 − x1 . (16. we have x3 = 1 − x1 − x2 and x3 = 1 x3 = 0 so that 1 1−x1 ⇒ x1 = x2 = 0.69) dx1 dx2 dx3 δ(1 − x1 − x2 − x3 ) = dx1 0 0 dx2 .70) Furthermore.

Let us first look at the graphs with non-overlapping divergence of the form shown in Fig.74) Let us next calculate the self-energy at two loops. Let us look at the two integrals in (16. 16. (16.59). 1 (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. = 2(4π)3 ǫ ǫ (16. Using the form of the one loop self-energy in (16.12 (there would also be diagrams with a one loop self-energy and counter term insertion on the other internal line) k k + p k+p p k+p k p p Figure 16.68) leads to the one loop counter term (the factor of arises from doing the integrals over x1 .16.75) has the form .62).75) separately.73) 1 2 Equation (16. the first integral in (16. x2 ) ig3 µ 2 2 .4 BPHZ renormalization 699 where in the new variables x1 (1 − x1 )Q2 = x1 (1 − x1 )(p + up1 )2 + u(1 − u)p2 . = + −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 .12: Non-overlapping two loop self-energy diagrams in the φ3 theory.

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

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 .) 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.64) as well as (15. let us look at the second integral in (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 .85).76) and (16.60) in the intermediate steps. + − 2− 72(4π)6 ǫ ǫ 4πµ2 12 (16.76) Here we have used (16. (The y integration is related to the appropriate beta 2 function. Next. they will require non-local counter terms to cancel them.

78) I1 + I2 = ig4 p2 2 11 + finite .75) takes the form (−p2 ) 43 2 2 ig4 p2 − ln + −γ − 2− 72(4π)6 ǫ ǫ 4πµ2 12 + = (−p2 ) 8 2 4 ln − − + γ + finite ǫ2 ǫ 4πµ2 3 (16. 72(4π)6 ǫ2 ǫ 4πµ2 3 As a result.702 ≃ − 16 Renormalization theory g 4 p2 2 i 2 − + (γ − 1) − ln 4π 3 ǫ (4π)3 12(4π) ǫ × ǫ (−p2 ) 5ǫ 1 1 − ln + 6 2 µ2 6 ≃ − = ig4 p2 2 2 (−p2 ) 5 − + ln − + (γ − 1) + O(ǫ) 72(4π)6 ǫ ǫ 4πµ2 3 (16.77). − 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. takes the form ℓ k p ℓ+p k+p p . Let us next look at the overlapping divergent graph in the self energy at two loops which. upon using the form of the one loop vertex function in (16. adding (16.76) and (16.77) ig4 p2 4 (−p2 ) 8 2 − ln − − + γ + finite .67) (with x2 expressed in terms of u as in (16.71)). the sum of the two graphs in (16.

73) (with appropriate momenta).79) n −3 n where Q2 is defined in (16. x1 (16.4 BPHZ renormalization 703 3ǫ = (−igµ 2 ) 2 × ǫ i i dn ℓ ig3 µ 2 dx1 du 2 − n (2π)n ℓ (ℓ + p)2 (4π) 2 n 2 n 2 3− 2 − x1 (1 − x1 )Q n (1 − x1 )Γ 3 − n g4 µ2ǫ (−1) 2 −3 Γ 3− = n 2 2(4π) 2 dn ℓ x 2 (1 − x1 ) 2 −2 dx1 du 1 n . we can write (16. (16.81) . (2π)n ℓ2 (ℓ + p)2 (Q2 )3− 2 (16.80) n 2− n 2 × δ(1 − y1 − y2 − y3 ) D2 5− n 2 where we have identified D2 = y1 Q2 + y2 (ℓ + p)2 + y3 ℓ2 = y1 (ℓ + up)2 + = u(1 − u) 2 p + y2 (ℓ + p)2 + (1 − y1 − y2 )ℓ2 x1 2 ℓ + (uy1 + y2 )p + y1 (1 − y1 )u2 + y2 (1 − y2 ) − 2uy1 y2 + y1 u(1 − u) 2 p . Using (16.16.64) to combine denominators.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 .

704 16 Renormalization theory The divergence of the two loop integral in (16. in fact.81) has the form D2 = ℓ2 + (1 − y1 )(y1 (u − v)2 + v(1 − v)) + = ℓ2 + d(x1 . we can do the integration over y3 using the delta function in (16. However. where d is defined in (16. let us evaluate the integral in (16.80) in some detail.83).80).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).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. This is a special feature of overlapping divergent graphs and to appreciate this.84) . First we note that we can shift the variable of integration ℓ → ℓ − (uy1 + y2 )p.83) (16. Finally. v)p2 . To see the divergence . the divergence of the subdiagram (one loop vertex) has been transformed to the Feynman parametric integrals. Next. we can redefine the variable y2 (as we have already done in (16. so that the (shifted) denominator in (16. y1 . y1 u(1 − u) 2 p x1 (16.82) Substituting this into (16. u. The one loop pole structure is manifest in the overall multiplicative gamma function. this is not right and.71)) as y2 = (1 − y1 )v.

84) gives the value of the overlapping self-energy diagram to be .84) can be evaluated exactly using Gegenbauer polynomials and the only modification from the leading order result in (16.4 BPHZ renormalization 705 structure from the parametric integration. The parametric integration in (16.85) the leading term in the parametric integration in (16.86) This demonstrates how the subdivergence is hidden in the parametric integration.86) is that the finite constant 1 inside the parenthesis changes from −γ + 6 to 4.84) 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 .16. let us look at only the leading order term in (x1 d)1−ǫ . Using the standard results 1 0 1 0 dt tα−1 (1 − t)β−1 = 1 du u(1 − u) = . substituting this result into (16. 6 1 Γ(α)Γ(β) = B(α. β). −γ+ 6 ǫ 6 (16. Γ(α + β) 0 1 dudv (u − v)2 = . Thus. 6 (16.

we also have a contribution from the graph k p k+p = (−igµ 2 ) 2 g 4 µǫ 2 4(4π)3 ǫ g 4 µǫ 2 4(4π)3 ǫ n ǫ p i dn k i ig3 µ 2 2 n k 2 2(4π)3 ǫ (k + p)2 (2π) 1 dn k dx n (2π) ((k + xp)2 + x(1 − x)p2 )2 dx Γ 2− n (−1)2 i 2 n n (4π) 2 Γ(2) (−x(1 − x)p2 )2− 2 dx x(1 − x) ǫ 2 Γ n 2 n −2 2 ǫ = − = − ig4 µǫ (−p2 ) 2 −2 2 ǫ = − Γ −1 + n 3 (4π) 2 ǫ 2 4(4π) ig4 p2 4(4π)6 2 ǫ (−p2 ) 4πµ2 1− − ǫ −2 = ≃ = Γ −1 + Γ(n − 2) 1 6 1+ 5ǫ 6 −1 2 ig4 p2 2 4(4π)6 ǫ ig4 p2 24(4π)6 ǫ (−p2 ) ln 2 4πµ2 ln 2 − + (γ − 1) ǫ 2 4 + 2 ǫ ǫ 8 (−p2 ) +γ− 2 4πµ 3 + finite.74)). (16.706 16 Renormalization theory = − = 1 (−p2 ) 1 2 ig4 p2 − + γ − 1 1 − ǫ ln + 4 + O(ǫ) 6 2(4π) ǫ 4πµ2 6 ǫ (16.87) ig4 p2 2 (−p2 ) 2 + − γ + 3 + finite .88) . with the one loop counter term for the three point function (see (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. On the other hand.

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. ǫ2 3ǫ We see that all the potentially dangerous nonlocal divergent terms have disappeared from the self-energy and the remaining divergences can be removed by adding a local two loop counter term. 16. BPHZ is a renormalization procedure where divergences are subtracted in the graph itself so that it works with any regularization.13: Overlapping two loop self-energy diagrams in the φ3 theory. (16.13 gives + + Figure 16. the graph with the counter term at the left vertex also contributes an equal amount k p k+p = ig4 p2 24(4π)6 − 2 4 + 2 ǫ ǫ ln 8 (−p2 ) +γ− 2 4πµ 3 + finite.16.90) 8 (−p2 ) +γ− 2 4πµ 3 = ig4 p2 12(4π)6 − 2 2 + + finite .89) p so that the sum of the three graphs in Fig. .85) in the intermediate step. Similarly.

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. (16.91) namely. (16.92) while γ1 and γ2 are called nested if one is contained inside the other. for the three loop self energy graph in the φ3 theory. if γ1 ⊂ γ2 . Thus. if it is superficially divergent. Fig. 16. An 1PI subgraph γ is called proper if it is different from the graph G itself and it is called a renormalization part if D(γ) ≥ 0. 16.15 shows examples of different ways of drawing boxes around the renormalization parts. namely. .14.14: Examples of proper renormalization parts in the two loop self-energy graph in the φ3 theory. two renormalization parts γ1 and γ2 are called overlapping if they share lines and vertices.708 16 Renormalization theory Consider an arbitrary Feynman graph G. It may contain one particle irreducible subgraphs that are superficially divergent. for example. (16.93) Finally. γ γ Figure 16. Two renormalization parts γ1 and γ2 are called disjoint if γ1 ∩ γ2 = 0. if none of the following holds. namely. or γ2 ⊂ γ1 . A couple of examples of proper subgraphs which correspond to renormalization parts are shown in Fig.

(16. γ1 ∩ γ2 = 0. Similarly. the complete set of laying down of boxes (forests) containing disjoint renormalization parts is shown in Fig.94) A set of such laying down of boxes around renormalization parts is called a forest. Thus. γ2 ⊂ γ1 . .17 shows the nested forests in the two loop self-energy graph in the φ3 theory. γ1 ⊂ γ2 . let us add to the set D the graph with no boxes. let N denote the set of laying down of boxes (forests) containing only nested renormalization parts.4 BPHZ renormalization 709 Figure 16. a forest is called full if there is a box around G.16. 16. Fig. 16. A forest is called empty if there is no box around any subdiagram. for the two loop self-energy graph in the φ3 theory. Let D denote the set of laying down of boxes which contain only disjoint renormalization parts.15: Examples of forests in the three loop self-energy graph in the φ3 theory. On the other hand.16. A forest is called normal if there is no box around the complete graph G. Furthermore.

for example.17: The complete set of nested forests for the two loop self-energy graph in the φ3 theory.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 introduce an operator tγ (the Taylor operator or the Taylor expansion operator) for any graph γ such that acting on the integrand of the graph. 16. Figure 16.18 for the two loop self-energy graph in the φ3 theory. we define O to denote the set of laying down of boxes (forests) containing overlapping renormalization parts. In the conventional BPHZ method. The reason why we draw boxes around renormalization parts is because these sub-diagrams are superficially divergent and to make the full diagram finite. shown in Fig. it Taylor expands . we must subtract out the divergences from each of these sub-diagrams. Figure 16. Figure 16.710 16 Renormalization theory Furthermore.

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

the tγ operation is defined to be (it separates out the pole terms) −1 m=−n t FG = G am ǫm . (16.103) . in fact. (16. a graph (integral) at n-loops.98) In such a case. in the case of dimensional regularization.712 16 Renormalization theory away potentially divergent parts inside the integral. Therefore.100) Once again. this makes it clear that for a superficially convergent graph (with no pole term) tG FG = 0. ∞ m=0 (16. (16. am ǫm = finite parts. has the general form ∞ m=−n FG = am ǫm . In particular. compatible with any regularization. if the integral has the form ∞ m=−n Iγ = am ǫm .101) Given a renormalization part γ in a Feynman diagram.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.99) so that the finite part of the graph can be identified with ∞ m=0 1−t G FG = am ǫm = F G = finite. (16. (16. it never uses any particular regularization and is.

the graph would now be finite. (16. However.105) We note that if G is not superficially divergent. (16. (16. γs as subdiagrams. and (16. one needs a final subtraction to render the diagram finite. . from the definition of RG IG in (16.106) RG IG = RG IG . this method is correspondingly more general. . by Weinberg’s theorem. one defines s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . 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. graph by graph this procedure can be applied and every graph can be made finite. the effect of the term (−tγ Iγ ) can be thought of as that of adding counter terms which subtract the divergence.4 BPHZ renormalization 713 Thus. the graph G itself may be superficially divergent in which case. the subtraction must be carried out . γ2 . Thus. However. makes all the subdiagrams (proper renormalization parts) in G finite. Then. Thus. since in this method we do not explicitly use counter terms. The only understanding here is that for nested diagrams.107) but this is not true otherwise.16. then.105). by definition tG RG IG = 0. In any case. of course. Let us consider an arbitrary Feynman diagram G with proper renormalization parts γ1 .104) This. 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 . Fig. this does not address the issue of locality of counter terms since we are subtracting out overlapping divergences which. For disjoint or overlapping divergences. The proof of renormalization by local counter terms. Here Rγk is the operator (16. for example.714 16 Renormalization theory inside out. we can now write RG IG = = 1 − tG RG IG 1 − tG (−tγk ) IG . of course. (16. Zimmermann’s solution of the BPH recursion relation (16.108) as well as . can lead to non-local counter terms. Although this method of making a Feynman diagram finite is absolutely correct. 16.108) 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. With this. the order of the subtraction is irrelevant.108) is obtained as follows.109) where Φi is the complete set of proper renormalization parts that are non-overlapping. as we have seen. This is the BPH formula. Instead of the set of disjoint proper renormalization parts (subgraphs) of a Feynman graph G. Φi γk ∈Φi (16. Thus. The BPH recursion relation (16.104) associated with the superficially divergent renormalization part γk . Equation (16.19 defines all the disjoint and nested proper renormalization parts of the three loop self-energy diagram. requires that the overlapping divergences be taken care of by lower order counter terms. This is seen through Bogoliubov’s R-operator which defines RG IG = = 1 − tG RG IG 1 − tG Φi γk ∈Φi −tγk Rγk IG . That this is true will be seen shortly through examples. Its effect is to give IG itself). let us look at the complete set of subgraphs that include only disjoint and nested proper renormalization parts.

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

716 16 Renormalization theory RG IG = IG + IG/γ −tγ Rγ Iγ = IG + IG/γ (−tγ Iγ ) .113) . Let us next look at the two loop self-energy graphs shown in Fig.110) also as RG IG = IG + (−tγ IG ) = (1 − tγ ) IG . there is no subdiagram in this case. RG IG = (16. Rγ Iγ = Iγ because there is no superficially divergent proper subdiagram of γ. by definition. (16. (16.105) and. in fact. From the relation IG = IG/γ Iγ .112) This is exactly the original formula (16.21: The complete set of disjoint proper renormalization parts of the two loop self-energy graph in the φ3 theory.111) 1 − tG RG IG = 1 − tG (1 − tγ ) IG .108)) we can write γ1 γ2 Figure 16. this is how we are supposed to renormalize the theory. (16. in fact. we can write (16.21 containing only proper disjoint renormalization parts. Here we have overlapping divergences and following BPH (see (16.110) where IG/γ is the part of the diagram which does not contain the subdiagram Iγ . We note that. 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 . 16.

16.117) tγ1 tγ2 = −tγ1 = − = = ǫ ig3 2 γ1 1 −igµ 2 t 3 ǫ 2(4π) 2 dn k (i)2 (2π)n k2 (k + p)2 − 2 ǫ (16.115) which we recognize to correspond to the counter term associated with an overlapping divergence if it did not vanish.118) 2 i (−p2 ) g4 4(4π)3 ǫ (4π)3 6 ig4 p2 2 . At one loop we had found that (see (16. It is such terms that complicate the proof of renormalizability. (16. sets the factor of µ 2 = 1 in the graph at one loop) = so that ig3 2 = (−tγ2 Iγ2 ) .74). tγ which is supposed to isolate ǫ pole terms in ǫ. Let us recall some of the calculations we had done earlier. 12(4π)6 ǫ2 .105) RG IG = 1 − tG (1 − tγ1 ) (1 − tγ2 ) IG .114) This does not exactly coincide with the original formula (16. The difference between the two is given by 1 − tG tγ1 tγ2 IG .116) (16. (16.4 BPHZ renormalization 717 so that RG IG = 1 − tG (1 − tγ1 − tγ2 ) IG . 2(4π)3 ǫ (16.

there is a theorem that says that if γ1 and γ2 denote two proper renormalization parts of G that are overlapping.119) Physically. (16.87)).122) i where Φi ’s define forests containing only disjoint or nested proper renormalization parts. γi ’s are only disjoint or nested proper renormalization parts.105) and Zimmermann’s solution (16. it is best to study an example.121) involving subtractions of only non-overlapping proper renormalization parts. If we recall our calculation for the two-loop overlapping divergence (see(16. (16. namely. physically because it says that we do not need to subtract overlapping divergences. The product of the factors in (16. then.120) This result is very important. In general. Φi γk ∈Φi (16.718 16 Renormalization theory where we have used the fact that n = 6 − ǫ and that tγ1 acts on ǫ the integral as well as on µ 2 to isolate only pole terms. This makes the connection between (16. that 1 − tG tγ1 tγ2 IG = 0. 2 therefore. To make contact with the formula of BPH.121) can also be written as (1 − tγi ) ≡ (−tγk ) . we note that using this. this is the statement that we do not need subtractions associated with overlapping divergences. the ǫ1 term precisely has the same coefficient and it follows. but more importantly. (16. Let us recall that the three loop diagram has eight . we can now write s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . then we can always find a renormalization part γ12 which contains both γ1 and γ2 and for which (1 − tγ12 ) tγ1 tγ2 IG = 0.109).

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 . + (tγ1 Rγ1 )(tγ2 Rγ2 ) IG = (16.γ2 } (−tγ1 Rγ1 Iγ1 )(−tγ2 Rγ2 Iγ2 ) + IG/γ4 (−tγ4 ) Rγ4 Iγ4 = 1 − tG 1 − tG 1 − tγ1 Rγ1 − tγ2 Rγ2 − tγ3 Rγ3 − tγ4 Rγ4 −tγk Rγk IG .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 . = Iγ2 . (16.γ2 } (−tγ1 Iγ1 ) (−tγ2 Iγ2 ) + IG/γ3 −tγ3 Iγ3 /γ1 (−tγ1 Iγ1 ) + IG/γ4 −tγ4 Iγ4 /γ2 (−tγ2 Iγ2 ) .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. 16.124) Using this. (16.123) as RG IG = (1 − tG ) IG + IG/γ1 (−tγ1 Rγ1 Iγ1 ) + IG/γ2 (−tγ2 Rγ2 Iγ2 ) + IG/γ3 (−tγ3 ) Iγ3 + Iγ3 /γ1 −tγ1 Rγ1 Iγ1 + IG/{γ1 .125) Φi γk ∈Φi . Thus.4 BPHZ renormalization 719 forests consisting of disjoint and nested proper renormalization parts which are shown in Fig. we can rewrite (16.

it is equivalent to making any Feynman graph finite by local subtractions (counter terms). This analysis. We would have divergence structures with more and more numbers of external lines.128) not only would we have a finite number of types of graphs that can be divergent.20) and (16.21)) 2F + 3 δi = di + bi + fi − 4. In contrast.720 16 Renormalization theory where Φi is the complete set of proper renormalization parts consisting of only disjoint graphs.127) then. of course. However. namely. Such theories are known as renormalizable theories. Such theories are called nonrenormalizable theories. 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 .108). 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. This is. On the other hand if δi > 0. (16.126) 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 this case. Of course. through Weinberg’s theorem. (16. therefore. 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. So far. if the index of divergence were negative. In this case. shows how the BPH formula leads to Zimmermann’s solution and how. 2 (16. δi < 0. it is clear that the superficial degree of divergence of a graph would increase with increasing number of interaction vertices. we have talked about theories where the index of divergence of interactions in four dimensions (recall that D = 4 − B − 3 i ni δi . see (16. the BPH formula (16. we can also remove the divergences by adding local counter terms.

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

a ν µ ν 2 − Z4 g2 abc apq b c µ. Z1 = 1 + B. Namely. Z4 = 1 + C. ˜ Z1 = 1 + E. This is.a 4 2 ˜ ˜ (16.91)). if we redefine the fields and parameters as a(u) 2 Aµ = Z3 Aa . −3 1 1 g2 ˜ − ˜ −1 = Z1 Z3 2 Z3 g. But the main question is whether renormalization would preserve the gauge invariance – actually the BRST invariance – present in the original theory. µ where we have identified Z3 = 1 + A. as we have seen explicitly in the calculation in QED. the longitudinal part of the photon self-energy is not renormalized by quantum corrections.722 16 Renormalization theory +(1 + D)∂µ ca ∂ µ ca − (1 + E)gf abc (∂ µ ca )Ab cc µ = − Z3 ∂µ Aa (∂ µ Aν.130) + Z3 ∂µ ca ∂ µ ca − Z1 gf abc ∂ µ ca Ab cc . With these counter terms.131) It is worth emphasizing here again that we have not added any counter term for the gauge fixing terms because. all the divergences can be taken care of. g1 (u)2 (u) −2 = Z4 Z3 g2 .p ν.a − ∂ ν Aµ.a ) + Z1 gf abc Ab Ac ∂ µ Aν. a general feature of gauge theories following from the BRST invariance of the theory (see (13. of course. in fact. ˜2 ca(u) = Z3 ca . g(u) = Z1 Z3 2 g. µ 1 ˜2 ca(u) = Z3 ca . (16. −1 1 F a(u) = Z3 2 F a . ˜ Z3 = 1 + D. .q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ.

In this case. −2 2 −3 = Z4 Z3 g2 = Z1 Z3 g2 .91)) lead to relations between various amplitudes and we can show from these that they lead to relations among renormalization constants of the form (analogous to the relation Z1 = Z2 in QED following from the Ward-Takahashi identity) ˜ Z4 Z1 Z1 = = . (13.133) and with this identification we can write (16.134) and remove all the divergences from the theory. the same relation as in (16.134) Therefore. we will have 2 g(u) g1 2 −3 = Z1 Z3 g2 . The Slavnov-Taylor identities following from the BRST invariance (see. ˜3 Z3 Z1 Z (16. for example.136) This is.130) as . (16. g(u) = g1 (u) (u) = g2 .135) which would lead to g(u) 2 = g1 (u) 2 = g2 (u) 2 . 2 −3 ˜ 2 −1 ˜ −2 = Z1 Z3 Z3 g2 = Z1 Z3 g2 .132) ξ (u) = Z3 ξ. (u) 2 (u) 2 g2 (16.133) This would. But the real question is how can the renormalization process guarantee this. in turn require that the corresponding counter terms have to be related. if we choose a regularization scheme which respects the BRST symmetry. namely.5 Renormalization of gauge theories 723 (16.16. therefore. the counterterms will satisfy (16. of course. (16. 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.

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

e.142) then this would not be invariant under the chiral symmetry transformations (16.140) Here λ is the space time independent real infinitesimal parameter of transformation and (16.6 Anomalous Ward identity 725 is also invariant under the global transformation δψk = −iλγ5 ψk .a = ψ j γ µ (T a )jk ψk .143) is without the parameter of chiral transformation. Nonetheless.141) transforms covariantly under a gauge transformation and. δψ k = −iλψ k γ5 . (16.a = 0. will not be conserved. we can still define the axial vector current as in (16. µ ∂µ J5 = 0.. Dµ J µ. As is the case with continuous symmetries.140).140) describes the chiral symmetry transformations under which the massless theory is invariant. In this theory.141) We note here that the vector current in (16. (16. there are two conserved currents. if the fermions are massive. We can also derive identities from this relation that would relate various matrix elements of the . therefore. 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.141) which. i. therefore. (16. (16. the invariance of the theory leads to a conserved current. if there is a term in the Lagrangian density of the form Lm = −mψ k ψk . however. Let us note that. J µ. µ J5 = ψ k γ5 γ µ ψk .16.143) where the variation of the Lagrangian density in (16. rather it would satisfy the equation µ ∂µ J5 = −δLm = −2imψ k γ5 ψk . is covariantly conserved.

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

(16. (16. 16. in 1 + 1 dimensions the vector and the axial vector currents are related as µ J5 = ψγ5 γ µ ψ = ǫµν ψγν ψ = ǫµν Jν .16.151) Let us now calculate the one loop amplitude involving a photon and an axial vector current as shown in Fig.146) . γ µ γ ν = η µν + γ5 ǫµν .152) . but the ones that are directly of interest to us are γ5 γ µ = ǫµν γν . note that in addition to the diagonal metric tensor η µν = (+. 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) .148) There are various simple identities in 1 + 1 dimensions involving the Dirac matrices.147) Furthermore. 2 2 (k + p)2 (2π) k µν I5 (p) = −e2 (16.149) As a result of these identities. µ ∂µ J5 = 0. (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 = . with ǫ01 = 1. (16. (16.150) and at the classical (tree) level both these currents are conserved (recall that the fermions are massless in the Schwinger model and the gauge group in the present case is U (1)) ∂µ J µ = 0.

Furthermore.22: The amplitude with an axial vector current and a photon.73). Furthermore.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. Thus. we note that the amplitude in (16.152) 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. Γ(2) (−x(1 − x)p2 )2− 2 (16.36) in the limit m = 0. Here we have used the propagators for the fermions following from (15.153) We can evaluate the integral in (16. since .728 16 Renormalization theory γ5 γ µ k γν p k+p Figure 16. generalizing the integral to n = 2 − ǫ dimensions would lead to µν I5 (p) = − dn k kλ (k + p)ρ (2π)n k2 (k + p)2 kλ (k + p)ρ dn k dx n (2π) (k + xp)2 − x(1 − x)p2 dx dn k kλ kρ − x(1 − x)pλ pρ (2π)n (k2 − x(1 − x)p2 )2 n = − = − = − 2 Γ(1 − n ) 1 ηλρ iπ 2 2 − dx n n (2π) Γ(2) 2 (−x(1 − x)p2 )1− 2 − Γ(2 − n ) x(1 − x)pλ pρ 2 n . (16.71) and (15.

π (16.156) Since each of the terms is now completely finite. However.158) = − .155) where we have used the gamma matrix identity in n dimension in (15. We note from (16. in fact. η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 (16.149) and the final result is obtained to be (the x integration is trivial) µν I5 (p) = − ie2 λρ νµ pλ pρ . (16.86).6 Anomalous Ward identity 729 γ5 is a two dimensional quantity we have not yet evaluated the trace over the Dirac matrices. (16.16.154) that the second term inside the bracket is finite and hence for this term the trace over the Dirac matrices can be carried out in two dimensions (because any difference in the trace from extending to n diemnsions would be proportional to ǫ and hence would vanish in the limit ǫ → 0). using this relation. On the other hand. (16.157) The result in (16.154) takes the form (we now restore the multiplicative factor) µν I5 (p) = − ie2 n (4π) 2 − dx − Tr(γ5 γ µ γ λ γ ν γ ρ )x(1 − x)pλ pρ (−x(1 − x)p2 )1+ 2 ǫ ǫ 1 Tr(γ5 γ µ γ ν )ǫΓ( 2 ) ǫ 2 (−x(1 − x)p2 ) 2 . It is clear now that the factor (2 − n) = ǫ makes even the first term finite and. the trace over the Dirac matrices can be carried out in two dimensions using the identities (16. the first term is divergent and we have to be careful in evaluating the trace.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 . 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 ).

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

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

Muta. However. J. Aspects of symmetry. 75. J. Physical Review 128. 4. these topics are beyond the scope of these lectures. S. 6. Caswell and A. 392 (1982). Kennedy. Bogoliubov and O. J. N. W. ed G. F.732 16 Renormalization theory gauge group uniquely to be SO(32) or E(8) × E(8). 1848 (1969). S. C. Physical Review 25. Physical Review 75. Ward. S. 9. Wightman in “Renormalization Theory”. 227 (1957). E. S. K. Singapore (1987). Proceedings of Royal Society A64. 16. Proceedings of the International School of Mathematical Physics “Ettore Majorana”. 11. 47 (1969). Foundations of Quantum Chromodynamics. J.7 References 1. Bardeen. Physical Review 118. Wightman. Communications in Mathematical Physics 2. 10. 2426 (1969). Physical Review 184. . Physical Review 82. 1985). 2425 (1962). 8. Parasiuk. A. World Scientific. S. Adler. 426 (1951). 84. 54 (1951). Erice (1976). 5. N. 14. Salam. Weinberg. Velo and A. Dyson. 12. 838 (1960). Bell and R. 3. 7. Nuovo Cimento 60A. Physical Review 177. Coleman. Hepp. A. 1736 (1949). 13. Cambridge University Press (Cambridge. 2. 486 (1949). D. Schwinger. S. 301 (1966). 217 (1951). W. Jackiw. T. Acta Mathematica 97.

(Physical electric charge is determined from the scattering of on-shell electrons. 17.1 Gell-Mann-Low equation In the early 1950s. −→ V (r) = e2 αp . We can. Gell-Mann and Low were interested in studying how the electric force and the potential behave at extremely short distances. 1+ r 3π rme 733 (17. the Coulomb potential can also be obtained from the Born amplitude for the scattering of two electrons exchanging a photon.1) that at the lowest order the potential is independent of the mass of the fermion and it scales with distance as 1 . then the potential is obtained to have the form V (r) = 2αp αp 1 ln + ··· . We have seen earlier at the end of section 8.2) . In the same manner. namely.Chapter 17 Renormalization group and equation 17. ep denote the fine strucp c ture constant and the physical electric charge of the fermion respectively. of course. r (17.1) 1 where in CGS units α2 = p ≃ 137 and αp .) It is clear from (17. calculate quantum corrections to r this potential and if we add the one loop correction to the potential shown in Fig.1.10 that the Yukawa potential can be related to the Born amplitude for the scattering of two electrons exchanging a (pseudo) scalar meson.

(17. To understand the origin of this behavior. V (r) → 1 as r → 0 (at exr tremely short distances) as naive scaling would imply. When r is 1 1 large.2) that the potential no longer scales as 1 and furthermore. it also behaves as V (r) → α r 1− Λ 2α ln 3π me = αp . we see from (17. In fact. 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. in this limit.1: The one loop correction to the electromagnetic potential.4) where we have identified Λ 2α ln αp = α 1 − 3π me . we note that the one loop potential calculated with a cut off Λ leads to the form (together with the lowest order term) 2α α 1− ln r 3π 1 + r 2 Λ2 + ··· . it r depends on the mass of the electron in such a way that we cannot take the limit me → 0.734 17 Renormalization group and equation Figure 17. namely r ≫ me ≫ Λ . Furthermore.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 → 0.5) This relation can be inverted to express the renormalized parameter in terms of the physical parameter as . 1 + r 2 m2 e V (r) = (17. r (17. For example. It is clear that the one loop corrected potential has a very different character from the lowest order potential.

1+ r 3π rme This expression coincides with (17.6) Expressing the potential (17. on dimensional grounds we see that the potential can be expressed as .8) Furthermore.17.9) then. 1 r= m e (17. Let us next investigate what would happen if we define the electric charge at some arbitrary (intermediate) distance R and not as in (17.7) = 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 + ··· .1 Gell-Mann-Low equation 735 α = αp 1 + Λ 2αp ln 3π me + ··· . this derivation clarifies that the singularity in the potential at me → 0 arises purely because of the renormalization of charge at large distances.3) in terms of this (physical) parameter.8). (17. We also see that the naive scaling of the potential breaks down and mass dependence comes in due to renormalization effects. (17. For example. 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. in the limit Λ → ∞. if we define αR = RV (R).2) and also shows that the physical fine structure constant can be identified with αp = rV (r) .

(We note that rme = Rme × R so that it is not an independent quantity.3) as follows.13) × 1− = ≃ αR r αR r 1+ 1+ 2αR ln 3π 2αR R ln + · · · 3π r which is well behaved as we have already mentioned. 1 + R 2 m2 e (17.) This function should 1 not be singular as me → 0 which implies that for r.3) leads. in the limit Λ → ∞ and rme . 1 + r 2 m2 e + ··· 1 + R 2 m2 e (17. R .11) Inverting this relation we obtain α = αR 1 + 2αR ln 3π 1 + R2 Λ2 + ··· . the dependence of the potential on me can be neglected. r R (17. which can be seen from (17. Rme ≪ 1 (the second of these limits corresponds to me → 0).736 17 Renormalization group and equation r 1 V (r) = F αR . R ≪ me . We note that αR = RV (R) = α 1 − 2α ln 3π 1 + R2 Λ2 + ··· . Rme . to V (r) = αR r 1+ 2αR ln 3π 2αR ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· 1 + r 2 m2 e 1 + R2 Λ2 1 + r 2 Λ2 . Let us also note that if we define the electric charge (fine structure constant) as .10) where F is a dimensionless function (function of the independent r r dimensionless variables αR .12) Putting this back into the potential (17. . Rme ). 1 + R 2 m2 e (17.

14)) αr .16)) we see that the equation governing this change is given by . r V (r) = (17. then we see from (17. R ≪ 2αR ln 3π leads to the relation αr = αR 1 + = αR 1 + 1 + R2 Λ2 − ln 1 + R 2 m2 e .14) α = αR 1 + 2αR ln 3π 2αr ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· . . = F αR . = αr 1 + (17.18) (see also (17. 1 + r 2 m2 e 1 me .13) (see also the definition (17. write (see (17. in general.12) that we can identify (17.15) which.17.17) From this analysis we see that in the limit of me → 0 we can. 1 + r 2 Λ2 1 + r 2 m2 e + ··· (17.16) 2αR R ln + · · · 3π r and this is compatible with (17. R V (r) = αr (17.18) This shows that the charge (fine structure constant) changes (runs) with distance and from the second of the equations in (17.1 Gell-Mann-Low equation 737 αr = rV (r). r R r r . in the limit Λ → ∞ and r.10)) αr r 1 = F αR .

dαr 2α2 2α2 = − R = − r. (17. 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.20) and this is known as the Gell-Mann-Low equation.23) 1. if we let r.18) αr = F αR . αR does not exist and consequently.19) β(αr ) = − . In this case. then (17. We see from this discussion that through proper renormalization the dependence of the potential on the electron mass me disappears.22) There are two possibilities for the solution of the equation (17. x) ∂x (17. x) . αr= = αp .16) that. the theory develops ghosts and other problems. In this case. the bare charge becomes infinity. dr 3π 3π r (17. R → 0 with R = x fixed. ∂F (αr . r . (17.24) . R (17.18) leads to (α0 = αr=0 ) α0 = F (α0 . x=1 (17. in the present case. αR does have a limit which is nonzero and is inde1 pendent of the value 137 which arises only as an initial condir tion. As R → 0. we note that αp and me arise only as initial conditions in this evolution.738 17 Renormalization group and equation r where dαr dr = −β(αr ). namely. Gell-Mann-Low equation basically relates the electric charge at one distance (scale) to another (namely.21) 1 me We see explicitly from (17. 2. for example. As R → 0.

We will discuss these ideas in more detail in section 17.2 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. However. First.19) (also where the beta function vanishes). It is important to note that in a quantum field theory scale invariance is broken by the masses of particles. therefore. α0 → 0 which is obvious from (17. are different in the two renormalization schemes even though the regularization is the same (dimensional regularization). in dimensional regularization. we defined the charge at some length . In QED. On the other hand. the renormalization prescription makes a difference in the sense that how much of the finite part we subtract out along with the divergent parts changes the renormalized parameters of the theory. the effects of masses are negligible at high energies if we renormalize the theory in an appropriate way. Even in dimensional regularization. it corresponds to the second possibility. in the MS scheme we not only subtract out the pole parts (in ǫ) of an amplitude. Even within a given regularization. for example. For example. as r. 17. We can keep track of this by using a running coupling constant. a particular renormalization scheme may be preferrable if it makes the higher order contributions in perturbation theory smaller. therefore.16) (remember / r that R is fixed as the limit is taken) and.5. the MS scheme (minimal subtraction) corresponds to subtracting out only the pole parts (in ǫ) of an amplitude. 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). (In the example of QED that we discussed in the last section. R → 0. but some constants as well.17. The only remaining breaking of scale invariance is due to renormalization itself. There are two possible sources for this arbitrariness. The renormalized parameters. we saw that the coupling constants are defined with an arbitrary mass scale µ.2 Renormalization group 739 This is known as a fixed point of the evolution equation (17. This introduces an additional arbitrariness into the renormalized parameters depending on the arbitrariness in the mass scale used. such as the Euler’s constant etc.

(17. Namely. u ′ ′ λu = Z1 Z3−2 λ′ . On the other hand. λ. ′ ′ m2 = Zm Z3−1 m′ 2 . u −2 λu = Z1 Z3 λ. a different renormalization scale prescription would lead to φu = Z3 2 φ′ . λ′ .25) define a set of renormalized parameters. Namely. −1 m2 = Zm Z3 m2 .26) ′ where we are denoting Z3 = Z3 (µ′ ) and so on. we must have (S and S ′ denote representative S-matrix elements calculated in the two prescriptions) S ′ p. see the discussion after (16.) Let us consider a renormalizable theory. m. m′ . we are using the same regularization but different µ scales and the assumption here is that the unrenormalized field and the unrenormalized parameters are independent of the renormalization prescription. they should not only be finite. If we are calculating physical quantities such as the scattering amplitudes. µ). they must also have the same value independent of which set of parameters are used to calculate them. Calculations with either of these sets of parameters will lead to finite quantities. ′ 1 (17.27) so that if we expand the S matrix elements in a perturbation series . 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. µ′ = S(p.32)) 1 2 φu = Z3 φ. (17.740 17 Renormalization group and equation scale which illustrates the source of arbitrariness of the second kind.

25) and (17. m′ . the higher order corrections become negligible very quickly. Here we have identified z µ . λ. λ′ . In such a case.30) which leads to a relation between the two (17.33) .31) = Z3 (µ′ ) Z3 (µ) . (17.32) Similarly. in perturbation theory. If we can calculate the exact series. m′ . u −1 m′ 2 = Zm (µ′ )Z3 (µ′ )m2 . a′ λ′ n . µ φ. such as QED. we can only calculate up to a given order.17. λ′ . say the nth order. φ′ = z µ′ . in a theory like QCD (where the value of the coupling constant is not small).26) we can also write −1 m2 = Zm (µ)Z3 (µ)m2 .29) 1 In a theory. µ)φu . µ′ = (17. On the other hand. Let us next see that this arbitrariness in the renormalized parameters is due to a finite renormalization. from (17.28) 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 . (17. m. λ. µ) = n an λn . We can invert the relations in (17.2 Renormalization group 741 S(p.µ ′ −1 2 −1 φ′ = Z3 1 −2 λ′ . we can check this. n n S ′ p.25) and (17. where α ≃ 137 . m. µ′ − S(p. all we can say and check is that S ′ p. µ) = O λn+1 . the residual terms can be non-negligible.26) so that we can write the renormalized fields in the two prescriptions as φ = Z3 2 (λ. u (17. However. (17.

37) zλ (µ′ . Zm . (17. (17. Z1 ) differ only in finite parts (the divergent parts are the same since the “µ” dependence comes only in the finite parts). with (17. (17.35) and (17. parameters corresponding to two distinct renormalization prescriptions are related by a finite renormalization (which can be thought of as a finite transformation much like the symmetry transformations we have talked about in earlier chapters). with (17. Thus. it follows that (z. (17.26) imply −1 2 λ = Z1 Z3 λu . (17. ′ ′ λ′ = Z1−1 Z3 2 λu . from the definitions in (17.25) and (17.36) which leads to the relation λ′ = zλ λ. µ) = Z1 (µ′ ) Z1 (µ) −1 Z3 (µ′ ) Z3 (µ) 2 . zλ ) are completely finite.32).742 which implies that 17 Renormalization group and equation m′ 2 = zm (µ′ . Let us next show that these finite renormalizations (finite transformations) define a group. µ) = ′ Zm (µ′ ) Zm (µ) −1 Z3 (µ′ ) Z3 (µ) .34) zm (µ .38) ′ ′ ′ Since (Z3 . µ)m2 . Zm . In fact. Z1 ) and (Z3 .38) we note that .35) Finally. zm .

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

. . . m(µ). . . φu (xn )) = Z3 G(n) (x1 . . m(µ′ ). mu . µ)Γ(n) (p. G(n) (p1 . λ(µ′ ). . λ(µ). . this has the form Γ(n) (p. λu ) u = z 2 (µ′ .43) (17.42) where only (n−1) of the external momenta are independent (because of overall energy-momentum conservation). µ) = Z32 (µ)Γ(n) (p. . . u Written out explicitly. n n (17. λ(µ). pn ) = Z3 2 G(n) (p1 . λu ) . pn ) . mu . µ) . φ (xn )) −n 2 c c = Z3 T (φu (x1 ) .744 17 Renormalization group and equation 17. . xn ) .45) . . . namely. .3 Renormalization group equation To understand the consequences of the renormalization group of transformations described in the last section. u −n 2 (17. pn ) . . . . u n n (17. .44) which implies that the renormalized 1PI functions in two renormalization prescriptions are related as Γ(n) p. µ′ = Z32 (µ′ ) Γ(n) (p. . pn ) = Z32 Γ(n) (p1 . xn ) = T (φ (x1 ) . p2 . . The connected renormalized Green’s functions of the theory are defined as G(n) (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 . . . let us next look at the Green’s functions of the scalar field theory.41) Such a relation between the renormalized and the unrenormalized Green’s functions of the theory holds true in momentum space as well. . u −n (17. . m(µ). .

Thus. As we have noted this defines the Abelian group of transformations known as the renormalization group.48) that β(λ) = − ǫ + µ dZλ Zλ dµ λ. In dimensional regularization. Since λu is µ-independent (independent of the renormalization prescription). (17.48) −2 where we have identified Zλ = Z1 Z3 . it follows that (see also (17.50) . (17. 0 (17. Here µ0 is assumed to be fixed and µ is the variable renormalization scale.49) where we see from (17. the couplings are defined with an arbitrary mass scale. 0 0 or.17. λu = µ µ0 ǫ λZλ (µ). (17.32).39)) −1 −1 2 µǫ λ = Z1 (µ)Z3 (µ)µǫ λu = µǫ Zλ (µ)λu .47) Both λu and λ are dimensionless and are related as (compare with (16. Let us define the renormalized quartic coupling in the scalar field theory as µǫ λ.46) while the unrenormalized coupling can be defined as (in discussions in earlier chapters we had used µ0 = 1) µǫ λu .19)) µ dλ dµ = β(λ).3 Renormalization group equation 745 which follows from the definition of z in (17. we see that a change in the renormalization scale µ → µ′ gives rise to a finite multiplicative renormalization or a finite group of transformations acting on the vertex functions of the theory. (17. Here we have used p to denote collectively all the independent external momenta of the 1PI vertex function.

are independent of µ because the pole terms (in ǫ) are independent of µ as there are no overlapping divergences in a renormalizable theory. u u −1 (17. 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 + · · · . By dimensional reasoning then they cannot depend on m either. in fact.55) where the constants a. As a result.54) in the MS (MS) scheme. mu held fixed.746 17 Renormalization group and equation ∂Z ∂µ and the derivative on the right-hand side should be thought of as with λu .51) we obtain dm dµ µ where = −mγm .52) γm = (17. c. µ and. s. µ dZ m . . They are. . We note that this is a consequence of the fact that dimensional regularization is a mass independent regularization. . γm = γm (λ).52) and can be solved independently which we will analyze in the next section. the equation for the coupling constant (17. (17. their dependences are of the form (the µ dependence comes in only through the coupling) β = β(λ).53) Both β and γm are finite functions of µ as ǫ → 0 since the divergences in the numerator and denominator cancel out. 2Z m dµ (17. q. λ4 + · · · . b. (17.49) decouples from the mass equation (17. m. . Similarly. t. functions of λ. from −1 m2 = Zm (µ)Z3 (µ)m2 = Z m (µ)m2 . in general.

λu ) = 0.49) and (17. ∂µ ∂λ ∂m (17. m ∂µ (17. λ(µ).57) is known as the renormalization group equation for the 1PI vertex functions and expresses how amplitudes change when the renormalization scale is changed. dµ µ µ ∂ ∂λ(µ) ∂ ∂m(µ) ∂ +µ +µ − nγ Γ(n) = 0. where we have defined (the derivatives are taken with λu .3 Renormalization group equation 747 Let us next look at the 1PI amplitudes (17.56) which leads to the relation (since the unrenormalized quantities are independent of the renormalization prescription) dΓu (p. 2Z3 ∂µ We can also argue in the same way (as we have done for β and γm ) that γ is also finite and is a function of λ alone (in the MS/MS scheme). u −n (17. m(µ). mu fixed.17. µ) = 0.59) Equation (17. λu ) = Z3 2 (µ) Γ(n) (p. ∂µ ∂µ ∂λ(µ) ∂µ ∂m ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) = 0. λ(µ). mu .44) Γ(n) (p. see also (17. m(µ). µ) .52)) ∂λ . µ or. (17. . ∂µ µ ∂m . namely.58) β(λ) = µ γm = − γ = µ ∂Z3 .57) (n) µ or. γ = γ(λ). dµ d −n Z3 2 (µ) Γ(n) (p. or. mu .

57) describes how renormalized amplitudes (1PI vertex functions) change as the renormalization scale is changed ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) (pi .64) . ∂µ ∂λ µ (17.61) β(λ) = µ γm = − γ = µ ∂Z3 .4 Solving the renormalization group equation As we have seen.58) (the derivatives are taken with λu . the renormalization group equation (17. µ) = 0. λ. m. (17.24)) 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 ∂λ ∂m (17.62) Let us note from our earlier discussions (see (16.748 17 Renormalization group and equation 17. 2Z3 ∂µ For simplicity. ∂µ µ ∂m . µ) = 0. λ. 1 2 n (17.63) Consequently. if we define the Mandelstam variable associated with the n-point function as s = p2 + p2 + · · · + p2 . let us set m = 0 in which case. m ∂µ (17. the equation takes the form ∂ ∂ + β(λ) − nγ Γ(n) (pi .60) µ where pi denotes the independent external momenta and as we have seen in (17. mu fixed) ∂λ .

λ.62) can be written as ∂ ∂ + β(λ) − nγ Γ(n) = 0. f (n) is a dimensionless function of independent dimensionless variables which are Lorentz invariant. we can write f (n) pi · pj s .68) f (n) t.69) Solving the renormalization group equation (17.70) .66) so that for a fixed s. t) ∂ρ(x. λ. changing the renormalization scale merely corresponds to changing t. consequently. 2 µ (17.65) where we have used the fact that the n-point vertex function of a scalar field theory is Lorentz invariant and. s (17. λ. µ) = s 4−n 2 √ s. ∂µ ∂µ ∂t ∂t µ (17. µ2 s . ∂t ∂x (17.67) ∂t ∂ s 1 (where from (17.68) now becomes straightforward from the observation that a linear first order differential equation of the form ∂ρ(x.4 Solving the renormalization group equation 749 then. scaling all momenta by a factor of Γ(n) (pi .66) we have used ∂µ = − 1 ∂µ ln µ2 = µ ) so that the 2 renormalization group equation (17. t) + v(x) = L(x)ρ(x. or. ∂t ∂λ and Γ(n) = s 4−n 2 (17. It follows now that ∂t ∂ ∂ ∂ =µ = . 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. pi · pj . (17.17. it can be thought of as scaling the momentum) µ= √ set . t). t = − 1 s ln 2 .

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

80) then. Let us next note that if λ∗ is a fixed point of the coupling constant evolution equation (17. namely. for large α (large momentum). −t + ln α).68) requires solving the characteristic equation dλ = β(λ). (17.17. 1 2 s ln µ2 ) pi → αpi . the leading behavior of the amplitude is given by 4−n 2 Γ(n) → α2 s f (n) (λ(λ. (17. This is known as Weinberg’s theorem (the same theorem which also made its appearance in renormalization theory). )e s (17. (17.77).81) where D Γ(n) represents the superficial degree of divergence of the n-point function and ‘a’ is a constant which can be determined from the explicit form of the amplitude.t′ −ln α)) → α4−n (ln α)a · · · = (n) (α)D(Γ ) (ln α)a · · · . we can obtain Γ(n) = s 4−n 2 f (n) (λ(λ.77) and once we know the solution to this equation.t′ )) . dt (17.78) Note that if we scale all momenta (keeping µ fixed) as s → α2 s.4 Solving the renormalization group equation 751 so that the solution of the renormalization group equation (17. if . × en R ln α −t+ln α pi · pj ) s dt′ γ(λ(λ. −t).79) so that (remember that −t = t → t − ln α. 0 pi · pj n R−t dt′ γ(λ(λ.

under a scaling pi → αpi (17. namely. the n-point amplitude (17.752 17 Renormalization group and equation β (λ∗ ) = 0. if λ∗ = 0. at the fixed point. . = s 2 e 2 ∗ s 4−n 2 f (n) λ∗ .82) (17.83) Near the fixed point.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. (17. s (17. (17. then γ(λ∗ ) = 0. the n-point amplitude will scale. (17. near the fixed point of the renormalization group equation.24)). 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). If the fixed point of the beta function happens to be at the origin.87) (17.86) which follows from the fact that in perturbation theory γ(λ) is a power series in λ.78) behaves as Γ(n) = s pi · pj ntγ(λ∗ ) e s 4−n − n γ(λ∗ ) ln s pi · pj µ2 f (n) λ . as Γ(n) ≃ α4−n(1+γ(λ∗ )) Γ(n) .84) As a result. λ = λ∗ = constant. = s 4−n 2 s µ2 − n γ(λ∗ ) 2 f (n) λ∗ .79). pi · pj . then.

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.4 Solving the renormalization group equation 753 It is clear.90) λ = 0. t = 0) = λ) λ λ dλ′ β (λ′ ) t = 0 dt′ = t. the first thing that we notice is that (17. the behavior of β(λ) is shown in Fig.88) implies that (recall that λ(λ. let us analyze the qualitative behavior of the solution. In perturbation theory the beta function has the form of a power series in the coupling constant β(λ) = β0 λn + O λn+1 . λ(t) → (17. if β(λ) is negative (β < 0. We note that dλ dt = β λ . Qualitatively. In this case. this leads to the behavior of the β function as shown in Fig. (17. therefore. t → ∞. the behavior of the coupling constant is given by ∞. (17.17. 17. Therefore. and λ > 0).91) is a fixed point of the beta function.89) Before solving this equation explicitly. 0.92) On the other hand. 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). the function dλ is monotonically increasing for positive t and we say that dt the origin in the coupling constant space is an infrared fixed point. 17. qualitatively we have . (17.2.3 and in this case.

t → ∞.754 17 Renormalization group and equation β(λ) λ Figure 17. we have two distinct cases to consider (when λ∗ = 0. namely. (17.2: The graph showing that origin in the coupling constant space is an infrared fixed point of the β function. we recover the two cases we have discussed earlier).93) and we say that the origin in the coupling constant space is an ultraviolet fixed point. ∞. Let us next analyze the behavior of the solutions in some detail. t → −∞. . Let us assume that the beta function has a nontrivial fixed point at λ∗ on the positive axis. λ(t) → 0.3: The graph showing that the origin in the coupling constant space is an ultraviolet fixed point of the β function. β(λ) λ Figure 17. then expanding around this fixed point.

2. then as t → −∞.89) leads to . µ → ∞ and when t → −∞. (17. the arrows indicate the direction in which λ moves as t → ∞ (if we start near the fixed point). In the diagram Fig. if β0 < 0. In the second case (see (17.4 (which shows the behavior of a general β function). 17.97) 2. (17. λ → λ∗ . ′−λ ) β0 (λ β0 λ − λ∗ ∗ (17.98) The limits t → ∞ (t → −∞) are known respectively as the ultraviolet (infrared) limits and fixed points which arise in these limits are correspondingly known as ultraviolet (infrared) fixed points (note from (17.89) leads to λ t= λ dλ′ = β(λ′ ) λ λ or. β(λ) = β0 (λ − λ∗ )r + O (λ − λ∗ )r+1 . (17. λ − λ∗ = (λ − λ∗ ) eβ0 t .96) which implies that 1.17. µ → 0 for fixed s).94) r > 1. λ − λ∗ 1 dλ′ = ln . β(λ) = β0 (λ − λ∗ ) + O (λ − λ∗ )2 . if β0 > 0.66) that when t → ∞. λ → λ∗ .95) In the first case (17. (17. then as t → ∞.95)).4 Solving the renormalization group equation 755 1. (17.

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

99) g2 = 1 + 2|β0 |g2 t 1 . therefore. for QED we have from (17. at large momenta almost naive scaling sets in (see discussion after (17.99) e2 = 1 − 2β0 e2 t 1 .87)). In such theories. β0 < 0) we have from (17. it follows that . the coupling becomes weaker and weaker and ultimately vanishes. (17.38)) µ µ0 ǫ λu = λ 1+ 3λ 16π 2 ǫ . as we have seen.101) As t increases. 2β0 e2 t→ (17. Let us recall that with the MS (minimal subtraction) scheme at one loop we have (see (17. At infinite energy. on the other hand.104) Since the unrenormalized (bare) coupling does not depend on the renormalization scale. the theory behaves like a free theory and such theories are called asymptotically free theories. (g∗ = 0.17.46)-(17. we see that the coupling grows and. For QCD. − 2β0 t e2 (t) = 1 e2 (17. identifying the coupling with electric charge g = e and recalling β0 > 0).103) As t → ∞. Such a pole in the evolution of the coupling constant at a particular momentum scale is called the Landau pole. in particular. 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. diverges when 1 .4 Solving the renormalization group equation 757 the only fixed point – for QED it is the infra