BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

This page intentionally left blank

University of Rochester, USA

Ashok Das

World Scientific

NEW JERSEY

•

LONDON

•

SINGAPORE

•

BEIJING

•

SHANGHAI

•

HONG KONG

•

TA I P E I

•

CHENNAI

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

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

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

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

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

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

Printed in Singapore.

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

This page intentionally left blank

Preface

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

viii

Preface

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

Contents

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

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

2

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

3

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

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

x

Contents

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

5

6

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

Contents

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

7

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

8

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

9

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

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

10 Dirac 10.1 10.2 10.3 10.4 10.5 10.6

xii

Contents

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

Contents

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

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

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

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

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

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

This page intentionally left blank

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

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

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

one transforms inversely with respect to the other) and are known respectively as contravariant and covariant vectors. On a more fundamental level.11) From the forms of the contravariant and the covariant vectors in (1.10) as well as using (1. ηµν . xµ = (ct. (1. x). since they satisfy . we can immediately read out the metric tensors for the four dimensional Minkowski space which are diagonal with the signature (+. 0 0 0 −1 η µν (1. namely. η µν .13) The contravariant metric tensor. xµ = ηµν xν . we can write them in the matrix form as 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . −. 0 0 0 −1 1 0 0 0 0 −1 0 0 = 0 0 −1 0 . and the covariant metric tensor. xµ = η µν xν . are inverses of each other. Namely. −x).10) Here c represents the speed of light (necessary to give the same dimension to all the components) and we note that the two four vectors simply represent the two distinct possible ways space and time components can be embedded into the four vector. the two four vectors have distinct transformation properties under Lorentz transformations (in fact. (1. 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. −. −).12) ηµν (1. commonly known as the Minkowski space.11).

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

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

25) (2. (2. for the negative energy solutions we write p0 = E− = −E = − p 2 + m2 . u can be thought of as the independent solution. (2. we note that v must vanish in the rest frame while u remains ˜ ˜ arbitrary.18). 3 (2.9) becomes (E− − m)˜(p) + σ3 p3 v (p) = 0. (2.27) . from the second of the two two-component Dirac equations in (2.32 2 Solutions of the Dirac equation σ3 p 3 v (p) = − ˜ E+ + m and 1 0 = p3 −E + m + 0 .26) Choosing the independent solutions as 1 0 0 1 v (p) = ˜ . Thus.24) and the set of equations (2.22) v (p) = − ˜ σ3 p 3 E+ + m 0 1 = 0 p3 E+ + m . (The question of which components can be chosen independently follows from an examination of the dynamical equations. u ˜ σ3 p3 u(p) + (E− + m)˜(p) = 0. v (p) = ˜ .23) This determines the two positive energy solutions of the Dirac equation (remember that the energy eigenvalues are doubly degenerate). ˜ E− − m (2. for example. Thus. ˜ v We can solve these as u(p) = − ˜ σ3 p 3 v (p).) ˜ Similarly.

0 1 (2.32) .28) u(p) = − ˜ σ3 p 3 E− − m 0 1 = 0 p3 E− − m .27) are reminiscent of the spin up and spin down states of a two component spinor.31) p3 −E − m − 0 ↑ u− (p) = . (2.1 Plane wave solutions 33 we obtain respectively p3 −E − m − 0 σ3 p 3 u(p) = − ˜ E− − m and 1 0 = . 1 0 0 p3 ↓ u− = E− − m . from the fact that we can write σ p ˜ − E 3−3m v (p) − . (2. v (p) ˜ (2. − E+ + m 0 u↓ + = .21) and (2. Thus. The independent two component wave functions in (2.30) u+ (p) = u(p) ˜ . σ3 p 3 ˜ − E + m u(p) + u− (p) = the positive and the negative energy solutions have the explicit forms 1 0 1 0 p3 E+ + m u↑ (p) + 0 = p3 . (2.2.29) and these determine the two negative energy solutions of the Dirac equation.

36) Here α and β are normalization constants to be determined. we can write the solutions for motion along a general direction as u(p) ˜ . (The change in the sign in the dependent components comes from raising the index of the momentum. (We will determine the spin of 2 the Dirac particle shortly. it is also clear that.) 2. relativistic particle of spin s in the presence of both positive and negative energies follows to be 2(2s + 1). u− (p) = σ·p u(p) ˜ E+ + m σ·p ˜ E− − m v (p) . (The correct counting for the number of components of the wave function for a massive. pi = −pi . for the case of general motion. unlike the nonrelativistic counting (2s + 1).33) which can be explicitly veriﬁed.2 Normalization of the wave function Let us note that if we deﬁne E = E+ = p2 + m2 = −E− .34) v (p) ˜ (2. u− (p) = β σ·p ˜ E + m u(p) σ· − E + p v (p) m ˜ v (p) ˜ u+ (p) = α .) From the structure of the wave function.27) respectively are ˜ ˜ normalized as (for the same spin components) . (2. The two component spinors u(p) and v (p) in (2.21) and (2. the solutions take the forms (with p0 = E± = ± p2 + m2 ) u+ (p) = u(p) ˜ . (2. (2. namely. where p1 = p2 = 0.35) then.) 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 2 Solutions of the Dirac equation The notation is suggestive and implies that the wave function corresponds to that of a spin 1 particle.

Similarly.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).38) where we have used the familiar identity satisﬁed by the Pauli matrices (σ · A)(σ · B) = A · B + iσ · (A × B). this product vanishes. we can now calculate σ·p ˜ ˜ u† (p)u+ (p) = α∗ α u† (p) u† (p) E + m + = |α|2 u† (p)˜(p) + u† (p) ˜ u ˜ = |α|2 1 + = |α|2 = |α|2 = p2 (E + m)2 (2.2. ˜ u ˜ v For diﬀerent spin components. Given this. ˜ u E+m (2.2 Normalization of the wave function 35 u† (p)˜(p) = 1 = v † (p)˜(p).39) u† (p)u− (p) − v † (p) ˜ σ·p − E + m v (p) ˜ v (p) ˜ = |β|2 v † (p) ˜ = |β|2 = σ·p σ·p v (p) + v † (p)˜(p) ˜ ˜ v (E + m)2 v † (p)˜(p) ˜ v (2. for the negative energy solutions we have σ·p v = β β −˜† (p) E + m ∗ (2. ˜ v E +m .40) p2 +1 (E + m)2 2E |β|2 v † (p)˜(p).

36 2 Solutions of the Dirac equation It is worth remarking here that although we have seen in (2. for example. In dealing with the Dirac equation.37) that. for the same spin components. another wave function (known as the adjoint spinor) that plays an important role is deﬁned to be u(p) = u† (p)γ 0 . (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.40) simply because we have not speciﬁed their spin components.42) −˜† (p) . u† u = 1 = v † v .38) and (2. σ·p v u− (p) = β ∗ −˜† (p) E + m σ·p v = β ∗ −˜† (p) E + m v † (p) ˜ ½ 0 0 −½ (2. we have carried along ˜ ˜ ˜ ˜ these factors in (2.41) σ·p ˜ ˜ u+ (p) = α∗ u† (p) u† (p) E + m ½ 0 0 −½ σ·p ˜ u = α∗ u† (p) −˜† (p) E + m . Thus. We can also calculate the product σ·p ˜ u u+ (p)u+ (p) = α∗ α u† (p) −˜† (p) E + m = |α|2 u† (p)˜(p) − u† (p) ˜ u ˜ = |α|2 1 − = |α|2 u(p) ˜ σ·p ˜ E + m u(p) σ·p σ·k u(p) ˜ (E + m)2 p2 u† (p)˜(p) ˜ u (E + m)2 E 2 + m2 + 2Em − p2 † u (p)˜(p) ˜ u (E + m)2 . v Thus.

in the rest frame of the particle. m † d3 x ψp (x)ψp′ (x) = (2. − m u† (p)u+ (p) = + (2. ψp (x) = e−ik·x u(p) with p0 = ±E. a relativistically covariant normalization would be to require (for the same spin components) E = u† (p)u− (p).2 Normalization of the wave function 37 = |α|2 = 2m(E + m) † u (p)˜(p) ˜ u (E + m)2 (2.40) (for the same spin components when (2. + − (2. as in the non-relativistic case. By the way. must be positive.) The independent wave functions for a free particle. we can show that 2m |β|2 v † (p)˜(p). the motivation for such a normalization condition comes from the fact that. In fact. ˜ u E +m Similarly. with this normalization condition.38) and (2. Thus. this is not a relativistic normalization. would give (for the same spin components) E (2π)3 δ3 (p − p′ ). therefore. 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.46).43) 2m |α|2 u† (p)˜(p). we determine from (2.44) Our naive instinct will be to normalize the wave function.47) With the requirement (2.2.37) holds) . by requiring (for the same spin components) u† (p)u+ (p) = 1 = u† (p)u− (p). as we will see.45) However. ˜ v E+m u− (p)u− (p) = − (2. u† u. as we will see shortly. is related to the probability density which transforms like the time component of a four vector.

Let us also note here that by construction the positive and the negative energy solutions are orthogonal. E+m m E+m . σ·p ˜ E + m u(p) σ·p − E + m v (p) ˜ . 2m (2. with this normalization. appears to be a Lorentz invariant (scalar) and we will see later that this is indeed true.43) and (2. with this normalization.38 2 Solutions of the Dirac equation u† (p)u+ (p) = + or.48) or. α = α∗ = 2E E |α|2 = . we will obtain from (2. u† (p)u− (p) = − E+m m β = β∗ = E+m . Therefore. 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) ˜ . . v (p) ˜ (2. 2m E 2E |β|2 = . For example. therefore.49) It is also clear that. E+m E + m 2m 2m 2m E + m |β|2 = − = −1.44) (for the same spin components) u+ (p)u+ (p) = 2m 2m E + m |α|2 = = 1.50) u− (p)u− (p) = − This particular product. E+m E + m 2m (2.

however. ˜ E+m u+ (p)u− (p) = −2α∗ β u† (p) ˜ (2.51) Therefore. − + (2. we will obtain from (2. (There is no rest frame for a massless particle. that σ·p v (p) = 0.40) (for the same spin components) u† (p)u+ (p) = + or. an alternative normalization which works well for both massive and massless particles is given by u† (p)u+ (p) = E = u† (p)u− (p). 2 2E |β|2 = E.) The probability density has to be well deﬁned.2. let us note that it becomes meaningless for massless particles.38) and (2. Correspondingly. In this case. ˜ E+m σ·p u− (p)u+ (p) = −2β ∗ α v † (p) ˜ u(p) = 0. orthonormal solutions of the Dirac equation.54) or. u† (p)u− (p) = − E+m β = β∗ = E+m . 2 (2. .2 Normalization of the wave function 39 u† (p)u− (p) + = α∗ β = 0.52) While we will be using this particular normalization for massive particles.53) This still behaves like the time component of a four vector (m is a Lorentz scalar). Note. =α β ∗ u† (p) ˜ u† (p) ˜ σ·p E+m σ·p − E + m v (p) ˜ v (p) ˜ − u† (p) ˜ σ·p σ·p v (p) + u† (p) ˜ ˜ v (p) ˜ E+m E+m (2. E+m E+m . α = α∗ = 2E |α|2 = E. the solutions we have constructed correspond to four linearly independent.

(2. Let us note here parenthetically that.55) which vanishes for a massless particle. That this is indeed true 2 can be seen explicitly as follows.57) Namely. We can write the solution of the Dirac equation for a general motion (along an arbitrary direction) as d4 p a(p)δ(p2 − m2 ) e−ip·x u(p).56) where a(p) is a coeﬃcient which depends on the normalization of u(p) in such a way that the wave function would lead to a total probability normalized to unity. 2.40 2 Solutions of the Dirac equation Correspondingly. ψ(x) = (2. Let us note once again that this is a particularly convenient normalization for massless particles. in this case. d3 x ψ † (x)ψ(x) = 1. we obtain 2m |α|2 = m. it can be understood in light of what we have already pointed out earlier as follows. a particular choice of normalization for the u(p) is compensated for by a speciﬁc choice of the coeﬃcient function a(p) so that the total probability integrates to unity. The true normalization is really contained in the total probability. 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 . we will use the notations of three dimensional . This product continues to be a scalar. E +m 2m |β|2 = −m. while the arbitrariness in the normalization of u(p) may seem strange. u− (p)u− (p) = − E+m u+ (p)u+ (p) = (2.3 Spin of the Dirac particle As we have argued repeatedly. Let us deﬁne a four dimensional generalization of the Pauli matrices as (in this section.

(2. αi . let us note that .59) and write αi = ραi = αi ρ. 3. however.59) ρ= 0 ½ ½ 0 . that this deﬁnes a reducible representation of spin generators since the matrices are block diagonal. of course.58) It can. 2.2.101) through the relation 0 σi σi 0 αi = where = ρ˜ i = αi ρ.) Furthermore. αj ˜ ˜ = = This shows that 1 αi satisﬁes the angular momentum algebra (remem2˜ ber = 1) and this is why we call the matrices. (2. ˜ (2.98) and (1. the generalized ˜ Pauli matrices. σj ] 0 2iǫijk σk = 2iǫijk αk .61) From the structures of the matrices αi and αi we conclude that ˜ [σi . α ˜ (2. be checked readily that this is related to the αi matrices deﬁned in (1. ˜ (2.3 Spin of the Dirac particle 41 Euclidean space since we will be dealing only with three dimensional vectors) σi 0 0 σi αi = ˜ .60) We note that ρ2 = ½ so that we can invert the deﬁning relation (2. σj ] 0 2iǫijk σk 0 0 [σi . (Note. i = 1.62) αi .

(2. therefore. ½] 0 0 0 σi σj − σj σi 0 = [σi . β ˜ = = 0. the angular momentum operators which generate rotations should commute with the Dirac Hamiltonian. In other words. let us look at the free Dirac Hamiltonian in (1. αi .64) As we will see in the next chapter.42 2 Solutions of the Dirac equation αi .65) Calculating the commutator of this operator with the Dirac Hamiltonian. Lorentz transformations deﬁne a symmetry of the Dirac Hamiltonian and. k = 1.100) (remember that we are using three dimensional Euclidean notations in this section) H = α · p + βm = αi pi + βm. Consequently. we obtain . i. j. αj ˜ = σi 0 0 σi 0 0 σj σj 0 − 0 σj σj 0 σi 0 0 σi = σi σj − σj σi 0 [σi . rotations which correspond to a subset of the Lorentz transformations must also be a symmetry of the Dirac Hamiltonian. (2. the Dirac equation transforms covariantly under a Lorentz transformation. σj ] = 2iǫijk σk 0 −[σi . Let us recall that the orbital angular momentum operator is given by Li = ǫijk xj pk .63) With these relations at our disposal. ½] = 2iǫijk αk . (2. 2. 3. σj ] 0 2iǫijk σk [σi .

˜ 2 Ji = Li + (2. β m ˜ (2. H ˜ 2 1 αi . αj pj + αi . αℓ pℓ + βm] = [ǫijk xj pk . the total angular momentum which should commute with the Hamiltonian. we can identify the spin angular momentum operator with . αℓ pℓ ] = ǫijk αℓ [xj . can be identiﬁed with 1 αi . H] + = iǫijk αj pk − iǫijk αj pk = 0.67) = 2iǫijk αk pj = −2iǫijk αj pk .68) In other words. 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. we note that the orbital angular momentum operator does not commute with the Dirac Hamiltonian. so that combining this relation with (2. we note that αi . H] = [ǫijk xj pk . H ˜ = = αi . Thus. if rotations are a symmetry of the system. pℓ ]pk = ǫijk αℓ (iδjℓ ) pk = iǫijk αj pk . (2. αj pj + βm ˜ ˜ αi .66) we obtain 1 αi . (2.2. H ˜ 2 Li + = [Li . Consequently.69) In this case.66) Here we have used the fact that since β is a constant matrix and m is a constant. therefore. To determine the spin angular momentum.3 Spin of the Dirac particle 43 [Li .

4 Continuity equation The Dirac equation. 2 2.44 2 Solutions of the Dirac equation Si = 1 αi .72) with ψ † on the left and (2.73) with ψ on the right and subtracting the second from the ﬁrst.74) This is the continuity equation for the probability current density associated with the Dirac equation and we note that we can identify . (2.72) Taking the Hermitian conjugate of this equation.71) which has doubly degenerate eigenvalues ± 1 .70) Note. ∂t i (2. written in the Hamiltonian form. in particular. we obtain ← − ∂ψ † = ψ † iα · ∇ + βm). is given by ∂ψ = Hψ = − iα · ∇ + βm ψ. ∂t ∂t iψ † or. that 1 1 α3 = ˜ 2 2 σ3 0 0 σ3 S3 = . ∂t ∂ (ψ † ψ) = −∇ · ψ † αψ . i ∂ (ψ † ψ) = −i∇ · ψ † αψ . Therefore. ˜ 2 (2. ∂t (2. ∂t −i (2. we conclude 2 that the particle described by the Dirac equation corresponds to a spin 1 particle.73) where the gradient is assumed to act on ψ † . we obtain ∂ψ † ∂ψ +i ψ = −i ψ † α · ∇ψ + (∇ψ † ) · αψ . or. Multiplying (2.

must transform like the time coordinate under a Lorentz transformation. of course.2.) On the other hand. ψ † αψ). J) = (ψ † ψ.79) is a constant independent of any particular Lorentz frame. (We are. to write the continuity equation as ∂ρ = −∇ · J. yet to show that j µ transforms like a four vector which we will do in the next chapter.80) . the total probability d3 x ρ = d3 x ψ † ψ.76) so that the continuity equation can be written in the manifestly covariant form ∂µ J µ = 0. ρ. J = ψ † αψ = probability current density. and note that the Hermitian conjugate of ψ satisﬁes (2.77) (2. An alternative and more covariant way of deriving the continuity equation is to start with the covariant Dirac equation iγ µ ∂µ − m ψ = 0. (2. (2.75) (2. It is worth recalling that we have already used this Lorentz transformation property of ρ in deﬁning the normalization of the wave function.4 Continuity equation 45 ρ = ψ † ψ = probability density. P = (2.77)) and. is the time component of j µ (see (2.78) This. ∂t This suggests that we can write the current four vector as J µ = (ρ. shows that the probability density. in fact. therefore.

84) This is. we obtain ← − ψ − iγ 0 (γ µ )† γ 0 ∂µ − m) = 0. (2. 1. the covariant continuity equation and we can identify J µ = ψγ µ ψ. ← − ψ − iγ µ ∂µ − m = 0. ∂µ ψγ µ ψ = 0. (2.46 2 Solutions of the Dirac equation ← − ψ † − i(γ µ )† ∂µ − m = 0. or. γ 0 (γ µ )† γ 0 = γ µ . in fact.85) . we obtain ← − i ψγ µ ∂µ ψ + ψγ µ ∂µ ψ = 0. or.80) with ψ on the left and (2. 2. or.85) that J 0 = ψγ 0 ψ = ψ † γ 0 γ 0 ψ = ψ † ψ = ρ.82) where we have used the property of the gamma matrices that (for µ = 0. which is what we had derived earlier in (2. (2.83) Multiplying (2.81) Multiplying this equation with γ 0 on the right and using the fact that (γ 0 )2 = ½.86) (2. J = ψγψ = ψ † γ 0 γψ = ψ † αψ. i∂µ ψγ µ ψ = 0.82) with ψ on the right and subtracting the second from the ﬁrst.77). Note from the deﬁnition in (2. (2. (2. 3) γ 0 γ µ γ 0 = (γ µ )† .

even for this simple case of a free particle the energy spectrum has the form shown in Fig. Even when the probability density is consistently deﬁned. in such a case.1: Energy spectrum for a free Dirac particle. the probability density is independent of time derivative much like the Schr¨dinger equation. (2. because it is a ﬁrst order equation (particularly in the time derivative). 2. First of all. m −m Figure 2. can be deﬁned to be positive deﬁnite even in the presence of negative energy solutions.87) Thus. the presence of negative energy solutions leads to many conceptual diﬃculties. the energy eigenvalues are given by p0 = E± = ±E = ± p2 + m2 .2.38) and (2.1. 2. 2. This is rather diﬀerent from the case of the Klein-Gordon equation that we have studied in chapter 1.5 Dirac’s hole theory We have seen that Dirac’s equation leads to both positive and negative energy solutions. We note from Fig. Consequently.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).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. as we o have seen explicitly in (2. for example.40). we note that the energy spectrum . In the free particle case. the probability density.

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

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

To obtain the other basis matrices.93) .92) We recognize from (2. (2. Of course.50 2 Solutions of the Dirac equation Since these are 4 × 4 matrices.91).90) i ǫµνλρ γ µ γ ν γ λ γ ρ . the identity matrix will correspond to one of them. Note that in our particular representation for the γ µ matrices given in (1. Note that. (2. Let γ5 = iγ 0 γ 1 γ 2 γ 3 = − where ǫ0123 = 1 = −ǫ0123 . † γ5 = γ5 . 2 γ5 = ½. by deﬁnition. 4! (2. (2.89) represents the four-dimensional generalization of the Levi-Civita tensor.91) that we can identify this with the matrix ρ deﬁned earlier in (2.91) where we have used the property of the Pauli matrices σ1 σ2 σ3 = i½. a complete set of Dirac matrices must consist of 16 such matrices. let us deﬁne the following sets of matrices. (2. we obtain γ5 = i ½ 0 σ1 0 0 −σ1 σ1 0 0 0 −σ2 −σ2 σ3 σ2 0 0 −σ3 σ3 0 0 −½ σ1 0 0 −σ1 σ2 σ3 0 −i½ −i½ 0 = i −σ2 σ3 0 = i −σ1 σ2 σ3 0 0 = i = ½ ½ 0 .60).

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

98) we conclude. 4. From (2. therefore. σ µν . 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 . (This relation can be obtained from (2. in fact. ǫijk ǫℓjk = 2 δiℓ . Γ(S) Γ(V ) Γ(A) = = ½. 1. γµ.70) that the matrices 2 αi represent the ˜ spin operators for the Dirac particle. that the matrices 1 1 αi = ǫijk σ jk . 1.100) Γ(T ) = Γ(P ) = = γ5 where the numbers on the right denote the number of matrices in each category and these. the notation is suggestive and stands for the fact that ψΓ(S) ψ . 4. namely. ˜ (2.52 2 Solutions of the Dirac equation σ 0i = iγ 0 γ i = i 0 iσi iσi 0 ½ 0 0 −σi σi 0 0 −½ = iαi . provide a basis for all the 4 × 4 matrices. γ5 .99) can be identiﬁed with the spin operators for the Dirac particle.) We have thus constructed a set of sixteen Dirac matrices.98) 1 We have already seen in (2. ˜ 2 4 (2. ¯ Here. (2. γµ.98) using the identity for products of Levi-Civita tensors.

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

γν + + = 2i η µν γ λ − η µλ γ ν . we have used the fact that (2. BC] = [A. In this derivation.106) [A. = 0.108) However. C + . γλ = i γ µ. B] C + B [A.107) is more useful for our purpose. (2.54 2 Solutions of the Dirac equation γµ.107) can also be expressed in terms of commutators (instead of anti-commutators) as [A. γ µ . (2. (2. Similarly. γ ν γ λ = i γµ. γ µ + + = 2η µν ½. γν γ5 . C] . For example.107) We note here parenthetically that the commutator in (2. B + C − B A.105) We can also calculate various other commutation relations in a straightforward and representation independent manner. we obtain . σ νλ = γ µ . −i η νλ − γ ν γ λ γλ − γν γµ. for the commutator of two σ µν matrices. BC] = ABC − BCA = (AB + BA)C − B(AC + CA) = A. the form in (2. since γ µ matrices satisfy simple anti-commutation relations.

(2. for example. σ λρ = i γ µ γ ν . The various commutation and anti-commutation relations also lead to many algebraic simpliﬁcations in dealing with such matrices. σ λρ = iγ µ γ ν . This becomes particularly useful in calculating various amplitudes involving Dirac particles. (these relations are true only in 4-dimensions) γµ γ ν γ µ = γµ γ ν .110) (2. σ λρ γ ν = iγ µ 2iη νλ γ ρ − 2iη νρ γ λ + i 2iη µλ γ ρ − 2iη µρ γ λ γ ν = −2i η µλ i(η νρ − γ ρ γ ν ) + η νρ − i(η µλ − γ µ γ λ ) −η µρ i(η νλ − γ λ γ ν ) − η νλ (−i(η µρ − γ µ γ ρ ) = −2i η µλ σ νρ + η νρ σ µλ − η µρ σ νλ − η νλ σ µρ . + ν − γµγν (2.2.6 Properties of the Dirac matrices 55 σ µν . Similarly. Thus.109) Thus. and it follows now that. where we have used (γµ = ηµν γ ν ) γµ γ µ = 4 ½.112) . 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.111) (2. γµ A µ = γµ Aν γ ν γ µ = Aν γµ γ ν γ µ = −2Aν γ ν = −2A /γ /. σ λρ + i γ µ . 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. σ λρ = − i η µν − γ µ γ ν .

= 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ − Tr γ ν γ λ γ ρ γ µ . The commutation and anticommutation relations also come in handy when we are evaluating traces of products of such matrices. we know from the cyclicity of traces that Tr γ µ γ ν = Tr γ ν γ µ .114) Tr γ µ γ ν + Tr γ5 γ µ (2. or. (2. For example. 2 Tr γ µ γ ν γ λ γ ρ = 8η µν η λρ − 8η µλ η νρ + 8η νλ η µρ . we note that Tr γ µ γ ν γ λ γ ρ = Tr γµ. γ λ + = 8η µν η λρ − 8η µλ η νρ + Tr γ ν γ λ γ µ . 2 = Tr γ µ γ5 = −Tr γ5 γ µ = 0.116) . 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 . γ ν 2 2 1 = Tr 2η µν ½ = η µν Tr ½ = 4η µν . = (2.115) Even more complicated traces can be evaluated by using the basic relations we have developed so far. (2. For example. − γργµ Tr γ µ γ ν γ λ γ ρ = 4 η µν η λρ − η µλ η νρ + η νλ η µρ .113) = 2η λµ γµ γ ν + 2γ ν γ λ = 2γ λ γ ν + 2γ ν γ λ = 2 γ λ . γ µ + − γµγλ = 4η λν ½. it follows (in 4-dimensions) that 1 1 Tr γ µ γ ν + Tr γ ν γ µ = Tr γ µ .56 2 Solutions of the Dirac equation γµ γ ν γ λ γ µ = γµ γ ν γ λ . γ ρ or. Therefore. γ ν + and so on.

This is known as the Majorana representation and is quite useful in the study of Majorana fermions which are charge neutral fermions. 1 S = √ γ0 2 ½ + γ2 .119) σ µ = (½.117) It can be checked that the Dirac matrices in the Pauli-Dirac representation and the Majorana representation are related by the similarity (unitary) transformation µ γM = Sγ µ S −1 . (2. To conclude this section. ˜ (2. For example. It can be checked . However. let us note that we have constructed a particular representation for the Dirac matrices commonly known as the Pauli-Dirac representation. 3 γM = .6 Properties of the Dirac matrices 57 and so on.120) This is known as the Weyl representation for the Dirac matrices and is quite useful in the study of massless fermions. σ µ = (½. (2. there exists yet another representation for the γ µ matrices. 1 γM = . (2. −σ). there are other equivalent representations possible which may be more useful for a particular system under study. σ). the µ γM matrices have the forms 0 σ2 0 σ2 σ2 0 −σ2 0 iσ3 0 −iσ1 0 0 iσ3 0 −iσ1 0 γM = . Explicitly. We would use all these properties in the next chapter to study the covariance of the Dirac equation under a Lorentz transformation. namely. there exists a representation for the Dirac matrices where γ µ are all purely imaginary. 2 γM = .118) Similarly.2. 0 σµ ˜ σµ 0 γW = where µ .

it can be easily checked that Tr Γ(a) Γ(b) = 0. γµ . γ5 } . given this set of matrices. a = b. As a result.122) From the properties of the γ µ matrices. For example.126) With this. γ5 γ µ .122) is straightforward. it follows now that if .1 Fierz rearrangement. the linear independence of the set of matrices in (2. γ5 } . σ µν . A.123) where “Tr” denotes trace over the matrix indices. 4 (b) (2. (2. γ µ . P deﬁne a complete basis for 4 × 4 matrices.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 . 1 S=√ 2 ½ + γ5 γ 0 . −γ5 γµ .121) 2. T.124) Tr Γ(a) Γ(b) = δ(a) . Explicitly. (2. the sixteen Dirac matrices Γ(a) . As we have pointed out in (2. we can construct the inverse set of matrices as Γ(a) . V. We have explicitly constructed the sixteen matrices to correspond to the set Γ(a) = {½. Tr Γ(a) Γ(a) Γ(a) = such that a not summed. (2.6. (2. a = S. σµν . This is easily demonstrated by showing that they are linearly independent which is seen as follows.125) Γ(a) = (2. we can write the inverse set of matrices as 1 {½.100).

C(b) = 0.127) then.127) implies that all the coeﬃcients of expansion must vanishing which shows that the set of sixteen matrices Γ(a) in (2. V. Therefore. A. Substituting (2. (a) (2. (2. we obtain C(a) Tr Γ(b) Γ(a) = (a) (a) (M ) (M ) (a) Tr Γ(b) M = or. T. any arbitrary 4 × 4 matrix M can be expanded as a linear superposition of these matrices.129).128) for any b = S.127) with Γ(b) . M= (a) C(a) Γ(a) .6 Properties of the Dirac matrices 59 C(a) Γ(a) = 0.122) are linearly independent.129) Multiplying this expression with Γ(b) and taking trace over the matrix indices.125) we obtain C(a) Γ(a) = 0. As a result they constitute a basis for 4 × 4 matrices. P . (2.130) into the expansion (2.2. where b is arbitrary. multiplying (2. (a) (a) or.122) provide a basis for the 4 × 4 matrix space. (M ) (2. namely. (M ) C(a) δ(b) . and taking trace over the matrix indices and using (2. (2. (a) C(a) Tr Γ(b) Γ(a) = 0. C(a) δ(b) = 0. we obtain . or. Since the set of matrices in (2.130) C(b) = Tr Γ(b) M . (a) Tr Γ(b) or.

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

124) respectively.135) equivalently as . we have assumed that the spinors are ordinary functions. Let us note that using the explicit forms for Γ(a) and Γ(a) in (2. In deriving these identities. we can write the ﬁrst of the Fierz rearrangement identities in (2. (2. N and any spinors.135) Since this is true for any matrices M. (2.122) and (2.134) pick up a negative sign which arises from commuting the fermionic ﬁelds past one another.134) The two relations in (2.134) are known as the Fierz rearrangement identities which are very useful in calculating cross sections. if they correspond to anti-commuting fermion operators.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 can ˜ deﬁne a new spinor ψ4 = N ψ4 to write the identity in (2.6 Properties of the Dirac matrices 61 ¯ ψ1 M Γ(a) ψ4 ¯ ψ3 N Γ(a) ψ2 αδ ¯ = ψ1α M Γ(a) ¯ ψ4δ ψ3γ N Γ(a) αδ γβ ψ2β γβ ¯ ¯ = ψ1α ψ4δ ψ3γ ψ2β M Γ(a) N Γ(a) ¯ ¯ ¯ = ψ1α ψ4δ ψ3γ ψ2β Mαβ Nγδ = ψ1 M ψ2 ¯ ψ1 Γ(a) M ψ4 ¯ ψ3 Γ(a) N ψ2 γβ ¯ ψ3 N ψ4 . ¯ = ψ1γ Γ(a) M ¯ ψ4β ψ3α Γ(a) N γβ αδ ψ2δ αδ ¯ ¯ = ψ1γ ψ4β ψ3α ψ2δ Γ(a) M Γ(a) N ¯ ¯ ¯ = ψ1γ ψ4β ψ3α ψ2δ Mαβ Nγδ = ψ3 M ψ4 ¯ ψ1 N ψ2 .2. the right-hand sides of the identities in (2. On the other hand.

D. choose M = (½ − γ5 ) γ µ . Drell. McGraw-Hill. Relativistic Quantum Mechanics. Bjorken and S.137) then using various properties of the gamma matrices derived earlier as well as (2. 1964. 2.136) ¯ ¯ ψ1 (½ − γ5 ) γ µ ψ2 ψ3 (½ − γ5 ) γµ ψ4 ¯ ¯ = −ψ1 (½ − γ5 ) γ µ ψ4 ψ3 (½ − γ5 ) γµ ψ2 .110) and (2. Schiﬀ. for example.113). we obtain from (2.138) This is the well known fact from the weak interactions that the V −A form of the weak interaction Hamiltonian proposed by Sudarshan and Marshak is form invariant under a Fierz rearrangement (the negative sign is there simply because we are considering spinor functions and will be absent for anti-commuting fermion ﬁelds).7 References 1. (2. Quantum Mechanics. New York. New York. L. McGraw-Hill.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 . J. (2. if we which is often calculationally simpler. I. N = (½ − γ5 ) γµ .136) Thus. 2. (2. 1968. .

Zuber. 6. 1860 (1958). Itzykson and J-B. I. 5. 4.7 References 63 3. Lectures on Quantum Mechanics. Real representations of ﬁnite Cliﬀord algebras. Physical Review 109. India. Okubo. Das. E. E. 1657 (1991). New York. Quantum Field Theory. S. 1980. Classiﬁcation. (1957). Sudarshan and R. Hindustan Publishing. . Journal of Mathematical Physics 32. A. G. 2003. McGrawHill.2. C. Proceedings of PaduaVenice conference on mesons and newly discovered particles. C. New Delhi. Marshak.

This page intentionally left blank .

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

Under a Lorentz boost along the x-axis. 3 x′ = ǫx1 + x2 . We observe here that the matrix representing the inﬁnitesimal change under a rotation is anti-symmetric. We also note from (3. x′ 3 = x3 . (3. we also know that the coordinates transform as (boost velocity v = β since c = 1) xµ → x′ µ .7) We recognize that (3. R (RT R = ½).2) that an inﬁnitesimal rotation around the 3-axis (z-axis) takes the form x′ = x1 − ǫx2 .4) where we have identiﬁed θ = ǫ = inﬁnitesimal. x′ 1 = −γβx0 + γx1 . 1 x′ = x3 . with unit determinant.66 3 Properties of the Dirac equation rotation around the 3-axis (z-axis) or in the 1-2 plane is denoted by an orthogonal matrix. 2 (3.6) where the Lorentz factor γ is deﬁned in terms of the boost velocity to be 1 1 − β2 γ= . such that (3. x′ 2 = x2 .6) can also be written in the matrix form as .5) x′ 0 = γx0 − γβx1 . (3.

= 0 0 1 0 x2 0 0 0 1 x3 x′ 0 γ (3.8) 1 1 − β2 β .3. 1 − β2 1 − β2 (3.12) . we note that a Lorentz boost along the x-direction can be written as x′ µ = Λµν xν . we conclude from (3. Thus. (3.10) Λµν − sinh ω = 0 0 0 0 .11) 1 β2 − = 1. where cosh ω − sinh ω 0 0 cosh ω 0 0 (3.9) 1 − β2 cosh2 ω − sinh2 ω = Since the range of the boost velocity is given by −1 ≤ β ≤ 1.9) that −∞ ≤ ω ≤ ∞. 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 .

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. (Incidentally. as we have seen. 0 1 0 0 0 1 sinh ω (3. = 0 0 1 0 0 0 0 1 1 Λµν (3. (3. ω.) Therefore. (That these rotations become complex is related to the fact that the metric has opposite signature for time and space components. we can obtain. 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). Lorentz boosts correspond to noncompact transformations unlike space rotations. (3. as a result.13) The matrix representing the Lorentz transformation of the coordinates. x′ = Λµν xν .15) and also has a unit determinant. Let us ﬁnally note that if we are considering an inﬁnitesimal Lorentz boost along the 1-axis (or a rotation in the 0-1 plane).16) .) Furthermore.15) also shows that the covariant vector transforms in an inverse manner compared with the contravariant vector. µ (3.12) or (3. is easily seen from (3. much like the rotation matrix R in (3.68 From this. namely. (or the parameter of boost) can take any real value and.14) sinh ω = 0 0 0 0 cosh ω 0 0 . then we can write (ω = ǫ = inﬁnitesimal) −ǫ 0 0 −ǫ 1 0 0 = δµν + ǫµν .13) to be an ν orthogonal matrix in the sense that µ (ΛT )µν Λνλ = Λνµ Λνλ = δλ . Λµ (or Λµν ). the “angle” of rotation.3).

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

or. or. Lorentz transformations deﬁne the maximal symmetry of the space-time manifold which leaves the origin invariant. orthogonal means the corner that is straight up (perpendicular). then the transformation is called orthochronous. we note that ΛT as can be seen from (3. orthochronous means straight up in time. 2 = 1 + Λi 0 ≥ 1.23) = Λνµ and ΛT ν µ = Λνµ The set of homogeneous Lorentz transformations satisfying Λ00 ≥ 1.21) Therefore. Λ00 Λ00 2 2 − Λi 0 2 = 1. Incidentally. therefore. Λ00 ≤ −1. “gonia” in Greek means an angle or a corner and. such a Lorentz transformation does not change the direction of time. we can write out the relation (3. (3. (3. (3. of course. or.25) .22) If Λ00 ≥ 1.20) explicitly as Λ00 Λ00 + Λi 0 Λi 0 = 1. or.24) µ ν (3. Namely. and (3.15). Thus. we conclude that Λ00 ≥ 1. (The Greek preﬁx “ortho” means straight up.20)) (det Λ)2 = 1. det Λ = ±1.) Note also that since ΛT Λ = ½. we obtain (for clarity.70 3 Properties of the Dirac equation This is. what we have seen before in (3. Choosing ρ = σ = 0. an orthodontist is someone who can make your teeth straight. In the same spirit. det Λ = 1.

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

(3. 0 −1 0 (3. we say that it is covariant under a given transformation provided the transformed equation has the form L′ ψ ′ = 0. (3. These additional transformations. (3. the operator L.2 Covariance of the Dirac equation Given any dynamical equation of the form Lψ = 0. Under a Lorentz transformation xµ → x′ µ = Λµν xν . however.29) where L is a linear operator.31) where ψ ′ represents the transformed wavefunction and L′ stands for the transformed operator (namely. In simple terms.30) −1 0 0 0 −1 0 = 0 0 −1 0 0 0 0 Λprop .33) (3. 3.32) . we would refer to proper orthochronous Lorentz transformations as the Lorentz transformations. In these lectures. With this general deﬁnition.72 3 Properties of the Dirac equation Λ = Λspace−time Λprop with Λ00 ≤ −1 and det Λ = 1 (which is proper but not orthochronous). cannot be continuously connected to the identity matrix since they involve discrete reﬂections. with the transformed variables). covariance implies that a given equation is form invariant under a particular transformation (has the same form in diﬀerent reference frames). let us now consider the Dirac equation iγ µ ∂µ − m ψ(x) = 0.

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

(3.40) (3. from (3. so that.33) that ∂x′ µ = Λµν . on the wave function.37) where the wave function is recognized to be ψ(x) = x|ψ . or.35)) ψ ′ (x′ ) = x′ |ψ ′ = S(Λ)ψ(x). or. we can write iγ µ ∂µ − m ψ(x) = 0. Let us also note from (3.38) Namely. ∂xν deﬁne a set of real quantities.42) . ′ iΛµν γ ν ∂µ − m S −1 (Λ)ψ ′ (x′ ) = 0. A more physical way to understand this is to note that the Dirac wave function simply consists of four functions which do not change. Thus.39) (3. iγ µ (3.41) ∂x′ ν ∂ − m S −1 (Λ)ψ ′ (x′ ) = 0.35) we can write ψ(x) = S −1 (Λ)ψ ′ (x′ ).37) we obtain (see (3. ′ iΛµν Sγ ν S −1 ∂µ − m ψ ′ (x′ ) = 0. Since the Lorentz transformations are invertible. can be represented by a matrix S(Λ) which depends only on the parameter of transformation Λ and not on the space-time coordinates. (3. but get rotated by the S(Λ) matrix. (3. or. the matrix S(Λ) must possess an inverse so that from (3. ∂xµ ∂x′ ν iγ µ Λνµ ∂ν′ − m S −1 (Λ)ψ ′ (x′ ) = 0.74 = 3 Properties of the Dirac equation dx |Λx S(Λ)ψ(x). the eﬀect of the Lorentz transformation. or.

Λµν γ ν = S −1 γ µ S. (Sγ 0 S † γ 0 )γ µ (Sγ 0 S † γ 0 )−1 = γ µ . Next. or. let us note that since the parameters of Lorentz transformation are real (namely.43) Let us note that if we deﬁne γ ′ µ = Λµν γ ν . Λνσ γ σ (3. there must exist a matrix connecting the two representations.45) where we have used the orthogonality of the Lorentz transformations (see (3.15)). γ ′ µ. γ µ and γ ′µ . the matrices γ ′ µ satisfy the Cliﬀord algebra and.40).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. such that Λµν Sγ ν S −1 = γ µ . (3.2 Covariance of the Dirac equation 75 where we have used (3. (3. It now follows from (3.42) that the Dirac equation will be form invariant (covariant) under a Lorentz transformation provided there exists a matrix S(Λ). γ 0 (S −1 γ µ S)† γ 0 = S −1 γ µ S. we see from (3.46) † † † . γ σ = 2Λµρ Λνρ ½ = 2η = Λµρ Λνσ 2η ρσ ½ ½.44) + = + + µν = Λµρ Λνσ γ ρ . Sγ 0 S † γ 0 γ 0 γ µ γ 0 γ 0 S −1 γ 0 S −1 = γ µ . γ ′ ν Λµρ γ ρ . or. Therefore. by Pauli’s fundamental theorem. (Λ∗ )µν = Λµν ) γ 0 Λµν γ ν γ 0 = Λµν γ 0 (γ ν )† γ 0 = Λµν γ ν . Therefore. then. generating Lorentz transformations (for the Dirac wavefunction). (3.3. or. or.

we can determine the unique value of b in the following way. As a result. (3. or. or. b4 = 1.47).100) and calculating the commutator with γ µ ). b = b∗ . determines that the parameter b is real. 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. (3. Sγ 0 S † γ 0 = b∗ ½ = b ½. we can denote Sγ 0 S † γ 0 = b ½.49) We also note that det γ 0 = 1 and since we are interested in proper Lorentz transformations.46) that the matrix Sγ 0 S † γ 0 commutes with all the γ µ matrices and. namely.47) Taking the Hermitian conjugate of (3. It is clear from (3. (3. we obtain (Sγ 0 S † γ 0 )† = b∗ ½. Using these in (3.43) and the relations (γ 0 )† = γ 0 = (γ 0 )−1 as well as γ 0 (γ µ )† γ 0 = γ µ . or. . S † γ 0 = b γ 0 S −1 .50) The real roots of this equation are b = ±1. γ 0 (γ 0 Sγ 0 S † )γ 0 = b∗ ½. we determine det (Sγ 0 S † γ 0 ) = det (b ½). therefore. (3.48) which. therefore. (3.76 3 Properties of the Dirac equation Here we have used (3.51) In fact.47). or. det S = 1. or.

43) and (3. Since we are dealing with proper Lorentz transformations.56) (3. (3. Thus. it follows from (3.54) which implies (see (3. using (3. we are assuming Λ00 ≥ 1. Λ00 ≤ −1. S † γ 0 = γ 0 S −1 .2 Covariance of the Dirac equation 77 Let us note. or.3.52) which follows since S † S represents a non-negative matrix. (3. or. we conclude from (3.53) or.55) These are some of the properties satisﬁed by the matrix S which will be useful in showing that it provides a representation for the Lorentz transformations. therefore.51) that b = 1. b > 0. that S †S = S †γ 0γ 0S = bγ 0 S −1 γ 0 S = bγ 0 Λ0ν γ ν = bγ 0 Λ00 γ 0 + Λ0i γ i = b Λ00 + Λ0i γ 0 γ i . (3.53)) that b > 0 and. . The two solutions of this equation are obvious Λ00 ≥ 1.47). b < 0. Tr S † S = 4bΛ00 0.47) that Sγ 0 S † γ 0 = ½. (3.

. namely. therefore.62) (3. since ǫµν = −ǫνµ .57) From our earlier discussion in (3. ν ν ν (3. 4 (3. we can expand the matrix S(Λ) as S(Λ) = S(ǫ) = ½ − i Mµν ǫµν .78 3 Properties of the Dirac equation Next. We can also write S −1 (ǫ) = ½ + so that S −1 (ǫ)S(ǫ) = = = i Mµν ǫµν . ǫµν = −ǫνµ .63) ½ + O(ǫ2 ). 4 (3. let us consider an inﬁnitesimal Lorentz transformation of the form (ǫµν inﬁnitesimal) x′µ = Λµν xν = (δµ + ǫµ )xν = xµ + ǫµ xν . Mµν = −Mνµ .60) ½+ ½+ i Mµν ǫµν 4 ½− i Mστ ǫστ 4 i i Mµν ǫµν − Mµν ǫµν + O(ǫ2 ) 4 4 (3. Mµν denote matrices in the Dirac space). (3.18).61) (3. we recall that the inﬁnitesimal transformation matrix is antisymmetric.58) For an inﬁnitesimal transformation.59) where the matrices Mµν are assumed to be anti-symmetric in the Lorentz indices (for diﬀerent values of the Lorentz indices.

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

For completeness.72) .98)).69) with the generators of inﬁnitesimal Lorentz transformations for the 1 Dirac wave function. consequently. 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.71) A ﬁnite rotation by angle θ in the 1-2 plane would. 1 σµν . be constructed out of a series of inﬁnitesimal transformations and. S(ǫ) = ½ − i 1 i σ12 ǫ = ½ + γ1 γ2 ǫ = ½ − α3 ǫ. be obtained from an inﬁnite sequence of inﬁnitesimal transformations resulting in an exponentiation of the inﬁnitesimal generators as ˜ S(θ) = e− 2 α3 θ . of course.109) which the generators of the inﬁnitesimal transformations.43) and generates Lorentz transformations and as a result.80 3 Properties of the Dirac equation Let us note here from the form of S(ǫ) that we can identify 1 σµν . at least for inﬁnitesimal Lorentz transformations. ˜ 2 2 2 (3.70) with all other components of ǫµν vanishing. 2 (3. satisfy can 2 be identiﬁed with the Lorentz algebra (which also explains why they are closed under multiplication). Thus. then. In such a case (see also (2. the Dirac equation is form invariant (covariant) under such a Lorentz transformation. we have shown that there exists a S(Λ) which satisies (3.) We will see in the next chapter (when we study the representations of the Lorentz group) that the algebra (2. (The other factor of 2 is there to avoid double counting. i (3. A ﬁnite transformation can. (3. let us note that inﬁnitesimal rotations around the 3-axis or in the 1-2 plane would correspond to choosing ǫ12 = ǫ = −ǫ21 .

so that we can write (see also (2. rotations ˜i ˜ deﬁne unitary transformations. namely. This is. we are considering an inﬁnitesimal boost along the 1-axis.78) . In this case. in this case.98)) S(ǫ) = i 1 σ01 ǫ = ½ + γ0 γ1 ǫ 2 2 1 = ½ − α1 ǫ. namely.77) with all other components of ǫµν vanishing. 2 ½− (3. we can determine S(θ) = cos This shows that S(θ + 2π) = − cos S(θ + 4π) = S(θ). of course. is double valued and.74) and. θ θ − i˜ 3 sin α 2 2 = −S(θ). Furthermore.3. (3. we can identify ǫ01 = ǫ = −ǫ10 .2 Covariance of the Dirac equation 81 Note that since α† = αi . 0 ½ (3. Let us next consider an inﬁnitesimal rotation in the 0-1 plane. consistent with the fact that the Dirac equation describes spin 1 2 particles.73) α3 ˜ = ½ 0 = ½. the rotation operator. recalling that σ3 0 0 σ3 α3 = ˜ we have 2 .76) θ θ − i˜ 3 sin . we have S † (θ) = S −1 (θ). therefore. corresponds to a spinor representation. (3. therefore. α 2 2 (3. (3.75) That is.

operators deﬁning boosts are not unitary.80) and. the matrix for a ﬁnite boost. 0 ½ (3. therefore.79) α1 = . ω. Let us note next that.82) We note here that since α† = α1 .83) That is. (3. 2 2 S(ω) = cosh (3. since under a Lorentz transformation .82 3 Properties of the Dirac equation In this case. can be obtained through exponentiation as S(ω) = e− 2 α1 ω . we have shown how to construct the matrix S(Λ) for ﬁnite Lorentz transformations (for both rotations and boosts).3 Transformation of bilinears In the last section. (3. All the unitary representations are necessarily inﬁnite dimensional. This is related to the fact that Lorentz boosts are non-compact transformations and for such transformations. 3. 1 S † (ω) = S −1 (ω). α2 = 1 ½ 0 = ½. recalling that 0 σ1 σ1 0 1 (3. there does not exist any ﬁnite dimensional unitary representation. Furthermore. as 4 × 4 matrices). in this four dimensional space (namely.81) we can determine ω ω − α1 sinh .

Thus.86) ′ ψ(x)ψ(x) → = Namely. 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). we see that the adjoint wave function ψ(x) transforms inversely compared to the wave function ψ(x) under a Lorentz transformation. we see that if we deﬁne a current of the form j µ (x) = ψ(x)γ µ ψ(x). under a Lorentz transformation ψ(x)γ µ ψ(x) → = = ψ (x′ )γ µ ψ ′ (x′ ) ψ(x)S −1 (Λ)γ µ S(Λ)ψ(x) ψ(x)Λµν γ ν ψ(x) = Λµν ψ(x)γ ν ψ(x).50) and (2. (3. it would transform as a four vector under a proper Lorentz transformation. (3.56). ′ (3.55)). 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. namely. (3. Similarly. ψ (x′ ) = ψ ′ † (x′ )γ 0 = ψ † (x)S † (Λ)γ 0 = ψ † (x)γ 0 S −1 (Λ) = ψ(x)S −1 (Λ). J µ (x) = ψ(x)γ µ ψ(x) → Λµν J ν (x).84) (3.88) . it follows that ψ ′ † (x′ ) = ψ † (x)S † (Λ).43).3 Transformation of bilinears 83 ψ ′ (x′ ) = S(Λ)ψ(x).85) where we have used the relation (3. In other words.87) ′ where we have used (3.3.

completeness relation Let us note that the positive energy solutions of the Dirac equation satisfy (p − m)u+ (p) = γ 0 p0 − γ · p − m u+ (p0 . It is customary to identify (see (2.92) so that the equations satisﬁed by u(p) and v(p) (positive and negative energy solutions).89) and (3. (3.84 3 Properties of the Dirac equation This is. Namely.91). we note that in this way. −p) = u− (−p) = .90). p) = 0. we can determine the transformation properties of the other bilinears under a Lorentz transformation in a straightforward manner. the probability current density (see also (2. (3.93) . 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) = .91) with the same value of p0 as in (3. can be written as (p − m)u(p) = 0. (3. p) = 0. Finally.89) while the negative energy solutions satisfy − γ 0 p0 − γ · p − m u− (−p0 . (3. what we had observed earlier. E+m 2m σ ·p E+m v ˜ v ˜ v(p) = u− (−p0 . / (3.90) (3.4 Projection operators. of course.49).85)) transforms like a four vector so that the probability density transforms as the time component of a four vector. 3. / where p0 = ω = p2 + m2 .

/ (p + m)v(p) = 0.94) Given these equations.97) . 4.3. (3. can represent the spin projection of the two component spinors (in terms of which the four component solutions were obtained). or.95) where we have used (γ µ )† γ 0 = γ 0 γ µ (see (2.4 Projection operators. / (3. / or.96) where r. u(p)(p − m) = 0. 2. completeness relation 85 and − γ 0 p0 + γ · p − m v(p) = 0. α v r (p)v s (p) = α=1 r s r s v α (p)vα (p) = −δrs . v(p)(p + m) = 0. (−p − m)v(p) = 0. 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. Let us denote them by ur (p) and v r (p). Let us also note that each of the four solutions really represents a four component spinor. we can write down the Lorentz invariant conditions we had derived earlier from the normalization of a massive Dirac particle as (see (2. as we had seen earlier. or. / v † (p) (p)† + m γ 0 = 0. / or.83)). u (p)v (p) = 0 = v r (p)us (p). / (3. With these notations. 3. 2.50)) 4 u r (p)us (p) = α=1 4 r uα (p)us (p) = δrs . (3. As we have seen earlier there are two positive energy solutions and two negative energy solutions of the Dirac equation. Let us denote the spinor index by α = 1. r = 1.

of course.86 3 Properties of the Dirac equation Although we had noted earlier that u+ (p)u− (p) = 0. (3. 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). 2m (3. the last relation in (3. 2m p+m r / v (p) = 0. namely.101) .99) for any two spin components of the positive and the negative energy spinors.94).100) Λ− (p) = 2m These are. 2m −p + m r / u (p) = 0. it is clear that we can deﬁne projection operators for such solutions as p+m / . 2m −p + m r / v (p) 2m −p − m + 2m r / v (p) = v r (p).97) can be checked to be true simply because v(p) = u− (−p0 . This also allows us to write 2 r=1 [u r (p)ur (p) − v r (p)v r (p)] = 4. 4 × 4 matrices and their eﬀect on the Dirac spinors is quite clear. From the form of the equations satisﬁed by the positive and the negative energy spinors.93) and (3. 2m −p + m / . −p). (3. (3. because the direction of momentum changes for v(p) (see the derivation in (2. (3.98) For completeness we note here that it is easy to check u† (p)v(−p) = 0 = v † (−p)u(p).52)).

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

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). 4. (3. α α 4 s uβ (p)Pβα (p) β=1 2 2 s r uβ (p)ur (p)uα (p) β β=1 r=1 r s δrs uα (p) = uα (p). r=1 r ur (p)uβ (p) = α p+m / 2m . we see that the matrix P projects only on to the space of positive energy solutions and.104) This matrix has the property that acting on a positive energy spinor it gives back the same spinor. αβ (3. r=1 4 (u (p)P (p))α = s = = (P (p)v s (p))α = β=1 4 s Pαβ (p)vβ (p) 2 s r ur (p)uβ (p)vβ (p) = 0. 2 or. we can identify Pαβ (p) = (Λ+ (p))αβ . Namely.105) Thus. 3.106) Similarly. 2. α α. α β=1 r=1 = (3. β = 1. therefore. if we deﬁne .88 3 Properties of the Dirac equation 2 Pαβ (p) = r=1 r ur (p)uβ (p).

103)) . (3.109) The completeness relation for the solutions of the Dirac equation now follows from the observation that (see (3. Hence we can identify Qαβ = −(Λ− (p))αβ .3. αβ (3. β β=1 r=1 4 4 s Qαβ (p)vβ (p) = β=1 2 β=1 r=1 r s vα (p)(−δrs ) = −vα (p). the matrix Q projects only on to the space of negative energy solutions with a phase (a negative sign). (3. 2 or. 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 .107) then. 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 .108) Namely. completeness relation 89 2 Qαβ (p) = r=1 r r vα (p)v β (p).4 Projection operators.

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

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

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. by the eigenvalue of Sz ). however. that. namely. we note that since momentum commutes with the Dirac Hamiltonian. Thus.σ. On the other hand. 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. spin commutes with the Hamiltonian (since in this frame p = 0) and such solutions can be labelled by the spin projection.6. we have already seen that the spin operator 1 ˜ α. ˜ 2 [Si . In fact. unlike the case of non-relativistic systems where we specify a given energy state by the projection of spin along the z-axis (namely.115) = 1 Tr M Λ+ (p′ )γ 0 M † γ 0 Λ+ (p) . . it commutes with the total angular momentum).117) satisﬁes the commutation relation (see (2. 3.β.5 Helicity As we have seen. Note. H = −iǫijk αj pk . it can be easily checked that the plane wave solutions which we had derived earlier are not eigenstates of the spin operator.118) As a consequence. the Dirac Hamiltonian H = α · p + βm.67)) 1 αi .92 1 2 4 3 Properties of the Dirac equation = Mαβ Λ+ (p′ α. (3. (3. H] = (3.λ=1 βσ γ 0M †γ 0 σλ (Λ+ (p))λα (3. 2 σi 0 0 σi S= αi = ˜ . for a particle at rest.

120) Therefore. This is known as the helicity operator and we note that since the Hamiltonian commutes with helicity. α · p + βm] = 0. H] = [Si . h)v(p. The normalized operator h= S·p . |p | (3. h′ ).3. h′ ) = 0 = v(p. [Si pi .124) . h). the eigenvalues of the helicity operator. h)u(p. h)v(p. u(p.5 Helicity 93 [pi . can only be ± 2 and we can label the positive and the negative energy solutions also as u(p. h) with h = ± 1 (the two helicity 2 eigenvalues). the order of these operators in the product is not relevant). Note that S·p |p | 2 h2 = = 1 ½. this operator is a constant of motion. (3.121) measures the longitudinal component of the spin of the particle or the projection of the spin along the direction of motion. v(p. therefore.122) where we have used (this is the generalization of the identity satisﬁed by the Pauli matrices) (S · p )(S · p ) = 1 1 ˜ ˜ (α · p) (α · p) = p2 ½. h′ ) = δhh′ = −v(p. h′ ). the eigenstates of energy can also be labelled by the helicity eigenvalues. H]pi = −iǫijk αj pk pi = 0. Namely. (3. 4 (3.123) Therefore.119) the operator S · p does also (momentum and spin commute and. (3. 4 4 (3. for a Dirac parti1 cle. H] = [pi . The normalization relations in this case will take the forms u(p. h)u(p.

h)] = ½. h) − v(p. which can also be written as p0 u1 (p) − σ · p u2 (p) = mu1 (p). Let us represent the four component spinor (as before) as u1 (p) u2 (p) u(p) = . (3. (3.130) . h)v(p. (3. γ 0 p0 − γ · p − m u(p) = 0.126) where we are not assuming any relation between p0 and p as yet.125) 1 h=± 2 3.6 Massless Dirac particle Let us consider the free Dirac equation for a massive spin (γ µ pµ − m) u(p) = 0. this leads to the two (2-component) coupled equations p0 − m u1 (p) − σ · p u2 (p) = 0. (3.128) Explicitly.127) where u1 (p) and u2 (p) are two component spinors.111). In terms of u1 (p) and u2 (p). 1 2 particle. (3. the completeness relation. (3. can now be written as [u(p.110) or (3. the Dirac equation takes the form (p0 − m)½ σ·p −σ · p u1 u2 −(p0 + m)½ = 0.129) p0 u2 (p) − σ · p u1 (p) = −mu2 (p). (3. or. σ · p u1 (p) − p0 + m u2 (p) = 0.94 3 Properties of the Dirac equation Furthermore. h)u(p.

130).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). since vanishing of the mass which is a Lorentz scalar should not change the behavior of the equation under proper Lorentz transformations). we obtain p0 − σ · p (u1 (p) + u2 (p)) = m (u1 (p) − u2 (p)) . the equations in (3.6 Massless Dirac particle 95 Taking the sum and the diﬀerence of the two equations in (3. 2 1 (u1 (p) + u2 (p)) .3. are not invariant under parity or space reﬂection and are known as the Weyl equations. These equations. like the Dirac equation. p0 uL (p) + σ · p uL (p) = muR (p). Let us note that in the limit m → 0.132) then. (3. The corresponding two component spinors uL and uR are also known as Weyl spinors.131) p0 + σ · p (u1 (p) − u2 (p)) = m (u1 (p) + u2 (p)) .134) These two equations. can be shown to be covariant under proper Lorentz transformations (as they should be. (3.133) We note that it is the mass term which couples the two equations. . however. We note that if we deﬁne two new (2-component) spinors as 1 (u1 (p) − u2 (p)) .133) reduce to two (2-component) spinor equations which are decoupled and have the simpler forms p0 uR (p) = σ · p uR (p). the two equations in (3. 2 uL (p) = uR (p) = (3. (3. p0 uL (p) = −σ · p uL (p).

p0 uL (p) = −σ · p uL (p). (3. for a nontrivial solution of these equations to exist. p0 uR (p) = σ · p uR (p). |p | (3. Similarly.136) Thus. therefore. we can show that uL (p) also satisﬁes (p0 )2 − p 2 uL (p) = 0. while (3. namely. (p0 )2 uR (p) = σ · p p0 uR (p) (p0 )2 − p 2 uR (p) = 0. Recalling that 1 σ denotes the spin 2 . σ·p uR (p) = uR (p). the two diﬀerent Weyl equations really describe particles with opposite helicity. = (σ · p) (σ · p) uR (p) = p 2 uR (p). we note that p0 uR (p) = σ · p uR (p). |p | or.137) which is the Einstein relation for a massless particle. for the positive energy solutions. that for such solutions.135) or. σ·p uL (p) = −uL (p). we must have p0 = ±|p |. (3. we must have (p0 )2 − p 2 = 0.139) or. (3.138) For p0 = |p |. It is clear.96 Let us note that 3 Properties of the Dirac equation or. (3.140) In other words.

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

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

92)) u(p) ˜ σ·p u(p) ˜ E E (3.149) commutes with γ5 (in fact.100)) H = α · p. Since 2 γ5 = ½. the solutions of the massless Dirac equation can be classiﬁed according to the eigenvalues of γ5 also known as the chirality or the handedness. then γ5 u(p) (or γ5 v(p)) is also a solution.7 Chirality With the normalization for massless spinors discussed in (2. Therefore. we note that if u(p) (or v(p)) is a solution.151) .3. for example. (3. ˜ |p| v(p) = σ·p v (p) ˜ v (p) ˜ = σ ·p v (p) ˜ |p| v (p) ˜ . in the Pauli-Dirac representation γ5 = ρ deﬁned in (2. / / can be written as (see also (3.147) u(p) = E 2 E 2 = |p| 2 |p| 2 u(p) ˜ σ·p u(p) . (3. (2.61)).7 Chirality 99 3.54). (3.147).53) and (2. namely. (3. This can also be seen from the fact that the Hamiltonian for a massless Dirac fermion (see (1. see. γ5 vR (p) = vR (p). the solutions of the massless Dirac equation (m = 0) pu(p) = 0 = pv(p).148) From the structure of the massless Dirac equation (3. γ5 uR (p) = uR (p).150) it follows that the eigenvalues of γ5 are ±1 and spinors with the eigenvalue +1.60) and ρ commutes with α.

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

156) We note from (3. PL = .7 Chirality 101 1 1 PR = (½ + γ5 ). (3.158) . hχ(±) (p) = σ·p 1 2|p| 2 ½± ½± 1 1 = ± × 2 2 σ·p χ ˜ |p| 1 σ·p χ = ± χ(±) (p). From the deﬁnition of the helicity operator in (3. we will continue our discussion in the Pauli-Dirac representation of the Dirac matrices which we have used throughout.119). However.155) PR PL = 0 = PL PR . 2 −½ ½ ½ ½ (3.3. the four component spinors can be eﬀectively described by two component spinors. these projection operators have the explicit forms (see (2. PL = (½ − γ5 ).154) 2 2 We note that by deﬁnition these projection operators satisfy PR 2 = PR . which implies that any four component spinor can be uniquely decomposed into a right-handed and a left-handed component. we note that 2 spinors of the form χ(±) (p) = 1 σ·p 1± χ. ˜ |p| 2 (3. namely.153) that in the massless Dirac theory. (3.91)) 1 2 PR = ½ ½ 1 ½ −½ .121) (for the two component spinors S = 1 σ). PR + PL = 1. (In the Pauli-Dirac representation. PL 2 = PL . the Dirac equation reduces to two decoupled two component Weyl equations (recall that it is the mass term which generally couples these two spinors). The reducibility of the spinors is best seen in the Weyl representation for the Dirac matrices discussed in (2. ˜ 2 |p| (3. This is connected with our earlier observation (see section 3.6) that in the massless limit.157) correspond to states with deﬁnite helicity.

P (+) P (−) = 0 = P (−) P (+) .70) or (3. vR (p) = .153) are described by two component spinors with positive helicity while the lefthanded (four component) spinors are described in terms of two component spinors of negative helicity. therefore. (P (−) )2 = P (−) .102 3 Properties of the Dirac equation so that the right-handed (four component) spinors in (3.159) We note here that the operators P (±) = 1 2 ½± σ·p 1 ˆ = (½ ± σ · p). (3. uL (p) = .117)) P (±) 0 1 2( P4×4 (±) = 0 P (±) 0 1 2( = ˆ ½ ± σ · p) 0 ˆ ½ ± σ · p) . vL (p) = −v (−) (p) v (−) (p) .160) can also be written in a covariant notation as P (+) = 1 µ σ pµ .161) with σ µ .153) and (3. It is straightforward ˜ ˆ p to check that they satisfy the relations (P (+) )2 = P (+) .120) and pµ = (1.(3. Explicitly. ˆ 2 (3. They can be easily generalized to a reducible representation of operators acting on the four component spinors and have the form (see (2. ˜ ˆ 2 (3.162) and. |p| 2 P (−) = 1 µ σ pµ . σ µ deﬁned in (2. (3. deﬁne projection operators into the space of positive and negative helicity two component spinors.163) . −ˆ ). we see from (3. P (+) + P (−) = 1.157) 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) = .

156) and (3. (3. We note from (3. with the choice of the basis in (3.167) such that when p1 = p2 = 0.167).R .165) 2 2 Each of these spinors is one dimensional and together they span the two dimensional spinor space. Furthermore.168) . For example. ˜˜ ˜˜ (3.164) which is the reason the spinors can be simultaneous eigenstates of chirality and helicity (when mass vanishes). from (3.159) as well as (3. ˜ ˜ ˜ ˜ uu† + v v † = ½. (3. we can also deﬁne normalized spinors u(+) and v (+) . 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 .7 Chirality 103 and it is straightforward to check from (3. we can choose u= ˜ 1 .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 −1.3. P4×4 = 0. let us derive some properties of these spinors. We can choose u and v to be ˜ ˜ normalized so that we have u(+) (p) = u† u = 1 = v † v . v(+) (p) = ½ + σ · p v. 1 (3.163) that PL.166) For example.157) that we can write the positive and the negative energy solutions as (we will do this in detail for the right-handed spinors and only quote the results for the left-handed spinors) 1 1 ˆ ˜ ˆ ˜ ½ + σ · p u. v (+) (p) = . the helicity spinors simply reduce to eigenstates of σ3 . (±) (3. For completeness as well as for later use. 0 v= ˜ 0 . In fact.

Here we note that the second relation follows from the fact that a positive helicity spinor changes into an orthogonal negative helicity spinor when the direction of the momentum is reversed (which is also manifest in the projection operator). We note that with p0 = |p|.169) which can be checked from the explicit forms of the spinors in (3. (3. u(+)† (p)v (+) (−p) = 0 = v (+)† (−p)u(+) (p). (3.171) = P (+) P (+) P (+) P (+) . the positive helicity spinors satisfy u(+)† (p)u(+) (p) = v (+)† (p)v (+) (p) = 1. In general.159).170) can be simpliﬁed by noting the following identity. as well as (3. Given the form of the right-handed spinors in (3.170) The completeness relation in (3. u(+) (p)u(+)† (p) = v (+) (p)v (+)† (p) = 1 2 ˆ ½ + σ · p . R † u† (p)vR (−p) = vR (−p)uR (p) = 0.104 3 Properties of the Dirac equation However. 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 .169) it now follows in a straightforward manner that † u† (p)uR (p) = |p| = vR (p)vR (p). R † uR (p)u† (p) = vR (p)vR (p) = R |p| 2 P (+) P (+) P (+) P (+) . we do not need to use any particular representation for our discussions. (3.168).

4. (3.8 Non-relativistic limit of the Dirac equation 105 so that we can write the completeness relation in (3. 2m E + m E+m (3. (3. / 4 (3. ˜ 2m σ·p E+m σ·p u= ˜ uL (p). 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).3.172) which can also be derived using the methods in section 3.8 Non-relativistic limit of the Dirac equation Let us recall that the positive energy solutions of the Dirac equation have the form (see (2.170) as † uR (p)u† (p) = vR (p)vR (p) = R 1 0 pγ (½ + γ5 ). L † uL (p)u† (p) = vL (p)vL (p) = L 1 0 pγ (½ − γ5 ).176) .175) In (3.173) 3.174) we have deﬁned uL (p) = uS (p) = E+m u.49)) u ˜ σ·p ˜ E+mu uL (p) uS (p) u+ (p) = E+m 2m = .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. L † u† (p)vL (−p) = 0 = vL (−p)uL (p).

175) we have denoted vL (p) = vS (p) = − E+m v. vL (p) and vS (p) are also known as the large and the small components of the negative energy solution. m½ σ · p −m½ σ·p uL (p) uS (p) =E uL (p) uS (p) . 2m E + m E+m (3. (3. or.179) We note that the second equation in (3. (α · p + βm) u+ (p) = Eu+ (p). the component uS (p) is much smaller than (of the order of v ) uL (p) and c correspondingly. which satisfy the equation Hu+ (p) = Eu+ (p). in (3.106 3 Properties of the Dirac equation and we emphasize that the subscript “L” here does not stand for the left-handed particles introduced in the last section. Similarly. uL (p) and uS (p) are known as the large and the small components of the positive energy Dirac solution. (3. σ · p uL (p) = (E + m)uS (p).179) gives the relation σ·p uL (p). or.174). Let us next look at the positive energy solutions in (3.177) It is clear that in the non-relativistic limit. In the non-relativistic limit. E+m uS (p) = (3. we expect the large components to give the dominant contribution to the wave function.180) . Similarly. ˜ 2m σ·p E +m σ·p v=− ˜ vL (p).178) This would lead to the two (2-component) equations σ · p uS (p) = (E − m)uL (p). when |p | ≪ m.

and.9 Electron in an external magnetic field 107 while.181) p2 uL (p) ≃ (E − m)uL (p).3. of course.181) has the form p2 uL (p) = ENR uL (p). σ·p uL (p) = (E − m)uL (p). therefore.179) takes the form (σ · p) or. equation (3. |p| ≪ m. with the substitution of this.33) and remember that we are choosing = 1) .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.183) Namely. 2 3. Since the coordinate representation of pµ is given by (see (1. This preserves the gauge invariance associated with the Maxwell’s equations and corresponds to deﬁning pµ → pµ − eAµ . what we know for a free non-relativistic electron (spin 1 particle). the Dirac equation in this case reduces to the Schr¨dinger o equation for a two component spinor which we are familiar with. 2m (3. E+m (3. 2m where we have used the fact that for a non-relativistic system. then. Furthermore. if we identify the non-relativistic energy (without the rest mass term) as ENR = E − m.182) (3. (3. the ﬁrst equation in (3. This is. E ≈ m (recall that we have set c = 1).184) where e denotes the charge of the particle and Aµ represents the four vector potential of the associated electromagnetic ﬁeld.

189) leads to σ · (p − eA) σ · (p − eA) uL (p) ≃ uL (p).190) where in the last relation.108 3 Properties of the Dirac equation pµ → i∂µ . takes the form (α · (p − eA ) + βm) u(p) = Eu(p). (3. (3. In this case. (3.189) uS (p) = (3. m½ σ · (p − eA ) σ · (p − eA ) −m½ uL (p) uS (p) =E uL (p) uS (p) . B = (∇ × A ) . (3.187) where we are assuming that A = A(x). we can write the two (2-component) equations as σ · (p − eA ) uS (p) = (E − m)uL (p). we obtain . σ · (p − eA ) uL (p) = (E + m)uS (p). E+m 2m (3.186) Let us next consider an electron interacting with a time independent external magnetic ﬁeld.188) Explicitly. or. Substituting this back into the ﬁrst equation in (3. the second equation in (3. The Dirac equation for the positive energy electrons. we have A0 = 0 = φ.185) the minimal coupling prescription also corresponds to deﬁning (in the coordinate representation) ∂µ → ∂µ + ieAµ .189). we have used |p| ≪ m in the non-relativistic limit. in this case. In this case.

3.191) Let us simplify the expression on the left hand side of (3. when we can approximate the Dirac equation by that satisﬁed by the two component spinor uL (p). (3. equation (3.9 Electron in an external magnetic field 109 (σ · (p − eA)) (σ · (p − eA)) uL (p) ≃ (E − m)uL (p). 1 e (p − eA)2 − σ · B uL (p) ≃ ENR uL (p). in the non-relativistic limit. Ak ] = −iǫijk [∇j . (3. we are going to use purely three dimensional notation for simplicity) (p × A + A × p )i = ǫijk (pj Ak + Aj pk ) = ǫijk (pj Ak − Ak pj ) = ǫijk [pj .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) .192) Note that (here. We can use this in (3. 2m or.194) Consequently. Ak ] = −i (∇ × A )i = −i(B)i . 2m (3.192) to write (σ · (p − eA)) (σ · (p − eA)) = (p − eA )2 − eσ · B.191) takes the form 1 (p − eA)2 − eσ · B uL (p) = (E − m)uL (p). (3.193) = (p − eA )2 − ieσ · (−iB) (3.195) 2m 2m where we have identiﬁed (as before) .

change this value slightly and the experimental deviation of g from the value of 2 (g − 2 experiment) for the electron agrees exceptionally well with the theoretical predictions of quantum electrodynamics.195) to be the Schr¨dinger equation for a charged o electron with a minimal coupling to an external vector ﬁeld along with a magnetic dipole interaction with the external magnetic ﬁeld. 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. Thus. this shows that a point Dirac particle has a magnetic moment corresponding to a gyro-magnetic ratio g = 2.197) we obtain g = 2.110 3 Properties of the Dirac equation ENR = E − m. (3. Quantum mechanical corrections (higher order corrections) in an interacting theory such as quantum electrodynamics. (3. can have g-factors quite diﬀerent from 2.199) 1 Since S = 2 σ for a two component electron. comparing with (3. Namely. one says that there is an anomalous contribution to the magnetic moment.197) Of course. In this case. to a magnetic dipole interaction (recall that in the non-relativistic theory. Particles with a nontrivial structure (that is particles which are not point like and have extended structures). we know that the magnetic moments are given by . a minimally coupled Dirac particle automatically leads. however.198) Let us recall that the magnetic moment of a particle is deﬁned to be (c = 1) e S. in the non-relativistic limit. for the proton and the neutron. however. 2m µ=g (3.196) We recognize (3. 2m µ= (3. for example.

In the non-relativistic limit. 2m (3.10 Foldy-Wouthuysen transformation 111 µN µP = −1. 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 . 3. this method runs into diﬃculty. However.79 µnm .202) denotes the electromagnetic ﬁeld strength tensor and κ represents the anomalous magnetic moment of the particle. The process of eliminating the “small” . (3. if we were to calculate the electric dipole interaction of an electron in a background electromagnetic ﬁeld using the method described in the earlier sections. For example. at higher orders. the Hamiltonian becomes non-Hermitian). = 2. This puzzling feature can be understood in a simple manner as follows.3.10 Foldy-Wouthuysen transformation In the last two sections. the electric dipole 1 moment becomes imaginary at order m2 (namely.91 µnm . we have described how the non-relativistic limit of a Dirac theory can be taken in a simple manner. 2mP µnm = (3.201) with mP denoting the mass of the proton.200) where the nuclear magneton is deﬁned to be |e| . This is commonly known as the Pauli coupling or the Pauli interaction. the relevant expansion parameter is |p| and m the method works quite well in the lowest order of expansion. (3.203) eκ µν σ Fµν . as we have seen explicitly.

(3. for example.206) that it is not unitary and this is the reason that the Hamiltonian becomes non-Hermitian at higher orders in the inverse mass expansion (non-relativistic expansion). Thus. for example.206) It is clear from the form of the matrix in (3. This diﬃculty in taking a consistent non-relativistic limit 1 to any order in the expansion in m was successfully solved by Foldy and Wouthuysen and also independently by Tani which we describe below. has the form (see (3. In this case. (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 .207).180)) σ·p . let us look at the free Dirac theory where we know that the Hamiltonian has the form (see (1.100) as well as (1. 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. (3.206) is the source of the problem in taking the non-relativistic limit consistently.101)) H = α · p + βm = γ 0 (γ · p + m) .204) where the matrix A. in the case of the free Dirac equation.207) Let us next look for a unitary transformation that will diagonalize the Hamiltonian in (3. E+m A= (3. such a transformation would .205) for the positive energy spinors.204) has the form T = ½ −A 0 ½ . The matrix T that takes us to the two component “large” spinors in (3.

avoid the problem with “large” and “small” spinors). Let us consider a transformation of the type U (θ) = e 2m γ ·p θ . m).209) From the properties of the gamma matrices in (1.83) or (1. γ · pθ 2m 2n+1 = cos γ ·p sin |p| |p|θ 2m (3. (3.210) and using this we can simplify and write U (θ) = ∞ n=0 1 (2n)! |p|θ 2m γ · pθ 2m + 2n + 1 (2n + 1)! . θ = θ(|p|.208) where the real scalar parameter of the transformation is a function of p and m. 1 (3.10 Foldy-Wouthuysen transformation 113 also transform the spinor into two 2-component spinors that will be decoupled and we do not have to eliminate one in favor of the other (namely.3. (γ · p)2 = −p2 = −|p|2 . (3.212) . we note that (γ)† = −γ. (3.91).211) It follows now that U † (θ) = cos which leads to |p|θ 2m − γ ·p sin |p| |p|θ 2m .

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 .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 .208). the free Dirac Hamiltonian (3. Namely.208) is indeed unitary. (3.213) × cos = cos 2 γ·p sin |p| 2 |p|θ 2m |p|θ 2m γ ·p |p| sin2 = cos2 + sin2 |p|θ 2m = ½. Under the unitary 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. the transformation (3.

from (3. leads to the diagonalized Hamiltonian m2 p2 + m2 p2 p2 + m2 H ′ = γ0 + = γ0 p2 + m2 .219) . sin = p2 + m2 . 1 1 (3. tan θ= − m sin |p| . However. |p|θ m m arctan |p| (3.217) that the Hamiltonian is now diagonalized in the positive and the negative energy spaces. (3. (3.215) In this case. or.216) which.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). the two components of the transformed spinor u′ (p) 1 u′ (p) 2 u′ (p) = U (θ)u(p) = . we have |p|θ m m p2 + m2 |p|θ m |p| cos = .217) We see from (3. As a result. m |p| m . θ.214). it is clear from (3.3.10 Foldy-Wouthuysen transformation 115 So far. our discussion has been quite general and the parameter of the transformation. has been arbitrary.214) that we can choose the parameter of transformation to satisfy |p|θ m = |p|θ m |p| cos or. (3. if we want the transformation to diagonalize the Hamiltonian. = 0.

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

224) m which clearly has a natural expansion in powers of |p| (ultrarelativistic expansion). |p| m which has a natural expansion in powers of |p| .11 Zitterbewegung The presence of negative energy solutions for the Dirac equation leads to various interesting consequences. in this case. 3. let us consider the free Dirac Hamiltonian (1.223) p2 + m2 γ·p= |p| p2 + m2 α · p.211) has the form UCT (θ) = cos 1 arctan 2 m |p| − γ ·p sin |p| 1 arctan 2 m .225) In the Heisenberg picture.220) as transforming away the α · p term in the Hamiltonian (3.207) while the Cini-Touschek transformation rotates away the mass term βm from the Hamiltonian (3. |p| (3.3. the Heisenberg equations of motion take the forms ( = 1) .100) H = α · p + βm. For example. we can think of the Foldy-Wouthuysen transformation (3.207). Therefore. (3.11 Zitterbewegung 117 H ′ = γ0 = γ0 p2 p2 + m2 + m2 p2 + m2 γ·p |p| (3. In fact. the unitary transformation (3. where operators carry time dependence and states are time independent.

on the other hand.230) In other words. The ﬁrst equation. Secondly. pj ] = i i 1 j ij α (iδ ) = αi . H] = 0. α · p + βm] = 0. it follows that (3. xi = ˙ 1 1 i [p .228) (3. The second equation in (3. H] = [α. As a result. it follows that the . Let us recall that.101)).226) Here a dot denotes diﬀerentiation with respect to time. = i ˙ or. αj pj + βm] i i 1 1 i j j [x . p = 0.118 3 Properties of the Dirac equation 1 i 1 [x . identiﬁes α(t) with the velocity operator.101) we conclude that [α.229) [α(t). α p ] = αj [xi . (3. for example. (3. H] = [pi . i i ˙ or.226) shows that the momentum is a constant of motion as it should be for a free particle. α(t) = eiHt α e−iHt . αj pj + βm] = 0. the velocity is not. x = α. by deﬁnition. pi = ˙ (3. Furthermore.227) where we have denoted the operator in the Schr¨dinger picture by o α(0) = α. even though the momentum of a free particle is a constant of motion. (1. H] = [xi . using (1. since the eigenvalues of α are ±1 (see.

= 0. α(t) eiHt |ψ where we have identiﬁed |ψ ′ = eiHt |ψ . since the electron is a massive particle. α(t)|ψ ′ = λ |ψ ′ . then. or.231) where λ denotes the eigenvalue of the velocity operator α. it follows that eiHt αe−iHt eiHt |ψ = λ eiHt |ψ . = λ eiHt |ψ . H] i + ˙ α(t) = = −i 2α(t)H − α(t). β = 2δij ½. We note from Heisenberg’s equations of motion that the time derivative of the velocity operator is given by 1 [α(t).232) or. More explicitly. This would seem to imply that the velocity of an electron is equal to the speed of light which is unacceptable even classically. (3.233) Equation (3.11 Zitterbewegung 119 eigenvalues of α(t) are ±1 as well. H = −2iα(t)H + 2ip. (3. These peculiarities of the relativistic theory can be understood as follows.232) shows that the eigenvalues of α(t) are the same as those of α (only the eigenfunctions are transformed) and. (3.102)) αi . This is easily understood from the fact that the eigenvalues of an operator do not change under a unitary transformation. αj αi . (3. are ±1. we note that if α|ψ = λ|ψ .234) Here we have used the relations (see (1.235) + + . (3.3. therefore.

from (3.236) (3. + α(0) − α(t) = H H The ﬁrst term in (3. for a relativistic particle. E mγ (3.234) with respect to time.239) is quite expected. ˙ ˙ or. α(t) = α(0)e−2iHt .234) we have ˙ α(0) = −2iα(0)H + 2ip = −2i(α(0) − pH −1 )H. using this relation in (3. which is unexpected. (c = 1) E = mγ. Let us note next that both p and H are constants of motion. so that mγv p = = v. Therefore. we ﬁnally determine −2i(α(0) − p H −1 )He−2iHt = −2iα(t)H + 2ip. We note that. we obtain ˙ α(t) = −2i(α(0) − p H −1 )He−2iHt . or.120 3 Properties of the Dirac equation as well as the fact that momentum commutes with the Hamiltonian. α(t) = pH −1 + (α(0) − pH −1 )e−2iHt .237) (3.239).234).239) or.236). It is the second term. we obtain ¨ ˙ α(t) = −2iα(t)H. however. diﬀerentiating (3.240) which is the ﬁrst term in (3. (3.241) p = mγv. On the other hand. For example. Substituting this back into (3. (3. p −2iHt p e . in an p eigenstate of momentum it would have the form E which is the true relativistic expression for velocity.238) Furthermore. It represents an additional component to the . (3.

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

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

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

This page intentionally left blank .

i.1 Symmetry algebras Relativistic theories. 4. k = 1. for space-time symmetries. as we have discussed. In this chapter.1) where αk represents the inﬁnitesimal constant parameter of rotation around the k-th axis (there are three of them).Chapter 4 Representations of Lorentz and Poincar´ e groups 4. therefore. Furthermore. should be invariant under Lorentz transformations. experimentally we know that space-time translations also deﬁne a symmetry of physical theories. In studying the symmetry algebras of continuous symmetry transformations. we will study the symmetry algebras of the Lorentz and the Poincar´ groups as well as their representations e which are essential in constructing physical theories. Let us consider an arbitrary. the symmetry algebra associated with a continuous symmetry group is given by the algebra of the generators of inﬁnitesimal transformations. But. it is suﬃcient to study the behavior of inﬁnitesimal transformations since any ﬁnite transformation can be built out of inﬁnitesimal transformations. inﬁnitesimal three di- mensional rotation of the form δxi = ǫi jk xj αk . In addition. the symmetry algebras can be easily obtained from the coordinate representation of the symmetry generators and that is the approach we will follow in our discussions. 3.1 Rotation. It is worth noting here that. 2.1. let us start with rotations which we have already discussed brieﬂy in the last chapter. (Let us recall our 125 . (4. j.

7) .6) In other words.) ω i j = ǫi jk αk . then. (4.3) (4. 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 . (4.34) and (1.4) This is. (4. (4.35) here for clarity. we can write the inﬁnitesimal rotations also as δxi = −R(ω)xi = ω i j xj . we note that ω ij = −ω ji . of course.5) R(ω) = = ω ij xi ∂j = −ω ij xj ∂i . ω µν in order to avoid confusion. It follows now that i R(ω)xi = −ω kj xj ∂k xi = −ω kj xj δk = −ω ij xj = −ω i j xj .126 4 Representations of Lorentz and Poincar´ groups e notation from (1. but here we denote them by ω ij . the form of the rotation that we had discussed in the last chapter.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 . etc. Let us next deﬁne an inﬁnitesimal vector operator (also known as the tangent vector ﬁeld operator) for rotations (an operator in the coordinate basis) of the form 1 1 ij ω Mij = ω ij (xi ∂j − xj ∂i ) 2 2 (4. ǫi jk = η iℓ ǫℓjk . ǫµν .

namely. R(ω) in (4.5). we note that R(ω). R(ω) = 1 ij ω Mij . ω kℓ xk ∂ℓ = ω ij ω kℓ [xi ∂j . (4.1 Symmetry algebras 127 Namely. 2 (4. Alternatively.5) generates inﬁnitesimal rotations and the operators. we can calculate them directly as .8) = R(ω). from the algebra of the vector operators in (4. the vector operator. Using the form of R(ω) in (4. Mij . Thus. Such an algebra is called a non-Abelian (non-commutative) algebra. Mij .8). R(ω) = ω ij xi ∂j . ∂ℓ ] ∂j ) = ω ij ω kℓ (ηjk xi ∂ℓ − ηiℓ xk ∂j ) = ω ij ω jℓ xi ∂ℓ − ω ij ω ki xk ∂j = ω ij ωjk − ω ij ωjk xi ∂k (4. xk ] ∂ℓ + xk [xi . are known as the generators of inﬁnitesimal rotations. they give back a rotation.10) we can obtain the algebra satisﬁed by the generators of inﬁnitesimal rotations.4. two rotations do not commute. rather. ω]ij . xk ∂ℓ ] = ω ij ω kℓ (xi [∂j . The Lie algebra of the group of rotations can be obtained from the algebra of the vector operators themselves.9) Namely. where we have identiﬁed ω ij = ω ik ω kj − ω ik ωkj = −ω ji = [ω.

Mrs ] 4 i j 1 = ǫi pq ǫjrs (−ηpr Mqs − ηqs Mpr + ηps Mqr + ηqr Mps ) 4 1 1 = − ǫir q ǫjrs Mqs − ǫi p s ǫjrs Mpr 4 4 . (4.11) This is the Lie algebra for the group of rotations. Jj ] = = 1 pq rs ǫ ǫ [Mpq . xk ∂ℓ − xℓ ∂k ] = [xi ∂j . 3) 1 pq 1 ǫi Mpq . q. 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 . Mkℓ ] = [xi ∂j − xj ∂i . with a factor of “i”. (4. Using this. If we would like the generators to be Hermitian quantum mechanical operators corresponding to a unitary representation. xℓ ∂k ] − [xj ∂i . 2 2 which gives the familiar orbital angular momentum operators. then. (4. xk ∂ℓ ] + [xj ∂i . xk ∂ℓ ] − [xi ∂j . = (xi ∂j − xj ∂i ) = ǫijk Jk . we note that we can deﬁne (recall that in the four vector notation Ji = −(J)i ) Mij or.128 4 Representations of Lorentz and Poincar´ groups e [Mij .11) represents the Lie algebra of the group SO(3) or equivalently SU (2). But up to a rescaling. s = 1. To obtain the familiar algebra of the angular momentum operators.12) 1 1 Ji = − ǫi jk Mjk = − ǫijk (xj ∂k − xk ∂j ). we obtain (p. ǫjrs Mrs 2 2 [Ji . r. then we may deﬁne the operators. 2. Mij .

1. 4.14) which.18) . namely. In the same spirit.11) or (4.13) is. = −ǫri q ǫjsr Mqs .17) (4.2 Translation. 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). so that µ R(ǫ)xµ = −ǫν ∂ν xµ = −ǫν δν = −ǫµ . let us note that a constant in- ﬁnitesimal space-time translation of the form δxµ = ǫµ . (4. ǫri q ǫjsr + ǫrsq ǫij r + ǫrjq ǫsi r = 0. 2 (4. the Lie algebra of SO(3) or SU (2) (or the familiar algebra of angular momentum operators) up to a rescaling. of course.1 Symmetry algebras 129 1 p rs ǫ ǫ Mps 4 i r j ǫi q r ǫjsr Mqs 1 q sr ǫ ǫ Mqs 4 i r j (4. leads to (see (4. [Ji . Jj ] = ǫij r Jr .13) 1 + ǫis q ǫjrs Mqr + 4 1 1 = − ǫriq ǫjsr Mqs − 4 4 1 − ǫri q ǫjsr Mqs − 4 or. then.12)) 1 q ǫri ǫjsr Mqs = − ǫijr ǫrsq Mqs = −ǫijr Jr .4. (4.16) (4.15) The algebra of the generators in (4. can be generated by the inﬁnitesimal vector operator R(ǫ) = −ǫµ Pµ = −ǫµ ∂µ .

µ. R(ǫ) = [ǫµ ∂µ . ∂ν ] = 0.18). 3.23) As in the case of rotations. 4. as we have seen in (3.3 Lorentz transformation. translations form an Abelian (commuting) group while the three dimensional rotations form a non-Abelian group. ν = 0.20) In other words.19) The Lie algebra associated with translations is then obtained from R(ǫ). Pν ] = [∂µ . two translations commute and the corresponding relation for the generators is [Pµ . 1.24) then.130 and we can write 4 Representations of Lorentz and Poincar´ groups e δxµ = −R(ǫ)xµ . ǫ ν ∂ν ] = ǫµ ǫ ν [∂µ .1. As we have seen in the last chapter. (4. let us note that if we deﬁne an inﬁnitesimal vector operator 1 µν 1 ω Mµν = ω µν (xµ ∂ν − xν ∂µ ) 2 2 µν = ω xµ ∂ν = −ω µν xν ∂µ . R(ω) = (4. ∂ν ] = 0. (4. 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 .21) Namely. constant parameters of transformation satisfy ω µν = −ω νµ . (4. (4. the inﬁnitesimal. we obtain .22) where. (4. 2.

4. we can think of R(ω) as the vector operator generating inﬁnitesimal proper Lorentz transformations and the operators.12)) Mij = −Mji = xi ∂j − xj ∂i = ǫij k Jk . We also note that we can identify the inﬁnitesimal generators of spatial rotations with (see (4. Thus. R(ω) = = ω µν ω λρ [xµ ∂ν . λ (4. where. ω]µν . as in the case of rotations. Mµν = −Mνµ .25) Therefore. and the generators of inﬁnitesimal boosts with (4.1 Symmetry algebras 131 µ R(ω)xµ = −ω λν xν ∂λ xµ = −ω λν xν δλ = −ω µν xν = −ω µν xν = −δxµ .29) . 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 (4.26) M0i = −Mi0 = x0 ∂i − xi ∂0 = Ki . we have ω µν = −ω νµ = ω µλ ω λν − ω µ ω λν = [ω.28) = = ω µν xµ ∂ν = R(ω). (4. we can determine the group properties of the Lorentz transformations from the algebra of the vector operators generating the transformations. 1 µν 1 ω Mµν . ω λρ Mλρ 2 2 R(ω). as the generators of the inﬁnitesimal transformations.27) As before. (4.

109) coincides with (4. xλ ∂ρ − xρ ∂λ ] = [xµ ∂ν .30) This. This implies that. (4.132 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. the matrices σµν provide a representation for the generators of the Lorentz group.69) in connection with the discussion of covariance of the Dirac equation. 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. therefore. This is what we had seen explicitly in (3. we note that the number of generators for SO(4) which is (for SO(n). therefore. . the Lie algebra of the generators is isomorphic to that of the group SO(4). xλ ∂ρ ] − [xµ ∂ν . xρ ∂λ ] − [xν ∂µ . (Namely. up to a scaling. in this case. Lorentz transformations (boosts) are non-compact unlike rotations in Euclidean space.31) coincides exactly with the six generators we have (namely. in four dimensions and. 2 (4. it is n(n−1) ) 2 1 × 4 × (4 − 1) = 2 × 3 = 6.) We end this section by pointing out that the algebra in (2. 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 . 1). three rotations and three boosts).30) (up to a scaling). gives the Lie algebra associated with Lorentz transformations. As we have seen these transformations correspond to rotations. Mλρ ] = [xµ ∂ν − xν ∂µ . In fact. The algebra of the generators can also be calculated directly and has the form [Mµν . xλ ∂ρ ] + [xν ∂µ . However.

If. xρ ∂λ ] + ω µν ǫ λ [xν ∂µ . we also consider inﬁnitesimal translations. it generates inﬁnitesimal Poincar´ e transformations. if we deﬁne an inﬁnitesimal vector operator as R(ǫ.34) The algebra of the vector operators for the Poincar´ transformae tions can also be easily calculated as R(ǫ. The transformations in (4. ∂λ ] + ω µν ω λρ [xν ∂µ .4. ω µν denote respectively the parameters of inﬁnitesimal translation and Lorentz transformation.35) . xρ ∂λ ] = ǫµ ω λρ ηµρ ∂λ + ω µν ǫ λ (−ηνλ ) ∂µ + ω µν ω λρ (ηµρ xν ∂λ − ηνλ xρ ∂µ ) = − (ω µν ǫ ν − ω µν ǫν ) ∂µ + ω µλ ωλν − ω µλ ωλ ν xµ ∂ν = − (ǫµ ∂µ + ω µν xν ∂µ ) = R (ǫ. ω) . ω). in addition to inﬁnitesimal Lorentz e transformations. R(ǫ.33) then.1. Namely. the general transformation of the coordinates takes the form δxµ = ǫµ + ω µν xν . (4. (4. in this case.4 Poincar´ transformation. ǫ λ ∂λ + ω λρ xρ ∂λ = ǫµ ω λρ [∂µ .32) where ǫµ .1 Symmetry algebras 133 4. Clearly. (4. ω) = −ǫµ Pµ + 1 µν ω Mµν 2 = −ǫµ ∂µ + ω µν xµ ∂ν = −ǫµ ∂µ − ω µν xν ∂µ = − (ǫµ ∂µ + ω µν xν ∂µ ) . ω) = ǫµ ∂µ + ω µν xν ∂µ . acting on the coordinates. ω)xµ = − ǫν ∂ν + ω λν xν ∂λ xµ = −(ǫν δν + ω λν xν δλ ) = −(ǫµ + ω µν xν ) = −δxµ .32) are known as the (inﬁnitesimal) Poincar´ transformations or the inhomoe geneous Lorentz transformations. µ µ R(ǫ. (4.

Mλρ ] = −ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ .22) and (3.37) which simply shows that under a Lorentz transformation. For change in the coordinate four vector under an inﬁnitesimal Lorentz transformation. (4. Pν ] = 0.134 4 Representations of Lorentz and Poincar´ groups e where we have identiﬁed ǫµ = (ω µν ǫ ν − ω µν ǫν ) . (This is seen by recalling that 2 ω µν Mµν generates inﬁnitesimal Lorentz transformations. 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.21)). (4. (4. for example. Mνλ ] = [∂µ . ω µν = ω µλ ω λν − ω µλ ωλ ν = [ω. ω]µν . Therefore.36) We can also calculate the algebra of the generators of Poincar´ group. since the generae tors of translations do not commute with the generators of Lorentz . as well as the homogeneous Lorentz group. combining with our earlier results on the algebra of the translation group. [Pµ . we conclude that the Lie algebra associated with the Poincar´ e transformations (inhomogeneous Lorentz group) is given by [Pµ . e We already know the commutation relations [Mµν . Mνλ ] = ηµν Pλ − ηµλ Pν .38) [Mµν . xν ∂λ − xλ ∂ν ] = ηµν ∂λ − ηµλ ∂ν = ηµν Pλ − ηµλ Pν .30) and (4. However. We note that the algebra of translations deﬁnes an Abelian subalgebra of the Poincar´ algebra (4.57). the only relation that needs to be calculated is the commutator between the generators of translation and Lorentz transformations.38). (4. Pν ] (see (4. (4.) Thus.30). (4. Note that [Pµ . Mλρ ] as well as [Pµ . see.21). Pµ behaves 1 like a covariant four vector.

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.) 4. − 1 kℓ ǫ Mkℓ 2 j 1 = − ǫjkℓ (−ηiℓ M0k + ηik M0ℓ ) 2 1 1 k ǫ Kk − ǫji ℓ Kℓ = 2 ji 2 = ǫij k Kk = [Ji .12).26) and (4.30) [Mµν .27) that we can identify the angular momentum and boost operators as 1 Ji = − ǫi jk Mjk . (4. M0j ] = −Mij = −ǫij k Jk . Jj ] = M0i . [Ji . Namely.42) . (4. Kj ] = [M0i .40) We recall from (4. Kj ] .2 Representations of the Lorentz group 135 transformations. e (4. 3. (4. Poincar´ algebra cannot be written as a direct sum e of those for translations and Lorentz transformations.39) Rather. it is what is known as a semi-direct sum of the two algebras. 1).41) Written out in terms of these generators. 2 Ki = M0i . the Lorentz algebra takes the form [Ki . Poincar´ algebra = t4 ⊕ so(3. [Ki . Jj ] = ǫij k Jk . i = 1.4. (The general convention is to denote groups by capital letters while the algebras are represented by lower case letters. 2. (4. Mλρ ] = (−ηµλ Mνρ − ηνρ Mµλ + ηµρ Mνλ + ηνλ Mµρ ).

let us note from the form of the algebra in (4. (4. Ki† = Ki .47) . Ki . Ji† = −Ji .136 4 Representations of Lorentz and Poincar´ groups e where we have used (4.13) in the last relation. As a consequence of (4.45).44) Parenthetically. The opposite hermiticity property of the generators of boosts. Let us deﬁne a set of new generators as linear superpositions of Ji and Ki as (this is also known as changing the basis of the algebra) Ai = Bi = 1 2 1 2 (Ji + iKi ) . the ﬁnite dimensional representations of boosts are non-unitary (hence the opposite Hermiticity of Ki ).46) Namely. namely. † Mµν = i(xµ ∂ν − xν ∂µ ) = Mµν . (Ji − iKi ) .42) that we can assign the following hermiticity properties to the generators. the generators in the new basis are all anti-Hermitian. as can be seen from the hermiticity of the generators in the coordinate basis. (4. This is a set of coupled commutation relations. consequently. (4. Ki = −i (Ai − Bi ) .43) which also leads to the inverse relations Ji = (Ai + Bi ) . However. inﬁnite dimensional representations are unitary. namely. in choosing the coordinate representation for the generators. we have not been particularly careful about choosing Hermitian operators. (compared to Ji ) is connected with the fact that such transformations are non-compact and. we have A† = −Ai . if we deﬁne the generators with a factor of “i”. i † Bi = −Bi . (4. (4.45) This unconventional hermiticity for Ji arises because.

(Jj − iKj ) 2 2 [Ai . 2 ij 1 1 (Ji + iKi ) . Kj ] + i [Ki . the Lorentz group is double valued (doubly connected). Aj ] = = = = [Bi . the algebra separates into two angular momentum algebras which are decoupled.48) In other words. Bj ] = = = = [Ai . Kj ] − i [Ki . Jj ] + i [Ji .43). 1) ≃ so(3) ⊕ so(3) ≃ su(2) ⊕ su(2). Therefore. Jj ] − i [Ji . Kj ]) 4 1 ǫ k Jk + iǫij k Kk + iǫij k Kk + ǫij k Jk 4 ij 1 k ǫ (Jk + iKk ) = ǫij k Ak . Bj ] = = 1 ([Ji . in this new basis. (Jj + iKj ) 2 2 1 ([Ji . Mathematically. Jj ] + [Ki . Kj ]) 4 1 ǫ k Jk − iǫij k Kk − iǫij k Kk + ǫij k Jk 4 ij 1 k ǫ (Jk − iKk ) = ǫij k Bk .42) takes the form 1 1 (Ji + iKi ) . Jj ] − i [Ji . Kj ]) 4 1 ǫ k Jk − iǫij k Kk + iǫij k Kk − ǫij k Jk = 4 ij = 0. the Lorentz algebra (4. (4. (Jj − iKj ) 2 2 1 ([Ji . so(3. it is more . as we have already seen in the last chapter.2 Representations of the Lorentz group 137 In the new basis (4.4. Kj ] + i [Ki . 2 ij 1 1 (Ji − iKi ) . one says that the Lorentz algebra is isomorphic to the direct sum of two angular momentum algebras. Jj ] − [Ki . (4. Jj ] − [Ki .49) Incidentally.

1. . . we can write (This is the generalization of the S(Λ) matrix that we studied in (3. ) to describe the Lorentz transformations.50) An irreducible nonunitary representation of the homogeneous Lorentz group. . can be speciﬁed uniquely once we know the values of jA and jB and is labelled as D(jA . 2 2 (4. 2 2 3 1 jB = 0. 1) which is SL(2.jB ) “ ” (j ) (j ) i i −i θA Ai A +θB Bi B “ ” (j .j ) (j .138 4 Representations of Lorentz and Poincar´ groups e meaningful to consider the simply connected universal covering group of SO(3. . .35) in connection with the covariance of the Dirac equation. The ﬁnite dimensional unitary representations of each of the angular momentum algebras is well known. Namely. .jB ) (just as the representation of the rotation group is denoted by D(j) ). (4. . Denoting by jA and jB the eigenvalues of the Casimir operators A2 and B2 respectively for the two algebras.jB) (Λ)ψ (jA .j ) −i θ i Ji A B +δi Ki A B D (Λ) = e = e . much the same way we consider the universal covering group SU (2) of SO(3) to describe rotations. . .53) . . Explicitly. .jB ) (x).52) where the ﬁnite parameters of rotation and boost can be identiﬁed with θi = 1 i i θ + θB . therefore.jB ) (x) = D(jA ) (Λ)D(jB ) (Λ)ψ (jA .jB ) (x′ ) = D(jA . . this represents the operator implementing ﬁnite transformations on the Hilbert space of states or wave functions as ψ ′ (jA . 2 A (4. 2 A δi = i i i θ − θB . (4. 1.51) where Λ represents the ﬁnite Lorentz transformation parameter. we have 3 1 jA = 0.) (jA .

j = . (4.0) . dimensionality = 2. jB ) can lie between j = |jA − jB |. . jA + jB .jB ) = (2jA + 1) (2jB + 1) . dimensionality = 1. in fact.55) Consequently. jB = 2 . For jA = 1 .54) and its spin content follows from the fact that (see (4. . 2 (4. 2 (4. (4. D(0.4. (4.0) . jB ) will have the dimensionality D (jA .58) 1 corresponds to a two component spinor representation with spin 2 . dimensionality (4. dimensionality = 2. 1 1 D(0.56) The ﬁrst few low lying representations of the Lorentz group are as follows. j = .59) also corresponds to a two component spinor representation with spin 1 2 . 2 ) . 1 Similarly.44)) J = A + B. jB = 0. we conclude that the values of the spin in a given representation characterized by (jA . For jA = jB = 0. . |jA − jB | + 1. For jA = jB = 2 . These two representations are inequivalent and. for jA = 0. 2 1 1 D( 2 . . 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. . from our knowledge of the addition of angular momenta.57) corresponds to a scalar representation with zero spin (which acts on the wave function of a Klein-Gordon particle).2 Representations of the Lorentz group 139 Such a representation labelled by (jA . j = 0.

we have (We note here that the negative sign in the spin 2 k representation in (4.) Ai (0) = 0. Ki . namely.47).) From (4.50). 2 (4. j = 0. this leads to the ﬁrst two nontrivial representations for the angular momentum and boost operators of the forms 1 ( 1 . 4. Bi by a factor of i.60) is known as a four component vector representation and can be identiﬁed with a spin content of 0 and 1 for the components.63) . Ai . Let us now con- struct explicitly a few of the low order representations for the generators of the Lorentz group. = 0.62) Using (4. 2 (4. dimensionality = 2 × 2 = 4.44). 2 1 1 ( ) Bi 2 = − σi . 1. we scale all the generators Ji . 2 ) ⊗ D ( 2 . t and a spin 1 component x (under rotations) and the same is true for any other four vector. 2 (1) ( 1 .1 Similarity transformations and representations.) It may be puzzling as to where the four component Dirac spinor ﬁts into this description.0) (1) (0) Ji 2 = Ai 2 + Bi = − σi . It actually corresponds to a reducible representation of the Lorentz group of the form 1 1 1 D(0.48). we now consider Hermitian generators by letting Mµν → iMµν as in (4. (Namely. To compare with the results that we had derived earlier. j = .2. Bi (0) 1 (1) Ai 2 = − σi . 2 ··· .61) This discussion can similarly be carried over to higher dimensional representations. (Note that a four vector such as xµ has a spin zero component. (4.140 4 Representations of Lorentz and Poincar´ groups e D( 2 . 1 1 (4.0) . dimensionality = 4. we note that for the ﬁrst few low order represen1 tations.0) (0) = −i Ai 2 − Bi Ki 2 = i σi . 2 ) .62) arises because ǫij = −ǫijk in (4. ··· .

98) (or (3. from (4.64).2 Representations of the Lorentz group 141 and 1 (1) (0.0) = Ji 2 ⊕ Ji 2 = = Ki ( 1 . Therefore. For example.65) 1 then. From (4. 2 are inequivalent representations. we note that these do not resemble the generators of the Lorentz algebra deﬁned in (3.64) Equations (4. 0 and 0.65) would imply the existence of an invertible matrix S such that S −1 σi S = σi .63) and (4. S −1 σi S = −σi . Two representations are said to be equivalent.67) However. the two representations la1 belled by 1 .63) and (4. They 2 provide the representations of angular momentum and boost for the left-handed and the right-handed Weyl particles. 1 ) (0) Ji 2 = Ai + Bi 2 = − σi . if we can ﬁnd a similarity transformation S leading to (0. ( 1 . (4. 0 and 0.71) and (3.66) which is clearly impossible.0) 2 ⊕ Ki (0. if there exists a similarity transformation relating the two. (4. In fact. 1 ) Ki 2 = S −1 Ki 2 S. 1 ) (1) (0) Ki 2 = −i Ai − Bi 2 i = − σi .78)). 1 ) 2 = 1 − 2 σi Ji (Dirac) 0 1 − 2 σi 0 .0) (0. .63) and (4. we would say that the two representations 2 .0) Ji 2 = S −1 Ji 2 S. 2 (0. 1 are 2 equivalent. 1 ) ( 1 . 1 ) ( 1 . (4. we can obtain the representation of the Lorentz generators for the reducible four component Dirac spinors as (0.64) we see that the condition (4. (Dirac) Ki i 2 σi 0 0 i − 2 σi .64) give the two inequivalent representations of dimensionality 2 as we have noted earlier. 2 (4.69) and (2.4.

119). ij σW = i i j γ . give a representation of the Lorentz generators in the Weyl representation. We note that in the Weyl representation for the gamma matrices deﬁned in (2. we can apply the inverse similarity transformation in (2. On the other hand. consequently.67) and in our earlier discussion in (3.68) As a result. (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 . 2 0i Therefore.71) and (3.142 4 Representations of Lorentz and Poincar´ groups e This puzzle can be understood as follows. (4. γW = ǫijk 2 W . we note that the angular momentum and boost operators in (4.98) (see also (3.69) 2 µν and. if we would like the generators in the standard Pauli-Dirac representation (which is what we had used in our earlier discussions).70) = σi i − 2 σi 0 1 PD σ .69) and (2. 2 = S −1 Ki = 1 2 (Dirac) S i 2 ½ ½ −½ ½ 0 i −2 σi 0 = 0 i −2 σi ½ −½ ½ ½ (4.67) are obtained from 1 W σ . i 0 i γ .78)) are equivalent since they are connected by a similarity transformation that .γ = 2 W W −iσi 0 σk 0 0 iσi 0 σk 0i σW = . we note that the generators in (4.

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

for example. Das (Hindustan Book Agency.144 4 Representations of Lorentz and Poincar´ groups e It can be directly checked from the matrix structures in (4. Quantum Mechanics: A Modern Introduction. (4. This puzzle can be resolved by noting that there is a similarity transformation that connects the two representations and. 0 0 −1 0 −1 0 i = √ 1 0 −1 . let us note that the generators obtained (LA) from the Lie algebra are constructed by choosing the generator J3 See. j. J2 = 0 0 0 i −i 0 0 0 0 . A.75) (LA) J1 (LA) J2 (LA) J3 which look really diﬀerent from the generators in (4. i. To construct the similarity transformation (which actually is a unitary transformation).74) J1 = 0 0 0 0 0 0 −i 0 i . they are equivalent. therefore provide a representation for the generators of rotations. Jj ] = iǫijk Jk . the representation in the space of three vectors which would correspond to j = 1. A.73) in spite of the fact that they belong to the same representation for j = 1.72) and (4. C. Melissinos (Gordon and Breach). page 289 or Lectures on Quantum Mechanics. (4. 2. This is.73) that they satisfy [Ji .73) and. Das and A. 2 0 1 0 (4. On the other hand. in fact. New Delhi). 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 . k = 1. 3. therefore. 2 0 1 0 1 0 0 = 0 0 0 .72) and (4. page 182 (note there is a typo in the sign of the 23 element for L2 in this reference) 1 .

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

therefore. (4. 2 0 1 0 1 √ 2 i √ 2 S −1 J3 S √ 1 2 i = 0 0 −1 0 0 0 − √2 i 1 √ √ −i 0 0 0 − 2 0 2 1 i √ √ 0 0 −i 0 2 2 0 0 0 −1 0 = 0 i i 1 i − √2 0 √2 − √2 √2 0 0 −1 0 i (LA) = √ 1 0 −1 = J2 . √ 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 (4.73) and (4. related by a similarity transformation and. are equivalent.146 4 Representations of Lorentz and Poincar´ groups e = S −1 J2 S 0 1 0 1 (LA) √ 1 0 1 = J1 .79) This shows explicitly that the two representations for J corresponding to j = 1 in (4.72). 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 . . in fact.75) which look rather diﬀerent are.

This e would help us in understanding the kinds of theories we can consider and the nature of the states they can have. (4. Mµν ] + [Pλ . Pµ P 2 .38) e since it commutes with all the ten generators. Mµν P λ Pλ .3 Unitary representations of the Poincar´ group e Since we are interested in physical theories which are invariant under translations as well as homogeneous Lorentz transformations. Mµν ] P λ = P λ (ηλµ Pν − ηλν Pµ ) + (ηλµ Pν − ηλν Pµ ) P λ = Pµ Pν − Pν Pµ + Pν Pµ − Pµ Pν = 0. we seek to ﬁnd unitary representations in some inﬁnite dimensional Hilbert space where the generators Pµ . Pµ = 0. The operators Mµν generate inﬁnitesimal Lorentz transformations through commutation relations and the relation above. which is supposed to characterize the inﬁnitesimal transformation of P 2 . P 2 . Let us deﬁne a new vector operator.3 Unitary representations of the Poincar´ group e 147 4. (4.81) The last relation in (4. Mµν = = = P λ [Pλ . namely. simply implies that P 2 does not change under a Lorentz transformation (it is a Lorentz scalar) which is to be expected since it does not have any free Lorentz index. Therefore.81) can be intuitively understood as follows. Mµν act as Hermitian operators.4. In order to determine the unitary representations. P λ Pλ . it is useful to study the representations of the Poincar´ group. known as the Pauli-Lubanski operator. Since Poincar´ e group is non-compact (like the Lorentz group). 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. let us note that the operator P 2 = η µν Pµ Pν = P µ Pµ .80) deﬁnes a quadratic Casimir operator of the Poincar´ algebra (4.

83) which follows from the fact that the generators of translation commute. in general. It follows from (4.83) implies that the PauliLubanski operator has only three independent components (both in the massive and massless cases). Let us deﬁne the dual of the generators of Lorentz transformation as M µν = Mµν 1 µνλρ ǫ Mλρ . we can write (4.82) also as W µ = Pν M µν = M µν Pν . (4. 2 1 = − ǫµνλρ M λρ .83) that. Furthermore. First. (However. (4.85) where the order of the operators is once again not important. the vector Wµ is orthogonal to Pµ .82) The commutator between Pµ and Mνλ introduces metric tensors (see (4.38)) which vanish when contracted with the anti-symmetric LeviCivita tensor. 2 (4. the order of Pµ and Mνλ are irrelevant in the deﬁnition of the Pauli-Lubanski operator. therefore. 2 (4.148 4 Representations of Lorentz and Poincar´ groups e Wµ = 1 1 µνλρ ǫ Pν Mλρ = ǫµνλρ Mλρ Pν .84) With this. As a result.) In general. 2 2 (4. we have e . Let us next calculate the commutators between W µ and the ten generators of the Poincar´ group. we note that Pµ W µ = 1 µνλρ ǫ Pµ Pν Mλρ = 0. this is not true for massless theories as we will see shortly.

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

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

The irreducible representations of the Poincar´ group can be clase siﬁed into two distinct categories.4. Mµν M µν . cannot represent Casimir operators of the algebra. this would also represent a Casimir operator of the Poincar´ e algebra since Wµ commutes with the generators of translation (see (4. we conclude that if we deﬁne an operator W 2 = W µ Wµ . the representations can be labelled by the eigenvalues of these operators. pµ . labelled by the momentum eigenvalues.86)). it is easy to check that these do not commute with the generators of translation and. 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 ). Namely. consequently. we choose the basis vectors of the repe resentation to be eigenstates of the momentum operators.3. There are other Lorentz scalars that can be constructed from Pµ and Mµν such as Mµν M µν .3 Unitary representations of the Poincar´ group e 151 It follows now from (4. to be diagonal (they satisfy an Abelian subalgebra). 4. of course.89) that [W µ Wµ . It can be shown that P 2 and W 2 represent the only Casimir operators of the algebra and. Pµ . satisfying . The eigenstates of the momentum operators |p are. (4. (4. Mστ ] W µ = W µ (ηµσ Wτ − ηµτ Wσ ) + (ηµσ Wτ − ηµτ Wσ ) W µ = Wσ Wτ − Wτ Wσ + Wτ Wσ − Wσ Wτ = 0. L2 = Lµ Lµ with Lµ = P ν Mνµ . In fact. which we treat separately. Therefore.93) However. we can choose the momentum operators. therefore. without loss of generality. (4.1 Massive representation.91) which is to be expected since Wµ W µ is a Lorentz scalar. Mστ ] = W µ [Wµ . Mστ ] + [Wµ .92) then. To ﬁnd unitary irreducible representa- tions of the Poincar´ algebra.

P 2 |p = pµ pµ |p .96) Here m denotes the rest mass of the single particle state and we assume the rest mass to be non-zero. (4. However. where p 2 = p µ p µ = m2 . the eigenvalues of W 2 are not so obvious. using (4. we have 1 µνλρ ǫ Mλρ Pν ǫµστ ζ Mτ ζ Pσ 4 W 2 = W µ Wµ = = = 1 µνλρ στ ζ ǫ ǫµ Mλρ Pν Mτ ζ Pσ 4 1 −η νσ η λτ η ρζ − η λζ η ρτ − η ντ η λζ η ρσ − η λσ η ρζ 4 − η νζ η λσ η ρτ − η λτ η ρσ Mλρ Pν Mτ ζ Pσ 1 1 1 = − Mλρ Pν M λρ P ν − Mλρ Pν M νλ P ρ + Mλρ Pν M νρ P λ 2 4 4 1 1 − Mλρ Pν M ρν P λ + Mλρ Pν M λν P ρ 4 4 1 = − Mλρ Pν M λρ P ν − Mλρ Pν M νλ P ρ 2 . (4. let us study this operator in some detail. the eigenvalues of the operator P 2 = P µ Pµ are obvious. namely.152 4 Representations of Lorentz and Poincar´ groups e Pµ |p = pµ |p .94) and in this basis.88).97) Therefore. 2 2 (4. Therefore. We recall that Wµ = 1 µνλρ 1 ǫ Pν Mλρ = ǫµνλρ Mλρ Pν .95) (4.

4.99) and the operator W 2 acting on such a state. let us go to the rest frame of the massive particle.100) = − m2 M ij Mij . the Pauli-Lubanski operator (4. (4. Let us note that in the rest frame (4. 2 Recalling that (see (4.3 Unitary representations of the Poincar´ group e 153 1 λ ρ = − Mλρ M λρ P 2 + δν P ρ P ν − δν P λ P ν 2 λ −Mλρ M νλ Pν P ρ + 4P λ P ρ − δν P ν P ρ 1 = − M λρ Mλρ P 2 − Mλρ M νλ Pν P ρ .101) where Jk represents the total angular momentum of the particle. 0. 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 . 2 (4. 0) . p 2 = p µ p µ = m2 . (4. To understand the meaning of this operator.26)) Mij = ǫij k Jk .102) The result in (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. 0.82) has the form . In this frame.98) where we have simpliﬁed terms in the intermediate steps using the anti-symmetry of the Lorentz generators. (4.102) can also be derived in an alternative manner which is simpler and quite instructive. pµ = (m.99).

for a massive particle at rest.105) Thus. 2 m 1 iνλρ ǫ Mλρ pν = − ǫijk Mjk = 2 2 m jk = − ǫi Mjk = mJi . we ﬁnd eigenvalues of W 2 : w2 = −m2 s(s + 1).104) which is the result obtained in (4. we see that the representations with p2 = 0 can be labelled by the eigenvalues of the Casimir operators.102). (m.34) as well as (4. we can ask what Lorentz transformations would leave such a vector invariant. Clearly.154 4 Representations of Lorentz and Poincar´ groups e W0 = Wi 1 0ijk ǫ Mjk pi = 0. all the degenerate states can be labelled not . 0). 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). It is not hard to see that all possible 3dimensional rotations would leave such a vector invariant. Technically. would deﬁne the stability group of such vectors. Namely. That is. s). 0. 2 (4.103) where we have used (1. these would deﬁne an invariant subgroup of the Lorentz group and will lead to the degeneracy of states. this simply measures the spin of the particle. (4. The dimensionality of the representation can also be understood in an alternative manner as follows.12). 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. Therefore. therefore. It follows now that W 2 = W µ Wµ = W i Wi = m2 J i Ji = −m2 J2 . For a state at rest with momentum of the form pµ = (m. one says that the 3-dimensional rotations deﬁne the “little” group of a time-like vector and this method of determining the representation is known as the “induced” representation. 0. for a massive particle. (4. we can think of W 2 as being proportional to J 2 and in the rest frame of the particle. Therefore.

pµ = (p. Consequently. From the Lorentz algebra.2 Massless representation. 4. we note that (see (4. 0. s − 1. This can also be seen algebraically.30)) [P0 . 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. · · · . namely. 1 .4.3. In closing. Mij ] = 0. the three nontrivial Pauli-Lubanski operators correspond to the generators of symmetry of the “little group” in the rest frame. the operators Mij . (4. commute with P0 . 0. −s + 1.3 Unitary representations of the Poincar´ group e 155 just by the eigenvalue of the momentum. we note that (see (4.83) we see that our states in the representation should also satisfy Pµ W µ = 0.108) However. a state at rest is an eigenstate of the P0 operator. −s. from (4. which generate 3-dimensional rotations and are related to the angular momentum operators. let us note from (4.103) that.83)) P 2 = 0. The basic reason behind this is that the “little” group of a light-like vector is not so obvious. In contrast to the massive representa- tions of the Poincar´ group. acting on states in such a vector space.107) Consequently. p = 0. we would have (see (4. but also by the eigenvalues of three dimensional rotations.109) . (4. s = 0. Namely.95)) pµ pµ = 0. up to a normalization factor. In this case. −p). 1. This deﬁnes the 2s + 1 dimensional representation for a massive particle of spin s.106) Namely. (4. · · · and 2 ms = s. the representations for a massless partie cle are slightly more involved. (4.

p0 h= (4. the simultaneous eigenstates of P 2 and W 2 would .88)) 1 0ijk 1 ǫ pi Mjk = ǫijk pi ǫjkℓ Jℓ = −p · J = −J · p.111) we conclude that in this space J·p .114) This is nothing other than the helicity operator (since L · p = 0) and. therefore. namely. then it can be shown (we will see this at the end of this section) that the representations are inﬁnite dimensional with an inﬁnity of spin values. for a massless particle if W 2 = 0.112) from which it follows that acting on a general momentum basis state |p (not necessarily restricting to massless states). (4. W 2 = W µ Wµ = 0. W µ has the form W µ = −hpµ . or.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. 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. consequently. 2 (4. namely.113) W0 = Comparing with (4. acting on states in such a space. in the second case where W 2 = 0 on the states of the representation.111) where h represents a proportionality factor (operator). let us recall that Wµ = 1 µνλρ ǫ Pν Mλρ . 2 2 (4. To determine h. it would lead to (see (4. On the other hand.110) In the ﬁrst case. (4. namely. Such representations do not correspond to physical particles and. W 2 = W µ Wµ = 0.

2 W0 = W1 = W2 = W3 = (4.83).116) Let us now determine the dimensionality of the massless representations algebraically. indeed represents the helicity operator up to a normalization as we had noted in (4.107).103) and (4. −p) and we would like to determine the “little” group of symmetries associated with such a vector. We recognize that in this case. but also algebraically by recognizing that a light-like vector of the form being considered is an eigenstate of the operator P0 − P3 . W 0 . We also note from (4. Let us recall that we are considering a massless state with momentum of the form pµ = (p.115) that the Pauli-Lubanski operator indeed has only three independent components because of the transversality condition (4.117) . the Pauli-Lubanski operator (4.113). the set of Lorentz transformations which would leave this four vector invariant must include rotations around the z-axis. 2 1 3νλρ ǫ Mλρ pν = −pM12 = W 0 . in the massless case.115) We see from both (4.82) takes the form 1 0ijk 1 ǫ Mjk Pi = ǫijk Mjk pi = −pM12 . (P0 − P3 ) |p = 2p|p . as we had pointed out earlier. 2 1 2νλρ ǫ Mλρ pν = −p (M01 − M13 ) . 0. let us note here that in the light-like frame (4.115) that (the contributions from W 0 and W 3 cancel out) W 2 = W µ Wµ = W 1 W1 + W 2 W2 = −p2 (M01 − M13 )2 + (M02 − M23 )2 . namely. (4. For completeness.4. This can be seen intuitively from the fact that the motion of the particle is along the z axis. It follows now from (4. 0. 2 2 1 1νλρ ǫ Mλρ pν = p (M02 − M23 ) .3 Unitary representations of the Poincar´ group e 157 correspond to the eigenstates of momentum and helicity. (4.115) that.

(4. M01 − M13 ] |p = 2P1 |p = 0.119) p Π1 = M01 − M13 = Π2 = M02 − M23 = − It follows now that these operators commute with P0 − P3 in the space of light-like states. but can be easily understood as follows. we see that e [P0 − P3 .38). as we had also seen in the massive case. Thus. Π2 ] |p = [P0 − P3 . we say that the stability group or the “little” group of a light-like vector is E2 . These represent all the symmetries of the light-like vector (state). in fact. = [P0 − P3 .115). [Π1 . Π2 have the same commutation relations as those of translations in this two dimensional space. M01 − M13 ] = Π2 . M02 − M23 ] = 0. let us deﬁne two new operators as W2 . p W1 . namely. (4. Π2 ] = [M01 − M13 . M12 ] = 0. 3 in (4. [M12 .158 4 Representations of Lorentz and Poincar´ groups e Furthermore. To determine the other symmetries of a light-like vector. it is isomorphic to the algebra of the Euclidean group in two dimensions.118) so that rotations around the z-axis deﬁne a symmetry of the lightlike vector (state) that we are considering. 2. M02 − M23 ] = −Π1 . E2 (which consists of translations and rotation). [P0 − P3 . (4. This may seem puzzling.120) and. also deﬁne symmetries of light-like states. we see that up to a normalization. We note that the algebra of the symmetry generators takes the form [M12 . comparing with W i . (4. from the Poincar´ algebra in (4. Π2 ] = [M12 . the three independent Pauli-Lubanski operators are. We note from (4. Π1 ] |p [P0 − P3 . the generators of symmetry of the “little” group. M02 − M23 ] |p = 2P2 |p = 0.90) that in the . Π1 ] = [M12 . therefore. i = 1. Furthermore. M12 is the generator of rotations around the z axis or in the two dimensional plane and Π1 . Clearly.121) Namely.

the space of physical states would also include the state with the opposite helicity (recall that helicity changes sign under space reﬂection. what we have been investigating within the context of “little” groups. Π1 ∓ iΠ2 ] = ±i (Π1 ∓ iΠ2 ) . they generate transformations which will leave the momentum basis states invariant.121) that Π1 ∓ iΠ2 correspond respectively to raising and lowering operators for M12 . As a result. Furthermore.3 Unitary representations of the Poincar´ group e 159 momentum basis states (where pρ is a number). h = 0.4. namely. Clearly. Namely. therefore. As a result. we see that in the space of light-like momentum states W 2 ∝ Π2 . does not correspond to any physical system.111).116). follows from (4. (4. then. On the other hand.122) Let us also note for completeness that the Casimir of the E2 algebra is given by Π2 = Π2 + Π2 = (Π1 ∓ iΠ2 )(Π1 ± iΠ2 ). (4. the Pauli-Lubanski operators satisfy an algebra and. their eigenvalues can take any value. see (3. Π2 correspond to generators of “translation”. The meaning of the transformations. we can say that Π1 |p. as we have already pointed out. 1 2 (4. then it follows from (4. we note from (4.122) that spin can take an inﬁnite number of values which. if W 2 = 0 in this space. Let us note from (4. can be thought of as generators of some transformations.124) (Alternatively. such a state would correspond to the highest or the lowest helicity state.146)). if our theory is also invariant under parity (or space reﬂection). This is. h = 0 = Π2 |p.123) and comparing with (4. massless theories with nontrivial spin that are parity invariant would have two dimensional representations corresponding to the highest and the . of course.) This corresponds to the one dimensional representation of E2 known as the “degenerate” representation.123) that (h corresponds to the helicity quantum number) (Π1 ± iΠ2 ) |p. [M12 . if W 2 = 0 in this space of states. h and this is the reason for the assertion in (4. Since Π1 .86) as the transformations that leave pµ invariant.

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

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

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

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

ti (5. subject to the boundary conditions (5. Note that under an inﬁnitesimal change tf (5. We have .10) Here we have used Gauss’ theorem (as well as the vanishing of ﬁeld variables asymptotically (5.8) δφ (x.3)) to simplify the surface term. We can now generalize the variational principle of point particle mechanics and ask under what conditions will the action be stationary if we change the ﬁelds arbitrarily and inﬁnitesimally as φ(x) → φ′ (x) = φ(x) + δφ(x). tf ) = 0.164 5 Free Klein-Gordon field theory most on the ﬁrst order derivatives of the ﬁeld variables.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) . 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. ti ) = δφ (x.

As in point particle mechanics. the action will be stationary under an arbitrary change in the ﬁeld variable subject to the boundary conditions provided (namely. (5. ∂µ ∂ µ φ + m2 φ = 0.5. for the Klein-Gordon equation.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). tf ) .13) Therefore. namely. (5. ∂µ φ(x)).14) This is known as the Euler-Lagrange equation associated with the action S or the Lagrangian density L(φ. δφ (x. the change in the action is given by tf δS = ti d4 x ∂L ∂L − ∂µ ∂φ(x) ∂∂µ φ(x) δφ(x).15) . inﬁnitesimal change in the ﬁeld variable. we conclude that under an arbitrary. ∂φ(x) ∂∂µ φ(x) (5. Thus.11) We note that the “surface term” – the last term in the expression (5.10) – vanishes because of the boundary conditions (5. (5.9). ti ) = 0 = δφ (x. ∂L ∂L − ∂µ = 0. we can think of the dynamical equations as arising from the minimum principle associated with a given action.12) Thus. δS = 0 only if). (5.

Note that the Lagrangian density L in (5.18) In other words. 2 2 L= (5. ∂φ(x) ∂L = ∂ µ φ(x). ∂∂µ φ(x) Therefore. ∂φ(x) ∂∂µ φ(x) or. gives (5. the Euler-Lagrange equation. ∂µ ∂ µ + m2 φ(x) = 0. in this case.16) ∂L = −m2 φ(x). Similarly. . if we assume the ﬁelds to fall oﬀ rapidly at inﬁnite separation (see (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. Note that if we choose 1 m2 2 ∂µ φ∂ µ φ − φ .3)). in this case. the action does not change either. (5. Of course. 2 2 L= then (5. real scalar ﬁeld as its Euler Lagrange equation. It is worth noting here that one can add a constant to L without changing the equation of motion. we see that the Lagrangian density 1 m2 2 ∂µ φ∂ µ φ − φ .19) gives the Klein-Gordon equation for a free. adding a total divergence to L also does not change the equations of motion. This would simply correspond to adding a zero point energy.17) ∂L ∂L − ∂µ = −m2 φ(x) − ∂µ ∂ µ φ(x) = 0.19) is manifestly Lorentz invariant.

∂x ˙ p= (5. H} . x} = 0 = {p. x) . ˙ (5. the Lagrangian is a function of the coordinate and the velocity of the particle (a dot denotes a derivative with respect to t). p} . in turn. p} = 1. we can try to develop a Hamiltonian description for it (which we need for quantization) in complete parallel to what we do for point particle mechanics.5.19) (xµ in this case merely labels space-time coordinates) . on the other hand.23) These. p) = px − L (x.20) The conjugate momentum associated with the dynamical variable x is deﬁned as ∂L .22) The classical canonical Poisson bracket relations between the coordinate and the conjugate momentum have the familiar form {x. H} .24) In the case of the classical scalar (Klein-Gordon) ﬁeld theory. ˙ ˙ (5. {x. ˙ (5. L = L (x. ˙ p = {p. In the case of a point particle.21) and this leads to the deﬁnition of the Hamiltonian in the form (this is an example of a Legendre transformation) H(x.3 Quantization 167 5. lead to the dynamical equations in the Hamiltonian form as x = {x. x) .3 Quantization Since we have a Lagrangian description for the classical Klein-Gordon ﬁeld theory. (5. the basic variables are φ(x) and ∂µ φ(x) and we have from (5.

29) then. ˙ ∂ φ(x) Π(x) = (5. Π(y)}x0 =y0 .25) In an analogous manner to the point particle mechanics in (5.21).23)) {φ(x).28) If we further assume the equal time canonical Poisson bracket relations between the ﬁeld variable φ(x) and the conjugate momentum Π(x) to be (see (5. we can show that the dynamical equation (namely.168 5 Free Klein-Gordon field theory L = L (φ(x). ∂µ φ(x)) . H} . H} .30) Let us examine all of this in detail for the case of the free KleinGordon theory. φ(y)}x0 =y0 {φ(x). = δ3 (x − y). ˙ Π(x) = {Π(x). (5. and a Hamiltonian d3 x H. (5. the KleinGordon equation in this case) can be written in the Hamiltonian form as ˙ φ(x) = {φ(x). allow us to deﬁne a Hamiltonian density as ˙ H = Π(x)φ(x) − L. (5. (5.27) H= (5. Π(y)}x0 =y0 = 0 = {Π(x).26) which will. Note that the Lagrangian density in this case is given by . we deﬁne a momentum canonically conjugate to the ﬁeld variable φ(x) as ∂L . in turn.

φ − ∇φ · ∇φ − 2 2 2 (5. = δ3 (x − y).3 Quantization 169 L = = m2 2 1 ∂µ φ∂ µ φ − φ 2 2 1 ˙2 1 m2 2 φ . φ(y)}x0 =y0 {φ(x).5.32) Consequently.34) With the equal time canonical Poisson bracket relations deﬁned to be (see (5. Π(y)}x0 =y0 we obtain = 0 = {Π(x). 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) .33) 1 2 1 m2 2 Π (x) + ∇φ · ∇φ + φ (x).35) .29)) {φ(x). we can identify the conjugate momentum with ∂L ˙ = φ(x). 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 (5.31) Therefore. Π(y)}x0 =y0 . 2 2 2 (5. (5. ˙ ∂ φ(x) Π(x) = (5.

x0 = ∇ · ∇φ(x) − m2 φ(x). Π2 (y) d3 y Π(y) {φ(x). since the Hamiltonian is time independent (this can be checked easily using the equations of motion and will also be shown in (5. H} = Π(x). ∇y φ(y)} + m2 φ(y) {Π(x). (5.29). x0 · ∇y −δ3 (x − y) +m2 φ y. x0 δ3 (x − y) = Π(x).37) Thus. Furthermore. we also have ˙ Π(x) = {Π(x). φ(y)} x0 =y0 = d3 y ∇y φ y. the Hamiltonian equations (5. Π(y)} x0 =y 0 d3 y Π y.36) where we have used the fact that. H} = = = = φ(x). lead to ˙ φ(x) = Π(x). −δ3 (x − y) (5.37).73)). in this case. we can choose the time coordinate for the ﬁelds in its deﬁnition to coincide with x0 in order that we can use the equal time Poisson bracket relations (5.36) and (5. 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).170 5 Free Klein-Gordon field theory ˙ φ(x) = {φ(x). d3 y 1 m2 2 1 2 Π (y) + ∇y φ · ∇y φ + φ (y) 2 2 2 x0 =y 0 = d3 y ∇y φ(y) ·{Π(x). (5. ˙ Π(x) = ∇ · ∇φ(x) − m2 φ(x) = ∇ 2 φ(x) − m2 φ(x).38) .

Π(y)]x0 =y0 . This is conventionally referred to as second quantization.41) of the KleinGordon equation deﬁne a complete basis and. Π(y)]x0 =y0 = 0 = [Π(x). (5. therefore.39) or. (Namely. brings out the classical Hamiltonian description of the classical KleinGordon ﬁeld theory. This is.40) (which is a second order Euler-Lagrange equation) and this discussion. in the ﬁrst quantization we quantize x’s and p’s. ¨ φ − ∇ 2 φ(x) + m2 φ(x) = 0. = iδ3 (x − y). here we are treating φ(x) as an operator which is quantized and not as a wavefunction as we had done earlier.40) Note that this is the same Klein-Gordon equation as we had studied earlier in chapter 1. see (5. we are supposed to treat φ(x) and Π(x) as Hermitian operators (because we are dealing with real ﬁelds. However.) 5. We already know that the plane wave solutions (1. we obtain the second order equation ¨ ˙ φ(x) = Π(x) = ∇ 2 φ(x) − m2 φ(x). φ(y)]x0 =y0 [φ(x). or. the free Klein-Gordon equation (1.4 Field decomposition 171 These are ﬁrst order equations as Hamiltonian equations should be and from these.4 Field decomposition Given any complete set of solutions of the classical Klein-Gordon equation. ∂µ ∂ µ + m2 φ(x) = 0. the quantization of such a system is quite straightforward. we can expand the ﬁeld operator in this basis as .2)) satisfying the equal time commutation relations (remember = 1) [φ(x). In the quantum theory.5. therefore. we can expand the ﬁeld operator φ(x) in such a basis. we quantize the “wavefunction”. (5. of course. Once we have the classical Hamiltonian description of a physical system. In the second quantization. but not the wave function.

we obtain 1 (2π) or.44) where the operator a(k) is no longer constrained by the equation of motion. (5. We also emphasize here that the hermiticity properties of a given ﬁeld and its Fourier transform are. (Recall that 1 k · x = k0 x0 − k · x and the factor (2π)3/2 is introduced for later convenience. As a result.41) in the Klein-Gordon equation. we obtain 1 (2π) 3 2 φ(x) = d4 k δ k2 − m2 e−ik·x a(k).41).42) using the ﬁeld expansion (5.45) (5. Substituting this back into the expansion (5. (5.43) d4 k −k2 + m2 φ(k)e−ik·x = 0. the Fourier transform has non-vanishing contribution (support) only on the massshell of the particle deﬁned by k2 = (k0 )2 − k2 = m2 . (5. This shows that unless k2 = m2 . φ(k) = 0. in general. diﬀerent.46) . Namely.172 5 Free Klein-Gordon field theory φ(x) = 1 (2π) 3 2 d4 k e−ik·x φ(k). 1 (2π) 3 2 3 2 ∂µ ∂ µ + m2 d4 k e−ik·x φ(k) = 0. we can denote φ(k) = δ k2 − m2 a(k). (5.41) where we recognize that the operators φ(x) and φ(k) are Fourier transforms of each other (up to a multiplicative factor).) Since φ(x) satisﬁes the Klein-Gordon equation ∂µ ∂ µ + m2 φ(x) = 0. (5.

or. or. we can write δ k 2 − m2 2 = δ (k0 )2 − Ek = = 1 δ k0 − Ek + δ k0 + Ek 2|k0 | 1 δ k0 − Ek + δ k0 + Ek 2Ek . (5.47) Correspondingly. k0 > 0 k k 2 = m 2 . k0 < 0 Figure 5.46). We note that the argument of the delta function vanishes for k 2 = m2 . (5. (k0 )2 = k2 + m2 . the ﬁeld expansion becomes .2: Mass shell on which φ(k) has support. k0 = ± k2 + m2 ≡ ±Ek .4 Field decomposition k0 173 k2 = m2.48) Using this relation in (5.5.

we obtain a† (k) = a(−k).2)). 2k0 (5.51).49) Changing k → −k in the second term in (5. φ† (x) = φ(x). (5.53) . k) 2Ek 0 +ik·x + eiEk x a(−Ek . k) 0 x0 +ik·x = 1 (2π) 3 2 d3 k 1 0 e−iEk x +ik·x a(Ek . 1 (2π) = 3 2 d3 k ik·x † e a (k) + e−ik·x a† (−k) 2k0 1 3 (2π) 2 d3 k −ik·x e a(k) + eik·x a(−k) .51) Comparing the left-hand side and the right-hand side of (5. we obtain 1 (2π) 2 3 φ(x) = d3 k −ik·x e a(k) + eik·x a(−k) . k) . 2k0 (5.49) and identifying k0 = Ek > 0. (5. Therefore. so that we can write 1 (2π) 3 2 a† (−k) = a(k).50) Let us note that we are dealing with a Hermitian ﬁeld (see (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 .52) φ(x) = d3 k −ik·x e a(k) + eik·x a† (k) . or.

56) .5.55) and in terms of these operators we can write the ﬁeld operator (5. φ(x) = φ(+) (x) + φ(−) (x). Thus.5 Creation and annihilation operators Given the ﬁeld expansion (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).5 Creation and annihilation operators 175 Here we are supposed to understand that k0 = Ek = k2 + m2 > 0.56) This unique decomposition of the ﬁeld operator φ(x) into positive and negative energy (frequency) parts is quite signiﬁcant as we will see in the course of our discussions. 2k0 (5. (5. a(k) = √ 2k0 a† (k) a† (k) = √ . conventionally one deﬁnes a(k) . (5.54) It is clear from (5. (5.54) that a(k) and a† (k) are really functions of the three momentum k alone.58) 5. eik·x a† (k). (5. In fact.57) so that we can write the ﬁeld operator as a sum of its positive and negative energy parts.53) as d3 k (2π)3 2k0 φ(x) = e−ik·x a(k) + eik·x a† (k) .

Namely.56).59) and (5.32) to be ˙ Π(x) = φ(x) = −i d3 k k0 e−ik·x a(k) − eik·x a† (k) . the expressions on the righthand side must also be time independent. taking the time derivative of the ﬁrst relation in (5. For example. This can also be checked explicitly.60) 2(2π)3 The two deﬁning relations in (5.176 5 Free Klein-Gordon field theory φ(x) = d3 k (2π)3 2k0 e−ik·x a(k) + eik·x a† (k) .61) that since the left-hand sides are independent of time. (5. we note from these relations that we can write the operators a(k) and ˙ a† (k) in terms of φ(x) and Π(x) as (recall that Π(x) = φ(x)) a(k) = = a† (k) = = − 1 (2π)3 2k0 i (2π)3 2k0 1 (2π)3 2k0 i (2π)3 2k0 d3 x eik·x k0 φ(x) + iΠ(x) ← → d3 x eik·x ∂t φ(x).61). It is clear from (5.61) ← → where ∂t is deﬁned in (1. It is important to emphasize here that in the expressions in (5. (5.60) are invertible.59) we obtain the expansion for the conjugate momentum in (5.61) we obtain (neglecting the overall multiplicative factors) . d3 x e−ik·x k0 φ(x) − iΠ(x) ← → d3 x e−ik·x ∂t φ(x). it is assumed that k0 = Ek (namely. (5. k0 is not an independent variable).

61) are indeed time independent. Π(y)]x0 =y0 = − ′ ′ ′ x0 =y 0 (5. therefore. must be time independent).3)) and have √ made the identiﬁcation k0 = Ek = k2 + m2 . φ(y)]x0 =y0 = ′ × e−ik·x−ik ·y a(k). a† (k′ ) . [Π(x). a(k′ ) − e−ik·x+ik ·y a(k). a† (k′ ) + eik·x−ik ·y a† (k).5 Creation and annihilation operators 177 ∂t = = = = ← → d3 x eik·x ∂t φ(x) 2 2 d3 x eik·x (∂t φ(x)) − (∂t eik·x ) φ(x) 2 d3 x eik·x (∇2 − m2 )φ(x) − (∂t eik·x ) φ(x) 2 d3 x −∂t + ∇2 − m2 eik·x φ(x) d3 x (k0 )2 − k2 − m2 eik·x φ(x) = 0. a(k′ ) + eik·x+ik ·y a† (k). the time independence of the second expression in (5.40)) we obtain d3 k (2π)3 2k0 d3 k ′ (2π)3 2k′ 0 ′ [φ(x). a† (k′ ) = 0.62) Here.5.63) d kd k 3 3 ′ √ k0 k′ 0 2(2π)3 ′ × e−ik·x−ik ·y a(k).61) can also be derived in a simple manner (or note that the second relation is the Hermitian conjugate of the ﬁrst and. (5. Imposing now the quantization relations between φ(x) and Π(x) (see (5. integrated the space derivatives by parts assuming that the surface terms vanish (see (5. a(k′ ) + e−ik·x+ik ·y a(k). Similarly. This shows explicitly that the integrals in (5. in the intermediate steps. we have used the fact that φ(x) satisﬁes the Klein-Gordon equation.

[φ(x).66) This shows that the operators a(k) and a† (k). Thus. a† (k′ ) = δ3 k − k′ .55). Any other redeﬁnition will introduce a multiplicative factor into the commutation relations. a(k′ ) + eik·x+ik ·y a† (k). recalling that the integrals are ime independent.178 ′ 5 Free Klein-Gordon field theory − eik·x−ik ·y a† (k).64) i (2π)3 d3 k d3 k ′ k′ 0 4k0 ′ × e−ik·x−ik ·y a(k). a† (k′ ) = iδ3 (x − y).) a(k).61). Π(y)]x0 =y0 = − ′ ′ x0 =y 0 (5. k′ 0 φ(y) − iΠ(y) x0 =y 0 . (5. a(k′ ) − eik·x+ik ·y a† (k). a(k). a† (k′ ) = 0. which are the coeﬃcients of expansion of the ﬁeld operator in a plane wave basis. a† (k′ ) + eik·x−ik ·y a† (k). it appears that there is an inﬁnite number of such operators in the present case – one for every value of the momentum k. ′ ′ x0 =y 0 (5. for example. However. a(k′ ) − e−ik·x+ik ·y a(k).65) From these.66) can also be obtained more directly from the inversion formulae in (5. a(k′ ) = 0 = a† (k). we have a(k). a† (k′ ) . have commutation relations analogous to the annihilation and creation operators of a harmonic oscillator. The commutation relations in (5. we can deduce the fundamental commutation relations between the coeﬃcients of expansion to be (This nice form arises because of the redeﬁnition (5. a† (k′ ) = ′ d3 xd3 y √ eik·x−ik ·y (2π)3 4k0 k′ 0 × k0 φ(x) + iΠ(x).

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

.69) − a† (k)a(k) + e2ik 0 x0 a† (k)a† (−k) . Π(y)] − ik′ 0 [Π(x). To understand the meaning of these operators further. 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). independent of whether k′ = ±k. let us look at the Hamiltonian of the system. We note that d3 k d3 k ′ √ 0 ′ 0 k k 2(2π)3 ′ d3 x Π2 (x) = − ′ d3 x × e−i(k+k )·x a(k)a(k′ ) − e−i(k−k )·x a(k)a† (k′ ) − ei(k−k )·x a† (k)a(k′ ) + ei(k+k )·x a† (k)a† (k′ ) = − d3 k d3 k′ √ 0 ′ 0 −i(k0 +k′ 0 )x0 3 k k e δ k + k′ a(k)a(k′ ) 2 0 −k ′ 0 )x0 ′ ′ −e−i(k −ei(k + ei(k = − 1 2 δ3 k − k′ a(k)a† (k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) 0 x0 0 +k ′ 0 )x0 d3 k k0 e−2ik a(k)a(−k) − a(k)a† (k) (5.68) where we have used the fact that k′ 0 = k0 for k′ = −k. φ(y)] = = = (k0 − k′ 0 ) √ 4k0 k′ 0 (k0 − k′ 0 ) √ 4k0 k′ 0 d3 xd3 y −ik·x−ik′ ·y 3 e δ (x − y) (2π)3 d3 x −i(k+k′ )·x e (2π)3 x0 =y 0 x0 =y 0 (k0 − k′ 0 ) −i(k0 +k′ 0 )x0 3 √ δ (k + k′ ) = 0.180 5 Free Klein-Gordon field theory × k0 φ(x) − iΠ(x). Here. e 4k0 k′ 0 (5. we have used the fact that.

5. we can calculate d3 x ∇φ(x) · ∇φ(x) = − ′ 1 (2π)3 d3 x ′ d3 k d3 k ′ √ √ k · k′ 2k0 2k′ 0 × e−i(k+k )·x a(k)a(k′ ) − e−i(k−k )·x a(k)a† (k′ ) − ei(k−k )·x a† (k)a(k′ ) + ei(k+k )·x a† (k)a† (k′ ) = − d3 k d3 k ′ √ √ k · k′ 0 ′0 2k 2k 0 +k ′ 0 )x0 ′ ′ × e−i(k −e−i(k −ei(k + ei(k = − 1 2 δ3 k + k′ a(k)a(k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a(k)a† (k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) 0 +k ′ 0 )x0 d3 k k2 0 0 −e−2ik x a(k)a(−k) − a(k)a† (k) 0 k 0 x0 − a† (k)a(k) − e2ik d3 x φ2 (x) = 1 (2π)3 ′ a† (k)a† (−k) .5 Creation and annihilation operators 181 k′ 0 = (k′ )2 + m2 = k2 + m2 = k0 .70) Similarly. d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ (5.71) d3 x × e−i(k+k )·x a(k)a(k′ ) + e−i(k−k )·x a(k)a† (k′ ) +ei(k−k )·x a† (k)a(k′ ) + ei(k+k )·x a† (k)a† (k′ ) = d3 k d3 k ′ √ √ 2k0 2k′ 0 × e−i(k +e−i(k 0 +k ′ 0 )x0 ′ ′ δ3 k + k′ a(k)a(k′ ) 0 −k ′ 0 )x0 δ3 k − k′ a(k)a† (k′ ) . (5.

72) into the expression for the Hamiltonian (5.34). 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) .72) Substituting (5.69). (5. (5.182 +ei(k 0 −k ′ 0 )x0 5 Free Klein-Gordon field theory δ3 k − k′ a† (k)a(k′ ) δ3 k + k′ a† (k)a† (k′ ) + ei(k = 1 2 0 +k ′ 0 )x0 d3 k 0 0 1 e−2ik x a(k)a(−k) + a(k)a† (k) 0 k 0 x0 + a† (k)a(k) + e2ik a† (k)a† (−k) .71) and (5.

H = a† (k).5 Creation and annihilation operators 183 (5.73) = d3 k Ek a(k)a† (k) + a† (k)a(k) . we see from (5.54) k0 = k2 + m2 = Ek . we then conclude that we can think of this as the Hamiltonian for an inﬁnite collection of decoupled harmonic oscillators. Thus. H] = a(k). But since this is an additive constant .73).75) = −Ek a† (k). the zero point energy of such a system would be inﬁnite. It follows now that Ek′ a(k′ )a† (k′ ) + a† (k′ )a(k′ ) 2 [a(k). a† (k′ ) + a(k). a(k′ ) a† (k′ ) + a† (k′ ) a† (k). a† (k). = = d3 k ′ d3 k ′ d3 k ′ Ek′ a(k′ ) a(k). we can indeed think of them as the annihilation and creation operators of a harmonic oscillator system. 2 where we have used the relation (5. (5. = = d3 k ′ Ek′ 2 d3 k ′ Ek′ a(k′ )a† (k′ ) + a† (k′ )a(k′ ) 2 a† (k). a† (k′ ) a(k′ ) 2 Ek′ a(k′ )δ3 k − k′ + δ3 k − k′ a(k′ ) 2 = Ek a(k). From the form of the Hamiltonian for the Klein-Gordon ﬁeld in (5. a(k′ ) d3 k ′ Ek′ −δ3 k − k′ a† (k′ ) − a† (k′ )δ3 k − k′ 2 (5. Clearly.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 . The two relations in (5.74) Thus.5.75) show that the operators a(k) and a† (k) annihilate and create respectively a quantum of energy Ek .

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

N = = a(k). a(k′ ) = −a† (k′ )δ3 (k − k′ ). . It now follows from the deﬁnition of the number operators in (5. a† (k′ ) a(k′ ) = a(k′ )δ3 (k − k′ ). a† (k′ )a(k′ ) = a† (k′ ) a† (k).80) [a(k). a† (k).79) that a(k). a† (k).79) It is worth noting from the derivation in (5.77) how the normal ordering has redeﬁned away an inﬁnite zero point energy (which basically would arise from commuting a(k)a† (k) to bring it to the normal ordered form). (5.81) which reﬂects again the fact that a(k) and a† (k) merely lower or raise the number of quanta of momentum k by one unit. N (k′ ) = = a(k).5. a† (k′ )a(k′ ) a(k). Equation (5.80) then leads to (5. d3 k′ N (k′ ) d3 k′ a(k′ )δ3 (k − k′ ) = a(k).78) and (5.5 Creation and annihilation operators 185 N= d3 k N (k) = d3 k a† (k)a(k). N (k′ ) = a† (k). N ] = = a† (k). (5. d3 k′ N (k′ ) d3 k′ −a† (k′ )δ3 (k − k′ ) = −a† (k).

H = −Ek a† (k). E|H|E E| d3 k Ek a† (k)a(k)|E d3 k Ek E|a† (k)a(k)|E (5. we do not have the problem of negative energy states that we had in the single particle theory. consequently. which can be achieved by quantizing the system in a box. Let us next recall that the operators a(k) and a† (k) satisfy the commutation relations (see (5. (5. (5.6 Energy eigenstates 5 Free Klein-Gordon field theory From now on.85) . This shows that the eigenvalues of the Hamiltonian have to be positive semideﬁnite and in the second quantized ﬁeld theory. a† (k). the integrand is positive semideﬁnite. H] = Ek a(k). let us consider the normal ordered Hamiltonian (5.77) (as describing the quantum free Klein-Gordon theory) of the form H= d3 k Ek a† (k)a(k). It is clear from (5. (5.83) that E = = = ≥ 0.83) where we are assuming that the state |E is normalized and that the eigenvalue E is discrete for simplicity.77) and (5. the inner product represents the norm of the state a(k)|E ) and.82) Let us consider an energy eigenstate |E of this Hamiltonian satisfying H|E = E|E .75)) [a(k).84) This is because both Ek as well as the inner product of states is positive deﬁnite (namely.186 5.

= −Ek a† (k)|E . we have [a(k). such that a(k)|Emin = 0. (5. or. Therefore. or. raises the energy value. (5.88) and we cannot lower the energy any further. The annihilation operator a(k) lowers the energy eigenvalue as it should. if |E represents an energy eigenstate with eigenvalue E. therefore. d3 k Ek Emin |a† (k)a(k)|Emin Emin = = 0. But as we have seen in (5. Namely. it would appear that by applying the annihilation operator successively. (5.5. Let us next note that since a(k) acting on an energy eigenstate lowers the energy. a† (k)H|E H a† (k)|E = −Ek a† (k)|E . The creation operator a† (k). clearly. In such a state.89) This is. H |E or. therefore. there must exist a state (energy eigenstate) with a minimum energy. then {a(k)|E } is also an eigenstate of energy with the lower energy value E − Ek . a(k)H|E = Ek a(k)|E . H] |E or. we had redeﬁned . if |E is an energy eigenstate with energy eigenvalue E. the ground state of the system or the vacuum state (recall that by normal ordering the Hamiltonian. Similarly.86) − Ha(k)|E H {a(k)|E } = (E − Ek ) {a(k)|E } . we note that a† (k). the energy eigenvalue is bounded from below (E ≥ 0). we can lower the energy eigenvalue arbitrarily.87) − Ha† (k)|E = (E + Ek ) a† (k)|E In other words. = Ek a(k)|E . |Emin . (5. then a† (k) acting on it gives another energy eigenstate with the higher value of energy E+Ek .84). .6 Energy eigenstates 187 As a result.

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

nℓ . nℓ . and .) P= then d3 k k a† (k)a(k) = d3 k k N (k). kℓ . kℓ (5. kℓ . a† (k1 ) = n1 a† (k1 ) which also leads to n1 n1 δ3 (k − k1 ) . of course. k1 . the state |n1 . kℓ = d3 k Ek N (k)|n1 . the natural deﬁnition from considerations of covariance. nℓ .80)) N (k).6 Energy eigenstates 189 This relation can be easily shown using the identity (see also (5.5. kℓ contains n1 quanta with four µ µ momentum k1 .94) N (k)|n1 . 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. Namely. n2 . n2 . · · · . · · · . kℓ = (5. n2 quanta with four momentum k2 and so on. · · · . (5. k1 .97) H|n1 . k2 . nℓ . (a† (k1 ))n1 = a† (k) a(k). (5. nℓ . k1 . n2 .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).95) n1 δ3 (k − k1 ) + · · · + nℓ δ3 (k − kℓ ) |n1 . However. · · · . k2 . nℓ . · · · .98) = (n1 Ek1 + n2 Ek2 + · · · + nℓ Ekℓ ) |n1 . k1 . k2 . · · · . k1 .

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

k2 . it is clear that 0|φ(x)|0 = 0. (5. the only nonzero matrix element of the ﬁeld operator involving the vacuum state is given by k|φ(x)|0 = k|φ(x) . |φ(x) = φ(x)|0 = φ(−) (x)|0 . k1 . k1 ). as a result.108) where we have used (5. Note that the particles described by such states are necessarily identical (although with diﬀerent energy and momentum) and. Let us next consider the state that is produced by the ﬁeld operator acting on the vacuum. kℓ . Similarly. k2 ) and so on. n2 . (5. nℓ . k).107) This is. (5.105) as well as the fact that only the ﬁeld operator φ(x) depends on the space-time coordinates on which the .5. of course. therefore. This function satisﬁes ∂µ ∂ µ + m2 k|φ(x) = k| ∂µ ∂ µ + m2 φ(x)|0 = 0. µ n2 particles with k2 = (Ek2 . therefore. we can show that the state |n1 .106) In fact. ﬁeld theories naturally describe systems of many identical particles.105) Since φ(x) is linear in the creation and the annihilation operators. we can think of this as the one particle state with four momentum kµ = (Ek .104) µ can be thought of as the state with n1 particles with k1 = (Ek1 . (5. (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.7 Physical meaning of energy eigenstates 191 This state. · · · .

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

k′ |φ(−) (x1 ) φ(−) (x2 ) |0 . φ (x2 ) = k. by construction.109). (5.112) can be thought of as the one particle state in the conﬁguration space – describing the quantum mechanical state of the single particle at the coordinate x. The state = φ(−) (x)|0 . Furthermore. We can similarly construct multi-particle states in the conﬁguration space of the form |φ (x1 ) . k′ |φ (x1 ) .114) would give the probability for ﬁnding two Klein-Gordon particles with four momenta kµ = (Ek . Such states are physically meaningful only if the time coordinates are equal (namely. φ (x2 ) . They can also be shown to deﬁne a complete basis and describe what is known as the Fock space for the system.56)). (5. xn .7 Physical meaning of energy eigenstates 193 Thus. such a state contains a superposition of all possible momentum states (which follows from the ﬁeld decomposition in (5. the absolute square of the amplitude (with x0 = x0 ) 2 1 k. k) and k′ µ = (Ek′ . k′ ) at the coordinates x1 and x2 at a given time. (5. · · · .113) which will describe a state with n particles at coordinates x1 .5. These states naturally lead to quantum mechanical probabilities which are non-negative. φ (xn ) = φ(−) (x1 ) φ(−) (x2 ) · · · φ(−) (xn ) |0 . · · · .111) |φ(x) (5. As we have seen. for example. x0 = x0 = · · · = x0 ). positive semi-deﬁnite and there is no problem of negative . which is what we have seen explicitly in (5. the wave functional for a single particle with a deﬁnite energymomentum has the form ψ(φ(x)) = φ(x)|ψ = φ(x)|k . Such probabilities are. such conﬁguran 1 2 tion states are automatically symmetric as we would expect for a system of identical Bose particles. Thus.

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

(5. therefore. Thus. − m2 (5.117).118) d4 y −δ4 (x − y) J(y) (5.119) = J(x). (5. 4 (2π) (2π)4 G(k) = k2 1 .121) In other words. (2π)4 and substituting them back into (5. deﬁning the Fourier transforms δ4 (x − y) = G(x − y) = 1 (2π)4 d4 k e−ik·(x−y) .8 Green’s functions 195 then the particular solution of the inhomogeneous equation (5. or. 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. 1 1 −k2 + m2 G(k) = − . . the Fourier transformation turns the partial diﬀerential equation into an algebraic equation which is trivial to solve. The Green’s function is. 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).120) d4 k −ik·(x−y) e G(k).115) can be written as φ(x) = − so that ∂µ ∂ µ + m2 φ(x) = − = − µ d4 y ∂xµ ∂x + m2 G(x − y)J(y) d4 y G(x − y)J(y).5. we obtain ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). or.

or.121) back into (5. e Let us note that the integrand of the Green’s function in (5. 5.123) which lie on the real axis in the complex k0 -plane as shown in Fig. 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) . Mathematically. the Green’s function is not uniquely deﬁned until we specify a contour of integration in the complex k0 -plane and specifying a contour simply corresponds to specifying a boundary condition for the Green’s function. this is expressed by writing . (k0 )2 = k2 + m2 .122) G(x − y) = = This explicitly shows that the Green’s function is Poincar´ invariant. (2π)4 k2 − m2 (5. Im k 0 −Ek Ek Re k 0 Figure 5. substituting (5.4.122) has poles at k2 − m2 = 0.196 5 Free Klein-Gordon field theory Thus. Therefore.3.3: Poles of the integrand in the complex k0 -plane.120). If we choose a contour of the form shown in Fig. (5. or. then. k0 = ±Ek . this is equivalent to moving the poles inﬁnitesimally into the upper half plane. 5.

or.124) are in the upper half plane. if we enclose the contour in the lower half plane. for x0 − y 0 > 0.4: Choice of the contour in the complex k0 -plane for the advanced Green’s function. the poles of the integrand occur at k2 − m2 − ik0 η = (k0 )2 − k2 − m2 − ik0 η = 0. (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 . or. k0 − iǫ = ±Ek . GA (x − y) = lim η→0+ e−ik·(x−y) d4 k . the integral will vanish.126) For x0 − y 0 < 0. 2 Since in this case the poles in (5. then we must close the contour in the lower half plane for the damping of the exponential in (5. iη k − 2 0 2 ≃ k 2 + m2 . We note that if x0 − y 0 > 0. (5. we conclude that GA (x − y) = 0. (5.124). Therefore.8 Green’s functions 197 Im k 0 Im k 0 −Ek Ek Re k 0 → −Ek + iǫ Ek + iǫ Re k 0 Figure 5.5. or.125) where we have identiﬁed ǫ = η . k0 = ±Ek + iǫ.124) In this case. on the other hand.

For x0 − y 0 < 0.127) d3 k eik·(x−y) sin Ek x0 − y 0 . the poles of the integrand in (5.198 5 Free Klein-Gordon field theory case we can evaluate the integral by the method of residues. 5. we can write (we denote the argument of the Green’s function by x for simplicity) G (x) = A 0. for x0 < 0. GA (x) = θ −x0 d3 k eik·x sin Ek x0 .129) would appear at .129) In this case. 4 k 2 − m2 + ik 0 η (2π) GR (x) = lim η→0+ (5. (2π)3 Ek This is known as the advanced Green’s function and has support only in the past light cone.5 or equivalently if we push both the poles to the lower half plane. d3 k eik·x (2π)3 Ek for x0 > 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 ) (5.128) or. sin Ek x0 . (2π)3 Ek Similarly. Explicitly. if we choose the contour in the complex k0 -plane as shown in Fig. then. we can express this mathematically by deﬁning e−ik·x d4 k . (5.

on the other hand. we conclude that GR (x) = 0. or. therefore. Note that if x0 < 0. x0 < 0. k0 + iǫ ≃ ±Ek . In this case.130) where we have again identiﬁed ǫ = η .5: Choice of the contour in the complex k0 -plane for the retarded Green’s function. since both the 2 poles are in the lower half plane.131) For x0 > 0. k2 − m2 + ik0 η = (k0 )2 − k2 − m2 + ik0 η = 0. or. we have to close the contour in the upper half plane for the damping of the exponential in (5.5. 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. k0 = ±Ek − iǫ. k0 + iη 2 2 2 ≃ k2 + m2 = Ek .129) and. or. then the integral will vanish. (5. if we close the contour in the upper half plane.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.

132) arises because the contour is clockwise in the lower half plane. the average of the two. (2π)3 Ek (5. sin Ek x0 . 1 GA (x) + GR (x) .133) or. (5. Thus.) It is clear from (5. GR (x) = −θ x0 d3 k eik·x sin Ek x0 . On the other hand. we see that we can write 0. (2π)3 Ek Note that this Green’s function has support only in the future light cone and is known as the retarded Green’s function.128) and (5.117) with a delta function source. (5. the diﬀerence between GA (x) and GR (x) would deﬁne a function which would satisfy the homogeneous (Klein-Gordon) equation and . Therefore. 2 (5. for x0 > 0.133) that GR (x) = GA (−x).135) would also deﬁne a Green’s function corresponding to the contour picking up the principal values at the poles.132) e−i(−Ek −iǫ)x + −2Ek 0 +ik·x The overall negative sign in (5. (The advanced and the retarded Green’s functions have similar form except for their support and the overall sign. namely.134) Both the retarded and the advanced Green’s functions satisfy the inhomogeneous equation (5. − d3 k eik·x (2π)3 Ek GR (x) = for x0 < 0.200 = lim (−2πi) 5 Free Klein-Gordon field theory ǫ→0+ d3 k e−i(Ek −iǫ)x (2π)4 2Ek 0 +ik·x = − = − i 2 d3 k eik·x −iEk x0 0 − eiEk x e 3 E (2π) k d3 k eik·x sin Ek x0 .

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

this can be implemented by deﬁning d4 k e−ik·x . in this case. (2π)4 k2 − m2 + iη GF (x) = lim η→0+ (5.142) . (k0 )2 − Ek − k0 = ± Ek − iη 2Ek iη 2Ek 2 ≃ 0. or. (5. or. to the lower half plane. Mathematically.140) occur at k2 − m2 + iη = 0. (5. Since there are poles in both halves of the complex k0 -plane. 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. the poles of the integrand in (5.202 Im k 0 5 Free Klein-Gordon field theory Im k 0 −Ek Ek Re k 0 → −Ek + iǫ Ek − iǫ Re k 0 Figure 5. 2 (k0 )2 − Ek + iη = 0.6: Choice of the contour in the complex k0 -plane for the Feynman Green’s function.141) η where we have identiﬁed ǫ = 2Ek with η → 0+ . = ± (Ek − iǫ) . there will be a nontrivial contribution independent of whether we close the contour in the upper half or in the lower half plane. or.

where. Here the contour is clockwise leading to an overall negative sign. we have d3 k (2π)4 dk0 e−ik·x (k0 − Ek + iǫ) (k0 + Ek − iǫ) 0 (5.143) = G(−) (x). The Feynman Green’s function clearly has support in both the future as well as the past light cones and can be written as .5. if x0 > 0. 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 (5.144) 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 (5.145) = −G(+) (x). we have used k ↔ −k in the last step (and we understand that k0 = Ek ). Therefore. for x0 > 0. for x0 < 0. On the other hand. 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.8 Green’s functions 203 Therefore.

GF (−x) = GF (x).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. It is deﬁned as e−ik·x d4 k .136) as d3 k eik·x sin Ek x0 = G(+) (x) + G(−) (x). (2π)4 k2 − m2 − iη G(x) = lim η→0+ (5.143) and (5.143) 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). (2π)3 (5. However. Note from (5.147) Therefore. namely.150) The functions G(+) (x) and G(−) (x) (see (5.145) that i 2 d3 k eik·x = −G(−) (x). (2π)3 Ek G(x) = (5. We note from the structure of G(−) (x) in (5. (2π)3 d4 k θ(−k0 )δ(k2 − m2 ) e−ik·x .151) G(+) (x) = i G(−) (x) = −i .146) G(+) (−x) = (5. we will not go into this in more detail here.145) that we can write the Schwinger function (5. (5.204 5 Free Klein-Gordon field theory GF (x) = −θ x0 G(+) (x) + θ −x0 G(−) (x).148) There is yet another choice of the contour and. (2π)3 k0 (5.145)) can be written in the manifestly covariant forms d4 k θ(k0 )δ(k2 − m2 ) e−ik·x . the Feynman Green’s function is an even function.143) and G(+) (x) in (5. therefore.

128).57) that (we recall that k0 = Ek = 0) √ (5.154) which will be useful in our later discussions.153) where ǫ(x) = (θ(x) − θ(−x)) is known as the sign function (or the alternating step function).152) and all the Green’s functions can be expressed as linear combinations of these two fundamental Green’s functions. (2π)3 (5. 5.151) that the Schwinger function has the manifestly covariant representation G(x) = G(+) (x) + G(−) (x) = i d4 k ǫ(k0 )δ(k2 − m2 ) e−ik·x .155) k2 + m2 > . GR (x) = −θ(x0 )G(x) = −θ(x0 ) G(+) (x) + G(−) (x) .9 Covariant commutation relations To make contact between the quantum ﬁeld theory and the various Green’s functions that we have constructed. (5. They satisfy (see (5. For completeness. (5.9 Covariant commutation relations 205 and are correspondingly known as the positive and the negative energy (frequency) Green’s functions. where we note from (5.5. we note from (5. Let us recall from (5.151)) G(+) (x) ∗ = G(−) (x). (5.150) that we can write GA (x) = θ(−x0 )G(x) = θ(−x0 ) G(+) (x) + G(−) (x) . let us calculate some of the covariant commutation relations satisﬁed by the ﬁeld operators.58) that φ(x) = φ(+) (x) + φ(−) (x). We note from (5.133) and (5.

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

using the results in (5. It is now clear that the Schwinger function satisﬁes the homogeneous equation µ ∂x µ ∂x + m2 G(x − y) = 0. φ(y)] = φ(+) (x) + φ(−) (x). the commutator of two ﬁeld operators. . a† (k′ ) e 2k0 2k′ 0 d3 k d3 k′ −ik·x+ik′·y 3 √ √ e δ k − k′ 2k0 2k′ 0 d3 k −ik·(x−y) e 2k0 (5. (2π)3 Ek (5. φ(−) (y) + φ(−) (x). φ(+) (y) = iG(+) (y − x) = −iG(−) (x − y).158) = −iG(+) (x − y). φ(+) (y) + φ(−) (y) = φ(+) (x).158) and (5.158) and (5.40)) and.160) Namely.159). It follows now that φ(−) (x). φ(+) (y) = −iG(+) (x − y) − iG(−) (x − y) = −iG(x − y) = −i d3 k eik·(x−y) sin Ek x0 − y 0 . (5. at unequal times. we can evaluate [φ(x).161) simply because the ﬁeld operator φ(x) satisﬁes the Klein-Gordon equation.9 Covariant commutation relations 207 = = = 1 (2π)3 1 (2π)3 1 (2π)3 d3 k d3 k′ −ik·x+ik′·y √ √ a(k). is proportional to the Schwinger Green’s function. furthermore.5. (5.159) are not restricted to have equal time arguments any more (unlike in (5.159) Note that the ﬁeld operators in the commutators (5.

x0 =y 0 (5. φ(y)] = 0. this relation can also be seen to imply that [φ(x).) On the other hand.162) These commutation relations are known as covariant commutation relations simply because the (scalar) Green’s functions are invariant under Lorentz transformations. for x0 = y 0 d3 k eik·(x−y) sin Ek x0 − y 0 (2π)3 Ek G(x − y)|x0 =y0 = = 0.165) x0 =y 0 We note that by a Lorentz transformation. namely. Equation (5. φ(y)]x0 =y0 = −iG(x − y) = 0. Namely.40)) [φ(x). (5. two measurements at space-like separations should not inﬂuence each other. a light signal cannot connect two space-like points and.160) leads to the familiar relation (see (5. for (x − y)2 < 0.164) which is expected from the anti-symmetry of the Schwinger function (5. We also note that . (5. (5. This is known as the principle of microscopic causality.166) In other words. for space-like separations.163) where Λµν represents a Lorentz transformation. This is consistent with our intuitive expectation. (5.160) also shows that two ﬁeld operators φ(x) and φ(y) do not commute for arbitrary values of the coordinates. G(Λx) = G(x). the commutator of two ﬁeld operators vanishes.138) and (5.208 5 Free Klein-Gordon field theory ∂µ ∂ µ + m2 φ(x) = 0. as can be checked even in the simple classical Poisson bracket relations for a one dimensional free particle or a harmonic oscillator. therefore. (The non-commutativity is a reﬂection of the Hamiltonian dynamics.

φ(y)] = −iθ(x0 − y 0 ) G(+) (x − y) + G(−) (x − y) = −iθ(x0 − y 0 )G(x − y) = iGR (x − y). (5.5. Bjorken and S.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). namely. Π(y)]x0 =y0 = x0 =y 0 = −i ∂G(x − y) ∂y 0 x0 =y 0 = iδ3 (x − y). Drell.136). 5. ˙ φ(x).169) .150) and (5. McGrawHill.168) To conclude this section. New York. let us note from (5. J. (5. (5. 1964.10 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. D.167) d3 k ik·(x−y) e = −δ3 (x − y). (2π)3 This is consistent with our earlier ﬁeld quantization rule (5. φ(y)] = −iθ(y 0 − x0 ) G(+) (x − y) + G(−) (x − y) = −iθ(y 0 − x0 )G(x − y) = −iGA (x − y). Relativistic Quantum Fields.160) as well as the relations in (5. φ(y) [φ(x). θ(y 0 − x0 ) [φ(x). which will be useful later.10 References 1.40).

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

Obviously. ∂µ φ(x) → (6. we say that the dynamics described by the action (6.3) ′ ∂µ φ′ (x′ ) .1) is invariant under the transformations (6.2) φ(x) → φ′ (x′ ) .3) if the action does not change under these transformations. 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. in the present case.1) where.1 N¨ther’s theorem o In trying to extend the free Klein-Gordon ﬁeld theory to include interactions. Let us suppose that we have a dynamical system described by the action d4 x L. we assume L = L(φ(x). Namely. the ﬁrst question that we face is how we can choose one interaction term over another. (6.Chapter 6 Self-interacting scalar ﬁeld theory 6. Under a general transformation of the form xµ → x′ µ . if 211 . ∂µ φ(x)). S= (6.

the transformations in (6. Our discussion of symmetries applies to both spacetimetransformations where space-time coordinates are transformed as well as internal symmetry transformations where space-time coordinates are unaﬀected by the transformation. the Euler-Lagrange equations for the primed and the unprimed systems would remain form invariant. It is worth emphasizing here that (6. namely. ∂µ φ′ x′ ).3) deﬁne a symmetry of the system. ∂x (6.7) . Symmetries have interesting consequences for continuous transformations.5) and only the dynamical variables of the theory transform. Thus. ∂µ φ′ x′ δS = or. ∂µ φ(x)) = 0.6) which holds for most global space-time symmetries as well as internal symmetries. x′ µ = xµ . ∂µ φ′ (x) − L (φ(x). d4 x L φ′ (x). invariance of the action under the inﬁnitesimal forms of the transformations in (6. (6. ∂µ φ(x)) = 0.212 6 Self-interacting scalar field theory S= d4 x L (φ(x). In such a case. ∂µ φ′ (x) − d4 x L φ′ (x). or.4) then. − d4 x L (φ(x).3). ∂µ φ(x)) = 0. as indicated in (6. It is clear that in such a case. (6.3) would imply ′ d4 x′ L φ′ x′ . (6. ∂µ φ(x)) = ′ d4 x′ L(φ′ x′ . Such transformations are known as internal symmetry transformations to be contrasted with space-time transformations where space-time coordinates transform along with the dynamical variables. let us consider an inﬁnitesimal transformation with ∂x′ = 1. d4 x L (φ(x).3) includes a very interesting class of transformations where the space-time coordinates do not change.

(6.¨ 6. On the other hand.9) we can calculate explicitly (we remember that δφ(x) is inﬁnitesimal and. (6. ∂µ φ(x)) + δφ(x) −L (φ(x). In fact.11) ∂L ∂L + δ(∂µ φ(x)) ∂φ(x) ∂∂µ φ(x) = δφ(x)∂µ = ∂µ δφ(x) . keep only linear terms in δφ(x)) L φ′ (x). ∂µ φ′ (x) − L (φ(x). ∂µ φ(x)) = L (φ(x). this is indeed the case for internal symmetry transformations. ∂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) . ∂µ φ(x)) = ∂µ K µ . so that δ(∂µ φ(x)) = ∂µ φ′ (x) − ∂µ φ(x) = ∂µ δφ(x). ∂µ φ′ (x) − L (φ(x).8) which must hold independent of the use of equations of motion for the system under study. (6.1 Nother’s theorem 213 where we have identiﬁed the integration variable in the ﬁrst term on the right-hand side to be x (instead of x′ ) in the intermediate step. (6. therefore. Clearly.10) (6.7) will hold if L φ′ (x). This is quite general and it is possible to have symmetry transformations for which K µ = 0. as we will see later. 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).

x) . ∂µ J µ (x) = 0. on the other hand. namely. (Note that. not all such variations will deﬁne a symmetry. there is no restriction on δφ and. (6. ∂∂µ φ(x) J µ (x) = δφ(x) (6. Namely. we obtain ∂µ δφ(x) or.8) and (6. ∂µ δφ(x) ∂L ∂∂µ φ(x) = ∂µ K µ = 0. we can deﬁne a charge Q= d3 x J 0 (t. First. we note that . In fact. However.13) which is conserved. this is always true when the equations of motion are used. Its tensor structure is determined by the tensor structure of the inﬁnitesimal parameter of transformation. therefore. (6. the conserved current independent of the parameter of transformation is not always a vector. on φ′ .14) Several comments are in order here. for any inﬁnitesimal variation. in general. we can deﬁne a current ∂L − K µ.) Comparing (6.11). Second.15) which will be a constant (in time) and will generate the symmetry transformations. given a conserved current. (6.12) ∂L − Kµ ∂∂µ φ(x) This shows that whenever there is a continuous symmetry associated with a system.214 6 Self-interacting scalar field theory Here we have used the Euler-Lagrange equation in the intermediate steps. This is just the principle of least action. For a symmetry. φ′ must also satisfy the same equation of motion in the transformed frame as the original ﬁeld.

we are dealing with a scalar ﬁeld. with the usual assumptions on the asymptotic fall oﬀ of the ﬁeld operators (see (5. if we have a continuous symmetry in the theory (transformations under which the action is invariant). The converse is also true. Thus. the integral of a total divergence (∇ · J) vanishes. then it generates inﬁnitesimal transformations (through commutation relations) which deﬁne a symmetry of the theory. the time independence of Q also means that it commutes with the Hamiltonian of the theory [Q. (6. Q and H can have simultaneous eigenstates.1 Space-time translation. . In the operator language. (6. As an example of the consequences of N¨ther’s theorem.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. H] = 0. This is known as N¨ther’s theorem and is quite important in the study of symmeo tries in dynamical systems.18) where the parameter of translation ǫµ is assumed to be inﬁnitesimal and constant (global). there exists a conserved charge which is the generator of these inﬁnitesimal symmetry transformations. Namely.17) and. or.¨ 6. consequently. Since. (6. if there exists a conserved charge in a given theory. δxµ = x′µ − xµ = ǫµ . In this case. Q corresponds to the generator of the inﬁnitesimal symmetry transformations of the theory.3)).16) Here we have used the fact that. let us consider the simple case of an inﬁnitesimal o global space-time translation deﬁned by xµ → x′µ = xµ + ǫµ .1. 6.

then.22) . ∂µ φ′ (x) − L (φ(x).7) and (6.21) = −ǫµ ∂µ φ(x) = −ǫµ ∂µ L = ∂µ K µ .10). ∂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. = − φ′ x′ − φ′ (x) = −ǫµ ∂µ 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. We see that since the change in the Lagrangian density is a total divergence. we note that under the transformation (6. Next. We can now identify from (6.19) and in this case we obtain the change in the ﬁeld to correspond to (see (6. (6.8)) L φ′ (x).11) and (6.21) that (ǫµ is a constant parameter) K µ = −ǫµ L.216 6 Self-interacting scalar field theory φ′ x′ = φ (x) . we can now calculate explicitly the change in the Lagrangian density (see (6. (6. With this.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).20). where we have used (5. the action is invariant under inﬁnitesimal translations which deﬁne a symmetry of the system.20) where in the last step.

to be (see (6. (6.¨ 6. Therefore. even when the naive N¨ther o procedure does not lead to a symmetric stress tensor. that the conserved current independent of the parameter can be identiﬁed with ∂L − η µν L. (6.23) so that we obtain the current associated with the symmetry transformation.25) which can be easily checked to satisfy (using the equations of motion) ∂µ T µν = 0. namely.26) T µν is known as the stress tensor of the theory and can always be deﬁned to be symmetric (It is the source for the gravitational ﬁeld gµν just as the electromagnetic current j µ is the source for the electromagnetic potential Aµ .). ∂∂µ φ(x) ∂∂µ φ(x) (6. T µν = T νµ . 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. therefore. (6.1 Nother’s theorem 217 δφ(x) ∂L ∂L = −ǫν (∂ν φ(x)) .13)) ∂L − Kµ ∂∂µ φ(x) ∂L + ǫµ L ∂∂µ φ(x) ∂L − η µν L ∂∂µ φ(x) = −ǫν T µν . ∂∂µ φ(x) T µν = (∂ ν φ(x)) (6.27) . depending on the parameter of transformation.24) µ Jǫ (x) = δφ = −ǫν (∂ν φ(x)) = −ǫν (∂ ν φ(x)) It is clear.

which. (In general. is manifestly symmetric. let us recall that the Lagrangian density has the form L= m2 2 1 ∂µ φ∂ µ φ − φ .15)).) As a result. the conserved charges associated with the symmetry transformation (see (6.218 6 Self-interacting scalar field theory The conserved current. (6. P µ . in the present case. we obtain ∂L = ∂ µ φ(x).29) Therefore. We note that. for the free theory. is a second rank tensor simply because the parameter of transformation is a four vector.25) (6. the current is a tensor one rank higher than the parameter of transformation.30) T µν = ∂ ν φ(x) ∂L − η µν L ∂∂µ φ(x) (6. should generate inﬁnitesimal space-time translations. . would correspond to the components of a four vector which we identify with the energy-momentum operator as Pµ = d3 x T 0µ .31) = ∂ ν φ(x)∂ µ φ(x) − η µν L = T νµ . in this case. 2 2 (6. ∂∂µ φ(x) and this leads to the explicit form for the stress tensor (6. in this case. For the free Klein-Gordon theory.28) This is consistent with our physical intuition that the energy-momentum.

32) We see that P 0 = H coincides with the form of the Hamiltonian we had derived earlier in (5.2 Self-interacting φ4 theory 219 P0 = = = = H.34) . we note that the Lagrangian density for the interaction must be invariant under Lorentz transformations as well as translations (Poincar´ invariant).32) corresponds exactly to the former expression when normal ordered and expressed in terms of creation and annihilation operators.33) To include interactions.2 Self-interacting φ4 theory Let us denote the free part of the Klein-Gordon Lagrangian density as m2 2 1 ∂µ φ∂ µ φ − φ . We had also given an expression for the momentum operator in terms of creation and annihilation operators in (5. Pi = = P = − d3 x T 00 d3 x d3 x ˙ φ(x) 2 − 2 1 ˙ φ(x) 2 2 m2 2 1 φ + ∇φ · ∇φ + 2 2 1 ˙ φ(x) 2 1 m2 2 + ∇φ · ∇φ + φ 2 2 d3 x T 0i ˙ d3 x ∂ i φ(x)φ(x).97) and it can be checked that P in (6. ˙ d3 x ∇φ(x)φ(x). (6. (6. since the free Lae grangian density is invariant under the discrete transformation φ(x) ↔ −φ(x). Furthermore.34).6. 2 2 L0 = (6. 6.

the same canonical dimension as . in units of = c = 1 (which we have been using). a Hamiltonian) which is unbounded from below for any value of the coupling constant g. in fact. Consequently. (6.220 6 Self-interacting scalar field theory we would like to preserve this symmetry in the interactions as well. λ.34). With these conditions. would lead to a potential (and. let us introduce the concept of canonical dimensions. therefore. 4! LI = − λ > 0.35) The restriction on the coupling constant. is there so that the Hamiltonian will be positive deﬁnite. (6. . Let us also g note here that an interaction Lagrangian density of the form − 3! φ3 . the fully self-interacting theory of a real Klein-Gordon ﬁeld is described by a Lagrangian density L = L0 + LI . (6. then. To explain this brieﬂy. Thus.36) While there are other interaction Lagrangian densities that we can construct consistent with our symmetry requirements. In these units. we can show that the canonical dimension of any variable can be expressed in powers of an arbitrary mass dimension [M ]. the action is dimensionless. (We will see this shortly.35) represents the only meaningful interaction term for such a quantum ﬁeld theory. which does not respect the discrete symmetry (6. which in turn allows us to deﬁne a vacuum state of the theory. for various other reasons we can show that (6. Let us recall that the action for a quantum mechanical particle has the generic form S = = dt (p q + · · · ) ˙ (p dq + · · · ). the simplest interaction Lagrangian density involving only the scalar ﬁelds which we can think of has the form λ 4 φ (x).) As a result.37) and has.

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

(6. the coupling constant of the interaction Lagrangian density will have inverse dimensions of mass.42) so that from (6. ˙ ∂ φ(x) Π(x) = (6. consistent with our symmetry requirements (see. for n > 4. the φ4 interaction is the only physically allowed interaction in this case.32)) ∂L ˙ = φ(x). the canonical momentum conjugate to the ﬁeld variable of the theory continues to be (see (5. or. for example. restricts n ≤ 4 and. [g] = [M ]4−n . consequently. In such a case. (We will see this in a later chapter when we discuss renormalization of quantum ﬁeld theories. If we want to restrict to renormalizable theories which can give rise to meaningful physical predictions.34)). the coupling constants in the theory cannot have dimensions of inverse mass.32) we obtain the Hamiltonian density for the interacting theory to be ˙ H = Πφ − L = Π2 − = Π2 − 1 ˙2 1 m2 2 λ 4 φ − φ φ − ∇φ · ∇φ − 2 2 2 4! m2 2 λ 4 1 2 1 Π + ∇φ · ∇φ + φ + φ 2 2 2 4! .41) Therefore. [g] [M ]−4 [M ]n = [M ]0 .222 6 Self-interacting scalar field theory g ˜ SI = − n! d4 x φn . Such theories are known as non-renormalizable theories. 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. or.) This. ˜ SI = [g] d4 x [φn ] = [M ]0 . therefore. Since the interaction Lagrangian density does not involve derivatives of ﬁelds. (6.

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

(6.224 6 Self-interacting scalar field theory Schr¨dinger picture where the operators are time independent but o the state vectors carry all the time dependence and the time evolution of states is given by the Schr¨dinger equation governed by the o total Hamiltonian of the system. (6.46) In writing (6. (6. In the Schr¨dinger picture.45) O(S) = e−iHt O(H) (t)eiHt . for example. we have assumed that at t = 0 both the pictures coincide. for example.45)) and is. of course. It follows from (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. time independent. (6. t0 ) = e−iH(t−t0 ) .45). (6.49) . Thus. 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.48) The total Hamiltonian operator in the Schr¨dinger picture coino cides with that in the Heisenberg picture (see. The two pictures are related by the unitary transformation = e−iHt |ψ |ψ(t) (S) (H) . the Klein-Gordon ﬁeld o operator will be given by φ(S) (x) = e−iHt φ(H) (x. in the Schr¨dinger picture. |ψ(0) (S) = |ψ (H) . namely. 0). (6.47) O(S) = O(H) (0). t) eiHt = φ(H) (x.45) that (H) ψ|O(H) (t)|ψ (H) = (S) ψ(t)|O(S) |ψ(t) (S) .

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

49) and (6.53) Similarly. t0 ) |ψ (t0 ) . the time evolution of the state vectors. 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). i (6. In this picture. while the dynamical evolution of the operators is governed by the free Hamiltonian through the Heisenberg equations of motion. (6. it is easy to show that the time evolution operator satisﬁes the properties.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 . if we deﬁne the time evolution operator through the relation |ψ(t) = U (t.54) In other words.226 6 Self-interacting scalar field theory (IP HI ) (t) = eiH0 (S) t (S) HI e−iH0 (S) t .55) where the time evolution operator in the interaction picture in (6. From now on. This shows that once we deﬁne commutation relations for the operators in the free theory.56) then. (6. therefore. We note from (6.55) can be seen from (6. the ﬁeld operators can have a plane wave expansion. in the interaction picture they continue to hold even in the presence of interactions. is governed by the Schr¨dinger equation with the interaction o Hamiltonian playing the role of the Hamiltonian. (6. H0 ) . in this picture. . let us drop the superscript (IP ) with the understanding that we are nonetheless working in the interaction picture. H0 = 1 (IP ) (IP O (t).50) to have the explicit form U (t.

57). t0 ) t0 t t1 dt1 HI (t1 ) + (−i)2 dt1 t0 t0 dt2 HI (t1 ) HI (t2 ) +··· + ··· +(−i)n +··· . t0 ) . t0 ) = ½ − i = = dt1 HI (t1 ) U (t1 . ∂t ∂U (t.59) To bring this to a more convenient form.52) we see that the time evolution operator in (6. t0 ) = U (t2 . t t1 t0 tn−1 t0 dt1 t0 dt2 · · · dtn HI (t1 ) HI (t2 ) · · · HI (tn ) (6.) We can now solve (6. (6.3 Interaction picture and time evolution operator 227 U (t. it follows that the total Hamiltonian is time independent in the interaction picture. t0 ) t0 ½−i ½−i t dt1 HI (t1 ) t0 t t0 ½−i t1 dt2 HI (t2 ) U (t2 . t0 ) = U (t0 . to obtain t U (t. t) .51)).6. t0 ) = HI (t)U (t. (Since H0 is time independent in the interaction picture (see (6.58) We note from (6. from (6.56) with the use of (6. t) = ½. t0 ) = U −1 (t.50)) ∂|ψ(t) = HI (t)|ψ(t) .58) iteratively subject to the initial condition in (6. U † (t. t1 ) U (t1 .59) and note that . t0 ) . (6.55) satisﬁes the equation (this also follows from (6. i ∂t i or.53) that the interaction Hamiltonian in the interaction picture is time dependent.57) Furthermore. U (t2 . let us look at the second order term in (6.

t0 (6.59) can be written as .228 6 Self-interacting scalar field theory (−i)2 t t1 dt1 t0 t0 t dt2 HI (t1 ) HI (t2 ) t2 = (−i)2 = (−i)2 2 t dt2 t0 t t0 dt1 HI (t2 ) HI (t1 ) t1 dt1 t0 t2 t0 dt2 HI (t1 ) HI (t2 ) dt1 HI (t2 ) HI (t1 ) + t0 dt2 t0 t = (−i)2 2 t t1 dt1 t0 t2 t0 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + t0 dt2 t0 t dt1 θ (t2 − t1 ) HI (t2 ) HI (t1 ) t = (−i)2 2 t dt1 t0 t t0 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + t0 dt2 t0 t dt1 θ (t2 − t1 ) HI (t2 ) HI (t1 ) = (−i)2 2 t0 dt1 dt2 θ (t1 − t2 ) HI (t1 ) HI (t2 ) + θ (t2 − t1 ) HI (t2 ) HI (t1 ) = (−i)2 2! t t0 t dt1 dt2 T (HI (t1 ) HI (t2 )) . we can show that the n-th term in the series in (6. (6.60) where T denotes the time ordering operator and is conventionally deﬁned for two bosonic operators as T (A(t1 )B(t2 )) = θ(t1 − t2 )A(t1 )B(t2 ) + θ(t2 − t1 )B(t2 )A(t1 ). In a similar manner.61) with the operator at later time standing to the left of the operator at earlier time.

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

(6. which is the deﬁnition of the scattering amplitude.66) where we have identiﬁed S = U (∞. the S-matrix (or the scattering matrix) of the theory can be identiﬁed with the time evolution operator (6.67) which has . is obtained from (6. Let us denote the initial state at inﬁnite past as the free particle state |ψi (−∞) = |i . the interaction vanishes. −∞)|i = f |S|i . the state into which this will evolve at t = ∞ is deﬁned from (6. (6.64)) leads to the same result for the rate of transition in the physical scattering of particles.65) Then. (6. the probability amplitude for an initial state |i to be in the ﬁnal free particle state |f at t = ∞.68) Consequently. 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. 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. −∞)|i = S|i .67) Therefore. −∞).55) to be |ψi (∞) = U (∞. (6.64) can be thought of as the relevant boundary condition for the scattering problem under study. The condition (6. −∞)|ψi (−∞) = U (∞.66) to be Sf i = f |ψi (∞) = f |U (∞. As a result.230 6 Self-interacting scalar field theory so that in the inﬁnite past as well as in the inﬁnite future.

Schr¨dinger and interaction) coincide at t = 0 (see (6. These states can be constructed by noting that the three pictures (Heisenberg. not well deﬁned if we do not use an adiabatic interaction of the form. the linear contribution to the right-hand side of (6. say.51)) and o since the Heisenberg states are time independent. (6. Let us see this explicitly in the case of the linear term in the expansion of the time evolution operator in (6. in general.4 S-matrix 231 the explicit perturbative expansion of the form (see (6.57)). they are uniquely given by the states (in any picture) at t = 0.63). the S-matrix (the scattering matrix) is unitary since the time evolution operator is (see (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+ +··· . −∞)|Ψ(−∞) (in) . These are asymptotic free states as t → −∞ and t → ∞ respectively. In this case. Let us suppose that at t → −∞ we have a free incoming state denoted as |Ψ(−∞) (in) = |i .64) provides a regularized meaning to the oscillatory terms in (6.70) where the time evolution operator U (0.6. This also leads to the notion of “in” and “out” states which are quite important in a formal description of scattering theory.69) through the appropriate boundary conditions. It is clear from the expansion above that the adiabatic switching of the interaction in (6. in (6. −∞) is. this state would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (in) = U (0. Then.64).70) will have the form .69) Furthermore.63)) −i ∞ R S = U (∞. (6.

if we assume that as t → ∞. (6. ∞)|Ψ(∞) (out) . 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 ) . −∞). we have a free outgoing state |Ψ(∞) (out) . We also note that the integral in (6. (6. the “in” and the “out” states deﬁne a complete space of states at t → −∞ and t → ∞ respectively.71) would show that in this case. Rather.71) is not deﬁned at the lower limit in the absence of the adiabatic factor eηt .71) (IP where in the intermediate steps.64) naturally provides a regularization for the formal deﬁnition of the time evolution operator U (0. from their deﬁnitions in (6. From (6. ∞) would not be well deﬁned at the upper limit. the regularizing factor would have to be e−ηt without which U (0.70) and (6. Furthermore. This shows that the asymptotically free “in” and “out” states can be deﬁned in a unique manner by relating them to the Heisenberg state.72) we see that even though both the “in” and the “out” states are related to the same Heisenberg state.72) we see that they are related to each other as .72) and an analysis as in (6. then this would be related to the Heisenberg state as |Ψ (H) = |Ψ(0) (out) = U (0. This clariﬁes how the boundary condition (6. they are not identical because their deﬁnitions use diﬀerent regularizing factors.70) and (6.232 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) . In a parallel manner.

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

by deﬁnition. Thus. has the creation operator so that this prescription is equivalent to the deﬁnition of normal ordering in (5.77) It is clear. For example. 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). therefore. annihilated by the annihilation operator.75). N φ(+) (x)φ(−) (y) ≡ : φ(+) (x)φ(−) (y) : = φ(−) (y)φ(+) (x). Furthermore. that the vacuum expectation value of any normal ordered product vanishes.76) Since the vacuum state is.76) that . as far as bosonic operators are concerned. we note from (6. (6. 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). using the relations in (6. for the bosonic scalar ﬁeld.76)). N φ(−) (x)φ(+) (y) ≡ : φ(−) (x)φ(+) (y) : = φ(−) (x)φ(+) (y). (6. the order of the factors inside a normal ordered product is not important. φ(−) . we have N (φ(x)) ≡ : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x). (6.75) We note that. N φ(+) (x)φ(+) (y) ≡ : φ(+) (x)φ(+) (y) : = φ(+) (x)φ(+) (y).234 6 Self-interacting scalar field theory the negative energy part.

this also shows that if we can rewrite any product of operators (including normal ordered factors) in terms of normal ordered terms.81) . φ(−) (y) +φ(−) (x)φ(+) (y) + φ(−) (x)φ(−) (y) = : φ(x)φ(y) : −iG(+) (x − y). let us note that. (6.151) as φ(x)φ(y) = −iG(+) (x − y).79) as φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y). the calculation of S-matrix elements will simplify enormously.80) with the pairing of the two ﬁeld operators deﬁned to be related to the invariant positive energy Green’s function in (5.6.79) where we have used (5. (6. (6. does not carry any nontrivial quantum number.158) and (6. This gives another reason to choose the prescription of normal ordering for physical observables.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. To see how we can write a product of operators in normal ordered form. in the simple example of the product of two ﬁeld operators. 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). 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. Furthermore.78) Therefore. (6. Let us denote (6.76). therefore.

the second term in (6. This brings out yet another connection between Green’s functions and the quantum ﬁeld theory. (6.236 6 Self-interacting scalar field theory We note that G(+) (x − y) is not a symmetric function (so that the pairing should be read carefully as from left to right).86) we obtain . this will be necessary in developing a perturbation expansion for the quantum ﬁeld theory where the interaction Hamiltonian is normal ordered.85) and (6. Therefore.85) Thus adding the two terms in (6.86) (6. let us next see how we can combine a product of normal ordered factors into normal ordered terms. Since we choose the Hamiltonian to be normal ordered.82) because the vacuum expectation value of the normal ordered product in (6.80) that we can also identify φ(x)φ(y) = 0|φ(x)φ(y)|0 = −iG(+) (x − y). : φ(x) : : φ(y) : = : φ(x) : φ(+) (y) + : φ(x) : φ(−) (y).80) is zero. φ(−) (y) = = : φ(x)φ(−) (y) : + φ(+) (x).83) We recognize that the ﬁrst term in (6. On the other hand. but since it is a c-number function. (6. We know that by deﬁnition : φ(x) : = φ(x) = φ(+) (x) + φ(−) (x).84) gives : φ(x) : φ(−) (y) = φ(x)φ(−) (y) = φ(−) (y)φ(x) + φ(x).84) simply corresponds to a normal ordered product : φ(x) : φ(+) (y) = φ(x)φ(+) (y) = : φ(x)φ(+) (y) : . we note from (6.84) (6. (6. Clearly. φ(−) (y) : φ(x)φ(−) (y) : − iG(+) (x − y).

what we had seen earlier in (6. of course. namely.87) This is. let us note that .6.79) and (6.88) φ(x)φ(y) = = but (6. (6. φ(−) (z) = : φ(x)φ(y)φ(z) : + φ(+) (x).89) Finally. Next. φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + : φ(x)φ(y)φ(−) (z) : + φ(x)φ(y) + iG(+) (x − y) . (6.5 Normal ordered product and Wick’s theorem 237 : φ(x) : : φ(y) : = : φ(x) : φ(+) (y)+ : φ(x) : φ(−) (y) = : φ(x)φ(+) (y) : + : φ(x)φ(−) (y) : − iG(+) (x − y) = : φ(x)φ(y) : + φ(x)φ(y).80). (6. : φ(x)φ(y) : −iG(+) (x − y) : φ(x)φ(y) : + φ(x)φ(y). let us look at a product of the form : φ(x)φ(y) : : φ(z) : = : φ(x)φ(y) : φ(+) (z) + : φ(x)φ(y) : φ(−) (z) = : φ(x)φ(y)φ(+) (z) : + φ(−) (z) : φ(x)φ(y) : + : φ(x)φ(y) : . φ(−) (z) = : φ(x)φ(y)φ(z) : − iG(+) (x − z)φ(y) − iG(+) (y − z)φ(x) = : φ(x)φ(y)φ(z) : + φ(x)φ(z) : φ(y) : + φ(y)φ(z) : φ(x) : .87) shows how a product of two simple normal ordered factors can be expressed as a normal ordered term and a pairing. φ(−) (z) φ(y) +φ(x) φ(+) (y).

φ(−) (w)]φ(+) (z) + φ(−) (z)φ(−) (w) : φ(x)φ(y) : +[: φ(x)φ(y) :. φ(−) (w)]φ(+) (z) +φ(−) (z)[: φ(x)φ(y) :. (6.238 6 Self-interacting scalar field theory : φ(x)φ(y) : : φ(z)φ(w) : = : φ(x)φ(y) : φ(+) (z)φ(+) (w)+ : φ(x)φ(y) : φ(−) (z)φ(+) (w) + : φ(x)φ(y) : φ(−) (w)φ(+) (z)+ : φ(x)φ(y) : φ(−) (z)φ(−) (w) = : φ(x)φ(y)φ(+) (z)φ(+) (w) : +φ(−) (z) : φ(x)φ(y) : φ(+) (w) +[: φ(x)φ(y) :. φ(−) (z)φ(−) (w)] = : φ(x)φ(y)φ(+) (z)φ(+) (w) : + : φ(x)φ(y)φ(−) (z)φ(+) (w) : + : φ(x)φ(y)φ(+) (z)φ(−) (w) : + : φ(x)φ(y)φ(−) (z)φ(−) (w) : +[: φ(x)φ(y) :. φ(−) (z)]φ(+) (w) + φ(−) (w) : φ(x)φ(y) : φ(+) (z) +[: φ(x)φ(y) :. φ(−) (z)]φ(w) + [: φ(x)φ(y) :. 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).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 .

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.5 Normal ordered product and Wick’s theorem 239 factors can be written in terms of normal ordered products with all possible pairings between ﬁelds in diﬀerent factors (but no pairing between two ﬁelds within the same factor). (6.6.91) φ(x)φ(y)φ(z) = : φ(x)φ(y) : φ(z) + φ(x)φ(y)φ(z) = : φ(x)φ(y) : : φ(z) : + φ(x)φ(y) : φ(z) : = : φ(x)φ(y)φ(z) : + φ(x)φ(z) : φ(y) : + φ(y)φ(z) : φ(x) : + φ(x)φ(y) : φ(z) : . therefore. φ(x)φ(y)φ(z)φ(w) = = : φ(x)φ(y) : + φ(x)φ(y) : φ(z)φ(w) : + φ(z)φ(w) : φ(x)φ(y) :: φ(z)φ(w) : + φ(x)φ(y) : φ(z)φ(w) : + φ(z)φ(w) : φ(x)φ(y) : + φ(x)φ(y) φ(z)φ(w) = : φ(x)φ(y)φ(z)φ(w) : + φ(x)φ(z) : φ(y)φ(w) : + φ(x)φ(w) : φ(y)φ(z) : + φ(y)φ(z) : φ(x)φ(w) : . For example.80) that φ(x)φ(y) = : φ(x)φ(y) : + φ(x)φ(y).92) Similarly. that (6. 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. we already know from (6. It follows.

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). (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. we have already noted in (6.) An immediate consequence of this result is that the vacuum expectation value of an odd number of ﬁeld operators must vanish (since not all ﬁeld operators can be paired).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). (6. (6. For example. In general. 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.93) In other words. 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 ) : + ··· .89) that .94) This is quite signiﬁcant since it says that the vacuum expectation value of the product of any number of factors involving ﬁeld operators (whether the factors are normal ordered or not) is given by the terms where all the ﬁelds are completely paired.

6.6 Time ordered products and Wick’s theorem

241

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

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

: φ(x)φ(y) : φ(z)φ(w) =: φ(x)φ(y) :

: φ(z)φ(w) : + φ(z)φ(w)

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

T (φ(x)φ(y)) = T (φ(y)φ(x)) =

(6.97)

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

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

(6.98)

242

6 Self-interacting scalar field theory

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

θ(x) =

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

(6.99)

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

θ(x) = lim −

→0+

dk e−ikx . 2πi k + i

(6.100)

Of course, by deﬁnition

T (φ(x)) = φ(x) =: φ(x) : .

(6.101)

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

6.6 Time ordered products and Wick’s theorem

243

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

+θ x0 − z 0 θ z 0 − y 0 φ(x)φ(z)φ(y)

+θ y 0 − z 0 θ z 0 − x0 φ(y)φ(z)φ(x)

+θ y 0 − x0 θ x0 − z 0 φ(y)φ(x)φ(z)

+θ z 0 − x0 θ x0 − y 0 φ(z)φ(x)φ(y)

+θ z 0 − y 0 θ y 0 − x0 φ(z)φ(y)φ(x),

(6.102)

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

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

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

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

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

244

6 Self-interacting scalar field theory

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

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

(6.105)

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

φ(x)φ(y) = iGF (x − y).

(6.106)

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

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

(6.107)

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

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

(6.108)

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

6.6 Time ordered products and Wick’s theorem

245

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

φ(x)φ(y) = iGF (x − y) = φ(y)φ(x),

(6.110)

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

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

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

246

6 Self-interacting scalar field theory

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

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

6.7 Spectral representation and dispersion relation

247

0| [φ(x), φ(y)] |0

= −iG(x − y) =

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

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

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

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

n

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

= i

n

−eikn ·(x−y) 0|B(0)|n n|A(0)|0

,

(6.116)

248

6 Self-interacting scalar field theory

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

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

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

n

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

=

i (2π)3

n

= where

i (2π)3

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

(6.118)

ρ(q) = (2π)3

n

δ4 (q − kn )| 0|A(0)|0 |2 ,

(6.119)

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

6.7 Spectral representation and dispersion relation

249

GAA (x − y) = = =

i (2π)3

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

d4 q ρ(q 2 )

∞ 0

dσ 2 ρ(σ 2 ) i

=

0

∞

dσ 2 ρ(σ 2 ) G(x − y, σ 2 ),

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

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

∞ −∞

f (t) =

dk0 −ik0 t e f (k0 ). 2π

(6.123)

Using the integral representation of the step function

250

6 Self-interacting scalar field theory

θ(t) = lim −

ǫ→0+

1 2πi

∞ −∞

dk0

e−ik0 t , k0 + iǫ

(6.124)

**from (6.122) we obtain 1 2πi
**

∞ −∞

f (t) =

ǫ→0+

lim

−

dk0

e−ik0 t k0 + iǫ

′

∞ −∞

′ dk0 −ik0 t ′ ′ e f (k0 ) 2π

∞

=

ǫ→0+

lim −

−∞ ∞

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

=

ǫ→0+

lim − lim

−∞ ∞

=

ǫ→0+

−∞

dk0 −ik0 t e 2π

∞ −∞

′ ′ f (k0 ) dk0 ′ − k − iǫ . 2πi k0 0

(6.125)

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

∞ −∞ ′ dk0 ′ f (k0 ) , − k0 − iǫ

f (k0 ) = lim

ǫ→0+

′ k0

(6.126)

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

ǫ→0+

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

∞ −∞ ′ dk0 ′ f (k0 ) , ′ k0 − k0

f (k0 ) =

(6.128)

6.7 Spectral representation and dispersion relation

251

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

∞ −∞ ′ Im f (k0 ) , ′ k0 − k0

Re f (k0 ) =

′ dk0

Im f (k0 ) = −

∞

−∞

′ dk0

′ Re f (k0 ) . ′ k0 − k0

(6.129)

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

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

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

−∞

′ Im f (k0 ) , ′ k0 − k0 − iǫ

(6.130)

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

252

6 Self-interacting scalar field theory

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

|k0 |→∞

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

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

ǫ→0+

ǫ→0+

lim

−∞ ∞ −∞

′ dk0

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

ǫ→0+

lim

′ dk0

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

ǫ→0+

lim

1 ¯ π(k0 − k0 )

′ k0

×

1 1 . − ′ ¯ − k0 − iǫ k0 − k0 − iǫ

(6.133)

6.7 Spectral representation and dispersion relation

253

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

′ k0 ∞ −∞

¯ f (k0 ) − f (k0 ) =

ǫ→0+

lim ×

′ ′ ¯ dk0 Im (f (k0 ) − f (k0 ))

1 1 − ′ ¯ . − k0 − iǫ k0 − k0 − iǫ

(6.134)

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

Im f (k0 ) = −Im (−k0 ).

(6.135)

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

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

f (k0 ) − f (0) = lim

ǫ→0+

′ dk0

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

η→0+

(6.137)

This has the imaginary part given by

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

(6.138)

254

6 Self-interacting scalar field theory

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

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

ǫ→0+

lim = = = =

1 π

′ dk0

ǫ→0+

lim − lim −

π 2πEk

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

ǫ→0+

ǫ→0+

lim

1 2 (k0 + iǫ)2 − Ek 1

2 k0

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

η→0+

lim

−

2 Ek

+ ik0 η

,

(6.139)

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

6.8 References

255

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

This page intentionally left blank .

Chapter 7 Complex scalar ﬁeld theory 7.3) φ1 (x) = i φ2 (x) = − √ φ(x) − φ† (x) .2) can be inverted to give 1 √ φ(x) + φ† (x) . ∂µ ∂ µ + m2 φ† (x) = 0. 2 (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. the dynamical equations take the forms ∂µ ∂ µ + m2 φ(x) = 0. The relations in (7. there are now two distinct equations.1) Therefore. (7. φ† (x) = φ(x). 2 1 √ (φ1 (x) − iφ2 (x)) . spin zero (scalar) ﬁelds which are Hermitian. In this case. 2 (7.2) φ(x) = φ† (x) = where φ1 (x) and φ2 (x) are two distinct. The physical meaning of this system can be understood by noting that we can express the complex ﬁeld operator in terms of two real (Hermitian) ﬁeld operators as 1 √ (φ1 (x) + iφ2 (x)) . 2 257 .

and is manifestly Hermitian. therefore. then we can deﬁne the momenta conjugate to the two ﬁeld operators as Πi (x) = ∂L ˙ = φi (x).4) as the Euler-Lagrange equations is easily determined to be m2 2 1 m2 2 1 ∂µ φ1 ∂ µ φ1 − φ1 + ∂µ φ2 ∂ µ φ2 − φ 2 2 2 2 2 m2 1 ∂µ (φ1 − iφ2 ) ∂ µ (φ1 + iφ2 ) − (φ1 − iφ2 ) (φ1 + iφ2 ) 2 2 (7. Πj (y)]x0 =y0 . (7. (Note that we have not been careful about the order of the factors in rearranging terms in (7. two distinct particles) with degenerate mass.1) in terms of the real ﬁeld variables as ∂µ ∂ µ + m2 φ1 (x) = 0. [φi (x).) We can quantize this theory in one of the two equivalent ways.6) Consequently. (7. independent spin zero Klein-Gordon ﬁelds (and. Therefore.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.5) L = = = ∂µ φ† ∂ µ φ − m2 φ† φ. φj (y)]x0 =y0 = 0 = [Πi (x). the equal time canonical commutation relations follow to be [φi (x). ∂µ ∂ µ + m2 φ2 (x) = 0. 2. The Lagrangian density which will give the dynamical equations in (7. Πj (y)]x0 =y0 = iδij δ3 (x − y). 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. If we treat φ1 (x) and φ2 (x) as the independent variables.7) . ˙ ∂ φi (x) i = 1. such a system equivalently describes two free.1) or (7. (7.4) Namely.

if we treat φ(x) and φ† (x) as the independent dynamical variables.7. (7. 2 2 .1 Quantization 259 In this case. We see that the Hamiltonian density of the free complex scalar ﬁeld theory is simply the sum of the Hamiltonian densities for two non-interacting real Klein-Gordon ﬁeld theories with degenerate mass and H= d3 x H. then we can deﬁne the momenta conjugate to the ﬁeld operators from (7. the Hamiltonian density is obtained in a straightforward manner to be of the form 2 H = i=1 ˙ Πi φi − L 1 1 m2 2 = Π2 + Π2 − Π2 + ∇φ1 · ∇φ1 + φ 1 2 1 2 2 2 1 1 m2 2 1 φ − Π2 + ∇φ2 · ∇φ2 + 2 2 2 2 2 = 1 2 1 m2 2 Π1 + ∇φ1 · ∇φ1 + φ 2 2 2 1 1 1 m2 2 + Π2 + ∇φ2 · ∇φ2 + φ 2 2 2 2 2 2 = i=1 1 2 1 m2 2 Πi + ∇φi · ∇φi + φ . 2 2 2 i (7.8) where we have used the identiﬁcation in (7.6).5) to be Π(x) = = ∂L ˙ = φ(x) ˙ ∂ φ† (x) 1 1 ˙ ˙ √ φ1 (x) + iφ2 (x) = √ (Π1 (x) + iΠ2 (x)) .9) Alternatively.

(7.11) Furthermore. the equal time canonical commutation relations.5) and (7.12) = Π† Π + ∇φ† · ∇φ + m2 φ† φ. φ† (y) = φ† (x). φ(x). we can expand the ﬁeld φ(x) in the basis of plane wave solutions of the Klein-Gordon equation (see (5. in this case. φ† (y) = 0. (7. 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. of the form .10) 2 2 As a result.260 Π† (x) = ∂L ˙ = φ† (x) ˙ ∂ φ(x) 7 Complex scalar field theory 1 1 ˙ ˙ = √ φ1 (x) − iφ2 (x) = √ (Π1 (x) − iΠ2 (x)).8) and H= d3 x H. φ(y)] = φ(x). take the forms (we do not write the equal time condition explicitly for simplicity. in the present case.10). [Π(x). (7.56)). is manifestly Hermitian as it should be even though the dynamical variables are not. 7. Π† (y) = φ† (x). Π† (y) = 0. although it is to be understood that x0 = y 0 in all of the relations in (7.2 Field decomposition As in the case of the Hermitian (real) Klein-Gordon ﬁeld. Π(y)] = Π(x). Π(y) = iδ3 (x − y). Π† (y) = Π† (x).13) Note that the Hamiltonian. which can be compared with (7.11)) [φ(x). in this case. using (7.

a† (k′ ) = δij δ3 (k − k′ ). As before.15) Given the decomposition in (7. 2 1 √ (a1 (k) − ia2 (k)) . e−ik·x b(k) + eik·x a† (k) . The commutation relations between these creation and annihilation operators can then be obtained from the canonical quantization relations to be (see (5. can then be interpreted as the annihilation and the creation operators for the two kinds of quanta associated with the ﬁelds φi (x).14) with k0 = Ek = k2 + m2 > 0. it is clear that the ﬁeld operators φ(x) and φ† (x) can also be expanded in the plane wave basis as (see (7. 2 (7. i j ai (k). j (7.2)) φ(x) = = 1 √ (φ1 (x) + iφ2 (x)) 2 1 √ 2 d3 k (2π)3 2k0 e−ik·x (a1 (k) + ia2 (k)) +eik·x a† (k) + ia† (k) 1 2 = φ† (x) = d3 k (2π)3 2k0 d3 k (2π)3 2k0 e−ik·x a(k) + eik·x b† (k) .17) . a† (k′ ) . aj (k′ ) = 0 = a† (k). (7. 2. ai (k) and a† (k). with i i = 1. (7. i i = 1.2 Field decomposition 261 φi (x) = d3 k (2π)3 2k0 √ e−ik·x ai (k) + eik·x a† (k) .7.14).66)) ai (k).16) where we have deﬁned a(k) = b(k) = 1 √ (a1 (k) + ia2 (k)) . 2.

a† (k). we can write : H := d3 k Ek a† (k)a(k) + b† (k)b(k) . b† (k′ ) = 0.20) (We can equivalently normal order the Hamiltonian (7. in terms of the creation and the annihilation operators. a† (k′ ) = 0. b(k′ ) = 0. The commutation relations between the a(k). a† (k) and † (k) can now be obtained from (7.19) If we normal order the Hamiltonian in (7.11) and (7. a(k). (7. b(k).8) or (7. b† (k′ ) = b† (k). a(k′ ) = b(k). (7.19) with respect to the ai (k) operators and in that case the Hamiltonian would .262 7 Complex scalar field theory which correspond to the annihilation operators associated with the quanta for the two ﬁelds φ(x) and φ† (x) respectively. in two equivalent ways as H= = d3 x H d3 k Ek † a1 (k)a1 (k) + a1 (k)a† (k) 1 2 + a† (k)a2 (k) + a2 (k)a† (k) 2 2 = d3 k Ek † a (k)a(k) + a(k)a† (k) 2 + b† (k)b(k) + b(k)b† (k) .18) The Hamiltonian for the system (see (7. a† (k′ ) = b(k).16)) and take the form a(k).17) b (or directly from (7. (7. a(k).12)) can be written.15) and the deﬁnition in (7.19) with respect to the a(k) and b(k) operators. a† (k′ ) = a† (k). b(k′ ) = a(k). b† (k′ ) = b(k). b† (k′ ) = δ3 (k − k′ ).

(7. φ† (x) → φ′ † (x) = φ† (x)eiθ . there will now be two distinct one particle states degenerate in mass (and. The theory is.23) where θ is a real constant parameter of transformation (a global transformation). Namely. under the transformation φ(x) → φ′ (x) = e−iθ φ(x). of course. a(k)|0 = 0 = b(k)|0 = 0|a† (k) = 0|b† (k). as was also the case for a real Klein-Gordon theory. = b† (k)|0 .3 Charge operator To understand the physical meaning of the two distinct and degenerate one particle states. there is in addition an internal symmetry associated with the system. Such a transformation is not a space-time symmetry transformation since . therefore. manifestly Lorentz invariant and translation invariant. The degeneracy can be understood as follows. energy) given by |k = a† (k)|0 . Namely. as we know from our studies in non-relativistic quantum mechanics.) The vacuum state in such a case can be deﬁned to be the lowest energy state annihilated by both a(k) and b(k). in the present case. we can also construct higher order states which will also show a degeneracy that was not present in the case of the real KleinGordon theory.22) ˜ |k Similarly. However. After all.7. as we will see shortly. (7. the theory is invariant. 7. H|0 = 0. let us analyze the symmetry properties of this system.3 Charge operator 263 be a sum of two Klein-Gordon Hamiltonians that we have already studied in chapter 5.21) However. (7. degeneracies arise when there is a symmetry present in the system under study.

∂µ φ(x).24) where ǫ denotes the inﬁnitesimal constant parameter of transformation. the transformation in (7. φ† (x).13)) . Rather.5) is invariant under the global phase transformation in (7. Thus.23) (or the inﬁnitesimal form of it in (7.24). such a transformation is known as an internal symmetry transformation. ∂µ φ′ (x).23) takes the form (α = ǫ = inﬁnitesimal) δφ† (x) = φ′ † (x) − φ† (x) = iǫφ† (x).264 7 Complex scalar field theory the space-time coordinates are not changed by this transformation.23) (or the inﬁnitesimal form (7.8)) is given by L φ′ (x). (7. the change in the Lagrangian density (see (6. for inﬁnitesimal transformations in (7. ∂µ φ† (x) = ∂µ φ′ † ∂ µ φ′ − m2 φ′ † φ′ − ∂µ φ† ∂ µ φ + m2 φ† φ = ∂µ φ† eiθ e−iθ ∂ µ φ − m2 φ† eiθ e−iθ φ − ∂µ φ† ∂ µ φ + m2 φ† φ = 0. δφ(x) = φ′ (x) − φ(x) = −iǫφ(x).24)) showing that it is a symmetry of the system and in this case we have K µ = 0. ∂µ φ′ † (x) −L φ(x).24)). (7. the Lagrangian density (7. we have (see (6. in this case.26) On the other hand. We note that under the global phase transformation (7. (7.25) The derivatives do not act on the exponentials since θ is a constant parameter (independent of space-time coordinates). Inﬁnitesimally. φ′ † (x).

13)) ← → ∂L ∂L + δφ† − K µ = iǫφ† (x) ∂ µ φ(x). But here it is an operator. (7. Of course. under normal ordering the order of factors does not matter (for a bosonic theory). this is the probability current density (up to a normalization) which we had studied in the ﬁrst quantized theory. therefore. we normal order all observables such as currents and charges and. † ∂∂µ φ ∂∂µ φ µ Jǫ = δφ (7.) Therefore. we can show that the associated conserved charge of the theory can be written in terms of the creation and annihilation operators as .) Furthermore. Even in a quantum theory. (7.3 Charge operator 265 δφ ∂L ∂L + δφ† ∂∂µ φ ∂∂µ φ† = −iǫφ(x)∂ µ φ† (x) + iǫφ† (x)∂ µ φ(x) = iǫ φ† (x)∂ µ φ(x) − φ(x)∂ µ φ† (x) = iǫ φ† (x)∂ µ φ(x) − (∂ µ φ† (x))φ(x) ← → = iǫφ† (x) ∂ µ φ(x). we can deﬁne the N¨ther current associated with this o symmetry to be (see (6.29) This is a vector current since the parameter of transformation is a scalar.1) ∂µ J µ (x) = 0.30) Using the ﬁeld decomposition (7. as we have seen.27).28) and the parameter independent current follows to be ← → J µ = iφ† (x) ∂ µ φ(x).16). (7.7.27) (Note here that classically the order of the factors is not important. (In fact. we have freely rearranged the order of factors in (7. the current can be easily checked to be conserved using the equations of motion in (7.

Acting on the ﬁrst of the two one particle states in (7. the charge operator gives Q|k = = = = d3 k′ a† (k′ )a(k′ ) − b† (k′ )b(k′ ) a† (k)|0 d3 k′ a† (k′ )a(k′ )a† (k)|0 d3 k′ a† (k′ ) a(k′ ).32). (7.33) so that the vacuum does not carry any charge. (7. a† (k) + a† (k)a(k′ ) |0 (7. Here.32) annihilates the vacuum.266 7 Complex scalar field theory Q = = i = d3 x J 0 (x) = i ← → d3 x φ† (x) ∂ 0 φ(x) ˙ ˙ d3 x φ† (x)φ(x) − φ† (x)φ(x) d3 k a† (k)a(k) − b(k)b† (k) . (7.32) Since this charge operator is associated with a global phase transformation of the kind associated with electromagnetic interactions (namely. the charge operator takes the form : Q := d3 k a† (k)a(k) − b† (k)b(k) . . we are suppressing the normal ordering symbol for simplicity with the understanding that the charge operator has the normal ordered form in (7.34) d3 k′ a† (k′ )δ3 (k′ − k)|0 = a† (k)|0 = |k . Q|0 = d3 k a† (k)a(k) − b† (k)b(k) |0 = 0.31) If we normal order. We note that the normal ordered charge operator (7. a U (1) phase transformation) we can identify this operator as the electric charge operator.22).

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

107) that the Feynman Green’s function can be related to the vacuum expectation value of the time ordered product of two ﬁeld operators. 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. Since we have already constructed the Feynman Green’s function for the real Klein-Gordon ﬁeld theory in (5.146). to conclude this section. we note that the complex Klein-Gordon ﬁeld theory describes spin zero particles carrying electric charge while the real Klein-Gordon ﬁeld theory describes charge neutral spin zero particles. we will construct only the Feynman Green’s function for the complex KleinGordon ﬁeld theory. Therefore. let us discuss the Feynman Green’s function in the basis of the complex φ(x) and φ† (x) ﬁelds.16) as well as the commutation relations in (7. We note from (6. Thus. in this section.4 Green’s functions In chapter 6 we have already seen that the Feynman Green’s functions play the most important role in the calculations of the S-matrix in a quantum ﬁeld theory.268 7 Complex scalar field theory study in a later chapter. 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 ′ ′ ′ ′ . Using the ﬁeld decomposition in (7. 7.18) we now obtain 0|φ(x)φ(y)|0 = d3 kd3 k′ √ 0| e−ik·x a(k) + eik·x b† (k) 3 k0 k′ 0 2(2π) × e−ik ·y a(k′ ) + eik ·y b† (k′ ) |0 = 0.

7.21) as well as the deﬁnitions of the positive and negative energy Green’s functions in (5.143) and (5. Given the relations in (7.39) .4 Green’s functions 269 × e−ik ·y b(k ) + eik ·y a† (k ) |0 = = = d3 kd3 k √ e−ik·x+ik ·y 0|a(k)a† (k )|0 2(2π)3 k0 k 0 d3 kd3 k √ 2(2π)3 k0 k 0 e−ik·x+ik ·y δ3 (k − k ) d3 k e−ik·(x−y) (2π)3 2k0 = −iG(+) (x − y). 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 (7.145). Here we have used (7.38). 0|T (φ† (x)φ† (y))|0 = θ(x0 − y 0 ) 0|φ† (x)φ† (y)|0 + θ(y 0 − x0 ) 0|φ† (y)φ† (x)|0 = 0.38) = iG(−) (x − y). (7. 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.

we see that in this case.270 On the other hand. It is easy to check that the Lagrangian density with the quartic interaction λ † 2 (φ φ) .23). 7.40) Here we have used the deﬁnition of the Feynman Green’s function in (5.42) . 2 (7. namely.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. in this case we should look for interactions that would preserve the internal global symmetry in (7. deﬁning 1 φ(x) = √ (σ(x) + iχ(x)) .41) describes the most general (renormalizable) self-interacting theory for the complex Klein-Gordon ﬁeld. Thus. (7. (7. However. Denoting φ1 = σ and φ2 = χ in the deﬁnition in (7. there is only one independent Feynman Green’s function that is related to the vacuum expectation value of the time ordered product of φ(x)φ† (y). 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). 0|T (φ† (x)φ(y))|0 = 0|T (φ(y)φ† (x))|0 = iGF (y − x) = iGF (x − y).146). 4 L = ∂µ φ† ∂ µ φ − m2 φ† φ − λ > 0.

) λ † 2 (φ φ) 4 2 V = m2 φ† φ + = λ m2 2 σ + χ2 + σ 2 + χ2 2 16 .43) Namely. the minimum of the potential is obtained by requiring ∂V ∂V =0= .41) or (7.5 Spontaneous symmetry breaking and the Goldstone theorem 271 we note that the Lagrangian density (7.7. In the real basis of (7.42). 4 H= d3 x Π† Π + ∇φ† · ∇φ + m2 φ† φ + (7.44) The Hamiltonian for the interacting theory (7. this describes the theory of two real Klein-Gordon ﬁelds with quartic self-interactions and which are also interacting with each other.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 .41) has the form λ † 2 (φ φ) . (7.23) or the inﬁnitesimal form in (7. 16 L = (7.24) can be shown to have the form δσ(x) = ǫχ(x). ∂σ ∂χ (7.45) and we note that. δχ(x) = −ǫσ(x). the other (kinetic energy) terms would only add positively to the energy. the inﬁnitesimal transformations (7.46) In general. ∂V ∂V =0= . ∂φ ∂φ† or. The Lagrangian density (7. for constant ﬁeld conﬁgurations.43) is invariant under the global phase transformation (7. the minimum of energy coincides with the minimum of the potential (We note that if the ﬁeld φ is not constant. (7.24).

(7.24) and (7.44)).50) and (7.272 7 Complex scalar field theory In the present case. (7.51) Both the Lagrangian densities in (7. We note from our discussion of the real Klein-Gordon theory that the plane wave expansion for the ﬁeld operator in (5. σc = 0|σ(x)|0 = 0. or.48) As we know from studies in classical mechanics.56) has exactly this property (see (5. 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 .51) are clearly invariant under the global phase transformation in (7. (7. Changing the sign of the mass term does not change the .23) (or (7. since the potential in (7. and perturbation theory developed around such a vacuum would be stable. 16 L = (7. it is clear that the minimum occurs at (the subscript c is supposed to denote a classical ﬁeld.106)). this translates to the fact that we can deﬁne the ground state of the theory such that φ† = 0|φ† (x)|0 = 0.41) except that we change the sign of the mass term L = ∂µ φ† ∂ µ φ + m2 φ† φ − λ † 2 (φ φ) . 4 λ > 0. a concept that we do not get into here) φc = φ† = 0.46) is a sum of positive terms.49) φc = 0|φ(x)|0 = 0. c or. equivalently σc = χc = 0. perturbation around the minimum of the potential is stable.42).50) In terms of the real ﬁelds σ and χ deﬁned in (7. Let us next consider the same Lagrangian density for the complex scalar ﬁeld as in (7. In quantum theory. c χc = 0|χ(x)|0 = 0.

52) k2 = −m2 . in this case.7. the theory (7. Let us note that. (7.40) or (5. c = 1) leading to an acausal propagation (and. let us note that in the absence of interaction (λ = 0). (7.4)) − m2 φ(x) = 0. the particle will travel faster than the speed of light (recall that in our units. 2 − m2 ∂|k| k (7. has many undesirable features.51) has the form V = −m2 φ† φ + = − λ † 2 (φ φ) 4 2 m2 2 λ σ + χ2 + σ 2 + χ2 2 16 . Such particles are known as tachyons and a theory with tachyons. k0 = k2 − m2 .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. a violation of causality).55) The minimum of this potential is obtained from (7. corresponding to the Einstein relation (7.54) Namely. in general.48) . or.41) would lead to the dynamical equation (compare with (1.50) or (7. therefore. but it does change the Hilbert space structure of the theory in a very interesting way. for constant ﬁeld conﬁgurations the potential of the theory (7.5 Spontaneous symmetry breaking and the Goldstone theorem 273 symmetry behaviour. To begin with.

let us analyze the second derivatives of the potential at the two extrema in (7.χc =0 λ = σc =0=χc ∂2V ∂χ2 σc =0=χc = −m2 . namely.χc =0 λ (7. (7.274 7 Complex scalar field theory ∂V ∂σ = −m2 σ + λ σ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.χc =0 λ . σc =0=χc = −m2 + = −m2 + = 0. 3λ 4 λ 4 4m2 λ 4m2 λ = 2m2 . = σ −m2 + ∂V ∂χ = −m2 χ + λ χ σ 2 + χ2 4 λ 2 σ + χ2 4 = 0.60) 2m σc =± √ .56) = χ −m2 + There now appear to be two sets of solutions to the equations in (7. λ χc = 0. = 0. we can choose σc = 0 = χc . (7.59) = 0. (7. for example.58) To determine the true minimum.57) which lead to ∂2V ∂σ 2 ∂2V ∂σ∂χ while ∂2V ∂σ 2 ∂2V ∂χ2 ∂2V ∂σ∂χ 2m σc =± √ . (7.56). 2 σc + χ2 = c 2m σc = ± √ . 2m σc =± √ . 4m2 .57) λ For the second of these solutions.

5 Spontaneous symmetry breaking and the Goldstone theorem 275 In other words we note that out of the two solutions in (7. However. λ λ λ (7. First. the second solution in (7. χ) V =0 χ 2m −√ λ 2m √ λ σ Figure 7. On the other hand. V 2 ((σc + χ2 ) c m2 4m2 ) = − = λ 2 = − 4m2 λ λ + 16 4m2 λ 2 2m4 m4 m4 + =− .55) that V (σc = χc = 0) = 0.7. leads to a lower energy.1. Each point on this circle is related by the symmetry transformation (7. 7.57) is a true minimum and we conclude that the true minimum of the potential occurs for 2 2φ† φc = σc + χ2 = c c 4m2 . the ﬁrst corresponds to a local maximum.44). .61). There are several interesting things to note from this analysis. All of this is more easily seen from the graph of the potential in Fig.57).62) so that the second solution. V (σ. we note explicitly from the form of the potential in (7.1: Potential leading to spontaneous breaking of symmetry. λ (7.61) In fact. indeed. we note that there is an inﬁnity of minima (in the σ-χ ﬁeld space) of the potential given by the circle (7.

the action or the Hamiltonian of the theory is invariant under the symmetry transformation. This has to be contrasted with the case where an interaction term in the Hamiltonian can break the symmetry of a theory explicitly.44) can be written in terms of the commutators with the charge operator as δσ(x) = −iǫ [Q. χ(x)] = −ǫσ(x).66) which implies that . In the quantum theory.65) 0|δχ(x)|0 2m = −ǫ 0|σ(x)|0 = − √ ǫ. such a breaking manifests in the following manner. From (7. In this case. in the quantum theory this would imply (see (7. for example.63) then.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. but the ground state is not. λ 0|χ(x)|0 = 0. if we choose the minimum in (7. χ(x)] |0 (7. (7. (7. λ (7. In this case.44) spontaneously. λ χc = 0.64) We recall that conserved charges are the generators of inﬁnitesimal symmetry transformations of the theory and generate the transformations through commutators.49)) 2m 0|σ(x)|0 = √ . We recall that any choice for the minimum of the potential would translate into the properties of the vacuum state of the quantum theory. it now follows that = −iǫ 0| [Q. the inﬁnitesimal transformations in (7.64). δχ(x) = −iǫ [Q. Thus.61) to correspond to the point (in the σ-χ ﬁeld space) 2m σc = √ . σ(x)] = ǫχ(x).

H]|0 = 0.71) . or. The problem with this interpretation is that the state is not normalizable. (7.7.69) where we have denoted the energy of the vacuum |0 as E0 and this derivation shows that the state |χ deﬁned in (7.31) and (7. H|χ = E0 |χ . (7. namely. t)Q|0 d3 x 0|eiP ·x J 0 (0)e−iP ·x Q|0 . (7. HQ|0 = E0 Q|0 . H] = 0. let us denote Q|0 = |χ It follows now from (7.69) that [Q.68) Since Q does not annihilate the vacuum of the present theory.5 Spontaneous symmetry breaking and the Goldstone theorem 277 Q|0 = 0. This can be easily seen from (7.67) This explicitly shows that the vacuum state of the theory is not invariant under the symmetry transformation in this case. or. (7. or.69) as follows.69) would appear to be degenerate with the vacuum state |0 in energy. although the Hamiltonian of the theory is. Therefore.68) and (7. (note that Q is Hermitian) χ|χ = = = 0|QQ|0 0| d3 x J 0 (x. it is suggestive that we can think of this state as another vacuum state.70) (7. (QH − HQ)|0 = 0. [Q.

Q] = 0.72) χ|χ = = = d3 x 0|eiP ·x J 0 (0)Qe−iP ·x |0 d3 x 0|J 0 (0)Q|0 0|J 0 (0)Q|0 d3 x −→ ∞. therefore. it follows that the momentum operator also commutes with Q and.75) does not act unitarily on the Hilbert space.278 7 Complex scalar field theory We have already seen that the Hamiltonian commutes with Q expressing the fact that it is independent of time.74) In other words. we have [Pµ . (7. . the state |χ is not normalizeable and hence cannot be thought of as another vacuum.66) that χ(x) corresponds to the Nambu-Goldstone boson of the theory.72) leads to (7. (7.73) where we have used the property of the ground state Pµ |0 = 0. In fact. it is straightforward to show that the charge Q does not exist when there is spontaneous breakdown of the symmetry and the formally unitary operator in (7. The 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. This analysis also shows that the ﬁnite transformation operator U (θ) = e−iθQ . (7.75) takes a state out of the Hilbert space. (7. Since Q does not depend on spatial coordinates. and as a consequence.

λ χ = 0|χ(x)|0 = 0. is given by σ(x) = σ ′ (x) + σ with the later identiﬁcation σ ′ (x) = σ(x) as the dynamical variable.53) and (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. λ χ(x) → χ(x) + χ = χ(x).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) + √ .60) is as follows. the shift in (7. (7.60) shows that in this case of spontaneous breakdown of the symmetry. the Lagrangian density in (7. With the choice of the solutions for the minima as in (7. We note from Fig. 7.5 Spontaneous symmetry breaking and the Goldstone theorem 279 The second interesting observation that follows from the analysis in (7.51) becomes . We can see this quantitatively in the quantum theory as follows.7. say for the ﬁrst term.78) (Explicitly.77). (7.77). (7.77) so that the shifted ﬁelds have vanishing vacuum expectation values (vevs) 0|σ(x)|0 = 0 = 0|χ(x)|0 .1 that the massless mode would correspond to oscillations along the valley of minima while the massive mode would represent the orthogonal oscillations.) Under the shift (7.64) 2m σ = 0|σ(x)|0 = √ .54))). √ 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.

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

6 Electromagnetic coupling It is clear from our discussions so far that the complex Klein-Gordon ﬁeld theory describes spin zero charged particles whereas the real Klein-Gordon ﬁeld theory describes spin zero charge neutral particles. H = −g si · si+1 . we have eﬀectively chosen a deﬁnite orientation in the physical space. there are two such vacuum conﬁgurations allowed in one dimension. (7. we do not observe any massless spin 0 bosons in nature.7. The analogue of the massless Nambu-Goldstone bosons in this case correspond to the long range correlations associated with the spin waves.80) i where the negative sign in the Hamiltonian (g > 0) is for ferromagnets.53). it is clear that once we have chosen a particular ground state (vacuum) where all the spins are pointing in a given direction. we can ask how such a system can be coupled to a background electromagnetic ﬁeld. The prescription is again through the minimal coupling discussed in (1. when the spins are all pointing up or down.6 Electromagnetic coupling 281 magnet (for example. (The π meson which is the lightest among the spin zero bosons has a mass mπ ∼ 140 MeV. (7. namely.81) where “e” denotes the charge carried by the particle. (For example. Thus.) However. Given the complex Klein-Gordon ﬁeld theory. The Goldstone theorem is interesting in the sense that it can provide a possible reason for the existence of massless spin zero bosons if they are found in nature. the Lagrangian density describing the interaction of charged spin zero ﬁelds with a background electromagnetic ﬁeld (without the quartic self-interaction term) is given by . ∂µ → ∂µ + ieAµ . Unfortunately. in the case of the Ising model). the Hamiltonian is invariant under rotations. The ground state of the theory is clearly the state (spin conﬁguration) where all the spins are aligned.) 7. Consequently the rotational symmetry is spontaneously broken.

→ ((∂µ + ieAµ ) φ)† (∂ µ + ieAµ ) φ − m2 φ† φ = = (7.82) is invariant under a local (U (1)) phase transformation of the form φ(x) → φ′ (x) = e−iθ(x) φ(x). As we can verify readily.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. (7.84) where the parameter of transformation θ(x) is now a local function of space-time coordinates (unlike in (7. ′ Dµ φ(x) → Dµ φ′ (x) = = ∂µ + ieA′ (x) φ′ (x) µ ∂µ + ie Aµ (x) + 1 (∂µ θ(x)) e e−iθ(x) φ(x) = e−iθ(x) (−i(∂µ θ(x)) + ∂µ + ieAµ (x) + i(∂µ θ(x))) φ(x) = e−iθ(x) (∂µ + ieAµ (x)) φ(x) = e−iθ(x) Dµ φ(x). φ† (x) → φ′ † (x) = φ† (x)eiθ(x) . the Lagrangian density (7. e (7.82) The modiﬁed derivative Dµ introduced in (7. note that under this transformation.82) Dµ φ = ∂µ φ + ieAµ φ.23)). In fact.85) . (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. Aµ (x) → A′ (x) = Aµ (x) + µ 1 ∂µ θ(x).

the theory described by (7. is invariant under the local phase transformation (7.7 References 1. 154 (1961). Evanston (1961). 7. Of course.82). J. 2. φ† (x). Local phase transformations are commonly known as gauge transformations and we note that.84) L(φ(x). invariance under a local transformation necessarily requires additional ﬁelds known as gauge ﬁelds. 965 (1962). Jona-Lasinio. Aµ (x)) → L(φ′ (x). namely. Row. Y. Physical Review 122. φ† .) It follows from this observation that under the local transformation (7. This is a very general feature in a quantum ﬁeld theory. J. the covariant derivative acting on a ﬁeld transforms exactly like the ﬁeld itself (covariantly) under such a transformation (see (7.84)). 4. .7. Goldstone. φ′ † (x). 3. the local phase invariance (7. Aµ .7 References 283 In other words. note that ∂µ φ(x) does not transform covariantly. 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. Peterson. Weinberg. (In contrast. unlike the global phase invariance in (7. S. Introduction to Relativistic Quantum Field Theory. which is obtained by minimally coupling the complex scalar ﬁeld to the electromagnetic ﬁeld. Goldstone. Salam and S.84). (7.23).84) requires an additional ﬁeld (Aµ ). A. Nuovo Cimento 19. Schweber. 345 (1961). Physical Review 127. Nambu and G.86) Consequently. 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.

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

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

In fact. the anti-commutation relations in (8.3) = a† 1 − a† aF aF = a† aF = NF . the wave function of the system will have the necessary antisymmetry property characteristic of fermions. 2.6) where ψα (x). 1. This equation is manifestly Lorentz covariant and the adjoint equation has the form . (8. is a four component complex spinor function. 1 2 (8. 3.2 Quantization of the Dirac ﬁeld As we have seen in (1. It is clear. it leads to the fact that with the anti-commutation relations. can only be nF = 0. therefore.286 then. we obtain 2 NF = a† aF a† aF F F 8 Dirac field theory = a† F aF . 8. there can be at the most one quantum in a given state. in such a theory. Namely. α = 1. what the Pauli exclusion principle would say.96). we will be naturally dealing with anti-commuting variables (these variables are commonly known as Grassmann variables) and anti-commutation relations. a† F + − a† aF aF F (8. has the form (iγ µ ∂µ − m) ψ = 0. (8. F F F In other words. if this system is analyzed in more detail. that in dealing with fermionic ﬁeld operators. of course.4) which shows that the eigenvalues of the number operator.2) automatically lead to NF (NF − 1) = 0.5) This is. the Dirac equation for a massive spin particle. 4.

7) (8. consequently.6) and (8. ∂ψ ← − ∂L ∂L − ∂µ = mψ − ∂µ (−iψγ µ ) = ψ iγ µ ∂µ + m = 0.61)) describing fermionic particles and.8) In trying to second quantize the Dirac equation (theory). ∂ψ ∂∂µ ψ (8. we note that we can treat ψ(x) and ψ(x) as independent ﬁeld operators as in the case of the complex Klein-Gordon ﬁeld theory. However.10) . let us note that since ψ and ψ are anti-commuting (Grassmann) variables. (8. need to be treated as anti-commuting operators. In deriving the Euler-Lagrange equations. however. where the adjoint spinor is deﬁned to be (see also (2.8.41)) ψ(x) = ψ † (x)γ 0 .7) as the Euler-Lagrange equations. In general. we have to be careful about signs that may arise in taking derivative operators (with respect to these variables) past other Grassmann variables. the Euler-Lagrange equations following from the Lagrangian density (8.9) give ∂L = (iγ µ ∂µ − m) ψ = 0. as we have seen in chapter 3 as well as in chapter 4. the 1 Dirac ﬁeld operators are expected to belong to the (reducible) spin 2 representation of the Lorentz group (see (4. With this. the Lagrangian density of the form / L = ψ (iγ µ ∂µ − m) ψ = iψ∂ ψ − mψψ. we would like to decide on a Lorentz invariant Lagrangian density which would give rise to the two Dirac equations (8.2 Quantization of the Dirac field 287 ← − ψ iγ µ ∂µ + m = 0.3). we will always use the convention of taking fermionic (Grassmann) derivatives from the left.9) is manifestly Lorentz invariant (recall the transformation of the Dirac bilinears under a Lorentz transformation discussed in section 3. in this case. (8. In dealing with the Dirac ﬁeld theory. Clearly.

288 8 Dirac field theory which are exactly the two Dirac equations (8. Namely. 2 2 L′ = (8. (8. it is not Hermitian. the action associated with it is since a total divergence does not contribute to the action.8) as well as the fact that γ 0 is Hermitian. Hermitian and which can also be checked to give the two Dirac equations as the Euler-Lagrange equations has the form ← − 1 1 ψ (iγ µ ∂µ − m) ψ − ψ iγ µ ∂µ + m ψ.9) is not Hermitian. 2 .12) However.7) that we are interested in.83).9) is Lorentz invariant and gives the two Dirac equations as the Euler-Lagrange equations.6) and (8. In fact.12) diﬀers from L in (8.13) = L − ∂µ ψγ µ ψ . (8. It is worth noting here that even though the Lagragian density (8. we note that the Lagrangian density in (8.11) † where we have used (2. An alternate Lagrangian density which is Lorentz invariant.9) only by a total divergence. 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. Even though the Lagrangian density (8. note that L† = ψ (iγ µ ∂µ − m) ψ ← − † = ψ † − iγ µ † ∂µ − m ψ ← − = ψ † − iγ µ † ∂µ − m γ 0 ψ ← − = ψ − iγ µ ∂µ − m ψ = ψ (iγ µ ∂µ − m) ψ − i∂µ ψγ µ ψ = L.

˙ ∂ψ (8. −iψβ (y) + ψα (x). (Πψ )β (y) or. the second and the third relations in (8. it is more appropriate to use the Lagrangian density in (8. the equal time canonical anti-commutation relations for the ﬁeld variables and their conjugate momenta take the forms (we do not write the equal time condition explicitly for simplicity. ˙ ∂ψ ∂L = 0. † † ψα (x). [ψα (x). but we should understand that x0 = y 0 in all the relations in (8.9) to describe the Dirac theory (equations).8. ψβ (y) + + = 0. (Πψ )β (y)]+ = −iδαβ δ3 (x − y). (8.14). .14) Πψ = Πψ = Consequently.15) and (8.15) The negative sign in the last relation in (8.15) can also be written as (Πψ )α (x). we use L in (8.9) to describe the dynamics of the system. † ψα (x). (8. ψβ (y) + = −iδαβ δ3 (x − y). we can deﬁne momenta conjugate to the ﬁeld variables ψ and ψ as (we would be careful with taking derivatives from the left which is our convention for fermionic derivatives) ∂L = −iψγ 0 = −iψ † . ψβ (y)]+ = 0 = [(Πψ )α (x). If we work with the Lagrangian density (8.12) which naturally leads to a symmetric stress tensor (see (6.27)) for the theory.2 Quantization of the Dirac field 289 For this reason as well as for simplicity. in the presence of gravitation. (Πψ )β (y)]+ . = 0. However. Using the ﬁrst relation in (8.15) is a reﬂection of our choice of left derivatives. + † = ψα (x). (Πψ )β (y) or.16)) [ψα (x).16) = δαβ δ3 (x − y).

16) are all that we need for quantizing the Dirac theory. H = d3 y † ψα (x). (8. ψβ (y) γ 0 ψ(y) β x0 =y 0 = d3 y δαβ δ3 (x − y) −iγ 0 γ · ∇y ψ(y) + mδαβ δ3 (x − y) γ 0 ψ(y) β α β x0 =y 0 = −i γ 0 γ · ∇ψ(x) α + m γ 0 ψ(x) .17) Using the ﬁeld anti-commutation relations (8.) ˙ H = −Πψ ψ − L ˙ ˙ = − −iψ † ψ − iψ † ψ − iψγ · ∇ψ + mψψ = −iψγ · ∇ψ + mψψ.15) and (8. (8.290 8 Dirac field theory (We note that we have not written any anti-commutation relation for ψ α (x) and (Πψ )α (x) simply because the relations in (8.14) can be thought of as constraints on the dynamics of the system.9) has the form (The negative sign in the ﬁrst term arises from our convention of choosing left derivatives for fermions.) The Hamiltonian density for the Dirac ﬁeld theory (8.16).15) and (8. ˙ iψα (x) = ψα (x). We will discuss the method of Dirac in more detail in chapter 10.19) .18) (8. ψβ (y) + −iγ 0 γ · ∇y ψ(y) + β † + m ψα (x). Such systems can be systematically studied through the method of Dirac brackets and such an analysis leads to the conclusion that the relations (8. For example. leading to the Hamiltonian H= d3 x H = d3 x −iψγ · ∇ψ + mψψ . it is now easy to check that this Hamiltonian leads to the two Dirac equations as the Hamiltonian equations.

6) and (8. or. ← − ˙ iψγ 0 = −iψγ · ∇ − mψ.7) as the Hamiltonian (Heisenberg) equations. (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).20) Similarly. or. we obtain ˙† iψα (x) = = † ψα (x). (iγ µ ∂µ − m) ψ(x) = 0. (8.18) indeed reproduce the two Dirac equations (8. ˙ iγ 0 ψ(x) = −iγ · ∇ψ(x) + mψ(x). or.21) and this can be written in the matrix form as ← − ˙ iψ † = −iψγ · ∇ − mψ.16) as well as the Hamiltonian (8.8. ψβ (y) + † −mψ β (y) ψα (x). the anti-commutation relations (8. (8. or.15) and (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). ← − ψ iγ µ ∂µ + m = 0.22) Namely. . H d3 y i ψ(y)γ · ∇y β † ψα (x).

uα (k.23) has been chosen keeping in mind the massive normalization of the plane wave solutions in (2. s)uα (k. let us note that the ﬁeld decomposition in (8.46). s′ ) d† (k′ . s) and d† (k. s) are anti-commuting operators. s′ ) α ′ .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 3.5).23) automatically satisﬁes the Dirac equation in this basis (i∂ − m) ψ(x) = 0. s)ψα (x) α (2π)3 k0 d3 x d3 k ′ m ′ √ ei(k−k )·x u† (k. s) .292 8. Therefore. Taking advantage of our experience with the complex Klein-Gordon ﬁeld theory.93) and (3. s) + eik·x d† (k.23) × e−ik·x c(k. s) denote the positive and the negative energy spinors deﬁned in (3. s)vα (k.24) The speciﬁc form of the prefactor in (8. s) and vα (k. we write √ (k0 = Ek = k2 + m2 > 0) ψα (x) = 1 s=± 2 d3 k m (2π)3 k0 (8. s′ ) c(k′ .94) and the ﬁeld (8. / (8. we can expand the Dirac ﬁeld operator in the basis of these plane wave solutions.4). where c(k. s′ ) α 3 (2π) k0 k′ 0 +ei(k+k )·x u† (k. As in the case of the Klein-Gordon ﬁeld (see section 5.23) can be inverted to give √ (k0 = Ek = k2 + m2 > 0) d3 x = s′ m eik·x u† (k. s)uα (k′ . s)vα (k′ .

8. (2π)3 k0 m † eik·x ψα (x)vα (k.26) that m † e−ik·x ψα (x)uα (k. ψβ (y) † † ψα (x). s ) α (8. s)uα (k . (2π)3 k0 (8. s) in a straightforward manner from the quantization conditions (8. s)ψα (x) = d† (k. s) and d(k. s) = d3 x d3 x Furthermore. s). (8. s). c† (k. s).x0 =y 0 +. s)vα (k .28) Therefore. s). s ) = δss δ3 (k − k ) = d(k.27). (2π)3 k0 d3 x (8.25) and (8. k0 m where we have used the massive normalization for the spinors discussed in (2. s ) c(k . s).27) c† (k.28) and they correspond to c(k. d† (k.x0 =y 0 +. c† (k . = δαβ δ3 (x − y).16) to be ψα (x). s).25) m k0 δss c(k.46) (see also (3.3 Field decomposition 293 δ3 (k − k ) u† (k.25)-(8. s) = d(k. ψβ (y) † ψα (x).92)). = 0. ψβ (y) +. s ) α 0 x0 = s d3 k √ m k0 k 0 + δ3 (k + k ) e2ik = s u† (k.15) and (8.29) + . s). we can now obtain the anti-commutation relations satisﬁed by the operators c(k. s). s ) = c(k.x0 =y 0 = 0. s ) + . using the inversion relations (8. (8. d† (k . we know the equal time quantization conditions for the ﬁeld operators from (8. In a similar manner. s ) d† (k .26) It follows from (8. it can be shown that m † e−ik·x vα (k.

s′ c† (k′ s′ )eik ·x + v k′ . s′ )c† (k′ . s) + eik·x d† (k. If we now use the relations (3.s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m k0 ′ × u k′ . s) = (γ 0 k0 − γ · k − m)u(k. s). s′ )e−ik ·x × e−ik·x c(k.30) (k + m) v(k. s)u(k. s) = 0.93) and (3. The Hamiltonian (8. / we can simplify the Hamiltonian in (8. s′ d(k′ . d† (k. s)v(k. s) = 0. s′ )d(k′ . c† (k.18) of the theory. s) as well as d(k. s′ )eik ·x + v(k′ . s)(γ · k + m)u(k. s) + eik·x d† (k.294 8 Dirac field theory with all other anti-commutators vanishing.31) . / (8. s) = (γ 0 k0 − γ · k + m)v(k. s) as annihilation and creation operators for the two kinds of fermionic quanta in this theory (much like the complex Klein-Gordon ﬁeld theory).30) and write (k − m) u(k.94) (8. s). This shows that we can think of c(k. can be written out in terms of creation and annihilation operators as H= = d3 x −iψγ · ∇ψ + mψψ = d3 x (2π)3 d3 k ′ 1 s. s) . s′ )e−ik ·x × (−iγ · ∇ + m) × e−ik·x c(k.s′ =± 2 ′ d3 x ψ (−iγ · ∇ + m) ψ m k0 ′ m 3 d k k′ 0 × u(k′ . in this case. s) = 1 s. s)(−γ · k + m)v(k.

s)d(k′ . s) = 1 s. s′ )e−ik ·x ×γ 0 e−ik·x c(k. s′ )d† (k.46) (see also (3.s′ =± 2 = 1 s=± 2 d3 k Ek c† (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) −δ3 (k + k′ ) ei(k − e−i(k = 1 s. s) )x0 v k′ . s)c(k. s)d† (k. s′ γ 0 v(k. where we have used the orthogonality relations for the plane wave solutions following from (2. s)d(k. s) − eik·x d† (k. s)c† (k′ . s′ v(k. s) − v k′ . s′ )d† (k. s) d3 k u† k. s)u(k. s′ )d† (k. s) m − v † k. s′ c† (k′ . s′ γ 0 u(k. s) Ek m δss′ c† (k. s′ )c(k. −k. s′ d(k′ . s′ )d† (k. s)c† (k′ . s)v(k. s)c† (k. s′ )c(k. s) (8. s′ u(k. s′ )eik ·x + v k′ . s) v † k0 . s′ γ 0 u(k. s′ )d† (k.32) + e−2ik = 0 x0 m d3 k 1 s. The normal ordering for products of fermionic operators is deﬁned exactly the same . −k. We are yet to normal order the Hamiltonian. s′ γ 0 v(k. s) . s′ )c(k.s′ =± 2 ′ 0 +k 0 ′ 0 +k 0 )x0 u k′ . s) −e2ik 0 x0 u† k0 . s)d(k′ . s′ )c(k. s) − d(k. s)c† (−k.8.92)). s′ v(k. s′ u(k.3 Field decomposition 295 H= 1 s.s′ =± 2 d3 x 3 ′ 3 d kd k (2π)3 ′ m k′ 0 m 0 k k0 ′ × u k′ . s) − d(k.

33) so that the order of factors inside a product that is normal ordered becomes important. s′ ).36) |k.33). s′ ). (8. s′ ) : = d† (k. s)d(k′ . s)|0 . namely. every time we move a fermionic operator past another in order to bring the product into its normal ordered form. s)|0 H|0 The one particle states of the theory can now be deﬁned as (see (7. s)d(k. = 0 = 0|d† (k. (8.35) c(k. Using (8. s). s′ )d† (k.34) even though we do not put the normal ordering symbol explicitly. s) : = −d† (k. s) and d(k. s). we can deﬁne the vacuum state of the theory to correspond to the state which is annihilated by both c(k. (8.21)). s . s)d(k′ .34) As in the case of the complex (Klein-Gordon) scalar ﬁeld theory (see (7. s) . : d(k′ . we write products with the creation operators standing to the left of the annihilation operators. the normal ordered Hamiltonian for the free Dirac ﬁeld theory in (8. However.296 8 Dirac field theory way as for the bosonic operators. s)|0 d(k. s) (The Hamiltonian is understood to be normal ordered as in (8. we pick up a negative sign since they anti-commute (as opposed to bosonic variables). = d† (k. s)c(k. s |k. we now have : d† (k. (8. For example. = 0.) = 0 = 0|c† (k.32) is obtained to have the form : H := s=± 1 2 d3 k Ek c† (k. s)d(k′ . s)|0 . s) + d† (k.22)) = c† (k.

We note that under the global phase transformation (α is the constant parameter of transformation) ψ(x) → ψ ′ (x) = e−iθ ψ(x).9) L = ψ (iγ µ ∂µ − m) ψ.34) with energy eigenvalue k0 = Ek = k2 + m2 > 0. ′ (8. ψ(x) → ψ (x) = ψ(x)eiθ . there are two distinct one particle states with degenerate mass (energy) and to understand the meaning of these states. ψ). 8.4 Charge operator 297 Both these states can be checked to be eigenstates of the Hamil√ tonian (8.40) . as in the case of the complex scalar ﬁeld theory. the Lagrangian density (8. Therefore.12)).37) or the inﬁnitesimal form of it (ǫ is the constant inﬁnitesimal parameter of transformation) δǫ ψ = ψ ′ (x) − ψ(x) = −iǫψ(x).9) (or (8.8. ψ) → L(ψ ′ . δǫ ψ = ψ (x) − ψ(x) = iǫψ(x).4 Charge operator The Dirac 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. we need to analyze the symmetry properties of the theory.38) (8. transforms as L(ψ. ψ ) = ψ (iγ µ ∂µ − m) ψ ′ ′ ′ ′ (8. (8.39) = ψeiθ (iγ µ ∂µ − m) e−iθ ψ = ψ (iγ µ ∂µ − m) ψ = L(ψ.

38).43) where J µ (x) = ψ(x)γ µ ψ(x).29)) and it can be checked using the equations of motion (8.7) that this current is conserved. using the inﬁnitesimal transformation (8.9) is invariant under the global phase transformation (8.41) On the other hand. (8. where we have rearranged the fermionic operators keeping in mind their anti-commuting nature. It is a vector current as in the case of the complex Klein-Gordon ﬁeld theory (see (7.37) (or (8.37) of the system is given by (note that in the ﬁrst quantized theory. we obtain (see (6. the N¨ther current associated with the internal symmetry o transformation (8.37) (or (8.42) δǫ ψ α = (−iǫψ)α (−iψγ µ )α = ǫψ(x)γ µ ψ(x).44) represents the current independent of the parameter of transformation. (8.45) The conserved charge associated with the symmetry transformation (8. this would correspond to the total probability) . (8. Thus.26).13) and (7. we note that K µ = 0.298 8 Dirac field theory Namely. the Lagrangian density (8.38)). In this case. (8.38)) is obtained from (6. ∂µ J µ (x) = 0.42) 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. as in the case of the complex scalar ﬁeld theory in (7.6) and (8.13) to be µ Jǫ (x) = ǫψ(x)γ µ ψ(x) = ǫJ µ (x).

lepton number.37) (or (8. s′ ). Using the ﬁeld decomposition in (8. s′ )δss′ δ3 (k′ − k)|0 = c† (k. etc) which are associated with U (1) symmetry transformations as well. s′ )c(k′ . s′ ) |0 = 1 s′ =± 2 d3 k′ c† (k′ . = 1 s′ =± 2 d3 k′ c† (k′ .4 Charge operator 299 Q= d3 x J 0 (x) = d3 x ψ † (x)ψ(x). s) + − c† (k. s)c(k. s = 0. s)|0 = 1 s′ =± 2 d3 k′ c† (k′ . s′ ) c(k′ .46) in terms of creation and annihilation operators as : Q := s=± 1 2 d3 k c† (k. s)|0 = |k.46) is associated with an Abelian symmetry transformation (8.46) We note here that the charge (8.47) As in the case of the complex Klein-Gordon ﬁeld. Therefore. But. since we are dealing with a fermion system. s) − d† (k. (8. we can express (8.34)) Q|0 Q|k. as in the case of the complex Klein-Gordon ﬁeld we can associate the charge operator with that of the electric charge. s′ )d(k′ .38)).33) and (7. . s)c(k′ . s)d(k. s′ ) c† (k.8. s′ ) − d† (k′ . s . it is also possible to associate the charge operator with a fermion number (such as baryon number.23). we can now show that the charge operator (normal ordered) satisﬁes (see (7. c† (k. s) . (8.

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

Let us recall the identity (i∂ − m) (i∂ + m) = (i∂ )2 − m2 / / / = − 1 µ ν γ . (8.54) Therefore. from (5. we can identify (8.8.5 Green’s functions 301 (i∂ − m) S(x − y) = δ4 (x − y).53) holds for any Green’s function of the theory (retarded.).117) we know that the Green’s function for the Klein-Gordon equation satisﬁes ∂µ ∂ µ + m2 G(x − y) = −δ4 (x − y). Although the relation (8. S(x − y) is a 4 × 4 matrix function in the spinor space.53) (i∂ − m) S(x − y) = (i∂ − m) (i∂ + m) G(x − y) / / / = − ∂µ ∂ µ + m2 G(x − y) = δ4 (x − y). but it is the trivial 4 × 4 identity matrix (which we do not write explicitly). we would concentrate mainly on the Feynman Green’s function since that is the most useful . / so that (8. Furthermore.γ 2 = −γ µ ∂µ γ ν ∂ν − m2 = −γ µ γ ν ∂µ ∂ν − m2 ∂ ∂ + µ ν − m2 = −η µν ∂µ ∂ν − m2 (8.52) S(x − y) = (i∂ + m) G(x − y).50) Here. The right-hand side is also a matrix. etc.51) = − ∂µ ∂ µ + m2 . we do not have to calculate the Green’s function for the Dirac ﬁeld theory separately. From these then. advanced. Feynman. / (8.

53) can be written in the momentum space as S(k) = (k + m) G(k). / so that we can write. We note that the relation (8.145) √ where k0 = Ek = k2 + m2 > 0) S (+) (x) = (i∂ + m) G(+) (x) / = (i∂ + m) / = i 2 i 2 d3 k 1 −ik·x e (2π)3 k0 d3 k k + m −ik·x / e . 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 .58) We can express the Feynman Green’s function in terms of the positive and the negative energy Green’s functions as well (see (5.55) SF (k) = (k + m) GF (k) = lim / which can also be written as 1 .140)) k+m / .302 8 Dirac field theory in the calculation of S-matrix elements. k2 − m2 + iǫ (8.53) we can also obtain the positive and the negative energy Green’s functions for the Dirac ﬁeld theory as (see (5. (see (5. 3 (2π) k0 (8. k − m + iǫ / ǫ→0+ (8. for example.56) SF (k) = lim ǫ→0+ (8.143) and (5.146)) .57) From (8.

59) where we have used the fact that the terms involving derivatives of the step function cancel out (namely. s)vα (k. s)uα (k. s). other Green’s functions for the Dirac ﬁeld theory can also be easily obtained from the corresponding functions for the KleinGordon ﬁeld theory.62) (+) d3 k m e−ik·x d(k. (8.6 Covariant anti-commutation relations 303 SF (x) = (i∂ + m) GF (x) / = (i∂ + m) −θ x0 G(+) (x) + θ −x0 G(−) (x) / = −θ x0 S (+) (x) + θ −x0 S (−) (x). 8. where ψ α (x) = 1 s=± 2 (+) (−) (8. δ(x0 )(G(+) (x) + G(−) (x)) = 0). s)uα (k. s)v α (k. (2π)3 k0 m eik·x d† (k.60) where (+) ψα (x) = 1 s=± 2 d3 k m e−ik·x c(k. (8. s). (2π)3 k0 m eik·x c† (k. Similarly. (8. we can decompose the adjoint ﬁeld operator (8.8) also as ψ α (x) = ψ α + ψ α (x). (8.6 Covariant anti-commutation relations Let us decompose the ﬁeld operator into its positive and negative energy (frequency) parts as (+) (−) ψα (x) = ψα (x) + ψα (x).8.63) (2π)3 k0 ψ α (x) = 1 s=± 2 (−) d3 k . s).61) (2π)3 k0 (−) ψα (x) = 1 s=± 2 d3 k Similarly. s).

ψ β (y) (−) + ′ = s. s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 × e−ik·x+ik ·y uα (k. s′ ) d3 kd3 k′ (2π)3 m2 k0 k′ 0 . s).s′ =± 1 2 m2 k0 k′ 0 + × eik·x−ik ·y vα (k. s′ = s. d(k. s). (8. s′ = 1 s. s)uβ k′ . s)uβ k′ .s′ =± 1 2 m2 k0 k′ 0 + × e−ik·x+ik ·y uα (k.106) as well as the identiﬁcation in (8.64) where we have used the completeness relation in (3.304 8 Dirac field theory It is clear from the basic anti-commutation relations for the creation and the annihilation operators c(k. c† (k′ . c† (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). we have d3 kd3 k′ (2π)3 (−) ψα (x).58). s′ δss′ δ3 (k − k′ ) = 1 s=± 2 d3 k m −ik·(x−y) e uα (k. s). ψ β (y) (+) + = s.s′ =± 1 2 ′ c(k. s) and d† (k. s). s) in (8. 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). Similarly. s)uβ (k. d(k′ .s′ =± 2 ′ d† (k. s)v β k′ .

in rearranging terms to bring them to a particular form. s′ δss′ δ3 (k − k′ ) = 1 s=± 2 ′ d3 k m ik·(x−y) e vα (k. ψ β (y) (+) (−) (8.8. we can write (+) = ψα (x). we need to be careful about negative signs that can arise from changing the order of any two such operators. 8.63) that the positive and the negative energy parts of the ﬁeld operators contain annihilation and creation operators respectively) . ψ β (y) + + (−) + ψα (x).136) (see also (5. ψ β (y) (−) (+) + = −iSαβ (x − y) − iSαβ (x − y) = −i S (+) (x − y) + S (−) (x − y) = −iSαβ (x − y). Thus. by deﬁnition the normal ordered product of two ﬁeld operators yields (note from (8.7 Normal ordered and time ordered products The deﬁnitions of normal ordered and time ordered products for Dirac ﬁeld operators are still the same as discussed in sections 6.6. As a result.61) and (8. where we have used (3.109). However.5 and 6.9. s)v β k′ . since fermionic ﬁeld operators anti-commute (Grassmann variables). 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).7 Normal ordered and time ordered products 305 × eik·x−ik ·y vα (k.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 5.65) ψα (x).66) which is the analogue of the Schwinger function (5. s)v β (k. αβ (8.

68) as ψα (x)ψ β (y) = : ψα (x)ψ β (y) : −iSαβ (x − y).70) (8. it is clear that (+) (−) : ψα (x)ψ β (y) : = : ψα (x) + ψα (x) ψ β (y) + ψ β (y) : (+) (+) = : ψα (x)ψ β (y) + ψα (x)ψ β (y) (−) (−) + ψα (x)ψ β (y) + ψα (x)ψ β (y) : (+) (+) = ψα (x)ψ β (y) − ψ β (y)ψα (x) (−) (−) + ψα (x)ψ β (y) + ψα (x)ψ β (y) (+) = ψα (x)ψ β (y) − ψα (x). As a result.67) and we note that.68) where in the last step we have made the identiﬁcation in (8. using (8.67). we can write (8. Similarly. From the deﬁnition in (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) :. (+) (−) (+) (8. ψ β (y) (−) + (+) (−) (+) (−) (+) (−) (+) (−) (+) (−) = ψα (x)ψ β (y) + iSαβ (x − y). (+) (8. we see that we can denote 0|ψα (x)ψ β (y)|0 = −iSαβ (x − y) = ψα (x)ψ β (y).69) (8. (+) (+) (+) (+) (+) (8. unlike the bosonic case. it is easy to show that ψ β (y)ψα (x) = : ψ β (y)ψα (x) : −iSαβ (x − y).65). In other words.71) . here the order of factors inside normal ordering can lead to additional signs.64).

72) This can be expressed in terms of the normal ordered products as T ψα (x)ψ β (y) = θ(x0 − y 0 ) : ψα (x)ψ β (y) : −iSαβ (x − y) −θ(y 0 − x0 ) : ψ β (y)ψα (x) : −iSαβ (x − y) = θ(x0 − y 0 ) : ψα (x)ψ β (y) : + θ(y 0 − x0 ) : ψα (x)ψ β (y) : +i − θ(x0 − y 0 )Sαβ (x − y) + θ(y 0 − x0 )Sαβ (x − y) = = θ(x0 − y 0 ) + θ(y 0 − x0 ) : ψα (x)ψ β (y) : + iSF . (8. It follows now from (8. (8.104). .5 which we will not repeat here.αβ (x − y) = ψα (x)ψ β (y).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).73) (+) (−) (−) (+) where we have used the deﬁnition in (8.59) and which can be compared with (6.αβ (x − y). 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). The Wick’s theorem for normal ordered products of Dirac ﬁelds can now be developed in a way completely analogous to the discussions in section 6.8.74) Once again. 0|T ψα (x)ψ β (y) |0 = iSF .αβ (x − y) : ψα (x)ψ β (y) : + iSF .81). (8.7 Normal ordered and time ordered products 307 which can be compared with (6. we can develop Wick’s theorem for time ordered products of Dirac ﬁelds in a straightforward manner as discussed in section 6.6.

75) As a result. in this section we will discuss brieﬂy the quantization of such ﬁelds.164)). for example.8 Massless Dirac ﬁelds 8 Dirac field theory As we have seen in section 3.77) to obtain (here k0 = |k|) . Let us consider the ﬁeld quantization of a right-handed Dirac ﬁeld (Weyl ﬁeld). Using the deﬁnition of the chiral spinors in section 3.7. it is most natural to decompose it into its chirality states. the free Dirac Lagrangian density for a massless spinor ﬁeld naturally decomposes as / / / L = iψ∂ ψ = iψ R ∂ ψR + iψ L ∂ ψL .23) because we are using here the normalization (2. Unlike in (8. 2 1 (½ − γ5 ).77) where.53) and (2.23)) d3 k (2π)3 |k| ψR (x) = e−ik·x c(k)uR (k) + eik·x d† (k)vR (k) . the discussion following (3.76) Such massless left-handed and right-handed Dirac ﬁelds (Weyl ﬁelds) arise naturally in physical theories (as we will see. as before.3).7. we can expand the ﬁeld as (see (8. Correspondingly. in the case of the standard model in 14. we have identiﬁed k0 = Ek = |k|.308 8. (8. we note here that there is no sum over spin polarization because the chiral spinors are charaterized by their unique chirality or helicity (see.23). the multiplicative factors in (8. Using the orthonormality relations for the right-handed spinors described in (3. Therefore. for example. (8. 2 ψR (x) = ψL (x) = (8. Furthermore.170) we can invert the decomposition in (8.77) is diﬀerent from those in (8.54) which is appropriate for massless spinors. when the Dirac particle is massless. The discussion for the left-handed ﬁeld can be carried out in an analogous manner. we can also decompose a Dirac ﬁeld into a left-handed and a right-handed component as 1 (½ + γ5 )ψ(x).

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

77) and using the properties of the vacuum 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 .82) Putting in the ﬁeld decomposition 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 . We recall that the Feynman propagator is deﬁned as the time ordered product (see.74)) iSF .80) in evaluating the vacuum expectation values. let us determine the propagator for a righthanded fermion.83) Here we have used (8.Rαβ (x − y) = d3 kd3 k (2π)3 |k||k | × θ(x0 − y 0 )e−ik·x+ik ·y uRα (k)u† β (k ) 0|c(k)c† (k )|0 R † −θ(y 0 − x0 )eik·x−ik ·y vRα (k)vRβ (k ) 0|d(k )d† (k)|0 = d3 k e−ik·(x−y) θ(x0 − y 0 )uRα (k)u† β (k) R (2π)3 |k| † −eik·(x−y) θ(y 0 − x0 )vRα (k)vRβ (k) . (8.84) . for example. (8.310 8 Dirac field theory As an example. (8.81) we obtain (we assume k0 = |k| in the ﬁeld decomposition) iSF . (8. Dropping the spinor indices and using (3.

Using the integral representation for the theta function (see (6.100) and the limit → 0 is to be understood) we can write the ﬁrst term in the integrand as 1 0 0 θ(x0 − y 0 )(|k|γ 0 − γ · k)e−i|k|(x −y ) 4|k| = − = − = − = − dk0 (|k|γ 0 − γ · k) e−i(k +|k|)(x 2πi 4|k| k0 + i 0 0 0 0 0 −y 0 ) dk0 (|k|γ 0 − γ · k) e−ik (x −y ) 2πi 4|k| k0 − |k| + i dk0 (k0 γ 0 − γ · k) e−ik (x −y ) 2πi 4k0 k0 − |k| + i dk0 k e−ik (x −y ) / .8. 2πi 4k0 k0 − |k| + i 0 0 0 0 0 0 (8.84) separately.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. 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 .

the conventional covariant propagator (deﬁned as the time ordered product of ψR ψ R ) can be obtained from (8.R (k) = (8.R (k) = iSF . 8. adding the two results in (8.R (k)γ 0 = ik(½ − γ5 ) / ikPL / = 2 .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. (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.89) We note that although our discussion has been within the context of right-handed fermion ﬁelds.86) Therefore. 2(k2 + iǫ) k + iǫ (8.82) as † the time ordered product of ψR ψR and not as the conventional time ordered product of ψR ψ R (see.86) and substituting into (8.85) and (8. everything carries over in a straightforward manner to left-handed fermion ﬁelds.74)).87) iSF .R (x − y) = / d4 k −ik·(x−y) ikγ 0 (½ + γ5 ) e . However.88) by simply multiplying a factor of γ 0 on the right which leads to iS F .88) We note here that we have deﬁned the propagator in (8. The reason for this is that the adjoint ﬁeld ψ R has negative chirality. for example. 2πi 4k0 k0 + |k| − iǫ 0 0 0 8 Dirac field theory (8. 2(k2 + iǫ) k + iǫ iSF . 4 (2π) 2(k2 + iǫ) It follows now that the fermion propagator in the momentum space has the form ikγ 0 PR / ikγ 0 (½ + γ5 ) / = 2 .312 = − / dk0 k e−ik (x −y ) . For .

we can separate the Lagrangian density (8. As in (6.90) 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. ψψ is invariant under a Lorentz transformation and φ is a scalar ﬁeld) and is known as the Yukawa interaction..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. L = L0 + LI . the correct parity invariant trilinear interaction for a pseudoscalar meson should be (−gψγ5 ψφ). where 1 M2 2 / φ . (The self-interaction can be switched oﬀ at the lowest order by setting λ = 0. The trilinear coupling gψψφ is manifestly Lorentz invariant (as we have seen in (3. L0 = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 λ LI = −g ψψφ − φ4 .9 Yukawa interaction 313 example.86).36). Namely. we can denote the interaction Lagrangian density as (remember everything is normal ordered) .) In this case. since it leads to the Yukawa potential in the Born approximation and the coupling constant g represents the strength of the interaction. we are going to ignore this aspect of the meson and also switch oﬀ the self-interactions of the meson ﬁeld for simplicity. the π 0 meson) has mass M . 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. since here we are only interested in developing calculational methods in a quantum ﬁeld theory.90) into a free part and an interaction part. 4! (8. However. (8.91) (8. in this case.92). has the form 1 λ M2 2 / φ − g ψψφ − φ4 . The Lorentz invariant Lagrangian density.8.g.

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. Using Wick’s theorem (see sections 6.5 and 6.6).69). in this theory the S-matrix will have the expansion (see (6. (8.314 8 Dirac field theory As a result.94) +··· . the adiabatic switch-oﬀ is assumed) S = T e−i R∞ R −∞ LI = −g : ψψφ := −HI . we can write (8. (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′ ) + ······ .95) .

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

s) ˜ ˜ × 0|d(p′ . (2π)3 k′ 0 (8. In fact. energy conservation δ(k0 + k′ 0 + q 0 ) in (8.96). q 0 > 0.˜′ =± 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′ . The matrix element (8. s′ )c(p.95) also leads to a distinct physical process. k′ 0 . The delta function merely ensures the overall conservation of energy and momentum in the process. s)d† (k′ . However. s. note that since k0 .99) represents the process where a proton.˜′ =± 2 ˜s (2π)3 2˜ 0 p m (2π)3 p0 m (2π)3 p′ 0 × (2π)4 δ4 (p + p′ + p) v α (p′ . s′ )uα (p. s). s′ )uα (k. our goal in this section has been to describe the calculational methods in quantum ﬁeld theory. let us consider the second . k′ . an anti-proton along with a π 0 meson are annihilated.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) ˜ ˜ ˜ 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′ . 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. we note here that this lowest order matrix element in (8. (For completeness. For example.316 8 Dirac field theory 0|S1 |k. s′ )uα (p. s)a(˜ )c† (k.99) vanishes because energy-momentum conservation cannot be satisﬁed. q = −ig d4 x s.) Each term at order g2 in the expansion in (8. s′ )a† (q)|0 ˜ ˜ p = −ig d 3 p d 3 p d 3 p′ ˜ 1 s.

s3 |S2 |k1 . s1 . s3 )vα (k4 . Recalling that the Fourier transformation of the Green’s function is given by (see (5. s2 = − ig2 2 d4 xd4 x GF (x − x ) (−) (+) (+) (2) (−) × k4 .8. s4 )v β (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. In such a case.95) (2) S2 = − =− g2 2 g2 2 d4 xd4 x φ(x)φ(x ) : ψ(x)ψ(x)ψ(x )ψ(x ) : d4 xd4 x iGF (x−x ) : ψ(x)ψ(x)ψ(x )ψ(x ) : . s2 )uβ (k1 . the matrix element will have the form k4 . k2 . s1 . k2 . s4 .120)) d4 q −iq·(x−x ) e GF (q).101) × uα (k3 . s1 ). s1 . k3 . (8. k3 .9 Yukawa interaction 317 term in the expansion at order g2 in (8. k2 . s2 (−) (+) (+) = −ig2 × d4 xd4 x GF (x − x ) (−) (+) (+) (−) k4 . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x ) (−) + ψ α (x )ψα (x )ψ β (x)ψβ (x)|k1 . (2π)4 GF x − x = (8. s4 . s4 .100) A nontrivial matrix element of this operator would be obtained if the initial state contains a proton and an anti-proton of momenta k1 and k2 respectively and the ﬁnal state also contains a proton and an anti-proton of momenta k3 and k4 respectively. k3 .102) . s3 |ψ α (x)ψα (x)ψ β (x )ψβ (x )|k1 .

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

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

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. (anti-particle creation). for the Yukawa interaction (8. 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. the normalization factor for the external line √ (state) is denoted by (k0 = k2 + M 2 ) k = 1 (2π)3 2k0 . we . in a local quantum ﬁeld theory. s α = However. (anti-particle annihilation). (8.106) k.320 8 Dirac field theory and in such a case. v α (k. The interaction can be read out from the Lagrangian density LI and.107) k. is represented as a vertex (point of interaction) with no meaning attached to the external lines. s α = Note that it is the interaction between the diﬀerent ﬁelds which gives rise to nontrivial scattering.105) On the other hand.93). s). (8. s). (8. since the fermion ﬁeld carries a charge (and. (particle annihilation). therefore. s). vα (k. Thus. carries a direction for charge ﬂow). (particle creation). s).

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

s4 α β q δ γ k2 .2: A second 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′ . k′ . s1 k4 . s′ )uα (k. s′ . (2π)3 k′ 0 (8. the corresponding lowest order Feynman diagram will be given by Fig. The value of this diagram can be obtained from the Feynman rules to be 0|S1 |k. s3 Figure 8. therefore. with the external lines representing physical particles. 8.99). s). s) (2π)3 k0 m v β (k′ . Let us next note that if we are interested in the second order process where a proton and an anti-proton annihilate and create a π 0 meson which subsequently pair creates a proton and an anti-proton. q = −(2π)4 igδ4 (k + k′ + q)δαβ × 1 (2π)3 2q 0 m uα (k. 8. s.322 8 Dirac field theory In this language. the physical process where a proton.109) This is exactly the value of the ﬁrst order S-matrix element which we had evaluated in (8. k1 . s2 k3 . an anti-proton and a π 0 meson are annihilated (which we considered in (8.1. The .2. 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.

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

112) then.324 8 Dirac field theory −iV (q) = −ig2 GF (q) = ǫ→0+ lim − ig2 . Furthermore. q 2 − M 2 + iǫ q = k1 − k2 .113) Therefore. 4π |x| = (8.114) .113). q = (k1 − k2 ) = (0. we obtain d3 q iq·x e V (q) (2π)3 eiq·x d3 q (2π)3 −(q2 + M 2 ) + iǫ d|q||q|2 0 ∞ 0 ∞ −∞ π 2π −iV (x) = −ig2 = = = = = ǫ→0+ lim −ig2 ∞ 0 ig2 (2π)3 ig2 (2π)2 g2 dθ sin θ 0 dφ d|q| i|q||x|(|q|2 + M 2 ) zeiz|x| z2 + M 2 iM e−M |x| 2iM |q|2 ei|q||x| cos θ |q|2 + M 2 ei|q||x| − e−i|q||x| (2π)2 |x| dz g2 (2πi) (2π)2 |x| ig2 e−M |x| . we see that V (q) = V (q) and taking the Fourier transform of (8. k1 − k2 ). it follows from energy conservation that 0 0 k1 = k2 . k2 = −k3 . if we are in the center of mass frame where k1 = −k4 . (8.111) where V (q) represents the scattering amplitude in the Born approximation. 0 0 k2 = k3 . (8. (8. 0 0 k1 = k4 .

Shirkov. 8. N. Introduction to the theory of Quantized Fields. Introduction to Relativistic Quantum Field Theory. D. Moscow (1984). S.93) leads to the Yukawa potential in the Born approximation. Roman. Itzykson and J-B. 1980. 4. 6. N. 2. Bjorken and S. Bogoliubov and D. New York (1993). J.114) the potential between the fermions to be g2 e−M |x| . Drell. Peterson. 4π |x| V (x) = − (8. 3. Quantum Field Theory. 5. F. V. 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.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. 1964. C. John Wiley. John Wliley. . New York.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). Schweber. Relativistic Quantum Mechanics and Field Theory.11 References 1. McGrawHill. New York. Evanston (1961).8. Gross. New York (1969). McGrawHill. Nauka. Zuber. therefore. Relativistic Quantum Fields. we obtain from (8. P. Introduction to Quantum Field Theory. Row.

This page intentionally left blank .

Thus.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. ∂B . It is more convenient from our point of view to rewrite Maxwell’s equations (9.1) in a manifestly Lorentz covariant form.1) where E and B represent the electric and the magnetic ﬁelds respectively. the dynamical ﬁeld would naturally describe particles with spin 1.1) to write 327 . this is described by the vector potential Aµ (x) which belongs to the ( 1 .Chapter 9 Maxwell ﬁeld theory 9. ∂t ∂E ∇×B= . ∂t ∇×E=− ∇ · B = 0. 1 ) representation of the Lorentz group (see sec2 2 tion 4. Let us start with the classical Maxwell’s equations in vacuum which are given by (remember c = 1) ∇ · E = 0. 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. we note that we can solve the second and the third equations in (9. As we know. (9.2). To that end.

∂t (9.3) If we now deﬁne the four dimensional curl of the four vector potential as Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . the six independent components of the tensor Fµν are given by (three components each of) the electric and the magnetic ﬁelds. (9. ν = 0.328 9 Maxwell field theory B = ∇ × A. (9. denotes the electric ﬁeld vector. E2 . (9. the second . (9. 2. Fij = ∂i Aj − ∂j Ai = −ǫijk Bk . Similarly. Therefore. E = − ∂A − ∇Φ.6) (9. −A). In other words. where E = (E1 . A).7) (9. 3. we note that F0i = ∂0 Ai − ∂i A0 = Ei . B2 .2) where A(x) and Φ(x) are known as the vector and the scalar potentials respectively. E3 ).4) where ǫijk denotes the three dimensional Levi-Civita tensor and B = (B1 .5) µ. B3 ). of course. combine them into the four vector potential Aµ = (Φ. 1. Aµ = (Φ.8) denotes the magnetic ﬁeld vector. We can.

with the ﬁeld strength tensor deﬁned in (9. or. Similarly.1). for ν = 0.1 Maxwell’s equations 329 rank anti-symmetric tensor Fµν is also known as the ﬁeld strength tensor. we see that the set of four Maxwell’s equations (9.11) which is the last equation in (9. or. ∂Ej + ∂ i (−ǫijk Bk ) = 0. ∂ 0 F0j + ∂ i Fij = 0.4)) ∂i F i0 = 0. or. for ν = j. or. ∂t (9.12) .13) (9.9) takes the form (F 00 = 0 by deﬁnition (9.4) as Fµν = ∂µ Aν − ∂ν Aµ . or. or.9. ∂t ∂E = ∇ × B. It is now easy to check that the other two equations (the ﬁrst and the fourth) in (9.9) For example.1) are given by ∂µ F µν = 0.1). equation (9. (9. equation (9. Thus. ∂t ∂Ej = −ǫjik ∂ i Bk .1) can be written in the manifestly covariant form as ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0. ∂0 F 0j + ∂i F ij = 0. (9.9) leads to ∂µ F µj = 0. ∇ · E = 0. (9.10) which coincides with the ﬁrst of the equations in (9.

For example. we have to treat the four vector potential Aµ (x). Maxwell’s equations (9. we note that under a gauge transformation Aµ (x) → A′ (x) = Aµ (x) + ∂µ θ(x). 2 which leads to . the gauge invariance of Maxwell’s equations is obvious in this formulation.330 9 Maxwell field theory Furthermore. which denotes the dynamical variable of the theory. µ (9. µ or. (9. real.2 Canonical quantization In trying to second quantize the Maxwell ﬁeld theory. But ﬁrst.15) In other words.14) where α(x) is an arbitrary. Fµν ′ → Fµν = ∂µ A′ − ∂ν A′ µ ν = = ∂µ (Aν + ∂ν θ) − ∂ν (Aµ + ∂µ θ) ∂µ Aν − ∂ν Aµ = Fµν . let us examine whether there exists a Lagrangian density which will lead to Maxwell’s equations as its Euler-Lagrange equations. as an operator. 9. Let us consider the Lagrangian density 1 1 L = − Fµν F µν = − (∂µ Aν − ∂ν Aµ )(∂ µ Aν − ∂ ν Aµ ) 4 4 1 (9. the ﬁeld strength tensor is unchanged under the gauge transformation (9.16) = − ∂µ Aν (∂ µ Aν − ∂ ν Aµ ). local parameter of transformation. δAµ (x) = A′ (x) − Aµ (x) = ∂µ θ(x). Consequently.12) are also invariant under these transformations.14) of the four vector potential Aµ (x).

= − 1 1 1 (∂ µ Aν − ∂ ν Aµ ) − ∂ µ Aν + ∂ ν Aµ 2 2 2 (9.16). namely. Note that a Lagrangian density 1 1 Fµν F µν − Fµν (∂ µ Aν − ∂ ν Aµ ) .17) = − (∂ µ Aν − ∂ ν Aµ ) = −F µν . ∂µ or. ∂∂µ Aν (9. Fµν = ∂µ Aν − ∂ν Aµ . gives Maxwell’s equations as its Euler-Lagrange equations. ∂Aν ∂∂µ Aν or. 4 2 L= (9.20) ∂µ F µν = 0.16) when the ﬁeld variable Fµν is eliminated using the (ﬁrst) auxiliary ﬁeld equation in (9. ∂µ (∂ µ Aν − ∂ ν Aµ ) = ∂µ F µν = 0. ∂Fµν or. so that the Euler-Lagrange equations. take the form ∂L ∂L − ∂µ = 0.2 Canonical quantization 331 ∂L ∂Aν ∂L ∂∂µ Aν = 0. However.16) since the Lagrangian density (9. we will work with the simpler Lagrangian density in (9.19) is equivalent to (9. The Lagrangian density (9. ∂L = 0. It is manifestly Lorentz invariant. indeed.9). . (9. in this case. ∂L = 0.20).9.18) which coincides with the manifestly covariant Maxwell’s equations (9.19) with both Aµ and Fµν treated as independent dynamical variables would also lead to Maxwell’s equations as its Euler-Lagrange equations.

332 9 Maxwell field theory Furthermore. the momentum conjugate to the ﬁeld variable A0 (x) vanishes.23) This is equivalent to saying that.14) of the vector potential. would then appear to be [Ai (x). This is a consequence of the gauge invariance of the theory. for canonical quantization of such a system.25) . in this gauge. Therefore. (9. As a result.22) In other words. 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.21). (9. Π0 (x) = −F 00 = 0.15)). (9. ˙ ∂ Aµ (x) (9.21) 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. since the ﬁeld strength tensor Fµν is gauge invariant (see (9. x0 =y 0 . 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. in this gauge the nontrivial components of the conjugate momentum are given by ˙ Πi (x) = −F 0i = Ei = −(∂ 0 Ai − ∂ i A0 ) = −Ai (x). namely. this Lagrangian density is also invariant under the gauge transformation (9. Aj (y)]x0 =y0 Ai (x).24) The naive canonical quantization conditions.17)) Πµ (x) = ∂L = −F 0µ . Furthermore. Πj (y) x0 =y 0 = 0 = Πi (x). now must commute with all the ﬁeld variables in the theory. we note that A0 (x) which would have a nontrivial commutation relation only with Π0 (x). (9. Πj (y) j = iδi δ3 (x − y). If we treat Aµ (x) as the independent ﬁeld variable. we can deﬁne the associated conjugate momentum as (see (9.

2 Canonical quantization 333 However. ∂ 0 (∇ · A) = 0.23). or. Aj (y) j = −iδi TR (x − y). the canonical quantization condition (9.30) . (9. this is not quite consistent which can be seen as follows. The proper quantization condition consistent with the transversality condition (9. the dynamical ﬁeld variable A(x) must be transverse. lead to (for ν = 0) ∂µ (∂ µ A0 − ∂ 0 Aµ ) = 0. ¯ A(x) → A′ (x) = A(x) + ∇θ(x). or.23).18)) ∂µ (∂ µ Aν − ∂ ν Aµ ) = 0. x0 =y 0 x0 =y 0 (9. (9. Πj (y) or. in the gauge (9.27).27) (9. For example.29) which does not satisfy the transversality condition (9.) On the other hand.27) can be thought of as the gauge choice for this residual invariance. the dynamical equations (see (9. namely. we note that even after imposing the gauge condition (9.28) and the condition (9.25) leads to Ai (x). in this gauge.23). ˙ Ai (x). the theory can be checked to have a residual gauge invariance under a time independent gauge transformation which preserves (9.9. j = −iδi δ3 (x − y). or. (As a parenthetical remark. ∂ 0 (∂µ Aµ ) = 0. ∇ · A = 0. x0 =y 0 (9.26) Namely.27) is given by ˙ Ai (x). Aj (y) j = iδi δ3 (x − y).

31) so that (remember that ∂ i ∂i = −∇2 ) j j ∂ i δi TR = ∂ i δi + ∂i 1 j 2 ∂ ∇ (9.334 9 Maxwell field theory where the transverse delta function is formally deﬁned as the nonlocal j operator (whose coordinate representation is given by δi TR (x − y)) j j δi TR = δi + ∂i 1 ∂j . (2π)3 (9. ∇ and (9.27) is given by (9. When quantized systematically through the Dirac method. As we have pointed out in connection with the discussions in the Dirac ﬁeld theory. the Hamiltonian density for the Maxwell ﬁeld theory can be obtained to be .27) constitute constraints on the dynamics of the theory and systems with constraints are known as constrained systems. the transverse delta function in (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).22). Here 1 2 stands for the inverse of the Laplacian operator and for∇ mally denotes the Green’s function for ∇2 . (9. Let us note here that in momentum space.33) In the gauge (9. there exists a systematic procedure for deriving the correct quantization conditions for such systems.23).23) subject to (9.27). We note here that relations such as (9. known as the Dirac method. which we will discuss in the next chapter.23) and (9.30). the correct quantization condition (Dirac bracket) that results in the gauge (9.30) is consistent with the transversality condition in (9.32) = ∂ j + ∂ i ∂i 1 j j j 2 ∂ = ∂ − ∂ = 0. ∇2 (9.

2 (9. (9. Maxwell’s equations (9.35) H= Indeed this is what we know to be the Hamiltonian (energy) for the Maxwell theory even from studies in classical electrodynamics. 9.37) .27)) A0 = 0. (9.3 Field decomposition 335 ˙ H = Πi Ai − L = −Πi Πi + = = = = 1 Fµν F µν 4 1 2F0i F 0i + Fij F ij −Πi Πi + 4 1 1 −Πi Πi + Πi Πi + (−ǫijk Bk )(−ǫijℓ Bℓ ) 2 4 1 1 1 i Π2 + B2 − Π Πi + Bi Bi = 2 2 2 1 2 E + B2 .18) take the form ∂µ F µν = ∂µ (∂ µ Aν − ∂ ν Aµ ) = Aν = 0.23) (see also (9. (9.3 Field decomposition In the temporal gauge (9.9.34) and leads to the Hamiltonian d3 x H = = 1 2 1 2 d3 x Π2 + B2 d3 x E2 + B2 .36) which also lead to ∂ µ Aµ = 0.38) (9. ∇ · A = 0.

complex satisfying k · ǫ (k. λ′ ) = δλλ′ .44) ǫ∗ (k. k · ǫ (k) = 0. will have plane wave solutions of the form A(x) ∝ ǫ(k) e∓ik·x . we must have ∇ · A(x) = 0.38) vanishes identically (because of the choice of the gauge condition (9. (9. with k0 = Ek = √ k2 = |k |.39) Thus. We note from (9.41) 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. we have Ai = 0.23)) A0 ≡ 0. λ) = 0. λ = 1. for example.336 9 Maxwell field theory As we have seen. We can choose the two independent polarization vectors to be orthonormal and. λ) · ǫ(k. . in this case.42) (9. for ν = i. On the other hand. (9. the wave equation (9. the three components of the vector potential. (9. since the vector potential is transverse. (9. There can be two such independent directions.40) (9.1. for ν = 0. as shown in Fig. in general.27) that. 9. or. therefore. 2. satisfy the Klein-Gordon equation for a massless particle (wave equation) and.43) Namely. the polarization vector characterizing the vector potential must be transverse to the direction of propagation of the plane wave.

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

λ) + ǫ∗ (−k.47) and (9. we conclude that i d3 x ← → 2k0 eik·x ∂0 A(x) = √ (2π)3 2k0 d3 x (2π)3 2k0 2 ǫ (k. λ) = −i (9. λ)e2ik t a† (−k. we can show that d3 x (2π)3 2 ˙ eik·x A(x) 0 k0 ǫ(k. λ)e 2 a(k′ . λ)a(k.49) where we have used the normalization for the polarization vectors in (9.50) . a† (k. λ)a(k. λ)a(k. λ=1 or. (9. λ) 2k′ 0 0 +k ′ 0 )t +δ3 (k + k′ )ǫ∗ (k′ . λ)−ǫ∗ (−k. λ) = 1 0 √ ǫ(k. λ)ei(k 2 a† (k′ .338 9 Maxwell field theory d3 x (2π)3 2 e ik·x A(x) = λ=1 −i(k ′ −k)·x d3 k ′ d3 x √ 2k′ 0 (2π)3 ′ × ǫ(k′ . λ). λ) = λ=1 d3 k ′ √ δ3 (k − k′ )ǫ(k′ . (9. λ) = i ← → (eik·x ∂0 A(x)) · ǫ∗ (k. λ).44). λ). λ) . λ)ei(k +k)·x a† (k′ .48) 2 =−i λ=1 From (9. λ)a(k′ . λ)e2ik t a† (−k. Taking the Hermitian conjugate of (9. λ) . λ) + ǫ∗ (k′ .48).49) leads to d3 x (2π)3 2k0 ← → (e−ik·x ∂0 A(x)) · ǫ(k. a(k.47) 2k0 λ=1 Similarly. (9.

j = −iδi TR (x − y). Aj (y)]x0 =y0 ˙ [Ai (x). λ) behave exactly like the annihilation and the creation operators for a harmonic oscillator system. λ)a(k.50) denote the two inversion formulae for the Maxwell ﬁeld (analogous to (5. λ). The normal ordered forms for these operators can be shown to correspond to (see (5.50). λ).3 Field decomposition 339 Equations (9. The Hamiltonian and the momentum operators for the Maxwell theory can again be expressed in terms of the creation and the annihilation operators. λ) and a† (k.9. Given the quantization conditions (see (9. Aj (y)]x0 =y0 (9.49) and (9. λ)a(k. λ′ )] = δλλ′ δ3 (k − k′ ). λ).97) and we do not show the normal ordering symbol explicitly) 2 H = λ=1 2 d3 k Ek a† (k. Aj (y)]x0 =y0 . a† (k′ . (9.61) in the case of the Klein-Gordon ﬁeld).51) using (9.52) In other words. a(k′ .77) and (5. a† (k′ . then it satisﬁes .54) The Hilbert space for this theory can be built up in the standard manner.30)). λ). [Ai (x). (9. the operators a(k. we can easily show that [a(k.49) and (9. ˙ ˙ = 0 = [Ai (x). λ′ )]. (9. λ). P = λ=1 d3 k k a† (k. [a(k. λ′ )] = 0 = [a† (k.46)) that Ek = |k|.53) where we recall from the Einstein relation (see (9. If we denote the vacuum state of the theory by |0 .

λ = 0. This.57) Therefore. the energy of the system is positive semideﬁnite and the Hilbert space is the standard harmonic oscillator space. λ)|0 H|0 = 0 = 0|a† (k. λ = a† (k. the true physical degrees of freedom are two in number. since the A(x) ﬁeld is Hermitian like the charge neutral Klein-Gordon ﬁeld. Later. we will also . λ . by eliminating the extra degrees of freedom through non-covariant gauge choices (see (9.56) which satisﬁes H|k. the canonically quantized description which involves only true degrees of freedom is not manifestly Lorentz covariant although the ﬁnal result of any physical calculation will be. we achieved quantization. λ). the particles of the theory. describes a one photon state and the index λ merely labels the polarization of the photon. λ 2 = (Ek − k2 )|k. We can. do not carry any electric charge. (9. shows that even though the vector potential Aµ (x) has four independent ﬁeld degrees of freedom. there are two distinct one photon states corresponding to the two possible independent transverse polarizations a photon can have. 2) |k. the one particle state satisﬁes (H 2 − P2 )|k. namely the photons. (9. (9. λ = Ek |k. As a result. The one particle state is then obtained to be (λ = 1. = k |k.27)). in fact. involving the true degrees of freedom. λ)|0 . λ . First of all. Second. (9. show explicitly that the polarization is directly related to the spin of photon. consequently.23) and (9.55) = 0 = P|0 . There are several things to note here. λ P|k. Furthermore.58) This state.340 9 Maxwell field theory a(k. therefore. However.

62) describe respectively the left and the right circular polarization vectors. λ) = ex a† (k. e 2 1 ˆ ǫR (k) = √ ex + iˆy . (These one photon states are eigenstates of helicity with eigenvalue ±1. 1) + ey a† (k. 2) λ=1 = 1 (ˆx + iˆy )(a† (k.60) would then create photons which are left and right circularly polarized respectively. 2) as creating a photon polarized along the y-axis. R L R L where we have used the familiar identiﬁcation that 1 ˆ ǫL (k) = √ ex − iˆy . 1) + ia† (k. 2) . 1) − ia† (k.3 Field decomposition 341 quantize the Maxwell ﬁeld theory in a manifestly covariant manner which will bring out some other interesting aspects of such a theory. 2) = ey . 2 1 √ a† (k. 1) − ia† (k.59) then. ˆ ǫ(k. 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. in this language. 1) as creating a photon polarized along the x-axis and a† (k. 2 (9. 2) . . that the operators a† (k) = L a† (k) = R 1 √ a† (k. (9.9. It is clear. we can identify a† (k. λ)a† (k. 1) + ia† (k. 2)) e e 2 1 e e + (ˆx − iˆy )(a† (k. 1) = ex .) 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. 2)) 2 (9. e 2 (9.61) = ǫ∗ (k)a† (k) + ǫ∗ (k)a† (k).

if we choose the polarization vectors to be real as in (9. λ)+eik·x a† (k.27)). 3 2k 0 (2π) d3 k (2π)3 2k0 ǫµ (k. λ′ )e−ik·x+ik ·y δλλ′ δ3 (k − k′ ) . λ). λ)ǫν (k′ . ǫ(k.63) and furthermore. λ)). λ)ǫν (k′ . if we deﬁne the two polarization vectors for λ = 1. λ). λ)).52). a† (k′ .64) with k0 = |k|. (9. λ)(e−ik·x a(k.23) (see also (9. This allows us to decompose the ﬁeld into its positive and negative energy parts as 2 A(+) (x) µ = λ=1 2 d3 k ǫµ (k. λ)eik·x a† (k.λ′ =1 1 (2π)3 ′ d3 k d3 k ′ √ √ 2k0 2k′ 0 ×ǫµ (k. A(−) (y)] µ ν = λ. λ). λ) = (0. (9.23)) 2 Aµ (x) = λ=1 d3 k (2π)3 2k0 ǫµ (k.65) A(−) (x) = µ λ=1 From the commutation relations for a(k. λ) in (9.342 9.4 Photon propagator 9 Maxwell field theory Let us note that in the temporal gauge (9.λ′ =1 1 (2π)3 d3 k d3 k ′ √ √ 2k0 2k′ 0 ′ ×ǫµ (k. 2 as four vectors ǫµ (k.59) then we can write the ﬁeld decomposition (9. for simplicity. λ) and a† (k. it follows that the only nontrivial covariant commutation relations have the form (these are not at equal time) 2 [A(+) (x). λ′ )] 2 = λ.45) as (recall (9. λ)e−ik·x a(k. (9. λ′ )e−ik·x+ik ·y [a(k.

67) Aµ (x)Aν (y) = 0|Aµ (x)Aν (y)|0 = −iG(+) (x − y). we obtain : Aµ (x)Aν (y) : +[A(+) (x). (9. λ)e−ik·(x−y) 2k0 2 = λ=1 1 (2π)3 = 1 2 d3 k e−ik·(x−y) (2π)3 k0 ǫµ (k. namely.66) which are the analogs of the positive and the negative energy Green’s functions for the Maxwell ﬁeld (see (5.107)). µν [A(−) (x). λ) λ=1 = −iG(+) (x − y).68) From this. A(+) (y)] = −[A(+) (y). we can deﬁne the Feynman propagator in the standard manner (see (6. λ) λ=1 = −iG(−) (x − y). A(−) (x)] = iG(+) (y − x) µ ν ν µ νµ = − 1 2 d3 k eik·(x−y) (2π)3 k0 2 ǫµ (k.145)). λ)ǫν (k. A(−) (y)] ν µ : Aµ (x)Aν (y) : −iG(+) (x − y) µν : Aµ (x)Aν (y) : + Aµ (x)Aν (y). µν (9.4 Photon propagator 2 343 d3 k ǫµ (k.9.143) and (5. λ)ǫν (k. λ)ǫν (k. It follows now that Aµ (x)Aν (y) = (A(+) (x) + A(−) (x))(A(+) (y) + A(−) (y)) µ µ ν ν = A(+) (x)A(+) (y) + A(+) (x)A(−) (y) µ ν µ ν (−) +A(−) (x)A(+) (y) + A(−) (x)Aν (y) µ ν µ = = = Therefore. . µν (9.

therefore.69) µν By deﬁnition. λ)ǫν (k. (9.344 9 Maxwell field theory 0|T (Aµ (x)Aν (y)) |0 = iGF . 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. for example.µν (x − y) = −θ(x0 − y 0 )G(+) (x − y) + θ(y 0 − x0 )G(−) (x − y) µν µν = −i × d3 k (2π)3 2 ǫµ (k. (9. 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 ) .(9. To simplify this expression.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) = 0|θ(x0 − y 0 )Aµ (x)Aν (y) + θ(y 0 − x0 )Aν (y)Aµ (x)|0 = −iθ(x0 − y 0 )G(+) (x − y) − iθ(y 0 − x0 )G(+) (y − x) µν νµ = −iθ(x0 − y 0 )G(+) (x − y) + iθ(y 0 − x0 )G(−) (x − y) µν µν (−) = i(−θ(x0 − y 0 )G(+) (x − y) + θ(y 0 − x0 )Gµν (x − y)). λ)ǫν (k. we use the integral representation for the step function given in (6.70) 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. the Feynman Green’s function has the form GF .59)).

2πi (k′ 0 )2 − k2 + iǫ ′0 0 0 0 1 1 − k′ 0 − k0 + iǫ k′ 0 + k0 − iǫ 0 ǫ→0+ lim − (9.140)). let us note that if we introduce two new four vectors η µ = (1. we obtain GF .µν (k) = ǫ→0+ lim 1 k2 + iǫ 2 2 ǫµ (k. λ)ǫν (k.9. λ) .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 ) . |k| (9. λ)ǫν (k. 0.73) depends on the polarization sum and to evaluate the polarization sum.71) where we have used the Einstein relation (9.46). λ) . λ) λ=1 = GF (k) λ=1 ǫµ (k. λ=1 (9.µν (x − y) = lim ǫ→0+ d4 k e−ik·(x−y) (2π)4 k2 + iǫ 2 ǫµ (k. ˆ kµ = kµ − (k · η)η µ (k · η)2 − k2 = 0. we can identify the momentum space Feynman Green’s function for the Maxwell (photon) ﬁeld to be GF .72) Therefore. (9. Substituting this into (9.70) and denoting the variable of integration k′ 0 → k0 .74) .73) where GF (k) denotes the Feynman Green’s function for a massless scalar ﬁeld (see (5. 0). The Feynman Green’s function in (9. 0. k . λ)ǫν (k.

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

83) . which follows from (9.5 Quantum electrodynamics 347 GF . The Lagrangian density for this theory is given by 1 L = − Fµν F µν + iψγ µ Dµ ψ − mψψ 4 1 = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ. ψ(x) → ψ (x) = ψ(x)eiθ(x) .81) (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)) and we have separated the Lagrangian density in (9. Aµ (x) → A′ (x) = Aµ (x) + µ ′ 1 ∂µ θ(x).74).9. e (9.80) denotes the covariant derivative (see (7. where Dµ ψ(x) = (∂µ + ieAµ (x))ψ(x). (9.0µ (k) = 0.82) The total Lagrangian density (9. (9. 4 LI = −eψγ µ ψAµ .79) Quantum electrodynamics (QED) is the theory describing the interaction of electrons and positrons with the electromagnetic ﬁeld.80) into the free and the interaction parts as 1 / L0 = − Fµν F µν + iψ∂ ψ − mψψ. 9.5 Quantum electrodynamics (9. 4 = L0 + LI .

(9.86) .80). ψ.348 9 Maxwell field theory where θ(x) is the real and ﬁnite local parameter of transformation or under the inﬁnitesimal form of the transformation (9. we need to examine the invariance of the minimally coupled Dirac Lagrangian density in (9.81) transforms covariantly under a gauge transformation (see (7. Aµ ) → LDirac (ψ ′ . Aµ ).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). = iψeiθ(x) γ µ e−iθ(x) Dµ ψ − mψeiθ(x) e−iθ(x) ψ (9. A′ ) µ ′ = iψ γ µ Dµ ψ ′ − mψ ψ ′ ′ ′ = iψγ µ Dµ ψ − mψψ = LDirac(ψ.85) It follows from this that under a (ﬁnite) gauge transformation.83) (or (9.84) where ǫ(x) denotes the inﬁnitesimal local parameter of transformation. the minimally coupled Dirac Lagrangian density transforms as ′ LDirac (ψ. let us note that the covariant derivative in (9. δǫ Aµ (x) = A′ (x) − Aµ (x) = µ ′ 1 ∂µ ǫ(x). ψ . Therefore. First. e (9. we have already seen that the Lagrangian density for the Maxwell theory is invariant under the gauge transformation in (9.83) (with θ(x) = ǫ(x) =inﬁnitesimal and keeping terms to linear order in ǫ) δǫ ψ(x) = ψ ′ (x) − ψ(x) = −iǫ(x)ψ(x).84)). In fact. ψ. δǫ ψ(x) = ψ (x) − ψ(x) = iψ(x)ǫ(x).

In addition. there is a conserved current.80) r µβ α q p = −(2π)4 ie (γ µ )βα δ4 (p+q +r).5 Quantum electrodynamics 349 Namely.107).83) and.88) The Feynman rules for QED can be derived in the standard manner.104) and in (8.91) . In this theory.µν (k) = k.64) so that the external line in (9. (9. The form of the propagator in (9. the minimally coupled Dirac Lagrangian density is also invariant under the gauge transformation (9. therefore.90) can represent both the annihilation as well as the creation of a photon.89) lim − 1 ˆ ˆ i(ηµν − ηµ ην + kµ kν ) .9.87) In momentum space the current conservation takes the form kµ J µ (k) = 0.106)-(8. ∂µ J µ (x) = 0. J µ (x) = ψγ µ ψ.77)) correponds to the temporal gauge (9.90) (2π)3 2k0 where we have chosen a real polarization vector as in (9. (9. the complete QED Lagrangian density (9.89) (or (9. in QED we have the photon propagator as well as the photon external line factor which take the forms. λ µ = ǫ→0+ (9. (9. The interaction vertex for QED can also be read out from the Lagrangian density (9. µ k ν = iGF . λ). We have already derived the fermion propagator as well as the fermion external line factors in (8. We note here that the photon propagator is a gauge dependent quantity.80) is gauge invariant.27)). k2 + iǫ ǫµ (k. (9.23) (see also (9.

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

(9. λ′ ) +v (k2 . depending on whether we are interested in summing over all spin states of the fermions or not. (9. / ˜ − m + iǫ K / (9. λ) ǫν k4 . λ) / 1 ǫ (k3 . s1 ) / K − m + iǫ / 1 ǫ k4 . Furthermore. The diﬀerential cross-section will be proportional to the absolute square of the matrix element. s2 ) ǫ (k4 . s1 ) . λ′ )|S (2) |e− (k1 ). s2 ) ǫ (k3 . λ′ u (k1 . we have deﬁned ˜ K = k1 − k4 . λ)u (k1 . the γ-algebra can be simpliﬁed and a simple formula for the diﬀerential cross-section can be obtained (as discussed in section 3.4). K = k1 − k3 . e+ (k2 ) = 1 1 4 (2π)6 (2π) 1 1 0 2k3 0 2k4 m 0 k1 m 0 k2 ×ǫµ (k3 . λ′ uα (k1 .94) and the limit ǫ → 0+ is understood. 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 . γ(k4 . for simplicity.δγ (K)(−(2π)4 ie) (γ µ )γα ˜ −(2π)4 ie (γ µ )βδ iSF .9.93) where. λ).6 Physical processes 351 γ(k3 .95) .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 . s1 ) v β (k2 .

352 9 Maxwell field theory are given to the lowest order by the same diagrams as in Fig.94) K = k1 − k4 and we have introduced Q = k1 + k2 .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 (9.93) (or (9. Therefore. (9. s1 ) / Q − m + iǫ / 1 ǫ (k4 .96) / ˜ − m + iǫ K / ˜ where as deﬁned in (9.2 except for the labeling of the momenta.93) to be e− (k3 ). the internal line represents a photon propagator signifying that the electrons scatter by exchanging (emitting and absorbing) a photon.98) To the lowest order. λ)u(k1 . we can obtain the S-matrix element for Compton scattering directly from the results in (9. γ(k4 )) to be outgoing.97) c) M¨ller scattering: M¨ller scattering is the study of elastic scato o tering of two electrons e− (k1 ) + e− (k2 ) → e− (k3 ) + e− (k4 ). γ(k4 . λ′ )u (k1 . this process is described by the two Feynman diagrams shown in Fig. (9. γ(k2 . λ) / 1 ǫ (k2 . 9. λ) = − ie2 (2π)2 1 0 2k2 1 0 2k4 m 0 k1 m 0 k3 / × u(k3 . but . s1 ) .3.93). λ′ ) +u(k3 . In this case. 9. this amplitude will differ from (9. Thus. namely. s3 )ǫ (k4 . λ′ )|S (2) |e− (k1 ). s3 )ǫ (k2 . we will assume the ﬁnal state electron and photon (e− (k3 ). (9.90) is the same for the creation or the annihilation of a photon). instead of an electron and a positron being annihilated as was the case in (9. Here. The amplitude diﬀers from the case of the pair production of photons by only the external line factors.

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

s1 ).4: The two lowest order Feynman diagrams for Bhabha scattering. e+ (k4 . but in the present case we will treat the ﬁnal state electron and the positron as outgoing (e− (k3 ). s2 ) = ie2 (2π)2 m 0 k1 m 0 k2 m 0 k3 m 0 k4 × u(k3 . the ﬁrst diagram in Fig. the second diagram simply describes the scattering of an electron and a positron by exchanging a photon.4 represents the annihilation of an electron and a positron to create a photon which subsequently creates an electron and a positron. s1 ) 2 + iǫ Q ˆ ˆ (ηµν − ηµ ην + Kµ Kν ) u(k3 . 9. We see that topologically these diagrams are the same as in Fig. s4 ) ˆ ˆ (ηµν − ηµ ην + Qµ Qν ) v(k2 . 9. s1 ) . we have e− (k3 .4. s4 ) +v(k2 . s3 )γ µ v(k4 . s2 )γ µ v(k4 . e+ (k4 )). s3 )γ ν u(k1 .100) k4 + k1 + k2 k1 k3 k1 k1 − k3 k2 Figure 9. s2 )γ ν u(k1 . In this case. s4 )|S (2) |e− (k1 .354 9 Maxwell field theory e− (k1 ) + e+ (k2 ) → e− (k3 ) + e+ (k4 ).101) . The lowest order Feynman diagrams for this process are shown in Fig.3. 2 + iǫ K (9. 9. s3 ). On the other hand. k2 k4 k3 (9. We can calculate this matrix element as before and omitting the overall delta function representing conservation of energy-momentum. e+ (k2 .

In simple language this identity implies that there exists a relation between the fermion two point function (self-energy) and the vertex function in QED and the relation can be traced to the gauge invariance of the theory. the Lagrangian density (9. e (9. depends on the choice of gauge.104) .97) K = k1 − k3 and Q = k1 + k2 . This gives a ﬂavor of low order calculations in QED. let us note that the fermion propagator in QED can be represented as i = p−m / iSF (p) = p .7 Ward-Takahashi identity in QED 355 where as deﬁned in (9.7 Ward-Takahashi identity in QED Before we discuss more complicated symmetries. 4 (9. As we have seen. (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.80) for QED has the form 1 ¯ ¯ L = − Fµν F µν + iψγ µ (∂µ + ieAµ ) ψ − mψψ. 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. Here we would only give a very simple derivation of this relation in QED. It follows from (9. −p. 9. 0) (iSF (p)) . of course. let us discuss very brieﬂy the Ward-Takahashi identity in QED.103) where the i factor in the denominator is understood. which we will study later. independent of the choice of gauge.9. However.94) and (9.102) The structure of the photon propagator in the theory.

108) . by inverting (9. As a result the right-hand side of (9.104) we can obtain a relation between the fermion two point function and the three point vertex function as 1 Γµ (p.104) in QED resulting from the structure of the fermion propagator has the form ∂ ∂pµ p =− 1 e p k=0 p . k = 0).91) in QED with a zero momentum photon (without the delta function and the factor of (2π)4 ). −p. −p. µ F ∂p e (9.104) can be diagrammatically represented as i i iγµ p−m / p−m / 1 e = − p k=0 p (9. a diagrammatic representation for the basic identity (9. (9.356 9 Maxwell field theory where Γµ (p. −p. 0) denotes the three point vertex (9. ∂pµ F (9.106) In this case.107) ∂ (iSF (p)) (iSF (p))−1 ∂pµ = −i which can also be written as i 1 ∂ −1 S (p) = − Γµ (p. k = 0)(iSF (p)).105) 1 = − (iSF (p))Γµ (p. k = 0) = −(iSF (p))−1 e = ∂ (iSF (p))−1 ∂pµ ∂ −1 S (p). −p. e Consequently.

(The trick in obtaining this relation in a simple manner lies in channeling the external momentum only through the fermion lines. we see that (9.) At two loop order the fermion self energy diagrams have the forms shown in Fig. let us look at the fermion self-energy at one loop. Using the graphical identity in (9.106). such relations are called Slavnov-Taylor identities which we will discuss later). Once again. .108) holds for the one loop corrections to the amplitudes. without the external lines).109) represents the one loop correction to the fermion two point function or the self-energy (multiplied by a factor of i) whereas the right-hand side is the correction to the three point vertex function at one loop.9. For example. Such a relation can be seen to hold at any higher order in perturbation theory as well.109) Here we are looking only at the proper vertex parts of the diagrams (namely.5: Two loop corrections to the fermion self-energy in QED.5 (there are no external lines and we do not label the momenta for simplicity. we can write k ∂ ∂pµ p k+p p =− 1 e p k p k+p k+p q=0 (9. The diagram on the left-hand side in (9. but we understand that the external momentum is channelled only through the fermion lines) + Figure 9. 9.7 Ward-Takahashi identity in QED 357 This relates the three point vertex function to the fermion two point function (self-energy) in QED at the tree level and such relations are known as Ward-Takahashi identities (in non-Abelian gauge theories.

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

B.9.108) is a consequence of gauge invariance can be seen heuristically in the following way.85)) iaψD = iaψ(∂ + ieA /ψ / /)ψ.111) where Z1 .108) we conclude that these constants must be related as Z1 = Z2 .83). 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 + · · · .7 Ward-Takahashi identity in QED 359 −1 −1 SF (p. 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). then from (9. m = 0). That the Ward-Takahashi identity (9.112) This physically implies that if there are divergent parts in the fermion self-energy graphs at higher order. on the other hand. Γµ −→ Z1 Γµ . Thus comparing with (9.114) . m = 0) −→ Z2 SF (p. requires that the kinetic energy part of the fermion and the photon interaction term must combine into the form of a covariant derivative (see (9. we can identify a = 1 + B = 1 + C. Invariance under the local gauge transformations (9. Z2 are constants.113) −eCψA − Dψψ + · · · /ψ where A.115) (9. (9. (9.113). (9. C. D are constants and the dots denote other structures that may be induced due to quantum corrections. (9.

9.8 Covariant quantization of the Maxwell theory As we have seen.21) and (9. the Lagrangian density (9. (9. We recall that the Lagrangian density for Maxwell’s theory is given by 1 L = − Fµν F µν .15). The canonical momenta conjugate to the ﬁeld variables Aµ can be calculated from the Lagrangian density and take the forms (see (9.116) We will discuss these topics in more detail when we study renormalization chapter 16.118) As we have already seen in (9. then the equations of motion (see (9.117) where Fµν denotes the ﬁeld strength tensor. As a result.117) as well as the action for this theory are invariant under the gauge transformation δAµ = ∂µ α(x).18)) .360 which is another way of saying 9 Maxwell field theory Z2 = 1 + B = 1 + C = Z1 . since the ﬁeld strength tensor Fµν does not change under a shift of the vector potential by a gradient. ˙ ∂ Aµ Πµ = which implies the constraint Π0 = −F 00 = 0. canonical quantization of Maxwell’s theory results in a lack of manifest Lorentz invariance. 4 Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . (9. Let us recapitulate brieﬂy how this arises. (9. the present choice of gauge is known as the Coulomb gauge). 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 = 0.22)) ∂L = −F 0µ .

lead to (for ν = 0) ∂µ F µ0 = ∂i F i0 = 0. would lead to ∂µ F µ0 = J 0 .119) On the other hand. or. or. A0 = − 1 J0 .8 Covariant quantization of the Maxwell theory 361 ∂µ F µν = 0. ∇2 A0 = −J0 . or. ∇2 (9. ∇2 A0 = 0. in the Coulomb gauge ∇ · A = 0. however. the dynamical equations ∂µ F µν = J ν . if sources (charges and currents) are present. 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. The canonical quantization can now be carried out for the two physical degrees of freedom. A0 = 0. or. But. but we lose manifest Lorentz covariance in the process (because all components of Aµ are not being treated on a equal footing). . Let us emphasize here that the ﬁnal result for the calculation of any physical amplitude. ∂i ∂ i A0 − ∂ 0 Ai = 0.120) In either case. We recognize that this is a consequence of our choice of gauge. or. we lose manifest Lorentz covariance in the intermediate steps (which is highly desirable in any calculation). (9.9. remains manifestly Lorentz invariant in the canonical formalism.

117) for the Maxwell theory also as 1 1 L = − Fµν F µν = Aµ P µν Aν + total derivatives. the inverse of P µν does not exist (this also means that the determinant of P µν vanishes) and the Green’s function and. (9. P = 1 P µν deﬁnes the normalized projection operator. therefore.124) so that this is the transverse projection operator. (This is relevant in the discussion within the context of a path integral quantization of the system which we will discuss brieﬂy in chapter 12.122) (9.) Furthermore. − ∂µ∂λ (9. namely. (In fact. This implies that even if we can quantize the theory (say. the Feynman propagator for the theory cannot be deﬁned.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. we see that P µν can be thought of as µν a projection operator.123) With a suitable normalization.) We note that we can write the Lagrangian density (9. . 4 2 where P µν = η µν − ∂µ∂ν . we cannot carry out calculations in perturbation theory. we note that ∂µ P µν = ∂µ (η µν = (∂ ν − − ∂µ∂ν ) ∂ ν ) = 0 = ∂ν P µν . Consequently. (9. it projects on to the space of the components of any vector transverse (perpendicular) to the gradient operator ∂ µ . in the naive path integral formalism).

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

∂µ F µν + ∂ ν (∂ · A) = −J ν . therefore. ∂∂µ Aν ∂Aν −∂µ F µν − ∂ ν (∂ · A) − J ν = 0.129) ∂ · A = −∂ · J = 0. or. If we now write out the left-hand side explicitly. the presence of this additional term in the Lagrangian density would not change the physics of Maxwell’s theory.127). Furthermore. (∂ · A) = 0. (9. or. this is just the Maxwell’s equations in the presence of conserved sources.130) where we have used the anti-symmetry of the ﬁeld strength tensor as well as the conservation of J µ .127) Without the second term on the left-hand side in (9. (9. (9. ∂t χ = 0.128) (9. Thus we see that although the presence of the term − 1 χ2 in the 2 Lagrangian density (9. (9.127) takes the form ∂µ (∂ µ Aν − ∂ ν Aµ ) + ∂ ν (∂ · A) = −J ν .127). χ is a free ﬁeld and.364 9 Maxwell field theory ∂µ or.132) . (9.125) where χ = ∂ · A. Aν = −J ν . An alternate way to see this is to note that if we take the divergence of the equations of motion in (9.131) seems to modify the theory. we recognize that if we restrict classically to the initial value conditions ∂χ = 0. or. (∂ · A) = ∂ν J ν . we have ∂ν ∂µ F µν + or. ∂L ∂L − = 0. t = 0. (9.

125) with the supplementary condition ∂ · A = 0.133) (9. (9. (9. Aν (y) x0 =y 0 = −iηµν δ3 (x − y). ˙ ∂ Aµ Πµ = (9.9.40) 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).136) with all other commutators vanishing. ˙ Aµ (x). or. φ(y) = iδ3 (x − y). classically we can think of Maxwell’s theory as described by the modiﬁed Lagrangian density (9.8 Covariant quantization of the Maxwell theory 365 then χ = 0 at all times and we recover the familiar Maxwell’s theory.137) x0 =y 0 .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). Thus. 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.134) It is now obvious that the coeﬃcient matrix of highest derivatives is nonsingular in this case. Πν (y)]x0 =y0 = iδµ δ3 (x − y). We now see that this quantization relation is exactly like the quantization condition for four distinct scalar ﬁelds (see (5. 2 (9. We can deﬁne the canonical momenta as ∂L ˙ = −Aµ .

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

) As we would see shortly. (9. such a condition is hard to implement as an operator condition.136) are consistent. however. Πν (x′ ) ∂µ Aµ (x). A little bit of thinking tells us that although in the classical theory we can impose the condition ∂µ Aµ = 0. but the derivation needs to be done carefully in a limiting manner x0 → x′0 . (Note that since all ﬁeld components are dynamical in this modiﬁed theory. the naive “Hilbert” space is much larger than the physical Hilbert space of the Maxwell theory consisting of only the transverse photon degrees of freedom.9. namely. the non vanishing right-hand side would lead to an inconsistency.) We note that. we can weaken the supplementary condition (9. = i∂ ν δ3 (x − x′ ). Πν (x′ ) x0 =x′0 x0 =x′0 (This relation is essentially correct.136) would lead to Aµ (x). = i∂ ν δ3 (x − x′ ). where Aµ ’s are operators. Thus. or. ∂µ Aµ |phys = 0.133) and say that this condition is true only on the space of physical states of the theory.140) in the quantum theory.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 (9.133) as yet. We . even (9. = iη µν δ3 (x − x′ ). whereas the left-hand side of the above expression would be zero if the supplementary condition (9. It is easy to see that the commutation relation (9.133) were to hold as an operator equation. (9. (9. we can think of the supplementary condition as selecting out the subspace of the physical Hilbert space of the theory.142) is a very stringent condition and we have to relax this further. We have. The supplementary condition (9.142) In other words.141) x0 =x′0 ∂µ Aµ (x). Πν (x′ ) or.133) may be thought of as picking out the smaller physical subspace from the larger total vector space. not made use of the supplementary condition (9.

we note that if |ψ represents a physical state. λ) and a† (k. (9. 3. 1. −1. (9. 2. k0 = ω(k) = |k|. . we can always choose a basis such that the polarization vector for λ = 0 is time-like (recall η µ ). then ψ| ∂ · A(x). a† (k′ . λ = λ′ = 0. λ)fk (x)a† (k. λ′ ) = 0 = ′ a† (k. λ). λ) (9.141).74)). At this point we can quantize the coeﬃcients a(k. here they are not required to be transverse. λ)’s are the components of the polarization vector for a photon travelling along k and polarization index λ = 0.144) where fk (x) = 1 (2π)3 2ω e−ik·x . λ). λ′ ) a(k. let us make a plane wave expansion for the ﬁeld variables Aµ (x) = λ ∗ + ǫ∗ (k. µ d3 k ǫµ (k.) They are normalized as 1.368 9 Maxwell field theory can see this already from the commutator relation (9. 3. λ) as annihilation and creation operators satisfying a(k. λ)ǫ∗µ (k.146) ǫµ (k.147) = −η λλ δ3 (k − k′ ). (9. x0 =x′ 0 = ψ|i∂ ν δ3 (x − x′ )|ψ .145) Here ǫµ (k. 2. (Unlike the earlier discussion. a(k′ .143) which is inconsistent. λ). Basically. λ′ ) . ˆ while that for λ = 3 is longitudinal (recall kµ ) and the other two polarization vectors are transverse to the momentum four vector. λ = λ′ = i = 1. Πν (x′ ) |ψ or. a† (k′ . To proceed further. λ) . 0 = i∂ ν δ3 (x − x′ ). λ′ ) = η λλ = ′ As we have seen earlier (see (9. (9. λ)fk (x)a(k.

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

it is strange to note that there are four independent photon degrees of freedom. 0)a(k. 2)a(k. 3)a(k. 1. 2. However. The norm of this one photon state can now be calculated (9. λ) behave respectively like annihilation and creation operators for λ = 1. 1) +a† (k. it has the form :H: = d3 k ω(k) −a† (k. λ = 0)|0 . 3) . λ)|0 = 0.154) .370 9 Maxwell field theory Thus the commutation relations for the creation and the annihilation operators are indeed consistent with the quantization condition for the ﬁeld variables. 3. If we calculate the normal ordered Hamiltonian of the theory (recall J µ = 0). (9. 2) + a† (k.147). 0) + a† (k.152) Let us next consider the state with one time-like photon |1. We also note from (9.153) where F (k) is a suitable smearing function such that d3 k |F (k)|2 < ∞. (9. λ = 0 = d3 k F (k)a† (k. a(k. (9.147) that although a(k. 1)a(k.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. 2. For example. for the time-like component there is a relative negative sign in the commutation relation (9. λ) and a† (k. λ = 0. 3. let us denote by |0 the vacuum state of the theory so that 0|0 = 1.

3) .155) This shows that the vector space of the theory has an indeﬁnite metric – the norm for the one time-like photon state is negative although states containing only space-like photons have positive norm. 0)|0 .8 Covariant quantization of the Maxwell theory 371 1.158) and the energy of this one photon state is given by .157) In this case. 0)a† (k.156) |k.151)) :H: = d3 k ω(k) −a(k. We also realize that this negative norm is a consequence of the negative sign in the commutator (9. 0 = a(k.9. 0)|0 d3 kd3 k′ F ∗ (k)F (k′ ) 0|a† (k′ . 0)|0 = 0. Thus the natural modiﬁcation that comes to mind is to interchange the roles of these operators for the time-like photon. 0)a† (k′ . 0) + a† (k. λ = 0 = = = − d3 kd3 k′ F ∗ (k)F (k′ ) 0|a(k. (9. this also leads to trouble as we see below. 2) + a† (k. 2)a(k. (9.147) for the annihilation and the creation operator for a time-like photon. 1)a(k. However. 0) − δ3 (k − k′ )|0 d3 k |F (k)|2 < 0. 1) +a† (k. 0)a(k. so that the one time-like photon state is given by (9. 3)a(k. the normal ordered Hamiltonian will have the form (compare with (9. (9. Let us suppose that a† (k. λ = 0|1.

physical results are well behaved.372 9 Maxwell field theory H|k. So we are stuck with an indeﬁnite metric space (with the standard interpretation for annihilation and creation operators for the timelike photon) and the normalization of states in this space is given by (we suppress the momentum label for simplicity) n0 . n1 . This not only demands that certain . 0)]|0 (9. n2 . m2 . deﬁnite states of arbitrarily high negative energy values are possible and the Hamiltonian becomes unbounded from below. mλ denote number of photons for each of the polarizations λ = 0. the energy of the state with one time-like photon becomes negative. Thus states containing an odd number of time-like photons have negative norm. In fact it is obvious that with this deﬁnition. 0)|0 = [H. 1. The question that immediately comes to our mind is what happens to the probabilistic interpretation of the theory. m1 . 0)a† (k′ .161) However.159) = −ω(k)a(k. the physical states must satisfy ∂µ Aµ (x)|phys = 0. n3 |m0 . This is seen by using the supplementary condition (subsidiary condition) which we still have not imposed. m3 = (−1)n0 δn0 m0 δn1 m1 δn2 m2 δn3 m3 . Thus we see that if we exchange the roles of the creation and the annihilation operators for the time-like photon. 0 = Ha(k. 3.160) where nλ . as discussed earlier this is too stringent a condition to allow any state of the radiation ﬁeld. where we have used the fact that H|0 = 0. In a free theory we may impose suitable conditions to prohibit any unwanted state. a(k. Namely. 0 . a(k. As we will see shortly in spite of the presence of negative norm states. 2. (9. But in the presence of interactions such states may be excited. 0)|0 = −ω(k)|k. (9. 0). 0)]|0 = d3 k′ ω(k′ )[−a(k′ .

2. (9. the supplementary condition (9.163) ∂µ Aµ (x) is like a free scalar ﬁeld. Gupta and Bleuler weakened the supplementary condition to have the form ∂µ Aµ (+) (x)|phys = 0. λ or. λ) = 0 for λ = 1. Recalling that kµ ǫµ (k. (This is commonly known as the Gupta-Bleuler quantization. or. 3)a(k. 0)a(k. kµ ǫµ (k. 0) + ǫµ (k. λ = 0) − k2 a(k.165) (a(k. 0) − a(k. or.) We remark here that since (see (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. |k| (9. 3) |phys = 0. k0 a(k.9.129) or (9. Therefore. Let Vphys = {P } denote the set of all states which satisfy the supplementary condition and let |ψ represent such a state. 3))|phys = 0. λ)a(k. as we have seen.165) implies that .130)) ∂µ Aµ (x) = 0. where we have used the deﬁnitions and the properties of the polarization vectors in (9.164) the Gupta-Bleuler supplementary condition leads to kµ ǫµ (k.74) as well as the fact that for positive energy photons k0 = ω = |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. λ)|phys = 0. (9. λ = 3) |phys = 0. (9. Then. but it also requires that those photons cannot be emitted.

0) − a(k. 3)) |n0 . (9. n1 .n2 . 3)|ψ . n1 . of course.n3 Cn0 . All the coeﬃcients Cn0 . 3)|ψ . n3 √ − n3 |n0 . (9.n2 .n2 . n3 = 0. 0)|ψ = a(k.n1 .n3 ) ×|n0 − 1. n3 .n1 .n1 . 3)a(k.n1 .n2 . ψ|a† (k.n2 .170) .169) which leads to the relation in (9.167) 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 . or. n1 . n3 − 1 = 0.168). n2 .n2 .n2 . a superposition of diﬀerent states of the form |ψ = n0 .n3 = 0.n2 . (9.n1 .n3 − n0 |n0 − 1. √ Cn0 .n3 (a(k. 0)|ψ = ψ|a† (k.168) This can be seen as follows or.n3 |n0 . n2 . n1 .374 9 Maxwell field theory a(k.n3 in the expansion can be determined recursively from (9.n1 . A general physical state is. n2 . n3 − 1 = 0.n3 −1 + n3 Cn0 −1. n2 . (9.n3 −1 + n3 Cn0 −1. √ √ ( n0 Cn0 . (9. n1 .n1 . or.n2 .n1 .n2 . n2 .n1 .166) This leads to the fact that the physical states must contain superpositions of states with an equal number of time-like and longitudinal photons. 0)a(k.0 = 1. Cn0 .168) with the condition C0.

n2 . n2 . .n3 ’s determined recursively. n2 . n2 . n1 . n1 . n1 . For example. = φ0 |φ0 . (9. 0 . 2|0. = 0. n1 .175) . the physical states allowed by the supplementary condition have the forms |φ0 |φ1 |φ3 = |0. n1 . n1 . n1 . 1 + 2. n1 . n2 . 1|1. n2 of transverse photons as a linear superposition of states of the form |φ because φ|φ = |φ0 + b1 |φ1 + b2 |φ2 + .9. n0 = n3 = 0. 0|1.8 Covariant quantization of the Maxwell theory 375 It follows now that for a ﬁxed number n1 .171) and so on where we have used the values of Cn0 . (9. In other words. 0 . Let us note that because of the negative norm of states containing an odd number of time-like photons. n1 .174) we can assume that in a truly physical state of the radiation ﬁeld there are no time-like and longitudinal photons present. 1|0.n1 . n2 . we have φ1 |φ1 φ2 |φ2 = 0. n2 . n2 . n2 of transverse photons. n2 .172) and so on. n2 . (9. Thus although we can write a general physical state with a ﬁxed number n1 . 0 = 1 − 2 + 1 = 0. 1 + |2. 2 − 2|1. 1 − |1. n2 . n1 . n1 . n1 . all such states except for |φ0 have zero norm.n2 . n2 . n2 . = |0. . 0 = 1 − 1 = 0.173) (9. . n2 . n1 . 2 + 2 1. 1 + 1. √ = |0. n1 . n1 . 0 . n1 . 0|2. n2 . (9. n2 .

376 9 Maxwell field theory in such states. Without going into details we note here that the Feynman propagator for the photon in this theory has the simpler form iGF . We also point out here that the Gupta-Bleuler supplementary condition can be thought of as the quantum analog of the covariant Lorentz condition ∂µ Aµ (x) = 0. + iǫ (9. namely. V0 V phys = (9. there is no problem with a probabilistic interpretation). consequently. if we further mod out the states by the zero norm states. A diﬀerent way of saying this is to note that being orthogonal to every other state.165) contains states with positive semi-deﬁnite norm (negative norm states are eliminated by the supplementary condition or the physical state condition and. we have the true physical subspace of the theory where the norm of states is positive deﬁnite.176) where |ψ represents a physical state. The physical subspace of the theory selected by the supplementary condition (9. The physical S-matrix elements.µν (k) = lim − ǫ→0+ k2 iηµν .179) This is diﬀerent from (9. However their eﬀect cancels out in the ﬁnal result. Vphys . however. such states decouple.178) where V0 represents the set of states with zero norm.177) (9. on the space of physical states since ψ|∂µ Aµ |ψ = 0. If interactions are present time-like and longitudinal states may occur as intermediate states.89) simply because this can be thought of as quantizing the theory in a diﬀerent gauge (FeynmanFermi gauge). 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. are gauge .77) or (9. (9. Since the zero norm states are orthogonal to all the states including themselves.

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

This page intentionally left blank .

relates momenta to velocities (and coordinates) of the theory and if this relation is invertible. N . · · · . rather some of these variables have to satisfy constraints following from the structure of the theory. ∂ qi ˙ (10. thereby. In this case.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. let us consider a classical system of point particles described by the Lagrangian L(qi . qi ).Chapter 10 Dirac method for constrained systems 10. the dynamical phase space variables of these theories are not all independent. Given the Lagrangian of the theory. In this case. To appreciate the diﬃculties associated with constrained systems. velocities can be expressed in terms of momenta (and coordinates).1) Equation (10. we can go from the conﬁguration space of the system to the phase space 379 . carry out the quantization of a theory) and in this chapter we will discuss the Dirac method in some detail. Such systems are known as constrained systems and the naive passage to the Hamiltonian description for such a system starting from the Lagrangian description fails. we deﬁne the momenta canonically conjugate to the coordinates as pi = ∂L . 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. 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. 2. in general. i = 1.1).

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

8) then. (10. H}.1 Constrained systems 381 ˙ A(qi .9) However. (10. β {θα . we can go from the conﬁguration space to the phase space ˙ (θα .5).10) through the Legendre transformation (this is consistent with the convention of left derivatives) ˙ ˙ ˙ ˙ H(θα . we have to deﬁne the derivative in (10. · · · M with θα θβ = −θβ θα .10. θα ). Πβ } = −δα = {Πβ . pi ) = {A(qi . unlike the bosonic case in (10.12) The Poisson bracket between any two dynamical variables can now be obtained from (10. we can deﬁne the momenta conjugate to the coordinates as Πα = ∂L . we will choose the convention of taking derivatives from the left.7) This discussion can also be generalized to a classical system described by Grassmann (anti-commuting) variables. If the relations (10.9) and as discussed in chapter 8 (see (8. (10. if ˙ we have a classical system described by the Lagrangian L(θα . (10. For example. since the dynamical variables are now of Grassmann odd (fermionic) nature. Πα ) = θα Πα − L(θα . However. ˙ ∂ θα (10.11) The 2M dimensional phase space is endowed with a canonical Poisson bracket structure (with the convention of left derivatives) {θα . (10. θα ). θβ } = 0 = {Πα . θα ) = −Πα θα − L(θα . Πβ }.14)). pi ). . θα }. 2.9) can be inverted to express velocities in terms of the momenta. θα ) → (θα . α = 1.12). Πα ).

H} = − α . for example. {F1 . . Πβ } β − {Π .10) is invertible so that all the velocities can be expressed uniquely in terms of independent momenta leading to the . Πα = {Πα . B2 (θα .15) which can be compared with (10. θβ } α ∂θα ∂Π ∂Π ∂θβ ∂B1 ∂B2 ∂B1 ∂B2 + ∂θα ∂Πα ∂Πα ∂θα . Thus.6).14) the 2M ﬁrst order dynamical equations for the system can now be written in the Hamiltonian form as ∂H ˙ θα = {θα . B2 } = − = ∂B ∂B where we have used (10.13) {B1 . In a similar manner. Πα ) takes the form ∂B2 ∂B1 α ∂B2 ∂B1 {θα . the Poisson brackets between a Grassmann even and a Grassmann odd variable as well as the bracket between two Grassmann odd variables can be obtained to correspond to {F. Πα ).382 10 Dirac method for constrained systems the structure of the Poisson bracket now depends on whether the dynamical variables are Grassmann even (bosonic) or Grassmann odd (fermionic). (10.2) or (10. The discussion so far assumes that the transformation to phase space in (10.14) Using (10.12) as well as the fact that ∂θα and ∂Πα are Grassmann odd (fermionic) variables. F2 } = − ∂F1 ∂F2 ∂F1 ∂F2 + α ∂θα ∂Π ∂Πα ∂θα . H} = − ∂θα (10. ∂Π ∂H ˙ . the Poisson bracket between two bosonic dynamical variables B1 (θα . (10. B} = − {B. F } = ∂F ∂B ∂F ∂B + α ∂θα ∂Π ∂Πα ∂θα ∂B ∂F ∂B ∂F + ∂θα ∂Πα ∂Πα ∂θα . with the convention of left derivatives.

2) can be written in general in the matrix form as q p q q ˙ =A = ½ 0 ˜ b b q q ˙ . it follows that b = b(q). the diﬃculty in passage to a Hamiltonian description arises when this transformation is noninvertible. In this case. ˜=˜ b b(q).17) The inverse of the matrix A in (10.16) can be easily seen to have the form A−1 = ½ 0 −b−1˜ b−1 b . (10. it follows that the transformation is invertible (A−1 exists) if ∂2L = 0. b ˙ so that ∂pi ∂2L = .21) . the transformation to phase space in (10. (10.19) bij (q) = (10.18) so that A−1 exists only if the matrix b is invertible.10. From (10. (10. ∂ qj ˙ ∂ qj ∂ qi ˙ ˙ (10. ∂ qi ∂ qj ˙ ˙ det (10. However.16) where the elements b. ˜ denote N × N matrices.20) Therefore.16) we note that pi = ˜ij (q)qj + bij (q)qj . To understand the diﬃculty.1 Constrained systems 383 unique Hamiltonian of the system. For conventional b theories described by Lagrangians with quadratic velocity terms. let us concentrate on the bosonic theory for simplicity.

2.2 Dirac method and Dirac bracket In a physical system with constraints. As a result.2) is not invertible. b = 1. ∂ qi ∂ qj ˙ ˙ (10. may not represent the true brackets necessary for a Hamiltonian description of the system. we can determine only R of the N velocities in terms of coordinates and independent momenta as qa = fa (qi . there exist relations or constraints between various dynamical variables and such systems are known as constrained systems. In other words.4) need not satisfy the constraints of the theory and. pb ). the transformation (10. therefore. In this case. therefore. In this case. we can write . if the Lagrangian describing the theory satisﬁes ∂2L = 0.20) 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. ˙ a.24) and the other N − R velocities cannot be determined. Furthermore. it is not clear a priori how to deﬁne the Hamiltonian uniquely. Let us assume that the N × N matrix ∂2L . R. (10. · · · . ∂ qi ∂ qj ˙ ˙ det (10.22) then. the passage to a Hamiltonian description is achieved systematically through the method of Dirac which we describe in this section.23) of the physical system is of rank R < N . it is also clear that the naive canonical Poisson brackets in (10. On the other hand. 10. It follows from (10.384 10 Dirac method for constrained systems The naive passage to the Hamiltonian description can be carried out only in this case.

the canonical Hamiltonian does not depend on velocities as we would expect. we can deﬁne the canonical Hamiltonian of the theory as the Legendre transform Hcan = pi qi − L(qi . ˙ ˙ ˙ ˙ ˙ (10.28) that in spite of the fact that qα cannot be determined.26) and we recognize that such constraints will reduce the dimensionality of the true phase space of the system. namely. Hcan = Hcan (qi .25) It is clear that the second set of equations in (10. 2. For example.2 Dirac method and Dirac bracket 385 pa = ga (qi . pa ). if they are equal on the constrained hypersurface Γc . (If there are further constraints in the theory. qi ). namely. a. (10. pa ) = 0. qi ) = pa qa + gα qα − L(qi . from the canonical .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.25) deﬁne N − R constraints among dynamical variables which we denote by ϕα = pα − gα (qi .29) so that on the constrained hypersurface Γc . · · · .) Two dynamical variables F. the dimensionality of Γc would be further reduced. the Hamiltonian of the theory is no longer unique. · · · . In fact. qb ). ∂ qα ˙ ∂ qα ˙ (10.27) Γc As usual.10.2) (presence of constraints). (10.25). α = R + 1. F ≈ G.28) where we have used (10. b = 1. we note that the constraints in (10. We note from (10. (10. R. if (F − G) = 0. R + 2. pa ). ˙ ∂L ∂Hcan = gα − = gα − pα ≈ 0. G in Γ are said to be weakly equal. we also note that because of the non-invertibility of (10. ˙ pα = gα (qi . On the other hand. N.

Hp } = − ˙ and using (10. Hp } = {F.29). ∂pi ∂pi ∂Hp ∂(Hcan + λα ϕα ) =− .31) Furthermore.26) as (it is known as the primary Hamiltonian since it incorporates only the primary constraints in (10.32) qi ≈ {qi .35) . The time evolution of any dynamical variable F can now be written as ˙ F ≈ {F.30) with velocities λα ≈ qα . (10. ˙ (10.33) which remain undetermined.26)) Hp = Hcan + λα ϕα . λβ }ϕβ ≈ {ϕα . it follows that we can identify the Lagrange multipliers in (10. Hp } = ˙ pi ≈ {pi . ϕβ } ≈ 0. with the canonical Poisson brackets in (10.34) The constraints of the theory should be invariant under time evolution and using (10.34) this leads to ϕα ≈ {ϕα .386 10 Dirac method for constrained systems Hamiltonian of the theory. Hcan } + λβ {ϕα . Hcan } + λβ {ϕα . ∂qi ∂qi (10.4). Hcan + λα ϕα }. (10. we can write the Hamiltonian equations as ∂Hp ∂(Hcan + λα ϕα ) = .30) where λα denote undetermined Lagrange multipliers and it follows that Hp ≈ Hcan . Hcan + λβ ϕβ } ˙ = {ϕα . we can deﬁne the primary Hamiltonian associated with the system by incorporating the primary constraints of the theory (10. ϕβ } + {ϕα . (10. (10.

2n2 .27)) Poisson bracket with all the constraints (including themselves) are known as ﬁrst class constraints and are denoted as e ψ α ≈ 0. 1.10. · · · . n < 2N. (10. It is clear that (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. α = 1. 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. In general. . ) collectively as ϕα ≈ 0. 2. · · · . secondary. On the other hand.37) 2. Requiring the secondary constraints to be invariant under time evolution as in (10. (10. 2. · · · . . It is also worth noting that in this process of determining all the constraints that are time independent (evolution free).36)-(10.36) which can be divided into two distinct classes. α = 1. We continue with this process until all the constraints are determined to be evolution free. all of the Lagrange multipliers may be completely determined or some of them may remain undetermined. tertiary.35) may determine some of the Lagrange multipliers or may lead to new constraints known as secondary constraints. (10. The constraints which have weakly vanishing (see (10. α = 1.2 Dirac method and Dirac bracket 387 where we have used (10.38) that n1 + 2n2 = n < 2N .35) (with Hp as the Hamiltonian) may determine some other Lagrange multipliers or generate newer constraints (tertiary etc). n1 .26) in the last step. if there are primary constraints in the theory which are ﬁrst class. 2. Let us denote all the constraints of the theory (primary. .

388 10 Dirac method for constrained systems constraints have been determined. First.4) are not compatible with them. As we have seen in the case of Maxwell ﬁeld theory.37) into second class constraints. α = 1. vanish {F. The gauge ﬁxing conditions are generally denoted as e χα ≈ 0. 2(n1 + n2 ) < 2N. even though φA ≈ 0. after gauge ﬁxing we denote all the constraints of the theory (including the gauge ﬁxing conditions). 2. (10. we cannot yet impose them directly in the theory since the canonical Poisson brackets in (10. n1 . collectively as φA ≈ 0. a consistent Hamiltonian description.) The ﬁrst class constraints have a special signiﬁcance in that they are associated with gauge invariances (local invariances) in the theory.40) as . in general. · · · . in this case.39) and are chosen such that they convert the ﬁrst class constraints (10. / (10. (The number of undetermined Lagrange multipliers equals the number of ﬁrst class constraints among the primary constraints. Thus. · · · . let us note that we can construct the matrix of Poisson brackets of all the constraints in (10. Namely. (10.40) so that the true phase space of the theory is 2(N − n1 − n2 ) dimensional. A = 1. 2. requires gauge ﬁxing conditions and we must choose as many gauge ﬁxing conditions as there are ﬁrst class contraints. which are all second class. φA } ≈ 0. (10. Although we have identiﬁed all the constraints of the theory.42) This issue of incompatibility of the Poisson brackets can be addressed through the use of the Dirac brackets which are constructed as follows.41) the Poisson bracket of the constraints with any dynamical variable F does not.

(10. we can now deﬁne the Dirac bracket between any two dynamical variables as −1 {F. (10. φA } − {F. G}D = {F.43) This is an even dimensional anti-symmetric matrix and Dirac had −1 shown that it is nonsingular so that its inverse CAB exists.44) We note here parenthetically that in ﬁeld theories (where the variables depend on space coordinates as well). φA }D = {F. φA } − {F. (10. namely. G}. in contrast to the canonical Poisson brackets. The Dirac bracket has the interesting property . φA } −1 ≈ {F. we can show that the Dirac brackets are indeed the correct Poisson brackets of the theory subject to the constraints in (10. G} − {F. In fact. Jacobi identity) and in addition satisﬁes −1 {F.46) Namely. φA }CAB {φB . through the use of Lagrange brackets.10. (10. we can work with the Dirac brackets and set the constraints to zero in the theory (namely. As a result. the Dirac bracket is compatible with the constraints in the sense any dynamical variable of the theory has a vanishing Dirac bracket with a constraint. φB }CBD C DA = {F. the matrix of Poisson brackets in (10. A −1 −1 C AD CDB = δB = CBD C DA .2 Dirac method and Dirac bracket 389 {φA . 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.43) will be coordinate dependent and the product in (10. φA } = 0.44) will involve integration over the intermediate space coordinate.45) which can be shown to have all the properties of a Poisson bracket (anti-symmetry. φB }CBD {φD . φA } − {F.40). With this. φB } ≈ C AB . in the deﬁnitions of the Hamiltonian.

390 10 Dirac method for constrained systems that it can be constructed in stages. (10. If we denote the coordinate of the particle as qi . then the matrix of highest derivatives has the form .47) where summation over repeated indices is understood. · · · . 2. It is also worth emphasizing here that in dealing with ﬁeld theories where the matrix C AB is coordinate dependent. rather than dealing with a large matrix of Poisson brackets in (10. Namely. let us consider the motion of a point particle constrained to move on the surface of a ndimensional sphere of radius unity. 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. i = 1. therefore. · · · . We note that the dynamics of the system can be described by the Lagrangian 1 (qi qi − F (qi qi − 1)) . This simple system is an interesting example in the study of constrained quantization and is a prototype of the ﬁeld theoretic model known as the nonlinear sigma model.49) gives the constraint (10. when the number of constraints is large. 2. then the coordinates are constrained to satisfy (the metric in this simple theory is Euclidean and. the inverse needs to be deﬁned carefully with the appropriate boundary condition relevant for the problem. F into qa = (qi .43) and its inverse. n + 1.3 Particle moving on a sphere As a simple example of constrained systems. n. We note that if we combine the dynamical variables qi . ∂F 2 (10. F ). a = 1. 10.47) on the dynamics.48) where F is a Lagrange multiplier ﬁeld (an auxiliary ﬁeld without independent dynamics) whose Euler-Lagrange equation 1 ∂L = − (qi qi − 1) = 0. ˙ ˙ 2 L= (10. raising and lowering of indices can be carried out trivially) qi qi = 1.

52) The canonical Hamiltonian following from (10.22)) reﬂecting the fact that there are constraints in the theory and the naive Hamiltonian description would not hold.10.51) Therefore.55) . (10.50) which is indeed singular (see (10. {qi .51) has the form 1 i i ˙ Hcan = pi qi + pF F − L = ˙ p p + F (qi qi − 1) . we note that the conjugate momenta of the theory are given by ∂L = qi .3 Particle moving on a sphere 391 ∂2L = ∂ qa ∂ qb ˙ ˙ δij 0 0 0 . ˙ ∂ qi ˙ ∂L = 0. pF } = 0. {qi .48) and (10. {F. pF } = {pi . F } = {pi . qj } = {pi . (10.30)) 1 i i p p + F (qi qi − 1) + λ1 pF . To understand the passage to the Hamiltonian description in this simple case. (10. pj } = {F. 2 (10. F } = {pF . 2 leading to the primary Hamiltonian (see (10.53) Hp = Hcan + λ1 ϕ1 = (10. F } = {qi . pj } = δi . pF } = 1. pF } = 0.54) The equal-time canonical Poisson brackets of the theory are given by j {qi . ˙ ∂F pi = pF = (10. the theory has a primary constraint given by ϕ1 = pF ≈ 0.

we obtain ϕ2 ≈ {ϕ2 .58) which generates a new constraint ϕ3 = qi pi ≈ 0.59) to be time independent. Hp } ˙ 1 1 = { (qi qi − 1).60) . (pj pj + F (qj qj − 1) + λ1 pF )} 2 ≈ pi pi − F ≈ 0. 2 (10. = pi pj {qi .59) Requiring the new constraint (10. (pj pj + F (qj qj − 1) + λ1 pF )} 2 2 = qi pj {qi .57) Requiring (10.392 10 Dirac method for constrained systems With these we can now determine the time evolution of any dynamical variable. (pi pi + F (qi qi − 1) + λ1 pF } 2 1 = − (qi qi − 1) ≈ 0. pj } + qi F qj {pi .57) to be time independent. pj } = qi pi ≈ 0.56) Let us denote this secondary constraint as ϕ2 = 1 (qi qi − 1) ≈ 0. (10. Hp } ˙ 1 = {pF . Hp } ˙ 1 = {qi pi . (10. 2 (10. we obtain ϕ3 ≈ {ϕ3 . requiring the primary constraint to be independent of time leads to ϕ1 ≈ {ϕ1 . qj } = pi pi − F qi qi (10. In particular.

10. ϕ1 }.64) = −{ϕ3 . φB } = C AB . (10. the nontrivial equal-time Poisson brackets between the constraints take the forms {ϕ1 . (10. (10.3 Particle moving on a sphere 393 so that we can identify ϕ4 = pi pi − F ≈ 0. A = 1.61) to be time independent leads to ϕ4 ≈ {ϕ4 . It is easy to check that all the constraints of the theory (10. ϕ4 } = {pF . 4. pi pi − F } = 1 1 {ϕ2 . (pj pj + F (qj qj − 1) + λ1 pF )} 2 (10. qj pj } = qi qj {qi .61) 1 = {pi pi − F. ϕ3 } = { (qi qi − 1). we can calculate the matrix of the (equal-time) Poisson brackets of constraints {φA . ϕ2 }. . Collecting all the constraints into φA = (ϕ1 .61) are second class.52). Requiring (10.59) and (10. which takes the form (10. ϕ2 . In fact.63) where we have identiﬁed pi pi = p2 . = −{ϕ4 .57). ϕ3 }. pj pj − F } = 2pi pj {qi .62) = 2pi F qj {pi . which determines the Lagrange multiplier and the chain of constraints terminates. ϕ3 . Hp } ˙ (10. qj } − λ1 {F. ϕ4 ). ϕ4 } = {qi pi . pj } = qi qi ≈ 1 2 {ϕ3 . 2. pF } = −2F qi pi − λ1 ≈ −λ1 ≈ 0. pj } = 2pi pi = 2p2 = −{ϕ4 . 3.

qk pk }C32 { (qℓ qℓ − 1). (qj qj − 1)}C21 {pF .66) (10. pF }D = {F. G}. 0 0 (10. = −qi pj + pi qj . pF }D = 0. pj } − {qi . {qi . qj }D = 0. (qk qk − 1)}C23 {qℓ pℓ . pj } 2 j = δi − qi qj . F }D = {pF . φA }CAB {φB . pF }D = 0. 1 −1 {pi .66). pj } − {pi . {F. F } − {pi . −1 {qi . In particular. F } = 2pi . pj }D = {pi . G}D = {F. qk pk }C32 { (qℓ qℓ − 1).45)) −1 {F. pj pj − F }C41 {pF . F }D = {pi . F } − {qi . (10. 2 {qi . 2 2p = 0 1 0 . pF }D = {pi . F }D = {qi . pj }D = {qi .68) . = 0 −1 0 2p2 2 −1 0 −2p 0 0 0 0 1 (10. pj } 2 −1 1 −{pi . G} − {F. the Dirac brackets between the fundamental dynamical variables take the forms 1 −1 {pi . pj } 2 −1 1 {qi . F } = −2qi p2 .65) −1 CAB The Dirac brackets between any two dynamical variables can now be deﬁned as (see (10.394 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 .67) −1 with CAB given in (10.

pi ) (we note that F = pi pi because of the constraint (10. dτ pµ = muµ .69) where we have used (10.54)) 1 i i pp. (10.70) can be obtained as the EulerLagrange equation following from the action (recall that c = 1) dλ (xµ xµ ) 2 = ˙ ˙ 1 S=m ds = m dλ L. 10. dτ 2 m (10. dτ (10.4 Relativistic particle In the last section we studied a simple example of point particle dynamics which is constrained.61)) and the Hamiltonian for the theory is given by (see (10.71) then the dynamical equation (10.62). We note that if we deﬁne the relativistic four velocity and momentum as uµ = dxµ .4 Relativistic particle 395 With the use of the Dirac brackets. (10. we can set the constraints to zero so that the true independent dynamical variables are (qi .70) can also be written as dpµ = 0.72) The dynamical equation (10. Let us next study the Hamiltonian description of a free relativistic massive particle which is described by the equation d2 xµ = 0.10. 2 Hp = (10. The constraints of this theory were all second class.73) where λ is a parameter labelling the trajectory and we have identiﬁed . This is the starting point for the quantization of this theory.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.

let us proceed with the Hamiltonian description without choosing a gauge at this point. We note from (10. dλ xµ = ηµν xν .72) as the Euler-Lagrange equations.77) where τ denotes the proper time associated with the particle.70) or (10.396 10 Dirac method for constrained systems xµ = ˙ dxµ .78) which leads to . If we choose a gauge λ = τ. the action (10.76) Namely. then (10.73) will lead to the dynamical equation (10. (10.73) that ∂2L = ∂ xµ xν ˙ ˙ m xρ xρ ˙ ˙ xµ xν ˙ ˙ xσ xσ ˙ ˙ Pµν = ηµν − .73) is invariant under the diﬀeomorphism (10. ˙ ˙ (10. (10.75) which is an example of a local gauge transformation much like in the Maxwell ﬁeld theory. (10.74) We note that under a transformation xµ = ˙ dξ dxµ dxµ = .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. so that (10. dλ dλ dξ λ → ξ = ξ(λ).

82).81) are equivalent. To see that the two actions (10. Pµν is a projection operator and ∂2L = 0.81) where because of the diﬀeomorphism invariance of (10. For example.73) in the form 1 2 √ ˙ ˙ dλ g g−1 xµ xµ + m2 . We note that introducing an auxiliary variable g. ˙ (10.79) Therefore. ∂ xµ xν ˙ ˙ det Pµν = det (10.82) If we eliminate g in (10.73) can also be written in an alternate form that is more useful. it is not clear how to take the m = 0 limit in (10.81) are xµ and g.81). we can rewrite the action (10. g= xµ xµ ˙ ˙ . The two dynamical variables of the theory in (10. it is more convenient for us to work with the action (10. we note that the equation for the auxiliary ﬁeld g takes the form g−1/2 ∂L = −g−1 xµ xµ + m2 = 0. Viewed in this manner. We can obtain the momenta conjugate to these variables to be . then we obtain (10.81) using (10.80) as in the case of the Maxwell theory (see (9.10.81) is manifestly diﬀeomorphism invariant and the massless limit can now be taken in a straightforward manner.73) and (10.73) we can think of g as the metric on the one dimensional manifold of the trajectory of the particle. m2 (10.124)) and the naive passage to the Hamiltonian description fails (see (10.22)).73) to describe the motion of a massless particle. However. ˙ ˙ ∂g 4 or.73).4 Relativistic particle 397 xµ Pµν = ˙ m xρ xρ ˙ ˙ xν − ˙ xµ xµ xν ˙ ˙ ˙ xσ xσ ˙ ˙ = 0 = Pµν xν . Let us note that the action (10. the action (10. S= dλ L = (10.

pg } = 1. Hp = Hcan + λ1 ϕ = 2 1 (10. pg } = 0. pg } = {pµ . µ {xµ . {xµ .84). 2 µ (10.86) The equal time canonical Poisson brackets for the theory are given by {xµ .398 10 Dirac method for constrained systems pµ = ∂L = g−1/2 xµ . (10.88) ϕ ˙ 1 ≈ {ϕ . ∂g ˙ (10.85) Adding the primary constraint (10. g} = {pg . {g. pg } = 0. we obtain √ g µ p pµ − m2 + λ1 pg } 2 (10. 4 g which leads to the secondary constraint . g} = {pµ . ˙ ∂ xµ ˙ pg = ∂L = 0. g} = {xµ . pν } = {g. 1 1 = − √ pµ pµ − m2 ≈ 0.84) Hcan (10.84) and requiring the constraint to be time independent. pν } = δν . The canonical Hamiltonian of the theory follows to be √ g µ = pµ x + pg g − L = ˙ ˙ p p µ − m2 .87) With these we can now determine the evolution of the primary constraint (10. Hp } = {pg . we obtain the primary Hamiltonian for the theory to correspond to √ g µ p p µ − m 2 + λ1 p g .83) so that the primary constraint of the theory follows to be ϕ1 = pg ≈ 0. xν } = {pµ .

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

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

5 Dirac ﬁeld theory As we have seen in (8.10.22)). where α = 1. 2.97) On the other hand. from the deﬁnition of the inﬁnitesimal invariant length in the Minkowski space (see (1.20)). dxµ dxµ = 1.91) would imply dxµ dxµ = 1. it is obvious that the matrix of second order dervatives vanishes and the system is singular (see (10.5 Dirac field theory 401 the other hand. 10. we have (remember that c = 1) dτ 2 = dxµ dxµ .98) It follows. dτ dτ (10. or. dλ dλ xµ xµ = ˙ ˙ (10. (10. ψα .99) (10. that the ﬁrst of the gauge ﬁxing conditions corresponds to choosing λ = τ . This manifests in the deﬁnition of the conjugate momenta as (we have chosen the convention of left derivatives) . if we set the constraints strongly to zero.82) that the ﬁrst of the gauge ﬁxing conditions in (10. 3. therefore.100) † The dynamical variables of the theory can be chosen to be ψα . the free Dirac theory is described by the Lagrangian density / L = iψ∂ ψ − mψψ.) Let us also note from (10.89) in particular leads to H = p0 = p2 + m2 which can be taken as the Hamiltonian. where the adjoint spinor is given by ψ = ψ† γ 0 .9). 4 denote the Dirac indices and since the Lagrangian density is ﬁrst order in derivatives. (10.

(10.101) in the intermediate steps.104) Adding the primary constraints to the canonical Hamiltonian.30)) Hp = Hcan + d3 x φ† ξα + λ† ρα . Rather they correspond to the momenta conjugate to ψ and ψ † respectively. (10. The canonical Hamiltonian density of the theory (consistent with the convention of left derivatives) is obtained to be (see (10.105) . The canonical Hamiltonian is now obtained to be Hcan = d3 x Hcan = d3 x −iψγ · ∇ψ + mψψ .102) which in this case correspond to fermionic constraints. † ˙ † ˙ = iψα ψα − iψα ψα − iψγ · ∇ψ + mψψ (10.103) where we have used (10. ˙† ∂ ψα (10. α α ρα = Πα ≈ 0. we obtain the two sets of primary constraints for the theory to be † ϕ† = Π† + iψα ≈ 0. ˙ ∂ ψα ∂L = 0.) As a result.101) (Note that Π† is not the Hermitian conjugate of Π.402 10 Dirac method for constrained systems Π† = α Πα = ∂L † = −i(ψγ 0 )α = −iψα . we obtain the primary Hamiltonian for the Dirac ﬁeld theory as (see (10.11)) ˙ ˙† Hcan = −Π† ψα + ψα Πα − L α = −iψγ · ∇ψ + mψψ. α α (10.

β (10. Πβ (y)}. For example. we can identify x0 with the time variable of the ﬁeld operators in Hp leading to (we do not show equal-time explicitly for simplicity) {φ† (x). Hcan } α = = = = † d3 y {Π† (x) + iψα (x). −iψ(y)γ · ∇y ψ(y) + mψ(y)ψ(y)} α d3 y iψ(y)γ · ∇y − mψ(y) β {Π† (x).13)(10.12)) † {ψα (x). we can now calculate the time evolution of the primary constraints (10. Using (10.106) with all other brackets vanishing. Hcan } + α d3 y ξβ (y){φ† (x). Hp } α α ≈ {φ† (x).102). Π† (y) + iψβ (y)} α β .107) α β Let us evaluate each of these terms separately. λ† in the present case denote α fermionic variables. ˙ φ† (x) ≈ {φ† (x).10.14)). ρβ (y)} . Π† (y)} = −δαβ δ3 (x − y) = {ψα (x). φ† (y)} α β = † † d3 y ξβ (y){Π† (x) + iψα (x). φ† (y)} α β −λ† (y){φ† (x).5 Dirac field theory 403 where the Lagrange multipliers ξα .106) (see also (10. ψβ (y)} α d3 y iψ(y)γ · ∇y − mψ(y) i∇ψ(x) · γ + mψ(x) α β − δαβ δ3 (x − y) . (10. The equal-time canonical Poisson brackets for the ﬁeld variables take the forms (see (10. d3 y ξβ (y){φ† (x). Using the fact that the primary Hamiltonian is independent of time.

109) which determines the Lagrange multiplier λ† = − ∇ψ(x) · γ + imψ(x) α α . − 10 Dirac method for constrained systems d3 y λ† (y){φ† (x).107) and requiring the constraint to be independent of time leads to ˙ φ† (x) ≈ i∇ψ(x) · γ + mψ(x) α + iλ† (x) = 0. requiring the second set of primary constraints to be time independent we obtain ρα (x) ≈ {ρα (x). (10.108) = iλ† (x).111) which determines the other Lagrange multiplier ξα = γ 0 (γ · ∇ + im)ψ(x) α . (10. 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) = 0.110) Similarly. Hp } ˙ = = † d3 y {Πα (x). (10. Hp } = {Πα (x).404 = 0. Πβ (y)} α β d3 y λ† (y) − iδαβ δ3 (x − y) β (10. α Substituting these results into (10. ρβ (y)} α β = − = − † d3 y λ† (y){Π† (x) + iψα (x).112) . α α (10.

we obtain Hp = = d3 x −iψγ · ∇ψ + mψψ + φ† ξα + λ† ρα α α d3 x −iψγ · ∇ψ + mψψ † + (Π† + iψα ) γ 0 (γ · ∇ + im)ψ α α + = − ∇ψ · γ + imψ α Πα d3 x Πγ · ∇ψ + imΠψ + (−∇ψ · γ + imψ)Π . Πβ (y)} = 0. In fact. we can write α the matrix of Poisson brackets as . Πβ (y)} α α = −iδαβ δ3 (x − y) = {ρα (x). we have deﬁned Π = Π† γ 0 . for simplicity of notation. ρβ (y)} = {Π† (x) + iψα (x).105). ρα ). which constitute all the constraints of the theory. Substituting (10.102). deﬁne second class constraints (this is consistent with the fact that there is no undetermined Lagrange multiplier in the theory).112) into the primary Hamiltonian. we have (at equal time) † † {φ† (x).113) where. we see that requiring the two primary constraints to be time independent does not introduce any new constraint.5 Dirac field theory 405 In this case. α α β β † {φ† (x). φ† (y)} = {Π† (x) + iψα (x).115) Thus. β {ρα (x). (10. (10. (10.114) Let us note next that the two sets of primary constraints (10.10. combining the constraints into ΦA = (φ† . rather it determines both the Lagrange multipliers in the primary Hamiltonian (10.110) and (10. φ† (y)}. Π† (y) + iψβ (y)} = 0. ρβ (y)} = {Πα (x).

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

2. (10. 4 µ.121) which is the Hamiltonian we have used in the quantization of the Dirac ﬁeld theory (see (8.113) becomes Hp = d3 x −iψγ · ∇ψ + mψψ . 1.119) = −iδαβ δ3 (x − y). Π† (z) + iψγ (z)}{Πγ (z).120) and these are the relations we have used in quantizing the Dirac theory (see (8. where we have neglected the last term in (10. (10.16). Therefore.10. Similarly. ψβ (y)}D = 0 = {ψα (x). ψβ (y)}D = {ψα (x). Maxwell’s equations can be obtained as the Euler-Lagrange equations from the Lagrangian density 1 L = − Fµν F µν . we can show that † † {ψα (x).18)). ψβ (y)} γ d3 z(−δαγ δ3 (x − z))(−δγβ δ3 (z − y)) (10.16)).122) where the ﬁeld strength tensor (see (9. 10. 3. ψβ (y)}D .6 Maxwell ﬁeld theory Let us next consider the Maxwell ﬁeld theory which as we know exhibits gauge invariance.4)) .6 Maxwell field theory 407 † † {ψα (x). Furthermore. we recall that when using the Dirac brackets. ψβ (y)} −i = −i † † d3 z{ψα (x). As we have seen in (9. we can set the constraints to zero in the theory. ν = 0.118) in evaluating this bracket because Πγ has a vanishing Poisson bracket with ψα . the Hamiltonian of the theory (10. (10.

408 10 Dirac method for constrained systems Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . the matrix of quadratic derivatives becomes singular (see (9. i.128) We can now obtain the canonical Hamiltonian density of the theory to have the form .122)) and this manifests in the deﬁnition of the conjugate momenta as ∂L = −F 0µ .127) This determines that the theory has a primary constraint given by ϕ1 (x) = Π0 (x) ≈ 0. (10. ˙ ∂ Aµ Πµ = (10.123) is the four dimensional curl of the four vector potential and contains the electric and the magnetic ﬁelds as its components (see (9. 2. (10. k = 1. As a result of the gauge invariance of the theory. (10.7)) F0i = Ei . Fij = −ǫijk Bk .124) We also know that Maxwell’s theory is invariant under gauge transformations Aµ (x) → A′ (x) = Aµ (x) + ∂µ α(x). j. (10. ˙ Π = E = −(A + ∇A0 ). µ (10.125) where α(x) denotes the local parameter of gauge transformation.5) and (9.126) so that we have Π0 = 0. 3.

ν {Aµ (x). is given by Hcan = = d3 x Hcan d3 x 1 2 (Π + B2 ) − A0 ∇ · Π .132) Using these we can calculate the time evolution of the primary constraint (10. Aν (y)} = 0 = {Πµ (x).130) where we have neglected a surface term (total divergence) in integrating the last term by parts. The canonical Hamiltonian. 2 (10.128) Hp = Hcan + = d3 x d3 x λ1 ϕ1 1 2 (Π + B2 ) − A0 ∇ · Π + λ1 Π0 . (10. The equal-time canonical Poisson brackets of the theory are given by {Aµ (x). The primary Hamiltonian of the theory is now obtained by adding the primary constraint (10.128) and requiring the constraint to be time independent we obtain (the Poisson brackets are at equal-time which can be achieved using the time independence of Hp ) .6 Maxwell field theory 409 1 ˙ ˙ Hcan = Πµ Aµ − L = −Π · A + Fµν F µν 4 1 = −Π · (−Π − ∇A0 ) + (−E2 + B2 ) 2 1 2 (Π + B2 ) + Π · ∇A0 .127). Aµ (x)}.129) where we have used (10.10. Πν (y)} = δµ δ3 (x − y) = −{Πν (y). 2 (10.131) where λ1 denotes the Lagrange multiplier of the theory. Πν (y)}. therefore. = 2 (10.

This is consistent with our earlier observation that not all Lagrange multipliers in the primary Hamiltonian are determined when there are ﬁrst class constraints present. ˙ so that the chain of constraints terminates.134) It can now be checked that the secondary constraint is time independent. Hp } ≈ 0.136) and it is clear that both these constraints are ﬁrst class constraints. as we have seen. Therefore. We see that Maxwell’s theory has two constraints ϕ1 (x) = Π0 (x) ≈ 0. (10. the Lagrange multiplier λ1 remains as yet undetermined (there is one primary constraint which is ﬁrst class).133) = ∇ · Π(x) ≈ 0. (10. ϕ2 (x) = ∇ · Π(x) ≈ 0. we have a secondary constraint in the theory given by ϕ2 (x) = ∇ · Π(x) ≈ 0. namely. A0 (y)}∇ · Π(y) (10. In this case. Furthermore. Hp } ˙ = 1 d3 y {Π0 (x). Hp } = {∇ · Π(x). According to the Dirac procedure. ϕ2 (x) ≈ {ϕ2 (x).410 10 Dirac method for constrained systems ϕ1 (x) ≈ {ϕ1 (x). ﬁrst class constraints signal the presence of gauge invariances (local invariances) in the theory which we know very well in the case of Maxwell ﬁeld theory. (Π2 (y) + B2 (y)) 2 − A0 (y)∇ · Π(y) + λ1 (y)Π0 (y)} ≈ − d3 y {Π0 (x).135) (10. Let us choose . in the presence of ﬁrst class constraints we are supposed to add gauge ﬁxing conditions to convert them into second class constraints.

137) as the gauge ﬁxing conditions which clearly convert the constraints (10. relations (10.136) into second class constraints. y) = {φA (x). ϕ1 (y)}. Therefore.139) The inverse of this matrix is easily calculated to be 0 0 1 0 −1 CAB (x. (Π)j (y)} = (∇x )i (∇y )j (δij δ3 (x − y)) 2 = −∇x δ3 (x − y) = −{χ2 (x). A0 (y)} {ϕ2 (x).136) and (10. = −1 0 0 0 −2 0 −∇x 0 0 (10. χ1 (y)} = {Π0 (x). χα ).137) determine all the constraints of the theory.138) which leads to (x0 = y 0 ) C AB (x. χ2 (y)} = {∇ · Π(x). y) −2 0 0 0 ∇x 3 δ (x − y). ϕ2 (y)}. χ2 (x) = ∇ · A(x) ≈ 0. (10.10. ∇ · A(y)} = −δ3 (x − y) = −{χ1 (x).6 Maxwell field theory 411 χ1 (x) = A0 (x) ≈ 0. α = 1. 2. = 1 0 0 0 2 0 ∇x 0 0 (10. Combining the ﬁrst class constraints as well as the gauge ﬁxing conditions into φA = (ϕα . = (∇x )i (∇y )j {(Π)i (x). φB (y)} 0 0 −1 0 2 0 0 0 −∇x 3 δ (x − 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). (10.140) .

Π0 (z)}(δ3 (z − z )){A0 (¯). (10. with the usual asymptotic condition that ﬁelds vanish at spatial inﬁnity. Aj (y)}D = 0 = {Πi (x).142) ¯ 1 1 3 2 δ (x − y) = − 4π|x − y| . for the dynamical ﬁeld variables.141) and.143) Since we can now set the constraints (10. ∇x (10. Πν (y)}D . Πν (y)}D = {Aµ (x). Πν (y)} ¯ ν 0 ν = δµ δ3 (x − y) − δµ δ0 δ3 (x − y) − d3 zd3 z δµ δj ¯ i ν −2 ×((∇z )i δ3 (x − z))(∇z δ3 (z − z ))((∇z )j δ3 (¯ − y)) ¯ z ¯ ν 0 ν i ν = (δµ − δµ δ0 ) + δµ δj (∇x )i 1 3 j 2 (∇x ) δ (x − y). φA (z)}CAB (z.143) to be {Ai (x). . in particular. Πν (y)} ¯ ¯ z {Aµ (x). G(y)} − −1 ¯ z d3 zd3 z {F (x). Πj (y)}D .144) j = δi TR (x − y). d3 zd3 z {Aµ (x). z ){φB (¯). G(y)}D = {F (x). the relevant Dirac brackets for the Maxwell ﬁeld theory follow from (10. Πν (y)} − −2 ¯ z +{Aµ (x). Aν (y)}D = 0 = {Πµ (x). Πj (y)}D = δi + (∇x )i 1 3 j 2 (∇x ) δ (x − y) ∇x (10. j {Ai (x).137) to zero. it leads to {Aµ (x).412 10 Dirac method for constrained systems where.136) and (10. ∇x (10. ∇z · Π(z)}(∇z δ3 (z − z )){∇z · A(¯). G(y)}. the formal inverse of the Laplacian (Green’s function) has the explicit form −2 ∇x δ3 (x − y) = The equal-time Dirac bracket between any two dynamical variables can now be calculated from {F (x).

Teitelboim. A. Constrained Hamiltonian Systems.10. Roma (1976). 3. A. Berlin (1982). 1018 (1951). 269 (1987). Regge and C. 2 Hp = (10. Constrained Dynamics. A. Springer-Verlag. J. T.30) (along with the ﬁrst two of (9. Das and W. M. 6. Physical Review 83. 1 (1951).136) to zero. K. Acta Physica Polonica B18. 10. 4. ibid. Hanson. Bergmann. Barcelos-Neto. J. Anderson and P. 3. 129 (1950). Canadian Journal of Mathematics 2. we obtain the Hamiltonian for the theory (10. setting the constraints (10. 5. New York (1964). Sundermeyer. M. Furthermore. Yeshiva University. Accademia Nazionale dei Lincei. A. Scherer.7 References 1. Lectures on Quantum Mechanics. P. Dirac.131) to be d3 x 1 Π2 + B2 .145) which is the Hamiltonian we have used in the study of Maxwell’s theory in chapter 9. 2. L. Dirac. . P.7 References 413 which are the relations used in (9.25)) in quantizing Maxwell’s theory. G.

This page intentionally left blank .

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

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

6) from two equivalent points of view.1. Namely.6) holds. P −→ − P . simply corresponds to asking whether this group can be carried over to quantum mechanics without any obstruction. For a long time it was believed that the microscopic systems are also invariant under parity. let us assume that under the parity transformation. As is standard in quantum mechanics. of course. P P (11. under space reﬂection an arbitrary quantum mechanical state transforms as .6) where we have used the fact that classically vectors change sign under space reﬂection (see (11. the parity transformation (11. (11.1 Parity in quantum mechanics.1) would result in X −→ − X . shows that even that does not present any diﬃculty. but we know now that there exist processes in nature which do not respect parity.5) Deﬁning a symmetry in a quantum system. we can analyze the parity transformation in (11.2)) and that expectation values of quantum operators behave like classical objects (Ehrenfest theorem). let us consider a one dimensional quantum mechanical system. But a systematic analysis. the classical laws of physics are invariant under parity or space inversion. In this case. In other words.1 Parity 417 so that that even classically parity operation deﬁnes a group with two elements (a group of order two). In this case. quantum mechanical states change.11. we know that dynamical laws of physics (such as Newton’s equation) do not depend on the handedness of the coordinate frame. First. but not the operators such that (11. as we will see. 11. ½ and P with P 2 = ½. The only diﬃculty we may apprehend is in the case of spinor states since they are double valued. Classically. namely. let us assume that P represents the operator which implements the action of parity on the quantum mechanical states. For simplicity.

P is an idempotent operator and the only eigenvalues of P are ±1.9) that xP |y P = x|P † P|y . It is now obvious from (11. such that ψ|X|ψ ψ|P |ψ → → P P (11.11) In other words.10) which leads to P † P = ½.9) δ(x − y) = x|P † P|y . (11. Furthermore. acting on the coordinate basis P must lead to |x → |xP = P|x = | − x . (11. or. −x| − y = x|P † P|y . = P| − x = |x .13) Namely.12) P 2 |x which shows that P2 = ½.418 11 Discrete symmetries |ψ → |ψ P = P|ψ .8) Note that parity inverts space coordinates and. the parity operator is unitary. therefore. (11. since P|x = |−x . (11. or. Since P has real eigenvalues we conclude that the parity operator is Hermitian and thus we have . (11. P (11.7) ψ P |X|ψ P = ψ|P † XP|ψ = − ψ|X|ψ . ψ P |P |ψ P = ψ|P † P P|ψ = − ψ|P |ψ .

dx |x x|ψ dx ψ(x)P|x (11. (11.15) so that we obtain x|ψ P = ψ P (x) = ψ(−x). P P (11. then P|ψ = ±|ψ .11.17) Let us note further that since the eigenvalues of P are ±1. an eigenstate of parity has an associated wave function which is either even or odd depending on the eigenvalue of the parity .18) (11.19) Namely. ½. (11. we note from (11.7) that P |ψ → |ψ P = P|ψ = P = = P dx ψ(x)|x = dx ψ(x)| − x . P (11. under a parity transformation the wave function transforms as ψ(x) −→ ψ P (x) = ψ(−x). In other words. or.16) Namely.1 Parity 419 P2 = P † = P = P −1 . for such a state |ψ −→ |ψ P = P|ψ = ±|ψ . if |ψ is an eigenstate of parity. ψ(x) −→ ψ P (x) = ψ(−x) = ±ψ(x).14) To see how a scalar (not a multi-component) wave function transforms under parity.

P )P = O(−X.14). A general state of the system. ψ|P P |ψ = − ψ|P |ψ . the corresponding wave function need not have any such symmetry behavior. does not have to be an eigenstate of the parity operator and.22) where we have used (11. [P. however. rather under a transformation only the operators change as (this point of view may be more relevant within the context of ﬁeld theories) O −→ OP = P † OP. P ) −→ OP (X. namely. or. X]+ = 0. (11. In the other point of view. Similarly.6) holds.20) ψ|X|ψ ψ|P |ψ → → P ψ|X P |ψ = − ψ|X|ψ . P ) = P † O(X. P P (11. one can show that for any operator in this simple one dimensional theory. such that (11. we can obtain that [P. therefore. (11. or.23) In general.24) . P † P P (11. PX + XP = 0. P = O P † XP. −P ).420 11 Discrete symmetries operator for that state. we may assume that the states do not transform under the parity transformation. (11.21) It follows from this that P X −→ X P = P † XP = −X. O(X. P ]+ = 0.

It is obvious from simple systems like the one dimensional harmonic oscillator 1 P2 + mω 2 X 2 = H(−X.25) As a result. U (t)] = 0. (Recall that the Hermite polynomials. If parity is a symmetry of the theory.28) . or. (11. (11. 2m 2 P H(X.) If the Hamiltonian H of a quantum mechanical system is time independent.11. which are eigenfunctions of the Hamiltonian are even or odd functions of x depending on the value of the index n. P ) −→ P † H(X. and it follows from (11. indeed.26) that this is. H] = 0. −P ) = H(X.27) where U (t) is know as the time evolution operator. P ) = (11. (11.29) (11. −P ). H(X. [P. namely. Hn (x). we see that if parity is a symmetry of a quantum mechanical system. in such a case. P ). then as we have seen [P. P )] = 0. P and H can be simultaneously diagonalized and the eigenstates of the Hamiltonian would carry speciﬁc parity quantum numbers. P )P = H(−X. H(X. what happens. the eigenstates of the Hamiltonian would naturally decompose into even and odd states. the solution of the time dependent Schr¨dinger equation o can be written as |ψ(t) = e−iHt |ψ(0) = U (t)|ψ(0) .27) that [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. Therefore.

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

Namely. We have already seen this in the case of Lorentz transformations (see (3. We note that under a parity transformation a multi-component object would transform as P β ψα (x. t). t) −→ ψα (x.14) are more important. mix up the diﬀerent components of the object. we are led to the requirement that . P †P P = = = dxdy | − y y|P | − x x| dxdy − i dx − i d δ(x + y) | − y x| dy x| = −P dx |x x| (11. t) = ηψ Sα ψβ (−x. under a parity transformation.33) β where the matrix Sα can. in general. not only does x → −x. but the components of the object may mix with one another. Let us now go over to relativistic quantum systems. multi-component operators (recall the Dirac ﬁeld or the Maxwell ﬁeld). such systems can be described consistently only in the language of quantized ﬁelds which are. P (11.32) d |x dx = −P.11. On the other hand. in principle. we need to generalize our discussion of parity transformation to multicomponent objects. The important thing to note here is that such a representation is always nonlocal and not very useful from a practical point of view. As we have seen. Furthermore. the properties of the operator P such as in (11. Therefore.35) for transformation of wavefunction).1 Parity 423 = = − = −X dxdy (−xδ(x + y))| − y x| dx x|x x| = − dx X|x x| dx |x x| = −X. from the fact that two space inversions are equivalent to leaving the object unchanged. so that it truly gives a representation of the parity operator.

∂t2 or.2 Spin zero ﬁeld. In either of the cases. From our general discussion in (11.1. with ηφ = ±1. t) = φ(−x. t) −→ φP (x. 11.38) P (11. then φP (x. t).) 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.424 11 Discrete symmetries 2 ηψ S 2 = ½. Consequently. t) −→ 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. t) = P † ψα (x. t).35). ( + m2 )φ(x) = 0.37) If ηφ = 1 or the intrinsic parity of the spin zero ﬁeld is positive. (11. t). φ= (11. t). so is φ(−x. the equation is clearly invariant under space inversion. φ(x. we conclude that in this case. t) is a solution of the Klein-Gordon equation. (11. (11. t) = ηφ φ(−x.39) . (Remember that parity and the momentum operators do not commute and. Let us begin with the spin zero ﬁeld which is described by the Klein-Gordon equation ∂2 − ∇2 φ(x) = 0. if φ(x.34) In the language of quantized ﬁelds. (11. therefore. cannot have simultaneous eigenstates unless the eigenvalue of momentum vanishes. the above relation represents an operator relation P β ψα (x.34) and (11. t)P = ηψ Sα ψβ (−x.36) depending on whether the ﬁeld is massless or massive.

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

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

the current density has the explicit form J µ (x.53) P which is the correct transformation for the current four vector as we have seen in (11.) Thus we conclude that a charged spin zero particle and its antiparticle have the same intrinsic parity. t) − (∂ µ φ† (−x.29). see discussion in sections 7. ∗ (φP )† (x. t)∂ µ φ(−x. t) −→ J P µ (x.2 and 7. particles and antiparticles would have the same intrinsic parity. t). t) . t). t)∂ µ φP (x. t) = ηφ φ† (−x.51) Under a parity transformation J µ (x. t) φ(x.3.38)). (11. This result is. t))φP (x. (φ destroys a particle or creates an antiparticle whereas φ† destroys an antiparticle or creates a particle. t) − ∂ µ φ† (x. (11. t) = ηφ φ(−x. t)∂ µ φ(x.52) which leads explicitly to (recall that |ηφ |2 = 1) JP (x. t). t) = i φ† (x. t). (11.11. t) − (∂ µ (φP )† (x. t) = J 0 (−x. . in fact. Let us next look at the electromagnetic current associated with a charged Klein-Gordon system. t) = i (φP )† (x. since ηφ is assumed to take only real values ±1 (see (11. As we have seen in (7. t) .1 Parity 427 φP (x. t))φ(x. In quantum ﬁeld theory φ and φ† are associated with particles and antiparticles.50) On the other hand. quite general for the bosons and says that for bosons.2). J P 0 (x. t) = i|ηφ |2 φ† (−x.(11. the intrinsic parity of the ﬁeld φ† is the same as that of the ﬁeld φ. t) = −J(x.

Furthermore. ∂t ∂t B = ∇ × A. namely. t). we see that whereas the electric ﬁeld would transform like a vector. we must have E(x. t) −→ AP (x.428 11 Discrete symmetries 11. t) −→ −A(−x. A0 (x. E = − (11. t). B(x.54) under space reﬂection requires that under a parity transformation.57) This is very much like the transformations (11. t) = −A(−x.56) we conclude that under a parity transformation the components of the four vector potential would transform as A(x.2). t) −→ A0 (−x.1. t).58) P P . t) −→ −E(−x.3 Photon ﬁeld. P P (11. From Maxwell’s equations in the presence of sources (charges and currents) ∇ · E = ρ = J 0. t) = A0 (−x. t) −→ AP 0 (x. ∂A ∂A − ∇φ = − − ∇A0 . A0 (x. P P (11. t).54) we note that invariance of (11. ∂t (11.53) of the components of the current four vector under a space reﬂection. ∇×B = ∂E + J. t). (11. In the quantum theory. t). t) = ηA A(−x. t) −→ B(−x.55) Namely. t) = ηA0 A0 (−x. we would then generalize these as operator transformations A(x. the magnetic ﬁeld would behave like an axial vector if parity is a symmetry of the system. from the deﬁnitions of the electric and the magnetic ﬁelds in (9.

t)Aµ (x. This shows that the electromagnetic interaction is invariant under parity and. parity (quantum number) must be conserved in electromagnetic processes.60) we note that under a parity transformation Hem = → = = = = P d3 x J µ (x.33) or (11. t)) d3 x J µ (−x.1 Parity 429 Note that a physical photon has only transverse degrees of freedom (namely. t)AP 0 (x. (11. (11. namely A. t) = Hem .35). t) − JP (x. t)Aµ (x. t).4 Dirac ﬁeld. A0 is nondynamical).61) where in the last step we have let x → −x under the integral. t)A0 (−x. t)Aµ (x. t)AP (x. t) d3 x J µ (x. t) − (−J(x. we see that AP (x. In other words. (11.59) from which we conclude that the intrinsic parity of the photon ﬁeld (or the one photon state) is negative. 11. the photon is a vector particle. Thus according to our general discussion in (11. Let us recall that the Dirac wave function or the Dirac ﬁeld is described by a four component object. under a parity transformation . therefore. t) = −A(−x.11. t)) · (−A(x. From the parity transformation of the dynamical components. t)Aµ (−x.1. t) d3 x J 0 (−x. t) d3 x J P µ (x. t). t) · AP (x. From the form of the electromagnetic interaction Hamiltonian Hem = d3 x J µ (x. t) µ d3 x J P 0 (x.

t) = ψ(x. t)γ 0 = ηψ ψ † (−x. t).62) In matrix notation. we can always choose ηψ = ±1. t) = ηψ Sψ(−x. t)S † γ 0 ∗ = ηψ ψ † (−x. α = 1. S 2 = ½. With this choice (which also ∗ implies ηψ = ηψ ) we note that ∗ ψ (x. t)γ 0 S † γ 0 . (11. 3. t)γ µ ψ P (x. t) P P (11. we can write the transformation as ψ P (x. t)γ 0 γ 0 S † γ 0 P = ηψ ψ(−x. 4. 2.63) (11. t) = ηψ Sα ψβ (−x. t)γ 0 S † γ 0 γ µ Sψ(−x. (11.65) which will satisfy the above relation.64) Since we have assumed the parity eigenvalues to be ±1. P (11.67) 2 = ηψ ψ(−x. t). see (8. and under a parity transformation this would transform as J µ (x. t). t)γ µ ψ(x.430 11 Discrete symmetries P β ψα (x. t) −→ ψα (x. t)γ 0 S † γ 0 γ µ Sψ(−x. t) −→ J P µ (x.66) Let us recall that associated with the Dirac system is a conserved current (probability current or electromagnetic current. t) = (ψ P )† (x. with (see (11. t) = ψ (x.68) . t) = ψ(−x.34)) 2 ηψ S 2 = ½. t). (11. (11.44)) J µ (x.

J P 0 (x. we obtain ψ(−x. J P i (x. t). S = γ 0. t) = ψ(−x.74) It is now quite easy to see that if ψ(x. On the other hand. we obtain ψ(−x. or. t)γ 0 S † γ 0 γ i Sψ(−x. t). t) = ηψ γ 0 ψ(−x. t)γ 0 S † Sψ(−x. t). t) −→ ψ P (x. comparing (11.69) for µ = 0.72) Thus we determine the transformation of the Dirac ﬁeld (wave function) under a space reﬂection to be ψ(x.65) we have S = S † = S −1 . S † S = ½. ψ(x. (11. t) = −J i (−x. t) = ηψ ψ(−x. t) = −ψ(−x. (11. t) is a solution of the Dirac equation. namely.1 Parity 431 2 where we have used the fact that ηψ = 1.73) (11. or. we know the transformation properties of the current four vector under space inversion (see (11.71) Similarly.68) and (11.68) and (11. t) = J 0 (−x.53)).69) Comparing (11. t) also satisﬁes the Dirac equation in the new coordinates.70) so that using (11. t)γ 0 . t). t)γ 0 ψ(−x. Namely. if we deﬁne . P P P (11.69) for µ = i. (11. t)γ i ψ(−x.11. t). t) −→ ψ (x. with ηψ = ±1. then the parity transformed function ψ P (x. (11.

76) In other words.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. in fact.432 11 Discrete symmetries y µ = (t. then ∂ − m ψ P (y) = ∂y µ (11. the two photons move with equal and opposite momentum. as we have discussed in section 3. However.8 eV) e− + e+ → γ + γ. Let ǫ1 and ǫ2 denote the polarization vectors for the two photons while k represents their relative momentum.77) In the rest frame of the positronium.6.9). The polarization planes of the photons in this decay can. −x). This can also be seen by noting that parity is a symmetry of the free Dirac Lagrangian density (8. space inversion is a symmetry of the Dirac equation. (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. To determine the relative parity between Dirac particles and antiparticles. ∂xµ (11.78) where ‘a’ and ‘b’ are scalar functions of momentum and ψpositronium represents the wave function for the positronium. the two component Weyl fermions violate P. Since the electromagnetic interactions conserve parity. the scalar transition amplitude will have the general form Mi→f = (a ǫ1 · ǫ2 + b k · (ǫ1 × ǫ2 )) ψpositronium . be measured and it . (11.

Namely.81) we conclude that the positronium state must be a pseudoscalar under parity.83) In other words. (11. the total parity of the state is given by ηTOT = ηe− ηe+ . This result can be more directly seen when we study charge conjugation in the next section. it can also be seen in an intuitive manner from the structure of the spinor functions as follows. in the rest frame of the positronium. (11.1 Parity 433 is experimentally observed that the two are perpendicular. Let us choose two representative solutions (of positive and negative energy) of the Dirac equation in the rest frame (see (2. namely. for this decay we have ǫ1 · ǫ2 = 0. we must have ηTOT = ηe− ηe+ = −1. However. This is. (11. the transition amplitude for this decay takes the form Mi→f = (b k · (ǫ1 × ǫ2 )) ψpositronium . (11. the electron and the positron must have relative negative intrinsic parity. Let us recall from our earlier discussions that antiparticles are related to negative energy states. On the other hand. in fact.49)) .80) k · (ǫ1 × ǫ2 ) −→ −k · (ǫ1 × ǫ2 ) .79) Consequently. Since the transition amplitude is a scalar under parity while P (11.82) and for this state to represent a pseudoscalar state. a very general result for fermions.11. the relative intrinsic parity between Dirac particles and antiparticles is negative.

(11. t). For example. On the other hand.3) under parity. angular momentum).85) 0 P 0 = −→ ηγ 0 0 0 0 P −→ ηγ 0 = 1 0 1 0 0 = η . t)ψ(−x. t)γ5 ψ P (x. t) −→ ψ (x. t) ψ(−x. we note that ψ(x. t) = = P P |η|2 ψ(−x. t) −→ ψ (x. t) (11.100) (or in section 3. (11.86) which shows that this combination behaves like a scalar under parity. namely.84) introduced in (2. 0 0 0 0 0 0 0 0 = −η . Let us next calculate the transformation properties of the sixteen Dirac bilinears ψΓ(α) ψ. t). t) = = P P |η|2 ψ(−x. the only meaningful quantity as far as the fermions are concerned is the relative parity of a pair of fermions. t)ψ P (x. . t)γ5 ψ(−x. t)ψ(x. ψ(x. since fermions always appear in pairs in any process (because of conservation of fermion number. t)γ 0 γ 0 ψ(−x.87) −ψ(−x. t)γ 0 γ5 γ 0 ψ(−x. t)γ5 ψ(x.434 11 Discrete symmetries ψp0 >0 ψp0 <0 This analysis also brings out another feature involving the fermions. 1 1 0 0 1 1 (11.

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

let us go back to Maxwell’s equations in the presence of charges and currents ∇ · E = ρ. J → −J. 11. if we know the interaction Lagrangian or Hamiltonian density for a given process. (11. ∇ · B = 0. B → −B.94)) and is denoted by C. ∂t ∂E + J. ∇×E = − ∇×B = ∂B .95) .2 Charge conjugation To understand the discrete symmetry known as charge conjugation. 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. E → −E. ∂t B = ∇ × A.93) We note that this set of equations is invariant under the discrete transformations ρ → −ρ. since ∂A − ∇A0 . E = − classically we see that charge conjugation leads to (11.436 11 Discrete symmetries ψγ5 ψφ. ∂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 (11.92) Conversely. Furthermore. (11.

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

(11. C (11. would deﬁne a transformation under which J µ (x) −→ i (φC )† ∂ µ φC − (∂ µ (φC )† )φC = = i|ηφ |2 φ∂ µ φ† − (∂ µ φ)φ† −i φ† ∂ µ φ − (∂ µ φ† )φ = −J µ (x). We have already discussed (see sections 7. Therefore.438 11 Discrete symmetries J µ = i φ† ∂ µ φ − (∂ µ φ† )φ . ∗ φ† (x) −→ (φC )† (x) = ηφ φ(x). (11.100) with |ηφ |2 = 1. then C φ(x) −→ φC (x) = ηφ φ† (x). we continue to use “†” instead of “∗” so that the passage to quantum theory will be straightforward).101) C Thus the ﬁeld transformations in (11.99) We note that if we treat φ(x) as a classical function (even though the functions are classical. under a charge conjugation. we have (|ηφ |2 = 1) 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).2 and 7.102) . this further supports the fact that (11. C C C Aµ (x) −→ −Aµ (x).100) would deﬁne the appropriate charge conjugation transformation for the current. (11. As a result. It follows now that under such a transformation. ∗ φ† (x) −→ ηφ φ(x).100) is the appropriate charge conjugation transformation for complex scalar ﬁeld.

99). we have to choose an operator ordering which we have not taken care of in the above discussion.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 .2 Charge conjugation 439 Fµν (x) −→ −Fµν (x). For example. C C (11.105) C since the operators φ and φ† do not commute with ∂µ φ† and ∂µ φ respectively. of course. then under the transformation (11. we note that if we deﬁne the electromagnetic current in the quantum theory as in (11. spin zero particles interacting with photons described by 1 L = (Dµ φ)† (Dµ φ) − m2 φ† φ − Fµν F µν .103) Therefore. All of the above discussion carries over to the quantum theory if we are careful about the operator ordering of φ and φ† . the Lagrangian density of charged. (11. Dµ φ(x) = (∂µ + ieAµ )φ(x) −→ ηφ (∂µ − ieAµ )φ† = ηφ (Dµ φ)† . C ∗ ∗ (Dµ φ(x))† = (∂µ − ieAµ )φ† (x) −→ ηφ (∂µ + ieAµ )φ = ηφ Dµ φ. As a result.11. As is known. 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 . can be traced to the fact that in going from a classical theory to the quantum theory.104) is invariant under charge conjugation. 4 (11.100) do not lead to the correct transformation properties for the current operator. The problem. it would appear as if the ﬁeld transformations (11.

106) leads to (this should be compared with the normal ordered charge deﬁned in section 7.106) 2 µ Jsym (x) = and it is easy to check that this current will transform properly (namely. if a spin-zero ﬁeld theory describes charge neutral particles which are their own antiparticles.107) µ the eﬀect of the inﬁnite constant in Jsym in (11. (11. we obtain i φ† ∂ µ φ + (∂ µ φ)φ† − (∂ µ φ† )φ − φ∂ µ φ† . (11. under a charge conjugation (11. As we have discussed earlier. In such a case.110) .440 11 Discrete symmetries normal ordering the product). 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). 2 µ Jsym (x) = (11.109) which brings out the particle-antiparticle symmetry in a more direct way.3) µ 0|Jsym (x)|0 = 0.108) This is esthetically more pleasing and treats the vacuum in a more symmetrical way. If we symmetrize the products in the deﬁnition of the current. (11. will change sign) under charge conjugation. Note that we can also write the symmetrized current as 1 µ (J (x) − J Cµ (x)) . it is described by a Hermitian ﬁeld operator φ† (x) = φ(x). But this has the interesting consequence that while (remember the Dirac vacuum picture) 0|J µ (x)|0 −→ ∞.

ηφ = ±1.114) Such ﬁelds or particles are known as self-charge-conjugate and simply describe particles that are their own antiparticles.2 Charge conjugation 441 φ(x) −→ φC (x) = ηφ φ† (x) = ηφ φ(x). one can show that for all Hermitian spin zero ﬁelds that we know of (scalar or pseudoscalar) ηφ = 1.115) since the charge parity of the photon is (−1). we conclude that the π 0 meson must also have charge parity (+1) and similar arguments hold for other mesons.112) In this case. we expect 2 ηφ = 1. (11. the sign of the charge parity cannot be ﬁxed from theory.113) φ(x) −→ φC (x) = φ(x).11. let us note that in the decay π 0 → 2γ. (11. the total charge parity of the ﬁnal state is (+1). As an example. . or. C (11. C (11.111) From the fact that two charge conjugation operations should return the ﬁeld to itself. But by analyzing various physical processes where charge conjugation is a symmetry or by analyzing the invariance of various interaction terms under charge conjugation. If charge parity is to be conserved in this process (this corresponds to an electromagnetic process since photons are involved). so that in these cases (11.

T T T (11.116) +m = 0.2.442 11 Discrete symmetries 11. Let us next analyze a system of Dirac fermions in- teracting with an electromagnetic ﬁeld. To see the relation between ψ C (x) and ψ(x). from (11. T (11.120) (11.2 Dirac ﬁeld. (11. it would satisfy the equation (iγ µ (∂µ − ieAµ ) − m) ψ C (x) = 0. let us note from (11. (iγ µ (∂µ − ieAµ ) − m) Cψ = 0.117). the equation of motion takes the form (iγ µ (∂µ + ieAµ ) − m) ψ(x) = 0.119) where we have identiﬁed the charge conjugate function ψ C as describing the positron. (11. In fact. iC(γ µ )T C −1 (∂µ − ieAµ ) + m Cψ = 0.118) and (11.119) that the two equations are very similar in structure.117) Taking the transpose of the adjoint equation (11. or. we obtain i(γ µ )T (∂µ − ieAµ ) + m ψ = 0.118) Let us recall that a positron which is the antiparticle of the electron is also a Dirac particle with the same mass but carries an opposite electric charge. or. Thus. The adjoint equation is easily obtained to correspond to ← ψ(x) iγ µ ∂µ −ieAµ (11.121) . interacting with an electromagnetic ﬁeld. we note that if we can ﬁnd a matrix C satisfying C(γ µ )T C −1 = −γ µ . If the fermions are minimally coupled to the photons.118) we would have C i(γ µ )T (∂µ − ieAµ ) + m ψ = 0.

C T = C.126) (11. C −1 γ µ C = −(γ µ )T . T (11. Therefore we conclude that if we can ﬁnd a matrix C with the property C(γ µ )T C −1 = −γ µ . namely. −(γ µ )T . we can identify (with |ηψ |2 = 1) ψ C (x) = ηψ Cψ . (11.127) (11.123) That such a matrix exists can be deduced from the fact that −(γ µ )T also satisﬁes the Cliﬀord algebra. (This property depends on the dimensionality of space-time. −(γ µ )T + = 2η µν ½.119). or.) To see this. from C −1 γ µ C = −(γ µ )T . we can also show that σ µν C = − (σ µν C)T .122) (11. The important symmetry property that C must have is that it should be antisymmetric. (11.92)). namely. Similarly.124) and hence from the generalized Pauli theorem (see (1.125) . we obtain γ µ C = −C(γ µ )T = −C T (γ µ )T = − (γ µ C)T .128) (11. In this case. let us assume that instead C is symmetric.2 Charge conjugation 443 which can be compared with (11.11. it follows that there must exist a similarity transformation which relates −(γ µ )T and γ µ . In 4-dimension C T = −C.

C T = −C. the structure of the matrix C depends on the representation of the Dirac matrices and. in the PauliDirac representation that we have been using. therefore. (11. C † C = 1) ψ(x) −→ ψ C (x) = ηψ Cψ . Therefore.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 . ∗ ψ(x) −→ ψ (x) = −ηψ ψ T C −1 .444 11 Discrete symmetries so that if C T = C.129) Unlike the parity transformation.132) In general. But these are 4 × 4 matrices and. it follows that C must be antisymmetric.44)) . however. we have (with |ηψ |2 = 1.133) The fermion ﬁelds. we can construct 10 linearly independent antisymmetric matrices. there can be only six linearly independent antisymmetric matrices and C cannot be symmetric. (11. it is easy to determine that C = γ 0γ 2. satisfy anticommutation relations to reﬂect the Fermi-Dirac statistics that the underlying particles obey.131) (11. Thus. T (11. if we do not restrict to any particular representation. In this case. It now follows that ψ C = ηψ Cψ = ηψ γ 0 γ 2 (γ 0 )T ψ ∗ = −ηψ γ 2 ψ ∗ . C C T (11. given the classical current (see (8. the appropriate operator ordering for the current is antisymmetrization (this is equivalent to normal ordering in this case). as quantum mechanical operators.

(11. Therefore. this current transforms properly under charge conjugation. we can obtain the antisymmetrized current as µ Janti (x) = (11. Let us note here again that the antisymmetrized current diﬀers from the original current by an inﬁnite constant whose eﬀect is to make µ 0|Janti (x)|0 = 0. where we have used (11.138) Thus this current treats the vacuum more symmetrically which can also be seen from the fact that we can write the current in the form .11.134) = 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) .136) µ = −Janti (x).122) in the intermediate steps.2 Charge conjugation 445 J µ (x) = ψ(x)γ µ ψ(x).135) 2 It can now be easily checked that under charge conjugation the antisymmetrized current transforms as µ Janti (x) −→ C 1 C C ψ (x)γ µ ψ C (x) − (ψ C )T (x)(γ µ )T (ψ )T (x) 2 = = = 1 T |ηψ |2 −ψ T (x)C −1 γ µ Cψ (x) + ψ C T (γ µ )T (C −1 )T ψ 2 1 T T ψ T (x)(γ µ )T ψ (x) + ψ(x) C −1 γ µ C ψ(x) 2 1 T T ψ T (x)γ µ ψ (x) − ψ(x)γ µ ψ(x) 2 (11. (11. (11.137) whereas 0|J µ (x)|0 −→ ∞.

140) − (γ 1 )T T − (γ 2 )T T − (γ 3 )T T = γ5 . We have already seen in (11. . (11.141) i C −1 σ µν C = C −1 (γ µ γ ν − γ ν γ µ ) C 2 i = C −1 γ µ CC −1 γ ν C − C −1 γ ν CC −1 γ µ C 2 i − (γ µ )T − (γ ν )T − − (γ ν )T − (γ µ )T = 2 i (γ µ )T (γ ν )T − (γ ν )T (γ µ )T = 2 i ν µ = (γ γ − γ µ γ ν )T = −(σ µν )T .446 11 Discrete symmetries µ Janti (x) = 1 µ (J (x) − J Cµ (x)) .142) 2 Therefore. It follows from this and the deﬁnition of γ5 that C −1 γ5 C = C −1 iγ 0 γ 1 γ 2 γ 3 C = iC −1 γ 0 CC −1 γ 1 CC −1 γ 2 CC −1 γ 3 C = i − (γ 0 )T = i γ 3γ 2γ 1γ 0 Similarly. we obtain T C −1 γ5 γ µ C = C −1 γ5 CC −1 γ µ C = γ5 − (γ µ )T (11.122) that the charge conjugation matrix C satisﬁes C −1 γ µ C = −(γ µ )T . = i γ 0γ 1γ 2γ 3 (11. 2 (11. if we deﬁne properly antisymmetrized Dirac bilinear combinations as = − (γ µ γ5 )T = (γ5 γ µ )T .139) Let us next calculate how the Dirac bilinears transform under charge conjugation.

(11. α = V. A.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 . ψψ − ψ T ψ (11. (11.147) where the phase ǫα for the diﬀerent classes has the form 1. ǫα = −1. T.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 .) We have already seen that ψγ µ ψ −→ − ψγ µ ψ C anti anti . P. for α = S.148) . . (11. (ψψ)anti really measures the sum of probability densities (number densities) for particles and antiparticles and that should not change under charge conjugation (particle ↔ antiparticle). (11. for (11.2 Charge conjugation 447 ψΓ(α) ψ anti ≡ 1 2 ψΓ(α) ψ − ψ T Γ(α) T ψ T .144) 2 (If we think for a moment.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 . .

Thus P if ηψ represents the intrinsic parity of a Dirac particle.83). 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) = ηψ γ 0 ψ(−x. t) −→ ψ (x. let us note that since the charge conjugate wave function is deﬁned to be C ψ C = ηψ Cψ . P ψ(x. t) −→ ψ P (x. the antiparticle will transform as C ψ C (x. t). t) P −ηψ γ 0 ψ C (−x. t). t) = ηψ ψ(−x.152) Of course.150) under parity. P P P (11. we have already seen this from the analysis of the positronium decay in (11.151) Here we have used the fact that γ 0 is Hermitian in our metric.149) Since the antiparticle is described by C ψ C (x. but this derivation is simpler and more general. Next. T we obtain . (11. t) P P = = = = = C P ηψ ηψ C(ψ(−x. t)γ 0 . t) = ηψ Cψ (x. t) −→ (ψ C )P (x. T (11. t) = ηψ C(ψ )T (x. t)γ 0 )T C P ηψ ηψ Cγ 0 C −1 C(γ 0 )∗ ψ ∗ (−x.73) that under parity a Dirac particle transforms as P ψ(x.448 11 Discrete symmetries Let us next recall from (11. t). T T (11. t) P C −ηψ ηψ γ 0 C(γ 0 )T ψ ∗ (−x. t) C P −ηψ γ 0 ηψ Cψ (−x.

if γ5 ψ = −ψ. we have C γ5 ψ C = −ηψ Cψ = −ψ C . We have.2. (11.3 Majorana fermions. (p − m)ψ C = 0. of course. Let us recall that if a free Dirac particle satisﬁes the equation (p − m)ψ = 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) .157) In other words.119) we see that the antiparticle must also satisfy the same free equation.154) T (11. (11.6 and have used this in the construction of the standard model in section 14.153) As a result.158) then from (11.3.159) . T (11. / where (11. we see that if γ5 ψ = ψ. 11.11.155) On the other hand.156) T T C = ηψ Cψ = ψ C . / (11. already seen this in the study of neutrinos in section 3. then it follows that C γ5 ψ C = −ηψ C − ψ (11. the antiparticle of a particle with a given handedness has the opposite handedness. namely.

They can be massive and will have only two independent degrees of freedom (because of the constraint in (11. whether there are also fermionic particles that are their own antiparticles and if so how does one describe them.164) that for Majorana fermions (ψ = ψ C ) µ Janti (x) = 0. in fact. 2 = (11. Let us note from the deﬁnition of the antisymmetrized current in (11.135) µ Janti (x) = T 1 T ψ(x)γ µ ψ(x) − ψ T (x)γ µ ψ (x) 2 1 C ψ(x)γ µ ψ(x) − ψ (x)γ µ ψ C (x) . From the deﬁnition of charge conjugation in (11.162)). namely. (11. T 11 Discrete symmetries (11.450 ψ C = ηψ Cψ . These are charge neutral particles described by real quantum ﬁelds φ† = φ.163) T (11.123) we see that a Dirac particle which is its own antiparticle must satisfy ψ = ψ C = ηψ Cψ .165) so that the Majorana fermions are charge neutral and hence cannot have any electromagnetic interaction. . The answer to this question is. (11.162) Such particles are known as self charge conjugate fermions or Majorana fermions. / / (11.160) We have also seen that there are bosons in nature which are their own antiparticles.161) We can now ask the corresponding question in the case of the Dirac particles. quite obvious. and would also satisfy the free Dirac equation (p − m)ψ = (p − m)ψ C = 0.

However. then the antiparticle will have the intrinsic parity −(ηψ )∗ (although we had chosen the phase of the intrinsic parity for a Dirac fermion to be real.169) . (This is. it can in principle be complex). however. let us note that if we try to write a Lagrangian density for a massive Weyl fermion (say left-handed). (11.11. then χC = (ψL )C + ψL = χ. we can ﬁnd a simple relation between them. we note that if we deﬁne χ = ψL + (ψL )C . Since a Majorana fermion is its own antiparticle. this implies that it cannot have a well deﬁned handedness. we can have Majorana-Weyl spinors.167) (11. For a Majorana fermion then. not true if the space time dimensionality is not four. In particular for d = 2 mod 8.168) so that we can express a Majorana fermion in terms of Weyl fermions. as we have seen a particle with a deﬁnite handedness will lead to an antiparticle with the opposite handedness. 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. the naive mass term would have the form / L = iψ L ∂ ψL − mψ L ψL .166) which can be satisﬁed only if the intrinsic parity of a Majorana fermion is imaginary. the mass term would trivially vanish since (11. this would require P P ηψ = −(ηψ )∗ . To answer this.2 Charge conjugation 451 Secondly.) We have also seen that if the intrinsic parity of a Dirac particle P P is ηψ . In fact. Thus we cannot talk of a Majorana particle with a given helicity. (11. Let us also note here that since both the Weyl and the Majorana fermions have only two degrees of freedom.

/ (11.170) This is. the neutron and the neutrino are charge neutral and hence are prime candidates to be Majorana particles. Obviously.173) this would imply violation of lepton number which is also not seen yet. if (once again ν denotes the antineutrino) ν = ν. the essence of the statement that a Weyl fermion is massless. if we require conservation of fermion number. In fact. . Such a mass therm is known as a Majorana mass term and in writing this term we have obviously given up something to which we will return in a moment. note that any majorana particle would necessarily imply some sort of violation of a fermion number which is not seen in nature. in this case. But such an identiﬁcation is not compatible with the experimental results. if (here n represents the antineutron) n = n. Similarly.172) then this would imply violation of baryon number which is not seen at present. note that if we go beyond the conventional Dirac mass term in the Lagrangian density. 2 2 (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 . of course. a Weyl fermion cannot have mass. the mass term is not zero and we can give a mass to the Weyl fermion. However. (11. Obviously. Let us next ask whether there are any particles in nature which we can think of as Majorana particles. 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. namely.452 11 Discrete symmetries 1 1 ψ L ψL = ψ (1 + γ5 ) × (1 − γ5 ) ψ = 0.171) Clearly. (11.

. up to the experiments to decide whether this is really what happens. = QC|Q. . we have also seen that a Hermitian spin zero ﬁeld has charge parity +1. . Independent of whether supersymmetry is a symmetry of nature. mνµ < 190keV. means that a one photon state is an eigenstate of charge conjugation with eigenvalue −1. . . . . = Q|Q. . . then acting on an eigenstate it gives ˆ Q|Q. refer to other quantum numbers the state can depend on. .4 Eigenstates of charge conjugation. Another reason for the study of Majorana particles is that they are fundamental in the study of supersymmetry which we will not go into. 11. It is. so that comparing the two relations we obtain (11. . of course. . . .2 Charge conjugation 453 We can now ask why anyone would then like to study such particles or mass terms. We have seen that the pho- ton has odd charge conjugation parity. 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. (11. Similarly. . = Q| − Q. . of course. its theoretical studies would require the study of the properties of such particles. . = Q| − Q. .2MeV) (the present trend is to quote the lowest mass bound along with the diﬀerences of mass squares for the neutrinos determined from the neutrino oscillation experiments) suggest that the neutrinos may indeed have small masses. . mντ < 18. .175) (11.11. . This. The answer to this lies in the fact that while no one has really measured the mass of the neutrino. . It follows now that ˆ C Q|Q. . On the other hand. . . . . the direct experimental limits (mνe < 2ev.2. . .176) .174) where . let us note that if Q denotes the operator for charge (electric). we have ˆ ˆ QC|Q. To understand the eigenstates of charge ˆ conjugation in general. = −Q| − Q.

seen this already in the sense that the photon which does not carry charge has a well deﬁned charge parity and so does a Hermitian spin zero ﬁeld which describes charge neutral mesons. (11. 0 (11.180) so that such a state would be an eigenstate of charge conjugation with eigenvalue (−1)ℓ . = |K . ˆ C. the charge operator does not commute with the charge conjugation operator and.179) The ﬁnal state simply corresponds to the initial state with the two particles interchanged. we would have C|π − π + . of course. consequently. s respectively. (11. ℓ = |π + π − . We have. that while all eigenstates of C must be charge neutral. = |π + p . ℓ . not all charge neutral states would be eigenstates of charge conjugation operator.177) In other words. (11. for the positronium with orbital angular momentum and spin values ℓ. . ℓ = (−1)ℓ |π − π + . Thus if the state has well deﬁned angular momentum quantum number.178) which clariﬁes the point. It is easy to see that C|n C|π − p C|K 0 = |n . Note that C|π − π + = |π + π − . a state such as |π − π + can be an eigenstate of C. however. they cannot have simultaneous eigenstates unless the charge of the state vanishes. Similarly. Let us note. On the other hand.454 11 Discrete symmetries ˆ ˆ C Q = −QC. or. Q + = 0.

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

Therefore. any amplitude involving an odd number of photons must vanish if charge conjugation is a symmetry. electromagnetic interactions are invariant under charge conjugation. it is also known that charge conjugation symmetry holds true in strong interactions also. This result goes under the name of Furry’s theorem.1: An amplitude with N photons.188) In other words. As we have seen.189) where U denotes the time evolution operator.187) then conservation of charge parity would require (electromagnetic interactions are invariant under charge conjugation) Figure 11. (−1)n = (−1)N −n . then we must have . Similarly. and it follows from this that C † U C = U. (11. In fact. (−1)N = (−1)2n = 1.456 11 Discrete symmetries nγ → (N − n)γ. (11.190) (11. note that if a Hamiltonian is invariant under charge conjugation. (11. if the strong interaction Hamiltonian respects charge conjugation symmetry. or. then C † HC = H.

Thus we see that P and C (P for parity) are conserved in strong and electromagnetic interactions but are violated in weak interactions. On the other hand the product operation CP appears to be a good symmetry in most weak processes. 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 .2 Charge conjugation 457 π − p|U |π − p = π − p|C † U C|π − p = π + p|U |π + p . The simplest way to see this is to note that the weak current (see (14. it is violated in weak processes just like parity. Let us note here without going into details that while charge conjugation is a good symmetry for both strong and electromagnetic interactions. (11.191) In other words.192) 2 2 where ηJ denotes a phase. the weak current does not have a well deﬁned transformation property under charge conjugation and hence will violate C-invariance.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 . Namely.11. the scattering cross-section for π − p elastic scattering would be identical to the cross-section for π + p scattering which is experimentally observed. (11.

194) W 2 2 This has well deﬁned transformation properties under CP transformations and leads to CP invariance. time reversal or time reﬂection corresponds to the space time transformation where we reverse only the sign of time. x −→ x. (11. namely. Similarly.3 Time reversal Classically. ∇ × E(x) = − ∇ × B(x) = ∂B . Maxwell’s equations in the presence of sources ∇ · E(x) = ρ(x). 11.196) are also form invariant if we deﬁne .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 . (11.195) We note that the Newton’s equation is invariant under time reversal since it is second order in the time derivative. ∂t ∂E(x) + J. 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 . T (11. ∂t (11.193) 2 2 where ηTOT denotes the total phase under the transformation. ∇ · B(x) = 0. T t −→ −t.

is opposite from their behavior under parity in (11.199) The transformation of the Aµ (x) potentials under time reversal. the entropy or disorder increases with time.197) ∂A . we know that there is a sense of direction for time. t) −→ BT (x. However. we determine that under time reversal A0 (x. −t). t) −→ A0T (x. t) = E(x. Thus assuming that time inversion is also implemented by such an operator T . t) = −A(x. From the deﬁnition of the electric and the magnetic ﬁelds E = −∇A0 − B = ∇ × A. t) −→ AT (x. The ﬁrst problem which we face. t) = A0 (x. −t).3 Time reversal 459 ρ(x. −t). ∂t (11. however. −t).57). J(x.11. −t). we will have . T T T T T T (11. is that the time dependent Schr¨dinger equation is ﬁrst order in the time derivative just o like the diﬀusion equation and hence cannot be invariant under a naive extension of the time inversion operation to the quantum theory. −t). t) −→ ET (x. B(x. While some of the macroscopic equations are invariant under time reversal. E(x. t) = −B(x. in statistical mechanics we know that microscopic reversibility does hold. t) −→ ρT (x. Macroscopically. symmetry transformations normally are implemented by linear operators which are also unitary. t) = −J(x.198) (11. Thus we would like to investigate whether time reversal can be a symmetry of quantum mechanical systems. we know that macroscopic physics is not. namely. t) −→ JT (x. therefore. A(x. t) = ρ(x. Note that in quantum mechanics. The diﬀusion equation which is ﬁrst order in the time derivative is not symmetric under t → −t.

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

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

210) Thus from Ehrenfest’s theorem we would expect the corresponding operators to transform as T XT † = ηX X = X. Thus. T (11.462 11 Discrete symmetries In the time reversed process. we expect ψf |ψi −→ ψi |ψf .204) that ψ(x. −t).204) (see. (11. under time inversion.209) Let us derive the transformation properties of some of the operators under time inversion. T (11. In particular. t) = x|ψ(−t) = ψ ∗ (x. Pj ] . T ∗ (11. It follows now that T [Xi . T p −→ −p.211) T PT † = ηP P = −P. t) = x|ψ(t) −→ ψ T (x. ηP = −1. (11. Pj ] T † = T (Xi Pj − Pj Xi ) T † = Xi (−Pj ) − (−Pj )Xi = − [Xi . we know that x −→ x.207) (11. Classically. let us note from (11.208) which is indeed consistent with the deﬁnition of time inversion transformations given in (11. the role of the initial and the ﬁnal states are interchanged and the transition amplitude of interest would be ψi |ψf . It is reassuring to note that the canonical commutation relations are unchanged by the ηX = 1.211).212) where we have used (11. . on the other hand.204) as well as (11. in particular the third relation).

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

218) (11. which follows because under time inversion (say for m > 0) (11. (11.222) 11. φ). By convention. in general. the constant (phase) is chosen to be Cm = (−1)m = i2m .m (θ.221) It can be shown that the above relation holds true even for half integer angular momentum values. m (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ. we can indeed write T |ℓ. namely.223) . mj = T |j. mj = (i)2mj |j. We have already seen that under time reversal Aµ (x.464 11 Discrete symmetries |Cm |2 = 1.−m (θ. φ) = = T θ. φ|ℓ.3. T (11.m (θ. −t). (11.m (cos θ)e−imφ 4π (ℓ + |m|)! (11. t) −→ AµT (x. Thus.219) Yℓ. −mj . where we have used the fact that the spherical harmonics for m < 0 are deﬁned with a phase +1.m (cos θ)eimφ 4π ℓ + |m|)! ∗ −→ Yℓ. m = T |ℓ.1 Spin zero ﬁeld and Maxwell’s theory. t) = ηAµ Aµ (x. m = (i)2m |ℓ.220) (−1)m Yℓ. −m . we can write T |j. φ) = = (−1)m 2ℓ + 1 (ℓ − |m|)! Pℓ.

t). therefore.224) It follows. (11. t) −→ J µT (x. t) (11. invariant under time reversal as we have already seen and as can be checked directly. T Fij (x. ηA = −1. −t).3 Time reversal 465 where ηA0 = 1. −t) − J µ (x. t) − ∂ν AT (x. t) = T ∂µ Aν (x.225) T T −→ ∂µ AT (x. −t). of course. ηJ = −1.229) = d4 x S . where T (11. t) −→ Fij (x. (11.228) The Maxwell theory is. (11. ν µ so that T F0i (x.11. that Fµν (x.226) Similarly. the current four vector transforms as J µ (x. we have seen that under time reversal. t) = −Fij (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. −t) 4 1 = d4 x − Fµν (x)F µν (x) − J µ (x)Aµ (x) 4 = S. t) = ηJ µ J µ (x. −t) F µν (x.227) ηJ 0 = 1. T T (11. −t). t) −→ F0i (x − t) = F0i (x. −t)Aµ (x. t) − ∂ν Aµ (x.

235) . 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. (11. T (11. −t)∂ µ φ∗ (x. −t))φ(x.466 11 Discrete symmetries In general. For example. −t)) −→ i (φT ∗ (x)∂ µ φT (x) − (∂ µ φT ∗ (x))φT (x)) T = −i (φ∗ (x. −t))φ∗ (x. Furthermore. ∂t2 ∂2 − ∇2 + m2 φ∗ (x) = 0.231) (11. −t)∂ µ φ(x. ∂2 − ∇2 + m2 φT (x) ∂t2 ∂2 − ∇2 + m2 ηφ φ∗ (x.230) where the matrix T mixes the components of the function and |ηψ |2 = 1. Note that since the complex spin zero ﬁeld satisﬁes + m2 φ(x) = + m2 φ∗ (x) = it is clear that in this case + m2 φT (x) = = = 0. the Klein-Gordon equation is invariant under time inversion.232) That is. 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.233) T (11. ∂t2 (11.234) ∂2 − ∇2 + m2 φ(x) = 0. for a spin zero ﬁeld. −t)). −t) ∂t2 (11. −t). −t). we have φ(x) −→ φT (x) = ηφ φ∗ (x. −t) − (∂ µ φ∗ (x. −t) − (∂ µ φ(x.

∂(−t) ∂x (11.238) ∂ ∂ − iγ i∗ i − m ψ ∗ (x. −t). T (11. or. of course. ∂t ∂x . or. −t)∂ i φ(x. −t).3 Time reversal 467 Therefore. t) −→ −i φ∗ (x. the correct behavior of the current four-vector under time inversion as we have already seen in (11. −J i (x.236) J i (x. (Here ∂ we have used the fact that J 0 (x. t) = 0.11.) Thus we conclude that the spin zero charged particles interacting with photons can be described by a theory which will be time reversal invariant. −iγ 0∗ −iγ 0∗ iγ 0∗ ∂ ∂ − iγ i∗ i − m ψ ∗ (x.2 Dirac ﬁelds. −t). From our general discussion (see (11. we obtain J 0 (x. −t) − ( φ∗ (x. −t) ∂t ∂t J 0 (x. −t) = 0. −t) = which is. (11. or. −t) = T T ∂ ∂ φ(x. ∂t ∂x ∂ ∂ − iγ i∗ i − m ψ ∗ (x.3. −t) is deﬁned with − ∂t . t) −→ ψ T (x. we expect that under time reversal ψ(x. t) −→ −i φ∗ (x. −t) = 0. t) = ηψ T ψ ∗ (x. −t))φ(x. −t) − (∂ i φ∗ (x.197). −t))φ(x.230)). leads to (upon complex conjugation) (−iγ µ∗ ∂µ − m) ψ ∗ (x) = 0.239) (11.237) 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. 11.

T γ i∗ T −1 = −γ i .244) (11. −t) = 0. or.242) would satisfy the same Dirac equation as ψ(x. γ ˜ (11. we should have (iγ µ ∂µ − m) ψ T (x) = 0. or. (11. γ ν∗ } = 2η µν ½.243) γ 0 = γ 0. ∂t ∂x ∂ ∂ + iγ i i − m T ψ ∗ (x. γ ν } = 2η µν ½. if ψ T (x. ∂t ∂x (11. iγ 0 iγ 0 ∂ ∂ + iγ i i − m ψ T (x) = 0. and the matrices γ µ deﬁned as (˜ µ = γ µ† ) ˜ γ (11.245) .241) ψ T (x) = ηψ T ψ ∗ (x. t) were to be a solution of the Dirac equation. ˜ also satisfy the Cliﬀord algebra {˜ µ . then the time reversed wave function (11. We note that the matrices γ µ∗ satisfy the Cliﬀord algebra {γ µ∗ . ˜ γ i = −γ i .468 11 Discrete symmetries On the other hand.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).

They must also be related to each other through a similarity transformation. (11. CT −1 γ µ CT −1 −1 = −γ µ .122) that Cγ µT C −1 = −γ µ . ˜ leading to T γ 0∗ T −1 = γ 0 . (11. Let us also recall from (11.247) To determine the explicit form of T .248) Thus we see that the inverse relation can be written as T −1 γ µ T = γ µT .246) (11. the matrix CT −1 anticommutes with all the γ µ ’s and hence must be a multiple of γ5 .11. (11.250) CT −1 γ µ T C −1 = Cγ µT C −1 = −γ µ .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). Choosing . or. T = −γ i∗ = − γ i† = γ iT . Thus we can identify T γ µ∗ T −1 = γ µ .249) (11.251) In other words. T γ i∗ T −1 = −γ i . Combining the two results we obtain (11. let us invert these relations and write T T −1 γ 0 T T −1 γ i T = γ 0∗ = γ 0† = γ 0T .

the T matrix satisﬁes T = T † = T −1 . (11. and using the deﬁnition (11.256) so that the time inversion matrix in the spinor space is antisymmetric. −t). independent of any representation (the superscript denotes transposition) (T )T = −T. t) −→ ψ (x. (11. Let us next analyze the behavior of the current four-vector under time inversion . −t) = ηψ γ 0 T ψ (x. t) = ηψ T ψ ∗ (x. t) −→ ψ T (x. −t)T −1 γ 0 . Note that with this choice. T T T ∗ ψ(x. (11. Therefore we see that the time inversion transformations T ψ(x.253) (11.252) (11. we can show that in four space-time dimensions. we determine (in our representation) T = iγ 1 γ 3 .257) with (|ηψ |2 = 1) deﬁnes a symmetry of the Dirac equation.255) Furthermore.470 11 Discrete symmetries CT −1 = −γ5 = −iγ 0 γ 1 γ 2 γ 3 . t) = ηψ ψ T (x. −t) = T ηψ T γ 0T ψ (x.254) (11. as in the case of charge conjugation.130) C = γ 0γ 2.

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

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

Let us also note that if we are looking at a reaction of the form a + b −→ c + d. −t)σ ij ψ(x. t)σ 0i ψ(x. the tensor bilinears have precisely the same behavior under time reversal as the electromagnetic ﬁeld strength tensor F µν . We could have expected this from the deﬁnition (see (3. −t). namely.265) In other words. −t). It is also easy to see now that under time reversal. −mj .3 Consequences of T invariance. t) −→ −ψ(x. the chirality of a spinor does not change. j. 11.3 Time reversal 473 the helicity does not change. then the transition amplitude can be represented as (11. p. T = ηψ γ5 T ψ ∗ = ηψ T γ5 ψ ∗ † = ηψ T γ5 ψ ∗ γ5 ψ T = ηψ T (γ5 ψ)∗ . ψ(x. Let us recall that under time reversal. −t)σ 0i ψ(x.121)) s·p s · p T (−s) · (−p) −→ = = h. mj −→ (i)2mj |α. t)σ ij ψ(x. |p| |p| |p| h= (11.222)) T |α. (11.267) where α stands for all the other quantum numbers of the state.3.264) We can also show that ψ(x. −p.268) . T T (11. a state with given angular momentum transforms as (see (11.11. (11. t) −→ ψ(x. j.266) so that a left-handed spinor remains left-handed under time reversal and a right-handed spinor remains right-handed.

474 11 Discrete symmetries f |S|i .275) . where we are still using the same notation. On the other hand. |i = |a. d . This is known as the principle of detailed balance and has been used.274) (11.272) (11. and c + d −→ a + b. S † = S. b . (11. |f for the ﬁnal state and S denotes the S-matrix or the time evolution matrix. it does not follow a priori that | i|S|f |2 = | f |S|i |2 . (11. (11. experimentally this is seen to hold in many processes.273) On the other hand.270) Since the time evolution operator is not Hermitian. to obtain the spin of π-meson etc. |f = |c.271) (11. then the transition amplitude can be represented as i|S|f .269) where |i stands for the initial state. if we are looking at the reciprocal reaction c + d −→ a + b. (11. namely. It is worth emphasizing here that the detailed balance does not say that the rates for the two reactions a + b −→ c + d. namely. for example.

pi . the principle of detailed balance will hold. then we have | f |S|i | = 2π| f |Hint |i |δ (Ei − Ef ) = 2π| i|Hint |f |δ (Ei − Ef ) = | i|S|f |. Let us next assume that we are dealing with a system whose Hamiltonian is invariant under time reversal. Let us assume that we are dealing with a system where the interaction Hamiltonian can be treated as a perturbation. we can write (for distinct |i and |f states) f |S|i = −2πi f |Hint |i δ (Ei − Ef ) . That is.276) where ρf denotes the density of states for the ﬁnal state which may not be the same for the two processes. if the interaction Hamiltonian is Hermitian. mf .279) Let us choose the initial and the ﬁnal states to correspond to |i |f = |α. when the principle of detailed balance may hold. mi . = |β. from Fermi’s Golden rule we know that the rate for a process |i → |f is given by Wi→f ∝ 1 | f |S|i |2 ρf . In fact. therefore. (11. h (11.278) In this case. T ST † = S. (11. theoretically.11. (11. pf . (11.277) and furthermore.280) . In this case. It follows then (in the operator formulation) that the evolution operator or the S-matrix must also be invariant under time inversion. Let us next try to understand.3 Time reversal 475 have to be the same.

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

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

whereas the interaction involving the magnetic dipole moment is invariant under P and T . the electric dipole moment would be parallel to its spin. if electromagnetic interactions are invariant under P and T . B −→ −B. (11. so that we have σ · E −→ −σ · E. . Consequently.290) We have already seen that under parity and time reversal E −→ −E. (11. the electric dipole interaction changes sign under each one of the transformations. this is the Pauli coupling) Hint = µe σ · E + µm σ · B.289) Let us also note here that the only natural dimension associated with the neutron is its size which is of the order of 1F = 10−13 cm.63 × 10−25 e-cm. (µe )exp < 0. Let us assume that it also has an electric dipole moment. (11. Experiments have obtained extremely small limits on µe . T T (11. σ · B −→ σ · B. P P P P P E −→ E. then the electric dipole moment of the neutron must vanish.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. Since there is no natural axis for the neutron other than the spin axis. σ · B −→ σ · B.478 11 Discrete symmetries a magnetic dipole moment. σ −→ −σ.293) This basically measures the extent to which P and T invariances hold in electromagnetic interactions. σ −→ σ. Thus the neutron would interact with an electromagnetic ﬁeld through the interaction Hamiltonian (there is no minimal coupling.291) σ · E −→ −σ · E. T T T (11. B −→ B.

Rather than going into the technical proof of this theorem.297) . C and T in any order even though the individual transformations may not be a symmetry of the system. m. The operation of CP. m. of course. We can naturally ask whether there exists any combination of these symmetries which will be a symmetry of the weak Hamiltonian. (11. CPT (11.294) where m denotes the mass of the particle. m. s α. α denote the parameters associated with the antiparticle (in this section. m. m. m.4 CPT theorem 479 11. m. m. −s|(CPT )H(CPT )−1 |α. s = m|α. Let us consider the state of a particle at rest which satisﬁes H|α. (11.295) where m. m. −s = m. −s .11. Thus under the CPT transformation. 11. m. m. −s|. (11. s its spin and α denotes the other quantum numbers. m. then it follows that m = α.4 CPT theorem We have seen that some of the discrete symmetries appear to be violated in weak interactions. s|H|α. 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.1 Equality of mass for particles and antiparticles. quantities with bars would refer to antiparticle parameters).296) If the Hamiltonian is CPT invariant. s .4. let us discuss the obvious consequences of this theorem. takes a particle to the antiparticle state. s −→ η α. Thus we see that m = = = α. |α. −s|H|α. s|(CPT )−1 (CPT )H(CPT )−1 (CPT )|α. s α.

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

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

11. mj |H+ |α. j. and τ ) of particles and antiparticles. mj |2 ρβ | β. j. mj |2 ρβ (11. j. we have obtained | β. There are many other interesting consequences of the CPT theorem including the connection between spin and statistics which we cannot go into. j. j.482 11 Discrete symmetries Since the Hamiltonians H± are Hermitian we also have β. j.308) Here we have neglected an overall phase factor (depending on α. β) which is not relevant for the subsequent discussion. j. we note here that the individual decay channels may have diﬀerent life times for particle and anti-particle. This follows trivially from the equality of particle and antiparticle masses. j. From CPT invariance. 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. j. mj |H± |α. 984 (1958). However. mj |H− |α. (11.5 References 1. j. mj |H− |α. mj |2 + | β. mj |H± |α. j. mj ∗ . therefore. Physical Review 109. 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. mj |2 + | β. namely 2π 1 = τ = 2π β β | β. mj |2 .310) = 1 . mj |H± |α. j. it follows that the weak interactions must violate T -invariance by a small amount as well so that CPT will hold. (11. Okubo. j.309) which leads to a relation between the total life times (τ . j. j. mj = β. mj |2 = | β. S. mj |H+ |α. j. . mj |H± |α.

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

This page intentionally left blank .

We will now study gauge theories based on more complicated symmetries. δψ δψ = = where θ is a real constant parameter and inﬁnitesimally.3) . the Maxwell theory which is based on the Abelian gauge group U (1).37) and (8.1) where ψ provides a representation of U (1) (namely. iǫψ. we have talked about the simplest of gauge theories. it is a complex ﬁeld). we have −iǫψ. let us quickly review how the gauge ﬁelds come into the theory in the case of quantum electrodynamics (QED). this Lagrangian density has the global symmetry ψ → e−iθ ψ. (12. 485 (12.9) / L = iψ∂ ψ − mψψ. As we have seen in (8.Chapter 12 Yang-Mills theory 12. (12. namely.1 Non-Abelian gauge theories So far. 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.2) ψ → ψeiθ . Let us start with the Dirac Lagrangian density for a free fermion (8.38).

486 12 Yang-Mills theory with θ = ǫ = inﬁnitesimal. (12. (12. Neither the Lagrangian density (12.1) whose variation would cancel the ∂µ ǫ(x) term. the phase of ψ remains arbitrary. As we have seen earlier (see (9.4) since ∂µ ψ does not transform covariantly and there is no term in the Lagrangian density (12. If we now want to promote the symmetry to be local.1) nor the corresponding action is invariant under the local phase transformation in (12. namely. Consequently.4) = −ǫ(x)ψ∂ ψ + ψγ µ ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ / = (∂µ ǫ(x))ψγ µ ψ. δ (Dµ ψ) = −iǫ(x)Dµ ψ(x). we have δL = iδψ∂ ψ + iψγ µ ∂µ δψ − mδψψ − mψδψ / −imǫ(x)ψψ + imǫ(x)ψψ.81)).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). 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. Consequently. (12. δ (∂µ ψ) = ∂µ (δψ) = ∂µ (−iǫ(x)ψ) = −i ((∂µ ǫ(x)) + ǫ(x)∂µ ) ψ(x). the way out of this diﬃculty is to deﬁne a covariant derivative Dµ ψ such that it transforms covariantly under the local transformation. namely ǫ = ǫ(x) (or θ = θ(x)).5) (12.7) . δψ = iǫ(x)ψ. we note that δψ = −iǫ(x)ψ.

9) is invariant under the inﬁnitesimal local gauge transformations δψ = −iǫ(x)ψ(x). That is. is nondynamical in this theory. we note that since the gauge ﬁeld transforms longitudinally (by a gradient).86).13) . (12. e (12.10). the curl of the ﬁeld would be unchanged by such a transformation. if we deﬁne a ﬁeld strength tensor Fµν = ∂µ Aν − ∂ν Aµ = −Fνµ . the ﬁeld Aµ known as the gauge ﬁeld.8) We recognize (12. e δAµ (x) = (12. δAµ = 1 ∂µ ǫ(x). this would transform as δFµν = ∂µ δAν − ∂ν δAµ = 1 (∂µ ∂ν ǫ − ∂ν ∂µ ǫ) = 0.10) However.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. δψ = iǫ(x)ψ(x). (12.11) under the gauge transformation (12.12) Hence the ﬁeld strength tensor is gauge invariant and the gauge invariant Lagrangian density for the dynamical part of the gauge ﬁeld can be written as (quadratic in derivatives) 1 Lgauge = − Fµν F µν . e (12. 4 (12.7) as describing the minimal coupling of the charged fermions to the electromagnetic ﬁeld and as we have seen in (9. the Lagrangian density L = iψγ µ Dµ ψ(x) − mψψ.12.

a δψk = −iǫa Tkℓ ψℓ . This Lagrangian density is invariant under the global phase transformations aT a ψk → ψk → Uψ ψU † k = e−iθ ψ k. (12.80)) 1 LQED = − Fµν F µν + iψγ µ Dµ ψ − mψψ. we assume that ψk belongs to a nontrivial representation R of some internal symmetry group G and dim R denotes the dimensionality of the representation. (12. aT a k = ψU −1 = ψeiθ k . .15) where k is an internal symmetry index. · · · . dim G.16) where θ a denotes the real global parameter of transformation and inﬁnitesimally we have δψ k = iǫa ψ ℓ (T a )ℓk . / k = 1.17) with θ a = ǫa = inﬁnitesimal. . . dim R. 2.10). . . Let us now generalize these ideas to theories with more complicated symmetries. 2. a = 1. dim G represents the dimensionality of the symmetry group. Furthermore. 2.488 12 Yang-Mills theory The total Lagrangian density for QED is then given by (see also (9. dim G. . Namely. Here the T a ’s represent the generators of the internal symmetry group G and are assumed to be Hermitian. 4 (12. Let us consider a free Dirac theory described by the Lagrangian density (containing several complex fermion ﬁelds) L = iψ k ∂ ψk − mψ k ψk .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) . k a = 1. (We note here parenthetically that dim G = n2 − 1 for the group SU (n). . (12.

namely. c = 1. It is easy to see that the Lagrangian density is invariant under the transformations (12. (12.20) Once again. For example. a δψ k = iǫa (x)ψ ℓ Tℓk . namely. let us consider the inﬁnitesimal local phase transformations a δψk = −iǫa (x)Tkℓ ψℓ . a. 2. b. δL = δψ k (i∂ − m)ψk + ψ k (i∂ − m)δψk / / = iǫa ψ ℓ T a ℓk where the real constants f abc which are completely antisymmetric (antisymmetric under the exchange of any pair of indices) are known as the structure constants of the symmetry group. Let us now try to make this symmetry a local symmetry of our theory. . b. c = 1. the structure constants are nontrivial).1 Non-Abelian gauge theories 489 T a . f 147 = f 246 = f 257 = f 345 = 2 . f 156 = f 367 = √ a / (i∂ − m)ψk − iǫa ψ k (i∂ − m)Tkℓ ψℓ / a / / = iǫa ψ k (T a )kℓ (i∂ − m)ψℓ − iǫa ψ k (i∂ − m)Tkℓ ψℓ = 0. 2. That is. .12. we already know from the study of angular momenta.18) − 1 .) When the generators of the symmetry group 2 do not commute (namely. . .16)). (12.17) (or (12. dim G. a. . T b = if abc T c . f 458 = f 678 = 23 . (For semi-simple groups the structure constants can always be chosen to be completely antisymmetric. the symmetry group is known as a non-Abelian group. (12. we note that the ordinary derivative acting on the fermion ﬁeld does not transform covariantly under this transformation. the structure constants of the group SU (2) can be identiﬁed with f abc = ǫabc . (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. 3 while for the group SU (3) the structure constants are more complicated and the nontrivial ones 1 are given by f 123 = 1.19) Here we have used the fact that the symmetry generators are Hermitian.

We can compare this with QED where the generator is the identity matrix ½ which commutes with every operator. we write the covariant derivative (in the matrix notation) as Dµ ψ = (∂µ + igAµ ) ψ.24) explicitly as a a Dµ ψk = ∂µ ψk + igTkℓ Aa ψℓ = ∂µ δkℓ + igTkℓ Aa ψℓ .25) This is the generalization of the minimal coupling (in QED) to the present case and the T a ’s are the generators of the symmetry group belonging to the same representation as the fermions.23) Introducing a new ﬁeld (gauge ﬁeld). µ µ (12.21) so that under the local transformation (12.20).22). there is no other term in the Lagrangian density whose variation would cancel the ∂µ ǫa (x) term in (12.490 12 Yang-Mills theory a δ (∂µ ψk ) = ∂µ (−iǫa (x)Tkℓ ψℓ ) a a = −i (∂µ ǫa (x)) Tkℓ ψℓ − iǫa (x)Tkℓ ∂µ ψℓ . (12. Hence we deﬁne a covariant derivative (Dµ ψ)k such that under an inﬁnitesimal local transformation a δ (Dµ ψ)k = −iǫa (x)Tkℓ (Dµ ψ)ℓ . We can also write the covariant derivative in (12. the dimension of the Lie group.26) . µ (12. (12. We note that the number of gauge ﬁelds (that are needed to deﬁne a covariant derivative) is the same as dim G.24) (12. / / δL = δψ k (i∂ − m)ψk + ψ k (i∂ − m)δψk = (∂µ ǫa (x))ψ k γ µ (T a )kℓ ψℓ . where g denotes the appropriate coupling constant and Aµ = T a Aa .22) As in the case of the U (1) theory. (12.

28) can be written in the form a a a igTkℓ δ Aa ψℓ = igTkℓ δAa ψℓ + igTkℓ Aa δψℓ µ µ µ a a b = igTkℓ δAa ψℓ + igTkℓ Aa − iǫb Tℓm ψm µ µ a b a = igTkℓ δAa ψℓ + gǫa Tkℓ Ab Tℓm ψm . a δ (Dµ ψ)k = −iǫa Tkℓ (Dµ ψ)ℓ .26) that we should have a igTkℓ δ Aa ψℓ = δ(Dµ ψk ) − ∂µ δψk µ a = −iǫa Tkℓ (Dµ ψ)ℓ − ∂µ δψk a b a = −iǫa Tkℓ ∂µ ψℓ + igTℓm Ab ψm − ∂µ (−iǫa Tkℓ ψℓ ) µ a b = −iǫa Tkℓ ∂µ ψℓ + igTℓm Ab ψm µ a a +i (∂µ ǫa ) Tkℓ ψℓ + iǫa Tkℓ ∂µ ψℓ a b a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa Tkℓ Tℓm Ab ψm .30) a = i (∂µ ǫa ) Tkℓ ψℓ + igǫa f abc (T c )km Ab ψm .1 Non-Abelian gauge theories 491 Noting that we would like the covariant derivative to transform covariantly under an inﬁnitesimal local transformation. (12.28).29) Substituting this into (12. we obtain a igTkℓ δAa ψℓ µ a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a T b km Ab ψm µ −gǫa T b T a km Ab ψm µ km a = i (∂µ ǫa ) Tkℓ ψℓ + gǫa T a .28) We note that the left-hand side in (12. µ (12. T b Ab ψm µ (12.27) we conclude from (12. µ This can also be written as . µ µ (12.12.

a δψ k = iǫa (x)ψ ℓ Tℓk .16)) . we can write the Lagrangian density in (12. dim G. We can now deﬁne a unitary representation of the group as describing the ﬁnite symmetry transformation matrix (see (12. µ a = 1. (12.25)) Aµ = T a Aa . .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 . the Lagrangian density is a scalar) L = iψγ µ Dµ ψ − mψψ. µ g (12.33) in terms of matrices as (of course.33) with the covariant derivative deﬁned in (12. 2. . µ g (12.35) where the gauge ﬁelds Aµ are matrices belonging to the same representation of the group as the fermions (since the generators T a belong to this representation). . µ µ g g (12. .26) is invariant under the inﬁnitesimal local transformations a δψk = −iǫa (x)Tkℓ ψℓ (x).32) Thus the Lagrangian density L = iψ k γ µ (Dµ ψ)k − mψ k ψk .34) If we use the deﬁnition (see (12.492 12 Yang-Mills theory a igTkℓ ψℓ δAa − µ 1 ∂µ ǫa + f abc Ab ǫc = 0. (12. (12. δAa = µ 1 ∂µ ǫa − f abc Ab ǫc .36) with normal product rules for matrices.

ψ → ψU −1 . the tensor representing the Abelian ﬁeld strength (see (12.37) so that we can write the ﬁnite transformations (corresponding to (12.39) . U † = U −1 .12.34)) for the ﬁeld variables in the matrix form as ψ → U ψ. Aµ → U Aµ U −1 − 1 (∂µ U ) U −1 .1 Non-Abelian gauge theories 493 U = e−iθ a (x)T a . (12.38) To construct the Lagrangian density for the dynamical part of the gauge ﬁeld in a gauge invariant manner.11)) would transform as fµν = ∂µ Aν − ∂ν Aµ 1 (∂ν U ) U −1 ig 1 (∂µ U ) U −1 ig 1 1 (∂µ ∂ν U ) U −1 − (∂ν U ) ∂µ U −1 ig ig 1 1 (∂ν ∂µ U ) U −1 + (∂µ U ) ∂ν U −1 ig ig → ∂µ U Aν U −1 − −∂ν U Aµ U −1 − = ∂µ U Aν U −1 − −∂ν U Aµ U −1 + = 1 1 (∂µ U ) ∂ν U −1 − (∂ν U ) ∂µ U −1 + ∂µ U Aν U −1 ig ig +U ∂µ Aν U −1 + U Aν ∂µ U −1 − ∂ν U Aµ U −1 −U ∂ν Aµ U −1 − U Aµ ∂ν U −1 1 ∂µ U ∂ν U −1 − ∂ν U ∂µ U −1 + ∂µ U Aν U −1 ig = +U Aν ∂µ U −1 − ∂ν U Aµ U −1 − U Aµ ∂ν U −1 +U (∂µ Aν − ∂ν Aµ ) U −1 .38). ig (12. (12. we note that under the gauge transformation (12.

Fµν → U Fµν U −1 . we see that unlike in the case of QED. 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.40). in the present case. Aν ] . it is clear that.41) (12.40) Thus comparing (12. (12.38). Let us also note that under the gauge transformation ig [Aµ .39) and (12. (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 . if we deﬁne the ﬁeld strength tensor as Fµν = ∂µ Aν − ∂ν Aµ + ig [Aµ .42) so that Fµν transforms covariantly. here fµν = ∂µ Aν − ∂ν Aµ is neither invariant nor does it have a simple transformation under the gauge transformation (12. It is now easy to construct the gauge invariant Lagrangian density for the dynamical part of the gauge ﬁeld as (quadratic in derivatives) .38). then under the gauge transformation (12.

1 Non-Abelian gauge theories 495 1 1 a Lgauge = − Tr Fµν F µν = − Fµν F µν a . This Lagrangian density is invariant under the non-Abelian gauge symmetry transformations (12. T c ν µ µ ν = ∂µ Aa − ∂ν Aa T a − gf abc Ab Ac T a .34)) . a a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac = −Fνµ . therefore. a Fµν T a = ∂µ Aa − ∂ν Aa T a + igAb Ac T b .45) The complete Lagrangian density for QCD (quantum chromodynamics.38). (12. Aν . µ ν ν µ (12. given by 1 a LQCD = − Fµν F µν a + iψ k γ µ (Dµ ψ)k − mψ k ψk . see (12.12. 2 4 (12.38) which have the inﬁnitesimal form (see (12.41) is easily seen to take the form.59) and the comments fol2 lowing that equation). the ﬁeld strength tensor (12. Fµν = ∂µ Aν − ∂ν Aµ + ig Aµ .43) where a particular normalization for the trace of the generators is assumed (namely. 4 (12. more generally for fermions interacting with a non-Abelian gauge ﬁeld. In components.46) where we can set m = 0 if the fermions (quarks) are massless. ν µ µ ν or. or. T a T b = 1 δab .44) so that the gauge invariant Lagrangian density for the gauge ﬁeld can also be written in terms of component ﬁelds as 1 a Lgauge = − Fµν F µν a 4 1 = − ∂µ Aa − ∂ν Aa − gf abc Ab Ac ν µ µ ν 4 × ∂ µ Aνa − ∂ ν Aµa − gf apq Aµp Aνq . From the cyclicity of trace this Lagrangian density is easily seen to be invariant under the gauge transformations (12. QCD corresponds to the particular case when G = SU (3)) is.

(12.49) imposes a restriction on the structure constants of the group of the form if abp T p . has the form T a . T a . T a = 0.50) From the structure of the Lie algebra we know that we can write the ﬁnite group elements in a unitary representation as (recall ﬁnite rotations and the angular momentum operators and see (12. (12. as we have noted in (12. T b + if bcp T p .37) as well) U (x) = eiθ a (x)T a . or. T c + T c.496 12 Yang-Mills theory δψ k = iǫa (x)ψ ℓ (T a )ℓk .49) Using (12.18).47) Let us discuss brieﬂy some of the properties of the Lie algebra of the symmetry group G. T c . f abp f pcq + f cap f pbq + f bcp f paq = 0. The algebra of the Hermitian generators. T b + T b . T c + if cap T p . T b = if abc T c . In this case . (12. µ g (12.48) and the Jacobi identity associated with this algebra is given by T a. f abp f pcq T q + f cap f pbq T q + f bcpf paq T q = 0. or. 1 ∂µ ǫa − f abc Ab ǫc .48) it can be seen that the Jacobi identity (12. T a = 0. (12. δAa = µ δψk = −iǫa (x)(T a )kℓ ψℓ . then the Lie algebra breaks up into two commuting non-Abelian subalgebras. T b .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.

1 Non-Abelian gauge theories 497 the group G is a direct product of two independent non-Abelian Lie groups.12.52) We note that C2 is a normalization constant which determines the values of the structure constants.18) we obtain habc = Tr if abp T p T c = if abp Tr T p T c = if abp Kp δpc = iKc f abc . a = b. In all our discussions. is completely antisymmetric in all its indices. It depends on the representation but not on the indices a and b.54) Tr Tr T a T b T c − Tr T b T a T c . using the commutation relation (12. Furthermore. (12. Tr T a T b = (12. let us next note that the quantity habc = = T a. On the other hand. we will assume that the symmetry group G is simple.55) where the index c is ﬁxed (and not summed). A non-Abelian Lie group that cannot be so factorized is called a simple Lie group. A direct product of simple Lie groups is called semi-simple. 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.53) Using the cyclicity property of the trace. (12. For any representation of a simple Lie group we can write Tr T a T b = C2 δab . we note that . T b T c (12.

dim R.498 12 Yang-Mills theory hacb = Tr T a. · · · . 2.56) with the index b ﬁxed.55) and (12.61) . Furthermore. dim G. (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 (12.59) where the constant C2 depends only on the representation. since habc is completely antisymmetric. we have (T a T a )mn = C(R) δmn . This shows that we can write Tr (T a T b ) = C2 δab . (12. (12. we note here that it is (12. we have hacb = −habc . a. Similarly. T c T b = if acp Tr T p T b = if acp Kp δpb = iKb f acb = −iKb f abc . · · · .60) then T (R) is known as the index of the representation R.43). b = 1. (12.57) Using this and comparing (12. 2. m.58) (12.) We note that if we write Tr T a T b = T (R) δab . However.59) which is used to show that the structure constants for a semi-simple group are completely antisymmetric in the indices. n = 1. (12. (12.56) we conclude that Kb = Kc = K.

a particular representation that is very important as well as useful is given by Ta = −if abc . 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). As a result.62) We can determine the generators of the group in various representations much like in the case of angular momentum. (12.64) Furthermore. 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. (12.65) where we have used the anti-symmetry of the structure constants as well as (12. However. T a.12. (12.63) bc This is consistent with the hermiticity requirement for the generators. we conclude that the identiﬁcation . we can easily check that this representation satisﬁes the Lie algebra. 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 . namely. (12.

67) so that we can equivalently write (see.68) which shows that the gauge ﬁeld. This can also be seen from the transformation of the ﬁeld strength tensor in (12.66) indeed deﬁnes a representation of the Lie algebra known as the adjoint representation. Aµ .42).500 12 Yang-Mills theory a T(adj) bc = −if abc .69) . T b . (12. (12. (12. for example.34) δAa = µ = = = 1 ∂µ ǫa − gf abc Ab ǫc µ g 1 ∂µ ǫa + gf bac Ab ǫc µ g 1 ∂µ ǫa + ig − if bac Ab ǫc µ g 1 b ∂µ ǫa + ig T(adj) g ac Ab ǫc . which inﬁnitesimally has the form Fµν → U Fµν U −1 so that = Fµν + iǫb Fµν .26)) δAa = µ 1 (adj) a D θ . Let us next look at the transformation for the gauge ﬁelds in (12. µ (12. g µ (12. transforms according to the adjoint representation of the group.

(12.72) where Aµ = Aa T a and ǫ = ǫa T a (with T a in any representation). ǫ].) For completeness.67) (adj) Dµ ǫ a = ∂µ ǫa − gf abc Ab ǫc . (12. or. For example. let us scale the ﬁeld variables as . or. a b δFµν = −f abc Fµν ǫc b = −i − if cab ǫc Fµν c = −i T(adj) ab b ǫc Fµν c = −iǫc T(adj) Fb. T b a = iǫb Fµν T a .70) Comparing this with the transformation of the fermions in (12. ab µν (12.34) we conclude that the ﬁeld strength Fµν as well as the gauge ﬁeld Aµ transform according to the adjoint representation of the group. the gauge ﬁeld must transform in the adjoint representation.1 Non-Abelian gauge theories 501 δFµν = iǫb Fµν .73) Let us note the following features of this Lagrangian density. T b .12. µ Let us now concentrate only on the gauge ﬁeld (Yang-Mills) part of the Lagrangian density 1 a LYM = − Fµν F µν a 4 1 = − ∂µ Aa − ∂ν Aa − gf abc Ab Ac ν µ µ ν 4 × ∂ µ Aνa − ∂ ν Aµa − gf apq Aµp Aνq . µ (12. a a δFµν T a = iǫb Fµν if abc T c . let us note here that the covariant derivative in the adjoint representation (12.71) can also be written in the matrix form as (adj) Dµ ǫ = ∂µ ǫ + ig[Aµ . (It does not matter what representation the matter ﬁelds belong to.

here the gauge ﬁelds have self interaction (they interact with themselves) and. (12.75) LYM = − Tr Fµν F µν = − Fµν F µν a . in the case of non-Abelian symmetry they must couple to themselves and possess self interactions. therefore. Aν ].44)) Fµν a Fµν = ∂µ Aa − ∂ν Aa − gf abc Ab Ac . We note that the Lagrangian density for the Yang-Mills theory is given by (12.43) (or (12.41) and (12. the gauge ﬁelds carry the charge of the non-Abelian symmetry group (they have a nontrivial symmetry index) in contrast to the photon ﬁeld which is chargeless.45)) 1 a 1 (12. (12. Physically we understand this in the following way.76) . The other feature to note is that unlike the photon ﬁeld. 2 4 where the ﬁeld strength tensor is given by (see (12. ν µ µ ν g which leads to LYM → − 1 ∂µ Aa − ∂ν Aa − f abc Ab Ac ν µ µ ν 4g2 × ∂ µ Aνa − ∂ ν Aµa − f apq Aµp Aνq . the pure Yang-Mills theory is an interacting theory unlike the Maxwell theory. In the present case. 12. ν µ µ ν = ∂µ Aν − ∂ν Aµ + ig[Aµ . Since the gauge ﬁelds couple to any ﬁeld (particle) carrying charge of the symmetry group.502 12 Yang-Mills theory Aa → µ 1 a A . the coupling constant can be scaled out and written as an overall multiplicative factor in the Lagrangian density. g µ It follows then that 1 a Fµν → ∂µ Aa − ∂ν Aa − f abc Ab Ac .2 Canonical quantization of Yang-Mills theory Let us discuss the canonical quantization of Yang-Mills theory in the same spirit as the discussion of Maxwell’s theory in chapter 9.74) Thus we note that in this theory.

we can identify the non-Abelian electric and magnetic ﬁelds from (12. F µν ] = 0. µ (12.75) is the transverse projection operator just like in the case of Maxwell’s theory. 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) as 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a .75) and have the form ∂L ∂L − = 0.68)). −∂µ F µν a − −gf abc Ab F µν c = 0. Therefore.75) can also be written explicitly in terms of the non-Abelian electric and magnetic ﬁeld strength tensors (12. or.68) and we are suppressing the symbol “(adj)” to denote the covariant derivative in the adjoint representation for simplicity) .72)).80) Let us note that the coeﬃcient matrix of highest derivatives in (12. we can write the Euler-Lagrange equations as Dµ F µν = ∂µ F µν + ig[Aµ .2 Canonical quantization of Yang-Mills theory 503 As in the case of the Abelian theory (Maxwell theory).78) The Euler-Lagrange equation for the theory are obtained from (12. In the matrix notation (see (12.76) as a a F0i = ∂0 Aa − ∂i Aa − gf abc Ab Ac = Ei .77) The Lagrangian density (12. (12.79) where the covariant derivative is deﬁned to be in the adjoint representation of the group (see (12.67) and (12. j i i j (12.12. 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. ∂∂µ Aa ∂Aa ν ν ∂µ or.

81) In fact. Namely. The momentum conjugate to Aa does not exist. we have Π0 a (x) = −F 00 a (x) = 0.77). we choose Aa (x) = 0. This implies that 0 Aa is like a c-number quantity which commutes with every other 0 operator in the theory. Let us obtain the momenta canonically conjugate to the ﬁeld variables Aa from (12. Thus we can choose a gauge condition and set it equal to zero. as in the case of Maxwell’s theory and a Πi a (x) = −F 0i a = Ei (x). Di F i0 a = 0. we see that we have only 3N canonical momenta where N = dim G.83) a Noting that the ﬁeld strength Fµν is antisymmetric in the indices µ. namely. requiring that . (12.79) for ν = 0. ν.75) and this leads to µ Πµ a (x) = ∂L = −F 0µ a (x). (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.82) which is a constraint relation. Thus. ˙ a (x) ∂ Aµ (12.504 12 Yang-Mills theory δAa = µ 1 (Dµ ǫ(x))a . 0 or. the constrained structure of the theory is already obvious in the Euler-Lagrange equation of motion (12. A0 (x) = 0. For example.85) a where the non-Abelian electric ﬁeld Ei (x) is deﬁned in (12. (12. g (12.84) (12.

86) the Lagrangian density (12. However. 1 (∂0 U ) U −1 = 0. let us also note that the potential term in (12.2 Canonical quantization of Yang-Mills theory 505 A′ = U A0 U −1 − 0 or. U −1 ∂0 U = igA0 . we expect the physical dynamical degrees of freedom of a massless gauge ﬁeld to be transverse.91) ig ig so that there is a cyclic variable in the theory that needs to be separated. 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 . On the other hand.90) so that neither the gauge ﬁeld Ai nor the electric ﬁeld Ei is transverse ˙ (recall that Ei = Ai ). (12. (12. Ei ] = 0. ˙ Ei = Ai . in the present case. (12.88) (namely. (12.88) 2 2 4 LYM = Tr Ei Ei − where we have used the fact that in the gauge (12.75) takes the form 1 1 a a 1 a Fij F ij = Ei Ei − Fij F ij a 2 2 4 1 ˙ a ˙ a 1 a ij a 1 ˙ ˙ = Tr Ai Ai − Fij F ij = Ai Ai − Fij F .89) From our discussion in the case of the Maxwell theory.12.82) that Di Ei = ∂i Ei + ig[Ai .87) we can determine the parameter of transformation and we see that a gauge ﬁeld can always be transformed to satisfy (12. We note that the relation (12.91) can be inverted to give . ig (12. Let us note that with the gauge condition (12. i or.86) the non-Abelian electric ﬁeld in (12. we note from (12.77) takes the form a ˙ Ei = Aa .86).

We note next from the ﬁeld redeﬁnition in (12.72)) Di (A)G = ∂i G + ig[Ai .92) ig ig Keeping in mind that we would like the dynamical variables to be ¯ transverse. ¯ ∂i Ai = 0. 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).91). (12. (12. ig (12.95) This shows that if we can ﬁnd a gauge transformation which satisﬁes ¯ (12. G]. Imposing this condition in (12.506 12 Yang-Mills theory 1 1 ¯ Ai = S −1 Ai S + S −1 ∂i S = S −1 Ai + (∂i S)S −1 S.91) that . namely.94) where we have used (12.93).93) 1 1 (∂i S) (∂i S −1 ) − ∂i (∂i S)S −1 ig ig 1 ∂i (∂i S)S −1 − [Ai . let us require the new ﬁeld variable Ai to be transverse. then we can deﬁne a new ﬁeld variable Ai satisfying the transversality condition (12.92) in the intermediate step and the covariant derivative in the adjoint representation is deﬁned for a matrix function G as (see (12. (∂i S)S −1 ] ig 1 Di (A) (∂i S)S −1 .

100) .12. (12.86) we can write 1 ˙ ˙ ¯ ¯ Ei = Ai = S Ai − Di (A)f S −1 = SE i S −1 . Here Di (A) is the covariant derivative in the adjoint representation (see (12. ig (12. ig (12. where (12.98) (12.97) From (12.96) we note that in the gauge (12.2 Canonical quantization of Yang-Mills theory 507 1 ˙ ¯ Ai = ∂0 S Ai S −1 − (∂i S)S −1 ig ˙ ˙¯ ¯ ¯ ˙ = S Ai S −1 + S Ai S −1 − S Ai S −1 SS −1 − 1 ˙ ˙ (∂i S)S −1 − (∂i S)S −1 SS −1 ig 1 ˙ ¯ ¯ ˙ ˙ = S Ai − [Ai . ig where we can identify 1 ˙ ¯ ¯ E i = Ai − Di (A)f.95)) with respect to ¯ the gauge ﬁeld Ai and we have identiﬁed ˙ f = S −1 S.96) ˙ ¯ where we have used ∂0 S −1 = −S −1 SS −1 .99) Furthermore. 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 . decomposing E i into its transverse and longitudinal components as T E i = E i + ∂i φ.

T ∂i E i = ∂i ∂i φ = ∇2 φ. φ] = −ig[Ai . Ei ] ¯ = ∂i SE i S −1 + ig[S(Ai − 1 −1 S (∂i S))S −1 .101) we conclude from the deﬁnition in (12. Di (A)Ei = ∂i Ei + ig[Ai .99) that 1 1 ¯ ∂i Di (A)f.91) and (12.90) takes the form ¯ ¯ Di (A)E i = ∂i E i + ig[Ai . or.102) Let us next note that under the ﬁeld redeﬁnition (12. As a result of this relation. E i + ∂i φ ] = 0.104) T ¯ where we have used ∂i Ai = 0 in the intermediate step.103) where we have used (12. 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). ig ∇2 ˙ ¯T ¯ Ei = Ai . E i ] = 0. E i ] = 0. ¯ ∂i Di (A)φ = ρ. φ=− (12. (12.98). or.91). E i ]S −1 ig ¯ = S(∂i E i + ig[Ai . (12. E i ] = ρ. We can invert this relation to determine . (12. the constraint equation (12. or. T ¯ ¯ ∂i ∂i φ + ig∂i [Ai .508 12 Yang-Mills theory ∂i E i = 0. φ] + ig[Ai . T ¯ ∂i ∂i φ + ig[Ai . E i ])S −1 ¯ = S(Di (A)E i )S −1 . or. ¯ ¯ ∂i ∂i φ + ig[Ai .

∂i Di (A) 1 ¯ ρ.2 Canonical quantization of Yang-Mills theory 509 φ = 1 ¯ ρ.106) as well as the cyclicity of trace. ∂j Dj (A) (12.12. ∂i Di (A) = Di (A)∂i which has been used together with integration by parts). ¯ 2 ∂j Dj (A) (12. ¯ We have isolated the dynamical variables of the theory to be Aa i which are transverse and the conjugate momenta can be determined from (12.100) we obtain T ˙ ¯ E i = E i + ∂i φ = Ai + ∂i (12. the Lagrangian density (12.108) .107) where we have neglected total divergence terms in the intermediate ¯ ¯ steps (note also that because of (12. i ˙ ¯a ∂A i (12.91).98) and the decomposition in (12.107) to be Π ia = ∂LYM ˙ ¯ = Aa .88) takes the form LYM = Tr Ei Ei − 1 Fij (A)F ij (A) 2 1 ¯ ¯ = Tr E i E i − Fij (A)F ij (A) 2 1 1 ˙ ˙ ¯ ¯ = Tr (Ai + ∂i ¯ ρ)(Ai + ∂i ∂k Dk (A) ρ) ¯ ∂j Dj (A) 1 ¯ ¯ − Fij (A)F ij (A) 2 1 ˙ ˙ ¯ ¯ ¯ ¯ = Tr Ai Ai − Fij (A)F ij (A) 2 1 1 2 −ρ ¯ ∇ ∂k Dk (A) ρ ¯ ∂j Dj (A) 1 ¯a ¯a 1 ˙ ˙ ¯ ¯ = A A − Fij (A)F ij (A) 2 i i 4 1 1 1 a 2 − ρa ¯ ∇ ∂k Dk (A) ρ .105) so that from (12.93). (12.106) Using the deﬁnition in (12.

i (12. i.107) has exactly the same form as in the Abelian theory except for the last term.110) then we can make a time independent inﬁnitesimal gauge transformation (under which the theory is invariant) 1 ¯ǫ a ¯ ¯ Di (A)¯ . (12. Here we see that since the gauge ﬁelds are self interacting. Π (y)} = δab δi j (x − y).112) . This is related to the question of gauge ﬁxing. even in the absence of other matter ﬁelds there is a long range interaction.e. The last term is like the long range Coulomb interaction term in the case of QED (in an Abelian theory where the structure constants vanish and the adjoint covariant derivative reduces to an ordinary derivative. i ¯ ∂i Aa = 0.109) with all other brackets vanishing. this last term has the form ρ 1 2 ρ which corresponds to ∇ the long range Coulomb interaction). Here the transverse delta function corresponds to the one already deﬁned in the case of Maxwell’s theory in (9. of the theory) ¯ is. i TR jb (12. A′ a = Aa + i i g such that ¯i ∂i A′ a = 0. Thus the long range behavior of this theory is still not well deﬁned in the Coulomb gauge.111) (12. Let us note here that the long range behavior of the interaction (and. ¯ Gribov has shown that the operator ∂i Di (A) does possess zero modes or eigenvectors with zero eigenvalue. therefore. We note that the Lagrangian density in (12. This can be seen from the ¯ fact that if we have a ﬁeld conﬁguration Aa which is transverse.31). namely the Coulomb gauge does not uniquely deﬁne the gauge ﬁelds. therefore.510 12 Yang-Mills theory The equal-time canonical Poisson brackets for the theory can now be obtained as in the case of the Maxwell theory ¯ {Aa (x). controlled by the eigenvalues of the operator ∂i Di (A).

classically we can impose the condition ∂ · A = 0. The canonical quantization of Yang-Mills theories does not have manifest covariance very much like the Maxwell theory. (12. for example.12. there are serious diﬀerences from Maxwell’s theory. therefore. However. 4 2 (12.2 Canonical quantization of Yang-Mills theory 511 ¯ provided ǫa corresponds to a zero mode of ∂i Di (A). (12. consequently.115) .129)) (∂ · A) = 0. known as the Gribov ambiguity. makes the theory nonsingular.113) The additional term clearly breaks gauge invariance and. occurs in all gauges other than the axial gauge.125) with J µ = 0) 1 1 a L = − Fµν F µν a − (∂µ Aµa )2 .) g Hence such gauge ﬁeld conﬁgurations can only be reached nonperturbatively. we would not worry about it within the context of perturbation theory.114) namely. very much along the lines of the Abelian theory that we have discussed in chapter 9. Gribov ambiguity is a phenomenon associated with large gauge transformations and is an important issue since it leads to nonperturbative eﬀects.75) as (see (9. This shows that ¯ the Coulomb gauge does not really specify the gauge ﬁeld conﬁguration uniquely. in the present case. However. we recall that in QED (see (9. For example. by modifying the theory. As long as we are within perturbation theory we can neglect such conﬁgurations. But let us note from the form of A′ a in i (12. the hypersurface deﬁning the gauge choice intersects the gauge orbits more than once. The other way of saying this is that within the framework of perturbation theory we are interested in inﬁnitesimal gauge transformations. ∂ · A behaves like a free ﬁeld and. we can also try to quantize the non-Abelian gauge theory covariantly. On the other hand. We can show that this nonuniqueness. Therefore. Thus.111) that this has a singularity at g = 0. In other words. (It is of the form O 1 . There now exist prescriptions to ¯ take care of this problem. much like in the Maxwell theory. namely. let us modify the theory (12.

we need to analyze the question of modifying the theory in a more systematic and detailed manner. µ Contracting with ∂ν . 12. (12. since it is not invariant under time evolution.) Correspondingly the naive analog of the Gupta-Bleuler condition for non-Abelian gauge theories does not seem to exist.118) Thus. Therefore. nor can we think of the supplementary condition ∂ · Aa(+) |phys = 0. Furthermore. we obtain (∂ · Aa ) = gf abc ∂ν Ab F µν µ c (12. it cannot be uniquely decomposed into a positive and a negative frequency part (in a time invariant manner). We would see next how we can derive intuition on this important question from the path integral quantization of the theory. however.116) In the case of the Yang-Mills theory.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 . consequently. the physical subspace would keep changing with time which is not desirable. even classically the equations of motion are given by ∂µ F µν a − gf abc Ab F µν c + ∂ ν (∂ · Aa ) = 0. just modifying the Lagrangian density as in (12.113) does not seem to be suﬃcient in contrast to the Abelian gauge theory.117) = 0. In other words. (Namely.113) has truly modiﬁed the theory. (12. we note that (∂ · Aa ) does not behave like a free ﬁeld and. in contrast to the Abelian theory. since (∂ · Aa ) is not a free ﬁeld. the additional term in (12. (12.512 12 Yang-Mills theory which then translates to the Gupta-Bleuler condition on the physical states ∂ · A(+) |phys = 0.119) in a physically meaningful manner.

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. we will only go over some of the essential concepts involved in such a description.12. 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] . see section 6.120) where N is a normalization constant. The functional dependence of a quantity on a variable is generally denoted by a square bracket.4 for the deﬁnitions of “in” and “out” states) in the presence of interactions. Z[J] and W [J] are known as generating functionals for the Green’s functions of the theory. Like the derivative of a function.3 Path integral quantization of gauge theories 513 gauge theories. (12. 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 . Since this is a complete topic in itself. from the “in” vacuum to the “out” vacuum. 2 (12. For example.122) and various Green’s functions (n-point functions) are obtained from the generating functionals Z[J] and W [J] through functional diﬀerentiation. 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. J(x) is the external source to which the scalar ﬁeld is coupled and S[φ] = d4 x 1 ∂µ φ∂ µ φ − m2 φ2 . we will not go into the systematic details of this description. for a real scalar ﬁeld interacting only with an external source. . 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. rather. δφ(y) ǫ ǫ→0+ (12.121) Furthermore.

) The determinant in (12. the functional integration is deﬁned (up to a normalization constant) as Dφ = dφi .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. 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) .120) can be evaluated in a straightforward manner. Let us divide the entire space-time into inﬁnitesimal cells labelled by “i” of volume δVi and deﬁne 1 δVi d4 x φ(x).124) δVi →0 and in terms of these discretized ﬁeld variables. i (12. Basically.514 12 Yang-Mills theory The integration in (12. (12.126) is independent of ﬁeld variables and is a constant (possibly divergent) and can be incorporated into the normalization constant N .125) With these basics a simple functional integral for a quadratic action such as in (12. δVi φi = (12.123) Clearly in the limit of vanishing volume we recover lim φi = φ(x). G = (−∂µ ∂ µ − m2 )−1 . (12. (12.120) is known as a functional integration and is deﬁned as follows. When we have an .126) where the Green’s function G(x − y) is formally the inverse of the operator in the quadratic part of the action. namely.

(12. the generating functional in the path integral formalism is given by (J ) [A Z [Jµ ] = eiW [Jµ ] = N = N DAµ eiS µ] (12. we evaluate the functional integral perturbatively (expanding the interaction term in the exponent so that the functional integral becomes a series of integrals corresponding to diﬀerent moments of a Gaussian integration) and the perturbative expansion coincides with the perturbative expansion of amplitudes in the conventional canonical quantized ﬁeld theory. The advantage of using the path integral description lies in the fact that there are no operators in (12.3 Path integral quantization of gauge theories 515 interacting theory (not quadratic in the ﬁeld variables) the functional integral.120). let us go back to the Maxwell theory 1 L(J) = − Fµν F µν + J µ Aµ . In this case.124)).122)) we have identiﬁed P µν (x − y) = (η µν (J µ . Aµ ) = − ∂ µ ∂ ν ) δ4 (x − y).129) µν A ν )+(J µ .132) where we have used the notation in (12.130) where N is a normalization constant and (see (9.126) Z [Jµ ] = eiW [Jµ ] = N (det (−P µν ))− 2 e− 2 (J 1 i µ . in general. (12. 4 (12. With this brief introduction to path integral description. However.A µ) DAµ ei[ 2 (Aµ .128) where J µ represents a conserved current (source).12. This is a Gaussian functional integral and we can evaluate this using our earlier result in (12. as we have seen before (see (9. In this case. (12. everything is a classical function.P 1 ] .131) d4 x J µ (x)Aµ (x).123) and (9. the operator P µν is a projection . cannot be obtained in a closed form.P −1 J ν µν ).131).

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

µ (12. (12.12. the Coulomb gauge. here are a few of the familiar gauge ﬁxing conditions F [Aµ (x)] = ∂µ Aµ (x) = 0. (12. gauge invariant and do not depend on the choice of the hypersurface (gauge). The way this is carried out is by choosing a hypersurface which intersects each orbit only once. the axial gauge. Thus. 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.137) deﬁnes the hypersurface which intersects the gauge orbits once.138) has a unique solution for some θ(x). ∇ · A(x) = 0. if F [Aµ (x)] = 0. namely. then even if Aµ does not satisfy the condition. We recognize that we should not integrate over all gauge ﬁeld conﬁgurations because they are not really distinct.3 Path integral quantization of gauge theories 517 (In the non-Abelian case. (12. A0 (x) = 0. we pick up a representative gauge ﬁeld from each gauge orbit.139) is known as the gauge condition (or gauge ﬁxing condition). of course. we can ﬁnd a gauge trans(θ) formed Aµ which does and F [A(θ) (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. Rather we should integrate over each orbit only once. Physical quantities are.140) and so on. the temporal gauge. This procedure is known as gauge ﬁxing and the condition F [Aµ (x)] = 0. A3 (x) = 0. We can . the Lorentz/Landau gauge. for example. this should be replaced by the group invariant Haar measure x dU (x). In this way.

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

141) that . we should have the Haar measure which is gauge invariant.144) In the non-Abelian case.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] .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) . this gives the correct functional form for the generating functional.146) Furthermore. dθ(x) = (12. d(U U ′ ) = dU. µ under which the generating functional takes the form dθ(x) δ F [Aµ (x)] eiS x (J ) [A µ] (J ) [A µ] (12. (12. (12.147) Z[Jµ ] = N = N DAµ ∆FP [Aµ ] dθ(x) x DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS (J ) [A µ] = N DAµ ∆FP [Aµ ] δ F [Aµ (x)] eiS . To do this let us note from (12. However.148) where the (inﬁnite) gauge volume has been factored out and absorbed into the normalization constant N of the path integral. Therefore.12. we still have to determine what ∆FP [Aµ ] is. namely. (12. let us make an inverse gauge transformation Aµ → A(−θ) .

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.152) The Faddeev-Popov determinant can.520 12 Yang-Mills theory ∆−1 [Aµ ] = FP x dθ(x) δ F [A(θ) (x)] µ dF δ F [A(θ) (x)] µ x = = det det δθ δF (12.149)-(12. (12. be thought of as the Jacobian that goes with a given gauge choice. for gauge ﬁelds which satisfy the condition F [Aµ (x)] = 0.) We can further generalize our derivation by noting that a general equation of the hypersurface has the form (physical results are not sensitive to the choice of the hypersurface) F [Aµ (x)] = f (x). (Here we can completely ignore the problem of Gribov ambiguity associated with large gauge transformations. 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).151) we determine δF [Aµ (x)] ∆FP [Aµ ] = det δθ(y) (θ) (12.151) θ=0 .150) (12.149) δθ δF F [Aµ (x)]=0 (θ) .153) . (12. On the other hand. Thus. we have θ(x) = 0. therefore. using (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 . Hence we can multiply the generating functional by a weight factor and integrate over all f (x). (12. Thus the generating functional in this case is given by Z[Jµ ] = N DAµ ∆FP [Aµ ]δ F [Aµ (x)] − f (x) eiS (J ) [A µ] . The Faddeev-Popov determinant is unchanged by this modiﬁcation because f (x) does not depend on Aµ (x). we can now do the following (also known as the ’t Hooft trick).157) and ξ is known as the gauge ﬁxing parameter. We note that physical quantities are independent of f (x). Then we can insert the identity ∆FP [Aµ ] x (θ) dθ(x) δ F [Aµ (x)] − f (x) = 1. (12. (12.156) where we have deﬁned a gauge ﬁxing action as SGF = d4 x LGF = − d4 x (F [Aµ (x)])2 . (12.155) Following ’t Hooft. Thus. (12.158) .154) into the functional integral.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 ).

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

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). c] = S (J) [Aµ ] + SGF + Sghost = d4 x Leﬀ [Aµ .157)) has the form LGF = − 1 1 (F [Aµ (x)])2 = − (∂µ Aµ (x))2 . Let us now look at a simple gauge ﬁxing condition in Maxwell’s theory.164) In this case. 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. 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. 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. for example. the gauge ﬁxing Lagrangian density should make the quadratic part of the eﬀective Lagrangian density nonsingular. of course. 2ξ 2ξ (12.3 Path integral quantization of gauge theories 523 where Seﬀ [Aµ . Otherwise it fails its purpose and will not correspond to an acceptable gauge ﬁxing condition. It is worth noting at this point that we have modiﬁed our starting Lagrangian density by a series of formal manipulations. (12.12. namely. It is. However.153)) F [Aµ (x)] = ∂µ Aµ (x) = f (x). c. our responsibility to show that the physical interpretation of the theory has not changed.163) Thus we can summarize what we have done so far. the covariant condition (see (12. the gauge ﬁxing Lagrangian density (see (12. Of course. for the moment let us emphasize that the purpose of a gauge ﬁxing Lagrangian density is to break gauge invariance. We would show this in the next chapter. c. (J) (J) (12. c] .165) . this modiﬁes the starting theory.

To obtain the corresponding ghost Lagrangian density for this gauge choice. the ghost ﬁelds are interacting and have to be present.169) is nothing other than Maxwell’s theory in the Feynman-Fermi gauge (9. 4 2ξ (12.168) d4 x ∂µ c(x)∂ µ c(x) = where we have neglected total divergence terms. however. Thus our eﬀective Lagrangian density for the Maxwell theory with the choice of a covariant gauge ﬁxing condition becomes 1 1 (J) Leﬀ = − Fµν F µν − (∂µ Aµ )2 + ∂µ c∂ µ c + J µ Aµ .167) Absorbing the factor 1 into the normalization factor (alternatively e scaling the ghost ﬁelds).152)) δF [Aµ (x)] δθ(y) (θ) 1 µ ∂ θ(x) .166) = θ=0 1 e xδ 4 (x − y).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 . (12.125). we may neglect them and then for ξ = 1 we recognize that our eﬀective Lagrangian density (12. therefore. we note that F [A(θ) (x)] = ∂µ A(θ)µ (x) = ∂µ Aµ (x) + µ so that (see (12. In non-Abelian gauge theories.524 12 Yang-Mills theory It is clear that this provides longitudinal components to the quadratic terms in ﬁelds and hence breaks gauge invariance.169) We note here that the ghost ﬁelds are noninteracting in the case of the Maxwell theory in ﬂat space-time and. e (12. This . (12. the ghost action (12.

they anticommute even though they are scalar ﬁelds. 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. For example. namely.113) of the non-Abelian gauge theory failed to be suﬃcient unlike in the Maxwell theory (namely. It is also true that when Maxwell’s theory is coupled to a gravitational ﬁeld. the Aµ ﬁeld has four ﬁeld degrees of freedom.164). has only one ﬁeld degree of freedom (we will discuss the question of hermiticity of the ghost ﬁelds in the next chapter). since the ghost ﬁelds and the ghost action are really necessary for the unitarity of the S-matrix.170) 4 2 Here we have introduced an auxiliary ﬁeld F which does not have any dynamics. the ghost ﬁelds automatically couple to the geometry also.169) and ask how we can recover the Lorentz gauge from the Lorentz like covariant gauge condition in (12. This naive counting works pretty well as we will see later.) Let us now go back to the eﬀective Lagrangian density (12. To do this let us rewrite the eﬀective Lagrangian density (12. On the other hand each of the ghost ﬁelds. we cannot neglect them even if they are noninteracting particularly when we are doing calculations at ﬁnite temperature.171) . On the other hand. the ghost Lagrangian density was missing). (The ghost degrees of freedom subtract because they have the unphysical statistics. (12. we see that the equation of motion for F is given by ξF (x) = ∂µ Aµ (x). 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 (12.3 Path integral quantization of gauge theories 525 explains why the naive modiﬁcation in (12. Furthermore.169) as ξ 1 (J) Leﬀ = − Fµν F µν + F 2 −F (∂µ Aµ )+∂µ c∂ µ c+J µ Aµ . One way of looking at the ghost ﬁelds is as if they are there to subtract out the unphysical ﬁeld degrees of freedom. (12. Hence omitting the ghost Lagrangian density in such a case would lead to incorrect results.12.169). being a scalar.

Here we modify the theory so that all the ﬁeld degrees of freedom are independent. Here we explicitly eliminate the unphysical (dependent) ﬁeld degrees of freedom and then quantize the physical (independent) ﬁeld degrees of freedom. we see that there are two distinct quantization procedures. we can quantize the Abelian gauge theory canonically. 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 . Namely. In this case. From our discussions of the Abelian gauge theories thus far.172) Therefore. However. 4 (12. As a .170) with ξ = 0) 1 (J) Leﬀ = − Fµν F µν − F (∂µ Aµ ) + ∂µ c∂ µ c + J µ Aµ . therefore. in the limit ξ → 0. this equation leads to the Lorentz condition (or the Landau gauge) ∂µ Aµ (x) = 0. We select out the physical Hilbert space by imposing supplementary conditions on the state vectors in a Lorentz covariant way. (12. (12. in the intermediate states we ﬁnd time-like photon states which contribute negatively whereas the longitudinal states contribute an equal positive amount. Thus the vector space that we work with in this case is much larger than the physical Hilbert space. The second possibility is to use the GuptaBleuler quantization method to quantize the theory in a manifestly covariant manner. We maintain manifest Lorentz covariance and quantize all components of the ﬁeld as independent variables. in the process of eliminating the dependent ﬁeld degrees of freedom we lose manifest Lorentz covariance. The Hilbert space contains only photon states of physical polarization.174) and describes Maxwell’s theory in the Lorentz gauge. the larger vector space of the theory contains states of indeﬁnite norm and when we do perturbation theory in this formalism. we have the operator quantization and again there appear to be two distinct possibilities in this case.526 12 Yang-Mills theory and. First.173) ξ→0 4 2ξ which can also be written equivalently as (see (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. non-invertible. we note that gauge invariance puts a very strong constraint on the structure of the Lagrangian density for the gauge ﬁeld.176) In this way. the ghost contributions cancel those from the unphysical gauge ﬁeld degrees of freedom). The second method of quantizing the Abelian gauge theory is through the method of path integrals. as the Lagrangian density describing the .175) where AT denotes the transverse physical degrees of freedom. their contributions cancel out and eﬀectively we are left with only two physical transverse degrees of freedom. In summary. (12. is given by (J ) [AT ] µ Z[Jµ ] = N = N DAT eiS µ DAµ ∆FP [Aµ ]δ(∂ · A(x)) eiS (J ) [A µ] . the Faddeev-Popov determinant or the ghost action can be thought of as the Jacobian in transforming from the physical dynamical ﬁeld variables to all components of the ﬁeld variables through the constraint relation. In particular. The generating functional for physical Green’s functions is given by (J ) [AT ] µ Z[Jµ ] = N = N DAT eiS µ DAµ δ(∂ · A(x))eiS (J ) [A µ] . However. As a result. In this formalism the ﬁeld variables are treated as classical variables. the coeﬃcient matrix of the quadratic terms in the Lagrangian density is singular and. therefore. if we take Linv . Rather the extra contributions (from the unphysical degrees of freedom) are cancelled by the Faddeev-Popov determinant which we can write as a ghost action (namely. in this formalism there is no reference to the Hilbert space of the theory. The µ correct description for the generating functional. (12.12. according to Faddeev and Popov. In the path integral formalism.3 Path integral quantization of gauge theories 527 result.

then we cannot deﬁne propagators and the entire philosophy of doing perturbative calculations with Feynman diagrams breaks down. consists of adding to the invariant Lagrangian density a gauge ﬁxing Lagrangian density of the form . the Lagrangian density for the gauge ﬁelds is given by (see (12. in the non-Abelian gauge theory. The standard covariant gauge ﬁxing (12. As we have seen. On the other hand. µ a Fµν (x) = ∂µ Aa (x) − ∂ν Aa (x) − gf abc Ab (x)Ac (x). and to compensate for that we have to add a corresponding Lagrangian density for the ghost ﬁelds following the prescription of Faddeev and Popov.75)) 1 a Linv = − Fµν F µνa .528 12 Yang-Mills theory dynamics of the gauge ﬁeld theory.179) ν µ µ ν with g denoting the coupling constant of the theory. 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. 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.178) where ǫa (x) is the inﬁnitesimal parameter of transformation. g (ǫ)a Aa (x) → Aµ (x) = Aa (x) + µ µ (12. Here the covariant derivative in the adjoint representation as well as the non-Abelian ﬁeld strength tensors are deﬁned as (see (12. In order to circumvent this diﬃculty.177) which is invariant under the inﬁnitesimal gauge transformation 1 Dµ ǫa (x). Let us now apply these ideas to the study of the non-Abelian gauge theory. in general.76)) Dµ ǫa (x) = ∂µ ǫa (x) − gf abc Ab (x)ǫc (x). (12.164).67) and (12. 4 (12. adding a gauge ﬁxing Lagrangian density changes the theory.

(12. δǫb (y) (ǫ) (12.181) Here ξ represents a real arbitrary constant parameter known as the gauge ﬁxing parameter. we can write the ghost action corresponding to this gauge choice as (see (12.181) that we are using.180) which corresponds to a gauge ﬁxing condition of the form F a [Aµ (x)] = ∂µ Aµa (x) = f a (x).183) = 1 µ ab 4 ∂ D δ (x − y) g x xµ µ = ∂x ∂xµ δab − gf acb Ac (x) δ4 (x − y). 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 . µ µ g so that we have (see (12. it is redundant to set it equal to zero at the end since (ǫ) F a [Aµ (x)] is linear in ǫa ) Sghost = = − d4 x Lghost d4 xd4 y ca (x) δF a [Aµ (x)] b c (y). 2ξ (12.184) µ 1 Consequently.160) and note that since ǫa is an inﬁnitesimal parameter.3 Path integral quantization of gauge theories 529 LGF = − 1 (∂µ Aµa (x))2 . Following the prescription of Faddeev and Popov.182) We note that for the covariant gauge choice (12. we can write 1 (ǫ)a F a [A(ǫ) (x)] = ∂ µ Aµ (x) = ∂ µ Aa (x) + Dµ ǫa (x) . (12.122)) δF a [Aµ (x)] δǫb (y) (ǫ) (12.12.

in some cases we have to work through the details of this analysis rather carefully in order to obtain the correct result. (12. 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. the gauge ﬁxing and the ghost Lagrangian densities. let us consider the gauge theory of the antisymmetric tensor ﬁeld Aµν (x) . In a deeper sense. (12. As we have mentioned earlier. However. 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.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. in the path integral formulation.530 12 Yang-Mills theory Sghost = = − = d4 x Lghost µ ab d4 xd4 y ca (x) ∂x Dxµ δ4 (x − y) cb (y) d4 x ∂ µ ca (x) (Dµ c(x))a .185) where we have dropped total derivative terms (surface terms). cannot be neglected even in ﬂat space-time. therefore.142)) which does not change the physical content of the theory. With all these modiﬁcations.186) = − Fµν F µνa − 4 2ξ We note that the ghost ﬁelds in the present case are interacting unlike in Maxwell’s theory and. the total Lagrangian density for the non-Abelian gauge theory can be written in this covariant gauge as LTOT = Linv + LGF + Lghost 1 1 a (∂µ Aµa )2 + ∂ µ ca (Dµ c)a .

we also note that this Lagrangian density is invariant under the gauge transformation δAµν (x) = ∂µ θν (x) − ∂ν θµ (x).12. the canonical momenta are antisymmetric (like the ﬁeld variables). 1.191) The independent ﬁeld variables of the theory are A0i .190) so that not all ﬁeld variables are independent. 3. 2. Let us consider only the free Lagrangian density for this ﬁeld deﬁned by 1 L = − Fµνλ F µνλ .189) Naive counting shows that Aµν has six ﬁeld degrees of freedom. µ. (12. Aij and we note that Πµν = −Πνµ = −F 0µν . ˙ ∂ Aµν (x) Πµν (x) = (12. we conclude that . since Fµνλ is completely antisymmetric in all the indices. (12.187) Such tensor ﬁelds have been studied in connection with the question of conﬁnement (of quarks in QCD) and are known as Kalb-Ramond ﬁelds. However.188) where the ﬁeld strength tensor corresponds to the totally antisymmetric tensor Fµνλ = ∂[µ A νλ] = ∂µ Aνλ + ∂ν Aλµ + ∂λ Aµν . 6 (12.192) namely. Furthermore. (12. ν = 0.4 Path integral quantization of tensor fields 531 Aµν (x) = −Aνµ (x). (12. This can be seen from the deﬁnition of the conjugate momenta ∂L = −F 0µν (x).

the corresponding ﬁeld variables are like c-number quantities and we can choose the gauge condition A0i (x) = 0.196) Thus it would seem that the theory has truly three degrees of freedom.195) (12. 2.194) (12. 3. Π0µ = −F 00µ = 0. (12. Π0i = 0. 2.532 12 Yang-Mills theory or. k = 1. and the Lagrangian density (12. j. (12. i = 1. With these conditions we have ˙ Πij = −F0ij = −Aij . (12.188) takes the form 1 1 L = − Fµνλ F µνλ = − 3F0ij F 0ij + Fijk F ijk 6 6 1 = − 3Πij Πij + Fijk F ijk .198) This would seem like three constraints and hence we would naively conclude that this theory has no dynamical degrees of freedom. we note that the theory still possesses a residual (static) gauge invariance under the transformation δAij = ∂i θj (x) − ∂j θi (x).193) Since these momenta identically vanish. However.197) so that we can impose a Coulomb gauge condition of the form ∂i Aij (x) = 0. 6 (12. We can see this by writing out the gauge condition explicitly . However.198) actually represents only two independent conditions. 3. i. on closer inspection we notice that the Coulomb gauge condition (12.

∂ρ Fµνλ − ∂λ Fρµν + ∂ν Fλρµ − ∂µ Fνλρ = 0. Hence the theory has truly one physical degree of freedom and the antisymmetric tensor ﬁeld describes a scalar ﬁeld (it is a gauge theory describing a scalar degree of freedom). 12 2 (12.203) We recognize this as the Lagrangian density for a free. the Lagrangiandensity (12. the three conditions corresponding to j = 1. ∂1 A12 + ∂3 A32 = 0. 3 are (12. 2 (12.202) where φ(x) represents a real scalar ﬁeld. (12. massless real scalar ﬁeld describing a single degree of freedom. ∂1 A13 + ∂2 A23 = 0. Since Fµνλ represent a totally antisymmetric third rank tensor in four space-time dimensions satisfying the Bianchi identity (see (12.189)). The fact that the theory has only one degree of freedom can also be seen in the following manner. In this case. (12.199) ∂2 A21 + ∂3 A31 = 0. . 2.201) 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). 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.4 Path integral quantization of tensor fields 533 ∂1 A1j + ∂2 A2j + ∂3 A3j = 0.12.188) 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.

According to our earlier discussions. (12. 2ξ LGF = − (12. so that we have (12.207) (θ) δF ν [Aµσ (x)] δθ λ (y) (θ) θ=0 µ ν ν µ = ∂xµ δλ ∂x − δλ ∂x δ4 (x − y) ν = (δλ x ν − ∂xλ ∂x ) δ4 (x − y).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)).208) This is a matrix with two vector indices. (12.205) corresponding to the choice of gauge condition F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x). the starting gauge invariant Lagrangian density has the form 1 Fµνλ F µνλ .204) and we add to it the covariant gauge ﬁxing Lagrangian density (ξ is the gauge ﬁxing parameter) 1 (∂µ Aµν )2 .534 12 Yang-Mills theory Let us now look at the path integral quantization of this theory. 6 L=− (12. Hence to write the determinant as an action we need ghost ﬁelds which carry a vector index and we have .

12. (12. Let us start with the generating functional (J ) [A Z [Jµν ] = N DAµν eiS µν ] . The ghost ﬁelds being vectors have four degrees of freedom each and since they satisfy anticommutation relations. Thus the eﬀective Lagrangian density appears to have 6 − 2 × 4 = 6 − 8 = −2.212) . 2 (12.210) Let us now try our naive counting of degrees of freedom of the theory. (12.4 Path integral quantization of tensor fields 535 Sghost = − = − = = 1 2 d4 xd4 y cν (x) δF ν [Aµσ (x)] λ c (y) δθ λ (y) x ν − ∂xλ ∂x ) δ4 (x − y) cλ (y) (θ) ν d4 xd4 y cν (x) (δλ d4 x (∂µ cν (x) − ∂ν cµ (x)) (∂ µ cν (x) − ∂ ν cµ (x)) d4 x Lghost .209) Therefore. (12.211) eﬀective ﬁeld degrees of freedom. This does not seem right (does not agree with the analysis from canonical quantization) and we will show now that this is a consequence of our careless application of the Faddeev-Popov procedure. The ﬁeld variable Aµν has six degrees of freedom. the eﬀective Lagrangian density for the theory appears to be given by Leﬀ = L + LGF + Lghost 1 1 = − Fµνλ F µνλ − (∂µ Aµν )2 6 2ξ 1 + (∂µ cµ − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) . they subtract out ﬁeld degrees of freedom.

209) 1 2 ). the generating functional can be written as Z [Jµν ] = N DAµν Dcµ Dcν (J ) +S ghost × δ (F ν [Aµλ (x)] − f ν (x)) ei(S where as we have seen before in (12. (12. (12.214) where.154)) (θ) ∆FP [Aµν ] x dθ σ (x)δ F ν [Aµλ (x)] − f ν (x) = 1. (12.217) In the naive application of the quantization procedure.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 introduce the identity element as (see (12. as we have seen (see (12.215) Thus inserting this identity element into the functional integral.218) .213) Therefore. δF ν [Aµλ (x)] δθ σ (y) ν x δσ θ=0 ν − ∂x ∂xσ ) δ4 (x − y) . F ν [Aµλ (x)] = ∂µ Aµν (x) = f ν (x).216) Sghost = d4 x (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) . (12. (12. (θ) ∆FP [Aµν ] = det = det ( (12. we would use the ’t Hooft trick to write i − 2ξ Df µ δ (F ν [Aµλ (x)] − f ν (x)) e = e i − 2ξ R d4 x fν (x)f ν (x) R d4 x F ν [Aµλ (x)]Fν [Aµλ (x)] .152)).

(12. (12. implies that ∂ν ∂µ Aµν = ∂ν f ν (x) = 0. We take into account the transverse nature of f µ (x) exactly like the Maxwell ﬁeld and we apply the ’t Hooft trick by integrating over a transverse weight factor i − 2ξ Df µ δ (F ν [Aµλ (x)] − f ν (x)) e µν R d4 xd4 y fµ (x)P µν (x−y)fν (y) . for example. 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 . we note that the gauge condition ∂µ Aµν (x) = f ν (x). (12. the action for fµ in (12.4 Path integral quantization of tensor fields 537 However.221) has a gauge invariance (just like Maxwell’s theory) of the form δfµ (x) = ∂µ θ(x). (12.221) where P (x − y) is the normalized transverse projection operator deﬁned earlier (see.223) Thus one has to use a gauge ﬁxing condition and we choose again a covariant gauge condition 1 1 2 F [fµ (x)] = ∂µ f µ (x) = f (x).222) We see that because of the transverse projection operator.123)) as µν µ ν ∂x ∂x x P (x − y) = η µν − δ4 (x − y).12.219) That is. (12.220) (12. (9.

c) = − 2 2 d4 x c(x) c(x). (12. remembering that (see (12.f i µ) ∆FP [fµ ].225) where δF [fµ (x)] δθ(y) (θ) 1 2 ∆FP [fµ ] = det θ=0 = det xδ 4 (x−y) .228) det xδ 4 (x − y) 1 2 Dc(x) eiS ghost . c) .f µ )− 2 (c.F (fµ )) i µν fν ) = Df µδ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .Fν [Aµλ ])− 2 (c. c) i µλ ].229) .P × = i µν fν ) Df ∆FP [fµ ] δ F [fµ (x)] − f (x) e i i − 2ξ (f.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 .226)) ∆FP [fµ ] = = we determine 1 1 S ghost = − (c.226) Thus using the ’t Hooft trick we can write Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .227) Furthermore.f ) Df µ δ (F ν [Aµλ (x)] − f ν (x)) e− 2ξ (fµ .P ×∆FP [fµ ] e− 2ξ (F (fµ ). Substituting this into (12. (12. (12. (12. (12.

12.232) Counting the ﬁeld degrees of freedom we see that the eﬀective number of degrees of freedom seems to be 6 − 2 × 4 − 1 = −3. Thus using this modiﬁed ’t Hooft weighting factor.231) DAµν Dcµ Dcν Dc ei(S DAµν Dcµ Dcν Dc ei(S J +S ghost +S ghost i ) e− 2ξ (Fµ [Aλσ ].230) SGF = − = − Therefore.F µ[Aλσ ]) J +S GF +Sghost +S ghost ). (12.233) This is again not right and the agreement with canonical quantization seems to be worse than before.4 Path integral quantization of tensor fields 539 Note here that the ﬁeld c(x) is a real anticommuting scalar ﬁeld. (12. therefore. In ﬂat space-time this ghost Lagrangian density can be seen to give a total divergence and. This is diﬀerent from the usual Faddeev-Popov ghosts in the sense that the Faddeev-Popov ghosts seem to come in pairs. the generating functional takes the form Z [Jµν ] =N =N where 1 2ξ 1 2ξ d4 x Fµ [Aλσ (x)]F µ [Aλσ (x)] d4 x (∂µ Aµν (x)) ∂ λ Aλν . . may be neglected. but in the presence of gravitation it cannot be written as a total divergence and is quite relevant. The ghosts of the present form are known as Nielsen ghosts. (12. (12. we can write the eﬀective Lagrangian density as (J) Leﬀ 1 1 (∂µ Aµν ) ∂ λ Aλν = − Fµνλ F µνλ + J µν Aµν − 6 2ξ + 1 1 (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) − c 2 2 c.

We can again use the FaddeevPopov trick and write ∆FP [cµ ] x (λ) dλ(x) δ F [cµ (x)] − f (x) = 1.237) We are now ready to calculate the Faddeev-Popov determinants associated with these gauge ﬁxing conditions.239) (λ) = det δF [cµ (x)] δλ(y) . namely. λ=0 .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.234) where λ(x) and λ(x) are anticommuting scalar parameters. (12. under δcµ (x) = ∂µ λ(x).235) and ∆FP [cµ ] x dλ(x) δ F [c(λ) (x)] −f (x) = 1. and δcµ (x) = ∂µ λ(x). (12. (12. ∆−1 [cµ ] = FP x (λ) dλ(x)δ F [cµ (x)] − f (x) (λ) dF δ F [cµ (x)] − f (x) det (λ) = x δF [cµ (x)] δλ(y) (12. the ghost Lagrangian density Lghost is invariant. (12. the ghost Lagrangian density also possesses a gauge invariance.238) = ∂µ cµ (x) = f (x).236) Let us choose for simplicity Lorentz like gauge conditions F and F = ∂µ cµ (x) =f (x). µ (12.

e . the ghost ﬁelds c and c behave like commuting scalars.237) and (12. the ’t Hooft trick needs to be carried out rather carefully.241) We note here that since we are writing an inverse determinant as an action. (12.4 Path integral quantization of tensor fields 541 so that (λ) −1 λ=0 −1 ∆FP [cµ ] = = δF [cµ (x)] det δλ(y) det xδ 4 (x − y) . the Faddeev-Popov term is an inverse determinant and.240) where we have used the fact that for anticommuting variables. Rather. we cannot use weight factors of the forms i − 2χ e R e e d4 x f (x)f (x) . . the Jacobian for a change of variables is the inverse of the determinant. namely. 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.242) The gauge conditions (12.243) since the exponents vanish (because of the fermionic nature of the variables). Namely.12. Similarly we can show that 4 −1 ∆FP [cµ ] = = det xδ (x − y) Dc Dc e ” “ e c c e −i e. consequently. (12. e i − 2χ R e e e e d4 x f (x) f (x) . f and f are Grassmann (fermionic) functions. the appropriate weight factor in this case would correspond to .238) correspond to fermionic conditions. (12. (12.

542 12 Yang-Mills theory e i − 2χ R e e e d4 x f (x) f (x) . anticommuting. c and c being commuting scalars contribute positively to the counting of the number of degrees of freedom. (12.245) d4 x Leﬀ . 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 . commuting. c. (12. Aµν : 6 degrees of freedom.246) 1 1 = − Fµνλ F µνλ − (∂µ Aµν )2 + J µν Aµν 6 2ξ 1 1 + (∂µ cν − ∂ν cµ ) (∂ µ cν − ∂ ν cµ ) − c 2 2 1 − (∂ν cν ) (∂µ cµ ) − c c − c c. 2χ c (12.247) We are now ready to count the number of eﬀective ﬁeld degrees of freedom in the theory.244) Thus putting in these identity elements into the functional integral for the generating functional and using the ’t Hooft trick with the weight factor (12. (12. c : 1 degrees of freedom. c : 1 degrees of freedom. c : −1 degrees of freedom. c : 1 degrees of freedom. (12.248) The ghosts c. cµ : −4 degrees of freedom. cµ : −4 degrees of freedom. The other .

Physical Review D4. R. 11. Nuclear Physics B139. Zuber. Nuclear Physics B44. 170 (1980). New York. G. 12. Gribov. Popov. 6. Quantum Field Theory. A. CERN preprint (1973). New York (1965). Physical Review 101. 7. 1 (1978). L. Siegel. V. 191 (1954). ’t Hooft and M. 8. R. P. (12. 5. N. C. Utiyama. 378 (1971). Huang.12. N. Singapore (1982). Veltman. Thus counting the degrees of freedom. Feynman. L. Mills. D. Qualks. Physical Review 96. 12. K. World Scientiﬁc. W. N. McGraw-Hill. 29 (1967). Faddeev. Veltman. McGrawHill. Mohapatra. 1980. L. Quantum Mechanics and Path Integrals. D. Itzykson and J-B. 10. N. This matches exactly with the counting of the degrees of freedom from the canonical quantization. Diagrammar. Physics Letters 25B. we see that eﬀectively the theory has 6 − 2 × 4 − 1 + 4 × 1 = 6 − 8 − 1 + 4 = 1. Theoretical and Mathematical Physics 1.5 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 . 2. 9. Leptons and Gauge Fields. 1 (1970). Faddeev and V. Physics Letters 93B. 189 (1972). Yang and R. R. G.5 References 1. Hibbs and R.249) degree of freedom. C. 3. . ’t Hooft and M. 4. 1597 (1956).

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

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

The invariance of the Lagrangian density (13. 2 1 ω δ f abc cb cc = f abc δcb cc = f abc f bpq cp cq cc 2 2 ω abc bpq f f + f abp f bqc + f abq f bcp cp cq cc = 0. . = − µ gξ (13. (13.546 13 BRST invariance and its consequences δAa = µ δca δca ω (Dµ c)a . 2 ω ∂ µ Aa . while for ca it is µ true only on-shell (namely. ca . ca . only when the ghost equation of motion is used).5) Here we have used the Jacobi identity for the symmetry algebra (see (12.50)).2 correspond to transformations with the parameters ω1.4) where ω is an arbitrary anti-commuting constant parameter of the global transformations. we obtain δ ∂ µ Aa = µ ω µ ∂ Dµ ca = 0.4) we have δ2 δ1 φa = 0.6) when the ghost equation of motion is used. = 6 (13. This shows that under two successive transformations of the kind (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 respectively. ca independent of the parameters of transµ formations where δ1. We note that the nilpotency of the transformations holds oﬀ-shell only for the ﬁelds Aa . Similarly.7) for the ﬁelds φa = Aa . g ω abc b c = f cc. g (13.

5). the invariant Lagrangian density is trivially invariant under these transformations. The present formulation of the BRST symmetry. we need to worry only about the changes in the gauge ﬁxing and the ghost Lagrangian densities which lead to 1 δ (LGF + Lghost ) = − (∂ ν Aa ) ∂ µ δAa + (∂ µ δca ) Dµ ca ν µ ξ ω µ ν a ω ∂ (∂ Aν ) Dµ ca = − (∂ ν Aa ) ∂ µ Dµ ca − ν gξ gξ ω = −∂ µ ∂ ν Aa Dµ ca . This is 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. we have used the fact that δ (Dµ c)a = 0 which we have seen in (13.8) Consequently. δLinv = 0.1 BRST symmetry 547 The invariance of the Lagrangian density (13. is slightly unpleasant in the sense that the nilpotency of the anti-ghost ﬁeld transformation holds only on-shell. namely. this is a reﬂection of the fact that the theory is lacking in some auxiliary ﬁeld variables and once the correct auxiliary ﬁelds are incorporated the symmetry algebra will close oﬀ-shell (without the use of equations of motion). This shows that the action for the total Lagrangian density (13. Generally.170) we can write the gauge ﬁxing Lagrangian density by introducing an auxiliary ﬁeld.34)) with the parameter ǫa (x) = ωca (x) and. Since this will also be quite useful for our latter discussions. let us recall that the gauge ﬁxing Lagrangian density in (13. however. In this derivation. First. (13. ν gξ (13.13.4) can now be easily checked.2) under the transformations (13. we note that the transformation for Aa can really be thought of as an inﬁnitesiµ mal gauge transformation (see (12.2) can also be written equivalently as .9) so that the action is invariant. As we have seen earlier within the context of the Abelian gauge theory in (12. therefore.2) is invariant under the global transformations (13.4) with an anti-commuting constant parameter.

13) δF a = 0.10) diﬀers from that in (12. (The form of the gauge ﬁxing Lagrangian density in (13.14) . g ω abc b c = f cc. (13.4) take the form δAa = µ δca δca ω (Dµ c)a . the BRST transformations in (13. µ (13.) It is clear from the form of LGF in (13. and it is straightforward to check that these transformations are nilpotent oﬀ-shell. but it is this form that is very useful as we will see. δ2 δ1 φa = 0. Among other things LGF as written above allows us to take such gauge choices as the Landau gauge which corresponds to simply taking the limit ξ = 0. namely. (13. The total Lagrangian density can now be written as LTOT = Linv + LGF + Lghost ξ 1 a = − Fµν F µνa + F a F a + ∂ µ F a Aa + ∂ µ ca (Dµ c)a . g (13.11) and when we eliminate F a from the Lagrangian density using this equation. we recover the original gauge ﬁxing Lagrangian density (up to a total divergence term).12) µ 4 2 In this case. 2 ω = − F a.170) by a total divergence.10) where F a is an auxiliary ﬁeld. µ 2 (13.548 13 BRST invariance and its consequences LGF = ξ a a F F + (∂ µ F a )Aa .10) that the equation of motion for the auxiliary ﬁeld takes the form ξF a = ∂ µ Aa .

1 BRST symmetry 549 for all the ﬁeld variables φa = Aa . we will not pursue this symmetry further in our discussions. since these do not lead to any new constraint on the structure of the theory beyond what the BRST invariance provides. F a .13. ca .4).16) δAa = µ δca = δca = δF a = −ωf abc F b cc . 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. which deﬁne a residual global symmetry of the full theory. In some sense the BRST transformations. g ω abc b c f cc. These are known as the anti-BRST transformations. There is also a second set of fermionic transformations involving the anti-ghost ﬁelds of the form ω (Dµ c)a . replace the original gauge invariance of the theory and play a fundamental role in the study of non-Abelian gauge theories. We note that Linv is invariant under the BRST transformation as we had argued earlier in (13.9)). g ω F a + gf abc cb cc .12) invariant. F a represents µ the missing auxiliary ﬁeld that we had alluded to earlier. g g (13.15) Unlike in the formulation of BRST variations without the auxiliary ﬁeld in (13. We note here that the BRST and the anti-BRST transformations are not quite symmetric in the ghost and the anti-ghost ﬁelds which is a . However. 2 (13. which can also be easily checked to leave the total Lagrangian density (13.8) and the auxiliary ﬁeld F a does not transform at all which leads to δLTOT = δ (LGF + Lghost ) = = ∂ µ F a δAa + ∂ µ δca (Dµ c)a µ ω ω µ a ∂ F (Dµ c)a − ∂ µ F a (Dµ c)a = 0. ca . Therefore.

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

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. (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). therefore. “(+)”.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. QBRST . are conserved and hence can be used to deﬁne a physical Hilbert space which would remain invariant under the time evolution of the system. µ (13. not clear a priori if the conditions (13. the Gupta-Bleuler condition works because ∂ µ Aµ satisﬁes the free Klein-Gordon equation (9.118).19) are suﬃcient to reproduce even the GuptaBleuler condition in the case of the Abelian theory (namely. whether they can reduce the vector space suﬃciently enough to coincide with the physical space). as we have seen in (12.130) in the covariant gauge (12. We recognize that even though this looks like one condition. = 0. In the Abelian theory. On the other hand. (13. in reality it is an inﬁnite number of conditions since it has to hold for every value of the coordinates.162)) ∂ µ A(+) (x)|phys = 0. 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. The corresponding operator in a non-Abelian theory.13. the conditions in (13.) o Thus. stands for the positive frequency part of the ﬁeld. does not satisfy a free equation and hence it is not a suitable operator for identifying the physical subspace in a time invariant manner. It is. the generators of the BRST symmetry.19) Note that even in the case of Maxwell’s theory. Qc . and the ghost scaling symmetry.164) and hence the physical space so selected remains invariant under time evolution.18) where the superscript. .

g abc µ a b c f (∂ c ) c c . g 2 µ(ǫ) Jc = δca ∂LTOT ∂LTOT + δca ∂∂µ ca ∂∂µ ca = −ǫca (D µ c)a − ǫca ∂ µ ca = ǫ ((∂ µ ca )ca − ca (Dµ c)a ) . 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 . 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 . 2 (13. .552 13 BRST invariance and its consequences To see that these conditions indeed lead to the Gupta-Bleuler condition in Maxwell’s theory. (13. 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 .20) where we have used the fact that the auxiliary ﬁeld does not transform under the BRST transformations and the ﬁelds Aa . F a are inert µ under the scaling of ghost ﬁelds.22) d3 x − (∂ 0 F a )ca + F a (D 0 c)a + 0 d3 x Jc = ˙ d3 x ca ca − ca (D 0 c)a . From this.

(13.22) to correspond to ← → d3 x F ∂ 0 c. (13. denoting the physical states of the theory as |phys = |Aµ ⊗ |n. In particular.23) following from (13. we note that for the Abelian theory where f abc = 0 and there is no internal index “a”. ← → d3 x c ∂ 0 c. this allows for states containing an equal number of ghost and anti-ghost particles. (13. n .19) Qc |phys = 0. 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) . the BRST and the ghost scaling charge operators can be obtained from (13.25) Let us note that the condition (13. In principle.13.24) QBRST = Qc = ˙ d3 x − F c + F c = ˙ ˙ ˙ d3 x cc − cc = − If we use the ﬁeld decomposition for the ﬁelds and normal order the charges (so that the annihilation/positive frequency operators are to the right of the creation/negative frequency operators). (13. (13.12).26) implies that the physical states must have (net) zero ghost number. Thus.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 .27) .

19) are the right ones even for the non-Abelian theory. then this implies that (13. |phys = |Aµ ⊗ |0. for every momentum mode k). 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. 0 = |Aµ . and F (+) (k)|Aµ = 0.19) really select out the subspace of physical states. and must further satisfy (with k0 = |k|) F (+) (k)|phys = 0 = kµ Aµ(+) (k) |phys .11)).29) Namely.12) that we can obtain the canonical momenta conjugate to various ﬁeld variables of the theory to correspond to .30) (13. (13. (13. To investigate systematically whether the physical state conditions in (13. n = 0. 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). Thus. we note from the form of the Lagrangian density (13.554 13 BRST invariance and its consequences where n.31) where we have used the equation of motion for the auxiliary ﬁeld (see (13.28) c(+) (k)|n. we feel conﬁdent that the physical state conditions in (13. n = 0. n denote the (equal) number of ghost and anti-ghost particles. 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.

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

14)). The third is a statement of the fact that the BRST charge carries a ghost number of unity. c QBRST = − Qc = − (13. Qc ] = −iQBRST .33) [QBRST .556 13 BRST invariance and its consequences and we note that the currents (13. QBRST ]+ = 2Q2 BRST = 0. Qc ] = 0.38) This algebra can also be derived directly from the transformation laws for the ﬁeld variables in (13.38) is that . therefore. Qc ] = −ica (x). that (this basically describes the behavior of the ghost ﬁelds under a scaling (13. [QBRST . We note. c c 2 a d3 x (Πa ca + ca Πc ) . The conserved charges (13. (13.36) and we can calculate the algebra of charges as well as various other relations of interest using the (anti) commutation relations (13. The algebra of the conserved charges also follows in a straightforward manner from (13. we see that we can think of iQc as the ghost number operator (with the ghost number for ca being negative).38) simply reiterates in an operator language the fact that the BRST transformations are nilpotent (see (13. Qc ] = ica (x). An immediate consequence of the algebra in (13.36) and (13.21) and.34) for the ghost ﬁelds.33) for the ﬁeld variables.17)) [ca (x). Therefore.22) can now be expressed in terms of the ﬁelds and the conjugate momenta as (before integrating the ﬁrst term by parts in QBRST . The ﬁrst equation in (13. (13.22) are Hermitian with the assigned hermiticity conditions (13. for example.37) [ca (x). see (13.17). The second equation implies that the ghost scaling symmetry is Abelian.22)) g d3 x Πia (Di c)a − F a Πa + f abc Πa cb cc . the conserved charges (13. [Qc .13) and (13.

12) contains various unphysical states as well as states with negative norm in addition to the physical states. the total vector space V of the complete gauge theory (13.40). therefore. Namely. then. it is truly a BRST singlet (invariant) state.41) We note that there are two possible kinds of states which will satisfy the physical state condition (13. (13. must represent a truly gauge invariant ﬁeld variable.13. As we have discussed earlier. The ﬁeld operator which creates such a state must necessarily commute with QBRST and. we have QBRST |Ψ = 0. 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) We emphasize again that such an identiﬁcation of the physical subspace is invariant under the time evolution of the system since QBRST commutes with the Hamiltonian since the BRST transformations deﬁne a symmetry of the full theory.2 Covariant quantization of Yang-Mills theory 557 QBRST eπQc = eπ(Qc −i) QBRST = −eπQc QBRST . H] = 0. for every |Ψ ∈ Vphys . (13. (we assume that the vacuum state is a BRST singlet (invariant) state) . (13. or. First. we can select out a subspace Vphys by requiring that states in this space are annihilated by QBRST . [QBRST . Consequently. eπQc + = 0. (13. QBRST . namely.39) which is quite useful (particularly) in dealing with gauge theories at ﬁnite temperature.42) ˜ for some |Ψ and still satisﬁes the physical state condition. As is clear from our earlier discussions. if a state |Ψ cannot be written as ˜ |Ψ = QBRST |Ψ .

therefore. This implies that (13.13). also satisﬁes the above relation.45) and transverse ﬁelds would represent such operators which are gauge invariant asymptotically. (13. correspond to truly physical states of the theory and since gauge invariant degrees of freedom have physical (non-negative) commutation relations.) 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.40) can be written in the form ˜ |Ψ = QBRST |Ψ .558 13 BRST invariance and its consequences QBRST |Ψ = QBRST Ψ|0 = [QBRST .47) Such a state would also be orthogonal to any physical state (either of the ﬁrst or the second kind) satisfying the physical state condition . Ψ] = 0. namely. 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. it can be written as the BRST variation of the anti-ghost ﬁeld and. However.38)). Ψ] |0 = 0. Such states would.46) We see that the physical state condition is trivially satisﬁed in this case because of the nilpotency of QBRST (see (13. (13. as we have seen in (13. such states would have positive norm. Ψ|0 = |Ψ . (Note that the auxiliary ﬁeld does not transform under a BRST transformation and.34) for the ghost ﬁelds. has a diﬀerent character.44) [QBRST .) ˜ ˜ Ψ|Ψ = Ψ|QBRST QBRST |Ψ = 0. (13. therefore. (13.

13.49). then ˜ Ψ′ |Ψ = Ψ′ |QBRST |Ψ = 0. then. (13. we have ∞ n=0 P (n) = ½. V0 Every state in Vphys can. (13. which satisfy the physical state condition and are of the ﬁrst kind can.2 Covariant quantization of Yang-Mills theory 559 (13.48) which follows from the physical state condition.51) . deﬁning P (n) as the projection operator onto the n-unphysical particle sector.40) because if |Ψ represents a physical state (of either kind) while |Ψ′ is a state of the second kind. of course. µ We recall from (13.13) through (anti) commutation with the appropriate ﬁeld variables) F a ∼ [QBRST . Here.49) P (n) P (m) = δnm P (n) . we mean states containing quanta of the ghost ﬁelds. The true physical states. therefore. ca ]+ . this would contain all the zero norm states which would be orthogonal to Vphys itself. the longitudinal components of Aa and the auxiliary ﬁelds. If we denote all the states of Vphys of the second kind by V0 . be identiﬁed as belonging to the V quotient space V phys = phys and will have positive deﬁnite norm. by unphysical.49) that P (0) projects onto the truly physical states of the theory (namely. Thus. be decomposed into states containing ﬁxed numbers of unphysical particles. we note that we can write P (0) = ½ − P ′ . (13.13) that the auxiliary ﬁeld can be written as a BRST variation of the anti-ghost ﬁeld as (note that QBRST is the generator of the BRST transformations and generates the transformations in (13.50) It is clear from the deﬁnition in (13. (13. with no unphysical particles) and from the completeness of the projection operators in (13.

54) (13.55) which follows from the nilpotency of QBRST . (13. R]+ = 0. [QBRST . (13. P ′ = QBRST .52) We can. Namely. n ≥ 1.56) + P′ It is now easy to show that P (n) . so that QBRST . Consequently. however. of course. but this is not very illuminating. = 0.560 where ∞ n=1 13 BRST invariance and its consequences P = ′ P (n) . that P ′ must also commute with QBRST . P (0) QBRST . In fact. and. R]+ . the diﬀerence between the two lies in the fact that P ′ must necessarily involve unphysical ﬁelds since it projects onto unphysical states and. (13. we can write P (n) = QBRST . for n ≥ 1. It follows. we can even show that for every P (n) . cannot be truly gauge invariant. therefore. is the fact that P (0) projects onto the true physical states and. it must have the form (for some fermionic operator R. What is interesting.53) However. therefore. then. whose explicit form is not important for our discussions) P ′ = [QBRST . (13. we have QBRST . therefore. R(n) . P ′ = 0. must be gauge invariant and will commute with QBRST . construct the actual forms of all the projection operators explicitly. project onto the zero norm space V0 when acting on states |Ψ .

Ψ′ |P (n) |Ψ Ψ′ |P ′ |Ψ = = Ψ′ | QBRST . This V0 would correspond to the true physical subspace of the total vector space. with the hermiticity assignments for the ghost ﬁelds in (13. then. (13. R]+ |Ψ = 0. R(n) |Ψ = 0.19). the total Lagrangian . 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. and the norm of any such state is. then. let |Ψ . The question of the unitarity of the S-matrix in a gauge theory can be formulated in the following manner.57) + Ψ′ | [QBRST . It leads to a formal proof of unitarity of the theory when restricted to the subspace of the physical Hilbert space. which follows from the physical state condition (13. has a positive semi-deﬁnite norm as we should have for a physical vector space and the value of the norm depends on the truly physical component of the state. (13.13.3 Unitarity 561 which satisfy the physical state condition (13. For example. 13. Any vector |Ψ ∈ Vphys can now be written as |Ψ = P (0) |Ψ + P ′ |Ψ .34). obtained to be Ψ|Ψ = Ψ|P (0) |Ψ ≥ 0. We will only outline the proof of unitarity in this section. deﬁned by the physical state condition.40) (as well as the hermiticity of QBRST ). therefore.59) This.40). This also makes it clear that the V norm of a state |Ψ ∈ V phys = phys would be positive deﬁnite.58) (13. shows that Vphys .3 Unitarity The BRST invariance of a gauge theory is quite important from yet another consideration. We note that. |Ψ′ ∈ Vphys .

S † S = SS † = ½. any two operators which do not take us out of Vphys ) Ψ|P (0) |Φ Ψ|(½ − P ′ )|Φ = Ψ|Φ .12) is Hermitian so that the S-matrix of the theory is formally unitary. Let us note here from the discussions of the last section that for any |Ψ . P (0) acts as the identity operator.e. (13. Furthermore.562 13 BRST invariance and its consequences density (13.64) The two states. so that [QBRST .61) Given these the question that we would like to understand is whether one can deﬁne an operator Sphys which would act and correspond to the S-matrix in the physical subspace of states of the theory and which will also be unitary. 2 (13. the S-matrix is BRST invariant since the full theory is. we can deﬁne a unique state |Ψ ∈ V phys as |Ψ = P (0) |Ψ .62) which follow from (13.60) Furthermore. S] = 0.57).. |Ψ and |Ψ . P (0) |Ψ = P (0) |Ψ = P (0) |Ψ = |Ψ . diﬀer only by a zero norm state which is orthogonal to every state in Vphys so that the inner product of any two such states is the same . |Φ ∈ Vphys and any two operators A and B acting on Vphys (i. (13. This question can be systematically analyzed as follows. (13. (13. = = Ψ|AP (0) B|Φ Ψ|A(½ − P ′ )B|Φ = Ψ|AB|Φ .63) where we note that in the space V phys . namely. namely. we note that given any state |Ψ ∈ Vphys .

It follows from (13. |Ψ = P (0) |Ψ ∈ V phys as deﬁned in (13. In a similar manner. (13. also automatically leads to a formal proof of unitarity of the S-matrix in the subspace of the truly physical states of the theory.66) and (13.69) In other words.63). (13. we can also show that † Sphys Sphys = ½.65) where the last line follows from the ﬁrst of the conditions in (13.60). (13.70) .13.62) as well as the formal unitarity of S (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) .67) It is clear that since S is BRST invariant. (13. (13. it will leave the space Vphys invariant and hence Sphys will not take a state out of V phys .62). Furthermore.68) where we have used (13. the physical state condition (13.40) which naturally follows from the BRST invariance of the theory. 2 (13.66) such that for |Ψ ∈ Vphys .3 Unitarity 563 Ψ|Φ = Ψ| P (0) |Φ = Ψ|P (0) |Φ = Ψ|Φ .68) that † Sphys Sphys = ½. we now have † Ψ|Sphys Sphys |Φ † Ψ|P (0) Sphys Sphys P (0) |Φ = = = Ψ|S † P (0) S|Φ Ψ|S † S|Φ = Ψ|Φ = Ψ|Φ . Sphys |Ψ = Sphys P (0) |Ψ = P (0) S|Ψ = S|Ψ . Given this.

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. namely. − ca F a − ∂ µ ca Aa µ 2 + (13. 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 .564 13 BRST invariance and its consequences As another consequence of the BRST invariance of the theory. 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) .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. be written as a BRST variation (with the parameter of transformation taken out). It now follows from the physical state condition (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. = 0.40) that phys| (LGF + Lghost ) |phys′ = −g phys| QBRST . To show this. in fact. we note that these extra terms in the Lagrangian density (namely. = g QBRST . the gauge ﬁxing and the ghost Lagrangian densities) can. we can show that the gauge ﬁxing and the ghost Lagrangian densities lead to no physical consequences.

7. 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.74) where S[φ] denotes the dynamical action of the self-interacting scalar ﬁeld. let us consider the self-interacting ﬁeld theory of a real scalar ﬁeld coupled to an external source described by the action S J [φ] = S[φ] + d4 x J(x)φ(x).13. In this case. before we go into the actual derivation. For simplicity. However.4 Slavnov-Taylor identity The BRST invariance of the full theory leads to many relations between various scattering amplitudes of the theory.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.75) .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 . These are known as the Ward-Takahashi identities (in the Abelian case) or the SlavnovTaylor identities (in the non-Abelian case) of the theory and are quite essential in establishing the renormalizability of gauge theories. (13. 13.120). ca F a ]+ |0 J = 0. We have already given a simple diagrammatic derivation of such identities in the case of QED in section 9. (13. as we have seen in (12. However. the vacuum functional is given by 0|0 J = Z[J] = eiW [J] = N Dφ eiS J [φ] . (13. let us recapitulate brieﬂy some of the essential concepts from path integrals (a detailed discussion is beyond the scope of these lectures).

(13. the n-point Green’s functions would involve Feynman diagrams which consist of parts that are not connected and Z[J] (through (13. δJ(x) Z[J] δJ(x) (13. (13. This is more fundamental since the general Green’s functions can be constructed in terms of these. In general. In particular we note that 0|φ(x)|0 J δn W [J] δJ(x1 ) · · · δJ(xn ) J=0 . On the other hand.49)) and is a functional of J.79) . To clarify this distinction further. For example.78) is known as the classical ﬁeld (see (7.77) in the path integral formalism corresponds to a unique Feynman diagram in perturbation theory.77) involve only Feynman diagrams where all the parts of the diagram are connected.76) and (13.77) = φc (x) = (−i) δZ[J] δW [J] = . the Green’s functions calculated from W [J] (the logarithm of Z[J]) through (13. 0|T φ(x1 ) · · · φ(xn ) |0 = δn Z[J] (−i)n Z[J] δJ(x1 ) · · · δJ(xn ) . 0|T φ(x1 ) · · · φ(xn ) |0 c = (−i)n−1 leads to the connected n-point time ordered Green’s functions of the theory. On the other hand. Z[J] and W [J] are known as the generating functionals for the Green’s functions of the theory.76)) includes contributions of all diagrams (including the ones that are not connected) to a Green’s function. let us note here that each term in the calculation of the Green’s function in (13.77) iGF (x − y) = 0|T φ(x)φ(y) |0 = (−i) δ2 W [J] δJ(x)δJ(y) J=0 c = (−i)2 δ2 Z[J] Z[J] δJ(x)δJ(y) J=0 . The connected two point time ordered Green’s function which also corresponds to the Feynman propagator of the theory is similarly obtained from (13. (13.76) J=0 deﬁnes the n-point time ordered Green’s function of the theory.566 13 BRST invariance and its consequences As we have already mentioned in the last chapter.

82) .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). 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. 2 (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 . 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.80) It can now be checked using (13. let us consider the eﬀective Lagrangian density (at the lowest order or tree level) which consists of LTOT for the Yang-Mills theory (13.78) that δΓ[φc ] δφc (x) = d4 y δW [J] δJ(y) δJ(y) − φc (y) − δ4 (x − y)J(y) δJ(y) δφc (x) δφc (x) (13. With these basic ideas from path integrals. (13. the more fundamental concept in perturbation theory is the diagram without the external propagators known as the vertex function.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. However. Let us Legendre transform W [J] as (this is like going from the Lagrangian to the Hamiltonian) Γ[φc ] = W [J] − d4 x J(x)φc (x).

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

This immediately leads from (13.87) can also be written as δW δW δW = 0. (13.75) we note that (13. It is here that the . (13.86) the generating functional should be invariant (namely. δJ a (x) (13. − 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 .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. Therefore. it should not change under any ﬁeld redeﬁnition). if we redeﬁne the ﬁelds under the path integral as.85) to (see also (13.85) g 2 We note that the generating functional is deﬁned by integrating over all possible ﬁeld conﬁgurations. A → A + δBRST A.87) +iη a (x) The functional integration measure can be easily checked to be invariant under such a fermionic transformation and using (13.4 Slavnov-Taylor identity 569 in Leﬀ to be (remember that the parameter of the BRST transformation is anti-commuting) = J µa δAa + J a δF a + i (η a δca − δca η a ) µ = δLeﬀ g ω µa J (Dµ c)a + i − f abc η a cb cc + F a η a .13. since it does not depend on the ﬁeld variables.

however. δK a (13.80) and the ﬁeld variables are really the classical ﬁelds and we are dropping the superscript (c) for simplicity). From the deﬁnition of the generating functional for the proper vertices (see (13. it is clear that δΓ δAa µ δΓ δF a δΓ δca δΓ δca δΓ a δKµ δΓ δK a = −J µa . = iη a . we have Γ[A.570 13 BRST invariance and its consequences usefulness of the sources for the composite BRST variations becomes evident. = −J a .88) in terms of the generating functional for the proper vertices as .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 (13.81)). Most often. we are interested in the identities satisﬁed by the proper (1PI) vertices of the theory. = = δW . K] − d4 x J µa Aa + J a F a + i (η a ca − ca η a ) .90) Using these deﬁnitions. µ (13. a δKµ δW . = iη a . we see that we can rewrite the master equation (13. K] = W [J.

The BRST invariance.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. like the Slavnov-Taylor identities.93) − c δF b (p)δAa (−p) δcc (−p)δK µa (p) δc (−p)δcb (p) µ A simple analysis of this relation shows that the mixed two point vertex function involving the ﬁelds F -Aµ is related to the two point function for the ghost ﬁelds and. follow from the BRST invariance of the theory. let us note that we can write the master identity (13. 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. in this way. (We would discuss renormalization of ﬁeld theories in a later chapter. for example.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. This is essential in proving the renormalizability of gauge theories. consequently. (13.4 Slavnov-Taylor identity 571 d4 x δΓ δΓ δΓ δΓ δΓ + a − F a (x) a = 0. For example. Thus.) . a (x) δK µa (x) a (x) δAµ δc (x) δK δc (x) (13. is very fundamental in the study of gauge theories as far as renormalizability and gauge independence of physical observables are concerned. δ2 δF b (p)δcc (−p) d4 k (13. the counter terms (quantum corrections) should satisfy such a relation.13.92) and setting all Taking derivative of this with respect to ﬁeld variables to zero gives δ2 Γ δ2 Γ δ2 Γ = 0.

− 1− (13.5 Feynman rules 13 BRST invariance and its consequences Understanding gauge ﬁxing for gauge theories in the path integral formalism is quite essential in developing the perturbation theory for gauge theories.95) LQ = − = = 1 a µν A η 2 µ 1 a µν ab b A O Aν + ca M ab cb . (They are the Green’s functions of the free theory.572 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. it renders the theory nonsingular making it possible to deﬁne the propagator of the theory and derive the Feynman rules for the non-Abelian gauge theory.2) 1 1 a ∂ µ Aa LTOT = − Fµν F µνa − µ 4 2ξ 2 + ∂ µ ca Dµ ca .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). Let us recall that the total Lagrangian density after gauge ﬁxing (in the covariant gauge) has the form (13. ξ M ab = −δab . 2 µ where we have neglected total divergence terms and have identiﬁed Oµν ab = δab η µν 1 µ ν ∂ ∂ .) The inverses of the two point functions are easily calculated in the momentum space where the quadratic operators in (13.96) take the forms . We note that since gauge ﬁxing breaks gauge invariance. (13. These are the essential elements in carrying out any perturbative calculation in a quantum ﬁeld theory.

we can parameterize the most general form of the inverse as O−1ab (p) = δab α ηµν + β µν pµ pν .5 Feynman rules 573 M ab (p) = δab p2 . λ ξ ξ (13.99) where α. λ ξ (13.99) for the inverse and carrying out the multiplication explicitly we have η µν p2 − 1 − or. η µν p2 − 1 − 1 µ ν −1ac p p O νλ (p) = −δac δµ .100) Substituting the parameterization (13.13. or. p2 (13. ξ (13.98) while the derivation of the inverse of Oµν ab (p) is a bit more involved. λ or. λ ξ ξ .97) The inverse of M ab (p) in (13. p2 (13. With this parameterization. ηµν ). λ αp2 δµ − α 1 − λ 1 1 µ p pλ + β pµ pλ − β 1 − pµ pλ = −δµ . Therefore. let us derive this inverse systematically. β are arbitrary parameters to be determined (they are not necessarily constants and can be Lorentz invariant functions of the momentum). we note that the inverse is deﬁned to satisfy (repeated indices are summed) −1bc Oµν ab (p)Oνλ (p) = δac δµ . Oµν ab (p) = −δab η µν p2 − 1 − 1 µ ν p p .101) 1 β µ (αp2 + 1)δµ − α(1 − ) − p pλ = 0. Let us note that −1ab the Green’s function Oµν (p) is a symmetric second rank tensor (of rank 2) and with all the Lorentz structures available (pµ .97) is quite simple (it is like the propagator of a massless scalar ﬁeld) M −1ab (p) = δab . 1 µ ν p p ξ αηνλ + β pν pλ p2 = −δµ .

104) where ( . p2 1 β − = 0.126)) h −1ab i − 2 J µa (−p).99) as pµ pν δab ηµν + (ξ − 1) 2 . ξ ξ 1 (ξ − 1).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. β = α(ξ − 1) = − so that we can write the inverse in (13. 2 p p (13. it follows that 1 . (13.574 13 BRST invariance and its consequences Setting the coeﬃcients of each distinct Lorentz structure to zero.Oµν (p)J νb (p) +i η a (−p). .102) α=− α 1− or.M −1ab (p)ηb (p) Z0 = N e i . p2 (13. 2 p p −1ab Oµν (p) = − (13. 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 .103) With these.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. ) represents the integral over p and we deduce from this that (see (13.

when we are in the Landau gauge this propagator is transverse.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]. The propagators can be diagrammatically represented as = iGab (p) µν = − a p b (13. 2 p p iδab .107) µ.13.e. µ .110) −gf abc Aa ∂ µ cc cb . i. the gauge is known as the Feynman gauge where the propagator has a much simpler form which is more suitable for perturbative calculations. 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 . p2 (13.106) which deﬁnes the ghost propagator (which is sometimes also denoted by Dab (p)). a p ν. while for ξ = 1.5 Feynman rules 575 When ξ = 0. (13. p2 (13.109) 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. b pµ pν iδab ηµν + (ξ − 1) 2 ..

p2 . p3 ) = i δ3 S δAa (p1 )δAb (p2 )δAc (p3 ) µ ν λ = i δ3 Sint δAa (p1 )δAb (p2 )δAc (p3 ) µ ν λ = (2π)4 gf abc δ4 (p1 + p2 + p3 ) × η µν (p1 − p2 )λ + η νλ (p2 − p3 )µ + η λµ (p3 − p1 )ν .576 13 BRST invariance and its consequences Let us now derive the interaction vertices for the theory which are more useful in the momentum space (where Feynman diagram calculations are primarily carried out). We note that we can write the interaction action in the momentum space as Sint = d4 x Lint d4 p1 d4 p2 d4 p3 δ4 (p1 + p2 + p3 ) = (2π)4 gf abc × (−ip1µ ) Aa (p1 ) Aµb (p2 ) Aνc (p3 ) ν − (2π)4 g2 abp cdp f f 4 d 4 p1 d 4 p2 d 4 p3 d 4 p4 × δ4 (p1 + p2 + p3 + p4 )Aa (p1 )Ab (p2 )Aµc (p3 )Aνd (p4 ) µ ν −(2π)4 gf abc d4 p1 d4 p2 d4 p3 δ4 (p1 + p2 + p3 ) (13. p3 .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 ) . p2 . V µνλρ abcd (p1 . 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 . p4 ) = i (13.

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

A class of gauge conditions of the form n · Aa (x) = f a (x). 13.116) where nµ is an arbitrary vector does indeed achieve this. we recall (see (12.160)) that the total Lagrangian density for a non-Abelian gauge theory takes the form 1 1 a (n · Aa )2 − ca n · Dca . When n2 = 0. the diagrams with ghosts will be absent).117) . n2 = 0 n2 = 0. If possible. the cc ﬁeld.114)). 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. see for example (13. We will now discuss. this would.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. In a general axial-like gauge (13. It is interesting to ask if there exist gauge conditions in a non-Abelian theory where the ghost degrees of freedom may not be important much like in the Abelian theory. for n2 = 1 (normalized) it is called the temporal gauge while for n2 = −1 it is known as the axial gauge. The dot over the ghost line denotes where the derivative acts in the interaction term in the Lagrangian density (namely. 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. or (13. LTOT = − Fµν F µνa − 4 2ξ (13. in what sense these gauges become ghost free in a non-Abelian gauge theory. such a gauge is known as the light-cone gauge. from two distinct points of view.116). Let us ﬁrst discuss this from the diagrammatic point of view.6 Ghost free gauges In quantizing a non-Abelian gauge theory. Such a class of gauge choices is conventionally known as ghost free gauges. simplify the perturbative calculations enormously primarily because the number of diagrams to evaluate will be much smaller (namely. of course.

13. Therefore. (n · p)2 (13.6 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 .118) so that we have (independent of whether n2 = 0 or n2 = 0) pν . will have the structure (we are not writing the possible multiplicative Lorentz factors involving nµ which do not eﬀect the integral and we are evaluating the integral in D dimensions) Figure 13.1: Examples of diagrams with open and closed ghost lines. the integrand of any diagram involving ghosts. either open lines or closed loops. n·p iGab = (13.120) and the antighost-ghost-gauge vertex involves a multiplicative factor of nµ . the ghost propagator in the axial-like gauge (13. (n · p) nµ Gab = −ξδab µν nµ nν Gab = −ξδab . . µν (13.119) Similarly.116) in this theory has the form δab .

(13.123) Here G[f a ] is an arbitrary functional of f a which we normally choose to correspond to the ’t Hooft weight factor .580 13 BRST invariance and its consequences dD k = 1 (n · k) (n · (k + p1 )) . . An alternative way to see this is to note that the generating functional in the axial-like gauge (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 ) . Using the fact that the integral is Lorentz invariant. . (k2 )(m+1)/2 (13.122) and the vanishing of this integral (in the last step) is easily seen using dimensional regularization (that we will discuss in chapter 15.10)). in (15. 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. namely. (n · k)m+1 δ(1 − x0 − x1 − · · · − xm ) m+1 = 0 i=0 dxi (13. we can rewrite this also as 1 m dxi 0 i=0 δ(1 − x0 − x1 − · · · − xm ) (n2 )(m+1)/2 dD k = 0.116) 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 .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.

3. Rouet and R. 344 (1974). (13. 99 (1972). let us choose G[f a ] = det n · ∂δab + gf abc f c −1 − i R d4 x f a f a 2ξ e .116) in a non-Abelian gauge theory. Taylor. since the generating functional does not depend on the functional form of G [f a ] (up to irrelevant multiplicative constants). (13. correspond to ghost free gauges very much like in the Abelian gauge theories. the covariant gauge. Note that such a decoupling works only because nµ (in n · Aa inside the determinant) is a multiplicative operator. Slavnov.125) In this case. 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 . A. (13. 2. A. C. Nuclear Physics B33. C.126) This shows that in this gauge the ghosts decouple completely and can be absorbed into the normalization factor. Theoretical and Mathematical Physics 10. .7 References 1. 436 (1971).7 References 581 G[f a ] = e i − 2ξ R d4 x f a f a . Stora. Becchi.124) However. 13.13. Physics Letters 52B. It would not work in say. therefore. A. J. The axial-like gauge choices (13.

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

L = − Fµν F µν + 4 2 583 (14. 2 M 2 a µa Aµ A .2) . we also know of physical forces in nature (such as the weak force) that are short ranged and this would seem to suggest (intuitively) that the gauge ﬁelds associated with such forces may be massive. 12 and 13 we have studied gauge theories (both Abelian and non-Abelian) in great detail.Chapter 14 Higgs phenomenon and the standard model In chapters 9.14) or (12. a massless gauge ﬁeld is completely consistent with the observed fact that electromagnetic forces are long ranged.34) respectively.1) is not invariant under the gauge transformation (9. 2 or (14. One of the striking features of these theories is that the gauge ﬁelds are massless simply because the action for a mass term for the gauge ﬁeld in the Lagrangian density M2 Aµ Aµ .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 . However. Therefore. in this chapter we would discuss this important question of massive gauge ﬁelds concluding with the standard model of the electroweak interactions. Therefore. 14. As we have mentioned earlier gauge ﬁelds can be thought of as the carriers of physical forces.

5) deﬁnes the Proca equation (and (14.2) has the form ∂L ∂L − = 0.2) denotes the Proca Lagrangian density) and it is clear that this system of equations deﬁnes a massive spin 1 ﬁeld theory (a massive photon).2) does not have a gauge invariance (because of the mass term) unlike the Maxwell theory.6) Equation (14.4) with ∂ν and using the anti-symmetry of the ﬁeld strength tensor. section 13. (14.5) in a simple manner and in momentum space has the form . we obtain ∂ · A = 0.2) describes a massive photon which can be seen as follows. for example. Note that the ﬁeld variable Aµ has four ﬁeld degrees of freedom while the constraint (14. (14. The Euler-Lagrange equation following from (14. As has been mentioned earlier.6) together with the constraint (14. or.5) and substituting this back into the Euler-Lagrange equation in (14. ∂∂µ Aν ∂Aν ∂µ or. we also note here that the theory (14. (14.3) The Lagrangian density (14. or. + M 2 Aν = 0.4) Contracting (14. The propagator for this theory can be worked out (see.4) yields ∂µ (∂ µ Aν − ∂ ν Aµ ) + M 2 Aν = 0. (14. ∂µ F µν + M 2 Aν = 0.5) eliminates one degree of freedom leaving us with three ﬁeld degrees of freedom which is the correct number of dynamical ﬁeld degrees of freedom for a massive spin 1 ﬁeld. −∂µ F µν − M 2 Aν = 0.584 14 Higgs phenomenon and the standard model where the ﬁeld strength tensor is deﬁned as in the Maxwell theory Fµν = ∂µ Aν − ∂ν Aµ .

In contrast to the Proca theory (14. the St¨ckelberg theory u (14.8) = − Fµν F µν + 4 2 2 which.83) as well as for discussions in connection with the Higgs phenomenon in the next section). (14. To see the invariance of the full theory. but we have kept it for consistency with earlier discussion. M (14. e M θ(x). χ(x) → χ(x) − e (14.10) . 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 − . of course.8) was constructed to be invariant under a Maxwell-like gauge transformation Aµ (x) → Aµ (x) + 1 ∂µ θ(x). also contains a dynamical charge neutral spin zero ﬁeld (real scalar ﬁeld) which mixes with Aµ . in addition to the spin 1 ﬁeld.7) M2 1 1 Aµ Aµ + ∂µ χ∂ µ χ + M Aµ ∂µ χ.9).2). The limit M → 0 of the Proca theory is clearly quite subtle. 2 −M M2 (14.9) where e denotes the coupling constant of QED (this can be set to unity. As a result of these diﬃculties.1 Stuckelberg formalism 585 iGµν (p) = − p2 Thus. 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. it is suﬃcient to note that under the transformation (14.9) Aµ + M 1 1 1 ∂µ χ − ∂µ χ → Aµ + ∂µ θ + ∂µ θ M e M e 1 = Aµ + ∂µ χ.¨ 14. say in (9. This leads to diﬃculties in establishing renormalizability of the (interacting) Proca theory for massive photons. invariant under the gauge transformation (14. The ﬁeld strength tensor is.

2 −M M2 iGµν (p) = − p2 (14. M θ(x) = (14.586 14 Higgs phenomenon and the standard model Because of the gauge invariance of the theory. . the there is no diﬃculty in establishing renormalizability in such a gauge. in this unitary gauge.13) We recognize this to be the Proca theory (14. but would lead to propagators that are well behaved at high momentum. Furthermore. As a result. (14. section 12. for example. the massless limit M → 0 is now straightforward. if we choose a covariant gauge condition which leads to a gauge ﬁxing Lagrangian density of the form (see.2) and this makes it clear that the Proca theory can be thought of as the gauge ﬁxed St¨ckelberg theory.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 .14) has the unpleasant features described earlier. L = − Fµν F µν + 4 2 (14. 2ξ − (14. However.11) which can be achieved by the choice of the gauge transformation parameter e χ(x).15) this would induce a ghost action. This is typical of the behavior of theories in a unitary gauge. On the other hand.7)) i pµ pν ηµν − . we can choose a gauge and if we choose the unitary gauge χ(x) = 0. the gauge propu agator (see also (14.8) takes the form M2 1 Aµ Aµ .12) then the Lagrangian density (14.

(12. say.41)) Aµ → U Aµ U −1 − V → U V. Aν ] . (14.16) ig ig where the trace is over the fundamental representation of the group.18). SU (n) (as we have discussed earlier in section 12.¨ 14. ig (14. We will now brieﬂy describe the St¨ckelberg formalism for non-Abelian gauge u theories.1). 1 (∂µ U )U −1 . 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 .38) and (12. the Lagrangian density for the gauge ﬁeld is invariant under the transformation (14. = e M χ. for example. Here the ﬁeld strength tensor and the operator V are deﬁned as Fµν V = ∂µ Aν − ∂ν Aµ + ig [Aµ .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.17) where χ denotes a matrix valued scalar ﬁeld. ig (14. it is suﬃcient to note that .18) As we have already seen in chapter 12. This Lagrangian density can be easily checked to be invariant under the gauge transformations (see. To see that the complete Lagrangian density is also invariant.

18). 2 − M2 p M2 iGab (p) = − µν (14.20) which can be achieved by the choice of the gauge transformation matrix U (x) = V −1 (x).19) Using the cyclicity under trace it is now straightforward to see that the full theory is invariant under the gauge transformation in (14. ig (14. then the Lagrangian density (14.16) becomes 1 L = − Tr Fµν F µν + M 2 Tr Aµ Aµ . If we choose the unitary gauge condition V = ½. In this gauge. as before. 2 (14.2) to the non-Abelian case and describes a massive spin 1 ﬁeld belonging to SU (n). 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 − .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 .21) (14.23) .22) This can be thought of as a generalization of the Proca theory (14. (14.

(14. e δφ = −iǫ(x)φ(x). if we choose a covariant gauge ﬁxing condition (a generalization of (14.24) can be checked easily (see (7.85)) to be invariant under the inﬁnitesimal local transformations δφ† = iǫ(x)φ† (x). the propagator will have the correct asymptotic behavior for large values of the momenta.15)).2 Higgs phenomenon In section 7. (14. (14.26) .6) λ 1 2 L = − Fµν F µν + (Dµ φ)† (D µ φ) + m2 φ† φ − φ† φ . but now interacting with an Abelian gauge ﬁeld as well (scalar QED).83) Dµ φ = ∂µ φ + ieAµ φ. For simplicity we consider again the self-interacting theory of a complex scalar ﬁeld (as in section 7.5). The Lagrangian density for such a theory is given by (with λ > 0. 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).2 Higgs phenomenon 589 with the unpleasant features discussed earlier that are characteristics of the unitary gauge choice. On the other hand.14. This is known as the Nambu-Goldstone phenomena and the massless scalar ﬁeld is known as the Nambu-Goldstone boson.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. 14.25) The Lagrangian density (14. see also section 7.24) 4 4 where the theory has a “wrong” sign for the mass term and the covariant derivative is deﬁned as in (7. In that case.84) and (7. Let us next ask what happens if there is a spontaneous breakdown of a local symmetry. δAµ = 1 ∂µ ǫ(x).

let us rewrite the Lagrangian density (14.42)) 1 φ = √ (σ + iχ). as we have discussed in section 7.) Rather the true vacuum of the theory satisﬁes (see (7. (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 . this actually corresponds to the local maximum of the potential. On the other hand.5.24) in terms of the σ. 2 χc = χ = 0.29) For further analysis. there is actually an inﬁnity of minima of the potential. (As we have seen in section 7. since the mass term for the scalar ﬁeld has the “wrong” sign.28) (14.77)) .30) 2 16 Let us now rewrite the theory around the true vacuum by shifting the ﬁeld variables as (see (7. the normal vacuum for which φ = φ† = 0. λ where (see also (7. this is just one of them) 2m σc = σ = √ .64).26) does not allow a mass term for the gauge ﬁeld.27) does not correspond to the true vacuum.5. (14.590 14 Higgs phenomenon and the standard model The gauge boson in this theory is massless because the gauge symmetry (14.

Let us look at only the terms in (14. (14.31) where we have identiﬁed the vacuum expectation value (vev) of the ﬁeld with 2m σ =v= √ .2 Higgs phenomenon 591 2m σ → σ + σ = σ + √ = σ + v. (14. the Lagrangian density (14. 16 (14. λ In this case.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 (14.32) 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). λ χ → χ + χ = χ.34) 2 2 We note here that the χ ﬁeld appears to be massless in this theory as in (7. but it now mixes with the gauge ﬁeld Aµ so that we have .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 .33) where we have used the form of v in (14.32) to simplify some of the terms.

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 . (The Lagrangian density in (14.39) . ev Bµ = Aµ + then we have (14.38) and a massive gauge ﬁeld Bµ with mass (this is the analog of the unitary gauge in (14.35) ∂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. let us deﬁne 1 ∂µ χ.8). As in (7. σ (14.37) 2 We see from (14.) To achieve this.11)) 4e2 m2 .37) that the quadratic part of the shifted Lagrangian density is completely diagonalized.592 14 Higgs phenomenon and the standard model to diagonalize the theory in order to determine the true spectrum of the theory.36) Thus in terms of this new ﬁeld variable (14.34). (14.79).35) the quadratic Lagrangian density (14. it describes a massive scalar ﬁeld σ with mass m2 = 2m2 . ignoring the σ ﬁeld should be compared with the St¨ckelberg Lagrangian density u (14. λ m2 = M 2 = e2 v 2 = B (14. ev e v Aµ + Bµ B µ = (14.

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

it is not the right gauge to study renormalizability of the theory.39).41) e where M is deﬁned in (14. δχ(x) = −ǫ(x)σ(x). This is achieved by the ’t Hooft gauge ﬁxing which is described by the gauge ﬁxing Lagrangian density LGF = − = − 1 1 µ (∂ Aµ − ξevχ)2 = − (∂ µ Aµ − ξM χ)2 . 2ξ 2ξ ξM 2 2 1 µ (∂ Aµ )2 + M (∂ µ Aµ )χ − χ . The presence of the local symmetry (14. (14.33) which reﬂects a spontaneous breakdown of the local symmetry (14. e 1 ∂µ ǫ(x).594 14 Higgs phenomenon and the standard model Although the unitary gauge is useful for the counting of true dynamical degrees of freedom in the theory. First we note that the unshifted theory (14.40) e The shifted theory (14. To introduce the gauge choice.42) .34) (namely.84)) δσ(x) = ǫ(x)χ(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). eliminate the mixing term between the Aµ and the χ ﬁelds). 1 ∂µ ǫ(x). that is more useful for this purpose let us proceed as follows.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. commonly known as the ’t Hooft gauge or the Rξ gauge. as pointed out in the last section. δχ(x) = −ǫ(x)(v + σ(x)) = −ǫ(x) δAµ (x) = M + σ(x) .30) is invariant under the inﬁnitesimal gauge transformations (see (7. (14. 2ξ 2 (14.

It is clear that the mixing terms in (14. disappear from the action. The gauge ﬁxing Lagrangian density (14. − ξM 2 pµ pν i ηµν − (1 − ξ) 2 .2 Higgs phenomenon 595 which can be compared with (14. 2 (14. The complete Lagrangian density in this gauge has the form LTOT = L + LGF + Lghost .45) We see that unlike in the conventional covariant gauge ﬁxing (12. thereby.44)) leads to an interacting ghost Lagrangian density.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.164).34) together with the cross terms in (14.14. (14.43) and we note that in this gauge the propagators fall oﬀ for large values of the momenta unlike in the unitary gauge. the ’t Hooft gauge choice in (14. 2 −M p − ξM 2 (14.15). As a result. the propagators for the Aµ and the χ ﬁelds in this gauge are given respectively by iGµν (p) = − iGχ (p) = p2 p2 i . (14.42) combine into a total divergence and.170)) LGF = ξ 2 F − F (∂ · A − ξM χ).42) (or (14.46) and can be easily checked to be invariant under the BRST transformations .42) can also be written with an auxiliary ﬁeld as (see (12.

4 (see. the β decay of the neutron etc. M + σ c.47). (14. Therefore. This is the starting point for establishing the renormalizability of the gauge theory with a spontaneously broken symmetry (Higgs theory). δc = −ωF.47) Following the discussions in section 13.91)) 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. δF = 0. e δχ = −ω δc = 0. 2 (14. e δσ = ωχc. 14. At low energies.3 The standard model The standard model or the gauge theory of electroweak interactions was built on the results of years of earlier work on weak interactions.596 14 Higgs phenomenon and the standard model δAµ = 1 ω∂µ c.49) . let us review some of these essential earlier results before constructing the standard model (known as the Weinberg-SalamGlashow theory). δAµ (x) δσ(x) δPσ (x) δχ(x) δPχ (x) δc(x) (14. (13.48) where Pσ and Pχ denote respectively the sources for the composite BRST variations of the ﬁelds σ and χ in (14.) are described quite well by a speciﬁc form of the Fermi theory which involves four fermion interactions in the form of a current-current interaction Hamiltonian density GF † HI = √ Jµ J µ . weak interactions (for example. for example.

ψν denoting the fermion ﬁelds associated with neutron.140) for a deﬁnition of right and left-handed particles/ﬁelds).52) where Q denotes the electric charge of the particle. therefore. (14. It was known that all the weakly interacting particles (leptons and hadrons) can be described by multiplets belonging to the weak isospin group of SU (2). I3L the quantum number associated with the 3-component of the weak isospin generator and Y its weak hypercharge quantum number. The coupling constant (interaction strength) GF is known as the Fermi coupling and is determined from low energy experiments to have the value GF = 1.51) The weak interactions are. 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 . (14.14. Experimentally . (The V − A structure is connected with the fact that parity is violated maximally in weak interactions. it is clear from the V − A (vector minus axial vector) structure of the currents in (14.17 × 10−5 (GeV)−2 .50) with ψn . electron and neutrino respectively.139) and (3.3 The standard model 597 where the current is a sum of the hadronic and leptonic currents of the form (had) (lept) Jµ = Jµ + Jµ = ψ p γµ (1 − γ5 )ψn + ψ ν γµ (1 − γ5 )ψe + · · · . proton. Furthermore.50) have a weak isospin structure and are also known as the weak isospin currents. the currents in (14. ψp . quite weak and were also known to be extremely short ranged. As a result.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. ψe . 2 Q = I3L + (14.) In addition to the weak isospin quantum number.

) Therefore. therefore. we have already seen that gauge theories can describe physical forces and do not have any problem of renormalizability (we will discuss renormalizability and renormalizability of gauge theories in a later chapter). Furthermore. the V − A theory cannot be the fundamental theory underlying weak interactions (forces). since weak interactions are short ranged. (The nonrenormalizable character is already manifest in the fact that the coupling constant of the theory (14.53) where SUL (2) and UY (1) denote respectively the weak isospin and the weak hypercharge symmetry groups and we would like this symmetry group to spontaneously breakdown (at the weak interaction scale) in a way such that only the electromagnetic symmetry survives as the true symmetry of the low energy theory. We have seen in the last section that gauge bosons can become massive through the Higgs mechanism which would involve the spontaneous breakdown of some local symmetry. Namely.51) is dimensional with inverse dimensions of mass (energy). As we have discussed weak interactions violate both the weak isospin and the weak hypercharge symmetries and. it is clear that we should look for a gauge theory of weak interactions based on the local symmetry group SUL (2) × UY (1). 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. let us construct the standard model in several steps. On the other hand. it is not renormalizable. (14. As a result. we would like the gauge theory to describe the symmetry behavior (“SSB” denotes spontaneous symmetry breakdown) SSB SUL (2) × UY (1) −→ UEM (1). As a result we may try to formulate the weak interactions as a gauge theory such that it reduces to the V − A theory in the low energy limit. it is logical to associate the Higgs mechanism to the spontaneous breakdown of these symmetries.54) With these basics. Although the V − A theory with four fermion interactions works quite well at low energy.598 14 Higgs phenomenon and the standard model it was known that weak interactions violate both weak isospin and weak hypercharge quantum numbers. (14. .

the left-handed doublets have Y = −1 while for the righthanded singlets carry a hypercharge of Y = −2. These include both hadrons and leptons as we have already mentioned. its ﬁeld can be decomposed into a left-handed part and a right-handed part. The other lepton families (µ and the τ families) can be introduced through an obvious generalization. therefore. unlike the V − A theory (see (14. The standard model is expected to describe the weak and electromagnetic interactions of all fundamental particles that participate in the weak processes. namely. In contrast. 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.56) 2 ψ = ψL + ψR . Let us ﬁrst discuss the ﬁeld content of the stan- dard model.55) where the left. the electron and its neutrino. However. The matter ﬁelds associated with the electron family can be grouped as multiplets of SUL (2) as νe e L L= . . for simplicity we will consider the matter to consist of only one family of leptons.6). 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).3. However. The left-handed multiplets of the standard model are doublets of SUL (2) while the right-handed particles correspond to singlets.14. The only diﬀerence is that since all the quark ﬁelds are massive. We note that the electron is massive and. Thus.1 Field content. The quarks as well as the other families can be introduced in a straightforward manner. The quark families can also be described by such structures. (14.and the right-handed components of a spinor ﬁeld are deﬁned by (see also section 3.50) where the hadronic current was given in terms of neutron and proton ﬁelds) the matter ﬁelds in the standard model consist of quarks and leptons and they come in three families.3 The standard model 599 14. eR = R.7) 1 ψL = (½ − γ5 )ψ. the neutrino is experimentally known to be essentially left-handed (see section 3. (14. although the leptons are believed to be fundamental particles. we now know that the constituents of hadronic matter are the quarks.

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

Note that ǫabc denotes the structure constant of SUL (2) as we know from the study of angular momentum. (14. but it will be present in the matter Lagrangian density.44)) Fµν a Fµν = ∂µ Yν − ∂ν Yµ .52). 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. (The Abelian gauge ﬁelds are not self-interacting which is why g′ does not appear in the Lagrangian density for the gauge ﬁelds. let us note that the gauge invariant Lagrangian density for the gauge ﬁelds follows from our earlier discussions (see (9. 3.61) 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.3.4) and (12. a a b c = ∂µ Wν − ∂ν Wµ − gǫabc Wµ Wν .3 The standard model 601 Fields νe e eR φ+ φ0 L I3L 1 2 Y 2 1 −2 1 2 1 −2 I3L + − − 1 2 1 2 Y 2 Q 0 −1 −1 1 0 (14. 14.14. 2.59) =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. 4 4 a = 1.60) .60) where the ﬁeld strength tensors for the Abelian as well as the nonAbelian gauge ﬁelds are deﬁned as (see (9. (14.) The Lagrangian density (14.2 Lagrangian density.16) and (12.45)) to correspond to 1 1 a LG = − Fµν F µν a − Fµν F µν . First.

the Lagrangian density for the leptons takes the form ig′ σ · Wµ + Yµ L + iRγ µ ∂µ + ig′ Yµ R.602 14 Higgs phenomenon and the standard model is clearly invariant under the inﬁnitesimal gauge transformations of SUL (2) and UY (1) deﬁned by 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) . g g 1 ∂µ ǫ(x). An alternative a way to see this is to note that the ﬁelds Wµ do not carry weak hypercharge quantum number just as Yµ does not carry weak isospin quantum number and. therefore.62) a δWµ (x) = δYµ (x) = where ǫa (x). Since the two symmetry groups are commuta ing (direct products). 2.63) is that the right-handed fermions couple twice as strongly to the Yµ . they are inert under the corresponding transformations. 2 2 (14. We have also used a vector notation here to represent the three components of the isospin vectors which is often used interchangeably. the gauge ﬁeld Wµ for the group SUL (2) does not transform under the group UY (1) and vice versa. a = 1. 3 and ǫ(x) denote the local inﬁnitesimal parameters of symmetry transformations associated with the groups SUL (2) and UY (1) respectively. we simply have to remember that there are two local symmetries present in the theory. (We note here that within the context of isospin. The gauge invariant Lagrangian density for the matter ﬁelds (leptons) can be obtained through minimal coupling.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. g′ (14. sometimes it is conventional to denote the generators as τ by making the identiﬁcation 2 σ = τ .) The other thing to note from the structure of (14. Since the left-handed fermions belong to an isospin doublet while the right-handed fermions are isospin singlets.

In fact. we also need a Lagrangian density for the scalar Higgs ﬁelds which are expected to lead to a spontaneous breakdown of the local symmetries. 2. 3 into a vector ǫ(x) in (14.64). 2 2 δR(x) = −iǫ(x)R(x). The minimally coupled fermion Lagrangian density (14.62). a = 1.66) together with the gauge ﬁeld transformations in (14. δL(x) = −i (14. therefore. massless in this theory to begin with). Besides the gauge ﬁelds and the leptons. since we . 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. The Lagrangian density (14. the coupling to the UY (1) gauge ﬁeld is correspondingly with the opposite sign.64) together with the transformations in (14.65) can be checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)φ(x) + ǫ(x)φ(x).64) show explicitly that the right-handed fermions do not change under the SUL (2) transformations.62). A mass term as well as a potential for the Higgs ﬁeld can be introduced consistent with the SUL (2) symmetry. The transformations (14. Note also that the SUL (2) invariance forbids a mass term for the lepton ﬁelds (the electron and the neutrino are. 2 2 (14.63) can be easily checked to be invariant under the inﬁnitesimal local transformations i σ · ǫ(x)L(x) − ǫ(x)L(x).3 The standard model 603 ﬁeld as the left-handed ones simply reﬂecting the Y quantum numbers for the two ﬁelds.65) We note that since the hypercharge of the Higgs doublet is just the opposite of that of the left-handed lepton doublet. We note that we have combined the three components of ǫa (x). 2 2 δφ(x) = −i (14.

24).68) can be checked to be invariant under the transformations (14.63). δL(x) = −i δφ(x) = −i a δWµ (x) = i σ · ǫ(x)φ(x) + ǫ(x)φ(x). (14.68). (14. (14. 4 λ > 0.60). it is worth noting that in spite of the similarity in their forms for the potential.69) which is invariant under the inﬁnitesimal local gauge transformations i σ · ǫ(x)L(x) − ǫ(x)L(x). 2 2 1 1 b (Dµ ǫ(x))a = ∂µ ǫa (x) − gǫabc Wµ (x)ǫc (x) . (14. 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 . here the Higgs ﬁeld is represented by an isospin doublet) V φ. φ† .70) δYµ (x) = . collecting all the terms in (14. the Yukawa interaction leads to masses for fermions after the spontaneous breakdown of the symmetry. In addition.68) where h denotes the Yukawa coupling and the Lagrangian density (14.64) and (14. of course. Therefore. As we will see shortly. g′ (14. (14.66).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 (14.66). (14.67) which is invariant under the local transformations in (14. φ† = −m2 φ† φ + λ † 2 φφ .67) and (14. 2 2 δR(x) = −iǫ(x)R(x). the complete weak interaction Lagrangian density describing just one family of leptons (electron family) can be written as L = LG + Lf + LH + LY − V φ.65).

4 V φ.72) λ † φφ 2 As we have seen in the last section. λ (14.74) . 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. φ† = 0.2) that the ﬁrst solution in (14. φ† = −m2 φ† φ + λ > 0. the weak interactions are short ranged so that we would like the gauge bosons to be massive.73) represents a local maximum while the second solution corresponds to the true minimum of the potential.3. there are two solutions to (14.73) φ φ† φ = = Furthermore. λ (14.71) so that the minimum of the potential occurs at ∂V ∂φ ∂V ∂φ† λ † φφ 2 = φ† − m2 + = − m2 + = 0. 2m v=√ . (14. we have noted earlier in section 7.3 The standard model 605 14. furthermore.14. Furthermore. the factor of √2 below is a consequence of the deﬁnition in (14. 2m2 .29)) 0 v √ 2 φ = .5 (and also in section 14. (14. We know that both SUL (2) and UY (1) symmetries are violated in weak interactions (the associated quantum numbers are not conserved in weak processes).72). We have chosen the Higgs potential (14.67) in the form λ † 2 φφ . 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. namely. φ = 0.3 Spontaneous symmetry breaking.

606 14 Higgs phenomenon and the standard model Therefore. 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+ .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 (14.58) we obtain .77) σ · Wµ = 2 = 1 2 √ + 2Wµ φ+ φ0 . (Basically. it follows that 3 Wµ √ − 2Wµ (14. (14. this is the sector of the theory that involves the scalar ﬁeld and will be aﬀected by the shift.) We note that we can write explicitly in the matrix form 3 Wµ 2 1 Wµ − iWµ σ · Wµ φ = 2 = 1 2 1 2 1 2 3 Wµ + iWµ −Wµ √ + 3 2Wµ Wµ . the other parts will not change under the shift.75) However.68)). (14. √ − 3 2Wµ −Wµ (14.65).67) and (14. (14. as we have discussed earlier. before shifting the ﬁelds let us rewrite the Lagrangian densities for the Higgs and the Yukawa sector of the theory for simplicity (see (14.76) where we have deﬁned 1 ± 1 2 Wµ = √ (Wµ ∓ iWµ ). 2 As a result.

2 (14.74) can be done equivalently by letting v φ0 → φ0 + √ .14. 2 v 0 0 φ → φ +√ . 4 (14.79) in the Higgs and the Yukawa sector becomes LH + LY − V φ.80) under which the Lagrangian density (14. φ† = ∂µ + ig − 0 ig 3 ig′ v Yµ − Wµ φ− − √ Wµ φ + √ 2 2 2 2 .3 The standard model 607 LH + LY − V φ.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 .

However. 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 ).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 (14.60) and (14.63) respectively. (g′ Yµ + gWµ ) as well ± appear to have become massive. The spectrum of the theory can be obtained from the quadratic part of the Lagrangian density and the quadratic part of (14. since as the ﬁelds Wµ there is mixing between various ﬁelds.82) hv λv 2 0 0 2 φ +φ .81) gives ∂µ φ− ∂ µ φ+ + ∂µ φ ∂ µ φ0 + + + 0 igv + µ − igv − µ + Wµ ∂ φ − Wµ ∂ φ 2 2 iv g2 v 2 + µ− 0 3 Wµ W − √ (g′ Yµ + gWµ )∂ µ (φ − φ0 ) 4 2 2 v2 ′ 3 g Yµ + gWµ 8 g′ Y µ + gW µ3 (14.81) . we cannot truly determine the spectrum of the theory until we diagonalize the quadratic part of the Lagrangian density. Adding the quadratic terms from the gauge and the fermion Lagrangian densities in (14.

14. Yµ = Zµ sin θW + Aµ cos θW . Let us deﬁne g′ g2 + g′2 g g2 + g′2 Zµ = Yµ + 3 Wµ 3 = sin θW Yµ + cos θW Wµ . (14. (14.85) .84) to write 3 Wµ = Zµ cos θW − Aµ sin θW .83). We can invert the relations in (14. 8 2 It is now a straightforward matter to diagonalize the Lagrangian density (14.83) + λv 2 0 hv 0 2 − √ (eR eL + eL eR ) − φ +φ .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 (14.84) where θW is known as the Weinberg angle (also called the weak mixing angle). Aµ = g g2 + g′2 Yµ − g′ g2 + g′2 3 Wµ 3 = cos θW Yµ − sin θW Wµ .

83) as .86) 1 √ (σ 2 ∂µW ν 3 − ∂ν W µ 3 Similarly. 2 2 (14. (14.84)-(14. 2 cos θW hv √ . if we deﬁne the parameters gv . (14.29)) σ = χ = 1 0 √ φ + φ0 .87) so that we can write 0 ∂µ φ ∂ µ φ0 = 1 1 ∂µ σ∂ µ σ + ∂µ χ∂ µ χ. we can deﬁne (this is equivalent to deﬁning φ0 = iχ). 2 MW g2 + g′2 v = . φ = 0 1 √ (σ 2 + − iχ). 2 λ 4m2 λv 2 = 2 2 λ = 2m2 . see. 2 (14.89).89) MW MZ = = me = 2 MH = then together with (14. (14. for example.610 14 Higgs phenomenon and the standard model With this. 2 i 0 √ φ − φ0 . we can write the quadratic part of the Lagrangian density (14. 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 ) .88) Furthermore.

(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 .56)) e = eL + eR .14. eR to deﬁne the massive electron ﬁeld as (see (14.90) 2 2 where we have combined the two chirality states eL . If we now redeﬁne the (massive) gauge ﬁelds as (14.91) .

2 (14.94) while the neutrino remains massless as we would like. 2 (14. The charged ± vector (spin 1) bosons Wµ are massive with mass MW = gv . Zµ is also massive with mass .93) is now completely diagonalized and shows some very desirable features. MW 1 ∂µ χ → Zµ .612 14 Higgs phenomenon and the standard model ± Wµ ∓ i ± ∂µ φ± → Wµ .95) One of the two neutral vector bosons.92) Zµ − then the quadratic Lagrangian density (14. MZ (14. 1 2 2 M σ 2 H (14. namely. we see that after spontaneous breakdown of symmetry the electron has become massive with mass hv me = √ . The Higgs √ particle has the usual mass MH = 2m (see (14. Although the starting theory had a massless electron.38)).90) becomes completely diagonal with the form LQ = − 1 + + ∂µ Wν − ∂ν Wµ 2 2 + ∂ µ W ν − − ∂ ν W µ − + MW Wµ W µ − 1 M2 − (∂µ Zν − ∂ν Zµ ) (∂ µ Z ν − ∂ ν Z µ ) + Z Zµ Z µ 4 2 1 1 − (∂µ Aν − ∂ν Aµ ) (∂ µ Aν − ∂ ν Aµ ) + ∂µ σ∂ µ σ − 4 2 / / +e(i∂ − me )e + iν eL ∂ νeL .93) The quadratic Lagrangian density (14.

97) There is another neutral vector boson in the theory (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). Let us look at the minimally coupled fermion Lagrangian density (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 .93). Aµ which is massless. That this is indeed true can be seen as follows.14. 2 cos θW (14.96) where the Weinberg angle θW is deﬁned by (see (14. . namely.84)) g′ g2 + g′2 g g2 + g′2 sin θW = cos θW = . (14.

in addition to the correct electromagnetic current.98) We note from (14. We also note from (14.98) that the massless neutral vector boson Aµ couples only to the charged fermions as the photon should and we can identify the electric charge as gg′ g2 + g′2 e= = g sin θW = g′ cos θW . (14.614 − 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. 2 2 2 . (14.99) Thus. 2 2 2 g g + Jµ = − √ eL γµ νeL = − √ eL γµ (1 − γ5 ) νeL .98) that the standard model has. 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. both charged as well as neutral weak currents given by g g − Jµ = − √ ν eL γµ eL = − √ ν eL γµ (1 − γ5 ) eL .

8MW (14. If we look at the charged current interaction terms in (14.14.49) (for just one family of leptons) and comparing the two we determine G √F 2 g2 .1: Charged current-current interaction in the lowest order. 4 2GF 4 2GF sin2 θW = (14. two charged currents interact through the charged W gauge boson propagator (exchange) as shown in Fig.103) Experimental measurements lead to a value of the Weinberg angle as .101) This has exactly the form of the V − A four-fermion interaction in (14.102) which leads to 2 MW = (14. W Figure 14. 14.98).100) We recall that the V − A theory describes only charged current interactions. (14.1 and the propagator reduces 2 to a multiplicative factor of M12 in the low energy limit p2 ≪ MW ) W HI = g2 µ 2 ν eL γµ (1 − γ5 ) eL eL γ (1 − γ5 ) νeL . 2 8MW g2 e2 √ = √ . they would lead to a low energy eﬀective current-current interation Hamiltonian density in the lowest order of the form (basically.3 The standard model 0 Jµ = − 615 g2 (ν eL γ µ νeL − eL γ µ eL ) 1 2 g2 + g′2 +g′2 (ν eL γ µ νeL + eL γ µ eL + 2eR γ µ eR ) .

Search for Higgs particles is a top priority in many experiments at the upcoming LHC. e2 √ 4 2GF sin2 θW MW ≃ 90 GeV. the WeinbergSalam-Glashow theory does not predict a unique mass for the Higgs particle since its mass depends on the quartic coupling λ. Proca. Furthermore. 14. (14.23. However. A. although bounds on the value of its mass can be put from various other arguments. (14. (14.4 References 1.616 14 Higgs phenomenon and the standard model sin2 θW ≃ 0.49) 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. Such processes have also been experimentally observed. the standard model diﬀers from the conventional Fermi theory in (14. for example. .51) and recall that we have set = c = 1) e2 1 ≃ . et le Radium 7.104) and using the values of other constants (see. The standard model compares very well with the experimental results. 347 (1936). 4π 137 we can now determine MW MZ = = GF ≃ 10−5 (GeV)−2 . cos θW 1 2 ≃ 80 GeV. 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). J. de Phys. Once again these can be accomodated into extensions of the standard model that we will not go into here.105) These gauge bosons have been discovered and their masses are determined very close to the theoretically predicted values.

209 (1957). Physical Review 109. G. Physical Review Letters 13. Weinberg. 1860 (1958). 7. E. R. Proceedings of PaduaVenice conference on mesons and newly discovered particles. 11. E. Marshak. Ruiz-Altaba. C. Physical Review Letters 13. 8. P. St¨ckelberg. S.4 References 617 2. 9. C. G. H. Englert and R. Bernstein. C. 7 (1974). Gell-Mann. 3265 (2004). G. 1156 (1966). Helvetica Physica Acta 30. Hagen and T. ’t Hooft. 321 (1964). F. 193 (1958). 6. W. 585 (1964). E. Ruegg and M. 10. u 3. Higgs. J. P. International Journal of Modern Physics A19. 167 (1971). Brout. Sudarshan and R. (1957). Feynman and M. 4.14. Kibble. Physical Review 109. . Physical Review 145. W. G. 5. Nuclear Physics B35. R. Physical Review Letters 19. Guralnik. 1264 (1967). S. Reviews of Modern Physics 46. B.

This page intentionally left blank .

as we go to higher orders in perturbation. For example. we have to ﬁnd a way of extracting meaningful results from such a quantum ﬁeld theory. This procedure is known. the Feynman diagrams become topologically more complicated and may contain internal propagators whose momenta are not uniquely determined in terms of the external momenta.Chapter 15 Regularization of Feynman diagrams 15.6 there was only one internal propagator). the evaluation of Feynman diagrams may lead to divergences depending on the structure of the integral that is being evaluated.6. 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. any scattering matrix element (S matrix element) in a given quantum ﬁeld theory can be built out of 1PI (one particle irreducible) graphs. as renormalization of a quantum ﬁeld theory and involves several steps. As a result. The simple diagrams in 9. in general. 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. let us look at one of the simplest interacting ﬁeld 619 . Instead some of the internal propagators may involve additional momenta (loop momenta) that are integrated over all possible values. We will study this question in a systematic manner developing the necessary ideas in this chapter as well as in the next. In such a case.4). We have also carried out a few calculations of simple Feynman diagrams describing physical processes in section 9.1 Introduction So far we have discussed the basic structures of various quantum ﬁeld theories. However.

15. 2 2 4! L= (15. given these Feynman rules.1 (which topologically involve loops and. we can calculate any perturbative amplitude in the φ4 theory. 15. (15. With these Feynman rules. . for example. it is suﬃcient to concentrate on the 1PI graphs. Clearly the 1PI diagrams are fundamental since any other diagram can be built using them as basic elements.1) where the coupling constant λ is assumed to be positive. − M2 p p4 p3 = iGF (p) = p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ). the scalar φ4 theory described by the Lagrangian density (see (6.36)) M2 2 λ 4 1 ∂µ φ∂ µ φ − φ − φ .1: A few examples of 1PI diagrams in the φ4 theory. Therefore. In fact. are called loop diagrams) Figure 15. The Feynman rules for this theory are given by i . is not. namely. in studying quantum ﬁeld theories at higher orders. which are 1PI graphs whereas the graph in Fig.620 15 Regularization of Feynman diagrams theories.2) p1 p2 where all the momenta in the vertex are assumed to be incoming and we have not written the “iǫ” term in the propagator explicitly although it should be understood. therefore.2. we can construct the graphs in Fig.

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

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. is given by L = I − V + 1. 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).4) and (15. therefore. (15. but for a clear understanding of the loop expansion we are going to restore in this section only) i Z[J] = N Dφ e SJ . 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. (15.5) where the term (+1) reﬂects the fact that every 1PI vertex function has an overall energy-momentum conserving delta function associated with it.622 15 Regularization of Feynman diagrams P = I − V. we can eliminate I − V to obtain L = P + 1. From the two relations in (15.120). (12. (15. Let us recall that in the path integral description of a quantum ﬁeld theory.3). the generating functional is deﬁned as (see. we have set = 1 throughout our discussion.6) In other words.5). for example. . (15. therefore. be thought of as the parameter of expansion in the number of loops. the power of “a” in a diagram in such a theory also determines the number of loops in the diagram and can.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.

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. Loop expansion is a diﬀerent perturbative expansion. the expansion is unaﬀected by how one divides the total action into a free part and an interaction part. In such a case.3 Cut-oﬀ regularization To get an idea of how divergences arise in a quantum ﬁeld theory at the loop level. On the other hand.6). The reason why such an expansion is advantageous can be seen from the fact that since this parameter multiplies the whole action. the entire perturbation expansion has to be rearranged which is nontrivial. for example.15.3 Cut-off regularization 623 As a result. 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. the scalar ﬁeld picks up a vacuum expectation value which depends on the coupling constant of the theory. when we perturb around the true vacuum (by shifting the ﬁeld variable by its vacuum expectation value).1) which gives the one loop correction to the two point function. 15. In addition the small value of makes it a legitimate expansion (perturbation) parameter. Therefore. expanding in powers of in a quantum ﬁeld theory also corresponds to expanding in the number of loops as described in (15. as we have seen (see. for example. when we study theories where there is spontaneous breakdown of a symmetry. sections 7.5 and 14. This is quite important.2). We are usually accustomed to the idea of perturbation in powers of the coupling constant in the theory.

With this regularization. Furthermore.) Since large values of momentum correspond to small values of coordinate separation.8) that this integral is divergent and.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 .9) = 2π 2 .8) takes the form .8) dy 1 − Here we have used the fact that the Feynman diagram in (15. (This is the reason for the name cut-oﬀ regularization. y + M2 (15. We see from (15. the two point function in (15. such divergences are also known as ultraviolet divergences.8) has 1 an associated symmetry factor of 2 and have rotated the momentum to Euclidean space by letting k0 → ik0E in the intermediate step to facilitate the evaluation of the integral. we have used the value of the angular integral in four dimensions which can be easily determined as π π 2π dΩ = 0 π dθ1 sin2 θ1 0 dθ2 sin θ2 0 1 −1 dθ3 = 0 dθ1 1 (1 − cos 2θ1 ) 2 d cos θ2 × (2π) (15. therefore. 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.

2 ˜ 32π M Λ2 (15.577215665.15.11) ˜ Λ→∞ n γ = lim n→∞ k=1 1 − ln n k = 0. in this case this is also the result in the Minkowski space. Λ are scalar parameters).12) . We add here for completeness that the cut-oﬀ can also be implemented through a weight factor in the following manner.10) (and m. However. (15.10) The result at the end should be rotated back to the Minkowski space. (15. since there is no momentum vector in the ﬁnal result in (15. We note that we can write (15.3 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 .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( ) .

for example. (15. which gives (4π) in three dimensions. (5. Gradshteyn and Ryzhik 3. This result coincides with (15.14) does not depend on the angular variables.47)) Ek = |k|2 + M 2 . Ek (15. the integration over the angles can be carried out.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 + .352.10) with the identiﬁcation ˜ Λ2 = Λ2 − M 2 γ.4 and 8. for example. we also note here that the diagram in (15. the integral takes the form iλ (4π) 32π 3 ∞ 0 ¯ Λ I = − dy y2 y2 + M 2 y2 + M 2 − M2 y2 + M 2 y2 + M 2 ¯ Λ 0 iλ = lim − 2 ¯ 8π Λ→∞ = ¯ Λ→∞ dy 0 √ ¯ ¯ iλ ¯ ¯ 2 Λ + Λ2 + M 2 2 − M 2 ln = lim − Λ Λ +M ¯ 16π 2 M Λ→∞ 2 ¯ 1 4Λ iλ ¯ + O( ¯ 2 ) .1).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.14) d3 k where we have identiﬁed (see. Furthermore. (15.13) Furthermore.8) can be evaluated without rotating the momentum into Euclidean space as follows iλ 2 i d4 k (2π)4 k2 − M 2 + iǫ d3 kdk0 I = − = = λ 2(2π)4 λ (−2πi) 2(2π)4 1 (k0 − (Ek − iǫ))(k0 + Ek − iǫ) 1 iλ =− 2Ek 32π 3 d3 k 1 . Since the integrand in (15.214. deﬁning y = |k|.

4). 2.10) has been transformed into an infrared divergence in the Schwinger parameter τ .15.17) is well behaved for large values of τ .271.10) if we identify 1 ¯ (15. for example. We note that the integrand in (15. Thus.2 and 2.8) can also be evaluated in an alternative manner which is sometimes useful and so let us discuss the alternate method as well. The integral in (15. but is divergent at the lower limit τ = 0. We note that the integral in Euclidean space has the form iλ × 2π 2 2(2π)4 iλ 32π 2 iλ 32π 2 iλ 32π 2 iλ 32π 2 0 ∞ 0 ∞ 0 ∞ 0 ∞ ∞ 0 I = − = − = − = − = − k2 1 2 dkE 2 E 2 2 kE + M dy y y + M2 2 dydτ y e−τ (y+M ) dτ 1 2 Γ(2) e−τ M 2 τ (15.17) dτ −τ M 2 . e τ2 Here Γ(2) denotes the gamma function and the parameter τ is known as the Schwinger parameter (or the proper time parameter).271. the ultraviolet divergence in (15. Gradshteyn and Ryzhik. This result coincides with (15. 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 .16) Λ2 = Λ2 − M 2 (1 − ln 2) . 2 This derivation makes it clear that the evaluation of the amplitude is much easier in the Euclidean space.3 Cut-off regularization 627 Here we have used the standard table of integrals in the intermediate step (see.

We note that since the cutoﬀ Λ has to be actually taken to inﬁnity in the ﬁnal result.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 . this graph is divergent. but quantum ﬁeld theories can be deﬁned in any number of space-time dimensions).471.10) (or (15.18) can also be obtained by simply cut1 ting oﬀ the τ integral at the lower limit Λ2 (without the regularizing exponential factor). Let us next look at another one loop graph. we note from (15. for example. 15.3.18)) that the self-energy diagram at one loop contains a term that diverges quadratically as well as a term which is logarithmically divergent. However.9) to evaluate the integral over τ . Equation (15. For simplicity we would consider all the incoming momenta to vanish.10) and we see that both ways of evaluating the integral lead to the same result. Without going into details.18) can be compared with (15. we note that the result in (15. Gradshteyn and Ryzhik 3. 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. namely. 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. (15. regularizing the integral brings out the divergence structure of the Feynman graph. For example. This diagram is also divergent (in four dimensions) and with the cut-oﬀ regularization leads to (we assume p1 = p2 = p3 = p4 = 0 for simplicity so that each graph contributes the same amount and also because we are interested in looking at only the divergence structure of the graph) .

(15.3 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. 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 .19) Λ→∞ lim There are several things to note from this derivation. First we have factored out an overall momentum conserving delta function along .

Renormalization is the process of redeﬁning these inﬁnities that we encounter in perturbation theory. there is a net 4 − 4 = 0 power of momentum in the numerator reﬂected in the logarithmic divergence of the diagram.) Let us note here that these divergences are there in the quantum ﬁeld theory besides the zero point energy or the ground state energy divergence which. This is. for the ﬁrst graph in (15. of course. The divergence structure of any diagram can be determined simply by counting the powers of momentum in the numerator and in the denominator. for example. for the diagrams in Fig. the cutoﬀ regularizes and deﬁnes . equivalent to saying that since we only measure diﬀerences in the energy levels. (This can also be seen from the fact that the integral in (15.10) and (15. In fact.8) up to a multiplicative factor. Finally. in general. each of these diagrams has an associated symmetry factor of 1 2 and we have used the value of the angular integral from (15. is plagued with divergences at every (loop) order in perturbation theory (unless we are in lower dimensions). can be removed by normal ordering the Hamiltonian of the theory. Very broadly it consists of two essential parts.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. we are free to redeﬁne the value of the ground state energy to zero. Thus we see that this graph is also divergent (in four dimensions). 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.19) is ∂ given by − ∂M 2 of that in (15. However. but the divergence in this case is logarithmic. the present divergences are nontrivial and are present even after normal ordering of the theory. 15.3.9) in the intermediate step. First we introduce a regularization procedure (scheme) which gives a meaning to the divergent Feynman integrals by isolating the divergent parts of the diagram. Consequently. any given quantum ﬁeld theory. 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. as we have seen earlier.630 15 Regularization of Feynman diagrams with the factor of (2π)4 . Similarly. In the examples we have studied in (15.19). 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. As a result. Thus.

(15. in general. 15. depends analytically on the external momenta. − M2 p = iGF (p) = p2 p p3 p2 p1 p4 p3 = iSF (p) = i(p + m) / i = 2 . p−m / p − m2 = −(2π)4 igδ4 (p1 + p2 + p3 ). Let us note that any Feyn- man diagram (amplitude). But in this process the amplitude (and the theory) becomes cutoﬀ dependent.15.20) L = iψ∂ ψ − mψψ + ∂µ φ∂ µ φ − 2 2 4! This theory describes the interaction between a fermion and a scalar ﬁeld and the Feynman rules for this theory are given by i . the lack of dependence on the external momenta is simply due to our special choice of the external momenta to be all vanishing. the theory with a Yukawa coupling described by the Lagrangian density 1 λ M2 2 / φ − gψψφ − φ4 .21) . particularly in the second example (15.3. The second part of renormalization consists of removing the cutoﬀ dependence (or regularization dependence) of the theory which we will take up in the next chapter. In the previous examples. (15. namely.1 Calculation in the Yukawa theory. But let us examine the dependence of diagrams on external momenta (and this will be important in studying renormalization) in another theory.3 Cut-off regularization 631 the integrals. p1 p2 = −(2π)4 iλδ4 (p1 + p2 + p3 + p4 ).19).

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

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) .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 ) . (15.26) Using the Feynman combination formula (15.121).3 Cut-off regularization 633 useful. ( n xi Ai )n i=1 (15. the integral in (15. let us next calculate the two simplest nontrivial graphs involving loops in this theory.25) Here we have rotated the momenta as well as the gamma matrices 0 to Euclidean space as γ 0 → iγE so that 0 0 k = γ 0 k0 − γ · k = −γE kE − γ · k = −kE . The fermion self-energy at one loop has the form k p k+p p = (−ig)2 i i d4 k 4 k 2 − M 2 (k + p) − m (2π) / / = g2 = g2 = − 1 (k + p) + m / / d4 k 4 k 2 − M 2 (k + p)2 − m2 (2π) ig2 (2π)4 id4 kE 1 −(kE + pE ) + m / / 4 −k 2 − M 2 (2π) − (kE + pE )2 − m2 E d4 kE 2 (kE + M 2 ) (kE + pE )2 + m2 (kE + pE ) − m / / .23). / / (15. Equipped with the Feynman combination formula. Generalization of (15.15.24) which we have already used in (13.

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) / .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. 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 . (15.28) In (15.27) where we have deﬁned Q2 = x(1 − x)p2 + xm2 + (1 − x)M 2 . E (15.

/B the self energy graph in (15. M. Furthermore. Similarly the scalar self-energy graph is given by k p k+p d4 k i i 4 k − m (k + p) − m (2π) / / / (15.30) takes the form . Λ.30) p = −Tr (−ig)2 = −g2 Tr (k + m)((k + p) + m) / / / d4 k .31) Tr A = 0. 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. Using the trace identities Tr ½ = 4.29) = f (p. We also note here that we have used the integral from (15.15.3 Cut-off regularization 635 × ln Λ2 − x(1 − x)p2 + xm2 + (1 − x)M 2 −1 −x(1 − x)p2 + xm2 + (1 − x)M 2 (15. / Tr A / = 4A · B. m. (15.19) in the intermediate step and have rotated back to Minkowski space in the ﬁnal step. g). where it is understood that the limit Λ → ∞ is to be taken. 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).

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

for the fermion self-energy in (15. Similarly. for the scalar self-energy in (15. a Feynman amplitude with n external lines is an analytic function of (n − 1) external momentum variables. Λ). Thus. These two simple calculations show that Feynman amplitudes are functions of external momenta. 2 (15.3 Cut-off regularization 637 2 2 ig2 = lim Λ→∞ 4π 2 1 0 dx Λ − 3Q ln Λ2 + Q Q 2 2 + 2Q 2 = f (p. m.32) where we have identiﬁed Q2 = x(1 − x)p2 + m2 while Q = −x(1 − E x)p2 + m2 represents the function rotated to Minkowski space.15.33) +ﬁnite terms as Λ → ∞ . for example. In fact. This is because the overall momentum conservation eliminates one of the momentum variables. 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 (15.29). g.32) we can write . In the examples worked out above it is clear that one can Taylor expand the amplitudes around zero momentum of the external lines.

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

80) and (9. The Pauli-Villars regularization provides such a regularization scheme.125)) 1 1 /ψ L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 .µν (p) = − iηµν .15. let us consider QED (quantum electrodynamics) in the Feynman gauge (see (9. p2 p q p = iSF (p) = i(p + m) / . we can now calculate various 1PI amplitudes in QED. For example.4 Pauli-Villars regularization 639 a covariant regularization scheme which also respects gauge invariance. 2 − m2 p r µ = −(2π)4 ieγ µ δ4 (p + q + r).35) ν = iGF . To discuss the Pauli-Villars regularization scheme.(15.36) 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 . 4 2 which leads to the Feynman rules µ p (15.

37) ie2 8π 2 dx((1 − x)pE + 2m) / where we have deﬁned Q2 (m.) The momentum integral is. E (15. µ) − 2 2 kE + Q2 (m. µ) = x(1 − x)p2 + xm2 + (1 − x)µ2 .26). µ))2 dx ((1 − x)pE + 2m) / 2 dkE 1 Q2 (m. of course. µ))2 ∞ 0 2 dkE −1 .111) and (2.27)). 2 kE + Q2 (m.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.9) for the angular integral.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) (kE + Q2 (m. (The Dirac gamma matrices have also been rotated to Euclidean space as discussed in (15. µ))2 2 / kE (2(1 − x)pE + 4m) 2 (kE + Q2 (m. µ) (15. In the . We have also used gamma matrix identities in (2. divergent (logarithmically by power counting) as is the case in the earlier calculation of the fermion self-energy with the Yukawa coupling (see (15.112) as well as (15.

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. A′ are ﬁctitious heavy µ ﬁelds introduced to regularize diagrams. we can write the additional contribution to the fermion self-energy as ie2 8π 2 ∞ 0 2 dkE −1 . these additional interacting ﬁelds give an additional contribution to the fermion self-energy (15. without doing any further calculation. µ and ψ ′ .4 Pauli-Villars regularization 641 Pauli-Villars regularization. (The covariant derivative in (15. let us add to the Lagrangian density of QED (15. Λ) (15.37) except for a sign and µ → Λ.15. µ) (15.41) I (reg) = I + I ′ = − dx ((1 − x)pE + 2m) ln / . The idea is to take Λ → ∞ at the end and in that limit the heavy ﬁelds do not propagate and. / ′ L′ = (15. Λ) .35) another gauge invariant term of the form 1 ′ ′µν Λ2 ′ ′µ ′ / Fµν F − Aµ A − iψ D (A + A′ )ψ ′ 4 2 +Λψ ψ ′ + eψA ′ ψ. 2 kE + Q2 (m.39) where we assume that Λ ≫ m. we deﬁne the theory with a minimal coupling of the photon to another fermion which we assume to be heavy as well as introduce a second massive photon which couples to the normal as well as the heavy fermions.37) coming from the diagram where a heavy photon is being exchanged. Namely. However. therefore. The contribution from the heavy photon to the self-energy would be exactly the same as what we have already calculated in (15.) 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) is deﬁned with the gauge ﬁeld combination Aµ + A′ . Q2 (m.

µ) = −x(1 − x)p2 + xm2 + (1 − x)µ2 . (15.43) . We can also calculate the one loop photon self-energy in QED in a similar manner. (15. Λ) Q (m. The ﬁnal result can now be rotated back to Minkowski space and has the form ie2 = 2 8π Q (m.36) lead to k p k+p i(k + m) i(k + p + m) / / / d4 k γ γ 4 µ k 2 − m2 ν (k + p)2 − m2 (2π) p = iΠµν (p) 2 = −(−ie)2 Tr =− =− kµ (k + p)ν + kν (k + p)µ − ηµν k · (k + p) + m2 ηµν 4e2 d4 k (2π)4 (k2 − m2 )((k + p)2 − m2 ) 4e2 (2π)4 d4 k dx × =− kµ (k + p)ν + kν (k + p)µ − ηµν (k · (k + p) − m2 ) ((k + xp)2 + x(1 − x)p2 − m2 )2 d4 k dx 4e2 (2π)4 × =− 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − (x(1 − x)p2 + m2 )) (k2 + Q (m))2 2 4e2 (2π)4 dx d4 k 2 × 2x(1 − x)(ηµν p2 − pµ pν ) + 2kµ kν − ηµν (k2 + Q (m)) (k2 + Q (m))2 2 . µ) 2 2 I (reg) dx ((1 − x)p − 2m) ln / .642 15 Regularization of Feynman diagrams which is ﬁnite for any given value of Λ.42) where Q (m. The Feynman rules (15.

45) where Q2 (m) = x(1 − x)p2 + m2 . p0E ). 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 . To evaluate the momentum integral. We can now integrate each of the E terms in (15.44) The overall negative sign in (15.43) to 2 Euclidean space ((k0 . + y + Q2 (m) (y + Q2 (m))2 = d4 kE E 1 (−δµν ) + Q2 (m) . 2 (kE + Q2 (m))2 (15.45) individually. 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 . η00 → E 2 ΠE . p2 ).15.4 Pauli-Villars regularization 643 where we have used the gamma matrix identities in (2. (k2 . 2 (15.43) reﬂects the fact that we are evaluating a fermion loop.116) and have deﬁned Q (m) = x(1 − x)p2 − m2 . Π → (i)ΠE ) and we have δ00 . we can rotate (15. p2 ) → −(kE .115) and (2.

This integral.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) . recalling that the photon has coupling to ﬁctitious heavy fermions.9) in the intermediate steps. Putting this back into (15.46) =− π2 δµν 2 dy 1− (y + Q2 (m))2 Here we have used (15. we obtain for the self-energy of the photon (in Euclidean space). However. as we see.47) 1 + δµν − 2 dy + Q2 (m) . y + Q2 (m) d4 kE 2 (kE + Q2 (m)) 2 − k2 δµν 2 E + Q (m) (Q2 (m))2 . is quadratically divergent.45). (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.

Λ1 ) − ic2 ΠE (pE . for example. Then the regularized photon amplitude with these two sets of heavy fermions will have the form iΠE (reg) (pE ) = iΠE (pE . c2 e respectively. Λ) µν µν = − ie2 4π 2 dx 2x(1 − x) δµν p2 − pµE pν E ln E 1 − δµν Q2 (Λ) − Q2 (m) 2 Q2 (Λ) Q2 (m) . (15.4 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. m) − iΠE (pE . As a result. Λ2 and charges c1 e. 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 . Λ2 ) µν µν µν µν = − × ie2 4π 2 ∞ 0 dx 2x(1 − x) δµν p2 − pµE pν E E dy c1 c2 1 − − y + Q2 (m) y + Q2 (Λ1 ) y + Q2 (Λ2 ) −(1 − c1 − c2 ) 1 − δµν (1 − c1 − c2 ) 2 ∞ 0 dy − Q2 (m) − c1 Q2 (Λ1 ) − c2 Q2 (Λ2 ) = − ie2 4π 2 dx 2x(1 − x) δµν p2 − pµE pν E E . this has the unpleasant feature that the photon self-energy has developed a mass term (term proportional to δµν .15. m) − ic1 ΠE (pE . section 14. see.1) which would violate gauge invariance. the regularized photon self-energy will have the form iΠE (pE .48) This shows that the diagram is ﬁnite for any given value of Λ. However.

since the diagram for the photon self-energy in (15. (15. · · · . the result (15.52) We note here that even though we are working in the Feynman gauge.49) would cancel leading to ie2 δµν p2 − pµE pν E E 2π 2 iΠE (reg) = − µν dx x(1 − x) × c1 ln Q2 (Λ1 ) + c2 ln Q2 (Λ2 ) − ln Q2 (m) . c1 Λ2 + c2 Λ2 = m2 . Pauli-Villars regularization involves introducing.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 ) . In general. 1 2 (15. the quadratic divergences as well as the logarithmic divergences in (15.(15. a set of heavy ﬁelds with masses Λi and √ charges ( ci e) with i = 1. in a gauge invariant manner. n depending on the nature of the divergence in the diagram such that the regularized amplitude . therefore. that if the parameters satisfy c1 + c2 = 1.50) then.49) E 1 2 It is clear.43) does not involve the photon propagator.52) holds for any covariant gauge with an arbitrary value of the gauge ﬁxing parameter ξ. 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) .(15. 2.

it is by construction a gauge invariant regularization scheme. For example. 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. m) − ci Ii (p. i ··· .15. the integral d4 k . However.) We should mention here that dimensional regularization has its own problems which we will discuss at the end of . In fact. since gauge invariance is independent of the number of space time dimensions.54) will be ﬁnite and gauge invariant. This procedure is known as dimensional regularization. however. i=1 (15.4 which cannot be regularized by the Pauli-Villars method. (k2 − m2 ) ((k + p)2 − M 2 ) (15. there are graphs of the form shown in Fig.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. (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.53) with the conditions on the parameters given by ci = 1. Thus we see that a method of regularizing Feynman integrals can very well be to analytically continue the integral into n dimensions where it is well deﬁned (well behaved). In non-Abelian theories. 15. Λi ). 15. (15. it is quite useful in studying non-Abelian gauge theories for which the Pauli-Villars regularization does not work. i i ci Λ2 = m2 .5 Dimensional regularization 647 n I (reg) = I(p.55) is logarithmically divergent in four dimensions. in less than four dimensions it is ﬁnite. Furthermore.

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) 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. 2 [M ] = 1.648 15 Regularization of Feynman diagrams this section. 2 2 4! L= (15. but it is worth pointing out here that manipulations with dimensional regularization are extremely simple which is why it is quite useful. (15. the coupling constant for the quartic interaction in (15. [φ] = [λ] = 4 − n.4: One of the one loop graphs contributing to the gluon self-energy. Figure 15.56) for the theory in n dimensions as M 2 2 λµ4−n 4 1 ∂µ φ∂ µ φ − φ − φ . For example.57) Therefore. 2 2 4! L= (15.58) are given by . In dimensional regularization we analytically continue the theory to n dimensions. The Feynman rules for the theory (15.56) we can carry out the (canonical) dimensional analysis in n dimensions to determine n−2 .

therefore.61) I = = = = dn kE dn kE 1 . to calculate amplitudes in n dimensions. p1 p2 (15.15. (15.5 Dimensional regularization 649 p = p2 i . since each vector index takes n values. shifted the variable of integration and have deﬁned Q2 = M 2 − p2 . − M2 p4 p3 = −(2π)4 iλµ4−n δn (p1 + p2 + p3 + p4 ). we can separate out the angular part of the integral . For example. 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. it follows that in n dimensions we have µ ηµ = n. E To evaluate this integral.59) We need to be careful about manipulating tensors in n dimensions. 2 (kE + Q2 )α where we have rotated to Euclidean space.60) Furthermore. let us note that the integrand is spherically symmetric (independent of angular coordinates) and.

Let us consider the basic n dimensional Gaussian integral whose value is given by dn kE e− 2 kE 2 n = (2π) 2 .62) The integral over the angles dΩ can be evaluated from the simple Gaussian integral as follows. n ≥ 2. dθn−1 .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. .65) Let us compare this with known results in lower dimensions that we are familiar with. in n dimensions. We see that . 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 . where. This determines the value of the angular integral to be dΩ = 2π 2 (2π) 2 = . n −1 n Γ n 22 Γ 2 2 n n (15.64) n = 2 2 −1 Γ n 2 k2 dΩ.63) On the other hand. . (15. we can evaluate the integral in (15.650 15 Regularization of Feynman diagrams n−1 dn kE = kE dkE dΩ.

(15. q) = dt t2p−1 1 + t2 . 2 p= q =α−p=α− (15. for the basic integral (15. (15.5 Dimensional regularization 651 n=2: n=3: n=4: dΩ = dΩ = dΩ = dθ = 2π. Thus. (15.15.61) we have 1 dn k n (k 2 + 2k · p − M 2 )α (2π) dΩ 0 n 2 I = = = = i(−1)α (2π)n ∞ n−1 dkE kE (k2 E 1 + Q2 )α i(−1)α 2π (2π)n Γ n 2 i(−1)α 2π (2π)n Γ n 2 n ∞ 0 ∞ 0 n−1 dkE kE (k2 kE Q ∞ 0 1 2 α E + Q ) n−1 n 2 d kE Q Qn 1 2 2α kE Q 1 + Q2 α 1 i(−1)α π 2 = n × 2 n Γ n 2 )α− 2 (2π) 2 (Q dt tn−1 (1 + t2 )−α .66) which agree with what we already know from explicitly doing these integrals. we see that with the identiﬁcation n . dθ1 sin2 θ1 dθ2 sin θ2 dθ3 = 2π 2 . dθ1 sin θ1 dθ2 = 4π.68) Therefore.69) the integral (15.67) can be written as . 2 n .67) Let us recall that the beta function is deﬁned as Γ(p)Γ(q) =2 Γ(p + q) ∞ 0 −p−q B(p.

For example.71) by diﬀerentiation.71) This basic integral generates all other integrals that we need for evaluating amplitudes in n dimensions and.652 15 Regularization of Feynman diagrams I = = Γ i(−1)α π 2 1 n n Γ n 2 )α− 2 (2π) 2 (Q iπ 2 (−1)α Γ α − n 2 n .70) Rotating back to Minkowski space. we see that this gives us our basic integral as dn k 1 n (k 2 + 2k · p − M 2 )α (2π) n I = = iπ 2 (−1)α Γ α − n 2 n .71) 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 . (2π)n Γ (α) (Q2 )α− 2 n n n 2 Γ α− Γ(α) n 2 (15. in fact. any other formula can be obtained from (15. using (15.72) Similarly we can obtain . (2π)n Γ(α) (p2 + M 2 )α− 2 (15. n Γ α − (2π)n Γ(α) (p2 + M 2 )α− 2 2 n (15.

74) becomes .58). With these basic integration formulae.5 Dimensional regularization 653 Iµν = 1 iπ 2 (−1)α = n n Γ(α) (2π) (p2 + M 2 )α− 2 × pµ pν Γ α − 1 n n − ηµν p2 + M 2 Γ α − 1 − 2 2 2 . we are analytically continuing away from four dimensions (to a lower dimension where the integral is well deﬁned). (15.e. take the limit n → 4 which translates to ǫ → 0.15.74) Let us next set n = 4 − ǫ. the amplitude (15. With this. At the end of our calculations we should.8)) k p iλµ4−n 2 p i dn k n k2 − M 2 (2π) = − = λµ4−n iπ 2 (−1) Γ 1 − n 2 n 2 (2π)n Γ(1) (M 2 )1− 2 iλµ4−n π 2 Γ 1 − n 2 n . the one loop scalar self-energy takes the form (see also (15.. 2 (2π)n (M 2 )1− 2 n n = − (15. let us now calculate various amplitudes in the φ4 theory (15.73) n kµ kν dn k n (k 2 + 2k · p − M 2 )α (2π) and so on. First of all. i. of course.

the scalar self-energy (15.76) as well as (15.12). we have π2 (2π)n n = ≃ 1 1 n = ǫ (4π) 2 (4π)2− 2 ǫ 1 1 + ln 4π .77) Using (15. In the limit ǫ → 0. 16π 2 2 (15.76) where γ denotes the Euler’s constant deﬁned in (15. 2 (2π)n (M 2 )−1+ 2 n (15.75) takes the form .75) Let us next work out some of the identities involving the gamma functions that will be useful to us.77). ≃ 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.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 . we have n 2 n Γ 2− 2 Γ 3− ǫ ǫ ≃ 1 − γ. ǫ (15.

78) We note here that for M 2 = 0. we will set all the external momenta to vanish as we had done earlier in (15. In this case. for simplicity.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) . this graph would be regularized to zero in dimensional regularization which is the type of argument used in (13.122).19).15.3 and. Similarly. we can calculate the one loop correction to the vertex function shown in Fig. 15. the amplitude in n dimensions takes the form = = 3 (−iλµǫ )2 2 3λ2 µ2ǫ 2 dn k (2π)n dn k 1 n 2 − M 2 )2 (2π) (k i 2 − M2 k 2 . − + ln ǫ 4πµ2 (15.

ǫ 4πµ2 (15. we let e → e µ2 .35)) 1 1 L = − Fµν F µν + iψD − mψψ − (∂µ Aµ )2 . 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− . 15.82) . However.79) where we have used (15. as we take the limit ǫ → 0 (to go to four dimensions). /ψ 4 2 (15. the amplitudes diverge.81) As a result.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 .1 Calculations in QED. 2 (15. We note that these amplitudes are well behaved (regularized) for any ﬁnite value of ǫ.5.77).80) In n dimensions.76) and (15. to make the coupling constant dimensionless. 2 2 n−1 3 ǫ ψ = = − . ǫ (15. 2 2 2 ǫ . Let us next calculate one loop ampli- tudes in QED described by the Lagrangian density in the Feynman gauge (see (15.

83) As a result. 2 − m2 p r µ q p = −(2π)4 ieµ 2 γ µ δ4 (p + q + r).µν (p) = − iηµν .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. ǫ (15. p2 p = iSF (p) = i(p + m) / . (15.84) ǫ With these Feynman rules. we can calculate the fermion selfenergy at one loop which gives (see also (15.15. the Feynman rules of the theory in n = 4 − ǫ dimensions are given by µ p ν = iGF .85) = −ieµ 2 2 = −e2 µǫ / / dn k γ µ ((k + p) + m) γµ .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. n k 2 ((k + p)2 − m2 ) (2π) .

/ / (15. the one loop fermion self energy in (15.87) (2 − n)(k + p) + nm / / dn k n 2 + x(1 − x)p2 − xm2 )2 (2π) ((k + xp) / / dn k (2 − n)(k + (1 − x)p) + nm 2 − Q2 ) 2 (2π)n (k dn k (2 − n)(1 − x)p + nm / 2 − Q2 )2 (2π)n (k (15.89) .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 (15.79) in n dimensions as well as (15. ǫ 4πµ2 (15.658 15 Regularization of Feynman diagrams We note here that if we use the algebra of the gamma matrices (1.88) dx ((2 − n)(1 − x)p + nm) / dx ((1 − x)p − 2m) 1 − / × 2 −γ ǫ 1− ǫ 2 i (−1)2 Γ 2 − n 2 n n (4π) 2 Γ(2) (Q2 )2− 2 1+ ǫ ln 4π 2 ǫ Q2 ln 2 µ2 ǫ 2 2 − γ + ln 4π ǫ ≃ ie2 8π 2 dx ((1 − x)p − 2m) 1 − / × 1− ǫ Q2 ln 2 µ2 ≃ = ie2 8π 2 ie2 8π 2 × dx ((1 − x)p − 2m) / dx ((1 − x)p − 2m) / Q2 2 − ln 2 − γ − 1 + ln 4π ǫ µ xm2 − x(1 − x)p2 2 − ln −γ−1 . γµ kγ µ = (2 − n)k.86) Using this.60) we obtain γµ γ µ = n.

90) / / / dn k Tr γµ (k + m)γν (k + p + m) .5 Dimensional regularization 659 where we have deﬁned Q2 = −x(1−x)p2 +xm2 and have used (15.91) the photon self-energy (15. Tr A = 0. (15.115) and (2. /B Tr A /C = 0. Similarly the photon self energy at one loop can also be calculated to have the form (the overall negative sign is because of the fermion loop) k p k+p ǫ p = iΠµν (p) = − −ieµ 2 = −e2 µǫ 2 i(k + m) i(k + p + m) / / / dn k Tr γµ 2 γ n 2 ν (k + p)2 − m2 (2π) k −m (15.90) becomes (we factor out (−e2 µǫ ) for simplicity and will restore this later) dn k n(kµ (k + p)ν − ηµν k · (k + p) + kν (k + p)µ + m2 ηµν ) (2π)n (k2 − m2 )((k + p)2 − m2 ) dx dx dn k kµ (k + p)ν + kν (k + p)µ − ηµν (k · (k + p) − m2 ) (2π)n ((k + xp)2 + x(1 − x)p2 − m2 )2 dn k (2π)n = =n =n .42).15.76) as well as (15. This can be compared with (15.116)). /B / Tr γµ γν γλ γρ = n (ηµν ηλρ − ηµλ ηνρ + ηµρ ηνλ ) . / Tr A / = nA · B.77). n (k 2 − m2 ) ((k + p)2 − m2 ) (2π) Using the n-dimensional identities (see (2. Tr ½ = n.

660 × =n × 15 Regularization of Feynman diagrams 2kµ kν − 2x(1 − x)pµ pν − ηµν (k2 − x(1 − x)p2 − m2 ) (k2 − Q2 )2 dx dn k (2π)n ηµν (x(1 − x)p2 + m2 ) − 2x(1 − x)pµ pν + 2kµ kν − ηµν k2 . Let us 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 . (15.77).93) × ηµν x(1 − x)p2 + m2 − 2x(1 − x)pµ pν where we have used (15.76) as well as (15.92) contains three diﬀerent kinds of terms which can be integrated using the standard formulae in (15. Let us next look at the terms with kµ kν in the numerator separately which leads to .92) where we have deﬁned Q2 = −x(1 − x)p2 + m2 .71)(15. (k2 − Q2 )2 (15. The integral in (15.73).

the term with k2 in the numerator leads to = −ne2 µǫ = ne2 µǫ ηµν ≃− ie2 ηµν 4π 2 −ηµν k2 dn k (2π)n (k2 − Q2 )2 dx dx (−1)2−1 Γ 1 − n i n 2 n n 2 )1− 2 2 (4π) 2 Γ(2) (Q ǫ 4 2− ǫ 2 1+ ǫ ln 4π 2 dx Q2 1 − 2 × − +γ −1 ǫ ≃− ie2 ηµν 4π 2 1− ǫ Q2 ln 2 µ2 dx Q2 (2 + ǫ (−1 + ln 4π)) Q2 2 × − + ln 2 + γ − 1 ǫ µ .15.94) Q2 1 2 +γ− × − + ln 2 ǫ 4πµ 2 Similarly.5 Dimensional regularization 661 = −ne2 µǫ = −2ne2 µǫ ≃ ie2 ηµν 4π 2 dx 1 i (−1)2−1 Γ 1 − n 2 dx ηµν n n 2 (4π) 2 Γ(2) (Q2 )1− 2 dx Q2 1 − ǫ 4 1+ 1− ǫ ln 4π 2 2kµ kν dn k n 2 − Q2 ) 2 (2π) (k 2 × − + (γ − 1) ǫ ≃ ≃ = ie2 ηµν 4π 2 ie2 ηµν 4π 2 ie2 ηµν 4π 2 ǫ Q2 ln 2 µ2 1− ǫ Q2 ln 2 µ2 1 2 dx Q2 − + γ − − ln 4π ǫ 2 Q2 1 2 dx Q2 − + ln 2 + γ − − ln 4π ǫ µ 2 dx (−x(1 − x)p2 + m2 ) . (15.

the photon self energy in (15. p q=0 µ k+p p =(−ieµ 2 )3 ǫ k+p k iηλρ / / / / dn k λ i(k + p + m) µ i(k + p + m)γ ρ γ γ − 2 n 2 − m2 2 − m2 (2π) (k + p) (k + p) k .93)-(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.96) =− dxx(1 − x) This shows that the photon self energy graph is completely transverse which is consistent with gauge invariance (see also (15. this result holds in any covariant gauge with the gauge ﬁxing term ξ arbitrary.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. × − + 2 ln ǫ 4πµ2 Adding the contributions from (15. For simplicity we will put the momentum of the external photon equal to zero. 2 ǫ 4πµ 2 (15. Since there is no longitudinal term in the photon self-energy. since the photon self-energy does not involve the photon propagator.95).95) Q2 4 + 2γ .52)). Let us next calculate the one loop amplitude corresponding to vertex correction (charge renormalization). the gauge ﬁxing Lagrangian density does not receive any quantum correction which can also be seen from the BRST identities for QED.92) takes the ﬁnal form ie2 (ηµν p2 − pµ pν ) 2π 2 2 Q2 1 − ln −γ− .

n (2π) k2 ((k + p)2 − m2 )2 Using the n-dimensional identities (see also (2.97) = −e3 µ 2 3ǫ / / / / dn k γ λ (k + p + m)γ µ (k + p + m)γλ .97) as dn k (2π)n k2 ((k + p)2 − m2 )2 (15.100) Let us next use (15.5 Dimensional regularization 663 (15.15. we can write the amplitude (15.98) γ λ (k + p + m)γ µ (k + p + m)γλ / / / / = γ λ (k + p)γ µ (k + p)γλ + mγ λ (γ µ (k + p) + (k + p)γ µ )γλ / / / / / / / / +m2 γ λ γ µ γλ = (2 − n)(2(k + p)(k + p)µ − (k + p)2 γ µ ) + 2nm(k + p)µ / / +(2 − n)m2 γ µ = −(2 − n)γ µ ((k + p)2 − m2 ) +2((2 − n)(k + p) + nm)(k + p)µ .99) = −e3 µ 2 × 3ǫ −(2 − n)γ µ ((k + p)2 − m2 ) + 2((2 − n)(k + p) + nm)(k + p)µ / / 3ǫ = −e3 µ 2 −(2 − n)γ µ dn k (2π)n k2 ((k + p)2 − m2 ) + 2 (2 − n)(k + p) + nm (k + p)µ / / . 2 ((k + p)2 − m2 )2 k (15.23) as well as the identity .86) we have (15. / / Therefore.113)) γ λ A µ A λ = (2 − n)(2Aµ A − γ µ A2 ). /γ /γ / as well as (15.

101) to combine the denominators in (15. (15. Using this in (15. ((1 − x)A + xB)3 1 ((1 − x)A + xB)2 (15.100) which leads to 1 + p)2 − m2 ) 1 (k + xp)2 − Q2 2x (k + xp)2 − Q2 k2 ((k = = dx dx 2. 1 k2 ((k + p)2 − m2 )2 3.664 15 Regularization of Feynman diagrams 1 AB 2 = − = ∂ 1 ∂ =− ∂B AB ∂B dx dx 2x . we obtain for the vertex correction p q=0 µ k+p p = −e3 µ 2 × 3ǫ k+p k dx dn k −(2 − n)γ µ (2π)n (k2 − Q2 )2 / / dn k 4x (2 − n)kkµ + (1 − x) (2 − n)(1 − x)p + nm pµ 3 (2π)n (k2 − Q2 ) 3ǫ = −e3 µ 2 dx − (2 − n)γ µ i (−1)2 Γ 2 − n 2 n n (4π) 2 Γ(2) (Q2 )2− 2 .102) where we have deﬁned Q2 = −x(1 − x)p2 + xm2 .100) and shifting the variable of integration.

103) and (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. we see that the fermion self-energy and the vertex function are related as (see also (9.89) with (15. p k+p k+p p q=0 (15.100).105) . (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 (15.88) and (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 .87). −(1 − x)p2 + m2 It is worth noting from this derivation that at every step if we compare (15. (15.104) 2 (1 − x)p − 2m pµ / .

there exist several other important regularization schemes. However. (15.6 References 1. Physical Review 74. 15. J. We simply note here that we select a particular regularization scheme depending on the theory (or the problem) that we are analyzing. Hence if the theory or the physical quantity of interest involves such objects.) The power of dimensional regularization is quite obvious from these calculations. For example. Analytic continuation of these to other dimensions is nontrivial. the point splitting method. .107) There are several ways to address this issue with the γ5 matrix.666 15 Regularization of Feynman diagrams which we recognize as the Ward-Takahashi identity.6 where we discuss chiral anomaly. It is gauge invariant and extremely simple to manipulate with. the higher derivative method. In addition to the regularization methods that we have discussed in this chapter. we know that i 4! γ5 = iγ 0 γ 1 γ 2 γ 3 = − µνλρ γ µ ν λ ρ γ γ γ . we will not go into the details of these other methods here. for example. (Note that the divergent parts as well as the ﬁnite parts satisfy the identity which demonstrates that dimensional regularization preserves gauge invariance. 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. (15. naive dimensional regularization leads to incorrect answers. However. Schwinger. 1439 (1948). The lattice regularization is also quite useful. We will discuss this further in section 16.106) is deﬁned only in four dimensions (both γ5 and the Levi-Civita tensor µνλρ are manifestly four dimensional quantities). It is not naively applicable if the calculation involves quantities that typically exist only in four dimensions. the ζ-function regularization etc. it has its own drawbacks.

664 (1951). Giambiaggi. ’t Hooft and M. G. J. 3. 769 (1949). Physical Review 82. 6. 8. Pauli and F.6 References 667 2. Nauka. Villars. Veltman. 7. Physical Review 76. Moscow (1984). Reviews of Modern Physics 21. 189 (1972). C. Nuclear Physics B44. 20 (1972). G. Schwinger. 4. 5.15. Diagrammar. G. N. . Veltman. ’t Hooft and M. N. Introduction to the theory of Quantized Fields. V. 434 (1949). Nuovo Cimento 12B. Bogoliubov and D. Bollini and J. W. R. Shirkov. CERN preprint (1973). J. Feynman.

This page intentionally left blank .

in n dimensions it should be n). The next step in understanding the process of renormalization is to devise a method for determining which Feynman diagrams in a theory will be divergent as well as the nature of the divergence (without actually evaluating the integrals) and this is achieved through the notion of the superﬁcial degree of divergence of a graph. we have seen that loop diagrams in various theories become divergent and need to be regularized. To deﬁne this let us start with the Lagrangian density for a given theory L = L0 + Li .Chapter 16 Renormalization theory 16. Furthermore.2) .1) is dimensionless. we have also seen that since Feynman amplitudes are analytic functions of external momenta. the Lagrangian density must have canonical dimension 4 in these units (since we are in four space time dimensions. it is easy to check that [M ] = [L]−1 = 1. (16.1) i where L0 denotes the sum of the free Lagrangian densities for all the ﬁeld variables in the theory and each Li representing some interaction is a monomial in the basic ﬁeld variables and derivatives. 669 (16. As a result. In our units. the action for the theory (16.1 Superﬁcial degree of divergence In the last chapter. 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. In units of = c = 1 which we have been using.

Thus. / 3 .4) (16.6) Namely. (16. bi and di to denote respectively the number of fermions. Let us further introduce the notations fi . bosons and derivatives at an interaction vertex following from the interaction Lagrangian density Li .8) In fact.3) [∂µ ] = [L]−1 = 1. for the Yukawa interaction LY = gψψφ. and L0 we have [φ] = 1.670 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 . (The mass term also has the correct dimension with this assignment. the canonical dimension of a scalar ﬁeld (in our units and in four dimensions) is 1. 2 (16.) Similarly the free fermion Lagrangian density (ψ) L0 = iψ∂ ψ − mψψ. (φ) (16. 2 2 L0 = that since (φ) (16. (16.5) = 4.7) leads to [ψ] = [ψ] = (16. for example.9) . 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.

With these notations we are now ready to introduce the notion of the superﬁcial degree of divergence of a Feynman graph.12) LI = hψγµ ψ∂ µ φ. b = 1. Similarly. 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. On the other hand.14) and so on. d = 0. b = 1.10) LI = leads to λ 4 φ . 4! (16. κψσ µν ψFµν . for an interaction of the form (16. (16.1 Superficial degree of divergence 671 we have f = 2.13) f = 2. the φ4 interaction (16. we have or. d = 1.16. we can determine easily the superﬁcial degree of divergence of the following Feynman graphs as . For example. d = 0. b = 4.11) f = 0. (16.

D = 4 − 2 − 1 = 1. D = 8 − 5 × 2 = −2. diagrams with D = 1 or 2 are known . D = 8 − 3 × 2 = 2. D = 4 − 2 × 2 = 0. we say that the diagram is superﬁcially logarithmically divergent. D = 4 − 2 − 2 × 1 = 0.15) If D = 0 for a diagram.672 16 Renormalization theory D = 4 − 2 = 2. Similarly. D = 4 − 2 × 1 = 2. (16.

then the Feynman diagram is said to be superﬁcially convergent.1 Superficial degree of divergence 673 respectively to have superﬁcial linear or quadratic divergences. (16.18) The superﬁcial degree of divergence is easily seen to be given by . The meaning of the adjective “superﬁcial” becomes clear once we look at the last graph in (16.16) ni = There exist topological relations between these numbers. 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 . Although this graph is superﬁcially convergent. number of internal fermion lines. we have F + 2IF = i ni fi . Let us consider a Feynman diagram with B = IB = F IF = = number of external bosons lines.15). it is actually divergent since it contains a subgraph which is divergent. number of external fermion lines. number of vertices of the ith type from Li . Rather than calculating the superﬁcial degree of divergence for each graph.16. Similarly. let us develop a general formula for the superﬁcial degree of divergence of any connected Feynman graph. (16.17) The factor 2IB reﬂects the fact that it takes two boson lines at two diﬀerent vertices to form a propagator (internal boson line). as we will see it is the notion of superﬁcial degree of divergence which is useful in the study of renormalization theory. If D < 0 for a graph. number of internal boson lines. (16. However.

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-

16. mentum variables about zero four-momentum out to order D(γ) which denotes the superﬁcial degree of divergence of γ. We should only remember that when nested boxes exist then the tγ operation should be carried from inside out.4. We would see in detail how this works in section 16. p) p p 2! ∂pµ ∂pν tγ Γ(2) (k. then D(γ) = 0 and tγ simply evaluates the graph γ at zero external momentum. Note that the BPHZ method subtracts out the divergent parts in the integrand itself so that the renormalized integral is ﬁnite. if γ is a Γ(4) . we can obtain a renormalized Feynman integrand R(G) for the graph G given by the expression R(G) = (1 − tG ) (−tγ ) I(G). Then the assertion is that the renormalized Feynman graph is convergent. F ∈Φγ ∈F (16.45) where I(G) represents the integrand of the graph G and Φ represents the complete set of disjoint and nested forests associated with the graph (these concepts will be discussed in section 16.2 A brief history of renormalization 689 Figure 16.9: Forest diagrams associated with the simple two loop vertex correction diagram in the φ4 theory. p) = Γ(2) (k. p=0 (16.4). we have tγ H = 0. For a convergent graph H. . For example. With the notion of the Taylor operator. then (D(γ) = 2 and k denotes the generic internal momentum) 1 µ ν ∂ 2 Γ(2) (k. If γ denotes the two point function Γ(2) .44) Here we have neglected the linear term in the Taylor expansion since it would vanish upon integration because of antisymmetry. 0) + .

2 2 3! L= (16.3 Schwinger-Dyson equation Let us consider the φ3 theory described by the Lagrangian density m2 2 g 1 ∂µ φ∂ µ φ − φ − φ3 . Working out in detail. under an arbitrary ﬁeld redeﬁnition φ → φ + δφ inside the path integral.48) We note that since the generating functional does not depend on the ﬁeld variable (we are integrating over all ﬁeld conﬁgurations). (16. However.690 16 Renormalization theory Weinberg’s theorem. (16.49) leads to .47) This describes the dynamics of the system at the tree level. This theorem is extremely important in the study of renormalization. We recall that the generating functional for the theory in the presence of an external source is deﬁned as (see (12. the generating functional would be stationary leading to R δZ = 0 = N Dφ (F [φ] − J) ei(S[φ]+ Jφ) . (16. The integral of a Feynman graph G is abso- lutely convergent if the superﬁcial degree of divergence DH is negative for every subgraph H of G including the case when H = G. when we include quantum corrections.120)) R Z[J] = eiW [J] = N Dφ ei(S[φ]+ Jφ) .49) where the Euler-Lagrange operator F is deﬁned in (16.46) The Euler-Lagrange equation for the theory is given by δS g = F [φ] = ∂µ ∂ µ + m2 φ + φ2 = 0. this equation modiﬁes and the modiﬁed equation can be obtained from the generating functional for the theory as follows.47). δφ 2 − (16. 16.

g ig δφc ∂µ ∂ µ + m2 φc + φ2 − i − J = 0. F φc − i δJ (16. δJ δJ δ − J = 0. 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. or. c 2 2 δJ ig g ∂µ ∂ µ + m2 φc + φ2 + 2 c 2 δ2 Γ δφ2 c −1 + δΓ = 0. or. If we now take the functional derivative of (16.52) can be represented graphically as in Fig. δJ δW δ −i − J = 0. φc − i φc − i − J = 0.80). 16.52) where the restriction “|” stands for setting φc = 0 and Σ denotes the complete self-energy the φ ﬁeld (the self-energy is deﬁned to be the complete two point function minus the tree level contribution). (16. Σ= ig 2 δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c . For the φ3 theory described by (16.46).78). (16.3 Schwinger-Dyson equation 691 eiW [J] F − i or. F δ − J eiW [J] = 0. this can be written explicitly as g 2 δ δJ δ δJ ∂µ ∂ µ + m2 φc + or.51) with respect to φc and set φc = 0.10 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 . Equation (16.50) Here φc denotes the classical ﬁeld which is deﬁned in (13.16.

To see in some detail how the BPHZ method works. = + + Figure 16.692 16 Renormalization theory = . Figure 16. orders. 16.10: Schwinger-Dyson equation for the two point function. let us consider the φ3 theory in six dimensions described by the Lagrangian density .53) which can be represented graphically as in Fig. (We are using a very compact notation where integrations over intermediate coordinates/momenta are suppressed.11: Schwinger-Dyson equation for the three point function. (16.4 BPHZ renormalization We have seen in detail how multiplicative renormalization works.52) and (16.11. we obtain δ3 Γ = −g − ig δφ3 c ig + 2 δ2 Γ δφ2 c δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ4 Γ δφ4 c δ2 Γ δφ2 c δ2 Γ δφ2 c −1 δ3 Γ δφ3 c δ2 Γ δφ2 c −1 −1 −1 .53) are integral equations (they are written in a compact notation here) for the two point and the three point functions. Such equations (and for the higher point functions) are known as the Schwinger-Dyson equations.51) with respect to φc and setting φc = 0. The equations in (16. 16. by taking the second functional derivative of (16.) Similarly.

(16. the counting of the canonical dimension gives [φ] = 2. (16.4 BPHZ renormalization 693 L= 1 M2 2 g ∂µ φ∂ µ φ − φ − φ3 .56) . (Note that unlike the φ4 theory. 2 2 3! (16. Furthermore. for this theory. 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. since we are interested in the ultraviolet behavior of the theory.54) This is a simple quantum ﬁeld theory that is quite helpful in understanding various features of renormalization. let us set M = 0 for simplicity. in dimensional regularization which we will be using.) Although the zero point amplitude is divergent. the 2-point and the 3-point amplitudes are superﬁcially divergent.16. we can remove this divergence by normal ordering the theory. here the Lagrangian density is not invariant under φ → −φ. In six dimensions. Furthermore. the superﬁcial degree of divergence for any graph is given by B 0 1 2 3 D = 6 − 2B 6 4 2 0 and we see that the one point.55) Thus.

as before.59) . the naive dimensional analysis gives (n − 2) . Thus. in particular. In this case.57) so that we can introduce. 2 2 2 [φ] = [g] = n − (16. this theory should be renormalizable. Therefore. In this case. the one loop self-energy can be evaluated as k p k+p = = ǫ 1 − igµ 2 2 ǫ (16.694 16 Renormalization theory when n → 6. Using dimensional regularization we analytically continue to n dimensions and deﬁne ǫ = 6 − n. (16. let us calculate some amplitudes beyond the simple one loop. 2 3(n − 2) (6 − n) ǫ = = . g → gµ 2 .58) p 2 (i)2 dn k (2π)n k2 (k + p)2 1 dn k n 2 + x(1 − x)p2 )2 (2π) ((k + xp) n g 2 µǫ 2 g 2 µǫ 2 dx = Γ 2− n (−1)2 iπ 2 2 dx n Γ(2) (2π)n (−x(1 − x)p2 )2− 2 dx (−x(1 − Γ −1 + ǫ 2 n x)p2 )2− 2 = ig2 µǫ 1 n 2 (4π) 2 = ǫ ig2 µǫ 1 Γ −1 + 2 n n 2 (4π) 2 (−p2 )2− 2 dx 1 (x(1 − x))2− 2 n . an arbitrary mass scale to make the coupling constant dimensionless. In order to see how renormalization works in some detail and. only the 2-point and the 3-point amplitudes are divergent which can be made ﬁnite with conventional counter terms. namely. how the overlapping divergences are handled.

60) = 1 6 Using these. to understand the structure of the self energy at one loop. However. 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 . 2− 6−ǫ ǫ ǫ n =2− = 2 − 3 + = −1 + . 2 2 2 2 ǫ ≃ 2 2 − + (γ − 1) .16. let us simplify this.4 BPHZ renormalization 695 This form is quite suitable to carry out two loop calculations which we will do shortly. we obtain the one loop self energy of the theory in (16.76)). ǫ Γ −1 + 1 1 ǫ 1 + ln 4π . 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 .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 . (16. As we have discussed earlier (see (15.

(16.696 ≃ ≃ 16 Renormalization theory ig2 p2 2 (−p2 ) 5 + ln 4π − ln − γ −1− 3 ǫ 2 12(4π) µ 3 ig2 p2 2 (−p2 ) 8 − ln − γ− 3 ǫ 2 12(4π) 4πµ 3 . 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. but since the BPHZ method subtracts out divergences in every graph. (16.62) Similarly. 12(4π)3 ǫ =− (16.61) This leads to the one loop counter term ig2 p2 2 . i xi Ai ) × ( (16.64) .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 .

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

70) Furthermore.68) as ig3 µ 2 =− (4π)3 ǫ (16. we have x3 = 1 − x1 − x2 and x3 = 1 x3 = 0 so that 1 1−x1 ⇒ x1 = x2 = 0.67) dx1 0 0 dx2 × 1− ≃ − ig3 µ 2 (4π)3 ǫ ǫ (−x1 (1 − x1 )Q2 ) ln 2 4πµ2 (−x1 (1 − x1 )Q2 ) 2 − ln −γ . ǫ 4πµ2 (16.698 = 2g3 µ 2 3ǫ 16 Renormalization theory 1 1−x1 dx1 0 0 1 3ǫ dx2 1−x1 Γ 3− n i(−1)3 2 n 3− n 2 (−x (1 − x )Q2 ) 2 Γ(3)(4π) 1 1 (−x1 (1 − x1 )Q2 )3− 2 2 −γ ǫ Γ 3− n 2 n ig3 µ 2 = − n (4π) 2 ig3 µ 2 ≃ − (4π)3 ǫ dx1 0 1 0 dx2 1−x1 (16. we can rewrite (16.69) dx1 dx2 dx3 δ(1 − x1 − x2 − x3 ) = dx1 0 0 dx2 . (16.71) 1 1 dx1 0 0 du (1−x1 ) 2 (−x1 (1 − x1 )Q2 ) − ln −γ .72) . ⇒ x2 = 1 − x1 . ǫ 4πµ2 (16.68) 1 1−x1 dx1 0 0 dx2 In this derivation we have used the fact that because of the δ function involving the Feynman parameters. 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.

12: Non-overlapping two loop self-energy diagrams in the φ3 theory.68) leads to the one loop counter term (the factor of arises from doing the integrals over x1 .4 BPHZ renormalization 699 where in the new variables x1 (1 − x1 )Q2 = x1 (1 − x1 )(p + up1 )2 + u(1 − u)p2 .75) separately.74) Let us next calculate the self-energy at two loops. Let us ﬁrst look at the graphs with non-overlapping divergence of the form shown in Fig. 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 (16.12 (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. = 2(4π)3 ǫ ǫ (16.59).62). Let us look at the two integrals in (16. x2 ) ig3 µ 2 2 .73) 1 2 Equation (16. 16. the ﬁrst integral in (16. (16. = + −igµ 2 −igµ 2 ǫ ǫ 2 i i dn k i iΠ(1) (k2 ) 2 n k2 (2π) k (k + p)2 i dn k i (−ig2 k2 ) 2 i .16. Using the form of the one loop self-energy in (16.75) has the form .

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 .

16.76) Here we have used (16.76) and (16.85).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. (The y integration is related to the appropriate beta 2 function. see (16.4 BPHZ renormalization 701 2 1 2 (−p2 ) 5 − (2γ − 3) − ln + 2 2 ǫ ǫ 4πµ 2ǫ ǫ ≃ − ig4 p2 5ǫ 1+ 6 72(4π) 6 + ﬁnite = − 1 2 (−p2 ) 5 ig4 p2 2 − (2γ − 3) − ln + 6 ǫ2 2 72(4π) ǫ 4πµ 2ǫ ǫ + 5 + ﬁnite 3ǫ = (−p2 ) 43 2 2 ig4 p2 − ln − γ + ﬁnite . let us look at the second integral in (16.60) in the intermediate steps.64) as well as (15. Next. they will require non-local counter terms to cancel them. + − 2− 72(4π)6 ǫ ǫ 4πµ2 12 (16.

upon using the form of the one loop vertex function in (16.67) (with x2 expressed in terms of u as in (16. takes the form ℓ k p ℓ+p k+p p . adding (16. 72(4π)6 ǫ2 ǫ 4πµ2 3 As a result.78) I1 + I2 = ig4 p2 2 11 + ﬁnite .71)).702 ≃ − 16 Renormalization theory g 4 p2 2 i 2 − + (γ − 1) − ln 4π 3 ǫ (4π)3 12(4π) ǫ × ǫ (−p2 ) 5ǫ 1 1 − ln + 6 2 µ2 6 ≃ − = ig4 p2 2 2 (−p2 ) 5 − + ln − + (γ − 1) + O(ǫ) 72(4π)6 ǫ ǫ 4πµ2 3 (16. Let us next look at the overlapping divergent graph in the self energy at two loops which.77) ig4 p2 4 (−p2 ) 8 2 − ln − − + γ + ﬁnite .77). the sum of the two graphs in (16.75) takes the form (−p2 ) 43 2 2 ig4 p2 − ln + −γ − 2− 72(4π)6 ǫ ǫ 4πµ2 12 + = (−p2 ) 8 2 4 ln − − + γ + ﬁnite ǫ2 ǫ 4πµ2 3 (16.76) and (16. − 6 ǫ2 72(4π) 6ǫ We see that the potentially dangerous non-local divergences have cancelled in the non-overlapping divergent graphs and the remaining divergences can be cancelled by local counter terms.

Using (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 .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 (16.81) .73) (with appropriate momenta).79) as g4 µ2ǫ (−1) 2 −3 n Γ 3− n 2 2 2(4π) Γ 5− × Γ 3− = n 2 n 2 n n n n dn ℓ −3 2 dx1 du x1 (1 − x1 ) 2 −2 (2π)n = dy1 dy2 dy3 y1 2− n 2 δ(1 − y1 − y2 − y3 ) D2 n 5− n 2 g4 µ2ǫ (−1) 2 −3 n Γ 5− n 2 2(4π) 2 dn ℓ (2π)n dy1 dy2 dy3 y1 2 dx1 du x1 −3 (1 − x1 ) 2 −2 . (16.16. x1 (16.64) to combine denominators.79) n −3 n where Q2 is deﬁned in (16. (2π)n ℓ2 (ℓ + p)2 (Q2 )3− 2 (16.

84) .82) Substituting this into (16. let us evaluate the integral in (16.704 16 Renormalization theory The divergence of the two loop integral in (16. so that the (shifted) denominator in (16. Finally.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).81) has the form D2 = ℓ2 + (1 − y1 )(y1 (u − v)2 + v(1 − v)) + = ℓ2 + d(x1 . First we note that we can shift the variable of integration ℓ → ℓ − (uy1 + y2 )p. in fact. The one loop pole structure is manifest in the overall multiplicative gamma function. This is a special feature of overlapping divergent graphs and to appreciate this. where d is deﬁned in (16. y1 .80). we can do the integration over y3 using the delta function in (16.71)) as y2 = (1 − y1 )v. Next. However.80) in some detail.83) (16. this is not right and.80) and carrying out the ℓ integral we obtain n g4 µ2ǫ (−1) 2 −3 Γ 5− n 2 2(4π) 2 dy1 dv (1 − y1 )y1 ig4 p2 (−p2 ) 2(4π)6 4πµ2 2− n 2 2− n 2 n n 2 dx1 du x1 n = −3 (1 − x1 ) 2 −2 n × = − i(−1)5− 2 n (4π) 2 Γ 5 − n 2 Γ(5 − n) (−dp2 )5−n −ǫ Γ(−1 + ǫ) n × dx1 dudy1 dv x1 (1 − x1 ) 2 −2 y1 2− n 2 (1 − y1 )(x1 d)1−ǫ (16. we can redeﬁne the variable y2 (as we have already done in (16.83). y1 u(1 − u) 2 p x1 (16. v)p2 . the divergence of the subdiagram (one loop vertex) has been transformed to the Feynman parametric integrals. u. To see the divergence .

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

(16. On the other hand.74)).87) ig4 p2 2 (−p2 ) 2 + − γ + 3 + ﬁnite . − ln 12(4π)6 ǫ2 ǫ 4πµ2 We see again that the dangerous nonlocal divergent terms of the form (−p2 ) 1 ǫ ln 4πµ2 are present in this diagram. with the one loop counter term for the three point function (see (16.88) .706 16 Renormalization theory = − = 1 (−p2 ) 1 2 ig4 p2 − + γ − 1 1 − ǫ ln + 4 + O(ǫ) 6 2(4π) ǫ 4πµ2 6 ǫ (16. we also have a contribution from the graph k p k+p = (−igµ 2 ) 2 g 4 µǫ 2 4(4π)3 ǫ g 4 µǫ 2 4(4π)3 ǫ n ǫ p i dn k i ig3 µ 2 2 n k 2 2(4π)3 ǫ (k + p)2 (2π) 1 dn k dx n (2π) ((k + xp)2 + x(1 − x)p2 )2 dx Γ 2− n (−1)2 i 2 n n (4π) 2 Γ(2) (−x(1 − x)p2 )2− 2 dx x(1 − x) ǫ 2 Γ n 2 n −2 2 ǫ = − = − ig4 µǫ (−p2 ) 2 −2 2 ǫ = − Γ −1 + n 3 (4π) 2 ǫ 2 4(4π) ig4 p2 4(4π)6 2 ǫ (−p2 ) 4πµ2 1− − ǫ −2 = ≃ = Γ −1 + Γ(n − 2) 1 6 1+ 5ǫ 6 −1 2 ig4 p2 2 4(4π)6 ǫ ig4 p2 24(4π)6 ǫ (−p2 ) ln 2 4πµ2 ln 2 − + (γ − 1) ǫ 2 4 + 2 ǫ ǫ 8 (−p2 ) +γ− 2 4πµ 3 + ﬁnite.

= 2 2 ig4 p2 + 6 ǫ2 12(4π) ǫ − 4 2 + 2 ǫ ǫ − ln ln (−p2 ) −γ+3 4πµ2 + ﬁnite (16.89) p so that the sum of the three graphs in Fig. ǫ2 3ǫ We see that all the potentially dangerous nonlocal divergent terms have disappeared from the self-energy and the remaining divergences can be removed by adding a local two loop counter term. 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.13 gives + + Figure 16. 16.4 BPHZ renormalization 707 Here we have used (16. (16.13: Overlapping two loop self-energy diagrams in the φ3 theory.16.90) 8 (−p2 ) +γ− 2 4πµ 3 = ig4 p2 12(4π)6 − 2 2 + + ﬁnite . BPHZ is a renormalization procedure where divergences are subtracted in the graph itself so that it works with any regularization.85) in the intermediate step. . Similarly.

16. γ γ Figure 16.91) namely. namely.14: Examples of proper renormalization parts in the two loop self-energy graph in the φ3 theory. (16. or γ2 ⊂ γ1 . Fig. for example. 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. 16.708 16 Renormalization theory Consider an arbitrary Feynman graph G.15 shows examples of diﬀerent ways of drawing boxes around the renormalization parts. two renormalization parts γ1 and γ2 are called overlapping if they share lines and vertices. if it is superﬁcially divergent. (16. Two renormalization parts γ1 and γ2 are called disjoint if γ1 ∩ γ2 = 0. for the three loop self energy graph in the φ3 theory. (16. if γ1 ⊂ γ2 . It may contain one particle irreducible subgraphs that are superﬁcially divergent. A couple of examples of proper subgraphs which correspond to renormalization parts are shown in Fig. if none of the following holds. 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. Thus.14.93) Finally. . namely.92) while γ1 and γ2 are called nested if one is contained inside the other.

γ2 ⊂ γ1 . let us add to the set D the graph with no boxes. A forest is called normal if there is no box around the complete graph G.16. .4 BPHZ renormalization 709 Figure 16. 16. (16.17 shows the nested forests in the two loop self-energy graph in the φ3 theory.16. Let D denote the set of laying down of boxes which contain only disjoint renormalization parts. for the two loop self-energy graph in the φ3 theory.94) A set of such laying down of boxes around renormalization parts is called a forest. γ1 ∩ γ2 = 0. Similarly. Thus. a forest is called full if there is a box around G. let N denote the set of laying down of boxes (forests) containing only nested renormalization parts. 16. Furthermore. γ1 ⊂ γ2 .15: Examples of forests in the three loop self-energy graph in the φ3 theory. Fig. A forest is called empty if there is no box around any subdiagram. On the other hand. the complete set of laying down of boxes (forests) containing disjoint renormalization parts is shown in Fig.

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

(16. p=0 k γ: p k+p p2 k + p + p1 p1 = dn k I (k.97) (16.95) In other words.96) This is the conventional tγ operator in the BPHZ prescription and it is clear that using this operation. 0). p) = I(k.16. p) 2! ∂pµ ∂pν . 0) + p = dn k I(k. tγ I (k. Thus. p1 ) = I(k. Note that. (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. acting on a graph. if D(γ) < 0. p1 ) .4 BPHZ renormalization 711 it in the external momenta (external to the graph) up to terms of order D(γ). p. tγ I(k. we are simply able to throw . 0. (2π)n 1 µ ν ∂2 p p I(k. (16. tγ IG = 0. p). 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. (2π)n D(γ) = 6 − 6 = 0. p. by deﬁnition.

this makes it clear that for a superﬁcially convergent graph (with no pole term) tG FG = 0. in the case of dimensional regularization.99) so that the ﬁnite part of the graph can be identiﬁed with ∞ m=0 1−t G FG = am ǫm = F G = ﬁnite. (16.103) . (16. if the integral has the form ∞ m=−n Iγ = am ǫm . ∞ m=0 (16. it never uses any particular regularization and is. a graph (integral) at n-loops. compatible with any regularization.100) Once again. am ǫm = ﬁnite parts. (16.102) then as we have seen the Taylor operation is deﬁned such that −1 m=−n tγ Iγ = (1 − tγ ) Iγ = am ǫm = divergent parts. in fact.98) In such a case. In particular.712 16 Renormalization theory away potentially divergent parts inside the integral. (16.101) Given a renormalization part γ in a Feynman diagram. Therefore. the tγ operation is deﬁned to be (it separates out the pole terms) −1 m=−n t FG = G am ǫm . (16. has the general form ∞ m=−n FG = am ǫm .

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

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

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

716 16 Renormalization theory RG IG = IG + IG/γ −tγ Rγ Iγ = IG + IG/γ (−tγ Iγ ) . RG IG = (16.108)) we can write γ1 γ2 Figure 16.112) This is exactly the original formula (16. 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 . by deﬁnition. this is how we are supposed to renormalize the theory.21: The complete set of disjoint proper renormalization parts of the two loop self-energy graph in the φ3 theory. We note that.113) . (16. RG IG = IG + IG/γ1 −tγ1 Rγ1 Iγ1 + IG/γ2 −tγ2 Rγ2 Iγ2 = IG + IG/γ1 (−tγ1 Iγ1 ) + IG/γ2 (−tγ2 Iγ2 ) = IG + (−tγ1 IG ) + (−tγ2 IG ) = (1 − tγ1 − tγ2 ) IG . in fact.111) 1 − tG RG IG = 1 − tG (1 − tγ ) IG . (16. Let us next look at the two loop self-energy graphs shown in Fig. 16. we can write (16.110) also as RG IG = IG + (−tγ IG ) = (1 − tγ ) IG . (16. Rγ Iγ = Iγ because there is no superﬁcially divergent proper subdiagram of γ. there is no subdiagram in this case.21 containing only proper disjoint renormalization parts. From the relation IG = IG/γ Iγ .105) and. in fact.

sets the factor of µ 2 = 1 in the graph at one loop) = so that ig3 2 = (−tγ2 Iγ2 ) . 12(4π)6 ǫ2 . (16. The diﬀerence between the two is given by 1 − tG tγ1 tγ2 IG . (16.118) 2 i (−p2 ) g4 4(4π)3 ǫ (4π)3 6 ig4 p2 2 .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.4 BPHZ renormalization 717 so that RG IG = 1 − tG (1 − tγ1 − tγ2 ) IG .74).116) (16. Let us recall some of the calculations we had done earlier. tγ which is supposed to isolate ǫ pole terms in ǫ.114) This does not exactly coincide with the original formula (16.16.105) RG IG = 1 − tG (1 − tγ1 ) (1 − tγ2 ) IG . At one loop we had found that (see (16. 2(4π)3 ǫ (16. It is such terms that complicate the proof of renormalizability.

This makes the connection between (16. it is best to study an example. To make contact with the formula of BPH.121) involving subtractions of only non-overlapping proper renormalization parts. the ǫ1 term precisely has the same coeﬃcient and it follows. The product of the factors in (16. If we recall our calculation for the two-loop overlapping divergence (see(16. then.87)).109). 2 therefore. (16. γi ’s are only disjoint or nested proper renormalization parts. that 1 − tG tγ1 tγ2 IG = 0.122) i where Φi ’s deﬁne forests containing only disjoint or nested proper renormalization parts. we can now write s RG IG = 1 − tG RG IG = 1 − tG i=1 (1 − tγi ) IG . Let us recall that the three loop diagram has eight . (16. we note that using this. namely. then we can always ﬁnd a renormalization part γ12 which contains both γ1 and γ2 and for which (1 − tγ12 ) tγ1 tγ2 IG = 0. there is a theorem that says that if γ1 and γ2 denote two proper renormalization parts of G that are overlapping.120) This result is very important. (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.119) Physically.121) can also be written as (1 − tγi ) ≡ (−tγk ) .105) and Zimmermann’s solution (16. Φi γk ∈Φi (16. this is the statement that we do not need subtractions associated with overlapping divergences. In general. physically because it says that we do not need to subtract overlapping divergences. but more importantly.

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 . we can write Rγ1 Iγ1 Rγ2 Iγ2 = Iγ1 . 16. (16.16.125) Φi γk ∈Φi . (16.124) Using this.19. Thus.γ2 } (−tγ1 Rγ1 Iγ1 )(−tγ2 Rγ2 Iγ2 ) + IG/γ4 (−tγ4 ) Rγ4 Iγ4 = 1 − tG 1 − tG 1 − tγ1 Rγ1 − tγ2 Rγ2 − tγ3 Rγ3 − tγ4 Rγ4 −tγk Rγk IG . we can rewrite (16. = Iγ2 .4 BPHZ renormalization 719 forests consisting of disjoint and nested proper renormalization parts which are shown in Fig.123) Let us next recall that since γ1 and γ2 do not contain any proper renormalization parts.123) as RG IG = (1 − tG ) IG + IG/γ1 (−tγ1 Rγ1 Iγ1 ) + IG/γ2 (−tγ2 Rγ2 Iγ2 ) + IG/γ3 (−tγ3 ) Iγ3 + Iγ3 /γ1 −tγ1 Rγ1 Iγ1 + IG/{γ1 . + (tγ1 Rγ1 )(tγ2 Rγ2 ) IG = (16.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 .γ2 } (−tγ1 Iγ1 ) (−tγ2 Iγ2 ) + IG/γ3 −tγ3 Iγ3 /γ1 (−tγ1 Iγ1 ) + IG/γ4 −tγ4 Iγ4 /γ2 (−tγ2 Iγ2 ) .

the BPH formula (16. it is equivalent to making any Feynman graph ﬁnite by local subtractions (counter terms). We would have divergence structures with more and more numbers of external lines. This analysis. δi < 0. So far. if the index of divergence were negative. (16.127) then. This is. However. On the other hand if δi > 0. 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.108). it is clear that the superﬁcial degree of divergence of a graph would increase with increasing number of interaction vertices.20) and (16. (16.128) not only would we have a ﬁnite number of types of graphs that can be divergent. 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 . In this case. in this case. we have talked about theories where the index of divergence of interactions in four dimensions (recall that D = 4 − B − 3 i ni δi . Such theories are called nonrenormalizable theories. shows how the BPH formula leads to Zimmermann’s solution and how. see (16. namely. In contrast.21)) 2F + 3 δi = di + bi + fi − 4. through Weinberg’s theorem.720 16 Renormalization theory where Φi is the complete set of proper renormalization parts consisting of only disjoint graphs. 2 (16. 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. Such theories are known as renormalizable theories. Of course.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. we can also remove the divergences by adding local counter terms. of course. therefore.

Like QED.a ν µ ν 2 − g2 abc apq b c µ. we do not need counter terms for the gauge ﬁxing terms).q + F a F a + (∂ µ F a )Aµ. here also we can check that only the transverse part of the gauge self-energy diverges (we have seen this explicitly in the calculation of the photon self-energy) so that the longitudinal part is not renormalized (namely.16. However. the three point and the four point amplitudes involving gauge ﬁelds as well as the two point function for the ghosts and the three point ghost interaction vertex function.a − ∂ ν Aµ.5 Renormalization of gauge theories 721 of divergence) will be divergent.a µ ν 4 2 L=− − .a + ∂ µ ca Dµ ca 4 2 1 = ∂µ Aa (∂ µ Aν. Let us note here (as we have seen explicitly in the case of QED) that the true degree of divergence in a gauge amplitude is generally lower than its superﬁcial degree of divergence because of gauge invariance.129) + ∂µ ca ∂ µ ca − gf abc (∂ µ ca )Ab cc .p Aν. Let us discuss this brieﬂy with the example of the pure non-Abelian Yang-Mills theory. we can add counter terms to write (1 + A) ∂µ Aa (∂ µ Aν. As a result.a )+(1+B)gf abc Ab Ac ∂ µ Aν.q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ. Such theories are conventionally known as super-renormalizable theories. some subtleties arise in the case of gauge theories.a ) + gf abc Ab Ac ∂ µ Aν. The total Lagrangian density with the gauge ﬁxing and the ghost Lagrangian densities has the form (see (13.5 Renormalization of gauge theories The analysis in this chapter demonstrates renormalizability of standard ﬁeld theories involving scalar and fermion ﬁelds.12)) ξ 1 a L = − Fµν F µν a + F a F a + (∂µ F a )Aµ.p ν. µ Naive power counting shows that the only graphs that are superﬁcially divergent are the two point.a ν µ ν 2 g2 ξ (1 + C)f abc f apq Ab Ac Aµ. 16.a − ∂ ν Aµ.a 4 2 (16.

g1 (u)2 (u) −2 = Z4 Z3 g2 . a general feature of gauge theories following from the BRST invariance of the theory (see (13. ˜2 ca(u) = Z3 ca .a − ∂ ν Aµ.p ν. µ 1 ˜2 ca(u) = Z3 ca .130) + Z3 ∂µ ca ∂ µ ca − Z1 gf abc ∂ µ ca Ab cc . all the divergences can be taken care of. ˜ Z3 = 1 + D. Z1 = 1 + B. in fact.a 4 2 ˜ ˜ (16. the longitudinal part of the photon self-energy is not renormalized by quantum corrections. if we redeﬁne the ﬁelds and parameters as a(u) 2 Aµ = Z3 Aa . −1 1 F a(u) = Z3 2 F a . −3 1 1 g2 ˜ − ˜ −1 = Z1 Z3 2 Z3 g.q ξ a a f f Aµ Aν A A + F F + (∂ µ F a )Aµ. (16.a ν µ ν 2 − Z4 g2 abc apq b c µ. g(u) = Z1 Z3 2 g. Namely.a ) + Z1 gf abc Ab Ac ∂ µ Aν. of course. . This is. With these counter terms. µ where we have identiﬁed Z3 = 1 + A. as we have seen explicitly in the calculation in QED. But the main question is whether renormalization would preserve the gauge invariance – actually the BRST invariance – present in the original theory.91)). ˜ Z1 = 1 + E.131) It is worth emphasizing here again that we have not added any counter term for the gauge ﬁxing terms because. Z4 = 1 + C.722 16 Renormalization theory +(1 + D)∂µ ca ∂ µ ca − (1 + E)gf abc (∂ µ ca )Ab cc µ = − Z3 ∂µ Aa (∂ µ Aν.

(u) 2 (u) 2 g2 (16.134) and remove all the divergences from the theory. of course. we will have 2 g(u) g1 2 −3 = Z1 Z3 g2 .16. (16. the same relation as in (16.132) ξ (u) = Z3 ξ. g(u) = g1 (u) (u) = g2 . for example. if we choose a regularization scheme which respects the BRST symmetry. 2 −3 ˜ 2 −1 ˜ −2 = Z1 Z3 Z3 g2 = Z1 Z3 g2 . therefore. −2 2 −3 = Z4 Z3 g2 = Z1 Z3 g2 . in turn require that the corresponding counter terms have to be related. the counterterms will satisfy (16. (16.133) and with this identiﬁcation we can write (16.134) Therefore.130) as . But the real question is how can the renormalization process guarantee this.91)) lead to relations between various amplitudes and we can show from these that they lead to relations among renormalization constants of the form (analogous to the relation Z1 = Z2 in QED following from the Ward-Takahashi identity) ˜ Z4 Z1 Z1 = = . The Slavnov-Taylor identities following from the BRST invariance (see.133) This would. naively we would expect that the BRST invariance (or even the gauge invariance in the gauge unﬁxed theory) would require the diﬀerent (bare) unrenormalized couplings to be the same.136) This is. In this case. (13. namely.135) which would lead to g(u) 2 = g1 (u) 2 = g2 (u) 2 . ˜3 Z3 Z1 Z (16.5 Renormalization of gauge theories 723 (16.

a(u) L = − Fµν F µν a(u) + 4 2 +∂ µ ca(u) Dµ ca(u) . It is essential in identifying the physical states of the theory. It is for these reasons that in the case of gauge theories. 2. a + iψ k γ µ Dµ ψk . in studying nonAbelian gauge theories it is natural to discuss renormalizability using dimensional regularization. n.724 16 Renormalization theory ξ (u) a(u) a(u) 1 a(u) F F + (∂µ F a(u) )Aµ. k . 2. (16.139) . Otherwise. dimensional regularization is one of the regularization schemes (and it is quite simple) which respects gauge invariance and BRST symmetry and. · · · . n2 − 1 and k = 1. 4 (16.46)) 1 a L = − Fµν F µν.138) where a = 1.137) where the covariant derivative is deﬁned with g(u) . The BRST invariance of this Lagrangian density is now manifest (with unrenormalized ﬁelds and parameters).6 Anomalous Ward identity Let us next consider a non-Abelian gauge theory (belonging to the group SU (n)) interacting with massless fermions (quarks) described by the Lagrangian density (see (12. 16. µ δψ k = iǫa (x) ψT a δψk = −iǫa (x) (T a ψ)k . consequently. 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). We note that this theory. (16. in establishing unitarity as well as in showing the gauge independence of the physical matrix elements. the regularization scheme may introduce unwanted spurious eﬀects into the theory. · · · . As we have seen. in addition to having the inﬁnitesimal local gauge invariance δAa = Dµ ǫa (x). the choice of a regularization which preserves this symmetry is quite crucial.

(16.140) Here λ is the space time independent real inﬁnitesimal parameter of transformation and (16.e. Dµ J µ. the invariance of the theory leads to a conserved current. if the fermions are massive.a = 0.140). if there is a term in the Lagrangian density of the form Lm = −mψ k ψk . there are two conserved currents.a = ψ j γ µ (T a )jk ψk .141) which. we can still deﬁne the axial vector current as in (16. however.. is covariantly conserved. In this theory. µ J5 = ψ k γ5 γ µ ψk .141) We note here that the vector current in (16.142) then this would not be invariant under the chiral symmetry transformations (16. Nonetheless. 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.140) describes the chiral symmetry transformations under which the massless theory is invariant.143) is without the parameter of chiral transformation. δψ k = −iλψ k γ5 . J µ. We can also derive identities from this relation that would relate various matrix elements of the . (16. µ ∂µ J5 = 0.143) where the variation of the Lagrangian density in (16. Let us note that. (16. therefore. will not be conserved. rather it would satisfy the equation µ ∂µ J5 = −δLm = −2imψ k γ5 ψk . i. (16.16. therefore.6 Anomalous Ward identity 725 is also invariant under the global transformation δψk = −iλγ5 ψk . As is the case with continuous symmetries.141) transforms covariantly under a gauge transformation and.

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

150) and at the classical (tree) level both these currents are conserved (recall that the fermions are massless in the Schwinger model and the gauge group in the present case is U (1)) ∂µ J µ = 0. (16.147) Furthermore. (16. 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 . −). but the ones that are directly of interest to us are γ5 γ µ = ǫµν γν . γ µ γ ν = η µν + γ5 ǫµν . (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. with ǫ01 = 1.16. (16.151) Let us now calculate the one loop amplitude involving a photon and an axial vector current as shown in Fig. 16.149) As a result of these identities.148) There are various simple identities in 1 + 1 dimensions involving the Dirac matrices.152) .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 = . 2 2 (k + p)2 (2π) k µν I5 (p) = −e2 (16.146) .22 and this is given by / / / d2 k Tr γ5 γ µ kγ ν (k + p) . (16. µ ∂µ J5 = 0.

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.73). generalizing the integral to n = 2 − ǫ dimensions would lead to µν I5 (p) = − dn k kλ (k + p)ρ (2π)n k2 (k + p)2 kλ (k + p)ρ dn k dx n (2π) (k + xp)2 − x(1 − x)p2 dx dn k kλ kρ − x(1 − x)pλ pρ (2π)n (k2 − x(1 − x)p2 )2 n = − = − = − 2 Γ(1 − n ) 1 ηλρ iπ 2 2 − dx n n (2π) Γ(2) 2 (−x(1 − x)p2 )1− 2 − Γ(2 − n ) x(1 − x)pλ pρ 2 n . Here we have used the propagators for the fermions following from (15. (16. Furthermore. we note that the amplitude in (16.22: The amplitude with an axial vector current and a photon.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. Furthermore.728 16 Renormalization theory γ5 γ µ k γν p k+p Figure 16. Thus.153) We can evaluate the integral in (16. since .36) in the limit m = 0. Γ(2) (−x(1 − x)p2 )2− 2 (16.71) and (15.

157) The result in (16.16. On the other hand.155) where we have used the gamma matrix identity in n dimension in (15. (16. (16.157) is quite interesting for it leads to µν pµ I5 (p) = − pµ pλ pρ ie2 λρ νµ η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 ie2 νµ ǫ pµ .156) Since each of the terms is now completely ﬁnite.86). in fact. We note from (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 .158) = − . using this relation.149) and the ﬁnal result is obtained to be (the x integration is trivial) µν I5 (p) = − ie2 λρ νµ pλ pρ . the trace over the Dirac matrices can be carried out in two dimensions using the identities (16. It is clear now that the factor (2 − n) = ǫ makes even the ﬁrst term ﬁnite and. However. the ﬁrst term is divergent and we have to be careful in evaluating the trace.6 Anomalous Ward identity 729 γ5 is a two dimensional quantity we have not yet evaluated the trace over the Dirac matrices. η ǫ + η µλ ǫνρ + η νρ ǫµλ 2π p2 (16. π (16. (16. we note that the trace in the ﬁrst term has the form Tr (γ5 γ µ γ λ γ ν γρ )ηλρ = Tr (γ5 γ µ γ λ γ ν γλ ) = (2 − n)Tr (γ5 γ µ γ ν ).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).

This is very important in building models of particle interactions. we always choose a regularization scheme which would preserve gauge invariance in the quantum theory so that the chiral current becomes anomalous. 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.153). µ ∂µ J5 (x) = γλ γ5 γ µ γν Figure 16.730 16 Renormalization theory Recalling that this amplitude is the Fourier transform of the matrix element in (16. the same behavior is ob- . In some gauge theories. one of the two currents (or a linear combination of them) would always become anomalous because of quantum corrections.159) π 2π This shows that although the axial vector current is divergence free (conserved) at the tree level.159). In fact. On the other hand. since gauge invariance is so vital to the physical theory. (16. quantum corrections have made the conservation law anomalous and the anomaly in the divergence of the current is given by (16. Although our discussion here has been within the context of the two dimensional QED (Schwinger model).23: The triangle diagram in four dimensional QED contributing to the anomaly. this leads to the operator relation e2 e2 ǫµν ∂ µ Aν = ǫµν F µν . it is possible to show that there is no regularization which will simultaneously preserve gauge invariance and chiral invariance. the particle multiplets are carefully chosen such that the anomalous contributions coming from various multiplets cancel completely making the gauge current anomaly free. As a result.

such a current is not gauge invariant (although the associated charge is gauge invariant). 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 .160) corresponds to the anomaly in four dimensional QED. they have precisely this kind of coupling. For example. For example. if we have a theory where there are both vector and axial vector gauge couplings of the fermions.161) and the second term in (16.162) which would satisfy a nonanomalous divergence. We note from (16. anomalies have observable eﬀect.6 Anomalous Ward identity 731 tained in any dimension. in this case. as is clear from its structure. the correct pion life time can be calculated only if the chiral anomaly is taken into account. Therefore.160) µ ∂µ J5 = −2imψγ5 ψ + = −2imψγ5 ψ + where the dual ﬁeld strength tensor is deﬁned as 1 ˜ ǫµνλρ F λρ . However. As we have discussed in the last chapter.159) that a similar statement can be made in the case of the Schwinger model as well. In string theories similar anomaly free considerations ﬁx the . We note here that. 16. In grand uniﬁed theories where we treat both quarks and leptons as massless. in the four dimensional QED a graph of the form shown in Fig.16.23 leads to an anomaly in the chiral current. a consistent grand uniﬁed theory cannot be constructed unless we can cancel out the chiral anomaly and this leads to the study of groups and representations which are anomaly free. 8π 2 (16. Furthermore. Fµν = 2 (16. the model cannot be renormalized unless the anomaly is cancelled. it is possible to deﬁne a modiﬁed axial vector current as e2 µ ˜µ J5 = J5 − 2 ǫµνλρ Aν Fλρ . 8π (16.

Coleman. 2426 (1969). Physical Review 75. these topics are beyond the scope of these lectures. Communications in Mathematical Physics 2. 217 (1951). 6. Physical Review 82. J. 7. 1985). Kennedy. W. Erice (1976). 8. Schwinger. 16. 1736 (1949). K. 2425 (1962). However. Salam. Ward. 426 (1951). 84. F. Physical Review 184. 3. Dyson. D. S. Foundations of Quantum Chromodynamics. J. 13. 10. Wightman.732 16 Renormalization theory gauge group uniquely to be SO(32) or E(8) × E(8). 5. N. 75. Bardeen. S. S. Physical Review 128. 12. 392 (1982). Physical Review 25. 1848 (1969). 11.7 References 1. Singapore (1987). Hepp. W. Adler. 486 (1949). Weinberg. Proceedings of the International School of Mathematical Physics “Ettore Majorana”. J. Acta Mathematica 97. . 4. N. Bogoliubov and O. 9. 301 (1966). Bell and R. Parasiuk. S. S. T. 47 (1969). World Scientiﬁc. Aspects of symmetry. Jackiw. E. Physical Review 177. Caswell and A. 14. J. Muta. 2. ed G. 54 (1951). Cambridge University Press (Cambridge. A. C. 227 (1957). Physical Review 118. 838 (1960). A. Wightman in “Renormalization Theory”. S. Nuovo Cimento 60A. Proceedings of Royal Society A64. Velo and A.

17.) It is clear from (17. −→ V (r) = e2 αp . ep denote the ﬁne strucp c ture constant and the physical electric charge of the fermion respectively.1) that at the lowest order the potential is independent of the mass of the fermion and it scales with distance as 1 . namely. calculate quantum corrections to r this potential and if we add the one loop correction to the potential shown in Fig. Gell-Mann and Low were interested in studying how the electric force and the potential behave at extremely short distances. In the same manner. 1+ r 3π rme 733 (17. We have seen earlier at the end of section 8. r (17. (Physical electric charge is determined from the scattering of on-shell electrons.1.1) 1 where in CGS units α2 = p ≃ 137 and αp .Chapter 17 Renormalization group and equation 17. We can. then the potential is obtained to have the form V (r) = 2αp αp 1 ln + ··· . the Coulomb potential can also be obtained from the Born amplitude for the scattering of two electrons exchanging a photon.2) .1 Gell-Mann-Low equation In the early 1950s.10 that the Yukawa potential can be related to the Born amplitude for the scattering of two electrons exchanging a (pseudo) scalar meson. of course.

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 + ··· . When r is 1 1 large. it also behaves as V (r) → α r 1− Λ 2α ln 3π me = αp . Furthermore. in this limit. To understand the origin of this behavior.2) that the potential no longer scales as 1 and furthermore. V (r) → 1 as r → 0 (at exr tremely short distances) as naive scaling would imply.4) where we have identiﬁed Λ 2α ln αp = α 1 − 3π me . r (17.1: The one loop correction to the electromagnetic potential. namely r ≫ me ≫ Λ .734 17 Renormalization group and equation Figure 17. (17. it r depends on the mass of the electron in such a way that we cannot take the limit me → 0. we see from (17.5) This relation can be inverted to express the renormalized parameter in terms of the physical parameter as .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. 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 fact. It is clear that the one loop corrected potential has a very diﬀerent character from the lowest order potential. 1 + r 2 m2 e V (r) = (17. For example.

Let us next investigate what would happen if we deﬁne the electric charge at some arbitrary (intermediate) distance R and not as in (17. For example.6) Expressing the potential (17.8) Furthermore.17. on dimensional grounds we see that the potential can be expressed as . in the limit Λ → ∞.1 Gell-Mann-Low equation 735 α = αp 1 + Λ 2αp ln 3π me + ··· .9) then. We also see that the naive scaling of the potential breaks down and mass dependence comes in due to renormalization eﬀects. (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 + ··· . if we deﬁne αR = RV (R).8). rme ≪ 1 we obtain αp 1 + 2αp 3π Λ ln me V (r) = r αp 2αp 1+ r 3π ln 1− 2αp ln 3π 1 + r 2 Λ2 + ··· 1 + r 2 m2 e + ··· + ··· (17. 1+ r 3π rme This expression coincides with (17. this derivation clariﬁes that the singularity in the potential at me → 0 arises purely because of the renormalization of charge at large distances.2) and also shows that the physical ﬁne structure constant can be identiﬁed with αp = rV (r) . 1 r= m e (17. (17.3) in terms of this (physical) parameter.

736 17 Renormalization group and equation r 1 V (r) = F αR . 1 + R 2 m2 e (17. We note that αR = RV (R) = α 1 − 2α ln 3π 1 + R2 Λ2 + ··· .11) Inverting this relation we obtain α = αR 1 + 2αR ln 3π 1 + R2 Λ2 + ··· . Rme ≪ 1 (the second of these limits corresponds to me → 0).13) × 1− = ≃ αR r αR r 1+ 1+ 2αR ln 3π 2αR R ln + · · · 3π r which is well behaved as we have already mentioned. in the limit Λ → ∞ and rme . Rme ). the dependence of the potential on me can be neglected. 1 + r 2 m2 e + ··· 1 + R 2 m2 e (17.10) where F is a dimensionless function (function of the independent r r dimensionless variables αR . (We note that rme = Rme × R so that it is not an independent quantity.12) Putting this back into the potential (17. . r R (17.3) leads. Rme . which can be seen from (17.3) as follows. 1 + R 2 m2 e (17. R ≪ me . Let us also note that if we deﬁne the electric charge (ﬁne structure constant) as . R . to V (r) = αR r 1+ 2αR ln 3π 2αR ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· 1 + r 2 m2 e 1 + R2 Λ2 1 + r 2 Λ2 .) This function should 1 not be singular as me → 0 which implies that for r.

r R r r . 1 + r 2 Λ2 1 + r 2 m2 e + ··· (17.1 Gell-Mann-Low equation 737 αr = rV (r).16) 2αR R ln + · · · 3π r and this is compatible with (17.15) which. R V (r) = αr (17.16)) we see that the equation governing this change is given by . in general.10)) αr r 1 = F αR . R ≪ 2αR ln 3π leads to the relation αr = αR 1 + = αR 1 + 1 + R2 Λ2 − ln 1 + R 2 m2 e . = F αR . write (see (17.14)) αr .14) α = αR 1 + 2αR ln 3π 2αr ln 3π 1 + R2 Λ2 + ··· 1 + R 2 m2 e 1 + r 2 Λ2 + ··· .17. .18) (see also (17.13) (see also the deﬁnition (17. 1 + r 2 m2 e 1 me . in the limit Λ → ∞ and r.17) From this analysis we see that in the limit of me → 0 we can.12) that we can identify (17. r V (r) = (17.18) This shows that the charge (ﬁne structure constant) changes (runs) with distance and from the second of the equations in (17. then we see from (17. = αr 1 + (17.

then (17. namely. αr= = αp . 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. the bare charge becomes inﬁnity.18) leads to (α0 = αr=0 ) α0 = F (α0 .22) There are two possibilities for the solution of the equation (17. (17. ∂F (αr .16) that. In this case. (17. As R → 0. Gell-Mann-Low equation basically relates the electric charge at one distance (scale) to another (namely. We see from this discussion that through proper renormalization the dependence of the potential on the electron mass me disappears.24) . x=1 (17. x) . dαr 2α2 2α2 = − R = − r. x) ∂x (17. 2. As R → 0. in the present case. dr 3π 3π r (17.21) 1 me We see explicitly from (17. r .19) β(αr ) = − . R → 0 with R = x ﬁxed. αR does have a limit which is nonzero and is inde1 pendent of the value 137 which arises only as an initial condir tion. In this case. αR does not exist and consequently. the theory develops ghosts and other problems.23) 1.738 17 Renormalization group and equation r where dαr dr = −β(αr ). R (17. for example. we note that αp and me arise only as initial conditions in this evolution.18) αr = F αR .20) and this is known as the Gell-Mann-Low equation. if we let r.

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

Namely. u ′ ′ λu = Z1 Z3−2 λ′ . ′ 1 (17.26) ′ where we are denoting Z3 = Z3 (µ′ ) and so on. µ′ = S(p. If we are calculating physical quantities such as the scattering amplitudes. −1 m2 = Zm Z3 m2 . Namely. 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. λ.27) so that if we expand the S matrix elements in a perturbation series . Calculations with either of these sets of parameters will lead to ﬁnite quantities.740 17 Renormalization group and equation scale which illustrates the source of arbitrariness of the second kind. we must have (S and S ′ denote representative S-matrix elements calculated in the two prescriptions) S ′ p. λ′ . u −2 λu = Z1 Z3 λ.25) deﬁne a set of renormalized parameters. µ). ′ ′ m2 = Zm Z3−1 m′ 2 . they should not only be ﬁnite.) Let us consider a renormalizable theory.32)) 1 2 φu = Z3 φ. (17. m′ . a different renormalization scale prescription would lead to φu = Z3 2 φ′ . see the discussion after (16. 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 must also have the same value independent of which set of parameters are used to calculate them. (17. m. On the other hand.

31) = Z3 (µ′ ) Z3 (µ) .33) . n n S ′ p.25) and (17.25) and (17. (17. λ. On the other hand. µ φ. m′ . µ)φu . λ′ . say the nth order. µ′ = (17. m.26) we can also write −1 m2 = Zm (µ)Z3 (µ)m2 . (17. µ) = n an λn .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.2 Renormalization group 741 S(p. µ′ φu . m′ . However.32) Similarly. µ′ − S(p. Here we have identiﬁed z µ . If we can calculate the exact series. where α ≃ 137 . u −1 m′ 2 = Zm (µ′ )Z3 (µ′ )m2 . from (17. u (17. the higher order corrections become negligible very quickly. In such a case. λ′ . a′ λ′ n . in a theory like QCD (where the value of the coupling constant is not small). we can only calculate up to a given order.17. in perturbation theory. such as QED. (17.26) so that we can write the renormalized ﬁelds in the two prescriptions as φ = Z3 2 (λ. all we can say and check is that S ′ p. Let us next see that this arbitrariness in the renormalized parameters is due to a ﬁnite renormalization. λ. µ) = O λn+1 .µ ′ −1 2 −1 φ′ = Z3 1 −2 λ′ . the residual terms can be non-negligible. m.30) which leads to a relation between the two (17. φ′ = z µ′ .29) 1 In a theory. We can invert the relations in (17. we can check this.

′ ′ λ′ = Z1−1 Z3 2 λu .37) zλ (µ′ . (17.32). it follows that (z. (17.38) ′ ′ ′ Since (Z3 . µ) = Z1 (µ′ ) Z1 (µ) −1 Z3 (µ′ ) Z3 (µ) 2 .742 which implies that 17 Renormalization group and equation m′ 2 = zm (µ′ . Z1 ) diﬀer only in ﬁnite parts (the divergent parts are the same since the “µ” dependence comes only in the ﬁnite parts).35) and (17. (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). (17.34) zm (µ . zm .25) and (17. Zm . (17.38) we note that .35) Finally. Z1 ) and (Z3 . µ) = ′ Zm (µ′ ) Zm (µ) −1 Z3 (µ′ ) Z3 (µ) .36) which leads to the relation λ′ = zλ λ. from the deﬁnitions in (17. In fact. Let us next show that these ﬁnite renormalizations (ﬁnite transformations) deﬁne a group. Thus. µ)m2 .26) imply −1 2 λ = Z1 Z3 λu . with (17. with (17. zλ ) are completely ﬁnite. Zm .

These transformations further satisfy z(µ. −1 2 Z3 (µ′′ ) 2 Z3 (µ′ ) Zm (µ′′ ) Zm (µ) Z3 (µ′′ ) Z3 (µ) Z1 (µ′′ ) Z1 (µ′ ) −1 zλ µ′′ . µ = × = Z1 (µ′′ ) Z1 (µ) Z1 (µ′ ) Z1 (µ) −1 2 Z3 (µ′ ) 2 Z3 (µ) 2 Z3 (µ′′ ) = zλ µ′′ . Zm (µ′′ ) Zm (µ′ ) −1 −1 zm µ′′ . µ′ zλ µ′ . physical quantities cannot depend on the arbitrary scale introduced to carry out perturbation calculations in a theory. µ = × = Zm (µ′ ) Zm (µ) −1 Z3 (µ′′ ) Z3 (µ′ ) Z3 (µ′ ) Z3 (µ) = zm µ′′ . z(µ. 2 Z3 (µ) (17. therefore.µ z µ . µ). µ′ ) = zλ (µ′ . µ′ ) = zm (µ′ . −1 zλ (µ. µ′ ) = z −1 (µ′ . −1 zm (µ. Namely.40) The set of ﬁnite renormalizations. µ . µ) = 1 = zm (µ. µ).39) Namely. deﬁne an Abelian group known as the renormalization group and all physical quantities must be invariant under the renormalization group transformations. (17.17. two successive ﬁnite renormalizations lead to an equivalent single ﬁnite renormalization. µ . µ). µ) = zλ (µ.µ = = Z3 (µ′′ ) Z3 (µ) −1 2 ′′ ′ ′ Z3 (µ′′ ) Z3 (µ′ ) −1 2 Z3 (µ′ ) Z3 (µ) −1 2 = z µ′′ . µ′ zm µ′ . µ).2 Renormalization group 743 z µ . µ . .

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

3 Renormalization group equation 745 which follows from the deﬁnition of z in (17.47) Both λu and λ are dimensionless and are related as (compare with (16. 0 0 or.19)) µ dλ dµ = β(λ).32).39)) −1 −1 2 µǫ λ = Z1 (µ)Z3 (µ)µǫ λu = µǫ Zλ (µ)λu . (17. (17. λu = µ µ0 ǫ λZλ (µ). Thus. Here we have used p to denote collectively all the independent external momenta of the 1PI vertex function. As we have noted this deﬁnes the Abelian group of transformations known as the renormalization group. Here µ0 is assumed to be ﬁxed and µ is the variable renormalization scale. Since λu is µ-independent (independent of the renormalization prescription). (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. In dimensional regularization.50) . (17.46) while the unrenormalized coupling can be deﬁned as (in discussions in earlier chapters we had used µ0 = 1) µǫ λu . the couplings are deﬁned with an arbitrary mass scale.48) −2 where we have identiﬁed Zλ = Z1 Z3 .17.49) where we see from (17. 0 (17. Let us deﬁne the renormalized quartic coupling in the scalar ﬁeld theory as µǫ λ. it follows that (see also (17.48) that β(λ) = − ǫ + µ dZλ Zλ dµ λ.

49) decouples from the mass equation (17. b. They are. 2Z m dµ (17. from −1 m2 = Zm (µ)Z3 (µ)m2 = Z m (µ)m2 . . s.55) where the constants a. in general. By dimensional reasoning then they cannot depend on m either. Similarly. functions of λ. m. t. .54) in the MS (MS) scheme. As a result. λ4 + · · · . the equation for the coupling constant (17.51) we obtain dm dµ µ where = −mγm . c.52) and can be solved independently which we will analyze in the next section.746 17 Renormalization group and equation ∂Z ∂µ and the derivative on the right-hand side should be thought of as with λu . (17. (17. µ dZ m . q. γm = γm (λ).52) γm = (17. are independent of µ because the pole terms (in ǫ) are independent of µ as there are no overlapping divergences in a renormalizable theory. 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 + · · · .53) Both β and γm are ﬁnite functions of µ as ǫ → 0 since the divergences in the numerator and denominator cancel out. . We note that this is a consequence of the fact that dimensional regularization is a mass independent regularization. mu held ﬁxed. u u −1 (17. their dependences are of the form (the µ dependence comes in only through the coupling) β = β(λ). . in fact. µ and.

m(µ). (17.59) Equation (17. see also (17.56) which leads to the relation (since the unrenormalized quantities are independent of the renormalization prescription) dΓu (p. mu . λ(µ). or.17.57) is known as the renormalization group equation for the 1PI vertex functions and expresses how amplitudes change when the renormalization scale is changed. dµ µ µ ∂ ∂λ(µ) ∂ ∂m(µ) ∂ +µ +µ − nγ Γ(n) = 0.3 Renormalization group equation 747 Let us next look at the 1PI amplitudes (17. m ∂µ (17. ∂µ ∂µ ∂λ(µ) ∂µ ∂m ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) = 0. mu ﬁxed. ∂µ ∂λ ∂m (17.44) Γ(n) (p. ∂µ µ ∂m . µ) . µ or. . 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).52)) ∂λ . mu . γ = γ(λ). namely. where we have deﬁned (the derivatives are taken with λu . λu ) = Z3 2 (µ) Γ(n) (p. λu ) = 0. m(µ). λ(µ).58) β(λ) = µ γm = − γ = µ ∂Z3 . u −n (17. dµ d −n Z3 2 (µ) Γ(n) (p.57) (n) µ or.49) and (17. µ) = 0.

63) Consequently.57) describes how renormalized amplitudes (1PI vertex functions) change as the renormalization scale is changed ∂ ∂ ∂ + β(λ) − mγm − nγ Γ(n) (pi .748 17 Renormalization group and equation 17. µ) = 0.60) µ where pi denotes the independent external momenta and as we have seen in (17. ∂µ ∂λ µ (17. 1 2 n (17. if we deﬁne the Mandelstam variable associated with the n-point function as s = p2 + p2 + · · · + p2 . the renormalization group equation (17. λ.61) β(λ) = µ γm = − γ = µ ∂Z3 . let us set m = 0 in which case. m. the equation takes the form ∂ ∂ + β(λ) − nγ Γ(n) (pi . mu ﬁxed) ∂λ .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.4 Solving the renormalization group equation As we have seen. ∂µ µ ∂m . µ) = 0.64) . (17. λ.58) (the derivatives are taken with λu . m ∂µ (17. 2Z3 ∂µ For simplicity. ∂µ ∂λ ∂m (17.62) Let us note from our earlier discussions (see (16.

λ. µ2 s . ∂t ∂x (17. t) ∂ρ(x. scaling all momenta by a factor of Γ(n) (pi .17.66) so that for a ﬁxed s. ∂t ∂λ and Γ(n) = s 4−n 2 (17. or. Let us relate the renormalization scale µ to a new dimensionless variable t as (this has the advantage that for ﬁxed s. t). µ) = s 4−n 2 √ s.68) now becomes straightforward from the observation that a linear ﬁrst order diﬀerential equation of the form ∂ρ(x. a change in t can be thought of as changing µ or for a ﬁxed µ. pi · pj .69) Solving the renormalization group equation (17. we can write f (n) pi · pj s .66) we have used ∂µ = − 1 ∂µ ln µ2 = µ ) so that the 2 renormalization group equation (17.70) . λ. (17. t) + v(x) = L(x)ρ(x. It follows now that ∂t ∂ ∂ ∂ =µ = .65) where we have used the fact that the n-point vertex function of a scalar ﬁeld theory is Lorentz invariant and. t = − 1 s ln 2 .68) f (n) t. it can be thought of as scaling the momentum) µ= √ set .67) ∂t ∂ s 1 (where from (17.4 Solving the renormalization group equation 749 then. consequently. f (n) is a dimensionless function of independent dimensionless variables which are Lorentz invariant.62) can be written as ∂ ∂ + β(λ) − nγ Γ(n) = 0. 2 µ (17. s (17. λ. ∂µ ∂µ ∂t ∂t µ (17. changing the renormalization scale merely corresponds to changing t.

in this case. β(λ) ↔ v(x). dt with the boundary condition x(x. of course. are deﬁned by the equation dx(x. t)) = ρ (x(x.72) (17. t = 0)) e Rt 0 (17.750 17 Renormalization group and equation is of hydrodynamic type and is best solved by the method of characteristics. (17.76) . 0) = x.68) with the hydrodynamic problem (17. 0)) = ρ (x(x. λ ↔ x. −t)) e R0 −t R0 −t dt′ L(x(x. (17.73) dt′ L(x(x. Γ(n) ↔ ρ.70) as t ↔ t. Therefore.71).t′ )) .70) can be written as dρ(x) = L(x)ρ(x). where we have used (17. directly identify the quantities in the renormalization group equation (17. t). (17. dt which can be integrated to give ρ (x(x. −t)) e or. we note that the diﬀerential equation (17. we see that solving the equation (17.71) Using (17.74) Translating t by −t.75) dt′ L(x(x. nγ ↔ L. ρ(x) = ρ (x(x. The characteristics.t′ )) . t) = v(x). We can.70) clearly depends on determining its characteristics x(x. we then obtain ρ (x(x.72).t′ )) . (17.

we can obtain Γ(n) = s 4−n 2 f (n) (λ(λ. −t).81) where D Γ(n) represents the superﬁcial degree of divergence of the n-point function and ‘a’ is a constant which can be determined from the explicit form of the amplitude. × en R ln α −t+ln α pi · pj ) s dt′ γ(λ(λ.4 Solving the renormalization group equation 751 so that the solution of the renormalization group equation (17. namely. (17.77). dt (17.79) so that (remember that −t = t → t − ln α.68) requires solving the characteristic equation dλ = β(λ). −t + ln α). This is known as Weinberg’s theorem (the same theorem which also made its appearance in renormalization theory). 1 2 s ln µ2 ) pi → αpi . the leading behavior of the amplitude is given by 4−n 2 Γ(n) → α2 s f (n) (λ(λ.80) then. (17. for large α (large momentum). (17. if . Let us next note that if λ∗ is a ﬁxed point of the coupling constant evolution equation (17.t′ −ln α)) → α4−n (ln α)a · · · = (n) (α)D(Γ ) (ln α)a · · · . 0 pi · pj n R−t dt′ γ(λ(λ. )e s (17.17.78) Note that if we scale all momenta (keeping µ ﬁxed) as s → α2 s.77) and once we know the solution to this equation.t′ )) .

(17. near the ﬁxed point of the renormalization group equation.752 17 Renormalization group and equation β (λ∗ ) = 0.78) behaves as Γ(n) = s pi · pj ntγ(λ∗ ) e s 4−n − n γ(λ∗ ) ln s pi · pj µ2 f (n) λ . then γ(λ∗ ) = 0. .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. = s 2 e 2 ∗ s 4−n 2 f (n) λ∗ .82) (17. namely. If the ﬁxed point of the beta function happens to be at the origin.79).84) As a result.86) which follows from the fact that in perturbation theory γ(λ) is a power series in λ. the n-point amplitude will scale. if λ∗ = 0. at the ﬁxed point. then. the n-point amplitude (17.83) Near the ﬁxed point. pi · pj . λ = λ∗ = constant. = s 4−n 2 s µ2 − n γ(λ∗ ) 2 f (n) λ∗ . under a scaling pi → αpi (17. (17. (17. s (17. as Γ(n) ≃ α4−n(1+γ(λ∗ )) Γ(n) .87) (17. In such a case the amplitudes will have almost naive scaling behavior at very high energies (“almost” because f (n) s also depends on t through the explicit dependence on µ2 which we have not displayed).24)).

(17. In this case.92) On the other hand. t → ∞. 17. t → −∞.89) Before solving this equation explicitly. λ(t) → (17. therefore. t = 0) = λ) λ λ dλ′ β (λ′ ) t = 0 dt′ = t. the behavior of the coupling constant is given by ∞.17. if β(λ) is negative (β < 0.90) λ = 0. the function dλ is monotonically increasing for positive t and we say that dt the origin in the coupling constant space is an infrared ﬁxed point. let us analyze the qualitative behavior of the solution.2. the behavior of β(λ) is shown in Fig. (17. and λ > 0).88) implies that (recall that λ(λ. We note that dλ dt = β λ . the ﬁrst thing that we notice is that (17. Therefore. Qualitatively. this leads to the behavior of the β function as shown in Fig.4 Solving the renormalization group equation 753 It is clear. qualitatively we have . that the solution of the coupling constant equation is quite crucial in understanding the behavior of the renormalized amplitudes at high energy as well as under a change of the renormalization scale. 17. 0.3 and in this case.91) is a ﬁxed point of the beta function. (17. In perturbation theory the beta function has the form of a power series in the coupling constant β(λ) = β0 λn + O λn+1 . 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).

β(λ) λ Figure 17. . λ(t) → 0. Let us next analyze the behavior of the solutions in some detail. namely. Let us assume that the beta function has a nontrivial ﬁxed point at λ∗ on the positive axis. we have two distinct cases to consider (when λ∗ = 0.2: The graph showing that origin in the coupling constant space is an infrared ﬁxed point of the β function. t → −∞.3: The graph showing that the origin in the coupling constant space is an ultraviolet ﬁxed point of the β function. (17.93) and we say that the origin in the coupling constant space is an ultraviolet ﬁxed point. ∞. we recover the two cases we have discussed earlier). t → ∞.754 17 Renormalization group and equation β(λ) λ Figure 17. then expanding around this ﬁxed point.

4 (which shows the behavior of a general β function). (17. 17.94) r > 1. (17.96) which implies that 1. µ → 0 for ﬁxed s).89) leads to λ t= λ dλ′ = β(λ′ ) λ λ or.17.66) that when t → ∞. β(λ) = β0 (λ − λ∗ ) + O (λ − λ∗ )2 . 2.89) leads to . (17. (17. λ → λ∗ .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. if β0 < 0. the arrows indicate the direction in which λ moves as t → ∞ (if we start near the ﬁxed point).97) 2. In the diagram Fig. µ → ∞ and when t → −∞.95) In the ﬁrst case (17. if β0 > 0. λ → λ∗ . β(λ) = β0 (λ − λ∗ )r + O (λ − λ∗ )r+1 . In the second case (see (17.4 Solving the renormalization group equation 755 1. λ − λ∗ 1 dλ′ = ln . then as t → −∞. λ − λ∗ = (λ − λ∗ ) eβ0 t . (17. ′−λ ) β0 (λ β0 λ − λ∗ ∗ (17. then as t → ∞.95)).

the series starts at the cubic order in the coupling) β(g) = β0 g3 + O g5 . Thus. while for QCD.22)). 1 − (r − 1)β0 (λ − λ∗ )r−1 t (17. β0 < 0.99) Both in QED and QCD. (17. (17. 1 λ − λ∗ = or. = 1 − (r − 1)β0 t (λ − λ∗ )r−1 1 − (r − 1)β0 (λ − λ∗ )r−1 t .756 17 Renormalization group and equation β(λ) λ∗ λ β(λ) λ∗ λ Figure 17.100) where β0 is a constant calculable in perturbation theory. (λ − λ∗ )r−1 r−1 λ − λ∗ = (λ − λ∗ )r−1 . g = 0 is . in both cases. the beta function has the form (namely.4: Ultraviolet and Infrared ﬁxed points of the β function. The diﬀerence between the two theories is that for QED β0 > 0 (see. for example. The graph on the left shows an ultraviolet ﬁxed point at λ∗ while the one on the right corresponds to a nontrivial infrared ﬁxed point. 1 (r − 1)β0 r−1 − 1 (λ − λ∗ )r−1 . λ t= λ dλ′ = β(λ′ ) λ λ dλ′ β0 (λ′ − λ∗ )r 1 λ − λ∗ r−1 =− or.

So (setting g∗ = 0.87)).99) g2 = 1 + 2|β0 |g2 t 1 .104) Since the unrenormalized (bare) coupling does not depend on the renormalization scale. β0 < 0) we have from (17. therefore. it follows that . at large momenta almost naive scaling sets in (see discussion after (17. 2β0 e2 t→ (17. (g∗ = 0.102) and perturbation theory breaks down.99) e2 = 1 − 2β0 e2 t 1 .46)-(17. in particular. Such a pole in the evolution of the coupling constant at a particular momentum scale is called the Landau pole. identifying the coupling with electric charge g = e and recalling β0 > 0).38)) µ µ0 ǫ λu = λ 1+ 3λ 16π 2 ǫ .17. as we have seen. we see that the coupling grows and.4 Solving the renormalization group equation 757 the only ﬁxed point – for QED it is the infrared ﬁxed point while for QCD it is the ultraviolet ﬁxed point. (17. the theory behaves like a free theory and such theories are called asymptotically free theories. diverges when 1 . In such theories. the coupling becomes weaker and weaker and ultimately vanishes. For QCD. for QED we have from (17. At inﬁnite energy. − 2β0 t e2 (t) = 1 e2 (17.101) As t increases.48) as well as (16. Let us recall that with the MS (minimal subtraction) scheme at one loop we have (see (17.103) As t → ∞. + 2|β0 |t g 2 (t) = 1 g2 (17. on the other hand. 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.

− 1 1 3 µ + = ln . or. 2 λ(µ) λ(µ) 16π µ λ(µ) = 1 1 λ(µ) − 3 16π 2 µ ln µ . 16π 2 which is ﬁnite as discussed earlier. or. Therefore. ǫλ 1 + 3λ 16π 2 ǫ +µ ∂λ ∂µ 1+ 3λ 8π 2 ǫ = 0. ∂µ 16π 2 β(λ) = µ (17.108) . 2 λ(µ) λ(µ) 16π µ 1 1 3 µ = − ln . (17.106) = 3λ2 .107) so that we have λ(µ) λ(µ) µ µ dλ 3 = 2 λ 16π 2 dµ 3 µ = ln .758 17 Renormalization group and equation µ ∂λu = 0. ∂µ µ µ0 ǫ or. 2 µ 16π µ or. at one loop in the φ4 theory in four dimensions we obtain ∂λ 3λ2 = .105) This leads to −1 µ ∂λ ∂µ = lim −ǫλ 1 + ǫ→0 3λ 16π 2 ǫ 3λ 16π 2 ǫ 1+ 1− 3λ 8π 2 ǫ = lim −ǫλ 1 + ǫ→0 3λ + O(λ2 ) 8π 2 ǫ = lim −ǫλ 1 + ǫ→0 3λ 3λ − 2 + O(λ2 ) 2ǫ 16π 8π ǫ (17.(17.

(17. where we have identiﬁed λ 1 Linv = ∂µ φ∂ µ φ − φ4 . For an inﬁnitesimal scale transformation (α = ǫ = inﬁnitesimal). (In fact.110) The rationale for dividing the Lagrangian density in this way will become clear shortly. only theories involving non-Abelian gauge interactions may be asymptotically free.) Finally. we have δxµ = ǫxµ . we note that unlike the general renormalized amplitudes.109) which simply emphasizes the fact that they are invariant under a change of the renormalization scale. 2 (17.5 Callan-Symanzik equation Let us consider the φ4 theory in four dimensions described by the Lagrangian density L = Linv + Lm .5 Callan-Symanzik equation 759 This is precisely the same behavior as we ﬁnd in QED (see (17. 2 4! Lm = − m2 2 φ . (17. 17. We note that under a scale transformation the coordinates transform as xµ → eα xµ . (17. therefore.17.101)). physical quantities of the theory denoted by P (such as the physical masses of particles etc) satisfy the renormalization group equation ∂ ∂ + β(λ) ∂µ ∂λ µ P = 0. all other theories have the same character as QED.111) (17.113) .112) where α denotes the ﬁnite global parameter of scale transformation.

(17.111) that (here we will use d = 1 in (17. δφ = ǫ (xµ ∂µ + d) φ(x).115) for the scalar theory in four dimensions) λ 3 φ δφ 3! λ 3 φ ǫ (1 + xν ∂ν ) φ 3! δLinv = ∂µ φ∂ µ δφ − = ∂µ φ∂ µ (ǫ (1 + xν ∂ν ) φ) − = ǫ ∂µ φ∂ µ φ − λ 4 φ + ∂ µ φ∂µ