STEFAN POKORSKI

CAMBRIDGE UNIVERSITY PRESS

Gauge Field Theories, Second Edition Quantum ﬁeld theory forms the present theoretical framework for our understanding of the fundamental interactions of particle physics. This up-dated and expanded text examines gauge theories and their symmetries with an emphasis on their physical and technical aspects. Beginning with a new chapter giving a systematic introduction to classical ﬁeld theories and a short discussion of their canonical quantization and the discrete symmetries C, P and T , the book provides a brief exposition of perturbation theory, the renormalization programme and the use of the renormalization group equation. It then explores topics of current research interest including chiral symmetry and its breaking, anomalies, and low energy effective lagrangians and some basics of supersymmetry. A chapter on the basics of the electroweak theory is now included. P ROFESSOR P OKORSKI, a distinguished theoretical physicist, has presented here a self-contained text for graduate courses in physics, the only prerequisite is some grounding in quantum ﬁeld theory. Born in 1942, Professor Stefan Pokorski received his PhD in theoretical physics in 1967 from Warsaw University, where he is now holder of the Chair in Theoretical Particle Physics. He has been a member of the Polish Academy of Science since 1991 and was Director of the Institute for Theoretical Physics, University of Warsaw, from 1984 to 1994 and President of the Polish Physical Society from 1992 to 1994. A visiting scientist at CERN and the Max-Planck Institute for Physics, Munich, for periods of time totalling almost ten years, Professor Pokorski has also made regular visits to universities and laboratories around the world. Professor Pokorski has had many scientiﬁc articles published in leading international journals and in recent years has concentrated on physics beyond the standard model and supersymmetry.

This page intentionally left blank

**CAMBRIDGE MONOGRAPHS ON MATHEMATICAL PHYSICS General editors: P. V. Landshoff, D. R. Nelson, D. W. Sciama, S. Weinberg
**

J. Ambjørn, B. Durhuus and T. Jonsson Quantum Geometry: A Statistical Field Theory Approach A. M. Anile Relativistic Fluids and Magneto-Fluids J. A. de Azc´ rraga and J. M. Izquierdo Lie Groups, Lie Algebras, Cohomology and Some a Applications in Physics† J. Bernstein Kinetic Theory in the Early Universe G. F. Bertsch and R. A. Broglia Oscillations in Finite Quantum Systems N. D. Birrell and P. C. W. Davies Quantum Fields in Curved Space† S. Carlip Quantum Gravity in 2 + 1 Dimensions J. C. Collins Renormalization† M. Creutz Quarks, Gluons and Lattices† P. D. D’Eath Supersymmetric Quantum Cosmology F. de Felice and C. J. S. Clarke Relativity on Curved Manifolds† P. G. O. Freund Introduction to Supersymmetry† J. Fuchs Afﬁne Lie Algebras and Quantum Groups† J. Fuchs and C. Schweigert Symmetries, Lie Algebras and Representations: A Graduate Course for Physicists R. Gambini and J. Pullin Loops, Knots, Gauge Theories and Quantum Gravity† M. G¨ ckeler and T. Sch¨ cker Differential Geometry, Gauge Theories and Gravity† o u C. G´ mez, M. Ruiz Altaba and G. Sierra Quantum Groups in Two-dimensional Physics o M. B. Green, J. H. Schwarz and E. Witten Superstring Theory, volume 1: Introduction† M. B. Green, J. H. Schwarz and E. Witten Superstring Theory, volume 2: Loop Amplitudes, Anomalies and Phenomenology† S. W. Hawking and G. F. R. Ellis The Large-Scale Structure of Space-Time† F. Iachello and A. Aruna The Interacting Boson Model F. Iachello and P. van Isacker The Interacting Boson–Fermion Model C. Itzykson and J.-M. Drouffe Statistical Field Theory, volume 1: From Brownian Motion to Renormalization and Lattice Gauge Theory† C. Itzykson and J.-M. Drouffe Statistical Field Theory, volume 2: Strong Coupling, Monte Carlo Methods, Conformal Field Theory, and Random Systems† J. I. Kapusta Finite-Temperature Field Theory† V. E. Korepin, A. G. Izergin and N. M. Boguliubov The Quantum Inverse Scattering Method and Correlation Functions† M. Le Bellac Thermal Field Theory† N. H. March Liquid Metals: Concepts and Theory I. M. Montvay and G. M¨ nster Quantum Fields on a Lattice† u A. Ozorio de Almeida Hamiltonian Systems: Chaos and Quantization† R. Penrose and W. Rindler Spinors and Space-time, volume 1: Two-Spinor Calculus and Relativistic Fields† R. Penrose and W. Rindler Spinors and Space-time, volume 2: Spinor and Twistor Methods in Space-Time Geometry† S. Pokorski Gauge Field Theories, 2nd edition J. Polchinski String Theory, volume 1: An Introduction to the Bosonic String J. Polchinski String Theory, volume 2: Superstring Theory and Beyond V. N. Popov Functional Integrals and Collective Excitations† R. G. Roberts The Structure of the Proton† J. M. Stewart Advanced General Relativity† A. Vilenkin and E. P. S. Shellard Cosmic Strings and Other Topological Defects† R. S. Ward and R. O. Wells Jr Twistor Geometry and Field Theories†

† Issued as a paperback

This page intentionally left blank

**Gauge Field Theories Second Edition
**

STEFAN POKORSKI

Institute for Theoretical Physics, University of Warsaw

PUBLISHED BY CAMBRIDGE UNIVERSITY PRESS (VIRTUAL PUBLISHING) FOR AND ON BEHALF OF THE PRESS SYNDICATE OF THE UNIVERSITY OF CAMBRIDGE The Pitt Building, Trumpington Street, Cambridge CB2 IRP 40 West 20th Street, New York, NY 10011-4211, USA 477 Williamstown Road, Port Melbourne, VIC 3207, Australia http://www.cambridge.org © Cambridge University Press 1987, 2000 This edition © Cambridge University Press (Virtual Publishing) 2003 First published in printed format 1987 Second edition 2000 A catalogue record for the original printed book is available from the British Library and from the Library of Congress Original ISBN 0 521 47245 8 hardback Original ISBN 0 521 47816 2 paperback (Second edition)

ISBN 0 511 01746 4 virtual (netLibrary Edition)

In memory of Osterns – my mother’s family

This page intentionally left blank

Contents

Preface to the First Edition Preface to the Second Edition 0 0.1 0.2 Introduction Gauge invariance Reasons for gauge theories of strong and electroweak interactions QCD Electroweak theory Classical ﬁelds, symmetries and their breaking The action, equations of motion, symmetries and conservation laws Equations of motion Global symmetries Space-time transformations Examples Classical ﬁeld equations Scalar ﬁeld theory and spontaneous breaking of global symmetries Spinor ﬁelds Gauge ﬁeld theories U (1) gauge symmetry Non-abelian gauge symmetry From classical to quantum ﬁelds (canonical quantization) Scalar ﬁelds The Feynman propagator Spinor ﬁelds Symmetry transformations for quantum ﬁelds Discrete symmetries Space reﬂection Time reversal Charge conjugation Summary and the CPT transformation ¯ CP violation in the neutral K0 –K0 -system Problems

page xvii xviii 1 1 3 3 5 11 12 12 13 16 18 20 20 22 29 29 31 35 36 39 40 45 48 48 53 56 62 64 68

1 1.1

1.2

1.3

1.4

1.5

ix

x 2 2.1

Contents Path integral formulation of quantum ﬁeld theory Path integrals in quantum mechanics Transition matrix elements as path integrals Matrix elements of position operators Vacuum-to-vacuum transitions and the imaginary time formalism General discussion Harmonic oscillator Euclidean Green’s functions Path integral formulation of quantum ﬁeld theory Green’s functions as path integrals Action quadratic in ﬁelds Gaussian integration Introduction to perturbation theory Perturbation theory and the generating functional Wick’s theorem An example: four-point Green’s function in λ 4 Momentum space Path integrals for fermions; Grassmann algebra Anticommuting c-numbers Dirac propagator Generating functionals for Green’s functions and proper vertices; effective potential Classiﬁcation of Green’s functions and generating functionals Effective action Spontaneous symmetry breaking and effective action Effective potential Green’s functions and the scattering operator Problems Feynman rules for Yang–Mills theories The Faddeev–Popov determinant Gauge invariance and the path integral Faddeev–Popov determinant Examples Non-covariant gauges Feynman rules for QCD Calculation of the Faddeev–Popov determinant Feynman rules Unitarity, ghosts, Becchi–Rouet–Stora transformation Unitarity and ghosts BRS and anti-BRS symmetry Problems Introduction to the theory of renormalization Physical sense of renormalization and its arbitrariness Bare and ‘physical’ quantities Counterterms and the renormalization conditions 71 71 71 75 76 76 78 82 83 83 87 88 90 90 92 93 97 100 100 102 105 105 107 109 111 113 120 124 124 124 126 129 132 133 133 135 140 140 143 147 148 148 148 152

2.2

2.3

2.4

2.5

2.6

2.7 3 3.1

3.2

3.3

4 4.1

Contents Arbitrariness of renormalization Final remarks Classiﬁcation of the divergent diagrams Structure of the UV divergences by momentum power counting Classiﬁcation of divergent diagrams Necessary counterterms λ 4 : low order renormalization Feynman rules including counterterms Calculation of Fig. 4.8(b) Comments on analytic continuation to n = 4 dimensions Lowest order renormalization Effective ﬁeld theories Problems Quantum electrodynamics Ward–Takahashi identities General derivation by the functional technique Examples Lowest order QED radiative corrections by the dimensional regularization technique General introduction Vacuum polarization Electron self-energy correction Electron self-energy: IR singularities regularized by photon mass On-shell vertex correction Massless QED Dispersion calculation of O(α) virtual corrections in massless QED, in (4 ∓ ε) dimensions Self-energy calculation Vertex calculation Coulomb scattering and the IR problem Corrections of order α IR problem to all orders in α Problems Renormalization group Renormalization group equation (RGE) Derivation of the RGE Solving the RGE Green’s functions for rescaled momenta RGE in QED Calculation of the renormalization group functions β, γ , γm Fixed points; effective coupling constant Fixed points Effective coupling constant Renormalization scheme and gauge dependence of the RGE parameters

xi 153 156 157 157 159 161 164 164 166 168 170 173 175 177 179 179 181 184 184 185 187 190 191 194 196 197 198 200 200 205 208 209 209 209 212 214 215 216 219 219 222 224

4.2

4.3

4.4 5 5.1

5.2

5.3 5.4

5.5

6 6.1

6.2 6.3

6.4

xii Contents Renormalization scheme dependence Effective α in QED Gauge dependence of the β-function Problems 224 226 227 228 230 230 230 233 235 237 237 238 242 243 243 247 249 249 250 254 256 256 259 261 262 269 272 272 272 274 277 279 282 282 284 287 290 290 291 293 293 298 302 7 7.2 7.5 .3 8.1 Scale invariance and operator product expansion Scale invariance Scale transformations Dilatation current Conformal transformations Broken scale invariance General discussion Anomalous breaking of scale invariance Dimensional transmutation Operator product expansion (OPE) Short distance expansion Light-cone expansion The relevance of the light-cone Electron–positron annihilation Deep inelastic hadron leptoproduction Wilson coefﬁcients and moments of the structure function Renormalization group and OPE Renormalization of local composite operators RGE for Wilson coefﬁcients OPE beyond perturbation theory OPE and effective ﬁeld theories Problems Quantum chromodynamics General introduction Renormalization and BRS invariance.7 8 8. counterterms Asymptotic freedom of QCD The Slavnov–Taylor identities The background ﬁeld method The structure of the vacuum in non-abelian gauge theories Homotopy classes and topological vacua Physical vacuum -vacuum and the functional integral formalism Perturbative QCD and hard collisions Parton picture Factorization theorem Deep inelastic electron–nucleon scattering in ﬁrst order QCD (Feynman gauge) Structure functions and Born approximation Deep inelastic quark structure functions in the ﬁrst order in the strong coupling constant Final result for the quark structure functions 8.3 7.1 7.2 8.4 7.5 7.6 7.4 8.

5 9.4 9.6 10 10.1 .7 Hadron structure functions.Contents 8. light-like gauge Beyond the one-loop approximation Comments on the IR problem in QCD Problems Chiral symmetry.3 12 12.2 11.1 10.1 11.1 9.3 10. probabilistic interpretation Light-cone variables.2 10. spontaneous symmetry breaking Chiral symmetry of the QCD lagrangian Hypothesis of spontaneous chiral symmetry breaking in strong interactions Phenomenological chirally symmetric model of the strong interactions (σ -model) Goldstone bosons as eigenvectors of the mass matrix and poles of Green’s functions in theories with elementary scalars Goldstone bosons as eigenvectors of the mass matrix General proof of Goldstone’s theorem Patterns of spontaneous symmetry breaking Goldstone bosons in QCD Spontaneous and explicit global symmetry breaking Internal symmetries and Ward identities Preliminaries Ward identities from the path integral Comparison with the operator language Ward identities and short-distance singularities of the operator products Renormalization of currents Quark masses and chiral perturbation theory Simple approach Approach based on use of the Ward identity Dashen’s theorems Formulation of Dashen’s theorems Dashen’s conditions and global symmetry broken by weak gauge interactions Electromagnetic π + –π 0 mass difference and spectral function sum rules Electromagnetic π + –π 0 mass difference from Dashen’s formula Spectral function sum rules Results Higgs mechanism in gauge theories Higgs mechanism Spontaneous gauge symmetry breaking by radiative corrections Dynamical breaking of gauge symmetries and vacuum alignment Dynamical breaking of gauge symmetry Examples Problems Standard electroweak theory The lagrangian xiii 304 306 312 314 315 317 317 320 324 327 327 330 333 337 342 342 342 344 347 348 351 353 353 354 356 356 358 362 362 363 366 369 369 373 379 379 382 388 389 391 9 9.2 9.4 11 11.3 9.6 8.

3 13.2 13.3 12.4 Contents Electroweak currents and physical gauge boson ﬁelds Fermion masses and mixing Phenomenology of the tree level lagrangian Effective four-fermion interactions Z 0 couplings Beyond tree level Renormalization and counterterms Corrections to gauge boson propagators Fermion self-energies Running α(µ) in the electroweak theory Muon decay in the one-loop approximation Corrections to the Z 0 partial decay widths Effective low energy theory for electroweak processes QED as the effective low energy theory Flavour changing neutral-current processes QCD corrections to CP violation in the neutral kaon system Problems Chiral anomalies Triangle diagram and different renormalization conditions Introduction Calculation of the triangle amplitude Different renormalization constraints for the triangle amplitude Important comments Some physical consequences of the chiral anomalies Chiral invariance in spinor electrodynamics π 0 → 2γ Chiral anomaly for the axial U (1) current in QCD.2 12.xiv 12.2 .5 12.1 14.4 14 14.6 12.7 13 13. UA (1) problem Anomaly cancellation in the SU (2) × U (1) electroweak theory Anomaly-free models Anomalies and the path integral Introduction Abelian anomaly Non-abelian anomaly and gauge invariance Consistent and covariant anomaly Anomalies from the path integral in Euclidean space Introduction Abelian anomaly with Dirac fermions Non-abelian anomaly and chiral fermions Problems Effective lagrangians Non-linear realization of the symmetry group Non-linear σ -model Effective lagrangian in the ξa (x) basis Matrix representation for Goldstone boson ﬁelds Effective lagrangians and anomalies 394 398 402 403 406 407 407 411 418 419 422 430 435 438 441 445 456 457 457 457 459 464 465 469 469 471 473 475 478 478 478 480 481 484 486 486 488 491 492 495 495 495 500 502 504 12.1 13.

Contents Abelian anomaly The Wess–Zumino term Problems 15 15.4 Introduction to supersymmetry Introduction The supersymmetry algebra Simple consequences of the supersymmetry algebra Superspace and superﬁelds for N = 1 supersymmetry Superspace Superﬁelds Supersymmetric lagrangian.5 15.3 15. n 2 = 0 Convention for the logarithm Spence functions Appendix C: Feynman rules for the Standard Model Propagators of fermions Propagators of the gauge bosons Propagators of the Higgs and Goldstone bosons Propagators of the ghost ﬁelds Mixed propagators (only counterterms exist) Gauge interactions of fermions Yukawa interactions of fermions Gauge interactions of the gauge bosons Self-interactions of the Higgs and Goldstone bosons Gauge interactions of the Higgs and Goldstone bosons Gauge interactions of the ghost ﬁelds Interactions of ghosts with Higgs and Goldstone bosons xv 505 506 508 509 509 511 513 515 515 519 521 524 531 539 539 546 550 551 552 555 555 556 557 558 559 559 560 560 561 561 562 563 563 564 565 566 567 567 570 571 573 574 578 579 15.2 15. C) Solutions of the free Weyl and Dirac equations and their properties Parity Time reversal Charge conjugation Appendix B: Feynman rules for QED and QCD and Feynman integrals Feynman rules for the λ 4 theory Feynman rules for QED Feynman rules for QCD Dirac algebra in n dimensions Feynman parameters Feynman integrals in n dimensions Gaussian integrals λ-parameter integrals Feynman integrals in light-like gauge n · A = 0. Wess–Zumino model Supersymmetry breaking Supergraphs and the non-renormalization theorem Appendix A: Spinors and their properties Lorentz transformations and two-dimensional representations of the group S L(2.6 15.7 .1 15.

and four-point functions General expressions for the one-loop vector boson self-energies Appendix E: Elements of group theory Deﬁnitions Transformation of operators Complex and real representations Traces σ -model References Index 583 583 585 586 591 591 593 593 594 596 599 605 .xvi Contents Appendix D: One-loop Feynman integrals Two-point functions Three.

Marco Roncadelli. Jacek Prentki. Heinrich u o Leutwyler. I am grateful to Wojciech Debski. for instance at the level of two volumes by Bjorken & Drell (1964. Wilfried Buchm¨ ller. 1985 xvii . My warm thanks go to Antonio Bassetto. Krzysztof Meissner and Michał Spalinski. Physical problems are discussed only as illustrations of certain theoretical ideas and of computational methods. My thanks go to Andrzej Czechowski for his collaboration at the early stage of this project and for numerous useful discussions. Warsaw University. Peter Minkowski. and especially to Peter Landshoff for reading most of the preliminary manuscript. Marek ¸ Olechowski. 1965). No attempt has been made to review systematically the present status of the theory of fundamental interactions.Preface to the First Edition This book has its origin in a long series of lectures given at the Institute for Theoretical Physics. and particularly to Paweł Krawczyk for checking a large part of the calculations contained in this book. Maurice Jacob and Peter Landshoff o for their interest in this work and strong encouragement. The book is intended to be a relatively concise reference to some of the ﬁeld theoretical tools used in contemporary research in the theory of fundamental interactions. I am also grateful to several of my younger colleagues at the Institute for Theoretical Physics in Warsaw for their stimulating interactions. Andrzej Turski. It is a technical book and not easy reading. Wojciech Kr´ likowski. Olivier Piguet. I am grateful to Wojciech Kr´ likowski. Henri Ruegg and Wojtek Zakrzewski for reading various chapters of this book and for many useful comments. It is addressed to graduate students and to young research workers in theoretical physics who have some knowledge of quantum ﬁeld theory in its canonical formulation. Jacek Pawełczyk. Robert Budzynski. Stefan Pokorski Warsaw. Finally my thanks go to Zoﬁa Zi´ łkowska for her contribution to the preparation o of the manuscript.

Several smaller changes and corrections have been made in a number of places in the text. Chapter 1 makes the book more self-contained. including counterterms. I would also like to xviii . in which the modern approach to effective ﬁeld theories is presented. An important addition is Section 7. Appendices A. It gives a systematic introduction to classical ﬁeld theories and a brief discussion of their canonical quantization. P and T . Appendix C contains the complete set of Feynman rules for the Standard Model. Carlos Savoy and Kay Wiese for their numerous comments and corrections. I hope that the additions leave intact the main feature of the book: it is still not easy reading! I am grateful to many people for their help in the completion of this second edition. I am deeply indebted to Howie Haber for his careful reading of a large part of the new material. Thanks. which was missing in the ﬁrst edition. Jacek Pawełczyk. I also thank Mikolaj Misiak and Janusz Rosiek for their collaboration on Sections 7. This is an important completion for a modern book on quantum ﬁeld theory and fundamental interactions.7. The new Appendix A is a substantial extension of the previous Appendix C. In particular. Riccardo Guida. I am grateful to Zygmunt Ajduk. which is not easily available in the literature. The main additions are Chapter 1 and Chapter 12 and extended Appendices. Chapter 12 gives a concise but systematic and self-contained introduction to the electroweak theory. Ratindranath Akhoury. C and D are new.6. Marek Olechowski. Also in Chapter 1 the reader can ﬁnd a thorough discussion of discrete symmetries C. Very special thanks go to Piotr Chankowski. Krzysztof Meissner. Piotr.7 and 15.Preface to the Second Edition This new edition offers a substantial extension of topics covered by the book. as some intuition based on canonical quantization proves to be very useful even if the main emphasis is on the path integral approach. respectively. His help and collaboration in writing Chapters 1 and 12 and the Appendices was absolutely invaluable.

1999 .edu.fuw.edu.Preface to the Second Edition xix thank Peter Landshoff for his constant encouragement and patience during quite a few years before this edition ﬁnally materialized. Comments and corrections are welcome by e-mail at pokorski@fuw.pl/∼pokorski/ Stefan Pokorski Warsaw.pl They will be made accessible at the www page: http://www.

This page intentionally left blank .

A tiny mass for the photon would destroy the gauge invariance of electrodynamics but leave unaffected all its Earth-bound effects. nobody is seriously contesting electric charge conservation whereas experiments searching for proton decay belong to the present frontiers of physics.0 Introduction 0. In particular.3 × 1023 years (68% C.1 Gauge invariance It seems appropriate to begin this book by quoting the following experimental information (Review of Particle Properties 1996): electron life-time neutron life-time for the electric charge non-conserving decays (n → p + neutrals) proton life-time photon mass neutrino (νe ) mass > 4.L. It should be stressed at this point that U (1) gauge invariance implies global U (1) invariance but. of course. the opposite is not true.) 1019 years > 1032 years < 6 × 10−22 MeV < 15 × 10−6 MeV From the ﬁrst three lines of this table we see that the conservation of the electric charge is proved experimentally much more poorly than the conservation of the baryonic charge. such a theory is also renormalizable (Matthews 1949. Nevertheless. including the quantum ones. † The reader who is not familiar with notions of global symmetry and gauge symmetry is advised to read Chapter 1 ﬁrst. It also implies the masslessness of the photon and as seen from the table the present experimental limit on the photon mass is indeed many orders of magnitude better than for the other ‘massless’ particle: the electron neutrino. 1 . as long as the electric charge was conserved. Exact gauge invariance protects the conservation of the electric charge. The reason for this lies in the general conviction that the theory of electromagnetism has gauge symmetry† whereas no gauge invariance principle can be invoked to protect baryonic charge conservation.

however. non-abelian gauge bosons carry the group charges and their mass terms in the lagrangian would in general. In particular. this statement accounts not only for electrodynamics with its U (1) abelian gauge symmetry. This sounds unphysical and we are led to propose that this freedom of convention is present independently at every space-time point or is not present at all as an exact law of nature. It can be redeﬁned freely.2 0 Introduction Boulware 1970. The terminology ‘gauge invariance’ can be traced back to Weyl’s studies (Weyl 1919) of invariance under space-time-dependent changes of gauge (scale) in an attempt to unify gravity and electromagnetism. And non-abelian gauge symmetries are more restrictive and more profound than the U (1) symmetry. though this has been much better proved experimentally than electric charge conservation. ∂xµ → exp[ie f (x)/h c] ¯ . unless introduced by spontaneous symmetry breaking. The standard experimental evidence for gauge theories of weak and strong interactions is brieﬂy summarized in the next section. nevertheless. provided that all observers in the universe redeﬁne it in exactly the same way. We end this section with a short historical ‘footnote’ (Pauli 1933). Salam & Strathdee 1970) since longitudinally polarized photons decouple from the conserved current. because we expect that gauge invariance is behind it. As we know at present. and we doubt baryon charge conservation. but also for weak and strong interactions successfully described by gauge ﬁeld theories with non-abelian symmetry groups: SU (2) × U (1) and SU (3). In 1926. Those who remain sceptical should then remember that gauge theories give an economical description of the laws of nature based on well-deﬁned underlying principles which has been phenomenologically successful. destroy not only the gauge symmetry but also the current conservation and therefore the renormalizability of the theory. respectively.) This aesthetical argument may not convince everybody. (Approximate global symmetries may. be and are very useful in describing the fundamental interactions. We do not trust global symmetries as candidates for underlying ﬁrst principles! A global symmetry gives us a freedom of convention: choice of a reference frame (the phase of the electron wave-function for the U (1) symmetry of electrodynamics). This attempt proved. Thus one may ask what are the virtues of gauge invariance? We trust electric charge conservation. unsuccessful. Fock observed (Fock 1926) that one could base quantum electrodynamics (QED) of scalar particles on the operator −ih ¯ ∂ e − Aµ µ ∂x c where Aµ is the electromagnetic four-potential and that the equations were invariant under the transformation Aµ → Aµ + ∂ f (x) .

SU (6) is obtained by adjoining the group SU (2) of spin rotation to the internal symmetry group SU (3) for baryons (qqq) and mesons (q¯ ). successful as it was. appeared. the known hadrons indeed realize only those q representations of SU (3) which are given by the composite model. 0. the arrow denotes spin Sz = 1 for each quark. Thus. The quark model for hadrons. Similar ideas were also proposed much earlier by Klein (1939) and by Shaw (1955). Assigning quarks q to the fundamental representation of SU (3). colour states. Bardeen. otherwise indistinguishable. Zweig 1964). The latter was postulated as a rationale for the successful SU (3) classiﬁcation of hadrons (today one should say ﬂavour SU (3)). not realized by any known hadrons. In 1929. In particular. Nambu 1966. This is not the case in the + original quark model as can be immediately seen from inspection of the ++ ( 3 ) 2 wave-function which must be u↑u↑u↑. Assuming furthermore that (i) strong interactions are invariant under global SU (3)colour . The concept of gauge transformations was generalized to non-abelian gauge groups by Yang & Mills (1954). baryon wave-functions should be antisymmetric in spin and ﬂavour degrees of freedom. Han & Nambu 1965.0. QCD QCD emerged as a development of the Gell-Mann–Zweig quark model for hadrons (Gell-Mann 1964. Fritzsch & Gell-Mann 1973). London (1927) pointed out the similarity of Fock’s to Weyl’s earlier work: instead of Weyl’s scale change a local phase change was considered by Fock. and giving them spin one-half one obtains the phenomenologically successful SU (3) and SU (6) schemes. Weyl studied invariance under this phase change but he kept unchanged his earlier terminology ‘gauge invariance’ (Weyl 1929). to have difﬁculties in reconciling the Fermi statistics for quarks with the most natural assumption that in the lowest-lying hadronic states all the relative angular momenta among constituent quarks vanish (s-wave states).2 Reasons for gauge theories 3 which he called gradient transformation. 2 2 The difﬁculty can be resolved by postulating a new internal quantum number for quarks which has been called colour (Greenberg 1964. however.2 Reasons for gauge theories of strong and electroweak interactions We summarize very brieﬂy the standard arguments in favour of gauge theories in elementary particle physics. u denotes the quark with electric charge Q = 3 . Fermi statistics is saved by using a totally antisymmetric colour wave-function εabc u a ↑u b ↑u c ↑. Both quantum chromodynamics (QCD) and the Glashow–Salam–Weinberg theory are syntheses of our understanding of fundamental interactions progressing over many past years. If a quark of each ﬂavour has three.

only colour symmetry can be assumed to be an exact symmetry (ﬂavour . The concept of colour certainly underlies what we believe to be the true theory of strong interactions.1) = q 2 Qq = 3 × 11 3 4 9 + 1 9 + 4 9 + 1 9 + 1 9 + ··· (0.1 and the ratio R= is predicted to be R= e2 q 2 e 2 Qq σ (e+ e− → hadrons) σ (e+ e− → µ+ µ− ) (0. One is based on the parton model (Feynman 1972) approach to the reaction e+ e− → hadrons. we can understand why only qqq and q¯ states. Firstly. The concept of colour is also supported by at least two other. it is assumed that strong interactions act on the colour quantum numbers and only on them. namely QCD. exist in nature: the singlet representation appears only in the 3 × 3 × 3 ¯ and 3 × 3 products. Experimentally there is no evidence for any ﬂavour dependence of strong forces.e. In addition. though not satisfactorily understood yet. and not qq or q qqqq etc.4 0 Introduction Fig. is expected to be outside of QCD. transformations (the states may then be classiﬁed by their SU (3)colour representation) and (ii) physical hadrons are colourless. the theory also has several other basic features which are partly suggested by experimental observations and partly required by theoretical consistency. they are singlets under SU (3)colour (quark conﬁnement). However.. Here again the number of quark states matters in explaining the width of π 0 → 2γ . 0.1. This problem will be discussed in more detail in Chapter 13. strong arguments. i. The total cross section for this process is then given by the diagram in Fig. 0. 9 Yet another reason for colour is provided by the decay π 0 → 2γ . all ﬂavour-dependent effects can be explained by quark mass differences and the origin of the quark masses.2) = (including quarks up to b) The experimental value of R is in good agreement with this prediction and in poor agreement with the colourless prediction 11 . The process e+ e− → γ → hadrons in the parton model. The sum is taken over ¯ all hadronic states in the reaction e+ e− → γ → qq → hadrons.

Politzer 1973) and presumably. Asymptotic freedom means that the forces become negligible at short distances and consequently the interaction between quarks by exchange of non-abelian gauge ﬁelds (gluons) is consistent with the successful. An important line of argument in favour of gluons as vector bosons begins with approximate chiral symmetry of strong interactions (Chapter 9). Gell-Mann & Leutwyler 1973). To introduce the Glashow– Salam–Weinberg theory we recall ﬁrst that the effective Fermi lagrangian for the . of quarks and gluons. this cannot be fully satisfactory since experimenters collide hadrons and not quarks and gluons. at present. conﬁnement of quarks. at its present stage the theory provides us with calculable corrections to the free-ﬁeld behaviour of quarks and gluons in the parton model and can be tested in the deep inelastic region. In recent years there has been a lot of research in QCD perturbation theory neglecting the unsolved conﬁnement problem. Electroweak theory There is. Both are welcome features. Fritzsch. It has been shown that only non-abelian gauge theories are asymptotically free (Coleman & Gross 1973). as the ﬁrst approximation. Given the accuracy of the calculations and of the experimental data one cannot claim yet to have strongly positive experimental veriﬁcation of perturbative QCD predictions. i. conﬁnement of colour is desirable in view of the so far unsuccessful experimental search for free quarks and to avoid a proliferation of unwanted states. has profound implications: asymptotic freedom (Gross & Wilczek 1973. A theory with axial-vector gluons based on the SU (3) group cannot be consistently renormalized because of anomalies (Chapter 13). experimental results are certainly consistent with QCD (in particular jet physics already provides us with good evidence for the vector nature of gluons) and in view of its elegance and self-consistency there are few sceptical of its chance of being the theory of strong interactions. reﬂecting strongly singular structure of the non-abelian gauge theory in the IR region. impressive experimental evidence for the electroweak gauge theory with the gauge symmetry spontaneously broken. combined with the assumption that it is a gauge symmetry (Han & Nambu 1965. description of the deep inelastic scattering in the framework of the parton model. Thus. Of course. Coupling of fermions with vector and axial-vector ﬁelds. is chirally invariant. though not proven. Thus we arrive at vector gluon interaction. has not yet been proved to follow from QCD but it is likely to be true. Once we assume colourful quarks as elementary objects in hadrons. but not with scalars or pseudoscalars.2 Reasons for gauge theories 5 symmetry is evidently broken) and this. Conﬁnement of the colour quantum numbers. However.e.0. One can nevertheless argue that it is a justiﬁable approximation at short distances.

In terms of the lepton and quark ﬁelds it can be written as follows Jµ = f f ¯ L γ µT − f L (0. for instance. has conventionally been taken to be (for a systematic account of the weak interactions phenomenology see. Both factors . u d (0. valid at low energies.6 0 Introduction charged-current weak interactions. and (ii) always conserve strangeness to a very good accuracy.3) where the Fermi β-decay constant G µ = 1.5) The subscript L stands for the left-handed fermions. as known in the 1960s. f L = 1 (1 − γ5 ) 2 νe e− . for the weak interactions. 1973). νµ µ− . Gasiorowicz (1966) and more recently Abers & Lee (1973) and Taylor (1976)) √ † Leff (x) = −2 2 G µ Jµ (x)J µ (x) (0. However.4) with T ± = 1 (σ 1 ± σ 2 ) = T 1 ± iT 2 2 where σ a are Pauli matrices and for weak interactions. We now extend the Fermi–Cabibbo theory by several additional assumptions.2 occurring with a similar strength to the charged-current reactions. each with V–A structure. those in Fig.4) it is natural to postulate SU (2) symmetry and consequently the existence of the neutral current corresponding to the third generator of the SU (2) [T + . 0. Neutral-current weak transitions with the expected strength have been discovered at CERN (Hasert et al. The prime superscript indicates the existence of mixing of the quark ﬁelds observed in strong interactions (mass eigenstates): d = d cos C + s sin C where the angle C is known as the Cabibbo angle and has been measured in weak decays of strange particles. T − ] = 2T 3 (0. such as. they (i) are not of purely V–A character as expected from the SU (2) model. for example.165 × 10−5 GeV−2 (h = c = 1) and ¯ the charged current Jµ (x) is composed of several pieces. Firstly. we postulate the existence of a symmetry group. According to the existing experimental limits the strangeness non-conserving ¯ neutral-current transitions (like K0 → µ+ µ− or K0 ↔ K0 ) are suppressed by many L orders of magnitude as compared to the standard weak processes.6) which would induce transitions. Having the current (0. global for the time being.

A highly consistent scheme begins to be expected. .0. Iliopoulos & Maiani (1970) have discovered that the problem of the strangeness non-conserving neutral current is solved if the set of fermionic doublets f L is completed with a fourth one c s .† has emerged as an important property of weak interactions. call for further invention in searching for a realistic theory of weak interactions. Glashow 1961. Augustin et al. Also the doublet classiﬁcation of the left-handed fermions with an equal number of lepton and quark doublets. s = −d sin C + s cos C One can immediately check that with the s orthogonal to d the neutral current is diagonal in ﬂavour. This has profound implications for a successful extension of the effective model into a nonabelian local gauge ﬁeld theory: it ensures the cancellation of the chiral anomaly and consequently the renormalizability of the theory. 1974). Some neutral-current weak transitions. To achieve this in the most economical way we notice that for the left-handed doublets of fermions we can † The t quark was discovered by the CDF and D0 groups in Fermilab in 1994.2. Thus. Thus. they have predicted the existence of the charm quark discovered later at SLAC (Aubert et al. we want the electric charge Q also to be the generator of the symmetry group of our theory. The next step towards the ﬁnal form of the Glashow–Salam–Weinberg theory is a ‘uniﬁcation’ of weak and electromagnetic interactions (Schwinger 1957. Salam & Ward 1964). Glashow.2 Reasons for gauge theories 7 Fig. 1974. 0. now further conﬁrmed by the experimental discovery of the ντ τ and t b doublets.

15) below) and gives the W 3 –γ mixing in terms of g and g.7) such that each doublet of the left-handed fermions is an eigenvector of the operator Y (e.8) where L(R) are left. Therefore Y commutes with the generators of SU (2) and 2 including the right-handed fermions by the prescription Y = Q (they are singlets with respect to SU (2)) we arrive at the SU (2) × U (1) symmetry group for the electroweak interactions.(right-)handed chiral lepton and quark ﬁelds.11) couple directly to the experimentally observed weak currents. YνL = YeL = − 1 ). As follows from the formula 1 1 Q = (gT 3 ) + (g Y ) (0. the electromagnetic current can be written as follows µ µ Jem = JY + f f ¯ L γ µT 3 f f f L = Qγ µ ¯ f (0. The U (1) current reads µ JY = f f f ¯ L γ µY f L + f f ¯ R γ µY f R (0. which as we will see. With electromagnetism being described by a gauge ﬁeld Aµ our minimal model ‘unifying’ electromagnetic and weak interactions requires the α Yang–Mills gauge ﬁelds Wµ and Bµ to couple to the SU (2) ×U (1) currents giving the interaction µ α g Jµ W αµ + g JY Bµ (0. Therefore the signiﬁcance of the electroweak uniﬁcation comes from the choice of a minimal group structure which takes account of both currents. The charged gauge bosons 1 ± 1 2 Wµ = √ (Wµ ± iWµ ) 2 (0. Even with two independent coupling constants. reduces the number of free parameters of the theory. According to (0.g.8 0 Introduction deﬁne a new quantum number Y (weak hypercharge) Y = Q − T3 (0.12) g g . this ‘uniﬁcation’ has a predictive power: it determines the structure of the parity violation in the neutral weak current (see (0.9) The SU (2) × U (1) group is the minimal one which contains the electromagnetic and weak currents.10) Since we are dealing with a direct group product the couplings g and g are independent.7).

Then the boson masses are (at the tree level) MW = 1 vg. although less divergent than a theory with arbitrary couplings.2 Reasons for gauge theories 9 the photon ﬁeld Aµ (determined by the fact that it couples to the electromagnetic current) must be a combination of the W 3 and B bosons Aµ = The orthogonal combination Zµ = 1 1 + 2 2 g g −1/2 1 1 + 2 2 g g −1/2 1 1 3 Wµ + Bµ g g (0.16) (the last relation follows from the fact that the electromagnetic coupling constant is electric charge e.0.9) and we have introduced the standard notation in terms of the so-called Weinberg angle (Glashow 1961. apart from the fermion mass matrix and the scalar self-coupling we have a three-parameter (g. we must now introduce masses for intermediate vector boson ﬁelds W ± and Z to account for the weakness of the weak interactions as compared to electromagnetism.17) Thus. is not renormalizable unless the vector boson masses are introduced by means of the Higgs mechanism which breaks the non-abelian gauge invariance spontaneously (’t Hooft 1971a.15) µ where Jem is given by (0. Salam 1968) with one complex SU (2) doublet of scalar ﬁelds whose neutral component has a non-vanishing vacuum expectation value v. One knows that the massive Yang–Mills theory.13) 1 3 1 W − Bµ g µ g (0. e > 0). For instance. If initially all vector ﬁelds are massless. In the minimal model this is achieved (Weinberg 1967b. 2 so that MW /M Z cos W g2 g2 + g 2 or tan W = g g or e = g sin W = g cos W M Z = 1 v(g 2 + g 2 )1/2 2 =1 (0. the Fermi constant G µ and the Weinberg angle (parity non-conservation in all neutral-current transitions is consistent with the . g . b). knowing the electromagnetic coupling α = e2 /4π .14) z couples to the neutral weak current Jµ z Jµ = g cos W 3 (Jµ − sin2 em W Jµ ) (0. Weinberg 1967b) sin2 W = (0. v) electroweak theory.

They are also extensively studied in the hope of providing the uniﬁed description of all interactions (grand uniﬁcation. at the tree level. Non-abelian gauge theories are considered at present to be the most promising theoretical framework for fundamental interactions.15) and with sin2 ≈ 0. supergravity) going beyond the so-called standard SU (3) × SU (2) × U (1) model summarized in this section. apparently coupled to each other. We would like to stress the role of gauge symmetry and the uniﬁcation idea.23 one predicts. However. in this very successful description of the electroweak interactions.10 0 Introduction W form (0. W MW = M Z cos = πα √ 2G µ 1/2 1 sin W ≈ 80 GeV (0. the discovery of the Higgs particle still remains (in 1999) a challenge for future experiments.18) Vector bosons with the expected masses were discovered at CERN in 1982! This triumph of the spontaneously broken Yang–Mills theory has since been conﬁrmed by comparing higher accuracy data with higher order calculations. .

Quantization converts classical mechanics into quantum mechanics which describes the behaviour of particles at the quantum level. This logical structure of theories for fundamental interactions is illustrated by the diagram shown below: PARTICLES Classical mechanics (quantization) (wave-function (x) is interpreted as a physical ﬁeld and quantized) FORCES Classical ﬁeld theory (quantization) Quantum mechanics QUANTUM FIELD THEORY This chapter is devoted to a brief summary of classical ﬁeld theory. e. symmetries and their breaking Classically. The motion of particles in force ﬁelds is subject to the laws of classical mechanics. A state of a particle is described by a vector | (t) in the Hilbert space or by its concrete representation. The forces are described by classical ﬁelds. The reader should not be surprised by such formulations as. for example. the ‘classical’ Dirac ﬁeld which describes the spin 1 particle. we distinguish particles and forces which are responsible for interaction between particles. It is interpreted as a wave function for the 2 Dirac particle. such as electromagnetic forces. We then arrive at quantum ﬁeld theory as the universal physical scheme for fundamental interactions. x)) whose modulus squared is interpreted as the density of probability of ﬁnding the particle in point x at the time t.1 Classical ﬁelds. the wave-function (x) = x| (t) (x = (t. It is also reached by quantization of ﬁelds describing classical forces. The basic physical concept which underlies quantum ﬁeld theory is the equivalence of particles and forces. As a next step.g. the wave-function is interpreted as a physical ﬁeld and quantized. Our ultimate goal is quantum ﬁeld theory and our classical ﬁelds in 11 .

we always assume implicitly that real physical systems are described by wave packets localized in space. From the action we get: equations of motion (invoking Hamilton’s principle). (iv) S has to be real to account for the absence of absorption in classical physics and for conservation of probability in quantum physics. ∂µ (x) (1. 1. Such terms can usually be interpreted as a perturbation. It is our experience so far that in physically relevant theories the action satisﬁes several general principles such as: (i) Poincar´ invariance e (or general covariance for theories which take gravity into consideration). x) to the ﬁnal state (t1 . In general.12 1 Classical ﬁelds. .1 The action. the ﬁelds are multiplets under Lorentz transformations and in a space of internal degrees of freedom. conservations laws (from Noether’s theorem).1) (the volume V may be ﬁnite or inﬁnite. The action is then the most general functional which satisﬁes the above constraints. Path integrals are discussed in detail in Chapter 2. In classical ﬁeld theory the lagrangian density L is a function of ﬁelds and their derivatives. in the latter case we assume the system to be localized in space‡) are obtained from Hamilton’s principle. In this section we brieﬂy recall the ﬁrst two results. transition from classical to quantum physics (by using path integrals or canonical quantization). ‡ Although we shall often use plane wave solutions to the equations of motion. symmetries and their breaking this chapter are not necessarily only the ﬁelds which describe the classical forces observed in Nature. An ansatz for the action S = dt L = d4 x L can be regarded as a formulation of a theory. bilinearity in the derivatives ∂ν (x) (to get at most second order differential equations of motion). This ansatz (whose physical sense becomes clearer when classical theory is considered as a limit of quantum theory formulated in terms of path integrals) says: the dynamics of the system evolving from the initial state (t0 . (ii) locality of L and its. equations of motion. symmetries and conservation laws Equations of motion All fundamental laws of physics can be understood in terms of a mathematical construct: the action. The equations of motion for a system described by the action S= t1 dt t0 V d3 x L (x). at most.† (iii) invariance under all symmetry transformations which characterize the considered physical system. x) is † Later we shall encounter effective lagrangians that may contain terms with more derivatives.

3) which vanish on the boundary of ≡ ( t. 1.8) give the same classical equations of motion (due to the vanishing of variations of ﬁelds on the boundary of ). .4) (1.1 Action.1. . conformal transformations are discussed in . V ). 3 i = 1. is are internal quantum number indices). with taken as a Lorentz vector. Since in the variation the coordinates x do not change we have: δ(∂ν ) = ∂ν δ and (1. Let us consider a Lie group of continuous global (x-independent) inﬁnitesimal transformations (here we restrict ourselves to unitary transformations. equations of motion.5) The second term in (1.e.6) (1. the condition δS = 0 gives us the Euler–Lagrange equations of motion for the classical ﬁelds: ∂L ∂L =0 µ − ∂ν ∂ i ∂(∂ν iµ ) ν. δS = 0 for arbitrary variations (x) → (x) + δ (x) (1. µ = 0. 2.6) is a surface term which vanishes and. (1. .4) can be rewritten as follows: δS = d4 x ∂L ∂L − ∂ν δ ∂ ∂(∂ν ) + ∂ν ∂L δ ∂(∂ν ) (1. Global symmetries Eq. symmetries and conservation 13 such that the action (taken as a functional of the ﬁelds and their ﬁrst derivatives) remains stationary during the evolution. . i.7) (here we keep the indices explicitly. therefore. It is important to notice that lagrangian densities which differ from each other by a total derivative of an arbitrary function of ﬁelds L = L + ∂µ µ ( ) (1.2) (summation over all ﬁelds and for each ﬁeld over its Lorentz and ‘internal’ indices is always understood).6) can also be used to derive conservation laws which follow from a certain class of symmetries of the physical system. Explicit calculation of the variation δS gives: δS = d4 x ∂L δ ∂ + ∂L δ(∂ν ) ∂(∂ν ) (1. n (1.

in general are different in form from the original ones. ∂µ → Tr(T a T b ) = 1 δ ab 2 (1. ∂µ (x)) = L( (x). Note that a rotation of the reference frame by angle corresponds to active rotation by (− ).14 1 Classical ﬁelds. variation of the action generated by arbitrary variations of the ﬁelds (x) with boundary conditions (1.10) and i (x) is understood as the ith component of the ﬁeld (x) in the new reference frame or of the rotated ﬁeld in the original frame (active).9) on the ﬁelds to distinguish it from the variations δS and δS).) . ∂µ (x) (1. For the variations we have δS = δS. of course. symmetries and their breaking Chapter 7) acting in the space of internal degrees of freedom of the ﬁelds i .11) (x) given by (1.9) (for scale transformations see Chapter 7): L( or. ∂µ ) = 0 (1.9) the lagrangian density (x). ∂µ (x)).14) Indeed.6) (with → etc. Thus. Under the change of variables is transformed into L (x). the action remains numerically unchanged: δ S=S −S= d4 x L ( . The T a s are a set of hermitean matrices Tia satisfying the Lie algebra of the group G j T a . This is ensured if the density L is invariant under transformations (1. ∂µ (x)) = L( (x). ∂µ (x)) (1. if (x) describes a motion of a system. Transformations (1. (x). (x) is also a solution of the equations of motion for the transformed ﬁelds which. equivalently. a solution to the equations of motion after being transformed according to (1.3) is again given by (1.9) are symmetry transformations for a physical system if its equations of motion remain form-invariant in the transformed ﬁelds. however.9) where δ0 i (x) = −i a Tia j (1. ∂µ ) − L( .9) remains a solution of the same equations.13) (we use the symbol δ S for the change of the action under transformation (1.12) (x) ≡ L and. L( (x). In other words. T b = icabc T c . and the a s are x-independent. Under inﬁnitesimal rotation of the reference frame (passive view) or of the physical system (active view) in that space: i (x) → i (x) = i (x) + δ0 j (x) i (x) (1.

x) (1.17) can be written as Q a (t) = d3 x (t. x)(−iT a ) (t. Moreover. F2 (t. x) (1. Firstly.1.14) of the lagrangian density under the symmetry transformations: L( . ∂µ ) = L( . It follows from (1.15) and the equations of motion that the currents a jµ (x) = −i ∂L Ta ∂(∂ µ i ) i j j (1. Again.1 Action. y)} = −δ(x − y) { (t.17) (even if they are not conserved) are the generators of the transformation (1. { (t. such as. classical electrodynamics. ∂µ ) − L( . It is very important to notice that the charges deﬁned by (1. x). z) We get. H= ∂0 −L (1. x). they indeed generate transformations (1. y) {F1 (t. x). Secondly.16) are conserved and the charges Q a (t) = a d3 x j0 (t. x) ∂ F2 (t. the Lorentz and ‘internal space’ indices of the s and s are hidden. z) ∂ (t. (t. equations of motion. ). y)} = { (t. (t.17) are constants of motion.22) . they satisfy (in an obvious way) the same commutation relation as the matrices T a .21) The charge (1.10). x). (t.9). y)} = 0 (1. provided the currents fall off sufﬁciently rapidly at the space boundary of . y) ∂ F1 (t. these have to be discussed separately).6) with δ s given by (1. The hamiltonian of the system reads H= d3 x H( . x) ∂ F2 (t. ∂µ ) = 0 (1. there are interesting exceptions.18) (we assume here that = 0. for example. the change (1. in particular.15) is formally given by the integrand in (1.10) through Poisson brackets. Let us introduce conjugate momenta for the ﬁelds (x): (t. x) = ∂L ∂(∂0 (t. z) ∂ (t. z) ∂ (t. x)) (1. symmetries and conservation 15 and we derive the same equations for (x) as for (x). with = 0. ∂µ ) − L( . y)} = d3 z − (1.20) ∂ (t.19) and the Poisson bracket of two functionals F1 and F2 of the ﬁelds and is deﬁned as ∂ F1 (t.

∂µ (x) . In this case our formalism has to be slightly generalized. indeed the transformation (1. ∂µ (x ) = L (x). Thus. Moreover. we consider transformations (changes of reference frame) which act simultaneously on the coordinates and the ﬁelds.e.26) (x ) = ∂µ The lagrangian density L after the transformation is deﬁned by the equation (see. i. (t. for example. By this we mean that a change of the reference frame has no implications for the motion of the system. We consider an inﬁnitesimal transformation x µ = x µ + δx µ = x µ + εµ (x) (1. translations. x) i.27) implies δS = δS).24) (which acts on their Lorentz structure) into (x ) in the transformed frame: (x ) = exp[−iT (ε)] (x) (1.10).e.23) Q a (t). generators of symmetry transformations are conserved charges and vice versa. a transformed solution to the original equations remains a solution to the new equations (since (1. Trautman (1962.16 1 Classical ﬁelds. The well-known examples are.24) and δ µ is an arbitrary function of the ﬁelds) which is a sufﬁcient condition for the new equations of motion to be equivalent to the old ones. Space-time transformations Finally. 1996)) S = = d4 x L d4 x L (x ). there is an arbitrariness in . for instance. spatial rotations and Lorentz boosts.27) ( denotes the image of under the transformation (1. ∂µ (x ) µ (x). ∂µ (x) − ∂µ δ ( (x)) (1. In general det(∂ x /∂ x) = 1 and also L (x).24) and suppose that the ﬁelds (x) seen in the ﬁrst frame transform under some representation T (ε) of the transformation (1.25) (the Lorentz indices are implicit and α (x ) is understood as the αth component of the ﬁeld (x ) in the new reference frame). For inﬁnitesimal transformations we have (∂ denotes differentiation with respect to x ): (x ) = ∂µ (x) + δx µ ∂µ (x) + O(ε 2 ) (x) + δx ν ∂µ ∂ν (x) + O(ε 2 ) (1. symmetries and their breaking and for its Poisson bracket with the ﬁeld we get: (1. x) = (iT a ) (t.

we recover condition (1. The most famous example of symmetry transformations up to a non-vanishing total derivative is supersymmetry (see Chapter 15). the L deﬁned by (1.1 Action. symmetries and conservation 17 the choice of L due to the presence in (1. Using (1. therefore. ∂µ (x )) = L( (x ).31) + ∂L δ0 (∂µ ) + δx µ ∂µ L + O(ε 2 ) ∂(∂µ ) (1.32) Eq. ∂µ (x)) − ∂µ δ µ (x ).24). A sufﬁcient condition is that. since δ0 is a functional change. ∂µ (x )) (1.30) and.33) we ﬁnally get (using (1. if det(∂ x /∂ x) = 1. (1. ∂µ (x)) (1. equations of motion.30) can be rewritten in the following form: δL = δx µ ∂µ L + Since det ∂ xµ ∂ xν = 1 + ∂µ δx µ (1.34) ∂L ∂L − ∂µ δ0 ∂ ∂(∂µ ) + ∂µ (1.35) + O(ε 2 ) . for a certain choice of δ .14) up to the total derivative.27) of the total derivative of an arbitrary function δ µ ( (x)). equivalently.29) where x and x are connected by the transformation (1. δ0 ∂µ = ∂µ δ0 ∂L δ0 ∂(∂µ ) (1.28a). L( (x ).33) and the equations of motion): 0 = L∂µ δx µ + δx µ ∂µ L + ∂µ = Lδx µ + ∂L δ0 ∂(∂µ ) +δ ∂L δ0 ∂(∂µ ) µ + ∂µ δ µ + O(ε 2 ) (1. Symmetry transformations are again deﬁned as transformations which leave equations of motion form-invariant.26) we get δL = where δ0 (x) = (x) − (x) (1.1.27) satisﬁes the equation: L( or. The change of the action under symmetry transformations can be calculated in terms of δL = L( ∂L δ0 ∂ (x ). ∂µ (x )) − L( (x). (1.28) ( (x)) d4 x (1.28a) Note that. ∂µ (x )) d4 x = L( (x).

2. It is easy to check that transformations of the ﬁeld (x) under translations in time and space are given by {P µ . (x)} εµ . The energy–momentum tensor deﬁned by (1. note also that det(∂ x /∂ x) = 1).42) where f µν (x) are arbitrary functions satisfying the conservation law ∂µ f µν (x) = 0 (1. symmetries and their breaking Re-expressing δ0 that ∂µ j µ ≡ ∂µ in terms of δ Lg µ ρ − = (x )− (x) = δ0 +δx µ ∂µ δx ρ + ∂L δ ∂(∂µ ) +δ µ we conclude =0 ∂L ∂ρ ∂(∂µ ) (1.40) are the total energy of the system (ν = 0) and its momentum vector (ν = 1. 3). Examples We now consider several examples. translations Invariance of a physical system under (1.18 1 Classical ﬁelds.27) with ≡ 0 satisﬁes L (x ) = L(x ).39) is the so-called canonical energy–momentum tensor.38) is the energy–momentum tensor µν (x) = ∂L ∂ν ∂(∂µ ) − g µν L (1. It is important to notice that the physical interpretation of the energy–momentum tensor remains unchanged under the redeﬁnition µν (x) → µν (x) + f µν (x) (1.16) to space-time symmetries and both results are the content of the Noether’s theorem which states that symmetries of a physical system imply conservation laws.36) This is a generalization of the result (1.43) . The four constants of motion Pν = d3 x 0ν (x) (1.41) so that (x) = (x) + {P µ .37) x µ = x µ + εµ implies the conservation law ∂µ where µν µν (x) = 0 (1.39) (we have assumed that the considered system is such that L (x ) deﬁned by (1. (x)} = − ∂ (x) ∂ xµ (1.

2 The tensor M µν is not translationally invariant. This freedom can be used to make the energy–momentum tensor symmetric and gauge-invariant (see Problem 1.47) where µν is a spin matrix. (x)}.50) µρ µρ − xρ µν (1. The total angular momentum three-vector is obtained by constructing the Pauli–Luba´ ski vector n Wµ = 1 εµνρκ M νρ P κ 2 (1.νρ (x) = x ν for a ﬁeld with a non-zero spin: M µ. x) (1. For Lorentz scalars = 0.46) and 4 (1.38) and (1. symmetries and conservation 19 with vanishing charges d3 x f 0ν (x) = 0 Then (1. Inserting (1.νρ (t.48) − xρ µν + ∂L ∂(∂µ ) νρ (1.45) (1. An inﬁnitesimal Lorentz transformation reads xµ≡ µ ν ν (ω)x ≈ x µ + ωµν x ν µν (1.47) into (1. A solution to these constraints is f µν (x) = ∂ρ f µνρ (1. we get the following conserved currents: for a scalar ﬁeld: M µ.3).νρ (x) = x ν The conserved charges M νρ (t) = d3 x M 0.49) are identiﬁed with the total angular momentum tensor of the considered physical system. The charges M µν are the generators of Lorentz transformations on the ﬁelds (x) with {M µν .40) remain unchanged.36) (using (1. (x)} = −(x µ ∂ ν − x ν ∂ µ + µν ) (x) (1.46) (with ωµν antisymmetric) and it is accompanied by the transformation on the ﬁelds (x ) = exp 1 ωµν 2 µν (x) ≈ 1 + 1 ωµν 2 (x) (1.44) where f µνρ (x) is antisymmetric in indices (µρ).1 Action. equations of motion. γ ν ] (for the notation see Appendix A).51) so that (x) = (x) − 1 ωµν {M µν .1. for Dirac spinors µν = (−i/2)σ µν = 1 [γ µ .38) and replacing the canonical energy–momentum tensor by a symmetric one and ﬁnally using the antisymmetry of ωµν ).52) .

[ ] = g. 0) to the three-dimensional total angular momentum M k ≡ 1 εi jk M i j : M k = W k /m. the variations in and are independent): ( + m2) +λ + 1 ( m2 1 )2 + ··· = 0 (1. The conserved U (1) current jµ (x) = iq( ∂µ − ∂µ ) (1. 1 . We take these to be complex.20 1 Classical ﬁelds. −i χ=√ ( 2 − ) (1. In units c = h = 1 we have [L] = g4 .53) Thus. 2 1. unitary.55) 2 2 3m 1 where T is the kinetic energy density and V is the potential energy density. unimodular transformations): (x) = exp(−iq ) (x) ≈ (1 − iq ) (x) (1.2 Classical ﬁeld equations Scalar ﬁeld theory and spontaneous breaking of global symmetries In this section we illustrate our general considerations with the simplest example: the classical theory of scalar ﬁelds (x). It is sometimes also useful to introduce two real ﬁelds 1 φ=√ ( 2 + ). the action has the dimension [S] = g · cm = [h ]. [ ] = (g/cm)1/2 and ¯ the coefﬁcients [m 2 ] = cm−2 . [m 2 ] = g2 . symmetries and their breaking (where εµνρκ is the totally antisymmetric tensor deﬁned by ε0123 = −ε0123 = 1) which reduces in the rest frame P µ = (m.† Using the formalism of the previous section we derive the equation of motion (since we have two physical degrees of freedom.57) † In units c = 1.56) In the limit of vanishing couplings (free ﬁelds) we get the Klein–Gordon equation.54) The lagrangian density which satisﬁes all the constraints of the previous section reads: 1 1 L = ∂µ ∂ µ − m 2 − λ( )2 − ( )3 + · · · = T − V (1. [λ] = (g · cm)−1 . The terms which are higher than second powers of the bilinear combination require on dimensional grounds couplings with dimension of negative powers of mass. in order to be able to discuss a theory which is invariant under continuous global symmetry of phase transformations (U (1) symmetry group of one-parameter. ¯ 1 [m 2 ] = [m 2 ] = g2 and S and λ are dimensionless. the ﬁeld (x) (its complex conjugate (x)) carries the internal quantum number q (−q). Thus [L] = g · cm−3 .

The spin–spin interactions between neighbouring 2 dipoles cause them to align. Intuitively speaking. This tensor can be written in other forms.16) and (1. We say that the symmetry G is spontaneously broken if the ground state (usually called the vacuum) of a physical system with symmetry G (i. Thus. Let us return to the question of spontaneous breaking of the U (1) symmetry in the theory deﬁned by the lagrangian (1.55). by adding the term ∂ρ (gµν ∂ ρ − gρν ∂ µ ) and using the equation of motion. for m 2 < 0 (and m 2 > 0. The hamiltonian is rotationally invariant but the ground state is not: it is a state in which all the dipoles are aligned in some arbitrary direction. respectively. described by a lagrangian which is symmetric under G) is not invariant under the transformations G.1.1. The phenomenon of spontaneous breaking of global symmetries is well known in condensed matter physics. it can be spontaneously broken.2 Classical ﬁeld equations 21 and the canonical energy–momentum tensor µν = ∂ µ ∂ν + gµν m 2 + ∂ν ∂µ + − gµν (∂ρ ∂ ρ ) + ··· 2 ) (1. The minimum of the potential occurs for the classical ﬁeld conﬁgurations such that ∂V ∂V = =0 ∂ ∂ (1. spontaneous symmetry breaking can be realized only in the presence of scalar ﬁelds (fundamental or composite). So for an inﬁnite ferromagnet there is an inﬁnite degeneracy of the vacuum. Since we do not want spontaneously to break the Lorentz invariance.60) with V ( 0 ) = −(m 4 /2λ).59) For m 2 > 0 the minimum of the potential exists for = 0 = 0 and V ( 0 ) = 0. study the potential for m 2 < 0!) the solutions 1 1 to these equations are (note that λ must be positive for the potential to be bounded from below) 0 ( ) = exp(i ) −m 2 λ 1/2 1+O m2 λ2 m 2 1 (1. for example. The standard example of a physical theory with spontaneous symmetry breakdown is the Heisenberg ferromagnet. the vacuum is ﬁlled with scalars (with zero four-momenta) carrying the quantum number (s) of the broken symmetry.33). However. the U (1) symmetric state is no longer the ground state of this .e. Invariance under Lorentz transformation gives the conserved current (1. i.e. The U (1) symmetry of our scalar ﬁeld theory can be realized in the so-called Wigner mode or in the Goldstone–Nambu mode.58) 1 λgµν ( 2 are obtained from (1. The dependence of the potential on the ﬁeld is shown in Fig.48). 1. an inﬁnite array of spin 1 magnetic dipoles.

φ(x) and χ(x) (see (1. for example.22 1 Classical ﬁelds.55) around the true ground state. transforming. In this example we can choose any one of these vacua as the ground state of our theory with spontaneously broken U (1) symmetry. Spinor ﬁelds The basic objects which describe spin 1 particles are two-component anticom2 muting Weyl spinors λα and χ α . as two inequivalent 2 2 representations of the S L(2. 1 ) of the Lorentz group or. rewrite it in terms of the ﬁelds (we choose = 0 and put m 2 = ∞): 1 ˜ (x) = (x) − 0 . symmetries and their breaking Fig. C) group of complex two-dimensional matrices M . the Poisson brackets (or the commutators) of a charge with a ﬁeld can be deﬁned provided we integrate over the space after performing the operation at the level of the current (see. the potential has its minimum for an inﬁnite. have the mass parameters (−m 2 ) ˜ (i. Instead. theory. positive) and zero. Still.60) requires massless excitations to transform one vacuum into another one. as representations ¯˙ ( 1 . in other words. i. Physical interpretation of this model (and of the negative mass parameter m 2 ) is obtained if we expand the lagrangian (1. Bernstein (1974)). 0) and (0. respectively (see Chapter 11 for more details). the integral (1.17) taken over all space does not exist. 1. After a short calculation we see that the two dynamical ˜ degrees of freedom.e. respectively.54)). The χ (x) describes the so-called Goldstone boson – a massless mode whose presence ˜ is related to the spontaneous breaking of a continuous global symmetry.e. In theories with spontaneously broken symmetries the currents remain conserved but the charges corresponding to the broken generators of the symmetry group are only formal in the sense that. degenerate set of states (one often calls it a ‘ﬂat direction’) connected to each other by the U (1) rotations. We shall return to this subject later (Chapter 9).1. generically. The degenerate set of vacua (1.

65) where the two terms are equal up to a total derivative.61) and (1. Its complex 2 ¯˙ ¯ ˙ ˙˙ ¯˙ conjugates are λα ≡ (λα ) (see the text above (1. 1 ) representation.and four-dimensional matrices σ µν since it should not lead to any confusion in every concrete context.32) with the matrix M in (A. 0) and (0.64)) and λα = λβ ε β α which ˙˙ transforms as (0.1. We use the same symbol for two. Those unfamiliar with two-component spinors and their properties should read Appendix A before starting this subsection. C) can now be constructed as antisymmetric products ( 1 . for example. ε αβ λβ (σ µ )αβ χ β = λα (σ µ )αβ χ β or ˙ ¯ ˙ ¯ 2 2 χα (σ µ )αβ λβ (see Appendix A for the deﬁnitions of σ µ and σ µ ). 0) under the Lorentz group. Scalars of S L(2. Two other possible representations of the S L(2. 1 ) representations. M = exp −(i/4)ωµν σ µν (1. the structure ε αβ λβ φα = λα φα = 2 2 ˙ −λα φ α ≡ λ · φ or χ β ηα εβ α = χα ηα = −χ α ηα ≡ χ · η.61) and (1. for example. Thus. The distinction between the undotted and dotted indices is made to stress that the two representations are inequivalent.62) χ β (x) ¯ ˙ where x and x are connected by the transformation (1. (Note. ¯˙ ¯ ˙ ¯ We suppose now that a physical system can be described by a Weyl ﬁeld (or a set of Weyl ﬁelds) λα which transforms as ( 1 . The choice of (1. 0) and (0.62).) 2 It is clear that S L(2.46).2 Classical ﬁeld equations 23 of determinant 1: λα (x ) = Mα β λβ (x) χ α (x ) = M †−1 ¯ ˙ α ˙ ˙ β (1. χ α = χβ ε β α ¯˙ ¯˙ ˙ ˙ ˙ (1. the tensors ε αβ and ¯ ¯˙ ˙˙ ¯˙ ˙ ¯ ˙ ¯˙ ¯ ¯ αβ ˙˙ ε are metric tensors for the spinor representations.64) ˙ where the antisymmetric tensors εαβ and εαβ are deﬁned in Appendix A. respectively. Since transition from the representation (1. that numerically ε αβ = {εαβ } = iσ 2 . A four-vector is a product of ˙ ˙ ( 1 .61) to (1. 1 )⊗(0. 1 ).61) (1. The factor i has been .62) involves complex conjugation it is natural to denote one of them with a bar.62) as fundamental representations is dictated by our convention which identiﬁes (A.63) and the two-dimensional matrices σ µν are deﬁned in Appendix A. C)-invariant kinetic terms read ¯ ¯¯ L = iλσ µ ∂µ λ = iλσ µ ∂µ λ (1. The corresponding unitary transformations are given by: λα = ε αβ λβ . They have.16). 0) ⊗ 2 1 ( 2 . C) group denoted as λα and χα and transforming through matrices ¯˙ −1 T ∗ (M ) and M are unitarily equivalent to the representations (1. The pairwise equivalent representations correspond to lower and upper indices.

0) representations 2 of the Lorentz group but with λc transforming as the representation R . we remark that a system described by a single Weyl spinor has two degrees of freedom.67) We now consider the case of a system described by a set of Weyl spinors λiα which transform as a complex representation R (see Appendix E for the deﬁnition of complex and real representations) of a group of internal symmetry. form the representation R . 2 ) (or its unitary equivalent) under the Lorentz transformations. can be rotated away by the appropriate phase transformation. this is what we have to assume for these classical ﬁelds when we follow the path integral approach to ﬁeld quantization (see Chapter 2).e. the mass term −im (λλc − λλc ). 2 2 2 1/2 can be recast in the above form with m = (m 1 + m 2 ) by the transformation λ → exp(iϕ)λ with exp(2iϕ) = (m 1 − im 2 )/(m 2 + m 2 )1/2 . λiα . Then. Note that in this case the spinors λα (and λα ). symmetries and their breaking introduced for hermiticity of the lagrangian since one assumes that the spinor components are anticommuting c-numbers.68) ¯¯ Similarly to the case of Majorana mass. so both ﬁelds λ and λc enter the . one can construct the Lorentz ¯¯ ¯˙¯˙ scalars λλ = λα λα and λλ = λα λα and introduce the so-called Majorana mass term ¯¯ Lm = − 1 m λλ + λλ 2 (1.66) ¯¯ ¯¯ The most general hermitean combination. the kinetic lagrangian remains as in (1. The term Lm does not 1 2 vanish since λα s are anticommuting c-numbers. The mass term can then be introduced and the full lagrangian of such a system then reads: ¯¯ ¯ ¯¯ L = iλσ µ ∂µ λ + iλc σ µ ∂µ λc − m λλc + λλc (1. Finally. complex α conjugate to R. if present. Such theory is called vector-like. for example. Remember that ¯ ¯ ¯ λc σ µ ∂µ λc = λc σ µ ∂µ λc (up to a total derivative). complex conjugate to R. the spinors transforming as (0. Such theory is called chiral. if the ﬁeld λ carries no internal charges (or transforms as a real representation of an internal symmetry group). The operation of charge conjugation will be discussed in more detail in Section 1. i. and α the summation over them are always implicit).24 1 Classical ﬁelds.e. Furthermore.65) but no invariant ¯ ¯ mass term can be constructed since Lorentz invariants λi λ j and λi λ j are not ¯˙ ¯˙ invariant under the symmetry group. − 1 m 1 (λλ + λλ) − (i/2)m 2 (λλ − λλ).5. 1 i. We call such spinors charge conjugate to each other. both being ( 1 . Indeed. with the equation of motion obtained from the principle of minimal action (variations with respect to ¯ λ and λ are independent): iσ µ ∂µ λ ¯ α ˙ ¯˙ = m λα (1. There are also interesting physical systems that consist of pairs of independent sets of Weyl ﬁelds λα and λc (the internal symmetry indices i.

the mass term mixes positive energy solutions for λ with ¯α negative ones for λc (which are represented by positive energy solutions for λc ) and ˙ vice versa. 0) representations of S L(2. describe massless helicity + 1 states transforming as R . 2 ) representations of S L(2. C) and as R under the internal symmetry ¯ ˙ group can be given. the so-called Dirac mass. C). therefore. λiα s are called left-handed chiral ﬁelds. the left-handed i λα s and λci s. ± 2 . Negative energy solutions for λiα s and λci s describe. but transforms as a complex representation R under the full SU (2) × U (1) group. of course. this set forms a real representation of the electromagnetic 3 3 U (1) gauge group. In both cases. for example. a 2 2 matter of convention. i. In the second case (QCD) the fundamental ﬁelds of the theory. in this case the set λiα includes pairs of left-handed particles with opposite electric charges such as.) ¯ ˙ An equivalent description of the same theory in terms of the ﬁelds λi α transform1 ing as (0. α . complex representation R of the symmetry group G = SU (2) × U (1) (R = R ). transform as 3 and 3 of SU (3). 2 In the electroweak theory (see Chapter 12) all the left-handed physical ﬁelds are included in the set of λiα s which transforms then as a reducible.1. the electroweak theory is chiral and quantum chromodynamics is vector-like. Positive energy solutions for λiα s are identiﬁed with − 1 helicity states 2 of massless quarks and positive energy solutions for λci s describe − 1 helicity α 2 states of massless antiquarks. Thus. in the Dirac sea interpretation. In this formulation positive (negative) energy solutions for λi α s describe all + 1 (− 1 ) helicity massless states of the theory (formerly described as 2 2 negative (positive) energy solutions for λiα s). this can be interpreted as a mixing between different helicity states of the same massless particle.2 Classical ﬁeld equations 25 lagrangian in a fully symmetric way.68) are: iσ µ ∂µ λ ¯ α ˙ ¯α = m λc ˙ (1. e− . (Whether within a given reducible representation contained in R the − 1 or + 1 helicity states are called particles is. The equations of motion which follow from (1. These pairs consist of the states with opposite helicities. Each 2 pair of states described by λiα is connected by the C P transformation (see later) and can therefore be termed particle and antiparticle. The mass term.69) and similarly for λc . Negative energy classical solutions for λiα s. To be more speciﬁc. respectively. As we shall discuss in more detail later. is now invariant under the internal symmetry group. and are related by charge α conjugation.e. In accord with the discussion of Appendix A. e+ and quarks with electric charges ∓ 1 . it is convenient to take as the fundamental ﬁelds of the theory Weyl spinors which transform as ( 1 . We recall that (as discussed in Appendix A) 2 positive energy classical solutions for spinors λiα s describe massless fermion states with helicity − 1 and.

−im ¯ γ5 . λα ) to recover the mass term of (1. The invariants λα λα and λα λα can T T be written as L C L and ¯ L C ¯ L . for a set of Weyl spinors λα and their complex conjugate .68) and (1. 0) of the Lorentz group but as R and R 2 representations of an internal symmetry group. in fact. respectively.68) can be rewritten as L = i ¯ γ µ ∂µ −m ¯ (1.68) λ and λc can be interpreted as describing opposite helicity (chirality) states of the same massless particle which mix with each other due to the mass term) = λα ¯ cα λ ˙ or c = λc α ¯˙ λα (1.72) They satisfy the equations L = R 1 − γ5 2 1 + γ5 = 2 L ≡ PL ≡ PR L. it is convenient to use the Dirac ¯ four-component spinors (in the presence of the mass term in (1. R. antiquarks and quarks with helicity + 1 . respectively. Another possible hermitean mass term. By analogy with the nomenclature introduced for λα and χ α the ﬁelds L and R are called left¯˙ and right-handed chiral ﬁelds. is quite convenient for Weyl spinors λα and χ α transforming as ( 1 . both transforming as ( 1 .71) ¯˙ where we have introduced ¯ ≡ (λcα .68).71) we deﬁne Writing the matrices γ µ . Four-component notation can also be introduced and. Also in Appendix A we derive the Lorentz transformations for the Dirac spinors. ¯ = † γ 0 and comparing the kinetic terms of (1. symmetries and their breaking respectively. They are called Dirac matrices in the chiral representation and are given in Appendix A. Since the representation R is 2 real.73) R R which deﬁne the matrix γ5 in the chiral representation. Possible Lorentz invariants are of the ¯˙ ¯˙ ¯˙¯˙ form ¯ R L ≡ χ α λα and ¯ L R ≡ λα χ α .104). with charge conjugate pairs of Weyl spinors λ and λc .26 1 Classical ﬁelds. In particular. can be rotated away by the appropriate phase transformation of the spinors λ and λc . like QCD. For a vector-like theory. The lagrangian (1. respectively. under the group of internal symmetries. 1 ) under the Lorentz ¯˙ 2 2 group.70) transforming as R and R . with the matrix C deﬁned in (A. 0) and (0. We deﬁne four-component chiral fermion ﬁelds as follows L = λα 0 R = 0 χα ¯˙ =0 =0 (1. PR PL L (1. mass terms are possible.

71) but no mass term is allowed.78) for the Dirac spinors and with m = 0 for the chiral spinors.65) and (1.70). to scalar ﬁelds transforming properly under this group (see later). For a vector-like spectrum we have ci ci ) . In the latter case it is supplemented by the conditions (1.70) the spinors in (1. the sets† i L = λiα 0 ci R = 0 ¯ iα λ˙ (1.2 Classical ﬁeld equations 27 ¯˙ λα . under the internal symmetry group. for objects represented by a Weyl ﬁeld λα with no internal charges or transforming as a real representation of an internal symmetry group we can introduce the so-called Majorana spinor M = λα ¯˙ λα (1. γ µ = 0 (1. The kinetic part of the lagrangian for the chiral ﬁelds is the same as in (1.74) has been introduced by analogy with (1. cj j i ci Lorentz invariants L R and R L (or a combination of them) are generically not invariant under the internal symmetry group but such invariants can be constructed by coupling them. respectively. The solutions to the equations of motion for the Weyl spinors and for the four-component spinors in the chiral and Dirac representations for the γ µ matrices are given in Appendix A. in ordinary QED if L has a charge −1 (+1) under the U (1) symmetry group.76) In this notation the lagrangian given by (1. transforming as R and R . Thus. we can identify it with the c left-handed chiral electron (positron) ﬁeld. We note that γ 5 = γ5 = iγ 0 γ 1 γ 2 γ 3 and that the following algebra is satisﬁed: {γ µ . unlike the spinors in (1. Note.74) also transform as R and R .74) are deﬁned for both chiral and vector-like spectra of the fundamental ﬁelds. For instance. for example.75) Finally. † The notation in (1.77) The equations of motion for free four-dimensional spinors have the form iγ µ ∂µ − m =0 (1. respectively.66) reads LM = i ¯ M γ µ ∂µ 2 M − m ¯M 2 M (1. R R =( . γ ν } = 2g µν . γ 5.73). however.1. The R is then the chiral right-handed positron (electron) ﬁeld. that the superscript ‘c’ ci ¯˙ in the spinors R is now used even though λα and λα are not charge conjugate to each other (since they are in different Lorentz group representations).

. for instance. ¯¯ λλφ (1. symmetries and their breaking A few more remarks on the free Dirac spinors should be added here. . for a neutral (with respect to the internal quantum numbers) Weyl spinor λ interacting with a complex scalar ﬁeld φ (also neutral) the dimension four ([m]4 in units h = c = 1) Yukawa interaction terms are: ¯ λλφ.85) ¯¯ The absence of otherwise perfectly allowed coupling −g (φ λλ + φ λλ) is due ¯¯ to supersymmetry. χ. These may be internal symmetries or. would further constrain the allowed terms. . The simplest supersymmetric theory. describing interactions of one complex scalar ﬁeld φ(x) with one chiral fermion λ(x) has the following lagrangian† ¯¯ ¯¯ LWZ = iλσ · ∂λ − 1 m(λλ + λλ) + ∂µ φ ∂ µ φ − m 2 φ φ 2 ¯¯ − g(φλλ + φ λλ) − gmφ φ(φ + φ) − g 2 (φ φ)2 (1. .28 1 Classical ﬁelds.81) is obtained after using the equations of motion).84) Similar terms can be written down for several spinor ﬁelds (λ. ¯¯ λλφ . The angular momentum density i ∂ ∂ M κ. λλφ .76).82) − xν − σ µν ∂ xν ∂ xµ 2 (where four-dimensional σµν = (i/2)[γµ .) interacting with scalar ﬁelds. For instance. † This form of the lagrangian follows after eliminating auxiliary ﬁelds via their (algebraic) equations of motion.µν = i ¯ γ κ x µ (1. Additional symmetries. the so-called Wess–Zumino model. where M is given by (1.80) and the energy–momentum tensor density reads µν = i ¯ γµ (1.79) For the hamiltonian density we get H= ∂0 −L= γ γ 0 (−iγ · ∂ + m) ∂ ∂ xν (1. γν ]) gives the conserved angular momentum tensor M µν = d3 x M 0µν (1.83) The lagrangian (1. supersymmetry (see Chapter 15).65) can be extended to include various interaction terms.81) ((1. see Chapter 15. The momentum conjugate to is α = ∂L =i ∂(∂0 α ) † † α (1. if assumed. In the four-component notation we get (φλλ + φ λλ) = ¯ M PL M φ + ¯ M PR M φ .

with a scalar ﬁeld (x) or a vector ﬁeld Aµ (x). 1 ¯ γµ A µ M . the lagrangian (1.86) etc. If λi s transform as a representation R (with hermitean generators T a ) of an internal symmetry group. for example. The last term. (1.1.3 Gauge ﬁeld theories 29 In general. 1 ¯ M2 ¯ 1. It is straightforward to include the interaction terms in the equation of motion. It is often introduced as follows: consider a free-ﬁeld theory of n Dirac particles with . is called the Fermi interaction. are possible as well. For the abelian ﬁeld Aµ this is the same as the minimal coupling of the electron to the electromagnetic ﬁeld ( pµ → pµ + e Aµ ). A Lorentz-invariant and hermitean interaction part of the lagrangian density may consist.72).70).87) The ﬁrst two terms are Yukawa couplings (scalar and pseudoscalar).70)) this interaction can be added to the lagrangian (1. the interactions with vector ﬁelds Aa in the adjoint µ representation of the group (as for the gauge ﬁelds. Other interaction terms of a Dirac ﬁeld (1. with the mass scale denoted by M. The terms in (1. For a vector-like theory. (1. describing the self-interactions of a Dirac ﬁeld. see Appendix E) can be supplemented by terms like: j ¯ ¯ ¯ ¯ λi σ µ (T a )i λ j Aa + λic σ µ (−T a∗ )i j λc j Aa µ µ ¯ µ ¯ ¯ = λi σ µ (T a )i λ j Aa + λc j σ µ (T a ) j i λcj Aa µ j = ¯ i γ µ (T a )i j j Aa µ In the last form (in which (x) is given by (1.3 Gauge ﬁeld theories U (1) gauge symmetry As a well-known example. for example. 1 ¯ M .72) and the internal symmetry indices are properly contracted. in the four-component notation the Yukawa terms have the generic j j i i structure ( ¯ R L ± ¯ L R ) k ( k ) where the chiral ﬁelds are deﬁned by (1. of the operators: ¯ ¯ γ5 ¯ γ 5 γµ Aµ . see Section 1.71). . 1 ¯ M Aµ Aµ .87) require dimensionful coupling constants.68) in which λci transforms as a representation R ∗ of the group (with generators −T a∗ .3) take the form: j ¯ ¯ λi σ µ (T a )i λ j Aa = µ a j i µ L γ (T )i Lj Aa µ with L given by (1. we consider gauge invariance in electrodynamics.

Equivalently.93) where e is an arbitrary positive constant.1). By Noether’s theorem (see Section 1.88) implies the existence of the conserved current jµ (x) = i qi ¯ i γµ i (1. The covariant derivative is constructed by introducing a vector (gauge) ﬁeld Aµ (x) and deﬁning† Dµ i (x) = [∂µ + iqi e Aµ (x)] i (x) (1.88) is invariant under that group of transformations. . symmetries and their breaking the lagrangian density n L= i=1 ¯ i iγ µ ∂µ i −m ¯i i (1.88) Then deﬁne a U (1) group of transformations on the ﬁelds by i (x) = exp(−iqi ) i (x) (1. The transformation rule (1.88) is not invariant under gauge transformations because transformation of the derivatives of ﬁelds gives extra terms proportional to ∂µ (x).90) and therefore conservation of the corresponding charge (1. the U (1) symmetry of the lagrangian (1. To make the lagrangian invariant one must introduce a new term which can compensate for the extra terms.17).88). note that e < 0 in this reference). one should ﬁnd a modiﬁed derivative Dµ i (x) which transforms like i (x) [Dµ i (x)] = exp[−iqi (x)]Dµ i (x) (1.91) It is straightforward to verify that lagrangian (1. The derivative Dµ is called a covariant derivative.94) In particular † The sign convention in the covariant derivative is consistent with the standard minimal coupling in the classical limit (Bjorken and Drell (1964).92) is ensured if the gauge ﬁeld Aµ (x) transforms as: 1 Aµ (x) = Aµ (x) + ∂µ (x) e Covariant derivatives play an important role in gauge theories. We now consider gauge transformations (local phase transformations in which is allowed to vary with x) i (x) = exp[−iqi (x)] i (x) (1.89) where the parameter qi is an eigenvalue of the generator Q of U (1) and numbers the representation to which the ﬁeld i belongs.92) and replace ∂µ by Dµ in the lagrangian (1. The lagrangian (1. (1.30 1 Classical ﬁelds.

The ﬁeld strength tensor Fµν can be used to complete the lagrangian with the gauge-invariant kinetic energy term for the gauge ﬁeld itself (assume one fermion ﬁeld for deﬁniteness) / L = ¯ (iD − m) 2 − 1 Fµν 4 (1. Exactly analogous considerations apply to scalar ﬁeld theories with U (1) gauge symmetry and to theories with Weyl fermions.97) The Euler–Lagrange equations of motion are now ∂µ F µν = eq ¯ γ ν (i/ − m) ∂ or / (iD − m) =0 and identifying e (e > 0) with electric charge we recognize in our theory the Maxwell–Dirac electrodynamics. We start with the free-ﬁeld lagrangian for Dirac ﬁelds which transform according to a representation of some non-abelian Lie group (x) = exp(−i α / = eq A T α ) (x) (1.96) is gauge-invariant. Of course.11).1.98) where the T α are the hermitean matrix representations of the generators of the group.3 Gauge ﬁeld theories 31 one may construct new covariant objects by repeated application of covariant derivatives. Dν ] one gets [Dµ .95) we conclude that Fµν (x) = ∂µ Aν (x) − ∂ν Aµ (x) (1. It also implies absence of the gauge ﬁeld mass term m 2 Aµ Aµ . Dν ] i i = Dµ (Dν i) − Dν (Dµ i) = iqi e[∂µ Aν (x) − ∂ν Aµ (x)] i (1. . U (1) gauge invariance implies global U (1) invariance and the conservation of current (1. But such a term does not break the global U (1) symmetry. Non-abelian gauge symmetry To construct a non-abelian gauge ﬁeld lagrangian we repeat the same steps. appropriate for the ﬁelds and satisfying relations (1.17).90) and the corresponding charge (1.95) By comparing the gauge transformation properties of both sides in (1. For the antisymmetric product of two derivatives [Dµ .

103) and (1.105) that under global transformations (∂µ = 0) the gauge ﬁelds transform according to the adjoint representation of the group with (T α )βγ = −icαβγ .100) the condition (1. (x)] where (x) = +i α (1. The constant g is arbitrary.100) Thus we need gauge ﬁelds in the number given by the number of generators of the group.105) form a group. In classical physics. at the ¯ ¯ quantum level the normalization of the ﬁeld is ﬁxed by the normalization of the single particle state. It is also seen from (1. symmetries and their breaking The free-ﬁeld lagrangian is invariant under global group transformations (1. The corresponding transformation for the gauge ﬁelds Aα (x) reads µ δ Aα (x) = (1/g)∂µ µ α (x) + cαβγ β (x)Aγ (x) µ (1. this is convenient for embedding electromagnetism into the electroweak theory.93) we expect it to be given by a combination of the normal derivative and a transformation on the ﬁelds Dµ (x) = [∂µ + Aµ (x)] (x) where Aµ is a (antihermitean) element of the Lie algebra† Aµ (x) = +ig Aα (x)T α µ (1.102) (1.105) It can be checked that transformations (1.99) By analogy with (1.103) (x)T α (1. g is a dimensionful parameter and therefore can always be scaled to one (similar arguments apply to (1.32 1 Classical ﬁelds. Equivalently. With the form (1. Generalizing the U (1) case we seek a covariant derivative such that [Dµ (x)] = exp[−i α (x)T α ]Dµ (x) = U (x)Dµ (x) (1. † The sign convention for the non-abelian case is consistent with the convention (1.98).93).99) is satisﬁed if the gauge transformation rule for Aµ (x) is as follows: Aµ (x) = U (x)Aµ (x)U −1 (x) − [∂µU (x)]U −1 (x) = U (x)[Aµ (x) + ∂µ ]U −1 (x) or for an inﬁnitesimal transformation δ Aµ (x) = Aµ (x) − Aµ (x) = ∂µ (x) + [Aµ (x). . Let us now consider the extension of the group G to a group of local gauge transformations.104) is an inﬁnitesimal gauge parameter in the matrix notation. At the quantum level g is relevant since quantum theory contains the Planck constant h and g is dimensionless in units h = 1.101) (1.93)).

for instance. (x)] (1. We get.109).1. Aν (x)] G µν (x) = U (x)G µν (x)U −1 (x) For an inﬁnitesimal transformation we get δG µν (x) = [G µν (x). Dν ]G ρσ = [G µν . we can generalize the deﬁnition of the covariant derivative to apply it to any Lie algebra element ξ(x) = +iξ α (x)T α . µν Finally.106) is a covariant quantity transforming under gauge transformations as follows: (1.107) that the tensor G µν can be decomposed in terms of group generators G µν = igG α T α µν where G α = ∂µ Aα − ∂ν Aα − gcαβγ Aβ Aγ µν ν µ µ ν (1.108) It is obvious from (1. G ρσ ] (1. [Dµ .108) and (1.113a) (1.3 Gauge ﬁeld theories 33 We similarly can construct new covariant quantities by repeated application of the covariant derivative. Dν ] (x) = (∂µ Aν (x) − ∂ν Aµ (x) + [Aµ (x). It reads L=+ 1 / Tr[G µν G µν ] + ¯ (iD − m) 2g 2 (1.113) (1.107) (1. we have [Dµ .109) (1.111) The transformation rule for G α (x) follows from (1. Aν (x)]) (x) Thus. ξ ] Then the gauge transformation (1.112) + ξ(Dρ ) (1. the antisymmetric tensor (the ﬁeld strength) G µν (x) = ∂µ Aν (x) − ∂ν Aµ (x) + [Aµ (x).114) (Tr[G µν G µν ] is gauge-invariant because Tr[U G µν G µν U −1 ] = Tr[G µν G µν ]). .110) We are ready to write down the gauge-invariant lagrangian for a non-abelian gauge ﬁeld theory with fermions.103) can be written simply as δ Aµ (x) = Dµ (x) Moreover. By applying the covariant derivative Dρ = ∂ρ + Aρ (x) to the object ξ and insisting on the Leibnitz rule Dρ (ξ ) = (Dρ ξ ) we get Dµ ξ = ∂µ ξ + [Aµ .

h. Its gauge-invariant form then implies the presence of two gauge self-interaction terms of the order g and g 2 .34 1 Classical ﬁelds.107) under the variation (1. of (1. The variation of the gauge ﬁeld action is δ d4 x 1 2 Tr[G µν G µν ] = 2 2 2g g d4 x Tr[G µν (Dµ δ Aν )] (1.120) we may replace the covariant derivative by the ordinary .119) (antisymmetry of G µν in µ and ν has been used).s. One requires the action to be stationary with respect to small variations of the gauge ﬁelds Aµ → Aµ + δ Aµ and its derivative δ(∂ν Aµ ) = ∂ν (δ Aµ ) where we formally treat δ Aµ as a covariant quantity transforming in an adjoint representation of the group δ Aµ (x) = U (x)δ Aµ (x)U −1 (x) (1. Using the relation Dµ Tr[(G µν δ Aν )] = Tr[(Dµ G µν )δ Aν ] + Tr[G µν (Dµ δ Aν )] (1. The pure gauge ﬁeld term must be present in the lagrangian because we need the kinetic energy term to be quadratic in the gauge ﬁelds.120) and the invariance of Tr[G µν δ Aν ] under gauge transformations (therefore.116) can be expressed as follows: G µν → G µν + Dµ (δ Aν ) − Dν (δ Aµ ) (1.112) of a covariant derivative acting on a covariant object δ Aµ . symmetries and their breaking Terms of higher order in ﬁelds are not allowed for quantum ﬁeld theory if it is to be renormalizable (Chapter 4). respectively 1 Tr[G µν G µν ] = − 1 (∂µ Aα − ∂ν Aα )2 + gcαβγ Aβ Aγ ∂ µ Aνα ν µ µ ν 4 2g 2 − 1 g 2 cαβγ cασ δ Aβ Aγ Aµσ Aνδ µ ν 4 (1.117) (1. 2 We end this section with a derivation of the Euler–Lagrange equations for gauge ﬁelds. on the l. The term G µ µ is zero and the possibility of a parity non-conserving term εµνρδ G µν G ρδ in the lagrangian will be discussed later.116) The change of the ﬁeld strength (1.115) (we recall our normalization Tr[T α T β ] = 1 δ αβ ).118) Here we have used deﬁnition (1.

since gauge ﬁelds are not neutral their contribution must be included to ﬁnd conserved charges. We also see from (1. However.123) Combining (1.123) Dµ j µ (x) = 0 (1.122) / d4 x ¯ (iD − m) where the current jµ deﬁned by the variation (1. 1. this remark also helps to check relation (1. Itzykson & Zuber (1980)).4 From classical to quantum ﬁelds (canonical quantization) In this book we use the path integral approach to the quantization of ﬁelds (see Chapter 2). in this section we brieﬂy recall the canonical quantization. procedure (for more details see Bjorken & Drell (1964).124) one derives the covariant divergence equation for the current (1.125) Thus.124) Applying the covariant derivative to (1. for the sake of completeness and easy reference. that δ d4 x 1 2 Tr[G µν G µν ] = − 2 2 2g g d4 x Tr[(Dµ G µν )δ Aν ] (1.s. by integrating (1.121) and (1.122) reads α jµ (x) = i jµ (x)T α with α jµ = ¯ γµ T α (1.4 From classical to quantum ﬁelds 35 derivative..1. .120)) we conclude. gauge ﬁelds can only couple consistently to currents which are covariantly conserved.120) and neglecting the surface terms at inﬁnity on the l.122) we get the equation of motion (1/g 2 )Dµ G µν − j ν = 0 (1.h.125) that the non-abelian charge associated with the fermionic current is not a constant of motion.121) The remaining term δ can be written as δ / d4 x ¯ (iD − m) =2 d4 x Tr[ jµ δ Aµ ] (1.

130) (1.21) by commutators.36 1 Classical ﬁelds.128) that the (hermitean conjugate to each other) operators a(k) and a † (k) satisfy the following algebra: a(k). An arbitrary solution to the free Klein–Gordon equation (1. i (t.e. k0 = E = +(k2 + m 2 )1/2 and. for a real ﬁeld (x). x). a(k )] = a † (k). y)] j (t. symmetries and their breaking Scalar ﬁelds We begin with a scalar ﬁeld theory and consider only free ﬁelds. the ﬁelds and their conjugate momenta = ∂0 are promoted to hermitean operators in the Heisenberg picture (in a Hilbert space of state vectors) by postulating for them canonical equal-time commutation relations. the a(k) and a † (k) are complex conjugate to each other.129) The energy–momentum operator (1. Bjorken & Drell (1965). In the canonical quantization. It follows from the postulated relations (1. j (t. } → −i[ . we see that the vacuum energy is inﬁnite. ]: [ [ i (t. for example. y)] = −iδi j δ(x − y) =[ i (t.126) is a Lorentz-invariant measure. i. 0|0 = 1 (1.131) Since a(k)a † (k) = a † (k)a(k) + (2π )3 2k0 δ(0). For most physical questions (apart from the problem of the cosmological .40) can be rewritten in terms of the operators a(k) and a † (k): Pµ = d3 x 0µ (x) = 1 2 d3 k k µ a † (k)a(k) + a(k)a † (k) (2π )3 2k0 (1. The canonical quantization of interacting ﬁelds can be performed along similar lines in the so-called interaction picture (we refer the reader to the standard textbooks. a † (k ) = 0 The ground state (the vacuum) is deﬁned by the relations: a(k)|0 = 0. j (t. replacing the Poisson brackets in (1.56) may be expanded as a Fourier integral over plane-wave solutions: (x) = where d3 k = (2π )3 2k0 d4 k 2π δ(k 2 − m 2 )θ(k0 ) (2π )4 (1. x). j) are internal symmetry indices). a † (k ) = (2π )3 2k0 δ(k − k ) [a(k). see also Section 2. y)] =0 (1.128) (the indices (i. x).127) d3 k a(k) exp(−ik · x) + a † (k) exp(ik · x) (2π )3 2k0 (1. { .7).

a † (k) = k µ a † (k) and. The ﬁeld–particle duality now becomes evident.132) that P µ . : : = (−) (−) + 2 (−) (+) + (+) (+) . the wave-function (x) describes a state in the conﬁguration representation.e. we redeﬁne the hamiltonian as P 0 → P 0 − 0|(a † (k)a(k) + a(k)a † (k))|0 = The prescription 1 † (a (k)a(k) 2 d3 k k 0 a † (k)a(k) (2π )3 2k0 (1. it is its negative frequency part which creates a particle with positive energy. the a † (k) and a(k) are the creation and annihilation operators. the positive frequency part of the ﬁeld (x) annihilates that state into the vacuum and the negative frequency part creates two-particle state. Also. the ﬁeld (x) (its negative frequency part) acting on the vacuum creates a particle at point x. interpreted as a quantum ﬁeld.137) † Note that in quantum mechanics. consequently.126). with its momentum representation k| (x)|0 = x |k † . for example. Hence.† When acting on a particle at the point x. . in quantum mechanics the positive frequency part of (x) describes propagation with positive energy but. respectively. (y)] = 0 (1.1.135) (1. (x) = x | (t) . with continuous momentum spectrum and propagating as a plane wave (with positive energy).4 From classical to quantum ﬁelds 37 constant) the vacuum energy is unobservable. Hence. Simultaneous measurement of the ﬁeld strength at points x and y is possible only if [ (x). the state |k ≡ a † (k)|0 is an energy–momentum eigenstate.132) + a(k)a † (k)) → : 1 (a † (k)a(k) + a(k)a † (k)): ≡ a † (k)a(k) (1. P µ a † (k)|0 = k µ a † (k)|0 (1. For a product of the ﬁelds (x) = (+) (x) + (−) (x) we get.134) Thus.136) i. It follows from (1. we get k| (x)|0 = 0|a(k) (x)|0 = exp(ik · x) (1. for a quantum of four-momentum k µ . Its only effect is to subtract from the original expressions the inﬁnite vacuum energy. Therefore. where (+) ( (−) ) are positive (negative) frequency parts of (x) given by (1.133) 2 is called ‘normal (or Wick’s) ordering’ and is denoted by a double-dot symbol. whereas in quantum ﬁeld theory (x) |0 is a state |x of a particle localized at the space-time point x.129) and (1.

analogous to (1. j (t. a2 (k) and a1 (k). symmetries and their breaking By direct calculation we get [ (x). y) = −iδi j δ(x − y) (1. Fourier integrals.139) (y − x) vanishes for t1 − t2 = 0.141) we postulate i (t. They can easily be expressed in terms of the hermitean annihilation and † † creations operators a1 (k).38 1 Classical ﬁelds. † i = ∂0 i (1. Quantization of a complex scalar ﬁeld (x) proceeds in quite an analogous way. operator-valued functions. y) = † i (t. In terms of the complex ﬁeld and its conjugate momenta i = ∂0 † i.128) that the quanta of a scalar ﬁeld obey Bose–Einstein statistics. i. It follows from the postulated commutation relations (1. the procedure is exactly the same if we work with the two real ﬁelds φ and χ deﬁned in (1. x).54).138) is a solution of a free Klein–Gordon equation x + m2 (x − y) = 0 (x − y) = − (y − x) Since that (1. now read: d3 k † a(k) exp(−ik · x) + b (k) exp(ik · x) (x) = (2π )3 2k0 (1. (y)] = where ε(k0 ) = One checks that +1 −1 for k0 > 0 k0 < 0 d4 k 2π δ(k 2 − m 2 )ε(k0 ) exp[−ik · (x − y)] ≡ i (x − y) (2π )4 (1. consistently with the notion of locality and causality (two points with space-like intervals cannot be connected in a causal way). In fact. x).140) Thus the ﬁeld strength is simultaneously measurable in two points with a spacelike interval.142) and the other commutators vanish. from its Lorentz invariance we conclude (x − y) = 0 for all (x − y)2 < 0 (1. non-hermitean.143) d3 k † † b(k) exp(−ik · x) + a (k) exp(ik · x) (x) = (2π)3 2k0 where a(k) and b(k) are two independent. † j (t.126). a2 (k) for the quanta of the real ﬁelds .e.

using (1.144) Thus.146)). A generalization to non-abelian global symmetries is straightforward. the ﬁelds † carry opposite U (1) charges.55) is invariant under abelian global U (1) symmetry: → exp(−iq ) .146) Note that the charge of the state created by the ﬁeld (x) transforming as (x) → exp(−iq ) (x) is Q = −q. Therefore.17) can be and expressed in terms of the operators a(k) and b(k): Q = iq =q d3 x : † (x)∂0 (x) − ∂0 † (x) (x) : (1. . We consider the wave-function in the conﬁguration space of the one-particle state b† (k)|0 : φ(x) = x|b† (k)|0 = 0| † (x)b† (k)|0 and in the U (1) rotated basis (or of the rotated physical system in the same basis) φ (x) = 0| † (x)U b† (k)|0 . Moreover. A transparent interpretation of the quanta a and b is obtained by noting that the classical lagrangian (1. (1.† The conserved charge Q (1. a † (k ) = b(k). Using (1.145) we get [Q.143) that the ﬁeld (x) ( † (x)) creates quanta of type b (a) and annihilates those of type a (b). i. a(k). b† (k) are also annihilation and creation operators and it is clear from (1.23) with { . b† (k ) = (2π )3 2k0 δ(k − k ) (1.1.e. From (1.142) there follow the commutation relations for operators a(k) and b(k): a(k). b(k) and a † (k). The Feynman propagator For the sake of deﬁniteness.146) we get φ (x) = exp(−iq )φ(x). i. These equations can also be interpreted in the following way. † → exp(iq ) † .4 From classical to quantum ﬁelds 39 φ and χ . One-particle states a (Q = +1) and b (Q = −1) at † The value of the charge q relative to the charges of other ﬁelds can only be ﬁxed in the presence of interactions. (x)] = −q (x).145) d3 k a † (k)a(k) − b† (k)b(k) ≡ q(Na − Nb ) (2π)3 2k0 This equation deﬁnes the operators Na and Nb which are the number operators for the quanta a and b carrying opposite charges: Qa † (k)|0 = qa † (k)|0 Qb† (k)|0 = −qb† (k)|0 (1. } replaced by −i[ . We can introduce the unitary operator U = exp(i Q) ≈ 1 + i Q such that U b† (k)|0 = exp(−i q) b† (k)|0 and U (x)U −1 = exp(−i q) (x) (we have used (1. we again consider scalar ﬁeld theory with abelian U (1) symmetry and q = 1. ]. the symmetry transformation on the quantum ﬁelds is the analogue of the transformation on the classical ﬁelds interpreted as the wave-functions of one-particle states.e.

x) = (−) (t. x) are created by the operators † (t. x)|0 (t. x) ≡ θ(t − t) 0| (t . x) with t > t : θ (t − t ) 0| = θ(t − t ) 0| † (t. x ) 0 + θ(t − t ) 0| ≡ 0|T = † † † (t. x) (t . x) (t .151) = where T denotes the so-called chronological ordering of the ﬁeld operators. x ) = θ(t − t) 0| (+) † (t. x) (−) †(+) (t .147) (t. Another way of increasing the charge by +1 at (t. x)|0 (1. x ). x )|0 (1. x) = d3 k a † (k) exp(ik · x) 3 2E (2π ) d3 k b† (k) exp(ik · x) 3 2E (2π ) (1. x)|0 ≡ †(−) (−) (t.148) Let the particle a propagate to t > t: the probability amplitude to ﬁnd it at (t . x) (t . x )|0 (t.40 1 Classical ﬁelds. x) and its reabsorption into the vacuum at (t . x) and lowering it by −1 at (t . Spinor ﬁelds Quantization of a Weyl or Dirac ﬁeld can be based on the phenomenological Pauli’s exclusion principle for spin 1 particles or on the requirement of locality 2 .150) The Feynman propagator is deﬁned as superposition of the two amplitudes: G (2) (x . x)|0 †(−) (t . x)|0 (1. x ) and to propagate it to (t. symmetries and their breaking the space-time point (t. x ) |0 d3 k θ(t − t) exp[−ik · (x − x)] (2π )3 2E k + θ(t − t ) exp[ik · (x − x)] i d4 k exp[−ik · (x − x)] 4 k 2 − m 2 + iε (2π ) (1. x ) is given by θ(t − t) 0| (t .149) This amplitude can also be interpreted as the creation of the particle a at (t. x ) is to create a quantum of charge −1 at (t . x)|0 ≡ where †(−) (t. x ) (t. x )|0 (t. x)|0 (t.

155) and all other anticommutators vanish. Using the classical spinors introduced in Section 1. we postulate (the canonical momentum ˙ conjugate to λα (x) is α = iλ†˙ (σ 0 )βα ) ¯ β λα (t.e.152) into the analogue of (1. It can be also introduced by postulating supersymmetry (a certain symmetry between bosons and fermions. As for scalar ﬁelds. We are always consistently led to the conclusion that the Weyl and Dirac ﬁelds must be quantized by assigning for them certain anticommutation relations. this is the content of the theorem on the connection between spin and statistics).152) and for its hermitean conjugation λ† (x) (related to the right-handed ﬁeld through α ˙ ˙˙ ¯˙ λβ = λ† ε β α ): α ˙ λ† (x) = α ˙ d3 k b† (k)aα (k) exp(ik · x) + dR (k)bα (k) exp(−ik · x) ˙ ˙ (2π)3 2E L (1.156) we ﬁnd Pµ = = d3 k † † k µ bL (k)bL (k) − dR (k)dR (k) (2π)3 2E d3 k † † † k µ bL (k)bL (k) + dR (k)dR (k) − dR (k).1.153) where aα (k) and bα (k) are positive and negative energy chiral solutions to a free Weyl equation iσ µ ∂µ λ(x) = 0 (normalized as in (A. y) = (σ 0 )αβ δ(x − y) ˙ β (1.154) Other anticommutators vanish. dR (k ) = (2π )3 2k 0 δ(k − k ) (1.68)) and bL .128).81) for the left-handed Weyl ﬁelds. The expansion coefﬁcients become operators with the following anticommutation relations: † † bL (k).4 From classical to quantum ﬁelds 41 (i. dR are expansion ¯ coefﬁcients. see Chapter 15). µν ¯¯ = iλσ µ ∂ ν λ (1. λ†˙ (t. Inserting (1.2 and Appendix A.157) . dR (k) 3 2E (2π) (1. the classical Weyl ﬁelds are now promoted to operators but instead of relations (1. commutativity at the space-like intervals for observables constructed from the Dirac spinors. we can write down the general wave expansion for solutions of the free Weyl equations for the massless left-handed ﬁeld λα (x) λα (x) = d3 k † bL (k)aα (k) exp(−ik · x) + dR (k)bα (k) exp(ik · x) 3 2E (2π) (1. x). bL (k ) = dR (k).

The normal ordering for the Weyl ﬁelds is deﬁned as: (+) †(−) †(+) :λ† λβ : ≡ λ†(+) λ(+) + λ†(−) λβ + λα λ(−) − λ(−) λα β β β α ˙ α ˙ α ˙ ˙ ˙ (1. with the vacuum deﬁned by bL (k)|0 = dR (k)|0 = 0. After the normal ordering we get Pµ = = ¯¯ d3 x :iλσ 0 ∂ µ λ(x): d3 k † † k µ bL (k)bL (k) + dR (k)dR (k) (2π )3 2E (1. In consistency with the anticommutation relations (1.159) i. Indeed. † − NR (k) = dR (k)dR (k) d3 x :λ† σ 0 λ: = q ¯ d3 k † :b† (k)bL (k) + dR (k)dR (k): (2π )3 2E L (1. We redeﬁne the vacuum by subtracting this inﬁnite c-number. we see that λα (x) creates a helicity h = + 1 state with charge −q at the point x.e.160) d3 k + − NL (k) − NR (k) (2π)3 2E can be interpreted as the number operators for positive-energy state with U (1) charges +q and −q. k) which. After the normal ordering we get Q=q =q where † + NL (k) = bL (k)bL (k). x) can be expressed in terms of the creation and annihilation operators. respectively. Acting with the ﬁeld λ(x) on the vacuum and † projecting on the momentum eigenstates dR (k)|0 . symmetries and their breaking It follows from (1. have opposite charges under the global U (1) phase transformations. The conserved current j µ = q λσ µ λ µ µ 3 0 (∂ j /∂ x = 0) and the corresponding charge Q = d x j (t.153).159) that operators bL (k) and dR (k) have the interpretation of creation operators of positive-energy states with four-momentum (E.158) where the subscript ‘+’ (‘−’) denotes positive (negative) frequency parts of λ and λ† given by (1.152) and (1. It follows from the † † relation (1. we have redeﬁned the vacuum so that its energy is zero. h = + 1 = k. h = + 1 |λα (x)|0 2 2 = 0|dR (k)λα (x)|0 † † ˙ = b α (k) exp(−ik · x) . there is a minus sign for each interchange of λ and λ† required by the normal ordering. the lagrangian for a single Weyl ﬁeld has global U (1) symmetry generated by ¯¯ the transformations λ = exp(−iq )λ. its energy is not only inﬁnite (as for the scalar ﬁelds) but also is not positive deﬁnite. described by a superposition 2 of the plane wave solutions α ˙ x|k. This is accomplished by the normal ordering.155).157) and from the anticommutation relations (1.42 1 Classical ﬁelds. however.155) that.

h = − 1 = k. s)u(k. s) (2π )3 2E The operators b† (k. s = 1.162) (α and β are spinor indices) with the other anticommutators vanishing. s)u(k. The ﬁeld λ† (x) α ˙ creates a helicity − 1 state with charge +q at the point x – a state which 2 is a superposition of the plane wave solutions α x|k. s) exp(−ik · x) (2π )3 2E (1. s ) = (2π)3 2k 0 δss δ(k − k ) and all other anticommutators vanish. s). s) and v(k. s) we get b(k. s) have the interpretation of creation operators of positive-energy states with four-momentum (E.2 d3 k b(k. y) = δαβ δ(x − y) (1. we get for the momentum operator the expression Pµ = d3 x : 0µ (x): = s=1. Deﬁning the normal ordering for the Dirac ﬁelds : ¯α β: (+) = ¯α (+) β (−) + ¯α (+) β (−) + ¯α (−) β − (−) β (+) ¯α (1. s)γ0 exp(−ik · x) ¯ satisfying anticommutation relations α (t. that b α (k) ∼ a α (k) is the solution of (A. s)v(k.163) where the superscript ‘+’ (‘−’) denotes positive (negative) frequency parts of and ¯ given by (1. x). b† (k . s) exp(ik · x) † (x) = s=1. The functions u(k. † β (t. s)v(k. s ) = d(k.161) + d † (k. s) and d † (k. d † (k .161). Similar expansions can be written down in terms of the helicity amplitudes.60) with positive energy and positive helicity).2 d3 k b† (k. s) + d † (k. The conserved current .2 d3 k k µ b† (k. s)d(k. ˙ 2 α For the Dirac ﬁeld the procedure described above leads to the ﬁeld operators (x) = s=1. s) are positive and negative energy solutions for the four-component Dirac equation. s).1. h = 2 − 1 |λ† (x)|0 † = aα (k) exp(−ik · x). s) and d(k. For the operators b(k. s)b(k. 2 numbers the + 1 and − 1 spin projections 2 2 along the chosen quantization axis (see Appendix A). k) with opposite charges under the global U (1) phase transformations = exp(−iq ) . s)γ0 exp(ik · x) ¯ (2π )3 2E + d(k.4 From classical to quantum ﬁelds 43 ˙ ˙ with four-momenta k µ and positive energy (note.

s) + d(k. s) exp(−ik · x) with four-momenta k µ . creates a state of particle d with the wave-function given by α x|d(k. s)b(k. a (k). s) = d(k. (1.R u L. s)d † (k.L ) are positive (negative) energy solutions to the Dirac equation (1. in the chiral representation for the Dirac matrices we have a(k) 0 u L (k) = 0 .166) u R (k) = 0 0 a (k) L. symmetries and their breaking j µ = q ¯ γ µ (∂ j µ /∂ x µ = 0) and the corresponding charge Q = can be expressed in terms of the creation and annihilation operators Q=q =q s=1. spin s and positive energy. As for the scalar ﬁelds. s) = b (k.44 1 Classical ﬁelds. s) = b(k.2 d3 x j 0 (t. s) exp(−ik · x) For chiral ﬁelds in the four-component notation one has d3 k † bL. s) (2π)3 2E where N + (k.164) dk N + (k. s) = d † (k.R (x) = and similar expressions for vR and vL with a(k) and a (k) replaced by b(k) and b (k).L .78). x) d3 x : † :=q s=1. b(k). respectively. we † see that α (x) creates a state of particle b with the spin s and charge +q at the point x. PL. on the other hand. one can introduce the unitary operator U which realizes the U (1) transformations on the quantum states and on the quantum Weyl and Dirac ﬁelds. described by a superposition of the plane wave solutions α x|b(k. N − (k.R (k) exp(−ik · x) + dR. s)b(k.165) where u L.R .R vR. s) are interpreted as the number operators for positive-energy states with +q and −q U (1) charges.R (k)u L. The operator α (x). s)| = 0|d(k.L = vR. Obviously. s)d(k. respectively.R = u L. b (k) are solutions to the Weyl . s) − N − (k.2 3 d3 k :b† (k. s) α (x)|0 † † α (x)|0 = vα (k. s) = b† (k.R (vR. s). (We recall that a(k).L (k) exp(ik · x) (2π)3 2E (1. which satisfy the chirality conditions PL. s)|0 . s) † † α (x)|0 † † α (x)|0 = u α (k.L (k)vR. s)| = 0|b(k. † Similarly to the case of Weyl ﬁelds. s): (2π )3 2E (1. acting with the ﬁeld operator α (x) on the † vacuum and projecting on the momentum eigenstates |b(k.

s)γ0 exp(ik · x) ¯ (2π )3 2E + exp(−iρ)b(k.2 d3 k b(k.168) In the frame O the state vector |v describes the same physical state as does the state vector |v in O. We begin with the Lorentz transformations and consider two observers in two different reference frames O and O . s)v(k. a chiral left-handed massless ﬁeld L creates a particle with † charge (−q) and positive helicity and L creates a particle with charge (+q) and negative helicity (see the remarks below (A.146)). s) exp(ik · x) † M (x) = s=1. First of all.167) + exp(iρ)b† (k. s)u(k. (1.76) (neutral under global symmetry transformations) has the following expansion M (x) = s=1. The following results apply to free and interacting ﬁelds. it is obvious that a given (the same) physical quantum system is seen by both observers. It is described by them by the state vectors which form two isomorphic to each other but. We shall make here the following strong assumption which is a necessary (but not sufﬁcient) condition for the change of the reference frame to be a symmetry transformation: the two Hilbert spaces are identical and can be identiﬁed with each other (in consequence.4 From classical to quantum ﬁelds 45 equations. s) exp(−ik · x) (2π )3 2E (1. Therefore the Hilbert space used by observer O is mapped onto itself by a unitary operator U which is deﬁned by the equation: |v = U |v (1. in general. The state vectors form a representation of the Lorentz group. A self-conjugate four-component ﬁeld. s)u(k. s)γ0 exp(−ik · x) ¯ exp(iρ) is a phase factor depending on the choice of the phase for one-particle states (Kayser 1994). Symmetry transformations for quantum ﬁelds One way to discuss symmetry transformations for quantum ﬁelds is to interpret solutions to the classical ﬁeld equations as wave-functions of quantum states (see the remark below (1.) Thus.25) for wave-functions (in conﬁguration space) of quantum states.62)). and by analogy between classical and quantum physics we assume the transformation law (1. s)v(k. We note already here that the assumption about the identity of the two . the operators acting in one of them can also be identiﬁed with suitable operators acting in the other space).2 d3 k b† (k.1. different Hilbert spaces.

ω)|v (the index α refers to the original or to the new reference frame.171) simpliﬁes to: U Writing U (ε) = exp iεµ P µ i U (ω) = exp − ωµν M µν 2 (1.46 1 Classical ﬁelds.169) x| = 0| x | = 0| α (x) α (x ) (1. These equations suggest † Also. for example. space reﬂection (to be discussed later) which is not a symmetry of many physical systems.173) α (x)U −1 = α (x ) (1. .47) interpreted as the equation for the one-particle wave-functions reads: φα (x ) ≡ α x |v = exp( 1 ωµν 2 where α α µν ) β α β x|v (1. one often deﬁnes the operator α (x ) = U −1 α (x)U such that the matrices satisfy the equation β v2 | α (x )|v1 = exp( 1 ωµν µν ) α v2 | β (x)|v1 or the operator α (x) = U (x)U −1 such that 2 v2 | (x)|v1 = v2 | (x)|v1 . (1.41) and (1.170) the same ﬁeld operator appears since the Hilbert spaces of both observers are identical). symmetries and their breaking Hilbert spaces is not necessarily satisﬁed for such physically relevant changes of the reference frame as. Therefore. (1. Our goal is to ﬁnd the action of the operator U on the ﬁeld operators (x) which may carry a Lorentz index α. U α (x)U −1 = exp(− 1 ωµν 2 µν ) β α β (x ) (1.170) |v = U (ε.171) (Internal quantum number indices may be implicit. (x) = x µ ∂ ∂ − xν + ∂ xν ∂ xµ µν (1.†) For translations. in the ﬁrst and the second equation of (1.174) (x) (1.175) which replace the Poisson brackets (1. For Poincar´ transformations x µ = µν x ν ≈ x µ + e εµ + ωµν x ν .51).172) we obtain the inﬁnitesimal form of these transformations ∂ (x) i[Pµ . (x)] = ∂xµ and i M µν .

is Lorentz-invariant.1. the unitary operators U (ε.9) as the equation relating the wave-functions of one-particle states in the original and rotated basis built by the action of the ﬁeld operator (x) on the vacuum: |i = i (x)|0 (the index i refers to the original or rotated basis). ω) can be found and the theory. although not manifestly so.177) In deriving the last equation we have assumed that the vacuum is invariant under the considered group of transformations: U |0 = |0 . (1. Bjorken & Drell (1965). If so. respectively. The Lorentz boosts must be supplemented by additional gauge transformations (to recover the non-covariant gauge in the new frame). With U ( ) = exp(i a Q a ).176) where the states |v and |v = U |v describe the same physical system before and after the rotation of the basis (we again assume that the two Hilbert spaces can be identiﬁed with each other). We may follow the strategy of interpreting (1. using (1. The U (1) transformations in the internal quantum number space on the free quantum ﬁelds have already been discussed in the ﬁrst subsection of this section.176). See. Nevertheless.175) is the quantization of the gauge ﬁelds in non-covariant gauges (for example. Eq. The results obtained there remain generally valid for non-abelian groups of transformations and for interacting ﬁelds. we ﬁnd the transformation law for the operators U i (x)U −1 = exp(−i a T a) ij j (x) (1. for example. in the Coulomb gauge).174) and (1.132) or (1. In every concrete theory those operators can be constructed using equations like (1. We have then: φi (x) ≡ i|v = exp(−i a T a) ij j|v (1.174) and (1.4 From classical to quantum ﬁelds 47 therefore the identiﬁcation of the generators P µ and M µν with the four-momentum vector and the angular momentum tensor. Therefore.177) then remains valid for the original ﬁelds of the lagrangian but not for the physical ﬁelds deﬁned as perturbation around the true vacuum.175) follow from the commutation relations imposed in the quantization procedure. then the theory is Lorentz-invariant. . An interesting exception to relations (1.157) and one can check explicitly if (1.131) or (1. Later we shall also discuss theories with spontaneously broken global symmetries (see Chapter 9) where the ground state of the system (the physical vacuum determined by the equations of motion) is not invariant under the symmetry group of the lagrangian.

]. −ξ )]α ˙ λ P .e. 1 ) representation and vice 2 versa. 0) 2 representation of the Lorentz group to the one in the (0.48 1 Classical ﬁelds. J = J. Clearly. i. A group G of transformations in the space of internal quantum numbers is a symmetry group provided the action is invariant U L(x)U −1 = L(x). For scalar ﬁelds. p = −p. seen from the ˙ ¯ α transforming under Lorentz transformations reﬂected frame. the Klein–Gordon equations of motion for scalar ﬁelds remain form-invariant under those transformations. symmetries and their breaking where Q a are generators of G. total angular momentum J. ] = −T a (1. (1. are obvious: E = E.178) i. if two right-handed coordinate frames O1 and O2 are connected by the Lorentz transformation with parameters ηi . O1 and O2 are connected by the Lorentz transformation with parameters ηi . the three-momentum p. −ξi .23) with { . helicity h = (J · p)/|p| etc. h = −h (1. ξ )βα λα . We may also address the question of the transformation law for classical ﬁelds. Indeed. the most general form of transformation under space reﬂection is: P (x )= ± (x) exp(iγ ) (x) for a real ﬁeld for a complex ﬁeld (1. η Correspondingly.5 Discrete symmetries Space reﬂection Under the transformation of space reﬂection x µ → x µ with t = t. a spinor λα transforming as λβ = M(η . 1. ξi (see Appendix A for the deﬁnition of the parameters ηi and ξi ) then their images under space reﬂection. The Weyl spinors transform under space inversion from the one in the ( 1 . x = −x the transformation laws for classical physical observables such as energy E.181) β . } replaced by −i[ . is a dotted spinor λ P as: ˙ ˙ ¯α η ξ ˙ ¯β λ P = [M †−1 (η . (1. we get [Q a . they have the same form in the original and in the primed reference frames.e.180) where the phase γ is arbitrary for a free ﬁeld (a free scalar complex ﬁeld has no well-deﬁned parity).179) All internal charges are invariant with respect to space inversion.

in the reﬂected coordinate frame it describes. The Weyl equation is not invariant under space reﬂection.184). −ξ ) = †−1 η M (η . according to the representation 2 2 (0. −ξ )]αβ P βσ λσ (x) = P αβ [M(η . we get ¯ ¯ χ α (x) → χ Pα (x ) ≡ exp(iγ )Pαβ χ β (x) = exp(iγ )(σ 0 )αβ χ β (x) ¯˙ ˙ ¯ ˙ ¯ ˙ ˙ (1. 1 ).182) ¯ where the only role of P ≡ σ 0 is to change the indices from undotted to dotted ¯ ones. In this case ¯ P (x ) = exp(iγ )γ0 (x) (1. Thus. in accord with (1. The space reﬂection is a symmetry of the free Dirac fermion. P (x ) deﬁned in (1. For a four-component Dirac spinor in the chiral representation of the Dirac λ matrices. (1. have well-deﬁned properties under the parity transformation. Similarly. the transformation law follows directly from the transλβ formation of the Weyl spinors (1.185) satisﬁes the Dirac’s equation in the reﬂected frame) provided we choose γ = γ . 1 ) again in representation R (and not R ). ξ ) are numerically equal. ¯ γ5 . ξ )]βσ λσ (x) (1.e. and transforms. h = + 1 (− 1 ). ξ ). opposite helicity state. −ξ ) and M(η .1. Moreover. therefore. The positive (negative) energy solution of the Weyl equation iσ µ ∂µ λ = 0 (see Appendix A): ¯ ∼ exp(−ik · x)a(k) (∼ exp(ik · x)b(k)) upon substituting x = −x becomes the solution with positive (negative) energy of the equation iσ µ ∂µ λ = 0.182) satisﬁes the necessary relation: ˙ ¯˙ η ξ ˙˙ ¯ η [M †−1 (η . ¯ γµ .185) Such transformation preserves the form of the massive Dirac equation (i. P (x )= ¯ exp(iγ )(σ 0 λc )α ¯ ˙ exp(iγ )(σ 0 λ)β ¯ (x) (1.183) ¯ because P is numerically equal to the unit matrix. 2 Therefore parity transformation of the Weyl spinors can be a symmetry of a physical system only if it contains the corresponding pairs of fundamental ﬁelds.186) where the phase γ is still arbitrary. bilinear forms like ¯ . The transformation (1. ¯ γµ γ5 etc. 2 It is very important to notice that the parity transformation on the Weyl spinors transforms ﬁelds ( 1 .182).184) η ξ and under Lorentz transformations χ Pα (x ) transforms with M(η .179).5 Discrete symmetries 49 η ξ η since the matrices M †−1 (η . 0) in representation R of a group G acting in the space of 2 internal quantum numbers into (0. . = ¯ cα˙ . Thus: ˙ ¯˙ ¯P λα (x) → λα (x ) ≡ exp(iγ ) P αβ λβ (x) (1.

for one-particle states we have P|a. for example. see.169) reads: 0| P (x )|v P = exp(iγ ) 0| (x)|v (1. this equation is satisﬁed with P (x)P −1 = exp(−iγ ) (x ) (1. Thus. for example. in general. For complex scalar ﬁeld theory. there is no choice of the phases γ such that LP ( P ) = L( P ). assuming spin zero for the pion.190) (1. In quantum theory. by two Lorentz invariants ¯ and ¯ γ5 . This also means that the operators P (x) acting in H P can be identiﬁed with some operators acting in the original Hilbert space. For space reﬂection to be a symmetry of the physical system. if the invariance under space reﬂection is spontaneously broken we face the same situation as for the spontaneously broken global symmetries) and for the creation and annihilation operators we get Pa(k)P −1 = exp(−iγ )a(−k) Pb† (k)P −1 = exp(−iγ )b† (−k) Thus. Identifying P (x ) ≡ (x ). Werle (1966)). k ≡ Pa † (k)|0 = exp(iγ )|a. maps the Hilbert space onto itself. Bjorken & Drell (1965)).189) The operator P can be expressed in terms of the creation and annihilation operators (see.187) (for a real scalar ﬁeld exp(iγ ) = ±1) where |v P = P|v . i. the analogue of (1. k ≡ Pb† (k)|0 = exp(−iγ )|b. . We can then introduce a unitary linear operator P which transforms the states from one reference frame to another. interpreting the transformation on the classical ﬁelds as the transformation on the wave-functions.50 1 Classical ﬁelds.188) (we assume the vacuum to be invariant under space reﬂection. and study its action on the ﬁeld operators. −k (1. be different. If parity is not a symmetry of the lagrangian. For instance.e. symmetries and their breaking The interaction ﬁxes the relative parities of the different ﬁelds provided that space inversion is a good symmetry. From more detailed study of strong interactions it has been determined experimentally that the pion ﬁeld indeed has spin zero and negative parity (and space inversion is a symmetry of strong interactions. −k P|b. if parity is conserved in strong interactions. the pion–nucleon interaction can be described. the Hilbert spaces H of the observer O and H P of the observer O P (in the reﬂected coordinate frame) are isomorphic but can. it is necessary that these two Hilbert spaces are identical and can be identiﬁed with each other. the pion ﬁeld must be a scalar or a pseudoscalar (as seen in strong interactions).

for example. . where S P is a constant matrix and parity can be a symmetry of P the theory.182) (or (1.61). The action of the parity operator P on the left-handed Weyl ˙ ¯ ˙ ﬁeld λα (x) can now be deﬁned.187)). the system must also contain the right-handed ﬁeld operators λci α (x).87) of the classical solutions of the Weyl equations. respectively). † † PdR (k)P −1 = dL (−k) (1.184)) taken as the transformation law for the wave-functions. space reﬂection cannot be a symmetry of such a theory. 2 2 Thus.194) up to arbitrary and uncorrelated phases present in the classical solutions of the Weyl equations (A. λiα transforming as R and λci α transforming as R . the Weyl ﬁelds must be in a real representation. For the four-component Dirac ﬁelds the parity operation can be realized as a mapping of H onto itself. different (the states in H and in HP have opposite helicities − 1 and + 1 .193) ¯ ˙ ¯ ¯ ˙ P λcα (x)P −1 = exp(−iγ )(σ 0 )αβ λβ (x ) For the creation and annihilation operators. In other words. (A.5 Discrete symmetries 51 For a chiral theory containing left-handed Weyl ﬁelds λiα transforming as a complex representation R of the internal symmetry group the Hilbert spaces H P and H mapped into each other by the parity transformation are.152) and (1.64). respectively. In this case we can identify the operator P with (they have the same transformation properties under the internal symmetry group) † If the representation R is irreducible and real (like.1.191) where a (k) and b (k) are positive and negative energy chiral solutions to the free † ¯ Weyl equation iσ µ ∂µ χ(x) = 0 and bR and dL are the annihilation and creation operators. with the expansions analogous to the ones given in (1. using the properties (A. For this to be ¯ ˙ the case. transforming as (0. R = R ⊕ R with. we get† 0 ¯ Pλα (x)P −1 = exp(−iγ )σαβ λcβ (x ) ˙ ˙ (1. adjoint representation of SU (N )) then one ˙ ¯˙ can simply identify λα (x ) with S P λα (x ). Identifying λα (x ) ≡ λcα (x ) and following the P same arguments as for the scalar ﬁelds (see (1. in general.192) (1. we obtain PbL (k)P −1 = bR (−k). 1 ) and R under the Lorentz and internal symmetry group 2 transformations. For instance. with (1.153): d3 k † cα ˙ α ˙ α ˙ ¯ λ (x) = bR (k)a (k) exp(−ik · x) + dL (k)b (k) exp(ik · x) (2π)3 2E d3 k † ¯ bR (k)a α (k) exp(ik · x) + dL (k)b α (k) exp(−ik · x) λc†α (x) = 3 2E (2π) (1. for example.

s)P −1 = exp(−iγ )b(−k. s = ±i|−k.167) the application of the rule (1. symmetries and their breaking so that P (x)P −1 = exp(−iγ )γ 0 (x ) and Pb(p. One can also check that the free Dirac lagrangian is invariant under space reﬂection: PL(x)P −1 = L(x ). the (massive) particle and its antiparticle always have negative relative parity (assuming vacuum to be symmetric under the parity operation. γ 0 v(−p. The spinor properties γ 0 u(−p. dk2 . s) = u(p. s2 )|0 = −|b − k1 . d − k2 . s) have been used (see Appendix A).198) Note that this conclusion does not hold for massless Weyl states because the phases in (1. s)P † −1 (1. s) Pb† (k. s)P −1 = exp(−iγ )b(−p. s1 .195) gives two relations Pb(k. s2 (1. The parity of the electromagnetic ﬁeld is determined by its coupling to the classical current. a Majorana particle at rest (k = 0) is a parity eigenstate with purely imaginary eigenvalue. s) Pd (p. s). P|0 = +|0 ): P|bk1 . s i. Analogous relations for helicity creation and annihilation operators can be derived using (A. s) (1. s1 . s1 )d † (k2 .52 1 Classical ﬁelds.196) (1. s)P −1 = − exp(−iγ )b† (−k. In this case the theory containing the massive Majorana spinor may be parity-invariant. For a massive Majorana ﬁeld operator with the expansion given by (1.195) (1. The requirement of the invariance of the classical lagrangian Lint = −ejµ (x)Aµ (x) describing interaction of the charged four-component Dirac ﬁeld with the electromagnetic ﬁeld Aµ (x) . s2 ≡ Pb† (k1 .197) = − exp(−iγ )d (−p. jµ (x) = q ¯ (x)γµ (x). s) † and similarly for b† and d but with the complex conjugate phase exp(iγ ).194) are arbitrary and uncorrelated (the states with opposite helicities are not related by Lorentz transformations). It is also worth noting that for such a particle P|k.89) Although the phase factor exp(−iγ ) is arbitrary.e. s) = −v(p.199) which are compatible only if exp(−iγ ) = ±i.

203).204) where the phase factor can actually be only ±1 (since the real component ﬁelds cannot change). If the dynamics is invariant under such transformation. if (∂ 2 + m 2 ) (x) = 0 then also (∂ 2 + m 2 ) T (x ) = 0. x)) do not change whereas for the space coordinates of the current jia (−t.e.e. P Thus. Formally. If the ﬁelds carry internal quantum number index i and transform as (x) = exp(−i a T a ) (x) then the ﬁelds T (x ) transform the same way in spite of the complex conjugation because the change t → −t assures invariance of the charges.1. in quantum theory we identify A0 (x ) ≡ A0 (x ) = P A0 (x)P −1 P A P (x ) ≡ A(x ) = −PA(x)P −1 Time reversal The classical sense of time inversion is clear: the velocities of a system in some ﬁnal state are reversed and it is moving backwards.201) A P (x ) = −A(x) (1. All charges remain.200) Classical physical observables behave as follows (compare with (1. time reversal can be described by the change of coordinates† x µ → x µ with t = −t. J = −J. provided P A0 (x ) = A0 (x). x)T = − jia (t. in accord with (1. it is however easier to think in terms of the active transformation. The coupling with the electromagnetic ﬁeld is invariant under the transformation † Physically. h =h (1. We consider now a classical complex scalar ﬁeld (x). From (1. of course. Its transformation under time reversal can be inferred from the physical requirement that the charges a (i. j0 (t.16) (or (1. a positive energy plane wave with momentum p transforms into a positive energy plane wave in the reﬂected frame with momentum p = −p.179)): E = E. unchanged. Clearly.5 Discrete symmetries 53 together with the property P (x ) = exp(iγ )γ0 (x) leads to the conclusion that ¯ P (x )γµ P (x )Aµ (x ) = ¯ (x)γµ (x)Aµ (x). p = −p.57) for the abelian case) we get T (x ) = exp(iγ ) (x) (1. the system will eventually reach the original initial state. i. x =x (1.203) (all time derivatives change sign). x) (since the velocities are reversed). The Klein–Gordon equation is form-invariant under time reversal.202) (1. .

Moreover. T (x ) satisﬁes the Dirac equation in the primed variables) provided γ = γ .207) ˙ and χT (x ) transform again as (0. In this case ¯ (1.54 1 Classical ﬁelds. Taking proper account of the Lorentz indices we ﬁnally conclude that λT α (x ) = exp(iγ )(iσ 1 σ 3 )αβ (λ (x))β ¯ (1. one can argue that a Lorentz transformation described by parameters ηi . in the ‘time-reﬂected’ frame is described by the parameters ηi . In the chiral representation of the Dirac matrices we have (x) → T (x )= ¯ exp(iγ )iσ 1 σ 3 λ (x) 1 3¯c exp(iγ )iσ σ λ (x) ¯ ¯ (1.e. from the additional physical requirement that charges are invariant and ¯¯ the space components of the current jµ (x) = λσ µ λ change sign we conclude that the transformation λα → λT α must include the matrix σ 2 . . −ξ )αβ exp(iγ )(iσ 1 σ 3 )βκ (λκ ) ¯ (1.208) Such transformation preserves the form of the massive Dirac equation (i.204) and the simultaneous transformation of the potential four-vector: A0 (x ) = A0 (x). 1 ) under the Lorentz transformations. the right-handed spinors transform into ˙ ˙ χ α → χT (x ) = exp(iγ )(iσ 1 σ 3 )αβ (χ β (x)) ¯˙ ¯α ¯ ¯ ˙ ¯ ˙ (1. the necessary relation η exp(iγ )(iσ 1 σ 3 )αβ M(η . Since helicity does not change under time reversal. Thus.206) η η ξ is fulﬁlled because σ 1 σ 3 M (η . ξ )βκ λκ ¯ η ξ = M(η .and right-handed Weyl spinors.205) where the phase factor exp(iγ ) is arbitrary. Similarly. It is obvious that λT α (x ) ¯ ¯ satisﬁes the Weyl equation for the left-handed spinors in the primed variables: iσ · ¯ ∂ λT (x ) = 0. −ξ )σ 1 σ 3 . the transformation on the Weyl ﬁelds must involve their complex conjugation. symmetries and their breaking (1. the Lorentz transformation properties for the Weyl ﬁelds should not change. Furthermore. −ξi . Ai (x ) = −Ai (x). Because positive energy solutions with with momenta p must transform into positive energy solutions with −p. similarly to the space inversion. ξ ) = M(η . ¯α 2 The transformation law for the four-component Dirac spinors follows directly from the transformation properties of the left.209) with the phase γ still arbitrary. ξi .208) can be rewritten as T (x † T (x ) = exp(iγ )iγ 1 γ 3 ) = − exp(iγ )i † (x) (x)γ 3 γ 1 (1.

From (1.212) where we have applied‡ the rule (1. Therefore. The equation for the matrix elements similar to (1. the time reﬂection can be a symmetry of the physical system only if the two Hilbert spaces can be identiﬁed with each other.† By applying the antiunitarity condition successively to |v . ‡ Notice that since T is antiunitary 0T | = 0|T −1 . (x)]T −1 = T −i ∂ (x) T −1 ∂t (x ) is The physically motivated requirement that T H (t)T −1 = H (t ) again leads to the conclusion that if to satisfy (1.h.215) ˙ ¯ ˙ ¯˙ T λα (x)T −1 = exp(−iγ )(iσ 1 σ 3 )α ˙ λβ (x ) ¯ ¯ β † Antiunitarity of the T operator can be justiﬁed also by applying it to the time component of (1.212) leads to T λα (x)T −1 = exp(−iγ )(iσ 1 σ 3 )αβ λβ (x ) ¯ (1.214) and similarly for their hermitean conjugates.213) for the creation and annihilation operators we get: T a(k)T −1 = exp(−iγ )a(−k). For the Weyl ﬁelds we can identify HT with H and the operators λT α with λα .174) with primed variables then T must be antiunitary.h.211) to the state T (T −1 T (x )T |v ).s.210) to a state vector |v = α1 |v1 + α2 |v2 . |v1 + |v2 and to |w = |v1 + i|v2 one can show that (Wigner 1931) wT |vT = v|w = w|v The l. of (1.210) It follows that the operator T must be antilinear (antiunitary): T † T = 1 and T (α1 |v + α2 |w ) = α1 T |v + α2 T |w . The transformation on the quantum ﬁelds does not involve any complex conjugation and the consistency with the classical transformation is ensured due to the antilinearity of T . T b† (k)T −1 = exp(−iγ )b† (−k) (1.213) (for a real scalar ﬁeld we must have exp(−iγ ) = ±1). Interpreting the classical transformation (1. As for the space inversion.5 Discrete symmetries 55 In the quantum case we deﬁne an operator T such that |vT = T |v . we conclude that T (x)T −1 = exp(−iγ ) (x ) (1. .174): T [H.210) can now be written as 0T | T (x (1.210).s. This can be seen by applying (1.1.211) )|vT = 0T |T T −1 T (x )T |v = 0|T −1 T (x )T |v (1. with complex coefﬁcients αi . of (1. identifying T (x ) with (x ) and comparing with the r.204) as the transformation on the wave-function 0| (x)|v for a state |v we must have: 0T | T (x )|vT = exp(iγ ) 0| (x)|v (1.

The transformation properties of the electromagnetic ﬁeld under the operation of time reversal (already discussed) can be determined by its coupling to the classical current.167) the application of rule (1.218) one should also remember that T γ µ T −1 = γ µ . AT (x ) = −A(x) (1. † † T dR (k)T −1 = exp(−iγ )dR (−k) (1.217) leading. the requirement of the invariance of the lagrangian Lint = −ejµ (x)Aµ (x) together with the properties (1. s ) T d † (k. to the relations T b(k.218) and the consistency condition exp(−2iγ ) = exp(2iρ). upon using the properties of the classical solutions to the Dirac equation (summarized in (A. For the massive Majorana ﬁeld (1. s = 1(2) denotes the spin projection equal + 1 (− 1 ) with 2 2 respect to the chosen quantization axis).220) . T Thus.56 1 Classical ﬁelds. We consider a † In verifying the invariance of the quantum lagrangian by using (1. s)T −1 = i(−1)s exp(−iγ )d † (−k.91) of the solutions to the Weyl equations we get for the creation and annihilation operators T bL (k)T −1 = exp(−iγ )bL (−k).209) leads to the conclusion† that ¯ T (x )γµ T (x )Aµ (x ) T = ¯ (x)γµ (x)Aµ (x) provided A0 (x ) = A0 (x). For the four-component Dirac ﬁelds one gets T (x)T −1 = exp(−iγ )(iγ 1 γ 3 ) (x ) (1. in quantum theory A0 (x ) ≡ A0 (x ) = T A0 (x)T −1 T AT (x ) ≡ A(x ) = −T A(x)T −1 Charge conjugation The transformation of charge conjugation is genuinely connected to quantum physics.93)).219) (1.218) (s = 2(1) for s = 1(2).217) leads to the ﬁrst relation (1. As in the case of the parity transformation.216) (up to some arbitrary phase factors associated with the classical solutions for the Weyl spinors) and similar relations for the hermitean conjugate operators. It is clear that the operator T changes the direction of the momentum of the state leaving its helicity unchanged. symmetries and their breaking Using the properties (A. s ) (1. It can be deﬁned as a transformation on wave-functions (positive energy solutions to the classical wave equations) of quantum systems. jµ (x) = q ¯ (x)γµ (x). s)T −1 = −i(−1)s exp(−iγ )b(−k.

One should stress that from the fundamental fact of the existence of the negative energy solutions to the relativistic wave equations follows the existence of particle and antiparticle pairs (i.e. This means that the system seen by the observer in the new reference frame satisﬁes other equations of motion than in the original reference frame and the transformation of charge conjugation is not a symmetry of this particular system. This change of the ‘reference frame’ is called the transformation of charge conjugation (we could as well take the active view and replace the original system by a new one with ‘reversed’ charges). respectively): the negative energy solutions can always be reinterpreted as antiparticle (positive energy) wave-functions. . ‘opposite’ internal charges are assigned to the same system. Thus. the states whose wave-functions transform as R and R . obtained by a transformation on its negative energy solutions and usually also called charge conjugation. However.224) (see (E. Charge conjugation is a symmetry of the system if the charge conjugate wave function ic (x) can be identiﬁed with a solution of the relativistic equation of motion of the system. For an observer using this reference frame the same physical system is described by another wave-function ic (x) which transforms as representation R of G.221) = exp(−i T a) i j j pos (x) (1. This transformation may or may not be a symmetry of a given system.10) for the deﬁnition of complex and real representations).5 Discrete symmetries 57 physical system which carries some internal quantum numbers (charges) and is described by the wave-function i (x) transforming as a representation R of a group of transformations G: abelian or non-abelian i pos (x) pos (x) = exp(−iq ) a pos (x) (1. One can introduce a ‘charge conjugate’ reference frame (a ‘reﬂection’ in the space of internal charges).1.222) We assume that the group G is a global symmetry group of this system. complex conjugate to R but has the same space-time Lorentz transformation properties and the same dependence on the space-time variables as the wave-function i (x) c pos (x) = exp[−i(−q) ] c pos (x) (1. after such reinterpretation (often and somewhat misleadingly also called charge conjugation) the antiparticle wave-functions may satisfy different equations of motion than the negative energy solutions did and cannot be identiﬁed with the ‘charge conjugate’ wave-function of the system.223) or ci pos (x) = exp[−i a (−T a )] i j cj pos (x) (1.

neg i. if i (x) are solutions of the Klein–Gordon equation the same is true for the functions ic (x).58 1 Classical ﬁelds.e. It is trivial but important to notice that the operation (1. symmetries and their breaking Again. this does not change the fact that particles and antiparticles do exist in pairs. i. Instead. Moreover. µ µ For the Weyl equation there is no symmetry operation which transforms its neg ˙ c pos ˙ ¯ α neg ¯ cα pos (x)) of negative energy solutions λiα (x) (λi (x)) into solutions λiα (x) (λi the same equation with the same Lorentz transformation properties and the same pos ˙ ¯ α pos space-time dependence as λiα (x) (λi (x)) and with the internal charges reversed pos ¯˙ compared to the charges of λiα (x).225) Thus. We may extend our considerations to scalar ﬁelds with gauge symmetries.e. as predicted by the presence of the negative energy solutions to the equations of motion. We begin with a set of free complex scalar ﬁelds i (x) which satisfy the Klein–Gordon equation. It is easy to check that they are invariant under the operation i (x) → c i (x) = exp(iγ )( i (x)) Aa (x) → Aca (x) : Aca (x)(T a )∗ = −Aa (x)T a µ µ µ µ (1. transforms solutions i (x) which have no wave-function interpretation into c pos (x) which can be interpreted as wave-functions of antiparticles (with reversed i global quantum numbers). The equations of motion follow then from the lagrangian L = ∂ µ + ig Aaµ T a = ∗ ← 2 →µ ∂ µ − ig Aa T a ∂ + ig Aaµ T a µ (in going from the ﬁrst to the second line the hermiticity of the generators T a has been used). Thus. coupled in a gauge-invariant way to (abelian or non-abelian) gauge vector ﬁelds. 0) under . We now consider several examples. the operation λiα (x) → λiα (x) neg 1 transforms the negative energy solution λiα (x) (transforming as R and ( 2 . the existence of the symmetry operation i (x) and i (1. This equation is invariant under the operation i (x) → c i (x) ≡ exp(iγ ) i (x) (1.225) transforms negative energy solutions into positive energy ones and vice versa. the wave-function of the antiparticle in the charge conjugate vector potential Aca (x) has the same form as the particle wave-function in the potential Aa (x).225) for the relativistic scalar wave equation provides the interpretation of its negative energy solutions and one predicts the existence of particle–antiparticle pairs described by wave-functions with the same space-time form but opposite internal charges.226) i. the space-time dependences of the functions pos c pos (x) are the same.e.

respectively) into λi (x) which satisﬁes the Weyl equation for right-handed spinors (i. s) and C v T (p. ¯ For the four-component spinors we have C u T (p. Aa → Aca (x) µ µ (1. λi c ¯˙ ¯ ˙ ( 1 . Such a system is equivalent to a Dirac ﬁeld (x) = λiα (x) ˙ ¯ cβ λi (x) c λiα (x) ˙ ¯β λi (x) (1. 2 The same conclusion remains valid for the Weyl ﬁelds coupled to the abelian or non-abelian gauge ﬁelds.e. The operation (x) → c (x) = exp(iγ ) (1. neg (x) is the same as the helicity of λiα (x). for a quantum system described by a set of left.5 Discrete symmetries 59 ˙ ¯ α pos internal and Lorentz groups. For a Dirac ﬁeld coupled to (abelian or non-abelian) gauge ﬁelds the lagrangian / L = ¯ i/ − g Aa T a − m ∂ is invariant under simultaneous transformation → c . s).226) (remember that µ ¯ † Note that the helicity of the solution λi α pos ˙ s are anticommuting c-numbers.227) which satisﬁes the Dirac equation. (0. 1 ) there exists the symmetry operation λiα → λiα .228) transforms the negative energy solutions neg (x) into positive energy c pos (x) which transform as R and have the same space-time dependence as pos (x). 0) and R .and 2 2 pos ˙ ¯ cα pos (x) transforming as R. right-handed Weyl equations with solutions λiα (x). For a general Dirac ﬁeld (a solution to the massive Dirac equation) there exists the symmetry operation c = exp(iγ )Cγ 0 T = exp(iγ )C ¯ T . . Therefore the positive energy solution to a Weyl equation is a wave-function of a particle with h = − 1 (+ 1 ) and the ‘charge conjugate’ 2 2 form of the negative energy solution is the wave-function of its antiparticle with h = + 1 (− 1 ).† However.229) where the 4×4 matrix C (its explicit form and properties are given in Appendix A) satisﬁes (Cγ0 )γ µ (Cγ0 )−1 = −γ µ .1. s) = ¯ u(p.230) with Aca (x) deﬁned in (1. with the same Lorentz and space˙ pos ¯ α pos time dependence as λi (x) and λiα (x). 1 )). transforming as R and (0. C −1 ¯ c = − exp(−iγ ) (1. respectively. s) = v(p. so that c indeed satisﬁes the Dirac equation. λicα → λiα 2 2 which transforms the negative energy solutions of each of the equations into the positive energy solutions of the other equations.

We can then deﬁne a linear (unitary) operator C. Cb† (k)C −1 = exp(−iγ )a † (k) (1. We use the fact that for the wave-functions 0| c (x)|v c = exp(iγ ) 0| (x)|v . k) into the antiparticle state with the same fourmomentum.229): ¯1 2 = ¯2 1 .243)). After such transformations.126) and Ca(k)C −1 = exp(−iγ )b(k). In the presence of interactions this freedom can be used to deﬁne the maximal discrete symmetry of the system. we can identify the operators c (x) ≡ † (x) (1. we see that.231) getting† C i (x)C −1 = exp(−iγ ) † i (x) (1.225) with (1. the space-time dependence of the solutions (x) and c (x) remains the same. .232) The presence in (1.234) (1. i.60 1 Classical ﬁelds. |v c = C|v . in agreement with the transformation properties under internal symmetry group.143). † It is worth stressing the difference between charge conjugation and time reversal. This difference follows from the antilinearity of T .143) satisfy the following relations Ca(k)C −1 = ±a(k) for real scalar ﬁelds (1. Moving now to the description of the physical system in terms of the state vectors (rather than their representation in the conﬁguration space) the operation of charge conjugation maps the original Hilbert space H onto the Hilbert space Hc which describes the physical system in the ‘charge conjugate’ frame.229) is unphysical and can be taken as zero. This operation can be a symmetry only if H and Hc are identical and can be identiﬁed with each other (in consequence. compare (1. symmetries and their breaking the Dirac lagrangian is invariant up to a total derivative). The states in Hc have the same space-time description but opposite internal charges.213) with (1. Next. ¯1 γµ 2 = − ¯2 γµ 1 etc.232).233) for complex scalar ﬁelds (1. The transformation properties of the spinor c c c c bilinears follow from (1. The creation and annihilation operators deﬁned in (1.126) and (1. the arbitrary phase γ in (1.e.232) of an arbitrary phase reﬂects the freedom to redeﬁne any quantum state by multiplying it by a phase factor.225) and (1. they have the same space-time dependence. we would like to ﬁnd the action of C on the ﬁeld operators. consistently with its deﬁnition.204) and (1. With few exceptions (see (1. charge conjugation transforms a particle state with four-momentum (E. the operators acting in H can be identiﬁed with some operators acting in Hc ). Remembering our discussion in Section 1.4 we see that. For scalar ﬁeld theory.

With λα and λc transforming as R and R .239) up to arbitrary phases present in the classical solutions for the Weyl equations. It is clear that a particle state with momentum k and helicity h is transformed under charge conjugation into an antiparticle state with the same momentum and helicity.235) are eigenvectors of the charge conjugation operator. For the Weyl ﬁelds. ¯ The K0 –K0 mixing will be discussed in more detail at the end of this section and in Section 12. The presence in this equation of the phase factor exp(iγ ) is crucial for a phase-convention-free conclusion. when C (and/or C P) is a good symmetry and the system is neutral in all conserved charges (the latter is necessary since. Assuming the vacuum to be invariant under C. For instance. the Hilbert space HC can be identiﬁed with H (and C can be a symmetry operations) only if the set of those ﬁelds transforms as a real representation of the group G. and following the same arguments as for the (1.5 Discrete symmetries 61 It is sometimes very useful to consider the eigenstates of the charge conjugation operator C (similar comments apply to the combined transformation C P).238) (1. with eigenvalues ±1. Note also that the state |kC can be written as |kC = exp(iφ± ) † a (k)|0 ± Ca † (k)|0 √ 2 (1.7.236) In a very similar manner one can construct the states at rest invariant under the C P transformation even when C and P are not conserved separately. −k . † † CdR (k)C −1 = exp(−iγ )bR (k). respectively.1. .232) we get Cλα C −1 = exp(−iγ )λc α (1. for example. for abelian charges C Q = −QC. k = exp(iγ )|b. so no common eigenstates exist) then the description in terms of the C (or CP) eigenstates is convenient. (1. it then follows that the states given by the linear superposition of the particle and antiparticle states |kC = exp(iφ± ) † a (k)|0 ± exp(iγ )b† (k)|0 √ 2 (1. α respectively.237) and a similar equation for λ† with the opposite phase. Important physical examples are the positronium state (electromagnetic interactions conserve both C and Q but positronium is ¯ ¯ electrically neutral) and the K0 –K0 (or B0 –B0 ) mixing by electroweak interactions (CP is almost conserved but the ﬂavour charges like strangeness or beauty are not).152) and (1.153) we arrive at: α CbL (k)C −1 = exp(−iγ )dL (k). Using the expansions of the α ˙ ﬁelds λα and λc given in (1. For the states with momentum k the C P acts as follows: CP|a.

62 1 Classical ﬁelds.240) which leads to (1. s) Cb(k. Weyl and Dirac particles described by state vectors in some Hilbert space. For those transformations which are symmetries of the system one can introduce unitary operators acting on quantum ﬁelds.241) (1.167).245) It is its own charge conjugate up to a sign. We have learned that a system of free Weyl particles which transforms as a complex representation R of some group of internal symmetries is invariant under the time reﬂection but not under separate . time and charge inversions) on classical ﬁelds. s)C −1 = exp(−iρ) exp(−iγ )b† (k. We have identiﬁed the requirements which are necessary for these transformations to be symmetries of free (or interacting with the gauge ﬁelds) quantum systems of scalar. For a Majorana ﬁeld with expansion (1. it follows that a Majorana particle is an eigenstate of the charge conjugation operator with eigenvalue ±1. s)C −1 = exp(−iγ )d(k. s) and v(k. a state of such a particle at rest may be a CP eigenstate with eigenvalues ±i (see (1.199)). symmetries and their breaking Following the same arguments. s) Consistency of these two equations requires exp(−iρ) exp(−iγ ) = ±1 Thus.244) =± M (x) (1. Summary and the C P T transformation We have discussed the action of the discrete transformations (space. s)C −1 = exp(−iρ) exp(−iγ )b(k. It is interesting that even free particle systems are not all invariant under the three discrete transformations. the action of C on the four-component Dirac ﬁeld operators is given by c (x) ≡ C ¯ † = exp(iγ )C (x)C −1 Cb(k. Also. s) (1. s) Cd † (k.243) (1. the Majorana ﬁeld satisﬁes C M (x)C −1 (1. Thus.46) for spinors u(k. s). s)C −1 = exp(−iγ )b† (k.242) for the creation and annihilation operators. We have used the transformation properties (A.229) we obtain Cb† (k. using the deﬁnition (1.

Lorentz invariance and unitarity.248) independently of the form of the matrix m AB . it can be checked that Lmass remains invariant under the C P T transformation CPT Lmass (CPT )−1 = Lmass (1. and equal magnetic moments. Consider the theory of Weyl fermions in which each set of λs transforming as R has its counterpart λc transforming as R of the internal symmetry group. Similar considerations apply to complex Yukawa couplings. Similarly. We conclude that all the considered systems are invariant under the C P and C P T transformations.246) A particle state with momentum k and helicity h is transformed under C P transformation into an antiparticle state with opposite momentum and helicity (as expected from the Dirac sea interpretation of the negative energy solutions to the Weyl equation). The discrete symmetries can easily be broken by interactions or even mass terms. the ﬁeld λC P (x) satisﬁes the same equation as the ﬁeld ¯˙ λα (x). It is natural to ask what happens to such a system under simultaneous C and P transformations. The most general hermitean mass term of the lagrangian is ¯ ¯B Lmass = −m AB λ A λc − m AB λ A λc B (1. −x).5 Discrete symmetries 63 transformations of space reﬂection and charge conjugation.e.247) It is straightforward to check that if Lmass is to conserve P we must have m AB = m B A . i. the mass matrix m AB must be hermitean. for C to be conserved the mass matrix must be symmetric. i. It turns out that the Weyl equation is invariant under the replacement λα (x) → λC P (x) = ¯˙ exp(−iγ )λα (t. A number of experimental upper limits on violation of the C P T theorem can be found in the Particle Data Book (1996). Among the consequences of the C P T theorem are equal masses for particles and their antiparticles.e. −x) and then using (1. One should also stress that often there exists some arbitrariness in the ﬁeld phase conventions. Finally. The importance of the C P T transformation follows from a fundamental theorem of quantum ﬁeld theory which says that C P T invariance follows from the basic requirements of locality.153) we can infer the transformation properties of the annihilation and creation operators: CPbL (k)(CP)−1 = exp(iγ )dR (−k) † † C PdR (k)C P −1 = exp(iγ )bL (−k) (1.1. For C P to be conserved m AB must be real. equal lifetimes. They include several measurements . For quantized ﬁelds CPλα (x)(CP)−1 = exp(−γ )λα (t. equal and opposite electric charges.152) and (1. which can be used to make the maximal symmetry of the lagrangian explicit.

We recall that both pions and neutral kaons belong to the lightest quark– antiquark bound states (mesons) which are stable with respect to strong interactions.) ¯ C P violation in the neutral K0 –K0 -system In this subsection. down to the Planck scale. Mavromatos & Nanopoulos (1996). On the other hand. . 1964). symmetries and their breaking on electrons and positrons. The names of KL(ong) and KS(hort) originate from their drastically different lifetimes τL 5 × 10−8 s and τS 9 × 10−11 s. we shall discuss the effects of the C P violation in the K0 – ¯ K0 mixing. π − and π 0 are physically observable particles.64 1 Classical ﬁelds.255) We choose our phase convention for normalization of the neutral kaon states so ¯ ¯ that CP|K0 = |K0 and CPT |K0 = |K0 . The most stringent bound exists ¯ for the mass difference of the K0 and K0 : m K /m K < 9 × 10−19 . see Ellis. (For a review of speculations on the possibility of violation of the C P T theorem.254) (1.250) (1. it is of great interest if the theorem remains valid at even smaller scales of length.249) (1. Their ﬂavour structure is the following: |π + |π − ¯ = |u d = |d u ¯ = 1 ¯ ¯ √ |u u − |d d 2 (1. They were the ﬁrst processes in which violation of the C P symmetry was observed (Christenson et al.251) (1. The physical processes under consideration here will be neutral kaon decays to two pions. They decay only due to electroweak interactions.253) |π 0 |K0 ¯ |K0 = |d s ¯ ¯ = |s d While the pion states π + . muons and protons. respectively. their masses m L mS 498 MeV are almost equal: m ≡ m L − m S 3. More details about C P violation in the neutral kaon system can be found in Jarlskog (1989) and Bernabeu (1997). In spite of the tremendous accuracy with which the C P T theorem has been veriﬁed.5 × 10−12 MeV. the ¯ neutral kaon states K0 and K0 combine to the following physical ones: |KL |KS = = 1 2(1 + |¯ |2 ) ε 1 2(1 + 1/2 |¯ |2 ) ε 1/2 (1 + ε)|K0 − (1 − ε)|K0 ¯ ¯ ¯ (1 + ε)|K0 + (1 − ε)|K0 ¯ ¯ ¯ (1.252) (1.

260) Hi j = Mi j − ij 2 i|Hw |n n|Hw | j 3 + O(Hw ) (1.5 Discrete symmetries 65 If C P was conserved in Nature.262) are obtained by sending t0 to (−∞) and turning Hw adiabatically off at t → ±∞ (see.† The matrix Hi j in (1. KL could not decay to two pions which are C P-even in the S-wave.257) and bn (t = t0 ) = 0.262) Here.258) | (t) = (Hs + Hw )| (t) dt where Hw is the electroweak interaction hamiltonian.259) is not hermitean. KL and ¯ KS would be C P-odd and C P-even. Then. (1959) §43).256) where Hs |K0 = m K |K0 . KL indeed decays much more often to three than to two pions. we write it as a linear combination of eigenstates of the strong interaction hamiltonian Hs ¯ | (t) = a(t)|K0 + a(t)|K0 + ¯ n bn (t)|n (1. respectively. It parametrizes the decay rates of the kaons to other strong interaction eigenstates. The states |n stand for all the eigenstates of Hs which are ¯ different from |K0 and |K0 . (1. because the coefﬁcients bn (t) have been removed from the evolution equation. † Eqs. the parameter ε would vanish.1. . = † ) (1. Hs |n = E n |n (1. while P stands for the principal value. ¯ ¯ Hs |K0 = m K |K0 . In such a case. This is the reason why the absorptive part i j arises in Hi j .261) Mi j = m K δi j + i|Hw | j + P m K − En n ij = 2π n 3 i|Hw |n n|Hw | j δ(m K − E n ) + O(Hw ) (1. Landau & Lifshitz. we ﬁnd the time evolution equation for a(t) and a(t) ¯ i i d dt a(t) a(t) ¯ = H11 H21 H12 H22 a(t) a(t) ¯ (1. for example. This explains its long lifetime. the sum over n denotes the sum-integral over all states different from |K0 ≡ ¯ |1 and |K0 ≡ |2 . For t > t0 .261) and (1. Solving this equation perturbatively in Hw .259) expressed in terms of matrix elements of Hw between strong interaction eigenstates: i (M = M † . The state | (t) satisﬁes the Schr¨ dinger equation o d (1. because decays to three pions are strongly phase-space suppressed. In reality. Let us consider a state which at the initial time t0 is a neutral kaon at rest.

66 1 Classical ﬁelds. m L.265) Here. i. masses and decay widths of the physical neutral kaons KL and KS .S − i 2 L. Although the latter equality is violated in Nature. The considered state | (t) can now be written as follows: | (t) = cL exp −im L t − 1 2 Lt 1 2 |KL St + cS exp −im S t − |KS + n bn (t)|n (1.S are. while kaon decays can proceed via only single W ± boson exchange. too.† Once we know the eigenbasis of Hi j .S and L. it is usually convenient to verify what happens in this limit with the relations we shall discuss below. The eigenvalues of the matrix Hi j are λL.259) is trivial. then the off-diagonal elements H12 and H21 would be equal. solving (1. 12. The observable we would like to consider is the ratio of decay amplitudes of KL and KS into two-pion states with vanishing total isospin εK = A[KL → (π π ) I =0 ] A[KS → (π π ) I =0 ] (1. H11 = H22 ≡ H (1.S = H ± H12 H21 1/2 1/2 (1.263) If electroweak interactions were symmetric under C P.17). . we have assumed that H21 tends continuously to H12 when H21 tends to H12 in the limit of no C P violation.255). This is because the leading contribution to i|Hw | j originates from double W ± boson exchange (Fig. respectively.266) with some time-independent coefﬁcients cL and cS .267) † This may not hold only in some really peculiar phase conventions for the parametrization of the Cabibbo– Kobayashi–Maskawa matrix and/or for the complex square root. The C P T symmetry of electroweak interactions implies that the diagonal elements of the matrix Hi j are equal.254) and (1. The parameter ε which occurs in these equations is related to ¯ the matrix Hi j as follows: ε= ¯ 1/2 H12 − H21 H12 + H21 1/2 1/2 1/2 1/2 .S = m L. symmetries and their breaking 2 It is important to note that the O(Hw ) and O(Hw ) parts of Hi j are actually of the same order in the electroweak coupling constants.e. respectively. Thus. 1/2 (1.264) while the corresponding eigenvectors |KL and |KS are given in (1.

respectively. In the .265). 12 is approximately ¯ proportional to A∗ A0 = A∗2 .270) which can be equivalently rewritten as H12 A2 − H21 A∗2 0 0 2 (1 − εK ) 1/2 1/2 2 [2(m L − m S ) − i( L − S )]|A0 |2 4H12 H21 |A0 | (1. Consequently. Therefore.260) of Hi j into hermitean and antihermitean parts Mi j and i j . Now. 0 0 0 εK Im(M12 A2 ) 0 exp(iπ/4). Next.273) In general. Consequently. We know from experiment that εK 1 and L 2(m L − m S ) ≡ 2 m. The C P T symmetry implies that A0 = A∗ . it is time to make some approximations. the phase of the amplitude A0 can take any value between 0 and 2π.272) where we have used the decomposition (1. This means that the sum representing i j in (1. depending on phase conventions in the parametrization of the CKM matrix.271) The advantage of the latter formula is that it is manifestly independent of sign conventions used for complex square roots and of phase conventions in the parametrization of the Cabibbo–Kobayashi–Maskawa (CKM) matrix. Using this fact together with 0 (1. we take into account the experimental fact that KS decays dominantly to isospin-zero two-pion states.262) is dominated by (ππ) I =0 intermediate states. √ 2 m|A0 |2 (1.269) (1.269).254). (1. (1.268) where δ0 is the strong phase shift in elastic scattering of two pions in the isospin ¯ zero state. S εK = 2 (1 − εK ) = H12 A2 − H21 A∗2 0 0 εK 2i Im(M12 A2 ) + Im( 12 A2 ) 0 0 .5 Discrete symmetries 67 where 1 |(ππ) I =0 = √ |π + π − + |π − π + − |π 0 π 0 3 It is convenient to express εK in terms of the decay amplitudes A[K0 → (π π ) I =0 ] ≡ A0 exp(iδ0 ) ¯ ¯ A[K0 → (π π ) I =0 ] ≡ A0 exp(iδ0 ) (1. 2(1 + i) m|A0 |2 (1.255).267) and (1. one ﬁnds after a short calculation that εK = H12 A0 − H21 A∗ 0 1/2 1/2 1/2 1/2 H12 A0 + H21 A∗ 0 (1.1. and Im( 12 A2 ) can be neglected. (1.

the quantity Im(M12 A2 ) is a 0 0 measure of C P violation in the neutral kaon mixing. momentum p and orbital momentum L = r×p are constants of motion if the potential is symmetric under translations and rotations.265). when C P is violated.274) 2 m The equality of ε and εK for real A0 follows directly from comparing (1.275) However. Riazuddin & Ryan (1969). corresponding to the transformation t = t + εt 2 . However. As we have observed before.255) are not C P eigenstates. . unlike εK . the value of ε ¯ is bounded from below when C P is violated.273). When it is ¯ small. if εK differs from zero. Prove that energy E. On the other hand. ε does not even need to be small. ¯ The value of εK is measured quite precisely nowadays. the parameter ε is. dependent on phase conventions. we cannot neglect Im 12 before specifying the phase convention. Problems 1. On the other hand.270) and ¯ (1.254) and (1.272) εK 2i Im M12 + Im 2(1 + i) m 12 (1. symmetries and their breaking standard phase convention (see Chapter 12). and we can write Im M12 ε εK √ ¯ exp(iπ/4). 1.68 1 Classical ﬁelds. t1 . we can write in analogy to (1.2 Write down the Newton equation d2 x/dt 2 + ∂ V /∂ x = 0 in the new reference frame. in general. t2 ) = t2 dt t1 1 dx m 2 dt − V (x) where [x] denotes a functional of the path x(t). In other words. ¯ Therefore ε is not an observable. and it is of order 10−3 (for the description of some of the methods of measuring εK see Marshak. then the mass eigenstates KL and KS in (1. A0 is approximately real. (1. as follows from (1. Construct the lagrangians which give both equations and compare with (1. Calculate the variation S under the transformation x(t) → x(t) + δx(t) ∇ Derive the Newton equation m d2 x/dt 2 = −∇ V from the principle of the minimal action (subject to the constraint δx(t1 ) = δx(t2 ) = 0). Thus.27).1 Consider the action for a motion of a particle in a potential V (x): S([x]. then either Im M12 or Im 12 must be non-vanishing. Consequently. the phase of 12 is approximately opposite to the phase of A2 .

3 Use (1. Thus gauge invariance leads to no new conservation laws as compared to global invariance.94) is S µ = ∂ρ (F ρµ (x)) and that the associated conserved charge Qs = d3 x S 0 (x.4 Consider a closed system of a charged classical particle interacting with an electromagnetic ﬁeld.5 Show that for an abelian gauge theory the Noether current corresponding to the gauge symmetry transformation (1. ˜ 1.35) to study conservation laws in classical electrodynamics (in the absence of sources) with L = − 1 F µν Fµν . 1. Express µν in terms of the electric E and magnetic B ﬁelds. In particular show that the energy density is 00 = 1 (E2 + B2 ) and the momentum 2 density 0i = (E×B)i is the Poynting vector. where F µν = ∂ µ Aν − ∂ ν Aµ . t) either vanishes or reduces to the electric charge associated with global U (1) invariance provided approaches an angle-independent limit at spatial inﬁnity.6 In abelian gauge theory the dual ﬁeld strength tensor F µν is deﬁned as ˜ F µν = 1 εµνρσ Fρσ . Derive the canonical 4 energy–momentum tensor and bring into the symmetric and gauge-invariant form µν ν = F µα Fα + 1 g µν Fαβ F αβ 4 Check that it is traceless. Finally. derive the angular momentum vector for the electromagnetic ﬁeld and split it into orbital and intrinsic orbital momentum parts. 2 ˜ Show that Fµν F µν is a total divergence ˜ F µν Fµν = ∂µ K µ where K µ = εµνρσ Aν Fρσ In a non-abelian case a matrix-valued dual ﬁeld strength is ˜ G µν = 1 εµνρσ G ρσ 2 and ˜ Tr[G µν G µν ] = ∂µ K µ where K ρ = ερσ µν Tr[G σ µ Aν + 2 Aσ Aµ Aν ] 3 ε0123 = −1 . The conservation law ∂0 00 +∂i 0i = 0 connects the change in the energy density to the Poynting vector. Write down the action in terms of the electromagnetic potential Aµ and derive the equations of motion and the energy–momentum conservation laws. 1.91) and (1.Problems 69 1.

Using Noether’s theorem derive the conserved symmetry current: a a Jµ = jµ + cabc Ab G cν ν µ a a where jµ is given by (1. [Dµ . provided Yja falls off sufﬁciently rapidly at large |x|. Dρ ]] + [Dν . t) is time-independent. the fall-off requirement restricts the large |x| behaviour of gauge transformation. Since Y ja is not gauge-invariant. Dµ ]] + [Dρ . symmetries and their breaking 1. [Dν . Dν ]] = 0 and the Bianchi identity Dµ G νσ + Dν G σ µ + Dσ G µν = 0 or ˜ Dµ G µν = 0 1.124). Show that the ordinary conservation of Jµ is equivalent a to the covariant conservation of jµ and the gauge ﬁeld equation (1. Show that the charge Qa = a d3 x Y j0 (x. Show that the current Y j generates no new charges provided approaches an angle-independent limit at spatial inﬁnity. Show that µν ν ∂µ G a = g Ja .7 Check the Jacobi identity for the covariant derivatives [Dµ . Show that Q a is gauge-covariant against gauge transformations which µ approach a deﬁnite angle-independent limit as |x| → ∞.123).70 1 Classical ﬁelds. [Dρ .8 In pure Yang–Mills theory the Noether current for global symmetry transformations is Y a jµ = cabc Ab G cν ν µ 2 Tr[G νµ Dν ] g2 and for gauge transformations Y µ j = Using ﬁeld equations Dµ G µν = 0 show that both currents are conserved. Include fermion ﬁelds.

t = x | exp[−(i/h )H (t − t)]|x ¯ (2. with the integrand dependent on the action integral. t ¯ where H denotes the hamiltonian of the system. the need for integration is obvious: the position operator does not commute with the momentum or the hamiltonian. and is a Green’s function: deﬁne |t by H |t = ih (∂/∂t)|t . t |x. Consequently. according to the superposition principle. Intuitively. t X S |x = x|x with the relation |x = exp[−(i/h )H t]|x.1 Path integrals in quantum mechanics Our ﬁrst aim is to show that the matrix elements of quantum mechanical operators can be written as functional (path) integrals over all trajectories. time evolution changes the position eigenstate into one in which position is not determined. The matrix element x . ¯ then x|t = dx x| exp[−(i/h )H (t − t)]|x ¯ 71 x |t . The eigenstates of the position operator are introduced as follows X H (t)|x.1) Heisenberg picture Schr¨ dinger picture o corresponds to the transition from the eigenstate |x at the moment of time t to the state |x at the time t . The quantum system has no deﬁnite trajectory and it is necessary to take a sum over all possible ones. t = x|x.2 Path integral formulation of quantum ﬁeld theory 2. Transition matrix elements as path integrals Consider a quantum mechanical system with one degree of freedom.

x j−1 )(t j − t j−1 )] (2. . x .4) where H ( p. t |x. t j | = 1 together with i x j . tn+1 are to be understood as x.72 2 Path integrals The matrix element (2.3) where x0 . xn+1 . (2. t j |x j−1 . .4). we use the completeness relation at each of the times t j : dx j |x j . t0 . Using (2. respectively. . t . x) is now the classical c-number hamiltonian.1): n n+1 x . x j−1 ) + O ε 2 exp h h 2π h ¯ ¯ ¯ (2. x j−1 ) 2π h h ¯ ¯ (2. t j−1 = dpj i i p j (x j − x j−1 ) 1 − ε H ( p j . X ) to be of the form H = f (P) + g(X ) we can write x j |H |x j−1 = = d p j x j | p j p j |H |x j−1 dpj i exp p j (x j − x j−1 ) H ( p j . t j−1 = x j exp − ε H x j−1 h ¯ iε x j |H |x j−1 + O ε 2 = x j |x j−1 − h ¯ (2. t. Choosing the hamiltonian H = H (P. . t j |x j−1 .2) (2.5) and we obtain the following expression for the matrix element (2.1) we shall ﬁrst represent as a multiple integral which will then be used to deﬁne the functional integral by a limiting procedure. x j−1 ) + O ε2 exp h h 2π h ¯ ¯ ¯ dpj i i = p j (x j − x j−1 ) − ε H ( p j .6) . t j x j . n) Next.3) becomes x j . t = lim n→∞ dx j j=1 dpj i exp h 2π h ¯ ¯ j=1 n+1 [ p j (x j − x j−1 ) j=1 −H ( p j . First we divide the time interval (t − t) into (n + 1) equal parts of length ε t = (n + 1)ε + t t j = jε + t ( j = 1.

8) we have derived path integral representation (2. however. Shifting the integration variables: p j → p j − m( x j /ε) we obtain p2 dpj i j exp ε pj xj − 2π h 2m h ¯ ¯ where x j = x j − x j−1 and 1 = Nj d p j exp − i p2 j ε h 2m ¯ = 1 1 i m exp ε N j 2π h h 2 ¯ ¯ xj ε 2 (2. for instance. theories where the lagrangian is L= 1 2 f (x)x 2 + g(x)x − V (x) ˙ ˙ (2. t . x(t ) = x implied in this case. t = t 1 N Dx i S[x] exp h 2π h ¯ ¯ (2. (2. namely. x)] dτ ˙ t (2. x) = 2 m x − V (x) is the Lagrange function and the normalization factor N ˙ ˙ is given by 1 = N D p exp − i h ¯ t t p dτ 2m Starting with the canonically quantized theory described by hamiltonian (2. t |x. to deﬁne the quantum theory by functional integral (2.7) (Dx D p/2π h ) ≡ ¯ ¯ τ [dx(τ ) d p(τ )/2π h ] is called a functional integration over all phase space.8).11) . x) dτ is the action integral over the trajectory x(τ ).10). t |x.10) Here S[x] = t L(x. systems for which canonical quantization is ambiguous due to non-commutativity of operators P and X .9) The ﬁnal result has the form of a functional integral over conﬁguration space: x . Here belong. If the hamiltonian is of a simple form H = (1/2m)P 2 + V (X ) (2. where ˙ 1 2 L(x.10).1 Path integrals in quantum mechanics 73 where the limit n → ∞ (ε → 0) has been taken and the O ε 2 terms neglected.7) is the promised path integral representation of x .2. with the boundary conditions x(t) = x. t |x. t = i Dx D p exp 2π h h ¯ ¯ t [ p x − H ( p. we can choose the path integral formulation as the quantization prescription for a system with the classical hamiltonian in the form (2. i. There are. Eq.8). This result we shall write in the compact form x .6).e. Then our derivation proves the equivalence of the path integral and canonical quantization methods for systems described by hamiltonian (2. We can use another approach.8) it is convenient to perform the momentum integrations in (2.

in fact. if of interest. [ f (x) p]S = 1 [ f (x) p + p f (x)].11). A lagrangian of this kind appears. can be done at least in the framework of perturbation theory. The modiﬁcation of the functional measure Dx into Dx = lim comes from the integral 1 = Nj d p j exp − i −1 f (x j ) p 2 ε j 2h ¯ i→∞ dxi f 1/2 (xi ) i which now depends on x j . The functional integral (2. which in the classical limit gives the theory (2. ¯ The integral is then dominated by trajectories close to the classical one xc (t). in the following we shall consider quantum theories deﬁned by the path integral formulation. 2 Theories like (2. must be veriﬁed case by case.74 2 Path integrals On the other hand.12) (see Problem 2. supplemented by the rule that. the integrals over p j are to be performed before the x-integration. . Their correspondence to canonically quantized theories.e. Therefore L= m 2 1+ dc dx 2 x 2 − V (x.13) ˙ ˙ 2 2 subject to a constraint equation y = c(x). t ∼ lim j→∞ tj dx(t j ) 1/2 i S[x(t j )] f (x(t j )) exp 2π h h ¯ ¯ (2.11). for instance. In summary. t |x.11) are of physical interest.14) We shall encounter a similar situation in Chapter 3 when quantizing gauge ﬁeld theories. which satisﬁes δS[xc (τ )] = 0: i. for example.1).10) by use of the saddle point approximation. For ﬁeld theories considered in this book this. Integrating over p j one derives (2. It can be shown (Cheng 1972) that the appropriately generalized path integral reads x . is unambiguously deﬁned by the path integral ansatz in conﬁguration space. insensitive to the ordering of P and X . y) (2. namely the symmetric ordering. whenever there is an ambiguity.7) with the c-number hamiltonian corresponding to lagrangian (2.12) deﬁnes the same dynamics as the Schr¨ dinger equation with one particular ordering of the operators P and o X . one can postulate (2. c(x)) ˙ (2.12) Equivalently. for a system of two interacting particles m m L = x 2 + y 2 − V (x. If S[x(τ )] h we can evaluate (2. quantum theory. the path integral formulation allows a relatively simple understanding of the classical and semi-classical limit.

Also.h.6) we obtain (2. t .16) corresponds to the matrix element x .2. x(t N ) 2π h ¯ i t [ p x − H ( p. t |xn . t . . x)] dτ ˙ × exp h t ¯ (2. t = Dx D p x(t1 ) .17) As before.15). t J = i Dx D p exp h 2π h ¯ ¯ t t [ p x − H ( p.15) for the product of two operators: X (τ1 )X (τ2 ) at τ1 > τ2 . t |x. t |X (t2 )X (t1 )|x. τ2 = ti2 and apply the completeness relation at each ti . Note that (2. . . . t |X (τ2 )X (τ1 )|x. t |xn . t1 > t 2 = x . For the time-ordered product of N such operators the following expression can be shown to hold x .15) Let us check (2. We obtain x . let us note that the transition amplitude in the presence of an external source J (τ ) x . ti1 −1 · · · × xi2 .16) is true for τ1 > τ2 . t1 |x. Therefore the path integral. t (2. X (t N )|x.15). t |T X (t1 ) . tn · · · x1 . like (2. deﬁnes the matrix element of the time-ordered product of the position operators Dx D p i t x(t1 ) x(t2 ) exp [ p x − H ] dτ ˙ h t 2π h ¯ ¯ x . tn · · · xi1 . t1 < t 2 (2. corresponding to the ﬁeld operator of quantum ﬁeld theory. of (2. ti2 −1 · · · x1 . t = i dxi x .16) Proceeding exactly as when deriving (2. Again we divide the time axis into small intervals. ti1 |X (τ1 )|xi1 −1 . t = i dxi xi1 xi2 x . ti2 |X (τ2 )|xi2 −1 . t |x. it is also possible to change from phase space path integrals to path integrals over conﬁguration space. . t |X (t1 )X (t2 )|x. choosing t1 . t |X (τ1 )X (τ2 )|x. . the r. t1 |x. tn in such a way that τ1 = ti1 .1 Path integrals in quantum mechanics 75 Matrix elements of position operators The matrix element x . In view of further applications of functional formalism to quantum ﬁeld theories it is also important to know the path integral representation of the matrix elements of the position operators. t determines all transition probabilities between quantum mechanical states.18) . t .s. When τ1 < τ2 . x) + h J (τ )x(τ )] dτ ˙ ¯ (2.

. . t |x.23) J =0 which compares with the Taylor expansion of the usual functions. X (t N )|x. qn ) J (q1 ) . .20) the functional derivative with respect to J is simply ( f can be chosen to be symmetric in all variables) δ F[J ] = δ J (q) dq1 . t|0 ¯ ¯ (2. the ground state or the vacuum is described by the function 0 (x) = x|0 . . which are then given by its functional derivatives with respect to J (τ ) x . . . t J (2. The energy eigenstates correspond to the wave-functions n (x) = x|n . can be used as a generating functional of the ¯ matrix elements of the position operators. . . qn ) = δ n [J ] δ J (q1 ) . . .76 2 Path integrals corresponding to the usual transition amplitude with the hamiltonian modiﬁed by a source term: H → H − h J x. t) deﬁned as 0 (x. Assume that the lagrangian L of the system is time-independent. t) = exp[−(i/h )E 0 t] x|0 = x| exp[−(i/h )H t]|0 = x. . . . dqn φn (q1 . t = 1 i N δN δ J (t1 ) . . J (qn−1 ) n f (q1 . . . J (qn ) (2. . . . . qn ) J (q1 ) . .22) (2. . . δ J (t N ) J =0 x . 2.19) Instead of a formal deﬁnition of the functional derivative. . for our purpose it is sufﬁcient to know that.21) If the This corresponds to the usual rule of differentiating the monomials. dqn f (q1 . functional is deﬁned instead by the series [J ] = then φn (q1 . qn−1 q) (2. t |T X (t1 ) . . .2 Vacuum-to-vacuum transitions and the imaginary time formalism General discussion In ﬁeld theoretical applications we shall mainly deal with the matrix elements of the ﬁeld operators taken between the vacuum states: the Green’s functions. δ J (qn ) ∞ n=1 1 n! dq1 .24) . dqn−1 J (q1 ) . In particular. in the case of the functional F[J ] deﬁned as F[J ] = dq1 . J (qn ) (2. It will be convenient to use 0 (x. Let us ﬁrst consider the analogous problem in quantum mechanics.

t ) x . t 0 (x. . It is very important that the generating functional W [J ] can also be derived in another way. The considered vacuum expectation value can also be obtained from a generating functional 0|T X (t1 ) . t x .25) and for the matrix element x . t|x1 . . T2 |x1 . t|x1 . t the functional form (2. t |x. . t J given by (2. t ) x . The only T -dependent terms are now the factors exp[−(i/h )E n (t − T1 )] and we can continue to T1 → +i∞ explicitly. T2 |x . Recalling ¯ that E 0 is the lowest energy eigenvalue we get T1 →+i∞ lim exp[−(i/h )E 0 T1 ] x. .2.26) = dx dx ∗ 0 (x 0 (x. T2 |x . X (tn )|x. To derive (2. T2 |x1 . T2 |x . t |x. T1 J at any given x1 . Then we can write x2 . t) ∗ 0 (x 1 ) ∗ 0 (x 1 ) and. T1 (2.2 Vacuum-to-vacuum transitions 77 We are interested in the matrix element 0|T X (t1 ) . t ) with T2 > t > t > T1 . .15) can be used. t|x1 .28) let us choose that the source J (t) vanishes outside the time interval (t. . t = ¯ ∗ 0 (x .t ) 0 (x 2 ) . in fact. t J W [J ] J =0 (2. determined by the transition amplitude x2 . x2 (for instance x1 = x2 = 0) provided that the analytic continuation to the imaginary values of T1 and T2 is performed. T1 = ¯ = ¯ 0 (x) exp[−(i/h )E 0 t] 0 (x. . . δ J (t N ) . T1 = x| exp[−(i/h )H (t − T1 )]|x1 ¯ = n n (x) ∗ ¯ n (x 1 ) exp[−(i/h )E n (t J = dx dx x2 . t |T X (t1 ) .27) with x . X (t N )|0 = where W [J ] = 0|0 J 1 i N δN δ J (t1 ) . X (tn )|x. t) (2. t |x. t |T X (t1 ) . . . in the same way T2 →−i∞ lim exp[(i/h )E 0 T2 ] x2 . t . X (tn )|0 . It reads 0|T X (t1 ) .29) − T1 )] and similarly for x2 . T1 where x. T2 |x1 . t J x. . . t) (2.18).28) This implies that W [J ] is. T1 ∗ (x1 ) 0 (x2 ) T1 →+i∞ 0 lim T2 →−i∞ J (2. We shall show that W [J ] = exp[(i/h )E 0 (T2 − T1 )] ¯ x2 . X (tn )|0 = dx dx ∗ 0 (x .

t ) x(t ) = T2 T1 −x x − ω2 x 2 dt ¨ (2. Harmonic oscillator As an example of particular interest in view of further ﬁeld theoretical applications we shall consider the case of the harmonic oscillator L = 1 x 2 − 1 ω2 x 2 = 1 (d/dt)(x x) − 1 x x − 1 ω2 x 2 ˙ ˙ ¨ 2 2 2 2 2 The action integral. δ J (t N ) T2 T1 W [J ] (2.32) where x1 . because we can always consider quantities like 1 0|T X (t1 ) . .29).33) where ( f. Notice also that (2. X (t N )|0 = exp[(i/h )E 0 (T2 − T1 )] ¯ ∗ T1 →+i∞ 0 (x 1 ) 0 (x 2 ) lim (2. . . . x) ¯ 2 (2. x2 . x) + h J x] dt ˙ ¯ (2. The J -independent factors are irrelevant. Ax) + (h J.28) follows. . . .30) T2 →−i∞ × x2 .34) so that A(t.28) we can write W [J ] = T1 →+i∞ T2 →−i∞ lim x(T1 )=x1 x(T2 )=x2 Dx exp (i/h ) ¯ [L(x. g) = T2 T1 dt f (t) g(t) and the operator A is deﬁned by the equation T2 T1 dt dt x(t) A(t.35) . X (t N )|x1 .28) and (2. . given by (2. g) denotes the ‘scalar product’ ( f. including the source term. T2 |T X (t1 ) . T1 We conclude that the vacuum matrix elements can be calculated by taking functional derivatives of the generating functional W [J ]. instead of (2.26) imply 0|T X (t1 ) .31) J ≡0 Consequently. can be written as follows S J [x] = T2 T1 [L + h J x] dt = 1 x x ˙ ¯ 2 T2 T1 + 1 (x. t ) = − d2 dt 2 + ω2 δ(t − t) (2. are arbitrary.27) and (2.28). X (t N )|0 = 0|0 1 i N 1 δN W [0] δ J (t1 ) . (2.78 2 Path integrals After employing (2.

33) in terms of the classical trajectory and the new integration variable z(t) we get S J [xc + z] = 1 xc xc ˙ 2 T2 T1 + 1 (z. xc (T2 ) = x2 .36). we have z(T1 ) = z(T2 ) = 0.2. x0 ) ¯ × L(z) dt T1 (2.37) so that G = −A−1 . G J ) 2 2¯ 2¯ (xc .32) by introducing z(t) x(t) = xc (t) + z(t) where xc (t) is a solution of the classical equation of motion (in the presence of the source) x + ω2 x = h J ¨ ¯ Requiring that xc (t) satisﬁes the boundary conditions xc (T1 ) = x1 . The classical trajectory xc (t) can be written as a superposition of a solution x0 (t) of the homogeneous equation. . x0 ) + 1 h 2 (J.39) z(T1 )=0 z(T2 )=0 The important observation is that if the Green’s function (2.† Re-expressing the action integral (2.40) † In the space of functions x(t) with zero modes x0 (t) of A excluded. The resulting W [J ] is of the following form W [J ] = T1 →+i∞ T2 →−i∞ lim exp (i/2h )h 2 (J. G J ) exp (i/2h ) xc xc ˙ ¯ ¯ ¯ Dz exp (i/h ) ¯ T2 T2 T1 + h (J.2 Vacuum-to-vacuum transitions 79 and Ax ≡ T2 T1 dt A(t. t ) x(t ) Let us now change the functional integration variable in (2.36) Here G(t − t ) is the Green’s function of the classical equation of motion d2 dt 2 + ω2 G(t − t ) = δ(t − t ) (2. ¯ together with (2. t→±i∞ dG(t)/dt −→ 0 t→±i∞ (2. and the special solution of the complete equation xc (t) = x0 (t) + h ¯ dt G(t − t ) J (t ) (2. z) − z xc ˙ T2 T1 (2. Az) = (Axc .38) where we have used the relations Axc = −h J. Az) + 1 h (J.37) is chosen to satisfy the conditions G(t) −→ 0.

80

2 Path integrals

**the only functional dependence of W [J ] on J is given by the factor W [J ] ∼ exp (i/2h )h 2 (J, G J ) ¯ ¯
**

T

(2.41)

To prove this assertion we should show that xc xc T2 and (J, x0 ) are J -independent ˙ 1 whenever the boundary conditions (2.40) are satisﬁed. In fact (2.36) and (2.40) imply xc xc ˙

T2 T −→ x x 2 ˙ T1 T →+i∞ 0 0 T1 1 T2 →−i∞

with x0 obviously J -independent. For (J, x0 ) the argument is somewhat more involved. Consider the solution of the homogeneous equation x0 (t) = A exp(iωt) + B exp(−iωt) with A, B chosen to satisfy the boundary conditions x0 (T1 ) = x1 , x0 (T2 ) = x2 x2 exp(−iωT1 ) − x1 exp(−iωT2 ) , exp[iω(T2 − T1 )] − exp[−iω(T2 − T1 )] x1 exp(iωT2 ) − x2 exp(iωT1 ) B= exp[iω(T2 − T1 )] − exp[−iω(T2 − T1 )] A= In the limit T1 → +i∞, T2 → −i∞ we have A ≈ x2 exp(−ω|T2 |), B ≈ x1 exp(−ω|T1 |)

implying that (J, x0 ) also vanishes in this limit. Note that in (2.41) the limit T1 → +i∞, T2 → −i∞ has in fact been omitted. This could be done because the integral

T2 T2

dt

T1 T1

dt J (t) G(t − t ) J (t ) ≡ (J, G J )

is independent of the limits of integration due to the requirement that J (t) vanishes outside some ﬁnite time interval (t1 , t2 ) such that T2 > t2 > t1 > T1 . The dependence on T1 and T2 can now occur only in the factors which are J -independent because of the boundary conditions (2.40), and therefore irrelevant for our discussion. To complete the case of the harmonic oscillator we must ﬁnd the Green’s function G(t − t ) consistent with the boundary conditions speciﬁed by (2.40). ˜ Introducing the Fourier transform G(ν) G(t) =

∞ −∞

dν ˜ G(ν) exp(−iνt) 2π

2.2 Vacuum-to-vacuum transitions

81

Fig. 2.1. Possible prescriptions for bypassing the singularity at ν = ±ω in the integral (2.42) in the complex ν plane.

**we obtain the formal solution of (2.37) G(t) = −
**

∞ −∞

1 dν exp(−iνt) 2π ν 2 − ω2

(2.42)

This result represents, in fact, four distinct solutions depending on the prescription for bypassing the singularity at ν = ±ω, which may be one of those shown in Fig. 2.1. Cases (a) and (b) are easily seen to be in conﬂict with the boundary conditions (2.40). It sufﬁces to perform the analytic continuation of (2.42) by deforming the contour as shown in the picture and then to integrate for imaginary t. The two remaining solutions agree with (2.40): we again continue analytically to imaginary t by deforming the contour of the ν-integration from the real to the imaginary axis. They correspond to adding to the denominator of the integrand in (2.42) (+iε) and (−iε), respectively G(t) = −

∞ −∞

dν 1 exp(−iνt) 2 − ω2 ± iε 2π ν

(2.43)

Adding (+iε) to the Green’s function denominator is equivalent to introducing the extra term ±iεx 2 to the lagrangian function. Only for +iεx 2 is the functional integral well deﬁned, so that the ﬁnal solution for the Green’s function is speciﬁed uniquely ∞ 1 dν G(t) = − exp(−iνt) (2.44) 2π ν 2 − ω2 + iε −∞ After the contour integration this becomes (for real t) i [exp(−iωt) (t) + exp(iωt) (−t)] (2.45) 2ω The physical interpretation of the (+iε) prescription is thus made clear: it corresponds to the positive frequency solutions propagating into the future. G(t) =

82

2 Path integrals

Fig. 2.2. Rotation of the integration contour from the real to the imaginary axis in the complex ν-plane which deﬁnes the analytic continuation of G E (τ ) to τ = it.

Euclidean Green’s functions Now let us observe that the same Green’s function can also be derived in another way, i.e. by the analytic continuation from the imaginary time region. To this end we ﬁrst deﬁne the ‘Euclidean’ (i.e. imaginary time) Green’s function G E (τ ), which obeys the equation − d2 dτ 2 + ω2 G E (τ ) = δ(τ ) Consequently, G E (τ ) =

∞ −∞

(2.46)

dν 1 exp(−iντ ) 2 + ω2 2π ν

(2.47)

As the singularities of the integrand appear at ν = ±iω, away from the integration contour, G E (τ ) is uniquely speciﬁed to be G E (τ ) = − 1 exp(−ω|τ |) 2ω (2.48)

in agreement with the boundary conditions speciﬁed for G(t). We shall now show that the real time Green’s function G(t) is given by the analytic continuation of G E (τ ) to τ = it G(t) = −iG E (it) (2.49) In order to perform this continuation the integration contour in (2.47) should be deformed to keep the integral convergent during the τ = it; such a rotation is illustrated in Fig. 2.2. The result is G(t) = −

∞ −∞

dν 1 exp(−iνt) 2 − ω2 + iε 2π ν

which coincides with (2.44). It is also possible to continue to τ = −it. In this case, however, the resulting real

2.3 Path integral formulation of quantum ﬁeld theory

83

time Green’s function appears with the opposite, (−iε), prescription. We conclude that the analytic continuation is correct for τ = it. One way to see the difference is to make the substitution t = ∓iτ under the functional integral. The action changes as follows: i i S[x] = h h ¯ ¯ dt 1 dx 2 dt

2

− V (x) −→

t=±iτ

i (±i) h ¯

dτ −

1 dx 2 dt

2

− V (x)

1 (2.50) = ± SE [x] h ¯ where SE [x] is the ‘Euclidean’ action. The resulting functional integral is of the form Dx exp{±(1/h )SE [x]} ¯ and only for the (−) sign is it well deﬁned. For the generating functional WE [J ] one gets WE [J ] = Dx exp −(1/h )SE [x] + ¯ dτ J (τ ) x(τ ) ∼ exp

1 h (J, G E J ) 2¯

(2.51)

where G E (τ ) is deﬁned by (2.46). 2.3 Path integral formulation of quantum ﬁeld theory Green’s functions as path integrals The results of Sections 2.1 and 2.2 are easily generalized to the case of more than one degree of freedom. If the number of degrees of freedom is to be n, the coordinate x should be replaced by an n-component vector. The functional integral would now correspond to the sum over all trajectories in the n-dimensional conﬁguration space, satisfying appropriate boundary conditions. In the ﬁeld theory, the trajectory x(t) is replaced by a ﬁeld function (x, t). The degrees of freedom are now labelled by the continuous index x; the number of degrees of freedom is obviously inﬁnite. To deﬁne the appropriate path integral one can start from a multiple integral on a discrete and, for a beginning, ﬁnite lattice of space-time points. This amounts to deﬁning the quantum ﬁeld theory as a limit of a theory with only a ﬁnite number of degrees of freedom. The limit of an inﬁnite lattice, related to the thermodynamical limit of statistical mechanics, already deﬁnes a theory with an inﬁnite number of degrees of freedom. However, this lattice theory does not have enough space-time invariance and a continuous theory must be deﬁned. The latter limit is accompanied by inﬁnities, the ‘UV divergences’ of quantum ﬁeld theory. The deﬁnition of the functional integral in quantum ﬁeld theory is thus more ambiguous than in the case of quantum

84

2 Path integrals

mechanics. Nevertheless, the functional formalism in quantum ﬁeld theory is of great heuristic value. It is a very convenient tool for studying perturbation theory and allows a natural description of some non-perturbative phenomena. The quantum ﬁeld theory is usually formulated in terms of the vacuum expectation values of the chronologically ordered products of the ﬁeld operators, the Green’s functions: G (n) (x1 , . . . , xn ) = 0|T (x1 ) . . . (xn )|0 (2.52)

Using our experience from the previous section we shall write down the path integral representation for them. In particular, it is important to remember the role played by (2.28) and (2.30) in getting rid of the vacuum wave functions originally present in (2.25) as the boundary conditions. Thus, by analogy with the results of the previous section, we postulate the following path integral representation G (n) (x1 , . . . , xn ) ∼ D (x1 ) . . . (xn ) exp (i/h ) ¯ d4 x L (2.53)

(D denotes integration over all functions (x, t) of space and time, because, for each value of x, (x, t) corresponds to a separate degree of freedom; L is the lagrangian density). Eq. (2.30) suggests that any boundary conditions are, in fact, irrelevant here, provided that we take the imaginary time limit. In particular, the trajectories may be not constrained at all. In analogy with (2.30), the rotation of the time contour away from the real axis should start somewhere at large absolute values of time, so that the arguments of the Green’s functions, t1 . . . tn , stay real. If instead the whole time axis is rotated, ti = −iτi , the result is a Euclidean Green’s function (see the end of Section 2.2). The latter has a particularly convenient path integral representation, because the weight factor in the integrand, exp(−SE /h ), is then non-negative. This Euclidean path ¯ integral formalism can be used to deﬁne the Minkowski space Green’s functions by an analytic continuation of the Euclidean ones. An expression like (2.53) would then be understood as an analytic continuation in the variables t1 , . . . , tn of the analogous Euclidean formula. Equivalently one can work in Minkowski space and use the (+iε) prescription (see (2.62)). Eq. (2.53) should be regarded as the formulation of the theory. What is the relation between the path integral and the usual (canonical) operator formulation of quantum ﬁeld theory based on the same lagrangian L? For our purposes these are equivalent if they imply the same perturbation theory Feynman rules. This has to be checked in each case of interest. However, the derivation of the perturbation theory rules is much simpler in the functional framework, particularly in the case of gauge ﬁeld theories. In the following we shall often use the path integral formulation,

2.3 Path integral formulation of quantum ﬁeld theory

85

referring to the operator formalism only if it helps to make the presentation more concise, like in the case of the scattering operator discussed in Section 2.7. It is convenient to normalize the Green’s functions by factorizing out the vacuum amplitude G (n) (x1 , . . . , xn ) = 0|T (x1 ) . . . (xn )|0 / 0|0 =N where 1/N = D exp (i/h ) ¯ d4 x L ∼ 0|0 (2.55) D (x1 ) . . . (xn ) exp (i/h ) ¯ d4 x L (2.54)

This removes the extra factors like those appearing in (2.30). The T -product deﬁned by (2.54) is a covariant quantity (see Section 10.1 on ambiguities of T -products). The derivatives of the Green’s functions are functional integrals of derivatives of ﬁelds, e.g.

xG (2)

(x, y) ∼

D

x

(x) (y) exp (i/h ) ¯

d4 x L

(2.56)

The Green’s functions (2.54) satisfy the standard equations of motion which can be derived using the invariance of the functional integral under a shift of variables (x) → (x) + ε f (x) (see Problem 2.2): for example, in a scalar massless ﬁeld theory deﬁned by the lagrangian L = 1 ∂µ ∂ µ 2 we have

xG (2)

− V( )

(x, y) = − 0|T V ( (x)) (y)|0 − iδ(x − y)

(2.56a)

We observe also that given a theory formulated in terms of the Green’s functions (2.54) quantities like 0|T x (x) (y)|0 remain a priori undeﬁned. We can deﬁne them by requiring that they satisfy certain constraints like, for example, for a scalar massless ﬁeld theory

x

0|T (x) (y)|0 = 0|T

x

(x) (y)|0 − iδ(x − y)

where the l.h.s. is given by (2.56a), which amounts to specifying a regularization prescription for them. The Green’s functions (2.54) are given by the functional derivatives of the functional W [J ] equivalent to the vacuum transition amplitude in the presence of the external source J (x) W [J ] = N D exp (i/h ) ¯ d4 x [L + h J (x) (x)] ¯ (2.57)

86

2 Path integrals

**Expanding in powers of J and using (2.54) we can rewrite W [J ] as follows: W [J ] = Consequently, G
**

(n) ∞ n=0

in n!

dx1 . . . dxn G (n) (x1 , . . . , xn ) J (x1 ) . . . J (xn )

(x1 , . . . , xn ) =

1 i

n

δn δ J (x1 ) . . . δ J (xn )

W [J ]

J ≡0

(2.58)

The Green’s functions can also be considered as the analytic continuation of those obtained from the generating functional deﬁned in the Euclidean space with ˆ ˆ x0 = −ix4 , where x4 is real: WE [J ] = N D exp −(1/h )SE [ (x)] + ˆ ¯ d3 x d x 4 J ˆ

**where, for example, for a free scalar ﬁeld theory SE [ (x)] = ˆ
**

1 2

ˆ d3 x dx4

∂ ∂ x4 ˆ

2

∇ + (∇ )2 + m 2

2

.

**The Euclidean Green’s functions are given by G (n) (x1 , . . . , xn ) = ˆ E ˆ δ δ ... δ J (x1 ) ˆ δ J (xn ) ˆ WE [J ]
**

J ≡0

Using path integral formalism one can derive equations of motion for them (Problem 2.2): for example, in free scalar ﬁeld theory − ∂2 − ∇ 2 + m 2 G (2) (x, y ) = δ(x4 − y4 )δ(x − y) ˆ ˆ E ˆ ˆ ∂ x4 ˆ2

analogous to those in Minkowski space. The Minkowski space Green’s functions are given by analytic continuation G (n) (x1 , . . . , xn ) = (i)n G (n) (x1 , . . . , xn ) ˆ E ˆ where ˆ x = x, x0 = −ix4 ˆ

Since in Euclidean space the exponent in the generating functional WE [J ] is bounded from above we can evaluate the functional integral by the saddle point method (see Problem 2.8). We recall that the saddle point approximation rests on the fact that the integral I = dx exp[−a(x)]

2.3 Path integral formulation of quantum ﬁeld theory

87

can be successfully approximated by I ∼ exp[−a(x0 )] = dx exp − 1 (x − x0 )2 a (x0 ) 2

where x0 satisﬁes a (x0 ) = 0, if a (x0 ) > 0 and if the points away from the minimum do not contribute much. We remark also that we can deﬁne the ﬁeld operator evolution for imaginary time (x) = exp(iH x0 ) (0, x) exp(−iH x0 ) = exp(H x4 ) (0, x) exp(−H x4 ) ˆ ˆ ˆ and also the operator ordering T with respect to x4 analogously to the T -product ˆ deﬁnition, for example, ˆ ˆ T (x) ( y ) = ˆ (x4 − y4 ) (x) ( y ) + ˆ ˆ ˆ ˆ ( y4 − x4 ) ( y ) (x) ˆ ˆ ˆ ˆ

whose vacuum expectation value then satisﬁes the same equation of motion as G (2) (x, y ) deﬁned by the generating functional in Euclidean space. E ˆ ˆ

**Action quadratic in ﬁelds In the case in which the classical action is quadratic in the ﬁeld variable S=
**

1 2

(x) (2.59)

d4 x d4 y

(x) A(x, y) (y)

the generating functional W0 [J ] (index 0 corresponds to the speciﬁc form (2.59) of the action) is easily obtained in a closed form. Repeating the steps that led us to the derivation of formula (2.41) we obtain W0 [J ] = exp i h2 ¯ 2h ¯ d4 x d4 y J (x) G(x, y) J (y) (2.60)

where G(x, y) is the Green’s function of the classical ﬁeld equation A(x, y) (y) d4 y = −h J (x) ¯ (2.61)

satisfying conditions (2.40). One example of a theory with quadratic action is the theory of a free ﬁeld, with the lagrangian density L= The extra term 1 iε 2

2 1 2

∂µ ∂ µ

− m2

2

+ 1 iε 2

2

(2.62)

has been added in accordance with the discussion following

88

2 Path integrals

(2.43): it makes the functional integral (2.57) well-deﬁned and simultaneously ensures the correct boundary conditions for the Green’s function G(x, y). As ∂µ ∂ µ = ∂µ ∂µ − ∂µ ∂ µ (2.63)

and neglecting in the action the surface term at inﬁnity, the action can be written in the form S= i.e. A(x, y) = − ∂µ ∂ µ + m 2 − iε δ(x − y) The classical ﬁeld equation ∂µ ∂ µ + m 2 − iε is then solved by (x) = h ¯ d4 y G(x − y) J (y) (2.67) (x) = h J (x) ¯ (2.66) (2.65)

1 2

d4 x d4 y

(x) −∂µ ∂ µ − m 2 + iε δ(x − y) (y)

(2.64)

**with the Green’s function G(x − y) satisfying
**

µ ∂µ ∂(x) + m 2 − iε G(x − y) = δ(x − y)

(2.68)

˜ Introducing the Fourier transform G(k) G(x − y) = we obtain from (2.68) ˜ G(k) = − and therefore G(x − y) = 1 (2π)4 d4 k k2 1 exp[−ik(x − y)] − m 2 + iε (2.71) 1 k 2 − m 2 + iε (2.70) 1 (2π)4 ˜ d4 k G(k) exp[−ik(x − y)] (2.69)

Again, the Green’s function agrees with the analytic continuation from the Euclidean region.

Gaussian integration We end this section with a comment on gaussian integration. Having deﬁned functional integrals as an appropriate limit of multiple integrals, one encounters

2.3 Path integral formulation of quantum ﬁeld theory

89

**integrals of the type I =
**

∞ −∞

**dx exp −bεx 2 + ibx 2 = 2
**

0

∞

dx exp −bx 2 (ε − i)

(2.72)

with ε → 0, where the ε-term corresponds to the (+iε) prescription. Rotating the contour of integration into x = x exp(iϕ) such that exp(2iϕ)(ε − i) = 1 one easily ﬁnds I = exp i 1 π − 1 ε (π/b)1/2 −→(iπ/b)1/2 4 2

ε→0

(2.73)

**which is the analytic continuation of the gaussian integral
**

∞ −∞

dx exp −ax 2 = (π/a)1/2

(2.74)

for complex a (and Re a > 0). A similar result holds for integration over the complex variable z = x + iy dz ∗ dz exp(−az ∗ z) ≡ 2

∞ ∞

dx

−∞ −∞

dy exp −a x 2 + y 2

= 2π/a

(2.75)

for complex a (and Re a > 0). We write dz ∗ dz for 2 d Re z d Im z by convention. For many degrees of freedom z 1 , . . . , z n we can deﬁne the complex vector z z1 . z= . . zn and denote by (z ∗ , Az) the scalar product of vectors z and Az, where A is a (n × n)-dimensional complex matrix. Then we have the following generalization of relation (2.75):

∗ dz 1 dz 1 . . . ∗ dz n dz n exp[i(z ∗ , Az)] = (2π )n in / det A

(2.76)

for any positive-deﬁnite matrix A which can be diagonalized by unitary transformations. The result is obvious for a diagonal matrix A (it follows from multiple use of (2.75)). For the general case one has to show that the integration measure is invariant under the unitary transformation diagonalizing A (the scalar product (z ∗ , Az) is obviously invariant). Denote the unitary transformation by U = A+iB, with A and B real matrices. From the unitarity of U : UU † = 1 it follows that A AT + B B T = 1, B AT − AB T = 0 (2.77)

90

2 Path integrals

The change of integration variables z = U z can be written as x y = A −B B A x y =M x y (2.78)

and using (2.77) we conclude that the transformation matrix M is orthogonal (M M T = 1). Therefore the integration measure is indeed invariant under the unitary transformation U . For integration over n real variables x ≡ (x1 , . . . , xn ), one has dx1 . . . dxn exp[i(x, Ax)] = π n/2 in/2 /[det A]1/2 (2.79)

In the limit n → ∞ integral (2.79) is the J -independent path integral in (2.39). Although we do not have to evaluate it explicitly, it is worth remembering its compact form (2.79). Another point is that in addition to results (2.76) and (2.79) we can also easily calculate integrals in which the product (z ∗ , Az) is replaced by an expression (z ∗ , Az) + (b∗ , z) + (z ∗ , b) ≡ F(z) (2.80)

where b is a constant vector, and matrix A is now hermitean. Indeed, expression (2.80) can be written as F(z) = (ω∗ , Aω) − b∗ , A−1 b (2.81)

where ω = z − z 0 and z 0 = −A−1 b is the minimum of F(z). Actually, (2.81) has been used in the derivation of (2.41) and (2.60) for the generating functional W0 [J ]. Finally, we recall that for any matrix L which can be diagonalized by a unitary transformation det(1 − L) = exp Tr ln(1 − L) where Tr ln(1 − L) = − Tr L + 1 L 2 + 1 L 3 + · · · 2 3 2.4 Introduction to perturbation theory Perturbation theory and the generating functional We shall discuss ﬁrst a simple case of a scalar ﬁeld theory described by the lagrangian λ 4 L = 1 ∂µ (x)∂ µ (x) − m 2 2 (x) − (x) (2.82) 2 4! where x denotes four coordinates in Minkowski space. We seek a method for calculating an arbitrary Green’s function in such a theory. For that purpose it is

2.4 Introduction to perturbation theory

91

convenient to use the functional formulation of the theory in which the Green’s functions are given by functional derivatives of the generating functional W [J ] G (n) (x1 , . . . , xn ) = where W [J ] = N D exp (i/h ) ¯ d4 x [L + h J (x) (x)] ¯ (2.84) 1 i

n

δ δ ... δ J (x1 ) δ J (xn )

W [J ]

J ≡0

(2.83)

and factor N is deﬁned by (2.55). We recall at this point that, so far, we have been able to calculate exactly the generating functional W0 [J ] for a theory of non-interacting scalar ﬁelds with the action S0 given by S0 = d4 x Lfree = d4 x

1 2

∂µ ∂ µ

− m2

2

(2.85)

or, more generally, containing terms at most quadratic in ﬁelds. The analogous, exact, solution is not known for the full theory (2.82) and the method to be used for calculating the Green’s functions (2.83) is a perturbative expansion in terms of powers of SI deﬁned as follows SI = S − S0 where S= d4 x L (2.86)

Such an expansion should be understood as an expansion in a neighbourhood of the vacuum state for which (x) = 0, so that the action SI can be regarded as a small parameter. It is convenient to note the following functional identity D exp i S0 [ ] + SI [ ] + h J (x) (x) d4 x ¯ h ¯ i 1 δ i = exp SI S0 [ ] + h D exp ¯ h h i δJ ¯ ¯

J (x) (x) d4 x (2.87)

so that (2.84) can be rewritten as 1 δ i SI W0 [J ] i 1 δ h i δJ ¯ SI W [J ] = = N exp 1 δ i h i δJ ¯ W0 [J ] exp SI i δJ h ¯ J ≡0 exp

W0 [J ]

(2.88)

The perturbation series is generated by expanding the exponential factor

. x2 ) = −ih G(x1 − x2 ) ¯ 0 (2.90).91) Referring again to (2. xn ) is given by (2. For the free propagator 2 one immediately gets G (2) (x1 . . .k. The n-point Green’s function G (n) (x1 . j. where G(x − y) is the Green’s function for the free classical Klein–Gordon equation G(x − y) = − 1 (2π )4 d4 k exp[−ik(x − y)] k 2 − m 2 + iε (2.83) D G (n) (x1 . . Each term can be written as a respective derivative of the functional W0 [J ].92) (the symmetry G(x1 − x2 ) = G(x2 − x1 ) has been taken into account). . . (Notice that all G (n) -functions with n odd equal zero.89) If SI can be written as an integral over polynomials in ﬁelds (as in our example) any Green’s function can be calculated term by term using (2.90) into sums over discrete points and write i4 G (4) (x1 .90) up to the overall J -independent normalization factor which cancels out in the deﬁnition of the Green’s functions. .l (2.92 2 Path integrals exp{(i/h ) SI [(1/i)(δ/δ J )]} in powers of SI and performing the functional differ¯ entiations as indicated.93) . (xn ) ∞ N =0 ∞ N =0 1 i SI N! h ¯ N N exp i S0 h ¯ i S0 h ¯ 1 i SI N! h ¯ exp (2. .83) one observes that the n-point Green’s function is given by the 1 n th term of the Taylor expansion of exponent (2. . . therefore. . x4 ) = 0 δ4 δ J1 δ J2 δ J3 δ J4 i h ¯ 2 2 J =0 1 Ji J j Jk Jl G i j G kl 2! i.89). The case n = 4 is somewhat more complicated. . the following general formula for the Green’s functions (2. This is equivalent to expanding exp{(i/h )SI [ ]} under the ¯ path integral. .) It is convenient to convert the integrals in (2. 0 From the previous considerations we know that W0 [J ] = exp 1 ih 2 ¯ d4 x d4 y J (x) G(x − y) J (y) (2. xn ) = D (x1 ) .83) with W [J ] replaced by W0 [J ]. Wick’s theorem As an introduction to the detailed exposition of perturbative rules we consider again the case of non-interacting scalar ﬁelds and prove Wick’s theorem. . In perturbation theory one gets.

4 Introduction to perturbation theory 93 where Ji = J (xi ). J4 leads to all possible assignments of subscripts 1–4 to subscripts i. j. .94) This is the content of Wick’s theorem (see. . .89) in the ﬁrst order in SI which reads D (x1 ) . . Bjorken & Drell (1965)) which states that any n-point free Green’s function can be written as a sum over all possible products of 1 n two-point Green’s function G (2) (xi . . . x3 )G (2) (x2 . x3 ) 0 0 (2.2. .95) . l. x j ). .93) can be represented diagrammatically by the ‘prototype’ diagram ≡ (−ih G i j ) × (−ih G kl ) ¯ ¯ and the differentiation with respect to J1 . . . G i j = G(xi − y j ) The structure of (2. Let us consider the four-point Green’s function and discuss the numerator of (2. One gets So ﬁnally (4) G 0 (x1 . x4 ) + G (2) (x1 . for example. x4 )G (2) (x2 . 0 2 An example: four-point Green’s function in λ 4 Coming back to our main problem we shall now discuss several examples in order to derive general rules for perturbative calculations in the theory of interacting ﬁelds. (x4 ) (−iλ) 1 4!h ¯ d4 y 4 (y) exp i S0 h ¯ (2. x2 )G (2) (x3 . k. x4 ) 0 0 0 0 + G (2) (x1 . x4 ) = G (2) (x1 .

(x2 . n. Let us ﬁrst assume that pairs of points joined by the propagators are speciﬁed. To make the combinatorics transparent we have again replaced the integrations in (2. 2. The differentiations of (2. we have an additional combinatorial factor following from the ‘horizontal’ symmetry (for example. p.97) and this happens for any term of expansion (2. l) etc.96) It is clear that the non-zero contribution to (2. x2 . using the generating functional W0 [J ]. 2. y2 . x3 . y3 ). 2.89) (N = 1 in our examples) and about 1/4! present in the deﬁnition of the coupling constant and. y1 = j and x1 = j.5 where the dot represents the integral of a product of two propagators taken at the same space-time points y (notice that the combinatorial factors in front of the second and the third diagrams include different . x4 and four points y (it is convenient to split them into y1 . (x1 .4 with self-evident classiﬁcation of different possibilities.97) now provide all possible assignments of points x1 . j) or (k.89).97) and can be represented graphically by the prototype diagram in Fig. for example. or. In the limit y1 = y2 = y3 = y4 = y we get the diagrams shown in Fig.3. remembering that the limit y1 = y2 = y3 = y4 should be understood) to the points i. These factors cancel with the factor 1/(2n n!) present in (2. 2.94 2 Path integrals Fig. y4 ). y1 = i) which is 2n for n pairs. and another factor from the ‘vertical’ symmetry ((x1 . l. y1 ) can be assigned to (i. k.3. It is easy to discuss them in a systematic fashion. y4 . So we have to remember about 1/N ! present in (2. Then. (x4 .97) by discrete sums. (x3 .96) comes from the term in the Taylor expansion of (2. y1 ).) which is n! (again for n pairs). o. m. The result is as shown in Fig. y2 ). x1 = i. on the other hand.90) which contains four propagators 4 1 ih 2 ¯ 1 4! 4 d4 x d4 y J (x) G(x − y) J (y) (2. (−iλ) 1 4!h ¯ d4 y δ δ4 δ ··· δ J (x1 ) δ J (x4 ) δ J (y) W0 [J ] J ≡0 (2. j. y3 . we have to take account of all possible choices of pairs of points xi and yi joined by propagators.

x2 . it happens to be just the denominator of (2.) So we may ignore the denominator of (2. y) 0 0 0 0 (2.8 which give non-trivial S-matrix elements for G (4) (x1 . In momentum space this corresponds to assigning to each vertex deﬁnite external momenta. They deﬁne the connected Green’s functions G (n) .5.89) and simultaneously all the diagrams containing vacuum subgraphs. which are joined by propagators to each of the interaction points. These problems are related since G (n) involves in N principle a quotient of terms of arbitrary order. 2.6. each of which can be discussed as follows. . Actually. The second bracket contains the sum of the so-called vacuum amplitudes which do not depend on the Green’s function considered and. y) G (2) (x3 . In the following we shall understand the central diagram to be the one in which the pairs of external points. for a given Green’s function we shall be interested only in connected diagrams like those in Fig.7. for example. most importantly.89). y) G (2) (x4 .4. the central diagram can be conn easily read off from the numerator of (2. x4 ). Fig. y) G (2) (x2 . the SI2 term in the numerator is given by the diagrams (a)–( j) in Fig.2. The analytic expression for the ﬁrst diagram reads (−iλ/h ) ¯ d4 y G (2) (x1 . 2.89). We observe that. The amplitude of. (The proof of this again requires some attention to the combinatorial factors.98) It remains to extend our considerations to terms SIN . In this sense we have three different central diagrams.4 Introduction to perturbation theory 95 Fig.89) can be represented by a product of two inﬁnite series of amplitudes as shown in Fig.89). 2. x3 . 2. permutations of x1 . There are twelve ﬁelds under the functional integral. x4 ). 2. . for example. one can convince oneself that the numerator of (2. . In general. have been speciﬁed. . N > 1 and to discuss the role of the denominator in (2. .

there are two combinations for the remaining yl joining z k (factor 2) and ﬁnally y can be interchanged with z in connections with external points (factor 2). No two external points can be joined by a propagator.7.9 and. We can have therefore the situation depicted in Fig. 2. in addition. 2. Fig. In addition two propagators must connect y and z.6.96 2 Path integrals Fig. y1 can be replaced by any yi (factor 4). the same for z 4 and z 3 (factor 4 × 3). 2. so the ‘prototype’ diagram with six propagators is relevant for combinatorics. We get 4! × 4! to cancel 1/4! × 4! from the deﬁnition of the coupling constant.) Since N = 2 there is . (The factors from the ‘vertical’ and ‘horizontal’ symmetries cancel as usual. y2 by any y j = yi (factor 3). Two external points must be joined with the interaction point at y and the other two with the interaction point at z.

positive 0 frequencies propagate into future xn → +∞ (see (2. . . .99) Our examples can be generalized to the following rules. z) G (2) (x4 . . . 2. We ˜ deﬁne G (n) ( p1 . . Attach (−iλ/h ) to each vertex and G 0 to each line. pn ) by the relation G (n) (x1 . xn → +∞ . y) G (2) (x2 . .100) so that all momenta are treated as outgoing (due to (+iε)) in propagators. pn ) (2.45)).89). . . y) G (2) (y. In actual physical transitions 0 0 x1 . xn ) = d4 p 1 d4 pn ··· exp(−i p1 x1 ) · · · exp(−i pn xn ) (2π)4 (2π )4 ˜ ×G (n) ( p1 . Fig. .9. . . To calculate to order N in SI draw all connected diagrams consistent with the structure of the numerator in (2. z) G (2) (x3 . xk → −∞ 0 0 xk+1 . . . 2. .2. ¯ Momentum space One is usually interested in Green’s functions in momentum space deﬁned as Fourier transforms of the respective Green’s functions in conﬁguration space. . z) 0 0 1 (−iλ/h )2 ¯ 2 ×G (2) (y.4 Introduction to perturbation theory 97 Fig. z) 0 0 0 G (n) (x1 . . For each diagram calculate the combinatorial (2) (symmetry) factor. . . xn ) conn (2.8. . an extra factor 1 2 (we will call it the symmetry factor) and the amplitude reads (2) d4 y d4 z G 0 (x1 .

2. Usually.99) 4 1 (−iλ/h )2 ¯ 2 i=1 h ¯ i 2 pi − m 2 + iε d4 k1 d4 k2 (2π )4 δ( p1 + p2 + k1 + k2 ) (2π )4 (2π )4 2 k1 ×(2π)4 δ( p3 + p4 − k1 − k2 )h ¯ i i h 2 + iε ¯ k 2 − m 2 + iε −m 2 (2. Observe that each vertex contributes a factor h −1 and ¯ . .98) the momentum space amplitude reads 4 4 (2π)4 δ i=1 pi (−iλ/h ) ¯ i=1 h ¯ pi2 i − m 2 + iε (2. represented by expression (2. . . . 2.101) For the diagram shown in Fig. Fig. .11. n i 2 p1 − m 2 + iε In momentum space we get from (2.98 2 Path integrals Fig. the diagrams as above are assumed to represent amplitudes without the propagators on external lines (Feynman diagrams).100) are related to the physical momenta pi ( pi0 > 0) of the initial and ﬁnal particles as follows ˜ ˜ ˜ pi = − pi ˜ pi = pi i = 1. .102) and for the diagram shown in Fig. and therefore the momenta p1 . . . . . k i = k + 1.92) and (2. pn deﬁned by (2. 2.11 we get from (2. 2.10.91) ˜0 G (2) ( p1 . . .103) The Feynman rules in momentum space are now obvious. p2 ) = (2π )4 δ( p1 + p2 )h ¯ (2.10.

An important modiﬁcation of the lagrangian (2. each internal line gives a factor h . but with a number of counterterms.2.4 Introduction to perturbation theory 99 Fig.82) follows if we specify that all products of the ﬁeld operators in it are normal-ordered L= 1 2 :∂µ (x)∂ µ (x): − 1 m 2 : 2 2 (x): − (λ/4!) : 4 (x): (2. 2. x) : 0 (x): + const. 2. Again. For N vertices there are (2N −1)(2N −3) . (x): + 4 (2. as in Fig. So we again get the symmetry factor 1 for a Feynman diagram in 2 momentum space. From Wick’s theorem it follows that (see Section 2. the factor (2N − 1) .89) and an extension of our discussion of the N = 2 case gives (3 × 4) N ×2 N −1 × N !. . 1 gives the number of different groupings of 2N momenta in N pairs and (N − 1)! gives the number of their different orderings along the loop. In the resulting perturbation expansion . (2. for example. .104) This can be rewritten in a form similar to (2. remember that under the path integral the ﬁelds are c-numbers and the only way of specifying their order is by introducing appropriate counterterms. (2. Thus a Feynman diagram with L loops is of the ¯ order O h L−1 . ¯ It is easy to extend these considerations to a loop with N vertices. with momenta speciﬁed along all the internal lines.106) so that : (x): = (x) − 6G (2) (x. this corresponds to a well-deﬁned situation in momentum space.12.12.105) 2 6G (2) (x. because L = (number of internal lines − number of vertices + 1). We have the factor (1/N !)(1/4!) N from the general formula (2.7): 2 4 (x) = : (x) = : 4 2 4 (x): + const. 1×(N −1)! different graphs in momentum space.107) and the other counterterms are constants. assume that the pairs of external points which are joined to each vertex as well as the ordering of vertices along the loop are speciﬁed.82). . x) 0 2 (x) − const. .

Let us take η to be a Grassmann variable.112) . 2. like that shown in Fig. 0|T (x1 ) ¯ (x2 )|0 = − 0|T ¯ (x2 ) (x1 )|0 (2. the diagram is now cancelled by the −6G (2) (x.109) For instance if f (η) is taken to be an ordinary number. that the higher order corrections to this diagram. however. One deﬁnes the left and right derivatives by the equation d d η=η =1 dη dη Therefore.5 Path integrals for fermions. The anticommuting c-number variables. d d f (η) = f 1 = − f (η) dη dη ← ← (2. 2. then f 0 and f 1 are ordinary and Grassmann numbers. are not cancelled.108) the fermion classical ﬁelds in the path integral must be taken to be anticommuting quantities. x) 2 (x) counterterm. 0 Note. when f 1 is a Grassmann number.111) (2. are said to form Grassmann algebra.53). for example.100 2 Path integrals Fig. η} = 0 Any function f (η) can be written as f (η) = f 0 + η f 1 (2. or functions. There follows a brief introduction to the subject of Grassmann algebra.13. {η. The constant counterterms cancel some vacuum diagrams.13. Grassmann algebra Anticommuting c-numbers The path integral quantization method involves classical ﬁelds only and the theory is formulated directly in terms of Green’s functions given by path integrals like (2. respectively.110) or η2 = 0 (2. 2. Since for Fermi ﬁelds we must have.

.76) we now have ∗ ∗ dη1 dη1 .117) is valid for an arbitrary A. we deﬁne the integration by the requirement of linearity and the following relations dη 1 = 0.117) follows immediately from the integration rules for Grassmann variables. Aη) = i.114) Consider now the change of integration variable η → η = a + bη. . Aη) = det A (2. the standard Jacobian appears inverted. It can be shown that instead of (2. All the rules can be easily generalized to the case of n real Grassmann variables.113) applies when dη and η are next to each other. . In the following chapter the Gaussian integral for complex Grassmann variables will be shown to be very useful. Grassmann algebra 101 Furthermore. The corresponding formula also holds for an integral over real variables (see Problem 2. One gets dη f (η) = dη dη dη −1 f (η(η )) (2.111) we see that the integrals and the left derivatives are identical dη f (η) = The integral of the derivative vanishes: dη d d2 f (η) = 2 f (η) = 0 dη dη (2. j ηi∗ Ai j η j Eq.113) The ﬁrst relation follows from the requirement that the property of a convergent integral over commuting numbers ∞ −∞ dx f (x) = ∞ −∞ dx f (x + a) valid for any ﬁnite a. The second relation is the normalization convention.115) d f (η) = f 1 dη (2. dηn dηn exp(η∗ .116) i.2.4). For a diagonal A the proof of (2. For a complex Grassmann variable the real and imaginary parts can be replaced by η and η∗ as independent generators of Grassmann algebra.113) with (2.e.117) where (η∗ . holds for an integral over anticommuting numbers as well. (2. We also add a speciﬁcation that rule (2. dη η = 1 (2.5 Path integrals for fermions. Comparing (2.

The generating functional W0 [α] for a free-fermion ﬁeld described by the lagrangian ∂ L = ¯ (i/ − m) . z and η. We can deﬁne a linear transformation in superspace z η = A C D B z z ≡M η η where matrices A and B are commuting.102 2 Path integrals Gaussian integrals can be used to deﬁne the determinant of a matrix acting in superspace. respectively. Az) − (z ∗ . and C and D are anticommuting. Vectors in superspace are decomposed in terms of the variables z and η.9. Dirac propagator The classical Dirac ﬁelds (x) and ¯ (x) are taken to be elements of an inﬁnite-dimensional Grassmann algebra.117) we get det M = If we shift z and z ∗ we get det M = det A det B − C A−1 D 1 A−1 D 0 B − C A−1 D A − D B −1 C B −1 C 0 1 det A − D B −1 C det B To evaluate the integral we make a shift in integration variables These results for the superdeterminant are not surprising if we write M= or M= 1 D 0 B A 0 C 1 Several other properties of the superdeterminant are listed in Problem 2. C z) − (η∗ . Coordinates in superspace are commuting and anticommuting variables. Dη) − (η∗ .76) and (2. Bη)] 2π η = η − B −1 C z η∗ = η∗ − z ∗ D B −1 and using (2. The superdeterminant of a superspace matrix M can be constructed by calculating a generalized Gaussian integral (det M)−1 = dz dz ∗ ∗ dη dη exp[−(z ∗ .

−A−1 β (2.123) reads SF (x − y) = and using (2. α] as follows: ¯ W0 [α.126) (2. 0] and α(x) and α(x) are sources of fermion ﬁelds.119) we can rewrite W0 [α.120) one gets SF (x − y) = −ih S(x − y) ¯ Imagine now a theory of interacting fermions with the action S= d4 x L (x) + d4 x ¯ (x) V (x) (x) (2. α] = N D ¯ D ¯ exp (i/h ) d4 x [L (x) + h α(x) (x) + h ¯ (x)α(x)] ¯ ¯¯ ¯ (2. Aη) + i(β ∗ . η) + i(η∗ .2. β)] = det(iA) exp i β ∗ .118) where N −1 = 0|0 = W0 [0.5 Path integrals for fermions. α] δ α(x) ¯ δα(y) ← (2.124) α=α=0 ¯ = S0 + SI .120) (2.125) 1 i 2 d4 x d4 y α(x) S(x − y) α(y) ¯ (2. Grassmann algebra 103 is W0 [α.81) but for Grassmann ﬁelds Dη Dη∗ exp[i(η∗ . α] = exp ih ¯ ¯ where (i/ − m)S(x − y) = −1 ∂ lδ(x − y) Therefore S(x − y) = − d4 k /+m k exp[−ik(x − y)] 1 l 4 k 2 − m 2 + iε (2π) D¯ D (x) ¯ (y) d4 x L (x) (2. Using a formula analogous to ¯ (2.121) (2.122) The Feynman propagator deﬁned as SF (x − y) = 0|T (x) ¯ (y)|0 / 0|0 = × exp (i/h ) ¯ d4 x L (x) D D ¯ exp (i/h ) ¯ δ δ ¯ W0 [α.

for example.104 2 Path integrals Fig.127) which correspond to fermion loops like that shown in Fig. . In the expansion in powers of SI of. 2. .127) can be calculated explicitly by differentiating W0 [α. .4 we can develop perturbation theory with the interaction term as the perturbation. 2. rearrange the ﬁelds as follows V (x1 ) (x1 ) ¯ (x2 )V (x2 ) (x2 ) ¯ (x3 ) . . . i δ α(xi ) ¯ → (xi ) → ¯ (xi ) → 1 δ i δα(xi ) ← (2. .128) we must.120) with respect to α and α. α] ¯ given by (2.120) and (2. however.14. dxk ¯ (x1 ) V (x1 ) (x1 ) ¯ (x2 ) V (x2 ) (x2 ) .89)) we ﬁnd under the path integral the following products of the and ¯ ﬁelds: dx1 . ¯ (xk ) V (xk ) (xk ) (2. From (2.128) we obtain for the closed fermion loop the following result (−)V (x1 )SF (x1 − x2 )V (x2 )SF (x2 − x3 ) . Before making the substitution ¯ 1 δ . V (xk ) (xk ) ¯ (x1 ) This reordering introduces the familiar minus sign for a closed fermion loop. .129) . the vacuumto-vacuum amplitude (the denominator in (2. V (xk )SF (xk − x1 ) (2. Following Section 2. . The path integral with insertion (2.14.

. A connected Green’s function is obtained by disregarding all terms which factorize into two or more functions with no overlapping arguments. . .2. . . with the exception of the vacuum diagrams which are cancelled by the normalization factor. d4 xn G (n) (x1 . To each type of the Green’s functions we assign the corresponding generating functional.132) (x). . . including disconnected ones. . . Observe that in the generating functional [ ] we denote the arguments as . they remain connected after an arbitrary internal line is cut. The Feynman diagrams contributing to G (n) are all connected. . xn ) = 0|T (x1 ) . . . . . . (xn ) (2. (xn )|0 / 0|0 = × D (x1 ) . .6 Generating functionals for Green’s functions and proper vertices. . J (xn ) d4 x1 . . These are deﬁned as follows W [J ] = ∞ Z [J ] = h ¯ and n=0 ∞ in n! in−1 n! d4 x1 . i. . . d4 xn G (n) (x1 . xn ) J (x1 ) . . The connected Green’s functions G (n) (x1 . In perturbation theory the full Green’s functions G (n) are given by the sum of all diagrams with n external legs. also called the oneparticle-irreducible (1PI) Green’s functions. .6 Generating functionals 105 2. For a Fourier transform this implies that no subset of n external momenta is conserved separately. . xn ) (x1 ) . . . . conn The connected proper vertex functions (n) (x1 . d4 x n (n) (x1 .e.54) with appropriate generalization to the case of other ﬁelds. are given by the Feynman diagrams which are one-particle-irreducible. . . . In the lowest order (no loops) the connected proper vertex functions coincide when appropriately normalized with the vertices of the original lagrangian. . for simplicity. . J (xn ) conn (2. . xn ) are deﬁned as the connected conn part of G (n) . xn ) J (x1 ) . . . xn ). The full Green’s functions are deﬁned by (2.130) (2. . effective potential Classiﬁcation of Green’s functions and generating functionals There are three basic sorts of Green’s functions. in this section we shall always consider the case of a single ﬁeld. . .54) G (n) (x1 . (xn ) exp{(i/h )S[ ]} ¯ D exp{(i/h )S[ ]} ¯ −1 (2. .131) n=1 [ ]= ∞ n=1 1 n! d 4 x 1 . . .

. . .132) that G (n) (x1 .136) We want to show that the functionals Z [J ] and relations: [ ] are given by the following (2.4 and 2.138) where the ‘classical’ ﬁeld combination is deﬁned as the ﬁeld which minimizes the d4 x J (x) (x) [ ]+h ¯ or by the equation δ [ ] δ (x) δ Z [J ] = δ J (x) = = −h J (x) ¯ CL (2. δ J (xn ) (2.141) . xn ) = conn and (n) 1 1 h i ¯ δ n Z [J ] δ J (x1 ) . . .138) with respect to J (x) we also ﬁnd d4 y δ [ CL ] δ CL (y) +h ¯ δ CL (y) δ J (x) CL (x) +h ¯ d4 y δ CL (y) δ J (x) J (y) (2. .137) W [J ] = exp{(i/h )Z [J ]} ¯ and [ CL ] = Z [J ] − h ¯ CL (x) d4 x J (x) CL (x) (2. .134) J ≡0 (x1 . . . . xn ) = 1 i n δ n W [J ] δ J (x1 ) .139). δ J (xn ) n−1 (2.106 2 Path integrals not J (x): this is convenient because of the relationship between [ ] and the action integral S[ ] which we shall discuss below. .130)–(2. with (2. . It is given by the path integral formula W [J ] = D exp i S[ ] + h ¯ h ¯ d4 x J (x) (x) D exp i S[ ] h ¯ (2. . It follows from deﬁnitions (2.135) ≡0 The generating functional of the full Green’s functions has been studied in Sections 2.139) By differentiating (2. . . δ (xn ) (2.133) J ≡0 G (n) (x1 . . .140) or. xn ) = δn [ ] δ (x1 ) .5. CL (x) = 0| (x)|0 1 δ Z [J ] = h δ J (x) 0|0 J ¯ J (2. .

It remains to prove that (2.68) ∂µ ∂ µ + m 2 G(x − y) = δ(x − y) (2.130)–(2.6 Generating functionals 107 Eq. This can be checked by induction (see.h. CL takes the value 0 and vice versa. The value of (2.141) when the external source is turned off (J (x) ≡ 0) is the vacuum expectation value of the ﬁeld (x) which. where ∂µ ∂ µ + m 2 − (λ/4!) 4 S[ ] = d4 x − 1 2 Eq.142). Effective action Consider the classical ﬁeld equation (we take the case of the λ deﬁniteness) ∂µ ∂ µ + m 2 δ δ (do not confuse c c 4 theory for (2.2.132).146) where G(x − y) is deﬁned by (2.138) can be written as a functional of CL (x) which determines [ CL ]. of (2. In particular.144) follows from the action principle S[ ] + h ¯ c = d4 x J (x) (x) = 0 (2. we assume to be zero: δ Z [J ] δ J (x) J ≡0 =0 (2.141). Eq. (2.143) CL ≡0 because when J ≡ 0.145) and CL ). (2. it implies that δ [ CL ] δ CL (x) =0 (2. In the zeroth order in λ we obtain (0) c (x) = d4 y [−ih G(x − y)]iJ (y) ¯ (2.144) can be solved perturbatively. for example.142) The case of a non-zero ﬁeld vacuum expectation value reﬂecting spontaneous symmetry breaking will be discussed later in this section.138) hold for the generating functionals deﬁned by (2.141) to determine J (x) in terms of CL (x) so that the r.138) is the Legendre transform.141) and (2. (2. (2. Here we shall use another argument which allows a simple understanding of the relationship between [ ] and S[ ]. ¯ λ>0 Eq. together with (2. Abers & Lee (1973)).68) .139) expresses J (x) in terms of CL (x) and in this sense it is the inverse of (2.137) and (2. for the time being.s.144) = −(λ/3!) 3 c + h J (x). If Z [J ] is known we can use (2.

The result reads c (1) c (x) = (0) c (x) 3 + d4 y [−ih G(x − y)](−iλ/h 3!) ¯ ¯ d4 z [−iG(y − z)]iJ (z) (2. given by (2. (the iε term should always be remembered).136).144). When the loop corrections are taken into account the connected Green’s functions as well as the generating functional Z [J ] can still be understood as given by the tree diagrams but with the elementary vertices replaced by 1PI Green’s functions. This corresponds to replacing (x) in the integrand of (2. It is easy to see that only tree diagrams appear and that they are all connected. Consider now the generating functional W [J ]. from (2. 2. again make in the integrand the ‘classical’ approximation (x) → CL (x) with CL (x) deﬁned by (2. and we get W [J ] = exp (i/h ) S[ ¯ c] +h ¯ d4 x J (x) c (x) (2. This can be seen by calculating δ n Z /δ J (x1 ) . If in (2. δ c (xn )| c =0 generates proper vertex functions in the tree approximation). To obtain the ﬁrst order correction we insert (0) in the r. Only the numerator then contributes because J ≡ 0 in the denominator. we .138) is indeed the generating function for the 1PI Green’s functions (because δ n S[ c ]/δ c (x1 ) .148) Thus. . of (2. in the classical approximation.147) which can be represented diagrammatically as shown in Fig. δ J (xn )| J ≡0 using (2. in the classical approximation [ CL ] = S[ c ] so in this approximation [ c ] as given by (2.139).137).136) by the classical solution c (x). .s.h.136) we replace the action S[ ] by the functional [ ] and instead of doing the path integration exactly. in the classical approximation Z c [J ] reads Z c [J ] = S[ c] +h ¯ d4 x J (x) c (x) (2.138) and (2. 2. if we assume (2. Repeating the iterations we obtain the perturbation series for the classical ﬁeld c (x).15.108 2 Path integrals Fig.147) for the c (x).15.141). . .149) and the expansion (2.149) and it is indeed given by the connected tree diagrams. Moreover.

137) is indeed connected (the connected diagrams exponentiate).2. Also (2. It is worth observing explicitly that differentiating (2. We conclude that to all orders W [J ] = exp (i/h ) ¯ [ CL ] +h ¯ d4 x J (x) CL (x) (2. In this sense [ ] can be treated as the effective action including the loop corrections. . . xn )s are n-point non-local vertices of this effective action. Z [J ] given by (2. from translational invariance. ˜ ˜ (2) ( p)G (2) ( p) = ih ¯ (we draw attention to the factor i). so that δ [ CL ] δ CL (x) =0 (2. as [ CL ] and CL are both given by connected diagrams. (n) (2) (x − z) G (2) (z − x) = ih δ(x − y) ¯ conn (2. (n) (x1 . .152) s are the Spontaneous symmetry breaking and effective action The formalism of the effective action can be extended to the case of a nonvanishing ﬁeld vacuum expectation value which corresponds to spontaneous symmetry breaking (the latter is discussed in detail in Chapter 9). strictly speaking.141) with respect to CL (y). by induction.139). i 1PI functions given directly in terms of Feynman diagrams. for example.6 Generating functionals 109 shall obtain the exact result. .155) CL =v .137).139) and setting CL (x) = 0 one gets d4 z or. .151) (2.150) and therefore.136) and (2. . ¯ (xn )|0 (2.142) is replaced by 0| (x)|0 1 δ Z [J ] = h δ J (x) 0|0 ¯ =v (2. δ J (xn ) J ≡0 = 0|T ¯ (x1 ) .154) The effective action is deﬁned by (2. It can be proved.153) J ≡0 where v = const.138) holds to all orders. . that higher derivatives of Z [J ] at J ≡ 0 are connected Green’s functions of the ﬁeld ¯ = − v whose vacuum expectation value vanishes: 1 1 h i ¯ n−1 δ n Z [J ] δ J (x1 ) . This is because all loop corrections are already included in [ ]. However. The functionals W [J ] and Z [J ] are taken as given by (2. respectively. For higher n. . we now suppose that (2. using (2. in momentum space.138) and (2.

. . . . xn . see (2. .138) and (2. . δ CL (xn ) = (n) CL ≡0 (x1 . by induction. .159) J ≡J0 One can ﬁrst check by induction that higher derivatives of Z [J ] taken for J ≡ J0 give connected Green’s functions 0|T . . (2. we can prove. xn+m ) . CL (x n ) in terms of [ CL ] deﬁned by (2. . .157) It is important to observe that even for 0| (x)|0 = v = 0 the effective action [ CL ] can also be expanded in terms of the original ﬁelds (x). .139). . .110 2 Path integrals This follows from (2. . . .132). . xn ) (2. . . . This can be proved. . .158) are the 1PI Green’s functions for the original ﬁelds (x). The coefﬁcients of such an expansion δ n [ CL ] δ CL (x1 ) . Then we differentiate the appropriate number of times the relation δ Z [J ]/δ J (x) = CL (x) with respect to n CL and express the derivatives δ J (x)/δ CL (x 1 ) . x n ) n f (x)|x=a = am d m! dx m=0 ∞ n+m f (x)|x=0 = ∞ (1/m!)v m (n+m) d4 xn+1 . d4 xn+m (2. . . . . ( (n) ¯ = −v (x 1 . that [ CL ] given by (2. . Also. x n ) (2. |0 conn . . In the limit J ≡ J0 and CL ≡ 0 we get the desired result. . d 4 x n − v) . by repeating the inductive arguments leading to (2.160) m=0 × (x1 .156) 1 n! d4 x 1 .141) and (2. . . . δ CL (xn ) and [ CL ] = ∞ n=1 CL =v = (n) ¯ = −v (x 1 . . where δ Z [J ] δ J (x) =0 (2. .139). . for example. x n ) CL (x n ) ×( CL (x 1 ) − v) (2. CL takes the value v and vice versa. .156) but for CL = 0 and correspondingly for J (x) = J0 .153) because when J ≡ 0. The relation between the Green’s functions (n) −v and (n) is given by the ¯= analogue of the relation d dx Thus we have (n) ¯ = −v (x 1 . .138) generates 1PI vertices for the ﬁeld ¯ = − v δ n [ CL ] δ CL (x1 ) . . for example.

This method of calculation is equivalent to computing directly the derivative of V and demanding that it vanishes. the negative sum of all non-derivative terms in the lagrangian density. in momentum space. pn ) is the Fourier transform of the proper vertex (n) (x1 .155) has very important implications in studies of spontaneous symmetry breaking: it tells us that the ﬁeld vacuum expectation value is the value of CL (x) which extremizes [ CL ]. 0)ϕ . It is deﬁned by the equation [ (x) = ϕ = const. . . without computing V ﬁrst. . . by demanding that all tadpole diagrams sum to zero. . (2. (In the tree approximation V is just the ordinary potential.e. . xn ) ˜ (n) ( p1 .155) now translates into ϕ=v =0 (2. . .161) m=0 where ( p1 . . one can rewrite the lagrangian of the theory in terms of shifted ﬁelds ¯ = − v considering v as an extra free parameter. xn ) Eq. pn ) = ¯ ˜ (n) ∞ (1/m!)v m ˜ (n+m) ( p1 . . . .164) so that the nth derivative of V at ϕ = 0 is the sum of all 1PI diagrams for the original ﬁelds of the lagrangian. . . . with n vanishing external momenta. pn . not a functional).156). 0. Condition (2. ˜ (n) ( p1 .2. .155) and (2. ϕ (2. . (2. . Using representation (2. . . . .132) for the [ ] we get V (ϕ) = − ∞ n=2 (1/n!) ˜ (n) (0. .6 Generating functionals 111 or. develop the appropriate Feynman rules and ﬁx v directly from (2.] = −(2π )4 δ(0)V (ϕ) dV dϕ (2. . . .) Alternatively. but it makes use of the . . i. 0. . pn )(2π)4 δ( p1 + · · · + pn ) = n d4 x exp(i pi xi ) i=1 (n) (x1 .161) expresses the 1PI Green’s functions in a theory with spontaneously broken symmetry in terms of the 1PI Green’s functions calculated as in the symmetric mode. . Effective potential Eq. These implications can be discussed more conveniently if we introduce the so-called effective potential V (ϕ).163) which is more useful than the original one because the effective potential V (ϕ) can be calculated perturbatively in a systematic way. . . . 0) m (2. . .162) (V (ϕ) is a function.

1](n − 1)! 1 = (1/2n) 1 . It has been seen at the end of Section 2. thus we get (1/2n!)[(2n − 2 n 1)(2n − 3) .166) exhibits the 2 2 UV divergence typical for perturbative calculations in quantum ﬁeld theories.16. . 2. To the lowest order (tree approximation) the only nonvanishing 1PI Green’s functions are ˜ (2) ( p. formulation of the theory in which the underlying spontaneously broken symmetry is not explicit (see Problem 4.165) To the next order (one-loop approximation) we have the inﬁnite series of diagrams shown in Fig.164).112 2 Path integrals Fig. The integral in (2. A sensible way of organizing this double series is the loop expansion. and we get V0 (ϕ) = + 1 m 2 ϕ 2 + (λ/4!)ϕ 4 2 (2. .4 that loop expansion is an expansion in powers of Planck’s constant h : a Feynman diagram ¯ with L loops is O h L−1 . In principle. ¯ ¯ The calculation of the effective potential for the λ 4 theory in the one-loop approximation is simple. using (2. Thus. (1/h )L → L corresponds to a change of units into h = 1. 1](n − 1)! Feynman diagrams each with the symmetry factor 1 . the loop expansion preserves any symmetry of ¯ the lagrangian and it is unaffected by shifts of ﬁelds and by the redeﬁnition of the division of the lagrangian into free and interacting parts associated with such shifts.166) The factor (1/2n) 1 can be easily understood by recalling the discussion at the 2 end of Section 2. .16 which gives the following contribution V1 (ϕ) = i ∞ n=1 1 2n 1 d4 k ϕ2 λ 2 (2π )4 k − m 2 + iε 2 n = −1i 2 n 1 λϕ 2 d4 k ln 1 − 2 2 2 (2π )4 k − m + iε (2. Knowledge of the effective potential means that one knows the structure of spontaneous symmetry breaking.1).4: each diagram represents [(2n − 1)(2n − 3) . . p) = p 2 − m 2 and ˜ (4) = −λ. 2. . Indeed expansion in powers of h is an expansion in a parameter that multiplies the ¯ total Lagrange density. we can calculate the effective potential in perturbation theory but the calculation involves a double sum: over all 1PI Green’s functions and for each 1PI Green’s function there is the expansion in powers of the coupling constant.

169) dt HI (t) t U (t.170) S = U (∞. t) = exp(iH0 t) (x.171) It transforms the ‘incoming’ (t → −∞) states into the ‘outgoing’ (t → +∞) ones. In the interaction picture the time evolution of operators is governed by the free.170) can be expressed in terms of the interaction picture ﬁeld operators. let us consider the case of a single scalar ﬁeld. −∞) = T exp −i ∞ −∞ dt HI (t) (2.166) we refer the reader to Section 11. (2. Coleman (1974)) that V (ϕ) is the expectation value of the energy density in a certain state for which the expectation value of the ﬁeld is ϕ. H0 .172) . If V (ϕ) has several local minima. it is only the absolute minimum that corresponds to the true ground state of the theory. for example.168) with the formal solution |t = U (t. It can be shown (see. described by the lagrangian density L= 1 2 ∂µ ∂ µ − m2 2 − V( ) (2. t )|t t (2. 0) exp(−iH0 t) (2. Formally. For simplicity. 2.7 Green’s functions and the scattering operator 113 Its evaluation requires setting up the renormalization programme and for further discussion of V1 given by (2. remark in this section deals with the physical meaning of the effective potential.2. Our ﬁnal. for example.167) while the state vectors obey the equation i ∂ t = HI (t)|t .163) for the ﬁeld vacuum expectation value 0| (x)|0 = v. part of the complete hamiltonian operator H = H0 + HI of the system (from now on we put h = 1) ¯ (x. The interaction hamiltonian HI (t) in (2.7 Green’s functions and the scattering operator The Green’s functions of quantum ﬁeld theory can be used to determine the matrix elements of the operator S which are directly related to scattering amplitudes and therefore experimental results. t ) = T exp −i We then deﬁne the scattering operator to be (2.2. the operator S can be obtained as an inﬁnite time limit of the evolution operator in the interaction picture. This interpretation of V (ϕ) is not surprising: it is already suggested by the classical limit of V (ϕ) and by. but very important. ∂t h I (t) = exp(iH0 t)HI exp(−iH0 t) (2.

173) where (x. by the hamiltonian H= 1 2 d3 x 2 ∇ (x. t)] (2π)3 2k0 (2. k 0 = + k2 + m 2 1/2 (2. . a † (k) are the annihilation and creation operators of scalar particles with three-momentum k satisfying [a(k). j i= j (−i)G(xi − x j ) : k=i. t) + [∇ (x)]2 + m 2 2 (x) + d3 x V ( (x)) = H0 + HI (2.114 2 Path integrals or. (xm ) = : (x1 ) .176) and a(k). . j1 .174) so that at any moment of time we can write (x. t). a † (k )] = (2π )3 2k0 δ(k − k ) (see Section 1. equivalently.. t) = ∂L/∂ ˙ (x. (2. t) = exp[−i(k0 t0 − k · x)]. . t) + a † (k) f k∗ (x. (2. t) = d3 k [a(k) f k (x.4). (xm ): + i.. jn (−i)G(xi1 − x j1 ) . . (−i)G(xin − x jn ) when when m = 2n m = 2n + 1. We can now use (2.171).175) to express S in terms of the annihilation and creation operators and calculate the matrix elements between states containing scalar particles. Eq. When applied to the T -ordered product of m ﬁeld operators Wick’s theorem has the well-known form (Bjorken & Drell 1965) T (x1 ) .167) implies that the interaction picture ﬁeld operator obeys the Klein– Gordon equation of a free ﬁeld + m2 (x) = 0 (2. This is done by using Wick’s theorem.177) × 1 : (xl ): . with all creation operators standing to the left of the annihilation operators. When calculating the matrix elements it is convenient to deal with normalordered products (in the interaction picture). . j (xk ): + · · · + i 1 . . l = i 1 = · · · = jn (2.173) and (2.175) where f k are the plane wave solutions of the Klein–Gordon equation f k (x. i n .

It is a straightforward generalization of the usual functional differentiation: the only difference is that one must keep in mind that the ﬁeld operators for different values of time do not commute.2. the corollary of Wick’s theorem used to derive the perturbation theory Feynman rules in the previous section.181) δ J (y) so that (2.e.182) Here G −1 (x − y) is the inverse operator to G(x − y) d4 y G −1 (x − y) G(y − z) = δ(x − z) i.178) between the vacuum states.178) where F[ ] is some functional of the ﬁeld operator and G(x − y) is the Green’s function given by (2. Now let us take the interaction hamiltonian to be HI (t) = d3 x V ( ) = − d3 x J (x) (x) (2. Functional differentiation with respect to the operator has been introduced here.179) where J (x) is a c-number function.7 Green’s functions and the scattering operator 115 A more general and compact formulation is T F[ ] = exp − 1 i 2 δ δ G(x − y) d4 x d4 y :F[ ]: δ (x) δ (y) (2.183) . If we assume F[ ] to be a product of four ﬁeld operators. the result is exactly (2. This corresponds to the case of scattering from an external source.91). after performing the functional differentiation in (2. and take the matrix element of (2. the Klein–Gordon operator.178): T exp i d4 x J = exp 1 i 2 d4 y d4 x J (x) G(x − y) J (y) d4 x J : : (2.94). We observe that δ J (x)W0 [J ] = −i d4 y G −1 (x − y) W0 [J ] (2.180) can be rewritten as follows T exp i d4 x J = : exp d4 x d4 y (x) G −1 (x − y) δ : W0 [J ] δ J (y) (2. We obtain. (2.180) × :exp i = W0 [J ] :exp i d4 x J where W0 [J ] is the previously introduced generating functional of the free-ﬁeld Green’s functions. (x1 ) (x2 ) (x3 ) (x4 ).

186) × iG (x1 − y1 ) . .187) It is convenient to assume that ki = k j for all i. the generating functional of the previously introduced Green’s functions of the interacting theory. . j pairs. by (2. . .185) J ≡0 We have been dealing with ﬁelds in the interaction picture but in W [J ] we can recognize. Finally. . km | = c 0|a(k1 ) . (xn ): The factor N −1 contains all the vacuum-to-vacuum transitions and will be omitted in the following. c are the normalization constants) |k1 . . the S operator of the interacting scalar ﬁeld can be written in the form S = T exp −i × exp −i = N −1 :exp d4 x V ( ) = : exp d4 x V 1 δ i δJ W0 [J ] J ≡0 d 4 x d4 y (x) G −1 (x − y) δ : δ J (y) d4 x d4 y (x) G −1 (x − y) δ : W [J ] δ J (y) (2. Expanding the exponential we obtain S = N −1 ∞ n=0 −1 1 n! d4 x1 d4 y1 . . . The calculation of matrix elements is now straightforward. see (2. kn = ca + (k1 ) . . there are n!m! ways of matching the annihilation and m . a(km ) (2. (xn+m ): will then contribute.54). . . . iG −1 (xn − yn ) : (x1 ) .116 2 Path integrals The generalization to arbitrary V ( ) follows from the formula T exp −i d4 x V ( ) = exp −i d4 x V 1 δ i δJ T exp i d4 x J J ≡0 (2. . .184) Consequently. . . a + (kn )|0 k1 . . . d4 xn d4 yn G (n) (y1 . Only the term in the series with : (x1 ) . We have (c. Let the initial state involve n particles and the ﬁnal state m particles. . . Of the (n + m) ﬁeld operators n will act as the annihilation and m as the creation operators: this gives a combinatorial n+m factor . yn ) (2.88). deﬁned as the vacuum matrix elements of the T -ordered products of the ﬁeld operator in the Heisenberg picture.

192) . kn = cc × d4 x1 . . . Its connected part is proportional to the on-shell limit of the so-called connected truncated Green’s function.188) With the usual invariant normalization |k1 . . d4 yn exp(ikn yn ) G (n) (y1 . . . . Fields satisfying (2. external lines of G conn At this point we must observe that in the derivation of (2. (2. .7 Green’s functions and the scattering operator 117 creation operators with the external particles. d4 xn+m f k∗ (x1 ) . . . km |S|k1 . km .e. −k1 . . The result is k1 . .190) (2. i.191) We see that the S matrix element is equal to the multiple on-shell residue of the Fourier transformed Green’s function.189) ki2 →m 2 i ˜ ×G (n+m) (k1 . d4 yn+m iG −1 (x1 − y1 ) · · · ×iG −1 (xn+m − yn+m )G (n+m) (y1 .175)) that the one-particle state created by (x) from the vacuum has the standard normalization k| (x)|0 = exp[i(k0 x0 − k · x)] with k| normalized as in (2. . . yn ) (2. . kn = a † (k1 ) . . . . . . obtained from ˜ conn ˜ G (n) by factorizing out the exact two-point functions G (2) (ki ) for each of the ˜ (n) . . . kn = lim 1 2 k − m2 i i (2. . More general cases will be discussed in Chapter 4. .192) holds for the ﬁeld operators both in the interaction and in the Heisenberg pictures. . . .190) we have made the assumption (following from (2. f kn (xn+m ) 1 d4 y1 . .192) we shall call the ‘physical’ ﬁelds. km |S|k1 . . −kn ) where ˜ G (n) (k1 . .193) (2. . . kn ) = d4 y1 exp(ik1 y1 ) . yn+m ) (2. a † (kn )|0 this is just the on-shell limit of the Fourier transformed function k1 . . . .189).2. . . . . k| = 0|a(k) or k|k = (2π )3 2k0 δ(k − k ) Normalization (2.

. We conclude therefore that 2 k1 →m 2 ˜ lim G (n) (k1 . kn ) = m d4 y1 . .190).118 2 Path integrals The above formal considerations leading to the Lehman–Symanzik– Zimmermann reduction theorem (2. . . .196) |1 1| ≡ . . . .191) of the Green’s function G (n) (y1 .195) 0 0 k 1 + pn 0 0 exp i k1 − pn t 2 2 k1 − pn + iε 2 Thus. .195). . . Let us consider the Fourier transform (2. Among the various terms which ˜ contribute to G (n) (k1 . .192). . (yn )|0 + ··· (2. (yn )|0 . . . . It is important to observe that the pole arises due to the limit 0 y1 → +∞ in the integral (2. We can insert a complete set of intermediate states between (y1 ) and the remaining time-ordered product getting ˜ G (n) (k1 . which by themselves can actually be taken as a proof of (2. d4 yn exp(ik1 y1 ) . kn ) = 2 k1 i k1 . (yn )|0 − m2 d3 p |p p| (2π )3 2 p0 (2. . yn × 0| (y1 )|m m|T (y2 ) . . . a single particle state of mass m contributes a pole at k1 = m 2 while no other state |n and none of the remaining terms arising from other time-orderings can have such a pole. .190) can be supplemented by the following arguments. and assume that the ﬁeld (x) is normalized according to (2.194) The y1 -integration can now be carried out using 0| (y1 )|n = 0| (0)|n exp(−i pn y1 ) and the integral representation for the 0 y1 − t = -function ∞ 0 dω exp −iω y1 − t ω + iε i 2π −∞ One then gets the following d4 y1 exp[i(k1 − pn )y1 ] 0 y1 − t = iδ(k1 − pn )(2π )3 = i(2π )3 δ(k1 − pn ) 0 k1 1 0 0 exp i k1 − pn t 0 − pn + iε (2. kn ) there is a time-ordering in which (y1 ) stands furthest to the left. . . . yn ) = 0|T (y1 ) . . . . . t = +∞|T (y2 ) . where we have taken 0|0 = 1. . exp(ikn yn ) 0 0 0 y1 − max y2 .

p2 |S|k1 . kn . . Weinberg (1995)). The latter approach is sometimes quite convenient.190) can also be developed either in the path integral approach. . as in Section 2. remembering that k1 .4. .2. Perturbation theory for calculating the Green’s functions present in (2. . (1. k2 = 0|a(p1 )a(p2 )T exp −i λ 4! d4 x : 4 (2.126). Commuting the creation and annihilation operators (using (1. using (2.177) that contain contractions do not contribute to the connected amputated part of the matrix element. kn ) = 2 k1 i 0|T (y2 ) . km |S|k1 . The S-matrix elements can be directly calculated as matrix elements of the operator in (2. for instance.188) and (2.126) we rederive (2. . In the lowest order. For instance.130)) and performing the integral over d4 x as well as the integrals over momenta in the ﬁeld operators (1. . in particular for quick tree level manipulations.190) consider elastic scattering of two identical scalar particles in the 4 theory. . km . is valid for both the path integral and canonical formulations of quantum ﬁeld theory. . In this order the terms in (2. . a † (k2 )|0 4! λ = −i d4 x 0|a(p1 )a(p2 ) + (x) + (x) − (x) − (x) 4 ×a † (k1 )a † (k2 )|0 (2. (2.7 Green’s functions and the scattering operator 119 Similarly.197) By repeating these steps we prove (2. . t = −∞ − m2 (2.190).190). (yn )|k1 . .200) where + ( − ) stands for positive (negative) frequency parts of the ﬁeld operators (1. .102) we get (S − 1)2→2 = −iλ(2π )4 δ (4) ( p1 + p2 − k1 − k2 ) (2. t = +∞|k1 . . or in the (canonical) operator formalism (see.129). k2 = −i λ d4 x 0|a(p1 )a(p2 ) : 4 (x):a † (k1 ). 2 k1 →m 2 ˜ lim G (n) (−k1 . for 2 → 2 scattering in the 4 theory we have: S2→2 = p1 .198) The basic result.199) : a † (k1 )a † (k2 )|0 In the lowest order the contribution to the transition matrix is p1 . .171) (expressed in terms of ﬁelds satisfying free equations of motion) taken between free particle states.198). kn = k1 . p2 |S − 1|k1 .188) and (2. Bjorken & Drell (1965). . t = −∞ As an explicit example illustrating (2.

s) (u(k. s4 ). In the lowest order we have f 4 (p4 . Using the Wick theorem (2. s2 ).177) and repeating the same steps as in the previous example one arrives at (S − 1) f1 f¯2 → f¯3 f4 = −ig si d3 q 4 . The operator ¯ method is convenient to reach this conclusion quickly. s2 ) (2. s2 ) ¯ ×[u(p4 . s3 )|S − 1| f¯2 (p2 . s)). s1 )] = −ig(2π )4 δ( p4 + p3 − p2 − p1 ) ¯ B v(p3 . s1 )b(q1 .12) from (2.201) (where A(B) are some constant matrices in the spinor space) and compute the matrix elements of S − 1 for the process f 1 f¯2 → f¯3 f 4 .202) where we have used the abbreviations b4 ≡ b(p4 . it is clear that the absolute sign of the amplitude containing more than one fermion (or antifermion) in the initial and/or ﬁnal states is purely conventional as it depends on the arbitrary ordering of fermionic creation operators deﬁning these states.7).. Problems 2. b1 |0 A u(p1 .2 For a scalar ﬁeld theory deﬁned by a lagrangian L = 1 ∂ µ (x)∂µ (x) − V ( (x)) 2 .. Also. 2. f 1 (p1 . s4 ) ¯ (q3 . s)) and the initial ¯ (ﬁnal) antifermion with momentum k and spin s by v(k. f¯3 (p3 . s4 ) ¯ ×[v(q2 . s3 )][v(p2 . integrating over pi s ﬁrst. s1 )]: d2 . s2 )d(q2 .120 2 Path integrals As a second example consider the so-called four-fermion interaction lagrangian of the generic type LI = −g :( ¯ f4 (x) A f 3 (x))( ¯ f2 (x) B f 1 (x)): (2. s) (v(k. s4 )b† (q4 . The relative signs of several different contributions to the transition amplitudes (if present) and various combinatorial factors are easy to ﬁx by operator manipulations. s3 )] † † A u(q1 .1 Derive (2. s1 ) = −ig d4 x 0|b4 d3 : † 0 f 4 (x)γ A f 3 (x) † 0 f 2 (x)γ B f 1 (x) † † : d2 . s3 )d † × 0|b4 d3 :[u(q4 . (2π )3 2E q4 d3 q 1 (2π )3 2E q1 × d4 x exp[−ix · (q1 + q2 − q3 − q4 )] B v(q3 .203) It follows from this example that the initial (ﬁnal) Dirac fermion with momentum k and spin s is described by the wave-function u(k. b1 |0 (2. s4 ) etc.

ηi Ai j η j .4 For real Grassmann variables ηi prove the formula dη1 .4 calculate the following symmetry factors S in the λ theory 4 2. 2.158).6 Prove (2. .56) and the invariance of the generating functional (2. (2. Aη) = 2.54) and (2.5 Prove (2. .3 Using the rules of Section 2. 2. 2.10) calculate the Green’s function for the harmonic oscillator x | exp[−i(t − t)H ]|x = where H = − 1 ∂ 2 ∂ x 2 + 1 ω2 x 2 2 2 ω 2 S(τ ) = x + x 2 cos ωτ − 2x x 2 sin ωτ ω 2iπ sin ω(t − t) 1/2 exp[iS(t − t)] .57) under an inﬁnitesimal change of variables (x) → where f (x) is a function of x D exp iS[ + ε f ] + i d4 x ( + ε f ) J = D exp iS[ ] + i d4 x J (x) + ε f (x) Repeat the same in Euclidean space. dηn exp 1 2 (η Aη) = (det A)1/2 i. where (η.154).7 Using representation (2.156) and (2. j for any antisymmetric matrix A.Problems and the corresponding equation of motion +V ( )=0 derive the equation of motion for the Green’s function 0|T (x) (y)|0 : x 121 0|T (x) (y)|0 = − 0|T V ( (x)) (y)|0 − iδ (4) (x − y) using deﬁnitions (2.178).

9 For a matrix M acting in superspace z η =M z = η A C D B z η where z and η are commuting and anticommuting coordinates.122 2 Path integrals Note that S(τ ) is the classical action (along the classical path). The steps are: (a) Expand the exponent around 0 (often called the background ﬁeld) which is the solution of the classical equation of motion in the presence of the external source J (x) ˆ ˆ −∂ µ ∂ µ + m 2 to write WE [J ] = NE exp −SE [ × 0] + 0 +V ( 0) − J =0 (2.166). Calculate the effective potential and prove (2.205) × [ (x) − ˆ (b) In the classical approximation ˆ 0 ( x)][ ˆ 0 ( y )] + · · · WE [J ] = NE exp −SE [ 0] + d4 x J ˆ 0 Evaluate Z [J ] and [ ] explicitly by solving (2.204) J 4 ˆ 0d x D exp − 1 2 d4 x d4 y ˆ ˆ (y) − ˆ δ 2 SE [ 0 ] δ (x)δ ( y ) ˆ ˆ (2. Check that in this approximation [ ] is the classical action. (b) Show that the supertrace deﬁned as Tr M = Tr A − Tr B satisﬁes the cyclicity property Tr[M1 M2 ] = Tr[M2 M1 ] .205) by using det A = exp(Tr ln A) and expanding the logarithm in powers of λ.2 with V ( ) = λ 4 calculate the effective potential [ ] by evaluating the generating functional WE [J ] in Euclidean space WE [J ] = NE D exp −SE [ ] + J d4 x ˆ by the method of steepest descent.204) perturbatively in powers of λ. 2. Evaluate the Gaussian integral in (2. respectively: (a) Write det M instead of (det M)−1 as a Gaussian integral.8 For a scalar ﬁeld theory as in Problem 2. 2. (c) Calculate [ ] in the saddle point approximation.

as for ordinary matrices. B]]. [A. B] + · · · 4! The exponent and the logarithm of a matrix in superspace are deﬁned by a series expansion. B] + 1 [A. . [A. B] + 2! 3! 1 2 [[A. prove that ln det M = Tr ln M and det(M1 M2 ) = det M1 det M2 (e) Write the Jacobian of a transformation in superspace. B]] 2 1 [[A. 123 B]. (d) Writing M= A C 0 1 1 0 A−1 D B − C A−1 D and using property (c).Problems (c) Show that Tr ln M1 M2 = Tr ln M1 + Tr ln M2 Hint: use the Campbell–Backer–Hausdorff formula for ordinary matrices exp(A) exp(B) = exp A + B + + 1 1 [A.

1) For brevity we write Aµ instead of Aα .1) is constant along any orbit 124 .x dAα (x).3 Feynman rules for Yang–Mills theories 3. which implies multiple counting of the physically equivalent conﬁgurations (equivalent up to a gauge transformation).e. gauge-invariant. Aµ denotes the result of this transformation when applied to Aµ . corresponding to a physical. it must have the following property: g D Aµ = D Aµ g where g is an arbitrary transformation from the gauge group. Consider a path integral over gauge ﬁeld Aµ . The simplest formal ansatz for the integration measure which has this property is D Aµ = µ.α.1. i. L is the µ lagrangian density. where α is the gauge group index. As explained before. 3.1) runs over all possible conﬁgurations Aµ (x). the path integral in (3. The integrand of (3. quantity f (Aµ ) D Aµ f (Aµ ) exp i d4 x L (3.1 The Faddeev–Popov determinant Gauge invariance and the path integral As discussed in the Introduction gauge quantum ﬁeld theories are of special physical interest in our attempts to describe fundamental interactions. In this chapter we shall derive Feynman rules for gauge theories. This construction can be represented graphically as shown in Fig. Let us divide the g conﬁguration space {Aµ (x)} into the equivalence classes {Aµ (x)} called the orbits of the gauge group. the µ straightforward generalization of the integration measure introduced previously for scalar ﬁelds. The integration measure D Aµ we assume to be invariant under gauge transformations: i. An orbit of the group includes all the ﬁeld conﬁgurations which result when all possible transformations g from the gauge group G are applied to a given initial ﬁeld conﬁguration Aµ (x).e.

. the integral as it stands is proportional to an inﬁnite constant (the volume of the total gauge group G). we assume that the equation g F Aµ = 0 (3.3. because the inﬁnity can always be cancelled by a normalization constant. i. for any initial ﬁeld Aµ . where α is the group index of the adjoint representation and x a space-time point.1) after the inﬁnite constant has been factorized out. Consequently. In the conﬁguration space {Aµ (x)} the equation F[Aµ ] = 0 must then deﬁne a surface that crosses any of the orbits of the group exactly once. This is not an important difﬁculty in itself. As for the functional F[Aµ ]. (3. We shall now describe a procedure by means of which this factorization can be achieved. In fact to ﬁx a gauge we need (at each space-time point) one equation for each parameter of the group.4) deﬁnes a ‘gauge’. g0 . over the physically distinct ﬁeld conﬁgurations) which follows from (3. of the gauge group. One way to resolve this problem is to apply perturbation theory to the functional integral over the coset space of the orbits of the group (in other words.2) [Aµ ] deﬁned by the following equation [Aµ ] g Dg δ F Aµ (3.1) perturbatively because local gauge symmetry implies that the quadratic part of the gauge ﬁeld action density has zero modes and therefore cannot be inverted in the conﬁguration space {Aµ (x)} so that the propagator of the gauge ﬁeld cannot be deﬁned. one at each space-time point. [Aµ ]).3) Here δ[ f (x)] represents the product of the usual Dirac δ-functions: x δ( f (x)). Let us observe that the functional F[Aµ ] should be of the general form F α (x. 3. The problem appears when we want to calculate (3.1. and let us introduce a functional 1= Dg = x dg(x) (3.1 The Faddeev–Popov determinant 125 Fig. Let Dg denote an invariant measure on the gauge group G Dg = D(gg ).e.4) has exactly one solution.

From (3. Unlike (3.5) where we have used the invariance property of the group measure Dg.1).7) which deﬁnes the theory as formulated in the F[Aµ ] = 0 gauge. Our aim is to replace integration over all ﬁeld conﬁgurations by integration restricted to the hypersurface F[Aµ ] = 0. [Aµ ] = det g δ F Aµ δg (3. We start by inserting (3.7) can be used as a starting point for perturbative calculations. the functional [Aµ ] depends only on the orbit to which Aµ belongs. (3. In this case each orbit will contribute only one ﬁeld conﬁguration and we shall have an integration over physically distinct ﬁelds. The result is Dg D Aµ g [Aµ ] f (Aµ ) δ F Aµ exp{iS[Aµ ]} (3. Faddeev–Popov determinant We have still to calculate −1 [Aµ ]. Next we change the order of integrations.3) we obtain.8) F[Aµ ]=0 g . the result g D Aµ g g g Aµ f Aµ δ F Aµ g exp iS Aµ g can be made manifestly g-independent by a change in notation: Aµ → Aµ . This can be shown as follows −1 g Aµ = gg Dg δ F Aµ g Dg δ F Aµ = = gg D(gg ) δ F Aµ −1 = [Aµ ] (3.3) (which is equal to one) under the path integral in (3.e. the group integration Dg factorizes out to produce an inﬁnite constant: the volume of the full gauge group. f Aµ and S Aµ : [Aµ ]. To show this we can use the gauge invariance of D Aµ . after formal manipulations DF det g δ F Aµ −1 [Aµ ] = δg δ[F] i.1). Consequently.126 3 Feynman rules for Yang–Mills theories [Aµ ] is invariant under gauge transformations. We obtain ﬁnally Dg D Aµ [Aµ ] δ[F[Aµ ]] f (Aµ ) exp{iS[Aµ ]} (3.6) An important observation is that the complete expression under the Dg integral is in fact independent of g. In effect. f (Aµ ) and S[Aµ ] to replace them by D Aµ . g g g g Aµ .

11) g Here Aµ stands for Aµ and α. [Aµ ]) 1 (z) d4 z cρβγ Aγ (z) + δρβ ∂ν δ(z − y) ν g δ Aρ (z) ν (3. Then in (3. F α (x.105) for the inﬁnitesimal transformation of the gauge ﬁeld. y) = = d4 z δ F α (x. [Aµ ]) + d4 y M αβ (x.10) where U ( ) = U ( )U ( ) and is arbitrary. It is convenient to use the gauge invariance property of [Aµ ] to choose Aµ which already satisﬁes the gauge g condition F[Aµ ] = 0. g=1 F[Aµ ] = 0 (3. For inﬁnitesimal transformations ≈ + and therefore det(d /d ) = 1.9) Near g = 1 we only have to deal with the inﬁnitesimal transformations: U ( ) = 1 − iT α α (x) (where α (x) 1) and the invariant group measure Dg takes the α simple form D ≡ α. so that (3. y).13) [Aµ ] = det M.15) For the subsequent applications the following observations will be helpful. with all relevant indices [Aµ ] = det δ F α x. we can also write M αβ (x.14) Recalling (1.12) M (x. y) δ F α x. Aµ so that = F α (x. [Aµ ]) δ Aρ (z) ν δ β (y) θ =0 δ Aρ (z) ν δ F α (x.x d (x). We can now rewrite (3. This matrix appears in the expansion of F α x.1 The Faddeev–Popov determinant 127 [Aµ ] is usually called the Faddeev–Popov determinant. which simpliﬁes the practical calculations [Aµ ] = det g δ F Aµ δg . y) = and αβ (3. β are the gauge group indices.3.10) is proved. First . We have to calculate a determinant of a matrix in both space-time and the group indices M αβ (x. Aµ δ β (y) =0 . [Aµ ]) = 0 (3.8) we can replace the constraint F Aµ = 0 by g = 1.9) in a more explicit form. F[Aµ ] = 0 (3. It is easy to check the invariance property of D D =D (3. Aµ δ β (y) =0 β (y) + · · · (3. Aµ in powers of the inﬁnitesimal parameters β (y) F α x.

7) becomes Dg D Aµ det M δ[F[Aµ ] − C(x)] f (Aµ ) exp{iS[Aµ ]} (3. because the condition F[Aµ ] = 0 is already enforced by the δ-functional which ﬁxes the gauge.16) is the only C-dependent term. of a gauge-invariant quantity can be rewritten in a ﬁxed gauge. we obtain D Aµ det M f (Aµ ) exp{iS[Aµ ]} = DC δ[F[Aµ ] − C] G[C] (3. so that the perturbation theory may be set up.1). Observing that the δ-functional in (3. We can integrate (3. consequently.16) functionally over DC with an arbitrary weight functional G[C].7) by some other functional of the gauge condition. which may be more convenient for practical calculations. In particular the generating functional W [J ] in the F[Aµ ] = 0 gauge is given by the following .7) can be replaced by det M. because the quantity we calculate is gauge-invariant by assumption. this is obviously independent of C(x).17) D Aµ det M f (Aµ ) exp{iS[Aµ ]}G[F[Aµ ]] A popular choice for G[C] is G[C] = exp − i 2α d4 x [C(x)]2 (3. where C(x) is an arbitrary function of a space-time point. This is equivalent to replacing the original lagrangian density by 1 Leff = L − (F[Aµ ])2 (3.18) approaches the δ-functional (up to the α-dependent normalization factor which for our purposes is irrelevant). but the ﬁnal results will not. The rules of this perturbation theory will depend on the gauge. Consider a class of gauge conditions of the form F[Aµ ] − C(x) = 0. For all gauge conditions belonging to this class det M is the same because C(x) is unaffected by g the gauge transformation from Aµ to Aµ .18) where α is a real constant.16) Being gauge-invariant. Notice also that in the limit α → 0 (3.16) only by an overall normalization constant. We have shown how a path integral representation.19) 2α which is no longer gauge-invariant. We can make use of this feature in order to replace the δ-functional in (3. The derivation of Feynman rules in gauges ﬁxed by δ-functionals can often be simpliﬁed by taking this limit. In the following we shall also deal with gauge-dependent quantities. perturbation theory may apply in this case.128 3 Feynman rules for Yang–Mills theories let us note that [Aµ ] under a path integral like (3. (3. the result will differ from (3. In the F[Aµ ]−C(x) = 0 gauge (3.

y)-matrix deﬁned by (3. which is irrelevant.1 The Faddeev–Popov determinant 129 expression: W [J ] = N D Aµ [Aµ ] δ(F[Aµ ]) exp i α d4 x L + Jµ Aµ α (3. an element of the S-matrix is obtained from the corresponding Green’s functions by removing single-particle propagators corresponding to external lines.13) is then of the form 1 M(x. x2 ) corresponding to 0 different gauge conditions.7. To deal with the δ-functional δ ∂µ Aµ which ﬁxes the gauge we shall (3. Although in a sense unphysical. Examples As a simple example let us consider the case of quantum electrodynamics (QED). We shall calculate the free photon propagators G µν (x1 . ’t Hooft & Veltman (1973)).3. the S-matrix suffers from IR divergences and may not even be deﬁned. In particular. taking the Fourier transform of the resulting ‘amputated’ Green’s function and placing external momenta on the mass-shell. for example.21) . Although we have not discussed the renormalization programme yet. if not with a spontaneously broken gauge symmetry.20) α Due to the presence of ﬁxed external sources Jµ . ﬁnite quantities. Slavnov & Faddeev (1980). the Green’s functions of the gauge ﬁeld. consequently it can be included into the overall normalization factor. These formal arguments are rigorous except for the fact that in a gauge theory. First we assume the so-called Lorentz gauge ∂µ Aµ = 0 The M(x. The equivalence of different gauges can be explicitly checked for various cases (see. as we know from Section 2. this is not a gauge-invariant object and the same applies to its functional derivatives with respect to J . In such cases one can discuss IR regularized theory but the regulator must not break gauge invariance. Abers & Lee (1973). One can show that only the renormalization constants attached to each external line depend on the gauge but that is of no consequence for the renormalized S-matrix. IR dimensional regularization can be used as an intermediate step to prove gauge independence of the physical. the Green’s functions are often useful at intermediate stages of calculations and physical quantities may be conveniently deﬁned in terms of them. it is in order to mention here that the renormalized S-matrix elements are gauge-independent. y) = + ∂ 2 δ(x − y) e where ∂ 2 = ∂µ ∂ µ and the Faddeev–Popov determinant is independent of the ﬁeld Aµ . The other gauge-dependent terms do not contribute to the poles of the Green’s functions at pi2 = m 2 .

23) Thus the longitudinal Aµ does not appear in the action and in the Lorentz gauge L δ ∂µ Aµ makes the integration over D AL trivial. Perturbation theory is.22) (the extra term 1 iε AT g µν AT in the action should always be kept in mind). µ AL = (gµν − Pµν )Aν µ ∂µ A µ = 0 T so that the gauge-ﬁxing δ-functional does not affect the integration over the transverse component of the gauge ﬁeld. it requires that ∂ 2 = 0. In Euclidean space we can do it by assuming that the gauge ﬁeld Aµ vanishes at Euclidean inﬁnity. We can simply set Aµ = 0. let us observe that the residual gauge freedom can be removed by an appropriate choice of the boundary conditions. This is the residual gauge freedom allowed by (3. are separated by ﬁnite gauge transformations. In the case of the Lorentz gauge it appears only in non-abelian gauge theories. There is also another sort of gauge ambiguity (the Gribov ambiguity) with more physical implications. however. i. Clearly we have (3. instead. Taking this all into account we obtain the following path integral formula for W0 [J ] in the Lorentz gauge (in the following the fermion degrees of freedom are always suppressed): W0 [J ] ∼ D AT exp i µ d4 x 1 2 T AT ∂ 2 g µν AT + Jµ Aµ µ ν T where Pµν = gµν − ∂µ ∂ν /∂ 2 is the projection operator. Here µ ν 2 we have made use of the fact that the lagrangian density of a free electromagnetic ﬁeld can be written as follows − 1 Fµν F µν = 1 Aµ ∂ 2 g µν − ∂ µ ∂ ν /∂ 2 Aν = 1 AT ∂ 2 g µν AT ν 4 2 2 µ (3. However.130 3 Feynman rules for Yang–Mills theories make use of a decomposition of the ﬁeld Aµ into transverse AT and longitudinal µ AL parts µ AT = Pµν Aν .22) is given by W0 [J ] = exp 1 i 2 T d4 x d4 y Jµ (x) D µν (x − y) JνT (y) (3. unaffected by this ambiguity. The existence of this residual gauge freedom contradicts our previous assumpg tion that the gauge condition F Aµ = 0 has exactly one solution g0 for any Aµ . This is the case when the points of intersection of a group orbit with the F = 0 surface are at a ﬁnite distance from each other. The longitudinal ﬁelds are of the form AL = ∂µ (‘pure gauge’). We L L know that the J -dependence of a path integral like (3. Such boundary conditions are in fact implicit in the usual derivation of perturbation theory rules from path integrals.21).24) .e. The Lorentz gauge condition does not set them exactly µ to zero.

With F[Aµ ] = ∂µ Aµ we obtain the effective lagrangian Leff = − 1 Fµν F µν − 4 1 ∂µ A µ 2α 2 = 1 Aµ gµν ∂ 2 − 1 − 2 1 ∂µ ∂ν Aν α (3. as long as we deal only with transverse sources. Again we can write W0 [J ] = exp 1 i 2 d4 x d4 y J µ (x) Dµν (x − y) J ν (y) (3. so that we have the following equation for Dµν : ∂ 2 − iε Dµν (x − y) = − gµν − ∂µ ∂ν /∂ 2 δ(x − y) After a Fourier transformation this becomes ˜ −k 2 − iε Dµν (k) = − gµν − kµ kν /k 2 so that ˜ Dµν (k) = and Dµν (x − y) = d4 k gµν − kµ kν /k 2 exp[−ik(x − y)] (2π )4 k 2 + iε (3. In transverse space the unit operator is equivalent to Pµν δ(x − y).29) The quadratic action is now no longer proportional to a projection operator and therefore can be inverted over all conﬁguration space. We can now write the formula for the free electromagnetic ﬁeld propagator in the Lorentz gauge G µν (x1 .17)–(3.1 The Faddeev–Popov determinant 131 where −Dµν (x − y) is the inverse of the quadratic action.27).3. Note also that in transverse space.28) where Dµν (x1 − x2 ) is given by (3.25) The prescription of the k 2 pole in the kµ kν /k 2 term can be arbitrary.26) (3. x2 ) = 0 1 i 2 δ 2 W0 [J ] δ Jµ (x1 )δ Jν (x2 ) J ≡0 = −iD µν (x1 − x2 ) (3. A generalization of this result follows if we make use of another procedure for ﬁxing the gauge. we can always replace Pµν by gµν .31) .19). summarized in (3.27) 1 kµ kν gµν − 2 2 + iε k k = ˜ Pµν k 2 + iε (3.30) where Dµν (x − y) satisﬁes the equation ν ∂ 2 gµν − ∂µ ∂ν (1 − 1/α) − iε Dλ (x − y) = −gµλ δ(x − y) (3.

Free propagators of the gauge ﬁeld in the Coulomb and axial gauges can be formally obtained following the procedure summarized in (3. .19).35) Propagator (3. 8 (3. deﬁned by the condition ∂µ Aµ − n µ ∂ µ n µ Aµ = 0. 0) (3.34) n µ = (1. Only in this limit do the propagators behave as k −2 . also called the radiation gauge (Bjorken & Drell 1965). Our formal derivation of (3. The only modiﬁcation is that they are diagonal matrices in the non-abelian quantum number space. In QCD one often uses axial gauges n µ Aµ = 0.132 3 Feynman rules for Yang–Mills theories with the solution kµ kν 1 exp[−ik(x − y)] k 2 + iε k 2 + iε (3.33) which allow the quanta of the vector ﬁelds to be interpreted as partons. inverting the term of the action quadratic in the gauge ﬁelds and taking the limit α → 0.32) In the limit α → 0 this coincides with (3. 0. . The case α = 1 corresponds to the Feynman gauge. Dµν (x − y) = gµν − +α d4 k (2π )4 kµ kν k 2 + iε Non-covariant gauges Non-covariant gauges.36) neglects subtle problems raised by the presence of the (n · k) singularity. 0.28) and (3. The free propagators for non-abelian gauge ﬁelds in covariant gauges are also given by (3. A more complete treatment of n 2 < 0 and n 2 = 0 gauges has been given in the framework of . .17)–(3. α n 2 = 0 or n 2 < 0. α = 1.36) δ αβ k · n(kµ n ν + kν n µ ) − kµ kν gµν − 2 + iε k (k · n)2 − k 2 (3. are also frequently used. It is easy to check that in axial gauges the Faddeev–Popov determinant is Aα independent even µ in a non-abelian theory. depending on an arbitrary four-vector n µ .27) (the Lorentz or Landau gauge). For instance the standard canonical quantization of QED uses the Coulomb gauge. .35) can be split into the transverse propagator and the static Coulomb interaction term (Bjorken & Drell 1965). For the Coulomb gauge one gets ˜ αβ Dµν = and for axial gauges ˜ αβ Dµν = n2 δ αβ kµ n ν + kν n µ + kµ kν gµν − k 2 + iε k·n (k · n)2 (3.32).

3. Dalbosco. Bassetto. It then follows from (3. assume (3.38) follows from (3.15). In the space-like case n 2 < 0 one can show (Bassetto.20) is a consequence of quantization with constraints: gauge-ﬁxing conditions. equivalently.38) Relation (3.2 Feynman rules for QCD Calculation of the Faddeev–Popov determinant In order to derive Feynman rules for a non-abelian gauge theory we have to take proper account of the Faddeev–Popov determinant whose presence under the functional integral (3. = 2 2 n·k k + k⊥ + iε n = (1.37) This prescription has been proposed earlier without any canonical justiﬁcation as one which allows for Wick rotation (Mandelstam 1983. However. y) = (1/g) δ αβ ∂ 2 + gcαβγ Aµ (x)∂µ δ(x − y) = (1/g) Dµ ∂ µ γ αβ δ(x − y) (3. Lazzizzera & Soldati 1985. 0.15) that the Faddeev–Popov matrix reads M αβ (x. Lazzizzera & Soldati 1985). With both kinds of prescription dimensional regularization is usually adopted (Pritchard & Stirling 1979 and Appendix B). 1) (3. The treatment of the light-cone gauge n 2 = 0 is quite different (Bassetto et al. Leibbrandt 1984). 0. . 1985). Lazzizzera & Soldati 1985) that this residual gauge symmetry is directly connected to the spatial asymptotic behaviour of the potentials and can be eliminated by imposing a speciﬁc boundary condition. The matrix M αβ .2 Feynman rules for QCD 133 canonical quantization (Bassetto. Nevertheless a hamiltonian canonical quantization leads to the well-deﬁned prescription for the singularity (n · k): 1 k0 + k3 . For the sake of deﬁniteness we choose a class of covariant gauges ∂µ Aµ = c(x) or. the general proof of the renormalizability of the theory in the light-cone gauge is still lacking. In QCD calculations the principal value prescription has also been extensively used. Choice of this condition gives rise to a speciﬁc prescription for the spurious singularity in the free boson propagator. γ depends on gauge ﬁelds Aµ . The (n · k) singularities are related to the residual gauge freedom still present after the axial gauge condition is imposed.19). This time the residual gauge freedom involving functions which do not depend on the variable n µ x µ cannot be eliminated: we are not allowed to impose boundary conditions which would interfere with the time evolution of the system. not only on the hypersurface ∂µ Aµ = 0 but also for ∂µ Aµ = c(x) because det ∂µ D µ = det Dµ ∂ µ . 3.

41) (x − y) = δ αβ d4 k 1 exp[−ik(x − y)] 4 k 2 + iε (2π ) 2 (3.40) + β ∗ α (x)ηα (x) + η∗ α (x)β α (x) Using standard methods. y) we can take account of det M αβ (x.39) and (3. The Feynman rules for ghost ﬁelds follow in a straightforward way from (3. y) is to replace it by an additional functional integration over some auxiliary complex ﬁelds η(x) (ghost ﬁelds) which are Grassmann variables. .38) of the matrix M αβ (x.42) One can deﬁne the ‘propagator’ of the ghost ﬁeld η G 0 (x. The ghost ﬁeld propagator can be calculated from the generating η functional W0 [β] W0 [β] ∼ η Dη Dη∗ exp i d4 x η∗ α (x)δ αβ ∂ 2 ηβ (x) (3.134 3 Feynman rules for Yang–Mills theories The standard method of dealing with the Faddeev–Popov determinant det M αβ (x.43) β=β ∗ =0 (x − y) (3.39) and the explicit form (3.5 we can write det M αβ (x. Having (3.38) is also included in it) and the integration over η and η∗ is equivalent to integration over the real and the imaginary parts of the ﬁeld η. y) = and expanding (3. y) = −i Graphically. ηαβ αβ ηαβ 1 i δ δβ ∗ α (x) W0 [β] η δ δβ β (y) ← (3.38).41) we get G 0 (x. y) in the framework of our general perturbative strategy for calculating various Green’s functions. y) = C Dη Dη∗ exp i d4 x d4 y η∗ α (x) M αβ (x. y) ηβ (y) (3. propagator ηαβ G 0 (x. the β-dependence of W0 [β] reads explicitly W0 [β] = exp i where αβ η η d4 x d4 y β ∗ α (x) αβ (x − y) β β (y) (3. Using the results of Section 2.44) y) will be denoted by with an arrow pointing towards the source of η.39) where C is some unimportant constant (the factor (1/g) in relation (3.

3. ]= d x 4 (x) + ¯ (x) α(x) (3. . The propagators 0|T Aα (x)Aβ (y)|0 . α.42).47) + ¯ (x) iγµ ∂ µ − m and into the interaction term SI SI [A. 0|T α (x) ¯ β (y)|0 µ ν and 0|T ηα (x)η∗ β (y)|0 are obtained by multiplying the respective D-. ] + S η [A. η] + SG ] α d4 x Jµ (x) Aµ (x) + β ∗ α (x) ηα (x) + η∗ α (x) β α (x) α + α(x) ¯ where S[A.32). η] = d4 x gcαβγ ∂µ Aα Aµβ Aνγ − 1 g 2 cαβγ cαρσ Aβ Aγ Aµρ Aνσ ν µ ν 4 − g ¯ γµ Aµα T α + gcαβγ Aγ η∗ α ∂ µ ηβ µ (3.46) SG = −(1/2α) Dirac spinors . α. The ‘free’ generating functional reads W0 = exp i d4 x d4 y αβ µν 1 α J (x)Dαβ (x 2 µ − y)Jνβ (y) + α(x)α Sαβ (x − y)αβ (y) ¯ (3. respectively. ¯ describe quarks and are vectors in the three-dimensional colour space.122) and (3.49) + β ∗ α (x) (x − y)β β (y) µν where the Green’s functions Dαβ .and -functions by (−i). η] = d4 x 1 2 Aα (x)δ αβ ∂ 2 g µν − ∂ µ ∂ ν /∂ 2 Aβ (x) µ ν (x) + η∗ α (x)δ αβ ∂ 2 ηβ (x) + SG (3. β] = N +i D Aα D µ D ¯ Dη Dη∗ exp i[S[A. β] for QCD reads as follows: W [J. S. . We split the action into the term S0 describing non-interacting theory (bilinear in ﬁelds and including SG ) S0 [A. η] = d4 x η∗ α (x) δ αβ ∂ 2 + gcαβγ Aµ (x)∂µ ηβ (x) γ d4 x ∂ µ Aα µ 2 (3. .48) Matrices T α are SU (3) colour generators in the fundamental representation. Sαβ and αβ are given by (3.45) − 1 G α G µν 4 µν α + ¯ iγµ D µ −m ¯ S η [A. (2.2 Feynman rules for QCD 135 Feynman rules The full generating functional W [J.

51) .136 3 Feynman rules for Yang–Mills theories Fig. . . ¯ |0 ¯ ηη∗ ] Aα . . x3 ) ≡ 0|T Aa (x1 )Ab (x2 )Ac (x3 )|0 µ ν λ (3.50) eliminates all vacuum-tovacuum diagrams. We shall now consider several Green’s functions in the ﬁrst order in SI to derive Feynman rules for vertices in QCD. One is usually interested in Green’s functions in momentum space. .50) We recall that the presence of the denominator in (3. Any Green’s function in QCD can be calculated perturbatively from the expansion 0|T Aα . 3. x2 . Aβ µ ν ¯ ηη∗ ] N D[A . .2.. Aβ µ ν = D[A .. ¯ N N (iSI ) N exp(iS0 ) N! (iSI ) exp(iS0 ) N! (3. We begin with G µνλ (3g)abc (x1 . .

y2 . x2 . κ) and the derivative is with respect to y.3. In the ﬁrst order in SI one has G µνλ (3g)abc (x1 . x3 ) = 1 δ3 i c a i6 δ Jµ (x1 )δ Jνb (x2 )δ Jλ (x3 ) × (d) d4 y gcde f ∂ρ δ3 W0 [J. x3 and y1 . κ). x2 .52) illustrated in Fig. 3. x3 ) = aa bb d4 y −iDµµ (x1 − y) −iDνν (x2 − y) cc × −iDλλ (x3 − y) Vaµbνcλ (y) (3. Following the standard method we assign points x1 . x3 ) = igcde f (d) d4 y ∂ρ ad be −iDµκ (x1 − y) −iDνρ (x2 − y) −iDλ (x3 − y) cf κ × (3.55) ˜ In momentum space we deﬁne G µνλ G µνλ (3g)abc (3g)abc ( p. r ) as (x1 .52) we get (d) abc (d) (d) Vµνλ = igcde f δ ad δ be δ c f gµλ ∂ν + δ a f δ bd δ ce gµν ∂λ + δ ae δ b f δ cd gνλ ∂µ (d) (d) (d) + δ ad δ b f δ ce gµν ∂λ + δ ae δ bd δ c f gνλ ∂µ + δ a f δ be δ cd gµλ ∂ν (3. q. The factor 1/i6 i3 = (−i)3 is included in the three propagators. q. x3 ) = d4 p d4 q d4r exp(−i p · x1 ) exp(−iq · x2 ) (2π )4 (2π )4 (2π )4 ˜ (3g)abc × exp(−ir · x3 ) G µνλ ( p. x2 . (e.56) . Writing the result (3. y3 → y to the prototype diagram For a connected diagram there are six different orderings in pairs joined by propagators.2. α]| J =β=α=0 d e δ Jκ (y)δ Jρ (y)δ J f κ (y) (3. The ﬁnal result reads: G µνλ (3g)abc (x1 . β.2 Feynman rules for QCD 137 which deﬁnes the full symmetrized triple gluon (3g) vertex by means of the relation G µνλ (3g)abc (x1 . x2 . r ) (3.53) (d) (the superscript ‘d’ in ∂ρ means that the component Ad is differentiated with κ respect to y) and the cubic term in the expansion of W0 contributes to the ﬁnal result. x2 .54) in the form (3. ρ) and ( f. For each assignment there is a 23 3! symmetry factor which cancels with factors coming from the expansion of W0 .54) where the sum denotes all permutations of sets of indices (d.

r ) ˜ ˜ G µνλ ( p. . q. x2 .55). q.138 3 Feynman rules for Yang–Mills theories µνλ and the 3g vertex Vabc ( p. x4 ) = (−i)4 aa bb d4 y Dµµ (x1 − y) Dνν (x2 − y) cc dd ×Dλλ (x3 − y) Dσ σ (x4 − y) Vaµbνcλdσ (3. .60) abcd one can calculate Vµνλσ . The additional 4!24 symmetry factor is cancelled by the same factor coming from the expansion of W0 . x4 . q. one can derive the expression for the 4g vertex. x2 . There are 24 different assignments of pairs of points x1 . x4 ) ≡ 0|T Aa (x1 )Ab (x2 )Ac (x3 )Ad (x4 )|0 µ ν λ σ which in the ﬁrst order in SI deﬁnes the 4g vertex according to the following relation: G µνλσ (4g)abcd (x1 . y4 to the prototype diagram.61) . integrating over d4 y and ﬁnally comparing with (3. r ) by the relation ˜ (3g)abc ˜ (2g)aa ˜ (2g)bb (q)G (2g)cc (r )V µ ν λ ( p. x3 .52) in terms of their Fourier transforms (3.59) Writing G µνλσ (4g)abcd (x1 . The starting point is the Green’s function G µνλσ (4g)abcd (x1 .56) and (3. Similarly. Writing the propagators in relation (3. q.57). x3 . x4 ) = 1 δ4 c a d i8 δ Jµ (x1 )δ Jνb (x2 )δ Jλ (x3 )δ Jσ (x4 ) δ4 h δ Jκ δ Jρ δ J mκ δ J nρ (y) f ×i d4 y − 1 g 2 ce f h cemn 4 W0 | J =β=α=0 (3. (3.57) etc.57) one gets the ﬁnal result ˜ abc Vµνλ ( p. Remembering that the Lorentz contraction in the 4g vertex always occurs between colour indices f and m and h and n one gets the ﬁnal result: abcd Vµνλσ = −ig 2 [cabe ccde (gµλ gνσ − gµσ gνλ ) + cace cbde (gµν gλσ − gµσ gνλ ) + cade cbce (gµν gλσ − gµλ gνσ )] which is obviously also true in momentum space. r ) = −(2π )4 δ( p + q + r )gcabc × [(r − p)ν gµλ + (q − p)λ gµν + (r − q)µ gνλ ] (3.58) (where all momenta are outgoing). y1 . x3 . r ) = G µµ ( p)G νν λλ abc where ˜ (2g)aa G µµ ( p) = p2 −i pµ pµ gµµ − (1 − α) 2 δ aa + iε p (3. differentiating according to (3. x2 . . .

3. Fig.2 Feynman rules for QCD 139 Fig. respectively.3. 3. In a similar way one can 2 obtain the other factors. Following the previous calculations closely we . 3.3. 3.3 contribute with additional factors 1 . e. 1 and 1 .4.g.4.58) 36 combinations we are left with an additional factor 1 . Following the functional approach one can also easily ﬁnd that loop diagrams in Fig. Noticing that the product of two vertices gives us according to (3. respectively. The ﬁnal task is to ﬁnd expressions for the ghost and fermion vertices in Fig. They can be obtained from the Green’s functions 0|T ηc (x3 )ηb (x2 )Aa (x1 )|0 and µ 0|T j (x3 ) ¯ i (x2 )Aa (x1 )|0 .62) The number of different assignments to the prototype diagram is 3 × 2 × 3 × 2 × 4!24 . In 2 2 6 particular diagrams (a) and (b) occur in the second order in SI .63) and similarly for the ¯ A vertex. the expression corresponding to the ﬁrst diagram is of the form (iSI )2 1 1 J DJ J DJ J DJ J DJ 2! 4! 24 (3. written as µ aa d4 y −iDµµ (x1 − y) −i bb (x2 − y) −i cc (x3 − y) Vaµb c (y) (3.

the . for instance. An important point is that ﬁxing the gauge is usually not sufﬁcient to totally remove the unphysical degrees of freedom from the theory (an exception is the axial gauge with n 2 < 0). choose another vector n µ such that k · n = 0. Since the three four-vectors kµ . for example. Becchi–Rouet–Stora transformation Unitarity and ghosts A spin-one massless particle has only two physical polarization states εµ (λ.3 Unitarity. k) do not span four-dimensional space.4(a): Diagram 3. A well-known example of their presence in the theory is the Gupta–Bleuler formulation of QED in the covariant gauge ∂µ Aµ = 0. All Feynman rules are given in Appendices B and C. In general. k) = −gµν + λ=1. this restriction does not prevent one from having unphysical states propagating as virtual states in the intermediate steps of perturbative calculations. The polarization sum reads ∗ εµ (λ.67) n·ε =0 (3. such degrees of freedom can be seen explicitly if we construct the Fock space of the theory: a longitudinal photon state and a negative norm scalar photon state are present. Feynman rules for QED in the covariant gauge can be derived analogously. λ = 1.65) Of course.66) specify the vectors ε(λ.68) In ﬁeld theory every particle is associated with a ﬁeld with deﬁnite transformation properties under the homogeneous Lorentz group. de Rafael (1979)). εµ (λ. To make them unique we must sacriﬁce manifest Lorentz covariance and. k) uniquely. k) = −δλ1 λ2 (3. However. 2. Indeed. k)ε(λ2 . the usual on-shell Lorentz condition k · ε = 0 for spin-one massive particles is not enough to determine the photon polarization vectors uniquely. 3.66) conditions (3.64) (3. Together with equations k · ε = 0. ε ∗ (λ1 . Thus the four-vector gauge ﬁeld Aµ (x) actually represents only two physical degrees of freedom.4(b): ˜ abc Vµ = −gccab rµ (2π )4 δ(r + q − p) ˜ abc Vµ = −igγµ T a ji (2π )4 δ(r + q − p) (3.140 3 Feynman rules for Yang–Mills theories obtain in momentum space Diagram 3. ghosts.2 n 2 kµ kν kµ n ν + kν n µ ≡ Pµν − n·k (k · n)2 (3. k)εν (λ. A physically sensible theory can be deﬁned in Hilbert space restricted to the transverse photons (see. k).

12). 3. be quite subtle as . The requirement that the S-matrix is unitary implies that the scattering amplitude T . for the gauge boson propagators in the Feynman gauge (see Section 3. for example. In fact one can expect that gauge invariance.3 Unitarity.s.70).h. gauge boson propagators are gauge-dependent. ghosts. Here we sum over physical states only. the unphysical intermediate states are included in our calculation. An explicit veriﬁcation of this point may.70) where the sum is taken over all physical states for which transitions to the initial and ﬁnal states are allowed by quantum number conservation laws.3. of the unitarity relation (3.1) αβ Dµν = δ αβ k2 gµν → −2πiδ αβ gµν δ k 2 + iε (k0 ) (3. On the other hand we can calculate the absorptive part of the scattering amplitude from the r. Section 18.5.69) (3. Fig. 3. BRS transformation 141 Fig.71) It is clear that in gauges which do not remove the unphysical degrees of freedom. is obtained from the Feynman amplitude by replacing the propagators in this intermediate state by their imaginary parts.70) is just the imaginary part of the amplitude Tii and in the general case it is the absorptive part of Tif (see. Bjorken & Drell (1965). For instance. ensures at the same time their decoupling. which reﬂects the presence of the redundant degrees of freedom. deﬁned as Sif = δif + i(2π )4 δ( pi − pf )Tif satisﬁes the relation Tif − Tﬁ∗ = i n ∗ Tin Tfn (2π )4 δ( pi − pn ) (3. like in the Feynman gauge. The question then arises about the unitarity of the theory. however.5. of (3. The absorptive part of the scattering amplitude can be calculated perturbatively by means of the so-called Landau–Cutkosky rule: the contribution due to a Feynman diagram with a given intermediate state.s.h. For the forward elastic scattering the l. Thus the unitarity condition demands that the unphysical intermediate states do not contribute to the amplitude when also calculated by means of the Landau–Cutkosky rule.

However.8.h. 3.7 with only physical gluons as the state n (Problem 3. amplitudes Tin and Tfn are given by the diagrams in Fig. Nevertheless in QED there is no problem due to the fact that photons always couple to conserved currents kµ M µν = 0 Therefore Pµν given by (3. These are examples of Ward identities for a non-abelian gauge theory. shown in Fig.6) calculated in a covariant gauge (3. 3.s.72) . for solving the equations of motion we need a gauge-ﬁxing term which breaks the manifest invariance. the so-called ghost term (3. The sum of the diagrams with only gauge particles in the intermediate state (Fig. In our speciﬁc example the cancellation occurs due to the contribution of the diagram with the ghost loop. respectively. where Aµ and η are the gauge and µ ν ghost ﬁelds.1) shows that certain relations between the f f¯ → Aα Aβ and f f¯ → ηα ηβ amplitudes.2 that a consistent quantization of a non-abelian gauge theory requires the addition to the lagrangian of an extra term.73) where α is an arbitrary quantity.19) does not satisfy ∗ the unitarity condition (3.142 3 Feynman rules for Yang–Mills theories Fig.1). where the r. are responsible for the cancellation. Aitchison & Hey 1982).70). It turns out that these additional unphysical degrees of freedom cancel the gauge ﬁeld unphysical polarization states exactly.39). 3. In non-abelian gauge theories. 3. An explicit perturbative calculation (Problem 3. the situation is more complicated as can be illustrated by the standard example of the fermion–antifermion scattering in the lowest non-trivial order in the gauge-coupling constant (Feynman 1977. we have learned in Sections 3.68) can be replaced by Pµν → −gµν + αkµ kν (3. leading to a perfectly unitary theory.6.1 and 3. (3.

8. ﬁelds with unphysical statistics. 1976) let us consider the action including the gauge-ﬁxing term and the ghost term 1 F α [x. BRS and anti-BRS symmetry We have introduced ghosts as a technical device to express the modiﬁcation of the functional integral measure required to compensate for gauge-dependence introduced by the gauge-ﬁxing term. Such a symmetry.3. i. BRS transformation 143 Fig. are necessary to compensate for effects due to the quantum propagation of the unphysical states of the gauge ﬁelds. however.7. To introduce BRS symmetry (Becchi.74) . Fig. We shall ﬁrst follow this historical path and later comment on the extension which is independent of the notion of a lagrangian and remains in the spirit of the above outlined programme. 3.3 Unitarity. 3. the so-called BRS and anti-BRS symmetry. that it is very natural to adopt the point of view that ghosts. Provided one is able to express the gauge symmetry in a form involving both the classical and ghost ﬁelds one can introduce both ﬁelds from the very beginning as fundamental ﬁelds of a gauge theory and construct its lagrangian requiring invariance under this extended symmetry principle.e. ghosts. has indeed been discovered. It turns out that given any gauge symmetry one can always ﬁnd symmetry transformations acting on gauge and ghost ﬁelds such that invariance under this symmetry is equivalent to gauge invariance of physical quantities. Rouet & Stora 1974. The discussion in the previous subsection suggests. originally as a symmetry of the Faddeev–Popov lagrangians with speciﬁcally chosen gauge-ﬁxing conditions. A(x)] 2α Seff = S + d4 x − 2 + η∗ Mη = S + SG + Sη (3.

y) = and ρβ Dµ = ∂µ δ ρβ + gcρβγ Aγ µ d4 y Mαβ (x. We observe that for the classical ﬁelds Aµ and the BRS transformation has the form of a gauge transformation with α (x) = +g ηα (x) (3. The full action (3.79) and η are nilpotent (see Problem 3.39) as hermitean conjugates: see Problem 3.76) δBRS ηα = + 1 gcαβγ ηβ ηγ ≡ sηα 2 δ η∗ α = +(1/α) F α ≡ sη∗ α BRS where is a space-time-independent inﬁnitesimal Grassmann parameter (note that it is not necessary to think of η and η∗ in (3.144 3 Feynman rules for Yang–Mills theories where (Mη)α = and M αβ (x.75) Thus (Mη)α = δ Fα [x.2) (3. A(x)] ρβ Dµ (z)δ(z − y) δ Aρ (z) µ d4 z (3.74) is invariant under the global transformation αβ δBRS Aα = Dµ ηβ ≡ s Aα µ µ α α δBRS = −igT η ≡ (s ) (3. We note also that the measure D(A ηη∗ ) is invariant under the BRS transformations. As intermediate steps we record the relation δBRS (Mη) = 0 and the fact that the transformations on Aα .77) Therefore the BRS transformation leaves the original action invariant without the gauge-ﬁxing and the ghost terms.4). µ s 2 Aα = s 2 µ (3. . Thus it remains to show that δBRS (SG + Sη ) = 0 (3. A(x)] where δ Fα is the variation of the functional Fα under a gauge transformation with β (x) replaced by gηβ (x) .78) This can be checked by a straightforward calculation.80) = s 2 ηα = 0 (but s 2 η∗ α = (Mη)α ). y) ηβ (y) δ F α [x.

sbα = 0 (3.81) as our gauge-ﬁxing condition. The nilpotency of BRS symmetry can now be expressed as s2 = 0 (3.3.86) .83) where the minus sign occurs if there is an odd number of ghosts and antighosts in A.3 Unitarity. The BRS transformation can be written down in a form more symmetric with respect to the ﬁelds η and η∗ if we introduce an auxiliary ﬁeld b. ghosts.84) It has also been discovered (Curci & Ferrari 1976.74) invariant provided the gauge-ﬁxing functional F α is linear in ﬁelds.76) depend on the gauge-ﬁxing term of the lagrangian).82) The action of s on an arbitrary function of the ﬁelds follows from their action on ﬁeld polynomials s(AB) = (s A)B ± As B (3. These anti-BRS transformations read αβ s A α = Dµ η ∗ β ¯ µ s η∗ α = 1 gcαβγ η∗ β η∗ γ ¯ 2 α α αβγ ∗ β γ (3.81) is equivalent to the standard gauge ﬁxing term 2 −(1/2α) ∂ µ Aα . Using the freedom we have to choose any gauge condition without affecting the S-matrix we can take LG = bα ∂µ Aµ + 1 αbα bα α 2 (3. In this case the ghost ﬁeld is a free real scalar ﬁeld and the covariant derivative in the ﬁrst equation in (3.85) s η = −b + gc η η ¯ α αβγ ∗ β γ s b = gc η b ¯ α ∗α s = −igT η ¯ and ss + ss = s2 = 0 ¯ ¯ ¯ (3.76) is replaced by the normal derivative (Problem 3. Using the auxiliary ﬁeld b the BRS transformation for η∗ and b µ reads sη∗ α = bα . Ojima 1980) that in addition to BRS symmetry there is another symmetry which leaves the quantum Yang–Mills action (3. This also allows formulation of the notion of BRS symmetry in a lagrangian-independent way (transformations (3. To be speciﬁc let us discuss the case of a covariant gauge. BRS transformation 145 The BRS symmetric lagrangian can also be written down in QED. This step is very important for promoting BRS symmetry to the fundamental symmetry of gauge theories. After taking into account the equation of motion for the ﬁeld b we see that (3.3).

Thus. η] sη∗ = b s η = −[η∗ . which is simultaneously BRS and anti-BRS invariant. in supergravity. Indeed.or/and s -invariant and ¯ Lorentz-invariant function (with ghost number zero) of physical and ghost ﬁelds. absent in the Faddeev–Popov approach. in a BRS-invariant lagrangian four-ghost or higher ghost interactions are present which cannot be obtained from the ordinary Faddeev–Popov procedure. shows that the four-point ghost function is divergent. Actually. have been shown to be necessary (Kallosch 1978. One can also prove that full BRS invariance of the lagrangian leads to the gauge independence of physics. Thus we are led to construct the quantum lagrangian of a gauge theory as the most general s.86). the symmetry generators s and s become independent of the notion of ¯ the lagrangian and satisfy the nilpotency relations (3. η] − b ¯ ∗ sb = 0 s b = −[η . in such cases. In terms of the ﬁelds. η ] ¯ sη = − 2 [η. e In fact.146 3 Feynman rules for Yang–Mills theories We have introduced the BRS and anti-BRS transformations as symmetry transformations of a Yang–Mills lagrangian with a gauge-ﬁxing term linear in ﬁelds. the latter is a consistent approach only for four-dimensional Yang–Mills theories when one chooses a linear gauge condition. This construction simultaneously generates gauge-ﬁxing and ghost terms in the lagrangian. † In Problem 3.84) and (3. . Specifying the general BRS-invariant lagrangian to this special case we indeed recover the lagrangian of Faddeev and Popov. four-ghost interaction terms should have been introduced at the tree level. which are matrix-valued in the Lie algebra of the gauge group. Such couplings are in fact necessary for a consistent renormalizable Yang–Mills theory if the gauge-ﬁxing functional F α is chosen to be non-linear in ﬁelds. Also.87) s η = − 2 [η . we have s A µ = Dµ η s A µ = Dµ η ∗ ¯ ∗ s = −η s = −η ¯ 1 1 ∗ ∗ ∗ (3.5 this is discussed for a less general lagrangian. b] ¯ An important next step is to promote full BRS (BRS or/and anti-BRS) symmetry to the fundamental symmetry of a gauge theory (Alvarez-Gaum´ & Baulieu 1983). With the auxiliary ﬁeld b interpreted as a Lagrange multiplier for the gauge-ﬁxing condition. one-loop calculation. It is not just a formal improvement of the Faddeev–Popov approach since it also gives consistent quantum theories in cases for which the Faddeev– Popov method is inapplicable. quartic ghost interactions.† In the general case. based on the Faddeev–Popov lagrangian generating only quadratic ghost interactions. one can show that starting from a set of inﬁnitesimal gauge transformations building up a closed algebra (possibly with ﬁeld-dependent structure functions) with a Jacobi identity one can always build an associated nilpotent full BRS algebra.

e.5 Check that the most general Yang–Mills and Lorentz scalar lagrangian which is BRSand anti-BRS-invariant and has ghost number zero can be written as follows (Baulieu & Thierry Mieg 1982) L = − 1 (Fµν )2 − 1 (Dµ )2 + ¯ iD / 4 2 + Vinv ( ) + s s ¯ 1 2 2 Aµ + Yukawa terms + εµνρσ Fµν Fρσ + 1 αb2 2 + βη∗ η + γ where α. Hirshfeld & Leschke 1981) to construct the BRS.1 Calculate in a covariant gauge the absorptive part of the amplitude corresponding to the sum of diagrams in Fig. . 3. as hermitean ﬁelds.70). Townsend & van Nieuwenhuizen 1978).76).39) the Faddeev–Popov term Lη can be made hermitean. i.3 Check the invariance of the generalized QED lagrangian L = − 1 Fµν F µν + ¯ (iD − m) / 4 − (1/2α) ∂µ Aµ 2 − 1 (∂µ ω) ∂ µ ω 2 where ω(x) is a free scalar ﬁeld. 3. Check that.80). Sterman. T b = icabc T c with T a bc = −icabc . Check that due to the phase arbitrariness of the exponent in (3. γ are arbitrary gauge parameters and is the vacuum expectation value of the boson ﬁeld .2 Prove the invariance of (3.) 3. Problems 3. Follow the same method in supersymmetric Yang–Mills theory (Falck. under the transformation δ = +ie ω(x) (x) ∂µ ω(x) δ Aµ = ∂ω = (1/α) ∂µ Aµ 3. 3.79) and (3. They are naturally generated by determining the quantum lagrangian from the requirement of its BRS invariance.6 Use the superspace formulation of the Yang–Mills theory (Bonora & Tonin 1981. full BRS invariance indeed emerges as the fundamental symmetry of the quantum theory associated with any given underlying gauge invariance. (Use the relation for the structure constants which follows from the commutation relations T a . Thus. Hirshfeld & Kubo 1983) and in supergravity. 3. they can be taken as real and imaginary parts of η.Problems 147 Nielsen 1978. Check (3.4 Note that the ﬁelds η and η∗ in (3. in particular.6 and Fig. β.39) can be any anticommuting ﬁelds. 3.and anti-BRS-invariant lagrangians. When β = 0 one recovers the usual Faddeev–Popov lagrangian for linear gauges.8 and check the unitarity condition (3.74) under transformation (3.

We shall now discuss this problem in some detail. the ‘bare’ coupling constant λB is not a correct expansion parameter.1) are the regularization of divergent integrals and the subsequent renormalization of the parameters and the Green’s functions. like λB and the ‘bare’ mess m B (as well as the Green’s functions of the ‘bare’ ﬁeld B ) are not directly related to observable quantities. most likely. After renormalization.4 Introduction to the theory of renormalization 4. inﬁnite. for instance. Two basic steps in interpreting a theory like (4. The renormalization procedure accounts for the fact that.1 Physical sense of renormalization and its arbitrariness Bare and ‘physical’ quantities Let us consider a theory of interacting ﬁelds deﬁned by a given lagrangian density. the coefﬁcients of the perturbative series are.1) We know already how to derive the Feynman rules which allow us to calculate the Green’s functions of this theory in the form of the perturbative expansion in powers of the coupling constant λB . for deﬁniteness. reﬂecting the fact that. our theory does not properly describe the physics at very short distances. the observable quantities are expressed in terms of experimentally measured parameters and our ignorance about the short-distance physics is hidden in those free parameters of the theory determined by experiment. let us take the case of a scalar ﬁeld with quartic coupling L = 1 (∂µ 2 B) 2 − 1 m2 2 B 2 B − λB 4! 4 B (4. This is not so unnatural as it seems: the ‘bare’ parameters of the theory. in general. However. some of the loop momentum integrations in the resulting Feynman diagrams are divergent in the UV. Does this mean that the lagrangian (4. A regularization procedure is necessary to make sense of the intermediate steps of perturbative calculations. 148 . Consequently.1) cannot be used to deﬁne a theory? In the following we shall show that this conclusion is not necessarily true.

y) = 0|T (x) (y)|0 and insert the complete set of states (2. The normalization of a ﬁeld operator (x) is directly related to the on-shell residue of its two-point Green’s function in momentum space. ﬁrst). The on-shell truncated Green’s functions of the physical ﬁeld have a straightforward physical interpretation: they are equal to the appropriate S-matrix elements. Consider thus the two-point Green’s function of the ﬁeld G (2) (x. If now follows from (2. agree with the result of the perturbation theory calculation for ‘bare’ Green’s function: G (2) (x.5) for the ‘physical’ ﬁeld F does not.189) between the the one-particle intermediate state is then of the form (4.5) The result (4. in general. The contribution of d3 k [ 0| (x)|k k| (y)|0 (x0 − y0 ) (2π)3 2k0 + 0| (y)|k k| (x)|0 (y0 − x0 )] d4 k (k0 )δ(k 2 − m 2 ){ (x0 − y0 ) exp[−ik(x − y)] = F (2π)3 + (y0 − x0 ) exp[ik(x − y)]}| 0| (0)|k |2 i d4 k = exp[−ik(x − y)]| 0| (0)|k |2 (2π)4 k2 − m 2 + iε F (4.2) s.1 Physical sense of renormalization and arbitrariness 149 First let us recall the deﬁnition of the ‘physical’ ﬁeld k| F (x)|0 F given in Section 2. Therefore.4) F (x) note that 0| (0)|k is a constant from relativistic invariance. y) = 0|T B (x) B (y)|0 (with some regularization introduced to keep B .192) = exp[i(k0 x0 − k · x)].4. the Fourier-transformed two-point function near the mass-shell behaves as ˜ G (2) (k) ≈ k 2 →m 2 F i | 0| (0)|k |2 k 2 − m 2 + iε F (x) = (4. This is convenient because we prefer to deal with Green’s functions and not ﬁeld operators throughout.h.192) that ˜F G (2) (k) ≈ k 2 →m 2 F it i k 2 − m 2 + iε F (4.3) (the last step can be easily checked by calculating the r.7: (2.s. k0 = (k2 + m 2 )1/2 F The ‘physical’ ﬁeld F is thus deﬁned by a normalization condition on the one-particle state created by F . Off-shell the equality also holds for analytic continuations.

in momentum space. .7) which implies a simple relation between the Green’s functions (connected or not): (n) G F (x1 . the physical Green’s function. . xn ) = Z 3 −n/2 G (n) (x1 . xn ) B (n) B (x 1 . We expect that a relation between observables should be free from inﬁnities.150 4 Introduction to the theory of renormalization the loop integrals ﬁnite). . . the ‘physical’ Green’s functions are observable and must come out ﬁnite in perturbation theory. Suppose that we deﬁne the coupling constant λF as the value of the connected (4) proper vertex function ˜ F ( pi ) at some speciﬁed point.10) (4) ˜B We can use the regularized bare perturbation expansion to calculate point.9) which also holds for the truncated Green’s functions that appear in the expression (2. Consequently. . x n ) (4.11) . We shall ﬁnd instead that ˜B G (2) (k) ≈ where Z3 = 1 and is actually divergent in the absence of regularization. In this way we would obtain the expression of one observable quantity. . We can then attempt to express other physical Green’s functions as perturbation series in powers of λF .190) for the S-matrix elements. . The ‘bare’ Green’s functions by themselves are not observable. the relation between the S-matrix and ‘bare’ Green’s functions involves the renormalization constant Z 3 .8) and its counterpart for the connected proper vertex functions (n) F (x 1 . . Let us denote the renormalization point by µ. . . the coefﬁcients of the resulting perturbation expansion should be ﬁnite for observable quantities if the perturbation theory applies. . Consequently. The ‘bare’ ﬁeld B is thus not ‘physical’.6) = Z3 1/2 F (4. The result we shall write as (4) ˜ B |µ = at the same λB Z1 (4. . in terms of other observable quantities: λF and the physical mass. This is not yet the case in the ‘bare’ perturbation expansion in λB . the ‘renormalization point’. . however. On the other hand. We can. . x n ) = Z3 n/2 (4. . express the original ﬁeld B in terms of the physical ﬁeld F B k 2 →m 2 F Z3 k 2 − m 2 + iε F (4. . By deﬁnition we have (4) ˜ F |µ = λ F (4.

(4.1 Physical sense of renormalization and arbitrariness 151 Fig. Deﬁned as a pole of the scattering amplitude. between ‘bare’ and ‘physical’ quantities. where the central bubble denotes the full propagator.12) The physical mass m 2 is deﬁned as the square of the four-momentum of a freely F propagating particle (see (2. 4. 4. m 2 is an F experimentally measurable parameter.1. which deﬁnes the new renormalization constant Z 1 .13) which can be either ‘bare’ or ‘physical’. This deﬁnition is convenient because it ˜ avoids multiple poles present in G (2) ( p) calculated to any ﬁnite order. Using these relations we can now rewrite the original .70). Using (4.4. Again. (3. we have three independent renormalization constants: Z 0 . Therefore the full propagator must have a pole at p 2 = m 2 F ˜ G (2) ( p) = R + ··· − m2 F p2 and the physical mass can also be deﬁned as the value of p 2 for which the full propagator has a pole. expressed in terms of the scattering amplitude Ti f . In perturbation theory the general F structure of the amplitude with a one-particle intermediate state corresponds to the diagram in Fig. it then follows that the amplitude Ti f for the process i → f which can proceed via a one-particle intermediate state has a pole at p 2 = m 2 .69) and (3. Equivalently.12) and (4.9) we now obtain 2 λB = (Z 1 /Z 3 )λF (4. we shall deﬁne m 2 as the position of the zero of the inverse F propagator (see (2.14).192)).7). (4. Z 1 and Z 3 corresponding to three relations. From the unitarity condition for the S-matrix.1. we 2 can calculate m F in the regularized bare perturbation expansion and deﬁne the mass renormalization constant Z 0 by the relation m 2 = (Z 3 /Z 0 )m 2 F B (4.152)) ˜ (2) ( p 2 = m 2 ) = 0 F (4.14) Altogether.

A physically important example is. We have introduced the renormalization constants as multiplicative constants. and . for instance. The counterterms can be properly deﬁned only in a regularized theory. B F −2 λB = Z 1 Z 3 λF (4. The perturbation theory based on this part of the lagrangian alone.15) we have divided the lagrangian into two parts.1) in terms of new ﬁelds L = 1 (∂µ 2 − + where B B) 2 and parameters m F . . we say that the theory is renormalizable. λB = Z 3 (λF + δλ). unless subject to some regularization procedure. This is not necessary. renormalized by the fermion loop contributing to the proper 4 vertex function and this contribution is proportional to g 4 but not to λ. with the masses m and M. the renormalization of m 2 requires an additive term δm 2 which has a part proportional to M 2 and vice versa. with the couplings λ 4 and g ) (Yukawa coupling).15) = Z3 1/2 F. Z i -dependent part of the lagrangian: the ‘counterterms’. −1 m2 = Z0 Z3 m2 . for example.1) with the original parameters replaced by physical ones.152 4 Introduction to the theory of renormalization F lagrangian (4. We expect that these divergences will be cancelled order-by-order if we take into account the second. is more appropriate. in fact. The ﬁrst part is identical to (4. −1 −2 2 2 2 m B = Z 3 (m F +δm ) and λB = Z 3 (λF +δλ) is. If this procedure works in any order of perturbation theory with ﬁnite number of counterterms. φ and . After the cancellation of the regulator-dependent terms. would give rise to divergences analogous to those of the ‘bare’ perturbation expansion. Thus. respectively. Counterterms and the renormalization conditions In (4.16) The resulting ‘renormalized’ lagrangian we shall use as the basis for the formulation of the new perturbation expansion parametrized by the ‘physical’ quantities: m F and λF . In this case. general enough to include both cases. It is this ‘physical’ perturbation theory and its generalizations described below that will be used in calculations: we shall not refer to the ‘bare’ lagrangian explicitly. Another important example is a theory of scalar and fermion ﬁelds. The latter notation. The ﬁnal important remark is the following. and for the parameters of the lagrangian (couplings and masses) it is not always convenient. The former is. a scalar ﬁeld theory with two ﬁelds. an additive −2 renormalization of λ. the regulator can be removed. λF 1 Z (∂ )2 − 1 Z 0 m 2 2 F F 2 3 µ F 2 2 − (λF /4!) 4 F F 2 − (Z 1 − 1)(λF /4!) 4 F F 2 4 B − (λB /4!) B = Z 1 (λF /4!) 4 = 1 (∂µ F )2 − 1 m 2 F 2 2 F 1 (Z 3 − 1)(∂µ F )2 − 1 (Z 0 − 1)m 2 F 2 2 − 1 m2 2 B (4.

in fact.18) where (2) ˜ F ( p) = p 2 − m 2 − F (4. present in the Feynman rules of the ‘physical’ perturbation expansion.15) is assumed to be the ‘physical’ ﬁeld. Observe that we now use m F and λF as free parameters rather than m B and λB : this is also more natural physically.17) F ∂ p2 F which implies (2) ˜ F ( p) −→ p 2 − m 2 F p 2 =m 2 F in accord with the fact that ﬁeld and F in (4.1) is renormalizable if considered in the space-time of dimension d ≤ 4. we shall ﬁx the counterterms by imposing the ‘renormalization conditions’ on the physical Green’s functions. We shall always use the ‘physical’ perturbation expansion and never refer to the ‘bare’ parameters at any stage of calculations. Z 1 and Z 0 by relations like (4. can be determined from the renormalization conditions order-by-order. The counterterms. Instead. (4.7).1 Physical sense of renormalization and arbitrariness 153 The theory deﬁned by (4. To ﬁx all counterterms we need two more conditions. The renormalization constants Z 0 . We have. However. which we choose as follows ∂ (2) ˜ ( p)| p2 =m 2 = 1 (4. Now let us make the following important observation concerning the practical application of our results. Arbitrariness of renormalization We have outlined the procedure which replaces the original ‘bare’ perturbation expansion with divergent coefﬁcients by a new perturbation expansion in which the coefﬁcients are ﬁnite.12) and (4. From this point of view to deﬁne Z 3 .10). =0 F ( p) F ( p)| p 2 =m 2 F (4. How unique is this procedure? We have used the physical observability of new parameters and Green’s functions as a guide in order to ensure that they are ﬁnite and in ﬁnite relation to each other. if the perturbation theory parametrized by m F and λF is ﬁnite we would not expect to lose this property . not observable quantities and even if expressed in the new ‘physical’ perturbation expansion in terms of λF and m F are divergent order-by-order after the regularization is removed. Z 1 and Z 3 are. already used one such condition to deﬁne λF : it is (4.4. of course.14) is inconvenient because these relations involve ‘bare’ quantities.19) This choice is exactly equivalent to our previous deﬁnitions.

11). As a result. First of all we can change the renormalization point µ: µ → µ . it will now have to be expressed in terms of the new parameter λF . in powers of λF . Let us now change the value of the mass renormalization constant Z 0 : Z 0 → Z 0 keeping Z 3 unchanged. This renormalization symmetry of the theory is very important. must change: Z 1 → Z 1 . We thus have an inﬁnite. we do not have to require that the coupling constant parameter is equal (4) to the value of ˜ F at any renormalization point. it is the basis of the renormalization group and the renormalization group equations. and not only because it allows one to choose the most convenient renormalization prescription. deﬁned as m 2 = (Z 3 /Z 0 )m 2 R B (4. λR can be expressed in terms of λF in the λF -perturbation expansion. the value of λF changes: λF → λF . or vice versa. deﬁned by (4. The Green’s functions are still ‘physical’ but have a new perturbation expansion. In other words once we have speciﬁed the counterterms which cancel the inﬁnities we can make a ﬁnite change in them. It is possible to do this transformation without affecting Z 3 or Z 0 : the ﬁeld F is still given by (4. Instead of F we can use the ‘renormalized’ ﬁeld R provided that their relative normalization is perturbatively . As m 2 is ﬁxed and the physical content of the theory should not change we must B conclude that the new mass parameter. As the bare parameters are ﬁxed.e. In terms of the ‘renormalization conditions’ discussed before this may correspond to replacing (4. We can use any deﬁnition as long as the transition from λF to the new parameter λR is perturbatively ﬁnite: i.18) by R ( p)|µ =0 (4.20) is no longer the physical mass. n-parameter class of different ‘renormalization prescriptions’ (n is the number of counterterms). One important example of a renormalization procedure which makes use of this generalization is the minimal subtraction scheme. however.21) where now (2) ˜ R = p2 − m 2 − R R ( p) (4.7) and therefore does not change. which we shall discuss in detail in the following sections. Finally we can relax the requirement of the ‘physical’ renormalization of the two-point renormalized Green’s function. also the renormalization constant Z 1 . ‘physical’ scheme. Consider some examples.22) and µ is some renormalization point. Similarly. So far we have kept to the original.154 4 Introduction to the theory of renormalization if we change the parametrization and use parameters which differ from m F and λF by a ﬁnite amount. the ‘renormalized mass’.

.15) can be rewritten in terms of the new renormalized ﬁelds and parameters L = L(Z 3 = L( where in the λ 4 2 2 R . . (Z 1 /Z 3 )λR .23) F R where Z R = Z 3 /Z 3 is ﬁnite in the perturbation expansion.) 2 R . . . . . xn ) R (4.28). . kn = lim ki2 →m 2 F i −(n+m)/2 ˜ (n+m) GR ×Z R (k1 . . . . The counterterms provide vertices. . . .29) The Feynman rules of the renormalized perturbation theory follow from the renormalized lagrangian (4. (Z 0 /Z 3 )m R . we have R = ZR 1/2 F (4.1 Physical sense of renormalization and arbitrariness 155 ﬁnite. km . The relation between the S-matrix and the renormalized Green’s functions has the following form k1 . The counterterms are constructed according to a renormalization prescription. This does not make the renormalized Green’s functions very ‘unphysical’ because the Z R factors are ﬁnite order-by-order.24) 2 →m 2 k 2 − m 2 + iε k F F or.4. The value of Z R can be determined if we know the on-shell residue of the two-point renormalized Green’s function iZ R ˜R G (2) (k) ≈ (4. xn ) = Z R −n/2 (4. .e.27) which generalizes (2.28) theory R) 2 L = 1 (Z 3 − 1)(∂µ 2 − 1 (Z 0 − 1)m 2 R 2 2 R − 1/4!(Z 1 − 1)λR 4 R (4. . m R . with Z dependent coupling constants: (Z 3 − 1) p 2 . km |S|k1 . (Z 0 − 1)m 2 . i. the lagrangian (4. . Note that R the counterterms are treated strictly perturbatively: when calculating the nth order correction it is enough to know the counterterms up to this order only. .190). . equivalently ∂ (2) 1 ˜ (k)|k 2 =m 2 = 2 R F ∂k ZR Due to the relation (n) G F (x1 . Of course. . λR . . −k1 . . One possibility is the ‘physical’ renormalization prescription described before. . . −kn ) 1 2 (k − m 2 ) F i i (4. A .) + L (4. (Z 1 − 1)λR . .25) G (n) (x1 .26) the factors of Z R will also appear in the relation between the renormalized Green’s functions and the S-matrix elements. . .

The reason is that the ˜R perturbation expansion for G (2) (k) has the form ˜R G (2) (k) = i i i i + 2 [−i R (k)] 2 + 2 [−i 2 2 2 − mR k − mR k − mR k − m2 R i i × [−i R (k)] 2 + ··· = 2 2 2 k − mR k − m R − R (k) k2 R (k)] 2 k i − m2 R (4. ∂k 2 R R (k)|µ = 0. (2) Assume that ˜ R (k) calculated up to order n in some renormalization scheme is regular at k 2 = m 2 and expand F R (k 2 )= 0 + (k 2 − m 2 ) F 1 + (k 2 − m 2 )2 F 2 + ··· (4.31).s.30) Clearly the number of the equations must be equal to the number of the renormalization constants. This procedure is particularly convenient because the counterterms can be determined from the divergent parts of the diagrams alone.32) where m 2 is understood as the value of k 2 at which the nth order approximation to F . denoting −i R (k). We see that. To obtain a single pole one must sum the whole geometric series (4.31) corresponding to the diagrams with the blobs on the r. ˜ G (2) (k) has multiple poles at k 2 = m 2 instead of a single pole at k 2 = m 2 required R F by unitarity.h. (4) ˜ R (k)|µ = λR (4. we determine the physical mass m 2 from the requirement that the inverse F (2) propagator ˜ R (k) = k 2 − m 2 − R (k) vanishes at k 2 = m 2 .22)) ∂ (2) ˜ (k)|µ = 1. We shall discuss this scheme in detail in Section 4. We use this condition R F ˜R in preference to the one involving the propagator G (2) (k). after the introduction of dimensional regularization.156 4 Introduction to the theory of renormalization more general case corresponds to specifying ﬁnite values for some renormalized Green’s functions at a renormalization point.3. to any ﬁnite order. This is the case in the minimal subtraction scheme. As explained before. Final remarks Let us ﬁnish this section with some remarks concerning the calculation of the scattering amplitudes in the renormalized perturbation expansion. For instance ( R is deﬁned by (4. Another possibility is to assume some trial form for the counterterms and adjust the parameters to make the renormalized Green’s functions ﬁnite order-by-order.

. Here we shall give only a general idea of the proof.35) In the lowest non-trivial order (one-loop approximation) we can equivalently replace (1 − 1 )−(n+m)/2 by (1 + 1 (n + m) 1 ) so that we then have the correction 2 (+ 1 1 ) for each external line. . Some examples from QED and λ 3 theories will also be considered.25) ZR = Consequently.2 Classiﬁcation of the divergent diagrams (2) ˜ R (k) vanishes. also called the 1PI Green’s functions. . . . we should still prove that the counterterms induced by the renormalization. . −k1 . km . They appear to have the same form as the counterterms induced by the renormalization described in Section 4. However.33) and. We shall concentrate on the structure of UV divergences which appear in Feynman diagrams of λ 4 perturbation theory (with counterterms not yet included) and use the results to determine the form of counterterms from the requirement that these divergences are cancelled. .34) 1 (n+m)/2 1) ˜R ×G (n+m) (k1 . hence 0 157 = m 2 − m 2 .2 Classiﬁcation of the divergent diagrams Structure of the UV divergences by momentum power counting In the previous section we have described the procedure of renormalization.4. (4. The basic blocks of a renormalization programme are the connected proper vertex functions (n) . We shall often refer to this rule in one-loop 2 calculations.27) becomes k1 . We obtain F R 1 (2) ˜ R (k) = (k 2 − m 2 )[1 − F − (k 2 − m 2 ) F 2 − · · ·] (4. −kn ) ˜ = (1 − 1 )−(n+m)/2 G (n+m) R(trunc) (4. . 4. . kn 1 2 (ki − m 2 ) (1 − = lim F 2 →m 2 i ki F i 1 1− (4. Let us recall that these are truncated Green’s functions which remain connected when an arbitrary internal line is cut. are enough to cancel UV divergences due to loop momentum integrations. In other words. . which are of the same general structure as vertices of the original lagrangian. .1. . In a renormalization programme these are important for two . using (4. we have not given the proof that this procedure actually leads to ﬁnite results: we still have to show that a theory like λ 4 is renormalizable. km |S|k1 .

First. so that the momentum integrations in the two distinct parts are independent of each other. Let us next assume that only some subset S of the integration momenta is large and of the order . → ∞. In this way we can assign a subdiagram and a degree of divergence to each loop subintegration. For a diagram with V vertices and I internal lines the number of independent loops is L = I − V + 1. The integrand will then behave as a power of . This is because the connected proper vertex functions can themselves be understood as generalized vertices of an effective lagrangian which includes all loop corrections: the generating functional [ ].158 4 Introduction to the theory of renormalization reasons. Suppose that all integration momenta become large simultaneously: ki → ki . This fundamental result will be the basis of our discussion. It is easy to see that D S can also be regarded as a superﬁcial degree of divergence of a subdiagram built up of all the internal lines of the complete Feynman diagram with momenta depending on those in S: L S is then the number of loops in this subdiagram. One deﬁnes the so-called superﬁcial degree of divergence of a Feynman diagram D = 4L − N (4. states that a Feynman integral converges if the degree of divergence of the diagram as well as the degree of divergence associated with each possible subintegration over loop momenta is negative. For example. all loops and therefore all divergences are present already in the 1PI Green’s functions. say −N . In the above example diagram (a) has D = 2 while (b) and (c) both have D = 0. The ﬁrst thing to note is that in λ 4 theory (and any other interacting quantum ﬁeld theory in four dimensions) diagrams with D ≥ 0 are present. where L S is the number of integration momenta belonging to S. each corresponding to a d4 ki momentum integration. each loop included entirely in one of the 1PI parts. in the case of the two-loop diagram (a) in Fig. The UV divergences we are concerned with come from the region of integration where some of the integration momenta are large. Consequently. The integrand will now be proportional to another power of : −N S . . A theorem due to Weinberg.36) where L is the number of independent loops.2 the subdiagrams (b) and (c) correspond to the subintegrations over l and k. We can introduce the superﬁcial degree of divergence D S of this subintegration: D S = 4L S − N S . 4. Some information about the possible structure of these divergences can be obtained by the simple procedure of counting the powers of momentum in the Feynman integrand. any other Green’s function is made up of 1PI Green’s functions joined together to form a tree. The second reason is that the identiﬁcation of the necessary counterterms which are to be interpreted as additional vertices in the renormalized lagrangian is particularly simple if one looks at the divergences of the 1PI functions. respectively.

[ ] = [m](n−1)/2 (4. Some .36) reads D = 4L − N = 4L − 2I (4. respectively. It ∂ follows that [ ] = [m](n−2)/2 . We set h = c = 1 so that the ¯ action integral must be dimensionless. IF are the numbers of photon and fermion internal lines.36).38) to the case of an n-dimensional space.37) which is applicable to the case of fermion QED. 4.39) In the same manner the dimensions of a coupling constant follow from the requirement that the corresponding term in the action is dimensionless. respectively.4. Note that by replacing 4L with n L we obtain a generalization of (4. are dimensionless. (4. while length and mass are measured in inverse units: [x]−1 = [m].37) and (4. We shall see that D is determined by the number of external lines of the diagram. which are.2. The canonical dimensions of Bose and Fermi ﬁelds are determined from the requirement that the kinetic terms in the action. We can also write a generalization of (4. again with restriction to 1PI diagrams D = 4L − 2IB − IF (4. Then. The argument is based on dimensional analysis so let us ﬁrst recall the dimensions of some quantities in quantum ﬁeld theory. Consider the case of a scalar ﬁeld theory without derivative terms in the interaction (λ 4 is one example).38) Here IB .2 Classiﬁcation of the divergent diagrams 159 Fig. for any 1PI diagram (4.37) where I is the number of internal lines. of the form dn x ∂ 2 and dn x ¯ i/ . Classiﬁcation of divergent diagrams To ﬁnd out what counterterms are required we must examine all 1PI Green’s functions with D ≥ 0 which appear in the theory. In this formula the restriction to 1PI diagrams is due to the fact that propagators corresponding to the internal lines which do not belong to loops are independent of the integration momenta and consequently do not contribute to N : in 1PI diagrams such lines are absent.

One is that in any theory with coupling constants all of positive dimension in the units of mass (λ 3 for example) the number of 1PI diagrams with D ≥ 0 is ﬁnite. 4. Here (a).4).42) (also see Problem 4. the superﬁcial degree of divergence coincides with the dimension of the diagram in the unit of mass. the single elementary vertex for which the superﬁcial degree of divergence is undeﬁned. If the diagram also contains vertices with dimensional coupling constants of total dimension in the units of mass (for example. (d) is a constant ‘tadpole’ which contributes to the vacuum expectation value of the ﬁeld and can be disregarded in the = 0 case (diagrams of this kind will be important in theories with spontaneous symmetry breaking) and (e) produces an inﬁnite mass renormalization counterterm which is the only divergent counterterm in λ 3 theory with = 0. [g] = [m]2−n/2 (4.3. As all contributions to a given Green’s function must be of the same dimension it then follows that with dimensionless couplings. as 1/k 2 and 1/k. A theory like this is called super-renormalizable. [m]−n [λ][m]4(n−2)/2 = 1. This formula has some important consequences.160 4 Introduction to the theory of renormalization examples are λ λ 3 4 : : [m]−n [λ][m]3(n−2)/2 = 1. Assume that at large momenta the Bose and Fermi propagators behave. = K for K fermion mass insertions or cubic scalar couplings) D = 4− B − 3F − 2 (4. i. for any 1PI Feynman diagram built up of vertices with dimensionless coupling constants only. The exception is the possible lowest order no-loop contribution. In four dimensions / the quartic (λ 4 ) and QED (g ¯ A ) coupling constants are dimensionless while the cubic (λ 3 ) coupling constant has the dimension of mass. which is [m]4−B−3F/2 in four dimensions. Then. this applies to 1PI diagrams with dimensionless couplings only. the superﬁcial degree of divergence of a diagram is determined solely by the external lines.40) where [m]−n is the dimension of the dn x integration measure. / g ¯ A : [m]−n [g][m]2(n−1)/2+(n−2)/2 [λ] = [m]3−n/2 [λ] = [m]4−n = 1.e. The canonical dimension of a 1PI diagram with B external boson lines and F external fermion lines is equal to the dimension of a coupling constant corresponding to the vertex of this structure. (b) and (c) are irrelevant vacuum diagrams: they cancel with the normalization factor of our Green’s functions. In particular. respectively.41) As explained. there are no D ≥ 0 1PI . In the case of the λ 3 theory the diagrams with D ≥ 0 are as shown in Fig. Consequently D = 4− B − 3F 2 (4.

41) applies: all diagrams contributing to a speciﬁc 1PI Green’s function have the same degree of divergence. like λ 4 in more than four space-time dimensions or G( ¯ )2 in four-dimensional space.43) . any coupling constant of negative dimension. the perturbation expansion of any 1PI Green’s function must include D ≥ 0 diagrams when the order is high enough.2 Classiﬁcation of the divergent diagrams 161 Fig. Each 1PI Green’s function requires new counterterms and new renormalization conditions: consequently. Let { f } D denote the sum of ﬁrst (D + 1) terms of a Taylor expansion of the Feynman integrand f in powers of the external momenta with D equal to the superﬁcial degree of divergence of the diagram considered.4 (the triple photon proper vertex also has D = 1 but it vanishes because the photon has odd charge conjugation). For the λ 4 theory and for QED we have the D ≥ 0 connected proper vertex functions shown in Fig.3. 4. Necessary counterterms To make the divergent diagrams ﬁnite we must ﬁnd a way to lower the associated degrees of divergence.4. diagrams with three external lines and consequently no divergent coupling constant renormalization counterterm. An important example is Fermi’s theory of weak interactions. the theory is non-renormalizable. so that (4. If we replace the integrand of this diagram by a subtracted expression f − { f }D (4. 4. Consider a 1PI Feynman diagram with D ≥ 0. Let us now return to theories with dimensionless coupling constants. On the other hand in a theory which involves. This can be achieved by making a subtraction in the Feynman integrand. among others.

after the integration it will produce a polynomial ‘counterterm’. As the second term is a polynomial in the external momenta. or a combination of counterterms. We recognize the counterparts of the wave-function renormalization constant ( p 2 is equivalent to −∂ 2 ) and the mass counterterm. . Obviously the counterterms will diverge when the intermediate regularization is removed.162 4 Introduction to the theory of renormalization Fig. We then have two D ≥ 0 Green’s functions: ˜ (2) ( p) with D = 2 and ˜ (4) ( p1 . This produces two counterterms Ap 2 + B The third one. Consider the case of the λ 4 scalar ﬁeld theory.43) separately. proportional to pµ . from Weinberg’s theorem. The twopoint function has D = 2 and consequently requires subtraction of the ﬁrst three terms of the Taylor expansion of the integrand. A similar procedure should be applied to divergent subdiagrams.4. In a renormalizable theory we should be able to understand these subtractions as due to the inclusion of counterterms. 4. the superﬁcial degree of divergence will become negative.43) to ultimately give a ﬁnite result provided that we have dealt with all divergent subdiagrams. The four-point function has D = 0 and only one (constant) counterterm which corresponds to the coupling constant counterterm. p4 ) with D = 0. We conclude that the subtractions necessary to lower the degree of divergence of the D ≥ 0 Green’s functions are . . With some procedure for the regularization of divergent integrals we can calculate the Feynman integrals of the two terms in (4. . vanishes because of relativistic invariance: all counterterms must be scalars. they must combine with the other part of (4. however. .

equivalent to including the counterterms of structure as implied by renormalization of the original lagrangian.2 overlap each other.6.5.5. Can these counterterms. We ﬁrst identify all lowest order (one-loop) divergent subdiagrams. where . Note that in this case the dot is not the complete lowest order coupling constant counterterm.6. Another example of such ‘overlapping divergences’ is the QED diagram shown in Fig. 4. this argument runs into difﬁculties: one problem is that we cannot always make subtractions independently in different subdiagrams.2 Classiﬁcation of the divergent diagrams 163 Fig. One example is shown in Fig.4. subdiagrams (b) and (c) in Fig.7. Fig. For instance. This is because we make the subtraction in the indicated subdiagrams. In each of these we make the appropriate subtractions. We can envisage the following procedure. if included in the interaction lagrangian. Consider a diagram with no counterterm vertices which includes some D ≥ 0 subdiagrams. also deal with all divergences due to the D ≥ 0 subdiagrams? Again (in the λ 4 case) these are the subdiagrams with two (D = 2) or four (D = 0) external lines. so that no new types of counterterm are apparently required. The boxes indicate the divergent subdiagrams we are considering. Unfortunately. 4. 4. 4. 4. The effect of these subtractions should be equivalent to adding to the original diagram all diagrams obtained by replacing some of the considered lowest order subdiagrams by the corresponding lowest order counterterms. 4. and not in complete one-loop vertex functions which are shown in Fig. This procedure should then be repeated with respect to higher order D ≥ 0 subdiagrams and counterterms until all divergences are removed.

However.7. We choose the minimal subtraction scheme to illustrate our general consideration of Section 4. the BPHZ construction (Bogoliubov– Parasiuk–Hepp and Zimmermann).1. Renormalization constants Z i will be determined by the renormalization conditions in each order of the perturbation theory Zi = 1 + n (Z i − 1)(n) .164 4 Introduction to the theory of renormalization Fig. All counterterms are treated as interaction terms since we want to use counterterms order-by-order in the perturbation theory.28).3 λ 4 : low order renormalization Feynman rules including counterterms In this section we introduce in some detail the dimensional regularization method (’t Hooft & Veltman 1972. 4. The problem of overlapping divergences requires more detailed treatment and we shall not discuss it here. We must also note another approach. the ultimate answer is positive: the counterterms in the renormalized lagrangian do indeed cancel all UV divergences. The Feynman rules for the λ 4 theory are collected in Table 4. the D = 0 divergent subdiagram can be identiﬁed in two distinct ways. namely the renormalization constants Z i do not depend on mass parameters. 4. Bollini & Giambiagi 1972) and the minimal subtraction renormalization scheme using the λ 4 theory as an example. It also shows diagrams corresponding to the counterterms present in lagrangian (4.1 since this method of renormalization is particularly convenient in applications of the renormalization group equation: this is the so-called mass-independent renormalization. In this case the subtractions are performed systematically by means of an ‘R-operation’ which is based on a somewhat different identiﬁcation of subdiagrams.

e. In four space-time dimensions this .1.8. 4. i.4. 4.8(b). with Fig. propagator loop integration p2 i − m 2 + iε d4 k (2π )4 vertex −iλ.3 λ 4: low order renormalization 165 Table 4. p + q + r + s = 0 symmetry factors S S= S= S= 1 2 1 6 1 2 S= 1 4 vertex counterterm −iλ(Z 1 − 1)(n) mass counterterm wave-function renormalization counterterm −i(Z 0 − 1)(n) m 2 +i(Z 3 − 1)(n) p 2 where (Z i − 1)(n) = f i(n) λn It is clear from the previous section that the main task for the lowest order renormalization is to calculate the proper (1PI) Green’s functions corresponding to the three divergent diagrams of Fig. We begin with vertex renormalization. The calculation will be based on the dimensional regularization method and we introduce it now brieﬂy.

166 4 Introduction to the theory of renormalization Fig. if we do the integration in 4 − ε dimensions the result is ﬁnite.9 gives the following expression for the amplitude: A1 = 1 2 i dn k i (−iλµε )2 n k 2 − m 2 + iε (k − P)2 − m 2 + iε (2π) (4. is logarithmically divergent ∼ d4 k/k 4 . so it is convenient to write it as µε λ. The ﬁrst diagram of Fig.9.8(b) in 4 − ε dimensions and then discuss some subtle points of the method.1 are changed accordingly. cancelling all terms divergent when ε → 0 by properly adjusting the renormalization constants Z i (renormalization) and continuing the result analytically to four dimensions.44) . 4. 4. Fig.8. 4. However. 4. where µ is an arbitrary mass scale and λ is dimensionless. This suggests the possibility of considering the analytic continuation of the amplitude to 4 − ε dimensions (regularization). 4.8(b) We ﬁrst calculate the diagram shown in Fig.1) is deﬁned by the action S= dn x 1 (∂ φ)2 2 µ − 1 m2φ2 − 2 λµε 4 φ + counterterms 4! where n = 4 − ε. The kinematics are deﬁned in Fig. In n = 4 − ε dimensions the coupling constant λ is dimensionful.9. Calculation of Fig. Thus. The Feynman rules in Table 4. the dimensionally regularized theory (4. 4.

10. Changing variables ik0 → k0 dk0 → −i dk0 2 2 k0 − k2 → −k0 − k2 −i∞ +i∞ dk0 → +i ∞ −∞ dk0 one gets the following expression A1 = 1 λ2 µ2ε 2 1 dx 0 dn k (+i) (2π )n [k 2 − sx(1 − x) + m 2 − iε]2 (4. We can change the contour to integrate over the imaginary axis.3 λ 4: low order renormalization 167 Fig.47) The integration over dk0 dn−1 k can be easily done after the Wick rotation. since k0 = ±[k2 − sx(1 − x) + m 2 − iε]1/2 . We then get A1 = 1 λ2 µ2ε 2 1 (4.4.45) and change the variables as follows k − Px → k which is legitimate for a convergent integral.46) dx 0 dn k 1 n [k 2 + sx(1 − x) − m 2 + iε]2 (2π ) (4. We introduce the standard Feynman parameters 1 = ab 1 0 dx [ax + b(1 − x)]2 (4. We note that the poles of the integrand in the k0 -plane are located as shown in Fig.48) .10. 4. 4.

577 is Euler’s constant. after integration. Note also the presence in (4.50) and (4. The next step is to use the formula (in Euclidean space) dn k 1 1 bn/2−α (α − 1 n) 2 = (2π)n (k 2 + b)α (4π )n/2 (α) which leads us to an almost ﬁnal result A1 = 1 (λµε )i 2 ( 1 ε) λ 2 (4π )n/2 (2) 1 (4.50) To proceed we require the expression (x) −→ 1/x − γ + O(x) x→0 where γ = limn→∞ (1 + 1 + · · · + 1/n − ln n) = 0.51) of the factor λµε (not expanded in ε) since this is the coupling constant in n dimensions and it enters into the deﬁnition of the counterterm −iλµε (Z 1 − 1).51) The presence of the pole term 1/ε reﬂects the divergence of the integral in four dimensions. The remaining contribution is ﬁnite and reads. First. and we 2 also use the expansion a ε = exp(ε ln a) = 1 + ε ln a + O(ε 2 ) In the limit ε → 0 we get m 2 − sx(1 − x) + γ − ln 4π + O(ε) µ2 0 (4.168 4 Introduction to the theory of renormalization where k is a vector in n-dimensional Euclidean space.53) Comments on analytic continuation to n = 4 dimensions Before proceeding further a comment on the explored analytic continuation to n = 4 dimensions should be made. A1 = iλµε 1 λ 1 λ − iλµε 16π 2 ε 2(4π )2 dx ln A1 (s) = −i m2 (λµε )λ γ − ln 4π − 2 + ln 2 2(4π)2 µ 1 4 +2 − m /s 2 1/2 ln 1 2 1 2 + − 1 4 1 4 − m 2 /s − m 2 /s 1/2 1/2 (4. consider the space with one time-like .52) The inclusion of the crossed diagrams gives the ﬁnal result A1 = i(λµε )λ 3 1 + A1 (s) + A1 (t) + A1 (u) 16π 2 ε ε (4.49) dx 0 m 2 − sx(1 − x) − iε µ2 −ε/2 (4.

In addition we would like to be able to continue In to values n > 4. 2. .3 λ 4: low order renormalization 169 dimension and n − 1 = 3 space-like dimensions. This can be achieved using the identity I = d 1 d k0 + r 2 dk0 dr and by integrating each term by parts over k0 and r .4. . The above procedure may be repeated to show that In is of the form In = 1 (4 − n) I˜n (4. For 1 < n < 4 the surface terms are zero and one gets In ∼ ∞ −∞ ∞ dk0 0 ∞ −∞ dr r n−2 ∞ 2 (−2k0 + 2r 2 ) (k 2 + b)3 1 + 2 so that dk0 0 dr r n−2 (n − 2) = In + 1 (n − 2)In 2 (k 2 + b)2 (4.54) where d n−1 is an element of the solid angle in (n − 1) dimensions = 0 2π π n−1 dθ1 0 sin θ2 dθ2 sin2 θ3 dθ3 . Furthermore. n integer). 0 π sinn−3 θn−2 dθn−2 = 2π (n−1)/2 1 (n − 1) 2 Integral (4. . So one has constructed an analytic function In with simple poles at n = 4. . . 2 k 2 = k0 − r 2 (4. 6.55) In = − 1 I n−4 n where In exists for 1 < n < 5. 8. by partially integrating N times over dr 2 one can ﬁnd explicit expressions for the analytic continuation of In to (1 − 2N ) < n < 4 dimensions. . . respectively.54) can be used to deﬁne an analytic function In in the range 1 < n < 4 which coincides with the original In for n = 2. The basic integral In = dn k 1 (k 2 + b)2 makes sense for n = 1. 3 and can be written as In = 1 2 ∞ −∞ ∞ dk0 0 dr 2 (r 2 )(n−3)/2 1 (k 2 + b)2 π 0 dn−1 . 3 (the solid angle for a non-integer n is the analytic continuation of n−1 . Dimensional regularization assumes the physical theory to be given by In for n → 4.56) 2 where I˜n is well behaved for arbitrarily large n.

The amplitude of Fig.57) ε 16π 2 and the renormalized (ﬁnite) amplitude of Fig.53) can be cancelled by the vertex counterterm in Fig. 4. Lowest order renormalization Coming back to the result (4.58) The divergence is cancelled by the mass counterterm with ≡ (−i)(Z 0 − 1)(1) m 2 1 λ (4. 4. by properly adjusting Z 1 . With such a prescription one then gets to order λ 1 3λ (4. 4. 4. The amplitude of Fig.8 require mass and wave-function renormalization. The so-called MS scheme in which (Z 1 − 1)(1) = (Z 1 − 1)(1) = 1 3λ 2 − γ + ln 4π 2 16π 2 ε is also often used.53).8(a) reads 1 (−iλ)µε 2 dn k i (2π)n k 2 − m 2 + iε 1 (4π )ε/2 (m 2 )1−ε/2 1 ε − 1 = 1 (−iλ)µε 2 2 2 (4π) 2 2 2λm 1 1 λm m2 − i = 1i −1 + γ − ln 4π + ln 2 2 (4π)2 ε 2 (4π )2 µ (4.53) one can now renormalize the amplitude by including counterterms. .11. 4.59) (Z 0 − 1)(1) = 2 16π ε in the minimal subtraction renormalization scheme.e. It is clear from the structure of the counterterm that the divergences of (4. i.170 4 Introduction to the theory of renormalization Fig. The other two diagrams of Fig.11. The minimal subtraction (MS) prescription assumes that Z i have only (multiple) pole terms in ε and the residua are ﬁxed to cancel all the divergences for ε → 0.8(c) requires both mass and wave-function renormalization.12.8(b) is given by A1 (s) + A1 (t) + A1 (u) of (4. It is of order λ2 and therefore in order-by-order renormalization one must include simultaneously the other diagrams of order λ2 given in Fig. 4. 4.

(iii) both k and l large.3 λ 4: low order renormalization 171 Fig.61) is lengthy (Collins 1974).12.12(c): (−i)λ2 m2 2 1 2γ m2 1 2 − ln + − ε 4πµ2 ε ε (16π 2 )2 2 ε2 + O(ε0 ) (4. 4. 4.12(b): iλ2 Diagram 4.8(c). We tabulate ﬁrst the results for Fig.57) and (4.8(c) and those of Fig.60b) m2 1 1 γ m2 3 2 − ln + − 2 )2 2 ε 2 2 ε 4πµ ε ε (16π + O(ε0 ) (4. 4. 4. This is what is called an overlapping divergence because several overlapping loop subintegrations lead to a divergent result.59) it is easy to get the following (we write down explicitly only pole terms): Diagram 4.12(a): iλ2 Diagram 4.60a) The only non-trivial contribution is due to Fig. The structure of pole terms turns out to be the following I = 1 λ2 µ2ε i 6 I =i λ2 1 m2 1 2 m2 + − 2 + 2 m 2 ln p + (γ − 3 )m 2 2 (16π 2 )2 ε ε 4π µ2 12 .12(a)–(c). l large.60c) 1 γ m2 1 2 m2 − ln − 2 )2 2 ε 2 ε 4πµ2 ε (16π + O(ε0 ) (4. 4. k large. Using (4. The divergences which remain after adding the amplitude of Fig. The amplitude reads 1 1 dn k dn l 1 n (2π)n k 2 − m 2 + iε l 2 − m 2 + iε ( p + l − k)2 − m 2 + iε (2π) (4. The calculation of amplitude (4.4.61) which is divergent for (i) l ﬁxed.12(a)–(c) should be then cancelled by new λ2 terms of the mass and wave-function renormalization counterterms ≡ −i(Z 0 − 1)(2) m 2 ≡ +i(Z 3 − 1)(2) p 2 (notice that in λ 4 theory (Z 3 − 1)(1) = 0). (ii) k ﬁxed.

Z m = Z 0 /Z 3 B R λB = Z λ λR .13. 4. 4. We end this section with two comments. Let us illustrate that fact with one example.172 4 Introduction to the theory of renormalization + (1) + (1) + (1) (1) Fig. However.13. they cancel in the ﬁnal result for constants Z i and we see explicitly that at least in that order of perturbation theory the minimal subtraction scheme is indeed a mass-independent scheme. Z λ = Z 1 /Z 3 (4. it should be remembered that one-particle reducible diagrams do not require new counterterms. i. We see in particular that to order λ2 Zm = 1 + 1 2λ2 λ λ2 1 5 + 2 − ε 16π 2 12 (16π 2 )2 ε (16π 2 )2 (4. Firstly.62) (2) Combining (4.64) In the literature one can also often ﬁnd the constants Z m and Z λ deﬁned by the relations m 2 = Z m m 2 .62) one gets the following results for Z 0 and Z 3 (to order λ2 ) Z0 = 1 + λ2 λ 1 11 2 − + 16π 2 ε (16π 2 )2 ε 2 2 ε 1 λ2 Z3 = 1 − 2 )2 ε 12(16π (4. We observe that . Consider the diagrams of Fig.66) 2 i.65) Notice that there are logarithms of mass in the residua of the poles in each diagram.59) with (4. Adding (4.e.e.63) (4.60) one gets i so that (Z 0 − 1) (Z 3 − 1) (2) 1 1 2 11 2 1 λ2 p − m 2m 2 + 2 )2 ε 2 (16π ε 12 ε2 λ2 11 2 = − 2 )2 ε 2 (16π 2ε λ2 1 =− 12(16π 2 )2 ε (4.

i. Consider. mass loop diagrams ≡ (1/ε) f 1 + f 2 + ε f 3 (1) ≡ −(1/ε) f 1 2 so that in the second order + (1) ≡ f 22 However. The number of free parameters is related to the number of so-called counterterms in the lagrangian. which are taken from experiment. for example. in general. typical for local ﬁeld theories.70). 4.4 Effective ﬁeld theories 173 the following relation holds. It is very important to realize that this number may be ﬁnite . see (3.13 = + (1) 2 which illustrates our statement. which are required for the procedure of renormalization. 4. In perturbative calculations we face the problem of divergent results. Thus. we are still able to predict results of measurements in terms of other measurable quantities. to absorb all inﬁnities.4. this problem is also present in classical ﬁeld theories. Diagrams of Fig. for the two-loop diagram alone we have ≡ (1/ε 2 ) f 12 + f 22 + ε2 f 32 + 2(1/ε) f 1 f 2 + 2 f 1 f 3 + 2 f 2 f 3 ε Discarding poles in the last result one obtains f 22 + 2 f 1 f 3 = f 22 which is wrong because we insist on unitarity being satisﬁed order-by-order in perturbation theory and therefore the ﬁnite part of the two-loop diagram should read f 22 as can be seen by considering the imaginary part of the amplitude.e. obtained by simply discarding poles and multiple poles at n = 4 but have to be worked out by order-by-order renormalization procedure. In fact.4 Effective ﬁeld theories The basic tool for calculations in quantum ﬁeld theory is perturbation theory. The second remark is devoted to the fact that the ﬁnite parts of Feynman integrals are not. The predictive power of a theory is retained due to the procedure of renormalization: divergent contributions are absorbed into free parameters of the theory.

that it provides a tool for easy resummation of large logarithmic corrections O(α n lnn (E/MW )) (Chapter 12) to weak processes. We have described here a situation in which the complete renormalizable theory at the scale M is known and the effective theory can be derived from it by perturbative methods. At the energy scale m. Perturbative QCD is not applicable in that region but we may construct. i.174 4 Introduction to the theory of renormalization or inﬁnite! Traditionally. For a theory to be renormalizable in this traditional sense. after integrating out heavy degrees of freedom (see Chapters 7 and 12). A different situation occurs when we try to describe strong interactions of hadrons at low momentum transfer. Moreover. We encounter the latter case. These are described by the weak effective lagrangian with dimensionful coupling. Since E/MW 1. As we shall see. or less. If we neglect all physical effects of those corrections. for example. it cannot contain couplings with the dimension of negative powers of mass. we also know now that they should be regarded as only effective theories. weak processes (Chapter 12). However. if we want to calculate QED and QCD corrections to low energy. valid in a limited energy range. quantum electrodynamics (QED) considered as the effective low energy part of the electroweak theory (Chapter 12). Effective ﬁeld theories appear naturally in the physical situation characterized by the presence of two hierarchical mass scales M m. too. This certainly makes such a theory very predictive: all results are obtained in terms of a ﬁnite (usually small) number of measurable parameters. the virtue of the effective theory is not only that it offers a quickly truncated expansion in powers of E/MW but. for instance. an effective pion nucleon theory which respects the chiral symmetry of QCD but is not derivable perturbatively from it. it is sufﬁcient to use it only in the Born approximation. It is the content of the Appelquist–Carrazone decoupling theorem (Appelquist & Carrazone 1976) that all effects depending on positive powers of M or logarithmic in M can be absorbed into the redeﬁnition of the bare parameters of the effective renormalizable theory. E ∼ 1 GeV. be interested in the O(m/M) corrections. however. one obtains an effective theory whose lagrangian may consist of operators of dimension four or less (the ‘renormalizable’ part) and higher dimension operators suppressed by inverse powers of large mass M. for instance. We know that such theories are often very good approximations to the real world in some energy range and therefore very interesting to study. then physics at m is described by the ‘renormalizable’ part. the corresponding lagrangian may contain only operators of dimension four. ﬁrst of all. In this case. Suppose that physics at scale M m is described by a renormalizable theory. we may. a theory is called renormalizable only if a ﬁnite number of counterterms is sufﬁcient to renormalize it. A wellknown example is. the physical cut-off for that theory is of similar . with no need for higher order perturbation in it.e.

The latter still has a lot of predictive power (based on such principles as analyticity. However. Planck scale) but we do not know the correct theory at that scale. It is then a very natural hypothesis that physics at lower scales is described by an effective theory which contains all higher dimension operators consistent with symmetries of the low energy theory and suppressed by powers of the large scale. It is of fundamental importance that Nature can be described by ﬁeld theories that obey the decoupling theorem. without the (unrealistic) quest for the theory of everything. (a) Construct the most general renormalizable lagrangian invariant under the symmetry φ → −φ. The above comments on effective ﬁeld theories anticipate a lot of knowledge which the reader is going to acquire in the following chapters. we may consider the physical situation when we know about (or expect) the existence of a large physical scale (for example. although a renormalizable theory can formally be used for calculations at any energy scale (since the dependence on larger scales is totally absent). this does not mean that it is correct at any scale.1 Discuss a model with two scalar ﬁelds φ and respectively. unitarity. Hint: expand φ = φ − v where . although the number of free parameters increases with the order of perturbation theory. φ → − . such that m M. with the masses m and M. A good example is again QED. At the scale M Z it must be embedded into the electroweak theory. This makes it possible to understand fundamental laws of physics step by step. (c) Assuming that the discrete symmetry is spontaneously broken calculate the vacuum expectation value v at one-loop level. . On the other hand. it must be stressed that there is no fundamental difference between a theory which is renormalizable with a ﬁnite or inﬁnite number of counterterms (but only those consistent with the symmetry of the bare action). as such effects are always strongly suppressed. It should be emphasized that. we cannot learn about that by studying electromagnetic processes at low energy. Finally. (b) Study spontaneous breaking of this discrete symmetry at the classical level. they may be a useful guide to further reading. However. Therefore. unless we have reached the precision of sensitivity to O(E/M Z ) corrections.Problems 175 order of magnitude as the maximal energy of its applicability so that the E/ expansion cannot be so quickly truncated and higher order iteration of the effective lagrangian is necessary. Problems 4. symmetries). calculate the one-loop effective ˜ potential by integrating the tadpole amplitudes. it makes it difﬁcult to learn about new physics at higher scales. E M Z .

where s f = 0 (1/2) for spin 0 (1/2) particles.4 Calculate the superﬁcial degree of divergence in n dimensions of a Feynman diagram with I f (E f ) internal (external) lines of type f and Ni vertices of type i. Instead of performing Wick’s rotation in the Feynman integrals it is more convenient to begin with the theory formulated in the Euclidean space. (b) Calculate one-loop self-energy diagrams for the ﬁeld ψ and compare the corrections to its mass with the analogous corrections to the mass of the ﬁeld in Problem 4.67) + 4 (4. respectively. Show that D = f I f (2s f − 2 + n) + i Ni (di − n) + n. the vacuum expectation ( value v is ﬁxed by the equation ( + = 0). V1 /dv = i (d) Single out from the effective potential the corrections to the mass m and to the self-coupling of the ﬁeld φ which depend on the heavy ﬁeld and compare them with the result of the direct diagrammatical calculation. pay attention to the structure of quantum corrections from the exchange of the ﬁeld . If all i ≥ 0 then D ≤ Dmax and the theory is renormalizable with a ﬁnite number of counterterms. 4. such that m < M.2 Discuss a model with a scalar ﬁeld φ and a fermion ﬁeld ψ. the presence of the φ 3 coupling generates tree level ) and one-loop ( ) tadpole diagrams. where p 4 is the ‘time’ component of the momentum four-vector in the ˆ Euclidean space. with di derivatives. Using 2I F + E f = i n i f Ni . (a) Construct the most general renormalizable lagrangian. 4. .68) Check that the n-point Green’s functions in the Minkowski space are obtained from those in the Euclidean space by analytic continuation of their momentum variables: ˆ pi4 → −i pi0 . one gets D = n − f E f (s f − 1 + n/2) − i Ni i with i = n −di − f n i f (s f −1−n/2). where n i f is the number of lines of type f in the vertex of type i. (e) Formulate the effective theory at the scale µ M.176 4 Introduction to the theory of renormalization φ = v is to be found from the minimum of the effective potential. For the 4 theory we then have the generating functional W [J ] = where SE = d4 x E 1 1 (∂ )2 + m 2 2 2 2 [d ] exp −S E − d4 x J (x) (x) E λ 24 (4. Only the last term in the equation for D depends on the internal structure of diagrams. 4.1. rewrite the ˜ ˜ lagrangian in terms of the φ.3 It is convenient to perform multi-loop calculations in Euclidean space. calculate the one-loop contribution V1 to the effective potential by integrating the equation and identifying φ ≡ v. with the masses M and m.

are gauge-independent. It is important that there exist regularization procedures which preserve the gauge symmetry of the theory. so that LG = − 1 (∂µ Aµ )2 B 2aB To begin. We will work in the class of covariant gauges. All physical quantities. about perturbative expansion for the bare Green’s function) belongs to this category. the gauge symmetry of the theory leads to certain constraints on the Green’s functions: there exist relations between them known as the Ward–Takahashi identities. to derive the Feynman rules for a gauge theory it is necessary to ﬁx the gauge by adding some gauge-ﬁxing conditions. In terms of the Green’s functions these are exactly the conditions which protect the physical equivalence of different gauges. This fact. Nevertheless. crucial for which is the renormalization programme described in Chapter 4 which is usually formulated in terms of the Green’s functions. The Green’s functions are not.5 Quantum electrodynamics We begin our discussion of gauge theories with the simplest and the best-known one: QED. Thus. for instance. we consider the bare. physical objects and they are gauge-dependent. The QED lagrangian is based on the minimal coupling ansatz and in terms of the bare ﬁelds and parameters it reads µν B / L = − 1 Fµν + FB + ¯ B (i/ − qeB AB ) ∂ 4 B − mB ¯ B B + LG (5. however. which is necessary for a consistent theory. has to be checked in each case.1) where eB > 0 and LG is a gauge-ﬁxing term. So far most of the physical applications of gauge theories rely on perturbation theory. As we know from Chapter 3. the results of 177 . The ’t Hooft–Veltman dimensional regularization deﬁned for perturbative expansion (we can think. and the S-matrix elements in particular. regularized (ﬁnite) Green’s functions.

The gauge-invariant structure of the QED lagrangian is not destroyed when each gauge-invariant piece of it is multiplied by a constant. compare with QCD. there is no need for a gauge-ﬁxing counterterm).1) in terms of the renormalized quantities assuming the most general form consistent with gauge invariance after renormalization and we derive Ward identities for the renormalized Green’s functions (actually.178 5 Quantum electrodynamics our formal manipulations should always be understood as valid order-by-order in perturbation theory.3) on renormalized ﬁelds. that the theory is renormalizable without destroying its gauge invariance. Section 8.1) we can derive the Ward–Takahashi identities for the bare. µ 1/2 B = Z2 1/2 .3). Apart from the gauge-ﬁxing term. for example. Leaving aside these formal considerations (consult.2) is invariant under the gauge transformations (1. Actually.3) 1 Aµ (x) = Aµ (x) + (Z 2 /Z 1 )∂µ (x) e If a free scalar ghost ﬁeld is included the full lagrangian is invariant under BRS symmetry transformation (see Problem 3. regularized Green’s functions of our theory. Slavnov & Faddeev (1980). .91) and (1.1) written in terms of the renormalized quantities reads† / ∂ L = − 1 Z 3 Fµν F µν +Z 2 ¯ [i/ −qe(Z 1 /Z 2 )A] −Z 0 m ¯ 4 −(1/2a)(∂µ Aµ )2 (5. Bogoliubov & Shirkov (1959). † Strictly speaking one should include the scalar ghost ﬁeld and refer to the BRS invariance (see Problem 3.94) (x) = exp[−iq (x)] (x) (5.3) are ﬁnite (make this argument precise by considering BRS invariance and taking into account that the ghost ﬁeld is not renormalized. lagrangian (5. in most popular renormalization schemes Z 1 = Z 2 including ﬁnite terms. Using these identities one can then prove. allowing all parameters and ﬁelds to become renormalized. eB = (Z 1 /Z 2 Z 3 )e aB = Z 3 a 1/2 m B = (Z 0 /Z 2 )m = (m − δm)/Z 2 . This result will be derived formally using the Ward– Takahashi identities. In addition the parameter e can be arbitrarily changed. Our discussion in Chapter 4 of the renormalization programme applies to the present case.2) (as we shall see shortly. From the lagrangian (5. Collins (1984)) we simply rewrite the lagrangian (5. Thus the most general gauge-invariant form for the lagrangian (5. the ghost ﬁeld. being free.1). using these Ward identities one can also prove a posteriori that the renormalized Green’s functions are ﬁnite). The renormalization constants Z i must be ﬁxed by imposing certain renormalization conditions. The relations between bare and renormalized quantities are as follows AB = Z 3 Aµ . Therefore Z 1 = Z 2 up to ﬁnite terms. In addition it is clear that Z 1 /Z 2 must be ﬁnite as all other quantities in (5. does not undergo renormalization.

the so-called on-shell renormalization scheme is the one most commonly used in QED calculations. The integration measure is invariant under inﬁnitesimal gauge ¯ transformation and the only gauge-dependent terms in the integrand are source terms and the gauge-ﬁxing term. respectively. the physical mass of the fermion and the ‘physical’ ﬁelds in the sense of Chapter 4. for instance.1 Ward–Takahashi identities General derivation by the functional technique We derive the Ward–Takahashi identities in QED using the functional technique. Therefore we conclude that δW/δ (y)|θ =0 = 0 (5. in each order of perturbation theory there exist deﬁnite relations between sets of parameters corresponding to different renormalization schemes.2) as the electric charge measured. the mass parameter m and the renormalized ﬁelds which appear in the lagrangian (5. Under the functional integral one can perform an inﬁnitesimal gauge transformation = [1 − iq ] (5.1 Ward–Takahashi identities 179 Mainly for historical reasons. when all the four-momenta approach their on-shell values. ) is given by (5. this can be achieved by demanding that in the on-shell limit. The generating functional for the full Green’s functions is the following W [J.6) 1 A µ = A µ + ∂µ e ¯ where e = eZ 1 /Z 2 . However. an inﬁnite number of other renormalization prescriptions is possible. the fermion and the photon propagator as well as the fermion–fermion–photon vertex function approach their zeroth order form. On the other hand.4) cannot change the value of the integral. Generally speaking. in the Thomson scattering (see (12. In addition.7) . In this scheme we choose to interpret the coupling constant e.5.4) and the lagrangian L(Aµ . a change of the integration variables in (5. ) + Jµ Aµ + α ¯ + ¯ α] (5. α] = ¯ where Seff = d4 x [L(Aµ .2). α.5) D(Aµ ¯ ) exp(iSeff ) (5. 5.86)). The renormalization conditions are then imposed so as to ensure this interpretation of the renormalized quantities in any order of perturbation theory. The renormalized quantities lose their previous interpretation but the theory can be formulated in terms of them in a totally equivalent way. in any order of perturbation expansion.

¯ ] the identity (5. α α] − ¯ δZ δZ δZ ¯ =− .10) is the general form of the Ward–Takahashi identities in QED. (x).12) δ δ δ Jµ = − µ . α.8) 1 (x) − (1/a)∂µ Aµ (x)∂ 2 (x) e ¯ Integrating expression (5. = Aµ .6.3) we get δW 1 2 1 δW δW ∂ ∂µ + ∂µ J µ (y)W + q eα(y) + qe ¯ α(y) = 0 ¯¯ a i δ Jµ (y) δ α(y) ¯ δα(y) → ← (5.6) so that [Aµ (x).11) = Z [J. in a class of covariant gauges. .9) (remember that ( δ /δα)W = −W ( δ /δα) or in terms of the generating functional Z [J. α = .13) where (left derivatives only) .180 5 Quantum electrodynamics where δW [J. α. ¯ (x)] d4 x [Jµ (x)Aµ (x) + α(x) (x) + ¯ (x)α(x)] ¯ (5.10) can be rewritten as follows: 1 2 (y) µ δ δ δ (y) ∂(y) ∂µ A (y) − ∂µ + iq e ¯ (y) + iq e ¯ (y) ¯ =0 a δ Aµ (y) δ (y) δ ¯ (y) (5. It is also very useful to have the analogous relation for the 1PI Green’s functions.10) Eq. = δ Jµ δα ¯ δα (5. α= ¯ δA δ δ¯ In terms of the functional [Aµ . (5. neglecting the surface terms at inﬁnity and remembering that functional integrals of derivatives of ﬁelds are derivatives of the Green’s functions (see Section 2.8) by parts. ] = i ¯ × d4 x D(Aµ ¯ ) exp(iSeff ) 1 Jµ (x)∂ µ (x) − iq (x)α(x) (x) + iq (x) ¯ (x)α(x) ¯ e ¯ (5. α. for the connected Green’s functions. α] for the connected Green’s functions (W = exp(iZ )) ¯ δZ 1 2 δZ δZ ∂ ∂µ + ∂µ J µ (y) + iq eα(y) + iq e ¯ α(y) = 0 ¯¯ a δ Jµ (y) δ α(y) ¯ δα(y) (5. In the present case its arguments are the ‘classical’ boson ﬁeld Aµ (x) and the ‘classical’ fermion ﬁelds (x) and ¯ (x) (it is convenient to use the same notation for the arguments of as for the ﬁelds in the lagrangian. The generating functional for the latter has been deﬁned in Section 2. see Section 2.

13) with several examples. We observe that the longitudinal (gauge-dependent) part is not altered by interactions (compare with (3.10) with respect to δ δ δ δ =− δ α(x) δα(z) ¯ δα(z) δ α(z) ¯ and setting J = α = α = 0.18) .17) (5.16) gives zero contribution to the relation (5.16). Our next example is the well-known relation between the vertex and the fermion propagators. Differentiating relation (5. one gets ¯ δ 1 2 (y) δ2 ∂(y) ∂µ Z a δ α(x)δα(z) δ Jµ (Y ) ¯ δ2 δ2 Z = iq e ¯ δ(y − x) + iq e ¯ δ(y − x) δα(z)δ α(y) ¯ δ α(x)δα(y) ¯ Since 0|T (x) ¯ (y)|0 = i δ2 Z δ α(x)δα(y) ¯ J =α=α=0 ¯ (5. This means.14) Using (2.15) reads ˜ ˜ µν G µν (k) = −ia(kµ kν /k 4 ) + G T (k) where the transverse projection ˜ µν G T (k) = (gµν − kµ kν /k 2 ) f (k 2 ) (5.134) and Fourier transforming into momentum space we obtain the following relation: ˜ i(1/a)k 2 k µ G µν (k) = kν (5.15) constrains the full renormalized propagator to the form (5. Examples As the ﬁrst example we consider the Ward identity for the longitudinal part of the connected photon propagator.1 Ward–Takahashi identities 181 In the following we shall illustrate the general relations (5. The solution to (5. The Ward identity (5.10) with respect to Jν (z) and putting J = α = α = 0 one gets ¯ 1 2 µ δ2 Z − ∂(y) ∂(y) µ a δ J (y)δ Jν (z) J =0 ν = ∂(y) δ(y − z) (5.15) ˜ Here G µν (k) is the Fourier transform of the connected part of the photon propagator G µν (x1 − x2 ).32)). among other things.10) and (5.15). Differentiating the general identity (5. that the renormalization programme does not require any counterterms for the gauge-ﬁxing term.5.

q) (5.19) in conﬁguration space.19) can be represented by Fig.2.182 5 Quantum electrodynamics Fig.19) This is the Ward identity in conﬁguration space.1 and relation (5. 5. .1. where the cross denotes −(1/a)∂ 2 ∂ µ or −(1/a)q 2 q µ .21) in momentum space by Fig.21) Graphically.21) in momentum space can be immediately deduced from the graphical representation of relation (5. respectively.18) can be rewritten as follows µ 2 −(1/a)∂(y) ∂(y) 0|T Aµ (y) (x) ¯ (z)|0 = q e 0|T (y) ¯ (z)|0 δ(y − z) − q e 0|T (x) ¯ (y)|0 δ(y − z) ¯ ¯ (5.20) d4 k exp[−ik(y − z)]iS(k) 0|T (y − z) ¯ (0)|0 = (2π )4 4 dk ¯ (0)|0 = exp[−ik(x − y)]iS(k) 0|T (x − y) 4 (2π ) then inserting (5. Notice that the form (5.20) into (5.19) and integrating over dx dy dz exp(i p x) × exp(iq y) exp(ik z) to get rid of the δ-functions. 5. relation (5. and 0|T Aµ (y) (x) ¯ (z)|0 = δ3 Z δ Jµ (y)δ α(x)δα(z) ¯ J =α=α=0 ¯ relation (5. 5. q) = q eS( p + q) − q eS( p) ¯ ¯ (5. Using translational invariance and Fourier transforming into momentum space 0|T Aµ (y) (x) ¯ (z)|0 = 0|T Aµ (y − z) (x − z) ¯ (0)|0 4 4 d p dq exp[−i p(x − z)] = 4 (2π )4 (2π ) × exp[−iq(y − z)]Vµ ( p. we obtain the Ward identity in momentum space −(1/a)q 2 q µ Vµ ( p.

This is convenient (but not necessary) because the parameter e is then universal for all fermions in the theory.22) By analogy with (5. y)δ(y − z) + iq e ¯ (2) (y.21) and (5. z) − iq e ¯ (2) (x. k As an interesting consequence of the Ward identities (5. including e. i. q) = −iq e ˜ (2) ( p + q) + iq e ˜ (2) ( p) ¯ ¯ (5. Thus the ﬁniteness of the ratio Z 1 /Z 2 implies that order-by-order in perturbation theory Z 1 = Z 2 up to the ﬁnite terms. when ε → 0 in dimensional regularization) and possibly also ﬁnite pieces. We recall (2. This point will be discussed in more detail in Section 6.e. z)δ(y − x) = 0 (5. S(k) = 1/(/ − m). 5. It is also worth remembering that only in the on-shell renormalization scheme is e the electric charge. the Z i s are (n) n ∞ power series in the coupling constant α = e2 /4π: (Z i − 1) = n=1 f i α . In perturbative calculations. using a regularization procedure which preserves gauge invariance. In practice.2.21) can also be written in terms of the 1PI Green’s functions. for example. for the free propagator. (n) The coefﬁcients f i contain pieces which are divergent in the limit where regularization is absent (for example.23) we notice that the gauge parameter e = eZ 1 /Z 2 and consequently the ratio Z 1 /Z 2 must ¯ be ﬁnite if the theory is renormalizable.5. in momentum space relation (5. Indeed all other quantities. e2 /4π = 1/137.22) reads qµ ˜ (3) ( p. x. in these equations are renormalized quantities. in all the commonly used renormalization schemes Z 1 = Z 2 including the ﬁnite terms.19). when Z 1 = Z 2 the counterterms form a gauge-invariant set and the Ward identities for .21) or (5.1 Ward–Takahashi identities 183 Fig.4.23) where the ˜ (n) s are Fourier transforms of the (n) s. We take the derivative of (5. In addition.135) generalized to include fermions to get the following: µ ∂(x) (3) µ (y. The Ward identity (5.152): ˜ (2) = ˜ i[G (2) ( p)]−1 = +S −1 (k) where.13) with respect to δ δ δ ¯ (x) δ (z) set Aµ = ¯ = = 0 and use the deﬁnitions (2.

We shall calculate now all the elements which are necessary to discuss the lowest order radiative corrections to QED processes such as.1). for example. This is not the case for the standard dimensional regularization convention used here. be derived in a similar way to the renormalized Ward identities. In n dimensions we can assume Tr[γµ γν ] = 1 2 Tr[γµ γν + γν γµ ] = gµν Tr 1 = 4gµν l (5. in supersymmetric theories a dimensional regularization method called dimensional reduction (Siegel 1979) is used which preserves supersymmetry. 5.2) in the on-shell and minimal subtraction renormalization schemes. and deﬁne g µ µ = g µν gµν = n This convention is the one most frequently used in calculations but other conventions are also possible (Collins 1984). We notice that. In particular.2 Lowest order QED radiative corrections by the dimensional regularization technique General introduction We recall the QED lagrangian (q = −1. for example.184 5 Quantum electrodynamics the renormalized Green’s functions are identical to those for the bare regularized Green’s functions. Dimensional regularization in QED in the presence of γ -matrices requires an extension of previous rules. . The only difference is that one must start with the lagrangian expressed in terms of the bare quantities. in the relation γµ γν + γν γµ = 2gµν 1 l the only space-time-dependent object is the tensor gµν . we still consider γµ s as 4 × 4 matrices in the fermion and antifermion spin space. The Feynman rules are given in Appendix B. Several other relations and integrals useful in the subsequent calculations are collected in Appendix B.24) (there is no counterterm to the gauge-ﬁxing term: see Section 5. of course. e > 0) in a covariant gauge / L = − 1 Fµν F µν + ¯ (i/ + e A) − m ¯ − (1/2a)(∂µ Aµ )2 − (Z 3 − 1) 1 Fµν F µν ∂ 4 4 ¯ i/ − (Z 0 − 1) ¯ m + (Z 1 − 1)e ¯ A / + (Z 2 − 1) ∂ (5.25) i. the electron scattering (or e+ e− pair production) in a static Coulomb potential. The latter can.e. An important exercise is to calculate the Thomson limit of the Compton scattering (Problem 5.

From consideration of Section 5. 5.29) + counterterm .2 Lowest order QED radiative corrections 185 Fig.1 we know that only the transverse part of the photon propagator has higher order corrections. This is a oneloop correction to the free photon propagator.26) In the one-loop approximation the transverse part of the photon propagator reads −iPµν (−iPµρ ) ˜ µν G T (q) = + i q2 q2 ˜ µν G T (q) = The scalar function (q 2 ) is given by g µν µν (q) ρσ (q) (−iPσ ν ) −iPµν = [1 + q2 q2 (q 2 )] Summing over all one-particle reducible diagrams one gets q 2 [1 −iPµν − (q 2 )] = (n − 1)q 2 (q 2 ) (5. after taking the trace Tr[γµ γα γν γβ ] = 4(gµα gνβ + gµβ gνα − gµν gαβ ) and using the Feynman parametrization 1/ab = gets (n − 1)q 2 (q 2 ) = ie2 dn p (2π)n 1 1 0 dx (1/[xa + (1 − x)b]2 ) one dx 0 4(2 − n)( p 2 − p · q) + 4nm 2 {x( p 2 − m 2 ) + (1 − x)[( p − q)2 − m 2 ]}2 (5.28) × γν ( p − / + m)] + i(Z 3 − 1)(1) (qµ qν − q 2 gµν ) / q Therefore. On general grounds one can write it as follows: i µν (q) = iP µν = i g µν (q 2 ) q µq ν − 2 q 2 (q 2 ) q T (5.5. Vacuum polarization We calculate the amplitude corresponding to the diagrams of Fig.3.3.27) where from the Feynman rules we obtain i µν (q) = −(ie)2 dn p i i Tr[γµ ( p + m) / n p 2 − m 2 ( p − q)2 − m 2 (2π) (5. 5.

from (5. can easily be evaluated in the small q 2 and large q 2 limits and reads (up to constant terms)† m2 q4 α α q2 1 2 q →0 ln +O − 3π µ2 π m 2 15 m4 (5. . In this case one demands: ON (q 2 ) = 0 for q 2 = 0 Therefore. where µ is an arbitrary mass parameter).30) (Z 3 − 1)(1) = − and ON α 21 1 m2 + 3 (ln 4π − γ ) − 1 ln 2 3 π 3ε µ m 2 − x(1 − x)q 2 − iε m2 (5. given by (5. The pole in ε reﬂects the UV divergence in four dimensions. (q 2 = 0) = 0 for µ2 = m 2 . and α is dimensionless in 4 − ε dimensions: α → αµε .20) we get ﬁnally 1 α m 2 − x(1 − x)q 2 − iε −ε/2 (q 2 ) = − (4π)ε/2 1 ε dx 2x(1 − x) 2 π µ2 0 1 α 12 1 m 2 − x(1 − x)q 2 − iε + (ln 4π − γ ) − =− dx 2x(1 − x) ln π ε3 3 µ2 0 + O(ε) − (Z 3 − 1)(1) (5.31) and the ﬁnite part of the vacuum polarization.33) (q 2 ) = α 2 π 1 0 dx x(1 − x) ln (5. In the large q 2 limit.30).30) (we have taken n = 4 − ε. Hence α ON = α(m 2 ) (see also Chapters 6 and 12).32) |q 2 | α ln 2 |q 2 | m2 3π µ The renormalization scheme most frequently used in QED is the on-shell renormalization. It is useful to notice that in the MS. α = e2 /4π . In the minimal subtraction scheme (Z 3 − 1)(1) = − α 21 π 3ε (5. in the MS scheme the sum over all one-loop one particle reducible diagrams gives ˜ µν G T (q) = −iPµν 2 2 1 − α ln |q | q 3π µ2 This is the so-called leading logarithmic approximation for the photon propagator.18)–(B.34) † The so-called MS scheme in which the counterterms include the constant terms: 2/ε → η = 2/ε − γ + ln 4π is also often used. It is cancelled by adding the Z 3 counterterm.186 5 Quantum electrodynamics Using the integrals (B.

Taking into account the symmetry y → −y of the integrand we can write ON (q 2 ) = α1 π2 α1 π2 1 dy 0 ∂ q 2 (1 − y 2 ) (y − 1 y 3 ) ln 1 − 3 ∂y 4m 2 − iε (5.38) which is a so-called once subtracted dispersion relation.34) into y = 1 − 2x.40) .35) ON α (q ) −→ 2 /m 2 →∞ π −q 2 1 3 m2 |q 2 | 5 ln 2 − + O m 9 q2 As a ﬁnal comment we rewrite ON (q 2 ) in the form of a dispersion integral (the dispersion technique will be discussed in some detail later). Firstly. The proper two-point function (2) ( p ) reads / (2) ( p) = p − m − / / ( p .2 Lowest order QED radiative corrections 187 Fig.5.4. Asymptotically one obtains ON q4 α q2 1 (q ) −→ − +O π m 2 15 m4 q 2 →0 2 (5.39) The propagator is the inverse of the proper two-point function so we get ˆ G (2) ( p ) = / i p−m− / (5. 5.37) Another change of variable t = 4m 2 /(1 − y 2 ) gives the result ON α (q 2 ) = − q 2 π ∞ 4m 2 1 1 2m 2 dt 1+ t t − q 2 − iε 3 t 1− 4m 2 t 1/2 (5. Electron self-energy correction We consider the electron self-energy truncated diagram and the necessary counterterms which are shown in Fig. m) / (5.36) and integrate it by parts to get ON (q 2 ) = − 1 0 dy 2y(y − 1 y 3 ) 3 q2 4m 2 − q 2 (1 − y 2 ) − iε (5. we change variables in the integral (5.4. 5.

43) In this section we shall calculate ( p ) in the Feynman gauge (unlike µν (q).188 5 Quantum electrodynamics The general structure of the electron self-energy correction is obviously the following ( p . We recall that 1 (actually 1 1 : see 2 Section 4.1) appears as a correction to an external line of the on-shell scattering amplitude. It can be calculated as ( p 2 = p 2 ) / where m 2 is the zero of F 2 V (m F )]} to the order 1 = ∂ ∂p / = p=m F ∂ m ∂ p2 p 2 =m F 2m F + 2 2 V (m F ) + 2m F ∂ V (p ∂ p2 2 ) p 2 =m 2 F (5. The functions and V ( p 2 ) are given by Tr ( p ) = 4 / m( p 2 ) ) V (p 2 Tr[ p ( p )] = 4 p 2 / / Introducing ) (5.46) dx 0 1 [k 2 + x p · 2k + x( p 2 − m 2 )]2 (5. The calculation proceeds as follows. m) = / V (p 2 )p + / m( p 2 )I (5.41) It is often convenient to write it as an expansion ( p . We can then ﬁnd the one-loop relation between the pole mass m F and the renormalized mass parameter m as well as the correction 1 .42) the inverse propagator = {[m + m (m F )]/[1 − considered. the / electron self-energy is gauge-dependent).47) .45) N = γ µ ( p + / + m)γ ν gµν / k we ﬁnd (after some algebra) Tr N = 4nm Tr[ p N ] = 4[ p 2 (2 − n) + ( p · k)(2 − n)] / Next.44) m( p 2 − im(Z 0 − 1)(1) + i(Z 2 − 1)(1) p / As before we use dimensional regularization: n = 4 − ε. m) + ( p − m F )2 / m2 F / 2 ( p . m) + ( p − m F ) / / 1( p = m F . The Feynman integral reads. −i ( p ) = (ie)2 / dn k µ −igµν i γ 2 γν n (2π ) k + iε p + / − m + iε / k (5. m) (5. m) = / / 0( p = m F . we use Feynman parametrization for the denominators 1 1 = k 2 ( p + k)2 m 2 1 (5.

48) with respect to p 2 one obtains the following result ∂ m = nmC − 1 ε 2 ∂ p2 1 dx 0 x(1 − x) 1+2/ε X µ2 (5.51) which is. their derivatives contributing to 1 are not. differentiating the regularized expression (5. Indeed.50) and similarly for V ( p 2 ).52) .49) α 1 (1) (Z 0 − 1) = − 2π ε The ﬁnal (ﬁnite) expressions for m and V can then be easily written down.5. UV ﬁnite (Z i are constants) and in the limit p 2 → m 2 gives (we can take m 2 = m 2 because we work in the ﬁrst order in α) F ∂ m ∂ p2 =n ε 1 α 1−γ m 4π 2 m2 4π µ2 −ε/2 p 2 =m 2 (−ε) (2) (2 − ε) (5. m( p 2 α )=− m π dx ln xm 2 − x(1 − x) p 2 +γ + 4π µ2 1 2 (5.18)–(B. for example. We leave it to the reader to check that in the MS scheme at the one-loop level α(µ) m F = m(µ) 1 + π We observe that m and V are IR ﬁnite (when p 2 → m 2 ).20) to perform the integration over momentum k to get the following results: 1 (1) dx X − (Z 0 − 1) m m = nmC 0 (5.48) 1 (1) dx (1 − x)X − (Z 2 − 1) V = (2 − n)C 0 where α −1 C≡ 4π 4π −ε/2 1 ε 2 −ε/2 x( p 2 − m 2 ) − x 2 p 2 X= µ2 Using standard tricks to single out the pole terms in ε we immediately ﬁnd the renormalization constants in the minimal subtraction scheme 2α 1 (1) (Z 0 − 1) = π ε (5. of course.2 Lowest order QED radiative corrections 189 and (B. However.

The ﬁnal result reads (εIR > 0) ∂ m ∂ p2 and for ∂ = 1 1 α m2 + 1 γ − 1 + 1 ln 4 2 2 m 4π εIR 4π µ2 +1 (5. but this time with ε < 0 (changing ε from a negative to a positive value is legitimate since we are already dealing with a UV ﬁnite quantity).53) p 2 =m 2 2 V /∂ p ∂ V ∂ p2 In addition p 2 =m 2 =− 1 α 1 m2 − 1 + 1 γ + 1 ln 2 2 2m 2 π εIR 4π µ2 (5.55) so for 1.190 5 Quantum electrodynamics The integration over Feynman parameters has introduced a new (IR) singularity which has again been regularized by keeping n = 4 − ε. Instead of using the dimensional method one can regularize the IR singularities by giving the photon a small mass λ. we get = ∂ ∂p / = p=m 1 α 1 m2 + 3 ln − 1 + 3γ 4 π εIR 4 4π µ2 (5.54) V (p 2 = m2) = m2 α +γ −2 ln 4π 4π µ2 (5. it nevertheless has the so-called mass singularities (when p 2 → m 2 → 0). In order to keep trace of the origin of the singular terms here and in the following we shall often distinguish between εUV (n = 4 − εUV . where eventually λ → 0 (actually this is the most standard method). Electron self-energy: IR singularities regularized by photon mass The IR singularities encountered above are due to the zero mass of the photon. εIR > 0). For the electron self-energy correction we now have −i ( p ) = −e2 / dn k 1 1 γn ( p + / + m)γ µ 2 / k (2π)n k − λ2 ( p + k)2 − m 2 .56) Our ﬁnal remark is that although V ( p 2 ) does not have IR singularities (when p 2 → m 2 ). εUV > 0) and εIR (n = 4 + εIR .

58) p 2 =m 2 m2 1 α − 2 ln 2 + const. On-shell vertex correction We consider Fig. The derivatives of m and V can now be calculated in an elementary way.59) = m2 α 1 m2 α1 ln 2 + ln + const.5.2 Lowest order QED radiative corrections 191 Fig.50) one gets: α m ( p ) = −m π 2 2 V (p ) = x 2 p 2 − x( p 2 − m 2 + λ2 ) + λ2 − iε + const.5. 2 4πµ (5. Following the previous calculation. instead of (5. π 2m 2 λ (5. π2 λ π 4 4π µ2 (5.61) .5 µ ( p. dx ln 4πµ2 dx (1 − x) ln α 2π x 2 p 2 − x( p 2 − m 2 + λ2 ) + λ2 − iε + const. 5.60) Comparing with the dimensional regularization of IR singularities we observe the correspondence 2/εIR → ln(m 2 /λ2 ). 5. p ) = (ie)2 1 dn k i i γµ γβ (−ig αβ ) 2 (Z 1 − 1)(1) γµ γα (2π)n p + / − m p + / − m / k / k k (5.57) These integrations can be done by writing the logarithm as ln p 2 (x − x1 )(x − x2 ). We get in particular ∂ m ∂ p2 ∂ V ∂ p2 and ﬁnally 1 p 2 =m 2 =m = α 1 m2 ln 2 + const. where x1 and x2 are the roots of the quadratic expressions. π 4m λ (5.

we can write the a a integral as (1) 2 µ ( p.62) where N = γµ [4 p · p + (n − 2)k 2 + 4( p + p ) · k] k k − 4( p + p )µ/ + kµ [4m − 2(n − 2)/ ] (5. We get ﬁnally 2 µ = UV µ + IR µ where (in n = 4 − εUV dimensions) −q 2 x(1 − x) + m 2 α 1 1 − (2 + γ ) − 1 dx ln 4 π 2εUV 4 4πµ2 α m + ( p + p )µ I1 + γµ (Z 1 − 1)(1) 2π 2 and (in n = 4 + εIR dimensions) UV µ = γµ (5.66) IR µ = γµ α α 1 (2q 2 − 4m 2 ) γµ ( p + p )2 I1 + 1 γ I1 + 1 I2 + 2 2 4π εIR 2π α − (5. The other integrals are at most only IR divergent. kα . To simplify notation we write the ﬁnal results in terms of the integrals 1 1 dx dx x I1 = I3 = −q 2 x(1 − x) + m 2 −q 2 x(1 − x) + m 2 0 0 (5. p ) − (Z 1 − 1) γµ = −ie dn k N n [( p + k)2 − m 2 ][( p + k)2 − m 2 ]k 2 (2π ) (5.63) We now introduce Feynman parametrization for the propagators and then have to perform the following integrations 1 2 0 dx dy y dn k [1. kα kβ ] n {k 2 + [ p x y + p(1 − x)y]2k}3 (2π ) (5.64) Only the integral with two powers of k in the numerator is UV divergent (but IR ﬁnite). so we take n = 4 + εIR there. We regularize it by taking n = 4 − εUV .192 5 Quantum electrodynamics We work in the Feynman gauge and assume that the external electron lines are on-shell p 2 = p 2 = m 2 .66a) ( p + p )µ m I1 2π .65) 1 −q 2 x(1 − x) + m 2 dx I2 = ln 2 2 4π µ2 0 −q x(1 − x) + m where q = p − p (we assume q 2 ≤ 0) and µ2 is an arbitrary mass scale. Using the identity γ µ / γµ = (2 − n)/ . We see that I3 = 1 I1 .

p ) = γµ F1 (q 2 ) + 1 (i/m)σµν q ν F2 (q 2 ). γν ] 2 (5.5.73) Notice that the radiative correction (5. we have Z 1 = Z 2 to this order.71) (Z 1 − 1)(1) = − 2π εUV and. Correction (5. Indeed from (5.67) Using the so-called Gordon decomposition ¯ ¯ u( p )γµ u( p) − 1 (i/m)u( p )σ µν qν u( p) = 1 (1/m)u( p )( p + p )µ u( p) ¯ 2 2 and the relation ( p + p )2 = 4m 2 − q 2 .2). One can check that for the on-shell renormalization given by the conditions ∂ ∂p / = 0.68) dx ln −q 2 x(1 − x) + m 2 + m 3 I3 4π µ2 1 + 1 γ I1 + 1 I2 2 εIR 2 (5. In our calculation we have reproduced the famous Schwinger result (Schwinger 1949) for the anomalous magnetic moment of the electron F2 (q 2 = 0) = α/2π (5.70) 2π We can now make several observations. one gets the following: F1 (q 2 ) = 1 1 α 1 − (2 + γ ) − π 2εUV 4 4 + (Z 1 − 1)(1) + + (m 2 − 1 q 2 )I1 2 F2 (q 2 ) α π 1 2 q 2 (5.73) is . At this point we encourage the reader to calculate the Thomson cross-section in the on-shell renormalization scheme (Problem 5. In the minimal subtraction scheme α 1 (5.56) and (5.49).2 Lowest order QED radiative corrections 193 The above result is often rewritten in terms of the Dirac and Pauli formfactors µ ( p. comparing with (5. 2 σµν = 1 i[γµ . p=m F1 (q 2 = 0) = 1 the relation Z 1 = Z 2 also holds. It depends on the renormalization scheme through the value of α(µ2 ) but that dependence is of the order O(α 2 ).69) it follows that Z1 = Z2 = 1 − 1 α m2 1 + 3 γ + 1 − 3 ln − 4 4 2 π 2εUV 4π µ εIR (5.72) in this case (the renormalization constants now become IR divergent).73) for the anomalous magnetic moment is ﬁnite before renormalization.69) − m2 α 2 m I1 (5.

Here the ﬁrst two terms are the dimensional regularization form of the famous double logarithms due to the simultaneous presence of IR and mass singularities. A straightforward calculation of the leading double logarithms gives a result which corresponds to the replacement 2/εIR → ln(m 2 /λ2 ). We recalculate now in the Feynman gauge the mass and vertex corrections in this approximation. We have 1 dn k 1 −i ( p ) = −e2 / N 2 (5. contains only single logarithms). .74) −1 ln from IR . In particular I2 can be written in terms of the Spence functions. for instance. the photon mass. in the on-shell renormalization.3 Massless QED For some high energy reactions it is useful to have results in the m = 0 approximation. As a ﬁnal remark we notice that the double logarithm terms do not depend on the renormalization scheme (Z i . Consider ﬁrst the self-energy correction.194 5 Quantum electrodynamics ﬁnite because the operator ¯ σµν F µν has dimension ﬁve and cannot be present in the original renormalizable lagrangian. It is generated by radiative corrections as a non-local term which becomes local in the low energy limit q 2 = 0 (see also Chapter 12).75) n (2π ) k ( p + k)2 where N = (2 − n)( p + / ) / k † It is easy to see that I1 ≈ I2 ≈ −1 m2 2 ln 2 −q −q 2 −1 −q 2 m2 1 2 ln ln + 2 2 −q 4πµ −q 2 −q 2 ln2 1 −q 2 + 1 π3 O 4 3 m2 q m 2 . 5. As the next point we discuss the structure of the leading logarithmic terms in our result for the vertex corrections in the limit |−q 2 | m 2 . In the minimal subtraction scheme we get α −q 2 − ln 4π 4π µ2 as a ﬁnite contribution to F1 from α 1 m2 1 2 m2 ln − ln + π εIR −q 2 4 −q 2 1 γ 2 UV and† m2 1 −q 2 m2 + 2 ln ln −q 2 4π µ2 −q 2 (5. The logarithms can also be when |q 2 | ξ 1 obtained simply by writing 0 = 0 + ξ1 and approximating the integrands in both integrals. The IR singularities can also be regularized by λ.

in the minimal subtraction scheme ( p) = p / / − p2 α1 + const. Using Feynman parametrization and the integrals a a (B. we get† in the minimal subtraction scheme V (p 2 = 0) = − α 2 4π εIR (5. µ2 part and the ﬁnal result is simply . Finally.79) where the pole in εIR reﬂects the IR singularity.80) where N = γµ [4 p · p + (n − 2)k 2 + 4( p + p ) · k] − ( p + p )µ 4/ − 2(n − 2)kµ/ k k ∞ † This result is immediate if we note that µ2 −(Z 2 − 1) is already UV ﬁnite and we can write it in (4 + εIR ) ∞ dimensions by changing ε → −εIR .18)–(B.76) in the following form ( p) = / α (2 − n) p / 4π × 0 1 ∞ − p2 q2 dq 2 q 2 4π µ2 −ε/2 1 + 1ε 2 (5.78) dx x −ε/2 (1 − x)1−ε/2 − (Z 2 − 1)(1) p / µ2 ∞ For p 2 = 0 we split the integral into two: 0 + µ2 which are IR and UV divergent. so we take ε → −εIR ). Then the µ2 part cancels the 0 −(Z 2 − 1) written in terms of εIR . We rewrite (5. respectively.77) The other way is to take external lines on-shell p 2 = 0 and to use dimensional regularization of the IR singularity. Then. when ε → 0 (to regularize the ﬁrst integral we work in n = 4 + εIR dimensions. after cancelling the UV divergence by Z 2 . One is to keep external lines off-shell p 2 < 0 to regularize the IR singularity. Note that for massless QED the ﬁnite wavefunction renormalization 1 is given solely by V ( p 2 = 0) and not its derivative. We have µ = −ie2 dn k N n ( p + k)2 ( p + k)2 k 2 (2π ) (5. It is also straightforward to repeat the calculation of the vertex correction in the massless electron case.3 Massless QED 195 (remember γ µ / γµ = (2 − n)/ ).5.76) We can now proceed in either of two ways.20) one gets (n = 4 − ε and α → αµε ) ( p) = / α1 1 p (2 − n) / π4 4π (1) / − (Z 2 − 1) p ≡ −ε/2 − p2 µ2 2 / V (p )p −ε/2 1 ε 2 1 0 x −ε/2 (1 − x)1−ε/2 dx (5. ln π 4 4π µ2 (5.

we now get (n = 4 − εUV ) UV µ α = γµ 2π × = γµ 1+ 1 ε 2 UV (2 − εUV ) 1 εUV −1 −q 2 4π µ2 −εUV /2 dx dy y 1−εUV [x(1 − x)]−εUV /2 + γµ (Z 1 − 1)(1) α −q 2 2π 4π µ2 −εUV /2 1 − 1 εUV 1 2 + γµ (Z 1 − 1)(1) (1 − εUV ) εUV (5. ﬁnite) in n = 4 + εIR dimensions µ by changing εUV → −εIR . Both have been regularized by working in n = 4 + εIR dimensions. 5. of course.e. to add counterterms.196 5 Quantum electrodynamics Repeating the steps which led us to the result (5. However. once (Z 1 − 1)(1) is explicitly determined. has poles in ε and we have to perform next the standard renormalization procedure. Landshoff. the calculated amplitude. in 4 − ε dimensions Feynman amplitudes are UV convergent. The dispersive technique has been extensively used in particle physics. for example. As we know.66).4 Dispersion calculation of O(α) virtual corrections in massless QED. So we can always write for them the dispersion relation without subtraction. We do not attempt its systematic discussion here.81) and† (n = 4 + εIR ) IR µ = γµ − × = γµ α 2π −q 2 4πµ2 εIR /2 1 − 1 εIR 2 dx [x(1 − x)]εIR /2−1 1 + 1 εIR 2 (2 + εIR ) 2 εIR 1+ π2 6 2 1 ε 2 IR dy y εIR −1 (1 − y) α − 2π −q 2 4πµ2 εIR /2 (5. we can rewrite UV (which is. in the minimal subtraction scheme. † (1 − ε) ∼ (1) − = (1)ε + 1 2 (1)ε2 = 1 + γ ε + 1 (γ 2 + 1 π 2 )ε2 . Eden. Then the full vertex is written in (4 + εIR ) dimensions. i. We note also that.82) The double pole structure in IR is due to the simultaneous occurrence of IR µ (massless photons) and mass (massless electrons) singularities. in (4 ∓ ε) dimensions It is instructive to repeat the calculation of the electron self-energy and vertex corrections using dispersion relations. Our purpose is mainly a technical one: to show yet another method of calculating Feynman diagrams which in some cases happens to be instructive. of course. referring the interested reader to other books (Bjorken & Drell 1965. 2 6 . Olive & Polkinghorne 1966).

86) We now introduce polar coordinates for the Euclidean vector r in n − 1 dimensions dn−1 r = |r|n−2 d|r| d n−1 = |r|n−2 d|r| dθ (sin θ)n−3 d n−2 (5. 5.6.6 we can write Im where d 2 = V (p 2 ) = − 1 e2 (2 − n) ( p 2 ) 2 d 2 (1 − p · k/ p 2 ) (5. 5.4 Dispersion calculation of O(α) virtual corrections 197 Fig.41) reads V (p 2 = 0) = 1 π d p2 Im p2 ) (5.83) The imaginary part of a Feynman diagram amplitude is obtained by the replacement p2 1 → −2πiδ( p 2 − m 2 ) ( p0 ) ≡ −2πiδ+ ( p 2 − m 2 ) − m 2 + iε in the Feynman integral (Landau–Cutkosky rule). Therefore for the diagram shown in Fig. Self-energy calculation The dispersion relation for the function V (q 2 V (p 2 ) deﬁned by (5.87) .5. l) centre of mass system it can be written as follows in terms of momenta p = k +l and r = 1 (k −l): 2 d 2 1 dn r δ( 1 p 2 + pr + r 2 )δ( 1 p 2 − pr + r 2 ) 4 4 (2π)n−2 1 dn r 1 δ( 1 p 2 + r 2 )δ( pr ) = 2 4 n−2 (2π) 1 1 = dn−1 r δ( 1 p 2 − r2 ) n−2 (2π) 2 p0 4 = (5.84) dn k dn l 2πδ+ (k 2 ) 2π δ+ (l 2 )(2π)n δ (n) (k + l − p) (2π)n (2π )n dn k = 2πδ+ (k 2 )2π δ+ (( p − k)2 ) (2π)n (5. In the (k.85) is the two-body phase space element in n dimensions.

78).90) We are now ready to calculate Im V ( p 2 ) and V (q 2 = 0).91) which is equivalent to (5. We shall limit ourselves to the UV ﬁnite part IR to understand better the origin of the IR and mass µ singularities.93) . 5. in n = 4 − ε dimensions. Vertex calculation Calculation of the vertex correction proceeds similarly.92) and using the kinematical variables deﬁned in Fig. Writing IR µ = γµ 2q 2 (q 2 ) (5. Since in cms p · k = 1 2 p we obtain (as usual.89) 2 (we specify n = 4 − ε and remember that r0 = 0 and p0 = ( p2 )1/2 in cms).198 5 Quantum electrodynamics where 0 ≤ θ ≤ π and due to an azimuthal symmetry of the problem we integrate over all but one angle θ to get dn−1 r = (r2 )(n−3)/2 dr2 (sin θ)n−3 dθ π (n−2)/2 ( 1 (n − 2)) 2 (5.7 we have Im (q 2 ) = −e2 d 2 1+ 4k · ( p − p ) 1 2q 2 k2 (5.88) For the phase space element integrated over angles one then gets d 2 = π 1−ε/2 1 (2π)2−ε (1 − 1 ε) 2 p2 4 −ε/2 (sin θ)n−3 dθ (5. we deﬁne dimensionless α by 2 α → αµε ) V (q 2 = 0) = − α (2 − n) 1 4π (1 − 1 ε) 2 2 ∞ 0 d p2 p2 p 2 4π µ2 −ε/2 0 1 dy [y(1 − y)]−ε/2 (5. Finally one can write dθ (sin θ)n−3 = dz (1 − z 2 )−ε/2 = dy 2[4y(1 − y)]−ε/2 where z = cos θ and y = 1 (1 + z) to get 2 d 2 = 1 p2 8π 4π −ε/2 1 dy [y(1 − y)]−ε/2 (1 − 1 ε) 2 (5.

(the term proportional to / when integrated must give A p + B p .96) The integral over dy gives [ (1 + 1 ε) ( 1 ε)/ (1 + 1 ε) − 2 (1 + 1 ε)/ (2 + ε)] 2 2 2 2 and it is divergent for ε → 0. l ) cms system (l + l = p + p = 0) r = l. the exchanged photon has to have zero momentum. i.e.97) . In the present case r = 1 (l − l ) and in the (l. Taking the z axis 2 in the direction p we have k 2 = −2l · p = −2l0 p0 + 2l · p = −2( 1 q0 )2 + 2( 1 q0 )z = − 1 q 2 (1 − z) 2 2 2 where z = cos θ and ﬁnally (n = 4 + ε) (Q 2 ) = α 1 1 2 4π 4πµ (1 + 1 ε) 2 × 0 1 ∞ 0 dq 2 q2 q 2 − Q 2 4πµ2 ε/2−1 dy 1 − 2 [y(1 − y)]ε/2 1−y (5. We then get the ﬁnal result ∞ 0 dq 2 (q 2 )ε/2−1 = (−Q 2 )ε/2−1 (1 − 1 ε) ( 1 ε) 2 2 q 2 − Q2 (5.94) Therefore Im (q 2 ) = −e2 d 2 (5. To integrate over q 2 (for Q 2 < 0) we change variables twice: q 2 = −Q 2 x and u = 1/(1 + x).95) The next step is to use the expression (5.90) for d 2 .7. so it does not k / / contribute to the on-shell limit).5. 5.4 Dispersion calculation of O(α) virtual corrections 199 Fig. In our speciﬁc reference frame z = 1 corresponds to l p. The divergence comes from the region y ≈ 1 and therefore z ≈ 1. where d 2 = dn k 2πδ+ (( p + k)2 )2π δ+ (( p + k)2 ) (2π)n dn l dn l = 2πδ+ (l 2 ) 2π δ+ (l 2 )(2π )n δ (n) (q + l + l ) (2π)n (2π )n 2 1 + 2 2 k q (5.

We use the static approximation to the general formulae (see. According to the general rules the cross section reads 1 d3 p (0) dσ0 = (5.e. i. 5.82). for example.8. using (5. Finally. Halzen & Martin (1984)): ﬂux−1 = (2π)4 δ(P + p − P − p ) 1 2E p 2E P |v − V| → 1 1 2E p |v| 2|p| d3 P d3 p d3 p → 2π δ(E − E) 2E P (2π )3 2E p (2π )3 2E p (2π )3 .98) which can be seen to coincide with (5. The pole in ε is due to the lower limit of integration over q 2 .99) |M0 |2 2π δ(E − E) 2|p| (2π )3 2E where (0) |M0 |2 = 1 2 |u( p )γ 0 u( p)|2 ¯ pol Z 2 e4 |q|4 (5. due to a massless electron–positron intermediate state (mass singularity).92) we get (n = 4 + ε) IR µ = γµ − α 2π −Q 2 4πµ2 ε/2 2 (1 − 1 ε) 2 ε (1 + 1 ε) ( 1 ε) 2 2 − (1 + ε) 2 (1 + 1 ε) 2 (2 + ε) (5. 5.5 Coulomb scattering and the IR problem Corrections of order α In the Born approximation the scattering of an electron from a static Coulomb potential A0 (x) = Z e/4π|x|. 5. Ai (x) = 0 is described by the diagram in Fig.200 5 Quantum electrodynamics Fig.100) and Z e2 /|q|2 is the Fourier transform of the A0 (x).8.

. we use the relativistic normalization of 2E particles in a unit volume V = 1 – see Appendix A).103) we have neglected the terms O(α 2 ) and the corrections which are regular in the IR limit: (q 2 ) and the Pauli formfactor 2 F2 (q ). vacuum polarization and the vertex.102) where β = |p|/E and p · p = 2|p|2 cos θ.) Including virtual corrections to order α we obtain therefore (1) (0) dσ0 = dσ0 [1 + 2 1 + 2F1 (q 2 )] (5.106) The above expressions are valid for |q 2 | m 2 and in the minimal subtraction 2 scheme.105) (one-quarter (one-half) of the ln(m 2 /4π µ2 ) comes from the dimensional regularization of the UV (IR) divergence) and 2F1 (q 2 ) = −2 m2 −q 2 α1 α 1 −q 2 1 −2 ln 2 + ln2 ln π 4 4π µ2 π εIR m 4 −q 2 −q 2 −q 2 1 −q 2 + ( 1 γ − 1) ln 2 + ln ln 2 2 m 2 4π µ2 m (5. (In the amplitude for bremsstrahlung of a real photon we shall only be interested in the limit of a soft photon. The ﬁrst term in F1 (q ) comes from the dimensional regularization of the UV divergences.103) ≈ u( p )γ0 u( p)[1 + ¯ + F1 (q 2 )] In the ﬁnal expression (5. p )](1 + 1 2 1 )u( p)[1 + (q 2 )] (5. Together with the analogous term in 1 it gives (α/4π) ln(−q 2 /m 2 ).5. As usual 1 2 pol |u( p )γ 0 u( p)|2 = 1 ( p + m)αβ (γ 0 )βδ ( p + m)δγ (γ 0 )γ α = ¯ / / 2 1 2 Tr[· · ·] (5.101) Performing the necessary calculations we ﬁnd (0) dσ0 = 1 Z 2 α 2 4 1 |p|4 sin4 ( 1 θ)/E 2 2 [1 − β 2 sin2 ( 1 θ)] d dE δ(E − E ) 2 (5.104) where using the results of Section 5. The remaining µ2 dependence is due only to our method of regularization of the IR singularity and should cancel out in every cross section free of such singularities. We shall now study real and virtual corrections of order α to the Born cross section. Virtual corrections of order α are easily included using our results for the electron self-energy.5 Coulomb scattering and the IR problem 201 for P = P = 0 (remember. The amplitude now reads (0) M0 = u( p )(1 + ¯ 1 2 1 )[γ0 + 1 0 ( p.3 2 1 =2 m2 α 1 3 + ln − 1 + 3γ 4 π εIR 4 4π µ2 (5.

as before. Later we shall discuss this problem in more general terms. 5. and dσ1γ is the differential cross section with a photon in the ﬁnal state. (1) (1) Working to order α we have to include dσ0 given by (5. Using the standard rules we get the following: (1) (1) dσ1γ = |M1γ |2 1 d3 p d3 k 2π δ(E − E − k0 ) |v| (2π)3 2k0 (2π )3 (5.100) and the sum over photon polarizations has been taken. (1) The origin of the IR singularity encountered in our cross section dσ0 is in the masslessness of the photon. 5. The ﬁrst term comes from the interference of the two diagrams .109) (0) In (5. One feels that a summation over experimentally indistinguishable ﬁnal states is necessary. In consequence the system has a high degree of degeneracy: a ﬁnal state consisting of a single electron state is not distinguishable by any measurement from an e + n soft (k → 0) γ s state.9. E and E are. Therefore the previously deﬁned cross section dσ0 is not a physically meaningful (measurable) quantity.9.104) and dσ1γ which is the cross section for a single photon emission given by the Feynman diagrams shown in Fig. or in other words in the long range nature of the electromagnetic interactions.107) where E is the energy resolution of the detection equipment. the energies of the initial and ﬁnal electrons in the laboratory system. We shall explicitly check that the cross section σ is free of the IR singularity.109) the amplitude |M0 |2 is deﬁned by (5.108) where in the soft photon approximation (momentum k neglected in the numerators of the Feynman expressions) (1) (0) |M1γ |2 = |M0 |2 e2 2p · p m2 m2 − − k · pk · p (k · p)2 (k · p )2 (5. We observe that the corrections due to the soft photon emission factorize. E = E − E is the energy of the undetected real photon. Following Bloch and Nordsieck and working to order α we introduce a ‘measurable’ cross section σ as follows E σ = 0 dE d (1) (1) (σ + σ1γ ) dE 0 (5.202 5 Quantum electrodynamics Fig.

Let us consider I2 ﬁrst. . .5 Coulomb scattering and the IR problem 203 shown in Fig. the result dσ (0) dσ = 0 [1 + 2 d d where. At this point we get for the cross section. σ . . Our last task is to calculate the corrections Ii . We replace d3 k dn−1 k → 2|k|(2π )3 2|k|(2π )n−1 2 2 where |k| = (k1 + · · · + kn−1 )1/2 and introduce spherical coordinates in n dimensions (see (5. dθn−2 π n/2−1 ( 1 n − 1) 2 d|k| |k|n−2 1 −1 dz (1 − z 2 )n/2−2 (5. 5.110) where the last step is legitimate in the soft photon limit (we denote |k| = |k| = k0 ). They are IR divergent and for consistency with the calculation of virtual corrections we must regularize them by working in 4 + εIR dimensions.113) where z = cos θ1 and the integration over the remaining angles has been done in the last step (assuming the integrand depends only on θ1 ).5. We see also that the dominant contribution to I2 comes from that region of phase space where k p. sin θn−3 dθ1 . that part of the radiation tends to be collimated around the direction of motion .111) d3 k 2p · p 2|k|(2π )3 k · p k · p (5. Writing (k · p)2 = |k|2 (E − |p|z)2 we can easily perform the necessary integrations.112) and I2 and I3 are self-evident. We now proceed as follows E dE 0 (1) dσ1γ E dE = 0 dE E d3 k E |p |2 d (1) 2π δ(E − E − k0 ) |M1γ |2 2k0 (2π )3 |p| (2π)3 ∼ = (0) |k|< E dσ0 d3 k e2 dE 0 2|k|(2π )3 0 2p · p m2 m2 × − − k · pk · p (k · p)2 (k · p )2 dE (5. So. I1 = e 2 0 |k|< E 1 + 2F1 (q 2 ) + I1 + I2 + I3 ] (5.9. for example. We see that the probability of a soft photon emission into an interval d|k| is proportional to d|k|/|k| which is easy to understand on dimensional grounds (soft emission does not depend on the electron momentum p). .88)): dn−1 k = =2 d|k| |k|n−2 sinn−3 θ1 sinn−4 θ2 . the next two are just the respective amplitudes squared.

depend on the method of regularization of the IR singularity. where α ON = 1/137. of course. exponentiate. The transition probability decreases with decreasing E. (5.114) For I3 we get the same result with E changed into E . The coupling constant α is deﬁned by the minimal subtraction renormalization scheme and α = α ON [1 + O(α ON )]. In this frame ˜ ˜ ˜ p · k = |k|( E − |p| cos θ) ˜ E + |p| cos θ) ˜ ˜ p · k = |k|( where the sign ˜ denotes quantities in the Breit frame and the condition |k| < E ˜ in the laboratory system translates into |k| < E sin( 1 θ). π εIR 2 4π µ2 2 m (5. To calculate the term I1 it is convenient to choose the reference frame in which p + p = 0 (the Breit frame).204 5 Quantum electrodynamics of the initial electron. The ﬁnal result reads I2 = − α 1 1 ( E)2 1 4E 2 + ln − ln 2 + const. Adding all the contributions we obtain the ﬁnal result (0) dσ0 α −q 2 dσ ( E)2 −q 2 3 −q 2 + ln ln 2 + ln 2 = 1+ ln d d π ( E)2 −q 2 m 2 m + f (θ) + const. This can best be formulated if we represent .116) (remember that −q 2 ∼ 4E 2 ). the splitting into real and virtual contributions does. Our calculation is valid for |(α/π) ln[−q 2 /( E)2 ]| 1. where θ is the scattering 2 angle in the laboratory (indeed pi · pf = p · p = 2E 2 sin2 ( 1 θ) in the laboratory 2 ˜ ˜ and pi · pf = 2 E 2 in the Breit frame. as we shall see later. Otherwise the correction becomes large and we must include higher order contributions which. The cross section dσ/d is indeed free of = the IR singularity and depends on the experimental resolution E. So working to order α we can neglect the difference. The limit |q 2 | m 2 can also be easily obtained. In the Breit system the 2 integration over the angle can be performed with the help of Spence’s functions and we get α 1 −q 2 1 ( E)2 −q 2 1 2 m 2 −q 2 I1 ∼ 2 ln 2 + ln ln 2 + ln + 1 γ ln 2 + f (θ) = π εIR m 2 4π µ2 m 4 −q 2 2 m (5. It should be noted that although the ﬁnal result (5.116) does not.115) for |q 2 | m 2 . Finally we observe certain systematics in the cancellation of the IR singularities between virtual and real corrections. The results presented here are for |q 2 | m 2 . so E = E sin( 1 θ)).

p ) (5. the cross section for real emissions by cut diagrams as shown in Fig.10.5.10. Let the lowest order matrix element for our process be M0 ( p.1 and the factor 2 takes account of 2 two identical interference terms. 5. p ) = ∞ n=0 (n) M0 ( p. separately. Frautschi & Suura 1961). The complete matrix element is then M0 ( p. Let us consider a process in which there are some electrons and non-soft photons in the initial and ﬁnal states and we represent their momenta by p and p .5 Coulomb scattering and the IR problem 205 Fig.117) . where the factor 1 follows from the rule of Section 4. p ) be the contribution corresponding to all diagrams in which there are in addition n virtual photons. The cancellation occurs between contributions to the cross section represented by the diagrams in each column. 5. p ) (n) and let M0 ( p. An important fact is that in QED the IR divergent terms due to virtual corrections and to real soft emissions exponentiate and cancel each other in the physically meaningful (observable) cross section. (0) respectively. IR problem to all orders in α The previous results can be generalized to all orders in the electromagnetic coupling constant α (Yennie.

ln(−q 2 /m 2 ) and ln(−q 2 /λ2 ). For the Coulomb elastic scattering we get from (5.119) is to convince oneself that IR divergences originate from only those virtual photons which have both ends terminating on external lines. where E is the energy resolution of the detectors. The differential cross section .106). 1961) that the M (n) s have the following structure (0) M0 = m 0 (1) M0 = m 0 α B + m 1 (2) 2 2 M0 = m 0 (α B) /2 + m 1 α B + m (5.120) ˆ where M is free of IR singularities.118) . in the socalled double logarithmic approximation. Thus we get M0 = exp(α B) ∞ j=0 ˆ m j = exp(α B) M0 (5.104)–(5. . . Let us now consider an inclusive cross section summed over any number of undetected real photons in the ﬁnal state. An important intermediate step in proving (5.120) we have changed the dimensional IR regularization into a cut-off by a small photon mass λ: 1/εIR → 1 ln(m 2 /λ2 ) (see Section 5. The exponent in (5. We observe that due to the IR divergences encountered in order-by-order calculation the complete amplitude for purely elastic scattering is zero in the limit λ → 0.206 5 Quantum electrodynamics It has been shown (Yennie et al.106).119) and the IR divergent exponent can be read off from our lowest order calculation.120) is called the on-shell electron formfactor in the double logarithmic approximation. This is just a reﬂection of the fact that electron scattering is always accompanied by soft photon radiation and the cross section for elastic scattering is not a physical observable. (5. Mel = exp − α −q 2 −q 2 ˆ ln 2 ln 2 Mel π m λ (5. Actually. this is to be expected on physical grounds since long wave photons should not see charge distribution at short distance.2) in order to follow the traditional notation and in the 2 exponent we have only retained terms quadratic in large logarithms.119) and (5. To write (5. n (n) i M0 = m n−i (α B) /i! i=0 (0) where m j is an IR ﬁnite function (independent of n) of order α j relative to M0 and each factor α B contains the IR contribution from one virtual photon. with total energy E such that E ≤ E.

physical quantity. k1 . as mentioned before.5. However. Virtual corrections to all orders in α are included in (5.126) d3 ki p(ki )[exp(−iyki0 ) − 1] 2ki0 (5.121) where p and p represent.127) Thus all IR singularities due to real emissions are collected in the exponential factor ˜ ˜ exp(2α B). each σm is zero due to the IR divergences. .5 Coulomb scattering and the IR problem 207 for emission of m undetected real photons with total energy E reads 1 dσm = exp(2α B) dE m! m i=1 d3 ki δ E− 2ki0 ki0 ρm ( p. Soft real photons must terminate exclusively on external lines in order to contribute IR singularities.122) turns out to be free of IR singularities and thus it is a ﬁnite. . an inclusive cross section deﬁned as ∞ dσ dσm = dE dE m=0 (5. momenta of all particles in the initial state and of the detected particles in the ﬁnal state. where B is given by our order α calculation. p . p . as in our order α calculation.109) ρm ( p. . Writing (Yennie et al.121) and. .124) where p(k1 ) is given by the expression in the square brackets in (5. as before.125) d3 ki p(ki ) 2ki0 (5.116)): ˜ 2α(B + B) = α −q 2 E2 −q 2 ln 2 ln 2 + ln π E −q 2 m (5.109) and. and they cancel with virtual IR singularities in B giving together (see (5.128) .123) using a generalization of (5. km ) i (5. 1961) δ E− i ki0 = 1 2π ∞ −∞ dy exp iyE − iy i ki0 (5. k1 . km ) = σ0 p(k1 ) p(km ) ˆ (5. neglecting for soft photon radiation the energy–momentum conservation constraints one gets the following result: dσ ˜ ˆ 1 = exp[2α(B + B)]σ0 dE 2π where ˜ 2α B = and D= |k|<E |k|<E ∞ −∞ dy exp[iyE + D(y)] (5.

15) and (5.120)) and the off-shell electron formfactors in the double logarithmic approximation using the light-like gauge n · A = 0 where n 2 = 0.4 Calculate the off-shell electron formfactor (Sudakov formfactor) in the double logarithmic approximation µ ( p1 . | p2 | m2 5. where P( p 2 ) = p 2 ( p 2 ) and ( p 2 ) is deﬁned by (5. Hint: apply the BRS transformation to the Green’s functions 0|T Aµ (x)ω(y)|0 = 0 and 0|T (x) ¯ (y)ω(z)|0 = 0 where ω(x) is the free auxiliary scalar ﬁeld.21) using the invariance of the QED Green’s functions under the BRS transformations (Problem 3. Hint: calculate the difference P( p 2 + i ) − P( p 2 − i ). Compare with the calculation in the Feynman gauge. as expected. p2 . study the pole structure of the integrand as a function of k0 and integrate over dk0 ﬁrst. . For a detector resolution E the integration over E up to E is necessary. For the forward Compton amplitude and in the limit of the photon momentum k → 0 (the Thomson limit) express the amplitude in terms of the electron propagator. where q 2 ˆ is some characteristic for the process four-momentum squared and µ2 is the renormalization point. 5. in general. Show that in the on-shell renormalization scheme the coupling constant e can be deﬁned by means of the Compton amplitude in the Thomson limit. One should also remember that the IR ﬁnite part σ0 contains.121) is α = α(µ2 ). 0). 5. The coupling constant α in (5.5 Calculate the on-shell (see (5. Problems 5.3).1 Derive the Ward identities (5. terms like ln(q 2 /µ2 ). 5. q) = γµ exp − |q 2 | |q 2 | α ln 2 ln 2 2π | p1 | | p2 | |q 2 | 2 2 | p1 |.208 5 Quantum electrodynamics The cross section dσ/dE is ﬁnite.3 Study the singularity structure of the photon propagator in the one-loop approximation in the complex p 2 -plane and derive the Landau–Cutkosky rule.26): choose the cms frame p = ( p0 .2 Derive the Ward identity for the truncated connected Green’s function of the Compton amplitude.

of course. and also in this section. Each renormalization prescription can be interpreted as a particular reordering of the perturbative expansion and expressing it in terms of the new renormalized constants. Once the inﬁnities of the theory have been subtracted out by a renormalization prescription R one is still free to perform further ﬁnite renormalizations resulting in each case in a different effective renormalization R . For a single-mass-scale-dependent subset of renormalization transformations we derive in the following a differential equation which controls the changes in the renormalized parameters induced by changing µ: the RGE. only subsets of arbitrary renormalization prescription transformations which can be parametrized by a single mass scale parameter µ are discussed. A change in renormalization prescription is compensated by simultaneous changes of the parameters of the theory so as to leave.) Most often. (In practice. the set of renormalization prescription transformations {R} does not always have a group structure so the name is partly due to historical reasons. The latter are. by construction.1 Renormalization group equation (RGE) Derivation of the RGE The spirit of the renormalization group approach lies in the observation discussed in Chapter 4 that in a speciﬁc theory the renormalized constants such as the couplings or the masses are mathematical parameters which can be varied by arbitrarily changing the renormalization prescription. all exact physical results renormalization-invariant. the mass deﬁned as the pole of the propagator is a physical constant) which are directly measurable and therefore renormalization-invariant. for example.6 Renormalization group 6. the value of the subtraction point in the µ-subtraction schemes or the dimensionful parameter µ in the minimal subtraction prescription. 209 . in each case related differently to physical constants (for example. there is a residual renormalization scheme dependence in each order of perturbation theory. Actually. The parameter can be.

R) = Z (R . unless R j = Rk . in addition.1) dx 2x(1 − x) ln µ2 x(1 − x) + m 2 2 µ2 x(1 − x) + m 2 1 (6. Suppose we work in the µ-subtraction scheme and we want to relate two renormalizations performed at subtraction points µ1 and µ2 . to each element Z (R . R) for arbitrary R and R . R).3) This representation of Z obeys the composition law (6.2) is not obeyed. Rl ) (6.1) is not always deﬁned. an element of the set Z (R . Among the elements of this set there is the composition law Z (R . unless m = 0. R )Z (R . however. Their relation to the bare Green’s functions B will be R = Z (R) B where Z (R) denotes the appropriate product of renormalization constants deﬁned within the R scheme.2) is not. R) = Z (R )/Z (R) Let us now consider the set of all possible Z (R . The relevant quantity will be Z (µ1 . R) R where Z (R . for arbitrary pairs of Z s: the product Z (Ri . If we choose another renormalization scheme R then. R) there is an inverse Z −1 (R . R) = 1 The composition law (6. in general.1) but the group multiplication law (6. R) = Z (R. µ2 ) which in the one-electron loop approximation reads Z (µ2 . namely R = Z (R . . R) and. R ) and we can deﬁne the unit element as Z (R. µ1 ) = 1 + α 2π 1 0 (6.210 6 Renormalization group Let us assume a particular renormalization scheme R and call R the Rrenormalized 1PI Green’s functions. A simple example of the above considerations is provided by the photon wave-function renormalization constant Z 3 (de Rafael 1979). there exists a relation between both renormalized Green’s functions. R = Z (R ) B Obviously. R j )Z (Rk .

These include the minimal subtraction prescription and. i. ε) = = 1µ ln(Z 3 Z 0 ) −→ γm (λ R ) 2 ε→0 m R dµ dµ (6. for example.e.4) exists since the theory is renormalizable. in 4 − ε dimensions (n) R ( pi . λB .5) We can deﬁne now the following functions:† β(λ R . . of course. λ R . ε) = 1 µ ln Z 3 −→ γ (λ R ) 2 ε→0 dµ µ dm R d −1 γm (λ R . µ.8) † Similar considerations apply to additive renormalization.7) (6. in a trivial way any prescription for a massless theory. m B . λ 4 theory can be derived as follows.6) (6. λB .1).4) (for the deﬁnition of the renormalization constants Z i see Section 4. µ. ε) ∂µ dµ ∂λ R dµ ∂m R d n/2−1 (n) = 1 n Z3 µ Z 3 B ( pi . i. ε) where λ R is deﬁned as dimensionless. λ R . m R . ε) = µ (6. m B .4) with respect to the renormalization scale µ for ﬁxed λB . ε) (6. The ﬁnite limit (for ε → 0) of (6. µ.6. ε) R 2 Z 3 dµ d d 3 −1 λR = λR µ Z3 Z1 − ελ R −→ β(λ R ) ε→0 dµ dµ d γ (λ R . The RGE will be derived by differentiating relation (6. dimensionful in 4 − ε dimensions. ε) is in an obvious way independent of the renormalization scale µ. m B and ε. −2 λB µ−ε = Z 3 Z 1 λ R Similarly. m B . The (n) bare Green’s function B ( pi .1 Renormalization group equation (RGE) 211 In this section we shall discuss the RGE in renormalization schemes such that renormalization constants Z i are mass-independent (mass-independent renormalization schemes). m R .4) and using d µ = 0. λB . ε) = Z3 n/2 (n) B ( pi . The RGE in. λ R . We begin with the relation between the renormalized and bare 1PI Green’s functions in the theory regularized by the dimensional method. we have −1 m2 = Z 3 Z 0m2 B R B /dµ Differentiating (6. In the mass-independent renormalization scheme we have Z i = Z i (λ R . m R . ε) 2 dµ 1 d = 1 n µ Z 3 (n) ( pi .e. one obtains dλ R ∂ ∂ dm R ∂ (n) + + R ( pi .

m R .6) and (6.11) as differential equations ¯ ¯ ∂ λ(t. γm and γ depend on λ R only.λ) dx/β(x) = t λ m(t.8) exist since the theory is renormalizable) we obtain the ﬁnal result µ ∂ ∂ ∂ + β(λ R ) + γm (λ R )m R − nγ (λ R ) ∂µ ∂λ R ∂m R (n) R ( pi . λ. λ R . m(t.9) describes the change of renormalized parameters and multiplicative scaling factors for the renormalized Green’s functions which compensates a change of the renormalization scale µ. λ)/∂t = β(λ) ¯ ¯ ∂ m(t. with the boundary conditions ¯ λ(0. Solving the RGE Eq. and µ0 is some ﬁxed scale. λ. λ.λ) λ dx (γm (x)/β(x)) = m exp 0 1 ¯ γm (λ(t )) dt (6. In general they also can depend on the ratio m R /µ. in the µ-subtraction scheme. m. m(0. λ).5) (ﬁnite limits of relations (6. m) = m ¯ (6. We then have a homogeneous equation µ(d/dµ) (n) )( pi . λ. This happens.9) It is important to remember that only in mass-independent renormalization schemes or for m = 0 do the functions β. λ.11) Taking the partial derivative ∂/∂t one can rewrite (6.8) simply express the µ-independence of the bare parameters λB and m B .9) suppose for a moment that γ (λ R ) ≡ 0. As we shall see the meaning of this equation is that the dependence of the function (n) on t and hence on µ can be R ¯ expressed solely through the functions λ(t. m)/∂t = γm (λ)m ¯ in accord with (6. (6. m) deﬁned by the equations ¯ ¯ λ(t. Taking the limit ε → 0 in (6. µ) = 0 which R can be written in the form (subscript R in λ R and m R will be omitted from now on): − ∂ ∂ ∂ + β(λ) + γm (λ)m ∂t ∂λ ∂m (n) R ( pi . for instance.212 6 Renormalization group The differential equations (6.6) and (6. λ R . We recall also that the parameters λ and m are renormalized at µ. λ) = λ.8). µ) =0 (6. m) = m exp ¯ ¯ λ(t.10) where µ = exp(−t)µ0 . µ0 exp(−t)) = 0 (6. m R .5)–(6.12) . To solve (6.

13) we can easily check that any function f (λ. m.6. µ0 ) = 0 (6. λ). λ) = 0 (6. µ).15) ¯ ¯ where λ(t) stands for λ(t. λ. µ) = (n) ¯ ¯ R ( pi . m(t).9) reads (n) R ( pi .19) . λ. using (6. ¯ differentiating λ(t.11) with respect to λ one gets ∂ ∂ ¯ − + β(λ) λ(t. µ0 ) is µ-independent and satisﬁes the equation µ ∂ ∂ d f = − + β(λ) dµ ∂t ∂λ exp(−t)) = µ ¯ f (λ(t. m(t). µ exp(t)). λ). m.16) as can be checked by applying to it the differential operator of (6.16) relates a Green’s function (n) ( pi . λ. The renormalized parameters are renormalization-scale dependent. m. we can write a Laurent expansion for F ( p 2 . µ) − m2 F (6. λ. The solution to the inhomogeneous (6. λ. µ). µ0 (n) ¯ ¯ R ( pi . µ) = 0 Assuming that R + ˜ ( p 2 . m(t). λ) given by the ﬁrst equation of (6. µ0 ) µ exp(t) (6.9). (6. Being by deﬁnition renormalization-scheme-independent P must satisfy the homogeneous RGE: µ d ∂ ∂ ∂ p= µ + β(λ) + γm (λ)m P(λ. µ exp(t)) exp −n 0 1 ¯ γ (λ(t )) dt (6. µ0 ) = f (λ(λ. m.1 Renormalization group equation (RGE) 213 We also observe that (6. Let us consider the scalar propagator ( p 2 ) which satisﬁes the equation [R + 2γ (λ)] ( p 2 . µ) = 0 dµ ∂µ ∂λ ∂m (6. m. λ(t).13) ∂t ∂λ ¯ Using (6. µ) ≡ RP(λ. t.14) Therefore the solution to (6. where λ(t) and m(t) can be understood as the renormalized parameters corresponding to the scale µ exp(t).18) : has a pole at p 2 = m 2 . for instance.17) We shall now check. m. λ. λ. m. Any physical quantity P (renormalization-scheme-independent) can be calculated in terms of them: P = P(m. that this is indeed the case for the physical mass deﬁned as the position of a pole in a propagator and for renormalized S-matrix elements. λ(t). m. λ(t). t).10) can be written as follows (n) R ( pi . to the Green’s function (n) ¯ ¯ ¯ ¯ R ( pi . Eq. with the renormalized parameters R equal to λ and m at the renormalization scale µ. µ) = p2 (6.11) can be written in the form of the RGEs.9).

λ. m F (λ. µ) and the residue of the pole R(λ.20). Using the basic property of such functions we can write for it the following equation: ∂ ∂ ∂ ρ (6. ρ and R µ.19) in (6. m.23) +m +µ − D (n) (ρpi . m.27)). λ. m. λ.9) and m i2 and R satisfy (6. λ) (6. µ) = 0 (6. µ) = 0 (6. µ) = µ D f (ρ 2 pi · p j /µ2 . m. λ.23) one gets (t = ln ρ) − ∂ ∂ ∂ + β(λ) + (γm − 1)m − nγ (λ) + D ∂t ∂λ ∂m (n) R (ρpi . On dimensional grounds it can be therefore written as follows R (n) R (ρpi .21) One must stress that it is an exact S-matrix element which is renormalizationscheme-independent.9) and (6. In perturbation theory truncated at order N there is always a residual renormalization scheme dependence O(λ N +1 ).18) and equating the residua of the poles at p 2 = m 2 F Rm 2 (m. µ) = 0 F [R + 2γ (λ)]R(λ.20) We consider next an arbitrary n-scalar particle S-matrix element. Let us take a Green’s function of dimension D : (n) ∼ [M] D . we have R lim R n/2 2 pi2 →m i (n) = lim RR n/2 2 pi2 →m i (n) = lim [nγ (λ) − nγ (λ)]R n/2 2 pi2 →m i (n) =0 (6. µ) = 0 R ∂ρ ∂m ∂µ Combining (6. We notice ﬁrst that for m = 0 and for a non-interacting theory (β ≡ γ ≡ 0) the solution to (6. namely to discuss the change in the Green’s functions when all momenta are rescaled pi → ρpi (but for ﬁxed µ). µ) = ρD (n) R ( pi . λ.25) . Green’s functions for rescaled momenta The RGE is used most frequently in a somewhat different context. µ) (6. m. Since (n) satisﬁes (6.24) reads (n) R (ρpi .24) which is the result we are looking for.22) We observe that (n) . µ) themselves satisfy RGEs which can be derived by using (6. m.214 6 Renormalization group The position of the pole. m. λ. This is obtained from the 1PI Green’s function (n) by going to the physical poles pi2 = m i2 and multiplying by R n/2 (see (4. is a homogeneous function of order D of variables m. m/µ.

λ(t).s. 2 pi < 0.s. exp(t)µ) × exp −n = ρD γ dt −n γ dt (n) ¯ ¯ R ( pi . λ.16) and the last step is a change of the mass scale by a factor ρ.h. unlike (6. relates Green’s functions which both have the same renormalization scale µ. exp(t)µ) exp −n γ (t ) dt (n) ¯ ¯ R (exp(t) pi .26) gives us Green’s functions for the rescaled momenta exp(t) pi (off¯ mass shell) in terms of the ‘effective’ parameters λ(t) and m(t) which are solutions ¯ of (6. µ) = (n) ¯ ¯ R ( pi . Strictly speaking the RGE solution (6. λ(t).9) and it reads (n) R (exp(t) pi . m. RGE in QED We choose to work in the set of mass-independent renormalization schemes and in a covariant gauge deﬁned by the gauge-ﬁxing term: −(1/a)(∂µ Aµ )2 .24) can be written using arguments similar to those used to solve (6. m (t) in the r. and λ. To use these results in physical applications one must know the functions β(λ). away from possible IR problems and new thresholds associated with timelike momenta. the Green’s functions exhibit canonical scaling (see Chapter 7). but different values of the renormalized couplings: λ. therefore. λ(t).e. m (t). that they are Green’s functions of different theories. λ(t). The general solution to (6. Result (6. The RGE . m (t). We notice that (6. One can say. m(t). µ) exp t Dt − n 0 ¯ γ (λ(t )) dt (6.11). γ (λ) and γm (λ) which can only be calculated in perturbation theory.26). m.h.6.16).26) applies in the deep Euclidean region. µ) exp The ﬁrst step corresponds to a change of the renormalization point according to (6.26) where ¯ m (t) = m(t)/ρ ¯ ¯ and λ(t) and m(t) satisfy (6. µ) = = (n) ¯ ¯ R (exp(t) pi . m ¯ ¯ in the l.1 Renormalization group equation (RGE) 215 i. λ. exp(t) exp(−t)m(t).11). This solution can also be obtained directly by the ¯ following steps (n) R (exp(t) pi .

30) Using our previous experience we can also derive the RGE analogous to (6.6)–(6. a). α) = a ¯ (6. B = Z 2 (for deﬁnitions of the renormalization constants. a)∂t = δ(α. m B = Z m m.n F ) ( pi . a) = 0 (6. ε) = lim µ ε→0 ε→0 d λ dµ (6. The only new element is the −1 µ-dependence of the gauge-ﬁxing parameter α = Z 3 aB which is responsible for the term δ(α.16) with the running parameters given by (6.31) ﬁxed λB . We recall the deﬁnitions (6. a) = lim β(α. a) = lim µ ε→0 d da = lim −aµ ln Z 3 dµ ε→0 dµ (6. for example. a) = lim 1 µ 2 ε→0 dµ d γF (α.216 6 Renormalization group can then be derived following the steps outlined previously and we get µ ∂ ∂ ∂ ∂ + β(α. ε) = lim µ ε→0 ε→0 (n γ . a) ∂µ ∂α ∂m ∂a − n F γF (α.8).6)–(6. ¯ ¯ ¯ a(0. in addition. a) where in analogy to (6. 6. β(λ) = lim β(λ.2 Calculation of the renormalization group functions β. ∂ a(t. a) + γm (α. AµB = Z 3 Aµ . m. µ. α.27) is obtained by the obvious modiﬁcation of (6. where α = e2 /4π. see Chapter 5). a) − n γ γγ (α.28) −1 Derivatives are taken for ﬁxed αB and ε. αB = µε Z 3 α in 1/2 1/2 −1 n = 4 − ε dimensions. a) = lim 1 µ ln Z 2 2 ε→0 dµ γm (α. β(λ).29) The solution to (6.26) for QED. a)∂/∂a with δ(α.8) β(α. α. a)m + δ(α. a. γm We address ourselves to the problem of calculating functions.11) and. γ (λ) and γm (λ) in a mass-independent renormalization scheme. a) = lim µ (6.27) d α dµ d ln Z m ε→0 dµ d ln Z 3 γγ (α.ε . γ .

32) εν ν −2 where. be multiple poles. . am (λ) = am1 λ + am2 λ2 + · · · + amn λn Using (6. the simple pole in ε. ε) = −ελ + β(λ) 2 da1 β(λ) = λ (6. for ε → 0. . exists). therefore. This reﬂects an important aspect of renormalization theory.6) we get −1 β(λ. The coefﬁcients aν (λ) are a1 (λ) = a11 λ + a12 λ2 + · · · + a1n λn . Equating the residua of the poles in ε we obtain the following ﬁnal result β(λ. ε)Z λ + ελZ λ + β(λ. at two important conclusions (i) β(λ) is totally determined by the residua of the simple poles in ε of Z λ .33) Since µ dZ λ dλ dZ λ d Zλ = µ = β(λ.e.6. ε) = −ελ − λZ λ µ d Zλ dµ (6. ε)λ dZ λ /dλ = 0 (6. for instance. of course.2 Calculation of renormalization group functions 217 In perturbation theory we can calculate the renormalization constants Z i which relate λ to λB aν (λ) λB = µε Z λ λ = µε 1 + λ (6. ε). One can easily check in particular that amn = 0 for m < n.34) we look for the solution for β(λ. Z λ = Z 3 Z 1 for the λ the series in λ 4 theory. ε) = ν βν εν which is regular at ε = 0 (since our theory is renormalizable a ﬁnite limit of β(λ.32) in (6. ε) in the form of a series β(λ. There will. .35) dλ d daν+1 = β(λ) (λaν ) λ2 dλ dλ We arrive. i.34) Substituting expansion (6. ε) dµ dµ dλ dλ we can rewrite (6. up to 1/ε N in the N th order but their residua are determined by lower order perturbation expansion. In each successive order of perturbation theory only one new divergence arises.33) in the following form β(λ. (ii) The residua of the higher order poles of Z λ are totally determined by a1 (λ).

218 6 Renormalization group In a quite analogous way we also ﬁnd the results for γ (λ) and γm (λ): γ (λ) = − 1 λ 2 where Z3 = 1 + ν (ν) Z3 εν d (1) Z (λ) dλ 3 (6.32) is replaced by αB = µε Z α α with Z α = Z 3 .31) we get Zα = 1 + α 21 + π 3ε α π 2 b2 + · · · .39) 2 16π 2 12 (16π 2 )2 2 1 λ 3 γ (λ) = + O(λ ) 2 12 16π −1 For QED relation (6.37) one obtains the following renormalization group functions in the λ 4 theory: 3λ2 3 β(λ) = + O(λ ) 16π 2 2 λ 1 λ 5 3 − + O(λ ) γm (λ) = (6.3 for the renormalization constants Z i in the λ 4 theory 3λ Z1 = 1 + 16π 2 ε 2 2 λ 2λ λ (6. Using (5. (6.38) Z0 = 1 + − + 16π 2 ε 2(16π 2 )ε (16π 2 )2 ε 2 λ2 Z3 = 1 − 12(16π 2 )ε and with help of (6.35).36) and (6. Using the results of Section 4. B Zm = 1 + ν (ν) Zm εν (1) dZ m dλ (6.37) −1 (Z m = Z 0 Z 3 in the λ 4 theory).36) and γm (λ) = 1 λ 2 where m2 = Z mm2.

3 Fixed points. If t→∞ ¯ ˜ lim λ(t.35). effective coupling constant 219 and therefore.42) it is called a UV stable ﬁxed point. the second order contribution (α/π)2 b2 has been singled out for the sake of the discussion later.40) Though not calculated explicitly. if t→−∞ ¯ ˜ lim λ(t. First we assume (6. from (6. ¯ The solution λ(t.3 Fixed points.43) it is called an IR stable ﬁxed point.11) where β(λ) can be calculated perturbatively from (6. Otherwise the renormalization scale µ could not vary arbitrarily and the theory would need some natural cut-off .11) to be valid in the whole range −∞ < t < ∞. λ) = λ ≡ λ+ (6. new interactions switch on at a mass scale ) is not necessarily uninteresting physically and we shall come back to this point later.41) 6. Note also that β(e) deﬁned as β(e) = µ(d/dµ)e is given by β(e) = 2πβ(α)/e. or small. as follows ¯ from (6. Let λ be the ﬁxed point nearest to λ.λ) λ dx/β(x) = t (6. A zero of β(x) ˜ is called a ﬁxed point. Eq. λ) of (6.26) by the behaviour of the ‘effective’ coupling constant λ(t) as t → ±∞. ˜ In both cases one says that λ is in the domain of attraction of the ﬁxed point λ. (6. for example. Therefore β(e) = e3 4 e5 8b2 + · · · + 16π 2 3 (16π 2 )2 (6. The latter is determined by ¯ λ(t.11) provides a basis for a physically very important classiﬁcation of different theories. λ) = λ ≡ λ− (6.11) must for t → ∞ and t → −∞ approach the zero of β(x) nearest to λ or go to inﬁnity if there is no zero to approach. We shall discuss here only theories with one coupling constant. effective coupling constant Fixed points The main use of the renormalization group is in discussing the large.6. A . A renormalizable theory with perturbative expansion valid only in a certain energy range cut-off by a scale parameter (if.35) β(α) = α α2 + π3 α π 2 b2 + · · · (6. momentum behaviour of a quantum ﬁeld theory which is determined.

. λ) → λ for t → ∞ if β (λ)|λ < 0. if it is a simple zero. has two different theories (or two different phases of one theory). λ is an IR stable ﬁxed point. strictly speaking. for two different values of the coupling λ lying in two different domains one. the only place that we can compute β(λ) in perturbation theory is near λ = 0 and its behaviour for increasing λ is essentially a speculation. In an asymptotically free theory the effective coupling vanishes for inﬁnite momenta. by the sign of the derivative β (λ) at ˜ λ = λ.1. Of course. ˜ ˜ β (λ)|λ > 0.11) the effective coupling ¯ ˜ ˜ ˜ λ(t. In Fig.2 we show some examples of what β(λ) might look like. A theory is said to be asymptotically free if λ = 0 is a UV stable ﬁxed point and to be IR stable if it is an IR stable ﬁxed point. domain is a region which lies between two zeros of β. 6. a speculation. then the λ is a UV ˜ β (λ)|λ > 0.) (IR) stable ﬁxed point for β(λ)/(λ − λ) It is obvious but important to notice that for all single coupling theories the value λ = 0 is a ﬁxed point (β(0) = 0 in perturbation theory). and for t → −∞ in QCD also remains. From (6. The nature of the ﬁxed point is determined.1 it is clear that for ˜ β (λ)|λ < 0. Consequently. The λ 4 theory (with λ ≥ 0) and QED are IR stable (β(λ) is positive for small λ) whereas QCD is asymptotically free (β(λ) is negative for small λ). 6. λ) → λ for t → −∞ if ˜ ˜ is a multiple zero of β(λ). λ is a UV stable ﬁxed point. 6. both for λ < λ and λ > λ. Of course. the behaviour of the ¯ coupling constant λ(t. ˜ ˜ ˜ In other words. (If λ ˜ ˜ M < 0 (> 0). and λ(t. for a given theory. to satisfy (6. All single coupling theories are either asymptotically free or IR stable.220 6 Renormalization group Fig. in general. For theories with several couplings the origin may be UV stable for some couplings and IR stable for others.11) and Fig. of odd order M. λ) for t → ∞ in λ 4 and QED.

It is generally believed (but not proved) that IR slavery in QCD is responsible for conﬁnement. 6. if λ is in the region II then the theory is ¯ not asymptotically free. On the other hand.3 Fixed points.2. instead λ(t. Fig. 6. Finally.2(a) we have distinguished two domains of attraction. λ) must diverge when t → −∞ and β(x) must be such that λ dx/β(x) is divergent. For such a . 6. effective coupling constant 221 Fig. λ) → λ+ when t → ∞.2(d) shows what we hope happens in QCD. 6.2(d) there is no IR stable ﬁxed point. To satisfy (6.11) ¯ λ(−∞.39)).2(c) shows another alternative for QED and Fig. In Fig. suppose that the β-function is always positive (as in Fig.2(b) shows what might happen in QED: IR stability and ﬁxed point λ+ for t → ∞. 6.6. 6. If λ is in region I. In Fig. 6.2(c)) and that it diverges at inﬁnity in such a way that ∞ λ dx/β(x) = tc < ∞ 4 It is likely that this situation occurs in the λ theory (see (6. Fig. If the effective coupling becomes inﬁnite in the IR region one often refers to this situation as IR slavery. the theory is asymptotically free and for t → −∞ (low momentum) λ approaches the ﬁxed point λ− .λ) λ dx/β(x) = −∞ ∞ ¯ λ(t.

We expand α(t) = m=1 (6. µ0 ) (6.46) and β(α(t)) = n=1 α n+1 (t)bn /π n (6.44) For a given c the equation gives upper bounds for the low energy coupling ¯ constants (assuming λ( c . can be transformed into an upper bound for the number of elementary particles in the theory or their masses (see Problem 6. a1 = 1 (6.50) . The theory must be modiﬁed by new interactions which switch on at some mass scale c which provides a natural cut-off or must undergo a phase transition at c . . an (t) = (b1 t/π)n−1 + O(t n−2 ) so that α(t) = α 1 + n=1 (6.48) Comparing the coefﬁcients at the same powers of α and integrating over t one gets a2 (t) = b1 t/π a3 (t) = (b1 t/π)2 + b2 t/π . in turn. .4).222 6 Renormalization group theory the renormalized perturbative expansion cannot be consistent for all energies (unless λ(µ0 ) = 0 for any µ0 ). λ) = ∞) which.45) am (t)α m . Effective coupling constant We will now study the explicit behaviour of the effective coupling constant in perturbation theory.47) Therefore dam (t) m α = dt m=2 n=1 bn am (t)α m α+ πn m=2 n+1 (6.12) order-by-order in α. For the theory to be valid up to c the following equation must be then satisﬁed ¯ λ( c . where α = λ for λ 4 and α = e2 /4π for QED and QCD.49) (b1 tα/π )n + O(t n−1 α n ) (6. With β(α).λ) λ(µ0 ) dx/β(x) = ln( c . given by an expansion β(α) = α[(α/π)b1 + (α/π)2 b2 + (α/π)3 b3 + · · ·] we can calculate α(t) integrating (6.

α(t) = − t→−∞ π b2 1 1 π + 3 2 ln + O 2 b1 t t t b1 t (6. .53) where a1 (t) is called the leading-logarithm-result (6. In the second approximation (b1 = 0. bn = 0 for n > 1) we recover the leading-logarithm-result (6. λ) ¯ m(t.52) Cn = (−1)n+1 Truncating 1/β(x) at a given order in x gives us the corresponding approximate solution for α(t). we can consider the so-called leading logarithm approximation in which we neglect terms O(t n−1 α n ) as compared to the terms O(t n α n ). . In this way one gets the series b1 t = where Bn = bn+1 /b1 and B1 1 0 . Taking 3 2 the limit t → −∞ in (6. . α = 0 is a UV stable ﬁxed point.51).. m. . Another renormalization group parameter is the running mass m(t. bn = 0 for n > 2) we get the following α(t) = α1 (t) 1 − α1 (t) b2 α ln 1 − b1 t π b1 π (6.λ) λ [γm (x)/β(x)] dx (6.6. .16) or (6.51) 1 − (α/π)b1 t The same result can be obtained by expanding 1/β(x) in the integral (6.. 0 B2 · · · Bn+1 B1 · · · Bn 1 B1 · · · Bn−1 .55) . effective coupling constant 223 2 For large t = 1 ln(µ2 /µ2 ) or t = 1 ln( p 2 / p0 ).11) in powers of x and directly integrating over x. . 0 0 1 B1 π Cn π α − + + B1 ln α α(t) α(t) n=1 n α(t) π n − α π n (6.51).26).2) and b2 = 1 (de Rafael & Rosner 1974).54) we see explicitly that the point α = 0 is an IR stable ﬁxed point. In this approximation the effective coupling constant reads α α(t) = . In QED b1 = + 2 (Section 5. .3 Fixed points. In QCD b1 < 0 and b2 < 0. b2 = 0. . In the ﬁrst approximation (b1 = 0. depending on whether we use 0 2 2 (6. α = α(t = 0) (6. therefore.53). λ) = m exp ¯ ¯ λ(t. µ.

57) In the tree approximation the coupling is scheme-independent. For asymptotically free theories λ+ = 0 (and γm (0) = 0) so only the integral is responsible for the large t limit of m(t).224 6 Renormalization group The behaviour of the Green’s functions for rescaled momenta exp(+t) pi is controlled by exp(−t)m(t). We . another one will give a coupling λ λ = F(λ) = λ + O(λ2 ) for λ 4 = λ + O(λ3 ) for trilinear coupling theories (6. for ¯ example. For small enough λ we can assume that dF/dλ = 0 and Fi = 0. mass corrections are suppressed by exp(−t)t p as t → ∞. If λ is the coupling in one scheme. and. for QCD. with p > 0. In particular. 6. in a theory with the UV ﬁxed point λ+ ¯ one gets exp(−t)m(t) = m exp{−t[1 − γm (λ+ )]} exp − ¯ ¯ λ(t) λ [dx/β(x)][γm (λ+ ) − γm (x)] (6.4 Renormalization scheme and gauge dependence of the RGE parameters Renormalization scheme dependence The last problem we would like to discuss in this chapter is the renormalizationscheme dependence within the class of mass-independent renormalization procedures and the gauge dependence of the perturbative results for the functions β. γm and γ . modiﬁcations for λ 4 are trivial). The parameters of one renormalization scheme can be determined in terms of those in another scheme.56) Whether in the high energy limit masses can be neglected or not depends on γm (λ+ ) and on the integral in the exponent. The wave-function renormalization Z 3 and mass renormalization Z m will be given in the new scheme by Z m (λ ) = Z m (λ)Fm (λ). Z 3 (λ ) = Z 3 (λ)F3 (λ) (6.58) where Fi (λ) = 1 + O(λ2 ) (for the sake of deﬁniteness we restrict our discussion to the case of the trilinear coupling theories (QED. QCD).

the ﬁrst two coefﬁcients of the expansion for β are for mass-independent schemes. IR stability) is scheme-independent. (iv) If γ (λ) = γ0 λ2 + O(λ4 ) then γ (λ ) = γ0 λ 2 + O(λ 4 ) and similarly for γm (λ).26) and (6. α(q 2 ) is scheme-independent. We stress that these properties hold only in the class of mass-independent renormalization schemes.ε γm (λ ) = γm (λ) + 1 d ln Fm β(λ) 2 dλ 1 d ln F3 β(λ) 2 dλ ˜ ˜ ˜ Using (6.59) λB. and have physical signiﬁcance.59) the following results can easily be obtained (let λ and λ = F(λ) be ﬁxed points in both schemes) (i) dβ dλ (ii) If β(λ) = −2b0 λ3 − 2b1 λ5 + · · · then β (λ ) = −2b0 λ 3 − 2b1 λ 5 + · · · (6.11).6.61) ˜ λ = dβ dλ ˜ λ (6.e. 0 This is again important since γ0 and γm determine the leading behaviour for t → ±∞ of the exponents in (6. Properties (i) and (ii) are important to ensure that the nature of the ﬁxed points (asymptotic freedom.4 Renormalization scheme and gauge dependence 225 get therefore d β (λ ) = µ dµ γ (λ ) = 1 µ 2 λ = (dF/dλ)β(λ) ln Z 3 (λ ) = γ (λ) + (6. renormalization-scale-independent. ˜ ˜ γm (λ ) = γm (λ) This again must be so since the values of γ and γm at a ﬁxed point determine the scaling behaviour and masses. γi depend on the ratio m/µ and obviously cannot be equal to those in the mass-independent schemes. In mass-dependent schemes the renormalization constants Z i and consequently the coefﬁcients bi . (iii) ˜ ˜ γ (λ ) = γ (λ). In particular within this class of renormalization schemes and in the leading logarithm approximation.ε d dµ λB. .60) i. respectively (scale factors for the ﬁeld and the mass).

α) = αB dB (q 2 ) (6. α) and αB dB (q 2 ) involves uncancelled renormalization constants. α) µν (q) q µ q ν αd(q 2 /µ2 .64) Indeed.63) does not hold in QCD because then the Ward identities do not imply Z 1 = Z 2 . α(q 2 )) Using (5.66) .65) (q 2 /µ2 . where α(m 2 ) is deﬁned in the MS scheme. Since the product α(µ)d(q 2 /µ2 .32) for αinv (q 2 ) = α(q 2 ) q 2 →∞ (6. To this end we consider the photon propagator renormalized in some mass-independent renormalization scheme α D µν (q) = −i g µν − where α = α(µ) and d(q 2 /µ2 . α) = 1− 1 (q 2 /µ2 . α) is then renormalization-scheme-dependent and αB dB (q 2 ) is not ﬁnite.63) and (5. we see that in QED the product αB dB (q 2 ) is ﬁnite. α(µ)) is renormalizationscheme-independent we can change µ2 → q 2 to get αinv (q 2 ) = α(q 2 )d(1. Incidentally. in QCD the relation between αd(q 2 . Thus. α) q µq ν − αai 2 2 q2 q2 (q ) (6.62) The function is the scalar factor of the vacuum polarization tensor deﬁned in (5.64). Using (6. Turning again to QED we observe that α ON = lim αinv (q 2 ) q 2 →0 (6. in the on-shell renormalization scheme ON (q 2 = 0) = 0 and given the scheme independence of αinv deﬁned by (6. µ2 .63) is renormalization-scheme-independent because the bare quantities are. µ2 . α(µ2 )) in the ﬁrst order in α and with µ2 = q 2 one gets (6. αd(q 2 . An important observation is that in QED the quantity αinv (q 2 ) = αd(q 2 /µ2 .26).11) We recall ﬁrst that α(q 2 ) = 0 q 2 →0 (it is an IR ﬁxed point). α) = (α/Z 3 )Z 3 d(q 2 /µ2 . Relation (6.226 6 Renormalization group Effective α in QED It is of some interest to ﬁnd a link between the effective coupling constant in QED and the ‘on-shell’ a ON = 1/137 deﬁned by (12. It remains to ﬁnd the relation between the αinv (q 2 ) and the running α(q 2 ) deﬁned as the solution to (6.32) we also see that α ON = α(m 2 ).63) we arrive at the result (6.86).

for example. It renormalizes the gauge-invariant operator Fµν F µν or. (ii) in the minimal subtraction scheme β(α) is gaugeindependent.67) The inverse relation to ﬁrst order in α ON = αinv (0) reads α(q 2 ) q 2 →0 α ON ln(|q 2 |/m 2 ) 1 = α ON 1 + α ON |q 2 | ln 2 3π m (6.4 Renormalization scheme and gauge dependence 227 This result actually holds to any order in α in the leading approximation because (see Problem 6. in other words. Now we consider the change of both . in general. 2 →0 1 − [α(q 2 )/3π] ln(|q 2 |/m 2 ) 0 q (6. Nevertheless one can prove that: (i) in any renormalization scheme the coefﬁcient b1 in the expansion α α β(α) = α b1 + b2 π π 2 + ··· (6.6.32) as the ﬁnite limit of the expression αinv (q 2 ) = 0 α(q 2 ) = . the transverse part of the propagator. as the value of the three-point function at some value µ2 of the kinematical invariants or according to the minimal subtraction scheme. we can expand α (µ) = α(µ) + O(α 2 ) (6.1) and β(α) is. This follows immediately from the fact that the renormalization constant Z 3 is gauge-independent.70) α = α in the zeroth order but the coefﬁcients of the higher order terms might depend. To prove the ﬁrst theorem we note that if we change the gauge we might change α(µ) to α (µ). both can be deﬁned.6). gauge-dependent.5) (q 2 /µ2 . α(µ2 )) = A1 α(µ2 ) ln(|q 2 |/µ2 ) + n=2 An α n (µ2 ) lnn−1 (|q 2 |/µ2 ) + non-leading terms The value of αinv (q 2 ) in the limit q 2 = 0 can be obtained in terms of α(q 2 ) from knowledge of the ﬁrst terms of the perturbation expansion again using (5.69) is gauge-independent. which is gauge-invariant (see also Problem 6. In non-abelian gauge theories the renormalization programme is more complicated (see Section 8. in general. on the gauge parameter a.68) Gauge dependence of the β-function In QED the β(α)-function is gauge-independent. However.

3 Consider the QED of n h ‘heavy’ fermions h . Thus in the minimal subtraction scheme the renormalized coupling constant α(µ) is gauge-independent and consequently the same is true for the β(α) = µ(d/dµ)α.74) and in particular in this scheme Z α does not contain any constant term C(α. in the one-loop approximation by ‘integrating out’ the heavy fermions from the Green’s functions with light ﬁeld external legs: take the .75) + Zα + α α +O 2 =0 da ε da da ε (1) The only way this equation can be satisﬁed is if dα/da = 0 and dZ α /da = 0. Problems 6. a)/ε + Z α (α. Obtain the effective ﬁeld theory for the light ﬁelds Aµ and l . each with non-vanishing mass M. Inserting the expansion (6. γm in QED in the momentumsubtraction renormalization scheme. Write down the system of coupled RGEs for the gauge coupling g and for the meson self-coupling λ.73) In the minimal subtraction scheme (1) (2) Z α = 1 + Z α (α.aB . and n l ‘light’ fermions l each with zero mass. a) which would destroy the following argument.2 Consider a two-coupling theory: a non-abelian gauge theory with a scalar multiplet transforming according to the representation R of the gauge group. By its deﬁnition it is a ﬁxed.ε ﬁxed = Zα (6.) 6. This proves theorem (i). γ . Therefore d αB da ε ﬁxed = daB d αB = 0 da daB d d α + α Zα da da (6.72) and using the relation αB = µε Z α α we have 0= d (Z α α) da αB . a)/ε 2 + · · · (6. (See also Gross (1976). to the lowest order in perturbation theory.1 Calculate the renormalization group functions β. gauge-independent parameter.228 6 Renormalization group α (µ) and α(µ) under a change of the renormalization point µ to calculate β (α ) and β(α) β (α ) = µ d α dµ αB .74) into (6. and study the effect of the meson selfinteraction on the asymptotic freedom of the theory.73) we get dα 1 dα (1) 1 dZ (1) (6. 6.ε ﬁxed =µ d [α(µ) + O(α 2 )] = β(α) + O(a 3 ) dµ (6. Theorem (ii) can be proved as follows: consider the bare coupling constant αB of the theory.71) because α(µ + dµ) = α(µ) + O(α 2 ).

Problems 229 photon propagator as an example and check the validity of the Appelquist–Carazzone decoupling theorem in the on-shell renormalization scheme (Appelquist & Carazzone 1975) and in the minimal subtraction scheme (Ovrut & Schnitzer 1981).1 (taken now for m = M) has two coupling constants. 6.6 Using the functional integral representation for the bare full two-photon Green’s function in QED. reads (q 2 /µ2 .5 Using the RGE in a mass-independent renormalization scheme show that the vacuum polarization in QED. 6. 4 theory in the Euclidean space (Problem 4. 6.44) and using the lowest order result for the β-function estimate the upper bound for the Higgs particle mass (Dasher & Neuberger 1983). Assume that for consistency the relation m H ≤ c must be satisﬁed. In the latter case understand the role of a ﬁnite renormalization of the coupling constant in absorbing the effect of the heavy mass.4 Consider the standard electroweak theory based on the gauge group SU (2) × U (1) with the gauge symmetry spontaneously broken (see Chapter 12) by one complex scalar doublet. 6. Study the RGE for the Green’s functions of the effective theory. 6.3). α(µ2 )) = A1 α(µ2 ) ln(|q 2 |/µ2 ) + n=2 An α n (µ2 ) lnn−1 (|q 2 |/µ2 ) + non-leading terms where the coefﬁcients An are determined in terms of the b1 and b2 of the β-function. What symmetry does it possess when two couplings are equal? Discuss the ﬁxed point structure for these couplings. Calculate the two-loop .26). (5. Adopting the point of view expressed by (6.7 The model discussed in Problem 4.8 Consider λ β-function. show by taking the derivative with respect to the gauge parameter a that the transverse part of the propagator is gauge-independent.

1 Scale invariance Scale transformations An important concept in quantum ﬁeld theory is that of scale invariance. t). Scale invariance is the invariance under the group of scale transformations. Its lagrangian does not contain dimensionful constants and should be invariant under the transformation (7.7 Scale invariance and operator product expansion 7. † (7.5) (y. Among other things. Obviously. t)} = δ(x − y) and the scale dimension d is then deﬁned so that the commutation relations (7. t)] = iδ(x − y) ˙ { (x.4) (7. For free ﬁelds one has the canonical equal-time commutation relations [ϕ(x. ϕ(y.4) U (ε) (x)U −1 (ε) = T −1 (ε) (eε x) (7. it is closely related to the famous Bjorken scaling in the deep inelastic lepton–hadron scattering.1) local ﬁelds (x) (in this section we denote by (x) any scalar ﬁeld ϕ(x) or spinor ﬁeld (x)) will be. t). Scale transformation for the coordinates is deﬁned as follows x → x = exp(ε)x (7.4) 230 .3) where the constant d is called the scale dimension of the ﬁeld (x). in general.2).2) where T (ε) is a ﬁnite-dimensional representation of the group (for classical ﬁelds (x ) = T (ε) (x)). Let us consider ﬁrst a free massless ﬁeld theory. all such transformations form an abelian group. We assume it to be fully reducible and therefore we can write T (ε) = exp (−d ε) (7.1) where ε is a real number. If the conﬁguration variables are scaled down as in (7. subject to a unitary transformation U (ε) (compare with Section 1.

Multiplying relations (7.2) one sees that invariance of (7.4) and (7.8) and neglecting. scale transformation is a transformation of dynamical variables (the ﬁelds) but not of the dimensionful parameters such as masses since we want to . for instance.5) for bare ﬁelds.2). Take.5) implies that these so-called canonical scale dimensions are dϕ = 1 for scalar ﬁelds and d = 3 for spinor ﬁelds.8) + =ϕ. d = δL = U LU −1 − L = L(exp(dφ ε)φ(x )) − L(φ(x)) = it is easy to calculate (see (1. if added to (7.4) and (7. Therefore the corresponding action integral is invariant under scale transformations δI = d4 x δL(x) = 0 (7.9) (one can also integrate explicitly using (7. = ε(4 + xµ (∂/∂ xµ ))L(x) where we have used the relation δ ∂ ∂ xν (x) = ∂ ∂ δφ(x) = ε d + 1 + x · (∂/∂ xν ) (x) ∂ xν ∂x This result is expected on general grounds since in a lagrangian which does not contain dimensionful constants each term has a canonical dimension equal to its ordinary dimension which is four. would.7) Using (7.2) (of course. break the scale invariance of the lagrangian as their scale dimension is two.5) remain invariant under the scale transformation.6) and the values dϕ = 1.29)) ∂L δ ∂ ∂L ∂ δ ∂(∂ /∂ xµ ) ∂ xµ (7. Notice the difference between scale dimension and ordinary dimension which is four for the term m 2 2 .7. as usual. strictly speaking. of course. it is the action integral which is scale-invariant).2) is as follows: U (ε) (x)U −1 (ε) = (x) + ε(d + xµ (∂/∂ xµ )) (x) + O(ε2 ) (7. In an inﬁnitesimal form the scale transformation (7.7).4) and (7. one can easily check that a lagrangian which does not contain dimensionful constants is invariant under the scale transformation (7.5) by operators U and U −1 and using (7. the following lagrangian density L=i¯/ ∂ + 1 (∂µ ϕ)(∂ µ ϕ) + g ¯ γ5 ϕ − (λ/4!)ϕ 4 2 3 2 (7. Terms like m 2 2 (x). the surface terms at inﬁnity).6) For interacting ﬁeld theories one still has the canonical commutation relations (7.1 Scale invariance 231 and (7. We see that the canonical 2 scale dimensions of ﬁelds coincide with their ordinary dimensions deﬁned on purely dimensional grounds (see Section 4. Deﬁning scale dimensions for bare ﬁelds as canonical ones.

one gets U (ω)U (ε) = U (ε)U (ω) and [D.12) From (7.14) . in general. (x)] = (d + xµ (∂/∂ xµ )) (x) (7. Turning back to an exactly scale-invariant world we can introduce the generator D of the scale transformation U (ε) = exp(iε D) (7. Terms like m 2 2 (x) break scale invariance explicitly. Mµν ] = 0 (7.13) (7. However.6) we get the following differential representation of the generator D i[D.11) and 1 µν U (ω) = exp(− 2 iω Mµν ) By comparing two sequences of dilatation and translation (x) → exp(d ε) (exp(ε)x) → exp(d ε) (exp(ε)x + aµ ) ε aµ and (x) → (x + aµ ) → exp(d ε) (exp(ε)(x + aµ )) aµ ε we conclude that U (ε)U (a) = U (exp(ε)a)U (ε) or i[D. This is due to the necessity of renormalization which introduces a dimensionful parameter. it is already worth recognizing at this point that in quantum ﬁeld theory described by a scale-invariant lagrangian the scale invariance is. also broken.10) and study its commutation relations with generators of the Poincar´ group (for e conventions see Problem 7.1) U (a) = exp(ia µ Pµ ) (7.232 7 Scale invariance and OPE stay in the framework of the same physical theory. Pµ ] = Pµ Similarly.

) (7.7. ∂µ µν =0 and Pµ = Mµν = d3 x can 0µ = 0ν d3 x − xν 0µ 0µ ) d3 x (xµ . Let us consider ﬁrst a scalar ﬁeld theory and deﬁne µν as follows: µν = can µν − 1 (∂µ ∂ν − gµν )ϕ 2 6 (7. The dilatation current can be written in a more compact way if we introduce a modiﬁed energy–momentum tensor µν .1 Scale invariance 233 Dilatation current The dilatation generator D can be written in terms of the Noether dilatation current Dµ (x). The conserved dilatation current can be deduced by comparing an inﬁnitesimal change. The current is conserved in the presence of the symmetry. but it may. obtained without the use of the Euler–Lagrange equations of motion.16) = µ ∂ ∂xν (x) − gµν L is the canonical energy–momentum tensor. under the scale transformation of the lagrangian density of a scale-invariant theory (7. The Noether dilatation current is therefore the following Dµ (x) = or Dµ (x) = where can µν µd µ (d + xν (∂/∂ xν )) (x) − xµ L (x) + x ν can µν (7. be deﬁned even in the absence of scale invariance.15) where µ is the conjugate momentum. of course.17) The new tensor has the following properties µν = νµ .8) δL = ε 4 + xµ ∂ ∂ (xµ L) L=ε ∂ xµ ∂ xµ with an alternate expression for δL which does use Euler–Lagrange equations δL = (∂/∂ xµ )( µδ ) = (ϕ.

16) (d = 1) reads µν Dµ = x ν µν − 1 ∂ν (xµ ∂ ν − x ν ∂µ )ϕ 2 6 The last term is the divergence of an antisymmetric tensor.e.20a) in the context of searching for an energy–momentum tensor whose matrix elements between physical states are ﬁnite. in general. (7. must be separately veriﬁed (see. Collins & Duncan (1977) for the complete discussion. 1970): µν = Tµν − 1 (∂µ ∂ν − gµν )ϕ 2 6 (7.20) for a scale-invariant scalar ﬁeld theory (and µ µ = m2ϕ2 (7.19) An important property of the new tensor is that its trace vanishes µ µ =0 (7.46) to the µ µ from the anomalous scale invariance breaking.20a) if a mass term is added) while the trace of can contains singular derivative terms. Coleman & Jackiw (1970) when looking for an object with the ‘soft’ divergences (7. so it does not contribute to ∂ µ Dµ and to D.18) (7. Coleman & Jackiw of the ﬁniteness of matrix elements of the µν is incomplete because it ignores the non-soft contribution (7. In a renormalizable ﬁeld theory the ﬁniteness of matrix elements of the T-ordered products of local currents is an additional requirement. Thus we redeﬁne Dµ (x) = x ν and ∂ µ Dµ (x) = µ µ µν (x) (7. µν The new tensor was invented by Callan. µ is a legitimate energy–momentum tensor of our theory. the proof of Callan.21) where Tµν = µ ∂ν − gµν L + 1 ∂ λ ( 2 λ µν − µ λν − ν λµ ) (7. Eq. Using the relation = ∂µ ϕ it is easy to check that in terms of µν the Dµ of (7. see Adler. Section 10.17) can be generalized to include ﬁelds with non-zero spin (Callan et al.234 7 Scale invariance and OPE i. for instance. Actually.22) . which.1).

7. angle-preserving) transformation must leave invariant the ratio dx α dyα |dx| |dy| So it must have the property that ds 2 = gαβ dx α dx β = gαβ (∂ x α /∂ x γ )(∂ x β /∂ x δ ) dx γ dx δ (7. where λµ λν Mλµν = x µ Tcan − x ν Tcan + and M µν = we get µν νµ Tcan − Tcan + ∂λ ( λ µν ) = 0 d3 x M0µν Eq.21) has the same properties as the tensor (7. The spin matrix µν is deﬁned by the transformation properties of the ﬁelds under an inﬁnitesimal Lorentz transformation δ = − 1 (xµ ∂ν 2 − xν ∂µ + µν )ωµν (7. which in addition to translations. µν = − 1 iσµν = 1 [γµ .23) and reads.17) deﬁned for scalar ﬁelds only. A conformal (i. Lorentz transformations and scale transformation also contains special conformal transformations. for example. γν ] for 2 4 spin one-half ﬁelds. µν = 0 for spin 0 ﬁelds. Conformal transformations It turns out that many scale-invariant theories are invariant under a larger symmetry group of transformations. For inﬁnitesimal transformations x µ = x µ + εµ (x) λ µν † It can be constructed by noting that from conservation of the angular momentum current ∂λ Mλµν = 0.e. (7.24) = f (x) gαβ dx α dx β ≡ f (x) ds 2 where f (x) is a certain scalar function. the so-called conformal group.22) follows if we require ∂µ T µν = ∂ν T µν = 0. In particular its trace is a measure of scale invariance breaking. . In terms of Tµν the generators of the Lorentz transformation are Mµν = d3 x (xµ T0ν − xν T0µ ) The tensor µν deﬁned by (7.1 Scale invariance 235 is the conventional symmetric energy–momentum tensor† and where all ﬁelds of the theory are assembled into a vector whereas the scalar ﬁelds are denoted by ϕ.

1.2.27) we ﬁnd explicitly the following four generators of the special conformal transformations Kα = d 3 x ( j α )0 = d3 x (x 2 α0 − 2x α xν ν0 ) (7. It can be checked that such a µν exists in all renormalizable ﬁeld theories of ﬁelds of spin ≤ 1. ε. . However.27) are conserved for any εν = εν (x. The general solution for εµ (x) satisfying our constraints reads εµ (x) = a µ + εx µ + ωµν xν + (−2b · x x µ + bµ x 2 ). = e+ε x µ . Conformal invariance implies scale invariance since the commutator of an inﬁnitesimal conformal transformation and an inﬁnitesimal translation contains a scale transformation. a α . † The case n = 2. ωαβ .26) and (7. ωµν = −ωνµ It includes (listed are the ﬁnite transformations) translations scale transformation Lorentz transformations special conformal transformations x x x x µ µ µ µ (7. x µ + bµ x 2 = .1). bα ) satisfying (7. and only if. This equation gives (n − 2)∂α ∂β h(x) = 0 ∂ µ ∂µ εα (x) = 1 (n − 2)∂α h(x) 2 where n is the dimension of the space-time. see Problem 7.25) (check it) and indeed scale invariance goes together with the special conformal symmetry. We see that for n > 2. In this case the currents ( j ε )µ (x) = µν (x)εν (x) (7.25) (7. From (7. Scale invariance implies conformal invariance if.26) = x µ + aµ. the opposite is not true and examples of ﬁeld theories which are scale-invariant but not conformally invariant can be constructed.28) For the transformation of ﬁelds under special conformal transformations see Problem 7. consequently.† h(x) is at most linear in x and. = µν (ω)xν .236 7 Scale invariance and OPE one therefore obtains the following condition ∂ν εµ (x) + ∂µ εν (x) = h(x)gµν where h(x) = f (x) − 1. requires a separate discussion. there exists in the considered theory a symmetric traceless conserved energy–momentum tensor so that ∂ µ Dµ = µ µ = 0. εµ (x) is at most quadratic in x. 1 − 2b · x + b2 x 2 h(x) = 0 h(x) = 2ε h(x) = 0 h(x) = 4b · x which in n dimensions form together the (n + 1 + n(n − 1)/2 + n)-parameter conformal group (see Problem 7. interesting for statistical mechanics and string theories.

31) means that the state exp(−iε D)| p is a quantum of the same ﬁeld as the state | p but with a rescaled momentum (if the vacuum was not unique then the state exp(−iε D)| p would belong to a different Hilbert space than the state | p and our ﬁnal conclusion would be avoided) and therefore. but the ground state is asymmetric.e. Here the dynamical equations are completely symmetric. exponentiating. Indeed.2 Broken scale invariance General discussion Scale invariance cannot be an exact symmetry of the real world.30) i.g. We believe this idea to be realized for chiral symmetry in strong interactions.29) Acting on a single particle state | p with four-momentum p(P 2 | p = p 2 | p ) one gets P 2 exp(−iε D)| p = exp(2ε) p 2 exp(−iε D)| p (7. it follows from commutation relation (7. exp(iε D)P 2 exp(−iε D) = exp(2ε)P 2 (7. We will learn in the next chapters that there are several ways to account for the symmetry breaking. the state exp(−iε D)| p is an eigenstate of P 2 with eigenvalue exp(2ε) p 2 . Another concept is that of spontaneous symmetry breaking. if scale invariance is not spontaneously broken) exp(−iε D)|0 = |0 then we conclude that exp(−iε D)| p = exp(−iε D)a † ( p)|0 = exp(da )a † (eε p)|0 (7.30). all particles must be massless or the mass spectrum must be continuous. by (7. One assumes that there are terms in the lagrangian (e. Result (7. the Noether currents associated with the symmetry are conserved and the Ward identities are still valid. The simplest idea for symmetry breaking is that of ‘approximate’ symmetries. Finally there is a third way: anomalous breaking of symmetries.7. If we assume in addition that the vacuum is unique under scale transformations (i. If it were. mass terms in the case of scale invariance) which violate the symmetry. Whenever a classical theory possesses a symmetry. P 2 ] = −2iP 2 or.12) that [D.31) where a † ( p) is the creation operator for the considered particle with momentum p and da is its dimension. but that they are ‘small’.e. but there is no way of quantizing the theory so as to preserve that .2 Broken scale invariance 237 7. all particles would have to be massless or their mass spectra continuous.

. . δ (xi ) .32) δ(xi − y) 0|T (x1 ) . (xn )|0 (7. . . in reality the scale invariance may be spontaneously broken but simultaneously the corresponding Goldstone bosons (dilatons) get masses due to anomalies. . By integrating with respect to variable y the l. .s. we say that there are anomalies in the conservation equations for the symmetry currents. Equivalently. . . . .33) µ = i nd + x1 ∂ µ ∂ µ + · · · + xn µ ∂ x1 ∂ xn . . (xn )|0 = 0|T ∂ yµ −i i µ µ (y) (x1 ) . .14). In quantum ﬁeld theory. quantum ﬁeld theoretical calculations must be regularized to avoid inﬁnities and anomalies occur when there does not exist a regularization procedure which respects the classical symmetry and the corresponding Ward identity. the anomalous scale invariance breaking is quite expected. (xn )|0 where δ s are given by (7.11) ∂ 0|T Dµ (y) (x1 ) . More technically. For a discussion of the spontaneous breaking of scale invariance we refer the reader to the literature. an unavoidable renormalization procedure necessarily introduces a scale at which the theory is renormalized and this breaks scale invariance. In QCD the effects of the anomalous explicit breaking of scale symmetry are dominant. δ (xi ) . (xn )|0 0|T (x1 ) . integrating with respect to y is equivalent to taking the Fourier transform and letting k → 0. and there is no remaining consequence of the original classical scale symmetry. even with a scale-invariant lagrangian. . Of course.h. . In the rest of this section we shall study the ‘anomalous’ response of the theory to scale transformations by comparing the naive Ward identities with the renormalization group approach. vanishes: it gives surface terms which can be neglected if there are no zero mass particles coupled to µ µ . (xn )|0 (7.238 7 Scale invariance and OPE symmetry. . Scale invariance requires that there be no dimensionful parameters whereas the regularization of the quantum theory is effected by introducing a dimensionful cut-off or dimensionful coupling constants in the dimensional regularization procedure. . (xn )|0 = i i 0|T (x1 ) . Anomalous breaking of scale invariance Ward identities for the dilatation current (trace identities) can be obtained from the general formula (10. a given symmetry may be broken by several of these mechanisms. simultaneously. We then obtain d4 y 0|T µ µ (y) (x1 ) .

39) . .37) one can derive analogous equations for the 1PI Green’s functions (n) ( p1 .36) as follows (t = ln ρ) n−1 − pi [−∂/∂t + D]G (n) (exp(t) p1 . .2 Broken scale invariance 239 In the last step of (7. . .38) where D = 4 − nd is the dimension of the 1PI Green’s function (n) . dxn exp i i = pi xi 0|T (x1 ) . (x2 )]|0 ∂ x1 ∂ x2 0 0 0 + x2 δ(x1 − x2 ) 0|[ (x2 ). . . . for a scale-invariant lagrangian. . . exp(t) pi ) (7. pn−1 )(G (n) ∼ M D ). pn−1 ) = −iG (n) (0. pn−1 ) and (n) (0. for n = 2 we have µ 0|T (x1 )x2 ∂ µ ∂ µ (x 2 )|0 + 0|T x 1 µ (x 1 ) (x 2 )|0 ∂ x2 ∂ x1 µ ∂ µ ∂ 0 0 0 = x1 µ + x2 µ 0|T (x1 ) (x2 )|0 + x1 δ(x1 − x2 ) 0|[ (x1 ). (7. we have only n − 1 four-dimensional integrations in (7. (xn )|0 we get from (7. . .e. . pn−1 ) ∂ pi i=1 (7. .35) and analogously for G (n) (k. exp(t) pn−1 ) (7.38) are the Ward identities following from the Noether relation ∂µ D µ = µ µ and from the canonical ﬁeld transformation rule (7. . . for µ µ = 0. . . cancel in pairs. .33) the following Ward identity ∂ + n(d − 4) + 4 G (n) ( p1 . . . . . p1 . .14). . . . . pn−1 ) [−∂/∂t + D] (n) (exp(t) pi ) = −i (n) (0. . p1 . .33) we have brought the derivatives outside the T -product which results in further equal-time commutator terms which. .37) From (7. pn−1 ) dx1 . For example. (x1 )]|0 (7. i.34) Deﬁning the momentum-dependent Green’s functions n (2π)4 δ 1 pi G (n) ( p1 . .38) reads (n) (ρpi ) = ρ D (n) ( pi ) (7. . .7. . . (xn )|0 (7. . . . .35). if we parametrize all momenta by the scaling factor pi = ρ pi so that i pi ∂/∂ pi = ˆ ρ∂/∂ρ we can rewrite (7. however. pn−1 ) as the Fourier transform of the 0|T µ µ (y) (x1 ) . . exp(t) pn−1 ) = −iG (n) (0. . effectively. We observe that for d = d can the nd − 4n + 4 is just the dimension D of the Green’s function G (n) ( p1 .36) Summation over n − 1 momenta and the additional term 4 in the square bracket are due to the fact that.37) and (7. In addition. the solution to (7. In particular. p1 . . exp(t) p1 . Eq.

scale-invariant.e.1 only for β(λ) = γ (λ) = 0.24). λ(t).39) to (7.39) with D → D − nγ (λ+ ). Imagine now a theory with a UV ﬁxed point λ+ . namely solution (7. µ) exp −n ¯ 0 t ¯ γ (λ(t )) dt (7.24) (n) (exp(t) pi . do we recover the canonical scaling. for a non-interacting theory. m. The asymptotic behaviour of . Comparing (6.240 7 Scale invariance and OPE In this case the Green’s functions exhibit canonical scaling. or both. As mentioned in Section 6. i.41a) = (1/ρ)m exp ¯ λ(t) λ Consider ﬁrst a massless m = 0 theory with a conserved Noether dilatation current ∂µ D µ = 0.24) with the Ward identity (7. µ) = exp(Dt) (n) ¯ ( pi . this happens when λ+ is a simple zero of β(x) : β(λ+ ) = 0. The exponential factor in (7. we have already mentioned that there is no reason to expect these identities to be valid in quantum ﬁeld theory since the scale invariance is always broken by a dimensionful cut-off. m(t). β (λ+ ) < 0 (we assume that γ (x) is differentiable).42) If the integral deﬁning ε(t) is convergent.40) where ¯ λ(t) λ dx/β(x) = t (7.41) and t m(t) = (1/ρ)m exp ¯ 0 ¯ γm (λ(t )) dt dx [γm (x)/β(x)] (7. then ε(t) = O(1/t) and the theory is asymptotically.38) we see that the previously mentioned anomalous breaking of canonical scale invariance occurs in quantum ﬁeld theory.38). For easy reference we recall the solution to (6. for t → ∞. Either the dilatation current has an anomalous divergence. Apart from a trivial case when λ is exactly equal to λ+ and the RGE can then be solved directly and gives (7. However.40) can then be written as t exp −n 0 ¯ γ (λ(t )) dt = ρ −nγ (λ+ )+ε(t) where ε(t) = −(1/t) ¯ λ(t) λ dx n[γ (x) − γ (λ+ )]/β(x) (7. The response of a quantum ﬁeld theory to the scale transformation is given by the RGE (6. or the ﬁeld transforms anomalously. λ.

The mass effects in general are summarized in the solutions (7. The identity (for bosons. λ+ . Duncan & Joglekar 1977. λ.42).46) is rather lengthy (Adler. µ) ∼ ρ D−nγ (λ+ ) (n) ( pi .47) implies that the operation m∂/∂m with external momenta and the coupling constant λ ﬁxed (we assume that there is only one mass and one dimensionless coupling constant) is equivalent to the insertion of a new vertex −im 2 2 at zero-momentum transfer. with µ µ = m 2 2 . λ (n) (ρpi .41) to the RGE.44) ( pi . µ) (7. Thus we again see that for β = γm = γ = 0 (6. to be speciﬁc) m ∂ i i i = 2 (−2im 2 ) 2 2 − m2 2 ∂m p p −m p − m2 (7. µ)[1 + O(ρ −|β (λ+ )| )] Asymptotic scale invariance. µ) is obtained by expanding λ(t) about λ+ . λ. In this case β(x) = −bx 3 + O(x 5 ) and for most of the interesting operators γ (x) − γ (0) = 2cx 2 + O(x 4 ). µ) and of the exponential factor (7. More speciﬁcally one gets ( pi . Collins & Duncan 1977. We now extend our discussion to theories with scale invariance explicitly broken by mass terms. possibly with anomalous dimensions.7.2 Broken scale invariance (n) 241 ¯ (exp(t) pi .43) The function γ (λ+ ) is called the anomalous dimension of the ﬁeld. λ.24) and (7.43) arise from O(λ(t) − λ+ ) terms in the expansion of (n) ¯ (t). So the integral (7.42) is logarithmically divergent and it causes logarithmic deviations from the asymptotic ¯ scale invariance (7. is not a feature of asymptotically free (λ+ = 0) gauge theories.37) are identical. The leading term is (n) (exp(t) pi .40) and (7.45) We also note an operator equation for the anomalous divergence of the dilatation current in gauge theories reﬂecting the anomalous scale invariance breaking µ µ µν = (β(λ)/2λ3 )G a G a µν (7. Nielsen 1977). The lead¯ ing corrections to (7. in particular. µ) = ρ D−nγ (λ+ ) exp − × (n) λ+ λ dx [γ (x) − γ (λ+ )]n/β(x) (7.40) is controlled by the asymptotic behaviour of . The ρ → ∞ limit of (7. Collins. λ+ .43) (remember that λ2 (t) ∼ 1/2bt) ¯ λ(t) exp λ 2c dx b x ∼ (2bλ2 t)−c/b (7.46) Rigorous proof of relation (7. We see again that only for λ = λ+ = 0 do we recover exact scale invariance.

To introduce this concept let us consider any dimensionless observable quantity A(Q) which can depend only on one dimensionful variable Q. Eq.49) is automatically satisﬁed if the theory is asymptotically free because the relevant anomalous dimension γm vanishes. y and which is dimensionless and whose deﬁnition does not involve any massive constants must be a function of the ratio x/y only.3 Dimensional transmutation The phenomenon of dimensional transmutation is closely related to scale invariance broken by the renormalization of a quantum ﬁeld theory. So if it does not depend on y it must be a constant. As we know from our experience with quantum ﬁeld theories the conclusion (7. dimensional analysis tells us that. gB ). 7.48) −→ mρ γm (λ+ )−1 × {logs of ρ} Now. (7. the predictions of the theory cannot be expressed directly in terms of gB because . for example. Otherwise the mass insertion would effectively correspond to a hard operator and would inﬂuence the asymptotic behaviour of the Green’s functions.242 7 Scale invariance and OPE ¯ both λ(t) and m(t) (we recall that t = ln ρ) ¯ m(ρ) = mρ γm (λ+ )−1 exp ¯ ρ→∞ ¯ λ λ dx [γm (x) − γm (λ+ )]/β(x) (7. for example. for the mass effects to be asymptotically non-leading. for example.50) is. it is essential that m(ρ) −→ 0 or ¯ ρ→∞ γm (λ+ ) < 1 m Introducing the scale dimension d d we get m m of the mass operator m=m¯ m = 2m2 = 3 + γm (λ+ ) 2 + 2γm (λ+ ) d <4 m (7.49) as the condition for ‘softly’ broken scale invariance (Wilson 1969a). Moreover. The reason why pure dimensional analysis breaks down in quantum ﬁeld theory is this: our quantity A actually also depends on the dimensionless free parameter(s) of the theory. bare coupling constant(s): A = A(Q. a function f (x.50) Indeed. If we assume our theory to be described by a scale-invariant lagrangian (no dimensionful parameters) then from dimensional analysis we must conclude that A(Q) = const. y) which depends on two massive variables x. in general. (7. wrong. A(Q) = Q 2 · σtot .

53a) which tells us that given the strong coupling constant α(Q) at some Q the theory predicts the conﬁnement scale .4 Operator product expansion (OPE) 243 the theory requires renormalization.51) where Q = exp(t)Q 0 . in the limit of . 7.51).53) can be also rewritten in the following form /Q = exp{−π/[|b1 |α(Q)]} (7.4 Operator product expansion (OPE) Short distance expansion The OPE technique originates in studies of the relevance of scale invariance in quantum ﬁeld theory (Wilson 1969a). be re-established. An important illustration of these general considerations is provided by the behaviour of the running coupling constant α(Q 2 ) in QCD (we forget here about a subtlety: α(Q 2 ) is not really an observable. Eq. it depends on the deﬁnition of the renormalization scheme). The physics is independent of the details of the renormalization prescription but not of the necessity of renormalization which produces an effective scale. If we choose Q 0 so that F(α(Q 0 )) = 0 and call it then α(Q) = F −1 [ln(Q/ )] and in the leading logarithm approximation α(Q) = −π/[b1 ln(Q/ )] (7. α = g 2 /4π and β(α) = (b1 /π)a 2 + O(α 3 ). One-parameter ambiguity for α(Q 2 ) is reﬂected in the fact that the theory speciﬁes the ﬁrst derivative of α(Q 2 ) rather than α(Q 2 ) itself dα/dt = β(α) (7. at least up to logarithmic corrections.51) we get ln(Q/Q 0 ) = F(α(Q)) − F(α(Q 0 )) where F(x) = dx/β(x) Thus. In the previous section we have discussed how mass terms and renormalizable interactions break this invariance which may. however. is the conﬁnement scale: α( ) = ∞. (7. Integrating (7. deﬁned by the condition F(α( )) = 0. Dimensional transmutation is a process which exploits the above to introduce dimensionful parameters into the predictions of the theory.7.52) . Dimensionless α(Q) and dimensionful can be traded for each other.53) (7. an effective scale Q 0 appears as a constant of integration of the RGE (7.

The expansion is valid in the weak sense.4) and in any order of perturbation theory for interacting ﬁelds (Zimmermann 1970).56) where dn . Here we shall introduce the technique of the OPE. According to Wilson’s hypothesis a product of two local operators A(x) and B(y). to vanishing separation for the operators in the matrix elements.e. Relation (7.58) Asymptotic scale invariance at short distance implies that relation (7. commuting the dilatation generator D with (7. On (y) by expanding On ( 1 (x + y)) in a Taylor series about y. A(x)B(y) − n C n (x − y)On ((x + y)/2) = 0 AB (7. A and B. may be written as† A(x)B(y) = n C n (x − y)On ( 1 (x + y)) AB 2 (7.57) C n (x − y) = (dn − d A − d B )C n (x − y) AB ∂(x − y) AB and therefore C n is homogeneous of degree dn − d A − d B in its argument AB C n (x) ∼ x dn −d A −d B AB (7.58) holds at least in the limit x → 0. In QCD asymptotic scale invariance is broken by logarithmic corrections (see Section 7. In position space this corresponds to short distances. OPE and scale invariance arguments allow us to determine the short distance singularity structure of operator products. Assuming scale invariance to be an approximate symmetry at short distance one gets the leading behaviour of the Cs for small values of their arguments.2) so relation (7. when xµ is near to yµ . in terms of the scale dimensions of the operators.54) we get D. Since the operators On can be chosen to be independent we deduce that ∂ (x − y) (7. The physical importance of studying operator products at short distance will be discussed later. d A and d B are scale dimensions of the operators On . hopefully up to at most logarithmic corrections. say. Indeed.58) is modiﬁed but only by logarithmic terms. AB i.e. respectively. when sandwiched between physical states.55) and in an exactly scale-invariant theory dn − d A − d B − (x − y) n ∂ C n (x − y)On ( 1 (x + y)) = 0 2 ∂(x − y) AB (7. i.244 7 Scale invariance and OPE momenta going to inﬁnity.54) where {On (x)} is a complete set of hermitean local normal-ordered operators and {C n (x)} is a set of c-number functions. We see that the most singular contribution to A(x)B(y) † The product also can be written in terms of. 2 .54) is true in free-ﬁeld theory (see Problem 7.

Thus for T -products we conclude that scalar factors in Cs depend on x 2 − iε.4) T J (x)J (0) = −2 x →0 ˙ 2 2 F (x. To do this let us refer to free scalar ﬁeld theory. (0)]|0 = ε(x0 )2 Im(i F ) = −(1/2π )ε(x0 )δ(x 2 ) (7. Notice that : (x) (0): may be expanded in a Taylor series since it has no singularity. B(0)] we then get (1/(x ±iε) = P(1/x)∓iπ δ(x))† x2 1 1 1 − 2 = 2π iδ(x 2 )ε(x0 ) = ε(x0 )2i Im 2 − iεx0 x + iεx0 x − iε = −i 0|[ (x).7. the anomalous dimensions vanish (see Section 10. Since F = −i 0|T (x) (0)|0 = T A(x)B(0) = A(x)B(0) (x0 ) + B(0)A(x) (−x0 ) and −x 2 1 1 1 (x0 ) + (−x0 ) = − 2 2 − iεx + iεx0 −x x − iε 0 (7. However. m )I F (x. i. broken by operators with scale dimensions d < 4.e. simple products. x 2 ). m 2 + 4i + 4i F (x.2 that the scale dimensions d A .61) for simple products we have the following prescription C n (x. equal to the canonical dimensions of the corresponding operators. For instance. We have learned in Section 7.1). x 2 − AB AB iεx0 ).60) 4π x − iε x 2 →0 is the free particle propagator in position space. Anomalous dimensions (see (7. For commutators [A(x). commutators etc. The Lorentz transformation properties of the Wilson coefﬁcients C n (x) require AB them to be polynomials in the components of x times scalar functions of x 2 : C n = C n (x.4 Operator product expansion (OPE) 245 as x → y is given by the operator On having the lowest dimension. We have to determine prescriptions for handling light-cone AB AB singularities of the coefﬁcients C n for different types of products: time-ordered AB products. m F (x. for several interesting operators like conserved currents or currents of softly broken symmetries. m 2 2 ) : (x) (0): + : )J (0) 2 (x) 2 (0): = −2 + 4i )xµ : (0)(∂ µ )(0): + · · · (7.62) † Note in particular that: . d B and dn are not.43)) can be calculated in perturbation theory. for the T -product of two currents J (x) = : (x) (x): in free scalar ﬁeld theory we have (see Problem 7. x 2 ) ≡ C n (x.59) where 1 1 d4 k exp(−ikx) 2 4 (2π ) k − m 2 + iε i 1 = − 2 2 (7. m )I 2 2 F (x. in general.

is consistent with causality. This is interesting because the appearance in the commutator of the so-called Schwinger terms (proportional to derivatives of the δ-functions) is linked to the singularity structure of operator products at short distances. B(0)] vanishes for space-like separation x 2 < 0: the prescription x 2 → x 2 − iεx0 . x)(ρ(0) + x · ∇ρ(0) + · · ·) (7. Using the expansion of the form [A(x). x 2 − iεx0 ) − C n (x. The . x)x etc.65) where n 0 FAB (x 0 ) n 1 F AB (x 0 ) = =− ˜ AB d3 x C n (x0 . the difference (−x 2 + iεx0 )− p − (−x 2 − iεx0 )− p appearing in a commutator [A(x). Following Wilson (1969a) we can see that ˜ AB the coefﬁcients C n are equivalent to a sum of δ-functions n ˜ AB C n = 0 FAB (x0 )δ (3) (x) + 1 Fn (x0 ) · ∇ δ (3) (x) + · · · AB (7.246 7 Scale invariance and OPE and by differentiating both sides with respect to x 2 1 2 + iεx −x 0 where ε(x) = 1 −1 for x > 0 for x < 0 n − 1 2 − iεx −x 0 n =− 2πi δ (n−1) (x 2 )ε(x0 ) (n − 1)! (7.63) In a general case in which the light-cone singularity (7. To get (7.66) where ρ(x) is a differentiable trial function and the Taylor expansion is justiﬁed ˜ AB for small x0 because C n vanishes for |x| ≥ |x0 |. x 2 + iεx0 ) AB AB and letting x0 → 0 one can study the equal-time commutators.66) and from the deﬁnitions of the δ-function and its derivatives. x)ρ(x) = ˜ AB d3 x C n (x0 .65) follows immediately from (7. (7.64) where ˜ AB C n = C n (x.65) one considers an integral ˜ AB d3 x C n (x0 . Eq. B(0)] = n ˜ AB C n On (0) (7.63) is (x 2 )− p with some non-integer power p. by placing the cut along the positive real axis. x) ˜ AB d3 x C n (x0 .

0).67) when ˜ AB C n ∼ x −2 pn The equal-time commutator is obtained in the limit x0 → 0. (∂µ ∂ν )(0) etc.4 Operator product expansion (OPE) 247 n dependence of the functions i FAB (x0 ) on x0 is determined by dimensional analysis to be n i FAB (x 0 ) ∼ x0 −2 pn +3+i (7. the Lorentz tensor operators (∂µ )(0).54) which are ﬁnite in the limit xµ → 0 may be singular in the limit x 2 → 0 with xµ ﬁnite. Some terms may vanish due to rotational invariance. Ji (0)] (but not for two J0 s): [J0 (x. Let us look in more detail at the expansion of a product of two Lorentz scalar operators. (7. It is obvious. We will see that the vacuum expectation value of this term is related to the e+ e− annihilation cross-section calculated under the assumption that the leading short distance singularities are those of the free-ﬁeld theory. however. two currents in free scalar ﬁeld theory.4) Jµ (x)Jv (0) ∼ x 2 gµν − 2xµ xν 1 + O(x −3 ) l π 4 (x 2 − iεx0 )4 (7.69) (7. Thus for p > 1. Therefore we must expect an inﬁnity of terms to contribute to the leading singularity on the light-cone. Light-cone expansion The OPE can be extended to the light-cone region as well. that terms in (7.70) where Ci j is quadratically divergent.59).68) The ﬁrst term gives the familiar Schwinger term in the commutator [J0 (x). Ji (0)] ∼ Ci j ∂ j δ (3) (x) (7. As mentioned before. cannot be neglected on the light-cone and the general structure of the expansion is as . for instance.7. As an example let us take the product of two electromagnetic currents Jµ (x) = : ¯ (x)γµ (x): In free fermion ﬁeld theory one has (see Problem 7.5 the coefﬁcient of the δ-function is inﬁnite and for p > 2 derivatives of the δ-functions appear in the commutator. in contrast to the situation at x = 0 where the leading behaviour is given by one term. Such terms are usually not revealed by direct calculation of the commutator of two operators based on equal-time canonical commutation relations between ﬁelds and their conjugate momenta.

The leading singularity on the light-cone comes from operators with the lowest twist.. . . We see that the behaviour of the expansion for small x 2 is determined by scalar functions of x 2 . the C n (x 2 ) in the AB neighbourhood of x = 0.µ j (0) AB n follows: A( 1 x)B(− 1 x) = 2 2 (7.71).. one usually takes the free-ﬁeld theory classiﬁcation of operators as a guide in writing down the OPE. ∂ µ j ↔ ↔ ↔ and ↔ ↔ ¯ γµ ∂ µ1 . The scalar ﬁeld and the fermion ﬁeld as well as their derivatives have twist one. .248 7 Scale invariance and OPE n C n (x 2 )x µ1 . In a similar way we can obtain the light-cone expansion for two electromagnetic currents.. As an exercise let us expand ∗ − x 2 x 2 = 1 ← ← x µ1 x µ2 x µ1 + ∂ µ1 ∂ µ2 − ··· 2 2! 2 2 x ν1 → 1 x ν 1 x ν2 → → × 1+ ∂ ν1 + ∂ν ∂ν + · · · (0) 2 2! 2 2 1 2 1 x µ1 x µ2 x µn ∗ ↔ ↔ ↔ = (0)∂ µ1 ∂ µ2 · · · ∂ µn (0) ··· 2 2 n n! 2 ∗ (0) 1 − ∂ µ1 ← (7.74) into (7.73) The difference (dimension of an operator – its spin) is called twist.74) Inserting (7. This should not lead to any confusion.72) (7.59) we get the leading terms in the light-cone expansion of the product of two scalar currents in free-ﬁeld theory.) As long as there are no extra divergences introduced by the summation over n. Nevertheless. ∂ µ j When dimensions are modiﬁed by the interactions it is not obvious that they are correlated with the spin in this way. .µ j to be symmetric traceless tensors with j Lorentz indices.. since operating on a ﬁeld with a derivative raises the spin and the dimension by one unit simultaneously. (Note that we use the same notation for complete Wilson coefﬁcients in (7. they are irreducible tensors of spin j. In free-ﬁeld theory there exist inﬁnite series of operators with ﬁxed twist. Examples of twist two operators are ∗ ∂ µ1 ∂ µ2 . x µ j Oµ1 . . . short distance behaviour and light-cone behaviour are connected C n (x 2 ) = (x 2 )(dn − jn −d A −d B )/2 + higher orders in x 2 AB so that the degree of singularity on the light-cone is determined by τn = dn − jn (7.54) and for their scalar part in (7. .71) n We have taken the bases Oµ1 .

1) the amplitude is Tn = e2 v( p+ )γµ u( p− )(1/q 2 ) n|Jµ (0)|0 ¯ (7. (7.79) Eq. In the one-photon exchange approximation (see Fig.5 The relevance of the light-cone Electron–positron annihilation We brieﬂy discuss now the relevance of the light-cone for high energy phenomena. 7.76) by the term −(2π)4 δ(q + pn ) 0|Jν (0)|n n|Jµ (0)|0 . we observe ﬁrst that Wµν (q) can be written in terms of the current commutator Wµν (q) = d4 x exp(iq x) 0|[Jµ (x).78) To see what happens when q 2 → ∞.5 The relevance of the light-cone 249 Fig. It is a process widely investigated experimentally and interesting in many respects as a laboratory for important discoveries in particle physics. Thus the total cross section is determined by the tensor Wµν (q) = n (2π)4 δ(q − pn ) 0|Jµ (0)|n n|Jν (0)|0 d4 x exp(iq x) 0|Jµ (x)Jν (0)|0 (7. Jν (0)]|0 (7. 7.1.77) It can be checked that the total annihilation cross section for unpolarized initial leptons reads σ (q 2 ) = 8π 2 α 2 ρ(q 2 )/q 2 (7.75) where Jµ is the hadronic electromagnetic current. Consider ﬁrst the process of the electron-positron annihilation into hadrons.76) = Jµ is conserved and therefore one may write Wµν (q) = ρ(q 2 )(q 2 gµν − qµ qν ) (7. We focus only on the total cross section which is of the order of a few tens of nanobarns and is comparable to the cross section for the e+ e− → µ+ µ− .7.79) differs from (7. 7.

P is the four-momentum of the hadronic target (usually a proton). Since q 2 > 0 we may choose a frame in which q has only a time component and use the fact that the commutator vanishes for x 2 < 0. It is shown in Fig. 0 which is zero because q0 and pn are positive. 7. k are lepton four-momenta.250 7 Scale invariance and OPE Fig.70).83) This result is not surprising: a free-ﬁeld theory is asymptotically scale-invariant and the cross section must behave like 1/q 2 on dimensional grounds. 7. E.69) which is responsible for the Schwinger term (7.63) and the relation d4 x exp(ikx)δ (n) (x 2 )ε(x0 ) = 22−2n 2 2 n−1 (k 2 )ε(k0 ) π i(k ) (n − 1)! (7. Using (7. Jν (0)]|0 (7.2. Then in this frame Wµν (q) = ∞ −∞ dx0 exp[i(q 2 )1/2 x0 ] |x|<x0 d3 x 0|[Jµ (x).81) we easily ﬁnd (for a single fermion with charge ±1) Wµ µ (q) = (1/2π )q 2 q 2 →∞ (7.82) and consequently σ (q 2 ) → 4π α 2 /3q 2 (7. Deep inelastic hadron leptoproduction Let us now consider another famous process: the production of hadrons in lepton–hadron collisions with very large momentum transfer.80) Since the dominant contribution to the integral over x0 comes from regions with the least rapid oscillations this Fourier transform is determined by the singularity of the commutator at x = 0. E and θ are the lepton energies and lepton scattering angle in the laboratory . In a free-ﬁeld theory the leading singularity of the commutator originates from the term (7.2 which also contains some deﬁnitions of kinematical variables: k.

7. The passage from product to commutator can be justiﬁed in a way similar to that for the annihilation reaction.85) (7.87) then it follows from the spectral representations (see also (7. the physical region for the process is therefore 0 < xBJ < 1 (7. We collect together some properties of the tensor Wµν (P.62)) that Wµν (q 2 . Jν (0)]|P (7. q3 ) . Let us choose a frame in which P = (m. Note that only for 0 < xBJ < 1 is the Wµν (q 2 .88) where Disc Tµν is the discontinuity of the Tµν along its branch cuts. P · q) = i d4 x exp(iq x) P|T Jµ (x)Jν (0)|P (7. Now we are ready to discuss the relevance of the light-cone in our process. q = (q0 . If we introduce the forward Compton scattering amplitude for photons with mass q 2 . 0. respectively. As we shall see it is also very useful to deﬁne the so-called Bjorken variable xBJ = −q 2 /2P · q For elastic scattering s = m 2 ⇒ 2P · q + q 2 = 0 ⇒ xBJ = 1 In terms of the Bjorken variable. Tµν (q 2 . 0. for ﬁxed q 2 .86) = = where Jµ is the hadron electromagnetic current. q). 0. 0. P · q) = 2 Disc Tµν (q 2 . As in case of the e+ e− annihilation into hadrons in the one-photon exchange approximation the cross section for this process is determined by the tensor Wµν (P.84) To be speciﬁc we consider the process of electroproduction. q) = n (2π)4 δ(q + P − pn ) P|Jµ (0)|n n|Jν (0)|P d4 x exp(iq x) P|Jµ (x)Jν (0)|P d4 x exp(iq x) P|[Jµ (x). which are important for further discussion. They correspond to a cut in xBJ for |xBJ | < 1. P · q) = 2 Im Tµν (7. P · q) given by the deep inelastic cross section. The amplitude Tµν has normal threshold cuts in s = (P + q)2 and u = (P − q)2 for m 2 < s < ∞ and m 2 < u < ∞.5 The relevance of the light-cone 251 frame. 0).

.90) q 2 → −∞. in the region with the least oscillations.91) in the Bjorken limit. . The idea of studying the limit (7. x+ ﬁnite. J (0)]|P (7. P · q) = d4 x exp(iq x) P|[J (x).252 7 Scale invariance and OPE and introduce the so-called light-cone variables x± = x0 ± x3 .59)) the leading singularities of the operator products on the light-cone are given by the twist two operators P|[J (x).e. The general form of the matrix n × P| (0)∂µ1 . In the limit (7.92) where we have used (7.89) d2 x⊥ P|[Jµ (x). q) = × dx− exp(iq+ x− ) 2 x⊥ ≤x+ x− dx+ exp(iq− x+ ) (7.90) is due to Bjorken (1969) and it has turned out to be one of the most creative ideas in modern elementary particle physics.90) the behaviour of Wµν is determined by the behaviour of the integrand for x− → 0. We see that in the limit q+ → ∞ q− ﬁxed (7. . J (0)]|P = 1 2πiδ(x 2 )ε(x0 ) π2 1 xµ . . xµn n! 1 (7.74). x− → 0 for x+ ﬁnite corresponds 2 to x → 0 (but not to xµ → 0) and Wµν is determined by the singularity of the current commutator on the light-cone. 2 Then Wµν (P. Since the integrand is 2 non-vanishing only for x⊥ < x+ x− (causality). Jν (0)]|P where the limit on the x⊥ -integration comes from the vanishing of the commutator 2 for x 2 = x+ x− − x⊥ < 0. For the realistic case of particles with spin. ν → ∞ and xBJ is ﬁxed (xBJ = −q 2 /2ν). x2 ) ν = q0 m q± = 1 (q0 ± q3 ) = (ν/2m)[1 ± (1 − m 2 q 2 /ν 2 )1/2 ]. Spin complications are not essential for general orientation. i. x⊥ = (x1 . Apart from the c-number contribution which is irrelevant for the considered matrix element (ﬁrst term in (7. twist two operators are again the leading ones. For simplicity let us study the scalar current analogue of the electroproduction tensor W (q 2 . ∂µn (0)|P + less singular terms .63) and (7.

So we have 2i P|[J (x). Pµn + · · · (7.− 2 2ν m (7. . P · q). .92) is P|Oµ1 .92) they give additional factors x 2 and therefore are less singular. Thus only the ξ+ is relevant and in terms of the Bjorken variable our result reads νW (q 2 .94) π where 1 (x · P)n An f (x P) = n n! Deﬁning the Fourier transform f (x P) = (1/2π ) and using the identity† i we get W (q 2 .95) dξ exp(iξ x · P) f˜(ξ ) The roots of the argument of the δ-function are ξ± = − ν ν q 2m2 ± 2 1− 2 m2 m ν 1/2 −→ BJ 2ν −q 2 . As has already been explained. .93) Terms having at least one Kronecker delta may be ignored because when contracted with the xµ s in (7.7. (0)]|0 = − = 1 d4 k exp(ikx)δ(k 2 − m 2 )ε(k0 ) (2π )3 i δ(x 2 )ε(x0 ) 2π . being the imaginary part of the amplitude for forward off-mass-shell Compton scattering. P · q) = 2 f˜(xBJ ) (7..62): 0|[ (x). P · q) = 4 dξ δ((ξ P + q)2 )ε(q0 + ξ P0 ) f˜(ξ ) d4 x exp(ikx)δ(x 2 )ε(x0 ) = (2π )2 ε(k0 )δ(k 2 ) (7.µn |P = An Pµ1 .5 The relevance of the light-cone 253 element in (7. is non-vanishing for |−q 2 /2ν| < 1.95) follows from the spectral representation for the free-ﬁeld commutator (Bjorken & Drell 1965) and the relation (7. .96) where ‘BJ’ means Bjorken limit. J (0)]|P = δ(x 2 )ε(x0 ) f (x P) + less singular terms (7.. W (q 2 . Pµn + δµ1 µ2 Pµ3 .97) In the free-ﬁeld theory with twist two operators giving the leading singularity on † Identity (7.

The standard procedure is as follows. we obtain the famous Bjorken scaling for the structure function W (q 2 . . Therefore ˜k ˜k in the case of canonical scaling Cn (q 2 ) = Cn are constants. An important step in this direction is to establish some relation between the Wilson coefﬁcients and measurable quantities which is not limited to the canonical scaling case.254 7 Scale invariance and OPE the light-cone. Remembering that it is an analytic function of xBJ in the complex xBJ -plane with a branch cut along the real axis for .92) let us consider T J ( 1 x)J (− 1 x) ≈ 2 2 x 2 →0 µ k Cn (x 2 )xµ1 . Consider ﬁrst the forward scattering amplitude for scalar currents T (q 2 . . x µn Cn (x 2 ) (7. Using (7. P · q) = 1 −q 2 −n ˜ k xBJ Cn (q 2 )Ak n n. Wilson coefﬁcients and moments of the structure function Use of the OPE is by no means limited to the free-ﬁeld theory case. P · q) = i d4 x exp(iq x) P|T J ( 1 x)J (− 1 x)|P 2 2 (7.. . P · q) if C we integrate T (q 2 .93) and deﬁning 2n q µ1 .91) is the discontinuity of T .µn n.. for ﬁxed q 2 .k (7. see (7.99) where the sum is taken over all twist two operators Ok and all even spins n. . The coefﬁcients k ˜ n (q 2 ) can be related to the moments of the structure function W (q 2 . Instead of (7. beyond that contained in the hypothesis of canonical dimensions.k (7. .92). An important ﬁnal observation is that only even spin operators are allowed in the expansion (7. P · q) in the Bjorken limit can be written as T (q 2 . Its validity for renormalized operators has been proved in perturbation theory and therefore it can be used to obtain further information about the scaling limit of the structure functions. q µn ˜ k 2 Cn (q ) = i (−q 2 )n+1 k d4 x exp(iq x)x µ1 .98) The structure function (7. Indeed W is real (it is a cross section) so f˜(ξ ) must be real and therefore f (x P) must be even in x.101) ˜k Note that Cn (q 2 ) are deﬁned so that they are dimensionless for twist two operators in the expansion of the product of two currents with dimension d J = 3.100) the leading contribution to T (q 2 . P · q). xBJ ) over xBJ . . xµn Ok 1 .88).

0 < x < 1 (for odd m the integral (7.104) for even m in terms of the integral over the physical region in the electroproduction process.104) vanishes). x) = C 1 4π 1 −1 dx x m−1 W (q 2 . This may not be the case for low moments: for ﬁxed q 2 the limit x → 0 corresponds to ν → ∞.3. g)Ak (µ2 .104) we have written down explicitly the dependence on the renormalization point and the coupling constant. x) = −W (q 2 . P · q) we have (x ≡ xBJ ) 2 1 2π i dx x m−1 T (q 2 . −x) (7. x) = k ˜k Cm (q 2 .3. 7. |xBJ | < 1 we integrate over the contour shown in Fig. .103) dx x m νW (q 2 . where ν = P · q ≈ s and s is the cms energy squared in the forward Compton amplitude. Thus we get a relationship between moments of the deep inelastic structure function and the coefﬁcient functions.5 The relevance of the light-cone 255 Fig.105) we can rewrite (7. x) ∼ x −1 . µ2 .7. x) = C 1 −q 2 ˜k Cm (q 2 )Ak m k (7. At high energy the imaginary part of this amplitude is expected to behave as s α .102) and on the other hand 1 2πi Finally we get Mm (q 2 ) = (1/2π) 1 −1 dx x m−1 T (q 2 . Using the crossing properties of the structure function under q → −q or x → −x W (q 2 . g) (7. Since the discontinuity of the T is just 1 W (q 2 .104) is meaningful for values of m such that the integral converges. so that W (q 2 .104) m In (7. x) (7. 7. We also note that (7. α ≈ 1.

256 7 Scale invariance and OPE An important virtue of the OPE is that it allows us to factorize out the q 2 dependence in quantities like Mm which depend on q 2 and P 2 = m 2 .108) The generating functional formalism and perturbation theory rules can be extended to such cases if we introduce sources (x) coupled to operators O(x). xn .106) Interactions can modify this behaviour. The counterterms present in the lagrangian for the renormalization of the Green’s functions involving only elementary ﬁelds are not sufﬁcient for eliminating divergences in the Green’s . in general. Feynman rules for vertices involving elementary ﬁelds and the composite operators O(x) can be derived in exactly the same way as in Chapter 2 or Chapter 3.109) generates these new Green’s functions. 7. . . This dependence can then be studied by means of the RGE. In the next section we shall study the q 2 dependence of the Wilson coefﬁcients using RGE. p) n (7. The new Green’s functions calculated perturbatively. The functional W [J. P → ρ P relates Mn s for different q 2 s and m 2 s. . . x) = 0|T (x1 ) . x) O (7. (xn )O(x)|0 O or in momentum space ˜ (n) (2π)4 δ( p1 + · · · + pn + p)G O ( p1 .107) = d x exp(i px) i=1 4 d4 xi exp(i pi xi )G (n) (x1 . . . As we know. xn . . . Connected Green’s functions are obtained from the functional Z and the Legendre transformation performed on sources J (only) generates 1PI Green’s functions with O insertions. . Similar analysis is not possible directly for Mn (q 2 . pn . exact Bjorken scaling gives ˜k Cn (q 2 ) = const.6 Renormalization group and OPE Renormalization of local composite operators We are used by now to working with Green’s functions involving not only elementary ﬁelds but also composite operators. . . require renormalization. for example. . P 2 = m 2 ) because the scale transformation q → ρq. . As usual one has to take derivatives with respect to and then set = 0. ] = D exp iS + i d4 x (J + O) (7. G (n) (x1 . . (7.

110) and (7. Of course.112) is the new counterterm. (7. .113) can where dO is the canonical dimension of O. As an example let us consider λ 4 theory and discuss a single insertion of O(x) = 1 2 (x) (Itzykson & Zuber 1980) 2 or an insertion of O(x) = ¯ (x) (x) in QCD.2) DO of O differs from D of (n) and reads can DO = D + (dO − 4) (7. 0) = 1 and (2) ¯ (0.110) then the 1PI npoint Green’s functions with a single O insertion satisfy (n) O. The superﬁcial degree of divergence (n) (see Section 4. 0) = 1 (7. .115) Simple multiplicative renormalization. 7. for example.114) in agreement with the lowest order.B ( p1 . µ.109). λB . Additional operator renormalization constants are needed OB (x) = Z O OR (x) (7. λ. If an operator O is multiplicatively renormalized as in (7. p. We have taken O(x) to be of the ith order in the elementary ﬁelds (x). The divergent Green’s functions with this insertion are (2) and (2) with DO = 0. ( 2 B )B = Z 2( 2 B )R =Z 2 Z 3( 2 R )R (7.R ( p1 .1).110) If we want to express OR in terms of the renormalized ﬁelds the wave-function renormalization constants will explicitly appear. . The last term in (7. −1 pn . it may not eliminate divergences from some Green’s functions involving more then one insertion of O. It is an easy exercise to calculate 2 ¯ Z O in these simple cases in the lowest order of perturbation theory.111) Only in exceptional cases like conserved currents or currents of softly broken symmetries does Z O = 1 (see Section 10. we may impose (2) 2 (0. .6 Renormalization group and OPE 257 functions with composite operator insertions.111) correspond to a replacement (x)OB ( i B (x)) = OR ( i R) + (Z O Z 3 − 1)OR ( i/2 i R) (7. (7.110). p. A well-known example is the vacuum . where the vertex ⊗ is in both cases just one and for simplicity we can take the momenta of the external legs to be equal. . .4. new renormalization constants require new renormalization conditions.112) in the generating functional. For instance. m B ) (7. is actually not always sufﬁcient. pn . The Z O can be calculated in perturbation theory by calculating the divergent Green’s functions with operator insertions. For instance. . m) = Z O Z 3 n/2 (n) O. The relevant Feynman diagrams are shown in Fig. Relations (7. .7.

1).117) Note that the matrix Z i j is not symmetric: Z i j = 0 if dim Oi < dim O j . Thus. From (7.118) d −1 (Z O )ik (Z O )k j dµ .R ( p1 . expectation value of two electromagnetic currents Jµ (x) which requires a subtraction although Z J = 1 (see Section 10.4. (7. Using the arguments of Section 4. . for instance. −1 pn . (7. p. λB . λ(µ). Therefore if there exist several operators with the same quantum numbers as the operator O and with their canonical dimensions lower or equal to the dimension of the operator O. .116) ∂ ∂ ∂ ij + β(λ) + γm − nγ + D δi j + γO ∂t ∂λ ∂m (n) t O j (e pi . . Another complication arises for the following reason. . they will mix with O under the renormalization.9) and (6.2 concerning the necessary counterterms in the standard renormalization programme we can convince ourselves that an operator in the lagrangian with dimension d requires. m) = (Z O )nk Z 3 n/2 (n) Ok . − where γO = − µ ij (n) On . µ. p. .B ( p1 . .24).110) takes a matrix form OB = (Z O )nk Ok n and correspondingly pn . . in general.258 7 Scale invariance and OPE Fig. λ. m) = 0 (7. If we limit our discussion to Green’s functions involving only one insertion of an operator O the same is true for the renormalization of such composite operators. . µ. 7. m B ) (7. all possible counterterms with dimensions ≤d allowed by the symmetries of the bare lagrangian.117) we can derive for Green’s functions with an operator insertion the RGE analogous to (6.

the renormalization constants Z 0 and Z 1 . µ) (n) Oi (0. in fact. γm interpreted now as the anomalous dimension of the operator 2 (7.24).1): LB 4 → LB 4 + n n B and we can renormalize the new coupling constants: Cn = Z O Cn . In the opposite direction.1). the renormalization of local composite operators discussed here can be phrased in the language of Chapter 4. putting Cn = 1. pn . for example. We can interpret the operators On as a perturbation of the lagrangian (4. (4. . p1 .47)) and we arrive to the same RG equations (6. p1 . The derivatives m 2 ∂/∂m 2 and λ∂/∂λ count then the number of such insertions (see also (7. . µ) = i ˜ CiAB (q.49).9) and (6. Let us recall that in general A( 1 x)B(− 1 x) ≈ 2 2 x 2 →0 CiAB (x. . with. . and of the composite operators. for example. At this point it is worthwhile to observe that in perturbation theory there is no fundamental difference between the renormalization of the original lagrangian. g(µ). g(µ). the n-particle Green’s function is. µ) (7. a superposition of Green’s functions with the vertex and mass operator insertions (we can treat the mass as one of the parameters in perturbation theory). µ)Oi (0) i (7. we get. (7.120) By applying the differential operator D=− ∂ ∂ + β(g) ∂t ∂g . . According to (2.89). deﬁned in Chapter 4 as renormalization of the lagrangian parameters m B and λB . .24) and. .118). . For instance. pn . We then derive (n) the RGE for the On by a complete analogy with the derivation of (6. B d4 x Cn OB n RGE for Wilson coefﬁcients We now turn back to the OPE and study the q 2 dependence of the Wilson coefﬁcient functions by means of the RGE.9) and (6.6 Renormalization group and OPE 259 is the anomalous dimension matrix for the operators Ok and D is the dimension of the Green’s function (equal to the canonical dimension of all operators involved) and the mass dependence has been neglected.7. could as well be deﬁned as the renormalization constants for the operators 2 and 4 . Both can be formulated in the same language. g(µ).119) and therefore for any Green’s function with insertion of the product AB we have (n) AB (q.

µ). g(µ). Firstly take Ci dimensionless. In asymptotically free theories the UV ﬁxed point g+ = 0. This is in particular trivially true for the electromagnetic currents. In addition put γ A = γ B = 0 as for conserved currents.26). Putting q 2 = µ2 . β(x) = (2/16π 2 )b1 x 3 + · · · and γi (x) = (4/16π 2 )γi0 x 2 + · · ·. t = 1 ln(q 2 /µ2 ) and expanding 2 ˜ ˜ Ci (µ. µ)[g 2 (t)/g 2 (µ2 )]−γi /b1 (7. g(t). 0. µ) = exp[−γi (g+ )t] exp − g g(t) ˜ dx [γi (x) − γi (g+ )]/β(x) (7.260 7 Scale invariance and OPE to (7. µ) = Ci (µ.122) ˜ ¯ × Ci (q. Eq. γ B and γO are their anomalous dimensions.53). The solution to the RGE for the Wilson coefﬁcient is analogous to (6. For a non-zero UV ﬁxed point g+ = 0 (7. i.122) gives a power-like violation of scaling ˜ ˜ Ci (q. µ) = Ci (µ.124) 2 In the last step we have used (7. γi (g) and β(g) in perturbation theory. For the no-mixing case ij γO = δ i j γi we then have ˜ Ci (et q.124) is valid for both q0 and q 2 large as 2 compared to . 0.e. µ) ≈ (q 2 /µ2 )−γi (g+ ) Ci (µ. g(t).120) with all momenta rescaled by exp(t) and using (7. g(µ). We have assumed here that operators A and B do not mix with other operators when undergoing renormalization.121) where D A . µ) ˜ ¯ To proceed further one has to calculate Ci (q. g(µ).99)) Mi (q 2 ) g 2 (q 2 ) ¯ = 2 2 2 Mi (q0 ) g (q0 ) ¯ −γi0 /b1 ln(q 2 / = 2 ln(q0 / ) 2) 2 +γi0 /b1 (7. Let us write it down ˜ for some speciﬁc cases. D A + D B − DO = 0. µ) (7.123) Thus for the moments of the structure function we have (from (7. g+ . D B and DO are canonical dimensions of A. g(t).118) we get − ∂ ∂ ij ˜ + β(g) + D A + D B − DO + γ A + γ B δi j − γO CiAB (et q. µ)[1 + Ci1 g 2 (t)/4π + · · ·] ¯ ¯ in the leading order in g 2 we get 0 ˜ ˜ ¯ Ci (q. (7. if there is only one twist two operator in expansion (7.125) . µ) ∂t ∂g =0 (7.e.104) when k = 1. i. B and O respectively and ij γ A . g(µ).

goes beyond perturbation theory. is deﬁned as follows ˆ ˆ g Tg exp dx g ¯ γ (x) ˆ =1+ β(x) 1 + 2! g dx g ¯ g γ (x) ˆ β(x) x dx g ¯ g ¯ dy ˆ γ (x) γ (y) ˆ + ··· β(x) β(y) (7. Nevertheless we assume the validity AB of the OPE and. . that the coefﬁcient functions Cn are calculable in QCD perturbation theory.6 Renormalization group and OPE 261 In the case of operator mixing the solution to the RGE can be written in the matrix form (take g+ = 0) ˆ C(q.7. very suggestive as the physical sense of the OPE formalism. . when the strong interactions are described by QCD. which is discussed in this section. necessary since [γ (x1 ). Vainshtein & Zakharov (1979) and extensively discussed in the recent years. The same idea underlies the use of the OPE to account for the non-perturbative structure of the QCD vacuum. Cn γ = [γi j ] ˆ and the Tg ordering. µ) = Tg exp where g(µ2 ) g(q 2 ) ¯ dx γ (x)/β(x) ˆ ˆ C(µ. It is also in agreement with the possibility. One can then expect that to a good approximation the coefﬁcient functions are calculable in perturbation theory while the non-perturbative effects are accounted for by the matrix elements of the operators On . Non-perturbative effects manifest themselves through non-trivial vacuum expectation values of various operators which vanish in perturbation theory 0| :On : |0 = 0 if On = I . the matrix elements of the operators taken between hadronic states contain non-perturbative effects responsible for the binding of quarks and gluons into hadrons.126) C1 ˆ . g(µ).127) OPE beyond perturbation theory Strictly speaking. pioneered by Shifman. This approach is implicitly based on the parton model assumption discussed in Section 8. µ) ¯ (7. g(q 2 ). that the contribution from short distances can be separated from that of large distances. γ (x2 )] = 0. C = . moreover. the use of the OPE to study the moments of the structure function in deep inelastic lepton–hadron scattering.4. This is because in the realistic case.

Lϕ = L(ϕ) . 1975). . Let us consider a renormalizable action containing some light ﬁeld(s) ϕ with mass m and some heavy ﬁeld(s) with mass M (such that m M) S(ϕ. we assume for simplicity that the interaction lagrangian Lϕ is ˜ linear in . ) + i Dϕ D exp i ik k! dx J (x)ϕ(x) dx Lϕ (x) + J (x)ϕ(x) dx L (x) . (7. The Green’s functions for processes involving only light particles are generated by the functional (recall (2.5). ) = dx Lϕ (x) + L (x) + Lϕ (x) (7. This use of the OPE has been successful phenomenologically but its theoretical status and possible limitations are still subject to some discussion (see. Important for this programme is the theorem that the coefﬁcient functions in the OPE manifestly reﬂect the symmetries of the lagrangian whether or not they are symmetries of the vacuum (Bernard et al.e.129) dy1 . (The derivatives acting on can be removed . . 7. . (1984) and references therein). i.84)) W [J ] = N =N × ∞ k=0 Dϕ D exp iS(ϕ. dyk Lϕ (y1 ) .262 7 Scale invariance and OPE The best-known examples are quark and gluon condensates 0| : ¯ : |0 and µν 0| :G a G a : |0 µν the ﬁrst one is responsible for the spontaneous breaking of chiral symmetry in QCD (Chapter 9). Novikov et al. AB The coefﬁcient functions Cn (q) can be found by calculating the appropriate Feynman diagrams (Problem 7. . for instance.7 OPE and effective ﬁeld theories OPE is a convenient tool for discussing effective ﬁeld theories.128) We are interested in calculating amplitudes involving only the light particles ϕ with external momenta much smaller than M. Lϕ (yk ) exp i From now on.

the OPE has to be performed. belonging to a complete basis of such monomials {Ok (x)}. if A or B involve derivatives.105)–(2. which is due to the features described below (2. . . . . it corresponds only to OPE for T -products. . . . we can write W [J ] = N × × =N × ∞ k=0 ∞ k=0 Dϕ exp i ik k! dx Lϕ (x) + J (x)ϕ(x) ˜ ˜ dy1 . L(yk )G (y1 .54). . .130a) ˜ ˜ dy1 . The important point is that the effective region of integration over y involves distances much shorter than the region of integration over x. dyk L(y1 ) .7. Let A(x) and B(x) be some monomials in the ﬁelds ϕ(x) and their derivatives. yk ) We have thus arrived at the effective theory described by a non-local action. Then there exists a set of Wilson coefﬁcients {C k (x)} AB such that† Dϕ A(x)B(y) exp iS[ϕ] + i = k dx J (x)ϕ(x) x+y 2 exp iS(ϕ) + i dx J (x)ϕ(x) (7. . .h. dyk L(y1 ) . . dyk L(y1 ) .131) C k (x − y) AB Dϕ Ok † Strictly speaking.131) is not in a one-to-one correspondence with that of (7. All the local ﬁeld products here are tacitly assumed to involve counterterms which in the operator language would correspond to their normal ordering (see (2. the series in the r.s. Moreover. of (7. it is perfectly equivalent to the the underlying theory (7. . .) In such a case.128). . Now.54) can be formulated in the path integral language as follows. L(yk )G conn (y1 . (2. .130). . (yk ) exp i dx L (x) (7.130) D Dϕ exp i ik k! dx Lϕ (x) + J (x)ϕ(x) ˜ ˜ dy1 . .107)).131) and (2. . . As far as the processes involving only external ϕ particles are concerned. .7 OPEs and effective ﬁeld theories 263 by integration by parts. The OPE hypothesis (7.56). yk ) Comparison with (2. .137) allows us to ‘exponentiate’ with simultaneous replacement of ordinary Green’s functions by the connected ones W [J ] = N + ∞ k=0 Dϕ exp i ik k! dx Lϕ (x) + J (x)ϕ(x) (7. L(yk ) (y1 ) . .

. one arrives at the conclusion that it is possible to replace the products ˜ ˜ L(y1 ) . . i.130a) by sums of the form ˜ ˜ L(y1 ) . It remains to show that Lϕ (which may contain M independent terms. In a similar way. . . M n(l) n(l) ≥ 1.s.131) to each term on the r. Lϕ stands for terms which would have n(l) ≤ 0.133): in (7. This latter conclusion would be rather obvious in the operator language. L(yk ) −→ l Cl(k) (y1 . and applying (7.136) The factors M n(l) have been extracted above just to keep Cl dimensionless. .132) Finally.h. (7. one can show that if A(x)B(y) corresponds k to a series C k Ok and C(x)D(y) corresponds to a series CC D Ok . which can be interpreted as additional independent coupling constants for the effective interaction vertices Ol .133) in spite of the fact that they stand under the exponent. the coefﬁcients Cl are constants. logarithms and positive powers of large mass M) can be absorbed into Lϕ by . . .264 7 Scale invariance and OPE One can prove the validity of a similar expansion for multiple products like A(x)B(y)C(z) by appropriate differentiations of (7. . then AB A(x)B(y)C(z)D(w) corresponds to the product C k (x − y)Ok AB k x+y 2 n CC D (z − w)On n z+w 2 (7. . .135) ik k! dy1 . L(yn ) in (7. . . dyk Cl(k) (y1 . . However. . Therefore. The results of this integration are independent of ymean . one should stress the important difference with respect to the original (true) Wilson coefﬁcients deﬁned by (7. we integrate the coefﬁcients Cl(k) with the connected Green’s functions keeping all the ymean s ﬁxed. In this way we obtain W [J ] = N with Seff given by Seff = and Cl = M n(l) k Dϕ exp iSeff + i dx [ Lϕ (x) + J (x)ϕ(x)] (7. yk )Ol (ymean ) (7. yk )G conn (y1 . . .e. .134) dx Lϕ (x) + l 1 Cl Ol (x) .136) the latter are integrated with the Green’s functions for heavy particles.131) with respect to J (z) and z (according to the structure of C(z)). performing only k − 1 integrations for each k-point function. which follows from translational invariance. yk )δ(ymean ) (7. The coefﬁcients Cl are customarily also called Wilson coefﬁcients. In the next step. .

128). An important point omitted in the above discussion is checking whether the integrals present in (7. The action Seff in (7.136) are not divergent in spite of their proper behaviour at M → ∞. but not for ϕ (see Weinberg (1980)). The parameters (couplings) Cl undergo renormalization and also evolve according to their RGEs which do not depend on the anomalous dimensions of the original † A renormalizable theory can be deﬁned by specifying its ﬁeld content and its symmetries. Determining the behaviour of the Cl s in the limit M → ∞ is based on the fact that the (Euclidean) connected Green’s functions for the particles exponentially vanish when splitting among their arguments becomes much larger than 1/M.‡ This is equivalent to saying that they behave at most logarithmically. up to logarithmic corrections. ‡ Provided the interaction L ϕ does not contain coupling constants growing with M. and the symmetries deﬁning† Lϕ are also the symmetries of S including the possible gauge ﬁxing terms for . yk ) at points where all the yi almost coincide.131). since the contributions to the amplitudes from higher-dimensional Ol s are suppressed§ by powers of (external momenta)/M. (ii) asymptotically scale-invariant.135) to make the Cl s dimensionless. they are UV divergent. . Assumption (ii) tells us that the latter behaviour is independent of dimensionful parameters present in Lϕ . but these divergences are expected to cancel among different integrals when counterterms for the full theory deﬁned by the lagrangian (7. Hence. By assumption (i) Lϕ contains only terms allowed by symmetries of S (we assume the absence of anomalies). . up to logarithmic corrections. .7. In fact. Such a situation occurs when Higgs–fermion couplings in the Standard Model are considered in the process of heavy fermion decoupling. This requires the introduction of some restrictions on the action S in (7. in principle. only a ﬁnite number of Ol s needs to be considered in practice. We are going to bypass this point here. This guarantees that Lϕ can be absorbed into Lϕ by appropriate redeﬁnition of ﬁelds and parameters in Lϕ . not every set of ﬁelds and symmetries can give us a renormalizable theory. as we will see below. the leading behaviour of the Cl s at M → ∞ has the same property. non-renormalizable because it contains terms of arbitrary dimension. . § This can be easily shown on dimensional grounds. and that Cl s are convergent (or behaving at most logarithmically) in the limit M → ∞. because powers of 1/M have been already extracted in (7. However. We assume that the theory described by Lϕ is (i) renormalizable.128) are properly included. Checking this would require expressing Cl(k) in terms of the ϕ-particle Green’s functions (and their derivatives) with the help of (7. Moreover. The action is built from all possible terms of dimension ≤ 4 that are invariant under the symmetries.7 OPEs and effective ﬁeld theories 265 renormalization of its ﬁelds and parameters. Of course. Thus. . the behaviour of the Cl s at M → ∞ is determined by the leading behaviour of (Euclidean versions of) Cl(k) (y1 .135) is. the renormalization programme can be effectively carried out.

135).135).. when expressed in terms of renormalized quantities.135) is the fact that the perturbative expansion of low energy amplitudes generated by (7.135) combined with the RGE evolution allows us to extract large logarithms from all orders of the perturbation series. in principle. Replacing (7. The renormalization scale µ has then to be set equal to µ ∼ M. only a ﬁnite number of renormalization constants has to be found when one works up to a given order in 1/M. in practice. It is easy to convince oneself that Z l1 . . one recovers the renormalization constants in Seff determined by the requirement of ﬁniteness of amplitudes generated by Seff . ‡ We restrict ourselves to the µ-dependent renormalization schemes such as MS or the µ-subtraction scheme. Therefore. if the full theory is known.128) and (7. be carried on in the theory described by (7. . This is why the renormalization programme can. but not on the coefﬁcients Cl .‡ The third step is deriving and solving the RGEs for the effective action parameters.l j l are additional renormalization constants necessary to cancel divergences in diagrams with j Ol -vertices. .135). There are four main steps of this procedure: First. the superﬁcial degree of divergence of a diagram with Ol1 .42).266 7 Scale invariance and OPE operators in the OPE (7. Only in this sense are they dependent on m/M and other parameters present in (7. they can be recovered by requiring the equality of amplitudes generated by (7. where p = n(l1 ) + · · · + n(l j ) − n(l). Such Ol s have n(l) ≤ D + B +(3/2)F −4.. in order to avoid large logarithms.137) l ˜ where the Z l s are products of renormalization constants of the ﬁelds present in Ol s. and Z l1 . Such counterterms are generated by Ol s involving B bosonic ﬁelds. . and therefore n(l1 ) + · · · + n(l j ) − n(l) ≥ 0. takes the form 1 M n(l) Z l1 l Cl1 + l1 l1 l2 ˜ Z l1 l2 l Cl1 Cl2 + · · · Z l Ol (7. It vanishes for p < 0 which can be justiﬁed† using the arguments of Section 4. F fermionic ﬁelds and no more than D derivatives. also the renormalized parameters in Lϕ ) by requiring equality of amplitudes generated by (7. The sum over l in (7. They can be considered to be new free parameters to be determined from experiment or.. which results in expressing these parameters renormalized at the scale µF ∼ † According to (4. Ol j -vertices is equal to D = 4 − B − 3 F + n(l1 ) + · · · + n(l j ).128).128).128) by (7. one ﬁnds the coefﬁcients Cl (and. Next.133). .l j l is proportional to M − p . The reason for introducing the effective action (7. The renormalization constants in Lϕ may be different than in the underlying theory (7.. All the Z s in the above expression depend on m/M and the couplings present in Lϕ .128) and (7.128) is in practice an expansion in powers of (coupling constants)· ln[(external momenta)/M] which may be too large to be expansion parameters. it may require counterterms which 2 are polynomials in external momenta only of order ≤ D. After removing subdivergences.135).2.

only the ﬁrst of the sums in the bracket of (7. As already mentioned. . the outlined procedure allows us to extract large logarithms from all orders of the perturbation series. For simplicity. we get µ d Cl (µ) = dµ γkl (g(µ))Ck (µ).7 OPEs and effective ﬁeld theories 267 (external momenta) in terms of their counterparts renormalized at µI ∼ M. let us restrict our attention to the leading order in 1/M. it is convenient to solve the RGEs using the variable v = g 2 ln(µ2 /µ2 ) instead of µ itself. Sometimes this requires changing the normalization of some of the Ol s. The coupling constant g(µ) is therefore determined.1) in order to keep them dimensionless. Therefore. interpreting Cl s as new coupling constants. dµ (7.2 have to be modiﬁed.137) needs to be considered and. Buras & Harlander 1990).7. As long as one works in the framework of dimensional regularization. Then. the renormalized Cl s should be rescaled by appropriate powers of µ (as has been done with λR in Section 6. In the following. their number is usually very large. In principle. The variable v is treated I as independent of g(µI ).139) The last step is calculating the desired amplitudes with help of Seff renormalized at µF ∼ (external momenta). Then the methods of Section 6.2 are used for determining the µ-derivatives of renormalization constants. and all the RGEs are solved perturbatively in g(µI ). One should also remember that even the coefﬁcients Cl vanishing at µI can acquire non-zero values at µF .128) are expected to appear in the expansion (7. Even at the leading order in 1/M.135). all possible terms allowed by symmetries of (7. and the methods† of Section 6. up to a given order in g(µI ). multiplying them by m. the meaning of this statement will be made more precise. for example. The latter have been already found by ‘matching’ in the previous step and serve as initial conditions for the RGEs. by † It may not always be possible to deﬁne a mass-independent renormalization scheme in a theory described by Seff . from the equation d/dµ Seff = 0. This is done either perturbatively or with use of some non-perturbative methods (Buchalla. for example. at the beginning.138) where the anomalous dimension matrix γ is determined by ˆ γkl (g) = − j µ d Z k j (Z −1 ) jl . The RGEs are then derived.137). Here we consider only those cases in which it is possible to avoid the appearance of dimensionful parameters. elements of all the four steps are performed simultaneously in order to select the set of operators Ol that may contribute to the desired amplitudes at the desired order (even by their inﬂuence on the RGE evolution of some directly contributing Ol ). Let us assume that Lϕ contains only one (dimensionless) coupling constant g. k (7. Then. due to the non-diagonal character of the mixing in (7.

Grinstein. for example. There. for example.6)) µ d g(µ) = β(g). First of them covers low energy electroweak processes involving hadrons (see. but this is often a much better approximation than truncating the series in g(µI ) which has to be done anyway. dµ (7. Gilman & Wise (1979)). (More details of the effective theory approach are presented in Sections 12. the role of heavy particles is played by the W± and Z0 bosons. • the ﬁnal amplitude (last step) is calculated up to nth order in g 2 (µF ) and expanded with help of (7. we obtain the ﬁnal answer up to nth order in g 2 (µI ) and exactly in vF = 2g 2 ln(µI /µF ). knowledge of γ and β up to (n + 1)th ˆ 2 order in g is necessary for determining the Cl s up to nth order in g 2 (µI ) and exactly in v. There. The other domain covers phenomena taking place much above the electroweak scale (see. (7. It is important to notice that after decoupling the resulting effective theory usually has less symmetries than the original one (as in the examples mentioned above).6 and 12. This is the precise meaning of the statement that all large logarithms are extracted from the perturbation series. the b and c quarks are also treated as heavy.140) where f n (v) are certain functions found with help of the equation (see (6. the coupling constants are smaller but the evolution is usually longer. . Springer & Wise (1990)). In some processes. if • the renormalization constants (ﬁrst step) are found up to (n + 1)th order in g 2 .140) is then inserted into the RGE for the coefﬁcients Cl . The price we have to pay is the necessity of truncating the series in 1/M. (7. even the coupling constant splitting after the spontaneous symmetry breakdown in GUTs can be most easily understood in terms of the effective theory arising after the heavy gauge boson decoupling (Weinberg 1980). • the matching at µI (second step) is performed up to nth order in g 2 (µI ). Since µd/dµ = −2g 2 (µI )d/dv.141) The coupling g(µ) from (7. The coupling which becomes too large after multiplication by the logarithm is the QCD gauge coupling.128) itself. Thus. There are two main domains of applications of the described technique. Then the decoupling is done successively and the matching between several ‘effective theories’ is performed (see.7). Such a result could never be obtained with help of perturbative calculus based on the action (7. Wilczek & Zee (1979)). In fact.140). the top quark and the Higgs bosons. for example.138).268 7 Scale invariance and OPE ∞ n=0 the series g(µ) = g(µI ) g 2n (µI ) f n (v).

f (x)] = ∂µ f i[M µν . U (b) = exp(ibµ K µ ) exp(− 1 iωµν Mµν ).3 When the dilatation current is not conserved one may still deﬁne a time-dependent dilatation charge by D(t) = d3 x x µ 0µ (x) = t P 0 + d3 x xi 0i (x) Check that D(t) generates the same dilatation transformation (7. Pν ] = 2i(−gµν D + Mµν ) [M µν . U (ω) = for x µ = x µ + εµ (x). Obtain the . f (x)] = (x 2 gµν − 2xν xµ )∂ ν f where f (x) is a dimensionless scalar function and f (x ) = U f (x)U −1 where U (a) = exp(iaµ P µ ). (b) Check the algebra [Mµν . D] = iK µ [Mµν . Check the differential representation of the generators i[Pµ .14) on the ﬁelds as in the scale-invariant theory (use the canonical formula (7. D] = iPµ [K µ .26) is closed under commutation. 7. Polyakov & Zamolodchikov 1984). K λ ] = i(gµλ K ν − gνλ K µ ) [K µ . f (x)] = x µ ∂µ f i[K µ . 2 under the special (x) 7. f (x)] = (x µ ∂ ν − x ν ∂ µ ) f i[D. K ν ] = [D. Pν ] = [K µ . D] = 0 [Pµ .Problems 269 Problems 7. M µν ] = −i(g αµ M βν − g βµ M αν + g αν M µβ − g βν M µα ) (c) Check the inﬁnitesimal transformation law for a ﬁeld conformal transformation δµ (x) = (x 2 ∂µ − 2xµ xν ∂ ν − 2xµ d − 2x ν µν ) U (ε) = exp(iε D).2 Study conformal symmetry in two-dimensional quantum ﬁeld theories (Belavin.16) for Dµ (x)).1 (a) Check that the set of 15 inﬁnitesimal transformations deﬁned by (7. P λ ] = −i(g µλ P ν − g νλ P µ ) [M αβ . D] = [Pµ .

P α ] = P α − g α0 i[D(t). prove (7. M αβ ] = d3 x ∂µ D µ (x) d3 x (g α0 x β − g β0 x α )∂µ D µ (x) 7. The ﬁrst term is given by the diagram 0| jµ jν |0 = = CI (0) 0|I |0 .5 Obtain the expansion d4 x exp(iq x)T jµ (x) jν (0) = (qµ qν − q 2 gµν ) − 1 −q 2 2m ¯ ln 2 + 4 ¯ 4π 2 µ q +··· = 1 (dabc + icabc )λc 2 i where jµ = ¯ γµ . λa are SU (3) ﬂavour matrices 1 4 λa λb γµ γρ γν = (Sµνρσ + iεµρνσ γ5 )γ σ Sµρνσ = (gµρ gνσ + gρν gµσ − gµν gρσ ) Also derive the light-cone expansion for commutators of these currents (for example.4 (a) Using the Wick expansion and then the Taylor expansion of the well-deﬁned normal-ordered operator products.59) in the free scalar ﬁeld theory. 7. Ellis (1977)). (b) In the free quark model show that a b T Jµ (x)Jν (0) ∼ xµ →0 3δab (gµν x 2 − 2xµ xν ) dabc εµναβ x α Ac (0) I+ (2π)4 (x 2 − iε)4 2π 2 (x 2 − iε)2 ρS σ (0) cabc x µρνσ Jc + 2 + ··· 2π (x 2 − iε)2 dabc εµναβ x α Jc (0) cabc x ρ Sµρνσ Aσ (0) c + + ··· 2π 2 (x 2 − iε)2 2π 2 (x 2 − iε)2 β β T where a Jµ (x)Ab (0) ν xµ →0 ∼ a 1 Jµ (x) = : ¯ c (x)γµ 2 λa c (x): c (x): Aa (x) µ =: ¯ c (x)γµ γ5 1 λa 2 the repeated index c implies a sum over colours.270 7 Scale invariance and OPE commutators of D(t) with the Poincar´ generators e i[D(t).

is given by the zeroth order calculation. and ﬁnd C G . (0) (1) . Extend the calculation to ﬁrst order in α.Problems the second by 271 ¯ | jµ jν | = = C¯ (0) ¯|¯ | = C¯ (0) · =C¯ (0) Observe that to this order in α taking the appropriate matrix element singles out one particular contribution to the OPE. in addition (1) (1) to CI and C ¯ given by the same matrix elements (now taken to the ﬁrst order in α) calculate also the two-gluon matrix element g| jµ jν |g = C ¯ g| ¯ |g + C G g|G α G µν |g µν α where C ¯ .

In this section we discuss the theory formulation in covariant gauges. In terms of bare quantities it reads L = − 1 G α G αµν + 4 µν f ¯ f (iD − m f ) / f + gauge-ﬁxing term + Faddeev–Popov term (8. Bassetto et al.2. Let us therefore rewrite lagrangian (8.3 and 3.1) in terms of the renormalized quantities assuming the most general form consistent with BRS 272 .8 Quantum chromodynamics 8. the crucial issue is the gauge invariance or strictly speaking BRS invariance of the theory. The use of the axial gauge n · Aα = 0. As in QED. n 2 < 0. counterterms QCD is a theory of interactions of quarks and gluons. Collins (1984)) that the lagrangian is BRS-invariant after renormalization. where n is in each case some ﬁxed four-vector. Its lagrangian density has been discussed in Sections 1. n 2 = 0. is also often very convenient in perturbative QCD calculations.1 General introduction Renormalization and BRS invariance.1) where the sum is over all quark ﬂavours and the last two terms are given in (3. The quarks are assumed to transform according to the fundamental representation: each ﬂavour of quark is a triplet of the colour group SU (3). or light-like gauge. Let us concentrate on the renormalization programme. In light-like gauges the general proof of renormalizability is still lacking but order-by-order calculations have been consistently performed. QCD formulation in the axial gauge n 2 < 0 can be given in a similarly precise form to that in covariant gauges (for example. (1985)). It can be proved (for example.46) for the class of covariant gauges. Gauge bosons transform according to the adjoint representation. so that there are eight gluons.

3) follows from the requirement of the BRS invariance for the lagrangian (8.76) δBRS B = −igB T α α B (∂µ ηB α B ηB α B ηB B ≡ B (s B (s β B) δBRS ¯ B = igB ¯ B T α δBRS Aα = Bµ ≡ ≡ ≡ ¯ B) γ α B (s ABµ ) + gB cαβγ ηB ABµ ) ≡ B α B (sηB ) ∗α B (sηB ) (8.4) gB = g Z 1YM /Z 3 3/2 = g Z 1 /Z 2 Z 3 ˜ ˜˜ ˜ = g Z 1/ Z 2 Z 3 ˜ 1/2 (8.6) α δBRS ηB = 1 gB cαβγ ηB ηB 2 ∗α δBRS ηB = −(1/a) β γ µα B ∂µ AB † See also the beginning of Chapter 5.1) is the following Aα = Z 3 Aα µ Bµ 1/2 B = Z2 1/2 aB = Z 3 a m B = m Z 0 /Z 2 1/2 ˜ 1/2 ηB = Z 2 η ∗ ˜ 1/2 ηB = Z 2 η ∗ (8.2). Relation (8.1) is invariant under the BRS transformation (3.2) where ˜˜ ˜ ˜ ˜ g Z 1YM /Z 3 = g Z 1 /Z 2 = g Z 1 / Z 2 (8.8. Formally it can be derived from the Slavnov–Taylor identities (Ward identities for a non-abelian gauge theory) which themselves follow from the BRS invariance. As in QED there is no need for a gauge-ﬁxing counterterm. The correspondence between the renormalized quantities and the bare quantities of lagrangian (8.3) If we use a renormalization scheme such as the minimal subtraction scheme in which the renormalized couplings satisfy ˜ g=g=g ˜ ˜ then (8.3) implies the corresponding equality for the ratios of the renormalization constants.† We get (for one ﬂavour) L = − 1 Z 3 [∂µ Aα − ∂ν Aα − g(Z 1YM /Z 3 )cαβγ Aβ Aγ ]2 ν µ µ ν 4 ¯ iγµ [∂ µ + ig(Z 1 /Z 2 )Aαµ T α ] − m Z 0 ¯ + Z2 ˜ ∗α αβ 2 ˜ Z 1 / Z 2 )cαβγ Aγ ∂ µ ]ηβ − (1/2a)(∂µ Aβµ )2 ˜ + Z 2 η [δ ∂ + g( ˜ ˜ ˜ µ (8. .1 General introduction 273 invariance.5) Lagrangian (8.

They are given in Appendix B. In this context the behaviour of observable cross sections has been discussed in Chapter 7 in the framework of the OPE which provides a systematic way of factorizing effects at different mass scales.1) under (8. the notion of the effective coupling constant and its role in determining the behaviour of a quantum ﬁeld theory at different momentum scales have been extensively discussed in Chapter 6.2) under δBRS ˜ = −ig Z 1 T α ηα γ ˜ ˜ (∂µ ηα Z 2 + g Z 1 cαβγ ηβ Aµ ) ˜ δBRS ¯ = ig Z 1 ¯ T α ηα δBRS Aα = µ ˜ δBRS ηα = 1 g Z 1 cαβγ ηβ ηγ 2 δBRS η∗α = −(1/a) ∂µ Aµα (8.8) have been derived in Section 3. Thus. (8. This is actually what is necessary for ﬁniteness of the composite operators δ .3).7).4) and (8.2.6) and if is identiﬁed as 1/2 ˜ 1/2 = (1/Z 3 Z 2 ) B ˜ We note the presence of the divergent renormalization constants Z i in the BRS transformation (8. the behaviour of the Green’s . µ ˜ Lagrangian (8.2) can be rewritten introducing counterterms (we take g = g = g) ˜ ˜ / L = − 1 G α G µν + ¯ iD 4 µν α −m ¯ − (1/2a)(∂µ Aµ )2 β − m(Z 0 − 1) ¯ + η∗α (δ αβ ∂ 2 + gcαβγ Aµ ∂µ )ηβ γ ¯ i/ − (Z 1 − 1)g ¯ γ µ Aα T α + (Z 2 − 1) ∂ µ − 1 (Z 3 − 1)(∂µ Aα − ∂ν Aα )2 + 1 (Z 1YM − 1)g(∂µ Aα − ∂ν Aα )cαβγ Aµ Aν ν µ ν µ β γ 4 2 −1 2 − 1 g 2 (Z 1YM Z 3 − 1)cαβγ Aβ Aγ cαρσ Aρµ Aσ ν µ ν 4 ˜ ˜ + ( Z 2 − 1)η∗α ∂ αβ ∂ 2 ηβ + ( Z 1 − 1)gη∗α cαβγ Aµ ∂µ ηβ γ (8.5) are inserted into (8.8) The Feynman rules for the vertices of (8. δ Aα and δηα (Collins 1984).7) Invariance of the lagrangian (8. The complete set which includes the counterterms is given in Appendix C (in the notation of Chapter 12). δ ¯ .2) under (8.7) can be checked explicitly but it also follows immediately from the invariance of bare lagrangian (8.6) if relations (8. QCD is an asymptotically free theory: the coupling constant decreases as the scale at which it is deﬁned is increased. Asymptotic freedom of QCD The RGE.274 8 Quantum chromodynamics and lagrangian (8.

9) Diagrams contributing to different renormalization constants are determined by the structure of counterterms in lagrangian (8. Tr[λa λb ] = 2δ ab and in the regular (adjoint) eight-dimensional representation (T a )bc = −icabc where cabc s are the SU (3) structure constants. Both appear in the QCD vertices.8).10) .1 General introduction 275 functions when all the momenta are scaled up with a common factor is governed by a theory where g → 0.2) the new features in QCD Feynman diagram computation are the group theory factors present for each diagram. Tadpole diagrams have been ignored because they do not contribute when dimensional regularization is used. The β-function in QCD deﬁned as β(α) = µ(d/dµ)α(µ2 )|αB ﬁxed . One can use the most convenient way of calculating Z α .8.33) and (6.45) β(α) = α α b1 + π α π 2 α = g 2 /4π b2 + α π 3 b3 + · · · (6. The basic ingredients are SU (3) generators in the fundamental three-dimensional representation conventionally taken as (T a )bc = 1 (λa )bc 2 where λs are the Gell-Mann matrices. is given by expansion (6.1 (sums over ﬂavours and over different permutations are omitted).35): β(α) is totally determined by the residua of the simple UV pole of Z α (beyond the one-loop order we need to calculate a non-leading UV divergence) where according to (8. For instance at the one-loop level we have the contributions shown in Fig. Only theories with a non-abelian gauge symmetry can be asymptotically free in four space-time dimensions (Coleman & Gross 1973).45) and the coefﬁcients bi are calculable perturbatively. Using dimensional regularization and a mass-independent renormalization scheme we have (6.5) one can use one of the relations 2 3 2 2 ˜2 ˜2 Z α = Z 1YM /Z 3 = Z 1 /Z 2 Z 3 = Z 1 / Z 2 Z 3 (8. 8. This can explain the success of the parton model in describing the scaling phenomena in the deep inelastic processes. and in the diagram calculation one often encounters the following combinations: a a 2 (TR )bc (TR )cd = (TR )bd = CR δbd a. As compared to QED (Section 5.c (8.

11) CR is the eigenvalue of the quadratic Casimir operator in the representation R and . 8. and a b a b (TR )dc (TR )cd = Tr[TR TR ] = TR δ ab c.1.d (8.276 8 Quantum chromodynamics Fig.

One can proceed as in Section 5. α. expression (6. Vladimirov & Zharkov (1980). s and sη that appear in the BRS † The coefﬁcient b1 was ﬁrst calculated by Politzer (1973) and Gross & Wilczek (1973). Marciano & Pagels 1978).11) (Abers & Lee 1973.1 General introduction 277 for SU (N ) CF = (N 2 − 1)/2N CA = N The trace TR is TF = 1 2 F ≡ fundamental A ≡ adjoint TA = N So far the ﬁrst three terms in the expansion (6. their analogue in a non-abelian gauge theory – the Slavnov–Taylor identities for bare regularized or for renormalized Green’s functions – can be derived in several different ways. because of the presence of ghost ﬁelds it is very complicated to obtain the analogous relation for the 1PI generating functional (A. s ¯ .53) in the next-to-leading approximation. In addition to sources J. the coefﬁcient b2 by Caswell (1974) and Jones (1974) and the coefﬁcient b3 by Tarasov. ¯ L. .3. respectively.1: apply gauge transformation to the generating functional for full or connected Green’s functions which is written as a functional of the physical ﬁeld sources only.51) in the leading logarithm approximation. ). It is often convenient to express the result of a QCD calculation in terms of the dimensional parameter instead of the dimensionless α. This has been discussed in Section 7. α for the ﬁelds A. and generally by (6. Then one derives the non-abelian analogue of (5. The Slavnov–Taylor identities As with the Ward identities in QED. ¯ .52).45) have been calculated and in the minimal subtraction scheme they are (for SU (N ) and n ﬂavours)† −b1 = 11 CA − 2 TF n 6 3 17 2 1 5 −b2 = 12 CA − 2 CF TF n − 6 CA TF n (8. ¯ and . one ¯ introduces sources cα and c∗α for the ghost ﬁeld η∗α and ηα as well as sources K . A more elegant method is based on the BRS transformation (Lee 1976). This form generates all the identities for the connected Green’s functions but.8. L and M for the composite operators s A.12) 79 3 2 −b3 = 2857 CA − 1415 CA TF n + 432 CA TF2 n 2 − 205 CA CF TF n 1728 864 288 44 1 2 + 288 CF TF2 n 2 + 16 CF TF n The QCD effective coupling constant is given by expression (6.

15) . Using (8. α. K . One can convince oneself that in the ¯ presence of additional sources K .13) ¯ + c∗ η + η∗ c + K s A + s ¯ L + Ls Changing variables in the functional integral according to the BRS transformation and using the invariance δSeff /δ as well as the nilpotent relations s 2 A = s 2η = s 2 = s2 ¯ = 0 =0 one obtains the desired general relation generating the Slavnov–Taylor identities. M] ¯ = D(A ¯ ηη∗ ) exp iSeff + i d4 x (J A + α ¯ + Msη) + ¯α (8. The simplest way. L. The general relations whose derivation has just been brieﬂy described contain all the information about BRS invariance and are useful in formal manipulations. L . c.14) where ηα = δ Z /δc∗α . let us consider the regularized Green’s function 0|T Aα (x)η∗β (y)|0 . This is to obtain identities that are linear in derivatives with respect to the sources. M] = Z [J. c∗ .278 8 Quantum chromodynamics transformation (8. . c∗ . K . K . Since the ghost ﬁelds are treated on equal footing with the physical ﬁelds it is now straightforward to obtain similar relation-generating identities for the 1PI Green’s functions. α. L and M the functional still generates 1PI Green’s functions. c∗α = −δ /δηα etc. α. For instance.6). Next one deﬁnes the generating functional ¯ W [J. L . L . L. L . M] ¯ − d4 x (J A + α ¯ + ¯ α + c∗ η + η∗ c) (8. ¯ . α. L. however. As usual this is achieved by means of the Legendre transform ¯ ¯ [A. Llewellyn Smith (1980)). c. to get speciﬁc identities for the Green’s functions of interest is based on the observation that since BRS invariance is an exact symmetry of the theory all the Green’s functions are BRS invariant (for example.6) we have µ δBRS 0|T Aα (x)η∗β (y)|0 = 0|T (∂µ ηα (x) + gcαβγ ηβ (x)Aγ (x))η∗β (y)|0 µ µ − (1/a) 0|T Aα (x)∂ν Aνβ (y)|0 = 0 µ (8. η. η∗ .

α ¯α Aµ = δ Z /δ Jµ (8.21) implies invariance under two kinds of transformations on Aµ and Aµ which give the same δ Aµ quantum δAµ = 0 δ Aµ = Dµ (x) + [Aµ .20) where Aα is the background ﬁeld and Aα is the quantum ﬁeld (we recall that µ µ Aµ (x) = ig Aα (x)T α .14)) the longitudinal part of the propagator is unchanged by the higher order corrections. Dµ = ∂µ + [Aµ .18) (8.8. as usual.16) Result (8. Let us consider ﬁrst the gauge-invariant part of the action and introduce the background ﬁeld by the replacement A µ = A µ + Aµ (8. Z [J ] = −i ln W [J ] and the effective action ¯ ¯ F[ A] = Z [J ] − J · A.2 The background ﬁeld method This method is based on the idea of quantizing ﬁeld theory in the presence of a background ﬁeld chosen so that certain symmetries of the generating functionals are restored. µ µ Invariance of the action under the transformation (1.22) .17) and deﬁne.] (8. (x)] (8. . (x)]. 8. (x) = i α (x)T α and we deﬁne Aµ (x) = igAα (x)T α ).2 The background ﬁeld method 279 Differentiating (8.16) implies that as in QED (see (5. we can restore it by introducing a background ﬁeld with properly chosen transformation properties. However. J·A = α d4 x Jµ Aαµ (8. Let us consider gauge theories. .56a) we conclude that µ ν (1/a)∂(x) ∂(y) 0|T Aα (x)Aβ (y)|0 = −iδ(x − y)δ αβ µ ν (8.19) Manifest gauge invariance is lost because of the gauge-ﬁxing procedure. We start with the gauge-ﬁxed functional integral (the dependence on other ﬁelds and on the ghost sources is not indicated explicitly) W [J ] = D Aµ exp(iSeff + iJ · A ).15) with respect to xµ and using the equation of motion for the Green’s function 0|T ηα (x)η∗β (y)|0 which can be derived analogously to (2. .113) δ Aµ = Dµ (x) = ∂µ (x) + [Aµ (x).

Aµ ]: Aµ → Aµ + Aµ one gets ˜ W [J. Aµ ] in the presence of the background ﬁeld. Aµ ))2 = −(1/2a)(Dµ Aµ )2 = −(1/2a)(∂ µ Aα − gcαβγ Aµ Aγ )2 (8. . (x)] (8. For instance. Aµ ] − J · A α a ˜ ˜ Aµ = δ Z [J.] δ Aµ = [Aµ (x).25) for W [J. . . The generating functional ˜ W [J.27) (8.24) α µ β µ The Faddeev–Popov ghost term then takes account of ˜ det Mαβ = det(δ F α /δ β ) ˜ where δ F α corresponds to the quantum gauge transformation δ Aα = (1/g)∂µ µ α + cαβγ β (Aγ + Aγ ) µ µ and it also transforms covariantly under the background gauge transformation.29) . Aµ ] = D Aµ exp{iSeff [Aµ + Aµ ] + iJ · A} (8.23) We can maintain the manifest gauge invariance of Seff with respect to the ˜ background transformations if the gauge-ﬁxing function F α [A. Aµ ] = −i ln W [J. Changing the integration ˜ variables in (8. Aµ ]/δ Jµ if ˜µ δ Aα = cαβγ β (8.25) is manifestly invariant under the background gauge transformations if we complete them with the transformation α δ Jµ = cαβγ β γ Jµ (8. Aµ ] = Z [J. choose ˜ ( F α (Aµ .280 8 Quantum chromodynamics background δAµ = Dµ (x).28) ˜µ Aγ ¯ The next step is to ﬁnd the relationship between the original [ A] and the ˜ ˜ [ Aµ . Aµ ] and for ˜ ˜ ˜ ˜ [ Aµ .26) The same is true for ˜ ˜ Z [J. Aµ ] = W [J ] exp(−iJ · Aµ ) (8. A] transforms covariantly under these transformations. Dµ = ∂µ + [Aµ .

It should also be remembered that this [ Aµ ] [A α α ˜ corresponds to the gauge-ﬁxing term F [A ] = F [A − A. Aµ ] (check it). restricting ˜ ˜ µ lines.32) A very important fact is that ˜ [0. ˜ ˜ In particular it is convenient to calculate ˜ [ Aµ . The background ﬁeld method is extensively used in supergravity theories (Gates. Aµ ] is invariant under the transformation δAµ = Dµ (x). Aµ ] given by (8. Aµ ] = Z [J ] − J · Aµ Therefore ˜ ˜ ˜ ˜ [ Aµ . that the divergent terms and the counterterms are manifestly invariant with respect to the background gauge transformation.2 The background ﬁeld method 281 and correspondingly ˜ Z [J. These ‘ Aµ -vacuum’ diagrams are ourselves to diagrams with no external A ˜ the 1PI subset of diagrams obtained from W [0. Aµ ] and completes the evaluation of the β-function in the oneloop approximation (Problem 8. i. A] which may be an unusual gauge in the standard approach. it is a gauge-invariant functional of Aµ . i. Aµ ] is the usual effective action ¯ ˜ ¯ ¯ µ ] evaluated at Aµ = Aµ + Aµ . Grisaru. the calculation of the β-function in the one-loop approximation simpliﬁes to the calculation of the diagrams in Fig.e. It implies.e. This property is maintained in a perturbative loop-by-loop calculation of ˜ [0.e. Thus the inﬁnities appearing in ˜ [0. Aµ ] = Z [J ] − J · Aµ − J · Aµ = [Aµ + Aµ ] (8. Aµ ] (8. ghost and other non-gauge ﬁeld lines.3) that the renormalization constants (Aµ )B = Z A Aµ gB = Z g g satisfy Zg = ZA −1/2 1/2 (8. for instance.30) ˜ and we conclude that the effective action ˜ [ Aµ . where µν Gα = ∂µ Aα − ∂ν Aα − gcαβγ Aβ Aγ µν ν µ µ ν It is then straightforward to show (Problem 8.2. It is recommended that the reader derives the Feynman rules necessary for the calculation of ˜ [0. in particular.31) (8.34) and. Roˇ ek & Siegel 1983). diagrams with only internal Aµ lines and external Aµ . 8.25). Aµ ] must take the form of a divergent constant times (Gα )2 . i. We get [Aµ ] = ˜ [0.8. c .3).33) (8. Aµ ] for Aµ = 0.

Thus the vacuum deﬁned by G µν = 0 restricts the potentials to ‘pure gauge’ conﬁgurations Ai (r) = −U −1 (r)∂i U (r) Further discussion relies on an additional restriction that r →∞ (8. The matrix functions U provide a mapping of S 3 into the manifold of the SU (2) gauge group. Since any element M in the SU (2) group can be written as M = a + ib · σ where σ s are Pauli matrices and where real a and b satisfy a 2 + b2 = 1 . for example. the so-called gauge-ﬁxing procedure used in perturbation theory usually leaves some residual gauge freedom.36) where U∞ is a global (position-independent) gauge transformation. In pure Yang–Mills theory one can argue (Jackiw 1980). but not prove.36) which ensures that the non-abelian charge is well deﬁned (see Problem 1. see.3 The structure of the vacuum in non-abelian gauge theories Homotopy classes and topological vacua For the moment we shall consider the pure Yang–Mills gauge theory with SU (2) as the gauge group. For instance we can use the gauge A0 = 0 (for the present discussion this is convenient but not necessary. As far as the gauge functions U (r) are concerned the three-dimensional spatial manifold with points at inﬁnity identiﬁed is topologically equivalent to S 3 – the surface of a four-dimensional Euclidean sphere labelled by three angles. Equivalently.8). we assume that all vector potentials fall faster than 1/r at large distances.282 8 Quantum chromodynamics Fig.2. 8. As we know. Coleman (1979)) which obviously leaves room for further gauge transformations U (r) depending on spatial variables only.35) lim U (r) = U∞ (8. 8. that physically admissible gauge transformations satisfy (8.

there exists a continuous deformation of maps F(x. Let X and Y be two topological spaces and f 0 . They can be classiﬁed into homotopy classes. and only if. Thus. Hence the integer N is called the winding number and each homotopy class is characterized by its winding number. This discussion can be generalized by taking X = S n (the n-dimensional sphere) or the topologically equivalent n-dimensional cube with all its boundaries identiﬁed and equivalent to some x0 of S n . The group of homotopy classes of mappings from S 1 into the manifold of G is called the ﬁrst homotopy group of G and is denoted by 1 (G). 0 ≤ t ≤ 1. Take the mappings from S 1 into G ≡ U (1) represented by a set of unimodular complex numbers z = exp(iu) which is itself topologically equivalent to S 1 . They are said to be homotopic if they are continuously deformable into each other.3 The structure of the vacuum 283 we conclude that the manifold of the SU (2) group elements is topologically equivalent to S 3 .8. such that F(x. The classes of mappings S n → S n with . One can think of u(θ) as a mapping of a circle into another circle such that N points of the ﬁrst circle are mapped into one point of the second circle (winding N times around the second circle). gauge transformations U (r) provide mappings S 3 → S 3 . We are interested in mappings from S 1 into the manifold of a Lie group G. with θ and θ + 2π identiﬁed.1). t). For this example 1 (U (1)) = 1 (S 1 )=Z where Z denotes an additive group of integers.38) The mappings of any winding number can be obtained by taking powers of u (1) (θ) = exp(iθ) (8. i.39) Instead of the unit circle S 1 we can consider the whole real axis −∞ < x < ∞ with the points x = ±∞ identiﬁed which is topologically equivalent to S 1 (see Problem 8. We can divide all mappings of X into Y into homotopic classes: two mappings are in the same class if they are homotopic. The function F(x. Let us consider a simple example. f 1 be two continuous mappings from X into Y . A group structure can be deﬁned on the set of homotopy classes.37) form a homotopic class for different values of a and a ﬁxed integer N . Thus S 1 → S 1 and the continuous functions u(θ ) = exp[i(N θ + a)] (8. t) is called the homotopy. 0) = f 0 (x) and F(x. Let S 1 be a unit circle parametrized by the angle θ. if. 1) = f 1 (x).e. We observe that in our example the winding number N can be written as N = −i 0 2π 1 du dθ 2π u(θ) dθ (8.

44) The hamiltonian H= d3 x H(x). is every mapping of S 3 into U (1) continuously deformable into the constant map f (x) = y0 corresponding to N = 0.42) where the Ai s are given by (8. Only if the group is U (1).35).38) and (8.e. H= a a i ∂0 A i −L (8. In the Aa = 0 gauge the 0 canonical variables are the Aia s and their conjugate momenta a i = δL = G a = ∂0 Aia = −E ia 0i δ(∂0 Aia ) (8.1 the winding number for gauge transformations providing the mapping S 3 → G is given by N = −(1/24π 2 ) d3 x Tr[εi jk Ai A j Ak ] (8. mappings S n → S n are classiﬁed by the number of times one n-sphere covers the other. It can be shown that analogously to (8.41) and this together with relations (8.36) implies an inﬁnity of topologically distinct vacua for the SU (2) Yang–Mills theory.35) and (8. Thus 3 (SU (2)) = 3 (S 3 )=Z (8.43) To see this let us ﬁrst recall the canonical formalism.1. The same result holds for an arbitrary simple Lie group (for SU (N ) in particular) due to a theorem (Bott 1956) which says that any continuous mapping of S 3 into G can be continuously deformed into a mapping into an SU (2) subgroup of G.168) in Problem 8. For prototype mapping with N = 1. As we have already mentioned the group SU (2) is also equivalent to S 3 .45) .40) i. see Problem 8. We have n (S n )=Z (8. Ordinary space R 3 with all points at ∞ identiﬁed is equivalent to S 3 . Physical vacuum The inequivalence of topologically distinct vacua can be understood as a consequence of Gauss’ law Diab G b = 0 0i (8.284 8 Quantum chromodynamics one ﬁxed point f (x0 ) = y0 form a group called the nth homotopy group of S m and designated by n (S m ).

although we obtain a consistent quantum theory. E b (y. We cannot set Diab E ib to zero as this operator does not commute with the canonical variables. Actually.43) is not found. Thus. as the next step.53).49) (8. Aia ] = −E ia b ∂0 E ia = i[H.54) ˆ where T (0) is the unitary transformation corresponding to U = exp(iQ G ) in the class N = 0. t)] = iδ ab δi j δ(x − y) j the hamiltonian equations give ∂0 Aia = i[H. Integrating by parts and noting that for the N = 0 class one gets QG = d3 x [−Diab G a 0i b (x)] (8. that only gauge transformations which belong to the homotopy class N = 0 can be built up by iterating the inﬁnitesimal gauge transformations (see Problem 8. t). for a gauge transformation in the class N = 1 we have T (1) |N = |N + 1 (8.48) (8.46) (8. However.55) .53) we conclude that such transformations leave the vacuum state |N invariant: T (0) |N = |N (8. Ab (y. t).47) When the canonical commutation relations are imposed [Aia (x.8. The generator of these gauge transformations is (see Problem 1. Ab (y.53) Using (8. Gauss’ law can be incorporated into the theory by demanding that the physical states are only those which are annihilated by Diab G b 0i Diab G b | 0i =0 (8.50) but Gauss’ law (8. t)] = [E ia (x.3 The structure of the vacuum 285 is the energy H= 1 2 2 d3 x (E a + Ba ) = d3 x 00 2 a E ia = G i0 = − ia . as seen from (8.8) QG = d3 x G a Diab 0i b (x) a (8. it generates inﬁnitesimal space-dependent gauge transformations. t).2).51) and (8.52) (x) vanishes at x → ∞ ˆ and U = exp(iQ G ). E ia ] = εi jk D ab Bk j (8. it is different from the desired gauge theory. Bia = − 1 εi jk G a jk 2 (8.51) Having (8.51) we can return to our main question and recognize. t)] = 0 j j [E ia (x.

(8.286 8 Quantum chromodynamics Since the states |N are functionals of Aµ . ˆ Indeed. for example. U N ] = 0 then 0= and ˆ | O| = 0 if = (8. the degeneracy between different -vacua is lifted if there exists tunnelling between topological vacua.55) follows from the transformation properties of the winding number N (A) given by (8. 0 ≤ ≤ 2π. U N ]| = [exp(−iN ) − exp(−iN )] ˆ | O| . Dashen & Gross (1976)). For instance.58) = ∞ N =−∞ exp(iN )|N (8. when there are massless fermions in the theory. In the absence of tunnelling. Thus the deﬁnitions of the winding number as well as of the topological vacua are gauge-dependent. if O is any gauge-invariant operator ˆ ˆ [ O.42) −1 −1 N (Un AUn − Un ∂i Un ) = N (A) + n (8. In our gauge. the -vacua remain degenerate and the angle has no physical meaning. Gauge invariance led us to the notion of the physical | vacua as superpositions of the topological vacua (see also Callan. pure gauge conﬁgurations for which E and B are zero are analogous to inﬁnitely degenerate zero-energy conﬁgurations in a quantum mechanical problem with periodic potential.57) The physical vacuum is characterized by which is a constant of motion and each | vacuum is the ground state of an independent sector of the Hilbert space. ˆ ˆ |[ O. The classical vacuum will be unique. of course.56) where Un is in the class N = n and Aµ is pure gauge: A = H −1 ∂µ H . Now the true vacuum must be at least phase-invariant under all gauge transformations because they commute with observables. and the physical vacuum is given by | so that T (M) | = exp(−iM )| (8. the physical | -vacuum is. However. This interpretation of the | -vacuum is certainly gauge-dependent (Bernard & Weinberg 1977). one could choose a physical gauge where Aµ is uniquely determined in terms of G µν with no residual gauge freedom at all. In pure Yang–Mills theories such tunnelling is due to instantons. Because T is unitary its eigenvalues are exp(−i ).59) We want to stress that we have been working in a ﬁxed gauge A0 = 0. As in the quantum mechanical problem.

7) and no surface terms arise from the second term.6) ˜ −(1/16π 2 ) Tr[G µν G µν ] = ∂µ K µ (8. We are interested in transitions from one classical vacuum to another so we require that the initial t → −∞ and ﬁnal t → +∞ ﬁeld conﬁgurations are of the pure gauge form.60). -vacuum and the functional integral formalism Let us consider a vacuum-to-vacuum transition given by the functional integral 0| exp(−iH t)|0 ∼ D Aµ exp i d4 x (LYM + LG + LFP ) (8. Since (see Problem 1.63) ˜ d4 x Tr[G µν G µν ] (8. This will become clear in the next subsection where we discuss the functional integral representation for vacuum-to-vacuum transitions.8.62) The ﬁrst term vanishes from the Bianchi identities (Problem 1. we shall require that gauge potentials tend to a pure gauge at large distances in all four directions Aµ −→ U −1 ∂µU |xµ |→∞ (8. Let us now consider the integral Q = −(1/16π 2 ) where ˜ G µν = 1 εµνρσ G ρσ 2 Q is gauge-invariant and is called the topological charge or the Pontryagin index. Firstly. Q is a topological invariant. we must pay some attention to the possible boundary conditions on the ﬁelds in the integral (8. It is stationary with respect to any local variation δ Aµ whether or not the equation of motion is satisﬁed δQ ∼ = =− 1 2 ˜ d4 x Tr[G µν (D µ δ Aν − D ν δ Aµ )] ˜ d4 x Tr[G µν D µ δ Aν ] ˜ d4 x Tr[Dµ G µν δ Aν ] + ˜ d4 x ∂ µ Tr[G µν δ Aν ] (8.64) . Actually.61) For the time being we leave the gauge unspeciﬁed.3 The structure of the vacuum 287 a gauge-invariant concept.60) It remains to be understood what one means by ‘vacuum’ for the transition given by (8.60).

The gauge-dependent notion of . tends (we always also assume (8.67) to be ﬁnite) and the functional integral (8.66 Q= d3 x K 0 ∞ −∞ = N (+∞) − N (−∞) (8. Therefore in this gauge the topological charge Q can be interpreted as the winding number of the pure gauge conﬁguration to which Ai .62).67) Using the remaining gauge freedom we can set N (−∞) = 0.65) For potentials Aµ satisfying (8.42) and 8. The result turns out to have the gauge-invariant form for which we are looking. We have | exp(−iH t)| = m. Then from (8.n exp[i(n − m) ] exp[in( − ) Q − )] m| exp(−iH t)|n = 2πδ( ∼ Q exp(−iQ ) N + Q| exp(−iH t)|N D A Q exp i d4 x (LYM + LG + LFP ) exp(−iQ ) D Aµ exp i = d4 x {LYM + LG + LFP (8.61) Q may be written in terms of U by inserting the asymptotic form of K µ in (8. In the gauge A0 = 0 we are now ready to construct the functional integral corresponding to the | → | transition.62).60).68) ˜ + (1/16π 2 ) Tr[G µν G µν ]} where we have used (8. As promised the ﬁnal result is gauge-independent and it deﬁnes the theory having a well-deﬁned vacuum state | The | -vacuum is deﬁned by the vacuum expectation values of operators which in turn are determined by the functional integral with a given value of .288 8 Quantum chromodynamics where K µ = −(1/16π 2 )εµαβγ Tr[G αβ Aγ + 2 Aα Aβ Aγ ] 3 we can write Q as an integral over the surface S at inﬁnity Q= dSµ K µ (8.36) which is necessary for the integral in (8. as the transition amplitude between topological vacua |N = 0 and |N = Q or more generally between |N and |N + Q . with the integral restricted to the Q sector deﬁned by (8.66) To make contact with the discussion earlier in this section let us take the gauge A0 = 0.65) Q = −(1/24π 2 ) dSµ εµαβγ Tr[U −1 ∂α UU −1 ∂β UU −1 ∂γ U ] (8.

Thus tunnelling takes place and the -vacua are not degenerate.8. take fermions to be massless. This is no longer true when fermions are massive since then chiral symmetry is also broken explicitly. . The new term in the lagrangian is a total divergence so it does not inﬂuence the classical equations of motion. Dine. So far no convincing explanation exists for the smallness of (Peccei & Quinn 1977. Wilczek 1978.36) and having different winding numbers as x4 → +∞ and x4 → −∞. However. The -term violates P and C P conservation so one could argue that we should set = 0 in the strong interaction theory. It is convenient to work in Euclidean space and to look at Minkowski Green’s functions as the analytic continuation of the Euclidean ones. However. Weinberg 1978. it modiﬁes quantum dynamics which depends on the action.3 The structure of the vacuum 289 the topological vacua does not appear in this formulation. The instantons are classical E = 0 trajectories in Euclidean space-time. Fischler & Srednicki 1981). using the path integral representation (8. Firstly. New interesting aspects then appear which cannot be considered in detail before discussion of chiral symmetry and anomalies. Then the -vacua turn out to be degenerate: they are the degenerate vacua of spontaneously broken axial UA (1) global symmetry of a theory with massless fermions.68) we can always write the physical vacuum as a superposition | = dν exp(iν )|ν ˜ where |ν is an eigenstate of the operator Tr[G µν G µν ] with the eigenvalue ν. With massless fermions changes with chiral rotations and can be rotated away. Let us summarize brieﬂy only the most important facts (see also Chapter 13). This mechanism does not imply the existence of the Goldstone boson coupled to the gauge-invariant UA (1) current and it offers a prototype solution to the so-called UA (1) problem. However. We face a situation where on one hand the ﬁeld conﬁgurations with Q = 0 are welcome to solve the UA (1) problem and on the other hand the experimentally allowed value of is surprisingly small as for a strong interaction parameter: from the upper limit on the neutron dipole moment < 10−19 . Also it does not contribute to the energy–momentum tensor. hence in Euclidean space A4 = 0) one can cast the instanton solution into a form which tends to pure gauge conﬁgurations satisfying (8. Finally in the presence of fermions the spontaneous breakdown of the UA (1) and of the chiral † By a gauge transformation which removes A4 (we are in the gauge A0 = 0.† Fermions must be included in a realistic theory. Explicit solutions for ﬁeld conﬁgurations giving Q = n = 0 are known (instantons). this does not help since C P-violating weak interactions renormalize to a non-zero value. The spontaneous breakdown of the UA (1) is the combined effect of the anomaly in the divergence of the axial current and of the existence of ﬁeld conﬁgurations with a non-zero topological charge Q.

The life-time of a virtual state with quasi-free constituents is P/( E)2 and it becomes long compared to the collision interaction time which is O(1/q0 ) (here E is the scale of binding effects: E ∼ O(1/r ) where r is the radius of the hadron). It can be used for any hard scattering process. However. No such connection has so far been derived from QCD itself. If in a given process there are several large momenta O(q 2 ) then we assume the ratio of these variables to be ﬁxed as q 2 → ∞.290 8 Quantum chromodynamics SU (n) × SU (n) are phenomena which are connected to each other by the chiral Ward identities based on the spontaneous breaking of the SU (n) × SU (n) and on the anomalous non-conservation of the UA (1) current. The usual argument to support this conjecture is as follows: view the scattering in a frame in which the hadron (or hadrons) is fast: P → ∞. By the parton picture one means. for instance. 8. Parton picture There are two basic ingredients in applications of perturbative QCD to hadronic processes. This is. The time scale of the bound state effects is dilated as P rises. The ﬁrst one is to assume the parton picture which provides a connection between QCD and hadrons. The parton picture is implicit in the OPE analysis of the deep inelastic lepton– hadron scattering in Section 7. speaking most generally. for example. However. . One then simply sets = 0.5. the assumption that in processes with large momentum transfer q hadrons can be pictured as ensembles of partons (quarks and gluons) and that there is no interference between the long time and distance physics responsible for the binding of quarks and gluons into hadrons and the short distance effects relevant for the large momentum transfer process under consideration and occurring at the parton level. in the QCD calculation of the † This point has been particularly emphasized by Crewther (1979).† The structure of the QCD vacuum brieﬂy discussed in this section is of fundamental physical importance.4 Perturbative QCD and hard collisions Historically.5 and. it is usually neglected in applications of perturbative QCD to short distance phenomena. the Bjorken limit in the deep inelastic lepton–hadron scattering discussed in Section 7. in which there is a large momentum scale q 2 . non-perturbative in nature. not necessarily light-cone dominated. the light-cone expansion discussed in Chapter 7 was the original foundation for perturbative QCD applications. this OPE combined with the renormalization group techniques are applicable to a limited number of problems and a more general approach based on direct summation of Feynman diagrams has been widely developed.

Factorization theorem The second central issue for perturbative QCD calculations is the factorization theorem (Ellis et al.e. In most applications. as we expect from our QED examples. P − m 2 /4P) 2 2 2 2 (8. It is a generalization of the factorization encountered in the OPE approach and applies to the dσ QCD which under the parton model assumption is to be calculated perturbatively as the sum of appropriate quark–gluon Feynman diagrams. with the cross section for the hard subprocess which involves only partons and can be studied perturbatively. q) (8.2.4 Perturbative QCD and hard collisions 291 total cross section for the e− e+ annihilation into hadrons which is identiﬁed with the total cross section e− e+ → quarks and gluons. the perturbative expression for dσ QCD contains singular contributions coming from certain regions of phase space in which the quark and gluon momenta are small or parallel. We assume that in the Bjorken limit dσ had (P. To be speciﬁc let us consider the deep inelastic electron–proton scattering depicted in Fig. p⊥ . parallelly moving partons and f i (z) dz as the average number of partons of type ‘i’ having momentum fraction z. In the limit P → ∞. Sachrajda 1983b). The problem is that.8.70) where σiQCD is the cross section for deep inelastic scattering on the parton ‘i’ as the target and f i is the four-momentum distribution of partons of type ‘i’ in the proton. In this limit we envisage the proton as a collection of massless. it is related to the square of the soft wave-function.70) has a very interesting physical interpretation (see (8. p 2 .69) P = (z P + ( p⊥ + p⊥ )/4z P. p⊥ . but not the only one . q) = i dz d2 p⊥ d p 2 f i (z. p⊥ . and use the so-called light-cone parametrization of the four-momenta P = (P + m 2 /4P. this parton model assumption takes a more concrete form. We call P and p the proton and parton momenta. z + dz of the total proton momentum. respectively. We shall see shortly that in the inﬁnite momentum frame (8. O. p 2 )σiQCD (z P. p⊥ and p 2 ﬁnite (inﬁnite momentum frame) we see that p = z P up to corrections O( p⊥ /z P). z P − ( p⊥ + p )/4z P) d4 p = 1 (dz/|z|) d2 p⊥ d p 2 2 which deﬁnes z as z = ( p0 + p3 )/2P and is very convenient for this discussion.129) and the discussion following it). i. 7. We assume that the hard scattering cross section can be written as a convolution of soft hadronic wave functions. 1979. A typical example. We expect that both p⊥ and p 2 are determined by soft binding effects. however. which are process-independent.

is large logarithms like α ln(q 2 / p 2 ) coming from collinear parton radiation where q 2 is the large scale of the process and p 2 is the parton mass. however. q) = dy Fi (y. In this second case it is usually convenient to let M 2 be the scale q 2 itself and simultaneously be the renormalization scale. Buras 1980. Dokshitzer.70) and (8. It is. The cross sections dσi are ﬁnite in ˆ the limit p 2 → 0 and M 2 is an arbitrary mass introduced to factorize out the long distance effects of perturbation theory.72) where 0 < y < 1 and the Fi are universal process-independent functions containing all the logarithms singular in the limit p 2 → 0 (mass singularities or more generally all long distance singularities).73) In the following we explicitly get (8. In the following we present a sample of lowest order QCD calculations and then brieﬂy mention possible strategies for all order extensions. q 2 /M 2 ) + O( p 2 /M 2 ) ˆ (8. This may concern the same process at different momentum transfers. for example.73) in the ﬁrst order perturbation theory. For instance. again for the lepton–parton deep inelastic scattering dσiQCD ( p. for example. Then. the theory has a clear predictive power by relating one cross section to another. absolutely crucial for any relevance of perturbative QCD to hard hadronic processes that the factorization theorem holds asymptotically to all orders in perturbation theory.71) and the theorem which has been proved to all orders in perturbation theory up to power corrections O( p 2 /q 2 ) states that. .72) the hadronic cross section can be written as a convolution dσ had = i f i ∗ Fi dσi = ˆ i ˜ ˆ Fi dσi (8. The idea of factorization can be illustrated by the identity [1 + α ln(q 2 / p 2 ) + · · ·] = [1 + α ln(M 2 / p 2 ) + · · ·][1 + α ln(q 2 /M 2 ) + · · ·] (8.72) and (8.g. scaling violation in the deep inelastic lepton–hadron scattering or two different processes that involve the same ˜ distribution functions Fi . For more details both technical and phenomenological the reader should consult some of the excellent review articles (e. Sachrajda 1983a. one can compute the cross section for lepton pair production in p p collisions using ¯ distribution functions deﬁned and measured in the inclusive lepto-production. p 2 /M 2 ) dσi (yp. Reya 1981. Dyakonov & Troyan 1980) on this subject. although we cannot calculate the cross sections dσ had or even dσ QCD from ﬁrst principles. Using (8.292 8 Quantum chromodynamics (Sachrajda 1983b).

7. ν). Writing 1 4 pol ˜ |M X |4 = L µν W µν (8.77) d3 pn (2π )4 δ (4) P + q − 2E pn (2π )3 n=1 d4 x exp(iq · x) P|Jµ (x)Jν (0)|P ˜ pn W µν n = pol (8. The unpolarized cross section for the process eN → eX has the form dσ = 1 d3 k Flux 2E k (2π)3 d3 pn 2E pn (2π )3 n=1 pn n N N ×(2π)4 δ (4) k + P − k − e4 1 q4 4 |M X |4 pol (8.78) The most general Lorentz and parity-invariant form of the tensor Wµν reads Wµν = −(gµν −qµ qν /q 2 )W1 +[Pµ −qµ (P ·q/q 2 )][Pν −qν (P ·q/q 2 )] W2 (8.79) m2 where Wi = Wi (q 2 . The cross section in the laboratory frame then reads (Flux = 4(P · k) = 4Em) dσ α 2 16E 2 dσ mEE = 4 [cos2 ( 1 θ )W2 + 2 sin2 ( 1 θ )W1 ] (8.5 Deep e–n scattering 293 8.8.2.5 Deep inelastic electron–nucleon scattering in ﬁrst order QCD (Feynman gauge) Structure functions and Born approximation We consider this process in the one-photon exchange approximation.74) where pn are four-momenta of the ﬁnal particles and X includes all the necessary spin indices for a given ﬁnal state (we neglect all particle masses with the exception of the initial nucleon mass). The kinematical variables are deﬁned in Fig.75) where the tensor L µν ≡ 1 2 Tr(/ γµ/ γν ) = 2(kµ kν + kν kµ − k · k gµν ) k k (8.80) = 2 2 d dE π d|q 2 | dν q 4m .76) comes from the lepton vertex we can rewrite the cross section as follows (q ≡ k − k ): dσ = where we deﬁne 4π W µν ≡ N e4 1 d3 k 4π L µν W µν q 4 Flux 2E k (2π )3 N (8. they are Lorentz scalars. i.e.

We also introduce another often used notation F1 = W1 .74)–(8.70).82) are written for the general case of n space-time dimensions. We get ﬁnally 1 Pµ Pν µν F1 = −g + 2x Wµν n−2 P ·q (8.81) region |q 2 | ∼ s are free of kinematical singularities when m → 0. Relations (8. F2 = P · qW2 /m 2 (8. Calculating the structure functions F1 and F2 we also neglect the nucleon mass as we are interested in the kinematical m 2 . (8.78) are real dimensionless functions which are measurable experimentally. k ) (8.81) and express the structure functions F1 and F2 in terms of the tensor Wµν as follows: 1 g µν Wµν = −(n − 1)F1 + F2 2x (8.70) the lepton–hadron scattering is an incoherent mixture of elementary scatterings dσ had = 0 1 dz i f i (z) dσiQCD ( p = z P.79). According to (8. Structure functions Fi deﬁned by (8.84) where dσiQCD is the cross section for the lepton scattering on a parton of type ‘i’. (8.85) .82) Pµ Pν 1 Wµν = −F1 + F2 2x P ·q 2x where x = −q 2 /2P · q is the Bjorken variable.80) we see that the structure functions W1 and W2 deﬁned by (8.294 8 Quantum chromodynamics This result follows directly from relations (8.84) is a short-hand notation for: in the case of the proton as a target: ¯ f i (z) dσiQCD = [ 4 (u p + u p ) + 1 (dp + dp ) + 1 (sp + sp )] dσ q + G p (z) dσ G ¯ ¯ 9 9 9 i (8. of course. remember that hadrons consist of quarks with different ﬂavours and of gluons. Eq. We assume that partons in a nucleon have some momentum distributions f i (z) dz. In the actual physical applications of (8. k.75).84) we must.75) and (8. where p = z P.83) Pµ Pν 1 F2 = 2F1 + 4x Wµν x P ·q We want to calculate the structure functions Fi in perturbative QCD under the parton model assumption (8. From the expression (8.79) and (8.

u ↔ d) gives u p (z) = dn (z) ≡ u(z) dp (z) = u n (z) ≡ d(z) sp (z) = sn (z) ≡ s(z) G p (z) = G n (z) ≡ G(z) and then for protons (8. There are several obvious constraints on the q N (z) determined by the quantum numbers of the nucleon.91) It is convenient to decompose the quark distribution functions into valence quark and sea quark distributions q(z) = qv (z) + qs (z) where we assume ¯ u s = u s = ds = ds = ss = ss ≡ S(z) ¯ ¯ (8.92) . For instance. The dσ G vanishes in the zeroth order in the strong coupling constant α but contributes in higher orders.88) ¯ { 2 [u p (z) − u p (z)] − 1 [dp (z) − dp (z)]} dz = 1 ¯ 3 3 etc.90) and for neutrons ¯ f i (z) dσiQCD = [ 4 (d + d) + 1 (u + u) + 1 (s + s )] dσ q + G(z) dσ G ¯ ¯ 9 9 9 i (8.86) where the numerical coefﬁcients are the squared electric charges of quarks.8. q N = q N (z) denotes the momentum distribution of quark q in hadron N and G N (z) denotes the gluon distribution in hadron N .5 Deep e–n scattering 295 in the case of the neutron as a target: ¯ f i (z) dσiQCD = [ 4 (u n + u n ) + 1 (dn + dn ) + 1 (sn + sn )] dσ q + G n (z) dσ G ¯ ¯ 9 9 9 i (8. Isospin symmetry (p ↔ n.93) (8.87) (8.89) ¯ f i (z) dσiQCD = [ 4 (u + u) + 1 (d + d) + 1 (s + s )] dσ q + G(z) dσ G ¯ ¯ 9 9 9 i (8. Once the squared electric charges of quarks are factorized out the dσ q is ﬂavour-independent (and averaged over colours of the initial and summed over colours of the ﬁnal partons). 1 0 1 0 [sp (z) − sp (z)] dz = 0 ¯ (strangeness) (charge) (8.

q 2 ) q2 1 p·q p·q q + pν − qν 2 F2 (y.3. We have q 4π Wµν = dn−1 p1 (2π )n δ (n) ( p + q − p1 ) 1 Tr[ p γµ p1 γν ] / / 2 0 (2π)n−1 2 p1 (8. (8. q 2 ) (8.2 (y.81) q Wµν = − gµν − qµ qν q F1 (y. q 2 ) as follows: F had (x. q 2 ) stands for F1 (y. 8. q 2 ) = 0 1 (dz/z) i f i (z)F q (x/z. q 2 ) − dσ en (x.95) One also often talks about ﬂavour non-singlet cross sections sensitive only to some combinations of the valence quark distributions.105) to calculate the cut diagram shown in Fig. q 2 ) or (1/y)F2 (y. for example.96) ¯ and ﬂavour singlet cross sections sensitive to the distributions 1 (qs + qs ) = S(z) 2 and G(z). dσ ep (x.79) and (8.98) is the Bjorken variable deﬁned for electron–quark scattering. In the Born approximation the quark structure functions are determined by elastic electron–quark scattering. q 2 ). . q 2 ) pµ − q µ 2 p·q q q y = −q 2 /2 p · q = x/z (8.296 8 Quantum chromodynamics Then for protons f i (z) dσiQCD = [ 4 u v + 1 dv + 4 S(z)] dσ q + G(z) dσ G 9 9 3 i (8. In the following we calculate dσ q in the zeroth and ﬁrst orders in the strong coupling α and study ﬂavour non-singlet hadron structure functions. q 2 ) = dz 1 [u v (z) − dv (z)] dσ q (z P.100) We can also use relation (8. We can deﬁne the quark structure functions for the elementary subprocess again by (8.94) and for neutrons f i (z) dσiQCD = [ 4 dv + 1 u v + 4 S(z)] dσ q + G(z) dσ G 9 9 3 i (8. q 2 ) 3 (8.97) where (8.99) where F(y.84). Using (8.77) and remembering about the normalizing factor 1/z P0 in dσ q we can express the q hadron structure functions in terms of F1.

The Bjorken parameter is therefore determined by the speciﬁc kinematic conﬁguration in our experiment. is the kinematic parameter of the reaction. Using (8.103) where x = −q 2 /2P · q and p = z P.102) where y = −q 2 /2 p · q. which together with q = k − k. in terms of the total nucleon momentum P. 8. Then dn p1 2 1 (2π)n δ (n) ( p1 − p − q)(−2πi)δ+ ( p1 )(−i)2 2 Tr[γµ i p1 γν p ] / / (2π)n (8.100)) that our result (8. Indeed. as before. It is clear from calculation of the elastic cross section (see (8. This will no longer be the case when we include radiative corrections.83) we easily get the quark structure functions Fi in the Born approximation q Wµν = − 1 4π (1/y)F2 = 2F1 = 2 p · qδ(2 pq + q 2 ) = δ(1 − y) q q (8. we know already that y = 1 for elastic scattering. For the hadronic structure functions we get ﬁnally had had (1/x)F2 (x. q 2 ) = 2F1 (x. We encounter here the so-called Bjorken scaling law: structure functions.100).104) is deﬁned. which in general can depend on y and q 2 . In terms of the Bjorken variable x deﬁned for the hadronic process we have 1 q x F x/z 2 z = 2F1 q x z =δ 1− x z (8.104) holds up to corrections O( p⊥ /z P). We arrive at a very interesting conclusion: it follows from (8.3.5 Deep e–n scattering 297 Fig. In particular we have learned from experiment this way that half of the nucleon momentum is carried by gluons. q 2 ) = (dz/z) i f i (z)δ(1 − x/z) = i f i (x) (8.8.104) that in measuring the hadron structure functions as a function of x we effectively study the momentum distribution of the hadron constituents. depend on y only.101) n n n (n) (we have used the identity 1 = [(d p1 )/(2π ) ](2π) δ ( p1 − p − q)) which indeed coincides with (8. .104) The Bjorken variable x in (8.

Deep inelastic quark structure functions in the ﬁrst order in the strong coupling constant We now calculate the dσ q in the ﬁrst order in α strong .5 show the diagrams for the virtual and real corrections. 8.4. p⊥ ﬁnite. .104) of the hadron structure functions and of the Bjorken variable x (for x = 0) is valid in a so-called inﬁnite momentum frame in which P → ∞. Figs. where p⊥ is the typical transverse momentum of partons in the hadron.5. 8. So the corrections are frame-dependent and the transparent physical interpretation (8. We use the notation of cut diagrams. 8.4 and 8. Fig.298 8 Quantum chromodynamics Fig.

8.5 Deep e–n scattering

299

Fig. 8.6.

where Wµν = 1 4π (8.105)

The calculation is performed in the Feynman gauge. The presence for each diagram of the group theoretical factor CF = 3 should be kept in mind. It is omitted in most 4 of the following expressions. All the quantities are for lepton–quark scattering but the superscript ‘q’ will also be omitted. Here and in the following we shall be mainly concerned with the quantity σ =− 1 gµν Wµν 1 + 1ε 2 (n = 4 + ε) (8.106)

The term p µ p ν Wµν in (8.83) is much simpler to calculate. The ladder diagram in Fig. 8.6 gives the following contribution to the cross section σ σL = 1 αµ−ε 1 + 1ε 2 d

1 2 2

Tr[ p γα ( p − /)γµ/ γ µ ( p − /)γ α ] / / l k / l

1 ( p − l)2

2

(8.107) with d

2

=

dn l dn k 2πδ+ (l 2 ) 2π δ+ (k 2 )(2π)n δ (n) (q + p − l − k) (2π)n (2π )n

**(contrary to the calculation of Section 5.5 we do not assume the gluon to be soft). The trace gives
**

1 2

Tr[· · ·] = 4(1 + 1 ε)2 W 2 [−( p − l)2 ] = 4(1 + 1 ε)2 2 2

1−x 2 Q [−( p − l)2 ] x (8.108)

where W = q + p, Q 2 = −q 2 and x = −q 2 /2 p · q is the Bjorken variable for the

300

8 Quantum chromodynamics

lepton–quark scattering. Using expression (5.90) for the two-body phase space in the ( p, q) cms we get the following: σL = α 1 8π × Q 2 1 − x ε/2 1 (1 + 1 ε)4 2 4π µ2 x (1 + 1 ε) 2 1−x 2 −1 dy [y(1 − y)]ε/2 Q x ( p − l)2

(8.109)

with y = 1 (1 − z) and z = cos θ, where θ is the polar angle between the three2 vectors l and p. We observe that in the ( p, q) cms 2 p · l = 2 p0l0 (1 − z) = p·W (1−z) = (Q 2 /x)y and the denominator in (8.107) gives a singularity at y = 0, i.e. for l p and any value of the momentum |l| allowed by the conservation law q + p = k +l (4|l|2 = −q 2 (1− x)/x in cms). This is the so-called mass singularity. We ﬁnally get σL = α 2π Q2 4πµ2

ε/2

(1 + 1 ε) 2

( 1 ε) (1 + 1 ε) 1−x 2 2 (1 − x) 1 (1 + ε) x (1 + 2 ε)

ε/2

(8.110) The pole in ε reﬂects the mass singularity regularized by working in n = 4 + ε dimensions. The behaviour near x = 1 is IR behaviour (|l| → 0 when x → 1). The cut ladder diagram in the Feynman gauge exhibits no IR singularity.† However, we expect that the full cross section for a single photon production is IR divergent. As we know from Section 5.5 in the properly deﬁned physical cross section the IR divergence is cancelled by virtual corrections to the elastic scattering, i.e. Aδ(1− x) dx + σ (x, Q 2 ) dx is ﬁnite. In order to demonstrate this cancellation explicitly we write the cross section σ in terms of the so-called regularized distribution. For any function f (x) we deﬁne the distribution f + (x) as follows f + (x) = f (x) − δ(1 − x)

0 1

dy f (y)

(8.111)

**i.e. we explicitly single out a possibly singular behaviour of the function f (x) as x → 1. We can then write σL = α 2π Q2 4πµ2
**

ε/2

1 + 1 + 1 γ δ(1 − x) 2 ε 1−x (1 − x) x

ε/2

2 + +1+γ ε

(8.112)

+

**The contribution of the longitudinal term pµ pν W µν can also be easily included.
**

† Note the difference with Section 5.5: now m = 0.

8.5 Deep e–n scattering

301

Fig. 8.7.

**The ladder diagram gives 1 F2 x
**

long

=

α 3 α 3x = [ δ(1 − x) + (3x)+ ] 2π 2π 2

(8.113)

and the contribution of the rest of the diagrams to pµ pν W µν vanishes. We proceed to calculate cut self-energy corrections. The contribution of the diagram in Fig. 8.7 reads σ

c

1 αµ−ε = 1 1 + 2ε

d

2

1 t2

2 1 2

Tr[ p γ µ/γ α / γα /γµ ] / t k t

(8.114)

As before, d 2 is the two-body (l, k) phase space and we work in n = 4 + ε dimensions. The trace gives 2(1 + 1 ε)2 2(2 p · l)2k · l = 2(1 + 1 ε)2 2(t 2 )2 y/(1 − x), 2 2 where y = 1 (1 + cos θ) and θ is the angle between vectors p and k in the (k, l) cm 2 frame. We see that the propagators are cancelled and we get ﬁnally σ

c

=α =

Q2 1 − x 1 8π 4πµ2 x

2 ε/2

ε/2

4(1 + 1 ε) 2 1−x x

1 (1 + 1 ε) 2

ε/2

1 0

dy [y(1 − y)]ε/2

y 1−x (8.115)

Q α 2π 4πµ2

(1 + 1 ε) 2

1 1−x

(2 + 1 ε) 2 (3 + 1 ε) 2

In the present case the integration over y is ﬁnite: there is no mass singularity. The result is, however, divergent for x → 1 (IR divergence). Using the regularized distribution we get: σ

c

=

Q2 α 2π 4πµ2

ε/2

1 1 1 1 + 2 (γ − 1) δ(1 − x) + ε 2 1−x

(8.116)

+

**Finally we consider the diagrams in Fig. 8.8 and the corresponding cross section σ
**

c

= 2αµ−ε

1 1 + 1ε 2

d

2 2 t (p

1 1 Tr[ p γ µ/γ α / γµ ( p − /)γα ] / t k / l − l)2 2

(8.117)

302

8 Quantum chromodynamics

Fig. 8.8.

**where d 2 is as before. Using γ µ / bcγµ = −2cb/ − ε/ bc and γ µ / bγµ = 4ab + a // a // a/ //a ε/ b we evaluate the trace and get a/
**

1 2

Tr[· · ·] = −{4(2 p · k)(2 p · t − 2t · l) + 1 ε4(2 p · t)(2k · p − 2k · l) 2 /t k / l + ε Tr[ p//( p − /)] + 1 ε2 Tr[ p// ( p − /)]} /kt / l 2 (8.118)

The last two terms do not contribute (one is exactly zero and the second gives ε2 (−8)( p · l)(k · l), i.e. both propagators are cancelled and there is no divergence to cancel ε 2 ). In the (k, l) cm frame we have 2 p · t = Q 2 /x, 2 p · k = 2 p · (t − l) = Q 2 /x + ( p − l)2 , 2t · l = 2k · l = (Q 2 /x)(1 − x) and σ

c

= 2α

1 1 + 1ε 2

d

2

4

ε x 1 x (1 + 1 ε) − 1+ 2 1−x y 1−x 2x

(8.119)

We see that the cut vertex has (in the Feynman gauge) both IR and mass singularity. Performing the integration over d 2 we obtain the following σ =2 + α 2π Q2 4π µ2

ε/2

c

(1 + 1 ε) 1 2 (1 + ε) 1 + ε

ε/2 +

2 ε

2

+

π2 − 1 δ(1 − x) 6 (8.120)

2 −1 ε

x 1−x 1−x x

Final result for the quark structure functions Virtual corrections to the considered cross section can be easily calculated using the results of Chapter 5 for the electron self-energy correction ( p 2 ) and the vertex correction (q 2 ). For instance, the ﬁrst two diagrams in Fig. 8.4 give together (in

8.5 Deep e–n scattering

303

n = 4 + ε dimensions) σ =− gµν W µν dn k −1 1 (2π )n δ (n) (k − p − q)2πδ+ (k 2 )(−i)2 = 1 1 4π (2π )n 1 + 2ε 1 + 2ε i × 1 Tr p γ µ/ γµ [−i ( p 2 = 0) p ] / k / 2 p / α 1 = ( p 2 = 0)δ(1 − x) = − δ(1 − x) (8.121) 2π ε

The same result holds for the next two diagrams of Fig. 8.4. Also, with ( p 2 ) changed into (q 2 ), each of the vertex correction diagrams can be calculated in a similar way. We are now ready to collect all the results (n = 4 + ε everywhere): Real σL = α 2π Q2 4πµ2

ε/2

1 + 1 + 1 γ δ(1 − x) + 2 ε

ε/2 +

2 +1+γ ε

× (1 − x) σ σ = =2 α 2π α 2π

1−x x

2

c

Q 4πµ2 Q 4πµ2

2

ε/2

1 1 1 1 + (γ − 1) δ(1 − x) + ε 2 2 1−x (1 + (1 + ε)

ε/2 + 1 ε) 2

ε/2

c

2 ε

2

+

π −1 6

2

2 + −1 ε 1 2 F x Virtual

long

x 1−x 1−x x

=

α 3 [ δ(1 − x) + (3x)+ ] 2π 2

1 δ(1 − x) 1+ε (8.122)

+

α 1 σ =2 − δ(1 − x) 2π ε σ =2 × α 1 1 (1 + 1 ε) Q 2 2 − 2π ε ε (1 + ε) 4π µ2 2 ε

2 ε/2

+

Q2 4πµ2

ε/2

+

π2 6

δ(1 − x)

1 (1 + 2 ε) (−1) (1 + ε) 1 + ε (8.123)

The ﬁrst two terms in σ correspond to the renormalized

UV

(Section 5.3) written

304

8 Quantum chromodynamics

**in n = 4+ε dimensions. The ﬁnal result for (1/x)F2 in the lepton–quark scattering reads 1 α F2 = CF x 2π 1 1 + 2 1−x 2 +γ ε + 3x
**

+

1 + x2 1−x Q2 4π µ2

+

ε/2

1 + x2 1−x ln − 1 + 2(1 − x) 1−x x + δ(1 − x) + O(α 2 ) (8.124)

We now add several important comments on the result obtained to clarify its physical sense. As we already know from Section 5.5, the IR singularities cancel between real and virtual corrections and they are not present in (8.124). However, this is not the case with mass singularities (due to the decay of a massless particle into two parallelly moving massless particles) which explicitly reﬂect themselves in the ﬁnal result as poles in ε. We certainly have to cut them off by some physical parameter before inserting result (8.124) into (8.99) to compare our calculation with experiment. The trick is to ‘factorize out’ the mass singularities by writing the result (8.124) in the limit ε → 0 as (see (8.71)) 1 F2 (x, Q 2 ) = x

1 0

dz z

x 1 , Q2, Q2 F2 (z, Q 2 ) + O(α 2 ) 0 0 z z Q2 α ln 2 2π Q0 1 + t2 1−t

(8.125)

**where Q 2 is an arbitrary parameter and from (8.124) 0 (t, Q 2 , Q 2 ) = δ(1 − t) + CF 0
**

+

+ O(α)

(8.126)

where the O(α) terms are ﬁnite when Q 2 → ∞. The meaning of (8.125) is clear: the structure function for any Q 2 is expressed in terms of the structure function for Q 2 and of the theoretically calculated function (x/z, Q 2 , Q 2 ) free of mass 0 0 singularity. Hadron structure functions; probabilistic interpretation To calculate hadronic structure functions we use the convolution (8.99) and get 1 had F (x, Q 2 ) = x 2

1 x

z x dz f (z) F2 , Q 2 z x z

1

1 = x

dt

x

(t, Q 2 , Q 2 ) 0

1 x/t

dz f (z)F2

x , Q2 0 zt

(8.127)

**where x = Q 2 /2Pq is the Bjorken variable for the lepton–hadron scattering. Consequently 1 had F (x, Q 2 ) = x 2
**

1 x

dz z

x 1 had , Q2, Q2 F (z, Q 2 ) 0 0 z z 2

(8.128)

8.5 Deep e–n scattering

305

and interestingly enough we again recover the result (8.125). The effect of the had intrinsic quark distribution in the nucleon is included in F2 (z, Q 2 ). We do not 0 have to know it theoretically to compare the theory with experiment as far as Q 2 evolution of the structure functions is concerned. By introducing an intermediate scale Q 2 we escape two problems: the intrinsic 0 parton distribution which originates from non-perturbative effects need not be known theoretically and the perturbative mass singularities are not present in had our ﬁnal formula. They have been absorbed into F2 (z, Q 2 ), which we take 0 as the experimentally measured structure function. Our theoretical prediction is had the evolution function (x/z, Q 2 , Q 2 ) and we do not worry that F2 (z, Q 2 ) is 0 0 actually singular in our calculation since this is due to our ignorance of how to include correctly the soft binding effects. It is also worth observing that the leading Q 2 dependence of the result (8.124) for the quark structure functions and therefore also for the hadron structure functions (8.128) can be interpreted in a very transparent probabilistic parton-like way. To this end let us formally rewrite (8.124) as follows 1 F2 (x, Q 2 ) = x

1 0

dz 1 x F2 (z, Q 2 )δ 1 − z z z

(8.129)

Formally, we obtain an expression for F2 which is the same as (8.104) for the structure function of a composite object consisting of a set of partons which scatter elastically at large Q 2 . We are tempted to interpret the quantity (1/z)F(z, Q 2 ) as the average number of partons, with momentum in the interval zp, (z + dz) p, seen in the initial quark with momentum p by a probe (electron) in the process of deep inelastic scattering with momentum transfer Q 2 . The theory predicts its evolution with Q 2 . Let us assume for the moment that our interpretation is sensible. had The hadronic structure function (1/x)F2 (x, Q 2 ) is then a convolution of the intrinsic parton momentum distribution in a hadron (Q 2 -independent) and the parton momentum distribution in a parton when probed with large Q 2 . Technically, the second effect is present in perturbation theory, whereas the ﬁrst one has its origin in non-perturbative binding effects. Is our interpretation of (8.129) sensible? It is for the leading Q 2 dependence. This point will become clearer in the next section when we repeat our calculation in a so-called axial gauge. The reason is that interference diagrams with a gluon emitted or absorbed by the ﬁnal quark contribute to the ﬁnal result in the present calculation. It turns out, however, that with the proper choice of gauge the leading effect comes from the ladder diagram and so our interpretation is possible. It applies, however, only to the leading Q 2 dependence because to interpret (8.129) in terms of the elastic scattering of a set of partons with momentum distribution (1/z)F2 (z, Q 2 ) we have to worry, as before, about corrections O( p⊥ /zp). But

306

8 Quantum chromodynamics

now the transverse momentum of a parton is limited only by kinematics and can be as large as O((Q 2 )1/2 ). For the probabilistic interpretation we have not only to stay in an inﬁnite momentum frame but also in a limited region of phase space ( p⊥ small). This we can do by writing ln(Q 2 /Q 2 ) = ln(ε Q 2 /Q 2 ) + ln(Q 2 /ε Q 2 ), 0 0 where ε is some ﬁxed arbitrarily small number. The ﬁrst logarithm comes from the region where p⊥ (ε Q 2 )1/2 is small and at the same time it gives the leading Q 2 dependence correctly (for Q 2 → ∞). The function P(z) = 1 + z2 1−z (8.130)

+

is called the Altarelli–Parisi function. One often considers the Q 2 evolution of moments of the structure function, e.g. M N (Q 2 ) =

0 1

dx x N −1 F(x, Q 2 )

(8.131)

**where again F(x) = F1 (x) or F(x) = (1/x)F2 (x). From (8.128) and (8.126) one gets MN =
**

0 1

dx x N −1

0

1

1

dz

0

dt

(t, Q 2 , Q 2 )δ(x − zt)(1/z)F2 (z, Q 2 ) 0 0 (8.132) α Q2 PN ln 2π Q2 0

or in the leading Q 2 dependence approximation M N (Q 2 ) = M N (Q 2 ) CF 0 +1 (8.133)

**where PN is the N th moment of the Altarelli–Parisi function: PN =
**

0 1

dx x N −1

1 + x2 1−x

+

=

0

1

dx

1 + x 2 N −1 − 1) (x 1−x

(8.134)

8.6 Light-cone variables, light-like gauge In this section we describe brieﬂy the calculation of dσ q given by the diagrams in Figs. 8.4 and 8.5 using the light-cone parametrization of the momenta and the light-like gauge. To be speciﬁc let us consider ﬁrst the ladder diagram in Fig. 8.6. The gluon propagator in the n 2 = 0 gauge is −i gµν − lµ n ν + lν n µ δαβ l ·n (8.135)

8.6 Light-cone variables, light-like gauge

307

**and we parametrize the momenta as follows p = (P + p 2 /4P, O⊥ , P − p 2 /4P)
**

2 2 l = (z P + l⊥ /4z P, l⊥ , z P − l⊥ /4z P) 2 2 t = (y P + (t 2 + t⊥ )/4y P, t⊥ , y P − (t 2 + t⊥ )/4y P)

(8.136)

q = (η P + q 2 /4η P, O⊥ , η P − q 2 /4η P) y = 1 − z,

2 t 2 z = −l⊥

For the gauge four-vector n we take n = ( p · n/2P, O, − p · n/2P) (8.137)

We can interpret (8.136) and (8.137) noticing that P is an arbitrary boost parameter deﬁning the frame (invariant quantities are P-independent) so that we may start, for example, from the p rest frame (4P 2 = p 2 ) and make an inﬁnite boost P → ∞ in the direction opposite to n. We observe that:

(i) l · n/ p · n = z; (ii) η = −x for p 2 = 0 because 2 p · q = q 2 /η;

2 2 (iii) 1 − z = x + O(l⊥ (−q 2 )) because ( p − l)2 = −l⊥ /z and ( p − l + q)2 = 0.

**The two-body phase space d
**

2

=

dn l dn k 2πδ+ (l 2 ) 2π δ+ (k 2 )(2π )n δ (n) (q + p − l − k) (2π)n (2π)n

(8.138)

**can be written as d From
**

2 δ+ (( p − t)2 ) = δ(−(1 − y)t 2 /y − t⊥ /y), 2

=

dn t δ+ (( p − t)2 )δ+ ((q + t)2 ) (2π)n−2

(8.139)

( p − t)0 > 0

2 δ+ ((q + t)2 ) = δ((y − x)(t 2 /y − q 2 /x) − xt⊥ /y), (q + t)0 > 0

(8.140)

one concludes that x <y<1 y − x q2 t2 = , 1−x x

q2 < t2 < 0 x

(8.141)

308

8 Quantum chromodynamics

**Furthermore, dn t 1 dy n−2 = dt 2 d t⊥ n−2 n−2 (2π) (2π ) 2|y| and dn−2 t⊥ =
**

n−3 t⊥ dt⊥ d n−2

(8.142)

=

π (n−2)/2 1 (n − 2) 2

2 2 dt⊥ (t⊥ )(n−4)/2

(8.143)

2 So ﬁnally, integrating over d(t⊥ /y) by means of the δ(( p − t)2 ), we get (n = 4 + ε)

2

=

1 8π 1 8π

1 (1 + 1 ε) 2

0

1

dt 2

q 2 /x x

dy (1 − y)|t 2 | y y 4π

ε/2

δ t 2 (1 − x) −

ε/2

(y − x)q 2 x (8.144) (8.145)

=

1 1 1 (1 + 2 ε) 1 − x

1

dy

x

(1 − y)(y − x) −q 2 x(1 − x) 4π

For the Feynman part of the ladder diagram, using (8.107) and (8.108) together with (8.144) we recover the previous result (8.110). The axial part of the ladder diagram

A σL = CF

1 αµ−ε 1 + 1ε 2 d

1 2 2

×

1 1 1 1 / // l Tr p/ γµ (/ + / )γ µ n + p n γµ (/ + / )γ µ / /l t q t q / t / t / t / t

−

1 l ·n (8.146)

**can be calculated using the expansion
**

2 t 2 + t⊥ n / /= yp + t / + /⊥ t 2y p · n

(8.147)

**/ −q 2 n / = −x p + q / x 2p · n and (8.144). After some effort one gets
**

A σL = CF

α 2π

−q 2 4π µ2 2 +γ ε

ε/2

2

(1 + 1 ε) 2 (1 + ε)

2 ε

ε/2

2

−

2 π2 + ε 6 (8.148)

×δ(1 − x)

2x 1−x 1−x x

+

8.6 Light-cone variables, light-like gauge

309

**We see that
**

F σL

+

A σL

= CF

α 2π

2 +γ ε

1 + x2 1−x

+

+ A(ε)δ(1 − x) + · · ·

−q 2 ε/2 4π µ2 (8.149)

where the neglected terms do not have a pole in ε and are polynomials in x and logarithms in x and 1 − x. Thus in the axial gauge the ladder diagram by itself gives the leading logarithmic contribution (8.126) to the quark structure function. (Remember that we are interested in the ﬁnite x = 0 region so ln x and ln(1 − x) can be dropped.) We expect therefore that the contribution from the remaining diagrams with real emissions is non-leading. By explicit evaluation of the axial parts of these amplitudes one gets: σ Ac = so that

F A F σL + σL + σ Fc + σ Ac + σ Fc + σ Ac = σL + σ Fc + σ Fc

1 2 F x

long

= 0,

A σ Ac = −σL

(8.150)

and (σ Fc + σ Ac ) does not contribute to the leading logarithm, i.e. it is free of mass singularities. Actually, one can support this conclusion by the following general argument. The leading logarithm reﬂecting the presence of mass singularity obviously comes from the integration over t 2 in the region εq 2 < t 2 < 0, i.e. from the region where t and l are almost collinear with p: t ∼ yp. Part of the = Dirac algebra involves 2y lµ p /γµ p ∼ y p γµ p = y2 pµ p = t /= / / / / 1−y

(8.151)

and since in the axial gauge the on-shell gluons have only physical polarization the contraction with the gluon tensor dµν (l) = λ=1,2 ε ∗µ (λ, l)ε ν (λ, l), l 2 = 0 gives lµ dµν (l) = 0 (8.152)

Actually, one can check that the vertex vanishes linearly in θ, where θ is the gluon emission angle. So the ladder diagram gives d

2

(1/t 2 )2 θ 2 ∼

dθ θθ 2 /(θ 2 )2 ∼ ln θ

(8.153)

310

8 Quantum chromodynamics

**whereas the cut-vertex diagram gives d
**

2

(1/t 2 )θ ∼

dθ θ(θ/θ 2 )

(8.154)

and hence no logarithm. The poles in ε in the coefﬁcient of the δ(1 − x) are cancelled by the virtual corrections. Correspondingly the situation with real emissions in the axial gauge, mass singularity is present only in the quark self-energy correction. It is instructive to calculate the virtual corrections in the light-like gauge and using the light-cone parametrization of the momenta. This we leave as an exercise for the reader limiting ourselves to the following two remarks. Take, for instance, the self-energy correction = −i where can be decomposed as follows = Ap + B / with n / 2p · n pn p // / A + B = 1 Tr 2 2p · n A = 1 Tr 2 The full propagator reads G( p 2 ) = i p / n / B − 1 − (A + B) p 2 1 − A 2 p · n ip / = [1 + (A + B) + · · ·] 2 + · · · p i p− / = (8.156) 1 p2 (8.157) p2 n / 2p · n (8.155)

(8.158)

Apart from IR and mass singularities is also UV divergent, i.e. contains poles in εUV where n = 4 − εUV . The necessary counterterms are of the form a p + bn / // The term proportional to n reﬂects the pathology of the n 2 = 0 gauge because / d4l 1 (n · l)l 2 ( p − l)2

In fact this result has the following structure α σL ( p. However. 8.8.6 Light-cone variables. we can replace 2 the denominator by δ(z(1 − z)( p 2 − l 2 /z) − l⊥ ). q) (8. Going to the complex l⊥ -plane.† and it makes the general proof of the renormalizability in this gauge difﬁcult. Our second remark is that an explicit calculation of the virtual corrections using the light-cone parametrization of the momenta is called the Sudakov technique. see Fig.9.9. It consists of noting that the integral dn l 1 1 ∼ (2π)n l 2 + iε ( p − l)2 + iε dl 2 dz 2 2 (n−4)/2 dl (l ) l 2 2|z| ⊥ ⊥ z × 2 − l 2 /z) − l 2 + iε z(1 − z)( p ⊥ (8. q) = δ(1 − x/y) † In the Feynman parametrization such integrals are discussed in Appendix B. 8. light-like gauge 311 Fig. . Since in the leading logarithmic approximation. since there is a pole at 2 l⊥ = z(1 − z)( p 2 − l 2 /z) + iε (8. its probabilistic interpretation is quite natural. q ) = CF 2π 2 q2 p2 dt 2 t2 1 x dy 1 + y 2 y 1−y + σL (yp. which is at inﬁnity. should be convergent on dimensional grounds. vanishes. in order-by-order calculation one can check explicitly that such terms cancel out in the full vertex function. only the ladder diagram contributes to the ﬁnal result.159) 2 2 can be evaluated by integrating over dl⊥ ﬁrst.161) where σL (yp.160) and the integral over the semi-circle.

Another one. this extension is. Parisi & Petronzio (1979)) and to the so-called doubly-logarithmic region of the structure and fragmentation n functions (x ∼ 0). This coincides with (8. . In the context of QCD pioneered by Furma´ ski. for instance. with the OPE.129). Dokshitzer.7 Beyond the one-loop approximation Since αstrong ∼ O(1) the relevance of perturbative QCD to the description of the actual experimental data often relies on our ability to go beyond the one-loop approximation. It is very interesting that these results can also be given a probabilistic interpretation. Dokshitzer. For the regions x = 0 and x = 1. Ciafaloni.312 8 Quantum chromodynamics and we have chosen to regularize the mass singularity with p 2 = 0. Dyakonov & Troyan 1980).127)–(8.129) and Pq→q (y) 1 + y2 1−y + can be interpreted as the variation of the probability density per unit t 2 of ﬁnding a quark in a quark with fraction y of its momentum. important for studying new particle production in hard collisions in very high energy accelerators. whenever possible. 8. the summation of the so-called leading logarithms α n lnn (q 2 / p 2 ) and also of the next-to-leading α n lnn−1 (q 2 / p 2 ) terms has been explicitly performed and compared. (8. Let us only brieﬂy indicate the techniques that have been developed to include higher order effects in αstrong . applicable to a wider class of problems. consists of summing inﬁnite sets of Feynman diagrams by cleverly picking out the leading contribution. Dyakonov & Troyan (1980). Of course. neglecting ln x and ln(1 − x) terms as compared to ln(q 2 / p 2 ) terms. i. One approach is based on the OPE and on the RGE for the Wilson coefﬁcient functions (Chapter 7). Petronzio & = Pokorski (1979) and Bassetto. the next-to-leading one and so on (Gribov & Lipatov 1972. Marchesini & Mueller (1982). our previous discussion concerning the hadron structure functions. is still valid. Further very interesting and more recent progress consists of extending the QCD perturbative techniques to processes where there are two types of large logarithm (for example. In the leading logarithm approximation it is concisely formulated by means of the Altarelli–Parisi equations (Altarelli & Parisi 1977). Ciafaloni & Marchesini (1980) and fully clariﬁed by Bassetto. Llewellyn Smith 1978. This vast area of technical and phenomenological activity goes far beyond the scope of the present book and can by itself be a subject of an extended monograph.e. In addition we must also include in considerations the singlet and the gluon structure and fragmentation functions.

ladder diagrams with vertex and self-energy insertions. Thus in the leading logarithm approximation the cross section is a generalization of (8. First we note that CF 2π n |q 2 | p2 2 dtn 2 α(tn ) 2 tn 2 dtn 2 α(tn ) 2 tn × |q 2 | p2 1 2 α(tn−1 ) · · · → (CF /2)n 2 n! tn−1 |q 2 | dt 2 ln(−q 2 / 1 n−1 2 α(tn−1 ) · · · = ln 2 n! ln(− p 2 / tn−1 p2 p2 2 |tn | 2 dtn−1 2 ) 2) n − CF n b1 (8. 8. i. It turns out that the dominant diagrams in the axial gauges are the generalized ladder diagrams. it can be shown that the generalized ladder can be replaced by a ladder with the coupling constant α(ti2 ) at each vertex.162) we can easily calculate the moments of (1/x)F2 .10.162) 2 |t2 | Using (8.161) and reads n σL = |q 2 | p2 2 2 dtn α(tn ) CF 2 tn 2π 2 |tn | 2 dtn−1 2 tn−1 CF p2 2 α(tn−1 ) ··· 2π × y1 2 2 dt1 α(t1 ) 1 dy2 y2 CF dy1 Pq→q (y1 ) Pq→q ··· 2 2π x y1 y1 t1 x p2 yn−2 yn−1 1 dyn−1 dyn yn−1 yn Pq→q Pq→q × δ(yn − x) yn−2 yn−2 yn−1 yn−1 yn x x (8.163) .e.7 Beyond the one-loop approximation 313 Fig.8. Let us be a little bit more explicit concerning the summation of the leading logarithm terms α n lnn (q 2 / p 2 ) generated by mass singularities in the quark structure function. Fig. The dominant region of integration is where | p2 | 2 |t1 | 2 |t2 | ··· 2 |tn | |q 2 | Furthermore. 8.10.

165) M N (Q 2 ) ln(Q 2 / = M N (Q 2 ) ln(Q 2 / 0 0 where 0 −γ N = CF 1 0 ) 2) 2 0 γ N /b1 . . . For instance.e.. of course. Sachrajda (1983b)). z n = yn 1 i 0 dx x N 0 i dz i Pq→q (z i ) 1 δ(x − z 1 . z n = yn /yn−1 z 1 z 2 .166) dz z N −1 1 + z2 1−z + =− N −2 1 2 2 +4 1− 3 N (N + 1) j j=2 (8. q 2 ) x 1 ln(−q 2 / 2 ) ln n! ln(− p 2 / 2 ) 2 − CF b1 1 0 dz z N −1 Pq→q (z) n = and ln(−q 2 / ln(− p 2 / ) 2) −(CF /b1 ) 1 N −1 P q→q (z) 0 dz z (8. we get 1 )]. Changing variables 3 z 2 = y2 /y1 . for example. . −b1 = . Q 2 = −q 2 (8. . . in the process qq → l+ l− X these divergences arise at the two-loop level. . i. However.. 11 2 − 1 n f . Comments on the IR problem in QCD The Bloch–Nordsieck theorem in QED states that IR divergences cancel for physical processes.. for processes with an arbitrary number of undetectable soft photons.124) obtained using the OPE. . zn dz i z iN −1 Pq→q (z i ) (8. it has been shown by means of an explicit example that such a cancellation does not occur for processes with two coloured particles in the initial state even if an average over different colour states is taken.167) The same result holds. In QCD it has been demonstrated that IR divergences cancel for processes with none or one coloured particle in the initial state (for a review see. ﬁnally MN = 0 1 = n 1 dx x N F2 (x. for the moments of the hadron structure function 0 and it coincides with (7. We see that γ N are the g 2 coefﬁcients in the expansion of the anomalous dimensions of the operators in (7. z n ) = z1 .314 8 Quantum chromodynamics 2 where α(t 2 ) = −2π/[b1 ln(t 2 / z 1 = y1 . Of course.99) in the coupling constant.164) Therefore.

U (x) belongs to N = 1 class.168) N= 2π −∞ α(x) dx Check that N = 1 for α(x) = exp[iπ x/(x 2 + a 2 )1/2 ]. In the context of the IR problem in QCD it is worth recalling the Kinoshita– Lee–Nauenberg (KLN) theorem (Kinoshita 1962. in principle. belongs to continuously deformable to = 0. spoil the factorization property discussed in Section 8. Fortunately enough it has been shown that such large IR logarithms are suppressed by at least one power of Q 2 and asymptotically the factorization property is still maintained. Clearly.Problems 315 in physical processes coloured partons never appear as free on-shell particles and these divergences are regulated by a mass-scale characteristic of the hadronic wave function which is of the order of the inverse hadronic radius. Lee & Nauenberg 1964).4 which states that all long distance effects can be absorbed into the quark and gluon distribution functions and is crucial for the sensible use of perturbative QCD. (b) For S 3 → SU (2) the transformation Aµ → exp(− 1 i 2 with a (x) ≡ 0 or N = 0. as a consequence of unitarity. transition probabilities are ﬁnite when the sum over all degenerate states (ﬁnal and initial) is taken. The physical meaning of this theorem seems. Problems 8. This is true order-by-order in perturbation theory in bare quantities or if the minimal subtraction renormalization is used (to avoid IR or mass singularities in the renormalization constants). however. However. then a (x) a (∞) a a a σ )(Aµ + ∂µ ) exp( 1 i 2 a a σ ) ≡ 0. residual large logarithms remain and could. with a continuously deformable to it. to be unclear. The KLN theorem is a mathematical theorem saying that.1 (a) Consider mappings of the real axis with the end-points identiﬁed into a set of unimodular complex numbers z = exp[iα(x)]. Check that with (x) = 2π x a /(x2 + a 2 )1/2 a (x) or. What is U∞ ? Can such a transformation be built up from inﬁnitesimal transformations a (x) → δ a (x) for all x? . In QED the cancellation of IR singularities occurs separately in the ﬁnal state and there is no need to invoke the KLN theorem. Check that the winding number can be written as ∞ −i dα(x) 1 dx (8. In non-abelian gauge theories this does not happen but the initial state is determined by the non-perturbative conﬁnement effects and the relevance of the KLN theorem is doubtful.

Use the shifted action S[Aµ + Aµ ]. show that exp(iQ G )W (A) exp(−iQ G )W (A) contradicting the correct result exp(iQ G )W (A) exp(−iQ G ) = W (U −1 AU − U −1∇ U ) = W (A) + N (Use δW (A)/δ Ai = (1/8π 2 )Bia .2 Show that gauge transformations which belong to the homotopy class N = 0 cannot be built up by iterating the inﬁnitesimal gauge transformations.64). d3 x K 0 i d3 x Dab b · i a .3 Derive the Feynman rules for calculating the 1PI Green’s functions generated by ˜ [0. the gauge-ﬁxing term (8. Aµ ] deﬁned in Section 8.39) with F α → ˜ F α [Aµ . Aµ ] and δ Aµ given by (8.15) and (3.316 8 Quantum chromodynamics 8. Prove (8.) a 8.24) and the Faddeev–Popov term obtained from (3.34) and complete the calculation of the β-function in Yang–Mills theories in the one-loop approximation (Abbott 1982). For this assume that a ‘large’ (N = 0) gauge transformation U = exp(− 1 i 2 where a (x) a a σ ) approaches a non-zero limit at spatial inﬁnity is given by the operator exp(iQ G )A exp(−iQ G ) = U −1 AU − U −1∇ U where exp(iQ G ) = exp −i Applying it to W (A) = where K 0 is given by (8.2.22).

to ﬁnd the dynamical theory of the fermion mass matrix which might eventually lead to the true theory of fundamental interactions.1) . the masses of the light quarks are small in a sense which will be speciﬁed later on and we shall assume in the following that the theory with the quark masses neglected is a correct ﬁrst approximation. its importance for particle physics originates in the idea that SU (2)L × SU (2)R is an approximate symmetry of the strong interactions. In fact. Consider the general massless Dirac lagrangian of fermion ﬁelds ri coupled to gauge ﬁelds Aα µ L= r n(r ) i ¯ ri iγ µ Dµ 317 ri .1 Chiral symmetry of the QCD lagrangian We begin our discussion by considering the QCD lagrangian with two quarks u and d in the theoretical limit in which the quark mass parameters in the lagrangian are put equal to zero. spontaneous symmetry breaking Chiral symmetry plays an important role in the dynamics of fermions coupled to gauge ﬁelds. Much of the progress of particle physics since 1960 is related to understanding the phenomenological aspects of spontaneous and explicit chiral symmetry breaking. 9. The successes of gauge theories in describing the fundamental interactions further stimulate interest in chiral symmetry and this is for at least two reasons: one may hope ﬁrstly to understand the underlying mechanism of spontaneous chiral symmetry breaking in the strong interactions and in gauge theories in general and. realized as spontaneously broken symmetry with pions being Goldstone bosons in the symmetry limit. secondly. Historically. = (x) (9. This chapter serves as an introduction to the subject of chiral symmetry and to techniques used in its exploration.9 Chiral symmetry.

2). 2. i = 1. 2. 2. The axial currents read 1 Aa = ¯ γµ γ5 2 σ a µ (9. For QCD with two ﬂavours r ≡ 3. Introducing = L= ¯ 1 2 we can write / iD / iD (9. 2 and i (we now omit the subscript r ) denotes a vector in the colour space.3) where a = 0.7) . Indeed.dimensional ﬂavour space.2) also has another global symmetry. spontaneous symmetry breaking where Dµ = ∂µ + ig Aα trα µ and the index α runs over the generators of the gauge group.4) and the SU (2) × U (1) invariance gives the isospin and the baryon number Qa = d3 x ¯ (x)γ0 1 σ a (x). γµ } = 0 the exponentials again cancel when (9.2) which is invariant under SU (2) × U (1) transformations 1 → exp(−iα a 2 σ a ) ¯ → ¯ exp(+iα a 1 σ a ) 2 (9.6) is introduced into lagrangian (9. the massless theory (9. involving γ5 1 → exp(−iα a 2 σ a γ5 ) ¯ → ¯ exp(−iα a 1 σ a γ5 ) 2 (9. The corresponding conserved currents are a 1 Vµ = ¯ γµ 2 σ a (9.1) is invariant under the general unitary transformation in the two. 2 a = 0. 1. 2. 3 and σ 0 = 1. 3. since {γ5 . 1.318 9 Chiral symmetry. 3 and σ a are Pauli matrices for a = 1.5) conservation. 1. However. 3 (9. In this case lagrangian (9. the matrices trα represent these generators in the representation r of the gauge group to which the fermions are assigned and the index i runs over all n(r ) ﬂavours belonging to the same representation r .6) where a = 0.

12) j i i i R → R = (UR ) j R where indices i.2) we introduce some useful notation. 3 L.9. An 5 5 5 5 important property of the generators Q a and Q a is that they transform into each L R other under the parity operation: P Q a P −1 = Q a . Before we proceed to explore the chiral symmetry of the massless QCD lagrangian (9.8) together with the charges (9. Q b ] = 0.11) means invariance under the transformations j i i i L → L = (UL ) j L (9.R ] = iε Q L. L + R = 1 (1 − γ5 ) 2 + 1 (1 + γ5 ) 2 = PL + PR (9. a = 0. Q b ] = [Q a .9) [Q L. Q L.11) (1 is unit matrix in the ﬂavour space). 3 L R 5 5 2 2 a b abc c (9. their commutators exist provided we ﬁrst commute under the space integrals and then do the integrations.R .5) and (9. 1. Q b ] = iεabc Q c ). b. .2) can then be rewritten as / l L = ¯ L iD 1 L R (9. l Chiral symmetry of the lagrangian (9. Even so.L L. b = 0.R transformations.10) has a clear interpretation in terms of the free particle solutions of the massless Dirac equation. j are ﬂavour indices and the unitary transformations UL and UR read a a UL = exp(−iαL TL ) (9.8) and the equal-time canonical commutation relations between the ﬁelds) Q a = 1 (Q a − Q a ). a. [Q a .1 Chiral symmetry of the QCD lagrangian 319 and the axial charges† Qa = 5 d3 x ¯ (x)γ0 γ5 1 σ a (x) 2 (9. This remark is relevant for the case of spontaneous symmetry breakdown. 2. 3 [Q a .R R (for completeness: [Q a . a. c = 1.R . Q a = 1 (Q a + Q a ).13) a a UR = exp(−iαR TR ) † Charges may be formal in the sense that the integral over all space may not exist. ‡ Decomposition (9. 2. This is discussed in Appendix A.5) form the chiral SUL (2) × SUR (2) × U (1) × UA (1) algebra (as can be seen using (9. hence the name chiral R. Q b ] = iεabc Q c .10) PL PR = 0 / l + ¯ R iD 1 and the lagrangian (9. Any spinor ﬁeld can be decomposed into ‘chiral’ ﬁelds‡ as follows = where 2 2 PL + PR = 1. 2. 1.

320 9 Chiral symmetry.e. 513): ‘a symmetry of the vacuum is a symmetry of the world’ or Q i |0 = 0 ⇒ [Q i . Some useful properties of the chiral ﬁelds L and R are given in Appendix A. in grand uniﬁcation schemes) only L (or only R ) ﬁelds can be placed in irreducible multiples of the gauge group. At this level we should also conclude that because of the U (1) × UA (1) invariance the fermion numbers of R and L (corresponding to U (1) ± UA (1) charges) are separately conserved. Now let us imagine in addition that the physical vacuum deﬁned by the minimum of the expectation values of the hamiltonian 0|H |0 = H min is invariant under the chiral transformations. For the time being we will be concerned with the chiral SU (2) × SU (2) only. In many problems we work with one type of chiral ﬁeld R or L only. For instance in ‘chiral’ gauge theories such that R and L transform differently under the gauge group (for example.15) We can invoke Coleman’s theorem (see. Indeed. Itzykson & Zuber (1980). for example. ¯ R γµ L currents vanish and we would not be able to introduce symmetry operations between R and L .2 Hypothesis of spontaneous chiral symmetry breaking in strong interactions We have assumed that the massless QCD with its chiral SUL (2) × SUR (2) global symmetry is a good approximation to the real world.16) and the transformation properties of the ﬁeld . i. The problem of the UA (1) symmetry will be brieﬂy discussed in Chapter 13. We see that multiplet L ( R ) is a singlet with respect to SUR (2)(SUL (2)) transformations. Another notation worth mentioning is the one using the two-component Weyl spinors.L be seen by constructing them from the vacuum by means of the ﬁeld operators and then using the properties (9. The description in terms of. We refer the reader to Appendix A for further details on both subjects.14) Then we can formulate our theory in terms of L and ˜ L . The matrices T form a representation of the corresponding generators Q in the space of fermion multiplets .16) to conclude that the physical states in the spectrum of H can be classiﬁed according to the irreducible representations of the chiral group generated by Q a (as can R. H ] = 0 (9. the left-handed ﬁelds can be obtained if we replace R by their charge conjugate partners ( R) C =( C )L = ˜ L df (9. for example. 9. spontaneous symmetry breaking a a with TL = T a ⊗ PL and TR = T a ⊗ PR . p. Q a |0 = Q a |0 = 0 R L (9.

or Nambu–Goldstone realization (Nambu 1960.2 Spontaneous chiral symmetry breaking 321 operators under the chiral group). the problem of calculating the spontaneous chiral symmetry breaking in QCD has not yet been satisfactorily solved. Unfortunately.15) and (9.17) is no longer free for us. The proof of the theorem will be given later. scalar or pseudoscalar.17) and explore its consequences in the hope of ﬁnding support for the assumption. The most important consequence of the spontaneous symmetry breakdown is summarized in Goldstone’s theorem: if a theory has a global continuous symmetry of the lagrangian which is not a symmetry of the vacuum there must be a massless (Goldstone) boson. 5 Q a |0 = 0 (9. It is easy then to see that all isospin multiplets would have to have at least one degenerate partner of opposite parity.17) One then talks about spontaneous breakdown. Let the state | be an energy and parity eigenstate: H| = E| and P| =| Since Q L and Q R commute with the hamiltonian we also have H QL| In addition P QL | R and therefore the state | degenerate with | 1 = √ (Q R − Q L )| 2 = P Q L P + P| R = QR| L = E QL| and H QR| = E QR| is also a parity eigenstate P| = −| The presence of parity degenerate states is not a general feature of the known hadron spectrum. Therefore. in principle. of the chiral symmetry SUL (2) × SUR (2) into isospin symmetry SU (2) generated by Q a . A very simple heuristic argument is as follows: . Instead we assume that the physical vacuum is not invariant (but the hamiltonian remains invariant) under the full chiral group and Q a |0 = 0. tell us whether the symmetry or part of it is spontaneously broken or not. corresponding to each generator (with its quantum numbers) which does not leave the vacuum invariant. Of course. if we wish to maintain our assumption about massless QCD being a good approximation to the strong interaction we must give up assumption (9. The theory itself should.9. It is then interesting to begin as we do: assume the spontaneous breakdown (9.15). Goldstone 1961). given a theory like massless QCD the choice between (9.

18) and (9.322 9 Chiral symmetry. spin and parity of Q a must exist to account for the degeneracy. the π ± decay constant √ is 2 f π ). Another consequence of spontaneous chiral symmetry breaking which agrees with the experimental data is the Goldberger–Treiman relation. π → lν. That they are not exactly massless (but nevertheless much lighter than other hadrons) is attributed to an explicit chiral symmetry breaking by the quark mass terms in the lagrangian. The derivation of the Goldberger–Treiman relation in the limit of exact chiral symmetry (with quark masses and pion mass neglected) which is spontaneously broken. H ] = 0. from Lorentz covariance. π (2) = −(i/ 2)(π + − π − ). Assuming C invariance and isospin † This interpretation requires some caution (Bernstein 1974) since Q a |0 is not a normalizable state. spontaneous symmetry breaking if Q a |0 = 0 then there is a state Q a |0 degenerate with the vacuum† because 5 5 [Q a . . A massless boson with internal quantum numbers. This is based on the identiﬁcation of the axial current of the two-ﬂavour QCD with the axial current of the weak interactions for the ﬁrst generation of quarks. In the Born approximation for the electroweak decays (see Chapter 12) the pion decay.20) This is Goldstone’s theorem: f π = 0 therefore m 2 = 0.18) √ The constant f π is called the pion decay constant and π (1) = (1/ 2)(π + + √ π − ). amplitude is proportional to the matrix element (9. It means 5 that U = exp(−iα a Q a ) is not a unitary operator. By successive application of Q a one can construct an inﬁnite number 5 5 of degenerate states. In our approximation the axial current is conserved: ∂ µ Aa (x) = 0 and from (9. Nevertheless.19) we get µ 0|∂ µ Aa (x)|π b (q) = exp(−iq x) f π m 2 δ ab µ π (9. Restricting ourselves to the SU (2) × SU (2) case we see that the triplet of pions has the right quantum numbers for such an interpretation. 5 If the exponentials are expanded. then goes as follows: consider the matrix element 0|Aa (x)|π b (q) = exp(−iq x) 0|Aa (0)|π b (q) µ µ where.18) (we neglect the Cabbibo angle). expressions like U AU + can be meaningful. therefore f π = 0. We assume the axial charge does not annihilate the vacuum. 0|Aa (0)|π b (q) = i f π qµ δ ab µ (9. π 3 = π 0 (therefore. the resulting commutators may be well deﬁned if we ﬁrst commute currents and then integrate over space. This identiﬁcation is explicit in the SU (2) × U (1) Glashow–Salam–Weinberg theory of weak interactions. Next consider nucleon π matrix elements of the axial current (β-decay). 5 Very plausible candidates for Goldstone bosons in the strong interactions with spontaneously broken chiral symmetry are pseudoscalar mesons.19) (9.

If pions are Goldstone bosons they contribute such a pole as can be seen from (9.20). Relation (9.2 Spontaneous chiral symmetry breaking 323 invariance one has ¯ N( p )|A− (x)|P( p) = exp(iq x)u( p ) 1 σ − [γµ γ5 gA (q 2 ) + qµ γ5 h(q 2 )]u( p) µ 2 (9.21) where √ σ − = (1/ 2)(σ 1 − iσ 2 ). from β-decay gA (0) = 1. the PCAC approximation has been formulated in another.24) happens to be satisﬁed within 7%. and it is assumed that the pion ﬁeld so deﬁned is .9.22) where m N is the nucleon mass. π 2 Experimentally. from π N scattering gπN /4π = 14. equivalent.23) where gπN is the pion–nucleon coupling constant with the same normalization convention as for f π . One can now ask whether the relation (9.6. as can be seen from (9. Originally.24 and relation (9.25) where m 2 is the physical pion mass. Relating physical quantities to the results obtained in the exact symmetry limit m 2 = 0 is called the PCAC (partially conserved axial vector current) π approximation. way as an operator equation ∂ µ Aa (x) = m 2 f π π a (x) µ π (9. For q 2 = 0 this relation can be satisﬁed either for gA (0) = 0 or if h(q 2 ) has a pole for q 2 = 0. π a (x) is the pion ﬁeld normalized as follows: π π a ( p)|π b (x)|0 = δ ab exp(i px).24) is a good approximation to the real world where m 2 = 0. Hence m N gA (0) = gπN f π (9.25) is an identity on the pion massshell. This agreement supports our main assumption: massless QCD with spontaneously broken chiral symmetry is a good approximation for the strong interactions. q = p − p Current conservation implies 2m N gA (q 2 ) + q 2 h(q 2 ) = 0 (9. from (physical) π → lν | f π | = 93 MeV.24) which is the Goldberger–Treiman relation derived from the assumption of spontaneously broken chiral symmetry of strong interactions.23) = √ √ 2gπN (i/q 2 )i 2 f π qπ uγ5 u exp(iq x) ¯ (9.

calculating spontaneous chiral symmetry breaking in QCD is a difﬁcult problem which has not yet been solved. therefore. 2) and (2. In this model the scalar ﬁelds are introduced as elementary ﬁelds and their interactions are arranged to produce the spontaneous breakdown of the symmetry. i. 9. the so called σ -model. It has. 1) under SUL (2) × SUR (2). isotopic spin group α π → π + δα × π σ →σ under axial vector rotations (9.3 Phenomenological chirally symmetric model of the strong interactions (σ -model) As we have already mentioned. σ 2 . The fourth ﬁeld σ has been introduced to get a chirally symmetric theory. massless to start with so in fact we have two doublets NR and NL transforming like (1. for example. σ = (σ 1 .26) This is the usual pseudoscalar pion–nucleon coupling. π 2 .27) π to see that for chiral invariance of the lagrangian the (σ + iπ · σ ) must transform as the (2. the triplet pion ﬁeld π (0− ) and σ -a scalar 0+ meson. σ 3 ) (9. spontaneous symmetry breaking a good interpolating ﬁeld off-shell at p 2 = 0.25) and its connection to the chiral derivation see. For a more extensive discussion of the derivation of the Goldberger–Treiman relation based on relation (9. The elementary ﬁelds in the σ -model are: the nucleon doublet. Pagels (1975).28) α π → π + δα σ α σ → σ − δα · π . 2) representation of the SU (2) × SU (2) (remember that 2∗ ≡ 2).e. π 3 ).324 9 Chiral symmetry. The postulated lagrangian is as follows ¯ ∂ ¯ σ L = N i/ N + g N (σ + iσ · π γ5 )N + 1 [(∂µπ )2 + (∂µ σ )2 ] 2 − 1 µ2 (σ 2 + π 2 ) − 1 γ (σ 2 + π 2 )2 2 4 where π = (π 1 . This corresponds to the following transformations (see Appendix E): under vector SU (2). proved instructive in several aspects to study a phenomenological chirally symmetric ﬁeld theory model of the strong interactions. We rewrite ¯ ∂ ¯ ∂ ¯ ∂ N i/ N = NR i/ NR + NL i/ NL ¯ ¯ ¯ π π π N (σ + iπ · σ γ5 )N = NL (σ + iπ · σ )NR + NR (σ − iπ · σ )NL (9.

The physical vacuum must then satisfy (9. The vacuum is deﬁned by the minimum of the expectation values of the hamiltonian. whereas for µ2 /λ < 0 it exists for σ 2 + π 2 = |µ2 /λ| (9. j since on the other hand the operator relation i → i = exp(iα a Q a ) i (−iα a Q a ) ∼ i + iδα a [Q a . The physical vacuum is chosen (deﬁned) from the degenerate ground states given by the condition (9.30) [Q a . it must be that [Q a . π ). π b ] = 1 iεabc π c ± 1 iδ ab σ L 2 2 R Note (Appendix E) that in general if a ﬁeld multiplet transforms as i → i = (δi j − iδα a Tia ) j under linear transformations generated by generators Q a then. σ ] = ∓ 1 iπ a L 2 R (9. i ] = i (x)] (9. π ) = − 1 λ(σ 2 + π 2 + µ2 /λ)2 4 (9. In particular we would like to preserve unbroken isospin symmetry. As we know. = −Tia j j (x) (9.33) to see that for µ2 /λ > 0 the minimum occurs for σ = π = 0.34) With this second choice the model accounts for the spontaneous breakdown of chiral symmetry. Q L .3 σ -model 325 α (δα is an inﬁnitesimal vector in the isospin space). Q R and using canonical commutation 5 relations for ﬁelds one gets [Q a .29) Constructing then charges Q a .34) and also Q a |0 = 0. We can rewrite the potential as V (σ . from the effective potential formalism.9. in the tree approximation the vacuum expectation values of the ﬁelds are equal to the values of the classical ﬁelds in the minimum of the potential V (σ . for example.32) (matrices T a form a representation of generators Q a ). Q a |0 = 0 5 (9.34). They read 1 a ¯ π Vµ (x) = N (x)γµ 2 σ a N (x) + [π (x) × ∂µ π(x)]a ¯ Aa (x) = N (x)γµ γ5 1 σ a N (x) + σ (x)∂µ π a (x) − π a (x)∂µ σ (x) µ 2 (9.35) . Chiral symmetry implies conserved axial and vector currents.31) holds. Q a . It is now necessary to establish the correct ground state of the model.

The pattern of the spontaneous symmetry breaking from such a general viewpoint will be discussed in Section 9. one can see that 0|σ |0 = 0 implies Q a |0 = 0 (and vice versa) by taking the 5 vacuum expectation value of the commutator [Q a .19) one gets f π = − 0|σ |0 The v or f π is the only mass scale of the model.34) is no longer arbitrary if we want to keep the isospin unbroken. give to them and to the ﬁelds the desired physical interpretation.5.37) . from (9.29) give a vanishing contribution to this matrix element.36) we must.26) can now be rewritten in terms of the ﬁeld σ ¯ ¯ ∂ ¯ σ L = N i/ N + g N (σ + iσ · π γ5 )N + gv N N + 1 [(∂µπ )2 + (∂µ σ )2 ] 2 π − |µ2 |σ 2 − 1 λ(σ 2 + π 2 )2 − λv(σ 3 + σπ 2 ) + const. Another way 5 is to consider 0| exp(iα a Q a )σ exp(−iα a Q a )|0 to conclude that generators of the 5 5 transformations which do not leave 0|σ |0 invariant are spontaneously broken.36) π Indeed the vacuum must be an isospin singlet (hence 0|π |0 = 0) but not a chiral singlet (hence 0|σ |0 = 0.326 9 Chiral symmetry. π b ] = −iδ ab σ . spontaneous symmetry breaking These conditions are satisﬁed with the choice π 0|π |0 = 0. ﬁnd the broken and unbroken generators of the theory. 4 (9. In addition we observe that in the tree approximation. deﬁne the physical σ ﬁeld as σ =σ −v The lagrangian (9.29) 0|Aa (x)|π b (q) = 0|σ |0 0|∂µ π a (x)|π b (q) µ = − 0|σ |0 iqµ δ ab exp(−iq x) In the tree approximation the other terms in the axial current (9. 0|σ |0 = −(|µ2 /λ|)1/2 ≡ v (9. Formally. In view of (9. where σ belongs to the chiral doublet). we could also proceed in another way: deﬁne any of the a priori equivalent degenerate ground states as the physical vacuum. a σ meson of mass m σ = 2(−µ2 /2)1/2 = (2v 2 λ)1/2 and the isospin triplet of massless pions. However. To do this in an (9.38) It describes nucleons of mass m N = g(|µ2 /λ|)1/2 = −gv. In the basis so deﬁned the choice of the physical vacuum among the degenerate ground states given by the condition (9. Our last remark in this section is as follows: the model can be extended to account for an explicit breakdown of the chiral symmetry. Comparing with (9. It is worth stressing at this point that in our discussion of the σ -model we have ﬁxed at the very beginning the physical interpretation of the group generators and of the ﬁelds. to ensure the orthogonality of the vacuum to the one-particle state.

the theory contains elementary scalar ﬁelds i . The corresponding eigenvectors are Goldstone boson ﬁelds. Goldstone bosons as eigenvectors of the mass matrix First we generalize the discussion of Section 9.9.39) which is the operator PCAC relation (9. .33) plus the extra term one ﬁnds instead of (9. Salam & Weinberg 1962) which does not rely on perturbation theory. i = 1.4 Goldstone’s theorem 327 isotopically invariant manner one adds to L a term −εσ .36) the following equation for the vacuum expectation value v: −ε − λv 3 − µ2 v = 0 With the shift (9. Here we study the problem of Goldstone bosons in theories with elementary scalars (QCD-like theories are discussed in Section 9. Next we give the general proof of Goldstone’s theorem (Goldstone.25) introduced earlier. This proof can also be easily extended to theories without elementary scalars. .6) in a more general context. any complex .3 and show in the tree approximation that in a theory with spontaneously broken continuous global symmetry the mass matrix of the scalar sector has eigenvalues zero. Another effect of the εσ term is that it deﬁnes the unique physical vacuum.4 Goldstone bosons as eigenvectors of the mass matrix and poles of Green’s functions in theories with elementary scalars We have learned in the previous section that spontaneous breakdown of the chiral symmetry in the σ -model gives massless Goldstone bosons in the physical spectrum. Among other things. 9.40) (9. n. .37) one now gets m 2 = µ2 + λv 2 = −ε/v = ε/ f π π The axial current is no longer conserved ∂ µ Aa (x) = επ a (x) µ and ﬁnally we get ∂ µ Aa (x) = m 2 f π π a (x) µ π (9. . Consider a theory described by a lagrangian L which has some global continuous symmetry G generated by charges Q a . This vacuum alignment problem will be discussed in more detail later on. Then minimizing the potential (9. which we take as real.

vn )T is a solution of (9.42) is equivalent to the following commutation relation between charges Q a and ﬁelds i (see (9. Of course. The transformation rule (9. t) (9. Notice that if the vector v = (v1 .44) In the tree or classical approximation. Thus. in principle. give physically .45) then it follows from the G invariance of the potential V that the vector v = exp(−i a T a )v (9. for a given potential V which gives the spontaneous breakdown of the symmetry G the vacuum is inﬁnitely degenerate and any choice of it is physically equivalent (a choice of a solution v is a choice of the vacuum). spontaneous symmetry breaking representation can always be turned into a real one by doubling the number of the basis vectors. if the theory is to be renormalizable. T a is purely imaginary and being hermitean it is antisymmetric.328 9 Chiral symmetry. The V ( i ) is a real polynomial in ﬁelds i .32)): [Q a (t).41) where the unspeciﬁed terms.45) ∂ i ∂ i must have solutions with i = vi = 0 for some of the i s. different potentials may. The piece Ls of the lagrangian containing the scalar ﬁelds i is Ls = 1 ∂µ 2 i∂ µ i − V( i) + ··· (9. when the energy density of the vacuum state is just given by the minimum of the potential V ( i ).42) Matrices T a form the representation of charges Q a in the n-dimensional space of scalar ﬁelds i . . of the fourth order at most. one then concludes that for the spontaneous breakdown of the symmetry G the condition for the minimum of V ( i) ∂V ∂V a a 0 = δV ( i ) = δ i = −i Ti j j (9. irrelevant for our discussion in the tree approximation. We assume furthermore that the set i forms a multiplet transforming under some irreducible representation of the group G of the global continuous symmetry of the lagrangian L (if the original ﬁelds form a reducible multiplet we change the basis to write it as a direct sum of irreducible multiplets) i → i ≈ i −i a Tia j i (9.43) We are interested in theories in which the global symmetry G is spontaneously broken by the vacuum which is such that some of the ﬁelds i have non-vanishing vacuum expectation values 0| i |0 = vi = 0 (9. Because iT a must be real.46) is another solution to this equation. i (x. t)] = −Tia j j (x. . are possible couplings of the scalar ﬁelds to other ﬁelds of the theory.

The generators which remain unbroken must form the algebra . . N (note the values of the index i) we have X i v = 0 and (9. n. .47) and to rewrite in terms of them the potential V V( where 2 Mik = i) 2 = const + 1 Mik ( 2 − v)i ( − v)k + · · · (9. . We conclude these considerations with several useful remarks.45) one gets the following result: 2 Mik iTia v j = 0 j (9. Denoting their matrix representation by X i . one has Y I v = 0. non-degenerate. for each broken generator of G. . . Differentiating (9.48) ∂2V ∂ i∂ (9.1. .4 Goldstone’s theorem 329 different. Obviously.50) implies the existence. .51) We recall at this point that our discussion is for spontaneously broken global symmetries.48) because of (9.45) as the physical vacuum of our theory it is convenient to deﬁne the physical ﬁelds i = i − vi (9. . This so-called Higgs mechanism is discussed in Section 11. of the eigenvectors X i v corresponding to zero eigenvalues of the mass matrix. some generators of G must be broken. vacua.9. i = n + 1.50) which suggests that the mass matrix has eigenvectors T a v corresponding to zero eigenvalues. The term linear in ﬁelds does not appear in (9. Choosing some solution v of (9.45). The ﬁrst one consists of those which remain unbroken with our choice of the vacuum. To discuss this point more precisely we divide the generators of G into two groups. The massless boson ﬁelds (normal coordinates) are then the following: l = i(X l v)i i (9.49) k =v is the mass matrix for the physical scalar ﬁelds. To the second group belong generators which are spontaneously broken. Possible patterns of spontaneous symmetry breaking are discussed in more detail in the next section. i = 1. For spontaneously broken gauge symmetries the Goldstone boson ﬁelds can be gauged away and do not appear in the physical spectrum. These degrees of freedom are used for longitudinal components of the massive gauge boson. the vacuum is invariant with respect to Y i and (9. if some of the ﬁelds have non-vanishing vacuum expectation values.50) does not give any new information. Denoting their matrix representation by Y i .

for example.56) General proof of Goldstone’s theorem We turn now to the general proof of Goldstone’s theorem (Goldstone.h. k≤n (9. Then there is one more useful and obvious relation [X i . Therefore ci jl = 0. i ≤ n. k ≤ n.1. i. X j ] = ici jk Y k .54) cannot contain a term ci jl Y l . Finally. spontaneous symmetry breaking of some subgroup H of group G (it may. for chiral symmetry. Indeed.58) . l>n (9. X j ] = ici jk X k .53) and from the antisymmetry of the structure constants in indices i. j. Y j ] = ici jk Y k + ici jl X l . j > n. µ. l ≤ n. l>n (9. where i. The Green’s function G a (x − y) satisﬁes a Ward identity obtained by differentiating (9.57) with proper µ. see Section 9. j ≤ n. One can show that spontaneous breakdown of some global continuous symmetry G of the lagrangian implies poles at p 2 = 0 in certain Green’s functions and consequently implies the existence of massless bosons in the physical spectrum. k.330 9 Chiral symmetry. j. P X i P −1 = −X i (9.52) on the vector v we get a contradiction. Let us consider the Green’s function a G a (x − y) = 0|T jµ (x) µ. Thus. Salam & Weinberg 1962) in the class of theories speciﬁed in the beginning of this section. Eqs.s. Acting with both sides of (9.k account of the -functions involved in the deﬁnition of the T -product µ a ∂(x) G a (x − y) = δ(x 0 − y 0 ) 0|[ j0 (x). be that H ≡ 1).54) and (9. Furthermore. of course.51) taken together tell us that Goldstone bosons transform under some representation of the unbroken subgroup H . (9. it is easy to see that broken generators transform under some representation of the subgroup H [Y i . i. j > n. the r.k k (y)]|0 (9. i. j. if it is not the case then [Y i .k k (y)|0 (9. of (9.52) with some of the ci jl different from zero.57) a where jµ (x) is the current corresponding to a generator Q a of the symmetry G and k (y) belongs to an irreducible multiplet of real scalar ﬁelds. in many physical problems one can deﬁne a parity operation P which leaves the Lie algebra of the group G invariant and such that PY i P −1 = Y i .54) This follows from the property (9. k > n (9.55) This is the case.53) and the unbroken generators form the algebra of a group.

However. derived for bare currents and ﬁelds for which transformation properties (9.63) (1/ p 2 ) (9.62) This last equation implies that if some of the vacuum expectation values 0| j |0 ˜ µ.k j (0)|0 (9.60) the following: ˜ µ. Indeed.9.59). in principle.65) where the vector |π k ( p) describes a particle of mass m k which is a quantum of the .k exp[−i p(x − y)]G a ( p) (2π )4 j (0)|0 (9.60) holds both for bare and renormalized Green’s functions. Thus.60) A few words are worthwhile here about renormalization properties of (9.k d4 p ˜ µ. The Ward identity is.59) which follows from the assumed transformation properties.4 Goldstone’s theorem 331 This is an example of the so-called non-anomalous Ward identity (see Section 10. are assumed. see (9.60) (see also the next chapter). of ﬁelds i under the group G.k renormalized quantities.43).k i p µ G a ( p) = Tkaj 0| (9. t)δ(x − y) (9. related to these 0| j |0 by (9. t)] = −Tkaj j (y.62). the conserved currents are not subject to renormalization and the ﬁeld renormalization 1/2 R 1/2 R constants iB = Z i i = Z i do not carry the group index because the lagrangian is G invariant. one gets the following Ward identity (translational invariance is assumed) µ ∂(x) G a (x − y) = −∂(x − y)Tkaj 0| µ. Introducing the Fourier transform G a ( p) G a (x − y) = µ. the renormalized ﬁelds transform under the group G in the same way as the bare ﬁelds and the same Ward identity (9.k G a ( p) = pµ Fka ( p 2 ) and therefore Fka ( p 2 ) = −iTkaj 0| j (0)|0 (9. from Lorentz invariance the general ˜ µ. To this end we consider the matrix element a 0| jµ (x)|π k ( p) = i f ka pµ exp(−i px) (9. Let us take it as a relation for ˜ µ.k do not vanish then there are poles at p 2 = 0 in the Green’s functions G a ( p). for the discussion of anomalies in theories with fermions see Chapter 13.1). Using the relation a [ j0 (x.61) one gets from (9.k structure of G a ( p) can be written as ˜ µ.64) Our last step is to show that poles at p 2 = 0 imply the existence of massless bosons in the physical spectrum. t). k (y.

k kk (9.k = lim −i exp(−i px)G a ( p)( p 2 − m 2 ) k Comparison with (9.k lim G a ( p)( p 2 − m 2 ) = − f ka pµ k (9.k If G a ( p) is one of those Green’s functions which (according to (9.66) the following result: a ˜ µ.71) and f ka = f a δak .67) Writing exp(−i pz) as exp(−i px) exp[i p(x − y)] exp[i p(y − z)] we get from (9.69) implies m 2 = 0 and k f ka = 0: f ka = iTkaj 0| j (0)|0 (9.k ˜ kk 0| jµ (x)|π k ( p) = lim exp(−i px)G a ( p)iG −1 ( p) p 2 →m 2 k p 2 →m 2 k ˜ µ.70) Thus there must be massless bosons | a ( p) = iTkaj 0| j (0)|0 |π k ( p) in the physical spectrum corresponding to each broken generator which does not leave the vacuum invariant Tkaj 0| j (0)|0 = 0.66) where f p (z) = exp(−i pz) and −δk k d4 q G k k (y − z) = exp[−iq(y − z)] 4 q 2 − m 2 + iε (2π ) k d4 y G −1 (x − y)G(y − z) = δ(x − z) (9. spontaneous symmetry breaking ﬁeld k .69) ˜ µ.65) gives us our ﬁnal equation: p 2 →m 2 k (9.k element 0| jµ (x)|π k ( p) to the Green’s function G a ( p) a 0| jµ (x)|π k ( p) = d4 y d4 z f p (z)G a (x − y)iG −1 (y − z) µ. If the broken generators and the corresponding currents form a single real irreducible representation of the unbroken subgroup H which leaves the vacuum invariant then from Schur’s lemma a 0| jµ |π k ( p) ∼ δ ak (9.332 9 Chiral symmetry.68) ˜ µ.64)) have for spontaneously broken symmetry a pole at p 2 = 0 then (9.7) one can relate the matrix a ˜ µ. Using the reduction formula (see Section 2.

70). The question we would like to discuss here is what are the possible patterns of spontaneous breaking of a given symmetry G.1). Our ﬁrst example is a group of orthogonal transformations.2). Gauge ﬁxing then requires a speciﬁcation of some four-vector n µ and the general form of the matrix element involving the gauge source current is no longer given by (9. correspond to different potentials V ( ). In theories with elementary scalars the non-vanishing vacuum expectation values may already appear in the tree approximation as a solution minimizing the assumed potential V ( ) or may be generated by radiative corrections (Section 11. for example. and the scalar ﬁelds are assumed to form a real multiplet = ( 1 . Different choices of the physical vacuum among the degenerate vacua give the same pattern of symmetry breaking. are usually introduced by assumption. For a discussion in a covariant gauge see. f π = − 0|σ |0 (9. they simply correspond to different orientations of the unbroken subgroup H in the group G which can be transformed one into the other by a redeﬁnition of the basis. As we have already mentioned in the previous section for any choice of the vacuum speciﬁed by the non-vanishing vacuum expectation values there exists a set of physically equivalent vacua. Goldstone’s theorem does not hold for spontaneously broken gauge symmetries (see Section 11.63): a term proportional to n µ can be present. Bernstein (1974). Although different patterns. 9. with matrices Tkaj from Appendix E. patterns of spontaneous symmetry breaking are determined by the properties of the vacuum deﬁned by the non-vanishing of the vacuum expectation values of some ﬁelds. 0| ¯ |0 ) which. 2 . in general. In theories without elementary scalars the symmetry is spontaneously broken by some condensates (for example. obtained from the original one by the set of unitary transformations belonging to G. if they exist. which also minimize the same potential V ( ).5 Patterns of spontaneous symmetry breaking For a theory with some global symmetry G and a given multiplet content of ﬁelds. O(3) for deﬁniteness. no explicit reference to the potential is necessary for our discussion.9. in the absence of reliable methods for calculating non-perturbative effects. The breakdown of the presented proof is most easily seen if we work in a ‘physical’ gauge in which the gauge ﬁeld has only physical degrees of freedom. be such that some subgroup H of G remains unbroken. for the σ -model with 0|σ |0 = 0 we get from (9. must.5 Patterns of spontaneous symmetry breaking 333 For instance.72) as the generalization of the result obtained in the tree approximation. A choice of the vacuum speciﬁes the pattern of the spontaneous symmetry breaking which may. in general. 3 )T .

Any other choice of the vacuum expectation value v1 v2 0| |0 = (9. There are two Goldstone bosons corresponding to two linearly independent vectors iTx 0 and iTy 0 .74) breaks spontaneously the other two generators: 0 −v iTx 0 = v . This would not be the case if there was.73) 0 i 0 −i 0 0 0 0 0 A possible choice for the vacuum expectation value of the multiplet instance. Consider now unitary groups with scalars in a real representation. 0 0| |0 = 0 ≡ 0 v is.32) we get = 0 (9. The matrix representation of generators Q i in the space of s is 0 0 0 0 0 i 0 −i 0 Tx = 0 0 −i Ty = 0 0 0 Tz = i 0 0 (9. in space. a physically distinguished direction.334 9 Chiral symmetry.75) 0 0 0 0 0 O| exp(iα Q z ) exp(−iα Q z )|0 ∼ = + iα 0|[Q z .76) v3 which can be obtained from the original one by unitary transformations belonging to G gives the same pattern of symmetry breaking but this time with the combination v1 Tx + v2 Ty + v3 Tz unbroken.74) However. the vacuum deﬁned by (9. Any complex ﬁeld can be written in a real basis by deﬁning two real ﬁelds √ ϕi = (1/ 2)( i + i∗ ) (9. that is if the O(3) symmetry was also broken explicitly.77) √ χi = (1/ 2i)( i − i∗ ) . for The vacuum deﬁned this way is invariant under the transformations generated by Q z : taking the transformed vacuum exp(−iα Q z )|0 and using (9. This point will be discussed in detail in the next chapter. spontaneous symmetry breaking transforming under the vector representation. Both theories are physically equivalent and transform one into the other by a redeﬁnition of the basis. ]|0 = − iαTz Thus the group O(3) is broken to the group O(2) of rotations around the z axis. iTy 0 = 0 (9.

in this case. All other possible non-vanishing vacuum expectation values can be obtained from the one above by unitary SU (2) transformations and. 0)T .9. T1 = T2 = i i 0 −i i i 0 −i T3 = −i 0 i (9. . In the real representation ϕ1 χ r = 1 ϕ2 χ2 the generators of the SU (2) transformations are −i i 0 −i −i . there is just one pattern of spontaneous symmetry breaking and any non-zero choice of the vacuum expectation value 0| r |0 = (v1 v2 )T is equally good. Our next example is the group SU (2) with a doublet of complex ﬁelds 1 and 2 . . v. give the same pattern of symmetry breaking.78) and T is purely imaginary T = 0 i −i 0 because r must be real. thus. 0.79) Let us take. Obviously. Following the discussion of our ﬁrst example we see that the group is broken completely because the three vectors iT a 0| r |0 = 0 are linearly independent. 0| r |0 = (0.5 Patterns of spontaneous symmetry breaking 335 For the group U (1) of the unitary transformations → the real representation is r = exp(−i ) → r = exp(−i T ) ϕ χ r where r = (9. for instance.

However. 5 T 3 and T 1 remain unbroken and we see from the commutation relations (9. 0. thus obtaining the same theory as before. of course.80) It is convenient to write the scalar multiplet in the matrix form ˆ = 1 2 b b λ . 0. For instance. for 0| a a the generators 5 T are broken and the T s remain unbroken: SUL (2) × SUR (2) → SUL+R (2). for 0| r |0 = (0.9) that they also form an SU (2) algebra. spontaneous symmetry breaking It is easy to extend the analysis to the SU (2) × U (1) group with one complex doublet of scalar ﬁelds (Glashow–Salam–Weinberg model).336 9 Chiral symmetry.81) . v)T 5 T collected in Appendix E we see that. for example. v. This corresponds to the standard choice 0|σ |0 = 0 and to the π s forming a triplet under the SU (2). We can call this group the isospin group and interpret as physical pions those combinations of the ﬁelds π i and σ which transform as a triplet under this new SU (2). 0. 0)T . 0. The spontaneous breaking of chiral symmetry in the σ -model can be discussed in a similar manner. again only one pattern of symmetry breaking is possible. . Take. 0)T we have T 0| r |0 = −T 3 0| r |0 and the combination T + T 3 remains unbroken. not always the case. b = 1. Then the generators 5 T 2 . Other choices of the vacuum give only different unbroken combinations of the generators T and T a but a redeﬁnition of the basis transforms one case into another. . 0. In this case we have one more generator i 0 −i T = i 0 −i but the vector iT 0| r |0 is not linearly independent from the other three vectors iT a 0| r |0 . . for example. where a Tbc ≡ −icabc b] a = −Tbc c (9. Thus. This is. . So far we have been discussing examples with only one possible pattern of spontaneous symmetry breaking. 8 (9. any other choice of vacuum among the SU (2) × SU (2) degenerate set gives the same pattern of symmetry breaking. 0| r |0 = (v. Using the real basis r = (π1 π2 π3 σ )T and the matrices T a and a r |0 = (0. As our last example we consider the group SU (3) with a real multiplet of scalar ﬁelds transforming under the adjoint representation of the SU (3) [Q a .

Considerable phenomenological evidence has then been summarized that the group SU (2) × SU (2) is an approximate symmetry of the strong interactions which is spontaneously broken by the vacuum.6. if v3 = + 3v8 √ √ the λ6 . v8 = 0 the √ generators λ1 .83) First of all we notice that in our example the maximal possible breaking is SU (3) → U3 (1) × U8 (1) because it is clear from (9. One should be aware. 1 (λ3 + 3λ ) (SU (2)) and 1 ( 3λ3 − λ8 ) (U (1)) 2 2 are unbroken.80). However. another pattern of symmetry breaking is also possible. If we assume v3 = 0. as we do for the σ -model in Section 9. for each pattern there is an inﬁnitely degenerate set of possible vacua obtained from (9. λ7 . ˆ] (9.81) it is equivalent to → = exp(−i a T a) with T a given by (9. In this case we have six Goldstone bosons.83) (with speciﬁed v3 and v8 ) by unitary SU (3) transformations. that our understanding of .6 Goldstone bosons in QCD 337 and λb are the Gell-Mann matrices.1. λ5 . namely SU (3) → SU (2) × U (1).3 and for QCD in Section 9. Chiral symmetry of the massless QCD lagrangian has been discussed in Section 9. For a given pattern of spontaneous symmetry breaking one usually works from the beginning in a convenient basis with a speciﬁed physical interpretation of the group generators and of the ﬁelds. The matrix ˆ is hermitean and ˆ ii = 0 so it has eight independent elements.6 Goldstone bosons in QCD The purpose of this section is to extend the proof of Goldstone’s theorem to theories like QCD. Taking them as λ3 and λ8 we can write the most general form of the vacuum expectation values of the scalar ﬁelds as follows: 0| ˆ |0 = 1 v3 λ3 + 1 v8 λ8 2 2 (9. of the fact. if 2 √2 √ 8 √ v3 = − 3v8 the λ4 . λ3 (SU (2)) and λ8 (U (1)) remain unbroken.82) and because of (9. λ2 . However. 1 (λ3 − 3λ8 )(SU (2)) and 1 ( 3λ3 + λ8 ) (U (1)) are unbroken. 9.82) that λ3 and λ8 remain unbroken for any choice of v3 and v8 .83) and (9. It is also worth noticing that a vacuum expectation value giving SU (3) → SU (2) × U (1) cannot be transformed by unitary transformations belonging to SU (3) to one which gives SU (3) → U (1) × U (1) and they must correspond to different potentials V ( ). Because the matrix is hermitean it can be diagonalized by unitary transformations and written in terms of commuting generators of SU (3). The transformation rule for the matrix ˆ is ˆ → ˆ = exp(−i a1 a λ ) 2 ˆ exp(i a1 a λ ) 2 ∼ ˆ −i = a 1 a [2λ . however. with no elementary scalars.9. for the multiplet .

we shall assume the spontaneous breakdown of chiral symmetry and then show that it implies the existence of massless bosons (composite states) in the physical spectrum. spontaneous symmetry breaking the underlying mechanism of spontaneous symmetry breaking in theories without elementary scalars is not yet satisfactory. We still expect the ground state to be invariant under Lorentz transformations so these pairs must have zero total momentum and angular momentum.85) i j |0 = 0. Assuming that it becomes arbitrarily strong we expect that the ground state of the theory has an indeﬁnite number of massless fermion pairs which can be created and annihilated by the strong coupling. Thus we ﬁnd that the vacuum |0 has the property that operators which destroy or create such a fermion pair have nonzero vacuum expectation values 0| ¯ R L |0 = 0. ( † )i j = ¯ R j Li (9. 0| ¯ L R |0 = 0. 2 are ﬂavour indices). This relies on the fact that the gauge coupling in QCD becomes strong at large distances. In the following we assume that the chiral symmetry of QCD is indeed spontaneously broken by non-vanishing vacuum expectation values of some fermion condensates. It has at least been realized that only asymptotically free theories allow spontaneous chiral symmetry breaking at reasonable momentum scales (Lane 1974.338 9 Chiral symmetry. We consider QCD with two ﬂavours and the chiral symmetry SUL (2) × SUR (2). 0| † i j |0 =0 0| i j |0 = 0| † i j |0 . we omit indices here. It is nevertheless useful to give at least an intuitive argument in favour of the spontaneous breaking of chiral symmetries in asymptotically free theories.84) (i. Let us introduce scalar operators and † ij = ¯ Lj Ri . From the transformation properties of the ﬁelds Li and R j under the chiral group Ri † Lj = (UR )ik = Rk † † Lk (UL )k j it follows that † = UR UL ( We assume 0| and in addition † ) = UL † † UR (9. It is not our aim here to review partial results suggesting that this indeed happens. j = 1. Gross & Neveu 1974).

86) (we always work in a speciﬁed basis of chiral ﬁelds). From now on we always refer to renormalized quantities. being a hermitean 2 × 2 matrix. For the proof of Goldstone’s theorem we use a Ward identity for Green’s functions involving composite operator a (x) deﬁned as follows: a j0L a a ( j0R ) (x) = Tr[ (x) · T a ] . Notice the difference with the σ -model with one complex doublet.85) a [ j0L (x. t)δ(x − y) (9. t)T a δ(x − y) 0R where the are the zeroth components of the currents of the SUL (2) (SUR (2)) transformations and the matrices T a represent the generators of the SUL (2) (when to the right of the operator ) and of the SUR (2) (when to the left of the ) in the space of chiral doublets L and R .83)) 0| |0 = v1 = v3 σ 3 l (9. † (y.4 and it is based on Ward identities for certain appropriate Green’s functions. We can also write 0| i j |0 = 1 0| ¯ |0 δi j (9. t). where the SU (2) × SU (2) → SU (2) is the only possible pattern of symmetry breaking. The vacuum expectation value 0| |0 . v3 = 0 in (9. respectively. The above commutation relations are valid both for the bare and the renormalized quantities because chiral symmetry is an exact symmetry of the lagrangian (see also the next chapter). t)T a δ(x − y) a a [ j0R (x. now the 0| |0 is a 2 × 2 complex hermitean matrix. Therefore we assume v = 0. t)] = (x. We want the chiral SUL (2) × SUR (2) symmetry to be broken to the SUL+R (2) symmetry which is the isospin symmetry in our basis. The proof is similar to that in Section 9.6 Goldstone bosons in QCD 339 The latter equality ensures that parity is not spontaneously broken since then 0| ¯ i γ5 j |0 = 0|( − † ) ji |0 = 0 (9. or four real ﬁelds.87) where σ 3 is the diagonal Pauli matrix.9. t)] = −T (x.87): it is clear from (9. t). t)δ(x − y) [ j a (x. (y.87a) 2 where ¯ = ¯1 1 + ¯2 2 We are ready to prove Goldstone’s theorem. (y. t)] = −T a † (x.88) a [ j0L (x.85) that the vacuum expectation value v1 is invariant under the simultaneous UL = UR l transformation. t).59) and (9. † (y. t)] = † (x. First we collect the useful equal-time commutation relations following from (9. can be most generally written as (see the discussion preceding (9. t).

consider the Green’s function G ab (x − y) = 0|T a (x) b (y)|0 (9.4. µ b (y)|0 = d4 p ˜ µ. t) + † (y.h. † = (9.88) we get 2 the following result for the axial currents: [Aa (x. σ a are Pauli matrices).93) Following the steps (9. G ab ( p) = pµ 2 δ ab = pµ 2 δ ab p p (9.92) (it is unaltered by the chiral anomalies discussed in Chapter 13).87) one gets 2 µ ∂(x) 0|T Aa (x) µ b (y)|0 = −2viδ(x − y)δ ab This result is analogous to (9. t)}δ(x − y) (9.89) (x) = i( ij † a i j )T ji = i ¯ T aγ 5 (T a = 1 σ a .60) and it implies a pole at p 2 = 0 in the Fourier ˜ µ.95) A single particle state with appropriate quantum numbers gives a pole at p 2 = m 2 in the Fourier transform of (9. means anticommutator. spontaneous symmetry breaking where (x) = i( and therefore a − − † ). (y)+ † (y)}T b ]|0 (9. t).340 9 Chiral symmetry.95) d4 x G ab (x) exp(i px) ≈ p 2 →m 2 iZ 2 δ ab p2 − m 2 (9.91) Using (9. Using this last result we ﬁnally arrive at the desired Ward identity: µ ∂(x) 0|T Aa (x) µ b (y)|0 = −iδ(x − y) 0| Tr[{T a .s.96) . Indeed.64) we get ¯ 2v ˜ µ. 0 (y. Tr[T a T b ] = 1 δ ab and (9. exp[−i p(x − y)]G ab ( p) (2π )4 (9. transform G ab ( p) of the Green’s function G ab (x − y) µ. G ab (x − y) = 0|T Aa (x) µ.94) ˜ µ.90) where the curly bracket on the r. t)] = −i{T a . Using the commutation relations (9. The proof goes as in Section 9.62)–(9. (y. The pole at p 2 = 0 in the Green’s function G ab ( p) implies the existence in the physical spectrum of the massless Goldstone boson with the quantum numbers of the operators a . The only modiﬁcation is that some care must be taken about the interpretation of the operator (x).

in the sense of Chapter 2.68) 0|Aa (x)|π b ( p) µ to get p 2 →m 2 p 2 →m 2 a (x) (9. ﬁeld π a (x): π a (z) = Z −1 we repeat the step (9. (9.98) ˜ µ. It does not carry the group index because G is an exact symmetry of the lagrangian.101) This completes the proof of Goldstone’s theorem.9.100) ˜ µ.99) ˜ µ.100) implies m 2 = 0 and Z fπ = − ¯ (9. Introducing the physical. Since G ab ( p) has the pole (9. lim G ab ( p)( p 2 − m 2 ) = −Z f π δ ab pµ (9.97) (9.66)–(9. . In this case we have two low energy parameters: f π and ¯ . ≈ −i exp(−i px)Z −1 G ab ( p)( p 2 − m 2 ) (9.65) 0|Aa (x)|π b ( p) = i f π δ ab pµ exp(−i px) µ and the steps (9.94) due to the spontaneous breakdown of chiral symmetry.6 Goldstone bosons in QCD 341 where Z is a ﬁnite dimensionful constant ( a is a renormalized operator).

Furthermore. where L0 is invariant under global internal symmetry G and L1 explicitly breaks this symmetry.1) stands for boson or fermion ﬁelds). Several interesting physical problems fall into this category. We begin with a systematic discussion of Ward identities between Green’s functions involving global symmetry currents.1 Internal symmetries and Ward identities Preliminaries We are interested in theories described by lagrangians L such that L = L0 + L1 (or H = H0 + H ). In addition the symmetry G may or may not be spontaneously broken.10 Spontaneous and explicit global symmetry breaking So far we have been discussing theories described by lagrangians having some exact global symmetry G spontaneously broken by the vacuum. 342 . 10. This will also be useful as a supplement to our use of Ward identities in the earlier chapters and as an introduction to Chapter 13 devoted to the problem of anomalies. We assume that L1 can be treated as a perturbation whose effect can be calculated as a perturbative expansion in an appropriate small parameter. let the bare ﬁelds in the lagrangian L form a basis for some deﬁnite representation R of the symmetry group G of the lagrangian L0 δ ( i = −i a Tia j j ≡ a a δ i (10. Now we would like to introduce the possibility that the symmetry G of the lagrangian is also broken explicitly by a small perturbation.

using the classical equations of motion. i. The same discussion applies when the symmetry G is spontaneously broken since in this case also only symmetric counterterms are needed: this follows from (2.1 Internal symmetries and Ward identities a The bare currents jµ (x) constructed according to Noether’s theorem a jµ (x) = −i 343 ∂L Ta ∂(∂µ i ) i j j (10. we assume this to be the case.2) which are conserved currents of the G symmetric classical theory described by L0 . Whether G is spontaneously broken or not we are able to derive the same Ward identities.7) We want to derive Ward identities for bare regularized or for renormalized Green’s functions following from the symmetries of the lagrangian.3) and that L1 does not contain derivatives of the ﬁelds we get δL(x) = a a (x)δ a L(x) + jµ (x)∂ µ a (x) (10.3) the lagrangian L transforms as follows: L (x) = L(x) + a a δ L(x) = L(x) + a a δ L1 (x) (10. L1 = 0. Under an inﬁnitesimal global symmetry transformation i (x) = i (x) + a a δ i (x) (10. .e. As we will see in Chapter 13 this is not true in theories with chiral fermions. Both bare regularized and renormalized Green’s functions are G symmetric. Let us ﬁrst consider the case in which the symmetry G is an exact symmetry of the lagrangian.5) In the following we shall also need the variation of the same lagrangian L under local transformations i (x) = i (x) + a (x)δ a i (x) (10. remain unchanged when L0 → L if L1 does not contain derivatives of the ﬁelds a present in the jµ (x). The renormalized ﬁelds transform under the same representation R of G as the bare ﬁelds: the renormalization constants are G-invariant. a ∂ µ jµ (x) = δ a L1 (x) (10.161). In such theories some of the Ward identities derived here are afﬂicted with anomalies.10. If the regularization preserves this symmetry then the necessary counterterms are also G-invariant. We assume in this section that the UV regularization of the theory does preserve its classical symmetries.6) Remembering that L0 is invariant under global transformations (10.4) and.

Then the counterterms required by L1 . δ a L(x) and δ a i (x).8) The integral is invariant under the change of integration variables deﬁned by (10.7) we get 0= d4 x ×[ a D i exp iS + i a d4 x J i i a a (x)δ a L + jµ (x)∂ µ (x) + J i (x) (x)δ a i (x)] (10. This is true when generators T a in (10.s. the wave-function renormalization constants are mass-independent and thus the same as in the chirally symmetric theory. † We consider cases in which det(D i /D j ) = 1 + O( 2 ). for such a ‘soft’ breaking the wave-function renormalization constants.s.1) are traceless. . remain symmetric under G (strictly speaking there exists a renormalization scheme such that they remain symmetric) and the renormalized ﬁelds of the theory with softly broken symmetry G transform under G in the same way as the bare ﬁelds. This is so-called soft explicit symmetry breaking. Thus. Ward identities from the path integral Given all these preliminary constraints we can proceed to derive Ward identities using the functional integral formulation of quantum theory. In other words. included in counterterms of dimension 4. for example. If the integration measure is invariant.344 10 Spontaneous and explicit global symmetry breaking With L1 present we consider the case in which symmetry breaking operators have dimension d < 4 (L1 then involves dimensionful constants).6). One should remember that the quantity on the l. We now deﬁne the Green’s functions 0|T A(x) i i (yi )|0 ∼ D i A(x) i i (yi ) exp(iS) (10. for bare regularized Green’s functions W [J ] = N D i exp i d4 x [L( i) + Ji i] (10. being of dimension lower than or equal to d (see the discussion in Section 4.10) a where A(x) denotes one of the jµ (x). of (10.† using (10.2) do not affect counterterms of dimension 4. Consider the generating functional W [J ]. Our general remarks can be illustrated with the minimal subtraction renormalization scheme used for QCD with the mass term m ¯ . It breaks down for chiral transformations in theories with chiral fermions (see Chapter 13).9) The crucial point in deriving (10. in this case there is no regularization of the path integral and no renormalization prescription for the Green’s functions which preserve the full chiral symmetry and anomalies are present.6). We also assume the absence of anomalies.10) is just a short-hand notation for its r.h.9) is the invariance of the integration measure under the transformation (10.h.

J. K ] = S0 [ .10. K ]} . In our case the background ﬁelds are the sources j i .12) where S0 [ ] is the part of the action symmetric under global transformation (10. of (10.11) Ward identity (10.1 Internal symmetries and Ward identities 345 The notation has been chosen. J. by construction.s. an action which is exactly invariant under certain symmetry transformations. A] + d4 x (J i i + K · L1 ) (10. If the symmetry G is an exact global symmetry of the lagrangian the ﬁrst term on the r. µ a jµ and L1 .11) is valid irrespective of whether the symmetry is spontaneously broken or not. to make easy contact with the operator language.13) = −J i δ δ K · L1 = −K · δL1 The generating functional W [A.9) by parts (remember the remarks following (2. Aa and K for the operators i . We take the latter to have well-deﬁned transformation properties under the symmetry group. A.6). This is achieved if S[ .6) if simultaneous transformations on A. The most compact way to derive such Ward identities is based on the general idea (see Chapter 8) of using background ﬁelds to get. We may also be interested in Ward identities for Green’s functions involving a several symmetry currents jµ and/or the symmetry breaking operator L1 . A. J.12) is invariant under (10. We want to introduce them in such a way that the action S[ . respectively. Action µ (10. J and K deﬁned as δ Aa = +∂µ µ δJi i a (x) + cabc i.55)) differentiating it functionally with respect to the sources Ji and then setting them to zero we get the following general Ward identity: a (∂/∂ xµ ) 0|T jµ (x) i i (yi )|0 = 0|T δ a L(x) i i (yi )|0 i (yi ) j=i j (y j )|0 −i i δ(x − yi ) 0|T δ a (10.11) vanishes. Integrating (10. A. K ] is invariant under the local transformation (10. J.1) and the background ﬁelds Aa have been introduced as gauge ﬁelds by changing µ the normal derivatives in S0 into covariant derivatives ∂µ → ∂µ + iAa T a . K ] = N D i exp{iS[ .h. b Ac (10. however. Under the constraint of explicit breaking being only soft the Ward identity is valid both for bare regularized and for the renormalized Green’s functions.

In particular renormalization of Green’s functions involving products of composite operators. J. for instance. K ] reads 0= d4 x δW δW δW δ J i (x) + δ Aa (x) + δ K (x) δ Ji (x) δ Aa (x) µ δ K (x) µ (10. K and Aa and setting J i = Aa = 0.14) a or specifying jµ . K ] = W [A .346 10 Spontaneous and explicit global symmetry breaking is then invariant under transformations (10. Ward identities like (10.13) of the background ﬁelds W [A. (10. The method can be generalized to generate Ward identities involving other composite operators with well-deﬁned transformation properties under the global symmetry group if we introduce sources for such operators. to ﬂavour fermionic currents.15) generates the desired Ward identities valid to any order in L1 by differentiating with respect to sources J i . ]}(J i δ a i +K ·δ a L1 −∂µ j µa +cabc Ab j µc ) µ a (x) (10. For inﬁnitesimal transformations the variation of the functional W [A. K ] The last statement contains all the information equivalent to Ward identities which follow from the global symmetry of the original theory. J. Finally we observe that in the special cases of no massless Goldstone bosons α coupled to the current jµ .17) Note that in the latter case the symmetry may also be spontaneously broken since . K . 0 = d4 x D i exp{iS[A.16) which simply means an exact invariance of the Green’s functions under the symmetry transformations. in addition to multiplicative renormalization. We stress again that the validity of the symmetry relations for the Green’s functions relies on the existence of the regularization and renormalization procedures which preserve this symmetry.11) can be integrated over spacetime and the surface terms neglected.15) Eq. Thus in the case of exact symmetry which is not spontaneously broken one gets δ a 0|T i i (yi )|0 =0 (10. J . may also require subtractions in the form of polynomials in momenta in momentum space or of δ-functions and their derivatives in position space. µ µ K = 1. When the symmetry is explicitly broken by perturbation L1 we have δ a 0|T i i (yi )|0 = −i d4 x 0|T δ a L i i (yi )|0 i i (yi )|0 ≡ −i 0|T δ a S (10. J.

already be seen at the level of the tree approximation. t). t). Q b (t)] = icabc Q c (t) (10. t). Deﬁning Q a (t) = we get from (10.22) where H is the symmetry breaking part of the hamiltonian. j0 (y. for instance.10. t). This can. Q a (t)] (10. i (y.18)–(10. Comparison with the operator language The equal-time bare current commutation relations following from the equal-time canonical commutation relations for the bare ﬁeld operators are a b c [ j0 (x.21) If the symmetry breaking is soft (10. t)] = −Tia j j (x.1) we have furthermore [Q a (t). However.23) (the surface terms at inﬁnity can be neglected because due to perturbation L1 the would-be Goldstone bosons acquire masses). Q a (t)] (10. and therefore a ∂ µ jµ (x. t) = i dt 0 d3 x [H (x. t) = i[H (x.18) [Q a (t).19) a d3 x j0 (x.1 Internal symmetries and Ward identities 347 in the presence of perturbation L1 the massless Goldstone bosons acquire masses. or a [ j0 (x. t)] i (x. i.e. t) (10. t)δ(x − y) (10.20) = −Tia j j (y.18) are true in the symmetric theory and also with the symmetry breaking term L1 included as long as such terms do not contain derivatives of the ﬁelds.18) Relations (10. t)] = icabc j0 (x. The time dependence of the charges is given by dQ a = dt d3 x d a j (x.21) hold for the renormalized quantities as well because wave-function renormalization constants can remain invariant (for properly chosen renormalization schemes) and as will be seen shortly the currents are not subject to renormalization. . by analysing the classical lagrangian. t)δ(x − y) (10. t) In accordance with the transformation properties (10. in the presence of the symmetry breaking term L1 the currents are no longer conserved and the corresponding charges are time-dependent.

24).11). For instance a a (∂/∂ xµ ) 0|T jµ (x) (y)|0 = 0|T ∂ µ jµ (x) (y)|0 a + δ(x 0 − y 0 ) 0|[ j0 (x).24). We shall come back to these problems in some detail in the following. for instance. at x0 = y0 .e.11) is the same as one derived for the time-ordered product of operators deﬁned as a a T jµ (x) (y) = jµ (x) (y) (x 0 − y 0 ) + a (y) jµ (x) (y 0 − x 0 ) (10.5) for the operator ∂ µ jµ we recover the Ward identity (10. (y)]|0 (10.21) together with (10. Therefore there is an apparent ambiguity T A(x)B(0) → T A(x)B(0) + c1 δ(x) + c2 ∂x δ(x) + · · · (10. whereas the latter are not and they may differ by local terms. This relation is obtained by differentiating the T -product deﬁned by (10.24) is clear everywhere except at coinciding points. i.25) with (x) replaced by any operator O(x). Similarly. the so-called seagull terms.24) (with a straightforward generalization to products of more than two operators) by direct differentiation of the Green’s function and use of the commutation relations (10. in many cases these terms certainly do not vanish (see Jackiw (1972) for a systematic discussion) and one may wonder about the compatibility of the two methods. generally speaking.26) (Lorentz and other symmetry group indices must.13) or (10.25) a and using (10. of course. The answer is that the Green’s functions deﬁned by the path integral do not always coincide with the vacuum expectation values of the T -products deﬁned by (10.21).348 10 Spontaneous and explicit global symmetry breaking The Ward identity (10. the Ward identity (10. Otherwise anomalies appear.15) if we neglect in the current commutators the presence of the derivatives of the δ-functions (the so called Schwinger terms). Ward identities and short-distance singularities of the operator products Consider. The question thus arises of the role of such terms which in momentum space are polynomials in momentum. be properly taken into account). The path integral approach tells us that it is always possible to arrange such cancellation by a suitable renormalization prescription if there exists one which preserves the symmetry of the classical lagrangian. which cancel the Schwinger terms. However. the meaning of time-ordering as given by (10. . one can derive Ward identities for Green’s functions involving products of currents and they are the same as the relations following from (10. The former are always understood to be regularized or renormalized covariant quantities. However.

Using the deﬁnition (10.10.h.25) can be written as − lim ε→0 d4 x ∂µ f (x) 0|T jµ (x)O(0)|0 (|x0 | − ε) (10. so that the lightcone singularities of the operator product collapse to the origin.25).28) where f˜(q) is the Fourier transform of f (x) and use the notation ˜ T (q) = d4 x exp(iq x)T (x) The Fourier transform (∂µ T ) of ∂µ T (x) is (2π)−4 d4 q f˜(−q)(∂µ T ) = =− i. In particular. they hold in the sense that both sides are integrated over d4 x f (x) with f (x) being any ‘smooth’ test function vanishing at xµ → ±∞. let us rederive (10.31) (10.25) are for distributions.30) d4 x f (x)∂µ T (x) d4 x ∂µ f (x)T (x) (10. of (10.25) to be related to the short-distance singularity structure of the distributions on both sides of (10. Then the integrand . To see this clearly.s. ˜ (∂µ T ) = −iqµ T (q) In the other notation d4 x exp(iq x)∂µ T (x) = −iqµ d4 x exp(iq x)T (x) (10.32) Performing Wick’s rotation we can work in the Euclidean space.e. i.29) We expect the meaningfulness of the Ward identity (10.27) We also deﬁne the Fourier transform of a distribution (2π)−4 ˜ d4 q f˜(−q)T (q) = d4 x f (x)T (x) (10.27) the l.25) paying explicit attention to the singularities at x ∼ 0.1 Internal symmetries and Ward identities 349 To discuss this point we recall ﬁrst that equations like (10. the derivative ∂µ T (x) of the distribution T (x) is deﬁned as follows: d4 x f (x)∂µ T (x) ≡ df d4 x [−∂µ f (x)]T (x) (10.e.

34) using the OPE.4) [ jµ (x. we may consider the Fourier transform d4 x exp(i px)∂ µ 0|T jµ (x)O(0)|0 = lim ε→0 ε→0 d4 x [−∂ µ exp(i px)] 0|T jµ O|0 (|x0 | − ε) = − lim + lim ε→0 d4 x exp(i px) 0|T j0 (x)O(0)|0 [δ(x0 + ε) − δ(x0 − ε)] d4 x exp(i px) 0|T ∂µ j µ (x)O(0)|0 (|x0 | − ε) (10.37) is a well-deﬁned relation. operators translates into terms additional to the δ(x) term with a ﬁnite coefﬁcient in .37) (no Schwinger terms. Similarly. 0).33) or (10.36) we also get (see Section 7.35) with the equal-time commutator given by (10. 0)δ(x) (10.350 10 Spontaneous and explicit global symmetry breaking in (10. after excluding x = 0.32) is ﬁnite and unambiguous for ε > 0. One can immediately convince oneself that the limit exists if 1 T jµ (x)O(0) ∼ Cµ (x)O1 (0) + less singular terms (10. We have also assumed the absence of massless particles. we have ˜ −i pµ T ( p) = ( p) + d3 x exp(ipx) 0|[ j0 (x. Integrating by parts we get − lim −ε −∞ ε→0 ε→0 + ∞ ε dx0 d3 x ∂µ f (x) 0|T j µ O|0 = − lim + lim ε→0 d3 x f (x) 0|O(0) j0 (x.35) where ( p) is the Fourier transform of 0|T ∂ µ jµ (x)O(0)|0 . em em Often T jµ (x)O(0) is more singular than O(1/x 3 ) (for example T jµ (x) jν (0) ∼ (1/x 6 ) · 1 Stronger than O(1/x 3 ) singularity in the OPE of the T -product of l). Indeed.34) If the limit ε → 0 exists.34) are convergent. From (10. then the surface terms at ±∞ do not contribute. and similarly for ∂ µ jµ (x).33) In the last term the operator ∂µ has been interchanged with T because. T jµ (x)O(0) is regular. with no additional subtractions the Ward identity (10. with O(1/x 3 ) singularity the integrals in (10.33) and (10. Thus.33) and (10. O(0)] = a 1 O1 (x. a 1 ﬁnite).36) 1 where Cµ (x) ∼ O(1/x 3 ). ε)O(0)|0 d4 x f (x) 0|T ∂ µ jµ (x)O(0)|0 (|x0 | − ε) (10. −ε) − j0 (x. 0). O(0)]|0 (10. Let us discuss the limit ε → 0 in (10.

The ﬁnite constant c is to be ﬁxed by some renormalization conditions.4).11): µ a ∂(x) 0|T jµ (x) b a (y) ¯ c (z)|0 = 0|T ∂ µ jµ (x) a − δ(x − y)Tbd b (y) ¯ c (z)|0 0|T d (y) ¯ c (z)|0 b a + δ(x − z)Tdc 0|T (y) ¯ d (z)|0 (10. the following speciﬁc case of (10. Let us consider. at the same time subtractions at x = 0 are necessary to have a well-deﬁned limit when ε → 0 in the integrals in (10. without subtractions these integrals may depend on the order of integration over x0 and x (and be then non-covariant) or may be divergent. as if the Schwinger terms were absent.11) we discuss the renormalization properties of currents and their divergences. Our last remark is that in applications of Ward identities to physical problems one should pay attention to the fact that part of their content may be subtraction dependent. This subtractive renormalization of the Green’s functions comes in addition to the multiplicative renormalization of operators.10. However. Subtractions are of the form c1 δ(x) + c2 ∂(x) δ(x) + · · · possibly with divergent ci . Renormalization of currents As the ﬁrst application of (10. let T A(x)B(0) ∼ 1 2 − iε x 2 1 + ··· l Performing Wick’s rotation and integrating in Euclidean space we conclude that for a well-deﬁned ε → 0 limit in the l. For instance.38) . The path integral approach teaches us that when regularization and renormalization schemes exist which preserve the classical symmetry of the lagrangian it is possible to deﬁne the renormalized T -products so that they satisfy Ward identities in a form like (10.33) a subtraction of the form iπ 2 ln εδ(0) + cδ(0) is necessary. Depending on the nature of the singularity.33) and (10.11). In other words by suitable renormalization the Schwinger terms can be cancelled by subtraction terms.34).e.1 Internal symmetries and Ward identities 351 their equal-time commutators (see Section 7.h. integral in (10. i. These are the so-called Schwinger terms. for instance.s.

.38) that a a jµB (x) = jµR (x) up to a ﬁnite multiplicative constant which is ﬁxed as one by the current commutation relations.s.42) a for the divergence of the axial currents j5µ we have (from the equations of motion) a ∂ µ j5µ = 2mi ¯ T a γ5 (10.s. even if the symmetry is broken explicitly but only softly the currents and their divergences are not subject to renormalization.41) In the case of explicit symmetry breaking by soft (d < 4) terms one has ∂ jµ (x) = 0 but.352 10 Spontaneous and explicit global symmetry breaking We take (10. For instance. arguing as before. Thus the conserved currents are not subject to renormalization.h.43) and the non-renormalization theorem implies Z m = Z −1 ¯γ where ( ¯ γ5 T a )B = Z ¯ γ5 ( ¯ γ5 T a )R (10. Next. Writing in a a a a general jµB = Z j jµR and ∂ µ jµB = Z D ∂ µ jµR we ﬁrst conclude. of course. a jµB = ¯ B T a γµ B = Z 2 ¯ R T a γµ R a = jµR (10.38) to be for bare quantities. Expressing the bare fermion ﬁelds in (10. In the symmetry limit a ∂ µ jµ = 0 a B (10.40) where Z 2 is invariant under symmetry transformations. as discussed before. as discussed earlier. Fourier transforming (10. In conclusion. that Z j = Z D .h. for chiral symmetry broken by a mass term µ L1 = −m ¯ (d = 3) (10.44) Absence of wave-function renormalization for conserved or partially conserved currents does not.45) 5 (10. for example.39) a R = 1/2 Z2 (10.38) and taking the limit p = 0 ( p is conjugate to x) we get Z D = 1 since in the limit p = 0 the Fourier transform of the l.38) by the renormalized ﬁelds and dividing both sides by Z 2 we immediately conclude from the ﬁniteness of the r.40) can still be assumed. of (10. (10. exclude subtractive renormalizations for time-ordered products of such currents. vanishes (explicit symmetry breaking accompanies the spontaneous symmetry breaking and therefore there are no massless bosons in the spectrum a coupled to jµ currents).

Pagels 1975). It is worth mentioning at this point that the subject of chiral perturbation theory is not limited to the problem of the pseudoscalar meson masses but covers the whole low energy physics of pions and kaons. 1985a. Several techniques have been developed to expand the bound state masses.91) does not need subtractions µ and therefore in the chiral limit the relation (9. one can now calculate masses of the would-be (pseudo-) Goldstone bosons in terms of the explicit symmetry breaking parameter: quark mass m. The same Ward identity gave us the relation (9.) Imagine now a realistic case in which the chiral symmetry is also broken explicitly by the mass term (10.2 Quark masses and chiral perturbation theory 353 10.101) is free of any subtractiondependent parameters.101) between the parameters f π and ¯ of spontaneous breaking of chiral symmetry Z fπ = − ¯ (9. m d and m s (Dashen 1969. Gasser & Leutwyler 1984.2 Quark masses and chiral perturbation theory Simple approach Using the Ward identity for the Green’s function involving the axial current Aa (x) µ and the pseudoscalar density a (x) we have proved in Section 9.97) and (9. The most systematic approach proposed in a very interesting series of papers (Leutwyler 1984.98) 0| and 0|Aa (x)|π b ( p) = i f π δ ab pµ exp(−i px) µ respectively. b) relies on the effective lagrangian technique (see Chapter 14). that the Ward identity (9.10.5 that in chirally symmetric QCD with chiral symmetry spontaneously broken there exist massless Goldstone particles in the physical spectrum.43a) µ Instead. Green’s functions and any other quantity in the low energy pseudoscalar meson sector (which in the limit of exact chiral symmetry depend on only two parameters f π and ¯ ) in powers of the light quark masses m u . Dashen & Weinstein 1969. Then the proof of Goldstone’s theorem breaks down because ∂ µ Aa (x) = 2mi ¯ T a γ5 = 0 (10. (In view of our discussion in the previous section check.101) where the constants f π and Z are deﬁned by (9.42). a (x)|π b ( p) = Z δ ab exp(−i px) . using the OPE for T Aa (x) b (y). Such calculation is based on perturbative treatment of the symmetry breaking interaction (chiral perturbation theory) but takes account of all orders of the interactions described by the chirally symmetric part L0 of the lagrangian.

47) π which is the desired result valid to the lowest order in chiral perturbation theory. However. m ¯ are renormalization-scheme-independent. where µ is some renormalization point. and operators ¯ γ5 T a and ¯ are also renormalized at the same point.46) Working to the ﬁrst order in m we can now use (9. d) and m u = m d = m. The answer is almost immediate. however. Approach based on use of the Ward identity The result (10.47) can be extended to the case of three ﬂavours with m u = m d = m s (for heavy ﬂavours the breaking of chiral symmetry by the mass terms is too strong to be reliably considered as a perturbation). Eq. Therefore m = m(µ). Our discussion is always in terms of renormalized quantities. valid in the chiral limit.4. to get 2 m 2 = −(1/ f π )2m 0| ¯ |0 (10.354 10 Spontaneous and explicit global symmetry breaking Here we shall only be concerned with the pseudoscalar meson masses calculated in the ﬁrst order in the quark mass m.43a) sandwiched between the vacuum and the one-pion state together with the deﬁnitions (9. For a mass-independent renormalization procedure ratios of the renormalized masses do not depend on µ. for example. 0|uu|0 = 0|dd|0 = 0|¯ s|0 To extract reliable values for the quark mass ratios we need. The result (10. It is the pion mass. This approach can be used in other .97) and (9. We recall that interactions described by the chirally invariant part L0 of the lagrangian are included to all orders in our discussion. (10.98) imply the exact relation f π m 2 = 2m Z π (10. to include the electromagnetic mass corrections which will be considered in Section 10.1 the products m ¯ γ5 T a . We get the following: 2 m 2 0 = m 2 ± = (m u + m d )C + O(m q ) π π 2 m K+ = (m u + m s )C + · · · (10.101).47) can also be derived using a Ward identity directly for the realistic case with the chiral symmetry broken explicitly. as discussed in Section 10. The pion mass has been expressed in terms of the parameters of chiral symmetry breaking: spontaneous 0| ¯ |0 and explicit m.48) m 2 0 = m 2 0 = (m d + m s )C + · · · ¯ K K 2 ¯ ¯ ¯ s C = −(2/ f π ) 0|uu|0 . in SU (2) × SU (2) symmetric QCD with = (u. Finally we would like to stress that the quark masses we talk about are the renormalized quark mass parameters of the lagrangian (the so-called current quark masses).

Q a (t)] (10. t) = i[H (x.23). In general ab (q) is given by the pole and the continuum contribution 5 starting at s = 9m 2 π ab 2 5 (q ) = 2 f π m 4 δ ab π + m2 − q2 π ∞ ds (3m π )2 ρ(s. of (10. ∂ ν Ab (0)]|0 5 ν (10. m 2 ) π s − q2 (10.51) which in a general case.52) For us H = m ¯ .49) can then be regarded as an identity for the renormalized Green’s functions. ∂ ν Ab (0)]|0 µ ν 0 ν (10.50) as the spectral µ π . can also be written as ab 5 (q = 0) = − 0|[Q a (t = 0).49) and taking the limit q → 0 one gets the following relation: 0= ab 5 (q = 0) + i 0|[Q a (t = 0). similar considerations hold for any other current corresponding to a generator broken both spontaneously and explicitly when according to (10.s. H (0)]]|0 5 5 (10.2 Quark masses and chiral perturbation theory 355 problems too and its generalizations are used to calculate higher order corrections in the chiral perturbation theory.23) The Ward identity reads µ ab ∂(x) 0|T Aa (x)∂ ν Ab (0)|0 = D5 (x) + δ(x0 ) 0|[Aa (x). ab Deﬁning the Fourier transform ab (q) of the D5 (x) 5 ab 5 (q) =i ab d4 x exp(iq x)D5 (x) (10.49) where ab D5 (x) = 0|T ∂ µ Aa (x)∂ ν Ab (0)|0 µ ν The divergence ∂ µ Aµ is soft: the identity (10. where Aa (x) is the axial current and ∂ ν Ab stands µ ν µ ν for 2mi ¯ γ5 T b . In the limit q → 0 the l.53) We have used 0|∂ µ Aa (0)|π b = f π m 2 δ ab and have written (10. using (10.52) can be used to calculate the would-be Goldstone boson (pion) masses in the ﬁrst order in chiral perturbation theory if we approximate the Fourier transform ab (q) by the 5 pion pole.50) Fourier transforming both sides of (10. [Q b (t = 0).23) a ∂ µ jµ (x.h. The relation (10. The starting point is the Ward identity for the Green’s function 0|T Aa (x)∂ ν Ab (0)|0 .49) is zero because in the present case (explicitly broken symmetry) there are no massless bosons in the physical spectrum and therefore no poles at q 2 = 0.10. t).

For q 2 = 0 one gets ab 2 5 (q 2 = 0) = f π m 2 δ ab + O(m 4 ) π π (10. If we introduce generators X a and Y a which are. Bjorken & Drell (1965)).47).54) The leading term in m 2 is given by the pion pole because this is the only term where π in the chiral symmetry limit the zero of the numerator in (10.54) it is O(m 4 ) and does not affect the ﬁrst order π results. The set of degenerate vacua is isomorphic to the coset G/H space. in higher orders the subtraction constants appear as additional free parameters. However. from commutation relations (9.3 Dashen’s theorems Formulation of Dashen’s theorems We have learned in Section 9. H (0)]]|0 5 5 2 fπ (10. In our speciﬁc problem. for consistency. it can be seen from the OPE (see Chapter 7) that the product T ∂ µ Aa (x)∂ ν Ab (0) µ ν has a singularity O(m 2 x −6 ). The pion mass measures the departure from chiral symmetry induced by perturbation H (x). broken and unbroken by the |0 then the transformation (10.4 that if a global symmetry G of a theory is spontaneously broken by the vacuum state |0 to a subgroup H then there is a set of degenerate vacua corresponding to different orientations of H in G. for example. If the vacuum |0 is left invariant by H then | (g) is left invariant by the equivalent G-rotated subgroup g H g −1 . Zero mass particles . as in the chiral limit.53) does require subtractions.90).56) where U (g) is a unitary operator corresponding to a group element g.57) Transformations in G/H space correspond to variations of the vacuum expectation values for which the effective action is level at its minimum. The subtraction ambiguity is O(m 2 ) or according to our discussion following (10. so expansion in m 2 is the chiral perturbation theory. [Q b (t = 0). π In the order O(m 2 ) we get the so-called Dashen’s relation π δ ab m 2 = − π 1 0|[Q a (t = 0). respectively. This set of vacua is generated by the action of the G transformations on the state |0 | (g) = U (g)|0 (10.55) where f π and |0 should be taken.356 10 Spontaneous and explicit global symmetry breaking integral using the standard technique (see.56) can be written as | ( ) = exp(−i a X a )|0 (10.53) partially cancels with the denominator. ab It is also worth observing that the Green’s function D5 (x) or equivalently the integral in the spectral representation (10. For instance. we easily get (10. 10.

ensure the vacuum stability.59) also follows from the requirement of the translational invariance of the vacuum: the vacuum value of a divergence then vanishes. To the leading order in the perturbation H (x) the energy density in each of the degenerate vacua is given by X a )|0 (10. in agreement with (9. Q a ].59) (10. (10.59) says that π a |H |0 = 0.e. . H (0)]]|0 ≥ 0 a In view of the relation ∂ µ jµ (x) = i[H (x).60) has a straightforward interpretation too: up to the renormalization factor the second derivative of the potential with respect to is the Goldstone boson mass matrix (see also Chapter 14) and (10.10.3 Dashen’s theorems 357 are quantized excitations – one for each orthogonal direction in G/H (see (9.55) referring.60) ensures that in the ﬁrst order 2 perturbation in H (x) the pseudo-Goldstone boson masses m ab are positive.60) give for the σ -model with chiral symmetry explicitly broken by the H (x) = +εσ term. [X a . Let us ﬁrst see what the conditions (10. From the commutation relations (9. So we rederive Dashen’s formula (10. A small symmetry breaking perturbation H (x) may change the above picture in a qualitative way: it may lift the degeneracy of the vacuum. the Goldstone boson tadpoles vanish. i.59) and (10. however.39) for the pseudo-Goldstone boson mass.58) To identify |0 as the true vacuum in the presence of the perturbation H (x).36). E( ) should have a minimum (at least a local one) for = 0. Eq. Since charges X a which do not annihilate the vacuum couple to Goldstone bosons 0|X a |π a = 0.30) we get 0|π|0 = 0 −ε 0|σ |0 ≥ 0 (10.51) and Chapter 14). H (0)]|0 = 0 (10.60) =0 = − 0|[X b . condition (10. (10. to the previous derivation for the ﬁxing of the normalization factor. The criteria for identifying the correct vacuum in the presence of a small perturbation H (x) are called Dashen’s theorems.61) for the correct vacuum. We therefore get the following two conditions: a a E( ) = ( )|H (0)| ( ) = 0| exp(i X a )H (0) exp(−i ∂ ∂ ∂2 ∂ a∂ b a E( )| E( )| =0 = i 0|[X a . Using the normalization given in (10.55) we also rederive the result (9. and generally speaking.

the chiral symmetry which is spontaneously broken.65) . Dashen’s conditions ﬁnd very interesting applications in models of the dynamical breaking of the gauge symmetry of weak interactions (technicolour and extended technicolour models). Explicit symmetry breaking lifts the degeneracy and leaves (10. where = v1 as in l (9.64) where µν αβ (x) is the G W gauge boson propagator µν αβ (x) = i 0|T Aµ (x)Aν (0)|0 ≡ δ αβ α β µν (x) (10.86) which we have taken as the deﬁnitions of the vacuum in the chiral limit and in a speciﬁed basis for the chiral ﬁelds. Imagine then that some of the G-symmetry currents couple to some of the gauge bosons of the group G W of weak interactions (in the rest of this section all sums over the group indices are explicitly written down) L1 = − α µ g α Aα (x) jα (x) µ (10.63) µ where jα (x) are linear combinations of some of the G currents. Here we give a general technical introduction to such problems which will also be useful in the next section where we calculate the π + − π 0 mass difference.61) and (10. for example.β 1 α β g g 2 d4 x µν α β αβ (x)T jµ (x) jν (0) (10.84).358 10 Spontaneous and explicit global symmetry breaking For the SUL (2) × SUR (2) symmetric QCD lagrangian with chiral symmetry + broken spontaneously to SU (2) by a condensate UR UL . In the ﬁrst non-vanishing order of perturbation theory in g α the G-symmetry breaking term of such a theory is given by the following effective hamiltonian: H (0) = − α. operators corresponding to quantum numbers commuting with G are also possible but irrelevant for our discussion. In particular we recover (9. Some of these ideas will be discussed in Section 11.62) as the only possibilities.62) This follows from the commutation relation (9. Dashen’s conditions and global symmetry broken by weak gauge interactions A physically very interesting case of an explicit global symmetry breaking by a small perturbation is the following one: consider a strongly interacting theory (a gauge theory) with some exact global symmetry G.88) and from the relations ¯ = Tr[ + † ] and ¯ γ5 = Tr[ − † ]. where is deﬁned by (9.3.87) and explicitly by the mass term H = m ¯ one gets 0| ¯ γ5 |0 = 0 − 0| ¯ |0 > 0 (10.

First of all the counterterms affected by H (0) are easily enumerated (by assumption H (0) is used only in the ﬁrst order perturbation): they involve only those operators of the expansion which are accompanied by coefﬁcients that produce non-integrable singularities.64). Section 10. We assume here that the expression (10. It turns out.64) and (10.c a gα R ac ( ) jcµ (x) (10.10. The non-singlets with d = 4 would affect our discussion but in several interesting cases they do not appear (see. do not require renormalization and are irrelevant for our discussion.68) where R( ) is the adjoint representation of G. The problem can be studied by means of the operator expansion of the product of the two currents in (10.67) (α is the G W index.66) expressed in terms of (10. notice the difference in the case if H (0) was considered in any order of perturbation theory.s. and at ﬁrst glance it appears that it could affect the wave-function renormalization of ﬁelds.3 Dashen’s theorems 359 The question is whether our considerations in Section 10. Dashen’s conditions (10. With perturbation H (0) given by (10. for example.h.1 based on the assumed softness of H apply to this case when the symmetry breaking term H (0) is not soft (d = 6). It is clear from (10. We recall that (10.66) where U is any unitary operator representing the group G. Among the operators of dimension 4 there may be singlets under G which are again irrelevant. The G W currents are linear combinations of the G currents µ gα jα (x) = a a µ gα ja (x) (10. that this does not necessarily happen.60) can be rewritten in a more useful form if we express the vacuum energy density E( ) given by (10.58) in terms of the G currents (Peskin 1981. a is the G index and there is no sum over α on the l.58) and (10.59) and (10. however.64). The higher dimension operators in the α β expansion of jµ jν do not cause a non-integrable singularity at x = 0.65) (remember that µν (x) ∼ 1/x 2 for x → 0) that these are operators of dimension ≤ 4.64) converges without subtractions which are non-singlets under the group G. Because the vacuum is H -invariant c d (G → H spontaneously) only the H -invariant part of the product jµ jν contributes to (10.4 for the breaking of the chiral group by H given by the product of two electromagnetic currents).) and consequently µ gα U † jα (x)U = a a µ gα U † ja (x)U = a. Preskill 1981).68).64) together give the following expression E( ) = − 1 2 d4 x µν (x) α 2 α α gα 0|T U † jµ (x)UU † jν (0)U |0 (10. For any given vacuum from the degenerate set the generators of G split into two groups: the generators Y a such that Y a |0 = 0 .

360 10 Spontaneous and explicit global symmetry breaking and X a where X a |0 = 0.72) ˆ ˆ + Tr[ XJαU X JαU ] 0|T X jµ (x) X jν (0)|0 } ˆ where JαU is the linear combination of the G generators corresponding to the α current U † gα jµ (x)U with gα included. Tr[Y a X b ] = 0). Correspondingly.R a b Eqs. Since Y a span a single real representation of H (the adjoint representation).73) . Taking all this into account we obtain for E( ) the following: E( ) = − 1 2 d4 x µν (x) α ˆ ˆ {Tr[YJαU Y JαU ] 0|T Y jµ (x)Y jν (0)|0 (10.68) can be split into the Y part and the X part µ µ µ U † gα jα (x)U = U † gα jα (x)U |Y + U † gα jα (x)U |x µ µ ≡ Y jαU (x) + X jαU (x) (10. where Y jµ denotes any single current Y jµ . R.71) (the last line deﬁnes our notation) such that only Y (X ) currents from the sum a ac c a. there is no summation in T Y jµ Y jν . true since P Q a P −1 = Q a .L L. (10. by c d Schur’s lemma the only H -invariant term in the product Y jµ Y jν is proportional to δ cd : c d 0|T Y jµ (x)Y jν (0)|0 = δ cd 0|T Y jµ Y jν |0 = Tr[Y c Y d ] 0|T Y jµ Y jν |0 (10.c gα R jµ contribute to the ﬁrst (second) component.70) also imply that the joint expectation values of one Y jµ and one X jν are zero by parity conservation. such that PY a P −1 = Y a . From the orthogonality of X a and Y b (Tr[X a Y b ] = 0) ˆ ˆ ˆ ˆ Tr[YJαU Y JαU ] = Tr[YJαU JαU ] ˆ ˆ ˆ ˆ Tr[ XJαU X JαU ] = Tr[ XJαU JαU ] Using in addition an obvious relation ˆ ˆ ˆ ˆ ˆ ˆ Tr[YJαU JαU ] = Tr[ JαU JαU ] − Tr[ XJαU JαU ] (10.69) (here we normalize the broken and unbroken generators to Tr[Y a Y b ] = a Tr[X a X b ] = δ ab . We will limit our discussion to cases in which in the product X a X b also the only H -invariant term is δ ab . of course. The decomposition in (10.74) (10. It can be shown (Peskin 1981) that this restriction implies the existence of a parity operation P which preserves the Lie algebra of G.70) are. P X a P −1 = −X a (10. each G current carries index Y or X .70) For chiral symmetry (10.

3 Dashen’s theorems 361 we ﬁnally get E( ) = − 1 2 −1 2 × α d4 x d4 x µν (x) α ˆ ˆ {Tr[ JαU JαU ] 0|T Y jµ (x)Y jν (0)|0 } µν (x) 0|T [ X jµ (x) X jν (0) − Y jµ (x)Y jν (0)]|0 (10.10. from (10. the vacuum orientation is a stationary point of E if ˆ ˆ [YJα . Jα ] XJα ] = 2i α ˆ ˆ Tr[X a [YJα . X Jα ] = 0 for all αs.75) ˆ ˆ Tr[ XJαU X JαU ] The ﬁrst term in the sum is -independent (without subtraction it may even be inﬁnite) and in the second term -dependence has been factored out from the current matrix elements. X Jα ]] ∂2 1 2 ∂ a∂ b α ˆ Tr( XJαU )2 |U =1 × M 2 ≥ 0 (10. one gets the preferred vacuum simply by minimizing the quantity α ˆ ˆ Tr[ XJαU X JαU ] ˆ i. We can rewrite Dashen’s conditions (10. X ] = iY following from (10. The second Dashen’s condition reads 2 m ab = α ˆ ˆ Tr[ X [X a . for projections of transformed currents on the subspace of the original broken generators. Assuming that the coefﬁcient of the last trace in (10.77) where.78) and Y jµ ( X jµ ) denotes. regarding the vacuum |0 as ﬁxed so that the rotation is applied to the perturbation H . by minimizing the trace for projections of currents J α on the subspace of a a possible equivalent representations of the X and Y or.e.70). C]]. any single current corresponding to an unbroken . as before. Thus.75) 2 M 2 = (1/ f π ) d4 x µν (x) 0|T [Y jµ (x)Y jν (0) − X jµ (x) X jν (0)]|0 (10.59) and (10.76) where the last result follows from the commutation relations [X.75) is negative as can be argued in some cases. orthogonality of Y and X and the trace property Tr[[A. Writing U = exp(−i a X a ) one has ∂ ∂ a α ˆ Tr( XJαU )2 |U =1 = 2i (10. One sees that the preferred vacuum orientation corresponds to the minimal number of ˆ broken generators in the decomposition of J α into the generators of G.60) in the present notation. B]C] = Tr[A[B.

64) with jµ (x) being the electromagnetic current and µν the photon propagator. Thus. including the vacuum. the only d = 4 operator (G α G µν ) is a singlet under µν α the chiral group and our considerations of Sections 10.3 we notice that the expansion of the operator product T jµ (x) j µ (x) reads T jµ (x) j µ (x) = C1 (x)1 + l q Cq (x)m q qq + C G (x)G α G µν ¯ µν α + operators with d > 4 + non-scalar operators (10.80) where G α is the gluon ﬁeld strength. 10.64).79) = −2 =2 α ˆ ˆ ˆ ˆ Tr[[YJα . The latter can often be studied by means of the spectral function sum rules. Jα ] X [X b . Only scalar operators contribute to the µν integral in (10. We also see that in the chiral limit m q = 0 all the divergent terms in Hem are singlets under the chiral group and may be disregarded: they induce a universal energy shift for all states.3 apply.3. [ XJα . [YJα . The expressed as follows ∂2 ∂ a∂ b -dependence is only in the trace which can be α α ˆ T r ( XJαU )2 |U =1 ˆ ˆ ˆ ˆ {Tr[ X [X a . Following our discussion in Section 10.1 and 10. The quark masses will be neglected in the calculation of the electromagnetic masses in the spirit of the ﬁrst order perturbation in any explicit symmetry breaking term. in this case the electromagnetic pion mass . X a ]]X b − [ XJα .4 Electromagnetic π + –π 0 mass difference and spectral function sum rules Electromagnetic π + –π 0 mass difference from Dashen’s formula In Section 10. This section is devoted to a study of the electromagnetic contribution to the pion masses. Another source of such a breaking is the electromagnetic interaction of quarks which couples the T 3 generator of the isospin group to the electromagnetic gauge ﬁeld: Q = T 3 + Y . Jα ]] XJα ] + Tr[ X [X a .60) or equivalently the positivity of the pseudo-Goldstone bosons mass matrix depends on the signs of the trace (10. X α ]]X b ] Dashen’s condition (10.2 we have studied an explicit chiral symmetry breaking in QCD caused by non-zero current quark masses.362 10 Spontaneous and explicit global symmetry breaking (broken) generator of G.79) and of the integral (10. Several applications of the results of this section will be discussed in the next section and in Section 11. [X b . As mentioned. The effective hamiltonian is given by (10. Jα ]]} (10.78).

To calculate the integral (10. can be discussed in a general way by invoking the OPE (Bernard.e. (10.82) was derived by Das et al.83) .79) and the fact that JY = eT 3 .6 we deﬁne the vacuum of the SUL (2) × SUR (2) ˆα symmetric QCD so that the SUL+R (2) remains unbroken. The calculation can be performed by means of Dashen’s formula.77) in the previous section). Formula (10. Lo Secco & Weinberg 1975).82) (see the derivation of (10.82) one can use the spectral function sum rules.81) where M 2 can be explicitly rewritten in terms of the vector and axial currents 2 M 2 = (1/ f π ) d4 x µν 3 3 (x) 0|T [Vµ (x)Vν (0) − A3 (x)A3 (0)]|0 µ ν (10. We also remember that both equations are valid only in the ﬁrst order in the symmetry breaking.82) We recall again that all. then J X = 0 and Dashen’s condition for the stationary vacuum is satisﬁed in the presence of the electromagnetic interactions) we immediately get the following result for the one-photon exchange contribution to the pion mass: (m 2 + )γ = e2 M 2 . Spectral function sum rules Spectral function sum rules. Let us consider the K¨ llen–Lehmann spectral representation (Bjorken & Drell 1965) for the timea ordered product of two currents ∞ 0 a b 0|T jµ (x) jν (0)|0 = dµ2 i d4 k exp(−ikx) 2 ρ ab (k. originally derived by Weinberg (1967a) for the SU (2) × SU (2) chiral symmetry. π (m 2 0 )γ = 0 π (10. k 2 = µ2 ) (2π )4 k − µ2 + iε µν (10. Duncan. Non-zero quark masses introduce second order corrections O(αm q ) which need renormalization and are therefore renormalization-scale-dependent. (1967). ˆα ˆα Using (10.77). singlet contributions to the electromagnetic masses have been disregarded in (10.4 π + –π 0 mass difference 363 difference is ﬁnite without renormalization. i. possibly inﬁnite.10. for m q = 0. J X = 0 for the electromagnetic current (as in Section 9.81) and (10.

In looking for such products one can use the global symmetry of the lagrangian which is respected by the OPE coefﬁcient functions even if the symmetry is spontaneously broken.T. If one constructs a linear combination of currents such that the Fourier transform (10.86) and expanding it formally in powers of 1/k 2 we get F.T.85) Taking the Fourier transform of (10.88) 0 ∞ (0) ds sρ (s) = 0 0 ab and these are the spectral function sum rules. = i kµ kν k2 ∞ 0 ∞ 0 dµ2 (1) ρab (µ2 ) gµν (0) + ρab (µ2 ) − i 2 2 µ k ∞ 0 (1) dµ2 ρab (µ2 ) kµ kν +i 2 2 (k ) (1) (0) dµ2 [ρab (µ2 ) + µ2 ρab (µ2 )] + · · · (10.87) This formal expansion can be meaningful only when the Fourier transform a b d4 x exp(ikx) 0|T jµ (x) jν (0)|0 behaves for large k 2 as O(1) or softer.86) is for large k 2 softer than O(1). In our problem we are interested in .83) F. Otherwise the coefﬁcients of the expansion must be divergent.87) must vanish. then the ﬁrst few terms in the expansion (10.84) a b 0| jµ (0)|n n| jν (0)|0 δ(k − kn ) = kµ kν ρ (0)ab (µ2 ) (2π)3 n spin 0 (10. k 2 = µ2 ) k 2 − µ2 (10. The OPE is useful in ﬁnding appropriate combinations of current products with soft high energy behaviour. for example.364 10 Spontaneous and explicit global symmetry breaking where the spectral function ρ can be decomposed into the spin-one and spin-zero parts: (2π)3 n spin 1 a b 0| jµ (0)|n n| jν (0)|0 δ(k − kn ) = −(gµν − kµ kν /µ2 )ρ (1)ab (µ2 ) (10. for softer than (1/k 2 ) behaviour ∞ (1) (0) ds [ρab (s)/s + ρab (s)] = 0 0 ∞ (1) ds ρab (s) = 0 (10. = d x exp(ikx) 0|T 4 a b jµ (x) jν (0)|0 =i 0 ∞ dµ 2 ab ρµν (k.

3) under SU (2) × SU (2). Representing the symmetry breaking interaction by a spurion which ¯ ¯ ¯ must also belong to (2. G µν ∼ (k ) up to logarithms and the sum rules (10.89) ∞ (1) (1) ds [ρ (s) − ρ (s)] = 0 0 V ab Aab which are valid in the chirally symmetric theory. According to (10. whereas the operator ¯ ¯ qq belongs to (2.4 π + –π 0 mass difference 365 the group G = SUL (2) × SUR (2) broken to the H = SUL+R (2). because the expansion coefﬁcient functions respect the G symmetry. 3) under SU (2) × SU (2). H -invariant. The asymptotic behaviour of G ab (k) is determined by the lowest-dimension µν a b operator in the expansion of jLµ jRν which has a non-zero vacuum expectation value. Thus. Taking into account that 0|Aa |π b = ikµ f π δ ab µ and therefore (0) 2 ρAab (µ2 ) = δab f π δ(µ2 ) + · · · (0) and neglecting any continuum contribution to ρA which should be small we get from (10. 2)). 2) we see that combining one spurion with the qq ¯ operator one can produce the (3. so its ¯ . the symmetry is broken by the mass ¯ term m qq. The second sum ¯ rule is then no longer valid: the coefﬁcient of this m qq term scales as x −2 . The lowest-dimension operator satisfying these requirements is ¯ ¯ a four-fermion operator of dimension (mass)6 ( ¯ transforms as (2. gauge-invariant. 2) ⊕ (2. i.88) hold for the ab spectral functions ρLR .e. a singlet under SUL+R (2).82) we are interested in the combination VV − AA = 4 jL jR . However. This can be seen by a spurion analysis (Wilson 1969a): the ¯ current product belongs to (3.10. 2). ab 2 −2 Therefore. This operator must be Lorentz-invariant. If the chiral symmetry is broken by non-zero quark masses the OPE for the difference VV − AA contains a term proportional to m qq. 2) ⊕ (2.88) the following sum rules: ∞ (1) (1) 2 (ds/s)[ρV ab (s) − ρAab (s)] = f π δab 0 (10. in the expansion of the two-current product we need one power of the symmetry breaking parameter m for the coefﬁcient function in front of the operator qq to be non-zero. it must transform as the adjoint representation (3. 3) of the SU (2) × SU (2). 3) representation. 2) ⊕ (2. Spectral function sum rules can be derived by studying the high momentum limit of G ab (k) = µν a b d4 x exp(ikx) 0|T jµL (x) jνR (0)|0 which transforms as the adjoint representation (3. and.

µ2 1 1 1 + ln 0 + const. however. Results We are now ready to calculate M 2 .92) with the two lowest-lying narrow resonances. (4π)2 ε (4π )2 µ2 (10. expected to be small because the masses m u and m d are very small in comparison to the scale of QCD.82).366 10 Spontaneous and explicit global symmetry breaking Fourier transform behaves as k −2 . Finally we can calculate (10.91) However.89) one gets 2 2 2 f ρ = f A1 + f π .90) The momentum integral is logarithmically divergent and gives. The remaining ﬁnite term gives (m 2 + )γ = π 3α 4π( f π )2 ∞ ds ln 0 µ2 (1) (1) 0 [ρV (s) − ρA (s)] s (10. Taking 2 ρV (s) = f ρ sδ(s − m 2 ) ρ 2 ρA (s) = f A1 sδ(s − m 2 1 ) A and using both sum rules (10. one (ρ) coupled to the vector current and the other (A1 ) to the axial current. Introducing explicitly µν (x) = gµν d4 k exp(−ik x) (2π )4 k 2 + iε we get M2 = − i 2 fπ ∞ 0 (1) (1) dµ2 3[ρV (µ2 ) − ρA (µ2 )] d4 k 1 4 k 2 (k 2 − µ2 ) (2π ) (10.92) and on account of the second sum rule (10.89) that the coefﬁcient of the divergent term vanishes. This just reﬂects the already discussed fact that the ﬁrst order electromagnetic contribution is ﬁnite after the singlet piece under SU (2) × SU (2) has been disregarded.89) it is renormalization point (µ2 ) 0 independent. The corrections are.92) assuming that it is reasonable to saturate the integral (10. it follows from the second sum rule (10. using dimensional regularization. (10.

Completing our discussion given at the end of . Extending our considerations to the SU (3) × SU (3) case we get (m 2 + )γ = K (m 2 + )γ and (m 2 0 )γ = 0.2 and of this section.6 MeV. π K m 2 + = C(m u + m s ) + (m 2 + )γ K π m 2 0 = C(m d + m s ) K (10.94) and (10.4 π + –π 0 mass difference 367 and 2 2 m 2 f ρ = f A1 m 2 1 ρ A Therefore we can eliminate the A1 parameters to get the ﬁnal answer in terms of the ρ parameters: (m 2 + )γ = π 3α 2 m 4π ρ fρ fπ 2 ln 2 fρ 2 2 fρ − fπ (10. In the ﬁrst order in both symmetry breaking perturbations. as already mentioned. one gets m π + − m π 0 = (4. corrections of order αm q require renormalization but are also expected to be small (Gasser & Leutwyler 1982).9 ± 0. quark masses and electromagnetic interactions.93) To calculate the π + –π 0 mass difference we have to combine the results of Section 10.10.2) MeV. Therefore.48). using the ﬁrst equation (10.95) 2 C = −(2/ f π ) 0|uu|0 ¯ (10.48).95) taken together can be used to estimate the quark mass ratios (Weinberg 1977). Corrections to this result due to the π 0 –η mixing caused by m u = m d can be estimated to be negligibly small (Gasser & Leutwyler 1982). Eqs. we have. Although the error due to the resonance saturation is hard to estimate accurately this result should be regarded as at least in qualitative agreement with the experimental value 4. m 2 + = m 2 0 + (m 2 + )γ π π π where m 2 0 = C(m u + m d ).94) and the K+ –K0 mass difference calculated in the ﬁrst order of the explicit chiral symmetry breaking is not purely electromagnetic but it also reﬂects a possible m u – m d difference. π and therefore m π + − m π 0 ≈ (m 2 + )γ /2m π 0 π Using f π = 93 MeV and f ρ = (145 ± 8) MeV. again using (10. Finally. (10.

2 we get m2 0 − m2 + + m2 + md K π K = = 1.1 md m2 0 − m2 + + m2 + K π K We stress that these results are valid in the ﬁrst order of the chiral symmetry breaking. .368 10 Spontaneous and explicit global symmetry breaking Section 10.8 mu 2m 2 0 + m 2 + − m 2 0 − m 2 + K π π K m2 0 + m2 + − m2 + ms K π = K = 20.

2) 2 4 This lagrangian describes the interaction of the massive particle ϕ and the massless Goldstone boson χ. the theory contains states of negative norm. 369 . for instance.1 Higgs mechanism The proof of Goldstone’s theorem given in Chapter 9 for theories with spontaneously broken global symmetry requires Lorentz invariance and Hilbert space with positive-deﬁnite scalar products.1) = exp(−i ) . For a U (1) symmetric theory it has to be a complex scalar ﬁeld. 0|χ |0 = 0 and deﬁning the physical ﬁelds ϕ = ϕ − v. In consequence. For the real ﬁelds √ = (ϕ + iχ )/ 2 we can choose. The lagrangian for a single self-interacting complex ﬁeld L = ∂µ ∗ µ ∂ − λ( ∗ + m 2 /2λ)2 (11. 0|ϕ |0 = v = (−m 2 /λ)1/2 . Abers & Lee (1973) and references therein).2) for m 2 < 0. This example illustrates all the essential points. The U (1) is invariant under global U (1) transformation: U symmetry is spontaneously broken (see Section 1. Goldstone’s theorem does not hold and the so-called Higgs mechanism operates (see.11 Higgs mechanism in gauge theories 11. In a covariant gauge. for instance. The Higgs mechanism requires the presence of a scalar ﬁeld coupled to the gauge ﬁeld. We shall consider the simplest example of the U (1) gauge theory. In a gauge in which the theory has only states of positive norm it is not manifestly covariant. Gauge theories do not obey both requirements simultaneously. χ = χ we get L = 1 (∂µ ϕ∂ µ ϕ + ∂µ χ∂ µ χ) − v 2 λϕ 2 − 1 λ(ϕ 2 + χ 2 )2 − λϕv(ϕ 2 + χ 2 ) (11.

Choosing the gauge function to be −η (x) = −η(x)/v and using (11. the particle interpretation of the ﬁelds ρ and η is the same as that of ϕ and χ . This is easiest to see working with the angular variables.4) and in terms of the ﬁelds ρ = ρ − v and η = vη the lagrangian (11.8) µ 4 2 2 4 )∗ (Dµ ) − λ( ∗ + m 2 /2λ)2 (11. a lagrangian invariant under local gauge transformation is L = − 1 F µν Fµν + (Dµ 4 where Dµ = ∂µ + ig Aµ The gauge freedom has profound implications for the mechanism of spontaneous symmetry breaking. T = 0 i −i 0 as (11.7) AU (x) = Aµ (x) + (1/gv)∂µ η(x) µ Since the lagrangian (11.370 11 Higgs mechanism in gauge theories It is also useful to introduce ‘angular’ variables and parametrize the ﬁeld follows (see (9.4) we get (in the real basis) 0 U U (x) − 0| (x)|0 = [ρ (x) − v] 1 (11.3) Using the freedom in our choice of the vacuum expectation value among the U (1) degenerate set (Section 1. There is a scalar meson ρ with the bare mass (2λv 2 )1/2 and a massive vector meson with mass gv and no particle corresponding to the . called the unitary gauge (introduced in Section 1. the particle spectrum is evident.1) reads L = 1 (∂µ ρ∂ µ ρ + ∂µ η∂ µ η) − v 2 λρ 2 + cubic and higher order terms 2 (11.78)) (x) = ρ (x) exp[−iη (x)T ] 0 1 = ρ sin η ρ cos η .6) is gauge-invariant we can write it in terms of ﬁelds U and AU and then µ U L = − 1 F Uµν Fµν + 1 ∂µ ρ∂ µ ρ + 1 g 2 (ρ + v)2 AU AUµ − 1 λ(ρ 2 + 2ρv)2 (11.6) where ρ =ρ −v In this gauge. Now let us introduce a gauge ﬁeld Aµ .5) Thus.3).2) we can write (x) − 0| (x)|0 = [ρ (x) − v] exp[−iη (x)T ] 0 1 (11.

the gauge symmetry is spontaneously broken and no massless Goldstone boson remains in the physical spectrum.11) where ( p 2 ) is deﬁned by the gauge-invariant vacuum polarization tensor. Let us ﬁrst take a gauge ﬁxed by the standard gauge-ﬁxing term −(∂µ Aµ )2 /2a. Returning to the variables ϕ and χ we get from (11. Such divergences vanish in the S-matrix elements but this we know from the proof of renormalizability of a spontaneously broken gauge theory given in covariant gauges (Abers & Lee (1973) and references therein).10) ϕ(massive) : i/( p 2 − 2λv 2 + iε) 2 χ(massless) : i/( p + iε) The full gauge boson propagator µν ( p) = −iDµν 1 1− ( p2 ) (11. Because of the massive vector meson propagator which for large k grows like kµ kν /m 2 k 2 the Green’s functions contain divergences that cannot be removed by the renormalization counterterms. ≡i µν ( p 2 ) = i(gµν p 2 − pµ pν ) ( p 2 ) (11. now massive. Quantization in the unitary gauge gives a theory which is manifestly unitary order-by-order in perturbation theory but not manifestly renormalizable.6) the following lagrangian: L = − 1 F µν Fµν + 1 (∂µ ϕ∂ µ ϕ + ∂µ χ∂ µ χ) − λv 2 ϕ 2 − (1/2a)(∂µ Aµ )2 4 2 + 1 g 2 v 2 Aµ Aµ + gv Aµ ∂ µ χ 2 − g Aµ ∂ µ ϕχ + g Aµ ∂ µ χϕ + 1 g 2 Aµ Aµ (ϕ 2 + 2ϕv) 2 + 1 g 2 Aµ Aµ χ 2 − 1 λ(ϕ 2 + χ 2 )2 − λ(ϕ 2 + χ 2 )ϕv 2 4 (S0 ) ˜ ( S0 ) (SI ) (11.1 Higgs mechanism 371 ﬁeld η which has disappeared from the lagrangian. With the standard methods of Chapters 2 and 3 we get then the following free particle propagators (for example.12) gets a contribution from the remaining terms in the lagrangian and in particular .9) One can now proceed in several different ways. in the Landau gauge a → 0): Aµ (massless) : −i(gµν − pµ pν / p 2 )/( p 2 + iε) ≡ −iDµν (11. The η degree of freedom has been traded for the longitudinal component of the vector ﬁeld. The Higgs mechanism is one of the fundamental ideas for condensed matter physics and it is also the crucial part of uniﬁed electroweak theory (see Chapter 12). Thus. Hence it is useful to extend our discussion to covariant gauges.11. One possibility is to take S0 as the free lagrangian of the theory with the corresponding deﬁnitions of the free particle states (which may not be physical states).

16) Note also that the χ propagator remains massless in the Landau gauge.3 where other mechanisms for spontaneous gauge symmetry breaking are discussed. Indeed. One sometimes says that the coupling (11.17) 2ξ Then. Secondly. such an interpretation is certainly procedure˜ and gauge-dependent.15) which can be rewritten as a combination of the standard propagator for a massive vector boson and of a term corresponding to a scalar negative norm boson coupled to the source of the vector meson gradiently µν ( p) = −i gµν − pµ pν g2v2 p2 1 1 pµ pν −i 2 2 2 2 v 2 + iε −g g v p + iε (11. In this context one should observe that the massless Goldstone boson contributes only to the pµ pν term in the µν but by gauge invariance the gµν term must then be present. Finally we can get rid of the coupling gv Aµ ∂ µ χ in the lagrangian by a clever choice of gauge. χ) = − (∂ µ Aµ − ξ gvχ)2 (11. although quadratic in ﬁelds too.13a) (11. This remark will be particularly relevant in Section 11. The presence of the mixing term gv Aµ ∂ χ leads to a system of coupled differential equations for Aµ and χ propagators and in the limit a → 0 we get (11.13a) of a gauge boson to a massless Goldstone boson is the origin of the gauge boson mass and of the spontaneous breaking of gauge invariance. This so-called Rξ gauge is in wide use. is counted as an interaction term.14) = −i(gµν − pµ pν / p 2 )/( p 2 − g 2 v 2 ) (11.18) . integrating by parts when necessary. χ) = 1 g 2 v 2 Aµ Aµ − 1 ξ g 2 v 2 χ 2 − (1/2ξ )(∂µ Aµ )2 2 2 (11. Take the gauge-ﬁxing term as 1 f (Aµ . we can as well take S0 + S0 as our free particle µ lagrangian. The coupling disappears from the theory as well as its previous interpretation.15) as the free boson propagator.372 11 Higgs mechanism in gauge theories ˜ from S0 which. ˜ S0 + f (Aµ . ˜ In the lowest order the S0 contribution is ≡ −i( pµ pν / p 2 )g 2 v 2 ≡ +ig 2 v 2 gµν Hence ( p2 ) = g2 v 2 / p2 and consequently µν ( p) (11.13b) (11.

11.2 Spontaneous gauge symmetry breaking

373

and the free propagators are −i pµ pν gµν − (1 − ξ ) 2 2 + iε −M p − ξ M2 gµν − pµ pν /M 2 pµ pν /M 2 = −i 2 −i 2 p − M 2 + iε p − ξ M 2 + iε 2 2 χ : i/( p − ξ M + iε) Aµ : p2 ϕ: i/( p 2 − 2λv 2 + iε) where M = gv. In the limit ξ = 0 we get the Landau gauge and for ξ = 1 the so-called ’t Hooft–Feynman gauge. The unitary gauge is recovered in the limit ξ → ∞. For any ﬁnite value of ξ there are unphysical poles at p 2 = ξ M 2 in the gauge boson propagator and in the χ propagator. They cancel, however, in the S-matrix elements which are obtained from the Green’s functions by removing external lines, setting external momenta on the mass-shell and contracting tensor indices with appropriate physical polarization vectors. This can be shown, for instance, by proving that the renormalized S-matrix is independent of ξ (Abers & Lee 1973).

11.2 Spontaneous gauge symmetry breaking by radiative corrections In the model of the previous section gauge symmetry has already been spontaneously broken at the tree level. Here we discuss spontaneous gauge symmetry breakdown by radiative corrections (Coleman & Weinberg 1973). This can occur when elementary scalar ﬁelds of zero mass are present in the theory (Georgi & Pais 1977). In the tree approximation there is then no spontaneous symmetry breaking because in this approximation a non-zero vacuum expectation value of the scalar ﬁelds can appear only as a result of interplay between the 4 interaction terms and the scalar mass terms. Spontaneous symmetry breaking is, however, produced by non-zero vacuum expectation values induced by the higher order corrections. This is also an example of a dimensional transmutation: a dimensionful parameter 0| |0 is generated in a quantum ﬁeld theory whose classical limit is described by a scale-invariant lagrangian. We consider ﬁrst λ 4 theory with a massless meson ﬁeld and discuss spontaneous breaking of global reﬂection symmetry → − . The lagrangian density for this theory written in terms of the renormalized quantities L = 1 (∂µ )(∂ µ ) − (λ/4!) 2

4

+ 1 A(∂µ )(∂ µ ) − 1 B 2 2

2

− 1/4!C

4

(11.19)

contains the usual wave-function and coupling-constant renormalization counterterms. Also a mass renormalization counterterm is present, even though we are

374

11 Higgs mechanism in gauge theories

studying the massless theory, because the theory possesses no symmetry which would guarantee vanishing bare mass in the limit of vanishing renormalized mass (the scale invariance is broken by anomalies). To study spontaneous symmetry breaking in such a theory we will use the effective potential formalism developed in Section 2.6. To the one-loop approximation the effective potential is given by (2.165) and (2.166) (with m 2 = 0 and with the contributions from the mass and coupling-constant counterterms included) V = λ 4 1 2 1 1 ϕ + 2 Bϕ + Cϕ 4 + 4! 4! 2 d4 k λϕ 2 ln 1 + 2 (2π )4 2k (11.20)

In this expression we have performed the usual Wick rotation so the integral is in Euclidean space: d4 k = 1 k 2 dk 2 d 4 , 4 = 2π 2 . Observe also that, even though 2 for m 2 = 0 the series in (2.166) is disastrously IR divergent, its sum exhibits only a logarithmic singularity at ϕ = 0. Cutting off the integral in (11.20) at k 2 = 2 (exceptionally we use the UV cut-off rather than the dimensional regularization method), performing the elementary integration and throwing away terms which vanish as 2 goes to inﬁnity we obtain V = λϕ 2 λ 4 1 2 1 λ 2 2 λ2 ϕ 4 1 ln ϕ + − ϕ + 2 Bϕ + Cϕ 4 + 2 2 2 4! 4! 64π 256π 2 2 (11.21)

Imposing now the deﬁnitions of the renormalized mass and coupling constant we determine the value of the counterterms. Deﬁning the renormalized squared mass of the ﬁeld (x) as the value of the inverse propagator at zero momentum and recalling the interpretation following (2.164) of the derivatives of the effective potential we get d2 V /dϕ 2 |ϕ=0 = 0 (11.22) as the mass renormalization condition. We take the derivative at ϕ = 0 because we are interested in the inverse propagator of the original (not shifted) ﬁeld (x); note also that the perturbative calculation of V (ϕ) makes use of the Feynman rules for the lagrangian written in terms of the ﬁeld (x) and therefore it is based on (2.164) even though the theory may be spontaneously broken. From (11.21) and (11.22) we get B = −λ

2

/32π 2

A possible deﬁnition of the renormalized coupling constant would be to identify it with the value of the 1PI four-point function in the original ﬁelds (x) at zero external momenta d4 V /dϕ 4 |ϕ=0 = λ Unfortunately, for a massless ﬁeld we encounter the logarithmic IR singularities.

11.2 Spontaneous gauge symmetry breaking

375

In the standard renormalization procedure one deﬁnes in such a case the coupling constant at some off-mass-shell position in momentum space. Using the effective potential formalism it is much more convenient to adopt an alternative deﬁnition, namely d4 V /dϕ 4 |ϕ=M = λ(M) (11.23)

where M is some arbitrary number with the dimension of mass. It is clear from our discussion in Section 2.6 (see (2.156)) that the fourth derivative of V (ϕ) at ϕ = M is the 1PI Green’s function, at zero external momenta, for the shifted ﬁelds (x) = (x) − M. In particular if we take M = 0| |0 it is the four-point function taken at zero external momenta for the physical massive ﬁeld ¯ (x): 0| ¯ (x)|0 = 0. Imposing condition (11.23) on the effective potential V (ϕ) given by (11.21) we ﬁnally get V = λ(M) 4 λ2 (M)ϕ 4 25 ϕ2 ϕ + ln 2 − 2 4! 256π M 6 (11.24)

We see that the one-loop radiative corrections have turned the minimum at the origin ϕ = 0 into a maximum and caused a new minimum of V to appear at λ(M) ln( 0| |0 /M) = − 32 π 2 + O(λ) 3 (11.25)

Thus, the one-loop corrections have generated spontaneous breaking of reﬂection symmetry. However, in our example, the new minimum lies outside the range of validity of perturbation theory as higher orders will give higher powers of λ ln(ϕ 2 /M 2 ) which according to (11.25) is large for ϕ = 0| |0 . Physically interesting theories in which spontaneous symmetry breaking can be studied in perturbation theory are gauge theories with elementary massless scalar ﬁelds and possibly fermion ﬁelds coupled to the gauge ﬁelds. Let us consider the case of a single gauge ﬁeld coupled to a charged scalar to see qualitatively what happens. In the one-loop approximation the effective potential of the scalar sector gets a contribution from the scalar loops which, aside from irrelevant numerical √ factors (there are now two real scalar ﬁelds in the theory: = ( 1 + i 2 )/ 2; ∗ 2 ) ; the effective potential can only depend on the scalar self-coupling is λ( 2 2 2 ϕ = (ϕ1 + ϕ2 ) because the theory is U (1)-invariant), have the same structure as before, and from the gauge ﬁeld loops in Fig. 11.1 arising from the minimal coupling |(∂µ + ig Aµ ) |2 . The trilinear terms lead to diagrams such as

which vanish if we work in the Landau gauge, where the gauge boson propagator

376

11 Higgs mechanism in gauge theories

Fig. 11.1.

is Dµν = −i gµν − kµ kν /k 2 k 2 + iε

This is because we calculate diagrams with zero external momenta, and therefore the momentum of the internal scalar meson is the same as that of the internal gauge boson. It is obvious now that the gauge boson loop contribution to the effective potential has the same structure as the scalar loop contribution, so we get V = λ(M)ϕ 4 + [(1/16π 2 )λ2 (M) + bg 4 (M)]ϕ 4 [ln(ϕ 2 /M 2 ) − where b is some numerical factor. If it happens that λ(M) neglect the O(λ2 ) term and then dV /dϕ = 0 for | ϕ | = ( 0|

1 |0 2 25 ] 6

(11.26) 1 we can

g 2 (M)

+ 0|

2 |0

2 1/2

)

= M exp

1 λ(M) 11 − 6 2b g 4 (M)

(11.27)

Thus, in the one-loop approximation we ﬁnd in our theory spontaneous breaking of the U (1) symmetry and, this time, the new minimum is in the range of validity of our approximation: due to our choice of the renormalization point such that λ(M) g 2 (M), (11.27) constrains only the value of ln( ϕ 2 /M 2 ) and for small λ and g the terms λ ln( ϕ 2 /M 2 ) and g 2 ln( ϕ 2 /M 2 ) are small. A non-vanishing ﬁeld vacuum expectation value generated by radiative corrections is an example of the dimensional transmutation phenomenon. This can be seen more clearly if we choose M = ϕ . Then, from (11.27) λ( ϕ ) =

11 bg 4 ( 3

ϕ )

(11.28)

and we recall that M = ϕ corresponds to on-shell renormalization for the shifted ﬁelds. We can take dimensionless λ and g or g and dimensionful ϕ as independent parameters of our theory. Spontaneous gauge symmetry breaking leads to the physical consequences discussed in Section 11.1: the would-be Goldstone boson combines with the gauge

11.2 Spontaneous gauge symmetry breaking

377

boson to make a massive vector meson m 2 (V ) = g 2 ϕ 2 and the remaining scalar ﬁeld, after shifting, corresponds to a scalar meson of mass m 2 (S) = V ( ϕ ) ∼ g4 ϕ 2. Physics cannot depend on the arbitrary parameter M but one choice of M can be more convenient than the others, particularly since we are working in perturbation theory and have to control higher order corrections. We have already learned that the perturbative calculation of ϕ is reliable when the renormalization point M is such that λ(M) g 2 (M) 1. It is also worth noting that from (11.27) ϕ = M0 e11/6 (11.29)

where λ(M0 ) = 0. Usually, we suppose that λ(M) and g(M) are given at some ˜ scale M = M, for example, at the grand uniﬁcation scale, and we can then use renormalization group arguments to look for a renormalization scale suitable for our calculation (for example, M0 ). In our renormalization procedure the scalar quartic coupling is deﬁned in terms of the fourth derivative of V . Thus, to study λ(M) one should derive the RGE for this fourth derivative (see Coleman & Weinberg (1973) and Gildener (1976)). We know, however, that the function β(λ) in the one-loop approximation does not depend on the renormalization conventions. Therefore to the lowest order we can use the results of calculations which adopt momentum space renormalization conventions. The one-loop RGEs for λ and g then read 16π 2 M(dλ/dM) = Aλ2 (M) + B λ(M)g 2 (M) + Cg 4 (M) 16π 2 M(dg 2 /dM) = −bg 4 (M) (11.30)

where the dimensionless coefﬁcients A, B and C are given, respectively, by the diagrams shown in Fig. 11.2. To solve (11.30) we introduce α = λ/g 2 . Then we get (t = ln M) 16π 2 dα dα = −bg 2 (t) 2 = Aα 2 (t) + Bα(t) + C 2 (t) dt g dg = A(α − α1 )(α − α2 ), α1 < α2

(11.31)

where B = B + b. For real roots there is a UV stable ﬁxed point at α = α1 < α2 ˜ ˜ ˜ and an IR stable ﬁxed point at α = α2 . Thus, if 0 < α( M) = λ( M)/g 2 ( M) < α1 we expect the RGEs (11.31) to have a solution α(M) = 0 for M = 0. A direct integration gives exp −b

0 ˜ α( M)

dα Aα 2 + Bα + C

=

˜ 1 2 ˜ g 2 ( M) M =1+ g ( M)b ln (11.32) ˜ g 2 (M) 16π 2 M

378

11 Higgs mechanism in gauge theories

Fig. 11.2.

**or, more explicitly, α(M) = 0 for 16π 2 ˜ ˜ F(α( M)) M = M exp − ˜ g 2 ( M) where ˜ F(α( M)) = 1 1 − exp −b b
**

˜ α( M) 0

(11.33)

Aα 2

dα + Bα + C

(11.34)

˜ ˜ ˜ ˜ If α( M) is of the order O(1) (i.e. if λ( M) ∼ g 2 ( M)) then F(α( M)) is also of order unity and M O(1) = exp − ˜ ˜ g 2 ( M) M (11.35)

˜ Thus, using (11.29) we conclude that for a reasonably small value of g 2 ( M), the dimensional transmutation can generate a scale M of spontaneous symmetry ˜ breaking which is enormously smaller than the scale M at which as we suppose ˜ is given. This small ratio of mass scales is not put into the theory by hand α( M) but arises automatically from the assumption of a vanishing scalar mass. Thus in order for this to happen some massless scalars which escape getting masses ˜ ˜ of order M must be available in our theory. For instance, if M is the grand uniﬁcation scale they must remain massless after the spontaneous breakdown of the grand uniﬁcation group. Such a situation can perhaps arise naturally from ˜ supersymmetries which are unbroken at M. The mechanism described here very much resembles the generation of the scale at which QCD coupling becomes large (see (7.53a)). It can also be generalized to the case of several quartic scalar couplings present in the theory (Gildener & Weinberg 1976). The generation of a very small ratio of mass scales by radiative corrections and ˜ the dimensional transmutation phenomenon is to be contrasted with making M/ M very small in the tree approximation. The latter is highly ‘unnatural’. To be more speciﬁc, we have in mind the following: consider a gauge group G which we

11.3 Dynamical breaking of gauge symmetries

379

˜ want to be spontaneously broken to G 1 at some scale M, and G 1 to be broken to ˜ G 2 at M with M/ M very small. It turns out that one can obtain such a pattern of spontaneous symmetry breaking in the tree approximation by a proper choice of the scalar potential, but only at the expense of tuning the parameters of the potential to an accuracy much higher than the neglected one-loop corrections. This is called the gauge hierarchy problem. The formalism presented in this section can be extended to non-abelian gauge theories with scalars and also fermions (see Coleman & Weinberg (1973) and Problem 11.3).

11.3 Dynamical breaking of gauge symmetries and vacuum alignment Dynamical breaking of gauge symmetry We consider now the possibility of spontaneous gauge symmetry breaking in the absence of elementary scalar ﬁelds. The mechanism presented here is often termed dynamical gauge symmetry breaking. Our discussion in this section is closely connected with that in Section 10.3. Imagine a spontaneously broken chiral global symmetry G of some strong gauge interactions with gauge group G H (we shall call it the hypercolour group) also to be broken explicitly by weak interactions with a gauge symmetry group G W ; some of the chiral currents of G couple to gauge bosons of G W . By assumption, it is a good approximation to treat G W interaction to the lowest order (see (10.64)). In Section 10.3 we have learned how to ﬁnd the preferred vacuum which minimizes the vacuum energy in the presence of the perturbation G W . Assuming the positivity condition for certain current matrix elements the correct vacuum corresponds to the minimal value of the quantity (see (10.75))

α

ˆα ˆα Tr[ XJU XJU ]

(11.36)

built of the transformed currents projected onto the subspace of the broken ˆ generators in the original vacuum; we recall that J α is the linear combination of the α ˆ ˆ G generators corresponding to the G W current gα jµ (x) and XJ α is that part of J α α which contains only broken generators X . Equivalently, the correct vacuum must satisfy Dashen’s conditions (10.59) and (10.60). Assume that the correct vacuum has been found: it gives a speciﬁc orientation of the spontaneously unbroken subgroup H in the G, a set of broken generators X α and the corresponding Goldstone bosons, some of which get masses induced by the perturbation G W . In general one might also single out another subgroup of G: the maximal subgroup S of elements of G which commute with all the generators of G W . Then the coupling of gauge bosons to the currents of G breaks G explicitly to G W × S and we get

380

11 Higgs mechanism in gauge theories

Fig. 11.3.

Fig. 11.4.

the picture of the G symmetry breaking shown in Fig. 11.3 where the orientation of different subgroups is illustrated in Fig. 11.4. In the zeroth order in the G W couplings there are Goldstone bosons associated with each spontaneously broken generator (regions I + II + III). In presence of the G W interactions the Goldstone bosons split into three different groups which are easy to classify. The G W × S symmetry remains the exact symmetry of the lagrangian and therefore the Goldstone bosons in regions I and II remain massless in any order in the G W interactions, as can be proved following the arguments of Section 9.6. The generators in region III do not correspond to exact symmetries of the full lagrangian and the corresponding Goldstone bosons acquire, in general, masses which in the ﬁrst order in G W are given by (10.55) or (10.77) and (10.78). From the calculation of the π + –π 0 mass difference we can see that the masses are of the order g (m ρ ∼ ), where g is the coupling constant of the weak interactions and is the conﬁnement scale of the hypercolour interactions. The Goldstone bosons in regions I and II still split into two groups: clearly, those in region II remain in the physical spectrum; however, those in region I are absorbed by the G W gauge bosons which get masses through the Higgs mechanism. One then talks about the dynamical breaking of the G W gauge symmetry. Let us discuss the mass generation for the G W gauge bosons somewhat more explicitly. The α Goldstone bosons in region I couple to the G W currents jµ

α 0| jµ (0)|π a = i pµ f αa

(11.37)

11.3 Dynamical breaking of gauge symmetries

381

αβ µν

**and therefore the gauge-invariant vacuum polarization tensor i
**

αβ 2 µν ( p )

deﬁned by

= gα gβ

α β d4 x exp(i px) 0|T jµ (x) jν (0)|0 αβ

**= i(gµν p 2 − pµ pν ) has a pole at p 2 = 0 (see also Section 11.1)
**

p 2 →0 αβ 2 µν ( p )

( p2 )

(11.38)

lim

= −(gµν p 2 − pµ pν )

a

g α g β f αa f βa / p 2

(11.39)

It is important to remember the role of gauge invariance in writing (11.39): the Goldstone bosons contribute only to the pµ pν term of the polarization tensor; the gµν term must be present by gauge invariance and we do not need to specify its origin more explicitly. The full gauge boson propagator αβ reads† µν

αβ µν

= −iDµν [δ αβ −

αβ

( p 2 )] = −iDµν (1 +

)−1αβ + higher order terms (11.40)

We are working in the Landau gauge and −iDµν δ αβ is the free propagator in this gauge. Thus, the residue (11.39) gives the vector boson mass matrix m2 = αβ

a

gα gβ f αa f βa

The Goldstone boson decay constants f αa can be expressed in terms of the appropriate traces. Under assumption (10.70) we have

a 0| jµ (0)|π b = i pµ f π δ ab = i pµ f π Tr[Xa Xb ] a where jµ is a G current and consequently 2 m 2 = fπ αβ a 2 ˆ ˆ ˆ ˆ Tr[ J α X a ] Tr[ J β X a ] = f π Tr[ X J α X J β ]

(11.41)

(11.42)

(where no sum has been taken over α and β). This result has been obtained for the vacuum state |0 . For any other vacuum state related by the unitary transformation U |0 we ﬁnd 2 ˆα ˆβ m 2 = f π Tr[ X JU X JU ] (11.43) αβ Comparing with (10.75) we see that E( ) ∼ Tr m 2 (11.44)

The conclusion is that the energetically preferred vacuum alignment and the relative orientation of the groups G W and H in the presence of the perturbation

†

αβ αβ αβ (−i λρ,β γ )(−iD δ γ β )). ρν µν = (−iDµν δ ) + (−iDµλ δ

382

11 Higgs mechanism in gauge theories

G W are such that the trace of the vector boson mass matrix is the minimal one (the gauge symmetry is broken as little as possible). The weak interactions determine their own pattern of symmetry breaking by the choice of the vacuum orientation. We note also that the vector boson mass matrix obeys, in general, certain relations which are consequences of H invariance. The above ideas may be relevant for the gauge symmetry breaking in the Glashow–Salam–Weinberg theory of the weak interactions (see, for instance, Fahri & Susskind (1981)), however, no convincing model has been constructed yet. We will illustrate them with two simple examples. Examples Consider a strong hypercolour gauge group G H with 2N left-handed fermions and 2N right-handed fermions transforming as the same representation of G H . The theory has G = SUL (2N ) × SUR (2N ) chiral symmetry which, we assume, breaks down spontaneously to H = SU (2N ). The embedding of H in G is characterized by a fermion condensate 0| ¯ Li R j |0 = i j (11.45) As in Section 9.6, assuming i j = vδi j , the chiral group is broken to SUL+R (2N ). Any equivalent orientation of the unbroken SU (2N ) corresponds to a condensate obtained from (11.45) by a unitary transformation

† † UR UL = vURUL

(11.46)

† (remember that U † U = UR UL ; U is an unitary operator, UL(R) are unitary † matrices), i.e. it is speciﬁed by a unimodular unitary matrix URUL . We take the weak gauge group to be G W = SUL (2) × U (1) with fermions in N left-handed G W doublets and 2N right-handed singlets, and with the U (1) charge assignments as follows (see Chapter 12):

Y =

Ui Di q

UiR

L

DiR

1 2

q+

q−

1 2

Consider ﬁrst the case N = 1. Denoting the gauge bosons of the Weinberg– α Salam model by Wµ , Bµ the weak coupling is L=−

α α gWµ ¯ L γ µ T α L

− g Bµ ( ¯ R γ µ T 3

R

+ q ¯ γµ )

(11.47)

where T α is an SU (2) generator normalized to Tr[T α T β ] = 1 δ αβ and q is the mean 2 electric charge of the doublet. Assuming the condensate (11.45) and expressing the

11.3 Dynamical breaking of gauge symmetries

383

α G W currents jµ in terms of the G currents corresponding to the unbroken subgroup (vector currents) and to the broken generators (axial currents) jL = 1 (V −A), jR = 2 1 (V + A) one gets in the ( L , R ) basis the following expression: 2

Tα ˆ Jα = g +

1 g 2

0 Tα

+g −T

α

q q+T + 1g 2

3

= −T 3

1 g 2

Tα T

α

+ 1g 2

2q + T 3 2q + T 3 (11.48)

T3

**Inserting (11.48) into (11.42) gives the familiar vector boson matrix W1 W2 W3 B After diagonalization one gets m W1,2 = 1 g f π , 2 mγ = 0 where
**

W,

W1 1 2 2 g fπ 4

W2

1 2 2 g fπ 4

W3

B

1 2 2 g fπ 4 1 2 − 4 gg f π

2 − 1 gg f π 4 1 2 2 g fπ 4

mZ =

1 f (g 2 2 π

+ g 2 )1/2 = m W / cos

W

(11.49)

**the Weinberg angle, is given by cos
**

W

=

g (g + g 2 )1/2

2

**and γ = cos Z = − sin
**

WB

+ sin + cos

W W3 W W3

WB

In this model all the Goldstone bosons have been used to give masses to the weak gauge bosons. There is no problem of vacuum alignment since all the equivalent vacua (11.46) can be transformed into (11.45) by the gauge transformations in the weak sector. For N > 1 left-handed doublets with the same U (1) charge assignment the situation is more complex. If we introduce the 4N -plet = (U1L D1L . . . D N L D N L U1R D1R . . . U N R D N R ) then the Salam–Weinberg coupling is l Tα ⊗ 1 α L = −gWµ ¯ γ µ 0 − g Bµ ¯ γ µ q q + T3 ⊗ 1 l (11.50)

where the 2N × 2N block T α ⊗ 1 is a direct product of the 2 × 2 weak isospin l and the N × N matrix on the space of N doublets. The weak coupling with

384

11 Higgs mechanism in gauge theories

the gauge symmetry G W = SUL (2) × U (1) breaks the original chiral symmetry G = SUL (2N ) × SUR (2N ) of the strong interaction sector into G → G W × S where S = SUL (N ) × SUR (N ) × SUR (N ) with the ﬁrst SU (N ) corresponding to horizontal transformations among N left-handed doublets, and the second (the third) SU (N ) corresponding to transformations among N right-handed singlets UiR (DiR ); UiR and DiR have different charges and therefore the original SUR (2N ) is broken into SUR (N ) × SUR (N ). The spontaneous breakdown of G into SUL+R (2N ) induced by (11.46) leaves the subgroup SUL+R (N ) of S unbroken. The spontaneous breakdown of the original SU (2N ) × SU (2N ) chiral symmetry produces 4N 2 − 1 Goldstone bosons. Three of them are used to give masses to the gauge bosons. Again, there is no vacuum alignment problem. The breaking of S into SU (N ) which remains the exact symmetry of the full theory leaves 2(N 2 − 1) exactly massless Goldstone bosons. They are electrically neutral because they correspond to generators of transformations among fermions with the same charge. There remain 2(N 2 − 1) charged Goldstone bosons which do not correspond to generators of exact symmetries of the full lagrangian and they should, therefore, receive masses from the G W interactions. Splitting the weak currents into the ˆ ˆ unbroken YJ α and the broken parts XJ α as in (11.48) Tα ⊗ 1 2q + T 3 ⊗ 1 l l 1 1 Y ˆα J ≡ 2g + 2g α 3 T ⊗1 l 2q + T ⊗ 1 l (11.51) −T α ⊗ 1 −T 3 ⊗ 1 l l 1 1 X ˆα + 2g J ≡ −2g Tα ⊗ 1 l T3 ⊗ 1 l it is evident from (10.79) that the Goldstone bosons receive no mass in the ﬁrst order in G W . The broken generators X a are of the block-diagonal form −A A where the As are 4N 2 − 1 matrices which can be easily constructed for any N , and ˆ ˆ therefore for YJ and XJ given by (11.51) the two contributions to the trace in (10.79) cancel each other. Actually, one can see that if G W commutes with one of the chiral groups SUL (2N ) or SUR (2N ) then there will be pseudo-Goldstone bosons which remain massless to the lowest order in G W . Indeed, one sees directly from (10.66), using H invariance as in (10.69), that the -dependent (U -dependent) part of the vacuum energy E( ) relevant for the pseudo-Goldstone boson masses (10.79) comes from the term E( ) = − 1 2 d4 x

µν

(x)

α

† α † α 2 gα 0|T UL jµL (x)ULUR jνR (0)UR |0

(11.52)

α α α α whereas jµL jνL and jµR jνR terms contribute only the

-independent constants.

for example. Let us ﬁrst assume that the SU (4) × SU (4) symmetry of the G H lagrangian is broken by the ¯ ¯ 0|ULi UR j |0 = 0|DLi DR j |0 = vδi j (11. Such a model is very instructive because then the U (1) boson exchange contributes the -dependent term to the E( ) and the pattern of the G W breakdown depends on the U (1) charge assignments. To be speciﬁc we consider a model with four left-handed and four right-handed fermions with the G W representation content given by U1 D1 Y = U2 D2 − .11.54) where in the product A ⊗ B the ﬁrst matrix is in the 2 × 2 weak space and the second one is on the space of doublets. 11. ( + 1 ). as in Fig. so it is SUL (2N )× SUR (2N )-invariant and hence -independent. The charged pseudo-Goldstone bosons receive masses in the next order in G W due to the mass splitting of the γ and Z which has to be taken into account in that order.55) . 2 U2R (− + 1 ).5. D2R (− R L (11. The U (1) contribution is therefore SUL (2N )invariant and also SUL+R (2N )-invariant because the vacuum is. U1R L D1R ( − 1 ). In the model discussed above there remain massless neutral Goldstone bosons in the physical spectrum because the SU (N ) group of transformations between doublets remains the exact symmetry of the theory. This can be avoided by giving the left-handed doublets different U (1) charge.5.53) 1 ) 2 . 2 L. 2 − Similarly to (11.3 Dynamical breaking of gauge symmetries 385 Fig.50) the G W coupling in the (U1L D1L U2L D2L U1R D1R U2R D2R )) is Tα ⊗ 1 l α L = −gWµ ¯ γ µ + 1 ⊗ T3 l 1 ⊗ T3 l 0 − g Bµ ¯ γ µ basis ( = 0 T3 ⊗ 1 l (11. 11. In our example the U (1) current also does not contribute to E( ) (although it has both the left-handed and the right-handed parts) because the left-handed U (1) current is an SUL (2N ) singlet.

79).79) cancels. as before the contribution of the SU (2) current to the two terms under the trace in (10. l example.386 11 Higgs mechanism in gauge theories condensates.57) are degenerate in mass and m 2 1 U2 = m 2 1 D2 ∼ g 2 2 M 2 U D 2 2 2 m U1 D2 ∼ g ( + 1)M (11. D1 ) with one of (U2 . Assuming that the integral M 2 is dominated by the lowest-lying vector mesons of the correct quantum numbers we ﬁnd. (11.56) There are 15 Goldstone bosons corresponding to the broken generators X= 1 −A A 2 where A represents T a ⊗ 1 2T a ⊗ T 3 . We can now split the G W currents into unbroken and broken parts in the vacuum (11.56) and (11.79) and (11.53). for l. It is clear from our previous discussion that only the U (1) current is interesting to us.55). Therefore. An explicit calculation l based on (10. therefore. We must.57) √ √ 1 0 i 0 2 2 0 0 0 0 0 0 linking one of (U1 . us that Goldstone bosons corresponding to broken generators T a ⊗ 1 T a ⊗ T 3 and 1 ⊗ T 3 receive no mass in the ﬁrst order in G W . ﬁnd the correct . This deﬁnes the vacuum and the breaking SUL (4) × SUR (4) → SUL+R (4) in the basis (11.57) gives the following masses for the remaining pseudo-Goldstone bosons which are denoted by the ﬂavours linked by the corresponding generators: non-diagonal terms of the mass matrix vanish and bosons corresponding to the two generators like (11. D2 ). we ﬁnd an interesting result: for | | > 1 Dashen’s condition is satisﬁed by masses (11.58) 2 m D1 U2 ∼ g 2 ( − 1)M 2 where M 2 is the integral given by (10. as for the π + –π 0 mass difference in Section 10.4. For the U (1) current we have ˆ J = 1g 2 T3 ⊗ 1 + 2 1 ⊗ T3 l l T3 ⊗ 1 + 2 1 ⊗ T3 l l + 1g 2 −T 3 ⊗ 1 l T3 ⊗ 1 l (11.55) turns out to be unstable.56) tells l. for the U (1) charge assignments for doublets such that | | < 1 the vacuum (11.55) is the preferred vacuum in the presence of perturbation G W .78). 1 ⊗ T 3 and eight matrices of the form. However. A look at (10. M 2 > 0.58) and the vacuum (11. 1 0 −i 0 0 0 1 1 0 0 0 0 or (11.

54) is ‘as far as possible’ SUL+R (4)-invariant. for | | < 1 the right transformation is the one which gives ˆ U† JU = − −1 − − 1+ 1− −1 + . is the U (1) current in the original basis (11. Since the trace of B 2 should be the minimal one. for | | < 1 one can imagine ˆ that a transformation U † J U . According to (11. simultaneously transforms the other term of the U (1) current to such a form that the full transformed current has ‘better’ SUL+R (4) symmetry than the original one. Before we ﬁnd this transformation let us notice that in the original basis (11. However. It is then clear that for | | > 1 that basis is the right one: the L given by (11. respectively.3 Dynamical breaking of gauge symmetries 387 vacuum in the set of vacua obtained from (11. the transformed current should have both 4 × 4 blocks as similar as possible (A + B ≈ A − B ≈ A). Hence.55) by all unitary transformations U of SU (4) × SU (4). split into unbroken and broken parts.s. The right transformation can be found by noting that the transformed U (1) current.54).h.11.54) the term in the lagrangian proportional to is SUL+R (4)-invariant. is as follows: U† − U −1 − − 1+ −1 + 1− = A A + B −B where the matrix on the l. which breaks the SUL+R (4)-invariance of the term.36) the right transformation is the one which ˆ2 ˆ2 minimizes the Tr XJU or maximizes the Tr YJU of the projection of the transformed current on the subspace of the original broken (X a ) and unbroken (Y a ) generators.

2 Check by explicit calculation that at the tree level the unphysical ξ -dependent poles cancel in the S-matrix for the process ϕ1 + ϕ2 → A1 + A2 where ϕ and A are the physical scalar particle and the massive vector meson of the U (1) theory in Section 11. respectively. one can easily ﬁnd a set of broken and unbroken generators corresponding to the new vacuum as well as all the gauge boson masses and pseudo-Goldstone boson masses. Calculate the Higgs particle (physical scalar) mass in terms of the W and Z masses in the SU (2) × U (1) electroweak theory with one massless scalar complex doublet and taking fermions as massless. Thus.1. The new vacuum is then deﬁned by the condensate ¯ ¯ ¯ ¯ D1L U2L D2L U 1L 1 0 0 0 † UL = v 0 0 1 0 0 1 0 0 0 0 0 1 U1R D1R U2R D2R UR ¯ so the electrically charged composite ﬁeld U2 D1 acquires a non-zero vacuum expectation value. Problems 11.388 11 Higgs mechanism in gauge theories From (11. for dynamical reasons. the U (1)em subgroup of the gauge group SU (2) × U (1) is broken and the photon has a mass when | | < 1.3 Extend the discussion of Section 11.1 Derive the remaining Feynman rules for the U (1) theory of Section 11. 11.1 in the Rξ gauge. .2 to non-abelian gauge theories including fermions. lead to different theories depending on the speciﬁc assignment of the chiral multiplets to the weak group representations. 11.54) one sees that this is the transformation D1R ↔ U2R . An interesting conclusion of the above exercise is that the coupling of a weak gauge group to the chiral currents of a spontaneously broken chiral symmetry of strong interactions may. Of course.

(Any reader unfamiliar with the two-component spinor notation should consult Chapter 1 and Appendix A.1) 1 c 3 . It will acquire the proper meaning of charge conjugation (as in Chapter 1) after the electroweak symmetry breaking to the electromagnetic U (1). supplemented by the idea of minimal uniﬁcation with electromagnetic forces. 3 is the generation index. It is just used to denote the left-handed Weyl ﬁelds which are singlets of SUL (2). respectively. in accordance with general remarks in the Introduction and as required by uniﬁcation with electromagnetism. The complete SUc (3) × SUL (2) × UY (1) theory is called the Standard Model of elementary interactions. 1. 2. 1. The underlying symmetry group SUL (2) × UY (1) emerges from study of the charged weak currents. − 1 2 ec (1.) The ﬁrst and second entries show the transformation properties under the colour SU (3) and the weak SUL (2) groups.12 Standard electroweak theory The basic concepts of the electroweak theory (Glashow 1961. and T 3 is the third component of its weak isospin SUL (2). 2. + 3 dA c uA 3 . 2. + 6 qA (12. The last entry shows the hypercharge of each ﬁeld. In terms of the left-handed Weyl ﬁelds each generation consists of: lA 1. We recall that the hypercharge is deﬁned as Y = Q − T 3 . − 2 3 where A = 1. where Q is the electric charge of the ﬁeld. The hermitean conjugates of the ﬁelds carrying it will then combine with appropriate † The SU (3) gauge group of QCD was extensively discussed in Chapter 8. 1. At this stage the superscript ‘c’ does not have the same meaning as in Chapter 1. The symmetry is assumed to be local gauge symmetry. a Thus. +1) A 1 3. Weinberg 1967b. we have the following gauge vector ﬁelds:† Wµ for SUL (2) and Bµ for UY (1). The matter ﬁelds consist of quarks and leptons forming three generations. 389 . Salam 1968) have been outlined in the Introduction.

e. It does not destroy the renormalizability of the theory (’t Hooft 1971a. which are strictly massless in the symmetry limit.72): l iA 0 i . We can also introduce the four-component chiral fermion ﬁelds. 1 ) 2 with respect to the full SU (3) × SUL (2) × UY (1) gauge group. We recall that the unbarred (barred) two-component spinors carry lower (upper) Lorentz indices. dA uc ¯A 0 where i = 1. the one-doublet theory is what is called the Standard Model. more precisely.1) must be invoked. 2.390 12 Standard electroweak theory components of the doublets to form Dirac ﬁelds. The Higgs mechanism is the origin of the masses of the W ± and Z 0 vector a bosons (which are linear combinations of the original Wµ and Bµ ﬁelds) and of the masses of elementary fermions. conserved) charges. u2 ≡ c . The left-(right-)handed chiral ﬁelds L(R) ≡ 1 (1 ∓ 2 γ5 ) L(R) are doublets (singlets) under SUL (2) and the electric and colour charges of the left. 2. ( d A )R ≡ ¯ c . c d2 ≡ s c . ( l A )L ≡ ec ¯A 0 (12. c d3 ≡ b c for quarks.and right-handed ﬁelds are pairwise the same. called the Higgs doublet. q2 ≡ . We shall also use the notation ( l1A )L ≡ ( ν A )L . with charge conjugation acting only on their electric and colour (i. νµ ντ . as in (1. Since the presence of more doublets (for constraints on the presence of other representations of scalars see Problem 12. ( e A )R ≡ . u3 ≡ t c c d1 ≡ d c .1) has no particular theoretical motivation (apart from the supersymmetric models which require an even number of Higgs doublets). the so-called Higgs mechanism (discussed in Section 11. . l2 ≡ e c e1 ≡ ec . Writing explicitly the SUL (2) doublet components and using the conventional terminology we have l1 ≡ νe . The gauge symmetry SUL (2) × UY (1) must be spontaneously broken to the electromagnetic UEM (1) or. One Higgs doublet is the minimal set of scalar ﬁelds which allows for the required symmetry breaking. l3 ≡ µ τ c c c e2 ≡ µ .2) q iA 0 0 i ( q A )L ≡ ( u A )R ≡ . e3 ≡ τ c for leptons and q1 ≡ u c t . with the quantum numbers (1. To this end one introduces a doublet of scalar ﬁelds H . b). q3 ≡ d s b c c c c c u1 ≡ u .

mainly by the LEP experiments.1 The lagrangian With the ideas and the formalism introduced so far in this book we are well equipped to formulate the electroweak theory (Glashow 1961. as the correct theory of elementary interactions for the energy range up to O(100 GeV).4). spontaneously broken symmetries can be intuitively depicted as the property of the vacuum which is a big reservoir of the scalars H 0 . 12.3).3) and the components H + and H 0 are complex ﬁelds. the Higgs boson mass range can extend up to O(1 TeV). As discussed in Chapter 9. By proper redeﬁnition of the group generator basis the remaining U (1) can always be interpreted as the electromagnetic UEM (1). The electroweak theory has been conﬁrmed experimentally with the precision of one per mille. and is extensively searched for in present and planned experiments.4) H + (x) H 0 (x) (12. Therefore. In particular. . Thus. the vacuum carries quantum numbers of the component H 0 . However. Weinberg 1967b. This is different from quantum vacuum ﬂuctuations which are singlets with respect to all symmetries.12.and right-handed chiral matter ﬁelds. full experimental clariﬁcation of the Higgs sector is expected to provide an important bridge to physics beyond the Standard Model. according to (12.1 The lagrangian 391 Moreover. As discussed in Chapters 1 and 9. the Higgs scalar has not yet (1999) been discovered. the pattern SU (2) × U (1) → U (1) is unique if the Higgs mechanism is generated by vacuum expectation values of the components of a complex scalar SU (2) doublet ﬁeld H (x).and partly two-loop corrections to the tree level predictions have been tested (see later). interestingly enough. Higgs doublets and only Higgs doublets can have renormalizable Yukawa couplings to the left.4).1) the desired pattern is obtained with H (x) = where 1 0|H (x)|0 = √ 2 0 v (12. they can generate the Dirac mass terms for those particles. With the basis ﬁxed by the charge assignment (12. It may be useful to stress that. after symmetry breaking. It can be elementary or composite or even only the effective description of a strongly interacting (at the scale O(1 TeV)) gauge boson sector (Section 11. From the point of view of the electroweak theory considered in the range up to O(100 GeV). the quantum one. and are eliminated by normal ordering (a redeﬁnition of the vacuum energy) which is understood also to be present in (12.

8) . The generation index A is summed from 1 to 3.ﬁx + Lghosts The gauge boson part has the standard form: a Lgauge = − 1 Wµν W aµν − 1 Bµν B µν 4 4 (12.7) can be rewritten in the more familiar form i a Lmatter = i( l A )L γ µ ∂µ + igWµ T a − g Bµ ( l A )L 2 + i( + i( + i( + i( e A )R γ q A )L γ d A )R γ u A )R γ µ µ ∂µ − ig Bµ ( e A )R q A )L µ µ i a ∂µ + igWµ T a + g Bµ ( 6 i ∂µ − g Bµ ( d A )R 3 2 ∂µ + i g Bµ ( u A )R 3 (12.6) a a a b c where Wµν = ∂µ Wν − ∂ν Wµ − gεabc Wµ Wν and Bµν = ∂µ Bν − ∂ν Bµ are the ﬁeld a strengths of the Wµ and Bµ gauge ﬁelds and εabc are the SUL (2) group structure constants.392 12 Standard electroweak theory Salam 1968. In terms of four-component chiral fermions deﬁned in (12.2) the lagrangian (12. respectively. ’t Hooft 1971a. a a Lkin = 1 Wµ g µν ∂ 2 − ∂ µ ∂ ν Wν + 1 Bµ g µν ∂ 2 − ∂ µ ∂ ν Bν gauge 2 2 b c a b b b Lint = gεabc ∂ µ W aν Wµ Wν − 1 g 2 W aµ W bν Wµ Wν − W aµ W aν Wµ Wν gauge 4 The matter ﬁeld lagrangian in the two-component spinor notation reads i a Lmatter = il¯A σ µ ∂µ + igWµ T a − g Bµ l A ¯ 2 c µ c + ie A σ ∂µ + ig Bµ e A ¯ ¯ i a + iq A σ µ ∂µ + igWµ T a + g Bµ q A ¯ ¯ 6 i c ¯c ¯ + id A σ µ ∂µ + g Bµ d A 3 2 + iu c σ µ ∂µ − i g Bµ u c ¯A ¯ A 3 (12.7) where T a are the SUL (2) generators in the fundamental representation and g. g are the two coupling constants associated with SUL (2) and UY (1) groups. Its kinetic and interaction parts are. b) as a gauge theory.5) (12. respectively. The full classical gauge-invariant lagrangian consists of several parts: LSM = Lgauge + Lmatter + LHiggs + LYukawa + Lg.

2) of physical ﬁelds does not contain the right-handed neutrinos.5): c LYukawa = −YlB A Hi li A ec − YdB A Hi qi A d B − YuB A εi j Hi q j A u c + h. are sterile (T3 = Y = 0) with respect to the electroweak interactions and their inclusion would be a no-cost extension of the theory. Note.† Using the four-component chiral ﬁelds this can be rewritten as LYukawa = −YlB A Hi ( e B )R ( i l A )L − YdB A Hi ( + h. Indeed. (12.13) √ √ 2 2 where Y is the hypercharge generator. 3.c.7). Y AB are 3 × 3 complex Yukawa coupling matrices and summation over the SUL (2) indices i. The SUL (2) × UY (1) symmetry is then broken down to U (1) (which is identiﬁed with the electromagnetic U (1) with Q = T 3 + Y ) because 0 0 T a v = 0. d B )R ( i q A )L − YuB A εi j Hi ( u B )R ( j q A )L (12.8) compared to (12.11) When m 2 < 0 is chosen in (12.9) the Higgs doublet acquires the vacuum expectation value. Therefore. † In the lagrangian (12. that the right-handed neutrinos. 2. if they exist. for a = 1.) The lagrangian for the Higgs ﬁeld has the form i ← a LHiggs = H † ∂ µ − igT a Wµ − g Bµ 2 λ 2 H†H − m2 H † H − 2 i → a ∂ µ + igT a Wµ + g Bµ H 2 (12.10) we have not included the Yukawa couplings for neutrinos since our list (12. (T 3 + Y ) v = 0 (12. i. Such an extension may be necessary in view of the accumulating evidence for neutrino masses. as 2 of SUL (2).12.10) B B where ε12 = −ε21 = −1. however. . the minimum of the potential is for H†H = − v2 m2 ≡ λ 2 (12.12) By SUL (2) rotation we can always redeﬁne the vacuum so that only the vacuum expectation value of the lower component of the Higgs doublet is non-vanishing.9) Interactions of the Higgs doublet with the matter ﬁelds are described by the Yukawa part of the full lagrangian (12. j and over the fermion generation indices A and B is understood. The neutrino Yukawa terms and explicit Majorana mass terms for the righthanded neutrinos can then be added to the electroweak lagrangian without destroying the renormalizability of the theory. (iτ2 H ) j q j = εi j Hi q j is also an invariant of SUL (2).1 The lagrangian 393 (Note the change of signs of the U (1) charges of the SUL (2) singlets in (12.e.c. We have also used the fact that the two-dimensional representation of SU (2) is real (see Appendix E) and iτ2 H transforms as H .

or from the requirement that its coupling to fermions is vector-like with the strength given by their electric charges.394 12 Standard electroweak theory The gauge-ﬁxing term and the Fadeev–Popov ghost term of the lagrangian (12.2 Electroweak currents and physical gauge boson ﬁelds The gauge boson couplings to matter follow from Lmatter and have the form µ a Lmatter ⊃ −g J aµ Wµ − g JY Bµ (12.17) where the Weinberg angle is deﬁned as sin2 θW ≡ g2 .20) . from the tree level structure of the lagrangian. explicit V –A structure. Both left.16) + f f ¯ R γ µY f The SUL (2) current has.14) where (see (12. i.8)) J aµ = f µ JY = f f ¯ L γ µY f f L f ¯ L γ µT a f L f R (12. This combination can be easily found either from the requirement that the photon does not couple to the neutral component of the Higgs ﬁeld. g2 + g 2 tan θW ≡ g g (12. First.19) The orthogonal combination deﬁnes another electrically neutral ﬁeld of the theory: 0 Zµ = 1 3 3 (gWµ − g Bµ ) ≡ cos θW Wµ − sin θW Bµ (g 2 + g 2 )1/2 (12.e.and right-handed fermions contribute to it. 12. Note that the U (1) current is chiral. too. of course. It follows from the uniﬁcation hypothesis that the photon ﬁeld Aµ is a com3 bination of the Yang–Mills ﬁelds Wµ and Bµ . This is a potential source of chiral anomalies (see Chapter 13).5) will be discussed later. we shall discuss the predictions of the theory which follow from the underlying gauge symmetry and the Higgs mechanism. but with different U (1) charges.15) (12.18) and the electric charge is (e > 0) e≡ (g 2 gg = g sin θW + g 2 )1/2 (12. We ﬁnd Aµ = (g 2 1 3 3 (g Bµ + g Wµ ) ≡ cos θW Bµ + sin θW Wµ + g 2 )1/2 (12.

9) L that the minimum of the scalar potential occurs for v 2 = −2m 2 /λ. The W ± and Z 0 boson masses are related by MW = cos θW MZ (12. Moreover.24) The photon remains massless. however. Such a relation is.25) (actually the same result holds with any number of Higgs doublets.12.21) + f f ¯ Rγ µQ f f R (12.14): g g µ 0 µ 0 JZ Z µ = (J 3µ − sin2 θW JEM )Z µ cos θW cos θW with µ JEM = f f ¯ L γ µQ f f L (12.4). for example. 4 2 M Z = 1 (g 2 + g 2 )v 2 4 (12. The physical gauge and Higgs boson spectrum of the electroweak theory is most easily seen in the so-called unitary gauge introduced in Chapter 11. Using the ‘angular’ variables for the Higgs ﬁeld 1 ηa (x) a H (x) = √ exp i T v 2 0 v + h 0 (x) (12. but is just a measure of this departure. which obtain the masses 2 MW = 1 g 2 v 2 . Also from (12. It is worth stressing that the departure in the neutral current from pure V –A Lorentz structure is the basic prediction of the uniﬁed theory. Furthermore. in accord with the fact that the U (1)EM gauge symmetry remains exact. and it depends crucially on the SUL (2)×UY (1) representation content of the Higgs ﬁelds. . obtained after the W ± and Z 0 bosons are given masses by the Higgs mechanism. we again see from (12. three of the four degrees of freedom present in the Higgs doublet H (x) are 1 ± 1 exchanged for the longitudinal components of the gauge bosons Wµ ≡ √2 (Wµ ∓ 2 iWµ ) and Z 0 .23) and choosing the gauge rotations a (x) = −ηa (x)/v.22) µ The mixing angle θW measures the departure of the weak neutral current J Z from pure V –A structure! It is measured experimentally. the reader is invited to check this by explicit calculation). the value of the Weinberg angle has a priori nothing to do with the vector boson masses. and together with the electric charge can be used as independent input parameter ﬁxing the couplings g and g .9) we see that the fourth degree of freedom originally present in the ﬁeld H (x) remains in the physical 2 spectrum as an electrically neutral massive Higgs scalar with the mass Mh = λv 2 . in atomic parity violation experiments (for other measurements of sin2 θW – see Section 12.2 Electroweak currents 395 Its couplings to fermions then follow from (12.

28) The ﬁelds G 0 and G ± are massless at this stage but they will be given masses by the gauge-ﬁxing term. however. the unphysical Higgs ﬁelds (the three Goldstone bosons ‘eaten up’ in the unitary gauge by the longitudinal components of the gauge boson ﬁelds) remain in the lagrangian (and in the Feynman rules). contribute to the S-matrix elements. In lagrangian (12.9) we also have the canonical kinetic terms for h 0 .3) we have renamed the Goldstone boson ﬁelds which in the unitary gauge are ‘eaten up’ by the longitudinal components of the physical massive gauge boson ﬁelds. .396 12 Standard electroweak theory Since in the unitary gauge the theory is not manifestly renormalizable.26) in the scalar potential in (12. Also note that the neutral ﬁeld is split into two real ﬁelds. as discussed in Section 11. one ﬁnds the mass terms for the gauge bosons 1 2 3 LHiggs = · · · + 1 g 2 v 2 (W 1µ Wµ + W 2µ Wµ ) + 1 v 2 (gW 3µ − g B µ )(gWµ − g Bµ ) 8 8 2 − 2 0 ≡ MW W +µ Wµ + 1 M Z Z 0µ Z µ 2 (12.27) In the so-called Rξ gauge these terms are. The terms (12. of course. G 0 and G + ﬁelds as well as their self-interactions 0 ± and interactions with the gauge ﬁelds Aµ . Before ﬁxing the gauge we write† in general H= G+ 1 √ (v + h 0 + iG 0 ) 2 (12. setting external momenta on the mass-shell. However. These terms are straightforward to obtain and their derivation is left as an exercise for the reader. In this class of gauges. the unphysical poles of the Green’s functions do not.ﬁx part of the lagrangian. in the following we choose to work with the covariant gauge-ﬁxing condition.26) where h 0 is assumed to have a vanishing vacuum expectation value. Finally.1. cancelled by the appropriate choice of the Lg.9) one ﬁnds (with v 2 = −2m 2 /λ ﬁxed as usual in its minimum) that the ﬁelds G 0 and G ± remain in the lagrangian (12. We recall that S-matrix elements are obtained from the Green’s functions by removing external lines.27) which mix W ± and Z 0 vector bosons with the Goldstone † Compared to (12. We complete this section by discussing some aspects of the lagrangian in covariant (renormalizable) gauges. Inserting (12. and contracting tensor indices with appropriate physical polarization vectors. Z µ and Wµ .9) and mix with the gauge bosons (v can always be taken as a real parameter): i + − 0 LHiggs = · · · + gv(Wµ ∂ µ G − − Wµ ∂ µ G + ) − 1 (g 2 + g 2 )1/2 v Z µ ∂ µ G 0 2 2 (12.

32) a δY Bµ = ∂µ δ Y . δY W µ = 0 δ B = 0. The ﬁrst three terms are gauge-dependent corrections to the kinetic terms for the gauge bosons.ﬁx = 1 0 µ ν 0 1 1 + − Z µ∂ ∂ Z ν + A µ ∂ µ ∂ ν A ν + Wµ ∂ µ ∂ ν Wν 2ξ 2ξ ξ 1 2 2 − ξ M Z G 0 G 0 − ξ MW G + G − 2 1 i 0 + − + (g 2 + g 2 )1/2 v Z µ ∂ µ G 0 − gv Wµ ∂ µ G − − Wµ ∂ µ G + 2 2 (12. Performing inﬁnitesimal gauge transformations on the ﬁelds in F a in (12. Finally.12.2 Electroweak currents 397 ﬁelds are cancelled by the following gauge-ﬁxing lagrangian† Lg. δL H = ig H δ L T (12.ﬁx = − with F 0 ≡ ∂ µ Bµ − iξ g H † Y H − H † Y H a F a ≡ ∂ µ Wµ − iξ g H † T a H − H † T a H 1 2ξ 3 FaFa a=0 (12. more explicitly.27). one gets Lg. 3) (12.30): δY H = −ig δ Y Y H. and the remaining two terms are mass terms for the unphysical Goldstone bosons. 2. 1 F 0 ≡ ∂ µ Bµ − ξ g vG 0 2 1 3 µ 3 F ≡ ∂ Wµ + ξ gvG 0 2 i 1 F 1 ≡ ∂ µ Wµ − √ ξ gv(G − − G + ) 2 2 1 2 F 2 ≡ ∂ µ Wµ − √ ξ gv(G − + G + ) 2 2 Integrating these terms by parts. one can introduce different gauge-ﬁxing parameters ξ for different components of the W a ﬁeld and for the B ﬁeld. δL H = −igδ a T a H L a a † † † † δY H = ig δ Y H Y.29) (a = 1.31) The last two terms indeed cancel the terms in (12. the ghost lagrangian can be obtained using the general procedure described in Chapter 3. δ W a = −gεabc W b δ c + ∂ δ a L µ L µ µ L µ L † In general. We follow here the simplest approach and introduce only one such parameter.30) or. .

y) ≡ = ξ gg H T Y H + H Y T H δ (x − y) a δ L (y) (12. one arrives at Lghost generating the Feynman rules collected in Appendix C. y)ηY (y) ¯ (12. Clearly. 0 Z µ . 1 1 2 η± = √2 ηL ∓ iηL .398 12 Standard electroweak theory we compute the corresponding Fadeev–Popov determinants (the variations are taken always at = 0): δF 0 (x) 2 2 † † (4) MYY (x. Aµ and h 0 .3 Fermion masses and mixing The Yukawa terms (12. y)ηL (y) ¯ ¯a ←µ a a a ¯a + ηY (x)MYL (x. all those ﬁelds have well-deﬁned symmetry properties under SUL (2) × UY (1) and we call them to be in an ‘electroweak basis’.33) through the physical ﬁelds Wµ . this can be written as an integral of a local lagrangian density Sghost = Finally. deﬁning dx Lghost (x) (12. (12. y)ηL (y) + ηL (x)MLY (x. 12. We shall work in the so-called ’t Hooft–Feynman gauge with ξ = 1. y) ≡ = ∂ + ξ g H Y Y H + H Y Y H δ (x − y) δ Y (y) a δF (x) ←µ 2 abc b 2 † c a ac MLL (x.35) 3 η Z = cos θW ηL − sin θW ηY . 3 η A = sin θW ηL + cos θW ηY .34) Since the variations of F a contain factors δ (4) (x − y).36) ± and expressing the functional derivatives (12. y)ηY (y) + ηL (x)MLL (x. The ghost action then reads Sghost = dx ab a dy ηY (x)MYY (x. y) ≡ = ξ gg H † Y T a H + H † T a Y H δ (4) (x − y) δ Y (y) 0 δF (x) † a † a (4) a MYL (x. y) ≡ = ∂ + gc ∂ Wµ + ξ g H T T H c δ L (y) † a c (4) + H T T H δ (x − y) δF a (x) a MLY (x.33) (the symbol ∂ means that the derivative acts on the object standing to its left in the full expression for the ghost action).10) in the electroweak lagrangian couple the left-handed Weyl fermion doublets of SUL (2) with the left-handed Weyl fermion singlets and the Higgs boson. With .

2): v v Lmass = − √ YlB A ( e B )R ( e A )L − √ YdB A ( d B )R ( d A )L 2 2 v BA − √ Yu ( u B )R ( u A )L + h.and left-handed ﬁelds. An attractive and straightforward extension is the inclusion of the right-handed neutrinos or. One may hope that the future. l A .1) . ec . As mentioned in Section 12.10) is that neutrinos remain massless.c. deﬁned in (12. They can be diagonalized by two unitary transformations (check it): V † [m]U → [m]diagonal with real diagonal entries. 2. the latter follows from the fact that the transformations V and U can include phase rotations independently for the leftand right-handed ﬁelds.12. u c .39) In general the Yukawa matrices can be the most general complex 3 × 3 matrices. τ ). of a set of new left-handed Weyl ﬁelds νL in the list (12. Fisher. which are mass eigenstates. (12. to diagonalize the mass matrices we perform unitary transformations on the right. e. more fundamental theory will ﬁx those parameters in some electroweak basis. d B )L . independently for each category of ﬁelds q A . The accumulating experimental evidence for neutrino masses and oscillations (for a review see.40) and analogously for the down quarks and charged leptons e A (i. ( ( u A )R d A )R = (UR ) AB ( = (DR ) AB ( u B )R d B )R (12. Thus.38) 2 The mass matrices are: v [m e ] = √ [Ye ].26) generates fermion mass terms: v v v c Lmass = − √ YlB A e A ec − √ YdB A d A d B − √ YuB A u A u c + h.e.c. for example. µ. Yukawa couplings are free parameters of the present theory. 2 v [m u ] = √ [Yu ] 2 (12.37) in terms of the four-component chiral ﬁelds. 2 v [m d ] = √ [Yd ]. The primed ﬁelds describe physical particles: after diagonalization we can combine chiral ﬁelds into Dirac ﬁelds = L + R . independently for up and down quarks and charged leptons: ( ( u A )L d A )L = (UL ) AB ( = (DL ) AB ( u B )L . A non-zero vacuum expectation value of the Higgs ﬁeld (12. (12. Kayser & McFarland (1999)) may call for some extension of the theory.1 one consequence of the lagrangian (12. d A (A = A A 1.3 Fermion masses and mixing 399 three families of fermions. c strictly speaking. this basis is deﬁned up to rotations in three-dimensional c ﬂavour space.37) B B 2 2 2 The physical interpretation of these terms becomes more transparent when we rewrite the lagrangian (12. 3).

43). Indeed.c.41) c (from tree diagrams with νL exchange) which. Assuming that the entries of the matrix M are much larger than the electroweak scale one can integrate out the heavy right-handed neutrino ﬁelds. This see-saw mechanism (Gell-Mann. cB neutrino Yukawa terms YνB A i j Hi l jA νL and the right-handed neutrino Majorana c cB ˆ mass matrix M B A νLA νL + h.400 12 Standard electroweak theory with four-component ( ν A )R given by the relation analagous to (12.43) † and VCKM ≡ UL DL is called the Cabibbo–Kobayashi–Maskawa (CKM) mixing matrix. the neutrino ﬁelds can always be rotated in such a way that there is no lepton mixing in (12. One then obtains the standard electroweak theory supplemented by the effective dimension ﬁve operator ˆ ˆ ˆ Yν M −1 YνT ( i j Hi l j )( km Hk l m ) (12. We can now rewrite the matter couplings (12. It is also important that the matrices UR and DR drop out completely from the ﬁnal lagrangian. after the Higgs ﬁeld aquires a nonzero vacuum expectation value. (12.40) in terms of the left-handed Weyl ﬁelds) − Jµ = A u A σµ d A + ν A σµ e A ¯ ¯ ¯ ¯ u A σµ (VCKM ) AB d B + ¯ ¯ AB A = ≡ AB ν A σµ e A ¯ ¯ dB ¯ u γµ PL (VCKM ) AB A + A ¯ ν γµ PL A eA . The neutral weak current and the electromagnetic current have the same form in the mass eigenstate and in the electroweak basis. in particular. namely.2). Finally. the Yukawa couplings of the physical scalar h 0 and of the unphysical pseudoscalar Goldstone boson G 0 (present in the Rξ gauge) become diagonal in ﬂavour in the mass eigenstate basis. Note. that due to the doublet structure of all left-handed quark multiplets.14) to the gauge bosons in terms of the physical fermion ﬁelds: g g µ 0 µ − + Lmatter ⊃ − √ (Jµ W +µ + Jµ W −µ ) − J Z Z µ − e JEM Aµ cos θW 2 (12.42) where (returning ﬁrst to the Weyl spinor notation – rewrite (12. because the right-handed ﬁelds only have diagonal couplings to the Bµ gauge ﬁeld. In (12.37) neutrinos are massless and. they are determined by the same Yukawa coupling . Ramond & Slansky 1979) can naturally explain the neutrino masses in the experimentally suggested range provided the Majorana mass M is of the order of the Grand Uniﬁcation scale. gives masses to the almost pure Majorana lefthanded neutrinos. the neutral weak current remains diagonal in ﬂavour in the mass eigenstate basis. New renormalizable terms can then be added to the electroweak lagrangian. therefore.

and right-handed quarks have to be the same to preserve real diagonal mass terms. and VCKM ≈ 3 1 − 1 λ2 2 −λ − iAλ5 η Aλ (1 − ρ − iη) 2 λ 1 − 1 λ2 2 −Aλ − iAλ η 4 Aλ3 (ρ − iη) Aλ2 1 (12. Thus.3 Fermion masses and mixing 401 matrices which give the masses. and are diagonalized simultaneously with the latter. Indeed. We end this section with a few more remarks. The Feynman rules for the full lagrangian can be derived by the standard methods of Chapter 2. and from (12.† The coupling of the unphysical charged Goldstone boson G + to physical quarks involves the CKM matrix as for the charged weak currents. unless for some symmetry reasons only one doublet couples to the up-type quarks and one to the down-type. γ and the phase δ ﬁxed by experiment. the matrix VCKM remains invariant when we change all quarks by the same phase. We know from experiment that λ ≈ 0. This completes the discussion of the electroweak lagrangian in the Rξ gauge. n 2 −2n +1− 1 n(n −1) = 1 (n −1)(n −2) is the 2 2 number of phases. In the world with three generations of quarks the C P symmetry is explicitly broken if this phase is different from zero. β.44) with the angles θ. The Cabibbo–Kobayashi– Maskawa quark mixing matrix is a 3 × 3 unitary matrix. For n = 3 there is one physical phase. the n × n unitary mixing matrix has 2n 2 − n 2 = n 2 real parameters. The two most often used parametrizations of VCKM read VCKM = −sθ cγ − cθ sβ sγ exp (iδ) sθ sγ − cθ sβ cγ exp (iδ) c θ cβ sθ cβ cθ cγ − sθ sβ sγ exp (iδ) −cθ sγ − sθ sβ cγ exp (iδ) sβ exp (−iδ) c β sγ c β cγ (12. . This is another reason (apart from economy) to consider one-doublet theory. Notice that the phase transformations on the left. but not all of them are observable.12. † This is no longer true if the Higgs sector consists of several Higgs doublets. Therefore. of which 1 n(n −1) are angles of 2 n ×n orthogonal matrix.23 (the Cabibbo angle) and A ≈ 1. So only 2n − 1 phases can be removed. with several Higgs doublets one generically has phenomenologically dangerous ﬂavour changing neutral-current transitions mediated by Higgs boson exchange. In the general case of n quark doublets.45) The second one (Wolfenstein 1983) is only approximate (in this parametrization VCKM is unitary only up to order O(λ4 )). 2n − 1 phases can be removed by phase redeﬁnitions of the physical quark mass eigenstates. and are summarized in Appendix C. The number of real parameters is therefore n 2 −2n +1.45) we clearly see the order of magnitude of the elements of the VCKM matrix. For n = 2 no physical phase in the CKM matrix is present. Moreover. but is sufﬁcient for the present level of accuracy of the measurements.

which will reach the c. the Higgs boson mass and the strong coupling constant.m. The parameters of the electroweak gauge sector of the Standard Model lagrangian are the gauge couplings g and g and the vacuum expectation value v of the Higgs ﬁeld. g2 + g 2 α= g 2 sin2 θW . For processes induced by the weak forces there are two regions of energy scale which have been studied experimentally particularly thoroughly. the so-called LEP2 is operating. Of course. . sin θW is ﬁxed in terms of G µ . as we 2 ˆ Z.† We are going to concentrate on these two regions and to provide sample calculations in the standard electroweak theory. energy of 200 GeV. Any further measurement is already a test of its predictions. At present the ˆ best measured electroweak quantities are αEM . and a convenient choice is. These are: (a) very small momentum.0359895(61) is the ﬁne structure constant. G µ and M Z . in a search for some departures from the Standard Model at those high energies. where αEM = 1/137. processes (mainly weak decays of leptons and hadrons). with non-renormalizable four-fermion lagrangian introduced on purely phenomenological grounds in 1933 (Fermi theory). G µ = 1. M Z = M Z . i. αEM and M The complete set of the electroweak parameters also includes fermion masses. they can be replaced by some combinations of them. (b) the region of the Z 0 resonance at s 1/2 ≈ 91 GeV studied experimentally with very high precision in e+ e− collisions at LEP1.002 GeV is the physical Z 0 -boson mass (we distinguish it by the hat from the mass introduced as the parameter of the lagrangian.46) Three independent measurements are thus necessary to ﬁx these input parameters of the theory (beyond tree level they are renormalization-scale-dependent). We shall use these observables as the three measurements ﬁxing the parameters of the lagrangian order by order in perturbation theory.e.46). 4π 1 2 M Z = (g 2 + g 2 )v 2 4 (12. sin2 θW = g2 . we need to calculate them to the required accuracy in the electroweak theory parametrized by the set (12.402 12 Standard electroweak theory 12. Moreover.4 Phenomenology of the tree level lagrangian The electroweak theory discussed in this chapter correctly describes all known phenomena induced by the electromagnetic and/or weak forces up to the energy scale O(100 GeV). CKM angles.1 that at the tree level we get α = αEM . It is clear from ˆ Section 12.16639(2)×10−5 GeV−2 is the so-called Fermi constant directly related to the muon decay width (see ˆ (12. q ≈ 0.50)) and M Z = 91. The fermion masses are known with good (or very good) accuracy for leptons and less precisely † At present (1999) the second phase of the LEP accelerator. shall see. This will illustrate the basic logic of such calculations. this distinction becomes important beyond the tree level). for example.1867 ± 0. One should also remember that weak decays of elementary particles played a crucial role in the development of the electroweak theory.

The origin of this approximate locality is in the decoupling of the heavy particles (W ± . since we can discuss the whole class of processes at once.e. ¯ corresponds to the diagram in Fig. are much lighter than the W ± and Z 0 bosons. and with their momenta q MW.132). The V –A current is always built of a pair of fermions which form an SUL (2) doublet. which is approximately local for the considered range of phase space. for the decay µ− → eνµ νe . and used relation (12.47).Z . Charge current processes with four external light fermions. for example. ﬁrst we have to calculate the µ− decay widths. This is a convenient way to organize certain calculations in the complete electroweak theory. the coefﬁcient G F is universal for all four-fermion operators present in the lagrangian (12. The concept of the effective low energy theory. in its full sense introduced in Chapter 7. Effective four-fermion interactions We proceed with the discussion of the tree level predictions.47) accounts for the amplitude that. can be described in the tree level approximation by the effective four-fermion lagrangian √ − LCC (x) = −2 2 G F Jµ (x)J µ+ (x) eff where √ 2 GF = πα 2 s 2 c2 M Z (12.1. At the tree level.4 Phenomenology of the tree level lagrangian 403 for quarks.43). It is important to stress that we introduce the effective four-fermion lagrangian as the four-point part of the integrand in the effective action (2. The lagrangian (12.48) We have abbreviated our notation.12. One should also remember that the predictions for the electroweak processes in the hadronic sector generically have in addition uncertainties coming from our limited knowledge of the quark current matrix elements taken between the hadronic states (i. It can be determined in terms of the muon decay widths. and includes the appropriate CKM matrix element. As advocated above.25) (this relation is actually strictly valid for the parameters of the lagrangian in any order of perturbation theory in the MS scheme). Calculating the muon decay (µ → eνµ νe ) width in the tree approximation ¯ . s ≡ sin θW . The currents are the same as in (12. from the lack of precise theory of hadrons as bound states of quarks and gluons). 12. c ≡ cos θW . Z 0 and Higgs bosons and the top quark).6. and then we shall be able to discuss the predictions of the theory.47) (12. will be discussed in Section 12. except for the top quark. All known elementary fermions.

Fig. 12. Tree level diagram for the charged currents interactions in the Standard Model. The absence of the IR and mass singularities follows from the Lee–Nauenberg–Kinoshita theorem (Chapter 5). Historically. the standard attitude is that (12.49) where τµ and µ are the muon life time and its full decay width.50) where αEM is the ﬁne structure constant. From this formula and the measured muon lifetime τµ = (2.16639(2) × 10−5 GeV−2 is extracted. 12. i. . which will be denoted by G µ .19703 ± 0. Kinoshita & Sirlin 1959). the so-called Fermi constant. 12. taking into account muon decays with one photon in the ﬁnal state. In four-dimensional regularization schemes the counterterms shown in Fig.404 12 Standard electroweak theory Fig.00004) × 10−6 s the value G µ = 1.1. 12.3 cancel out. Computing the full set of diagrams shown in Figs. Once G F is determined by its identiﬁcation with G µ .2 and 12.3 and integrating over the whole phase space. f (x) = 1 − 8x + 8x 3 − x 4 − 12x 2 ln x and G F has been renamed G µ . the G F in (12. respectively.2. Tree level diagram for muon decay in the Fermi theory.48) is expressed in terms of the muon lifetime after correcting the theoretical prediction for O(αEM ) QED radiative effects. we get 1/τµ = µ = G 2 m 5 /192π 3 F µ (12. Actually.50) deﬁnes a new fundamental constant. this calculation has been performed in the Fermi theory. its comparison with † This result is ﬁnite. in a four-dimensional regularization scheme and with the on-shell renormalization (Berman 1958. one gets the following result:† 1/τµ = µ = G2 m5 µ µ 192π 3 × f m2 e m2 µ 1+ αEM 2π 25 − π2 4 (12.e. Traditionally (but not quite consistently for a tree level calculation).

e− νµ → e− νµ ) in the limit q ≈ 0. Numerically.3. we get MW = 80.94 GeV (the superscript (0) means the tree level prediction for the physical W -mass). for the W ± -boson mass one gets (using again (12. The recent (1997) combination of different measurements at LEP and Tevatron colliders gives ˆ MW = 80. for vanishing fermion masses and in the tree approximation.25)) 1/2 ˆ (0) ˆ MW = M Z 1 1 + (1 − 4A)1/2 (12.4 Phenomenology of the tree level lagrangian 405 Fig. The crossed circles denote counterterms.48). can then be used to exchange the third parameter sin2 θW for the accurately measured G µ . The relation (12.12. The neutral-current transitions (like. any other experimentally measured decay constant G ll (deﬁned by the analogue of (12.52) LNC = −ρ f f 2 2 G µ J Z J Z µ eff . can be described by the effective four-fermion lagrangian √ µ (12. For instance. As said earlier.08.51) 2 √ ˆ (0) ˆ2 where A ≡ παEM / 2 G µ M Z . for example. QED corrections to the muon decay in the Fermi theory. at the tree level the parameters α and M Z of the lagrangian have to be identiﬁed ˆ with the ﬁne structure constant αEM and with the physical Z -boson mass M Z .50) for beta decay of lepton l into lepton l ) is a tree level test of the electroweak theory (with the O(αEM ) QED corrections included). With the three input parameters taken from experiment we can predict at the tree level the values of all other electroweak observables. 12. with G F = G µ . The discrepancy signals the need of including radiative corrections (see next section).39 ± 0.

12. the differences are easy to understand) √ ˆ µ 0 L Z f f = −(ρ f 2 G µ )1/2 2 M Z J Z Z µ (12.52). f f gR = −Q f sin2 θ eff f f (12.406 12 Standard electroweak theory Fig. . Z 0 couplings For the on-shell Z 0 boson coupled to leptons and quarks we have (by analogy with (12. For instance.006 ± 0. we obtain the remarkable tree level prediction of the SM† ρff ≡ ρ = 1 (12.52).034 (Vilain et al. the parameter ρ measured in the leptonic sector of the neutral-current transitions is ρEXP = 1.4.56) µ where.4 and using (12. These predictions can be compared with experimental determination of the parameters ρ (strictly speaking ρ f f ) and sin2 θ eff f (low) deﬁned by the lagrangian (12.25) we see that at the tree level 2 sin2 θ eff f (low) = sin θW (12.25) (used in the derivation of (12. as before.53) Calculating the Feynman amplitude corresponding to Fig. 1994).53) with gL = T f3 − Q f sin2 θ eff . Tree level diagram for neutral-current interactions in the Standard Model. interpreted as a low energy phenomenological lagrangian.21) and (12.48)). In this case the Higgs sector has the so-called custodial SUV (2) symmetry (see Problem 12. 12. where the neutral current is µ JZ = f ¯ f γ µ PL T f3 f − f Q f sin2 θ eff ¯ f γ µ f (low) f f ≡ f f f ¯ f γ µ (gL PL + gR PR ) (12. J Z is given by (12.3).55) This follows from the relation (12. which in turn is the consequence of using only Higgs doublets to break SUL (2) × UY (1) symmetry.57) √ 2 † Note the factor of 2 in the denominator in the effective coupling 2 2 G µ ρ = g 2 /2M Z cos2 θW obtained from the tree level diagram. moreover. It follows from the identity of the two currents.54) and.

58) Also. which suggests the existence of a still deeper (more uniﬁed?) theory of elementary interactions.5 Beyond tree level 407 (one also often deﬁnes vector and axial couplings gV = gL + gR . there is a consensus at least with regard to that latter point. higher order contributions.5 Beyond tree level Renormalization and counterterms The electroweak theory is a renormalizable ﬁeld theory. after the experimental results of LEP1.82 MeV.51).21215. Crucial for renormalization is spontaneous (and not explicit) breaking of the gauge symmetry and also the matter content. The commonly . part of the renormalization scheme is the choice of the input observables that are used to ﬁx the parameters of the lagrangian. b). In this book we shall be mainly concerned with practical aspects of the renormalization programme to provide a basis for systematic higher order calculations. The interplay of quarks and leptons in obtaining the anomaly-free theory is a big puzzle of the Standard Model.00022 with the tree level 2 eff 2 1 prediction sin θ f = sin θW = 2 [1 − (1 − 4A)1/2 ] = 0. The experimental values for the parameters ρ f and sin2 θ eff can f be obtained by measuring the widths and various asymmetries in the Z 0 → f¯ f decays. which should be compared with the LEP measurement 83.59) where Nc = 1(3) for leptons (quarks). gA = gL − gR ). A formal proof of this fact can be found in ’t Hooft (1971a. At the tree level one gets 2 sin2 θ eff = sin2 θ eff f f (low) = sin θW (12. then ρ f = ρ = 1. The latter ensures the absence of chiral anomalies (Chapter 13).10 MeV. At present.91 ± 0.23151±0. Strictly speaking. at the tree level we get (Z 0 → e+ e− ) = 84.12. The complexity of the theory together with arbitrariness in the choice of the ‘best’ renormalization scheme often make it difﬁcult to follow and to compare different approaches. The importance of higher order effects can be judged from the comparison eff of the experimental world average sin2 θe = 0. The decay width of Z 0 into a fermion–antifermion pair is √ ˆ3 4m 2f 1/2 Nc ρ f 2 G µ M Z (Z 0 → f¯ f ) = 1− ˆ2 12π MZ × (gL + gR ) 1 − f2 f2 m 2f ˆ2 MZ + 6gL gR f f m 2f ˆ2 MZ (12. A vast literature exists on precision calculations and precision tests of the electroweak theory which include all one-loop effects and also. 12. partially. where A is deﬁned below (12. Thus.

and 2 (µ) and M (µ). s 2 (µ) and M Z (µ) obtained in step (2). G µ and M Z by inserting into the result of step (3) the 2 expressions for the parameters α(µ). We get (1) (0) (1) (0) ˆ 2(1) ˆ 2(0) ˆ2 αEM = αEM + δαEM . Once we know the theory is renormalizable. at the tree the equations can then be solved for α(µ). keep only the terms which are ﬁnite in the limit ε → 0. G µ and M Z . As in the rest of the book. M Z ). Therefore. . Complete ﬁxing of the renormalization scheme then requires only a choice of the renormalization scale µ. M Z = M Z + δ M Z . without keeping track of the necessary counterterms. Of course. G (0) = √ (12. G µ . ˆ We use the set ai = (αEM . s 2 (µ) and M Z (µ) deﬁned in (12. (2) Compute all input observables ai . The Minimal Subtraction scheme is a convenient reference frame for comparison of different approaches. s Z level we have πα (0) ˆ (0) MZ = MZ . G µ = G µ + δG µ . we choose to work in the Minimal Subtraction scheme (actually in the MS scheme). The remaining freedom in the renormalization scheme is the choice of the renormalization conditions that ﬁx the counterterms in the intermediate steps of the calculation. Thus.5) as it stands simply without explicitly listing the counterterms. G µ and 2(1) ˆ2 ˆ M Z are then identiﬁed with experimentally measured values αEM . we are assured that calculation of any physical observable in terms ˆ of the input physical parameters (such as αEM . in the one-loop calculations considered in this book. dimensionally regularized.60) αEM = α. in terms of the parameters of lagrangian (12. in terms of the (1) (1) same set of the renormalized lagrangian parameters. in one-loop calculations we can proceed as follows: (1) Choose the set of the input observables in terms of which the others will be calculated. M Z ) must give a ﬁnite answer. Any other renormalization scheme (for example. (4) Obtain ﬁnite predictions for the computed observables in terms of the chosen physical ˆ input observables αEM . we can as well subtract them in the dimensionally regularized result. keep only the terms which are ﬁnite in the limit ε → 0.5) interpreted as the renormalization scale-dependent renormalized parameters in the MS (we shall use the set α(µ).61) (3) Compute in the one-loop approximation any other observable quantity of interest in terms of the same set of the renormalized parameters of the lagrangian.408 12 Standard electroweak theory ˆ used input parameters are αEM . G µ and the mass M Z of the Z 0 boson.46)). The quantities αEM . the most pragmatic attitude to the renormalization programme (useful in a theory as complex as the electroweak theory) is to use lagrangian (12. µ 2 (1 − s 2 )M 2 2s Z and in this order one gets c2 = 1 − s 2 = 1 4παEM 1+ 1− √ ˆ2 2 2G µ M Z 1/2 2 ≡ c0 (12. Since we know the inﬁnities cancel out. the often used on-shell OS scheme) is obtained by ﬁnite renormalization. G µ .

s 2 (µ)) (12. M Z ) (0) (0) (0) ∂M ∂M ∂M ˆ2 − δαEM − δG µ − δ M Z (12.63) 2 ∂αEM ∂G µ ∂ MZ For higher order calculations it is necessary to go beyond the ‘short cut’ approach described above and to introduce the counterterms explicitly (see the end of Section 4. we can rewrite the last equation as follows: ˆ ˆ ˆ M(1) (αEM .12. G µ . It is helpful at this point to recall the notation for QED in Chapter 5: there eB = .5 Beyond tree level 409 The above prescriptions deﬁne the one-loop perturbative calculation up to some terms of the second order which may appear depending on the particular way of realizing this programme.62) 2 ˆ 2(0) ˆ2 ˆ2 ˆ Using the relations M Z ≡ M Z = M Z −δ M Z . s 2 (µ)) + δM(α(µ). where M Z is the measured physical (0) mass.4). G µ . the superscript B means ‘bare’): 1/2 G aB = Z G G a µ µ 1/2 a aB Wµ = Z W Wµ 1/2 B Bµ = Z B Bµ 1/2 B (12. The gauge invariance of the theory is easily taken care of if we introduce the renormalization constants at the level of the original lagrangian (12. G µ .64) H = ZH H −3/2 B gs = Z G Z s gs −3/2 g B = Z B Zg g −3/2 B g = ZW Zgg The renormalization constants for the non-abelian couplings g and gs are deﬁned as in Chapter 8. M Z (µ). respectively. It is convenient to deﬁne the abelian Z g in the same way. One way of organizing such calculations is as follows (Barbieri 1993): calculate a renormalized matrix element (or any observable quantity) 2 2 M(1) = M(0) (α(µ).3). This is also helpful in getting a more profound ﬂavour of the renormalization programme (for example. M Z ) + δM(αEM . G a denote the gluon ﬁelds.5). Ward identities) and is necessary in studying the renormalization group equations for the parameters of the theory (Problem 12. M Z (µ). α(µ) ≡ αEM = αEM − δαEM and similarly replacing s 2 (µ) by the Fermi constant. We summarize here the renormalization constants of the complete Standard Model. For the gauge ﬁelds and couplings we follow the deﬁnitions introduced in Chapter 8 for QCD but we change the notation in the QCD sector for a more convenient one in the complete Standard Model framework (gs is the QCD coupling constant. the counterterms Z 3 and Z 1Y M are µ renamed as Z G and Z S . M Z ) = M(0) (αEM .

The Yukawa coupling renormalization is determined by the vertex counterterms Z Y s B B B deﬁned by the matrix equations. for the Higgs boson mass parameter m 2 . Z Yu Yu = Yu + δYu . ˜ we get Z e = Z 3 . where Z 1 . and it follows that the relation Z B = Z g can be imposed. † With massive neutrinos it is still convenient to work in the basis with diagonal matrix YiA and non-diagonal effective neutrino mass operator m AB ( i j l iA H j )( i j l i H j ).and right-handed chiral parts of the UY (1) lagrangian. electron and photon selfenergy renormalization constants.† The quark wave-function renormalization constants are hermitean. Yu ¯ R L H B = Z Yu Yu ¯ R L H . eB = Z 3 Z e e. For fermion ﬁelds we use the chiral notation of (12. for example.66) λ = Z H (λ + δλ) −1 2 B 2 2 (m ) = Z (m + δm ) H The additive notation is again useful. the Yukawa matrices Yl A are diagonal. In Appendix C the vertex counterterms are then expressed in terms of δY s. Replacing the unrenormalized ﬁelds and parameters in the original lagrangian (12. Z 2 . in particular since λ is additively renormalized (for example.64). In the present case the same argument applies separately to the left. The U (1) Ward identity ensures −1/2 −3/2 ˜ Z 1 = Z 2 and Z e = Z 3 . Finally. and it is convenient to deﬁne the renormalization constant matrices δY s. for example.65) B 1/2 AC d A R = (Z d ) dC R −1/2 −1/2 A −1/2 A A B A A Yl = Z H (Z e ) (Z Yl Yl )(Z l ) B −1/2 −1/2 AX −1/2 Y C AC XY Yu = Z H (Z u ) (Z Y u Y ) (Z q ) B −1/2 −1/2 AX −1/2 Y C AC XY Yd = Z H (Z d ) (Z Y d Y ) (Z q ) Since there is no ﬂavour mixing in the lepton sector.2): B 1/2 A l A L = (Z l ) lA L B 1/2 A = (Z e ) eA R eA R B 1/2 AC q A L = (Z q ) qC L B 1/2 AC u A R = (Z u ) uC R (12. by fermionic loops proportional to the Yukawa couplings). B ‡ In the Rξ gauge it is convenient but not necessary to introduce the counterterm δv for the Higgs vacuum expectation value.410 −1/2 12 Standard electroweak theory (Z 1 /Z 2 )Z 3 e ≡ Z e e. its self-coupling λ and the vacuum expectation value v we write‡ 1/2 v B = Z H (v − δv) −2 B (12. Z 3 are the vertex. . respectively.5) with the help of the above relations and splitting all Z s: Z i = 1 + δ Z i . If for QED we redeﬁne as in (12.

Finally. in this book we shall never call the latter a sine of the Weinberg angle. in any order of perturbation theory. corrections to the vertices and corrections . we recall that.4. the lagrangian parameter s 2 (µ) ≡ sin2 θW deﬁned by (12. Likewise. however. as we shall see.12. and then consider a sample of the most relevant physical applications. we must carefully distinguish parameters of the lagrangian which depend on the renormalization scale (and may also be gauge-dependent) and the physical quantities. 2 2 c2 (µ) = MW (µ)/M Z (µ).5) are also sufﬁcient to renormalize this theory after spontaneous breaking of the gauge symmetry. it will be convenient also to recast one-loop corrections to the selected processes in the form of corrections to the effective lagrangians introduced in Section 12. Corrections to gauge boson propagators In general. We recall that in the MS scheme. ﬁnite order results can numerically depend on the choice of the input parameters. To avoid any confusion. (We do not explicitly write down the renormalization constants for the ghost ﬁelds but they are included in the Feynman rules given in Appendix C. a given observable is calculated in terms of two different sets of input observables. of higher order. In particular.18) should be distinguished from several quantities directly related to experimental measurements and often also called in the literature the Weinberg angles. and process-dependent corrections consisting of corrections to the external fermion lines. In the following we discuss certain building blocks of the one-loop quantum corrections in the electroweak theory. It is worth stressing once again that.4 can be conveniently divided into two classes: universal corrections to the gauge boson propagators (called sometimes the ‘oblique’ corrections) which in most cases give the dominant contribution. the result does not depend on the remaining freedom in the choice of the renormalization scheme.5 Beyond tree level 411 we can generate the necessary counterterms. Although this dependence is.24). to estimate the uncertainty one would have to perform full next order calculation.) We are already familiar with the fact that the counterterms which have been introduced at the level of the original lagrangian (12. and in the four-component Dirac notation for fermions (appropriate after spontaneous gauge symmetry breaking) are presented in Appendix C.4 and the ˆ2 ˆ2 ratio of the physical gauge boson masses 1 − MW / M Z . For instance. The Feynman rules derived from the lagrangian including the counterterms. in principle. ˆ ˆ MW and M Z must be distinguished from the µ-dependent lagrangian parameters MW and M Z (the same remark applies to the other masses too) which are given by (12. going beyond the tree level. Important examples are various sin2 θ eff introduced in Section 12. if the same physical quantity is calculated in terms of the same input physical observables to the same order in perturbation theory. If. radiative corrections to the processes described in Section 12.

for example. 12. in which only 1/ε terms are subtracted. Feynman (ξ = 1) or Landau (ξ = 0) gauge. The tadpole diagrams must be included if one wants to verify gauge invariance of the mass counterterms.‡ † Note the sign convention which is different from the one used for fermion self-energies. Moreover. . in the calculation in the MS scheme. They cancel out. Note also that.412 12 Standard electroweak theory Fig. the longitudinal terms L1 V2 of the V gauge boson self-energies do not contribute to the physical matrix elements. however. much smaller than the ‘oblique’ corrections. Contributions to gauge boson self-energy. the dependence on the renormalization scale µ of the loop corrections to gauge boson propagators alone does not disappear.5. One should. like. for on-shell amplitudes. Process-dependent corrections are usually (but not always) in the commonly used gauges. The net result is equivalent to using the propagators in the unitary gauge. for example. with the vertex corrections. In the general case these terms cancel (order by order in perturbation theory) Goldstone boson self-energy amplitudes as a consequence of the Slavnov–Taylor identities. in some schemes (like the MS scheme) inclusion of at least part of the process-dependent corrections may be necessary to avoid spurious singularities in the amplitudes in some kinematical limits (see below).67) where Pµν and L µν are transverse and longitudinal projection operators. however. in general. that the splitting into universal and process-dependent parts is gauge-dependent and to get physical. This is immediately seen for processes with external vector bosons (εµ k µ = 0) and with external fermions in the limit m 2f → 0 (conserved currents). the corrections to the gauge boson propagators are not separately renormalization scheme-independent (in a given order) since they share the counterterms. gauge-independent results one should include both types of correction. will be denoted by† i µν V1 V2 ( p) = iP µν T 2 V1 V2 ( p ) + iL µν L 2 V1 V2 ( p ) (12. We begin with the process-independent corrections to the gauge boson propagators. The full vector-boson self-energy amplitude. T s is cancelled by the vertex corrections. after summing up one2 particle reducible diagrams. and are therefore frequently neglected. which consists in general of two parts corresponding to diagrams shown in Fig. in physical amplitudes and will be omitted in the following. Moreover. keep in mind. 12. the self-energy amplitudes at k 2 = MV are corrections to the physical masses. generated by box diagrams. This is convenient since the free fermion and boson propagators have opposite signs and in both cases. Also. ‡ The gauge dependence of the gauge-dependent parts of the tree level propagators cancel with the gauge-dependent Goldstone boson propagators. Thus.5.

M2 ) + (4 p 2 + M1 + M2 )b0 ( p 2 . with the dimensionless couplings g → gµε/2 ) (ig1 )(ig2 )µε dn k −ig σ σ −ig λλ 2 2 (2π)n (k + p)2 − M1 k 2 − M2 × g µσ (k + 2 p)λ + g µλ (k − p)σ − g σ λ (2k + p)µ × g νσ (k + 2 p)λ + g νλ (k − p)σ − g σ λ (2k + p)ν where the couplings are.12. A V and B given by (D. M1 . M1 . g1 g2 T 2 2 2 10A( p 2 . With some algebra (and remembering that g µν gνµ = n) we ﬁnd −ig1 g2 µε dn k i g µν (2k 2 + 2kp + 5 p 2 ) n [(k + p)2 − M 2 ][k 2 − M 2 ] (2π) 1 2 + (4n − 6)k µ k ν + (n − 6) p µ p ν + (2n − 3)(k µ p ν + p µ k ν ) T 2 V1 V2 ( p ) and L1 V2 ( p 2 ) can now be expressed in terms of the functions a. We recall in particular that.1)–(D. in principle. 12. For instance. The diagram to be evaluated is shown in Fig.5 Beyond tree level 413 Fig.6. we explicitly compute in the ’t Hooft–Feynman gauge the contribution of the gauge boson loop to the gauge boson self-energies. renormalization scale-dependent but to this order of calculation they can be replaced by their tree level values. 12. Higgs boson and mixed Higgs–vector boson loops) are given in Appendix D. As an illustration. using our experience from the chapters on the scalar ﬁeld theory and on QED.6 (in n = 4 − ε dimensions.11). b0 . Vector boson contribution to the vector boson self-energy. 12. M2 ) V1 V2 ( p ) = − (4π)2 2 2 + a(M1 ) + a(M2 ) + 2M1 + 2M2 − 2 p 2 3 With some labour the other contributions to the vector boson self-energies can be obtained in a fairly straightforward manner. due to UEM (1) .6. The formulae for other contributions (from fermion. Using the trilinear vector boson couplings given in Appendix C we get then for Fig.

71) It is important that T Z (0) does not contain a fermionic contribution. but to ˆ2 this order of calculation they can be replaced by the physical parameters αEM . the two MW -dependent terms are the W ± and the Goldstone boson contributions. respectively. L γ ( p 2 ) ≡ 0 and T γ (0) = 0. γ Z ( p ). M Z and the appropriate function of G µ . δ Z H and δv. Z Z ( p ). 3 for quarks) and charge of the fermion. It will prove useful to decompose it as follows T 2 γ Z(p ) = T γ Z (0) + p2 ˜ γ Z ( p2 ) (12.2) i µν γ γ ( p) = i g µν − pµ pν p2 p2 ˜ γ γ ( p2 ) (12.69) Here Nc f and Q f are the colour factor (1 for leptons.1(a)–(c) at p 2 = 0. Of 2 2 course. δ Z g .2(a)–(c) and D. This is a non-trivial and practical test of the computation.70) In the case of the photon self-energy. At one loop it is given by the same set of Feynman diagrams (Figs D. which is useful for us one gets the following result: ˜ γ γ (0) = 23 1 5 M2 α 1 M2 − η + − ln W + ln W + 3π 4 2 2 µ2 4 µ2 Nc f Q 2f ln f m 2f µ2 (12. one easily ﬁnds T γ Z (0) = −2 M2 α c 2 M Z −η + ln W 4π s µ2 (12.2(d) separately vanish at p 2 = 0. which depend on ﬁve independent renormalization constants δ Z W . the photon self-energy can be written as (see Section 5. Using the results of Section 5. in the limit p 2 = 0. δ Z B . only the transverse part of the photon propagator is renormalized and. moreover.414 12 Standard electroweak theory gauge invariance.68) i.2(d) (with W + G − and W − G + circulating in the loop) does not cancel the contribution of D. Another self-energy which requires a comment is the γ –Z 0 mixing amplitude. the contributions of the diagrams shown in Figs D. for instance.2 we can include γ γ the fermion loop corrections to the photon propagator and. γ T(L) T(L) T(L) 2 2 2 T(L) 2 Eight self-energies: γ γ ( p ) are made W W ( p ). MW = c2 M Z and all the parameters are renormalized at the scale µ.1(a)–(c) plus D.1 and D.e. Comparing with the T γ case and isolating the combination of the two point functions which γ does vanish at p 2 = 0. ﬁnite by including counterterms shown in Appendix C. For the γ –Z 0 mixing amplitude diagram D. µ is the MS renormalization scale and the divergence η ≡ 2/(4 − n) + ln(4π) − γ E may be removed by adding appropriate counterterms. .2) as the corrections to the photon self-energy with appropriate changes in one of the vertices (see Appendix C).

inserted into four-fermion Green’s functions (Fig. it cannot be incorporated as it stands into the neutral-current effective lagrangian.72) (12. 2 In the limit p 2 ≈ 0 and neglecting terms O(m 2f /MW ).73) where. Fig.8(b). However. Corrections to elastic neutrino scattering which are singular at p2 = 0. 12. LCC (Born) ≡ − eff 2π α 2 s 2 c2 M Z − Jµ J µ+ . Being singular at p 2 = 0. 12.7.7).8. ‘Oblique’ correction to muon decay and elastic neutrino scattering.5 Beyond tree level 415 Fig. 12. numerically small and can be neglected. It is important to remark that an additional contribution which should in principle be included is the one shown in Fig. The sum of these two graphs is ﬁnite at p 2 = 0. as it is clear from (12. 12. for example. the one-loop corrections to the gauge boson propagators.8(a) cancels against the similar singularity contained in the vertex correction shown in Fig. can be summarized in the form of corrections to the effective fourfermion lagrangians: LCC (oblique) = 1 − eff LNC (oblique) = 1 − eff T W W (0) 2 MW T Z Z (0) 2 MZ × LCC (Born) eff × LNC (Born) eff (12. the singularity at p 2 = 0 of the diagram shown in Fig. .47) and (12. 12. 12.12.48).8(a).

m t ) t t 3 3 9 T(b. It comes from the third generation fermion loops. ρ f . The latter contribution is ﬁnite by itself since it vanishes when the splitting vanishes.74) is ﬁnite by itself. therefore. With the help of the formulae for the fermionic contribution to the gauge boson self-energies collected in Appendix D one ﬁnds Nc α 1 1 m2m2 m2 t b t (m 2 + m 2 ) − = ln 2 b 4π s 2 12 t 6 m2 − m2 mb t b (12.416 12 Standard electroweak theory The ﬁrst interesting physical result which we can already get at this point is that. M Z .t) Z Z (0) = Nc α 4π 2s 2 c2 32 8 1 − s2 + s4 3 9 2 2 − a(m t ) − m 2 3 3 t 1 8 32 − m 2 b0 (0. m t ) + − s 2 + s 4 m 2 b0 (0. as mentioned earlier. The largest violation of the custodial SUV (2) originates from the large top quark–bottom quark mass splitting.b) W W (0) 1 1 1 1 1 − a(m t ) − a(m b ) − (m 2 + m 2 )b0 (0. as any other physical observable. G µ and the Higgs boson mass.e. the dominant contribution to (12. m b . This symmetry is explicitly broken by the g coupling. Still. m b . m t . a systematic application of the renormalization programme gives a ﬁnite (and. is calculable (ﬁnite) in terms ˆ of αEM . as a generic situation in the case of some truncations in this programme. we have already noted that ρ f = 0 in the limit of exact custodial SUV (2) symmetry. Indeed. m b ) − m 2 − m 2 t b t 3 3 6 3 3 b T(t. the dominant contribution to ρ is readily calculable. the Higgs sector and the fermion mass splittings within each generation.74) Of course.b) Z Z (0) = Nc α 4π 2s 2 c2 8 4 1 − s2 + s4 3 9 2 2 − a(m b ) − m 2 3 3 b 1 4 8 − m 2 b0 (0. m b ) b 3 b 3 9 . with one-loop accuracy and neglecting corrections other than the ones to the W ± and Z 0 gauge boson propagators. m t . i.75) T(t. However. m t . Thus. m b ) + − s 2 + s 4 m 2 b0 (0. we have T W W (0) 2 MW T Z Z (0) 2 MZ ρf ≡ ρf − 1 ≈ − (12. the above combination of the self-energy amplitudes is not separately renormalization-scale-independent. renormalization scaleindependent) result for ρ f in each order of perturbation theory.

where a systematic perturbation theory result is insufﬁcient.6 ± 5. however. the e+ e− scattering close to M Z (in the resonance region). G µ and M Z . It is easy to check that T(t.76) This result is indeed renormalization-scale-independent (as for a ﬁnite contribum b (experimentally m t = 175. equivalently.b) W W (0) ˆ2 MW − T(t. T 2 W W (k ) 2 = k 2 − MW − T 2 W W (k ) (12.0096. At each order the physical mass of the W ± boson is given 2 in terms of the running mass MW (µ) by the requirement that the real part of the two-point non-local vertex of the effective action (also called 1PI two-point Green’s function). for example. we need the one-loop relation between MW (µ) and the above three measured parameters. we see that it indeed has the Breit–Wigner form with the decay width of W ± given by 2 Im T W (k 2 = MW ) (we ignore here the subtleties related to the deﬁnition of the W mass of unstable particles and to the gauge-independent deﬁnition of their width).77) Expressing the Weinberg angle in terms of G µ (in a one-loop calculation the tree level formula can be used).152) (or.12.5 GeV whereas tion). For that. introduced in Section 2. So far we have discussed the self-energies of the vector bosons in a systematic perturbation theory.b) Z Z (0) ˆ2 MZ = α Nc ˆ2 4π 4s 2 c2 M Z m2 + m2 − t b 2m 2 m 2 m2 t b t ln 2 2 2 mt − mb mb (12. and should be replaced by the Breit–Wigner form of the full propagator. .6. There are.5 Beyond tree level 417 where Nc = 3 is the colour factor. which will be discussed later in this section.† This condition applies to the transverse part of W W . can be obtained by summing all 1PI diagrams). † We have not yet given all the calculations necessary for solving this equation with one-loop accuracy in terms 2 ˆ2 of αEM .t) Z Z (0) ˆ2 MZ − T(b. The imaginary part of T W (k 2 ) can be obtained by applying the Landau–Cutkosky W rule to the Feynman integrals or by taking the imaginary part of the ﬁnal result (see Appendix D). and since m t m b ≈ 5 GeV) it can be approximated as ρ f ≈ Nc αEM m 2 t ˆ2 16πs 2 c2 M Z (12. important physical situations such as the phenomena at energies in the vicinity of the physical masses of unstable particles. we obtain in our approximation the ﬁrst one-loop prediction of the electroweak theory: ρ f = 0.78) ˆ2 vanishes at p 2 = MW . Since the transverse part of the full W ± propagator is given by (2.

9(a): −i = (i)2 µε −igµν dn k i (2π )n [(k + p)2 − m 2 ] [k 2 − M 2 ] µ k / ×γ (cV − cA γ 5 ) (/ + p + m) γ ν (cV − cA γ 5 ) (12.81) we have for Fig.82) Performing the Dirac algebra and using the formulae (D.80) Fermion self-energies We compute also the gauge boson and scalar contributions to the fermion selfenergy. Of course.9 as −i ≡ −i ( / L p PL + / R p PR + m) ≡ −i( / Vp + / A γ5 p + m) (12.8) we obtain the .1) and (D. We begin with the 2 × 2 matrix for the inverse propagator T 2 Z γ (k ) = k2 − − T 2 γ γ (k ) T 2 γ Z (k ) − T 2 γ Z (k ) T 2 Z Z (k ) 2 k2 − MZ − (12. The twopoint vertex function is now a 2 × 2 matrix and the longitudinal part of the full propagator is obtained by inverting this matrix. 12. 12. The same procedure is necessary for the Z 0 boson propagator because of the 0 Z –γ mixing.418 12 Standard electroweak theory For the longitudinal part of the W ± propagator the situation is more complicated as it receives a contribution from the W ± –Goldstone boson mixing. Denoting the amplitude corresponding to the general self-energy diagrams shown in Fig. the physical mass and the width of W obtained from this procedure are the same (due to Ward identities) as those deﬁned for the transverse part.79) Using 2 × 2 matrix inversion we ﬁnd for the transverse parts of the Z 0 –γ propagators 1 1 T 2 G Z Z (k ) = ∼ 2 2 [ T Z (k 2 )]2 k − M Z − T Z (k 2 ) γ Z T 2 − M2 − 2) − k Z Z Z (k 2− 2) k (k γγ 1 1 T 2 G γ γ (k ) = ∼ 2 T 2 2 k − T γ (k 2 ) [ γ Z (k )] γ k 2 − T γ (k 2 ) − 2 γ 2 T 2) k − M Z − Z Z (k 2 2 (k ) (k ) γZ γZ T Gγ Z = 2 ∼ 2 2 [k − (k 2 )][k 2 − M 2 − T (k 2 )] − [ T (k 2 )]2 k (k − M 2 ) γγ Z ZZ γZ Z (12.

These corrections enter into the calculation of any other (predicted) observable if accuracy beyond the tree level is needed. M) (12. M) Similarly. General vector boson and scalar contributions to a fermion self-energy. m. It is also important to stress that the renormalization-scale-dependent coupling α(µ) in the complete electroweak theory is different from the analogous coupling in QED.12. m. applying the general formula (12. M Z and G F . M) 2 p2 × p |cV + cA |2 PL + |cV − cA |2 PR / − m(|cV |2 − |cA |2 ) n b0 ( p 2 . M) 2 p2 × p cL cL PL + cR cR PR / + m cL cR PL + cR cL PR b0 ( p 2 . result (4π)2 = n−2 a(M) − a(m) + ( p 2 + m 2 − M 2 )b0 ( p 2 . m.5 Beyond tree level 419 Fig.83) to the W ± and Z 0 corrections to the electron self-energy we get in the limit m e MW. which is the low energy effective theory obtained after integrating out the weak . 12. we now have to calculate loop corrections to the input observables to relate the renormalization scale-dependent lagrangian paramˆ eters to the measured αEM .9.9(b)) we ﬁnd (4π)2 = 1 a(M) − a(m) + ( p 2 + m 2 − M 2 )b0 ( p 2 . as usual. m.85) 2 MW α s2 1 + ln 2 e R (0) = (4π) c2 2 µ All the parameters are. for a scalar contribution (Fig.84) (12. the running parameters.83) For example. Running α(µ) in the electroweak theory Following our general strategy.Z : 2 2 MW 1 1 MZ α (1 − 2s 2 )2 1 + ln 2 + + ln 2 eL (0) = 2 2 c2 (4π) 2s 2 µ 4s 2 µ (12. 12.

It is convenient to summarize the result of this calculation.86) In any order of perturbation theory the lagrangian parameter α(µ) can be expressed ˆ2 in terms of αEM and M Z by calculating the Thomson scattering in the limit of k → 0 (k is the photon four-momentum) in the complete electroweak theory renormalized at some scale µ.6. Here we want to ﬁnd the relation between αSM (µ) and the ﬁne structure constant αEM = 1/137. and the subscript ‘SM’ will be omitted. in the limit k = 0. When necessary. sector.10(a) but with the photon replacing Z 0 . 12. By explicit calculation one ﬁnds that the result obtained in the complete electroweak theory. in a form such that it will then be easy to compare it with (12. The ﬁne structure constant can be deﬁned by the classical cross section for Thomson scattering (see. 12. 12.† The diagram similar to Fig. is reproduced by the following tree level lagrangian: L R Llow = − 1 (1 + δz γ )Fρκ F ρκ + (1 + δz 2 ) ¯ e i/ PL e + (1 + δz 2 ) ¯ e i/ PR ∂ ∂ eff 4 L / − e(µ)(1 + δz 1 ) ¯ e A PL e − (δm + mδz L ) ¯ e PL e R / − e(µ)(1 + δz 1 ) ¯ e A PR e e − (δm + mδz R ) ¯ e PR e (12. At the one-loop level one has to compute the diagrams shown in Fig. we shall keep the distinct notation αSM (µ) and αEM (µ).86). This point will be discussed in detail in Section 12. at k = 0 and with the electrons on-shell is exactly cancelled by the photon contribution to the external line renormalization (as demonstrated in Chapter 5). performed in the MS scheme.10.420 12 Standard electroweak theory Fig. Bjorken & Drell 1964) σ = 2 8π αEM 3 m2 e (12. for example.87) † Box diagrams with a gauge boson (γ or Z 0 ) attached to the external electron lines vanish in the limit k = 0. . Corrections to the photon–electron interaction. The external photon is real with the four-momentum k ≈ 0.10 (the black blobs denote one-loop diagrams) and one-loop W ± and Z 0 contributions to the electron self-energies.

. One-loop W ± and Z 0 contributions to the electron self-energies have already been computed and are given by (12.91) can be identiﬁed† with αEM . In this regime strong interaction corrections become large and non-perturbative.86) is immediate if we rescale the ﬁelds in the lagrangian (12.5 Beyond tree level 421 All the ﬁelds and parameters are. those of the complete theory. δz 1 − δz 2 = δz 1 − δz 2 = 0 because of the weak contributions.71).91) are worth noting: it is an equivalence relation (as we checked by explicit calculation) L L R R due to the unbroken UEM (1) symmetry and. They give 2 2 MW (1 − 2s 2 )2 1 MZ e2 (µ) 1 1 δz L = − + ln 2 + + ln 2 (4π)2 2s 2 2 µ 4s 2 c2 2 µ (12. of course. In the MS scheme and for k = 0 we get δz γ = − ˜ γ γ (0) L e(µ)δz 1 = − e(µ)(1 − 2s 2 ) 2sc 3 2s 2 1 + ln 6 s c + T γ Z (0) 2 MZ 2 MW µ2 T γ Z (0) 2 MZ − e3 (µ) (1 − 2s 2 )2 (4π )2 4s 2 c2 2 MZ 1 + ln 2 2 µ (12.85). moreover.69) and γ Z by (12.12.90) 2 MZ e2 (µ) s 2 1 δz R = − + ln 2 (4π)2 c2 2 µ Comparison with (12. The effective charge L L R R e(µ) 1 + δz 1 − δz 2 − 1 δz γ ≡ e(µ) 1 + δz 1 − δz 2 − 1 δz γ 2 2 (12.92) We remind the reader once again that α(µ) ≡ e2 (µ)/4π ≡ αSM (µ).92) is that ˜ γ γ (0) contains light quark contribution. To deal † Two facts about (12. and the result obtained for it in perturbation theory is not reliable.88) e3 (µ) s 2 (4π )2 c2 R e(µ)δz 1 = e(µ) − 2 MZ 1 + ln 2 2 µ (12. One complication in using (12. and we get the one-loop relation αEM = α(µ) 1 + ˜ γ γ (0) + 2 = α(µ) 1 + s c T γ Z (0) 2 MZ 2 MW 4 + 2 µ 3 α 2 − 7 ln 4π 3 Nc f Q 2f ln f m 2f µ2 (12.87) to get the canonical form of the kinetic terms.89) where ˜ γ γ (0) is given by (12.

50) as a fundamental constant. The dispersion relation reads ˆ had (s) = γγ 1 s π ∞ ds 4m 2 π Im ˆ (s ) s (s − s) (12. As we shall see.93) is the photon self-energy scalar function in the on-shell scheme. This can be achieved if we succeed in splitting the full result into QED corrections to the tree level effective lagrangian and the genuine one-loop electroweak corrections to the coupling G F . in practice. Muon decay in the one-loop approximation The most straightforward approach to using the muon decay width for ﬁxing the parameters of the electroweak theory at the one-loop level is simply to express them in terms of the width itself. we have to subtract from the full one-loop Green’s function the Green’s function obtained by calculating the diagrams in Figs. and the ﬁnite γγ quantity 2 2 ˆ had (M Z ) = ˜ had (M Z ) − ˜ had (0) γγ γγ γγ (12.2 and 12.422 12 Standard electroweak theory with this problem one writes ˜ γ γ (0) = ˜ had (0) + ˜ remainder (0) γγ γγ 2 2 = − ˆ had (M Z ) + ˜ had (M Z ) + ˜ remainder (0) γγ γγ γγ 2 where ˜ had (M Z ) can already be computed in perturbation theory. to write down the result of the one-loop calculation in the electroweak theory in the same form and to express the lagrangian parameters in terms of the constant G µ . Thus. It can be computed via the dispersion relation with one subtraction at p 2 = 0 (since ˆ (0) = 0). much more elegant to consider the constant G µ deﬁned by (12.3. the effective four-fermion lagrangian has to be written as a sum of ﬂavour-dependent .0280 ± 0. Evaluation of the integral gives the value ˆ had (M Z ) = γγ 3 0. the remaining piece can indeed be summarized in the form of the effective fourfermion lagrangian (in fact this is true in any order of perturbation theory). This is true for all charged current processes with four external fermions but. however. 12. in the limit q ∼ 0 and neglecting fermion masses.94) and Im ˆ ( p 2 ) can be expressed in terms of the measured ratio R(s) = σ (e+ e− → hadrons)/σ (e+ e− → µ+ µ− ) (where s is the centre of mass energy of the process): 2 Im ˆ (s) = 1 αEM R(s). It is.0007 (Eidelman & Jegerlehner 1995). now.

with four-dimensional regularization prescriptions. As mentioned earlier.D √ 2 2 G f ¯ u A γ µ PL (VCKM ) AB dB † ¯ dC γµ PL (VCKM )C D uD (12. Another consequence is more delicate.C. This is. One obvious consequence is that our ﬁnal result (suppose it has the form (12. O1 = √ + const. for example.5 Beyond tree level 423 pieces: LCC = − eff A. for instance.50) will be replaced by α(µ). we have to complete the calculation of one-loop corrections to the relevant piece of the effective lagrangian (12. The subscript f of the couplings G denotes the dependence on all the fermion ﬂavours of a given four-fermion operator. This difference is a higher order effect and can be neglected at one-loop accuracy.3 cancel out. [4O2 + (O1 − 4O2 )] ≡ √ π 2 ε π 2 ε νe where O2 ≡ ¯ e γλ PL energy correction give ¯ νµ γ λ PL α GF √ π 2 µ .95) or. In the following we shall use G µ (best measured experimentally) as the input observable and the others will be the predictions of the theory at each order of perturbation. more precisely. the lepton universality is no longer strictly true and each decay: µ → eνe νµ .12. be expressed as the values of the parameters κ f deﬁned by the equations: G f = κ f G µ . O2 ε . in general. the vertex counterterm and the wave-function counterterms in Fig.50)) will depend on the renormalization scale µ: αEM in (12. The subtle point of our subtraction procedure is that the result (12. 12. The diagrams with the fermion self- −4 1 + const.50) has been obtained with a four-dimensional regularization and in the on-shell renormalization scheme. they are no longer universal and become fermion quantum number dependent. which describes muon decay. These predictions can. to the coefﬁcient of the four-fermion operator ¯ e γµ PL νe ¯ νµ γ µ PL µ . ¯ e γµ γν γρ PL νe ¯ νµ γρ γν γµ PL µ ε α GF 1 α GF 1 + const.95) with similar terms (with obvious simpliﬁcations) for the leptonic and semi-leptonic processes. τ → µ¯ µ ντ and τ → eνe ντ has its own Fermi ¯ ν ¯ constant.47) has the structure α GF √ π 2 1 + const. whereas now we work with the dimensional regularization and in the MS scheme. no longer true in dimensional regularization. Thus. Due to the process-dependent corrections. For muon decay.B. By explicit calculation one can check that the vertex correction to the lagrangian (12.

therefore.97) Our calculation requires. in addition. the original MS scheme must be supplemented by this renormalization condition. In the four-component notation it follows from the identity γµ γρ γν = γµ gρν + γν gµρ − γρ gµν − iεµρνσ γ σ γ5 εαβγ µ ε αβγ ν = −6gµν .96) is not valid in 4 − ε dimensions and instead for the sum of the vertex and the self-energy corrections we get an additional piece 1 + const. It is possible to chose the subtraction scheme so that one subtracts all the 1/ε poles and.11.41). However.98).96) was used in n dimensions. 12. after the Fierz rearrangement ¯ e γ λ PL νe ¯ νµ γλ PL µ = ¯ e γ λ PL µ ¯ νµ γλ PL νe (12. The ﬁnite result would then be a ﬁnite correction to (12. and the ﬁnite part in (12. † In the Weyl spinor notation it has the form e σ λ σ κ σ ρ νe ¯¯ ¯ ν µ σ ρ σ κ σ λ µ = 4 e σ κ νe ¯ ¯ ¯ ¯¯ νµ σ κ µ ¯ ¯ and follows directly from the completeness relation for the σ matrices summarized in Appendix A (A. the identity (12.e. i.424 12 Standard electroweak theory In four dimensions there is the identity† ¯ e γ λ γ κ γ ρ PL νe ¯ νµ γ ρ γ κ γ λ PL µ = 4 ¯ e γ κ PL νe ¯ νµ γ κ PL µ (12.98) which again follows easily from (A. With the above background we can proceed with the subtraction procedure. 12. Thus. There is then no need to calculate matrix elements of the (so-called evanescent) operator (O1 − 4O2 ).3 is in the S M replaced by the box diagram shown in Fig. which imply (γµ γρ γν PL )i j (γ ν γ ρ γ µ PL )kl = 4(γµ PL )i j (γ µ PL )kl . (O1 − 4O2 ) ε (12. these ﬁnite pieces (Dugan & Grinstein 1991).47).41). The ﬁnite result is the same as if (12.3 are equal to their obvious analogue in the full amplitude and the latter should be simply omitted. The ﬁnite parts of the last four graphs shown in Fig. Another way of reaching the same result is to perform the one-loop QED calculations starting from the outset with the tree level effective lagrangian written in the ‘charge retention’ form (12. The vertex graph in Fig. 12. an additional counterterm in the form of the operator (O1 − 4O2 ) and an additional renormalization condition.96) and the two divergent contributions would cancel each other.97) depends on the choice of it.

MW .99) where LCC (Born) now denotes only that part of the tree level effective lagrangian eff (12. the result for that diagram should be subtracted from the one for the box diagram.100) where µ is the renormalization scale. Computing in the same limit (again introducing the photon mass) the effective lagrangian generated by the ﬁrst diagram shown in Fig.12. 12. W –γ box in the Standard Model.11. this result is obtained either by starting with the charge retention form of the four-fermion effective . The corresponding amplitude is ﬁnite and is reproduced by the following effective lagrangian: LCC (γ W ) = i eff e4 2s 2 k k d4 k ¯ νµ γ λ PL (/ + m µ )γ ρ µ ¯ e γρ (/ + m e )γλ PL 4 2 − m 2 ][k 2 − m 2 ][k 2 − M 2 ]k 2 (2π) [k µ e W e In the limit of zero fermion masses and introducing the photon mass λγ to regularize the integral in the infrared one gets LCC (γ W ) = eff d4 k ik µ k ν 2 (2π )4 k 4 [k 2 − MW ][k 2 − λ2 ] γ ¯ νµ γ λ γµ γ ρ PL µ ¯ e γρ γν γλ PL × e4 2s 2 e Replacing k µ k ν → (1/n)k 2 g µν → 1 k 2 g µν under the integral and using the identity 4 (12.47) which is relevant for the muon decay (we have retained only terms singular in the limit λγ → 0).96) we get (see (D.18)) LCC (γ W ) = eff e4 F(0. 12. As explained earlier. λγ ) ¯ νµ γ λ PL 2s 2 2(4π) 2 MW α = ln 2 × LCC (Born) eff 4π λγ µ ¯ e γλ PL e (12. 12. We calculate the contribution to the effective lagrangian from the box diagram shown in Fig.3 one gets − α 4π λ2 1 γ + ln 2 2 µ × LCC eff (12.11 in the limit of zero external momenta. Hence.5 Beyond tree level 425 Fig.15) and (D.

these corrections give LCC (vertex) = eff × LCC (Born) eff (12. lagrangian.101) Note that the dependence on the IR cut-off has disappeared. after using the identity ¯ e γ λ γ κ γ ρ PL ¯ νµ γ λ γ κ γ ρ PL = 16 ¯ e γ κ PL ¯ νµ γ κ PL νe µ νe µ (12. The other two diagrams give amplitudes which are equal to each other and. The computation of the ﬁrst two makes use of identity (12. these four box diagrams generate the effective lagrangian LCC (W Z boxes) = − eff α 4π 1− M2 5 5 + 4 ln W × LCC (Born) eff 2 s 2 2s MZ (12. 12. In the limit of vanishing external momenta and neglecting the fermion masses.12. or by imposing the particular renormalization condition for the new counterterm which is necessary in n dimensions if we work with the lagrangian (12. 12. Remaining box diagrams for the muon decay in the Standard Model. Subtracting (12.12.47). 12.426 12 Standard electroweak theory Fig. Together. when inserted in the full diagram for the muon decay.96).102) 2 2 each give −(e2 /16π 2 )(2/s 4 )(1 − 2s 2 ) ln(MW /M Z ) as the coefﬁcient of the effective Born lagrangian. The remaining box diagrams which have to be evaluated are shown in Fig.13.103) ≡ BW Z × LCC (Born) eff Vertex corrections to be computed are shown in Fig.104) .100)) we ﬁnally get eff LCC (γ W ) = eff M2 α 1 + ln W × LCC (Born) eff 4π 2 µ2 ≡ Bγ W × LCC (Born) eff (12. The result is again the Born effective 2 2 lagrangian multiplied by the coefﬁcients −(e2 /16π 2 )(1/4s 4 ) ln(MW /M Z ) and 2 2 2 2 2 2 4 −(e /16π )[(1 − 2s ) /4s ] ln(MW /M Z ).99) (and identifying to this order LCC with eff LCC (Born) in (12. respectively.100) from (12.

5 Beyond tree level 427 Fig. vertex and external wave-function renormalization corrections we also neglect the contributions of the Higgs and Goldstone bosons whose couplings are proportional to the masses of fermions to which they couple.107) . up to terms O(m 2f /MW ). Vertex corrections.105) (a) α 1 − 2s 2 = 4π 4s 2 c2 =− 2 MZ 1 + ln 2 2 µ (b) 2 MZ α 3 c2 M 2 5 − + ln 2 − 2 ln W 2 4π 2s 2 6 µ s MZ (c) 2 MZ 5 α 3 − 6s 2 c2 M 2 − + ln 2 − 2 ln W =− 2 4π 2s 2 6 µ s MZ (d) =− 3α 4π 2 M2 5 MZ − + ln 2 + ln W 2 6 µ MZ Diagrams renormalizing external fermion lines are shown in Fig. For the neutrino self-energy.9. where ≡ (a) + (b) + (c) + (d) and (12.12.13. 12. the contribution of the external line renormalization to the effective lagrangian for muon decay reads LCC (ext lines) = eff 1 2 µL + νµ L + eL + νe L × LCC (Born) eff (12.85).106) where we have again neglected the charged lepton masses. one gets νµ (νe )L (0) = α 4π 1 2s 2 M2 1 + ln W 2 µ2 + 1 4s 2 c2 2 MZ 1 + ln 2 2 µ (12. Due to the presence of the projections PL in the µW νµ and eW νe vertices only L s contribute to the muon 2 decay amplitude. In the calculation of the box. 12. using the general formula (12. The W ± and Z 0 gauge boson contributions to the muon and electron self-energies are given by (12. We conclude that.83).

111) can be used to express s 2 (µ) in terms of G µ .79) and (12. (12. s 2 (µ) and M Z (µ) in 2 ˆ terms of αEM . We get 1 2A A 2 = c0 − 2 1 + (1 − 4A)1/2 − 2 1/2 2 (1 − 4A) c0 − s 0 (12. A ≡ παEM / 2 G µ M Z = s0 c0 .61). M Z and G µ . adding the correction (12.113) √ 2 2 ˆ2 and. (12.110) 2 The three equations can be solved for the parameters α(µ).428 12 Standard electroweak theory Collecting together all the corrections we get (note that corrections to the electron and muon ‘sides’ are equal in the limit m µ.112) 2 2 and c0 and s0 are given by (12. This task is simpliﬁed if we notice that.111) + −2 s0 c0 T γ Z (0) ˆ2 MZ + µL − ˜ γ γ (0) + Bγ W + BW Z + 2 + νµ L (12. For c2 (µ) = 1 − s 2 (µ) = .92) and (12. to the order we are working.80) we have 2 ˆ2 M Z = M Z (µ) + T 2 Z Z (M Z ) (12. Eq. as in (12.Z + 2 √ = 2 2 2 2 2 2s c M Z (µ) c M Z (µ) + µL + νµ L (12. 2 Inserting the obtained α(µ) and M Z (µ) into (12.92). The third one is the one-loop correction to the physical mass of the Z 0 boson.109) After (12.e → 0) Bγ W + BW Z + 2 × α = −4 ln 4π s 2 +2× 2 MZ µ2 1 2 µL + νµ L 2 7 − 12s 2 MW +6+ ln 2 2s 2 MZ (12.110) can immediately be inverted. this is the second equation which relates the measured quantities to the running parameters. From (12. Eqs.109) we get π αEM Gµ (1 + √ = ˆ2 2 2s 2 c2 M Z where with one-loop precision (µ) = − T W W (0) 2 ˆ2 c0 M Z T ˆ2 Z Z (MZ ) ˆ2 MZ ) (12. we can replace the running parameters by the measured ones in all one-loop corrections.51).108) Finally.72) to the W ± boson propagator computed earlier we get T (0) Gµ πα(µ) 1 − 2 WW + Bγ W + BW. With these results we can predict any other electroweak observable.

Those relations contain logarithms. We shall show there that it is possible to calculate αSM (µ ≈ M Z ) in terms of αEM with all large logarithms summed up to all orders. It should be remembered that the important step was the use of the strictly one-loop relations (12. (12.113).115)) which can make all of them simultaneously small. we can obtain the one-loop prediction for the W ± mass.115) we have deﬁned the correction r .115) This result is.109) is an important improvement. Then using that αSM (M Z ) in (12. In the last step of (12.114) and expressing c2 (µ) in terms of G µ . We shall return to this point in Section 12. renormalization-scale-independent.117) . From the previous equations we get r =− T W W (0) 2 ˆ2 c0 M Z + − ˜ γ γ (0) − 2 T 2 ˆ2 W W (c0 M Z ) 2 ˆ2 c0 M Z T s0 γ Z (0) − 2 c0 2 s0 T 2 ˆ2 W W (c0 M Z ) 2 ˆ2 c0 M Z − µL T ˆ2 Z Z (MZ ) ˆ2 MZ c0 ˆ2 MZ + Bγ W + BW Z + 2 + + νµ L (12.116) W Z can be approximated as r = ··· − 2 c0 ρ + ··· 2 s0 (12.116) Numerically. and there is no choice of the scale µ (which is arbitrary from the point of the ﬁnal result (12. Using the equation 2 ˆ2 MW = MW (µ) + T 2 W W (M W ) ˆ2 = c2 M Z + T 2 ˆ2 W W (c0 M Z ) 2 − c0 T ˆ2 Z Z (MZ ) (12. µ dependence in (12. the most important contributions to r are contained in factors 2 2 2 2 ˜ γ γ (0) and [ T W (MW )/MW ] − [ T Z (M Z )/M Z ].12.115) suffers from considerable inaccuracy introduced when expressing αSM (µ) in terms of αEM . the numerical accuracy is improved since these are the largest corrections to MW . which is the common notation in the literature.5 Beyond tree level 429 instance. Thus. Although it introduces only partial summation of higher order effects and. as such.92) and (12. the result (12.69) expressing αSM (µ) in terms of αEM . as it should be when a physical observable is expressed in a given order of perturbation theory in terms of other physical observables. one gets the ﬁnal one-loop result: 2 ˆ2 ˆ2 M W = c0 M Z 1 − 2 ˆ2 ≡ c0 M Z 1 − T ˆ2 Z Z (MZ ) ˆ2 MZ 2 s0 r 2 2 c0 − s0 + T 2 ˆ2 W W (c0 M Z ) 2 ˆ2 c0 M Z − 2 s0 2 2 c0 − s0 (12.115). The second term in (12.6. of course.

using the formulae for the b0 function.120) 2 2 The terms ∼Mh /MW cancel out.55) are modiﬁed. Higher order effects can be parametrized by the effective parameters ρ f G µ and sin2 θ eff .e.52) and relations (12. This is an illustration of the important screening of heavy Higgs boson effects (Veltman 1994). The custodial SUV (2) symmetry is broken by the Yukawa couplings and by the UY (1) current.56).119) For the dominant contribution of the Higgs boson to r≈ 2 Mh αEM 11 ln 2 2 ˆ2 16πs0 3 c0 M Z r one ﬁnds (12. different for different ﬂavour pieces of the lagrangian (12.54) and (12. They become ﬂavour-dependent.52). Two-body Z 0 decays can be described by a local effective lagrangian since they depend on kinematical invariants whose values are ﬁxed in terms of the external particle masses. All these observables are calculable from the effective lagrangian for the on-shell Z 0 coupled to a pair of fermions. Indeed.118) With a bit more work. † For several years CERN LEP and linear SLAC electron–positron colliders operated at the centre of mass √ energy s = M Z producing millions of Z 0 s at rest and allowing for detailed study of its couplings to light fermions. introduced in (12.430 12 Standard electroweak theory because T 2 W W (M W ) 2 MW − T 2 Z Z (M Z ) 2 MZ ≈ T W W (0) 2 MW − T Z Z (0) 2 MZ ≈ ρ (12. . loop corrections to ﬂavour conserving neutral-current processes can be summarized in terms of the parameters ρ f G µ and sin2 θ eff f (low) already introduced in the lagrangian (12. In the limit of q ∼ 0 and vanishing fermion masses.† It measures the partial decay widths of the Z 0 boson and various asymmetries of the decay products. it is possible to show that the leading m t dependence of r is: r ≈− 2 c0 3αEM m 2 αEM mt t + ln 2 2 2 ˆ2 2 ˆ s0 16πs0 c0 M Z 12πs0 c0 M Z (12. i. higher order corrections can also be summarized in terms of that effective lagrangian. which are different for different fermions and different from the f analogous parameters present in the low energy neutral-current lagrangian. Corrections to the Z 0 partial decay widths A very precise test of the electroweak theory comes from the measurements of the Z 0 boson properties at CERN and SLAC e+ e− colliders. Below we will compute radiative corrections to the effective Z 0 f f¯ lagrangian.

The relevant diagrams are shown in Fig.81) and .14. T Z (M Z ) is Z 2 2 the derivative at q = M Z of its transverse self-energy. Denoting the momentum of the Z 0 boson by q and the momenta of the outgoing fermion and antifermion by p1 and p2 . Diagrams for Z 0 boson decay. we have for these diagrams the following amplitudes (the factor T f3 − Q f s 2 in the couplings can be written as ± 1 (1 − 2|Q f |s 2 ) with the sign + 2 (−) for up (down) type fermions) (a) = ∓ ie u( p1 )γ λ (1 − 2|Q f |s 2 )PL + (−2|Q f |s 2 )PR v( p2 )ελ ¯ 2sc T 2 ie γ Z (M Z ) v( p2 )ελ (b) = ∓ u( p1 )γ λ 2sc|Q f | ¯ 2 2sc MZ (c) = ∓ ie 1 u( p1 )γ λ (1 − 2|Q f |s 2 ) ¯ 2sc 2 T 2 Z Z (M Z )PL + (−2|Q f |s 2 ) (d) = ∓ 1 2 T 2 Z Z (M Z )PR v( p2 )ελ ie u( p1 )γ λ [FL PL + FR PR ] v( p2 )ελ ¯ 2sc ie u( p1 )γ λ (1 − 2|Q f |s 2 )δZL PL + (−2|Q f |s 2 )δZR PR v( p2 )ελ ¯ (e) + ( f ) = ∓ 2sc The factors FL and FR denote one-loop vertex amplitudes (note that e/2sc has 2 been factored out). and δZL. ελ (q) is the polarization vector of the Z 0 boson.14.12.R (m f ) + 2m 2f 2 V (m f ) + 2m f 2 m (m f ) (12. 12.5 Beyond tree level 431 Fig.121) are the external fermion wave-function renormalization factors (see (12. respectively.R = 2 L. 12.

Its presence originates from expressing with one-loop accuracy the factor e(µ)/s(µ)c(µ) through measurable quantities. π (12. .125) The ﬁnal result for the partial decay width is still subject to QED and (for f = q) QCD radiative corrections.77). For the effective mixing angle we ﬁnd sin2 θ eff = s 2 1 − f c0 s0 T 2 γ Z (M Z ) 2 MZ − c0 s0 T γ Z (0) 2 MZ ˆ − FL + 1 − 1 2 2|Q f |s0 ˆ FR (12.126) with 0 given by (12.123) for the renormalized vertex corrections. Since the effective coefﬁcient of ρ is positive.† and is given by (12. (Z 0 → f f¯) increases with increasing top-quark mass. the ﬁnal one-loop result can be written in the form (12. Collecting all the factors together. For all fermions f except for the b quark‡ the vertex corrections are small. † Indeed.74) and (12. With O(αEM αs ) accuracy (including virtual and real emissions) we get (Z 0 → f f¯) = 0 (Z 0 3αEM 2 Qf → f f¯) 1 + 4π 1− αs .R ).56) with ρf = 1 − Here c0 2 ˆ FL ≡ FL + (1 − 2|Q f |s0 )δZL − 2 s0 ˆ FR ≡ FR − 2 2|Q f |s0 δZR T γ Z (0) + T ˆ2 Z Z (MZ ) +2 c0 s0 T γ Z (0) ˆ2 MZ ˆ ˆ + 2( FL − FR ) (12. in the popular on-shell renormalization scheme.123) are the ‘renormalized’ (scale-independent) vertex corrections. c and 2 e through the measurable quantities (as we did in the case of r ) and replacing the amplitude by the corresponding contribution to the effective action. one ﬁnds precisely the combinations (12.42).432 12 Standard electroweak theory compare with (5.124) where 2 s 2 = s0 1 + 2 c0 2 2 c0 − s0 (12.118) – which contains a term 2 proportional to m 2 /MW (12. ¯ ‡ For kinematical reasons Z 0 cannot decay into a t t quark pair.122) ˆ2 MZ (12. where the counterterms also contain ﬁnite parts (ﬁxed by some physical conditions). and which originates solely from expressing the t parameters s and c through s0 and c0 .112). expressing the MS parameters s. The corrections to ρ f and sin2 θ eff are then dominated by f the factor ρ present in – see (12. each external wave-function renormalization factor is 1 δZL.59).

16 and Appendix C reveals that only diagrams (e) and ( f ) contain 2 the factor m 2 /MW . The reason is that the top-quark mass is generated by spontaneous breaking of the gauge symmetry and. and the logarithmically divergent loop integrals can at most give ln m t . . Finally. We will compute the leading term. compensating factors m t /MW coming from the G + tb vertices. The relevant vertex corrections are shown in Fig.15. e mt +i √ 2s MW PR − mb PL mt e mt +i √ 2s MW PL − mb PR mt Fig. larger mass requires larger Yukawa coupling. This is an interesting example of when the Appelquist–Carrazzone decoupling theorem does not work. Inspection of the Feynman rules in Fig. 12. 12.5 Beyond tree level 433 ¯ Fig. m t -dependent vertex corrections to the Z 0 bb decay. 12. Charged Goldstone boson couplings to the (t. coming from the Yukawa coupling G + tb of the Goldstone ﬁeld t G + . 12. b) quark pair.12. 12.16. Diagrams (c) and (d) are ﬁnite. therefore. diagrams (a) and (b) do not have any m t /MW factors in their vertices.14).15. The loop integral goes as 1/m t . 2 m 2 /MW t ¯ For the bb pair in the ﬁnal state there is an additional contribution of order eff ˆ to the parameters ρb and sin2 θb . It comes from FL (the vertex corrections) and from the renormalization of the external b-quark line (diagrams (d)–( f ) in Fig.

12. Comparing with ρ.15( f ) we ﬁnd i e (f) e e2 m 2 t FL = 2 2sc 2sc 2s 2 MW × = 4 4 1 1 − s 2 PL − s 2 PR PL 3 3 / − mt k dn k k2 2 (2π )n (k 2 − MW )(k 2 − m 2 )2 t e3 m 2 4 2−n t − s2 3c M 2 4s 3 n W 4 + 1 − s2 3 dn k m2 t γ λ PL 2 (2π )n (k 2 − MW )(k 2 − m 2 )2 t 1 dn k n (k 2 − M 2 )(k 2 − m 2 ) (2π ) t W = 4 2−n e3 m 2 t − s2 3c M 2 4s 3 n W 2 + 1 − s2 3 dn k m2 t γ λ PL 2 (2π )n (k 2 − MW )(k 2 − m 2 )2 t m2 4 2 1 t s −η − + ln 2 3 2 µ 2 + 2 1 − s2 3 γ λ PL = −i e e2 m 2 t 2 (4π)2 8s 3 c MW 2 Finally. and neglecting the b quark mass Fig. we ﬁnd that for b quarks the new negative contribution (12. the full contribution of order m 2 /MW to the renormalization factor δZL ≈ t ± L can be obtained by calculating the contribution of G to the self-energy of the b quark. 12. c by s0 . One obtains G± L (0) = 1 e2 m 2 t 2 (4π )2 4s 2 MW −η − m2 3 + ln t2 2 Q +O 1 mt (12. . for the Fig. for vanishing external momenta. t (12. i.434 12 Standard electroweak theory In the limit MW.127) Collecting all contributions together (and replacing s.128) 2 ˆ There is no similar contribution of order m 2 /MW to FR .128) of order † We also approximate the CKM matrix element Vtb by 1.Z m t . c0 ) we get 2 2 ˆ 2 FL = 2 FL + 1 − s0 3 G± L + ··· = − α m2 t 2 2 ˆ2 4π s0 c0 M Z (12.77).15(e) gives† i e (e) e3 1 − 2s 2 m 2 t FL = 2 2 2sc 2s 2sc MW = −i dn k 1 2k λ PR PL 2 2 (2π )n / − m t (k − MW )2 k γ λ PL m2 3 e3 1 − 2s 2 m 2 1 t t −η − + ln 2 + O 2 8s 3 c M 2 (4π) 2 µ mt W dn k 1 1 PR γλ n k2 − M 2 (2π ) / − mt k W Similarly.e.

12. An effective theory for light particles can be obtained by calculating the Green’s functions with only light ﬁeld external legs at small momenta (compared to heavy masses) and with all heavy and mixed light–heavy particle loops. namely. They can be considered as given in terms of the parameters of the original theory (matching conditions) or as free. decrease of sin2 θb with m t . It is also very interesting to look at the previous results from another perspective. µ . The same correction (12. As a result.4 and 12. The UV divergences arising in these graphs must be renormalized by appropriate counterterms (for example. apart from those present in the running coupling α(µ). In a renormalizable theory with several mass scales. which one encounters when calculating low energy.50) one can convince oneself that it does not contain large 2 logarithms. Suppose we calculate at one-loop level the muon decay amplitude in the complete theory. renormalizable parameters of the local effective theory. ln(MW /m 2 ). and to compare it with the systematic perturbative calculation performed in the previous section. The main virtue of such a separation of mass scales is the possibility of easy resummation to all orders of large logarithms.12. (Z 0 → bb) decreases as m t grows. summing up the terms O(α ln(MW /m f )) is less important than the terms O(αs ln(MW /m f )) (QCD corrections) discussed in the next section.7. but by inspection of (12. effective. to introduce the effective low energy theory in its full sense deﬁned in Section 7.109) and (12. compared to effective eff Weinberg angles for other fermions. This result has not been explicitly written down. but it is interesting as an illustration of the method. with the muon decay amplitude as an example. In general. E M. the subsequent decoupling of heavy particles gives ﬁeld theories which are valid in a smaller and smaller energy range. parameters. It is interesting to discuss this technique in the electroweak theory. ln(E/M).6 Effective low energy theory 435 2 m 2 /MW to the parameter ρb overcompensates the positive universal contribution t ¯ of the same order. We have used the language of effective lagrangians. they are described by lagrangians which consist of dimension four or less (renormalizable part) and higher dimension operators.5 we have presented a sample of systematic tree level and one-loop calculations in the standard electroweak theory. Of course. in the MS scheme).6 Effective low energy theory for electroweak processes In Sections 12. processes in the full theory containing the large mass scale M. In the ﬁnal step one redeﬁnes the light ﬁelds and parameters so that the whole dependence on the heavy ﬁelds is absorbed into those new. but that was merely as a local approximation to the effective action for the considered range of phase space.128) is also responsible for the slightly slower.

we have to require equality of the full SM one-loop amplitude and the amplitude generated by lagrangian (12. . 2 J − J µ+ . by decoupling heavy particles: vector and Goldstone bosons. With 2 † A systematic expansion in inverse powers of MW would also give higher dimension operators. since the counterterm diagrams in Fig 12. let us proceed with the construction of the effective theory. to be able to identify the Wilson coefﬁcient at one-loop accuracy. and (12.129). pathological.129) where G F is given by (12. Putting aside (for a moment) the pathology of our example. from the point of view of minimizing higher order corrections to the Wilson coefﬁcient G F (µ). and for the light fermions are negligible compared to the f corrections O(αG µ ) to the Wilson coefﬁcients G f . they are suppressed by O(m 2 /MW ).49).† Thus.e. The local.436 12 Standard electroweak theory This is pathological from the point of view of perturbative calculations in the full theory. The basic idea is then to apply the RGE in the effective theory to evolve G F (µ) down to the scale µ ∼ m µ . In the present. which generically give such logarithms of two different mass scales. we know already the one-loop corrections to the Wilson coefﬁcient G F : they are given by (12. Compared mf µ 2 to (12.109)! And here comes an important observation: the correction to the Wilson coefﬁcient depends on the renormalization scale in the effective theory (remember that in the calculation of the previous section the scale µ is the renormalization scale for QED corrections to the four-fermion operator) and involves a logarithm which can be large. In fact. operators present in lagrangian (12. the tree level effective lagrangian must be interpreted as the lagrangian at the decoupling scale µ ∼ MW . for example. as well as terms with different chirality structure (V + A currents and scalar currents).3 cancel out. describing muon decay. after decoupling heavy particles. for instance. with QED corrections included. Clearly. with one-loop QED corrections included. and has now the interpretation of the Wilson coefﬁcient. obtained for four-fermion processes in the limit q → 0. Higgs bosons and the top quark.129). such a procedure gives the effective theory described by the lagrangian √ − L = LQED − 2 2 G F J +µ Jµ (12. In consequence. The Wilson coefﬁcients of the effective four-fermion operators can be calculated in each order of perturbation by comparing the leading term in the expansion in powers of 2 (1/MW )n of the amplitudes calculated in the complete electroweak theory with the amplitude generated by (12. dimension six. and to calculate the muon-decay matrix element at that scale.129) are the lowest dimension operators describing weak transitions. i.48) is the equation for the Wilson coefﬁcient G F (µ) at µ ∼ MW .129). example G F (µ) is renormalization-scale-independent at the one-loop level (a fact which is directly connected to the absence of logarithms of two separate scales in complete one-loop calculation). It is minimized for µ ∼ MW . at the tree level. and these are the logarithms that one would like to resum to all orders.

obtained by the Fierz rearrangement. and in consequence in (12. that our example is pathological). for example. In this way we would resum all terms α(α ln(MW /m f ))n ) of the full theory. in terms of the low energy observable G µ . The perturbation theory based on lagrangian (12. therefore. (12. In our pathological example G F is renormalization-scale-independent at any order. Such contributions are effective dimension > 6 operators. and of course each new coupling requires new renormalization constant (the theory is not renormalizable with a ﬁnite number of counterterms). however.12. and in the present case we use LCC only in the Born approximation. In the next order of perturbation theory. One can see it by the following argument.6 Effective low energy theory 437 G F (MW ) = G F (m µ ).98). we expect to minimize two-loop corrections.51). at the tree level that matrix element does not depend on the renormalization scale but it should be understood as taken at the scale µ ∼ m µ . Again. This is a product of the neutrino current (which does not interact with photons) and the current which. (12. be taken as µ ∼ m µ to minimize two-loop corrections to the matrix element (remember. The power of the effective theory approach resides in the truncation of this series. Finally we calculate the muon decay matrix element by calculating ﬁnite parts of the one-loop QED corrections in the effective theory. Although those corrections do not contain logarithms of the renormalization scale. also do not contain any logarithms of the renormalization scale. if they had been present. we include one-loop corrections to G F . in principle. This is an important improvement for calculating at the tree level electroweak observables at the W ± mass scale. in the limit of vanishing fermion masses.7). is now α(M Z ).48). we should evolve G F (MW ) to G F (m µ ) with two-loop RGE in the effective theory.) The main difference of this approach compared to the tree level calculation in the full theory is that α in (12. is conserved (and. of course. . (Here we encounter the same pathology: one-loop corrections to the matrix element. the W ± mass. does not undergo renormalization). to minimize higher order corrections to it.129) would include higher order contributions generated by perturbation in the four-fermion operator itself. with new dimensionful coupling constants ≤ O(G 2f ). Choosing µ ∼ M Z . Suppose we calculate higher order QED corrections to the four-fermion operator written in the charge retention form ¯ e γ λ PL µ ¯ νµ γ λ PL νe . the latter should. Thus.109). we can ﬁnally calculate the muon-decay matrix element at the tree level (to match the overall accuracy). eff consistent with the fact that operators with dimension > 6 have already been neglected in the construction of the effective theory. for consistency. Eventually we can express G F (MW ) in terms of G µ with the next-to-leading logarithm accuracy. all inﬁnities generated by this four-fermion vertex can be renormalized by the ﬁeld wave-function renormalization constants and the Wilson coefﬁcient is renormalization-scale-independent (see Section 7. Next.

n u are. our goal is to use each effective QED only in the energy range m n < E < m n+1 . s. by its own coupling constant αEM. the main virtue of effective theories (which is easy resummation of large logarithms) is achieved if this theory is used only in the ˆ energy range m b < E < M Z . in particular. starting with the decoupling of the weak sector and the even heavier top quark we obtain QED describing the interaction of photons. down. it still contains hugely different mass scales (fermion masses). However.n u ) (µ) 1 + α 3π n e . where m 2 is the next mass threshold).n d .129). This is the usual bottom-up attitude in which we study physics in a certain energy range without even knowing the more complete theory. c. d. Since the theory is renormalizable. τ ) and ﬁve quark ﬂavours (u. Eventually. we can integrate out subsequent quark and lepton ﬂavours (in the order of their heaviness). by repeating in these simpler cases the procedure used for calculating αEM in the complete electroweak theory).n d . and we are able to prove by the decoupling procedure that the usual bottom-up attitude is correct (Appelquist & Carrazzone 1975). respectively. and ﬁnally arriving at the QED of photons and electrons valid in the range 0 < E < m µ . obtaining each time a new renormalizable effective theory valid in a smaller and smaller energy range. This effective QED is valid for ˆ the energy range 0 < E < M Z . its structure does not contain any information about the energy range of its applicability. The result is .e.and up-type quarks. even at the level of pure QED we can discuss several steps of ‘effectiveness’: in the top-down approach.n u Nc f Q 2f ln m 2f µ2 +O m2 n m2 (n+1) (12. the numbers of the charged leptons. µ. In the next step. present in the effective theory. In the one-loop approximation we get: (n e αEM = αEM. b).438 12 Standard electroweak theory QED as the effective low energy theory It is now worthwhile to discuss in more detail the connection between the complete electroweak theory and the renormalizable part of the effective theory deﬁned by lagrangian (12.130) These equations are obtained by calculating in each effective QED the k → 0 limit of the Thomson scattering (i. The ﬁrst point to stress is that the renormalized (in the M S scheme) coupling constant for each effective QED can be ﬁxed in terms of αEM . three lepton ﬂavours (e. namely the QED of Chapter 5.n u ) (µ). In the energy range of interest. In the present case we are fortunate to know both the complete theory and the effective theory. the corresponding ‘minimal’ effective renormalizable theory is not only sufﬁcient but also best suited to describe physics in that range (up to corrections O(q 2 /m 2 ). where n e . Actually. Thus. n d . Each effective QED is described by its proper scale-dependent lagrangian (n e parameters and.n d .

Of course.0) αEM (µ) = 1 (1. and in this theory improved accuracy can be achieved only by calculating higher order corrections to this equation.130) reﬂects quite clearly the problem of large logarithm resummation. We can then switch to the next effective QED parametrized by the (2. n d .2) αEM (µ).133) d (1. we have µ where (in general) b1 (n e . we obtain αEM at the scale µ = m µ . they are also absent in higher order corrections to this relation used in the effective QED of only photons and electrons.0) of αEM (µ2 ≈ m 2 ) in terms of αEM is O(α 2 ) since no large logarithm appears e in this case in the one-loop (12. n u ) = 4 (n e + 1 n d + 4 n u ) 3 3 3 (12.130) can take any value in this range. i.0) valid. 0.132) is the ﬁrst coefﬁcient of the β-function β(n e .0) Thus.130) is valid up to non-perturbative corrections to the photon propagator discussed in Section 12. It is clear from (12. we can proceed in a different way. n u the renormalized coupling is deﬁned only in the range 0 < µ < m h . αEM .130) that (1. When we calculate the low energy observable.131) at some µ0 ≈ m e .0. n u ) with n e . and the µ scale in (12.0) running αEM (µ) (with β(2. between αEM (µ) and αEM (µ) for µ ≈ m 2 (for the sake of µ . where m h is the mass of the lightest decoupled fermion. (In fact. (12.0. the solution reads: 1 (1.0) αEM (µ) = b1 (1. The accuracy of the determination (1. in the one-loop approximation for the β-functions. (12.e.3. for ﬁxed n e . 0) ln 2 4π µ0 (12. n d .0.0) αEM = αEM (µ ≈ m e ) + O(α 2 ).0) αEM (µ0 ) − µ2 1 b1 (1. 0)). up to the next threshold etc. (12. in an effective QED with several ﬂavours. Eq.0.0. n d . However.0.131) the result already known from Chapters 5 and 6. We also remark that (12.0.130) contains large logarithms for any value of µ. Summing up leading logarithms.0.134) (1.0) 1 (1.0) (2. In order to accomplish this programme. we have to know the relations between the couplings of two different effective QEDs in the range of µ where both are (1.0. n u ‘active’ fermions.0) Once αEM (µ) is ﬁxed by (12. until we reach (3.0.0. 0. its µ dependence in the MS is determined by the RGE. 0. for example.) (1. with all terms [α ln(m µ /m e )]n summed up.6 Effective low energy theory 439 given by the corrections to the photon propagator. For any µ in the range of the validity of the effective theory.130).12. n d . 0 < µ < m µ . 0)(αEM )2 dµ 2π (12..5.

The one-loop approximation. and (12. The ﬁnal step is to use the matching condition (3. but this is not a necessary choice of µ.92) and to express αSM (µ) in terms of αEM with all the large logarithms summed up to all orders.2) 2) 3π αEM (µ) αEM (m e 1 1 Nc f Q 2f f =t ln m 2f µ2 (12. (12. therefore. for µ ≈ M Z . It is.0) αEM (µ) = αEM (µ) 1 + m2 α µ ln 2 3π µ +O m2 e .0.5. as in Section 12.2) = αEM (µ) + O(α 2 ).135) can 1.† (Another way to get (12.0.0.2) αSM (µ) = αEM (µ) 1 − α 4π M2 2 16 m 2 t − 7 ln W + ln 2 3 µ2 9 µ (12. for µ = m µ we get αEM = αEM be used for any µ such that (α/3π) ln(m 2 /µ2 ) µ (1.92).0. renormalizable parameters from the point of view of the latter theory. Such corrections can be neglected as long as we work with one-loop RGE.0) αEM (m 2 ) = αEM ).136) i.135) is to begin with two-lepton theory and to follow the 1.0.135) for µ ≈ m µ .0) (2.137) are small we can rewrite † Of course. These are the so-called matching conditions.0) . Taking it as αEM . the leading logarithms are summed up to all orders (we remember that (1. .137) (3.) It is very easy to see that our procedure ﬁnally gives 1 = (1. the two couplings are equal up to corrections O(α 2 ). αEM (µ) is the effective coupling constant of the ‘effective’ one-lepton theory obtained from the full two-lepton theory. with one-loop accuracy we get from (12. Suppose that. (12.0) (2.0) + (3. e The procedure which we have outlined here is important in the following context. Then. We see that in this simple example.3.440 12 Standard electroweak theory our examples we assume that all quarks are heavier than leptons).135) illustrates the general feature of decoupling in the MS scheme: the dependence on the heavy states is absorbed into effective parameters of the lower energy theory. very desirable to go beyond the accuracy of (12. Since at µ ≈ M Z the logarithms in (12. Eq. which are free.e.0.3. One way to get these relations is to calculate the same physical quantity in both theories.0 decoupling procedure. This is indeed achieved by using our effective theory approach. m2 µ (12. we want to ﬁx αSM (µ) in terms of αEM . for µ ≈ m µ . contains large logarithms (there is no scale µ for which all of them can be small).130): (1.3.0.

only the global baryon number is conserved.3 that the structure of the electroweak theory is such that it ensures the absence of the tree level ﬂavour changing neutral currents. for C P violation in the K –K system). see for example (Chankowski. This follows from the SUL (2) × UY (1) gauge invariance. One is that often several mass scales must be decoupled at the same time. If one wishes. Then. of course. It reﬂects the Ue (1) × Uµ (1) × Uτ (1) global symmetry of the lagrangian (12. in theories of grand uniﬁcation. in 0 ¯0 particular. In such cases the two-loop evolution of coupling constants is discontinuous.92). Our ﬁnal comment is that (12. For quarks. ln(µ/M Z ) ln(m f /M Z ) and the accuracy of (12.5). the lepton number conservation (separately for each lepton ﬂavour) is exact. the generic examples of ﬂavour changing neutral-current ¯ ¯ ¯ transitions are the reactions d s → ds ( S = 2). while the quark mixing explicitly breaks the quark ﬂavour U (1)s. b → sγ ( B = 1). This is the desired improvement over the one-loop formula (12. it is interesting to discuss the predictions for them in the electroweak theory.7 Flavour changing neutral-current processes 441 our result in the compact form: 1 1 1 + = αSM (M Z ) αEM (4π) M2 2 4 − 7 ln W + 2 3 3 MZ Nc f Q 2f ln f m 2f 2 MZ (12. In some very interesting physical questions. If both logarithms are small enough the whole procedure is self-consistent.137) (the so-called ﬁnite threshold correction) 3 by using Dimensional Reduction (DR) scheme instead of the MS. with a real or virtual photon. for radiative ﬂavour changing .138) requires for consistency the two-loop renormalization group evolution in that large µ range.138) where the sum extends to all fermions including the top quark. Moreover. one can avoid the constant ( 2 ) in (12. Both neutral gauge boson and Higgs boson couplings are diagonal in the ﬂavour mass eigenstate basis. for example. Thus. Płuciennik & Pokorski 1995). Such processes are. Since they are strongly suppressed in Nature. At the quark level. 12. Such transitions ¯ ¯ are responsible for physical processes like K0 –K0 and B0 –B0 mixing (and.137) illustrates two generic points. calculable without any new counterterms.12. bd → bd ( B = 2) and s → ¯ dγ ( S = 1). renormalizability and the particle content of the theory.7 Flavour changing neutral-current processes We have learned already in Section 12. Another general fact is the presence of some constant factors in the one-loop matching conditions. the ﬂavour changing neutral-current processes involving quarks are generated in higher orders in the electroweak interactions. αSM (M Z ) is used as input parameter in the renormalization group evolution up to scales µ ≈ 1016 GeV.

12. 12. Suppose we want to . j=u. For the sum of the two diagrams 12. We shall be interested in the off-shell 1PI Green’s functions which build the effective action. it is approximately local and can be interpreted as an effective lagrangian. in the ’t Hooft–Feynman gauge.442 12 Standard electroweak theory ¯ Fig.t d] q q d4 q [ ¯ s γµ PL (/ + m qi )γν PL d ][ ¯ s γ ν PL (/ + m q j )γ µ PL 2 2 2 (2π)4 (q 2 − m qi )(q 2 − m q j )(q 2 − MW )2 (12. Similar processes occur in the up-quark sector. (2.139) The colour indices are contracted within square brackets. too. for example.17(a) (G denote the charged unphysical Goldstone bosons). but are less interesting phenomenologically. for example. The corresponding contribution to the effective action is ﬁnite and has dimension [mass]−2 .c. (b) penguin diagram contribution to a S = 1 process.132). we have ¯ ± the contribution from the diagrams shown in Fig. Predictions for physical processes are obtained by convoluting the quark offshell amplitudes with the hadron wave-functions or (in practice – as we shall see) by calculating the hadronic matrix elements of operators present in the effective lagrangian. we have Leff = 1 2 × e √ 2s 4 Vis Vid V js V jd i. decays of strange and bottom mesons (like. for the s d → s d transition. In the limit of small (compared to the gauge boson mass) external momenta which interests us.17. the celebrated b → sγ decay) and for decays like K → π e+ e− or B → K e+ e− . we remain at the quark level and estimate qualitatively the expected order of magnitude for the S(B) = 2 and S(B) = 1 amplitudes. For the time being.17(a) with the W + W − exchange. as can be checked by general techniques developed in Chapter 2 for obtaining the Feynman rules from a lagrangian. The two sets of diagrams enter with a relative minus sign. due to Fermi statistics of the quark ﬁelds. ¯ For the S(B) = 2 transitions. (a) Leading Standard Model contribution to K0 –K0 mixing.

in fact. The other two contractions give the same result.139) into (2. the physics which determines the quark Yukawa couplings). 2 mq j ∼ αG F Vts Vtd +O i. .139) is strongly suppressed by its very small mixing with the ﬁrst two generations of quarks. u.17(a).140) In the last line.e. this prediction follows not only from the structure of the theory but also depends on the empirical pattern of the quark masses and mixing angles. (As we shall see. Such a mechanism of suppression of the ﬂavour changing neutral-current amplitudes is known as the Glashow–Iliopoulos–Maiani (GIM) mechanism. the successful predictions for the ﬂavour changing neutral-current processes should be considered as fairly accidental and. i. We insert the effective lagrangian (12. Thus. j=u. because couplings of G ± to light quarks are suppressed by m q /MW .c 2 2 m qi 2 2 MW MW 2 mq 2 MW .12. We see that the strong suppression of the ﬂavour changing neutral-current transitions is indeed a prediction of the Standard Model. it is probably the (unknown) physics beyond the Standard Model which is responsible for it (namely. The contribution from the top quark exchange in (12.and/or c-quark exchange: Auc ∼ e √ 2s 4 1 2 MW Vis Vid V js V jd 1 + O i. ¯ On dimensional grounds. we need an extra combinatorial factor of 1 in 2 front of (12. For the box diagram there is no logarithmic term. we can take the limit m qi MW . j=u. which are proportional to larger CKM angles. with the relative minus sign. Therefore.c Vis Vid V js V jd (12.202). j=u. this suppression is just the right one for accounting for the measurements). However. We shall return to this point shortly. 2 † The dimensionless coefﬁcient of the term 1/MW could also include the term ln(m q /MW ). they are O αG F m q /MW .† The same refers to contributions from G ± exchange. to recover the result of the calculation in the original theory. we then get the following estimate for the s d → s d ¯ transition amplitude with double W -boson. unitarity of the CKM matrix has been used: i. The remaining terms.139) with the external ﬁelds that contribute to the 1PI four point Green’s function. and make use of unitarity of the Cabibbo–Kobayashi–Maskawa (CKM) matrix.c Vis Vid = −Vts Vtd . Two contractions give the two sets of Fig. Consequently.139) to the Green’s function 0|T s ¯ d s ¯ d |0 . the leading αG F term is suppressed by the same very small CKM angles as the double top quark exchange contribution.7 Flavour changing neutral-current processes 443 recover the contribution of the effective vertex (12.139). from the point of view of the Standard Model. There are four possible ways of contracting the ﬁelds in Leff (12. As far as the u and c exchange is concerned. are in turn suppressed by light quark 2 2 masses. 12.

since the masses of the up and charm quarks are quite different. On 4 dimensional grounds. Since the ﬂavour changing neutral-current transitions involve strongly interacting quarks. In processes where the GIM mechanism is only logarithmic. Auc ∼ αG F Vid Vis ln + O(Vtd Vts ) = αG F Vud Vus ln 2 2 † The leading term behaves like 1/MW . perturbative calculations can be trustworthy only if all the quark masses under logarithms are larger than the QCD conﬁnement scale. (In practice.141) could be used in a quantitative calculation only if the up-quark mass was larger than ∼1 GeV. 2 m qi . they receive contributions from box diagrams and also from the so-called penguin diagrams like. They are often important phenomenologically. which means that the power-like GIM mechanism occurs. we obtain a UV-divergent one-loop diagram with a single four-quark interaction.141) can be considered only as an indication that non-perturbative strong interactions play an important role in the quark-level s → de+ e− transition amplitude. but only logarithmic in the case of certain penguin diagrams. In the ﬁrst case we get either a tree level (UV-ﬁnite) diagram or a one-loop diagram with two four-quark vertices. the one shown in Fig. j=u. We often say that the GIM mechanism is power-like in the case of box diagrams. 2 m2 MW c i. At one-loop.141) This time. In the penguin case. Since this is not the case in Nature. The corresponding amplitude behaves as m2 u + O(Vtd Vts ). the dimensionless coefﬁcient of the αG F term† contains logarithms of light quark masses. respectively.) This qualitative difference between penguin and box diagrams can be seen either by explicit calculation or by the following heuristic argument.444 12 Standard electroweak theory It is also interesting to comment on the S(B) = 1 transitions at the qualitative level.c (12. The 1/k 2 from the photon propagator is cancelled by the k 2 /MW factor coming from the vertex function. for example. no additional suppression except for the usual αG F factor occurs in this case. The MW dependence of the amplitude can be traced back to the behaviour of the effective ﬁeld theory diagrams which are obtained by contracting the heavy internal lines in the box and penguin diagrams. the latter diagram is proportional to 1/MW . In the ﬁnal subsection of this chapter we describe these techniques using the concrete example of C P violation in the neutral kaon system. The logarithmic part of the vertex function is independent of quark masses and vanishes due to unitarity of the CKM matrix. and their calculation strongly relies on the effective theory approach. Thus (12.17(b). 12. one of the most interesting ﬁeld theoretical aspects are the QCD corrections to them. This signals the possibility of only logarithmic GIM mechanism. unlike the previously considered box diagrams. (12.

In the following.5). Summation over indices i. 12.t d ¯ s γ µ PL d (12. x j ) − 2xi x j B(xi .143) 2 G 2 MW λi λ j F A(xi . The complete result for the Leff given by (12. 2 (1 − xi ) (1 − xi )(1 − x j ) (12.144) where B(xi . x j )] (12. x j ) 4 (12. G F is given by (12.139) accounts only for the W – W exchange diagrams) with either two unphysical Goldstone scalars in the loop or one such scalar and one W ± -boson are calculated in an analogous way. the diagrams shown in Fig.147) 2 G 2 MW F 4π 2 ˜ λi λ j S(xi . up to O(external momenta/MW ). we shall ﬁrst discuss the calculation of these diagrams and QCD corrections to them.139) is easy to obtain.17(a).48) and xi = m qi /MW . Next. j = u. 12. t is understood in (12.17(a) can be reproduced by the following effective four-quark interaction lagrangian Leff (box) = − where ˜ S(xi . this amplitude is generated by the s d → s d amplitude ¯ which receives the leading-order contribution from the box diagrams presented in Fig. c. j=u. ¯ At the quark level. it can be written in the following form: Leff (W W ) = − with A(xi . The products of the CKM matrix elements ∗ have been denoted shortly by λi = Vis Vid and colour indices are contracted within brackets. x j ) replaced by 1 x x [A(xi . x j ) = 1 + 1 xi x j A(xi . x j ) 4 i j − 8B(xi .142) 2 2 To this order. Using the formulae from Appendix D and the identity (12.12. The sum of these diagrams is reproduced by an effective lagrangian which has the same form as Leff (W W ) with A(xi .96). x j ) = 1 xi − x j xi2 ln xi 1 − (i → j) + .145) Thus.c. we shall turn to the explicit form of their relation to the C P violation parameter εK . x j ) ¯ s γµ PL 4π 2 d ¯ s γ µ PL d (12. The remaining box diagrams (remember that (12.146) .7 Flavour changing neutral-current processes 445 QCD corrections to C P violation in the neutral kaon system We are interested in C P violation in the kaon–antikaon transition amplitude (for the general discussion of C P violation in the kaon system consult Section 1. x j ) = 1 xi − x j 1 xi ln xi − (i ↔ j) + 2 (1 − xi ) (1 − xi )(1 − x j ) (12. x j ) ¯ s γµ PL i.142).

They are described by diagrams obtainable from those shown in Fig. In other words. The kaon–antikaon transition amplitude we look for is not perturbatively calculable even when large QCD logarithms are resummed. One could then ask why we intend to perform the perturbative logarithm resummation instead of applying non-perturbative methods immediately. It is here where the main virtue of using the effective theory is. . 12. Q 2 stands for all possible Lorentz low invariants made out of momenta and masses of external particles in the considered amplitude. i. one needs to use non-perturbative methods like. 12.17(a) by connecting the quark lines by an arbitrary number of gluon lines. the logarithms are large. and the ordinary expansion in αs breaks down.18. 12. gluons and quarks with momenta close to MW or m t (i.18. An important property of these corrections is that their ordinary perturbative ex2 pansion is actually an expansion in powers of αs ln(MW /Q 2 ) and αs ln(m 2 /Q 2 ) t low low rather then in powers of αs itself. for instance. ¯ the purely perturbative resummation of large logarithms in the s d → s d amplitude ¯ will automatically provide us with quantities suitable for applying non-perturbative methods. like that shown in Fig.e.17(a) is subject to QCD ¯ corrections.e. since we can then easily apply the (perturbative) renormalization group methods for resummation of large logarithms. This procedure will be described in some detail below. Sample QCD correction to K0 –K0 mixing. it is at most of order of the neutral kaon mass m K 0.446 12 Standard electroweak theory ¯ Fig. In order to ﬁnd it. for example. As we shall see. very much larger than QCD ) cannot be accounted for by non-perturbative methods in the case when external states are light hadrons at rest. 12. The answer to this question is that the accessible non-perturbative methods are not suited for calculating quantities which receive signiﬁcant contributions from rapidly varying ﬁeld conﬁgurations.5 GeV squared. Since MW 80 GeV and m t 175 GeV. lattice calculations. Here. ¯ The s d → s d amplitude presented in Fig. We need to resum these large logarithms from all orders of the perturbation series.

The Wilson coefﬁcient C(µ) depends on the scale µ at which the effective theory ﬁelds and parameters are renormalized (see Section 7.7). Each of the two diagrams ¯ ¯ has a combinatoric factor of 2.18 on the Standard Model side.17(a) with only top quarks in the electroweak loop. At the leading order in QCD. Tree level d s → s d amplitude generated by Leff . At the leading order in QCD.12. 12. At the next-to leading order in QCD.148) 2 up to non-leading corrections in Q 2 /(MW or m 2 ) and higher order electroweak t low corrections. µ0 denotes the renormalization scale used in the effective theory when the comparison of the full and the effective theory amplitudes is performed.19.149) Here. one needs to keep in mind that we intend to effectively resum large .7): Leff = LQCD (gluons and light quarks) G2 M 2 − F 2W λ2 C(µ) ¯ s γµ PL d t 4π ¯ s γ µ PL d (12. 12. this means requiring equality of the box diagrams in ¯ Fig. and by all the QCD corrections to this diagram. i. 12. On the effective theory side we would have to include one-loop QCD corrections to the vertex generated by Leff (see Fig.e. 12.19). it may seem strange that we specify this renormalization scale even though the tree level diagram in Fig.7 Flavour changing neutral-current processes 447 Fig. At the ﬁrst glance. for the amplitude generated by the diagrams in Fig.146) immediately implies that ˜ C(µ0 ) = S(xt . the result we have already obtained in (12. For simplicity. xt ) + O(αs ) (12. 12.20).17(a) (with only top quarks in the loop) and the tree level d s → s d ¯ amplitude generated by Leff (see Fig.19 obviously requires no renormalization. 12. 12. The fact of basic importance here is that this amplitude can be reproduced to all orders in QCD by the following effective lagrangian (see Section 7. we would need to take into account diagrams like the one shown in Fig. However. The numerical value of C(µ) can be found perturbatively in QCD by requiring equality of the effective theory and full Standard Model amplitudes up to O(external momenta/MW ). we shall discuss the procedure of large logarithm resummation in detail only for the top quark contribution.

up to 2 non-leading corrections in Q 2 /(MW or m 2 ). After renormalization.19. Let us have a closer look at how the size of µ0 is being speciﬁed. The dependence on ln Q 2 cancels in the low difference of the full and effective theory amplitudes to all orders in QCD. 12. the QCD corrections to the leading order result for the coefﬁcient 2 C(µ0 ) depend on αs ln(MW /µ2 ) and αs ln(m 2 /µ2 ). 12. Each of the diagrams shown has its partner in which fermion lines in the four-fermion vertex are differently contracted. On the other hand. Consequently. shown in Fig. 12. not much larger than unity). It turns out that such large logarithms are absent so long as the renormalization scale µ0 is chosen to be of the same order as the heavy particle masses MW and m t in the full Standard Model diagrams.e.18 contains logarithms ln(MW /Q 2 ) and ln(m 2 /Q 2 ).e.20. In order to make the latter two t 0 0 logarithms small (i. 12. the t low low very logarithms we would like to resum. and which enters with a relative minus sign due to the Fermi statistics of quark ﬁelds. we choose µ0 of the same order as MW and m t . they contain logarithms ln(µ2 /Q 2 ). even if we restrict ourselves to the leading terms in the evaluation of C(µ0 ). as in Fig.20. are UV-divergent.18 with those in Fig. .20. 12. 12. the box 0 low 2 diagram in Fig. (This is a general fact: Wilson t low coefﬁcients are insensitive to low energy behaviour of the considered theory). Thus. we must make sure that no large logarithms reside in the neglected higher order QCD corrections to this coefﬁcient coming from comparing amplitudes like the one in Fig. One-loop QCD corrections to the d s → s d amplitude generated by the ¯ ¯ effective lagrangian. logarithms from all orders of the QCD perturbation series for the full Standard Model amplitudes.448 12 Standard electroweak theory Fig. The QCD corrections to the tree level diagram. i.

In order to render the amplitudes generated by Leff ﬁnite.153) † The factor BK (µ) is deﬁned so that BK (µ) = 1 when only the vacuum state is introduced as an intermediate state between the two currents in the matrix elements in (12.7. 0 low Thus. This requires ﬁnding the matrix element of the four-quark operator present in Leff between the kaon and antikaon states at rest.12.7 Flavour changing neutral-current processes 449 ¯ Once the Wilson coefﬁcient C(µ) is found. (7. we need to evaluate the K0 → K0 transition amplitude generated by Leff . Performing renormalization group evolution of the Wilson coefﬁcient would be necessary even if the matrix element was perturbatively calculable. Non-perturbative methods used for evaluation of BK (µ) are not applicable when µ is arbitrarily large. The where m K is the kaon mass. the argument for using low µ in calculation of the matrix element would be identical to the one we have used to ﬁx the size of µ0 .150). i.139): µ d − γc C(µ) = 0 dµ (12. it is still a function of the renormalization scale µ used for UV renormalization of amplitudes generated by the considered four-quark interaction.e.150) 161 MeV is its decay constant.151) ¯ s γ µ PL d 2 → Z 2 ¯ s γµ PL d ¯ s γ µ PL d (12. We would then argue that the presence of large renormalization scale µ0 worsens the perturbative expansion of the matrix element due to large logarithms ln(µ2 /Q 2 ). maximal energy scales available in lattice calculations are much smaller than QCD × (size of the lattice)/(lattice spacing). .4)).152) The constant Z 2 in the above equation is the usual QCD renormalization constant of the quark wave-function (see (8. renormalization group evolution of the Wilson coefﬁcient from µ0 ∼ MW to µlow = (a few)× QCD must be performed before calculation of the ﬁnal low energy amplitude. (7. we need to perform the standard renormalization of the LQCD part of it. This matrix element is usually parametrized as follows (for states normalized to unity): 2 ¯ s d3 x K0 |(¯L γµ dL )(¯L γ µ dL )|K0 = 1 BK (µ)m K f K s 3 (12. and f K parameter BK (µ) cannot be calculated perturbatively because it depends on QCD interactions at energy scales close to QCD .† Nevertheless. as well as renormalize couplings and ﬁelds in the effective part: C(µ) → Z c C(µ) and ¯ s γµ PL d (12. ‡ For instance.138). We recall the RGE for C(µ). in the so-called vacuum saturation approximation. The RGE for the Wilson coefﬁcient C(µ) is discussed in Section 7.‡ Consequently.

(b) are found in an analogous manner and yield the contribution to the effective lagrangian: L2 pole =− 2 Xgs ¯ s γν T a γρ γµ PL γ ρ γ ν T a 16π 2 ε d ¯ s γ µ PL d .154) We are going to use this equation for evolving C(µ) from µ0 down to µlow at the leading order in QCD (i.155) 2 where X = G 2 MW λ2 C(µ)/4π 2 abbreviates the factor standing in front of the t F four-quark interaction in Leff . respectively. 12. At the leading order in QCD. 12. 12. 12. 12. dµ Z c (12. the only integral we need is dn q qρqσ ig ρσ + ﬁnite terms = (2π)n (q 2 − m 2 )(q 2 − m 2 )q 2 32π 2 ε s d 2 Xgs ¯ s γµ PL γρ γν T a 2ε 16π (12.20. (d) reads L1 = 2 × (−X )(−igs )2 (−i)i2 µε q q dn q [ ¯ s γµ PL (/ + m d )γν T a d ][ ¯ s γ ν T a (/ + m s )γ µ PL × 2 (2π)n (q 2 − m d )(q 2 − m 2 )q 2 s d] (12.202) to build the Green’s function 0|T s ¯ d s ¯ d |0 (observe that for L1 all contributions are distinguishable and see Fig. we need to ﬁnd the explicit O(αs ) contribution to the renormalization constant Z c .20(a).148) with the external ﬁelds after its insertion into (2.450 12 Standard electroweak theory where γc = −µ dZ c 1 .156) Substituting this result into the considered amplitude. By power counting. The factor of 2 is the combinatorial factor for the diagrams summarized by L1 . 12. Thus.20(c). one ﬁnds the pole part of L1 L1 pole =− d ¯ s γ ν γ ρ γ µ PL T a d (12. The SU (3) generators and gauge coupling are denoted by T a and gs .19).157) The pole parts of Fig. 2 Up to O(Q 2 /MW ). low (b) and the ‘negative partners’ of Fig. colour indices are contracted within brackets. Thus. we can verify that only the double / term in the numerator q can contribute to UV divergence. the renormalization constant Z c in the MS scheme is found from pole parts of one-loop diagrams shown in Fig. the contribution to the effective lagrangian of Fig. in the so-called leading logarithmic approximation).20(a). It can be traced back to the possibility of two different (but indistinguishable) contractions of the ﬁelds in the four-fermion operator in (12.e. (d) and the ‘negative partners’ of Fig. As usually.20(c).

7 Flavour changing neutral-current processes 451 Pole parts of Fig.5 on muon decay).12.160) together with the Fierz rearrangement formula (12.96). but we are allowed to use four-dimensional Dirac algebra so long as we are interested only in calculating one-loop renormalization constants in the MS scheme (see also the discussion in Section 12.98) for anticommuting ﬁelds. It holds in four dimensions only. ( f ) (and their ‘negative’ companions) are summarized by L3 pole =+ L4 pole 2 Xgs ¯ s γν T a γρ γµ PL 32π 2 ε 2 Xgs ¯ s γµ PL γρ γν T a =+ 32π 2 ε d ¯ s γ ν T a γ ρ γ µ PL ¯ s γ µ PL γ ρ γ ν T a d d d Compared with L1 and L2 .158) In order to simplify the Dirac algebra in the ﬁrst expression. we obtain the ﬁnal result 4 Lk k=1 pole =− 2 Xgs ¯ s γ µ PL 12π 2 ε d ¯ s γ µ PL d (12. we can get rid of the SU (3) generators T a by applying the identity (T a )αβ (T a )γ δ = 1 δαδ δγβ − 1 δαβ δγ δ 2 6 (12. 12. which gives pairs of indistinguishable contractions). The second term in (12. L3 and L4 have no combinatorial factors of 2 since two indistinguishable contractions appear when L3 (L4 ) is used in (2. we apply the identity (12. we require that the divergent . Summing up all contributions one ﬁnds for the pole part 4 Lk k=1 pole =− 2 Xgs 16π 2 ε ¯ s γµ γρ γν PL T a d ¯ s γ ν γ ρ γ µ PL T a d d d + ¯ s γν γρ γµ γ ρ γ ν PL T a T a d ¯ s γ µ PL − ¯ s γµ γρ γν PL T a d ¯ s γ µ γ ρ γ ν PL T a (12.202) (we recall once again that for L1 and L2 all contractions are distinguishable and the factor of 2 is necessary to match the combinatorics of the tree level diagrams described by the operator in (12. After all these simpliﬁcations.20(e).158) is simpliﬁed by using γν γρ γµ γ ρ γ ν = 4γµ + O(ε) and (T a T a )αβ = 4 δαβ . Now.159) Finally. After all that we ﬁnd 3 4 Lk k=1 pole = 2 Xgs 12 ¯ s γ µ T a PL 16π 2 ε d ¯ s γ µ T a PL d d − 16 ¯ s γ µ PL 3 d ¯ s γ µ PL (12.148).161) with colour indices contracted within brackets.

Finally. the charm quark can be .45)): µ dαs αs 2 = αs b1 + O(αs ) dµ π (12.452 12 Standard electroweak theory part of the one-loop diagrams we have calculated is cancelled by the counterterm 2 Leff = −X (Z c Z 2 − 1) ¯ s γ µ PL d ¯ s γ µ PL d (12. γc (µ) = αs (µ) 2 + O(αs ).166) 2 gs αs 4 2 + O(gs ) = 1 + + O(αs ). (12.162) which implies that 2 Zc Z2 − 1 = − 2 gs 4 + O(gs ) 12π 2 ε (12.167) Thus. the top quark is already absent. Below µb .168) where b1 is the coefﬁcient in one-loop RGE for the QCD coupling constant (see Chapters 6. π (12.165) The solution to the RGE (12. we substitute N f = 4 to b1 . we substitute N f = 5 to b1 when running the Wilson coefﬁcient from µ = µ0 to µ = µb ∼ m b .164) the renormalization constant Z c must be equal to Zc = 1 + Consequently. we have N f = 6. 4π 2 ε πε (12. Consequently. At this latter scale.169) The value of b1 = − 1 (11 − 2 N f ) depends on the number N f of quark ﬂavours 2 3 active in the considered effective theory.163) Since the quark wave-function renormalization constant is known to be (see Chapter 8) Z2 = 1 − 2 gs 4 + O(gs ) 6π 2 ε (12. In the effective theory introduced at the scale µ0 .153) is easily found to be C(µ) = exp ln µ ln µ0 d(ln µ ) γc (µ ) C(µ0 ). the b quark should be decoupled. 8 and (6. In the full Standard Model. we obtain (up to higher orders in αs ) C(µ) = exp = exp 1 π 1 b1 ln µ ln µ0 d(ln µ )αs (µ ) C(µ0 ) dαs αs (µ0 ) C(µ0 ) = αs αs (µ) −1/b1 αs (µ) αs (µ0 ) C(µ0 ) (12.

Equating the two amplitudes renormalized at the scale µ0 ∼ MW . changing them by. 12. a factor of 2 is still allowed. decoupled at µ = µc ∼ m c . αs ln(µb /µc ) and αs ln(µc /µlow ). An example of such a correction is shown in Fig. For instance. The ultimate expression for the Wilson coefﬁcient at scales µlow < µc is thus C(µlow ) = αs (µc ) αs (µlow ) 6/27 (12. On the effective theory side. i. next-to-leading contributions to the matching on the full theory side are given by two-loop diagrams like the one shown in Fig. µb and µc . we have resummed powers of αs ×(large logarithm) from all orders of the usual perturbation series. These scales have to be of the same order of magnitude as MW . Sample two-loop QCD corrections to the effective vertex. respectively.21. As mentioned earlier. xt ) + O(αs ). when µ0 in (12. Buras & Harlander (1990). At the next-to-leading order. Details of the next-to-leading QCD calculation are presented in Buchalla. one ﬁnds C(µ0 ) including the O(αs ) part of it.7 Flavour changing neutral-current processes 453 Fig. 12. Achieving this resummation has been the main reason for introducing the effective lagrangian (12.148). 12. αs ln(µb /µc ) and αs ln(µc /µlow ).12. too.170) automatically reduces its dependence on the arbitrariness in the choice of scales µ0 . .170) The expression we have obtained for C(µlow ) is perturbative in αs but exact in αs ln(µ0 /µb ).170) is replaced by 2µ0 . it requires calculating the O(αs ) part of C(µlow ) exactly in αs ln(µ0 /µb ). the leading order (LO) contribution to the αs (µb ) αs (µc ) 6/25 αs (µ0 ) αs (µb ) 6/23 ˜ S(xt . In other words. m b and m c .18. and to the running of the Wilson coefﬁcient down to the scale µlow .e. However. Running the Wilson coefﬁcient down to the low energy scale µlow at the next-to-leading order in QCD requires knowing the two-loop contribution to the anomalous dimension γc .20. In the MS-scheme. one has to evaluate ﬁnite parts of the diagrams presented in Fig. for example. summing all terms αs (αs ln(µ0 /µb ))n etc. Resummation of large logarithms can be systematically performed at higher orders in QCD. 12. one needs to include O(αs ) contributions to the matching between the full and effective theories at the scale µ0 . In order to achieve this. it is given by pole parts of two-loop corrections to the effective four-quark vertex. Including the O(αs ) part of C(µlow ) in (12.21.

173) As follows from (1. We recall that (1. the amplitude A0 is approximately real. On the other hand. then the parameters εK in (1.170) is found. The same phenomenon can be proven to happen at any order in perturbation theory. when the O(αs ) term is calculated explicitly.44).171) π Before the O(αs ) term in (12.273) and A0 is deﬁned in (1. In order to simplify further discussion. the effect of changing the scale by a factor of 2 can always be absorbed into unknown higher order corrections. Our approximate expression for εK now becomes proportional to Im M12 εK Im M12 exp(iπ/4) √ 2 m (12. Thus.273) C LO (µlow ) = εK Im(M12 A2 ) 0 exp(iπ/4). the imaginary part of M12 is equal to 2 2 ¯ ¯ Im M12 = Im K0 |HW |K0 + O(HW ) = − Im K0 |HW |K0 + O(HW ) (12. 2 the µ0 dependence of C(µlow ) becomes a O(αs ) effect (see. Buras. A0 is to a good approximation proportional to V12 V11 which is real in the standard parametrization (12. Choosing the proper order of magnitude for the renormalization scale is important for (asymptotic) convergence of the perturbation series. what matters in practical calculations are only orders of magnitude of renormalization scales. not their precise values.172) where M12 is the hermitean part of the weak hamiltonian.270) and ε in (1. √ 2 m|A0 |2 (12. m is introduced in (1. When this parametrization is used. Misiak.265) would be equal. On the other hand.44) of the CKM matrix. Note that ¯ if A0 was exactly real. This is ¯ because the dominant contribution to A0 originates from the tree level s → u ud ∗ weak transition. one ﬁnds that its µ0 dependence is precisely such that it cancels the µ0 dependence of the leading term. for example.270). Therefore. The renormalization-scale-dependence is always formally a higher order effect. we shall brieﬂy describe how the evaluated Wilson coefﬁcient is used to ﬁnd the C P-violating observable εK (deﬁned in Chapter 1). In the remainder of this subsection.261).174) . Consequently. M¨ nz & Pokorski (1995) for an extensive discussion of this point in the u case of the decay b → sγ ). we shall assume that we use the parametrization of the CKM matrix as in (12.454 12 Standard electroweak theory coefﬁcient C(µlow ) changes by αs (2µ0 ) 6/23 − 1 C LO (µlow ) αs (µ0 ) αs (µ0 ) 2 =− (ln 2) · C(µlow ) + O(αs ) (12. the above dependence on µ0 can serve as an estimate of the error we make by neglecting this very term.

Using (12. 12. xt ) = S(xc . because |λc | contributions are included and the unitarity of the CKM matrix is used. Namely η2 = [αs (µc )] 9 2 αs (µb ) αs (µc ) 6/25 αs (µ0 ) αs (µb ) 6/23 + O(αs ) (12.t λi λ j M12 (i j) (12. we can write it as a sum of contributions proportional to different products of CKM matrix elements Im M12 = Im i.175) The part proportional to Im λ2 in the above expression is given by the matrix t element of the effective four-quark interaction present in (12. one obtains a result which is usually written in the following form Im M12 = − 2 G 2 MW F 2 BK m K f K Im(λ2 )S(xc )η1 c 12π 2 2 + Im(λt )S(xt )η2 + 2 Im(λt λc )S(xc .179) The quantity η2 which parametrizes the short-distance QCD effects in the double top quark contribution is directly related to the calculated expression for the coefﬁcient C(µ).174) is generated by the box diagram in Fig. 0) = xc − ln xc + xt2 − 8xt + 4 3xt 2 ln xt − + O(xc ) 2 4(1 − xt ) 4(1 − xt ) and the µ-independent parameter BK is deﬁned by BK = BK (µ)[αs (µ)]−2/9 [1 + O(αs )] (12.17 and by QCD corrections to it.170). Similarly to (12. we ﬁnd Im M12 = − 2 G 2 MW F 2 Im(λ2 )C(µlow )BK (µlow )m K f K t 12π 2 + (contributions proportional to other products λi λ j ) (12.h.146). xt ) + S(0.174).176) with C(µlow ) given by (12.c. x) + S(0.178) where −3x 3 ln x x 3 − 11x 2 + 4x ˜ ˜ ˜ S(x) = S(x. When the other other contributions cannot be neglected. xt )η3 (12. of (12.s.12.177) (12.150) and (12. the |λt |. Although the double top-quark contribution is really the largest in Im M12 . 0) − 2 S(x. 0) − S(xc .148) between the kaon and antikaon states. 0) − S(xt .7 Flavour changing neutral-current processes 455 The matrix element in the r. j=u.180) . 0) = + 2(1 − x)3 4(1 − x)2 ˜ ˜ ˜ ˜ S(xc .

. ¯ 12.7 Show that (h 0 → γ γ ) ∼ (g 2 /MW )α 2 Mh .2 Show that a convenient measure of the C P violation in the K0 –K0 mixing is the parameter J = Im(Vub Vcd Vud Vcb ). Show that it can be expressed in terms of four moduli of the elements of the CKM matrix: 4J 2 = 4|Vub |2 |Vcb |2 |Vud |2 |Vcd |2 2 2 2 2 2 2 2 (12. due to a singularity in the phase space integration.e. 12.9 Calculate the box diagram for K0 –K0 mixing in the unitary gauge. even in the limit of zero fermion masses (i. with vacuum expectation values vn . 12. . G 0 . 12.8 Calculate the forward–backward and left–right asymmetries for the process e+ e− → Z 0 → f¯ f for s 1/2 ≈ M Z .) = A(G ± .1 Show that with N Higgs multiplets with weak isospin Tn2 = tn (tn +1).182) − (1 − |Vud | − |Vcd | − |Vcb | − |Vub | + |Vud | |Vcb | + |Vub | |Vcd |2 )2 12. with help of an effective theory and renormalization group evolution of Wilson coefﬁcients. . one gets the tree level relation: ρ≡ 2 MW 2 cos2 θW M Z = n 2 2 tn (tn + 1) − t3n vn 2 2 2 n t3n vn (12. i.6 Calculate the partial decay width for h 0 → W + W − . . Show that in the limit Mh MW the same result is obtained by calculating the decay width for h 0 → G + G − where G ± are the Goldstone bosons in the ’t Hooft–Feynman gauge. it is technically much more involved. . This is an exam± 0 ple of the ‘equivalence theorem’: A(WL . 12. 3 2 12.181) ¯ 12. Construct the effective lagrangian for this process. .456 12 Standard electroweak theory The calculation of η1 and η3 is performed with the use of the same technique as η2 . Problems 12. . However. for vanishing G ± f f coupling).4 Calculate the β-functions for the renormalization group evolution of the gauge couplings in the Standard Model.3 Check that the relation ρ = 1 follows from the SUV (2) symmetry of the Higgs potential after spontaneous symmetry breaking (like in the σ model).5 Derive RGE for Yukawa couplings in the Standard Model. Show that the equivalence theorem also holds for decays h 0 → W − W + → 4 f and h 0 → G + G − → 4 f (for MW < Mh < 2MW ). . . Z L . . n = 1. . N .) + O(MW /E).e.

although external with respect to the strong interaction gauge group. the conservation of the U (1) current corresponds to baryon number conservation whereas the conservation of the UA (1) current is a problem (the so-called UA (1) problem): it can be seen that the spontaneous breakdown of the SU (N ) × SU (N ) implies the same for the UA (1) but there is no good candidate for the corresponding Goldstone boson in the particle spectrum. Its existence can be traced back to the short-distance singularity structure of products of local operators. Bell & Jackiw 1969). as we shall see. To be speciﬁc let us consider QCD.1 Triangle diagram and different renormalization conditions Introduction Anomalies have already been mentioned in this book on several occasions.13 Chiral anomalies 13. The axial non-abelian currents couple to Goldstone bosons (pseudoscalar mesons) and the left-handed chiral currents couple to the intermediate vector bosons. apart from being invariant under gauge transformations in the colour space its lagrangian is also invariant under the global SU (N ) × SU (N ) × U (1) × UA (1) chiral group of transformations acting in the ﬂavour space. they are gauge currents of the weak interaction gauge group. representation of the gauge group and for N > 2 to a complex representation of the chiral group (see Appendix E). acquire important dynamical sense. this is irrelevant for the discussion of anomalies. 457 . real. It is clear from the preceding discussion that Green’s functions involving chiral or axial currents are of direct physical interest. Moreover. We know from Chapters 9 and 10 that the Noether currents corresponding to the global ﬂavour symmetry.e. i. In the framework of the renormalization programme in QCD the ﬁniteness of the matrix elements of † We take quarks as massless. As discussed in Chapter 9.† Fermions belong to a vector-like.e. For massive quarks the chiral group is softly broken but. In this chapter we systematically discuss the fermion anomaly in (3 + 1)-dimensional quantum ﬁeld theory (Adler 1969. i.

The triangle diagram is anomalous because it is impossible to add a polynomial so that all the classical symmetries are simultaneously preserved.1. the requirement of unitarity in each channel makes any one-loop amplitude well-deﬁned up to a polynomial in the external momenta. We consider the U (1) and UA (1) currents of one-fermion free-ﬁeld theory. in general. To complete the introductory remarks it is worth mentioning that anomalies are also of basic importance for most theories going beyond the Standard Model. In other words an anomaly means a deviation of the renormalized theory from the canonical behaviour. The basic structures to study are the triangle diagrams depicted in Fig. being a generalization of Gauss’ theorem. However. In the next section we will see explicitly that the anomaly constitutes a breakdown of gauge invariance. This freedom is used in the renormalization procedure when we get rid of all inﬁnite pieces. must be separately veriﬁed. In this way anomalies can be seen experimentally. All internal quantum numbers of fermions are suppressed in this calculation. in perturbation theory. Anomalies originate in the short-distance singularities of the free fermion propagator which generate renormalization ambiguities in the Green’s functions containing chiral currents. In particular the problem of chiral anomalies in more than four space-time dimensions and of gravitational anomalies as well as new techniques for their calculation . as in the standard SU (2) × U (1) model of weak interactions. 13.458 13 Chiral anomalies the T -ordered products of such local composite operators as chiral currents is an additional requirement which. Relations obtained in this way between various renormalized Green’s functions involving currents can then be summarized in operator language: certain currents have anomalous divergences. Anomalies reﬂect the absence of a renormalization procedure such that Green’s functions respect all the symmetries of the lagrangian. these combinations must be anomaly-free. Historically. Ward identities. also provide a beautiful manifestation of the UV–IR interrelation in the theory: short-distance singularities manifest themselves in low energy theorems. It turns out that in four space-time dimensions the Green’s function which cannot be renormalized preserving all the classical symmetries of the lagrangian is the one containing two vector and one axial currents or three chiral currents. the anomaly spoils the conservation of the currents coupled to gauge bosons and gauge theories with gauge ﬁelds coupled to non-conserved currents are inconsistent. the main tool for studying anomalies is the Ward identities in perturbation theory. Otherwise. They may require an additional renormalization. As we know. In the following we calculate explicitly the triangle amplitude and study its Ward identities. if gauge ﬁelds are coupled to linear combinations of vector and axial-vector ﬂavour currents. All the necessary generalizations can be easily introduced at the end. This is an important constraint on any theory.

q) =0 (13. and two vector currents jµ (x).1.13.1 Triangle diagrams 459 Fig. jµ (y. t). they read −( p + q)µ 5 µνδ ( p. (y.1 and the fact that the canonical equal5 5 5 time commutators [ j0 (x. q) =0 (13.3) . t)] and [ j0 (x. 13. q) = pδ 5 µνδ ( p. t)]. t). AlvarezGaum´ & Witten (1983). or of one 5 axial vector current jµ (x). For the time being we treat them as abelian currents of the one-fermion free-ﬁeld theory which. at the classical level. have been vigorously studied (for example. [ j0 (x. Wu & Zee (1984)). t). for example.1) for m = 0 or. say. q) 5 µνδ ( p. t) vanish one can derive the standard Ward identities for the matrix elements L L 5 0|T jµ (z) jν (y) jδL (x)|0 and 0|T jµ (z) jν (y) jδ (x)|0 . jµ . In momentum space. We shall come back to some e of these subjects later on. q) = 2m 5 νδ ( p. are conserved. q) = qν 5 µνδ ( p. Calculation of the triangle amplitude L We consider a vacuum matrix element of three chiral currents jµ (x). q) (13.2) = qν 5 µνδ ( p. Frampton & Kephart (1983). jµ (y. Zumino. If we allow for a Dirac mass term in the lagrangian then the axial-vector current is partially conserved 5 L ∂ µ jµ (x) = −2∂ µ jµ (x) = 2mi ¯ γ5 Using the techniques of Section 10. for m = 0 −( p + q)µ and pδ 5 µνδ ( p.

13.5) = Fµνδ ( p.s. It depends on our choice of the renormalization constraints. the anomaly is massindependent. q) (13. p) (13. 5 Fourier transform of 0|T jµ (z) jν (y) jδ (x)|0 5 4 µνδ ( p. We shall ﬁrst consider the diagram on the r. a contribution from the triangle diagrams shown in Fig. respectively.460 13 Chiral anomalies 5 µνδ ( p. dictated by the physics of the problem. anticommuting γµ γ5 = −γ5 γµ and reversing the order of factors Tr[γα γν γβ γδ γγ γµ γ5 ] = Tr[γµ γγ γδ γβ γν γα γ5 ] (13.6) We take here a non-zero fermion mass but. For clarity of the following discussion it is useful to ﬁrst write down the most 5 general form of the amplitude µνδ ( p. The amplitude corresponding to this diagram reads Fµνδ ( p.1)–(13.1. q) is the Fourier transform of 0|T P(z) jν (y) jδ (x)|0 with P = ¯ γ5 . p ↔ q and δ ↔ ν. too.2) and (13.8) We also notice that the sign of the mass factor is irrelevant because the only nonvanishing mass-dependent terms in the trace are proportional to m 2 . q) = 2Fµνδ ( p. Thus the result we will obtain is valid for the m = 0 case. if currents are coupled to gauge ﬁelds with some coupling constant g. In a quantized theory. 13.3).1) or (13. q) + Fµδν (q.3). q) where p and q are momenta conjugate to x and y.h. as we shall see. at the one-loop level. the current matrix elements get. p) (13. q) = (−1) and d4 k i i i Tr γν γδ γ µ γ5 (2π)4 /+/ −m /−m /− p−m k q k k / 5 µνδ ( p. As we shall see. We also keep in mind that. q) = Fµδν (q.7) This follows from (13. q) is superﬁcially linearly divergent. q)(2π) δ(r is the + p + q) (13.4) = 5 d4 x d4 y d4 z exp[−i( px + qy + r z)] 0|T jµ (z) jν (y) jδ (x)|0 5 and νδ ( p. By power counting the amplitude Fµνδ ( p. The ﬁrst observation worth making is the symmetry property Fµνδ ( p. an extra factor (−ig) appears in each such vertex. these contributions destroy the naive Ward identities (13. q) consistent with the . of Fig.1 and from crossed diagrams with the external lines interchanged. It is impossible to renormalize this amplitude preserving simultaneously all three identities (13. where the anomaly is actually placed.5) by using the change of variables k = −k.

q) pδ εµναβ p α q β + B4 ( p. Given this identity. q)qδ εµναβ p α q β (13. One is to regularize the amplitude (13. Subtractions are. Imagine. satisfying different renormalization constraints. q) = −A2 (q.13. p) implies that A1 ( p. q) 5 µνδ . these functions can be expressed in terms of Bs when deﬁnite renormalization conditions are imposed. that we insist on the vector current conservation qν for the renormalized vertex 5 µνδ ( p. q)εµνδα q α + B1 ( p. Thus we insist on vector 5 current conservation for the renormalized vertex µνδ 5 The Pauli–Villars regularization of the amplitude µνδ ( p. p) follows so that only one of the equations (13. q) = A1 ( p. q) = −B3 (q. The requirement of Bose symmetry under interchange 5 5 of the two vector currents µνδ ( p. q) =0 (13. p). B1 ( p. q) = −A2 (q. Let us follow this approach choosing to work with the Pauli–Villars regularization (Itzykson & Zuber 1980) of the amplitude (13. Dimensional regularization in this case has some problems with ambiguities in the extension of γ5 to n dimensions (for a review of this matter see.10) Then we must have (13. B2 ( p. q) pν εµδαβ p α q β + B2 ( p. However. p).11) determine the divergent amplitudes Ai in terms of ﬁnite Bi s.11) A1 + p · q B1 + q 2 B2 = 0 A2 + p 2 B3 + p · q B4 = 0 and indeed the conditions (13. for instance. q) = µδν (q.11) is independent. q) = −B4 (q. The terms pν ενδαβ p α q β and qµ ενδαβ p α q β are not linearly independent. It reads 5 µνδ ( p. q)qν εµδαβ p α q β + B3 ( p. for instance.1 Triangle diagrams 461 requirements of parity and Lorentz invariance.9) where As and Bs are scalar functions. q) consists of considering it to be a function of the mass of the fermion circulating in the loop: . by additional ﬁnite subtractions one can always transform it into another anomalous Ward identity. q)εµνδα p α + A2 ( p. whereas the functions Bi are ﬁnite.5) by some regularization prescription and to calculate Ward identities for the 5 regularized µνδ . = pδ 5 µνδ ( p. Let us choose as our renormalization conditions the relations (13. Ovrut (1983) and references therein). There are many ways to calculate anomalies. From Bose symmetry A1 ( p.10). p). If the regularization procedure satisﬁes the constraints which we want to impose as our renormalization constraints then in the limit of no regulator we obtain the anomalous Ward identity we are looking for. in fact only logarithmically divergent. Using again power counting arguments it is clear that only the functions A1 and A2 are divergent. necessary for the functions A1 and A2 .5). in principle.

q. M). m. m.13) (13. This can easily be checked by expanding the integrand in the integrals over d4 k in powers of 1/k 2 : the linear 5 5 divergence of µνδ ( p. q. M) is ﬁnite. q. M) (13. q) 13 Chiral anomalies 5 = µνδ ( p. 4 M) dk 1 1 γν γδ γ5 − (m → M) Tr 4 (2π) /+/ −m /−m k q k d4 k 1 1 + 2i Tr γν γδ γ5 − (m → M) (2π)4 /−m /− p−m k k / d4 k 1 1 1 γν γδ 2mγ5 − (m → M) + 2i Tr 4 (2π) /+/ −m /−m /− p−m k q k k / (13.15) All the integrals are ﬁnite. M) = 2i d4 k 1 1 Tr γν γµ γ5 − (m → M) 4 (2π) /+/ −m /− p−m k q k / d4 k 1 1 − 2i γν γµ γ5 − (m → M) Tr 4 (2π ) /+/ −m /−m k q k (13. q. m) is cancelled by µνδ ( p. At the end one lets the mass M go to inﬁnity and the ﬁnal answer is ﬁnite with no additional subtractions.14) pδ 5 µνδ ( p. Using p = / (/ − m) − (/ − p − m) this becomes k k / pδ 5 µνδ ( p.13).16) The last term vanishes for the same reason as above. q. M) = q ν 5 µνδ ( p. m. q. q. q. The ﬁrst term also vanishes because we can shift the momentum integration variable to k = k − p and . q. q. M). m. using the relation p + / = (/ + / − m) − (/ + p − m) we can write down / q q k k / the following equation: −( p + q)µ = 2i 5 µνδ ( p. q.462 5 µνδ ( p.12) 5 The regularized amplitude µνδ ( p. m) − 5 µνδ ( p. M) = 2m 5 νδ ( p. m). q. M) = 5 µνδ ( p. m. The last term gives 5 just the Ward identity (13. m) − 2M 5 νδ ( p. q. The ﬁrst two terms are second rank pseudotensors depending on only one momentum variable and hence vanish. Consider now p δ µνδ ( p. A regulated amplitude is deﬁned as the difference between the given amplitude and the same amplitude taken at some other value M of the mass (M is the regulator) 5 µνδ ( p. m. m. M) = 0 Indeed. q. M) (13. q. m. because it satisﬁes our renormalization conditions ﬁxing Ai in terms of Bi First we show that the regularized amplitude satisﬁes the ‘normal’ Ward identities −( p + q)µ 5 µνδ ( p.

1 Triangle diagrams 463 again apply the same argument. M) = +(i/2π 2 )ενδβα pβ qα and the ﬁnal form of the axial Ward identity consistent with our renormalization constraints reads −( p + q)µ pδ 5 µνδ 5 µνδ ( p. m) = 2i 1 1 d4 k 1 Tr γν γδ γ5 4 (2π) /+/ −m /−m /− p−m k q k k / = 2 × 4miενδβα pβ qα F( p.1).21) The anomalous term is m-independent. q. m) = i (13. m. our regularization procedure is consistent with our renormalization requirements (13. † Note the difference between a shift of the integration variable by a constant and a change of variables k → −k used in proving (13. If it were linearly divergent the translation of the integration variable might involve a non-vanishing surface term which would spoil the argument† (see Problem 13.13) in that limit. m) = − 1 16π 2 1 1−x dx 0 0 dy [q 2 y(1 − y) + p 2 (1 − x)x + 2 pq x y − m 2 ]−1 (13.20) It is now straightforward to calculate the limit M→∞ lim 2M 5νδ ( p.19) (13. This we leave as an exercise for the reader. This shift of the integration variable is legitimate because the integral is ﬁnite.17) d4 k 1 1 1 4 (k + q)2 − m 2 k 2 − m 2 ( p − k)2 − m 2 (2π) where F( p.13). The νδ ( p. q. The anomaly results from the non-vanishing of the regulator term in 5 (13.16) and integrating over d4 k one gets F( p. q.21) is to calculate explicitly the ﬁnite functions Bi starting with the unregularized amplitude (13. q. M) = 0 Thus.7). q.5) and then to use renormalization constraints (13. . q. m) vertex can be calculated explicitly since the Feynman integral is convergent: 5νδ ( p. Another way to obtain the anomalous Ward identity (13.18) and the factor 2 in front of the integral in (13. Our last step is to take the limit M → ∞ in the Ward identity (13. q. m) 5 µνδ = 2m 5 νδ ( p. m) (13. In a quite analogous way we can show that qν 5 µνδ ( p.11).13. q. m) − (i/2π 2 )ενδβα p β q α = qν =0 (13. q. Using the Feynman parametrization (B.10).17) takes account of the crossed diagram.

1.23) (13. q) (13.22) as the renormalization condition one gets pδ qν 5 µνδ ( p. m) 5 µνδ ( p.25) One cannot satisfy simultaneously all three Ward identities (13.h.21) and (13.h.3). q. 13.1 with the left-handed currents at each vertex and in the limit m = 0. m) (13.2) and (13.22) as our renormalization constraint. q.21). p) from the Bose symmetry under interchange of the two vector currents discussed earlier. m) = 2m 5 νδ ( p. q) (which we denote 5 by ˜ µνδ ( p. q) to distinguish it from the AVV amplitude considered earlier) under interchange of any pair of vertices. q) formally satisﬁes the following relation L µνδ ( p. This will be illustrated in the next section. q) − 1 2 5 µνδ ( p.464 13 Chiral anomalies Different renormalization constraints for the triangle amplitude The renormalization conditions (13. our previous renormalization conditions are inappropriate since now for the Green’s function of three identical currents we 5 must require Bose symmetry of the renormalized µνδ ( p. of Fig.22). L Thus formally the anomaly of µνδ ( p. instead. Whether we should impose them or not depends on the physics of the problem. q) = 1 2 µνδ ( p. of Fig. However.21). q) is the three-vector current amplitude which has no anomaly. q. m) (13. q) = −a2 (q. Adding a ﬁnite polynomial . It may happen that we must.24) = −(i/4π 2 )εµναβ p α q β = (i/4π 2 )εµδαβ p α q β (13. that the latter is satisﬁed for a1 − a2 = i/2π 2 Since a1 ( p. The divergent amplitudes A1 and A2 are again uniquely speciﬁed by the condition (13. with an additional ﬁnite subtraction for the A1 and A2 a1 εµνδα p α + a2 εµνδα q α we see.s. We can get the right answer by applying a ﬁnite renormalization to the previous result (13. Finally let us consider the triangle on the l.26) where µνδ ( p. with (13. 13.10) lead to the result (13. At the level of the divergent L Feynman integral the bare amplitude µνδ ( p. by comparing (13. q.22). we conclude that a1 = −a2 = i/4π 2 Thus. Indeed. insist on the relation −( p + q)µ 5 µνδ ( p.s. q) can also be calculated from the triangle on the r.

q) so we combine (13.27) Insisting on symmetry under the interchange −( p + q) ↔ p. Fµν = ∂µ Aν − ∂ν Aµ . we consider two models given by the lagrangians / L = Z 2 ¯ iD and R L = Z 2 ¯ R i/ ∂ R L / + Z 2 ¯ L iD L − 1 Z 3 Fµν F µν 4 − 1 Z 3 Fµν F µν 4 (13. q) = (−i/2π 2 + a1 − a2 )ενδαβ p α q β δ ˜5 α β p µνδ ( p.28) (13.29) (13.23). to the amplitudes A1 and A2 we get (for m = 0) 5 −( p + q)µ ˜ µνδ ( p. µ ↔ δ and p ↔ q.30) Thus.26). Important comments In the next section we will discuss in more detail the anomaly problem in the context of speciﬁc physical theories. q) = −(i/6π 2 )ενδαβ p α q β 5 p δ ˜ µνδ ( p. Here. to illustrate the meaning of arbitrariness in renormalization conditions for the triangle diagram. (13. q) = a1 εµδαβ p q (13. the same triangle amplitude leads to three different sets of anomalous Ward identities depending on the imposed renormalization conditions.1 Triangle diagrams 465 term. q) = −(i/6π 2 )εµναβ p α q β ν ˜5 2 α β q µνδ ( p. The latter must be chosen according to the physical requirements of the problem under consideration. L We recall that our present interest is in the amplitude µνδ ( p. q) = a2 εµναβ p q ν ˜5 α β q µνδ ( p.31) (13. q) = +(i/6π )εµδαβ p q (13.30) with (13.13.32) respectively. where Dµ = ∂µ + ig Aµ . δ ↔ ν one gets the following equations −i/2π 2 + a1 − a2 = a2 a1 = −a2 Therefore a1 = −a2 = i/6π 2 and 5 −( p + q)µ ˜ µνδ ( p.

at this level Z 5 = 1. conserved in the Noether sense. for instance vector and axial-vector currents in the theory (13. but Z 5 is not ﬁnite (we will come back to this point shortly). is not coupled to any gauge ﬁeld.2. This conclusion is no longer valid in the presence of the anomaly. 13. as we know from the previous subsection. Thus.36) (see Section 7. We deﬁne the bare currents. Considering. Thus for the triangle diagram involving 5 jµ and two vector currents we must impose the renormalization conditions (13. for example. and the lagrangians are written in terms of renormalized quantities.10) which lead to the anomalous Ward identity (13. Gauge invariance of the theory requires the vector current conservation. In operator language this corresponds to an anomaly in the divergence of the axial-vector current 5 ∂ µ jµ (x) = (g 2 /16π 2 )εµνρδ Fµν (x)Fρδ (x) (13.466 13 Chiral anomalies Fig. at the one-loop level the current matrix elements are ﬁnite after imposing certain renormalization conditions.1 we have proved that a conserved or partially conserved current satisfying normal Ward identities is not renormalized. In Section 10. which in terms of the renormalized ﬁelds are as follows: 5 jµ (x) = Z 2 ¯ (x)γµ γ5 (x) (13.31).34) A comment is in order here on the renormalization of currents in presence of the axial anomaly.35) can be veriﬁed using the Feynman rules for the operator insertion εµνρδ Fµν (x)Fρδ (x) = 4ε µνρδ ∂µ Aν (x)∂ρ Aδ (x) (13.6). (13. The current is still 5 5 multiplicatively renormalized R jµ = Z 5 jµ . the Green’s function 0|T ∂µ Aν (z) . However.21).35) Eq. The vector current is coupled to the electromagnetic ﬁeld and the axial-vector current.33) and jµ (x) = Z 2 ¯ (x)γµ (x) (13. The ﬁrst theory is just QED of a single fermion ﬁeld. without additional counterterms.

Using relations (13.11) of Ward identities. 13. as usual.3 and in the lowest non-vanishing order in g 2 the anomaly is given by the diagram in Fig.1 we conclude that Z 5 is not ﬁnite as the operator Fµν F µν does not have ﬁnite matrix elements: e. Our second example.2 (propagators are. is a theory in which only the left-handed chiral current couples to the gauge ﬁelds.3. Closely following the ˜ procedure of Section 10.21)† (with m = 0) of the divergence 5 ∂ µ 0|T jµ jν jδ |0 is given by the Feynman diagram in Fig. Therefore. ∂ρ Aδ (z)Aα (y)Aβ (x)|0 in momentum space one gets the Feynman diagram shown in Fig. 13.35) follows if we recall the general form (10. the diagram in Fig. 13.1 Triangle diagrams 467 Fig. in order g 2 the Fourier transform (13. For instance.32).3) can be used to study the renormalization properties of the axial-vector current.13. 13. .35) we can calculate perturbatively other anomalous Ward identities involving axial-vector 5 current jµ (x).4. Fig. Using (13.4. 13.30) we now get † Now with (−ig) at each vector vertex included.4 is logarithmically divergent. The triangle diagram with gauge source currents at each vertex should be evaluated imposing symmetry under interchange of any pair of indices.g. the lagrangian (13.26) and (13. The anomalous Ward identity for 5 (derive ¯ it using the path integral technique of Sections 10.2 and (13. removed from the external lines). the derivative of the Green’s function 5µ = ¯ 0|T (x) j 5µ (0) ¯ (y)|0 is anomalous because the latter gets a contribution from the diagram in Fig.2 and 13. 13. 13.

the scale invariance anomaly discussed in Chapter 7 does obtain corrections because renormalization violates scale invariance.) The Adler–Bardeen theorem also holds in non-abelian gauge theory but it may be regularization-scheme-dependent (Bardeen 1972). i. (13.37) Because of the anomaly the gauge source current is no longer conserved and the theory is not gauge invariant. all the triangle diagram anomalies are absent. (On the contrary. λb . does not interfere with chiral symmetry.35) for the bare current we have used the result obtained in the lowest order in g 2 in perturbation theory. .e. λc }]. For the vector abelian theory considered here this has been proved to all orders in perturbation theory (Adler & Bardeen 1969). One believes that the anomalies are not modiﬁed by higher order corrections. In gauge theories with chiral gauge symmetry the anomalies must be absent for a consistent theory. 1 2 3 2 where the trace corresponds to summing over different fermions in the loop and the anticommutator is due to the symmetrization over the two diagrams in Fig.468 13 Chiral anomalies the anomaly in the chiral current L ˜ ∂ µ jµ (x) = −(g 2 /48π 2 )Fµν F µν (13. λb . and only if.5 The necessary modiﬁcations of operator equations. Let the currents in the vertices have coupling matrices λa . Our last remark in this section is concerned with generalizing the triangle diagram calculation to the case in which several species of fermion ﬁelds which have some internal degrees of freedom are present. of photon insertions in the triangle diagram. Equivalently one can say that the gauge ﬁeld has been treated as a classical background ﬁeld. 13. i. λc . For the triangle diagram we get then an extra factor 1 Tr[λa {λb .31). According to the Adler–Bardeen theorem generalized to the non-abelian case a theory is completely free of anomalies if.e. particularly in supersymmetric theories.35). The proof is technical but its main point can be summarized as follows: the necessity of renormalization of the higher order corrections in g 2 . so the chiral anomaly should arise only from the lowest order fermion loops. Writing the operator equation (13. like (13. λc may transform non-trivially under different internal 1 2 3 symmetry groups and be block-diagonal in the remaining internal symmetry spaces. The currents may be coupled to gauge ﬁelds or 1 2 3 not and the matrices λa . will be discussed later.35) is exact to any order in g 2 for the bare current in the regularized theory or for the renormalized operators (see also the next section) with g and Fµν being the renormalized quantities deﬁned by the lagrangian (13.

this renormalized axial-vector current does not generate chiral transformations.31) of spinor electrodynamics is invariant both under U (1) and axial UA (1) transformations. it turns out that chiral invariance remains a symmetry of the quantum theory based on the lagrangian (13. the triangle diagram amplitude turns out to be ﬁnite even in the limit of no cut-off but the higher order corrections induce inﬁnities in the axial-current matrix elements. Following Adler (1970) we ﬁrst deﬁne a new bare .13.1).2 Some physical consequences of the chiral anomalies Chiral invariance in spinor electrodynamics The lagrangian (13.35) in terms of the renormalized operators 5 ˜ ∂ µ R jµ (x) = (g 2 /8π 2 )[Fµν (x) F µν (x)]R (13.35). Nevertheless.39) However. The axial charge is not conserved.38) we may write (13. 13.31) and it is possible to construct the corresponding charge Q 5 . in quantum theory if we want to preserve vector gauge invariance an anomaly appears in the divergence of the axial-vector current. Not all of these can be absorbed in the renormalization constants present in the lagrangian (13. it does not commute with the photon ﬁeld and does not have the correct equal-time commutation relations with the renormalized spinor ﬁeld.31). Therefore due to the anomaly the axial-vector current is renormalized: R 5 jµ 5 = Z 5 jµ and we must re-express (13. 13. If we introduce a cut-off which preserves photon gauge invariance then to any order in g 2 the divergence of the unrenormalized axial-vector current is given by (13.5. due to the presence of a dimension four operator ˜ (Fµν F µν )R in the divergence of the current (see Section 10. As discussed in the previous section. However. Deﬁning 5 ˜ ˜ (g 2 /8π 2 )[Fµν (x) F µν (x)]R = (g 2 /8π 2 )Fµν (x) F µν (x) + (Z 5 − 1)∂ µ jµ (x) (13.35) in terms of the renormalized operators.2 Physical consequences of the chiral anomalies 469 Fig.

is conserved S ∂ µ j5µ = 0 (13. But the associate charge Q 5 (t) = S d3 x j50 (x.35).40) which. Under the gauge transformation δ (x) = −ig (x) (x) δ Aµ (x) = ∂µ (x) the symmetry current transforms as follows: S ˜ ˜ δ j5µ = (g 2 /4π 2 )∂ ν (x) Fµν (x) = (g 2 /4π 2 )∂ ν ( (x) Fµν (x)) (13. (x.470 13 Chiral anomalies current: symmetry current S 5 ˜ j5µ (x) = jµ (x) − (g 2 /4π 2 )Aν (x) Fµν (x) (13. t)] = −γ5 (x) it commutes with the photon ﬁeld and it is gauge-invariant. Let us check the last property. calculated by use of (13. compare (13. . its commutator with the canonical commutation relations.42) (x).41) time-independent.21) and (13.3) where not all ﬁnite gauge transformations can be reached by iterating inﬁnitesimal transformations.46) Gauge invariance of Q 5 in spinor electrodynamics follows from the vanishing ˜ of (x) F0k (x) at spatial inﬁnity. The current j5µ is conserved but it is explicitly gauge-dependent due to the presence of the gauge ﬁeld in its deﬁnition and therefore is not an observable operator.43) is from (13.45) Therefore δ Q 5 (t) = S d3 x δ j50 (x) ∼ ˜ d3 x ∂ ν ( (x) F0ν (x)) = ˜ d3 x ∂ k ( (x) F0k (x)) = 0 (13. The situation is different in non-abelian gauge theories (see Section 8. t) (13.41) Its existence is directly related to our discussion of different renormalization conditions of the triangle diagram.44) (13. We conclude that in spite of the presence of the axial anomaly chiral symmetry remains a good symmetry in massless spinor electrodynamics. This current is a S ﬁnite operator (Bardeen 1974) and does not need renormalization.22). is [Q 5 (t). from (13.

of (13.h. Next.68) and the equations of motion for the gauge ﬁelds Aµ (x).47) is given by the results of the previous section with the triangle diagram speciﬁed as shown in Fig.188).48) where O(k1 .49) plus terms of at least third order in ki from the second diagram: the factor k1 k2 comes from the gauge-invariant coupling (through the ﬁeld strength tensor Fµν ) of external photons and an extra factor (k1 +k2 ) comes from the anomalous divergence term. In this limit there is a pion pole contribution π to the l.196)) q 2 → p2 lim 53 d4 x exp(−iq x) γ1 (k1 )γ2 (k2 )|T S jµ (x)|0 = f π ( pµ / p 2 ) γ1 γ2 |S|π 0 + O(k1 . We now take the divergence of (13.194).47) we have used (2.6.47) 53 ν δ × exp(+ik2 z) exp(−iq x) 0|T jµ (x) jν (y) jδ (z)|0 ε1 (k1 )ε2 (k2 ) In (13. Thus. (2. (2. k2 ) α = (i/2π 2 ) Tr[ 1 σ 3 Q 2 ]ενδβα k1 k2 2 β (13.1) acquire dynamical sense through their coupling to pions. The divergence of the r. for µνδ deﬁned by (13. see (9.47) (see. for example.47). In this case. k2 ) (13. we use the PCAC (partial conservation of the axial-vector current) approximation relating physical quantities to the results obtained in the exact symmetry limit p 2 = m 2 = 0. The effect of the last diagram is only the coupling-constant renormalization. The fermion lines are quark lines.50) . With u and d quarks of three colours we get Tr[ 1 σ 3 Q 2 ] = 3 × 1 ( 4 − 1 )e2 = 1 e2 2 2 9 9 2 (13. 13. the Pauli matrix σ 3 is the neutral axial-vector coupling matrix and Q is the diagonal quark electric charge 5 matrix. the axial-vector ﬂavour currents of the SU (2)× SU (2) symmetric QCD lagrangian (9.65) 5a 0| jµ (x)|π b = i f π pµ δ ab exp(−i px) where pµ is the pion four-momentum. To calculate the π 0 → 2γ decay rate we 5a consider the amplitude ( jµ (x) is the renormalized axial isospin current) 53 d4 x exp(−iq x) γ1 (k1 )γ2 (k2 )|T S jµ (x)|0 = d4 x d4 y d4 z exp(+ik1 y) (13.195) and (2. (2. k2 ) has no pole in p 2 .s.2 Physical consequences of the chiral anomalies 471 γ π 0 → 2γ The relevance of the chiral anomaly to the decay π 0 → 2γ relies on our identiﬁcation of pions as the Goldstone bosons of the spontaneously broken chiral symmetry of QCD (see Chapter 9).h. of (13.13.4) we get the following: pµ 5 µνδ (k1 .s.

47) and (13. 13. of course. k2 . 13.7. respectively.51) are at least second and third order in momenta.53) .7. Therefore. k2 ) in (13.48) γ1 γ2 |S|π 0 = 1 e2 β iε k k α εν εδ (2π )4 ∂( p − k1 − k2 ) + O(k1 .48) and O(k1 k2 .6. p) 2 νδβα 1 2 1 2 f π 4π (13. From Lorentz and parity invariance α ν δ γ1 γ2 |S|π 0 = f ( p 2 )iενδβα k1 k2 ε1 ε2 (2π )4 δ( p − k1 − k2 ) β (13. from (13. Fig. 13.51) Our calculation of the divergence of (13. equivalent to calculating 5a perturbatively the matrix element γ1 γ2 |T S∂ µ jµ (x)|0 with the divergence given by 5a ˜ ∂ µ jµ (x) = (e2 /8π 2 ) 1 Tr[ 1 σ a {Q.52) The contributing diagrams are those shown in Fig. Q}]Fµν F µν 2 2 (13.47) is. p) in (13.472 13 Chiral anomalies Fig. Terms O(k1 .

A derivation of the equation D µ jµ = 0 in a quantized gauge-invariant theory is given in the next section.63 eV 2 64π 3 f π (Bose statistics implies integration over a 2π solid angle only).8.10). 13. We make two important observations. We study now the anomaly of the UA (1) current.13.ε2 d 3 k 1 d3 k 2 β α ν δ | f (0)ενδβα k1 k2 ε1 ε2 |2 (2π )4 δ( p − k1 − k2 ) 2π 6 4k10 k20 (13. 13. Secondly. Comparing (13.5 eV). The UA (1) current is a singlet under the colour group and the a relevant triangle diagram is shown in Fig.7 ± 0. 5 the divergence −( p + q) µνδ ( p. by analogy with (13.51) and the last equation we conclude that f (0) = and the rate for π 0 → 2γ is = = 1 2m π 2 1 fπ e2 4π 2 = α fπ π (13.1): chiral SU (n) × SU (n) invariance and invariance under abelian vector U (1) and axial-vector UA (1) transformation.21). In the absence of the anomaly one would get f (0) = 0 and PCAC could not be reconciled with the π 0 → 2γ decay rate. UA (1) problem Global symmetries of the QCD lagrangian with massless quarks are (see Section 9. For the triangle diagram 5 a we effectively get the condition ∂ ν 0|T jµ jν jδb |0 = 0 and therefore we must impose the renormalization constraints (13.2 Physical consequences of the chiral anomalies 473 Fig.55) is entirely given by the axial anomaly. Thus. The result (13. The experimental value is Exp = (7.8. We must have D µ jµ = 0 to a maintain gauge invariance. q) of the Fourier transform of the matrix element .55) α m3 π = 7. Chiral anomaly for the axial U (1) current in QCD. the agreement with the experimental value relies on the existence of three colour states of quarks.54) ε1 .

As in the ﬁrst subsection of this section we observe that the very existence of the anomaly does not destroy the UA (1) symmetry since the conserved current S j5µ = j5µ − 2n K µ . The operator equation (13. due to the self-coupling of the non-abelian gauge ﬁelds. µν 2 A new feature of (13. ˜ ∂ µ K µ = −(1/16π 2 ) Tr[G µν G µν ] and the conserved charge Q5 = S d3 x j50 .56) where n is the number of quark ﬂavours.56) is 5 ˜ ∂ µ jµ (x) = −(n/8π 2 ) Tr[G µν G µν ] (13. As we know from Chapter 9 the massless QCD lagrangian has the chiral (U (1) ∓ UA (1)) symmetry. However. Aviv & Zee (1972). The resolution of the problem is in the existence of the anomaly (13.111). there is no good candidate for the UA (1) Goldstone boson in the particle spectrum.57). All the anomalous Ward identities follow from the operator equation (13.474 5 a 0|T jµ jν jδb |0 reads 13 Chiral anomalies −( p + q)µ 5 µνδ ( p. q) 1 = (ig 2 /2π 2 )n Tr[ 1 λa 2 λb ]ενδβα p α q β 2 (13. square diagrams also contribute to the anomaly for the divergence of the axial current. As with the SU (n) × SU (n) it does not seem to be the symmetry of the hadron spectra so UA (1) must be broken (U (1) manifests itself in baryon number conservation).57) is that. For an explicit discussion of the square diagrams in the spirit of the previous section see.57) of the axial-vector current is crucial for resolving the so-called UA (1) problem in QCD. see (1. The only colour-invariant operator which for the triangle diagram gives the result (13.35) can also be easily generalized to the present case if we remember that UA (1) current is a colour singlet. for instance.58) Thus. the anomalous divergence of the UA (1) current involves not only terms quadratic in gauge ﬁelds but also terms of the third order in gauge ﬁelds 5 ∂ µ jµ (x) = −(n/16π 2 ) Tr[4∂µ Aν ∂ρ Aσ + 4∂µ Aν [Aρ . T a = 1 λa . However.57) where Tr is in the internal quantum number space and G µν = igG a T a . Aσ ]]ε µνρσ = −(n/4π 2 )εµνρσ ∂µ Tr[Aν ∂ρ Aσ + 2 Aν Aρ Aσ ] 3 (13.57) and of ﬁeld conﬁgurations with a non-zero topological charge (see Section 8. it is determined by the gauge invariance of the axial-vector current. this contribution is not independent of the triangle anomaly: given the latter. The anomaly (13.3). unlike the case of the spontaneously broken SU (n) × SU (n). These properties offer a mechanism of breaking of the UA (1) which does not imply the existence of the massless Goldstone boson coupled to the gauge-invariant UA (1) current.

anomaly-free (Gross & Jackiw 1972.58) . Different -vacua are degenerate states since for massless fermions H commutes with Q 5 . We can now systematically list possible anomalous triangle diagram contributions to the divergences of the gauge source a currents. However.3 we also remember that GN | Thus we get G N exp(i or exp(i Q 5 )| =| + 2n The conclusion is that the UA (1) rotation by an angle changes the | vacuum vacuum.2 Physical consequences of the chiral anomalies 475 can be deﬁned. It will also be convenient to remember that by construction the fermion electric charge operator is given by Q = T 3 + Y . it can be seen from the chiral Ward identities that the spontaneous breakdown occurring because of the axial-vector anomaly and the existence of ﬁeld conﬁgurations with non-zero topological charge (hence the set of the -vacua) is consistent with the assumption 5 of no Goldstone boson coupled to the gauge-invariant current jµ .9 with the jLµ in one of the vertices and the corresponding Q 5 )| = exp[−iN ( + 2n )] exp(i Q 5 )| = exp(−iN )| (8. see also Jackiw 1984 and our discussion in the next section). however. Now. i.65) and (8. Anomaly cancellation in the SU (2) × U (1) electroweak theory We discuss this problem at the level of the triangle diagrams relegating to the next section the construction of the full anomalous divergence of a non-abelian current coupled to gauge ﬁelds. This does not limit the generality of our discussion: a theory is anomaly-free if.3 and in particular (8. 13. UA (1) symmetry is spontaneously broken because into the | + 2n the vacuum is not UA (1)-invariant.67) and Problem 8. Take ﬁrst the SU (2) current jLµ : we can have the triangle diagrams a shown in Fig.2) −1 G N Q 5 G N = Q 5 − 2n N From Section 8.and right-handed currents but with different U (1) charges Y .e. In the electroweak theory the SU (2) gauge bosons couple only to the left-handed currents and the U (1) gauge boson couples to both the left. Thus. For consistency of a quantum gauge theory the gauge source currents must be covariantly conserved. the triangle diagram anomalies are absent. We have (see Section 8. Q 5 is not gauge-invariant: it is not gauge-invariant under large gauge transformations G N with the winding number N .13. and only if. YR = Q and YL = (Q − T 3 ).

Y symmetrized diagrams ( jµL is the left-handed part of the U (1) current).59) where the sum is taken over the charges of all fermions in the loop.9(b) does not contribute because [Y.9. We recall that the U (1) coupling is not vector-like because of different Y quantum number assignments to the left. where the sum is over all fermions in the theory. with left. T b }] for Fig. 13.60) . T b }] and 1 2 Tr[Y {Y. T c }] Tr[T a {Y. i ( L γµ Y L + can be studied in a similar way. T a }] = δ ab Tr[Q(T a )2 ] = 1 δ ab 2 i Q i (13. 13. T c }] = 0 for the group SU (2). If we express Y in terms of Q the remaining contributions are 1 2 Tr[T a {Q.and right-handed fermions. It is clear from our discussion in the previous section that the eventual anomalous contribution of these diagrams after summing over all fermions in the loops is proportional to 1 2 1 2 1 2 Tr[T a {T b . T a ] = 0 and Tr T a = 0. We calculate 1 2 Tr[Y {Y. T a }].9(b) for Fig. 13. Y The anomalous contribution to the divergence of the U (1) current jµ = i i i i ¯ ¯ R γµ Y R ).9(c) (T a = σ a are SU (2) generators).26)). 1 2 Tr[Y {T a . 13. We immediately see that Fig. Y }] Tr[T a {Y.9(a) for Fig. T b }] = 1 2 Tr[Q{T b .and right-handed ﬁelds. which contribute with opposite signs because of the (1 ± γ5 ) factors (see (13. In this case all possible triangle diagrams are proportional to one of the following factors: 1 2 Tr[Y {Y. Y }] = Tr Y 3 = Tr(Q − T3 )3 = Tr[Q 3 − 3Q 2 T3 + 3Q(T3 )2 − (T3 )3 ] (13. Y }] The ﬁrst two traces have already been calculated.476 13 Chiral anomalies Fig. and Tr[T a {T b . The last one appears in two types of loop. 13.

the Q 3 term is cancelled separately for each fermion when the sum over the two types of loop is taken.s. ∂ µ jµ = −(1/16π 2 ) Tr[ε µνρσ G µν G ρσ ] (13. Insisting on chiral gauge current conservation we are formally back to the renormalization conditions (13.61) since Tr[Q 2 T 3 ] = Tr[(T3 + Y )2 T3 ] = Tr[(T3 )3 + 2Y (T3 )2 + Y 2 T3 ] = Tr Y = Tr Q. The remaining terms give again Qi = 0 i (13. An interesting point to mention in the context of the SU (2) ×U (1) theory is that once we insist on conservation of the chiral source currents for the electroweak gauge ﬁelds (one needs suitable sets of fermion multiplets to ensure this property) the baryon number current. The quarks and leptons discovered so far strikingly satisfy this condition: to each SU (2) doublet of leptons corresponds a doublet of quarks.62) where G µν = igG a T a is the SU (2) ﬁeld strength tensor. One would then expect ‘topological’ baryon number non-conservation. Notice that the condition of vanishing divergence of two currents can always be imposed on triangle diagrams. . This is clear if we consider the triangle diagram with the baryon current and two gauge source currents. Since U (1) current is an SU (2) singlet.10) and the anomaly is placed in the vector current.62) does not vanish due to some non-perturbative effects like tunnelling.61) is the only condition for the anomaly cancellation. for instance. fractionally charged but coming in three colours. An interesting µν possibility is that the vacuum expectation value of the r. acquires an anomaly. analogously to (13.13. or monopoles. in the operator notation we have.35). Thus (13.h. Restrictions on the fermion representations appear when we consider triangle diagrams with three gauge currents. U (1) vector current. of (13.2 Physical consequences of the chiral anomalies 477 Since Q is the same for fermions with either of the chiralities. We conclude that the SU (2) × U (1) theory is anomaly-free if the sum of charges of all fermions in the theory vanishes. with instantons as the dominant ﬁeld conﬁgurations.

for an anomaly-free theory the quantity Dabc must vanish. The other case is when the group is not safe. for all representations of the group. The transformation properties of the fermion and the gauge ﬁelds under inﬁnitesimal † Since in this section ﬁelds Aµ are treated as background ﬁelds.e. the theory is vector-like (Appendix E).and right-handed couplings. These are called ‘safe’ groups.64) and the T α s are the generators of the gauge group in fermion space. They include SU (2). One can distinguish three cases of anomaly-free models (Georgi & Glashow 1972).3 Anomalies and the path integral Introduction For a systematic discussion of chiral anomalies it is convenient to use the generating functional and the path integral technique. i.A Aµ = iAV.A T α αµ = ¯ iγ µ αµ αµ − jV AV − jA AA αµ αµ (13.63) where T a . Two cases may be distinguished here. the coupling constant g is included in the deﬁnition of Aµ . 13.478 13 Chiral anomalies Anomaly-free models We conclude with more general remarks. but there exist some ‘safe’ representations for which Dabc vanishes. It may be that. We consider a theory of massless fermions coupled to non-abelian gauge ﬁelds. . Depending on the physical problem at hand we may need to use vector and axial-vector couplings or left. all orthogonal groups except S O(6) ≈ SU (4) and all symplectic groups. We take all three currents to be gauge source currents. If this is not the case. Dabc = 0. T b and T c are generators of the group of transformations of currents attached to the three vertices. like SU (N ) with N > 2 or SU (2)×U (1). The right-handed fermion loop and the left-handed fermion loop anomalies may cancel. In the ﬁrst case the gauge-invariant lagrangian density reads† L = ¯ iγ µ (∂µ + AV + γ5 AA ) µ µ where V. For the purpose of the present discussion it is convenient to treat the gauge ﬁelds as classical background ﬁelds. T c }] (13. As we know the triangle diagram anomaly is given by a multiple of Dabc = 1 2 Tr[T a {T b . This happens when fermions with either of the chiralities transform according to the same representation of the gauge group. This then gives a limitation on the allowed fermion representations.

It is often more convenient to use the left. The lagrangian (13. We see.65) = (1 − γ5 δ δ A A A) A.3 Anomalies and the path integral 479 vector and axial-vector gauge transformations are = (1 − δ δ V V V) V A V = ∂µ µ AA = −[ µ −[ AA ] µ V. the gauge ﬁelds in the lagrangian (13.67) also depends on the problem.R = 0 (see (1. by assumption. no gauge bosons are coupled to the corresponding abelian currents.R V A classical level the U (1) currents are conserved: ∂ µ jµ = ∂ µ jµ = 0 and the αµ non-abelian currents are covariantly conserved: Dµ jL. only two types of Feynman diagram appear in our theory: diagrams in which the external ﬁelds couple to a free spinor line and diagrams in which the external ﬁelds couple . the gauge ﬁelds are classical background ﬁelds. in particular.64) or (13.R γµ TL. Depending on the physical problem considered. In addition the same fermions may couple to other gauge ﬁelds.R where α α jL.R iγ µ ∂µ L.64) is also invariant under global U (1) and UA (1) transformations. Of course. However.13. Since. A AV = −[ µ A A = ∂µ µ AA ] µ AV ] µ −[ (13. for instance.3).R − jµ Aµ − jµ Aµ = Lα Rα L. This is.66) respectively. the case with the ﬂavour symmetry currents in QCD.Rµ = ¯ L. too. V V. introduced to write down the generating functional for the anomalous Ward identities involving currents which in the framework of the considered theory are not coupled to any dynamical gauge ﬁelds. At the L.67) can be regarded as dynamical gauge ﬁelds or as auxiliary background ﬁelds.R Invariance under gauge transformations on the chiral ﬁelds requires standard transformation properties for the gauge ﬁelds Aµ (see Section 1.and right-handed ﬁelds and then L = ¯ L iγ µ (∂µ + iAα T α ) L + ¯ R iγ µ (∂µ + iAα T α ) R Lµ Rµ (13.125)). that in a non-abelian theory a pure axial-vector interaction with Dirac fermions would not be gauge-invariant. the physical interpretation of the non-abelian structure in the lagrangians (13. where as usual = i α T α . corresponding to a gauge group acting in a different space. (13.67) Lα Rα ¯ L.R L.64) and (13. A V ] = Dµ µ A.

η. ¯ (x) → ¯ (x) exp[−iγ5 (x)] (13.1. µ Thus we consider the axial abelian anomaly in a QCD-like theory.9). Naively we then get (10.70) One may ask where the anomalous term comes from in the path integral formalism. are not altered). Eq.69). We use the lagrangian (13. has no impact on the generality of the discussion of chiral anomalies due to the Adler–Bardeen non-renormalization theorem quoted in Section 13.1 to derive Ward identities we take the functional exp{iZ [Aµ . (13. This. η and η and set η = η = 0. Applying the general formalism developed in Section 10. η]/δ | ¯ 1 = exp{iZ [A. It reads 0 = δ Z [A.70) generates the desired Ward identities when we differentiate it functionally with respect to the Aα . η. η]} = ¯ D D ¯ exp i µ d4 x (Lkin − jα Aα + η + ¯ η) ¯ µ (13. Using the latter we can anticipate the correct modiﬁcation of (10. The answer has been given by Fujikawa (1980) who noticed that the anomaly is due to the non-invariance of the fermionic path integral measure under the transformation (13. We shall come back to this point in Section 13. η. Abelian anomaly Let us ﬁrst consider the generating functional for Ward identities involving the U (1) axial-vector current.68) and use its independence of the choice of the integration variables changed according to the transformation (x) → exp[−iγ5 (x)] (x).57) in particular.64) and specify AA = 0.9) which generates µ 5 Ward identities corresponding to ∂ µ jµ = 0 and is wrong in view of our perturbative calculations and of (13.4.480 13 Chiral anomalies to a spinor loop.69) (the gauge ﬁelds Aα . however. η]} ¯ × exp i =0 d4 z D D¯ + ¯ η) d4 x ( ¯ i/ ∂ µ − jα Aα + η ¯ µ 5 ˜ ¯ × {−(n/8π 2 ) Tr[G µν (z)G µν (z)] − ∂ µ jµ (z) + η(−iγ5 ) + (−i ¯ γ5 )η} (13. For instance ¯ ¯ µ β 5 α the anomalous Ward identity for the matrix element 0|T jµ (z) jν (y) jδ (x)|0 is obtained from (13.70) by taking the derivative ∂ ∂ Aα (y) ν ∂ ∂ Aδ (x) β .

if we use chiral ﬁelds.66).13. Differentiating with respect to η and η and setting η = η = 0 one gets the anomalous Ward identity for ¯ ¯ 5 ¯ the matrix element 0|T jµ |0 .65) and (13.64) or (13.R −[ L.R (13. On the other hand.R ] = Dµ µ L. First.73) and 1 δ Z [Aµ ] = δ Aµ (x) exp{iZ [Aµ ]} D ˆ and jµ (x) is the current coupled to the ﬁeld Aµ (x) whereas J µ (x) is deﬁned by (13.67). If we work with vector and axial-vector gauge ﬁelds.74) . we study the gauge dependence of the functional Z [Aµ ].R = ∂µ µ L. AL.and right-handed gauge ﬁelds. The explicit formulae (13. We again consider the functional exp{iZ [Aµ ]} = D D ¯ exp i / d4 x ¯ iD (13.3 Anomalies and the path integral 481 and setting η = η = 0. of course. The gauge transformation of the fermion ﬁelds is again just a change of integration variables but under the accompanying transformation of the gauge ﬁelds Aµ (x) → Aµ (x) + δ Aµ (x) the functional Z [Aµ ] changes as follows Z [Aµ ] → Z [Aµ ] − 2 where d4 x Tr δ Z [Aµ ] δ Aµ (x) δ Aµ (x) (13. other ﬁelds are suppressed. respectively.71)–(13. on whether we use vector and axial-vector ﬁelds or chiral ﬁelds.72) δ δ = (iT α ) δ Aµ δ Aα µ D ¯ j µ (x) exp i / d4 x ¯ iD ˆ ≡ J µ (13.73) and that for δ Aµ depend. Non-abelian anomaly and gauge invariance Our next task is to study the generating functional for the anomalous Ward identities involving non-abelian currents.73).56). One gets ¯ 5 α α ˜ ∂ µ 0|T jµ (z) jν (y) jδ (x)|0 = −(n/8π 2 ) 0|T Tr[G µρ G µρ ] jν jδ |0 β β and in momentum space this is just equivalent to (13.71) corresponding to the lagrangian (13. The stands for Dirac or chiral fermion ﬁelds and Aµ for vector and axial-vector gauge ﬁelds or left.72) includes both AV and µ AA which transform under vector or axial-vector gauge transformation according µ to (13.R .and right-handed ﬁelds separately form a group and we have δ AL. the gauge transformations on the left. then Aµ (x) in (13.

R (x) µ = Dµ δ δ AL. µ δ δ AV (x) µ + AA (x).72) and (13. µ δ δ AA (x) µ + AA (x). † It is convenient to use a test functional.482 13 Chiral anomalies In each case we may rewrite (13.R (Y )] = cαβγ X L. X V (y)] = cαβγ X V δ(x − y) β α αβγ γ [X V (x).R (x). X A (Y )] = cαβγ X V δ(x − y) γ β α [X L. X L. applied to Z [Aµ ]. µ δ δ AV (x) µ (13.76) = −X α (x)Z [Aµ ] (13. µ δ δ AA (x) µ (13.81) γ β α [X A (x). With the help of (13.R δ(x − y) which. .82) expressing the group property of the gauge transformations.R (x) µ (13.R (x). X A (y)] = c X A δ(x − y) (13.75) which is easy to verify.80) for chiral gauge transformations on the chiral gauge ﬁelds (all commutators are for matrices T α and not for ﬁelds which are c-numbers). are equivalent to the relation δ 1 δ 2 Z [Aµ ] − δ 2 δ 1 Z [Aµ ] = δ[ 1.R (x) = ∂µ δ δ AL. and X L.78) α (x)| (x)=0 d4 x Tr[X (x)Z [Aµ ] (x)] (13. µ δ δ AL. 2] Z [Aµ ] (13.79) for the axial-vector gauge transformations on the vector and axial-vector ﬁelds.75) we get δ Z =2 or δ Z [Aµ ]/δ where X V (x) = ∂µ δ δ AV (x) µ + AV (x). X A (x) = ∂µ δ δ AA (x) µ + AV (x). By explicit calculation† one veriﬁes that the X s satisfy the gauge group commutation relations γ β α [X V (x).73) using the following relation ˆ d4 x Tr[ J µ (x)Dµ (x)] = − ˆ d4 x Tr[D µ Jµ (x) (x)] (13.R (x) µ + AL.77) (where X = iX α T α ) for the vector gauge transformations on the vector and axialvector ﬁelds.

76). (13. Firstly.R (13. To study gauge invariance of the complete quantum theory it is sufﬁcient to study the gauge invariance of the Z [Aµ ].81) or (13.85) or equivalently α −X L. by the absence of ¯ the functional integration over D Aµ and of the gauge ﬁeld kinetic terms which are gauge-invariant. (13. from the triangle diagram calculation in Section 13.R D µ jµ = 0 (13.80) depending on the considered gauge transformation.R (x) (13.R ] ν σ and −(1/2π 2 )ε µνρσ Tr[T α (∂µ AV ∂ρ AV + 1 ∂µ AA ∂ρ AA )] ν σ ν σ 3 (13.83) where the X is one of the X s given by (13.89) .88) which are (at α α the triangle diagram level D µ = ∂ µ ) ±(1/12π 2 )ε µνρσ Tr[T α ∂µ AL.86) and the G A (x) in (13. The anomaly means that the variation of Z [Aµ ] under gauge transformation does not vanish L. The non-abelian source currents for gauge ﬁelds must be covariantly conserved at any order of perturbation theory for gauge invariance to be maintained in quantum theory.90) (13.78)–(13. η.88) Indeed.77) and (13.R (x)Z [Aµ ] = G α (x) L. η are the gauge and fermion ﬁeld sources. η.3 on the classical level.82) have several implications. for invariance of our quantum theory under one of the previously deﬁned gauge transformations we must have δ Z [Aµ ] = 0 ⇒ X (x)Z [Aµ ] = 0 (13.R (x) in (13.R D µ jµ (x) = G L.13. where J.R ∂ρ AL.3 Anomalies and the path integral 483 Eqs. In a theory with only vector or only chiral gauge ﬁelds. using (13.1 we know partial contributions to the G L. With dynamical gauge ﬁelds the generating functional Z [Aµ ] differs from the Z [J.84) obtained in Section 1.86) Similarly.87) or α −X A (x)Z [Aµ ] = G α (x) A (13.73) we recover the equation V D µ jµ = 0 or L. if we work with vector and axial-vector currents and insist on the vector V current conservation D µ jµ (x) = 0 then the divergence of the axial current is anomalous and A A A D µ jµ = ∂ µ jµ + [Aµ . jµ ] = G A (x) V (13. η] of the complete ¯ theory.

R L. Naively. Using (13. this current would be expected to transform covariantly under gauge transformations.87).86) or (13.R L. in addition to the gauge variation δ A µ = Dµ 4 4 ˆµ (x)] δ Z [A] = −2 d x Tr[(δ Z /δ Aµ )δ Aµ (x)] = 2 d x Tr[Dµ J 4 α α =− d x (x)G (A) (13.82) and (13. Therefore the Wess–Zumino condition completely determines G α [Aµ ] once the term (13.R L.R (x)G α (y) − X L. After some calculation we can get G α (x) = ±(1/12π 2 ) Tr[T α ∂ µ εµνρσ (Aν ∂ ρ Aσ + 1 Aν Aρ Aσ )] (13.89) is given.R (x)δ(x − y) L.R (y)G L. Different numerical coefﬁcients reﬂect different renormalization conditions for the two diagrams: conservation of the vector currents for the ﬁrst one and symmetry between all three vertices for the second one. Observe that as long as for the anomaly the minimal form depending only upon the gauge ﬁelds and not other external ﬁelds is taken.88) follow from (13.R The expression for the axial anomaly must satisfy the following consistency conditions (again.85) we get for the anomaly G L. Consistent and covariant anomaly Finally we study the transformation properties of the anomalous current.81)) V X α G A (y) = cαβγ G A δ(x − y) γ β A A X α G A (y) − X β G A (y) = 0 α β (13. All the anomalous Ward identities follow from the relations (13. One way to get the full anomalous terms G α (A) is to use the so-called Wess– Zumino consistency condition (Wess & Zumino 1971).93) and it is much more complicated (Bardeen 1969).R (x) the equation α α X L.R β β γ (13. In the second expression the ﬁrst term comes from the AVV diagram and the second one from the AAA diagram. (13. To determine the effect of the anomalies.92) L.R 2 L. see (13. since the G α (A)s factorize out of the path integral which cancels with the exp{−iZ [A]}.85) and (13.81) or (13.94) .484 13 Chiral anomalies respectively.R L.R (x) = −cαβγ G L.86) and (13.91) The importance of this condition follows from the fact that the operator X is nonlinear in the gauge potential Aµ .88) (we may need to introduce additional sources to the functional Z ). respectively.

h.Rµ in gauge ﬁelds Its explicit form can be found by searching for a polynomial Pα ˆµ with an anomalous gauge transformation rule opposite to that of the current Jα . The covariant current may have some physical signiﬁcance when coupled to other external non-gauged ﬁelds.101) are called consistent and covariant anomalies.92) there exists the covariant non-abelian current L.98) The ﬁrst term on the r.96) Applying this operator to the functional Z [A] we obtain ˜ ˜ (δ B δ − δ δ B )Z [A] = = ˜ d x {[δ B G α (A)] 4 α µ α ˆα )Bµ (x)} (x) − (δ J (13. J µ ]α Bµ − [δ B G α (A)] α } (13.101) The anomalies (13. (x)]}α The gauge transformation properties of the non-abelian current follow from (13. ] d4 x Tr[(δ Z /δ Aµ )δ Aµ ] = −2 ˆ d4 x Tr[ J µ Bµ (x)] (13.Rµ Pα = ∓(1/24π 2 )εµνρσ Tr[Tα (Aν G ρσ + G ρσ Aν − Aν Aρ Aσ )] (13.Rµ )α = G α (A) = ±(1/16π 2 )εµνρσ Tr[Tα G µν G ρσ ] (13.95) (13. .Rµ = Jα Jα (13.92) and (13.100) ˜α The current Jµ is also anomalous ˜ ˜ (Dµ J L.Rµ + Pα (A) ˜L.Rµ ˆL. By deﬁnition ˜ δ B Z [A] = −2 and ˜ δ B Aµ (x) = δ Aµ ≡ Bµ (x) The following operator equation holds ˜ ˜ ˜ δ B δ − δ δ B = δ[B. Bardeen & Zumino (1984) show that L.97) ˆµ d4 x Jα {[Bµ (x).s. respectively.97) ˆµ α d4 x (δ Jα )Bµ (x) = − α ˆ ˜ d4 x {[ . the second term is due to the anomaly. gives the usual transformation property of the current. Each can be used to study anomaly cancellation. The physical sense of the original non-abelian current has been discussed earlier. One can show (Bardeen & Zumino 1984) that in the case of the anomaly (13.3 Anomalies and the path integral 485 we introduce the variation which deﬁnes the current.13.99) L.

3).486 13 Chiral anomalies 13. We deﬁne the Euclidean vector potential Aµ as (13. For the spatial coordinates we have (see Section 2.104) ˆ so that Aµ form a Euclidean four-vector. γν } = 2δµν ˆ ˆ and explicitly γ4 = γ0 . 2. ˆ ˆµ Dµ = −∂/∂ x µ + ig Aa T a ˆ ˆ 4i G a = −iG a 0i (13.103) ˆ where again µ = 1. . ˆ The matrix γ5 is taken as ˆ ˆ ˆ ˆ ˆ γ5 = −γ1 γ2 γ3 γ4 ˆ (13.107) .110) γi = −iγi . ˆ Ai = − Ai (13. . The fermion ﬁelds and ¯ are independent variables in the path integral and in Euclidean space we deﬁne ˆ = . ν = 1. − p y . .102) where µ = 1. −y. (13. ˆ i = 1. 2. −z) ˆ x µ = −xµ ˆ ˆ pi = (− px . First let us summarize the rules of continuation from Minkowski to Euclidean space. 4. ˆ =i¯ ¯ (13.109) ˆ / and it is also hermitean.105) The ﬁeld strength tensor is ˆ G iaj = G iaj where (i. 3. 3. . −∂/∂ xi )) µ ˆ D0 = i D4 . Thus for the covariant derivative Dµ = ∂µ + Aµ = ∂µ + ig Aa T a we get (∂µ = (∂/∂ x0 .3) x0 = −ix4 ˆ ˆ xi = xi p0 = −i p4 ˆ ˆ pi = pi xi = (−x. − pz ) ˆ pµ = i(∂/∂ x µ ) ˆ ˆ ˆ A0 = i A4 .4 Anomalies from the path integral in Euclidean space Introduction In this section we discuss the techniques of anomaly calculation using the path integrals in Euclidean space (Fujikawa 1980). j = 1. 4 (13. 2.108) µ.106) ∂ ˆ ∂ ˆ ˆ µν ˆµ ˆν G a = − µ Aa + ν Aa + gcabc Ab Ac ν ∂x ˆ ∂x µ ˆ We deﬁne the Euclidean γ -matrices to be hermitean matrices γn obeying ˆ {γµ . The Dirac operator i D is hermitean. 4 and correspondingly (13. 3 (13. 2. ˆ Di = − Di .

115) Thus non-vanishing eigenvalues always occur in pairs of opposite sign.113) ˆ ¯ / + ˆ (i D + iM) ˆ To complete this introduction let us discuss the eigenvalue problem for the ˆ ˆ / / hermitean Dirac operator i D in Euclidean space.116) (13. We observe ˆ / also that for λn = 0 chiral ﬁelds are not eigenvectors of i D ˆ / i D (1 ± γ5 ) ˆ n = λn (1 ∓ γ5 ) ˆ n ˆ ˆ (13.118) In the rest of this section we work in Euclidean space and the caret is omitted. i D } = 0 ˆ / we get ˆ /ˆ ˆ i D γ5 ˆ n = −λn γ5 ˆ n (13.13. The Euclidean action is (13. ˆ and δ ˆ † = − 1 ˆ † (γµ γν − γν γµ )ωµν ˆ ˆ ˆ ˆ ˆ 8 (13. For simplicity we assume that i D has a purely discrete spectrum.112) † ¯ so that ˆ 1 ˆ 2 is a scalar and we identify the transformation properties of ˆ as those † † of ˆ (in Minkowski space 1 γ0 2 is a scalar). This corresponds to stereographically projecting Euclidean four-space onto a four-sphere and changes the determinant but only by ˆ / a factor which is independent of the gauge ﬁeld.114) and because ˆ {γ5 .111) ω0i = iω4i ˆ ˆ iS = − S where ˆ S= ˆ d4 x 1 ˆa ˆa G G 4 µν µν (13. .4 Anomalies from the path integral in Euclidean space 487 Under inﬁnitesimal rotations in Euclidean space ˆ and ˆ † transform as follows: δ ˆ = 1 (γµ γν − γν γµ )ωµν ˆ ˆ ˆ ˆ ˆ ˆ 8 where ωi j = ωi j . Since γ52 = 1 its eigenvalues ˆ are ±1 and γ5 (1 ± γ5 ) ˆ n = ±(1 ± γ5 ) ˆ n ˆ ˆ ˆ (13.117) On the other hand eigenfunctions of the vanishing eigenvalue can always be chosen to be eigenfunctions of γ5 and thus to be chiral ﬁelds. Since i D is hermitean it has real eigenvalues λn ˆ / i D ˆ n = λn ˆ n (13.

119) where and ¯ are two independent variables describing Dirac fermions coupled to gauge potentials transforming under a vector-like gauge group. However.120) (13. Under local inﬁnitesimal axial transformations (13.119) the lagrangian is transformed into µ L (x) = L(x) + ∂µ j5 (x) and therefore / d4 x ¯ iD → / d4 x ¯ iD − d4 x µ (x)∂µ j5 (x) (13. We get (x) = n an ϕn (x) = n an exp[i (x)γ5 ]ϕn (x) (13.122) up to an irrelevant arbitrary normalization factor. The measure becomes D D¯ = n dan m ¯ dbm (13.121) † / iD ϕn (x) = λn ϕn (x). ϕn (x)bn (x) = n n (13.124) or am = n † d4 x ϕm (x) exp[i (x)γ5 ]ϕn (x)an = n Cmn an (13. the measure (13.123) If the measure were invariant under axial transformations then using the invariance of the path integral under a change of variables one could derive (10.122) is not invariant and following Fujikawa we can explicitly evaluate the Jacobian of the change of variables.125) .11) (with δ a L(x) = 0) implying the conservation of axial current at the quantum level. d4 x ϕn (x)ϕm (x) = δmn ¯ The coefﬁcients an and bn are Grassmann variables. can be deﬁned / precisely by expanding and ¯ in terms of the eigenfunctions ϕn of iD † ¯ ¯ (x) = an ϕn (x).488 13 Chiral anomalies Abelian anomaly with Dirac fermions As mentioned in the previous section the anomaly of the ungauged axial U (1) 5 current jµ = ¯ γ µ γ5 can be understood as a non-invariance of the path integral measure under local transformations on fermion ﬁelds = exp[i (x)γ5 ] ≈ (1 + i (x)γ5 ) ¯ = ¯ exp[i (x)γ5 ] ≈ ¯ (1 + i (x)γ5 ) The generating functional in the Euclidean space exp{−Z [A]} = D D ¯ exp − / d4 x ¯ iD (13.

4 Anomalies from the path integral in Euclidean space 489 Thus (see (2.e. Hence the measure is multiplied by exp −2i d4 x (x) n † ϕn (x)γ5 ϕn (x) (13. In our problem we want to choose a regularization procedure which preserves the invariance under the vectorlike gauge group with gauge potentials Aµ . Going over to a .128) with a gaussian cut-off for large eigenvalues λn d4 x (x) n † ϕn (x)γ5 ϕn (x) ≡ lim M→∞ d4 x d x d4 x 4 (x) n † ϕn (x)γ5 ϕn (x) exp(−λ2 /M 2 ) n † / Tr[γ5 exp[−(iD )2 /M 2 ]ϕn (x)ϕn (x)] n ∗ [ϕn (x)]lb [(ϕn (y)]a k n = lim M→∞ (x) = lim M→∞ / (x) lim {γ5 exp[−(iD x )2 /M 2 ]}ab kl y→x (13.116)) dam = det−1 C m n dan (13. with this fact in mind we have / been working in the basis (13.121) of the eigenfunctions of iD = i(/ + A) which ∂ / are gauge-invariant. gauge transformed eigenfunctions are eigenfunctions of the gauge transformed operator. Actually.128) which is ill deﬁned as it stands and must be regularized. i.129) where Tr means the trace over the group and the Dirac indices.126) and the Jacobian for D ¯ gives the identical factor.13. Furthermore. Hence we may simply regularize the expression (13.127) where the trace is taken over the whole Hilbert space. we have C = exp[i (x)γ5 ] and det C = exp{i Tr[ (x)γ5 ]} = exp i n d4 x † (x)ϕn (x)γ5 ϕn (x) (13.

490 13 Chiral anomalies plane wave basis and using the completeness relation for the ϕn (x) we get ∗ [ϕn (x)]lb [ϕn (y)]a k n = δ ab δ kl n x|n n|y = δ ab δ kl x|y d4 k x|k k|y (2π )4 d4 k exp(−ikx) exp(+iky) (2π )4 = δ ab δ kl = δ ab δ kl and therefore† d4 x = lim (x) n † ϕn (x)γ5 ϕn (x) (13.128) and invariance of the path integral under the change of variables we ﬁnally get 5 ˜ ∂ µ jµ (x) = (i/8π 2 ) Tr[G µν G µν ] (13. As a by-product of this derivation of the abelian anomaly we can easily / understand the content of the so-called index theorem for the Dirac operator iD .132) which corresponds to the previous result (13.123).133) vanishes for eigenfunctions with eigenvalue λn = 0 since γ5 ϕn has eigenvalue −λn and is thus orthogonal to ϕn . observing that d4 x n † ϕn (x)γ5 ϕn (x) (13. (13.131) for (x) independent of x.131) 1 16π 2 where the last trace is over the group indices only. and writing † † † ϕn (x)γ5 ϕn (x) = 1 ϕn (1 + γ5 )2 ϕn − 1 ϕn (1 − γ5 )2 ϕn 4 4 † In Euclidean space we deﬁne ε 1234 = −1 and then Tr[γ5 γ µ γ ν γ ρ γ δ ] = 4ε µνρσ . Using (13.57) obtained in Minkowski space by the triangle diagram calculation.130) M→∞ d4 x d4 x d4 x = lim = M→∞ d4 k / Tr[γ5 exp(+ikx) exp[−(iD )2 /M 2 ] exp(−ikx)] (2π )4 d4 k 1 21 (x) Tr[γ5 ([γ µ . The theorem relates the number of positive chirality minus the number of negative / chirality zero modes of the operator iD to the topological charge (Pontryagin index) of the background gauge ﬁeld Aµ . γ ν ]G µν )2 exp(+k µ kµ /M 2 ) 4M 2 2! (2π )4 (x) ˜ (x) Tr[G µν G µν ] (13. . Using the result (13.

i. it can always be regularized in a gauge-invariant anomaly-free manner (in other words positive and negative chirality anomalies cancel each other). . In Section 13. as we know from the previous discussion.3 we have understood the non-abelian anomaly as a breakdown of the gauge invariance of the functional Z [A]. This suggests a possible approach to its calculation using the path integral formalism: calculate / Z [A] = − ln det iD (A) (13.3 we assume here that the background gauge ﬁelds rapidly approach the pure gauge conﬁguration of |xµ | → ∞ so that the Euclidean action is ﬁnite and the topological charge Q is well deﬁned.13. As in Section 8.4 Anomalies from the path integral in Euclidean space 491 for the eigenfunctions with eigenvalue λn = 0.e. The topological charge density is connected to the anomaly density.135) The left-handed fermions L and the right-handed fermions ¯ L are independent variables. leftand right-handed chiralities transform under the same representation. Then we can formulate our theory in terms of the Dirac fermions = L + R and in Euclidean / space we are back to the hermitean eigenvalue problem iD n = λn n with real / λn .136) and its change under gauge transformation. Non-abelian anomaly and chiral fermions We consider now the structure of the non-abelian anomaly in theories of chiral fermions coupled to an external dynamical or auxiliary gauge ﬁeld Aµ and described by the effective action exp{−Z [A]} = D LD ¯ L exp − / d4 x ¯ L iD L (13.62). we get n+ − n− = Q (13. However.134) where n + (n − ) is the number of positive chirality (negative chirality) zero modes of / the operator iD and Q is the topological charge deﬁned by (8. In this case det iD is real and. as we know from the / introduction to this section the operator iD (A) maps positive chirality spinors to negative chirality spinors / iD (A) L → R and consequently it does not have a well-deﬁned eigenvalue problem and a welldeﬁned determinant on the space of chiral ﬁelds. The only exception is when chiral fermions transform as a real representation of the gauge group.

Leutwyler 1984). and different regularization methods we refer the reader to the original literature (Fujikawa 1984. performing a shift of the integration variable such that (13. The main difﬁculty is to invent an appropriate regularization procedure. For actual calculation. One can circumvent the / problem of deﬁning det iD observing that it sufﬁces to calculate the change of / ln det iD under a gauge transformation. The overall phase of the determinant is irrelevant for us and so is the dependence on A.2 Show that Ward identities for the Green’s function of the three-vector currents are anomaly-free. for example.138) / / is meaningful since D −1 D maps a space of fermion ﬁelds with given chirality into itself.21) starting with the unregularized integral representation (13.5). Of course.139) Problems 13. 13. if det iD was deﬁned in the space of chiral fermions. Discuss this point also in Euclidean space using the material of Section 13. We arrive at an interesting observation (Alvarez-Gaum´ & Witten 1983): in e Euclidean space only the imaginary part of the functional Z R [A] may suffer from the anomalies which give its anomalous variation under gauge transformation.1 Derive the anomalous Ward identity (13. Nair & Trahern 1982.93) which in Euclidean space reads G α = Dµ L ∂Z i µνρσ ε Tr[T α ∂ µ (Aν ∂ρ Aσ − 1 iAν Aρ Aσ )] α = − 2 ∂ AµL 12π 2 (13. We can always imagine an extended theory with additional fermions in R ∗ so that Z R+R ∗ [A] = Z R [A] + Z R ∗ [A] = Z R [A] + Z ∗ [A] = 2 Re Z R [A] R (13. / Equivalently. Marmo. 13. A = 0. Thus we see that the calculation of the non-abelian anomaly from the path integral is a well-deﬁned problem. . we could say / that only the phase of iD may be gauge-non-invariant.137) But R + R ∗ is a real representation and consequently Z R+R ∗ [A] is gauge-invariant.492 13 Chiral anomalies Now consider chiral fermions in a complex representation R.3 In Fig.6 replace photons by gluons and check that in QCD there is no anomaly for the axial isospin current.4. 13. the result e agrees with (13. And / / / / ln det iD (A g ) − ln det iD (A) = ln det[−iD −1 (A)iD (A g )] (13. Alvarez-Gaum´ & Ginsparg 1984. Balachandran. we can put.3) is satisﬁed for this integral representation and studying the surface term induced by this change of the integration variable in the linearly divergent integral.

y.4 Check that V V G α = −(1/4π 2 )εµνσ τ Tr[λα [ 1 Fµν Fσ τ + A 4 V V + 2 (Aµ Aν Fσ τ − Aµ Fνσ Aτ + 3 1 A A 12 Fµν Fσ τ V Fµν Aσ Aτ ) − 8 Aµ Aν Aσ 3 493 Aτ ]] where µν FV µν FA = ∂ µ V ν − ∂ ν V µ + [V µ . V ν ] + [Aµ . y. τ ).Problems 13. Aµ = +iλi Ai µ satisﬁes the consistency conditions (13. τ ) and h(x. τ = 0) = δ(x − y) ∞ 0 1 (t) dτ τ t−1 h(x. V ν ] + [Aµ . y. x. x. y) ≡ n ∗ f n (x) f n (y) = t λn h(x. τ ) = −(∂/∂τ )h(x. y. Aν ] i Vµ = +iλi Vµ . y. t)] n t→0 y→x . Aν ] = ∂ µ Aν − ∂ ν Aµ + [V µ . y. τ ) = n ∗ exp(−λn τ ) f n (x) f n (y) = x| exp(−Aτ )|y 1 (t) ∞ 0 dτ τ t−1 dx h(x. deﬁned as ζ A (t) = n 1/λt n converges for Re t > 2 and can be analytically extended to a meromorphic function of t with poles only at t = 1 and t = 2 and is regular at t = 0. τ ) obeys the so-called ‘heat equation’ A x h(x. y. Note also that ˜ ζ A (t. Using the integral representation for the -function (t) = 0 ∞ x t−1 exp(−x) dx check that the ζ A -function can be calculated as ζ A (t) = where h(x.5 The determinant of an operator A with real positive discrete and increasing without bound eigenvalues λn : A f n = λn f n can be deﬁned as det A = n λn = exp[−(dζ A /dt)(0)] where the so-called ζ -function.93). 13. τ ) The last equation can be used to regularize quantities such as + ˜ ϕn (x)γ5 ϕn (x) = lim Tr[γ5 ζ A (x.

7 Prove (13. 13. τ ) = for small τ . y. The heat kernel h(x. Wu & Zee 1984. Check that ˜ ζ ˆ 2 (x.134) starting with the anomalous divergence in Euclidean space 5 ∂ µ jµ = 2m ¯ γ5 − 2iQ(x) (Integrate over the Euclidean volume.494 13 Chiral anomalies with A = (D )2 . y. 0) = (1/16π 2 )a2 (x.) . x. Zumino. take the vacuum expectation value and then take the limit m → 0. τ ) then has the asymptotic expansion / (Nielsen.6 Determine the abelian and non-abelian chiral anomalies in 2n-dimensional spacetime (Frampton & Kephart 1983. R¨ mer & van Nieuwenhuizen 1978) o h(x. y)τ n 13. x) D (x − y)2 1 exp − 4τ 16π 2 τ 2 ∞ n=0 an (x. Grisaru. Leutwyler 1984).

2) under the chiral group splits then into massless Goldstone bosons and a massive σ = σ0 − v meson which transform under the unbroken subgroup SU (2) as a triplet and a singlet.14 Effective lagrangians 14. an inspection of the lagrangian (9.1 Non-linear realization of the symmetry group Non-linear σ -model To become familiar with the effective lagrangian technique it is useful to follow the original approach and to begin with the discussion of the SU (2) × SU (2)-invariant σ -model in which the chiral symmetry has been spontaneously broken: SUL (2) × SUR (2) → SUV (2). E m σ . Can we ﬁnd an effective low energy description in terms of the Goldstone bosons only which would be SU (2) × SU (2)-invariant? Our goal is to implement the SU (2) × SU (2) symmetry in such a way that the two irreducible multiplets.38) explicitly breaks the SU (2) × SU (2) invariance of the lagrangian. decoupling of the σ means setting σ0 = v and π this breaks the invariance since (π . The σ mass is proportional to the vacuum expectation value v and can be made arbitrarily heavy. Equivalently. However. for example. Imagine we are interested in the physics at low energy. respectively. σ0 ) does no longer transform as a quadruplet under SU (2) × SU (2). The quadruplet of real ﬁelds 0 = (π 1 π 2 π 3 σ0 )T transforming as (2. the triplet and the singlet of the unbroken SU (2) which are contained in the quadruplet 0 do not mix under general SU (2)× SU (2) 495 . Both the ππ ππ and the π πσ 2 couplings originate from the SU (2) × SU (2) invariant term (σ0 + π 2 )2 in the original lagrangian. the σ -exchange contribution to the low energy π π → π π scattering cannot be neglected when m σ → ∞. with the σ meson decoupled.38) shows that this does not happen: the ππσ coupling is proportional to v and therefore. It is natural to expect that the low energy region can be effectively described in terms of the Goldstone boson ﬁelds. The reason is that a naive decoupling of the σ ﬁeld by its elimination from the lagrangian (9. only. We consider the σ -model with bosons only.

Any chiral transformation on the vector 0 (x) can then be written as follows: π1 (x) 0 π (x) 0 (x) = exp(−iαa G a ) 0 (x) = 2 = exp[−iξa (ξ.2) (14.1) and (14. Tr[G j G i ] = δ ji and X a and Ya are broken and unbroken generators of SU (2)×SU (2). .1) where the X a s are broken generators in (14. σ0 (x))T in the originally deﬁned vacuum.5) among the degenerate set |α |α = exp(+iαa X a )|0 The corresponding transformation of the ﬁeld operators reads = exp(iαa X a ) exp(−iαa X a ) or in the matrix form = exp(−iαa X a ) (14.2) and their matrix π representation in (14. α)X a ] 0 0 π3 (x) σ (x) σ (x) 0 (14.3) a rotation of the vacuum by the angle ξ (x) sets the new Goldstone boson ﬁelds to zero locally at x. respectively.38) for the generators we can write 1 π (x) 0 −σ (ξ1 /ξ ) sin ξ π 2 (x) 0 −σ (ξ /ξ ) sin ξ 2 = exp(−iξa X a ) = 0 (x) = 3 π (x) 0 −σ (ξ3 /ξ ) sin ξ σ0 (x) where 2 2 2 ξ = (ξ1 + ξ2 + ξ3 )1/2 σ (x) σ cos ξ (14. Ya ).3).4) and 2 σ 2 (x) = π 2 (x) + σ0 (x) and therefore due to (14. In fact using the real representation (E.496 14 Effective lagrangians transformations. First of all we observe that at a given point x the Goldstone boson ﬁelds can always be set to zero by a suitable redeﬁnition of the vacuum. This can be achieved if we use the freedom of choice of the broken symmetry vacuum (Section 9. Let us take some 0 = (π (x).3) (14.5) where G a ≡ (X a .

ξa (x)s indeed transform linearly under the SUV (2). vacuum.e. every group element can be decomposed uniquely into a product of the form g = exp(iξa X a ) exp(iu a Ya ) (14. Since σ (x) is an SUV (2) singlet we suspect that in general a chiral transformation can be factorized into an unbroken-SU (2)-subgroup transformation (which in our case acts on the SU (2) singlet σ (x)) and a rotation of the local vacuum orientation. belong to the same representation as X a s and transform according to R T . Wess & Zumino 1969.14.6) R is the matrix representation of the unbroken SU (2) under which the matrices X a transform exp(−iu c Yc )X a exp(iu c Yc ) = Rab X b i. Thus. for the purpose of further considerations let us check the transformation properties of the ξa (x)s under the unbroken SU (2) isospin group 0 0 exp(−iu a Ya ) 0 = exp(−iu a Ya ) exp(−iξa X a ) exp(iu a Ya ) exp(−iu a Ya ) 0 0 0 = exp(−iξa Rab X b ) 0 = exp(−iξb X b ) σ Hence ξb = Rab ξa 0 0 0 σ σ (14. a general chiral transformation on 0 (x) can be represented ξ by the new angles ξa (ξ .4) because σ 2 (x) = π 2 (x) + σ0 (x) is invariant under chiral rotations.1 Non-linear realization of the symmetry group 497 2 with the same σ (x) as in (14. in some neighbourhood of the identity of G. From the properties of the exponentials it follows that. These results can be generalized as follows (Coleman.7) . However. connected. Coleman. The vector (0 0 0 σ (x))T refers now to the new. Wess & Zumino 1969). We consider a theory with some symmetry group G which is spontaneously broken to a subgroup H . Let G be a compact. rotated. before we generalize our discussion. Callan. α ) specifying the vacuum orientation which sets the transformed Goldstone boson ﬁelds locally to zero. semi-simple Lie group.

given by the vacuum-to-vacuum amplitude in the † The scalar ﬁelds ξ(x) are coordinates of the manifold of left cosets G/H at each point of space-time.9) ξ ξ and ξ (ξ . T 0 0 = 1. . We say that the ﬁelds (ξ.‡ Thus. Eq.11) Imagine that using the lagrangian (14. ) provide a non-linear realization of the group G. for example.11) we want to calculate matrix elements involving Goldstone bosons.10) ξ ξ = exp[−iξa (ξ . on the multiplet .7). After rescaling 0 /v → 0 we obtain the so-called non0 linear σ -model described by the lagrangian L= 1 2 f ∂ 2 π µ T µ 0∂ 0. introducing the external gauge ﬁelds by changing the ordinary derivatives acting on the linearly transforming ﬁelds 0 into covariant derivatives and expanding the generating functional. we can recast it as 0 (x) = exp[−iξa (x)X a ] (x) ≡ U (14. α )X a ] exp[−iu a (ξ .5) we observe that the ﬁeld σ (x) can be decoupled at E m σ in a chirally invariant manner by setting σ (x) = v or equivalently by imposing on the ﬁeld 0 (x) a chirally invariant constraint T 0 = v 2 . α ) are non-linear functions of ξ and α . α )Yi ] 0 (x) (14. Given the transformation rule (14.8) 0 (x)) (X a is the matrix representation of the broken generator X a appropriate for where under a general G transformation g(α) ξ ξ = ξ (ξ . Let us go back to the σ -model. Under transformations belonging to the unbroken subgroup H the ﬁelds ξi (x) transform linearly.7) means that once we have a parametrization l(ξ ) each group element g can be uniquely decomposed into a product g = l · h. g).498 14 Effective lagrangians where X a s and Ya s are broken and unbroken generators of the group G. where h ∈ H .6). The set of group elements l(ξ ) = exp(iξa X a ) parametrizes this manifold. The l is the representative member of the coset to which g belongs and h connects l to g within the coset.† If 0 is a ﬁeld which transforms according to a linear representation of G.7) = exp(−iαa G a ) 0 (x) = exp(−iαa G a ) exp[−iξa (x)X a ] (x) (14. Gasser & Leutwyler (1984)) and consists in gauging the full chiral group. where ξ = ξ (ξ. 2 fπ = v2 (14. respectively. according to the generalized (14. The ﬁrst method is more general (see. g) and h = h(ξ. We can work either with Green’s functions of currents or with the on-mass-shell matrix elements. Indeed. ‡ A product of g with an arbitrary group element and in particular with some l(ξ ) deﬁnes another l(ξ ) and an h according to g · l(ξ ) = l(ξ ) · h. α )Ya ] (x) since exp(−iα ·G) exp(−iζ · X ) belongs to G and can be decomposed as in (14. (14. α ) ξ = exp[−iu i (ξ . a general global transformation belonging to the group G is realized as a non-linear transformation on the ﬁelds ξa (x) and as a gauge (local) transformation. belonging to the unbroken H . using (14. α ) and u(ξ .

chiral multiplets split into H -multiplets with quite different masses in general.11) by freezing the σ degree of freedom in the linear σ -model.† 0 Thus we can think about 0 as representing the degrees of freedom of composite Goldstone bosons in the SU (2) × SU (2)-invariant QCD as well. Due to spontaneous symmetry breaking. and possibly other states also with well-deﬁned transformation properties under H . namely for exploiting its underlying chiral symmetry. Imagine we are interested in the low energy description of the strong interactions. There is also no reason to expect that the linear σ -model is a good approximation to the low energy QCD. one ﬁnds that beyond the tree level our theory has UV divergences which require counterterms of higher and higher order in derivatives of ﬁelds reﬂecting the non-renormalizability of the non-linear σ -model. However. † T∂ 0 µ 0 vanishes and T∂µ∂ µ 0 0 = −∂µ T∂µ 0 0. parity. if explicit breaking of chiral symmetry is taken into account). This happens at the expense of a set of new unknown constants at every level of the low energy expansion. We have arrived at the lagrangian (14. Working to order p 4 or higher we anyway have to supplement the lagrangian (14. The real virtue of the effective lagrangians is that they provide the most general systematic low energy expansion in the light particle sector. Its chiral symmetry is spontaneously broken into the isovector subgroup H . Such a redeﬁnition of the pion ﬁeld does not change the S-matrix elements if the redeﬁnition is local and leaves the origin of ﬁeld space invariant and as long as the new object has the right quantum numbers. Different deﬁnitions lead to different matrix elements off the mass-shell but they all give the same results on the mass-shell. If we are interested in the on-mass-shell matrix elements only. In this context the non-renormalizability of (14. Irrespective of the method. .11) is not a problem.1 Non-linear realization of the symmetry group 499 presence of external gauge ﬁelds. We are not able yet to get it directly from QCD in terms of its fundamental parameter (and quark masses. in powers of external ﬁelds. On the other hand the non-linear effective lagrangian is a tool for exploiting QCD as much as possible without solving the conﬁnement problem. This leads us to the following important comment on the physical signiﬁcance of the effective lagrangian.14. as the pion ﬁelds (Weinberg 1968. invariant under full symmetry G of the underlying theory. There are Goldstone bosons. one can convince oneself that this lagrangian is the most general one in order p 2 which is Lorentz-. Let us again consider QCD. We shall brieﬂy discuss this second method. however.11) with additional terms with free parameters. rather than the πa (x)s. Coleman. Wess & Zumino 1969). it is best to interpret the transforming non-linearly ξa (x)s.and chiral-invariant and can be constructed in terms of the real O(4) vector 0 of unit length T 0 = 1. composites of quark ﬁelds which form a multiplet under H .

α )Ya ])X c (14.16) (14.15) The transformation properties under G of both sides in (14. expand as follows: ∂µ (U ) = exp[−iξa (x)X a ]{i∂µ ξa (x)Dab (ξ )X b + [i∂µ ξa (x)E ab (ξ )Yb + ∂µ ] } (14. α )Yc ] Identifying (Dµ ξ )b = i∂µ ξa Dab (ξ ) ≡ Tr[U † ∂µU X b ] Dµ = [∂µ + i∂µ ξa E ab (ξ )Yb ] ≡ {∂µ + Tr[U † ∂µU Yb ]Yb } (14. α )Ya ] = Rbc (exp[−iu a (ξ .14) are the same as those of the 0 . The covariant derivative can be constructed explicitly by considering the derivative ∂µ 0 which transforms under G transformations in the same way as 0 does. respectively. Because of the constraint T 0 = 1 the form 0 (14.8) for a vector 0 we can. We can introduce the covariant derivative of the ﬁeld ξ(x).14) and correspondingly for ∂µ (U ) with ξa → ξa . . Using the non-linear notation (14. The functions u a (ξ. on general grounds. α )Yc ]) where the matrix Rba is deﬁned by the relation ξ ξ ξ exp[−iu a (ξ .9). They can be derived easily if we work in the ξa (x) basis.9). α )Dab (ξ )X b + (i∂µ ξa E ab (ξ )Yb + ∂µ )] ξ = exp(−iu c Yc )[i∂µ ξa Dab X b + (i∂µ ξa E ab Yb + ∂µ )] exp[iu c (ξ . E ab → E ab . α) are deﬁned by (14. where the functions D(ξ ) and E(ξ ) are deﬁned by† exp(iξa X a )∂µ exp(−iξa X a ) = i∂µ ξa [Dab (ξ )X b + E ab (ξ )Yb ] → (14.11) is not convenient for constructing the Feynman rules. so the object in the braces must be a multiplet of H and have the local gauge transformation law (14. . Dab → Dab .500 14 Effective lagrangians Effective lagrangian in the ξa (x) basis We return to the problem mentioned earlier of calculating the on-mass.17) † Those who are familiar with differential geometry will recognize that in the expansion (14.13) (14. deﬁned by its transformation properties under general transformation G(α) as follows ξ [(Dµ ξ )a (x)] ≡ (Dµ ξ )b Rba (exp[−iu c (ξ . α )Ya ]X b exp[iu a (ξ .12) and Ya and X a are representation matrices of unbroken and broken generators of G. Thus ξ [i∂µ ξa (ξ .shell matrix elements from effective lagrangians. in the representation to which the 0 belongs.15) Dab (ξ ) represents the vielbein in G/H and E ab (ξ ) is an H -connection.

15) in power series one can check the identity exp(iξa X a )∂µ exp(−iξa X a ) = ∞ n=0 (−i)n+1 [. Dµ and (Dµ ξ )b which is invariant under H transformations is also automatically invariant under G transformations.. icuwz ξw G z . ξw · X w ] (n + 1)! 1 2 . ξb · X b ]. n (14. G b ] = icabc G c is the G group algebra and G z is a broken generator X for n even. Similarly we also get ξ (Dµ ) = exp[−iu a (ξ .14.15). An important conclusion following from these transformation rules is that a function of .23) ab .. .1 Non-linear realization of the symmetry group 501 and equating coefﬁcients of X a we recover for the object (Dµ ξ )b the transformation law (14. expanding the l. . (14.. Dµ and (Dµ ξ )b all transform according to some ﬁnite matrix representations of the group H . .19) The commutators give ∂µ ξα · icabc ξb · iccde ξd . . ξd · X d ] .12)..h. . First.20) (−i)n+1 n ∂µ ξa (c · ξ )az G z (n + 1)! (14. Denoting icabc ξb = (c · ξ )ac one gets exp(iξa · X a )∂µ exp(−iξa · X a ) = and summing the terms even in n Dab (ξ ) = Similarly E ab (ξ ) = i(−1)m 1 − cos c · ξ 2m+1 (c · ξ )ab = i (2m + 2)! c·ξ m=0 ∞ ∞ n=0 (14.22) ab (14. and unbroken Y for n odd. n 1 2 where [G a . .s. The explicit form of the covariant derivative can be obtained from (14. α )Ya ]Dµ (14. [[∂µ ξa · X a . under general transformation G(α) functions .21) (−1)m+1 sin c · ξ 2m (c · ξ )ab = − (2m + 1)! c·ξ m=0 ∞ (14.18) Thus.

Dµ (x) and (Dµ ξ(x))a is automatically G-invariant. Any H -invariant form built of (x). there is no dependence on the ξ ﬁeld other than that present in the covariant derivative. . Any such dependence would break G invariance because one may always set ξ(x) to zero at any given x by rotating the vacuum. ¯ γµ D µ . where gab is given in terms of the vielbein Dab (ξ ) and is a G-invariant metric 2 on G/H . Matrix representation for Goldstone boson ﬁelds In QCD-like theories with chiral U (N )×U (N ) symmetry spontaneously broken to SUV (N ) × UV (1)‡ it is very convenient to collect the Goldstone boson ﬁelds into a unitary matrix U transforming linearly under chiral U (N )×U (N ) transformations.502 14 Effective lagrangians Using the transformation properties of the covariant derivative (Dµ ξ )a the effective lagrangian of the non-linear σ -model can be written down in the ξ(x) basis. ¯ γµ γ5 Y a (D µ ξ )a . Using the explicit form of the covariant derivative we can now easily read off the Feynman rules. The presence of the ﬁrst term reﬂects the fact that the chiral symmetry of the lagrangian tells us nothing about the fermion masses as the symmetry may be spontaneously broken. ). According to the previously described general approach a multiplet 0 transforming linearly under G is replaced by the ﬁelds (ξ(x). ‡ See Section 13. For deﬁniteness let us take to be a fermion ﬁeld and G to be the chiral symmetry spontaneously broken to the isospin subgroup H . . Let us consider a general matrix ˜ = ¯ Lj Ri (14.24) As expected. . It reads† L= 1 2 f (Dµ ξ )a (D µ ξ )a 2 π (14. We can extend the effective lagrangian description to include the interaction of Goldstone bosons with other ﬁelds which may be relevant for the effective low energy theory.2 for the discussion of the spontaneous breaking of UA (1). If we want to use the effective lagrangian beyond the tree approximation. terms of higher order in derivatives must be systematically included. The leading low energy terms are ¯ .25) 2 † Or L = 1 f π gab ∂µ ξ a ∂ µ ξ b . where 0 = exp[−iξa (x)X a ] providing a non-linear realization of the group G which is linear on the unbroken subgroup H . The second term contains the fermion–fermion–pion vertex and the effective gradient coupling is known as the Adler zero for the soft pion emission. .

N 2 − 1 and T 0 = 1 and T a . We check that any matrix ˜ written as a product of a unitary and hermitean matrix = U · h transforms under general chiral transformations as follows: † † † = URU hUL = URUUL UL hUL = U h (14. is not a Goldstone boson (see Section 13. 1.27) a UL.R are chiral fermion ﬁelds with ﬂavour i. The single component ﬁeld α0 (x) is related to the determinant of U det U = exp(iα0 ) (14. . Since (14. .26) R Matrices UR and UL act in the space of fermions the form and L. . 1.1 Non-linear realization of the symmetry group 503 i where L. The matrix = exp(2iαa Ta )1 ≡ U · 1 l l (14. 2. the unitary matrix U (x) contains N 2 ﬁelds U (x) = exp[ 1 iα0 (x)] exp[2iαa (x)Ta ] 3 (14. 2 The transformations under the unbroken vector group UV (N ) are represented by αL = αR and axial transformations by αL = −αR . l are the SU (N ) generators in the fermion representation with Tr[T a T b ] = 1 δ ab . . respectively. . and have (14. N 2 − 1. . .2). .87) is an ansatz for the spontaneous breaking SU (N ) × SU (N ) → SUV (N ). .14. For instance. N 2 − 1 and T a are hermitean and traceless.31) where a = 1. . Thus once is written in this product form any full UV (N ) multiplet can be decoupled from the matrix h in the chirally invariant way. due to the abelian anomaly. Imagine now that we take ˜ = 1 which according l to (9. Observe that now we always work with SU (N ) generators T a and do not need to introduce explicitly the broken and unbroken generators of the full chiral SU (N ) × SU (N ).32) . . Actually. . 3.29) is in the product form it provides us with the desired chirally invariant decoupling of all but Goldstone degrees of freedom.30) so U also transforms linearly and h undergoes only vector transformations. where ˜ = exp(iαa Ta ) ˜ exp(iαa Ta ) (14. Under chiral transformations ˜ transforms linearly and according to (9.R T a ) where a = 0. .85) † ˜ = UR ˜ UL (14.29) corresponds to a rotated vacuum and represents the degrees of freedom necessary to describe the Goldstone boson sector. . Thus a change of the vacuum orientation among the degenerate set corresponds to ˜ → ˜ . . in the case of U (3) × U (3) it includes the η degree of freedom which. a = 1.R = exp(−iαL. N 2 − 1. .28) where a = 0.

Because U is an unitary matrix UU † = 1. On the basis of arguments which are not discussed in this book such as the Okubo–Zweig–Iizuka rule and the large Nc (number of colours) limit of QCD (Veneziano 1979. Since U −1 ∂µU ≡ i∂µ αa Dab (αc )Tb ≡ (Dµ α)b Tb (14. .35) where Tb form a group.36) 2 = 1 f π Tr[U −1 ∂µU Ta ] Tr[(U −1 ∂ µU )† Ta ] 2 † 1 2 = 8 f π (Dµ α)b (D α)b Using the deﬁnition (14. Witten 1979. One can also easily check that αa s transform linearly under vector transformations UR = UL . βL . h) provide a non-linear realization of this chiral group.2 Effective lagrangians and anomalies In QCD and in models of composite quarks and leptons one starts from a fundamental lagrangian describing a non-abelian gauge theory with elementary .30) one can easily check that for general chiral transformations the ‘covariant derivative’ (Dµ α)a transforms only under global rotations of the unbroken SUV (N ). . the effective lagrangian can now be written in one of the following equivalent ways 2 L = 1 f π Tr[∂µU ∂ µU † ] 4 1 2 −1 −1 µ †