Springer Tracts in Modern Physics Volume 211 Managing Editor: G. H¨ohler, Karlsruhe Editors: J. K¨uhn, Karlsruhe Th. M¨u...

Author:
Vladimir A. Smirnov

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Springer Tracts in Modern Physics Volume 211 Managing Editor: G. H¨ohler, Karlsruhe Editors: J. K¨uhn, Karlsruhe Th. M¨uller, Karlsruhe A. Ruckenstein, New Jersey F. Steiner, Ulm J. Tr¨umper, Garching P. W¨olﬂe, Karlsruhe

Starting with Volume 165, Springer Tracts in Modern Physics is part of the [SpringerLink] service. For all customers with standing orders for Springer Tracts in Modern Physics we offer the full text in electronic form via [SpringerLink] free of charge. Please contact your librarian who can receive a password for free access to the full articles by registration at: springerlink.com If you do not have a standing order you can nevertheless browse online through the table of contents of the volumes and the abstracts of each article and perform a full text search. There you will also f ind more information about the series.

Springer Tracts in Modern Physics Springer Tracts in Modern Physics provides comprehensive and critical reviews of topics of current interest in physics. The following ﬁelds are emphasized: elementary particle physics, solid-state physics, complex systems, and fundamental astrophysics. Suitable reviews of other ﬁelds can also be accepted. The editors encourage prospective authors to correspond with them in advance of submitting an article. For reviews of topics belonging to the above mentioned ﬁelds, they should address the responsible editor, otherwise the managing editor. See also springeronline.com

Managing Editor

Solid-State Physics, Editors

Gerhard H¨ohler

Andrei Ruckenstein Editor for The Americas

Institut f¨ur Theoretische Teilchenphysik Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 33 75 Fax: +49 (7 21) 37 07 26 Email: [email protected] www-ttp.physik.uni-karlsruhe.de/

Elementary Particle Physics, Editors

Department of Physics and Astronomy Rutgers, The State University of New Jersey 136 Frelinghuysen Road Piscataway, NJ 08854-8019, USA Phone: +1 (732) 445 43 29 Fax: +1 (732) 445-43 43 Email: [email protected] www.physics.rutgers.edu/people/pips/ Ruckenstein.html

Johann H. K¨uhn

Peter W¨olﬂe

Institut f¨ur Theoretische Teilchenphysik Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 33 72 Fax: +49 (7 21) 37 07 26 Email: [email protected] www-ttp.physik.uni-karlsruhe.de/∼jk

Institut f¨ur Theorie der Kondensierten Materie Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 35 90 Fax: +49 (7 21) 6 08 77 79 Email: woelﬂ[email protected] www-tkm.physik.uni-karlsruhe.de

Thomas M¨uller Institut f¨ur Experimentelle Kernphysik Fakult¨at f¨ur Physik Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 35 24 Fax: +49 (7 21) 6 07 26 21 Email: [email protected] www-ekp.physik.uni-karlsruhe.de

Fundamental Astrophysics, Editor Joachim Tr¨umper Max-Planck-Institut f¨ur Extraterrestrische Physik Postfach 13 12 85741 Garching, Germany Phone: +49 (89) 30 00 35 59 Fax: +49 (89) 30 00 33 15 Email: [email protected] www.mpe-garching.mpg.de/index.html

Complex Systems, Editor Frank Steiner Abteilung Theoretische Physik Universit¨at Ulm Albert-Einstein-Allee 11 89069 Ulm, Germany Phone: +49 (7 31) 5 02 29 10 Fax: +49 (7 31) 5 02 29 24 Email: [email protected] www.physik.uni-ulm.de/theo/qc/group.html

Vladimir A. Smirnov

Evaluating Feynman Integrals With 48 Figures

123

Vladimir A. Smirnov II. Institut f¨ur Theoretische Physik Universit¨at Hamburg Luruper Chaussee 149 22761 Hamburg, Germany E-mail: [email protected]

Lomonosov Moscow State University Skobeltsyn Institute of Nuclear Physics Moscow 119992, Russia E-mail: [email protected]

Library of Congress Control Number: 2004115458

Physics and Astronomy Classiﬁcation Scheme (PACS): 12.38.Bx, 12.15.Lk, 02.30.Gp

ISSN print edition: 0081-3869 ISSN electronic edition: 1615-0430 ISBN 3-540-23933-2 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable for prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springeronline.com © Springer-Verlag Berlin Heidelberg 2004 Printed in Germany The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting: by the author and TechBooks using a Springer LATEX macro package Cover concept: eStudio Calamar Steinen Cover production: design &production GmbH, Heidelberg Printed on acid-free paper

SPIN: 10985380

56/3141/jl

543210

Preface

The goal of this book is to describe in detail how Feynman integrals1 can be evaluated analytically. The problem of evaluating Lorentz-covariant Feynman integrals over loop momenta originated in the early days of perturbative quantum ﬁeld theory. Over a span of more than ﬁfty years, a great variety of methods for evaluating Feynman integrals has been developed. This book is a ﬁrst attempt to summarize them. I understand that if another person – in particular one actively involved in developing methods for Feynman integral evaluation – made a similar attempt, he or she would probably concentrate on some other methods and would rank the methods as most important and less important in a diﬀerent order. I believe, however, that my choice is reasonable. At least I have tried to concentrate on the methods that have been used in the past few years in the most sophisticated calculations, in which world records in the Feynman integral ‘sport’ were achieved. The problem of evaluation is very important at the moment. What could be easily evaluated was evaluated many years ago. To perform important calculations at the two-loop level and higher one needs to choose adequate methods and combine them in a non-trivial way. In the present situation – which might be considered boring because the Standard Model works more or less properly and there are no glaring contradictions with experiment – one needs not only to organize new experiments but also perform rather nontrivial calculations for further crucial high-precision checks. So I hope very much that this book will be used as a textbook in practical calculations. I shall concentrate on analytical methods and only brieﬂy describe numerical ones. Some methods are also characterized as semi-analytical, for example, the method based on asymptotic expansions of Feynman integrals in momenta and masses which was described in detail in my previous book. In this method, it is also necessary to apply some analytical methods of evaluation which were described there only very brieﬂy. So the present book can be considered as Volume 1 with respect to the previous book, which might be termed Volume 2, or the sequel. 1

Let us point out from beginning that two kinds of integrals are associated with Feynman: integrals over loop momenta and path integrals. We will deal only with the former case.

VI

Preface

Although all the necessary deﬁnitions concerning Feynman integrals are provided in the book, it would be helpful for the reader to know the basics of perturbative quantum ﬁeld theory, e.g. by following the ﬁrst few chapters of the well-known textbooks by Bogoliubov and Shirkov and/or Peskin and Schroeder. This book is based on the course of lectures which I gave in the winter semester of 2003–2004 at the Universities of Hamburg and Karlsruhe as a DFG Mercator professor in Hamburg. It is my pleasure to thank the students, postgraduate students, postdoctoral fellows and professors who attended my lectures for numerous stimulating discussions. I am grateful very much to B. Feucht, A.G. Grozin and J. Piclum for careful reading of preliminary versions of the whole book and numerous comments and suggestions; to M. Czakon, M. Kalmykov, P. Mastrolia, J. Piclum, M. Steinhauser and O.L. Veretin for valuable assistance in presenting examples in the book; to C. Anastasiou, K.G. Chetyrkin and A.I. Davydychev for various instructive discussions; to P.A. Baikov, M. Beneke, K.G. Chetyrkin, A. Czarnecki, A.I. Davydychev, B. Feucht, G. Heinrich, A.A. Penin, A. Signer, M. Steinhauser and O.L. Veretin for fruitful collaboration on evaluating Feynman integrals; to M. Czakon, A. Czarnecki, T. Gehrmann, J. Gluza, T. Riemann, K. Melnikov, E. Remiddi and J.B. Tausk for stimulating competition; to Z. Bern, L. Dixon, C. Greub and S. Moch for various pieces of advice; and to B.A. Kniehl and J.H. K¨ uhn for permanent support. I am thankful to my family for permanent love, sympathy, patience and understanding. Moscow – Hamburg, October 2004

V.A. Smirnov

Contents

1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

Feynman Integrals: Basic Deﬁnitions and Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Feynman Rules and Feynman Integrals . . . . . . . . . . . . . . . . . . . . 2.2 Divergences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Alpha Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Properties of Dimensionally Regularized Feynman Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

4

1 8 9 11 11 14 18 20 24 29

Evaluating by Alpha and Feynman Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Simple One- and Two-Loop Formulae . . . . . . . . . . . . . . . . . . . . . 3.2 Auxiliary Tricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Recursively One-Loop Feynman Integrals . . . . . . . . . . . . 3.2.2 Partial Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Dealing with Numerators . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Feynman Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Two-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31 31 34 34 35 36 38 41 43 52

Evaluating by MB Representation . . . . . . . . . . . . . . . . . . . . . . . . 4.1 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Multiple MB Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 More One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Two-Loop Massless Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Two-Loop Massive Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Three-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 More Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.8 MB Representation versus Expansion by Regions . . . . . . . . . . .

55 56 63 65 71 81 92 98 102

VIII

Contents

4.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 5

IBP and Reduction to Master Integrals . . . . . . . . . . . . . . . . . . . 5.1 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Two-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Reduction of On-Shell Massless Double Boxes . . . . . . . . . . . . . . 5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

Reduction to Master Integrals by Baikov’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Basic Parametric Representation . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Constructing Coeﬃcient Functions. Simple Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 General Recipes. Complicated Examples . . . . . . . . . . . . . . . . . . . 6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

109 109 114 120 127 130 133 133 138 146 152 162 163

7

Evaluation by Diﬀerential Equations . . . . . . . . . . . . . . . . . . . . . . 7.1 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Two-Loop Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

A

Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 A.1 Table of Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 A.2 Some Useful Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

B

Some Special Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

C

Summation Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C.1 Some Number Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C.2 Power Series of Levels 3 and 4 in Terms of Polylogarithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C.3 Inverse Binomial Power Series up to Level 4 . . . . . . . . . . . . . . . C.4 Power Series of Levels 5 and 6 in Terms of HPL . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

D

165 165 170 173 176

191 192 197 198 200 204

Table of MB Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 D.1 MB Integrals with Four Gamma Functions . . . . . . . . . . . . . . . . . 207 D.2 MB Integrals with Six Gamma Functions . . . . . . . . . . . . . . . . . . 214

Contents

E

F

IX

Analysis of Convergence and Sector Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.1 Analysis of Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.2 Practical Sector Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

221 221 229 232

A Brief Review of Some Other Methods . . . . . . . . . . . . . . . . . . F.1 Dispersion Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.2 Gegenbauer Polynomial x-Space Technique . . . . . . . . . . . . . . . . F.3 Gluing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.4 Star-Triangle Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.5 IR Rearrangement and R∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.6 Diﬀerence Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.7 Experimental Mathematics and PSLQ . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

233 233 234 235 236 237 240 241 243

List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

1 Introduction

The important mathematical problem of evaluating Feynman integrals arises quite naturally in elementary-particle physics when one treats various quantities in the framework of perturbation theory. Usually, it turns out that a given quantum-ﬁeld amplitude that describes a process where particles participate cannot be completely treated in the perturbative way. However it also often turns out that the amplitude can be factorized in such a way that diﬀerent factors are responsible for contributions of diﬀerent scales. According to a factorization procedure a given amplitude can be represented as a product of factors some of which can be treated only non-perturbatively while others can be indeed evaluated within perturbation theory, i.e. expressed in terms of Feynman integrals over loop momenta. A useful way to perform the factorization procedure is provided by solving the problem of asymptotic expansion of Feynman integrals in the corresponding limit of momenta and masses that is determined by the given kinematical situation. A universal way to solve this problem is based on the so-called strategy of expansion by regions [3, 10]. This strategy can be itself regarded as a (semianalytical) method of evaluation of Feynman integrals according to which a given Feynman integral depending on several scales can be approximated, with increasing accuracy, by a ﬁnite sum of ﬁrst terms of the corresponding expansion, where each term is written as a product of factors depending on diﬀerent scales. A lot of details concerning expansions of Feynman integrals in various limits of momenta and/or masses can be found in my previous book [10]. In this book, however, we shall mainly deal with purely analytical methods. One needs to take into account various graphs that contribute to a given process. The number of graphs greatly increases when the number of loops gets large. For a given graph, the corresponding Feynman amplitude is represented as a Feynman integral over loop momenta, due to some Feynman rules. The Feynman integral, generally, has several Lorentz indices. The standard way to handle tensor quantities is to perform a tensor reduction that enables us to write the given quantity as a linear combination of tensor monomials with scalar coeﬃcients. Therefore we shall imply that we deal with scalar Feynman integrals and consider only them in examples. A given Feynman graph therefore generates various scalar Feynman integrals that have the same structure of the integrand with various distributions V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 1–9 (2004) c Springer-Verlag Berlin Heidelberg 2004

2

1 Introduction

of powers of propagators (indices). Let us observe that some powers can be negative, due to some initial polynomial in the numerator of the Feynman integral. A straightforward strategy is to evaluate, by some methods, every scalar Feynman integral resulting from the given graph. If the number of these integrals is small this strategy is quite reasonable. In non-trivial situations, where the number of diﬀerent scalar integrals can be at the level of hundreds and thousands, this strategy looks too complicated. A well-known optimal strategy here is to derive, without calculation, and then apply some relations between the given family of Feynman integrals as recurrence relations. A well-known standard way to obtain such relations is provided by the method of integration by parts1 (IBP) [6] which is based on putting to zero any integral of the form ∂f dd k1 dd k2 . . . µ ∂ki over loop momenta k1 , k2 , . . . , ki , . . . within dimensional regularization with the space-time dimension d = 4 − 2ε as a regularization parameter [4, 5, 7]. Here f is an integrand of a Feynman integral; it depends on the loop and external momenta. More precisely, one tries to use IBP relations in order to express a general dimensionally regularized integral from the given family as a linear combination of some irreducible integrals which are also called master integrals. Therefore the whole problem decomposes into two parts: a solution of the reduction procedure and the evaluation of the master Feynman integrals. Observe that in such complicated situations, with the great variety of relevant scalar integrals, one really needs to know a complete solution of the recursion problem, i.e. to learn how an arbitrary integral with general integer powers of the propagators and powers of irreducible monomials in the numerator can be evaluated. To illustrate the methods of evaluation that we are going to study in this book let us ﬁrst orient ourselves at the evaluation of individual Feynman integrals, which might be master integrals, and take the simple scalar oneloop graph Γ shown in Fig. 1.1 as an example. The corresponding Feynman integral constructed with scalar propagators is written as dd k 2 2 . (1.1) FΓ (q , m ; d) = (k 2 − m2 )(q − k)2 1

As is explained in textbooks b on integral calculus, b the method of IBP is applied with the help of the relation a dxuv = uv|ba − a dxu v as follows. One tries to represent the integrand as uv with some u and v in such a way that the integral on the right-hand side, i.e. of u v will be simpler. We do not follow this idea in the case of Feynman integrals. Instead we only use the fact that an integral of the derivative of some function is zero, i.e. we always neglect the corresponding surface terms. So the name of the method looks misleading. It is however unambiguously accepted in the physics community.

1 Introduction

3

Fig. 1.1. One-loop self-energy graph. The dashed line denotes a massless propagator

The same picture Fig. 1.1 can also denote the Feynman integral with general powers of the two propagators, dd k . (1.2) FΓ (q 2 , m2 ; a1 , a2 , d) = 2 2 (k − m )a1 [(q − k)2 ]a2 Suppose, one needs to evaluate the Feynman integral FΓ (q 2 , m2 ; 2, 1, d) ≡ F (2, 1, d) which is ﬁnite in four dimensions, d = 4. (It can also be depicted by Fig. 1.1 with a dot on the massive line.) There is a lot of ways to evaluate it. For example, a straightforward way is to take into account the fact that the given function of q is Lorentz-invariant so that it depends on the external momentum through its square, q 2 . One can choose a frame q = (q0 , 0), introduce spherical coordinates for k, integrate over angles, then over the radial component and, ﬁnally, over k0 . This strategy can be, however, hardly generalized to multi-loop2 Feynman integrals. Another way is to use a dispersion relation that expresses Feynman integrals in terms of a one-dimensional integral of the imaginary part of the given Feynman integral, from the value of the lowest threshold to inﬁnity. This dispersion integral can be expressed by means of the well-known Cutkosky rules. We shall not apply this method, which was, however, very popular in the early days of perturbative quantum ﬁeld theory, and only brieﬂy comment on it in Appendix F. Let us now turn to the methods that will be indeed actively used in this book. To illustrate them all let me use this very example of Feynman integrals (1.2) and present main ideas of these methods, with the obligation to present the methods in great details in the rest of the book. First, we will exploit the well-known technique of alpha or Feynman parameters. In the case of F (2, 1, d), one writes down the following Feynmanparametric formula: 1 ξdξ 1 = 2 . (1.3) 2 2 2 2 2 2 2 3 (k − m ) (q − k) 0 [(k − m )ξ + (1 − ξ)(q − k) + i0] Then one can change the order of integration over ξ and k, perform integration over k with the help of the formula (A.1) (which we will derive in Chap. 3) and obtain the following representation: 1 dξ ξ −ε d/2 F (2, 1, d) = −iπ Γ (1 + ε) . (1.4) 2 2 1+ε 0 [m − q (1 − ξ) − i0] 2

Since the Feynman integrals are rather complicated objects the word ‘multiloop’ means the number of loops greater than one ;-)

4

1 Introduction

This integral is easily evaluated at d = 4 with the following result: ln 1 − q 2 /m2 . F (2, 1, 4) = iπ 2 q2

(1.5)

In principle, any given Feynman integral F (a1 , a2 , d) with concrete numbers a1 and a2 can similarly be evaluated by Feynman parameters. In particular, F (1, 1, d) reduces to 1 dξ ξ −ε d/2 F (1, 1, d) = iπ Γ (ε) . (1.6) 2 2 ε 0 [m − q (1 − ξ) − i0] There is an ultraviolet (UV) divergence which manifests itself in the ﬁrst pole of the function Γ (ε), i.e. at d = 4. The integral can be evaluated in expansion in a Laurent series in ε, for example, up to ε0 : 1 − ln m2 + 2 F (1, 1, d) = iπ d/2 e−γE ε ε m2 q2 − 1 − 2 ln 1 − 2 + O(ε) , (1.7) q m where γE is Euler’s constant. We shall study the method of Feynman and alpha parameters in Chap. 3. Another method which plays an essential role in this book is based on the Mellin–Barnes (MB) representation. The underlying idea is to replace a sum of terms raised to some power by the product of these terms raised to certain powers, at the cost of introducing an auxiliary integration that goes from −i∞ to +i∞ in the complex plane. The most natural way to apply this representation is to write down a massive propagator in terms of massless ones. For F (2, 1, 4), we can write +i∞ 1 (m2 )z 1 = dz Γ (2 + z)Γ (−z) . (1.8) 2 2 2 (m − k ) 2πi −i∞ (−k 2 )2+z Applying (1.8) to the ﬁrst propagator in (1.2), changing the order of integration over k and z and evaluating the internal integral over k by means of the one-loop formula (A.7) (which we will derive in Chap. 3) we arrive at the following onefold MB integral representation: +i∞ 2 z m iπ d/2 Γ (1 − ε) 1 dz F (2, 1, d) = − (−q 2 )1+ε 2πi −i∞ −q 2 Γ (1 + ε + z)Γ (−ε − z)Γ (−z) . (1.9) × Γ (1 − 2ε − z) The contour of integration is chosen in the standard way: the poles with a Γ (. . . + z) dependence are to the left of the contour and the poles with a Γ (. . . − z) dependence are to the right of it. If |ε| is small enough we can choose this contour as a straight line parallel to the imaginary axis with −1 < Rez < 0. For d = 4, we obtain

1 Introduction

F (2, 1, 4) = −

iπ 2 1 q 2 2πi

+i∞

dz −i∞

m2 −q 2

5

z Γ (z)Γ (−z) .

(1.10)

By closing the integration contour to the right and taking a series of residues at the points z = 0, 1, . . ., we reproduce (1.5). Using the same technique, any integral from the given family can similarly be evaluated. We shall study the technique of MB representation in Chap. 4 where we shall see, through various examples, how, by introducing MB integrations in an appropriate way, one can analytically evaluate rather complicated Feynman integrals. Let us, however, think about a more economical strategy based on IBP relations which would enable us to evaluate any integral (1.2) as a linear combination of some master integrals. Putting to zero dimensionally regularized ∂ ∂ ·kf (a1 , a2 ) and q· ∂k f (a1 , a2 ), where f (a1 , a2 ) is the integrand integrals of ∂k in (1.2), and writing down obtained relations in terms of integrals of the given family we obtain the following two IBP relations: d − 2a1 − a2 − 2m2 a1 1+ − a2 2+ (1− − q 2 + m2 ) = 0 , −

−

a2 − a1 − a1 1 (q + m − 2 ) − a2 2 (1 − q + m ) = 0 , +

2

2

+

2

2

(1.11) (1.12)

in the sense that they are applied to the general integral F (a1 , a2 ). Here the standard notation for increasing and lowering operators has been used, e.g. 1+ 2− F (a1 , a2 ) = F (a1 + 1, a2 − 1). Let us observe that any integral with a1 ≤ 0 is zero because it is a massless tadpole which is naturally put to zero within dimensional regularization. Moreover, any integral with a2 ≤ 0 can be evaluated in terms of gamma functions for general d with the help of (A.3) (which we will derive in Chap. 3). The number a2 can be reduced either to one or to a non-positive value using the following relation which is obtained as the diﬀerence of (1.11) multiplied by q 2 + m2 and (1.12) multiplied by 2m2 : (q 2 − m2 )2 a2 2+ = (q 2 − m2 )a2 1− 2+ −(d − 2a1 − a2 )q 2 − (d − 3a2 )m2 + 2m2 a1 1+ 2− . (1.13) Indeed, when the left-hand side of (1.13) is applied to F (a1 , a2 ), we obtain integrals with reduced a2 or, due to the ﬁrst term on the right-hand side, reduced a1 . Suppose now that a2 = 1. Then we can use the diﬀerence of relations (1.11) and (1.12), d − a1 − 2a2 − a1 1+ (2− − q 2 + m2 ) = 0 ,

(1.14)

by writing down a1 (q 2 − m2 )1+ through the rest terms, and reduce the index a1 to one or the index a2 to zero. We see that we can now express any integral of the given family as a linear combination of the integral F (1, 1) and simple integrals with a2 ≤ 0 which can be evaluated for general d in terms of gamma functions. In particular, we have

6

1 Introduction

F (2, 1) =

m2

1 [(1 − 2ε)F (1, 1) − F (2, 0)] . − q2

(1.15)

At this point, we can stop our activity because we have already essentially solved the problem. In fact, we shall later encounter several examples of non-trivial calculations where any integral is expressed in terms of some complicated master integrals and families of simple integrals. However, mathematically (and aesthetically), it is natural to be more curious and wonder about the minimal number of master integrals which form a linearly independent basis in the family of integrals F (a1 , a2 ). We will do this in Chaps. 5 and 6. In Chap. 5, we shall investigate various examples, starting from simple ones, where the reduction of a given class of Feynman integrals can be performed by solving IBP recurrence relations. If we want to be maximalists, i.e. we are oriented at the minimal number of master integrals, we expect that any Feynman integral from a given family, F (a1 , a2 , . . .) can be expressed linearly in terms of a ﬁnite set of master integrals: ci (F (a1 , a2 , . . .))Ii , (1.16) F (a1 , a2 , . . .) = i

These master integrals Ii cannot be reduced further, i.e. expressed as linear combinations of other Feynman integrals of the given family. There were several attempts to systematize the procedure of solving IBP recurrence relations. Some of them will be described in the end of Chap. 5. One of the corresponding methods [1, 2, 11] is based on an appropriate parametric representation which is used to construct the coeﬃcient functions ci (F (a1 , a2 , . . .)) ≡ ci (a1 , a2 , . . .) in (1.16). The integrand of this representai , where the integration parameters tion consists of the standard factors x−a i xi correspond to the denominators of the propagators, and a polynomial in these variables raised to the power (d − h − 1)/2, where h is the number of loops for vacuum integrals and some eﬀective loop number, otherwise. This polynomial is constructed for the given family of integrals according to some simple rules. An important property of such a representation is that it automatically satisﬁes IBP relations written for this family of integrals, provided one can use IBP in this parametric representation. For example, for the family of integrals F (a1 , a2 ) we are dealing with in this chapter, the auxiliary representation takes the form dx1 dx2 [P (x1 , x2 )](d−3)/2 , (1.17) ci (a1 , a2 ) ∼ xa1 1 xa2 2 with the basic polynomial P (x1 , x2 ) = −(x1 − x2 + m2 )2 − q 2 (q 2 − 2m2 − 2(x1 + x2 )) .

(1.18)

As we shall see in Chap. 6, such auxiliary representation provides the possibility to characterize the master integrals and construct algorithms for the evaluation of the corresponding coeﬃcient functions. When looking for

1 Introduction

7

candidates for the master integrals one considers integrals of the type (1.17) with indices ai equal to one or zero and tries to see whether such integrals can be understood non-trivially. According to a general rule, which we will explain in Chap. 6, the value ai = 1 of some index forces us to understand the integration over the corresponding parameter xi as a Cauchy integration contour around the origin in the complex xi -plane which in turn reduces to taking derivatives of the factor P (d−3)/2 in xi at xi = 0. If an index ai is equal to zero one has to understand the corresponding integration in some sense, which implies the validity of IBP in the integration over xi . In our present example, let us therefore consider the candidates F (1, 1), F (1, 0), F (0, 1) and F (0, 0). Of course, we neglect the last two of them because they are massless tadpoles. Thus we are left with the ﬁrst two integrals. According to the rule formulated above, the coeﬃcient function of F (1, 1) is evaluated as an iterated Cauchy integral over x1 and x2 . It is therefore constructed in a non-trivial (non-zero) way and this integral is recognized as a master integral. For F (1, 0), only the integration over x1 is understood as a Cauchy integration, and the representation (1.17) gives, for the corresponding coeﬃcient function, a linear combination of terms (d−3)/2−l dx2

−(m2 − q 2 )2 + 2(m2 + q 2 )x2 − x22 , (1.19) j x2 with integer j and non-negative integer l. When j ≤ 0, the integration can be taken between the roots of the quadratic polynomial in the square brackets. Thus one can again construct a non-zero coeﬃcient function and the integral F (1, 0) turns out to be our second (and the last) master integral. We shall see in Chap. 6 how (1.17) can be understood for j > 0; this is indeed necessary for the construction of the coeﬃcient function c2 (a1 , a2 ) at a2 > 0. We shall also learn other details of this method illustrated though various examples. Anyway, the present example shows that this method enables an elegant and transparent classiﬁcation of the master integrals: the presence of (only two) master integrals F (1, 1) and F (1, 0) in the given recursion problem is seen in a very simple way, as compared with the complete solution of the reduction procedure outlined above. One more powerful method that has been proven very useful in the evaluation of the master integrals is based on using diﬀerential equations (DE) [8, 9]. Let us illustrate it again with the help of our favourite example. To evaluate the master integral F (1, 1) let us observe that its in m2 derivative 2 2 2 2 is nothing but F (2, 1) (because ∂/(∂m ) 1/(k − m ) = 1/(k − m2 )2 ) which is expressed, according to our reduction procedure, by (1.15). Therefore we arrive at the following diﬀerential equation for f (m2 ) = F (1, 1):

∂ 1 (1 − 2ε)f (m2 ) − F (2, 0) , f (m2 ) = 2 ∂m2 m − q2

(1.20)

where the quantity F (2, 0) is a simpler object because it can be evaluated in terms of gamma functions for general ε. The general solution to this equation

8

1 Introduction

can easily be obtained by the method of the variation of the constant, with ﬁxing the general solution from the boundary condition at m = 0. Eventually, the above result (1.7) can successfully be reproduced. As we shall see in Chap. 7, the strategy of the method of DE in much more non-trivial situations is similar: one takes derivatives of a master integral in some arguments, expresses them in terms of original Feynman integrals, by means of some variant of solution of IBP relations, and solves resulting diﬀerential equations. However, before studying the methods of evaluation, basic deﬁnitions are presented in Chap. 2 where tools for dealing with Feynman integrals are also introduced. Methods for evaluating individual Feynman integrals are studied in Chaps. 3, 4 and 7 and the reduction problem is studied in Chaps. 5 and 6. In Appendix A, one can ﬁnd a table of basic one-loop and two-loop Feynman integrals as well as some useful auxiliary formulae. Appendix B contains deﬁnitions and properties of special functions that are used in this book. A table of summation formulae for onefold series is given in Appendix C. In Appendix D, a table of onefold MB integrals is presented. Appendix E contains analysis of convergence of Feynman integrals as well a description of a numerical method of evaluating Feynman integrals based on sector decompositions. Some other methods are brieﬂy characterized in Appendix F. These are mainly old methods whose details can be found in the literature. If I do not present some methods, this means that either I do not know about them, or I do not know physically important situations where they work not worse than than the methods I present. I shall use almost the same examples in Chaps. 3–7 and Appendix F to illustrate all the methods. On the one hand, this will be done in order to have the possibility to compare them. On the other hand, the methods often work together: for example, MB representation can be used in alpha or Feynman parametric integrals, the method of DE requires a solution of the reduction problem, boundary conditions within the method of DE can be obtained by means of the method of MB representation, auxiliary IBP relations within the method described in Chap. 6 can be solved by means of an algorithm originated within another approach to solving IBP relations. Basic notational conventions are presented below. The notation is described in more detail in the List of Symbols. In the Index, one can ﬁnd numbers of pages where deﬁnitions of basic notions are introduced.

1.1 Notation We use Greek and Roman letters for four-indices and spatial indices, respectively: xµ = (x0 , x) ,

References

9

q·x = q 0 x0 − q·x ≡ gµν q µ xν . The parameter of dimensional regularization is d = 4 − 2ε . The d-dimensional Fourier transform and its inverse are deﬁned as ˜ f (q) = dd x eiq·x f (x) , 1 f (x) = dd q e−ix·q f˜(q) . (2π)d In order to avoid Euler’s constant γE in Laurent expansions in ε, we pull out the factor e−γE ε per loop.

References 1. P.A. Baikov, Phys. Lett. B 385 (1996) 404; Nucl. Instrum. Methods A 389 (1997) 347. 6 2. P.A. Baikov and M. Steinhauser, Comput. Phys. Commun. 115 (1998) 161. 6 3. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 1 4. C.G. Bollini and J.J. Giambiagi, Nuovo Cim. B 12 (1972) 20. 2 5. P. Breitenlohner and D. Maison, Commun. Math. Phys. 52 (1977) 11, 39, 55. 2 6. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 2 7. G. ’t Hooft and M. Veltman, Nucl. Phys. B 44 (1972) 189. 2 8. A.V. Kotikov, Phys. Lett. B 254 (1991) 158; B 259 (1991) 314; B 267 (1991) 123. 7 9. E. Remiddi, Nuovo Cim. A 110 (1997) 1435. 7 10. V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 1 11. V.A. Smirnov and M. Steinhauser, Nucl. Phys. B 672 (2003) 199. 6

2 Feynman Integrals: Basic Deﬁnitions and Tools

In this chapter, basic deﬁnitions for Feynman integrals are given, ultraviolet (UV), infrared (IR) and collinear divergences are characterized, and basic tools such as alpha parameters are presented. Various kinds of regularizations, in particular dimensional one, are presented and properties of dimensionally regularized Feynman integrals are formulated and discussed.

2.1 Feynman Rules and Feynman Integrals In perturbation theory, any quantum ﬁeld model is characterized by a Lagrangian, which is represented as a sum of a free-ﬁeld part and an interaction part, L = L0 + LI . Amplitudes of the model, e.g. S-matrix elements and matrix elements of composite operators, are represented as power series in coupling constants. Starting from the S-matrix represented in terms of the time-ordered exponent of the interaction Lagrangian which is expanded with the application of the Wick theorem, or from Green functions written in terms of a functional integral treated in the perturbative way, one obtains that, in a ﬁxed perturbation order, the amplitudes are written as ﬁnite sums of Feynman diagrams which are constructed according to Feynman rules: lines correspond to L0 and vertices are determined by LI . The basic building block of the Feynman diagrams is the propagator that enters the relation T φi (x1 )φi (x2 ) = : φi (x1 )φi (x2 ) : +DF,i (x1 − x2 ) .

(2.1)

Here DF,i is the Feynman propagator of the ﬁeld of type i and the colons denote a normal product of the free ﬁelds. The Fourier transforms of the propagators have the form iZi (p) ˜ F,i (p) ≡ d4 x eip·x DF,i (x) = D , (2.2) 2 (p − m2i + i0)ai where mi is the corresponding mass, Zi is a polynomial and ai = 1 or 2 (for the gluon propagator in the general covariant gauge). The powers of the propagators al will be also called indices. For the propagator of the scalar ﬁeld, we have Z = 1, a = 1. This is not the most general form of the propagator. For example, in the axial or Coulomb gauge, the gluon propagator has another form. We usually omit the causal i0 for brevity. Polynomials V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 11–30 (2004) c Springer-Verlag Berlin Heidelberg 2004

12

2 Feynman Integrals: Basic Deﬁnitions and Tools

associated with vertices of graphs can be taken into account by means of the polynomials Zl . We also omit the factors of i and (2π)4 that enter in the standard Feynman rules (in particular, in (2.2)); these can be included at the end of a calculation. Eventually, we obtain, for any ﬁxed perturbation order, a sum of Feynman amplitudes labelled by Feynman graphs1 constructed from the given type of vertices and lines. In the commonly accepted physical slang, the graph, the corresponding Feynman amplitude and the integral are all often called the ‘diagram’. A Feynman graph diﬀers from a graph by distinguishing a subset of vertices which are called external. The external momenta or coordinates on which a Feynman integral depends are associated with the external vertices. Thus quantities that can be computed perturbatively are written, in any given order of perturbation theory, through a sum over Feynman graphs. For a given graph Γ , the corresponding Feynman amplitude

qi FΓ (q1 , . . . , qn ) (2.3) GΓ (q1 , . . . , qn+1 ) = (2π)4 i δ i

can be written in terms of an integral over loop momenta L ˜ F,l (pl ) , D FΓ (q1 , . . . , qn ) = d4 k1 . . . d4 kh

(2.4)

l=1

where d4 ki = dki0 dki , and a factor with a power of 2π is omitted, as we have agreed. The Feynman integral FΓ depends on n linearly independent external momenta qi = (qi0 , q i ); the corresponding integrand is a function of L internal momenta pi , which are certain linear combinations of the external momenta and h = L − V + 1 chosen loop momenta ki , where L, V and h are numbers of lines, vertices and (independent) loops, respectively, of the given graph. After some tensor reduction2 one can deal only with scalar Feynman integrals. To do this, various projectors can be applied. For example, in the case of Feynman integrals contributing to the electromagnetic formfactor (see Fig. 2.1) Γ µ (p1 , p2 ) = γ µ F1 (q 2 ) + σ µν qν F2 (q 2 ), where q = p1 − p2 , γ µ and σ µν are γ- and σ-matrices, respectively, the following projector can be applied to extract scalar integrals which contribute to the formfactor F1 in the massless case (with F2 = 0): 1 When dealing with graphs and Feynman integrals one usually does not bother about the mathematical deﬁnition of the graph and thinks about something that is built of lines and vertices. So, a graph is an ordered family {V, L, π± }, where V is the set of vertices, L is the set of lines, and π± : L → V are two mappings that correspond the initial and the ﬁnal vertex of a line. By the way, mathematicians use the word ‘edge’, rather than ‘line’. 2 In one-loop, the well-known general reduction was described in [23]. Steps towards systematical reduction at the two-loop level were made in [1].

2.1 Feynman Rules and Feynman Integrals

13

p1 q µ p2 Fig. 2.1. Electromagnetic formfactor

F1 (q 2 ) =

Tr [γµ p2 Γ µ (p1 , p2 ) p1 ] , 2(d − 2) q 2

(2.5)

where p = γ µ pµ and d is the parameter of dimensional regularization (to be discussed shortly in Sect. 2.4). Anyway, after applying some projectors, one obtains, for a given graph, a family of Feynman integrals which have various powers of the scalar parts of the propagators, 1/(p2l −m2l )al , and various monomials in the numerator. The denominators p2l can be expressed linearly in terms of scalar products of the loop and external momenta. The factors in the numerator can also be chosen as quadratic polynomials of the loop and external momenta raised to some powers. It is convenient to consider both types of the quadratic polynomials on the same footing and treat the factors in the numerators as extra factors in the denominator raised to negative powers. The set of the denominators for a given graph is linearly independent. It is natural to complete this set by similar factors coming from the numerator in such a way that the whole set will be linearly independent. Therefore we come to the following family of scalar integrals generated by the given graph: 4 d k1 . . . d4 kh (2.6) F (a1 , . . . , aN ) = · · · aN , E1a1 . . . EN where ki , i = 1, . . . , h, are loop momenta, ai are integer indices, and the denominators are given by 2 Aij (2.7) Er = r p i · pj − mr , i≥j≥1

with r = 1, . . . , N . The momenta pi are either the loop momenta pi = ki , i = 1, . . . , h, or independent external momenta ph+1 , . . . , ph+n of the graph. For a usual Feynman graph, the denominators Er determined by some matrix A are indeed quadratic. However, a more general class of Feynman integrals where the denominators are linear with respect to the loop and/or external momenta also often appears in practical calculations. Linear denominators usually appear in asymptotic expansions of Feynman integrals within the strategy of expansion by regions [2, 29]. Such expansions provide a useful link of an initial theory described by some Lagrangian with various eﬀective theories where, indeed, the denominators of propagators can be linear with

14

2 Feynman Integrals: Basic Deﬁnitions and Tools

respect to the external and loop momenta. For example, one encounters the following denominators: p · k, with an external momentum p on the light cone, p2 = 0, for the Sudakov limit and with p2 = 0 for the quark propagator of HQET [14, 19, 22]. Some non-relativistic propagators appear within threshold expansion and in the eﬀective theory called NRQCD [4, 18, 35], for example, the denominator k0 − k2 /(2m).

2.2 Divergences As has been known from early days of quantum ﬁeld theory, Feynman integrals suﬀer from divergences. This word means that, taken naively, these integrals are ill-deﬁned because the integrals over the loop momenta generally diverge. The ultraviolet (UV) divergences manifest themselves through a divergence of the Feynman integrals at large loop momenta. Consider, for example, the Feynman integral corresponding to the one-loop graph Γ of Fig. 2.2 with scalar propagators. This integral can be written as d4 k , (2.8) FΓ (q) = 2 (k 2 − m1 )[(q − k)2 − m22 ] where the loop momentum k is chosen as the momentum of the ﬁrst line. Introducing four-dimensional (generalized) spherical coordinates k = rkˆ in (2.8), where kˆ is on the unit (generalized) sphere and is expressed by means of three angles, and counting powers of propagators, we obtain, in the limit of ∞ large r, the following divergent behaviour: Λ dr r−1 . For a general diagram, a similar power counting at large values of the loop momenta gives 4h(Γ ) − 1 from the Jacobian that arises when one introduces generalized spherical coordinates in the (4 × h)-dimensional space of h loop four-momenta, plus a contribution from the powers of the ∞propagators and the degrees of its polynomials, and leads to an integral Λ dr rω−1 , where nl (2.9) ω = 4h − 2L + l

is the (UV) degree of divergence of the graph. (Here nl are the degrees of the polynomials Zl .)

Fig. 2.2. One-loop self-energy diagram

2.2 Divergences

15

This estimate shows that the Feynman integral is UV convergent overall (no divergences arise from the region where all the loop momenta are large) if the degree of divergence is negative. We say that the Feynman integral has a logarithmic, linear, quadratic, etc. overall divergence when ω = 0, 1, 2, . . ., respectively. To ensure a complete absence of UV divergences it is necessary to check convergence in various regions where some of the loop momenta become large, i.e. to satisfy the relation ω(γ) < 0 for all the subgraphs γ of the graph. We call a subgraph UV divergent if ω(γ) ≥ 0. In fact, it is suﬃcient to check these inequalities only for one-particle-irreducible (1PI) subgraphs (which cannot be made disconnected by cutting a line). It turns out that these rough estimates are indeed true – see some details in Sect. E.1. If we turn from momentum space integrals to some other representation of Feynman diagrams, the UV divergences will manifest themselves in other ways. For example, in coordinate space, the Feynman amplitude (i.e. the inverse Fourier transform of (2.3)) is expressed in terms of a product of the Fourier transforms of propagators L

DF,l (xli − xlf )

(2.10)

l=1

integrated over four-coordinates xi corresponding to the internal vertices. Here li and lf are the beginning and the end, respectively, of a line l. The propagators in coordinate space, 1 ˜ F,l (p)e−ix·p , DF,l (x) = d4 p D (2.11) (2π)4 are singular at small values of coordinates x = (x0 , x). To reveal this singularity explicitly let us write down the propagator (2.2) in terms of an integral over a so-called alpha-parameter 2 2 1 ∂ (−i)al ∞ 2iul ·p ˜ DF,l (p) = i Zl dαl αlal −1 ei(p −m )αl . e 2i ∂ul Γ (al ) ul =0

0

(2.12) which turns out to be a very useful tool both in theoretical analyses and practical calculations. To present an explicit formula for the scalar (i.e. for a = 1 and Z = 1) propagator ∞ 2 2 ˜ DF (p) = dα ei(p −m )α (2.13) 0

in coordinate space we insert (2.13) into (2.11), change the order of integration over p and α and take the Gaussian integrations explicitly using the formula 2 2 d4 k ei(αk −2q·k) = −iπ 2 α−2 e−iq /α , (2.14)

16

2 Feynman Integrals: Basic Deﬁnitions and Tools

which is nothing but a product of four one-dimensional Gaussian integrals: ∞ 2 π −iq02 /α+iπ/4 e dk0 ei(αk0 −2q0 k0 ) = , α −∞ ∞ π iqj2 /α−iπ/4 −i(αkj2 −2qj kj ) e dkj e = , j = 1, 2, 3 (2.15) α −∞ (without summation over j in the last formula). The ﬁnal integration is then performed using [26] or in MATHEMATICA [37] with the following result: m √ DF (x) = K1 m −x2 + i0 4π 2 −x2 + i0 1 1 + O m2 ln m2 , (2.16) =− 2 2 4π x − i0 where K1 is a Bessel special function [12]. The leading singularity at x = 0 is given by the value of the coordinate space massless propagator. Thus, the inverse Fourier transform of the convolution integral (2.8) equals the square of the coordinate-space scalar propagator, with the singularity (x2 − i0)−2 . Power-counting shows that this singularity produces integrals that are divergent in the vicinity of the point x = 0, and this is the coordinate space manifestation of the UV divergence. The divergences caused by singularities at small loop momenta are called infrared (IR) divergences. First we distinguish IR divergences that arise at general values of the external momenta. A typical example of such a divergence is given by the graph of Fig. 2.2 when one of the lines contains the second power of the corresponding propagator, so that a1 = 2. If the mass of this line is zero we obtain a factor 1/(k 2 )2 in the integrand, where k is chosen as the momentum of this line. Then, keeping in mind the introduction of generalized spherical coordinates and performing power-counting at small k (i.e. when all the components of the four-vector k are small), we again enΛ counter a divergent behaviour 0 dr r−1 but now at small values of r. There is a similarity between the properties of IR divergences of this kind and those of UV divergences. One can deﬁne, for such oﬀ-shell IR divergences, an IR degree of divergence, in a similar way to the UV case. A reasonable choice is provided by the value ω ˜ (γ) = −ω(Γ/γ) ≡ ω(γ) − ω(Γ ) ,

(2.17)

where γ ≡ Γ \γ is the completion of the subgraph γ in a given graph Γ and Γ/γ denotes the reduced graph which is obtained from Γ by reducing every connectivity component of γ to a point. The absence of oﬀ-shell IR divergences is guaranteed if the IR degrees of divergence are negative for all massless subgraphs γ whose completions γ include all the external vertices in the same connectivity component. (See details in [8, 27] and Sect. E.1.) The oﬀ-shell IR divergences are the worst but they are in fact absent in physically

2.2 Divergences

17

meaningful theories. However, they play an important role in asymptotic expansions of Feynman diagrams (see [29]). The other kinds of IR divergences arise when the external momenta considered are on a surface where the Feynman diagram is singular: either on a mass shell or at a threshold. Consider, for example, the graph Fig. 2.2, with the indices a1 = 1 and a2 = 2 and the masses m1 = 0 and m2 = m = 0 on the mass shell, q 2 = m2 . With k as the momentum of the second line, the corresponding Feynman integral is of the form d4 k FΓ (q; d) = . (2.18) 2 2 k (k − 2q·k)2 At small values of k, the integrand behaves like 1/[4k 2 (q·k)2 ], and, with the help of power counting, we see that there is an on-shell IR divergence which would not be present for q 2 = m2 . If we consider Fig. 2.2 with equal masses and indices a1 = a2 = 2 at the threshold, i.e. at q 2 = 4m2 , it might seem that there is a threshold IR divergence because, choosing the momenta of the lines as q/2 + k and q/2 − k, we obtain the integral d4 k , (2.19) 2 (k + q·k)2 (k 2 − q·k)2 with an integrand that behaves at small k as 1/(q · k)4 and is formally divergent. However, the divergence is in fact absent. (The threshold singularity at q 2 = 4m2 is, of course, present.) Nevertheless, threshold IR divergences do exist. For example, the sunset3 diagram of Fig. 2.3 with general masses at threshold, q 2 = (m1 + m2 + m3 )2 , is divergent in this sense when the sum of the integer powers of the propagators is greater than or equal to ﬁve (see, e.g. [11]).

Fig. 2.3. Sunset diagram

The IR divergences characterized above are local in momentum space, i.e. they are connected with special points of the loop integration momenta. Collinear divergences arise at lines parallel to certain light-like four-vectors. A typical example of a collinear divergence is provided by the massless triangle graph of Fig. 2.4. Let us take p21 = p22 = 0 and all the masses equal to zero. The corresponding Feynman integral is 3

called also the sunrise diagram, or the London transport diagram.

18

2 Feynman Integrals: Basic Deﬁnitions and Tools

Fig. 2.4. One-loop triangle diagram

(k 2

d4 k . − 2p1 ·k)(k 2 − 2p2 ·k)k 2

(2.20)

At least an on-shell IR divergence is present, because the integral is divergent when k → 0 (componentwise). However, there are also divergences at nonzero values of k that are collinear with p1 or p2 and where k 2 ∼ 0. This follows from the fact that the product 1/[(k 2 − 2p·k)k 2 ], where p2 = 0 and p = 0, generates collinear divergences. To see this let us take residues in the upper complex half plane when integrating this product over k0 . For example, taking the residue at k0 = −|k| + i0 leads to an integral containing 1/(p·k) = 1/[p0 |k|(1 − cos θ)], where θ is the angle between the spatial components k and p. Thus, for small θ, we have a divergent integration over angles because of the factor d cos θ/(1−cos θ) ∼ dθ/θ. The second residue generates a similar divergent behaviour – this can be seen by making the change k → p − k. Another way to reveal the collinear divergences is to introduce the lightcone coordinates k± = k0 ± k3 , k = (k1 , k2 ). If we choose p with the only non-zero component p+ , we shall see a logarithmic divergence coming from the region k− ∼ k 2 ∼ 0 just by power counting. These are the main types of divergences of usual Feynman integrals. Various special divergences arise in more general Feynman integrals (2.6) that can contain linear propagators and appear on the right-hand side of asymptotic expansions in momenta and masses and in associated eﬀective theories. For example, in the Sudakov limit, one encounters divergences that can be classiﬁed as UV collinear divergences. Another situation with various non-standard divergences is provided by threshold expansion and the corresponding eﬀective theories, NRQCD and pNRQCD, where special power counting is needed to characterize the divergences.

2.3 Alpha Representation A useful tool to analyse the divergences of Feynman integrals is the so-called alpha representation based on (2.12). It can be written down for any Feynman integral. For example, for (2.8), one inserts (2.12) for each of the two propagators, takes the four-dimensional Gaussian integral by means of (2.14) to obtain

2.3 Alpha Representation

FΓ (q) = iπ 2 0

∞

∞

0

× exp iq 2

19

dα1 dα2 (α1 + α2 )−2

α1 α2 − i(m21 α1 + m22 α2 ) . α1 + α2

(2.21)

For a usual general Feynman integral, this procedure can also explicitly be realized. Using (2.12) for each propagator of a general usual Feynman integral (i.e., with usual propagators (2.2)) one takes (see, e.g., [20]) 4hdimensional Gauss integrals by means of a generalization of (2.14) to the case of an arbitrary number of loop integration momenta: Aij ki ·kj + 2 qi ·ki d4 k1 . . . d4 kh exp i i,j

= i−h π 2h (det A)−2 exp −i

i

. A−1 ij qi ·qj

(2.22)

i,j

Here A is an h × h matrix and A−1 its inverse.4 The elements of the inverse matrix involved here are rewritten in graphtheoretical language (see details in [5, 20]), and the resulting alpha representation takes the form [6] i−a−h π 2h FΓ (q1 , . . . , qn ; d) = Γ (al ) ∞ l∞ 2 a −1 × dα1 . . . dαL αl l U −2 ZeiV/U −i ml αl , 0

0

where a = al , and U and V are the well-known functions αl , U= T ∈T 1 l∈T

V=

2 αl q T .

(2.23)

l

(2.24) (2.25)

T ∈T 2 l∈T

In (2.24), the sum runs over trees of the given graph, i.e. maximal connected subgraphs without loops, and, in (2.25), over 2-trees, i.e. subgraphs that do not involve loops and consist of two connectivity components; ±q T is the sum of the external momenta that ﬂow into one of the connectivity components of the 2-tree T . (It does not matter which component is taken because of the conservation law for the external momenta.) The products of the alpha parameters involved are taken over the lines that do not belong to the 4

In fact, the matrix A involved here equals eβe+ with the elements of an arbitrarily chosen column and row with the same number deleted. Here e is the incidence matrix of the graph, i.e. eil = ±1 if the vertex i is the beginning/end of the line l, e+ is its transpose and β consists of the numbers 1/αl on the diagonal – see, e.g., [20].

20

2 Feynman Integrals: Basic Deﬁnitions and Tools

given tree T . The functions U and V are homogeneous functions of the alpha parameters with the homogeneity degrees h and h + 1, respectively. The factor Z is responsible for the non-scalar structure of the diagram: 1 ∂ Z= Zl , (2.26) ei(2B−K)/U 2i ∂ul l

u1 =...uL =0

where (see, e.g., [27, 38]) B= ul qT αl , l

K=

T ∈Tl1

T ∈T 0 l∈T

αl

l ∈T

±ul

(2.27)

2 .

(2.28)

l

In (2.27), the sum is taken over trees Tl1 that include a given line l, and qT is the total external momentum that ﬂows through the line l (in the direction of its orientation). In (2.28), the sum is taken over pseudotrees T 0 (a pseudotree is obtained from a tree by adding a line), and the sum in l is performed over the loop (circuit) of the pseudotree T , with a sign dependent on the coincidence of the orientations of the line l and the pseudotree T . The alpha representation of a general h-loop Feynman integral is useful for general analyses. In practical calculations, e.g. at the two-loop level, one can derive the alpha representation for concrete diagrams by hand, rather than deduce it from the general formulae presented above. Still, even in practice, such general formulae can provide advantages because the evaluation of the functions of the alpha representation can be performed on a computer. Let us stress that this terrible-looking machinery for evaluating the determinant of the matrix A that arises from Feynman integrals, as well as for evaluating the elements of the inverse matrix, together with interpreting these results from the graph-theoretical point of view, is exactly the same as that used in the problem of the solution of Kirchhoﬀ’s laws for electrical circuits, a problem typical of the nineteenth century. Recall, for example, that the parameters αl play the role of ohmic resistances and that the expression (2.24) for the function U as a sum over trees is a Kirchhoﬀ result. Explicit formulae for Feynman integrals (2.6) with more general propagators which can be linear are not known. In this situation, one can derive alpha representation for any given concrete Feynman integral using formulae like (2.12) and performing Gaussian integration as in the case of Feynman integrals with standard propagators. We will follow this way in Chap. 3.

2.4 Regularization The standard way of dealing with divergent Feynman integrals is to introduce a regularization. This means that, instead of the original ill-deﬁned Feynman

2.4 Regularization

21

integral, we consider a quantity which depends on a regularization parameter, λ, and formally tends to the initial, meaningless expression when this parameter takes some limiting value, λ = λ0 . This new, regularized, quantity turns out to be well-deﬁned, and the divergence manifests itself as a singularity with respect to the regularization parameter. Experience tells us that this singularity can be of a power or logarithmic type, i.e. lnn (λ − λ0 )/(λ − λ0 )i . Although a regularization makes it possible to deal with divergent Feynman integrals, it does not actually remove UV divergences, because this operation is of an auxiliary character so that sooner or later it will be necessary to switch oﬀ the regularization. To provide ﬁniteness of physical observables evaluated through Feynman diagrams, another operation, called renormalization, is used. This operation is described, at the Lagrangian level, as a redeﬁnition of the bare parameters of a given Lagrangian by inserting counterterms. The renormalization at the diagrammatic level is called R-operation and removes the UV divergence from individual Feynman integrals. It is, however, beyond the scope of the present book. (See, however, some details in Sect. F.5, where the method of IR rearrangement is brieﬂy described.) An obvious way of regularizing Feynman integrals is to introduce a cutoﬀ at large values of the loop momenta. Another well-known regularization procedure is the Pauli–Villars regularization [24], which is described by the replacement 1 1 1 → 2 − 2 p 2 − m2 p − m2 p − M2 and its generalizations. For ﬁnite values of the regularization parameter M , this procedure clearly improves the UV asymptotics of the integrand. Here the limiting value of the regularization parameter is M = ∞. If we replace the integer powers al in the propagators by general complex numbers λl we obtain an analytically regularized [30] Feynman integral where the divergences of the diagram are encoded in the poles of this regularized quantity with respect to the analytic regularization parameters λl . For example, power counting at large values of the loop momentum inthe analytically ∞ regularized version of (2.8) leads to the divergent behaviour Λ dr rλ1 +λ2 −3 , which results in a pole 1/(λ1 + λ2 − 2) at the limiting values of the regularization parameters λl = 1. For example, in the case of the analytically regularized integral of Fig. 2.2, we obtain αλ1 −1 α2λ2 −1 e−iπ(λ1 +λ2 +1)/2 π 2 ∞ ∞ FΓ (q; λ1 , λ2 ) = dα1 dα2 1 Γ (λ1 )Γ (λ2 ) (α1 + α2 )2 0 0 α1 α2 2 2 2 − i(m1 α1 + m2 α2 ) . × exp iq (2.29) (α1 + α2 ) After the change of variables η = α1 + α2 , ξ = α1 /(α1 + α2 ) and explicit integration over η, we arrive at

22

2 Feynman Integrals: Basic Deﬁnitions and Tools

FΓ (q; λ1 , λ2 ) = eiπ(λ1 +λ2 ) ×

ξ λ1 −1 (1 − ξ)λ2 −1

1

dξ 0

iπ 2 Γ (λ1 + λ2 − 2) Γ (λ1 )Γ (λ2 ) λ1 +λ2 −2

[m21 ξ + m22 (1 − ξ) − q 2 ξ(1 − ξ) − i0]

.

(2.30)

Thus the UV divergence manifests itself through the ﬁrst pole of the gamma function Γ (λ1 + λ2 − 2) in (2.30), which results from the integration over small values of η due to the power η λ1 +λ2 −3 . The alpha representation turns out to be very useful for the introduction of dimensional regularization, which is a commonly accepted computational technique successfully applied in practice and which will serve as the main kind of regularization in this book. Let us imagine that the number of space– time dimensions diﬀers from four. To be more precise, the number of space dimensions is considered to be d − 1, rather than three. (But, of course, we still think of an integer number of dimensions!) The derivation of the alpha representation does not change much in this case. The only essential change is that, instead of (2.14), we need to apply its generalization to an arbitrary number of dimensions, d: 2 2 (2.31) dd k ei(αk −2q·k) = eiπ(1−d/2)/2 π d/2 α−d/2 e−iq /α . So, instead of (2.21), we have the following in d dimensions: ∞ ∞ FΓ (q; d) = e−iπ(1+d/2)/2 π d/2 dα1 dα2 (α1 + α2 )−d/2 0 0 2 α1 α2 × exp iq − i(m21 α1 + m22 α2 ) . α1 + α2

(2.32)

The only two places where something has been changed are the exponent of the combination (α1 + α2 ) in the integrand and the exponents of the overall factors. Now, in order to introduce dimensional regularization, we want to consider the dimension d as a complex number. So, by deﬁnition, the dimensionally regularized Feynman integral for Fig. 2.2 is given by (2.32) and is a function of q 2 as given by this integral representation. We choose d = 4−2ε, where the value ε = 0 corresponds to the physical number of the space–time dimensions. By the same change of variables as used after (2.29), we obtain ∞ −iπ(1+d/2)/2 d/2 FΓ (q; d) = e π dη η ε−1 ×

0 1

dξ exp iq 2 ξ(1 − ξ)η − i[m21 ξ + m22 (1 − ξ)]η .

(2.33)

0

This integral is absolutely convergent for 0 < Re ε < Λ (where Λ = ∞ if both masses are non-zero and Λ = 1 otherwise; this follows from an IR analysis of convergence, which we omit here) and deﬁnes an analytic function of ε, which

2.4 Regularization

23

is extended from this domain to the whole complex plane as a meromorphic function. After evaluating the integral over η, we arrive at the following result: 1 dξ (2.34) FΓ (q; d) = iπ d/2 Γ (ε) ε . 2 2 2 0 [m1 ξ + m2 (1 − ξ) − q ξ(1 − ξ) − i0] The UV divergence manifests itself through the ﬁrst pole of the gamma function Γ (ε) in (2.34), which results from the integration over small values of η in (2.33). This procedure of introducing dimensional regularization is easily generalized [6, 7, 8] to an arbitrary usual Feynman integral. Instead of (2.22), we use Aij ki ·kj + 2 qi ·ki dd k1 . . . dd kh exp i i,j

= eiπh(1−d/2)/2 π hd/2 (det A)−d/2 exp −i

i

, A−1 ij qi ·qj

(2.35)

i,j

and the resulting d-dimensional alpha representation takes the form [6, 7] eiπ[a+h(1−d/2)]/2 π hd/2 FΓ (q1 , . . . , qn ; d) = (−1)a l Γ (al ) ∞ ∞ 2 a −1 × dα1 . . . dαL αl l U −d/2 ZeiV/U −i ml αl . 0

0

(2.36)

l

Let us now deﬁne5 the dimensionally regularized Feynman integral by means of (2.36), treating the quantity d as a complex number. This is a function of kinematical invariants constructed from the external momenta and contained in the function V. In addition to this, we have to take care of polynomials in the external momenta and the auxiliary variables ul hidden in the factor Z. We treat these objects qi and ul , as well as the metric tensor gµν , as elements of an algebra of covariants, where we have, in particular, 5

An alternative deﬁnition of algebraic character [16, 32, 36] (see also [10]) exists and is based on certain axioms for integration in a space with non-integer dimension. It is unclear how to perform the analysis within such a deﬁnition, for example, how to apply the operations of taking a limit, diﬀerentiation, etc. to algebraically deﬁned Feynman integrals in d dimensions, in order to say something about the analytic properties with respect to momenta and masses and the parameter of dimensional regularization. After evaluating a Feynman integral according to the algebraic rules, one arrives at some concrete function of these parameters but, before integration, one is dealing with an abstract algebraic object. Let us remember, however, that, in practical calculations, one usually does not bother about precise deﬁnitions. From the purely pragmatic point of view, it is useless to think of a diagram when it is not calculated. On the other hand, from the pure theoretical and mathematical point of view, such a position is beneath criticism. ;-)

24

2 Feynman Integrals: Basic Deﬁnitions and Tools

∂ ∂uµl

uνl = gµν δl,l ,

gµµ = d .

This algebra also includes the γ-matrices with anticommutation relations γµ γν + γν γµ = 2gµν so that γ µ γµ = d, the tensor εκµνλ , etc. Thus the dimensionally regularized Feynman integrals are deﬁned as linear combinations of tensor monomials in the external momenta and other algebraic objects with coeﬃcients that are functions of the scalar products qi ·qj . However, this is not all, because we have to see that the α-integral is well-deﬁned. Remember that it can be divergent, for various reasons. The alpha representation is not only an important technique for evaluating Feynman integrals but also a very convenient tool for the analysis of their convergence. This analysis is outlined in Sect. E.1. It is based on decompositions of the alpha integral into so-called sectors where new variables are introduced in such a way that the integrand factorizes, i.e. takes the form of a product of some powers of the sector variables with a non-zero function. Eventually, in the new variables, the analysis of convergence reduces to power counting (for both UV and IR convergence) in one-dimensional integrals. As a result of this analysis, any Feynman integral considered at Euclidean external momenta qi , i.e. when any sum of incoming momenta is spacelike, is deﬁned as meromorphic function of d with series of UV and IR poles [7, 25, 27, 31, 33]. Here it is also assumed that there are no massless detachable subgraphs, i.e. massless subdiagrams with zero external momenta. For example, a tadpole, i.e. a line with coincident end points, is a detachable subgraph. However, such diagrams are naturally put to zero in case they are massless – see a discussion below. Unfortunately, there are no similar mathematical results for Feynman integrals on a mass shell or a threshold which are really needed in practice and which be mainly considered in this book. However, in every concrete example considered below, we shall see that every Feynman diagram is indeed an analytical function of d, both in intermediate steps of a calculation and, of course, in our results. Still it would be nice to have also a mathematical theorem on the convergence of general Feynman integrals. On the other hand, there is a practical algorithm [3] based on some sector decompositions that can provide the resolution of the singularities in ε for any given Feynman integral in the case where all the non-zero kinematical invariants have the same sign (and, possibly, are on a mass shell or at a threshold). This algorithm is described in Sect. E.2.

2.5 Properties of Dimensionally Regularized Feynman Integrals We can formally write down dimensionally regularized Feynman integrals as integrals over d-dimensional vectors ki :

2.5 Properties of Dimensionally Regularized Feynman Integrals

FΓ (q1 , . . . , qn ; d) =

dd k1 . . .

dd kh

L

˜ F,l (pl ) . D

25

(2.37)

l=1

In order to obtain dimensionally regularized integrals with their dimension independent of ε, a factor of µ−2ε per loop, where µ is a massive parameter, is introduced. This parameter serves as a renormalization parameter for schemes based on dimensional regularization. Therefore, we obtain logarithms and other functions depending not only on ratios of given parameters, e.g. q 2 /m2 , but also on q 2 /µ2 etc. However, we shall usually omit this µ-dependence for brevity (i.e. set µ = 1) so that you will meet sometimes quantities like ln q 2 which should be understood in the sense of ln(q 2 /µ2 ). We have reasons for using the notation (2.37), because dimensionally regularized Feynman integrals as deﬁned above possess the standard properties of integrals of the usual type in integer dimensions. In particular, – the integral of a linear combination of integrands equals the same linear combination of the corresponding integrals; – one may cancel the same factors in the numerator and denominator of integrands. These properties follow directly from the above deﬁnition. A less trivial property is that – a derivative of an integral with respect to a mass or momentum equals the corresponding integral of the derivative. This is also a consequence (see [8, 27]) of the deﬁnition of dimensionally regularized Feynman integrals based on the alpha representation and the corresponding analysis of convergence presented in Sect. E.1. To prove this statement, one uses standard algebraic relations between the functions entering the alpha representation [7, 20]. (We note again that these are relations quite similar to those encoded in the solutions of Kirchhoﬀ’s laws for a circuit deﬁned by the given graph.) A corollary of the last property is the possibility of integrating by parts and always neglecting surface terms: L ∂ ˜ d d DF,l (pl ) = 0 , i = 1, . . . , h . (2.38) – d k1 . . . d kh µ ∂ki l=1

This property is the basis for solving the reduction problem for Feynman integrals using IBP relations [9] – see Chaps. 5 and 6. The next property says that – any diagram with a detachable massless subgraph is zero. This property can also be shown to be a consequence of the accepted deﬁnition [8, 27], by use of an auxiliary analytic regularization, using pieces of the α-integral considered in diﬀerent domains of the regularization parameters. Let us consider, for example, the massless tadpole diagram, which

26

2 Feynman Integrals: Basic Deﬁnitions and Tools

can be reduced by means of alpha parameters to a scaleless one-dimensional integral: ∞ d d k ε d/2 = −i π dα αε−2 . (2.39) k2 0 We divide this integral into two pieces, from 0 to 1 and from 1 to ∞, integrate these two integrals and ﬁnd results that are equal except for opposite signs, which lead to the zero value.6 It should be stressed here that the two pieces that contribute to the right-hand side of (2.39) are convergent in diﬀerent domains of the regularization parameter ε, namely, Re ε > −1 and Re ε < −1, with no intersection, and that this procedure here is equivalent to introducing analytic regularization and considering its parameter in diﬀerent domains for diﬀerent pieces. But let us distinguish between two qualitatively diﬀerent situations: the ﬁrst when we have to deal with a massless Feynman integral, with a zero external momentum, which arises from the Feynman rules, and the second when we obtain such scaleless integrals after some manipulations: after using partial fractions, diﬀerentiation, integration by parts, etc. We can also include in this second class all such integrals that appear on the right-hand side of explicit formulae for (oﬀ-shell) asymptotic expansions in momenta and masses [2, 29]. In the ﬁrst situation, the only possibility is to use the ad hoc prescription of setting the integral to zero. In the second situation, we can start with an alpha representation, introduce an auxiliary analytic regularization [8, 27] and use the fact that it is convergent in some non-empty domain of these parameters (see Sect. E.1). A very important point here is that all the properties of dimensionally regularized integrals given above, apart from the last one, can be justiﬁed in a purely algebraic way [8, 27], through identities between functions in the alpha representation. Then, using sector decompositions described in Sect. E.1, with a control over convergence at hand, one can see that all the resulting massless Feynman integrals with zero external momenta indeed vanish – see details in [8, 27]. Let us now remind ourselves of reality and observe that it is necessary to deal in practice with diagrams on a mass shell or at a threshold. What about the properties of dimensionally regularized Feynman integrals in this case? At least the algebraic proof of the basic properties of dimensionally regularized Feynman integrals is not sensitive to putting the external momenta in any particular place. However, as we noticed above, a general analysis of the convergence of such integrals, even in speciﬁc cases, is still absent, so that we do not have control over convergence. Technically, this means that the sectors used for the analysis of the convergence in the oﬀ-shell case are no longer suﬃcient for the resolution of the singularities of the integrand of the alpha 6

These arguments can be found, for example, in [17], and even in a pure mathematical book [13]. Well, let us not take the latter example seriously ;-)

2.5 Properties of Dimensionally Regularized Feynman Integrals

27

representation. These singularities are much more complicated and can even appear (e.g. at a threshold) at non-zero, ﬁnite values of the α-parameters. However, the good news is that numerous practical applications have shown that there is no sign of breakdown of these properties for on-shell or threshold Feynman integrals. Although on-shell and threshold Feynman integrals have been already mentioned many times, let us now be more precise in our deﬁnitions. We must realize that, generally, an on-shell or threshold Feynman integral is not the value of the given Feynman integral FΓ (q 2 , . . .), deﬁned as a function of q 2 and other kinematical variables, at a value of q 2 on a mass shell or at a threshold. Consider, for example, the Feynman integral corresponding to Fig. 2.2, with m1 = 0, m2 = m, a1 = 1, a2 = 2. We know an explicit result for the diagram given by (1.5). There is a logarithmic singularity at threshold, q 2 = m2 , so that we cannot strictly speak about the value of the integral there. Still we can certainly deﬁne the threshold Feynman integral by putting q 2 = m2 in the integrand of the integral over the loop momentum or over the alpha parameters. And this is what was really meant and will be meant by ‘on-shell’ and ‘threshold’ integrals. In this example, we obtain an integral which can be evaluated by means of (A.13) (to be derived in Chap. 3): Γ (ε) dd k = iπ d/2 . (2.40) 2 2 k (k − 2q·k)2 2(m2 )1+ε This integral is divergent, in contrast to the original Feynman integral deﬁned for general q 2 . Thus on-shell or threshold dimensionally regularized Feynman integrals are deﬁned by the alpha representation or by integrals over the loop momenta with restriction of some kinematical invariants to appropriate values in the corresponding integrands. In this sense, these regularized integrals are ‘formal’ values of general Feynman integrals at the chosen variables. Note that the products of the free ﬁelds in the Lagrangian are not required to be normal-ordered, so that products of ﬁelds of the same sort at the same point are allowed. The formal application of the Wick theorem therefore generates values of the propagators at zero. For example, in the case of the scalar free ﬁeld, with the propagator i e−ix·k 4 k , (2.41) d DF (x) = (2π)4 k 2 − m2 which satisﬁes (2 + m2 )DF (x) = −iδ(x), we have T φ(x)φ(x) = : φ2 (x) : +DF (0) .

(2.42)

The value of DF (x) at x = 0 does not exist, because the propagator is singular at the origin according to (2.16). However, we imply the formal value at the origin rather than the ‘honestly’ taken value. This means that we set x to zero in some integral representation of this quantity. For example, using the

28

2 Feynman Integrals: Basic Deﬁnitions and Tools

inverse Fourier transformation, we can deﬁne DF (0) as the integral (2.41) with x set to zero in the integrand. Thus, by deﬁnition, i d4 k DF (0) = . (2.43) 4 2 (2π) k − m2 This integral is, however, quadratically divergent, as Feynman integrals typically are. So, we understand DF (0) as a dimensionally regularized formal value when we put x = 0 in the Fourier integral and obtain, using (A.1) (which we will derive shortly), dd k = −iπ d/2 Γ (ε − 1)(m2 )1−ε . (2.44) 2 k − m2 This Feynman integral in fact corresponds to the tadpole φ4 theory graph shown in Fig. 2.5. The corresponding quadratic divergence manifests itself through an UV pole in ε – see (2.44).

Fig. 2.5. Tadpole

Observe that one can trace the derivation of the integrals tabulated in Sect. A.1 and see that the integrals are convergent in some non-empty domains of the complex parameters λl and ε and that the results are analytic functions of these parameters with UV, IR and collinear poles. Before continuing our discussion of setting scaleless integrals to zero, let us present an analytic result for the one-loop massless triangle integral with two on-shell external momenta, p21 = p22 = 0. Using (A.28) (which we will derive in Chap. 3), we obtain 2 dd k d/2 Γ (1 + ε)Γ (−ε) = −iπ . (2.45) (k 2 − 2p1 ·k)(k 2 − 2p2 ·k)k 2 Γ (1 − 2ε)(−q 2 )1+ε A double pole at ε = 0 arises from the IR and collinear divergences. A similar formula with a monomial in the numerator can be obtained also straightforwardly: µ µ 2 dd k k µ d/2 Γ (ε)Γ (1 − ε) p1 + p2 = iπ . (k 2 − 2p1 ·k)(k 2 − 2p2 ·k)k 2 Γ (2 − 2ε) (−q 2 )1+ε (2.46) Now only a simple pole is present, because the factor k µ kills the IR divergence. Consider now a massless one-loop integral with the external momentum on the massless mass shell, p2 = 0:

References

dd k . (p − k)2 k 2

29

(2.47)

If we write down the alpha representation for this integral we obtain the same expression (2.39) as for p = 0 because only p2 , equal to zero in both cases, is involved there. In spite of this obvious fact, there is still a qualitative diﬀerence: for p = 0, there are UV and IR poles which enter with opposite signs and, for p2 = 0 (but with p = 0 as a d-dimensional vector), there is a similar interplay of UV and collinear poles. Now we follow the arguments presented in [21] and write down the following identity for (2.47), with p = p1 : dd k 2 (k − 2p1 ·k)k 2 dd k 2p2 ·k dd k − = , (k 2 − 2p1 ·k)(k 2 − 2p2 ·k) (k 2 − 2p1 ·k)(k 2 − 2p2 ·k)k 2 (2.48) where p22 = 0 and p1·p2 = 0. We then evaluate the integrals on the right-hand side by means of (A.7) and (2.46), respectively, and obtain a zero value. This fact again exempliﬁes the consistency of our rules. Thus we are going to systematically apply the properties of dimensionally regularized Feynman integrals in any situation, no matter where the external momenta are considered to be. Moreover, we will believe that these properties are also valid for more general Feynman integrals given by the dimensionally regularized version of (2.6) which can contain linear propagators. Let us also point out that the rule to put all scaleless integrals to zero is rather general and, as far as I know, never causes contradictions. In particular, it is applied in asymptotic expansions of Feynman integrals in various limits of momenta and masses within expansion by regions [2, 29], where such integrals are always put to zero, even if they are not regulated by dimensional regularization. We will follow this rule also in Chap. 6 where we will put to zero scaleless integrals which appear in auxiliary parametric representations when constructing coeﬃcient functions at master integrals.

References 1. S. Actis, A. Ferroglia, G. Passarino, M. Passera and S. Uccirati, hepph/0402132. 12 2. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 13, 26, 29 3. T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 24 4. G.T. Bodwin, E. Braaten and G.P. Lepage, Phys. Rev. D 51 (1995) 1125; Phys. Rev. D 55 (1997) 5853. 14 5. N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 19

30 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38.

2 Feynman Integrals: Basic Deﬁnitions and Tools C.G. Bollini and J.J. Giambiagi, Nuovo Cim. B 12 (1972) 20. 19, 23 P. Breitenlohner and D. Maison, Commun. Math. Phys. 52 (1977) 11, 39, 55. 23, 24, 25 K.G. Chetyrkin and V.A. Smirnov, Teor. Mat. Fiz. 56 (1983) 206. 16, 23, 25, 26 K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 25 J.C. Collins, Renormalization (Cambridge University Press, Cambridge, 1984). 23 A.I. Davydychev and V.A. Smirnov, Nucl. Phys. B 554 (1999) 391. 17 A. Erd´elyi (ed.), Higher Transcendental Functions, Vols. 1 and 2 (McGraw-Hill, New York, 1954). 16 I.M. Gel’fand and G.E. Shilov, Generalized Functions, Vol. 1 (Academic Press, New York, London, 1964). 26 A.G. Grozin, Heavy Quark Eﬀective Theory (Springer, Berlin, Heidelberg, 2004). 14 K. Hepp, Commun. Math. Phys. 2 (1966) 301. G. ’t Hooft and M. Veltman, Nucl. Phys. B 44 (1972) 189. 23 G. Leibbrandt, Rev. Mod. Phys. 47 (1975) 849. 26 G.P. Lepage et al., Phys. Rev. D 46 (1992) 4052. 14 A.V. Manohar and M.B. Wise, Heavy Quark Physics (Cambridge University Press, Cambridge 2000). 14 N. Nakanishi, Graph Theory and Feynman Integrals (Gordon and Breach, New York, 1971). 19, 25 W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 29 M. Neubert, Phys. Rep. 245 (1994) 259. 14 G. Passarino and M. Veltman, Nucl. Phys. B 160 (1979) 151. 12 W. Pauli and F. Villars, Rev. Mod. Phys. 21 (1949) 434. 21 K. Pohlmeyer, J. Math. Phys. 23 (1982) 2511. 24 A.P. Prudnikov, Yu.A. Brychkov and O.I. Marichev, Integrals and Series, Vols. 1–3 (Gordon and Breach, New York, 1986–1990). 16 V.A. Smirnov, Renormalization and Asymptotic Expansions (Birkh¨ auser, Basel, 1991). 16, 20, 24, 25, 26 V.A. Smirnov, Phys. Lett. B 491 (2000) 130; Phys. Lett. B 500 (2001) 330. V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 13, 17, 26, 29 E.R. Speer, J. Math. Phys. 9 (1968) 1404. 21 E.R. Speer, Commun. Math. Phys. 23 (1971) 23; Commun. Math. Phys. 25 (1972) 336. 24 E.R. Speer, in Renormalization Theory, eds. G. Velo and A.S. Wightman (Reidel, Dodrecht, 1976) p. 25. 23 E.R. Speer, Ann. Inst. H. Poincar´e 23 (1977) 1. J.B. Tausk, Phys. Lett. B 469 (1999) 225. 24 B.A. Thacker and G.P. Lepage, Phys. Rev. D 43 (1991) 196. K.G. Wilson, Phys. Rev. D 7 (1973) 2911. 14 S. Wolfram, The Mathematica Book, 4th edition (Wolfram Media and Cambridge University Press, Cambridge, 1999). 23 O.I. Zavialov, Renormalized Quantum Field Theory (Kluwer Academic Publishers, Dodrecht, 1990). 16 20

3 Evaluating by Alpha and Feynman Parameters

Feynman parameters1 are very well known and often used in practical calculations. They are closely related to alpha parameters introduced in Chap. 2 so that we shall study both kinds of parametric representations of Feynman integrals in one chapter. The use of these parameters enables us to transform Feynman integrals over loop momenta into parametric integrals where Lorentz invariance becomes manifest. Using alpha parameters we shall ﬁrst evaluate one and two-loop integrals with general complex powers of the propagators, within dimensional regularization, for which results can be written in terms of gamma functions for general values of the dimensional regularization parameter. We shall show then how these formulae, together with simple algebraic manipulations, enable us to evaluate some classes of Feynman integrals. We then turn to various characteristic one-loop examples where results cannot be written in terms of gamma functions. In such situations, we shall be usually oriented at the evaluation in expansion in powers of ε up to some ﬁxed order. We then introduce Feynman parameters and present the so-called Cheng–Wu theorem which provides a very useful trick that can greatly simplify the evaluation. Finally, we proceed at the two-loop level by presenting rather complicated examples of evaluating Feynman integrals by Feynman and alpha parameters.

3.1 Simple One- and Two-Loop Formulae A lot of one- and two-loop formulae can be derived, using alpha and Feynman parameters, for general complex indices with results expressed in terms of gamma functions. A collection of such formulae is presented in Sect. A.1. Let us evaluate, for example, the dimensionally regularized massive tadpole Feynman diagram of Fig. 2.5 with a general power of the propagator, dd k . (3.1) FΓ (q; λ, d) = (−k 2 + m2 )λ We apply the alpha representation of the analytically regularized scalar propagator given by (2.12) with Z = 1, i.e. 1

See, e.g., textbooks [18] and [7].

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 31–53 (2004) c Springer-Verlag Berlin Heidelberg 2004

32

3 Evaluating by Alpha and Feynman Parameters

1 iλ = (−k 2 + m2 )λ Γ (λ)

∞

dα αλ−1 ei(k

2

−m2 )α

,

(3.2)

0

change the order of integration over k and α, take the Gaussian k integral by means of (2.31), again apply (3.2) written in the reverse order, i.e. ∞ Γ (λ) i−λ dα αλ−1 e−iAα = , (3.3) (A − i0)λ 0 and arrive at (A.1). In particular, this table formula gives (2.44). Let us now turn to the dimensionally regularized Feynman diagram of Fig. 2.2 with general powers of the propagators, dd k . (3.4) FΓ (q; λ1 , λ2 , d) = (−k 2 + m21 )λ1 [−(q − k)2 + m22 ]λ2 From now on, we shall use the following convention: when powers of propagators are integers we use them with +k 2 +i0, but when they are non-integral or complex, we take the opposite sign, i.e. −k 2 −i0. The second choice is more natural if we wish to obtain a Euclidean, −q 2 , dependence of the results (see, e.g., (3.6) below). We shall also prefer to use al for integer and λl for general complex indices. In the latter case, the alpha representation is obtained from (2.36) by replacing al by λl and dropping out the factor (−1)a . Starting from the alpha representation of Fig. 2.2, with the basic functions U = α1 +α2 and V = α1 α2 q 2 , and using the change of variables α1 = ξη, α2 = η(1 − ξ) we obtain the dimensionally regularized version of (2.30), i.e. FΓ (q; λ1 , λ2 , d) = iπ d/2 × 0

1

Γ (λ1 + λ2 + ε − 2) Γ (λ1 )Γ (λ2 ) dξ ξ λ1 −1 (1 − ξ)λ2 −1

[m21 ξ + m22 (1 − ξ) − q 2 ξ(1 − ξ) − i0]

λ1 +λ2 +ε−2

.

(3.5)

Suppose that the masses are zero. In this case the integral over ξ can be evaluated in terms of gamma functions, and we arrive at the following result: dd k G(λ1 , λ2 ) = iπ d/2 , (3.6) (−k 2 )λ1 [−(q − k)2 ]λ2 (−q 2 )λ1 +λ2 +ε−2 where G(λ1 , λ2 ) =

Γ (λ1 + λ2 + ε − 2)Γ (2 − ε − λ1 )Γ (2 − ε − λ2 ) . Γ (λ1 )Γ (λ2 )Γ (4 − λ1 − λ2 − 2ε)

(3.7)

The one-loop formula (3.6) can graphically be described by Fig. 3.1. In the case where the powers of propagators are equal to one, we have dd k Γ (ε)Γ (1 − ε)2 = iπ d/2 . (3.8) 2 2 k (q − k) Γ (2 − 2ε)(−q 2 )ε Note that although the indices of the diagrams are integral at the beginning, non-integral indices shifted by amounts proportional to ε appear after intermediate integration, e.g. after the use of (3.8) inside a bigger diagram.

3.1 Simple One- and Two-Loop Formulae λ2

= iπ d/2 G(λ1 , λ2 )×

33

λ1 + λ2 − d/2

λ1 Fig. 3.1. Graphical interpretation of (3.6)

Another formula that can be derived from (3.5) gives a result for the integral dd k . (−k 2 + m2 )λ1 (−k 2 )λ2 Indeed, we set q = 0, m1 = m and m2 = 0, take an integral over ξ and obtain (A.4). Consider now the following integral that arises in calculations in Heavy Quark Eﬀective Theory [12, 15, 17]: dd k . (−k 2 )λ1 (2v·k + ω − i0)λ2 Since the denominator of one of the propagators is not quadratic we cannot use the general formula of the alpha representation. Still we proceed by alpha parameters, i.e. apply (3.2) to the ﬁrst propagator and a similar Fourier representation ∞ iλ 1 = dα αλ−1 eiAα , (3.9) (−A − i0)λ Γ (λ) 0 with A = −2v · k − ω, to the second propagator. Changing the order of integration as above and evaluating a Gaussian integral over k we then apply 2 2 (3.3) to take the integral of α1λ1 +ε−3 e−iα2 v /α1 over α1 and, ﬁnally, an integral over α2 , and arrive at (A.25). The following one-loop integral is typical for the evaluation of the one-loop quark potential: dd k . 2 λ 1 (−k ) [−(q − k)2 ]λ2 (−2v·k − i0)λ3 Here v · q = 0. (Typically, one chooses q = (0, q) and v = (1, 0).) One of the propagators is again not quadratic so that we proceed by alpha parameters and represent each of the three factors as an alpha integral. After taking a Gaussian integral over k we obtain

3 iλ1 +λ2 +λ3 +ε−1 π d/2 ∞ ∞ ∞ λl −1 αl dαl (α1 + α2 )ε−2 0 0 0 l Γ (λl ) l=1 2 q α1 α2 − v 2 α32 × exp i . α1 + α2

34

3 Evaluating by Alpha and Feynman Parameters

√ Then the integral over α3 can be evaluated by the change α3 = t and (3.3). After that the integration over α1 and α2 is taken, as before, by introducing the variables η = α1 + α2 , ξ = α1 /(α1 + α2 ), with the result (A.27). Using alpha parameters one can also derive the formula (A.40) for the formal Fourier transformation within dimensional regularization. This formula provides another way to derive (3.6). In fact, the initial integral is nothing but the convolution of the two functions, f˜i = 1/(−k 2 − i0)λi , i = 1, 2. Then one uses the well-known mathematical formula f˜1 ∗ f˜2 (q) = (2π)d (f1˜f2 ) for the convolution of two Fourier transforms, applies (A.40) and arrives at (3.6).

3.2 Auxiliary Tricks 3.2.1 Recursively One-Loop Feynman Integrals Massless integrals are often evaluated with the help of successive application of the one-loop formula (3.6). In addition one can use the fact that a sequence of two lines with scalar propagators with the same mass and the indices a1 and a2 can be replaced by one line with index a1 + a2 . Consider, for example, the two-loop diagram shown in Fig. 3.2. The internal one-loop integral can be evaluated by use of (3.8) and is eﬀectively replaced, according to Fig. 3.1, by a line with index ε. Then the sequence of two massless lines with indices 1 and ε is replaced by one line with index 1 + ε, and the one-loop diagram so obtained, which has indices 2 and 1 + ε, is evaluated by means of the oneloop formula (3.6), with the following result expressed in terms of gamma functions: G(1, 1)G(2, 1 + ε)/(−q 2 )1+2ε . The class of Feynman diagrams that can be evaluated in this way by means of (3.6) can be called recursively one-loop.

Fig. 3.2. A recursively one-loop diagram

Another example where two tabulated one-loop integration formulae can successively be applied is given by the two-loop scalar diagram of Fig. 3.3 with general complex indices and two zero masses, dd k dd l . (−k 2 )λ1 [−(k + l)2 ]λ2 (m2 − l2 )λ3

3.2 Auxiliary Tricks

35

1 2 3 Fig. 3.3. Vacuum two-loop diagram with the masses 0, 0 and m

Here one can ﬁrst apply the one-loop massless integration formula (3.6), then apply (A.4) and obtain (A.39). 3.2.2 Partial Fractions When evaluating dimensionally regularized Feynman integrals one uses their properties, in particular the possibility of manipulations based on the properties listed in Sect. 2.5. Here the following standard decomposition proves to be useful: a 1 −1 a2 − 1 + i (−1)i 1 = (x + x1 )a1 (x + x2 )a2 (x2 − x1 )a2 +i (x + x1 )a1 −i a2 − 1 i=0 a 2 −1 a1 − 1 + i (−1)a1 + , (3.10) (x2 − x1 )a1 +i (x + x2 )a2 −i a1 − 1 i=0 where a1 , a2 > 0 and n n! = j j!(n − j)! is a binomial coeﬃcient. For example, the vacuum one-loop Feynman integral with two diﬀerent masses, dd k , (k 2 − m21 )(k 2 − m22 ) can be evaluated by (3.10) and (A.1), with the result iπ d/2 Γ (ε − 1)

− m2−2ε m2−2ε 2 1 . m21 − m22

If one of the indices, e.g. a2 is non-positive, a similar decomposition is performed by expanding (x + x2 )−a2 in powers of x + x1 . Let us note that if one proceeds by MATHEMATICA [22], one can use, for given integer values of a1 and a2 , the command Apart to perform partial fractions decompositions.

36

3 Evaluating by Alpha and Feynman Parameters

3.2.3 Dealing with Numerators As we have agreed we suppose that a tensor reduction for a given class of Feynman integrals was performed so that we start with evaluating scalar integrals. Let us, however, mention that one can also evaluate integrals with Lorentz indices. A lot of one-loop Feynman integrals with numerators can be found in Sect. A.1. One can reduce evaluating such a one-loop integral to an integral with a product k α1 . . . k αN . Then one can switch to traceless monomials and back using (A.41a) and (A.41b). An integral with a traceless monomial independent of other Lorentz indices is again traceless. If it depends on one external momentum it should be proportional to its traceless monomial. This is how tabulated integrals for traceless monomials, e.g. (A.8), can be derived. Then one can turn back to usual monomials using (A.41b). (In Sect. A.2, one can ﬁnd also other useful formulae for various traceless monomials.) In the case of a general h-loop Feynman integral with standard propagators, let us observe that the function (2.26) in (2.36) can be taken into account by shifting the space–time dimension d and indices al of a given diagram because any factor that arises after the diﬀerentiation with respect to the auxiliary parameters ul is a sum of products of positive integer powers of the α-parameters and negative integer powers of the function U. In particular, the factor 1/U n is taken into account by the shift d → d + 2n. Then the shift of a power of a parameter αl can be translated into a shift of the power of the corresponding propagator, in particular, a multiplication by αl can be described by the operator ial l+ where l+ increases the index al by one, the multiplication by αl2 can be described by the operator −al (al + 1)l++ , etc. This observation enables us to express any given Feynman integral with numerators through a linear combination of scalar integrals with shifted indices and shifted dimensions. Systematic algorithms oriented towards realization on a computer, with a demonstration up to two-loop level, have been constructed in [20]. We shall come back to this point in Chap. 5 when solving IBP recurrence relations. At the one-loop level, this property has been used [9] to derive a general formula for the Feynman integrals kα1 . . . kαn ) (λ , . . . , λ , d) = dd k N , (3.11) Fα(N 1 N 1 ...αn 2 λi 2 i=1 [−(qi − k) + mi ] depending on the external momenta q1 − q2 , . . . , qN − q1 and the general masses mi : (−1)r ) Fα(N (λ1 , . . . , λN , d) = 1 ...αn 2r r,κ1 ,...,κN : 2r+

×{{[g] [q1 ] r

κ1

κN

. . . [qN ]

κi =n

}α1 ...αn

N

(λi )κi

i=1

3.2 Auxiliary Tricks

37

p1 a1 a3 a2 p2

Fig. 3.4. Triangle diagram with the masses 0, 0, m, external momenta p21 = p22 = 0 and general indices of the propagators

×F (N ) (λ1 + κ1 , . . . , λN + κN , d + 2(n − r)) ,

(3.12)

where {[g] [q1 ] . . . [qN ] }α1 ...αn is symmetric in its indices and is composed of the metric tensor and the vectors qi . Tabulated formulae with numerators presented in Appendix A can be derived by means of (3.12). Let us now present a simple one-loop example and illustrate the trick with turning to integrals without numerators. Consider the Feynman integral corresponding to Fig. 3.4 with a numerator F (q 2 , m2 ; a1 , a2 , a3 , n, d) dd k (l·k)n = , (3.13) 2 a (k − 2p1 ·k) 1 (k 2 − 2p2 ·k)a2 (k 2 − m2 )a3 where l is a momentum not related to p1 and p2 . The alpha representation (2.36) takes the form ia1 +a2 +a3 +ε−1 π d/2 F (q 2 , m2 ; a1 , a2 , a3 , n, d) = (−1)a l Γ (al ) ∞ ∞ ∞ a −1 × dα1 dα2 dα3 αl l U −d/2 exp iV/U − im2 α3 r

κ1

0

×

κN

0

1 ∂ 2i ∂r

n

0

l

exp

! i[2rl·(α1 p1 + α2 p2 ) + r2 l2 ] , α1 + α2 + α3 r=0

(3.14)

where U = α1 + α2 + α3 ,

V = q 2 α1 α2 .

Taking into account the arguments above we see, for example, that 1 F (a1 , a2 , a3 , 1, d) = − [a1 l·p1 F (a1 + 1, a2 , a3 , 0, d + 2) π +a2 l·p2 F (a1 , a2 + 1, a3 , 0, d + 2)] , l2 F (a1 , a2 , a3 , 0, d + 2) F (a1 , a2 , a3 , 2, d) = 2π 1

+ 2 a1 (a1 + 1)(l·p1 )2 F (a1 + 2, a2 , a3 , 0, d + 4) π +2a1 a2 (l·p1 )(l·p2 )F (a1 + 1, a2 + 1, a3 , 0, d + 4) +a2 (a2 + 1)(l·p2 )2 F (a1 , a2 + 2, a3 , 0, d + 4) .

(3.15)

(3.16)

38

3 Evaluating by Alpha and Feynman Parameters

Such a reduction of numerators can be performed for any Feynman integral. The corresponding algebraic manipulations can easily be implemented on a computer.

3.3 One-Loop Examples Let us present examples of evaluation of Feynman diagrams by means of alpha parameters with results which are not written in terms of gamma functions for general d. We ﬁrst turn to the example considered in the introduction. Example 3.1. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. We apply (3.5) to obtain F3.1 (q 2 , m2 ; a1 , a2 , d) = iπ d/2 (−1)a1 +a2 × 0

For example, we have

1

Γ (a1 + a2 + ε − 2) Γ (a1 )Γ (a2 )

dξ ξ a2 −1 (1 − ξ)1−a2 −ε [m2 − q 2 ξ − i0]

a1 +a2 +ε−2

.

(3.17)

dd k (k 2 − m2 )2 (q − k)2 1 (1 − ξ)−ε dξ d/2 = −iπ Γ (1 + ε) 1+ε . 2 2 0 [m − q ξ − i0]

F3.1 (q , m ; 2, 1, d) ≡ 2

2

(3.18)

Suppose that we are interested only in the value of this (ﬁnite) integral exactly in four dimensions. The integral over ξ is then evaluated easily at ε = 0 with the result (1.5). Similarly, Feynman integrals corresponding to Fig. 1.1 with various integer indices ai can be evaluated. In particular, we obtain (1.7). The next one-loop example is Example 3.2. The triangle diagram of Fig. 3.4. The Feynman integral for Fig. 3.4 with general integer indices looks like (3.13) with n = 0, i.e. F3.2 (q 2 , m2 ; a1 , a2 , a3 , d) dd k = , 2 a 2 (k − 2p1 ·k) 1 (k − 2p2 ·k)a2 (k 2 − m2 )a3

(3.19)

where q = p1 − p2 , q 2 ≡ −Q2 = −2p1 ·p2 . The alpha representation (2.36) takes the form (3.14) with n = 0. Introducing variables α1 = ξ1 η, α2 = ξ2 η and α3 = (1 − ξ1 − ξ2 )η and integrating over η we obtain

3.3 One-Loop Examples

F3.2 (q 2 , m2 ; a1 , a2 , a3 , d) = ×

1

dξ1 0

iπ d/2 (−1)a1 +a2 +a3 Γ (a + ε − 2) l Γ (al )

1−ξ1

dξ2 0

39

ξ1a1 −1 ξ2a2 −1 (1 − ξ1 − ξ2 )a3 −1 . [Q2 ξ1 ξ2 + m2 (1 − ξ1 − ξ2 )]a+ε−2

(3.20)

This can be a reasonable starting point for the evaluation of integrals with any given indices ai . Let us evaluate the integral with a1 = a2 = a3 = 1 at d = 4. Then the integral is ﬁnite: 1 1−ξ1 dξ2 . F3.2 (q 2 , m2 ; 1, 1, 1, 4) = −iπ 2 dξ1 2 Q ξ1 ξ2 + m2 (1 − ξ1 − ξ2 ) 0 0 A straightforward integration gives the following result: F3.2 (q 2 , m2 ; 1, 1, 1, 4) iπ 2 1 2 π2 = 2 Li2 (x) − ln x + ln x ln(1 − x) − , Q 2 3

(3.21)

where Li2 (x) is the dilogarithm (see (B.7)) and x = m2 /Q2 . Example 3.3. The massless on-shell box diagram of Fig. 3.5, i.e. with p2i = 0, i = 1, 2, 3, 4.

p1 2

p2

p3

1 4 3

p4

Fig. 3.5. Box diagram

With the loop momentum chosen as the momentum of line 1, the Feynman integral takes the form F3.3 (s, t; a1 , a2 , a3 , a4 , d) dd k = , 2 a 2 a (k ) 1 [(k + p1 ) ] 2 [(k + p1 + p2 )2 ]a3 [(k − p3 )2 ]a4

(3.22)

where s = (p1 + p2 )2 and t = (p1 + p3 )2 are Mandelstam variables. The trees and 2-trees relevant to the functions U and V are shown in Figs. 3.6 and 3.7. Four more existing 2-trees, for example the 2-tree with the component consisting of the lines 1 and 2 and the component consisting of the isolated vertex with the external momentum p4 , do not contribute to the function V because the product α3 α4 is multiplied by the corresponding external momentum squared which is zero. We have (2.36) with

40

3 Evaluating by Alpha and Feynman Parameters

Fig. 3.6. Trees contributing to the function U for the box diagram

Fig. 3.7. 2-trees contributing to the function V for the massless on-shell box diagram

U = α1 + α2 + α3 + α4 , V = tα1 α3 + sα2 α4 .

(3.23)

Introducing new variables by α1 = η1 ξ1 , α2 = η1 (1 − ξ1 ), α3 = η2 ξ2 , α4 = η2 (1 − ξ2 ), with the Jacobian η1 η2 , and evaluating an integral over η2 due to the delta function and an integral over η1 in terms of gamma functions we obtain F3.3 (s, t; a1 , a2 , a3 , a4 , d) Γ (a + ε − 2)Γ (2 − ε − a1 − a2 )Γ (2 − ε − a3 − a4 ) = (−1)a iπ d/2 Γ (4 − 2ε − a) Γ (al ) 1 1 ξ a1 −1 (1 − ξ1 )a2 −1 ξ2a3 −1 (1 − ξ2 )a4 −1 × dξ1 dξ2 1 . (3.24) [−sξ1 ξ2 − t(1 − ξ1 )(1 − ξ2 ) − i0]a+ε−2 0 0 where a = a1 + a2 + a3 + a4 . Consider, for example, the master integral2 with all the indices equal to one. We have F (s, t; d) ≡ F3.3 (s, t; 1, 1, 1, 1, d) = iπ d/2

1

× 0

0

1

Γ (2 + ε)Γ (−ε)2 Γ (−2ε)

dξ1 dξ2 . [−tξ1 ξ2 − s(1 − ξ1 )(1 − ξ2 ) − i0]2+ε

(3.25)

Then the integration over ξ2 results in Γ (1 + ε)Γ (−ε)2 Γ (−2ε)

dξ (−t)−1−ε ξ −1−ε − (−s)−1−ε (1 − ξ)−1−ε . s − (s + t)ξ

F (s, t; d) = −iπ d/2

1

× 0

(3.26)

The singularity at s − (s + t)ξ = 0 is absent because the rest of the integrand is zero at this point. To calculate this integral in expansion in ε one needs, however, to separate the two terms in the square brackets. In order not to run into divergence due to the denominator one can perform an auxiliary subtraction at s − (s + t)ξ = 0. We obtain 2

We shall see in Chaps. 5 and 6 that this is indeed an irreducible Feynman integral.

3.4 Feynman Parameters

F (s, t; d) = −iπ d/2

Γ (1 + ε)Γ (−ε)2 [f (s, t; ε) + f (t, s; ε)] , Γ (−2ε)

where −1−ε

f (s, t; ε) = (−t)

0

1

" −1−ε # s dξ −1−ε ξ . − s − (s + t)ξ s+t

41

(3.27)

(3.28)

To expand the function f in a Laurent series in ε one needs to perform another subtraction, at ξ = 0, which we make by the replacement 1 (s + t)ξ 1 → + . s − (s + t)ξ s(s − (s + t)ξ) s

(3.29)

Then the integral with the ﬁrst term can be evaluated by expanding the integrand in ε while the second term is integrated explicitly. Eventually, we arrive at the following result: iπ d/2 e−γE ε 4 2 F (s, t; d) = − [ln(−s) + ln(−t)] st ε2 ε 2 4π +2 ln(−s) ln(−t) − + O(ε) . (3.30) 3 Here and in all the expansions in ε below we pull out the factor e−γE ε , with Euler’s constant γE , per loop in order to avoid it in our results. Although we are oriented at calculations in expansion in ε, let us, for completeness, present a simple result for general ε [16] which can straightforwardly be obtained from (3.27): t iπ d/2 Γ (−ε)2 Γ (ε) −ε F (s, t; d) = − (−t) 2 F1 1, −ε; 1 − ε; 1 + stΓ (−2ε) s $ s , (3.31) +(−s)−ε 2 F1 1, −ε; 1 − ε; 1 + t where 2 F1 is the Gauss hypergeometric function (see (B.1)).

3.4 Feynman Parameters Let us now present the alpha representation of scalar dimensionally regular ized integrals in a modiﬁed form by making the change of variables αl = ηαl , where αl = 1. Starting from (2.36) with Z = 1, performing the integration over η from 0 to ∞ explicitly and omitting primes from the new variables, we obtain d/2 h iπ Γ (a − hd/2) FΓ (q1 , . . . , qn ; d) = (−1)a l Γ (al ) ∞ ∞ U a−(h+1)d/2 αal −1 l l dα1 . . . dαL δ αl − 1 . (3.32) × a−hd/2 0 0 (−V + U m2l αl )

42

3 Evaluating by Alpha and Feynman Parameters

A folklore Cheng–Wu theorem [5] (see also [2]) says that the same formula (3.32) holds with the delta function

αl − 1 , (3.33) δ l∈ν

where ν is an arbitrary subset of the lines 1, . . . , L, when the integration over the rest of the α-variables, i.e. for l∈ν, is extended to the integration from zero to inﬁnity. Observe that the integration over αl for l ∈ ν is bounded at least by 1 from above, as in the case where all the α-variables are involved in the sum in the argument of the delta function. One can prove this theorem straightforwardly by changing variables and calculating the corresponding Jacobian. But a simpler way to prove it3 is to start from the alpha representation (2.36), introduce new variables by αl = ηαl for all l = 1, 2, . . . , L, where η = l∈ν αl , and immediately arrive at (3.32) with the delta function (3.33). Let us stress that this theorem holds not only for (3.32) corresponding to Feynman diagrams with standard propagators but also for the alpha representation derived for Feynman diagrams with various linear propagators. As we will see below in multiple examples, an adequate choice of the delta function in (3.32) can greatly simplify the evaluation. Note that one can use various homogeneous substitutions which keep the form of the delta function in (3.32) – see Sect. 3.1 of [10] and references therein. In addition to alpha parameters, the closely related Feynman parameters are often used. For a product of two propagators, one writes down the following relation: 1 (m21 − p21 )λ1 (m22 − p22 )λ2 dξ ξ λ1 −1 (1 − ξ)λ2 −1 Γ (λ1 + λ2 ) 1 = . (3.34) 2 2 Γ (λ1 )Γ (λ2 ) 0 [(m1 − p1 )ξ + (m22 − p22 )(1 − ξ)]λ1 +λ2 This relation is usually applied to a pair of appropriately chosen propagators if an explicit integration over a loop momentum then becomes possible. Then new Feynman parameters can be introduced for other factors in the integral, etc. In fact, any choice of the Feynman parameters can be achieved by starting from the alpha representation (3.32) and making certain changes of variables. However, the possibility of an intermediate explicit loop integration of the kind mentioned above can be hidden in the alpha integral. The generalization of (3.34) to an arbitrary number of propagators is of the form 1 λ −1 δ ( ξl − 1) Γ ( λl ) 1 1 , = dξ . . . dξ ξl l (3.35) λl 1 L λl Γ (λl ) 0 Al 0 ( Aξ) l

where Al = 3

m2l

−

l l

p2l .

Thanks to A.G. Grozin for pointing out this possibility!

3.5 Two-Loop Examples

43

For the evaluation of diagrams with a small number of loops, the choice of applying either alpha or Feynman parameters is usually just a matter of taste. In particular, if we apply (3.35) to a two-loop diagram and then integrate over two loop momenta, with the help of (A.1) and its generalizations to integrals with numerators, we obtain the same result as that obtained starting from (3.32). For completeness, here is a one more parametric representation which is related to Feynman parameters and is often used in practice: xλ2 −1 dx Γ (λ1 + λ2 ) 1 1 = . (3.36) λ λ A 1B 2 Γ (λ1 )Γ (λ2 ) 0 (A + Bx)λ1 +λ2

3.5 Two-Loop Examples At the two-loop level, we ﬁrst consider the Example 3.4. Two-loop vacuum diagram of Fig. 3.8 with the masses m, 0, m and general complex powers of the propagators.

1 2 3 Fig. 3.8. Vacuum two-loop diagram with the masses m, 0 and m

The Feynman integral is written as F3.4 (m2 ; λ1 , λ2 , λ3 , d) dd k dd l = . (−k 2 + m2 )λ1 [−(k + l)2 ]λ2 (−l2 + m2 )λ3

(3.37)

The two basic functions in the alpha representation are U = α1 α2 +α2 α3 + α3 α1 and V = 0. We apply (3.32) to obtain

∞ ∞ ∞ 3 2 Γ (λ + 2ε − 4) λl −1 d/2 αl dαl F3.4 = iπ Γ (λl )(m2 )λ+2ε−4 0 0 0 l=1

(α1 α2 + α2 α3 + α3 α1 )ε−2 ×δ αl − 1 . (3.38) (α1 + α3 )λ+2ε−4 l

Now we exploit the freedom provided by the Cheng–Wu theorem and choose the argument of the delta function as α1 + α3 − 1. The integration over α2 is

44

3 Evaluating by Alpha and Feynman Parameters

performed from 0 to ∞. Resulting integrals are evaluated in terms of gamma functions for general ε and we arrive at the table formula (A.38). Consider now Example 3.5. Two-loop massless propagator diagram of Fig. 3.9 with arbitrary integer powers of the propagators,

Fig. 3.9. Two-loop propagator diagram

F3.5 (q 2 ; a1 , a2 , a3 , a4 , a5 , d) dd k dd l = . (k 2 )a1 [(q − k)2 ]a2 (l2 )a3 [(q − l)2 ]a4 [(k − l)2 ]a5

(3.39)

The sets of trees and 2-trees relevant to the two basic functions in the alpha representation are shown in Figs. 3.10 and 3.11

Fig. 3.10. Trees contributing to the function U for Fig. 3.9

Fig. 3.11. 2-trees contributing to the function V for Fig. 3.9

Correspondingly, we have U = (α1 + α2 + α3 + α4 )α5 + (α1 + α2 )(α3 + α4 ) , V = [(α1 + α2 )α3 α4 + α1 α2 (α3 + α4 ) + (α1 + α3 )(α2 + α4 )]q ≡ Vq 2 .

(3.40) 2

(3.41)

As we will see in Chaps. 5 and 6, any diagram of this class can be evaluated for general ε in terms of gamma functions. This is however hardly seen from

3.5 Two-Loop Examples

45

its alpha representation. In spite of the fact that the evaluation by alpha parameters is not an optimal method for this class of integrals, let us evaluate, for the sake of illustration, this diagram for all powers of the propagators equal to one, using its alpha representation. It is ﬁnite at d = 4, both in the UV and IR sense. Representation (3.32) takes the form ∞ δ ( αl − 1) (iπ 2 )2 ∞ dα . . . dα . (3.42) F3.5 (q 2 ; 1, 1, 1, 1, 1, 4) = 1 5 q2 UV 0 0 We exploit the Cheng–Wu theorem by choosing the delta function δ (α5 − 1), with the integration over the rest of the four variables from zero to inﬁnity. Then one can delegate the integration procedure to MATHEMATICA [22] and obtain the well-known result4 : 2 2 iπ F3.5 (q 2 ; 1, 1, 1, 1, 1, 4) = 6ζ(3) , (3.43) q2 where ζ(z) is the Riemann zeta function. In the rest of this chapter, we shall consider just two more examples which are, however, more complicated than the previous ones. Example 3.6. Two classes of two-loop integrals5 with integer powers of the propagators: dd k dd l . (3.44) F± (q 2 ; a1 , a2 , a3 ) = 2 a 1 (k + q·k) (l2 + q·l)a2 [(k ± l)2 ]a3 It turns out that the F− is simple. Indeed we rewrite the ﬁrst denominator k 2 + q ·k as (k + q/2)2 − q 2 /4 and similarly the second denominator, make the change of variables k = k − q/2, l = l − q/2 and recognize F− as a two-loop vacuum diagram with the mass m2 = q 2 /4 shown in Fig. 3.8 which was evaluated in Example 3.4 – see (A.38). The integrals F+ are, however, not so simple. Using the same manipulation as above we see that they are graphically recognized as sunset diagrams of Fig. 3.12 at threshold, i.e. q 2 = 4m2 . We start from the alpha representation (2.36) with Z = 1. The two basic functions are U = α1 α2 + α2 α3 + α3 α1 , V = α1 α2 α3 q 2 . 2

(3.45)

2

After using the threshold condition m = q /4 we obtain

4

This result was ﬁrst obtained in [19] by means of expansion in Chebyshev polynomials in momentum space. In [6], it was reproduced using Gegenbauer polynomials in coordinate space. 5 They were involved, in particular, in the calculation [1, 8] of two-loop matching coeﬃcients of the vector current in QCD and Non-Relativistic QCD (NRQCD) [3, 14, 21].

46

3 Evaluating by Alpha and Feynman Parameters

1 2 3 Fig. 3.12. Sunset diagram with the masses m, m, 0

(−1)a ia+2ε−2 Γ (al )

! ∞ ∞ ∞ 3 q2 W al −1 ε−2 αl dαl U exp −i × , 4U 0 0 0

F+ (q 2 ; a1 , a2 , a3 ) =

(3.46)

l=1

where W = (α1 + α2 )α1 α2 + α3 (α1 − α2 )2 .

(3.47)

Proceeding as with the general alpha representation we come to 2 (−1)a iπ d/2 Γ (a + 2ε − 4) 2 F+ (q ; a1 , a2 , a3 ) = Γ (al ) (q 2 /4)a+2ε−4

3 ∞ ∞ ∞ U a+3ε−6 al −1 δ αl − 1 αl dαl . × W a+2ε−4 0 0 0

(3.48)

l=1

We continue to exploit the Cheng–Wu theorem in an appropriate way. We choose the delta function in (3.48) as δ (α1 + α2 − 1) and obtain an integral over ξ = α1 from 0 to 1, with α2 = 1 − ξ, and an integral over t = α3 from 0 to ∞: 2 (−1)a iπ d/2 Γ (a + 2ε − 4) F+ (q 2 ; a1 , a2 , a3 ) = Γ (al ) (q 2 /4)a+2ε−4 ∞ 1 a3 −1 [t + ξ(1 − ξ)]a+3ε−6 t dξ ξ a1 −1 (1 − ξ)a2 −1 dt . (3.49) × [t(1 − 2ξ)2 + ξ(1 − ξ)]a+2ε−4 0 0 This two-parametric integral representation can be used for the evaluation of any diagram of the given class in expansion in ε. Let us show how the integral with all the indices equal to one can be evaluated in expansion in ε up to the ﬁnite part. We start with (3.49) which gives d/2 2 iπ Γ (2ε − 1) 2 F+ (q ; 1, 1, 1) = − 2 (q /4)2ε−1 1 ∞ [t + ξ(1 − ξ)]3ε−3 × dξ dt . (3.50) [t(1 − 2ξ)2 + ξ(1 − ξ)]2ε−1 0 0 Observe that the integrand is invariant under the transformation ξ → 1 − ξ. We write the integral as √ twice the integral from 0 to 1/2 over ξ, change the variable ξ by ξ = (1 − 1 − x)/2 and rescale t → t/4 to obtain

3.5 Two-Loop Examples

47

2

F+ (q 2 ; 1, 1, 1) = − iπ d/2 Γ (2ε − 1)(q 2 /2)1−2ε 1 ∞ dx [t(1 − x) + x]1−2ε √ × dt . (t + x)3−3ε 1−x 0 0

(3.51)

Remember that our integral is UV divergent. The overall divergence is quadratic since the UV degree of divergence is ω = 2, and there are three oneloop logarithmically divergent subgraphs, so that, presumably, there should be poles up to the second order in ε. One source of the poles is the overall gamma function Γ (2ε − 1). Another power of 1/ε comes from the integration over t and x in (3.51), namely from the region of small t and x. To have the possibility to perform an expansion in ε we have to reveal the singularity at ε = 0. Similarly to what we did in Example 3.3, let us perform a subtraction according to the identity [t(1 − x) + x]1−2ε = [t(1 − x) + x]1−2ε − (t + x)1−2ε + (t + x)1−2ε . Now, the integral with the expression in braces can be evaluated by expanding the integrand in a Laurent series in ε, while the last term can be integrated by hand with a result expressed in terms of gamma functions which can be, of course, expanded in ε after the evaluation: √ 1 ∞ πΓ (ε) dx dt √ . = 2−ε (t + x) (1 − ε)Γ (ε + 1/2) 1 − x 0 0 The integration of the subtracted part up to order ε0 can straightforwardly be done by MATHEMATICA [22]. Finally, we obtain the following result: 2 q 2 1−2ε F+ (q 2 ; 1, 1, 1) = iπ d/2 e−γE ε 4 1 1 2 11π 2 − + O(ε) . (3.52) × 2+ + ε ε 12 2 Consider now Example 3.7. Non-planar two-loop massless vertex diagram of Fig. 3.13 with p21 = p22 = 0. The Feynman integral can be written as dd k dd l 2 F3.7 (Q ; a1 , . . . , a6 , d) = [(k + l)2 − 2p1 ·(k + l)]a1 1 × , (3.53) 2 a 2 2 [(k + l) − 2p2 ·(k + l)] (k − 2p1 ·k)a3 (l2 − 2p2 ·l)a4 (k 2 )a5 (l2 )a6 where Q2 = −(p1 − p2 )2 = 2p1 ·p2 , and the loop momenta are chosen as the momenta ﬂowing through lines 5 and 6. Let us proceed by Feynman parameters following [11] where some integrals of this class were calculated. (They were also evaluated in [13] and [16].)

48

3 Evaluating by Alpha and Feynman Parameters

Fig. 3.13. Non-planar vertex diagram

We write down Feynman parametric formula (3.34) for the pairs of the propagators (3, 5) and (4, 6): 1 (−1)a3 +a5 Γ (a3 + a5 ) = (k 2 − 2p1 ·k)a3 (k 2 )a5 Γ (a3 )Γ (a5 ) 1 dξ1 ξ1a3 −1 (1 − ξ1 )a5 −1 × 2 a3 +a5 0 [−(k − ξ1 p1 ) − i0]

(3.54)

and, similarly, for the second pair, with the replacements ξ1 → ξ2 , p1 → p2 , k → l, a3 → a4 , a5 → a6 . Then we change the integration variable l → r = k + l and integrate over k by means of our one-loop tabulated formula (3.6): dk [−(k − ξ1 p1 )2 ]a3 +a5 [−(r − ξ2 p2 − k)2 ]a4 +a6 G(a3 + a5 , a4 + a6 ) = iπ d/2 . (3.55) [−(r − ξ1 p1 − ξ2 p2 )2 ]a3 +a4 +a5 +a6 +ε−2 Then we apply Feynman parametric formula (3.35) to the propagators 1 and 2 and the propagator resulting from the right-hand side of (3.55), with a resulting integral over r evaluated by (A.1): dd r 2 2 [−(r − Q A(ξ1 , ξ2 , ξ3 , ξ4 ))]a+ε−2 1 Γ (a + 2ε − 4) = iπ d/2 , (3.56) Γ (a + ε − 2) (Q2 )a+2ε−4 A(ξ1 , ξ2 , ξ3 , ξ4 )a+2ε−4 where a = a1 + . . . + a6 and A(ξ1 , ξ2 , ξ3 , ξ4 ) = ξ3 ξ4 + (1 − ξ3 − ξ4 )[ξ2 ξ3 (1 − ξ1 ) + ξ1 ξ4 (1 − ξ2 )] . Thus we arrive at the following intermediate result valid for general powers of the propagators:

3.5 Two-Loop Examples

49

2 (−1)a iπ d/2 Γ (2 − ε − a35 )Γ (2 − ε − a46 ) (Q2 )a+2ε−4 Γ (al )Γ (4 − 2ε − a3456 ) 1 1 dξ1 . . . dξ4 ξ1a3 −1 (1 − ξ1 )a5 −1 ξ2a4 −1 (1 − ξ2 )a6 −1 ×Γ (a + 2ε − 4)

F3.7 (Q2 ; a1 , . . . , a6 , d) =

0

0

a3456 +ε−3 ×ξ3a1 −1 ξ4a2 −1 (1 − ξ3 − ξ4 )+ A(ξ1 , ξ2 , ξ3 , ξ4 )4−2ε−a .

(3.57)

We use the shorthand notation a35 = a3 + a5 , a3456 = a3 + a4 + a5 + a6 . As usually, X+ = X for X > 0 and X+ = 0 otherwise. This four-parametric integral representation can be used for the evaluation of Feynman integrals of this class with various indices. Let us use it in the case a1 = . . . = a6 = 1 and evaluate the corresponding Feynman integral in expansion in ε up to the ﬁnite part. We have d/2 2 iπ Γ (2 + 2ε)Γ (−ε)2 2 F3.7 (Q ; 1, . . . , 1, d) = 2 2+2ε (Q ) Γ (−2ε) 1 1 (1 − ξ3 − ξ4 )1+ε + × dξ1 . . . dξ4 . (3.58) 2+2ε A(ξ , ξ , ξ , ξ ) 1 2 3 4 0 0 We introduce new variables by ξ3 = ξη, ξ4 = (1 − ξ)η and integrate over ξ2 to obtain d/2 2 iπ Γ (1 + 2ε)Γ (−ε)2 1 2 dη η −1−2ε (1 − η)ε F3.7 (Q ; 1, . . . , 1, d) = − 2 2+2ε (Q ) Γ (−2ε) 0 1 1 dξdξ1 −1−2ε ξ × [(1 − ξ)η + (1 − η)(1 − ξ1 )]−1−2ε 0 0 ξ − ξ1 (3.59) −(1 − ξ)−1−2ε [ξη + (1 − η)ξ1 ]−1−2ε . The singularity of the denominator at ξ = ξ1 is spurious because the numerator is zero at this point. We notice that, due to the symmetry of the integrand, the integral over ξ and ξ1 equals twice the integral over the domain 0 ≤ ξ1 ≤ ξ ≤ 1. Following [11] again, we turn to the variable z by ξ1 = zξ, make the changes η → 1 − η, z → 1 − z and come to d/2 2 iπ Γ (1 + 2ε)Γ (−ε)2 2 f (ε) , (3.60) F3.7 (Q ; 1, . . . , 1, d) = −2 2 2+2ε (Q ) Γ (−2ε) where

f (ε) = × 0

0 1

1

dη η ε (1 − η)−1−2ε

1

dξ ξ −1−2ε

0

dz [1 − ξ(1 − ηz)]−1−2ε − (1 − ξ)−1−2ε (1 − ηz)−1−2ε . z

(3.61)

At this point it is claimed in [11] that, in principle, it is possible to evaluate this integral, in expansion in ε up to the ﬁnite part, performing appropriate subtractions of the integrand. Still another way was chosen: to expand

50

3 Evaluating by Alpha and Feynman Parameters

various quantities of the type (1 − X)λ in a binomial series, with subsequent integration and summing up resulting multiple series. (This procedure can be qualiﬁed as another method of evaluation.) Let us, however, realize the possibility of making subtractions. Indeed, the situation is complicated because we are dealing with a three-parametric integral so that several subtractions that would reveal the singularities that generate poles in ε are necessary. Since the prefactor in (3.60) involves a simple pole in ε we have to evaluate the function f (ε) given by (3.61) up to order ε1 . There are several sources of the poles: the points ξ = 0, ξ = 1, η = 0, η = 1, and z = 1. The following strategy of subtractions is suitable for the calculation. Let us ﬁrst decompose f into the sum f1 + f2 according to the subtraction of the braces in (3.61) at η = 0, i.e.

(1 − ξ(1 − ηz))−1−2ε − (1 − ξ)−1−2ε

(3.62) +(1 − ξ)−1−2ε 1 − (1 − ηz)−1−2ε . Let us start with f1 . We perform subtraction of the integrand at η = 1 according to the decomposition of the ﬁrst part of (3.62) into

(1 − ξ(1 − z))−1−2ε − (1 − ξ)−1−2ε

(3.63) + (1 − ξ(1 − ηz))−1−2ε − (1 − ξ(1 − z))−1−2ε . The ﬁrst term in (3.63) does not depend on η so that the corresponding integration over η is performed in terms of gamma functions. Then the integral 1 1 dz [1 − ξ(1 − z)]−1−2ε − (1 − ξ)−1−2ε dξ ξ −1−2ε z 0 0 appears. We need a subtraction at ξ = 1 here because when ξ → 1 the factor z −1−2ε generating a pole in ε arises. So we replace ξ −1−2ε by 1+ ξ −1−2ε − 1 . The ﬁrst term corresponding to unity, after integration over ξ, gives the following integral evaluated in terms of gamma functions 1 dz 1 − z −1−2ε = ψ(−2ε) + γE , 0 1−z where ψ(z) is the logarithmical derivative of the gamma function, i.e. ψ(z) = Γ (z)/Γ (z). Thus we obtain the following contribution to our result: Γ (1 + ε)Γ (−2ε) 2εΓ (1 − ε) 3ζ(3) 3π 4 π2 1 − − ε + O(ε2 ) . (3.64) = 3− 8ε 24ε 4 80 Starting from the second term we obtain an integral which can be evaluated by expanding the integrand in ε and performing the integration, e.g., in MATHEMATICA [22], with the following contribution: f11 = −

f12 =

43π 4 π2 + 5ζ(3) + ε + O(ε2 ) . 12ε 180

(3.65)

3.5 Two-Loop Examples

51

In the second part of (3.63), we make the same replacement (with the same motivation) as before, i.e. ξ −1−2ε → 1 + ξ −1−2ε − 1 . The second part here again produces an integral which can be evaluated by expanding the integrand in ε, with the following contribution: 11π 4 ε + O(ε2 ) . (3.66) 120 The unity gives a part where the integration over ξ is explicitly taken. The corresponding result is proportional to the sum of these two two-parametric integrals: 1 1 dηdzη ε (1 − η)−1−2ε 1 − η −1−2ε 0 0 1 1 −2ε 1 − z −2ε ε −1−2ε 1 − (ηz) − dηdzη (1 − η) + . (3.67) 1 − ηz 1−z 0 0 f13 = ζ(3) +

The ﬁrst integral can be evaluated in terms of gamma functions, with the following contribution: Γ (1 − ε) Γ (−2ε) Γ (1 + ε) − f14 = 4ε2 Γ (1 − ε) Γ (1 − 3ε) 2 π4 π − ζ(3) − ε + O(ε2 ) . (3.68) =− 12ε 36 In the second integral, one can expand the integrand in ε. Here is the corresponding contribution: π4 ε + O(ε2 ) . (3.69) 72 Let us now deal with f2 deﬁned by the second part of (3.62). The integration over ξ is performed explicitly, and the following integral over z arises: 1 dz

(1 − ηz)−1−2ε − 1 . z 0 f15 = −ζ(3) −

When z → 1 a factor (1 − η)−1−2ε appears so that we need a subtraction at z = 1. We make the replacement 1/z → 1 + (1 − z)/z. The unity generates a part which is integrated explicitly over z and then over η. The resulting contribution is then Γ (−2ε)2 Γ (ε) 1 Γ (−4ε) Γ (−2ε) Γ (−2ε) − f21 = − + Γ (−4ε) 2ε Γ (−3ε) Γ (−ε) Γ (−ε) 2 2 4 29π π 1 π 1 − + 2ζ(3) + − 7ζ(3) ε + O(ε2 ) . = 3+ 2+ 8ε 2ε 12ε 6 360 (3.70) Starting from the second term and performing one more subtraction we obtain the following integral

52

3 Evaluating by Alpha and Feynman Parameters

0

1

1−z dηdzη ε (1 − η)−1−2ε z 0

−1−2ε − (1 − z)−1−2ε + (1 − z)−1−2ε − 1 . × (1 − ηz) 1

(3.71)

For the part corresponding to the second square brackets, one can explicitly integrate over η and then expand the integrand in ε and integrate over z with the following resulting contribution: Γ (−2ε)3 Γ (1 + ε) 1 f22 = − + 1 − ψ(−2ε) − γE Γ (−4ε)Γ (1 − ε) 2ε 4 π π2 π2 1 + − 2ζ(3) + + 7ζ(3) ε + O(ε2 ) . (3.72) =− 2 − 2ε 6ε 6 90 For the part corresponding to the ﬁrst square brackets in (3.71), one can expand the integrand in ε and integrate over z and η with the following resulting contribution: 19π 4 π2 − 9ζ(3) + ε + O(ε2 ) . (3.73) 6ε 45 Collecting all the eight contributions obtained and taking into account the prefactor in (3.60) we arrive at the well-known analytical result [11] d/2 −γ ε 2 iπ e E 2 F3.7 (Q ; 1, . . . , 1, d) = (Q2 )2+2ε 1 π2 83ζ(3) 59π 4 − × 4− 2 − + O(ε) . (3.74) ε ε 3ε 120 f23 = −

In [11], a similar algorithm based on Feynman parameters has been developed for the evaluation of planar massless two-loop vertex diagrams. It has turned out that the evaluation, by Feynman parameters, in the planar case is more complicated. As we will see in Chaps. 5 and 6, there is, however, a better choice of an appropriate method in this situation and the planar vertex diagrams of this class are in fact much simpler than the non-planar ones.

References 1. M. Beneke, A. Signer and V.A. Smirnov, Phys. Rev. Lett. 80 (1998) 2535. 45 2. K.S. Bjoerkevoll, P. Osland and G. Faeldt, Nucl. Phys. B 386 (1992) 303. 42 3. G.T. Bodwin, E. Braaten and G.P. Lepage, Phys. Rev. D 51 (1995) 1125; Phys. Rev. D 55 (1997) 5853. 45 4. N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 5. H. Cheng and T.T. Wu, Expanding Protons: Scattering at High Energies (MIT Press, Cambridge, MA, 1987). 42 6. K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Nucl. Phys. B 174 (1980) 345. 45

References 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

53

J.C. Collins, Renormalization (Cambridge University Press, Cambridge, 1984). 31 A. Czarnecki and K. Melnikov, Phys. Rev. Lett. 87 (2001) 013001. 45 A.I. Davydychev, Phys. Lett. B 263 (1991) 107. 36 A.I. Davydychev and R. Delbourgo, J. Math. Phys. 39 (1998) 4299. 42 R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542. 47, 49, 52 A.G. Grozin, Heavy Quark Eﬀective Theory (Springer, Berlin, Heidelberg, 2004). 33 G. Kramer and B. Lampe, J. Math. Phys. 28 (1987) 945. 47 G.P. Lepage et al., Phys. Rev. D 46 (1992) 4052. 45 A.V. Manohar and M.B. Wise, Heavy Quark Physics (Cambridge University Press, Cambridge, 2000). 33 W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 41, 47 M. Neubert, Phys. Rep. 245 (1994) 259. 33 M.E. Peskin and D.V. Schroeder, An Introduction to Quantum Field Theory (Perseus, Reading, MA, 1995). 31 J.L. Rosner, Ann. Phys. 44 (1967) 11. 45 O.V. Tarasov, Nucl. Phys. B 480 (1996) 397; Phys. Rev. D 54 (1996) 6479. 36 B.A. Thacker and G.P. Lepage, Phys. Rev. D 43 (1991) 196. 45 S. Wolfram, The Mathematica Book, 4th edition (Wolfram Media and Cambridge University Press, Cambridge, 1999). 35, 45, 47, 50

4 Evaluating by MB Representation

One often uses Mellin integrals1 when dealing with Feynman integrals. These are integrals over contours in a complex plane along the imaginary axis of a product and ratio of gamma functions. In particular, the inverse Mellin transform is given by such an integral. We shall, however, deal with a very speciﬁc technique in this ﬁeld. The key ingredient of the method presented in this chapter is the MB representation used to replace a sum of two terms raised to some power by the product of these terms raised to some powers. Our goal is to use such a factorization in order to achieve the possibility to perform integrations in terms of gamma functions, at the cost of introducing extra Mellin integrations. Then one obtains a multiple Mellin integral of gamma functions in the numerator and denominator. The next step is the resolution of the singularities in ε by means of shifting contours and taking residues. It turns out that multiple MB integrals are very convenient for this purpose. The ﬁnal step is to perform at least some of the Mellin integrations explicitly, by means of the ﬁrst and the second Barnes lemma and their corollaries and/or evaluate these integrals by closing the integration contours in the complex plane and summing up corresponding series. In Sect. 4.1 we start with simple one-loop examples. In Sect. 4.2 we discuss general properties of multiple MB integrals we are going to deal with. We continue in Sect. 4.3 with typical one-loop examples. In fact we shall illustrate the method of MB representation mainly by the same characteristic examples as in the case of the method of alpha and Feynman parameters in Chap. 3. Let us stress, however, that, for double and triple boxes, complete analytical calculations strictly by means of alpha and Feynman parameters, or, by some other techniques, are not known. We turn to various two-loop examples of massless and massive diagrams in Sects. 4.4 and 4.5, respectively. We then consider three- and even four-loop examples in Sects. 4.6 and 4.7. In Sect. 4.8, we discuss how multiple MB integrals can be used to obtain asymptotic expansions of Feynman integrals in various limits and compare this procedure with expansion by regions [4, 27]. In the last section, we also discuss some other results obtained by means of MB integrals and summarize basic characteristic features of the method presented in this chapter. 1

First examples of application of Mellin integrals to Feynman integrals can be found in [5, 34]. V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 55–107 (2004) c Springer-Verlag Berlin Heidelberg 2004

56

4 Evaluating by MB Representation

4.1 One-Loop Examples Our basic tool is the following formula: +i∞ 1 1 1 Yz = dz Γ (λ + z)Γ (−z) . (X + Y )λ Γ (λ) 2πi −i∞ X λ+z

(4.1)

Here the contour of integration is chosen in the standard way: the poles with a Γ (. . . + z) dependence (let us call them left poles, for brevity) are to the left of the contour and the poles with a Γ (. . . − z) dependence (right poles) are to the right of it. See Fig. 4.1, where a possible contour C is shown in the case of λ = −1/4 − i/2. (This terminology is useful and, although it often happens that the ﬁrst right pole is to the left of the ﬁrst left pole of a given integrand, this, hopefully, will not cause misunderstanding.) Im z C

−λ − 2

-2

−λ − 1

-1

2

1 −λ 0

1

Re z 2

-1

-2 Fig. 4.1. Possible integration contour in (4.1) for λ = −1/4 − i/2

We shall use decompositions X + Y of various functions in integrals over Feynman and alpha parameters. But a more transparent way2 to apply this representation is to write down a massive propagator in terms of massless ones: +i∞ 1 1 (m2 )z 1 = dz Γ (λ + z)Γ (−z) . (4.2) 2 2 λ (m − k ) Γ (λ) 2πi −i∞ (−k 2 )λ+z Our ﬁrst example is the same as Example 3.1: 2

Historically, it was ﬁrst advocated and applied in [8].

4.1 One-Loop Examples

57

Example 4.1. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. We insert (4.2) with λ = a1 into (1.2), apply (3.6) and obtain the following result: F4.1 (q 2 , m2 ; a1 , a2 , d) =

iπ d/2 (−1)a1 +a2 Γ (2 − ε − a2 ) Γ (a1 )Γ (a2 )(−q 2 )a1 +a2 +ε−2 +i∞ 2 z m 1 × dz Γ (a1 + a2 + ε − 2 + z) 2πi −i∞ −q 2 Γ (2 − ε − a1 − z)Γ (−z) × . (4.3) Γ (4 − 2ε − a1 − a2 − z)

The rules for choosing an integration contour that goes from −i∞ to +i∞ in the complex z-plane are the same as before: the right poles (in Γ (. . . − z)) are to the right of the contour and the left poles (in Γ (. . . + z)) are to left. This representation can be used to evaluate any integral of this family in a Laurent expansion in ε. In particular, for F4.1 (q 2 , m2 ; 2, 1, d), we obtain (1.9) and, at d = 4 come to +i∞ 2 z m Γ (1 + z)Γ (−z)2 iπ 2 1 (4.4) dz F4.1 (2, 1, 4) = 2 q 2πi −i∞ −q 2 Γ (1 − z) with an integration contour at −1 < Rez < 0. Using properties of the gamma function we obtain (1.10). Here is a subtle point: if we look at (1.10) we observe that there is a product Γ (z)Γ (−z) which would be bad if it was present from the beginning because we could not satisfy our agreement about choosing the integration contours. Indeed, here the right and left poles at ε = 0 glue together and there is no space between them. However, the situation is unambiguous because we have ﬁxed an integration contour with −1 < Rez < 0 and we are free to perform identical transformations of the integrand after that. A moral of this discussion is the recipe to derive the MB representation for general powers of the propagators al and ﬁx appropriate integration contours at this point. Then, for concrete integer indices al , we are allowed to make transformations like Γ (1 + z)Γ (−z) = −Γ (z)Γ (1 − z), but it is necessary to remember about the choice of the contours made before this. The integral (1.10) can be evaluated, according to the Cauchy theorem, by closing the integration contour to the right and taking a series of residues (with the minus sign, at the points z = 0, 1, 2, . . .. The residue of course) at z = 0 gives iπ 2 ln −q 2 /m2 /q 2 and the residues at z = 1, 2, . . . give the series n ∞ iπ 2 1 m2 . − 2 q n=1 n q 2 As a result, we reproduce (1.5).

58

4 Evaluating by MB Representation

In the case of the indices equal to one we use (4.3) to obtain iπ 2 Γ (1 − ε) F4.1 (q 2 , m2 ; 1, 1, d) = (−q 2 )ε 2 z m Γ (ε + z)Γ (−z)Γ (1 − ε − z) 1 . × dz 2πi C −q 2 Γ (2 − 2ε − z)

(4.5)

To evaluate MB integrals in a Laurent expansion in ε the ﬁrst point is to analyse how singularities in ε are generated. We know in advance that the given integral has a pole in ε because the diagram is UV-divergent. There are no explicit functions with singularities in ε so that the pole is generated by the MB integration. Indeed, the product Γ (ε + z)Γ (−z) generates a singularity in ε when ε → 0 because the ﬁrst left pole, i.e. at z = −ε, and the ﬁrst right pole, i.e. z = 0, glue together when ε = 0, and there is no place for a contour between these poles. Possible integration contours C in (4.5) in the cases Re ε > 0 and Re ε < 0 are shown in Figs. 4.2 and 4.3, respectively. In the former case, a contour can be chosen as a straight line parallel to the imaginary axis, while in the latter case, there is no such choice. However, no matter which value of ε we can imagine, we shall use the same procedure to reveal the pole in ε: we write down the integral (4.5) as the sum of a similar integral over a new contour, C , which goes to the left of the pole at z = −ε and the residue at this point. In the integral over the shifted contour, the nature of the pole at z = −ε changes, and it becomes right, rather than left, in our terminology.

C

Im z C

2

1 −ε

−ε − 1 -2

-1

1−ε 0

1

Re z 2

-1

-2 Fig. 4.2. Possible integration contour in (4.5) in the case Re ε > 0

4.1 One-Loop Examples

59

Im z

C, C

2

1 −ε − 2 -2

−ε − 1 -1

C −ε

C 0

1−ε 1

Re z 2

-1

-2 Fig. 4.3. Possible integration contour in (4.5) in the case Re ε < 0

The crucial point is that, in the integral over C , we can safely expand the integrand in a Laurent series in ε. (In this particular example, this is just a Taylor series.) As to the residue, it is equal to iπ 2

Γ (ε) (m2 )ε (1 − ε)

and can explicitly be expanded in ε. For the integral over the shifted contour C , with −1 < Rez < 0, we obtain, at ε = 0, 2 z m Γ (z)Γ (−z) 2 1 . iπ dz 2πi C −q 2 1−z This MB integral can be evaluated by closing the integration contour to the right in the complex z-plane, as in the previous example. Combining the corresponding result with the residue calculated above we arrive at (1.7). In fact, we could similarly proceed by moving the contour C across the right pole at z = 0 and, correspondingly, taking minus residue at this point. Then the integral over the new contour C would be at 0 < Rez < 1. The next example is the same as Example 3.2: Example 4.2. The triangle diagram of Fig. 3.4. We again exploit the MB representation in the simplest way, i.e. apply (4.2) to the only massive propagator in (3.19), and evaluate the resulting massless triangle integral by (A.28) to obtain the following result:

60

4 Evaluating by MB Representation

(−1)a iπ d/2 Γ (al )(Q2 )a+ε−2 +i∞ 2 z m 1 × dz Γ (a3 + z)Γ (a + ε − 2 + z) 2πi −i∞ Q2 Γ (2 − ε − a1 − a3 − z)Γ (2 − ε − a2 − a3 − z)Γ (−z) × , Γ (4 − 2ε − a − z)

F4.2 (Q2 , m2 ; a1 , a2 , a3 , d) =

(4.6)

where a = a1 + a2 + a3 and Q2 = −(p1 − p2 )2 as above. Consider, as in Chap. 3, the diagram with the powers of the propagators equal to one: F4.2 (Q2 , m2 ; 1, 1, 1, d) = − ×

1 2πi

+i∞

dz −i∞

m2 Q2

z

iπ d/2 (Q2 )1+ε

Γ (1 + ε + z)Γ (1 + z)Γ (−ε − z)2 Γ (−z) . (4.7) Γ (1 − 2ε − z)

If we want to calculate this integral at ε = 0, we observe that we can safely set ε = 0 in the integrand because the right and left poles in the complex z-plane are well separated. We obtain F4.2 (Q2 , m2 ; 1, 1, 1, 4) =

iπ 2 (Q2 ) 1 × 2πi

+i∞

dz −i∞

m2 Q2

z

Γ (1 + z)2 Γ (−z)2 , (4.8) z

where the integration contour can be chosen with −1 < Rez < 0. The integral can be evaluated by the same procedure as before, with the known result (3.21). Any integral (3.19) with integer indices can be evaluated using (4.6). For example, +i∞ 2 z m iπ d/2 1 2 2 dz F4.2 (Q , m ; 2, 1, 1, d) = (Q2 )2+ε 2πi −i∞ Q2 Γ (2 + ε + z)Γ (1 + z)Γ (−1 − ε − z)Γ (−ε − z)Γ (−z) . (4.9) × Γ (−2ε − z) We know in advance that there should be an IR pole in ε because of the second power of the ﬁrst massless propagator so that we anticipate that a pole is generated by the MB integration. Indeed, we observe that the only source of the singularity in ε is the product Γ (1+z)Γ (−1−ε−z). When ε → 0 the ﬁrst left pole (from Γ (1 + z)) and the ﬁrst right pole (from Γ (−1 − ε − z)) tend to each other and there is no place for an integration contour to go between them. To evaluate (4.9) in expansion in ε we apply the strategy formulated above: we turn to the integral over a shifted contour which goes to the left of the ﬁrst pole of Γ (−1 − ε − z) so that this pole changes its nature, i.e. becomes left. According to the Cauchy theorem, (4.9) equals the integral over

4.1 One-Loop Examples

61

the shifted contour minus residue of the integrand at the point z = −1 − ε. Then the integral is evaluated by closing the contour (which can again be taken at −1 < Rez < 0) to the right and summing up a series of residues at the points z = 0, 1, 2, . . .). We thus obtain iπ d/2 e−γE ε F4.2 (Q2 , m2 ; 2, 1, 1, d) = − Q2 1 1 ln(−m2 /Q2 ) 2 − ln m × + O(ε) . + m2 ε m2 − Q2

(4.10)

As before, we again had two options: to change the nature of the ﬁrst pole of Γ (−1−ε−z) or the ﬁrst pole of Γ (1+z). Let us agree, for deﬁniteness, that we shall always try to obtain MB integrals expanded in ε at −1 < Rez < 0. The next example is the same as Example 3.3: Example 4.3. The massless on-shell box diagram of Fig. 3.5, i.e. with p2i = 0, i = 1, 2, 3, 4. Up to now we applied MB representation using (4.2). Let us start with (3.24). The natural idea here is to apply (4.1) to the denominator of the integrand. We do this with X = −sξ1 ξ2 . After that we change the order of integration over z and the parameters ξ1 and ξ2 and evaluate the parametric integrals in terms of gamma functions: F4.3 (s, t; a1 , a2 , a3 , a4 , d) =

(−1)a iπ d/2 Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞ z t 1 dz Γ (a + ε − 2 + z)Γ (a2 + z)Γ (a4 + z)Γ (−z) 2πi −i∞ s ×Γ (2 − a1 − a2 − a4 − ε − z)Γ (2 − a2 − a3 − a4 − ε − z) , (4.11)

×

where a = a1 + a2 + a3 + a4 . One can use this representation to evaluate any box with integer powers of the propagators in expansion in ε. In particular, F (s, t; d) ≡ F4.3 (s, t; 1, 1, 1, 1, d) = ×

1 2πi

+i∞

dz −i∞

iπ d/2 Γ (−2ε)(−s)2+ε

z t Γ (2 + ε + z)Γ (1 + z)2 Γ (−1 − ε − z)2 Γ (−z) . (4.12) s

The way how poles in ε are generated is already familiar: we immediately identify the product Γ (1 + z)2 Γ (−1 − ε − z)2 responsible for that. The only diﬀerence with the previous cases is that the left poles in Γ (1 + z)2 and the right poles in Γ (−1 − ε − z)2 are of the second order. After this analysis we proceed as before: take minus residue at z = −1 − ε and turn to the integral over the contour which goes to the right of it. The contribution of the residue is

62

4 Evaluating by MB Representation

iπ

d/2

Γ (1 + ε)Γ (−ε)2 t ln + 2ψ(−ε) − ψ(1 + ε) + γE , Γ (−2ε)s(−t)1+ε s

(4.13)

where ψ(z) is the logarithmical derivative of the Γ -function. There is no gluing of left and right poles in the integral over the shifted contour so that it can be expanded safely in a Taylor series in ε. Every term of this expansion can be integrated by closing the integration contour to the right, taking residues at the points z = 0, 1, 2, . . ., and summing up the resulting series. Combining this contribution with (4.13) we obtain F (s, t; d) = −

iπ d/2 e−γE ε cj (x) εj , (−s)1+2ε t j=−2

(4.14)

where x = t/s. To calculate the ﬁrst coeﬃcients c−2 , . . . , c1 , it is enough to use MATHEMATICA for summing up the series involved. However, starting from c2 , it does not work. In this case, one can use summation formulae (C.83)– (C.94) [14]. One can also do this automatically, using the package SUMMER [39] implemented in FORM [38]. We have c−2 = 4 , c−1 = −2 ln x , c0 = −

4π 2 , 3

(4.15)

c1 = 2 (Li3 (−x) − ln x Li2 (−x)) 1 7π 2 34ζ(3) ln x − π 2 + ln2 x ln(1 + x) − , (4.16) + ln3 x + 3 6 3 c2 = 2 (S2,2 (−x) − Li4 (−x) + ln(1 + x)Li3 (−x) − ln x S1,2 (−x)) π2 (ln x − ln(1 + x))2 + ln x (ln x − 2 ln(1 + x)) Li2 (−x) − 2 2 1 1 ln x ln(1 + x) − ln2 (1 + x) − ln2 x + ln2 x 3 2 6 41π 4 2 , (4.17) + (10 ln x − 3 ln(1 + x))ζ(3) − 3 360 where, in addition to polylogarithms, we encounter generalized polylogarithms Sa,b [12, 20] (see (B.8)). One indeed needs to know expansions of one-loop Feynman integrals up to order ε2 if one wants to perform calculations in two loops because some two-loop contributions factorize and one-loop diagrams enter with coeﬃcients that have poles up to 1/ε2 . On the other hand, the functions that enter ε2 terms of expansion of one-loop Feynman integrals should be present in genuine two-loop contributions, although the ‘true’ two-loop world is, of course, much more complicated than the ε2 -expansion of the one-loop world so that, usually, two-loop results involve functions that are not present in one-loop. Any on-shell massless box with integer indices can be evaluated by a similar procedure. Generally, one encounters several right and left poles which tend to each other when ε → 0. For example, we have

4.2 Multiple MB Integrals

63

iπ d/2 Γ (−1 − 2ε)(−s)3+ε +i∞ z t 1 × dz Γ (3 + ε + z) 2πi −i∞ s

F4.3 (s, t; 2, 1, 1, 1, d) = −

×Γ (1 + z)2 Γ (−2 − ε − z)Γ (−1 − ε − z)Γ (−z) .

(4.18)

Here the ﬁrst two left poles of Γ (1 + z) glue, when ε → 0, with the ﬁrst two right poles of the product Γ (−2−ε−z)Γ (−1−ε−z). However the generalization of the above procedure to such situations is straightforward: one shifts the initial contour across the poles at z = −1 − ε and z = −2 − ε and takes two residues (with the minus sign) at these points. The procedure of evaluating any given Feynman integral from this class can easily be implemented on a computer. 2

4.2 Multiple MB Integrals Up to now we were dealing with one-parametric MB integrals. To resolve the singularities in ε we analysed the integrand, and then shifted contours and took residues, in an appropriate way. In the end of this procedure we obtained either explicit expressions for general ε or integrals where a Laurent expansion of the integrand in ε was possible. In fact, we are going to use a similar procedure for multiple MB integrals which arise when evaluating more complicated Feynman integrals. Of course, the resolution of singularities in ε in such multi-dimensional MB integrals is more complicated than in the one-dimensional case. Usually, the poles in ε are not visible at once, at a ﬁrst integration over one of the MB variables. However, the rule for ﬁnding a mechanism of the generation of poles is just a straightforward generalization of the rule used in the previous one-loop examples with one-parametric MB integrals. For example, for the massless master on-shell box, we observed that the product of Γ (1 + z) and Γ (−1 − ε − z) generated a pole of the type Γ (−ε) (this is nothing but the value of one of these gamma functions at the pole of the other gamma function). Suppose now that we are dealing with a multiple MB integral and we start from the integration over one of the variables, z. We shall analyse various products Γ (a + z)Γ (b − z), where a and b depend on the rest of the variables, with the understanding that this integration generates a pole of the type Γ (a + b). Indeed, if we shift an initial contour of integration over z across the point z = −a we obtain an integral over a new contour which is not singular at a + b = 0, while the corresponding residue involves an explicit factor Γ (a + b). (Well, sometimes it turns out that it is cancelled by a factor in the denominator.) This observation shows that any contour of one of the next integrations over the rest of the MB variables should be chosen according to this dependence, Γ (a + b). We continue this analysis, in a similar way, with various

64

4 Evaluating by MB Representation

next integrations of the second level, etc. In other words, we consider various orders of integrations over given MB variables and analyse whether a singular dependence on ε in the form of some gamma function, e.g. Γ (−ε), is generated in a given order. After this ﬁrst step, we can identify some gamma functions (in the numerator of the integrand) that are essential for the generation of poles in ε. Then we proceed with one of the MB integrations as in the case of one-dimensional MB integrals by shifting contour and taking residue. In the integral over the shifted contour, we continue this procedure by taking care of another key gamma function etc. The corresponding residue has one integration less. We deal with it exactly like with the initial integral, i.e. perform an analysis of generation of poles and then shift contours and take residues. In the end of our procedure, we are left with MB integrals which can be expanded in a Laurent series in ε under the sign of integration. We shall usually evaluate such expanded MB integrals by means of the table of one-dimensional MB integrals presented in Appendix D. All these formulae are corollaries of the ﬁrst and the second Barnes lemmas (D.1) and (D.47). Typically, the integration over the last variable is performed, as in the previous examples, by shifting the contour to the right (or left) and taking a series of residues. These series are summed up by means of summation formulae of Appendix C. There is an alternative strategy [2, 33] for the evaluation of multiple MB integrals. First, one chooses a domain of the regularization parameter ε in such a way that all the integrations over the MB variables can be performed over straight lines parallel to imaginary axis. Then one lets ε → 0, and whenever a pole of some gamma function is crossed one takes into account the corresponding residue. It is simple to organize this procedure in such a way that no more than one pole is crossed at the same time. For every resulting residue, which involves one integration less, a similar procedure is applied, and so on. We shall not, however, use this strategy. In fact, we are going to be pragmatic and not bother whether the change of the order of integration over MB variables is legitimate. Well, usually, at least at large values in the complex plane, the convergence of MB integrals is perfect3 because gamma functions have exponential decrease in both imaginary directions. This property can be used for numerical checks. Moreover, in complicated situations, one can decompose a given integrand into pieces and choose an order of integration for every piece in a special way, with the possibility to integrate explicitly, using table formulae of Appendix D.

3

However, in some situations, e.g. in a MB integral for the Gauss hypergeometric function, the asymptotic exponents of gamma functions cancel each other so that the convergence is deﬁned by the value of the argument x which is present in the MB integral as xz . Depending on whether |x| < 1 or |x| > 1, one has to close the integration contour to the right or to the left. Closing the contours to the diﬀerent sides corresponds to an analytical continuation with respect to the argument x.

4.3 More One-Loop Examples

65

We shall apply some standard properties of integration for multiple MB integrals. We shall use changes of variables of the type z → ±z + z0 . When doing this we shall, of course, trace how the nature of various poles is transformed. Note that, after such a change, z → −z, right poles become left poles. The IBP is also possible in multiple MB integrals, although it is reasonable to apply it in rare situations. Still sometimes it is useful. For example, tabulated formulae of Appendix D with the factor 1/z 2 were derived using the IBP identity f (z) f (z) . (4.19) dz 2 = dz z z C C The word ‘multiple’ will mean, in examples below, the number of MB integrations from two to eight (and even ten, in some restricted sense) which is indeed a big number. Still even in such situations, an explicit integration becomes possible, probably, because multiple MB integrals arising in the evaluation of Feynman integrals are very ﬂexible, both in the procedure of resolving the structure of singularities in ε and when evaluating ﬁnite integrals after expansion in ε. Before evaluating a Feynman integral by means of MB integrals, we shall need to derive an appropriate MB representation. Of course, we shall try to have a minimal number of MB integrations. In every case, we shall derive MB representations for general powers of the propagators. This is useful and important for several reasons. First, if we obtain a MB representation for general indices which we might imagine as complex we will certainly have unambiguous prescriptions for choosing integration contours. Second, such general formulae can be checked using various partial simple cases. Finally, starting from a general formula we can derive a lot of formulae by setting some indices to zero and thereby turning to graphs where the corresponding lines are contracted to a point. We will illustrate all these features through multiple examples below.

4.3 More One-Loop Examples We now turn to a class of one-loop Feynman integrals with two more parameters. Example 4.4. The massless box diagram of Fig. 3.5 with two legs on shell, p23 = p24 = 0, and two legs oﬀ shell, p21 , p22 = 0. We proceed like in the pure on-shell case, using alpha parameters, and obtain

66

4 Evaluating by MB Representation

Γ (a + ε − 2) F4.4 (s, t, p21 , p22 ; a1 , . . . , a4 , d) = iπ d/2 (−1)a Γ (al )

4 ∞ ∞ 4 al −1 ... αl dαl δ αl − 1 × 0

0

l=1

×(−sα1 α3 − tα2 α4 −

p21 α1 α2

−

l=1 2 p2 α2 α3

− i0)2−a−ε .

(4.20)

We have chosen the delta function of the sum of all the α-variables so that the factor with a power of the function U is equal to one. Now we need a generalization of (4.1) to the case of several terms which is easily obtained by induction: +i∞ +i∞ n 1 1 1 = . . . dz . . . dz Xizi 2 n (X1 + . . . + Xn )λ Γ (λ) (2πi)n−1 −i∞ −i∞ i=2 ×X1−λ−z2 −...−zn Γ (λ + z2 + . . . + zn )

n

Γ (−zi ) .

(4.21)

i=2

We use (4.21) to replace the last factor in (4.20) by a product of four factors thus separating terms with t, p21 and p22 from s. After that we introduce new variables by α1 = η1 ξ1 , α2 = η1 (1 − ξ1 ), α3 = η2 ξ2 , α4 = η2 (1 − ξ2 ) and arrive at a product of three parametric integrals evaluated in terms of gamma functions. Eventually we obtain the following threefold MB representation of a general Feynman integral of the given class: F4.4 (s, t, p21 , p22 ; a1 , . . . , a4 , d) =

iπ d/2 (−1)a Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞ +i∞ +i∞ (−p21 )z2 (−p22 )z3 (−t)z4 1 dz2 dz3 dz4 3 (2πi) −i∞ −i∞ −i∞ (−s)z2 +z3 +z4 ×Γ (a + ε − 2 + z2 + z3 + z4 )Γ (a2 + z2 + z3 + z4 )Γ (a4 + z4 ) ×

×Γ (2 − ε − a234 − z3 − z4 )Γ (2 − ε − a124 − z2 − z4 ) ×Γ (−z2 )Γ (−z3 )Γ (−z4 ) .

(4.22)

In this chapter, we continue to use our notation: a124 = a1 + a2 + a4 , etc. with a = a1234 . This representation can be, of course, used for evaluating these Feynman integrals. We shall use it, however, in the next section only as an auxiliary result when deriving an MB representation for the massless on-shell double box diagrams. One of the advantages of general formulae is that they provide a lot of partial cases. For example (4.22) immediately gives a twofold MB representation for Example 4.5. The massless box diagram of Fig. 3.5 with three legs on shell, p22 = p23 = p24 = 0, and one leg oﬀ shell, p21 = 0. Indeed we put p22 to zero in the ‘naive’ sense, i.e. in the integrand of the corresponding Feynman integral or in some parametric representation. This

4.3 More One-Loop Examples

67

is equivalent to setting p22 to zero in the sense of the leading term of the hard part of the asymptotic expansion in the limit p22 → 0 (see details in [27]), which corresponds to taking residues (with the minus sign) of the poles of Γ (−z3 ). So we just take minus residue of the integrand at z3 = 0. Thus we obtain F4.5 (s, t, p21 ; a1 , . . . , a4 , d) =

iπ d/2 (−1)a Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞ +i∞ (−p21 )z2 (−t)z4 1 dz2 dz4 Γ (a + ε − 2 + z2 + z4 ) 2 (2πi) −i∞ −i∞ (−s)z2 +z4 ×Γ (a2 + z2 + z4 )Γ (a4 + z4 )Γ (2 − ε − a234 − z4 ) ×Γ (2 − ε − a124 − z2 − z4 )Γ (−z2 )Γ (−z4 ) . (4.23)

×

Let us now turn to massive diagrams. Example 4.6. The on-shell box with two massive and two massless lines shown in Fig. 4.4, with p21 = . . . = p24 = m2 .

p1 2

p2

p3

1 4 3

p4

Fig. 4.4. On-shell box with two massive and two massless lines. The solid lines denote massive, the dotted lines massless particles

The derivation of the corresponding MB representation is quite straightforward. The combination that is involved in the corresponding integral over alpha or Feynman parameters has now an additional piece as compared with the massless case: V −U m2l αl = sα1 α3 + tα2 α4 − m2 (α1 + α3 )2 . This term can be separated from the rest terms at the cost of introducing one more MB integration according to (4.21). This time, let us introduce new parametric variables in a slightly diﬀerent way, α1 = η1 ξ1 , α2 = η2 ξ2 , α3 = η1 (1 − ξ1 ), α4 = η2 (1 − ξ2 ), in order to make (α1 + α3 )2 simpler. Evaluating the parametric integrals we arrive at the following massive generalization of (4.11): F4.6 (s, t, m2 ; a1 , a2 , a3 , a4 , d) = 1 × (2πi)2

+i∞

−i∞

(−1)a iπ d/2 Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞

−i∞

dz1 dz2

(−t)z1 (m2 )z2 Γ (a + ε − 2 + z1 + z2 ) (−s)z1 +z2

68

4 Evaluating by MB Representation

×Γ (a2 + z1 )Γ (a4 + z1 )Γ (−z1 )Γ (−z2 )Γ (2 − a124 − ε − z1 − z2 ) Γ (4 − a122344 − 2ε − 2z1 ) , (4.24) ×Γ (2 − a234 − ε − z1 − z2 ) Γ (4 − a122344 − 2ε − 2z1 − 2z2 ) where a122344 = a1 + 2a2 + a3 + 2a4 , etc. Observe that the onefold representation (4.11) in the massless case follows from (4.24) when we put m to zero. As it was discussed above we do this by taking the limit m → 0 in the sense of the leading term of the hard part of the expansion. Here this means that we just take minus residue at z2 = 0 with respect to the variable z2 which enters the integrand as the exponent of m2 . In particular, we have F4.6 (s, t, m2 ; 1, 1, 1, 1, d) = ×

1 (2πi)2

+i∞

−i∞

+i∞

−i∞

dz1 dz2

(−1)a iπ d/2 Γ (−2ε)(−s)2+ε (−t)z1 (m2 )z2 Γ (2 + ε + z1 + z2 )Γ (−z1 ) (−s)z1 +z2

×Γ (−z2 )Γ (−1 − ε − z1 − z2 )2

Γ (1 + z1 )2 Γ (−2 − 2ε − 2z1 ) . Γ (−2 − 2ε − 2z1 − 2z2 )

(4.25)

The resolution of singularities in ε can be performed here as in the onedimensional case because only the product Γ (1 + z1 )2 Γ (−2 − 2ε − 2z1 ) is responsible for the generation of poles. To see this, we use properties of the gamma function and write Γ (−2−2ε−2z1 ) as Γ (−1−ε−z1 )Γ (−1/2−ε−z1 ) up to a factor so that we obtain the product Γ (1 + z1 )2 Γ (−1 − ε − z1 ) which involves gluing of the left pole at z1 = −1 and the right pole at z1 = −1 − ε when ε → 0. We proceed as in Sect. 4.1 by taking minus residue at the point z1 = −1 − ε and shifting the integration contour over z1 across this point. The residue gives 2 z2 +i∞ m Γ (1 + z2 )Γ (−z2 )3 Γ (1 + ε)Γ (−ε)2 1 . (4.26) − dz 2 1+ε 2s(−t) Γ (−2ε) (2πi) −i∞ −s Γ (−2z2 ) This integral can be evaluated by closing the contour to the left and taking residues at the points z2 = −1, −2, . . . with summing up this inverse binomial series by the summation formulae of Sect. C.3. As to the integral over the shifted contour, it does not have poles in ε. If we need to expand (4.25) only up to ε0 this integral does not contribute because of the overall Γ (−2ε) in the denominator, so that we are left with the contribution of the residue: F4.6 (s, t, m2 ; 1, 1, 1, 1, d)

−t 1 2iπ d/2 e−γE ε 1−x − ln + O(ε) , (4.27) =− ln 2 2 ε 2 m 1+x (m ) t −s(4m − s) ε where x = 1/ 1 − 4m2 /s, in agreement with [3]. The general MB representation (4.24) can be used to derive an MB representation for the triangle diagram shown in Fig. 4.5. This class of Feynman integrals is obtained from the corresponding box integrals if we set a4 = 0.

4.3 More One-Loop Examples

69

p1 1 3 2 p2 Fig. 4.5. Triangle diagram with the masses m, m, 0 and external momenta on-shell, p21 = p22 = m2 . A dotted line denotes a massless propagator

If we do this blindly in (4.24) we obtain a zero result due to Γ (a4 ) in the denominator. This is, of course, wrong. Let us think of a4 as a complex number and analyse the behaviour in the limit a4 → 0 similarly to what we do when analysing how singularities in ε are generated. We identify the product Γ (a4 + z1 )Γ (−z1 ) responsible for the generation of the singularity when a4 → 0. To reveal this singularity we can take minus residue at the point z1 = 0 and shift the integration contour over z1 . The contribution of the new integral is indeed zero because of the factor 1/Γ (a4 ). The contribution of the residue produces Γ (a4 ) which cancels this factor in the denominator, and we put a4 to zero after that. Changing the numbering 2 ↔ 3, for convenience, we obtain the following onefold MB representation4 for integrals corresponding to Fig. 4.5: (−1)a iπ d/2 Γ (4 − 2ε − a1 − a2 − 2a3 ) Γ (4 − 2ε − a1 − a2 − a3 )Γ (a1 )Γ (a2 )(−s)a+ε−2 +i∞ 2 z m 1 × dz Γ (a + ε − 2 + z)Γ (−z) 2πi −i∞ −s Γ (2 − a1 − a3 − ε − z)Γ (2 − a2 − a3 − ε − z) × . (4.28) Γ (4 − 2ε − a1 − a2 − 2a3 − 2z) Observe that if we want to have a representation for massive propagatortype diagrams by setting a3 = 0 we shall not reduce the number of integrations: there is no Γ (a3 ) in the denominator and, on the other hand, no singularities in the limit a3 → 0 are generated. So, one can simply apply (4.28) with a3 = 0 for this class of diagrams. The general MB representation (4.24) provides in a very similar way a MB representation for another triangle diagram obtained from Fig. 4.4. We shrink the line 3 to a point and obtain Fig. 4.6. The corresponding onefold MB representation takes the form

4

In [11], it was demonstrated that this Feynman integral reduces, for any values of the three indices, to a two-point function in the shifted dimension d − 2a3 .

70

4 Evaluating by MB Representation

p1 2 1 4 p3 Fig. 4.6. Triangle diagram with the masses m, 0, 0 and external momenta on-shell, p21 = p23 = m2 , obtained from the box of Fig. 4.4

(−1)a iπ d/2 Γ (4 − 2ε − a)Γ (a1 )Γ (a2 )Γ (a4 )(m2 )a+ε−2 z +i∞ −t 1 × dz Γ (a + ε − 2 + z)Γ (−z) 2πi −i∞ m2 ×Γ (a2 + z)Γ (a4 + z)Γ (4 − 2ε − a1 − 2a2 − 2a4 − 2z) ,

(4.29)

where t = (p1 + p3 )2 . Among other partial cases of the massive on-shell boxes let us mention the case where a1 = a2 = 0. Then we obtain a massless one-loop propagatortype diagram which is evaluated by (3.6). On the other hand, one can see that to perform the limit a1 , a2 → 0 it is necessary to take two residues in the integrand and somehow compensate the corresponding gamma functions in the denominator. Eventually one arrives at the known result. This procedure is just an additional check for the initial MB representation (4.24). The representation (4.24) can straightforwardly be generalized to various oﬀ-shell cases, similarly to how we obtained the generalizations (4.22) and (4.23). Here are three results which we shall use in Sect. 4.4. For the box of Fig. 4.4 with two massive and two massless lines, two legs on shell, p23 = p24 = m2 , and two legs oﬀ shell we obtain the following fourfold MB representation: +i∞ +i∞ 4 1 (−1)a iπ d/2 (−s)2−a−ε ... dzj Γ (−zj ) Γ (4 − 2ε − a) Γ (al ) (2πi)4 −i∞ −i∞ j=1 (m2 − p21 )z1 (m2 − p22 )z2 (−t)z3 (m2 )z4 Γ (a2 + z1 + z2 + z3 )Γ (a4 + z3 ) (−s)z1 +z2 +z3 +z4 ×Γ (2 − a124 − ε − z1 − z3 − z4 )Γ (2 − a234 − ε − z2 − z3 − z4 ) Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 ) × Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 − 2z4 ) ×Γ (a + ε − 2 + z1 + z2 + z3 + z4 ) . (4.30) ×

For the box of Fig. 4.4 with two legs on shell, p22 = p24 = m2 , and two legs oﬀ shell, we obtain:

4.4 Two-Loop Massless Examples

1 (−1)a iπ d/2 (−s)2−a−ε Γ (4 − 2ε − a) Γ (al ) (2πi)4 ×

(m − 2

+i∞

+i∞

... −i∞

p21 )z1 (m2 − p23 )z2 (−t)z3 (m2 )z4 (−s)z1 +z2 +z3 +z4

−i∞

4

71

dzj Γ (−zj )

j=1

Γ (a2 + z1 + z3 )Γ (a4 + z2 + z3 )

×Γ (2 − a124 − ε − z1 − z2 − z3 − z4 )Γ (2 − a234 − ε − z3 − z4 ) Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 ) × Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 − 2z4 ) ×Γ (a + ε − 2 + z1 + z2 + z3 + z4 ) . p21

p24

(4.31) 2

Finally, for the box of Fig. 4.4 with two legs on shell, = = m , and two legs oﬀ shell, we obtain: +i∞ +i∞ 4 1 (−1)a iπ d/2 (−s)2−a−ε ... dzj Γ (−zj ) Γ (4 − 2ε − a) Γ (al ) (2πi)4 −i∞ −i∞ j=1 (m2 − p23 )z1 (m2 − p22 )z2 (−t)z3 (m2 )z4 Γ (a2 + z2 + z3 )Γ (a4 + z1 + z3 ) (−s)z1 +z2 +z3 +z4 ×Γ (2 − a124 − ε − z1 − z3 − z4 )Γ (2 − a234 − ε − z2 − z3 − z4 ) Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 ) × Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 − 2z4 ) ×Γ (a + ε − 2 + z1 + z2 + z3 + z4 ) . (4.32) ×

4.4 Two-Loop Massless Examples Our ﬁrst two-loop example is the same as Example 3.7: Example 4.7. Non-planar two-loop massless vertex diagram of Fig. 3.13 with p21 = p22 = 0. We are again dealing with two-loop vertex Feynman integrals (3.53). We start with the four-parametric representation (3.57) obtained within the method of Feynman parameters in the previous chapter. Let us turn to the variables ξ3 = ξη, ξ4 = (1 − ξ)η and apply (4.1) to the resulting denominator in the integrand: Γ (a + 2ε − 4) a+2ε−4

[ηξ(1 − ξ) + (1 − η)(ξξ2 (1 − ξ1 ) + (1 − ξ)ξ1 (1 − ξ2 ))] +i∞ dz1 Γ (−z1 )η z1 ξ z1 (1 − ξ)z1 1 = 2πi −i∞ (1 − η)a+2ε−4+z1 Γ (a + 2ε − 4 + z1 ) × a+2ε−4+z1 . [ξξ2 (1 − ξ1 ) + (1 − ξ)ξ1 (1 − ξ2 )] Then we again apply (4.1) to transform the last line of (4.33) into

(4.33)

72

4 Evaluating by MB Representation

1 2πi

+i∞

−i∞

dz2 Γ (a + 2ε − 4 + z1 + z2 )Γ (−z2 )ξ z2 ξ2z2 (1 − ξ1 )z2 . (1 − ξ)a+2ε−4+z1 +z2 ξ1a+2ε−4+z1 +z2 (1 − ξ2 )a+2ε−4+z1 +z2

After that all the integrals over the parameters ξ1 , ξ2 , ξ, η can be evaluated in terms of gamma functions, and we come to the following twofold MB representation of (3.53) with general powers of the propagators: 2 (−1)a iπ d/2 Γ (2 − ε − a35 ) 2 F4.7 (Q ; a1 , . . . , a6 , d) = (Q2 )a+2ε−4 Γ (6 − 3ε − a) Γ (al ) +i∞ +i∞ Γ (2 − ε − a46 ) 1 × dz1 dz2 Γ (a + 2ε − 4 + z1 + z2 ) Γ (4 − 2ε − a3456 ) (2πi)2 −i∞ −i∞ ×Γ (−z1 )Γ (−z2 )Γ (a4 + z2 )Γ (a5 + z2 )Γ (a1 + z1 + z2 ) Γ (2 − ε − a12 − z1 )Γ (4 − 2ε + a2 − a − z2 ) × Γ (4 − 2ε − a1235 − z1 )Γ (4 − 2ε − a1246 − z1 ) ×Γ (4 − 2ε + a3 − a − z1 − z2 )Γ (4 − 2ε + a6 − a − z1 − z2 ) . (4.34) As in Chap. 3 let us evaluate the integral with all indices equal to one. We have d/2 2 iπ 2 F (ε) , (4.35) F4.7 (Q ; 1, . . . , 1, d) = (Q2 )2+2ε with F (ε) =

Γ (−ε)2 V (ε) Γ (−3ε)Γ (−2ε)

and V (ε) =

1 (2πi)2

+i∞

−i∞

+i∞

−i∞

dz1 dz2 Γ (2 + 2ε + z1 + z2 )Γ (1 + z1 + z2 )

× Γ (1 + z2 )2 Γ (−z1 )Γ (−z2 )

Γ (−ε − z1 ) Γ (−2ε − z1 )2

× Γ (−1 − 2ε − z2 )Γ (−1 − 2ε − z1 − z2 )2 .

(4.36)

After the useful change of variables z1 → −1 − z1 − z2 , we obtain +i∞ +i∞ Γ (1 + z1 + z2 )Γ (1 − ε + z1 + z2 ) 1 V (ε) = dz1 dz2 2 (2πi) −i∞ −i∞ Γ (1 − 2ε + z1 + z2 )2 × Γ (−2ε + z1 )2 Γ (−z1 )Γ (1 + 2ε − z1 ) × Γ (1 + z2 )2 Γ (−1 − 2ε − z2 )Γ (−z2 ) .

(4.37)

The analysis of the integrand shows that the poles in ε are generated by the two products Γ (−2ε + z1 )2 Γ (−z1 ) and Γ (1 + z2 )2 Γ (−1 − 2ε − z2 ) so that the situation is somehow factorized and we can proceed like in the onedimensional cases taking care of the integrations over z1 and z2 separately. So, let us ﬁrst deal with the ﬁrst pole of Γ (−1 − 2ε − z2 ). We have minus the residue at z2 = −1 − 2ε,

4.4 Two-Loop Massless Examples

F1 (ε) =

Γ (1 + 2ε)Γ (−2ε)Γ (−ε)2 1 Γ (−3ε) 2πi

×

73

+i∞

−i∞

dz1 Γ (1 + 2ε − z1 )

Γ (−2ε + z1 )3 Γ (−3ε + z1 )Γ (−z1 ) , Γ (−4ε + z1 )2

(4.38)

and the integral F0 (ε) with the opposite nature of the ﬁrst pole at z2 = −1 − 2ε. For (4.38), we analyse how singularities in ε are generated. The situation is quite familiar and we come to the conclusion that they come from the product Γ (−2ε + z1 )3 Γ (−3ε + z1 )Γ (−z1 ). We take residues at the points z1 = 2ε and z1 = 3ε and turn to the integral F10 with the same integrand as (4.38) but with the opposite nature of these poles. The sum of these two residues gives, in expansion in ε, 1 π2 211ζ(3) π 4 −2γE ε + − 2 − F11 = e + O(ε) . (4.39) ε4 ε 6ε 80 The integral F10 can be evaluated by expanding the integrand in ε and subsequently closing the contour to the right and summing up a series of residues. Here one can apply summation formulae of Appendix C for summing up this number series. The result is 2 π 3ζ(3) 41π 4 − + F10 = e−2γE ε + O(ε) . (4.40) 4ε2 ε 48 Now we have to calculate (4.37) with the opposite nature of the ﬁrst pole of Γ (−1 − 2ε − z2 ). Let us take care of the ﬁrst pole of Γ (−2ε + z1 )2 . We take the residue at this point which is an integral F01 over z2 without gluing of poles of diﬀerent nature and thereby can be evaluated directly in expansion in ε. The resulting expanded integral is evaluated similarly to F10 . We obtain 9ζ(3) 31π 4 π2 −2γE ε + F01 = e − 2+ + O(ε) . (4.41) 4ε 2ε 60 The remaining piece is the integral F00 with the integrand of (4.37) where the ﬁrst poles of Γ (−2ε+z1 )2 and Γ (−1−2ε−z2 ) have changed their nature. There is no gluing anymore so that we can expand the integrand in ε: +i∞ +i∞ 6 dz1 dz2 Γ (z1 )2 Γ (−z1 )Γ (1 − z1 ) F00 = (2πi)2 −i∞ −i∞ × Γ (1 + z2 )2 Γ (−1 − z2 )Γ (−z2 ) + O(ε) ,

(4.42)

where the integration contours are at −1 < Rez1,2 < 0. The integral is a product of one-dimensional MB integrals which can be evaluated by the same procedure as above. We obtain π2 + O(ε) . (4.43) 6 Summing up the four pieces (4.39), (4.40), (4.41) and (4.43) we reproduce the result (3.74) obtained in [15]. F00 = −

74

4 Evaluating by MB Representation

p1

1 2

p2

p3

6 7

3

5 4

p4

Fig. 4.7. Double box

Let us now consider Example 4.8. Massless on-shell planar double box diagram of Fig. 4.7. As in Example 4.3. we have p2i = 0, i = 1, 2, 3, 4. Let us consider double boxes with the irreducible numerator (k +p1 +p2 +p4 )2 and the routing of the external momenta as in [2]. Then the general double box Feynman integral takes the form dd k dd l K(s, t; a1 , . . . , a8 , ε) = 2 a 1 (k ) [(k + p1 )2 ]a2 [(k + p1 + p2 )2 ]a3 [(k + p1 + p2 + p4 )2 ]−a8 × , (4.44) 2 a [(l + p1 + p2 ) ] 4 [(l + p1 + p2 + p4 )2 ]a5 (l2 )a6 [(k − l)2 ]a7 As usual, we consider the factor corresponding to the irreducible numerator as an extra propagator but, really, we are interested only in non-positive integer values of a8 . In fact, there are two possible independent irreducible numerators but the derivation of the MB representation is simple only when we take one of them into account. In order to derive a MB representation for (4.44) it is possible to start from the alpha representation and then apply (4.1) to the corresponding functions U and V. This is not, however, an optimal way. In particular, this was done in the ﬁrst calculation of the master double box [23] but a resulting MB representation turned out to be ﬁvefold, with essential complications in the calculations. We will see that one can proceed using a fourfold MB representation. Let us mention, however, that in the case of non-planar onshell double boxes it was possible to achieve [33] the minimal number of integrations equal to four starting from the global alpha representation. So, we follow (as in [2]) the strategy of [35], where MB integrations were, ﬁrst, introduced, in a suitable way, after the integration over one of the loop momenta, l, and complete this procedure after the integration over the second loop momentum, k. To do this, let us observe that (4.44) can be represented as dd k [(k + p1 + p2 + p4 )2 ]−a8 K(s, t; a1 , . . . , a8 , ε) = (k 2 )a1 [(k + p1 )2 ]a2 [(k + p1 + p2 )2 ]a3 ×F4.4 (s, (k + p1 + p2 + p4 )2 , k 2 , (k + p1 + p2 )2 ; a6 , a7 , a4 , a5 , d) , (4.45) where the integral of four propagators dependent on l has been recognized as the box with two legs oﬀ shell. Then we can use (4.22). After inserting it into

4.4 Two-Loop Massless Examples

75

(4.45) we obtain the massless on-shell box with the indices a1 −z2 , a2 , a3 , a8 − z4 for which we apply our representation (4.11). After these straightforward manipulations, we change the variables z2 → z2 −z4 , z3 → z3 −z4 , z4 → z1 +z4 , and arrive at the following fourfold MB representation of (4.44) (see also [2]): d/2 2 iπ (−1)a K(s, t; a1 , . . . , a8 , ε) = a−4+2ε l=2,4,5,6,7 Γ (al )Γ (4 − a4567 − 2ε)(−s) z1 +i∞ +i∞ 4 t 1 × ... dzj Γ (a2 + z1 )Γ (−z1 ) 4 (2πi) −i∞ s −i∞ j=1 Γ (z2 + z4 )Γ (z3 + z4 )Γ (a1238 − 2 + ε + z4 )Γ (a7 + z1 − z4 ) Γ (a1 + z3 + z4 )Γ (a3 + z2 + z4 )Γ (4 − a1238 − 2ε + z1 − z4 ) Γ (a8 − z2 − z3 − z4 )Γ (a5 + z1 + z2 + z3 + z4 )Γ (−z1 − z2 − z3 − z4 ) × Γ (a8 − z1 − z2 − z3 − z4 ) ×Γ (a4567 − 2 + ε + z1 − z4 )Γ (2 − a128 − ε + z2 )Γ (2 − a238 − ε + z3 ) ×Γ (2 − a567 − ε − z1 − z2 )Γ (2 − a457 − ε − z1 − z3 ) . (4.46) ×

Let us apply (4.46) to the evaluation, in expansion in ε up to the ﬁnite part, of the double box without numerator and with all powers of the propagators equal to one. We know in advance that it has poles up to the fourth order in ε, due to IR and collinear divergences. In fact, at least the highest pole can be predicted without calculation. Representation (4.46) gives d/2 2 iπ F (x, ε) , (4.47) K(s, t; 1, . . . , 1, 0, ε) = − (−s)3+2ε where x = t/s and F (x, ε) =

1 1 Γ (−2ε) (2πi)4

+i∞

+i∞

... −i∞

−i∞

4

dzj xz1

j=1

Γ (1 + z1 )Γ (−z1 )Γ (−1 − ε − z1 − z2 )Γ (−1 − ε − z1 − z3 ) Γ (1 + z2 + z4 )Γ (1 + z3 + z4 )Γ (1 − 2ε + z1 − z4 ) ×Γ (2 + ε + z1 − z4 )Γ (1 + z1 + z2 + z3 + z4 )Γ (1 + z1 − z4 )

×

×Γ (z2 + z4 )Γ (z3 + z4 )Γ (−ε + z2 )Γ (−ε + z3 ) ×Γ (1 + ε + z4 )Γ (−z2 − z3 − z4 ) .

(4.48)

Observe that, because of the presence of the factor Γ (−2ε) in the denominator, we are forced to take some residue in order to arrive at a non-zero result at ε = 0, so that the integral is eﬀectively threefold. Here is an example of the procedure of generating poles in the integral (4.48). The product Γ (−1 − ε − z1 − z2 )Γ (−ε + z2 ) generates, due to the integration over z2 , a pole of the type Γ (−1 − 2ε − z1 ). Then the product of this gamma function with Γ (1 + z1 ) generates a pole of the type Γ (2ε) due to the integration over z1 .

76

4 Evaluating by MB Representation

After such a preliminary analysis we conclude that the key gamma functions that are responsible for the generation of poles in ε are Γ (−ε + z2 ), Γ (−ε + z3 ) and Γ (1 + z1 − z4 ). This gives a hint for the construction of a complete procedure of the resolution of the singularities in ε, with the goal to decompose the given integral into pieces where the Laurent expansion of the integrand in ε becomes possible. One can proceed as follows. We ﬁrst take care of the gamma functions Γ (−ε + z2 ) and Γ (−ε + z3 ), i.e. take residues at z2 = ε and z3 = ε and shift contours across these poles. As a result, (4.48) is decomposed as F = F11 + F10 + F01 + F00 , where F11 corresponds to taking the two residues, F00 is deﬁned by the same expression (4.48) but with both ﬁrst poles of the selected two gamma functions treated in the opposite way, and the two intermediate contributions deﬁned by taking one of the residues and changing the nature of the ﬁrst pole of the other gamma function. The contribution F11 takes the form +i∞ +i∞ 1 1 F11 = dz1 dz4 xz1 Γ (1 + z1 ) Γ (−2ε) (2πi)2 −i∞ −i∞ ×Γ (−1 − 2ε − z1 )2 Γ (−z1 )Γ (1 + z1 − z4 )Γ (2 + ε + z1 − z4 ) Γ (1 + 2ε + z1 + z4 ) . (4.49) ×Γ (ε + z4 )2 Γ (−2ε − z4 ) Γ (1 − 2ε + z1 − z4 )Γ (1 + ε + z4 ) The contributions F10 and F01 are equal to each other because of the symmetrical dependence of the integrand on z2 and z3 . We have +i∞ +i∞ +i∞ 1 1 dz1 dz2 dz4 xz1 Γ (1 + z1 ) F01 = Γ (−2ε) (2πi)3 −i∞ −i∞ −i∞ ×Γ (−1 − 2ε − z1 )Γ (−z1 )Γ (−1 − ε − z1 − z2 )Γ ∗ (−ε + z2 ) Γ (1 + z1 − z4 )Γ (2 + ε + z1 − z4 )Γ (ε + z4 )Γ (z2 + z4 ) × Γ (1 − 2ε + z1 − z4 )Γ (1 + z2 + z4 ) ×Γ (1 + ε + z1 + z2 + z4 )Γ (−ε − z2 − z4 ) , (4.50) where the ﬁrst pole of Γ (−ε + z2 ) is of the opposite nature. We indicate this by asterisk, as in Appendix D. For all these contributions, further decompositions are necessary. One can proceed as follows. In the case of F11 , take care of Γ (−1 − 2ε − z1 ). We decompose F11 as F111 + F110 , where the additional index 1 corresponds to the residue at z1 = −1 − 2ε (with the minus sign) and 0 to the integral where the ﬁrst pole of Γ (−1 − 2ε − z1 ) is left. Take care of Γ (z4 ) and Γ (z4 + ε) by decomposing F111 as F111 = F1111 + F1110 , where the additional index 1 corresponds to the residues at z4 = 0 and z4 = ε given by an explicit expression in terms of gamma and psi functions, and 0 to the one-dimensional MB integral where the ﬁrst pole of each of these gamma functions is right. For F110 , take care of Γ (z4 + ε) to obtain F110 = F1101 + F1100 , where 1 denotes the residue at z4 = −ε. The F1101 is a one-dimensional MB integral

4.4 Two-Loop Massless Examples

77

over z1 which is calculated by expanding in ε. The F1100 starts from ε1 and therefore gives a zero contribution. For F01 , take care of Γ (−1 − 2ε − z1 ) and obtain the decomposition F01 as F011 + F010 similar to the case of F11 . For F011 , let us consecutively take care of the ﬁrst poles of the gamma functions Γ (z2 + z4 ) and Γ (z2 + z4 − ε) with respect to the variable z2 and obtain F011 = F0111 + F0112 + F0110 , where 1 denotes the residue at z2 = −z4 , 2 denotes the residue at z2 = ε − z4 and 0 denotes the integral with ﬁrst poles of these gamma functions to be right. Then we obtain F0111 = F01111 + F01110 , similarly taking care of Γ (ε + z4 )2 , F0112 = F01121 + F01120 taking care of Γ (ε + z4 )Γ (z4 ), and F0110 = F01101 + F01100 taking care of Γ (ε + z4 ). For F010 , we turn to the decomposition F010 = F0101 +F0100 where 1 stands for the residue at z4 = −z2 and 0 for the integral with the ﬁrst right pole of Γ (z2 + z4 ). Finally, we turn to F0101 = F01011 + F01010 , where 1 stands for the residue at z2 = −1 − ε − z1 and 0 for the integral with the ﬁrst left pole of Γ (−1 − ε − z1 − z2 ). For F00 , we take care of the ﬁrst poles of the gamma functions Γ (−1 − ε − z1 − z2 ) and Γ (−1 − ε − z1 − z3 ). The only non-zero contribution arises when taking both residues. As a result we obtain either explicit expressions in terms of gamma functions and their derivatives, or one-dimensional integrals over straight lines parallel to the imaginary axis of ratios of gamma functions which can be of two types: integrals over z1 or some other z-variable. The integrals over z1 can be calculated by closing the contour to the right, taking residues at the points z1 = 0, 1, 2, . . . and summing up resulting series with the help of the table of formulae [14] presented in Appendix C. The one-dimensional MB integrals over z2 or z3 or z4 can be calculated with the help of formulae of Appendix D which are all corollaries of the ﬁrst and the second Barnes lemma (D.1) and (D.47). For example, this is the twofold MB integral that appears in F01100 : +i∞ +i∞ 1 dz2 dz4 Γ ∗ (z2 )Γ (−z2 )Γ (1 + z4 )Γ (−z4 ) (2πi)2 −i∞ −i∞ ×

Γ ∗ (z2 + z4 )2 Γ (−z2 − z4 ) , Γ (1 + z2 + z4 )

(4.51)

where asterisks denote, as in Appendix D, the opposite nature of the ﬁrst poles of the corresponding gamma functions, i.e. the poles z2 = 0 and z4 = −z2 are considered right here. The internal integral over z4 is then evaluated with the help of (D.51), with λ1 = 1, λ2 = z2 , λ3 = 0, λ4 = 1 + z2 , and a resulting onefold MB integral is evaluated as other integrals of this kind. Collecting all the contributions we reproduce the result of [23]: d/2 −γ ε 2 iπ e E t f ;ε , (4.52) K(s, t; 1, . . . , 1, 0, ε) = − (−s)2+2ε t s where

78

4 Evaluating by MB Representation

4 5 ln x 5 2 1 2 π + − 2 ln x − ε4 ε3 2 ε2 1 2 3 11 65 ln x + π 2 ln x − ζ(3) − 3 2 3 ε 4 4 29 88 + ln x + 6π 2 ln2 x − ζ(3) ln x + π 4 3 3 30 2

2 − 2 Li3 (−x) − 2 ln x Li2 (−x) − ln x + π 2 ln(1 + x) ε −4 [S2,2 (−x) − ln x S1,2 (−x)] + 44 Li4 (−x)

f (x, ε) = −

−4 [ln(1 + x) + 6 ln x] Li3 (−x) 10 2 2 +2 ln x + 2 ln x ln(1 + x) + π Li2 (−x) 3 2 2 2 + ln x + π ln (1 + x) 2

− 4 ln3 x + 5π 2 ln x − 6ζ(3) ln(1 + x) + O(ε) . (4.53) 3 This result is in agreement with the leading behaviour in the (Regge) limit t/s → 0 obtained in [32] by use of the strategy of expansion by regions [4, 27, 30]. Keeping the two leading powers of x we have 1 5 ln x 5 4 f (x, ε) = − 4 + 3 − 2 ln2 x − π 2 ε ε 2 ε2 1 2 3 11 65 ln x + π 2 ln x − ζ(3) − 3 2 3 ε 29 4 4 88 + ln x + 6π 2 ln2 x − ζ(3) ln x + π 4 3 3 30 1 2 ln x − 2 ln x + π 2 + 2 +2x ε 1 − 4 ln3 x + 3 ln2 x + (5π 2 − 36) ln x + 2[33 + 5π 2 − 3ζ(3)] 3 +O(x2 ln3 x, ε) .

(4.54)

Using known formulae that relate polylogarithms and generalized polylogarithms with arguments z and 1/z [12, 20, 21] one can rewrite this and similar results for the master double boxes in terms of the same class of functions depending on the inverse ratio s/t. Let us now illustrate the point discussed in the end of Sect. 4.2. The general fourfold representation (4.46) contains a lot of information. In particular, it is very easy to derive MB representations for the two classes of Feynman integrals corresponding to the graphs shown in Fig. 4.8. The integrals for the box with a one-loop insertion are obtained from the double box integrals at a4 = a6 = 0. (For simplicity, we consider the case a8 = 0.) There are Γ (a4 ) and Γ (a6 ) in the denominator of (4.46) but, of course, the limit a4 , a6 → 0 is not zero. Indeed, we can distinguish the product

4.4 Two-Loop Massless Examples 1

p1 2

p2

6

p3 7

3

(a)

5

2

p4

79

7

5

3

(b)

Fig. 4.8. Boxes with a one-loop insertion (a) and boxes with a diagonal (b) obtained from Fig. 4.7

Γ (a4567 − 2 + ε + z1 − z4 )Γ (2 − a567 − ε − z1 − z2 )Γ (z2 + z4 ) which generates, due to integration over z2 and z4 , the singularity of the type Γ (a4 ) – remember our discussion in Sect. 4.2. So, to perform this limit we take a residue at z4 = −z2 and minus residue at z2 = 2 − a567 − ε − z1 and then set a4 = 0. We still have Γ (a6 ) in the denominator, but there is also the product Γ (a567 − 2 + ε + z1 + z3 )Γ (2 − a57 − ε − z1 − z3 ) which generates the singularity of the type Γ (a6 ). Therefore, we take minus residue at z3 = 2 − a57 − ε − z1 , then set a6 = 0 and arrive at the following onefold MB representation: d/2 2 iπ (−1)a Γ (2 − a5 − ε)Γ (2 − a7 − ε) K(a1 , a2 , a3 , 0, a5 , 0, a7 , 0) = Γ (al )Γ (4 − a57 − 2ε)Γ (6 − a − 3ε) z +i∞ t 1 1 dz Γ (a − 4 + 2ε + z)Γ (a57 − 2 + ε + z) × (−s)a−4+2ε 2πi −i∞ s ×Γ (a2 + z)Γ (4 − a1257 − 2ε − z)Γ (4 − a2357 − 2ε − z)Γ (−z) . (4.55) The integrals for the box with a diagonal are obtained from the double box integrals at a1 = a4 = 0. We start from the limit a4 → 0 as in the previous case. Then we observe that there is no Γ (a1 ) in the denominator and no gluing of right and left poles when a1 → 0. So, we just set a1 = 0. After that the integration over z3 involves only four gamma functions Γ (2 − a23 − ε + z3 )Γ (a5 + z1 + z3 )Γ (2 − a57 − ε − z1 − z3 )Γ (−z3 ) . The integral is evaluated by the ﬁrst Barnes lemma (D.1), and we obtain d/2 2 iπ Γ (2 − a23 − ε)Γ (2 − a56 − ε) K(0, a2 , a3 , 0, a5 , a6 , a7 , 0) = Γ (al )Γ (4 − a237 − 2ε)Γ (4 − a567 − 2ε) z +i∞ a t 1 (−1) Γ (2 − a7 − ε) dz Γ (a − 4 + 2ε + z) × Γ (6 − a − 3ε)(−s)a−4+2ε 2πi −i∞ s ×Γ (a2 + z)Γ (a5 + z)Γ (−z) ×Γ (4 − a2357 − 2ε − z)Γ (4 − a2567 − 2ε − z) .

(4.56)

80

4 Evaluating by MB Representation

So, these two classes of integrals are rather simple because they are given only by onefold MB representations. Each of them can be evaluated by decomposing the integral into ‘singular’ and ‘regular’ parts. The singular parts correspond to the residues necessary to reveal the singular behaviour in ε while the regular parts are given by integrals where expansion in ε in the integrand is possible. For the boxes with a one-loop insertion, the singular part is written as minus the sum of the residues of the integrand at the points j−2ε, with j = −max{a1 , a3 } − a257 + 4, . . . , −1, plus the sum of the residues of the integrand at the points j − 2ε for j = 0, . . . , 4 − a. For the diagonal crossed boxes, the singular part is written as minus the sum of the residues of the integrand at the points j − 2ε, with j = −max{a3 , a6 } − a257 + 4, . . . , −1, plus the sum of the residues of the integrand at the points j−2ε for j = 0, . . . , 4−a. The regular parts can be written as MB integrals for −1 < Re z < 0 with an integrand expanded in a Laurent series in ε up to a desired order. Then these integrals are straightforwardly evaluated by closing the contour of integration to the right and taking residues at the points z = 0, 1, 2, . . .. At this step, one can use the collection of formulae for summing up series presented in Appendix C. The evaluation of both the singular and the regular parts can easily be implemented on a computer. Let us, for example, present an analytical result [32] for the box with a diagonal with all indices equal to one: d/2 −γ ε 2 iπ e E F0 (s, t, ε) , (4.57) K(s, t; 0, 1, 1, 0, 1, 1, 1, 0, ε) = − s+t where 1 F0 (s, t, ε) = − ln2 x + π 2 2ε2

+ 2Li3 (−x) − 2 ln x Li2 (−x) − ln2 x + π 2 ln(1 + x) 1 2 3 2 2 + ln x + ln(−s) ln x + π ln(−t) − 2ζ(3) 3 ε +4 (S2,2 (−x) − ln xS1,2 (−x)) − 4Li4 (−x) +4 (ln(1 + x) − ln(−s)) Li3 (−x) +2 ln2 x + 2 ln(−s) ln x − 2 ln x ln(1 + x) Li2 (−x) 2 3 2 2 ln x + ln(−s) ln x + π ln(−t) − 2ζ(3) ln(1 + x) +2 3 1 4 − ln2 x + π 2 ln2 (1 + x) − ln4 x − ln(−s) ln3 x 2 3 11 2 2 − ln (−s) + π ln2 x − π 2 ln2 (−s) − 2π 2 ln(−s) ln x 12 π4 , (4.58) +4ζ(3) ln(−t) − 20 and x = t/s.

4.5 Two-Loop Massive Examples

81

Concerning non-trivial checks of general formulae discussed in the end of Sect. 4.2 let us observe that, if we start from (4.46), we have to obtain, in the limit a1,3,4,6 → 0 with a8 = 0, the massless sunset diagram with the indices a2 , a5 , a7 . Indeed, we can start from (4.55) and perform the limit a3 → 0 by taking minus the residue at z1 = 4 − a1257 − 2ε in order to take into account the singularity of the integral of Γ (a − 4 + 2ε + z1 )Γ (4 − a1257 − 2ε − z1 ). Then we can set a1 = 0 and reproduce a known result. On the other hand, we should obtain the product of two one-loop massless propagator-type integrals with the indices (a1 , a3 ) and (a4 , a6 ) in the limit a2,5,7 → 0 with a8 = 0. Yes, we do this by a similar analysis and similar manipulations: take minus residue at z1 = 0 and set a2 = 0, then take minus residue at z4 = −z2 − z3 and set a5 = 0, then take residues at z2 = 0 and z3 = 0 and set a7 = 0. Representation (4.46) can be used for the evaluation, in expansion in ε, of any massless on-shell planar double box. See, e.g., [2] where it was applied to the evaluation of a double box with a numerator, K(s, t; 1, . . . , 1, −1, ε). Let us mention that, in this case, one meets a spurious singularity in MB integrals which can be cured by introducing an auxiliary analytic regularization. To do this, one can choose a8 = −1 + λ. Then the singularities in the MB integrals are ﬁrst resolved with respect to λ and then with respect to ε when λ and ε tend to zero. (The singularities in λ are indeed cancelled.) Non-planar double boxes can also be evaluated by MB representation – see [33].

4.5 Two-Loop Massive Examples Our next two-loop example is Example 4.9. Massive on-shell double box diagrams shown in Figs. 4.9 and 4.10. p1

1 2

p2

p3

6 7

3

5 4

(a)

p4 (b)

Fig. 4.9. Planar massive on-shell double boxes: (a) ﬁrst type, (b) second type. The solid lines denote massive, the dotted lines massless particles

This is an important class of Feynman integrals with one more parameter, with respect to the massless on–shell double boxes. In particular, it is relevant for Bhabha scattering.

82

4 Evaluating by MB Representation 6 1 5

2

7

3 4 Fig. 4.10. Non-planar massive on-shell double box

The general double box Feynman integral of the ﬁrst type (see Fig. 4.9a) takes the form dd k dd l 2 BPL,1 (s, t, m ; a1 , . . . , a8 , ε) = (k 2 − m2 )a1 [(k + p1 )2 ]a2 [(k + p1 + p2 + p4 )2 ]−a8 × [(k + p1 + p2 )2 − m2 ]a3 [(l + p1 + p2 )2 − m2 ]a4 [(l + p1 + p2 + p4 )2 ]a5 1 × 2 , (4.59) (l − m2 )a6 [(k − l)2 ]a7 where we consider a (non-negative) power −a8 of the factor (k +p1 +p2 +p4 )2 in the numerator as in the massless case. To derive an appropriate MB representation for (4.59) we proceed similarly to the massless case, i.e. recognize the internal integral over l as a massive box with two legs oﬀ-shell for which we use representation (4.30). After that the integral over k can be recognized as the massive on-shell box represented by (4.24), and we obtain the following sixfold MB representation [26]: d/2 2 iπ (−1)a (−s)4−a−2ε 2 BPL,1 (s, t, m ; a1 , . . . , a8 , ε) = j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε) 2 z1 +z5 w +i∞ +i∞ 5 m t 1 ... dw dzj Γ (a2 + w)Γ (−w) × (2πi)6 −i∞ −s s −i∞ j=1 Γ (z2 + z4 )Γ (z3 + z4 )Γ (4 − a13 − 2a28 − 2ε + z2 + z3 )Γ (a7 + w − z4 ) Γ (a1 + z3 + z4 )Γ (a3 + z2 + z4 ) Γ (a1238 − 2 + ε + z4 + z5 )Γ (a4567 − 2 + ε + w + z1 − z4 ) × Γ (4 − a46 − 2a57 − 2ε − 2w − 2z1 − z2 − z3 ) Γ (a8 − z2 − z3 − z4 )Γ (−w − z2 − z3 − z4 )Γ (2 − a238 − ε + z3 − z5 ) × Γ (4 − a1238 − 2ε + w − z4 )Γ (a8 − w − z2 − z3 − z4 ) Γ (a5 + w + z2 + z3 + z4 )Γ (2 − a567 − ε − w − z1 − z2 ) × Γ (4 − a13 − 2a28 − 2ε + z2 + z3 − 2z5 ) ×Γ (2 − a457 − ε − w − z1 − z3 )Γ (2 − a128 − ε + z2 − z5 ) (4.60) ×Γ (4 − a46 − 2a57 − 2ε − 2w − z2 − z3 )Γ (−z1 )Γ (−z5 ) . ×

4.5 Two-Loop Massive Examples

83

This general formula can be used to evaluate various Feynman integrals of the given family. Let us consider the example of the Feynman integral without numerator and ai = 1 for i = 1, 2, . . . , 7. Then (4.60) takes the form d/2 2 iπ (0) 2 2 B (s, t, m , ε) ≡ BPL,1 (s, t, m ; 1, . . . , 1, 0, ε) = − Γ (−2ε)(−s)3+2ε +i∞ +i∞ 5 z +z w m2 1 5 t 1 × ... dw dzj (2πi)6 −i∞ −s s −i∞ j=1 Γ (1 + w)Γ (−w)Γ (2 + ε + w + z1 − z4 )Γ (−1 − ε − w − z1 − z2 ) Γ (1 − 2ε + w − z4 )Γ (1 + z2 + z4 )Γ (1 + z3 + z4 ) Γ (−1 − ε − w − z1 − z3 )Γ (−z1 )Γ (−ε + z2 − z5 )Γ (−ε + z3 − z5 ) × Γ (−2ε + z2 + z3 − 2z5 )Γ (−2 − 2ε − 2w − 2z1 − z2 − z3 ) ×Γ (1 + ε + z4 + z5 )Γ (−z5 )Γ (−2ε + z2 + z3 )Γ (1 + w − z4 ) ×Γ (1 + w + z2 + z3 + z4 )Γ (−2 − 2ε − 2w − z2 − z3 )

×

×Γ (z2 + z4 )Γ (z3 + z4 )Γ (−z2 − z3 − z4 ) .

(4.61)

Observe that, because of the presence of the factor Γ (−2ε) in the denominator, we are forced to take some residue in order to arrive at a non-zero result at ε = 0, so that the integral is eﬀectively ﬁvefold. Let us apply our strategy of shifting contours and taking residues, with the goal to decompose (4.61) into pieces where the Laurent expansion ε of the integrand becomes possible. We shall evaluate this integral in expansion in ε up to a ﬁnite part. We know in advance that the poles in ε are now only of the second order because collinear divergences are absent. This is how such procedure can be performed in this case [26]: 1. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then a residue at z4 = 0, then a residue at z2 = 0, expand in a Laurent series in ε up to a ﬁnite part. Let us denote the resulting integral over z1 and z5 by B1 . 2. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then a residue at z4 = 0, and change the nature of the ﬁrst pole of Γ (z2 ) (choose a contour from the opposite side, i.e. the pole z2 will be now right), then expand in ε. Denote this integral over z1 , z2 and z5 by B2 . 3. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then change the nature of the ﬁrst pole of Γ (z4 ), then take a residue at z2 = −z4 , then take a residue at z4 = −ε and expand in ε. This resulting integral over z1 and z5 is denoted by B3 . 4. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then change the nature of the ﬁrst pole of Γ (z4 ), then take a residue at z2 = −z4 , then change the nature of the ﬁrst pole of Γ (2(ε + z4 )) and expand in ε. The resulting integral over z1 , z4 and z5 is denoted by B4 .

84

4 Evaluating by MB Representation

5. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then change the nature of the ﬁrst pole of Γ (z4 ), then change the nature of the ﬁrst pole of Γ (z2 + z4 ) and expand in ε. The resulting integral over z1 , z2 , z4 and z5 is denoted by B5 . 6. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then change the nature of the ﬁrst pole of Γ (−2(1 + 2ε + w)), then take minus residue at z4 = 1 + w, then minus residue at z2 = −1 − 2ε − w and expand in ε. The resulting integral over w, z1 and z5 is denoted by B6 . 7. Change the nature of the ﬁrst pole of Γ (−2 − 2ε − 2w − z2 − z3 ), then take minus residue at z4 = −z2 − z3 , then a residue at z3 = 2ε − z2 , then take a residue at z2 = 2ε and expand in ε. The resulting integral over w, z1 and z5 is denoted by B7 . One can see that all the other contributions vanish at ε = 0. By a suitable change of variables, one can observe that B7 = B6 . In fact, the dependence of the ﬁrst ﬁve contributions on the Mandelstam variable t is trivial: they are just proportional to 1/t. The two-dimensional integrals B1 and B3 are products of one-dimensional integrals which can be evaluated by closing the contour to the left and summing up resulting series with the help of formulae [11] of Appendix C. To evaluate the three-parametric integral B4 it is reasonable to observe that the integrand only changes its sign after the transformation {z4 → −z4 , z1 → z5 , z5 → z1 }. If we take into account that the change of variables z4 → −z4 implies that the initial integration contour −1 < Rez4 < 0 becomes 0 < Rez4 < 1 we will obtain a simple equation for B4 and conclude that the value of the integral equals 1/2 times the residue at z4 = 0. The latter quantity turns out to be a factorized integral over z1 and z5 which is evaluated like B1 and B3 . The three-dimensional integral B2 is evaluated by closing the integration contours over z1 and z5 to the left, summing up resulting series and applying a similar procedure to a ﬁnal integral in z2 . The corresponding result is naturally expressed in terms of polylogarithms, up to Li3 , depending on s and m2 in terms of the variable "√ √ #2 4m2 − s + −s v= √ . √ 4m2 − s − −s The form of this result provides a hint about a possible functional dependence of the result for the four-dimensional integral B5 , and a heuristic procedure which was explicitly formulated in [14] turns out to be successfully applicable here. First, all the contributions, in particular B4 , are analytic functions of s in a vicinity of the origin. One can observe that any given term of the Taylor expansion can be evaluated straightforwardly because the corresponding integrals over z2 and z4 are taken recursively. It is, therefore, possible to evaluate enough ﬁrst terms (say, 30) of this Taylor expansion.

4.5 Two-Loop Massive Examples

85

Then one takes into account the type of the functional dependence mentioned above, turns to a new Taylor series in terms of the variable v − 1 and assumes that the n-th term of this Taylor series is a linear combination, with unknown coeﬃcients, of the following quantities of levels 1, 2, 3, and 4, respectively: 1 , (4.62) n 1 S1 (n) , (4.63) , n2 n 1 S1 (n) S2 (n) S1 (n)2 , , (4.64) , , n3 n2 n n 1 S1 (n) S2 (n) S1 (n)2 , , , , n4 n3 n2 n2 S3 (n) S12 (n) S1 (n)S2 (n) S1 (n)3 , , , . (4.65) n n n n n where Sk (n) = j=1 j −k , etc. are nested sums (see Appendix C). Using the information about the ﬁrst terms of the Taylor series one solves a system of linear equations, ﬁnds those unknown coeﬃcients and checks this solution with the help of the next Taylor coeﬃcients. This experimental mathematics has turned out to be quite successful for the evaluation of B5 . Finally, the contribution B6 is a product of a onedimensional integral over z1 , which is easily evaluated, and a two-dimensional integral over w and z5 which involves a non-trivial dependence on t and is evaluated by closing the integration contour in z5 to the left, summing up a resulting series in terms of Gauss hypergeometric function for which one can apply the parametric representation (B.5). After that the internal integral over w is taken by the same procedure and, ﬁnally, one takes the parametric integral. The ﬁnal result takes the following form [26]: d/2 −γ ε 2 2 iπ e E x (0) 2 B (s, t, m ; ε) = − 2 1+2ε s (−t) b2 (x) b1 (x) + b × + (x) + b (x, y) + O(ε) , (4.66) 01 02 ε2 ε where x = 1/ 1 − 4m2 /s, y = 1/ 1 − 4m2 /t, and b2 (x) = 2(mx − px )2 , 1−x 1+x −2x b1 (x) = −8 Li3 + Li3 + Li3 2 2 1−x 2x 1−x −2x +Li3 + 4(mx − px ) Li2 − Li2 1+x 2 1−x −(4/3)m3x + 4m2x px − 6mx p2x + (2/3)p3x + 4l2 (mx px + p2x )

(4.67)

86

4 Evaluating by MB Representation

−2l22 (mx + 3px ) − (π 2 /3)(4l2 − mx − 3px ) + (8/3)l23 + 14ζ(3) , (4.68) 1+x b01 (x) = −8(mx − px ) Li3 (x) − Li3 (−x) − Li3 2 1−x 2x −2x +Li3 − Li3 + Li3 2 1+x 1−x 1−x +16Li2 (Li2 (x) − Li2 (−x)) 2 " 2 # 1−x 2 2 − 8Li2 (x) Li2 (−x) +4 Li2 (x) + Li2 (−x) + 4Li2 2 1−x 2 2 −(8/3)[π − 6l2 + 6lx px − 6mx (lx + px − 2l2 )]Li2 2 −(4/3)[π 2 − 6l22 + 3m2x + 6mx (2l2 − 2lx − px ) + 12lx px − 3p2x ] 2x ×(Li2 (x) − Li2 (−x)) + 8(mx − px ) (px − mx + 2l2 )Li2 1+x −2x +2(lx − mx + l2 )Li2 − 8(mx − px )(2lx − px − 5mx + 4l2 ) 1−x ×(−mx px + l2 (mx + px ) − l22 + π 2 /6) −(20/3)m4x + (164/3)m3x px − 40m2x p2x − (4/3)mx p3x − (8/3)p4x +8mx lx (m2x − 3mx px + 2p2x ) −4l2 (7m3x + 21m2x px − 4mx lx px − 23mx p2x + 4lx p2x − p3x ) −π 2 ((17/3)m2x − (4/3)mx lx − 2mx px + (4/3)lx px − (7/3)p2x ) +l22 (84m2x − 8mx lx − 16mx px + 8lx px − 44p2x ) −(8/3)l2 (6l22 − π 2 )(3mx − 2px ) − (4/3)π 2 l22 + 4l24 + π 4 /9 .

(4.69)

The last piece of the ﬁnite part comes from B6 and B7 : 1−x 1+x b02 (x, y) = 2(px − mx ) 4 Li3 − Li3 2 2 (1 − x)y −(1 + x)y −(1 − x)y +Li3 − Li3 + Li3 1 − xy 1 − xy 1 + xy (1 + x)y (1 + x)(1 − y) (1 − x)(1 + y) −Li3 + 2 Li3 − Li3 1 + xy 2(1 − xy) 2(1 − xy) (1 − x)(1 − y) (1 + x)(1 + y) −Li3 + Li3 2(1 + xy) 2(1 + xy) +2(my + py − mxy − pxy ) −2x 2x × 2Li2 (x) − 2Li2 (−x) + Li2 − Li2 1−x 1+x 1−x +4(mxy − pxy )(Li2 (−y) − Li2 (y)) − 4(mx + px − 2l2 )Li2 2

4.5 Two-Loop Massive Examples

1−y (1 − x)y − 4(mx + ly − mxy )Li2 2 1 − xy −(1 + x)y +4(px + ly − mxy )Li2 1 − xy −(1 − x)y −4(mx + ly − pxy )Li2 1 + xy (1 + x)y +4(px + ly − pxy )Li2 1 + xy (1 − x)(1 + y) +2(mx + px + my + py − 2mxy − 2l2 )Li2 2(1 − xy) (1 − x)(1 − y) +2(mx + px + my + py − 2pxy − 2l2 )Li2 2(1 + xy)

87

−4(mxy − pxy )Li2

+2p2x (my + py − mxy − pxy ) + 2px (2(my ly + my py + ly py ) +mxy (−my − 2ly − 3py + 3mxy ) + pxy (−3my − 2ly − py + 3pxy )) +2mx (2px + my − 2ly + py )(my + py − mxy − pxy ) − p2y (mxy + pxy ) +2py (2m2xy + p2xy ) + m2y (2py − mxy − pxy ) +2my (p2y + m2xy + 2p2xy − py (3mxy + pxy )) − 2(m3xy + p3xy ) +2l2 ((4my + 4py − 3mxy )mxy + (2my + 2py − 3pxy )pxy −2(px + 2mx )(my + py − mxy − pxy ) − m2y − 4my py − p2y ) +2l22 (3(my + py ) − 2(2mxy + pxy )) −(π 2 /3)(my + py − 8mxy + 6pxy ) .

(4.70)

The following abbreviations are used here: lz = ln z for z = x, y, 2, pz = ln(1 + z) and mz = ln(1 − z) for z = x, y, xy. This result is presented in such a way that it is manifestly real at small negative values of s and t. From this Euclidean domain, it can easily be continued analytically to any other domain. The result (4.66)–(4.70) is in agreement with the leading power behaviour in the (Sudakov) limit of the ﬁxed-angle scattering, m2 |s|, |t| which can be alternatively obtained [26] by use of the strategy of expansion by regions [4, 27]: d/2 −γ ε 2 iπ e E (0) 2 B (s, t, m ; ε) = − 2 s (−t)1+2ε 1 L2

× 2 2 − (2/3)L3 + (π 2 /3)L + 2ζ(3) ε ε −(2/3)L4 + 2 ln(t/s)L3 − 2(ln2 (t/s) + 4π 2 /3)L2

+ 4Li3 (−t/s) − 4 ln(t/s)Li2 (−t/s) + (2/3) ln3 (t/s)

−2 ln(1 + t/s) ln2 (t/s) + (8π 2 /3) ln(t/s) − 2π 2 ln(1 + t/s) + 10ζ(3) L +π 4 /36 + O(m2 L3 , ε) , (4.71)

88

4 Evaluating by MB Representation

where L = ln(−m2 /s). This asymptotic behaviour is reproduced when one starts from the result (4.66)–(4.70). Another check of such a complicated result came from the numerical integration based on a method of sector decompositions in the space of alpha parameters [7] (to be discussed in Sect. E.2). Let us stress that, in the present case with a non-zero mass, there are no collinear divergences and the poles in ε are only up to the second order, so that the resolution of singularities in ε in the MB integrals is relatively simple. Therefore, it looks promising to use the technique presented, starting from (4.60), for the evaluation of any given master integral. For example, the integral BPL,1 (s, t, m2 ; 1, . . . , 1, −1, ε) was evaluated in [31]. There is the same problem as in the massless case discussed above and connected with spurious singularities in MB integrals. It can also be cured in the same way, by introducing an auxiliary analytic regularization, e.g. with a8 = −1+λ. The singularities in the corresponding MB integral are ﬁrst resolved with respect to λ and then with respect to ε when λ and ε tend to zero. In the result [31], one meets not only usual polylogarithms but also a harmonic polylogarithm (HPL) [22] (see Appendix C), H−1,0,0,1 (−(1 − x)/(1 + x)) with x deﬁned after (4.66). Let us turn to the massive double boxes of the second type shown in Fig. 4.9b: dd k dd l BPL,2 (s, t, m2 ; a1 , . . . , a8 , ε) = 2 2 (k − m )a1 [(k + p1 )2 ]a2 [(k + p1 + p2 + p4 )2 ]−a8 × 2 2 a [(k + p1 + p2 ) − m ] 3 [(l + p1 + p2 )2 ]a4 [(l + p1 + p2 + p4 )2 − m2 ]a5 1 × 2 a6 . (4.72) (l ) [(k − l)2 − m2 ]a7 To derive a MB representation for (4.72) let us straightforwardly generalize the derivation of (4.60). For the subintegral over l we now use representation (4.31) of the massive box with two legs oﬀ-shell in the second variant. Then the integral over k can be recognized as the massive on-shell box (4.24). We therefore obtain the following sixfold MB representation [31]: d/2 2 iπ (−1)a (−s)4−a−2ε 2 BPL,2 (s, t, m ; a1 , . . . , a8 , ε) = j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε) 2 z5 +z6 z1 +i∞ +i∞ 6 6 m t 1 . . . dz Γ (−zj ) × j (2πi)6 −i∞ −s s −i∞ j=1 j=1 Γ (a4 + z2 + z4 )Γ (4 − a445667 − 2ε − z2 − z3 − 2z4 )Γ (a6 + z3 + z4 ) Γ (4 − a445667 − 2ε − z2 − z3 − 2z4 − 2z5 )Γ (6 − a − 3ε − z4 − z5 ) Γ (a2 + z1 )Γ (8 − a13 − 2a245678 − 4ε − 2z1 − z2 − z3 − 2z4 − 2z5 ) × Γ (8 − a13 − 2a245678 − 4ε − 2z1 − z2 − z3 − 2z4 − 2z5 − 2z6 ) ×

4.5 Two-Loop Massive Examples

89

Γ (2 − a456 − ε − z4 − z5 )Γ (2 − a467 − ε − z2 − z3 − z4 − z5 ) Γ (a45678 − 2 + ε + z2 + z3 + z4 + z5 )Γ (a1 − z3 )Γ (a3 − z2 ) ×Γ (a4567 + ε − 2 + z2 + z3 + z4 + z5 )Γ (a − 4 + 2ε + z1 + z4 + z5 + z6 )

×

×Γ (4 − a1245678 − 2ε − z1 − z2 − z4 − z5 − z6 ) ×Γ (4 − a2345678 − 2ε − z1 − z3 − z4 − z5 − z6 ) ×Γ (a45678 − 2 + ε + z1 + z2 + z3 + z4 + z5 ) ,

(4.73)

This representation was used in [31] to calculate the master planar double box of the second type BP L,2 (s, t, m2 ; 1, . . . , 1, 0, ε). The resolution of the singularities in ε was performed similar to the previous cases. The number of resulting MB integrals where an expansion in ε can be performed in the integrand is again equal to six. This time, some of the contributions turned out to be hardly evaluated in terms of known functions. Some two-parametric integrals of elementary functions entered the result in [31]. This result was controlled similarly to the previous case, by numerical evaluation of ﬁnite MB integrals and numerical evaluation by the method of [7] (to be discussed in Sect. E.2). We shall come back to the discussion of the problem of the evaluation of the massive on-shell double boxes in Chap. 7. To conclude this section let us turn to the non-planar graph of Fig. 4.10. Its MB representation can again be derived by using an MB representation for the subdiagram consisting of the lines (4, 5, 6, 7). This time, we can use (4.32). For the subsequent integral over the second loop momentum, we need the following MB representation for this auxiliary one-loop integral: dd k 2 2 a 2 1 (k − m ) [(k + p1 ) ]a2 [(k + p1 + p2 )2 − m2 ]a3 1 (−1)a iπ d/2 (−s)2−a−ε = 2 a 2 a [(k + p1 + p2 + p4 ) ] 4 [(k − p4 ) ] 4 Γ (4 − 2ε − a) Γ (al ) +i∞ +i∞ 4 2 z2 z3 z4 1 (m ) (−t) (−u) . . . dz Γ (−z ) × j j (2πi)4 −i∞ (−s)z2 +z3 +z4 −i∞ j=1

×

Γ (a245 + z1 + 2z3 + 2z4 ) Γ (a245 + z1 + z3 + 2z4 ) ×Γ (a2 + a4 + z1 + z3 + z4 )Γ (−a4 − z1 − z3 − z4 )Γ (a4 + z1 + z3 )

×Γ (a + ε − 2 + z2 + z3 + z4 )Γ (a5 + z4 )

×Γ (2 − a1245 − ε − z2 − z3 − z4 )Γ (2 − a2345 − ε − z2 − z3 − z4 ) Γ (4 − a12234455 − 2ε − 2z3 − 2z4 ) , (4.74) × Γ (4 − a12234455 − 2ε − 2z2 − 2z3 − 2z4 ) where u = (p1 + p4 )2 is a Mandelstam variable. It can be derived similarly to the previous MB representations for one-loop Feynman integrals. Using (4.74) one arrives at the following eightfold MB representation [31]:

90

4 Evaluating by MB Representation

iπ d/2

2

(−1)a (−s)4−a−2ε j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε) z +z z +i∞ +i∞ 8 7 m2 5 6 t 7 u z8 1 . . . dz Γ (−zj ) × j (2πi)8 −i∞ −s s s −i∞ j=1 j=1

BNP (s, t, u, m2 ; a1 , . . . , a8 , ε) =

Γ (a5 + z2 + z4 )Γ (a7 + z3 + z4 )Γ (4 − a455677 − 2ε − z2 − z3 − 2z4 ) Γ (a1 − z2 )Γ (a3 − z3 )Γ (a8 − z4 ) Γ (2 − a567 − ε − z2 − z4 − z5 )Γ (2 − a457 − ε − z3 − z4 − z5 ) × Γ (4 − a455677 − 2ε − z2 − z3 − 2z4 − 2z5 ) Γ (a8 + z1 − z4 + z7 )Γ (4 − a2345678 − 2ε − z2 − z5 − z6 − z7 − z8 ) × Γ (6 − a − 3ε − z5 ) Γ (8 − a13 − 2a245678 − 4ε − z2 − z3 − 2z5 − 2z7 − 2z8 ) × Γ (8 − a13 − 2a245678 − 4ε − z2 − z3 − 2z5 − 2z6 − 2z7 − 2z8 ) Γ (4 − a1245678 − 2ε − z3 − z5 − z6 − z7 − z8 ) × Γ (a245678 − 2 + ε + z1 + z2 + z3 + z5 + z7 + 2z8 ) ×Γ (a4567 + ε − 2 + z2 + z3 + z4 + z5 + z8 )Γ (−a8 − z1 + z4 − z7 − z8 ) ×Γ (a245678 − 2 + ε + z1 + z2 + z3 + z5 + 2z7 + 2z8 ) ×

×Γ (a − 4 + 2ε + z5 + z6 + z7 + z8 )Γ (a28 + z1 − z4 + z7 + z8 ) .

(4.75)

Representation (4.75) can be checked for various simple partial cases as it was explained above. Although the number of integrations is rather high one can proceed also in this case. However, it turns out that the massive non-planar case is rather complicated. A description of preliminary results for the master planar double box can be found in [31]. Let us now again illustrate the fact that general MB representations accumulate a lot of information so that MB representations for various classes of Feynman integrals can be derived in a very simple way from an initial global representation. Suppose we want to consider Example 4.10. Sunset diagrams of Fig. 3.12 with one zero mass and two equal non-zero masses at a general value of the external momentum squared. Remember that we have already considered such Feynman integrals at threshold, q 2 = 4m2 – see Example 3.6. There is no need to derive an appropriate MB representation from the beginning. Let us observe that such Feynman integrals, with the massive propagators 5 and 7 and the massless propagator 2, can be obtained from the massive on-shell double boxes of Fig. 4.9b at a1 = a3 = a4 = a6 = 0. As usual such a limit results in taking some residues. We ﬁrst let a4 → 0 and observe that Γ (a4 ) in the denominator can be cancelled only if we take into account the gluing in the product Γ (a4 + z2 + z4 )Γ (−z2 )Γ (−z4 ). Thus we are forced to take the two residues at

4.5 Two-Loop Massive Examples

91

z4 = 0 and z2 = 0. Then the limit a6 → 0 can similarly be taken, because of the presence of Γ (a6 + z3 )Γ (−z3 )/Γ (a6 ), by taking minus residue at z3 = 0. Then we take the limit a1 → 0 by observing that the only way to cancel Γ (a1 ) in the denominator is to take into account the gluing in the product Γ (a123578 − 4 + 2ε + z1 + z5 + z6 )Γ (4 − a23578 − 2ε − z1 − z5 − z6 ) and take a residue, e.g. at z6 = 4 − a23578 − 2ε − z1 − z5 (with the minus sign). Finally, we let a3 → 0 by distinguishing the product Γ (a23578 − 4 + 2ε + z1 + z5 )Γ (8 − a3 − 2a2578 − 4ε − 2z1 − 2z5 ) which generates Γ (a3 ) and cancels this factor in the denominator. After relabelling the lines, substituting t → q 2 and expressing the irreducible numerator in terms of the loop momenta of the sunset diagram, we obtain F4.10 (q 2 , m2 ; a1 , a2 , a3 , a4 , d) dd k dd l [(k + l)2 ]−a4 = (k 2 − m2 )a1 (l2 − m2 )a2 [(q − k − l)2 ]a3 d/2 2 +i∞ 2 z iπ (−1)a Γ (2 − a3 − ε) 1 q = dz Γ (a1 )Γ (a2 )Γ (a3 )(m2 )a−4+2ε 2πi −i∞ m2 Γ (a − 4 + 2ε + z)Γ (a3 + z)Γ (−z)Γ (2 − a34 − ε − z) × Γ (a12 + 2a34 − 4 + 2ε + 2z)Γ (2 − ε + z)Γ (2 − a3 − ε − z) ×Γ (a134 − 2 + ε + z)Γ (a234 − 2 + ε + z) . (4.76) If we evaluate the integral in (4.76) for general ε by closing the contour and taking a series of residues we shall reproduce the result of [8] in terms of the hypergeometric series 4 F3 . We are oriented, however, at the evaluation in expansion in ε and will evaluate integrals (4.76), for concrete values of the indices, by resolving singularities in ε and then closing the contour and summing up the corresponding series. For example, (4.76) gives 2 F4.10 (q 2 , m2 ; 1, 1, 1, 0, d) = − iπ d/2 Γ (1 − ε)(m2 )1−2ε +i∞ 2 z q Γ (2ε − 1 + z)Γ (ε + z)2 Γ (1 + z)Γ (−z) 1 . (4.77) × dz 2 2πi −i∞ m Γ (2ε + 2z)Γ (2 − ε + z) The resolution of the singularities in ε is standard: we distinguish the factor Γ (2ε − 1 + z) as the source of poles. We have to take care of its ﬁrst two poles, i.e. take residues at z = 1 − 2ε and z = −2ε. The calculation of the integral with the opposite nature of these two poles is performed by closing the integration contour to the right and summing up series, with the following result which can be found in [11, 13]: 2 1 1 q2 F4.10 (q 2 , m2 ; 1, 1, 1, 0, d) = iπ d/2 (m2 )1−2ε 2 + 3 − ε 4m2 ε π2 11 13(1 + x2 ) 1 + 2x − x2 + + + + ln x 6 4 8x 2x

92

4 Evaluating by MB Representation

1 − x + x2 2 2 ln x − − ln x + O(ε) , (4.78) 1−x (1 − x)2 where x = ( 4m2 − q 2 − −q 2 )/( 4m2 − q 2 + −q 2 ). (Please, note that the letter x is used in various ways: this is another function in Examples 4.6, 4.9, while, for massless double and triple boxes, this is simply t/s.)

4.6 Three-Loop Examples Our next example is already at three-loop level: Example 4.11. The massless on-shell triple box diagram of Fig. 4.11.

p1

1 2

p2

6 7

3

5 4

p3

10 9 8

p4

Fig. 4.11. Triple box

The general planar triple box Feynman integral without numerator takes the form dd k dd l dd r T (s, t; a1 , . . . , a10 , ε) = (k 2 )a1 [(k + p2 )2 ]a2 [(k + p1 + p2 )2 ]a3 1 × 2 a 4 [(l + p1 + p2 ) ] [(r − l)2 ]a5 (l2 )a6 [(k − l)2 ]a7 1 × . (4.79) 2 a 8 [(r + p1 + p2 ) ] [(r + p1 + p2 + p4 )2 ]a9 (r2 )a10 To derive a suitable MB representation for (4.79) we proceed like in the derivation of (4.46). We recognize the internal integral over the loop momentum r as a box with two legs oﬀ-shell given by (4.22). After inserting it into (4.79) we obtain an MB integral of the on-shell double box with certain indices dependent on MB integration variables. These straightforward manipulations lead [29] to the following sevenfold MB representation of (4.79): d/2 3 iπ (−1)a (−s)6−a−3ε T (s, t; a1 , . . . , a10 , ε) = j=2,5,7,8,9,10 Γ (aj )Γ (4 − a589(10) − 2ε) z1 +i∞ +i∞ 7 t 1 Γ (a2 + z1 )Γ (−z1 )Γ (z2 + z4 ) × ... dzj 7 (2πi) −i∞ s Γ (a1 + z3 + z4 )Γ (a3 + z2 + z4 ) −i∞ j=1

4.6 Three-Loop Examples

93

Γ (2 − a12 − ε + z2 )Γ (2 − a23 − ε + z3 )Γ (a7 + z1 − z4 )Γ (−z5 )Γ (−z6 ) Γ (4 − a123 − 2ε + z1 − z4 )Γ (a6 − z5 )Γ (a4 − z6 ) ×Γ (z3 + z4 )Γ (a123 − 2 + ε + z4 )Γ (z1 + z2 + z3 + z4 − z7 )

×

×Γ (2 − a59(10) − ε − z5 − z7 )Γ (2 − a589 − ε − z6 − z7 ) ×Γ (a467 − 2 + ε + z1 − z4 − z5 − z6 − z7 )Γ (a5 + z5 + z6 + z7 ) ×Γ (4 − a467 − 2ε + z5 + z6 + z7 )Γ (a589(10) − 2 + ε + z5 + z6 + z7 ) ×Γ (2 − a67 − ε − z1 − z2 + z5 + z7 )Γ (a9 + z7 ) ×Γ (2 − a47 − ε − z1 − z3 + z6 + z7 ) , (4.80) 10 where a = i=1 ai , a589(10) = a5 + a8 + a9 + a10 , etc. In the case of the master triple box, we set ai = 1 for i = 1, 2, . . . , 10 to obtain T (0) (s, t, ε) ≡ T (1, . . . , 1; s, t, ε) d/2 3 z1 +i∞ +i∞ 7 iπ t 1 ... dzj Γ (1 + z1 ) = Γ (−2ε)(−s)4+3ε (2πi)7 −i∞ s −i∞ j=1 Γ (−z1 )Γ (−ε + z2 )Γ (−ε + z3 )Γ (1 + z1 − z4 )Γ (−z2 − z3 − z4 ) Γ (1 + z2 + z4 )Γ (1 + z3 + z4 )Γ (1 − 2ε + z1 − z4 ) Γ (z2 + z4 )Γ (z3 + z4 )Γ (−z5 )Γ (−z6 )Γ (z1 + z2 + z3 + z4 − z7 ) × Γ (1 − z5 )Γ (1 − z6 )Γ (1 − 2ε + z5 + z6 + z7 ) ×Γ (2 + ε + z5 + z6 + z7 )Γ (−1 − ε − z5 − z7 )Γ (−1 − ε − z6 − z7 ) ×Γ (1 + z7 )Γ (1 + ε + z1 − z4 − z5 − z6 − z7 )Γ (−ε − z1 − z2 + z5 + z7 ) ×

×Γ (1 + ε + z4 )Γ (−ε − z1 − z3 + z6 + z7 )Γ (1 + z5 + z6 + z7 ) .

(4.81)

Observe that, because of the presence of the factor Γ (−2ε) in the denominator, we are forced to take some residue in order to arrive at a non-zero result at ε = 0, so that the integral is eﬀectively sixfold. Then our standard procedure of taking residues and shifting contours can be applied, with the goal to obtain a sum of integrals where one may expand integrands in Laurent series in ε. The analysis of the integrand shows that the following four gamma functions play a crucial role for the generation of poles in ε: Γ (−ε + z2,3 ) and Γ (−1 − ε − z6,5 − z7 ). The ﬁrst decomposition of the integral (4.81) arises when one either takes a residue at the ﬁrst pole of one of these gamma functions or shifts the corresponding contour, i.e. changes the nature of this pole. As a result (4.81) is decomposed as 2T0001 + 2T0010 + 2T0011 +T0101 +2T0110 +2T0111 +T1010 +2T1011 +T1111 where the symmetry of the integrand is taken into account. Here the value 1 of an index means that a residue is taken and 0 means a shifting of a contour. The ﬁrst two indices correspond to the gamma functions Γ (−ε + z2 ) and Γ (−1 − ε − z5 − z7 ) and the second two indices to Γ (−ε + z3 ) and Γ (−1 − ε − z6 − z7 ), respectively. The term T0000 is absent because it is zero at ε = 0 due to Γ (−2ε) in the denominator.

94

4 Evaluating by MB Representation

Each of these terms is further decomposed appropriately and, eventually, one is left with integrals where integrands can be expanded in ε. These resulting terms involve up to ﬁve integrations. Taking some of these integrations with the help of the table of formulae presented in Appendix D, one can reduce all the integrals to no more than twofold MB integrals of gamma functions and their derivatives. In some of them, one more integration can be performed also in terms of gamma functions. Then the last integration, over z1 , is performed by taking residues and summing up resulting series, in terms of HPL. Keeping in mind the Regge limit, t/s → 0, let us, for deﬁniteness, decide to close the contour of the ﬁnal integration, over z1 , to the right and obtain power series in t/s. The coeﬃcients of these series are (up to of 1/n6 , S1 (n)/n5 , . . . , S1 (n)S3 (n)/n2 , . . ., where (−1)n ) linear n combinations −k , etc. (see Appendix C). Summing up these series with Sk (n) = j=1 j the help of tabulated formulae of Appendix C gives results in terms of HPL of the variable −t/s which can be continued analytically to any domain from the region |t/s| < 1. In the twofold MB integrals where one more integration (over a variable diﬀerent from z1 ) can hardly be performed in terms of gamma functions, one performs it with z1 in a vicinity of an integer point z1 = n = 0, 1, 2, . . ., in expansion in z = z1 − n, with a suﬃcient accuracy. Then one obtains power series where, in addition to nested sums with one index, various nested sums (see Appendix C) appear. These series are also summed up in terms of HPL. Eventually one arrives at the following result [29]: d/2 −γ ε 3 6 cj (x, L) iπ e E (0) , (4.82) T (s, t; ε) = − 3 1+3ε s (−t) εj j=0 where x = −t/s, L = ln(s/t), and c6 =

16 5 3 , c5 = − L , c4 = − π 2 , 9 3 2

3 c3 = 3(H0,0,1 (x) + LH0,1 (x)) + (L2 + π 2 )H1 (x) 2 11 2 131 ζ(3) , − π L− 12 9

(4.83)

(4.84)

c2 = −3 (17H0,0,0,1 (x) + H0,0,1,1 (x) + H0,1,0,1 (x) + H1,0,0,1 (x)) 3 −L (37H0,0,1 (x) + 3H0,1,1 (x) + 3H1,0,1 (x)) − (L2 + π 2 )H1,1 (x) 2 3 3 23 2 2 2 L + 8π H0,1 (x) − L + π L − 3ζ(3) H1 (x) − 2 2 1411 4 49 π , (4.85) + ζ(3)L − 3 1080

4.6 Three-Loop Examples

95

c1 = 3 (81H0,0,0,0,1 (x) + 41H0,0,0,1,1 (x) + 37H0,0,1,0,1 (x) + H0,0,1,1,1 (x) +33H0,1,0,0,1 (x) + H0,1,0,1,1 (x) + H0,1,1,0,1 (x) + 29H1,0,0,0,1 (x) +H1,0,0,1,1 (x) + H1,0,1,0,1 (x) + H1,1,0,0,1 (x)) + L (177H0,0,0,1 (x) +85H0,0,1,1 (x) + 73H0,1,0,1 (x) + 3H0,1,1,1 (x) + 61H1,0,0,1 (x) +3H1,0,1,1 (x) + 3H1,1,0,1 (x)) 119 2 139 2 47 2 2 L + π H0,0,1 (x) + L + 20π H0,1,1 (x) + 2 12 2 35 2 3 2 L + 14π 2 H1,0,1 (x) + L + π 2 H1,1,1 (x) + 2 2 23 3 83 2 L + π L − 96ζ(3) H0,1 (x) + 2 12 3 3 L + π 2 L − 3ζ(3) H1,1 (x) + 2 9 4 25 2 2 13 L + π L − 58ζ(3)L + π 4 H1 (x) + 8 8 8 73 2 301 503 4 π L + π ζ(3) − ζ(5) , − 1440 4 15 c0 = − (951H0,0,0,0,0,1 (x) + 819H0,0,0,0,1,1 (x) + 699H0,0,0,1,0,1 (x) +195H0,0,0,1,1,1 (x) + 547H0,0,1,0,0,1 (x) + 231H0,0,1,0,1,1 (x) +159H0,0,1,1,0,1 (x) + 3H0,0,1,1,1,1 (x) + 363H0,1,0,0,0,1 (x) +267H0,1,0,0,1,1 (x) + 195H0,1,0,1,0,1 (x) + 3H0,1,0,1,1,1 (x) +123H0,1,1,0,0,1 (x) + 3H0,1,1,0,1,1 (x) + 3H0,1,1,1,0,1 (x) +147H1,0,0,0,0,1 (x) + 303H1,0,0,0,1,1 (x) + 231H1,0,0,1,0,1 (x) +3H1,0,0,1,1,1 (x) + 159H1,0,1,0,0,1 (x) + 3H1,0,1,0,1,1 (x) +3H1,0,1,1,0,1 (x) + 87H1,1,0,0,0,1 (x) + 3H1,1,0,0,1,1 (x) +3H1,1,0,1,0,1 (x) + 3H1,1,1,0,0,1 (x)) −L (729H0,0,0,0,1 (x) + 537H0,0,0,1,1 (x) + 445H0,0,1,0,1 (x) +133H0,0,1,1,1 (x) + 321H0,1,0,0,1 (x) + 169H0,1,0,1,1 (x) +97H0,1,1,0,1 (x) + 3H0,1,1,1,1 (x) + 165H1,0,0,0,1 (x) +205H1,0,0,1,1 (x) + 133H1,0,1,0,1 (x) + 3H1,0,1,1,1 (x) +61H1,1,0,0,1 (x) + 3H1,1,0,1,1 (x) + 3H1,1,1,0,1 (x)) 311 2 619 2 531 2 89 2 L + π H0,0,0,1 (x) − L + π H0,0,1,1 (x) − 2 4 2 12 71 2 247 2 307 2 2 L + π H0,1,0,1 (x) − L + 32π H0,1,1,1 (x) − 2 12 2 107 2 151 2 197 2 2 L − π H1,0,0,1 (x) − L + 50π H1,0,1,1 (x) − 2 12 2

(4.86)

96

4 Evaluating by MB Representation

35 2 3 2 L + 14π 2 H1,1,0,1 (x) − L + π 2 H1,1,1,1 (x) 2 2 119 3 317 2 L + π L − 455ζ(3) H0,0,1 (x) − 2 12 47 3 179 2 L + π L − 120ζ(3) H0,1,1 (x) − 2 12 35 3 35 2 L + π L − 156ζ(3) H1,0,1 (x) − 2 12 3 3 L + π 2 L − 3ζ(3) H1,1,1 (x) − 2 69 4 101 2 2 559 4 L + π L − 291ζ(3)L + π H0,1 (x) − 8 8 90 27 5 25 2 3 9 4 25 2 2 13 L + π L − 58ζ(3)L + π 4 H1,1 (x) − L + π L − 8 8 8 40 8 183 131 4 37 ζ(3)L2 + π L − π 2 ζ(3) + 57ζ(5) H1 (x) − 2 60 12 223 2 167 624607 6 π ζ(3) + 149ζ(5) L + ζ(3)2 − π . (4.87) + 12 9 544320

−

The above result was conﬁrmed with the help of numerical integration in the space of alpha parameters [7]. Another natural check of the result is its agreement with the leading power Regge asymptotic behaviour [28] which was evaluated by an independent method based on the strategy of expansion by regions [4, 27]. The procedure described above can be applied, in a similar way, to the calculation of any massless planar on-shell triple box. At a ﬁrst step, one has to take care of the following four gamma functions in (4.80): Γ (2 − a12 − ε + z2 ), Γ (2 − a23 − ε + z3 ), Γ (2 − a59(10) − ε − z5 − z7 ), Γ (2 − a589 − ε − z6 − z7 ) . This procedure gives a decomposition similar to 2T0001 + 2T0010 + . . .. Next steps will be also generalizations of the corresponding steps in the evaluation of (4.81). The result presented above shows that analytical calculations of fourpoint on-shell massless Feynman diagrams at the three-loop level are quite possible so that one may think of evaluating three-loop virtual corrections to various scattering processes. Let us now consider a more complicated fourpoint three-loop diagram: Example 4.12. The massless on-shell tennis court5 diagram of Fig. 4.12. 5

Well, this is only one half of the court for singles. One also can call it ‘window’.

4.6 Three-Loop Examples

p1

97

p3

8 9

10

1 2

6 7

3

p2

5 4

p4

Fig. 4.12. Three-loop tennis court graph

To derive an appropriate MB representation we can proceed again quite straightforwardly. Here we need an auxiliary MB representation for the double box with two legs oﬀ shell applied to the double box subintegral in Fig. 4.12 and inserted into the MB representation for the on-shell box. As a result, an eightfold MB representation can be derived for the general diagram W (s, t; a1 , . . . , a11 , ε) of Fig. 4.12 with the eleventh index corresponding to the numerator (l1 ·l3 )−a11 which involves the scalar product of the momenta l1,3 ﬂowing though lines 1 and 3 in the same direction. Feynman integrals corresponding to Fig. 4.12 and many others will be indeed necessary to perform three-loop calculations of various scattering processes. It turns out that triple boxes are necessary right now in order to check some relations between diﬀerent loop orders in N = 4 supersymmetric gauge theories. The N = 4 theory has attracted considerable interest because of its remarkably simple structure and central role in the AdS/CFT correspondence. As was recently emphasized in [1], one needs, in addition to the result (4.87) for the triple box considered above, just one more triple box [6], namely, W (s, t; 1, . . . , 1, −1, ε). For this integral, one has d/2 3 iπ W (s, t; 1, . . . , 1, −1, ε) = − Γ (−2ε)(−s)1+3ε t2 w +i∞ +i∞ 7 t 1 × ... dw dz1 dzj Γ (−zj ) Γ (1 + 3ε + w) (2πi)8 −i∞ s −i∞ j=2 Γ (−3ε − w)Γ (1 + z1 + z2 + z3 )Γ (−1 − ε − z1 − z3 )Γ (1 + z1 + z4 ) Γ (1 − z2 )Γ (1 − z3 )Γ (1 − z6 )Γ (1 − 2ε + z1 + z2 + z3 ) Γ (−1 − ε − z1 − z2 − z4 )Γ (2 + ε + z1 + z2 + z3 + z4 ) × Γ (−1 − 4ε − z5 )Γ (1 − z4 − z7 )Γ (2 + 2ε + z4 + z5 + z6 + z7 ) ×Γ (−ε + z1 + z3 − z5 )Γ (2 − w + z5 )Γ (−1 + w − z5 − z6 )

×

×Γ (z5 + z7 − z1 )Γ (1 + z5 + z6 )Γ (−1 + w − z4 − z5 − z7 ) ×Γ (−ε + z1 + z2 − z5 − z6 − z7 )Γ (1 − ε − w + z4 + z5 + z6 + z7 ) ×Γ (1 + ε − z1 − z2 − z3 + z5 + z6 + z7 ) . (4.88)

98

4 Evaluating by MB Representation

There is again the factor Γ (−2ε) in the denominator, so that the integral is eﬀectively sevenfold. The evaluation of this integral in expansion in ε is in progress. Here is a preliminary result up to 1/ε3 : d/2 −γ ε 3 6 cj iπ e E W (s, t; 1, . . . , 1, −1, ε) = − , (4.89) 1+3ε 2 (−s) t εj i=0 where 16 13 19 1 , c5 = − ln x , c4 = − π 2 + ln2 x 9 6 12 2 5 7 3 5 ln x − ln2 x ln(1 + x) c3 = [Li3 (−x) − ln x Li2 (−x)] + 2 12 4 157 2 5 2 241 π ln x − π ln(1 + x) − ζ(3) + 72 4 18 with x = t/s. c6 =

(4.90)

4.7 More Loops One can proceed in the same style even in higher loops. Let us illustrate this point by considering Example 4.13. The four-loop ladder massless on-shell diagram shown in Fig. 4.13.

p1

1 2

p2

6 7

3

10 5

4

9 8

p3

13 12 11

p4

Fig. 4.13. Four-loop ladder diagram

We start with the derivation of an appropriate MB representation for general powers of the propagators. As before we use this general strategy because it provides a lot of checks and gives the possibility to obtain MB representations for various diagrams which result from the given diagram when contracting some lines. As in the previous example, we need an auxiliary MB representation for the double box with two legs oﬀ shell but in a diﬀerent situation (two left legs rather than two upper legs oﬀ shell). It can easily be derived by the technique described and takes the form

4.7 More Loops

99

d/2 2 iπ (−1)a a−4+2ε j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε)(−s) +i∞ +i∞ 6 1 (−t)z4 (−p21 )z5 (−p22 )z6 × . . . dzj Γ (−zj ) 6 (2πi) −i∞ (−s)z4 +z5 +z6 −i∞ j=1

K2 (s, t; a1 , . . . , a8 , ε) =

Γ (a4567 + ε − 2 + z1 + z2 + z3 )Γ (a5 + z1 )Γ (a7 + z1 + z2 + z3 ) Γ (4 − 2ε − a1238 + z1 + z2 + z3 )Γ (a1 − z2 )Γ (a3 − z3 )Γ (a8 − z1 ) ×Γ (a1238 + ε − 2 − z1 − z2 − z3 + z4 + z5 + z6 )Γ (a2 + z4 + z5 + z6 ) ×Γ (2 − ε − a457 − z1 − z3 )Γ (2 − ε − a567 − z1 − z2 )Γ (a8 − z1 + z4 )

×

×Γ (2 − ε − a128 + z1 + z2 − z4 − z5 ) ×Γ (2 − ε − a238 + z1 + z3 − z4 − z6 ) .

(4.91)

Then, similarly to the derivation of the multiple MB representation for the triple box when we inserted the MB representation of the box with two legs oﬀ shell into the MB representation of the on-shell double box, let us now insert (4.91) instead. We come to the following tenfold MB representation of the four-loop ladder diagram: d/2 2 iπ (−1)a (−s)8−a−4ε Q(s, t; a1 , . . . , a13 , ε) = j=2,5,7,9,11,12,13 Γ (aj )Γ (4 − a9,11,12,13 − 2ε) z7 +i∞ +i∞ 10 t 1 . . . dz Γ (−zj ) × j 10 (2πi) s −i∞ −i∞ j=1 j=2,3,5,6,7,8,9 Γ (a1,2 + z1 )Γ (a2 + z7 )Γ (z7 − z10 )Γ (z10 − z4 )Γ (z4 − z1 ) Γ (a10 − z2 )Γ (a8 − z3 )Γ (a6 − z5 )Γ (a4 − z6 )Γ (a1 − z8 )Γ (a3 − z9 ) Γ (2 − ε − a9,11,12 − z1 − z3 )Γ (2 − ε − a9,12,13 − z1 − z2 ) × Γ (4 − 2ε − a5,8,10 + z1 + z2 + z3 )Γ (4 − 2ε − a4,6,7 + z4 + z5 + z6 ) Γ (a9 + z1 + z2 + z3 )Γ (a9,11,12,13 + ε − 2 + z1 + z2 + z3 ) × Γ (4 − 2ε − a1,2,3 + z8 + z9 + z10 ) ×Γ (2 − ε − a5,10 + z1 + z2 − z4 − z5 )Γ (2 − ε − a5,8 + z1 + z3 − z4 − z6 ) ×Γ (a5 + z4 + z5 + z6 )Γ (a5,8,10 + ε − 2 − z1 − z2 − z3 + z4 + z5 + z6 ) ×

×Γ (2 − ε − a6,7 + z4 + z5 − z8 − z10 )Γ (2 − ε − a1,2 + z8 + z10 − z7 ) ×Γ (2 − ε − a4,7 + z4 + z6 − z9 − z10 )Γ (2 − ε − a2,3 + z9 + z10 − z7 ) ×Γ (a1,2,3 + ε − 2 − z8 − z9 − z10 + z7 )Γ (a7 + z8 + z9 + z10 ) ×Γ (a4,6,7 + ε − 2 − z4 − z5 − z6 + z8 + z9 + z10 ) ,

(4.92)

where we separate indices in a9,11,12,13 = a9 + a11 + a12 + a13 etc. by commas because they are now two-digit. One can check this monster representation as before, using partial cases: when we put the indices a2 , a5 , a7 , a9 , a12 to zero we reproduce a known analytical result for the product of four one-loop propagator diagrams with

100

4 Evaluating by MB Representation

Fig. 4.14. The ‘N in O’ diagram

the indices (a1 , a3 ), (a4 , a6 ), (a8 , a10 ) and (a11 , a13 ). When we put the indices a1 , a3 , a4 , a6 , a8 , a10 , a11 , a13 to zero we reproduce a known analytical result for the four-loop water melon diagram with the indices a2 , a5 , a7 , a9 , a12 and the external momentum square t. Representation (4.92) contains a lot of information. Let us use it in order to calculate the ‘N in O’ diagram6 shown in Fig. 4.14 exactly in four dimensions, i.e. at ε = 0. This is nothing but N (q 2 ) = Q(s, t; 1, 0, 1, 1, 1, 0, 1, 0, 1, 1, 1, 0, 1, 0) which is, of course, independent of t and proportional to 1/q 2 . The limit a2 , a12 → 0 is achieved as described above, due to four residues with respect to some of the integration variables. Then one can simply set a6 = a8 = 0 and obtain 4 C (4.93) N (q 2 ) = iπ 2 q2 with the constant C given by a ﬁnite ﬁvefold MB integral. Three of these ﬁve integrations can be performed explicitly with the help of tabulated formulae of Appendix D, and one can obtain the following twofold MB integral: +i∞ +i∞ dz1 dz2 1 Γ (z1 + z2 )Γ (1 − z1 − z2 )Γ (z2 )Γ (−z2 ) C= 2 (2πi) −i∞ −i∞ 2z12 z2 ×Γ (1 − z1 )Γ (z1 ) [z1 (ψ(1 − z1 ) + ψ(z1 ) − ψ(1 − z1 − z2 ) − ψ(z1 + z2 )) −z2 (ψ(1 − z1 − z2 ) − ψ(−z2 ) − ψ(z2 ) + ψ(z1 + z2 ))]

× ψ(z1 )2 − 2ψ(z1 )ψ(1 − z1 − z2 ) + 2ψ(1 − z1 − z2 )ψ(z1 + z2 ) −ψ(z1 + z2 )2 − ψ (z1 ) + ψ (z1 + z2 ) ,

(4.94)

where the poles at z1 = 0 and z2 = 0 are considered left so that one can choose 0 < Rez1 , Rez2 < 1 with Rez1 + Rez2 < 1 for the integration contour. One can check numerically, with a high accuracy, that the known result which will be presented shortly is successfully reproduced. 6

This diagram was a challenge in the eighties in renormalization group calculations. In the ﬁrst result on the ﬁve-loop β-function in the φ4 theory [9] (see [19] for a corrected later version) the contribution of this diagram was treated numerically. The analytical value of this diagram was predicted and later proven in [18] using a technique based on functional equations – see more details in Appendix F.

4.7 More Loops

101

The twofold MB integral (4.94) can be converted into a sum of two twofold series of expressions consisting of nested sums (see Appendix C). The ﬁrst of them is obtained by taking residues at the points z2 = 1, 2, . . . and then at z1 = 1, 2, . . .. The second of them is obtained by taking residues at the points z2 = 1 − z1 + n2 with n2 = 1, 2, . . . and then at z1 = 1, 2, . . .. Then one can perform one of the summations using the package SUMMER [39] and arrive at the following onefold series: C=

5 ∞ cj,n n=1 j=1

nj

,

(4.95)

where c5,n = 5π 2 /6 − 6S12 − 27S2 , c4,n = 5π 2 S1 /2 + 3S13 − 18S12 + 12S1 S2 − 6S3 + 12ζ(3) ,

(4.96) (4.97)

c3,n = π 4 /5 − 4π 2 S12 /3 − S14 /2 − 28S112 + 20S1 S12 − 10S13 −19π 2 S2 /6 − S12 S2 + 37S22 /2 + 4S1 S3 + 11S4 + 6S1 ζ(3) ,

(4.98)

c2,n = π 4 S1 /10 + π 2 S13 /6 − 2S1112 − 18S113 − 17π 2 S12 /6 + 16S1 S13 +11S14 + 4π 2 S1 S2 /3 − 2S13 S2 /3 + 6S12 S2 − S1 S22 − 13S212 +19S23 − π 2 S3 /3 − 5S12 S3 + 2S2 S3 /3 − 6S1 S4 − 4S5 − 2π 2 ζ(3)/3 −3S12 ζ(3) − S2 ζ(3) + 14ζ(5) , (4.99) 2 c1,n = 61π 6 /2520 − 16S1113 + π 2 S112 /3 + 4S1 S113 + 14S114 − 3S12 +10S123 − 3π 2 S13 − 5S1 S14 + 8S15 + 3π 4 S2 /20 − π 2 S12 S2 /6 +6S112 S2 − S1 S12 S2 + 10S13 S2 − 5π 2 S22 /6 + S12 S22 + 5S23 /3 − 8S2112 +S1 S212 − 3S1 S23 + 18S24 + 10π 2 S1 S3 /3 + 2S13 S3 /3 − 4S12 S3 −7S1 S2 S3 + 10S32 /3 − π 2 S4 /6 − 3S12 S4 − 31S2 S4 − 9S1 S5 − 80S6 /3 −4S12 ζ(3) + 4S1 S2 ζ(3) + 14S3 ζ(3) − 9ζ(3)2 ,

(4.100)

and we omit the argument n−1 in all the nested sums involved, i.e. S1 stands for S1 (n − 1) etc. Summation of the terms with 1/n5 , . . . ,1/n2 can be performed with the help of formulae (C.51)–(C.82) implemented in SUMMER [39]. The terms with 1/n are also successfully summed up by SUMMER, and we arrive at the wellknown result [18]: 1 4 441 ζ(7) . (4.101) N (q 2 ) = 2 iπ 2 q 8 I cannot say that the derivation of this result outlined above is simpler than that of [18]. Let me, however, stress that the present derivation involves a lot of steps that are performed automatically, and a lot of other similar results (e.g. for diagrams which can be obtained from the four-loop ladder diagram by shrinking other lines to points) can be obtained quite similarly.

102

4 Evaluating by MB Representation

4.8 MB Representation versus Expansion by Regions To expand a given Feynman integral in some limit, where certain masses and/or kinematical invariants are large with respect to the rest of these parameters, one can successfully apply expansion by regions [4, 30], as explained in the book [27] in detail. An alternative technique for solving the problem of asymptotic expansion is provided by multiple MB representations. Let us see how it works using some of our previous examples. For Example 4.1, we have derived the MB representation (4.3). Let us use it to expand such Feynman integrals in the two diﬀerent limits, m2 /q 2 → 0 and q 2 /m2 → 0. Consider, for example, F4.1 (2, 1, 4) represented by (4.4). This is an integral over the variable z, with the ratio m2 /q 2 present in the form (m2 /q 2 )z . The initial integration contour is at −1 < Rez < 0. Let us observe that if we follow the procedure used to evaluate this integral, i.e. close the integration contour to the right and pick up (minus) residues at z = 0, 1, 2, . . . , n, . . . we shall obtain terms of the asymptotic expansion in the limit m2 /q 2 → 0. Indeed, one can prove that the remainder of this expansion determined by picking up the (n + 1)-st residue is of order (m2 )n+1 . Thus we obtain iπ 2 m2 m4 −q 2 − ... . (4.102) F4.1 (2, 1; 4) = 2 ln 2 − 2 − q m q 2(q 2 )2 If we are interested in the opposite limit, q 2 /m2 → 0, the natural idea is to close the integration contour to the left and take residues at the points z = −1, −2, . . . to obtain iπ 2 (q 2 )2 q2 + + ... . (4.103) F4.1 (2, 1; 4) = − 2 1 + m 2m2 3m4 Consider now Example 4.3, where IR and collinear divergences are present. We can use MB representation (4.11) for expanding Feynman integrals with various indices in the two diﬀerent limits, t/s → 0 and s/t → 0. There is again the typical dependence of the ratio of t and s on z of the form (t/s)z . The procedure of using (4.11) to obtain an asymptotic expansion in the limit t/s → 0 is standard: to shift the integration contour to the right. For the integral with given indices al , the points where it is necessary to take (minus) residues are given by the right poles of the gamma functions, in our terminology: at z = 0, 1, 2, . . . and at z = 2 − max{a1 , a3 } − a2 − a4 − ε + n with n = 0, 1, 2, . . .. For example, for F (s, t; d) = F4.3 (s, t; 1, 1, 1, 1, d) represented by (4.12), these are the two series of residues at z = 0, 1, 2, . . . and z = −1−ε, −ε, 1−ε, . . . which reproduce the hard and collinear contributions, respectively, to the asymptotic expansion within expansion by regions – see Chap. 8 of [27]. We obtain Γ (1 + ε)Γ (−ε)2 t iπ d/2 ln + 2ψ(−ε) − ψ(1 + ε) + γE F (s, t; d) = Γ (−2ε) s(−t)1+ε s

4.8 MB Representation versus Expansion by Regions

103

Γ (ε)Γ (1 − ε)2 t − ln + 2ψ(1 − ε) − ψ(ε) − 1 + γE s2 (−t)ε s Γ (2 + ε)Γ (−1 − ε)2 + (−s)2+ε 3 Γ (ε − 1)Γ (2 − ε)2 (−t)1−ε t + ln + 2ψ(2 − ε) − ψ(ε − 1) − + γE 2s3 s 2 ! 2 Γ (3 + ε)Γ (−2 − ε) t + + ... . (4.104) (−s)3+ε To obtain the asymptotic expansion in the opposite limit, s/t → 0, one shifts the integration contour to the left and takes residues at the left poles at z = 2 − min{a2 , a4 } − n and at z = 2 − a − ε − n with n = 0, 1, 2, . . .. For F (s, t; d), these are the two series of residues at z = −1, −2, . . . and z = −2 − ε, −3 − ε, −4 − ε, . . .. One can check that the resulting expansion is nothing but (4.104) with the interchange s → t, t → s – this should be the case because of the symmetry of the initial integral. In these two examples, terms of asymptotic expansions were obtained as residues in onefold MB integrals. As a non-trivial example with a multiple MB integration let us turn again to Example 4.8 of massless on-shell double boxes. Let us evaluate the leading asymptotic behaviour of the K(s, t; 1, . . . , 1, 0, ε) in the Regge limit, t/s → 0, using representation (4.48). The starting point of the evaluation of this quantity in expansion in ε was the analysis of gluing of right and left poles which showed the way how the poles in ε are generated. Now, our starting point is to look at the integration over the variable z1 which enters as the power of the ratio t/s and try to understand what right poles with respect to z1 are. One source of such poles is obvious: this is Γ (−z1 ) corresponding to the hard part within expansion by regions – see Chap. 8 of [27]. This part, however, starts only with order (t/s)0 which is subleading, as we will see shortly. Other sources are not visible at once, similarly to the poles in ε. However, the experience obtained in our previous examples when analysing the singular behaviour in ε shows how the poles in z1 appear after integrating over z2 , z3 and z4 . Let us use the rule formulated in Sect. 4.2 and systematically applied in our examples and analyse the integrand of (4.48) from the point of view of generating right poles in z1 . Apart from Γ (−z1 ), there are only two gamma functions that can generate a singularity of the type Γ (. . . − z1 ): Γ (−1 − ε − z1 − z2 ) and Γ (−1 − ε − z1 − z3 ) . Indeed, the singularity of the type Γ (−1 − ε − z1 ) is generated, due to the integration over z2 , because of the presence of Γ (−ε + z2 ), and, due to the integration over z3 , because of the presence of Γ (−ε+z3 ). Thus, to reveal this singularity, we can take a residue at the ﬁrst pole of Γ (−ε+z2 ) or Γ (−ε+z3 ). Therefore, we start with the same decomposition F = F11 +F10 +F01 +F00 as in Sect. 4.4. Now, in F11 represented by (4.49) and in F01 represented by

104

4 Evaluating by MB Representation

(4.50), the function Γ (−1−2ε−z1 ) is already explicitly present. The term F00 does not contribute now because it cannot generate the leading asymptotic behaviour in the given limit. To evaluate the leading asymptotics, let us, ﬁrst, consider F11 and take (minus) residue at z1 = −1 − 2ε to obtain +i∞ Γ (1 − ε − z4 )Γ (−2ε − z4 )2 Γ (1 + 2ε) 1 dz f11 = 4 x1+2ε 2πi −i∞ Γ (1 + ε + z4 )Γ (−4ε − z4 ) ×Γ (ε + z4 )2 Γ (z4 ) [2γE + ln x + ψ(−2ε) − ψ(1 + 2ε) − ψ(−4ε − z4 ) +ψ(−2ε − z4 ) + ψ(1 − ε − z4 ) + ψ(z4 )] . (4.105) Observe that this quantity is nothing but the contribution F111 that we have met in Sect. 4.4. It was evaluated in expansion in ε by taking residues at z4 = 0 and z4 = ε and shifting the integration contour over z4 . Starting from F01 and taking (minus) residue at z1 = −1 − 2ε we obtain +i∞ +i∞ Γ (1 + 2ε) 1 dz2 dz4 Γ ∗ (−ε + z2 )Γ (ε − z2 ) f01 = − x1+2ε (2πi)2 −i∞ −i∞ Γ (ε + z4 )Γ (−2ε − z4 )Γ (1 − ε − z4 ) × Γ (−4ε − z4 ) Γ (z2 + z4 )Γ (−ε + z2 + z4 )Γ (−ε − z2 − z4 ) . (4.106) × Γ (1 + z2 + z4 ) where the asterisk denotes, as in Appendix D, the opposite nature of the ﬁrst pole of Γ (−ε + z2 ). Now we observe that this is nothing but the contribution F011 of Sect. 4.4, where it was explained how it can be evaluated in expansion in ε. Summing up results for F111 and F011 we reproduce the leading part of (4.54), e.g. the terms of order 1/t modulo logarithms. So, we see that the evaluation of the leading asymptotic behaviour in the Regge limit, using MB representation, is a (simple) part of the global evaluation. Observe that the evaluation of the triple box in Example 4.11 is also organized in such a way that the leading Regge asymptotics can be extracted from this evaluation. On the other hand, it was also evaluated using expansion by regions [28]. It is not clear in advance which way is simpler: expanding by MB representation, or, by regions. My experience tells me that, usually, expanding by regions is certainly preferable, but sometimes, it looks more convenient to derive an appropriate MB representation and proceed as described in this section. But I can imagine that, sometimes, this is just a matter of taste. In complicated situations, the two strategies can successfully be combined. In particular, extracting the leading asymptotic behaviour from a general MB representation can show what kind of contributions one gets and will help detecting all regions which contribute. There are a lot of papers where the asymptotic behaviour was evaluated using MB representations – see, e.g., [16].

4.9 Conclusion

105

4.9 Conclusion Mellin integrals were used for the evaluation of Feynman integrals in various ways. For example, in [35], the ﬁrst analytical result for massless double boxes of Fig. 4.7 was obtained in the case where all the external legs are oﬀ-shell so that these are functions depending on many variables, s, t and p2i for i = 1, 2, 3, 4. Nevertheless it was possible to evaluate the double box for all powers of the propagators equal to one exactly in four dimensions. The following nice mathematical result was obtained: 2 2 iπ C(p21 p24 , p22 p23 , st) , s2 t where 1 C(x1 , x2 , x3 ) = (6 [Li4 (−ρx) + Li4 (−ρy)] λ y y 1 +3 ln [Li3 (−ρx) − Li3 (−ρy)] + ln2 [Li2 (−ρx) + Li2 (−ρy)] x 2 x 1 2 π2 2 y 7π 4 π2 2 + ln (ρx) ln (ρy) + ln(ρx) ln(ρy) + ln + , (4.107) 4 2 12 x 60 (1 − x − y)2 − 4xy , 2 , ρ ≡ ρ(x, y) = 1 − x − y + λ(x, y)

λ ≡ λ(x, y) =

(4.108) (4.109)

and x = x1 /x3 , y = x2 /x3 . Moreover, a similar analytical result was obtained [36] also for a general oﬀ-shell h-loop ladder planar diagram, in particular, for the oﬀ-shell triple box.7 In [37], an oﬀ-shell result for the non-planar two-loop three-point diagram was also obtained using the MB representation. Other examples of results obtained by this technique are analytical expressions for n-point oneloop massive Feynman integrals for general d [10]. Let me summarize the basic features that distinguish the technique of MB representation presented in this chapter and oriented at the evaluation in ε-expansion from other approaches based on Mellin integrals. – An appropriate multiple MB representation for a given class of integrals is derived for general powers of the propagators and irreducible numerators. 7 Well, one can hardly expect that explicit analytical results can be obtained for other (even double-box) Feynman integrals of this purely oﬀ-shell class, in particular, with a double power of some propagator, with some irreducible numerator, or where one of the lines other than rungs is contracted to a point. The possibility to obtain such a nice mathematical result for such a complicated object depending on so many variables in the case of all indices equal to one was later understood by making an interesting mathematical link with some problem of conformal quantum mechanics – see [17].

106

– –

–

–

–

–

–

4 Evaluating by MB Representation

In order to achieve the minimal number of MB integrations it is recommended to derive an MB representation for a sub-loop integral, insert it in the given integral over the loop momenta, etc. There is always the possibility to check multiple MB representations, which are sometimes rather cumbersome, by using simple partial cases. Multiple MB integrals are very ﬂexible for the resolution of the singularities in ε. This procedure reduces to shifting contours, in an appropriate way, and taking corresponding residues. After the resolution of the singularities in ε, at least some of the integrations can be performed explicitly by tabulated formulae of Appendix D, with results in terms of gamma and psi functions. One can usually have an easy numerical control on ﬁnite (in ε) MB integrals: it is enough to integrate from −5i to +5i along the imaginary axis to have a very good accuracy. When the integration in multiple MB integrals is hardly performed explicitly, one can convert them into multiple series and apply such packages as SUMMER [39] for summation. Onefold MB integrals can be summed up by closing the integration contour and summing up corresponding residues. Here one can apply summation formulae of Appendix C and/or SUMMER. All the manipulations with MB integrals can be done on a computer. (For example, I use MATHEMATICA for this.)

The technique of multiple MB representations is not always optimal. This holds at least for non-planar double boxes with one leg oﬀ-shell. Although ﬁrst analytical results were obtained with its help [24, 25] the adequate technique here turned out to be the method of diﬀerential equations which will be studied in Chap. 7.

References 1. C. Anastasiou, Z. Bern, L.J. Dixon and D.A. Kosower, Phys. Rev. Lett. 91 (2003) 251602. 97 2. C. Anastasiou, J.B. Tausk and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 262. 64, 74, 75, 81 3. W. Beenakker and A. Denner, Nucl. Phys. B 338 (1990) 349. 68 4. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 55, 78, 87, 96, 102 5. M.C. Berg`ere and Y.-M.P. Lam, Commun. Math. Phys. 39 (1974) 1. 55 6. Z. Bern, J.S. Rozowsky and B. Yan, Phys. Lett. B 401 (1997) 273. 97 7. T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 88, 89, 96 8. E.E. Boos and A.I. Davydychev, Theor. Math. Phys. 89 (1991) 1052, [Teor. Mat. Fiz. 89 (1991) 56]. 56, 91 9. K.G. Chetyrkin, S.G. Gorishnii, S.A. Larin and F.V. Tkachov, Phys. Lett. B 132 (1983) 351. 100 10. A.I. Davydychev, J. Math. Phys. 32 (1991) 1052; 33 (1992) 358. 105

References 11. 12. 13. 14. 15. 16.

17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39.

107

A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 699 (2004) 3. 69, 84, 91, 173 A. Devoto and D.W. Duke, Riv. Nuovo Cim. 7, No. 6 (1984) 1. 62, 78 J. Fleischer, M.Yu. Kalmykov and A.V. Kotikov, Phys. Lett. B 462 (1999) 169. 91, 173 J. Fleischer, A.V. Kotikov and O.L. Veretin, Nucl. Phys. B 547 (1999) 343. 62, 77, 84 R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542. 73 C. Greub, T. Hurth and D. Wyler, Phys. Rev. D 54 (1996) 3350; C. Greub and P. Liniger, Phys. Rev. D 63 (2001) 054025; H.H. Asatryan, H.M. Asatrian, C. Greub and M. Walker, Phys. Rev. D 65 (2002) 074004; K. Bieri, C. Greub and M. Steinhauser, Phys. Rev. D 67 (2003) 114019. 104 A.P. Isaev, Nucl. Phys. B 662 (2003) 461. 105 D.I. Kazakov, Theor. Math. Phys. 62, 84 (1985) [Teor. Mat. Fiz. 62, 127 (1984)]. 100, 101 H. Kleinert, J. Neu, V. Schulte-Frohlinde, K.G. Chetyrkin and S.A. Larin, Phys. Lett. B 272, 39 (1991) [Erratum-ibid. B 319, 545 (1993)]. 100 K.S. K¨ olbig, J.A. Mignaco and E. Remiddi, BIT 10 (1970) 38; K.S. K¨ olbig, Math. Comp. 39 (1982) 647. 62, 78 L. Lewin, Polylogarithms and Associated Functions (North-Holland, Amsterdam, 1981). 78 E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725. 88 V.A. Smirnov, Phys. Lett. B 460 (1999) 397. 74, 77 V.A. Smirnov, Phys. Lett. B 491 (2000) 130. 106 V.A. Smirnov, Phys. Lett. B 500 (2001) 330. 106 V.A. Smirnov, Phys. Lett. B 524 (2002) 129. 82, 83, 85, 87 V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 55, 67, 78, 87, 96, 102, 103 V.A. Smirnov, Phys. Lett. B 547 (2002) 239. 96, 104 V.A. Smirnov, Phys. Lett. B 567 (2003) 193. 92, 94 V.A. Smirnov and E.R. Rakhmetov, Teor. Mat. Fiz. 120 (1999) 64; V.A. Smirnov, Phys. Lett. B 465 (1999) 226. 78, 102 V.A. Smirnov, hep-ph/0406052; G. Heinrich and V.A. Smirnov, hepph/0406053. 88, 89, 90 V.A. Smirnov and O.L. Veretin, Nucl. Phys. B 566 (2000) 469. 78, 80 J.B. Tausk, Phys. Lett. B 469 (1999) 225. 64, 74, 81 N.I. Ussyukina, Teor. Mat. Fiz. 22 (1975) 300. 55 N.I. Ussyukina and A.I. Davydychev, Phys. Lett. B 298 (1993) 363. 74, 105 N.I. Ussyukina and A.I. Davydychev, Phys. Lett. B 305 (1993) 136. 105 N.I. Ussyukina and A.I. Davydychev, Phys. Lett. B 332 (1994) 159. 105 J.A.M. Vermaseren, Symbolic Manipulation with FORM (CAN, Amsterdam, 1991). 62 J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037. 62, 101, 106

5 IBP and Reduction to Master Integrals

The next method in our list is based on integration by parts1 (IBP) [15] within dimensional regularization, i.e. property (2.38). The idea is to write down various equations (2.38) for integrals of derivatives with respect to loop momenta and use this set of relations between Feynman integrals in order to solve the reduction problem, i.e. to ﬁnd out how a general Feynman integral of the given class can be expressed linearly in terms of some master integrals. In contrast to the evaluation of the master integrals, which is performed, at a suﬃciently high level of complexity, in a Laurent expansion in ε, the reduction problem is solved at general d, and the expansion in ε does not provide simpliﬁcations here. The reduction can be stopped whenever one arrives at suﬃciently simple integrals. On the other hand, one could try to solve the reduction problem in the ultimate mathematical sense, i.e. to reduce a given integral to true irreducible integrals which cannot be reduced further. To illustrate the procedure of solving IBP relations we shall begin in Sect. 5.1 with very simple one-loop examples. Usually, we shall indeed stop the reduction if we obtain integrals that can be expressed in terms of gamma functions for general values of the parameter of dimensional regularization, d. In Sect. 5.2, we shall proceed in two loops. We shall also study some general tricks within the method of IBP such as the triangle rule and shifting dimension. One of the two-loop examples, the reduction of massless on-shell double boxes, will be considered separately in Sect. 5.3. We shall conclude in Sect. 5.4 with brief bibliographic remarks and a description of attempts of making systematic the procedure of solving IBP recurrence relations.

5.1 One-Loop Examples The ﬁrst example is very simple: Example 5.1. One-loop vacuum massive Feynman integrals 1 For one loop, IBP was used in [34]. The crucial step – an appropriate modiﬁcation of the integrand before diﬀerentiation, with an application at the two-loop level (to massless propagator diagrams) – was taken in [15] and, in a coordinate-space approach, in [51]. The case of three-loop massless propagators was treated in [15].

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 109–132 (2004) c Springer-Verlag Berlin Heidelberg 2004

110

5 IBP and Reduction to Master Integrals

F5.1 (a) =

dd k . (k 2 − m2 )a

(5.1)

In this chapter, we are concentrating on the dependence of Feynman integrals on the powers of the propagators so that we will usually omit dependence on dimension, masses and external momenta. Let us forget that we know the explicit result (A.1) and try to exploit information following from IBP. Let us use the IBP identity 1 ∂ =0, (5.2) dd k ·k 2 ∂k (k − m2 )a with (∂/(∂k))·k = (∂/(∂kµ ))kµ , and write down resulting quantities in terms of integrals (5.1). We obtain (d − 2a)F (a) − 2am2 F (a + 1) = 0 .

(5.3)

This gives the following recurrence relation: F (a) =

d − 2a + 2 F (a − 1) . 2(a − 1)m2

(5.4)

We see that any Feynman integral with integer a > 1 can be expressed recursively in terms of one integral F (1) ≡ I1 which we therefore consider as a master integral. (Observe that all the integrals with non-positive integer indices are zero since they are massless tadpoles.) This can be done explicitly here: (−1)a (1 − d/2)a−1 I1 , (5.5) F (a) = (a − 1)!(m2 )a−1 where (x)a is the Pochhammer symbol and the only master integral is I1 = −iπ d/2 Γ (1 − d/2)(m2 )d/2−1 .

(5.6)

As in Chap. 3 let us consider Example 5.2. Massless one-loop propagator Feynman integrals dd k F5.2 (a1 , a2 ) = . (k 2 )a1 [(q − k)2 ]a2

(5.7)

(As we have agreed, the dependence on q 2 and d is omitted.) For integer powers of the propagators, these integrals are zero whenever one of the indices is non-positive. Let us forget the explicit result (3.6) and try to apply the IBP identity 1 ∂ =0. (5.8) dd k ·k 2 a1 ∂k (k ) [(q − k)2 ]a2 We recognize diﬀerent terms resulting from the diﬀerentiation as integrals (5.7) and obtain the following relation

5.1 One-Loop Examples

d − 2a1 − a2 − a2 2+ (1− − q 2 ) = 0

111

(5.9)

which is understood as applied to the general integral F (a1 , a2 ) with the standard notation for increasing and lowering operators, e.g. 2+ 1− F (a1 , a2 ) = F (a1 − 1, a2 + 1). We rewrite it as a2 q 2 2+ = a2 1− 2+ + 2a1 + a2 − d

(5.10)

and obtain the possibility to reduce the sum of the indices a1 + a2 . Explicitly, applying (5.10) to the general integral and shifting the index a2 , we have 1 [(d − 2a1 − a2 + 1)F (a1 , a2 − 1) (a2 − 1)q 2 −(a2 − 1)F (a1 − 1, a2 )] ,

F (a1 , a2 ) = −

(5.11)

Indeed, a1 + a2 on the right-hand side is less by one than on the left-hand side. This relation can be applied, however, only when a2 > 1. Suppose now that a2 = 1. Then we use the symmetry property F (a1 , a2 ) = F (a2 , a1 ) and apply (5.11) interchanging a1 and a2 and setting a2 = 1: F (a1 , 1) = −

d − a1 − 1 F (a1 − 1, 1) . (a1 − 1)q 2

(5.12)

This relation enables us to reduce the index a1 to one and we see that the two relations (5.11) and (5.12) provide the possibility to express any integral of the given family in terms of the only master integral I1 = F (1, 1) given by (3.8), i.e. F (a1 , a2 ) = c(a1 , a2 )I1 , and the corresponding coeﬃcient function c(a1 , a2 ) is constructed as a rational function of d. Let us now complete the analysis for the example considered in the introduction, i.e. once again consider our favourite example: Example 5.3. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. We stopped in Chap. 1 at the point where we were able to express any integral (1.2) in terms of the master integral I1 = F (1, 1) and integrals with a2 ≤ 0 which can be evaluated for general d in terms of gamma functions by means of (A.3). Let us now try to understand what the true master integrals are. They should be really irreducible, i.e. they cannot be expressed linearly in terms of other integrals. Suppose that a2 ≤ 0, Then we can apply (1.11) to reduce a1 to one. In the case a1 = 1, we use relation (1.11) multiplied by 2− to express the term 2m2 a1 1+ 2− in (1.13). Thus, we obtain the following relation (d − a2 − 1)2− = (q 2 − m2 )2 a2 2+ + (q 2 + m2 )(d − 2a2 − 1)

(5.13)

that can be used to increase the index a2 to zero or one starting from negative values. We come to the conclusion that there are two irreducible integrals I1 = F (1, 1) given by (1.7) and I2 = F (1, 0) which equals the right-hand side of (5.6), and any integral from our family can be expressed linearly in terms

112

5 IBP and Reduction to Master Integrals

of them. This reduction procedure to I1 and I2 can easily be implemented on a computer. Observe that the integrals I1 and I2 cannot be linearly expressed through each other because, at general d, I1 is a non-trivial function of q 2 and m2 while I2 is independent of q 2 . This was the last example in this chapter, where we solve the reduction problem in the maximal way, i.e. in the sense of reduction to irreducible integrals. In the rest of the examples, we shall not be so curious and will stop the reduction whenever we arrive at suﬃciently simple classes of integrals. In Chap. 6, however, the reduction will be performed in the ultimate sense. Some other approaches with this property will be characterized in Sect. 5.4. The next example is again our old one. Example 5.4. The triangle diagrams of Fig. 3.4 given by (3.19). Writing down IBP relations with p1,2 · (∂/(∂k)) and (∂/(∂k)) · k we obtain the following three equations: a3 − a1 + a1 1+ (3− + m2 ) − a2 2+ (1− − 3− + Q2 − m2 ) −a3 3+ (1− − m2 ) = 0 , −

−

−

a3 − a2 + a2 2 (3 + m ) − a1 1 (2 − 3 + Q − m ) −a3 3+ (2− − m2 ) = 0 , +

2

+

−

2

d − a1 − a2 − 2a3 − (a1 1 + a2 2 )(3 + m ) − 2m a3 3 = 0 , +

+

2

(5.14)

2

2

+

(5.15) (5.16)

where Q2 = −q 2 = −(p1 − p2 )2 . Let us observe that the integrals (3.19) can be evaluated in terms of gamma functions if at least one of the indices is non-positive. In the case of a1 ≤ 0 or a2 ≤ 0, we can apply (A.6) and, in the case of a3 ≤ 0, we can apply (A.12). Let us now assume that all the indices are positive. Let us apply (5.14)–(5.16) to the general integral F (a1 , a2 , a3 ) and solve the corresponding linear system of the three equations with respect to F (a1 + 1, a2 , a3 ), F (a1 , a2 + 1, a3 ) and F (a1 , a2 , a3 + 1). We shall obtain an expression of these quantities in terms of integrals with the sum of the indices equal to a1 + a2 + a3 . Using the ﬁrst part of this solution we obtain a relation that expresses F (a1 , a2 , a3 ) in terms of integrals with a1 less by one and can be used in the case a1 > 1. Similarly, the second and the third parts of the solution give the possibility to reduce a2 > 1 and a3 > 1 to one. Therefore, we see that any given Feynman integral (3.19) can be reduced to I1 = F (1, 1, 1) and a family of simple integrals which can be expressed in terms of gamma functions. For example, we have F (1, 1, 2) =

(d − 4)(2m2 − Q2 ) I1 2m2 (m2 − Q2 )

2 1 Q (F (1, 2, 0) + F (2, 1, 0)) + 2 2 2 2m (m − Q )

−m2 (F (0, 1, 2) + F (0, 2, 1) + F (1, 0, 2) + F (2, 0, 1)) , (5.17)

5.1 One-Loop Examples

113

where all the integrals with an index equal to zero can be evaluated simply by (A.4) and (A.7) Observe that the coeﬃcient at I1 in (5.17) is proportional to ε. According to [18], where the reduction in the massless case was performed and in the case of general masses analysed, this is a general phenomenon, i.e. this property holds for any F (a1 , a2 , a3 ) with a1 + a2 + a3 > 3 in the case of general masses ml and indices. As a result, such integrals involve only elementary functions (no polylogarithms) in the expansion in ε up to the ﬁnite part – this was noticed very much time ago [35]. Let us again consider the massless on-shell boxes which we analysed in Examples 3.3 and 4.3. For convenience, we change the numbering of the lines as compared with Chaps. 3 and 4. Example 5.5. The massless on-shell box Feynman integrals of Fig. 5.1 with p2i = 0, i = 1, 2, 3, 4 and general integer powers of the propagators.

p1 3

p2

p3

1 4 2

p4

Fig. 5.1. Box diagram

Let us ﬁrst observe that whenever one of the indices is non-positive, the integrals can be evaluated in terms of gamma functions for general ε. In particular, if some index is zero, e.g., a4 = 0, one can apply (A.28). Suppose now that all the indices are positive. Starting from the IBP identity with the operator (∂/∂k)·k acting on the integrand and choosing the loop momentum k to be the momentum of each of the four lines, we obtain the following four IBP relations: a1 s1+ = a1 + 2a2 + a3 + a4 − d + (a1 1+ + a3 3+ + a4 4+ )2− = 0 , (5.18) a2 s2+ = 2a1 + a2 + a3 + a4 − d + (a2 2+ + a3 3+ + a4 4+ )1− = 0 , (5.19) a3 t3+ = a1 + a2 + a3 + 2a4 − d + (a1 1+ + a2 2+ + a3 3+ )4− = 0 , (5.20) a4 t4+ = a1 + a2 + 2a3 + a4 − d + (a1 1+ + a2 2+ + a4 4+ )3− = 0 , (5.21) where s = (p1 + p2 )2 and t = (p1 + p3 )2 are Mandelstam variables, as above. These equations can be used to reduce the indices al to one. For example, when (5.18) is applied to the general integral, we have, on the right hand side, terms with a1 less by one, with the exception of one term corresponding to a1 1+ 2− . This term, however, decreases a2 . Anyway, the sum of the indices corresponding to the right-hand side of (5.18)–(5.21) is less by one than corresponding to the left-hand side.

114

5 IBP and Reduction to Master Integrals

Therefore we come to the conclusion that any given Feynman integral F5.5 (a1 , a2 , a3 , a4 ) can be expressed linearly in terms of the master integral I1 ≡ F5.5 (1, 1, 1, 1) and a family of integrals where some indices are nonpositive. We again stop reduction here and do not try to reduce various integrals with non-positive indices to true master integrals. In Chap. 6, however, we will see what these true master integrals are.

5.2 Two-Loop Examples Let us now see how IBP relations can be used for the reduction of the massless Feynman integrals corresponding to Fig. 3.9. We have already considered these diagrams in Example 3.5 in Chap. 3. Example 5.6. Two-loop massless propagator Feynman integrals of Fig. 3.9 with integer powers of the propagators. First, we observe that if a5 = 0 the integrals over k and l decouple and can be evaluated in terms of gamma functions by use of (3.6): 2 F5.6 (a1 , a2 , a3 , a4 , 0) = (−1)a1 +a2 +a3 +a4 iπ d/2 ×

G(a1 , a2 )G(a3 , a4 ) . (−q 2 )a1 +a2 +a3 +a4 +2ε−4

(5.22)

When some other index al is zero, the integral becomes recursively oneloop (see Sect. 3.2.1), i.e. it can be evaluated in terms of gamma functions by successively applying the same one-loop formula, for example, 2 F5.6 (a1 , a2 , a3 , 0, a5 ) = (−1)a1 +a2 +a3 +a5 iπ d/2 ×

G(a3 , a5 )G(a2 , a1 + a3 + ε − 2) . (−q 2 )a1 +a2 +a3 +a5 +2ε−4

(5.23)

Suppose now that all the indices are positive integers. Let us write down the following IBP identity: kµ − lµ ∂ dd k dd l =0. (5.24) (l2 )a3 [(q − l)2 ]a4 ∂kµ (k 2 )a1 [(q − k)2 ]a2 [(k − l)2 ]a5 Taking derivatives, using identities such as 2k ·(k − l) = k 2 + (k − l)2 − l2 , and recognizing terms on the left-hand side as integrals (3.39), we arrive at the following relation: (5.25) (a1 + a2 + 2a5 − d) − a1 1+ 3− − 5− − a2 2+ 4− − 5− = 0 . Equation (5.25) can be used as a recurrence relation for the given family of integrals. Indeed, applying it to the general integral, we obtain

5.2 Two-Loop Examples

115

1 a1 + a2 + 2a5 − d × [a1 (F5.3 (a1 + 1, a2 , a3 − 1, a4 , a5 ) − F5.3 (a1 + 1, a2 , a3 , a4 , a5 − 1)) +{1 ↔ 2, 3 ↔ 4}] . (5.26)

F5.6 (a1 , a2 , a3 , a4 , a5 ) =

On the right-hand side, we encounter integrals where the sum a3 + a4 + a5 is less by one than that on the left-hand side. Thus, successive application of this relation reduces any given integral to integrals with some index equal to zero, where (5.22) and (5.23) can be used. In fact, in case one of the indices is negative, generalizations of the explicit formulae (5.22) and (5.23) can be derived. To do this, one applies (A.12). Therefore we come to the conclusion that any given integral (3.39) with integer indices can be evaluated in terms of gamma functions for general values of d. If we are not too curious we can stop our analysis at this point and not bother about the minimal number of master integrals. We could consider any integral with a non-positive index as a master integral because they can be expressed explicitly in terms of gamma functions. Otherwise it is necessary to continue to exploit IBP relations and obtain a solution of the reduction problem in the strict sense, i.e. with a minimal family of the master integrals. Usually, people are lazy in such situations and indeed stop the reduction. In this particular example, we shall see, in Chap. 6, what the true master integrals are. For example, the integral with all indices equal to one, is evaluated by means of (5.26) as follows: 1 [F5.6 (2, 1, 0, 1, 1) − F5.6 (2, 1, 1, 1, 0)] ε 1 1 = G(1, 1) [G(2, 1) − G(2, 1 + ε)] ε (−q 2 )1+2ε 2 2 4 iπ π + 12ζ(3) ε = 6ζ(3) + q2 10 4 π 2 + + (24 − π )ζ(3) + 42ζ(5) ε2 + . . . , 5

F5.6 (1, 1, 1, 1, 1) =

(5.27)

so that the well-known result [14, 42] at order ε0 is again (as in Sect. 3.5) reproduced. In this simple example, it was suﬃcient to use only one IBP relation which, in fact, follows from an IBP identity for the triangle diagram of Fig. 5.2 with general indices, m3 = 0 and general masses m1 and m2 . The general Feynman integral for this graph is dd k F (a1 , a2 , a3 ) = . (5.28) [(k + p1 )2 − m21 ]a1 [(k + p2 )2 − m22 ]a2 (k 2 )a3 Let us write down the IBP identity with the operator (∂/∂k) · k acting on the integrand of (5.28). Then we obtain the following ‘triangle’ rule:

116

5 IBP and Reduction to Master Integrals

Fig. 5.2. Triangle diagram with general integer indices

1=

1 d − a1 − a2 − 2a3

× a1 1+ 3− − (p21 − m21 ) + a2 2+ 3− − (p22 − m22 ) .

(5.29)

This identity can be applied to a triangle as a subgraph in a bigger graph. Suppose that the external upper right line in Fig. 5.2 has the mass m1 and the external lower right line has the mass m2 but these are internal lines for the bigger graph. Then the factors (p21 − m21 ) and (p22 − m22 ) eﬀectively reduce the indices of the corresponding lines (with the momenta p1 and p2 ) by one. For example, if we consider the triangle rule in the massless case and apply it to the left triangle in Fig. 3.9 we shall obtain (5.25). The triangle rule derived above is very well known. Let us derive another triangle rule from it. Consider the case where (p1 −p2 )2 = 0 and m1 = m2 = 0. Starting from the IBP identity with the operator (∂/∂k) · k acting on the integrand and choosing the loop momentum k to be the momentum of each of the three lines, we obtain the following three IBP relations: d − 2a1 − a2 − a3 − a2 2+ 1− − a3 3+ (1− − p21 ) = 0 , + −

−

d − a1 − 2a2 − a3 − a1 1 2 − a3 3 (2 0, d − a1 − a2 − 2a3 − a1 1+ (3− − p21 ) − a2 2 (3 − p22 ) = 0 . +

− p22 ) = + −

(5.30) (5.31) (5.32)

We form the combination (5.30) times a1 1+ plus (5.31) times a2 2+ minus (5.32) times a3 3+ and arrive at the following extra triangle relation: (d − 2a3 − 2)a3 3+ = (d − 2a1 − 2a2 − 2)(a1 1+ + a2 2+ ) .

(5.33)

There was a subtle point when multiplying quantities like 3+ and a3 which have algebraic properties similar to creation and annihilation operators. For example, the additional terms −2 in the brackets of (5.33) appear due to this multiplication. Consider now Example 5.7. Planar two-loop massless vertex diagrams with p21 = p22 = 0 and general integer powers of the propagators. The general scalar Feynman integral corresponding to Fig. 5.3 can be written as

5.2 Two-Loop Examples

117

p1 3 q

1 6

5

2 4 p2 Fig. 5.3. Planar vertex diagram

dd l (l2 )−a7 − 2p1 ·l)a1 (l2 − 2p2 ·l)a2 dd k × , (k 2 − 2p1 ·k)a3 (k 2 − 2p2 ·k)a4 (k 2 )a5 [(k − l)2 ]a6

F5.7 (a1 , . . . , a7 ) =

(l2

(5.34)

where k and l are loop momenta of the box and triangle subgraphs, respectively. There is one irreducible numerator, which cannot be expressed linearly in terms of the factors in the denominator, chosen as l2 . We are interested only in non-positive values of a7 . As it was mentioned in Chap. 3, the evaluation of such Feynman integrals by Feynman parameters is rather cumbersome. It turns out that using IBP provides the possibility to reduce any integral of this family to very simple integrals. As we will see shortly, any given integral can be expressed in terms of gamma functions for general values of d. We shall not, however, write down various IBP relations for (5.34). As it was noticed in [36] it is enough to use just one tool, the triangle rule (5.29), for the evaluation of these integrals. Suppose that all the indices a1 , . . . , a6 are positive and a7 = 0. Let us apply (5.29) to the triangle subgraph, i.e. with the lines (1, 2, 6). We obtain

1 1= a1 1+ 6− − 3− + a2 2+ 6− − 4− (5.35) d − a1 − a2 − 2a6 as acting on F5.7 (a1 , . . . , a6 , 0). Since the sum a1 + a2 + a6 on the right-hand side of the corresponding relation is less by one, it provides the possibility to reduce one of the indices a4 , a5 , a6 to zero. In the case where a6 = 0 the Feynman integral factorizes and is evaluated by (A.7) and (A.28): 2 F5.7 (a1 , . . . , a5 , 0, 0) = (−1)a1 +...+a5 iπ d/2 ×

G(a1 , a2 )G3 (a3 , a4 , a5 ) . (−q 2 )a1 +...+a5 +2ε−4

(5.36)

where the function G3 is deﬁned as the coeﬃcient of the right-hand side of (A.28) at iπ d/2 (−q 2 )−λ1 −λ2 −λ3 −ε+2 .

118

5 IBP and Reduction to Master Integrals

Suppose now that a3 or a4 is zero. Let it be a4 so that the line 4 is reduced to a point. Then we apply (5.29) to the triangle subgraph, with the lines (5, 6, 3). We obtain

1 a5 5+ 3− + a6 6+ 3− − 1− (5.37) 1= d − a5 − a6 − 2a3 as acting on F5.7 (a1 , a2 , a3 , 0, a5 , a6 , 0). (There is one term less as compared with (5.35) because of the on-shell condition p21 = 0.) This relation provides the possibility to reduce either a1 or a3 to zero. In both cases, resulting integrals become recursively one-loop and can be evaluated again by (A.7) and (A.28). We have 2 F5.7 (0, a2 , a3 , 0, a5 , a6 , 0) = (−1)a2 +a3 +a5 +a6 iπ d/2 G(a2 , a6 )G3 (a3 , a2 + a6 + ε − 2, a5 ) (5.38) (−q 2 )a2 +a3 +a5 +a6 +2ε−4 2 F5.7 (a1 , a2 , 0, 0, a5 , a6 , 0) = (−1)a1 +a2 +a5 +a6 iπ d/2 ×

×

G(a5 , a6 )G3 (a1 , a2 , a5 + a6 + ε − 2) . (5.39) (−q 2 )a1 +a2 +a5 +a6 +2ε−4

Therefore, any integral with positive indices can be evaluated by this procedure. For example, we reproduce the well-known result [28, 36, 41] for F5.7 (1, . . . , 1, 0): (iπ d/2 )2 1 1 G2 (2, 2)G3 (2 + ε, 1, 1) (Q2 )2+2ε ε 2ε 1 G3 (2, 1, 1 + ε) + G3 (1, 1, 1) −G2 (2, 1) ε d/2 −γE ε 2 1 ) 5π 2 29ζ(3) 3π 4 (iπ e + + O(ε) . (5.40) + + = (Q2 )2+2ε 4ε4 24ε2 6ε 32 In fact, a similar reduction procedure can be developed for general Feynman integrals with an irreducible numerator, i.e. for a7 < 0, and with general integer indices (not only positive). This can be done by using generalizations of the triangle rule to the case with a numerator. In fact, a general recursive procedure for such integrals (and integrals with another oﬀ-shell external momentum, p21 = 0 instead of q 2 = 0) with general numerators was developed in [20], with boundary integrals written in terms of terminating hypergeometric series of the unit argument. Another possibility in this situation is to get rid of the numerator and negative indices using the technique of shifting dimension which we will discuss shortly. Then we shall come back to this point. We now turn, following [3], to the two classes of integrals already studied in Chap. 4 which are partial cases of massless on-shell double boxes: boxes with a one-loop insertion and boxes with a diagonal shown in Fig. 5.4. For

5.2 Two-Loop Examples 1

p1 3

p2

3

p3 4

2

5

119

7

p4

(a)

6

5

2

(b)

Fig. 5.4. (a) Box with a one-loop insertion. (b) Box with a diagonal

convenience, we again change the numbering of the lines: In Fig. 5.4a we adjust it to that of Fig. 5.1 and, in Fig. 5.4b, to a new numbering for the double box which will be studied in the next section. So, the next is Example 5.8. Reduction of boxes with a one-loop insertion. Let us, ﬁrst, assume that we are dealing with the boxes with a one-loop insertion without numerator, B5.8 (a1 , . . . , a5 ) (In the given case, there are two independent scalar products that cannot be linearly expressed in terms of the denominators of the propagators.) In fact, the integration in the oneloop insertion in Fig. 5.4a can be taken explicitly by (A.7) and, graphically, this insertion can be replaced by a line with the index a4 + a5 + ε − 2 – see Fig. 3.1. Therefore, the problem reduces to the boxes of Fig. 5.1 in the case where the index of the line 4 is not integer. Still if one of the ﬁrst three indices is non-positive we obtain a quantity evaluated in terms of gamma functions by (A.28). Suppose now that a1 , a2 , a3 > 0. Then we can apply (5.18) and (5.19) to reduce a1 and a2 to one, as in the case of the box with integer indices. To take care of a3 let us form the new relation as a4 4+ times (5.20) minus a3 3+ times (5.21): (d − a1233 )a3 3+ = (d − a1244 − 2)a4 4+ +(a3 − a4 )(a1 1+ + a2 2+ ) ,

(5.41)

where we keep our notation of Chap. 4, e.g. a1233 = a1 + a2 + 2a3 etc. Observe now that (5.41) can be used to reduce the index a3 to one because a1 1+ and a2 2+ in the last term can be replaced immediately according to (5.18) and (5.19). Let us therefore assume that a1 = a2 = a3 = 1. Now we can apply (5.21), where the term with a3 3− gives integrals expressed in terms of gamma functions, to have control on a4 = a4 + ε which has an amount proportional to ε because of the one-loop integration. For example, one can shift a4 to a4 = 0: this choice corresponds to I1 = B5.8 (1, . . . , 1). In the case with numerators, one can get rid of them by shifting indices and dimension [47], as outlined in Subsect. 3.2.3. Then the previous procedure

120

5 IBP and Reduction to Master Integrals

provides the possibility to express any given box, with dimension d shifted by a positive even number, in terms of the master box with a one loop insertion I1 (d + 2n) in the same dimension and a family of simpler integrals expressed in terms of gamma functions. To complete this reduction procedure we need to know how to express these integrals in terms of I1 (d). To do this, let us apply the general relation (−1)h − U(α1 , . . . , αL ) , (5.42) d = π + αl →ial l

where U given by (2.24) is one of the two basic functions present in the alpha representation (2.36). (The factors (−1)h and 1/π come from the overall coeﬃcient in (2.36).) In particular, for Fig. 5.4a, this gives 1

a4 a5 4+ 5+ + (a1 1+ + a2 2+ + a3 3+ )(a4 4+ + a5 5+ ) . d− = (5.43) π We have d− I1 (d + 2) = I1 (d). On the other hand, applying the right-hand side of (5.43) to I1 (d + 2) we obtain a linear combination of integrals in dimension d + 2 with shifted indices for which we can use the reduction procedure described above. As a result, we obtain a desired linear relation of the type I1 (d) = A(d)I1 (d + 2) + B(d) , where A(d) is a rational function (of d, s and t) and B(d) comes from various integrals with some zero indices and can be evaluated in terms of gamma functions. Thus, any integral I1 (d + 2n) can be expressed recursively in terms of the master integral I1 (d) and a collection of simpler integrals. This completes our reduction procedure. Let us remember about the vertex diagrams of Example 5.7 which we considered without numerator. Now, we can get rid of any numerator as described above and then apply our reduction procedure formulated for nonnegative indices. However, since the corresponding results are expressed in terms of gamma functions for general d, there is no problem to make any shift d → d + 2n in them. We shall consider the reduction of the boxes with a diagonal in the next section.

5.3 Reduction of On-Shell Massless Double Boxes Let us turn, following [45], to Example 5.9. Reduction of on-shell massless double boxes. Let us follow the strategy [47] characterized in Subsect. 3.2.3 that enables us to express any integral with a numerator as a linear combination of integrals with shifted indices and dimension d. So, let us deal with Fig. 5.5 and the corresponding Feynman integrals

5.3 Reduction of On-Shell Massless Double Boxes

p1

1 7

p2

p3

3 6

2

121

5 4

p4

Fig. 5.5. Double box

dd k dd l (k 2 + 2p1 ·k)a1 (k 2 − 2p2 ·k)a2 (l2 + 2p1 ·l)a3 1 × 2 . (5.44) a 4 (l − 2p2 ·l) [(l + p1 + p3 )2 ]a5 [(k − l)2 ]a6 (k 2 )a7

K(a1 , . . . , a7 , d) =

where all indices al are non-negative. For convenience, we have changed the routing of the external momenta as well as the numbering of the lines in order to take into account the symmetry of the graph. (In Chap. 4, the numbering was oriented at insertions of boxes into double boxes.) Let us ﬁrst analyse situations, where one of the indices is zero. For a6 = 0, we obtain a product of two triangles which can be evaluated by (A.28) in terms of gamma functions. If a5 = 0 or a7 = 0 we obtain planar vertex diagrams analysed in Example 5.7. They are all evaluated in terms of gamma functions. Consider now the four symmetrical cases, where one of the other four indices is zero. Let it be a4 ; graphically, this means that the line 4 is contracted to a point – see Fig. 5.5. In this reduced graph, we can apply the triangle rule (5.29) to the resulting triangle with the lines 5, 6 and 3. After that we reduce either a3 or a1 to zero. Therefore, we arrive at a box with a one-loop insertion, in the former case, or a box with a diagonal, in the latter case – see Fig. 5.4. We conclude that, whenever one of the indices is zero, a given integral becomes a linear combination of the boxes with a one-loop insertion or a diagonal, or integrals expressed in terms of gamma functions. Let us call all these integrals boundary integrals. For the boxes with a oneloop insertion, we already know how to perform the reduction further, due to Example 5.8. Let us forget about this for a while and decide that all these boundary integrals are simple enough to stop the reduction here (as this was done in [45]). To perform the reduction for a given double box with positive indices, let ∂ ·(k − p2 ) which gives us start from the IBP relation with ∂k sa1 1+ = a7 7+ 2− + a6 6+ (2− − 4− ) + a1 1+ 2− −(d − 2a2 − a1 − a7 − a6 ) .

(5.45)

Three similar relations can be obtained from (5.45) by the two symmetry transformations: (1 ↔ 3, 2 ↔ 4, 5 ↔ 7) and (1 ↔ 2, 3 ↔ 4). The so-obtained four relations can be used to reduce the indices a1 , a2 , a3 , a4 to one. To reduce a5 to one we shall need one more IBP relation which is the ∂ · k times a5 5+ and the relation diﬀerence of the relation obtained with ∂k

122

5 IBP and Reduction to Master Integrals

obtained with

∂ ∂k ·(k

− l) times a6 6+ :

(d − a3455 − 2)a5 5+ = (d − a3466 2)a6 6+ + (a5 − a6 )(a3 3+ + a4 4+ ) +a3 a6 1− 3+ 6+ + a4 a6 2− 4+ 6+ . (5.46) The symmetrical relation applied to reduce a7 to one is (d − a1277 − 2)a7 7+ = (d − a1266 )a6 6+ + (a7 − a6 )(a1 1+ + a2 2+ ) +a1 a6 1+ 3− 6+ + a2 a6 2+ 4− 6+ .

(5.47)

Using the above recurrence relations we can bring the indices of the lines 1,2,3,4,5,7 all to one so that only a6 can now be greater than one. An appropriate relation for the reduction of a6 is [45] t(d − 6 − 2a6 )(a6 + 1)a6 6++ = t −(d − 5 − a6 ) 3d − 14 − 2a6 + 2a6 a6 6+ s 2 + (d − 4 − a6 )2 (d − 5 − a6 ) s % 2 t + (2+ + 7+ ) − (d − 4 − a6 )(d − 5 − a6 ) + 2 a26 6+ s s &

− 2t(a6 + 1)a6 6++ + 2(d − 4 − a6 )a6 6+ 3+ 1− +(d − 6)7− d− ,

(5.48)

−

where d is the operator that shifts dimension by −2, as before. This relation is valid only if it is applied to an integral with a1 = . . . = a5 = 1 and a7 = 1 (since some terms that are zero in this case are dropped out). The operator d− can be substituted explicitly using (5.42) with U = (α1 + α2 + α7 )(α3 + α4 + α5 ) +α6 (α1 + α2 + α3 + α4 + α5 + α7 ) ,

(5.49)

so that 1 + (1 + 2+ + 6+ )(3+ + 4+ + 5+ ) + 6+ (1+ + 2+ ) . (5.50) π The relation (5.48) can be derived as follows. Let us start with an integral with the numerator 2k · p2 . Since 2k · p2 = k 2 − (k 2 − 2p2 · k), such an integral is the diﬀerence of integrals where a7 or a2 is reduced by one. On the other hand, we can express this integral with the numerator in terms of integrals with shifted dimension and indices. Using an exponentiation of this numerator, similarly to how this is done for polynomials in the propagators (see (2.12)) and modifying the derivation of the alpha representation for the scalar double box in this case, we see (similarly to (3.16)) that the insertion of the numerator and shifting dimension by −2 can be described either by d− = −

5.3 Reduction of On-Shell Massless Double Boxes

123

the diﬀerence of the operators 7− − 2− times d− , or (up to a coeﬃcient with π) by the operator

(5.51) s a1 1+ (a6 6+ + a4 4+ + a5 5+ ) + a3 a6 3+ 6+ − ta5 a6 5+ 6+ . On the right-hand side of the so-obtained equation, we apply the reduction formulae (5.45)–(5.47) to reduce indices increased by the operators in (5.51). After some transformation, we then arrive at (5.48). Observe that on the left-hand side of (5.48) there is 6++ , rather than 6+ . This means that (5.48) enables us to reduce a6 to 1 or 2. Thus, after the application of the recurrence relations presented above, we reduce a given integral, up to our boundary integrals, to a linear combination of the two integrals, K1 (d) = K(1, 1, 1, 1, 1, 1, 1, d) and K2 (d) = K(1, 1, 1, 1, 1, 2, 1, d). However, these integrals generally appear, in the course of the reduction, in shifted dimensions so that we obtain the two families of integrals instead: K1 (d, n) = K1 (d + 2n) and K2 (d, n) = K2 (d + 2n) with K1 (d, 0) = K1 (d) and K2 (d, 0) = K2 (d). Of course, if we had results for general d for the master integrals (even expressed in terms of gamma functions), there would be no problem to shift the dimension in such analytical results. However, we are at a rather high level of complexity and are able to obtain results (at least for the master integrals) only in a Laurent expansion in ε, where expansions of the master integrals at d = 4 − 2ε and, say, at d = 6 − 2ε, when ε → 0, are not related to each other. To derive appropriate relations for the reduction of K1,2 (d, n) to K1,2 (d, 0), one can use the same trick with shifting dimension [47] as above, i.e. to write down equations K1,2 (d, n) = d− K1,2 (d, n + 1) with d− given by (5.50) and perform the reduction of the indices, which are increased after the action of d− , using (5.45)–(5.48). Solving the resulting linear system of equations one arrives at the following recurrence relations [45] which can be used to come back to dimension d = 4 − 2ε in the two master integrals: 1 ' (d) a22 K1 (d, n − 1) − f1 K1 (d, n) K1 (d, n) = ∆ $ (d)

, −a12 K2 (d, n − 1) − f2 K2 (d, n) ' 1 (d) K2 (d, n) = −a21 K1 (d, n − 1) − f1 K1 (d, n) ∆ $ (d)

+a11 (K2 (d, n − 1) − f2 K2 (d, n)) , (d)

where operators fj (d)

f1

(5.52)

(5.53)

are given by

2 + + (2 3 + 2+ 4+ + 2+ 6+ + 4+ 6+ + 4+ 7+ + 3+ 7+ ) s 2 4 + (2+ 5+ + 5+ 6+ + 5+ 7+ ) − 2 (d − 5)(3s + 2t)(2+ + 7+ ) s s t ! 2 2 + + + + + 3 6 7 − (3s(d − 5) + t(3d − 14)) 3 6 + 1− d−6 st(d − 6)

=

124

5 IBP and Reduction to Master Integrals

3 + 7− d− , t

(5.54)

2 + + 4 (2 3 + 2+ 4+ + 3+ 7+ + 4+ 7+ )6+ + (2+ + 4+ )6++ s s 1 2(2d − 13) + 4 + + + + + (2 + 7 + 2 6 )5 6 + +3 + 7+ 6++ s(d − 6) d−6 s 2(d − 5)(d − 7) (s(3d − 20) + 2t(d − 6)) (2+ + 7+ )6+ − 2 s t(d − 6)(d − 8) 2(d − 5)(d − 7) s (3s(3d − 20) + 4t(2d − 13)) 2+ + 7+ + 3+ 6+ + 2 2 s t (d − 8) d−6 ! 4 5d − 34 (3d − 20)(2d − 13) + + 3+ 6++ 1− d−8 s t(d − 6) ! 3d − 20 + d−7 6 − 2 (3s(3d − 20) + 4t(2d − 13)) 7− d− , (5.55) + t(d − 6) st (d − 8) (d)

f2

=

2 (d − 5)2 (3s + 2t), s2 t 3 2 = − (4d − 21) − (3d − 16), s t (d − 5)2 (d − 7) 8(2d − 13) 6(3d − 20) + , =− st(d − 8) s t d−7 2 3s (3d − 16)(3d − 20) + 6st(5d2 − 59d + 172) = 2 2 s t (d − 8) +4t2 (d − 5)(d − 6) ,

a11 =

(5.56)

a12

(5.57)

a21 a22

16(s + t)(d − 5) (d − 6)(d − 7) . s4 t(d − 8)

(5.58)

(5.59)

3

∆=

(5.60)

Thus, we are already able to reduce any double box to the two master integrals K1 (d) and K2 (d) and a family of our boundary integrals. For the ﬁrst master double box, K1 (d), we know the result given by (4.52) and (4.53), in expansion in ε, derived by MB representation in Chap. 4. To evaluate the second master double box, K2 (d), let us use alpha representation (2.36), where the function U is given by (5.49) and the second basic function (2.25) by V = [α1 α2 (α3 + α4 + α5 ) + α3 α4 (α1 + α2 + α7 ) +α6 (α1 + α3 )(α2 + α4 )] s + α5 α6 α7 t ,

(5.61)

We exploit this very simple dependence of this function on t to derive the following two relations by diﬀerentiating in t and implementing the factor α5 α6 α7 /U by shifting indices and dimension:

5.3 Reduction of On-Shell Massless Double Boxes

125

∂ 1 K(s, t; 1, . . . , 1, d) = − K(s, t; 1, 1, 1, 1, 2, 2, 2, d + 2) , (5.62) ∂t π 2 ∂ K(s, t; 1, 1, 1, 1, 1, 2, 1, d) = − K(s, t; 1, 1, 1, 1, 2, 3, 2, d + 2) . (5.63) ∂t π Then we apply the reduction procedure described above and express the righthand side of these equations in terms of the two master double boxes and a family of our boundary integrals (around ﬁfty terms in each case). In fact, the boundary integrals are simple enough here: a simple procedure based on the onefold MB representations (4.55) and (4.56) (see comments after these formulae) implemented on a computer can provide their ε-expansions up to order ε2 which is necessary here because the boundary integrals sometimes enter with coeﬃcients involving 1/ε2 . Then we insert (4.53) into (5.62) and use this equation to obtain a similar result for the second master double box. 2

K(1, 1, 1, 1, 1, 2, 1, d) =

(ie−γE ε ) f2 (t/s, ε) (−s)2+2ε t2

(5.64)

with

1 4 1 5 2 2 f2 (x, ε) = 4 − 5 (ln x − 2) 3 + 2 ln x − 14 ln x − (π + 4) 2 ε ε 2 ε 1 2 3 11 65 ln x + 8 ln2 x + π 2 + 14 ln x − 2 − 3π 2 − ζ(3) + 3 2 3 ε 2 2 4 3 88 − ln x(ln x + 1) − 2 3π + 4 ln x + 10 + 9π 2 + ζ(3) ln x 3 3 29 4 +20 + 12π 2 − π 4 + ζ(3) 30 3 7 1 21 2 1 +x − 3 + (8 ln x − 33) 2 + 26 ln x + 6 + π ε ε 2 ε 1 3 2 2 + −32 ln x − 4(21 + 26π ) ln x + 180 + 209π + 904ζ(3) 6 2

+ 2Li3 (−x) − 2 ln xLi2 (−x) − ln2 x + π 2 ln(1 + x) ε

−4x 8 (Li3 (−x) − ln xLi2 (−x)) − 4 ln2 x + π 2 ln(1 + x) +4 (S2,2 (−x) − ln xS1,2 (−x)) − 44Li4 (−x) +4 (ln(1 + x) + 6 ln x − 2) Li3 (−x) − ln2 x + π 2 ln2 (1 + x) 10 −2 ln2 x + 2 ln x ln(1 + x) − 4 ln x + π 2 Li2 (−x) 3 8 3 10 ln x + 4 ln2 x + π 2 ln x + 4π 2 − 4ζ(3) ln(1 + x) . (5.65) + 3 3

Proceeding in the same way with the second recurrence relation (5.63) and inserting there our analytical results for the two master double boxes we obtain the possibility to check these two results.

126

5 IBP and Reduction to Master Integrals

Although boxes with a one-loop insertion and a diagonal are simple quantities one can reduce them further. In the former case, the reduction was described in Example 5.8. Let us now do this for the latter case and consider, following [3], Example 5.10. Reduction of boxes with a diagonal shown in Fig. 5.4b. We imply that we have already got rid of the numerators as before, by shifting dimension and indices. Applying our auxiliary triangle rule (5.33) to the triangles (3, 5, 6) and (2, 7, 6) in Fig. 5.4b we obtain (d − 2a27 − 2)a2 2+ = (d − 2a6 − 2)a6 6+ − (d − 2a27 − 2)a7 7+ , (d − 2a35 − 2)a5 5+ = (d − 2a6 − 2)a6 6+ − (d − 2a35 − 2)a3 3+ .

(5.66) (5.67)

These relations can be used to reduce a2 and a5 to one. Then the following IBP relations derived in [3] can be used to reduce a3 and a7 to one: s(d − 2a35 − 2)a3 3+ = −(d − a356 − 1)(3d − 2a223567 ) +2(d − a356 − 1)a7 2− 7+ + (d − 2a6 − 2)a6 2− 6+ ,

(5.68)

t(d − 2a27 − 2)a7 7+ = −(d − a267 − 1)(3d − 2a235567 ) +2(d − a267 − 1)a3 5− 3+ + (d − 2a6 − 2)a6 5− 6+ .

(5.69)

To reduce a6 to one, the following relation valid for a2 = a3 = a5 = a7 = 1 and derived in [3] can be used: st(d − 2a6 − 2)a6 6+ = −(s + t)(d − a6 − 3)(3d − 2a6 − 10) +2(d − a6 − 3)(t2− 7+ + s2+ 7− ) + (d − 2a6 − 2)a6 6+ (t2− + s7− ) . (5.70) Finally, we have to express the master box with a diagonal, B5.10 (1, . . . , 1, d+ 2n), in the shifted dimension in terms of B5.10 (1, . . . , 1, d) which is given by (4.58) in expansion in ε. This can be done by the same trick with shifting dimension as above: we write down relation (5.42) for the box with a diagonal, i.e. where the function U is given by U = (α2 + α7 )(α3 + α5 ) + α6 (α2 + α3 + α5 + α7 ) ,

(5.71)

according to (2.24), and apply it to B5.10 (1, . . . , 1, d). Then we proceed exactly as in Example 5.8 and arrive at a desired recurrence relation. The algorithm presented above enables us to reduce any massless double box in terms of the two master integrals K1 and K2 , two master boxes with a one-loop insertion and a diagonal and a family of integrals (two-loop planar vertices and products of triangles) expressed in terms of gamma functions. As was pointed out later [27] the choice of the second master integral K2 as the integral with a dot on the sixth line brought complications in practical calculations because one obtained a linear combination of K1 and K2 with a coeﬃcient involving 1/ε, but the calculation of the master integrals in one more order in ε looked rather nasty (at that time ;-)). Two solutions of

5.4 Conclusion

127

this problem have appeared immediately. In [23], this very combination of the master integrals was indeed calculated using the method of diﬀerential equations (to be studied in Chap. 7), while in [5] another choice of the master integrals was made: instead of K(1, 1, 1, 1, 1, 2, 1, 0), the authors have taken the integral K(1, 1, 1, 1, 1, 1, 1, −1) as the second complicated master integral. This was a more successful choice because, according to the calculational experience, no negative powers of ε occur as coeﬃcients at these two new master integrals.

5.4 Conclusion When solving the problem of the reduction to master integrals, one tries to use all possible IBP relations. For h-loop Feynman integrals over the loop momenta ki depending on n independent external momenta pj , all possible IBP relations with derivatives (∂/∂ki ) · pj and (∂/∂ki ) · kj are used. For example, for the double boxes, this gives 10 IBP relations. In addition to the IBP relations, one can use the so-called Lorentz-invariance (LI) identities [24]. They follow from the fact that scalar Feynman integrals are invariant under inﬁnitesimal Lorentz transformations of the external momenta, pµi → pµi + εµν pνi . For example, in the case of four-point Feynman integrals (in particular, double boxes) with three independent external momenta, this provides the following relation, in addition to 10 IBP relations: 3 ∂ ∂ pn,µ ν − pn,ν µ = 0 (5.72) (pµ1 pν2 − pν1 pµ2 ) ∂pn ∂pn n=1 as well as the other two relations obtained by the cyclic permutations from (5.72). Well, if we turn to alpha or Feynman parameters, the Lorentz invariance becomes manifest and the equations (5.72) trivially hold (in contrast to the IBP relations), so that one might think that the LI equations follow from the IBP relations. However, explicitly, this statement has not been proven. Anyway, the LI identities can be certainly practically very useful. One can consider them together with the IBP relations and not bother about whether they are linear combinations of some IBP relations. There are a lot of papers where reduction problems for various classes of Feynman integrals were solved, in some way, with the help of IBP relations. Here is a very short list of some of them, starting from the two-loop level. Historically, IBP relations were ﬁrst successfully applied in [15] to threeloop massless propagators diagrams shown in Fig. 5.6. The corresponding algorithm [29] called MINCER was implemented in FORM [52]. In [12, 19, 22, 30], the problem of reduction for two-loop on-shell diagrams was solved: in [30], relevant recurrence relations were derived and used to ﬁnd all necessary integrals, and, in [12], a general algorithm implemented in the REDUCE [33]

128

5 IBP and Reduction to Master Integrals

Fig. 5.6. Three-loop massless planar, non-planar and Mercedez–Benz propagator diagrams

package Recursor was constructed. The reduction in the three-loop case was developed in [38] and, completely, in [39] with an implementation in FORM [52] (although no details of the reduction procedure were presented, as in many other cases). The reduction of two-loop bubble integrals with diﬀerent masses was solved in [21]. Three-loop vacuum diagrams with one mass were considered in [6, 12, 46]. The corresponding computer package MATAD was developed in [46]. The reduction problem for the massless on-shell double boxes in the nonplanar case (Fig. 4.9b where all lines are massless) was solved, using IBP and LI relations, in [2] and, in the case of (simpler) pentabox diagrams, in [3]. The general algorithm for the massless on-shell double boxes resulted in a series of NNLO calculations of various scattering processes – see, e.g., [26] for a review. The reduction of two- and three-loop propagator diagrams in Heavy Quark Eﬀective Theory was solved in [13, 31]. A pedagogical introduction to recursion problems oriented at HQET can be found in a recent review [32]. Unfortunately, the way how IBP relations are solved is not often explained. A typical example of such a situation is solving the reduction problem for two-loop vertex diagrams at threshold, q 2 = 4m2 : two independent algorithms were constructed [7, 17] but never published. The examples presented in this chapter and the papers cited above show how IBP relations can be solved without systematization. In other words, if it is necessary to solve a new problem, one can use the experience obtained in these examples and then analyse the new situation with the hope to solve somehow corresponding IBP relations. Still the complexity of unsolved calculational problems requires a systematization in this ﬁeld. One might hope that a systematization can be achieved within the technique based on shifting dimension [47]. Typical tricks were described in the previous section. Some prescriptions of this technique were presented in [48, 49]. Another example of its applications [43] is provided by the calculation of Feynman integrals relevant to the two-loop quark potential (to be considered within another technique in Chap. 6). It was also used to solve the reduction problem for two-loop propagator integrals with arbitrary masses [47]. Anyway, this technique provides the possibility to get rid of the numerators (which, of course, make the problem of the reduction more complicated) from the beginning.

5.4 Conclusion

129

Another attempt of a systematization was initiated in [25, 37, 38]. It is based on the observation that the total number of IBP and Lorentz invariance equations grows faster than the number of independent Feynman integrals, labelled by the powers of propagators and the powers of independent scalar products in the numerators, when the total dimension of the denominator and numerator in Feynman integrals associated with the given graph is increased. Therefore this system of resulting equations sooner or later becomes overconstrained, and one obtains the possibility of performing a reduction to master integrals. To be formal let us modify our notation for the Feynman integrals a little bit. Consider now, as a general Feynman integral, b H1b1 . . . HNN22 , (5.73) F (a1 , . . . , aN1 ; b1 , . . . , bN2 ) = · · · dd k1 . . . dd kh a1 a E1 . . . ENN1 1 instead of the dimensionally regularized version of (2.6). Now, we consider all the indices ai and bi to be positive or zero, both in the denominator and numerator. As before, all the quantities Ei and Hi are considered linear or quadratic with respect to the loop momenta. So, the idea [37, 38] is to write all possible IBP and LI relations for Feynman integrals (5.73) with a ﬁxed N1 + N2 = N . Our experience tells us that starting from some large N this will be an overconstrained linear system of equations which will be solved successfully (using a computer, of course). A breakthrough in the implementation of this idea came due to the following two publications: the ﬁrst practical successful implementation was achieved for the reduction of massless double box diagrams with one leg oﬀ-shell [25] (which was applied for NNLO calculations of the process e+ e− → 3jets – see [40] for a review), and detailed prescriptions for the implementation of this method in a general situation were presented in [37]. These two important works have resulted in a series of various calculations at the two-loop level – see, e.g. [1, 8, 9, 10, 11, 16, 44]. The implementation of this method on a computer in non-trivial situations was hardly possible, say, ten years ago. Indeed, for example, in the case of the double boxes with one leg oﬀ-shell, it was necessary [25] to solve linear systems of dozens of thousands of equations for dozens of thousands of variables. It is not clear at the moment what the practical limits of applications of this algorithm are, for example, whether it can be applied successfully to such problems as the reduction of triple boxes or four-loop massless propagator diagrams. This method is rather pragmatic and is a kind of experimental mathematics because its analysis from the mathematical point of view is absent. In particular, it is not known which linear equations of the method are really independent. It is not clear in advance which will be master integrals in a given problem: this becomes clear after solving the corresponding system of equations. The authors of [1, 8, 9, 10, 11, 16, 25, 44] constructed various computer implementations of this method. Fortunately, a ﬁrst public version called AIR

130

5 IBP and Reduction to Master Integrals

which can be applied to any problem, with the hope to obtain a solution of a concrete reduction problem, has recently appeared [4]. Now, to solve a new reduction problem, one can try to adjust this general computer algorithm, rather than solve IBP relations oneself. Well, if it turns out that this algorithm does not work, for some reasons (e.g. the lack of time or computer memory), then one could still try to solve the reduction problem in some way. One more option is described in the next chapter, where we will study a method which does not resemble any previously developed technique in this ﬁeld. The explicit and detailed recipes for solving overdetermined systems of equations presented in [37] are more optimal than the simple Gauss elimination. In fact, the Gauss elimination is present there, but only after the initial system is ordered according to some criteria. Then diﬀerent terms of the equations are characterized by a relative weight of their complexity, and the equations are solved starting from the most complicated terms. One could still look for more optimal strategies. In particular, one could hope to use a Gr¨ obner basis in this situation. This idea was already discussed in [48, 50] and applied to the case of two-loop propagator integrals with general masses. In this case, it was possible to use an existing Gr¨obner basis for diﬀerential equations with coeﬃcients independent of the arguments because, in the case of general non-zero masses, the initial problem of solving IBP type equations can be reduced to solving some systems of diﬀerential equations. Unfortunately, one usually needs physical cases, where zero masses are unavoidable. Solving the reduction problem with general non-zero masses and taking a massless (and on-shell or threshold) limit in the corresponding solution is not a natural procedure, because coeﬃcients at master integrals in this general solution are singular in these limits. Another variant here is to try to construct a Gr¨ obner basis adequate to deal with IBP equations which are diﬀerence equations with respect to the indices. This is, however, an open mathematical problem.

References 1. U. Aglietti and R. Bonciani, Nucl. Phys. B 668 (2003) 3; hep-ph/0401193; U. Aglietti, R. Bonciani, G. Degrassi and A. Vicini, Phys. Lett. B 595 (2004) 432; hep-ph/0407162. 129 2. C. Anastasiou, T. Gehrmann, C. Oleari, E. Remiddi and J.B. Tausk, Nucl.Phys. B 580 (2000) 577. 128 3. C. Anastasiou, E.W.N. Glover and C. Oleari, Nucl. Phys. B 575 (2000) 416 [Erratum-ibid. B 585 (2000) 763]. 118, 126, 128 4. C. Anastasiou and A. Lazopoulos, JHEP 0407 (2004) 046. 130 5. C. Anastasiou, J.B. Tausk and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 262. 127 6. L.V. Avdeev, Comput. Phys. Commun. 98 (1996) 15. 128 7. M. Beneke, A. Signer and V.A. Smirnov, Phys. Rev. Lett. 80 (1998) 2535. 128

References

131

8. W. Bernreuther et al., hep-ph/0406046. 129 9. T.G. Birthwright, E.W.N. Glover and P. Marquard, hep-ph/0407343. 129 10. R. Bonciani, A. Ferroglia, P. Mastrolia, E. Remiddi and J. J. van der Bij, Nucl. Phys. B 681 (2004) 261. 129 11. R. Bonciani, P. Mastrolia and E. Remiddi, Nucl. Phys. B 661 (2003) 289; B 676, 399 (2004); B 690, 138 (2004). 129 12. D.J. Broadhurst, Z. Phys. C 54 (1992) 599. 127, 128 13. D.J. Broadhurst and A.G. Grozin, Phys. Lett. B 267 (1991) 105. 128 14. K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Nucl. Phys. B 174 (1980) 345. 115 15. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 109, 127 16. M. Czakon, J. Gluza and T. Riemann, hep-ph/0406203. 129 17. A. Czarnecki and K. Melnikov, Phys. Rev. Lett. 87 (2001) 013001. 128 18. A.I. Davydychev, J. Phys. A 25 (1992) 5587. 113 19. A.I. Davydychev and A.G. Grozin, Phys. Rev. D 59 (1999) 054023. 127 20. A.I. Davydychev and P. Osland, Phys. Rev. D 59 (1999) 014006. 118 21. A.I. Davydychev and J.B. Tausk, Nucl. Phys. B 397 (1993) 123. 128 22. J. Fleischer and O.V. Tarasov, Comput. Phys. Commun. 71 (1992) 193; J. Fleischer and M.Yu. Kalmykov, Comp. Phys. Comm. 128 (2000) 531. 127 23. T. Gehrmann and E. Remiddi, Nucl. Phys. Proc. Suppl. 89 (2000) 251. 127 24. T. Gehrmann and E. Remiddi, Nucl. Phys. B 580 (2000) 485. 127 25. T. Gehrmann and E. Remiddi, Nucl. Phys. B 601 (2001) 248; Nucl. Phys. B 601 (2001) 287. 129 26. E.W.N. Glover, Nucl. Phys. Proc. Suppl. 116 (2003) 3. 128 27. E.W.N. Glover and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 196. 126 28. R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542. 118 29. S.G. Gorishny, S.A. Larin, L.R. Surguladze and F.V. Tkachov, Comput. Phys. Commun. 55 (1989) 381; S.A. Larin, F.V. Tkachov and J.A.M. Vermaseren, Preprint NIKHEF-H/91-18 (Amsterdam 1991). 127 30. N. Gray, D.J. Broadhurst, W. Grafe and K. Schilcher, Z. Phys. C 48 (1990) 673; D.J. Broadhurst, N. Gray and K. Schilcher, Z. Phys. C 52 (1991) 111. 127 31. A.G. Grozin, JHEP 0003 (2000) 013. 128 32. A.G. Grozin, Int. J. Mod. Phys. A 19 (2004) 473. 128 33. A.C. Hearn, REDUCE User’s Manual, Version 3.7 (ZIB, Berlin, 1999). 127 34. G. ’t Hooft and M. Veltman, Nucl. Phys. B 160 (1979) 151. 109 35. G. K¨ allen and A. Wightman, Mat. Fys. Skr. Dan. Vid. Selsk. 1 (No.6) (1958) 1. 113 36. G. Kramer and B. Lampe, J. Math. Phys. 28 (1987) 945. 117, 118 37. S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 129, 130 38. S. Laporta and E. Remiddi, Phys. Lett. B 379 (1996) 283. 128, 129 39. K. Melnikov and T. van Ritbergen, Phys. Lett. B 482 (2000) 99; Nucl. Phys. B 591 (2000) 515. 128 40. S. Moch, P. Uwer and S. Weinzierl, Nucl. Phys. Proc. Suppl. 116 (2003) 8. 129 41. W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 118 42. J.L. Rosner, Ann. Phys. 44 (1967) 11. 115 43. Y. Schr¨ oder, Phys. Lett. B 447 (1999) 321; Ph.D. thesis (Hamburg, 1999), DESY–THESIS–1999–021. 128 44. Y. Schr¨ oder, Nucl. Phys. Proc. Suppl. 116 (2003) 402; Y. Schr¨ oder and A. Vuorinen, hep-ph/0311323. 129 45. V.A. Smirnov and O.L. Veretin, Nucl. Phys. B 566 (2000) 469. 120, 121, 122, 123

132 46. 47. 48. 49. 50. 51.

5 IBP and Reduction to Master Integrals

M. Steinhauser, Comput. Phys. Commun. 134 (2001) 335. 128 O.V. Tarasov, Nucl. Phys. B 480 (1996) 397; Phys. Rev. D 54 (1996) 6479. 119, 120, 123, 128 O.V. Tarasov, Acta Phys. Polon. B 29 (1998) 2655. 128, 130 O.V. Tarasov, Nucl. Phys. (Proc. Suppl.) 89 (2000) 237. 128 O.V. Tarasov, hep-ph/0403253. 130 A.N. Vassiliev, Yu.M. Pis’mak and Yu.R. Khonkonen, Teor. Mat. Fiz. 47 (1981) 291. 109 52. J.A.M. Vermaseren, Symbolic Manipulation with FORM (CAN, Amsterdam, 1991). 127, 128

6 Reduction to Master Integrals by Baikov’s Method

In the previous chapter, we solved IBP relations [7] in a non-systematic way. Now we are going to do this systematically following Baikov’s method1 [2, 4, 5, 14]. Our goal is to solve the reduction problem, i.e. to develop an algorithm that would enable us to express any Feynman integral of a given family of Feynman integrals which are labelled by powers of the propagators (indices) as a linear combination of some master integrals. A characteristic feature of this method is the reduction to a minimal number of master integrals. In Sect. 6.1, the basic parametric representation which is an essential ingredient of this method will be described. In Sect. 6.2, this representation will be applied to formulate a strategy for identifying master integrals and constructing the corresponding coeﬃcient functions. As usual, we shall end up, in Sects. 6.2 and 6.3, with a lot of instructive examples starting from very simple ones. We shall continue to use mainly the examples considered in the previous chapters. In conclusion, applications and open problems of the method will be characterized.

6.1 Basic Parametric Representation Suppose that we are dealing with a family d d k1 . . . dd kh F (a) = · · · aN , E1a1 . . . EN

(6.1)

of h-loop dimensionally regularized Feynman integrals, where the factors in the denominator are given by (2.7) with r = 1, . . . , N = h(h + 1)/2 + hn. The denominators are quadratic or linear with respect to the loop momenta pi = ki , i = 1, . . . , h, and the independent external momenta ph+1 , . . . , ph+n of the graph. The ai are integer indices. Underlined letters denote collections of variables, i.e. a = (a1 , . . . , aN ), etc. 1 In [2], it was characterized as a ‘non-recursive’ solution of IBP recurrence relations. As we will see shortly, solving some recurrence relations is necessary within this method. However, these auxiliary recurrence relations are simpler than the initial IBP recurrence relations for a given family of Feynman integrals.

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 133–163 (2004) c Springer-Verlag Berlin Heidelberg 2004

134

6 Reduction to Master Integrals by Baikov’s Method

Some of the factors in the denominator are associated with irreducible numerators (which cannot be expressed linearly in terms of the given set of the denominators), so that the corresponding indices ai are considered only non-positive. We are going to solve the reduction problem in a maximal way, i.e. to be able to represent a given Feynman integral as a linear combination of a minimal number of some true master (or, irreducible) integrals, ci (a)Ii , (6.2) F (a) = i

with the natural normalization conditions ci (Ij ) = δij

(6.3)

which simply mean that any master integral cannot be expressed in terms of other master integrals. In fact, the master integrals are integrals of the given family, Ii = F (ai ), where ai = (ai1 , . . . , aiN ) are some concrete sets of indices. In the approach under consideration, the master integrals have indices air equal to one, or zero, or a negative value. Mathematically, if the reduction problem has been solved, we know a basis in the linear space of the given Feynman integrals. Then we could turn to some other basis. In particular, we could choose all the master integrals which have only positive indices. Consider, for example, the propagator integrals of Example 5.3 and choose, instead of I1 = F (1, 1) and I2 = F (1, 0), say, I1 = F (1, 1) and I2 = F (2, 1), why not? Well, practically, this is an unnatural choice. According to our experience of solving IBP relations and our standard attempts to reduce complicated integrals to simpler integrals, we imply that the master integrals must have as many non-positive indices as possible, so that we always keep this hierarchy in mind. Therefore, when we say that a given integral is irreducible, we omit the words to simpler integrals, in this sense, i.e. that have more non-positive indices. Our experience of solving IBP recurrence relations, in particular, the examples of Chap. 5, shows that the coeﬃcient functions ci (a) are rational functions of everything, i.e. of dimension, masses and external kinematical invariants. This property is a useful postulate that can be used in the calculation of the coeﬃcient functions. Within the approach of [2, 14], every coeﬃcient function in (6.2) satisﬁes, by construction, the initial IBP relations for (6.1) so that these relations for the given Feynman integrals are automatically satisﬁed. Let us start with the case of vacuum Feynman integrals which are functions of some masses and are deﬁned by (6.1) with 2 Aij (6.4) Er = r ki · kj − mr , h≥i≥j≥1

with r = 1, . . . , N = h(h + 1)/2.

6.1 Basic Parametric Representation

135

The IBP relations in the vacuum case originate from the following N equations: kj ∂ d d · = 0, i ≥ j . (6.5) · · · d k1 . . . d kh aN ∂ki E1a1 . . . EN We proceed, in this general situation, like in multiple examples in the previous chapter, i.e. perform diﬀerentiation and then express the resulting scalar products ki ·kj in terms of the denominators Er . When we invert the relations (6.4) we obtain a matrix which is inverse, in some sense, to the matrix Aij r . So, we write down the IBP relations in the following form: ji − A¯ir i A˜r r + m2r ar r+ = (d − h − 1)δij /2 , (6.6) r,r ,i ij ij ji where A¯ij r = Ar for i = j, Ar /2 for i > j and Ar /2 for i < j. The matrix A˜ is deﬁned as follows. Take the quadratic N × N matrix A, where the ﬁrst index is labelled by pairs (i, j) with i ≥ j, and the second index is r. The corresponding inverse matrix (A−1 )ij r (with i ≥ j) satisﬁes N

−1 i j Aij )r = δii δjj . r (A

(6.7)

r=1 −1 ij )r to all values i, j. Then A˜ij r is the symmetrical extension of (A + − Moreover, the operators r and r in (6.6) are our usual operators that increase and lower indices:

r+ F (. . . , ar , . . .) = F (. . . , ar + 1, . . .) , r− F (. . . , ar , . . .) = F (. . . , ar − 1, . . .) .

(6.8a) (6.8b)

We extensively exploited these operators in Chap. 5 for various concrete values of r. To construct the coeﬃcient functions ci (a) in the vacuum case, the following basic representation [2] is applied: dx1 . . . dxN (d−h−1)/2 [P (x )] , (6.9) ... xa1 1 . . . xaNN where the parameters x = (x1 , . . . , xN ) are obtained from x = (x1 , . . . , xN ) by the shift xi = xi + m2i . Integration over the parameters xi is understood in some way, with the requirement that the IBP in this parametric integral is valid. In this case, such objects satisfy the initial IBP relations (6.6). This property can be veriﬁed straightforwardly if we take into account that the operator ar r+ is transformed into the diﬀerential operator ∂/∂xr and the operator r− is transformed into the multiplication by xr . Now, the basic polynomial P of x which enters (6.9) is [2] N

ij ˜ A xr . (6.10) P (x) = det ij

r

r=1

136

6 Reduction to Master Integrals by Baikov’s Method

Here are simple practical prescriptions for evaluating the basic polynomials: 1. Solve the system Aij r ki · kj = Er ,

r = 1, . . . , N

i≥j≥1

with respect to ki · kj , i ≥ j; 2. Replace Er by xr on the right-hand side of this solution; 3. Extend this expression to all values of i and j in the symmetrical way; 4. Take the determinant of this matrix to obtain P . In fact, the basic polynomial is deﬁned up to a normalization factor independent of the variables xj . This will be clear when constructing the coefﬁcient functions which will be themselves normalized at some point. For general Feynman integrals, the problem can be reduced to the vacuum case [2, 4]. If there is one external momentum, q, so that we are dealing with a family of propagator-type integrals, one involves into the game coeﬃcients of the Taylor expansion of F (a) in q 2 , F (q ; a1 , . . . , aN ) ∼ 2

∞

(q 2 − m2N +1 )aN +1 −1 F (a1 , . . . , aN , aN +1 ) .

aN +1 =1

(6.11) It turns out [2, 4] that the so deﬁned objects F (a1 , . . . , aN , aN +1 ) (with some overall rescaling factor which is not important in the examples in this chapter) satisfy vacuum IBP relations. To formulate a prescription for corresponding basis polynomials in the non-vacuum case, we need ﬁrst to present a preliminary discussion of constructing master integrals. To identify candidates for master integrals in a ﬁrst approximation, we shall analyse integrals where the indices corresponding to irreducible numerators are set to zero and other indices are either zero or one. Let F (ai ) with aij = 1 or 0 be a candidate to be considered as a master integral. Let us remember the examples of Chap. 5, where the reduction always goes down: our experience tells us that a master integral Ii = F (ai ) = F (ai1 , . . . , air , . . . , aiN ) never appears in the decomposition of a given Feynman integral in terms of master integrals F (a) = . . . + ci (a1 , . . . , ar , . . . , aN )Ii + . . . if ar ≤ 0 and air > 0. Therefore, we come to the natural condition for the coeﬃcient function ci (a) of F (ai ): if air = 1 then ci (a1 , . . . , ar , . . . , aN ) = 0 for ar ≤ 0. This condition can be realized easily [2] in an automatic way by treating the integration over xj as a Cauchy integral around the origin in the complex xj -plane,

6.1 Basic Parametric Representation

1 2πi

(

dxj a xj j

137

. . . [P (x)]

(d−h−1)/2

.

(6.12)

According to the Cauchy theorem, this expression reduces to the Taylor expansion of order aj − 1 of the integrand in xj so that it becomes a linear combination of terms dxj z−nd (6.13) . . . [Pi (x)] n , x j j:a ≤0 j ij

where z = (d − h − 1)/2, and Pi (x) is obtained from P (x) by setting to zero all the variables xj with j such that aij = 1. We shall use nj instead of aj for powers of xj in auxiliary parametric integrals. Observe that the parameter nd in such integrals plays the role of the shift of the dimension. Suppose that we are not interested in higher terms of the Taylor expansion in powers of (q 2 − m2N +1 ) in (6.11), i.e. we need just the value at q 2 = m2N +1 , i.e. the term with aN +1 = 1. Then the integration over xN +1 should be understood in the sense of Cauchy integration so that, eﬀectively, xN +1 is set to zero. So, if Pˆ (x1 , . . . , xN , xN +1 ) is the basic polynomial for the corresponding vacuum problem, then the basic polynomial for the initial propagator-type problem is obtained as P (x) ≡ P (x1 , . . . , xN ) = Pˆ (x1 , . . . , xN , 0) .

(6.14)

In the case of n independent external momenta q1 , . . . , qn , one includes into the procedure all the terms of the formal Taylor expansions in the scalar products qi · qj . One is usually interested only in the value at some qi · qj and not in the derivatives at these points. (Otherwise, it would be necessary to deal with a generalization of (6.11), where the initial Feynman integrals are rescaled by the Gram determinant det(pi · pj ) which is raised to the power (h + n + 1 − d)/2 – see [2, 4].) Then the transition to the vacuum problem, which eﬀectively increases the number of loops, h → h + n, can be performed as follows: 1. Introduce a complete set of invariants by considering, in addition to ki ·kj , i ≥ j and ki · qj , also invariants generated by the external momenta, i.e. the scalar products qi · qj , i ≥ j. Let pi = ki , i = 1, . . . , h and pi = qi , i = h + 1, . . . , h + n so that the total number of the kinematical invariants ˆ = (h + n)(h + n + 1)/2. becomes N 2. Introduce, in some way, the corresponding new propagators. 3. Solve the system ˆ Aij r pi · pj = Er , r = 1, . . . , N i≥j≥1

with respect to pi · pj . 4. Evaluate the basic polynomial Pˆ for such a vacuum problem. 5. Obtain P (x) ≡ P (x1 , . . . , xN ) = Pˆ (x1 , . . . , xN , 0, . . . , 0).

138

6 Reduction to Master Integrals by Baikov’s Method

Observe that the method under consideration is based only on the IBP relations so that the LI identities discussed in Sect. 5.4 are not used at all.

6.2 Constructing Coeﬃcient Functions. Simple Examples Now, we want to apply the basic parametric representation for two closely related purposes: – identifying master integrals, – constructing the corresponding coeﬃcient functions. According to the discussion above, let us consider integrals where the indices corresponding to irreducible numerators are set to zero and other indices are either zero or one. Let Ii = F (ai ) = F (ai1 , . . . , air , . . . , aiN ). For indices equal to one, we understand the corresponding integration over xj in the basic parametric representation (6.9) in the Cauchy sense. This leads to a Taylor expansion of order aj − 1 of the integrand in xj and gives a linear combination of (6.13). Let us try to understand whether a given candidate can be considered as a master integral. Suppose that Pi = 0. Then there is no other way as to consider the coeﬃcient function equal to zero. Therefore, this integral cannot be a master integral and has to be recognized as a reducible integral within the reduction problem. Let us assume a weaker condition: the parametric integral involves an integral without scale which we put, by deﬁnition, to zero. Then, again, we cannot construct the coeﬃcient function in a non-trivial way so that the corresponding integral is considered reducible. Let us stress that such a scaleless integral can appear not only immediately but also after some preliminary non-trivial integrations. After such analysis, we obtain a preliminary list of master integrals. Sometimes one has to consider master integrals which diﬀer from F (ai ) by some indices aij < 0. The number of such additional master integrals is connected with the degree of the polynomial Pi with respect to some of the parameters xj . Let us now turn to examples and see how the basic parametric representation enables us to solve the reduction problem. Many examples will be the same as in Chap. 5, in particular, the ﬁrst one. Example 6.1. One-loop vacuum massive Feynman integrals given by the right-hand side of (5.1). We have one propagator with the denominator E = k 2 − m2 and one kinematical invariant k 2 . The equation E = k 2 is solved as k 2 = E. Therefore, the resulting basic polynomial is P (x) = x and the polynomial that enters

6.2 Constructing Coeﬃcient Functions. Simple Examples

139

(6.9) is P (x ) = x + m2 . There is one master integral I1 = F6.1 (1) given by the right-hand side of (5.6). According to (6.9) the corresponding coeﬃcient function is ( dx 1 dx 2 (d−2)/2 c(a) ∼ (x + m ) = (x + m2 )(d−2)/2 . (6.15) xa 2πi xa At a = 1 we have ( 1 dx (x + m2 )(d−2)/2 = (x + m2 )(d−2)/2 = (m2 )(d−2)/2 . 2πi x x=0 To satisfy the normalization c(1) = 1 we deﬁne ( (m2 )(2−d)/2 dx c(a) = (x + m2 )(d−2)/2 2πi xa a−1 ' $ (m2 )(2−d)/2 ∂ (x + m2 )(d−2)/2 = . (a − 1)! ∂x x=0

(6.16)

for a = 1, 2, . . .. So, we have F6.1 (a) = c(a)I1 , in agreement with (5.5) and the explicit result (A.1). As in Chaps. 3 and 5 let us consider Example 6.2. Massless one-loop propagator Feynman integrals given by the right-hand side of (5.7). The transition to the corresponding vacuum problem reduces to adding a new propagator, 1/(q 2 − m2 )a3 , with an eﬀective mass m. The eﬀective number of loops that is involved in the exponent in (6.9) is h = 2. We want to consider the value of our diagram at some general point and are not interested in higher terms of the Taylor expansion in q 2 . Therefore, we consider only the value a3 = 1 so that, according to our agreements, the integration contour for the corresponding variable x3 is taken as a Cauchy contour around the origin, and x3 is set to zero. Thus, using (6.14), we obtain the basic polynomial P (x1 , x2 ) = (q 2 )2 − 2q 2 (x1 + x2 ) + (x1 − x2 )2 .

(6.17)

The only possible candidate for a master integral is I1 = F6.2 (1, 1) = iπ d/2 (−q 2 )d/2−2

Γ (2 − d/2)Γ 2 (d/2 − 1) . Γ (d − 2)

(6.18)

because integrals with one non-positive index are zero. The corresponding coeﬃcient function is 2 (d−3) q c1 (a1 , a2 ) = (a1 − 1)!(a2 − 1)! a1 −1 a2 −1 ∂ ∂ (d−3)/2 [P (x1 , x2 )] , (6.19) × ∂x1 ∂x2 xi =0

140

6 Reduction to Master Integrals by Baikov’s Method

where the normalization condition c1 (1, 1) = 1 was immediately implemented. One can check that this result is in agreement with what we had in Example 5.2 when explicitly solving recurrence relations. Let us now turn to Example 6.3. One-loop diagram for the heavy quark potential shown in Fig. 6.1.

3

2

1 Fig. 6.1. One-loop diagram for the heavy quark potential. A wavy line denotes a propagator for the static source

The corresponding general Feynman integral is dd k , F6.3 (a1 , a2 , a3 ) = (k 2 )a1 [(k − q)2 ]a2 (v·k + i0)a3

(6.20)

with v·q = 0. In addition to k 2 , q · k and v · k, we consider q 2 , v · q and v 2 as external kinematical invariants so that the eﬀective loop number is h = 3. The choice of additional propagators is arbitrary. We choose the following extended set of the denominators: E1 = k 2 , E2 = (k − q)2 , E3 = k·v + v 2 , E4 = v 2 , E5 = q 2 , E6 = (q + v)2 . The basic polynomial is given by the determinant of the matrix (x1 − x2 + x5 )/2 x3 − x4 x1 (x1 − x2 + x5 )/2 x5 (−x4 − x5 + x6 )/2 . (−x4 − x5 + x6 )/2 x4 x3 − x4

(6.21)

(6.22)

The variables xi are then shifted by the corresponding eﬀective masses, x3 → x3 + v 2 , x4 → x4 + v 2 , x5 → x5 + q 2 , x6 → x6 + (q + v)2 . We are not interested in higher order Taylor coeﬃcients of the additional kinematical invariants so that, eﬀectively, we set x4 = x5 = x6 = 0. Thus, we obtain

2 P (x1 , x2 , x3 ) = (q 2 )2 v 2 + v 2 (x1 − x2 ) + 2q 2 v 2 (x1 + x2 ) − 2x23 , Observe that integrals (6.20) are zero whenever a1 or a2 are non-positive. After analysing various integrals with the indices 1 and 0 and corresponding

6.2 Constructing Coeﬃcient Functions. Simple Examples

141

reduced polynomials we see that the coeﬃcient functions can be constructed non-trivially for the following two integrals which can be evaluated by (A.27) and which we consider as master: I1 = F6.3 (1, 1, 1)

√ (−q 2 )d/2−5/2 π Γ (5/2 − d/2)Γ (d/2 − 3/2)2 , v Γ (d − 3) Γ (2 − d/2)Γ (d/2 − 1)2 . I2 = F6.3 (1, 1, 0) = iπ d/2 (−q 2 )d/2−2 Γ (d − 2) = −iπ d/2

(6.23) (6.24)

The coeﬃcient function c1 is simply calculated without integration. For the coeﬃcient function c2 , we need the following integrals: a α g1 (k3 , α) = dx3 xk33 a2 − x23 . (6.25) −a

Here k3 is an integer but α depends on d. This integral can be interpreted in the sense of the principal value, with 2 α+k/2+1/2 Γ (k/2 + 1/2)Γ (α + 1) a for even k . (6.26) g1 (k, α) = Γ (α + k/2 + 3/2) 0 for odd k Let us imply that these and similar integrals below are understood as convergent integrals in an appropriate domain of analytical parameters, such as α in (6.26), with analytic continuation to the whole complex plane of α on the right-hand side. We obtain the following decomposition of the general integral of the given class: F6.3 (a1 , a2 , a3 ) = c1 (a1 , a2 , a3 )I1 + c2 (a1 , a2 , a3 )I2 .

(6.27)

One can check that this procedure is in agreement with the explicit result (A.27) evaluated in Sect. 3.1. Let us now consider again Example 6.4. Two-loop massless propagator Feynman integrals of Fig. 3.9 with integer powers of the propagators given by the right-hand side of (3.39). The transition to vacuum integrals is similar to Example 6.2. Now we have h = 3. The basic polynomial can be obtained straightforwardly: P (x1 , . . . , x5 ) = −x1 x2 x3 + x22 x3 + x2 x23 + x21 x4 − x1 x2 x4 −x1 x3 x4 − x2 x3 x4 + x1 x24 + x1 x3 x5 − x2 x3 x5 −x1 x4 x5 + x2 x4 x5 + q 2 [−x1 x2 + x2 x3 + x1 x4 −x3 x4 + x1 x5 + x2 x5 + x3 x5 + x4 x5 − x25 ] + (q 2 )2 x5 . (6.28)

142

6 Reduction to Master Integrals by Baikov’s Method

After analysing various candidates with the indices 1 and 0 we conclude that the corresponding integrals (6.13) with reduced polynomials Pi can be interpreted non-trivially only in the following three cases two of which are symmetrical to each other: F6.4 (1, 1, 1, 1, 0) = I1 , F6.4 (0, 1, 1, 0, 1) = F6.4 (1, 0, 0, 1, 1) = I2 . Thus, we qualify them as master integrals. The values of these integrals can be obtained from (5.22) and (5.23), respectively: Γ (2 − d/2)2 Γ (d/2 − 1)4 , Γ (d − 2)2 Γ (3 − d)Γ (d/2 − 1)3 . I2 = −(iπ d/2 )2 (−q 2 )d−3 Γ (3d/2 − 3) I1 = (iπ d/2 )2 (−q 2 )d−4

(6.29) (6.30)

The corresponding coeﬃcient functions are constructed using the values of the following integrals that appear in (6.13). For c1 , we use q2 2 β g2 (α, β) = dx5 xα 5 (q − x5 ) 0

α+β+1 Γ (α + 1)Γ (β + 1) . = q2 Γ (α + β + 2) For c2 , we use

∞

∞

g3 (α1 , α4 , β) = 0

(6.31)

2 β 1 α4 dx1 dx4 xα 1 x4 (q + x1 + x4 )

0

α1 +α4 +β+2 Γ (α1 + 1)Γ (α4 + 1)Γ (−α1 − α4 − β − 2) . = q2 Γ (−β)

(6.32)

The decomposition of an arbitrary integral is F6.4 (a1 , a2 , a3 , a4 , a5 ) = c1 (a1 , a2 , a3 , a4 , a5 )I1 + [c2 (a1 , a2 , a3 , a4 , a5 ) + c2 (a2 , a1 , a4 , a3 , a5 )] I2 .

(6.33)

We again consider Example 6.5. Two-loop massless vertex Feynman integrals (5.34) of Fig. 5.3 with integer powers of the propagators. This is also a relatively simple example which can be treated almost like the previous examples. We shall deal with the following extended set of the denominators of the propagators: E1 = l2 − 2l·p1 + p21 ,

E2 = l2 − 2l·p2 + p22 ,

E3 = k 2 − 2k·p1 + p21 , E4 = k 2 − 2k·p2 + p22 , E5 = k 2 , E6 = k 2 − 2k·l + l2 , E7 = l2 , E8 =

p21

,

E9 = p1 ·p2 ,

E10 =

p22

.

(6.34) (6.35)

6.2 Constructing Coeﬃcient Functions. Simple Examples

143

The basic polynomial is straightforwardly evaluated, as a determinant of the corresponding 4 × 4-matrix (6.10). The eﬀective number of loops to be used in (6.9) is h = 4. Since we are not interested in higher terms of expansion in the external kinematical invariants2 p21 , p22 and p1 ·p2 , as usual, the parameters x8 , x9 and x10 are set to zero, and we obtain the following basic polynomial, according to the last rule in Sect. 6.1: P (x) = x22 x23 − 2x1 x2 x3 x4 + x21 x24 + 4Q2 x1 x2 x5 − 2Q2 x2 x3 x5 +2x1 x2 x3 x5 − 2x22 x3 x5 − 2Q2 x1 x4 x5 − 2x21 x4 x5 + 2x1 x2 x4 x5 +(Q2 )2 x25 + 2Q2 x1 x25 + x21 x25 + 2Q2 x2 x25 − 2x1 x2 x25 + x22 x25 +2Q2 x2 x3 x6 + 2Q2 x1 x4 x6 − 2(Q2 )2 x5 x6 − 2Q2 x1 x5 x6 − 2Q2 x2 x5 x6 +(Q2 )2 x26 − 2Q2 x2 x3 x7 − 2x2 x23 x7 − 2Q2 x1 x4 x7 + 4Q2 x3 x4 x7 +2x1 x3 x4 x7 + 2x2 x3 x4 x7 − 2x1 x24 x7 − 2(Q2 )2 x5 x7 − 2Q2 x1 x5 x7 −2Q2 x2 x5 x7 − 2Q2 x3 x5 x7 − 2x1 x3 x5 x7 + 2x2 x3 x5 x7 − 2Q2 x4 x5 x7 +2x1 x4 x5 x7 − 2x2 x4 x5 x7 − 2(Q2 )2 x6 x7 − 2Q2 x3 x6 x7 − 2Q2 x4 x6 x7 +4Q2 x5 x6 x7 + (Q2 )2 x27 + 2Q2 x3 x27 + x23 x27 + 2Q2 x4 x27 −2x3 x4 x27 + x24 x27 ,

(6.36)

where Q2 = −(p1 − p2 )2 as before. After a straightforward analysis of candidates we identify the following set of the master integrals: F (1, 1, 0, 0, 1, 1, 0) = I1 , F (1, 1, 1, 1, 0, 0, 0) = I2 and F (0, 1, 1, 0, 0, 1, 0) = F (1, 0, 0, 1, 0, 1, 0) = I3 . To construct the coeﬃcient function c1 we have to deal with integrals (6.13), where the reduced polynomial is

P1 (x3 , x4 , x7 ) = x7 ((Q2 )2 + (x3 − x4 )2 (6.37) +2Q2 (x3 + x4 ))x7 + 4Q2 x3 x4 . One can observe that in the cases, where n4 ≤ 0 (n3 ≤ 0) in the corresponding integral (6.13), one can straightforwardly integrate over x4 (x3 ) and then over x3 (x4 ) and x7 , using ∞ β g4 (α, β) = dx xα (x + a) 0 α+β+1 Γ (1

=a

+ α)Γ (−α − β − 1) . Γ (−β)

(6.38)

Suppose now that n3 , n4 > 0 in (6.13). Then we can use a trick based on the following integration formula obtained by IBP in a one-parametric integral: ∞ dx (Ax + B)z−n n+γ x 0 2

Observe that this is a formal expansion for p21 and p22 and a Taylor expansion for p1 ·p2 .

144

6 Reduction to Master Integrals by Baikov’s Method

=

(z − n )B z − n − n − γ + 1

0

∞

dx (Ax + B)z−n −1 . xn+γ

(6.39)

Applying it to the integration over x7 we can reduce either n3 or n4 to zero because here B = 4Q2 x3 x4 . The coeﬃcient function c2 can easily be constructed because the corresponding integral (6.13) over x7 can be evaluated by means of the following explicit formula x2 g5 (k, α1 , α2 ) = dx xk (x − x1 )α1 (x2 − x)α2 x1

=

k r=0

x1k−r (x2 − x1 )α1 +α2 +r+1

Γ (1 + α2 )Γ (1 + α1 + r) k! , (6.40) (k − r)!r! Γ (α1 + α2 + r + 2)

and then over x5 and x6 by means of (6.38). A similar procedure, without tricks, can be developed for the coeﬃcient function c3 . If I3 = F (0, 1, 1, 0, 0, 1, 0), this is achieved by integrating over x7 (which always can be done because n7 ≤ 0), and then over x5 , x1 and x4 . For the second copy of I3 , the coeﬃcient function is symmetrically obtained. Let us again turn to our favourite example which illustrates all the basic methods. Example 6.6. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. The transition to the corresponding vacuum problem reduces to adding a new propagator, 1/(q 2 − s)a3 . We again consider these integrals at general q 2 and are not interested in derivatives so that, eﬀectively, the corresponding index will be a3 = 1 and the corresponding variable x3 is set to zero. The resulting basic polynomial is P (x1 , x2 ) = −(x1 − x2 + m2 )2 − q 2 (q 2 − 2m2 − 2(x1 + x2 )) .

(6.41)

Of course, at m = 0 it coincides with the polynomial (6.17) for Example 6.2. There are two master integrals F6.6 (1, 1) = I1 given by (1.5) and F6.6 (1, 0) = I2 given by the right-hand side of (5.6). We want to construct the corresponding coeﬃcient function with the normalization conditions (6.3), i.e. c1 (1, 1) = 1 , c1 (1, 0) = 0 , c2 (1, 1) = 0 ,

c2 (1, 0) = 1 .

The coeﬃcient function of I1 is simply obtained similar to the massless case 2 (d−3) q − m2 c1 (a1 , a2 ) = (a1 − 1)!(a2 − 1)! a1 −1 a2 −1 ∂ ∂ (d−3)/2 [P (x1 , x2 )] . (6.42) × ∂x1 ∂x2 xi =0

6.2 Constructing Coeﬃcient Functions. Simple Examples

145

For the coeﬃcient function c2 (a1 , a2 ) of I2 , we obtain linear combinations of one-parametric integrals dx (d−3)/2−n2 f (n1 , n2 ) = [P2 (x)] , (6.43) xn1 where P2 (x) = P (x1 , x)|x1 =0 = αx2 + βx + γ

(6.44)

with α = −1, β = 2(m2 + q 2 ), γ = −(m2 − q 2 )2 . Consider ﬁrst the case a2 ≤ 0. Then n1 is always non-positive here, and f (n1 , n2 ) can be understood as an integral between the roots 2 (1,2) = m ∓ q2 x2 of the quadratic polynomial P2 (x2 ), using (6.40). The evaluation at a1 = 1 and a2 = 0 provides a normalization factor to satisfy the normalization condition c2 (1, 0) = 1, and we obtain the following expression for c2 (a1 , a2 ) at a2 ≤ 0: c02 (a1 , a2 ) = c02 (a1 , a2 ) ≡ 1 × (a1 − 1)!

(2) x2 (1)

x2

Γ (d − 1) 4d−2 (m2 q 2 )(d−2)/2 Γ ((d

dx2 xn2

∂ ∂x1

a1 −1

− 1)/2)2

[P (x1 , x2 )]

(d−3)/2

.

(6.45)

x1 =0

In the case a2 > 0, the integrals f (n1 , n2 ) appear also with n1 > 0. When taken seriously they can be evaluated in terms of a Gauss hypergeometric function. Instead of doing this, let us apply IBP to our parametric integrals f (n1 , n2 ). This gives the relation f (n1 , n2 ) =

(d − 3)/2 − n2 n1 − 1 ×(2αf (n1 − 2, n2 + 1) + βf (n1 − 1, n2 + 1))

(6.46)

which can be used to reduce n1 to one or zero. Moreover, the identity (d−3)/2−n2

P2

(d−3)/2−n2 −1

= P2

P2

leads to the relation 1 f (1, n2 ) = (f (1, n2 − 1) − αf (−1, n2 ) − βf (0, n2 )) γ

(6.47)

which can be used to reduce n2 to zero. This means that we can express any f (n1 , n2 ) as a linear combination of an auxiliary master integral f (1, 0) and integrals f (n1 , n2 ) with n1 ≤ 0 which can be evaluated in terms of gamma functions. We believe that the coeﬃcient functions are rational functions of everything. The only chance to satisfy this property here is to construct c2 (a1 , a2 ) as a linear combination of c02 (a1 , a2 ) and the ﬁrst coeﬃcient function c1 (a1 , a2 ):

146

6 Reduction to Master Integrals by Baikov’s Method

c2 (a1 , a2 ) = c02 (a1 , a2 ) + Ac1 (a1 , a2 ) .

(6.48)

The constant A is determined by the normalization condition c2 (1, 1) = 0: A = −c02 (1, 1) .

(6.49)

After this, the dependence on f (1, 0) drops out and c2 (a1 , a2 ) indeed turns out to be a rational function. Observe that integrating over some real domain, in particular between the roots of a quadratic polynomial when constructing coeﬃcient functions, with a subsequent normalization, is in fact equivalent to solving IBP relations for our auxiliary parametric integrals. If there is such a possibility to understand a given parametric integral it is reasonable to use it. If there is no such possibility, e.g. one meets a polynomial of the third degree, or, an integration over one of the x-variables leads to inconvenient integrals over the rest variables, then there is no other way as to treat the auxiliary parametric integrals in a pure algebraic way by solving the corresponding IBP relations. We shall meet such situations in the examples below. As to the example above, the situation with a2 ≤ 0 could be treated algebraically, by IBP in the initial two-parametric integral, but integrating over x2 has simpliﬁed the situation.

6.3 General Recipes. Complicated Examples Let us extend what was done in the previous example to the general situation. After a preliminary analysis, with the help of (6.9), we obtain a preliminary list of candidates for the master integrals. Let us deﬁne the relation of partial ordering of the master integrals as follows: F (a1 ) < F (a2 ) if a1j ≤ a2j for all j , and the strict inequality holds at least for one index. The master integrals can be grouped into families characterized by their maximal integrals. Let us start from the master integrals which have most non-negative indices. Usually, the corresponding parametric integral for the coeﬃcient function can be understood in such a way that it results in integrations in terms of gamma functions. Consider now a situation with two master integrals with F (a2 ) < F (a1 ), and suppose that we already know c1 . If a2i = 1 we have also a1i = 1. To construct an algorithm for the coeﬃcient function c2 (a) we start with the case of negative indices aj for those indices j where a1j = 1 since in this case we have c1 (a) = 0. Experience shows that the integrations for c2 (a) result in ratios of gamma functions which in particular can be used to satisfy the normalization c2 (a2 ) = 1. In a next step one considers the case aj > 0. Then the corresponding parametric representation usually leads to integrals which cannot be evaluated in terms of gamma functions. (See the previous example.) Thus at ﬁrst

6.3 General Recipes. Complicated Examples

147

sight it looks hopeless to achieve that the coeﬃcient functions have to be rational functions of d. The way out is to look for an expression for the coeﬃcient function c2 (a) which is a linear combination of c1 (a) and the basic parametric representation for c2 (a) denoted by c02 (a) c2 (a) = c02 (a) + Ac1 (a) .

(6.50)

The constant A is determined by the normalization condition c2 (a1 ) = 0 which gives A = −c02 (a1 ) .

(6.51)

Then IBP is applied to the parametric integrals and the corresponding relations are used to express any given parametric integral in terms of auxiliary (parametric) master integrals and expressions which are straightforwardly evaluated in terms of gamma functions. The dependence on the new auxiliary master integrals has to drop out3 in order to provide a rational dependence of the coeﬃcient functions on d. In fact, this strategy can be generalized to the case of several master integrals with more complicated hierarchies. Let us proceed with examples, where we shall meet such situations. These will be mainly our old examples considered in Chaps. 3–5. Example 6.7. Feynman integrals (3.19) corresponding to the triangle diagram of Fig. 3.4. Almost all the steps can straightforwardly be performed, as above. The basic polynomial is P (x1 , x2 , x3 ) = (x1 − x3 )(x2 − x3 ) − Q2 x3 −m2 (Q2 + x1 + x2 − 2x3 ) + m4 ,

(6.52)

where again Q = −(p1 − p2 ) with = = 0. We obtain the following list of the master integrals: F (1, 1, 1) = I1 , F (1, 1, 0) = I2 and F (0, 0, 1) = I3 . When testing various candidates to be master integrals we consider, in particular, F (1, 0, 1) with the corresponding reduced polynomial P1,0,1 (x2 ) = m2 − Q2 − x2 linearly dependent on x2 . Let us try to understand the corresponding integrals (6.13) dx2 (6.53) (m2 − Q2 − x2 )z−nd n2 x2 2

2

p21

p22

in a non-trivial way. (Here we have z = (d − 4)/2 = −ε because the eﬀective number of loops is h = 3.) We do not consider the Cauchy integration around the origin in the complex plane because this choice corresponds to the value a2 = 1 in the master integral so that we are looking for other options. We 3

This cancellation serves as a good check of the algorithm, similarly to cancellations of spurious poles in ε on the right-hand side of various asymptotic expansions in momenta and/or masses [6].

148

6 Reduction to Master Integrals by Baikov’s Method

cannot integrate from x2 = 0 because we have integer negative powers of x2 . Still it looks like there is a chance to obtain a new non-trivial understanding of the integral by choosing to integrate from −∞ to m2 −Q2 Here we suppose that m2 − Q2 < 0 in order to have no singularity in the integration domain. However, this choice brings nothing new! One can check that, after the normalization by the equation c1,0,1 (1, 0, 1) = 1, one obtains the same expression as in the case of the Cauchy integration corresponding to other values of the index a2 . Therefore, we conclude that we cannot interpret (6.53) in a new non-trivial way so that the integral F (1, 0, 1) is not a master integral. A more general recipe is that, whenever we obtain in a linear dependence of a reduced polynomial in (6.13) on some variable, we shall usually4 conclude that this cannot be a master integral. The coeﬃcient function of I1 can be constructed trivially because it does not involve integration. The coeﬃcient function of I2 , with the corresponding polynomial P2 = (m2 + x3 )(m2 − Q2 + x3 ), is also simple (at least simpler than in Example 6.6). If n3 ≤ 0 in the corresponding integral (6.13), we can integrate between the roots of this polynomial using (6.40). In the case of n3 > 0, one can use the IBP relation with respect to x3 in order to reduce n3 to one and the relation following from the identity P2z−nd = P2z−nd −1 P2 to adjust the dimension. For the coeﬃcient function of I3 , we obtain integrals (6.13) with P3 (x1 , x2 ) = x1 x2 − m2 (Q2 + x1 + x2 ) + m4 . If one of the indices n1 and n2 in this integral is negative the integration over the corresponding variable, e.g. over x2 , can be performed but one obtains a power of (m2 − x1 ) not regularized by z. So, in this situation, it is necessary to proceed in a pure algebraic way and solve the corresponding IBP relations, together with the relation that follows from the identity P3z−nd = P3z−nd −1 P3 , in order to reduce any given integral to auxiliary master integrals. There is, however, one more option5 : to use the package AIR [1] based on the algorithm of [11] and designed to solve genuine IBP relations for Feynman integrals as discussed in the end of the previous chapter. It turns out that this program can be applied to the auxiliary IBP relations for integrals (6.13). As a result of this procedure, an algorithm for c3 can be constructed. In particular, we obtain 1 1 (d − 4)(2m2 − Q2 )I1 F (1, 1, 2) = 2 2 m (m − Q2 ) 2 4

Well, up to some pathological situations, where one has chances to obtain a new meaning for such integrals by considering the integration over xi in the sense of a distribution with respect to the variables on which coeﬃcients of the corresponding linear polynomial depend. 5 Thanks to J. Piclum who implemented the corresponding algorithm on a computer, also for the Example 6.10 below.

6.3 General Recipes. Complicated Examples

+(d − 3)I2 +

2−d I3 2m2

149

,

(6.54)

in agreement with (5.17), where several integrals expressed in terms of gamma functions were involved on the right-hand side. Let us again consider massless on-shell boxes which we have already analysed in Examples 3.3, 4.3 and 5.4. Example 6.8. The massless on-shell box Feynman integrals of Fig. 5.1 with p2i = 0, i = 1, 2, 3, 4 and general integer powers of the propagators. The basic polynomial is now P (x1 , x2 , x3 , x4 ) = s2 t2 + t2 (x1 − x2 )2 − 2st2 (x1 + x2 ) +s2 (x3 − x4 )2 − 2s2 t(x3 + x4 ) −2st[2x1 x2 + 2x3 x4 − (x1 + x2 )(x3 + x4 )] .

(6.55)

The eﬀective number of loops to be used in (6.9) is now h = 4. Using the strategy formulated above, we reveal the following three master integrals: F (1, 1, 1, 1) = I1 and F (1, 1, 0, 0) = F (0, 0, 1, 1) = I2 . The coeﬃcient function of I1 can be constructed trivially. In the case of I2 (the ﬁrst of the two symmetric variants), the integration in the corresponding integral (6.13) over x4 and then over x3 can be performed in terms of gamma functions if n4 ≤ 0, and, in the opposite order, in the case of n3 ≤ 0. One can then proceed similarly to Example 6.6 by introducing an auxiliary parametric integral and using IBP relations to reduce n3 or n4 to one or zero. Then, to deﬁne the coeﬃcient function c2 , one involves a linear combination with the coeﬃcient function c1 so that the dependence on this auxiliary integral drops out. Now we turn to a massive generalization of this example. Example 6.9. The on-shell boxes with two massive and two massless lines shown in Fig. 6.2, with p21 = . . . = p24 = m2 . p1 3

p2

p3

1 4 2

p4

Fig. 6.2. On-shell box with two massive and two massless lines. The solid lines denote massive, the dotted lines massless particles

As in Example 6.8, we have changed the numbering of the lines with respect to Chap. 4. The procedure is again straightforward. One can identify the master integral with four lines, F (1, 1, 1, 1) = I1 , two symmetrical master integrals with

150

6 Reduction to Master Integrals by Baikov’s Method

three lines, F (1, 0, 1, 1) = I21 , F (0, 1, 1, 1) = I22 , two master integrals with two lines, F (1, 1, 0, 0) = I31 , F (0, 0, 1, 1) = I32 and two symmetrical master integrals with one line, F (1, 0, 0, 0) = I41 , F (0, 1, 0, 0) = I42 . These master integrals are graphically shown in Fig. 6.3. We have the following hierarchy relations: I41 , I42 < I31 < I1 and I32 < I21 , I22 < I1 .

I1

I2

I31

I32

I4

Fig. 6.3. Master integrals for Fig. 6.2

The coeﬃcient function c1 is trivial. The coeﬃcient function c21 can be constructed, using (6.13), ﬁrst in the case of n2 ≤ 0, where it can be obtained by an explicit integration. Then, for n2 > 0, one applies IBP to these auxiliary integrals, introduces an auxiliary master integral and mixes such a solution with c1 . To construct the coeﬃcient function of I31 , one uses a straightforward integration in the case n1 ≤ 0 and general n2 and, similarly, for n2 ≤ 0 and general n1 . In the case of n1,2 > 0, one can apply auxiliary IBP relations with the introduction of an auxiliary master integral for n1 = n2 = 1 which is cancelled when mixing the so constructed coeﬃcient function with c1 . In the cases of the master integrals I32 , I41 and I42 , we have a tower of three hierarchical master integrals. Still the case of I32 is quite similar to I31 and does not provide complications. To construct the coeﬃcient function of I42 one uses a straightforward integration over x1 , x3 , x4 in the case of n1 ≤ 0, n3 ≤ 0, and over x1 , x4 , x3 in the case of n1 ≤ 0, n4 ≤ 0. In the case of n1 ≤ 0, n3,4 > 0, one integrates over x1 and uses, for resulting integrals over x3 and x4 , auxiliary recurrence relations, with an introduction of a master integral for n3 = n4 = 1 which cancels when mixing with the coeﬃcient function c22 . Quite similarly, one can explicitly integrate over x3 or x4 when n3 ≤ 0 or/and n4 ≤ 0 and reduce resulting integrals. Finally, in the case of n1,3,4 > 0, one solves corresponding auxiliary IBP relations and introduces a master integral for n1 = n3 = n4 = 1 which cancels when mixing with the coeﬃcient function c1 . Here is an example of the reduction of massive boxes to the master integrals: F (2, 1, 1, 1) =

(d − 4)(4m2 − t) d−3 d−5 I I2 − 2 I32 + 1 4m2 − s 2m2 (4m2 − s)t m (4m2 − s)t (d − 4)(d − 2) I4 . − (6.56) 2(d − 5)m4 (4m2 − s)t

6.3 General Recipes. Complicated Examples

151

We shall consider another example with a tower of three hierarchical master integrals in the next section. The last example in this section is Example 6.10. Sunset diagrams of Fig. 3.12 with one zero mass and two equal non-zero masses at a general value of the external momentum squared. We are dealing with the following family of integrals: dd kdd l (2q·k)−a3 (2q·l)−a4 F6.10 (a) = , (k 2 − m2 )a1 (l2 − m2 )a2 [(q − k − l)2 ]a5

(6.57)

where a = (a1 , a2 , a3 , a4 , a5 ) with a3,4 ≤ 0. The strategy presented above reveals the following preliminary list of the master integrals: F (1, 1, 0, 0, 1) = I1 and F (1, 1, 0, 0, 0) = I2 . The coeﬃcient function c2 can be constructed using the strategy described above: for n5 ≤ 0, an integration in terms of gamma functions is used and, for n5 > 0, a simple recursion is applied. It turns out that one can use the package AIR [1] to solve the recurrence relations for the auxiliary parametric integrals (6.13) corresponding to c1 , z−nd dx3 dx4 [P1 (x3 , x4 )] , (6.58) f (n3 , n4 , nd ) = xn3 3 xn4 4 where z = (d − 4)/2 = −ε and P1 (x3 , x4 ) = m2 (x3 + x4 − 2q 2 )2 −(x3 − q 2 )(x4 − q 2 )(x3 + x4 − q 2 ) .

(6.59)

Remember that we have n3 , n4 ≤ 0 so that we can perform a useful change of variables, x3,4 = x3,4 + q 2 and deal with integrals in these variables where the basic polynomial looks simpler. When solving the corresponding IBP relations (together with the relation following from the identity P1z−nd = P1z−nd −1 P1 ) z−nd which is it is useful to apply Euler’s theorem to the factor [P1 (x3 , x4 )] 2 2 a homogeneous functions of the four variables, x3 , x4 , q , m (although it is clear that the resulting relation is nothing but a special combination of the IBP relations). A general solution to these relations is determined by the two auxiliary master integrals, f (0, 0, 0) and f (−1, 0, 0). Therefore, it is necessary to introduce an extra master integral, I¯1 = F (1, 1, −1, 0, 1). As a result, an algorithm for the evaluation of all the three coeﬃcient functions, c1 , c¯1 and c2 , can be constructed. The dependence on the auxiliary master integrals drops out in expressions for the coeﬃcient functions. We have, in particular,

1 (d − 3)m2 − (d − 2)q 2 I1 F (2, 1, 0, 0, 1) = 2 m (4m2 − q 2 ) 3 1 ¯ + (d − 2)I1 + (d − 2)I2 , (6.60) 2 2

152

6 Reduction to Master Integrals by Baikov’s Method

F (2, 1, −1, 0, 1) =

F (2, 1, 0, −1, 1) =

2 − 2(d − 3)m2 + (d − 1)q 2 I1 2 −q +3(d − 2)I¯1 + (d − 2)I2 . 4m2

4(d − 3)m4 − (d − 2)(q 2 )2 I1 − 3 2¯ 2 2 + (d − 2)q I1 − (d − 2)(2m − q )I2 . 2

(6.61)

1

m2 (4m2

q2 )

(6.62)

Let us consider, following [14], a more complicated example in a separate section.

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential Example 6.11. Two-loop Feynman integrals for the heavy quark potential corresponding to Fig. 6.4. 6

7

6

1

2

2

1

7

5 5 3

4 (A)

3

4 (B)

Fig. 6.4. Feynman diagrams corresponding to case A and case B. Wavy lines denote propagators for the static source

The numbering of the lines in Fig. 6.4 is changed as compared with Fig. 3.9 in order to take into account the symmetry. There are two classes of such Feynman integrals which we denote A and B: dd kdd l FA (a) = (k 2 )a1 (l2 )a2 [(k − q)2 ]a3 [(l − q)2 ]a4 [(k − l)2 ]a5 1 × , (6.63) (v·k)a6 (v·l)a7 dd kdd l FB (a) = (k 2 )a1 (l2 )a2 [(k − q)2 ]a3 [(l − q)2 ]a4 [(k − l)2 ]a5 1 × , (6.64) (v·k)a6 [v·(k − l)]a7 where v·q = 0.

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

153

The Feynman integrals necessary for the evaluation of the two-loop potential were calculated in [12]. In [13], a procedure for the evaluation of arbitrary integrals (6.63) and (6.64) was developed, using the technique of shifting dimension [15] discussed in Chap. 5. However, not all the necessary relations were published. Another version of partial calculation of integrals (6.63) and (6.64) was used in [9] for the evaluation of 1/m corrections to the two-loop potential. In this algorithm, IBP was used without systematization, as in Chap. 5, and the reduction always stopped at integrals expressed in terms of gamma functions so that a lot of boundary integrals, sometimes involving up to fourfold ﬁnite summations, entered the reduction. Now, we are going to apply the method of this chapter to these integrals. We will, therefore, obtain a minimal set of master integrals. The basic polynomials are straightforwardly obtained: PA (x1 , . . . , x7 ) = −[x2 x6 − x4 x6 + (−x1 + x3 )x7 ]2 +v 2 {x21 x4 + x3 (x22 + x2 (x3 − x4 − x5 ) + x4 x5 ) −x1 [x2 (x3 + x4 − x5 ) + x4 (x3 − x4 + x5 )]} +(q 2 )2 [v 2 x5 − (x6 − x7 )2 ] + q 2 {v 2 [(x3 + x4 − x5 )x5 +x2 (x3 − x4 + x5 ) + x1 (−x3 + x4 + x5 )] + 2[x2 x6 (−x6 + x7 ) + x4 x6 (−x6 + x7 ) +x7 (x1 x6 + x3 x6 − 2x5 x6 − x1 x7 − x3 x7 )]} , PB (x1 , . . . , x7 ) = PA (x1 , x2 , x3 , x4 , x5 , x6 , x6 − x7 ) .

(6.65) (6.66)

The two cases A and B are considered separately. Case A. The application of the procedures described above to case A leads to the following families of master integrals which are shown in Fig. 6.5. As far as the notation is concerned the ﬁrst index labels the diﬀerent master integrals. In case the master integrals are equal we introduce a second index for further speciﬁcation. If Ij is a master integral with indices 1 and 0 then we shall denote by I¯j the master integral which diﬀers from Ij by one index −1 instead of 0. – Family A1 consists of the four master integrals with the hierarchy I1 > {I21 , I22 } > I3 : I1 = FA (1, 1, 1, 1, 0, 1, 1) , I21 = FA (1, 1, 1, 1, 0, 0, 1) , I22 = FA (1, 1, 1, 1, 0, 1, 0) , I3 = FA (1, 1, 1, 1, 0, 0, 0) . – Family A2 consists of the four master integrals with the hierarchy I51 > {I71 , I81 } > I41 :

154

6 Reduction to Master Integrals by Baikov’s Method

Family A1: I1

I21

I22

I3

I51

I71

I81

I41

Family A2:

Family A4: I61

Fig. 6.5. Feynman diagrams corresponding to the master integrals of case A. In addition to I61 , there is also a master integral I¯61 containing an irreducible numerator

I51 = FA (1, 0, 0, 1, 1, 1, 1) , I71 = FA (1, 0, 0, 1, 1, 0, 1) , I81 = FA (1, 0, 0, 1, 1, 1, 0) , I41 = FA (1, 0, 0, 1, 1, 0, 0) . – Family A3 is symmetrical to Family A2 with respect to the transformation 1 ↔ 2, 3 ↔ 4, 6 ↔ 7. It contains the master integrals I52 , I72 , I82 and I42 . – Family A4 contains the master integrals I61 = FA (0, 1, 0, 1, 1, 1, 0) , I¯61 = FA (0, 1, 0, 1, 1, 1, −1) . – Family A5 is symmetrical to Family A4 with respect to the transformation 1 ↔ 2, 3 ↔ 4, 6 ↔ 7. It contains the master integrals I62 and I¯62 . As has already become clear from the examples discussed so far, one expects the appearance of complicated expressions for the coeﬃcient functions of simplest master integrals. Indeed, in the case of the coeﬃcient function c1 , six out of seven indices can be treated with the help of diﬀerentiations and the remaining one-dimensional integral can be understood in the sense of integration (6.31). The situation is similar for c22 (and c21 which can be obtained by exploiting the symmetry) where the remaining two-fold integration over x7 and x5 can be understood with the help of the integrals (6.40) and (6.26). To construct c3 we have to understand, in some way, three integrations, over x5 , x6 , x7 . In case one of the indices n5 , n6 or n7 is less or equal to zero one can use various combinations of the auxiliary integrals gi (i = 1, . . . , 4) listed above. Thereby it is advantageous to perform the integration corresponding to the negative index ﬁrst. If, on the contrary, n5 , n6 and n7 are

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

155

positive an immediate integration seems not to be possible. However, from the corresponding three-parametric integral representation it is simple to derive recurrence relations which shift at least one of the indices to zero, eventually at the cost of increasing the dimension nd . The latter does not constitute a problem since the whole formulation of our procedure is in d dimensions. Thus, also in this case the integration can be performed in terms of gamma functions. In principle one could be forced to introduce three auxiliary master integrals and build the proper linear combinations with c1 , c21 and c22 . However, it turns out that the corresponding constants in such combinations are zero. For the coeﬃcient function c51 , only two non-trivial integrations over x2 and x3 are involved which can be performed with the help of (6.32). For c81 , one can use the symmetry: c81 (a1 , a2 , a3 , a4 , a5 , a6 , a7 ) = c71 (a4 , a3 , a2 , a1 , a5 , a7 , a6 ) . The most complicated coeﬃcient function is certainly c41 since there are four non-trivial integrations over x2 , x3 , x6 and x7 left. If n6 or n7 are less than or equal to zero the integrations can be performed in terms of gamma functions with the help of the formulae provided above. However, for n6 ≥ 1 and n7 ≥ 1 this is not possible. In this case, the idea is to use IBP in order to reduce the four-parametric auxiliary integrals dx2 dx3 dx6 dx7 A,aux (n2 , n3 , n6 , n7 , nd ) = . . . I41 xn2 2 xn3 3 xn6 6 xn7 7 × [P41 (x2 , x3 , x6 , x7 )]

z−nd

(6.67)

(with z = (d − h − 1)/2 = (d − 5)/2) to the auxiliary master integral A,aux I41 (1, 1, 1, 1, 0). Here P41 is obtained from PA by setting x1 , x4 and x5 to zero. Observe that the corresponding recurrence procedure is signiﬁcantly simpler than the original one which involves seven denominators. Furthermore, if during the recursion either n6 or n7 becomes negative the corresponding expressions can immediately be expressed in terms of gamma functions. The A,aux (1, 1, 1, 1, 0) ﬁve IBP relations which are useful for the reduction to I41 can be obtained by either diﬀerentiating the integrand with respect to xi z−nd z−nd −1 (i = 2, 3, 6, 7) or by writing down the identity P41 = P41 P41 and inserting the explicit result for the last factor. The proper combination of these relations leads to new ones which allow the following steps to be performed in an automatic way: 1. 2. 3. 4. 5.

Reduce n6 and n7 to one. Reduce n2 , n3 > 0 to n2 , n3 ≤ 0. Use IBP recurrence relations to obtain n2 = n3 . Reduce n2 = n3 < 0 to n2 = n3 = 0. Adjust the dimension, i.e. reduce nd to zero.

A,aux A,aux A simple relation transforms I41 (0, 0, 1, 1, 0) to I41 (1, 1, 1, 1, 0).

156

6 Reduction to Master Integrals by Baikov’s Method

At this point one constructs the ﬁnal coeﬃcient function c41 by considering the linear combination with c51 , c71 and c81 . Since c41 (a71 ) = c41 (a81 ) = 0, we are left with c41 (a) = c041 (a) − c041 (a51 )c51 (a) ,

(6.68)

where 1 4(d − 3)(3d − 14)(3d − 10)(3d − 8) q2 v2 (d − 4)2 (3d − 13)(3d − 11) (d − 5)2 A,aux (q 2 )2 I41 (1, 1, 1, 1, 0) . + (3d − 13)(3d − 11)

c041 (a51 ) = −

A,aux In this combination the auxiliary master integral I41 (1, 1, 1, 1, 0) cancels and c41 (n) turns out to be a rational function in d. The master integral I61 forms a family by its own. However, as the polynomial P61 is quadratic in x7 and thus the corresponding recurrence relation shifts n7 only in steps of two, it is necessary to introduce in addition the master integral I¯61 where a7 = −1. The very calculation of the coeﬃcient function is identical for I61 and I¯61 . For n3 ≤ 0, it can be done in terms of gamma functions with the integration order x3 , x1 , x7 . On the other hand, for n3 > 0, a simple one-step relation reduces n3 to zero. Let us now turn to Case B. As one can see from (6.66) the basic polynomial is quite similar to the one of case A which can be used while computing the coeﬃcient functions. However, the symmetry can only be exploited if n7 ≤ 0 as for n7 > 0 the factor (x6 − x7 ) would appear in the denominator. Altogether there are four families which, however, show a more complicated structure than in case A – see Fig. 6.6. More precisely one has

– Family B1. There are twelve master integrals which obey the hierarchies I1B > {I2B , I22 } > I3 and I1B > I2B > {I6i , I¯6i } (i = 3, 4, 5, 6) and are given by I1B = FB (1, 1, 1, 1, 0, 1, 1) , I2B = FB (1, 1, 1, 1, 0, 0, 1) , I22 = FB (1, 1, 1, 1, 0, 1, 0) , I3 = FB (1, 1, 1, 1, 0, 0, 0) , I63 = FB (1, 1, 1, 0, 0, 0, 1) , I64 = FB (1, 1, 0, 1, 0, 0, 1) , I65 = FB (1, 0, 1, 1, 0, 0, 1) , I66 = FB (0, 1, 1, 1, 0, 0, 1) . There are four master integrals with a6 = −1:

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

157

Family B1: I1B

I2B

I22

I3

I63

I64

I65

I66

I9

I82

I81

I41

I53

I83

I72

I42

Family B2:

Family B3:

Family B4: I67

Fig. 6.6. Feynman diagrams corresponding to the master integrals of case B. In addition to I6i (i = 3, . . . , 7) there are also master integrals I¯6i containing irreducible numerators

I¯63 = FB (1, 1, 1, 0, 0, −1, 1) , I¯64 = FB (1, 1, 0, 1, 0, −1, 1) , I¯65 = FB (1, 0, 1, 1, 0, −1, 1) , I¯66 = FB (0, 1, 1, 1, 0, −1, 1) . – Family B2. There are four master integrals which obey the following hierarchy: I9 > {I82 , I81 } > I41 with I9 = FB (1, 0, 0, 1, 1, 1, 1) , I82 = FB (1, 0, 0, 1, 1, 0, 1) , I81 = FB (1, 0, 0, 1, 1, 1, 0) , I41 = FB (1, 0, 0, 1, 1, 0, 0) . – Family B3. Similarly to Family B2, there are four master integrals obeying the hierarchy I53 > {I83 , I72 } > I42 with I53 = FB (0, 1, 1, 0, 1, 1, 1) , I83 = FB (0, 1, 1, 0, 1, 0, 1) ,

158

6 Reduction to Master Integrals by Baikov’s Method

I72 = FB (0, 1, 1, 0, 1, 1, 0) , I42 = FB (0, 1, 1, 0, 1, 0, 0) . – Family B4 consists of the two master integrals I67 = FB (0, 1, 0, 1, 1, 1, 0) , I¯67 = FB (0, 1, 0, 1, 1, 1, −1) . It is similar to the Families A4 and A5 of case A. B The construction of the coeﬃcient functions cB 1 , c2 and c22 of the family B1 proceeds along the same lines as in case A. In the case of c3 , we have to deal with integrals I3B,aux (n5 , n6 , n7 , nd ) which are deﬁned similarly to (6.67). There is a slight complication as, in contrast to case A, c3 (a1 ) = 0. As a consequence an auxiliary master integral, I3B,aux (0, 1, 1, 0), has to be introduced which is only cancelled after considering the proper linear combination with c1 . The reduction to I3B,aux (0, 1, 1, 0) is straightforward. Family B1 has four more members, I63 , I64 , I65 and I66 , which belong to the four hierarchies I1B > I2B > I6i (i = 3, 4, 5, 6). Thus, in order to obtain the coeﬃcient functions c6i one has to consider the linear combination B 0 B B c6i = c06i − c06i (aB 1 )c1 (a) − c6i (a2 )c2 (a) .

(6.69)

Let us in the following restrict the discussion to c63 since the results for the other three coeﬃcients can be obtained by exploiting the symmetry. The corresponding auxiliary integrals are given by an integral representation of the form z−nd dx4 dx5 dx6 , (6.70) c063 ∼ . . . [P63 (x4 , x5 , x6 )] xn4 4 xn5 5 xn6 6 with

P63 = (q 2 )2 v 2 x5 + q 2 v 2 x4 x5 − x25 − 4q 2 x5 x26 − x24 x26 .

(6.71)

B For n4 ≤ 0, where we have cB 1 (n) = c2 (n) = 0, the integrals in (6.70) can be taken analytically in the order x4 , x5 , x6 using (6.40) for x4 , the formula (6.40) for x5 and (6.26) extended to non-integer k3 for x6 . Let n4 > 0. Then we need to introduce two auxiliary master integrals, B,aux B,aux (1, 0, 0, 0) and I63 (1, 0, 1, 0). The reduction of the auxiliary paraI63 metric integrals (6.70) can be performed as follows:

1. Reduce n4 to one. 2. Reduce n5 to zero. 3. The reduction of n6 can only be performed in steps of two. Thus one ends up with n6 = 0 or n6 = −1. 4. Adjust the dimension, i.e. reduce nd to zero. The corresponding recurrence relations are derived easily from (6.71). It B,aux is interesting to note that in (6.69) the master integral I63 (1, 0, 1, 0) is B,aux B B cancelled from c1 and I63 (1, 0, 0, 0) from c2 . Observe that, due to the

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

159

structure of the reduced polynomial (6.71), in addition to I63 also a master integral with n6 = −1, I¯63 , has to be introduced which, however, has the same coeﬃcient function as I63 . Observe also that, for c63 and c65 , the master integrals I6 and I¯6 are needed, while for c64 and c66 , the integrals I6 and I¯6B are necessary. Families B2, B3 and B4 are similar to the families A2, A3 and A4, respectively, so that the corresponding coeﬃcient functions are similarly constructed. The procedure described above was implemented in a MATHEMATICA package [14]. Let us now list all occurring master integrals in both cases A and B. They have been obtained with the help of the program package developed for the calculation performed in [9] where IBP recurrence relations have been ‘nonsystematically’ solved. d/2 2 iπ π Γ (5/2 − d/2)2 Γ (d/2 − 3/2)4 , I1 = Q2+4ε v 2 Γ (d − 3)2 d/2 2 √ iπ π Γ (2 − d/2)Γ (5/2 − d/2)Γ (d/2 − 1)2 Γ (d/2 − 3/2)2 I2 = − , Q1+4ε v Γ (d − 3)Γ (d − 2) 2 Γ (2 − d/2)2 Γ (d/2 − 1)4 I3 = iπ d/2 , Q4ε Γ (d − 2)2 2 Γ (3 − d)Γ (d/2 − 1)3 , I4 = − iπ d/2 Q2−4ε Γ (3d/2 − 3) 2 π 2 e−2γE ε 7 2 2 d/2 2 I5 = iπ − − 4 + −24 + π ε + O(ε ) , Q4ε v 2 3ε 9 2 √πQ1−4ε I6 = iπ d/2 v 2d−2 Γ (3 − d)Γ (7/2 − d)Γ (d/2 − 1)Γ (d − 5/2)2 , × Γ (2 − d/2)Γ (2d − 5) 2 √ 2d−2 Γ (3 − d)2 Γ (d/2 − 1)Γ (d − 2)2 I¯6 = − iπ d/2 , πQ2−4ε Γ (3/2 − d/2)Γ (2d − 4) √ 2 πQ1−4ε I7 = iπ d/2 v Γ (7/2 − d)Γ (d/2 − 1)2 Γ (d/2 − 3/2)Γ (d − 5/2) , × Γ (d − 2)Γ (3d/2 − 4) I8 = I7 , I9 = I5 , 1 I1B = I1 , 2 2 π 2 e−2γE ε I2B = iπ d/2 Q1+4ε v

160

6 Reduction to Master Integrals by Baikov’s Method

5 2 π − 16 ln 2 − 4 ln2 2 + O(ε2 ) , × −4 ln 2 + ε 3 B ¯ ¯ I6 = −I6 , where Q = −q 2 . The fact that I5 = I9 and I7 = I8 can be seen immediately by a simple change of the loop momenta. Since I7 = I8 , we have in both cases one master integral less. So, in case A, we have eight master integrals, I1 , . . . , I7 and I¯6 , and, in case B, ten master integrals I2 , . . . , I7 , I9 , I¯6B , I1B and I2B , Only two of the master integrals are not known in terms of gamma functions. Their results are given in expansion in ε up to ε1 . For example, they can be evaluated by the method of MB representation described in Chap. 4. Here are some examples of results for the coeﬃcient functions: FA (2, 2, 1, 1, 1, 1, 1) = c1 I1 + c3 I3 + (c41 + c42 )I4 + (c51 + c52 )I5 +(c61 + c62 )I¯6 d/2 2 iπ 2 4 2 16 368 2 + π − + π − 8ζ(3) + O(ε) , = 8+4ε 2 Q v 3ε 3ε 9 45 with 2(d − 5)(d − 4) 8(d − 5)(d − 3)2 , c = , 3 q6 (d − 4)q 8 v 2 −3(d − 3)(3d − 16)(3d − 14)(3d − 10)(3d − 8) = c42 = (d − 9)(d − 8)(d − 7)(d − 6)2 (d − 4)2 q 10 v 2

c1 = c41

c51 = c52 c61 = c62

×(5d3 − 93d2 + 588d − 1264) , −3(3d − 17)(3d − 13)(3d − 11) = , (d − 9)(d − 7)q 8 −32(2d − 13)(2d − 11)(2d − 9)(2d − 7)(2d − 5) = . (d − 9)(d − 7)(d − 6)(d − 4)q 10 v 2

FB (2, 2, 1, 1, 1, 1, 1) = c1 I1B + c3 I3 + (c41 + c42 )I4 + c53 I5 +(c63 + c65 )I¯6 + (c64 + c66 + c67 )I¯6B + c9 I9 d/2 2 iπ 4 2 8 368 2 1 π + 4ζ(3) + O(ε) , = 8+4ε 2 − + π + + Q v 3ε 3ε 9 45 with 2(d − 5)(d − 4) −4(d − 5)(d − 3)2 , c3 = , 6 q (d − 4)q 8 v 2 3(d − 3)(3d − 16)(3d − 14)(3d − 10)(3d − 8) = (d − 9)(d − 8)(d − 7)(d − 6)2 (d − 4)2 q 10 v 2

cB 1 = c41

c42

×(7d3 − 117d2 + 654d − 1232) , −6(d − 3)(3d − 16)(3d − 14)(3d − 10)(3d − 8) = (d − 9)(d − 8)(d − 7)(d − 6)2 (d − 4)2 q 10 v 2

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

c53 c63

×(d3 − 12d2 + 33d + 16) , −3(3d − 17)(3d − 13)(3d − 11) = , (d − 9)(d − 7)q 8 4(2d − 7)(2d − 5) = c64 = − (d − 9)(d − 7)(d − 6)(d − 4)q 10 v 2

×(15d4 − 304d3 + 2240d2 − 7093d + 8118) , 4(2d − 7)(2d − 5)(d2 − 17d + 55) , c65 = c66 = (d − 7)(d − 4)q 10 v 2 −32(2d − 13)(2d − 11)(2d − 9)(2d − 7)(2d − 5) c67 = , (d − 9)(d − 7)(d − 6)(d − 4)q 10 v 2 −3(3d − 17)(3d − 13)(3d − 11) c9 = . (d − 9)(d − 7)q 8 FA (1, 1, 2, 1, 1, −1, 1) = c3 I3 + (c41 + c42 )I4 + c62 I¯6 d/2 2 iπ 3 1 = − + − 2ζ(3) + O(ε) , Q4+4ε 2ε 2 2(d − 3) −3(3d − 10)(3d − 8)(d2 − 5d + 2) , c41 = , 4 (d − 4)q 2(d − 6)(d − 5)(d − 4)2 q 6 3(d − 5)(d − 2)(3d − 10)(3d − 8) = , 2(d − 6)(d − 4)2 q 6 4(2d − 9)(2d − 7)(2d − 5) = . (d − 5)(d − 4)q 6

c3 = c42 c62

FB (1, 1, 2, 1, 1, −1, 1) = c3 I3 + (c41 + c42 )I4 + (c63 + c65 )I¯6 +(c64 + c66 )I¯6B d/2 2 iπ 1 1 = − + + O(ε) , Q4+4ε 2ε 2 (d − 5)(d − 3) −3(3d − 10)(3d − 8)(d2 − 9d + 22) , c = , 41 (d − 6)q 4 2(d − 6)2 (d − 5)(d − 4)q 6 3(3d − 10)(3d − 8)(d2 − 11d + 26) = , 2(d − 6)2 (d − 4)q 6 (2d − 11)(2d − 7)(2d − 5) = , (d − 6)(d − 5)q 6 −(2d − 7)(2d − 5) (2d − 7)2 (2d − 5) = , c = , 65 (d − 6)(d − 5)q 6 (d − 6)(d − 5)q 6 −(2d − 7)(2d − 5)(4d − 19) = . (d − 6)(d − 5)q 6

c3 = c42 c63 c64 c66

161

162

6 Reduction to Master Integrals by Baikov’s Method

6.5 Conclusion Let us observe that since a given problem of solving IBP relations is always reduced, in the present method, to the corresponding problem for vacuum Feynman integrals, it turns out that diﬀerent initial problems can have the same vacuum ‘image’. As it was demonstrated in [4], this property can be used when a solution of some reduction problem is known and another reduction problem has the same vacuum image with it. For example, solving IBP relations for the two-loop massless vertex diagrams (of Fig. 5.3, Fig. 3.13 and a Mercedez-Benz type) can be reduced to solving IBP relations for the three-loop propagator diagrams that was done in [7] and implemented in [8]. The method of this chapter has a feature opposite to the method of shifting dimension [15] discussed in Chap. 5. Indeed, the ﬁrst point in the latter is to get rid of numerators, with the primary idea to simplify the situation. In contrast to this, the numerators play a crucial role in the present method: each irreducible numerator results in an integration over the corresponding xvariable in the basic parametric representation. One more diﬀerence of these two methods is that master integrals with indices ai > 1 usually appear in a reduction with shifting dimension, while there are no such master integrals in the present method. (The same feature holds for the modern realization of the method of diﬀerential equations to be discussed in the next chapter.) On the other hand, shifting dimension is also an intrinsic feature of the present method because the dimension d enters the basic representation in a very simple way and it is necessary to put the shift of dimension under control when solving the auxiliary IBP relations. The method of this chapter was successfully applied, due to the reduction presented in Sect. 6.3, in [10], where various two-loop diagrams associated with the two-loop quark potential were necessary. A breakthrough in another direction – the evaluation of general four-loop propagator diagrams (i.e. one loop above [7]!) was also achieved with its help [3]. Another branch of this method was, however, applied there. It is based on an expansion at large d which is somehow introduced when constructing the coeﬃcient function of the master integrals starting from (6.9). Unfortunately, no details of this branch have been published up to now. This method is now at the level of experimental mathematics, as well as many other techniques discussed in this book. One tries to follow the prescriptions formulated in this chapter and, hopefully, arrives at a solution of a given reduction problem. One always believes in the rational dependence of the coeﬃcient functions on everything, and this is one of possible consistency checks. The validity of the reduction so obtained can be checked by explicit evaluation of various Feynman integrals of the given class. On the other hand, one can check that the initial IBP equations are satisﬁed for the so constructed coeﬃcient functions. Anyway, after successful checks, one can conclude that the obtained solution of the IBP relations is valid and apply it for practical purposes.

References

163

I hope, however, that this method can be put on a solid mathematical ground and, moreover, some interesting mathematics is behind it.

References 1. C. Anastasiou and A. Lazopoulos, JHEP 0407 (2004) 046. 148, 151 2. P.A. Baikov, Phys. Lett. B 385 (1996) 404; Nucl. Instrum. Methods A 389 (1997) 347. 133, 134, 135, 136, 137 3. P.A. Baikov, K.G. Chetyrkin, and J.H. K¨ uhn, Phys. Rev. Lett. 88 (2002) 012001; Phys. Rev. D 67 (2003) 074026; Phys. Lett. B 559 (2003) 245; hepph/0311137. 162 4. P.A. Baikov and V.A. Smirnov, Phys. Lett. B 477 (2000) 367. 133, 136, 137, 162 5. P.A. Baikov and M. Steinhauser, Comput. Phys. Commun. 115 (1998) 161. 133 6. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 147 7. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 133, 162 8. S.G. Gorishny, S.A. Larin, L.R. Surguladze and F.V. Tkachov, Comput. Phys. Commun. 55 (1989) 381; S.A. Larin, F.V. Tkachov and J.A.M. Vermaseren, Preprint NIKHEF-H/91-18 (Amsterdam 1991). 162 9. B.A. Kniehl, A.A. Penin, V.A. Smirnov, and M. Steinhauser, Phys. Rev. D 65 (2002) 091503. 153, 159 10. B.A. Kniehl, A.A. Penin, V.A. Smirnov and M. Steinhauser, Nucl. Phys. B 635 (2002) 357; Phys. Rev. Lett. 90 (2003) 212001; B.A. Kniehl, A.A. Penin, A. Pineda, V.A. Smirnov and M. Steinhauser, Phys. Rev. Lett. 92 (2004) 242001; A.A. Penin, A. Pineda, V.A. Smirnov and M. Steinhauser, hep-ph/ 0403080, hep-ph/0406175. 162 11. S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 148 12. M. Peter, Phys. Rev. Lett. 78 (1997) 602; Nucl. Phys. B 501 (1997) 471. 153 13. Y. Schr¨ oder, Phys. Lett. B 447 (1999) 321. 153 14. V.A. Smirnov and M. Steinhauser, Nucl. Phys. B 672 (2003) 199. 133, 134, 152, 159 15. O.V. Tarasov, Nucl. Phys. B 480 (1996) 397; Phys. Rev. D 54 (1996) 6479. 153, 162

7 Evaluation by Diﬀerential Equations

The method of diﬀerential equations (DE) suggested in [20] and developed in [23] and later works (see references below) is a method of evaluating individual Feynman integrals. We have agreed that, at the present level of complexity of unsolved important problems, it looks unavoidable to decompose the problem of evaluating Feynman integrals of a given family into the reduction to some master integrals and the problem of evaluating these master integrals. Thus, this basic method is oriented at the evaluation of the master integrals. Moreover, in contrast to other methods of evaluating individual Feynman integrals, it is assumed within this method that a solution of the reduction problem is already known. The idea is to take some derivatives of a given master integral with respect to kinematical invariants and masses. Then the result of this diﬀerentiation is written in terms of Feynman integrals of the given family and, according to the known reduction, in terms of the master integrals. Therefore, one obtains a system of diﬀerential equations for the master integrals which can be solved with appropriate boundary conditions. To illustrate basic recipes of this method we shall consider only four examples. The fact is that, for complicated examples, all the calculations can be done only on a computer and intermediate formulae usually happen to be very cumbersome. We shall consider typical one-loop examples in Sect. 7.1 and a two-loop characteristic example in Sect. 7.2. The status of the method, i.e. its perspectives and open problems will be discussed in Sect. 7.3. together with a brief review of its applications.

7.1 One-Loop Examples Of course, we start with our favourite example. Example 7.1. One-loop propagator diagram corresponding to Fig. 1.1. After solving the corresponding reduction problem in Chaps. 5 and 6, we know that there are two master integrals, F (1, 1) = I1 and F (1, 0) = I2 . The second one is a simple one-scale integral given by the right-hand side of V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 165–177 (2004) c Springer-Verlag Berlin Heidelberg 2004

166

7 Evaluation by Diﬀerential Equations

(5.6). We have started to evaluate I1 in Chap. 1, by diﬀerentiating in m2 and arrived at the equation (1.20) for f (m2 ) = F (1, 1). To be very pedantic, let us rewrite it in terms of our true master integrals, ∂ 1 1−ε 2 2 f (m ) = ) − I (1 − 2ε)f (m , (7.1) 2 ∂m2 m2 − q 2 m2 although this does not make an essential diﬀerence here. Let us turn to the new function by f (m2 ) = iπ d/2 (m2 )−ε y(m2 ). We obtain the following diﬀerential equation for it: y −

Γ (ε) m2 (1 − ε) − εq 2 y=− 2 . 2 2 2 m (m − q ) m − q2

(7.2)

It can be solved by the method of the variation of the constant. The general solution to the corresponding homogeneous equation, with a zero on the right-hand side of (7.2), is y(m2 ) = C(m2 − q 2 )1−2ε (m2 )−ε .

(7.3)

Then we make C = C(m2 ) dependent on m2 , solve this equation and obtain " # m2 dx x−ε 2 d/2 2 2 1−2ε + C1 , −Γ (ε) (7.4) f (m ) = iπ (m − q ) (x − q 2 )2−2ε 0 where the constant C1 can be determined from the boundary value f (0) which is a massless one-loop diagram evaluated by means of (A.7). This gives f (m2 ) = −iπ d/2 (m2 − q 2 )1−2ε Γ (ε) # " 2 m dx x−ε Γ (1 − ε)2 − . × (x − q 2 )2−2ε Γ (2 − 2ε)(−q 2 )1−ε 0

(7.5)

If we turn to expansion in ε and take terms up to ε0 into account we shall reproduce (1.7). The next example is also an old one. Example 7.2. The triangle diagram of Fig. 3.4. The reduction problem was solved in Examples 5.4 and 6.7. The only master integral that is not expressed in terms of gamma functions for general d is F (1, 1, 1) = I1 = f (m2 ). We have already calculated it in Examples 3.2 and 4.2. Let us now do this by DE. As in the previous example, we take the ∂ 2 derivative ∂m 2 f (m ) and obtain F (1, 1, 2) for which we apply the relation (6.54), according to our reduction procedure. Let us again, as above, conﬁne ourselves to the evaluation up to the ﬁnite part in ε. Then the ﬁrst term on the right-hand side of (6.54) is irrelevant because it is proportional to ε. So, we obtain, at ε = 0, 2 2 ∂ 2 2 ln(m /Q ) . f (m ) = iπ ∂m2 m2 (m2 − Q2 )

(7.6)

7.1 One-Loop Examples

167

Thus, the evaluation of I1 at d = 4 reduces to taking an integral of the righthand side of (7.6). The boundary condition is simple: this function vanishes in the large mass limit. This can be seen, for example, by examining this behaviour using the MB representation (4.7) as explained in Sect. 4.8. (To do this, one takes a residue at the point z = −1.) Consequently, the known result (3.21) is once again reproduced. If one needs to evaluate I1 at general ε, or obtain higher terms of expansion in ε by DE, one can start from (6.54) and solve the so-obtained diﬀerential equation, applying the method of the variation of the constant quite similarly to Example 7.1. Let us now turn, following [8], to Example 7.3. The on-shell box diagram with two massive and two massless lines shown in Fig. 6.2, with p21 = . . . = p24 = m2 . These are functions of the three variables s, t and m2 . The following combinations naturally arise in the problem: √ √ √ √ 4m2 − s − −s 4m2 − t − −t , y=√ . (7.7) x= √ √ √ 4m2 − s + −s 4m2 − t + −t We again assume that we know a solution of the corresponding reduction problem. It was brieﬂy described in Example 6.9. The reduction based on the algorithm of [16, 21, 22] which was discussed in Sect. 5.4 also leads [8] to the same family of the master integrals shown in Fig. 6.3: I1 = F (1, 1, 1, 1), I2 = F (1, 0, 1, 1) = F (0, 1, 1, 1), I31 = F (1, 1, 0, 0), I32 = F (0, 0, 1, 1) and I4 = F (1, 0, 0, 0) = F (0, 1, 0, 0), where I2 and I4 are present in two copies. Suppose that we want to evaluate I1 by DE. Therefore, we assume that all the master integrals with the number of lines less than four are already known. The integrals I4 and I32 are given by (2.44) and (3.8). The value of the master integral I31 = F (1, 1, 0, 0) is very well-known and can be obtained by various methods. To be self-consistent, let us observe that one can apply MB representation (4.28), set a1 = a2 = 1, a3 = 0 and evaluate this integral by closing the integration contour and summing up the resulting series. Within the method of DE, it is important to present this and later results in terms of the variables (7.7): 1 1 iπ d/2 e−γE ε 1 +2−2 − (7.8) I31 = H0 (x) + O(ε) . (m2 )ε ε 2 1−x Here and in subsequent formulae, usual logarithms and polylogarithms are written in terms of HPL [25] – see Appendix B. Moreover, it is necessary to rewrite the quantity q 2 in (3.8) in terms of these variables, i.e. make the substitution q 2 → t → −(1−y)2 /(m2 y) in the factor (−q 2 )ε and then expand it in ε. Finally, we need I2 which can be obtained using (4.29) at a1 = a2 = a4 = 1 and evaluating this integral by closing the integration contour to the right.

168

7 Evaluation by Diﬀerential Equations

In [8], this result was obtained by DE. It is also naturally written in terms of the variables (7.7): 1 1 2 2 iπ 2 − π I2 = + H (y) + 2H (y) + O(ε) . (7.9) 0,0 0,1 2m2 1 + y 1 − y 3 Observe that higher terms of this and other expansions in ε can be found in [8]. The starting point is to take derivatives in s or t and write them down as a linear combination of integrals of the given class. In order to do this, one observes that taking derivatives in the external momenta reduces to taking derivatives in s and t: ∂ ∂sr ∂ = pi · , ∂pj ∂p j ∂sr r=1 6

pi ·

(7.10)

where si = p2i , i = 1, 2, 3, 4, are invariants with the on-shell condition, si = m2 , and s5 = s, s6 = t. This linear system of six equations can easily be solved, i.e. the derivatives ∂/∂sr can be expressed linearly in terms of the derivatives pi · ∂/∂pj with i, j = 1, 2, 3 – see [8]. One can use here the following expressions [12] which are equivalent to that of [8] due to the on-shell conditions: 1 ∂ s ∂ = (p2 − p3 ) · s , (7.11) p1 + p2 − ∂s 2 4m2 − s − t ∂p2 1 t ∂ ∂ (p2 − p3 ) · . (7.12) t = p3 − p1 − ∂t 2 4m2 − s − t ∂p3 So, we take partial derivatives of I1 = f (s, t) with respect to s and t, using (7.11) and (7.12), and obtain, on the right-hand side, a linear combination of integrals corresponding to Fig. 6.2. Every integral can be written in terms of the master integrals, according to the reduction procedure, and we obtain 1 1 d−5 d−4 ∂f =− + − (7.13) f + g1 , ∂s 2 s 4m2 − s 4m2 − s − t 1 d−6 d−4 ∂f = + (7.14) f + g2 , ∂t 2 t 4m2 − s − t where

# 4m2 − t) 1 1 − + I2 g1 = −(d − 4) 4m2 s 4m2 t(4m2 − s) t(4m2 − s − t) " # 1 2(d − 3) 1 1 − + + I31 t (4m2 − s)2 t(4m2 − s) t(4m2 − s − t) " # 1 d−3 1 + − 2 I32 2m − t s 4m2 − s "

7.1 One-Loop Examples

"

169

#

1 1 d−2 1 − I4 , + 2 2 2 2 2 m t (4m − s) t(4m − s) t(4m − s − t) " # 1 d−4 1 + I2 g2 = − 2 4m − s t 4m2 − s − t # " 1 2(d − 3) 1 + I31 − (4m2 − s)2 t 4m2 − s − t " # 1 d−2 1 + − 2 I4 . m (4m2 − s)2 t 4m2 − s − t +

(7.15)

(7.16)

It is suﬃcient to use one of the two equations to evaluate f (s, t). Let it be (7.13). Then (7.14) can be used for a non-trivial check. One needs also a boundary condition when solving (7.13): it can be obtained using the fact that the function f (s, t) is regular at s = 0. Multiplying (7.13) by s and taking the limit s → 0 one obtains d−3 d−4 I2 + 2 I32 . 2m2 m t Equation (7.13) can be solved in a Laurent expansion in ε, fj (s, t)εj . f (s, t) =

f (0, t) = −

(7.17)

(7.18)

j=−1

As a result, one obtains a set of nested diﬀerential equations from (7.13), dfj 1 1 1 =− + (7.19) fj + hj , ds 2 s 4m2 − s where the functions hj involve, in addition to the corresponding term of the expansion of the function g1 , a piece coming from fj−1 . These equations can be solved by the method of the variation of the constant. The homogeneous equation corresponding to (7.19), which is the same for all fj , takes the following form in the new variables, x and y: d 1 1 1 − + − (7.20) f (0) (x) = 0 , dx x 1 + x 1 − x with the solution f (0) (x) =

x . (1 − x)(1 + x)

(7.21)

Then the solution of the j-th diﬀerential equation in (7.19) can be written as

fj (x, y) = f

(0)

(x) Aj +

hj (x, y) dx (0) f (x)

,

(7.22)

where Aj is a constant which can be ﬁxed by imposing the boundary condition (7.17) expanded in ε.

170

7 Evaluation by Diﬀerential Equations

Observe that the combinations of the kinematical invariants involved on the right-hand side of (7.13) and (7.15) and, therefore, present in hj can be represented as 4m2 − s = m2

(1 + x)2 , x

4m2 − s − t = m2

(x + y)(1 + xy) . xy

(7.23)

After that the integration in (7.22), order by order in ε, becomes straightforward. All the quantities are prepared in such a form that the integration is taken in terms of HPL of the next level, also of the arguments x and y. So, one arrives at (4.27). However, keeping in mind that this very master integral can be needed when evaluating other master integrals in two loops, also by the method of DE, it is reasonable to present it in the same form as its ingredients were presented: 1 1 1 1 iπ d/2 e−γE ε I1 = − − H0 (x) (m2 )2+ε 1 + x 1 − x 1 − y (1 − y)2 1 + H0 (y) + 2H1 (y) + O(ε) . (7.24) × ε Further terms of this expansion in ε can be found in [8].

7.2 Two-Loop Example We turn again to Feynman integrals considered in Examples 4.10 and 6.10. Example 7.4. Sunset diagram of Fig. 3.12 with one zero mass and two equal non-zero masses at a general value of the external momentum squared. The general Feynman integral of this class is given by (6.57), so that there are two irreducible numerators in the problem. According to Example 6.10, we know a solution of the reduction problem, and that there are three master integrals, I1 = F (1, 1, 0, 0, 1), I¯1 = F (1, 1, −1, 0, 1) and I2 = F (1, 1, 0, 0, 0). The last of them is the square of the massive tadpole given by the right-hand side of (2.44). Let us now evaluate I1 and I¯1 by DE. For convenience, let us use, instead of I¯1 , the integral with a1 = a2 = a5 = 1 and the numerator equal to the product of the momenta (ﬂowing in the same direction) of the massless and one of the massive lines, 1 2 q I1 − I¯1 − I2 . (7.25) I˜1 = 2 We start with taking derivatives. We use the homogeneity of the integrals I1 and I˜1 with respect to q 2 and m2 , with the help of Euler’s theorem, set q 2 = s and obtain ∂ sf (s) = (1 − 2ε)f (s) − f (s) , (7.26) ∂m2 ∂ ˜ sf˜ (s) = 2(1 − ε)f˜ (s) − f (s) , (7.27) ∂m2

7.2 Two-Loop Example

171

where f (s) = I1 and f˜(s) = I˜1 , and we have already put m2 = 1 after diﬀerentiating with respect to the mass which results in indices equal to 2 instead of 1 on one of the massive lines. We apply (6.60)–(6.62) to these integrals with the indices equal to two in order to obtain only the master integrals on the right-hand side. Therefore, we arrive at the following diﬀerential equations for the functions f (s) and f˜(s): sf (s) =

1 [(3s − 2 − 4ε(s − 1)) f (s) s−4 $ +4(ε − 1)(h(s) + 3f˜(s)) ,

(7.28) $ 1 (7.29) sf˜ (s) = (ε − 1) h(s) − sf (s) + 2f˜(s) , 2 where h originates from I2 . As in the previous example, it is convenient to turn to the new variable x given by (7.7), or, vice versa, '

(1 − x)2 . x Then we obtain the following equations:

1 3 − 4x + 3x2 − 4ε(1 − x + x2 ) f (x) f (x) = x(x2 − 1) $ −4(ε − 1)x(h(x) + 3f˜(x)) ,

(7.31)

1 (ε − 1)(1 + x) 2x2 (x − 1) ' $ × (x − 1)2 f (x) + x(h(x) + 2f˜(x)) .

(7.32)

s=−

f˜ (x) =

(7.30)

The second function f˜(x) can be eliminated from this system in order to obtain a separate equation for the ﬁrst one: (3ε(x − 1)2 + 6x − 2) f (x) x(x2 − 1) 2(ε − 1)2 (2ε − 1)(2x + ε(1 − 4x + x2 )) f (x) + h(x) = 0 . + 2 2 x (x − 1) x(x − 1)2

f (x) +

(7.33)

Then we turn to solving this equation in expansion in ε, as in the previous examples, f−2 (x) f−1 (x) + f0 (x) + . . . . + (7.34) ε2 ε As usual, we need a general solution of the corresponding homogeneous equation at ε = 0: 2(3x − 1) 2 f (x) − f (x) + f (x) = 0 . (7.35) x(x2 − 1) x(x − 1)2 f (x) =

172

7 Evaluation by Diﬀerential Equations

Two independent solutions are 1 − x + x2 , (x − 1)2 4x(1 − x + x2 )H0 (x) − 1 + 7x − 3x2 − x3 + x4 , φ2 (x) = x(x − 1)2

φ1 (x) =

(7.36) (7.37)

with the Wronskian w(x) =

(x + 1)4 . x2 (x − 1)2

(7.38)

The solutions are presented in a form similar to the previous example, in terms of HPL. The equation for f−2 has the inhomogeneous term r−2 (x) = −

2 . x(x − 1)2

Its solution is written as φ2 (x)r−2 (x) f−2 (x) = c1 − dx φ1 (x) w(x) φ1 (x)r−2 (x) + c2 + dx φ2 (x) , w(x)

(7.39)

(7.40)

where c1 and c2 are integration constants. We obtain

1 x(c1 (1 − x + x2 ) − x) − c2 (1 − 7x + 3x2 + x3 − x4 ) f−2 (x) = 2 x(x − 1) +4c2 x(1 − x + x2 )H0 (x) . (7.41) The integration constants are evaluated from the regular behaviour of the √ solution at x → 0 so that 1/x and x in the asymptotic expansion of (7.41) are forbidden. This gives the values c1 = 1 and c2 = 0, with f−2 (x) = 1 .

(7.42)

The inhomogeneous term for f1 (x) is r−1 (x) =

1 − 8x + x2 . x2 (x − 1)2

(7.43)

Proceeding in a similar way we obtain the following solution:

1 1 − 6x − x2 − 2x3 + 2c1 x(1 − x + x2 ) f−1 (x) = 2x(x − 1)2

−2c2 (1 − 7x + 3x2 + x3 − x4 ) + 2(4c2 − 1)x(1 − x + x2 )H0 (x) . (7.44)

The regularity condition at x = 0 gives c1 = 13/4 and c2 = 1/4, with f−1 (x) =

1 + 10x + x2 . 4x

(7.45)

7.3 Conclusion

173

Finally, for f0 , we have the inhomogeneous term r0 (x) = −

3 − 9x + 2(48 + π 2 )x2 − 9x3 + 3x4 . 6x3 (x − 1)2

(7.46)

Similarly, we obtain the following solution:

1 (x − 1)2 (39 + 66x + 4π 2 x + 39x2 ) f0 (x) = 2 24x(x − 1)

+12(1 − 4x + 4x3 − x4 )H0 (x) − 48x(1 − x + x2 )H0,0 (x) . (7.47)

The second function f˜−2 (x) f˜−1 (x) f˜ = + f˜0 (x) + . . . . + (7.48) ε2 ε can be now obtained in a pure algebraic way, with the following results: 2

1+x , f˜−2 (x) = − 4x 1 + 11x + 11x3 + x4 f˜−1 (x) = − , 24x2

1 −(x − 1)2 (2π 2 − 11)x(1 + x2 ) f˜0 (x) = 2 2 48x (x − 1) +13(1 + x4 ) + 44x2 − 4 1 − 9x(1 − x2 )(1 − x + x2 ) − x6 H0 (x) +24x(1 − 2x + 4x2 − 2x3 + x4 )H0,0 (x) . (7.49) The corresponding result for the master integral I¯1 can be obtained easily from (7.42), (7.44), (7.47) and (7.49), using (7.25). It can be evaluated also using the onefold MB representation (4.76) (with another choice of the numerator). These results are in agreement with [11, 13], where another choice of the master integrals was used (with higher powers of the propagators, instead of integrals with numerators).

7.3 Conclusion At ﬁrst sight, the method of DE cannot be applied to integrals dependent on one scale since the dependence on the only scale parameter is trivial and can be obtained immediately by power counting. However, one can introduce, for a one-scale integral, an additional scale parameter, apply the corresponding diﬀerential equation, get the boundary condition at a diﬀerent, more suitable point and then return to the single scale value. An example of this strategy can be found in [5]. I admit that it might seem, from the previous examples1 , that the method of DE is not optimal. In particular, the results for Example 7.4 can be, probably, derived by MB representation in a simpler way. However, the method of 1

Simple instructive examples can be found also in the review [1].

174

7 Evaluation by Diﬀerential Equations

DE is indeed very powerful and, in some situations, the very best one. An important feature of the strategy outlined above is that it can straightforwardly be generalized to more complicated classes of multiloop Feynman integrals, with a computer implementation of all the steps. The method of DE, coupled with solving the reduction problem by use of IBP and LI relations by means of the algorithm of [16, 21, 22], has become, by now, a powerful industry for obtaining results for various phenomenologically important classes of Feynman integrals – see, e.g., [2, 3, 6, 7, 8, 9, 15, 27]. The method of DE was also successfully applied [10, 24] for the analytical evaluation of various (generalized) sunset diagrams.2 However, the ﬁrst impressive example of this technique was evaluating master integrals by DE for the massless double boxes with one leg oﬀ-shell, p21 = 0, p22 = p23 = p24 = 0, performed in [16]. Another important feature of the method of DE is that it provides a natural solution in the situation where results obtained can be hardly expressible in terms of known special functions of mathematical physics. The very form of results obtained when applying DE, by means of iterative integrations, naturally leads, in such a situation, to the idea to introduce new functions which would be adequate to express the results for the given class of the integrals. This is how two-dimensional HPL (2dHPL) [16], new special functions of mathematical physics introduced and studied by physicists, have appeared. They are natural generalizations of HPL to the case of functions of two variables. To deﬁne them [16] one uses, instead of the functions (B.10), the following set of functions of the two variables x and y labelled by the four indices 0, −1, −y and −1/y: g(0; x) =

1 1 1 , g(−1; x) = , g(−y; x) = , x 1+x x+y 1 g(−1/y; x) = . x + 1/y

(7.50) (7.51)

Then 2dHPLs are deﬁned as the set of functions generated by repeated integrations with these functions similarly to (B.9). Some basic properties of these new functions were studied and packages for the numerical evaluation were provided [17, 18]. These are 2dHPL that have turned out to be adequate functions to express results for the double boxes with one leg oﬀ shell [16]. This strategy of inventing new special functions, in situations where one fails to express results in terms of the known functions3 , has already become 2 For generalized sunset diagrams (i.e. with an arbitrary number of lines between two external vertices), a successful alternative technique is based on the coordinate space representation, where any such diagram is just a product of the propagators in coordinate space given by a Bessel function – see (2.16). Then, in order to go back to momentum space, it is necessary to evaluate a one-dimensional (but complicated) integral of this product of the Bessel functions with one more Bessel function – see [19] and references therein. 3 Of course, we already consider HPL and 2dHPL as known functions.

7.3 Conclusion

175

standard. In 2004, at least two types of new functions were introduced: generalized HPL in [3] which were necessary to evaluate some two-loop massive Feynman diagrams and some generalized 2dHPL [7] which were necessary to evaluate two-loop massless diagrams with three oﬀ-shell legs. Pragmatically, the introduction of new functions is just a way to parameterize the results obtained. Then one has at least a deﬁnite procedure for the numerical evaluation of any of the calculated integrals with a reasonable accuracy. Mathematically, if one introduces a new class of functions, there is an implicit obligation to describe their properties and present procedures for their numerical evaluation. Of course, it is natural to try to represent results in known functions. Observe that, in the above examples where the new functions were introduced, at least some of the new functions can be expressed in terms of the standard special functions. Consider, for example, the generalized HPL of various types which were deﬁned in [3] similarly to the HPL, with other basic functions, in particular 1/ t(t + 4). Observe that the new generalized HPL x dt H(−r, −1; x) = (7.52) t(t + 4) 0 equals 1 1 π2 Li2 −y 3 − Li2 (−y) + ln2 y − , (7.53) 3 2 18 √ √ √ √ where y = ( 4 + x − x)/( 4 + x − x). For more complicated generalized HPL, similar representations can hardly be found. Still nobody has proven a no go theorem for this situation. Moreover, it is not clear how to take into account all possible choices of special combinations of the initial variables such as the y(x) above. Anyway, physicists are naturally impatient to report on their results and apply them for the evaluation of physical quantities, so that, I hope, mathematicians will not blame them for this, keeping in mind that the mathematicians themselves seem not to bother about these interesting mathematical problems at the moment. Let us now remember about the evaluation of the massive on-shell QEDtype double boxes of Figs. (4.9) and (4.10). Two of our four examples were in fact oriented at this problem: its one-loop prototype and the sunset diagrams that can be obtained from the massive double boxes – see Sect. 4.5. In [12], it was reported about the solution of the reduction problem, by an authors’ implementation of the algorithm of [16, 21, 22]. The number of master integrals is 22 in the ﬁrst planar case, 35 in the second planar case, and 47 in the non-planar case. The diagrams with three reduced lines and some of the diagrams with two reduced lines have been calculated by DE [12]. When applying the method of DE to diagrams with six and seven lines, serious problems arise because diﬀerential equations of third order and higher are encountered there. One may still hope to solve such equations or choose

176

7 Evaluation by Diﬀerential Equations

another strategy. This could be the method of MB representation which was used to evaluate the most complicated planar master integrals – see Example 4.9 and [26]. So, hopefully, the problem of the evaluation of the massive on-shell double boxes will be completely solved in the nearest future, as well as other phenomenologically important calculational problems at least at the two-loop level.

References 1. U. Aglietti, hep-ph/0408014. 173 2. U. Aglietti and R. Bonciani, Nucl. Phys. B 668 (2003) 3; U. Aglietti, R. Bonciani, G. Degrassi and A. Vicini, hep-ph/0407162. 174 3. U. Aglietti and R. Bonciani, hep-ph/0401193; U. Aglietti, R. Bonciani, G. Degrassi and A. Vicini, Phys. Lett. B 595 (2004) 432. 174, 175 4. C. Anastasiou and A. Lazopoulos, JHEP 0407 (2004) 046. 5. M. Awramik, M. Czakon, A. Freitas and G. Weiglein, hep-ph/0407317. 173 6. W. Bernreuther et al., hep-ph/0406046. 174 7. T.G. Birthwright, E.W.N. Glover and P. Marquard, hep-ph/0407343. 174, 175 8. R. Bonciani, A. Ferroglia, P. Mastrolia, E. Remiddi and J.J. van der Bij, Nucl. Phys. B 681 (2004) 261. 167, 168, 170, 174 9. R. Bonciani, P. Mastrolia and E. Remiddi, Nucl. Phys. B 661 (2003) 289; B 676, 399 (2004); B 690, 138 (2004). 174 10. M. Caﬀo, H. Czy˙z, S. Laporta and E. Remiddi, Nuovo Cim. A 111 (1998) 365; Acta Phys. Polon. B 29 (1998) 2627; M. Caﬀo, H. Czy˙z and E. Remiddi, Nucl. Phys. B 581 (2000) 274; B 611 (2001) 503. 174 11. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 12. M. Czakon, J. Gluza and T. Riemann, hep-ph/0406203. 168, 175 13. A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 699 (2004) 3. 14. J. Fleischer, M.Yu. Kalmykov and A.V. Kotikov, Phys. Lett. B 462 (1999) 169. 15. T. Gehrmann and E. Remiddi, Nucl. Phys. B 580 (2000) 485. 174 16. T. Gehrmann and E. Remiddi, Nucl. Phys. B 601 (2001) 248; Nucl. Phys. B 601 (2001) 287. 167, 174, 175 17. T. Gehrmann and E. Remiddi, Comput. Phys. Commun. 144 (2002) 200; 141 (2001) 296. 174 18. T. Gehrmann and E. Remiddi, Nucl. Phys. B 640 (2002) 379. 174 19. S. Groote, J.G. Korner and A.A. Pivovarov, Eur. Phys. J. C 36 (2004) 471. 174 20. A.V. Kotikov, Phys. Lett. B 254 (1991) 158; B 259 (1991) 314; B 267 (1991) 123; Mod. Phys. Lett. A 6 (1991) 677; 3133; Int. J. Mod. Phys. A 7 (1992) 1977. 165 21. S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 167, 174, 175 22. S. Laporta and E. Remiddi, Phys. Lett. B 379 (1996) 283. 167, 174, 175 23. E. Remiddi, Nuovo Cim. A 110 (1997) 1435. 165 24. E. Remiddi, Acta Phys. Polon. B 34 (2003) 5311; M. Argeri, P. Mastrolia and E. Remiddi, Nucl. Phys. B 631 (2002) 388; P. Mastrolia and E. Remiddi, Nucl. Phys. B 657 (2003) 397; S. Laporta, P. Mastrolia and E. Remiddi, Nucl. Phys. B 688 (2004) 165; S. Laporta and E. Remiddi, hep-ph/0406160. 174 25. E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725. 167

References

177

26. V.A. Smirnov, hep-ph/0406052; G. Heinrich and V.A. Smirnov, hep-ph/ 0406053. 176 27. Y. Schr¨ oder, Nucl. Phys. Proc. Suppl. 116 (2003) 402; Y. Schr¨ oder and A. Vuorinen, hep-ph/0311323. 174

A Tables

A.1 Table of Integrals Each Feynman integral presented here can be evaluated straightforwardly by use of alpha or Feynman parameters. Results are presented for the ‘Euclidean’ dependence, −k 2 , of the denominators, which is more natural when the powers of propagators are general complex numbers. As usual, −k 2 is understood in the sense of −k 2 − i0, etc. Moreover, denominators with a linear dependence on k are also understood in this sense, e.g. 2p · k → 2p · k − i0, although sometimes this i0 dependence is explicitly indicated to avoid misunderstanding. 1 Γ (λ + ε − 2) dd k = iπ d/2 . (A.1) 2 2 λ 2 (−k + m ) Γ (λ) (m )λ+ε−2 Γ (λ − n + ε − 2) (−1)n gsα1 ...α2n k α1 . . . k α2n = iπ d/2 , (A.2) dd k 2 2 λ (−k + m ) 2n Γ (λ) (m2 )λ−n+ε−2 where gsα1 ...α2n = g α1 α2 . . . g α2n−1 α2n +. . . (with (2n−1)!! terms in the sum) is a combination symmetrical with respect to the permutation of any pair of indices. If the number of monomials in the numerator is odd, the corresponding integral is zero. (2l·k)2n dd k (−k 2 + m2 )λ (l2 )n Γ (λ − n + ε − 2) = iπ d/2 (−1)n (2n − 1)!! . (A.3) 2 Γ (λ) (m )λ−n+ε−2

dd k + m2 )λ1 (−k 2 )λ2 Γ (λ1 + λ2 + ε − 2)Γ (−λ2 − ε + 2) 1 = iπ d/2 . 2 λ +λ 1 2 +ε−2 Γ (λ1 )Γ (2 − ε) (m ) (−k 2

(A.4)

k α1 . . . k α2n (−k 2 + m2 )λ1 (−k 2 )λ2 (−1)n Γ (λ1 + λ2 − n + ε − 2)Γ (n − λ2 − ε + 2) = iπ d/2 n gsα1 ...α2n . 2 Γ (λ1 )Γ (n − ε + 2)(m2 )λ1 +λ2 −n+ε−2 dd k

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 179–186 (2004) c Springer-Verlag Berlin Heidelberg 2004

(A.5)

180

A Tables

dd k

(2l·k)2n = iπ d/2 (−1)n (2n − 1)!! (−k 2 + m2 )λ1 (−k 2 )λ2 Γ (λ1 + λ2 − n + ε − 2)Γ (n − λ2 − ε + 2)(l2 )n . × Γ (λ1 )Γ (n − ε + 2)(m2 )λ1 +λ2 −n+ε−2

(A.6)

dd k (−k 2 )λ1 [−(q − k)2 ]λ2 Γ (2 − ε − λ1 )Γ (2 − ε − λ2 ) Γ (λ1 + λ2 + ε − 2) = iπ d/2 . Γ (λ1 )Γ (λ2 )Γ (4 − λ1 − λ2 − 2ε) (−q 2 )λ1 +λ2 +ε−2

(A.7)

Let k (α1 ...αn ) = k α1 . . . k αn + . . . be traceless with respect to any pair of indices, i.e. gαi αj k (α1 ...αn ) = 0 – see (A.41b) below. Then (α1 ...αn ) k (α1 ...αn ) d/2 AT (λ1 , λ2 ; n)q = iπ , (A.8) dd k (−k 2 )λ1 [−(q − k)2 ]λ2 (−q 2 )λ1 +λ2 +ε−2 where AT (λ1 , λ2 ; n) =

Γ (λ1 + λ2 + ε − 2)Γ (n + 2 − ε − λ1 )Γ (2 − ε − λ2 ) . Γ (λ1 )Γ (λ2 )Γ (4 + n − λ1 − λ2 − 2ε) (A.9)

For pure monomials, the corresponding formula has one more ﬁnite summation: k α1 . . . k αn dd k (−k 2 )λ1 [−(q − k)2 ]λ2 [n/2] iπ d/2 1 = ANT (λ1 , λ2 ; r, n) r (q 2 )r {[g]r [q]n−2r }α1 ...αn , (−q 2 )λ1 +λ2 +ε−2 r=0 2

(A.10) where ANT (λ1 , λ2 ; r, n) Γ (λ1 + λ2 + ε − 2 − r)Γ (n + 2 − ε − λ1 − r)Γ (2 − ε − λ2 + r) , = Γ (λ1 )Γ (λ2 )Γ (4 + n − λ1 − λ2 − 2ε) (A.11) and {[g]r [q]n−2r }α1 ...αn is symmetric in its indices and is composed of the metric tensor and the vector q. (2l·k)n iπ d/2 = dd k (−k 2 )λ1 [−(q − k)2 ]λ2 (−q 2 )λ1 +λ2 +ε−2

[n/2]

×

r=0

ANT (λ1 , λ2 ; r, n)

n! (q 2 )r (l2 )r (2q·l)n−2r , r!(n − 2r)!

(A.12)

A.1 Table of Integrals

181

dd k (−k 2 )λ1 (−k 2 + 2p·k)λ2 1 Γ (λ1 + λ2 + ε − 2)Γ (−2λ1 − λ2 − 2ε + 4) = iπ d/2 . Γ (λ2 )Γ (−λ1 − λ2 − 2ε + 4) (p2 )λ1 +λ2 +ε−2 (A.13)

dd k

k (α1 ...αn ) p(α1 ...αn ) d/2 = iπ B (λ , λ ; n) , T 1 2 λ 2 + 2p·k) 2 (p )λ1 +λ2 +ε−2 (A.14)

(−k 2 )λ1 (−k 2

where BT (λ1 , λ2 ; n) = dd k

Γ (λ1 + λ2 + ε − 2)Γ (−2λ1 − λ2 + n − 2ε + 4) . Γ (λ2 )Γ (−λ1 − λ2 + n − 2ε + 4)

(A.15)

k α1 . . . k αn iπ d/2 = (−k 2 )λ1 (−k 2 + 2p·k)λ2 (p2 )λ1 +λ2 +ε−2

[n/2]

×

BNT (λ1 , λ2 ; r, n)

r=0

(−1)r 2 r (p ) {[g]r [p]n−2r }α1 ...αn , 2r

(A.16)

where BNT (λ1 , λ2 ; r, n) Γ (λ1 + λ2 + ε − 2 − r)Γ (−2λ1 − λ2 + n − 2ε + 4) . = Γ (λ2 )Γ (−λ1 − λ2 + n − 2ε + 4) dd k

(2l·k)n iπ d/2 = λ 2 λ + 2p·k) 2 (q ) 1 +λ2 +ε−2

(−k 2 )λ1 (−k 2

[n/2]

×

(A.17)

BNT (λ1 , λ2 ; r, n)(−1)r

r=0

n! (p2 )r (l2 )r (2p·l)n−2r . r!(n − 2r)!

(A.18)

Let p·q = 0. Then (p·k)b1 (q·k)b2 dd k (−k 2 )λ1 [−(l − k)2 ]λ2 iπ d/2 = (−l2 )λ1 +λ2 +ε−2

min{r,[b1 /2]}

×

r1 =max{0,r−[b2 /2]}

and

[(b1 +b2 )/2]

r=0

ANT (λ1 , λ2 ; r, b1 + b2 )

b1 !b2 ! 2 r (l ) 4r

(p·l)b1 −2r1 (q·l)b2 −2r+2r1 (p2 )r1 (q 2 )r−r1 , r1 !(r − r1 )!(b1 − 2r1 )!(b2 − 2r + 2r1 )!

(A.19)

182

A Tables

dd k

(p·k)b1 (q·k)b2 (−k 2 )λ1 (−k 2 + 2q·k)λ2 = iπ d/2

(p2 )b1 /2 (q 2 )λ1 +λ2 +ε−2−b1 /2−b2

Bpq (λ1 , λ2 ; b1 , b2 ) ,

(A.20)

for even b1 (and are equal to zero for odd b1 ), where Bpq (λ1 , λ2 ; b1 , b2 )

b1 /2+[b2 /2]

=

r=b1 /2

(A.21)

dd k (−k 2 + m2 )λ1 (2p·k)λ2 Γ (λ2 /2)Γ (λ1 + λ2 /2 + ε − 2) iπ d/2 . 2 λ /2 2 λ +λ /2+ε−2 2 1 2 2Γ (λ1 )Γ (λ2 ) (p ) (m )

= dd k

(−k 2

= iπ d/2

(−1)r b1 !b2 ! BNT (λ1 , λ2 ; r, b1 + b2 ) . r 4 (b1 /2)!(r − b1 /2)!

(A.22)

k (α1 ,...,αn ) + m2 )λ1 (2p·k)λ2 Γ ((λ2 + n)/2) Γ (λ1 + (λ2 − n)/2 + ε − 2) p(α1 ,...,αn ) . 2Γ (λ1 )Γ (λ2 ) (m2 )λ1 +(λ2 −n)/2+ε−2 (p2 )(λ2 +n)/2 (A.23)

dd k (−k 2 + 2p·k)λ1 (2p·k)λ2 =

iπ d/2 (p2 )λ1 +λ2 +ε−2

Γ (λ1 + λ2 + ε − 2)Γ (2λ1 + λ2 + 2ε − 4) . Γ (λ1 )Γ (2λ1 + 2λ2 + 2ε − 4)

(A.24)

dd k + ω − i0)λ2 Γ (2 − λ1 − ε)Γ (2λ1 + λ2 + 2ε − 4) 2 λ1 +ε−2 −2λ1 −λ2 −2ε+4 (v ) = iπ d/2 ω . Γ (λ1 )Γ (λ2 ) (A.25) (−k 2 )λ1 (2v·k

dd k ×

k (α1 ,...,αn ) = iπ d/2 ω −2λ1 −λ2 +n−2ε+4 (−k 2 )λ1 (2v·k + ω − i0)λ2

v (α1 ,...,αn ) (v 2 )−λ1 +n−ε+2

Let v·q = 0. Then

Γ (2 − λ1 + n − ε)Γ (2λ1 + λ2 − n + 2ε − 4) . Γ (λ1 )Γ (λ2 )

(A.26)

A.1 Table of Integrals

dd k (−k 2 )λ1 [−(q − k)2 ]λ2 (−2v·k − i0)λ3 Γ (−λ1 − λ3 /2 − ε + 2)Γ (−λ2 − λ3 /2 − ε + 2) = iπ d/2 Γ (−λ1 − λ2 − λ3 − 2ε + 4) Γ (λ1 + λ2 + λ3 /2 + ε − 2)Γ (λ3 /2) . × 2Γ (λ1 )Γ (λ2 )Γ (λ3 )(−q 2 )λ1 +λ2 +λ3 /2+ε−2 (v 2 )λ3 /2

Let p21 = p22 = 0, q = p1 − p2 . Then dd k 2 λ (−k + 2p1 ·k) 1 (−k 2 + 2p2 ·k)λ2 (−k 2 )λ3 Γ (−λ1 − λ3 − ε + 2)Γ (−λ2 − λ3 − ε + 2) = iπ d/2 Γ (λ1 )Γ (λ2 )Γ (−λ1 − λ2 − λ3 − 2ε + 4) Γ (λ1 + λ2 + λ3 + ε − 2) , × (−q 2 )λ1 +λ2 +λ3 +ε−2 (−k 2

183

(A.27)

(A.28)

Γ (−λ1 − ε + 2) dd k = iπ d/2 + 2p1 + 2p2 ·k)λ2 (2p2 ·k)λ3 Γ (λ1 )Γ (λ2 ) Γ (λ1 + λ2 + ε − 2)Γ (−λ2 − λ3 − ε + 2) , (A.29) × Γ (−λ1 − λ2 − λ3 − 2ε + 4)(−q 2 )λ1 +λ2 +λ3 +ε−2 ·k)λ1 (−k 2

dd k (2p1 + 2p2 ·k)λ2 (−k 2 + m2 )λ3 Γ (λ2 − λ1 )Γ (λ2 + λ3 + ε − 2)Γ (−λ2 − ε + 2) = iπ d/2 , Γ (λ2 )Γ (λ3 )Γ (−λ1 − ε + 2)(−q 2 )λ1 (m2 )λ2 +λ3 +ε−2 ·k)λ1 (−k 2

(A.30)

dd k + 2p2 + m2 )λ3 (Q2 − 2p1 ·k)λ4 Γ (λ2 − λ1 )Γ (λ2 + λ3 + ε − 2)Γ (−λ2 − λ4 − ε + 2) = iπ d/2 Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ4 − ε + 2) 1 × 2 λ1 +λ4 2 λ2 +λ3 +ε−2 , (Q ) (m )

(2p1

·k)λ1 (−k 2

·k)λ2 (−k 2

(A.31)

dd k (2p1 ·k + m2 )λ1 (2p2 ·k + m2 )λ2 (−k 2 )λ3 Γ (λ1 + λ3 + ε − 2)Γ (λ2 + λ3 + ε − 2)Γ (−λ3 − ε + 2) = iπ d/2 . Γ (λ1 )Γ (λ2 )Γ (λ3 )(−q 2 )−λ3 −ε+2 (m2 )λ1 +λ2 +2λ3 +2ε−4

(A.32)

Let p21 = 0, p22 = −m2 , q = p1 − p2 . Then Γ (λ2 + λ3 + ε − 2) dd k = iπ d/2 (2p1 ·k)λ1 (−k 2 + 2p2 ·k + m2 )λ2 (−k 2 )λ3 (m2 )λ2 +λ3 +ε−2 Γ (−λ1 − λ3 − ε + 2)Γ (−λ2 − ε + 2) × , (A.33) Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ2 − λ3 − 2ε + 4)(−q 2 )λ1

184

A Tables

dd k (2p1 ·k)λ1 (−k 2 + 2p2 ·k − m2 )λ2 (−k 2 )λ3 (−q 2 − 2p1 ·k)λ4 Γ (λ2 + λ3 + ε − 2) = iπ d/2 (m2 )λ2 +λ3 +ε−2 Γ (−λ1 − λ3 − ε + 2)Γ (−λ2 − λ4 − ε + 2) × . Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ2 − λ3 − λ4 − 2ε + 4)(−q 2 )λ1 +λ4

(A.34)

Let P 2 = M 2 , p2 = 0, (P − p)2 = 0. Then dd k 2 λ (−k + 2P ·k) 1 (−k 2 + 2p·k)λ2 (−k 2 )λ3 Γ (−λ1 − λ2 − 2λ3 − 2ε + 4)Γ (λ1 + λ2 + λ3 + ε − 2) = iπ d/2 Γ (λ1 )Γ (−λ1 − λ2 − λ3 − 2ε + 4) Γ (−λ2 − λ3 − ε + 2) . (A.35) × Γ (−λ3 − ε + 2)(M 2 )λ1 +λ2 +λ3 +ε−2 Let p21 = 0, p22 = m2 , Q2 = 2p1 ·p2 . Then dd k (2p1 ·k)λ1 (−k 2 + 2p2 ·k)λ2 (−k 2 )λ3 (Q2 − 2p1 ·k)λ4 Γ (λ3 − λ4 )Γ (−λ1 − λ2 − 2λ3 − 2ε + 4) = iπ d/2 Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ2 − λ3 − λ4 − 2ε + 4) Γ (λ2 + λ3 + ε − 2) × 2 λ1 +λ4 2 λ2 +λ3 +ε−2 , (Q ) (m )

iπ d/2 dd k = (2p1 ·k)λ1 (−k 2 + 2p2 ·k)λ2 (−k 2 )λ3 (Q2 )λ1 (m2 )λ2 +λ3 +ε−2 Γ (λ2 + λ3 + ε − 2)Γ (−λ1 − λ2 − 2λ3 − 2ε + 4) . × Γ (λ2 )Γ (−λ1 − λ2 − λ3 − 2ε + 4)

The following integrals are related to two-loop diagrams: dd k dd l 2 2 λ 1 (−k + m ) [−(k + l)2 ]λ2 (−l2 + m2 )λ3 2 Γ (λ + λ + ε − 2)Γ (λ + λ + ε − 2)Γ (2 − ε − λ ) 1 2 2 3 2 = iπ d/2 Γ (λ1 )Γ (λ3 ) Γ (λ1 + λ2 + λ3 + 2ε − 4) , × Γ (λ1 + 2λ2 + λ3 + 2ε − 4)Γ (2 − ε)(m2 )λ1 +λ2 +λ3 +2ε−4

(A.36)

(A.37)

(A.38)

dd k dd l (−k 2 )λ1 [−(k + l)2 ]λ2 (m2 − l2 )λ3 2 Γ (λ + λ + λ + 2ε − 4) 1 2 3 = iπ d/2 (m2 )λ1 +λ2 +λ3 +2ε−4 Γ (λ1 + λ2 + ε − 2)Γ (2 − ε − λ1 )Γ (2 − ε − λ2 ) . × Γ (λ1 )Γ (λ2 )Γ (λ3 )Γ (2 − ε)

(A.39)

A.2 Some Useful Formulae

185

This is the (inverse) Fourier transformation of (−q 2 − i0)−λ in d dimensions: 1 1 e−ix·q iΓ (d/2 − λ) d q = λ d/2 . (A.40) d d 2 λ 2 (2π) (−q − i0) 4 π Γ (λ) (−x + i0)d/2−λ

A.2 Some Useful Formulae To traceless expressions and back: k

α1

αN

...k

[N/2] 1 1 = (k 2 )r {[g]r [k](N −2r) }α1 ...αN , N ! r=0 2r (d/2 + N − 2r)r

(A.41a) 1 1 (k 2 )r {[g]r [k]N −2r }α1 ...αN , r N ! r=0 2 (2 − N − d/2)r [N/2]

k (α1 ...αN ) =

(A.41b) where {[g]r [k]N −2r }α1 ...αN is deﬁned after (A.11) and (a)n is the Pochhammer symbol (B.2). Furthermore,

[N/2] N

(k·p)

=

aN,r (k 2 )r (p2 )r (k·p)(N −2r) ,

(A.42)

r=0

[N/2] (N )

(k·p)

=

bN,r (k 2 )r (p2 )r (k·p)N −2r ,

(A.43)

(d − 2)N (k 2 )N , 2N ((d − 2)/2)N

(A.44)

r=0

k(α1 ...αN ) k (α1 ...αN ) =

where (k · p)(N ) = k(α1 ...αN ) p(α1 ...αN ) and N! , − 2r)!(d/2 + N − 2r)r 1 = r . 4 r!(N − 2r)!(2 − N − d/2)r

aN,r = bN,r

(A.45)

4r r!(N

(A.46)

Summation formulae: α

α

[(k1 )m (k2 )n ∗ gs ] ≡ k1α1 . . . k1αm k2 m+1 . . . k2 m+n gs, α1 ...αm+n

min{m,n}

=

j≥0, j+min{m,n} even

m!n! 2(m+n)/2−j ((m − j)/2)!((n − j)/2)!j!

×(k12 )(m−j)/2 (k22 )(n−j)/2 (k1 ·k2 )j ,

(A.47)

186

A Tables

[(k1 )m (k2 )n ∗ {[g]r [k3 ]m+n−2r }]

min{2r,m}

=

min{r1 ,2r−r1 }

r1 =max{0,2r−n} j≥0, j+r1 even

1 (m − r1 )!(n − 2r + r1 )!

m!n!

×

2r−j ((r1 − j)/2)!(r − (r1 + j)/2)!j! ×(k1 ·k2 )j (k1 ·k3 )m−r1 (k2 ·k3 )n−2r+r1

(k12 )(r1 −j)/2 (k22 )r−(r1 +j)/2 .

(A.48)

In particular, [(k1 )m (k2 )n ∗ {[g]r [k3 ]N −2r }] n = (k2 ·k3 )N −2r [(k1 )m (k2 )n−N +2r ∗ gs ] , N − 2r

(A.49)

where k1 ·k3 = 0, N = m + n, and [pb1 q b2 ∗ {[g]r [l]n−2r }] b1 !b2 ! = 2r

min{r,[b1 /2]}

r1 =max{0,r−[b2 /2]}

(p·l)b1 −2r1 (q·l)b2 −2r+2r1 (p2 )r1 (q 2 )r−r1 , r1 !(r − r1 )!(b1 − 2r1 )!(b2 − 2r + 2r1 )! (A.50)

where p·q = 0 and n = b1 + b2 . [(k1 )m (k2 )n (k3 )l−m−n ∗ gs ] =

a(l, m, n, j1 , j2 , j3 )

j1 ≥0, j1 +m even j2 ≥0, j2 +n even j3 ≥0, j3 +l−m−n even

×(k12 )(m−j1 )/2 (k22 )(n−j2 )/2 (k32 )(l−m−n−j3 )/2 ×(k1 ·k2 )(j1 +j2 −j3 )/2 (k1 ·k3 )(j1 −j2 +j3 )/2 (k2 ·k3 )(−j1 +j2 +j3 )/2 , 2(j1 +j2 +j3 −l)/2 m!n!(l − m − n)! ((m − j1 )/2)!((n − j2 )/2)!((l − m − n − j3 )/2)! θ(j1 + j2 − j3 )θ(j1 − j2 + j3 )θ(−j1 + j2 + j3 ) , (A.51) × ((j1 + j2 − j3 )/2)!((j1 − j2 + j3 )/2)!((−j1 + j2 + j3 )/2)!

a(l, m, n, j1 , j2 , j3 ) =

where θ(n) = 1 for n ≥ 0 and θ(n) = 0 otherwise.

B Some Special Functions

The Gauss hypergeometric function [3] is deﬁned by the series ∞ (a)n (b)n n z , F (a, b; c; z) = 2 1 (c)n n! n=0

(B.1)

where (x)n = Γ (x + n)/Γ (x)

(B.2)

is the Pochhammer symbol. This power series has the radius of convergence equal to one. It is analytically continued to the whole complex plane, with a cut, usually chosen as [1, ∞). The analytic continuation to values of z where |z| > 1 is given by Γ (c)Γ (b − a) 1 −a (−z) 2 F1 a, 1 − c + a; 1 − b + a; 2 F1 (a, b; c; z) = Γ (b)Γ (c − a) z Γ (c)Γ (a − b) 1 (−z)−b 2 F1 b, 1 − c + b; 1 − a + b; + . (B.3) Γ (a)Γ (c − b) z Another formula for the analytic continuation is z −a F (a, b; c; z) = (1 − z) F a, c − b; c; . (B.4) 2 1 2 1 z−1 This is a useful parametric representation: 1 Γ (c) F (a, b; c; z) = dx xb−1 (1 − x)c−b−1 (1 − zx)−a . (B.5) 2 1 Γ (b)Γ (c − b) 0 The polylogarithms [6] and generalized (Nielsen) polylogarithms [2, 5] are deﬁned by ∞ zn (B.6) Lia (z) = na n=1 1 a−1 (−1)a ln t dt (B.7) = (a − 1)! 0 t − 1/z and (−1)a+b−1 1 lna−1 t lnb (1 − zt) dt , (B.8) Sa,b (z) = (a − 1)!b! 0 t where a and b are positive integers. V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 187–190 (2004) c Springer-Verlag Berlin Heidelberg 2004

188

B Some Special Functions

The harmonic polylogarithms [8] Ha1 ,a2 ,...,an (x) ≡ H(a1 , a2 , . . . , an ; x), (HPL) with ai = 1, 0, −1, are deﬁned recursively by x Ha1 ,a2 ,...,an (x) = dtf (a1 ; t)H(a2 , . . . , an ; t) , (B.9) 0

where 1 1 , f0 (x) = , 1∓x x H±1 (x) = ∓ ln(1 ∓ x), H0 (x) = ln x , f±1 (x) =

(B.10) (B.11)

and at least one of the indices ai is non-zero. For all ai = 0, one has 1 n ln x . (B.12) n! Up to level 4, HPL with the indices 0 and 1 can be expressed in terms of usual polylogarithms [8]: H0,0,...,0 (x) =

H0 (x) = ln x , H1 (x) = − ln(1 − x) , 1 H0,0 (x) = ln2 x , 2! H0,1 (x) = Li2 (x) , H1,0 (x) = − ln x ln(1 − x) − Li2 (x) , 1 H1,1 (x) = ln2 (1 − x) , 2! 1 H0,0,0 (x) = ln3 x , 3! H0,0,1 (x) = Li3 (x) , H0,1,0 (x) = −2Li3 (x) + ln x Li2 (x) , H0,1,1 (x) = S1,2 (x) , 1 H1,0,0 (x) = − ln(1 − x) ln2 x − ln x Li2 (x) + Li3 (x) , 2 H1,0,1 (x) = −2S1,2 (x) − ln(1 − x)Li2 (x) , 1 H1,1,0 (x) = S1,2 (x) + ln(1 − x) Li2 (x) + ln x ln2 (1 − x) , 2 1 3 H1,1,1 (x) = − ln (1 − x) , 3! 1 4 H0,0,0,0 (x) = ln x , 4! H0,0,0,1 (x) = Li4 (x) , H0,0,1,0 (x) = ln x Li3 (x) − 3Li4 (x) , H0,0,1,1 (x) = S2,2 (x) , 1 H0,1,0,0 (x) = ln2 x Li2 (x) − 2 ln x Li3 (x) + 3Li4 (x) , 2

(B.13) (B.14) (B.15) (B.16) (B.17) (B.18) (B.19) (B.20) (B.21) (B.22) (B.23) (B.24) (B.25) (B.26) (B.27) (B.28) (B.29) (B.30) (B.31)

References

189

1 2 H0,1,0,1 (x) = −2S2,2 (x) + Li2 (x) , (B.32) 2 1 2 H0,1,1,0 (x) = ln x S1,2 (x) − Li2 (x) , (B.33) 2 (B.34) H0,1,1,1 (x) = S1,3 (x) , 1 3 1 2 H1,0,0,0 (x) = − ln x ln(1 − x) − ln x Li2 (x) 6 2 (B.35) + ln x Li3 (x) − Li4 (x) , 1 2 (B.36) H1,0,0,1 (x) = − Li2 (x) − ln(1 − x)Li3 (x) , 2 H1,0,1,0 (x) = 2 ln(1 − x)Li3 (x) − ln x ln(1 − x)Li2 (x) − 2 ln x S1,2 (x) 1 2 + Li2 (x) + 2S2,2 (x) , (B.37) 2 (B.38) H1,0,1,1 (x) = − ln(1 − x)S1,2 (x) − 3S1,3 (x) , 1 H1,1,0,0 (x) = ln2 x ln2 (1 − x) − ln(1 − x)Li3 (x) 4 (B.39) + ln x ln(1 − x)Li2 (x) + ln x S1,2 (x) − S2,2 (x) , 1 2 H1,1,0,1 (x) = ln (1 − x)Li2 (x) + 2 ln(1 − x)S1,2 (x) + 3S1,3 (x) , (B.40) 2 1 1 H1,1,1,0 (x) = − ln x ln3 (1 − x) − ln2 (1 − x) Li2 (x) 6 2 (B.41) − ln(1 − x)S1,2 (x) − S1,3 (x) , 1 H1,1,1,1 (x) = ln4 (1 − x) . (B.42) 4! Analytic properties of HPL (and 2dHPL) which allow to continue them to any domain are described in [18]. The HPL are partial cases of the socalled Z- and S-sums which are deﬁned similarly to the nested sums (see Appendix C) but with the factor xj – see, e.g., [7]. The set of Z- or S-sums can be equipped with an operation of multiplication in such a way that they (as well as HPL) form a Hopf algebra – see, e.g., [1, 8].

References 1. J. Bl¨ umlein, Comput. Phys. Commun. 159 (2004) 19. 189 2. A. Devoto and D.W. Duke, Riv. Nuovo Cim. 7, No. 6 (1984) 1. 187 3. A. Erd´elyi (ed.), Higher Transcendental Functions, Vols. 1 and 2 (McGraw-Hill, New York, 1954). 187 4. T. Gehrmann and E. Remiddi, Nucl. Phys. B 640 (2002) 379. 5. K.S. K¨ olbig, J.A. Mignaco and E. Remiddi, BIT 10 (1970) 38; K.S. K¨ olbig, Math. Comp. 39 (1982) 647. 187

190

B Some Special Functions

6. L. Lewin, Polylogarithms and Associated Functions (North-Holland, Amsterdam, 1981). 187 7. S. Moch, P. Uwer and S. Weinzierl, J. Math. Phys. 43 (2002) 3363. 189 8. E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725. 188, 189

C Summation Formulae

Nested sums are deﬁned as follows [17]: Si (n) = Sikl (n) =

n n 1 Sk (j) , S (n) = , ik i j ji j=1 j=1

n Skl (j) j=1

ji

, Siklm (n) =

n Sklm (j) j=1

ji

,

(C.1)

(C.2)

etc. Properties and algorithms for the nested sums (also for negative indices which are deﬁned with (−1)j ) are presented in [17]. In particular, for positive indices, we have Sj,k (n) + Sk,j (n) = Sj (n)Sk (n) + Sj+k (n) .

(C.3)

The nested sums are closely connected with multiple ζ-values – see, e.g., [1, 2, 11, 19] and the review [7]. The sums with one index are connected with the ψ function (the logarithmical derivative of the gamma function) as ψ(n) = S1 (n − 1) − γE , ψ (n) = (−1)k k! (Sk+1 (n − 1) − ζ(k + 1)) , k = 1, 2 , . . . , (k)

(C.4) (C.5)

where ζ(z) is the Riemann zeta function ζ(z) =

∞ 1 . nz n=1

(C.6)

All the summation formulae of this Appendix, apart from the inverse binomial series1 , are implemented in the package called SUMMER [17] which is written in FORM [16]. This powerful package was successfully used in nontrivial calculations – see, e.g., [12, 13, 14]. There is also another package operating with the nested sums [18]. Nested sums are closely connected with expansions of hypergeometric series in its parameters – see, e.g., [3, 4, 11]. For example, the expansion of the Gauss hypergeometric function 2 F1 (1 + a1 ε, 1 + a2 ε; 3/2 + bε; z) is connected with inverse binomial series [3]. 1

The authors of SUMMER are planning to include the inverse binomial series into this package. V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 191–205 (2004) c Springer-Verlag Berlin Heidelberg 2004

192

C Summation Formulae

C.1 Some Number Series These are series up to level 6 with at least 1/n2 dependence:

∞

∞ 1 π2 , = n2 6 n=1

(C.7)

∞ 1 = ζ(3) , n3 n=1

(C.8)

1 = ζ(3) , n2

(C.9)

S1 (n − 1)

n=1

∞ 1 π4 , = n4 90 n=1 ∞

S1 (n − 1)

1 π4 , = n3 360

(C.11)

S1 (n − 1)2

1 11π 4 , = 2 n 360

(C.12)

1 π4 , = 2 n 120

(C.13)

∞ 1 = ζ(5) , 5 n n=1

(C.14)

n=1 ∞ n=1 ∞

S2 (n − 1)

n=1

∞

S1 (n − 1)

1 π 2 ζ(3) , = 2ζ(5) − n4 6

(C.15)

S2 (n − 1)

1 π 2 ζ(3) 11ζ(5) − , = 3 n 2 2

(C.16)

S1 (n − 1)2

1 π 2 ζ(3) 3ζ(5) − , = 3 n 6 2

(C.17)

S3 (n − 1)

1 9ζ(5) π 2 ζ(3) − , = n2 2 3

(C.18)

S1 (n − 1)3

1 π 2 ζ(3) 15ζ(5) + , = n2 6 2

(C.19)

1 7ζ(5) π 2 ζ(3) − , = 2 n 2 6

(C.20)

1 2π 2 ζ(3) , = 9ζ(5) − 2 n 3

(C.21)

n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)S2 (n − 1)

n=1 ∞ n=1

(C.10)

S12 (n − 1)

C.1 Some Number Series ∞ 1 π6 , = n6 945 n=1 ∞

ζ(3)2 1 π6 − , = 5 n 1260 2

(C.23)

S2 (n − 1)

1 π6 + ζ(3)2 , = −4 4 n 2835

(C.24)

1 37π 6 − ζ(3)2 , = n4 22680

(C.25)

S3 (n − 1)

ζ(3)2 1 π6 + , = − n3 1890 2

(C.26)

S4 (n − 1)

1 5π 6 − ζ(3)2 , = 2 n 2268

(C.27)

S13 (n − 1)

1 61π 6 , = n2 45360

(C.28)

S2 (n − 1)2

1 59π 6 − ζ(3)2 , = n2 22680

(C.29)

S1 (n − 1)3

1 11π 6 + 2ζ(3)2 , = − n3 5040

(C.30)

1 121π 6 + 2ζ(3)2 , =− 3 n 45360

(C.31)

1 41π 6 − ζ(3)2 , = n3 22680

(C.32)

3ζ(3)2 1 167π 6 − , = n2 45360 2

(C.33)

1 23π 6 − ζ(3)2 , = 2 n 3780

(C.34)

1 859π 6 + 3ζ(3)2 , = 2 n 22680

(C.35)

1 17π 6 − ζ(3)2 , = n2 4536

(C.36)

1 313π 6 − 2ζ(3)2 . = n2 45360

(C.37)

n=1 ∞

S1 (n − 1)2

n=1 ∞ n=1 ∞ n=1 ∞

n=1 ∞ n=1 ∞

S1 (n − 1)S2 (n − 1)

n=1 ∞

S12 (n − 1)

n=1 ∞

S1 (n − 1)S3 (n − 1)

n=1 ∞

S1 (n − 1)2 S2 (n − 1)

n=1 ∞

S1 (n − 1)4

n=1 ∞

S112 (n − 1)

n=1 ∞ n=1

(C.22)

S1 (n − 1)

n=1 ∞

n=1 ∞

193

S1 (n − 1)S12 (n − 1)

194

C Summation Formulae

Series up to level 6 with the factor 1/n where the convergence is provided by other factors: ∞

ψ (n + 1)

1 = ζ(3) , n

(C.38)

ψ (n + 1)S1 (n)

7π 4 1 = , n 360

(C.39)

π4 1 =− , n 180

(C.40)

π 2 ζ(3) 1 = , n 3

(C.41)

1 5π 2 ζ(3) = − 9ζ(5) , n 6

(C.42)

2π 2 ζ(3) 1 =− + 7ζ(5) , n 3

(C.43)

ψ (n + 1)

1 = −π 2 ζ(3) + 12ζ(5) , n

(C.44)

ψ (n + 1)

2π 6 1 =− + 12ζ(3)2 , n 105

(C.45)

ψ (n + 1)S1 (n)

π6 1 = , n 1512

(C.46)

ψ (n + 1)S1 (n)2

1 π6 = − 8ζ(3)2 , n 90

(C.47)

ψ (n + 1)2 S1 (n)

π6 1 =− + 2ζ(3)2 , n 432

(C.48)

ψ (n + 1)S1 (n)3

269π 6 1 = , n 22680

(C.49)

61π 6 1 = − 2ζ(3)2 . n 22680

(C.50)

n=1 ∞ n=1 ∞

ψ (n + 1)

n=1 ∞

ψ (n + 1)S1 (n)2

n=1 ∞

ψ (n + 1)2

n=1 ∞

ψ (n + 1)S1 (n)

n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

ψ (n + 1)ψ (n + 1)

n=1

Series of level 7 with at least 1/n2 dependence: ∞ 1 = ζ(7) , 7 n n=1 ∞ n=1

S1 (n − 1)

1 π 2 ζ(5) π 4 ζ(3) − , = 3ζ(7) − n6 6 90

(C.51) (C.52)

C.1 Some Number Series ∞

1 5π 2 ζ(5) π 4 ζ(3) + , = −11ζ(7) + n5 6 45

(C.53)

1 π 2 ζ(5) π 4 ζ(3) − , = −ζ(7) + 5 n 6 180

(C.54)

1 5π 2 ζ(5) , = 17ζ(7) − 4 n 3

(C.55)

1 119ζ(7) π 2 ζ(5) 11π 4 ζ(3) + − , = n4 16 3 120

(C.56)

1 61ζ(7) π 2 ζ(5) π 4 ζ(3) − + = , n4 16 3

(C.57)

1 141ζ(7) 5π 2 ζ(5) π 4 ζ(3) − − , = 4 n 8 4 24

(C.58)

S4 (n − 1)

1 5π 2 ζ(5) π 4 ζ(3) + , = −18ζ(7) + n3 3 90

(C.59)

S13 (n − 1)

1 73ζ(7) 5π 2 ζ(5) π 4 ζ(3) + + , = − n3 4 3 72

(C.60)

1 85ζ(7) 11π 2 ζ(5) π 4 ζ(3) + + , = − n3 8 12 72

(C.61)

S2 (n − 1)2

1 13ζ(7) 5π 2 ζ(5) 11π 4 ζ(3) − + , = 3 n 8 6 180

(C.62)

S1 (n − 1)S12 (n − 1)

1 113ζ(7) 7π 2 ζ(5) π 4 ζ(3) + + , = − n3 16 12 72

(C.63)

S1 (n − 1)2 S2 (n − 1)

1 77ζ(7) π 2 ζ(5) 7π 4 ζ(3) − + , = − n3 8 3 60

(C.64)

S2 (n − 1)

n=1 ∞

S1 (n − 1)2

n=1 ∞

S3 (n − 1)

n=1 ∞

S1 (n − 1)3

n=1 ∞

S1 (n − 1)S2 (n − 1)

n=1 ∞

S12 (n − 1)

n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)S3 (n − 1)

n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)4

n=1 ∞

1 109ζ(7) 5π 2 ζ(5) 37π 4 ζ(3) − + , (C.65) =− 3 n 8 6 180

S112 (n − 1)

1 61ζ(7) 5π 2 ζ(5) π 4 ζ(3) + + , =− 3 n 4 4 40

(C.66)

S1 (n − 1)S4 (n − 1)

1 173ζ(7) 3π 2 ζ(5) π 4 ζ(3) − − , = n2 16 4 60

(C.67)

S1 (n − 1)S13 (n − 1)

1 61ζ(7) 3π 2 ζ(5) π 4 ζ(3) − + , = n2 4 2 36

(C.68)

n=1 ∞ n=1 ∞ n=1

195

196

C Summation Formulae ∞

S1 (n − 1)2 S3 (n − 1)

1 301ζ(7) 3π 2 ζ(5) π 4 ζ(3) − − , = n2 16 4 15

(C.69)

S1 (n − 1)S2 (n − 1)2

1 77ζ(7) 13π 2 ζ(5) π 4 ζ(3) + − , =− 2 n 16 12 30

(C.70)

S1 (n − 1)2 S12 (n − 1)

1 423ζ(7) π 2 ζ(5) 37π 4 ζ(3) − − , = 2 n 16 6 360

(C.71)

1 307ζ(7) 5π 2 ζ(5) 13π 4 ζ(3) + − , = n2 16 12 180

(C.72)

n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)3 S2 (n − 1)

n=1 ∞

S1 (n − 1)5

n=1

1 1855ζ(7) 19π 2 ζ(5) + = n2 16 4

11π 4 ζ(3) , 30 ∞ 1 73ζ(7) 3π 2 ζ(5) π 4 ζ(3) − − , S1 (n − 1)S112 (n − 1) 2 = n 4 4 30 n=1 +

∞

S5 (n − 1)

n=1 ∞

S14 (n − 1)

n=1 ∞

S2 (n − 1)S3 (n − 1)

n=1 ∞

S23 (n − 1)

n=1 ∞

S2 (n − 1)S12 (n − 1)

n=1 ∞

(C.74)

1 2π 2 ζ(5) π 4 ζ(3) − , = 10ζ(7) − n2 3 45

(C.75)

1 141ζ(7) 19π 2 ζ(5) π 4 ζ(3) − − , = 2 n 8 12 360

(C.76)

1 19ζ(7) 5π 2 ζ(5) 7π 4 ζ(3) + − , = 2 n 16 12 180

(C.77)

1 131ζ(7) 4π 2 ζ(5) 7π 4 ζ(3) + − , = − n2 16 3 180

(C.78)

1 141ζ(7) 5π 2 ζ(5) 19π 4 ζ(3) + − , (C.79) = − n2 16 3 360

S113 (n − 1)

1 113ζ(7) π 2 ζ(5) − , = 2 n 16 2

(C.80)

S212 (n − 1)

1 169ζ(7) π 2 ζ(5) 7π 4 ζ(3) − − , = n2 16 2 180

(C.81)

S1112 (n − 1)

1 141ζ(7) 7π 4 ζ(3) 2 − π . = ζ(5) − n2 8 180

(C.82)

n=1 ∞ n=1 ∞ n=1

(C.73)

C.2 Power Series of Levels 3 and 4 in Terms of Polylogarithms

197

C.2 Power Series of Levels 3 and 4 in Terms of Polylogarithms The formulae of this section can be found in [6]. ∞

S2 (n − 1)

n=1 ∞ n=1 ∞

zn = −2S1,2 (z) − ln(1 − z)Li2 (z) , n

S1 (n − 1)2 S1 (n − 1)

n=1 ∞

(C.83)

zn 1 = −2S1,2 (z) − ln(1 − z)Li2 (z) − ln3 (1 − z) , (C.84) n 3

zn = S1,2 (z) , n2

(C.85)

zn = Li3 (z) , n3 n=1 ∞ n=1 ∞

S3 (n − 1)

(C.86)

1 zn 2 = − Li2 (z) − ln(1 − z)Li3 (z) , n 2

S12 (n − 1)

n=1

(C.87)

zn 1 2 = 3S1,3 (z) − ln(1 − z)Li3 (z) − Li2 (z) n 2

1 + ln2 (1 − z)Li2 (z) + 2 ln(1 − z)S1,2 (z) , (C.88) 2 ∞ 1 zn 2 = − Li2 (z) + ln(1 − z)(S1,2 (z) − Li3 (z)) S1 (n − 1)S2 (n − 1) n 2 n=1 1 + ln2 (1 − z)Li2 (z) , 2 ∞ n 1 z 3 2 = − Li2 (z) + ln2 (1 − z)Li2 (z) S1 (n − 1)3 n 2 2 n=1 + ln(1 − z)(3S1,2 (z) − Li3 (z)) + ∞ n=1 ∞ n=1 ∞ n=1 ∞

S2 (n − 1)

zn 1 2 = −2S2,2 (z) + Li2 (z) , n2 2

S1 (n − 1)2 S1 (n − 1)

1 4 ln (1 − z) , 4

zn 1 2 = 2S1,3 (z) − 2S2,2 (z) + Li2 (z) , n2 2

zn = S2,2 (z) , n3

zn = Li4 (z) . n4 n=1

(C.89)

(C.90) (C.91) (C.92) (C.93) (C.94)

198

C Summation Formulae

C.3 Inverse Binomial Power Series up to Level 4 The formulae of this section (as well as other similar formulae) can be found in [3]. See a table of formulae for the corresponding number series in [8]. Let √ √ 4 − z − −z y=√ . √ 4 − z + −z Then ∞

1−y 1 zn 2n = ln y, n 1+y n n=1 ∞

1 zn 1 2n 2 = − ln2 y, n 2 n n=1

(C.95) (C.96)

∞

1 zn 2n 3 = 2Li3 (y) − 2 ln y Li2 (y) − ln2 y ln(1 − y) n n n=1 ∞

1 + ln3 y − 2ζ(3) , 6

(C.97)

1 zn 2n 4 = 4S2,2 (y) − 4Li4 (y) − 4S1,2 (y) ln y n n n=1 +4Li3 (y) ln(1 − y) + 2Li3 (y) ln y − 4Li2 (y) ln y ln(1 − y) 1 1 4 ln y − ln2 y ln2 (1 − y) + ln3 y ln(1 − y) − 3 24 −4 ln(1 − y)ζ(3) + 2 ln y ζ(3) + 3ζ(4) , (C.98) ∞

1 zn 1−y 2n S1 (n − 1) = n 1+y n n=1

1 2 × −2Li2 (−y) − 2 ln y ln(1 + y) + ln y − ζ(2) , 2 ∞ 1 zn 1−y 2n S1 (n − 1)2 = 8S1,2 (−y) − 4Li3 (−y) n 1 +y n n=1

(C.99)

+8Li2 (−y) ln(1 + y) + 4 ln2 (1 + y) ln y − 2 ln(1 + y) ln2 y

1 3 + ln y + 4ζ(2) ln(1 + y) − 2ζ(2) ln y − 4ζ(3) , (C.100) 6 ∞

1 zn 1−y 2n S2 (n − 1) = − ln3 y , n 6(1 + y) n n=1 ∞

1 zn 1 4 2n 2 S2 (n − 1) = ln y , n 24 n n=1

1 zn 1 − y 1 4 2n S3 (n − 1) = ln y + 6Li4 (y) + ln2 y Li2 (y) n 1 + y 24 n n=1 ∞

(C.101) (C.102)

C.3 Inverse Binomial Power Series up to Level 4

199

−2ζ(3) ln y − 4 ln y Li3 (y) − 6ζ(4) , 1 zn 1 − y 1 3 1 4 2n S1 (n − 1)S2 (n − 1) = ln y ln(1 + y) − ln y n 1 + y 3 24 n n=1 ∞

1 + ζ(2) ln2 y + ln2 y Li2 (−y) + ln2 y Li2 (y) + ζ(3) ln y − 4 ln y Li3 (−y) 2 −4 ln y Li3 (y) + ζ(4) + 8Li4 (−y) + 6Li4 (y) ,

(C.103)

∞

1 zn 2n 2 S1 (n − 1) = 4Li3 (−y) − 2Li2 (−y) ln y n n n=1 ∞

1 − ln3 y + 3ζ(3) + ζ(2) ln y , 6

(C.104)

1 zn 2n 2 S1 (n − 1)2 = −8S1,2 (−y) ln y + 4Li3 (−y) ln y n n n=1 2

−2Li2 (−y) ln2 y + 4Li2 (−y) −

1 4 ln y + 4ζ(2)Li2 (−y) 24

5 +ζ(2) ln2 y + 4ζ(3) ln y + ζ(4) , 2 ∞ 1 zn 2n 3 S1 (n − 1) = 4H−1,0,0,1 (−y) + S2,2 y 2 n n n=1

(C.105)

−4S2,2 (y) − 4S2,2 (−y) − 6Li4 (−y) − 2Li4 (y) + 4S1,2 (−y) ln y +4S1,2 (y) ln y − 2S1,2 y 2 ln y + 4Li3 (−y) ln(1 − y) +2Li3 (−y) ln y + 2Li3 (y) ln y − Li2 (y) ln2 y 1 1 4 ln y −4Li2 (−y) ln y ln(1 − y) − ln3 y ln(1 − y) + 3 24 1 +2ζ(2)Li2 (y) − ζ(2) ln2 y + 2ζ(2) ln y ln(1 − y) 2 +6ζ(3) ln(1 − y) − 3ζ(3) ln y − 4ζ(4) , (C.106) ∞ n 1 z 1−y 2n S1 (n − 1)3 = −48S1,2 (−y) ln(1 + y) − 48S1,3 (−y) n 1 +y n n=1 +24S2,2 (−y) − 12ζ(2) ln2 (1 + y) − 24 ln2 (1 + y)Li2 (−y) +24ζ(3) ln(1 + y) + 24 ln(1 + y)Li3 (−y) − 8 ln y ln3 (1 + y) +12ζ(2) ln y ln(1 + y) + 6 ln2 y ln2 (1 + y) − ln3 y ln(1 + y) 3 1 + ln4 y − ζ(2) ln2 y + 3 ln2 y Li2 (−y) 24 2 + ln2 y Li2 (y) − 5ζ(3) ln y − 12 ln y Li3 (−y) − 4 ln y Li3 (y) 3 (C.107) + ζ(4) + 12Li4 (−y) + 6Li4 (y) . 2

200

C Summation Formulae

C.4 Power Series of Levels 5 and 6 in Terms of HPL ∞ zn = H0,0,0,0,1 (z) , n5 n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

(C.108)

S1 (n − 1)

zn = H0,0,0,1,1 (z) , n4

(C.109)

S2 (n − 1)

zn = H0,0,1,0,1 (z) , n3

(C.110)

S1 (n − 1)2 S3 (n − 1)

zn = H0,0,1,0,1 (z) + 2H0,0,1,1,1 (z) , n3

zn = H0,1,0,0,1 (z) , n2

S1 (n − 1)3

n=1

S1 (n − 1)S2 (n − 1)

n=1

n=1 ∞ n=1 ∞ n=1 ∞

S12 (n − 1) S4 (n − 1)

z = H0,1,0,0,1 (z) + H0,1,1,0,1 (z) , n2

zn = H1,0,0,0,1 (z) + H1,1,0,0,1 (z) , n

S1 (n − 1)S3 (n − 1)

n=1

n=1 ∞ n=1

S2 (n − 1)2

(C.115) (C.116) (C.117)

zn = H1,0,0,0,1 (z) + H1,0,0,1,1 (z) n +H1,1,0,0,1 (z) ,

∞

(C.114)

n

zn = H1,0,0,0,1 (z) , n

S13 (n − 1)

(C.113)

zn = H0,1,0,0,1 (z) + H0,1,0,1,1 (z) n2 +H0,1,1,0,1 (z) ,

∞

(C.112)

zn = H0,1,0,0,1 (z) + 3H0,1,0,1,1 (z) n2 +3H0,1,1,0,1 (z) + 6H0,1,1,1,1 (z) ,

∞

(C.111)

(C.118)

n

z = H1,0,0,0,1 (z) + 2H1,0,1,0,1 (z) , n

S1 (n − 1)S12 (n − 1)

(C.119)

zn = H1,0,0,0,1 (z) + H1,0,0,1,1 (z) n

+H1,0,1,0,1 (z) + 2H1,1,0,0,1 (z) + H1,1,0,1,1 (z) + 2H1,1,1,0,1 (z) , (C.120) ∞ zn = H1,0,0,0,1 (z) + 2H1,0,0,1,1 (z) S1 (n − 1)2 S2 (n − 1) n n=1

C.4 Power Series of Levels 5 and 6 in Terms of HPL

201

+2H1,0,1,0,1 (z) + 2H1,0,1,1,1 (z) + 2H1,1,0,0,1 (z) +2H1,1,0,1,1 (z) + 2H1,1,1,0,1 (z) , ∞

S1 (n − 1)4

n=1

(C.121)

n

z = H1,0,0,0,1 (z) + 4H1,0,0,1,1 (z) + 6H1,0,1,0,1 (z) n +12H1,0,1,1,1 (z) + 4H1,1,0,0,1 (z) + 12H1,1,0,1,1 (z) +12H1,1,1,0,1 (z) + 24H1,1,1,1,1 (z) , (C.122)

∞

S112 (n − 1)

n=1

zn = H1,0,0,0,1 (z) + H1,0,1,0,1 (z) + H1,1,0,0,1 (z) n +H1,1,1,0,1 (z) ,

∞

zn = H0,0,0,0,0,1 (z) , n6 n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

(C.124)

S1 (n − 1)

zn = H0,0,0,0,1,1 (z) , n5

(C.125)

S2 (n − 1)

zn = H0,0,0,1,0,1 (z) , n4

(C.126)

S1 (n − 1)2 S3 (n − 1)

zn = H0,0,0,1,0,1 (z) + 2H0,0,0,1,1,1 (z) , n4

zn = H0,0,1,0,0,1 (z) , n3

S1 (n − 1)3

n=1

S1 (n − 1)S2 (n − 1)

n=1

n=1 ∞ n=1 ∞ n=1 ∞ n=1

S12 (n − 1) S4 (n − 1)

zn = H0,0,1,0,0,1 (z) + H0,0,1,1,0,1 (z) , n3

zn = H0,1,0,0,0,1 (z) + H0,1,1,0,0,1 (z) , n2

S1 (n − 1)S3 (n − 1)

(C.129)

z = H0,0,1,0,0,1 (z) + H0,0,1,0,1,1 (z) n3

zn = H0,1,0,0,0,1 (z) , n2

S13 (n − 1)

(C.128)

n

+H0,0,1,1,0,1 (z) , ∞

(C.127)

zn = H0,0,1,0,0,1 (z) + 3H0,0,1,0,1,1 (z) n3 +3H0,0,1,1,0,1 (z) + 6H0,0,1,1,1,1 (z) ,

∞

(C.123)

zn = H0,1,0,0,0,1 (z) + H0,1,0,0,1,1 (z) n2

(C.130) (C.131) (C.132) (C.133)

202

C Summation Formulae

+H0,1,1,0,0,1 (z) , ∞ n=1 ∞

S2 (n − 1)2

zn = H0,1,0,0,0,1 (z) + 2H0,1,0,1,0,1 (z) , n2

S1 (n − 1)S12 (n − 1)

n=1

(C.134) (C.135)

zn = H0,1,0,0,0,1 (z) + H0,1,0,0,1,1 (z) n2 +H0,1,0,1,0,1 (z) + 2H0,1,1,0,0,1 (z) +H0,1,1,0,1,1 (z) + 2H0,1,1,1,0,1 (z) , (C.136)

∞

S1 (n − 1)2 S2 (n − 1)

n=1

n

z = H0,1,0,0,0,1 (z) + 2H0,1,0,0,1,1 (z) n2

+2H0,1,0,1,0,1 (z) + 2H0,1,0,1,1,1 (z) + 2H0,1,1,0,0,1 (z) +2H0,1,1,0,1,1 (z) + 2H0,1,1,1,0,1 (z) , ∞ n=1

S1 (n − 1)4

(C.137)

n

z = H0,1,0,0,0,1 (z) + 4H0,1,0,0,1,1 (z) n2

+6H0,1,0,1,0,1 (z) + 12H0,1,0,1,1,1 (z) + 4H0,1,1,0,0,1 (z) +12H0,1,1,0,1,1 (z) + 12H0,1,1,1,0,1 (z) + 24H0,1,1,1,1,1 (z) , (C.138) ∞ n=1

S112 (n − 1)

zn = H0,1,0,0,0,1 (z) + H0,1,0,1,0,1 (z) n2 +H0,1,1,0,0,1 (z) + H0,1,1,1,0,1 (z) ,

∞ n=1

S1 (n − 1)S4 (n − 1)

z = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n +H1,1,0,0,0,1 (z) ,

∞

(C.139)

n

S1 (n − 1)S13 (n − 1)

n=1

(C.140)

n

z = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n +H1,0,1,0,0,1 (z) + 2H1,1,0,0,0,1 (z) +H1,1,0,0,1,1 (z) + 2H1,1,1,0,0,1 (z) ,

∞

S1 (n − 1)2 S3 (n − 1)

n=1

(C.141)

n

z = H1,0,0,0,0,1 (z) + 2H1,0,0,0,1,1 (z) n

+H1,0,0,1,0,1 (z) + 2H1,0,0,1,1,1 (z) + H1,0,1,0,0,1 (z) +2H1,1,0,0,0,1 (z) + 2H1,1,0,0,1,1 (z) + 2H1,1,1,0,0,1 (z) , (C.142) ∞ n=1

S1 (n − 1)S2 (n − 1)2

zn = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n

+2H1,0,0,1,0,1 (z) + 2H1,0,1,0,0,1 (z) + 2H1,0,1,0,1,1 (z) +2H1,0,1,1,0,1 (z) + H1,1,0,0,0,1 (z) + 2H1,1,0,1,0,1 (z) ,

(C.143)

C.4 Power Series of Levels 5 and 6 in Terms of HPL ∞

S1 (n − 1)2 S12 (n − 1)

n=1

203

zn = H1,0,0,0,0,1 (z) + 2H1,0,0,0,1,1 (z) n

+2H1,0,0,1,0,1 (z) + 2H1,0,0,1,1,1 (z) + 3H1,0,1,0,0,1 (z) +2H1,0,1,0,1,1 (z) + 3H1,0,1,1,0,1 (z) + 3H1,1,0,0,0,1 (z) +4H1,1,0,0,1,1 (z) + 4H1,1,0,1,0,1 (z) + 2H1,1,0,1,1,1 (z) +6H1,1,1,0,0,1 (z) + 4H1,1,1,0,1,1 (z) + 6H1,1,1,1,0,1 (z) , (C.144) ∞

S1 (n − 1)3 S2 (n − 1)

n=1

zn = H1,0,0,0,0,1 (z) + 3H1,0,0,0,1,1 (z) n

+4H1,0,0,1,0,1 (z) + 6H1,0,0,1,1,1 (z) + 4H1,0,1,0,0,1 (z) +6H1,0,1,0,1,1 (z) + 6H1,0,1,1,0,1 (z) + 6H1,0,1,1,1,1 (z) +3H1,1,0,0,0,1 (z) + 6H1,1,0,0,1,1 (z) + 6H1,1,0,1,0,1 (z) +6H1,1,0,1,1,1 (z) + 6H1,1,1,0,0,1 (z) +6H1,1,1,0,1,1 (z) + 6H1,1,1,1,0,1 (z) , ∞

S1 (n − 1)5

n=1

(C.145)

n

z = H1,0,0,0,0,1 (z) + 5H1,0,0,0,1,1 (z) + 10H1,0,0,1,0,1 (z) n

+20H1,0,0,1,1,1 (z) + 10H1,0,1,0,0,1 (z) + 30H1,0,1,0,1,1 (z) +30H1,0,1,1,0,1 (z) + 60H1,0,1,1,1,1 (z) + 5H1,1,0,0,0,1 (z) +20H1,1,0,0,1,1 (z) + 30H1,1,0,1,0,1 (z) + 60H1,1,0,1,1,1 (z) +20H1,1,1,0,0,1 (z) + 60H1,1,1,0,1,1 (z) +60H1,1,1,1,0,1 (z) + 120H1,1,1,1,1,1 (z) , ∞

S1 (n − 1)S112 (n − 1)

n=1

(C.146)

n

z = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n

+H1,0,0,1,0,1 (z) + 2H1,0,1,0,0,1 (z) + H1,0,1,0,1,1 (z) +2H1,0,1,1,0,1 (z) + 2H1,1,0,0,0,1 (z) + H1,1,0,0,1,1 (z) +2H1,1,0,1,0,1 (z) + 3H1,1,1,0,0,1 (z) +H1,1,1,0,1,1 (z) + 3H1,1,1,1,0,1 (z) , ∞ n=1 ∞ n=1 ∞ n=1

S5 (n − 1)

z = H1,0,0,0,0,1 (z) , n

S14 (n − 1)

(C.147)

n

zn = H1,0,0,0,0,1 (z) + H1,1,0,0,0,1 (z) , n

S2 (n − 1)S3 (n − 1)

(C.148) (C.149)

zn = H1,0,0,0,0,1 (z) + H1,0,0,1,0,1 (z) n +H1,0,1,0,0,1 (z) ,

(C.150)

204

C Summation Formulae ∞ n=1 ∞

S23 (n − 1)

zn = H1,0,0,0,0,1 (z) + H1,0,1,0,0,1 (z) , n

S12 (n − 1)S2 (n − 1)

n=1

zn = H1,0,0,0,0,1 (z) + 2H1,0,0,1,0,1 (z) n +H1,0,1,0,0,1 (z) + H1,0,1,1,0,1 (z) +H1,1,0,0,0,1 (z) + 2H1,1,0,1,0,1 (z) ,

∞

S113 (n − 1)

n=1

S212 (n − 1)

n=1

z = H1,0,0,0,0,1 (z) + H1,0,1,0,0,1 (z) n z = H1,0,0,0,0,1 (z) + H1,0,0,1,0,1 (z) n

S1112 (n − 1)

n=1

(C.153)

n

+H1,0,1,0,0,1 (z) + H1,0,1,1,0,1 (z) , ∞

(C.152)

n

+H1,1,0,0,0,1 (z) + H1,1,1,0,0,1 (z) , ∞

(C.151)

(C.154)

n

z = H1,0,0,0,0,1 (z) + H1,0,0,1,0,1 (z) + H1,0,1,0,0,1 (z) n

+H1,0,1,1,0,1 (z) + H1,1,0,0,0,1 (z) + H1,1,0,1,0,1 (z) +H1,1,1,0,0,1 (z) + H1,1,1,1,0,1 (z) .

(C.155)

References 1. J. Bl¨ umlein, Comput. Phys. Commun. 159 (2004) 19. 191 2. J.M. Borwein, D.M. Bradley and D.J. Broadhurst, Electronic J. Combinatorics, 4(2) (1997) R5; J.M. Borwein, D.M. Bradley, D.J. Broadhurst and P. Lisonˇek, Electronic J. Combinatorics, 5(1) (1998) R38; Trans. Amer. Math. Soc. 355 (2001) 907. 191 3. A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 699 (2004) 3. 191, 198 4. A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 605 (2001) 266; Phys. Rev. D 61 (2000) 087701. 191 5. A. Devoto and D.W. Duke, Riv. Nuovo Cim. 7, No. 6 (1984) 1. 6. J. Fleischer, A.V. Kotikov and O.L. Veretin, Nucl. Phys. B 547 (1999) 343. 197 7. A.G. Grozin, Int. J. Mod. Phys. A 19 (2004) 473. 191 8. M.Yu. Kalmykov and O. Veretin, Phys. Lett. B 483 (2000) 315. 198 9. K.S. K¨ olbig, J.A. Mignaco and E. Remiddi, BIT 10 (1970) 38; K.S. K¨ olbig, Math. Comp. 39 (1982) 647. 10. L. Lewin, Polylogarithms and Associated Functions (North-Holland, Amsterdam, 1981). 11. S. Moch, P. Uwer and S. Weinzierl, J. Math. Phys. 43 (2002) 3363. 191 12. S. Moch, P. Uwer and S. Weinzierl, Phys. Rev. D 66 (2002) 114001. 191 13. S. Moch and J. A. M. Vermaseren, Nucl. Phys. B 573 (2000) 853. 191 14. S. Moch, J. A. M. Vermaseren and A. Vogt, Nucl. Phys. B 688 (2004) 101; A. Vogt, S. Moch and J.A.M. Vermaseren, Nucl. Phys. B 691 (2004) 129. 191 15. E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725.

References

205

16. J.A.M. Vermaseren, Symbolic Manipulation with FORM (CAN, Amsterdam, 1991). 191 17. J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037. 191 18. S. Weinzierl, Comput. Phys. Commun. 145 (2002) 357. 191 19. S. Weinzierl, J. Math. Phys. 45 (2004) 2656. 191

D Table of MB Integrals

D.1 MB Integrals with Four Gamma Functions This is the ﬁrst Barnes lemma: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ Γ (λ1 + λ3 )Γ (λ1 + λ4 )Γ (λ2 + λ3 )Γ (λ2 + λ4 ) = . Γ (λ1 + λ2 + λ3 + λ4 )

(D.1)

Results for integrals with ψ(λ1 + z), . . . are obtained from (D.1) by differentiating with respect to λ1 , . . .. Second derivatives give, in a similar way, results for integrals with products of two diﬀerent functions ψ(λi ± z) and with the combinations ψ (λi ± z) + ψ(λi ± z)2 . Various corollaries can be derived from (D.1). For example, +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z) 2πi −i∞ = Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ(λ1 − λ2 ) − ψ(λ1 + λ3 )] , (D.2) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z) = Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ(λ2 + λ3 ) − ψ(λ1 + λ3 )] .

(D.3)

The asterisk is used to indicate that the ﬁrst pole of the corresponding gamma function is of the opposite nature, i.e. the ﬁrst pole of Γ (λ2 + z) in (D.2) is considered right and the ﬁrst pole of Γ (−λ2 − z) in (D.3) is considered left. These are four formulae with the psi function with the same condition as in (D.2): +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(λ1 + z) 2πi −i∞

= Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ1 − λ2 )2 − ψ(λ1 − λ2 )ψ(λ1 + λ3 ) +ψ (λ1 − λ2 ) − ψ (λ1 + λ3 )] , (D.4)

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 207–219 (2004) c Springer-Verlag Berlin Heidelberg 2004

208

D Table of MB Integrals

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(λ2 + z)

1 = − Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ1 − λ2 )2 − ψ(λ1 + λ3 )2 2 +2ψ(λ1 − λ2 )(γE − ψ(λ2 + λ3 )) − 2ψ(λ1 + λ3 )(γE − ψ(λ2 + λ3 )) +ψ (λ1 − λ2 ) + ψ (λ1 + λ3 )] , (D.5) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(−λ2 − z)

1 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ1 − λ2 )2 + 2γE ψ(λ1 + λ3 ) 2 +ψ(λ1 + λ3 )2 − 2ψ(λ1 − λ2 )(γE + ψ(λ1 + λ3 )) +ψ (λ1 − λ2 ) − ψ (λ1 + λ3 )] ,

=

1 2πi

+i∞

−i∞

(D.6)

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(λ3 − z)

= Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ(λ1 − λ2 )ψ(λ2 + λ3 ) −ψ(λ1 + λ3 )ψ(λ2 + λ3 ) − ψ (λ1 + λ3 )] .

(D.7)

These are four formulae with the psi function with the same condition as in (D.3): +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(λ1 + z) 2πi −i∞ = −Γ (λ1 − λ2 )Γ (λ2 + λ3 ) × [ψ(λ1 − λ2 )(ψ(λ1 + λ3 ) − ψ(λ2 + λ3 )) + ψ (λ1 + λ3 )] , (D.8) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(λ2 + z)

1 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) (ψ(λ1 + λ3 ) − ψ(λ2 + λ3 ))2 2 +2γE (ψ(λ1 + λ3 ) − ψ(λ2 + λ3 )) − ψ (λ1 + λ3 ) + ψ (λ2 + λ3 )] , (D.9)

=

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(−λ2 − z)

1 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) 2 × [2(ψ(λ1 − λ2 ) − γE )(ψ(λ2 + λ3 ) − ψ(λ1 + λ3 )) +ψ(λ1 + λ3 )2 − ψ(λ2 + λ3 )2 − ψ (λ1 + λ3 ) − ψ (λ2 + λ3 ) , (D.10)

=

D.1 MB Integrals with Four Gamma Functions

1 2πi

209

+i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(λ3 − z)

= Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ2 + λ3 )2 − ψ(λ1 + λ3 )ψ(λ2 + λ3 ) −ψ (λ1 + λ3 ) + ψ (λ2 + λ3 )] , (D.11)

−i∞

This is an example with the gluing of two poles: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗∗ (−1 − λ2 − z)Γ (λ3 − z) 2πi −i∞ = Γ (λ1 − λ2 − 1)Γ (λ2 + λ3 ) [1 − λ1 + λ2 +(λ1 + λ3 − 1)(ψ(λ1 + λ3 − 1) − ψ(λ2 + λ3 ))] ,

(D.12)

where the ﬁrst two poles of Γ (−1 − λ2 − z), i.e. z = −λ2 and z = −λ2 − 1, are considered left, with the corresponding change in notation. Here it is implied that λ1 + λ3 = 1. In the case λ1 + λ3 = 1, we have +i∞ 1 dz Γ (1 − λ1 + z)Γ (λ2 + z)Γ ∗∗ (−1 − λ2 − z)Γ (λ1 − z) 2πi −i∞ = (λ1 + λ2 − 1)Γ (λ1 + λ2 )Γ (−λ1 − λ2 ) . (D.13) Here is one more example of such an integral: +i∞ 1 dz Γ (1 − λ1 + z)Γ ∗ (λ2 + z)Γ ∗ (−1 − λ2 − z)Γ (λ1 − z) 2πi −i∞ = Γ (λ1 + λ2 )Γ (−λ1 − λ2 ) × [(λ1 + λ2 )(ψ(−λ1 − λ2 ) − ψ(1 + λ1 + λ2 )) − 1] . (D.14) Furthermore, we have +i∞ 1 dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z) 2πi −i∞ = Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [2γE + ψ(λ1 − λ2 ) + ψ(λ2 − λ1 )] ,

(D.15)

where the poles z = −λ1 and z = −λ2 are right. These are four more formulae with these conditions: +i∞ 1 dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(λ1 + z) 2πi −i∞

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4ψ(λ1 − λ2 )ψ(λ2 − λ1 ) 4 +4γE (ψ(λ2 − λ1 ) − 2ψ(λ1 − λ2 )) − 4ψ(λ1 − λ2 )2 − 4ψ (λ1 − λ2 ) +2ψ(λ2 − λ1 )2 + 2ψ (λ2 − λ1 ) , (D.16) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(λ2 + z)

210

D Table of MB Integrals

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 2ψ(λ1 − λ2 )2 4 +4ψ(λ1 − λ2 )(γE − ψ(λ2 − λ1 )) − 8γE ψ(λ2 − λ1 ) − 4ψ(λ2 − λ1 )2 +2ψ (λ1 − λ2 ) − 4ψ (λ2 − λ1 )] , (D.17) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(−λ2 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 2ψ(λ1 − λ2 )2 4 −4ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) − 2ψ (λ1 − λ2 )] ,

1 2πi

+i∞

−i∞

(D.18)

dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(−λ1 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 2ψ(λ2 − λ1 )2 4 −4(γE + ψ(λ1 − λ2 ))ψ(λ2 − λ1 ) − 2ψ (λ2 − λ1 )] .

(D.19)

There are similar formulae with diﬀerent understanding of the nature of the poles: +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ = 2Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [γE + ψ(λ1 − λ2 )] , (D.20) where the pole z = −λ1 is left and the pole and z = −λ2 is right, and +i∞ 1 dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z) 2πi −i∞ = 2Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [γE + ψ(λ2 − λ1 )] , (D.21) where the pole z = −λ1 is right and the pole and z = −λ2 is left. These are four more formulae with these conditions: +i∞ 1 dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(λ1 + z) 2πi −i∞

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 4γE ψ(λ2 − λ1 ) + 2ψ(λ2 − λ1 )2 4 −8ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) + 2ψ (λ2 − λ1 )] , (D.22) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(λ2 + z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4γE ψ(λ2 − λ1 ) 4 (D.23) −6ψ(λ2 − λ1 )2 − 6ψ (λ2 − λ1 ) ,

D.1 MB Integrals with Four Gamma Functions

1 2πi

+i∞

−i∞

211

dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(−λ2 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 4γE ψ(λ2 − λ1 ) + 2ψ(λ2 − λ1 )2 4 −8ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) + 2ψ (λ2 − λ1 )] , (D.24) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(−λ1 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4γE ψ(λ2 − λ1 ) 4 (D.25) −6ψ(λ2 − λ1 )2 − 6ψ (λ2 − λ1 ) .

Furthermore, we have +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ = Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [2γE + ψ(λ1 − λ2 ) + ψ(λ2 − λ1 )] ,

(D.26)

where the poles z = −λ1 and z = −λ2 are left. These are four more formulae with these conditions: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(λ1 + z) 2πi −i∞

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 2ψ(λ1 − λ2 )2 4 −4ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) − 2ψ (λ1 − λ2 )] , (D.27) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(λ2 + z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4(γE + ψ(λ1 − λ2 ))ψ(λ2 − λ1 ) 4 (D.28) −2ψ(λ2 − λ1 )2 − 2ψ (λ2 − λ1 ) , 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(−λ2 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4ψ(λ1 − λ2 )2 4 +4γE ψ(λ2 − λ1 ) + 2ψ(λ2 − λ1 )2 − 4ψ(λ1 − λ2 )(2γE + ψ(λ2 − λ1 )) −4ψ (λ1 − λ2 ) + 2ψ (λ2 − λ1 )] , 1 2πi

+i∞

−i∞

(D.29)

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(−λ1 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 2ψ(λ1 − λ2 )2 4

212

D Table of MB Integrals

+4ψ(λ1 − λ2 )(γE − ψ(λ2 − λ1 )) − 8γE ψ(λ2 − λ1 ) −4ψ(λ2 − λ1 )2 + 2ψ (λ1 − λ2 ) − 4ψ (λ2 − λ1 ) .

(D.30)

We also have +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 2πi −i∞

= −Γ (λ1 − λ2 )ψ (λ1 − λ2 ) ,

(D.31)

where the pole z = −λ2 is right. These are three more formulae with this condition: +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 ψ(λ1 + z) 2πi −i∞ = −Γ (λ1 − λ2 ) [ψ(λ1 − λ2 )ψ (λ1 − λ2 ) + ψ (λ1 − λ2 )] ,

1 2πi

1 2πi

+i∞

−i∞

+i∞

−i∞

(D.32)

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 ψ(λ2 + z) = Γ (λ1 − λ2 )ψ (λ1 − λ2 ) [2γE + ψ(λ1 − λ2 )] ,

(D.33)

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 ψ(−λ2 − z) 1 Γ (λ1 − λ2 ) [2γE ψ (λ1 − λ2 ) − ψ (λ1 − λ2 )] . 2

(D.34)

We also have +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)2 2πi −i∞

1 = Γ (λ1 − λ2 ) π 2 + 2(γE + ψ(λ1 − λ2 ))2 − 2ψ (λ1 − λ2 ) , 4 where the pole z = −λ2 is left, +i∞ 1 dz Γ (λ1 + z)2 Γ ∗ (−λ1 − z)Γ (λ2 − z) 2πi −i∞

(D.35)

=

= −Γ (λ1 + λ2 )ψ (λ1 + λ2 ) ,

(D.36)

where the pole z = −λ1 is left, and +i∞ 1 dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z) 2πi −i∞

1 = Γ (λ1 + λ2 ) 2(γE + ψ(λ1 + λ2 ))2 + π 2 − 2ψ (λ1 + λ2 ) , (D.37) 4 where the pole z = −λ1 is right. These are three more formulae with this condition:

D.1 MB Integrals with Four Gamma Functions

1 2πi

213

+i∞

dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z)ψ(λ1 + z) π2 1 3 2 = Γ (λ1 + λ2 ) ψ(λ1 + λ2 ) + 3ψ(λ1 + λ2 ) ψ (λ1 + λ2 ) − γE + 6 6 −2γE3 − γE π 2 + 6γE ψ (λ1 + λ2 ) − 4ζ(3) − 2ψ (λ1 + λ2 ) , (D.38)

1 2πi

−i∞

+i∞

−i∞

dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z)ψ(−λ1 − z)

1 Γ (λ1 + λ2 ) 12γE ψ(λ1 + λ2 )2 + 2ψ(λ1 + λ2 )3 12 π2 2 − 2ψ (λ1 + λ2 ) +3ψ(λ1 + λ2 ) 6γE + 3

=−

+2(4γE3 + 2γE π 2 − 6γE ψ (λ1 + λ2 ) + 8ζ(3) + ψ (λ1 + λ2 )) ,

1 2πi

+i∞

−i∞

(D.39)

dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z)ψ(λ2 − z)

1 = Γ (λ1 + λ2 ) 4γE ψ(λ1 + λ2 )2 + 2ψ(λ1 + λ2 )3 + 4γE ψ (λ1 + λ2 ) 4 (D.40) +ψ(λ1 + λ2 )(2γE2 + π 2 + 2ψ (λ1 + λ2 )) − 2ψ (λ1 + λ2 ) . In some situations, it is possible to evaluate MB integrals with higher derivatives of the ψ function. Here are some examples: +i∞ 1 dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z) 2πi −i∞ = 1 2πi

Γ (λ1 + λ2 )4 [2ψ(λ1 + λ2 ) − ψ(2(λ1 + λ2 ))] , Γ (2(λ1 + λ2 ))

(D.41)

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z)2 Γ (λ1 + λ2 )4 4ψ(λ1 + λ2 )2 − 4ψ(λ1 + λ2 )ψ(2(λ1 + λ2 )) Γ (2(λ1 + λ2 )) +ψ(2(λ1 + λ2 ))2 − ψ (2(λ1 + λ2 )) , (D.42)

=

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ (λ1 + z) =2

1 2πi

Γ (λ1 + λ2 )4 ψ (λ1 + λ2 ) , Γ (2(λ1 + λ2 ))

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z)ψ(λ2 − z)

(D.43)

214

D Table of MB Integrals

Γ (λ1 + λ2 )4 4ψ(λ1 + λ2 )2 − 4ψ(λ1 + λ2 )ψ(2(λ1 + λ2 )) Γ (2(λ1 + λ2 )) +ψ(2(λ1 + λ2 ))2 + ψ (λ1 + λ2 ) − ψ (2(λ1 + λ2 )) , (D.44)

=

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z)2 ψ(λ2 − z)

Γ (λ1 + λ2 )4 8ψ(λ1 + λ2 )3 − 12ψ(λ1 + λ2 )2 ψ(2(λ1 + λ2 )) Γ (2(λ1 + λ2 ))

=

+2ψ(λ1 + λ2 )(3ψ(2(λ1 + λ2 ))2 + 2ψ (λ1 + λ2 ) − 3ψ (2(λ1 + λ2 ))) +ψ(2(λ1 + λ2 ))(3ψ (2(λ1 + λ2 )) − 2ψ (λ1 + λ2 )) −ψ(2(λ1 + λ2 ))3 − ψ (2(λ1 + λ2 )) , (D.45) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ (λ1 + z)ψ(λ2 − z) Γ (λ1 + λ2 )4 [4ψ(λ1 + λ2 )ψ (λ1 + λ2 ) Γ (2(λ1 + λ2 )) −2ψ(2(λ1 + λ2 ))ψ (λ1 + λ2 ) + ψ (λ1 + λ2 )] ,

=

(D.46)

D.2 MB Integrals with Six Gamma Functions This is the second Barnes lemma: +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ (λ4 − z)Γ (λ5 − z) dz 2πi −i∞ Γ (λ6 + z) Γ (λ1 + λ4 )Γ (λ2 + λ4 )Γ (λ3 + λ4 )Γ (λ1 + λ5 ) = Γ (λ1 + λ2 + λ4 + λ5 )Γ (λ1 + λ3 + λ4 + λ5 ) Γ (λ2 + λ5 )Γ (λ3 + λ5 ) , × Γ (λ2 + λ3 + λ4 + λ5 )

(D.47)

where λ6 = λ1 + λ2 + λ3 + λ4 + λ5 . Here is a collection of its corollaries: +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ ∗ (−λ3 − z)Γ (λ4 − z) dz 2πi −i∞ Γ (λ5 + z) Γ (λ1 − λ3 )Γ (λ2 − λ3 )Γ (λ3 + λ4 ) [ψ(λ1 + λ2 − λ3 + λ4 ) = Γ (λ1 + λ2 − λ3 + λ4 ) (D.48) +ψ(λ3 + λ4 ) − ψ(λ1 + λ4 ) − ψ(λ2 + λ4 )] , where λ5 = λ1 + λ2 + λ4 and the pole z = −λ3 is considered left, +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (λ3 + z)Γ (−λ3 − z)Γ (λ4 − z) dz 2πi −i∞ Γ (λ5 + z)

D.2 MB Integrals with Six Gamma Functions

215

Γ (λ1 − λ3 )Γ (λ2 − λ3 )Γ (λ3 + λ4 ) [ψ(λ1 − λ3 ) + ψ(λ2 − λ3 ) Γ (λ1 + λ2 − λ3 + λ4 ) −ψ(λ1 + λ4 ) − ψ(λ2 + λ4 )] , (D.49)

=

where λ5 = λ1 + λ2 + λ4 and the pole z = −λ3 is considered right, +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ3 + z)Γ (λ3 − z)2 dz 2πi −i∞ Γ (λ4 + z) Γ (λ1 + λ3 )Γ (λ2 + λ3 ) [ψ (λ1 + λ3 ) + ψ (λ2 + λ3 )] , (D.50) =− Γ (λ1 + λ2 + 2λ3 ) where λ4 = λ1 + λ2 + λ3 and the pole z = λ3 is considered right, +i∞ 1 Γ (λ1 + z)Γ ∗ (λ2 + z)2 Γ (−λ2 − z)Γ (λ3 − z) dz 2πi −i∞ Γ (λ4 + z) 2 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) π + (γE − ψ(λ1 − λ2 ) + ψ(λ1 + λ3 ) = 2Γ (λ1 + λ3 ) 2 +ψ(λ2 + λ3 ))2 + ψ (λ1 − λ2 ) + ψ (λ1 + λ3 ) − ψ (λ2 + λ3 ) , (D.51) where λ4 = λ1 + λ2 + λ3 and the pole z = −λ2 is considered right, +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)2 Γ ∗ (−λ2 − z)Γ (λ3 − z) dz 2πi −i∞ Γ (λ4 + z) Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ (λ1 + λ3 ) − ψ (λ2 + λ3 )] , (D.52) = Γ (λ1 + λ3 ) where λ4 = λ1 + λ2 + λ3 and the pole z = −λ2 is considered left. The integrals (D.47) can be evaluated recursively in the case where the diﬀerence λ6 − (λ1 + λ2 + λ3 + λ4 + λ5 ) is a positive integer. In particular, we have +i∞ Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ (λ4 − z)Γ (−z) 1 dz 2πi −i∞ Γ (λ5 + z) (Γ (1 + λ2 + λ3 + λ4 ))−1 Γ (λ1 )Γ (λ3 )Γ (λ2 + λ4 ) = Γ (1 − λ1 − λ3 − λ4 )Γ (1 + λ1 + λ2 + λ4 )Γ (λ1 + λ3 + λ4 ) × [Γ (1 + λ2 )Γ (1 − λ1 − λ3 − λ4 )Γ (λ1 + λ4 )Γ (λ3 + λ4 ) −Γ (λ2 )Γ (−λ1 − λ3 − λ4 )Γ (1 + λ1 + λ4 )Γ (1 + λ3 + λ4 )] , (D.53) where λ5 = λ1 + λ2 + λ3 + λ4 + 1, and +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ (λ4 − z)Γ (−z) dz 2πi −i∞ Γ (λ5 + z) −1 (Γ (2 + λ2 + λ3 + λ4 )) Γ (λ1 )Γ (λ3 )Γ (λ2 + λ4 ) = Γ (1 − λ1 − λ3 − λ4 )Γ (2 + λ1 + λ2 + λ4 )Γ (λ1 + λ3 + λ4 ) × [Γ (2 + λ2 )Γ (1 − λ1 − λ3 − λ4 )Γ (λ1 + λ4 )Γ (λ3 + λ4 ) −2Γ (1 + λ2 )Γ (−λ1 − λ3 − λ4 )Γ (1 + λ1 + λ4 )Γ (1 + λ3 + λ4 ) +Γ (λ2 )Γ (−1 − λ1 − λ3 − λ4 )Γ (2 + λ1 + λ4 )Γ (2 + λ3 + λ4 )] , (D.54) where λ5 = λ1 + λ2 + λ3 + λ4 + 2.

216

D Table of MB Integrals

Here are more corollaries of the second Barnes lemma: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) = Γ (1 − λ1 )Γ (1 − λ2 ) × [Γ (1 − λ1 )Γ (1 − λ2 ) − Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 )] , (D.55) where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered left, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z = −Γ (λ1 )Γ (λ2 )Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 ) Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) + Γ (1 − λ1 )Γ (1 − λ2 ) × [Γ (1 − λ1 )Γ (1 − λ2 ) − Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 )] , (D.56) where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered right, +i∞ 1 Γ ∗ (λ + z)2 Γ ∗ (z)Γ (−z)Γ (−λ − z) dz 2πi −i∞ Γ (λ + 1 + z) +i∞ dz 1 Γ (λ + z)Γ (z)Γ ∗ (−z)Γ ∗ (−λ − z) =− 2πi −i∞ z

1 = Γ (λ)Γ (−λ) 12(γE + ψ(λ)) + 2λπ 2 6λ +3λ((ψ(λ) − ψ(−λ))2 − ψ (λ) + ψ (−λ)) , (D.57) where the nature of the poles at z = 0 and z = −λ is indicated by asterisks, according to our conventions, +i∞ 1 Γ (λ + z)2 Γ (z)Γ ∗ (−z)Γ ∗ (−λ − z) dz 2πi −i∞ Γ (λ + 1 + z) +i∞ dz ∗ 1 1 Γ (λ + z)Γ ∗ (z)Γ (−z)Γ (−λ − z) = 2 Γ (λ)Γ (−λ) =− 2πi −i∞ z λ

π2 2 × 1 + λ(ψ(λ) + ψ(−λ) + 2γE ) − λ ψ (λ) − , (D.58) 6 1 2πi

Γ (λ + z)2 Γ ∗ (z)Γ (−z)Γ ∗ (−λ − z) Γ (λ + 1 + z) +i∞ dz 1 Γ (λ + z)Γ ∗ (z)Γ (−z)Γ ∗ (−λ − z) =− 2πi −i∞ z

π2 1 = Γ (λ)Γ (−λ) 2(γE + ψ(λ)) − λ ψ (λ) − . λ 6

+i∞

dz

−i∞

(D.59)

D.2 MB Integrals with Six Gamma Functions

217

We also have +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z 2 Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 ) = Γ (2 − λ1 )Γ (1 − λ2 ) ×Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) [1 + (λ1 − 1)(ψ(2 − λ1 ) + ψ(1 − λ2 ) −ψ(2 − λ1 − λ2 − λ3 ) − ψ(λ3 ))] ,

(D.60)

where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered left, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z 2 = Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 ) [−Γ (λ1 )Γ (λ2 )(ψ(λ1 ) + ψ(λ2 ) −ψ(2 − λ1 − λ2 − λ3 ) − ψ(λ3 )) Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) + Γ (2 − λ1 )Γ (1 − λ2 ) × [1 + (λ1 − 1)(ψ(2 − λ1 ) + ψ(1 − λ2 ) −ψ(2 − λ1 − λ2 − λ3 ) − ψ(λ3 ))]] ,

(D.61)

where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered right, +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z 1 = − 2 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [2λ1 − λ2 λ1 λ2 +λ1 (λ1 + λ2 )(γE + ψ(λ1 − λ2 )) − λ1 (λ1 − λ2 ) ×(ψ(−λ1 ) − ψ(−λ2 ) + ψ(λ2 − λ1 ) − ψ(1 − λ1 + λ2 ))] , (D.62) where the pole at z = 0 is left and the nature of the ﬁrst poles of the gamma functions is shown by asterisks, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z

1 = 2 2 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) λ21 − λ1 λ2 + λ22 λ1 λ2 −λ1 λ2 (λ1 + λ2 )γE + λ1 (λ1 − λ2 )λ2 (ψ(−λ1 ) − ψ(−λ2 )) −λ1 λ2 (λ2 ψ(λ1 − λ2 ) + λ1 ψ(λ2 − λ1 ))] , where the pole at z = 0 is left, +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z 2

1 = 3 2 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2(λ21 + λ1 λ2 − λ22 ) λ1 λ2

(D.63)

218

D Table of MB Integrals

+λ1 (λ21 + λ22 )(ψ(λ1 − λ2 ) + γE ) −λ1 (λ21 − λ22 )(ψ(−λ1 ) − ψ(−λ2 ) + ψ(−λ1 + λ2 ) − ψ(1 − λ1 + λ2 )) −λ21 λ2 (λ1 − λ2 )(ψ (−λ1 ) − ψ (−λ2 )) , (D.64) where the pole at z = 0 is left, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z 2

1 = − 3 3 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) (λ1 + λ2 )(2λ21 − 3λ1 λ2 + 2λ22 ) λ1 λ2 −λ1 λ2 (λ21 + λ22 )γE + λ1 λ2 (λ21 − λ22 )ψ(−λ1 ) −λ1 λ32 (ψ(λ1 − λ2 ) − ψ(−λ2 )) − λ31 λ2 (ψ(−λ2 ) + ψ(λ2 − λ1 )) +λ31 λ22 (ψ (−λ1 ) − ψ (−λ2 )) − λ21 λ32 (ψ (−λ1 ) − ψ (−λ2 )) , where the pole at z = 0 is left, +i∞ 1 dz Γ (λ + z)Γ (z)Γ ∗ (−z)Γ ∗ (−λ − z) 2πi −i∞ z 2 1 = − 3 Γ (λ)Γ (−λ) [12 − 6λ(2γE + ψ(−λ) + ψ(λ)) 6λ +λ2 (π 2 − 6ψ (−λ)) − 3λ3 (ψ (−λ) + 2ζ(3)) ,

(D.65)

(D.66)

where the pole at z = 0 is left, +i∞ 1 dz Γ (λ + z)Γ ∗ (z)Γ (−z)Γ ∗ (−λ − z) 2πi −i∞ z 2

1 = 3 Γ (λ)Γ (−λ) −12 + 6λ(2γE + ψ(−λ) + ψ(λ)) − λ2 (π 2 − 6ψ (−λ)) 6λ −λ3 (π 2 (ψ(−λ) − ψ(λ)) + (ψ(−λ) − ψ(λ))3 − 2ψ (−λ) − ψ (λ) (D.67) +3(ψ(−λ) − ψ(λ))(ψ (−λ) + ψ (λ)) − 6ζ(3))] , where the pole at z = 0 is right, +i∞ 1 dz Γ (λ + z)2 Γ ∗ (−λ − z)2 2πi −i∞ z 1 = − 4 6 + λ2 (π 2 − 6ψ (−λ)) + 12λ3 ζ(3) , 6λ where the pole at z = 0 is left, +i∞ 1 dz Γ (λ + z)2 Γ ∗ (−λ − z)2 2πi −i∞ z 2 1 = 5 12 + λ2 (π 2 − 6ψ (−λ)) − 3λ3 (ψ (−λ) − 2ζ(3)) , 3λ where the pole at z = 0 is left,

(D.68)

(D.69)

D.2 MB Integrals with Six Gamma Functions

1 2πi

+i∞

−i∞

219

dz Γ (λ + 1 + z)2 Γ (−λ − z)2 z2

= 2Γ (1 + λ)2 Γ (−λ)2 (ψ(−λ) − ψ(1 + λ)) − ψ (−λ) , where the pole at z = 0 is right.

(D.70)

E Analysis of Convergence and Sector Decompositions

In this appendix, the analysis of convergence of Feynman integrals based on the alpha representation is brieﬂy described. The UV divergences come from the region of small values of the α-parameters in (2.36), while the oﬀ-shell IR divergences arise from the integration over large αl . To reveal these divergences, the integration region is divided into so-called ‘sectors’, where new integration variables are introduced, with the goal to obtain a factorization of the integrand. Then the analysis of convergence reduces to power counting in one-dimensional integrals. However, this mathematical analysis of convergence is restricted to the cases where the external momenta are Euclidean. Generalizations of these results connected with the analysis of convergence and dimensional regularization to Feynman integrals at a mass shell or at a threshold are not known. On the other hand, it turns out that, in these important cases, one can introduce some practical sector decompositions and corresponding sectors [5] that give the possibility to have control on the convergence and, in particular, provide a powerful method of evaluating Feynman integrals in situations with strong UV, IR and collinear divergences. The corresponding algorithm is described in Sect. E.2.

E.1 Analysis of Convergence We obtain the alpha representation of an analytically and dimensionally regularized Feynman integral corresponding to a graph Γ starting from the alpha representation (2.36) and substituting the powers of propagators al by al + λl with general complex numbers λl . For simplicity, let us assume the scalar case and that the powers of propagators are equal to one. (If al > 1, one can represent such a line by a sequence of al lines.) In this case the alpha representation takes a simpler form FΓ (q, m; d, λ) ∞ λ −d/2 2 l = dα αl U(α) exp iV(q, α)/U(α) − i ml αl , 0

l

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 221–232 (2004) c Springer-Verlag Berlin Heidelberg 2004

l

(E.1)

222

E Analysis of Convergence and Sector Decompositions

where the functions U and V are given by (2.24) and (2.25), and from now on we omit the coeﬃcient (−1)L eiπ( λl +h(1−d/2))/2 π hd/2 / Γ (λ + 1) , l

l

which is irrelevant to the analysis of convergence. In this appendix (as in Chap. 6), families of variables are denoted by underlined letters, i.e. q = (q1 , . . . , qn ), m = (m1 , . . . , mL ), λ = (λ1 , . . . , λL ), α = (α1 , . . . , αL ), etc., with dα = dα1 . . . dαL . Let us also assume here and later that the limit of integration refers to all of the integration variables involved. The alpha parameters have dimension −2 in mass units. By making the change of variables αl → µ−2 αl , where µ is a massive parameter, we can transform to dimensionless alpha parameters. For simplicity, let us take µ = 1 in this appendix. To separate the analysis of the UV and IR convergence as much as possible let us decompose the integration from 0 to ∞ over each alpha parameter into two regions: from 0 to 1 and from 1 to ∞. The integral (E.1) is then divided into 2L pieces, each of which is determined by a decomposition of the set of lines L of the given graph into two subsets, Lα and Lβ , corresponding to the integrations over the UV region (from 0 to 1) and the IR region (from 1 to ∞), respectively. For a given piece generated by a subset Lα , let us change the variables αl for l ∈ Lβ according to αl = 1/βl . The corresponding integral then takes the form 1 λ −λ −ε Lα dα dβ αl l βl l U(α, β)−d/2 FΓ (q, m; d, λ) = 0

l∈Lα

l∈Lβ

× exp iV(q, α, β)/U(α, β) − i

m2l αl − i

l∈Lα

m2l /βl .

(E.2)

l∈Lβ

For brevity, the new functions U and V are denoted by the same letters, although they are now of the form U(α, β) = βl U(α)|αl →1/βl ,l∈Lβ l∈Lβ

=

T ∈T 1

V(q, α, β) =

T ∈T 2

βl ,

(E.3)

l∈Lβ ∩T

βl V(q, α)α →1/β ,l∈L l

αl

l∈Lα \T

l∈Lβ

=

l∈Lα \T

αl

l

l∈Lβ ∩T

β

2 βl q T .

(E.4)

E.1 Analysis of Convergence

223

Remember that ±q T is the sum of the external momenta that ﬂow into one of the connectivity components of a 2-tree T . For a given piece FΓLα , let us change the numbering of the lines in such a way that the UV lines (i.e. those with αl ≤ 1) have smaller numbers. Thus we perform integration in the domain 0 ≤ αl ≤ 1, 1 ≤ l ≤ ¯l and 0 ≤ βl ≤ 1, ¯l + 1 ≤ l ≤ L, where ¯l = |Lα |. If S is a ﬁnite set, we denote by |S| the number of its elements. As we shall see, the analysis of UV and IR convergence is now decoupled. To analyse the UV convergence let us divide the domain of integration over αl into sectors. In the following, we shall use sectors of two types associated with nests and forests, respectively. The sectors connected with nests of subgraphs, (i.e. that γ ⊂ γ or γ ⊂ γ for any pair of the subgraphs of any nest; let us call them N -sectors) [14] are deﬁned by α1 ≤ . . . ≤ α¯l

(E.5)

and similar inequalities obtained by permutations. Without loss of generality, let us consider only the sector (E.5). Let us then change the integration variables according to αl = tl . . . t¯l .

(E.6)

The new (N -sector) variables tl are expressed in terms of αl by αl /αl+1 if l < ¯l . (E.7) tl = if l = ¯l α¯l The corresponding Jacobian equals tll−1 . The decomposition of the IR integration, over βl , is performed in a quite similar way. The following are the corresponding analogues of N -sectors and sector variables: βL ≥ . . . ≥ β¯l+1 , βl = τ¯l+1 . . . τl , βl /βl−1 if l > ¯l + 1 τl = , β¯l+1 if l = ¯l + 1

(E.8) (E.9) (E.10)

and the corresponding Jacobian is τlL−l . So, the initial integral is eventually divided into (L + 1)! sectors απ(1) ≤ . . . ≤ απ(¯l) ≤ 1 ≤ απ(¯l+1) ≤ απ(L) ,

(E.11)

which are labelled by permutations π of the numbers 1, . . . , L and the number ¯l. As we have stated, we consider only the contribution of the identical permutation, i.e. π(l) = l, l = 1, . . . , L. Although these sectors provide a resolution of the singularities of the integrand, they can turn out to be too rough for analysing convergence. A more sophisticated set of sectors corresponds to the maximal UV and IR forests. A set f of 1PI subgraphs and single lines with non-coincident end

224

E Analysis of Convergence and Sector Decompositions

points is called a UV forest [8, 22, 16] if the following conditions hold: (i) for any pair γ, γ ∈ f , we have either γ ⊂ γ , γ ⊂ γ or L(γ ∩ γ ) = ∅; (ii) if γ 1 , . . . , γ n ∈ f and L(γ i ∩ γ j ) = ∅ for any pair from this family, the subgraph ∪i γ i is one-vertex-reducible (i.e. can be made disconnected by deleting a vertex). Let F be a maximal UV forest (i.e. there are no UV forests that include F) of a given graph Γ . An element γ ∈ F is called trivial if it consists of a single line and is not a loop line. Any maximal UV forest has h(Γ ) nontrivial and L − h(Γ ) trivial elements. Let us deﬁne the mapping σ : F → L such that σ(γ) ∈ L(γ) and σ(γ) ∈ L(γ ) for any γ ⊂ γ, γ ∈ F. Its inverse σ −1 uniquely determines the minimal element σ −1 (l) of the UV forest F that contains the line l. Let us denote by γ+ the minimal element of F that strictly includes the given element γ. For a given maximal UV forest F, let us deﬁne the corresponding sector (F -sector) as (E.12) DF = α|αl ≤ ασ(γ) ≤ 1, l ∈ γ ∈ F . The intersection of two diﬀerent F -sectors has zero measure and the union of all the sectors gives the whole integration domain of the UV alpha parameters (i.e. αl ≤ 1) (see [8, 16, 18, 22]). For a given F -sector, let us introduce new variables labelled by the elements of F, tγ , (E.13) αl = γ∈F : l∈γ

L(γ)−1 where the corresponding Jacobian is γ tγ . The inverse formula is ασ(γ) /ασ(γ+ ) if γ is not maximal . (E.14) tγ = ασ(γ) if γ is maximal Consider, for example, the two-loop self-energy diagram of Fig. 3.9 and the following maximal UV forest F consisting of γ 1 = {1}, γ 2 = {2}, γ 3 = {3}, γ 4 = {1, 2, 5}, γ 5 = Γ . The mapping σ is σ(γ 1 ) = 1, σ(γ 2 ) = 2, σ(γ 3 ) = 3, σ(γ 4 ) = 5, σ(γ 5 ) = 4. The sector associated with this maximal UV forest is given by DF = {α1,2 ≤ α5 ≤ α4 , α3 ≤ α4 } and the sector variables are tγ 1 = α1 /α5 , tγ 2 = α2 /α5 , tγ 3 = α3 /α4 , tγ 4 = α5 /α4 , tγ 5 = α4 . The IR F -sectors and variables are introduced in a quite analogous way. New variables τγ are associated with maximal IR forests composed of IRirreducible subgraphs – see [18]. (A subgraph γ of Γ is called IR irreducible [10, 18] if the reduced graph Γ/γ is one-vertex-irreducible. (As in Chap. 2, Γ/γ is obtained from Γ by reducing every connectivity component of γ to a point.) The UV and IR maximal forests Fα and Fβ , composed of lines Lα and Lβ , respectively, are then combined in pairs to generate ‘generalized maximal forests’, with corresponding variables {tγ , τγ }, γ ∈ Fα , γ ∈ Fβ . As a result, the initial integration domain is divided into F -sectors associated with generalized maximal forests.

E.1 Analysis of Convergence

225

In each of the N - or F -sectors, the function (E.3) takes a factorized form in the new variables [8, 16, 18, 22, 24]: ¯ L l L−l+1−h(Γ/γ ) h(γ ) l−1 U = tl l [1 + PN (t, τ )] τl (E.15) =

l=1

l=¯ l+1

th(γ) γ

γ∈Fα

τγL(γ)−h(Γ/γ) [1 + PF (t, τ )] ,

(E.16)

γ∈Fβ

where PN and PF are non-negative polynomials, γl denotes the subgraph consisting of the lines {1, . . . , l}, and again γ = Γ \γ. The factorization of the function (E.4) in the N -sector variables is of the form ¯ L l −1 L−l+1−h(Γ/γl−1 ) h(γ ) l τ¯l+1 . . . τl0 V= tl τl l=¯ l+1

l=1

2 × q T0 + P0 (q, t, τ ) ,

(E.17)

where l0 denotes the number such that all the external vertices belong to the same connectivity component of the subgraphs γl for l ≥ l0 . In the Euclidean domain, where 2 qi −1 and in the sense of the limit δ → +0 with m2 → m2 − iδ (with identical resulting prescriptions in both these variants). In particular, such integrals are well deﬁned for the integer values λ = −1, −2, . . . Thus we have IR convergence when either the subgraph γl (or just γ) has at least one non-zero mass or its completion γl−1 (or γ) does not have all the external vertices in the same connectivity component. Therefore it is suﬃcient to check the IR convergence for the other IR-irreducible subgraphs. The domain of the regularization parameters λl and ε where these sector integrals are convergent is determined by the inequalities Re λ(γ) + h(γ) Re ε > [ω(γ)/2] , Re λ(γ) − h(Γ/γ) Re ε < [(ω(Γ ) − ω(γ) + 1)/2] ,

(E.23a) (E.23b)

which correspond, respectively, to UV-irreducible subgraphs and massless IRirreducible subgraphs whose completions γ contain all the external vertices in the same connectivity component. It turns out that this domain is non-empty for any graph without massless detachable subgraphs, i.e. massless subgraphs with zero external momenta. This statement can be proven [22] by observing that the parameters |T 1 | (0) λl = (2 − ε) 1 + δ − l1 − 1 , (E.24) |T | where Tl1 is the set of trees containing the line l, satisfy (E.23a) and (E.23b) for suﬃciently small δ > 0. (As before, | . . . | is the number of elements in the corresponding ﬁnite set.) Here again the scalar case is assumed. The generalization to a general diagram is straightforward: one adds nl /2 to the right-hand side of (E.24), where nl is the degree of the polynomial in the numerator of the lth propagator. In order to see that the Feynman integral can be continued from the above domain of mutual convergence to the whole hypercomplex plane of the variables (λ, ε) let us use the well-known property of the integrals ∞ F (λ) = dx xλ φ(x) . (E.25) 0

(In distributional language, this is the analytic property of the distribution xλ+ – see [12].) Indeed, the integral (E.25) with an inﬁnitely diﬀerentiable

228

E Analysis of Convergence and Sector Decompositions

function φ which has a compact support (or, a fast decrease at large values of x – see details in [12]) is absolutely convergent for all complex values of λ with Re λ > −1 so that it deﬁnes an analytic function of λ in this domain. This function can be continued analytically to the whole complex plane of λ with simple poles at λ = −1, −2, . . .. To perform the analytical continuation to the domain Re λ > −2 one decomposes the integral (E.25) into the two integrals, from 0 to 1 and from 1 to ∞, and uses an appropriate subtraction in the ﬁrst of them, i.e. represents φ(x) in (E.25) as (φ(x) − φ(0)) + φ(0) and takes the integral with the second term explicitly to obtain ∞ 1 φ(0) + dx xλ (φ(x) − φ(0)) + dx xλ φ(x) . (E.26) F (λ) = λ + 1 0 1 The ﬁrst integral on the right-hand side is now absolutely convergent at Re λ > −2 so that we obtain, from (E.26), an explicit analytic continuation of the function F (λ) to this domain. We also see that this function has a simple pole at λ = −1 with the residue φ(0).1 This procedure can naturally be generalized for the analytic continuation to the whole complex plane. To do this, one makes more subtractions2 : 1 n n (j) φ (0) j φ(j) (0) x + F (λ) = dx xλ φ(x) − j! j!(λ + j + 1) 0 j=0 j=0 ∞ + dx xλ φ(x) . (E.27) 1

Let us come back to our sector integrals. It follows from the factorizations (E.20), when they are written for all the sectors, that the Feynman integral can be continued from the above domain of mutual convergence to the whole hypercomplex plane of the variables (λ, ε) as a meromorphic function, with series of UV and IR poles. It is also clear that, in the case where there is no non-empty mutual-convergence domain, the contribution from any sector can be made convergent by choosing the absolute values of the real parts of the In distributional language, this means that the functional xλ+ has the pole at λ = −1 with the residue δ(x). By the way, in the domain −2 < Re λ < −1, the ∞ value φ(0)/(λ+1) can be rewritten as −φ(0) 1 dx xλ . After we combine it with the last integral in (E.26) we obtain the followingcompact expression for the analytic ∞ continuation of (E.25) to this band: F (λ) = 0 dx xλ (φ(x) − φ(0)). However, in our case of factorized expressions resulting from sector integrals, this is not relevant because we are dealing with ﬁnite regions of integration. 2 With the help of this procedure, the analytic continuation of (E.25) to the band −n − 1 < Re λ < −n − 1 takes the form [12]: 1

!

∞ λ

dx x

F (λ) = 0

φ(x) −

n φ(j) (0) j=0

j!

"

j

x

.

E.2 Practical Sector Decompositions

229

UV/IR analytic-regularization parameters to be suﬃciently large (positive and negative for l ≤ ¯l and l > ¯l, respectively). The analytic regularization can then be switched oﬀ, by analytic continuation, and one obtains [9] a dimensionally regularized Feynman integral as the sum of its sector contributions, which were deﬁned in their own initial analyticity domains using the auxiliary analytic regularization. Therefore, we obtain a deﬁnition of dimensional regularization for any Feynman integral at Euclidean external momenta.

E.2 Practical Sector Decompositions The sector decompositions of the previous section are simpler than the sectors of [22]. However, if we want to apply sectors for the numerical evaluation of Feynman integrals the initial decomposition of the integration domain over every alpha parameter in the two regions is not optimal at all because we obtain 2L pieces from the beginning. So, the natural idea is to apply the sectors of [22]. Presumably, this procedure can be implemented on a computer, but no such examples are known. The bad news is that, although the sector decompositions discussed above can successfully be used for proving theorems on renormalization [14, 21, 24] and on asymptotic expansions in limits of momenta and masses typical of Euclidean space (see [17, 18] and Appendix B of [19]), they are not suﬃcient for resolving the singularities of the integrand in the case of Feynman integrals on a mass shell or at a threshold. Let us consider again Example 3.3 of Sect. 3.3, with the basic functions U and V given by (3.23), and try to apply the N -sectors to resolve the singularities of the alpha integral in the region of large αl . To do this, let us turn to the variables βl = 1/αl , as in the previous section, where we obtain the functions U(β) = β1 β2 β3 + β1 β2 β4 + β1 β3 β4 + β2 β3 β4 , V(β) = tβ2 β4 + sβ1 β3 .

(E.28) (E.29)

Consider now the N -sector β2 ≤ β1 ≤ β3 ≤ β4 and introduce the variables (E.10), i.e. by means of the relations β2 = τ1 τ2 τ3 τ4 ,

β1 = τ2 τ3 τ4 ,

β3 = τ3 τ4 , β4 = τ4 .

(E.30)

In these sector variables, the function (E.28) factorizes, in a suitable way, according to (E.15), but the function (E.29) does not: V(τ ) = τ2 τ3 τ42 (sτ1 + tτ3 ) .

(E.31)

Such a phenomenon would never happen for Feynman integrals considered at Euclidean external momenta – see the general result (E.17). So, we do not have a nice factorization property similar to (E.17) for the contribution of the sector under consideration. In order to perform the analysis of convergence, the factor sτ1 + tτ3 raised to some power dependent

230

E Analysis of Convergence and Sector Decompositions

on ε has to be further factorized. The natural idea here is to perform a next sector decomposition, using N -sectors, then proceed further if we do not immediately succeed, etc. However, this procedure looks awful from the practical point of view: to have L! contributions at the ﬁrst step, then (L!)2 at the second step is a very bad idea if we think of a computer implementation. Still the idea to introduce, recursively, more and more sectors has turned out to be quite successful and easily implemented in practice. A suitable algorithm based on sector decompositions for resolving singularities of general Feynman integrals, in particular, considered on a mass-shell or at a threshold, possibly, with severe UV, IR and collinear divergences, was developed in [5]. On the one hand, this algorithm makes the analysis of the singularities in ε possible for any given Feynman integral. On the other hand, it gives a powerful universal numerical method for evaluating Feynman integrals. The starting point of the algorithm of [5] is representation (3.32), where the sum of all the parameters αl is implied in the δ-function. It is supposed that all the kinematical invariants and the masses have the same sign, i.e. if there is a non-zero mass, all the invariants are non-positive. Then one introduces the following primary sectors ∆l labelled by the number l = 1, . . . , L: αi ≤ αl ,

l = i = 1, 2, . . . , L

(E.32)

and turns, in a given sector ∆l , to the variables αi /αl if i = l ti = . if i = l αl

(E.33)

Then the integration over tl is taken due to the δ-function, and one obtains the integral 1 L−(h+1)d/2 U . (E.34) Fl = dti L−hd/2 V tl =1 i =l 0

Here we used the fact that the functions U and V are homogeneous functions of the alpha parameters with the homogeneity degrees h and h + 1, respectively. The goal of the introduction of the sector decompositions is to obtain a perfect factorization, i.e. of the form (E.15) for U and of the form (E.17) for 2 −V + U m2l αl , where, instead of q T0 , there is some positive combination of the kinematical invariants and masses. So, if the perfect factorization is not achieved, for the contribution of the given sector ∆l , the next natural step is to introduce a second decomposition in a similar way, i.e. over L − 1 sectors ∆lj , ti ≤ tj , i = 1, 2, . . . , L ,

i = j, l ,

j = l .

(E.35)

and new variables ti similarly to (E.33). One may hope that sooner or later a perfect factorization will be achieved. If this is the case, one obtains a sum of parametric integrals, over some sector variables ti , where the singularities are factorized, i.e. the integrand is a product of ti raised to some powers

E.2 Practical Sector Decompositions

231

λi = ni + hi ε, with integer ni and hi = 0, and the two functions (also raised to similar powers) which result from U and V and are positive in the integration region. In such a ‘perfect’ situation, the analysis of convergence reduces to counting powers of the variables ti . This reminds again, as in the end of the previous section, the analysis of the distribution xλ+ – see [12]. Explicitly, we have integrations over sector variables (of some level of iterations) of the form 1 dt tn+hε φ(t) , (E.36) G(ε) = 0

where t is one of the sector variables, n and h = 0 are integer numbers and φ(t) is a function with φ(0) = 0 which involves similar factorized integrations over the rest of the sector variables. If n ≥ 0, the integration over t does not generate poles in ε. Suppose that n is negative. The procedure outlined in the end of the previous section suggests a similar subtraction: 1 −n−1 φ(j) (0) tj dt tn+hε φ(t) − G(ε) = j! 0 j=0 +

−n−1 j=0

φ(j) (0) . j!(n + hε + j + 1)

(E.37)

After performing such manipulations with integrations over all the sector variables ti with ni < 0 one obtains a linear combination of integrals where one can perform an expansion in a Laurent series in ε. This provides the possibility to formulate an algorithm for the numerical evaluation of any term of expansion of the given Feynman integral in ε. Numerous practical calculations have shown [5] that this algorithm works for complicated Feynman integrals with multiple IR and collinear divergences. For example, analytical results for double and triple boxes [3, 20, 23] were numerically conﬁrmed by means of this algorithm. Once again, this is a method with experimental mathematics. It is not guaranteed, as in a mathematical theorem, that the process of the recursive introduction of the sector decompositions described above will stop at some point with a perfect factorization. Moreover, practical calculations have shown that one has to avoid possible closed loops in the algorithm. However, this is the only working general algorithm at the moment, applicable at any loop order, with applications restricted only by the computer time. One may hope that the algorithm can be generalized to the cases without restrictions on the signs of the kinematical invariants and the masses. Observe, however, that another important generalization, to the case of phase-space integrals, was already developed and successfully applied in practice in [1, 2, 6, 11, 13].

232

E Analysis of Convergence and Sector Decompositions

References 1. C. Anastasiou, K. Melnikov and F. Petriello, Phys. Rev. D 69 (2004) 076010; Phys. Rev. Lett. 93 (2004) 032002. 231 2. C. Anastasiou, K. Melnikov and F. Petriello, hep-ph/0409088. 231 3. C. Anastasiou, J.B. Tausk and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 262. 231 4. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 5. T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 221, 230, 231 6. T. Binoth and G. Heinrich, Nucl. Phys. B 693 (2004) 134. 231 7. N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 8. P. Breitenlohner and D. Maison, Commun. Math. Phys. 52 (1977) 11, 39, 55. 224, 225 9. K.G. Chetyrkin and V.A. Smirnov, Teor. Mat. Fiz. 56 (1983) 206. 229 10. K.G. Chetyrkin and V.A. Smirnov, Phys. Lett. B 144 (1984) 419. 224 11. A. Gehrmann-De Ridder, T. Gehrmann and G. Heinrich, Nucl. Phys. B 682, 265 (2004). 231 12. I.M. Gel’fand and G.E. Shilov, Generalized Functions, Vol. 1 (Academic Press, New York, London, 1964). 227, 228, 231 13. G. Heinrich, Nucl. Phys. Proc. Suppl. 116, 368 (2003). 231 14. K. Hepp, Commun. Math. Phys. 2 (1966) 301. 223, 229 15. N. Nakanishi, Graph Theory and Feynman Integrals (Gordon and Breach, New York, 1971). 16. K. Pohlmeyer, J. Math. Phys. 23 (1982) 2511. 224, 225 17. V.A. Smirnov, Commun. Math. Phys. 134 (1990) 109. 229 18. V.A. Smirnov, Renormalization and Asymptotic Expansions (Birkh¨ auser, Basel, 1991). 224, 225, 229 19. V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 229 20. V.A. Smirnov, Phys. Lett. B 491 (2000) 130; B 500 (2001) 330; B 524 (2002) 129; B 567 (2003) 193; hep-ph/0406052; G. Heinrich and V.A. Smirnov, hepph/0406053. 231 21. E.R. Speer, J. Math. Phys. 9 (1968) 1404. 229 22. E.R. Speer, Ann. Inst. H. Poincar´e 23 (1977) 1. 224, 225, 227, 229 23. J.B. Tausk, Phys. Lett. B 469 (1999) 225. 231 24. O.I. Zavialov, Renormalized Quantum Field Theory (Kluwer Academic, Dodrecht, 1990). 225, 229

F A Brief Review of Some Other Methods

In this appendix, some methods which were not considered in Chaps. 3–7 are brieﬂy reviewed. The method based on dispersion relations was successfully used from the early days of quantum ﬁeld theory. The Gegenbauer Polynomial x-Space Technique [13], the method of gluing [15] and the method based on star-triangle uniqueness relations [16, 23, 36] are methods for evaluating massless diagrams. The method of IR rearrangement [38], also in a generalized version based on the R∗ -operation [14, 34], is a method oriented at renormalization-group calculations. The recently developed method of diﬀerence equations [27] is also brieﬂy described. It is not analytical, although based on non-trivial mathematical analysis. It enables us to obtain numerical results with extremely high precision, with hundreds of digits. Finally, some methods which could be characterized as based on experimental mathematics are discussed. In particular, this is the integer relation algorithm called PSLQ [18] which provides the possibility to obtain a result for a given one-scale Feynman integral, when we strongly suspect that it is a linear combination of some transcendental numbers with rational coeﬃcients, provided we know the result numerically with a high accuracy.

F.1 Dispersion Integrals A given propagator scalar Feynman integral can be written as ∞ 1 ∆F (s) , ds F (q 2 ) = 2πi s0 s − q 2 − i0

(F.1)

where the discontinuity ∆F (s) = 2i Im(F (s + i0)) is given, according to Cutkosky rules, by a sum over cuts in a given channel of integrals, where the propagators i/(k 2 − m2 + i0) in the cut are replaced by 2πi θ(k0 )δ(k 2 − m2 ), while the propagators to the left of the cut stay the same, and the propagators to the right of the cut change the causal prescription and become −i/(k 2 − m2 − i0). Let us again consider our favourite example of Fig. 1.1, with the indices equal to one. This time, let us include al the necessary factors of i from each V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 233–244 (2004) c Springer-Verlag Berlin Heidelberg 2004

234

F A Brief Review of Some Other Methods

propagator and the factor −i corresponding to the deﬁnition of the Feynman integral with i on the right-hand side of (2.3). We have 2 2 dd k θ(k0 )δ(k 2 − m2 )θ(q0 − k0 )δ[(q − k)2 ] ∆F (q ) = 4π ! " 2 q0 q02 − m2 2π 2 d−2 2 Ωd−1 dr r δ −r = q0 2q0 0 =

24−d π (d+3)/2 (q 2 − m2 )d−3 + , Γ ((d − 1)/2) (q 2 )(d−2)/2

(F.2)

where X+ = X for X > 0 and X+ = 0 otherwise, as usual. We have chosen q = (q0 , 0) and introduced (d − 1)-dimensional spherical coordinates with the surface of the unit sphere in d dimensions equal to 2π d/2 . Γ (d/2) For d = 4, this gives Ωd =

∆F (s) =

2π 3 (q 2 − m2 )+ . q2

(F.3)

(F.4)

Integrating from the threshold s0 = m2 in the dispersion integral (F.1) (where a subtraction is needed) leads to the ﬁnite part of (1.7) (where the factors of i mentioned above were dropped) up to a renormalization constant. In this calculation, a phase-space integral corresponding to a two-particle cut with the masses m and 0 was evaluated. The evaluation of three- and four-particle phase-space integrals is much more complicated. Although we have less integrations in integrals corresponding to cuts, because of the δfunctions, resulting integrals are still rather nasty so that the evaluation of Feynman integrals via their imaginary part by means of Cutkosky rules (see [29] for a typical example) was successful only up to some complexity level. On the other hand, the phase-space integrals are needed for the calculation of the real radiation. It has turned out that the development of methods of evaluating Feynman integrals resulted in similar techniques for the phase-space integrals. Now, one applies, for the evaluation of the phase-space integrals, the strategy of the reduction to master integrals, using IBP, and DE applied for the evaluation of the master integrals – see, e.g., [1, 2]. Moreover, the technique of the sector decompositions of [7] (see Sect. E.2) is also applicable here and was successfully applied in NNLO calculations – see references in the end of Appendix E.

F.2 Gegenbauer Polynomial x-Space Technique The Gegenbauer polynomial x-space technique (GPXT) [13] is based on the SO(d) symmetry of Euclidean Feynman integrals. According to (A.40), the dimensionally regularized scalar massless propagator in coordinate space is

F.3 Gluing

DF (x1 − x2 ) =

1 (2π)d

dd q

e−ix·q Γ (1 − ε) = , q2 4π d/2 [(x1 − x2 )2 ]1−ε

235

(F.5)

where x2 = x20 + x2 . It can be expanded in Gegenbauer polynomials [17] as 1 1 = 2λ [(x1 − x2 )2 ]λ (max{|x1 |, |x2 |}) n/2 ∞ min{|x1 |, |x2 |} × Cnλ (ˆ x1 · x ˆ2 ) , (F.6) max{|x1 |, |x2 |} n=0 √ where |x| = x2 , λ = 1 − ε and x ˆ = x/|x|. The polynomials Cnλ are orthogonal on the unit sphere [17]: λ λ δn,m Cnλ (ˆ x1 · x ˆ2 ) Cm (ˆ x2 · x ˆ3 ) = x1 · x ˆ3 ) . (F.7) dˆ x2 Cnλ (ˆ n+λ The normalization is such that dˆ x = 1. So, the strategy of GPXT is to turn to coordinate space, represent each propagator by (F.6), evaluate integrals over angles by (F.7) and sum up resulting multiple series. First results for non-trivial multiloop diagrams within dimensional regularization were obtained by GPXT: for example, the value of the non-planar diagram (see the second diagram of Fig. 5.6 with all the powers of the propagators equal to one), with the famous result proportional to 20ζ(5) [13]. The GPXT as well as the method of gluing (see below) were crucial in many important analytical calculations, for example, of the three-loop ratio R(s) in QCD [12] and the ﬁve-loop β-function in the φ4 theory [11]. More details on the GPXT can be found in the review [25].

F.3 Gluing The dependence of an h-loop dimensionally regularized scalar propagator massless Feynman integral corresponding to a graph Γ on the external momentum can easily be found by power counting: h (F.8) FΓ (q; d) = iπ d/2 CΓ (ε)(q 2 )ω/2−hε , where ω is the degree of divergence given by (2.9) and CΓ (ε) is a meromorphic function which is ﬁnite at ε = 0 if the integral is convergent, both in the UV and IR sense. (Of course, there are no collinear divergences in propagator integrals.) It turns out that the values CΓ (0) are the same for graphs connected by some transformations based on gluing. The gluing can be of two types: by vertices and by lines. Let Γ be a graph with two external vertices. Let us denote by Γˆ the graph obtained from it by identifying these vertices, and by Γ¯ the graph obtained from it by adding a new line which connects them. Then the following properties hold [15]:

236

F A Brief Review of Some Other Methods

– Gluing by vertices. Let us suppose that two UV- and IR-convergent graphs, Γ1 and Γ2 , have degrees of divergence ω1 = ω2 = −4 and that Γˆ1 and Γˆ2 are the same. Then CΓ1 (0) = CΓ2 (0). – Gluing by lines. Let us suppose that two UV- and IR-convergent graphs, Γ1 and Γ2 , have degrees of divergence ω1 = ω2 = −2 and that Γ¯1 and Γ¯2 are the same. Then CΓ1 (0) = CΓ2 (0). For example, the ﬁrst and the second diagrams in Fig. 5.6 with all the indices equal to one produce the same graph after the gluing the external vertices. It is shown in Fig. F.1. Therefore, one could obtain the value of the more complicated non-planar diagram (proportional to 20ζ(5)) from a simpler planar diagram [15].

Fig. F.1. The graph Γˆ obtained by gluing of vertices

The method of gluing was successfully applied in the combination with GPXT – see the references above.

F.4 Star-Triangle Relations The method based on star-triangle uniqueness relations can be applied to massless diagrams. As in the case of GPXT, the coordinate space language is used, where the propagators have the form 1/(x2 )λ up to a coeﬃcient depending on ε – see, e.g., (F.5). The basic uniqueness relation [16, 36] connects diagrams with diﬀerent numbers of loops. It is graphically shown in Fig. F.2, where λi = d/2 − λi and Γ (d/2 − λi ) v(λ1 , λ2 , λ3 ) = π d/2 . (F.9) Γ (λi ) i This equation holds when the vertex on the left-hand side is unique, i.e. λ1 +λ2 +λ3 = d. The triangle on the right-hand side, with λ1 +λ2 +λ3 = d/2, is also called unique. Remember that, in coordinate space, the triangle diagram does not involve integration and is just a product of the three propagators, [(x1 − x2 )2 ]−λ3 [(x2 − x3 )2 ]−λ1 [(x3 − x1 )2 ]−λ2 , while the star diagram is an integral over the coordinate corresponding to the central vertex.

F.5 IR Rearrangement and R∗

237

λ1 λ2

λ3

=

v(λ1 , λ2 , λ3 ) ×

λ3

λ2 λ1

Fig. F.2. Uniqueness equation

The relation (F.9) can be used to simplify a given diagram. Almost unique relations introduced in [35], with λ1 +λ2 +λ3 = d−1, can be also useful. Sometimes one introduces an auxiliary analytic regularization, to satisfy (almost) unique relations, which can be switched oﬀ in the end of the calculation. For example, using (almost) unique relations, the general ladder massless scalar propagator diagram with an arbitrary number of loops, h, with all the indices ai equal to one (see the ﬁrst diagram of Fig. 5.6 and imagine a general number of rungs), was evaluated [5] with a result proportional to ζ(2h − 1). Another example of applications of the uniqueness relations is the evaluation of the diagram of Fig. 4.14 where they were coupled with functional equations [23]. In this calculation, the initial problem was reduced to the problem of expansion of the propagator diagram of Fig. 3.9 with the indices a1 = . . . = a4 = 1, , a5 = 1 + λ in a Taylor series in λ up to λ4 . This diagram, at various indices, was investigated in many papers starting from the old result for all indices equal to one [33] which was later reproduced [13] by GPXT, an analytical result for this diagram with general values of the indices a1 and a2 and other integer indices [13], an analysis of this diagram from the group-theoretical point of view [9], an extension of the previous results with the help of GPXT [24], etc. As a more recent paper, with updated references to the previous works, let us cite [6], where the expansion of this diagram at indices ai = ni + hi ε, with integer hi , in ε was further studied.

F.5 IR Rearrangement and R∗ The method of IR rearrangement is a special method for the evaluation of UV counterterms which are necessary to perform renormalization. The counterterms are introduced into the Lagrangian, i.e. the dependence of the bare parameters (coupling constants, masses, etc.) of a given theory on a regularization parameter (e.g., d within dimensional regularization) is adjusted in such a way that the renormalized physical quantities become ﬁnite when the regularization is removed. The renormalization can be described at the diagrammatic level, i.e. the renormalized Feynman integrals can be obtained by applying the so-called R-operation which removes the UV divergence from

238

F A Brief Review of Some Other Methods

individual Feynman integrals. Thus, for any R-operation, the quantity RFΓ is UV ﬁnite at d = 4. As is well known, the requirement for the R-operation to be implemented by inserting counterterms into the Lagrangian leads to the following structure [8]: ∆(γ1 ) . . . ∆(γj )FΓ ≡ R FΓ + ∆(Γ ) FΓ , (F.10) RFΓ = γ1 ,...,γj

where ∆(γ) is the corresponding counterterm operation, and the sum is over all sets {γ1 , . . . , γj } of disjoint UV-divergent 1PI subgraphs, with ∆(∅) = 1. The ‘incomplete’ R-operation R , by deﬁnition, includes all the counterterms except the overall counterterm ∆(Γ ). For example, if a graph is primitively divergent, i.e. does not have divergent subgraphs, the R-operation is of the form RFΓ = [1 + ∆(Γ )] FΓ . The action of the counterterm operations is described by ∆(γ) FΓ = FΓ/γ ◦ Pγ ,

(F.11)

where FΓ/γ is the Feynman integral corresponding to the reduced graph Γ/γ, and the right-hand side of (F.11) denotes the Feynman integral that diﬀers from FΓ/γ by insertion of the polynomial Pγ in the external momenta and internal masses of γ into the vertex vγ to which the subgraph γ was reduced. The degree of each Pγ equals the degree of divergence ω(γ). It is implied that a UV regularization is present in (F.10) and (F.11) because these quantities are UV-divergent. The coeﬃcients of the polynomial Pγ are connected in a straightforward manner with the counterterms of the Lagrangian. A speciﬁc choice of the counterterm operations for the set of the graphs of a given theory deﬁnes a renormalization scheme. In the framework of dimensional renormalization, i.e. renormalization schemes based on dimensional regularization, the polynomials Pγ have coeﬃcients that are linear combinations of pure poles in ε = (4 − d)/2. In the minimal subtraction (MS) scheme [21], these polynomials are deﬁned recursively by equations of the form ˆ ε R Fγ Pγ ≡ ∆(γ) Fγ = −K

(F.12)

ˆ ε is the operator that picks up for the graphs γ of the given theory. Here K the pole part of the Laurent series in ε. The modiﬁed MS scheme [4] (MS) is obtained from the MS scheme by the replacement µ2 → µ2 eγE /(4π) for the massive parameter of dimensional regularization that enters through the factors of µ2ε per loop. If Γ is a logarithmically divergent diagram the corresponding counterterm is just a constant. To simplify its calculation it is tempting to put to zero the masses and external momenta. This is, however, a dangerous procedure because it can generate IR divergences. Consider, for example, the two-loop graph of Fig. F.3a. It contributes to the mass renormalization in the φ4 theory. To evaluate the corresponding counterterm it is necessary to compute R Fγ ,

F.5 IR Rearrangement and R∗

q

q

q

239

q

q (a)

(b)

(c)

Fig. F.3. (a) A two-loop graph contributing to the mass renormalization. (b) A possible IR rearrangement. (c) A three-loop graph contributing to the β-function

according to (F.12). Here R = 1+∆1 , where ∆1 is the counterterm operation for the logarithmically divergent subgraph of Fig. F.3a. We consider each of the two resulting terms separately. The last term is simple. The ﬁrst one is just the pole part of the given diagram. If we put the mass to zero we shall obtain an IR divergence. There is another option which is safe: we put the mass to zero and let the external momentum q ﬂow in another way through the graph: from the bottom vertex, rather than from the right vertex – see Fig. F.3b. Then the resulting Feynman integral is IR-convergent and, at the same time, much simpler because it is now recursively one-loop and can be evaluated in terms of gamma functions. This is a simple example of the trick called IR rearrangement and invented in [38]. In a general situation, one tries to put as many masses and external momenta to zero as possible and, probably, let the external momentum ﬂow through the graph in such a way that the resulting diagram is IR-convergent and simple for calculation. Consider now the three-loop graph of Fig. F.3c contributing to the β-function in the φ4 theory. It is also logarithmically divergent. When calculating its counterterm, it is dangerous to put the masses to zero and let the external momentum ﬂow from the bottom to the top vertex, because we run into IR divergences either due to the left or the right pair of the lines. Still there is a possibility not to generate IR divergences: to put the masses of the central loop and the external momentum to zero. The resulting three-loop Feynman integral is evaluated in terms of gamma functions, ﬁrst, by integrating the massless subintegral by (A.7) and then by (A.38). At a suﬃciently high level, such a safe IR rearrangement is not always possible. However, there is a way to put as many masses and momenta to zero and still have control on IR divergences. Formally, we have ˆ ε R∗ Fγ (q) , Pγ = −K

(F.13)

where it is implied that all the masses are put to zero, and one external momentum is chosen to ﬂow through the diagram in an appropriate way. (Another version is to put all the external momenta to zero and leave one non-zero mass.)

240

F A Brief Review of Some Other Methods

The operation R∗ removes not only UV but also (oﬀ-shell) IR divergences in a similar way [14], i.e. by a formula which generalizes (F.10). Now, it ˜ includes IR counterterms ∆(γ) which are deﬁned in a full analogy to the UV counterterms ∆(γ). They are deﬁned for subgraphs irreducible in the IR sense, with the IR degree of divergence given by (2.17). Now, they are local in momentum space. For example, the IR counterterm corresponding to the logarithmically divergent (in the IR sense, i.e. with the IR degree of divergence ω ˜ (γ) = 0) factor 1/(k 2 )2 for the two lower lines in Fig. F.3a (when they are massless) is proportional to δ (d) (k)/ε. More details on the R∗ -operation can be found in [34]. So, according to (F.13), one can safely put to zero all the momenta and masses but one, in a way which is the simplest for the calculation, at the cost of generating IR divergences which should be removed with the help of IR counterterms. Finally, the problem of the evaluation of the UV counterterms for graphs with positive degrees of divergence can be reduced, by diﬀerentiating in momenta and masses, to the case ω = 0. The R∗ -operation was successfully applied in renormalization group calculations – see, e.g., [11].

F.6 Diﬀerence Equations A new method based on diﬀerence equations has recently appeared. Basic prescriptions of this method can be found in [27] and an informal introduction in [28]. It is analytical in nature but is used to obtain numerical results with extremely high precision. The starting point of this approach is to choose a propagator, in an arbitrary way, treat its power, n, as the basic integer variable and ﬁx other powers of the propagators (typically, equal to one). Then the general Feynman integral (5.73) of a given family is written as H , (F.14) F (n) = · · · dd k1 . . . dd kh n E1 E2 . . . EN where H is a numerator. After combining various IBP relations, one can obtain a diﬀerence equation for F (n): c0 (n)F (n) + c1 (n)F (n + 1) + . . . + cr (n)F (n + r) = G(n) ,

(F.15)

where the right-hand side contains Feynman integrals F1 , F2 , . . . which have one or more denominators E2 , E3 , . . . less with respect to (F.14). These integrals are treated in a similar way, by means of equations of the type (F.15) so that one obtains a triangular system of diﬀerence equations. This system is solved, starting from the simplest integrals that have the minimum number of denominators, with the help of an Ansatz in the form of a factorial series, µn

∞ l=0

bl n! , Γ (n − K + l + 1)

(F.16)

F.7 Experimental Mathematics and PSLQ

241

where the values of parameters µ, bl and K are obtained from these values for the factorial series corresponding to the right-hand side of (F.15). This method was successfully applied, with a precision of several dozens up to hundreds of digits, to the calculation of various multiloop Feynman integrals [26, 27]. Observe that, although this method is numerical, it requires serious mathematical eﬀorts. The same feature holds for any modern method of numerical evaluation. One can say that the boarder between analytical and numerical methods becomes rather vague at the moment. Remember about new results obtained in terms of new functions discussed in the end of Chap. 7 – in a narrow sense, these new functions can be regarded as tools to obtain numerical results at various points. Another numerical method based on non-trivial mathematical analysis was described in Sect. E.2. For completeness, here are some references to modern methods of numerical evaluation of Feynman integrals: [30, 31, 32]. Observe that such methods are often called semianalytical. Sometimes it is claimed that sooner or later we shall achieve the limit in the process of analytical evaluation of Feynman integrals so that we shall be forced to proceed only numerically (see, e.g., [30]). However, the dramatic progress in the ﬁeld of analytical evaluation of Feynman integrals shows that we have not yet exhausted our abilities. So, the natural strategy is to combine available analytical and numerical methods in an appropriate way.

F.7 Experimental Mathematics and PSLQ When evaluating Feynman integrals, various tricks are used. One usually does not bother about mathematical proofs of the tricks, partially, because of the pragmatical orientation and strong competition and, partially, because, now, there are a lot of possibilities to check obtained results, both in the physical and mathematical way. An example of such ‘experimental mathematics’ suggested in [20] was described in Sect. 4.5, where it was supposed that the nth coeﬃcient of the Taylor series cn of a piece of the result for the master massive double box is a linear combination of the 15 functions (4.62)–(4.65) of the variable n. Then the possibility to evaluate the ﬁrst 15 coeﬃcients c1 , c2 , . . . , c15 was used and the corresponding linear system for unknown coeﬃcients in the given linear combination was solved. At this point, a pure mathematician could say that there is no mathematical proof of this procedure and its validity is not guaranteed at all even after we (successfully) check it by calculating more terms of the Taylor expansion, starting from the 16th and comparing it with what we have from the obtained solution. Still I believe that this pure mathematician will believe in the result when he/she looks at some details of the calculation. Indeed, suppose that we forget about just one of the functions in (4.62)–(4.65) and follow our procedure. Then we indeed obtain a diﬀerent solution of our system of 14 equations but it blows up and

242

F A Brief Review of Some Other Methods

looks so ugly, in terms of rational numbers with hundreds of digits in the numerator and denominator, that this pure mathematician will say that our previous solution, with nice rational numbers, is true and there is no need for mathematical proofs. Of course, an important point here is to understand what we can expect in the result. Another example is given by taking a sum when going from (4.94) to (4.95) when evaluating the diagram of Fig. 4.14. Instead of using SUMMER [39], we can suppose that the general term of the Taylor series (4.95) is a linear combination, with unknown coeﬃcients, of (4.62)–(4.65) and similar terms up to level 7. (For example, at level 7, one can use the structures with a 1/n2 dependence present on the left-hand side of (C.51)–(C.82).) Then one obtains a system of 63 linear equations for these coeﬃcients and solves it using information about the ﬁrst 63 terms which can be obtained from the two-fold series following from (4.94). There are a lot of other elements of experimental mathematics in dealing with Feynman integrals. Indeed, we never hesitate to change the order of integration over alpha and Feynman parameters and over MB parameters, it is not known in advance which IBP equations within the algorithm formulated in [27] are really independent, there is no mathematical justiﬁcation of the prescriptions of Chap. 6, etc. One more example of experimental mathematics1 is provided by the so-called PSLQ algorithm [18]. It can be applied when we evaluate a one-scale Feynman integral in expansion in ε. Let us suppose that, in a given order of expansion in ε, we understand which transcendental numbers can appear in the result and that we can obtain the result numerically with a high accuracy. For example, in the ﬁnite part of the ε-expansion in two loops we can expect at least xi−1 = ζ(i) with i = 2, 3, 4 or, equivalently, x1 = π 2 , x2 = ζ(3) and x3 = π 4 . Then the PSLQ algorithm could be of use. In this particular example, it gives the possibility to estimate whether or not a given number, x can be expressed linearly as x = c1 x1 + c2 x2 + c3 x3 with rational coeﬃcients ci . The PSLQ is an example of an ‘integer relation algorithm’. If x1 , x2 , · · · , xn are some real numbers, it gives the possibility to ﬁnd the n integers ci such that c1 x1 +c2 x2 +· · ·+cn xn = 0 or provide bounds within which this relation is impossible. (In the above situation, we consider our numerical result as x4 , in addition to the xi , i = 1, 2, 3.) More formally, suppose that xi are given with the precision of ν decimal digits. Then we have an integer relation with the norm bound N if |c1 x1 + . . . + cn xn | < ε ,

(F.17)

provided that max|ci | < N , where ε > 0 is a small number of order 10−ν . With a given accuracy ν, a detection threshold ε and a norm bound N as an input,

1

The very term ‘experimental mathematics’ can be found on the web page where, in particular, the PSLQ algorithm is described [39].

References

243

the PSLQ algorithm enables us to ﬁnd out whether the relation (F.17) exists or not at some conﬁdence level (see details in [18]). The PSLQ algorithm has been successfully applied in the evaluation of various single-scale Feynman integrals – see, e.g., [3, 10, 19, 22]. The experience obtained in these calculations shows that one needs around ten digits for each independent transcendental number.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29.

C. Anastasiou and K. Melnikov, Nucl. Phys. B 646 (2002) 220. 234 C. Anastasiou and K. Melnikov, Phys. Rev. D 67, 037501 (2003) 234 D.H. Bailey and D.J. Broadhurst, Math. Comput. 70 (2001) 1719. 243 W.A. Bardeen, A.J. Buras, D.W. Duke and T. Muta, Phys. Rev. D 18 (1978) 3998. 238 V.V. Belokurov and N.I. Ussyukina, J. Phys. A 16 (1983) 2811. 237 I. Bierenbaum and S. Weinzierl, Eur. Phys. J. C 32 (2003) 67 237 T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 234 N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 238 D.J. Broadhurst, Z. Phys. C 32 (1986) 249; D.T. Barfoot and D.J. Broadhurst, Z. Phys. C 41 (1988) 81. 237 D.J. Broadhurst, Eur. Phys. J. C 8 (1999) 311 243 K.G. Chetyrkin, S.G. Gorishnii, S.A. Larin and F.V. Tkachov, Phys. Lett. B 132, 351 (1983). 235, 240 K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Phys. Lett. B 85 (1979) 277. 235 K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Nucl. Phys. B 174 (1980) 345. 233, 234, 235, 237 K.G. Chetyrkin and V.A. Smirnov, Phys. Lett. B 144 (1984) 419. 233, 240 K.G. Chetyrkin and F.V. Tkachov, Phys. Lett. B 192 (1981) 159. 233, 235, 236 M. D’Eramo, L. Peliti and G. Parisi, Lett. Nuovo Cim. 2 (1971) 878. 233, 236 A. Erd´elyi (ed.), Higher Transcendental Functions, Vols. 1 and 2 (McGraw-Hill, New York, 1954). 235 H.R.P. Ferguson and D.H. Bailey, RNR Technical Report, RNR-91-032; H.R.P. Ferguson, D.H. Bailey and S. Arno, NASA Technical Report, NAS96-005. 233, 242, 243 J. Fleischer and M. Y. Kalmykov, Phys. Lett. B 470 (1999) 168; Comput. Phys. Commun. 128 (2000) 531. 243 J. Fleischer, A.V. Kotikov and O.L. Veretin, Nucl. Phys. B 547 (1999) 343. 241 G. ’t Hooft, Nucl. Phys. B 61 (1973) 455. 238 M.Yu. Kalmykov and O. Veretin, Phys. Lett. B 483 (2000) 315. 243 D.I. Kazakov, Theor. Math. Phys. 58 (1984) 223 [Teor. Mat. Fiz. 58 (1984) 343]; 62, 84 (1985) [Teor. Mat. Fiz. 62, 127 (1984)]. 233, 237 A.V. Kotikov, Phys. Lett. B 375 (1996) 240. 237 A.V. Kotikov, hep-ph/0102177. 235 S. Laporta, Phys. Lett. B 504, 351 (1983); B 523 (2001) 95; B 549 (2002) 115 241 S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 233, 240, 241, 242 S. Laporta, Acta Phys. Polon. B 34 (2003) 5323. 240 W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 234

244

F A Brief Review of Some Other Methods

30. G. Passarino, Nucl. Phys. B 619 (2001) 257. 241 31. G. Passarino and S. Uccirati, Nucl. Phys. B 629 (2002) 97; A. Ferroglia, G. Passarino, S. Uccirati and M. Passera, Nucl. Instrum. Meth. A 502 (2003) 391; A. Ferroglia, M. Passera, G. Passarino and S. Uccirati, Nucl. Phys. B 680 (2004) 199. 241 32. A. Ghinculov and Y. Yao, Phys. Rev. D 63 (2001) 054510; Nucl. Phys. B 516 (1998) 385. 241 33. J.L. Rosner, Ann. Phys. 44 (1967) 11. 237 34. V.A. Smirnov, Renormalization and Asymptotic Expansions (Birkh¨ auser, Basel, 1991). 233, 240 35. N.I. Ussyukina, Teor. Mat. Fiz. 54 (1983) 124. 237 36. A.N. Vassiliev, Yu.M. Pis’mak and Yu.R. Honkonen, Teor. Mat. Fiz. 47 (1981) 291. 233, 236 37. J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037. 38. A.A. Vladimirov, Teor. Mat. Fiz. 43 (1980) 210. 233, 239 39. http://www.cecm.sfu.ca 242

List of Symbols

Aij r – matrix which deﬁnes denominators of the propagators al – power of a propagator (index) ci (a1 , . . . , aN ) – coeﬃcient function of a master integral Ii ˜ F – propagator in coordinate space D DF , DF,i – propagator in momentum space d – space-time dimension Er – denominator of propagator FΓ – Feynman integral 2 F1 (a, b; c; z) – Gauss hypergeometric function G(λ1 , λ2 ) – function in one-loop massless integration formula gµν – metric tensor Ha1 ,a2 ,...,an (x) – harmonic polylogarithm (HPL) h – number of loops Ii – master integral k – loop momentum L – number of lines Lia (z) – polylogarithm l – loop momentum m – mass P (x1 , . . . , xN ) – basic polynomial p – external or internal momentum Q2 = −q 2 – Euclidean external momentum squared q – external momentum Sa,b (z) – generalized polylogarithm Sj , Sjk ,. . . – nested sums s = (p1 + p2 )2 – Mandelstam variable

T – tree, 2-tree, pseudotree t = (p1 + p3 )2 – Mandelstam variable tl – sector variable U – function in the alpha representation u = (p1 + p4 )2 – Mandelstam variable ul – auxiliary parameter V – number of vertices V – function in the alpha representation w – variable in MB integrals x – coordinate xi – variable in the basic parametric representation Zl – polynomial in propagator z, zi – variable in MB integrals αl – alpha parameter βl = 1/αl – inverse alpha parameter Γ – graph Γ (x) – gamma function (ﬁrst Euler integral) γ – subgraph γE = 0.577216 . . . – Euler’s constant δ(x) – delta function ε = (4 − d)/2 – parameter of dimensional regularization ζ(z) – Riemann zeta function λl – parameter of analytic regularization ξ, ξi – Feynman parameter τl – sector variable ψ(x) = Γ (z)/Γ (z) – logarithmical derivative of the gamma function ω – degree of UV divergence

Index

alpha parameters 15 auxiliary master integral Baikov’s method

method of diﬀerence equations 240 method of diﬀerential equations (DE) 7, 165 momentum Euclidean 225 external 12 internal 12 loop 12

145

133

Cheng–Wu theorem

42

degree of UV divergence dispersion integral 233 divergence 14 collinear 17 IR 16 on-shell IR 17 threshold IR 17 UV 14

14

nested sums

partial fractions 35 Pochhammer symbol polylogarithm 187 propagator 11 PSLQ 241

Feynman amplitude 12 Feynman integral 12 Feynman parameters 41 ﬁrst Barnes lemma 207

188

index (power of a propagator) 11 integer relation algorithm 242 integration by parts (IBP) 2, 65, 109 IR rearrangement 237 left poles

187

recursively one-loop diagrams regularization 20 analytic 21 dimensional 22, 23 Pauli–Villars 21 Riemann zeta function 191 right poles 56

Gauss hypergeometric function 187 Gegenbauer polynomial x-space technique (GPXT) 234 generalized polylogarithm 187 gluing 235 graph 12 harmonic polylogarithm (HPL)

191

second Barnes lemma 214 sectors 223 shifting dimension 36, 120 subgraph detachable 24 divergent 15 one-particle-irreducible (1PI) one-vertex-reducible 224

34

15

56

Mandelstam variables 39 master integral 2, 109, 133 Mellin–Barnes (MB) representation 55, 56

tadpole 24, 28 tree 19 two-dimensional HPL (2dHPL) 4, uniqueness relations

236

174

Starting with Volume 165, Springer Tracts in Modern Physics is part of the [SpringerLink] service. For all customers with standing orders for Springer Tracts in Modern Physics we offer the full text in electronic form via [SpringerLink] free of charge. Please contact your librarian who can receive a password for free access to the full articles by registration at: springerlink.com If you do not have a standing order you can nevertheless browse online through the table of contents of the volumes and the abstracts of each article and perform a full text search. There you will also f ind more information about the series.

Springer Tracts in Modern Physics Springer Tracts in Modern Physics provides comprehensive and critical reviews of topics of current interest in physics. The following ﬁelds are emphasized: elementary particle physics, solid-state physics, complex systems, and fundamental astrophysics. Suitable reviews of other ﬁelds can also be accepted. The editors encourage prospective authors to correspond with them in advance of submitting an article. For reviews of topics belonging to the above mentioned ﬁelds, they should address the responsible editor, otherwise the managing editor. See also springeronline.com

Managing Editor

Solid-State Physics, Editors

Gerhard H¨ohler

Andrei Ruckenstein Editor for The Americas

Institut f¨ur Theoretische Teilchenphysik Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 33 75 Fax: +49 (7 21) 37 07 26 Email: [email protected] www-ttp.physik.uni-karlsruhe.de/

Elementary Particle Physics, Editors

Department of Physics and Astronomy Rutgers, The State University of New Jersey 136 Frelinghuysen Road Piscataway, NJ 08854-8019, USA Phone: +1 (732) 445 43 29 Fax: +1 (732) 445-43 43 Email: [email protected] www.physics.rutgers.edu/people/pips/ Ruckenstein.html

Johann H. K¨uhn

Peter W¨olﬂe

Institut f¨ur Theoretische Teilchenphysik Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 33 72 Fax: +49 (7 21) 37 07 26 Email: [email protected] www-ttp.physik.uni-karlsruhe.de/∼jk

Institut f¨ur Theorie der Kondensierten Materie Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 35 90 Fax: +49 (7 21) 6 08 77 79 Email: woelﬂ[email protected] www-tkm.physik.uni-karlsruhe.de

Thomas M¨uller Institut f¨ur Experimentelle Kernphysik Fakult¨at f¨ur Physik Universit¨at Karlsruhe Postfach 69 80 76128 Karlsruhe, Germany Phone: +49 (7 21) 6 08 35 24 Fax: +49 (7 21) 6 07 26 21 Email: [email protected] www-ekp.physik.uni-karlsruhe.de

Fundamental Astrophysics, Editor Joachim Tr¨umper Max-Planck-Institut f¨ur Extraterrestrische Physik Postfach 13 12 85741 Garching, Germany Phone: +49 (89) 30 00 35 59 Fax: +49 (89) 30 00 33 15 Email: [email protected] www.mpe-garching.mpg.de/index.html

Complex Systems, Editor Frank Steiner Abteilung Theoretische Physik Universit¨at Ulm Albert-Einstein-Allee 11 89069 Ulm, Germany Phone: +49 (7 31) 5 02 29 10 Fax: +49 (7 31) 5 02 29 24 Email: [email protected] www.physik.uni-ulm.de/theo/qc/group.html

Vladimir A. Smirnov

Evaluating Feynman Integrals With 48 Figures

123

Vladimir A. Smirnov II. Institut f¨ur Theoretische Physik Universit¨at Hamburg Luruper Chaussee 149 22761 Hamburg, Germany E-mail: [email protected]

Lomonosov Moscow State University Skobeltsyn Institute of Nuclear Physics Moscow 119992, Russia E-mail: [email protected]

Library of Congress Control Number: 2004115458

Physics and Astronomy Classiﬁcation Scheme (PACS): 12.38.Bx, 12.15.Lk, 02.30.Gp

ISSN print edition: 0081-3869 ISSN electronic edition: 1615-0430 ISBN 3-540-23933-2 Springer Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable for prosecution under the German Copyright Law. Springer is a part of Springer Science+Business Media springeronline.com © Springer-Verlag Berlin Heidelberg 2004 Printed in Germany The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Typesetting: by the author and TechBooks using a Springer LATEX macro package Cover concept: eStudio Calamar Steinen Cover production: design &production GmbH, Heidelberg Printed on acid-free paper

SPIN: 10985380

56/3141/jl

543210

Preface

The goal of this book is to describe in detail how Feynman integrals1 can be evaluated analytically. The problem of evaluating Lorentz-covariant Feynman integrals over loop momenta originated in the early days of perturbative quantum ﬁeld theory. Over a span of more than ﬁfty years, a great variety of methods for evaluating Feynman integrals has been developed. This book is a ﬁrst attempt to summarize them. I understand that if another person – in particular one actively involved in developing methods for Feynman integral evaluation – made a similar attempt, he or she would probably concentrate on some other methods and would rank the methods as most important and less important in a diﬀerent order. I believe, however, that my choice is reasonable. At least I have tried to concentrate on the methods that have been used in the past few years in the most sophisticated calculations, in which world records in the Feynman integral ‘sport’ were achieved. The problem of evaluation is very important at the moment. What could be easily evaluated was evaluated many years ago. To perform important calculations at the two-loop level and higher one needs to choose adequate methods and combine them in a non-trivial way. In the present situation – which might be considered boring because the Standard Model works more or less properly and there are no glaring contradictions with experiment – one needs not only to organize new experiments but also perform rather nontrivial calculations for further crucial high-precision checks. So I hope very much that this book will be used as a textbook in practical calculations. I shall concentrate on analytical methods and only brieﬂy describe numerical ones. Some methods are also characterized as semi-analytical, for example, the method based on asymptotic expansions of Feynman integrals in momenta and masses which was described in detail in my previous book. In this method, it is also necessary to apply some analytical methods of evaluation which were described there only very brieﬂy. So the present book can be considered as Volume 1 with respect to the previous book, which might be termed Volume 2, or the sequel. 1

Let us point out from beginning that two kinds of integrals are associated with Feynman: integrals over loop momenta and path integrals. We will deal only with the former case.

VI

Preface

Although all the necessary deﬁnitions concerning Feynman integrals are provided in the book, it would be helpful for the reader to know the basics of perturbative quantum ﬁeld theory, e.g. by following the ﬁrst few chapters of the well-known textbooks by Bogoliubov and Shirkov and/or Peskin and Schroeder. This book is based on the course of lectures which I gave in the winter semester of 2003–2004 at the Universities of Hamburg and Karlsruhe as a DFG Mercator professor in Hamburg. It is my pleasure to thank the students, postgraduate students, postdoctoral fellows and professors who attended my lectures for numerous stimulating discussions. I am grateful very much to B. Feucht, A.G. Grozin and J. Piclum for careful reading of preliminary versions of the whole book and numerous comments and suggestions; to M. Czakon, M. Kalmykov, P. Mastrolia, J. Piclum, M. Steinhauser and O.L. Veretin for valuable assistance in presenting examples in the book; to C. Anastasiou, K.G. Chetyrkin and A.I. Davydychev for various instructive discussions; to P.A. Baikov, M. Beneke, K.G. Chetyrkin, A. Czarnecki, A.I. Davydychev, B. Feucht, G. Heinrich, A.A. Penin, A. Signer, M. Steinhauser and O.L. Veretin for fruitful collaboration on evaluating Feynman integrals; to M. Czakon, A. Czarnecki, T. Gehrmann, J. Gluza, T. Riemann, K. Melnikov, E. Remiddi and J.B. Tausk for stimulating competition; to Z. Bern, L. Dixon, C. Greub and S. Moch for various pieces of advice; and to B.A. Kniehl and J.H. K¨ uhn for permanent support. I am thankful to my family for permanent love, sympathy, patience and understanding. Moscow – Hamburg, October 2004

V.A. Smirnov

Contents

1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

Feynman Integrals: Basic Deﬁnitions and Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Feynman Rules and Feynman Integrals . . . . . . . . . . . . . . . . . . . . 2.2 Divergences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Alpha Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Properties of Dimensionally Regularized Feynman Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

4

1 8 9 11 11 14 18 20 24 29

Evaluating by Alpha and Feynman Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Simple One- and Two-Loop Formulae . . . . . . . . . . . . . . . . . . . . . 3.2 Auxiliary Tricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Recursively One-Loop Feynman Integrals . . . . . . . . . . . . 3.2.2 Partial Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Dealing with Numerators . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Feynman Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Two-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31 31 34 34 35 36 38 41 43 52

Evaluating by MB Representation . . . . . . . . . . . . . . . . . . . . . . . . 4.1 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Multiple MB Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 More One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Two-Loop Massless Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Two-Loop Massive Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Three-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 More Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.8 MB Representation versus Expansion by Regions . . . . . . . . . . .

55 56 63 65 71 81 92 98 102

VIII

Contents

4.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 5

IBP and Reduction to Master Integrals . . . . . . . . . . . . . . . . . . . 5.1 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Two-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Reduction of On-Shell Massless Double Boxes . . . . . . . . . . . . . . 5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6

Reduction to Master Integrals by Baikov’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Basic Parametric Representation . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Constructing Coeﬃcient Functions. Simple Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 General Recipes. Complicated Examples . . . . . . . . . . . . . . . . . . . 6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

109 109 114 120 127 130 133 133 138 146 152 162 163

7

Evaluation by Diﬀerential Equations . . . . . . . . . . . . . . . . . . . . . . 7.1 One-Loop Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Two-Loop Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

A

Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 A.1 Table of Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 A.2 Some Useful Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

B

Some Special Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

C

Summation Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C.1 Some Number Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C.2 Power Series of Levels 3 and 4 in Terms of Polylogarithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C.3 Inverse Binomial Power Series up to Level 4 . . . . . . . . . . . . . . . C.4 Power Series of Levels 5 and 6 in Terms of HPL . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

D

165 165 170 173 176

191 192 197 198 200 204

Table of MB Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 D.1 MB Integrals with Four Gamma Functions . . . . . . . . . . . . . . . . . 207 D.2 MB Integrals with Six Gamma Functions . . . . . . . . . . . . . . . . . . 214

Contents

E

F

IX

Analysis of Convergence and Sector Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.1 Analysis of Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.2 Practical Sector Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

221 221 229 232

A Brief Review of Some Other Methods . . . . . . . . . . . . . . . . . . F.1 Dispersion Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.2 Gegenbauer Polynomial x-Space Technique . . . . . . . . . . . . . . . . F.3 Gluing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.4 Star-Triangle Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.5 IR Rearrangement and R∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.6 Diﬀerence Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . F.7 Experimental Mathematics and PSLQ . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

233 233 234 235 236 237 240 241 243

List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247

1 Introduction

The important mathematical problem of evaluating Feynman integrals arises quite naturally in elementary-particle physics when one treats various quantities in the framework of perturbation theory. Usually, it turns out that a given quantum-ﬁeld amplitude that describes a process where particles participate cannot be completely treated in the perturbative way. However it also often turns out that the amplitude can be factorized in such a way that diﬀerent factors are responsible for contributions of diﬀerent scales. According to a factorization procedure a given amplitude can be represented as a product of factors some of which can be treated only non-perturbatively while others can be indeed evaluated within perturbation theory, i.e. expressed in terms of Feynman integrals over loop momenta. A useful way to perform the factorization procedure is provided by solving the problem of asymptotic expansion of Feynman integrals in the corresponding limit of momenta and masses that is determined by the given kinematical situation. A universal way to solve this problem is based on the so-called strategy of expansion by regions [3, 10]. This strategy can be itself regarded as a (semianalytical) method of evaluation of Feynman integrals according to which a given Feynman integral depending on several scales can be approximated, with increasing accuracy, by a ﬁnite sum of ﬁrst terms of the corresponding expansion, where each term is written as a product of factors depending on diﬀerent scales. A lot of details concerning expansions of Feynman integrals in various limits of momenta and/or masses can be found in my previous book [10]. In this book, however, we shall mainly deal with purely analytical methods. One needs to take into account various graphs that contribute to a given process. The number of graphs greatly increases when the number of loops gets large. For a given graph, the corresponding Feynman amplitude is represented as a Feynman integral over loop momenta, due to some Feynman rules. The Feynman integral, generally, has several Lorentz indices. The standard way to handle tensor quantities is to perform a tensor reduction that enables us to write the given quantity as a linear combination of tensor monomials with scalar coeﬃcients. Therefore we shall imply that we deal with scalar Feynman integrals and consider only them in examples. A given Feynman graph therefore generates various scalar Feynman integrals that have the same structure of the integrand with various distributions V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 1–9 (2004) c Springer-Verlag Berlin Heidelberg 2004

2

1 Introduction

of powers of propagators (indices). Let us observe that some powers can be negative, due to some initial polynomial in the numerator of the Feynman integral. A straightforward strategy is to evaluate, by some methods, every scalar Feynman integral resulting from the given graph. If the number of these integrals is small this strategy is quite reasonable. In non-trivial situations, where the number of diﬀerent scalar integrals can be at the level of hundreds and thousands, this strategy looks too complicated. A well-known optimal strategy here is to derive, without calculation, and then apply some relations between the given family of Feynman integrals as recurrence relations. A well-known standard way to obtain such relations is provided by the method of integration by parts1 (IBP) [6] which is based on putting to zero any integral of the form ∂f dd k1 dd k2 . . . µ ∂ki over loop momenta k1 , k2 , . . . , ki , . . . within dimensional regularization with the space-time dimension d = 4 − 2ε as a regularization parameter [4, 5, 7]. Here f is an integrand of a Feynman integral; it depends on the loop and external momenta. More precisely, one tries to use IBP relations in order to express a general dimensionally regularized integral from the given family as a linear combination of some irreducible integrals which are also called master integrals. Therefore the whole problem decomposes into two parts: a solution of the reduction procedure and the evaluation of the master Feynman integrals. Observe that in such complicated situations, with the great variety of relevant scalar integrals, one really needs to know a complete solution of the recursion problem, i.e. to learn how an arbitrary integral with general integer powers of the propagators and powers of irreducible monomials in the numerator can be evaluated. To illustrate the methods of evaluation that we are going to study in this book let us ﬁrst orient ourselves at the evaluation of individual Feynman integrals, which might be master integrals, and take the simple scalar oneloop graph Γ shown in Fig. 1.1 as an example. The corresponding Feynman integral constructed with scalar propagators is written as dd k 2 2 . (1.1) FΓ (q , m ; d) = (k 2 − m2 )(q − k)2 1

As is explained in textbooks b on integral calculus, b the method of IBP is applied with the help of the relation a dxuv = uv|ba − a dxu v as follows. One tries to represent the integrand as uv with some u and v in such a way that the integral on the right-hand side, i.e. of u v will be simpler. We do not follow this idea in the case of Feynman integrals. Instead we only use the fact that an integral of the derivative of some function is zero, i.e. we always neglect the corresponding surface terms. So the name of the method looks misleading. It is however unambiguously accepted in the physics community.

1 Introduction

3

Fig. 1.1. One-loop self-energy graph. The dashed line denotes a massless propagator

The same picture Fig. 1.1 can also denote the Feynman integral with general powers of the two propagators, dd k . (1.2) FΓ (q 2 , m2 ; a1 , a2 , d) = 2 2 (k − m )a1 [(q − k)2 ]a2 Suppose, one needs to evaluate the Feynman integral FΓ (q 2 , m2 ; 2, 1, d) ≡ F (2, 1, d) which is ﬁnite in four dimensions, d = 4. (It can also be depicted by Fig. 1.1 with a dot on the massive line.) There is a lot of ways to evaluate it. For example, a straightforward way is to take into account the fact that the given function of q is Lorentz-invariant so that it depends on the external momentum through its square, q 2 . One can choose a frame q = (q0 , 0), introduce spherical coordinates for k, integrate over angles, then over the radial component and, ﬁnally, over k0 . This strategy can be, however, hardly generalized to multi-loop2 Feynman integrals. Another way is to use a dispersion relation that expresses Feynman integrals in terms of a one-dimensional integral of the imaginary part of the given Feynman integral, from the value of the lowest threshold to inﬁnity. This dispersion integral can be expressed by means of the well-known Cutkosky rules. We shall not apply this method, which was, however, very popular in the early days of perturbative quantum ﬁeld theory, and only brieﬂy comment on it in Appendix F. Let us now turn to the methods that will be indeed actively used in this book. To illustrate them all let me use this very example of Feynman integrals (1.2) and present main ideas of these methods, with the obligation to present the methods in great details in the rest of the book. First, we will exploit the well-known technique of alpha or Feynman parameters. In the case of F (2, 1, d), one writes down the following Feynmanparametric formula: 1 ξdξ 1 = 2 . (1.3) 2 2 2 2 2 2 2 3 (k − m ) (q − k) 0 [(k − m )ξ + (1 − ξ)(q − k) + i0] Then one can change the order of integration over ξ and k, perform integration over k with the help of the formula (A.1) (which we will derive in Chap. 3) and obtain the following representation: 1 dξ ξ −ε d/2 F (2, 1, d) = −iπ Γ (1 + ε) . (1.4) 2 2 1+ε 0 [m − q (1 − ξ) − i0] 2

Since the Feynman integrals are rather complicated objects the word ‘multiloop’ means the number of loops greater than one ;-)

4

1 Introduction

This integral is easily evaluated at d = 4 with the following result: ln 1 − q 2 /m2 . F (2, 1, 4) = iπ 2 q2

(1.5)

In principle, any given Feynman integral F (a1 , a2 , d) with concrete numbers a1 and a2 can similarly be evaluated by Feynman parameters. In particular, F (1, 1, d) reduces to 1 dξ ξ −ε d/2 F (1, 1, d) = iπ Γ (ε) . (1.6) 2 2 ε 0 [m − q (1 − ξ) − i0] There is an ultraviolet (UV) divergence which manifests itself in the ﬁrst pole of the function Γ (ε), i.e. at d = 4. The integral can be evaluated in expansion in a Laurent series in ε, for example, up to ε0 : 1 − ln m2 + 2 F (1, 1, d) = iπ d/2 e−γE ε ε m2 q2 − 1 − 2 ln 1 − 2 + O(ε) , (1.7) q m where γE is Euler’s constant. We shall study the method of Feynman and alpha parameters in Chap. 3. Another method which plays an essential role in this book is based on the Mellin–Barnes (MB) representation. The underlying idea is to replace a sum of terms raised to some power by the product of these terms raised to certain powers, at the cost of introducing an auxiliary integration that goes from −i∞ to +i∞ in the complex plane. The most natural way to apply this representation is to write down a massive propagator in terms of massless ones. For F (2, 1, 4), we can write +i∞ 1 (m2 )z 1 = dz Γ (2 + z)Γ (−z) . (1.8) 2 2 2 (m − k ) 2πi −i∞ (−k 2 )2+z Applying (1.8) to the ﬁrst propagator in (1.2), changing the order of integration over k and z and evaluating the internal integral over k by means of the one-loop formula (A.7) (which we will derive in Chap. 3) we arrive at the following onefold MB integral representation: +i∞ 2 z m iπ d/2 Γ (1 − ε) 1 dz F (2, 1, d) = − (−q 2 )1+ε 2πi −i∞ −q 2 Γ (1 + ε + z)Γ (−ε − z)Γ (−z) . (1.9) × Γ (1 − 2ε − z) The contour of integration is chosen in the standard way: the poles with a Γ (. . . + z) dependence are to the left of the contour and the poles with a Γ (. . . − z) dependence are to the right of it. If |ε| is small enough we can choose this contour as a straight line parallel to the imaginary axis with −1 < Rez < 0. For d = 4, we obtain

1 Introduction

F (2, 1, 4) = −

iπ 2 1 q 2 2πi

+i∞

dz −i∞

m2 −q 2

5

z Γ (z)Γ (−z) .

(1.10)

By closing the integration contour to the right and taking a series of residues at the points z = 0, 1, . . ., we reproduce (1.5). Using the same technique, any integral from the given family can similarly be evaluated. We shall study the technique of MB representation in Chap. 4 where we shall see, through various examples, how, by introducing MB integrations in an appropriate way, one can analytically evaluate rather complicated Feynman integrals. Let us, however, think about a more economical strategy based on IBP relations which would enable us to evaluate any integral (1.2) as a linear combination of some master integrals. Putting to zero dimensionally regularized ∂ ∂ ·kf (a1 , a2 ) and q· ∂k f (a1 , a2 ), where f (a1 , a2 ) is the integrand integrals of ∂k in (1.2), and writing down obtained relations in terms of integrals of the given family we obtain the following two IBP relations: d − 2a1 − a2 − 2m2 a1 1+ − a2 2+ (1− − q 2 + m2 ) = 0 , −

−

a2 − a1 − a1 1 (q + m − 2 ) − a2 2 (1 − q + m ) = 0 , +

2

2

+

2

2

(1.11) (1.12)

in the sense that they are applied to the general integral F (a1 , a2 ). Here the standard notation for increasing and lowering operators has been used, e.g. 1+ 2− F (a1 , a2 ) = F (a1 + 1, a2 − 1). Let us observe that any integral with a1 ≤ 0 is zero because it is a massless tadpole which is naturally put to zero within dimensional regularization. Moreover, any integral with a2 ≤ 0 can be evaluated in terms of gamma functions for general d with the help of (A.3) (which we will derive in Chap. 3). The number a2 can be reduced either to one or to a non-positive value using the following relation which is obtained as the diﬀerence of (1.11) multiplied by q 2 + m2 and (1.12) multiplied by 2m2 : (q 2 − m2 )2 a2 2+ = (q 2 − m2 )a2 1− 2+ −(d − 2a1 − a2 )q 2 − (d − 3a2 )m2 + 2m2 a1 1+ 2− . (1.13) Indeed, when the left-hand side of (1.13) is applied to F (a1 , a2 ), we obtain integrals with reduced a2 or, due to the ﬁrst term on the right-hand side, reduced a1 . Suppose now that a2 = 1. Then we can use the diﬀerence of relations (1.11) and (1.12), d − a1 − 2a2 − a1 1+ (2− − q 2 + m2 ) = 0 ,

(1.14)

by writing down a1 (q 2 − m2 )1+ through the rest terms, and reduce the index a1 to one or the index a2 to zero. We see that we can now express any integral of the given family as a linear combination of the integral F (1, 1) and simple integrals with a2 ≤ 0 which can be evaluated for general d in terms of gamma functions. In particular, we have

6

1 Introduction

F (2, 1) =

m2

1 [(1 − 2ε)F (1, 1) − F (2, 0)] . − q2

(1.15)

At this point, we can stop our activity because we have already essentially solved the problem. In fact, we shall later encounter several examples of non-trivial calculations where any integral is expressed in terms of some complicated master integrals and families of simple integrals. However, mathematically (and aesthetically), it is natural to be more curious and wonder about the minimal number of master integrals which form a linearly independent basis in the family of integrals F (a1 , a2 ). We will do this in Chaps. 5 and 6. In Chap. 5, we shall investigate various examples, starting from simple ones, where the reduction of a given class of Feynman integrals can be performed by solving IBP recurrence relations. If we want to be maximalists, i.e. we are oriented at the minimal number of master integrals, we expect that any Feynman integral from a given family, F (a1 , a2 , . . .) can be expressed linearly in terms of a ﬁnite set of master integrals: ci (F (a1 , a2 , . . .))Ii , (1.16) F (a1 , a2 , . . .) = i

These master integrals Ii cannot be reduced further, i.e. expressed as linear combinations of other Feynman integrals of the given family. There were several attempts to systematize the procedure of solving IBP recurrence relations. Some of them will be described in the end of Chap. 5. One of the corresponding methods [1, 2, 11] is based on an appropriate parametric representation which is used to construct the coeﬃcient functions ci (F (a1 , a2 , . . .)) ≡ ci (a1 , a2 , . . .) in (1.16). The integrand of this representai , where the integration parameters tion consists of the standard factors x−a i xi correspond to the denominators of the propagators, and a polynomial in these variables raised to the power (d − h − 1)/2, where h is the number of loops for vacuum integrals and some eﬀective loop number, otherwise. This polynomial is constructed for the given family of integrals according to some simple rules. An important property of such a representation is that it automatically satisﬁes IBP relations written for this family of integrals, provided one can use IBP in this parametric representation. For example, for the family of integrals F (a1 , a2 ) we are dealing with in this chapter, the auxiliary representation takes the form dx1 dx2 [P (x1 , x2 )](d−3)/2 , (1.17) ci (a1 , a2 ) ∼ xa1 1 xa2 2 with the basic polynomial P (x1 , x2 ) = −(x1 − x2 + m2 )2 − q 2 (q 2 − 2m2 − 2(x1 + x2 )) .

(1.18)

As we shall see in Chap. 6, such auxiliary representation provides the possibility to characterize the master integrals and construct algorithms for the evaluation of the corresponding coeﬃcient functions. When looking for

1 Introduction

7

candidates for the master integrals one considers integrals of the type (1.17) with indices ai equal to one or zero and tries to see whether such integrals can be understood non-trivially. According to a general rule, which we will explain in Chap. 6, the value ai = 1 of some index forces us to understand the integration over the corresponding parameter xi as a Cauchy integration contour around the origin in the complex xi -plane which in turn reduces to taking derivatives of the factor P (d−3)/2 in xi at xi = 0. If an index ai is equal to zero one has to understand the corresponding integration in some sense, which implies the validity of IBP in the integration over xi . In our present example, let us therefore consider the candidates F (1, 1), F (1, 0), F (0, 1) and F (0, 0). Of course, we neglect the last two of them because they are massless tadpoles. Thus we are left with the ﬁrst two integrals. According to the rule formulated above, the coeﬃcient function of F (1, 1) is evaluated as an iterated Cauchy integral over x1 and x2 . It is therefore constructed in a non-trivial (non-zero) way and this integral is recognized as a master integral. For F (1, 0), only the integration over x1 is understood as a Cauchy integration, and the representation (1.17) gives, for the corresponding coeﬃcient function, a linear combination of terms (d−3)/2−l dx2

−(m2 − q 2 )2 + 2(m2 + q 2 )x2 − x22 , (1.19) j x2 with integer j and non-negative integer l. When j ≤ 0, the integration can be taken between the roots of the quadratic polynomial in the square brackets. Thus one can again construct a non-zero coeﬃcient function and the integral F (1, 0) turns out to be our second (and the last) master integral. We shall see in Chap. 6 how (1.17) can be understood for j > 0; this is indeed necessary for the construction of the coeﬃcient function c2 (a1 , a2 ) at a2 > 0. We shall also learn other details of this method illustrated though various examples. Anyway, the present example shows that this method enables an elegant and transparent classiﬁcation of the master integrals: the presence of (only two) master integrals F (1, 1) and F (1, 0) in the given recursion problem is seen in a very simple way, as compared with the complete solution of the reduction procedure outlined above. One more powerful method that has been proven very useful in the evaluation of the master integrals is based on using diﬀerential equations (DE) [8, 9]. Let us illustrate it again with the help of our favourite example. To evaluate the master integral F (1, 1) let us observe that its in m2 derivative 2 2 2 2 is nothing but F (2, 1) (because ∂/(∂m ) 1/(k − m ) = 1/(k − m2 )2 ) which is expressed, according to our reduction procedure, by (1.15). Therefore we arrive at the following diﬀerential equation for f (m2 ) = F (1, 1):

∂ 1 (1 − 2ε)f (m2 ) − F (2, 0) , f (m2 ) = 2 ∂m2 m − q2

(1.20)

where the quantity F (2, 0) is a simpler object because it can be evaluated in terms of gamma functions for general ε. The general solution to this equation

8

1 Introduction

can easily be obtained by the method of the variation of the constant, with ﬁxing the general solution from the boundary condition at m = 0. Eventually, the above result (1.7) can successfully be reproduced. As we shall see in Chap. 7, the strategy of the method of DE in much more non-trivial situations is similar: one takes derivatives of a master integral in some arguments, expresses them in terms of original Feynman integrals, by means of some variant of solution of IBP relations, and solves resulting diﬀerential equations. However, before studying the methods of evaluation, basic deﬁnitions are presented in Chap. 2 where tools for dealing with Feynman integrals are also introduced. Methods for evaluating individual Feynman integrals are studied in Chaps. 3, 4 and 7 and the reduction problem is studied in Chaps. 5 and 6. In Appendix A, one can ﬁnd a table of basic one-loop and two-loop Feynman integrals as well as some useful auxiliary formulae. Appendix B contains deﬁnitions and properties of special functions that are used in this book. A table of summation formulae for onefold series is given in Appendix C. In Appendix D, a table of onefold MB integrals is presented. Appendix E contains analysis of convergence of Feynman integrals as well a description of a numerical method of evaluating Feynman integrals based on sector decompositions. Some other methods are brieﬂy characterized in Appendix F. These are mainly old methods whose details can be found in the literature. If I do not present some methods, this means that either I do not know about them, or I do not know physically important situations where they work not worse than than the methods I present. I shall use almost the same examples in Chaps. 3–7 and Appendix F to illustrate all the methods. On the one hand, this will be done in order to have the possibility to compare them. On the other hand, the methods often work together: for example, MB representation can be used in alpha or Feynman parametric integrals, the method of DE requires a solution of the reduction problem, boundary conditions within the method of DE can be obtained by means of the method of MB representation, auxiliary IBP relations within the method described in Chap. 6 can be solved by means of an algorithm originated within another approach to solving IBP relations. Basic notational conventions are presented below. The notation is described in more detail in the List of Symbols. In the Index, one can ﬁnd numbers of pages where deﬁnitions of basic notions are introduced.

1.1 Notation We use Greek and Roman letters for four-indices and spatial indices, respectively: xµ = (x0 , x) ,

References

9

q·x = q 0 x0 − q·x ≡ gµν q µ xν . The parameter of dimensional regularization is d = 4 − 2ε . The d-dimensional Fourier transform and its inverse are deﬁned as ˜ f (q) = dd x eiq·x f (x) , 1 f (x) = dd q e−ix·q f˜(q) . (2π)d In order to avoid Euler’s constant γE in Laurent expansions in ε, we pull out the factor e−γE ε per loop.

References 1. P.A. Baikov, Phys. Lett. B 385 (1996) 404; Nucl. Instrum. Methods A 389 (1997) 347. 6 2. P.A. Baikov and M. Steinhauser, Comput. Phys. Commun. 115 (1998) 161. 6 3. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 1 4. C.G. Bollini and J.J. Giambiagi, Nuovo Cim. B 12 (1972) 20. 2 5. P. Breitenlohner and D. Maison, Commun. Math. Phys. 52 (1977) 11, 39, 55. 2 6. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 2 7. G. ’t Hooft and M. Veltman, Nucl. Phys. B 44 (1972) 189. 2 8. A.V. Kotikov, Phys. Lett. B 254 (1991) 158; B 259 (1991) 314; B 267 (1991) 123. 7 9. E. Remiddi, Nuovo Cim. A 110 (1997) 1435. 7 10. V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 1 11. V.A. Smirnov and M. Steinhauser, Nucl. Phys. B 672 (2003) 199. 6

2 Feynman Integrals: Basic Deﬁnitions and Tools

In this chapter, basic deﬁnitions for Feynman integrals are given, ultraviolet (UV), infrared (IR) and collinear divergences are characterized, and basic tools such as alpha parameters are presented. Various kinds of regularizations, in particular dimensional one, are presented and properties of dimensionally regularized Feynman integrals are formulated and discussed.

2.1 Feynman Rules and Feynman Integrals In perturbation theory, any quantum ﬁeld model is characterized by a Lagrangian, which is represented as a sum of a free-ﬁeld part and an interaction part, L = L0 + LI . Amplitudes of the model, e.g. S-matrix elements and matrix elements of composite operators, are represented as power series in coupling constants. Starting from the S-matrix represented in terms of the time-ordered exponent of the interaction Lagrangian which is expanded with the application of the Wick theorem, or from Green functions written in terms of a functional integral treated in the perturbative way, one obtains that, in a ﬁxed perturbation order, the amplitudes are written as ﬁnite sums of Feynman diagrams which are constructed according to Feynman rules: lines correspond to L0 and vertices are determined by LI . The basic building block of the Feynman diagrams is the propagator that enters the relation T φi (x1 )φi (x2 ) = : φi (x1 )φi (x2 ) : +DF,i (x1 − x2 ) .

(2.1)

Here DF,i is the Feynman propagator of the ﬁeld of type i and the colons denote a normal product of the free ﬁelds. The Fourier transforms of the propagators have the form iZi (p) ˜ F,i (p) ≡ d4 x eip·x DF,i (x) = D , (2.2) 2 (p − m2i + i0)ai where mi is the corresponding mass, Zi is a polynomial and ai = 1 or 2 (for the gluon propagator in the general covariant gauge). The powers of the propagators al will be also called indices. For the propagator of the scalar ﬁeld, we have Z = 1, a = 1. This is not the most general form of the propagator. For example, in the axial or Coulomb gauge, the gluon propagator has another form. We usually omit the causal i0 for brevity. Polynomials V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 11–30 (2004) c Springer-Verlag Berlin Heidelberg 2004

12

2 Feynman Integrals: Basic Deﬁnitions and Tools

associated with vertices of graphs can be taken into account by means of the polynomials Zl . We also omit the factors of i and (2π)4 that enter in the standard Feynman rules (in particular, in (2.2)); these can be included at the end of a calculation. Eventually, we obtain, for any ﬁxed perturbation order, a sum of Feynman amplitudes labelled by Feynman graphs1 constructed from the given type of vertices and lines. In the commonly accepted physical slang, the graph, the corresponding Feynman amplitude and the integral are all often called the ‘diagram’. A Feynman graph diﬀers from a graph by distinguishing a subset of vertices which are called external. The external momenta or coordinates on which a Feynman integral depends are associated with the external vertices. Thus quantities that can be computed perturbatively are written, in any given order of perturbation theory, through a sum over Feynman graphs. For a given graph Γ , the corresponding Feynman amplitude

qi FΓ (q1 , . . . , qn ) (2.3) GΓ (q1 , . . . , qn+1 ) = (2π)4 i δ i

can be written in terms of an integral over loop momenta L ˜ F,l (pl ) , D FΓ (q1 , . . . , qn ) = d4 k1 . . . d4 kh

(2.4)

l=1

where d4 ki = dki0 dki , and a factor with a power of 2π is omitted, as we have agreed. The Feynman integral FΓ depends on n linearly independent external momenta qi = (qi0 , q i ); the corresponding integrand is a function of L internal momenta pi , which are certain linear combinations of the external momenta and h = L − V + 1 chosen loop momenta ki , where L, V and h are numbers of lines, vertices and (independent) loops, respectively, of the given graph. After some tensor reduction2 one can deal only with scalar Feynman integrals. To do this, various projectors can be applied. For example, in the case of Feynman integrals contributing to the electromagnetic formfactor (see Fig. 2.1) Γ µ (p1 , p2 ) = γ µ F1 (q 2 ) + σ µν qν F2 (q 2 ), where q = p1 − p2 , γ µ and σ µν are γ- and σ-matrices, respectively, the following projector can be applied to extract scalar integrals which contribute to the formfactor F1 in the massless case (with F2 = 0): 1 When dealing with graphs and Feynman integrals one usually does not bother about the mathematical deﬁnition of the graph and thinks about something that is built of lines and vertices. So, a graph is an ordered family {V, L, π± }, where V is the set of vertices, L is the set of lines, and π± : L → V are two mappings that correspond the initial and the ﬁnal vertex of a line. By the way, mathematicians use the word ‘edge’, rather than ‘line’. 2 In one-loop, the well-known general reduction was described in [23]. Steps towards systematical reduction at the two-loop level were made in [1].

2.1 Feynman Rules and Feynman Integrals

13

p1 q µ p2 Fig. 2.1. Electromagnetic formfactor

F1 (q 2 ) =

Tr [γµ p2 Γ µ (p1 , p2 ) p1 ] , 2(d − 2) q 2

(2.5)

where p = γ µ pµ and d is the parameter of dimensional regularization (to be discussed shortly in Sect. 2.4). Anyway, after applying some projectors, one obtains, for a given graph, a family of Feynman integrals which have various powers of the scalar parts of the propagators, 1/(p2l −m2l )al , and various monomials in the numerator. The denominators p2l can be expressed linearly in terms of scalar products of the loop and external momenta. The factors in the numerator can also be chosen as quadratic polynomials of the loop and external momenta raised to some powers. It is convenient to consider both types of the quadratic polynomials on the same footing and treat the factors in the numerators as extra factors in the denominator raised to negative powers. The set of the denominators for a given graph is linearly independent. It is natural to complete this set by similar factors coming from the numerator in such a way that the whole set will be linearly independent. Therefore we come to the following family of scalar integrals generated by the given graph: 4 d k1 . . . d4 kh (2.6) F (a1 , . . . , aN ) = · · · aN , E1a1 . . . EN where ki , i = 1, . . . , h, are loop momenta, ai are integer indices, and the denominators are given by 2 Aij (2.7) Er = r p i · pj − mr , i≥j≥1

with r = 1, . . . , N . The momenta pi are either the loop momenta pi = ki , i = 1, . . . , h, or independent external momenta ph+1 , . . . , ph+n of the graph. For a usual Feynman graph, the denominators Er determined by some matrix A are indeed quadratic. However, a more general class of Feynman integrals where the denominators are linear with respect to the loop and/or external momenta also often appears in practical calculations. Linear denominators usually appear in asymptotic expansions of Feynman integrals within the strategy of expansion by regions [2, 29]. Such expansions provide a useful link of an initial theory described by some Lagrangian with various eﬀective theories where, indeed, the denominators of propagators can be linear with

14

2 Feynman Integrals: Basic Deﬁnitions and Tools

respect to the external and loop momenta. For example, one encounters the following denominators: p · k, with an external momentum p on the light cone, p2 = 0, for the Sudakov limit and with p2 = 0 for the quark propagator of HQET [14, 19, 22]. Some non-relativistic propagators appear within threshold expansion and in the eﬀective theory called NRQCD [4, 18, 35], for example, the denominator k0 − k2 /(2m).

2.2 Divergences As has been known from early days of quantum ﬁeld theory, Feynman integrals suﬀer from divergences. This word means that, taken naively, these integrals are ill-deﬁned because the integrals over the loop momenta generally diverge. The ultraviolet (UV) divergences manifest themselves through a divergence of the Feynman integrals at large loop momenta. Consider, for example, the Feynman integral corresponding to the one-loop graph Γ of Fig. 2.2 with scalar propagators. This integral can be written as d4 k , (2.8) FΓ (q) = 2 (k 2 − m1 )[(q − k)2 − m22 ] where the loop momentum k is chosen as the momentum of the ﬁrst line. Introducing four-dimensional (generalized) spherical coordinates k = rkˆ in (2.8), where kˆ is on the unit (generalized) sphere and is expressed by means of three angles, and counting powers of propagators, we obtain, in the limit of ∞ large r, the following divergent behaviour: Λ dr r−1 . For a general diagram, a similar power counting at large values of the loop momenta gives 4h(Γ ) − 1 from the Jacobian that arises when one introduces generalized spherical coordinates in the (4 × h)-dimensional space of h loop four-momenta, plus a contribution from the powers of the ∞propagators and the degrees of its polynomials, and leads to an integral Λ dr rω−1 , where nl (2.9) ω = 4h − 2L + l

is the (UV) degree of divergence of the graph. (Here nl are the degrees of the polynomials Zl .)

Fig. 2.2. One-loop self-energy diagram

2.2 Divergences

15

This estimate shows that the Feynman integral is UV convergent overall (no divergences arise from the region where all the loop momenta are large) if the degree of divergence is negative. We say that the Feynman integral has a logarithmic, linear, quadratic, etc. overall divergence when ω = 0, 1, 2, . . ., respectively. To ensure a complete absence of UV divergences it is necessary to check convergence in various regions where some of the loop momenta become large, i.e. to satisfy the relation ω(γ) < 0 for all the subgraphs γ of the graph. We call a subgraph UV divergent if ω(γ) ≥ 0. In fact, it is suﬃcient to check these inequalities only for one-particle-irreducible (1PI) subgraphs (which cannot be made disconnected by cutting a line). It turns out that these rough estimates are indeed true – see some details in Sect. E.1. If we turn from momentum space integrals to some other representation of Feynman diagrams, the UV divergences will manifest themselves in other ways. For example, in coordinate space, the Feynman amplitude (i.e. the inverse Fourier transform of (2.3)) is expressed in terms of a product of the Fourier transforms of propagators L

DF,l (xli − xlf )

(2.10)

l=1

integrated over four-coordinates xi corresponding to the internal vertices. Here li and lf are the beginning and the end, respectively, of a line l. The propagators in coordinate space, 1 ˜ F,l (p)e−ix·p , DF,l (x) = d4 p D (2.11) (2π)4 are singular at small values of coordinates x = (x0 , x). To reveal this singularity explicitly let us write down the propagator (2.2) in terms of an integral over a so-called alpha-parameter 2 2 1 ∂ (−i)al ∞ 2iul ·p ˜ DF,l (p) = i Zl dαl αlal −1 ei(p −m )αl . e 2i ∂ul Γ (al ) ul =0

0

(2.12) which turns out to be a very useful tool both in theoretical analyses and practical calculations. To present an explicit formula for the scalar (i.e. for a = 1 and Z = 1) propagator ∞ 2 2 ˜ DF (p) = dα ei(p −m )α (2.13) 0

in coordinate space we insert (2.13) into (2.11), change the order of integration over p and α and take the Gaussian integrations explicitly using the formula 2 2 d4 k ei(αk −2q·k) = −iπ 2 α−2 e−iq /α , (2.14)

16

2 Feynman Integrals: Basic Deﬁnitions and Tools

which is nothing but a product of four one-dimensional Gaussian integrals: ∞ 2 π −iq02 /α+iπ/4 e dk0 ei(αk0 −2q0 k0 ) = , α −∞ ∞ π iqj2 /α−iπ/4 −i(αkj2 −2qj kj ) e dkj e = , j = 1, 2, 3 (2.15) α −∞ (without summation over j in the last formula). The ﬁnal integration is then performed using [26] or in MATHEMATICA [37] with the following result: m √ DF (x) = K1 m −x2 + i0 4π 2 −x2 + i0 1 1 + O m2 ln m2 , (2.16) =− 2 2 4π x − i0 where K1 is a Bessel special function [12]. The leading singularity at x = 0 is given by the value of the coordinate space massless propagator. Thus, the inverse Fourier transform of the convolution integral (2.8) equals the square of the coordinate-space scalar propagator, with the singularity (x2 − i0)−2 . Power-counting shows that this singularity produces integrals that are divergent in the vicinity of the point x = 0, and this is the coordinate space manifestation of the UV divergence. The divergences caused by singularities at small loop momenta are called infrared (IR) divergences. First we distinguish IR divergences that arise at general values of the external momenta. A typical example of such a divergence is given by the graph of Fig. 2.2 when one of the lines contains the second power of the corresponding propagator, so that a1 = 2. If the mass of this line is zero we obtain a factor 1/(k 2 )2 in the integrand, where k is chosen as the momentum of this line. Then, keeping in mind the introduction of generalized spherical coordinates and performing power-counting at small k (i.e. when all the components of the four-vector k are small), we again enΛ counter a divergent behaviour 0 dr r−1 but now at small values of r. There is a similarity between the properties of IR divergences of this kind and those of UV divergences. One can deﬁne, for such oﬀ-shell IR divergences, an IR degree of divergence, in a similar way to the UV case. A reasonable choice is provided by the value ω ˜ (γ) = −ω(Γ/γ) ≡ ω(γ) − ω(Γ ) ,

(2.17)

where γ ≡ Γ \γ is the completion of the subgraph γ in a given graph Γ and Γ/γ denotes the reduced graph which is obtained from Γ by reducing every connectivity component of γ to a point. The absence of oﬀ-shell IR divergences is guaranteed if the IR degrees of divergence are negative for all massless subgraphs γ whose completions γ include all the external vertices in the same connectivity component. (See details in [8, 27] and Sect. E.1.) The oﬀ-shell IR divergences are the worst but they are in fact absent in physically

2.2 Divergences

17

meaningful theories. However, they play an important role in asymptotic expansions of Feynman diagrams (see [29]). The other kinds of IR divergences arise when the external momenta considered are on a surface where the Feynman diagram is singular: either on a mass shell or at a threshold. Consider, for example, the graph Fig. 2.2, with the indices a1 = 1 and a2 = 2 and the masses m1 = 0 and m2 = m = 0 on the mass shell, q 2 = m2 . With k as the momentum of the second line, the corresponding Feynman integral is of the form d4 k FΓ (q; d) = . (2.18) 2 2 k (k − 2q·k)2 At small values of k, the integrand behaves like 1/[4k 2 (q·k)2 ], and, with the help of power counting, we see that there is an on-shell IR divergence which would not be present for q 2 = m2 . If we consider Fig. 2.2 with equal masses and indices a1 = a2 = 2 at the threshold, i.e. at q 2 = 4m2 , it might seem that there is a threshold IR divergence because, choosing the momenta of the lines as q/2 + k and q/2 − k, we obtain the integral d4 k , (2.19) 2 (k + q·k)2 (k 2 − q·k)2 with an integrand that behaves at small k as 1/(q · k)4 and is formally divergent. However, the divergence is in fact absent. (The threshold singularity at q 2 = 4m2 is, of course, present.) Nevertheless, threshold IR divergences do exist. For example, the sunset3 diagram of Fig. 2.3 with general masses at threshold, q 2 = (m1 + m2 + m3 )2 , is divergent in this sense when the sum of the integer powers of the propagators is greater than or equal to ﬁve (see, e.g. [11]).

Fig. 2.3. Sunset diagram

The IR divergences characterized above are local in momentum space, i.e. they are connected with special points of the loop integration momenta. Collinear divergences arise at lines parallel to certain light-like four-vectors. A typical example of a collinear divergence is provided by the massless triangle graph of Fig. 2.4. Let us take p21 = p22 = 0 and all the masses equal to zero. The corresponding Feynman integral is 3

called also the sunrise diagram, or the London transport diagram.

18

2 Feynman Integrals: Basic Deﬁnitions and Tools

Fig. 2.4. One-loop triangle diagram

(k 2

d4 k . − 2p1 ·k)(k 2 − 2p2 ·k)k 2

(2.20)

At least an on-shell IR divergence is present, because the integral is divergent when k → 0 (componentwise). However, there are also divergences at nonzero values of k that are collinear with p1 or p2 and where k 2 ∼ 0. This follows from the fact that the product 1/[(k 2 − 2p·k)k 2 ], where p2 = 0 and p = 0, generates collinear divergences. To see this let us take residues in the upper complex half plane when integrating this product over k0 . For example, taking the residue at k0 = −|k| + i0 leads to an integral containing 1/(p·k) = 1/[p0 |k|(1 − cos θ)], where θ is the angle between the spatial components k and p. Thus, for small θ, we have a divergent integration over angles because of the factor d cos θ/(1−cos θ) ∼ dθ/θ. The second residue generates a similar divergent behaviour – this can be seen by making the change k → p − k. Another way to reveal the collinear divergences is to introduce the lightcone coordinates k± = k0 ± k3 , k = (k1 , k2 ). If we choose p with the only non-zero component p+ , we shall see a logarithmic divergence coming from the region k− ∼ k 2 ∼ 0 just by power counting. These are the main types of divergences of usual Feynman integrals. Various special divergences arise in more general Feynman integrals (2.6) that can contain linear propagators and appear on the right-hand side of asymptotic expansions in momenta and masses and in associated eﬀective theories. For example, in the Sudakov limit, one encounters divergences that can be classiﬁed as UV collinear divergences. Another situation with various non-standard divergences is provided by threshold expansion and the corresponding eﬀective theories, NRQCD and pNRQCD, where special power counting is needed to characterize the divergences.

2.3 Alpha Representation A useful tool to analyse the divergences of Feynman integrals is the so-called alpha representation based on (2.12). It can be written down for any Feynman integral. For example, for (2.8), one inserts (2.12) for each of the two propagators, takes the four-dimensional Gaussian integral by means of (2.14) to obtain

2.3 Alpha Representation

FΓ (q) = iπ 2 0

∞

∞

0

× exp iq 2

19

dα1 dα2 (α1 + α2 )−2

α1 α2 − i(m21 α1 + m22 α2 ) . α1 + α2

(2.21)

For a usual general Feynman integral, this procedure can also explicitly be realized. Using (2.12) for each propagator of a general usual Feynman integral (i.e., with usual propagators (2.2)) one takes (see, e.g., [20]) 4hdimensional Gauss integrals by means of a generalization of (2.14) to the case of an arbitrary number of loop integration momenta: Aij ki ·kj + 2 qi ·ki d4 k1 . . . d4 kh exp i i,j

= i−h π 2h (det A)−2 exp −i

i

. A−1 ij qi ·qj

(2.22)

i,j

Here A is an h × h matrix and A−1 its inverse.4 The elements of the inverse matrix involved here are rewritten in graphtheoretical language (see details in [5, 20]), and the resulting alpha representation takes the form [6] i−a−h π 2h FΓ (q1 , . . . , qn ; d) = Γ (al ) ∞ l∞ 2 a −1 × dα1 . . . dαL αl l U −2 ZeiV/U −i ml αl , 0

0

where a = al , and U and V are the well-known functions αl , U= T ∈T 1 l∈T

V=

2 αl q T .

(2.23)

l

(2.24) (2.25)

T ∈T 2 l∈T

In (2.24), the sum runs over trees of the given graph, i.e. maximal connected subgraphs without loops, and, in (2.25), over 2-trees, i.e. subgraphs that do not involve loops and consist of two connectivity components; ±q T is the sum of the external momenta that ﬂow into one of the connectivity components of the 2-tree T . (It does not matter which component is taken because of the conservation law for the external momenta.) The products of the alpha parameters involved are taken over the lines that do not belong to the 4

In fact, the matrix A involved here equals eβe+ with the elements of an arbitrarily chosen column and row with the same number deleted. Here e is the incidence matrix of the graph, i.e. eil = ±1 if the vertex i is the beginning/end of the line l, e+ is its transpose and β consists of the numbers 1/αl on the diagonal – see, e.g., [20].

20

2 Feynman Integrals: Basic Deﬁnitions and Tools

given tree T . The functions U and V are homogeneous functions of the alpha parameters with the homogeneity degrees h and h + 1, respectively. The factor Z is responsible for the non-scalar structure of the diagram: 1 ∂ Z= Zl , (2.26) ei(2B−K)/U 2i ∂ul l

u1 =...uL =0

where (see, e.g., [27, 38]) B= ul qT αl , l

K=

T ∈Tl1

T ∈T 0 l∈T

αl

l ∈T

±ul

(2.27)

2 .

(2.28)

l

In (2.27), the sum is taken over trees Tl1 that include a given line l, and qT is the total external momentum that ﬂows through the line l (in the direction of its orientation). In (2.28), the sum is taken over pseudotrees T 0 (a pseudotree is obtained from a tree by adding a line), and the sum in l is performed over the loop (circuit) of the pseudotree T , with a sign dependent on the coincidence of the orientations of the line l and the pseudotree T . The alpha representation of a general h-loop Feynman integral is useful for general analyses. In practical calculations, e.g. at the two-loop level, one can derive the alpha representation for concrete diagrams by hand, rather than deduce it from the general formulae presented above. Still, even in practice, such general formulae can provide advantages because the evaluation of the functions of the alpha representation can be performed on a computer. Let us stress that this terrible-looking machinery for evaluating the determinant of the matrix A that arises from Feynman integrals, as well as for evaluating the elements of the inverse matrix, together with interpreting these results from the graph-theoretical point of view, is exactly the same as that used in the problem of the solution of Kirchhoﬀ’s laws for electrical circuits, a problem typical of the nineteenth century. Recall, for example, that the parameters αl play the role of ohmic resistances and that the expression (2.24) for the function U as a sum over trees is a Kirchhoﬀ result. Explicit formulae for Feynman integrals (2.6) with more general propagators which can be linear are not known. In this situation, one can derive alpha representation for any given concrete Feynman integral using formulae like (2.12) and performing Gaussian integration as in the case of Feynman integrals with standard propagators. We will follow this way in Chap. 3.

2.4 Regularization The standard way of dealing with divergent Feynman integrals is to introduce a regularization. This means that, instead of the original ill-deﬁned Feynman

2.4 Regularization

21

integral, we consider a quantity which depends on a regularization parameter, λ, and formally tends to the initial, meaningless expression when this parameter takes some limiting value, λ = λ0 . This new, regularized, quantity turns out to be well-deﬁned, and the divergence manifests itself as a singularity with respect to the regularization parameter. Experience tells us that this singularity can be of a power or logarithmic type, i.e. lnn (λ − λ0 )/(λ − λ0 )i . Although a regularization makes it possible to deal with divergent Feynman integrals, it does not actually remove UV divergences, because this operation is of an auxiliary character so that sooner or later it will be necessary to switch oﬀ the regularization. To provide ﬁniteness of physical observables evaluated through Feynman diagrams, another operation, called renormalization, is used. This operation is described, at the Lagrangian level, as a redeﬁnition of the bare parameters of a given Lagrangian by inserting counterterms. The renormalization at the diagrammatic level is called R-operation and removes the UV divergence from individual Feynman integrals. It is, however, beyond the scope of the present book. (See, however, some details in Sect. F.5, where the method of IR rearrangement is brieﬂy described.) An obvious way of regularizing Feynman integrals is to introduce a cutoﬀ at large values of the loop momenta. Another well-known regularization procedure is the Pauli–Villars regularization [24], which is described by the replacement 1 1 1 → 2 − 2 p 2 − m2 p − m2 p − M2 and its generalizations. For ﬁnite values of the regularization parameter M , this procedure clearly improves the UV asymptotics of the integrand. Here the limiting value of the regularization parameter is M = ∞. If we replace the integer powers al in the propagators by general complex numbers λl we obtain an analytically regularized [30] Feynman integral where the divergences of the diagram are encoded in the poles of this regularized quantity with respect to the analytic regularization parameters λl . For example, power counting at large values of the loop momentum inthe analytically ∞ regularized version of (2.8) leads to the divergent behaviour Λ dr rλ1 +λ2 −3 , which results in a pole 1/(λ1 + λ2 − 2) at the limiting values of the regularization parameters λl = 1. For example, in the case of the analytically regularized integral of Fig. 2.2, we obtain αλ1 −1 α2λ2 −1 e−iπ(λ1 +λ2 +1)/2 π 2 ∞ ∞ FΓ (q; λ1 , λ2 ) = dα1 dα2 1 Γ (λ1 )Γ (λ2 ) (α1 + α2 )2 0 0 α1 α2 2 2 2 − i(m1 α1 + m2 α2 ) . × exp iq (2.29) (α1 + α2 ) After the change of variables η = α1 + α2 , ξ = α1 /(α1 + α2 ) and explicit integration over η, we arrive at

22

2 Feynman Integrals: Basic Deﬁnitions and Tools

FΓ (q; λ1 , λ2 ) = eiπ(λ1 +λ2 ) ×

ξ λ1 −1 (1 − ξ)λ2 −1

1

dξ 0

iπ 2 Γ (λ1 + λ2 − 2) Γ (λ1 )Γ (λ2 ) λ1 +λ2 −2

[m21 ξ + m22 (1 − ξ) − q 2 ξ(1 − ξ) − i0]

.

(2.30)

Thus the UV divergence manifests itself through the ﬁrst pole of the gamma function Γ (λ1 + λ2 − 2) in (2.30), which results from the integration over small values of η due to the power η λ1 +λ2 −3 . The alpha representation turns out to be very useful for the introduction of dimensional regularization, which is a commonly accepted computational technique successfully applied in practice and which will serve as the main kind of regularization in this book. Let us imagine that the number of space– time dimensions diﬀers from four. To be more precise, the number of space dimensions is considered to be d − 1, rather than three. (But, of course, we still think of an integer number of dimensions!) The derivation of the alpha representation does not change much in this case. The only essential change is that, instead of (2.14), we need to apply its generalization to an arbitrary number of dimensions, d: 2 2 (2.31) dd k ei(αk −2q·k) = eiπ(1−d/2)/2 π d/2 α−d/2 e−iq /α . So, instead of (2.21), we have the following in d dimensions: ∞ ∞ FΓ (q; d) = e−iπ(1+d/2)/2 π d/2 dα1 dα2 (α1 + α2 )−d/2 0 0 2 α1 α2 × exp iq − i(m21 α1 + m22 α2 ) . α1 + α2

(2.32)

The only two places where something has been changed are the exponent of the combination (α1 + α2 ) in the integrand and the exponents of the overall factors. Now, in order to introduce dimensional regularization, we want to consider the dimension d as a complex number. So, by deﬁnition, the dimensionally regularized Feynman integral for Fig. 2.2 is given by (2.32) and is a function of q 2 as given by this integral representation. We choose d = 4−2ε, where the value ε = 0 corresponds to the physical number of the space–time dimensions. By the same change of variables as used after (2.29), we obtain ∞ −iπ(1+d/2)/2 d/2 FΓ (q; d) = e π dη η ε−1 ×

0 1

dξ exp iq 2 ξ(1 − ξ)η − i[m21 ξ + m22 (1 − ξ)]η .

(2.33)

0

This integral is absolutely convergent for 0 < Re ε < Λ (where Λ = ∞ if both masses are non-zero and Λ = 1 otherwise; this follows from an IR analysis of convergence, which we omit here) and deﬁnes an analytic function of ε, which

2.4 Regularization

23

is extended from this domain to the whole complex plane as a meromorphic function. After evaluating the integral over η, we arrive at the following result: 1 dξ (2.34) FΓ (q; d) = iπ d/2 Γ (ε) ε . 2 2 2 0 [m1 ξ + m2 (1 − ξ) − q ξ(1 − ξ) − i0] The UV divergence manifests itself through the ﬁrst pole of the gamma function Γ (ε) in (2.34), which results from the integration over small values of η in (2.33). This procedure of introducing dimensional regularization is easily generalized [6, 7, 8] to an arbitrary usual Feynman integral. Instead of (2.22), we use Aij ki ·kj + 2 qi ·ki dd k1 . . . dd kh exp i i,j

= eiπh(1−d/2)/2 π hd/2 (det A)−d/2 exp −i

i

, A−1 ij qi ·qj

(2.35)

i,j

and the resulting d-dimensional alpha representation takes the form [6, 7] eiπ[a+h(1−d/2)]/2 π hd/2 FΓ (q1 , . . . , qn ; d) = (−1)a l Γ (al ) ∞ ∞ 2 a −1 × dα1 . . . dαL αl l U −d/2 ZeiV/U −i ml αl . 0

0

(2.36)

l

Let us now deﬁne5 the dimensionally regularized Feynman integral by means of (2.36), treating the quantity d as a complex number. This is a function of kinematical invariants constructed from the external momenta and contained in the function V. In addition to this, we have to take care of polynomials in the external momenta and the auxiliary variables ul hidden in the factor Z. We treat these objects qi and ul , as well as the metric tensor gµν , as elements of an algebra of covariants, where we have, in particular, 5

An alternative deﬁnition of algebraic character [16, 32, 36] (see also [10]) exists and is based on certain axioms for integration in a space with non-integer dimension. It is unclear how to perform the analysis within such a deﬁnition, for example, how to apply the operations of taking a limit, diﬀerentiation, etc. to algebraically deﬁned Feynman integrals in d dimensions, in order to say something about the analytic properties with respect to momenta and masses and the parameter of dimensional regularization. After evaluating a Feynman integral according to the algebraic rules, one arrives at some concrete function of these parameters but, before integration, one is dealing with an abstract algebraic object. Let us remember, however, that, in practical calculations, one usually does not bother about precise deﬁnitions. From the purely pragmatic point of view, it is useless to think of a diagram when it is not calculated. On the other hand, from the pure theoretical and mathematical point of view, such a position is beneath criticism. ;-)

24

2 Feynman Integrals: Basic Deﬁnitions and Tools

∂ ∂uµl

uνl = gµν δl,l ,

gµµ = d .

This algebra also includes the γ-matrices with anticommutation relations γµ γν + γν γµ = 2gµν so that γ µ γµ = d, the tensor εκµνλ , etc. Thus the dimensionally regularized Feynman integrals are deﬁned as linear combinations of tensor monomials in the external momenta and other algebraic objects with coeﬃcients that are functions of the scalar products qi ·qj . However, this is not all, because we have to see that the α-integral is well-deﬁned. Remember that it can be divergent, for various reasons. The alpha representation is not only an important technique for evaluating Feynman integrals but also a very convenient tool for the analysis of their convergence. This analysis is outlined in Sect. E.1. It is based on decompositions of the alpha integral into so-called sectors where new variables are introduced in such a way that the integrand factorizes, i.e. takes the form of a product of some powers of the sector variables with a non-zero function. Eventually, in the new variables, the analysis of convergence reduces to power counting (for both UV and IR convergence) in one-dimensional integrals. As a result of this analysis, any Feynman integral considered at Euclidean external momenta qi , i.e. when any sum of incoming momenta is spacelike, is deﬁned as meromorphic function of d with series of UV and IR poles [7, 25, 27, 31, 33]. Here it is also assumed that there are no massless detachable subgraphs, i.e. massless subdiagrams with zero external momenta. For example, a tadpole, i.e. a line with coincident end points, is a detachable subgraph. However, such diagrams are naturally put to zero in case they are massless – see a discussion below. Unfortunately, there are no similar mathematical results for Feynman integrals on a mass shell or a threshold which are really needed in practice and which be mainly considered in this book. However, in every concrete example considered below, we shall see that every Feynman diagram is indeed an analytical function of d, both in intermediate steps of a calculation and, of course, in our results. Still it would be nice to have also a mathematical theorem on the convergence of general Feynman integrals. On the other hand, there is a practical algorithm [3] based on some sector decompositions that can provide the resolution of the singularities in ε for any given Feynman integral in the case where all the non-zero kinematical invariants have the same sign (and, possibly, are on a mass shell or at a threshold). This algorithm is described in Sect. E.2.

2.5 Properties of Dimensionally Regularized Feynman Integrals We can formally write down dimensionally regularized Feynman integrals as integrals over d-dimensional vectors ki :

2.5 Properties of Dimensionally Regularized Feynman Integrals

FΓ (q1 , . . . , qn ; d) =

dd k1 . . .

dd kh

L

˜ F,l (pl ) . D

25

(2.37)

l=1

In order to obtain dimensionally regularized integrals with their dimension independent of ε, a factor of µ−2ε per loop, where µ is a massive parameter, is introduced. This parameter serves as a renormalization parameter for schemes based on dimensional regularization. Therefore, we obtain logarithms and other functions depending not only on ratios of given parameters, e.g. q 2 /m2 , but also on q 2 /µ2 etc. However, we shall usually omit this µ-dependence for brevity (i.e. set µ = 1) so that you will meet sometimes quantities like ln q 2 which should be understood in the sense of ln(q 2 /µ2 ). We have reasons for using the notation (2.37), because dimensionally regularized Feynman integrals as deﬁned above possess the standard properties of integrals of the usual type in integer dimensions. In particular, – the integral of a linear combination of integrands equals the same linear combination of the corresponding integrals; – one may cancel the same factors in the numerator and denominator of integrands. These properties follow directly from the above deﬁnition. A less trivial property is that – a derivative of an integral with respect to a mass or momentum equals the corresponding integral of the derivative. This is also a consequence (see [8, 27]) of the deﬁnition of dimensionally regularized Feynman integrals based on the alpha representation and the corresponding analysis of convergence presented in Sect. E.1. To prove this statement, one uses standard algebraic relations between the functions entering the alpha representation [7, 20]. (We note again that these are relations quite similar to those encoded in the solutions of Kirchhoﬀ’s laws for a circuit deﬁned by the given graph.) A corollary of the last property is the possibility of integrating by parts and always neglecting surface terms: L ∂ ˜ d d DF,l (pl ) = 0 , i = 1, . . . , h . (2.38) – d k1 . . . d kh µ ∂ki l=1

This property is the basis for solving the reduction problem for Feynman integrals using IBP relations [9] – see Chaps. 5 and 6. The next property says that – any diagram with a detachable massless subgraph is zero. This property can also be shown to be a consequence of the accepted deﬁnition [8, 27], by use of an auxiliary analytic regularization, using pieces of the α-integral considered in diﬀerent domains of the regularization parameters. Let us consider, for example, the massless tadpole diagram, which

26

2 Feynman Integrals: Basic Deﬁnitions and Tools

can be reduced by means of alpha parameters to a scaleless one-dimensional integral: ∞ d d k ε d/2 = −i π dα αε−2 . (2.39) k2 0 We divide this integral into two pieces, from 0 to 1 and from 1 to ∞, integrate these two integrals and ﬁnd results that are equal except for opposite signs, which lead to the zero value.6 It should be stressed here that the two pieces that contribute to the right-hand side of (2.39) are convergent in diﬀerent domains of the regularization parameter ε, namely, Re ε > −1 and Re ε < −1, with no intersection, and that this procedure here is equivalent to introducing analytic regularization and considering its parameter in diﬀerent domains for diﬀerent pieces. But let us distinguish between two qualitatively diﬀerent situations: the ﬁrst when we have to deal with a massless Feynman integral, with a zero external momentum, which arises from the Feynman rules, and the second when we obtain such scaleless integrals after some manipulations: after using partial fractions, diﬀerentiation, integration by parts, etc. We can also include in this second class all such integrals that appear on the right-hand side of explicit formulae for (oﬀ-shell) asymptotic expansions in momenta and masses [2, 29]. In the ﬁrst situation, the only possibility is to use the ad hoc prescription of setting the integral to zero. In the second situation, we can start with an alpha representation, introduce an auxiliary analytic regularization [8, 27] and use the fact that it is convergent in some non-empty domain of these parameters (see Sect. E.1). A very important point here is that all the properties of dimensionally regularized integrals given above, apart from the last one, can be justiﬁed in a purely algebraic way [8, 27], through identities between functions in the alpha representation. Then, using sector decompositions described in Sect. E.1, with a control over convergence at hand, one can see that all the resulting massless Feynman integrals with zero external momenta indeed vanish – see details in [8, 27]. Let us now remind ourselves of reality and observe that it is necessary to deal in practice with diagrams on a mass shell or at a threshold. What about the properties of dimensionally regularized Feynman integrals in this case? At least the algebraic proof of the basic properties of dimensionally regularized Feynman integrals is not sensitive to putting the external momenta in any particular place. However, as we noticed above, a general analysis of the convergence of such integrals, even in speciﬁc cases, is still absent, so that we do not have control over convergence. Technically, this means that the sectors used for the analysis of the convergence in the oﬀ-shell case are no longer suﬃcient for the resolution of the singularities of the integrand of the alpha 6

These arguments can be found, for example, in [17], and even in a pure mathematical book [13]. Well, let us not take the latter example seriously ;-)

2.5 Properties of Dimensionally Regularized Feynman Integrals

27

representation. These singularities are much more complicated and can even appear (e.g. at a threshold) at non-zero, ﬁnite values of the α-parameters. However, the good news is that numerous practical applications have shown that there is no sign of breakdown of these properties for on-shell or threshold Feynman integrals. Although on-shell and threshold Feynman integrals have been already mentioned many times, let us now be more precise in our deﬁnitions. We must realize that, generally, an on-shell or threshold Feynman integral is not the value of the given Feynman integral FΓ (q 2 , . . .), deﬁned as a function of q 2 and other kinematical variables, at a value of q 2 on a mass shell or at a threshold. Consider, for example, the Feynman integral corresponding to Fig. 2.2, with m1 = 0, m2 = m, a1 = 1, a2 = 2. We know an explicit result for the diagram given by (1.5). There is a logarithmic singularity at threshold, q 2 = m2 , so that we cannot strictly speak about the value of the integral there. Still we can certainly deﬁne the threshold Feynman integral by putting q 2 = m2 in the integrand of the integral over the loop momentum or over the alpha parameters. And this is what was really meant and will be meant by ‘on-shell’ and ‘threshold’ integrals. In this example, we obtain an integral which can be evaluated by means of (A.13) (to be derived in Chap. 3): Γ (ε) dd k = iπ d/2 . (2.40) 2 2 k (k − 2q·k)2 2(m2 )1+ε This integral is divergent, in contrast to the original Feynman integral deﬁned for general q 2 . Thus on-shell or threshold dimensionally regularized Feynman integrals are deﬁned by the alpha representation or by integrals over the loop momenta with restriction of some kinematical invariants to appropriate values in the corresponding integrands. In this sense, these regularized integrals are ‘formal’ values of general Feynman integrals at the chosen variables. Note that the products of the free ﬁelds in the Lagrangian are not required to be normal-ordered, so that products of ﬁelds of the same sort at the same point are allowed. The formal application of the Wick theorem therefore generates values of the propagators at zero. For example, in the case of the scalar free ﬁeld, with the propagator i e−ix·k 4 k , (2.41) d DF (x) = (2π)4 k 2 − m2 which satisﬁes (2 + m2 )DF (x) = −iδ(x), we have T φ(x)φ(x) = : φ2 (x) : +DF (0) .

(2.42)

The value of DF (x) at x = 0 does not exist, because the propagator is singular at the origin according to (2.16). However, we imply the formal value at the origin rather than the ‘honestly’ taken value. This means that we set x to zero in some integral representation of this quantity. For example, using the

28

2 Feynman Integrals: Basic Deﬁnitions and Tools

inverse Fourier transformation, we can deﬁne DF (0) as the integral (2.41) with x set to zero in the integrand. Thus, by deﬁnition, i d4 k DF (0) = . (2.43) 4 2 (2π) k − m2 This integral is, however, quadratically divergent, as Feynman integrals typically are. So, we understand DF (0) as a dimensionally regularized formal value when we put x = 0 in the Fourier integral and obtain, using (A.1) (which we will derive shortly), dd k = −iπ d/2 Γ (ε − 1)(m2 )1−ε . (2.44) 2 k − m2 This Feynman integral in fact corresponds to the tadpole φ4 theory graph shown in Fig. 2.5. The corresponding quadratic divergence manifests itself through an UV pole in ε – see (2.44).

Fig. 2.5. Tadpole

Observe that one can trace the derivation of the integrals tabulated in Sect. A.1 and see that the integrals are convergent in some non-empty domains of the complex parameters λl and ε and that the results are analytic functions of these parameters with UV, IR and collinear poles. Before continuing our discussion of setting scaleless integrals to zero, let us present an analytic result for the one-loop massless triangle integral with two on-shell external momenta, p21 = p22 = 0. Using (A.28) (which we will derive in Chap. 3), we obtain 2 dd k d/2 Γ (1 + ε)Γ (−ε) = −iπ . (2.45) (k 2 − 2p1 ·k)(k 2 − 2p2 ·k)k 2 Γ (1 − 2ε)(−q 2 )1+ε A double pole at ε = 0 arises from the IR and collinear divergences. A similar formula with a monomial in the numerator can be obtained also straightforwardly: µ µ 2 dd k k µ d/2 Γ (ε)Γ (1 − ε) p1 + p2 = iπ . (k 2 − 2p1 ·k)(k 2 − 2p2 ·k)k 2 Γ (2 − 2ε) (−q 2 )1+ε (2.46) Now only a simple pole is present, because the factor k µ kills the IR divergence. Consider now a massless one-loop integral with the external momentum on the massless mass shell, p2 = 0:

References

dd k . (p − k)2 k 2

29

(2.47)

If we write down the alpha representation for this integral we obtain the same expression (2.39) as for p = 0 because only p2 , equal to zero in both cases, is involved there. In spite of this obvious fact, there is still a qualitative diﬀerence: for p = 0, there are UV and IR poles which enter with opposite signs and, for p2 = 0 (but with p = 0 as a d-dimensional vector), there is a similar interplay of UV and collinear poles. Now we follow the arguments presented in [21] and write down the following identity for (2.47), with p = p1 : dd k 2 (k − 2p1 ·k)k 2 dd k 2p2 ·k dd k − = , (k 2 − 2p1 ·k)(k 2 − 2p2 ·k) (k 2 − 2p1 ·k)(k 2 − 2p2 ·k)k 2 (2.48) where p22 = 0 and p1·p2 = 0. We then evaluate the integrals on the right-hand side by means of (A.7) and (2.46), respectively, and obtain a zero value. This fact again exempliﬁes the consistency of our rules. Thus we are going to systematically apply the properties of dimensionally regularized Feynman integrals in any situation, no matter where the external momenta are considered to be. Moreover, we will believe that these properties are also valid for more general Feynman integrals given by the dimensionally regularized version of (2.6) which can contain linear propagators. Let us also point out that the rule to put all scaleless integrals to zero is rather general and, as far as I know, never causes contradictions. In particular, it is applied in asymptotic expansions of Feynman integrals in various limits of momenta and masses within expansion by regions [2, 29], where such integrals are always put to zero, even if they are not regulated by dimensional regularization. We will follow this rule also in Chap. 6 where we will put to zero scaleless integrals which appear in auxiliary parametric representations when constructing coeﬃcient functions at master integrals.

References 1. S. Actis, A. Ferroglia, G. Passarino, M. Passera and S. Uccirati, hepph/0402132. 12 2. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 13, 26, 29 3. T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 24 4. G.T. Bodwin, E. Braaten and G.P. Lepage, Phys. Rev. D 51 (1995) 1125; Phys. Rev. D 55 (1997) 5853. 14 5. N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 19

30 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38.

2 Feynman Integrals: Basic Deﬁnitions and Tools C.G. Bollini and J.J. Giambiagi, Nuovo Cim. B 12 (1972) 20. 19, 23 P. Breitenlohner and D. Maison, Commun. Math. Phys. 52 (1977) 11, 39, 55. 23, 24, 25 K.G. Chetyrkin and V.A. Smirnov, Teor. Mat. Fiz. 56 (1983) 206. 16, 23, 25, 26 K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 25 J.C. Collins, Renormalization (Cambridge University Press, Cambridge, 1984). 23 A.I. Davydychev and V.A. Smirnov, Nucl. Phys. B 554 (1999) 391. 17 A. Erd´elyi (ed.), Higher Transcendental Functions, Vols. 1 and 2 (McGraw-Hill, New York, 1954). 16 I.M. Gel’fand and G.E. Shilov, Generalized Functions, Vol. 1 (Academic Press, New York, London, 1964). 26 A.G. Grozin, Heavy Quark Eﬀective Theory (Springer, Berlin, Heidelberg, 2004). 14 K. Hepp, Commun. Math. Phys. 2 (1966) 301. G. ’t Hooft and M. Veltman, Nucl. Phys. B 44 (1972) 189. 23 G. Leibbrandt, Rev. Mod. Phys. 47 (1975) 849. 26 G.P. Lepage et al., Phys. Rev. D 46 (1992) 4052. 14 A.V. Manohar and M.B. Wise, Heavy Quark Physics (Cambridge University Press, Cambridge 2000). 14 N. Nakanishi, Graph Theory and Feynman Integrals (Gordon and Breach, New York, 1971). 19, 25 W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 29 M. Neubert, Phys. Rep. 245 (1994) 259. 14 G. Passarino and M. Veltman, Nucl. Phys. B 160 (1979) 151. 12 W. Pauli and F. Villars, Rev. Mod. Phys. 21 (1949) 434. 21 K. Pohlmeyer, J. Math. Phys. 23 (1982) 2511. 24 A.P. Prudnikov, Yu.A. Brychkov and O.I. Marichev, Integrals and Series, Vols. 1–3 (Gordon and Breach, New York, 1986–1990). 16 V.A. Smirnov, Renormalization and Asymptotic Expansions (Birkh¨ auser, Basel, 1991). 16, 20, 24, 25, 26 V.A. Smirnov, Phys. Lett. B 491 (2000) 130; Phys. Lett. B 500 (2001) 330. V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 13, 17, 26, 29 E.R. Speer, J. Math. Phys. 9 (1968) 1404. 21 E.R. Speer, Commun. Math. Phys. 23 (1971) 23; Commun. Math. Phys. 25 (1972) 336. 24 E.R. Speer, in Renormalization Theory, eds. G. Velo and A.S. Wightman (Reidel, Dodrecht, 1976) p. 25. 23 E.R. Speer, Ann. Inst. H. Poincar´e 23 (1977) 1. J.B. Tausk, Phys. Lett. B 469 (1999) 225. 24 B.A. Thacker and G.P. Lepage, Phys. Rev. D 43 (1991) 196. K.G. Wilson, Phys. Rev. D 7 (1973) 2911. 14 S. Wolfram, The Mathematica Book, 4th edition (Wolfram Media and Cambridge University Press, Cambridge, 1999). 23 O.I. Zavialov, Renormalized Quantum Field Theory (Kluwer Academic Publishers, Dodrecht, 1990). 16 20

3 Evaluating by Alpha and Feynman Parameters

Feynman parameters1 are very well known and often used in practical calculations. They are closely related to alpha parameters introduced in Chap. 2 so that we shall study both kinds of parametric representations of Feynman integrals in one chapter. The use of these parameters enables us to transform Feynman integrals over loop momenta into parametric integrals where Lorentz invariance becomes manifest. Using alpha parameters we shall ﬁrst evaluate one and two-loop integrals with general complex powers of the propagators, within dimensional regularization, for which results can be written in terms of gamma functions for general values of the dimensional regularization parameter. We shall show then how these formulae, together with simple algebraic manipulations, enable us to evaluate some classes of Feynman integrals. We then turn to various characteristic one-loop examples where results cannot be written in terms of gamma functions. In such situations, we shall be usually oriented at the evaluation in expansion in powers of ε up to some ﬁxed order. We then introduce Feynman parameters and present the so-called Cheng–Wu theorem which provides a very useful trick that can greatly simplify the evaluation. Finally, we proceed at the two-loop level by presenting rather complicated examples of evaluating Feynman integrals by Feynman and alpha parameters.

3.1 Simple One- and Two-Loop Formulae A lot of one- and two-loop formulae can be derived, using alpha and Feynman parameters, for general complex indices with results expressed in terms of gamma functions. A collection of such formulae is presented in Sect. A.1. Let us evaluate, for example, the dimensionally regularized massive tadpole Feynman diagram of Fig. 2.5 with a general power of the propagator, dd k . (3.1) FΓ (q; λ, d) = (−k 2 + m2 )λ We apply the alpha representation of the analytically regularized scalar propagator given by (2.12) with Z = 1, i.e. 1

See, e.g., textbooks [18] and [7].

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 31–53 (2004) c Springer-Verlag Berlin Heidelberg 2004

32

3 Evaluating by Alpha and Feynman Parameters

1 iλ = (−k 2 + m2 )λ Γ (λ)

∞

dα αλ−1 ei(k

2

−m2 )α

,

(3.2)

0

change the order of integration over k and α, take the Gaussian k integral by means of (2.31), again apply (3.2) written in the reverse order, i.e. ∞ Γ (λ) i−λ dα αλ−1 e−iAα = , (3.3) (A − i0)λ 0 and arrive at (A.1). In particular, this table formula gives (2.44). Let us now turn to the dimensionally regularized Feynman diagram of Fig. 2.2 with general powers of the propagators, dd k . (3.4) FΓ (q; λ1 , λ2 , d) = (−k 2 + m21 )λ1 [−(q − k)2 + m22 ]λ2 From now on, we shall use the following convention: when powers of propagators are integers we use them with +k 2 +i0, but when they are non-integral or complex, we take the opposite sign, i.e. −k 2 −i0. The second choice is more natural if we wish to obtain a Euclidean, −q 2 , dependence of the results (see, e.g., (3.6) below). We shall also prefer to use al for integer and λl for general complex indices. In the latter case, the alpha representation is obtained from (2.36) by replacing al by λl and dropping out the factor (−1)a . Starting from the alpha representation of Fig. 2.2, with the basic functions U = α1 +α2 and V = α1 α2 q 2 , and using the change of variables α1 = ξη, α2 = η(1 − ξ) we obtain the dimensionally regularized version of (2.30), i.e. FΓ (q; λ1 , λ2 , d) = iπ d/2 × 0

1

Γ (λ1 + λ2 + ε − 2) Γ (λ1 )Γ (λ2 ) dξ ξ λ1 −1 (1 − ξ)λ2 −1

[m21 ξ + m22 (1 − ξ) − q 2 ξ(1 − ξ) − i0]

λ1 +λ2 +ε−2

.

(3.5)

Suppose that the masses are zero. In this case the integral over ξ can be evaluated in terms of gamma functions, and we arrive at the following result: dd k G(λ1 , λ2 ) = iπ d/2 , (3.6) (−k 2 )λ1 [−(q − k)2 ]λ2 (−q 2 )λ1 +λ2 +ε−2 where G(λ1 , λ2 ) =

Γ (λ1 + λ2 + ε − 2)Γ (2 − ε − λ1 )Γ (2 − ε − λ2 ) . Γ (λ1 )Γ (λ2 )Γ (4 − λ1 − λ2 − 2ε)

(3.7)

The one-loop formula (3.6) can graphically be described by Fig. 3.1. In the case where the powers of propagators are equal to one, we have dd k Γ (ε)Γ (1 − ε)2 = iπ d/2 . (3.8) 2 2 k (q − k) Γ (2 − 2ε)(−q 2 )ε Note that although the indices of the diagrams are integral at the beginning, non-integral indices shifted by amounts proportional to ε appear after intermediate integration, e.g. after the use of (3.8) inside a bigger diagram.

3.1 Simple One- and Two-Loop Formulae λ2

= iπ d/2 G(λ1 , λ2 )×

33

λ1 + λ2 − d/2

λ1 Fig. 3.1. Graphical interpretation of (3.6)

Another formula that can be derived from (3.5) gives a result for the integral dd k . (−k 2 + m2 )λ1 (−k 2 )λ2 Indeed, we set q = 0, m1 = m and m2 = 0, take an integral over ξ and obtain (A.4). Consider now the following integral that arises in calculations in Heavy Quark Eﬀective Theory [12, 15, 17]: dd k . (−k 2 )λ1 (2v·k + ω − i0)λ2 Since the denominator of one of the propagators is not quadratic we cannot use the general formula of the alpha representation. Still we proceed by alpha parameters, i.e. apply (3.2) to the ﬁrst propagator and a similar Fourier representation ∞ iλ 1 = dα αλ−1 eiAα , (3.9) (−A − i0)λ Γ (λ) 0 with A = −2v · k − ω, to the second propagator. Changing the order of integration as above and evaluating a Gaussian integral over k we then apply 2 2 (3.3) to take the integral of α1λ1 +ε−3 e−iα2 v /α1 over α1 and, ﬁnally, an integral over α2 , and arrive at (A.25). The following one-loop integral is typical for the evaluation of the one-loop quark potential: dd k . 2 λ 1 (−k ) [−(q − k)2 ]λ2 (−2v·k − i0)λ3 Here v · q = 0. (Typically, one chooses q = (0, q) and v = (1, 0).) One of the propagators is again not quadratic so that we proceed by alpha parameters and represent each of the three factors as an alpha integral. After taking a Gaussian integral over k we obtain

3 iλ1 +λ2 +λ3 +ε−1 π d/2 ∞ ∞ ∞ λl −1 αl dαl (α1 + α2 )ε−2 0 0 0 l Γ (λl ) l=1 2 q α1 α2 − v 2 α32 × exp i . α1 + α2

34

3 Evaluating by Alpha and Feynman Parameters

√ Then the integral over α3 can be evaluated by the change α3 = t and (3.3). After that the integration over α1 and α2 is taken, as before, by introducing the variables η = α1 + α2 , ξ = α1 /(α1 + α2 ), with the result (A.27). Using alpha parameters one can also derive the formula (A.40) for the formal Fourier transformation within dimensional regularization. This formula provides another way to derive (3.6). In fact, the initial integral is nothing but the convolution of the two functions, f˜i = 1/(−k 2 − i0)λi , i = 1, 2. Then one uses the well-known mathematical formula f˜1 ∗ f˜2 (q) = (2π)d (f1˜f2 ) for the convolution of two Fourier transforms, applies (A.40) and arrives at (3.6).

3.2 Auxiliary Tricks 3.2.1 Recursively One-Loop Feynman Integrals Massless integrals are often evaluated with the help of successive application of the one-loop formula (3.6). In addition one can use the fact that a sequence of two lines with scalar propagators with the same mass and the indices a1 and a2 can be replaced by one line with index a1 + a2 . Consider, for example, the two-loop diagram shown in Fig. 3.2. The internal one-loop integral can be evaluated by use of (3.8) and is eﬀectively replaced, according to Fig. 3.1, by a line with index ε. Then the sequence of two massless lines with indices 1 and ε is replaced by one line with index 1 + ε, and the one-loop diagram so obtained, which has indices 2 and 1 + ε, is evaluated by means of the oneloop formula (3.6), with the following result expressed in terms of gamma functions: G(1, 1)G(2, 1 + ε)/(−q 2 )1+2ε . The class of Feynman diagrams that can be evaluated in this way by means of (3.6) can be called recursively one-loop.

Fig. 3.2. A recursively one-loop diagram

Another example where two tabulated one-loop integration formulae can successively be applied is given by the two-loop scalar diagram of Fig. 3.3 with general complex indices and two zero masses, dd k dd l . (−k 2 )λ1 [−(k + l)2 ]λ2 (m2 − l2 )λ3

3.2 Auxiliary Tricks

35

1 2 3 Fig. 3.3. Vacuum two-loop diagram with the masses 0, 0 and m

Here one can ﬁrst apply the one-loop massless integration formula (3.6), then apply (A.4) and obtain (A.39). 3.2.2 Partial Fractions When evaluating dimensionally regularized Feynman integrals one uses their properties, in particular the possibility of manipulations based on the properties listed in Sect. 2.5. Here the following standard decomposition proves to be useful: a 1 −1 a2 − 1 + i (−1)i 1 = (x + x1 )a1 (x + x2 )a2 (x2 − x1 )a2 +i (x + x1 )a1 −i a2 − 1 i=0 a 2 −1 a1 − 1 + i (−1)a1 + , (3.10) (x2 − x1 )a1 +i (x + x2 )a2 −i a1 − 1 i=0 where a1 , a2 > 0 and n n! = j j!(n − j)! is a binomial coeﬃcient. For example, the vacuum one-loop Feynman integral with two diﬀerent masses, dd k , (k 2 − m21 )(k 2 − m22 ) can be evaluated by (3.10) and (A.1), with the result iπ d/2 Γ (ε − 1)

− m2−2ε m2−2ε 2 1 . m21 − m22

If one of the indices, e.g. a2 is non-positive, a similar decomposition is performed by expanding (x + x2 )−a2 in powers of x + x1 . Let us note that if one proceeds by MATHEMATICA [22], one can use, for given integer values of a1 and a2 , the command Apart to perform partial fractions decompositions.

36

3 Evaluating by Alpha and Feynman Parameters

3.2.3 Dealing with Numerators As we have agreed we suppose that a tensor reduction for a given class of Feynman integrals was performed so that we start with evaluating scalar integrals. Let us, however, mention that one can also evaluate integrals with Lorentz indices. A lot of one-loop Feynman integrals with numerators can be found in Sect. A.1. One can reduce evaluating such a one-loop integral to an integral with a product k α1 . . . k αN . Then one can switch to traceless monomials and back using (A.41a) and (A.41b). An integral with a traceless monomial independent of other Lorentz indices is again traceless. If it depends on one external momentum it should be proportional to its traceless monomial. This is how tabulated integrals for traceless monomials, e.g. (A.8), can be derived. Then one can turn back to usual monomials using (A.41b). (In Sect. A.2, one can ﬁnd also other useful formulae for various traceless monomials.) In the case of a general h-loop Feynman integral with standard propagators, let us observe that the function (2.26) in (2.36) can be taken into account by shifting the space–time dimension d and indices al of a given diagram because any factor that arises after the diﬀerentiation with respect to the auxiliary parameters ul is a sum of products of positive integer powers of the α-parameters and negative integer powers of the function U. In particular, the factor 1/U n is taken into account by the shift d → d + 2n. Then the shift of a power of a parameter αl can be translated into a shift of the power of the corresponding propagator, in particular, a multiplication by αl can be described by the operator ial l+ where l+ increases the index al by one, the multiplication by αl2 can be described by the operator −al (al + 1)l++ , etc. This observation enables us to express any given Feynman integral with numerators through a linear combination of scalar integrals with shifted indices and shifted dimensions. Systematic algorithms oriented towards realization on a computer, with a demonstration up to two-loop level, have been constructed in [20]. We shall come back to this point in Chap. 5 when solving IBP recurrence relations. At the one-loop level, this property has been used [9] to derive a general formula for the Feynman integrals kα1 . . . kαn ) (λ , . . . , λ , d) = dd k N , (3.11) Fα(N 1 N 1 ...αn 2 λi 2 i=1 [−(qi − k) + mi ] depending on the external momenta q1 − q2 , . . . , qN − q1 and the general masses mi : (−1)r ) Fα(N (λ1 , . . . , λN , d) = 1 ...αn 2r r,κ1 ,...,κN : 2r+

×{{[g] [q1 ] r

κ1

κN

. . . [qN ]

κi =n

}α1 ...αn

N

(λi )κi

i=1

3.2 Auxiliary Tricks

37

p1 a1 a3 a2 p2

Fig. 3.4. Triangle diagram with the masses 0, 0, m, external momenta p21 = p22 = 0 and general indices of the propagators

×F (N ) (λ1 + κ1 , . . . , λN + κN , d + 2(n − r)) ,

(3.12)

where {[g] [q1 ] . . . [qN ] }α1 ...αn is symmetric in its indices and is composed of the metric tensor and the vectors qi . Tabulated formulae with numerators presented in Appendix A can be derived by means of (3.12). Let us now present a simple one-loop example and illustrate the trick with turning to integrals without numerators. Consider the Feynman integral corresponding to Fig. 3.4 with a numerator F (q 2 , m2 ; a1 , a2 , a3 , n, d) dd k (l·k)n = , (3.13) 2 a (k − 2p1 ·k) 1 (k 2 − 2p2 ·k)a2 (k 2 − m2 )a3 where l is a momentum not related to p1 and p2 . The alpha representation (2.36) takes the form ia1 +a2 +a3 +ε−1 π d/2 F (q 2 , m2 ; a1 , a2 , a3 , n, d) = (−1)a l Γ (al ) ∞ ∞ ∞ a −1 × dα1 dα2 dα3 αl l U −d/2 exp iV/U − im2 α3 r

κ1

0

×

κN

0

1 ∂ 2i ∂r

n

0

l

exp

! i[2rl·(α1 p1 + α2 p2 ) + r2 l2 ] , α1 + α2 + α3 r=0

(3.14)

where U = α1 + α2 + α3 ,

V = q 2 α1 α2 .

Taking into account the arguments above we see, for example, that 1 F (a1 , a2 , a3 , 1, d) = − [a1 l·p1 F (a1 + 1, a2 , a3 , 0, d + 2) π +a2 l·p2 F (a1 , a2 + 1, a3 , 0, d + 2)] , l2 F (a1 , a2 , a3 , 0, d + 2) F (a1 , a2 , a3 , 2, d) = 2π 1

+ 2 a1 (a1 + 1)(l·p1 )2 F (a1 + 2, a2 , a3 , 0, d + 4) π +2a1 a2 (l·p1 )(l·p2 )F (a1 + 1, a2 + 1, a3 , 0, d + 4) +a2 (a2 + 1)(l·p2 )2 F (a1 , a2 + 2, a3 , 0, d + 4) .

(3.15)

(3.16)

38

3 Evaluating by Alpha and Feynman Parameters

Such a reduction of numerators can be performed for any Feynman integral. The corresponding algebraic manipulations can easily be implemented on a computer.

3.3 One-Loop Examples Let us present examples of evaluation of Feynman diagrams by means of alpha parameters with results which are not written in terms of gamma functions for general d. We ﬁrst turn to the example considered in the introduction. Example 3.1. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. We apply (3.5) to obtain F3.1 (q 2 , m2 ; a1 , a2 , d) = iπ d/2 (−1)a1 +a2 × 0

For example, we have

1

Γ (a1 + a2 + ε − 2) Γ (a1 )Γ (a2 )

dξ ξ a2 −1 (1 − ξ)1−a2 −ε [m2 − q 2 ξ − i0]

a1 +a2 +ε−2

.

(3.17)

dd k (k 2 − m2 )2 (q − k)2 1 (1 − ξ)−ε dξ d/2 = −iπ Γ (1 + ε) 1+ε . 2 2 0 [m − q ξ − i0]

F3.1 (q , m ; 2, 1, d) ≡ 2

2

(3.18)

Suppose that we are interested only in the value of this (ﬁnite) integral exactly in four dimensions. The integral over ξ is then evaluated easily at ε = 0 with the result (1.5). Similarly, Feynman integrals corresponding to Fig. 1.1 with various integer indices ai can be evaluated. In particular, we obtain (1.7). The next one-loop example is Example 3.2. The triangle diagram of Fig. 3.4. The Feynman integral for Fig. 3.4 with general integer indices looks like (3.13) with n = 0, i.e. F3.2 (q 2 , m2 ; a1 , a2 , a3 , d) dd k = , 2 a 2 (k − 2p1 ·k) 1 (k − 2p2 ·k)a2 (k 2 − m2 )a3

(3.19)

where q = p1 − p2 , q 2 ≡ −Q2 = −2p1 ·p2 . The alpha representation (2.36) takes the form (3.14) with n = 0. Introducing variables α1 = ξ1 η, α2 = ξ2 η and α3 = (1 − ξ1 − ξ2 )η and integrating over η we obtain

3.3 One-Loop Examples

F3.2 (q 2 , m2 ; a1 , a2 , a3 , d) = ×

1

dξ1 0

iπ d/2 (−1)a1 +a2 +a3 Γ (a + ε − 2) l Γ (al )

1−ξ1

dξ2 0

39

ξ1a1 −1 ξ2a2 −1 (1 − ξ1 − ξ2 )a3 −1 . [Q2 ξ1 ξ2 + m2 (1 − ξ1 − ξ2 )]a+ε−2

(3.20)

This can be a reasonable starting point for the evaluation of integrals with any given indices ai . Let us evaluate the integral with a1 = a2 = a3 = 1 at d = 4. Then the integral is ﬁnite: 1 1−ξ1 dξ2 . F3.2 (q 2 , m2 ; 1, 1, 1, 4) = −iπ 2 dξ1 2 Q ξ1 ξ2 + m2 (1 − ξ1 − ξ2 ) 0 0 A straightforward integration gives the following result: F3.2 (q 2 , m2 ; 1, 1, 1, 4) iπ 2 1 2 π2 = 2 Li2 (x) − ln x + ln x ln(1 − x) − , Q 2 3

(3.21)

where Li2 (x) is the dilogarithm (see (B.7)) and x = m2 /Q2 . Example 3.3. The massless on-shell box diagram of Fig. 3.5, i.e. with p2i = 0, i = 1, 2, 3, 4.

p1 2

p2

p3

1 4 3

p4

Fig. 3.5. Box diagram

With the loop momentum chosen as the momentum of line 1, the Feynman integral takes the form F3.3 (s, t; a1 , a2 , a3 , a4 , d) dd k = , 2 a 2 a (k ) 1 [(k + p1 ) ] 2 [(k + p1 + p2 )2 ]a3 [(k − p3 )2 ]a4

(3.22)

where s = (p1 + p2 )2 and t = (p1 + p3 )2 are Mandelstam variables. The trees and 2-trees relevant to the functions U and V are shown in Figs. 3.6 and 3.7. Four more existing 2-trees, for example the 2-tree with the component consisting of the lines 1 and 2 and the component consisting of the isolated vertex with the external momentum p4 , do not contribute to the function V because the product α3 α4 is multiplied by the corresponding external momentum squared which is zero. We have (2.36) with

40

3 Evaluating by Alpha and Feynman Parameters

Fig. 3.6. Trees contributing to the function U for the box diagram

Fig. 3.7. 2-trees contributing to the function V for the massless on-shell box diagram

U = α1 + α2 + α3 + α4 , V = tα1 α3 + sα2 α4 .

(3.23)

Introducing new variables by α1 = η1 ξ1 , α2 = η1 (1 − ξ1 ), α3 = η2 ξ2 , α4 = η2 (1 − ξ2 ), with the Jacobian η1 η2 , and evaluating an integral over η2 due to the delta function and an integral over η1 in terms of gamma functions we obtain F3.3 (s, t; a1 , a2 , a3 , a4 , d) Γ (a + ε − 2)Γ (2 − ε − a1 − a2 )Γ (2 − ε − a3 − a4 ) = (−1)a iπ d/2 Γ (4 − 2ε − a) Γ (al ) 1 1 ξ a1 −1 (1 − ξ1 )a2 −1 ξ2a3 −1 (1 − ξ2 )a4 −1 × dξ1 dξ2 1 . (3.24) [−sξ1 ξ2 − t(1 − ξ1 )(1 − ξ2 ) − i0]a+ε−2 0 0 where a = a1 + a2 + a3 + a4 . Consider, for example, the master integral2 with all the indices equal to one. We have F (s, t; d) ≡ F3.3 (s, t; 1, 1, 1, 1, d) = iπ d/2

1

× 0

0

1

Γ (2 + ε)Γ (−ε)2 Γ (−2ε)

dξ1 dξ2 . [−tξ1 ξ2 − s(1 − ξ1 )(1 − ξ2 ) − i0]2+ε

(3.25)

Then the integration over ξ2 results in Γ (1 + ε)Γ (−ε)2 Γ (−2ε)

dξ (−t)−1−ε ξ −1−ε − (−s)−1−ε (1 − ξ)−1−ε . s − (s + t)ξ

F (s, t; d) = −iπ d/2

1

× 0

(3.26)

The singularity at s − (s + t)ξ = 0 is absent because the rest of the integrand is zero at this point. To calculate this integral in expansion in ε one needs, however, to separate the two terms in the square brackets. In order not to run into divergence due to the denominator one can perform an auxiliary subtraction at s − (s + t)ξ = 0. We obtain 2

We shall see in Chaps. 5 and 6 that this is indeed an irreducible Feynman integral.

3.4 Feynman Parameters

F (s, t; d) = −iπ d/2

Γ (1 + ε)Γ (−ε)2 [f (s, t; ε) + f (t, s; ε)] , Γ (−2ε)

where −1−ε

f (s, t; ε) = (−t)

0

1

" −1−ε # s dξ −1−ε ξ . − s − (s + t)ξ s+t

41

(3.27)

(3.28)

To expand the function f in a Laurent series in ε one needs to perform another subtraction, at ξ = 0, which we make by the replacement 1 (s + t)ξ 1 → + . s − (s + t)ξ s(s − (s + t)ξ) s

(3.29)

Then the integral with the ﬁrst term can be evaluated by expanding the integrand in ε while the second term is integrated explicitly. Eventually, we arrive at the following result: iπ d/2 e−γE ε 4 2 F (s, t; d) = − [ln(−s) + ln(−t)] st ε2 ε 2 4π +2 ln(−s) ln(−t) − + O(ε) . (3.30) 3 Here and in all the expansions in ε below we pull out the factor e−γE ε , with Euler’s constant γE , per loop in order to avoid it in our results. Although we are oriented at calculations in expansion in ε, let us, for completeness, present a simple result for general ε [16] which can straightforwardly be obtained from (3.27): t iπ d/2 Γ (−ε)2 Γ (ε) −ε F (s, t; d) = − (−t) 2 F1 1, −ε; 1 − ε; 1 + stΓ (−2ε) s $ s , (3.31) +(−s)−ε 2 F1 1, −ε; 1 − ε; 1 + t where 2 F1 is the Gauss hypergeometric function (see (B.1)).

3.4 Feynman Parameters Let us now present the alpha representation of scalar dimensionally regular ized integrals in a modiﬁed form by making the change of variables αl = ηαl , where αl = 1. Starting from (2.36) with Z = 1, performing the integration over η from 0 to ∞ explicitly and omitting primes from the new variables, we obtain d/2 h iπ Γ (a − hd/2) FΓ (q1 , . . . , qn ; d) = (−1)a l Γ (al ) ∞ ∞ U a−(h+1)d/2 αal −1 l l dα1 . . . dαL δ αl − 1 . (3.32) × a−hd/2 0 0 (−V + U m2l αl )

42

3 Evaluating by Alpha and Feynman Parameters

A folklore Cheng–Wu theorem [5] (see also [2]) says that the same formula (3.32) holds with the delta function

αl − 1 , (3.33) δ l∈ν

where ν is an arbitrary subset of the lines 1, . . . , L, when the integration over the rest of the α-variables, i.e. for l∈ν, is extended to the integration from zero to inﬁnity. Observe that the integration over αl for l ∈ ν is bounded at least by 1 from above, as in the case where all the α-variables are involved in the sum in the argument of the delta function. One can prove this theorem straightforwardly by changing variables and calculating the corresponding Jacobian. But a simpler way to prove it3 is to start from the alpha representation (2.36), introduce new variables by αl = ηαl for all l = 1, 2, . . . , L, where η = l∈ν αl , and immediately arrive at (3.32) with the delta function (3.33). Let us stress that this theorem holds not only for (3.32) corresponding to Feynman diagrams with standard propagators but also for the alpha representation derived for Feynman diagrams with various linear propagators. As we will see below in multiple examples, an adequate choice of the delta function in (3.32) can greatly simplify the evaluation. Note that one can use various homogeneous substitutions which keep the form of the delta function in (3.32) – see Sect. 3.1 of [10] and references therein. In addition to alpha parameters, the closely related Feynman parameters are often used. For a product of two propagators, one writes down the following relation: 1 (m21 − p21 )λ1 (m22 − p22 )λ2 dξ ξ λ1 −1 (1 − ξ)λ2 −1 Γ (λ1 + λ2 ) 1 = . (3.34) 2 2 Γ (λ1 )Γ (λ2 ) 0 [(m1 − p1 )ξ + (m22 − p22 )(1 − ξ)]λ1 +λ2 This relation is usually applied to a pair of appropriately chosen propagators if an explicit integration over a loop momentum then becomes possible. Then new Feynman parameters can be introduced for other factors in the integral, etc. In fact, any choice of the Feynman parameters can be achieved by starting from the alpha representation (3.32) and making certain changes of variables. However, the possibility of an intermediate explicit loop integration of the kind mentioned above can be hidden in the alpha integral. The generalization of (3.34) to an arbitrary number of propagators is of the form 1 λ −1 δ ( ξl − 1) Γ ( λl ) 1 1 , = dξ . . . dξ ξl l (3.35) λl 1 L λl Γ (λl ) 0 Al 0 ( Aξ) l

where Al = 3

m2l

−

l l

p2l .

Thanks to A.G. Grozin for pointing out this possibility!

3.5 Two-Loop Examples

43

For the evaluation of diagrams with a small number of loops, the choice of applying either alpha or Feynman parameters is usually just a matter of taste. In particular, if we apply (3.35) to a two-loop diagram and then integrate over two loop momenta, with the help of (A.1) and its generalizations to integrals with numerators, we obtain the same result as that obtained starting from (3.32). For completeness, here is a one more parametric representation which is related to Feynman parameters and is often used in practice: xλ2 −1 dx Γ (λ1 + λ2 ) 1 1 = . (3.36) λ λ A 1B 2 Γ (λ1 )Γ (λ2 ) 0 (A + Bx)λ1 +λ2

3.5 Two-Loop Examples At the two-loop level, we ﬁrst consider the Example 3.4. Two-loop vacuum diagram of Fig. 3.8 with the masses m, 0, m and general complex powers of the propagators.

1 2 3 Fig. 3.8. Vacuum two-loop diagram with the masses m, 0 and m

The Feynman integral is written as F3.4 (m2 ; λ1 , λ2 , λ3 , d) dd k dd l = . (−k 2 + m2 )λ1 [−(k + l)2 ]λ2 (−l2 + m2 )λ3

(3.37)

The two basic functions in the alpha representation are U = α1 α2 +α2 α3 + α3 α1 and V = 0. We apply (3.32) to obtain

∞ ∞ ∞ 3 2 Γ (λ + 2ε − 4) λl −1 d/2 αl dαl F3.4 = iπ Γ (λl )(m2 )λ+2ε−4 0 0 0 l=1

(α1 α2 + α2 α3 + α3 α1 )ε−2 ×δ αl − 1 . (3.38) (α1 + α3 )λ+2ε−4 l

Now we exploit the freedom provided by the Cheng–Wu theorem and choose the argument of the delta function as α1 + α3 − 1. The integration over α2 is

44

3 Evaluating by Alpha and Feynman Parameters

performed from 0 to ∞. Resulting integrals are evaluated in terms of gamma functions for general ε and we arrive at the table formula (A.38). Consider now Example 3.5. Two-loop massless propagator diagram of Fig. 3.9 with arbitrary integer powers of the propagators,

Fig. 3.9. Two-loop propagator diagram

F3.5 (q 2 ; a1 , a2 , a3 , a4 , a5 , d) dd k dd l = . (k 2 )a1 [(q − k)2 ]a2 (l2 )a3 [(q − l)2 ]a4 [(k − l)2 ]a5

(3.39)

The sets of trees and 2-trees relevant to the two basic functions in the alpha representation are shown in Figs. 3.10 and 3.11

Fig. 3.10. Trees contributing to the function U for Fig. 3.9

Fig. 3.11. 2-trees contributing to the function V for Fig. 3.9

Correspondingly, we have U = (α1 + α2 + α3 + α4 )α5 + (α1 + α2 )(α3 + α4 ) , V = [(α1 + α2 )α3 α4 + α1 α2 (α3 + α4 ) + (α1 + α3 )(α2 + α4 )]q ≡ Vq 2 .

(3.40) 2

(3.41)

As we will see in Chaps. 5 and 6, any diagram of this class can be evaluated for general ε in terms of gamma functions. This is however hardly seen from

3.5 Two-Loop Examples

45

its alpha representation. In spite of the fact that the evaluation by alpha parameters is not an optimal method for this class of integrals, let us evaluate, for the sake of illustration, this diagram for all powers of the propagators equal to one, using its alpha representation. It is ﬁnite at d = 4, both in the UV and IR sense. Representation (3.32) takes the form ∞ δ ( αl − 1) (iπ 2 )2 ∞ dα . . . dα . (3.42) F3.5 (q 2 ; 1, 1, 1, 1, 1, 4) = 1 5 q2 UV 0 0 We exploit the Cheng–Wu theorem by choosing the delta function δ (α5 − 1), with the integration over the rest of the four variables from zero to inﬁnity. Then one can delegate the integration procedure to MATHEMATICA [22] and obtain the well-known result4 : 2 2 iπ F3.5 (q 2 ; 1, 1, 1, 1, 1, 4) = 6ζ(3) , (3.43) q2 where ζ(z) is the Riemann zeta function. In the rest of this chapter, we shall consider just two more examples which are, however, more complicated than the previous ones. Example 3.6. Two classes of two-loop integrals5 with integer powers of the propagators: dd k dd l . (3.44) F± (q 2 ; a1 , a2 , a3 ) = 2 a 1 (k + q·k) (l2 + q·l)a2 [(k ± l)2 ]a3 It turns out that the F− is simple. Indeed we rewrite the ﬁrst denominator k 2 + q ·k as (k + q/2)2 − q 2 /4 and similarly the second denominator, make the change of variables k = k − q/2, l = l − q/2 and recognize F− as a two-loop vacuum diagram with the mass m2 = q 2 /4 shown in Fig. 3.8 which was evaluated in Example 3.4 – see (A.38). The integrals F+ are, however, not so simple. Using the same manipulation as above we see that they are graphically recognized as sunset diagrams of Fig. 3.12 at threshold, i.e. q 2 = 4m2 . We start from the alpha representation (2.36) with Z = 1. The two basic functions are U = α1 α2 + α2 α3 + α3 α1 , V = α1 α2 α3 q 2 . 2

(3.45)

2

After using the threshold condition m = q /4 we obtain

4

This result was ﬁrst obtained in [19] by means of expansion in Chebyshev polynomials in momentum space. In [6], it was reproduced using Gegenbauer polynomials in coordinate space. 5 They were involved, in particular, in the calculation [1, 8] of two-loop matching coeﬃcients of the vector current in QCD and Non-Relativistic QCD (NRQCD) [3, 14, 21].

46

3 Evaluating by Alpha and Feynman Parameters

1 2 3 Fig. 3.12. Sunset diagram with the masses m, m, 0

(−1)a ia+2ε−2 Γ (al )

! ∞ ∞ ∞ 3 q2 W al −1 ε−2 αl dαl U exp −i × , 4U 0 0 0

F+ (q 2 ; a1 , a2 , a3 ) =

(3.46)

l=1

where W = (α1 + α2 )α1 α2 + α3 (α1 − α2 )2 .

(3.47)

Proceeding as with the general alpha representation we come to 2 (−1)a iπ d/2 Γ (a + 2ε − 4) 2 F+ (q ; a1 , a2 , a3 ) = Γ (al ) (q 2 /4)a+2ε−4

3 ∞ ∞ ∞ U a+3ε−6 al −1 δ αl − 1 αl dαl . × W a+2ε−4 0 0 0

(3.48)

l=1

We continue to exploit the Cheng–Wu theorem in an appropriate way. We choose the delta function in (3.48) as δ (α1 + α2 − 1) and obtain an integral over ξ = α1 from 0 to 1, with α2 = 1 − ξ, and an integral over t = α3 from 0 to ∞: 2 (−1)a iπ d/2 Γ (a + 2ε − 4) F+ (q 2 ; a1 , a2 , a3 ) = Γ (al ) (q 2 /4)a+2ε−4 ∞ 1 a3 −1 [t + ξ(1 − ξ)]a+3ε−6 t dξ ξ a1 −1 (1 − ξ)a2 −1 dt . (3.49) × [t(1 − 2ξ)2 + ξ(1 − ξ)]a+2ε−4 0 0 This two-parametric integral representation can be used for the evaluation of any diagram of the given class in expansion in ε. Let us show how the integral with all the indices equal to one can be evaluated in expansion in ε up to the ﬁnite part. We start with (3.49) which gives d/2 2 iπ Γ (2ε − 1) 2 F+ (q ; 1, 1, 1) = − 2 (q /4)2ε−1 1 ∞ [t + ξ(1 − ξ)]3ε−3 × dξ dt . (3.50) [t(1 − 2ξ)2 + ξ(1 − ξ)]2ε−1 0 0 Observe that the integrand is invariant under the transformation ξ → 1 − ξ. We write the integral as √ twice the integral from 0 to 1/2 over ξ, change the variable ξ by ξ = (1 − 1 − x)/2 and rescale t → t/4 to obtain

3.5 Two-Loop Examples

47

2

F+ (q 2 ; 1, 1, 1) = − iπ d/2 Γ (2ε − 1)(q 2 /2)1−2ε 1 ∞ dx [t(1 − x) + x]1−2ε √ × dt . (t + x)3−3ε 1−x 0 0

(3.51)

Remember that our integral is UV divergent. The overall divergence is quadratic since the UV degree of divergence is ω = 2, and there are three oneloop logarithmically divergent subgraphs, so that, presumably, there should be poles up to the second order in ε. One source of the poles is the overall gamma function Γ (2ε − 1). Another power of 1/ε comes from the integration over t and x in (3.51), namely from the region of small t and x. To have the possibility to perform an expansion in ε we have to reveal the singularity at ε = 0. Similarly to what we did in Example 3.3, let us perform a subtraction according to the identity [t(1 − x) + x]1−2ε = [t(1 − x) + x]1−2ε − (t + x)1−2ε + (t + x)1−2ε . Now, the integral with the expression in braces can be evaluated by expanding the integrand in a Laurent series in ε, while the last term can be integrated by hand with a result expressed in terms of gamma functions which can be, of course, expanded in ε after the evaluation: √ 1 ∞ πΓ (ε) dx dt √ . = 2−ε (t + x) (1 − ε)Γ (ε + 1/2) 1 − x 0 0 The integration of the subtracted part up to order ε0 can straightforwardly be done by MATHEMATICA [22]. Finally, we obtain the following result: 2 q 2 1−2ε F+ (q 2 ; 1, 1, 1) = iπ d/2 e−γE ε 4 1 1 2 11π 2 − + O(ε) . (3.52) × 2+ + ε ε 12 2 Consider now Example 3.7. Non-planar two-loop massless vertex diagram of Fig. 3.13 with p21 = p22 = 0. The Feynman integral can be written as dd k dd l 2 F3.7 (Q ; a1 , . . . , a6 , d) = [(k + l)2 − 2p1 ·(k + l)]a1 1 × , (3.53) 2 a 2 2 [(k + l) − 2p2 ·(k + l)] (k − 2p1 ·k)a3 (l2 − 2p2 ·l)a4 (k 2 )a5 (l2 )a6 where Q2 = −(p1 − p2 )2 = 2p1 ·p2 , and the loop momenta are chosen as the momenta ﬂowing through lines 5 and 6. Let us proceed by Feynman parameters following [11] where some integrals of this class were calculated. (They were also evaluated in [13] and [16].)

48

3 Evaluating by Alpha and Feynman Parameters

Fig. 3.13. Non-planar vertex diagram

We write down Feynman parametric formula (3.34) for the pairs of the propagators (3, 5) and (4, 6): 1 (−1)a3 +a5 Γ (a3 + a5 ) = (k 2 − 2p1 ·k)a3 (k 2 )a5 Γ (a3 )Γ (a5 ) 1 dξ1 ξ1a3 −1 (1 − ξ1 )a5 −1 × 2 a3 +a5 0 [−(k − ξ1 p1 ) − i0]

(3.54)

and, similarly, for the second pair, with the replacements ξ1 → ξ2 , p1 → p2 , k → l, a3 → a4 , a5 → a6 . Then we change the integration variable l → r = k + l and integrate over k by means of our one-loop tabulated formula (3.6): dk [−(k − ξ1 p1 )2 ]a3 +a5 [−(r − ξ2 p2 − k)2 ]a4 +a6 G(a3 + a5 , a4 + a6 ) = iπ d/2 . (3.55) [−(r − ξ1 p1 − ξ2 p2 )2 ]a3 +a4 +a5 +a6 +ε−2 Then we apply Feynman parametric formula (3.35) to the propagators 1 and 2 and the propagator resulting from the right-hand side of (3.55), with a resulting integral over r evaluated by (A.1): dd r 2 2 [−(r − Q A(ξ1 , ξ2 , ξ3 , ξ4 ))]a+ε−2 1 Γ (a + 2ε − 4) = iπ d/2 , (3.56) Γ (a + ε − 2) (Q2 )a+2ε−4 A(ξ1 , ξ2 , ξ3 , ξ4 )a+2ε−4 where a = a1 + . . . + a6 and A(ξ1 , ξ2 , ξ3 , ξ4 ) = ξ3 ξ4 + (1 − ξ3 − ξ4 )[ξ2 ξ3 (1 − ξ1 ) + ξ1 ξ4 (1 − ξ2 )] . Thus we arrive at the following intermediate result valid for general powers of the propagators:

3.5 Two-Loop Examples

49

2 (−1)a iπ d/2 Γ (2 − ε − a35 )Γ (2 − ε − a46 ) (Q2 )a+2ε−4 Γ (al )Γ (4 − 2ε − a3456 ) 1 1 dξ1 . . . dξ4 ξ1a3 −1 (1 − ξ1 )a5 −1 ξ2a4 −1 (1 − ξ2 )a6 −1 ×Γ (a + 2ε − 4)

F3.7 (Q2 ; a1 , . . . , a6 , d) =

0

0

a3456 +ε−3 ×ξ3a1 −1 ξ4a2 −1 (1 − ξ3 − ξ4 )+ A(ξ1 , ξ2 , ξ3 , ξ4 )4−2ε−a .

(3.57)

We use the shorthand notation a35 = a3 + a5 , a3456 = a3 + a4 + a5 + a6 . As usually, X+ = X for X > 0 and X+ = 0 otherwise. This four-parametric integral representation can be used for the evaluation of Feynman integrals of this class with various indices. Let us use it in the case a1 = . . . = a6 = 1 and evaluate the corresponding Feynman integral in expansion in ε up to the ﬁnite part. We have d/2 2 iπ Γ (2 + 2ε)Γ (−ε)2 2 F3.7 (Q ; 1, . . . , 1, d) = 2 2+2ε (Q ) Γ (−2ε) 1 1 (1 − ξ3 − ξ4 )1+ε + × dξ1 . . . dξ4 . (3.58) 2+2ε A(ξ , ξ , ξ , ξ ) 1 2 3 4 0 0 We introduce new variables by ξ3 = ξη, ξ4 = (1 − ξ)η and integrate over ξ2 to obtain d/2 2 iπ Γ (1 + 2ε)Γ (−ε)2 1 2 dη η −1−2ε (1 − η)ε F3.7 (Q ; 1, . . . , 1, d) = − 2 2+2ε (Q ) Γ (−2ε) 0 1 1 dξdξ1 −1−2ε ξ × [(1 − ξ)η + (1 − η)(1 − ξ1 )]−1−2ε 0 0 ξ − ξ1 (3.59) −(1 − ξ)−1−2ε [ξη + (1 − η)ξ1 ]−1−2ε . The singularity of the denominator at ξ = ξ1 is spurious because the numerator is zero at this point. We notice that, due to the symmetry of the integrand, the integral over ξ and ξ1 equals twice the integral over the domain 0 ≤ ξ1 ≤ ξ ≤ 1. Following [11] again, we turn to the variable z by ξ1 = zξ, make the changes η → 1 − η, z → 1 − z and come to d/2 2 iπ Γ (1 + 2ε)Γ (−ε)2 2 f (ε) , (3.60) F3.7 (Q ; 1, . . . , 1, d) = −2 2 2+2ε (Q ) Γ (−2ε) where

f (ε) = × 0

0 1

1

dη η ε (1 − η)−1−2ε

1

dξ ξ −1−2ε

0

dz [1 − ξ(1 − ηz)]−1−2ε − (1 − ξ)−1−2ε (1 − ηz)−1−2ε . z

(3.61)

At this point it is claimed in [11] that, in principle, it is possible to evaluate this integral, in expansion in ε up to the ﬁnite part, performing appropriate subtractions of the integrand. Still another way was chosen: to expand

50

3 Evaluating by Alpha and Feynman Parameters

various quantities of the type (1 − X)λ in a binomial series, with subsequent integration and summing up resulting multiple series. (This procedure can be qualiﬁed as another method of evaluation.) Let us, however, realize the possibility of making subtractions. Indeed, the situation is complicated because we are dealing with a three-parametric integral so that several subtractions that would reveal the singularities that generate poles in ε are necessary. Since the prefactor in (3.60) involves a simple pole in ε we have to evaluate the function f (ε) given by (3.61) up to order ε1 . There are several sources of the poles: the points ξ = 0, ξ = 1, η = 0, η = 1, and z = 1. The following strategy of subtractions is suitable for the calculation. Let us ﬁrst decompose f into the sum f1 + f2 according to the subtraction of the braces in (3.61) at η = 0, i.e.

(1 − ξ(1 − ηz))−1−2ε − (1 − ξ)−1−2ε

(3.62) +(1 − ξ)−1−2ε 1 − (1 − ηz)−1−2ε . Let us start with f1 . We perform subtraction of the integrand at η = 1 according to the decomposition of the ﬁrst part of (3.62) into

(1 − ξ(1 − z))−1−2ε − (1 − ξ)−1−2ε

(3.63) + (1 − ξ(1 − ηz))−1−2ε − (1 − ξ(1 − z))−1−2ε . The ﬁrst term in (3.63) does not depend on η so that the corresponding integration over η is performed in terms of gamma functions. Then the integral 1 1 dz [1 − ξ(1 − z)]−1−2ε − (1 − ξ)−1−2ε dξ ξ −1−2ε z 0 0 appears. We need a subtraction at ξ = 1 here because when ξ → 1 the factor z −1−2ε generating a pole in ε arises. So we replace ξ −1−2ε by 1+ ξ −1−2ε − 1 . The ﬁrst term corresponding to unity, after integration over ξ, gives the following integral evaluated in terms of gamma functions 1 dz 1 − z −1−2ε = ψ(−2ε) + γE , 0 1−z where ψ(z) is the logarithmical derivative of the gamma function, i.e. ψ(z) = Γ (z)/Γ (z). Thus we obtain the following contribution to our result: Γ (1 + ε)Γ (−2ε) 2εΓ (1 − ε) 3ζ(3) 3π 4 π2 1 − − ε + O(ε2 ) . (3.64) = 3− 8ε 24ε 4 80 Starting from the second term we obtain an integral which can be evaluated by expanding the integrand in ε and performing the integration, e.g., in MATHEMATICA [22], with the following contribution: f11 = −

f12 =

43π 4 π2 + 5ζ(3) + ε + O(ε2 ) . 12ε 180

(3.65)

3.5 Two-Loop Examples

51

In the second part of (3.63), we make the same replacement (with the same motivation) as before, i.e. ξ −1−2ε → 1 + ξ −1−2ε − 1 . The second part here again produces an integral which can be evaluated by expanding the integrand in ε, with the following contribution: 11π 4 ε + O(ε2 ) . (3.66) 120 The unity gives a part where the integration over ξ is explicitly taken. The corresponding result is proportional to the sum of these two two-parametric integrals: 1 1 dηdzη ε (1 − η)−1−2ε 1 − η −1−2ε 0 0 1 1 −2ε 1 − z −2ε ε −1−2ε 1 − (ηz) − dηdzη (1 − η) + . (3.67) 1 − ηz 1−z 0 0 f13 = ζ(3) +

The ﬁrst integral can be evaluated in terms of gamma functions, with the following contribution: Γ (1 − ε) Γ (−2ε) Γ (1 + ε) − f14 = 4ε2 Γ (1 − ε) Γ (1 − 3ε) 2 π4 π − ζ(3) − ε + O(ε2 ) . (3.68) =− 12ε 36 In the second integral, one can expand the integrand in ε. Here is the corresponding contribution: π4 ε + O(ε2 ) . (3.69) 72 Let us now deal with f2 deﬁned by the second part of (3.62). The integration over ξ is performed explicitly, and the following integral over z arises: 1 dz

(1 − ηz)−1−2ε − 1 . z 0 f15 = −ζ(3) −

When z → 1 a factor (1 − η)−1−2ε appears so that we need a subtraction at z = 1. We make the replacement 1/z → 1 + (1 − z)/z. The unity generates a part which is integrated explicitly over z and then over η. The resulting contribution is then Γ (−2ε)2 Γ (ε) 1 Γ (−4ε) Γ (−2ε) Γ (−2ε) − f21 = − + Γ (−4ε) 2ε Γ (−3ε) Γ (−ε) Γ (−ε) 2 2 4 29π π 1 π 1 − + 2ζ(3) + − 7ζ(3) ε + O(ε2 ) . = 3+ 2+ 8ε 2ε 12ε 6 360 (3.70) Starting from the second term and performing one more subtraction we obtain the following integral

52

3 Evaluating by Alpha and Feynman Parameters

0

1

1−z dηdzη ε (1 − η)−1−2ε z 0

−1−2ε − (1 − z)−1−2ε + (1 − z)−1−2ε − 1 . × (1 − ηz) 1

(3.71)

For the part corresponding to the second square brackets, one can explicitly integrate over η and then expand the integrand in ε and integrate over z with the following resulting contribution: Γ (−2ε)3 Γ (1 + ε) 1 f22 = − + 1 − ψ(−2ε) − γE Γ (−4ε)Γ (1 − ε) 2ε 4 π π2 π2 1 + − 2ζ(3) + + 7ζ(3) ε + O(ε2 ) . (3.72) =− 2 − 2ε 6ε 6 90 For the part corresponding to the ﬁrst square brackets in (3.71), one can expand the integrand in ε and integrate over z and η with the following resulting contribution: 19π 4 π2 − 9ζ(3) + ε + O(ε2 ) . (3.73) 6ε 45 Collecting all the eight contributions obtained and taking into account the prefactor in (3.60) we arrive at the well-known analytical result [11] d/2 −γ ε 2 iπ e E 2 F3.7 (Q ; 1, . . . , 1, d) = (Q2 )2+2ε 1 π2 83ζ(3) 59π 4 − × 4− 2 − + O(ε) . (3.74) ε ε 3ε 120 f23 = −

In [11], a similar algorithm based on Feynman parameters has been developed for the evaluation of planar massless two-loop vertex diagrams. It has turned out that the evaluation, by Feynman parameters, in the planar case is more complicated. As we will see in Chaps. 5 and 6, there is, however, a better choice of an appropriate method in this situation and the planar vertex diagrams of this class are in fact much simpler than the non-planar ones.

References 1. M. Beneke, A. Signer and V.A. Smirnov, Phys. Rev. Lett. 80 (1998) 2535. 45 2. K.S. Bjoerkevoll, P. Osland and G. Faeldt, Nucl. Phys. B 386 (1992) 303. 42 3. G.T. Bodwin, E. Braaten and G.P. Lepage, Phys. Rev. D 51 (1995) 1125; Phys. Rev. D 55 (1997) 5853. 45 4. N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 5. H. Cheng and T.T. Wu, Expanding Protons: Scattering at High Energies (MIT Press, Cambridge, MA, 1987). 42 6. K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Nucl. Phys. B 174 (1980) 345. 45

References 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

53

J.C. Collins, Renormalization (Cambridge University Press, Cambridge, 1984). 31 A. Czarnecki and K. Melnikov, Phys. Rev. Lett. 87 (2001) 013001. 45 A.I. Davydychev, Phys. Lett. B 263 (1991) 107. 36 A.I. Davydychev and R. Delbourgo, J. Math. Phys. 39 (1998) 4299. 42 R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542. 47, 49, 52 A.G. Grozin, Heavy Quark Eﬀective Theory (Springer, Berlin, Heidelberg, 2004). 33 G. Kramer and B. Lampe, J. Math. Phys. 28 (1987) 945. 47 G.P. Lepage et al., Phys. Rev. D 46 (1992) 4052. 45 A.V. Manohar and M.B. Wise, Heavy Quark Physics (Cambridge University Press, Cambridge, 2000). 33 W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 41, 47 M. Neubert, Phys. Rep. 245 (1994) 259. 33 M.E. Peskin and D.V. Schroeder, An Introduction to Quantum Field Theory (Perseus, Reading, MA, 1995). 31 J.L. Rosner, Ann. Phys. 44 (1967) 11. 45 O.V. Tarasov, Nucl. Phys. B 480 (1996) 397; Phys. Rev. D 54 (1996) 6479. 36 B.A. Thacker and G.P. Lepage, Phys. Rev. D 43 (1991) 196. 45 S. Wolfram, The Mathematica Book, 4th edition (Wolfram Media and Cambridge University Press, Cambridge, 1999). 35, 45, 47, 50

4 Evaluating by MB Representation

One often uses Mellin integrals1 when dealing with Feynman integrals. These are integrals over contours in a complex plane along the imaginary axis of a product and ratio of gamma functions. In particular, the inverse Mellin transform is given by such an integral. We shall, however, deal with a very speciﬁc technique in this ﬁeld. The key ingredient of the method presented in this chapter is the MB representation used to replace a sum of two terms raised to some power by the product of these terms raised to some powers. Our goal is to use such a factorization in order to achieve the possibility to perform integrations in terms of gamma functions, at the cost of introducing extra Mellin integrations. Then one obtains a multiple Mellin integral of gamma functions in the numerator and denominator. The next step is the resolution of the singularities in ε by means of shifting contours and taking residues. It turns out that multiple MB integrals are very convenient for this purpose. The ﬁnal step is to perform at least some of the Mellin integrations explicitly, by means of the ﬁrst and the second Barnes lemma and their corollaries and/or evaluate these integrals by closing the integration contours in the complex plane and summing up corresponding series. In Sect. 4.1 we start with simple one-loop examples. In Sect. 4.2 we discuss general properties of multiple MB integrals we are going to deal with. We continue in Sect. 4.3 with typical one-loop examples. In fact we shall illustrate the method of MB representation mainly by the same characteristic examples as in the case of the method of alpha and Feynman parameters in Chap. 3. Let us stress, however, that, for double and triple boxes, complete analytical calculations strictly by means of alpha and Feynman parameters, or, by some other techniques, are not known. We turn to various two-loop examples of massless and massive diagrams in Sects. 4.4 and 4.5, respectively. We then consider three- and even four-loop examples in Sects. 4.6 and 4.7. In Sect. 4.8, we discuss how multiple MB integrals can be used to obtain asymptotic expansions of Feynman integrals in various limits and compare this procedure with expansion by regions [4, 27]. In the last section, we also discuss some other results obtained by means of MB integrals and summarize basic characteristic features of the method presented in this chapter. 1

First examples of application of Mellin integrals to Feynman integrals can be found in [5, 34]. V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 55–107 (2004) c Springer-Verlag Berlin Heidelberg 2004

56

4 Evaluating by MB Representation

4.1 One-Loop Examples Our basic tool is the following formula: +i∞ 1 1 1 Yz = dz Γ (λ + z)Γ (−z) . (X + Y )λ Γ (λ) 2πi −i∞ X λ+z

(4.1)

Here the contour of integration is chosen in the standard way: the poles with a Γ (. . . + z) dependence (let us call them left poles, for brevity) are to the left of the contour and the poles with a Γ (. . . − z) dependence (right poles) are to the right of it. See Fig. 4.1, where a possible contour C is shown in the case of λ = −1/4 − i/2. (This terminology is useful and, although it often happens that the ﬁrst right pole is to the left of the ﬁrst left pole of a given integrand, this, hopefully, will not cause misunderstanding.) Im z C

−λ − 2

-2

−λ − 1

-1

2

1 −λ 0

1

Re z 2

-1

-2 Fig. 4.1. Possible integration contour in (4.1) for λ = −1/4 − i/2

We shall use decompositions X + Y of various functions in integrals over Feynman and alpha parameters. But a more transparent way2 to apply this representation is to write down a massive propagator in terms of massless ones: +i∞ 1 1 (m2 )z 1 = dz Γ (λ + z)Γ (−z) . (4.2) 2 2 λ (m − k ) Γ (λ) 2πi −i∞ (−k 2 )λ+z Our ﬁrst example is the same as Example 3.1: 2

Historically, it was ﬁrst advocated and applied in [8].

4.1 One-Loop Examples

57

Example 4.1. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. We insert (4.2) with λ = a1 into (1.2), apply (3.6) and obtain the following result: F4.1 (q 2 , m2 ; a1 , a2 , d) =

iπ d/2 (−1)a1 +a2 Γ (2 − ε − a2 ) Γ (a1 )Γ (a2 )(−q 2 )a1 +a2 +ε−2 +i∞ 2 z m 1 × dz Γ (a1 + a2 + ε − 2 + z) 2πi −i∞ −q 2 Γ (2 − ε − a1 − z)Γ (−z) × . (4.3) Γ (4 − 2ε − a1 − a2 − z)

The rules for choosing an integration contour that goes from −i∞ to +i∞ in the complex z-plane are the same as before: the right poles (in Γ (. . . − z)) are to the right of the contour and the left poles (in Γ (. . . + z)) are to left. This representation can be used to evaluate any integral of this family in a Laurent expansion in ε. In particular, for F4.1 (q 2 , m2 ; 2, 1, d), we obtain (1.9) and, at d = 4 come to +i∞ 2 z m Γ (1 + z)Γ (−z)2 iπ 2 1 (4.4) dz F4.1 (2, 1, 4) = 2 q 2πi −i∞ −q 2 Γ (1 − z) with an integration contour at −1 < Rez < 0. Using properties of the gamma function we obtain (1.10). Here is a subtle point: if we look at (1.10) we observe that there is a product Γ (z)Γ (−z) which would be bad if it was present from the beginning because we could not satisfy our agreement about choosing the integration contours. Indeed, here the right and left poles at ε = 0 glue together and there is no space between them. However, the situation is unambiguous because we have ﬁxed an integration contour with −1 < Rez < 0 and we are free to perform identical transformations of the integrand after that. A moral of this discussion is the recipe to derive the MB representation for general powers of the propagators al and ﬁx appropriate integration contours at this point. Then, for concrete integer indices al , we are allowed to make transformations like Γ (1 + z)Γ (−z) = −Γ (z)Γ (1 − z), but it is necessary to remember about the choice of the contours made before this. The integral (1.10) can be evaluated, according to the Cauchy theorem, by closing the integration contour to the right and taking a series of residues (with the minus sign, at the points z = 0, 1, 2, . . .. The residue of course) at z = 0 gives iπ 2 ln −q 2 /m2 /q 2 and the residues at z = 1, 2, . . . give the series n ∞ iπ 2 1 m2 . − 2 q n=1 n q 2 As a result, we reproduce (1.5).

58

4 Evaluating by MB Representation

In the case of the indices equal to one we use (4.3) to obtain iπ 2 Γ (1 − ε) F4.1 (q 2 , m2 ; 1, 1, d) = (−q 2 )ε 2 z m Γ (ε + z)Γ (−z)Γ (1 − ε − z) 1 . × dz 2πi C −q 2 Γ (2 − 2ε − z)

(4.5)

To evaluate MB integrals in a Laurent expansion in ε the ﬁrst point is to analyse how singularities in ε are generated. We know in advance that the given integral has a pole in ε because the diagram is UV-divergent. There are no explicit functions with singularities in ε so that the pole is generated by the MB integration. Indeed, the product Γ (ε + z)Γ (−z) generates a singularity in ε when ε → 0 because the ﬁrst left pole, i.e. at z = −ε, and the ﬁrst right pole, i.e. z = 0, glue together when ε = 0, and there is no place for a contour between these poles. Possible integration contours C in (4.5) in the cases Re ε > 0 and Re ε < 0 are shown in Figs. 4.2 and 4.3, respectively. In the former case, a contour can be chosen as a straight line parallel to the imaginary axis, while in the latter case, there is no such choice. However, no matter which value of ε we can imagine, we shall use the same procedure to reveal the pole in ε: we write down the integral (4.5) as the sum of a similar integral over a new contour, C , which goes to the left of the pole at z = −ε and the residue at this point. In the integral over the shifted contour, the nature of the pole at z = −ε changes, and it becomes right, rather than left, in our terminology.

C

Im z C

2

1 −ε

−ε − 1 -2

-1

1−ε 0

1

Re z 2

-1

-2 Fig. 4.2. Possible integration contour in (4.5) in the case Re ε > 0

4.1 One-Loop Examples

59

Im z

C, C

2

1 −ε − 2 -2

−ε − 1 -1

C −ε

C 0

1−ε 1

Re z 2

-1

-2 Fig. 4.3. Possible integration contour in (4.5) in the case Re ε < 0

The crucial point is that, in the integral over C , we can safely expand the integrand in a Laurent series in ε. (In this particular example, this is just a Taylor series.) As to the residue, it is equal to iπ 2

Γ (ε) (m2 )ε (1 − ε)

and can explicitly be expanded in ε. For the integral over the shifted contour C , with −1 < Rez < 0, we obtain, at ε = 0, 2 z m Γ (z)Γ (−z) 2 1 . iπ dz 2πi C −q 2 1−z This MB integral can be evaluated by closing the integration contour to the right in the complex z-plane, as in the previous example. Combining the corresponding result with the residue calculated above we arrive at (1.7). In fact, we could similarly proceed by moving the contour C across the right pole at z = 0 and, correspondingly, taking minus residue at this point. Then the integral over the new contour C would be at 0 < Rez < 1. The next example is the same as Example 3.2: Example 4.2. The triangle diagram of Fig. 3.4. We again exploit the MB representation in the simplest way, i.e. apply (4.2) to the only massive propagator in (3.19), and evaluate the resulting massless triangle integral by (A.28) to obtain the following result:

60

4 Evaluating by MB Representation

(−1)a iπ d/2 Γ (al )(Q2 )a+ε−2 +i∞ 2 z m 1 × dz Γ (a3 + z)Γ (a + ε − 2 + z) 2πi −i∞ Q2 Γ (2 − ε − a1 − a3 − z)Γ (2 − ε − a2 − a3 − z)Γ (−z) × , Γ (4 − 2ε − a − z)

F4.2 (Q2 , m2 ; a1 , a2 , a3 , d) =

(4.6)

where a = a1 + a2 + a3 and Q2 = −(p1 − p2 )2 as above. Consider, as in Chap. 3, the diagram with the powers of the propagators equal to one: F4.2 (Q2 , m2 ; 1, 1, 1, d) = − ×

1 2πi

+i∞

dz −i∞

m2 Q2

z

iπ d/2 (Q2 )1+ε

Γ (1 + ε + z)Γ (1 + z)Γ (−ε − z)2 Γ (−z) . (4.7) Γ (1 − 2ε − z)

If we want to calculate this integral at ε = 0, we observe that we can safely set ε = 0 in the integrand because the right and left poles in the complex z-plane are well separated. We obtain F4.2 (Q2 , m2 ; 1, 1, 1, 4) =

iπ 2 (Q2 ) 1 × 2πi

+i∞

dz −i∞

m2 Q2

z

Γ (1 + z)2 Γ (−z)2 , (4.8) z

where the integration contour can be chosen with −1 < Rez < 0. The integral can be evaluated by the same procedure as before, with the known result (3.21). Any integral (3.19) with integer indices can be evaluated using (4.6). For example, +i∞ 2 z m iπ d/2 1 2 2 dz F4.2 (Q , m ; 2, 1, 1, d) = (Q2 )2+ε 2πi −i∞ Q2 Γ (2 + ε + z)Γ (1 + z)Γ (−1 − ε − z)Γ (−ε − z)Γ (−z) . (4.9) × Γ (−2ε − z) We know in advance that there should be an IR pole in ε because of the second power of the ﬁrst massless propagator so that we anticipate that a pole is generated by the MB integration. Indeed, we observe that the only source of the singularity in ε is the product Γ (1+z)Γ (−1−ε−z). When ε → 0 the ﬁrst left pole (from Γ (1 + z)) and the ﬁrst right pole (from Γ (−1 − ε − z)) tend to each other and there is no place for an integration contour to go between them. To evaluate (4.9) in expansion in ε we apply the strategy formulated above: we turn to the integral over a shifted contour which goes to the left of the ﬁrst pole of Γ (−1 − ε − z) so that this pole changes its nature, i.e. becomes left. According to the Cauchy theorem, (4.9) equals the integral over

4.1 One-Loop Examples

61

the shifted contour minus residue of the integrand at the point z = −1 − ε. Then the integral is evaluated by closing the contour (which can again be taken at −1 < Rez < 0) to the right and summing up a series of residues at the points z = 0, 1, 2, . . .). We thus obtain iπ d/2 e−γE ε F4.2 (Q2 , m2 ; 2, 1, 1, d) = − Q2 1 1 ln(−m2 /Q2 ) 2 − ln m × + O(ε) . + m2 ε m2 − Q2

(4.10)

As before, we again had two options: to change the nature of the ﬁrst pole of Γ (−1−ε−z) or the ﬁrst pole of Γ (1+z). Let us agree, for deﬁniteness, that we shall always try to obtain MB integrals expanded in ε at −1 < Rez < 0. The next example is the same as Example 3.3: Example 4.3. The massless on-shell box diagram of Fig. 3.5, i.e. with p2i = 0, i = 1, 2, 3, 4. Up to now we applied MB representation using (4.2). Let us start with (3.24). The natural idea here is to apply (4.1) to the denominator of the integrand. We do this with X = −sξ1 ξ2 . After that we change the order of integration over z and the parameters ξ1 and ξ2 and evaluate the parametric integrals in terms of gamma functions: F4.3 (s, t; a1 , a2 , a3 , a4 , d) =

(−1)a iπ d/2 Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞ z t 1 dz Γ (a + ε − 2 + z)Γ (a2 + z)Γ (a4 + z)Γ (−z) 2πi −i∞ s ×Γ (2 − a1 − a2 − a4 − ε − z)Γ (2 − a2 − a3 − a4 − ε − z) , (4.11)

×

where a = a1 + a2 + a3 + a4 . One can use this representation to evaluate any box with integer powers of the propagators in expansion in ε. In particular, F (s, t; d) ≡ F4.3 (s, t; 1, 1, 1, 1, d) = ×

1 2πi

+i∞

dz −i∞

iπ d/2 Γ (−2ε)(−s)2+ε

z t Γ (2 + ε + z)Γ (1 + z)2 Γ (−1 − ε − z)2 Γ (−z) . (4.12) s

The way how poles in ε are generated is already familiar: we immediately identify the product Γ (1 + z)2 Γ (−1 − ε − z)2 responsible for that. The only diﬀerence with the previous cases is that the left poles in Γ (1 + z)2 and the right poles in Γ (−1 − ε − z)2 are of the second order. After this analysis we proceed as before: take minus residue at z = −1 − ε and turn to the integral over the contour which goes to the right of it. The contribution of the residue is

62

4 Evaluating by MB Representation

iπ

d/2

Γ (1 + ε)Γ (−ε)2 t ln + 2ψ(−ε) − ψ(1 + ε) + γE , Γ (−2ε)s(−t)1+ε s

(4.13)

where ψ(z) is the logarithmical derivative of the Γ -function. There is no gluing of left and right poles in the integral over the shifted contour so that it can be expanded safely in a Taylor series in ε. Every term of this expansion can be integrated by closing the integration contour to the right, taking residues at the points z = 0, 1, 2, . . ., and summing up the resulting series. Combining this contribution with (4.13) we obtain F (s, t; d) = −

iπ d/2 e−γE ε cj (x) εj , (−s)1+2ε t j=−2

(4.14)

where x = t/s. To calculate the ﬁrst coeﬃcients c−2 , . . . , c1 , it is enough to use MATHEMATICA for summing up the series involved. However, starting from c2 , it does not work. In this case, one can use summation formulae (C.83)– (C.94) [14]. One can also do this automatically, using the package SUMMER [39] implemented in FORM [38]. We have c−2 = 4 , c−1 = −2 ln x , c0 = −

4π 2 , 3

(4.15)

c1 = 2 (Li3 (−x) − ln x Li2 (−x)) 1 7π 2 34ζ(3) ln x − π 2 + ln2 x ln(1 + x) − , (4.16) + ln3 x + 3 6 3 c2 = 2 (S2,2 (−x) − Li4 (−x) + ln(1 + x)Li3 (−x) − ln x S1,2 (−x)) π2 (ln x − ln(1 + x))2 + ln x (ln x − 2 ln(1 + x)) Li2 (−x) − 2 2 1 1 ln x ln(1 + x) − ln2 (1 + x) − ln2 x + ln2 x 3 2 6 41π 4 2 , (4.17) + (10 ln x − 3 ln(1 + x))ζ(3) − 3 360 where, in addition to polylogarithms, we encounter generalized polylogarithms Sa,b [12, 20] (see (B.8)). One indeed needs to know expansions of one-loop Feynman integrals up to order ε2 if one wants to perform calculations in two loops because some two-loop contributions factorize and one-loop diagrams enter with coeﬃcients that have poles up to 1/ε2 . On the other hand, the functions that enter ε2 terms of expansion of one-loop Feynman integrals should be present in genuine two-loop contributions, although the ‘true’ two-loop world is, of course, much more complicated than the ε2 -expansion of the one-loop world so that, usually, two-loop results involve functions that are not present in one-loop. Any on-shell massless box with integer indices can be evaluated by a similar procedure. Generally, one encounters several right and left poles which tend to each other when ε → 0. For example, we have

4.2 Multiple MB Integrals

63

iπ d/2 Γ (−1 − 2ε)(−s)3+ε +i∞ z t 1 × dz Γ (3 + ε + z) 2πi −i∞ s

F4.3 (s, t; 2, 1, 1, 1, d) = −

×Γ (1 + z)2 Γ (−2 − ε − z)Γ (−1 − ε − z)Γ (−z) .

(4.18)

Here the ﬁrst two left poles of Γ (1 + z) glue, when ε → 0, with the ﬁrst two right poles of the product Γ (−2−ε−z)Γ (−1−ε−z). However the generalization of the above procedure to such situations is straightforward: one shifts the initial contour across the poles at z = −1 − ε and z = −2 − ε and takes two residues (with the minus sign) at these points. The procedure of evaluating any given Feynman integral from this class can easily be implemented on a computer. 2

4.2 Multiple MB Integrals Up to now we were dealing with one-parametric MB integrals. To resolve the singularities in ε we analysed the integrand, and then shifted contours and took residues, in an appropriate way. In the end of this procedure we obtained either explicit expressions for general ε or integrals where a Laurent expansion of the integrand in ε was possible. In fact, we are going to use a similar procedure for multiple MB integrals which arise when evaluating more complicated Feynman integrals. Of course, the resolution of singularities in ε in such multi-dimensional MB integrals is more complicated than in the one-dimensional case. Usually, the poles in ε are not visible at once, at a ﬁrst integration over one of the MB variables. However, the rule for ﬁnding a mechanism of the generation of poles is just a straightforward generalization of the rule used in the previous one-loop examples with one-parametric MB integrals. For example, for the massless master on-shell box, we observed that the product of Γ (1 + z) and Γ (−1 − ε − z) generated a pole of the type Γ (−ε) (this is nothing but the value of one of these gamma functions at the pole of the other gamma function). Suppose now that we are dealing with a multiple MB integral and we start from the integration over one of the variables, z. We shall analyse various products Γ (a + z)Γ (b − z), where a and b depend on the rest of the variables, with the understanding that this integration generates a pole of the type Γ (a + b). Indeed, if we shift an initial contour of integration over z across the point z = −a we obtain an integral over a new contour which is not singular at a + b = 0, while the corresponding residue involves an explicit factor Γ (a + b). (Well, sometimes it turns out that it is cancelled by a factor in the denominator.) This observation shows that any contour of one of the next integrations over the rest of the MB variables should be chosen according to this dependence, Γ (a + b). We continue this analysis, in a similar way, with various

64

4 Evaluating by MB Representation

next integrations of the second level, etc. In other words, we consider various orders of integrations over given MB variables and analyse whether a singular dependence on ε in the form of some gamma function, e.g. Γ (−ε), is generated in a given order. After this ﬁrst step, we can identify some gamma functions (in the numerator of the integrand) that are essential for the generation of poles in ε. Then we proceed with one of the MB integrations as in the case of one-dimensional MB integrals by shifting contour and taking residue. In the integral over the shifted contour, we continue this procedure by taking care of another key gamma function etc. The corresponding residue has one integration less. We deal with it exactly like with the initial integral, i.e. perform an analysis of generation of poles and then shift contours and take residues. In the end of our procedure, we are left with MB integrals which can be expanded in a Laurent series in ε under the sign of integration. We shall usually evaluate such expanded MB integrals by means of the table of one-dimensional MB integrals presented in Appendix D. All these formulae are corollaries of the ﬁrst and the second Barnes lemmas (D.1) and (D.47). Typically, the integration over the last variable is performed, as in the previous examples, by shifting the contour to the right (or left) and taking a series of residues. These series are summed up by means of summation formulae of Appendix C. There is an alternative strategy [2, 33] for the evaluation of multiple MB integrals. First, one chooses a domain of the regularization parameter ε in such a way that all the integrations over the MB variables can be performed over straight lines parallel to imaginary axis. Then one lets ε → 0, and whenever a pole of some gamma function is crossed one takes into account the corresponding residue. It is simple to organize this procedure in such a way that no more than one pole is crossed at the same time. For every resulting residue, which involves one integration less, a similar procedure is applied, and so on. We shall not, however, use this strategy. In fact, we are going to be pragmatic and not bother whether the change of the order of integration over MB variables is legitimate. Well, usually, at least at large values in the complex plane, the convergence of MB integrals is perfect3 because gamma functions have exponential decrease in both imaginary directions. This property can be used for numerical checks. Moreover, in complicated situations, one can decompose a given integrand into pieces and choose an order of integration for every piece in a special way, with the possibility to integrate explicitly, using table formulae of Appendix D.

3

However, in some situations, e.g. in a MB integral for the Gauss hypergeometric function, the asymptotic exponents of gamma functions cancel each other so that the convergence is deﬁned by the value of the argument x which is present in the MB integral as xz . Depending on whether |x| < 1 or |x| > 1, one has to close the integration contour to the right or to the left. Closing the contours to the diﬀerent sides corresponds to an analytical continuation with respect to the argument x.

4.3 More One-Loop Examples

65

We shall apply some standard properties of integration for multiple MB integrals. We shall use changes of variables of the type z → ±z + z0 . When doing this we shall, of course, trace how the nature of various poles is transformed. Note that, after such a change, z → −z, right poles become left poles. The IBP is also possible in multiple MB integrals, although it is reasonable to apply it in rare situations. Still sometimes it is useful. For example, tabulated formulae of Appendix D with the factor 1/z 2 were derived using the IBP identity f (z) f (z) . (4.19) dz 2 = dz z z C C The word ‘multiple’ will mean, in examples below, the number of MB integrations from two to eight (and even ten, in some restricted sense) which is indeed a big number. Still even in such situations, an explicit integration becomes possible, probably, because multiple MB integrals arising in the evaluation of Feynman integrals are very ﬂexible, both in the procedure of resolving the structure of singularities in ε and when evaluating ﬁnite integrals after expansion in ε. Before evaluating a Feynman integral by means of MB integrals, we shall need to derive an appropriate MB representation. Of course, we shall try to have a minimal number of MB integrations. In every case, we shall derive MB representations for general powers of the propagators. This is useful and important for several reasons. First, if we obtain a MB representation for general indices which we might imagine as complex we will certainly have unambiguous prescriptions for choosing integration contours. Second, such general formulae can be checked using various partial simple cases. Finally, starting from a general formula we can derive a lot of formulae by setting some indices to zero and thereby turning to graphs where the corresponding lines are contracted to a point. We will illustrate all these features through multiple examples below.

4.3 More One-Loop Examples We now turn to a class of one-loop Feynman integrals with two more parameters. Example 4.4. The massless box diagram of Fig. 3.5 with two legs on shell, p23 = p24 = 0, and two legs oﬀ shell, p21 , p22 = 0. We proceed like in the pure on-shell case, using alpha parameters, and obtain

66

4 Evaluating by MB Representation

Γ (a + ε − 2) F4.4 (s, t, p21 , p22 ; a1 , . . . , a4 , d) = iπ d/2 (−1)a Γ (al )

4 ∞ ∞ 4 al −1 ... αl dαl δ αl − 1 × 0

0

l=1

×(−sα1 α3 − tα2 α4 −

p21 α1 α2

−

l=1 2 p2 α2 α3

− i0)2−a−ε .

(4.20)

We have chosen the delta function of the sum of all the α-variables so that the factor with a power of the function U is equal to one. Now we need a generalization of (4.1) to the case of several terms which is easily obtained by induction: +i∞ +i∞ n 1 1 1 = . . . dz . . . dz Xizi 2 n (X1 + . . . + Xn )λ Γ (λ) (2πi)n−1 −i∞ −i∞ i=2 ×X1−λ−z2 −...−zn Γ (λ + z2 + . . . + zn )

n

Γ (−zi ) .

(4.21)

i=2

We use (4.21) to replace the last factor in (4.20) by a product of four factors thus separating terms with t, p21 and p22 from s. After that we introduce new variables by α1 = η1 ξ1 , α2 = η1 (1 − ξ1 ), α3 = η2 ξ2 , α4 = η2 (1 − ξ2 ) and arrive at a product of three parametric integrals evaluated in terms of gamma functions. Eventually we obtain the following threefold MB representation of a general Feynman integral of the given class: F4.4 (s, t, p21 , p22 ; a1 , . . . , a4 , d) =

iπ d/2 (−1)a Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞ +i∞ +i∞ (−p21 )z2 (−p22 )z3 (−t)z4 1 dz2 dz3 dz4 3 (2πi) −i∞ −i∞ −i∞ (−s)z2 +z3 +z4 ×Γ (a + ε − 2 + z2 + z3 + z4 )Γ (a2 + z2 + z3 + z4 )Γ (a4 + z4 ) ×

×Γ (2 − ε − a234 − z3 − z4 )Γ (2 − ε − a124 − z2 − z4 ) ×Γ (−z2 )Γ (−z3 )Γ (−z4 ) .

(4.22)

In this chapter, we continue to use our notation: a124 = a1 + a2 + a4 , etc. with a = a1234 . This representation can be, of course, used for evaluating these Feynman integrals. We shall use it, however, in the next section only as an auxiliary result when deriving an MB representation for the massless on-shell double box diagrams. One of the advantages of general formulae is that they provide a lot of partial cases. For example (4.22) immediately gives a twofold MB representation for Example 4.5. The massless box diagram of Fig. 3.5 with three legs on shell, p22 = p23 = p24 = 0, and one leg oﬀ shell, p21 = 0. Indeed we put p22 to zero in the ‘naive’ sense, i.e. in the integrand of the corresponding Feynman integral or in some parametric representation. This

4.3 More One-Loop Examples

67

is equivalent to setting p22 to zero in the sense of the leading term of the hard part of the asymptotic expansion in the limit p22 → 0 (see details in [27]), which corresponds to taking residues (with the minus sign) of the poles of Γ (−z3 ). So we just take minus residue of the integrand at z3 = 0. Thus we obtain F4.5 (s, t, p21 ; a1 , . . . , a4 , d) =

iπ d/2 (−1)a Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞ +i∞ (−p21 )z2 (−t)z4 1 dz2 dz4 Γ (a + ε − 2 + z2 + z4 ) 2 (2πi) −i∞ −i∞ (−s)z2 +z4 ×Γ (a2 + z2 + z4 )Γ (a4 + z4 )Γ (2 − ε − a234 − z4 ) ×Γ (2 − ε − a124 − z2 − z4 )Γ (−z2 )Γ (−z4 ) . (4.23)

×

Let us now turn to massive diagrams. Example 4.6. The on-shell box with two massive and two massless lines shown in Fig. 4.4, with p21 = . . . = p24 = m2 .

p1 2

p2

p3

1 4 3

p4

Fig. 4.4. On-shell box with two massive and two massless lines. The solid lines denote massive, the dotted lines massless particles

The derivation of the corresponding MB representation is quite straightforward. The combination that is involved in the corresponding integral over alpha or Feynman parameters has now an additional piece as compared with the massless case: V −U m2l αl = sα1 α3 + tα2 α4 − m2 (α1 + α3 )2 . This term can be separated from the rest terms at the cost of introducing one more MB integration according to (4.21). This time, let us introduce new parametric variables in a slightly diﬀerent way, α1 = η1 ξ1 , α2 = η2 ξ2 , α3 = η1 (1 − ξ1 ), α4 = η2 (1 − ξ2 ), in order to make (α1 + α3 )2 simpler. Evaluating the parametric integrals we arrive at the following massive generalization of (4.11): F4.6 (s, t, m2 ; a1 , a2 , a3 , a4 , d) = 1 × (2πi)2

+i∞

−i∞

(−1)a iπ d/2 Γ (4 − 2ε − a) Γ (al )(−s)a+ε−2

+i∞

−i∞

dz1 dz2

(−t)z1 (m2 )z2 Γ (a + ε − 2 + z1 + z2 ) (−s)z1 +z2

68

4 Evaluating by MB Representation

×Γ (a2 + z1 )Γ (a4 + z1 )Γ (−z1 )Γ (−z2 )Γ (2 − a124 − ε − z1 − z2 ) Γ (4 − a122344 − 2ε − 2z1 ) , (4.24) ×Γ (2 − a234 − ε − z1 − z2 ) Γ (4 − a122344 − 2ε − 2z1 − 2z2 ) where a122344 = a1 + 2a2 + a3 + 2a4 , etc. Observe that the onefold representation (4.11) in the massless case follows from (4.24) when we put m to zero. As it was discussed above we do this by taking the limit m → 0 in the sense of the leading term of the hard part of the expansion. Here this means that we just take minus residue at z2 = 0 with respect to the variable z2 which enters the integrand as the exponent of m2 . In particular, we have F4.6 (s, t, m2 ; 1, 1, 1, 1, d) = ×

1 (2πi)2

+i∞

−i∞

+i∞

−i∞

dz1 dz2

(−1)a iπ d/2 Γ (−2ε)(−s)2+ε (−t)z1 (m2 )z2 Γ (2 + ε + z1 + z2 )Γ (−z1 ) (−s)z1 +z2

×Γ (−z2 )Γ (−1 − ε − z1 − z2 )2

Γ (1 + z1 )2 Γ (−2 − 2ε − 2z1 ) . Γ (−2 − 2ε − 2z1 − 2z2 )

(4.25)

The resolution of singularities in ε can be performed here as in the onedimensional case because only the product Γ (1 + z1 )2 Γ (−2 − 2ε − 2z1 ) is responsible for the generation of poles. To see this, we use properties of the gamma function and write Γ (−2−2ε−2z1 ) as Γ (−1−ε−z1 )Γ (−1/2−ε−z1 ) up to a factor so that we obtain the product Γ (1 + z1 )2 Γ (−1 − ε − z1 ) which involves gluing of the left pole at z1 = −1 and the right pole at z1 = −1 − ε when ε → 0. We proceed as in Sect. 4.1 by taking minus residue at the point z1 = −1 − ε and shifting the integration contour over z1 across this point. The residue gives 2 z2 +i∞ m Γ (1 + z2 )Γ (−z2 )3 Γ (1 + ε)Γ (−ε)2 1 . (4.26) − dz 2 1+ε 2s(−t) Γ (−2ε) (2πi) −i∞ −s Γ (−2z2 ) This integral can be evaluated by closing the contour to the left and taking residues at the points z2 = −1, −2, . . . with summing up this inverse binomial series by the summation formulae of Sect. C.3. As to the integral over the shifted contour, it does not have poles in ε. If we need to expand (4.25) only up to ε0 this integral does not contribute because of the overall Γ (−2ε) in the denominator, so that we are left with the contribution of the residue: F4.6 (s, t, m2 ; 1, 1, 1, 1, d)

−t 1 2iπ d/2 e−γE ε 1−x − ln + O(ε) , (4.27) =− ln 2 2 ε 2 m 1+x (m ) t −s(4m − s) ε where x = 1/ 1 − 4m2 /s, in agreement with [3]. The general MB representation (4.24) can be used to derive an MB representation for the triangle diagram shown in Fig. 4.5. This class of Feynman integrals is obtained from the corresponding box integrals if we set a4 = 0.

4.3 More One-Loop Examples

69

p1 1 3 2 p2 Fig. 4.5. Triangle diagram with the masses m, m, 0 and external momenta on-shell, p21 = p22 = m2 . A dotted line denotes a massless propagator

If we do this blindly in (4.24) we obtain a zero result due to Γ (a4 ) in the denominator. This is, of course, wrong. Let us think of a4 as a complex number and analyse the behaviour in the limit a4 → 0 similarly to what we do when analysing how singularities in ε are generated. We identify the product Γ (a4 + z1 )Γ (−z1 ) responsible for the generation of the singularity when a4 → 0. To reveal this singularity we can take minus residue at the point z1 = 0 and shift the integration contour over z1 . The contribution of the new integral is indeed zero because of the factor 1/Γ (a4 ). The contribution of the residue produces Γ (a4 ) which cancels this factor in the denominator, and we put a4 to zero after that. Changing the numbering 2 ↔ 3, for convenience, we obtain the following onefold MB representation4 for integrals corresponding to Fig. 4.5: (−1)a iπ d/2 Γ (4 − 2ε − a1 − a2 − 2a3 ) Γ (4 − 2ε − a1 − a2 − a3 )Γ (a1 )Γ (a2 )(−s)a+ε−2 +i∞ 2 z m 1 × dz Γ (a + ε − 2 + z)Γ (−z) 2πi −i∞ −s Γ (2 − a1 − a3 − ε − z)Γ (2 − a2 − a3 − ε − z) × . (4.28) Γ (4 − 2ε − a1 − a2 − 2a3 − 2z) Observe that if we want to have a representation for massive propagatortype diagrams by setting a3 = 0 we shall not reduce the number of integrations: there is no Γ (a3 ) in the denominator and, on the other hand, no singularities in the limit a3 → 0 are generated. So, one can simply apply (4.28) with a3 = 0 for this class of diagrams. The general MB representation (4.24) provides in a very similar way a MB representation for another triangle diagram obtained from Fig. 4.4. We shrink the line 3 to a point and obtain Fig. 4.6. The corresponding onefold MB representation takes the form

4

In [11], it was demonstrated that this Feynman integral reduces, for any values of the three indices, to a two-point function in the shifted dimension d − 2a3 .

70

4 Evaluating by MB Representation

p1 2 1 4 p3 Fig. 4.6. Triangle diagram with the masses m, 0, 0 and external momenta on-shell, p21 = p23 = m2 , obtained from the box of Fig. 4.4

(−1)a iπ d/2 Γ (4 − 2ε − a)Γ (a1 )Γ (a2 )Γ (a4 )(m2 )a+ε−2 z +i∞ −t 1 × dz Γ (a + ε − 2 + z)Γ (−z) 2πi −i∞ m2 ×Γ (a2 + z)Γ (a4 + z)Γ (4 − 2ε − a1 − 2a2 − 2a4 − 2z) ,

(4.29)

where t = (p1 + p3 )2 . Among other partial cases of the massive on-shell boxes let us mention the case where a1 = a2 = 0. Then we obtain a massless one-loop propagatortype diagram which is evaluated by (3.6). On the other hand, one can see that to perform the limit a1 , a2 → 0 it is necessary to take two residues in the integrand and somehow compensate the corresponding gamma functions in the denominator. Eventually one arrives at the known result. This procedure is just an additional check for the initial MB representation (4.24). The representation (4.24) can straightforwardly be generalized to various oﬀ-shell cases, similarly to how we obtained the generalizations (4.22) and (4.23). Here are three results which we shall use in Sect. 4.4. For the box of Fig. 4.4 with two massive and two massless lines, two legs on shell, p23 = p24 = m2 , and two legs oﬀ shell we obtain the following fourfold MB representation: +i∞ +i∞ 4 1 (−1)a iπ d/2 (−s)2−a−ε ... dzj Γ (−zj ) Γ (4 − 2ε − a) Γ (al ) (2πi)4 −i∞ −i∞ j=1 (m2 − p21 )z1 (m2 − p22 )z2 (−t)z3 (m2 )z4 Γ (a2 + z1 + z2 + z3 )Γ (a4 + z3 ) (−s)z1 +z2 +z3 +z4 ×Γ (2 − a124 − ε − z1 − z3 − z4 )Γ (2 − a234 − ε − z2 − z3 − z4 ) Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 ) × Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 − 2z4 ) ×Γ (a + ε − 2 + z1 + z2 + z3 + z4 ) . (4.30) ×

For the box of Fig. 4.4 with two legs on shell, p22 = p24 = m2 , and two legs oﬀ shell, we obtain:

4.4 Two-Loop Massless Examples

1 (−1)a iπ d/2 (−s)2−a−ε Γ (4 − 2ε − a) Γ (al ) (2πi)4 ×

(m − 2

+i∞

+i∞

... −i∞

p21 )z1 (m2 − p23 )z2 (−t)z3 (m2 )z4 (−s)z1 +z2 +z3 +z4

−i∞

4

71

dzj Γ (−zj )

j=1

Γ (a2 + z1 + z3 )Γ (a4 + z2 + z3 )

×Γ (2 − a124 − ε − z1 − z2 − z3 − z4 )Γ (2 − a234 − ε − z3 − z4 ) Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 ) × Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 − 2z4 ) ×Γ (a + ε − 2 + z1 + z2 + z3 + z4 ) . p21

p24

(4.31) 2

Finally, for the box of Fig. 4.4 with two legs on shell, = = m , and two legs oﬀ shell, we obtain: +i∞ +i∞ 4 1 (−1)a iπ d/2 (−s)2−a−ε ... dzj Γ (−zj ) Γ (4 − 2ε − a) Γ (al ) (2πi)4 −i∞ −i∞ j=1 (m2 − p23 )z1 (m2 − p22 )z2 (−t)z3 (m2 )z4 Γ (a2 + z2 + z3 )Γ (a4 + z1 + z3 ) (−s)z1 +z2 +z3 +z4 ×Γ (2 − a124 − ε − z1 − z3 − z4 )Γ (2 − a234 − ε − z2 − z3 − z4 ) Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 ) × Γ (4 − a122344 − 2ε − z1 − z2 − 2z3 − 2z4 ) ×Γ (a + ε − 2 + z1 + z2 + z3 + z4 ) . (4.32) ×

4.4 Two-Loop Massless Examples Our ﬁrst two-loop example is the same as Example 3.7: Example 4.7. Non-planar two-loop massless vertex diagram of Fig. 3.13 with p21 = p22 = 0. We are again dealing with two-loop vertex Feynman integrals (3.53). We start with the four-parametric representation (3.57) obtained within the method of Feynman parameters in the previous chapter. Let us turn to the variables ξ3 = ξη, ξ4 = (1 − ξ)η and apply (4.1) to the resulting denominator in the integrand: Γ (a + 2ε − 4) a+2ε−4

[ηξ(1 − ξ) + (1 − η)(ξξ2 (1 − ξ1 ) + (1 − ξ)ξ1 (1 − ξ2 ))] +i∞ dz1 Γ (−z1 )η z1 ξ z1 (1 − ξ)z1 1 = 2πi −i∞ (1 − η)a+2ε−4+z1 Γ (a + 2ε − 4 + z1 ) × a+2ε−4+z1 . [ξξ2 (1 − ξ1 ) + (1 − ξ)ξ1 (1 − ξ2 )] Then we again apply (4.1) to transform the last line of (4.33) into

(4.33)

72

4 Evaluating by MB Representation

1 2πi

+i∞

−i∞

dz2 Γ (a + 2ε − 4 + z1 + z2 )Γ (−z2 )ξ z2 ξ2z2 (1 − ξ1 )z2 . (1 − ξ)a+2ε−4+z1 +z2 ξ1a+2ε−4+z1 +z2 (1 − ξ2 )a+2ε−4+z1 +z2

After that all the integrals over the parameters ξ1 , ξ2 , ξ, η can be evaluated in terms of gamma functions, and we come to the following twofold MB representation of (3.53) with general powers of the propagators: 2 (−1)a iπ d/2 Γ (2 − ε − a35 ) 2 F4.7 (Q ; a1 , . . . , a6 , d) = (Q2 )a+2ε−4 Γ (6 − 3ε − a) Γ (al ) +i∞ +i∞ Γ (2 − ε − a46 ) 1 × dz1 dz2 Γ (a + 2ε − 4 + z1 + z2 ) Γ (4 − 2ε − a3456 ) (2πi)2 −i∞ −i∞ ×Γ (−z1 )Γ (−z2 )Γ (a4 + z2 )Γ (a5 + z2 )Γ (a1 + z1 + z2 ) Γ (2 − ε − a12 − z1 )Γ (4 − 2ε + a2 − a − z2 ) × Γ (4 − 2ε − a1235 − z1 )Γ (4 − 2ε − a1246 − z1 ) ×Γ (4 − 2ε + a3 − a − z1 − z2 )Γ (4 − 2ε + a6 − a − z1 − z2 ) . (4.34) As in Chap. 3 let us evaluate the integral with all indices equal to one. We have d/2 2 iπ 2 F (ε) , (4.35) F4.7 (Q ; 1, . . . , 1, d) = (Q2 )2+2ε with F (ε) =

Γ (−ε)2 V (ε) Γ (−3ε)Γ (−2ε)

and V (ε) =

1 (2πi)2

+i∞

−i∞

+i∞

−i∞

dz1 dz2 Γ (2 + 2ε + z1 + z2 )Γ (1 + z1 + z2 )

× Γ (1 + z2 )2 Γ (−z1 )Γ (−z2 )

Γ (−ε − z1 ) Γ (−2ε − z1 )2

× Γ (−1 − 2ε − z2 )Γ (−1 − 2ε − z1 − z2 )2 .

(4.36)

After the useful change of variables z1 → −1 − z1 − z2 , we obtain +i∞ +i∞ Γ (1 + z1 + z2 )Γ (1 − ε + z1 + z2 ) 1 V (ε) = dz1 dz2 2 (2πi) −i∞ −i∞ Γ (1 − 2ε + z1 + z2 )2 × Γ (−2ε + z1 )2 Γ (−z1 )Γ (1 + 2ε − z1 ) × Γ (1 + z2 )2 Γ (−1 − 2ε − z2 )Γ (−z2 ) .

(4.37)

The analysis of the integrand shows that the poles in ε are generated by the two products Γ (−2ε + z1 )2 Γ (−z1 ) and Γ (1 + z2 )2 Γ (−1 − 2ε − z2 ) so that the situation is somehow factorized and we can proceed like in the onedimensional cases taking care of the integrations over z1 and z2 separately. So, let us ﬁrst deal with the ﬁrst pole of Γ (−1 − 2ε − z2 ). We have minus the residue at z2 = −1 − 2ε,

4.4 Two-Loop Massless Examples

F1 (ε) =

Γ (1 + 2ε)Γ (−2ε)Γ (−ε)2 1 Γ (−3ε) 2πi

×

73

+i∞

−i∞

dz1 Γ (1 + 2ε − z1 )

Γ (−2ε + z1 )3 Γ (−3ε + z1 )Γ (−z1 ) , Γ (−4ε + z1 )2

(4.38)

and the integral F0 (ε) with the opposite nature of the ﬁrst pole at z2 = −1 − 2ε. For (4.38), we analyse how singularities in ε are generated. The situation is quite familiar and we come to the conclusion that they come from the product Γ (−2ε + z1 )3 Γ (−3ε + z1 )Γ (−z1 ). We take residues at the points z1 = 2ε and z1 = 3ε and turn to the integral F10 with the same integrand as (4.38) but with the opposite nature of these poles. The sum of these two residues gives, in expansion in ε, 1 π2 211ζ(3) π 4 −2γE ε + − 2 − F11 = e + O(ε) . (4.39) ε4 ε 6ε 80 The integral F10 can be evaluated by expanding the integrand in ε and subsequently closing the contour to the right and summing up a series of residues. Here one can apply summation formulae of Appendix C for summing up this number series. The result is 2 π 3ζ(3) 41π 4 − + F10 = e−2γE ε + O(ε) . (4.40) 4ε2 ε 48 Now we have to calculate (4.37) with the opposite nature of the ﬁrst pole of Γ (−1 − 2ε − z2 ). Let us take care of the ﬁrst pole of Γ (−2ε + z1 )2 . We take the residue at this point which is an integral F01 over z2 without gluing of poles of diﬀerent nature and thereby can be evaluated directly in expansion in ε. The resulting expanded integral is evaluated similarly to F10 . We obtain 9ζ(3) 31π 4 π2 −2γE ε + F01 = e − 2+ + O(ε) . (4.41) 4ε 2ε 60 The remaining piece is the integral F00 with the integrand of (4.37) where the ﬁrst poles of Γ (−2ε+z1 )2 and Γ (−1−2ε−z2 ) have changed their nature. There is no gluing anymore so that we can expand the integrand in ε: +i∞ +i∞ 6 dz1 dz2 Γ (z1 )2 Γ (−z1 )Γ (1 − z1 ) F00 = (2πi)2 −i∞ −i∞ × Γ (1 + z2 )2 Γ (−1 − z2 )Γ (−z2 ) + O(ε) ,

(4.42)

where the integration contours are at −1 < Rez1,2 < 0. The integral is a product of one-dimensional MB integrals which can be evaluated by the same procedure as above. We obtain π2 + O(ε) . (4.43) 6 Summing up the four pieces (4.39), (4.40), (4.41) and (4.43) we reproduce the result (3.74) obtained in [15]. F00 = −

74

4 Evaluating by MB Representation

p1

1 2

p2

p3

6 7

3

5 4

p4

Fig. 4.7. Double box

Let us now consider Example 4.8. Massless on-shell planar double box diagram of Fig. 4.7. As in Example 4.3. we have p2i = 0, i = 1, 2, 3, 4. Let us consider double boxes with the irreducible numerator (k +p1 +p2 +p4 )2 and the routing of the external momenta as in [2]. Then the general double box Feynman integral takes the form dd k dd l K(s, t; a1 , . . . , a8 , ε) = 2 a 1 (k ) [(k + p1 )2 ]a2 [(k + p1 + p2 )2 ]a3 [(k + p1 + p2 + p4 )2 ]−a8 × , (4.44) 2 a [(l + p1 + p2 ) ] 4 [(l + p1 + p2 + p4 )2 ]a5 (l2 )a6 [(k − l)2 ]a7 As usual, we consider the factor corresponding to the irreducible numerator as an extra propagator but, really, we are interested only in non-positive integer values of a8 . In fact, there are two possible independent irreducible numerators but the derivation of the MB representation is simple only when we take one of them into account. In order to derive a MB representation for (4.44) it is possible to start from the alpha representation and then apply (4.1) to the corresponding functions U and V. This is not, however, an optimal way. In particular, this was done in the ﬁrst calculation of the master double box [23] but a resulting MB representation turned out to be ﬁvefold, with essential complications in the calculations. We will see that one can proceed using a fourfold MB representation. Let us mention, however, that in the case of non-planar onshell double boxes it was possible to achieve [33] the minimal number of integrations equal to four starting from the global alpha representation. So, we follow (as in [2]) the strategy of [35], where MB integrations were, ﬁrst, introduced, in a suitable way, after the integration over one of the loop momenta, l, and complete this procedure after the integration over the second loop momentum, k. To do this, let us observe that (4.44) can be represented as dd k [(k + p1 + p2 + p4 )2 ]−a8 K(s, t; a1 , . . . , a8 , ε) = (k 2 )a1 [(k + p1 )2 ]a2 [(k + p1 + p2 )2 ]a3 ×F4.4 (s, (k + p1 + p2 + p4 )2 , k 2 , (k + p1 + p2 )2 ; a6 , a7 , a4 , a5 , d) , (4.45) where the integral of four propagators dependent on l has been recognized as the box with two legs oﬀ shell. Then we can use (4.22). After inserting it into

4.4 Two-Loop Massless Examples

75

(4.45) we obtain the massless on-shell box with the indices a1 −z2 , a2 , a3 , a8 − z4 for which we apply our representation (4.11). After these straightforward manipulations, we change the variables z2 → z2 −z4 , z3 → z3 −z4 , z4 → z1 +z4 , and arrive at the following fourfold MB representation of (4.44) (see also [2]): d/2 2 iπ (−1)a K(s, t; a1 , . . . , a8 , ε) = a−4+2ε l=2,4,5,6,7 Γ (al )Γ (4 − a4567 − 2ε)(−s) z1 +i∞ +i∞ 4 t 1 × ... dzj Γ (a2 + z1 )Γ (−z1 ) 4 (2πi) −i∞ s −i∞ j=1 Γ (z2 + z4 )Γ (z3 + z4 )Γ (a1238 − 2 + ε + z4 )Γ (a7 + z1 − z4 ) Γ (a1 + z3 + z4 )Γ (a3 + z2 + z4 )Γ (4 − a1238 − 2ε + z1 − z4 ) Γ (a8 − z2 − z3 − z4 )Γ (a5 + z1 + z2 + z3 + z4 )Γ (−z1 − z2 − z3 − z4 ) × Γ (a8 − z1 − z2 − z3 − z4 ) ×Γ (a4567 − 2 + ε + z1 − z4 )Γ (2 − a128 − ε + z2 )Γ (2 − a238 − ε + z3 ) ×Γ (2 − a567 − ε − z1 − z2 )Γ (2 − a457 − ε − z1 − z3 ) . (4.46) ×

Let us apply (4.46) to the evaluation, in expansion in ε up to the ﬁnite part, of the double box without numerator and with all powers of the propagators equal to one. We know in advance that it has poles up to the fourth order in ε, due to IR and collinear divergences. In fact, at least the highest pole can be predicted without calculation. Representation (4.46) gives d/2 2 iπ F (x, ε) , (4.47) K(s, t; 1, . . . , 1, 0, ε) = − (−s)3+2ε where x = t/s and F (x, ε) =

1 1 Γ (−2ε) (2πi)4

+i∞

+i∞

... −i∞

−i∞

4

dzj xz1

j=1

Γ (1 + z1 )Γ (−z1 )Γ (−1 − ε − z1 − z2 )Γ (−1 − ε − z1 − z3 ) Γ (1 + z2 + z4 )Γ (1 + z3 + z4 )Γ (1 − 2ε + z1 − z4 ) ×Γ (2 + ε + z1 − z4 )Γ (1 + z1 + z2 + z3 + z4 )Γ (1 + z1 − z4 )

×

×Γ (z2 + z4 )Γ (z3 + z4 )Γ (−ε + z2 )Γ (−ε + z3 ) ×Γ (1 + ε + z4 )Γ (−z2 − z3 − z4 ) .

(4.48)

Observe that, because of the presence of the factor Γ (−2ε) in the denominator, we are forced to take some residue in order to arrive at a non-zero result at ε = 0, so that the integral is eﬀectively threefold. Here is an example of the procedure of generating poles in the integral (4.48). The product Γ (−1 − ε − z1 − z2 )Γ (−ε + z2 ) generates, due to the integration over z2 , a pole of the type Γ (−1 − 2ε − z1 ). Then the product of this gamma function with Γ (1 + z1 ) generates a pole of the type Γ (2ε) due to the integration over z1 .

76

4 Evaluating by MB Representation

After such a preliminary analysis we conclude that the key gamma functions that are responsible for the generation of poles in ε are Γ (−ε + z2 ), Γ (−ε + z3 ) and Γ (1 + z1 − z4 ). This gives a hint for the construction of a complete procedure of the resolution of the singularities in ε, with the goal to decompose the given integral into pieces where the Laurent expansion of the integrand in ε becomes possible. One can proceed as follows. We ﬁrst take care of the gamma functions Γ (−ε + z2 ) and Γ (−ε + z3 ), i.e. take residues at z2 = ε and z3 = ε and shift contours across these poles. As a result, (4.48) is decomposed as F = F11 + F10 + F01 + F00 , where F11 corresponds to taking the two residues, F00 is deﬁned by the same expression (4.48) but with both ﬁrst poles of the selected two gamma functions treated in the opposite way, and the two intermediate contributions deﬁned by taking one of the residues and changing the nature of the ﬁrst pole of the other gamma function. The contribution F11 takes the form +i∞ +i∞ 1 1 F11 = dz1 dz4 xz1 Γ (1 + z1 ) Γ (−2ε) (2πi)2 −i∞ −i∞ ×Γ (−1 − 2ε − z1 )2 Γ (−z1 )Γ (1 + z1 − z4 )Γ (2 + ε + z1 − z4 ) Γ (1 + 2ε + z1 + z4 ) . (4.49) ×Γ (ε + z4 )2 Γ (−2ε − z4 ) Γ (1 − 2ε + z1 − z4 )Γ (1 + ε + z4 ) The contributions F10 and F01 are equal to each other because of the symmetrical dependence of the integrand on z2 and z3 . We have +i∞ +i∞ +i∞ 1 1 dz1 dz2 dz4 xz1 Γ (1 + z1 ) F01 = Γ (−2ε) (2πi)3 −i∞ −i∞ −i∞ ×Γ (−1 − 2ε − z1 )Γ (−z1 )Γ (−1 − ε − z1 − z2 )Γ ∗ (−ε + z2 ) Γ (1 + z1 − z4 )Γ (2 + ε + z1 − z4 )Γ (ε + z4 )Γ (z2 + z4 ) × Γ (1 − 2ε + z1 − z4 )Γ (1 + z2 + z4 ) ×Γ (1 + ε + z1 + z2 + z4 )Γ (−ε − z2 − z4 ) , (4.50) where the ﬁrst pole of Γ (−ε + z2 ) is of the opposite nature. We indicate this by asterisk, as in Appendix D. For all these contributions, further decompositions are necessary. One can proceed as follows. In the case of F11 , take care of Γ (−1 − 2ε − z1 ). We decompose F11 as F111 + F110 , where the additional index 1 corresponds to the residue at z1 = −1 − 2ε (with the minus sign) and 0 to the integral where the ﬁrst pole of Γ (−1 − 2ε − z1 ) is left. Take care of Γ (z4 ) and Γ (z4 + ε) by decomposing F111 as F111 = F1111 + F1110 , where the additional index 1 corresponds to the residues at z4 = 0 and z4 = ε given by an explicit expression in terms of gamma and psi functions, and 0 to the one-dimensional MB integral where the ﬁrst pole of each of these gamma functions is right. For F110 , take care of Γ (z4 + ε) to obtain F110 = F1101 + F1100 , where 1 denotes the residue at z4 = −ε. The F1101 is a one-dimensional MB integral

4.4 Two-Loop Massless Examples

77

over z1 which is calculated by expanding in ε. The F1100 starts from ε1 and therefore gives a zero contribution. For F01 , take care of Γ (−1 − 2ε − z1 ) and obtain the decomposition F01 as F011 + F010 similar to the case of F11 . For F011 , let us consecutively take care of the ﬁrst poles of the gamma functions Γ (z2 + z4 ) and Γ (z2 + z4 − ε) with respect to the variable z2 and obtain F011 = F0111 + F0112 + F0110 , where 1 denotes the residue at z2 = −z4 , 2 denotes the residue at z2 = ε − z4 and 0 denotes the integral with ﬁrst poles of these gamma functions to be right. Then we obtain F0111 = F01111 + F01110 , similarly taking care of Γ (ε + z4 )2 , F0112 = F01121 + F01120 taking care of Γ (ε + z4 )Γ (z4 ), and F0110 = F01101 + F01100 taking care of Γ (ε + z4 ). For F010 , we turn to the decomposition F010 = F0101 +F0100 where 1 stands for the residue at z4 = −z2 and 0 for the integral with the ﬁrst right pole of Γ (z2 + z4 ). Finally, we turn to F0101 = F01011 + F01010 , where 1 stands for the residue at z2 = −1 − ε − z1 and 0 for the integral with the ﬁrst left pole of Γ (−1 − ε − z1 − z2 ). For F00 , we take care of the ﬁrst poles of the gamma functions Γ (−1 − ε − z1 − z2 ) and Γ (−1 − ε − z1 − z3 ). The only non-zero contribution arises when taking both residues. As a result we obtain either explicit expressions in terms of gamma functions and their derivatives, or one-dimensional integrals over straight lines parallel to the imaginary axis of ratios of gamma functions which can be of two types: integrals over z1 or some other z-variable. The integrals over z1 can be calculated by closing the contour to the right, taking residues at the points z1 = 0, 1, 2, . . . and summing up resulting series with the help of the table of formulae [14] presented in Appendix C. The one-dimensional MB integrals over z2 or z3 or z4 can be calculated with the help of formulae of Appendix D which are all corollaries of the ﬁrst and the second Barnes lemma (D.1) and (D.47). For example, this is the twofold MB integral that appears in F01100 : +i∞ +i∞ 1 dz2 dz4 Γ ∗ (z2 )Γ (−z2 )Γ (1 + z4 )Γ (−z4 ) (2πi)2 −i∞ −i∞ ×

Γ ∗ (z2 + z4 )2 Γ (−z2 − z4 ) , Γ (1 + z2 + z4 )

(4.51)

where asterisks denote, as in Appendix D, the opposite nature of the ﬁrst poles of the corresponding gamma functions, i.e. the poles z2 = 0 and z4 = −z2 are considered right here. The internal integral over z4 is then evaluated with the help of (D.51), with λ1 = 1, λ2 = z2 , λ3 = 0, λ4 = 1 + z2 , and a resulting onefold MB integral is evaluated as other integrals of this kind. Collecting all the contributions we reproduce the result of [23]: d/2 −γ ε 2 iπ e E t f ;ε , (4.52) K(s, t; 1, . . . , 1, 0, ε) = − (−s)2+2ε t s where

78

4 Evaluating by MB Representation

4 5 ln x 5 2 1 2 π + − 2 ln x − ε4 ε3 2 ε2 1 2 3 11 65 ln x + π 2 ln x − ζ(3) − 3 2 3 ε 4 4 29 88 + ln x + 6π 2 ln2 x − ζ(3) ln x + π 4 3 3 30 2

2 − 2 Li3 (−x) − 2 ln x Li2 (−x) − ln x + π 2 ln(1 + x) ε −4 [S2,2 (−x) − ln x S1,2 (−x)] + 44 Li4 (−x)

f (x, ε) = −

−4 [ln(1 + x) + 6 ln x] Li3 (−x) 10 2 2 +2 ln x + 2 ln x ln(1 + x) + π Li2 (−x) 3 2 2 2 + ln x + π ln (1 + x) 2

− 4 ln3 x + 5π 2 ln x − 6ζ(3) ln(1 + x) + O(ε) . (4.53) 3 This result is in agreement with the leading behaviour in the (Regge) limit t/s → 0 obtained in [32] by use of the strategy of expansion by regions [4, 27, 30]. Keeping the two leading powers of x we have 1 5 ln x 5 4 f (x, ε) = − 4 + 3 − 2 ln2 x − π 2 ε ε 2 ε2 1 2 3 11 65 ln x + π 2 ln x − ζ(3) − 3 2 3 ε 29 4 4 88 + ln x + 6π 2 ln2 x − ζ(3) ln x + π 4 3 3 30 1 2 ln x − 2 ln x + π 2 + 2 +2x ε 1 − 4 ln3 x + 3 ln2 x + (5π 2 − 36) ln x + 2[33 + 5π 2 − 3ζ(3)] 3 +O(x2 ln3 x, ε) .

(4.54)

Using known formulae that relate polylogarithms and generalized polylogarithms with arguments z and 1/z [12, 20, 21] one can rewrite this and similar results for the master double boxes in terms of the same class of functions depending on the inverse ratio s/t. Let us now illustrate the point discussed in the end of Sect. 4.2. The general fourfold representation (4.46) contains a lot of information. In particular, it is very easy to derive MB representations for the two classes of Feynman integrals corresponding to the graphs shown in Fig. 4.8. The integrals for the box with a one-loop insertion are obtained from the double box integrals at a4 = a6 = 0. (For simplicity, we consider the case a8 = 0.) There are Γ (a4 ) and Γ (a6 ) in the denominator of (4.46) but, of course, the limit a4 , a6 → 0 is not zero. Indeed, we can distinguish the product

4.4 Two-Loop Massless Examples 1

p1 2

p2

6

p3 7

3

(a)

5

2

p4

79

7

5

3

(b)

Fig. 4.8. Boxes with a one-loop insertion (a) and boxes with a diagonal (b) obtained from Fig. 4.7

Γ (a4567 − 2 + ε + z1 − z4 )Γ (2 − a567 − ε − z1 − z2 )Γ (z2 + z4 ) which generates, due to integration over z2 and z4 , the singularity of the type Γ (a4 ) – remember our discussion in Sect. 4.2. So, to perform this limit we take a residue at z4 = −z2 and minus residue at z2 = 2 − a567 − ε − z1 and then set a4 = 0. We still have Γ (a6 ) in the denominator, but there is also the product Γ (a567 − 2 + ε + z1 + z3 )Γ (2 − a57 − ε − z1 − z3 ) which generates the singularity of the type Γ (a6 ). Therefore, we take minus residue at z3 = 2 − a57 − ε − z1 , then set a6 = 0 and arrive at the following onefold MB representation: d/2 2 iπ (−1)a Γ (2 − a5 − ε)Γ (2 − a7 − ε) K(a1 , a2 , a3 , 0, a5 , 0, a7 , 0) = Γ (al )Γ (4 − a57 − 2ε)Γ (6 − a − 3ε) z +i∞ t 1 1 dz Γ (a − 4 + 2ε + z)Γ (a57 − 2 + ε + z) × (−s)a−4+2ε 2πi −i∞ s ×Γ (a2 + z)Γ (4 − a1257 − 2ε − z)Γ (4 − a2357 − 2ε − z)Γ (−z) . (4.55) The integrals for the box with a diagonal are obtained from the double box integrals at a1 = a4 = 0. We start from the limit a4 → 0 as in the previous case. Then we observe that there is no Γ (a1 ) in the denominator and no gluing of right and left poles when a1 → 0. So, we just set a1 = 0. After that the integration over z3 involves only four gamma functions Γ (2 − a23 − ε + z3 )Γ (a5 + z1 + z3 )Γ (2 − a57 − ε − z1 − z3 )Γ (−z3 ) . The integral is evaluated by the ﬁrst Barnes lemma (D.1), and we obtain d/2 2 iπ Γ (2 − a23 − ε)Γ (2 − a56 − ε) K(0, a2 , a3 , 0, a5 , a6 , a7 , 0) = Γ (al )Γ (4 − a237 − 2ε)Γ (4 − a567 − 2ε) z +i∞ a t 1 (−1) Γ (2 − a7 − ε) dz Γ (a − 4 + 2ε + z) × Γ (6 − a − 3ε)(−s)a−4+2ε 2πi −i∞ s ×Γ (a2 + z)Γ (a5 + z)Γ (−z) ×Γ (4 − a2357 − 2ε − z)Γ (4 − a2567 − 2ε − z) .

(4.56)

80

4 Evaluating by MB Representation

So, these two classes of integrals are rather simple because they are given only by onefold MB representations. Each of them can be evaluated by decomposing the integral into ‘singular’ and ‘regular’ parts. The singular parts correspond to the residues necessary to reveal the singular behaviour in ε while the regular parts are given by integrals where expansion in ε in the integrand is possible. For the boxes with a one-loop insertion, the singular part is written as minus the sum of the residues of the integrand at the points j−2ε, with j = −max{a1 , a3 } − a257 + 4, . . . , −1, plus the sum of the residues of the integrand at the points j − 2ε for j = 0, . . . , 4 − a. For the diagonal crossed boxes, the singular part is written as minus the sum of the residues of the integrand at the points j − 2ε, with j = −max{a3 , a6 } − a257 + 4, . . . , −1, plus the sum of the residues of the integrand at the points j−2ε for j = 0, . . . , 4−a. The regular parts can be written as MB integrals for −1 < Re z < 0 with an integrand expanded in a Laurent series in ε up to a desired order. Then these integrals are straightforwardly evaluated by closing the contour of integration to the right and taking residues at the points z = 0, 1, 2, . . .. At this step, one can use the collection of formulae for summing up series presented in Appendix C. The evaluation of both the singular and the regular parts can easily be implemented on a computer. Let us, for example, present an analytical result [32] for the box with a diagonal with all indices equal to one: d/2 −γ ε 2 iπ e E F0 (s, t, ε) , (4.57) K(s, t; 0, 1, 1, 0, 1, 1, 1, 0, ε) = − s+t where 1 F0 (s, t, ε) = − ln2 x + π 2 2ε2

+ 2Li3 (−x) − 2 ln x Li2 (−x) − ln2 x + π 2 ln(1 + x) 1 2 3 2 2 + ln x + ln(−s) ln x + π ln(−t) − 2ζ(3) 3 ε +4 (S2,2 (−x) − ln xS1,2 (−x)) − 4Li4 (−x) +4 (ln(1 + x) − ln(−s)) Li3 (−x) +2 ln2 x + 2 ln(−s) ln x − 2 ln x ln(1 + x) Li2 (−x) 2 3 2 2 ln x + ln(−s) ln x + π ln(−t) − 2ζ(3) ln(1 + x) +2 3 1 4 − ln2 x + π 2 ln2 (1 + x) − ln4 x − ln(−s) ln3 x 2 3 11 2 2 − ln (−s) + π ln2 x − π 2 ln2 (−s) − 2π 2 ln(−s) ln x 12 π4 , (4.58) +4ζ(3) ln(−t) − 20 and x = t/s.

4.5 Two-Loop Massive Examples

81

Concerning non-trivial checks of general formulae discussed in the end of Sect. 4.2 let us observe that, if we start from (4.46), we have to obtain, in the limit a1,3,4,6 → 0 with a8 = 0, the massless sunset diagram with the indices a2 , a5 , a7 . Indeed, we can start from (4.55) and perform the limit a3 → 0 by taking minus the residue at z1 = 4 − a1257 − 2ε in order to take into account the singularity of the integral of Γ (a − 4 + 2ε + z1 )Γ (4 − a1257 − 2ε − z1 ). Then we can set a1 = 0 and reproduce a known result. On the other hand, we should obtain the product of two one-loop massless propagator-type integrals with the indices (a1 , a3 ) and (a4 , a6 ) in the limit a2,5,7 → 0 with a8 = 0. Yes, we do this by a similar analysis and similar manipulations: take minus residue at z1 = 0 and set a2 = 0, then take minus residue at z4 = −z2 − z3 and set a5 = 0, then take residues at z2 = 0 and z3 = 0 and set a7 = 0. Representation (4.46) can be used for the evaluation, in expansion in ε, of any massless on-shell planar double box. See, e.g., [2] where it was applied to the evaluation of a double box with a numerator, K(s, t; 1, . . . , 1, −1, ε). Let us mention that, in this case, one meets a spurious singularity in MB integrals which can be cured by introducing an auxiliary analytic regularization. To do this, one can choose a8 = −1 + λ. Then the singularities in the MB integrals are ﬁrst resolved with respect to λ and then with respect to ε when λ and ε tend to zero. (The singularities in λ are indeed cancelled.) Non-planar double boxes can also be evaluated by MB representation – see [33].

4.5 Two-Loop Massive Examples Our next two-loop example is Example 4.9. Massive on-shell double box diagrams shown in Figs. 4.9 and 4.10. p1

1 2

p2

p3

6 7

3

5 4

(a)

p4 (b)

Fig. 4.9. Planar massive on-shell double boxes: (a) ﬁrst type, (b) second type. The solid lines denote massive, the dotted lines massless particles

This is an important class of Feynman integrals with one more parameter, with respect to the massless on–shell double boxes. In particular, it is relevant for Bhabha scattering.

82

4 Evaluating by MB Representation 6 1 5

2

7

3 4 Fig. 4.10. Non-planar massive on-shell double box

The general double box Feynman integral of the ﬁrst type (see Fig. 4.9a) takes the form dd k dd l 2 BPL,1 (s, t, m ; a1 , . . . , a8 , ε) = (k 2 − m2 )a1 [(k + p1 )2 ]a2 [(k + p1 + p2 + p4 )2 ]−a8 × [(k + p1 + p2 )2 − m2 ]a3 [(l + p1 + p2 )2 − m2 ]a4 [(l + p1 + p2 + p4 )2 ]a5 1 × 2 , (4.59) (l − m2 )a6 [(k − l)2 ]a7 where we consider a (non-negative) power −a8 of the factor (k +p1 +p2 +p4 )2 in the numerator as in the massless case. To derive an appropriate MB representation for (4.59) we proceed similarly to the massless case, i.e. recognize the internal integral over l as a massive box with two legs oﬀ-shell for which we use representation (4.30). After that the integral over k can be recognized as the massive on-shell box represented by (4.24), and we obtain the following sixfold MB representation [26]: d/2 2 iπ (−1)a (−s)4−a−2ε 2 BPL,1 (s, t, m ; a1 , . . . , a8 , ε) = j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε) 2 z1 +z5 w +i∞ +i∞ 5 m t 1 ... dw dzj Γ (a2 + w)Γ (−w) × (2πi)6 −i∞ −s s −i∞ j=1 Γ (z2 + z4 )Γ (z3 + z4 )Γ (4 − a13 − 2a28 − 2ε + z2 + z3 )Γ (a7 + w − z4 ) Γ (a1 + z3 + z4 )Γ (a3 + z2 + z4 ) Γ (a1238 − 2 + ε + z4 + z5 )Γ (a4567 − 2 + ε + w + z1 − z4 ) × Γ (4 − a46 − 2a57 − 2ε − 2w − 2z1 − z2 − z3 ) Γ (a8 − z2 − z3 − z4 )Γ (−w − z2 − z3 − z4 )Γ (2 − a238 − ε + z3 − z5 ) × Γ (4 − a1238 − 2ε + w − z4 )Γ (a8 − w − z2 − z3 − z4 ) Γ (a5 + w + z2 + z3 + z4 )Γ (2 − a567 − ε − w − z1 − z2 ) × Γ (4 − a13 − 2a28 − 2ε + z2 + z3 − 2z5 ) ×Γ (2 − a457 − ε − w − z1 − z3 )Γ (2 − a128 − ε + z2 − z5 ) (4.60) ×Γ (4 − a46 − 2a57 − 2ε − 2w − z2 − z3 )Γ (−z1 )Γ (−z5 ) . ×

4.5 Two-Loop Massive Examples

83

This general formula can be used to evaluate various Feynman integrals of the given family. Let us consider the example of the Feynman integral without numerator and ai = 1 for i = 1, 2, . . . , 7. Then (4.60) takes the form d/2 2 iπ (0) 2 2 B (s, t, m , ε) ≡ BPL,1 (s, t, m ; 1, . . . , 1, 0, ε) = − Γ (−2ε)(−s)3+2ε +i∞ +i∞ 5 z +z w m2 1 5 t 1 × ... dw dzj (2πi)6 −i∞ −s s −i∞ j=1 Γ (1 + w)Γ (−w)Γ (2 + ε + w + z1 − z4 )Γ (−1 − ε − w − z1 − z2 ) Γ (1 − 2ε + w − z4 )Γ (1 + z2 + z4 )Γ (1 + z3 + z4 ) Γ (−1 − ε − w − z1 − z3 )Γ (−z1 )Γ (−ε + z2 − z5 )Γ (−ε + z3 − z5 ) × Γ (−2ε + z2 + z3 − 2z5 )Γ (−2 − 2ε − 2w − 2z1 − z2 − z3 ) ×Γ (1 + ε + z4 + z5 )Γ (−z5 )Γ (−2ε + z2 + z3 )Γ (1 + w − z4 ) ×Γ (1 + w + z2 + z3 + z4 )Γ (−2 − 2ε − 2w − z2 − z3 )

×

×Γ (z2 + z4 )Γ (z3 + z4 )Γ (−z2 − z3 − z4 ) .

(4.61)

Observe that, because of the presence of the factor Γ (−2ε) in the denominator, we are forced to take some residue in order to arrive at a non-zero result at ε = 0, so that the integral is eﬀectively ﬁvefold. Let us apply our strategy of shifting contours and taking residues, with the goal to decompose (4.61) into pieces where the Laurent expansion ε of the integrand becomes possible. We shall evaluate this integral in expansion in ε up to a ﬁnite part. We know in advance that the poles in ε are now only of the second order because collinear divergences are absent. This is how such procedure can be performed in this case [26]: 1. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then a residue at z4 = 0, then a residue at z2 = 0, expand in a Laurent series in ε up to a ﬁnite part. Let us denote the resulting integral over z1 and z5 by B1 . 2. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then a residue at z4 = 0, and change the nature of the ﬁrst pole of Γ (z2 ) (choose a contour from the opposite side, i.e. the pole z2 will be now right), then expand in ε. Denote this integral over z1 , z2 and z5 by B2 . 3. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then change the nature of the ﬁrst pole of Γ (z4 ), then take a residue at z2 = −z4 , then take a residue at z4 = −ε and expand in ε. This resulting integral over z1 and z5 is denoted by B3 . 4. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then change the nature of the ﬁrst pole of Γ (z4 ), then take a residue at z2 = −z4 , then change the nature of the ﬁrst pole of Γ (2(ε + z4 )) and expand in ε. The resulting integral over z1 , z4 and z5 is denoted by B4 .

84

4 Evaluating by MB Representation

5. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then minus residue at w = −1 − 2ε, then change the nature of the ﬁrst pole of Γ (z4 ), then change the nature of the ﬁrst pole of Γ (z2 + z4 ) and expand in ε. The resulting integral over z1 , z2 , z4 and z5 is denoted by B5 . 6. Take minus residue at z3 = −2 − 2ε − 2w − z2 , then change the nature of the ﬁrst pole of Γ (−2(1 + 2ε + w)), then take minus residue at z4 = 1 + w, then minus residue at z2 = −1 − 2ε − w and expand in ε. The resulting integral over w, z1 and z5 is denoted by B6 . 7. Change the nature of the ﬁrst pole of Γ (−2 − 2ε − 2w − z2 − z3 ), then take minus residue at z4 = −z2 − z3 , then a residue at z3 = 2ε − z2 , then take a residue at z2 = 2ε and expand in ε. The resulting integral over w, z1 and z5 is denoted by B7 . One can see that all the other contributions vanish at ε = 0. By a suitable change of variables, one can observe that B7 = B6 . In fact, the dependence of the ﬁrst ﬁve contributions on the Mandelstam variable t is trivial: they are just proportional to 1/t. The two-dimensional integrals B1 and B3 are products of one-dimensional integrals which can be evaluated by closing the contour to the left and summing up resulting series with the help of formulae [11] of Appendix C. To evaluate the three-parametric integral B4 it is reasonable to observe that the integrand only changes its sign after the transformation {z4 → −z4 , z1 → z5 , z5 → z1 }. If we take into account that the change of variables z4 → −z4 implies that the initial integration contour −1 < Rez4 < 0 becomes 0 < Rez4 < 1 we will obtain a simple equation for B4 and conclude that the value of the integral equals 1/2 times the residue at z4 = 0. The latter quantity turns out to be a factorized integral over z1 and z5 which is evaluated like B1 and B3 . The three-dimensional integral B2 is evaluated by closing the integration contours over z1 and z5 to the left, summing up resulting series and applying a similar procedure to a ﬁnal integral in z2 . The corresponding result is naturally expressed in terms of polylogarithms, up to Li3 , depending on s and m2 in terms of the variable "√ √ #2 4m2 − s + −s v= √ . √ 4m2 − s − −s The form of this result provides a hint about a possible functional dependence of the result for the four-dimensional integral B5 , and a heuristic procedure which was explicitly formulated in [14] turns out to be successfully applicable here. First, all the contributions, in particular B4 , are analytic functions of s in a vicinity of the origin. One can observe that any given term of the Taylor expansion can be evaluated straightforwardly because the corresponding integrals over z2 and z4 are taken recursively. It is, therefore, possible to evaluate enough ﬁrst terms (say, 30) of this Taylor expansion.

4.5 Two-Loop Massive Examples

85

Then one takes into account the type of the functional dependence mentioned above, turns to a new Taylor series in terms of the variable v − 1 and assumes that the n-th term of this Taylor series is a linear combination, with unknown coeﬃcients, of the following quantities of levels 1, 2, 3, and 4, respectively: 1 , (4.62) n 1 S1 (n) , (4.63) , n2 n 1 S1 (n) S2 (n) S1 (n)2 , , (4.64) , , n3 n2 n n 1 S1 (n) S2 (n) S1 (n)2 , , , , n4 n3 n2 n2 S3 (n) S12 (n) S1 (n)S2 (n) S1 (n)3 , , , . (4.65) n n n n n where Sk (n) = j=1 j −k , etc. are nested sums (see Appendix C). Using the information about the ﬁrst terms of the Taylor series one solves a system of linear equations, ﬁnds those unknown coeﬃcients and checks this solution with the help of the next Taylor coeﬃcients. This experimental mathematics has turned out to be quite successful for the evaluation of B5 . Finally, the contribution B6 is a product of a onedimensional integral over z1 , which is easily evaluated, and a two-dimensional integral over w and z5 which involves a non-trivial dependence on t and is evaluated by closing the integration contour in z5 to the left, summing up a resulting series in terms of Gauss hypergeometric function for which one can apply the parametric representation (B.5). After that the internal integral over w is taken by the same procedure and, ﬁnally, one takes the parametric integral. The ﬁnal result takes the following form [26]: d/2 −γ ε 2 2 iπ e E x (0) 2 B (s, t, m ; ε) = − 2 1+2ε s (−t) b2 (x) b1 (x) + b × + (x) + b (x, y) + O(ε) , (4.66) 01 02 ε2 ε where x = 1/ 1 − 4m2 /s, y = 1/ 1 − 4m2 /t, and b2 (x) = 2(mx − px )2 , 1−x 1+x −2x b1 (x) = −8 Li3 + Li3 + Li3 2 2 1−x 2x 1−x −2x +Li3 + 4(mx − px ) Li2 − Li2 1+x 2 1−x −(4/3)m3x + 4m2x px − 6mx p2x + (2/3)p3x + 4l2 (mx px + p2x )

(4.67)

86

4 Evaluating by MB Representation

−2l22 (mx + 3px ) − (π 2 /3)(4l2 − mx − 3px ) + (8/3)l23 + 14ζ(3) , (4.68) 1+x b01 (x) = −8(mx − px ) Li3 (x) − Li3 (−x) − Li3 2 1−x 2x −2x +Li3 − Li3 + Li3 2 1+x 1−x 1−x +16Li2 (Li2 (x) − Li2 (−x)) 2 " 2 # 1−x 2 2 − 8Li2 (x) Li2 (−x) +4 Li2 (x) + Li2 (−x) + 4Li2 2 1−x 2 2 −(8/3)[π − 6l2 + 6lx px − 6mx (lx + px − 2l2 )]Li2 2 −(4/3)[π 2 − 6l22 + 3m2x + 6mx (2l2 − 2lx − px ) + 12lx px − 3p2x ] 2x ×(Li2 (x) − Li2 (−x)) + 8(mx − px ) (px − mx + 2l2 )Li2 1+x −2x +2(lx − mx + l2 )Li2 − 8(mx − px )(2lx − px − 5mx + 4l2 ) 1−x ×(−mx px + l2 (mx + px ) − l22 + π 2 /6) −(20/3)m4x + (164/3)m3x px − 40m2x p2x − (4/3)mx p3x − (8/3)p4x +8mx lx (m2x − 3mx px + 2p2x ) −4l2 (7m3x + 21m2x px − 4mx lx px − 23mx p2x + 4lx p2x − p3x ) −π 2 ((17/3)m2x − (4/3)mx lx − 2mx px + (4/3)lx px − (7/3)p2x ) +l22 (84m2x − 8mx lx − 16mx px + 8lx px − 44p2x ) −(8/3)l2 (6l22 − π 2 )(3mx − 2px ) − (4/3)π 2 l22 + 4l24 + π 4 /9 .

(4.69)

The last piece of the ﬁnite part comes from B6 and B7 : 1−x 1+x b02 (x, y) = 2(px − mx ) 4 Li3 − Li3 2 2 (1 − x)y −(1 + x)y −(1 − x)y +Li3 − Li3 + Li3 1 − xy 1 − xy 1 + xy (1 + x)y (1 + x)(1 − y) (1 − x)(1 + y) −Li3 + 2 Li3 − Li3 1 + xy 2(1 − xy) 2(1 − xy) (1 − x)(1 − y) (1 + x)(1 + y) −Li3 + Li3 2(1 + xy) 2(1 + xy) +2(my + py − mxy − pxy ) −2x 2x × 2Li2 (x) − 2Li2 (−x) + Li2 − Li2 1−x 1+x 1−x +4(mxy − pxy )(Li2 (−y) − Li2 (y)) − 4(mx + px − 2l2 )Li2 2

4.5 Two-Loop Massive Examples

1−y (1 − x)y − 4(mx + ly − mxy )Li2 2 1 − xy −(1 + x)y +4(px + ly − mxy )Li2 1 − xy −(1 − x)y −4(mx + ly − pxy )Li2 1 + xy (1 + x)y +4(px + ly − pxy )Li2 1 + xy (1 − x)(1 + y) +2(mx + px + my + py − 2mxy − 2l2 )Li2 2(1 − xy) (1 − x)(1 − y) +2(mx + px + my + py − 2pxy − 2l2 )Li2 2(1 + xy)

87

−4(mxy − pxy )Li2

+2p2x (my + py − mxy − pxy ) + 2px (2(my ly + my py + ly py ) +mxy (−my − 2ly − 3py + 3mxy ) + pxy (−3my − 2ly − py + 3pxy )) +2mx (2px + my − 2ly + py )(my + py − mxy − pxy ) − p2y (mxy + pxy ) +2py (2m2xy + p2xy ) + m2y (2py − mxy − pxy ) +2my (p2y + m2xy + 2p2xy − py (3mxy + pxy )) − 2(m3xy + p3xy ) +2l2 ((4my + 4py − 3mxy )mxy + (2my + 2py − 3pxy )pxy −2(px + 2mx )(my + py − mxy − pxy ) − m2y − 4my py − p2y ) +2l22 (3(my + py ) − 2(2mxy + pxy )) −(π 2 /3)(my + py − 8mxy + 6pxy ) .

(4.70)

The following abbreviations are used here: lz = ln z for z = x, y, 2, pz = ln(1 + z) and mz = ln(1 − z) for z = x, y, xy. This result is presented in such a way that it is manifestly real at small negative values of s and t. From this Euclidean domain, it can easily be continued analytically to any other domain. The result (4.66)–(4.70) is in agreement with the leading power behaviour in the (Sudakov) limit of the ﬁxed-angle scattering, m2 |s|, |t| which can be alternatively obtained [26] by use of the strategy of expansion by regions [4, 27]: d/2 −γ ε 2 iπ e E (0) 2 B (s, t, m ; ε) = − 2 s (−t)1+2ε 1 L2

× 2 2 − (2/3)L3 + (π 2 /3)L + 2ζ(3) ε ε −(2/3)L4 + 2 ln(t/s)L3 − 2(ln2 (t/s) + 4π 2 /3)L2

+ 4Li3 (−t/s) − 4 ln(t/s)Li2 (−t/s) + (2/3) ln3 (t/s)

−2 ln(1 + t/s) ln2 (t/s) + (8π 2 /3) ln(t/s) − 2π 2 ln(1 + t/s) + 10ζ(3) L +π 4 /36 + O(m2 L3 , ε) , (4.71)

88

4 Evaluating by MB Representation

where L = ln(−m2 /s). This asymptotic behaviour is reproduced when one starts from the result (4.66)–(4.70). Another check of such a complicated result came from the numerical integration based on a method of sector decompositions in the space of alpha parameters [7] (to be discussed in Sect. E.2). Let us stress that, in the present case with a non-zero mass, there are no collinear divergences and the poles in ε are only up to the second order, so that the resolution of singularities in ε in the MB integrals is relatively simple. Therefore, it looks promising to use the technique presented, starting from (4.60), for the evaluation of any given master integral. For example, the integral BPL,1 (s, t, m2 ; 1, . . . , 1, −1, ε) was evaluated in [31]. There is the same problem as in the massless case discussed above and connected with spurious singularities in MB integrals. It can also be cured in the same way, by introducing an auxiliary analytic regularization, e.g. with a8 = −1+λ. The singularities in the corresponding MB integral are ﬁrst resolved with respect to λ and then with respect to ε when λ and ε tend to zero. In the result [31], one meets not only usual polylogarithms but also a harmonic polylogarithm (HPL) [22] (see Appendix C), H−1,0,0,1 (−(1 − x)/(1 + x)) with x deﬁned after (4.66). Let us turn to the massive double boxes of the second type shown in Fig. 4.9b: dd k dd l BPL,2 (s, t, m2 ; a1 , . . . , a8 , ε) = 2 2 (k − m )a1 [(k + p1 )2 ]a2 [(k + p1 + p2 + p4 )2 ]−a8 × 2 2 a [(k + p1 + p2 ) − m ] 3 [(l + p1 + p2 )2 ]a4 [(l + p1 + p2 + p4 )2 − m2 ]a5 1 × 2 a6 . (4.72) (l ) [(k − l)2 − m2 ]a7 To derive a MB representation for (4.72) let us straightforwardly generalize the derivation of (4.60). For the subintegral over l we now use representation (4.31) of the massive box with two legs oﬀ-shell in the second variant. Then the integral over k can be recognized as the massive on-shell box (4.24). We therefore obtain the following sixfold MB representation [31]: d/2 2 iπ (−1)a (−s)4−a−2ε 2 BPL,2 (s, t, m ; a1 , . . . , a8 , ε) = j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε) 2 z5 +z6 z1 +i∞ +i∞ 6 6 m t 1 . . . dz Γ (−zj ) × j (2πi)6 −i∞ −s s −i∞ j=1 j=1 Γ (a4 + z2 + z4 )Γ (4 − a445667 − 2ε − z2 − z3 − 2z4 )Γ (a6 + z3 + z4 ) Γ (4 − a445667 − 2ε − z2 − z3 − 2z4 − 2z5 )Γ (6 − a − 3ε − z4 − z5 ) Γ (a2 + z1 )Γ (8 − a13 − 2a245678 − 4ε − 2z1 − z2 − z3 − 2z4 − 2z5 ) × Γ (8 − a13 − 2a245678 − 4ε − 2z1 − z2 − z3 − 2z4 − 2z5 − 2z6 ) ×

4.5 Two-Loop Massive Examples

89

Γ (2 − a456 − ε − z4 − z5 )Γ (2 − a467 − ε − z2 − z3 − z4 − z5 ) Γ (a45678 − 2 + ε + z2 + z3 + z4 + z5 )Γ (a1 − z3 )Γ (a3 − z2 ) ×Γ (a4567 + ε − 2 + z2 + z3 + z4 + z5 )Γ (a − 4 + 2ε + z1 + z4 + z5 + z6 )

×

×Γ (4 − a1245678 − 2ε − z1 − z2 − z4 − z5 − z6 ) ×Γ (4 − a2345678 − 2ε − z1 − z3 − z4 − z5 − z6 ) ×Γ (a45678 − 2 + ε + z1 + z2 + z3 + z4 + z5 ) ,

(4.73)

This representation was used in [31] to calculate the master planar double box of the second type BP L,2 (s, t, m2 ; 1, . . . , 1, 0, ε). The resolution of the singularities in ε was performed similar to the previous cases. The number of resulting MB integrals where an expansion in ε can be performed in the integrand is again equal to six. This time, some of the contributions turned out to be hardly evaluated in terms of known functions. Some two-parametric integrals of elementary functions entered the result in [31]. This result was controlled similarly to the previous case, by numerical evaluation of ﬁnite MB integrals and numerical evaluation by the method of [7] (to be discussed in Sect. E.2). We shall come back to the discussion of the problem of the evaluation of the massive on-shell double boxes in Chap. 7. To conclude this section let us turn to the non-planar graph of Fig. 4.10. Its MB representation can again be derived by using an MB representation for the subdiagram consisting of the lines (4, 5, 6, 7). This time, we can use (4.32). For the subsequent integral over the second loop momentum, we need the following MB representation for this auxiliary one-loop integral: dd k 2 2 a 2 1 (k − m ) [(k + p1 ) ]a2 [(k + p1 + p2 )2 − m2 ]a3 1 (−1)a iπ d/2 (−s)2−a−ε = 2 a 2 a [(k + p1 + p2 + p4 ) ] 4 [(k − p4 ) ] 4 Γ (4 − 2ε − a) Γ (al ) +i∞ +i∞ 4 2 z2 z3 z4 1 (m ) (−t) (−u) . . . dz Γ (−z ) × j j (2πi)4 −i∞ (−s)z2 +z3 +z4 −i∞ j=1

×

Γ (a245 + z1 + 2z3 + 2z4 ) Γ (a245 + z1 + z3 + 2z4 ) ×Γ (a2 + a4 + z1 + z3 + z4 )Γ (−a4 − z1 − z3 − z4 )Γ (a4 + z1 + z3 )

×Γ (a + ε − 2 + z2 + z3 + z4 )Γ (a5 + z4 )

×Γ (2 − a1245 − ε − z2 − z3 − z4 )Γ (2 − a2345 − ε − z2 − z3 − z4 ) Γ (4 − a12234455 − 2ε − 2z3 − 2z4 ) , (4.74) × Γ (4 − a12234455 − 2ε − 2z2 − 2z3 − 2z4 ) where u = (p1 + p4 )2 is a Mandelstam variable. It can be derived similarly to the previous MB representations for one-loop Feynman integrals. Using (4.74) one arrives at the following eightfold MB representation [31]:

90

4 Evaluating by MB Representation

iπ d/2

2

(−1)a (−s)4−a−2ε j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε) z +z z +i∞ +i∞ 8 7 m2 5 6 t 7 u z8 1 . . . dz Γ (−zj ) × j (2πi)8 −i∞ −s s s −i∞ j=1 j=1

BNP (s, t, u, m2 ; a1 , . . . , a8 , ε) =

Γ (a5 + z2 + z4 )Γ (a7 + z3 + z4 )Γ (4 − a455677 − 2ε − z2 − z3 − 2z4 ) Γ (a1 − z2 )Γ (a3 − z3 )Γ (a8 − z4 ) Γ (2 − a567 − ε − z2 − z4 − z5 )Γ (2 − a457 − ε − z3 − z4 − z5 ) × Γ (4 − a455677 − 2ε − z2 − z3 − 2z4 − 2z5 ) Γ (a8 + z1 − z4 + z7 )Γ (4 − a2345678 − 2ε − z2 − z5 − z6 − z7 − z8 ) × Γ (6 − a − 3ε − z5 ) Γ (8 − a13 − 2a245678 − 4ε − z2 − z3 − 2z5 − 2z7 − 2z8 ) × Γ (8 − a13 − 2a245678 − 4ε − z2 − z3 − 2z5 − 2z6 − 2z7 − 2z8 ) Γ (4 − a1245678 − 2ε − z3 − z5 − z6 − z7 − z8 ) × Γ (a245678 − 2 + ε + z1 + z2 + z3 + z5 + z7 + 2z8 ) ×Γ (a4567 + ε − 2 + z2 + z3 + z4 + z5 + z8 )Γ (−a8 − z1 + z4 − z7 − z8 ) ×Γ (a245678 − 2 + ε + z1 + z2 + z3 + z5 + 2z7 + 2z8 ) ×

×Γ (a − 4 + 2ε + z5 + z6 + z7 + z8 )Γ (a28 + z1 − z4 + z7 + z8 ) .

(4.75)

Representation (4.75) can be checked for various simple partial cases as it was explained above. Although the number of integrations is rather high one can proceed also in this case. However, it turns out that the massive non-planar case is rather complicated. A description of preliminary results for the master planar double box can be found in [31]. Let us now again illustrate the fact that general MB representations accumulate a lot of information so that MB representations for various classes of Feynman integrals can be derived in a very simple way from an initial global representation. Suppose we want to consider Example 4.10. Sunset diagrams of Fig. 3.12 with one zero mass and two equal non-zero masses at a general value of the external momentum squared. Remember that we have already considered such Feynman integrals at threshold, q 2 = 4m2 – see Example 3.6. There is no need to derive an appropriate MB representation from the beginning. Let us observe that such Feynman integrals, with the massive propagators 5 and 7 and the massless propagator 2, can be obtained from the massive on-shell double boxes of Fig. 4.9b at a1 = a3 = a4 = a6 = 0. As usual such a limit results in taking some residues. We ﬁrst let a4 → 0 and observe that Γ (a4 ) in the denominator can be cancelled only if we take into account the gluing in the product Γ (a4 + z2 + z4 )Γ (−z2 )Γ (−z4 ). Thus we are forced to take the two residues at

4.5 Two-Loop Massive Examples

91

z4 = 0 and z2 = 0. Then the limit a6 → 0 can similarly be taken, because of the presence of Γ (a6 + z3 )Γ (−z3 )/Γ (a6 ), by taking minus residue at z3 = 0. Then we take the limit a1 → 0 by observing that the only way to cancel Γ (a1 ) in the denominator is to take into account the gluing in the product Γ (a123578 − 4 + 2ε + z1 + z5 + z6 )Γ (4 − a23578 − 2ε − z1 − z5 − z6 ) and take a residue, e.g. at z6 = 4 − a23578 − 2ε − z1 − z5 (with the minus sign). Finally, we let a3 → 0 by distinguishing the product Γ (a23578 − 4 + 2ε + z1 + z5 )Γ (8 − a3 − 2a2578 − 4ε − 2z1 − 2z5 ) which generates Γ (a3 ) and cancels this factor in the denominator. After relabelling the lines, substituting t → q 2 and expressing the irreducible numerator in terms of the loop momenta of the sunset diagram, we obtain F4.10 (q 2 , m2 ; a1 , a2 , a3 , a4 , d) dd k dd l [(k + l)2 ]−a4 = (k 2 − m2 )a1 (l2 − m2 )a2 [(q − k − l)2 ]a3 d/2 2 +i∞ 2 z iπ (−1)a Γ (2 − a3 − ε) 1 q = dz Γ (a1 )Γ (a2 )Γ (a3 )(m2 )a−4+2ε 2πi −i∞ m2 Γ (a − 4 + 2ε + z)Γ (a3 + z)Γ (−z)Γ (2 − a34 − ε − z) × Γ (a12 + 2a34 − 4 + 2ε + 2z)Γ (2 − ε + z)Γ (2 − a3 − ε − z) ×Γ (a134 − 2 + ε + z)Γ (a234 − 2 + ε + z) . (4.76) If we evaluate the integral in (4.76) for general ε by closing the contour and taking a series of residues we shall reproduce the result of [8] in terms of the hypergeometric series 4 F3 . We are oriented, however, at the evaluation in expansion in ε and will evaluate integrals (4.76), for concrete values of the indices, by resolving singularities in ε and then closing the contour and summing up the corresponding series. For example, (4.76) gives 2 F4.10 (q 2 , m2 ; 1, 1, 1, 0, d) = − iπ d/2 Γ (1 − ε)(m2 )1−2ε +i∞ 2 z q Γ (2ε − 1 + z)Γ (ε + z)2 Γ (1 + z)Γ (−z) 1 . (4.77) × dz 2 2πi −i∞ m Γ (2ε + 2z)Γ (2 − ε + z) The resolution of the singularities in ε is standard: we distinguish the factor Γ (2ε − 1 + z) as the source of poles. We have to take care of its ﬁrst two poles, i.e. take residues at z = 1 − 2ε and z = −2ε. The calculation of the integral with the opposite nature of these two poles is performed by closing the integration contour to the right and summing up series, with the following result which can be found in [11, 13]: 2 1 1 q2 F4.10 (q 2 , m2 ; 1, 1, 1, 0, d) = iπ d/2 (m2 )1−2ε 2 + 3 − ε 4m2 ε π2 11 13(1 + x2 ) 1 + 2x − x2 + + + + ln x 6 4 8x 2x

92

4 Evaluating by MB Representation

1 − x + x2 2 2 ln x − − ln x + O(ε) , (4.78) 1−x (1 − x)2 where x = ( 4m2 − q 2 − −q 2 )/( 4m2 − q 2 + −q 2 ). (Please, note that the letter x is used in various ways: this is another function in Examples 4.6, 4.9, while, for massless double and triple boxes, this is simply t/s.)

4.6 Three-Loop Examples Our next example is already at three-loop level: Example 4.11. The massless on-shell triple box diagram of Fig. 4.11.

p1

1 2

p2

6 7

3

5 4

p3

10 9 8

p4

Fig. 4.11. Triple box

The general planar triple box Feynman integral without numerator takes the form dd k dd l dd r T (s, t; a1 , . . . , a10 , ε) = (k 2 )a1 [(k + p2 )2 ]a2 [(k + p1 + p2 )2 ]a3 1 × 2 a 4 [(l + p1 + p2 ) ] [(r − l)2 ]a5 (l2 )a6 [(k − l)2 ]a7 1 × . (4.79) 2 a 8 [(r + p1 + p2 ) ] [(r + p1 + p2 + p4 )2 ]a9 (r2 )a10 To derive a suitable MB representation for (4.79) we proceed like in the derivation of (4.46). We recognize the internal integral over the loop momentum r as a box with two legs oﬀ-shell given by (4.22). After inserting it into (4.79) we obtain an MB integral of the on-shell double box with certain indices dependent on MB integration variables. These straightforward manipulations lead [29] to the following sevenfold MB representation of (4.79): d/2 3 iπ (−1)a (−s)6−a−3ε T (s, t; a1 , . . . , a10 , ε) = j=2,5,7,8,9,10 Γ (aj )Γ (4 − a589(10) − 2ε) z1 +i∞ +i∞ 7 t 1 Γ (a2 + z1 )Γ (−z1 )Γ (z2 + z4 ) × ... dzj 7 (2πi) −i∞ s Γ (a1 + z3 + z4 )Γ (a3 + z2 + z4 ) −i∞ j=1

4.6 Three-Loop Examples

93

Γ (2 − a12 − ε + z2 )Γ (2 − a23 − ε + z3 )Γ (a7 + z1 − z4 )Γ (−z5 )Γ (−z6 ) Γ (4 − a123 − 2ε + z1 − z4 )Γ (a6 − z5 )Γ (a4 − z6 ) ×Γ (z3 + z4 )Γ (a123 − 2 + ε + z4 )Γ (z1 + z2 + z3 + z4 − z7 )

×

×Γ (2 − a59(10) − ε − z5 − z7 )Γ (2 − a589 − ε − z6 − z7 ) ×Γ (a467 − 2 + ε + z1 − z4 − z5 − z6 − z7 )Γ (a5 + z5 + z6 + z7 ) ×Γ (4 − a467 − 2ε + z5 + z6 + z7 )Γ (a589(10) − 2 + ε + z5 + z6 + z7 ) ×Γ (2 − a67 − ε − z1 − z2 + z5 + z7 )Γ (a9 + z7 ) ×Γ (2 − a47 − ε − z1 − z3 + z6 + z7 ) , (4.80) 10 where a = i=1 ai , a589(10) = a5 + a8 + a9 + a10 , etc. In the case of the master triple box, we set ai = 1 for i = 1, 2, . . . , 10 to obtain T (0) (s, t, ε) ≡ T (1, . . . , 1; s, t, ε) d/2 3 z1 +i∞ +i∞ 7 iπ t 1 ... dzj Γ (1 + z1 ) = Γ (−2ε)(−s)4+3ε (2πi)7 −i∞ s −i∞ j=1 Γ (−z1 )Γ (−ε + z2 )Γ (−ε + z3 )Γ (1 + z1 − z4 )Γ (−z2 − z3 − z4 ) Γ (1 + z2 + z4 )Γ (1 + z3 + z4 )Γ (1 − 2ε + z1 − z4 ) Γ (z2 + z4 )Γ (z3 + z4 )Γ (−z5 )Γ (−z6 )Γ (z1 + z2 + z3 + z4 − z7 ) × Γ (1 − z5 )Γ (1 − z6 )Γ (1 − 2ε + z5 + z6 + z7 ) ×Γ (2 + ε + z5 + z6 + z7 )Γ (−1 − ε − z5 − z7 )Γ (−1 − ε − z6 − z7 ) ×Γ (1 + z7 )Γ (1 + ε + z1 − z4 − z5 − z6 − z7 )Γ (−ε − z1 − z2 + z5 + z7 ) ×

×Γ (1 + ε + z4 )Γ (−ε − z1 − z3 + z6 + z7 )Γ (1 + z5 + z6 + z7 ) .

(4.81)

Observe that, because of the presence of the factor Γ (−2ε) in the denominator, we are forced to take some residue in order to arrive at a non-zero result at ε = 0, so that the integral is eﬀectively sixfold. Then our standard procedure of taking residues and shifting contours can be applied, with the goal to obtain a sum of integrals where one may expand integrands in Laurent series in ε. The analysis of the integrand shows that the following four gamma functions play a crucial role for the generation of poles in ε: Γ (−ε + z2,3 ) and Γ (−1 − ε − z6,5 − z7 ). The ﬁrst decomposition of the integral (4.81) arises when one either takes a residue at the ﬁrst pole of one of these gamma functions or shifts the corresponding contour, i.e. changes the nature of this pole. As a result (4.81) is decomposed as 2T0001 + 2T0010 + 2T0011 +T0101 +2T0110 +2T0111 +T1010 +2T1011 +T1111 where the symmetry of the integrand is taken into account. Here the value 1 of an index means that a residue is taken and 0 means a shifting of a contour. The ﬁrst two indices correspond to the gamma functions Γ (−ε + z2 ) and Γ (−1 − ε − z5 − z7 ) and the second two indices to Γ (−ε + z3 ) and Γ (−1 − ε − z6 − z7 ), respectively. The term T0000 is absent because it is zero at ε = 0 due to Γ (−2ε) in the denominator.

94

4 Evaluating by MB Representation

Each of these terms is further decomposed appropriately and, eventually, one is left with integrals where integrands can be expanded in ε. These resulting terms involve up to ﬁve integrations. Taking some of these integrations with the help of the table of formulae presented in Appendix D, one can reduce all the integrals to no more than twofold MB integrals of gamma functions and their derivatives. In some of them, one more integration can be performed also in terms of gamma functions. Then the last integration, over z1 , is performed by taking residues and summing up resulting series, in terms of HPL. Keeping in mind the Regge limit, t/s → 0, let us, for deﬁniteness, decide to close the contour of the ﬁnal integration, over z1 , to the right and obtain power series in t/s. The coeﬃcients of these series are (up to of 1/n6 , S1 (n)/n5 , . . . , S1 (n)S3 (n)/n2 , . . ., where (−1)n ) linear n combinations −k , etc. (see Appendix C). Summing up these series with Sk (n) = j=1 j the help of tabulated formulae of Appendix C gives results in terms of HPL of the variable −t/s which can be continued analytically to any domain from the region |t/s| < 1. In the twofold MB integrals where one more integration (over a variable diﬀerent from z1 ) can hardly be performed in terms of gamma functions, one performs it with z1 in a vicinity of an integer point z1 = n = 0, 1, 2, . . ., in expansion in z = z1 − n, with a suﬃcient accuracy. Then one obtains power series where, in addition to nested sums with one index, various nested sums (see Appendix C) appear. These series are also summed up in terms of HPL. Eventually one arrives at the following result [29]: d/2 −γ ε 3 6 cj (x, L) iπ e E (0) , (4.82) T (s, t; ε) = − 3 1+3ε s (−t) εj j=0 where x = −t/s, L = ln(s/t), and c6 =

16 5 3 , c5 = − L , c4 = − π 2 , 9 3 2

3 c3 = 3(H0,0,1 (x) + LH0,1 (x)) + (L2 + π 2 )H1 (x) 2 11 2 131 ζ(3) , − π L− 12 9

(4.83)

(4.84)

c2 = −3 (17H0,0,0,1 (x) + H0,0,1,1 (x) + H0,1,0,1 (x) + H1,0,0,1 (x)) 3 −L (37H0,0,1 (x) + 3H0,1,1 (x) + 3H1,0,1 (x)) − (L2 + π 2 )H1,1 (x) 2 3 3 23 2 2 2 L + 8π H0,1 (x) − L + π L − 3ζ(3) H1 (x) − 2 2 1411 4 49 π , (4.85) + ζ(3)L − 3 1080

4.6 Three-Loop Examples

95

c1 = 3 (81H0,0,0,0,1 (x) + 41H0,0,0,1,1 (x) + 37H0,0,1,0,1 (x) + H0,0,1,1,1 (x) +33H0,1,0,0,1 (x) + H0,1,0,1,1 (x) + H0,1,1,0,1 (x) + 29H1,0,0,0,1 (x) +H1,0,0,1,1 (x) + H1,0,1,0,1 (x) + H1,1,0,0,1 (x)) + L (177H0,0,0,1 (x) +85H0,0,1,1 (x) + 73H0,1,0,1 (x) + 3H0,1,1,1 (x) + 61H1,0,0,1 (x) +3H1,0,1,1 (x) + 3H1,1,0,1 (x)) 119 2 139 2 47 2 2 L + π H0,0,1 (x) + L + 20π H0,1,1 (x) + 2 12 2 35 2 3 2 L + 14π 2 H1,0,1 (x) + L + π 2 H1,1,1 (x) + 2 2 23 3 83 2 L + π L − 96ζ(3) H0,1 (x) + 2 12 3 3 L + π 2 L − 3ζ(3) H1,1 (x) + 2 9 4 25 2 2 13 L + π L − 58ζ(3)L + π 4 H1 (x) + 8 8 8 73 2 301 503 4 π L + π ζ(3) − ζ(5) , − 1440 4 15 c0 = − (951H0,0,0,0,0,1 (x) + 819H0,0,0,0,1,1 (x) + 699H0,0,0,1,0,1 (x) +195H0,0,0,1,1,1 (x) + 547H0,0,1,0,0,1 (x) + 231H0,0,1,0,1,1 (x) +159H0,0,1,1,0,1 (x) + 3H0,0,1,1,1,1 (x) + 363H0,1,0,0,0,1 (x) +267H0,1,0,0,1,1 (x) + 195H0,1,0,1,0,1 (x) + 3H0,1,0,1,1,1 (x) +123H0,1,1,0,0,1 (x) + 3H0,1,1,0,1,1 (x) + 3H0,1,1,1,0,1 (x) +147H1,0,0,0,0,1 (x) + 303H1,0,0,0,1,1 (x) + 231H1,0,0,1,0,1 (x) +3H1,0,0,1,1,1 (x) + 159H1,0,1,0,0,1 (x) + 3H1,0,1,0,1,1 (x) +3H1,0,1,1,0,1 (x) + 87H1,1,0,0,0,1 (x) + 3H1,1,0,0,1,1 (x) +3H1,1,0,1,0,1 (x) + 3H1,1,1,0,0,1 (x)) −L (729H0,0,0,0,1 (x) + 537H0,0,0,1,1 (x) + 445H0,0,1,0,1 (x) +133H0,0,1,1,1 (x) + 321H0,1,0,0,1 (x) + 169H0,1,0,1,1 (x) +97H0,1,1,0,1 (x) + 3H0,1,1,1,1 (x) + 165H1,0,0,0,1 (x) +205H1,0,0,1,1 (x) + 133H1,0,1,0,1 (x) + 3H1,0,1,1,1 (x) +61H1,1,0,0,1 (x) + 3H1,1,0,1,1 (x) + 3H1,1,1,0,1 (x)) 311 2 619 2 531 2 89 2 L + π H0,0,0,1 (x) − L + π H0,0,1,1 (x) − 2 4 2 12 71 2 247 2 307 2 2 L + π H0,1,0,1 (x) − L + 32π H0,1,1,1 (x) − 2 12 2 107 2 151 2 197 2 2 L − π H1,0,0,1 (x) − L + 50π H1,0,1,1 (x) − 2 12 2

(4.86)

96

4 Evaluating by MB Representation

35 2 3 2 L + 14π 2 H1,1,0,1 (x) − L + π 2 H1,1,1,1 (x) 2 2 119 3 317 2 L + π L − 455ζ(3) H0,0,1 (x) − 2 12 47 3 179 2 L + π L − 120ζ(3) H0,1,1 (x) − 2 12 35 3 35 2 L + π L − 156ζ(3) H1,0,1 (x) − 2 12 3 3 L + π 2 L − 3ζ(3) H1,1,1 (x) − 2 69 4 101 2 2 559 4 L + π L − 291ζ(3)L + π H0,1 (x) − 8 8 90 27 5 25 2 3 9 4 25 2 2 13 L + π L − 58ζ(3)L + π 4 H1,1 (x) − L + π L − 8 8 8 40 8 183 131 4 37 ζ(3)L2 + π L − π 2 ζ(3) + 57ζ(5) H1 (x) − 2 60 12 223 2 167 624607 6 π ζ(3) + 149ζ(5) L + ζ(3)2 − π . (4.87) + 12 9 544320

−

The above result was conﬁrmed with the help of numerical integration in the space of alpha parameters [7]. Another natural check of the result is its agreement with the leading power Regge asymptotic behaviour [28] which was evaluated by an independent method based on the strategy of expansion by regions [4, 27]. The procedure described above can be applied, in a similar way, to the calculation of any massless planar on-shell triple box. At a ﬁrst step, one has to take care of the following four gamma functions in (4.80): Γ (2 − a12 − ε + z2 ), Γ (2 − a23 − ε + z3 ), Γ (2 − a59(10) − ε − z5 − z7 ), Γ (2 − a589 − ε − z6 − z7 ) . This procedure gives a decomposition similar to 2T0001 + 2T0010 + . . .. Next steps will be also generalizations of the corresponding steps in the evaluation of (4.81). The result presented above shows that analytical calculations of fourpoint on-shell massless Feynman diagrams at the three-loop level are quite possible so that one may think of evaluating three-loop virtual corrections to various scattering processes. Let us now consider a more complicated fourpoint three-loop diagram: Example 4.12. The massless on-shell tennis court5 diagram of Fig. 4.12. 5

Well, this is only one half of the court for singles. One also can call it ‘window’.

4.6 Three-Loop Examples

p1

97

p3

8 9

10

1 2

6 7

3

p2

5 4

p4

Fig. 4.12. Three-loop tennis court graph

To derive an appropriate MB representation we can proceed again quite straightforwardly. Here we need an auxiliary MB representation for the double box with two legs oﬀ shell applied to the double box subintegral in Fig. 4.12 and inserted into the MB representation for the on-shell box. As a result, an eightfold MB representation can be derived for the general diagram W (s, t; a1 , . . . , a11 , ε) of Fig. 4.12 with the eleventh index corresponding to the numerator (l1 ·l3 )−a11 which involves the scalar product of the momenta l1,3 ﬂowing though lines 1 and 3 in the same direction. Feynman integrals corresponding to Fig. 4.12 and many others will be indeed necessary to perform three-loop calculations of various scattering processes. It turns out that triple boxes are necessary right now in order to check some relations between diﬀerent loop orders in N = 4 supersymmetric gauge theories. The N = 4 theory has attracted considerable interest because of its remarkably simple structure and central role in the AdS/CFT correspondence. As was recently emphasized in [1], one needs, in addition to the result (4.87) for the triple box considered above, just one more triple box [6], namely, W (s, t; 1, . . . , 1, −1, ε). For this integral, one has d/2 3 iπ W (s, t; 1, . . . , 1, −1, ε) = − Γ (−2ε)(−s)1+3ε t2 w +i∞ +i∞ 7 t 1 × ... dw dz1 dzj Γ (−zj ) Γ (1 + 3ε + w) (2πi)8 −i∞ s −i∞ j=2 Γ (−3ε − w)Γ (1 + z1 + z2 + z3 )Γ (−1 − ε − z1 − z3 )Γ (1 + z1 + z4 ) Γ (1 − z2 )Γ (1 − z3 )Γ (1 − z6 )Γ (1 − 2ε + z1 + z2 + z3 ) Γ (−1 − ε − z1 − z2 − z4 )Γ (2 + ε + z1 + z2 + z3 + z4 ) × Γ (−1 − 4ε − z5 )Γ (1 − z4 − z7 )Γ (2 + 2ε + z4 + z5 + z6 + z7 ) ×Γ (−ε + z1 + z3 − z5 )Γ (2 − w + z5 )Γ (−1 + w − z5 − z6 )

×

×Γ (z5 + z7 − z1 )Γ (1 + z5 + z6 )Γ (−1 + w − z4 − z5 − z7 ) ×Γ (−ε + z1 + z2 − z5 − z6 − z7 )Γ (1 − ε − w + z4 + z5 + z6 + z7 ) ×Γ (1 + ε − z1 − z2 − z3 + z5 + z6 + z7 ) . (4.88)

98

4 Evaluating by MB Representation

There is again the factor Γ (−2ε) in the denominator, so that the integral is eﬀectively sevenfold. The evaluation of this integral in expansion in ε is in progress. Here is a preliminary result up to 1/ε3 : d/2 −γ ε 3 6 cj iπ e E W (s, t; 1, . . . , 1, −1, ε) = − , (4.89) 1+3ε 2 (−s) t εj i=0 where 16 13 19 1 , c5 = − ln x , c4 = − π 2 + ln2 x 9 6 12 2 5 7 3 5 ln x − ln2 x ln(1 + x) c3 = [Li3 (−x) − ln x Li2 (−x)] + 2 12 4 157 2 5 2 241 π ln x − π ln(1 + x) − ζ(3) + 72 4 18 with x = t/s. c6 =

(4.90)

4.7 More Loops One can proceed in the same style even in higher loops. Let us illustrate this point by considering Example 4.13. The four-loop ladder massless on-shell diagram shown in Fig. 4.13.

p1

1 2

p2

6 7

3

10 5

4

9 8

p3

13 12 11

p4

Fig. 4.13. Four-loop ladder diagram

We start with the derivation of an appropriate MB representation for general powers of the propagators. As before we use this general strategy because it provides a lot of checks and gives the possibility to obtain MB representations for various diagrams which result from the given diagram when contracting some lines. As in the previous example, we need an auxiliary MB representation for the double box with two legs oﬀ shell but in a diﬀerent situation (two left legs rather than two upper legs oﬀ shell). It can easily be derived by the technique described and takes the form

4.7 More Loops

99

d/2 2 iπ (−1)a a−4+2ε j=2,4,5,6,7 Γ (aj )Γ (4 − a4567 − 2ε)(−s) +i∞ +i∞ 6 1 (−t)z4 (−p21 )z5 (−p22 )z6 × . . . dzj Γ (−zj ) 6 (2πi) −i∞ (−s)z4 +z5 +z6 −i∞ j=1

K2 (s, t; a1 , . . . , a8 , ε) =

Γ (a4567 + ε − 2 + z1 + z2 + z3 )Γ (a5 + z1 )Γ (a7 + z1 + z2 + z3 ) Γ (4 − 2ε − a1238 + z1 + z2 + z3 )Γ (a1 − z2 )Γ (a3 − z3 )Γ (a8 − z1 ) ×Γ (a1238 + ε − 2 − z1 − z2 − z3 + z4 + z5 + z6 )Γ (a2 + z4 + z5 + z6 ) ×Γ (2 − ε − a457 − z1 − z3 )Γ (2 − ε − a567 − z1 − z2 )Γ (a8 − z1 + z4 )

×

×Γ (2 − ε − a128 + z1 + z2 − z4 − z5 ) ×Γ (2 − ε − a238 + z1 + z3 − z4 − z6 ) .

(4.91)

Then, similarly to the derivation of the multiple MB representation for the triple box when we inserted the MB representation of the box with two legs oﬀ shell into the MB representation of the on-shell double box, let us now insert (4.91) instead. We come to the following tenfold MB representation of the four-loop ladder diagram: d/2 2 iπ (−1)a (−s)8−a−4ε Q(s, t; a1 , . . . , a13 , ε) = j=2,5,7,9,11,12,13 Γ (aj )Γ (4 − a9,11,12,13 − 2ε) z7 +i∞ +i∞ 10 t 1 . . . dz Γ (−zj ) × j 10 (2πi) s −i∞ −i∞ j=1 j=2,3,5,6,7,8,9 Γ (a1,2 + z1 )Γ (a2 + z7 )Γ (z7 − z10 )Γ (z10 − z4 )Γ (z4 − z1 ) Γ (a10 − z2 )Γ (a8 − z3 )Γ (a6 − z5 )Γ (a4 − z6 )Γ (a1 − z8 )Γ (a3 − z9 ) Γ (2 − ε − a9,11,12 − z1 − z3 )Γ (2 − ε − a9,12,13 − z1 − z2 ) × Γ (4 − 2ε − a5,8,10 + z1 + z2 + z3 )Γ (4 − 2ε − a4,6,7 + z4 + z5 + z6 ) Γ (a9 + z1 + z2 + z3 )Γ (a9,11,12,13 + ε − 2 + z1 + z2 + z3 ) × Γ (4 − 2ε − a1,2,3 + z8 + z9 + z10 ) ×Γ (2 − ε − a5,10 + z1 + z2 − z4 − z5 )Γ (2 − ε − a5,8 + z1 + z3 − z4 − z6 ) ×Γ (a5 + z4 + z5 + z6 )Γ (a5,8,10 + ε − 2 − z1 − z2 − z3 + z4 + z5 + z6 ) ×

×Γ (2 − ε − a6,7 + z4 + z5 − z8 − z10 )Γ (2 − ε − a1,2 + z8 + z10 − z7 ) ×Γ (2 − ε − a4,7 + z4 + z6 − z9 − z10 )Γ (2 − ε − a2,3 + z9 + z10 − z7 ) ×Γ (a1,2,3 + ε − 2 − z8 − z9 − z10 + z7 )Γ (a7 + z8 + z9 + z10 ) ×Γ (a4,6,7 + ε − 2 − z4 − z5 − z6 + z8 + z9 + z10 ) ,

(4.92)

where we separate indices in a9,11,12,13 = a9 + a11 + a12 + a13 etc. by commas because they are now two-digit. One can check this monster representation as before, using partial cases: when we put the indices a2 , a5 , a7 , a9 , a12 to zero we reproduce a known analytical result for the product of four one-loop propagator diagrams with

100

4 Evaluating by MB Representation

Fig. 4.14. The ‘N in O’ diagram

the indices (a1 , a3 ), (a4 , a6 ), (a8 , a10 ) and (a11 , a13 ). When we put the indices a1 , a3 , a4 , a6 , a8 , a10 , a11 , a13 to zero we reproduce a known analytical result for the four-loop water melon diagram with the indices a2 , a5 , a7 , a9 , a12 and the external momentum square t. Representation (4.92) contains a lot of information. Let us use it in order to calculate the ‘N in O’ diagram6 shown in Fig. 4.14 exactly in four dimensions, i.e. at ε = 0. This is nothing but N (q 2 ) = Q(s, t; 1, 0, 1, 1, 1, 0, 1, 0, 1, 1, 1, 0, 1, 0) which is, of course, independent of t and proportional to 1/q 2 . The limit a2 , a12 → 0 is achieved as described above, due to four residues with respect to some of the integration variables. Then one can simply set a6 = a8 = 0 and obtain 4 C (4.93) N (q 2 ) = iπ 2 q2 with the constant C given by a ﬁnite ﬁvefold MB integral. Three of these ﬁve integrations can be performed explicitly with the help of tabulated formulae of Appendix D, and one can obtain the following twofold MB integral: +i∞ +i∞ dz1 dz2 1 Γ (z1 + z2 )Γ (1 − z1 − z2 )Γ (z2 )Γ (−z2 ) C= 2 (2πi) −i∞ −i∞ 2z12 z2 ×Γ (1 − z1 )Γ (z1 ) [z1 (ψ(1 − z1 ) + ψ(z1 ) − ψ(1 − z1 − z2 ) − ψ(z1 + z2 )) −z2 (ψ(1 − z1 − z2 ) − ψ(−z2 ) − ψ(z2 ) + ψ(z1 + z2 ))]

× ψ(z1 )2 − 2ψ(z1 )ψ(1 − z1 − z2 ) + 2ψ(1 − z1 − z2 )ψ(z1 + z2 ) −ψ(z1 + z2 )2 − ψ (z1 ) + ψ (z1 + z2 ) ,

(4.94)

where the poles at z1 = 0 and z2 = 0 are considered left so that one can choose 0 < Rez1 , Rez2 < 1 with Rez1 + Rez2 < 1 for the integration contour. One can check numerically, with a high accuracy, that the known result which will be presented shortly is successfully reproduced. 6

This diagram was a challenge in the eighties in renormalization group calculations. In the ﬁrst result on the ﬁve-loop β-function in the φ4 theory [9] (see [19] for a corrected later version) the contribution of this diagram was treated numerically. The analytical value of this diagram was predicted and later proven in [18] using a technique based on functional equations – see more details in Appendix F.

4.7 More Loops

101

The twofold MB integral (4.94) can be converted into a sum of two twofold series of expressions consisting of nested sums (see Appendix C). The ﬁrst of them is obtained by taking residues at the points z2 = 1, 2, . . . and then at z1 = 1, 2, . . .. The second of them is obtained by taking residues at the points z2 = 1 − z1 + n2 with n2 = 1, 2, . . . and then at z1 = 1, 2, . . .. Then one can perform one of the summations using the package SUMMER [39] and arrive at the following onefold series: C=

5 ∞ cj,n n=1 j=1

nj

,

(4.95)

where c5,n = 5π 2 /6 − 6S12 − 27S2 , c4,n = 5π 2 S1 /2 + 3S13 − 18S12 + 12S1 S2 − 6S3 + 12ζ(3) ,

(4.96) (4.97)

c3,n = π 4 /5 − 4π 2 S12 /3 − S14 /2 − 28S112 + 20S1 S12 − 10S13 −19π 2 S2 /6 − S12 S2 + 37S22 /2 + 4S1 S3 + 11S4 + 6S1 ζ(3) ,

(4.98)

c2,n = π 4 S1 /10 + π 2 S13 /6 − 2S1112 − 18S113 − 17π 2 S12 /6 + 16S1 S13 +11S14 + 4π 2 S1 S2 /3 − 2S13 S2 /3 + 6S12 S2 − S1 S22 − 13S212 +19S23 − π 2 S3 /3 − 5S12 S3 + 2S2 S3 /3 − 6S1 S4 − 4S5 − 2π 2 ζ(3)/3 −3S12 ζ(3) − S2 ζ(3) + 14ζ(5) , (4.99) 2 c1,n = 61π 6 /2520 − 16S1113 + π 2 S112 /3 + 4S1 S113 + 14S114 − 3S12 +10S123 − 3π 2 S13 − 5S1 S14 + 8S15 + 3π 4 S2 /20 − π 2 S12 S2 /6 +6S112 S2 − S1 S12 S2 + 10S13 S2 − 5π 2 S22 /6 + S12 S22 + 5S23 /3 − 8S2112 +S1 S212 − 3S1 S23 + 18S24 + 10π 2 S1 S3 /3 + 2S13 S3 /3 − 4S12 S3 −7S1 S2 S3 + 10S32 /3 − π 2 S4 /6 − 3S12 S4 − 31S2 S4 − 9S1 S5 − 80S6 /3 −4S12 ζ(3) + 4S1 S2 ζ(3) + 14S3 ζ(3) − 9ζ(3)2 ,

(4.100)

and we omit the argument n−1 in all the nested sums involved, i.e. S1 stands for S1 (n − 1) etc. Summation of the terms with 1/n5 , . . . ,1/n2 can be performed with the help of formulae (C.51)–(C.82) implemented in SUMMER [39]. The terms with 1/n are also successfully summed up by SUMMER, and we arrive at the wellknown result [18]: 1 4 441 ζ(7) . (4.101) N (q 2 ) = 2 iπ 2 q 8 I cannot say that the derivation of this result outlined above is simpler than that of [18]. Let me, however, stress that the present derivation involves a lot of steps that are performed automatically, and a lot of other similar results (e.g. for diagrams which can be obtained from the four-loop ladder diagram by shrinking other lines to points) can be obtained quite similarly.

102

4 Evaluating by MB Representation

4.8 MB Representation versus Expansion by Regions To expand a given Feynman integral in some limit, where certain masses and/or kinematical invariants are large with respect to the rest of these parameters, one can successfully apply expansion by regions [4, 30], as explained in the book [27] in detail. An alternative technique for solving the problem of asymptotic expansion is provided by multiple MB representations. Let us see how it works using some of our previous examples. For Example 4.1, we have derived the MB representation (4.3). Let us use it to expand such Feynman integrals in the two diﬀerent limits, m2 /q 2 → 0 and q 2 /m2 → 0. Consider, for example, F4.1 (2, 1, 4) represented by (4.4). This is an integral over the variable z, with the ratio m2 /q 2 present in the form (m2 /q 2 )z . The initial integration contour is at −1 < Rez < 0. Let us observe that if we follow the procedure used to evaluate this integral, i.e. close the integration contour to the right and pick up (minus) residues at z = 0, 1, 2, . . . , n, . . . we shall obtain terms of the asymptotic expansion in the limit m2 /q 2 → 0. Indeed, one can prove that the remainder of this expansion determined by picking up the (n + 1)-st residue is of order (m2 )n+1 . Thus we obtain iπ 2 m2 m4 −q 2 − ... . (4.102) F4.1 (2, 1; 4) = 2 ln 2 − 2 − q m q 2(q 2 )2 If we are interested in the opposite limit, q 2 /m2 → 0, the natural idea is to close the integration contour to the left and take residues at the points z = −1, −2, . . . to obtain iπ 2 (q 2 )2 q2 + + ... . (4.103) F4.1 (2, 1; 4) = − 2 1 + m 2m2 3m4 Consider now Example 4.3, where IR and collinear divergences are present. We can use MB representation (4.11) for expanding Feynman integrals with various indices in the two diﬀerent limits, t/s → 0 and s/t → 0. There is again the typical dependence of the ratio of t and s on z of the form (t/s)z . The procedure of using (4.11) to obtain an asymptotic expansion in the limit t/s → 0 is standard: to shift the integration contour to the right. For the integral with given indices al , the points where it is necessary to take (minus) residues are given by the right poles of the gamma functions, in our terminology: at z = 0, 1, 2, . . . and at z = 2 − max{a1 , a3 } − a2 − a4 − ε + n with n = 0, 1, 2, . . .. For example, for F (s, t; d) = F4.3 (s, t; 1, 1, 1, 1, d) represented by (4.12), these are the two series of residues at z = 0, 1, 2, . . . and z = −1−ε, −ε, 1−ε, . . . which reproduce the hard and collinear contributions, respectively, to the asymptotic expansion within expansion by regions – see Chap. 8 of [27]. We obtain Γ (1 + ε)Γ (−ε)2 t iπ d/2 ln + 2ψ(−ε) − ψ(1 + ε) + γE F (s, t; d) = Γ (−2ε) s(−t)1+ε s

4.8 MB Representation versus Expansion by Regions

103

Γ (ε)Γ (1 − ε)2 t − ln + 2ψ(1 − ε) − ψ(ε) − 1 + γE s2 (−t)ε s Γ (2 + ε)Γ (−1 − ε)2 + (−s)2+ε 3 Γ (ε − 1)Γ (2 − ε)2 (−t)1−ε t + ln + 2ψ(2 − ε) − ψ(ε − 1) − + γE 2s3 s 2 ! 2 Γ (3 + ε)Γ (−2 − ε) t + + ... . (4.104) (−s)3+ε To obtain the asymptotic expansion in the opposite limit, s/t → 0, one shifts the integration contour to the left and takes residues at the left poles at z = 2 − min{a2 , a4 } − n and at z = 2 − a − ε − n with n = 0, 1, 2, . . .. For F (s, t; d), these are the two series of residues at z = −1, −2, . . . and z = −2 − ε, −3 − ε, −4 − ε, . . .. One can check that the resulting expansion is nothing but (4.104) with the interchange s → t, t → s – this should be the case because of the symmetry of the initial integral. In these two examples, terms of asymptotic expansions were obtained as residues in onefold MB integrals. As a non-trivial example with a multiple MB integration let us turn again to Example 4.8 of massless on-shell double boxes. Let us evaluate the leading asymptotic behaviour of the K(s, t; 1, . . . , 1, 0, ε) in the Regge limit, t/s → 0, using representation (4.48). The starting point of the evaluation of this quantity in expansion in ε was the analysis of gluing of right and left poles which showed the way how the poles in ε are generated. Now, our starting point is to look at the integration over the variable z1 which enters as the power of the ratio t/s and try to understand what right poles with respect to z1 are. One source of such poles is obvious: this is Γ (−z1 ) corresponding to the hard part within expansion by regions – see Chap. 8 of [27]. This part, however, starts only with order (t/s)0 which is subleading, as we will see shortly. Other sources are not visible at once, similarly to the poles in ε. However, the experience obtained in our previous examples when analysing the singular behaviour in ε shows how the poles in z1 appear after integrating over z2 , z3 and z4 . Let us use the rule formulated in Sect. 4.2 and systematically applied in our examples and analyse the integrand of (4.48) from the point of view of generating right poles in z1 . Apart from Γ (−z1 ), there are only two gamma functions that can generate a singularity of the type Γ (. . . − z1 ): Γ (−1 − ε − z1 − z2 ) and Γ (−1 − ε − z1 − z3 ) . Indeed, the singularity of the type Γ (−1 − ε − z1 ) is generated, due to the integration over z2 , because of the presence of Γ (−ε + z2 ), and, due to the integration over z3 , because of the presence of Γ (−ε+z3 ). Thus, to reveal this singularity, we can take a residue at the ﬁrst pole of Γ (−ε+z2 ) or Γ (−ε+z3 ). Therefore, we start with the same decomposition F = F11 +F10 +F01 +F00 as in Sect. 4.4. Now, in F11 represented by (4.49) and in F01 represented by

104

4 Evaluating by MB Representation

(4.50), the function Γ (−1−2ε−z1 ) is already explicitly present. The term F00 does not contribute now because it cannot generate the leading asymptotic behaviour in the given limit. To evaluate the leading asymptotics, let us, ﬁrst, consider F11 and take (minus) residue at z1 = −1 − 2ε to obtain +i∞ Γ (1 − ε − z4 )Γ (−2ε − z4 )2 Γ (1 + 2ε) 1 dz f11 = 4 x1+2ε 2πi −i∞ Γ (1 + ε + z4 )Γ (−4ε − z4 ) ×Γ (ε + z4 )2 Γ (z4 ) [2γE + ln x + ψ(−2ε) − ψ(1 + 2ε) − ψ(−4ε − z4 ) +ψ(−2ε − z4 ) + ψ(1 − ε − z4 ) + ψ(z4 )] . (4.105) Observe that this quantity is nothing but the contribution F111 that we have met in Sect. 4.4. It was evaluated in expansion in ε by taking residues at z4 = 0 and z4 = ε and shifting the integration contour over z4 . Starting from F01 and taking (minus) residue at z1 = −1 − 2ε we obtain +i∞ +i∞ Γ (1 + 2ε) 1 dz2 dz4 Γ ∗ (−ε + z2 )Γ (ε − z2 ) f01 = − x1+2ε (2πi)2 −i∞ −i∞ Γ (ε + z4 )Γ (−2ε − z4 )Γ (1 − ε − z4 ) × Γ (−4ε − z4 ) Γ (z2 + z4 )Γ (−ε + z2 + z4 )Γ (−ε − z2 − z4 ) . (4.106) × Γ (1 + z2 + z4 ) where the asterisk denotes, as in Appendix D, the opposite nature of the ﬁrst pole of Γ (−ε + z2 ). Now we observe that this is nothing but the contribution F011 of Sect. 4.4, where it was explained how it can be evaluated in expansion in ε. Summing up results for F111 and F011 we reproduce the leading part of (4.54), e.g. the terms of order 1/t modulo logarithms. So, we see that the evaluation of the leading asymptotic behaviour in the Regge limit, using MB representation, is a (simple) part of the global evaluation. Observe that the evaluation of the triple box in Example 4.11 is also organized in such a way that the leading Regge asymptotics can be extracted from this evaluation. On the other hand, it was also evaluated using expansion by regions [28]. It is not clear in advance which way is simpler: expanding by MB representation, or, by regions. My experience tells me that, usually, expanding by regions is certainly preferable, but sometimes, it looks more convenient to derive an appropriate MB representation and proceed as described in this section. But I can imagine that, sometimes, this is just a matter of taste. In complicated situations, the two strategies can successfully be combined. In particular, extracting the leading asymptotic behaviour from a general MB representation can show what kind of contributions one gets and will help detecting all regions which contribute. There are a lot of papers where the asymptotic behaviour was evaluated using MB representations – see, e.g., [16].

4.9 Conclusion

105

4.9 Conclusion Mellin integrals were used for the evaluation of Feynman integrals in various ways. For example, in [35], the ﬁrst analytical result for massless double boxes of Fig. 4.7 was obtained in the case where all the external legs are oﬀ-shell so that these are functions depending on many variables, s, t and p2i for i = 1, 2, 3, 4. Nevertheless it was possible to evaluate the double box for all powers of the propagators equal to one exactly in four dimensions. The following nice mathematical result was obtained: 2 2 iπ C(p21 p24 , p22 p23 , st) , s2 t where 1 C(x1 , x2 , x3 ) = (6 [Li4 (−ρx) + Li4 (−ρy)] λ y y 1 +3 ln [Li3 (−ρx) − Li3 (−ρy)] + ln2 [Li2 (−ρx) + Li2 (−ρy)] x 2 x 1 2 π2 2 y 7π 4 π2 2 + ln (ρx) ln (ρy) + ln(ρx) ln(ρy) + ln + , (4.107) 4 2 12 x 60 (1 − x − y)2 − 4xy , 2 , ρ ≡ ρ(x, y) = 1 − x − y + λ(x, y)

λ ≡ λ(x, y) =

(4.108) (4.109)

and x = x1 /x3 , y = x2 /x3 . Moreover, a similar analytical result was obtained [36] also for a general oﬀ-shell h-loop ladder planar diagram, in particular, for the oﬀ-shell triple box.7 In [37], an oﬀ-shell result for the non-planar two-loop three-point diagram was also obtained using the MB representation. Other examples of results obtained by this technique are analytical expressions for n-point oneloop massive Feynman integrals for general d [10]. Let me summarize the basic features that distinguish the technique of MB representation presented in this chapter and oriented at the evaluation in ε-expansion from other approaches based on Mellin integrals. – An appropriate multiple MB representation for a given class of integrals is derived for general powers of the propagators and irreducible numerators. 7 Well, one can hardly expect that explicit analytical results can be obtained for other (even double-box) Feynman integrals of this purely oﬀ-shell class, in particular, with a double power of some propagator, with some irreducible numerator, or where one of the lines other than rungs is contracted to a point. The possibility to obtain such a nice mathematical result for such a complicated object depending on so many variables in the case of all indices equal to one was later understood by making an interesting mathematical link with some problem of conformal quantum mechanics – see [17].

106

– –

–

–

–

–

–

4 Evaluating by MB Representation

In order to achieve the minimal number of MB integrations it is recommended to derive an MB representation for a sub-loop integral, insert it in the given integral over the loop momenta, etc. There is always the possibility to check multiple MB representations, which are sometimes rather cumbersome, by using simple partial cases. Multiple MB integrals are very ﬂexible for the resolution of the singularities in ε. This procedure reduces to shifting contours, in an appropriate way, and taking corresponding residues. After the resolution of the singularities in ε, at least some of the integrations can be performed explicitly by tabulated formulae of Appendix D, with results in terms of gamma and psi functions. One can usually have an easy numerical control on ﬁnite (in ε) MB integrals: it is enough to integrate from −5i to +5i along the imaginary axis to have a very good accuracy. When the integration in multiple MB integrals is hardly performed explicitly, one can convert them into multiple series and apply such packages as SUMMER [39] for summation. Onefold MB integrals can be summed up by closing the integration contour and summing up corresponding residues. Here one can apply summation formulae of Appendix C and/or SUMMER. All the manipulations with MB integrals can be done on a computer. (For example, I use MATHEMATICA for this.)

The technique of multiple MB representations is not always optimal. This holds at least for non-planar double boxes with one leg oﬀ-shell. Although ﬁrst analytical results were obtained with its help [24, 25] the adequate technique here turned out to be the method of diﬀerential equations which will be studied in Chap. 7.

References 1. C. Anastasiou, Z. Bern, L.J. Dixon and D.A. Kosower, Phys. Rev. Lett. 91 (2003) 251602. 97 2. C. Anastasiou, J.B. Tausk and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 262. 64, 74, 75, 81 3. W. Beenakker and A. Denner, Nucl. Phys. B 338 (1990) 349. 68 4. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 55, 78, 87, 96, 102 5. M.C. Berg`ere and Y.-M.P. Lam, Commun. Math. Phys. 39 (1974) 1. 55 6. Z. Bern, J.S. Rozowsky and B. Yan, Phys. Lett. B 401 (1997) 273. 97 7. T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 88, 89, 96 8. E.E. Boos and A.I. Davydychev, Theor. Math. Phys. 89 (1991) 1052, [Teor. Mat. Fiz. 89 (1991) 56]. 56, 91 9. K.G. Chetyrkin, S.G. Gorishnii, S.A. Larin and F.V. Tkachov, Phys. Lett. B 132 (1983) 351. 100 10. A.I. Davydychev, J. Math. Phys. 32 (1991) 1052; 33 (1992) 358. 105

References 11. 12. 13. 14. 15. 16.

17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39.

107

A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 699 (2004) 3. 69, 84, 91, 173 A. Devoto and D.W. Duke, Riv. Nuovo Cim. 7, No. 6 (1984) 1. 62, 78 J. Fleischer, M.Yu. Kalmykov and A.V. Kotikov, Phys. Lett. B 462 (1999) 169. 91, 173 J. Fleischer, A.V. Kotikov and O.L. Veretin, Nucl. Phys. B 547 (1999) 343. 62, 77, 84 R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542. 73 C. Greub, T. Hurth and D. Wyler, Phys. Rev. D 54 (1996) 3350; C. Greub and P. Liniger, Phys. Rev. D 63 (2001) 054025; H.H. Asatryan, H.M. Asatrian, C. Greub and M. Walker, Phys. Rev. D 65 (2002) 074004; K. Bieri, C. Greub and M. Steinhauser, Phys. Rev. D 67 (2003) 114019. 104 A.P. Isaev, Nucl. Phys. B 662 (2003) 461. 105 D.I. Kazakov, Theor. Math. Phys. 62, 84 (1985) [Teor. Mat. Fiz. 62, 127 (1984)]. 100, 101 H. Kleinert, J. Neu, V. Schulte-Frohlinde, K.G. Chetyrkin and S.A. Larin, Phys. Lett. B 272, 39 (1991) [Erratum-ibid. B 319, 545 (1993)]. 100 K.S. K¨ olbig, J.A. Mignaco and E. Remiddi, BIT 10 (1970) 38; K.S. K¨ olbig, Math. Comp. 39 (1982) 647. 62, 78 L. Lewin, Polylogarithms and Associated Functions (North-Holland, Amsterdam, 1981). 78 E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725. 88 V.A. Smirnov, Phys. Lett. B 460 (1999) 397. 74, 77 V.A. Smirnov, Phys. Lett. B 491 (2000) 130. 106 V.A. Smirnov, Phys. Lett. B 500 (2001) 330. 106 V.A. Smirnov, Phys. Lett. B 524 (2002) 129. 82, 83, 85, 87 V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 55, 67, 78, 87, 96, 102, 103 V.A. Smirnov, Phys. Lett. B 547 (2002) 239. 96, 104 V.A. Smirnov, Phys. Lett. B 567 (2003) 193. 92, 94 V.A. Smirnov and E.R. Rakhmetov, Teor. Mat. Fiz. 120 (1999) 64; V.A. Smirnov, Phys. Lett. B 465 (1999) 226. 78, 102 V.A. Smirnov, hep-ph/0406052; G. Heinrich and V.A. Smirnov, hepph/0406053. 88, 89, 90 V.A. Smirnov and O.L. Veretin, Nucl. Phys. B 566 (2000) 469. 78, 80 J.B. Tausk, Phys. Lett. B 469 (1999) 225. 64, 74, 81 N.I. Ussyukina, Teor. Mat. Fiz. 22 (1975) 300. 55 N.I. Ussyukina and A.I. Davydychev, Phys. Lett. B 298 (1993) 363. 74, 105 N.I. Ussyukina and A.I. Davydychev, Phys. Lett. B 305 (1993) 136. 105 N.I. Ussyukina and A.I. Davydychev, Phys. Lett. B 332 (1994) 159. 105 J.A.M. Vermaseren, Symbolic Manipulation with FORM (CAN, Amsterdam, 1991). 62 J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037. 62, 101, 106

5 IBP and Reduction to Master Integrals

The next method in our list is based on integration by parts1 (IBP) [15] within dimensional regularization, i.e. property (2.38). The idea is to write down various equations (2.38) for integrals of derivatives with respect to loop momenta and use this set of relations between Feynman integrals in order to solve the reduction problem, i.e. to ﬁnd out how a general Feynman integral of the given class can be expressed linearly in terms of some master integrals. In contrast to the evaluation of the master integrals, which is performed, at a suﬃciently high level of complexity, in a Laurent expansion in ε, the reduction problem is solved at general d, and the expansion in ε does not provide simpliﬁcations here. The reduction can be stopped whenever one arrives at suﬃciently simple integrals. On the other hand, one could try to solve the reduction problem in the ultimate mathematical sense, i.e. to reduce a given integral to true irreducible integrals which cannot be reduced further. To illustrate the procedure of solving IBP relations we shall begin in Sect. 5.1 with very simple one-loop examples. Usually, we shall indeed stop the reduction if we obtain integrals that can be expressed in terms of gamma functions for general values of the parameter of dimensional regularization, d. In Sect. 5.2, we shall proceed in two loops. We shall also study some general tricks within the method of IBP such as the triangle rule and shifting dimension. One of the two-loop examples, the reduction of massless on-shell double boxes, will be considered separately in Sect. 5.3. We shall conclude in Sect. 5.4 with brief bibliographic remarks and a description of attempts of making systematic the procedure of solving IBP recurrence relations.

5.1 One-Loop Examples The ﬁrst example is very simple: Example 5.1. One-loop vacuum massive Feynman integrals 1 For one loop, IBP was used in [34]. The crucial step – an appropriate modiﬁcation of the integrand before diﬀerentiation, with an application at the two-loop level (to massless propagator diagrams) – was taken in [15] and, in a coordinate-space approach, in [51]. The case of three-loop massless propagators was treated in [15].

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 109–132 (2004) c Springer-Verlag Berlin Heidelberg 2004

110

5 IBP and Reduction to Master Integrals

F5.1 (a) =

dd k . (k 2 − m2 )a

(5.1)

In this chapter, we are concentrating on the dependence of Feynman integrals on the powers of the propagators so that we will usually omit dependence on dimension, masses and external momenta. Let us forget that we know the explicit result (A.1) and try to exploit information following from IBP. Let us use the IBP identity 1 ∂ =0, (5.2) dd k ·k 2 ∂k (k − m2 )a with (∂/(∂k))·k = (∂/(∂kµ ))kµ , and write down resulting quantities in terms of integrals (5.1). We obtain (d − 2a)F (a) − 2am2 F (a + 1) = 0 .

(5.3)

This gives the following recurrence relation: F (a) =

d − 2a + 2 F (a − 1) . 2(a − 1)m2

(5.4)

We see that any Feynman integral with integer a > 1 can be expressed recursively in terms of one integral F (1) ≡ I1 which we therefore consider as a master integral. (Observe that all the integrals with non-positive integer indices are zero since they are massless tadpoles.) This can be done explicitly here: (−1)a (1 − d/2)a−1 I1 , (5.5) F (a) = (a − 1)!(m2 )a−1 where (x)a is the Pochhammer symbol and the only master integral is I1 = −iπ d/2 Γ (1 − d/2)(m2 )d/2−1 .

(5.6)

As in Chap. 3 let us consider Example 5.2. Massless one-loop propagator Feynman integrals dd k F5.2 (a1 , a2 ) = . (k 2 )a1 [(q − k)2 ]a2

(5.7)

(As we have agreed, the dependence on q 2 and d is omitted.) For integer powers of the propagators, these integrals are zero whenever one of the indices is non-positive. Let us forget the explicit result (3.6) and try to apply the IBP identity 1 ∂ =0. (5.8) dd k ·k 2 a1 ∂k (k ) [(q − k)2 ]a2 We recognize diﬀerent terms resulting from the diﬀerentiation as integrals (5.7) and obtain the following relation

5.1 One-Loop Examples

d − 2a1 − a2 − a2 2+ (1− − q 2 ) = 0

111

(5.9)

which is understood as applied to the general integral F (a1 , a2 ) with the standard notation for increasing and lowering operators, e.g. 2+ 1− F (a1 , a2 ) = F (a1 − 1, a2 + 1). We rewrite it as a2 q 2 2+ = a2 1− 2+ + 2a1 + a2 − d

(5.10)

and obtain the possibility to reduce the sum of the indices a1 + a2 . Explicitly, applying (5.10) to the general integral and shifting the index a2 , we have 1 [(d − 2a1 − a2 + 1)F (a1 , a2 − 1) (a2 − 1)q 2 −(a2 − 1)F (a1 − 1, a2 )] ,

F (a1 , a2 ) = −

(5.11)

Indeed, a1 + a2 on the right-hand side is less by one than on the left-hand side. This relation can be applied, however, only when a2 > 1. Suppose now that a2 = 1. Then we use the symmetry property F (a1 , a2 ) = F (a2 , a1 ) and apply (5.11) interchanging a1 and a2 and setting a2 = 1: F (a1 , 1) = −

d − a1 − 1 F (a1 − 1, 1) . (a1 − 1)q 2

(5.12)

This relation enables us to reduce the index a1 to one and we see that the two relations (5.11) and (5.12) provide the possibility to express any integral of the given family in terms of the only master integral I1 = F (1, 1) given by (3.8), i.e. F (a1 , a2 ) = c(a1 , a2 )I1 , and the corresponding coeﬃcient function c(a1 , a2 ) is constructed as a rational function of d. Let us now complete the analysis for the example considered in the introduction, i.e. once again consider our favourite example: Example 5.3. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. We stopped in Chap. 1 at the point where we were able to express any integral (1.2) in terms of the master integral I1 = F (1, 1) and integrals with a2 ≤ 0 which can be evaluated for general d in terms of gamma functions by means of (A.3). Let us now try to understand what the true master integrals are. They should be really irreducible, i.e. they cannot be expressed linearly in terms of other integrals. Suppose that a2 ≤ 0, Then we can apply (1.11) to reduce a1 to one. In the case a1 = 1, we use relation (1.11) multiplied by 2− to express the term 2m2 a1 1+ 2− in (1.13). Thus, we obtain the following relation (d − a2 − 1)2− = (q 2 − m2 )2 a2 2+ + (q 2 + m2 )(d − 2a2 − 1)

(5.13)

that can be used to increase the index a2 to zero or one starting from negative values. We come to the conclusion that there are two irreducible integrals I1 = F (1, 1) given by (1.7) and I2 = F (1, 0) which equals the right-hand side of (5.6), and any integral from our family can be expressed linearly in terms

112

5 IBP and Reduction to Master Integrals

of them. This reduction procedure to I1 and I2 can easily be implemented on a computer. Observe that the integrals I1 and I2 cannot be linearly expressed through each other because, at general d, I1 is a non-trivial function of q 2 and m2 while I2 is independent of q 2 . This was the last example in this chapter, where we solve the reduction problem in the maximal way, i.e. in the sense of reduction to irreducible integrals. In the rest of the examples, we shall not be so curious and will stop the reduction whenever we arrive at suﬃciently simple classes of integrals. In Chap. 6, however, the reduction will be performed in the ultimate sense. Some other approaches with this property will be characterized in Sect. 5.4. The next example is again our old one. Example 5.4. The triangle diagrams of Fig. 3.4 given by (3.19). Writing down IBP relations with p1,2 · (∂/(∂k)) and (∂/(∂k)) · k we obtain the following three equations: a3 − a1 + a1 1+ (3− + m2 ) − a2 2+ (1− − 3− + Q2 − m2 ) −a3 3+ (1− − m2 ) = 0 , −

−

−

a3 − a2 + a2 2 (3 + m ) − a1 1 (2 − 3 + Q − m ) −a3 3+ (2− − m2 ) = 0 , +

2

+

−

2

d − a1 − a2 − 2a3 − (a1 1 + a2 2 )(3 + m ) − 2m a3 3 = 0 , +

+

2

(5.14)

2

2

+

(5.15) (5.16)

where Q2 = −q 2 = −(p1 − p2 )2 . Let us observe that the integrals (3.19) can be evaluated in terms of gamma functions if at least one of the indices is non-positive. In the case of a1 ≤ 0 or a2 ≤ 0, we can apply (A.6) and, in the case of a3 ≤ 0, we can apply (A.12). Let us now assume that all the indices are positive. Let us apply (5.14)–(5.16) to the general integral F (a1 , a2 , a3 ) and solve the corresponding linear system of the three equations with respect to F (a1 + 1, a2 , a3 ), F (a1 , a2 + 1, a3 ) and F (a1 , a2 , a3 + 1). We shall obtain an expression of these quantities in terms of integrals with the sum of the indices equal to a1 + a2 + a3 . Using the ﬁrst part of this solution we obtain a relation that expresses F (a1 , a2 , a3 ) in terms of integrals with a1 less by one and can be used in the case a1 > 1. Similarly, the second and the third parts of the solution give the possibility to reduce a2 > 1 and a3 > 1 to one. Therefore, we see that any given Feynman integral (3.19) can be reduced to I1 = F (1, 1, 1) and a family of simple integrals which can be expressed in terms of gamma functions. For example, we have F (1, 1, 2) =

(d − 4)(2m2 − Q2 ) I1 2m2 (m2 − Q2 )

2 1 Q (F (1, 2, 0) + F (2, 1, 0)) + 2 2 2 2m (m − Q )

−m2 (F (0, 1, 2) + F (0, 2, 1) + F (1, 0, 2) + F (2, 0, 1)) , (5.17)

5.1 One-Loop Examples

113

where all the integrals with an index equal to zero can be evaluated simply by (A.4) and (A.7) Observe that the coeﬃcient at I1 in (5.17) is proportional to ε. According to [18], where the reduction in the massless case was performed and in the case of general masses analysed, this is a general phenomenon, i.e. this property holds for any F (a1 , a2 , a3 ) with a1 + a2 + a3 > 3 in the case of general masses ml and indices. As a result, such integrals involve only elementary functions (no polylogarithms) in the expansion in ε up to the ﬁnite part – this was noticed very much time ago [35]. Let us again consider the massless on-shell boxes which we analysed in Examples 3.3 and 4.3. For convenience, we change the numbering of the lines as compared with Chaps. 3 and 4. Example 5.5. The massless on-shell box Feynman integrals of Fig. 5.1 with p2i = 0, i = 1, 2, 3, 4 and general integer powers of the propagators.

p1 3

p2

p3

1 4 2

p4

Fig. 5.1. Box diagram

Let us ﬁrst observe that whenever one of the indices is non-positive, the integrals can be evaluated in terms of gamma functions for general ε. In particular, if some index is zero, e.g., a4 = 0, one can apply (A.28). Suppose now that all the indices are positive. Starting from the IBP identity with the operator (∂/∂k)·k acting on the integrand and choosing the loop momentum k to be the momentum of each of the four lines, we obtain the following four IBP relations: a1 s1+ = a1 + 2a2 + a3 + a4 − d + (a1 1+ + a3 3+ + a4 4+ )2− = 0 , (5.18) a2 s2+ = 2a1 + a2 + a3 + a4 − d + (a2 2+ + a3 3+ + a4 4+ )1− = 0 , (5.19) a3 t3+ = a1 + a2 + a3 + 2a4 − d + (a1 1+ + a2 2+ + a3 3+ )4− = 0 , (5.20) a4 t4+ = a1 + a2 + 2a3 + a4 − d + (a1 1+ + a2 2+ + a4 4+ )3− = 0 , (5.21) where s = (p1 + p2 )2 and t = (p1 + p3 )2 are Mandelstam variables, as above. These equations can be used to reduce the indices al to one. For example, when (5.18) is applied to the general integral, we have, on the right hand side, terms with a1 less by one, with the exception of one term corresponding to a1 1+ 2− . This term, however, decreases a2 . Anyway, the sum of the indices corresponding to the right-hand side of (5.18)–(5.21) is less by one than corresponding to the left-hand side.

114

5 IBP and Reduction to Master Integrals

Therefore we come to the conclusion that any given Feynman integral F5.5 (a1 , a2 , a3 , a4 ) can be expressed linearly in terms of the master integral I1 ≡ F5.5 (1, 1, 1, 1) and a family of integrals where some indices are nonpositive. We again stop reduction here and do not try to reduce various integrals with non-positive indices to true master integrals. In Chap. 6, however, we will see what these true master integrals are.

5.2 Two-Loop Examples Let us now see how IBP relations can be used for the reduction of the massless Feynman integrals corresponding to Fig. 3.9. We have already considered these diagrams in Example 3.5 in Chap. 3. Example 5.6. Two-loop massless propagator Feynman integrals of Fig. 3.9 with integer powers of the propagators. First, we observe that if a5 = 0 the integrals over k and l decouple and can be evaluated in terms of gamma functions by use of (3.6): 2 F5.6 (a1 , a2 , a3 , a4 , 0) = (−1)a1 +a2 +a3 +a4 iπ d/2 ×

G(a1 , a2 )G(a3 , a4 ) . (−q 2 )a1 +a2 +a3 +a4 +2ε−4

(5.22)

When some other index al is zero, the integral becomes recursively oneloop (see Sect. 3.2.1), i.e. it can be evaluated in terms of gamma functions by successively applying the same one-loop formula, for example, 2 F5.6 (a1 , a2 , a3 , 0, a5 ) = (−1)a1 +a2 +a3 +a5 iπ d/2 ×

G(a3 , a5 )G(a2 , a1 + a3 + ε − 2) . (−q 2 )a1 +a2 +a3 +a5 +2ε−4

(5.23)

Suppose now that all the indices are positive integers. Let us write down the following IBP identity: kµ − lµ ∂ dd k dd l =0. (5.24) (l2 )a3 [(q − l)2 ]a4 ∂kµ (k 2 )a1 [(q − k)2 ]a2 [(k − l)2 ]a5 Taking derivatives, using identities such as 2k ·(k − l) = k 2 + (k − l)2 − l2 , and recognizing terms on the left-hand side as integrals (3.39), we arrive at the following relation: (5.25) (a1 + a2 + 2a5 − d) − a1 1+ 3− − 5− − a2 2+ 4− − 5− = 0 . Equation (5.25) can be used as a recurrence relation for the given family of integrals. Indeed, applying it to the general integral, we obtain

5.2 Two-Loop Examples

115

1 a1 + a2 + 2a5 − d × [a1 (F5.3 (a1 + 1, a2 , a3 − 1, a4 , a5 ) − F5.3 (a1 + 1, a2 , a3 , a4 , a5 − 1)) +{1 ↔ 2, 3 ↔ 4}] . (5.26)

F5.6 (a1 , a2 , a3 , a4 , a5 ) =

On the right-hand side, we encounter integrals where the sum a3 + a4 + a5 is less by one than that on the left-hand side. Thus, successive application of this relation reduces any given integral to integrals with some index equal to zero, where (5.22) and (5.23) can be used. In fact, in case one of the indices is negative, generalizations of the explicit formulae (5.22) and (5.23) can be derived. To do this, one applies (A.12). Therefore we come to the conclusion that any given integral (3.39) with integer indices can be evaluated in terms of gamma functions for general values of d. If we are not too curious we can stop our analysis at this point and not bother about the minimal number of master integrals. We could consider any integral with a non-positive index as a master integral because they can be expressed explicitly in terms of gamma functions. Otherwise it is necessary to continue to exploit IBP relations and obtain a solution of the reduction problem in the strict sense, i.e. with a minimal family of the master integrals. Usually, people are lazy in such situations and indeed stop the reduction. In this particular example, we shall see, in Chap. 6, what the true master integrals are. For example, the integral with all indices equal to one, is evaluated by means of (5.26) as follows: 1 [F5.6 (2, 1, 0, 1, 1) − F5.6 (2, 1, 1, 1, 0)] ε 1 1 = G(1, 1) [G(2, 1) − G(2, 1 + ε)] ε (−q 2 )1+2ε 2 2 4 iπ π + 12ζ(3) ε = 6ζ(3) + q2 10 4 π 2 + + (24 − π )ζ(3) + 42ζ(5) ε2 + . . . , 5

F5.6 (1, 1, 1, 1, 1) =

(5.27)

so that the well-known result [14, 42] at order ε0 is again (as in Sect. 3.5) reproduced. In this simple example, it was suﬃcient to use only one IBP relation which, in fact, follows from an IBP identity for the triangle diagram of Fig. 5.2 with general indices, m3 = 0 and general masses m1 and m2 . The general Feynman integral for this graph is dd k F (a1 , a2 , a3 ) = . (5.28) [(k + p1 )2 − m21 ]a1 [(k + p2 )2 − m22 ]a2 (k 2 )a3 Let us write down the IBP identity with the operator (∂/∂k) · k acting on the integrand of (5.28). Then we obtain the following ‘triangle’ rule:

116

5 IBP and Reduction to Master Integrals

Fig. 5.2. Triangle diagram with general integer indices

1=

1 d − a1 − a2 − 2a3

× a1 1+ 3− − (p21 − m21 ) + a2 2+ 3− − (p22 − m22 ) .

(5.29)

This identity can be applied to a triangle as a subgraph in a bigger graph. Suppose that the external upper right line in Fig. 5.2 has the mass m1 and the external lower right line has the mass m2 but these are internal lines for the bigger graph. Then the factors (p21 − m21 ) and (p22 − m22 ) eﬀectively reduce the indices of the corresponding lines (with the momenta p1 and p2 ) by one. For example, if we consider the triangle rule in the massless case and apply it to the left triangle in Fig. 3.9 we shall obtain (5.25). The triangle rule derived above is very well known. Let us derive another triangle rule from it. Consider the case where (p1 −p2 )2 = 0 and m1 = m2 = 0. Starting from the IBP identity with the operator (∂/∂k) · k acting on the integrand and choosing the loop momentum k to be the momentum of each of the three lines, we obtain the following three IBP relations: d − 2a1 − a2 − a3 − a2 2+ 1− − a3 3+ (1− − p21 ) = 0 , + −

−

d − a1 − 2a2 − a3 − a1 1 2 − a3 3 (2 0, d − a1 − a2 − 2a3 − a1 1+ (3− − p21 ) − a2 2 (3 − p22 ) = 0 . +

− p22 ) = + −

(5.30) (5.31) (5.32)

We form the combination (5.30) times a1 1+ plus (5.31) times a2 2+ minus (5.32) times a3 3+ and arrive at the following extra triangle relation: (d − 2a3 − 2)a3 3+ = (d − 2a1 − 2a2 − 2)(a1 1+ + a2 2+ ) .

(5.33)

There was a subtle point when multiplying quantities like 3+ and a3 which have algebraic properties similar to creation and annihilation operators. For example, the additional terms −2 in the brackets of (5.33) appear due to this multiplication. Consider now Example 5.7. Planar two-loop massless vertex diagrams with p21 = p22 = 0 and general integer powers of the propagators. The general scalar Feynman integral corresponding to Fig. 5.3 can be written as

5.2 Two-Loop Examples

117

p1 3 q

1 6

5

2 4 p2 Fig. 5.3. Planar vertex diagram

dd l (l2 )−a7 − 2p1 ·l)a1 (l2 − 2p2 ·l)a2 dd k × , (k 2 − 2p1 ·k)a3 (k 2 − 2p2 ·k)a4 (k 2 )a5 [(k − l)2 ]a6

F5.7 (a1 , . . . , a7 ) =

(l2

(5.34)

where k and l are loop momenta of the box and triangle subgraphs, respectively. There is one irreducible numerator, which cannot be expressed linearly in terms of the factors in the denominator, chosen as l2 . We are interested only in non-positive values of a7 . As it was mentioned in Chap. 3, the evaluation of such Feynman integrals by Feynman parameters is rather cumbersome. It turns out that using IBP provides the possibility to reduce any integral of this family to very simple integrals. As we will see shortly, any given integral can be expressed in terms of gamma functions for general values of d. We shall not, however, write down various IBP relations for (5.34). As it was noticed in [36] it is enough to use just one tool, the triangle rule (5.29), for the evaluation of these integrals. Suppose that all the indices a1 , . . . , a6 are positive and a7 = 0. Let us apply (5.29) to the triangle subgraph, i.e. with the lines (1, 2, 6). We obtain

1 1= a1 1+ 6− − 3− + a2 2+ 6− − 4− (5.35) d − a1 − a2 − 2a6 as acting on F5.7 (a1 , . . . , a6 , 0). Since the sum a1 + a2 + a6 on the right-hand side of the corresponding relation is less by one, it provides the possibility to reduce one of the indices a4 , a5 , a6 to zero. In the case where a6 = 0 the Feynman integral factorizes and is evaluated by (A.7) and (A.28): 2 F5.7 (a1 , . . . , a5 , 0, 0) = (−1)a1 +...+a5 iπ d/2 ×

G(a1 , a2 )G3 (a3 , a4 , a5 ) . (−q 2 )a1 +...+a5 +2ε−4

(5.36)

where the function G3 is deﬁned as the coeﬃcient of the right-hand side of (A.28) at iπ d/2 (−q 2 )−λ1 −λ2 −λ3 −ε+2 .

118

5 IBP and Reduction to Master Integrals

Suppose now that a3 or a4 is zero. Let it be a4 so that the line 4 is reduced to a point. Then we apply (5.29) to the triangle subgraph, with the lines (5, 6, 3). We obtain

1 a5 5+ 3− + a6 6+ 3− − 1− (5.37) 1= d − a5 − a6 − 2a3 as acting on F5.7 (a1 , a2 , a3 , 0, a5 , a6 , 0). (There is one term less as compared with (5.35) because of the on-shell condition p21 = 0.) This relation provides the possibility to reduce either a1 or a3 to zero. In both cases, resulting integrals become recursively one-loop and can be evaluated again by (A.7) and (A.28). We have 2 F5.7 (0, a2 , a3 , 0, a5 , a6 , 0) = (−1)a2 +a3 +a5 +a6 iπ d/2 G(a2 , a6 )G3 (a3 , a2 + a6 + ε − 2, a5 ) (5.38) (−q 2 )a2 +a3 +a5 +a6 +2ε−4 2 F5.7 (a1 , a2 , 0, 0, a5 , a6 , 0) = (−1)a1 +a2 +a5 +a6 iπ d/2 ×

×

G(a5 , a6 )G3 (a1 , a2 , a5 + a6 + ε − 2) . (5.39) (−q 2 )a1 +a2 +a5 +a6 +2ε−4

Therefore, any integral with positive indices can be evaluated by this procedure. For example, we reproduce the well-known result [28, 36, 41] for F5.7 (1, . . . , 1, 0): (iπ d/2 )2 1 1 G2 (2, 2)G3 (2 + ε, 1, 1) (Q2 )2+2ε ε 2ε 1 G3 (2, 1, 1 + ε) + G3 (1, 1, 1) −G2 (2, 1) ε d/2 −γE ε 2 1 ) 5π 2 29ζ(3) 3π 4 (iπ e + + O(ε) . (5.40) + + = (Q2 )2+2ε 4ε4 24ε2 6ε 32 In fact, a similar reduction procedure can be developed for general Feynman integrals with an irreducible numerator, i.e. for a7 < 0, and with general integer indices (not only positive). This can be done by using generalizations of the triangle rule to the case with a numerator. In fact, a general recursive procedure for such integrals (and integrals with another oﬀ-shell external momentum, p21 = 0 instead of q 2 = 0) with general numerators was developed in [20], with boundary integrals written in terms of terminating hypergeometric series of the unit argument. Another possibility in this situation is to get rid of the numerator and negative indices using the technique of shifting dimension which we will discuss shortly. Then we shall come back to this point. We now turn, following [3], to the two classes of integrals already studied in Chap. 4 which are partial cases of massless on-shell double boxes: boxes with a one-loop insertion and boxes with a diagonal shown in Fig. 5.4. For

5.2 Two-Loop Examples 1

p1 3

p2

3

p3 4

2

5

119

7

p4

(a)

6

5

2

(b)

Fig. 5.4. (a) Box with a one-loop insertion. (b) Box with a diagonal

convenience, we again change the numbering of the lines: In Fig. 5.4a we adjust it to that of Fig. 5.1 and, in Fig. 5.4b, to a new numbering for the double box which will be studied in the next section. So, the next is Example 5.8. Reduction of boxes with a one-loop insertion. Let us, ﬁrst, assume that we are dealing with the boxes with a one-loop insertion without numerator, B5.8 (a1 , . . . , a5 ) (In the given case, there are two independent scalar products that cannot be linearly expressed in terms of the denominators of the propagators.) In fact, the integration in the oneloop insertion in Fig. 5.4a can be taken explicitly by (A.7) and, graphically, this insertion can be replaced by a line with the index a4 + a5 + ε − 2 – see Fig. 3.1. Therefore, the problem reduces to the boxes of Fig. 5.1 in the case where the index of the line 4 is not integer. Still if one of the ﬁrst three indices is non-positive we obtain a quantity evaluated in terms of gamma functions by (A.28). Suppose now that a1 , a2 , a3 > 0. Then we can apply (5.18) and (5.19) to reduce a1 and a2 to one, as in the case of the box with integer indices. To take care of a3 let us form the new relation as a4 4+ times (5.20) minus a3 3+ times (5.21): (d − a1233 )a3 3+ = (d − a1244 − 2)a4 4+ +(a3 − a4 )(a1 1+ + a2 2+ ) ,

(5.41)

where we keep our notation of Chap. 4, e.g. a1233 = a1 + a2 + 2a3 etc. Observe now that (5.41) can be used to reduce the index a3 to one because a1 1+ and a2 2+ in the last term can be replaced immediately according to (5.18) and (5.19). Let us therefore assume that a1 = a2 = a3 = 1. Now we can apply (5.21), where the term with a3 3− gives integrals expressed in terms of gamma functions, to have control on a4 = a4 + ε which has an amount proportional to ε because of the one-loop integration. For example, one can shift a4 to a4 = 0: this choice corresponds to I1 = B5.8 (1, . . . , 1). In the case with numerators, one can get rid of them by shifting indices and dimension [47], as outlined in Subsect. 3.2.3. Then the previous procedure

120

5 IBP and Reduction to Master Integrals

provides the possibility to express any given box, with dimension d shifted by a positive even number, in terms of the master box with a one loop insertion I1 (d + 2n) in the same dimension and a family of simpler integrals expressed in terms of gamma functions. To complete this reduction procedure we need to know how to express these integrals in terms of I1 (d). To do this, let us apply the general relation (−1)h − U(α1 , . . . , αL ) , (5.42) d = π + αl →ial l

where U given by (2.24) is one of the two basic functions present in the alpha representation (2.36). (The factors (−1)h and 1/π come from the overall coeﬃcient in (2.36).) In particular, for Fig. 5.4a, this gives 1

a4 a5 4+ 5+ + (a1 1+ + a2 2+ + a3 3+ )(a4 4+ + a5 5+ ) . d− = (5.43) π We have d− I1 (d + 2) = I1 (d). On the other hand, applying the right-hand side of (5.43) to I1 (d + 2) we obtain a linear combination of integrals in dimension d + 2 with shifted indices for which we can use the reduction procedure described above. As a result, we obtain a desired linear relation of the type I1 (d) = A(d)I1 (d + 2) + B(d) , where A(d) is a rational function (of d, s and t) and B(d) comes from various integrals with some zero indices and can be evaluated in terms of gamma functions. Thus, any integral I1 (d + 2n) can be expressed recursively in terms of the master integral I1 (d) and a collection of simpler integrals. This completes our reduction procedure. Let us remember about the vertex diagrams of Example 5.7 which we considered without numerator. Now, we can get rid of any numerator as described above and then apply our reduction procedure formulated for nonnegative indices. However, since the corresponding results are expressed in terms of gamma functions for general d, there is no problem to make any shift d → d + 2n in them. We shall consider the reduction of the boxes with a diagonal in the next section.

5.3 Reduction of On-Shell Massless Double Boxes Let us turn, following [45], to Example 5.9. Reduction of on-shell massless double boxes. Let us follow the strategy [47] characterized in Subsect. 3.2.3 that enables us to express any integral with a numerator as a linear combination of integrals with shifted indices and dimension d. So, let us deal with Fig. 5.5 and the corresponding Feynman integrals

5.3 Reduction of On-Shell Massless Double Boxes

p1

1 7

p2

p3

3 6

2

121

5 4

p4

Fig. 5.5. Double box

dd k dd l (k 2 + 2p1 ·k)a1 (k 2 − 2p2 ·k)a2 (l2 + 2p1 ·l)a3 1 × 2 . (5.44) a 4 (l − 2p2 ·l) [(l + p1 + p3 )2 ]a5 [(k − l)2 ]a6 (k 2 )a7

K(a1 , . . . , a7 , d) =

where all indices al are non-negative. For convenience, we have changed the routing of the external momenta as well as the numbering of the lines in order to take into account the symmetry of the graph. (In Chap. 4, the numbering was oriented at insertions of boxes into double boxes.) Let us ﬁrst analyse situations, where one of the indices is zero. For a6 = 0, we obtain a product of two triangles which can be evaluated by (A.28) in terms of gamma functions. If a5 = 0 or a7 = 0 we obtain planar vertex diagrams analysed in Example 5.7. They are all evaluated in terms of gamma functions. Consider now the four symmetrical cases, where one of the other four indices is zero. Let it be a4 ; graphically, this means that the line 4 is contracted to a point – see Fig. 5.5. In this reduced graph, we can apply the triangle rule (5.29) to the resulting triangle with the lines 5, 6 and 3. After that we reduce either a3 or a1 to zero. Therefore, we arrive at a box with a one-loop insertion, in the former case, or a box with a diagonal, in the latter case – see Fig. 5.4. We conclude that, whenever one of the indices is zero, a given integral becomes a linear combination of the boxes with a one-loop insertion or a diagonal, or integrals expressed in terms of gamma functions. Let us call all these integrals boundary integrals. For the boxes with a oneloop insertion, we already know how to perform the reduction further, due to Example 5.8. Let us forget about this for a while and decide that all these boundary integrals are simple enough to stop the reduction here (as this was done in [45]). To perform the reduction for a given double box with positive indices, let ∂ ·(k − p2 ) which gives us start from the IBP relation with ∂k sa1 1+ = a7 7+ 2− + a6 6+ (2− − 4− ) + a1 1+ 2− −(d − 2a2 − a1 − a7 − a6 ) .

(5.45)

Three similar relations can be obtained from (5.45) by the two symmetry transformations: (1 ↔ 3, 2 ↔ 4, 5 ↔ 7) and (1 ↔ 2, 3 ↔ 4). The so-obtained four relations can be used to reduce the indices a1 , a2 , a3 , a4 to one. To reduce a5 to one we shall need one more IBP relation which is the ∂ · k times a5 5+ and the relation diﬀerence of the relation obtained with ∂k

122

5 IBP and Reduction to Master Integrals

obtained with

∂ ∂k ·(k

− l) times a6 6+ :

(d − a3455 − 2)a5 5+ = (d − a3466 2)a6 6+ + (a5 − a6 )(a3 3+ + a4 4+ ) +a3 a6 1− 3+ 6+ + a4 a6 2− 4+ 6+ . (5.46) The symmetrical relation applied to reduce a7 to one is (d − a1277 − 2)a7 7+ = (d − a1266 )a6 6+ + (a7 − a6 )(a1 1+ + a2 2+ ) +a1 a6 1+ 3− 6+ + a2 a6 2+ 4− 6+ .

(5.47)

Using the above recurrence relations we can bring the indices of the lines 1,2,3,4,5,7 all to one so that only a6 can now be greater than one. An appropriate relation for the reduction of a6 is [45] t(d − 6 − 2a6 )(a6 + 1)a6 6++ = t −(d − 5 − a6 ) 3d − 14 − 2a6 + 2a6 a6 6+ s 2 + (d − 4 − a6 )2 (d − 5 − a6 ) s % 2 t + (2+ + 7+ ) − (d − 4 − a6 )(d − 5 − a6 ) + 2 a26 6+ s s &

− 2t(a6 + 1)a6 6++ + 2(d − 4 − a6 )a6 6+ 3+ 1− +(d − 6)7− d− ,

(5.48)

−

where d is the operator that shifts dimension by −2, as before. This relation is valid only if it is applied to an integral with a1 = . . . = a5 = 1 and a7 = 1 (since some terms that are zero in this case are dropped out). The operator d− can be substituted explicitly using (5.42) with U = (α1 + α2 + α7 )(α3 + α4 + α5 ) +α6 (α1 + α2 + α3 + α4 + α5 + α7 ) ,

(5.49)

so that 1 + (1 + 2+ + 6+ )(3+ + 4+ + 5+ ) + 6+ (1+ + 2+ ) . (5.50) π The relation (5.48) can be derived as follows. Let us start with an integral with the numerator 2k · p2 . Since 2k · p2 = k 2 − (k 2 − 2p2 · k), such an integral is the diﬀerence of integrals where a7 or a2 is reduced by one. On the other hand, we can express this integral with the numerator in terms of integrals with shifted dimension and indices. Using an exponentiation of this numerator, similarly to how this is done for polynomials in the propagators (see (2.12)) and modifying the derivation of the alpha representation for the scalar double box in this case, we see (similarly to (3.16)) that the insertion of the numerator and shifting dimension by −2 can be described either by d− = −

5.3 Reduction of On-Shell Massless Double Boxes

123

the diﬀerence of the operators 7− − 2− times d− , or (up to a coeﬃcient with π) by the operator

(5.51) s a1 1+ (a6 6+ + a4 4+ + a5 5+ ) + a3 a6 3+ 6+ − ta5 a6 5+ 6+ . On the right-hand side of the so-obtained equation, we apply the reduction formulae (5.45)–(5.47) to reduce indices increased by the operators in (5.51). After some transformation, we then arrive at (5.48). Observe that on the left-hand side of (5.48) there is 6++ , rather than 6+ . This means that (5.48) enables us to reduce a6 to 1 or 2. Thus, after the application of the recurrence relations presented above, we reduce a given integral, up to our boundary integrals, to a linear combination of the two integrals, K1 (d) = K(1, 1, 1, 1, 1, 1, 1, d) and K2 (d) = K(1, 1, 1, 1, 1, 2, 1, d). However, these integrals generally appear, in the course of the reduction, in shifted dimensions so that we obtain the two families of integrals instead: K1 (d, n) = K1 (d + 2n) and K2 (d, n) = K2 (d + 2n) with K1 (d, 0) = K1 (d) and K2 (d, 0) = K2 (d). Of course, if we had results for general d for the master integrals (even expressed in terms of gamma functions), there would be no problem to shift the dimension in such analytical results. However, we are at a rather high level of complexity and are able to obtain results (at least for the master integrals) only in a Laurent expansion in ε, where expansions of the master integrals at d = 4 − 2ε and, say, at d = 6 − 2ε, when ε → 0, are not related to each other. To derive appropriate relations for the reduction of K1,2 (d, n) to K1,2 (d, 0), one can use the same trick with shifting dimension [47] as above, i.e. to write down equations K1,2 (d, n) = d− K1,2 (d, n + 1) with d− given by (5.50) and perform the reduction of the indices, which are increased after the action of d− , using (5.45)–(5.48). Solving the resulting linear system of equations one arrives at the following recurrence relations [45] which can be used to come back to dimension d = 4 − 2ε in the two master integrals: 1 ' (d) a22 K1 (d, n − 1) − f1 K1 (d, n) K1 (d, n) = ∆ $ (d)

, −a12 K2 (d, n − 1) − f2 K2 (d, n) ' 1 (d) K2 (d, n) = −a21 K1 (d, n − 1) − f1 K1 (d, n) ∆ $ (d)

+a11 (K2 (d, n − 1) − f2 K2 (d, n)) , (d)

where operators fj (d)

f1

(5.52)

(5.53)

are given by

2 + + (2 3 + 2+ 4+ + 2+ 6+ + 4+ 6+ + 4+ 7+ + 3+ 7+ ) s 2 4 + (2+ 5+ + 5+ 6+ + 5+ 7+ ) − 2 (d − 5)(3s + 2t)(2+ + 7+ ) s s t ! 2 2 + + + + + 3 6 7 − (3s(d − 5) + t(3d − 14)) 3 6 + 1− d−6 st(d − 6)

=

124

5 IBP and Reduction to Master Integrals

3 + 7− d− , t

(5.54)

2 + + 4 (2 3 + 2+ 4+ + 3+ 7+ + 4+ 7+ )6+ + (2+ + 4+ )6++ s s 1 2(2d − 13) + 4 + + + + + (2 + 7 + 2 6 )5 6 + +3 + 7+ 6++ s(d − 6) d−6 s 2(d − 5)(d − 7) (s(3d − 20) + 2t(d − 6)) (2+ + 7+ )6+ − 2 s t(d − 6)(d − 8) 2(d − 5)(d − 7) s (3s(3d − 20) + 4t(2d − 13)) 2+ + 7+ + 3+ 6+ + 2 2 s t (d − 8) d−6 ! 4 5d − 34 (3d − 20)(2d − 13) + + 3+ 6++ 1− d−8 s t(d − 6) ! 3d − 20 + d−7 6 − 2 (3s(3d − 20) + 4t(2d − 13)) 7− d− , (5.55) + t(d − 6) st (d − 8) (d)

f2

=

2 (d − 5)2 (3s + 2t), s2 t 3 2 = − (4d − 21) − (3d − 16), s t (d − 5)2 (d − 7) 8(2d − 13) 6(3d − 20) + , =− st(d − 8) s t d−7 2 3s (3d − 16)(3d − 20) + 6st(5d2 − 59d + 172) = 2 2 s t (d − 8) +4t2 (d − 5)(d − 6) ,

a11 =

(5.56)

a12

(5.57)

a21 a22

16(s + t)(d − 5) (d − 6)(d − 7) . s4 t(d − 8)

(5.58)

(5.59)

3

∆=

(5.60)

Thus, we are already able to reduce any double box to the two master integrals K1 (d) and K2 (d) and a family of our boundary integrals. For the ﬁrst master double box, K1 (d), we know the result given by (4.52) and (4.53), in expansion in ε, derived by MB representation in Chap. 4. To evaluate the second master double box, K2 (d), let us use alpha representation (2.36), where the function U is given by (5.49) and the second basic function (2.25) by V = [α1 α2 (α3 + α4 + α5 ) + α3 α4 (α1 + α2 + α7 ) +α6 (α1 + α3 )(α2 + α4 )] s + α5 α6 α7 t ,

(5.61)

We exploit this very simple dependence of this function on t to derive the following two relations by diﬀerentiating in t and implementing the factor α5 α6 α7 /U by shifting indices and dimension:

5.3 Reduction of On-Shell Massless Double Boxes

125

∂ 1 K(s, t; 1, . . . , 1, d) = − K(s, t; 1, 1, 1, 1, 2, 2, 2, d + 2) , (5.62) ∂t π 2 ∂ K(s, t; 1, 1, 1, 1, 1, 2, 1, d) = − K(s, t; 1, 1, 1, 1, 2, 3, 2, d + 2) . (5.63) ∂t π Then we apply the reduction procedure described above and express the righthand side of these equations in terms of the two master double boxes and a family of our boundary integrals (around ﬁfty terms in each case). In fact, the boundary integrals are simple enough here: a simple procedure based on the onefold MB representations (4.55) and (4.56) (see comments after these formulae) implemented on a computer can provide their ε-expansions up to order ε2 which is necessary here because the boundary integrals sometimes enter with coeﬃcients involving 1/ε2 . Then we insert (4.53) into (5.62) and use this equation to obtain a similar result for the second master double box. 2

K(1, 1, 1, 1, 1, 2, 1, d) =

(ie−γE ε ) f2 (t/s, ε) (−s)2+2ε t2

(5.64)

with

1 4 1 5 2 2 f2 (x, ε) = 4 − 5 (ln x − 2) 3 + 2 ln x − 14 ln x − (π + 4) 2 ε ε 2 ε 1 2 3 11 65 ln x + 8 ln2 x + π 2 + 14 ln x − 2 − 3π 2 − ζ(3) + 3 2 3 ε 2 2 4 3 88 − ln x(ln x + 1) − 2 3π + 4 ln x + 10 + 9π 2 + ζ(3) ln x 3 3 29 4 +20 + 12π 2 − π 4 + ζ(3) 30 3 7 1 21 2 1 +x − 3 + (8 ln x − 33) 2 + 26 ln x + 6 + π ε ε 2 ε 1 3 2 2 + −32 ln x − 4(21 + 26π ) ln x + 180 + 209π + 904ζ(3) 6 2

+ 2Li3 (−x) − 2 ln xLi2 (−x) − ln2 x + π 2 ln(1 + x) ε

−4x 8 (Li3 (−x) − ln xLi2 (−x)) − 4 ln2 x + π 2 ln(1 + x) +4 (S2,2 (−x) − ln xS1,2 (−x)) − 44Li4 (−x) +4 (ln(1 + x) + 6 ln x − 2) Li3 (−x) − ln2 x + π 2 ln2 (1 + x) 10 −2 ln2 x + 2 ln x ln(1 + x) − 4 ln x + π 2 Li2 (−x) 3 8 3 10 ln x + 4 ln2 x + π 2 ln x + 4π 2 − 4ζ(3) ln(1 + x) . (5.65) + 3 3

Proceeding in the same way with the second recurrence relation (5.63) and inserting there our analytical results for the two master double boxes we obtain the possibility to check these two results.

126

5 IBP and Reduction to Master Integrals

Although boxes with a one-loop insertion and a diagonal are simple quantities one can reduce them further. In the former case, the reduction was described in Example 5.8. Let us now do this for the latter case and consider, following [3], Example 5.10. Reduction of boxes with a diagonal shown in Fig. 5.4b. We imply that we have already got rid of the numerators as before, by shifting dimension and indices. Applying our auxiliary triangle rule (5.33) to the triangles (3, 5, 6) and (2, 7, 6) in Fig. 5.4b we obtain (d − 2a27 − 2)a2 2+ = (d − 2a6 − 2)a6 6+ − (d − 2a27 − 2)a7 7+ , (d − 2a35 − 2)a5 5+ = (d − 2a6 − 2)a6 6+ − (d − 2a35 − 2)a3 3+ .

(5.66) (5.67)

These relations can be used to reduce a2 and a5 to one. Then the following IBP relations derived in [3] can be used to reduce a3 and a7 to one: s(d − 2a35 − 2)a3 3+ = −(d − a356 − 1)(3d − 2a223567 ) +2(d − a356 − 1)a7 2− 7+ + (d − 2a6 − 2)a6 2− 6+ ,

(5.68)

t(d − 2a27 − 2)a7 7+ = −(d − a267 − 1)(3d − 2a235567 ) +2(d − a267 − 1)a3 5− 3+ + (d − 2a6 − 2)a6 5− 6+ .

(5.69)

To reduce a6 to one, the following relation valid for a2 = a3 = a5 = a7 = 1 and derived in [3] can be used: st(d − 2a6 − 2)a6 6+ = −(s + t)(d − a6 − 3)(3d − 2a6 − 10) +2(d − a6 − 3)(t2− 7+ + s2+ 7− ) + (d − 2a6 − 2)a6 6+ (t2− + s7− ) . (5.70) Finally, we have to express the master box with a diagonal, B5.10 (1, . . . , 1, d+ 2n), in the shifted dimension in terms of B5.10 (1, . . . , 1, d) which is given by (4.58) in expansion in ε. This can be done by the same trick with shifting dimension as above: we write down relation (5.42) for the box with a diagonal, i.e. where the function U is given by U = (α2 + α7 )(α3 + α5 ) + α6 (α2 + α3 + α5 + α7 ) ,

(5.71)

according to (2.24), and apply it to B5.10 (1, . . . , 1, d). Then we proceed exactly as in Example 5.8 and arrive at a desired recurrence relation. The algorithm presented above enables us to reduce any massless double box in terms of the two master integrals K1 and K2 , two master boxes with a one-loop insertion and a diagonal and a family of integrals (two-loop planar vertices and products of triangles) expressed in terms of gamma functions. As was pointed out later [27] the choice of the second master integral K2 as the integral with a dot on the sixth line brought complications in practical calculations because one obtained a linear combination of K1 and K2 with a coeﬃcient involving 1/ε, but the calculation of the master integrals in one more order in ε looked rather nasty (at that time ;-)). Two solutions of

5.4 Conclusion

127

this problem have appeared immediately. In [23], this very combination of the master integrals was indeed calculated using the method of diﬀerential equations (to be studied in Chap. 7), while in [5] another choice of the master integrals was made: instead of K(1, 1, 1, 1, 1, 2, 1, 0), the authors have taken the integral K(1, 1, 1, 1, 1, 1, 1, −1) as the second complicated master integral. This was a more successful choice because, according to the calculational experience, no negative powers of ε occur as coeﬃcients at these two new master integrals.

5.4 Conclusion When solving the problem of the reduction to master integrals, one tries to use all possible IBP relations. For h-loop Feynman integrals over the loop momenta ki depending on n independent external momenta pj , all possible IBP relations with derivatives (∂/∂ki ) · pj and (∂/∂ki ) · kj are used. For example, for the double boxes, this gives 10 IBP relations. In addition to the IBP relations, one can use the so-called Lorentz-invariance (LI) identities [24]. They follow from the fact that scalar Feynman integrals are invariant under inﬁnitesimal Lorentz transformations of the external momenta, pµi → pµi + εµν pνi . For example, in the case of four-point Feynman integrals (in particular, double boxes) with three independent external momenta, this provides the following relation, in addition to 10 IBP relations: 3 ∂ ∂ pn,µ ν − pn,ν µ = 0 (5.72) (pµ1 pν2 − pν1 pµ2 ) ∂pn ∂pn n=1 as well as the other two relations obtained by the cyclic permutations from (5.72). Well, if we turn to alpha or Feynman parameters, the Lorentz invariance becomes manifest and the equations (5.72) trivially hold (in contrast to the IBP relations), so that one might think that the LI equations follow from the IBP relations. However, explicitly, this statement has not been proven. Anyway, the LI identities can be certainly practically very useful. One can consider them together with the IBP relations and not bother about whether they are linear combinations of some IBP relations. There are a lot of papers where reduction problems for various classes of Feynman integrals were solved, in some way, with the help of IBP relations. Here is a very short list of some of them, starting from the two-loop level. Historically, IBP relations were ﬁrst successfully applied in [15] to threeloop massless propagators diagrams shown in Fig. 5.6. The corresponding algorithm [29] called MINCER was implemented in FORM [52]. In [12, 19, 22, 30], the problem of reduction for two-loop on-shell diagrams was solved: in [30], relevant recurrence relations were derived and used to ﬁnd all necessary integrals, and, in [12], a general algorithm implemented in the REDUCE [33]

128

5 IBP and Reduction to Master Integrals

Fig. 5.6. Three-loop massless planar, non-planar and Mercedez–Benz propagator diagrams

package Recursor was constructed. The reduction in the three-loop case was developed in [38] and, completely, in [39] with an implementation in FORM [52] (although no details of the reduction procedure were presented, as in many other cases). The reduction of two-loop bubble integrals with diﬀerent masses was solved in [21]. Three-loop vacuum diagrams with one mass were considered in [6, 12, 46]. The corresponding computer package MATAD was developed in [46]. The reduction problem for the massless on-shell double boxes in the nonplanar case (Fig. 4.9b where all lines are massless) was solved, using IBP and LI relations, in [2] and, in the case of (simpler) pentabox diagrams, in [3]. The general algorithm for the massless on-shell double boxes resulted in a series of NNLO calculations of various scattering processes – see, e.g., [26] for a review. The reduction of two- and three-loop propagator diagrams in Heavy Quark Eﬀective Theory was solved in [13, 31]. A pedagogical introduction to recursion problems oriented at HQET can be found in a recent review [32]. Unfortunately, the way how IBP relations are solved is not often explained. A typical example of such a situation is solving the reduction problem for two-loop vertex diagrams at threshold, q 2 = 4m2 : two independent algorithms were constructed [7, 17] but never published. The examples presented in this chapter and the papers cited above show how IBP relations can be solved without systematization. In other words, if it is necessary to solve a new problem, one can use the experience obtained in these examples and then analyse the new situation with the hope to solve somehow corresponding IBP relations. Still the complexity of unsolved calculational problems requires a systematization in this ﬁeld. One might hope that a systematization can be achieved within the technique based on shifting dimension [47]. Typical tricks were described in the previous section. Some prescriptions of this technique were presented in [48, 49]. Another example of its applications [43] is provided by the calculation of Feynman integrals relevant to the two-loop quark potential (to be considered within another technique in Chap. 6). It was also used to solve the reduction problem for two-loop propagator integrals with arbitrary masses [47]. Anyway, this technique provides the possibility to get rid of the numerators (which, of course, make the problem of the reduction more complicated) from the beginning.

5.4 Conclusion

129

Another attempt of a systematization was initiated in [25, 37, 38]. It is based on the observation that the total number of IBP and Lorentz invariance equations grows faster than the number of independent Feynman integrals, labelled by the powers of propagators and the powers of independent scalar products in the numerators, when the total dimension of the denominator and numerator in Feynman integrals associated with the given graph is increased. Therefore this system of resulting equations sooner or later becomes overconstrained, and one obtains the possibility of performing a reduction to master integrals. To be formal let us modify our notation for the Feynman integrals a little bit. Consider now, as a general Feynman integral, b H1b1 . . . HNN22 , (5.73) F (a1 , . . . , aN1 ; b1 , . . . , bN2 ) = · · · dd k1 . . . dd kh a1 a E1 . . . ENN1 1 instead of the dimensionally regularized version of (2.6). Now, we consider all the indices ai and bi to be positive or zero, both in the denominator and numerator. As before, all the quantities Ei and Hi are considered linear or quadratic with respect to the loop momenta. So, the idea [37, 38] is to write all possible IBP and LI relations for Feynman integrals (5.73) with a ﬁxed N1 + N2 = N . Our experience tells us that starting from some large N this will be an overconstrained linear system of equations which will be solved successfully (using a computer, of course). A breakthrough in the implementation of this idea came due to the following two publications: the ﬁrst practical successful implementation was achieved for the reduction of massless double box diagrams with one leg oﬀ-shell [25] (which was applied for NNLO calculations of the process e+ e− → 3jets – see [40] for a review), and detailed prescriptions for the implementation of this method in a general situation were presented in [37]. These two important works have resulted in a series of various calculations at the two-loop level – see, e.g. [1, 8, 9, 10, 11, 16, 44]. The implementation of this method on a computer in non-trivial situations was hardly possible, say, ten years ago. Indeed, for example, in the case of the double boxes with one leg oﬀ-shell, it was necessary [25] to solve linear systems of dozens of thousands of equations for dozens of thousands of variables. It is not clear at the moment what the practical limits of applications of this algorithm are, for example, whether it can be applied successfully to such problems as the reduction of triple boxes or four-loop massless propagator diagrams. This method is rather pragmatic and is a kind of experimental mathematics because its analysis from the mathematical point of view is absent. In particular, it is not known which linear equations of the method are really independent. It is not clear in advance which will be master integrals in a given problem: this becomes clear after solving the corresponding system of equations. The authors of [1, 8, 9, 10, 11, 16, 25, 44] constructed various computer implementations of this method. Fortunately, a ﬁrst public version called AIR

130

5 IBP and Reduction to Master Integrals

which can be applied to any problem, with the hope to obtain a solution of a concrete reduction problem, has recently appeared [4]. Now, to solve a new reduction problem, one can try to adjust this general computer algorithm, rather than solve IBP relations oneself. Well, if it turns out that this algorithm does not work, for some reasons (e.g. the lack of time or computer memory), then one could still try to solve the reduction problem in some way. One more option is described in the next chapter, where we will study a method which does not resemble any previously developed technique in this ﬁeld. The explicit and detailed recipes for solving overdetermined systems of equations presented in [37] are more optimal than the simple Gauss elimination. In fact, the Gauss elimination is present there, but only after the initial system is ordered according to some criteria. Then diﬀerent terms of the equations are characterized by a relative weight of their complexity, and the equations are solved starting from the most complicated terms. One could still look for more optimal strategies. In particular, one could hope to use a Gr¨ obner basis in this situation. This idea was already discussed in [48, 50] and applied to the case of two-loop propagator integrals with general masses. In this case, it was possible to use an existing Gr¨obner basis for diﬀerential equations with coeﬃcients independent of the arguments because, in the case of general non-zero masses, the initial problem of solving IBP type equations can be reduced to solving some systems of diﬀerential equations. Unfortunately, one usually needs physical cases, where zero masses are unavoidable. Solving the reduction problem with general non-zero masses and taking a massless (and on-shell or threshold) limit in the corresponding solution is not a natural procedure, because coeﬃcients at master integrals in this general solution are singular in these limits. Another variant here is to try to construct a Gr¨ obner basis adequate to deal with IBP equations which are diﬀerence equations with respect to the indices. This is, however, an open mathematical problem.

References 1. U. Aglietti and R. Bonciani, Nucl. Phys. B 668 (2003) 3; hep-ph/0401193; U. Aglietti, R. Bonciani, G. Degrassi and A. Vicini, Phys. Lett. B 595 (2004) 432; hep-ph/0407162. 129 2. C. Anastasiou, T. Gehrmann, C. Oleari, E. Remiddi and J.B. Tausk, Nucl.Phys. B 580 (2000) 577. 128 3. C. Anastasiou, E.W.N. Glover and C. Oleari, Nucl. Phys. B 575 (2000) 416 [Erratum-ibid. B 585 (2000) 763]. 118, 126, 128 4. C. Anastasiou and A. Lazopoulos, JHEP 0407 (2004) 046. 130 5. C. Anastasiou, J.B. Tausk and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 262. 127 6. L.V. Avdeev, Comput. Phys. Commun. 98 (1996) 15. 128 7. M. Beneke, A. Signer and V.A. Smirnov, Phys. Rev. Lett. 80 (1998) 2535. 128

References

131

8. W. Bernreuther et al., hep-ph/0406046. 129 9. T.G. Birthwright, E.W.N. Glover and P. Marquard, hep-ph/0407343. 129 10. R. Bonciani, A. Ferroglia, P. Mastrolia, E. Remiddi and J. J. van der Bij, Nucl. Phys. B 681 (2004) 261. 129 11. R. Bonciani, P. Mastrolia and E. Remiddi, Nucl. Phys. B 661 (2003) 289; B 676, 399 (2004); B 690, 138 (2004). 129 12. D.J. Broadhurst, Z. Phys. C 54 (1992) 599. 127, 128 13. D.J. Broadhurst and A.G. Grozin, Phys. Lett. B 267 (1991) 105. 128 14. K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Nucl. Phys. B 174 (1980) 345. 115 15. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 109, 127 16. M. Czakon, J. Gluza and T. Riemann, hep-ph/0406203. 129 17. A. Czarnecki and K. Melnikov, Phys. Rev. Lett. 87 (2001) 013001. 128 18. A.I. Davydychev, J. Phys. A 25 (1992) 5587. 113 19. A.I. Davydychev and A.G. Grozin, Phys. Rev. D 59 (1999) 054023. 127 20. A.I. Davydychev and P. Osland, Phys. Rev. D 59 (1999) 014006. 118 21. A.I. Davydychev and J.B. Tausk, Nucl. Phys. B 397 (1993) 123. 128 22. J. Fleischer and O.V. Tarasov, Comput. Phys. Commun. 71 (1992) 193; J. Fleischer and M.Yu. Kalmykov, Comp. Phys. Comm. 128 (2000) 531. 127 23. T. Gehrmann and E. Remiddi, Nucl. Phys. Proc. Suppl. 89 (2000) 251. 127 24. T. Gehrmann and E. Remiddi, Nucl. Phys. B 580 (2000) 485. 127 25. T. Gehrmann and E. Remiddi, Nucl. Phys. B 601 (2001) 248; Nucl. Phys. B 601 (2001) 287. 129 26. E.W.N. Glover, Nucl. Phys. Proc. Suppl. 116 (2003) 3. 128 27. E.W.N. Glover and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 196. 126 28. R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542. 118 29. S.G. Gorishny, S.A. Larin, L.R. Surguladze and F.V. Tkachov, Comput. Phys. Commun. 55 (1989) 381; S.A. Larin, F.V. Tkachov and J.A.M. Vermaseren, Preprint NIKHEF-H/91-18 (Amsterdam 1991). 127 30. N. Gray, D.J. Broadhurst, W. Grafe and K. Schilcher, Z. Phys. C 48 (1990) 673; D.J. Broadhurst, N. Gray and K. Schilcher, Z. Phys. C 52 (1991) 111. 127 31. A.G. Grozin, JHEP 0003 (2000) 013. 128 32. A.G. Grozin, Int. J. Mod. Phys. A 19 (2004) 473. 128 33. A.C. Hearn, REDUCE User’s Manual, Version 3.7 (ZIB, Berlin, 1999). 127 34. G. ’t Hooft and M. Veltman, Nucl. Phys. B 160 (1979) 151. 109 35. G. K¨ allen and A. Wightman, Mat. Fys. Skr. Dan. Vid. Selsk. 1 (No.6) (1958) 1. 113 36. G. Kramer and B. Lampe, J. Math. Phys. 28 (1987) 945. 117, 118 37. S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 129, 130 38. S. Laporta and E. Remiddi, Phys. Lett. B 379 (1996) 283. 128, 129 39. K. Melnikov and T. van Ritbergen, Phys. Lett. B 482 (2000) 99; Nucl. Phys. B 591 (2000) 515. 128 40. S. Moch, P. Uwer and S. Weinzierl, Nucl. Phys. Proc. Suppl. 116 (2003) 8. 129 41. W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 118 42. J.L. Rosner, Ann. Phys. 44 (1967) 11. 115 43. Y. Schr¨ oder, Phys. Lett. B 447 (1999) 321; Ph.D. thesis (Hamburg, 1999), DESY–THESIS–1999–021. 128 44. Y. Schr¨ oder, Nucl. Phys. Proc. Suppl. 116 (2003) 402; Y. Schr¨ oder and A. Vuorinen, hep-ph/0311323. 129 45. V.A. Smirnov and O.L. Veretin, Nucl. Phys. B 566 (2000) 469. 120, 121, 122, 123

132 46. 47. 48. 49. 50. 51.

5 IBP and Reduction to Master Integrals

M. Steinhauser, Comput. Phys. Commun. 134 (2001) 335. 128 O.V. Tarasov, Nucl. Phys. B 480 (1996) 397; Phys. Rev. D 54 (1996) 6479. 119, 120, 123, 128 O.V. Tarasov, Acta Phys. Polon. B 29 (1998) 2655. 128, 130 O.V. Tarasov, Nucl. Phys. (Proc. Suppl.) 89 (2000) 237. 128 O.V. Tarasov, hep-ph/0403253. 130 A.N. Vassiliev, Yu.M. Pis’mak and Yu.R. Khonkonen, Teor. Mat. Fiz. 47 (1981) 291. 109 52. J.A.M. Vermaseren, Symbolic Manipulation with FORM (CAN, Amsterdam, 1991). 127, 128

6 Reduction to Master Integrals by Baikov’s Method

In the previous chapter, we solved IBP relations [7] in a non-systematic way. Now we are going to do this systematically following Baikov’s method1 [2, 4, 5, 14]. Our goal is to solve the reduction problem, i.e. to develop an algorithm that would enable us to express any Feynman integral of a given family of Feynman integrals which are labelled by powers of the propagators (indices) as a linear combination of some master integrals. A characteristic feature of this method is the reduction to a minimal number of master integrals. In Sect. 6.1, the basic parametric representation which is an essential ingredient of this method will be described. In Sect. 6.2, this representation will be applied to formulate a strategy for identifying master integrals and constructing the corresponding coeﬃcient functions. As usual, we shall end up, in Sects. 6.2 and 6.3, with a lot of instructive examples starting from very simple ones. We shall continue to use mainly the examples considered in the previous chapters. In conclusion, applications and open problems of the method will be characterized.

6.1 Basic Parametric Representation Suppose that we are dealing with a family d d k1 . . . dd kh F (a) = · · · aN , E1a1 . . . EN

(6.1)

of h-loop dimensionally regularized Feynman integrals, where the factors in the denominator are given by (2.7) with r = 1, . . . , N = h(h + 1)/2 + hn. The denominators are quadratic or linear with respect to the loop momenta pi = ki , i = 1, . . . , h, and the independent external momenta ph+1 , . . . , ph+n of the graph. The ai are integer indices. Underlined letters denote collections of variables, i.e. a = (a1 , . . . , aN ), etc. 1 In [2], it was characterized as a ‘non-recursive’ solution of IBP recurrence relations. As we will see shortly, solving some recurrence relations is necessary within this method. However, these auxiliary recurrence relations are simpler than the initial IBP recurrence relations for a given family of Feynman integrals.

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 133–163 (2004) c Springer-Verlag Berlin Heidelberg 2004

134

6 Reduction to Master Integrals by Baikov’s Method

Some of the factors in the denominator are associated with irreducible numerators (which cannot be expressed linearly in terms of the given set of the denominators), so that the corresponding indices ai are considered only non-positive. We are going to solve the reduction problem in a maximal way, i.e. to be able to represent a given Feynman integral as a linear combination of a minimal number of some true master (or, irreducible) integrals, ci (a)Ii , (6.2) F (a) = i

with the natural normalization conditions ci (Ij ) = δij

(6.3)

which simply mean that any master integral cannot be expressed in terms of other master integrals. In fact, the master integrals are integrals of the given family, Ii = F (ai ), where ai = (ai1 , . . . , aiN ) are some concrete sets of indices. In the approach under consideration, the master integrals have indices air equal to one, or zero, or a negative value. Mathematically, if the reduction problem has been solved, we know a basis in the linear space of the given Feynman integrals. Then we could turn to some other basis. In particular, we could choose all the master integrals which have only positive indices. Consider, for example, the propagator integrals of Example 5.3 and choose, instead of I1 = F (1, 1) and I2 = F (1, 0), say, I1 = F (1, 1) and I2 = F (2, 1), why not? Well, practically, this is an unnatural choice. According to our experience of solving IBP relations and our standard attempts to reduce complicated integrals to simpler integrals, we imply that the master integrals must have as many non-positive indices as possible, so that we always keep this hierarchy in mind. Therefore, when we say that a given integral is irreducible, we omit the words to simpler integrals, in this sense, i.e. that have more non-positive indices. Our experience of solving IBP recurrence relations, in particular, the examples of Chap. 5, shows that the coeﬃcient functions ci (a) are rational functions of everything, i.e. of dimension, masses and external kinematical invariants. This property is a useful postulate that can be used in the calculation of the coeﬃcient functions. Within the approach of [2, 14], every coeﬃcient function in (6.2) satisﬁes, by construction, the initial IBP relations for (6.1) so that these relations for the given Feynman integrals are automatically satisﬁed. Let us start with the case of vacuum Feynman integrals which are functions of some masses and are deﬁned by (6.1) with 2 Aij (6.4) Er = r ki · kj − mr , h≥i≥j≥1

with r = 1, . . . , N = h(h + 1)/2.

6.1 Basic Parametric Representation

135

The IBP relations in the vacuum case originate from the following N equations: kj ∂ d d · = 0, i ≥ j . (6.5) · · · d k1 . . . d kh aN ∂ki E1a1 . . . EN We proceed, in this general situation, like in multiple examples in the previous chapter, i.e. perform diﬀerentiation and then express the resulting scalar products ki ·kj in terms of the denominators Er . When we invert the relations (6.4) we obtain a matrix which is inverse, in some sense, to the matrix Aij r . So, we write down the IBP relations in the following form: ji − A¯ir i A˜r r + m2r ar r+ = (d − h − 1)δij /2 , (6.6) r,r ,i ij ij ji where A¯ij r = Ar for i = j, Ar /2 for i > j and Ar /2 for i < j. The matrix A˜ is deﬁned as follows. Take the quadratic N × N matrix A, where the ﬁrst index is labelled by pairs (i, j) with i ≥ j, and the second index is r. The corresponding inverse matrix (A−1 )ij r (with i ≥ j) satisﬁes N

−1 i j Aij )r = δii δjj . r (A

(6.7)

r=1 −1 ij )r to all values i, j. Then A˜ij r is the symmetrical extension of (A + − Moreover, the operators r and r in (6.6) are our usual operators that increase and lower indices:

r+ F (. . . , ar , . . .) = F (. . . , ar + 1, . . .) , r− F (. . . , ar , . . .) = F (. . . , ar − 1, . . .) .

(6.8a) (6.8b)

We extensively exploited these operators in Chap. 5 for various concrete values of r. To construct the coeﬃcient functions ci (a) in the vacuum case, the following basic representation [2] is applied: dx1 . . . dxN (d−h−1)/2 [P (x )] , (6.9) ... xa1 1 . . . xaNN where the parameters x = (x1 , . . . , xN ) are obtained from x = (x1 , . . . , xN ) by the shift xi = xi + m2i . Integration over the parameters xi is understood in some way, with the requirement that the IBP in this parametric integral is valid. In this case, such objects satisfy the initial IBP relations (6.6). This property can be veriﬁed straightforwardly if we take into account that the operator ar r+ is transformed into the diﬀerential operator ∂/∂xr and the operator r− is transformed into the multiplication by xr . Now, the basic polynomial P of x which enters (6.9) is [2] N

ij ˜ A xr . (6.10) P (x) = det ij

r

r=1

136

6 Reduction to Master Integrals by Baikov’s Method

Here are simple practical prescriptions for evaluating the basic polynomials: 1. Solve the system Aij r ki · kj = Er ,

r = 1, . . . , N

i≥j≥1

with respect to ki · kj , i ≥ j; 2. Replace Er by xr on the right-hand side of this solution; 3. Extend this expression to all values of i and j in the symmetrical way; 4. Take the determinant of this matrix to obtain P . In fact, the basic polynomial is deﬁned up to a normalization factor independent of the variables xj . This will be clear when constructing the coefﬁcient functions which will be themselves normalized at some point. For general Feynman integrals, the problem can be reduced to the vacuum case [2, 4]. If there is one external momentum, q, so that we are dealing with a family of propagator-type integrals, one involves into the game coeﬃcients of the Taylor expansion of F (a) in q 2 , F (q ; a1 , . . . , aN ) ∼ 2

∞

(q 2 − m2N +1 )aN +1 −1 F (a1 , . . . , aN , aN +1 ) .

aN +1 =1

(6.11) It turns out [2, 4] that the so deﬁned objects F (a1 , . . . , aN , aN +1 ) (with some overall rescaling factor which is not important in the examples in this chapter) satisfy vacuum IBP relations. To formulate a prescription for corresponding basis polynomials in the non-vacuum case, we need ﬁrst to present a preliminary discussion of constructing master integrals. To identify candidates for master integrals in a ﬁrst approximation, we shall analyse integrals where the indices corresponding to irreducible numerators are set to zero and other indices are either zero or one. Let F (ai ) with aij = 1 or 0 be a candidate to be considered as a master integral. Let us remember the examples of Chap. 5, where the reduction always goes down: our experience tells us that a master integral Ii = F (ai ) = F (ai1 , . . . , air , . . . , aiN ) never appears in the decomposition of a given Feynman integral in terms of master integrals F (a) = . . . + ci (a1 , . . . , ar , . . . , aN )Ii + . . . if ar ≤ 0 and air > 0. Therefore, we come to the natural condition for the coeﬃcient function ci (a) of F (ai ): if air = 1 then ci (a1 , . . . , ar , . . . , aN ) = 0 for ar ≤ 0. This condition can be realized easily [2] in an automatic way by treating the integration over xj as a Cauchy integral around the origin in the complex xj -plane,

6.1 Basic Parametric Representation

1 2πi

(

dxj a xj j

137

. . . [P (x)]

(d−h−1)/2

.

(6.12)

According to the Cauchy theorem, this expression reduces to the Taylor expansion of order aj − 1 of the integrand in xj so that it becomes a linear combination of terms dxj z−nd (6.13) . . . [Pi (x)] n , x j j:a ≤0 j ij

where z = (d − h − 1)/2, and Pi (x) is obtained from P (x) by setting to zero all the variables xj with j such that aij = 1. We shall use nj instead of aj for powers of xj in auxiliary parametric integrals. Observe that the parameter nd in such integrals plays the role of the shift of the dimension. Suppose that we are not interested in higher terms of the Taylor expansion in powers of (q 2 − m2N +1 ) in (6.11), i.e. we need just the value at q 2 = m2N +1 , i.e. the term with aN +1 = 1. Then the integration over xN +1 should be understood in the sense of Cauchy integration so that, eﬀectively, xN +1 is set to zero. So, if Pˆ (x1 , . . . , xN , xN +1 ) is the basic polynomial for the corresponding vacuum problem, then the basic polynomial for the initial propagator-type problem is obtained as P (x) ≡ P (x1 , . . . , xN ) = Pˆ (x1 , . . . , xN , 0) .

(6.14)

In the case of n independent external momenta q1 , . . . , qn , one includes into the procedure all the terms of the formal Taylor expansions in the scalar products qi · qj . One is usually interested only in the value at some qi · qj and not in the derivatives at these points. (Otherwise, it would be necessary to deal with a generalization of (6.11), where the initial Feynman integrals are rescaled by the Gram determinant det(pi · pj ) which is raised to the power (h + n + 1 − d)/2 – see [2, 4].) Then the transition to the vacuum problem, which eﬀectively increases the number of loops, h → h + n, can be performed as follows: 1. Introduce a complete set of invariants by considering, in addition to ki ·kj , i ≥ j and ki · qj , also invariants generated by the external momenta, i.e. the scalar products qi · qj , i ≥ j. Let pi = ki , i = 1, . . . , h and pi = qi , i = h + 1, . . . , h + n so that the total number of the kinematical invariants ˆ = (h + n)(h + n + 1)/2. becomes N 2. Introduce, in some way, the corresponding new propagators. 3. Solve the system ˆ Aij r pi · pj = Er , r = 1, . . . , N i≥j≥1

with respect to pi · pj . 4. Evaluate the basic polynomial Pˆ for such a vacuum problem. 5. Obtain P (x) ≡ P (x1 , . . . , xN ) = Pˆ (x1 , . . . , xN , 0, . . . , 0).

138

6 Reduction to Master Integrals by Baikov’s Method

Observe that the method under consideration is based only on the IBP relations so that the LI identities discussed in Sect. 5.4 are not used at all.

6.2 Constructing Coeﬃcient Functions. Simple Examples Now, we want to apply the basic parametric representation for two closely related purposes: – identifying master integrals, – constructing the corresponding coeﬃcient functions. According to the discussion above, let us consider integrals where the indices corresponding to irreducible numerators are set to zero and other indices are either zero or one. Let Ii = F (ai ) = F (ai1 , . . . , air , . . . , aiN ). For indices equal to one, we understand the corresponding integration over xj in the basic parametric representation (6.9) in the Cauchy sense. This leads to a Taylor expansion of order aj − 1 of the integrand in xj and gives a linear combination of (6.13). Let us try to understand whether a given candidate can be considered as a master integral. Suppose that Pi = 0. Then there is no other way as to consider the coeﬃcient function equal to zero. Therefore, this integral cannot be a master integral and has to be recognized as a reducible integral within the reduction problem. Let us assume a weaker condition: the parametric integral involves an integral without scale which we put, by deﬁnition, to zero. Then, again, we cannot construct the coeﬃcient function in a non-trivial way so that the corresponding integral is considered reducible. Let us stress that such a scaleless integral can appear not only immediately but also after some preliminary non-trivial integrations. After such analysis, we obtain a preliminary list of master integrals. Sometimes one has to consider master integrals which diﬀer from F (ai ) by some indices aij < 0. The number of such additional master integrals is connected with the degree of the polynomial Pi with respect to some of the parameters xj . Let us now turn to examples and see how the basic parametric representation enables us to solve the reduction problem. Many examples will be the same as in Chap. 5, in particular, the ﬁrst one. Example 6.1. One-loop vacuum massive Feynman integrals given by the right-hand side of (5.1). We have one propagator with the denominator E = k 2 − m2 and one kinematical invariant k 2 . The equation E = k 2 is solved as k 2 = E. Therefore, the resulting basic polynomial is P (x) = x and the polynomial that enters

6.2 Constructing Coeﬃcient Functions. Simple Examples

139

(6.9) is P (x ) = x + m2 . There is one master integral I1 = F6.1 (1) given by the right-hand side of (5.6). According to (6.9) the corresponding coeﬃcient function is ( dx 1 dx 2 (d−2)/2 c(a) ∼ (x + m ) = (x + m2 )(d−2)/2 . (6.15) xa 2πi xa At a = 1 we have ( 1 dx (x + m2 )(d−2)/2 = (x + m2 )(d−2)/2 = (m2 )(d−2)/2 . 2πi x x=0 To satisfy the normalization c(1) = 1 we deﬁne ( (m2 )(2−d)/2 dx c(a) = (x + m2 )(d−2)/2 2πi xa a−1 ' $ (m2 )(2−d)/2 ∂ (x + m2 )(d−2)/2 = . (a − 1)! ∂x x=0

(6.16)

for a = 1, 2, . . .. So, we have F6.1 (a) = c(a)I1 , in agreement with (5.5) and the explicit result (A.1). As in Chaps. 3 and 5 let us consider Example 6.2. Massless one-loop propagator Feynman integrals given by the right-hand side of (5.7). The transition to the corresponding vacuum problem reduces to adding a new propagator, 1/(q 2 − m2 )a3 , with an eﬀective mass m. The eﬀective number of loops that is involved in the exponent in (6.9) is h = 2. We want to consider the value of our diagram at some general point and are not interested in higher terms of the Taylor expansion in q 2 . Therefore, we consider only the value a3 = 1 so that, according to our agreements, the integration contour for the corresponding variable x3 is taken as a Cauchy contour around the origin, and x3 is set to zero. Thus, using (6.14), we obtain the basic polynomial P (x1 , x2 ) = (q 2 )2 − 2q 2 (x1 + x2 ) + (x1 − x2 )2 .

(6.17)

The only possible candidate for a master integral is I1 = F6.2 (1, 1) = iπ d/2 (−q 2 )d/2−2

Γ (2 − d/2)Γ 2 (d/2 − 1) . Γ (d − 2)

(6.18)

because integrals with one non-positive index are zero. The corresponding coeﬃcient function is 2 (d−3) q c1 (a1 , a2 ) = (a1 − 1)!(a2 − 1)! a1 −1 a2 −1 ∂ ∂ (d−3)/2 [P (x1 , x2 )] , (6.19) × ∂x1 ∂x2 xi =0

140

6 Reduction to Master Integrals by Baikov’s Method

where the normalization condition c1 (1, 1) = 1 was immediately implemented. One can check that this result is in agreement with what we had in Example 5.2 when explicitly solving recurrence relations. Let us now turn to Example 6.3. One-loop diagram for the heavy quark potential shown in Fig. 6.1.

3

2

1 Fig. 6.1. One-loop diagram for the heavy quark potential. A wavy line denotes a propagator for the static source

The corresponding general Feynman integral is dd k , F6.3 (a1 , a2 , a3 ) = (k 2 )a1 [(k − q)2 ]a2 (v·k + i0)a3

(6.20)

with v·q = 0. In addition to k 2 , q · k and v · k, we consider q 2 , v · q and v 2 as external kinematical invariants so that the eﬀective loop number is h = 3. The choice of additional propagators is arbitrary. We choose the following extended set of the denominators: E1 = k 2 , E2 = (k − q)2 , E3 = k·v + v 2 , E4 = v 2 , E5 = q 2 , E6 = (q + v)2 . The basic polynomial is given by the determinant of the matrix (x1 − x2 + x5 )/2 x3 − x4 x1 (x1 − x2 + x5 )/2 x5 (−x4 − x5 + x6 )/2 . (−x4 − x5 + x6 )/2 x4 x3 − x4

(6.21)

(6.22)

The variables xi are then shifted by the corresponding eﬀective masses, x3 → x3 + v 2 , x4 → x4 + v 2 , x5 → x5 + q 2 , x6 → x6 + (q + v)2 . We are not interested in higher order Taylor coeﬃcients of the additional kinematical invariants so that, eﬀectively, we set x4 = x5 = x6 = 0. Thus, we obtain

2 P (x1 , x2 , x3 ) = (q 2 )2 v 2 + v 2 (x1 − x2 ) + 2q 2 v 2 (x1 + x2 ) − 2x23 , Observe that integrals (6.20) are zero whenever a1 or a2 are non-positive. After analysing various integrals with the indices 1 and 0 and corresponding

6.2 Constructing Coeﬃcient Functions. Simple Examples

141

reduced polynomials we see that the coeﬃcient functions can be constructed non-trivially for the following two integrals which can be evaluated by (A.27) and which we consider as master: I1 = F6.3 (1, 1, 1)

√ (−q 2 )d/2−5/2 π Γ (5/2 − d/2)Γ (d/2 − 3/2)2 , v Γ (d − 3) Γ (2 − d/2)Γ (d/2 − 1)2 . I2 = F6.3 (1, 1, 0) = iπ d/2 (−q 2 )d/2−2 Γ (d − 2) = −iπ d/2

(6.23) (6.24)

The coeﬃcient function c1 is simply calculated without integration. For the coeﬃcient function c2 , we need the following integrals: a α g1 (k3 , α) = dx3 xk33 a2 − x23 . (6.25) −a

Here k3 is an integer but α depends on d. This integral can be interpreted in the sense of the principal value, with 2 α+k/2+1/2 Γ (k/2 + 1/2)Γ (α + 1) a for even k . (6.26) g1 (k, α) = Γ (α + k/2 + 3/2) 0 for odd k Let us imply that these and similar integrals below are understood as convergent integrals in an appropriate domain of analytical parameters, such as α in (6.26), with analytic continuation to the whole complex plane of α on the right-hand side. We obtain the following decomposition of the general integral of the given class: F6.3 (a1 , a2 , a3 ) = c1 (a1 , a2 , a3 )I1 + c2 (a1 , a2 , a3 )I2 .

(6.27)

One can check that this procedure is in agreement with the explicit result (A.27) evaluated in Sect. 3.1. Let us now consider again Example 6.4. Two-loop massless propagator Feynman integrals of Fig. 3.9 with integer powers of the propagators given by the right-hand side of (3.39). The transition to vacuum integrals is similar to Example 6.2. Now we have h = 3. The basic polynomial can be obtained straightforwardly: P (x1 , . . . , x5 ) = −x1 x2 x3 + x22 x3 + x2 x23 + x21 x4 − x1 x2 x4 −x1 x3 x4 − x2 x3 x4 + x1 x24 + x1 x3 x5 − x2 x3 x5 −x1 x4 x5 + x2 x4 x5 + q 2 [−x1 x2 + x2 x3 + x1 x4 −x3 x4 + x1 x5 + x2 x5 + x3 x5 + x4 x5 − x25 ] + (q 2 )2 x5 . (6.28)

142

6 Reduction to Master Integrals by Baikov’s Method

After analysing various candidates with the indices 1 and 0 we conclude that the corresponding integrals (6.13) with reduced polynomials Pi can be interpreted non-trivially only in the following three cases two of which are symmetrical to each other: F6.4 (1, 1, 1, 1, 0) = I1 , F6.4 (0, 1, 1, 0, 1) = F6.4 (1, 0, 0, 1, 1) = I2 . Thus, we qualify them as master integrals. The values of these integrals can be obtained from (5.22) and (5.23), respectively: Γ (2 − d/2)2 Γ (d/2 − 1)4 , Γ (d − 2)2 Γ (3 − d)Γ (d/2 − 1)3 . I2 = −(iπ d/2 )2 (−q 2 )d−3 Γ (3d/2 − 3) I1 = (iπ d/2 )2 (−q 2 )d−4

(6.29) (6.30)

The corresponding coeﬃcient functions are constructed using the values of the following integrals that appear in (6.13). For c1 , we use q2 2 β g2 (α, β) = dx5 xα 5 (q − x5 ) 0

α+β+1 Γ (α + 1)Γ (β + 1) . = q2 Γ (α + β + 2) For c2 , we use

∞

∞

g3 (α1 , α4 , β) = 0

(6.31)

2 β 1 α4 dx1 dx4 xα 1 x4 (q + x1 + x4 )

0

α1 +α4 +β+2 Γ (α1 + 1)Γ (α4 + 1)Γ (−α1 − α4 − β − 2) . = q2 Γ (−β)

(6.32)

The decomposition of an arbitrary integral is F6.4 (a1 , a2 , a3 , a4 , a5 ) = c1 (a1 , a2 , a3 , a4 , a5 )I1 + [c2 (a1 , a2 , a3 , a4 , a5 ) + c2 (a2 , a1 , a4 , a3 , a5 )] I2 .

(6.33)

We again consider Example 6.5. Two-loop massless vertex Feynman integrals (5.34) of Fig. 5.3 with integer powers of the propagators. This is also a relatively simple example which can be treated almost like the previous examples. We shall deal with the following extended set of the denominators of the propagators: E1 = l2 − 2l·p1 + p21 ,

E2 = l2 − 2l·p2 + p22 ,

E3 = k 2 − 2k·p1 + p21 , E4 = k 2 − 2k·p2 + p22 , E5 = k 2 , E6 = k 2 − 2k·l + l2 , E7 = l2 , E8 =

p21

,

E9 = p1 ·p2 ,

E10 =

p22

.

(6.34) (6.35)

6.2 Constructing Coeﬃcient Functions. Simple Examples

143

The basic polynomial is straightforwardly evaluated, as a determinant of the corresponding 4 × 4-matrix (6.10). The eﬀective number of loops to be used in (6.9) is h = 4. Since we are not interested in higher terms of expansion in the external kinematical invariants2 p21 , p22 and p1 ·p2 , as usual, the parameters x8 , x9 and x10 are set to zero, and we obtain the following basic polynomial, according to the last rule in Sect. 6.1: P (x) = x22 x23 − 2x1 x2 x3 x4 + x21 x24 + 4Q2 x1 x2 x5 − 2Q2 x2 x3 x5 +2x1 x2 x3 x5 − 2x22 x3 x5 − 2Q2 x1 x4 x5 − 2x21 x4 x5 + 2x1 x2 x4 x5 +(Q2 )2 x25 + 2Q2 x1 x25 + x21 x25 + 2Q2 x2 x25 − 2x1 x2 x25 + x22 x25 +2Q2 x2 x3 x6 + 2Q2 x1 x4 x6 − 2(Q2 )2 x5 x6 − 2Q2 x1 x5 x6 − 2Q2 x2 x5 x6 +(Q2 )2 x26 − 2Q2 x2 x3 x7 − 2x2 x23 x7 − 2Q2 x1 x4 x7 + 4Q2 x3 x4 x7 +2x1 x3 x4 x7 + 2x2 x3 x4 x7 − 2x1 x24 x7 − 2(Q2 )2 x5 x7 − 2Q2 x1 x5 x7 −2Q2 x2 x5 x7 − 2Q2 x3 x5 x7 − 2x1 x3 x5 x7 + 2x2 x3 x5 x7 − 2Q2 x4 x5 x7 +2x1 x4 x5 x7 − 2x2 x4 x5 x7 − 2(Q2 )2 x6 x7 − 2Q2 x3 x6 x7 − 2Q2 x4 x6 x7 +4Q2 x5 x6 x7 + (Q2 )2 x27 + 2Q2 x3 x27 + x23 x27 + 2Q2 x4 x27 −2x3 x4 x27 + x24 x27 ,

(6.36)

where Q2 = −(p1 − p2 )2 as before. After a straightforward analysis of candidates we identify the following set of the master integrals: F (1, 1, 0, 0, 1, 1, 0) = I1 , F (1, 1, 1, 1, 0, 0, 0) = I2 and F (0, 1, 1, 0, 0, 1, 0) = F (1, 0, 0, 1, 0, 1, 0) = I3 . To construct the coeﬃcient function c1 we have to deal with integrals (6.13), where the reduced polynomial is

P1 (x3 , x4 , x7 ) = x7 ((Q2 )2 + (x3 − x4 )2 (6.37) +2Q2 (x3 + x4 ))x7 + 4Q2 x3 x4 . One can observe that in the cases, where n4 ≤ 0 (n3 ≤ 0) in the corresponding integral (6.13), one can straightforwardly integrate over x4 (x3 ) and then over x3 (x4 ) and x7 , using ∞ β g4 (α, β) = dx xα (x + a) 0 α+β+1 Γ (1

=a

+ α)Γ (−α − β − 1) . Γ (−β)

(6.38)

Suppose now that n3 , n4 > 0 in (6.13). Then we can use a trick based on the following integration formula obtained by IBP in a one-parametric integral: ∞ dx (Ax + B)z−n n+γ x 0 2

Observe that this is a formal expansion for p21 and p22 and a Taylor expansion for p1 ·p2 .

144

6 Reduction to Master Integrals by Baikov’s Method

=

(z − n )B z − n − n − γ + 1

0

∞

dx (Ax + B)z−n −1 . xn+γ

(6.39)

Applying it to the integration over x7 we can reduce either n3 or n4 to zero because here B = 4Q2 x3 x4 . The coeﬃcient function c2 can easily be constructed because the corresponding integral (6.13) over x7 can be evaluated by means of the following explicit formula x2 g5 (k, α1 , α2 ) = dx xk (x − x1 )α1 (x2 − x)α2 x1

=

k r=0

x1k−r (x2 − x1 )α1 +α2 +r+1

Γ (1 + α2 )Γ (1 + α1 + r) k! , (6.40) (k − r)!r! Γ (α1 + α2 + r + 2)

and then over x5 and x6 by means of (6.38). A similar procedure, without tricks, can be developed for the coeﬃcient function c3 . If I3 = F (0, 1, 1, 0, 0, 1, 0), this is achieved by integrating over x7 (which always can be done because n7 ≤ 0), and then over x5 , x1 and x4 . For the second copy of I3 , the coeﬃcient function is symmetrically obtained. Let us again turn to our favourite example which illustrates all the basic methods. Example 6.6. One-loop propagator Feynman integrals (1.2) corresponding to Fig. 1.1. The transition to the corresponding vacuum problem reduces to adding a new propagator, 1/(q 2 − s)a3 . We again consider these integrals at general q 2 and are not interested in derivatives so that, eﬀectively, the corresponding index will be a3 = 1 and the corresponding variable x3 is set to zero. The resulting basic polynomial is P (x1 , x2 ) = −(x1 − x2 + m2 )2 − q 2 (q 2 − 2m2 − 2(x1 + x2 )) .

(6.41)

Of course, at m = 0 it coincides with the polynomial (6.17) for Example 6.2. There are two master integrals F6.6 (1, 1) = I1 given by (1.5) and F6.6 (1, 0) = I2 given by the right-hand side of (5.6). We want to construct the corresponding coeﬃcient function with the normalization conditions (6.3), i.e. c1 (1, 1) = 1 , c1 (1, 0) = 0 , c2 (1, 1) = 0 ,

c2 (1, 0) = 1 .

The coeﬃcient function of I1 is simply obtained similar to the massless case 2 (d−3) q − m2 c1 (a1 , a2 ) = (a1 − 1)!(a2 − 1)! a1 −1 a2 −1 ∂ ∂ (d−3)/2 [P (x1 , x2 )] . (6.42) × ∂x1 ∂x2 xi =0

6.2 Constructing Coeﬃcient Functions. Simple Examples

145

For the coeﬃcient function c2 (a1 , a2 ) of I2 , we obtain linear combinations of one-parametric integrals dx (d−3)/2−n2 f (n1 , n2 ) = [P2 (x)] , (6.43) xn1 where P2 (x) = P (x1 , x)|x1 =0 = αx2 + βx + γ

(6.44)

with α = −1, β = 2(m2 + q 2 ), γ = −(m2 − q 2 )2 . Consider ﬁrst the case a2 ≤ 0. Then n1 is always non-positive here, and f (n1 , n2 ) can be understood as an integral between the roots 2 (1,2) = m ∓ q2 x2 of the quadratic polynomial P2 (x2 ), using (6.40). The evaluation at a1 = 1 and a2 = 0 provides a normalization factor to satisfy the normalization condition c2 (1, 0) = 1, and we obtain the following expression for c2 (a1 , a2 ) at a2 ≤ 0: c02 (a1 , a2 ) = c02 (a1 , a2 ) ≡ 1 × (a1 − 1)!

(2) x2 (1)

x2

Γ (d − 1) 4d−2 (m2 q 2 )(d−2)/2 Γ ((d

dx2 xn2

∂ ∂x1

a1 −1

− 1)/2)2

[P (x1 , x2 )]

(d−3)/2

.

(6.45)

x1 =0

In the case a2 > 0, the integrals f (n1 , n2 ) appear also with n1 > 0. When taken seriously they can be evaluated in terms of a Gauss hypergeometric function. Instead of doing this, let us apply IBP to our parametric integrals f (n1 , n2 ). This gives the relation f (n1 , n2 ) =

(d − 3)/2 − n2 n1 − 1 ×(2αf (n1 − 2, n2 + 1) + βf (n1 − 1, n2 + 1))

(6.46)

which can be used to reduce n1 to one or zero. Moreover, the identity (d−3)/2−n2

P2

(d−3)/2−n2 −1

= P2

P2

leads to the relation 1 f (1, n2 ) = (f (1, n2 − 1) − αf (−1, n2 ) − βf (0, n2 )) γ

(6.47)

which can be used to reduce n2 to zero. This means that we can express any f (n1 , n2 ) as a linear combination of an auxiliary master integral f (1, 0) and integrals f (n1 , n2 ) with n1 ≤ 0 which can be evaluated in terms of gamma functions. We believe that the coeﬃcient functions are rational functions of everything. The only chance to satisfy this property here is to construct c2 (a1 , a2 ) as a linear combination of c02 (a1 , a2 ) and the ﬁrst coeﬃcient function c1 (a1 , a2 ):

146

6 Reduction to Master Integrals by Baikov’s Method

c2 (a1 , a2 ) = c02 (a1 , a2 ) + Ac1 (a1 , a2 ) .

(6.48)

The constant A is determined by the normalization condition c2 (1, 1) = 0: A = −c02 (1, 1) .

(6.49)

After this, the dependence on f (1, 0) drops out and c2 (a1 , a2 ) indeed turns out to be a rational function. Observe that integrating over some real domain, in particular between the roots of a quadratic polynomial when constructing coeﬃcient functions, with a subsequent normalization, is in fact equivalent to solving IBP relations for our auxiliary parametric integrals. If there is such a possibility to understand a given parametric integral it is reasonable to use it. If there is no such possibility, e.g. one meets a polynomial of the third degree, or, an integration over one of the x-variables leads to inconvenient integrals over the rest variables, then there is no other way as to treat the auxiliary parametric integrals in a pure algebraic way by solving the corresponding IBP relations. We shall meet such situations in the examples below. As to the example above, the situation with a2 ≤ 0 could be treated algebraically, by IBP in the initial two-parametric integral, but integrating over x2 has simpliﬁed the situation.

6.3 General Recipes. Complicated Examples Let us extend what was done in the previous example to the general situation. After a preliminary analysis, with the help of (6.9), we obtain a preliminary list of candidates for the master integrals. Let us deﬁne the relation of partial ordering of the master integrals as follows: F (a1 ) < F (a2 ) if a1j ≤ a2j for all j , and the strict inequality holds at least for one index. The master integrals can be grouped into families characterized by their maximal integrals. Let us start from the master integrals which have most non-negative indices. Usually, the corresponding parametric integral for the coeﬃcient function can be understood in such a way that it results in integrations in terms of gamma functions. Consider now a situation with two master integrals with F (a2 ) < F (a1 ), and suppose that we already know c1 . If a2i = 1 we have also a1i = 1. To construct an algorithm for the coeﬃcient function c2 (a) we start with the case of negative indices aj for those indices j where a1j = 1 since in this case we have c1 (a) = 0. Experience shows that the integrations for c2 (a) result in ratios of gamma functions which in particular can be used to satisfy the normalization c2 (a2 ) = 1. In a next step one considers the case aj > 0. Then the corresponding parametric representation usually leads to integrals which cannot be evaluated in terms of gamma functions. (See the previous example.) Thus at ﬁrst

6.3 General Recipes. Complicated Examples

147

sight it looks hopeless to achieve that the coeﬃcient functions have to be rational functions of d. The way out is to look for an expression for the coeﬃcient function c2 (a) which is a linear combination of c1 (a) and the basic parametric representation for c2 (a) denoted by c02 (a) c2 (a) = c02 (a) + Ac1 (a) .

(6.50)

The constant A is determined by the normalization condition c2 (a1 ) = 0 which gives A = −c02 (a1 ) .

(6.51)

Then IBP is applied to the parametric integrals and the corresponding relations are used to express any given parametric integral in terms of auxiliary (parametric) master integrals and expressions which are straightforwardly evaluated in terms of gamma functions. The dependence on the new auxiliary master integrals has to drop out3 in order to provide a rational dependence of the coeﬃcient functions on d. In fact, this strategy can be generalized to the case of several master integrals with more complicated hierarchies. Let us proceed with examples, where we shall meet such situations. These will be mainly our old examples considered in Chaps. 3–5. Example 6.7. Feynman integrals (3.19) corresponding to the triangle diagram of Fig. 3.4. Almost all the steps can straightforwardly be performed, as above. The basic polynomial is P (x1 , x2 , x3 ) = (x1 − x3 )(x2 − x3 ) − Q2 x3 −m2 (Q2 + x1 + x2 − 2x3 ) + m4 ,

(6.52)

where again Q = −(p1 − p2 ) with = = 0. We obtain the following list of the master integrals: F (1, 1, 1) = I1 , F (1, 1, 0) = I2 and F (0, 0, 1) = I3 . When testing various candidates to be master integrals we consider, in particular, F (1, 0, 1) with the corresponding reduced polynomial P1,0,1 (x2 ) = m2 − Q2 − x2 linearly dependent on x2 . Let us try to understand the corresponding integrals (6.13) dx2 (6.53) (m2 − Q2 − x2 )z−nd n2 x2 2

2

p21

p22

in a non-trivial way. (Here we have z = (d − 4)/2 = −ε because the eﬀective number of loops is h = 3.) We do not consider the Cauchy integration around the origin in the complex plane because this choice corresponds to the value a2 = 1 in the master integral so that we are looking for other options. We 3

This cancellation serves as a good check of the algorithm, similarly to cancellations of spurious poles in ε on the right-hand side of various asymptotic expansions in momenta and/or masses [6].

148

6 Reduction to Master Integrals by Baikov’s Method

cannot integrate from x2 = 0 because we have integer negative powers of x2 . Still it looks like there is a chance to obtain a new non-trivial understanding of the integral by choosing to integrate from −∞ to m2 −Q2 Here we suppose that m2 − Q2 < 0 in order to have no singularity in the integration domain. However, this choice brings nothing new! One can check that, after the normalization by the equation c1,0,1 (1, 0, 1) = 1, one obtains the same expression as in the case of the Cauchy integration corresponding to other values of the index a2 . Therefore, we conclude that we cannot interpret (6.53) in a new non-trivial way so that the integral F (1, 0, 1) is not a master integral. A more general recipe is that, whenever we obtain in a linear dependence of a reduced polynomial in (6.13) on some variable, we shall usually4 conclude that this cannot be a master integral. The coeﬃcient function of I1 can be constructed trivially because it does not involve integration. The coeﬃcient function of I2 , with the corresponding polynomial P2 = (m2 + x3 )(m2 − Q2 + x3 ), is also simple (at least simpler than in Example 6.6). If n3 ≤ 0 in the corresponding integral (6.13), we can integrate between the roots of this polynomial using (6.40). In the case of n3 > 0, one can use the IBP relation with respect to x3 in order to reduce n3 to one and the relation following from the identity P2z−nd = P2z−nd −1 P2 to adjust the dimension. For the coeﬃcient function of I3 , we obtain integrals (6.13) with P3 (x1 , x2 ) = x1 x2 − m2 (Q2 + x1 + x2 ) + m4 . If one of the indices n1 and n2 in this integral is negative the integration over the corresponding variable, e.g. over x2 , can be performed but one obtains a power of (m2 − x1 ) not regularized by z. So, in this situation, it is necessary to proceed in a pure algebraic way and solve the corresponding IBP relations, together with the relation that follows from the identity P3z−nd = P3z−nd −1 P3 , in order to reduce any given integral to auxiliary master integrals. There is, however, one more option5 : to use the package AIR [1] based on the algorithm of [11] and designed to solve genuine IBP relations for Feynman integrals as discussed in the end of the previous chapter. It turns out that this program can be applied to the auxiliary IBP relations for integrals (6.13). As a result of this procedure, an algorithm for c3 can be constructed. In particular, we obtain 1 1 (d − 4)(2m2 − Q2 )I1 F (1, 1, 2) = 2 2 m (m − Q2 ) 2 4

Well, up to some pathological situations, where one has chances to obtain a new meaning for such integrals by considering the integration over xi in the sense of a distribution with respect to the variables on which coeﬃcients of the corresponding linear polynomial depend. 5 Thanks to J. Piclum who implemented the corresponding algorithm on a computer, also for the Example 6.10 below.

6.3 General Recipes. Complicated Examples

+(d − 3)I2 +

2−d I3 2m2

149

,

(6.54)

in agreement with (5.17), where several integrals expressed in terms of gamma functions were involved on the right-hand side. Let us again consider massless on-shell boxes which we have already analysed in Examples 3.3, 4.3 and 5.4. Example 6.8. The massless on-shell box Feynman integrals of Fig. 5.1 with p2i = 0, i = 1, 2, 3, 4 and general integer powers of the propagators. The basic polynomial is now P (x1 , x2 , x3 , x4 ) = s2 t2 + t2 (x1 − x2 )2 − 2st2 (x1 + x2 ) +s2 (x3 − x4 )2 − 2s2 t(x3 + x4 ) −2st[2x1 x2 + 2x3 x4 − (x1 + x2 )(x3 + x4 )] .

(6.55)

The eﬀective number of loops to be used in (6.9) is now h = 4. Using the strategy formulated above, we reveal the following three master integrals: F (1, 1, 1, 1) = I1 and F (1, 1, 0, 0) = F (0, 0, 1, 1) = I2 . The coeﬃcient function of I1 can be constructed trivially. In the case of I2 (the ﬁrst of the two symmetric variants), the integration in the corresponding integral (6.13) over x4 and then over x3 can be performed in terms of gamma functions if n4 ≤ 0, and, in the opposite order, in the case of n3 ≤ 0. One can then proceed similarly to Example 6.6 by introducing an auxiliary parametric integral and using IBP relations to reduce n3 or n4 to one or zero. Then, to deﬁne the coeﬃcient function c2 , one involves a linear combination with the coeﬃcient function c1 so that the dependence on this auxiliary integral drops out. Now we turn to a massive generalization of this example. Example 6.9. The on-shell boxes with two massive and two massless lines shown in Fig. 6.2, with p21 = . . . = p24 = m2 . p1 3

p2

p3

1 4 2

p4

Fig. 6.2. On-shell box with two massive and two massless lines. The solid lines denote massive, the dotted lines massless particles

As in Example 6.8, we have changed the numbering of the lines with respect to Chap. 4. The procedure is again straightforward. One can identify the master integral with four lines, F (1, 1, 1, 1) = I1 , two symmetrical master integrals with

150

6 Reduction to Master Integrals by Baikov’s Method

three lines, F (1, 0, 1, 1) = I21 , F (0, 1, 1, 1) = I22 , two master integrals with two lines, F (1, 1, 0, 0) = I31 , F (0, 0, 1, 1) = I32 and two symmetrical master integrals with one line, F (1, 0, 0, 0) = I41 , F (0, 1, 0, 0) = I42 . These master integrals are graphically shown in Fig. 6.3. We have the following hierarchy relations: I41 , I42 < I31 < I1 and I32 < I21 , I22 < I1 .

I1

I2

I31

I32

I4

Fig. 6.3. Master integrals for Fig. 6.2

The coeﬃcient function c1 is trivial. The coeﬃcient function c21 can be constructed, using (6.13), ﬁrst in the case of n2 ≤ 0, where it can be obtained by an explicit integration. Then, for n2 > 0, one applies IBP to these auxiliary integrals, introduces an auxiliary master integral and mixes such a solution with c1 . To construct the coeﬃcient function of I31 , one uses a straightforward integration in the case n1 ≤ 0 and general n2 and, similarly, for n2 ≤ 0 and general n1 . In the case of n1,2 > 0, one can apply auxiliary IBP relations with the introduction of an auxiliary master integral for n1 = n2 = 1 which is cancelled when mixing the so constructed coeﬃcient function with c1 . In the cases of the master integrals I32 , I41 and I42 , we have a tower of three hierarchical master integrals. Still the case of I32 is quite similar to I31 and does not provide complications. To construct the coeﬃcient function of I42 one uses a straightforward integration over x1 , x3 , x4 in the case of n1 ≤ 0, n3 ≤ 0, and over x1 , x4 , x3 in the case of n1 ≤ 0, n4 ≤ 0. In the case of n1 ≤ 0, n3,4 > 0, one integrates over x1 and uses, for resulting integrals over x3 and x4 , auxiliary recurrence relations, with an introduction of a master integral for n3 = n4 = 1 which cancels when mixing with the coeﬃcient function c22 . Quite similarly, one can explicitly integrate over x3 or x4 when n3 ≤ 0 or/and n4 ≤ 0 and reduce resulting integrals. Finally, in the case of n1,3,4 > 0, one solves corresponding auxiliary IBP relations and introduces a master integral for n1 = n3 = n4 = 1 which cancels when mixing with the coeﬃcient function c1 . Here is an example of the reduction of massive boxes to the master integrals: F (2, 1, 1, 1) =

(d − 4)(4m2 − t) d−3 d−5 I I2 − 2 I32 + 1 4m2 − s 2m2 (4m2 − s)t m (4m2 − s)t (d − 4)(d − 2) I4 . − (6.56) 2(d − 5)m4 (4m2 − s)t

6.3 General Recipes. Complicated Examples

151

We shall consider another example with a tower of three hierarchical master integrals in the next section. The last example in this section is Example 6.10. Sunset diagrams of Fig. 3.12 with one zero mass and two equal non-zero masses at a general value of the external momentum squared. We are dealing with the following family of integrals: dd kdd l (2q·k)−a3 (2q·l)−a4 F6.10 (a) = , (k 2 − m2 )a1 (l2 − m2 )a2 [(q − k − l)2 ]a5

(6.57)

where a = (a1 , a2 , a3 , a4 , a5 ) with a3,4 ≤ 0. The strategy presented above reveals the following preliminary list of the master integrals: F (1, 1, 0, 0, 1) = I1 and F (1, 1, 0, 0, 0) = I2 . The coeﬃcient function c2 can be constructed using the strategy described above: for n5 ≤ 0, an integration in terms of gamma functions is used and, for n5 > 0, a simple recursion is applied. It turns out that one can use the package AIR [1] to solve the recurrence relations for the auxiliary parametric integrals (6.13) corresponding to c1 , z−nd dx3 dx4 [P1 (x3 , x4 )] , (6.58) f (n3 , n4 , nd ) = xn3 3 xn4 4 where z = (d − 4)/2 = −ε and P1 (x3 , x4 ) = m2 (x3 + x4 − 2q 2 )2 −(x3 − q 2 )(x4 − q 2 )(x3 + x4 − q 2 ) .

(6.59)

Remember that we have n3 , n4 ≤ 0 so that we can perform a useful change of variables, x3,4 = x3,4 + q 2 and deal with integrals in these variables where the basic polynomial looks simpler. When solving the corresponding IBP relations (together with the relation following from the identity P1z−nd = P1z−nd −1 P1 ) z−nd which is it is useful to apply Euler’s theorem to the factor [P1 (x3 , x4 )] 2 2 a homogeneous functions of the four variables, x3 , x4 , q , m (although it is clear that the resulting relation is nothing but a special combination of the IBP relations). A general solution to these relations is determined by the two auxiliary master integrals, f (0, 0, 0) and f (−1, 0, 0). Therefore, it is necessary to introduce an extra master integral, I¯1 = F (1, 1, −1, 0, 1). As a result, an algorithm for the evaluation of all the three coeﬃcient functions, c1 , c¯1 and c2 , can be constructed. The dependence on the auxiliary master integrals drops out in expressions for the coeﬃcient functions. We have, in particular,

1 (d − 3)m2 − (d − 2)q 2 I1 F (2, 1, 0, 0, 1) = 2 m (4m2 − q 2 ) 3 1 ¯ + (d − 2)I1 + (d − 2)I2 , (6.60) 2 2

152

6 Reduction to Master Integrals by Baikov’s Method

F (2, 1, −1, 0, 1) =

F (2, 1, 0, −1, 1) =

2 − 2(d − 3)m2 + (d − 1)q 2 I1 2 −q +3(d − 2)I¯1 + (d − 2)I2 . 4m2

4(d − 3)m4 − (d − 2)(q 2 )2 I1 − 3 2¯ 2 2 + (d − 2)q I1 − (d − 2)(2m − q )I2 . 2

(6.61)

1

m2 (4m2

q2 )

(6.62)

Let us consider, following [14], a more complicated example in a separate section.

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential Example 6.11. Two-loop Feynman integrals for the heavy quark potential corresponding to Fig. 6.4. 6

7

6

1

2

2

1

7

5 5 3

4 (A)

3

4 (B)

Fig. 6.4. Feynman diagrams corresponding to case A and case B. Wavy lines denote propagators for the static source

The numbering of the lines in Fig. 6.4 is changed as compared with Fig. 3.9 in order to take into account the symmetry. There are two classes of such Feynman integrals which we denote A and B: dd kdd l FA (a) = (k 2 )a1 (l2 )a2 [(k − q)2 ]a3 [(l − q)2 ]a4 [(k − l)2 ]a5 1 × , (6.63) (v·k)a6 (v·l)a7 dd kdd l FB (a) = (k 2 )a1 (l2 )a2 [(k − q)2 ]a3 [(l − q)2 ]a4 [(k − l)2 ]a5 1 × , (6.64) (v·k)a6 [v·(k − l)]a7 where v·q = 0.

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

153

The Feynman integrals necessary for the evaluation of the two-loop potential were calculated in [12]. In [13], a procedure for the evaluation of arbitrary integrals (6.63) and (6.64) was developed, using the technique of shifting dimension [15] discussed in Chap. 5. However, not all the necessary relations were published. Another version of partial calculation of integrals (6.63) and (6.64) was used in [9] for the evaluation of 1/m corrections to the two-loop potential. In this algorithm, IBP was used without systematization, as in Chap. 5, and the reduction always stopped at integrals expressed in terms of gamma functions so that a lot of boundary integrals, sometimes involving up to fourfold ﬁnite summations, entered the reduction. Now, we are going to apply the method of this chapter to these integrals. We will, therefore, obtain a minimal set of master integrals. The basic polynomials are straightforwardly obtained: PA (x1 , . . . , x7 ) = −[x2 x6 − x4 x6 + (−x1 + x3 )x7 ]2 +v 2 {x21 x4 + x3 (x22 + x2 (x3 − x4 − x5 ) + x4 x5 ) −x1 [x2 (x3 + x4 − x5 ) + x4 (x3 − x4 + x5 )]} +(q 2 )2 [v 2 x5 − (x6 − x7 )2 ] + q 2 {v 2 [(x3 + x4 − x5 )x5 +x2 (x3 − x4 + x5 ) + x1 (−x3 + x4 + x5 )] + 2[x2 x6 (−x6 + x7 ) + x4 x6 (−x6 + x7 ) +x7 (x1 x6 + x3 x6 − 2x5 x6 − x1 x7 − x3 x7 )]} , PB (x1 , . . . , x7 ) = PA (x1 , x2 , x3 , x4 , x5 , x6 , x6 − x7 ) .

(6.65) (6.66)

The two cases A and B are considered separately. Case A. The application of the procedures described above to case A leads to the following families of master integrals which are shown in Fig. 6.5. As far as the notation is concerned the ﬁrst index labels the diﬀerent master integrals. In case the master integrals are equal we introduce a second index for further speciﬁcation. If Ij is a master integral with indices 1 and 0 then we shall denote by I¯j the master integral which diﬀers from Ij by one index −1 instead of 0. – Family A1 consists of the four master integrals with the hierarchy I1 > {I21 , I22 } > I3 : I1 = FA (1, 1, 1, 1, 0, 1, 1) , I21 = FA (1, 1, 1, 1, 0, 0, 1) , I22 = FA (1, 1, 1, 1, 0, 1, 0) , I3 = FA (1, 1, 1, 1, 0, 0, 0) . – Family A2 consists of the four master integrals with the hierarchy I51 > {I71 , I81 } > I41 :

154

6 Reduction to Master Integrals by Baikov’s Method

Family A1: I1

I21

I22

I3

I51

I71

I81

I41

Family A2:

Family A4: I61

Fig. 6.5. Feynman diagrams corresponding to the master integrals of case A. In addition to I61 , there is also a master integral I¯61 containing an irreducible numerator

I51 = FA (1, 0, 0, 1, 1, 1, 1) , I71 = FA (1, 0, 0, 1, 1, 0, 1) , I81 = FA (1, 0, 0, 1, 1, 1, 0) , I41 = FA (1, 0, 0, 1, 1, 0, 0) . – Family A3 is symmetrical to Family A2 with respect to the transformation 1 ↔ 2, 3 ↔ 4, 6 ↔ 7. It contains the master integrals I52 , I72 , I82 and I42 . – Family A4 contains the master integrals I61 = FA (0, 1, 0, 1, 1, 1, 0) , I¯61 = FA (0, 1, 0, 1, 1, 1, −1) . – Family A5 is symmetrical to Family A4 with respect to the transformation 1 ↔ 2, 3 ↔ 4, 6 ↔ 7. It contains the master integrals I62 and I¯62 . As has already become clear from the examples discussed so far, one expects the appearance of complicated expressions for the coeﬃcient functions of simplest master integrals. Indeed, in the case of the coeﬃcient function c1 , six out of seven indices can be treated with the help of diﬀerentiations and the remaining one-dimensional integral can be understood in the sense of integration (6.31). The situation is similar for c22 (and c21 which can be obtained by exploiting the symmetry) where the remaining two-fold integration over x7 and x5 can be understood with the help of the integrals (6.40) and (6.26). To construct c3 we have to understand, in some way, three integrations, over x5 , x6 , x7 . In case one of the indices n5 , n6 or n7 is less or equal to zero one can use various combinations of the auxiliary integrals gi (i = 1, . . . , 4) listed above. Thereby it is advantageous to perform the integration corresponding to the negative index ﬁrst. If, on the contrary, n5 , n6 and n7 are

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

155

positive an immediate integration seems not to be possible. However, from the corresponding three-parametric integral representation it is simple to derive recurrence relations which shift at least one of the indices to zero, eventually at the cost of increasing the dimension nd . The latter does not constitute a problem since the whole formulation of our procedure is in d dimensions. Thus, also in this case the integration can be performed in terms of gamma functions. In principle one could be forced to introduce three auxiliary master integrals and build the proper linear combinations with c1 , c21 and c22 . However, it turns out that the corresponding constants in such combinations are zero. For the coeﬃcient function c51 , only two non-trivial integrations over x2 and x3 are involved which can be performed with the help of (6.32). For c81 , one can use the symmetry: c81 (a1 , a2 , a3 , a4 , a5 , a6 , a7 ) = c71 (a4 , a3 , a2 , a1 , a5 , a7 , a6 ) . The most complicated coeﬃcient function is certainly c41 since there are four non-trivial integrations over x2 , x3 , x6 and x7 left. If n6 or n7 are less than or equal to zero the integrations can be performed in terms of gamma functions with the help of the formulae provided above. However, for n6 ≥ 1 and n7 ≥ 1 this is not possible. In this case, the idea is to use IBP in order to reduce the four-parametric auxiliary integrals dx2 dx3 dx6 dx7 A,aux (n2 , n3 , n6 , n7 , nd ) = . . . I41 xn2 2 xn3 3 xn6 6 xn7 7 × [P41 (x2 , x3 , x6 , x7 )]

z−nd

(6.67)

(with z = (d − h − 1)/2 = (d − 5)/2) to the auxiliary master integral A,aux I41 (1, 1, 1, 1, 0). Here P41 is obtained from PA by setting x1 , x4 and x5 to zero. Observe that the corresponding recurrence procedure is signiﬁcantly simpler than the original one which involves seven denominators. Furthermore, if during the recursion either n6 or n7 becomes negative the corresponding expressions can immediately be expressed in terms of gamma functions. The A,aux (1, 1, 1, 1, 0) ﬁve IBP relations which are useful for the reduction to I41 can be obtained by either diﬀerentiating the integrand with respect to xi z−nd z−nd −1 (i = 2, 3, 6, 7) or by writing down the identity P41 = P41 P41 and inserting the explicit result for the last factor. The proper combination of these relations leads to new ones which allow the following steps to be performed in an automatic way: 1. 2. 3. 4. 5.

Reduce n6 and n7 to one. Reduce n2 , n3 > 0 to n2 , n3 ≤ 0. Use IBP recurrence relations to obtain n2 = n3 . Reduce n2 = n3 < 0 to n2 = n3 = 0. Adjust the dimension, i.e. reduce nd to zero.

A,aux A,aux A simple relation transforms I41 (0, 0, 1, 1, 0) to I41 (1, 1, 1, 1, 0).

156

6 Reduction to Master Integrals by Baikov’s Method

At this point one constructs the ﬁnal coeﬃcient function c41 by considering the linear combination with c51 , c71 and c81 . Since c41 (a71 ) = c41 (a81 ) = 0, we are left with c41 (a) = c041 (a) − c041 (a51 )c51 (a) ,

(6.68)

where 1 4(d − 3)(3d − 14)(3d − 10)(3d − 8) q2 v2 (d − 4)2 (3d − 13)(3d − 11) (d − 5)2 A,aux (q 2 )2 I41 (1, 1, 1, 1, 0) . + (3d − 13)(3d − 11)

c041 (a51 ) = −

A,aux In this combination the auxiliary master integral I41 (1, 1, 1, 1, 0) cancels and c41 (n) turns out to be a rational function in d. The master integral I61 forms a family by its own. However, as the polynomial P61 is quadratic in x7 and thus the corresponding recurrence relation shifts n7 only in steps of two, it is necessary to introduce in addition the master integral I¯61 where a7 = −1. The very calculation of the coeﬃcient function is identical for I61 and I¯61 . For n3 ≤ 0, it can be done in terms of gamma functions with the integration order x3 , x1 , x7 . On the other hand, for n3 > 0, a simple one-step relation reduces n3 to zero. Let us now turn to Case B. As one can see from (6.66) the basic polynomial is quite similar to the one of case A which can be used while computing the coeﬃcient functions. However, the symmetry can only be exploited if n7 ≤ 0 as for n7 > 0 the factor (x6 − x7 ) would appear in the denominator. Altogether there are four families which, however, show a more complicated structure than in case A – see Fig. 6.6. More precisely one has

– Family B1. There are twelve master integrals which obey the hierarchies I1B > {I2B , I22 } > I3 and I1B > I2B > {I6i , I¯6i } (i = 3, 4, 5, 6) and are given by I1B = FB (1, 1, 1, 1, 0, 1, 1) , I2B = FB (1, 1, 1, 1, 0, 0, 1) , I22 = FB (1, 1, 1, 1, 0, 1, 0) , I3 = FB (1, 1, 1, 1, 0, 0, 0) , I63 = FB (1, 1, 1, 0, 0, 0, 1) , I64 = FB (1, 1, 0, 1, 0, 0, 1) , I65 = FB (1, 0, 1, 1, 0, 0, 1) , I66 = FB (0, 1, 1, 1, 0, 0, 1) . There are four master integrals with a6 = −1:

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

157

Family B1: I1B

I2B

I22

I3

I63

I64

I65

I66

I9

I82

I81

I41

I53

I83

I72

I42

Family B2:

Family B3:

Family B4: I67

Fig. 6.6. Feynman diagrams corresponding to the master integrals of case B. In addition to I6i (i = 3, . . . , 7) there are also master integrals I¯6i containing irreducible numerators

I¯63 = FB (1, 1, 1, 0, 0, −1, 1) , I¯64 = FB (1, 1, 0, 1, 0, −1, 1) , I¯65 = FB (1, 0, 1, 1, 0, −1, 1) , I¯66 = FB (0, 1, 1, 1, 0, −1, 1) . – Family B2. There are four master integrals which obey the following hierarchy: I9 > {I82 , I81 } > I41 with I9 = FB (1, 0, 0, 1, 1, 1, 1) , I82 = FB (1, 0, 0, 1, 1, 0, 1) , I81 = FB (1, 0, 0, 1, 1, 1, 0) , I41 = FB (1, 0, 0, 1, 1, 0, 0) . – Family B3. Similarly to Family B2, there are four master integrals obeying the hierarchy I53 > {I83 , I72 } > I42 with I53 = FB (0, 1, 1, 0, 1, 1, 1) , I83 = FB (0, 1, 1, 0, 1, 0, 1) ,

158

6 Reduction to Master Integrals by Baikov’s Method

I72 = FB (0, 1, 1, 0, 1, 1, 0) , I42 = FB (0, 1, 1, 0, 1, 0, 0) . – Family B4 consists of the two master integrals I67 = FB (0, 1, 0, 1, 1, 1, 0) , I¯67 = FB (0, 1, 0, 1, 1, 1, −1) . It is similar to the Families A4 and A5 of case A. B The construction of the coeﬃcient functions cB 1 , c2 and c22 of the family B1 proceeds along the same lines as in case A. In the case of c3 , we have to deal with integrals I3B,aux (n5 , n6 , n7 , nd ) which are deﬁned similarly to (6.67). There is a slight complication as, in contrast to case A, c3 (a1 ) = 0. As a consequence an auxiliary master integral, I3B,aux (0, 1, 1, 0), has to be introduced which is only cancelled after considering the proper linear combination with c1 . The reduction to I3B,aux (0, 1, 1, 0) is straightforward. Family B1 has four more members, I63 , I64 , I65 and I66 , which belong to the four hierarchies I1B > I2B > I6i (i = 3, 4, 5, 6). Thus, in order to obtain the coeﬃcient functions c6i one has to consider the linear combination B 0 B B c6i = c06i − c06i (aB 1 )c1 (a) − c6i (a2 )c2 (a) .

(6.69)

Let us in the following restrict the discussion to c63 since the results for the other three coeﬃcients can be obtained by exploiting the symmetry. The corresponding auxiliary integrals are given by an integral representation of the form z−nd dx4 dx5 dx6 , (6.70) c063 ∼ . . . [P63 (x4 , x5 , x6 )] xn4 4 xn5 5 xn6 6 with

P63 = (q 2 )2 v 2 x5 + q 2 v 2 x4 x5 − x25 − 4q 2 x5 x26 − x24 x26 .

(6.71)

B For n4 ≤ 0, where we have cB 1 (n) = c2 (n) = 0, the integrals in (6.70) can be taken analytically in the order x4 , x5 , x6 using (6.40) for x4 , the formula (6.40) for x5 and (6.26) extended to non-integer k3 for x6 . Let n4 > 0. Then we need to introduce two auxiliary master integrals, B,aux B,aux (1, 0, 0, 0) and I63 (1, 0, 1, 0). The reduction of the auxiliary paraI63 metric integrals (6.70) can be performed as follows:

1. Reduce n4 to one. 2. Reduce n5 to zero. 3. The reduction of n6 can only be performed in steps of two. Thus one ends up with n6 = 0 or n6 = −1. 4. Adjust the dimension, i.e. reduce nd to zero. The corresponding recurrence relations are derived easily from (6.71). It B,aux is interesting to note that in (6.69) the master integral I63 (1, 0, 1, 0) is B,aux B B cancelled from c1 and I63 (1, 0, 0, 0) from c2 . Observe that, due to the

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

159

structure of the reduced polynomial (6.71), in addition to I63 also a master integral with n6 = −1, I¯63 , has to be introduced which, however, has the same coeﬃcient function as I63 . Observe also that, for c63 and c65 , the master integrals I6 and I¯6 are needed, while for c64 and c66 , the integrals I6 and I¯6B are necessary. Families B2, B3 and B4 are similar to the families A2, A3 and A4, respectively, so that the corresponding coeﬃcient functions are similarly constructed. The procedure described above was implemented in a MATHEMATICA package [14]. Let us now list all occurring master integrals in both cases A and B. They have been obtained with the help of the program package developed for the calculation performed in [9] where IBP recurrence relations have been ‘nonsystematically’ solved. d/2 2 iπ π Γ (5/2 − d/2)2 Γ (d/2 − 3/2)4 , I1 = Q2+4ε v 2 Γ (d − 3)2 d/2 2 √ iπ π Γ (2 − d/2)Γ (5/2 − d/2)Γ (d/2 − 1)2 Γ (d/2 − 3/2)2 I2 = − , Q1+4ε v Γ (d − 3)Γ (d − 2) 2 Γ (2 − d/2)2 Γ (d/2 − 1)4 I3 = iπ d/2 , Q4ε Γ (d − 2)2 2 Γ (3 − d)Γ (d/2 − 1)3 , I4 = − iπ d/2 Q2−4ε Γ (3d/2 − 3) 2 π 2 e−2γE ε 7 2 2 d/2 2 I5 = iπ − − 4 + −24 + π ε + O(ε ) , Q4ε v 2 3ε 9 2 √πQ1−4ε I6 = iπ d/2 v 2d−2 Γ (3 − d)Γ (7/2 − d)Γ (d/2 − 1)Γ (d − 5/2)2 , × Γ (2 − d/2)Γ (2d − 5) 2 √ 2d−2 Γ (3 − d)2 Γ (d/2 − 1)Γ (d − 2)2 I¯6 = − iπ d/2 , πQ2−4ε Γ (3/2 − d/2)Γ (2d − 4) √ 2 πQ1−4ε I7 = iπ d/2 v Γ (7/2 − d)Γ (d/2 − 1)2 Γ (d/2 − 3/2)Γ (d − 5/2) , × Γ (d − 2)Γ (3d/2 − 4) I8 = I7 , I9 = I5 , 1 I1B = I1 , 2 2 π 2 e−2γE ε I2B = iπ d/2 Q1+4ε v

160

6 Reduction to Master Integrals by Baikov’s Method

5 2 π − 16 ln 2 − 4 ln2 2 + O(ε2 ) , × −4 ln 2 + ε 3 B ¯ ¯ I6 = −I6 , where Q = −q 2 . The fact that I5 = I9 and I7 = I8 can be seen immediately by a simple change of the loop momenta. Since I7 = I8 , we have in both cases one master integral less. So, in case A, we have eight master integrals, I1 , . . . , I7 and I¯6 , and, in case B, ten master integrals I2 , . . . , I7 , I9 , I¯6B , I1B and I2B , Only two of the master integrals are not known in terms of gamma functions. Their results are given in expansion in ε up to ε1 . For example, they can be evaluated by the method of MB representation described in Chap. 4. Here are some examples of results for the coeﬃcient functions: FA (2, 2, 1, 1, 1, 1, 1) = c1 I1 + c3 I3 + (c41 + c42 )I4 + (c51 + c52 )I5 +(c61 + c62 )I¯6 d/2 2 iπ 2 4 2 16 368 2 + π − + π − 8ζ(3) + O(ε) , = 8+4ε 2 Q v 3ε 3ε 9 45 with 2(d − 5)(d − 4) 8(d − 5)(d − 3)2 , c = , 3 q6 (d − 4)q 8 v 2 −3(d − 3)(3d − 16)(3d − 14)(3d − 10)(3d − 8) = c42 = (d − 9)(d − 8)(d − 7)(d − 6)2 (d − 4)2 q 10 v 2

c1 = c41

c51 = c52 c61 = c62

×(5d3 − 93d2 + 588d − 1264) , −3(3d − 17)(3d − 13)(3d − 11) = , (d − 9)(d − 7)q 8 −32(2d − 13)(2d − 11)(2d − 9)(2d − 7)(2d − 5) = . (d − 9)(d − 7)(d − 6)(d − 4)q 10 v 2

FB (2, 2, 1, 1, 1, 1, 1) = c1 I1B + c3 I3 + (c41 + c42 )I4 + c53 I5 +(c63 + c65 )I¯6 + (c64 + c66 + c67 )I¯6B + c9 I9 d/2 2 iπ 4 2 8 368 2 1 π + 4ζ(3) + O(ε) , = 8+4ε 2 − + π + + Q v 3ε 3ε 9 45 with 2(d − 5)(d − 4) −4(d − 5)(d − 3)2 , c3 = , 6 q (d − 4)q 8 v 2 3(d − 3)(3d − 16)(3d − 14)(3d − 10)(3d − 8) = (d − 9)(d − 8)(d − 7)(d − 6)2 (d − 4)2 q 10 v 2

cB 1 = c41

c42

×(7d3 − 117d2 + 654d − 1232) , −6(d − 3)(3d − 16)(3d − 14)(3d − 10)(3d − 8) = (d − 9)(d − 8)(d − 7)(d − 6)2 (d − 4)2 q 10 v 2

6.4 Two-Loop Feynman Integrals for the Heavy Quark Potential

c53 c63

×(d3 − 12d2 + 33d + 16) , −3(3d − 17)(3d − 13)(3d − 11) = , (d − 9)(d − 7)q 8 4(2d − 7)(2d − 5) = c64 = − (d − 9)(d − 7)(d − 6)(d − 4)q 10 v 2

×(15d4 − 304d3 + 2240d2 − 7093d + 8118) , 4(2d − 7)(2d − 5)(d2 − 17d + 55) , c65 = c66 = (d − 7)(d − 4)q 10 v 2 −32(2d − 13)(2d − 11)(2d − 9)(2d − 7)(2d − 5) c67 = , (d − 9)(d − 7)(d − 6)(d − 4)q 10 v 2 −3(3d − 17)(3d − 13)(3d − 11) c9 = . (d − 9)(d − 7)q 8 FA (1, 1, 2, 1, 1, −1, 1) = c3 I3 + (c41 + c42 )I4 + c62 I¯6 d/2 2 iπ 3 1 = − + − 2ζ(3) + O(ε) , Q4+4ε 2ε 2 2(d − 3) −3(3d − 10)(3d − 8)(d2 − 5d + 2) , c41 = , 4 (d − 4)q 2(d − 6)(d − 5)(d − 4)2 q 6 3(d − 5)(d − 2)(3d − 10)(3d − 8) = , 2(d − 6)(d − 4)2 q 6 4(2d − 9)(2d − 7)(2d − 5) = . (d − 5)(d − 4)q 6

c3 = c42 c62

FB (1, 1, 2, 1, 1, −1, 1) = c3 I3 + (c41 + c42 )I4 + (c63 + c65 )I¯6 +(c64 + c66 )I¯6B d/2 2 iπ 1 1 = − + + O(ε) , Q4+4ε 2ε 2 (d − 5)(d − 3) −3(3d − 10)(3d − 8)(d2 − 9d + 22) , c = , 41 (d − 6)q 4 2(d − 6)2 (d − 5)(d − 4)q 6 3(3d − 10)(3d − 8)(d2 − 11d + 26) = , 2(d − 6)2 (d − 4)q 6 (2d − 11)(2d − 7)(2d − 5) = , (d − 6)(d − 5)q 6 −(2d − 7)(2d − 5) (2d − 7)2 (2d − 5) = , c = , 65 (d − 6)(d − 5)q 6 (d − 6)(d − 5)q 6 −(2d − 7)(2d − 5)(4d − 19) = . (d − 6)(d − 5)q 6

c3 = c42 c63 c64 c66

161

162

6 Reduction to Master Integrals by Baikov’s Method

6.5 Conclusion Let us observe that since a given problem of solving IBP relations is always reduced, in the present method, to the corresponding problem for vacuum Feynman integrals, it turns out that diﬀerent initial problems can have the same vacuum ‘image’. As it was demonstrated in [4], this property can be used when a solution of some reduction problem is known and another reduction problem has the same vacuum image with it. For example, solving IBP relations for the two-loop massless vertex diagrams (of Fig. 5.3, Fig. 3.13 and a Mercedez-Benz type) can be reduced to solving IBP relations for the three-loop propagator diagrams that was done in [7] and implemented in [8]. The method of this chapter has a feature opposite to the method of shifting dimension [15] discussed in Chap. 5. Indeed, the ﬁrst point in the latter is to get rid of numerators, with the primary idea to simplify the situation. In contrast to this, the numerators play a crucial role in the present method: each irreducible numerator results in an integration over the corresponding xvariable in the basic parametric representation. One more diﬀerence of these two methods is that master integrals with indices ai > 1 usually appear in a reduction with shifting dimension, while there are no such master integrals in the present method. (The same feature holds for the modern realization of the method of diﬀerential equations to be discussed in the next chapter.) On the other hand, shifting dimension is also an intrinsic feature of the present method because the dimension d enters the basic representation in a very simple way and it is necessary to put the shift of dimension under control when solving the auxiliary IBP relations. The method of this chapter was successfully applied, due to the reduction presented in Sect. 6.3, in [10], where various two-loop diagrams associated with the two-loop quark potential were necessary. A breakthrough in another direction – the evaluation of general four-loop propagator diagrams (i.e. one loop above [7]!) was also achieved with its help [3]. Another branch of this method was, however, applied there. It is based on an expansion at large d which is somehow introduced when constructing the coeﬃcient function of the master integrals starting from (6.9). Unfortunately, no details of this branch have been published up to now. This method is now at the level of experimental mathematics, as well as many other techniques discussed in this book. One tries to follow the prescriptions formulated in this chapter and, hopefully, arrives at a solution of a given reduction problem. One always believes in the rational dependence of the coeﬃcient functions on everything, and this is one of possible consistency checks. The validity of the reduction so obtained can be checked by explicit evaluation of various Feynman integrals of the given class. On the other hand, one can check that the initial IBP equations are satisﬁed for the so constructed coeﬃcient functions. Anyway, after successful checks, one can conclude that the obtained solution of the IBP relations is valid and apply it for practical purposes.

References

163

I hope, however, that this method can be put on a solid mathematical ground and, moreover, some interesting mathematics is behind it.

References 1. C. Anastasiou and A. Lazopoulos, JHEP 0407 (2004) 046. 148, 151 2. P.A. Baikov, Phys. Lett. B 385 (1996) 404; Nucl. Instrum. Methods A 389 (1997) 347. 133, 134, 135, 136, 137 3. P.A. Baikov, K.G. Chetyrkin, and J.H. K¨ uhn, Phys. Rev. Lett. 88 (2002) 012001; Phys. Rev. D 67 (2003) 074026; Phys. Lett. B 559 (2003) 245; hepph/0311137. 162 4. P.A. Baikov and V.A. Smirnov, Phys. Lett. B 477 (2000) 367. 133, 136, 137, 162 5. P.A. Baikov and M. Steinhauser, Comput. Phys. Commun. 115 (1998) 161. 133 6. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 147 7. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 133, 162 8. S.G. Gorishny, S.A. Larin, L.R. Surguladze and F.V. Tkachov, Comput. Phys. Commun. 55 (1989) 381; S.A. Larin, F.V. Tkachov and J.A.M. Vermaseren, Preprint NIKHEF-H/91-18 (Amsterdam 1991). 162 9. B.A. Kniehl, A.A. Penin, V.A. Smirnov, and M. Steinhauser, Phys. Rev. D 65 (2002) 091503. 153, 159 10. B.A. Kniehl, A.A. Penin, V.A. Smirnov and M. Steinhauser, Nucl. Phys. B 635 (2002) 357; Phys. Rev. Lett. 90 (2003) 212001; B.A. Kniehl, A.A. Penin, A. Pineda, V.A. Smirnov and M. Steinhauser, Phys. Rev. Lett. 92 (2004) 242001; A.A. Penin, A. Pineda, V.A. Smirnov and M. Steinhauser, hep-ph/ 0403080, hep-ph/0406175. 162 11. S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 148 12. M. Peter, Phys. Rev. Lett. 78 (1997) 602; Nucl. Phys. B 501 (1997) 471. 153 13. Y. Schr¨ oder, Phys. Lett. B 447 (1999) 321. 153 14. V.A. Smirnov and M. Steinhauser, Nucl. Phys. B 672 (2003) 199. 133, 134, 152, 159 15. O.V. Tarasov, Nucl. Phys. B 480 (1996) 397; Phys. Rev. D 54 (1996) 6479. 153, 162

7 Evaluation by Diﬀerential Equations

The method of diﬀerential equations (DE) suggested in [20] and developed in [23] and later works (see references below) is a method of evaluating individual Feynman integrals. We have agreed that, at the present level of complexity of unsolved important problems, it looks unavoidable to decompose the problem of evaluating Feynman integrals of a given family into the reduction to some master integrals and the problem of evaluating these master integrals. Thus, this basic method is oriented at the evaluation of the master integrals. Moreover, in contrast to other methods of evaluating individual Feynman integrals, it is assumed within this method that a solution of the reduction problem is already known. The idea is to take some derivatives of a given master integral with respect to kinematical invariants and masses. Then the result of this diﬀerentiation is written in terms of Feynman integrals of the given family and, according to the known reduction, in terms of the master integrals. Therefore, one obtains a system of diﬀerential equations for the master integrals which can be solved with appropriate boundary conditions. To illustrate basic recipes of this method we shall consider only four examples. The fact is that, for complicated examples, all the calculations can be done only on a computer and intermediate formulae usually happen to be very cumbersome. We shall consider typical one-loop examples in Sect. 7.1 and a two-loop characteristic example in Sect. 7.2. The status of the method, i.e. its perspectives and open problems will be discussed in Sect. 7.3. together with a brief review of its applications.

7.1 One-Loop Examples Of course, we start with our favourite example. Example 7.1. One-loop propagator diagram corresponding to Fig. 1.1. After solving the corresponding reduction problem in Chaps. 5 and 6, we know that there are two master integrals, F (1, 1) = I1 and F (1, 0) = I2 . The second one is a simple one-scale integral given by the right-hand side of V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 165–177 (2004) c Springer-Verlag Berlin Heidelberg 2004

166

7 Evaluation by Diﬀerential Equations

(5.6). We have started to evaluate I1 in Chap. 1, by diﬀerentiating in m2 and arrived at the equation (1.20) for f (m2 ) = F (1, 1). To be very pedantic, let us rewrite it in terms of our true master integrals, ∂ 1 1−ε 2 2 f (m ) = ) − I (1 − 2ε)f (m , (7.1) 2 ∂m2 m2 − q 2 m2 although this does not make an essential diﬀerence here. Let us turn to the new function by f (m2 ) = iπ d/2 (m2 )−ε y(m2 ). We obtain the following diﬀerential equation for it: y −

Γ (ε) m2 (1 − ε) − εq 2 y=− 2 . 2 2 2 m (m − q ) m − q2

(7.2)

It can be solved by the method of the variation of the constant. The general solution to the corresponding homogeneous equation, with a zero on the right-hand side of (7.2), is y(m2 ) = C(m2 − q 2 )1−2ε (m2 )−ε .

(7.3)

Then we make C = C(m2 ) dependent on m2 , solve this equation and obtain " # m2 dx x−ε 2 d/2 2 2 1−2ε + C1 , −Γ (ε) (7.4) f (m ) = iπ (m − q ) (x − q 2 )2−2ε 0 where the constant C1 can be determined from the boundary value f (0) which is a massless one-loop diagram evaluated by means of (A.7). This gives f (m2 ) = −iπ d/2 (m2 − q 2 )1−2ε Γ (ε) # " 2 m dx x−ε Γ (1 − ε)2 − . × (x − q 2 )2−2ε Γ (2 − 2ε)(−q 2 )1−ε 0

(7.5)

If we turn to expansion in ε and take terms up to ε0 into account we shall reproduce (1.7). The next example is also an old one. Example 7.2. The triangle diagram of Fig. 3.4. The reduction problem was solved in Examples 5.4 and 6.7. The only master integral that is not expressed in terms of gamma functions for general d is F (1, 1, 1) = I1 = f (m2 ). We have already calculated it in Examples 3.2 and 4.2. Let us now do this by DE. As in the previous example, we take the ∂ 2 derivative ∂m 2 f (m ) and obtain F (1, 1, 2) for which we apply the relation (6.54), according to our reduction procedure. Let us again, as above, conﬁne ourselves to the evaluation up to the ﬁnite part in ε. Then the ﬁrst term on the right-hand side of (6.54) is irrelevant because it is proportional to ε. So, we obtain, at ε = 0, 2 2 ∂ 2 2 ln(m /Q ) . f (m ) = iπ ∂m2 m2 (m2 − Q2 )

(7.6)

7.1 One-Loop Examples

167

Thus, the evaluation of I1 at d = 4 reduces to taking an integral of the righthand side of (7.6). The boundary condition is simple: this function vanishes in the large mass limit. This can be seen, for example, by examining this behaviour using the MB representation (4.7) as explained in Sect. 4.8. (To do this, one takes a residue at the point z = −1.) Consequently, the known result (3.21) is once again reproduced. If one needs to evaluate I1 at general ε, or obtain higher terms of expansion in ε by DE, one can start from (6.54) and solve the so-obtained diﬀerential equation, applying the method of the variation of the constant quite similarly to Example 7.1. Let us now turn, following [8], to Example 7.3. The on-shell box diagram with two massive and two massless lines shown in Fig. 6.2, with p21 = . . . = p24 = m2 . These are functions of the three variables s, t and m2 . The following combinations naturally arise in the problem: √ √ √ √ 4m2 − s − −s 4m2 − t − −t , y=√ . (7.7) x= √ √ √ 4m2 − s + −s 4m2 − t + −t We again assume that we know a solution of the corresponding reduction problem. It was brieﬂy described in Example 6.9. The reduction based on the algorithm of [16, 21, 22] which was discussed in Sect. 5.4 also leads [8] to the same family of the master integrals shown in Fig. 6.3: I1 = F (1, 1, 1, 1), I2 = F (1, 0, 1, 1) = F (0, 1, 1, 1), I31 = F (1, 1, 0, 0), I32 = F (0, 0, 1, 1) and I4 = F (1, 0, 0, 0) = F (0, 1, 0, 0), where I2 and I4 are present in two copies. Suppose that we want to evaluate I1 by DE. Therefore, we assume that all the master integrals with the number of lines less than four are already known. The integrals I4 and I32 are given by (2.44) and (3.8). The value of the master integral I31 = F (1, 1, 0, 0) is very well-known and can be obtained by various methods. To be self-consistent, let us observe that one can apply MB representation (4.28), set a1 = a2 = 1, a3 = 0 and evaluate this integral by closing the integration contour and summing up the resulting series. Within the method of DE, it is important to present this and later results in terms of the variables (7.7): 1 1 iπ d/2 e−γE ε 1 +2−2 − (7.8) I31 = H0 (x) + O(ε) . (m2 )ε ε 2 1−x Here and in subsequent formulae, usual logarithms and polylogarithms are written in terms of HPL [25] – see Appendix B. Moreover, it is necessary to rewrite the quantity q 2 in (3.8) in terms of these variables, i.e. make the substitution q 2 → t → −(1−y)2 /(m2 y) in the factor (−q 2 )ε and then expand it in ε. Finally, we need I2 which can be obtained using (4.29) at a1 = a2 = a4 = 1 and evaluating this integral by closing the integration contour to the right.

168

7 Evaluation by Diﬀerential Equations

In [8], this result was obtained by DE. It is also naturally written in terms of the variables (7.7): 1 1 2 2 iπ 2 − π I2 = + H (y) + 2H (y) + O(ε) . (7.9) 0,0 0,1 2m2 1 + y 1 − y 3 Observe that higher terms of this and other expansions in ε can be found in [8]. The starting point is to take derivatives in s or t and write them down as a linear combination of integrals of the given class. In order to do this, one observes that taking derivatives in the external momenta reduces to taking derivatives in s and t: ∂ ∂sr ∂ = pi · , ∂pj ∂p j ∂sr r=1 6

pi ·

(7.10)

where si = p2i , i = 1, 2, 3, 4, are invariants with the on-shell condition, si = m2 , and s5 = s, s6 = t. This linear system of six equations can easily be solved, i.e. the derivatives ∂/∂sr can be expressed linearly in terms of the derivatives pi · ∂/∂pj with i, j = 1, 2, 3 – see [8]. One can use here the following expressions [12] which are equivalent to that of [8] due to the on-shell conditions: 1 ∂ s ∂ = (p2 − p3 ) · s , (7.11) p1 + p2 − ∂s 2 4m2 − s − t ∂p2 1 t ∂ ∂ (p2 − p3 ) · . (7.12) t = p3 − p1 − ∂t 2 4m2 − s − t ∂p3 So, we take partial derivatives of I1 = f (s, t) with respect to s and t, using (7.11) and (7.12), and obtain, on the right-hand side, a linear combination of integrals corresponding to Fig. 6.2. Every integral can be written in terms of the master integrals, according to the reduction procedure, and we obtain 1 1 d−5 d−4 ∂f =− + − (7.13) f + g1 , ∂s 2 s 4m2 − s 4m2 − s − t 1 d−6 d−4 ∂f = + (7.14) f + g2 , ∂t 2 t 4m2 − s − t where

# 4m2 − t) 1 1 − + I2 g1 = −(d − 4) 4m2 s 4m2 t(4m2 − s) t(4m2 − s − t) " # 1 2(d − 3) 1 1 − + + I31 t (4m2 − s)2 t(4m2 − s) t(4m2 − s − t) " # 1 d−3 1 + − 2 I32 2m − t s 4m2 − s "

7.1 One-Loop Examples

"

169

#

1 1 d−2 1 − I4 , + 2 2 2 2 2 m t (4m − s) t(4m − s) t(4m − s − t) " # 1 d−4 1 + I2 g2 = − 2 4m − s t 4m2 − s − t # " 1 2(d − 3) 1 + I31 − (4m2 − s)2 t 4m2 − s − t " # 1 d−2 1 + − 2 I4 . m (4m2 − s)2 t 4m2 − s − t +

(7.15)

(7.16)

It is suﬃcient to use one of the two equations to evaluate f (s, t). Let it be (7.13). Then (7.14) can be used for a non-trivial check. One needs also a boundary condition when solving (7.13): it can be obtained using the fact that the function f (s, t) is regular at s = 0. Multiplying (7.13) by s and taking the limit s → 0 one obtains d−3 d−4 I2 + 2 I32 . 2m2 m t Equation (7.13) can be solved in a Laurent expansion in ε, fj (s, t)εj . f (s, t) =

f (0, t) = −

(7.17)

(7.18)

j=−1

As a result, one obtains a set of nested diﬀerential equations from (7.13), dfj 1 1 1 =− + (7.19) fj + hj , ds 2 s 4m2 − s where the functions hj involve, in addition to the corresponding term of the expansion of the function g1 , a piece coming from fj−1 . These equations can be solved by the method of the variation of the constant. The homogeneous equation corresponding to (7.19), which is the same for all fj , takes the following form in the new variables, x and y: d 1 1 1 − + − (7.20) f (0) (x) = 0 , dx x 1 + x 1 − x with the solution f (0) (x) =

x . (1 − x)(1 + x)

(7.21)

Then the solution of the j-th diﬀerential equation in (7.19) can be written as

fj (x, y) = f

(0)

(x) Aj +

hj (x, y) dx (0) f (x)

,

(7.22)

where Aj is a constant which can be ﬁxed by imposing the boundary condition (7.17) expanded in ε.

170

7 Evaluation by Diﬀerential Equations

Observe that the combinations of the kinematical invariants involved on the right-hand side of (7.13) and (7.15) and, therefore, present in hj can be represented as 4m2 − s = m2

(1 + x)2 , x

4m2 − s − t = m2

(x + y)(1 + xy) . xy

(7.23)

After that the integration in (7.22), order by order in ε, becomes straightforward. All the quantities are prepared in such a form that the integration is taken in terms of HPL of the next level, also of the arguments x and y. So, one arrives at (4.27). However, keeping in mind that this very master integral can be needed when evaluating other master integrals in two loops, also by the method of DE, it is reasonable to present it in the same form as its ingredients were presented: 1 1 1 1 iπ d/2 e−γE ε I1 = − − H0 (x) (m2 )2+ε 1 + x 1 − x 1 − y (1 − y)2 1 + H0 (y) + 2H1 (y) + O(ε) . (7.24) × ε Further terms of this expansion in ε can be found in [8].

7.2 Two-Loop Example We turn again to Feynman integrals considered in Examples 4.10 and 6.10. Example 7.4. Sunset diagram of Fig. 3.12 with one zero mass and two equal non-zero masses at a general value of the external momentum squared. The general Feynman integral of this class is given by (6.57), so that there are two irreducible numerators in the problem. According to Example 6.10, we know a solution of the reduction problem, and that there are three master integrals, I1 = F (1, 1, 0, 0, 1), I¯1 = F (1, 1, −1, 0, 1) and I2 = F (1, 1, 0, 0, 0). The last of them is the square of the massive tadpole given by the right-hand side of (2.44). Let us now evaluate I1 and I¯1 by DE. For convenience, let us use, instead of I¯1 , the integral with a1 = a2 = a5 = 1 and the numerator equal to the product of the momenta (ﬂowing in the same direction) of the massless and one of the massive lines, 1 2 q I1 − I¯1 − I2 . (7.25) I˜1 = 2 We start with taking derivatives. We use the homogeneity of the integrals I1 and I˜1 with respect to q 2 and m2 , with the help of Euler’s theorem, set q 2 = s and obtain ∂ sf (s) = (1 − 2ε)f (s) − f (s) , (7.26) ∂m2 ∂ ˜ sf˜ (s) = 2(1 − ε)f˜ (s) − f (s) , (7.27) ∂m2

7.2 Two-Loop Example

171

where f (s) = I1 and f˜(s) = I˜1 , and we have already put m2 = 1 after diﬀerentiating with respect to the mass which results in indices equal to 2 instead of 1 on one of the massive lines. We apply (6.60)–(6.62) to these integrals with the indices equal to two in order to obtain only the master integrals on the right-hand side. Therefore, we arrive at the following diﬀerential equations for the functions f (s) and f˜(s): sf (s) =

1 [(3s − 2 − 4ε(s − 1)) f (s) s−4 $ +4(ε − 1)(h(s) + 3f˜(s)) ,

(7.28) $ 1 (7.29) sf˜ (s) = (ε − 1) h(s) − sf (s) + 2f˜(s) , 2 where h originates from I2 . As in the previous example, it is convenient to turn to the new variable x given by (7.7), or, vice versa, '

(1 − x)2 . x Then we obtain the following equations:

1 3 − 4x + 3x2 − 4ε(1 − x + x2 ) f (x) f (x) = x(x2 − 1) $ −4(ε − 1)x(h(x) + 3f˜(x)) ,

(7.31)

1 (ε − 1)(1 + x) 2x2 (x − 1) ' $ × (x − 1)2 f (x) + x(h(x) + 2f˜(x)) .

(7.32)

s=−

f˜ (x) =

(7.30)

The second function f˜(x) can be eliminated from this system in order to obtain a separate equation for the ﬁrst one: (3ε(x − 1)2 + 6x − 2) f (x) x(x2 − 1) 2(ε − 1)2 (2ε − 1)(2x + ε(1 − 4x + x2 )) f (x) + h(x) = 0 . + 2 2 x (x − 1) x(x − 1)2

f (x) +

(7.33)

Then we turn to solving this equation in expansion in ε, as in the previous examples, f−2 (x) f−1 (x) + f0 (x) + . . . . + (7.34) ε2 ε As usual, we need a general solution of the corresponding homogeneous equation at ε = 0: 2(3x − 1) 2 f (x) − f (x) + f (x) = 0 . (7.35) x(x2 − 1) x(x − 1)2 f (x) =

172

7 Evaluation by Diﬀerential Equations

Two independent solutions are 1 − x + x2 , (x − 1)2 4x(1 − x + x2 )H0 (x) − 1 + 7x − 3x2 − x3 + x4 , φ2 (x) = x(x − 1)2

φ1 (x) =

(7.36) (7.37)

with the Wronskian w(x) =

(x + 1)4 . x2 (x − 1)2

(7.38)

The solutions are presented in a form similar to the previous example, in terms of HPL. The equation for f−2 has the inhomogeneous term r−2 (x) = −

2 . x(x − 1)2

Its solution is written as φ2 (x)r−2 (x) f−2 (x) = c1 − dx φ1 (x) w(x) φ1 (x)r−2 (x) + c2 + dx φ2 (x) , w(x)

(7.39)

(7.40)

where c1 and c2 are integration constants. We obtain

1 x(c1 (1 − x + x2 ) − x) − c2 (1 − 7x + 3x2 + x3 − x4 ) f−2 (x) = 2 x(x − 1) +4c2 x(1 − x + x2 )H0 (x) . (7.41) The integration constants are evaluated from the regular behaviour of the √ solution at x → 0 so that 1/x and x in the asymptotic expansion of (7.41) are forbidden. This gives the values c1 = 1 and c2 = 0, with f−2 (x) = 1 .

(7.42)

The inhomogeneous term for f1 (x) is r−1 (x) =

1 − 8x + x2 . x2 (x − 1)2

(7.43)

Proceeding in a similar way we obtain the following solution:

1 1 − 6x − x2 − 2x3 + 2c1 x(1 − x + x2 ) f−1 (x) = 2x(x − 1)2

−2c2 (1 − 7x + 3x2 + x3 − x4 ) + 2(4c2 − 1)x(1 − x + x2 )H0 (x) . (7.44)

The regularity condition at x = 0 gives c1 = 13/4 and c2 = 1/4, with f−1 (x) =

1 + 10x + x2 . 4x

(7.45)

7.3 Conclusion

173

Finally, for f0 , we have the inhomogeneous term r0 (x) = −

3 − 9x + 2(48 + π 2 )x2 − 9x3 + 3x4 . 6x3 (x − 1)2

(7.46)

Similarly, we obtain the following solution:

1 (x − 1)2 (39 + 66x + 4π 2 x + 39x2 ) f0 (x) = 2 24x(x − 1)

+12(1 − 4x + 4x3 − x4 )H0 (x) − 48x(1 − x + x2 )H0,0 (x) . (7.47)

The second function f˜−2 (x) f˜−1 (x) f˜ = + f˜0 (x) + . . . . + (7.48) ε2 ε can be now obtained in a pure algebraic way, with the following results: 2

1+x , f˜−2 (x) = − 4x 1 + 11x + 11x3 + x4 f˜−1 (x) = − , 24x2

1 −(x − 1)2 (2π 2 − 11)x(1 + x2 ) f˜0 (x) = 2 2 48x (x − 1) +13(1 + x4 ) + 44x2 − 4 1 − 9x(1 − x2 )(1 − x + x2 ) − x6 H0 (x) +24x(1 − 2x + 4x2 − 2x3 + x4 )H0,0 (x) . (7.49) The corresponding result for the master integral I¯1 can be obtained easily from (7.42), (7.44), (7.47) and (7.49), using (7.25). It can be evaluated also using the onefold MB representation (4.76) (with another choice of the numerator). These results are in agreement with [11, 13], where another choice of the master integrals was used (with higher powers of the propagators, instead of integrals with numerators).

7.3 Conclusion At ﬁrst sight, the method of DE cannot be applied to integrals dependent on one scale since the dependence on the only scale parameter is trivial and can be obtained immediately by power counting. However, one can introduce, for a one-scale integral, an additional scale parameter, apply the corresponding diﬀerential equation, get the boundary condition at a diﬀerent, more suitable point and then return to the single scale value. An example of this strategy can be found in [5]. I admit that it might seem, from the previous examples1 , that the method of DE is not optimal. In particular, the results for Example 7.4 can be, probably, derived by MB representation in a simpler way. However, the method of 1

Simple instructive examples can be found also in the review [1].

174

7 Evaluation by Diﬀerential Equations

DE is indeed very powerful and, in some situations, the very best one. An important feature of the strategy outlined above is that it can straightforwardly be generalized to more complicated classes of multiloop Feynman integrals, with a computer implementation of all the steps. The method of DE, coupled with solving the reduction problem by use of IBP and LI relations by means of the algorithm of [16, 21, 22], has become, by now, a powerful industry for obtaining results for various phenomenologically important classes of Feynman integrals – see, e.g., [2, 3, 6, 7, 8, 9, 15, 27]. The method of DE was also successfully applied [10, 24] for the analytical evaluation of various (generalized) sunset diagrams.2 However, the ﬁrst impressive example of this technique was evaluating master integrals by DE for the massless double boxes with one leg oﬀ-shell, p21 = 0, p22 = p23 = p24 = 0, performed in [16]. Another important feature of the method of DE is that it provides a natural solution in the situation where results obtained can be hardly expressible in terms of known special functions of mathematical physics. The very form of results obtained when applying DE, by means of iterative integrations, naturally leads, in such a situation, to the idea to introduce new functions which would be adequate to express the results for the given class of the integrals. This is how two-dimensional HPL (2dHPL) [16], new special functions of mathematical physics introduced and studied by physicists, have appeared. They are natural generalizations of HPL to the case of functions of two variables. To deﬁne them [16] one uses, instead of the functions (B.10), the following set of functions of the two variables x and y labelled by the four indices 0, −1, −y and −1/y: g(0; x) =

1 1 1 , g(−1; x) = , g(−y; x) = , x 1+x x+y 1 g(−1/y; x) = . x + 1/y

(7.50) (7.51)

Then 2dHPLs are deﬁned as the set of functions generated by repeated integrations with these functions similarly to (B.9). Some basic properties of these new functions were studied and packages for the numerical evaluation were provided [17, 18]. These are 2dHPL that have turned out to be adequate functions to express results for the double boxes with one leg oﬀ shell [16]. This strategy of inventing new special functions, in situations where one fails to express results in terms of the known functions3 , has already become 2 For generalized sunset diagrams (i.e. with an arbitrary number of lines between two external vertices), a successful alternative technique is based on the coordinate space representation, where any such diagram is just a product of the propagators in coordinate space given by a Bessel function – see (2.16). Then, in order to go back to momentum space, it is necessary to evaluate a one-dimensional (but complicated) integral of this product of the Bessel functions with one more Bessel function – see [19] and references therein. 3 Of course, we already consider HPL and 2dHPL as known functions.

7.3 Conclusion

175

standard. In 2004, at least two types of new functions were introduced: generalized HPL in [3] which were necessary to evaluate some two-loop massive Feynman diagrams and some generalized 2dHPL [7] which were necessary to evaluate two-loop massless diagrams with three oﬀ-shell legs. Pragmatically, the introduction of new functions is just a way to parameterize the results obtained. Then one has at least a deﬁnite procedure for the numerical evaluation of any of the calculated integrals with a reasonable accuracy. Mathematically, if one introduces a new class of functions, there is an implicit obligation to describe their properties and present procedures for their numerical evaluation. Of course, it is natural to try to represent results in known functions. Observe that, in the above examples where the new functions were introduced, at least some of the new functions can be expressed in terms of the standard special functions. Consider, for example, the generalized HPL of various types which were deﬁned in [3] similarly to the HPL, with other basic functions, in particular 1/ t(t + 4). Observe that the new generalized HPL x dt H(−r, −1; x) = (7.52) t(t + 4) 0 equals 1 1 π2 Li2 −y 3 − Li2 (−y) + ln2 y − , (7.53) 3 2 18 √ √ √ √ where y = ( 4 + x − x)/( 4 + x − x). For more complicated generalized HPL, similar representations can hardly be found. Still nobody has proven a no go theorem for this situation. Moreover, it is not clear how to take into account all possible choices of special combinations of the initial variables such as the y(x) above. Anyway, physicists are naturally impatient to report on their results and apply them for the evaluation of physical quantities, so that, I hope, mathematicians will not blame them for this, keeping in mind that the mathematicians themselves seem not to bother about these interesting mathematical problems at the moment. Let us now remember about the evaluation of the massive on-shell QEDtype double boxes of Figs. (4.9) and (4.10). Two of our four examples were in fact oriented at this problem: its one-loop prototype and the sunset diagrams that can be obtained from the massive double boxes – see Sect. 4.5. In [12], it was reported about the solution of the reduction problem, by an authors’ implementation of the algorithm of [16, 21, 22]. The number of master integrals is 22 in the ﬁrst planar case, 35 in the second planar case, and 47 in the non-planar case. The diagrams with three reduced lines and some of the diagrams with two reduced lines have been calculated by DE [12]. When applying the method of DE to diagrams with six and seven lines, serious problems arise because diﬀerential equations of third order and higher are encountered there. One may still hope to solve such equations or choose

176

7 Evaluation by Diﬀerential Equations

another strategy. This could be the method of MB representation which was used to evaluate the most complicated planar master integrals – see Example 4.9 and [26]. So, hopefully, the problem of the evaluation of the massive on-shell double boxes will be completely solved in the nearest future, as well as other phenomenologically important calculational problems at least at the two-loop level.

References 1. U. Aglietti, hep-ph/0408014. 173 2. U. Aglietti and R. Bonciani, Nucl. Phys. B 668 (2003) 3; U. Aglietti, R. Bonciani, G. Degrassi and A. Vicini, hep-ph/0407162. 174 3. U. Aglietti and R. Bonciani, hep-ph/0401193; U. Aglietti, R. Bonciani, G. Degrassi and A. Vicini, Phys. Lett. B 595 (2004) 432. 174, 175 4. C. Anastasiou and A. Lazopoulos, JHEP 0407 (2004) 046. 5. M. Awramik, M. Czakon, A. Freitas and G. Weiglein, hep-ph/0407317. 173 6. W. Bernreuther et al., hep-ph/0406046. 174 7. T.G. Birthwright, E.W.N. Glover and P. Marquard, hep-ph/0407343. 174, 175 8. R. Bonciani, A. Ferroglia, P. Mastrolia, E. Remiddi and J.J. van der Bij, Nucl. Phys. B 681 (2004) 261. 167, 168, 170, 174 9. R. Bonciani, P. Mastrolia and E. Remiddi, Nucl. Phys. B 661 (2003) 289; B 676, 399 (2004); B 690, 138 (2004). 174 10. M. Caﬀo, H. Czy˙z, S. Laporta and E. Remiddi, Nuovo Cim. A 111 (1998) 365; Acta Phys. Polon. B 29 (1998) 2627; M. Caﬀo, H. Czy˙z and E. Remiddi, Nucl. Phys. B 581 (2000) 274; B 611 (2001) 503. 174 11. K.G. Chetyrkin and F.V. Tkachov, Nucl. Phys. B 192 (1981) 159. 12. M. Czakon, J. Gluza and T. Riemann, hep-ph/0406203. 168, 175 13. A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 699 (2004) 3. 14. J. Fleischer, M.Yu. Kalmykov and A.V. Kotikov, Phys. Lett. B 462 (1999) 169. 15. T. Gehrmann and E. Remiddi, Nucl. Phys. B 580 (2000) 485. 174 16. T. Gehrmann and E. Remiddi, Nucl. Phys. B 601 (2001) 248; Nucl. Phys. B 601 (2001) 287. 167, 174, 175 17. T. Gehrmann and E. Remiddi, Comput. Phys. Commun. 144 (2002) 200; 141 (2001) 296. 174 18. T. Gehrmann and E. Remiddi, Nucl. Phys. B 640 (2002) 379. 174 19. S. Groote, J.G. Korner and A.A. Pivovarov, Eur. Phys. J. C 36 (2004) 471. 174 20. A.V. Kotikov, Phys. Lett. B 254 (1991) 158; B 259 (1991) 314; B 267 (1991) 123; Mod. Phys. Lett. A 6 (1991) 677; 3133; Int. J. Mod. Phys. A 7 (1992) 1977. 165 21. S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 167, 174, 175 22. S. Laporta and E. Remiddi, Phys. Lett. B 379 (1996) 283. 167, 174, 175 23. E. Remiddi, Nuovo Cim. A 110 (1997) 1435. 165 24. E. Remiddi, Acta Phys. Polon. B 34 (2003) 5311; M. Argeri, P. Mastrolia and E. Remiddi, Nucl. Phys. B 631 (2002) 388; P. Mastrolia and E. Remiddi, Nucl. Phys. B 657 (2003) 397; S. Laporta, P. Mastrolia and E. Remiddi, Nucl. Phys. B 688 (2004) 165; S. Laporta and E. Remiddi, hep-ph/0406160. 174 25. E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725. 167

References

177

26. V.A. Smirnov, hep-ph/0406052; G. Heinrich and V.A. Smirnov, hep-ph/ 0406053. 176 27. Y. Schr¨ oder, Nucl. Phys. Proc. Suppl. 116 (2003) 402; Y. Schr¨ oder and A. Vuorinen, hep-ph/0311323. 174

A Tables

A.1 Table of Integrals Each Feynman integral presented here can be evaluated straightforwardly by use of alpha or Feynman parameters. Results are presented for the ‘Euclidean’ dependence, −k 2 , of the denominators, which is more natural when the powers of propagators are general complex numbers. As usual, −k 2 is understood in the sense of −k 2 − i0, etc. Moreover, denominators with a linear dependence on k are also understood in this sense, e.g. 2p · k → 2p · k − i0, although sometimes this i0 dependence is explicitly indicated to avoid misunderstanding. 1 Γ (λ + ε − 2) dd k = iπ d/2 . (A.1) 2 2 λ 2 (−k + m ) Γ (λ) (m )λ+ε−2 Γ (λ − n + ε − 2) (−1)n gsα1 ...α2n k α1 . . . k α2n = iπ d/2 , (A.2) dd k 2 2 λ (−k + m ) 2n Γ (λ) (m2 )λ−n+ε−2 where gsα1 ...α2n = g α1 α2 . . . g α2n−1 α2n +. . . (with (2n−1)!! terms in the sum) is a combination symmetrical with respect to the permutation of any pair of indices. If the number of monomials in the numerator is odd, the corresponding integral is zero. (2l·k)2n dd k (−k 2 + m2 )λ (l2 )n Γ (λ − n + ε − 2) = iπ d/2 (−1)n (2n − 1)!! . (A.3) 2 Γ (λ) (m )λ−n+ε−2

dd k + m2 )λ1 (−k 2 )λ2 Γ (λ1 + λ2 + ε − 2)Γ (−λ2 − ε + 2) 1 = iπ d/2 . 2 λ +λ 1 2 +ε−2 Γ (λ1 )Γ (2 − ε) (m ) (−k 2

(A.4)

k α1 . . . k α2n (−k 2 + m2 )λ1 (−k 2 )λ2 (−1)n Γ (λ1 + λ2 − n + ε − 2)Γ (n − λ2 − ε + 2) = iπ d/2 n gsα1 ...α2n . 2 Γ (λ1 )Γ (n − ε + 2)(m2 )λ1 +λ2 −n+ε−2 dd k

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 179–186 (2004) c Springer-Verlag Berlin Heidelberg 2004

(A.5)

180

A Tables

dd k

(2l·k)2n = iπ d/2 (−1)n (2n − 1)!! (−k 2 + m2 )λ1 (−k 2 )λ2 Γ (λ1 + λ2 − n + ε − 2)Γ (n − λ2 − ε + 2)(l2 )n . × Γ (λ1 )Γ (n − ε + 2)(m2 )λ1 +λ2 −n+ε−2

(A.6)

dd k (−k 2 )λ1 [−(q − k)2 ]λ2 Γ (2 − ε − λ1 )Γ (2 − ε − λ2 ) Γ (λ1 + λ2 + ε − 2) = iπ d/2 . Γ (λ1 )Γ (λ2 )Γ (4 − λ1 − λ2 − 2ε) (−q 2 )λ1 +λ2 +ε−2

(A.7)

Let k (α1 ...αn ) = k α1 . . . k αn + . . . be traceless with respect to any pair of indices, i.e. gαi αj k (α1 ...αn ) = 0 – see (A.41b) below. Then (α1 ...αn ) k (α1 ...αn ) d/2 AT (λ1 , λ2 ; n)q = iπ , (A.8) dd k (−k 2 )λ1 [−(q − k)2 ]λ2 (−q 2 )λ1 +λ2 +ε−2 where AT (λ1 , λ2 ; n) =

Γ (λ1 + λ2 + ε − 2)Γ (n + 2 − ε − λ1 )Γ (2 − ε − λ2 ) . Γ (λ1 )Γ (λ2 )Γ (4 + n − λ1 − λ2 − 2ε) (A.9)

For pure monomials, the corresponding formula has one more ﬁnite summation: k α1 . . . k αn dd k (−k 2 )λ1 [−(q − k)2 ]λ2 [n/2] iπ d/2 1 = ANT (λ1 , λ2 ; r, n) r (q 2 )r {[g]r [q]n−2r }α1 ...αn , (−q 2 )λ1 +λ2 +ε−2 r=0 2

(A.10) where ANT (λ1 , λ2 ; r, n) Γ (λ1 + λ2 + ε − 2 − r)Γ (n + 2 − ε − λ1 − r)Γ (2 − ε − λ2 + r) , = Γ (λ1 )Γ (λ2 )Γ (4 + n − λ1 − λ2 − 2ε) (A.11) and {[g]r [q]n−2r }α1 ...αn is symmetric in its indices and is composed of the metric tensor and the vector q. (2l·k)n iπ d/2 = dd k (−k 2 )λ1 [−(q − k)2 ]λ2 (−q 2 )λ1 +λ2 +ε−2

[n/2]

×

r=0

ANT (λ1 , λ2 ; r, n)

n! (q 2 )r (l2 )r (2q·l)n−2r , r!(n − 2r)!

(A.12)

A.1 Table of Integrals

181

dd k (−k 2 )λ1 (−k 2 + 2p·k)λ2 1 Γ (λ1 + λ2 + ε − 2)Γ (−2λ1 − λ2 − 2ε + 4) = iπ d/2 . Γ (λ2 )Γ (−λ1 − λ2 − 2ε + 4) (p2 )λ1 +λ2 +ε−2 (A.13)

dd k

k (α1 ...αn ) p(α1 ...αn ) d/2 = iπ B (λ , λ ; n) , T 1 2 λ 2 + 2p·k) 2 (p )λ1 +λ2 +ε−2 (A.14)

(−k 2 )λ1 (−k 2

where BT (λ1 , λ2 ; n) = dd k

Γ (λ1 + λ2 + ε − 2)Γ (−2λ1 − λ2 + n − 2ε + 4) . Γ (λ2 )Γ (−λ1 − λ2 + n − 2ε + 4)

(A.15)

k α1 . . . k αn iπ d/2 = (−k 2 )λ1 (−k 2 + 2p·k)λ2 (p2 )λ1 +λ2 +ε−2

[n/2]

×

BNT (λ1 , λ2 ; r, n)

r=0

(−1)r 2 r (p ) {[g]r [p]n−2r }α1 ...αn , 2r

(A.16)

where BNT (λ1 , λ2 ; r, n) Γ (λ1 + λ2 + ε − 2 − r)Γ (−2λ1 − λ2 + n − 2ε + 4) . = Γ (λ2 )Γ (−λ1 − λ2 + n − 2ε + 4) dd k

(2l·k)n iπ d/2 = λ 2 λ + 2p·k) 2 (q ) 1 +λ2 +ε−2

(−k 2 )λ1 (−k 2

[n/2]

×

(A.17)

BNT (λ1 , λ2 ; r, n)(−1)r

r=0

n! (p2 )r (l2 )r (2p·l)n−2r . r!(n − 2r)!

(A.18)

Let p·q = 0. Then (p·k)b1 (q·k)b2 dd k (−k 2 )λ1 [−(l − k)2 ]λ2 iπ d/2 = (−l2 )λ1 +λ2 +ε−2

min{r,[b1 /2]}

×

r1 =max{0,r−[b2 /2]}

and

[(b1 +b2 )/2]

r=0

ANT (λ1 , λ2 ; r, b1 + b2 )

b1 !b2 ! 2 r (l ) 4r

(p·l)b1 −2r1 (q·l)b2 −2r+2r1 (p2 )r1 (q 2 )r−r1 , r1 !(r − r1 )!(b1 − 2r1 )!(b2 − 2r + 2r1 )!

(A.19)

182

A Tables

dd k

(p·k)b1 (q·k)b2 (−k 2 )λ1 (−k 2 + 2q·k)λ2 = iπ d/2

(p2 )b1 /2 (q 2 )λ1 +λ2 +ε−2−b1 /2−b2

Bpq (λ1 , λ2 ; b1 , b2 ) ,

(A.20)

for even b1 (and are equal to zero for odd b1 ), where Bpq (λ1 , λ2 ; b1 , b2 )

b1 /2+[b2 /2]

=

r=b1 /2

(A.21)

dd k (−k 2 + m2 )λ1 (2p·k)λ2 Γ (λ2 /2)Γ (λ1 + λ2 /2 + ε − 2) iπ d/2 . 2 λ /2 2 λ +λ /2+ε−2 2 1 2 2Γ (λ1 )Γ (λ2 ) (p ) (m )

= dd k

(−k 2

= iπ d/2

(−1)r b1 !b2 ! BNT (λ1 , λ2 ; r, b1 + b2 ) . r 4 (b1 /2)!(r − b1 /2)!

(A.22)

k (α1 ,...,αn ) + m2 )λ1 (2p·k)λ2 Γ ((λ2 + n)/2) Γ (λ1 + (λ2 − n)/2 + ε − 2) p(α1 ,...,αn ) . 2Γ (λ1 )Γ (λ2 ) (m2 )λ1 +(λ2 −n)/2+ε−2 (p2 )(λ2 +n)/2 (A.23)

dd k (−k 2 + 2p·k)λ1 (2p·k)λ2 =

iπ d/2 (p2 )λ1 +λ2 +ε−2

Γ (λ1 + λ2 + ε − 2)Γ (2λ1 + λ2 + 2ε − 4) . Γ (λ1 )Γ (2λ1 + 2λ2 + 2ε − 4)

(A.24)

dd k + ω − i0)λ2 Γ (2 − λ1 − ε)Γ (2λ1 + λ2 + 2ε − 4) 2 λ1 +ε−2 −2λ1 −λ2 −2ε+4 (v ) = iπ d/2 ω . Γ (λ1 )Γ (λ2 ) (A.25) (−k 2 )λ1 (2v·k

dd k ×

k (α1 ,...,αn ) = iπ d/2 ω −2λ1 −λ2 +n−2ε+4 (−k 2 )λ1 (2v·k + ω − i0)λ2

v (α1 ,...,αn ) (v 2 )−λ1 +n−ε+2

Let v·q = 0. Then

Γ (2 − λ1 + n − ε)Γ (2λ1 + λ2 − n + 2ε − 4) . Γ (λ1 )Γ (λ2 )

(A.26)

A.1 Table of Integrals

dd k (−k 2 )λ1 [−(q − k)2 ]λ2 (−2v·k − i0)λ3 Γ (−λ1 − λ3 /2 − ε + 2)Γ (−λ2 − λ3 /2 − ε + 2) = iπ d/2 Γ (−λ1 − λ2 − λ3 − 2ε + 4) Γ (λ1 + λ2 + λ3 /2 + ε − 2)Γ (λ3 /2) . × 2Γ (λ1 )Γ (λ2 )Γ (λ3 )(−q 2 )λ1 +λ2 +λ3 /2+ε−2 (v 2 )λ3 /2

Let p21 = p22 = 0, q = p1 − p2 . Then dd k 2 λ (−k + 2p1 ·k) 1 (−k 2 + 2p2 ·k)λ2 (−k 2 )λ3 Γ (−λ1 − λ3 − ε + 2)Γ (−λ2 − λ3 − ε + 2) = iπ d/2 Γ (λ1 )Γ (λ2 )Γ (−λ1 − λ2 − λ3 − 2ε + 4) Γ (λ1 + λ2 + λ3 + ε − 2) , × (−q 2 )λ1 +λ2 +λ3 +ε−2 (−k 2

183

(A.27)

(A.28)

Γ (−λ1 − ε + 2) dd k = iπ d/2 + 2p1 + 2p2 ·k)λ2 (2p2 ·k)λ3 Γ (λ1 )Γ (λ2 ) Γ (λ1 + λ2 + ε − 2)Γ (−λ2 − λ3 − ε + 2) , (A.29) × Γ (−λ1 − λ2 − λ3 − 2ε + 4)(−q 2 )λ1 +λ2 +λ3 +ε−2 ·k)λ1 (−k 2

dd k (2p1 + 2p2 ·k)λ2 (−k 2 + m2 )λ3 Γ (λ2 − λ1 )Γ (λ2 + λ3 + ε − 2)Γ (−λ2 − ε + 2) = iπ d/2 , Γ (λ2 )Γ (λ3 )Γ (−λ1 − ε + 2)(−q 2 )λ1 (m2 )λ2 +λ3 +ε−2 ·k)λ1 (−k 2

(A.30)

dd k + 2p2 + m2 )λ3 (Q2 − 2p1 ·k)λ4 Γ (λ2 − λ1 )Γ (λ2 + λ3 + ε − 2)Γ (−λ2 − λ4 − ε + 2) = iπ d/2 Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ4 − ε + 2) 1 × 2 λ1 +λ4 2 λ2 +λ3 +ε−2 , (Q ) (m )

(2p1

·k)λ1 (−k 2

·k)λ2 (−k 2

(A.31)

dd k (2p1 ·k + m2 )λ1 (2p2 ·k + m2 )λ2 (−k 2 )λ3 Γ (λ1 + λ3 + ε − 2)Γ (λ2 + λ3 + ε − 2)Γ (−λ3 − ε + 2) = iπ d/2 . Γ (λ1 )Γ (λ2 )Γ (λ3 )(−q 2 )−λ3 −ε+2 (m2 )λ1 +λ2 +2λ3 +2ε−4

(A.32)

Let p21 = 0, p22 = −m2 , q = p1 − p2 . Then Γ (λ2 + λ3 + ε − 2) dd k = iπ d/2 (2p1 ·k)λ1 (−k 2 + 2p2 ·k + m2 )λ2 (−k 2 )λ3 (m2 )λ2 +λ3 +ε−2 Γ (−λ1 − λ3 − ε + 2)Γ (−λ2 − ε + 2) × , (A.33) Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ2 − λ3 − 2ε + 4)(−q 2 )λ1

184

A Tables

dd k (2p1 ·k)λ1 (−k 2 + 2p2 ·k − m2 )λ2 (−k 2 )λ3 (−q 2 − 2p1 ·k)λ4 Γ (λ2 + λ3 + ε − 2) = iπ d/2 (m2 )λ2 +λ3 +ε−2 Γ (−λ1 − λ3 − ε + 2)Γ (−λ2 − λ4 − ε + 2) × . Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ2 − λ3 − λ4 − 2ε + 4)(−q 2 )λ1 +λ4

(A.34)

Let P 2 = M 2 , p2 = 0, (P − p)2 = 0. Then dd k 2 λ (−k + 2P ·k) 1 (−k 2 + 2p·k)λ2 (−k 2 )λ3 Γ (−λ1 − λ2 − 2λ3 − 2ε + 4)Γ (λ1 + λ2 + λ3 + ε − 2) = iπ d/2 Γ (λ1 )Γ (−λ1 − λ2 − λ3 − 2ε + 4) Γ (−λ2 − λ3 − ε + 2) . (A.35) × Γ (−λ3 − ε + 2)(M 2 )λ1 +λ2 +λ3 +ε−2 Let p21 = 0, p22 = m2 , Q2 = 2p1 ·p2 . Then dd k (2p1 ·k)λ1 (−k 2 + 2p2 ·k)λ2 (−k 2 )λ3 (Q2 − 2p1 ·k)λ4 Γ (λ3 − λ4 )Γ (−λ1 − λ2 − 2λ3 − 2ε + 4) = iπ d/2 Γ (λ2 )Γ (λ3 )Γ (−λ1 − λ2 − λ3 − λ4 − 2ε + 4) Γ (λ2 + λ3 + ε − 2) × 2 λ1 +λ4 2 λ2 +λ3 +ε−2 , (Q ) (m )

iπ d/2 dd k = (2p1 ·k)λ1 (−k 2 + 2p2 ·k)λ2 (−k 2 )λ3 (Q2 )λ1 (m2 )λ2 +λ3 +ε−2 Γ (λ2 + λ3 + ε − 2)Γ (−λ1 − λ2 − 2λ3 − 2ε + 4) . × Γ (λ2 )Γ (−λ1 − λ2 − λ3 − 2ε + 4)

The following integrals are related to two-loop diagrams: dd k dd l 2 2 λ 1 (−k + m ) [−(k + l)2 ]λ2 (−l2 + m2 )λ3 2 Γ (λ + λ + ε − 2)Γ (λ + λ + ε − 2)Γ (2 − ε − λ ) 1 2 2 3 2 = iπ d/2 Γ (λ1 )Γ (λ3 ) Γ (λ1 + λ2 + λ3 + 2ε − 4) , × Γ (λ1 + 2λ2 + λ3 + 2ε − 4)Γ (2 − ε)(m2 )λ1 +λ2 +λ3 +2ε−4

(A.36)

(A.37)

(A.38)

dd k dd l (−k 2 )λ1 [−(k + l)2 ]λ2 (m2 − l2 )λ3 2 Γ (λ + λ + λ + 2ε − 4) 1 2 3 = iπ d/2 (m2 )λ1 +λ2 +λ3 +2ε−4 Γ (λ1 + λ2 + ε − 2)Γ (2 − ε − λ1 )Γ (2 − ε − λ2 ) . × Γ (λ1 )Γ (λ2 )Γ (λ3 )Γ (2 − ε)

(A.39)

A.2 Some Useful Formulae

185

This is the (inverse) Fourier transformation of (−q 2 − i0)−λ in d dimensions: 1 1 e−ix·q iΓ (d/2 − λ) d q = λ d/2 . (A.40) d d 2 λ 2 (2π) (−q − i0) 4 π Γ (λ) (−x + i0)d/2−λ

A.2 Some Useful Formulae To traceless expressions and back: k

α1

αN

...k

[N/2] 1 1 = (k 2 )r {[g]r [k](N −2r) }α1 ...αN , N ! r=0 2r (d/2 + N − 2r)r

(A.41a) 1 1 (k 2 )r {[g]r [k]N −2r }α1 ...αN , r N ! r=0 2 (2 − N − d/2)r [N/2]

k (α1 ...αN ) =

(A.41b) where {[g]r [k]N −2r }α1 ...αN is deﬁned after (A.11) and (a)n is the Pochhammer symbol (B.2). Furthermore,

[N/2] N

(k·p)

=

aN,r (k 2 )r (p2 )r (k·p)(N −2r) ,

(A.42)

r=0

[N/2] (N )

(k·p)

=

bN,r (k 2 )r (p2 )r (k·p)N −2r ,

(A.43)

(d − 2)N (k 2 )N , 2N ((d − 2)/2)N

(A.44)

r=0

k(α1 ...αN ) k (α1 ...αN ) =

where (k · p)(N ) = k(α1 ...αN ) p(α1 ...αN ) and N! , − 2r)!(d/2 + N − 2r)r 1 = r . 4 r!(N − 2r)!(2 − N − d/2)r

aN,r = bN,r

(A.45)

4r r!(N

(A.46)

Summation formulae: α

α

[(k1 )m (k2 )n ∗ gs ] ≡ k1α1 . . . k1αm k2 m+1 . . . k2 m+n gs, α1 ...αm+n

min{m,n}

=

j≥0, j+min{m,n} even

m!n! 2(m+n)/2−j ((m − j)/2)!((n − j)/2)!j!

×(k12 )(m−j)/2 (k22 )(n−j)/2 (k1 ·k2 )j ,

(A.47)

186

A Tables

[(k1 )m (k2 )n ∗ {[g]r [k3 ]m+n−2r }]

min{2r,m}

=

min{r1 ,2r−r1 }

r1 =max{0,2r−n} j≥0, j+r1 even

1 (m − r1 )!(n − 2r + r1 )!

m!n!

×

2r−j ((r1 − j)/2)!(r − (r1 + j)/2)!j! ×(k1 ·k2 )j (k1 ·k3 )m−r1 (k2 ·k3 )n−2r+r1

(k12 )(r1 −j)/2 (k22 )r−(r1 +j)/2 .

(A.48)

In particular, [(k1 )m (k2 )n ∗ {[g]r [k3 ]N −2r }] n = (k2 ·k3 )N −2r [(k1 )m (k2 )n−N +2r ∗ gs ] , N − 2r

(A.49)

where k1 ·k3 = 0, N = m + n, and [pb1 q b2 ∗ {[g]r [l]n−2r }] b1 !b2 ! = 2r

min{r,[b1 /2]}

r1 =max{0,r−[b2 /2]}

(p·l)b1 −2r1 (q·l)b2 −2r+2r1 (p2 )r1 (q 2 )r−r1 , r1 !(r − r1 )!(b1 − 2r1 )!(b2 − 2r + 2r1 )! (A.50)

where p·q = 0 and n = b1 + b2 . [(k1 )m (k2 )n (k3 )l−m−n ∗ gs ] =

a(l, m, n, j1 , j2 , j3 )

j1 ≥0, j1 +m even j2 ≥0, j2 +n even j3 ≥0, j3 +l−m−n even

×(k12 )(m−j1 )/2 (k22 )(n−j2 )/2 (k32 )(l−m−n−j3 )/2 ×(k1 ·k2 )(j1 +j2 −j3 )/2 (k1 ·k3 )(j1 −j2 +j3 )/2 (k2 ·k3 )(−j1 +j2 +j3 )/2 , 2(j1 +j2 +j3 −l)/2 m!n!(l − m − n)! ((m − j1 )/2)!((n − j2 )/2)!((l − m − n − j3 )/2)! θ(j1 + j2 − j3 )θ(j1 − j2 + j3 )θ(−j1 + j2 + j3 ) , (A.51) × ((j1 + j2 − j3 )/2)!((j1 − j2 + j3 )/2)!((−j1 + j2 + j3 )/2)!

a(l, m, n, j1 , j2 , j3 ) =

where θ(n) = 1 for n ≥ 0 and θ(n) = 0 otherwise.

B Some Special Functions

The Gauss hypergeometric function [3] is deﬁned by the series ∞ (a)n (b)n n z , F (a, b; c; z) = 2 1 (c)n n! n=0

(B.1)

where (x)n = Γ (x + n)/Γ (x)

(B.2)

is the Pochhammer symbol. This power series has the radius of convergence equal to one. It is analytically continued to the whole complex plane, with a cut, usually chosen as [1, ∞). The analytic continuation to values of z where |z| > 1 is given by Γ (c)Γ (b − a) 1 −a (−z) 2 F1 a, 1 − c + a; 1 − b + a; 2 F1 (a, b; c; z) = Γ (b)Γ (c − a) z Γ (c)Γ (a − b) 1 (−z)−b 2 F1 b, 1 − c + b; 1 − a + b; + . (B.3) Γ (a)Γ (c − b) z Another formula for the analytic continuation is z −a F (a, b; c; z) = (1 − z) F a, c − b; c; . (B.4) 2 1 2 1 z−1 This is a useful parametric representation: 1 Γ (c) F (a, b; c; z) = dx xb−1 (1 − x)c−b−1 (1 − zx)−a . (B.5) 2 1 Γ (b)Γ (c − b) 0 The polylogarithms [6] and generalized (Nielsen) polylogarithms [2, 5] are deﬁned by ∞ zn (B.6) Lia (z) = na n=1 1 a−1 (−1)a ln t dt (B.7) = (a − 1)! 0 t − 1/z and (−1)a+b−1 1 lna−1 t lnb (1 − zt) dt , (B.8) Sa,b (z) = (a − 1)!b! 0 t where a and b are positive integers. V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 187–190 (2004) c Springer-Verlag Berlin Heidelberg 2004

188

B Some Special Functions

The harmonic polylogarithms [8] Ha1 ,a2 ,...,an (x) ≡ H(a1 , a2 , . . . , an ; x), (HPL) with ai = 1, 0, −1, are deﬁned recursively by x Ha1 ,a2 ,...,an (x) = dtf (a1 ; t)H(a2 , . . . , an ; t) , (B.9) 0

where 1 1 , f0 (x) = , 1∓x x H±1 (x) = ∓ ln(1 ∓ x), H0 (x) = ln x , f±1 (x) =

(B.10) (B.11)

and at least one of the indices ai is non-zero. For all ai = 0, one has 1 n ln x . (B.12) n! Up to level 4, HPL with the indices 0 and 1 can be expressed in terms of usual polylogarithms [8]: H0,0,...,0 (x) =

H0 (x) = ln x , H1 (x) = − ln(1 − x) , 1 H0,0 (x) = ln2 x , 2! H0,1 (x) = Li2 (x) , H1,0 (x) = − ln x ln(1 − x) − Li2 (x) , 1 H1,1 (x) = ln2 (1 − x) , 2! 1 H0,0,0 (x) = ln3 x , 3! H0,0,1 (x) = Li3 (x) , H0,1,0 (x) = −2Li3 (x) + ln x Li2 (x) , H0,1,1 (x) = S1,2 (x) , 1 H1,0,0 (x) = − ln(1 − x) ln2 x − ln x Li2 (x) + Li3 (x) , 2 H1,0,1 (x) = −2S1,2 (x) − ln(1 − x)Li2 (x) , 1 H1,1,0 (x) = S1,2 (x) + ln(1 − x) Li2 (x) + ln x ln2 (1 − x) , 2 1 3 H1,1,1 (x) = − ln (1 − x) , 3! 1 4 H0,0,0,0 (x) = ln x , 4! H0,0,0,1 (x) = Li4 (x) , H0,0,1,0 (x) = ln x Li3 (x) − 3Li4 (x) , H0,0,1,1 (x) = S2,2 (x) , 1 H0,1,0,0 (x) = ln2 x Li2 (x) − 2 ln x Li3 (x) + 3Li4 (x) , 2

(B.13) (B.14) (B.15) (B.16) (B.17) (B.18) (B.19) (B.20) (B.21) (B.22) (B.23) (B.24) (B.25) (B.26) (B.27) (B.28) (B.29) (B.30) (B.31)

References

189

1 2 H0,1,0,1 (x) = −2S2,2 (x) + Li2 (x) , (B.32) 2 1 2 H0,1,1,0 (x) = ln x S1,2 (x) − Li2 (x) , (B.33) 2 (B.34) H0,1,1,1 (x) = S1,3 (x) , 1 3 1 2 H1,0,0,0 (x) = − ln x ln(1 − x) − ln x Li2 (x) 6 2 (B.35) + ln x Li3 (x) − Li4 (x) , 1 2 (B.36) H1,0,0,1 (x) = − Li2 (x) − ln(1 − x)Li3 (x) , 2 H1,0,1,0 (x) = 2 ln(1 − x)Li3 (x) − ln x ln(1 − x)Li2 (x) − 2 ln x S1,2 (x) 1 2 + Li2 (x) + 2S2,2 (x) , (B.37) 2 (B.38) H1,0,1,1 (x) = − ln(1 − x)S1,2 (x) − 3S1,3 (x) , 1 H1,1,0,0 (x) = ln2 x ln2 (1 − x) − ln(1 − x)Li3 (x) 4 (B.39) + ln x ln(1 − x)Li2 (x) + ln x S1,2 (x) − S2,2 (x) , 1 2 H1,1,0,1 (x) = ln (1 − x)Li2 (x) + 2 ln(1 − x)S1,2 (x) + 3S1,3 (x) , (B.40) 2 1 1 H1,1,1,0 (x) = − ln x ln3 (1 − x) − ln2 (1 − x) Li2 (x) 6 2 (B.41) − ln(1 − x)S1,2 (x) − S1,3 (x) , 1 H1,1,1,1 (x) = ln4 (1 − x) . (B.42) 4! Analytic properties of HPL (and 2dHPL) which allow to continue them to any domain are described in [18]. The HPL are partial cases of the socalled Z- and S-sums which are deﬁned similarly to the nested sums (see Appendix C) but with the factor xj – see, e.g., [7]. The set of Z- or S-sums can be equipped with an operation of multiplication in such a way that they (as well as HPL) form a Hopf algebra – see, e.g., [1, 8].

References 1. J. Bl¨ umlein, Comput. Phys. Commun. 159 (2004) 19. 189 2. A. Devoto and D.W. Duke, Riv. Nuovo Cim. 7, No. 6 (1984) 1. 187 3. A. Erd´elyi (ed.), Higher Transcendental Functions, Vols. 1 and 2 (McGraw-Hill, New York, 1954). 187 4. T. Gehrmann and E. Remiddi, Nucl. Phys. B 640 (2002) 379. 5. K.S. K¨ olbig, J.A. Mignaco and E. Remiddi, BIT 10 (1970) 38; K.S. K¨ olbig, Math. Comp. 39 (1982) 647. 187

190

B Some Special Functions

6. L. Lewin, Polylogarithms and Associated Functions (North-Holland, Amsterdam, 1981). 187 7. S. Moch, P. Uwer and S. Weinzierl, J. Math. Phys. 43 (2002) 3363. 189 8. E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725. 188, 189

C Summation Formulae

Nested sums are deﬁned as follows [17]: Si (n) = Sikl (n) =

n n 1 Sk (j) , S (n) = , ik i j ji j=1 j=1

n Skl (j) j=1

ji

, Siklm (n) =

n Sklm (j) j=1

ji

,

(C.1)

(C.2)

etc. Properties and algorithms for the nested sums (also for negative indices which are deﬁned with (−1)j ) are presented in [17]. In particular, for positive indices, we have Sj,k (n) + Sk,j (n) = Sj (n)Sk (n) + Sj+k (n) .

(C.3)

The nested sums are closely connected with multiple ζ-values – see, e.g., [1, 2, 11, 19] and the review [7]. The sums with one index are connected with the ψ function (the logarithmical derivative of the gamma function) as ψ(n) = S1 (n − 1) − γE , ψ (n) = (−1)k k! (Sk+1 (n − 1) − ζ(k + 1)) , k = 1, 2 , . . . , (k)

(C.4) (C.5)

where ζ(z) is the Riemann zeta function ζ(z) =

∞ 1 . nz n=1

(C.6)

All the summation formulae of this Appendix, apart from the inverse binomial series1 , are implemented in the package called SUMMER [17] which is written in FORM [16]. This powerful package was successfully used in nontrivial calculations – see, e.g., [12, 13, 14]. There is also another package operating with the nested sums [18]. Nested sums are closely connected with expansions of hypergeometric series in its parameters – see, e.g., [3, 4, 11]. For example, the expansion of the Gauss hypergeometric function 2 F1 (1 + a1 ε, 1 + a2 ε; 3/2 + bε; z) is connected with inverse binomial series [3]. 1

The authors of SUMMER are planning to include the inverse binomial series into this package. V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 191–205 (2004) c Springer-Verlag Berlin Heidelberg 2004

192

C Summation Formulae

C.1 Some Number Series These are series up to level 6 with at least 1/n2 dependence:

∞

∞ 1 π2 , = n2 6 n=1

(C.7)

∞ 1 = ζ(3) , n3 n=1

(C.8)

1 = ζ(3) , n2

(C.9)

S1 (n − 1)

n=1

∞ 1 π4 , = n4 90 n=1 ∞

S1 (n − 1)

1 π4 , = n3 360

(C.11)

S1 (n − 1)2

1 11π 4 , = 2 n 360

(C.12)

1 π4 , = 2 n 120

(C.13)

∞ 1 = ζ(5) , 5 n n=1

(C.14)

n=1 ∞ n=1 ∞

S2 (n − 1)

n=1

∞

S1 (n − 1)

1 π 2 ζ(3) , = 2ζ(5) − n4 6

(C.15)

S2 (n − 1)

1 π 2 ζ(3) 11ζ(5) − , = 3 n 2 2

(C.16)

S1 (n − 1)2

1 π 2 ζ(3) 3ζ(5) − , = 3 n 6 2

(C.17)

S3 (n − 1)

1 9ζ(5) π 2 ζ(3) − , = n2 2 3

(C.18)

S1 (n − 1)3

1 π 2 ζ(3) 15ζ(5) + , = n2 6 2

(C.19)

1 7ζ(5) π 2 ζ(3) − , = 2 n 2 6

(C.20)

1 2π 2 ζ(3) , = 9ζ(5) − 2 n 3

(C.21)

n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)S2 (n − 1)

n=1 ∞ n=1

(C.10)

S12 (n − 1)

C.1 Some Number Series ∞ 1 π6 , = n6 945 n=1 ∞

ζ(3)2 1 π6 − , = 5 n 1260 2

(C.23)

S2 (n − 1)

1 π6 + ζ(3)2 , = −4 4 n 2835

(C.24)

1 37π 6 − ζ(3)2 , = n4 22680

(C.25)

S3 (n − 1)

ζ(3)2 1 π6 + , = − n3 1890 2

(C.26)

S4 (n − 1)

1 5π 6 − ζ(3)2 , = 2 n 2268

(C.27)

S13 (n − 1)

1 61π 6 , = n2 45360

(C.28)

S2 (n − 1)2

1 59π 6 − ζ(3)2 , = n2 22680

(C.29)

S1 (n − 1)3

1 11π 6 + 2ζ(3)2 , = − n3 5040

(C.30)

1 121π 6 + 2ζ(3)2 , =− 3 n 45360

(C.31)

1 41π 6 − ζ(3)2 , = n3 22680

(C.32)

3ζ(3)2 1 167π 6 − , = n2 45360 2

(C.33)

1 23π 6 − ζ(3)2 , = 2 n 3780

(C.34)

1 859π 6 + 3ζ(3)2 , = 2 n 22680

(C.35)

1 17π 6 − ζ(3)2 , = n2 4536

(C.36)

1 313π 6 − 2ζ(3)2 . = n2 45360

(C.37)

n=1 ∞

S1 (n − 1)2

n=1 ∞ n=1 ∞ n=1 ∞

n=1 ∞ n=1 ∞

S1 (n − 1)S2 (n − 1)

n=1 ∞

S12 (n − 1)

n=1 ∞

S1 (n − 1)S3 (n − 1)

n=1 ∞

S1 (n − 1)2 S2 (n − 1)

n=1 ∞

S1 (n − 1)4

n=1 ∞

S112 (n − 1)

n=1 ∞ n=1

(C.22)

S1 (n − 1)

n=1 ∞

n=1 ∞

193

S1 (n − 1)S12 (n − 1)

194

C Summation Formulae

Series up to level 6 with the factor 1/n where the convergence is provided by other factors: ∞

ψ (n + 1)

1 = ζ(3) , n

(C.38)

ψ (n + 1)S1 (n)

7π 4 1 = , n 360

(C.39)

π4 1 =− , n 180

(C.40)

π 2 ζ(3) 1 = , n 3

(C.41)

1 5π 2 ζ(3) = − 9ζ(5) , n 6

(C.42)

2π 2 ζ(3) 1 =− + 7ζ(5) , n 3

(C.43)

ψ (n + 1)

1 = −π 2 ζ(3) + 12ζ(5) , n

(C.44)

ψ (n + 1)

2π 6 1 =− + 12ζ(3)2 , n 105

(C.45)

ψ (n + 1)S1 (n)

π6 1 = , n 1512

(C.46)

ψ (n + 1)S1 (n)2

1 π6 = − 8ζ(3)2 , n 90

(C.47)

ψ (n + 1)2 S1 (n)

π6 1 =− + 2ζ(3)2 , n 432

(C.48)

ψ (n + 1)S1 (n)3

269π 6 1 = , n 22680

(C.49)

61π 6 1 = − 2ζ(3)2 . n 22680

(C.50)

n=1 ∞ n=1 ∞

ψ (n + 1)

n=1 ∞

ψ (n + 1)S1 (n)2

n=1 ∞

ψ (n + 1)2

n=1 ∞

ψ (n + 1)S1 (n)

n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

ψ (n + 1)ψ (n + 1)

n=1

Series of level 7 with at least 1/n2 dependence: ∞ 1 = ζ(7) , 7 n n=1 ∞ n=1

S1 (n − 1)

1 π 2 ζ(5) π 4 ζ(3) − , = 3ζ(7) − n6 6 90

(C.51) (C.52)

C.1 Some Number Series ∞

1 5π 2 ζ(5) π 4 ζ(3) + , = −11ζ(7) + n5 6 45

(C.53)

1 π 2 ζ(5) π 4 ζ(3) − , = −ζ(7) + 5 n 6 180

(C.54)

1 5π 2 ζ(5) , = 17ζ(7) − 4 n 3

(C.55)

1 119ζ(7) π 2 ζ(5) 11π 4 ζ(3) + − , = n4 16 3 120

(C.56)

1 61ζ(7) π 2 ζ(5) π 4 ζ(3) − + = , n4 16 3

(C.57)

1 141ζ(7) 5π 2 ζ(5) π 4 ζ(3) − − , = 4 n 8 4 24

(C.58)

S4 (n − 1)

1 5π 2 ζ(5) π 4 ζ(3) + , = −18ζ(7) + n3 3 90

(C.59)

S13 (n − 1)

1 73ζ(7) 5π 2 ζ(5) π 4 ζ(3) + + , = − n3 4 3 72

(C.60)

1 85ζ(7) 11π 2 ζ(5) π 4 ζ(3) + + , = − n3 8 12 72

(C.61)

S2 (n − 1)2

1 13ζ(7) 5π 2 ζ(5) 11π 4 ζ(3) − + , = 3 n 8 6 180

(C.62)

S1 (n − 1)S12 (n − 1)

1 113ζ(7) 7π 2 ζ(5) π 4 ζ(3) + + , = − n3 16 12 72

(C.63)

S1 (n − 1)2 S2 (n − 1)

1 77ζ(7) π 2 ζ(5) 7π 4 ζ(3) − + , = − n3 8 3 60

(C.64)

S2 (n − 1)

n=1 ∞

S1 (n − 1)2

n=1 ∞

S3 (n − 1)

n=1 ∞

S1 (n − 1)3

n=1 ∞

S1 (n − 1)S2 (n − 1)

n=1 ∞

S12 (n − 1)

n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)S3 (n − 1)

n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)4

n=1 ∞

1 109ζ(7) 5π 2 ζ(5) 37π 4 ζ(3) − + , (C.65) =− 3 n 8 6 180

S112 (n − 1)

1 61ζ(7) 5π 2 ζ(5) π 4 ζ(3) + + , =− 3 n 4 4 40

(C.66)

S1 (n − 1)S4 (n − 1)

1 173ζ(7) 3π 2 ζ(5) π 4 ζ(3) − − , = n2 16 4 60

(C.67)

S1 (n − 1)S13 (n − 1)

1 61ζ(7) 3π 2 ζ(5) π 4 ζ(3) − + , = n2 4 2 36

(C.68)

n=1 ∞ n=1 ∞ n=1

195

196

C Summation Formulae ∞

S1 (n − 1)2 S3 (n − 1)

1 301ζ(7) 3π 2 ζ(5) π 4 ζ(3) − − , = n2 16 4 15

(C.69)

S1 (n − 1)S2 (n − 1)2

1 77ζ(7) 13π 2 ζ(5) π 4 ζ(3) + − , =− 2 n 16 12 30

(C.70)

S1 (n − 1)2 S12 (n − 1)

1 423ζ(7) π 2 ζ(5) 37π 4 ζ(3) − − , = 2 n 16 6 360

(C.71)

1 307ζ(7) 5π 2 ζ(5) 13π 4 ζ(3) + − , = n2 16 12 180

(C.72)

n=1 ∞ n=1 ∞ n=1 ∞

S1 (n − 1)3 S2 (n − 1)

n=1 ∞

S1 (n − 1)5

n=1

1 1855ζ(7) 19π 2 ζ(5) + = n2 16 4

11π 4 ζ(3) , 30 ∞ 1 73ζ(7) 3π 2 ζ(5) π 4 ζ(3) − − , S1 (n − 1)S112 (n − 1) 2 = n 4 4 30 n=1 +

∞

S5 (n − 1)

n=1 ∞

S14 (n − 1)

n=1 ∞

S2 (n − 1)S3 (n − 1)

n=1 ∞

S23 (n − 1)

n=1 ∞

S2 (n − 1)S12 (n − 1)

n=1 ∞

(C.74)

1 2π 2 ζ(5) π 4 ζ(3) − , = 10ζ(7) − n2 3 45

(C.75)

1 141ζ(7) 19π 2 ζ(5) π 4 ζ(3) − − , = 2 n 8 12 360

(C.76)

1 19ζ(7) 5π 2 ζ(5) 7π 4 ζ(3) + − , = 2 n 16 12 180

(C.77)

1 131ζ(7) 4π 2 ζ(5) 7π 4 ζ(3) + − , = − n2 16 3 180

(C.78)

1 141ζ(7) 5π 2 ζ(5) 19π 4 ζ(3) + − , (C.79) = − n2 16 3 360

S113 (n − 1)

1 113ζ(7) π 2 ζ(5) − , = 2 n 16 2

(C.80)

S212 (n − 1)

1 169ζ(7) π 2 ζ(5) 7π 4 ζ(3) − − , = n2 16 2 180

(C.81)

S1112 (n − 1)

1 141ζ(7) 7π 4 ζ(3) 2 − π . = ζ(5) − n2 8 180

(C.82)

n=1 ∞ n=1 ∞ n=1

(C.73)

C.2 Power Series of Levels 3 and 4 in Terms of Polylogarithms

197

C.2 Power Series of Levels 3 and 4 in Terms of Polylogarithms The formulae of this section can be found in [6]. ∞

S2 (n − 1)

n=1 ∞ n=1 ∞

zn = −2S1,2 (z) − ln(1 − z)Li2 (z) , n

S1 (n − 1)2 S1 (n − 1)

n=1 ∞

(C.83)

zn 1 = −2S1,2 (z) − ln(1 − z)Li2 (z) − ln3 (1 − z) , (C.84) n 3

zn = S1,2 (z) , n2

(C.85)

zn = Li3 (z) , n3 n=1 ∞ n=1 ∞

S3 (n − 1)

(C.86)

1 zn 2 = − Li2 (z) − ln(1 − z)Li3 (z) , n 2

S12 (n − 1)

n=1

(C.87)

zn 1 2 = 3S1,3 (z) − ln(1 − z)Li3 (z) − Li2 (z) n 2

1 + ln2 (1 − z)Li2 (z) + 2 ln(1 − z)S1,2 (z) , (C.88) 2 ∞ 1 zn 2 = − Li2 (z) + ln(1 − z)(S1,2 (z) − Li3 (z)) S1 (n − 1)S2 (n − 1) n 2 n=1 1 + ln2 (1 − z)Li2 (z) , 2 ∞ n 1 z 3 2 = − Li2 (z) + ln2 (1 − z)Li2 (z) S1 (n − 1)3 n 2 2 n=1 + ln(1 − z)(3S1,2 (z) − Li3 (z)) + ∞ n=1 ∞ n=1 ∞ n=1 ∞

S2 (n − 1)

zn 1 2 = −2S2,2 (z) + Li2 (z) , n2 2

S1 (n − 1)2 S1 (n − 1)

1 4 ln (1 − z) , 4

zn 1 2 = 2S1,3 (z) − 2S2,2 (z) + Li2 (z) , n2 2

zn = S2,2 (z) , n3

zn = Li4 (z) . n4 n=1

(C.89)

(C.90) (C.91) (C.92) (C.93) (C.94)

198

C Summation Formulae

C.3 Inverse Binomial Power Series up to Level 4 The formulae of this section (as well as other similar formulae) can be found in [3]. See a table of formulae for the corresponding number series in [8]. Let √ √ 4 − z − −z y=√ . √ 4 − z + −z Then ∞

1−y 1 zn 2n = ln y, n 1+y n n=1 ∞

1 zn 1 2n 2 = − ln2 y, n 2 n n=1

(C.95) (C.96)

∞

1 zn 2n 3 = 2Li3 (y) − 2 ln y Li2 (y) − ln2 y ln(1 − y) n n n=1 ∞

1 + ln3 y − 2ζ(3) , 6

(C.97)

1 zn 2n 4 = 4S2,2 (y) − 4Li4 (y) − 4S1,2 (y) ln y n n n=1 +4Li3 (y) ln(1 − y) + 2Li3 (y) ln y − 4Li2 (y) ln y ln(1 − y) 1 1 4 ln y − ln2 y ln2 (1 − y) + ln3 y ln(1 − y) − 3 24 −4 ln(1 − y)ζ(3) + 2 ln y ζ(3) + 3ζ(4) , (C.98) ∞

1 zn 1−y 2n S1 (n − 1) = n 1+y n n=1

1 2 × −2Li2 (−y) − 2 ln y ln(1 + y) + ln y − ζ(2) , 2 ∞ 1 zn 1−y 2n S1 (n − 1)2 = 8S1,2 (−y) − 4Li3 (−y) n 1 +y n n=1

(C.99)

+8Li2 (−y) ln(1 + y) + 4 ln2 (1 + y) ln y − 2 ln(1 + y) ln2 y

1 3 + ln y + 4ζ(2) ln(1 + y) − 2ζ(2) ln y − 4ζ(3) , (C.100) 6 ∞

1 zn 1−y 2n S2 (n − 1) = − ln3 y , n 6(1 + y) n n=1 ∞

1 zn 1 4 2n 2 S2 (n − 1) = ln y , n 24 n n=1

1 zn 1 − y 1 4 2n S3 (n − 1) = ln y + 6Li4 (y) + ln2 y Li2 (y) n 1 + y 24 n n=1 ∞

(C.101) (C.102)

C.3 Inverse Binomial Power Series up to Level 4

199

−2ζ(3) ln y − 4 ln y Li3 (y) − 6ζ(4) , 1 zn 1 − y 1 3 1 4 2n S1 (n − 1)S2 (n − 1) = ln y ln(1 + y) − ln y n 1 + y 3 24 n n=1 ∞

1 + ζ(2) ln2 y + ln2 y Li2 (−y) + ln2 y Li2 (y) + ζ(3) ln y − 4 ln y Li3 (−y) 2 −4 ln y Li3 (y) + ζ(4) + 8Li4 (−y) + 6Li4 (y) ,

(C.103)

∞

1 zn 2n 2 S1 (n − 1) = 4Li3 (−y) − 2Li2 (−y) ln y n n n=1 ∞

1 − ln3 y + 3ζ(3) + ζ(2) ln y , 6

(C.104)

1 zn 2n 2 S1 (n − 1)2 = −8S1,2 (−y) ln y + 4Li3 (−y) ln y n n n=1 2

−2Li2 (−y) ln2 y + 4Li2 (−y) −

1 4 ln y + 4ζ(2)Li2 (−y) 24

5 +ζ(2) ln2 y + 4ζ(3) ln y + ζ(4) , 2 ∞ 1 zn 2n 3 S1 (n − 1) = 4H−1,0,0,1 (−y) + S2,2 y 2 n n n=1

(C.105)

−4S2,2 (y) − 4S2,2 (−y) − 6Li4 (−y) − 2Li4 (y) + 4S1,2 (−y) ln y +4S1,2 (y) ln y − 2S1,2 y 2 ln y + 4Li3 (−y) ln(1 − y) +2Li3 (−y) ln y + 2Li3 (y) ln y − Li2 (y) ln2 y 1 1 4 ln y −4Li2 (−y) ln y ln(1 − y) − ln3 y ln(1 − y) + 3 24 1 +2ζ(2)Li2 (y) − ζ(2) ln2 y + 2ζ(2) ln y ln(1 − y) 2 +6ζ(3) ln(1 − y) − 3ζ(3) ln y − 4ζ(4) , (C.106) ∞ n 1 z 1−y 2n S1 (n − 1)3 = −48S1,2 (−y) ln(1 + y) − 48S1,3 (−y) n 1 +y n n=1 +24S2,2 (−y) − 12ζ(2) ln2 (1 + y) − 24 ln2 (1 + y)Li2 (−y) +24ζ(3) ln(1 + y) + 24 ln(1 + y)Li3 (−y) − 8 ln y ln3 (1 + y) +12ζ(2) ln y ln(1 + y) + 6 ln2 y ln2 (1 + y) − ln3 y ln(1 + y) 3 1 + ln4 y − ζ(2) ln2 y + 3 ln2 y Li2 (−y) 24 2 + ln2 y Li2 (y) − 5ζ(3) ln y − 12 ln y Li3 (−y) − 4 ln y Li3 (y) 3 (C.107) + ζ(4) + 12Li4 (−y) + 6Li4 (y) . 2

200

C Summation Formulae

C.4 Power Series of Levels 5 and 6 in Terms of HPL ∞ zn = H0,0,0,0,1 (z) , n5 n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

(C.108)

S1 (n − 1)

zn = H0,0,0,1,1 (z) , n4

(C.109)

S2 (n − 1)

zn = H0,0,1,0,1 (z) , n3

(C.110)

S1 (n − 1)2 S3 (n − 1)

zn = H0,0,1,0,1 (z) + 2H0,0,1,1,1 (z) , n3

zn = H0,1,0,0,1 (z) , n2

S1 (n − 1)3

n=1

S1 (n − 1)S2 (n − 1)

n=1

n=1 ∞ n=1 ∞ n=1 ∞

S12 (n − 1) S4 (n − 1)

z = H0,1,0,0,1 (z) + H0,1,1,0,1 (z) , n2

zn = H1,0,0,0,1 (z) + H1,1,0,0,1 (z) , n

S1 (n − 1)S3 (n − 1)

n=1

n=1 ∞ n=1

S2 (n − 1)2

(C.115) (C.116) (C.117)

zn = H1,0,0,0,1 (z) + H1,0,0,1,1 (z) n +H1,1,0,0,1 (z) ,

∞

(C.114)

n

zn = H1,0,0,0,1 (z) , n

S13 (n − 1)

(C.113)

zn = H0,1,0,0,1 (z) + H0,1,0,1,1 (z) n2 +H0,1,1,0,1 (z) ,

∞

(C.112)

zn = H0,1,0,0,1 (z) + 3H0,1,0,1,1 (z) n2 +3H0,1,1,0,1 (z) + 6H0,1,1,1,1 (z) ,

∞

(C.111)

(C.118)

n

z = H1,0,0,0,1 (z) + 2H1,0,1,0,1 (z) , n

S1 (n − 1)S12 (n − 1)

(C.119)

zn = H1,0,0,0,1 (z) + H1,0,0,1,1 (z) n

+H1,0,1,0,1 (z) + 2H1,1,0,0,1 (z) + H1,1,0,1,1 (z) + 2H1,1,1,0,1 (z) , (C.120) ∞ zn = H1,0,0,0,1 (z) + 2H1,0,0,1,1 (z) S1 (n − 1)2 S2 (n − 1) n n=1

C.4 Power Series of Levels 5 and 6 in Terms of HPL

201

+2H1,0,1,0,1 (z) + 2H1,0,1,1,1 (z) + 2H1,1,0,0,1 (z) +2H1,1,0,1,1 (z) + 2H1,1,1,0,1 (z) , ∞

S1 (n − 1)4

n=1

(C.121)

n

z = H1,0,0,0,1 (z) + 4H1,0,0,1,1 (z) + 6H1,0,1,0,1 (z) n +12H1,0,1,1,1 (z) + 4H1,1,0,0,1 (z) + 12H1,1,0,1,1 (z) +12H1,1,1,0,1 (z) + 24H1,1,1,1,1 (z) , (C.122)

∞

S112 (n − 1)

n=1

zn = H1,0,0,0,1 (z) + H1,0,1,0,1 (z) + H1,1,0,0,1 (z) n +H1,1,1,0,1 (z) ,

∞

zn = H0,0,0,0,0,1 (z) , n6 n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞ n=1 ∞

(C.124)

S1 (n − 1)

zn = H0,0,0,0,1,1 (z) , n5

(C.125)

S2 (n − 1)

zn = H0,0,0,1,0,1 (z) , n4

(C.126)

S1 (n − 1)2 S3 (n − 1)

zn = H0,0,0,1,0,1 (z) + 2H0,0,0,1,1,1 (z) , n4

zn = H0,0,1,0,0,1 (z) , n3

S1 (n − 1)3

n=1

S1 (n − 1)S2 (n − 1)

n=1

n=1 ∞ n=1 ∞ n=1 ∞ n=1

S12 (n − 1) S4 (n − 1)

zn = H0,0,1,0,0,1 (z) + H0,0,1,1,0,1 (z) , n3

zn = H0,1,0,0,0,1 (z) + H0,1,1,0,0,1 (z) , n2

S1 (n − 1)S3 (n − 1)

(C.129)

z = H0,0,1,0,0,1 (z) + H0,0,1,0,1,1 (z) n3

zn = H0,1,0,0,0,1 (z) , n2

S13 (n − 1)

(C.128)

n

+H0,0,1,1,0,1 (z) , ∞

(C.127)

zn = H0,0,1,0,0,1 (z) + 3H0,0,1,0,1,1 (z) n3 +3H0,0,1,1,0,1 (z) + 6H0,0,1,1,1,1 (z) ,

∞

(C.123)

zn = H0,1,0,0,0,1 (z) + H0,1,0,0,1,1 (z) n2

(C.130) (C.131) (C.132) (C.133)

202

C Summation Formulae

+H0,1,1,0,0,1 (z) , ∞ n=1 ∞

S2 (n − 1)2

zn = H0,1,0,0,0,1 (z) + 2H0,1,0,1,0,1 (z) , n2

S1 (n − 1)S12 (n − 1)

n=1

(C.134) (C.135)

zn = H0,1,0,0,0,1 (z) + H0,1,0,0,1,1 (z) n2 +H0,1,0,1,0,1 (z) + 2H0,1,1,0,0,1 (z) +H0,1,1,0,1,1 (z) + 2H0,1,1,1,0,1 (z) , (C.136)

∞

S1 (n − 1)2 S2 (n − 1)

n=1

n

z = H0,1,0,0,0,1 (z) + 2H0,1,0,0,1,1 (z) n2

+2H0,1,0,1,0,1 (z) + 2H0,1,0,1,1,1 (z) + 2H0,1,1,0,0,1 (z) +2H0,1,1,0,1,1 (z) + 2H0,1,1,1,0,1 (z) , ∞ n=1

S1 (n − 1)4

(C.137)

n

z = H0,1,0,0,0,1 (z) + 4H0,1,0,0,1,1 (z) n2

+6H0,1,0,1,0,1 (z) + 12H0,1,0,1,1,1 (z) + 4H0,1,1,0,0,1 (z) +12H0,1,1,0,1,1 (z) + 12H0,1,1,1,0,1 (z) + 24H0,1,1,1,1,1 (z) , (C.138) ∞ n=1

S112 (n − 1)

zn = H0,1,0,0,0,1 (z) + H0,1,0,1,0,1 (z) n2 +H0,1,1,0,0,1 (z) + H0,1,1,1,0,1 (z) ,

∞ n=1

S1 (n − 1)S4 (n − 1)

z = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n +H1,1,0,0,0,1 (z) ,

∞

(C.139)

n

S1 (n − 1)S13 (n − 1)

n=1

(C.140)

n

z = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n +H1,0,1,0,0,1 (z) + 2H1,1,0,0,0,1 (z) +H1,1,0,0,1,1 (z) + 2H1,1,1,0,0,1 (z) ,

∞

S1 (n − 1)2 S3 (n − 1)

n=1

(C.141)

n

z = H1,0,0,0,0,1 (z) + 2H1,0,0,0,1,1 (z) n

+H1,0,0,1,0,1 (z) + 2H1,0,0,1,1,1 (z) + H1,0,1,0,0,1 (z) +2H1,1,0,0,0,1 (z) + 2H1,1,0,0,1,1 (z) + 2H1,1,1,0,0,1 (z) , (C.142) ∞ n=1

S1 (n − 1)S2 (n − 1)2

zn = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n

+2H1,0,0,1,0,1 (z) + 2H1,0,1,0,0,1 (z) + 2H1,0,1,0,1,1 (z) +2H1,0,1,1,0,1 (z) + H1,1,0,0,0,1 (z) + 2H1,1,0,1,0,1 (z) ,

(C.143)

C.4 Power Series of Levels 5 and 6 in Terms of HPL ∞

S1 (n − 1)2 S12 (n − 1)

n=1

203

zn = H1,0,0,0,0,1 (z) + 2H1,0,0,0,1,1 (z) n

+2H1,0,0,1,0,1 (z) + 2H1,0,0,1,1,1 (z) + 3H1,0,1,0,0,1 (z) +2H1,0,1,0,1,1 (z) + 3H1,0,1,1,0,1 (z) + 3H1,1,0,0,0,1 (z) +4H1,1,0,0,1,1 (z) + 4H1,1,0,1,0,1 (z) + 2H1,1,0,1,1,1 (z) +6H1,1,1,0,0,1 (z) + 4H1,1,1,0,1,1 (z) + 6H1,1,1,1,0,1 (z) , (C.144) ∞

S1 (n − 1)3 S2 (n − 1)

n=1

zn = H1,0,0,0,0,1 (z) + 3H1,0,0,0,1,1 (z) n

+4H1,0,0,1,0,1 (z) + 6H1,0,0,1,1,1 (z) + 4H1,0,1,0,0,1 (z) +6H1,0,1,0,1,1 (z) + 6H1,0,1,1,0,1 (z) + 6H1,0,1,1,1,1 (z) +3H1,1,0,0,0,1 (z) + 6H1,1,0,0,1,1 (z) + 6H1,1,0,1,0,1 (z) +6H1,1,0,1,1,1 (z) + 6H1,1,1,0,0,1 (z) +6H1,1,1,0,1,1 (z) + 6H1,1,1,1,0,1 (z) , ∞

S1 (n − 1)5

n=1

(C.145)

n

z = H1,0,0,0,0,1 (z) + 5H1,0,0,0,1,1 (z) + 10H1,0,0,1,0,1 (z) n

+20H1,0,0,1,1,1 (z) + 10H1,0,1,0,0,1 (z) + 30H1,0,1,0,1,1 (z) +30H1,0,1,1,0,1 (z) + 60H1,0,1,1,1,1 (z) + 5H1,1,0,0,0,1 (z) +20H1,1,0,0,1,1 (z) + 30H1,1,0,1,0,1 (z) + 60H1,1,0,1,1,1 (z) +20H1,1,1,0,0,1 (z) + 60H1,1,1,0,1,1 (z) +60H1,1,1,1,0,1 (z) + 120H1,1,1,1,1,1 (z) , ∞

S1 (n − 1)S112 (n − 1)

n=1

(C.146)

n

z = H1,0,0,0,0,1 (z) + H1,0,0,0,1,1 (z) n

+H1,0,0,1,0,1 (z) + 2H1,0,1,0,0,1 (z) + H1,0,1,0,1,1 (z) +2H1,0,1,1,0,1 (z) + 2H1,1,0,0,0,1 (z) + H1,1,0,0,1,1 (z) +2H1,1,0,1,0,1 (z) + 3H1,1,1,0,0,1 (z) +H1,1,1,0,1,1 (z) + 3H1,1,1,1,0,1 (z) , ∞ n=1 ∞ n=1 ∞ n=1

S5 (n − 1)

z = H1,0,0,0,0,1 (z) , n

S14 (n − 1)

(C.147)

n

zn = H1,0,0,0,0,1 (z) + H1,1,0,0,0,1 (z) , n

S2 (n − 1)S3 (n − 1)

(C.148) (C.149)

zn = H1,0,0,0,0,1 (z) + H1,0,0,1,0,1 (z) n +H1,0,1,0,0,1 (z) ,

(C.150)

204

C Summation Formulae ∞ n=1 ∞

S23 (n − 1)

zn = H1,0,0,0,0,1 (z) + H1,0,1,0,0,1 (z) , n

S12 (n − 1)S2 (n − 1)

n=1

zn = H1,0,0,0,0,1 (z) + 2H1,0,0,1,0,1 (z) n +H1,0,1,0,0,1 (z) + H1,0,1,1,0,1 (z) +H1,1,0,0,0,1 (z) + 2H1,1,0,1,0,1 (z) ,

∞

S113 (n − 1)

n=1

S212 (n − 1)

n=1

z = H1,0,0,0,0,1 (z) + H1,0,1,0,0,1 (z) n z = H1,0,0,0,0,1 (z) + H1,0,0,1,0,1 (z) n

S1112 (n − 1)

n=1

(C.153)

n

+H1,0,1,0,0,1 (z) + H1,0,1,1,0,1 (z) , ∞

(C.152)

n

+H1,1,0,0,0,1 (z) + H1,1,1,0,0,1 (z) , ∞

(C.151)

(C.154)

n

z = H1,0,0,0,0,1 (z) + H1,0,0,1,0,1 (z) + H1,0,1,0,0,1 (z) n

+H1,0,1,1,0,1 (z) + H1,1,0,0,0,1 (z) + H1,1,0,1,0,1 (z) +H1,1,1,0,0,1 (z) + H1,1,1,1,0,1 (z) .

(C.155)

References 1. J. Bl¨ umlein, Comput. Phys. Commun. 159 (2004) 19. 191 2. J.M. Borwein, D.M. Bradley and D.J. Broadhurst, Electronic J. Combinatorics, 4(2) (1997) R5; J.M. Borwein, D.M. Bradley, D.J. Broadhurst and P. Lisonˇek, Electronic J. Combinatorics, 5(1) (1998) R38; Trans. Amer. Math. Soc. 355 (2001) 907. 191 3. A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 699 (2004) 3. 191, 198 4. A.I. Davydychev and M.Yu. Kalmykov, Nucl. Phys. B 605 (2001) 266; Phys. Rev. D 61 (2000) 087701. 191 5. A. Devoto and D.W. Duke, Riv. Nuovo Cim. 7, No. 6 (1984) 1. 6. J. Fleischer, A.V. Kotikov and O.L. Veretin, Nucl. Phys. B 547 (1999) 343. 197 7. A.G. Grozin, Int. J. Mod. Phys. A 19 (2004) 473. 191 8. M.Yu. Kalmykov and O. Veretin, Phys. Lett. B 483 (2000) 315. 198 9. K.S. K¨ olbig, J.A. Mignaco and E. Remiddi, BIT 10 (1970) 38; K.S. K¨ olbig, Math. Comp. 39 (1982) 647. 10. L. Lewin, Polylogarithms and Associated Functions (North-Holland, Amsterdam, 1981). 11. S. Moch, P. Uwer and S. Weinzierl, J. Math. Phys. 43 (2002) 3363. 191 12. S. Moch, P. Uwer and S. Weinzierl, Phys. Rev. D 66 (2002) 114001. 191 13. S. Moch and J. A. M. Vermaseren, Nucl. Phys. B 573 (2000) 853. 191 14. S. Moch, J. A. M. Vermaseren and A. Vogt, Nucl. Phys. B 688 (2004) 101; A. Vogt, S. Moch and J.A.M. Vermaseren, Nucl. Phys. B 691 (2004) 129. 191 15. E. Remiddi and J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725.

References

205

16. J.A.M. Vermaseren, Symbolic Manipulation with FORM (CAN, Amsterdam, 1991). 191 17. J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037. 191 18. S. Weinzierl, Comput. Phys. Commun. 145 (2002) 357. 191 19. S. Weinzierl, J. Math. Phys. 45 (2004) 2656. 191

D Table of MB Integrals

D.1 MB Integrals with Four Gamma Functions This is the ﬁrst Barnes lemma: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ Γ (λ1 + λ3 )Γ (λ1 + λ4 )Γ (λ2 + λ3 )Γ (λ2 + λ4 ) = . Γ (λ1 + λ2 + λ3 + λ4 )

(D.1)

Results for integrals with ψ(λ1 + z), . . . are obtained from (D.1) by differentiating with respect to λ1 , . . .. Second derivatives give, in a similar way, results for integrals with products of two diﬀerent functions ψ(λi ± z) and with the combinations ψ (λi ± z) + ψ(λi ± z)2 . Various corollaries can be derived from (D.1). For example, +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z) 2πi −i∞ = Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ(λ1 − λ2 ) − ψ(λ1 + λ3 )] , (D.2) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z) = Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ(λ2 + λ3 ) − ψ(λ1 + λ3 )] .

(D.3)

The asterisk is used to indicate that the ﬁrst pole of the corresponding gamma function is of the opposite nature, i.e. the ﬁrst pole of Γ (λ2 + z) in (D.2) is considered right and the ﬁrst pole of Γ (−λ2 − z) in (D.3) is considered left. These are four formulae with the psi function with the same condition as in (D.2): +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(λ1 + z) 2πi −i∞

= Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ1 − λ2 )2 − ψ(λ1 − λ2 )ψ(λ1 + λ3 ) +ψ (λ1 − λ2 ) − ψ (λ1 + λ3 )] , (D.4)

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 207–219 (2004) c Springer-Verlag Berlin Heidelberg 2004

208

D Table of MB Integrals

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(λ2 + z)

1 = − Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ1 − λ2 )2 − ψ(λ1 + λ3 )2 2 +2ψ(λ1 − λ2 )(γE − ψ(λ2 + λ3 )) − 2ψ(λ1 + λ3 )(γE − ψ(λ2 + λ3 )) +ψ (λ1 − λ2 ) + ψ (λ1 + λ3 )] , (D.5) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(−λ2 − z)

1 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ1 − λ2 )2 + 2γE ψ(λ1 + λ3 ) 2 +ψ(λ1 + λ3 )2 − 2ψ(λ1 − λ2 )(γE + ψ(λ1 + λ3 )) +ψ (λ1 − λ2 ) − ψ (λ1 + λ3 )] ,

=

1 2πi

+i∞

−i∞

(D.6)

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (λ3 − z)ψ(λ3 − z)

= Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ(λ1 − λ2 )ψ(λ2 + λ3 ) −ψ(λ1 + λ3 )ψ(λ2 + λ3 ) − ψ (λ1 + λ3 )] .

(D.7)

These are four formulae with the psi function with the same condition as in (D.3): +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(λ1 + z) 2πi −i∞ = −Γ (λ1 − λ2 )Γ (λ2 + λ3 ) × [ψ(λ1 − λ2 )(ψ(λ1 + λ3 ) − ψ(λ2 + λ3 )) + ψ (λ1 + λ3 )] , (D.8) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(λ2 + z)

1 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) (ψ(λ1 + λ3 ) − ψ(λ2 + λ3 ))2 2 +2γE (ψ(λ1 + λ3 ) − ψ(λ2 + λ3 )) − ψ (λ1 + λ3 ) + ψ (λ2 + λ3 )] , (D.9)

=

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(−λ2 − z)

1 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) 2 × [2(ψ(λ1 − λ2 ) − γE )(ψ(λ2 + λ3 ) − ψ(λ1 + λ3 )) +ψ(λ1 + λ3 )2 − ψ(λ2 + λ3 )2 − ψ (λ1 + λ3 ) − ψ (λ2 + λ3 ) , (D.10)

=

D.1 MB Integrals with Four Gamma Functions

1 2πi

209

+i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (λ3 − z)ψ(λ3 − z)

= Γ (λ1 − λ2 )Γ (λ2 + λ3 ) ψ(λ2 + λ3 )2 − ψ(λ1 + λ3 )ψ(λ2 + λ3 ) −ψ (λ1 + λ3 ) + ψ (λ2 + λ3 )] , (D.11)

−i∞

This is an example with the gluing of two poles: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗∗ (−1 − λ2 − z)Γ (λ3 − z) 2πi −i∞ = Γ (λ1 − λ2 − 1)Γ (λ2 + λ3 ) [1 − λ1 + λ2 +(λ1 + λ3 − 1)(ψ(λ1 + λ3 − 1) − ψ(λ2 + λ3 ))] ,

(D.12)

where the ﬁrst two poles of Γ (−1 − λ2 − z), i.e. z = −λ2 and z = −λ2 − 1, are considered left, with the corresponding change in notation. Here it is implied that λ1 + λ3 = 1. In the case λ1 + λ3 = 1, we have +i∞ 1 dz Γ (1 − λ1 + z)Γ (λ2 + z)Γ ∗∗ (−1 − λ2 − z)Γ (λ1 − z) 2πi −i∞ = (λ1 + λ2 − 1)Γ (λ1 + λ2 )Γ (−λ1 − λ2 ) . (D.13) Here is one more example of such an integral: +i∞ 1 dz Γ (1 − λ1 + z)Γ ∗ (λ2 + z)Γ ∗ (−1 − λ2 − z)Γ (λ1 − z) 2πi −i∞ = Γ (λ1 + λ2 )Γ (−λ1 − λ2 ) × [(λ1 + λ2 )(ψ(−λ1 − λ2 ) − ψ(1 + λ1 + λ2 )) − 1] . (D.14) Furthermore, we have +i∞ 1 dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z) 2πi −i∞ = Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [2γE + ψ(λ1 − λ2 ) + ψ(λ2 − λ1 )] ,

(D.15)

where the poles z = −λ1 and z = −λ2 are right. These are four more formulae with these conditions: +i∞ 1 dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(λ1 + z) 2πi −i∞

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4ψ(λ1 − λ2 )ψ(λ2 − λ1 ) 4 +4γE (ψ(λ2 − λ1 ) − 2ψ(λ1 − λ2 )) − 4ψ(λ1 − λ2 )2 − 4ψ (λ1 − λ2 ) +2ψ(λ2 − λ1 )2 + 2ψ (λ2 − λ1 ) , (D.16) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(λ2 + z)

210

D Table of MB Integrals

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 2ψ(λ1 − λ2 )2 4 +4ψ(λ1 − λ2 )(γE − ψ(λ2 − λ1 )) − 8γE ψ(λ2 − λ1 ) − 4ψ(λ2 − λ1 )2 +2ψ (λ1 − λ2 ) − 4ψ (λ2 − λ1 )] , (D.17) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(−λ2 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 2ψ(λ1 − λ2 )2 4 −4ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) − 2ψ (λ1 − λ2 )] ,

1 2πi

+i∞

−i∞

(D.18)

dz Γ ∗ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ (−λ1 − z)ψ(−λ1 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 2ψ(λ2 − λ1 )2 4 −4(γE + ψ(λ1 − λ2 ))ψ(λ2 − λ1 ) − 2ψ (λ2 − λ1 )] .

(D.19)

There are similar formulae with diﬀerent understanding of the nature of the poles: +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ = 2Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [γE + ψ(λ1 − λ2 )] , (D.20) where the pole z = −λ1 is left and the pole and z = −λ2 is right, and +i∞ 1 dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z) 2πi −i∞ = 2Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [γE + ψ(λ2 − λ1 )] , (D.21) where the pole z = −λ1 is right and the pole and z = −λ2 is left. These are four more formulae with these conditions: +i∞ 1 dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(λ1 + z) 2πi −i∞

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 4γE ψ(λ2 − λ1 ) + 2ψ(λ2 − λ1 )2 4 −8ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) + 2ψ (λ2 − λ1 )] , (D.22) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(λ2 + z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4γE ψ(λ2 − λ1 ) 4 (D.23) −6ψ(λ2 − λ1 )2 − 6ψ (λ2 − λ1 ) ,

D.1 MB Integrals with Four Gamma Functions

1 2πi

+i∞

−i∞

211

dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(−λ2 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 4γE ψ(λ2 − λ1 ) + 2ψ(λ2 − λ1 )2 4 −8ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) + 2ψ (λ2 − λ1 )] , (D.24) 1 2πi

+i∞

−i∞

dz Γ ∗ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ (−λ1 − z)ψ(−λ1 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4γE ψ(λ2 − λ1 ) 4 (D.25) −6ψ(λ2 − λ1 )2 − 6ψ (λ2 − λ1 ) .

Furthermore, we have +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ = Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [2γE + ψ(λ1 − λ2 ) + ψ(λ2 − λ1 )] ,

(D.26)

where the poles z = −λ1 and z = −λ2 are left. These are four more formulae with these conditions: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(λ1 + z) 2πi −i∞

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 2ψ(λ1 − λ2 )2 4 −4ψ(λ1 − λ2 )(γE + ψ(λ2 − λ1 )) − 2ψ (λ1 − λ2 )] , (D.27) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(λ2 + z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4(γE + ψ(λ1 − λ2 ))ψ(λ2 − λ1 ) 4 (D.28) −2ψ(λ2 − λ1 )2 − 2ψ (λ2 − λ1 ) , 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(−λ2 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 − 4ψ(λ1 − λ2 )2 4 +4γE ψ(λ2 − λ1 ) + 2ψ(λ2 − λ1 )2 − 4ψ(λ1 − λ2 )(2γE + ψ(λ2 − λ1 )) −4ψ (λ1 − λ2 ) + 2ψ (λ2 − λ1 )] , 1 2πi

+i∞

−i∞

(D.29)

dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z)ψ(−λ1 − z)

1 = − Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2γE2 + π 2 + 2ψ(λ1 − λ2 )2 4

212

D Table of MB Integrals

+4ψ(λ1 − λ2 )(γE − ψ(λ2 − λ1 )) − 8γE ψ(λ2 − λ1 ) −4ψ(λ2 − λ1 )2 + 2ψ (λ1 − λ2 ) − 4ψ (λ2 − λ1 ) .

(D.30)

We also have +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 2πi −i∞

= −Γ (λ1 − λ2 )ψ (λ1 − λ2 ) ,

(D.31)

where the pole z = −λ2 is right. These are three more formulae with this condition: +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 ψ(λ1 + z) 2πi −i∞ = −Γ (λ1 − λ2 ) [ψ(λ1 − λ2 )ψ (λ1 − λ2 ) + ψ (λ1 − λ2 )] ,

1 2πi

1 2πi

+i∞

−i∞

+i∞

−i∞

(D.32)

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 ψ(λ2 + z) = Γ (λ1 − λ2 )ψ (λ1 − λ2 ) [2γE + ψ(λ1 − λ2 )] ,

(D.33)

dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)2 ψ(−λ2 − z) 1 Γ (λ1 − λ2 ) [2γE ψ (λ1 − λ2 ) − ψ (λ1 − λ2 )] . 2

(D.34)

We also have +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)2 2πi −i∞

1 = Γ (λ1 − λ2 ) π 2 + 2(γE + ψ(λ1 − λ2 ))2 − 2ψ (λ1 − λ2 ) , 4 where the pole z = −λ2 is left, +i∞ 1 dz Γ (λ1 + z)2 Γ ∗ (−λ1 − z)Γ (λ2 − z) 2πi −i∞

(D.35)

=

= −Γ (λ1 + λ2 )ψ (λ1 + λ2 ) ,

(D.36)

where the pole z = −λ1 is left, and +i∞ 1 dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z) 2πi −i∞

1 = Γ (λ1 + λ2 ) 2(γE + ψ(λ1 + λ2 ))2 + π 2 − 2ψ (λ1 + λ2 ) , (D.37) 4 where the pole z = −λ1 is right. These are three more formulae with this condition:

D.1 MB Integrals with Four Gamma Functions

1 2πi

213

+i∞

dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z)ψ(λ1 + z) π2 1 3 2 = Γ (λ1 + λ2 ) ψ(λ1 + λ2 ) + 3ψ(λ1 + λ2 ) ψ (λ1 + λ2 ) − γE + 6 6 −2γE3 − γE π 2 + 6γE ψ (λ1 + λ2 ) − 4ζ(3) − 2ψ (λ1 + λ2 ) , (D.38)

1 2πi

−i∞

+i∞

−i∞

dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z)ψ(−λ1 − z)

1 Γ (λ1 + λ2 ) 12γE ψ(λ1 + λ2 )2 + 2ψ(λ1 + λ2 )3 12 π2 2 − 2ψ (λ1 + λ2 ) +3ψ(λ1 + λ2 ) 6γE + 3

=−

+2(4γE3 + 2γE π 2 − 6γE ψ (λ1 + λ2 ) + 8ζ(3) + ψ (λ1 + λ2 )) ,

1 2πi

+i∞

−i∞

(D.39)

dz Γ ∗ (λ1 + z)2 Γ (−λ1 − z)Γ (λ2 − z)ψ(λ2 − z)

1 = Γ (λ1 + λ2 ) 4γE ψ(λ1 + λ2 )2 + 2ψ(λ1 + λ2 )3 + 4γE ψ (λ1 + λ2 ) 4 (D.40) +ψ(λ1 + λ2 )(2γE2 + π 2 + 2ψ (λ1 + λ2 )) − 2ψ (λ1 + λ2 ) . In some situations, it is possible to evaluate MB integrals with higher derivatives of the ψ function. Here are some examples: +i∞ 1 dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z) 2πi −i∞ = 1 2πi

Γ (λ1 + λ2 )4 [2ψ(λ1 + λ2 ) − ψ(2(λ1 + λ2 ))] , Γ (2(λ1 + λ2 ))

(D.41)

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z)2 Γ (λ1 + λ2 )4 4ψ(λ1 + λ2 )2 − 4ψ(λ1 + λ2 )ψ(2(λ1 + λ2 )) Γ (2(λ1 + λ2 )) +ψ(2(λ1 + λ2 ))2 − ψ (2(λ1 + λ2 )) , (D.42)

=

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ (λ1 + z) =2

1 2πi

Γ (λ1 + λ2 )4 ψ (λ1 + λ2 ) , Γ (2(λ1 + λ2 ))

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z)ψ(λ2 − z)

(D.43)

214

D Table of MB Integrals

Γ (λ1 + λ2 )4 4ψ(λ1 + λ2 )2 − 4ψ(λ1 + λ2 )ψ(2(λ1 + λ2 )) Γ (2(λ1 + λ2 )) +ψ(2(λ1 + λ2 ))2 + ψ (λ1 + λ2 ) − ψ (2(λ1 + λ2 )) , (D.44)

=

1 2πi

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ(λ1 + z)2 ψ(λ2 − z)

Γ (λ1 + λ2 )4 8ψ(λ1 + λ2 )3 − 12ψ(λ1 + λ2 )2 ψ(2(λ1 + λ2 )) Γ (2(λ1 + λ2 ))

=

+2ψ(λ1 + λ2 )(3ψ(2(λ1 + λ2 ))2 + 2ψ (λ1 + λ2 ) − 3ψ (2(λ1 + λ2 ))) +ψ(2(λ1 + λ2 ))(3ψ (2(λ1 + λ2 )) − 2ψ (λ1 + λ2 )) −ψ(2(λ1 + λ2 ))3 − ψ (2(λ1 + λ2 )) , (D.45) 1 2πi

+i∞

−i∞

dz Γ (λ1 + z)2 Γ (λ2 − z)2 ψ (λ1 + z)ψ(λ2 − z) Γ (λ1 + λ2 )4 [4ψ(λ1 + λ2 )ψ (λ1 + λ2 ) Γ (2(λ1 + λ2 )) −2ψ(2(λ1 + λ2 ))ψ (λ1 + λ2 ) + ψ (λ1 + λ2 )] ,

=

(D.46)

D.2 MB Integrals with Six Gamma Functions This is the second Barnes lemma: +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ (λ4 − z)Γ (λ5 − z) dz 2πi −i∞ Γ (λ6 + z) Γ (λ1 + λ4 )Γ (λ2 + λ4 )Γ (λ3 + λ4 )Γ (λ1 + λ5 ) = Γ (λ1 + λ2 + λ4 + λ5 )Γ (λ1 + λ3 + λ4 + λ5 ) Γ (λ2 + λ5 )Γ (λ3 + λ5 ) , × Γ (λ2 + λ3 + λ4 + λ5 )

(D.47)

where λ6 = λ1 + λ2 + λ3 + λ4 + λ5 . Here is a collection of its corollaries: +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ ∗ (−λ3 − z)Γ (λ4 − z) dz 2πi −i∞ Γ (λ5 + z) Γ (λ1 − λ3 )Γ (λ2 − λ3 )Γ (λ3 + λ4 ) [ψ(λ1 + λ2 − λ3 + λ4 ) = Γ (λ1 + λ2 − λ3 + λ4 ) (D.48) +ψ(λ3 + λ4 ) − ψ(λ1 + λ4 ) − ψ(λ2 + λ4 )] , where λ5 = λ1 + λ2 + λ4 and the pole z = −λ3 is considered left, +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (λ3 + z)Γ (−λ3 − z)Γ (λ4 − z) dz 2πi −i∞ Γ (λ5 + z)

D.2 MB Integrals with Six Gamma Functions

215

Γ (λ1 − λ3 )Γ (λ2 − λ3 )Γ (λ3 + λ4 ) [ψ(λ1 − λ3 ) + ψ(λ2 − λ3 ) Γ (λ1 + λ2 − λ3 + λ4 ) −ψ(λ1 + λ4 ) − ψ(λ2 + λ4 )] , (D.49)

=

where λ5 = λ1 + λ2 + λ4 and the pole z = −λ3 is considered right, +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ3 + z)Γ (λ3 − z)2 dz 2πi −i∞ Γ (λ4 + z) Γ (λ1 + λ3 )Γ (λ2 + λ3 ) [ψ (λ1 + λ3 ) + ψ (λ2 + λ3 )] , (D.50) =− Γ (λ1 + λ2 + 2λ3 ) where λ4 = λ1 + λ2 + λ3 and the pole z = λ3 is considered right, +i∞ 1 Γ (λ1 + z)Γ ∗ (λ2 + z)2 Γ (−λ2 − z)Γ (λ3 − z) dz 2πi −i∞ Γ (λ4 + z) 2 Γ (λ1 − λ2 )Γ (λ2 + λ3 ) π + (γE − ψ(λ1 − λ2 ) + ψ(λ1 + λ3 ) = 2Γ (λ1 + λ3 ) 2 +ψ(λ2 + λ3 ))2 + ψ (λ1 − λ2 ) + ψ (λ1 + λ3 ) − ψ (λ2 + λ3 ) , (D.51) where λ4 = λ1 + λ2 + λ3 and the pole z = −λ2 is considered right, +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)2 Γ ∗ (−λ2 − z)Γ (λ3 − z) dz 2πi −i∞ Γ (λ4 + z) Γ (λ1 − λ2 )Γ (λ2 + λ3 ) [ψ (λ1 + λ3 ) − ψ (λ2 + λ3 )] , (D.52) = Γ (λ1 + λ3 ) where λ4 = λ1 + λ2 + λ3 and the pole z = −λ2 is considered left. The integrals (D.47) can be evaluated recursively in the case where the diﬀerence λ6 − (λ1 + λ2 + λ3 + λ4 + λ5 ) is a positive integer. In particular, we have +i∞ Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ (λ4 − z)Γ (−z) 1 dz 2πi −i∞ Γ (λ5 + z) (Γ (1 + λ2 + λ3 + λ4 ))−1 Γ (λ1 )Γ (λ3 )Γ (λ2 + λ4 ) = Γ (1 − λ1 − λ3 − λ4 )Γ (1 + λ1 + λ2 + λ4 )Γ (λ1 + λ3 + λ4 ) × [Γ (1 + λ2 )Γ (1 − λ1 − λ3 − λ4 )Γ (λ1 + λ4 )Γ (λ3 + λ4 ) −Γ (λ2 )Γ (−λ1 − λ3 − λ4 )Γ (1 + λ1 + λ4 )Γ (1 + λ3 + λ4 )] , (D.53) where λ5 = λ1 + λ2 + λ3 + λ4 + 1, and +i∞ 1 Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 + z)Γ (λ4 − z)Γ (−z) dz 2πi −i∞ Γ (λ5 + z) −1 (Γ (2 + λ2 + λ3 + λ4 )) Γ (λ1 )Γ (λ3 )Γ (λ2 + λ4 ) = Γ (1 − λ1 − λ3 − λ4 )Γ (2 + λ1 + λ2 + λ4 )Γ (λ1 + λ3 + λ4 ) × [Γ (2 + λ2 )Γ (1 − λ1 − λ3 − λ4 )Γ (λ1 + λ4 )Γ (λ3 + λ4 ) −2Γ (1 + λ2 )Γ (−λ1 − λ3 − λ4 )Γ (1 + λ1 + λ4 )Γ (1 + λ3 + λ4 ) +Γ (λ2 )Γ (−1 − λ1 − λ3 − λ4 )Γ (2 + λ1 + λ4 )Γ (2 + λ3 + λ4 )] , (D.54) where λ5 = λ1 + λ2 + λ3 + λ4 + 2.

216

D Table of MB Integrals

Here are more corollaries of the second Barnes lemma: +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) = Γ (1 − λ1 )Γ (1 − λ2 ) × [Γ (1 − λ1 )Γ (1 − λ2 ) − Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 )] , (D.55) where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered left, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z = −Γ (λ1 )Γ (λ2 )Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 ) Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) + Γ (1 − λ1 )Γ (1 − λ2 ) × [Γ (1 − λ1 )Γ (1 − λ2 ) − Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 )] , (D.56) where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered right, +i∞ 1 Γ ∗ (λ + z)2 Γ ∗ (z)Γ (−z)Γ (−λ − z) dz 2πi −i∞ Γ (λ + 1 + z) +i∞ dz 1 Γ (λ + z)Γ (z)Γ ∗ (−z)Γ ∗ (−λ − z) =− 2πi −i∞ z

1 = Γ (λ)Γ (−λ) 12(γE + ψ(λ)) + 2λπ 2 6λ +3λ((ψ(λ) − ψ(−λ))2 − ψ (λ) + ψ (−λ)) , (D.57) where the nature of the poles at z = 0 and z = −λ is indicated by asterisks, according to our conventions, +i∞ 1 Γ (λ + z)2 Γ (z)Γ ∗ (−z)Γ ∗ (−λ − z) dz 2πi −i∞ Γ (λ + 1 + z) +i∞ dz ∗ 1 1 Γ (λ + z)Γ ∗ (z)Γ (−z)Γ (−λ − z) = 2 Γ (λ)Γ (−λ) =− 2πi −i∞ z λ

π2 2 × 1 + λ(ψ(λ) + ψ(−λ) + 2γE ) − λ ψ (λ) − , (D.58) 6 1 2πi

Γ (λ + z)2 Γ ∗ (z)Γ (−z)Γ ∗ (−λ − z) Γ (λ + 1 + z) +i∞ dz 1 Γ (λ + z)Γ ∗ (z)Γ (−z)Γ ∗ (−λ − z) =− 2πi −i∞ z

π2 1 = Γ (λ)Γ (−λ) 2(γE + ψ(λ)) − λ ψ (λ) − . λ 6

+i∞

dz

−i∞

(D.59)

D.2 MB Integrals with Six Gamma Functions

217

We also have +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z 2 Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 ) = Γ (2 − λ1 )Γ (1 − λ2 ) ×Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) [1 + (λ1 − 1)(ψ(2 − λ1 ) + ψ(1 − λ2 ) −ψ(2 − λ1 − λ2 − λ3 ) − ψ(λ3 ))] ,

(D.60)

where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered left, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ (λ3 − z)Γ (λ4 − z) 2πi −i∞ z 2 = Γ (2 − λ1 − λ2 − λ3 )Γ (λ3 ) [−Γ (λ1 )Γ (λ2 )(ψ(λ1 ) + ψ(λ2 ) −ψ(2 − λ1 − λ2 − λ3 ) − ψ(λ3 )) Γ (2 − λ1 − λ3 )Γ (1 − λ2 − λ3 )Γ (λ1 + λ3 − 1)Γ (λ2 + λ3 ) + Γ (2 − λ1 )Γ (1 − λ2 ) × [1 + (λ1 − 1)(ψ(2 − λ1 ) + ψ(1 − λ2 ) −ψ(2 − λ1 − λ2 − λ3 ) − ψ(λ3 ))]] ,

(D.61)

where λ1 + λ2 + λ3 + λ4 = 2, and the pole at z = 0 is considered right, +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z 1 = − 2 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) [2λ1 − λ2 λ1 λ2 +λ1 (λ1 + λ2 )(γE + ψ(λ1 − λ2 )) − λ1 (λ1 − λ2 ) ×(ψ(−λ1 ) − ψ(−λ2 ) + ψ(λ2 − λ1 ) − ψ(1 − λ1 + λ2 ))] , (D.62) where the pole at z = 0 is left and the nature of the ﬁrst poles of the gamma functions is shown by asterisks, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z

1 = 2 2 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) λ21 − λ1 λ2 + λ22 λ1 λ2 −λ1 λ2 (λ1 + λ2 )γE + λ1 (λ1 − λ2 )λ2 (ψ(−λ1 ) − ψ(−λ2 )) −λ1 λ2 (λ2 ψ(λ1 − λ2 ) + λ1 ψ(λ2 − λ1 ))] , where the pole at z = 0 is left, +i∞ 1 dz Γ (λ1 + z)Γ ∗ (λ2 + z)Γ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z 2

1 = 3 2 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) 2(λ21 + λ1 λ2 − λ22 ) λ1 λ2

(D.63)

218

D Table of MB Integrals

+λ1 (λ21 + λ22 )(ψ(λ1 − λ2 ) + γE ) −λ1 (λ21 − λ22 )(ψ(−λ1 ) − ψ(−λ2 ) + ψ(−λ1 + λ2 ) − ψ(1 − λ1 + λ2 )) −λ21 λ2 (λ1 − λ2 )(ψ (−λ1 ) − ψ (−λ2 )) , (D.64) where the pole at z = 0 is left, +i∞ 1 dz Γ (λ1 + z)Γ (λ2 + z)Γ ∗ (−λ2 − z)Γ ∗ (−λ1 − z) 2πi −i∞ z 2

1 = − 3 3 Γ (λ1 − λ2 )Γ (λ2 − λ1 ) (λ1 + λ2 )(2λ21 − 3λ1 λ2 + 2λ22 ) λ1 λ2 −λ1 λ2 (λ21 + λ22 )γE + λ1 λ2 (λ21 − λ22 )ψ(−λ1 ) −λ1 λ32 (ψ(λ1 − λ2 ) − ψ(−λ2 )) − λ31 λ2 (ψ(−λ2 ) + ψ(λ2 − λ1 )) +λ31 λ22 (ψ (−λ1 ) − ψ (−λ2 )) − λ21 λ32 (ψ (−λ1 ) − ψ (−λ2 )) , where the pole at z = 0 is left, +i∞ 1 dz Γ (λ + z)Γ (z)Γ ∗ (−z)Γ ∗ (−λ − z) 2πi −i∞ z 2 1 = − 3 Γ (λ)Γ (−λ) [12 − 6λ(2γE + ψ(−λ) + ψ(λ)) 6λ +λ2 (π 2 − 6ψ (−λ)) − 3λ3 (ψ (−λ) + 2ζ(3)) ,

(D.65)

(D.66)

where the pole at z = 0 is left, +i∞ 1 dz Γ (λ + z)Γ ∗ (z)Γ (−z)Γ ∗ (−λ − z) 2πi −i∞ z 2

1 = 3 Γ (λ)Γ (−λ) −12 + 6λ(2γE + ψ(−λ) + ψ(λ)) − λ2 (π 2 − 6ψ (−λ)) 6λ −λ3 (π 2 (ψ(−λ) − ψ(λ)) + (ψ(−λ) − ψ(λ))3 − 2ψ (−λ) − ψ (λ) (D.67) +3(ψ(−λ) − ψ(λ))(ψ (−λ) + ψ (λ)) − 6ζ(3))] , where the pole at z = 0 is right, +i∞ 1 dz Γ (λ + z)2 Γ ∗ (−λ − z)2 2πi −i∞ z 1 = − 4 6 + λ2 (π 2 − 6ψ (−λ)) + 12λ3 ζ(3) , 6λ where the pole at z = 0 is left, +i∞ 1 dz Γ (λ + z)2 Γ ∗ (−λ − z)2 2πi −i∞ z 2 1 = 5 12 + λ2 (π 2 − 6ψ (−λ)) − 3λ3 (ψ (−λ) − 2ζ(3)) , 3λ where the pole at z = 0 is left,

(D.68)

(D.69)

D.2 MB Integrals with Six Gamma Functions

1 2πi

+i∞

−i∞

219

dz Γ (λ + 1 + z)2 Γ (−λ − z)2 z2

= 2Γ (1 + λ)2 Γ (−λ)2 (ψ(−λ) − ψ(1 + λ)) − ψ (−λ) , where the pole at z = 0 is right.

(D.70)

E Analysis of Convergence and Sector Decompositions

In this appendix, the analysis of convergence of Feynman integrals based on the alpha representation is brieﬂy described. The UV divergences come from the region of small values of the α-parameters in (2.36), while the oﬀ-shell IR divergences arise from the integration over large αl . To reveal these divergences, the integration region is divided into so-called ‘sectors’, where new integration variables are introduced, with the goal to obtain a factorization of the integrand. Then the analysis of convergence reduces to power counting in one-dimensional integrals. However, this mathematical analysis of convergence is restricted to the cases where the external momenta are Euclidean. Generalizations of these results connected with the analysis of convergence and dimensional regularization to Feynman integrals at a mass shell or at a threshold are not known. On the other hand, it turns out that, in these important cases, one can introduce some practical sector decompositions and corresponding sectors [5] that give the possibility to have control on the convergence and, in particular, provide a powerful method of evaluating Feynman integrals in situations with strong UV, IR and collinear divergences. The corresponding algorithm is described in Sect. E.2.

E.1 Analysis of Convergence We obtain the alpha representation of an analytically and dimensionally regularized Feynman integral corresponding to a graph Γ starting from the alpha representation (2.36) and substituting the powers of propagators al by al + λl with general complex numbers λl . For simplicity, let us assume the scalar case and that the powers of propagators are equal to one. (If al > 1, one can represent such a line by a sequence of al lines.) In this case the alpha representation takes a simpler form FΓ (q, m; d, λ) ∞ λ −d/2 2 l = dα αl U(α) exp iV(q, α)/U(α) − i ml αl , 0

l

V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 221–232 (2004) c Springer-Verlag Berlin Heidelberg 2004

l

(E.1)

222

E Analysis of Convergence and Sector Decompositions

where the functions U and V are given by (2.24) and (2.25), and from now on we omit the coeﬃcient (−1)L eiπ( λl +h(1−d/2))/2 π hd/2 / Γ (λ + 1) , l

l

which is irrelevant to the analysis of convergence. In this appendix (as in Chap. 6), families of variables are denoted by underlined letters, i.e. q = (q1 , . . . , qn ), m = (m1 , . . . , mL ), λ = (λ1 , . . . , λL ), α = (α1 , . . . , αL ), etc., with dα = dα1 . . . dαL . Let us also assume here and later that the limit of integration refers to all of the integration variables involved. The alpha parameters have dimension −2 in mass units. By making the change of variables αl → µ−2 αl , where µ is a massive parameter, we can transform to dimensionless alpha parameters. For simplicity, let us take µ = 1 in this appendix. To separate the analysis of the UV and IR convergence as much as possible let us decompose the integration from 0 to ∞ over each alpha parameter into two regions: from 0 to 1 and from 1 to ∞. The integral (E.1) is then divided into 2L pieces, each of which is determined by a decomposition of the set of lines L of the given graph into two subsets, Lα and Lβ , corresponding to the integrations over the UV region (from 0 to 1) and the IR region (from 1 to ∞), respectively. For a given piece generated by a subset Lα , let us change the variables αl for l ∈ Lβ according to αl = 1/βl . The corresponding integral then takes the form 1 λ −λ −ε Lα dα dβ αl l βl l U(α, β)−d/2 FΓ (q, m; d, λ) = 0

l∈Lα

l∈Lβ

× exp iV(q, α, β)/U(α, β) − i

m2l αl − i

l∈Lα

m2l /βl .

(E.2)

l∈Lβ

For brevity, the new functions U and V are denoted by the same letters, although they are now of the form U(α, β) = βl U(α)|αl →1/βl ,l∈Lβ l∈Lβ

=

T ∈T 1

V(q, α, β) =

T ∈T 2

βl ,

(E.3)

l∈Lβ ∩T

βl V(q, α)α →1/β ,l∈L l

αl

l∈Lα \T

l∈Lβ

=

l∈Lα \T

αl

l

l∈Lβ ∩T

β

2 βl q T .

(E.4)

E.1 Analysis of Convergence

223

Remember that ±q T is the sum of the external momenta that ﬂow into one of the connectivity components of a 2-tree T . For a given piece FΓLα , let us change the numbering of the lines in such a way that the UV lines (i.e. those with αl ≤ 1) have smaller numbers. Thus we perform integration in the domain 0 ≤ αl ≤ 1, 1 ≤ l ≤ ¯l and 0 ≤ βl ≤ 1, ¯l + 1 ≤ l ≤ L, where ¯l = |Lα |. If S is a ﬁnite set, we denote by |S| the number of its elements. As we shall see, the analysis of UV and IR convergence is now decoupled. To analyse the UV convergence let us divide the domain of integration over αl into sectors. In the following, we shall use sectors of two types associated with nests and forests, respectively. The sectors connected with nests of subgraphs, (i.e. that γ ⊂ γ or γ ⊂ γ for any pair of the subgraphs of any nest; let us call them N -sectors) [14] are deﬁned by α1 ≤ . . . ≤ α¯l

(E.5)

and similar inequalities obtained by permutations. Without loss of generality, let us consider only the sector (E.5). Let us then change the integration variables according to αl = tl . . . t¯l .

(E.6)

The new (N -sector) variables tl are expressed in terms of αl by αl /αl+1 if l < ¯l . (E.7) tl = if l = ¯l α¯l The corresponding Jacobian equals tll−1 . The decomposition of the IR integration, over βl , is performed in a quite similar way. The following are the corresponding analogues of N -sectors and sector variables: βL ≥ . . . ≥ β¯l+1 , βl = τ¯l+1 . . . τl , βl /βl−1 if l > ¯l + 1 τl = , β¯l+1 if l = ¯l + 1

(E.8) (E.9) (E.10)

and the corresponding Jacobian is τlL−l . So, the initial integral is eventually divided into (L + 1)! sectors απ(1) ≤ . . . ≤ απ(¯l) ≤ 1 ≤ απ(¯l+1) ≤ απ(L) ,

(E.11)

which are labelled by permutations π of the numbers 1, . . . , L and the number ¯l. As we have stated, we consider only the contribution of the identical permutation, i.e. π(l) = l, l = 1, . . . , L. Although these sectors provide a resolution of the singularities of the integrand, they can turn out to be too rough for analysing convergence. A more sophisticated set of sectors corresponds to the maximal UV and IR forests. A set f of 1PI subgraphs and single lines with non-coincident end

224

E Analysis of Convergence and Sector Decompositions

points is called a UV forest [8, 22, 16] if the following conditions hold: (i) for any pair γ, γ ∈ f , we have either γ ⊂ γ , γ ⊂ γ or L(γ ∩ γ ) = ∅; (ii) if γ 1 , . . . , γ n ∈ f and L(γ i ∩ γ j ) = ∅ for any pair from this family, the subgraph ∪i γ i is one-vertex-reducible (i.e. can be made disconnected by deleting a vertex). Let F be a maximal UV forest (i.e. there are no UV forests that include F) of a given graph Γ . An element γ ∈ F is called trivial if it consists of a single line and is not a loop line. Any maximal UV forest has h(Γ ) nontrivial and L − h(Γ ) trivial elements. Let us deﬁne the mapping σ : F → L such that σ(γ) ∈ L(γ) and σ(γ) ∈ L(γ ) for any γ ⊂ γ, γ ∈ F. Its inverse σ −1 uniquely determines the minimal element σ −1 (l) of the UV forest F that contains the line l. Let us denote by γ+ the minimal element of F that strictly includes the given element γ. For a given maximal UV forest F, let us deﬁne the corresponding sector (F -sector) as (E.12) DF = α|αl ≤ ασ(γ) ≤ 1, l ∈ γ ∈ F . The intersection of two diﬀerent F -sectors has zero measure and the union of all the sectors gives the whole integration domain of the UV alpha parameters (i.e. αl ≤ 1) (see [8, 16, 18, 22]). For a given F -sector, let us introduce new variables labelled by the elements of F, tγ , (E.13) αl = γ∈F : l∈γ

L(γ)−1 where the corresponding Jacobian is γ tγ . The inverse formula is ασ(γ) /ασ(γ+ ) if γ is not maximal . (E.14) tγ = ασ(γ) if γ is maximal Consider, for example, the two-loop self-energy diagram of Fig. 3.9 and the following maximal UV forest F consisting of γ 1 = {1}, γ 2 = {2}, γ 3 = {3}, γ 4 = {1, 2, 5}, γ 5 = Γ . The mapping σ is σ(γ 1 ) = 1, σ(γ 2 ) = 2, σ(γ 3 ) = 3, σ(γ 4 ) = 5, σ(γ 5 ) = 4. The sector associated with this maximal UV forest is given by DF = {α1,2 ≤ α5 ≤ α4 , α3 ≤ α4 } and the sector variables are tγ 1 = α1 /α5 , tγ 2 = α2 /α5 , tγ 3 = α3 /α4 , tγ 4 = α5 /α4 , tγ 5 = α4 . The IR F -sectors and variables are introduced in a quite analogous way. New variables τγ are associated with maximal IR forests composed of IRirreducible subgraphs – see [18]. (A subgraph γ of Γ is called IR irreducible [10, 18] if the reduced graph Γ/γ is one-vertex-irreducible. (As in Chap. 2, Γ/γ is obtained from Γ by reducing every connectivity component of γ to a point.) The UV and IR maximal forests Fα and Fβ , composed of lines Lα and Lβ , respectively, are then combined in pairs to generate ‘generalized maximal forests’, with corresponding variables {tγ , τγ }, γ ∈ Fα , γ ∈ Fβ . As a result, the initial integration domain is divided into F -sectors associated with generalized maximal forests.

E.1 Analysis of Convergence

225

In each of the N - or F -sectors, the function (E.3) takes a factorized form in the new variables [8, 16, 18, 22, 24]: ¯ L l L−l+1−h(Γ/γ ) h(γ ) l−1 U = tl l [1 + PN (t, τ )] τl (E.15) =

l=1

l=¯ l+1

th(γ) γ

γ∈Fα

τγL(γ)−h(Γ/γ) [1 + PF (t, τ )] ,

(E.16)

γ∈Fβ

where PN and PF are non-negative polynomials, γl denotes the subgraph consisting of the lines {1, . . . , l}, and again γ = Γ \γ. The factorization of the function (E.4) in the N -sector variables is of the form ¯ L l −1 L−l+1−h(Γ/γl−1 ) h(γ ) l τ¯l+1 . . . τl0 V= tl τl l=¯ l+1

l=1

2 × q T0 + P0 (q, t, τ ) ,

(E.17)

where l0 denotes the number such that all the external vertices belong to the same connectivity component of the subgraphs γl for l ≥ l0 . In the Euclidean domain, where 2 qi −1 and in the sense of the limit δ → +0 with m2 → m2 − iδ (with identical resulting prescriptions in both these variants). In particular, such integrals are well deﬁned for the integer values λ = −1, −2, . . . Thus we have IR convergence when either the subgraph γl (or just γ) has at least one non-zero mass or its completion γl−1 (or γ) does not have all the external vertices in the same connectivity component. Therefore it is suﬃcient to check the IR convergence for the other IR-irreducible subgraphs. The domain of the regularization parameters λl and ε where these sector integrals are convergent is determined by the inequalities Re λ(γ) + h(γ) Re ε > [ω(γ)/2] , Re λ(γ) − h(Γ/γ) Re ε < [(ω(Γ ) − ω(γ) + 1)/2] ,

(E.23a) (E.23b)

which correspond, respectively, to UV-irreducible subgraphs and massless IRirreducible subgraphs whose completions γ contain all the external vertices in the same connectivity component. It turns out that this domain is non-empty for any graph without massless detachable subgraphs, i.e. massless subgraphs with zero external momenta. This statement can be proven [22] by observing that the parameters |T 1 | (0) λl = (2 − ε) 1 + δ − l1 − 1 , (E.24) |T | where Tl1 is the set of trees containing the line l, satisfy (E.23a) and (E.23b) for suﬃciently small δ > 0. (As before, | . . . | is the number of elements in the corresponding ﬁnite set.) Here again the scalar case is assumed. The generalization to a general diagram is straightforward: one adds nl /2 to the right-hand side of (E.24), where nl is the degree of the polynomial in the numerator of the lth propagator. In order to see that the Feynman integral can be continued from the above domain of mutual convergence to the whole hypercomplex plane of the variables (λ, ε) let us use the well-known property of the integrals ∞ F (λ) = dx xλ φ(x) . (E.25) 0

(In distributional language, this is the analytic property of the distribution xλ+ – see [12].) Indeed, the integral (E.25) with an inﬁnitely diﬀerentiable

228

E Analysis of Convergence and Sector Decompositions

function φ which has a compact support (or, a fast decrease at large values of x – see details in [12]) is absolutely convergent for all complex values of λ with Re λ > −1 so that it deﬁnes an analytic function of λ in this domain. This function can be continued analytically to the whole complex plane of λ with simple poles at λ = −1, −2, . . .. To perform the analytical continuation to the domain Re λ > −2 one decomposes the integral (E.25) into the two integrals, from 0 to 1 and from 1 to ∞, and uses an appropriate subtraction in the ﬁrst of them, i.e. represents φ(x) in (E.25) as (φ(x) − φ(0)) + φ(0) and takes the integral with the second term explicitly to obtain ∞ 1 φ(0) + dx xλ (φ(x) − φ(0)) + dx xλ φ(x) . (E.26) F (λ) = λ + 1 0 1 The ﬁrst integral on the right-hand side is now absolutely convergent at Re λ > −2 so that we obtain, from (E.26), an explicit analytic continuation of the function F (λ) to this domain. We also see that this function has a simple pole at λ = −1 with the residue φ(0).1 This procedure can naturally be generalized for the analytic continuation to the whole complex plane. To do this, one makes more subtractions2 : 1 n n (j) φ (0) j φ(j) (0) x + F (λ) = dx xλ φ(x) − j! j!(λ + j + 1) 0 j=0 j=0 ∞ + dx xλ φ(x) . (E.27) 1

Let us come back to our sector integrals. It follows from the factorizations (E.20), when they are written for all the sectors, that the Feynman integral can be continued from the above domain of mutual convergence to the whole hypercomplex plane of the variables (λ, ε) as a meromorphic function, with series of UV and IR poles. It is also clear that, in the case where there is no non-empty mutual-convergence domain, the contribution from any sector can be made convergent by choosing the absolute values of the real parts of the In distributional language, this means that the functional xλ+ has the pole at λ = −1 with the residue δ(x). By the way, in the domain −2 < Re λ < −1, the ∞ value φ(0)/(λ+1) can be rewritten as −φ(0) 1 dx xλ . After we combine it with the last integral in (E.26) we obtain the followingcompact expression for the analytic ∞ continuation of (E.25) to this band: F (λ) = 0 dx xλ (φ(x) − φ(0)). However, in our case of factorized expressions resulting from sector integrals, this is not relevant because we are dealing with ﬁnite regions of integration. 2 With the help of this procedure, the analytic continuation of (E.25) to the band −n − 1 < Re λ < −n − 1 takes the form [12]: 1

!

∞ λ

dx x

F (λ) = 0

φ(x) −

n φ(j) (0) j=0

j!

"

j

x

.

E.2 Practical Sector Decompositions

229

UV/IR analytic-regularization parameters to be suﬃciently large (positive and negative for l ≤ ¯l and l > ¯l, respectively). The analytic regularization can then be switched oﬀ, by analytic continuation, and one obtains [9] a dimensionally regularized Feynman integral as the sum of its sector contributions, which were deﬁned in their own initial analyticity domains using the auxiliary analytic regularization. Therefore, we obtain a deﬁnition of dimensional regularization for any Feynman integral at Euclidean external momenta.

E.2 Practical Sector Decompositions The sector decompositions of the previous section are simpler than the sectors of [22]. However, if we want to apply sectors for the numerical evaluation of Feynman integrals the initial decomposition of the integration domain over every alpha parameter in the two regions is not optimal at all because we obtain 2L pieces from the beginning. So, the natural idea is to apply the sectors of [22]. Presumably, this procedure can be implemented on a computer, but no such examples are known. The bad news is that, although the sector decompositions discussed above can successfully be used for proving theorems on renormalization [14, 21, 24] and on asymptotic expansions in limits of momenta and masses typical of Euclidean space (see [17, 18] and Appendix B of [19]), they are not suﬃcient for resolving the singularities of the integrand in the case of Feynman integrals on a mass shell or at a threshold. Let us consider again Example 3.3 of Sect. 3.3, with the basic functions U and V given by (3.23), and try to apply the N -sectors to resolve the singularities of the alpha integral in the region of large αl . To do this, let us turn to the variables βl = 1/αl , as in the previous section, where we obtain the functions U(β) = β1 β2 β3 + β1 β2 β4 + β1 β3 β4 + β2 β3 β4 , V(β) = tβ2 β4 + sβ1 β3 .

(E.28) (E.29)

Consider now the N -sector β2 ≤ β1 ≤ β3 ≤ β4 and introduce the variables (E.10), i.e. by means of the relations β2 = τ1 τ2 τ3 τ4 ,

β1 = τ2 τ3 τ4 ,

β3 = τ3 τ4 , β4 = τ4 .

(E.30)

In these sector variables, the function (E.28) factorizes, in a suitable way, according to (E.15), but the function (E.29) does not: V(τ ) = τ2 τ3 τ42 (sτ1 + tτ3 ) .

(E.31)

Such a phenomenon would never happen for Feynman integrals considered at Euclidean external momenta – see the general result (E.17). So, we do not have a nice factorization property similar to (E.17) for the contribution of the sector under consideration. In order to perform the analysis of convergence, the factor sτ1 + tτ3 raised to some power dependent

230

E Analysis of Convergence and Sector Decompositions

on ε has to be further factorized. The natural idea here is to perform a next sector decomposition, using N -sectors, then proceed further if we do not immediately succeed, etc. However, this procedure looks awful from the practical point of view: to have L! contributions at the ﬁrst step, then (L!)2 at the second step is a very bad idea if we think of a computer implementation. Still the idea to introduce, recursively, more and more sectors has turned out to be quite successful and easily implemented in practice. A suitable algorithm based on sector decompositions for resolving singularities of general Feynman integrals, in particular, considered on a mass-shell or at a threshold, possibly, with severe UV, IR and collinear divergences, was developed in [5]. On the one hand, this algorithm makes the analysis of the singularities in ε possible for any given Feynman integral. On the other hand, it gives a powerful universal numerical method for evaluating Feynman integrals. The starting point of the algorithm of [5] is representation (3.32), where the sum of all the parameters αl is implied in the δ-function. It is supposed that all the kinematical invariants and the masses have the same sign, i.e. if there is a non-zero mass, all the invariants are non-positive. Then one introduces the following primary sectors ∆l labelled by the number l = 1, . . . , L: αi ≤ αl ,

l = i = 1, 2, . . . , L

(E.32)

and turns, in a given sector ∆l , to the variables αi /αl if i = l ti = . if i = l αl

(E.33)

Then the integration over tl is taken due to the δ-function, and one obtains the integral 1 L−(h+1)d/2 U . (E.34) Fl = dti L−hd/2 V tl =1 i =l 0

Here we used the fact that the functions U and V are homogeneous functions of the alpha parameters with the homogeneity degrees h and h + 1, respectively. The goal of the introduction of the sector decompositions is to obtain a perfect factorization, i.e. of the form (E.15) for U and of the form (E.17) for 2 −V + U m2l αl , where, instead of q T0 , there is some positive combination of the kinematical invariants and masses. So, if the perfect factorization is not achieved, for the contribution of the given sector ∆l , the next natural step is to introduce a second decomposition in a similar way, i.e. over L − 1 sectors ∆lj , ti ≤ tj , i = 1, 2, . . . , L ,

i = j, l ,

j = l .

(E.35)

and new variables ti similarly to (E.33). One may hope that sooner or later a perfect factorization will be achieved. If this is the case, one obtains a sum of parametric integrals, over some sector variables ti , where the singularities are factorized, i.e. the integrand is a product of ti raised to some powers

E.2 Practical Sector Decompositions

231

λi = ni + hi ε, with integer ni and hi = 0, and the two functions (also raised to similar powers) which result from U and V and are positive in the integration region. In such a ‘perfect’ situation, the analysis of convergence reduces to counting powers of the variables ti . This reminds again, as in the end of the previous section, the analysis of the distribution xλ+ – see [12]. Explicitly, we have integrations over sector variables (of some level of iterations) of the form 1 dt tn+hε φ(t) , (E.36) G(ε) = 0

where t is one of the sector variables, n and h = 0 are integer numbers and φ(t) is a function with φ(0) = 0 which involves similar factorized integrations over the rest of the sector variables. If n ≥ 0, the integration over t does not generate poles in ε. Suppose that n is negative. The procedure outlined in the end of the previous section suggests a similar subtraction: 1 −n−1 φ(j) (0) tj dt tn+hε φ(t) − G(ε) = j! 0 j=0 +

−n−1 j=0

φ(j) (0) . j!(n + hε + j + 1)

(E.37)

After performing such manipulations with integrations over all the sector variables ti with ni < 0 one obtains a linear combination of integrals where one can perform an expansion in a Laurent series in ε. This provides the possibility to formulate an algorithm for the numerical evaluation of any term of expansion of the given Feynman integral in ε. Numerous practical calculations have shown [5] that this algorithm works for complicated Feynman integrals with multiple IR and collinear divergences. For example, analytical results for double and triple boxes [3, 20, 23] were numerically conﬁrmed by means of this algorithm. Once again, this is a method with experimental mathematics. It is not guaranteed, as in a mathematical theorem, that the process of the recursive introduction of the sector decompositions described above will stop at some point with a perfect factorization. Moreover, practical calculations have shown that one has to avoid possible closed loops in the algorithm. However, this is the only working general algorithm at the moment, applicable at any loop order, with applications restricted only by the computer time. One may hope that the algorithm can be generalized to the cases without restrictions on the signs of the kinematical invariants and the masses. Observe, however, that another important generalization, to the case of phase-space integrals, was already developed and successfully applied in practice in [1, 2, 6, 11, 13].

232

E Analysis of Convergence and Sector Decompositions

References 1. C. Anastasiou, K. Melnikov and F. Petriello, Phys. Rev. D 69 (2004) 076010; Phys. Rev. Lett. 93 (2004) 032002. 231 2. C. Anastasiou, K. Melnikov and F. Petriello, hep-ph/0409088. 231 3. C. Anastasiou, J.B. Tausk and M.E. Tejeda-Yeomans, Nucl. Phys. Proc. Suppl. 89 (2000) 262. 231 4. M. Beneke and V.A. Smirnov, Nucl. Phys. B 522 (1998) 321. 5. T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 221, 230, 231 6. T. Binoth and G. Heinrich, Nucl. Phys. B 693 (2004) 134. 231 7. N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 8. P. Breitenlohner and D. Maison, Commun. Math. Phys. 52 (1977) 11, 39, 55. 224, 225 9. K.G. Chetyrkin and V.A. Smirnov, Teor. Mat. Fiz. 56 (1983) 206. 229 10. K.G. Chetyrkin and V.A. Smirnov, Phys. Lett. B 144 (1984) 419. 224 11. A. Gehrmann-De Ridder, T. Gehrmann and G. Heinrich, Nucl. Phys. B 682, 265 (2004). 231 12. I.M. Gel’fand and G.E. Shilov, Generalized Functions, Vol. 1 (Academic Press, New York, London, 1964). 227, 228, 231 13. G. Heinrich, Nucl. Phys. Proc. Suppl. 116, 368 (2003). 231 14. K. Hepp, Commun. Math. Phys. 2 (1966) 301. 223, 229 15. N. Nakanishi, Graph Theory and Feynman Integrals (Gordon and Breach, New York, 1971). 16. K. Pohlmeyer, J. Math. Phys. 23 (1982) 2511. 224, 225 17. V.A. Smirnov, Commun. Math. Phys. 134 (1990) 109. 229 18. V.A. Smirnov, Renormalization and Asymptotic Expansions (Birkh¨ auser, Basel, 1991). 224, 225, 229 19. V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses (Springer, Berlin, Heidelberg, 2002). 229 20. V.A. Smirnov, Phys. Lett. B 491 (2000) 130; B 500 (2001) 330; B 524 (2002) 129; B 567 (2003) 193; hep-ph/0406052; G. Heinrich and V.A. Smirnov, hepph/0406053. 231 21. E.R. Speer, J. Math. Phys. 9 (1968) 1404. 229 22. E.R. Speer, Ann. Inst. H. Poincar´e 23 (1977) 1. 224, 225, 227, 229 23. J.B. Tausk, Phys. Lett. B 469 (1999) 225. 231 24. O.I. Zavialov, Renormalized Quantum Field Theory (Kluwer Academic, Dodrecht, 1990). 225, 229

F A Brief Review of Some Other Methods

In this appendix, some methods which were not considered in Chaps. 3–7 are brieﬂy reviewed. The method based on dispersion relations was successfully used from the early days of quantum ﬁeld theory. The Gegenbauer Polynomial x-Space Technique [13], the method of gluing [15] and the method based on star-triangle uniqueness relations [16, 23, 36] are methods for evaluating massless diagrams. The method of IR rearrangement [38], also in a generalized version based on the R∗ -operation [14, 34], is a method oriented at renormalization-group calculations. The recently developed method of diﬀerence equations [27] is also brieﬂy described. It is not analytical, although based on non-trivial mathematical analysis. It enables us to obtain numerical results with extremely high precision, with hundreds of digits. Finally, some methods which could be characterized as based on experimental mathematics are discussed. In particular, this is the integer relation algorithm called PSLQ [18] which provides the possibility to obtain a result for a given one-scale Feynman integral, when we strongly suspect that it is a linear combination of some transcendental numbers with rational coeﬃcients, provided we know the result numerically with a high accuracy.

F.1 Dispersion Integrals A given propagator scalar Feynman integral can be written as ∞ 1 ∆F (s) , ds F (q 2 ) = 2πi s0 s − q 2 − i0

(F.1)

where the discontinuity ∆F (s) = 2i Im(F (s + i0)) is given, according to Cutkosky rules, by a sum over cuts in a given channel of integrals, where the propagators i/(k 2 − m2 + i0) in the cut are replaced by 2πi θ(k0 )δ(k 2 − m2 ), while the propagators to the left of the cut stay the same, and the propagators to the right of the cut change the causal prescription and become −i/(k 2 − m2 − i0). Let us again consider our favourite example of Fig. 1.1, with the indices equal to one. This time, let us include al the necessary factors of i from each V.A. Smirnov: Evaluating Feynman Integrals STMP 211, 233–244 (2004) c Springer-Verlag Berlin Heidelberg 2004

234

F A Brief Review of Some Other Methods

propagator and the factor −i corresponding to the deﬁnition of the Feynman integral with i on the right-hand side of (2.3). We have 2 2 dd k θ(k0 )δ(k 2 − m2 )θ(q0 − k0 )δ[(q − k)2 ] ∆F (q ) = 4π ! " 2 q0 q02 − m2 2π 2 d−2 2 Ωd−1 dr r δ −r = q0 2q0 0 =

24−d π (d+3)/2 (q 2 − m2 )d−3 + , Γ ((d − 1)/2) (q 2 )(d−2)/2

(F.2)

where X+ = X for X > 0 and X+ = 0 otherwise, as usual. We have chosen q = (q0 , 0) and introduced (d − 1)-dimensional spherical coordinates with the surface of the unit sphere in d dimensions equal to 2π d/2 . Γ (d/2) For d = 4, this gives Ωd =

∆F (s) =

2π 3 (q 2 − m2 )+ . q2

(F.3)

(F.4)

Integrating from the threshold s0 = m2 in the dispersion integral (F.1) (where a subtraction is needed) leads to the ﬁnite part of (1.7) (where the factors of i mentioned above were dropped) up to a renormalization constant. In this calculation, a phase-space integral corresponding to a two-particle cut with the masses m and 0 was evaluated. The evaluation of three- and four-particle phase-space integrals is much more complicated. Although we have less integrations in integrals corresponding to cuts, because of the δfunctions, resulting integrals are still rather nasty so that the evaluation of Feynman integrals via their imaginary part by means of Cutkosky rules (see [29] for a typical example) was successful only up to some complexity level. On the other hand, the phase-space integrals are needed for the calculation of the real radiation. It has turned out that the development of methods of evaluating Feynman integrals resulted in similar techniques for the phase-space integrals. Now, one applies, for the evaluation of the phase-space integrals, the strategy of the reduction to master integrals, using IBP, and DE applied for the evaluation of the master integrals – see, e.g., [1, 2]. Moreover, the technique of the sector decompositions of [7] (see Sect. E.2) is also applicable here and was successfully applied in NNLO calculations – see references in the end of Appendix E.

F.2 Gegenbauer Polynomial x-Space Technique The Gegenbauer polynomial x-space technique (GPXT) [13] is based on the SO(d) symmetry of Euclidean Feynman integrals. According to (A.40), the dimensionally regularized scalar massless propagator in coordinate space is

F.3 Gluing

DF (x1 − x2 ) =

1 (2π)d

dd q

e−ix·q Γ (1 − ε) = , q2 4π d/2 [(x1 − x2 )2 ]1−ε

235

(F.5)

where x2 = x20 + x2 . It can be expanded in Gegenbauer polynomials [17] as 1 1 = 2λ [(x1 − x2 )2 ]λ (max{|x1 |, |x2 |}) n/2 ∞ min{|x1 |, |x2 |} × Cnλ (ˆ x1 · x ˆ2 ) , (F.6) max{|x1 |, |x2 |} n=0 √ where |x| = x2 , λ = 1 − ε and x ˆ = x/|x|. The polynomials Cnλ are orthogonal on the unit sphere [17]: λ λ δn,m Cnλ (ˆ x1 · x ˆ2 ) Cm (ˆ x2 · x ˆ3 ) = x1 · x ˆ3 ) . (F.7) dˆ x2 Cnλ (ˆ n+λ The normalization is such that dˆ x = 1. So, the strategy of GPXT is to turn to coordinate space, represent each propagator by (F.6), evaluate integrals over angles by (F.7) and sum up resulting multiple series. First results for non-trivial multiloop diagrams within dimensional regularization were obtained by GPXT: for example, the value of the non-planar diagram (see the second diagram of Fig. 5.6 with all the powers of the propagators equal to one), with the famous result proportional to 20ζ(5) [13]. The GPXT as well as the method of gluing (see below) were crucial in many important analytical calculations, for example, of the three-loop ratio R(s) in QCD [12] and the ﬁve-loop β-function in the φ4 theory [11]. More details on the GPXT can be found in the review [25].

F.3 Gluing The dependence of an h-loop dimensionally regularized scalar propagator massless Feynman integral corresponding to a graph Γ on the external momentum can easily be found by power counting: h (F.8) FΓ (q; d) = iπ d/2 CΓ (ε)(q 2 )ω/2−hε , where ω is the degree of divergence given by (2.9) and CΓ (ε) is a meromorphic function which is ﬁnite at ε = 0 if the integral is convergent, both in the UV and IR sense. (Of course, there are no collinear divergences in propagator integrals.) It turns out that the values CΓ (0) are the same for graphs connected by some transformations based on gluing. The gluing can be of two types: by vertices and by lines. Let Γ be a graph with two external vertices. Let us denote by Γˆ the graph obtained from it by identifying these vertices, and by Γ¯ the graph obtained from it by adding a new line which connects them. Then the following properties hold [15]:

236

F A Brief Review of Some Other Methods

– Gluing by vertices. Let us suppose that two UV- and IR-convergent graphs, Γ1 and Γ2 , have degrees of divergence ω1 = ω2 = −4 and that Γˆ1 and Γˆ2 are the same. Then CΓ1 (0) = CΓ2 (0). – Gluing by lines. Let us suppose that two UV- and IR-convergent graphs, Γ1 and Γ2 , have degrees of divergence ω1 = ω2 = −2 and that Γ¯1 and Γ¯2 are the same. Then CΓ1 (0) = CΓ2 (0). For example, the ﬁrst and the second diagrams in Fig. 5.6 with all the indices equal to one produce the same graph after the gluing the external vertices. It is shown in Fig. F.1. Therefore, one could obtain the value of the more complicated non-planar diagram (proportional to 20ζ(5)) from a simpler planar diagram [15].

Fig. F.1. The graph Γˆ obtained by gluing of vertices

The method of gluing was successfully applied in the combination with GPXT – see the references above.

F.4 Star-Triangle Relations The method based on star-triangle uniqueness relations can be applied to massless diagrams. As in the case of GPXT, the coordinate space language is used, where the propagators have the form 1/(x2 )λ up to a coeﬃcient depending on ε – see, e.g., (F.5). The basic uniqueness relation [16, 36] connects diagrams with diﬀerent numbers of loops. It is graphically shown in Fig. F.2, where λi = d/2 − λi and Γ (d/2 − λi ) v(λ1 , λ2 , λ3 ) = π d/2 . (F.9) Γ (λi ) i This equation holds when the vertex on the left-hand side is unique, i.e. λ1 +λ2 +λ3 = d. The triangle on the right-hand side, with λ1 +λ2 +λ3 = d/2, is also called unique. Remember that, in coordinate space, the triangle diagram does not involve integration and is just a product of the three propagators, [(x1 − x2 )2 ]−λ3 [(x2 − x3 )2 ]−λ1 [(x3 − x1 )2 ]−λ2 , while the star diagram is an integral over the coordinate corresponding to the central vertex.

F.5 IR Rearrangement and R∗

237

λ1 λ2

λ3

=

v(λ1 , λ2 , λ3 ) ×

λ3

λ2 λ1

Fig. F.2. Uniqueness equation

The relation (F.9) can be used to simplify a given diagram. Almost unique relations introduced in [35], with λ1 +λ2 +λ3 = d−1, can be also useful. Sometimes one introduces an auxiliary analytic regularization, to satisfy (almost) unique relations, which can be switched oﬀ in the end of the calculation. For example, using (almost) unique relations, the general ladder massless scalar propagator diagram with an arbitrary number of loops, h, with all the indices ai equal to one (see the ﬁrst diagram of Fig. 5.6 and imagine a general number of rungs), was evaluated [5] with a result proportional to ζ(2h − 1). Another example of applications of the uniqueness relations is the evaluation of the diagram of Fig. 4.14 where they were coupled with functional equations [23]. In this calculation, the initial problem was reduced to the problem of expansion of the propagator diagram of Fig. 3.9 with the indices a1 = . . . = a4 = 1, , a5 = 1 + λ in a Taylor series in λ up to λ4 . This diagram, at various indices, was investigated in many papers starting from the old result for all indices equal to one [33] which was later reproduced [13] by GPXT, an analytical result for this diagram with general values of the indices a1 and a2 and other integer indices [13], an analysis of this diagram from the group-theoretical point of view [9], an extension of the previous results with the help of GPXT [24], etc. As a more recent paper, with updated references to the previous works, let us cite [6], where the expansion of this diagram at indices ai = ni + hi ε, with integer hi , in ε was further studied.

F.5 IR Rearrangement and R∗ The method of IR rearrangement is a special method for the evaluation of UV counterterms which are necessary to perform renormalization. The counterterms are introduced into the Lagrangian, i.e. the dependence of the bare parameters (coupling constants, masses, etc.) of a given theory on a regularization parameter (e.g., d within dimensional regularization) is adjusted in such a way that the renormalized physical quantities become ﬁnite when the regularization is removed. The renormalization can be described at the diagrammatic level, i.e. the renormalized Feynman integrals can be obtained by applying the so-called R-operation which removes the UV divergence from

238

F A Brief Review of Some Other Methods

individual Feynman integrals. Thus, for any R-operation, the quantity RFΓ is UV ﬁnite at d = 4. As is well known, the requirement for the R-operation to be implemented by inserting counterterms into the Lagrangian leads to the following structure [8]: ∆(γ1 ) . . . ∆(γj )FΓ ≡ R FΓ + ∆(Γ ) FΓ , (F.10) RFΓ = γ1 ,...,γj

where ∆(γ) is the corresponding counterterm operation, and the sum is over all sets {γ1 , . . . , γj } of disjoint UV-divergent 1PI subgraphs, with ∆(∅) = 1. The ‘incomplete’ R-operation R , by deﬁnition, includes all the counterterms except the overall counterterm ∆(Γ ). For example, if a graph is primitively divergent, i.e. does not have divergent subgraphs, the R-operation is of the form RFΓ = [1 + ∆(Γ )] FΓ . The action of the counterterm operations is described by ∆(γ) FΓ = FΓ/γ ◦ Pγ ,

(F.11)

where FΓ/γ is the Feynman integral corresponding to the reduced graph Γ/γ, and the right-hand side of (F.11) denotes the Feynman integral that diﬀers from FΓ/γ by insertion of the polynomial Pγ in the external momenta and internal masses of γ into the vertex vγ to which the subgraph γ was reduced. The degree of each Pγ equals the degree of divergence ω(γ). It is implied that a UV regularization is present in (F.10) and (F.11) because these quantities are UV-divergent. The coeﬃcients of the polynomial Pγ are connected in a straightforward manner with the counterterms of the Lagrangian. A speciﬁc choice of the counterterm operations for the set of the graphs of a given theory deﬁnes a renormalization scheme. In the framework of dimensional renormalization, i.e. renormalization schemes based on dimensional regularization, the polynomials Pγ have coeﬃcients that are linear combinations of pure poles in ε = (4 − d)/2. In the minimal subtraction (MS) scheme [21], these polynomials are deﬁned recursively by equations of the form ˆ ε R Fγ Pγ ≡ ∆(γ) Fγ = −K

(F.12)

ˆ ε is the operator that picks up for the graphs γ of the given theory. Here K the pole part of the Laurent series in ε. The modiﬁed MS scheme [4] (MS) is obtained from the MS scheme by the replacement µ2 → µ2 eγE /(4π) for the massive parameter of dimensional regularization that enters through the factors of µ2ε per loop. If Γ is a logarithmically divergent diagram the corresponding counterterm is just a constant. To simplify its calculation it is tempting to put to zero the masses and external momenta. This is, however, a dangerous procedure because it can generate IR divergences. Consider, for example, the two-loop graph of Fig. F.3a. It contributes to the mass renormalization in the φ4 theory. To evaluate the corresponding counterterm it is necessary to compute R Fγ ,

F.5 IR Rearrangement and R∗

q

q

q

239

q

q (a)

(b)

(c)

Fig. F.3. (a) A two-loop graph contributing to the mass renormalization. (b) A possible IR rearrangement. (c) A three-loop graph contributing to the β-function

according to (F.12). Here R = 1+∆1 , where ∆1 is the counterterm operation for the logarithmically divergent subgraph of Fig. F.3a. We consider each of the two resulting terms separately. The last term is simple. The ﬁrst one is just the pole part of the given diagram. If we put the mass to zero we shall obtain an IR divergence. There is another option which is safe: we put the mass to zero and let the external momentum q ﬂow in another way through the graph: from the bottom vertex, rather than from the right vertex – see Fig. F.3b. Then the resulting Feynman integral is IR-convergent and, at the same time, much simpler because it is now recursively one-loop and can be evaluated in terms of gamma functions. This is a simple example of the trick called IR rearrangement and invented in [38]. In a general situation, one tries to put as many masses and external momenta to zero as possible and, probably, let the external momentum ﬂow through the graph in such a way that the resulting diagram is IR-convergent and simple for calculation. Consider now the three-loop graph of Fig. F.3c contributing to the β-function in the φ4 theory. It is also logarithmically divergent. When calculating its counterterm, it is dangerous to put the masses to zero and let the external momentum ﬂow from the bottom to the top vertex, because we run into IR divergences either due to the left or the right pair of the lines. Still there is a possibility not to generate IR divergences: to put the masses of the central loop and the external momentum to zero. The resulting three-loop Feynman integral is evaluated in terms of gamma functions, ﬁrst, by integrating the massless subintegral by (A.7) and then by (A.38). At a suﬃciently high level, such a safe IR rearrangement is not always possible. However, there is a way to put as many masses and momenta to zero and still have control on IR divergences. Formally, we have ˆ ε R∗ Fγ (q) , Pγ = −K

(F.13)

where it is implied that all the masses are put to zero, and one external momentum is chosen to ﬂow through the diagram in an appropriate way. (Another version is to put all the external momenta to zero and leave one non-zero mass.)

240

F A Brief Review of Some Other Methods

The operation R∗ removes not only UV but also (oﬀ-shell) IR divergences in a similar way [14], i.e. by a formula which generalizes (F.10). Now, it ˜ includes IR counterterms ∆(γ) which are deﬁned in a full analogy to the UV counterterms ∆(γ). They are deﬁned for subgraphs irreducible in the IR sense, with the IR degree of divergence given by (2.17). Now, they are local in momentum space. For example, the IR counterterm corresponding to the logarithmically divergent (in the IR sense, i.e. with the IR degree of divergence ω ˜ (γ) = 0) factor 1/(k 2 )2 for the two lower lines in Fig. F.3a (when they are massless) is proportional to δ (d) (k)/ε. More details on the R∗ -operation can be found in [34]. So, according to (F.13), one can safely put to zero all the momenta and masses but one, in a way which is the simplest for the calculation, at the cost of generating IR divergences which should be removed with the help of IR counterterms. Finally, the problem of the evaluation of the UV counterterms for graphs with positive degrees of divergence can be reduced, by diﬀerentiating in momenta and masses, to the case ω = 0. The R∗ -operation was successfully applied in renormalization group calculations – see, e.g., [11].

F.6 Diﬀerence Equations A new method based on diﬀerence equations has recently appeared. Basic prescriptions of this method can be found in [27] and an informal introduction in [28]. It is analytical in nature but is used to obtain numerical results with extremely high precision. The starting point of this approach is to choose a propagator, in an arbitrary way, treat its power, n, as the basic integer variable and ﬁx other powers of the propagators (typically, equal to one). Then the general Feynman integral (5.73) of a given family is written as H , (F.14) F (n) = · · · dd k1 . . . dd kh n E1 E2 . . . EN where H is a numerator. After combining various IBP relations, one can obtain a diﬀerence equation for F (n): c0 (n)F (n) + c1 (n)F (n + 1) + . . . + cr (n)F (n + r) = G(n) ,

(F.15)

where the right-hand side contains Feynman integrals F1 , F2 , . . . which have one or more denominators E2 , E3 , . . . less with respect to (F.14). These integrals are treated in a similar way, by means of equations of the type (F.15) so that one obtains a triangular system of diﬀerence equations. This system is solved, starting from the simplest integrals that have the minimum number of denominators, with the help of an Ansatz in the form of a factorial series, µn

∞ l=0

bl n! , Γ (n − K + l + 1)

(F.16)

F.7 Experimental Mathematics and PSLQ

241

where the values of parameters µ, bl and K are obtained from these values for the factorial series corresponding to the right-hand side of (F.15). This method was successfully applied, with a precision of several dozens up to hundreds of digits, to the calculation of various multiloop Feynman integrals [26, 27]. Observe that, although this method is numerical, it requires serious mathematical eﬀorts. The same feature holds for any modern method of numerical evaluation. One can say that the boarder between analytical and numerical methods becomes rather vague at the moment. Remember about new results obtained in terms of new functions discussed in the end of Chap. 7 – in a narrow sense, these new functions can be regarded as tools to obtain numerical results at various points. Another numerical method based on non-trivial mathematical analysis was described in Sect. E.2. For completeness, here are some references to modern methods of numerical evaluation of Feynman integrals: [30, 31, 32]. Observe that such methods are often called semianalytical. Sometimes it is claimed that sooner or later we shall achieve the limit in the process of analytical evaluation of Feynman integrals so that we shall be forced to proceed only numerically (see, e.g., [30]). However, the dramatic progress in the ﬁeld of analytical evaluation of Feynman integrals shows that we have not yet exhausted our abilities. So, the natural strategy is to combine available analytical and numerical methods in an appropriate way.

F.7 Experimental Mathematics and PSLQ When evaluating Feynman integrals, various tricks are used. One usually does not bother about mathematical proofs of the tricks, partially, because of the pragmatical orientation and strong competition and, partially, because, now, there are a lot of possibilities to check obtained results, both in the physical and mathematical way. An example of such ‘experimental mathematics’ suggested in [20] was described in Sect. 4.5, where it was supposed that the nth coeﬃcient of the Taylor series cn of a piece of the result for the master massive double box is a linear combination of the 15 functions (4.62)–(4.65) of the variable n. Then the possibility to evaluate the ﬁrst 15 coeﬃcients c1 , c2 , . . . , c15 was used and the corresponding linear system for unknown coeﬃcients in the given linear combination was solved. At this point, a pure mathematician could say that there is no mathematical proof of this procedure and its validity is not guaranteed at all even after we (successfully) check it by calculating more terms of the Taylor expansion, starting from the 16th and comparing it with what we have from the obtained solution. Still I believe that this pure mathematician will believe in the result when he/she looks at some details of the calculation. Indeed, suppose that we forget about just one of the functions in (4.62)–(4.65) and follow our procedure. Then we indeed obtain a diﬀerent solution of our system of 14 equations but it blows up and

242

F A Brief Review of Some Other Methods

looks so ugly, in terms of rational numbers with hundreds of digits in the numerator and denominator, that this pure mathematician will say that our previous solution, with nice rational numbers, is true and there is no need for mathematical proofs. Of course, an important point here is to understand what we can expect in the result. Another example is given by taking a sum when going from (4.94) to (4.95) when evaluating the diagram of Fig. 4.14. Instead of using SUMMER [39], we can suppose that the general term of the Taylor series (4.95) is a linear combination, with unknown coeﬃcients, of (4.62)–(4.65) and similar terms up to level 7. (For example, at level 7, one can use the structures with a 1/n2 dependence present on the left-hand side of (C.51)–(C.82).) Then one obtains a system of 63 linear equations for these coeﬃcients and solves it using information about the ﬁrst 63 terms which can be obtained from the two-fold series following from (4.94). There are a lot of other elements of experimental mathematics in dealing with Feynman integrals. Indeed, we never hesitate to change the order of integration over alpha and Feynman parameters and over MB parameters, it is not known in advance which IBP equations within the algorithm formulated in [27] are really independent, there is no mathematical justiﬁcation of the prescriptions of Chap. 6, etc. One more example of experimental mathematics1 is provided by the so-called PSLQ algorithm [18]. It can be applied when we evaluate a one-scale Feynman integral in expansion in ε. Let us suppose that, in a given order of expansion in ε, we understand which transcendental numbers can appear in the result and that we can obtain the result numerically with a high accuracy. For example, in the ﬁnite part of the ε-expansion in two loops we can expect at least xi−1 = ζ(i) with i = 2, 3, 4 or, equivalently, x1 = π 2 , x2 = ζ(3) and x3 = π 4 . Then the PSLQ algorithm could be of use. In this particular example, it gives the possibility to estimate whether or not a given number, x can be expressed linearly as x = c1 x1 + c2 x2 + c3 x3 with rational coeﬃcients ci . The PSLQ is an example of an ‘integer relation algorithm’. If x1 , x2 , · · · , xn are some real numbers, it gives the possibility to ﬁnd the n integers ci such that c1 x1 +c2 x2 +· · ·+cn xn = 0 or provide bounds within which this relation is impossible. (In the above situation, we consider our numerical result as x4 , in addition to the xi , i = 1, 2, 3.) More formally, suppose that xi are given with the precision of ν decimal digits. Then we have an integer relation with the norm bound N if |c1 x1 + . . . + cn xn | < ε ,

(F.17)

provided that max|ci | < N , where ε > 0 is a small number of order 10−ν . With a given accuracy ν, a detection threshold ε and a norm bound N as an input,

1

The very term ‘experimental mathematics’ can be found on the web page where, in particular, the PSLQ algorithm is described [39].

References

243

the PSLQ algorithm enables us to ﬁnd out whether the relation (F.17) exists or not at some conﬁdence level (see details in [18]). The PSLQ algorithm has been successfully applied in the evaluation of various single-scale Feynman integrals – see, e.g., [3, 10, 19, 22]. The experience obtained in these calculations shows that one needs around ten digits for each independent transcendental number.

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29.

C. Anastasiou and K. Melnikov, Nucl. Phys. B 646 (2002) 220. 234 C. Anastasiou and K. Melnikov, Phys. Rev. D 67, 037501 (2003) 234 D.H. Bailey and D.J. Broadhurst, Math. Comput. 70 (2001) 1719. 243 W.A. Bardeen, A.J. Buras, D.W. Duke and T. Muta, Phys. Rev. D 18 (1978) 3998. 238 V.V. Belokurov and N.I. Ussyukina, J. Phys. A 16 (1983) 2811. 237 I. Bierenbaum and S. Weinzierl, Eur. Phys. J. C 32 (2003) 67 237 T. Binoth and G. Heinrich, Nucl. Phys. B 585 (2000) 741; 680 (2004) 375. 234 N.N. Bogoliubov and D.V. Shirkov, Introduction to Theory of Quantized Fields, 3rd edition (Wiley, New York, 1983). 238 D.J. Broadhurst, Z. Phys. C 32 (1986) 249; D.T. Barfoot and D.J. Broadhurst, Z. Phys. C 41 (1988) 81. 237 D.J. Broadhurst, Eur. Phys. J. C 8 (1999) 311 243 K.G. Chetyrkin, S.G. Gorishnii, S.A. Larin and F.V. Tkachov, Phys. Lett. B 132, 351 (1983). 235, 240 K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Phys. Lett. B 85 (1979) 277. 235 K.G. Chetyrkin, A.L. Kataev and F.V. Tkachov, Nucl. Phys. B 174 (1980) 345. 233, 234, 235, 237 K.G. Chetyrkin and V.A. Smirnov, Phys. Lett. B 144 (1984) 419. 233, 240 K.G. Chetyrkin and F.V. Tkachov, Phys. Lett. B 192 (1981) 159. 233, 235, 236 M. D’Eramo, L. Peliti and G. Parisi, Lett. Nuovo Cim. 2 (1971) 878. 233, 236 A. Erd´elyi (ed.), Higher Transcendental Functions, Vols. 1 and 2 (McGraw-Hill, New York, 1954). 235 H.R.P. Ferguson and D.H. Bailey, RNR Technical Report, RNR-91-032; H.R.P. Ferguson, D.H. Bailey and S. Arno, NASA Technical Report, NAS96-005. 233, 242, 243 J. Fleischer and M. Y. Kalmykov, Phys. Lett. B 470 (1999) 168; Comput. Phys. Commun. 128 (2000) 531. 243 J. Fleischer, A.V. Kotikov and O.L. Veretin, Nucl. Phys. B 547 (1999) 343. 241 G. ’t Hooft, Nucl. Phys. B 61 (1973) 455. 238 M.Yu. Kalmykov and O. Veretin, Phys. Lett. B 483 (2000) 315. 243 D.I. Kazakov, Theor. Math. Phys. 58 (1984) 223 [Teor. Mat. Fiz. 58 (1984) 343]; 62, 84 (1985) [Teor. Mat. Fiz. 62, 127 (1984)]. 233, 237 A.V. Kotikov, Phys. Lett. B 375 (1996) 240. 237 A.V. Kotikov, hep-ph/0102177. 235 S. Laporta, Phys. Lett. B 504, 351 (1983); B 523 (2001) 95; B 549 (2002) 115 241 S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087. 233, 240, 241, 242 S. Laporta, Acta Phys. Polon. B 34 (2003) 5323. 240 W.L. van Neerven, Nucl. Phys. B 268 (1986) 453. 234

244

F A Brief Review of Some Other Methods

30. G. Passarino, Nucl. Phys. B 619 (2001) 257. 241 31. G. Passarino and S. Uccirati, Nucl. Phys. B 629 (2002) 97; A. Ferroglia, G. Passarino, S. Uccirati and M. Passera, Nucl. Instrum. Meth. A 502 (2003) 391; A. Ferroglia, M. Passera, G. Passarino and S. Uccirati, Nucl. Phys. B 680 (2004) 199. 241 32. A. Ghinculov and Y. Yao, Phys. Rev. D 63 (2001) 054510; Nucl. Phys. B 516 (1998) 385. 241 33. J.L. Rosner, Ann. Phys. 44 (1967) 11. 237 34. V.A. Smirnov, Renormalization and Asymptotic Expansions (Birkh¨ auser, Basel, 1991). 233, 240 35. N.I. Ussyukina, Teor. Mat. Fiz. 54 (1983) 124. 237 36. A.N. Vassiliev, Yu.M. Pis’mak and Yu.R. Honkonen, Teor. Mat. Fiz. 47 (1981) 291. 233, 236 37. J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037. 38. A.A. Vladimirov, Teor. Mat. Fiz. 43 (1980) 210. 233, 239 39. http://www.cecm.sfu.ca 242

List of Symbols

Aij r – matrix which deﬁnes denominators of the propagators al – power of a propagator (index) ci (a1 , . . . , aN ) – coeﬃcient function of a master integral Ii ˜ F – propagator in coordinate space D DF , DF,i – propagator in momentum space d – space-time dimension Er – denominator of propagator FΓ – Feynman integral 2 F1 (a, b; c; z) – Gauss hypergeometric function G(λ1 , λ2 ) – function in one-loop massless integration formula gµν – metric tensor Ha1 ,a2 ,...,an (x) – harmonic polylogarithm (HPL) h – number of loops Ii – master integral k – loop momentum L – number of lines Lia (z) – polylogarithm l – loop momentum m – mass P (x1 , . . . , xN ) – basic polynomial p – external or internal momentum Q2 = −q 2 – Euclidean external momentum squared q – external momentum Sa,b (z) – generalized polylogarithm Sj , Sjk ,. . . – nested sums s = (p1 + p2 )2 – Mandelstam variable

T – tree, 2-tree, pseudotree t = (p1 + p3 )2 – Mandelstam variable tl – sector variable U – function in the alpha representation u = (p1 + p4 )2 – Mandelstam variable ul – auxiliary parameter V – number of vertices V – function in the alpha representation w – variable in MB integrals x – coordinate xi – variable in the basic parametric representation Zl – polynomial in propagator z, zi – variable in MB integrals αl – alpha parameter βl = 1/αl – inverse alpha parameter Γ – graph Γ (x) – gamma function (ﬁrst Euler integral) γ – subgraph γE = 0.577216 . . . – Euler’s constant δ(x) – delta function ε = (4 − d)/2 – parameter of dimensional regularization ζ(z) – Riemann zeta function λl – parameter of analytic regularization ξ, ξi – Feynman parameter τl – sector variable ψ(x) = Γ (z)/Γ (z) – logarithmical derivative of the gamma function ω – degree of UV divergence

Index

alpha parameters 15 auxiliary master integral Baikov’s method

method of diﬀerence equations 240 method of diﬀerential equations (DE) 7, 165 momentum Euclidean 225 external 12 internal 12 loop 12

145

133

Cheng–Wu theorem

42

degree of UV divergence dispersion integral 233 divergence 14 collinear 17 IR 16 on-shell IR 17 threshold IR 17 UV 14

14

nested sums

partial fractions 35 Pochhammer symbol polylogarithm 187 propagator 11 PSLQ 241

Feynman amplitude 12 Feynman integral 12 Feynman parameters 41 ﬁrst Barnes lemma 207

188

index (power of a propagator) 11 integer relation algorithm 242 integration by parts (IBP) 2, 65, 109 IR rearrangement 237 left poles

187

recursively one-loop diagrams regularization 20 analytic 21 dimensional 22, 23 Pauli–Villars 21 Riemann zeta function 191 right poles 56

Gauss hypergeometric function 187 Gegenbauer polynomial x-space technique (GPXT) 234 generalized polylogarithm 187 gluing 235 graph 12 harmonic polylogarithm (HPL)

191

second Barnes lemma 214 sectors 223 shifting dimension 36, 120 subgraph detachable 24 divergent 15 one-particle-irreducible (1PI) one-vertex-reducible 224

34

15

56

Mandelstam variables 39 master integral 2, 109, 133 Mellin–Barnes (MB) representation 55, 56

tadpole 24, 28 tree 19 two-dimensional HPL (2dHPL) 4, uniqueness relations

236

174

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close