printed April 14, 2000

CLASSICS ILLUSTRATED

GROUP THEORY
Exceptional Lie groups as invariance groups
Predrag Cvitanovi c

Abstract

We oer the ultimate birdtracker guide to exceptional Lie groups.

Keywords: exceptional Lie groups, invariant theory, Tits magic square PRELIMINARY version of 30 March 2000 available on: www.nbi.dk/GroupTheory/ [email protected]

ii

Contents
1 Introduction 1

2 A preview 5 2.1 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.2 First example: SU (n) . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3 Second example: E6 family . . . . . . . . . . . . . . . . . . . . . . 12 3 Invariants and reducibility 3.1 Preliminaries . . . . . . . . . . . . . . . . . . . 3.1.1 Groups . . . . . . . . . . . . . . . . . . 3.1.2 Vector spaces . . . . . . . . . . . . . . . 3.1.3 Algebra . . . . . . . . . . . . . . . . . . 3.1.4 Dening space, tensors, representations 3.2 Invariants . . . . . . . . . . . . . . . . . . . . . 3.2.1 Algebra of invariants . . . . . . . . . . . 3.3 Invariance groups . . . . . . . . . . . . . . . . . 3.4 Projection operators . . . . . . . . . . . . . . . 3.5 Further invariants . . . . . . . . . . . . . . . . 3.6 Birdtracks . . . . . . . . . . . . . . . . . . . . . 3.7 Clebsch-Gordan coecients . . . . . . . . . . . 3.8 Zero- and one-dimensional subspaces . . . . . . 3.9 Innitesimal transformations . . . . . . . . . . 3.10 Lie algebra . . . . . . . . . . . . . . . . . . . . 3.11 Other forms of Lie algebra commutators . . . . 3.12 Irrelevancy of clebsches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 15 15 16 17 18 20 22 23 24 25 27 29 31 32 36 38 38

4 Recouplings 41 4.1 Couplings and recouplings . . . . . . . . . . . . . . . . . . . . . . . 41 4.2 Wigner 3n j coecients . . . . . . . . . . . . . . . . . . . . . . . 44 4.3 Wigner-Eckart theorem . . . . . . . . . . . . . . . . . . . . . . . . 45 5 Permutations 49 5.1 Permutations in birdtracks . . . . . . . . . . . . . . . . . . . . . . 49 5.2 Symmetrization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 iii

iv 5.3 5.4 5.5 5.6 5.7 5.8 Antisymmetrization . . . . . . Levi-Civita tensor . . . . . . . Determinants . . . . . . . . . . Characteristic equations . . . . Fully (anti)symmetric tensors . Young tableaux, Dynkin labels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

CONTENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 53 55 57 57 58 59 60 60 60 60 60 60 60 60 61 63 64 64 64 65 65 66 68 68 69 70 71 72 73 74 74 75 76 77 77 78 79 79 81 83

6 Casimir operators 6.1 Casimirs and Lie algebra . . . . . . . . . . 6.2 Independent casimirs . . . . . . . . . . . . 6.3 Casimir operators . . . . . . . . . . . . . . 6.4 Dynkin indices . . . . . . . . . . . . . . . 6.5 Quadratic, cubic casimirs . . . . . . . . . 6.6 Quartic casimirs . . . . . . . . . . . . . . 6.7 Sundry relations between quartic casimirs 6.8 Identically vanishing tensors . . . . . . . . 6.9 Dynkin labels . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

7 Group integrals 7.1 Group integrals for arbitrary representations . . . . . . . . . . . . 7.2 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Examples of group integrals . . . . . . . . . . . . . . . . . . . . . . 8 Unitary groups 8.1 Two-index tensors . . . . . . . . . . . . . . . . . . . . 8.2 Three-index tensors . . . . . . . . . . . . . . . . . . . 8.3 Young tableaux . . . . . . . . . . . . . . . . . . . . . . 8.3.1 Denitions . . . . . . . . . . . . . . . . . . . . 8.3.2 SU (n) Young tableaux . . . . . . . . . . . . . . 8.3.3 Reduction of direct products . . . . . . . . . . 8.4 Young projection operators . . . . . . . . . . . . . . . 8.4.1 A dimension formula . . . . . . . . . . . . . . . 8.4.2 Dimension as the number of strand colorings . 8.5 Reduction of tensor products . . . . . . . . . . . . . . 8.5.1 Three- and four-index tensors . . . . . . . . . . 8.5.2 Basis vectors . . . . . . . . . . . . . . . . . . . 8.6 3-j symbols . . . . . . . . . . . . . . . . . . . . . . . . 8.6.1 Evaluation by direct expansion . . . . . . . . . 8.6.2 Application of the negative dimension theorem 8.6.3 A sum rule for 3-j s . . . . . . . . . . . . . . . 8.7 Characters . . . . . . . . . . . . . . . . . . . . . . . . . 8.8 Mixed two-index tensors . . . . . . . . . . . . . . . . . 8.9 Mixed dening adjoint tensors . . . . . . . . . . . . 8.10 Two-index adjoint tensors . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . .

CONTENTS

8.11 Casimirs for the fully symmetric representations of SU (n) . . . . . 84 8.12 SU (n), U (n) equivalence in adjoint representation . . . . . . . . . 84 8.13 Dynkin labels for SU (n) representations . . . . . . . . . . . . . . . 84 9 Orthogonal groups 9.1 Two-index tensors . . . . . . . . . . . . 9.2 Three-index tensors . . . . . . . . . . . 9.3 Mixed dening adjoint tensors . . . . 9.4 Two-index adjoint tensors . . . . . . . . 9.5 Gravity tensors . . . . . . . . . . . . . . 9.6 Dynkin labels of SO(n) representations 10 Spinors 10.1 Spinograpy . . . . . . . 10.2 Fierzing around . . . . . 10.3 Fierz coecients . . . . 10.4 6j coecients . . . . . . 10.5 Exemplary evaluations . 10.6 Invariance of -matrices 10.7 Handedness . . . . . . . 10.8 Kahane algorithm . . . 85 86 86 86 86 86 86 89 90 90 90 90 90 90 90 90

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

11 Symplectic groups 91 11.1 Two-index tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 11.2 Mixed dening adjoint tensors . . . . . . . . . . . . . . . . . . . 93 11.3 Dynkin labels of Sp(n) representations . . . . . . . . . . . . . . . . 93 12 Negative dimensions 95 12.1 SU (n) = SU (n) . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 12.2 SO(n) = Sp(n) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 13 Spinsters 14 SU (n) family of invariance groups 14.1 Representations of SU (2) . . . . . . . . . . . . 14.2 SU (3) as invariance group of a cubic invariant . 14.3 Levi-Civita tensors and SU (n) . . . . . . . . . 14.4 SU (4) - SO(6) isomorphism . . . . . . . . . . . 15 G2 family of invariance groups 15.1 Jacobi relation . . . . . . . . . . . . . . . . . 15.2 Alternativity and reduction of f -contractions 15.3 Primitivity implies alternativity . . . . . . . . 15.4 Casimirs for G2 . . . . . . . . . . . . . . . . . 15.5 Hurwitzs theorem . . . . . . . . . . . . . . . 101 103 . 103 . 105 . 105 . 105 107 109 110 112 115 116

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

vi

CONTENTS 15.6 Representations of G2 . . . . . . . . . . . . . . . . . . . . . . . . . 118

16 E8 family of invariance groups 16.1 Two-index tensors . . . . . . . 16.2 Decomposition of Sym3 A . . . 16.3 Decomposition of |??| |??||??| 16.4 Diophantine conditions . . . . . 16.5 Generalized Young tableaux for 16.6 Conjectures of Deligne . . . . .

. . . . . . . . E8 . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

119 120 123 125 127 127 128 129 129 130 130 130 130 130 130 130 130 131 131 131 131 131 131

17 E6 family of invariance groups 17.1 Reduction of two-index tensors . . . . . . . . . . . 17.2 Mixed two-index tensors . . . . . . . . . . . . . . . 17.3 Diophantine conditions and the E6 family . . . . . 17.4 Three-index tensors . . . . . . . . . . . . . . . . . 17.4.1 Fully symmetric V 3 tensors . . . . . . . . 17.4.2 Mixed symmetry V 3 tensors . . . . . . . . 17.4.3 Fully antisymmetric V 3 tensors . . . . . . 17.5 Dening adjoint tensors . . . . . . . . . . . . . . 17.6 Two-index adjoint tensors . . . . . . . . . . . . . . 17.6.1 Reduction of antisymmetric 3-index tensors 17.7 Dynkin labels and Young tableaux for E6 . . . . . 17.8 Casimirs for E6 . . . . . . . . . . . . . . . . . . . . 17.9 Subgroups of E6 . . . . . . . . . . . . . . . . . . . 17.10Springer relation . . . . . . . . . . . . . . . . . . . 17.10.1 Springers construction of E6 . . . . . . . . 18 F4 family of invariance groups 18.1 Two-index tensors . . . . . . . . 18.2 Dening adjoint tensors . . . . 18.2.1 Two-index adjoint tensors 18.3 Jordan algebra and F4 (26) . . . . 19 E7 family of invariance groups

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

133 . 133 . 136 . 136 . 136 137

20 Exceptional magic 139 20.1 Magic triangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 21 Magic negative dimensions 143 21.1 E7 and SO(4) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 21.2 E6 and SU (3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 A Recursive decomposition 145

CONTENTS B Properties of Young Projections B.1 Uniqueness of Young projection operators B.2 Normalization . . . . . . . . . . . . . . . . B.3 Orthogonality . . . . . . . . . . . . . . . . B.4 The dimension formula . . . . . . . . . . . B.5 Literature . . . . . . . . . . . . . . . . . . Bibliography Index

vii 147 . 147 . 148 . 149 . 150 . 151 153 159

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

viii

CONTENTS

Acknowledgements
I would like to thank Tony Kennedy for coauthoring the work discussed in chapters on spinors, spinsters and negative dimensions; Henriette Elvang for coauthoring the chapter on representations of U (n); David Pritchard for much help with the early versions of this manuscript; Roger Penrose for inventing birdtracks (and thus making them respectable) while I was struggling through grade school; Paul Lauwers for the birdtracks rock-around-the-clock; Feza G ursey and Pierre Ramond for the rst lessons on exceptional groups; Sesumu Okubo for inspiring correspondence; Bob Pearson for assorted birdtrack, Young tableaux and lattice calculations; Bernard Julia for comments that I hope to understand someday (and also why does he not cite my work on the magic triangle?); M. Kontsevich for bringing to my attention the more recent work of Deligne, Cohen and de Man; R. Abdelatif, G.M. Cicuta, A. Duncan, E. Eichten, E. Cremmer, B. Durhuus, R. Edgar, M. G unaydin, K. Oblivia, G. Seligman, A. Springer, L. Michel, P. Howe, R.L. Mkrtchyan, P.G.O. Freund, T. Goldman, R.J. Gonsalves, P. Sikivie, H. Harari, D. Mili ci c, C. Sachrayda, G. Tiktopoulos and B. Weisfeiler for discussions (or correspondence). The appelation birdtracks is due to Bernice Durand who described diagrams on my blackboard as footprints left by birds scurrying along a sandy beach. I am grateful to Dorte Glass for typing most of the manuscript and drawing some of the birdtracks. Carol Monsrud, and Cecile Gourgues helped with typing the early version of this manuscript. The manuscript was written in stages in Chewton-Mendip, Paris, Buressur-Yvette, Rome, Copenhagen, Frebbenholm, Rros, Juelsminde, G oteborg Copenhagen train, Sjllands Odde, G oteborg, Cathay Pacic (Hong Kong Paris), Miramare and Kurkela. I am grateful to T. Dorrian-Smith, R. de la Torre, BDC, N.-R. Nilsson, E. Hsinen, family Cvitanovi c, U. Selmer and family Herlin for their kind hospitality along this long way.

Chapter 1

Introduction
One simple eld-theory question started this project; what is the group theoretic factor for the following QCD gluon self-energy diagram =? (1.1)

I rst computed the answer for SU (n). There was a hard way of doing it, using Gell-Mann fijk and dijk coecients. There was also an easy way, where one could doodle oneself to the answer in a few lines. This is the birdtracks method which will be described here. It works nicely for SO(n) and Sp(n) as well. Out of curiosity, I wanted the answer for the remaining ve exceptional groups. This engendered further thought, and that which I learned can be better understood as the answer to a dierent question. Suppose someone came into your oce and asked, On planet Z , mesons consist of quarks and antiquarks, but baryons contain three quarks in a symmetric color combination. What is the color group? The answer is neither trivial, nor without some beauty (planet Z quarks can come in 27 colors, and the color group can be E6 ). Once you know how to answer such group-theoretical questions, you can answer many others. This monograph tells you how. Like the brain, it is divided into two halves; the plodding half and the interesting half. The plodding half describes how group theoretic calculations are carried out for unitary, orthogonal and symplectic groups. Probably none of that is new, but the methods are helpful in carrying out theorists daily chores, such as evaluating Quantum Chromodynamics group theoretic weights, evaluating lattice gauge theory group integrals, computing 1/N corrections, evaluating spinor traces, evaluating casimirs, implementing evaluation algorithms on computers, and so on. The interesting half describes the exceptional magic (a new construction of exceptional Lie algebras) and the negative dimensions (relations between bosonic and fermionic dimensions). The methods used are applicable to grand unied theories and supersymmetric theories. Regardless of their immediate utility, the results are suciently intriguing to have motivated this entire undertaking. 1

CHAPTER 1. INTRODUCTION

There are two complementary approaches to group theory. In the canonical approach one chooses the basis, or the Clebsch-Gordan coecients, as simply as possible. This is the method which Killing [87] and Cartan [88] used to obtain the complete classication of semi-simple Lie algebras, and which has been brought to perfection by Dynkin [90]. There exist many excellent reviews of applications of Dynkin diagram methods to physics, such as the review by Slansky [71]. In the tensorial approach, the bases are arbitrary, and every statement is invariant under change of basis. Tensor calculus deals directly with the invariant blocks of the theory and gives the explicit forms of the invariants, Clebsch-Gordan series, evaluation algorithms for group theoretic weights, etc. The canonical approach is often impractical for physicists purposes, as a choice of basis requires a specic coordinatization of the representation space. Usually, nothing that we want to compute depends on such a coordinatization; physical predictions are pure scalar numbers (color singlets), with all tensorial indices summed. However, the canonical approach can be very useful in determining chains of subgroup embeddings. We refer reader to the Slansky review [71] for such applications; here we shall concentrate on tensorial methods, borrowing from Cartan and Dynkin only the nomenclature for identifying irreducible representations. Extensive listings of these are given by McKay and Patera [91] and Slansky [71]. To appreciate the sense in which canonical methods are impractical, let us consider using them to evaluate the group-theoretic factor (1.1) for the exceptional group E8 . This would involve summations over 8 structure constants. The Cartan-Dynkin construction enables us to construct them explicitly; an E8 structure constant has about 2483 /6 elements, and the direct evaluation of (1.1) is tedious even on a computer. An evaluation in terms of a canonical basis would be equally tedious for SU (16); however, the tensorial approach (described in the example at the end of this section) yields the answer for all SU (n) in a few steps. This is one motivation for formulating a tensorial approach to exceptional groups. The other is the desire to understand their geometrical signicance. The Killing-Cartan classication is based on a mapping of Lie algebras onto a Diophantine problem on the Cartan root lattice. This yields an exhaustive classication of simple Lie algebras, but gives no insight into the associated geometries. In the 19th century, the geometries, or the invariant theory was the central question and Cartan, in his 1894 thesis, made an attempt to identify the primitive invariants. Most of the entries in his classication were the classical groups SU (n), SO(n) and Sp(n). Of the ve exceptional algebras, Cartan [89] identied G2 as the group of octonion isomorphisms, and noted already in his thesis that E7 has a skew-symmetric quadratic and a symmetric quartic invariant. Dickinson [92] characterized E6 as a 27-dimensional group with a cubic invariant1 . The fact that the orthogonal, unitary and symplectic groups were invariance groups of real, complex and quaternion norms suggested that the exceptional groups were
1

I am indebted to G. Seligman for this reference.

printed April 14, 2000

DasGroup/book/chapter/intro.tex 14apr2000

3 associated with octonions, but it took more than another fty years to establish the connection. The remaining four exceptional Lie algebras emerged as rather complicated constructions from octonions and Jordan algebras, known as the Freudenthal-Tits construction. A mathematicians history of this subject is given in a delightful review by Freudenthal [93]. The subject has twice been taken up by physicists, rst by Jordan, von Neumann and Wigner [63], and then in the 1970s by G ursey and collaborators. Jordan et al.s eort was a failed attempt at formulating a new quantum mechanics which would explain the neutron, discovered in 1932. However, it gave rise to the Jordan algebras, which became a mathematics eld in itself. G ursey et al. took up the subject again in the hope of formulating a quantum mechanics of quark connement; the main applications so far, however, have been in building models of grand unication. Although beautiful, the Freudenthal-Tits construction is still not practical for the evaluation of group-theoretic weights. The reason is this; the construction involves [3 3] octonian matrices with octonian coecients, and the 248 dimensional dening space of E8 is written as a direct sum of various subspaces. This is convenient for studying subgroup embeddings [85], but awkward for grouptheoretical computations. The inspiration for the primitive invariants construction came from the axiomatic approach of Springer [94, 95] and Brown [96]: one treats the dening representation as a single vector space, and characterizes the primitive invariants by algebraic identities. This approach solves the problem of formulating ecient tensorial algorithms for evaluating group-theoretic weights, and also yields some intuition about the geometrical signicance of the exceptional Lie groups. Such intuition might be of use to quark-model builders. For example, because SU (3) has a cubic invariant abc qa qb qc , QCD based on this color group can accommodate 3-quark baryons. Are there any other groups that could accommodate 3-quark singlets? As we shall show, the dening representations of G2 , F4 and E6 are some of the groups with such invariants. Beyond being a mere computational aid, the primitive invariants construction of exceptional groups yields several unexpected results. First, it generates in a somewhat magical fashion a triangular array of Lie algebras, depicted in g. 1.1. This is a classication of Lie algebras dierent from Cartans classication; in particular, all exceptional Lie groups appear in the same series (the bottom line of g. 1.1). The second unexpected result is that many groups and group representations are mutually related by interchanges of symmetrizations and antisymmetrizations, and replacement of the dimension parameter n by n. I call this phenomenon negative dimensions. For me, the greatest surprise of all is that in spite of all the magic and the strange diagrammatic notation, the resulting manuscript is in essence not very dierent from Wigners [2] classic group theory book. Regardless of whether one is doing atomic, nuclear or particle physics, all physical predictions (spectroscopic levels) are expressed in terms of Wigners 3n j coecients, which can be
printed April 14, 2000 DasGroup/book/chapter/intro.tex 14apr2000

CHAPTER 1. INTRODUCTION
0 0 1 1 3 2 8 3

A1 A2 G2 D4 F4 E6

0 0 0 1 2 2 8

U(1) A1

0 0 0 1 3 5 8 6 21 14 52

3 3 9 4

14 7 28 8 52 26 78

0 0 0 2 2 3 9

2U(1) 3A 1
21

0 0 0 0 1 2 8 0 1 3 4 14 14

A1 A2 C3 F4

8 16 9

A2

14 35 15 66 32

C3 A5 D6 E7

2U(1)

2A2 A5 E6

0 0 3 3

35 20

27 133

U(1) A2

A1 G2

8 28 28

3A1 D4

78 78

133 133

56 248 248

E7 E8

A1

52

Figure 1.1:

The magic triangle for Lie algebras. The Freudenthal magic square is marked by the dotted line. The number in the lower left corner of each entry is the dimension of the dening representation. For more details consult chapter 20.

evaluated by means of recursive or combinatorial algorithms.

DasGroup/book/chapter/intro.tex 14apr2000

printed April 14, 2000

Chapter 2

A preview
This report on group theory had mutated greatly throughout its genesis. It arose from concrete calculations motivated by physical problems; but as it was written, the generalities were collected into introductory chapters, and the applications receded later and later into the text. As a result, the rst seven chapters are largely a compilation of denitions and general results which might appear unmotivated on the rst reading. The reader is advised to work through the examples, sect. 2.2 and sect. 2.3 in this chapter, jump to the topic of possible interest (such as the unitary groups, chapter 8, or the E8 family, chapter 16), and backtrack when necessary. The goal of these notes is to provide the reader with a set of basic grouptheoretic tools. They are not particularly sophisticated, and they rest on a few simple ideas. The text is long because various notational conventions, examples, special cases and applications have been laid out in detail, but the basic concepts can be stated in a few lines. We shall briey state them in this chapter, together with several illustrative examples. This preview presumes that the reader has considerable prior exposure to group theory; if a concept is unfamiliar, the reader is referred to the appropriate section for a detailed discussion.

2.1

Basic concepts

An average quantum theory is constructed from a few building blocks which we shall refer to as the dening representation. They form the dening multiplet of the theory - for example, the quark wave functions qa . The group-theoretical problem consists of determining the symmetry group, ie. the group of all linear transformations qa = Ga b q b a, b = 1, 2, . . . , n ,

which leave invariant the predictions of the theory. The [n n] matrices G form the dening representation of the invariance group G . The conjugate multiplet 5

6 (antiquarks) transforms as q a = Ga b q b .

CHAPTER 2. A PREVIEW

Combinations of quarks and antiquarks transform as tensors, such as

ef d p a q b r c = Ga c b , d pf q e r , ef Ga c b, d

= Gaf Gbe Gcd .

(see sect. 3.1.4). Tensor representations are plagued by a proliferation of indices. These indices can either be replaced by a few collective indices

c ab

, q ,

ef d

, (2.1)

q = G

or represented diagrammatically

(Diagrammatic notation is explained in sect. 3.6). Collective indices are convenient for stating general theorems; diagrammatic notation speeds up explicit calculations. A polynomial H (q, r, s, . . .) = hab... ...c q a rb . . . sc is an invariant if (and only if) for any transformation G G and for any set of vectors q, r, s, . . . (see sect. 3.3) H (Gq, Gr, Gs, . . .) = H (q, r, s, . . .) . (2.2)

An invariance group is dened by its primitive invariants, ie. by a list of the elementary singlets of the theory. For example, the orthogonal group O(n) is dened as the group of all transformations which leave the length of a vector invariant (see chapter 9). Another example is the color SU (3) of QCD which leaves invariant the mesons (q q ) and the baryons (qqq ) (see sect. 14.2). A complete list of primitive invariants denes the invariance group via the invariance conditions (2.2); only those transformations which respect them are allowed. It is not necessary to list explicitly the components of primitive invariant tensors in order to dene them. For example, the O(n) group is dened by the requirement that it leaves invariant a symmetric and invertible tensor gab = gba , det(g ) = 0. Such denition is basis independent, while a component denition g11 = 1, g12 = 0, g22 = 1, . . . relies on a specic basis choice. We shall dene all simple Lie groups in this manner, specifying the primitive invariants only by
DasGroup/book/chapter/preview.tex 14apr2000 printed April 14, 2000

2.1. BASIC CONCEPTS

their symmetry, and by the basis-independent algebraic relations that they must satisfy. These algebraic relations (which we shall call primitiveness conditions) are hard to describe without rst giving some examples. In their essence they are a, statements of irreducibility: for example, if the primitive invariant tensors are b e , as otherwise the dening habc and habc , then habc hcbe must be proportional to a representation would be reducible. (Reducibility is discussed in sect. 3.4, sect. 3.5 and chapter 4). The objective of physicists group-theoretic calculations is a description of the spectroscopy of a given theory. This entails identifying the levels (irreducible multiplets), the degeneracy of a given level (dimension of the multiplet) and the level splittings (eigenvalues of various casimirs). The basic idea that enables us to carry this program through is extremely simple: a hermitian matrix can be diagonalized. This fact has many names: Schurs lemma, Wigner-Eckart theorem, full reducibility of unitary representations, and so on (see sect. 3.4 and sect. 4.3). We exploit it by constructing invariant hermitian matrices M from the primitive invariant tensors. M s have collective indices (2.1) and act on tensors. Being hermitian, they can be diagonalized 1 0 0 CM C = . . . 0 1 0 0 0 1 2 ... , .

..

and their eigenvalues can be used to construct projection operators which reduce multiparticle states into direct sums of lower-dimensional representations (see sect. 3.4): .. . . . . ... 0 ... 0 1 0 ... 0 0 1 . . M j 1 . . C . (2.3) =C . . Pi = . . . . .. . i j . . j =i 0 ... 1 0 ... 0 ... . . . . .. An explicit expression for the diagonalizing matrix C (Clebsch-Gordan coecients, sect. 3.7) is unnecessary it is in fact often more of an impediment than an aid, as it obscures the combinatorial nature of group theoretic computations (see sect. 3.12). All that is needed in practice is knowledge of the characteristic equation for the invariant matrix M (see sect. 3.4). The characteristic equation is usually
printed April 14, 2000 DasGroup/book/chapter/preview.tex 14apr2000

CHAPTER 2. A PREVIEW

a simple consequence of the algebraic relations satised by the primitive invariants, and the eigenvalues i are easily determined. i s determine the projection operators Pi , which in turn contain all relevant spectroscopic information: the representation dimension is given by tr Pi , and the casimirs, 6j s, crossing matrices and recoupling coecients (see chapter 4) are traces of various combinations of Pi s. All these numbers are combinatoric; they can often be interpreted as the number of dierent colorings of a graph, the number of singlets, and so on. The invariance group is determined by considering innitesimal transformations Gb a
a b + i i (Ti )b a.

The generators Ti are themselves clebsches, elements of the diagonalizing matrix C for the tensor product of the dening representation and its conjugate. They project out the adjoint representation, and are constrained to satisfy the invariance conditions (2.2) for innitesimal transformations (see sect. 3.9 and sect. 3.10):
c... b c... c c ... + ... = 0 (Ti )a a ha b... + (Ti )b hab ... (Ti )c hab...
111111 000000 000000 111111 000000 111111 000000 111111 000000 111111 000000 111111 000000 111111 000000 111111 000000 111111 000000 111111 000000 111111
0000000 1111111 1111111 0000000 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111

0000000 1111111 1111111 0000000 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111 0000000 1111111

+ ... = 0.

(2.4)

As the corresponding projector operators are already known, we have an explicit construction of the symmetry group (at least innitesimally we will not consider discrete transformations). If the primitive invariants are bilinear, the above procedure leads to the familiar tensor representations of classical groups. However, for trilinear or higher invariants the results are more surprising. In particular, all exceptional Lie groups emerge in a pattern of solutions which we will refer to as a magic triangle. The logic of the construction can be schematically indicated by the following chains of subgroups (see chapter 15): primitive invariants qq qq qqq qqqq higher order invariance group SU (n) G2 + . . . E8 + . . . SO(n) F4 + . . . Sp(n) E6 + . . . E7 + . . .

In the above diagram the arrows indicate the primitive invariants which characterize a particular group. For example, E7 primitives are a sesquilinear invariant qq , a skew symmetric qp invariant and a symmetric qqqq (see chapter 19). The strategy is to introduce the invariants one by one, and study the way in which they split up previously irreducible representations. The rst invariant might be realizable in many dimensions. When the next invariant is added
DasGroup/book/chapter/preview.tex 14apr2000 printed April 14, 2000

2.2. FIRST EXAMPLE: SU (N )

(sect. 3.5), the group of invariance transformations of the rst invariant splits into two subsets; those transformations which preserve the new invariant, and those which do not. Such decompositions yield Diophantine conditions on representation dimensions. These conditions are so constraining that they limit the possibilities to a few which can be easily identied. To summarize; in the primitive invariants approach, all simple Lie groups, classical as well as exceptional, are constructed by (see chapter 20): i) dening a symmetry group by specifying a list of primitive invariants, ii) using primitiveness and invariance conditions to obtain algebraic relations between primitive invariants, iii) constructing invariant matrices acting on tensor product spaces, iv) constructing projection operators for reduced representation from characteristic equations for invariant matrices. Once the projection operators are known, all interesting spectroscopic numbers can be evaluated. The foregoing run through the basic concepts was inevitably obscure. Perhaps working through the next two examples will make things clearer. The rst example illustrates computations with classical groups. The second example is more interesting; it is a sketch of construction of irreducible representations of E6 .

2.2

First example: SU (n)

How do we describe the invariance group that preserves the norm of a complex vector? The list of primitives consists of a single primitive invariant
n a b p qa = m(p, q ) = b a=1 a is the only primitive invariant tensor. We can immediately The Kronecker b write down the two invariant matrices on the tensor product of the dening space and its conjugate:

(pa ) qa .

d identity :
c 1a d,b

c b c . b
DasGroup/book/chapter/preview.tex 14apr2000

a c b d

= a d

ac a c = d b = trace : Td,b

a
printed April 14, 2000

10

CHAPTER 2. A PREVIEW

The characteristic equation for T written out in the matrix, tensor and birdtrack notations is T 2 = nT
af ec a f e c ac Tf,b = d e f b = n Td,b Td,e

=n

e = n, the dimension of the dening vector space. The roots Here we have used e are 1 = 0, 2 = n, and the corresponding projection operators are

SU (n) adjoint rep:

P1 = =

T n1 0n

1 =1 n T

T 01 n1

1 n 1 n

(2.5) .

U (n) singlet:

P2 =

1 =n T =

Now we can evaluate any number associated with the SU (n) adjoint representation, such as its dimension and various casimirs. The dimensions of the two representations are computed by tracing the corresponding projection operators (see sect. 3.4) SU (n) adjoint: d1 = tr P1 = = n2 1 1 =1. n To evaluate casimirs, we need to x the overall normalization of the generators of SU (n). Our convention is to take singlet: d2 = tr P2 = ij = tr Ti Tj = birdTrack . (2.6) = 1 n
b a = b a

1 b a n a b

The value of the quadratic casimir for the dening representation is computed by substituting the adjoint projection operator
b = (Ti Ti )b SU (n) : CF a a =

=a
b

1 a n

n2 1 n

(2.7)

In order to evaluate the quadratic casimir for the adjoint representation, we need to replace the structure constants iCijk by their Lie algebra denition (see sect. 3.10) Ti Tj Tj Ti = iCijk =
printed April 14, 2000

DasGroup/book/chapter/preview.tex 14apr2000

2.2. FIRST EXAMPLE: SU (N )

11

Tracing with Tk we can express Cijk in terms of the dening representation traces: iCijk = tr (Ti Tj Tk ) tr (Tj Ti Tk ) =

The adjoint quadratic casimir Cimn C nmj is now evaluated by rst eliminating Cijk s in favor of the dening representation: ij CA =
i j

= 2

(2.8)

The remaining Cijk can be unwound by the Lie algebra commutator = (2.9)

We have already evaluated the quadratic casimir (2.7) in the rst term. The second term we evaluate by substituting the adjoint projection operator = 1 n = 1 n

1 a (Ti )b a (Tj )b n c The (Ti )a a (Tj )c term vanishes by the tracelessness of Ti s. This can be considered a consequence of the orthonormality of the two projection operators P1 and P2 in (2.5) (see (3.47)):
a c d a c tr (Ti Tk Tj Tk ) = (Ti )b a (P1 )d , b (Tj )c = (Ti )a (Tj )c

0 = P1 P2 = n2 1 1 + n n

tr Ti =

=0

(2.10)

Combining the above expressions we nally obtain CA = 2 = 2n . (2.11)

The problem (1.1) that started all this is evaluated the same way. First we relate the adjoint quartic casimir to the dening casimirs: = = = = = n2 1 n ... ... + + 2 n ... + 1 n + ...

printed April 14, 2000

DasGroup/book/chapter/preview.tex 14apr2000

12 and so on. The result is SU (n) : =n + +2

CHAPTER 2. A PREVIEW

(2.12)

(1.1) is now reexpressed in terms of the dening representation casimirs: = 2n2 +2n + + ... + 4 + ...

The rst two terms are evaluated by inserting the gluon projection operators SU (n) : = n2 1 n
2

1 n 1 n n 1 n + 1 n2 + 1 n2

(2.13)

= = =

1 1 2 n n 3 n2 3 + 2 n n2 2 +

and the remaining terms have already been evaluated. Collecting everything together, we nally obtain SU (n) : = 2n2 (n2 + 12) (2.14)

This example was unavoidably lengthy; the main point is that the evaluation is performed by a substition algorithm and is easiliy automated. Any graph, no matter how complicated, is eventually reduced to a polynomial in traces of a = n, ie. the dimension of the dening representation. a

2.3

Second example: E6 family

What invariance group preserves norms of complex vectors, as well as a symmetric cubic invariant D(p, q, r) = dabc pa qb rc = D(q, p, r) = D(p, r, q ) ? We analyze this case following the steps of the summary of sect. 2.1: i) primitive invariant tensors:
b =a a

b,

dabc = b

a c

dabc = (dabc ) = b

a c
printed April 14, 2000

DasGroup/book/chapter/preview.tex 14apr2000

2.3. SECOND EXAMPLE: E6 FAMILY

13

a , the only primitive twoii) primitiveness: daef def b must be proportional to b index tensor. We use this to x the overall normalization of dabc s:

= iii) invariant hermitian matrices: We shall construct here the adjoint representation projection operator on the tensor product space of the dening representation and its conjugate. All invariant matrices on this space are d
a c d b

c ,
a c b d

d a

c , b dace debd = .

They are hermitian in the sense of being invariant under complex conjugation and transposition of indices (see (3.18)). The adjoint projection operator must be expressible in terms of the four-index invariant tensors listed above:
d a d a d ade dbce ) (Ti )a b (Ti )c = A(c b + Bb c + Cd

= A

+B

+C

iv) invariance. The cubic invariant tensor satises (2.4) + + = 0.

Contracting with dabc we obtain +2 = 0.

Contracting next with (Ti )b a , we get an invariance condition on the adjoint projection operator: +2 = 0.

Substituting the adjoint projection operator yields the rst relation between the coecients in its expansion: 0 = n+B+C +2 0 = B+C +
printed April 14, 2000

+B

+C

n+2 . 3
DasGroup/book/chapter/preview.tex 14apr2000

14

CHAPTER 2. A PREVIEW

v) the projection operators should be orthonormal, P P = P . The adjoint projection operator is orthogonal to the singlet projection operator P constructed in sect. 2.2. This yields the second relation on the coecients: 0 = PA P1 1 0 = n

= 1 + nB + C .

Finally, the overall normalization factor A is xed by PA PA = PA : = =A 1+0 C 2 .

Combining the above 3 relations we obtain the adjoint projection operator for the invariance group of a symmetric cubic invariant = 2 9+n 3 + (3 + n) .

The corresponding characteristic equation, mentioned in the point iv of the summary of sect. 2.1 is given in (??). The dimension of the adjoint representation is obtained by tracing the projection operator N = ii = = = nA(n + B + C ) = 4n(n 1) n+9

This Diophantine condition is satised by a small family of invariance groups, discussed in chapter 17. The most interesting member of this family is the exceptional Lie group E6 , with n = 27 and N = 78.

DasGroup/book/chapter/preview.tex 14apr2000

printed April 14, 2000

Chapter 3

Invariants and reducibility

Basic group theoretic notions are introduced groups, invariants, tensors and the diagrammatic notation for invariant tensors. The basic idea is simple; a hermitian matrix can be diagonalized. If this matrix is an invariant matrix, it decomposes the representations of the group into direct sums of lower dimensional representations. The key results are the construction of projection operators from invariant matrices (3.45), the Clebsch-Gordan coecients representation of projection operators (3.73), the invariance conditions (3.91) and the Lie algebra relations (3.103).

3.1

Preliminaries

In this section we dene basic building blocks of the theory to be developped here: groups, vector spaces, algebras, etc. This material is covered in any introduction to group theory [7, 5]. Most of sect. 3.1.2 to sect. 3.1.4 is probably known to the reader, and protably skipped on the rst reading.

3.1.1

Groups

Denition. A set of elements g G forms a group with respect to multiplication G G G if (a) the set is closed with respect to multiplication; for any two elements a, b G , the product ab G . (b) the multiplication is associative (ab)c = a(bc) for any three elements a, b, c G . (c) there exists an identity element e G such that eg = g e for any g G 15

16

CHAPTER 3. INVARIANTS AND REDUCIBILITY

(d) for any g G there exists an inverse g 1 such that g 1 g = gg 1 = e . If the group is nite, the number of elements is called the order of the group and denoted |G|. If the multiplication ab = ba is commutative for all a, b G , the group is abelian. Two groups with the same multiplication table are said to be isomorphic. Denition. A subgroup H G is a subset of G that forms a group under multiplication. e is always a subgroup; so is G itself. Denition. A cyclic group is a group generated from one of its elements, called the generator of the cyclic group. If n is the minimum integer such that an = e, the set G = {e, a, a2 , , an1 } is the cyclic group. As all elements commute, cyclic groups are abelian. Every subgroup of a cyclic group is cyclic.

3.1.2

Vector spaces

Denition. A set V of elements x, y, z, . . . is called a vector (or linear) space over a eld F if (a) vector addition + is dened in V such that V is an abelian group under addition, with identity element 0. (b) the set is closed with respect to scalar multiplication and vector addition a(x + y) = ax + ay , (a + b)x = ax + bx a(bx) = (ab)x 1x = x, 0x = 0. a, b F , x, y V

Here the eld F will be either R, the eld of reals numbers, or C, the eld of complex numbers (quaternion and octonion elds are discussed in sect. 15.5). Denition. n-dimensional complex vector space V consists of all n-multiplets x = (x1 , x2 , . . . , xn ), xi C. The two elements x, y are equal if xi = yi for all 0 i n. The vector addition identity element is 0 = (0, 0, , 0). Denition. A complex vector space V is an inner product space if with every pair of elements x, y V there is associated a unique inner (or scalar) product (x, y ) C, such that (x, y ) = (y, x) (ax, by ) = a b(x, y ) , a, b C (z, ax + by ) = a(z, x) + b(z, y ) ,
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.1. PRELIMINARIES

17

where * denotes complex conjugation. Without any noteworthy loss of generality we shall here dene the scalar product of two elements of V by
n

(x, y ) =
j =1

x j yj .

(3.1)

3.1.3

Algebra

Denition. A set of elements t of a vector space T forms an algebra if, in addition to the vector addition and scalar multiplication (a) the set is closed with respect to multiplication T T T , so that for any two elements t , t T , the product t t also belongs to T : t t =
T

t t .

(3.2)

(b) the multiplication operation is bilinear (t + t ) t = t t + t t

t (t + t ) = t t + t t The set of numbers t are called the structure constants of the algebra. They form a matrix representation of the algebra (t ) = t (3.3)

whose dimension is the dimension of the algebra itself. Depending on what further assumptions one makes on the multiplication, one obtains dierent types of algebras. For example, if the multiplication is associative (t t ) t = t (t t ) , the algebra is associative. Typical examples of products are the matrix product
b c (t t )c a = (t )a (t )b

, t V V

(3.4)

and the Lie product

b c b a (t t )c a = (t )a (t )b (t )c (t )b

, t V V

(3.5)

which denes a Lie algebra. As a plethora of vector spaces, indices and conjugations looms large in our immediate future, it pays to streamline the notation now, by singling out one vector space as dening, and replacing complex conjugation by raised indices.
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

18

CHAPTER 3. INVARIANTS AND REDUCIBILITY

3.1.4

Dening space, tensors, representations

Denition. Let V be the dening n-dimensional complex vector space. Associate with the dening n-dimensional complex vector space V a conjugate (or = {x dual) n-dimensional vector space V |x V } obtained by complex conju by gation of elements x V . We shall denote the corresponding element of V raising the index xa = (xa ) , so the components of dening space vectors, resp. conjugate vectors, are distinguished by lower, resp. upper indices x = (x1 , x2 , . . . , xn ) , x V . V x = (x1 , x2 , . . . , xn ) , x

(3.6)

Repeated index summation: Throughout this text the repeated indices are always summed over
n

Gb a xb

=
b=1

Gb a xb ,

(3.7)

unless explicitly stated otherwise. Denition. Let G be a group of transformations acting linearly on V , with the action of a group element g G on a vector x V given by a unitary [nn] matrix G xa = Gb a xb a, b = 1, 2, . . . , n . (3.8)

We shall refer to Gb a as the dening representation of the group. The action of is given by the conjugate representation G g G on a vector q V x a = xb (G )a b,
b (G )a b (Ga ) .

(3.9)

by complex conjugation and inner product By dening the conjugate space V b as the invariant (3.1), we have already chosen (without any loss of generality) a tensor with the bilinear form (x, x) = xb xb . From this choice it follows that in the applications considered here, the group G is always assumed unitary
b b (G )c a Gc = a .

(3.10)

q is any object that transforms under the action Denition. A tensor x V p V of g G as

1 2 x b1 ...bp

a a ...aq

1 2 = G b1 ...bp

a a ...aq dp ...d1 , cq ...c2 c1

q 2 xd1 , 1 ...dp

c c ...c

(3.11)

q tensor representation of g G is dened by where the V p V

1 2 G b1 ...bq

a a ...ap dq ...d1 , cp ...c2 c1

q a2 p d1 1 (G )a c1 (G )c2 . . . (G )cp Gb1 . . . Gbq .

(3.12)
printed April 14, 2000

DasGroup/book/chapter/invar.tex 14apr2000

3.1. PRELIMINARIES Tensors can be combined into other tensors by (a) addition
ab...c ab...c = xab...c zd...e d...e + y d...e ,

19

, C ,

(3.13)

(b) product
abcd abc d zef g = xe yf g ,

(3.14)

(c) contraction: Setting an upper and lower index equal and summing over all of q1 without these indices: its values yields a tensor z V p1 V
bc...d = xabc...d ze...f e...af , ad d ze = xabc e ycb .

(3.15)

q transforms linearly under the action of g , so it can be A tensor x V p V . We can replace considered a vector in the d = np+q dimensional vector space V the array of its indices by one collective index:
1 2 x = xb1 ...bp

a a ...aq

(3.16)

One could be more explicit and give a table like

...1 21...1 nn...n x1 = x11 1...1 , x2 = x1...1 , . . . , xd = xn...n ,

(3.17)

but that is unnecessary, as we shall use the compact index notation only as a shorthand. Denition. Hermitian conjugation is eected by complex conjugation and index transposition:
edc (h )ab cde (hba ) .

(3.18)

Complex conjugation interchanges upper and lower indices, as in (3.6); transposition reverses their order. A matrix is hermitian if its elements satisfy
a (M )a b = Mb .

(3.19)

Denition. The tensor conjugate to x has form

1 x = xap q ...a2 a1 .

b ...b

(3.20)

Combined, the above denitions lead to the hermitian conjugation rule for collective indices: a collective index is raised or lowered by interchanging the upper and lower indices and reversing their order:

a1 a2 . . . aq b1 . . . bp

bp . . . b1 aq . . . a2 a1

(3.21)

This transposition convention will be motivated further by the diagrammatic rules of sect. 3.6. The tensor representation (3.12) can be treated as a [d d] matrix G =G
a1 a2 ...aq dp ...d1 b1 ...bp , cq ...c2 c1

(3.22)

and the tensor transformation (3.11) takes the usual matrix form x = G x .
printed April 14, 2000

(3.23)
DasGroup/book/chapter/invar.tex 14apr2000

20

CHAPTER 3. INVARIANTS AND REDUCIBILITY

3.2

Invariants

Denition. The vector q V is an invariant vector if for any transformation gG q = Gq . q is an invariant tensor if for any g G Denition. A tensor x V p V
1 2 xb1 ...bq

(3.24)

a a ...ap

q p d1 1 2 a2 1 = (G )a c1 (G )c2 . . . Gb1 . . . Gbq xd1 ...dq .

c c ...c

(3.25)

We can state this more compactly by using the notation of (3.22) x = G x .

a a ...a

(3.26)

p 1 2 as a vector in [d d] dimensional space, d = Here we treat the tensor xb1 ...bq p + q n . b y is invariant for all g G , the matrix If a bilinear form M ( x, y ) = xa Ma b

b b d = Gc Ma a (G )d Mc

(3.27)

is an invariant matrix . Multiplying with Ge b and using the unitary condition (3.10), we nd that the invariant matrices commute with all transformations g G: [G, M ] = 0 . If we wish to treat a tensor as a matrix
2 = Mb11...b M p

(3.28)

a a ...aq dp ...d1 , cq ...c2 c1

(3.29)

then the invariance condition (3.26) will takes the commutator form (3.28). Denition. We shall refer to an invariant relation between p vectors in V and which can be written as a homogeneous polynomial in terms of q vectors in V vector components, such as
e d c H (x, y, z , r , s ) = hab cde xb ya s r z ,

(3.30)

p (repeated indices, as always, summed over). In this as an invariant in V q V 3 2 example, the coecients hab cde are components of invariant tensor h V V , obeying the invariance condition (3.25). Diagrammatic represention of tensors, such as a hab cd = d (3.31) b c makes it easier to distinguish dierent types of invariant tensors. We shall explain in great detail our conventions for drawing tensors in sect. 3.6; sketching a few simple examples should suce for the time being.
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.2. INVARIANTS

21

The standard example of a dening vector space is our three-dimensional is the space of all three-component real vectors (n = 3), Euclidean space: V = V and examples of invariants are the length L(x, x) = ij xi xj and the volume V (x, y, z ) = ijk xi yj zk . We draw the corresponding invariant tensors as k ij = i j,
ijk

= i j

(3.32)

Denition. A composed invariant tensor can be written as a product and/or contraction of invariant tensors. Examples of composed invariant tensors are m i i , ij klm = ijm mn nkl = j k j l

l . k

(3.33)

The rst example corresponds to a product of the two invariants L(x, y )V (z, r, s). d The second involves an index contraction; we can write this as V x, y, dz V (z, r, s). In order to proceed, we need to distinguish the primitive invariant tensors from the innity of composed invariants. We begin by dening a nite basis for q: invariant tensors in V p V Denition. A tree invariant can be represented diagrammatically as a product of invariant tensors involving no loops of index contractions. We shall denote by T = {t0 , t1 . . . tr } a (maximal) set of r linearly independent tree invariants q . As any linear combination of t can serve as a basis, we clearly t V p V have a great deal of freedom in making informed choices for the basis tensors. Example: Tensors (3.33) are tree invariants. The tensor i hijkl =
ims jnm krn sr

l . (3.34) k

= j

is not a tree invariant, as it involves a loop. Denition. An invariant tensor is called a primitive invariant tensor if it cannot expressed as a combination of tree invariants composed from lower rank primitive invariant tensors. Let P = {p1 , p2 , . . . pk } be the set of all primitives. For example, the Kronecker delta and the Levi-Civita tensor (3.32) are the primitive invariant tensors of our three-dimensional space. The loop contraction (3.34) is not a primitive, because by the Levi-Civita completeness relation (5.32) it reduces to a sum of tree contractions: i i l l l i = + = ij kl + il jk . (3.35) j k k j k j
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

22

CHAPTER 3. INVARIANTS AND REDUCIBILITY

(the Levi-Civita tensor is discussed in sect. 5.4.) q can be Primitiveness assumption. Any invariant tensor h V p V q p expressed as a linear sum over the tree invariants T V V h=
T

h t .

(3.36)

In contradistinction to arbitrary composite invariant tensors, the number of tree invariants for a xed number of external indices is nite. For example, given the n = 3 dimensions primitives P = {ij , fijk }, any invariant tensor h V p (here denoted by a blob) must be expressible as birdTrack = A birdTrack = A birdTrack = A +B +C +D +E +F (3.38) (3.37)

birdTrack = birdTrack + birdTrack + . . . . . . =

3.2.1

Algebra of invariants

Any invariant tensor of matrix form (3.29)

2 = Mb11...b M p

a a ...aq dp ...d1 , cq ...c2 c1

p V q V p can be expanded in the basis (3.36). The basis which maps V q V p V q V p , and the matrix product of two t are themselves matrices in V q V p V q V p and can be expanded in basis elements is also an element of V q V the minimal basis: t t =
T

(t ) t .

(3.39)

As the number of tree invariants composed from the primitives is nite, under matrix multiplication the bases t form a nite algebra, with the coecients (t ) giving their multiplication table. The multiplication coecients (t ) form a [r r]-dimensional matrix representation of t acting on the vector (e, t1 , t2 , tr1 ). Given a basis, we can evaluate the matrices e , (t1 ) , (t2 ) , (tr1 ) and their eigenvalues. For at least one of these matrices all eigenvalues will be distinct (or we have failed to chose a minimal basis). The projection operator technique of sect. 3.4 will enable us to exploit this fact to p space into r irreducible subspaces. decompose the V q V
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.3. INVARIANCE GROUPS

23

This can be said in another way; the choice of basis {e, t1 , t2 , tr1 } is arbitrary, the only requirement being that the basis elements are linearly independent. Finding a (t ) with all eigenvalues distinct is all we need to construct an orthonormal basis {P0 , P1 , P2 , Pr1 }, where the basis matrices Pi are the projection operators, to be constructed below in sect. 3.4.

3.3

Invariance groups

So far we have dened invariant tensors as the tensors invariant under transformations of a given group. Now we proceed in the other direction: given a set of tensors, what is the group of transformations that leaves them invariant? Given a full set of primitives (3.30) P = {p1 , p2 , . . . , pk }, meaning that no other primitives exist, we wish to determine all possible transformations that preserve this gvien set of invariant relations. Denition. An invariance group G is the set of all linear transformations (3.25) which preserve the primitive invariant relations (and, by extension, all invariant relations) ) = p1 (Gx, y G ) p1 (x, y p2 (x, y, z, . . .) = p2 (Gx, Gy, Gz . . .) , ... (3.40)

Unitarity (3.10) guarantees that all contractions of primitive invariant tensors, and hence all composed tensors h H are also invariant under action of G . As G we consider is unitary, it follows from (3.10) that the list of primitives must always include the Kronecker delta. Example 1. If pa qa is an invariant of G
a p qa = pb (G G)c b qc = p qa , a

(3.41)

then G is the full unitary group U (n) (invariance group of the complex norm a ), whose elements satisfy |x|2 = xb xa b G G = 1 . (3.42)

Example 2. If we wish the z -direction to be invariant in our three-dimensional space, q = (0, 0, 1) is an invariant vector (3.24), and the invariance group is O(2), the group of all rotations in the x-y plane.

Remark 3.1 Which representation is dening?. 1. The dening space V need not carry the lowest dimensional representation of G ; it is merely the space in terms of which we chose to dene the primitive invariants.
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

24

CHAPTER 3. INVARIANTS AND REDUCIBILITY

b 2. We shall always assume that the Kronecker delta a is one of the primitive invariants, ie. that G is a unitary group whose elements satisfy (3.42). This restriction to unitary transformations is not essential, but it simplies proofs of full reducibility. The results, however, apply as well to the nite-dimensional representations of non-compact groups, such as the Lorentz group SO(3, 1).

3.4

Projection operators

For M a hermitian matrix, there exists a diagonalizing unitary matrix C such that: 1 0 0 0 0 1 2 0 . . . 0 0 2 0 0 . . . CM C = , i = j . (3.43) . .. . . . 0 . . . 2 . . . 3 0 0 . .. . . . Here i are the r distinct roots of the minimal characteristic polynomial
r

(M i 1) = 0 .
i=1

(3.44)

(the characteristic equations will be discussed in sect. 5.7.) In the matrix C (M 2 1)C the eigenvalues corresponding to 2 are replaced by zeroes: 1 2 1 2 1 2 0 .. . , 0 3 2 3 2 .. . and so on, so the product over all factors (M 2 1)(M 3 1) . . . with exception of the (M 1 1) factor has non-zero entries only in the subspace associated with 1 : 1 0 0 0 1 0 0 0 0 1 . 0 (M j 1)C = (1 j ) C 0 j =1 j =1 0 0 .. .
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.5. FURTHER INVARIANTS

25

In this way we can associate with each distinct root i a projection operator Pi Pi =
j =i

M j 1 , i j

(3.45)

which is identity on the ith subspace, and zero elsewhere. For example, the projection operator onto the 1 subspace is 1 1 1 C . 0 P1 = C (3.46) 0 0 .. . The matrices Pi are orthonormal Pi Pj = ij Pj , (no sum on j ) , (3.47)

and satisfy the completeness relation

r

Pi = 1 .
i=1

(3.48)

As tr (CPi C + ) = tr Pi , the dimension of the ith subspace is given by di = tr Pi . (3.49)

It follows from the characteristic equation (3.44) and the form of the projection operator (3.45) that i is the eigenvalue of M on Pi subspace: M Pi = i Pi , (no sum on i) . (3.50)

Hence any matrix polynomial f (M ) takes the scalar value f (i ) on the Pi subspace f (M )Pi = f (i )Pi . (3.51)

This, of course, is the real reason why one wants to work with irreducible representations: they render matrices and operators harmless c-numbers.

3.5

Further invariants

Suppose that there exist several linearly independent invariant [d d] hermitian matrices M1 , M2 , . . . and that we have used M1 to decompose the d-dimensional = Vi . Can M2 be used to further decompose Vi ? This vector space V is the standard problem of quantum mechanics (simultaneous observables), and
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

26

CHAPTER 3. INVARIANTS AND REDUCIBILITY

the answer is that further decomposition is possible if, and only if, the invariant matrices commute, [M1 , M2 ] = 0 , or, equivalently, if all projection operators commute Pi Pj = Pj Pi . (3.53) (3.52)

Usually the simplest choices of independent invariant matrices do not commute. In that case, the projection operators Pi constructed from M1 can be used to project commuting pieces of M2 : M2 = Pi M2 Pi ,
(i) (i)

(no sum on i) .

That M2 commutes with M1 follows from the orthogonality of Pi : [M2 , M1 ] =

j (i)

j [M2 , Pj ] = 0 .
(i)

(i)

(3.54)

Now the characteristic equation for M2 (if nontrivial) can be used to decompose Vi subspace. An invariant matrix M induces a decomposition only if its diagonalized form (3.43) has more than one distinct eigenvalue; otherwise it is proportional to the unit matrix, and commutes trivially with all group elements. A representation is said to be irreducible if all invariant matrices that can be constructed are proportional to the unit matrix. In particular, the primitiveness relation (3.37) is a statement that the dening representation is assumed irreducible. According to (3.28), an invariant matrix M commutes with group transformations [G, M ] = 0. Projection operators (3.45) constructed from M are polynomials in M , so they also commute with all g G : [G, Pi ] = 0 , (3.55)

(remember that Pi are also invariant [d d] matrices). Hence a [d d] matrix representation can be written as a direct sum of [di di ] matrix representations G = 1G1 =
i,j

Pi GPj =
i

Pi GPi =
i

Gi .

(3.56)

In the diagonalized representation (3.46), the matrix G has a block diagonal form: 0 G1 0 C i Gi Ci . (3.57) CGC = 0 G2 0 , G= .. i . 0 0
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.6. BIRDTRACKS

27

Representation Gi acts only on the di dimensional subspace Vi consisting of vec . In this way an invariant [d d] hermitian matrix M with r tors Pi q , q V distinct eigenvalues induces a decomposition of a d-dimensional vector space V into a direct sum of di -dimensional vector subspaces Vi M V1 V 2 . . . V r . (3.58) V For more detailed discussion of recursive reduction, consult appendix A.

3.6

Birdtracks

We shall often nd it convenient to represent aglomerations of invariant tensors by birdtracks, a group-theoretical version of Feynman diagrams. Tensors will be represented by vertices, and contractions by propagators. Diagrammatic notation has several advantages over the tensor notation. Diagrams do not require dummy indices, so explicit labelling of such indices is unnecessary. More to the point, for a human eye it is easier to identify topologically identical diagrams than to recognize equivalence between the corresponding tensor expressions. The main disadvantage of diagrammatic notation is lack of standardization, especially in the case of Clebsch-Gordan coecients. Many of the diagrammatic notations [97, 98, 73] designed for atomic and nuclear spectroscopy, are complicated by various phase conventions. In our applications, explicit constructions of clebsches are superuous, and we need no such conventions, confusing or otherwise. In the birdtrack notation, the Kronecker delta is a propagator:
a =b b

a.

(3.59)

, or up and down For a real dening space there is no distinction between V and V indices, and the lines do not carry arrows. Any invariant tensor can be drawn as a generalized vertex: x = xabc de = . (3.60)

Whether the vertex is drawn as a box or a circle or a dot is matter of taste. The orientation of propagators and vertices in the plane of the drawing is likewise irrelevant. The only rules are (1) Arrows point away from the upper indices and toward the lower indices; the line ow is downward, from upper to lower indices: a hcd ab = b
printed April 14, 2000

d . c
DasGroup/book/chapter/invar.tex 14apr2000

(3.61)

28

CHAPTER 3. INVARIANTS AND REDUCIBILITY

(2) Diagrammatic notation must indicate which in (out) arrow corresponds to the rst upper (lower) index of the tensor (unless the tensor is cyclically symmetric);

e = Rabcd

(3.62)

(3) The indices are read in the counterclockwise order around the vertex:

xbce ad =

(3.63)

(The upper and the lower indices are read separately in the counterclockwise order; their relative ordering does not matter.) In the examples of this section we index the external lines for readers convenience, but indices can always be omitted. An internal line implies a summation over corresponding indices, and for external lines the equivalent points on each diagram represent the same index in all terms of a diagrammatic equation. Hermitian conjugation (3.18) does two things: (a) it exchanges the upper and the lower indices, ie. it reverses the directions of the arrows (b) it reverses the order of the indices, ie. it transposes a diagram into its mirror image. For example, x , the tensor conjugate to (3.63), is drawn as

x = xed cba =

(3.64)

and a contraction of tensors x and y is drawn as

1 1 2 x y = xap q ...a2 a1 yb1 ...bp

b ...b

a a ...aq

(3.65)

DasGroup/book/chapter/invar.tex 14apr2000

printed April 14, 2000

3.7. CLEBSCH-GORDAN COEFFICIENTS

29

3.7

Clebsch-Gordan coecients
C .

Consider the product 0 0 1 1 1

(3.66)

0 0 0 ..

of the two terms in the diagonal representation of a projection operator (3.46). This matrix has non-zero entries only in the di rows of subspace Vi . We collect them in a [di d] rectangular matrix (Ci ) , = 1, 2, . . . d, = 1, 2, . . . di : 1 d (Ci )1 . . . (Ci )1 . . . . (3.67) Ci = . . d i . d (Ci )di
d The index in (Ci ) stands for all tensor indices associated with the d = np+q q . In the birdtrack notation these indices are dimensional tensor space V p V

explicit:
1 (Ci ) , ap q ...a2 a1 =

b ...b

(3.68)

Such rectangular arrays are called Clebsch-Gordan coecients (hereafter refered Vi . The conjugate to as clebsches for short). They are explicit mappings V is provided by the product mapping Vi V 0 0 1 1 1 C (3.69) 0 0 0 .. . which denes the [d di ] rectangular matrix (C i ) , = 1, 2, . . . d, = 1, 2, . . . di : i 1 di i (C )1 . . . (C )1 . a1 a2 ...aq . i . (C i )b1 = .(3.70) C = . . . d , ...bp , d (C i )di
di
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

30

CHAPTER 3. INVARIANTS AND REDUCIBILITY

The two rectangular Clebsch-Gordan matrices C i and Ci are related by hermitian conjugation. The tensors we have considered in sect. 3.6 transform as tensor products of the dening representation (3.11). In general, tensors transform as tensor products of various representations, with indices runnig over the corresponding representation dimensions: a1 = 1, 2, . . . , d1
p+q +1 xap 1 a2 ...ap

...a

a2 = 1, 2, . . . , d2 . . where . ap+q = 1, 2, . . . , dp+q . (3.71)

The action of transformation g on the index ak is given by the [dk dk ] matrix representation Gk . The Clebsch-Gordan coecients are notoriously index-overpopulated as they require a representation label and a tensor index for each representation in the tensor product. Diagrammatic notation alleviates this index plague in either of two ways: (i) one can indicate a representation label on each line:
a , = Ca

a a

(3.72)

(an index, if written, is written at the end of a line; a representation label is written above the line); (ii) one can draw the propagators (Kronecker deltas) for dierent representations with dierent kinds of lines. For example, we shall usually draw the adjoint representation with a thin line. By the denition of clebsches (3.46), a -representation projection operator can be written out in terms of Clebsch-Gordan matrices: C C : C C = P ,
a1 a2 ...ap dq ...d1 (C )b1 ...bq , (C ) , cp ...c2 c1

(no sum on )

(3.73)

= =

a1 a2 ...dp dq ...d1 (P )b1 ...bq , cp ...c2 c1

(3.74)

A specic choice of clebsches is quite arbitrary. All relevant properties of projection operators (orthonormality, completeness, dimensionality) are independent of the explicit form of the diagonalization transformation C . Any set of C is
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.8. ZERO- AND ONE-DIMENSIONAL SUBSPACES

31

acceptable, as long as it satises the orthogonality and completeness conditions. From (3.66) and (3.69) it follows that C are orthonormal:
1. C C =
1 2 (C ) , b1 ...bq

a a ...ap

q 1 (C )ap ...a = 2 a1 ,

b ...b

(3.75)

Here 1 is the [d d ] unit matrix, and C s are multiplied as [d d] rectangular matrices. The completeness relation (3.48) C C = 1 ,

1 2 (C )b1 ...bq

([d d] unit matrix) ,

d

a a ...ap

q 1 , (C ) , cp ...c2 c1

d ...d

q a1 a2 = c . . . bq 1 c2

(3.76)

and the orthonormality of projection operators and clebsches

C , C P = C , P C =

(no sum on , ) ,

(3.77)

follow immediately from (3.47) and (3.75).

3.8

Zero- and one-dimensional subspaces

If a projection operator projects onto a zero-dimensional subspace, it must vanish identically d = 0 P = = 0. (3.78)

This follows from (3.46); d is the number of 1s on the diagonal on the right-hand side. For d = 0 the right-hand side vanishes. The general form of P is
r

P =
k=1

ck M k

(3.79)

where Mk are the invariant matrices used in construction of the projector operators, and ck are numerical coecients. Vanishing of P therefore implies a relation among invariant matrices Mk .
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

32

CHAPTER 3. INVARIANTS AND REDUCIBILITY

If a projection operator projects onto a one-dimensional subspace, its expression in terms of the Clebsch-Gordan coecients (3.73) involves no summation, so we can omit the intermediate line di = 1 Pi =
i i

1 2 = (C i )b1 ...bq

a a ...ap

q 1 (Ci )cp ...c 2 c1 .

d ...d

(3.80)

For any subgroup of SU (n), the representations are unitary, with unit determinant. On the one-dimensional spaces the group acts trivially, G = 1. Hence if di = 1, the Clebsch-Gordan coecient Ci in (3.80) is an invariant tensor in q. VpV

...

...

3.9

Innitesimal transformations

A unitary transformation G which is innitesimally close to unity can be written as

b b Gb a = a + iDa , b| where D is a hermitian matrix with small elements, |Da g G on the conjugate space is given by a a (G )a b = b iDb .

(3.81) 1. The action of (3.82)

D can be parametrized by N n2 real parameters. N , the maximal number of independent parameters, is called the dimension of the group (also the dimension of the Lie algebra, or the dimension of the adjoint representation). We shall consider only innitesimal transformations, of form G = 1 + iD, a| 1. We do not study the entire group of invariances, but only the trans|Db formations (3.8) connected to the identity. For example, we shall not consider invariances under coordinate reections. The generators of innitesimal transformations (3.81) are hermitian matrices space. However, not any element of V V and belong to the Da V V
b

generates an allowed transformation; indeed, one of the main objectives of group theory is to dene the class of allowed transformations. In sect. 3.4 we have described the general decomposition of a tensor space into (ir)reducible subspaces. As a particular case, consider the decomposition of . The corresponding projection operators satisfy the completeness relation V V (3.76) 1 =
a c d b =

1 T + PA + n

P
=A c (P )a b, d =A
11 00

1 a c c + (PA )a b, d + n b d 1 n + +

111 000

(3.83)

printed April 14, 2000

DasGroup/book/chapter/invar.tex 14apr2000

3.9. INFINITESIMAL TRANSFORMATIONS

33

j decomposes into 2 subIf i is the only primitive invariant tensor, then V V spaces, and there are no other irreducible representations. However, if there are decomposes into more irreducible reprefurther primitive invariant tensors, V V sentations, and therefore the sum over . Examples will abound in what follows. a , c is The singlet projection operator T /n always gures in this expansion, as b d always one of the invariant matrices (see the example worked out in sect. 2.2). a must belong to at least one of the Furthermore, the innitesimal generators Db . irreducible subspaces of V V This subspace is called the adjoint space, and its special role warrants introduction of special notation. We shall refer to this vector space by letter A, in distinction to the dening space V of (3.6). We shall denote its dimension by N , label its tensor indices by i, j, k . . ., denote the corresponding Kronecker delta by a thin straight line

ij = i

j , i, j = 1, 2, . . . , N ,

(3.84)

and the corresponding Clebsch-Gordan coecients by 1 a (CA )i , a b = (Ti )b = i a a b a, b = 1, 2, . . . , n i = 1, 2, . . . , N . Matrices Ti are called the generators of innitesimal transformations. Here a is an (uninteresting) overall normalization xed by the orthogonality condition (3.75)
b (Ti )a b (Tj )a = tr (Ti Tj ) = a ij

= a

(3.85)

The scale of Ti is not set, as any overall rescaling can be absorbed into the normalization a. For our purposes it will be most convenient to use a = 1 as the normalization convention. Other normalizations are commonplace. For example, 1 SU (2) Pauli matrices Ti = 1 2 i and SU (n) Gell-Mann [8] matrices Ti = 2 i are conventionally normalized by xing a = 1/2: 1 tr (Ti Tj ) = ij . 2 (3.86)

The projector relation (3.73) expresses the adjoint representation projection operators in terms of the generators:
c (PA )a b, d =

1 1 c (Ti )a b (Ti )d = a a

(3.87)

Clearly, the adjoint subspace is always included in the sum (3.83) (there must b , or otherwise there is no group to exist some allowed innitesimal generators Da describe), but how do we determine the corresponding projection operator?
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

34

CHAPTER 3. INVARIANTS AND REDUCIBILITY

The adjoint projection operator is singled out by the requirement that the group transformations do not aect the invariant quantities. (Remember, the group is dened as the totality of all transformations that leave the invariants invariant.) For every invariant tensor q , the innitesimal group elements G = 1 + iD must satisfy the invariance condition (3.24). Parametrizing D as a projection into the adjoint space, D = PA H of an arbitrary hermitian matrix H V V : V V
a = Db

1 (Ti )a b i, a

1 tr (Ti H ) a

(3.88)

we obtain the invariance condition which the generators must satisfy: they annihilate invariant tensors Ti q = 0 . (3.89)

To state the invariance condition for an arbitrary invariant tensor, we need to dene the generators in the tensor representations. By substituting G = 1 + i T + O( 2 ) into (3.12) and keeping only the terms linear in , we nd that the generators of innitesimal transformations for tensor representations act by touching one index at a time:
1 2 (Ti )b1 ...bq

a a ...ap dq ...d1 , cp ...c2 c1

p d1 q 1 a2 = (Ti )a c 1 c 2 . . . c p b1 . . . bq

p d1 q p d1 q a1 a1 a2 2 +c (Ti )a c2 . . . cp b1 . . . bq + . . . + c1 c2 . . . (Ti )cp b1 . . . bq 1

p q p d1 q a1 a2 a1 a2 1 c . . . cp (Ti )d b1 . . . bq . . . c1 c2 . . . cp b1 . . . (Ti )bq . 1 c2

(3.90)

(3.91)

(with a relative minus sign between lines owing in opposite directions). In other words, the Leibnitz rule obscured by a forest of indices. Tensor representations of the generators decompose in the same way as the group representations (3.57) Ti =

C Ti C .

()

(3.92)

DasGroup/book/chapter/invar.tex 14apr2000

printed April 14, 2000

3.9. INFINITESIMAL TRANSFORMATIONS

35

The invariance conditions take a particularly suggestive form in the diagrammatic notation. (3.89) amounts to insertion of a generator into all external legs of the diagram corresponding to the invariant tensor q :

0 =

(3.93)

The insertions on the lines going into the diagram carry a minus sign relative to the insertions on the outgoing lines. Clebsch-Gordan coecients are also invariant tensors. Multiplying both sides of (3.57) with C and using orthogonality (3.75), we obtain C G = G C , (no sum on ) . (3.94) The Clebsch-Gordan matrix C is a rectangular [d d] matrix, hence g G acts on it with a [d d ] representation from the left, and a [d d] representation from the right. (3.45) is the statement of invariance for rectangular matrices, analogous to (3.27), the statement of invariance for square matrices: C = G C G . C = G C G ,

(3.95)

The invariance condition for the Clebsch-Gordan coecients is a special case of (3.93), the invariance condition for any invariant tensor: 0 = Ti C + C Ti
()

0 =

(3.96)

The orthonormality condition (3.75) now yields the generators in representation in terms of the dening representation generators

(3.97)

printed April 14, 2000

DasGroup/book/chapter/invar.tex 14apr2000

36

CHAPTER 3. INVARIANTS AND REDUCIBILITY

Remark 3.2 The reality of the adjoint representation. For hermitian generators the adjoint representation is real, and the upper and lower indices need not be distinguished; the propagator needs no arrow. For non-hermitian choices of generators, the adjoint representation is complex are equivalent, as indices can be (gluon lines carry arrows) but A and A raised an lowered by the Cartan-Killing form gij = Tr (Ti Tj ). The Cartan canonical basis D = i Hi + E + E is an example of a non-hermitian choice. Here we shall always assume that Ti are chosen hermitian.

3.10

Lie algebra

As the simplest example of computation of the generators of innitesimal transformations acting on spaces other than the dening space, consider the adjoint A adjoint representation Clebschrepresentation. Using (3.97) on the V V Gordan coecients (ie., generators Ti ) we obtain = (3.98)

b a c b a (Ti )jk = (Ti )c a (Tk )c (Tj )b (Ti )a (Tj )c (Tk )b .

Our convention is to always assume that the generators Ti have been chosen hermitian. That means that i in the expansion (3.88) are real, A is a real vector space, there is no distinction between upper and lower indices, and there is no need for arrows on the adjoint representation lines (3.84). However, the arrow on the adjoint representation generator (3.98) is necessary to dene correctly the overall sign. If we interchange the two legs, the right-hand side changes sign = (3.99)

(the generators for real representations are always antisymmetric). This arrow has no absolute meaning; its direction is dened by (3.98). Actually, as the right-hand side of (3.98) is antisymmetric under interchange of any two legs, it is convenient to replace the arrow in the vertex by a more symmetric symbol, such as a dot: = (3.100)

(Ti )jk iCijk = tr [Ti , Tj ]Tk ,

and replace the adjoint representation generators (Ti )jk by the fully antisymmetric structure constants iCijk . The factor i ensures their reality (in the case of
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.10. LIE ALGEBRA

37

hermitian generators Ti ), and we keep track of the overall signs by always reading indices counterclockwise around a vertex iCijk = i j . (3.101)
k

(3.102)

As all other clebsches, the generators must satisfy the invariance conditions (3.96): 0= + .

Redrawing this a little and replacing the adjoint representation generators (3.100) by the structure constants we nd that the generators obey the Lie algebra commutation relation = (3.103)

Ti Tj Tj Ti = iCijk Tk .

In other words, the Lie algebra is simply a statement that Ti , the generators of invariance transformations, are themselves invariant tensors. The invariance condition for structure constants Cijk is likewise 0= + + .

Rewriting this with the dot-vertex (3.100) we obtain = . (3.104)

This is the Lie algebra commutator for the adjoint representation generators, known as the Jacobi relation for the structure constants Cijm Cmkl Cljm Cmki = Ciml Cjkm . (3.105)

Hence the Jacobi relation is also an invariance statement, this time the statement that the structure constants are invariant tensors.
Remark 3.3 Sign convention for Cijk . A word of caution about using (3.103): vertex Cijk is an oriented vertex. If the arrows are reversed (matrices Ti , Tj multiplied in reverse order), the right-hand side gets an overall minus sign.
printed April 14, 2000 DasGroup/book/chapter/invar.tex 14apr2000

38

CHAPTER 3. INVARIANTS AND REDUCIBILITY

3.11

Other forms of Lie algebra commutators

Note that in our calculations we never need explicit generators; we use instead the projection operators for the adjoint representation. For representation they have the form
(PA )a b, =
a

a, b = 1, 2, . . . , n , = 1, . . . , d . (3.106)

The invariance condition for a projection operator is

i i i i

= 0.

(3.107)

a a Contracting with (Ti )a b and dening [d d ] matrices (Tb ) (PA )b , we obtain c c e c a e [Tba , Td ] = (PA )a b , e Td Te (PA )b , d
a b c d

(3.108)

This is a common way of stating the Lie algebra conditions for the generators in an arbitrary representation . For example, for U (n) the adjoint projection operator is simply a unit matrix (any hermitian matrix is a generator of unitary transformation, cf. chapter 8), and the right-hand side of (3.108) is given by U (n), SU (n) :
c c a a ] = b Td Tbc d . [Tba , Td

(3.109)

Another example is given by the orthogonal groups. The generators of rotations are antisymmetric matrices, and the adjoint projection operator antisymmetrizes generator indices: SO(n) : [Tab , Tcd ] = 1 2 gac Tbd gad Tbc gbc Tad + gbd Tac . (3.110)

Apart from the normalization convention, these are the familiar Lorentz group commutation relations (we shal return to this in chapter 9).

3.12

Irrelevancy of clebsches

As was emphasized in sect. 3.7, an explicit choice of clebsches is highly arbitrary; it corresponds to a particular coordinatization of the d -dimensional subspace V . For computational purposes clebsches are largely irrelevant. Nothing that a physicist wants to compute depends on an explicit coordinatization. For example, in QCD the physically interesting objects are color singlets and all color indices are summed over: one needs only an expression for the projection operators (3.87), not for the C s separately.
DasGroup/book/chapter/invar.tex 14apr2000 printed April 14, 2000

3.12. IRRELEVANCY OF CLEBSCHES

39

Again, a nice example is the Lie algebra generators Ti . Explicit matrices are often constructed (Gell-Mann matrices, Cartans canonical generators); however, in any singlet they always appear summed over the adjoint representation indices, as in (3.87). The summed combination of clebsches is just the gluon projection operator, a very simple object compared with explicit Ti matrices (PA is typically a combination of a few Kronecker deltas), and much simpler to use in explicit evaluations. As we shall show by many examples, all representation dimensions, casimirs, etc, are computable once the projection operators for the representations involved are known. Explicit clebsches are superuous from the computational point of view; we use them chiey to state general theorems, without recourse to any explicit realizations. However, if one has to compute non-invariant quantities, such as subgroup embeddings, explicit clebsches might be very useful. Gell-Mann [8] invented i matrices in order to embed SU (2) of isospin into SU (3) of the eightfold way. Cartans canonical form for generators, summarized by Dynkin labels of a representation is a very powerful tool in the study of symmetry breaking chains [71]. The same can be achieved with decomposition by invariant matrices (a nonvanishing expectation value for a direction in the dening space denes the little group of transformations in the remaining directions), but the tensorial technology in this context is still underdeveloped compared to the canonical methods.

printed April 14, 2000

DasGroup/book/chapter/invar.tex 14apr2000

40

CHAPTER 3. INVARIANTS AND REDUCIBILITY

DasGroup/book/chapter/invar.tex 14apr2000

printed April 14, 2000

Chapter 4

Recouplings
4.1 Couplings and recouplings

q onto a subspace . Clebsches discussed in sect. 3.7 project a tensor in V p V In practice one usually reduces a tensor step by step, decomposing a two-particle state at each step. We denote the Clebsches for by , P = . (4.1)

Here , , are representation labels, and the corresponding tensor indices are suppressed. Furthermore, if and are irreducible representations, the same clebsches can be used to project P = d d , (4.2)

and P = d d . (4.3)

Here the normalization factors come from P 2 = P condition. In order to draw the projection operators in a more symmetric way, we replace clebsches by 3-vertices: 1 a . (4.4)

In this denition one has to keep track of the ordering of the lines around the vertex. If in some context the birdtracks look better with two legs interchanged, one can use Yutsis (1962) notation . (4.5)

41

42

CHAPTER 4. RECOUPLINGS

While all sensible clebsches are normalized by the orthonormality relation (3.75), in practice no two authors ever use the same normalization for 3-vertices (in other guises known as 3j symbols, Gell-Mann matrices, Cartan roots, Dirac matrices, etc, etc). For this reason we shall usually not x the normalization

= a

a =

(4.6)

leaving the reader the option of substituting his favourite choice (such as a = 1 2 if the 3-vertex stands for Gell-Mann 1 2 i , etc). To streamline the discussion we shall drop the arrows and most of the representation labels in the remainder of this chapter - they can always easily be reinstated. The above three projection operators now take a more symmetric form: P = P = P = 1 a 1 a 1 a . (4.7)

In terms of 3-vertices, the completeness relation (3.73) is =

(4.8)

Any tensor can be decomposed by successive applications of the completeness relation: =

1 a 1 1 a a 1 1 1 a a a . (4.9)

=
,

=
,,

Hence, if we know clebsches for , we can also construct clebsches for . . . . However, there is no unique way of building up the clebsches; the above state can be equally well reduced by a dierent coupling scheme =
,,

1 1 1 a a a

(4.10)

DasGroup/book/chapter/recup.tex 14apr2000

printed April 14, 2000

4.1. COUPLINGS AND RECOUPLINGS

43

Consider now a process in which a particle in the representation interacts with a particle in the representation by exchanging a particle in the representation : (4.11) The nal particles are in representations and . To evaluate the contribution of this exchange to the spectroscopic levels of the particles system, we insert the Clebsch-Gordan series =

d d

(4.12)

By assumption is irreducible, so we have a recoupling relation between the exchanges in s and t channels:

(4.13)

as 3j coecients and as 6j coecients, committing us We shall refer to to no particular normalization convention. into the three-vertex and In atomic physics it is customary to absorb dene a 3j symbol (Racah 1942, Wigner 1931, 1959) = (1) 1 . (4.14)

Here = 1, 2, . . . , d etc are indices, , , representation labels, and the phase convention. Fixing a phase convention is a waste of time, as the phases cancel in summed-over quantities. All the ugly square roots one remembers from quantum into 3j symbols. The 6j symbols (Wigner mechanics come from sticking 1959) are related to our 6j coecients by (1)

(4.15)

The name 3n j coecient comes from atomic physics, where a recoupling involves 3n angular momenta j1 , j2 , . . . , j3n .
printed April 14, 2000 DasGroup/book/chapter/recup.tex 14apr2000

44 Skeletons 1j 3j 6j 9j 12j Vertex insertions

CHAPTER 4. RECOUPLINGS Self-energy insertions Total number 1 1 2 5 16

Table 4.1: Topologically distinct types of Wigner 3n j coecients.

Most of the textbook symmetries of and relations between 6j symbols are obvious from looking at the corresponding diagrams; others follow quickly from completeness relations. If we know the necessary 6j s, we can compute the level splittings due to single particle exchanges. In the next section we shall show that a far stronger claim can be made: given the 6j coecients, we can compute all multiparticle matrix element.

4.2

Wigner 3n j coecients

An arbitrary higher order contribution to a two-particle scattering process will give a complicated matrix element. The corresponding energy levels, crosssections, etc, are expresssed in terms of scalars obtained by contracting all tensor indices; diagrammatically they look like vacuum bubbles, with 3n internal lines. The topologically distinct vacuum bubbles in low orders are given in table 4.1. In group-theoretic literature, these diagrams are called 3n j symbols, and are studied in considerable detail. Fortunately, any 3n j symbol which contains as a sub-diagram a loop with, let us say, seven vertices

DasGroup/book/chapter/recup.tex 14apr2000

printed April 14, 2000

4.3. WIGNER-ECKART THEOREM Replace the dotted pair of vertices by the cross-channel sum (4.13):

45

(4.16)

Now the loop has six vertices. Repeating the replacement for the next pair of vertices, we obtain a loop of length ve: d =
,

d (4.17)

Repeating this process we can eliminate the loop altogether, producing 5-vertextrees times bunches of 6j coecients. In this way we have expressed the original 3n j coecients in terms of 3(n 1) j coecients and 6j coecients. Repeating the process for the 3(n 1) j coecients, we eventually arrive at the result that (3n j ) = products of . (4.18)

4.3

Wigner-Eckart theorem

For concreteness, consider an arbitrary invariant tensor with four indices:

T =

(4.19)

where , , and are representation labels, and indices and line arrows are suppressed. Now insert repeatedly the completeness relation (4.8) to obtain 1 a 1 a a 1 1 a2 d

=
,

(4.20)

In the last line we have used the orthonormality of projection operators - as in (4.13) or (4.23).
printed April 14, 2000 DasGroup/book/chapter/recup.tex 14apr2000

46

CHAPTER 4. RECOUPLINGS

In this way any invariant tensor can be reduced to a sum over clebsches (kinematics) weighted by reduced matrix elements:

< T > =

(4.21)

This theorem has many names, depending on how the indices are grouped. If T is a vector, then only the 1-dimensional representations (singlets) contribute
singlets

Ta =

(4.22)

If T is a matrix, and the representations , are irreducible, the theorem is called Schurs Lemma (for an irreducible representation an invariant matrix is either zero, or proportional to the unit matrix):

= Ta

(4.23)

If T is an invariant tensor operator, then the theorem is called the Wigner (1931, 1959) - Eckart (1930) theorem: (Ti )b a = d

(4.24)

(assuming that appears only once in Kronecker product). If T has many indices, as in our original example (4.19), the theorem is ascribed to Yutsis, Levinson and Vanagas (1962). The content of all these theorems is that they reduce spectroscopic calculations to evaluation of vacuum bubbles or reduced matrix elements (4.21). The rectangular matrices (C ) from (3.24) do not look very much like the clebsches from the quantum mechanics textbooks; neither does the Wigner-Eckart theorem in its birdtrack version (4.22). The dierence is merely a dierence of is written as notation. In the bra-ket formalism, a clebsch for 1 2 =< 1 2 m|1 m1 2 m2 > . (4.25)

DasGroup/book/chapter/recup.tex 14apr2000

printed April 14, 2000

4.3. WIGNER-ECKART THEOREM

47

Representing the [d d ] representation of a group element g diagrammatically by a black triangle

, (g ) = Dm,m

(4.26)

we can write the Clebsch-Gordan series (3.46) as =

1 2 Dm (g )Dm (g ) 1 m1 2 m2

=
m, , m 1

m 1 |1 m 2 m > . < 1 m1 2 m2 |1 2 > Dm m 1 (g ) < 1 2 m 1 2

An invariant tensor operator can be written as

< 2 m2 |Tm |1 m1 > =

(4.27)

In the bra-ket formalism, the Wigner-Eckart theorem (4.24) is written as

|1 m1 >=< 1 2 m2 |m1 m1 > T (, 1 2 ) , < 2 m2 |Tm

(4.28)

where the reduced matrix element is given by T (, 1 2 ) = 1 d2 1 d2

< n1 n1 |1 2 n2 >< 2 n2 |Tn |1 n1 > n1 ,n2 ,n

(4.29)

We do not nd the bra-ket formalism convenient for the group-theoretic calculations that will be discussed here. There is natural hierarchy to invariance groups that can perhaps already be grasped at this stage. Suppose that we have constructed the invariance group G1 which preserves primitives (17.3). Adding a new primitive, let us say a quartic invariant, means that we have imposed a new constraint; only those transformations of G1 which also preserve the additional primitive constitute G2 , the , , . Hence G2 is a subgroup of G1 , G2 G1 . Suppose invariance group of now that you think that the primitiveness assumption is too strong, and that some quartic invariant, let us say (3.34), cannot be reduced to a sum of tree invariants (3.38), ie. it is of form = + (rest of (3.38))

where is a new primitive, not included in the original list of primitives. By the above argument, G2 G1 . If G1 does not exist (the invariant relations are so stringent that there is no space on which they can be realized)
printed April 14, 2000 DasGroup/book/chapter/recup.tex 14apr2000

48

CHAPTER 4. RECOUPLINGS

DasGroup/book/chapter/recup.tex 14apr2000

printed April 14, 2000

Chapter 5

Permutations
The simplest example of invariant tensors are products of Kronecker deltas. On tensor spaces they represent index permutations. This is the way in which the symmetric group Sp , the group of permutations of p objects, enters into the theory of tensor representations. In this chapter, we introduce birdtracks notation for permutations, symmetrizations and antisymmetrizations and collect a few results which will be useful later on. These are the (anti)symmetrization expansion formulas (5.11) and (5.20), Levi-Civita tensor relations (5.32) and (5.34), the characteristic equations (5.54) and the invariance conditions (5.58) and (5.61).

5.1

Permutations in birdtracks

Operation of permuting tensor indices is a linear operation, and we can represent it by a [d d] matrix:
1 2 = b1 ...bp

a a ...aq dp ...d1 ,cq ...c2 c1

(5.1)

where (. . .) stands for the desired permutation of indices. As the covariant and contravariant indices have to be permuted separately, it is sucient to consider permutations of purely covariant tensors. For two-index tensors, there are two permutations identity: colour ip:
b c b = 1ab ,cd = a c d (12)ab ,cd = a b =

(5.2)

For three-index tensors, there are six permutations 111 000 b1 b2 b3 = 000 1 a 1 a 2 a 3 ,b 3 b 2 b 1 = a 111 1 a2 a3 111 000 000 111 11 00 b3 b2 b1 b2 b1 b3 000 111 (12)a1 a2 a3 , = a1 a2 a3 = 11 00 00 11 11 00 (23) = 000 111 11 00 (13) =
00 11 00 11 00 11 00 11 11 00 00 11 00 11 111 000 111 000 111 000

49

50

CHAPTER 5. PERMUTATIONS

11 00 (123) = 000 111 000 111 111 000 (132) = 00 11

000 111 111 000 000 111

000 111 111 000 000 111

(5.3)

Subscripts refer to the standard cycle notation. (In the above, and for the remainder of this chapter, we shall usually omit the arrows on the Kronecker delta lines.)

5.2

Symmetrization
1 b bp b1 b2 b1 b2 . . . ap a p + a 2 a 1 . . . a p + . . . 1 a2 p! 1 = + p + p + ... p p! p
... ...

The symmetric sum of all permutations Sa1 a2 ...ap ,bp ...b2 b1 =

yields the symmetrization operator S. In birdtrack notation, a white bar drawn across p lines will always denote symmetrization of the lines crossed. Factor 1/p! has been introduced in order that S satises the projection operator normalization S2 = S
... ... ...
... ...

...

...

S =

(5.4)

(5.5)

A subset of indices a1 , a2 , . . . aq , q < p can be symmetrized by symmetrization matrix S12...q (S12...q )a1 a2 ...aq ...ap ,bp ...bq ...b2 b1 1 b b1 b2 . . . aq q q ! a1 a2 = +
b1 b2 a . . . aq aq . . . ap q + ... p 2 a1 q +1
1 2

(5.6)
b b +1 b

...

...

S12...q =

(5.7)

Overall symmetrization also symmetrizes any subset of indices: SS12...q = S

...
... ...

...

(5.8)

Any permutation has eigenvalue 1 on the symmetric tensor space: S = S

... ... ... ...

...

...

(5.9)

Diagrammatically this means that legs can be crossed and un-crossed at will.
DasGroup/book/chapter/permu.tex 16/3-2000 printed April 14, 2000

5.2. SYMMETRIZATION

51

The denition (5.4) of the symmetrization operator as the sum of all p! permutations is inconvenient for explicit calculations - a recursive denition is more useful: Sa1 a2 ...ap ,bp ...b2 b1 = 1 b1 b1 Sa ...a ,bp ...b2 +a S ,bp ...b2 + . . . 2 a1 a3 ...ap p a1 2 p 1 1 + (21) + (321) + . . . + (p...321) S23...p p 1 + + + ... , p
... ... ... ... ... ...

S =
p ...

(5.10)

which involves only p terms. This equation says that if we start with the rst index, we end up either with the rst index, or the second index, and so on. The remaining indices are fully symmetric. Multiplying by S23 . . . p from the left, we obtain an even more compact recursion relation with two terms only: = 1 p + (p 1) p-1 . (5.11)

...

As a simple application, consider computation of a contraction of a single pair of indices:

p-1 p-2

...

=
... ...

1 p

...

p-1

...

+ (p 1)
... ...

...

...

= Sap ap1 ...a1 ,b1 ...bp1 ap =

n+p1 p n+p1 Sap1 ...a1 ,b1 ...bp1 . p

For a contraction in (p k ) pairs of indices we have

...
p k

...

...

(5.12)

(n + p 1)!k ! p!(n + k 1)!

(5.13)

...

...

The trace of the symmetrization operator yields the number of independent components of fully symmetric tensors: n+p1 p (n + p 1)! . p!(n 1)!

...

dS = tr

For example, for two-index symmetric tensors dS = n(n + 1) . 2 (5.15)

DasGroup/book/chapter/permu.tex 16/3-2000

printed April 14, 2000

...

...

...

(5.14)

52

CHAPTER 5. PERMUTATIONS

5.3

Antisymmetrization
1 b b1 b2 b1 b2 a . . . ap ,bp a . . . ap p + ... 1 a2 2 a1 p! 1 = p + p ... p p! p
... ...

The alternating sum of all permutations Aa1 a2 ...ap ,bp ...b2 b1 =

yields the antisymmetrization projection operator A. In birdtrack notation, antisymmetrization of p lines will always be denoted by a black bar drawn across the lines. As in the previous section A2 = A
... ... ...
... ...

= =

...

...

A =

(5.16)

...

and in addition SA = 0
... ... ...

= 0 = = 0. (5.18)
... ... ...

...

A transposition has eigenvalue 1 on the antisymmetric tensor space (i,i+1) A = A

... ... ... ...

...

...

...

(5.17)

...

...

(5.19)

Diagrammatically this means that legs can be crossed and uncrossed at will, but with a factor of 1 for a transposition of any two neighbouring legs. As in the case of symmetrization operators, the recursive denition is often computationally convenient
p ...

= =

1 p 1 p

...
...

... ...

p-1

This is useful for computing contractions such as

p-1 p-2

=
... ...

np+1 p

b1 ...bp1 a Aaa p1 ...a1

n p + 1 b1 ...bp1 Aap1 ...a1 . p

...

DasGroup/book/chapter/permu.tex 16/3-2000

...

(p 1) p-1

...

... ...

...

...

(5.20)

...

(5.21)
printed April 14, 2000

5.4. LEVI-CIVITA TENSOR

53

The number of independent components of fully antisymmetric tensors is given by n np+1np+2 ... p p1 1

dA = tr A =

n! np p!(n p)! = 0 n p. =

...

(5.22)

For example, for rank-two antisymmetric tensors the number of independent components is n(n 1) . 2 Tracing (p k ) pairs of indices yields dA =
...
p k

(5.23)

k !(n k )a5.23.b8 p!(n p)!

(5.24)

...

...

2 1 has non-vanishing components only if all The antisymmetrization tensor Aap 1 a2 ... lower (or upper) indices dier from each other. If the dening dimension is smaller than the number of indices, the tensor A has no non-vanishing components

...

...

b ...b b

1 2

=0

if p > n .

(5.25)

...

...

This identity implies that for p > n, not all combinations of p Kronecker deltas are linearly independent. A typical relation is the p = n + 1 case
...

...

0=

...
1 2 n+1

...

... ...

...

...

... .

(5.26)

for example, for n = 2 we have

f e d

n = 2; 0 =
a b c

f e d f e d f e d f e d f e d f e d 0 = a b c a c b b a c + b c a + c a b c b a . (5.27)

5.4

Levi-Civita tensor

An antisymmetric tensor with n indices in dening dimension n has only one independent component (by (5.22) dn = 1). The clebsches (3.72) are in this case proportional to the Levi-Civita tensor
n ...a2 a1 (CA )a 1

= C

...
a1 a2 an

an ...a2 a1

= =

a1 a2 an

(CA )a1 a2 ...an ,1 = C

printed April 14, 2000

a1 a2 ...an

(5.28)
DasGroup/book/chapter/permu.tex 16/3-2000

...

54

CHAPTER 5. PERMUTATIONS

with 12...n = 12...n = 1. This diagrammatic notation for the Levi-Civita tensor was introduced by R. Penrose [1]. The normalization factors C are physically irrelevant1 they adjust the phase and the overall normalization in order that the Levi-Civita tensors satisfy the projection operator (3.73) and orthonormality (3.75) conditions: 1 b b ...b a1 a2 ...an N! 1 2 n
... ...

= Ab1 b2 ...bn ,an ...a2 a1 = (5.32)

... ...

1 a a ...a a1 a2 ...an N! 1 2 n
...

= 1,1 = 1 = 1 (5.33)

Given n dimensions we cannot label more than n indices, so Levi-Civita tensors satisfy
...

0=
...
1 2 3 n +1

(5.34)

For example, for two colours n=2:0 = + (5.35)

d d d 0 = a bc b ac + c ab .

This is actually the same as the completeness relation (5.32), as can be seen by contracting (5.35) with cd and using n=2: =
b ac bc = a .
1

= 1/2 (5.36)

With our conventions C= in(n 1)/2 . n! (5.29)

The phase factor arises from the hermiticity condition (??) for clebsches (remember that indices are always read in the counterclockwise order around a diagram)
...
...
a1 a2 an

= = i

a1 a2 an

a1 a2 ...an

an ...a2 a1 .

(5.30)

Transposing the indices a1 a2 ...an = a2 a1 ...an = . . . = (1)n(n1)/2 an ...a2 a1 (5.31)

yields = n(n 1)/2. The factor (n!)1/2 is needed for the projection operator normalization (??)
DasGroup/book/chapter/permu.tex 16/3-2000 printed April 14, 2000

5.5. DETERMINANTS

55

This relation is one of a series of relations obtained by contracting indices in the completeness relation (5.32) and substituting (5.24): an ...ak+1 bk ...b1 an ...Ak+1 ak ...a1
...

= (n k )!k !Abk ...b1 ,a1 ...ak = k !(n l)! n! (5.37)

...

...

...

Such identities are familiar from relativistic calculations (n = 4):

gf e abcd agf e = bcd fe abcd abf e = 2cd e abcd abce = 6d

...

abcd abcd = 24 where the generalized Kronecker delta is dened by 1 b1 b2 ...bp a a ...a = Aa1 a2 ...ap ,bp ...b2 b1 . p! 1 2 p

(5.38)

(5.39)

5.5

Determinants

Consider a [np np ] matrix M dened by a direct product of [n n] matrices b Ma M = Ma1 a2 ...ap ,bp ...a2 a1 = Ma1
2 ...M p b1 Ma2 ap b b

M where

(5.40)

...

...

...

...

b = Ma

(5.41)

is given by The trace of the antisymmetric projection of M

tr p AM

a b d = Aabc...d ,d ...c b a Ma Mb . . . Md 1

... ...

(5.42)

The subscript p on tr p (. . .) distinguishes the traces on [np np ] matrices M from the [n n] matrix trace tr M . To derive a recursive evaluation rule for tr p AM use (5.20) to obtain 1 = (p 1) (5.43) p
... ...

...

printed April 14, 2000

DasGroup/book/chapter/permu.tex 16/3-2000

56 Iteration yields
...

CHAPTER 5. PERMUTATIONS

=
p-1
...

p-2 ...

+ ...

...

(5.44)

b we obtain Contracting with Ma

p k 1 (tr pk AM ) tr M k k=1 (1)

This formula enables us to compute recursively all tr p AM as polynomials in traces of powers of M: tr 0 AM tr 1 AM = 1, = = tr 2 AM = = tr 3 AM = = tr 4 AM = = tr M 1 , 2 1 (tr M )2 tr M 2 , 2 1 + 3 1 (tr M )3 3(tr M )(tr M 2 ) + 2tr M 3 , 3! 1 + 4 1 (tr M )4 6(tr M )2 tr M 2 4 +3(tr M 2 )2 + 8tr M 3 tr M 6tr M 4 . (5.46)

b are [n n] matrices) the antisymmetrized trace is the determinant For p = n (Ma a1 a2 an Mb . . . Mb . detM = tr n AM = Aa1 a2 ...an ,bn ...b2 b1 Mb n 1 2

The coecients in the above expansions are simple combinatoric numbers. A general term for (tr M 1 )1 (tr M 2 )s , with 1 loops of length 1 , 2 loops of length 2 and so on, is divided by the number of ways in which this pattern may be obtained 2 :
1 2 1 2
2

...

s s 1 !2 ! . . . s ! .

A.D. Kennedy, unpublished

printed April 14, 2000

DasGroup/book/chapter/permu.tex 16/3-2000

...

... ...

=
...

...

. . . (1)p

...

...

...

(5.45)

(5.47)

(5.48)

(5.49)

(5.50)

(5.51)

5.6. CHARACTERISTIC EQUATIONS

57

5.6

Characteristic equations

We have noted that the dimension of the antisymmetric tensor space is zero in n < p. This is rather obvious; antisymmetrization allows each label to be used at most once, and it is impossible to label more legs than there are labels. In terms of the antisymmetrization operator this is given by the identity A=0 if p > n . (5.52) This trivial identity has an important consequence: it guarantees that any [n n] matrix satises a characteristic (or Hamilton-Cayley) equation. Take p = n + 1 b n index pairs of A: and contract with M1
a1 a2 an Mb . . . Mb Aca1 a2 ...an ,bn ...b2 b1 d Mb n 1 2
c a

= 0 = 0. (5.53)

We have already expanded this in (5.44). For p = n + 1 we obtain the characteristic equation
n

0 =

(1)k (tr nk AM )M k ,
k=0 n

0 = M (tr M )M n1 + tr 2 M n2 . . . + (1)n detM 1 .

...

(5.54)

5.7

Fully (anti)symmetric tensors

As we shall often use fully symmetric and antisymmetric tensors, it is convenient to introduce special birdtrack symbols for them. We shall denote the fully symmetric tensors by small circles dabc...f =
.. .

(5.55)

a b c ... f

A symmetric tensor dabc...d = dbac...d = dacb...d = . . . satises Sd = d

. ..
. ..

=
...
...

(5.56)

If this tensor is also an invariant tensor, the invariance condition (??) can be written as 0 = = = p
printed April 14, 2000

+ +

+ + (p = number of indices) . (5.57)

DasGroup/book/chapter/permu.tex 16/3-2000

58

CHAPTER 5. PERMUTATIONS

Hence the invariance condition for symmetric tensors is 0=

. ..

(5.58)

...

The fully antisymmetric tensors with odd numbers of legs will be denoted by black dots fabc...d =
a b c .. . d

(5.59)

If the number of legs is even, an antisymmetric tensor is anticyclic fabc...d = fbc...da , (5.60)

and the birdtrack notation must distinguish the rst leg. A black dot is inadequate for the purpose. A bar, as for the Levi-Civita tensor (5.28), or a semicircle or similar notation xes the problem. For antisymmetric tensors, the invariance condition can be stated compactly as 0=
. ..

(5.61)

...

5.8

Young tableaux, Dynkin labels

It is standard to

DasGroup/book/chapter/permu.tex 16/3-2000

printed April 14, 2000

Chapter 6

Casimir operators
The construction of invariance groups developed elsewhere in this monograph is self-contained, and none of the material covered in this chapter is necessary for understanding the remainder of the report. We have argued in sect. 4.2 that all relevant group-theoretic numbers are given by vacuum bubbles (reduced matrix elements, 3n j coecients, etc), and we have described the algorithms for their evaluation. That is all that is really needed in applications. However, one often wants to cross-check ones calculation against the existing literature. In this chapter we discuss why and how one introduces casimirs (or Dynkin indices), we construct independent Casimir operators for the classical groups, and nally we compile values of a few frequently used casimirs. Our approach emphasizes the role of primitive invariants in constructing representations of Lie groups. Given a list of primitives, we present a systematic algorithm for constructing invariant matrices Mi and the associated projection operators (3.45). In the canonical, Cartan-Killing approach one faces a somewhat dierent problem. Instead of the primitives, one is given the generators Ti explicitly, and no other invariants. Hence the invariant matrices Mi can be constructed only from contractions of generators; typical examples are matrices M2 = birdTrack, , M4 = birdTrack, . . . (6.1)

where , could be any representations, reducible or irreducible. Such invariants are called Casimir operators. What is a minimal set of Casimir operators, sucient to reduce any representation to its irreducible subspaces? (Such bases can be useful, as the corresponding r Casimir operators label uniquely each irreducible representation by their eigenvalues 1 , 2 , . . . r ). The invariance condition for any invariant matrix (3.28) is 0 = [Ti , M ] = birdTrack birdTrack so all Casimir operators commute M2 M4 = birdTrack = birdTrack = M4 M2 59 (6.3) (6.2)

60

CHAPTER 6. CASIMIR OPERATORS

and, according to sect. 3.5, can be used to simultaneously decompose the representation . If M1 , M2 . . . have been used in the construction of projection operators (3.45), any matrix polynomial f (M1 , M2 . . .) takes value f (1 , 2 , . . .) on the irreducible subspace projected by Pi , so polynomials induce no further decompositions. Hence it is sucient to determine the nite number of Mi s which form a polynomial basis for all Casimir operators (6.1). Furthermore, as we show in the next section, it is sucient to restrict the consideration to the symmetrized casimirs. This observation enables us to explicitly construct in sect. 6.2 a set of independent casimirs for each classical group in sect. 6.2 . Exceptional groups are harder.

6.1 6.2 6.3

Casimirs and Lie algebra Independent casimirs Casimir operators

Most physicists would not refer to tr X k as a casimir. Casimirs (1931) quadratic operator and its generalizations (Racah 1950) are [d d ] matrices

6.4

Dynkin indices

As we have seen so far, there are many ways of dening casimirs; in practice it is usually quicker to directly evaluate a given birdtrack diagram than to relate it to standard casimirs. Still, it is good to have an established convention, if for no other reason than to be able to cross-check ones calculation against the tabulations available in the literature.

6.5

Quadratic, cubic casimirs

As the low-order Casimir operators appear so often in physics, it is useful to list them and their relations.

6.6 6.7 6.8

Quartic casimirs Sundry relations between quartic casimirs Identically vanishing tensors
printed April 14, 2000

DasGroup/book/chapter/casimirs.tex 14apr2000

6.9. DYNKIN LABELS

61

6.9

Dynkin labels

It is standard to identify a representation of a simple group of rank r by its Dynkin labels, a set of r integers (a1 a2 . . . ar ) assigned to the simple roots of the group by the Dynkin diagrams. The Dynkin diagrams, table 2.1, are the most concise summary of the Cartan-Killing construction of the semi-simple Lie Algebras. We list them here only to facilitate the identication of the representations, and do not attempt to derive or explain them. Dynkins canonical construction is described in Slanskys (1981) review. In this report we develop only the tensor techniques for constructing representations. However, in order to help the reader connect the two approaches, we will state the correspondence between the tensor representations (identied by the Young tableaux) and the canonical representations (identied by the Dynkin labels) for each group separately, in the appropriate chapters.

printed April 14, 2000

DasGroup/book/chapter/casimirs.tex 14apr2000

62

CHAPTER 6. CASIMIR OPERATORS

DasGroup/book/chapter/casimirs.tex 14apr2000

printed April 14, 2000

Chapter 7

Group integrals
In this chapter we discuss evaluation of group-theoretic integrals of form
g d e dgGb a Gc . . . G f Gh

(7.1)

where Gb a is the [nn] dening matrix representation of g Gc and the integration is over the entire range of g . As always, we assume that Gc is a compact Lie group, and Gb a is unitary. The integral (7.1) is dened by two rules: 1. As the overall normalization is rarely of interest in calculation of physically interesting group-averaged quantities, we normalize the group volume to 1: dg = 1 (7.2)

2. How do we dene dgGb a ? The action of g Gc is to rotate a vector xa 1 a b into xa = Ga xb The averaging smears x in all directions, hence the second integration rule dgGb a = 0, Gisanon trivial representationof g (7.3)

simply states that the average of a vector is zero. A representation is trivial if G = 1 for all colour rotations g. In this case no averaging is taking place, and the rst integration rule (7.2) applies. What happens if we average a pair of vectors x, y ? There is no reason why a a pair should average to zero; for example, we know that |x|2 = a xa x a = xa x is invariant (we are considering only unitary representations), so it cannot have a vanishing average. Therefore, in general
c dgGb a Gd = 0 .

(7.4)

To get a feeling for what the right-hand side looks like, let us work out a few examples 63

64

CHAPTER 7. GROUP INTEGRALS

7.1

Group integrals for arbitrary representations

Let Gb a be the dening [nxn] matrix representation of SU (n). The dening representation is non-trivial, so it averages to zero by (7.3). The rst non-vanishing average is the integral over G . G is the matrix representation of the action of g on the conjugate vector space, which we write as (3.4)
a Ga b = (G )b .

(7.5)

7.2

Characters

Physics calculations (such as lattice gauge theories) often involve group invariant quantities formed by contracting G with invariant tensors. Such invariants are of a Gb , where h stands for any invariant tensor. The trace of the form tr (hG) = Hb a an irreducible [d d] matrix representation of g is called the character of the representation: (g ) = tr G = Ga a. The character of the conjugate representation is (g ) = (g ) = tr G = (G )a a. (7.7) (7.6)

c Contracting (??) with two arbitrary invariant [d d] matrices ha d and (f )b we obtain the character orthonormality relation dg (hg ) (gf ) =

1 (hg ) d , irreducible representations (7.8)

dg birdTrack =

1 birdTrack d

The character orthonormality tells us that if two group variant quantities share a GG pair, the group averaging sews them into a single group invariant quantity. The replacement of Gb a by the trace (h g ) does not mean that any of the tensor b index structure is lost; Ga can be recovered by dierentiating Gb a = d (h g ) . dhb (7.9)

The birtracks and the characters are two equivalent notations for evaluating group integrals.

7.3

Examples of group integrals

DasGroup/book/chapter/grint.tex 4/7-98

printed April 14, 2000

Chapter 8

Unitary groups
(P. Cvitanovi c, H. Elvang, and A. D. Kennedy) U (n) is the group of all transformations which leave the norm of a complex a q b q invariant. For U (n) there are no other invariant tensors vector qq = b a beyond those constructed of products of Kronecker deltas. They can be used to decompose the tensor representations of U (n). For purely covariant or contravariant tensors, the symmetric group can be used to construct the Young projection operators. In sects. 8.18.2 we show how to do this for 2- and 3-index tensors by constructing the appropriate characteristic equations. For tensors with more indices it is easier to construct the Young projection operators directly from the Young tableaux. We use the projection operators so constructed to evaluate characters and 3-j coecients of U (n). For mixed tensors reduction also involves index contractions and the symmetric group methods alone do not suce. In sects. 8.88.10 the mixed U (n) tensors are decomposed by the projection operator techniques introduced in chapter 3.

8.1

Two-index tensors

Consider 2-index tensors q (1) q (2) V 2 . According to (5.1), all permutations are represented by invariant matrices. Here there are only two permutations, the identity and the color ip (5.2) = . 2 = ( + 1)( 1) = 0 . (8.1)

The color ip satises = 1, (8.2)

Hence the roots are 1 = 1, 2 = 1, and the corresponding projection operators (??) are P1 = 1 1 (1)1 = (1 + ) = 1 (1) 2 2 65 + , (8.3)

66 P2 = 1 1 1 = (1 ) = 1 1 2 2

CHAPTER 8. UNITARY GROUPS . (8.4)

We recognize the symmetrization, antisymmetrization operators (5.4), (5.16); P1 = S, P2 = A, with subspace dimensions d = n(n + 1)/2, d2 = n(n 1)/2. In other words, under general linear transformations the symmetric and the antisymmetric parts of a tensor xab transform separately: x = Sx + Ax , 1 1 xab = (xab + xba ) + (xab xba ) 2 2 = + The Dynkin indices for the two representations follow by (??) from 6j s: birdTrack =
1

(8.5)

= =

1 N birdTrack = 2 2 2 2 N d1 + n N 2 (n + 2)
1

(8.6) = CA = 2n was computed in

(The dening representation Dynkin index sect. ??). The result is = n+2 , 2n
2

n2 2n

(8.7)

8.2

Three-index tensors

Three-index tensors can be reduced to irreducible subspaces by adding the third index to each of the 2-index subspaces, the symmetric and the antisymmetric. The results of this section are summarized in table 8.2. We mix the third index into the symmetric 2-index subspace using the invariant matrix Q = S12 (23) S12 = . (8.8)

Here projection operators S12 ensure the restriction to the 2-index symmetric subspace, and the transposition (23) mixes in the third index. To nd the characteristic equation for Q, we compute Q2 : Q2 = S12 (23) S12 (23) S12 = = = 1 1 S12 + Q . 2 2 = 1 2 1 S12 + S12 (23) S12 2 + (8.9)

Hence Q satises (Q 1)(Q + 1/2)S12 = 0 ,

DasGroup/book/chapter/unitary.tex 14apr2000

(8.10)
printed April 14, 2000

8.2. THREE-INDEX TENSORS and the corresponding Young projection operators are P1 = Q+ 1 21 S12 = 1+ 1 2 = P2 = Q1 S12 = 1 2 1 1 + (123) + 1 S12 = S 3 (23) 1 ( + + ) = 3 4 4 S12 A23 S12 = 3 3 .

67

(8.11) (8.12)

Hence the symmetric 2-index subspace combines with the third index into a symmetric 3-index subspace (5.14) and a mixed symmetry subspace with dimensions d1 = tr P1 = d2 = tr P2 = n(n + 1)(n + 2) 3! 4 3 = n(n2 1) . 3 (8.13) (8.14)

The antisymmetric 2-index subspace can be treated in the same way using invariant matrix Q = A12 (23) A12 = . (8.15)

The resulting projection operators for the antisymmetric and mixed symmetry 3-index tensors are given in table 8.2. Symmetries of the subspace are indicated by the corresponding Young tableaux, table 8.2. For example, we have just constructed
1 2 3

= =

1 2 1 2 3 3

n2 (n + 1) 2

4 3 n(n + 1)(n + 2) n(n2 1) + . 3! 3 + Reduction stu

(8.16)

Table 8.1: Table summarizing the reduction procedure. Old table 6.4 Table 8.2: Table summarizing the reduction procedure.

printed April 14, 2000

DasGroup/book/chapter/unitary.tex 14apr2000

68

CHAPTER 8. UNITARY GROUPS Old table 6.5 Table 8.3: Table summarizing the reduction procedure.

8.3

Young tableaux

As we have seen in the above examples, the projection operators for two-index and three-index tensors can be constructed using the characteristic equations. This, however, becomes cumbersome when applied to tensors with more than 3 indices. We now show how to construct Young projection operators for the irreducible representations of U (n) directly from the Young tableaux.

8.3.1

Denitions

Partition k boxes into D subsets, so that the mth subset contains i boxes. Order the partition so the set = [1 , 2 , . . . , D ] fulls 1 2 . . . D 1 and D i=1 i = k . The diagram obtained by drawing the D rows of boxes on top of each other, left alligned, starting with 1 , is called a Young diagram Y. Examples: For k = 4 the ordered partitions for k = 4 are [4], [3, 1], [2, 2], [2, 1, 1] and [1, 1, 1, 1]. For the k = 7 partition [4, 2, 1] the Young diagram is for the k = 3 partition [1, 1, 1] it is . and

A box in a Young diagram can be assigned a coordinate (i, j ) such that Y = {(i, j ) Z2 |1 j i }. Here i label the rows and j the columns. Inserting a number from the set {1, . . . , n} into every box of a Young diagram Y in such a way that numbers increase when reading a column from top to bottom and numbers do not decrease when reading a row from left to right yields a Young tableaux Ya . The subscript a labels dierent tableaux derived from a given Young diagram, that is dierent admissible ways of inserting the numbers into the boxes. Denoting the number in the (i, j )th box by a (i, j ) we have Ya = {(a (i, j )) {1, . . . , n}k | (i, j ) Y, a (i, j + 1) a (i, j ), a (i + 1, j ) > a (i, j )} A Young tableaux with numbers inserted as above is called a standard arrangement. The monotonically ordered arrangement Ya = {(a (i, j )) {1, . . . , k} | (i, j ) Y, a (i, j + 1) > a (i, j ), a (i + 1, j ) > a (i, j )} is called a k -standard arrangement .
DasGroup/book/chapter/unitary.tex 14apr2000 printed April 14, 2000

8.3. YOUNG TABLEAUX

69

In the following we denote by Young diagram Y a box diagram without numbers, and by Young tableaux Ya a diagram lled with a standard arrangement. Often we simplify the notation by using Y, Z, . . . to denote both Young diagrams and Young tableaux. The transpose diagram Yt is obtained from Y by interchanging rows and columns. For example, the transpose of [3, 1] is [2, 1, 1], or in the alternative labelling () An alternative labelling of a Young diagram is to list the number bm of columns with m boxes as (b1 b2 . . .). Having k boxes we must have k m=1 mbm = k . As an example we see that [4, 2, 1] and (21100 . . .) label the same Young diagram. Similarly for [2, 2] and (020 . . .). This notation is handy when considering Dynkin labels.

8.3.2

SU (n) Young tableaux

We now show that a Young tableau with no more than n rows corresponds to an irreducible representation of SU (n). A k -index tensor is represented by a Young diagram with k boxes one may think of this as a k particle state. For SU (n) there are n one-particle states available and the irreducible k -particle states correspond to a Young tableaux obtained by inserting the numbers 1, . . . , n into the k boxes of the Young diagrams. Boxes in a row correspond to indices that are symmetric under interchanges (symmetric multiparticle states), and boxes in a column correspond to indices antisymmetric under interchanges (antisymmetric multiparticle states). Consider the reduction of a two-particle state, that is a two-index tensor, into a symmetric and an antisymmetric state (8.5). Using Young diagrams we would write this as = (8.17)

For the n = 2 case the Young tableaux of SU(2) are:

1 1 , 1 2 , 2 2

and

1 2

(8.18)

The dimension of an irreducible representation of SU (n) is found by counting the number of standard arrangements. Thus for SU(2) the symmetric state is 3 dimensional whereas the antisymmetric state is 1 dimensional, in agreement with the formulas (5.4) and (5.16) for the dimensions of the symmetry operators. In section sect. 8.4.1 we shall state and prove the dimension formula for a general irreducible U (n) representation. For SU (n) columns cannot contain more than n boxes, as it is impossible to antisymmetrize more than n labels. Columns of n boxes can be contracted away by means of the Levi-Civita tensor sect. ??. Hence the highest column is of height n 1, which is also the rank of SU (n). Furthermore, for SU (n) a column with k boxes (antisymmetrization of covariant k indices) can be converted by contraction
printed April 14, 2000 DasGroup/book/chapter/unitary.tex 14apr2000

70

CHAPTER 8. UNITARY GROUPS

with the Levi-Civita tensor into a column of (nk ) boxes (corresponding to (nk ) contravariant indices). This operation associates with each tableau a conjugate representation. Thus the conjugate of a SU (n) Young diagram Y is constructed from the missing pieces needed to complete the rectangle of n rows:

(8.19)

That is, add squares below the diagram of Y such that the resulting gure is rectangle with height n and width of the top row in Y. Remove the squares corresponding to Y and rotate the rest by 180 degrees. The result is the conjugate diagram of Y. For example, for SU (6), representation (20110)

(8.20)

has as its conjugate representation (01102). In general, the SU (n) representations (b1 b2 . . . bn1 ) and (bn1 . . . b2 b1 ) are conjugate. For example, if (10 . . . 0) stand for the dening representation, then its conjugate is represented by (00 . . . 01), ie. a column of n 1 boxes. We prefer to keep the conjugate representations conjugate, rather than replacing them by columns of (n 1) dening representations, as this will give us SU (n) expressions valid for any n.

8.3.3

Reduction of direct products

We now state the rules for reduction of direct products such as (8.17) in terms of Young diagrams: Draw the two diagrams next to one another and place in each box of the second diagram an ai , i = 1, . . . , k, such that the boxes in the rst row all have a1 in them, second row boxes have a2 in them etc. The boxes of the second diagram are now added to the rst diagram to create new diagrams in accordance to the rules 1. Each diagram must be a Young diagram. 2. The number of boxes in the new diagram must be equal to the sum of the number of boxes in the original two diagrams. 3. For SU(n) no diagram has more than n rows. 4. Making a journey through the diagram starting with the top row and entering each row from the right, at any point the number of ai s encountered in any of the attached boxes must not exceed the number of previously encountered ai1 s.
DasGroup/book/chapter/unitary.tex 14apr2000 printed April 14, 2000

8.4. YOUNG PROJECTION OPERATORS

71

5. The numbers must not increase when reading across a row from left to right. 6. The numbers must decrease when reading a column from top to bottom. The rules 4-6 ensure that states which were previously symmetrized are not antisymmetrized in the product and vice versa, and to avoid counting the same state twice.

8.4

Young projection operators

Given a Young tableau Y of U (n) with an k -standard arrangement we construct the corresponding Young projection operator PY in birdtrack notation by identifying each box in the diagram with a directed line. The operator PY is a block of symmetrizers to the left of a block of antisymmetrizers, all imposed on the n lines. The blocks of symmetry operators are dictated by the Young diagram whereas the attachment of lines to these operators follows from the k -standard arrangement. For a Young diagram Y with s rows and t columns we refer to the rows as S1 , S2 , . . . ,Ss and to the columns as A1 , A2 , . . . ,At . Each symmetry operator in PY is associated to a row/column in Y, hence we label a symmetry operator after the corresponding row/column,

S1

A1

A1 A2 A3 A4 A5

A2

...

S2 S3

...

S1

= Y

S2

A3

(8.21)

A4 S3 A5

We denote by |Si | or |Ai | the length of a row or column, respectively, that is the number of boxes it contains. Thus |Ai | also denotes the number of lines entering the antisymmetrizer Ai . In the above example we have |S1 | = 5, and |A2 | = 3, etc. An example of the construction of the Young projection operators: The Young tells us to use one symmetrizer of length three, one of length one, diagram one antisymmetrizer of length two, and two of length one. There are three distinct k -standard arrangements, each corresponding to a projection operator
1 2 3 4 1 2 4 3
printed April 14, 2000

= Y = Y

(8.22) (8.23)
DasGroup/book/chapter/unitary.tex 14apr2000

72
1 3 4 2

CHAPTER 8. UNITARY GROUPS = Y , (8.24)

where Y is a normalization constant. We use the convention that if the lines pass straight through the symmetry operators they appear in the same order as they entered. More examples of Young projection operators are given in sect. 8.5. The normalization is given by Y =
s i=1 |Si |! t j =1 |Aj |!

|Y|

(8.25)

where |Y| is a combinatoric number calculated by the following hook rule. For each box of the Young diagram Y write the number of boxes below and to the left of the box (including the box itself once). Then |Y| is the product of the numbers in all the boxes. For instance, Y= 4 3 1
2 1 6 5 3 1

(8.26)

has |Y| = 6!3. We prove that this is the correct normalization in appendix B. The normalization only depends on the Young diagram, not the particular tableau. For multidimensional irreducible representations the Young projection operators constructed as above will generally be dierent from the ones constructed from characteristic equations, see sects. 8.18.2, but the dierence amounts to a choice of basis, so they are equivalent. We prove in appendix B that the above construction indeed yields well-dened projection operators. Some of the properties of the Young projection operators:
2 =P . The Young projection operators are indeed projections, PY Y

The Young projection operators are orthogonal : If Y and Z are two dierent k -standard arrangement, then PY PZ = 0 = PZ PY . For a given k the Young projection operators constitute a complete set such that 1 = PY , where the sum is over all k -standard arrangements Y with k boxes and 1 is the [k k ] unit matrix. The dimension dY = tr PY of a Young projection operator PY can be calculated directly by tracing PY and expanding it using (5.11) and (5.20). In practice, this is unnecessarily laborious. Instead, we oer two simple ways of computing the dimension of an irreducible representation from its Young diagram.

8.4.1

A dimension formula

Let fY (n) be the polynomial in n obtained from the Young diagram Y by multiplying the numbers written in the boxes of Y, according to the following rules: 1. The upper left box contains an n.
DasGroup/book/chapter/unitary.tex 14apr2000 printed April 14, 2000

8.4. YOUNG PROJECTION OPERATORS 2. The numbers in a row increases by one when reading from left to right.

73

3. The numbers in a column decrease by one when reading from top to bottom. Hence, if k is the number of boxes in Y, fY (n) is a polynomial in n of degree k . For U(n) the dimension of the irreducible representation labelled by the Young diagram Y is dY = fY (n) |Y | (8.27)

Example: With Y = [4,2,1] we have

n+1 n+2 n+3 n

fY (n) = and

n-1 n-2

= n2 (n2 1)2 (n2 4)(n + 3),

|Y | = 3 1
1

6 4 2 1

= 144.

hence dY = n2 (n2 1)2 (n2 4)(n + 3) 144 (8.28)

This dimension formula is derived in appendix B. Next we give an intuitive interpretation of what this formula means.

8.4.2

Dimension as the number of strand colorings

The dimension of a Young projection operator PY of SU (n) can be calculated by counting the number of distinct ways in which the trace diagram of a Young projection operator can be colored. Draw the trace of the Young projection operator. Each line is strand, a closed path which we draw as passing straight through the symmetry operators. Order the paths in accordance to the k -standard arrangement (see example). The lines are colored in this order. Having n colors we can color the rst line in n dierent ways. Rule 1: If a path, which could be colored in k ways, enters an antisymmetrizer, the lines below it can be colored in k 1, k 2, . . . ways. Rule 2: If a path, which could be colored in k ways, enters a symmetrizer, the lines below it can be colored in k + 1, k + 2, . . . ways. Label each path with the number of ways it can be colored. The number of ways to color the trace diagram is the product of all the factors obtained above; but this is simply fY (n) dened in sect. 8.4.1. An example:
printed April 14, 2000 DasGroup/book/chapter/unitary.tex 14apr2000

74

CHAPTER 8. UNITARY GROUPS

n-1 n+1 n+2 n

dY =

1 2 3 6 fY (n) 1 = tr 4 5 7 = |Y| |Y| 8

n+3 n n+1 n-2

(8.29)

8.5

Reduction of tensor products

We now apply the rules for decomposition of direct products of Young diagrams/tableaux to several explicit examples. We use the tableaux to compute the dimensions and construct the Young projection operators. We have already treated the decomposition of the two-index tensor into the symmetric and the anti-symmetric tensors, but we shall reconsider the three-index tensor, since the projection operators will be dierent from those derived from the characteristic equations in sect. 8.2.

8.5.1

Three- and four-index tensors

According to the rules in sect. 17.10, the three-index tensor reduces to

1 3 = 1 2 3 1 2 1 3 2. (8.30) 3 2 3 The corresponding dimensions and Young projection operators are given in table 8.5.1. For simplicity, we neglect the arrows on the lines where this leads to no confusion. Let us check the completeness by an explicit computation. In the sum of the fully symmetric and the fully antisymmetric tensors all the odd permutions cancel, and we are left with 1 + = ( + + ) (8.31) 3 Expanding the two tensors of mixed symmetry, we obtain 2 1 1 4 = + . (8.32) 3 3 3 3 Adding (8.31) and (8.32) we get 4 4 + + + = , (8.33) 3 3 varifying the completeness relation. For four-index tensors the decomposition is performed as in the three-index case, resulting in table 8.5.1. 1 2 3 = 1 1 2 2
DasGroup/book/chapter/unitary.tex 14apr2000 printed April 14, 2000

8.5. REDUCTION OF TENSOR PRODUCTS Ya

1 2 3
1 2 3 1 3 2
1 2 3 1 2 3

75 PYa

d Ya n(n+1)(n+2) 6
n(n2 1) 3 n(n2 1) 3 (n2)(n1)n 6 4 3 4 3

n3

Table 8.4: Reduction of three-index tensor. The bottom row is the direct sum of the Young tableaux, the sum of the dimensions, and the sum of the projection operators (completeness).

8.5.2

Basis vectors

The Young projection operators as constructed above are also projection operators of the symmetric group Sn . If we let Y be a Young tableau labelling an irreducible representation of Sn , the dimension of the representation is dY = n! . |Y| (8.34)

For the two-index tensors we see that application of the projection operators project any group element to the subspace of the projection in question. For the three-index tensors the result is not as simple as that, because the Sn representation is two-dimensional. Instead, when the three-index projection operators are applied from the right, the group elements of Sn are projected to the set 4 3 4 3 4 3 , 4 3 , . (8.35)

of basis vectors. For higher-index tensors there are similar sets of basis vectors. The number of components in each basis vector is the dimension of the projection operator in Sn .

printed April 14, 2000

DasGroup/book/chapter/unitary.tex 14apr2000

76 Ya
1 2 3 4

CHAPTER 8. UNITARY GROUPS d Ya

n(n+1)(n+2)(n+3) 24

PYa

1 2 3 4

(n1)n(n+1)(n+2) 8

3 2

1 2 4 3

(n1)n(n+1)(n+2) 8

3 2

1 3 4 2

(n1)n(n+1)(n+2) 8

3 2

1 2 3 4

n2 (n2 1) 12

4 3

1 3 2 4 1 2 3 4 1 3 2 4 1 4 2 3
1 2 3 4

n2 (n2 1) 12

4 3

(n1)n(n+1)(n+2) 8

3 2

(n2)(n1)n(n+1) 8

3 2

(n2)(n1)n(n+1) 8

3 2

(n2)(n1)n(n+1) 8

1 2 3 4

n4

Table 8.5: Reduction of four-index tensor.

8.6

3-j symbols

The SU (n) three-vertex is written

P X

Y Z

PZ

DasGroup/book/chapter/unitary.tex 14apr2000

...

X Y Z

P Y a

...

(8.36)

printed April 14, 2000

8.6. 3-J SYMBOLS

77

in terms of the Young projection operators PX , PY , and PZ . If b + c = a the vertex vanishes; if a = b + c the vertex might be non-vanishing. The overall normalization is arbitrary, but X Y Z is a natural choice, see (8.25). A 3-j consists of two fully contracted three-vertices. We therefore have
...
X Y Z

...

P X

= X Y Z
...

P Y PZ
...

(8.37)

which we write tr (X Z) Y. As an example, take X= 1 2 ,

3

Y= 1 2 4 ,
3 5 6

and

Z= 4 5.
6

Then

X Y Z

4 4 2 3 3

For economy of notation we omit the arrows on the Kronecker delta lines.

8.6.1

Evaluation by direct expansion

The simplest 3-j s to evaluate are tr ( ) and tr ( ) . Any SU (n) 3-j may be evaluated by direct expansion of the symmetry operators, but the resulting number of terms grows combinatorially with the total number of boxes in the Young diagram Y, making brute force expansion an unattractive method. There is a slightly less brutal expansion method. Expanding one symmetry operator may lead to simplications of the diagram, for instance by using rules such as (5.8), (5.9), (5.18), and (5.19). An example of the application of this method is given in (Elvang). If Y is a Young diagram with a single row or a single column it is easily seen that the 3-j X Y Z is either 0 or dY .

8.6.2

Application of the negative dimension theorem

An SU (n) invariant scalar is a fully contracted object (vacuum bubble) consisting of Kronecker deltas and Levi-Civita symbols. Since there are no external legs,
printed April 14, 2000 DasGroup/book/chapter/unitary.tex 14apr2000

78

CHAPTER 8. UNITARY GROUPS

the Levi-Civitas appear only in pairs, making it possible to combine them into antisymmetrizers. In birdtrack notation, an SU (n) invariant scalar is therefore a vacuum bubble graph built only from symmetrizers and antisymmetrizers. The negative dimensionality theorem for SU (n) states that for any SU (n) invariant scalar exchanging symmetrizers and antisymmetrizers is equivalent to replacing n by n: SU(n) = SU(n) . (8.38)

where the bar on SU indicates transposition, ie. exchange of symmetrizations and antisymmetrizations. The theorem also applies to U (n) invariant scalars, since the only dierence between U (n) and SU (n) is the invariance of the Levi-Civita tensor in SU (n). The proof of this theorem is given in chapter 12. For the dimensions of the Young projection operators we have dYt (n) = dY (n) by the negative dimensionality theorem, where Yt is the transpose of the k -standard arrangement Y; hence it suces to compute the dimension once, either for Y or Yt . Now for k -standard arrangements X, Y, and Z, compare the diagram of Xt t Y Zt to that of X Y Z. The diagrams are related by a reection in a vertical line, reversal of the arrows on the lines, and interchange of symmetrizers and antisymmetrizers. The rst two operations do not change the value of the diagram, hence the value of Xt Yt Zt is the value of X Y Z with n n (and possibly an overall sign). Hence it is sucient to calculate approximately half of all 3-j s. Challenge We have seen that there is a coloring algorithm for the dimensionality of the Young projection operators. Find a coloring algorithm for the 3-j s of SU (n) open question.

8.6.3

A sum rule for 3-j s

Let Y be a k -standard arrangement with k boxes and let be the set of all k standard arrangements and p the set of k -standard arrangements with p boxes. Then
X Y

(k 1)dY .

(8.39)

(X,Z )

First of all, the sum is well-dened, ie. nite, because the 3-j is non-vanishing only if the number of boxes in X and Z add up to k , and this only happens for nitely many tableaux.
DasGroup/book/chapter/unitary.tex 14apr2000 printed April 14, 2000

8.7. CHARACTERS

79

To prove this, recall that the Young projection operators constitute a complete set, X p PX = 1, where 1 is the [p p] unit matrix. Hence,
...
X Y

...

k 1

P X P Y

=
m=1

X,Z

X m Z k m
...

...

PZ

...

...

k 1

k 1

=
m=1
...

P Y
...

=
m=1

dY = (k 1)dY .

(8.40)

This sum rule oers a cross-check on the individual 3-j calculations.

8.7

Characters

Now that we have explicit Young projection operators we should be able to compute any SU (n) invariant scalar. As an example we will consider calculations of characters of SU (n). Given an irreducible representation we have the corresponding Young tableau k -standard arrangement Y, which enable us to calculate the character Y (M ) = tr Y M , where M is a unitary [n n] matrix. i. Then Diagrammatically we shall denote M as Mij = j

Expanding the symmetry operators and collecting terms we nd

k

Y (M ) =
m=0

cm (tr M )m tr M km ,

...

Y (M ) =

P Y

(8.41)

(8.42)

where k is the number of boxes in Y and the cm s are coecients of the expansion.

8.8

Mixed two-index tensors

As the next trivial example consider mixed tensors q (1) q (2) V V . The Kronecker delta invariants are the same as in sect. 8.1, but now they are drawn
printed April 14, 2000 DasGroup/book/chapter/unitary.tex 14apr2000

80

CHAPTER 8. UNITARY GROUPS

dierently (we are looking at a cross channel):

c c b identity: 1 = 1b a,d = a d = bc b c = Ta,d = a d =

, . (8.43)

trace: T

The matrix has a trivial characteristic equation T2 = = nT , (8.44)

with roots 1 = 0, 2 = n; T (T n) = 0 . The corresponding projection operators (??) are P1 = 1 1 T = n n 1 T = n , 1 n (8.46) (8.47) (8.45)

P2 = 1 =

(8.48)

with dimensions d1 = tr P1 = 1, d2 = tr P2 = n2 1. (9.14) is the projection operator for the adjoint representation of SU (n). In this way the invariant matrix T has resolved the space of tensors xa b V V into singlet: P1 x = traceless part: 1 c b x , n c a 1 c b x . n c a (8.49) (8.50)

P2 x = xb a

a invariant, so the generators of the Both projection operators obviously leave b unitary transformations are given by their sum

U (n) :

1 a

(8.51)

a b = n2 . If we extend the list of and the dimension of U (n) is N tr PA = a b primitive invariants from the Kronecker delta to the Kronecker delta and the Levi-Civita tensor (5.28), the singlet subspace does not satisfy the invariance condition (5.61)

birdTrack = 0 . For the traceless subspace (8.48), the invariance condition is birdTrack 1 birdTrack = 0 . n

(8.52)

(8.53)
printed April 14, 2000

DasGroup/book/chapter/unitary.tex 14apr2000

8.9. MIXED DEFINING ADJOINT TENSORS

81

This is the same relation as (5.26). (Expand the antisymmetrization operator using (5.20) so the invariance condition is satised.) The adjoint representation is given by SU (n) : 1 a 1 n 1 a d a d = c b b . n c =

1 d (Ti )a b (Ti )c a

(8.54)

SU (n) is, by denition, the invariance group of the Levi-Civita tensor (hence special) and the Kronecker delta (hence unitary), and its dimension is N = n2 1. The dening representation Dynkin index follows from (??) and (??)
1

= 2n

(8.55)

(This was evaluated in the example of sect. ??). The Dynkin index for the singlet representation (8.49) vanishes identically (as it does for any singlet representation).

8.9

Mixed dening adjoint tensors

In this and the following section we generalize the reduction by invariant matrices to spaces other than the dening representation. Such techniques will be very useful later on, in our construction of the exceptional Lie groups. We consider the dening adjoint tensor space as a projection from V V space: = (8.56)

The following two invariant matrices acting on V 2 V space contract or interchange dening representation indices: R = Q = = (8.57) (8.58)

R projects onto the dening space, and satises characteristic equation R2 = = n2 1 R. n (8.59)

The corresponding projection operators are P1 = P4 =

printed April 14, 2000

n2

n 1 n2

, n 1 . (8.60)
DasGroup/book/chapter/unitary.tex 14apr2000

82

CHAPTER 8. UNITARY GROUPS

Q takes a single eigenvalue on the P1 subspace QR = 1 = R. n (8.61)

Q2 is computed by inserting the adjoint representation projection operator (8.54): Q2 = = 1 n . (8.62)

The projection on the P4 subspace yields the characteristic equation P4 (Q2 1) = 0 , with the associated projection operators P2 = = 1 1 P4 (1 + Q) = 2 2 1 2 + n n2 1 , + (8.64) (8.65) (8.66) P3 = = 1 P4 (1 Q) 2 1 2 (8.63)

1 n+1

1 n1

The dimensions of the two subspaces are computed by taking traces of their projection operators: d2 = tr P2 = birdTrack = = 1 2 birdTrack + birdTrack 1 birdTrack n+1

1 1 1 nN + N N n+1 2 n+1 n+1 (n 1)n(n + 2) = , (8.67) 2 and similarly for d3 . This is tabulated in table 8.5. Mostly for illustration purposes, let us now perform the same calculation by utilizing the algebra of invariants method outlined in sect. 3.3. A possible basis set picked from the V A V A linearly independent tree invariants consists of (e, R, Q) = , , . (8.68)

The multiplication table (3.39) has been worked out in (8.59), (8.61) and (8.62). For example, the (t ) matrix representation for Qt is e e 0 0 1 (8.69) (Q) t = Q R = 0 1/n 0 R T Q Q 1 1/n 0
DasGroup/book/chapter/unitary.tex 14apr2000 printed April 14, 2000

8.10. TWO-INDEX ADJOINT TENSORS

83

and similarly for R. In this way we obtain the [33] matrix representation of the algebra of invariants 0 1 0 0 0 1 1 0 0 1 {e, R, Q} = 0 1 0 , 0 n n 0 , 0 1/n 0 (8.70) 1 1/n 0 0 0 1 0 1/n 0
1 From (8.59) we already know that the eigenvalues of R are {0, 0, n n }. The last n eigenvalue yields the projection operator P1 = n2 1 , but the projection operator P4 yields a 2-dimensional degenerate representation. Q has 3 distinct eigenvalues 1 , 1, 1} and is thus more interesting; the corresponding projection operators { n 1 eigenspace projection operators is again fully decompose the V A space. n P1 , but P4 is split into 2 subspaces, verifying (8.66) and (8.64): 1 1) (Q + 1)(Q + n 1 = 1 2 (1 + 1)(1 + n )

P2 = P3 =

1+Q 1Q

1 R n+1 1 R n1 . (8.71)

1 1) (Q 1)(Q + n 1 1 = 2 (1 1)(1 + n )

We see that the matrix representation of the algebra of invariants is a ne tool for implementing the full reduction, and perhaps easies to implement as computation than out ans out birtracks manipulation. To summarize the invariant matrix R projects out the one-particle subspace P1 . The particle exchange matrix Q splits the remainder into the irreducible particle-adjoint subspaces P2 and P3 .

8.10

Two-index adjoint tensors

Consider the Kronecker product of two adjoint representations. We want to reduce the space of tensors xij A A, with i = 1, 2, . . . dA . The rst decomposition is the obvious decomposition (9.14) into the symmetric and antisymmetric subspaces 1 = S + + A . (8.72)

birdTrack =

As the adjoint representation is real, the symmetric part can be split into the trace and the traceless part, as in (??) S = = 1 T + PS dA 1 + dA

1 dA

(8.73)

printed April 14, 2000

DasGroup/book/chapter/unitary.tex 14apr2000

84

CHAPTER 8. UNITARY GROUPS

8.11

Casimirs for the fully symmetric representations of SU (n)

In this section we carry out a few explicit birdtrack Casimir evaluations.

8.12 8.13

SU (n), U (n) equivalence in adjoint representation Dynkin labels for SU (n) representations

DasGroup/book/chapter/unitary.tex 14apr2000

printed April 14, 2000

Chapter 9

Orthogonal groups
Orthogonal group SO(n) is the group of transformations which leave invariant a symmetric quadratic form (q, q ) = g q q : g = g = birdTrack If (q, q ) is an invariant, so is its complex conjugate (q, q ) = g q q , and g = g = birdTrack (9.2) (9.1)

must be proportional to unity, as is also an invariant tensor. Matrix A g g otherwise its characteristic equation would decompose the dening n-dimensional representation. A convenient normalization is

g g

birdTrack = birdTrack

(9.3)

As the indices can be raised and lowered at will, nothing is gained by keeping arrows. Our convention will be to perform all contractions with metric tensors with upper indices, and omit the arrows and the open dots: g birdTrack . (9.4)

All other tensors will have lower indices. For example, Lie group generators (Ti ) from (??) will be replaced by (Ti ) = birdTrack birdTrack . The invariance condition (??) for the metric tensor birdTrack + birdTrack = 0
(Ti ) g + (Ti ) g = 0

(9.5)

(9.6) 85

86

CHAPTER 9. ORTHOGONAL GROUPS

becomes in this notation a statement that the SO(n) generators are antisymmetric: birdTrack + birdTrack = 0 (Ti ) = (Ti ) (9.7)

Our analysis of the representations of SO(n) will depend only on the existence of a symmetric metric tensor and its invertability, and not on its eigenvalues. The resulting Clebsch-Gordan series applies both to the compact SO(n), and non-compact orthogonal groups such as the Minkowski group SO(1, 3). In this chapter, we outline the construction of SO(n) tensor representations. Spinor representations will be taken up in chapter ??

9.1 9.2 9.3 9.4 9.5

Two-index tensors Three-index tensors Mixed dening adjoint tensors Two-index adjoint tensors Gravity tensors

In a dierent application of the birdtracks, we now change the language, and construct irreducible rank-four gravity curvature tensors. (Birdtracks for Young projection operators had originally been invented by Penrose (1971) in this context.) The Riemann-Christoel curvature tensor has the following symmetries (Weinberg 1972): R = R R = R R + R + R = 0 . (9.8) (9.9) (9.10)

9.6

Dynkin labels of SO(n) representations

In general, one has to distinguish between the odd and the even dimensional orthogonal groups, as well as their spinor and ono-spinor representations. For SO(2r + 1) representations there are r Dynkin labels (a1 a2 . . . ar1 Z ). If Z is odd, the representation is spinor; if Z is even, it is tensor. In this chapter, we study only the tensor representations. For the tensor representations, the corresponding Young tableau (9.14) is given by (a1 a2 . . . ar1 Z ) (a1 a2 . . . ar1
DasGroup/book/chapter/orthog.tex 14apr2000

Z 00 . . .) . 2

(9.11)
printed April 14, 2000

9.6. DYNKIN LABELS OF SO(N ) REPRESENTATIONS For example, for SO(7) representation (102) have (102) (1010 . . .) = birdTrack

87

(9.12)

For orthogonal groups the Levi-Civita tensor can be used to convert a long column of k boxes into a short column of (2r + 1 k ) boxes. The highest column which cannot be shortened by this procedure has r boxes. r is the rank of SO(2r + 1). For SO(2r) representations, the last two Dynkin labels are spinor roots (a1 a2 . . . ar2 Y Z ). Tensor representations have Y + Z =even. However, as spinors are complex, tensor representations can also be complex, conjugate representations being related by (a1 a2 . . . Y Z ) = (a1 a2 . . . ZY ) . For Z Y, Z + Y even, the corresponding Young tableau is given by (a1 a2 . . . ar2 Y Z ) (a1 a2 . . . ar2 , Z Y , 0, 0, . . .) 2 (9.14) (9.13)

The Levi-Civita tensor can be used to convert long columns into short columns. For columns of r boxes, the Levi-Civita tensor splits 0(2r) representations into conjugate pairs of SO(n) representations. Of the various expressions for the dimensions of SO(n) tensor representations, we nd the formula of King (1972) and Murtaza and Rashid (1973) the most convenient.

printed April 14, 2000

DasGroup/book/chapter/orthog.tex 14apr2000

88

CHAPTER 9. ORTHOGONAL GROUPS

DasGroup/book/chapter/orthog.tex 14apr2000

printed April 14, 2000

Chapter 10

Spinors
In chapter 9 we have discussed only the tensor representation of orthogonal groups. However, the spinor representations of SO(n) play a fundamental role in particle physics, both as representations of space-time symmetries (Pauli spin matrices, Dirac gamma matrices, fermions in D-dimensional supergravities), and as representations of internal symmetries (SO(10) grand unied theory, for example). In calculations of radiative corrections, the QED colour weights (ie., spin traces) can easily run up to traces of some twelve gamma matrices (Kinoshita 1981), and ecient evaluation algorithms are of great practical importance. A most straightforward algorithm would evaluate such a trace in some 11!! = 11 9 7 5 3 10 000 steps. Even computers shirk such tedium. A good algorithm will do the job in some 62 100 steps. Spinors came to Cartan (1864) as an unexpected fruit of his labours on the complete classication of representations of the simple Lie groups. Dirac (1928) rediscovered them while looking for a linear version of the relativistic KleinGordon equation. He introduced matrices which were required to satisfy
2 2 (p0 0 + p1 1 + . . .)2 = (p2 0 p1 p2 . . .) .

(10.1)

For n = 4 he constructed gammas as [4 4] complex matrices. For SO(2r) and SO(2r + 1) gamma matrices were constructed explicitly as [2r 2r ] complex matrices by Weyl and Brauer (1935). In the early days, such matrices were taken as a literal truth, and Klein and Nishina (1929) are reputed to have computed their cross-section by multiplying by hand explicit [4 4] matrices. Nevertheless, all information that is actually needed for spin traces evaluation is contained in the Dirac algebraic condition (10.2), ie. the Cliord algebra of matrices { , } = 2g . (10.2)

Iterative application of this condition immediately yields a spin traces evaluation algorithm in which the only residue of gamma matrices is the normalization factor tr 1. However, this simple algorithm is inecient in the sense that it requires a combinatorially large number of evaluation steps. The most ecient algorithm on the market (for arbitrary n) appears to be the one given by Kennedy 89

90

CHAPTER 10. SPINORS

(1982) [Cvitanovi c and Kennedy (1982)]. In Kennedys algorithm, one views the spin trace to be evaluated as a 3n j coecient. Fierz (1934) identities are used to express this 3n j coecient in terms of 6j coecients (cf. sect. 9.14). Gamma matrices are [2n/2 2n/2 ] in even dimensions, [2(n1)/2 2(n1)/2 ] in odd dimensions, and at rst sight it is not obvious that a smooth analytic continuation in dimension should be possible for spin-traces. The reason why the Kennedy algorithm succeeds is that spinors are really not there at all. Their only role is to restrict the SO(n) Clebsch-Gordan series to fully antisymmetric representations. The corresponding 3j and 6j coecients are relatively simple combinatoric numbers, with analytic continuations in terms of gamma functions. The case of four spacetime dimensions is special because of the reducibility of SO(4) to SU (2) SU (2). Farrar and Neri (1983), who have (as of 18th April 1983) computed in excess of 58 149 Feynman diagrams, have used this structure to develop a very ecient method for evaluating SO(4) spinor expressions. An older technique is the Kahane (1968) algorithm, which implements diagrammatically the Chisholm (1963) identities. (REDUCE (Hearn 1970) uses the Kahane algorithm). Thrnblad (1967) has used SO(4) SO(5) embedding to speed-up evaluation of traces for massive fermions.

10.1

Spinograpy

Kennedy (1982) introduced diagrammatic notation for gamma matrices

10.2 10.3 10.4 10.5 10.6 10.7 10.8

Fierzing around Fierz coecients 6j coecients Exemplary evaluations Invariance of -matrices Handedness Kahane algorithm

For the case of four dimensions, there is a fast algorithm for trace evaluation, due to Kahane (1968).

DasGroup/book/chapter/spinor.tex 14apr2000

printed April 14, 2000

Chapter 11

Symplectic groups
Symplectic group SP (n) is the group of all transformations which leave invariant a skew symmetric quadratic form (p, q ) = gab pa q b : gab = gba
a b

a, b = 1, 2, . . . n n even (11.1)

The birdtrack notation is motivated by the need to distinguish the rst and the second index: it is a special case of the birdtracks for antisymmetric tensors of even rank (??). If (p, q ) is an invariant, so is its complex conjugate (p, q ) = g ba pa qb , and g ab = g ba
a b cb is also an invariant tensor. Matrix Ab a = gac g must be proportional to unity, as otherwise its characteristic equation would decompose the dening n-dimensional representation. A convenient normalization is b gac g cb = a

(11.2)

(11.3) = = (11.4)

Indices can be raised and lowered at will, so the arrows on lines can be dropped. However, omitting symplectic warts (the black half-circles) appears perilous, as without them it is hard to keep track of signs. Our convention will be to perform all contractions with g ab , and omit the arrows but not the warts: g ab =a
b

(11.5) 91

92

CHAPTER 11. SYMPLECTIC GROUPS

b

All other tensors will have lower indices. The Lie group generators (Ti )a be replaced by (Ti )ab = .

will

(11.6)

The invariance condition (??) for the symplectic invariant tensor is + = 0 (11.7)

(Ti )a c gcb + (Ti )b c gac = 0 .

A skew symmetric matrix gab has the inverse in (8.3) only if Det|g | = 0. That is possible only in even dimensions, so Sp(n) can be realized only for even n. In this chapter we shall outline the construction of Sp(n) tensor representations. They are obtained by contracting the irreducible tensors of SU (n) with the symplectic metric g ab and decomposing them into traces and traceless parts. The representation theory for Sp(n) is analogous in step-by-step fashion to the representation theory for SO(n). In chapter ??????? we shall show that this arises because the two groups are related by supersymmetry, and in chapter ??????? we shall exploit this connection by showing that all group-theoretic weights for the two groups are related by analytic continuation into negative dimensions.

11.1

Two-index tensors

The decomposition goes the same way as for SO(n), sect. 9.14. The matrix (??) is given by: T = (11.8)

and satises the same characteristic equation (11.11) as before. Now T is antisymmetric, AT = T , and only the antisymmetric subspace gets decomposed. The nal decomposition of Sp(n) two-index tensors is singlet: traceless antisymmetric: (P1 )ab,cd (P2 )ab,cd = = = symmetric: (P3 )ab,cd = =
DasGroup/book/chapter/symplect.tex 28/10-97 printed April 14, 2000

1 n gab gcd 1 2 (gad gbc

1 n

1 gac gbd ) n gab gcd

1 2 (gad gbc

1 n

(11.9)

+ gac gbd )

11.2. MIXED DEFINING ADJOINT TENSORS

93

The SU (n) adjoint representation (??) is now split into traceless symmetric and antisymmetric parts. The adjoint representation of Sp(n) is given by the symmetric subspace, as only P3 satises the invariance condition (11.7): + birdTrack = 0 Hence the projection operator for Sp(n) is given by 1 a = birdTrack (11.11) (11.10)

The dimension of Sp(n) is N = tr PA = birdTrack = n(n + 1) 2 (11.12)

Remember that all contractions are carried out by g ab - hence the extra warts in the trace expression. Dimensions of other representation and the Dynkin indices are listed in ???? .

11.2 11.3

Mixed dening adjoint tensors Dynkin labels of Sp(n) representations

printed April 14, 2000

DasGroup/book/chapter/symplect.tex 28/10-97

94

CHAPTER 11. SYMPLECTIC GROUPS

DasGroup/book/chapter/symplect.tex 28/10-97

printed April 14, 2000

Chapter 12

Negative dimensions
A cursory examination of the expressions for the dimensions and the Dynkin indices listed in the tables of chapter ??, chapter ?? and chapter ?? reveal intriguing symmetries under substitution n n. This kind of symmetry is best illustrated by the representations of SU (n); if stands for a Young tableau with p boxes, and for the transposed tableau obtained by ipping across the diagonal, (ie., exchanging symmetrizations and antisymmetrizations), then the dimensions of the two tableaux are related by SU (n) : d (n) = (1)p d (n) . (12.1)

This is evident from the standard recipe (??) for computing the SU (n) representation dimensions, as well as from the expressions listed in the tables of chapter ??. In all cases, exchanging symmetrizations and antisymmetrizations amounts to replacing n by n. Such relations have been noticed before, Parisi and Soulas (1979), have suggested that a Grassmann vector space of dimension n can be interpreted as an ordinary vector space of dimension n. Penrose (1971) has introduced the term negative dimensions in his construction of SU (2) Sp(2) representations as SO(2). King (1972) has proved that the dimension of any irreducible representation of Sp(n) is equal to that of SO(n) with symmetrizations exchanged with antisymmetrizations (ie. corresponding to the transposed Young tableau), and n replaced by n. Mkrtchyan (1981) has observed this relation for the QCD loop equations. With the advent of supersymmetries, n n relations have become commonplace, as they are built into the structure of groups such as the orthosympletic group OSp(b, f ). Some highly nontrivial examples of n n symmetries for the exceptional groups (Cvitanovi c 1981) will be discussed elsewhere in these notes. Here we shall prove the following: Theorem 1. For any SU (n) invariant scalar exchanging symmetrizations and antisymmetrizations is equivalent to replacing n by n: SU (n) = SU (n) . 95 (12.2)

96

CHAPTER 12. NEGATIVE DIMENSIONS

Theorem 2. For any SO(n) invariant scalar there exists the corresponding Sp(n) invariant scalar (and vice versa) obtained by exchanging symmetrizations and antisymmetrizations, replacing the SO(n) symmetric bilinear invariant dab by the Sp(n) antisymmetric bilinear invariant fab , and replacing n by n: SO(n) = Sp(n) , (12.3) Sp(n) = SO(n) , (12.4)

(the bars on SU , Sp, SO indicate transposition, ie. exchange of symmetrizations and antisymmetrizations). All previous n n relations are special cases of these general theorems, the general proof is much simpler than the published proofs for the special cases. As we have argued in sect. 9.14, all physical consequences of a symmetry (representation dimensions, level splittings, etc) can be expressed in terms of invariant scalars (3n j coecients, casimirs, etc), so it suces to prove the equivalence (12.2), (12.4) for arbitrary scalar invariants. The idea of the proof is illustrated by the following typical computation: evaluate, for example, the following SU (n) 9j coecient for recoupling of three antisymmetric rank-tworepresentations:

= n3 n 2 n 2 + n n 2 + n + n n 2 = n(n 1)(n 3) . (12.5)

Notice that in the expansion of the symmetry operators the graphs with an odd number of crossings give an even power of n, and vice versa. If we change the three symmetrizers into antisymmetrizers, the terms which change the sign are exactly those with an even number of crossings. The crossing in the original graph, which had nothing to do with any symmetry operator, appears in every term of the expansion, and this does not aect our conclusion; an exchange of symmetrizations and antisymmetrizations amounts to substitution n n.
DasGroup/book/chapter/negdim.tex 14apr2000 printed April 14, 2000

12.1. SU (N ) = SU (N )

97

(The overall sign is only a matter of convention; it depends on how we dene vertices in 3n j s.) The proof for the general SU (n) case is even simpler than the above example:

12.1

SU (n) = SU (n)

a and the The primitive invariant tensors of SU (n) are the Kronecker tensor b Levi-Civita tensor a1 an . All other invariants of SU (n) are built from these two objects. A scalar (3n j coecient, vacuum bubble) is a number which, in birdtrack notation, corresponds to a graph with no external legs. As the directed lines must end somewhere, the Levi-Civita tensors can be present only in pairs, and can always be eliminated by the identity (5.32). An SU (n) 3n j coecient, therefore, corresponds to a diagram made solely of closed loops of directed lines and symmetry projection operators, like the example (12.5). Consider the graph corresponding to an arbitrary SU (n) scalar and expand all its symmetry operators as in (12.5). The expansion can be arranged (in any of many possible ways) as a sum of pairs of form

... +

+ ...

(12.6)

with a plus sign if the crossing arises from a symmetrization, and a minus sign if it arises from an antisymmetrization. Each graph consists only of closed loops, ie. a denite power of n, and thus uncrossing two lines can have one of two consequences. If the two crossed line segments come from the same loop, then uncrossing splits this into two loops, whereas if they come from two loops, it joins them into one loop. The power of n is changed by the uncrossing: =n (12.7)

Hence the pairs in the expansion (12.6) always dier by n1 , and exchanging symmetrizations and antisymmetrizations has the same eect as substituting n n (up to an irrelevant overall sign). This completes the proof of (12.2). Some examples of n n relations for SU (n) representations: (i) Dimensions of the fully symmetric representations (5.14) and the fully antisymmetric representations (5.22) are related by the gamma-function analytic continuation formula (n + p 1)! n! = (1)p . (n p)! (n 1)!
printed April 14, 2000

(12.8)

DasGroup/book/chapter/negdim.tex 14apr2000

98

CHAPTER 12. NEGATIVE DIMENSIONS

(ii) The representations (??), (??) correspond to the 2??? symmetric, antisymmetric, respectively. Therefore, their dimensions in table 6.1 are related by n n. (iii) The representations (??) and (??) (see table ?? are related by n n for the same reason. (iv) n n symmetries in table ??. (v) Dimension formula (12.1).

12.2

SO(n) = Sp(n)

a and In addition to b ab...d , SO (n) preserves a symmetric bilinear invariant dab for which we have introduced birdtrack open circle notation in (7.1). Such open circles can occur in SO(n) 3n j graphs, ipping the line directions. The LeviCivita tensor still cannot occur, as directed lines starting on a tensor would have to end on a d tensor, which gives zero by symmetry. Sp(n) diers from SO(n) by having a skew symmetric bilinear tensor fab , for which we have introduced birdtrack wart notation in (??). A Levi-Civita tensor can appear in an Sp(n) 3n j graph, but as

det f ,

(12.9)

(an exercise for the reader), a Levi-Civita can always be replaced by an antisymmetrization = (det f ) 2
1

(12.10)

For any SO(n) scalar there exists a correponding Sp(n) scalar, obtained by exchanging the symmetrizations and antisymmetrizations and the dab s and fab s in the corresponding graphs. The proof that the two scalars are transformed into each other by replacing n by n is the same as for SU (n), except that the two line segments at a crossing could come from a new kind of loop containing dab s or fab s. In that case, equation (12.7) is replaced by =

=+

(12.11)

While now uncrossing the lines does not change the number of loops, changing dab s to fab s does provide the necessary minus sign. This completes the proof of
DasGroup/book/chapter/negdim.tex 14apr2000 printed April 14, 2000

12.2. SO(N ) = SP (N )

99

(12.4) for the tensor representations of SO(n) and Sp(n). To extend the proof to the spinor representations, we will rst have to invent the Sp(n) analog of spinor representations. We postpone this until the next chapter. Some examples of SO(n) = Sp(n) relations: (i) The SO(n) antisymmetric adjoint representation (??) corresponds to the Sp(n) symmetric adjoint representation (??) (ii) Compare table ?? and table ??. (iii) Penrose (1971) binors: SU (2) = SO(2).

printed April 14, 2000

DasGroup/book/chapter/negdim.tex 14apr2000

100

CHAPTER 12. NEGATIVE DIMENSIONS

DasGroup/book/chapter/negdim.tex 14apr2000

printed April 14, 2000

Chapter 13

Spinsters
This chapter is based on P. Cvitanovi c and A.D. Kennedy, Phys. Scripta 26, 5 (1982).

101

102

CHAPTER 13. SPINSTERS

DasGroup/book/chapter/negdim.tex 14apr2000

printed April 14, 2000

Chapter 14

SU (n) family of invariance groups

SU (n) preserves the Levi-Civita tensor, in addition to the Kronecker sect. ??. This additional invariant induces non-trivial decompositions of U (n) representations. In this chapter we show how the theory of SU (2) representations (the quantum mechanics textbooks theory of angular momentum) is developed by birdtracking; that SU (3) is the unique group with the Kronecker delta and a rank-three antisymmetric primitive invariant; that SU (4) is isomorphic to SO(6); and that for n 4, only SU (n) has the Kronecker and rank-n antisymmetric tensor primitive invariants.

14.1

Representations of SU (2)

For SU (2) we can construct an additional invariant matrix which would appear to induce a decomposition of n n representations
a c Eb, d =

1 ac bd 2 (14.1)
b d c a

(14.2)

However, by (5.32) this can be written as a sum over Kronecker deltas, and is not an independent invariant. So what does ac do? It does two things; it removes the distinction between quark and antiquark; (if qa transforms as a quark, then ab qb transforms as an antiquark), and it reduces the representations of SU (2) to the fully symmetric ones. Consider n n decomposition (??) 1 2 = birdTrack + (14.3) = + 103

104

CHAPTER 14. SU (N ) FAMILY OF INVARIANCE GROUPS (14.4) 22 = 23 21 + . 2 2 (14.5)

The antisymmetric representation is a singlet = (14.6)

Now consider the V 3 and V 4 space decompositions, obtained by adding successive quarks one at a time: = + (14.7) = + 3 4 + (14.8) 1 2 3 = 1 2 3 + 1 + 3 (14.9) = + 4 3 4 3 + (14.10) + 3 2 + + (14.11) 1 2 3 4 = 1 2 3 4 + 1 4 + 3 4 (14.12) + 1 2 +++ (14.13)

This is clearly leading us into the theory of SO(3) angular momentum addition, described in any quantum mechanics textbook. We shall anyway persist a little while longer, just to illustrate how birdtracks can be used to recover some familiar results. The projector for m-quark representation is
1

Pm =

2 m

(14.14)

...(2+m1) = m + 1 (usually m = 2j , where The dimension is trPm = 2(2+1)(2+2) m! j is the spin of the representation). The projection operator (??) for the adjoint representation (spin 1) is

1 2

(14.15)
printed April 14, 2000

DasGroup/book/chapter/suen.tex 14apr2000

14.2. SU (3) AS INVARIANCE GROUP OF A CUBIC INVARIANT (This can be rewritten as using (14.6)).

105

The quadratic casimir for the dening representation is birdTrack = Using = 1 2 = 1 2 , (14.17) 3 2 . (14.16)

we can compute the quadratic casimir for any representation

14.2

SU (3) as invariance group of a cubic invariant

We have proven that the only group that satises the conditions (i) - (iii) at the beginning of this section is SU (3). Of course, it is well-known that the colour group of physical hadrons is SU (3), and this result might appear rather trivial. That it is not so will become clear from the further examples of invariance groups, such as the G2 family of the next chapter.

14.3

Levi-Civita tensors and SU (n)

In chapter ????? we have shown that the invariance group for a skew-symmetric invariant f ab is Sp(n). In particular, for f ab = ab , the Levi-Civita tensor, the invariance group is SU (2) = Sp(2). In the preceding section, we have proven that the invariance group of a skew-symmetric invariant f abc is SU (3), and that f abc must be proportional to the Levi-Civita tensor. Now we shall show that for f abc...d with r indices, the invariance group is SU (r), and f is always proportional to the Levi-Civita tensor. r = 2 and r = 3 cases had to be treated separately because it was possible to construct from f ab and f abc tree invariants on the q q q q space which could reduce the colour group SU (n) to a subgroup. For f ab , n 4 this is indeed what happens: SU (n) Sp(n), for n even.

14.4

SU (4) - SO(6) isomorphism

printed April 14, 2000

DasGroup/book/chapter/suen.tex 14apr2000

106

CHAPTER 14. SU (N ) FAMILY OF INVARIANCE GROUPS

DasGroup/book/chapter/suen.tex 14apr2000

printed April 14, 2000

Chapter 15

G2 family of invariance groups

1

In this chapter we begin the construction of all invariance groups which posses a symmetric quadratic and an antisymmetric cubic invariant in the dening representation. The resulting classication is summarized in g. 15.1. I nd that the cubic invariant must satisfy either the Jacobi relation (15.7), or the alternativity relation (15.11). In the former case the invariance group can be any semi-simple Lie group in its adjoint representation; we pursue this possibility in the next chapter. The latter case is developed in this chapter; we nd that the invariance group is either SO(3) or the exceptional Lie group G2 . The problem of evaluation of 3n j coecients for G2 is solved completely by the reduction identity (15.14). As a byproduct of the construction I give a proof of the Hurwitzs theorem, sect. 15.5. I also demonstrate that the independent casimirs for G2 are of order 2 and 6, by explicitely reducing the 4-th order casimir in sect. 15.4. Consider the following list of primitive invariants:
a , so the invariance group is a subgroup of SU (n). (i) b

(ii) symmetric g ab = g ba , gab = gba , so the invariance group is a subgroup of SO(n). We take this invariant in its diagonal, Kronecker delta form ab . (iii) a cubic antisymmetric invariant fabc . Primitivness assumption requires that all other invariants can be expressed in terms of the tree contractions of ab , fabc . In the diagrammatic notation one keeps track of the antisymmetry of the cubic invariant by reading the indices o the vertex in a xed order: fabc = b
1

a c

= b

a c

= facb .

(15.1)

superceeds chapter 19 of ref. [13]

107

108

CHAPTER 15. G2 FAMILY OF INVARIANCE GROUPS

primitives:

two relations assume:

one relation

= SO(3)

=A

+B

no relations

n=7
Jacobi alternativity

n=6

=0
any adjoint representation

= G8

SU (3)

quartic primitive

no quartic primitive

SU (n), SO (n), Sp(n)

E8

family

Figure 15.1: Logical organization of chapter 15 and chapter 16. The invariance groups SO(3) and G2 are derived in this chapter, while the E8 family is derived in chapter 16.

The primitiveness assumption implies that the double contraction of a pair of f s is proportional to the Kronecker delta. We can use this relation to x the overall normalization of f s: fabc fcbd = ad = (15.2)

For convenience we shall often set = 1 in what follows. The next step in our construction is to identify all invariant matrices on V 2 , and construct the Clebsch-Gordan series for decomposition of two-index tensors. There are six such invariants, the three distinct permutations of indices of ab cd , and the three distinct permutations of free indices of fabe fecd . For reasons of clarity, we shall break up the discussion into two steps. In the rst step, sect. 15.1, we assume that a linear relation between these six invariants exists. Pure symmetry considerations, together with the invariance condition, completely x the algebra of invariants and restrict the dimension of the dening space to either 3 or 7. In the second step, sect. 15.3, we show that a relation assumed in the rst step must exist because of the invariance condition.
Remark 15.1 Quarks and hadrons. An example of a theory with above invariants would be QCD with the hadronic spectrum consisting of following singlets:
DasGroup/book/chapter/g2family.tex 14apr2000 printed April 14, 2000

15.1. JACOBI RELATION

(i) quark-antiquark mesons (ii) mesons built of two quarks (or antiquarks) in a symmetric color combination (iii) baryons built of three quarks (or antiquarks) in a fully antisymmetric color combination (iv) no exotics, ie. no hadrons built from other combinations of quarks and antiquarks As we shall now demonstrate, for this hadronic spectrum the color group is either SO(3), with quarks of three colors, or the exceptional Lie group G2 , with quarks of seven colors.

109

15.1

Jacobi relation

If the six invariant tensors mentioned above are not independent, they satisfy a relation of form 0=A +B +C +D +E +F (15.3)

Antisymmetrizing a pair of indices yields 0=A +E +F , (15.4)

and antisymmetrizing any three indices yields 0 = (E + F ) . (15.5)

If the tensor itself vanishes, f s satisfy the Jacobi relation (??) 0= + . (15.6)

If A = 0 in (15.4), the Jacobi relation relates the second and the third term 0= +E . (15.7)

The normalization condition (15.2) xes E = 1. = . (15.8)

Contracting with ab we obtain 1 = (n 1)/2, so n = 3. We conclude that if pair contraction of f s is expressible in terms of s, the invariance group is SO(3) and fabc is proportional to the 3-index Levi-Civita tensor. To spell it out; in 3
printed April 14, 2000 DasGroup/book/chapter/g2family.tex 14apr2000

110

CHAPTER 15. G2 FAMILY OF INVARIANCE GROUPS

dimensions an antisymmetric rank-3 tensor can take only one value, fabc = f123 which can be set equal to 1 by appropriate normalization convention (15.2). If A = 0 in (15.4), the Jacobi relation is the only relation we have, and the adjoint representation of any simple Lie group is a possible solution. I return to this case in chapter 16.

15.2

Alternativity and reduction of f -contractions

If the Jacobi relation does not hold we must have E = F in (15.5), and (15.4) takes form + = A . (15.9)

Contracting with ab xes A = 3/(n 1). Symmetrizing the top two lines and rotating the diagrams by 900 we obtain the alternativity relation: = 1 n1 (15.10)

The name comes from the octonian interpretation given in sect. 15.4. Adding the two equations we obtain + = 1 n1 2 + V 2 follows: (15.11)

The Clebsch-Gordon decomposition of = 1 n + n2 = 1 + + + 1 n

n(n 3) (n 1)(n + 2) +n+ 2 2

(15.12)

is reducible on the antisymmetric subspace. By (15.10) By (15.9) the invariant it is also reducible on the symmetric subspace. The only independent f f invariant is which, by the normalization (15.2), is already the projection operator which projects the antisymmetric two-index tensors onto the n-dimensional dening space. The dimensions of the representations are obtained by tracing the corresponding projection operators. of SO(n) is now split into two represenThe adjoint representation tations. Which one is the new adjoint representation? That we determine by to be the proconsidering (5.61), the invariance condition for fabc . If we take jection operator for the adjoint representation, we again get the Jacobi condition
DasGroup/book/chapter/g2family.tex 14apr2000 printed April 14, 2000

15.2. ALTERNATIVITY AND REDUCTION OF F -CONTRACTIONS

111

with SO(3) as the only solution. However, if we assume that the last term in (15.12) is the adjoint projection operator 1 a = 1 , (15.13)

the invariance condition becomes a non-trivial condition:

0=

(15.14)

The last term can be simplied by (15.9) and (5.20) 3 = 2 =3 +2 3 n1 .

Substituting back into (15.14) yields 2 n1

Expanding the last term and redrawing the equation slightly we have 2 n1 2 3 1 3

This equation is antisymmetric under interchange of the left and the right index pairs. Hence 2/(n 1) = 1/3, and the invariance condition is satised only for n = 7. Furthermore, the above relation gives us the G2 reduction identity = 2 + . (15.15) 3 This identity is the key result of this chapter: it enables us to recursively reduce all contractions of products of -functions and pairwise contractions fabc fcde , and thus completely solves the problem of evaluating any casimir or 3n j coecient of G2 . The invariance condition (15.14) for fabc implies that =
printed April 14, 2000

1 2

(15.16)
DasGroup/book/chapter/g2family.tex 14apr2000

112

CHAPTER 15. G2 FAMILY OF INVARIANCE GROUPS

The triangle graph for the dening representation can be computed in two ways, either by contracting (15.10) with fabc , or by contracting the invariance condition (15.14) with ab : = 4n n1 = 5n 4 (15.17)

So the alternativity and the invariance conditions are consistent i (n 3)(n 7) = 0, ie. only for 3 or 7 dimensions. In the latter case, the invariance group is the exceptional Lie group G2 , and the above derivation is also a proof of Hurwitzs theorem, see sect. 15.4. In this way symmetry considerations together with the invariance conditions, suce to determine the algebra satised by the cubic invariant. The invariance condition xes the dening dimension to n = 3 or 7. Having assumed only that a cubic antisymmetric invariant exists, we nd that if the cubic invariant is not a structure constant, it can be realized only in 7 dimensions, and its algebra is completely determined. The identity (15.14) plays the role analogous to that the Dirac relation { , } = 2g I plays for evaluation of traces of products of Dirac gamma-matrices. 2 Just as the Dirac relation obviates need for explicit representations of s, (15.14) reduces any f f f contraction to a sum of terms linear in f , and obviates any need for explicit construction of f s. The above results now enable us to compute any group-theoretic weight for G2 in two steps. First we replace all adjoint representation lines by the projection operators PA (15.13). The resulting expression contains Kronecker deltas and chains of contractions of fabc , which can then be reduced by systematic application of the reduction identity (15.15).

15.3

Primitivity implies alternativity

The only detail which remains to be proven is the assertion that the alternativity relation (15.10) follows from the primitiveness assumption. I complete the proof in this section. The proof is rather inellegant and can probably be easily streamlined. If no relation (15.3) between the three f f contraction is assumed, then by the primitiveness assumption the adjoint representation projection operator PA is of the form =A
2

+B

+C

(15.18)

The Klein-Nishina formula of quantum electrodynamics was computed by Klein and Nishina by explicitely multiplying [44] matrix representations of s and then summing over s. Day after day they would multiply away whole morning, and then meet in the Niels Bohr Institutes cafeteria to compare their results. Today the Klein-Nishina trace over Dirac s is a texbook exercise, reducible by several applications of { , } = 2g I.
DasGroup/book/chapter/g2family.tex 14apr2000 printed April 14, 2000

15.3. PRIMITIVITY IMPLIES ALTERNATIVITY

113

Assume that the Jacobi relation does not hold; otherwise this immediately reduces to SO(3). The generators must be antisymmetric, as the group is a subgroup of SO(n). Substitute the adjoint projection operator into the invariance condition (5.61) (or (15.14)) for fabc : 0= +B +C (15.19)

Resymmetrize this equation by contracting with

. This is evaluated expand-

ing with (5.20) and using a relation due to the antisymmetry of fabc : =0 The result is: 0= + C B 2 +B (15.21) (15.20)

Multiplying (15.19) by B , (15.21) by C and subtracting, we obtain C + B 0 = (B + C ) . 2 I return to the case B + C = 0 below, in (15.26). If B + C = 0, by contracting with fabc we get B C/2 = 1, and 0=

(15.22)

(15.23)

To prove that this is equivalent to the alternativity relation, we contract with , expand the 3-leg antisymmetrization, and obtain 0= 2 +2 (15.24)

The triangle subdiagram can be computed by adding (15.19) and (15.21) 1 + 0 = (B + C ) 2

printed April 14, 2000 DasGroup/book/chapter/g2family.tex 14apr2000

114

CHAPTER 15. G2 FAMILY OF INVARIANCE GROUPS . The result is

and contracting with 1 2

(15.25)

Substituting into (15.24) we recover the alternativity relation (15.10). Hence we have proven that the primitivity assumption implies the alternativity relation for the case B + C = 0 in (15.22). If B + C = 0, (15.19) becomes +B +C 0= (15.26) Using the normalization (??) and orthonormality = 1 a N = = 6n 9n 6 15 n 1 a = + 2(9 n) 15 n conditions we obtain (15.27) (15.28) (15.29)

4n(n 3) 15 n

The remaining antisymmetric representation = = has dimension d= = n(n 3)(7 n) . 2(15 n) (15.31) 9n 15 n 1 a + 3n 9n (15.30)

The dimension cannot be negative, so d 7. For n = 7 the projection operator (15.30) vanishes identically, and we recover the alternativity relation (15.10). The Diophantine condition (15.31) has two further solutions: n = 5 and n = 6. The n = 5 is eliminated by examining the decomposition of the traceless . By the primisymmetric subspace in (15.12) induced by the invariant Q = 2 tiveness assumption Q is reducible on the symmetric subspace 0 = +A +B 1 n

0 = (Q2 + AQ + B )P2 .
DasGroup/book/chapter/g2family.tex 14apr2000 printed April 14, 2000

15.4. CASIMIRS FOR G2 Contracting the top two indices with ab and (Ti )ab we obtain Q2 13n 5 6n Q I P2 = 0 29n 2 (2 + n)(9 n)

115

(15.32)

For n = 5 the roots of this equation are rational and the dimensions of the two representations induced by decomposition with respect to Q are not integers. Hence n = 5 is not a solution. We turn to the case n = 6 in appendix ??.

15.4

Casimirs for G2

In this section we prove that the independent casimirs for G2 are of order 2 and 6, as indicated in the table ??. As G2 is a subgroup of SO(7), its generators are antisymmetric and only even order casmirs are nonvanishing. The quartic casimir (in the notation of (??)) = tr X 4 =
ijkl

xi xj xk xl tr (Ti Tj Tk Tl )

can be reduced by manipulating it with the invariance condition (5.61) = 2 =2 +2

The last term vanishes by further manipulation with the invariance condition = =0 (15.33)

The remaining term is reduced by the alternativity relation (15.10) = = 1 6

This yields the explicit expression for the reduction of quartic casimirs in the dening representation of G2 = tr X 4 = 1 3 1 tr X 2 4
2

(15.34)

As the dening representation is 7-dimensional, the characteristic equation (??) reduces the 8th and all higher order casimirs. Hence the independent casimirs for G2 are of order 2 and 6.
printed April 14, 2000 DasGroup/book/chapter/g2family.tex 14apr2000

116

CHAPTER 15. G2 FAMILY OF INVARIANCE GROUPS

15.5

Hurwitzs theorem

Denition (Curtis 1963): A normed algebra A is an n + 1 dimensional vector space over a eld F with a product xy such that (i) (ii) x(cy ) = (cx)y = c(xy ) , x(y + z ) = xy + xz , (x + y )z = xz + yz, and a non-degenerate quadratic norm which permits composition (iii) N (xy ) = N (x)N (y ) , N (x) F. (15.35) cF x, y, z A

Here F will be the eld of real numbers. Let {e0 , e1 , . . . , en } be a basis of A over F: x = x0 e0 + x1 e1 + . . . + xn en , xa F , ea A . (15.36)

It is always possible to choose eo = I (see Curtis 1963). The product of remaining bases must close the algebra ea eb = dab I + fabc ec , The norm in this basis is N (x) = x2 0 + dab xa xb . From the symmetry of the associated inner product (Tits 1966) (x, y ) = (y, x) = N (x + y ) N (x) N (y ) , 2 (15.39) (15.38) dab , fabc F a, . . . , c = 1, 2, . . . , n (15.37)

it follows that dab = (ea , eb ) = (eb , ea ) is symmetric, and it is always possible to choose bases ea such that ea eb = ab + fabc ec . Furthermore, from (xy, x) = N (y + 1) N (y ) 1 N (xy + x) N (x)N (y ) = N (x) 2 2 = N (x)(y, 1), (15.41) (15.40)

it follows that fabc = (ea , eb , ec ) is fully antisymmetric. [In Tits (1966) notation, the multiplication tensor fabc is replaced by a cubic antisymmetric form (a, a , a ), his equation (14)]. The composition requirement (15.35) expressed in terms of bases (15.36) is 0 = N (xy ) N (x)N (y ) = xa xb yc yd (ac bd ab cd + face fcbd ) .
DasGroup/book/chapter/g2family.tex 14apr2000

(15.42)
printed April 14, 2000

15.5. HURWITZS THEOREM

117

To make a contact with sect. 15.2 we introduce diagrammatic notation (factor i 6/ adjusts the normalization to (15.2)) fabc = i 6 . (15.43)

Diagrammatically, (15.42) is given by 0= + 6 . (15.44)

This is precisely the relation (15.10) which we have proven to be nontrivially realizable only in 3 and 7 dimensions. The trivial realizations are n = 0 and n = 1, fabc = 0. So we have inadvertently proven Hurwitzs theorem (Curtis (1963): n + 1 dimensional normed algebras over reals exist only for n = 0, 1, 3, 7 (real, complex, quaternion, octonion). I call (15.10) the alternativity relation because it can also be obtained by substituting (15.40) into the alternativity condition for octonions (Schafer 1966) [xyz ] (xy )z x(yz ) , [xyz ] = [zxy ] = [yzx] = [yxz ] . (15.45)

Cartan (1894, 1952) was rst to note that G2 (7) is the isomorphism group of octonions, ie. the group of transformations of octonion bases (written here in the innitesimal form) ea = (ab + iDab )eb which preserve the octonionic multiplication rule (15.40). The reduction identity (15.15) was rst derived by Behrends et al. (1966) [in very dierent notation, their equation (16)]. Tits also constructed the adjoint representation projection operator for G2 (7) by dening the derivation on an octonion algebra as 3 1 Dz =< x, y > z = ((x y ) z ) + [(y, z )x (x, z )y ], 2 2 [Tits 1966, equation (23)] where ea eb fabc ec , (ea , eb ) ab . Substituting x = xa ea , we nd (Dz )d = 3xa yb 1 1 ab bd + fabe fecd zc 2 6 (15.48) (15.46) (15.47)

The term in the brackets is just the G2 (7) adjoint representation projection operator PA in (15.13) with normalization = 3.
printed April 14, 2000 DasGroup/book/chapter/g2family.tex 14apr2000

118

CHAPTER 15. G2 FAMILY OF INVARIANCE GROUPS

15.6

Representations of G2

G2 is characterized by the fully antisymmetric cubic primitive invariant fabc . Contracting with fabc , we are able to reduce any column higher than two boxes. Hence, representations of G2 are specied by Young tableaux (??) of form (qp00...). Patera and Sanko (1973) have chosen to label the simple roots in such a way that the correspondence is

DasGroup/book/chapter/g2family.tex 14apr2000

printed April 14, 2000

Chapter 16

E8 family of invariance groups

In this chapter we continue the construction of invariance groups characterized by a symmetric quadratic and an antisymmetric cubic primitive invariant. In the preceeding chapter we proved that the cubic invariant must either satisfy the alternativity relation (15.11), or the Jacobi relation (15.7). The rst case has SO(3) and G2 as the only solutions. Here we pursue the second possibility, restricted to the case of no quartic primitive invariant (see g. 15.1). The main results will be the Diophantine conditions (16.10)-(16.15) and the projection operators for E8 family, given in table 16.1. 1 As, by assumption, the dening representation satises the Jacobi relation (15.7), it is also the adjoint representation of some Lie group. Hence, in this chapter, we denote the dimension of the dening representation by N , the cubic invariant by the Lie algebra structure constants iCijk , and draw the invariants with the thin (adjoint) lines, as in (3.84). The assumption that the dening representation is irreducible means in this case that the Lie group is simple, and the quadratic casimir (Cartan-Killing tensor) is proportional to the identity = CA . (16.1)

In this chapter we shall usually chose normalization CA = 1. The Jacobi relation (??) reduces a loop with three structure constants 1 2

(16.2)

In order to reduce loops with four structure constants we turn to the reduction of the A A space.
1

The inspiration for construction of sect. 15.1 came from Okubo (1979).

119

120

CHAPTER 16. E8 FAMILY OF INVARIANCE GROUPS

16.1

Two-index tensors
A tensors. As in sect. ?? they immediately 1 CA 1 N (16.3)

Consider the decomposition of A decompose into four subspaces: = 1 CA + 1 N + +

|??|

1 = P|??| + P |??| + P + Ps . Consider A Qij,kl AA 1 = CA

i j k

A invariant matrix
l

(16.4)

By the Jacobi relation (??) or (15.6), Q has zero eigenvalue on the antisymmetric subspace QPa = Pa = 1 2 1 Pa = PA Pa = 0 2

so Q can decompose only the symmetric subspace. The assumption that there is no primitive quartic invariant is the dening relation for the E8 family. By this assumption Q2 is not linearly independent of Qij,k , fijm fmk and ij s. On the traceless symmetric subspace this implies that Q2 Ps satises a relationship of form 0 = +p +q 1 N (16.5)

0 = (Q2 + pQ + q )Ps .

The coecients p, q follow from symmetry considerations and the Jacobi relation. Rotate each term in the above equation by 90o and the project onto the traceless symmetric subspace; 0 = = +p + +q p 1+p+q N 1+p+q N Ps Ps .

+ q2

Jacobi relation (??) relates the second term to the rst: = 0 = 2 Q2 1 +p 2 + q2 1+p+q N Ps . Ps (16.6)
printed April 14, 2000

q 1+p+q 1 + 2p Q+ 4 2 N

DasGroup/book/chapter/e8family.tex 14apr2000

16.1. TWO-INDEX TENSORS

121

Comparing the coecients in (16.5) and (16.6) we obtain the characteristic equation for Q 1 5 Q2 Q 6 3(N + 2) Ps = 0 . (16.7)

We shall use this equation to obtain a Diophantine condition on admissible dimensions of the adjoint representation. Either eigenvalue of Q satises the characteristic equation 1 5 = 0, 2 6 3(N + 2) hence N can be expressed in terms of the eigenvalue N +2= 62 5 = 10 (6 1 ) 12 + 3( 1/6) 6 1 . (16.8)

It is convenient to reparametrize the two eigenvalues as = 1 , m6 = 1 m . 6m6 (16.9)

In terms of the parameter m, the dimension of the adjoint reperesentation is given by N = 122 + 10m + 360 m (16.10)

As N is an integer, allowed m are rationals built from any combination of subfactors of 360 = 23 32 5 in the numerator, and 1, 2, 5 or 10 in denominator, 45 distinct rationals in all. The existence of the pair of roots , is reected in the symmetry of (16.10) under interchange m/6 6/m, so we need to check only the 27 rationals m > 6; we postpone the Diophantine analysis to sect. 16.4. The associated projection operators are P|??| ??| = P|??| ??| =
*

= =

1 1

1 N .

(16.11) (16.12)

1 N

To compute the dimensions of the two subspaces we rst evaluate tr Ps Q = 1 N = N +2 . 2 (16.13)

The dimension of |??||??| is then given by d|??| ??| = tr P|??| ??| =

printed April 14, 2000

(N + 2)(1/ + N 1) , 2(1 /)

(16.14)

DasGroup/book/chapter/e8family.tex 14apr2000

122
E7 E7 E6 F4 D4 G2 A2 A1 Dimensions E8 E7 E6 F4 D4 G2 A2 A1

CHAPTER 16. E8 FAMILY OF INVARIANCE GROUPS

(10000000) (10000000) (1000000) (1000000) (000001) (000001) (1000) (1000) (0100) (0100) (10) (10) (11) (11) (2) (2) N2 272 272 272 152 152 92 62 8 32 = = = = = = = = = = = = = = = = = = (10000000) (1000000) (000001) (1000) (0100) (10) (11) (2) N 351 351 351 105 105 36 15 3 + + + + + + + + + + + + + + + + + + (01000000) (0100000) (001000) (0100) (1010) (03) (12) + (21) (0)
N (N 3) 2 1

27 27 27 15 15 9 6 0

+ + + + + + + + + + + + + + + + + +

1 ( ) (00000000) (0000000) (000000) (0000) (0000) (00) (00) (4) 1 351 351 351 105 105 36 15 1

A2 . The corresponding projection operators are listed in (16.3), (16.11) and (16.12). The parameter m is dened in (16.9). FILL THIS TABLE OUT!

Table 16.1: E8 family Clebsch-Gordon series for

and d|??| ??| is obtained by interchanging and . Substituting (16.10), (16.15) leads to d|??| ??| = d|??| ??| = 5(m 6)2 (5m 36)(2m 9) m(m + 6) 270(m 6)2 (m 5)(m 8) . m2 (m + 6)

(16.15)

The solutions that survive the Diophantine conditions form the E8 family, listed in table 16.1. To summarize, in absence of a primitive 4-index invariant, A A decomposes into 5 irreducible representations 1 = P|??| + P |??| + P + P|??| ??| + P|??| ??|
|??|

(16.16)

The decomposition is parametrized by integer m and is possible only if N and d|??| ??| satisfy Diophantine conditions (16.10), (16.15). The general strategy for decomposition of higher tensor products is as follows; the equation (16.6) reduces Q2 to Q, Pr weighted by the eigenvalues , . For higher tensor products we shall use the same result to decompose symmetric subspaces. We shall refer to a decomposition as boring if it brings no new Diophantine condition. As Q acts only on the symmetric subspaces, decompositions of antisymmetric subspaces will in general be boring, as was already the case in (16.3). We illustrate the technique by working out the decomposition of Sym3 A and |??||??| |??| in the next two sections.
DasGroup/book/chapter/e8family.tex 14apr2000 printed April 14, 2000

16.2. DECOMPOSITION OF SYM3 A

123

16.2

Decomposition of Sym3 A

Consider SymA fully symmetrized subspace of 3 A. As the rst step, project out the A and A A content of Sym3 A: P| ??| = 3 N +2 6(N + 1)(N 2 4) 5(N 2 + 2N 5)
r r

(16.17)

P |??|
|??|

(16.18)

P| ??| projects out Sym3 A A, and P |??| projects out the antisymmetric subspace (16.3) Sym3 A 2 A. The ugly prefactor is a normalization, and will play no role in what follows. We shall decompose the remainder of the Sym3 A space Pr = S P| ??| P |??| =
|??| |??|

(16.19)

by the invariant tensor Q restricted to the Pr remainder subspace Q= , = Q

r r

= Pr QPr Q

(16.20)

2 using (16.6) but symmetrization leads also to a new We can partially reduce Q invariant tensor 2 = 1 Q 3
r r

2 3

(16.21)

A calculation that requires applications of the Jacobi relation (3.105), symmetry identities such as
r r

=0

(16.22)

and relies on the fact that Pr contains no A 2A subspaces yields 3 = 1 Q 3

r r r

2 3

(16.23)

Reducing by (16.7) leads to 3 = ( + ) 1 Q 2 + 2 Q 3 3

printed April 14, 2000

r r

(16.24)

DasGroup/book/chapter/e8family.tex 14apr2000

124

CHAPTER 16. E8 FAMILY OF INVARIANCE GROUPS

The extra tensor can be eliminated by (16.21), and the result is a cubic equation (where we have substituted + = 1/6): for Q 1/18 0= Q Q Pr Q (16.25)

The projection operators for the corresponding three subspaces Q Q

P3 =

Pr (1/18 ) (1/18 ) m 162 (m 6)2 2 1Q Q = (m + 3)(m + 12) 6 6(m 6)2 1/18 Q Q

Pr

(16.26)

P4 = =

Pr ( 1/18) ( ) 54 (m 6)2 1 2 m 24 Q + Q (m + 3)(m + 6) 18(m 6) 18(m 6) Pr ( 1/18) ( ) m 108 (m 6)2 2 2(m 3) Q + Q (m + 6)(m + 12) 9(m 6) 108(m 6) 1/18 Q Q

Pr

(16.27)

P4

= =

Pr

(16.28)

The presumption is (still to be proven for a general tensor product) that only reductions occur in the symmetric subspaces, always via the Q characteristic equation. As overall scale of Q is arbitrary, there is only one rational parameter in the problem, either / or m, or whatever seems conveninent. Hence all dimensions and any coecients will be in m. To proceed, we follow the method outlined in appendix A. On P| ??|, P |??|
|??|

subspaces S Q has eigenvalues S QP| ??| = = 1 3 = ( + )

r a

| ??| = 1/3 |??| = 1/6 ,

|??|

(16.29)

S QP |??|

|??|

(16.30)

so the eigenvalues are | ??| = 1/3, |??| = 1/6, 3 = 1/18, 4 = , 4 = . The

|??|

dimension formulas (A.8) now require evaluation of tr S Q = = N (N + 2) 6 N (3N + 16) . 36 (16.31) (16.32)
printed April 14, 2000

tr (S Q)2 = birdTrack =
DasGroup/book/chapter/e8family.tex 14apr2000

16.3. DECOMPOSITION OF |??| |??||??|

125

Substituting into (A.8) we obtain the dimensions of the three representations: d3 = d4 = d4 = 27 (m 5) (m 8) (2m 15) (2m 9) (5m 36) (5m 24) (16.33) m2 (3 + m)(12 + m) 10(m 6)2 (m 5) (m 1) (2 m 9) (5 m 36) (5 m 24) (16.34) 3m2 (6 + m)(12 + m) 5 (m 5) (m 8) (m 6)2 (2m 15) (5m 36) (36 m) (16.35) m3 (3 + m)(6 + m)

The integer solutions of the above Diophantine conditions are listed in table 16.4. The main result of all this heavy birdtracking is that N > 248 is excluded by the positivity of d4 , and N = 248 is special, as P4 = 0 implies existence of a tensorial identity on the Sym3 A subspace. That dimensions should all factor into terms linear in m is althogehter not obvious at this point.

16.3

Decomposition of |??| |??| ??|

The decomposition of A2 tensors has split the traceless symmetric subspace into a pair of representations which we denoted by |??||??|, |??||??| . Now we turn to the decomposition of |??| |??||??| Kronecker product. We commence by identifying the A and A2 content of the |??||??||??| A3 Kronecker product. ??| components of |??| |??||??| are projected out by The |??|, |??||??| and | |??| P|??| = K|??| P|??| ??| = K|??| ??| P |??|
|??|

(16.36) (16.37) =
a

= K |??|
|??|

(16.38)

??| is the not-adjoint where the |??| |??||??| vertex is given by (16.11), and | |??| antisymmetric representation in (16.3). In this section double line denotes |??||??| representation, and K are normalization factors given by ratios of dimensions and appropriate Dynkin indices (4.7) (or 3j coecients (????)). As we shall not need them here, we do not write them out explicitly. We shall use the invariant tensor

R= to decompose the remainder subspace Pr = 1 P|??| P|??| ??| P |??| .

|??| printed April 14, 2000

(16.39)

(16.40)
DasGroup/book/chapter/e8family.tex 14apr2000

126

CHAPTER 16. E8 FAMILY OF INVARIANCE GROUPS

The eigenvalue of R on each of the above subspaces follows from invariance conditions (??) and the eigenvalue equation (3.50) QP|??| ??| = P|??| ??| (16.11): RP|??| = RP|??| ??| = RP |??| =
a

= (1 )P|??| 1 = P|??| ??| 2 = (1/2 )P |??| .

|??|

(16.41) (16.42) (16.43)

|??|

The characteristic equation for R projected to the remained subspace (cf. (3.54)) is obtained by evaluating R2 and R3 : R2 Pr = = Pr = 2 + Pr Pr Pr (16.44) (16.45)

2 + 2 ( + )R

+ 4( + ) 2 4 R R3 Pr = ( + )R

We have used (16.7), invariance, and the symmetry identity (16.22). =0 (16.46)

Eliminating the extra invariant tensor in (16.45) by (16.44) we nd that R satises a cubic equation symmetric under interchange 0 = (R ( + ))(R 2)(R 2 )Pr , (16.47)

so the eigenvalues of R on the six subspaces of |??||??||??| are |??| , |??| ??| , BB , 5 , 6 , 7 = 1 , 1/2, 1/2 , 1/6, 2, 2 . As in the preceeding section, this leads to decomposition of the remainder subspace Pr into three subspaces: P5 = P6 = P7 = 1 (R 2)(R 2 )Pr ( )2 1 (R ( + ))(R 2 )Pr 2( )2 1 (R ( + ))(R 2)Pr 2( )2 (16.48) (16.49) (16.50)

Dimension formulas of sect. ?? require that we evaluate tr 1 = N d |??| , tr R =

|??|

=0 = 2(1 )d|??| ??| (16.51)

printed April 14, 2000

tr R2 =

=2

DasGroup/book/chapter/e8family.tex 14apr2000

16.4. DIOPHANTINE CONDITIONS N d1 d2 d3 5 0 0 0 0 8 3 0 -3 0 9 8 1 0 27 10 14 7 64 189 12 28 56 700 1701 15 52 273 4096 10829 18 78 650 11648 34749 24 133 1463 40755 152152 30
190 1520 87040 392445

127 36 248 0 147250 779247

Table 16.2: All solutions of Diophantine conditions (16.5216.54); a bogus m = 30 solution still survives this set of conditions. This solution will be DG: MANGLER ET ORD HER, SOM JEG IKKE KAN LAESE eliminated by (??) which says that it does not exist for the F4 subgroup of E8 . Substituting into (A.8) we obtain the dimensions of the three representations d5 = d6 = d7 = 27(m 15)(2m 15)(m 8)(2m 9)(5m 24)(5m 36) (16.52) m2 (m + 3)(m + 12) 5(m 5)(2m 15)(m 6)2 (m 8)(5m 36) (36 m) (16.53) m3 (m + 3)(m + 6) 5120(m 5)(2m 15)(m 6)2 (m 9)(2m 9) (16.54) m3 (m + 6)(m + 12)

We see that nothing signicant is gained beyond the decomposition of Sym3 A of the preceeding section; we have recovered representations (??), (??). Representation (16.54) is new, but yields no new Diophantine condition. If we consider |??||??||??| instead, the only dierence is that (16.51) changes to 2(1 )d |??| ??| and we obtain 2 conjugate representations corresponding to m/6 6/m exchange; d 7 d 6 = = (16.55) (16.56)

16.4

Diophantine conditions

This Diophantine condition is satised only for m = 8, 9, 10, 12, 18, 20, 24, 30, 36, 40 and 45. As we shall show later, m = 20, 30 and 40 are bogus solutions, which do not survive further Diophantine conditions. The integer solutions of the above Diophantine conditions are listed in table 16.4. The formulas (16.52)-(16.54) yield, upon substitution of N , , the correct Clebsch-Gordan series for all members of the E8 family, table 16.4.
Remark 16.1 Remains to be done: P1 = 0 what special E8 relation? (reduction of 6-loops birdTrack?)

16.5

Generalized Young tableaux for E8

A very tedious table goes in here...?

printed April 14, 2000 DasGroup/book/chapter/e8family.tex 14apr2000

128

CHAPTER 16. E8 FAMILY OF INVARIANCE GROUPS

16.6

Conjectures of Deligne

The construction of the E8 family outlined in this chapter dates from early 1980s and was partially published in ref. [11]. In a 1995 paper Deligne [19] attributed to Vogel [18] the observation that for the 5 exceptional groups the antisymmetric 2 A and the symmetric Sym2 A adjoint representation tensor products P| ??| + P |??| and P + Ps in (16.3), respectively) can be decomposed into irreducible
|??|

representations in a uniform way, and that their dimensions and casimirs are rational functions of parameter a, related to the parameter m of (16.9) by a= 1 . m6 (16.57)

Here a is a = (, ), where is the largest weight of the representation, and the canonical bilinear form for the Lie algebra, in the notation of Bourbaki. Deligne conjectured that for A1 , A2 , G2 , F4 , E6 , E7 and E8 , the dimensions k A could likewise be expressed as rational of higher tensor representations functions of parameter a. The conjecture was checked on computer by Cohen and de Man [20] for dimensions and quadratic casimirs for all representations up to 4 A. They note that miraculously for all these rational functions both numerator and denominator factor in Q[a] as a product of linear factors. That can perhaps be deduced from the method of decomposing symmetric subspaces outlined in this chapter. Cohen and de Man have also observed that D4 should be added to Delignes list, in agreement with the denition of the E8 family construction here, consisting of A1 , A2 , G2 , D4 , F4 , E6 , E7 and E8 . Their algebra goes way beyond the results given in this chapter, which were originally obtained by paper and pencil birdtrack computations performed on trains while commuting between Gothenburg and Copenhagen. Cohen and de Man give formulas for 25 representations, 7 of which are also computed here. In current context a = = 1/6 is the symmetric space eigenvalue of the invariant tensor Q, my (16.9). The only role of the tensor Q is to split the traceless symmetric subspace, and its overall scale is arbitrary. In this chapter scale was xed by setting the adjoint representation quadratic casimir equal to unity, CA = 1 in (16.1). Deligne [19] and Cohen and de Man [20] x the scale of their , by setting + = 1, so their dimension formulas are stated in terms of a parameter related to the used here by CdM = 6. They refer to the interchange of the roots as involution . My (16.52) is their A My (16.53) is their Y3 My (16.54) is their C

DasGroup/book/chapter/e8family.tex 14apr2000

printed April 14, 2000

Chapter 17

E6 family of invariance groups

In this chapter we construct all invariance groups whose primitive invariant tena and fully symmetric d abc . The reduction of V 2 space yields a sors are b abc , d rule for evaluation of the loop contraction of four d-invariants (??). The reduction of V V yields the rst Diophantine condition (??) on the allowed dimensions of the dening representation. The reduction of V 3 tensors is straightforward, but the reduction of A V space yields the second Diophantine condition (d4 in table ??), and limits n to n 27. The solutions of Diophantine conditions form the E6 family consisting of E6 , A5 , A2 + A2 and A2 . For the interesting E6 (n = 27) case the cubic casimir (??) vanishes. This property of E6 enables us to evaluate loop contractions of six d-invariants (??), reduce A2 tensors (table ??) and investigate relations among the higher order casimirs of E6 (sect. ??). In sect. 17.7 we introduce a Young tableaux notation for any representation of E6 and exemplify its use in construction of Clebsch-Gordan series, table ??.

17.1

Reduction of two-index tensors

By assumption the primitive invariants set that we shall study

b = a a

b a = dbac = dacb c a (17.1) c

dabc = b dabc = b

Irreducibility of the dening n-dimensional representation implies

d dabc dbcd = a

= 129

(17.2)

130

CHAPTER 17. E6 FAMILY OF INVARIANCE GROUPS

The value of depends on the normalisation convention.1 We nd it convenient to set it to = 1. We can immediately write a Clebsch-Gordon series for the two-index tensors. The symmetric subspace in (8.5) is reduced by the dabc dcde invariant:

17.2 17.3

Mixed two-index tensors Diophantine conditions and the E6 family

The expressions for the dimensions of various representations (see tables in this chapter) are ratios of polynomials in n, the dimension of the dening representation. As the dimension of a representation should be a non-negative integer, these relations are the Diophantine conditions on the allowed values of n. The dimensions of the adjoint representation (??) is one such condition; the dimension of 4 from table ?? another. Furthermore, the positivity of the dimension 4 restricts the solutions to n 27. This leaves us with six solutions n = 3, 6, 9, 15, 21, 27. As we shall show this in chapter ??. Of these solutions only n = 21 is spurious - the remaining ve solutions are realized as the E6 row of the magic triangle, table ??. In the Cartan notation, the corresponding Lie algebras are A2 , A2 + A2 , A5 and E6 . We do not need to prove this, as for E6 Springer (see sect. ?? ). We call these invariance groups the E6 family, and list the corresponding dimensions, Dynkin labels and Dynkin indices in the tables of this chapter.

17.4

Three-index tensors

The V 3 tensor subspaces cf. SU (n) listed in table ?? are decomposed by invariant matrices constructed from the cubic primitive dabc in the following manner.

17.4.1 17.4.2 17.4.3

Fully symmetric V 3 tensors Mixed symmetry V 3 tensors Fully antisymmetric V 3 tensors

17.5

Dening adjoint tensors

We turn next to the determination of the Clebsch-Gordan series for V A representations. As always, this series contains the n-dimensional representation

17.6
1

Two-index adjoint tensors

Freudenthal (1954) takes = 5/2. Konstein (1977) and Kephart (1981) take = 10.
printed April 14, 2000

DasGroup/book/chapter/e6family.tex 14apr2000

17.7. DYNKIN LABELS AND YOUNG TABLEAUX FOR E6

131

17.6.1

Reduction of antisymmetric 3-index tensors

17.7 17.8 17.9

Dynkin labels and Young tableaux for E6 Casimirs for E6 Subgroups of E6

Why is A2 (6) in E6 family? The symmetric two-index representation (??) of SU (3) is 6 dimensional. The symmetric cubic invariant (17.2) can be constructed using a pair of Levi-Civita tensors Contractions of several dabc s can be reduced using the projection operator properties (5.32) of Levi-Civita tensors, yielding expressions like etc. The reader can check that, for example, the Springer relation (??) is satised. Why is A5 (15) in E6 family? The antisymmetric two-quark representation (??) of A5 = SU (6) is 15 dimensional. The symmetric cubic invariant (17.2) is constructed using the Levi-Civita invariant (5.28) for SU (6) The reader is invited to check the correctness of the primitivity assumption (??). All the other results of this chapter then follow. Is A2 + A2 (9) in E6 family? Exercise for the reader: unravel the A2 + A2 9-dimensional representation, construct the dabc invariant.

17.10

Springer relation

Substituting PA into the invariance condition (5.58) for dabc one obtains the Springer (1959, 1962) relation The Springer relation can be used to eliminate one of the three possible contractions of the three possible contractions of three dabc s.

17.10.1

Springers construction of E6

In the preceding sections we have given a self-contained derivation of the E6 family, in a form unfamiliar to most experts. Here we shall translate our results into more established notations, and identify those relations which have already been given by other authors.
printed April 14, 2000 DasGroup/book/chapter/e6family.tex 14apr2000

132

CHAPTER 17. E6 FAMILY OF INVARIANCE GROUPS

Consider the exceptional simple Jordan algebra A of Hermitian [3 3] matrices x with octonian matrix elements Freudenthal 1954, 1964), and its dual A (complex conjugate of A). Following Springer (1959, 1962), dene products (x, y ) = tr (xy ), xy = z 3 < x, y, z > = (x y, z ), and assume that they satisfy (x x) (x x) =< x, x, x > x Expanding x, x in ?? , we chose a normalization (ea , eb ) = b aa, b = 1, 2, ......, 27 and dene ea eb = dabc ec . (17.6) (17.5) (17.4) (17.3)

Substituting into ( ), we obtain ( ), with = 5 2 . Freudenthal and Springer prove that ( ) is satised if dabc is related to the usual Jordan product ea eb =??dabc ec , by 1 dabc ??dabc (ab tr (ec ) + ac tr (eb ) + bc tr (ea )) 2 1 (17.8) (ec ). + tr (ea ) tr (eb ) tr 2
b xa y E6 (27) is the group of isomorphisms which leave x, y = a and < b abc x, y, z >= d xa yb zc invariant. The derivation was constructed by Freudenthal (1954) (his equation (1.21)):

(17.7)

Dz < x, y > z = 2y x(xxz)

1 1 < x, y > z . 2 6

(17.9)

Substituting ( ) we obtain the projector ( ):

ac zc . (Dz )d = 3xa y b Pbd

(17.10)

The object < z, y > considered by Freudenthal is in our notation and the above factor 3 is the normalization ( ), Freudenthals equation (1.26). The invariance of the x-product is given by Freudenthal as < x, x x >>= 0. Substituting ( ) we obtain ( ) for dabc . (17.11)

DasGroup/book/chapter/e6family.tex 14apr2000

printed April 14, 2000

Chapter 18

F4 family of invariance groups

In this chapter we classify and construct all invariance groups whose primitive invariant tensors are a symmetric bilinear dab , and a symmetric cubic dabc , satisfying the relation (18.15). The result are summarized in table ??. Take as primitives a symmetric quadratic invariant dab and a symmetric cubic invariant dabc . As explained in chapter ??, we can use dab to lower all indices. In the birdtrack notation, we drop the circles denoting dab , and we drop arrows on all lines: dab = a b = (18.1)

dabc = dbac = dacb =

The dening n-dimensional representation is by assumption irreducible, so dabc dbcd = ad = dabb = 0 = 0 (18.3) (18.2)

to project out a 1-dimensional subspace). The (Otherwise, we could use value of depends on the normalisation convention (Schafer (1966) takes = 7/3).

18.1

Two-index tensors

dabc is a Clebsch-Gordan coecient for V V V , so V V space is decomposed into at least four subspaces: = + 1 n + 1 133

134

CHAPTER 18. F4 FAMILY OF INVARIANCE GROUPS 1 1 n (18.4)

1 = A + P3 + P1 We turn next to the decompositions induced by the invariant matrix Qab,cd = 1

(18.5)

We shall assume that Q does not decompose the symmetric subspace, ie. that its symmetrized projection can be expressed as 1 = A +B +C (18.6)

Together with the list of primitives (18.1), this assumption denes the F4 family1 . This corresponds to the assumption (15.3) in the construction of G2 . In the present case, we have not been able to construct the F4 family without assuming (18.6). Symmetrizing (18.6) in all legs we obtain 1A = (B + C ) Neither of the tensors can vanish, as contractions with s would lead to 0 = n+2=0 0 = =0 If the coecients were to vanish, 1 A = B + C = 0, we would have 1 B = (18.9) (18.8) (18.7)

Antisymmetrizing the top two legs we nd that in this case also the antisymmetric part of the invariant matrix Q (18.5) is reducible: 1 B = (18.10)

This would imply that the adjoint representation of SO(n) would also be the adjoint representation for the invariance group of dabc . However, the invariance condition 0=
1

(18.11)

Invariance groups with primitives dab , dabc which do not satisfy (18.6) also exist. The most familiar example is the adjoint representation of SU (n), n 4, where dabc is the Gell-Mann (?!) symmetric tensor.
DasGroup/book/chapter/f4family.tex 14apr2000 printed April 14, 2000

18.1. TWO-INDEX TENSORS cannot be satised for any positive dimension n : 0 = birdTrack 0 = birdTrack birdTrack n+1=0

135

(18.12)

Hence the coecients in (18.7) are non-vanishing, and are xed by tracing with ab : 1 = 2 n+2 (18.13)

Expanding the symmetrization operator we can write this relation as 1 + 1 2 = 2 n+2 + 1 n+2 (18.14)

(this xes A = 1/2, B = 2/(n + 2), C = 1/(n + 2) in (18.6)), or as + + = 2 + + n+2 2 (ab cd + ad bc + ac bd ) . (18.15) n+2

dabe decd + dade debc + dace debd =

In sect. 18.3 we shall show that this relation can be interpreted as the characteristic equation for [33] octonian matrices. This is the dening relation for the F4 family. The eigenvalue of the invariant matrix Q on the n-dimensional subspace can now be computed from (18.14) 1 1 + 1 2 = 2 n+2 (18.16)

1n2 2n+2

Let us now turn to the action of the invariant matrix Q on the antisymmetric subspace in (18.4). We evaluate Q2 with the help of the characteristic equation (18.14): = birdTrack + birdTrack = 1 2 = 1 2 n2 4 n+2 1 2 22 1 birdTrack + birdTrack birdTrack + 2 2 n+2 n+2 + 2 birdTrack + n+2 n+2 (18.17)

0 = A Q2
printed April 14, 2000

2 1n6 Q 2n+2 n+a

DasGroup/book/chapter/f4family.tex 14apr2000

136

CHAPTER 18. F4 FAMILY OF INVARIANCE GROUPS

The roots are A = 1/2, 5 = 4/(n + 2), and the associated projectors are PA = P5 = 8 n + 10 n+2 n + 10 + n+2 4 2 (18.18) (18.19)

The dimensions and Dynkin indices are listed in PA is the projector for the adjoint representation, as it satises the invariance condition (18.11): 1 birdTrack = birdTrack 2 1 (18.20) PA Q = PA 2

18.2
18.2.1

Dening adjoint tensors

Two-index adjoint tensors

18.3

Jordan algebra and F4 (26)

Consider the exceptional simple Jordan algebra of traceless Hermitian [33] matrices x with octonion matrix elements (Freudenthal 1964, Schafer 1966). The nonassociative multiplication rule for elements x can be written in a basis x = xa ea as ab e a e b = eb e a = I + dabc ec 3 a, b, c = 1, 2.......26. (18.21) where tr (ea ) = 0 and I is the [33] unit matrix. Traceless [33] matrices satisfy a characteristic equation 1 1 (18.22) x3 tr (x2 )x tr (x3 )I = 0 2 3 Substituting (?!) we obtain (?!) with normalization = 7 3 . Substituting (?!) into the Jordan identity (Schafer 1966) (xy )x2 = x(yx2 ) (18.23)

we obtain (?!). It is interesting to note that the Jordan identity (which denes Jordan algebra in the way Jacobi identity denes Lie algebra) is a trivial consequence of (?!). F4 (28) is the group of isomorphisms which leave forms tr (xy ) = ab xa xb and tr (xyz ) = dabc xa yb zc invariant. The derivation is given by Tits (1966) as Dz = (xz )y x(zy ) . Tits 1966, equation(28) (18.24) Substituting (18.21) we obtain the adjoint representation projection operator (18.18) (Dz )d = 3xa yb ad bc ac bd dbce dead dace debd + 9 3 zc (18.25)

DasGroup/book/chapter/f4family.tex 14apr2000

printed April 14, 2000

Chapter 19

E7 family of invariance groups

E7 family of invariance groups, negative dimensions: published (birdtrack free) as Negative dimensions and E7 symmetry, Nucl. Phys. B188, 373 (1981).

137

138

CHAPTER 19. E7 FAMILY OF INVARIANCE GROUPS

DasGroup/book/chapter/f4family.tex 14apr2000

printed April 14, 2000

Chapter 20

Exceptional magic
The study of invariance algebras pursued in the preceding chapters might appear to be a rather haphazard aair. Given a set of primitives, one gets some Diophantine equations, constructs the family of invariance algebras and moves onto the next set of primitives. However, a closer scrutiny of the Diophantine conditions leads to a surprise: most of the Diophantine equations are special cases of one and the same Diophantine equation, and they magically arrange all exceptional families into a single triangular pattern which we shall call the magic triangle.

20.1

Magic triangle

Our construction of invariance algebras has generated a series of Diophantine conditions which we now summarize. The adjoint representation conditions are: F4 family E6 family E7 family E8 family 360 n + 10 360 N = 4n 40 + n+9 360 N = 3n 45 + n/2 + 8 360 N = 10m 122 + m N = 3n 36 +

(20.1)

There is a striking similarity between the conditions for dierent families. If we dene F4 family E6 family E7 family m = n + 10 m=n+9 m = n/2 + 8 (20.2)

we can parametrize all the solutions of the above Diophantine conditions with a single integer m, see table 20.1. The Clebsch-Gordan series for A V Kronecker 139

140 m F4 E6 E7 E8 8 0 3 9 0 1 8 10 0 0 3 14 12 0 2 9 28 15 3 8 21 52 18 8 16 35 78 20 . . . .

CHAPTER 20. EXCEPTIONAL MAGIC 24 21 35 66 133 30 .

36 99 190

36 52 78 133 248

40 . . . .

360 . . . .

Table 20.1: All solutions of Diophantine conditions (20.1) not eliminated by other Diophantine conditions of chapter 15 through 18; those are marked by .

products also show a striking similarity. The characteristic equations (16.7), (??), (??) and (??) are the one and the same equation (Q 1) Q + 6 m Pr = 0 (20.3)

Here Pr removes the dening and V 2 subspaces, and we have rescaled the E8 operator Q (16.7) by factor 2. (Role of the Q operator is only to distinguish between two subspaces - we are free to rescale it as we wish). In the dimensions of the associated representations, eigenvalue 6/m introduces a new Diophantine denominator m + 6. For example, from (16.15), table ??, (??) and (??), the highest dimensional representation in V A has dimension (in terms of parametrization (20.2)): F4 family E6 family E7 family E8 family 15120 m+6 15120 2 4(m + 6) 188(m + 6) + 2928 m+6 15120 2 6(m + 6)2 246(m + 6) + 3348 m+6 27 360 11 15120 50m2 1485m + 19350 + m m+6 3(m + 6)2 156(m + 6) + 2673

(20.4)

These Diophantine conditions eliminate most of the spurious solutions of (20.1); only the m = 30, 60, 90 and 120 spurious solutions survive, but are in turn eliminated by further conditions. For the E8 family V V = V A = A A (the dening representation is the adjoint representation), hence the Diophantine condition (20.4) includes both 1/m and 1/(m + 6) terms. Not only can the four Diophantine conditions (20.1) be parametrized by a single integer m; the list of solutions table 20.1 turns out to be symmetric under the ip across the diagonal. F4 solutions are the same as those in the m = 15 column, and so on. This suggests that the rows be parametrized by an integer , in a fashion symmetric to the column parametrization by m. Indeed, the requirement of m symmetry leads to a unique expression which contains the four Diophantine conditions (20.1) as special cases: N= 360 360 ( 6)(m 6) 72 + + 3 m (20.5)
printed April 14, 2000

DasGroup/book/chapter/excMagic.tex 27/9-97

20.1. MAGIC TRIANGLE

141

We take m = 8, 9, 10, 12, 15, 24 and 36 as all the solutions allowed in table 20.1. By symmetry takes the same values. All the solutions ll up the magic triangle, table 20.1. Within each entry the number in the upper left corner is N , the dimension of the corresponding Lie algebra, and the number in the lower left corner is n, the dimension of the dening representation. The expressions for n for the top four rows are guesses. The triangle is called magic, partly because we arrived to it by magic, and partly because it contains Freudenthals (1964) magic square, marked by the dotted line in table 20.1.

printed April 14, 2000

DasGroup/book/chapter/excMagic.tex 27/9-97

142

CHAPTER 20. EXCEPTIONAL MAGIC

0 0 0 0 0 1 2 2 8 1 1

3 2 8 3 14

A1 A2 G2 D4 F4 E6

U(1) A1

0 0 0 0 0 2 2 3 9 0 1 3 5 8 6 21 14 52

3 3 9 4

7 28

2U(1) 3A 1
21

0 0 0 1 3 4 14 14

8 52 26 78

A1 A2 C3 F4

0 0 1 2 8

8 16 9

A2

14 35 15 66 32

C3 A5 D6 E7

2U(1)

2A2 A5 E6

0 0 3 3

35 20

27 133

U(1) A2

A1 G2

8 28 28

3A1 D4

78 78

133 133

56 248 248

E7 E8

A1

52

Table 20.2: Magic triangle. All exceptional Lie groups dening and adjoint representations form an array of the solutions of the Diophantine condition (20.5). Within each entry the number in the upper left corner is N , the dimension of the corresponding Lie algebra, and the number in the lower left corner is n, the dimension of the dening representation.

DasGroup/book/chapter/excMagic.tex 27/9-97

printed April 14, 2000

Chapter 21

Magic negative dimensions

21.1 21.2 E7 and SO(4) E6 and SU (3)

143

144

CHAPTER 21. MAGIC NEGATIVE DIMENSIONS

DasGroup/book/chapter/excMagic.tex 27/9-97

printed April 14, 2000

Appendix A

Recursive decomposition
This appendix deals with practicalities of computing eigenvalues, and is best skipped on rst reading. Let P stand for a projection onto a subspace, or the entire space (in which case P = 1). Assume that the subspace has already been reduced into m irreducible subspaces and a reminder
m

P =
=1

P + Pr

(A.1)

Now adjoin a new invariant matrix Q to the set of invariants. By assumption, Q does not reduce further the = 1, 2, . . . , m subspaces, ie. has eigenvalues 1 , 2 , . . . , m QP = P (no sum). (A.2) restricted to the reon the th subspace. We construct an invariant matrix Q maining (as yet not decomposed) subspace by
m

:= Pr QPr = P QP Q
=1

(A.3)

satises a minimal characteristic equaAs Pr is a nite dimensional subspace, Q tion of order n 2

n m+n

k = ak Q
k=0 =m+1

Pr ) = 0 (Q

(A.4)

with the corresponding projection operators (3.45). P =

=

Q Pr ,

= {m + 1, . . . , m + n}

(A.5)

Minimal in the above means that we drop repeated roots, so all eigenvalues are is an awkward object in computations, so we reexpress the projection distinct. Q operator in terms of Q as follows. 145

146

APPENDIX A. RECURSIVE DECOMPOSITION

) factor from Dene rst the polynomial obtained by deleting the (Q (A.4)
n1

(x ) =
= k=0

bk xk ,

, = m + 1, . . . m + n ,
( )

(A.6)

where the expansion coecient bk = bk depends on the choice of the subspace . Substituting Pr = P m =1 P and using the orthonomality of P we obtain an alternative formula for the projection operators m n1 1 k k b P P = ( P Q ) P (A.7) k bk k
k=0 =1

and dimensions d = tr P = 1 bk k
n1

bk

tr (P Q)k
=1

k d

. (A.8)

k=0

The utility of this formula lies in the fact that once the polynomial (A.6) is given only new data it requires are the traces tr (P Q)k , and those are simpler to k. evaluate than tr Q

DasGroup/book/chapter/appendInvar.tex 14apr2000

printed April 14, 2000

Appendix B

Properties of Young Projections

(H. Elvang and P. Cvitanovi c) In this appendix we prove the properties of the Young projections stated in sect. 8.4.

B.1

Uniqueness of Young projection operators

We now show that the Young projection operator PY is well-dened by proving the existence and uniqueness (up to sign) of a non-vanishing connection between the symmetrizers and antisymmetrizers in PY . The proof is induction over the number of columns t in the Young diagram Y. For t = 1 the Young projection operator consists of one antisymmetrizer of length s and s symmetrizers of length 1, and clearly the connection can only be made in one way, up to an overall sign. Assume the result to be valid for Young projections derived from Young diagrams with t 1 columns. Let Y be a Young diagram with t columns. The lines from A1 in PY must connect to dierent symmetrizers for the connection to be non-zero. Since there are exactly |A1 | symmetrizers in PY , this can be done in essentially one way, since which line goes to which symmetrizer is only a matter of an overall sign, and where a line enters a symmetrizer is irrelevant due to (5.9). After having connected A1 , connecting the symmetry operators in the rest of PY is the problem of connecting symmetrizers to antisymmetrizers in the Young projection PY , where Y is the Young diagram obtained from Y by slicing o the rst column. Thus Y has k 1 columns, so by the induction hypothesis the rest of the symmetry operators in PY can be connected in exactly one non-vanishing way (up to sign). 147

148

APPENDIX B. PROPERTIES OF YOUNG PROJECTIONS

The principles are illustrated below:

conn

o ectione! n

(B.1)

B.2

Normalization

We now derive the formula for the normalization factor Y such that the Young 2 = P . By the normalization of the projection operators are idempotent, PY Y symmetry operators, Young projection operators derived from fully symmetrical or antisymmetrical Young tableaux will be idempotent with Y = 1. 2 is simply P connected to P , hence it may be viewed as a set of outer PY Y Y symmetry operators connected by a set of inner symmetry operators. Expanding all the inner symmetrisers and using the uniqueness of the non-zero connection between the symmetrizers and antisymmetrizers of the Young projection, we nd that each term in the expansion is either 0 or a version of PY . In fact, the number of non-zero terms denote it Y is just the number |Y| dened in sect. 8.4. 1 For a Young diagram with s rows and t columns there will be a factor of |S i| 1 ( |Ai | ) for expansion of each inner (anti)symmetrizer, thus we nd
2 2 = Y PY

= |Y|
s i=1 |Si |!

2 Y s t i=1 |Si |! j =1 |Aj |! mess

sp

ag

he

tti

= Y

PY . t j =1 |Aj |!

(B.2)

Idempotency is then achieved by taking Y =

s i=1 |Si |! t j =1 |Aj |!

|Y|

(B.3)

Let Y be a Young tableau with |A1 | = s, |S1 | = t, |S2 | = t etc. We count in how many ways the lines entering the inner A1 pass through it to yield non-zero connections. We refer to
1st line 00000000000 11111111111

11111111111 000000000000 00000000000 111111111111

A1 S1 A1

111111 000000
...

A1 A2 S1 s+t -1 S2 s-1 +t-1

A t

At

S1

2nd li

ne

...

...

...

...

...

...

...

Y=

... ...

...
...

2 2 PY = Y
sth line

S2

(B.4)
...

... ...
Ss

...

...
DasGroup/book/chapter/appendUnitary.tex 25feb2000

St

printed April 14, 2000

B.3. ORTHOGONALITY

149

in the following. For each of the inner symmetrizers there must be exactly one from A1 . The rst line can pass through A1 in s ways and without loss of generality we may take it to pass straight through, connecting to S1 where it can pass through in t ways. Thus for the rst line, there were s + t 1 allowed roads through the inner symmetry operators. The second line may now pass through A1 in s 1 ways, and we can take it to pass straight through to S2 , where it has t possibilities. Thus we have found (s 1) + t 1 options for the second line. With a similar reasoning we nd (s 2) + t 1 allowed ways for the third line etc. Let wY be the number of ways of passing the m lines entering A1 through the inner symmetry operators. wY is then the product of the numbers found above, wY = (s + t 1)(s 1 + t 1)(s 2 + t 1) . Note that when calculating |Y| the product of the numbers in the rst column of the Young diagram is wY . We show Y = |Y| by induction on the number of columns t in the Young diagram Y. For a single column Young diagram, |Y| = |A1 |!, and the number of non-zero ways to connect the A1 symmetrizers to A1 in PY is |A1 |!, hence Y = |Y| for t = 1. Assume that Z = |Z| for any Young diagram Z with t 1 columns. Let Y be a Young diagram with t columns and let Y be the Young diagram obtained form Y by removal of the rst column. wY is the number of ways the lines 2 are allowed to pass through the entering the rst inner antisymmetrizer in PY inner symmetry operators. Finding the number of allowed paths for the rest of the lines is the problem of nding the number of allowed paths through the inner 2 , which is Y = |Y |. Now we have Y = Y w = symmetry operators of PY Y |Y |wY = |Y|.

B.3

Orthogonality

If Y and Z denote Young tableaux derived from the same Young diagram, then PY PZ = PZ PY = Y,Z PY , since there is a non-trivial permutation of the lines connecting the symmetry operators of Y with those of Z and by uniqueness of the non-zero connection the result is either P 2 = P or 0. Next, consider two dierently shaped Young diagrams Y and Z with the same number of boxes. Since at least one column must be bigger in (say) Y than in Z and the p lines from the corresponding antisymmetrizer must connect to dierent symmetrizers it is not possible to make a non-zero connection between the antisymmetry operators of PY to the symmetrizers in PZ , and hence PY PZ = 0. By a similar argument, PZ PY = 0.

printed April 14, 2000

DasGroup/book/chapter/appendUnitary.tex 25feb2000

150

APPENDIX B. PROPERTIES OF YOUNG PROJECTIONS

B.4

The dimension formula

The dimensions of the irreducible representations can be calculated recursively from the Young projection operators. Here is the recipe: Let Y be a Young diagram and Y the Young diagram obtained from Y by removal of the right-most box in the last row. Draw the Young projection operators corresponding to Y and Y and note that if we trace the last line of PY we obtain PY multiplied by a factor. Quite generally this contraction will look like
Rest of PY ...

(B.5)

Using (5.11) and (5.20) we have

k-1
k-1 k m

1 k

+ (k 1)

(B.6)

m-1 k-1

m-1

k-1

1 (n (m 1)) km

+ (k 1)

m-1 k-1

(k 1)(m 1)

Inserting (B.6) into (B.5) we see that the rst term is proportional to the projection PY . The second term vanishes:
Rest of P Y

m-1

k-1

m-1

k-1

nm+k km

(k 1)(m 1) km

k-1

S*

A*

lower loop

The lines going into S come from antisymmetrizers in the rest of the PY -diagram. One of these lines, from Aa , say, must pass from S through the lower loop to A
DasGroup/book/chapter/appendUnitary.tex 25feb2000 printed April 14, 2000

m-1

(B.7)

B.5. LITERATURE

151

and from A connect to one of the symmetrizers, say SS in the rest of the PY diagram. But due to the construction of the connection between symmetrizers and antisymmetrizers in a Young projection, there already is a line connecting Ss to Aa . Hence the diagram vanishes. The dimensionality formula follows by induction on the number of boxes in the Young diagrams with the dimension of a single box Young diagram being n. Let Y be a Young diagram with p boxes. We assume that the dimensionality formula is valid for any Young diagram with p 1 boxes. With PY obtained from PY as above, we have (using (B.6) and writing DY for the birdtrack diagram of PY ): dim PY = Y tr DY = nm+k Y tr DY km |Y | tr DY = (n m + k )Y |Y| fY fY = = (n m + k ) |Y| |Y| (B.8) (B.9) (B.10)

This completes the proof of the dimensionality formula (8.27).

B.5

Literature

This introduction to the Young tableaux is based on Lichtenberg [4], Hamermesh [5] and van der Waerden [6]. The rules for reduction of direct products: See Lichtenberg [4]. The rules are stated here as in (Elvang 1999). The method of constructing the Young projections directly from the Young tableaux is described in van der Waerden [6], who ascribes the idea to von Neumann. See also Kennedy slides [3]. Alternative labelling of Young diagrams: Fischler??.

printed April 14, 2000

DasGroup/book/chapter/appendUnitary.tex 25feb2000

152

APPENDIX B. PROPERTIES OF YOUNG PROJECTIONS

DasGroup/book/chapter/appendUnitary.tex 25feb2000

printed April 14, 2000

Bibliography
[1] R. Penrose 1971 [2] E. P. Wigner, Group Theory and Its Application to the Quantum Mechanics of Atomic Spectra (Academic Press, New York, 1959). [3] A. D. Kennedy, Group Algebras, Lie Algebras, and Cliord Algebras, Slides of colloquium given at the Moscow State University in 1997; www.ph.ed.ac.uk/adk/algebra-slides/all.html . [4] D. B. Lichtenberg, Unitary Symmetry and Elementary Particles (Academic Press, New York, 1970). [5] M. Hamermesh, Group Theory and Its Application to Physical Problems (Dover, New York, 1962). [6] B. L. van der Waerden, Algebra II 4th ed (Springer-Verlag, Berlin, 1959). [7] M. Tinkham, Group Theory and Quantum Mechanics (McGraw-Hill, New York, 1964). PCCOMPLETE: Gell-Mann matrices ref [8] M. Gell-Mann, Phys. Rev. ??, ?? (1964). [9] P. Cvitanovi c, Group theory for Feynman diagrams in non-Abelian gauge theories, Phys. Rev. D14, 1536 (1976). [10] P. Cvitanovi c, Classical and exceptional Lie algebras as invariance algebras, Oxford preprint 40/77 (June, 1977 - unpublished). [11] P. Cvitanovi c, Negative dimensions and E7 symmetry, Nucl. Phys. B188, 373 (1981). [12] P. Cvitanovi c and A.D. Kennedy, Spinors in negative dimensions, Phys. Scripta 26, 5 (1982). [13] P. Cvitanovi c, Group theory, part I (Nordita, February 1984); textbook in preparation. [14] P. Cvitanovi c, Group theory, part II, www.nbi.dk/GroupTheory/ (1984-1996); Webbook in preparation. [15] P. Cvitanovic, Group Theory For Feynman Diagrams In Nonabelian Gauge Theories: Exceptional Groups, Phys. Rev. D14, 1536 (1976).

153

154

BIBLIOGRAPHY

[16] P. Cvitanovic, P. G. Lauwers and P. N. Scharbach, Gauge Invariance Structure Of Quantum Chromodynamics, Nucl. Phys. B186, 165 (1981). [17] P. Cvitanovic, Negative Dimensions And The Emergence Of E7 Symmetry In Supergravity, NORDITA-80/39. [18] P. Vogel, Algebraic structures on modules of diagrams, (preprint 1995). [19] P. Deligne, La s erie exceptionelle de groupes de Lie, C.R. Acad. Sci. Paris, Ser. I, 322, 321 (1996). [20] A.M. Cohen and R. de Man, Computational evidence for Delignes conjecture regarding exceptional Lie groupes , C.R. Acad. Sci. Paris, Ser. I, 322, 427 (1996). [21] B. Giddings and J.M. Pierre, Some exact results in supersymmetric theories based on exceptional groups, eprint hep-th/9506196, Phys. Rev. D52, 6065 (1995). [22] B. G. Wybourne, Enumeration Of Group Invariant Quartic Polynomials In Higgs Scalar Fields, Austral. J. Phys. 33, 941 (1980). [23] P.H. Butler, R.W. Haase, B.G. Wybourne CALCULATION OF 3 JM FACTORS AND THE MATRIX ELEMENTS OF E7 GROUP GENERATORS Submitted to Austr. J. Phys Aug 1978 [24] P. Butera, G. M. Cicuta and M. Enriotti, Group Weight And Vanishing Graphs, Phys. Rev. D21, 972 (1980). [25] S. Okubo, Quartic Trace Identity For Exceptional Lie Algebras, J. Math. Phys. 20, 586 (1979). [26] S. Okubo, Gauge Groups Without Triangular Anomaly, Phys. Rev. D16, 3528 (1977). [27] S. Okubo, Constraint On Color Gauge Group, Phys. Rev. D16, 3535 (1977). [28] S. Okubo, Modied Fourth Order Casimir Invariants And Indices For Simple Lie Algebras, J. Math. Phys. 23, 8 (1982). [29] S. Okubo and J. Patera, Symmetrization Of Product Representations And General Indices And Simple Lie Algebras, J. Math. Phys. 24, 2722 (1983). [30] S. Okubo and J. Patera, General Indices Of Representations And Casimir Invariants, J. Math. Phys. 25, 219 (1984). [31] G. P. Canning, Diagrammatic Group Theory In Quark Models, Phys. Rev. D18, 395 (1978). [32] O. Aharony, S. S. Gubser, J. Maldacena, H. Ooguri and Y. Oz, Large N eld theories, string theory and gravity, hep-th/9905111. [33] N. Maru and S. Kitakado, Negative dimensional group extrapolation and a new chiral-nonchiral duality in N = 1 supersymmetric gauge theories, Mod. Phys. Lett. A12, 691 (1997) [hep-th/9609230]. [34] T. van Ritbergen, A. N. Schellekens and J. A. Vermaseren, Group theory factors for Feynman diagrams, Int. J. Mod. Phys. A14, 41 (1999) [hep-ph/9802376].
DasGroup/book/refs.tex 14apr2000 printed April 14, 2000

BIBLIOGRAPHY

155

[35] J. A. de Azcarraga, A. J. Macfarlane, A. J. Mountain and J. C. Perez Bueno, Invariant tensors for simple groups, Nucl. Phys. B510, 657 (1998) [physics/9706006]. [36] A. C. Pang and C. Ji, A Spinor technique in symbolic Feynman diagram calculation, J. Comput. Phys. 115, 267 (1994). [37] A. P. Kryukov and A. Y. Rodionov, Color: Program For Calculation Of Group Weights Of Feynman Diagrams In Nonabelian Gauge Theories, Comput. Phys. Commun. 48, 327 (1988). [38] P. Cvitanovic, Classics Illustrated: (NORDITA). Group Theory. Part 1, Print-84-0261

[39] P. H. Frampton and T. W. Kephart, Dynkin Weights And Global Supersymmetry In Grand Unication, Phys. Rev. Lett. 48, 1237 (1982). [40] T. W. Kephart and M. T. Vaughn, Tensor Methods For The Exceptional Group E6, Annals Phys. 145, 162 (1983). [41] A. D. Kennedy, Spinography: Diagrammatic Methods For Spinors In Feynman Diagrams, Phys. Rev. D26, 1936 (1982). [42] T. W. Kephart and M. T. Vaughn, Renormalization Of Scalar Quartic And Yukawa Couplings In Unied Gauge Theories Based On E6, Z. Phys. C10, 267 (1981). [43] T. P. Cheng, E. Eichten and L. Li, Higgs Phenomena In Asymptotically Free Gauge Theories, Phys. Rev. D9, 2259 (1974). [44] E. Cremmer and B. Julia, The SO(8) Supergravity, Nucl. Phys. B159, 141 (1979). [45] P. Cvitanovic, J. Greensite and B. Lautrup, The Crossover Points In Lattice Gauge Theories With Continuous Gauge Groups, Phys. Lett. B105, 197 (1981). [46] B. de Wit and G. t Hooft, Nonconvergence Of The 1/N Expansion For SU(N) Gauge Fields On A Lattice, Phys. Lett. B69, 61 (1977). [47] P. A. Dirac, The Quantum Theory Of Electron, Proc. Roy. Soc. Lond. A117 (1928) 610. [48] J. M. Droue, Transitions And Duality In Gauge Lattice Systems, Phys. Rev. D18, 1174 (1978). [49] G. R. Farrar and F. Neri, How To Calculate 35640 O (Alpha**5) Feynman Diagrams In Less Than An Hour, Phys. Lett. 130B, 109 (1983). [50] E. Fermi, An Attempt Of A Theory Of Beta Radiation. 1, Z. Phys. 88, 161 (1934). [51] P. H. Frampton and T. W. Kephart, Exceptionally Simple E(6) Theory, Phys. Rev. D25, 1459 (1982). [52] M. Gorn, Problems In Comparing Diagrams With Group Theory In Nonleptonic Decays, Nucl. Phys. B191, 269 (1981). [53] D. J. Gross and E. Witten, Possible Third Order Phase Transition In The Large N Lattice Gauge Theory, Phys. Rev. D21, 446 (1980).
printed April 14, 2000 DasGroup/book/refs.tex 14apr2000

156

BIBLIOGRAPHY

[54] G. t Hooft and M. Veltman, Regularization And Renormalization Of Gauge Fields, Nucl. Phys. B44, 189 (1972). [55] G. t Hooft, A Planar Diagram Theory For Strong Interactions, Nucl. Phys. B72, 461 (1974). [56] F. Gursey and P. Sikivie, E(7) As A Universal Gauge Group, Phys. Rev. Lett. 36, 775 (1976). [57] R. Dundarer, F. Gursey and C. Tze, Selfduality And Octonionic Analyticity Of S(7) Valued Antisymmetric Fields In Eight-Dimensions, Nucl. Phys. B266, 440 (1986). [58] R. Dundarer, F. Gursey and C. Tze, Generalized Vector Products, Duality And Octonionic Identities In D = 8 Geometry, J. Math. Phys. 25, 1496 (1984). [59] F. Gursey and M. Serdaroglu, E6 Gauge Field Theory Model Revisited, Nuovo Cim. 65A, 337 (1981). [60] M. Michel, V. L. Fitch, F. Gursey, A. Pais, R. U. Sexl, V. L. Telegdi and E. P. Wigner, Round table on the evolution of symmetries, In *Sant Feliu de Guixols 1983, Proceedings, Symmetries in physics (1600-1980)* 625-637. . [61] M. Evans, F. Gursey and V. Ogievetsky, From 2-D conformal to 4-D selfdual theories: Quaternionic analyticity, Phys. Rev. D47, 3496 (1993) [hep-th/9207089]. [62] G. t Hooft, Computation of the quantum eects due to a four-dimensional pseudoparticle, Phys. Rev. D14, 3432 (1976). [63] P. Jordan, J. von Neumann and E. Wigner, On An Algebraic Generalization Of The Quantum Mechanical Formalism, Annals Math. 35, 29 (1934). [64] A. D. Kennedy, Cliord Algebras In Two Omega Dimensions, J. Math. Phys. 22, 1330 (1981). [65] V. G. Kac, Lie Superalgebras, Adv. Math. 26, 8 (1977). [66] R. L. Mkrtchian, The Equivalence Of Sp(2n) And SO(2n) Gauge Theories, Phys. Lett. B105, 174 (1981). [67] G. Parisi and N. Sourlas, Random Magnetic Fields, Supersymmetry And Negative Dimensions, Phys. Rev. Lett. 43, 744 (1979). [68] J. Patera and R. T. Sharp, On The Triangle Anomaly Number Of SU(N) Representations, J. Math. Phys. 22, 2352 (1981). [69] S. Samuel, U(N) Integrals, 1/N, And The Dewit-T Hooft Anomalies, J. Math. Phys. 21, 2695 (1980). [70] A. Sirlin, A Class Of Useful Identities Involving Correlated Direct Products Of Gamma Matrices, Nucl. Phys. B192, 93 (1981). [71] R. Slansky, Group Theory For Unied Model Building, Phys. Rept. 79, 1 (1981). [72] G. E. Stedman, A Diagram Technique For Coupling Calculations In Compact Groups, J. Phys. A A8, 1021 (1975).
DasGroup/book/refs.tex 14apr2000 printed April 14, 2000

BIBLIOGRAPHY

157

[73] G. E. Stedman, UPDATE A Diagram Technique For ... (Cambridge University Press, Cambridge 19??). [74] F. Wilczek and A. Zee, Families From Spinors, Phys. Rev. D25, 553 (1982). [75] K. G. Wilson, Connement Of Quarks, Phys. Rev. D10, 2445 (1974). [76] B. G. Wybourne and M. J. Bowick, Basic Properties Of The Exceptional Lie Groups, Austral. J. Phys. 30, 259 (1977). [77] S. Okubo BRANCHING INDEX SUM RULES FOR SIMPLE LIE ALGEBRAS J.Math.Phys.26:2127,1985 [78] V.K. Agrawala MICU TYPE INVARIANTS OF EXCEPTIONAL SIMPLE LIE ALGEBRAS J.Math.Phys.21 (1980) 1577-1578 [79] A.M. Bincer and K. Riesselmann CASIMIR OPERATORS OF THE EXCEPTIONAL GROUP G(2) J.Math.Phys.34:5935-5941,1993 [80] A.M. Bincer CASIMIR OPERATORS OF THE EXCEPTIONAL GROUP F4: THE CHAIN B4 CONTAINS F4 CONTAINS D13 J.Phys.A27:3847-3856,1994 [81] J.A. Harvey PATTERNS OF SYMMETRY BREAKING IN THE EXCEPTIONAL GROUPS Nucl.Phys.B163:254,1980 [82] W.G. McKay, R.V. Moody and J. Patera DECOMPOSITION OF TENSOR PRODUCTS OF E8 REPRESENTATIONS Alg. Groups Geom. 3 ( 1986) 286-328 [83] M.A. Olshanetskii, V.B.K. Rogov ADJOINT REPRESENTATIONS OF EXCEPTIONAL LIE ALGEBRAS THEOR. MATH. PHYS. 72 (1987) 679-685 [84] P. Ramond IS THERE AN EXCEPTIONAL GROUP IN YOUR FUTURE? E(7) AND THE TRAVAILS OF THE SYMMETRY BREAKING Nucl.Phys.B126:509,1977 [85] P. Ramond INTRODUCTION TO EXCEPTIONAL LIE GROUPS AND ALGEBRAS (Cal Tech). CALT-68-577, Dec 1976. 58pp [86] Tetz Yoshimura SOME IDENTITIES SATISFIED BY REPRESENTATION MATRICES OF THE EXCEPTIONAL LIE ALGEBRA G2 Kings Coll., London). Print-75-0424 (KINGS COLL), (Received May 1975). 4pp [87] Killing 1888-1890 [88] Cartan 1894, 1952 [89] Cartan 1914 [90] Dynkin 1962 [91] McKay and Patera 1981 [92] Dickinson 1907 [93] Freudenthal 1964 [94] Springer 1959
printed April 14, 2000 DasGroup/book/refs.tex 14apr2000

158
[95] Springer 1962 [96] Brown 1969 [97] Yutsis 1962 [98] Brink 1968

BIBLIOGRAPHY

DasGroup/book/refs.tex 14apr2000

printed April 14, 2000

Index
, mathrmY 72 SU (n), 74 3-j , 76 3-j evaluation, 77 adjoint representation, 84 adjoint tensor, 81 basis vectors, 75 casimirs, 84 characters, 79 Dynkin labels, 84 representation, 69 three-vertex, 76 SU (n) representation dimensionality, 69 U (n) dimensionalities, 72 irreducible representations, 72 fY (n), 72 Young projection operator construction, 71 Young projection operators basis vectors, 75 k -standard arrangement, 68 3-j SU (n), 76 negative dimensions SU (n), 77 adjoint representation SU (n), 84 adjoint tensor SU (n), 81 two-index SU (n), 83 casimirs 159 SU (n), 84 characteristic equation unitary three-index tensor, 67 unitary two-index tensor, 65 characters SU (n), 79 color algorithm dimensionality, 73 completeness, 74 conjugate Young tableaux, 70 decomposition, 74 dimensionality SU (n) representation, 69 U (n), 72 color algorithm, 73 direct products Young tableaux, 70 Dynkin indices unitary two-index tensor, 66 Dynkin labels SU (n), 84 group unitary, 65, 147 inner product space, 16 invariant, 20 length row/column, 71 linear space, 16 mixed two-index tensor, 79 multi-particle state, 69

160 negative dimensions SU (n) 3-j evaluation, 77 application, 77 normalization Young proejction, 148 orthogonality Young projections, 149 representation SU (n), 69 standard arrangement, 69 SU(2) Young tableaux, 69 sum rule, SU (n) 3-j , 78 tableaux Young, 68 tensor three-index, 66 two-index, 65 tensors decomposition, 74 three-index tensor, 66 transpose Young tableaux, 69 two-index tensor adjoint SU (n), 83 mixed, 79 unitary, 65 Young tableaux, 69 unitary group, 65, 147 vector space, 16 Young diagram, 68 k -standard arrangement, 68 denition, 68 length, 71 Young projection, 71 , mathrmY 72
DasGroup/book/refs.tex 14apr2000

INDEX 3-j , 76 completeness, 74 decomposition, 74 normalization, 72, 148 orthogonality, 149 proof of properties, 147 properties, 72 three-vertex, 76 well-dened, 147 Young tableaux, 68 SU (n), 69 3-j , 76 conjugate, 70 decomposition, 74 denition, 68 dimensionality U (n), 72 direct products, 70 projection, 71 standard arrangement, 69 SU(2), 69 three-vertex, 76 transpose, 69 two-index tensor, 69

printed April 14, 2000