Electromagnetic Duality For Children

Electromagnetic Duality for Children
JM Figueroa-OFarrill [email protected] Version of 8 October 1998
Contents
I The Simplest Example: SO(3) 11
12 . 12 . 12 . 14 . . . . . . . . . . . . 16 18 18 20 24 26 27 28 30 30 31 33 39 40 40 42 44 44 45 47 48 49
1 Classical Electromagnetic Duality 1.1 The Dirac Monopole . . . . . . . . . . . . . . . . . . . . . . 1.1.1 And in the beginning there was Maxwell... . . . . . . 1.1.2 The Dirac quantisation condition . . . . . . . . . . . 1.1.3 Dyons and the ZwanzigerSchwinger quantisation condition . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 The t HooftPolyakov Monopole . . . . . . . . . . . . . . . 1.2.1 The bosonic part of the GeorgiGlashow model . . . 1.2.2 Finite-energy solutions: the t HooftPolyakov Ansatz 1.2.3 The topological origin of the magnetic charge . . . . 1.3 BPS-monopoles . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Estimating the mass of a monopole: the Bogomolnyi bound . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Saturating the bound: the BPS-monopole . . . . . . 1.4 Duality conjectures . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 The MontonenOlive conjecture . . . . . . . . . . . . 1.4.2 The Witten eect . . . . . . . . . . . . . . . . . . . . 1.4.3 SL(2, Z) duality . . . . . . . . . . . . . . . . . . . . .
2 Supersymmetry 2.1 The super-Poincar algebra in four dimensions . . . . . . . . . e 2.1.1 Some notational remarks about spinors . . . . . . . . . 2.1.2 The ColemanMandula and HaagLopuszaskiSohnius n theorems . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Unitary representations of the supersymmetry algebra . . . . . 2.2.1 Wigners method and the little group . . . . . . . . . . 2.2.2 Massless representations . . . . . . . . . . . . . . . . . 2.2.3 Massive representations . . . . . . . . . . . . . . . . . No central charges . . . . . . . . . . . . . . . . . . . . Adding central charges . . . . . . . . . . . . . . . . . . 1
[email protected]
draft version of 8/10/1998
2.3
N =2 2.3.1 2.3.2 2.3.3 2.3.4 2.3.5
2.4
N =4 2.4.1 2.4.2 2.4.3 2.4.4
Supersymmetric Yang-Mills . . . . . . . . . . . . . . . N =1 d=6 supersymmetric YangMills . . . . . . . . From N =1 in d=6 to N =2 in d=4 . . . . . . . . . . Higgsed N =2 supersymmetric YangMills . . . . . . N =2 avatar of the BPS-monopole . . . . . . . . . . . The supersymmetry bound is the Bogomolnyi bound The supersymmetry algebra in six dimensions . . . . The supersymmetry algebra in four dimensions . . . Supersymmetric Yang-Mills . . . . . . . . . . . . . . . N =1 d=10 supersymmetric YangMills . . . . . . . . Reduction to d=4: N =4 supersymmetric YangMills Monopoles and gauge bosons in N =4 supersymmetric YangMills . . . . . . . . . . . . . . . . . . . . . . . The mass bound for N =4 super YangMills . . . . . The supersymmetry algebra in ten dimensions . . . . The supersymmetry algebra in four dimensions . . .
. . . . . . . . . . . . . . .
51 52 57 59 60 62 63 64 66 67 69 74 76 77 78 80 81 81 82 84 85 87 88 89 90 92 96 99 100 103 107 109 111 111 113 116 116 119
3 Collective Coordinates 3.1 The metric on the moduli space . . . . . . . . . . . . . 3.1.1 The physical conguration space . . . . . . . . 3.1.2 The metric on the physical conguration space . 3.1.3 The metric on the moduli space . . . . . . . . . 3.1.4 The 1-monopole moduli space . . . . . . . . . . 3.1.5 The quantisation of the eective action . . . . . 3.1.6 Some general properties of the monopole moduli 3.2 dim Mk = 4k . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 The dimension as an index . . . . . . . . . . . . 3.2.2 Computing the index of D . . . . . . . . . . . . 3.2.3 Computing the current Ji (x, x) . . . . . . . . . 3.3 A quick motivation of hyperkhler geometry . . . . . . a 3.3.1 Riemannian geometry . . . . . . . . . . . . . . 3.3.2 Khler geometry . . . . . . . . . . . . . . . . . a 3.3.3 Ricci atness . . . . . . . . . . . . . . . . . . . 3.3.4 Hyperkhler geometry . . . . . . . . . . . . . . a 3.4 Mk is hyperkhler . . . . . . . . . . . . . . . . . . . . . a 3.4.1 Symplectic quotients . . . . . . . . . . . . . . . 3.4.2 Khler quotients . . . . . . . . . . . . . . . . . a 3.4.3 Hyperkhler quotients . . . . . . . . . . . . . . a 3.4.4 Mk as a hyperkhler quotient . . . . . . . . . . a 3.4.5 Another proof that Mk is hyperkhler . . . . . . a
. . . . . . . . . . . . . . . . . . . . . . . . space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
[email protected]
4 The Eective Action for N =2 Supersymmetric YangMills 122 4.1 Fermionic collective coordinates . . . . . . . . . . . . . . . . . 123 4.1.1 Computing the index . . . . . . . . . . . . . . . . . . . 124 4.1.2 Using supersymmetry . . . . . . . . . . . . . . . . . . . 125 4.2 The eective action . . . . . . . . . . . . . . . . . . . . . . . . 127 4.3 N =4 supersymmetry of the eective action . . . . . . . . . . . 130 4.3.1 N =4 supersymmetry in R4 : a toy model . . . . . . . . 131 4.3.2 N =4 supersymmetry in hyperkhler manifolds . . . . . 132 a 4.3.3 N =4 supersymmetry of Le . . . . . . . . . . . . . . . 134 4.4 A brief review of harmonic theory . . . . . . . . . . . . . . . . 136 4.4.1 Harmonic theory for riemannian manifolds . . . . . . . 136 4.4.2 Harmonic theory for Khler manifolds . . . . . . . . . 139 a and . . . . . . . . . . . . . . 142 4.4.3 Explicit formulas for 4.5 Quantisation of the eective action . . . . . . . . . . . . . . . 144 4.5.1 Canonical analysis . . . . . . . . . . . . . . . . . . . . 144 4.5.2 The quantisation of the eective hamiltonian . . . . . . 146 5 The Eective Action for N =4 Supersymmetric YangMills 149 5.1 Fermionic collective coordinates . . . . . . . . . . . . . . . . . 150
II
Arbitrary Gauge Groups
153
6 Monopoles for Arbitrary Gauge Groups 154 6.1 Topologically stable solutions . . . . . . . . . . . . . . . . . . 154 6.1.1 Some elements of homotopy . . . . . . . . . . . . . . . 156 The fundamental group . . . . . . . . . . . . . . . . . . 157 Higher homotopy groups . . . . . . . . . . . . . . . . . 162 6.1.2 Homotopy classication of nite-energy congurations . 166 Adding topological charges . . . . . . . . . . . . . . . . 171 6.2 The Dirac quantisation condition . . . . . . . . . . . . . . . . 174 6.3 Some facts about compact Lie groups and Lie algebras . . . . 177 6.3.1 Compact Lie groups . . . . . . . . . . . . . . . . . . . 177 6.3.2 The Weyl group . . . . . . . . . . . . . . . . . . . . . . 180 6.3.3 Root systems and simple Lie algebras . . . . . . . . . . 182 Reconstructing the group . . . . . . . . . . . . . . . . . 184 The centre of G . . . . . . . . . . . . . . . . . . . . . . 186 An example: A3 = D3 . . . . . . . . . . . . . . . . . . 188 Another example: D4 . . . . . . . . . . . . . . . . . . . 189 All the connected compact simple Lie groups . . . . . . 190 6.3.4 Some simple examples . . . . . . . . . . . . . . . . . . 192 3
[email protected]
The simple root system A2 . . . . . . . . . . . The simple root systems B2 = C2 . . . . . . . The simple root system G2 . . . . . . . . . . . 6.4 The magnetic dual of a compact Lie group . . . . . . 6.4.1 Some lattices and dual groups . . . . . . . . . H abelian . . . . . . . . . . . . . . . . . . . . H simple . . . . . . . . . . . . . . . . . . . . . An example: Spin(8) and its quotients . . . . Another example: Spin(12) and its quotients .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
192 194 195 195 197 198 198 200 202
List of Exercises
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 1.10 1.11 1.12 1.13 1.14 1.15 1.16 1.17 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 2.12 2.13 Angular momentum due to the electromagnetic eld The ZwanzigerSchwinger quantisation condition . Dyonic spectrum in CP non-violating theories . . . The spectrum of the model . . . . . . . . . . . . . . Boundary conditions on H and K . . . . . . . . . . The equations of motion for H and K . . . . . . . . Asymptotic form of the electromagnetic eld . . . . Gauge eld-strength in the Higgs vacuum . . . . . . Additivity of the magnetic charge g . . . . . . . . Dirac quantisation condition revisited . . . . . . . . The Bogomolnyi equation implies (1.10) . . . . . . The BPS-monopole . . . . . . . . . . . . . . . . . . The mass density at the origin is nite . . . . . . . (P )SL(2, Z) and its action on the upper half-plane SL(2, Z)-invariance of the mass formula . . . . . . . Orbifold points in the fundamental domain D . . . Properties of the mass matrix A( ) . . . . . . . . . The little groups for positive-energy particles . Helicity content of massless multiplets . . . . Highest spin in the multiplet . . . . . . . . . . Massive N =2 multiplets with s=0 and s=1/2 Short N =2 multiplets with s=0 and s=1/2 . . Short N =4 multiplets with s=0 . . . . . . . . N =1 supersymmetric YangMills . . . . . . . Supersymmetry variation of L . . . . . . . . . A Fierz rearrangement . . . . . . . . . . . . . Some -matrix identities . . . . . . . . . . . . ... and the proof of supersymmetry invariance The supersymmetry multiplet . . . . . . . . . L in a Majorana basis . . . . . . . . . . . . . 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 17 17 19 22 22 23 25 26 26 29 29 29 34 35 36 37 45 46 48 48 50 51 52 54 55 56 56 56 57
[email protected]
2.14 2.15 2.16 2.17 2.18 2.19 2.20 2.21 2.22 2.23 2.24 2.25 2.26 2.27 2.28 2.29 2.30 2.31 2.32 2.33 2.34 2.35 2.36 2.37 2.38 2.39 2.40 2.41 2.42 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11
Explicit N =2 supersymmetry transformations . . . . . . The SO(2) Noether current . . . . . . . . . . . . . . . . The perturbative spectrum of the model . . . . . . . . . Supersymmetric BPS-monopoles . . . . . . . . . . . . . . More -matrix identities . . . . . . . . . . . . . . . . . . Some euclidean -matrices . . . . . . . . . . . . . . . . . BPS-monopoles break one half of the supersymmetry . . Supersymmetric variation of the supercurrent . . . . . . The symmetric gauge-invariant energy momentum tensor The topological current . . . . . . . . . . . . . . . . . . . The momenta in the extra dimensions . . . . . . . . . The eective electromagnetic eld strength . . . . . . . . The space components of the topological charge . . . . . The Majorana condition . . . . . . . . . . . . . . . . . . Properties of Majorana and Weyl fermions . . . . . . . . Varying the lagrangian density . . . . . . . . . . . . . . . A ten-dimensional Fierz identity . . . . . . . . . . . . . . Some more -matrix identities . . . . . . . . . . . . . . . so(4) so(3) so(3) explicitly . . . . . . . . . . . . . . = Explicit realisation for I and J . . . . . . . . . . . . . The fundamental representation of su(4) . . . . . . . . . The fermionic terms in the lagrangian . . . . . . . . . . . N =4 supersymmetry transformations . . . . . . . . . . . su(4) invariance . . . . . . . . . . . . . . . . . . . . . . . The perturbative spectrum after higgsing . . . . . . . . . The supersymmetry algebra in ten dimensions . . . . . . Another topological current . . . . . . . . . . . . . . . . A self-dual 5-form . . . . . . . . . . . . . . . . . . . . . . Momentum and topological charge in this background . . The BPS-monopole as an instanton . . . . . . . D2 has no normalisable zero modes . . . . . . . The electric charge . . . . . . . . . . . . . . . . Two equations in one . . . . . . . . . . . . . . . D has no normalisable zero modes . . . . . . . A formula for the index of D . . . . . . . . . . . Another formula for I(M 2 ) . . . . . . . . . . . . Some properties of kernels . . . . . . . . . . . . Regularity properties of the propagator . . . . . The rst term doesnt contribute . . . . . . . . Another expression for the degree of the map 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
58 59 59 61 61 61 62 63 64 64 64 65 66 67 68 68 69 69 70 71 71 73 73 73 75 77 78 78 79 82 85 87 90 91 92 92 93 94 97 99
[email protected]
3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 3.20 3.21 3.22 3.23 3.24 3.25 3.26 3.27 3.28 3.29 3.30 3.31 3.32 3.33 3.34 3.35 3.36 3.37 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 4.12 4.13 4.14
A coordinate-free expression for . . . . . . . . The holonomy group of a connected manifold . The holonomy group of a riemannian manifold . The holonomy algebra of a riemannian manifold The Nijenhuis tensor . . . . . . . . . . . . . . . The holonomy group of a complex connection . The unitary group . . . . . . . . . . . . . . . . The Khler form . . . . . . . . . . . . . . . . . a Khler is Khler is Khler . . . . . . . . . . . . a a a The determinant of P . . . . . . . . . . . . . . The Ricci tensor is symmetric . . . . . . . . . . Curvature tensors in Khler manifolds . . . . . a An equivalent expression for the Ricci curvature Hyperkhler implies integrability . . . . . . . . a The holonomy group of a hyperkhler manifold a The Lie derivative acting on forms . . . . . . . The Poisson bracket . . . . . . . . . . . . . . . Triviality of the normal bundle . . . . . . . . . Integrability of the restricted complex structure ONeills formula . . . . . . . . . . . . . . . . . Hyperkhler structure of R4 . . . . . . . . . . . a Equivariance of the moment mapping . . . . . . The YangMillsHiggs eective action . . . . . A somewhat more explicit formula for Gab . . . The Christoel symbols . . . . . . . . . . . . . Explicit expressions for the complex structures . Fermion zero modes are chiral . . . . . From bosonic to fermionic zero modes . ... and back . . . . . . . . . . . . . . . A complex structure . . . . . . . . . . iA lives on the sphere . . . . . . . . . . Adjoint transformation . . . . . . . . . Computing . . . . . . . . . . . The rst kinetic term . . . . . . . . . . ... and the second kinetic term . . . . . N =2 supersymmetry in at space . . . The N =4 supersymmetry algebra . . . Another proof that 2 a = i a . . . . . Commuting supersymmetries . . . . . The supersymmetry charges . . . . . . 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
100 102 102 103 104 104 104 105 105 107 108 108 109 110 110 111 112 113 115 115 117 118 119 120 120 121 124 125 125 126 126 127 128 129 130 131 132 133 134 135
[email protected]
4.15 4.16 4.17 4.18 4.19 4.20 4.21 4.22 4.23 4.24 4.25 4.26 4.27 4.28 4.29 4.30 5.1 5.2 5.3 5.4 5.5 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9 6.10 6.11 6.12 6.13 6.14 6.15 6.16 6.17
The square of the Hodge . . . . . . . . The pointwise metric on forms . . . . . . The adjoint de Rham operator . . . . . . Harmonic forms . . . . . . . . . . . . . . Poincar duality . . . . . . . . . . . . . . e The Hodge and the complex structure The pointwise hermitian metric . . . . . Serre duality . . . . . . . . . . . . . . . . An explicit expression for d . . . . . . . An explicit expression for d . . . . . . . The canonical momenta . . . . . . . . . The eective hamiltonian . . . . . . . . . The supersymmetry charges revisited . . Some checks . . . . . . . . . . . . . . . . Classical N =4 supersymmetry algebra . Q is . . . . . . . . . . . . . . . . . . Euclidean Cliord algebra . . . . . . . . i and the Majorana condition . . . . . . The charge conjugation matrix explicitly MajoranaWeyl spinors . . . . . . . . Counting zero modes . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
137 138 138 138 139 140 141 142 142 143 145 145 145 146 146 148 150 150 151 152 152 154 155 156 157 159 161 162 163 164 166 168 174 175 186 187 187 199
Invariant metrics exist . . . . . . . . . . . . . . . . . The temporal gauge . . . . . . . . . . . . . . . . . . . 2 (S ) M0 lies in a single orbit . . . . . . . . . . . 1 (Y, y0 ) is a group . . . . . . . . . . . . . . . . . . . 1 (Y, y0 ) (Y, y1 ) . . . . . . . . . . . . . . . . . . . = SU (2) and SO(3) . . . . . . . . . . . . . . . . . . . . 1 (G) is abelian . . . . . . . . . . . . . . . . . . . . . 2 (X, x0 ) is a group . . . . . . . . . . . . . . . . . . . 2 (X, x0 ) 2 (X, x1 ) . . . . . . . . . . . . . . . . . . = Gauge invariance of the homotopy class . . . . . . . . Solving for . . . . . . . . . . . . . . . . . . . . . . The magnetic charge is covariantly constant . . . . . The non-abelian Stokes theorem . . . . . . . . . . . The centre is the intersection of all the maximal tori The central and root lattices are dual . . . . . . . . . Some facts about lattices . . . . . . . . . . . . . . . . The dual fundamental weights . . . . . . . . . . . . .
List of Figures
1.1 Fundamental domain (shaded) for the action of P SL(2, Z) on the upper half plane, and some of its P SL(2, Z) images. . . . . 36 1.2 Dyonic spectrum predicted by SL(2, Z) duality. Dots indicate dyons, crosses indicates holes in the dyonic spectrum. Only dyons with non-negative magnetic charge are shown. . . . . . 38 6.1 6.2 6.3 6.4 6.5 6.6 6.7 The two possible lifts to S n of a loop in A map S 2 G/H as a loop of loops. . A monopole. . . . . . . . . . . . . . . . Patching two monopoles. . . . . . . . . The root system A2 . . . . . . . . . . . The root system B2 . . . . . . . . . . . The root system G2 . . . . . . . . . . .
RPn . . . . . . . . . . 161
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 172 173 193 194 196
List of Tables
1.1 The perturbative spectrum after higgsing. . . . . . . . . . . . 20 6.1 Allowed angles between roots in a root system. . . . . . . . . 6.2 Dynkin diagrams of the simple root systems. . . . . . . . . . 6.3 Simple root systems, their Weyl groups, their complex Lie algebras, compact Lie groups and their centres. . . . . . . . 6.4 The connected compact simple Lie groups and their duals. . . 183 . 185 . 191 . 200
10
Part I The Simplest Example: SO(3)
11
Chapter 1 Classical Electromagnetic Duality

In this chapter we treat classical electromagnetic duality, and its manifestation (MontonenOlive duality) in some spontaneously broken gauge theories. We start by reviewing the Dirac monopole, and then quickly move on to the t HooftPolyakov monopole solution in the model described by the bosonic part of the SO(3) GeorgiGlashow model. We focus on the monopole solution in the PrasadSommereld limit and derive the Bogomolnyi bound for the mass of the monopole. We show that the classical spectrum of the model is invariant under electromagnetic duality. This leads to the conjecture of Montonen and Olive. We then discuss the Witten eect and show that the Z2 electromagnetic duality extends to an SL(2, Z) duality.
1.1
The Dirac Monopole
In this section we discuss the Dirac monopole and the Dirac(Zwanziger Schwinger) quantisation condition in the light of classical electromagnetic duality.
1.1.1
And in the beginning there was Maxwell...

E =0 E = B t B =0 B = E t (1.1)
Maxwells equations in vacuo, given by
12
[email protected]
are highly symmetric. In fact, they are invariant under both Lorentz transformations (in fact, conformal) and under electromagnetic duality: (E, B) (B, E) . (1.2)
Lorentz invariance can be made manifest by introducing the eld-strength F , dened by1 F 0i = F i0 = E i F ij =
ijk B k
In terms of F , Maxwells equations (1.1) become F = 0 where F =

1 F 2
F = 0 ,
(1.3)
with 0123 = +1. This formulation has the added virtue that the duality transformation (1.2) is simply F F F F , (1.4)
where the sign in the second equation is due to the fact that in Minkowski space 2 = 1.
In Minkowski space F = 0 implies that F = A A , for some electromag netic potential A . Similarly, F = 0 implies that F = A A , for some dual electromagnetic eld A . Notice however that the duality transformation relating A and A is nonlocal. It may be easier to visualise the following two-dimensional analogue, where the duality transformation relates functions and which satisfy = , where , take the values 0 and 1 now.
In the presence of sources, duality is preserved provided that we include both electric and magnetic sources: F = j F = k ,
and that we supplement the duality transformations (1.4) by a similar transformation of the sources: j k
1
k j .
In these lectures, we shall pretend to live in Minkowski space with signature (+). We will set c = 1 but will often keep explicit.
13
[email protected]
A charged point-particle of in the presence of an electromagnetic eld behaves according to the Lorentz force law. If the particle is also magnetically charge, the Lorentz law is then given by dx d2 x m = (qF + g F ) d 2 d where is the proper time, and m, q and g are the mass, the electric and magnetic charges, respectively. This formula is also invariant under duality provided we interchange the electric and magnetic charges of the particle: (q, g) (g, q). Problem: Derive the above force law from a particle action.
Notice that in the presence of magnetic sources, F = 0 whence there is no electromagnetic potential A . Nevertheless if at any given moment in time, the magnetic sources are localised in space, one may dene A in those regions where k = 0. The topology of such regions is generically nontrivial and therefore a nonsingular A need not exist throughout. Instead one solves for A locally, any two solutions being related, in their common domain of denition, by a gauge transformation. We will see this explicitly for the magnetic monopole.
1.1.2
The Dirac quantisation condition
Whereas a particle interacting classically with an electromagnetic eld does so solely via the eld-strength F , quantum mechanically the electromagnetic potential enters explicitly in the expression for the hamiltonian. Therefore the non-existence of the potential could spell trouble for the quantisation of, say, a charged particle interacting with the magnetic eld of a monopole. In his celebrated paper of 1931, Dirac [Dir31] studied the problem of the quantum mechanics of a particle in the presence of a magnetic monopole and found that a consistent quantisation forced a relation between the electric charge of the particle and the magnetic charge of the monopole: the so-called Dirac quantisation condition. We will now derive this relation. A magnetic monopole is a point-like source of magnetic eld. If we place the source at the origin in R3 , then the magnetic eld is given by B(r) = g r , 4 r3 (1.5)
where g is the magnetic charge. In these conventions, the magnetic charge is also the magnetic ux. Indeed, if denotes the unit sphere in R3 , then g=
B dS .
In the complement of the origin in R3 , B = 0, whence one can try to solve for a vector potential A obeying B = A. For example, we can 14
[email protected]
consider A+ (r) =
g 1 cos e , 4r sin
where (r, , ) are spherical coordinates. A+ = B everywhere but on the negative z-axis where = and hence A+ is singular. Similarly, A (r) = g 1 + cos e 4r sin
also obeys A = B everywhere but on the positive z-axis where = 0 and A is singular. It isnt that we havent been clever enough, but that any A which obeys A = B over some region will always be singular on some string-like region: the celebrated Dirac string. Over their common domain of denition (the complement of the z-axis in R3 ) (A+ A ) = 0, whence one would expect that there exists a function so that A+ A = . However the complement of the z-axis is not simply-connected, and need only be dened locally. For example, restricting ourselves to = , we nd that 2 A+ A = g g e = 2r 2 ,
but notice that since is an angle, the function is not continuous. It couldnt possibly be continuous, for if it were there would be no ux. Indeed, if again denotes the unit sphere in R3 , the upper and lower hemispheres respectively, and E the equator, the ux can be computed in terms of as follows: g=
B dS ( A+ ) dS +
+
= =
E
( A ) dS
A+ d
E
A d
=
E
= (2) (0) . Suppose now that we are quantising a particle of mass m and charge q in the eld of a magnetic monopole. The Schrdinger equation satised by the o wave-function is 2 2 =i 2m t 15
[email protected]
where = + ieA, for e = q/ . The Schrdinger equation is invariant o under the gauge-transformations: A A + and exp(ie) . This gauge invariance guarantees that solutions of the Schrdinger equation o obtained locally with a particular A will patch up nicely, provided that the wave-function be single-valued. This condition means that exp(ie) = exp(ieg/2) must be a single-valued function, which is equivalent to the Dirac quantisation condition: eg = 2 n for some n Z. (1.6)
The Dirac quantisation condition has the following physical interpretation. Classically, there is not much of a distinction between a a magnetic monopole and a very long and very thin solenoid. The eld inside the solenoid is of course, dierent, but in the limit in which the solenoid becomes innitely long (on one end only) and innitesimally thin, so that the inside of the solenoid lies beyond the probe of a classical experiment, the eld at the end of the solenoid is indistinguishable from that of a magnetic monopole. Quantum mechanically, however, one can in principle detect the solenoid through the quantum interference pattern predicted by the Bohm-Aharanov eect. The condition for the absence of the interference is precisely the Dirac quantisation condition.
1.1.3
Dyons and the ZwanzigerSchwinger quantisation condition
A quicker, more heuristic derivation of the Dirac quantisation condition (1.6) follows by invoking the quantisation of angular momentum. The orbital angular momentum L = r mr of a particle of mass m and charge q in the presence of a magnetic monopole (1.5) is not conserved. Indeed, using the Lorentz force law, dL = r m r dt = r qr B qg = r rr 4r3 d qg r = , dt 4 r whence the conserved quantity is instead qg r , 4 r a result dating to 1896 and due to Poincar. e J L 16
[email protected]
Exercise 1.1 (Angular momentum due to the electromagnetic eld) Show that the correction term is in fact nothing else but the angular momentum of the electromagnetic eld itself: J em =
R
d3 r r E B
where the E-eld is the one due to the charged particle.
If we now assume that the electromagnetic angular momentum is separately quantised, so that |J em | = 1 n 2 for some nZ,
we recover (1.6) again. The virtue of this derivation is that it provides a quick proof of the ZwanzigerSchwinger quantisation condition for dyons, as the following exercise asks you to show.
Exercise 1.2 (The ZwanzigerSchwinger quantisation condition) A dyon is a particle which possesses both electric and magnetic charge. Consider two dyons of charges (q=e , g) and (q =e , g ). Show that imposing the quantisation of the angular momentum of the resulting electromagnetic eld yields the following condition: eg e g = 2 n for some n Z . (1.7)
Notice that the existence of the electron (that is, a particle with charges (e, 0)) does not tell us anything about the electric charge of a monopole (q, g); although it does tell us something about the dierence between the electric charges of two such monopoles: (q, g) and (q , g). Indeed, (1.7) tells us immediately that g(qq ) = 2n for some integer n. If g has the minimum magnetic charge g = 2/e, then the dierence between the electric charges of the dyons (q, g) and (q , g) is an integer multiple of the electric charge of the electron: q q = ne for some integer n. But we cannot say anything further about the absolute magnitude of either q or q .
Exercise 1.3 (Dyonic spectrum in CP non-violating theories) Prove that if CP is not violated, then in fact there are only two (mutually exclusive) possibilities: either q = ne or q = ne + 1 e. 2 (Hint: use that under CP: (q, g) (q, g). Why?)
We will see later when we discuss the so-called Witten eect that this gets modied in the presence of a CP-violating term, and the electric charge of the dyon will depend explicitly on the angle measuring the extent of the CP violation. 17
[email protected]
1.2
The t HooftPolyakov Monopole
In 1974, t Hooft [tH74] and Polyakov [Pol74] independently discovered that the bosonic part of the Georgi-Glashow model admits nite energy solutions that from far away look like Dirac monopoles. In contrast with the Dirac monopole, these solutions are everywhere regular and do not necessitate the introduction of a source of magnetic chargethis being due to the twists in (the vacuum expectation value of) the Higgs eld.
1.2.1
The bosonic part of the GeorgiGlashow model
The GeorgiGlashow model was an early proposal to describe the electroweak interactions. We will be concerned here only with the bosonic part of the model which consists of an SO(3) YangMills eld theory coupled to a Higgs eld in the adjoint representation. The lagrangian density is given by L = 1 G G + 1 D D V () , 4 2 where the gauge eld-strength G is dened by G = W W eW W where W are gauge potentials taking values in the Lie algebra of SO(3), which we identify with R3 with the cross product for Lie bracket; the Higgs eld is a vector in the (three-dimensional) adjoint representation of SO(3), with components a = (1 , 2 , 3 ) which is minimally coupled to the gauge eld via the gauge-covariant derivative: D = eW ; the Higgs potential V () is given by V () = 1 2 a2 4
2
(1.8)
where 2 = and is assumed non-negative. The lagrangian density L is invariant under the following SO(3) gauge transformations: = g(x) 1 W W = g(x)W g(x)1 + g(x) g(x)1 , e 18 (1.9)
[email protected]
where g(x) is a possibly x-dependent 3 3 orthogonal matrix with unit determinant. The classical dynamics of the elds W and are determined from the equations of motion D G = e D and by the Bianchi identity D G = 0 , (1.11) D D = (2 a2 ) (1.10)
where G = 1 G . 2 The canonically conjugate momenta to the gauge eld W and the Higgs are given by Ei = G0i = D0 . (1.12) Dening Bi by Gij =
ijk B k
=+
ijk Bk ,
we can write the energy density as

1 1 1 H = 2 Ei Ei + 2 + 1 Bi Bi + 2 Di Di + V () 2
(1.13)
which is manifestly positive-semidenite and also gauge-invariant. We dene a vacuum conguration to be one for which the energy density vanishes. This means that G = 0 D = 0 V () = 0 .
For example, = a 3 and W = 0 is such a conguration, where ( a ) is e e an orthonormal basis for the three-dimensional representation space where the Higgs eld takes values. We also dene the Higgs vacuum as those congurations of the Higgs eld which satisfy the latter two equations above. Notice that in the Higgs vacuum, the Higgs eld obeys 2 = a2 . Any such vacuum conguration is not invariant under the whole SO(3), but only under an SO(2) U (1) subgroup, therefore this model exhibits spontaneous = symmetry breaking.
Exercise 1.4 (The spectrum of the model) Let = a + where a is a constant vector obeying a a = a2 . Expanding the lagrangian density in terms of , show that the model consists of a massless vector 1 boson A = a a W which we will identify with the photon, a massive scalar eld 1 = a a and two massive vector bosons W with the charge assignments given
19
[email protected]
in Table 1.1. (Hint: The masses are read o from the quadratic terms of the lagrangian density: L = +
1 2
MH
2 +
1 2
MW
+ W W +
whereas the charges are read o from the coupling to the photon. The photon couples minimally via the covariant derivative = + iQ/ A . By examining how this covariant derivative embeds in the SO(3) covariant derivative one can read o what Q are for the elds in the spectrum.)
Field A W
Mass 0 MH = a 2 MW = ae
Charge 0 0 e
Table 1.1: The perturbative spectrum after higgsing.
1.2.2
Finite-energy solutions: the t HooftPolyakov Ansatz
We now investigate the properties of nite-energy non-dissipative solutions to the equations of motion (1.10). But rst let us remark a few properties of arbitrary nite-energy eld congurations. The energy of a given eld conguration is the spatial integral E = d3 x H of the energy density H given by equation (1.13). Finite energy means that the integral exists, hence the elds must approach a vacuum conguration asymptotically. In particular the Higgs eld approaches the Higgs vacuum at spatial innity. If we think of the Higgs potential V as a function V : R3 R, let us dene M0 R3 as those points x R3 for which V (x) = 0. In the model at hand, M0 is the sphere of radius a, hence in any nite-energy conguration the Higgs eld denes a function from the sphere at spatial innity to M0 : ( ) lim (r) M0 . r
r We will assume that the resulting function is actually continuous. This would follow from some uniformity property of the limit and such a property has been proven by Taubes [JT80].
It is well-known that the space of continuous functions from a sphere to a sphere is disconnected: it has an innite number of connected components 20
[email protected]
indexed by an integer called the degree of the map. A constant map has degree zero, whereas the identity map has degree 1. Heuristically, the degree is the number of times one sphere wraps around the other. It is the direct two-dimensional generalisation of the winding number for maps from a circle to a circle.
Taking these remarks into account it is not dicult to construct maps of arbitrary degree. Consider the map fn : R3 \ {0} R3 dened by fn (r) = (sin cos n, sin sin n, cos ) , (1.14)
where (r, , ) are spherical coordinates. The map fn restricts to a map from the unit sphere in R3 to itself which has degree n.
The topological number of a nite-energy conguration is dened to be the degree of the map . The zero energy vacuum conguration W = 0, = a 3 has zero degree, since is constant. The topological number of e a eld congurationbeing an integeris invariant under any continuous deformation. In particular it is invariant under time evolution, and under gauge transformations, since the gauge group is connected. Hence if we set up a nite-energy eld conguration at some moment in time whose topological number is dierent from zero, it will never dissipate; that is, it will never evolve in time towards a trivial solution. In other words, in a sense it will be stable. We now investigate whether such stable solutions actually exist. We will narrow our search to spherically symmetric static solutions a solution is dened to be static if it is time-independent and in addition the timecomponent of the gauge eld W0 vanishes.
One may be tempted to think that the this latter condition is simply a choice of gauge. Indeed it is easy to show that W0 = 0 up to a gauge transformation, but the gauge transformation is actually time-dependent which is not allowed, since we are looking for time-independent solutions. Coming soon: More details on the explicit time-dependent gauge transformation.
It follows from (1.12) that for static eld congurations both E and vanish, and hence the energy agrees up to a sign with the lagrangian. This means that a eld conguration will be a solution to the classical equations of motion if and only if it extremises the energy. The t HooftPolyakov Ansatz for the monopole is given by (r) = r H(aer) er2 rj i Wa = aij 2 (1 K(aer)) er 0 Wa = 0 ,
(1.15)
for some arbitrary functions H and K. 21
[email protected]
Exercise 1.5 (Boundary conditions on H and K) Plugging the Ansatz into the expression for the energy density derive the following formula for the energy: E= 2 4a e dH + d
0 1 2
d 2 dH H d
2
1 2
K2 1
+ K 2H 2 +
H 2 2 4e2
. (1.16)
Deduce that the integral exists provided that the following boundary conditions hold: K 0 and H/ 1 suciently fast as K 1 O() and H O() as 0 . (1.17) This last equation means that H and K approach 0 and 1 respectively at least linearly in as 0.
Notice that with the above boundary conditions, r ( ) lim 2 H(aer) = a , r r r er which is (homotopic to) the identity map, and hence has degree 1. In other words, the topological number of such a eld conguration is 1. If such a solution exists it is therefore stable and non-dissipative.
Exercise 1.6 (The equations of motion for H and K) Work out the equations of motion for the functions H and K in either of two ways: either plug the Ansatz into the equations of motion (1.10) or else extremise the energy subject to the above boundary conditions. In either case you should get the following coupled nonlinear system of ordinary dierential equations: 2 d2 K = KH 2 + K(K 2 1) d 2 d2 H 2 2 = 2K 2 H + 2 H(H 2 2 ) . d e
(1.18)
Initial numerical studies of the above dierential equations for H and K together with the boundary conditions (1.17) suggested the existence of a solution. This was later proven rigourously by Taubes [JT80]. Notice that the asymptotic limit of the equations (1.18) in the limit yields: d2 K =K d 2 d2 h = 2 2h , 2 d e 22
[email protected]
where h H . The above equations can be solved for at once and one nds that the solutions compatible with the boundary conditions are K exp() = exp(MW r/ ) h exp(MH r/ ) , where MW and MH were obtained in Exercise 1.4. This means that the solution describes an object of nite size given by the largest of the Compton wavelengths /MH or /MW . In order to identify the solution provided by the t HooftPolyakov Ansatz we investigate the asymptotic electromagnetic eld. Recall that the electro1 magnetic potential is identied with A = a W , corresponding to the U (1) SO(3) dened as the stabiliser of . The electromagnetic eld can 1 therefore be identied with F = a G, . Because the t HooftPolyakov Ansatz corresponds to a static solution, there is no electric eld: F0i = 0. However, as the next exercise shows, there is a magnetic eld.
Exercise 1.7 (Asymptotic form of the electromagnetic eld) 1 Show that the asymptotic form of Fij = a Gij is given by Fij =
ijk
rk . er3
(1.19)
The form (1.19) of the electromagnetic eld shows that the asymptotic magnetic eld is that of a magnetic monopole: B= 1r . e r3
A quick comparison with equation (1.5) reveals that the magnetic charge of a t HooftPolyakov monopole is (up to a sign) twice the minimum magnetic charge consistent with the electric charge e and the Dirac quantisation condition; that is, twice the Dirac charge corresponding to e. This follows from the fact that the electromagnetic U (1) is embedded in SO(3) in such a way that the electric charge is the eigenvalue of the T3 isospin generator, which here is in the adjoint representation, which has integral isospin. The minimum electric charge is therefore emin = 1 e, relative to which the charge 2 of the t HooftPolyakov monopole is indeed one Dirac charge, again up to a sign.
In fact, there is another solution with the opposite magnetic charge. It is obtained from the t HooftPolyakov monopole by performing a parity transformation on the Ansatz. One might wonder whether there also exist dyonic solutions. These solutions would not be static in that W0 would be dierent from zero, but time-independent dyonic solutions have been found by Julia and Zee [JZ75] shortly after the results of t Hooft and Polyakov.
23
[email protected]
In summary, we see that the t HooftPolyakov solution describes an object of nite size which from far away cannot be distinguished from a Dirac monopole of charge 4/e. In contrast with the Dirac monopole, the t HooftPolyakov monopole is everywhere smooththis being due to the massive elds which become relevant as we approach the core of the monopole.
1.2.3
The topological origin of the magnetic charge
Although the t HooftPolyakov monopole is indistinguishable from far away from a Dirac monopole, as we approach its core the massive elds become relevant and the dierence becomes evident. In contrast to the Dirac monopole, which necessitates a singular point-like magnetic source at the origin, the t HooftPolyakov monopole is everywhere smooth and its magnetic charge is purely topological and, as we will see in this section, due completely to the behaviour of the Higgs eld far away from the core. Because of the exponential decay of the massive elds away from the core of the monopole, we notice that the Higgs eld approaches the Higgs vacuum. In other words, a large distance away from the core of the monopole, the Higgs eld satises D = 0 = a2 (1.20) (1.21)
up to terms of order O(exp(r/R)) where R is the eective size of the monopole, which is governed by the mass of the heavy particles.
Notice that equation (1.20) already implies that is a constant. Indeed, ( ) = 2 = 2e (W ) =0. What (1.21) tells us is that this constant is such that the potential attains its minimum. by (1.20)
It is therefore reasonable to assume that any nite-energy solution (not necessarily static or time-independent) of the YangMillsHiggs system (1.10) satises equations (1.20) and (1.21) except in a nite number of well-separated compact localised regions in space, which we shall call monopoles. In other words, we are considering a dilute gas of monopoles surrounded by a Higgs vacuum. Notice that in the Higgs vacuum, W = 1 , whence W is fully e determined except for the component in the -direction, which we denote by 24
[email protected]
A . Computing the components perpendicular to we nd that W = 1 1 + A . 2e a a
Exercise 1.8 (Gauge eld-strength in the Higgs vacuum) Show that the eld-strength in the Higgs vacuum points in the -direction and is 1 given by G = a F where F = 1 + A A . a3 e
Using the equations of motion (1.10) and the Bianchi identity (1.11) prove that F satises Maxwells equations (1.3).
Now let be a surface in the Higgs vacuum enclosing some monopoles in the volume it bounds. The magnetic ux through measures the magnetic charge. Notice that A doesnt contribute, and that we get: g
B dS 1 2ea2
ijk
j k dSi .
Notice that only the components of i tangential to contribute to the integral and therefore the magnetic charge only depends on the behaviour of on . Furthermore it only depends on the homotopy class of as map M0 ; in other words, the above integral is invariant under deformations of which preserve the Higgs vacuum: D = 0 and = 0 .
To see this, lets compute the variation of g under such a deformation of . Notice rst that ( 3
ijk ijk
(j k )) = (j k ) + 2
ijk j (
( k )) .
By Stokes theorem, the second term in the right-hand-side integrates to zero. Now, because j = 0, (j k ) = 0, whence j k is parallel to . Hence, (j k ) = 0. In other words, g = 0. This means that g is invariant under arbitrary deformations of and hence under any deformation which can be achieved by iterating innitesimal deformations: homotopies. Examples of homotopies are: 25
[email protected]
time-evolution of ; continuous gauge transformations on ; continuous changes of within the Higgs vacuum.
Exercise 1.9 (Additivity of the magnetic charge g ) Use the invariance of the magnetic charge under the last of the above homotopies, to argue that the magnetic charge is additive. (Hint: Use a contour deformation argument.)
Notice that the magnetic charge can be written as g = 4 N , where e N = 1 8a3 dSi
ijk
j k
(1.22)
which as the next exercise asks you to show, is the degree of the map : M0 .
Exercise 1.10 (Dirac quantisation condition revisited) Show that N is the integral of the jacobian of the map : M0 , which is the classical denition of the degree of the map. This means that N is an integer; a fact of which you may convince yourself by showing that if fn is the map dened by (1.14), then the value of N when is, say, the unit sphere in R3 , is equal to n. Taking this into account we recover again the Dirac quantisation condition: e g = 4N , (1.23)
with the same caveat as before about the fact that the minimum magnetic charge is twice the Dirac charge.
1.3
BPS-monopoles
Since the source for a Dirac monopole has to be put in by hand, its mass is a free parameter: it cannot be calculated. On the other hand, for the t Hooft Polyakov monopole there is no source, and the mass of the monopole is an intrinsic property of the YangMillsHiggs system and as such it should be calculable. In the next section we derive a lower bound for its mass. A natural question to ask is whether there are solutions which saturate this bound, and in the section after that such a solution is found: the BPS-monopole.
26
[email protected]
1.3.1
Estimating the mass of a monopole: the Bogomolnyi bound
In the centre of mass frame, all the energy of the monopole is concentrated in its mass. Therefore, taking equation (1.13) into account, M=
1 2
1 E 2 i
1 Ei + 1 Bi Bi + 2 + 1 Di Di + V () 2 2
R3
Ei Ei + Bi Bi + Di Di
where we have dropped some non-negative terms. We now redistribute the last term as follows: we introduce an angular parameter and we add and subtract Ei Di sin and Bi Di cos to the integrand. This yields M
1 2
Ei Di sin
R3 R3
+ Bi Di cos
R3 R3
+ sin sin
Di Ei + cos Di Ei + cos
Di Bi Di Bi ,
2
where we have introduced the obvious shorthand Vi notice that

R3
= Vi Vi . But now
Di Bi = =
R3
i Bi Bi dSi
by the Bianchi identity (1.11) by Stokes (1.24)
=a
B dS ag ,
where is the sphere are spatial innity and g is the magnetic charge of the solution. Notice that we have used the results of Exercise 1.8, which are valid since nite-energy demands that the sphere at spatial innity be in the Higgs vacuum. Similarly, using the equations of motion this time instead of the Bianchi identity, one nds out that
R3
Di Ei = a
E dS aq ,
(1.25)
where q is the electric charge of the solution. Therefore for all angles we have the following bound on the mass: M ag cos + aq sin . 27 (1.26)
[email protected]
The sharpest bound occurs when the right hand side is a maximum, which happens for q cos = g sin . In other words, tan = q/g. Plugging this back into (1.26), we nd the celebrated Bogomolnyi bound for the mass of a monopole-like solution in terms of the electric and magnetic charges: M a q2 + g2 , derived for the rst time in [Bog76] (see also [CPNS76]).
For the t HooftPolyakov monopole, which is electrically neutral, the Bogomolnyi bound yields M a|g| = 4a/e. But a/e = MW /e2 , whence M MW /, where 1/137 is the ne structure constant. If MW 90GeV, say, then MW 12TeVbeyond the present experimental range. This concludes the phenomenological part of these lectures!
(1.27)
1.3.2
Saturating the bound: the BPS-monopole
Having derived the Bogomolnyi bound, it is natural to ask whether there exist solutions which saturate the bound. We will follow custom and call such states BPS-states. We incurred in the inequalities for the mass by discarding certain terms from the mass formula. To saturate the bound, these terms would have to be equal to zero. Since they are all integrals of non-negative quantities, we must impose that these quantities vanish throughout space and not just asymptotically as the weaker requirement of nite-energy would demand. Let us concentrate on static solutions which saturate the bound. Static solutions satisfy Ei = 0 and D0 = 0. In particular they have no electric charge, so that sin = 0. This means that cos = 1 correlated to the sign of the magnetic charge. A quick inspection at the way we derived the bound reveals that for saturation we must also require that V () should vanish and that in addition the Bogomolnyi equation should hold: Bi = Di . (1.28)
Now the only way to satisfy V () = 0 and yet obtain a solution with nonzero magnetic charge, is for to vanish. Why? Because for = 0, 2 = a2 throughout space, and in particular, Di = i = 0. But using the Bogomolnyi equation (1.28), this means that Bi = 0, whence the solution carries no magnetic eld. One way to understand the condition = 0 is as a limiting value. We let 0, while at the same time retaining the boundary condition that at spatial innity satises (1.21). This is known as the PrasadSommereld limit [PS75].
28
[email protected]
Exercise 1.11 (The Bogomolnyi equation implies (1.10)) Show that the Bogomolnyi equation together with the Bianchi identity (1.11) implies the equations of motion (1.10) for the YangMillsHiggs system with = 0.
Of course, the advantage of the Bogomolnyi equation lies in its simplicity. In fact, it is not hard to nd an explicit solution to the Bogomolnyi equation in the t HooftPolyakov Ansatz, as the next exercise asks you to do.
Exercise 1.12 (The BPS-monopole) Show that the Bogomolnyi equation in the t HooftPolyakov Ansatz yields the following systems of equations for the functions H and K: dK = KH d dH = H + 1 K2 . d
Show that the following is a solution with the right asymptotic boundary conditions: H() = coth 1 . K() = sinh
Notice that the solution for the BPS-monopole is such that H() = 1 + O(exp()) , which does not contradict (1.17) because for = 0 the Higgs eld is massless. Its interactions are long range and hence the BPS-monopole can be distinguished from a Dirac monopole from afar. One consequence of the Bogomolnyi equation is that both the photon (through Bi ) and the Higgs (via Di ) contribute equally to the mass density. One can show that the longrange force exerted by the Higgs is always attractive and for static monopoles, it is equal in magnitude to the 1/r2 magnetic force. Therefore the forces add for oppositely charged monopoles, yet they cancel for equally charged monopoles. This is as it should be if static multi-monopole solutions saturating the Bogomolnyi bound are to exist. To see this, notice that the mass of a two-monopole system with charges g and g (of the same sign) is precisely equal to the sum of the masses of each of the BPS-monopoles. Hence there can be no net force between them.
Exercise 1.13 (The mass density at the origin is nite) Show that the mass density at the origin for a BPS-monopole is not merely integrable, but actually nite! (Hint: Notice that the mass density is given by Di 2 . Compute this for the BPS-monopole and expand as 0.)
29
[email protected]
1.4
Duality conjectures
In this section we discuss the observed duality symmetries between perturbative and nonperturbative states in the GeorgiGlashow model and the conjectures that this observation suggests. We start with the MontonenOlive conjecture and then, after introducing a CP-violating term in the theory, the Witten eect will suggest an improved SL(2, Z) duality conjecture.
1.4.1
The MontonenOlive conjecture
At = 0, the (bosonic) spectrum of the GeorgiGlashow model (including the BPS-monopoles) is the following: Electric Magnetic Charge Charge 0 0 0 0 q 0 0 g Spin/ Helicity 1 0 1 0
Particle Photon Higgs W boson M monopole
Mass 0 0 aq ag
where q = e . Two features are immediately striking: all particles satisfy the Bogomolnyi bound; and the spectrum is invariant under electromagnetic Z2 duality: (q, g) (g, q) provided that we also interchange the BPS-monopoles and the massive vector bosons. The invariance of the spectrum under electromagnetic duality is a consequence of the fact that the formula for the Bogomolnyi bound is invariant under electromagnetic duality and the fact that the spectrum saturates the bound. This observation prompted Montonen and Olive [MO77] to conjecture that there should be a dual (magnetic) description of this gauge theory where the elementary gauge particles are the BPS-monopoles and where the massive vector bosons appear as electric monopoles. This conjecture is reinforced by the fact that two very dierent calculations for the inter-particle force between the massive vector bosons (done by computing tree diagrams in the quantum eld theory) and between the BPS-monopoles (a calculation due to Manton) yield identical answers. Notice, however, that because of the Dirac quantisation condition, if the coupling constant e of the original theory is small, the coupling constant g of the magnetic theory must be large, and 30
[email protected]
viceversa. Hence the duality conjecture would imply that the strong coupling behaviour of a gauge theory could be determined by the weak coupling behaviour of its dual theorya very attractive possibility.
The formula for the Bogomolnyi bound is actually invariant under rotations in the (g, q)plane. But the quantisation of the electric and magnetic charges, actually breaks this symmetry down to the Z2 duality symmetry. In their paper, Montonen and Olive speculate that the massless Higgs could play the role of a Goldstone boson associated to the breaking of this SO(2) symmetry down to Z2 . I am not aware of any further progress in this direction.
The MontonenOlive conjecture suers from several drawbacks: there is no reason to believe that the duality symmetry of the spectrum is not broken by radiative corrections through a renormalisation of the Bogomolnyi bound; in order to understand the BPS-monopoles as gauge particles, we would expect that their spin be equal to oneyet it would seem naively that due to their rotational symmetry, they have spin zero; and the conjecture is untestable unless we get a better handle at strongly coupled theoriesof course, this also means that it cannot be disproved! We will see in the next chapter that supersymmetry solves the rst two problems. The third problem is of course very dicult, but we will now see that by introducing a CP violating term in the action, the duality conjecture will imply a richer dyonic spectrum which can be tested in principle.
1.4.2
The Witten eect
Exercise 1.3 asked you to compute the dyonic spectrum consistent with the quantisation condition (1.7) in a CP non-violating theory. You should have found that the electric charge q of a dyon with minimal magnetic charge g could take one of two sets of mutually exclusive values: either q = ne or 1 q = ne + 2 e, where n is some integer. We will see that indeed it is the former case which holds. Let N denote the operator which generates gauge transformations about the direction : 1 v = v a 1 W = D , ea 31
(1.29)
[email protected]
where v is any isovector. And consider the operator exp 2iN . In the background of a nite energy solution, the Higgs eld is in the Higgs vacuum at spatial innity, whence exp 2iN generates the identity transformation. On isovectors it generates a rotation about of magnitude 2||/a = 2, and on the gauge elds we notice that D = 0 in the Higgs vacuum. Since exp 2iN = 1, the eigenvalues of N are integral. To see what this means, we compute N . We can compute N since it is the charge of the Noether current associated with the transformations (1.29). Indeed, N= L
R3
0 W
W +
L 0
Using equation (1.29), and hence that = 0, we can rewrite N as N = 1 L Di . ae R3 0 Wi
Since the conjugate momentum to Wi is G0i = Ei = Ei , we nd that N= 1 q Ei Di = , ae R3 e (1.30)
where we have used the expression (1.25) for the electric charge q of the conguration. The quantisation of N then implies that q = ne for some integer n. Let us now introduce a -term in the action: L =
1 2
e2 32 2
G G =
e2 G G . 32 2
This term is locally a total derivative and hence does not contribute to the equations of motion. Its integral in a given conguration is an integral multiple (called the instanton number) of the parameter . is therefore an angular variable and parametrises inequivalent vacua. The Noether charge N gets modied in the presence of this term as follows: N N L 1 Di . ae R3 0 Wi
32
[email protected]
Computing this we nd N = e 0i G Di 16 2 a R3 e ijk = Gjk Di 2a 16 R3 e Bi Di = 2 8 a R3 e = 2 g, 8
where g, given by equation (1.24), is the magnetic charge of the conguration. In other words, q e N= + 2g. e 8 For the t HooftPolyakov monopole, eg = 4, hence the integrality of N means that e q = ne + for some n Z. (1.31) 2 This result, which was rst obtained by Witten in [Wit79], is of course consistent with the quantisation condition (1.7) since for a xed the dierence between any charges is an integral multiple of e.
1.4.3
SL(2, Z) duality
The action dened by L + L depends on four parameters: e, , and a. The dependence on the rst two can be unied into a complex parameter . To see this, let us rst rescale the gauge elds W eW . This has the eect of bringing out into the open all the dependence on e. The lagrangian is now L + L = 1 G G + G G + 1 D D V () , (1.32) 2 2 2 4e 32
where all the (e, )-dependence is now shown explicitly. We now dene a complex parameter 4 +i 2 , 2 e whose imaginary part is positive since e is real. To write the lagrangian explicitly in terms of it is convenient to introduce the following complex linear combination: (1.33) G G + i G . 33
[email protected]
It then follows that G G
= 2G G + 2iG G ,
whence the rst two terms in the lagrangian (1.32) can be written simply as
1 Im G G . (1.34) 32 Notice that because is an angular variable, it is only dened up to 2. This means that physics is invariant under +1. At = 0, the conjecture of electromagnetic duality says that e g = 4/e is a symmetry. But this duality transformation is just 1/ . We are therefore tempted to strengthen the conjecture of electromagnetic duality to say that for arbitrary , the physics should depend on only modulo the transformations:
T : +1 1 S: .
Exercise 1.14 ((P )SL(2, Z) and its action on the upper half-plane) The group SL(2, Z) of all 2 2 matrices with unit determinant and with integer entries acts naturally on the complex plane: a + b a b . = c d c + d Prove that this action preserves the upper half-plane, so that if Im > 0, so will its transform under SL(2, Z). Prove that the matrices 1 and 1 both act trivially (and are the only two matrices that do). Thus the action is not faithful, but it becomes faithful if we identify every matrix M SL(2, Z) with M . The resulting group is denoted P SL(2, Z) SL(2, Z)/{1}. The operations S and T dened above are clearly invertible and hence generate a discrete group. Prove that they satisfy the following relations: S2 = 1 and (ST )3 = 1 .
Prove that the group generated by S and T subject to the above relations is a subgroup of P SL(2, Z), by exhibiting matrices S and T whose action on coincides with the action of S and T . These matrices are not unique, since in going from P SL(2, Z) to SL(2, Z) we have to choose a sign. Nevertheless, for any choice of S and T , prove that the following matrix identities are satised: S 2 = 1 and (S T )6 = 1 . The matrices S and T thus generate a subgroup of SL(2, Z). Prove that this subgroup is in fact the whole group, which implies that S and T generate all of P SL(2, Z). (Hint: if you get stuck look in [Ser73].)
34
[email protected]
Is physics invariant under SL(2, Z)? Clearly this would be a bold conjecture, but no bolder than the original MontonenOlive Z2 conjecture, for in fact the evidence for both is more or less the same. Indeed, as we now show the mass formula for BPS-states is invariant under SL(2, Z). The mass of a BPS-state with charges (q, g) is given by the equality in formula (1.27). From formula (1.23) it follows that the allowed magnetic charges of the form g = nm 4/e, for some nm Z. As a consequence of the Witten eect, the allowed electric charges are given by q = ne e + nm e/2. The mass of BPS-states is then given by M 2 = 4a2 nt A( ) n , where n = (ne , nm )t Z Z and where A( ) = 1 Im 1 Re Re | |2 (1.35)
Exercise 1.15 (SL(2, Z)-invariance of the mass formula) Let a b M= SL(2, Z) . c d First prove that A(M ) = M 1
t
A( ) M 1 ,
and as a consequence deduce that the mass formula is invariant provided that we also transform the charges: nM n .
The improved MontonenOlive conjecture states that physics is SL(2, Z)invariant. If this is true, this means that the theories dened by two values of related by the action of SL(2, Z) are physically equivalent, provided that we are willing to relabel magnetic and electric charges by that same SL(2, Z) transformation. The action of P SL(2, Z) on the upper half-plane is well-known (see for example Serres book [Ser73]). There is a fundamental domain D dened by D = { C | Im > 0, | Re | 1 , | | 1} , 2 (1.36)
which has the property that its orbit under P SL(2, Z) span the whole upper half-plane and that no two points in its interior Int D = { C | Im > 0, | Re | < 1 , | | > 1} , 2 are related by the action of P SL(2, Z). 35
[email protected]
T -2
T -1
T -1S
ST
S
ST
-1
TS
Figure 1.1: Fundamental domain (shaded) for the action of P SL(2, Z) on the upper half plane, and some of its P SL(2, Z) images.
Exercise 1.16 (Orbifold points in the fundamental domain D) The fundamental domain D contains three orbifold points: i, = exp(i/3) and = exp(2i/3) which are xed by some nite subgroup of P SL(2, Z). Indeed, prove that i is xed by the Z2 -subgroup generated by S, whereas and are xed respectively by the Z3 -subgroups generated by T S and ST .
We end this section and this chapter with a discussion of the dyonic spectrum predicted by SL(2, Z)-duality. If we had believed in the electromagnetic Z2 -duality, we could have predicted the existence of the BPS-monopoles from the knowledge of the existence of the massive vector bosons (and viceversa). But this is as far as we could have gone with Z2 . On the other hand SL(2, Z) has innite order, and assuming that for all values of there are massive vector bosons in the spectrum, SL(2, Z)-duality predicts an innite number of dyonic states. This assumption is not as innocent as it seems, as the SeibergWitten solution to pure N =2 supersymmetric YangMills demonstrates; but it seems to hold if we have N =4 supersymmetry. But for now let us simply follow our noses and see what this assumption implies. Lets assume then that for all values of there is a state with quantum numbers n = (1, 0)t . The duality conjecture predicts the existence of one
36
[email protected]
state each with quantum numbers in the SL(2, Z)-orbit of n: M n= a b 1 c d 0 = a c .
Because M has unit determinant, a and c are not arbitrary integers: there exist integers b and d such that ad bc = 1. This means that a and c are coprime; that is, they dont have a common factor (other than 1). Indeed, if n were a common factor: a = na and c = nc for integers a and c , and we would have that n(a d bc ) = 1 which forces n = 1. We will now show that this arithmetic property of a and c actually translates into the stability of the associated dyonic state!
Exercise 1.17 (Properties of the mass matrix A( )) Notice that the matrix A( ) in the mass formula (1.35) enjoys the following properties for all in the upper half-plane: det A( ) = 1 and A( ) is positive-denite.
Prove that this latter property implies that the mass formula denes a distance function, so that in particular it obeys the triangle inequality. In other words, if 2 we dene n 2 Mn that is, the Bogomolnyi mass of a dyonic state with that charge assignmentthen prove that n+m n + m . (1.37)
Now lets consider a dyonic state q = (a, c)t . The triangle inequality (1.37) says that for any two dyonic states n and m which obey n + m = q, the mass of the q is less than or equal to the sum of the masses of n and m. But we claim that when a and c are coprime, the inequality is actually strict! Indeed, the inequality is only saturated when n and m, and hence q, are collinear. But if this is the case, a and c must have a common factor. Assume for a contradiction that they dont. If n = (p, q)t and m = (r, s)t , then we must have that both p and r are smaller in magnitude to a, and that q and s are smaller in magnitude to c. But collinearity means that pc = qa. Since a and c are relatively prime, it must be that a divides p so that there is some integer n such that p = an, which contradicts that p is smaller in magnitude to a. This also follows pictorially from the fact that a and c are coprime if and only if in the straight line from the origin to q Z2 R2 , q is the rst integral point. Therefore the dyonic state represented by q is is a genuine stable state which cannot be interpreted as a bound state of other dyonic states with smaller charges.
Contrast this with the case where the triangle inequality saturates. In this case, this means that the bound state of two dyons with collinear charges exhibits no net force between its constituents. Compare this with the discussion in section 1.3.2.
37
[email protected]
. . . . . . . . . . .
. . .
. .. . . . .. . . .... . .
. . . .
. . .
. . . .
Figure 1.2: Dyonic spectrum predicted by SL(2, Z) duality. Dots indicate dyons, crosses indicates holes in the dyonic spectrum. Only dyons with nonnegative magnetic charge are shown. The dyonic states in the SL(2, Z)-orbit of (1, 0) can be depicted as follows: Notice that for c=0 we have the original state and its charge conjugate. For c=1 we have the Julia-Zee dyons but with quantised electric charge: a can be an arbitrary integer. For c=2, we have that a must be odd. Notice that in every rational direction (that is, every half-line with rational slope emanating from the origin) only the rst integral point is present. As explained above these are precisely those points (m, n) whose coordinates are coprime. As we will see in the context of N =4 supersymmetric YangMills theory, the dyonic spectrum is in one-to-one correspondence with square-integrable harmonic forms on monopole moduli space. This is a fascinating prediction: it says that there is an action of the modular group on the (L2 ) cohomology of monopole moduli space.
38
Chapter 2 Supersymmetry
In this chapter we discuss the intimate relation between supersymmetry and the Bogomolnyi bound. The eect of supersymmetry is two-fold: rst of all, it enforces the bound since this is a property of unitary representations of the supersymmetry algebra; but it also protects the bound against quantum corrections, guaranteeing that if a state saturates the bound classically, it does so quantum mechanically. This last assertion follows because, as we will see, supersymmetry multiplets corresponding to BPS-states are smaller than the multiplets of states where the Bogomolnyi bound is not saturated. We rst discuss the supersymmetry algebra and its representations. For deniteness we shall work in four dimensions, but much of what well say can (and will) be used in dimensions other than four. It will be while studying (massive) representations with central charges that we will see the mechanism by which the Bogomolnyi bound follows from the algebra. We then illustrate this fact by studying a particular example: N =2 supersymmetric YangMills in four dimensions. We dene this theory by dimensional reduction from N =1 supersymmetric YangMills in six dimensions. This theory admits a Higgs mechanism by which the gauge symmetry is broken to U (1) while preserving supersymmetry. The higgsed spectrum falls into a massless gauge multiplet corresponding to the unbroken U (1) and two massive short multiplets. From the structure of the short N =2 multiplets we can deduce that the N =2 supersymmetry algebra admits central charges and, moreover, that the multiplets containing the massive vector bosons must saturate the mass bound. We will also see that this theory admits BPS-like solutions, which are shown to break one half of the supersymmetries. This implies that the BPS-monopole belongs to a short multiplet and suggests that the bound which follows abstractly from the supersymmetry algebra agrees with the Bogomolnyi bound for dyons given by equation (1.27). This is shown to be case. Nevertheless the short multiplets containing the massive vector 39
[email protected]
bosons and those containing the BPS-monopole have dierent spins, whence N =2 supersymmetric YangMills does not yet seem to be a candidate for a theory which is (MontonenOlive) self-dual. This problem will be solved for N =4 supersymmetric YangMills, which we study as the dimensional reduction of ten-dimensional N =1 supersymmetric YangMills. At a formal level, N =4 supersymmetric YangMills is qualitatively very similar to the N =2 theory; except that we will see that the short multiplets which contain the solitonic and the fundamental BPS-states have the same spin. This prompts the question whether N =4 supersymmetric YangMills is self-dual a conjecture that we will have ample opportunity to test as the lectures progress.
2.1
The super-Poincar algebra in four die mensions
In this section we will briey review the supersymmetric extension of the fourdimensional Poincar algebra. There are plenty of good references available e so we will be brief. We will follow for the most part the conventions in [Soh85], to where we refer the reader for the relevant references on supersymmetry.
2.1.1
Some notational remarks about spinors
The Lorentz group in four dimensions, SO(1, 3) in our conventions, is not simply-connected and therefore, strictly speaking, has no spinorial representations. In order to consider spinorial representations we must look to the corresponding spin group Spin(1, 3) which happens to be isomorphic to SL(2, C)the group of 2 2-complex matrices with unit determinant. From its very denition, SL(2, C) has a natural two-dimensional complex representation, which we shall call S. More precisely, S is the vector space C2 with the natural action of SL(2, C). If u S has components u = (u1 , u2 ) relative to some xed basis, and M SL(2, C), the action of M on u is dened simply by (M u) = M u . We will abuse the notation and think of the components u as the vector and write u S. This is not the only possible action of SL(2, C) on C2 , though. We could also dene an action by using instead of the matrix M , its complex conjugate M , its inverse transpose (M t )1 or its inverse hermitian adjoint (M )1 , since they all obey the same group multiplication law. These choices correspond, respectively to the conjugate representation S, the dual representation S , . We will use the following notation: and the conjugate dual representation S if u S, then u S, u S and u S . These representations 40
[email protected]
= are not all dierent, however. Indeed, we have that S S and S S , = which follows from the existence of : an SL(2, C)-invariant tensor (since M M = (det M ) and det M = 1) which allows us to raise and lower indices in an SL(2, C)-covariant manner: u = u , and u = u . We use conventions where 12 = 1 and 12 = 1. Because both the Lie algebra sl(2, C) (when viewed as a real Lie algebra) and su(2) su(2) are real forms of the same complex Lie algebra, one often employs the notation (j, j ) for representations of SL(2, C), where j and j are the spins of the two su(2)s. In this notation the trivial one dimensional representation is denoted (0, 0), whereas S = ( 1 , 0). The two su(2)s are 2 actually not independent but are related by complex conjugation, hence S = 1 (0, 2 ). In general, complex conjugation will interchange the labels. If the labels are the same, say ( 1 , 1 ), complex conjugation sends the representation 2 2 to itself and it makes sense to restrict to the sub-representation which is xed by complex conjugation. This is a real representation and in the case of the ( 1 , 1 ) representation of SL(2, C), it coincides with the dening representation 2 2 of the Lorentz group SO(1, 3): that is, the vector representation. Indeed, given a 4-vector P = (p0 , p) we can turn it into a bispinor as follows: p0 + p3 p1 ip2 P P = p1 + ip2 p0 p3 where = (1, ) with the Pauli matrices: 1 = 0 1 1 0 2 = 0 i i 0 3 = 1 0 0 1 .
Since the Pauli matrices are hermitian, so will be P provided P is real. The Pauli matrices have indices , which shows how SL(2, C) acts on this space. If M SL(2, C), then the action of M on such matrices is given by P M P M . This action is linear and preserves both the hermiticity of P and the determinant det( P ) = P 2 = p2 p p, just as we expect of 0 Lorentz transformations. We can summarise this discussion by saying that the are ClebschGordon coecients intertwining between the vector 1 1 and the ( 2 , 2 ) representations of SL(2, C). Notice also that both M and M at the same way on bispinors, which reiterates the fact that SL(2, C) is the double-cover of the Lorentz group SO(1, 3). Finally we discuss the adjoint representation of the Lorentz group, which is generated by antisymmetric tensors L = L . In terms of bispinors, such an L becomes a pair (L , L ) where L = L and similarly for L . In other words, L transforms as the (1, 0)(0, 1) representation of SL(2, C): notice that we need to take the direct sum because the representation is real. 41
[email protected]
2.1.2
The ColemanMandula and HaagLopusza ski n Sohnius theorems
Back in the days when symmetry was everything, physicists spent a lot of time trying to unify the internal symmetries responsible for the observed particle spectrum and the Poincar group into the same group: the holy e grail being the so-called relativistic quark model. However their hopes were dashed by the celebrated no-go theorem of Coleman and Mandula. In a nutshell, this theorem states that the maximal Lie algebra of symmetries of the S-matrix of a unitary local relativistic quantum eld theory obeying some technical but reasonable assumptions (roughly equivalent to demanding that the S-matrix be analytic), is a direct product of the Poincar algebra e with the Lie algebra of some compact internal symmetry group. Since Lie algebras of compact Lie groups are reductive: that is, the direct product of a semisimple and an abelian Lie algebras, the largest Lie algebra of symmetries of the S-matrix is a direct product: Poincar semisimple abelian. In e particular this implies that multiplets of the internal symmetry group consist of particles with the same mass and the same spin or helicity.
If all one-particle states are massless, then the symmetry is enhanced to conformal semisimple abelian; but the conclusions are unaltered: there is no way to unify the spacetime symmetries and the internal symmetries in a nontrivial way.
A wise person once said that inside every no-go theorem there is a yesgo theorem waiting to come out,1 and the ColemanMandula theorem is no exception. The trick consists, not in trying to relax some of the assumptions on the S-matrix of the eld theory, but in redening the very notion of symmetry to encompass Lie superalgebras. In a classic paper Haag, Lopuzaski n and Sohnius re-examined the result of Coleman and Mandula in this new light and found the most general Lie superalgebra of symmetries of an Smatrix. The ColemanMandula theorem applies to the bosonic sector of the Lie superalgebra, so this is given again by Poincar reductive. In e terms of representations of SL(2, C), these generators transform according 1 to the (0, 0), ( 2 , 1 ), (0, 1) and (1, 0) representations. The singlets are the 2 internal symmetry generators which we will denote collectively by B . The ( 1 , 1 ) generators correspond to the translations P , and the (1, 0) and (0, 1) 2 2 generators are the Lorentz generators: L and L . The novelty lies in the fermionic sector, which is generated by spinorial charges Q I in the ( 1 , 0) representation of SL(2, C) and their hermitian ad2 joints QI = (Q I ) in the (0, 1 ). Here I is a label running from 1 to some 2 positive integer N . The Lie superalgebra generated by these objects is called
1
and a wise guy said that we should call it a go-go theorem
42
[email protected]
the N -extended super-Poincar algebra. The important Lie brackets are given e by [B , Q I ] = b I J Q J [B , QI ] = I J QJ b I ] = 0 [P , Q I ] = 0 [P , Q (2.1) {Q I , Q J } = 2 ZIJ {QI , QJ } = 2 Z IJ {Q I , QJ } = 2I J P [ZIJ , anything] = 0 where ZIJ = zIJ m Bm , Z IJ = (ZIJ ) and the coecients b obey: b I K zKJ m + b J K zIK m = 0 .
I J
and zIJ m must (2.2)
This last condition is nontrivial and constraints the structure of that part of the internal symmetry group which acts nontrivially on the spinorial charges of the supersymmetry algebra, what we will call the internal automorphism group of the supersymmetry algebra. In the absence of central charges, the internal automorphism group of the supersymmetry algebra is U (N ), but in the presence of the central charges, it gets restricted generically to U Sp(N ), since condition (2.2) can be interpreted as the invariance under the internal automorphism group of each of the antisymmetric forms zIJ m , for each xed value of m. Notice that ZIJ = ZJI , whence central charges requires N 2.
The above Lie superalgebra is the most general symmetry of a local relativistic S-matrix in a theory describing point-particles. In the presence of extended objects: strings or, more generally, p-branes, the supersymmetry algebra receives extra terms involving topological conserved charges. These charges are no longer central since they fail to commute with the Lorentz generators; nevertheless they still commute with the spinorial charges and with the momentum generators. We will see an example of this later on when we discuss the six-dimensional N =1 supersymmetry algebra.
It is sometimes convenient, especially when considering supersymmetry algebras in dimensions other than 4, where there is no analogue to the isomorphism Spin(1, 3) SL(2, C), to work with 4-spinors. We can assemble = the spinorial changes Q I and QI into a Majorana spinor: QI = (Q I , Q I )t . The Dirac (=Majorana) conjugate is given by QI = (Q I , QI ), and the relevant bit of the supersymmetry algebra is now given by {QI , QJ } = 2IJ P + 2i(Im ZIJ + 5 Re ZIJ ) , where our conventions are such that = and = (1, ). 0 0 , (2.4) (2.3)
43
[email protected]
2.2
Unitary representations of the supersymmetry algebra
The construction of unitary representations of the super-Poincar algebra e can be thought of as a mild extension of the construction of unitary representations of the Poincar algebra. Because the Lorentz group is simple but e noncompact, any nontrivial unitary representation is innite-dimensional. The irreducible unitary representations are simply given by classical elds in Minkowski space subject to their equations of motion. Indeed the Klein Gordon and Dirac equations, among others, can be understood as irreducibility constraints on the elds. The method of construction for the Poincar e algebra is originally due to Wigner and was greatly generalised by Mackey. The method consists of inducing the representation from a nite-dimensional unitary representation of some compact subgroup. Let us review this briey.
2.2.1
Wigners method and the little group
The Poincar algebra has two casimir operators: P 2 and W 2 , where W = e 1 P L is the PauliLubansky vector. By Schurs lemma, on an irre2 ducible representation they must both act as multiplication by scalars. Lets focus on P 2 . On an irreducible representation P 2 = M 2 , where M is the rest-mass of the particle described by the representation. With our choice of metric, physical masses are real, whence M 2 0. We can thus distinguish two kinds of representations: massless for which M 2 = 0 and massive for which M 2 > 0. Wigners method starts by choosing a nonzero momentum k on the massshell: k 2 = M 2 . That is, this is a character (that is, a one-dimensional representation) of the translation subalgebra generated by the P . We let Gk denote the subgroup of the Lorentz group (or rather of its double-cover SL(2, C)) which leaves k invariant. Gk is known as the little group. Wigners method, which we will not describe in any more detail than this, consists in inducing a representation of the Poincar group from a nite-dimensional e unitary representation of the little group. This is done by boosting the representation to elds on the mass shell and then Fourier transforming to yield elds on Minkowski space subject to their equations of motion. In extending this method to the super-Poincar algebra all that happens e is that now the Lie algebra of the little group gets extended by the spinorial supersymmetry charges, since these commute with P and hence stabilise the chosen 4-vector. We will need to know about the structure of the little groups before 44
[email protected]
introducing supersymmetry. The little group happens to be dierent for massive and for massless representations, as the next exercise asks you to show.
Exercise 2.1 (The little groups for positive-energy particles) Let k be a 4-vector obeying k0 > 0, k 2 = M 2 0. Prove that the little group of k is isomorphic to: SU (2), for M 2 > 0; E2 , for M 2 = 0, where E2 SO(2) R2 , is the two-dimensional euclidean group and E2 = = Spin(2) R2 is its double cover. (Hint: argue that two momenta k which are Lorentz-related have isomorphic little groups. Then choose a convenient k in each case, examine the action of SL(2, C) on the bispinor k , and identify those M SL(2, C) for which M kM = k.)
The reason why we have restricted ourselves to positive-energy representations in this exercise, is that unitary representations of the supersymmetry algebra have non-negative energy. Indeed, for an arbitrary momentum k , the supersymmetry algebra becomes {Q I , QJ } = 2I J k0 + k3 k1 ik2 k1 + ik2 k0 k3 .
Therefore the energy k0 of any state |k with momentum k can be written as follows (for a xed but otherwise arbitrary I) k0 |k
2
= k| k0 |k = =
1 2 1 2
k| {Q1 I , QI } |k + 1 Q1 I |k
2
1 2
k| {Q2 I , QI } |k 2
2
1 2
Q2 I |k
1 2
(Q1 I ) |k
1 2
(Q2 I ) |k
whence k0 is positive, unless |k is annihilated by all the supersymmetry charges.
2.2.2
Massless representations
We start by considering massless representations. As shown in Exercise 2.1, the little group for the momentum k of a massless particle is noncompact. Therefore its nite-dimensional unitary representations must all come from its maximal compact subgroup Spin(2) and be trivial on the translation subgroup R2 . The unitary representations of Spin(2) are one-dimensional and indexed by a number 1 Z called the helicity. For CPT-invariance of 2 45
[email protected]
the spectrum, it may be necessary to include both helicities , but clearly all this does is double the states and we will not mention this again except to point out that some supersymmetry multiplets are CPT-self-conjugate. Lets choose k = (E, 0, 0, E), with E > 0. Then k = 2E 0 0 0 1 0 0 0
and the supersymmetry algebra becomes {Q I , QJ } = 4E I J .
In particular this means that {Q2 I , QJ } = 0. Because QJ = (Q2 J ) , it 2 2 follows that in a unitary representation Q2 I = 0 for all I. Indeed for any state | , 0 = | {Q2 I , (Q2 I ) } | = Q2 I |
2
+ (Q2 I ) |
Plugging this back into the supersymmetry algebra (2.1) we see that ZIJ = 1 {Q1 I , Q2 J } = 0, so that there are no central charges for massless represen2 tations. Let us now introduce qI (1/2 E) Q1 I , in terms of which the supersymmetry algebra becomes
{qI , qJ } = IJ {qI , qJ } = {qI , qJ } = 0 .
We immediately recognise this is as a Cliord algebra corresponding to a 2N dimensional pseudo-euclidean space with signature (N, N ). The irreducible representations of such Cliord algebras are well-known. We simply start with a Cliord vacuum | satisfying qI | = 0 for all I = 1, . . . , N ,
and we act repeatedly with the qI . Since {qI , qJ } = 0, we obtain a 2N dimensional representation spanned by the vectors: qI1 qI2 qIp | , where 1 I1 < I2 < < Ip N , and p = 0, . . . , N . The Cliord vacuum actually carries quantum numbers corresponding to the momentum k and also to the helicity: | = |k, . It may also contain quantum numbers corresponding to the internal symmetry generators B , but we ignore them in what follows.
Exercise 2.2 (Helicity content of massless multiplets) Paying close attention to the helicity of the supersymmetry charges, prove that Q1 I raises the helicity by 1 , whereas Q2 I lowers it by the same amount. Deduce 2 that the massless supersymmetry multiplet of helicity contains the following states:
46
[email protected]
States |k,
qI |k, qJ qI |k, qI1 qI2 qIp |k, q1 q2 qN |k,
Helicity + 1 2 +1 . . . + p/2 . . . + N/2
Number 1 N
N 2
. . . . . .
. . .
N p
. . . 1
Particularly interesting cases are the CPT-self-conjugate massless multiplets. First notice that CPT-self-conjugate multiplets can only exist for 1 N even. For N =2 we have the helicity = 2 multiplet, whose spectrum consists of Helicity Number -1/2 1 0 2 1/2 . 1
Then we have the N =4 gauge multiplet which has =1 and whose spectrum is given by: Helicity Number -1 1 -1/2 4 0 6 1/2 4 1 . 1
Pure (that is, without matter) N =4 supersymmetric YangMills in fourdimensions consists of several of these multipletsone for each generator of the gauge algebra. Finally, the third interesting case is the N =8 supergravity multiplet with = 2 and spectrum given by: Helicity Number -2 1 -3/2 8 -1 28 -1/2 56 0 70 1/2 56 1 28 3/2 8 2 . 1
2.2.3
Massive representations
We now consider massive representations. As shown in Exercise 2.1, the little group for the momentum k of a massive particle is SU (2). Its nitedimensional irreducible unitary representations are well-known: they are indexed by the spin s, where 2s is a non-negative integer, and have dimension 2s + 1. A massive particle can always be boosted to its rest frame, so that we can choose a momentum k = (M, 0, 0, 0). Then k = M 0 0 M 47
[email protected]
and the supersymmetry algebra becomes {Q I , QJ } = 2M I J 1 . No central charges In the absence of central charges, {Q I , Q J } = 0. Thus we can introduce q I (1/ 2M )Q,I , in terms of which the supersymmetry algebra is again a Cliord algebra:
{q I , q J } = IJ {q I , q J } = {q I , q J } = 0 ;
(2.5)
but where now the underlying pseudo-euclidean space is 4N -dimensional with signature (2N, 2N ). The unique irreducible representation of such a Cliord algebra is now 22N -dimensional and it is built just as before from a Cliord vacuum by acting successively with the q I . However unlike the case of massless representations, the Cliord vacuum is now degenerate since it carries spin: for spin s the Cliord vacuum is really a (2s + 1)-dimensional SU (2) multiplet. Notice that for xed I, q I transforms as a SU (2)-doublet of spin 1 . This must be taken into account when 2 determining the spin content of the states in the supersymmetry multiplet. Instead of simply adding the helicities like in the massless case, now we must use the ClebschGordon series to add the spins.
Exercise 2.3 (Highest spin in the multiplet) Prove that the highest spin in the multiplet will be carried by states of the form q1 1 q1 2 q1 N acting on the Cliord vacuum, and that their spin is s + N/2.
For example, if N = 1 and s = 0, then we nd the following spectrum: |k, 0 with spin 0, (q1 |k, 0 , q2 |k, 0 ) with spin 1/2 and q1 q2 |k, 0 which has spin 0 too. The supersymmetric eld theory describing this multiplet consists of a scalar eld, a pseudo-scalar eld, and a Majorana fermion: it is the celebrated WessZumino model and the multiplet is known as the massive WessZumino multiplet. Another example that will be important to us is the N =2 multiplets with spins s=0 and s=1/2, which we leave as an exercise.
Exercise 2.4 (Massive N =2 multiplets with s=0 and s=1/2) Work out the spin content of the massive N =2 multiplets without central charges
1 and with spins s=0 and s=1/2. Show that for s=0 the spin content is (05 , 2 , 1) 6 4 in the obvious notation, and for s=1/2 it is given by (3/2, 14 , 1 , 0 ). 2 4
48
[email protected]
Adding central charges Adding central charges changes the nature of the supersymmetry algebra, which now becomes {Q I , QJ } = 2M I J 1 Because ZIJ is antisymmetric, we automorphism of the rst of the takes a standard form: 0 z1 z1 0 0 z2 ZIJ = {Q I , Q J } = 2
ZIJ
can rotate the Q I unitarilywhich is an above bracketsin such a way that ZIJ z2 0
..
. .. . 0 zN/2 zN/2 0
where the zi can be chosen to be real and non-negative. To simplify the discussion we have assumed that N is even, but one should keep in mind that if N is odd, there will of course be a zero 1 1 block in the above normal form for ZIJ . Let us break up the index I into a pair (A, i) where A = 1, 2 and i = 1, . . . , N/2. In terms of these indices the supersymmetry algebra can be rewritten as {Q Ai , Q Bj } = 2M ij AB Dene the following linear combinations
S i 1 2
{Q Ai , Q Bj } = 2
AB ij zi
Q 1i
Q2i
where we have raised the spinor index in the second term in order to preserve covariance under the little group SU (2). In terms of these generators, the algebra is now: {S i , (S j ) } = ij (M zi ) with all other brackets being zero. Notice that acting on any state | , (M zi ) |
2
= |(M zi )|
= |{S1 i , (S1 i ) }| = S1 i | 2 + (S1 i ) | 2
49
[email protected]
from where it follows that M zi 0 for all i, or M |zi | for all i = 1, . . . , N/2 , (2.6)
which is reminiscent of the Bogomolnyi bound (1.27). Notice that this bound is an unavoidable consequence of having a unitary representation of the supersymmetry algebra. Therefore provided that supersymmetry is not broken quantum-mechanically, the bound will be maintained. Suppose that M > zi for all i. Then we can dene q i (1/ M zi )S i , in terms of which the supersymmetry algebra is again given by equation (2.5) once we recombine the indices (, i) into I. Therefore we are back in the case of massive representations without central charges, at least as far as the dimension of the representations is concerned. Suppose instead that some of the zi saturate the bound (2.6): zi = M for i = 1, . . . , q N/2. Then a similar argument as in the discussion of the massless representations allows us to conclude that the 2q generators S i for i = 1, . . . , q act trivially and can be taken to be zero. The remaining 2N 2q generators obey a Cliord algebra whose unique irreducible representation has dimension 22N 2q . Notice that the smallest representation occurs when all central charges saturate the bound (2.6), in which case all the S i = 0 and we are left only with 2N states, just as in the case of a massless multiplet. These massive multiplets are known as short multiplets. For example, in N =2 there is only one z = z1 . If z < M the massive multiplet contains 24 = 16 states, whereas if z = M the short multiplet only contains 22 = 4 states. For N =4, there are two zi . If both zi < M , then the massive multiplet has 28 = 256 states, whereas if both zi = M , then the short multiplet contains only 24 = 16 states. Half-way we nd the case when exactly one of the zi = M , in which case the dimension of the multiplet is 26 = 64. Strictly speaking we shouldnt call these numbers the dimension of the multiplet, but rather the degeneracy, since it may be that the Cliord vacuum is degenerate, in which case the dimension of the supersymmetry multiplet is the product of what weve been calling the dimension of the multiplet and that of the Cliord vacuum. Let us work out some examples. 1 We rst work out the case of N =2 and spins s=0 and s= 2 in the following exercise.
Exercise 2.5 (Short N =2 multiplets with s=0 and s=1/2) Prove that the spin contents of the short multiplet with s=0 is ( 1 , 02 ) and that of 2 the short multiplet with s=1/2 is (1, 1 , 0). Compare with the results of Exercise 2 2.4, which are the spin contents when the central charge does not saturate the bound. We will see that the s=0 multiplet contains the BPS-monopole, whereas the s=1/2 multiplet contains the massive vector bosons.
2
50
[email protected]
Next we take a look at the short N =4 multiplets with s=0. These will be the important ones when we discuss N =4 supersymmetric YangMills.
Exercise 2.6 (Short N =4 multiplets with s=0) 4 Prove that the spin content of the N =4 short multiplet with s=0 is (1, 1 , 05 ), 2 which totals the expected 16 states. As we will see later, this will be the multiplet containing both the BPS-monopole and the massive vector boson.
This dierence in the dimension of representations for which the bound (2.6) is saturated is responsible for the fact that if a multiplet saturates the bound classically, it will continue to do so when perturbative quantum corrections are taken into account. This is because perturbative quantum corrections do not alter the number of degrees of freedom, hence a short multiplet (that is, one which saturates the bound) cannot all of a sudden undergo the explosion in size required to obey the bound strictly.
2.3
N =2 Supersymmetric Yang-Mills
The supersymmetric bound (2.6) for massive representations with central charges may seem a little abstract, but it comes to life in particular eld theoretical models, where we can explicitly calculate the central charges in terms of the eld variables. We will see this rst of all in pure N =2 supersymmetric YangMills, which embeds the bosonic part of the GeorgiGlashow model. This result is due to Witten and Olive [WO78]. We could simply write the action down and compute the supersymmetry algebra directly as was done in [WO78], but it is much more instructive to derive it by dimensional reduction from the N =1 supersymmetric YangMills action in six dimensions. This derivation of N =2 supersymmetric YangMills by dimensional reduction was rst done in [DHdV78], and the six-dimensional computation of the central charges was rst done in [Oli79]. That there should be a N =1 supersymmetric YangMills theory in six dimensions is not obvious: unlike its nonsupersymmetric counterpart, supersymmetric YangMills theories only exist in a certain number of dimensions. Of course one can always write down the YangMills action in any dimension and then couple it to fermions, but supersymmetry requires a delicate balance between the bosonic and fermionic degrees of freedom. A gauge eld in d dimensions has d 2 physical degrees of freedom corresponding to the transverse polarisations. The number of degrees of freedom of a fermion eld depends on what kind fermion it is, but it always a power of 2. An unconstrained Dirac spinor in d dimensions has 2d/2 or 2(d1)/2 real degrees of freedom, for d even or odd respectively: a Dirac spinor has 2d/2 or 2(d1)/2 51
[email protected]
complex components but the Dirac equation cuts this number in half. In even dimensions, one can further restrict the spinor by imposing that it be chiral or Weyl. This cuts the number of degrees of freedom by two. Alternatively, in some dimensions (depending on the signature of the metric) one can impose a reality or Majorana condition which also halves the number of degrees of freedom. For a lorentzian metric of signature (1, d 1), Majorana spinors exist for d 1, 2, 3, 4 mod 8. When d 2 mod 8 one can in fact impose that a spinor be both Majorana and Weyl, cutting the number of degrees of freedom in four. The next exercise asks you to determine in which dimensions can supersymmetric YangMills theory exist based on the balance between bosonic and fermionic degrees of freedom.
Exercise 2.7 (N =1 supersymmetric YangMills) Verify via a counting of degrees of freedom that N =1 supersymmetric YangMills can exist only in the following dimensions and with the following types of spinors: d 3 4 6 10 Spinor Majorana Majorana or Weyl Weyl MajoranaWeyl
It is a curious fact that these are precisely the dimensions in which the classical superstring exists. Unlike superstring theory, in which only the ten-dimensional theory survives quantisation, it turns out that supersymmetric YangMills theory exists in each of these dimensions. Although we are mostly concerned with four-dimensional eld theories in these notes, the six-dimensional and ten-dimensional theories are useful tools since upon dimensional reduction to four dimensions they yield N =2 and N =4 supersymmetric YangMills, respectively.
2.3.1
N =1 d=6 supersymmetric YangMills
We start by setting some conventions. We will let uppercase Latin indices from the beginning of the alphabet A, B, . . . take the values 0, 1, 2, 3, 5, 6. Our metric AB is mostly minus; that is, with signature (1, 5). We choose the following explicit realisation of the Dirac matrices: = 0 0 5 = 0 5 5 0 6 = 0 1 1 0
52
[email protected]
where = 0, 1, 2, 3, and where are dened in (2.4) and 5 = 0 1 2 3 . The A obey the Cliord algebra {A , B } = 2 AB 1 . Weyl spinors are dened relative to 7 , which is dened by 7 = 0 1 6 = 1 0 0 1 .
We now write down the action for N =1 supersymmetric YangMills. We will take the gauge group to be SO(3) for deniteness, but it should be clear that the formalism is general. As before we will identify the Lie algebra so(3) with R3 but we will now drop the arrows on the vectors to unclutter the notation, hoping it causes no confusion. The lagrangian density is given by i L = 1 GAB GAB + 2 A D A , (2.7) 4 where GAB = A WB B WA e WA WB DA = A e WA , and where is a complex Weyl spinor obeying 7 = . The Dirac conjugate spinor is dened by = 0 , and obeys 7 = . Finally we have used the convenient shorthand D A to mean A D A = A DA DA A . The action dened by (2.7) is manifestly gauge invariant, but it is also invariant under supersymmetry. Let and be two constant anticommuting Weyl spinors of the same chirality as . Let us dene the following transformations: WA = iA = 0 = 1 GAB AB
2
WA = iA = 1 GAB AB = 0
2
(2.8)
where AB = 1 (A B B A ). We should remark that there is only one 2 supersymmetry in our theory: and are chiral. That is, there is only one spinorial charge Q, in terms of which the transformations and dened above can be understood as follows: = [Q, ] and = [Q, ] , 53
[email protected]
for any eld . Notice that it follows from this that the action of can be = ( ) (apart from the obvious change deduced from that of as follows: of to , of course). Keep in mind that we have chosen the Lie algebra structure constants to be real, whence the generators are antihermitian. We claim that L is invariant under and above up to a divergence. In order to derive the supersymmetry current, we will actually take and to depend on the position and simply vary the lagrangian density. We expect a total divergence plus a term with the current multiplying the derivative of the parameter. The calculations will take us until the end of the section and are contained in the following set of exercises.
Exercise 2.8 (Supersymmetry variation of L) Prove rst of all that for any derivation , GAB = DA WB DB WA , and conclude that the variation of the bosonic part of the action Lb is given by Lb = iGAB DA (B ) and Lb = iGAB DA (B ) .
Next we tackle the fermions. Prove the following identities: (DA ) = ie (A ) 1 DA = 2 DA GBC BC ie (A ) and (DA ) = 1 DA GBC BC + ie A 2 DA = ie A , and conclude that the variation of the fermionic part of the action Lf is given by
i i Lf = 4 DA GBC BC A 4 GBC BC A DA + e A A
and i i Lf = 4 A BC DA (GBC ) DA A BC GBC e A A . 4
Supersymmetry invariance demands, in particular, that the fermion trilinear terms in Lf should cancel. This requires a Fierz rearrangement, and this is as good a time as any to discuss this useful technique. Writing explicitly the Lie algebra indices on the fermions, the trilinear terms in Lf become c e abc A a A b . (2.9) 54
[email protected]
c Let us focus on the expression a . This is a bispinor. Since spinors in six dimensions have 8 components, bispinors form a 64-dimensional vector space spanned by the antisymmetrised products of -matrices:
1, A , AB , ABC , ABCD , ABCDE and ABCDEF ,

or equivalently
1, A , AB , ABC , AB 7 , A 7 and 7 .
(Notice that antisymmetrisation is dened by A1 A2 Ap = [A1 A2 Ap ] = 1 sign A(1) A(2) A(p) , p! S
p
so that it has strength one.) We will let {M } denote collectively these matrices. The above basis is orthogonal relative to the inner product dened by the trace: tr M M = c , which allows us to expand c a =
ac
b ac M ,
and to compute the coecients b simply by taking traces. Remembering that a are anticommuting, we nd 1 c b ac = M a . (2.10) c
Exercise 2.9 (A Fierz rearrangement) Using the above formula and computing the relevant traces, prove that
1 c c a = 8 A a A (1 + 7 ) 1 48
c ABC a ABC .
(Hint: use the fact that 7 = to discard from the start many of the terms in the general Fierz expansion.)
We now use this Fierz rearrangement to rewrite the trilinear term (2.9) as follows: 1e 8
abc
A B (1 + 7 )A b
c B a
1 e 48 abc
A BCD A b
c BCD a ,
which using that is Weyl, can be simplied to 1e 4

abc
A B A b
c B a
1 e 48 abc
A BCD A b
c BCD a .
55
[email protected] Exercise 2.10 (Some -matrix identities) Prove the following two identities: A B A = 4 B and
A BCD A = 0 ,
(2.11)
and deduce that the trilinear terms cancel exactly. The above identities are in fact the minor miracle that makes supersymmetric YangMills possible in six dimensions.
Up to a divergence, the remaining terms in the supersymmetric variation of the lagrangian density L are then: i L = iGAB DA (B ) GBC BC A DA 2 and i L = iGAB DA B GBC DA A BC . 2
Exercise 2.11 (... and the proof of supersymmetry invariance) Prove the following identity between Dirac matrices AB C = ABC + BC A AC B , and use it to rewrite the supersymmetric variations of L as i i L = A GBC BC A and L = A BC GBC A , (2.12) 2 2 again up to divergences and where we have used the Bianchi identity in the form ABC DA GBC = 0. This proves the invariance of L under the supersymmetry transformations (2.8).
From (2.12) we can read the expression for the supersymmetry currents: i i J A = GBC BC A and J A = A BC GBC . 2 2 As usual the spinorial supersymmetry charge is the space integral of the zero component of the current. Provided we already knew that L is supersymmetric, there is a more economical way to derive the expression of the supercurrent. This uses the fact that the supercurrent is part of a supersymmetry multiplet.
Exercise 2.12 (The supersymmetry multiplet) Prove that the lagrangian density (2.7) is invariant under the transformation exp(i) , exp(i) , and that the corresponding Noether current is given A . Prove that by jA = jA = iJA and jA = iJA .
56
[email protected]
The supersymmetry multiplet also contains the energy-momentum tensor, alone or in combination with other topological currents that may ap pear in the right hand side of {Q, Q} in the supersymmetry algebra. We will use this later to compute the supersymmetry algebra corresponding to six-dimensional supersymmetric YangMills. But rst we perform the dimensional reduction to four dimensions.
2.3.2
From N =1 in d=6 to N =2 in d=4
Let us single out two of the coordinates (x5 , x6 ) in six dimensions and assume that none of our elds depend on them: 5 6 0. This breaks SO(1, 5) Lorentz invariance down to SO(1, 3) SO(2). Let us therefore decompose our six-dimensional elds in a way that reects this. In fact, we will at rst ignore the SO(2) invariance and focus only on the behaviour of the components of the six-dimensional elds under the action of SO(1, 3). The gauge eld WA breaks up into a vector W , a pseudo-scalar P = W5 and a scalar S = W6 . In terms of these elds, the eld-strength breaks up as G , G5 = D P, G6 = D S and G56 = e SP. Meanwhile, the Weyl spinor breaks up as = , where is an unconstrained (complex) Dirac spinor. The 0 covariant derivative of the spinor then breaks up as (D , eP, eS). The lagrangian density now becomes L = Lb + Lf where
1 1 Lb = 4 G G + 2 D P D P + 1 D S D S 1 e2 P S 2 2 2
and Lf = i D + ie 5 P + ie S , (2.13)
where we see that P is indeed as pseudo-scalar as claimed. L is the lagrangian density of N =2 supersymmetric YangMills theory in four dimensions. The supersymmetry parameter , which in the six-dimensional theory is a Weyl spinor, becomes upon dimensional reduction a Dirac spinor. But in four dimensions the supersymmetry parameters are Majorana, hence this gives rise to N =2 supersymmetry. One can see this explicitly by breaking up the supersymmetry parameter into its Majorana components: simply choose a Majorana representation and split it into its real and imaginary parts. Each of these spinors is Majorana and generates one supersymmetry. Let us rst do this with . The next exercise shows the resulting fermion action in a Majorana basis.
57
[email protected] Exercise 2.13 (L in a Majorana basis) In a Majorana basis, let us split as follows:
1 = ( 1 i 2 ) . 2 Prove that relative to , = 1, 2, the fermionic part Lf of the lagrangian density becomes (up to a total derivative) i i Lf = 1 D 1 + 2 D 2 + e 1 5 P 2 + e 1 S 2 . (2.14) 2 2 (Hint: you may nd useful the following identities for anticommuting Majorana spinors in four dimensions: = 5 = 5 = , (2.15)
which you are encouraged to prove!)
We can do the same with the supersymmetry transformations (2.8), as the next exercise asks you to show.
Exercise 2.14 (Explicit N =2 supersymmetry transformations) Show that in a Majorana basis, the dimensional reduction of the supersymmetry transformations (2.8) becomes: 1 W = i 1 + 2 1 P = i 5 1 + 5 2 1 S = i 1 + 2 1 1 = D (S + 2 1 = 0
2 W = 1 + i 2 2 P = 5 1 + i 5 2 1 2 = 0 1 2 G
2 S = 1 + i 2 1 P5 ) + 2 e(S P)5 + 1 G 2 2 2 = D (S + P5 ) + 1 e(S P)5 2
The SO(2) invariance of (2.14) can be made manifest by rewriting Lf explicitly in terms of the SO(2) invariant tensors and . In fact, using the identities (2.15), one can rewrite (2.14) as: i e Lf = D + 2 2
(5 P + S) .
The SO(2) transformation properties of the four-dimensional elds can be succinctly written as follows: S + iP ei (S + iP) e5 /2 W W . 58 (2.16)
[email protected]
Exercise 2.15 (The SO(2) Noether current) Prove that the Noether current associated with the SO(2) transformations (2.16) is given by 5 i j = P D S S D P + 2 5 . Notice that this current contains the axial current, hence the notation. Problem: Is it anomalous in this theory?
2.3.3
Higgsed N =2 supersymmetric YangMills
The hamiltonian density corresponding to the N =2 supersymmetric Yang Mills theory dened by (2.13) is given by H = Hb + Hf . We focus on the bosonic part: Hb =
1 2
Ei
1 2
D0 S
+ 1 D0 P 2 2 + 1 Bi 2 + 1 Di S 2 2
1 2
Di P
+ 1 e2 P S 2
Demanding that the energy of a given eld conguration be nite doesnt necessarily imply that P and S acquire non-zero vacuum expectation values for the term P S 2 is already zero provided that P S = 0, which for so(3) means that they be parallel. Indeed, except for that term and the extra eld, Hb is nothing but the energy density (1.13) of (the bosonic part of) the GeorgiGlashow model in the limit of vanishing potential. We could add a 2 potential term ( P 2 + S 2 a2 ) to the lagrangian (2.13) to force S and P to acquire a nonzero vacuum expectation value, but such a term would break supersymmetry. Nevertheless we could then take the limit 0 while keeping the nonzero vacuum expectation values of S and P. This restores the supersymmetry provided that S and P are parallel, which would be the supersymmetric version of the PrasadSommereld limit. Since the potential depends only on the SO(2) invariant combination P 2 + S 2 , SO(2) is preserved and we could use this symmetry to choose P = 0 and S = a, where a is a xed vector with a 2 = a2 .
Exercise 2.16 (The perturbative spectrum of the model) We can analyse the perturbative spectrum of the model around such a vacuum in exactly the same way as we did in Exercise 1.4. Choosing for example the unitary gauge a = ae3 , show that there are now two massive multiplets ( , W , P ) of mass MW = ae , and a massless gauge multiplet corresponding to the unbroken 3 U (1): ( 3 , W , S 3 , P 3 ). Prove that the massless gauge multiplet is actually made out of two massless multiplets with helicities = 1 and = 0.
Now watch carefully: something curious has happened. From the analysis in section 2.2.3, we know that the generic massive representations of N =2 59
[email protected]
supersymmetry are sixteen-fold degenerate, and from Exercise 2.3 we know that they must have a state with spin 3/2. Yet the massive multiplets which have arisen out of higgsing the model contain maximum spin 1 and are only four-fold degenerate. This is only possible if the N =2 supersymmetry algebra in this model has central charges and these charges saturate the bound! Indeed, the only way to reconcile the above spectrum with the structure of massive representations of the N =2 supersymmetry algebra studied in section 2.2.3 is if it corresponds to the short multiplet with spin s=1/2 studied in Exercise 2.5. In the next section we will actually compute the supersymmetry algebra for this model and we will see that the central charges are precisely the electric and magnetic charges relative to the unbroken U (1). But before doing this let us check that the BPS-monopole is actually a solution of N =2 supersymmetric YangMills.
2.3.4
N =2 avatar of the BPS-monopole
We now show that this N =2 supersymmetric YangMills theory admits BPSmonopole solutions. We look for static solutions, so we put W0 = 0. Since the fermion equations of motion are linear, we can always put = 0 at the start. Applying supersymmetry transformations to such a solution, we will be able to generate solutions with nonzero fermions. Similarly, using the SO(2) invariance we can look for a solution with P = 0, and then obtain solutions with nonzero P by acting with SO(2). Having made these choices, we are left with Wi and S, which is precisely the spectrum of the bosonic part of the GeorgiGlashow model provided we identify S and . Furthermore, not just the spectrum, but also the lagrangian density agrees, with potential set to zero, of course. Therefore the BPS-monopole given by (1.15) with H and K given in Exercise 1.12 is a solution of N =2 supersymmetric YangMills. If we now apply an SO(2) rotation to this solution, we nd the following BPS-monopole solution: = W0 = 0 ra a S = 2 H() er ra a P = 2 H() er rj a Wi = aij 2 (K() 1) er
(2.17)
where as before = aer, where H and K are the same functions in Exercise 1.12, and 2 + 2 = 1. Putting = 0 we recover the BPS-monopole and anti-monopole for = 1, respectivelya result rst obtained in [DHdV78]. 60
[email protected]
Since (2.17) is a solution of the eld equations of a supersymmetric theory, supersymmetry transformations map solutions to solutions. Hence starting with (2.17) we can try to generate solutions with nonzero fermions by performing a supersymmetry transformation. We will actually assume a more general solution than the one above.
Exercise 2.17 (Supersymmetric BPS-monopoles) Prove that any BPS-monopole, that is, any static solution (Wi , ) of the Bogomolnyi equation (1.28), can be thought of as an N =2 BPS-monopole by setting S = , P = and = 0, with 2 + 2 = 1.
We will then take one such N =2 BPS-monopole as our starting point and try to generate other solutions via supersymmetry transformations. The supersymmetry transformation laws on the four-dimensional elds can be read o from those given in (2.8) for the six-dimensional elds. Since we start with a background in which = 0, the bosonic elds are invariant under supersymmetry. The supersymmetry transformation law of the fermion is given by = 1 G D ( + 5 ) , 2 where is an unconstrained (complex) Dirac spinor. Because the solution is staticW0 = 0 and all elds are time-independentthe above can be rewritten as = 1 Gij ij + Di i ( + 5 ) . 2 For deniteness we will assume that (Wi , ) describe a BPS-monopole (as 1 opposed to an anti-monopole) so that Di = + 2 ijk Gjk . Then we can rewrite the above transformation law once more as = Dk
1 2 ijk ij
+ k ( + 5 )
(2.18)
Exercise 2.18 (More -matrix identities) Prove the following identity:

1 2 ijk ij
= 0 5 k .
(2.19)
Exercise 2.19 (Some euclidean -matrices) Let i 0 i for i = 1, 2, 3, and let 4 = 0 ( + 5 ). Prove that they generate a euclidean Cliord algebra. Dene 5 1 2 3 4 = 0 (5 ). Prove that 5 is hermitian and that 5 = 1. 2
In terms of these euclidean Cliord algebra, and using (2.19), we can rewrite (2.18) as = 5 k Dk (1 5 ) . 61
[email protected]
Notice that 1 (1 5 ) is a projector. If we denote = 1 ( 5 ), then 2 2 the supersymmetric variation of in a BPS-monopole background is given simply by = 2 5 k Dk . This means that if has negative chirality relative to 5 , then we dont gen erate new solutions, yet if has positive chirality, then we do. Equivalently, supersymmetry transformations with negative chirality parameters preserve the solution, whereas those with positive chirality parameters break it.
Exercise 2.20 (BPS-monopoles break one half of the supersymmetry) Prove that the (1)-eigenspaces of 5 have the same dimension. Conclude that 1 the projector 2 (1 5 ) projects out precisely one half of spinors.
As a corollary of the above exercise we see that supersymmetric BPSmonopoles break half the supersymmetries. Notice that the parameter , being an unconstrained Dirac spinor has 4 complex (or 8 real) components, whereas only has 2 complex (or 4 real) components. Hence we expect that the BPS-monopole belongs to a fourfold degenerate multiplet. From our study in section 2.2.3 of massive representations of the N =2 supersymmetry algebra, we know that those massive multiplets preserving half the supersymmetries are necessarily short, and from Exercise 2.5 we see that the k=1 BPS-monopole given by (2.17) generates a short multiple with spin s=0. This multiplet contains two particles of spin 0 and one of spin 1/2, yet none of spin 1. Therefore although as we will see in the next section, N =2 supersymmetry solves the rst of the problems with the MontonenOlive conjecture mentioned at the end of section 1.4.1, it still does not address the second problem satisfactorily. As we will see later, the solution of this problem requires N =4 supersymmetry.
2.3.5
The supersymmetry bound is the Bogomolnyi bound

M 2 (aq)2 (ag)2 0 ,
The Bogomolnyi bound (1.27) can be suggestively rewritten as
which is begging us to add two spatial dimensions to our spacetime and interpret the above inequality as the positivity of mass. As explained in section 2.2.1, the positivity of the mass is a consequence of unitarity and the supersymmetry algebra. Therefore it would make sense to look for a six-dimensional supersymmetric explanation of the Bogomolnyi bound. The explanation of [WO78] used the central charges in four-dimensional N =2 62
[email protected]
supersymmetric YangMills theory, and as we have seen this theory comes induced from six-dimensional N =1 supersymmetric YangMills via dimensional reduction. It would make sense therefore to look for a direct sixdimensional explanation. This was done to a large extent in [Oli79] and we will now review this. The above heuristics suggest that we think of the electric and magnetic charges as momenta in the two extra spatial dimensions. However it isnt hard to see that this interpretation is not quite correct. If one computes the energy-momentum tensor TAB of the six-dimensional supersymmetric Yang Mills theory, and from there the momenta PA = T0A , then the positivity of mass formula in six-dimensions:
2 2 M 2 P5 + P6 ,
(2.20)
where M 2 = P P is the four-dimensional mass, does not agree with the Bogomolnyi bound (1.27). In fact one nds that the magnetic charge does not appear. What is wrong then? Simply that we have assumed that it is P which appears in the right hand side of {Q, Q} in the supersymmetry algebra, when in fact it is P + Z , where Z can be interpreted as the topological charge due to the presence of a string-like source in six-dimensions. We now nd out what Z is by computing the supersymmetry algebra. We rst do this in six dimensions and then reduce down to four. The supersymmetry algebra in six dimensions We start by noticing that the space integral of j0 is equal to {Q, Q} , A , which we naturally leave as an exercise. whence it is enough to compute j
Exercise 2.21 (Supersymmetric variation of the supercurrent) Prove that jA = i JA = 1 GBC GEF BC A EF + i (C DB ) A BC . 4 The fermion bilinear term has to be Fierzed, but we will not be concerned with the fermions in what follows: we are interested in computing the momenta in classical congurations like the BPS-monopole, where the fermions have been set to zero. Of course, it would be a good exercise in -matrix algebra to compute the fermionic terms, not that there is little -matrix algebra to be done. In fact, prove that setting = 0, jA is given by jA = 2 1 8
BC BCAEF D G 1 GEF + GBC GCA BD + 4 GBC GBC AD D . (2.21) = ABCDEF F 7 (prove it!) and use the fact that
(Hint: use that ABCDE 7 = .)
63
[email protected]
We see that there are two very dierent tensors appearing in the right hand-side of jA : AB = 1 8
C ABCDEF
GCD GEF
1 4 CD
(2.22) (2.23)
TAB = GA GCB + G
GCD AB .
Notice that TAB is symmetric, whereas AB is antisymmetric. In fact, TAB is (the bosonic part of) the energy-momentum tensor of the six-dimensional theory.
Exercise 2.22 (The symmetric gauge-invariant energy momentum tensor) Prove that the energy-momentum tensor of the six-dimensional supersymmetric YangMills theory is given by i TAB + 2 (A D B) AB
i 2
C D C .
Prove that TAB is gauge-invariant and that it is conserved on-shell. (Hint: Vary Lb with respect to an innitesimal translation xA xA + A (x) and determine the associated Noether current, which after symmetrisation is the energy-momentum tensor, by denition.)
Dening PA to be the space integral of T0A and ZA the space integral of 0A , we see that the supersymmetry algebra becomes {Q, Q}: {Q, Q} = 2A (PA + ZA ) . Only the rst of these terms is to be interpreted as the momentum, the other term is associated with a topologically conserved current.
Exercise 2.23 (The topological current) Prove that AB is gauge invariant and that it is conserved o-shell, that is, without imposing the equations of motion. This means that it is a topological current. (Hint: show that AB = C ABC where ABC is totally antisymmetric, though not gauge invariant.)
The supersymmetry algebra in four dimensions It is now time to dimensionally reduce the supersymmetry algebra. The following exercise asks you to compute P5 , P6 , Z5 , and Z6 (with fermions put to zero) after dimensional reduction.
64
[email protected]
Exercise 2.24 (The momenta in the extra dimensions) Prove that T05 = Di P G0i e (P S) D0 S T06 = Di S G0i + e (P S) D0 P 05 = 05 =
1 2 ijk Gij Dk S 1 ijk Gij Dk P 2 ijk Di Gjk
Using the Bianchi identity follows:

1 05 = 2 i ( ijk Gjk
= 0, we can rewrite 05 and 06 as

ijk Gjk
S)
1 and 05 = 2 i (
P)
whereas using the equations of motion (for zero fermions) Di G0i + e P D0 P + e S D0 S = 0 , we can rewrite T05 and T06 as follows: T05 = i (G0i P) and T05 = i (G0i S) .
We see that all the densities are divergences, whence their space integrals only receive contribution from spatial innity: P5 + Z5 =
P G0i + S G0i
1 S 2 ijk 1 P 2 ijk
Gjk di Gjk di .
P6 + Z6 =
To interpret these integrals we can proceed in either of two ways. The fastest way is to use the SO(2) invariance of the theory to choose P = 0 and S 2 = a2 at spatial innity. Comparing with (1.24) and (1.25), we see that P5 +Z5 = ag and P6 + Z6 = aq. The same reasoning follows without having to use SO(2) invariance, as the next exercise shows.
Exercise 2.25 (The eective electromagnetic eld strength) Dene the following eld strength: F 1 (S G + P G ) . a (2.24)
Prove that in the Higgs vacuum it obeys Maxwells equations, and deduce that P5 + Z5 = ag and P6 + Z6 = aq where g and q are, respectively, the magnetic and electric charges of this electromagnetic eld. (Hint: Compare with Exercise 1.8).
65
[email protected]
To prove that (2.20) is the Bogomolnyi bound (1.27), we can proceed in two ways. We can exploit the SO(2) invariance of the supersymmetry algebra in order to set P = 0, and then notice that Z = 0. Using the fact that P is indeed the honest momentum of the theory, namely the space integral of T0 , and plugging the expressions for PA + ZA into (2.20), we nally arrive at the Bogomolnyi bound (1.27)! Alternatively we can deduce that Z = 0 without having to set P = 0. This is the purpose of the following exercise.
Exercise 2.26 (The space components of the topological charge) Prove that 0i is given by 0i = whence Zi is given by Zi =
ijk ijk j
(P Dk S) ,
(P Dk S) dj .
Prove that this vanishes for a nite-energy conguration. (Hint: Notice that for a solution of the Bogomolnyi equation S = , P = with 2 + 2 = 1, P Dk S = 1 k 2 , and that the derivative k is tangential 2 to due to the ijk . Since 2 = a2 on , its tangential derivative vanishes.)
Dene the following complex linear combinations of elds (cf. (1.33)): G = G + i G = S + iP , in terms of which the eective electromagnetic eld strength dened in (2.24), becomes F = 1 Re G a .
Under an innitesimal SO(2) transformation, = i, and because i G = G , we can write F = F . In other words, SO(2) transformations become innitesimal duality transformations in the eective electromagnetic theory. Problem: Are anomalies responsible for the breaking of this symmetry in the quantum theory?
2.4
N =4 Supersymmetric Yang-Mills
We saw in Exercise 2.7 that 10 is the largest dimension in which N =1 supersymmetric YangMills theory can exist and that for it to exist we must impose that the spinors be both Weyl and Majoranaconditions which, luckily for us, can be simultaneously satised in ten-dimensional Minkowski space. In 66
[email protected]
this section we will prove that this theory exists and that upon dimensional reduction to four dimensions yields a gauge theory with N =4 supersymmetry. This theory admits Higgs phenomena and has room to embed the BPS-monopole and indeed, any solution of the Bogomolnyi equation, just as in the N =2 theory discussed in the previous section. We will see that both the massive fundamental states (e.g., vector bosons) and the solitonic states (e.g., BPS-monopoles) belong to isomorphic (short) multiplets saturating the supersymmetry mass bound which once again will be shown to agree with the Bogomolnyi bound for dyons.
2.4.1
N =1 d=10 supersymmetric YangMills
We start by setting up some conventions. We will let indices A, B, . . . from the start of the Latin alphabet run from 0 to 9. (No confusion should arise from the fact that in the previous section the very same indices only reached 6.) The metric AB is mostly minus and the 32 32 matrices {A } obey the Cliord algebra {A , B } = 2 AB 1. We let 11 0 1 9 ; it obeys 2 = 1. We shall also need the charge conjugation matrix C, which obeys 11 Ct = C and (CA )t = CA , from where it follows that (A )t = CA C1 . N =1 supersymmetric YangMills theory is dened by the following lagrangian density:
i L = 1 GAB GAB + 2 A DA , 4
(2.25)
where GAB = A WB B WA e WA WB , DA = A e WA , and where is a complex MajoranaWeyl spinor obeying 11 = (Weyl) and = 0 = t C (Majorana).
Exercise 2.27 (The Majorana condition) Prove that the Majorana condition above relates and its complex conjugate : = C0 , whence it can be considered a reality condition on the spinor.
The action dened above is clearly gauge invariant. We claim that it is also invariant under the following supersymmetry: WA = i A = i A = 1 GAB AB 2 = 1 AB GAB , 2 67 (2.26)
[email protected]
where is a constant anticommuting MajoranaWeyl spinor of the same chirality as , and where the second and third relations above imply each other. The proof that the action is invariant under the supersymmetry transformations (2.26) is very similar to the analogous statement for the sixdimensional theory, so we will not be as verbose. We start by varying the action with respect to (2.26). We dont take to be constant in order to be able to read o the form of the supersymmetry current from the variation of the lagrangian density. We will have proven invariance if we can show that up to a divergence, the variation of (2.25) is proportional to the derivative of the coecient being the supersymmetry current. Varying the lagrangian density we encounter two kinds of terms: terms linear in the fermions, and a term trilinear in the fermions and without derivatives, coming from the variation of the gauge eld inside the covariant derivative acting on the fermions. Before getting into the computation, it is useful to derive some properties of MajoranaWeyl fermions, which are left as an instructive exercise.
Exercise 2.28 (Properties of Majorana and Weyl fermions) Let and be anticommuting Majorana fermions in ten dimensions. Prove that A1 A2 Ak = ()k(k+1)/2 A1 A2 Ak . (2.27)
If, in addition, and are Weyl and of the same chirality, then prove that (even number of s) = 0 . (Hint: It may prove useful to rst prove the identity (CA1 A2 Ak )t = ()k(k+1)/2 CA1 A2 Ak , which will play a role also later on.) (2.28)
We now vary the lagrangian density.

Exercise 2.29 (Varying the lagrangian density) Prove that supersymmetric variation of the lagrangian density L is given, up to a divergence, by:
i L = 2 DC GAB AB C + iDA GAB B 1 i + 2 GAB C AB C + 2 e A (( A ) ) .
(Hint: Integrate by parts and use the identity (2.27) repeatedly.)
68
[email protected]
Using the Bianchi identity in the form ABC DC GAB = 0, it is easy to prove that the rst two terms in the above expression for L cancel out, leaving the trilinear terms and the term involving the supersymmetry current:
1 L = A J A + 2 eA ((A ) ) ,
where the supersymmetry current J A is given by

i J A = 2 GBC BC A .
(2.29)
Finally we tackle the trilinear terms, which as usual are the trickier ones. Just as in the six-dimensional theory, their vanishing will be seen to be a property of some identities between the -matrices. Writing the SO(3) indices explicitly, we nd that these terms are given by
1 e A a 2 abc
c A b ,
(2.30)
and we once again must use the Fierz identities to expand the bi-spinor c a .
Exercise 2.30 (A ten-dimensional Fierz identity) Prove that
1 c c a = 32 A a A (1 + 11 ) c + 1 ABC a ABC (1 + 11 ) 323!
a ABCDE 1 c 325! ABCDE
Using the results of Exercise 2.28in particular equation (2.27)and taking into account the antisymmetry of abc we see that only the rst and last terms on the right-hand side of the above Fierz identity contribute to (2.30).
Exercise 2.31 (Some more -matrix identities) Prove the following identities between ten-dimensional -matrices: A B A = 8 B and F ABCDE F = 0 , (2.31)
and use them to deduce that the trilinear terms (2.30) cancel exactly. (Compare these identities with those in Exercise 2.10.)
2.4.2
Reduction to d=4: N =4 supersymmetric Yang Mills
We now dimensionally reduce the d=10 N =1 supersymmetric YangMills theory described in the previous section down to four dimensions. From 69
[email protected]
now on we will let uppercase indices from the middle of the Latin alphabet: I, J, K, . . . run from 1 to 3 inclusive. It will be convenient to break up the ten-dimensional coordinates as xA = (x , x3+I , x6+J ), and by dimensional reduction we simply mean that we drop the dependence of the elds on (x3+I , x6+J ): 3+I 6+J 0. We also need to decompose the ten-dimensional -matrices. This is done as follows: = 14 3
3+I
C = C 14 12
6+J
= 14 1
I
= i5 J 3 ,
(2.32)
where C is the charge conjugation matrix in four-dimensional Minkowski space obeying C t = C and (C )t = C ; and {I } and { J } are 4 4 real antisymmetric matrices satisfying the following algebraic relations: [I , J ] = 2 [ I , J ] = 2 [I , J ] = 0 ; and where 1n denotes the n n unit matrix. (From now on we will drop the subscript when the dimension is clear from the context.) In the above decomposition, 11 takes the form: 11 = 1 1 2 . We can nd an explicit realisation for the matrices I and J as follows. Because they are real antisymmetric 4 4 matrices, they belong to so(4). Their commutation relations say that they each generate an so(3) subalgebra and moreover that these two so(3) subalgebras commute. Happily so(4) so(3) so(3), so that all we have to nd is an explicit realisation = of this isomorphism. This is found as follows. We say that a matrix A in so(4) is self-dual (respectively antiselfdual ), if its entries obey Aij = 1 ijkl Akl 2 1 (respectively, Aij = 2 ijkl Akl ). The next exercise asks you to show that the subspaces of so(4) consisting of (anti)self-dual matrices dene commuting subalgebras.
Exercise 2.32 (so(4) so(3) so(3) explicitly) = Prove that the commutator of two (anti)self-dual matrices in so(4) is (anti)selfdual, and that the commutator of a self-dual matrix and an antiselfdual matrix in so(4) vanishes. (Hint: Either compute this directly or use the fact that the duality operation is so(4) invariant since ijkl is an so(4)-invariant tensor, whence its eigenspaces are ideals.)
IJK IJK
K K
{I , J } = 2 IJ 14 { I , J } = 2 IJ 14
70
[email protected]
Using this result we can now nd a explicit realisation for the I and the J : we simply nd a basis for the (anti)self-dual matrices in so(4). This is the purpose of the next exercise.
Exercise 2.33 (Explicit realisation for I and J ) Prove that a matrix A in so(4) is (anti)self-dual if its entries are related in the following way: A12 = A34 A13 = A24 A14 = A23 ,
where the top signs are for the self-dual case and the bottom signs for the antiselfdual case. Conclude that explicit bases for the (anti)self-dual matrices are given by: e+ = i2 1 = 1 e+ = 3 i2 = 2 e+ = 1 i2 = 3 0 1 1 0 i2 0 0 i2 0 i2 i2 0 e = 1 i2 = 1 e = i2 3 = 2 e = i2 1 = 3 i2 0 0 i2 0 3 3 0 0 1 1 0 ,
where the {e+ } are self-dual and the {e } are antiselfdual. Prove that I = e I I I and J = eJ is a valid realisation. Exercise 2.34 (The fundamental representation of su(4)) Either abstractly or using the above explicit realisation, prove that the fteen (4 4)-matrices: AIJ =
IJK K
B IJ =
IJK K
C IJ = i{I , J }
(2.33)
are antihermitian and generate the su(4) Lie algebra. This is the fundamental representation of su(4).
The result of the above exercise and the above decomposition of the ten-dimensional -matrices mean that we have broken up a ten-dimensional spinor index (running from 1 to 32) into three indices: a four-dimensional spinor index (running from 1 to 4), an internal su(4) index in the fundamental representation (i.e., also running from 1 to 4), and an internal su(2) index also in the fundamental representation. We have chosen the above decomposition of the -matrices because it possesses two immediate advantages: 1. Because of the form of 11 , a Weyl spinor in ten-dimensions gives rise to a unconstrained Dirac spinor in four-dimensions with values in the fundamental representation of su(4). In other words, the chirality condition only aects the internal su(2) space and does not constrain the other degrees of freedom; and 71
[email protected]
2. Because of the form of the charge conjugation matrix, the Majorana condition in ten dimensions becomes the Majorana condition in four dimensions. Thus we see immediately that a MajoranaWeyl spinor in ten-dimensions yields a quartet of Majorana spinors in four-dimensions, or equivalently a Majorana spinor in four-dimensions with values in the fundamental representation of su(4).
This su(4) is a avour index of the four-dimensional theory; that is, su(4) is a global symmetry of N =4 supersymmetric YangMills theory, not a gauge symmetry. Of course, this avour symmetry is nothing but the residual Lorentz symmetry in ten-dimensions which upon dimensional reduction to four-dimensions breaks down to SO(1, 3) SO(6). The Lie algebras of SO(6) and SU (4) are isomorphic. In fact, SU (4) Spin(6), the = universal covering group of SO(6); and the four dimensional representations of SU (4) are precisely the spinorial representations of Spin(6) under which the supersymmetric charges transform.
We now want to write down the four-dimensional action obtained by the above dimensional reduction. We dene the scalar elds SI = W3+I and pseudoscalar elds PJ = W6+J . Together with the four-dimensional gauge elds W they comprise the bosonic eld content of the four-dimensional theory. As mentioned above, the chirality condition on a ten-dimensional spinor can be easily imposed. Let us write 1 = 2 1 i ,
where is a quartet of unconstrained Dirac spinors in four dimensions. From 1 1 the form of 11 it is easy so see that 11 = 2 2 = . The i Majorana condition says that is Majorana in four dimensions. Naturally, all elds are in the adjoint representation of the gauge group SO(3). In order to write down the action we need to dimensionally reduce the Dirac operator and the gauge eld-strength. We nd that GAB breaks up as G , G,3+I = D SI , G,6+I = D PJ , G3+I,3+J = eSI SJ , G3+I,6+J = eSI PJ , and G6+I,6+J = ePI PJ . This allows us to write the bosonic part of the lagrangian density immediately: Lb = 1 G G + 1 D SI D SI + 1 D PJ D PJ 4 2 2 2 1 2 1 2 4 e SI SJ 4 e PI PJ 2 1 e2 SI PJ 2
. (2.34)
The fermionic part of the action requires a bit more work, but it is nevertheless straightforward, and is left as an exercise. 72
[email protected]
Exercise 2.35 (The fermionic terms in the lagrangian) Using the explicit form of the -matrices, prove that the term equation (2.25), becomes
i 2
A DA in (2.35)
i e Lf = 2 D + 2 (I SI + J PJ 5 ) ,
from where we see that indeed we were justied in calling SI scalars and PJ pseudoscalars.
We now write down the supersymmetry transformations. In ten dimensions, the parameter of the supersymmetry transformation is a Majorana Weyl spinor. As we have seen, upon dimensional reduction, such a spinor yields a quartet of Majorana spinors in four-dimensions. Therefore the fourdimensional theory will have N =4 supersymmetry. Indeed the lagrangian density Lb + Lf , understood in four dimensions, denes N =4 supersymmetric YangMills theory. Since the supersymmetry parameter is a MajoranaWeyl spinor obey1 1 ing 11 = , we can write it as = 2 , where is a quartet of i four-dimensional anticommuting Majorana spinors.
Exercise 2.36 (N =4 supersymmetry transformations) Expand equation (2.26) in this reparametrisation to obtain the following supersymmetry transformations for the four-dimensional elds: W = i SI = I PJ = 5 J
1 = 2 G + iD SI I + iD PJ 5 J
e(SI PJ )5 I J + 1 e 2
IJK (SI
1 SJ )K + 2 e
IJK (PI
PJ ) K .
Finally we have the su(4) invariance of the action.

Exercise 2.37 (su(4) invariance) Prove that, for every choice of constant parameters (aIJ , bIJ , cIJ ) where aIJ = aJI and bIJ = bJI , the following transformations are a symmetry of N =4 supersymmetric YangMills theory: W = 0 SI = 2aIJ SJ + 2cIJ PJ PI = 2bIJ PJ 2cJI SJ
i = 1 aIJ AIJ 1 bIJ B IJ + 2 cIJ C IJ 5 , 2 2
73
[email protected]
where AIJ , B IJ and C IJ are the su(4) generators in the fundamental representation given by equation (2.33). (Hint: You may save some time by rst showing that these transformations are induced from Lorentz transformations in ten dimensions, and then using the Lorentz invariance of the ten-dimensional action.)
2.4.3
Monopoles and gauge bosons in N =4 supersymmetric YangMills
In section 2.3.4, we saw how any BPS-monopole could be thought of as a solution to the equations of motion of N =2 supersymmetric YangMills by setting the fermions to zero and aligning the scalar elds properly. Moreover we saw that such solutions break one half of the supersymmetry, so that these N =2 BPS-monopoles naturally belong to a short multiplet. In fact, they belong to the short multiplet with spin s=0. On the other hand, we had seen in section 2.3.3 that after higgsing, the perturbative spectrum of the theory arranged itself in a massless vector multiplet corresponding to the unbroken U (1) and massive short multiplets with spin s= 1 containing the 2 massive vector bosons. It therefore seemed unlikely that N =2 super Yang Mills would be self-dual, since the perturbative spectrum of the dual theory (i.e., the monopoles) now live in a dierent supersymmetry multiplet. And in fact, we now know from the results of Seiberg and Witten, that this theory is not self-dual. In this section we will see that this obstacle is overcome in N =4 supersymmetric YangMills theory. The discussion is very similar to that of sections 2.3.3 and 2.3.4, with the important distinction that the short multiplet containing the BPS-monopole and the one containing the massive vector boson are now isomorphic, being the one with spin s=0. This section and the next are based on the work of Osborn [Osb79]. The bosonic part of the hamiltonian density corresponding to the N =4 supersymmetric YangMills theory dened by (2.34) and (2.35) is given by: Hb =
1 2
Ei
1 2
1 D0 SI 2 + 1 D0 PJ 2 + 1 Bi 2 + 2 Di SI 2 + 1 Di PJ 2 2 2 2 2 2 1 2 1 2 1 2 + 2 e SI PJ + 4 e SI SJ + 4 e PI PJ 2 .
Demanding that the energy of a given eld conguration be nite doesnt necessarily imply that all the scalars PI and SJ acquire non-zero vacuum expectation values at spatial innity. Indeed, looking at the potential terms it is sucient (for so(3)) that they be parallel. This denes the supersymmetric PrasadSommereld limit as in N =2. In more detail, we add a potential term 2 ( SI 2 + PJ 2 a2 ) to the lagrangian (2.34) to force SI and PJ to acquire a nonzero vacuum expectation value, but since such a term would break 74
[email protected]
supersymmetry, we take the limit 0 while keeping the nonzero vacuum expectation values of SI and PJ . This restores the supersymmetry provided that SI and PJ are parallel. We could choose SI = aI and PJ = bJ where I (a2 + b2 ) = 1, and where has length a at innity. Since the I I potential depends only on the SO(6) invariant combination SI 2 + PJ 2 , we could use this symmetry to choose, say, bJ = a2 = a3 = 0, a1 = 1 and = a, where a is a xed vector with a 2 = a2 .
Exercise 2.38 (The perturbative spectrum after higgsing) We can analyse the spectrum of the model around such a vacuum in exactly the same way as we did in Exercise 1.4. Choosing for example the unitary gauge a = ae3 , show that there is now a massless gauge multiplet with helicity = 1, 3 3 3 corresponding to the unbroken U (1): (W , 3 , SI , PJ ); and two massive multiplets ( , W , PI , S1,2 ) of mass MW = ae . Conclude that these massive multiplets are actually short multiplets of spin s=0.
Now lets see to what kind of multiplets the N =4 BPS-monopoles belong. Let (Wi , ) be a BPS-monopole and let us set W0 = = 0, SI = aI , and PJ = bJ , where aI and bJ are real numbers satisfying I (a2 + b2 ) = 1. I I Because the fermions are zero, only the bosonic part of the lagrangian is nonzero. Plugging in these eld congurations into (2.34), we nd L = 1 Gij Gij 4
1 2
Di
after using that the elds are static and that SI and PJ are all collinear. But this is precisely the action for static solutions to the bosonic YangMills Higgs theory, hence it is minimised by BPS-monopoles. Therefore the above eld congurations minimise the equations of motion of N =4 supersymmetric YangMills. In other words, we have shown that any BPS-monopole can be embedded as a solution of the N =4 supersymmetric YangMills theory. (Compare with Exercise 2.17.) Now we will prove that such a solution breaks one half of the supersymmetry and hence it lives in a short multiplet. Because the fermions are put to zero, the supersymmetry transformation of the bosonic elds is automatically zero. From the results of Exercise 2.36 we can read o the expression for the supersymmetry transformation of the spinors in this background: =
1 G 2 ij ij
iDi i (aI I + bJ J 5 )
If we now use equation (2.19), and the Bogomolnyi equation in the form Gij = ijk Dk , takes the form: = k Dk 5 0 i(aI I + bJ J 5 ) = 5 k Dk (1 5 ) , 75
[email protected]
where i 0 i , 4 i0 (aI I + bJ J 5 ), and 5 1 2 3 4 . As expected, 1 the i generate a euclidean Cliord algebra, and it follows that 2 (1 5 ) projects out one half of the states: those which have positive chirality with respect to 5 . Hence we conclude that N =4 BPS-monopoles break one half of the supersymmetry. From the analysis of N =4 multiplets in section 2.2.3 and, in particular, from Exercise 2.6, we see that a multiplet where states break one half of the supersymmetries are short; and looking at the spectrum, we see that it is the short multiplet with spin s=0, which is the only short multiplet with spins not exceeding 1. Therefore the BPS-monopole and the massive vector boson belong to isomorphic multiplets. This solves the second problem with the MontonenOlive duality conjecture alluded to in section 1.4.1for, certainly, if N =4 supersymmetric YangMills is to be self-dual, the BPS-monopole and the massive vector boson should belong to isomorphic multiplets. Finally, we come to a minor point. The supersymmetry parameter , just like the fermion , is a quartet of Majorana spinors. The additional condition for the parameter to preserve the supersymmetry is that it be chiral with respect to 5 . One might be tempted to think that there is a prob lem since in four-dimensions (either with euclidean or lorentzian signature) there are no MajoranaWeyl spinors. However the Majorana condition is a condition in Minkowski spacetime, whereas the chirality condition is a condition relative to the euclidean 5 . We will see this more explicitly later on when we consider the eective action for the collective coordinates, but for now let us simply state without proof that these two conditions are indeed simultaneously realisable.
2.4.4
The mass bound for N =4 super YangMills
We end this chapter with a derivation of the mass bound for N =4 super YangMills. Keeping in mind the similar calculation for N =2 super Yang Mills, it should come as no surprise that the mass bound coincides once again with the Bogomolnyi bound. In order to derive the mass bound, we will rst write down the algebra obeyed by the supersymmetry charges in d=10 N =1 super YangMills. After dimensional reduction this will give us an explicit expression for the central charges appearing in the four-dimensional supersymmetry algebra. Naturally one could compute the supersymmetry algebra directly in four dimensions, but we nd it simple to dimensionally reduce the algebra in ten dimensions.
76
[email protected]
The supersymmetry algebra in ten dimensions The supersymmetry algebra can be derived by varying the supersymmetry current (2.29). Indeed, the supersymmetry algebra will be read o from the space integral of the supersymmetry variation of the timelike (zeroth) component of the supersymmetry current. Explicitly, if is a Majorana Weyl spinor just like , then {Q, Q} = i
space
J 0 ,
where the integral is over a spacelike hypersurface. We can get an idea of what to expect in the right-hand side of the supersymmetry algebra purely from the fact that Q is an anticommuting MajoranaWeyl spinor. From Exercise 2.28 we see that in the right-hand side of the supersymmetry algebra, we expect only terms consisting of an odd number of matrices and moreover only those bispinors for which C is symmetric, since so is the left-hand side of the supersymmetry algebra. Using equation (2.28), we see that only those terms with 1 and 5 matrices survive. We now turn to the computation, which is left as an exercise. We will only be interested in terms which survive in a BPS-monopole monopole background in which the fermions have been put to zero.
Exercise 2.39 (The supersymmetry algebra in ten dimensions) Prove that up to terms involving the fermions, the variation of the supersymmetry current is given by i J E = 1 AB E CD GAB GCD . 4 Perform the matrix algebra and, taking into account that and are Majorana Weyl, show that i J E = 1 ABCDE GAB GCD + 2GEA GAB B + 1 GAB GAB E 4 2 Prove the identity
1 ABCDE = 5! ABCDEF GHIJ
F GHIJ 11 ;
and using the fact that 11 = , conclude that i J E = 2

1 ABCDEF GHIJ GAB 85!
GCD F GHIJ .
E 1 + GEA GAB + 4 GCD GCD B B
77
[email protected]
We now dene the following tensors T AB = GAC GC B + 1 AB GCD GCD 4 ABCDEF =

1 ABCDEF GHIJ GGH 8
GIJ .
We recognise T as the bosonic part of the (improved) energy-momentum tensor of the super YangMills theory. The momentum is then given by the space integral of T 0A : PA =
space
T 0A .
How about ? Just as in the case of N =2, it is a topological current.

Exercise 2.40 (Another topological current) Prove that ABCDEF is gauge invariant and that it is conserved without imposing the equations of motion. (Hint: Compare with Exercise 2.23.)
We dene the topological charge associated to as the space integral of 0ABCDE : Z ABCDE = 0ABCDE .
space
In summary, the supersymmetry algebra remains as follows:

2 {Q, Q} = 2P A A + 5! Z ABCDE ABCDE .
(2.36)
In 10 dimensions, the 5-form Z ABCDE can be decomposed into a self-dual and an antiself-dual part. The next exercise asks you to show that only the self-dual part contributes to the algebra.
Exercise 2.41 (A self-dual 5-form) Using the fact that the supersymmetry charge has negative chirality, show that we can for free project onto the self-dual part of Z ABCDE in the left-hand side of the supersymmetry algebra.
This 5-form belies the existence of a 5-brane solution of ten dimensional supersymmetric YangMills. Under double dimensional reduction, it gives rise to the string-like solution of six-dimensional supersymmetric YangMills briey alluded to in section 2.3.4.
The supersymmetry algebra in four dimensions In order to write down the supersymmetry algebra in four dimensions, we need to dimensionally reduce both the momenta and the topological charge appearing in the ten-dimensional algebra (2.36). We will assume from the 78
[email protected]
start a BPS-monopole background where the fermions are put to zero. We will not demand that the solutions be static, since that is the only way we can generate electric charge. Moreover we will exploit the internal SO(6) symmetry to choose PJ = S2,3 = 0 and S1 = . In such a background, the only nonzero components of the eld strength GAB are G , G4 . This limits considerably the nonzero terms of the momentum P A and the topological charge Z ABCDE , as the next exercise shows.
Exercise 2.42 (Momentum and topological charge in this background) Prove that in the background chosen above, the only nonzero components of the momentum and topological charge densities are the following: T 0 , T 04 , 056789 . The rst term is of course simply the four-momentum density, whereas the other two terms are given by: T 04 = G0i Di = i (G0i ) 056789 = 1 2
ijk Gij 1 Dk = 2 k ( ijk Gij
) .
(Hint: In order to rewrite the right-hand sides of the equations, use the equations of motion in this background, and the Bianchi identity. (Compare with the discussion following Exercise 2.24.))
Taking into account the results of the previous exercise we can rewrite the supersymmetry algebra in four dimensions as follows: {Q, Q} = 2 P 24
G0i di 56789
ijk
Gij dk . (2.37)
But now notice that , 4 and 56789 = 5 6 7 8 9 generate a lorentzian Cliord algebra of signature (1,5), hence the supersymmetry algebra (2.37) is formally identical to one in six-dimensional Minkowski spacetime where the momenta in the extra two dimensions are given by: P5 = P6 = G0i di = aq
1 2 ijk
Gij dk = ag ,
and where we have used equations (1.24) and (1.25) to rewrite the extra momenta in terms of the electric and magnetic charges. Finally, as in the N =2 case, the mass bound is simply the positivity of the six-dimensional mass given by equation (2.20). Plugging in the expression for the extra momenta, we once again recover the Bogomolnyi bound (1.27).
79
Chapter 3 Collective Coordinates

BPS-monopoles are static: any motion, however small, increases their kinetic energy and makes their total energy strictly greater than the Bogomolnyi bound. Nevertheless, if we keep the velocity small and if the motion starts o tangent to the space of static BPS-monopoles, energy conservation will prevent the motion from taking the monopoles very far away from this space. Much like a point-particle moving slowly near the bottom of a potential well, the motion of slow BPS-monopoles may be approximated by motion on the space of static BPS-monopoles (i.e., along the at directions of the potential) and small oscillations in the transverse directions. We can trade the limit of velocities going to zero, for a limit in which the potential well becomes innitely steep. This suppresses the oscillations in the transverse directions (which become increasingly expensive energetically) and motion is eectively constrained to take place along the at directions, since this motion costs very little energy. Manton [Man82] showed that the motion along the at directions is geodesic relative to a metric on the moduli space of BPS-monopoles, which is induced naturally from the YangMillsHiggs action functional. Expanding the action functional around a BPS-monopole gives rise to an eective theory in terms of collective coordinates. These are the coordinates on the moduli space of BPS-monopoles and the eective action is nothing but a (1+0)-dimensional -model with target space the moduli space. In this chapter we will study the moduli space M of BPS-monopoles. Our aim is to prove that it is a hyperkhler manifold which, for a given magnetic a charge, is nite-dimensional, and to compute its (formal) dimension. For the simplest case of magnetic charge k=1, we will also work out the metric explicitly from the eld theory and hence the eective action. We quantise the eective action following the review in the introduction of [GM86]. This is as much as can be done directly with eld-theoretical methods. The only 80
[email protected]
other case in which the metric on moduli space is known exactly is the k=2 monopole sector. This metric was constructed by Atiyah and Hitchin [AH85] by indirect methods. We will eventually review their construction as well.
3.1
The metric on the moduli space
We start by constructing the metric on the true physical conguration space of the YangMillsHiggs theory. This will induce a metric on the moduli space of BPS-monopoles, which is a submanifold.
3.1.1
The physical conguration space
Let A denote the space of congurations (W , ) of YangMillsHiggs elds of nite energy in the PrasadSommereld limit. Recall that in this limit, the energy density is given by (1.13) setting V () = 0. Congurations which are related by a short-range gauge transformationthat is, gauge transformations which tend to the identity at innityare to be thought of as physically indistinguishable. Hence if we let G denote the group of shortrange gauge transformations, the true conguration space of the YangMills Higgs system is the quotient C A /G . = It is convenient to x the gauge partially by setting W0 = 0, that is by going to the temporal gauge. This still leaves the freedom of performing time-independent gauge transformations, since these are the gauge transformations which preserve the temporal gauge: e W0 = D0 = = 0 . We will therefore let G G denote the group of time-independent short-range gauge transformations. The temporal gauge W0 = 0 is preserved by the dynamics provided we impose its equation of motion, that is, Gausss law : Di Wi e = 0 . (3.1)
Therefore if we let A denote the space of nite-energy congurations (Wi , ) subject to Gausss law, then the conguration space C also admits the description C A/G . = 81
[email protected]
Notice that setting W0 = 0 means that there will be no static dyons (cf. the JuliaZee Ansatz). We will see that dyonic solutions emerge when we consider moving monopoles.
We can simplify many of the calculations by describing the space A in a dierent way. We introduce a fourth spatial coordinate x4 and interpret the Higgs eld as the fourth-component W4 of the gauge eld. Notice however that we must impose that nothing depends on the new coordinate: 4 0. We will write Wi = (Wi , W4 = ), where the underlined indices run from 1 to 4. Notice that the eld-strength has components Gij = (Gij , Gi4 = Di ). In this new notation, gauge transformations, Gausss law and the Bogomolnyi equation all have natural and simple descriptions.
Exercise 3.1 (The BPS-monopole as an instanton) Prove that innitesimal gauge transformations on the YangMillsHiggs system now take the form 1 Wi = Di ; e that Gausss law (3.1) becomes simply Di Wi = 0 ; (3.2)
and that the Bogomolnyi equation (1.28) is nothing but the (anti)self-duality equation Gij = 1 ijk Gk . (3.3) 2 In summary, this proves that BPS-monopoles in 3+1 dimensions are in one-to-one correspondence with static instantons in 4+1 dimensions which are translationally invariant in the fourth spatial direction.
Therefore in this description, the space A is given as those gauge elds Wi in 4+1 dimensions, independent of x4 , of nite energy per unit length in the x4 -direction, and whose time-dependence is subject to (3.2).
3.1.2
The metric on the physical conguration space
In the temporal gauge, the YangMillsHiggs lagrangian in the Prasad Sommereld limit is given as a dierence of two terms: L = T V , where the kinetic term T is the 3-space integral of
1 2
Wi
1 2
Wi
1 2
(3.4)
and the potential term V is the 3-space integral of

1 2
Bi
1 2
Di
1 2
Bi 82
Di
i ( Bi ) .
(3.5)
[email protected]
We will now show that the lagrangian is well-dened in the true conguration space C. The kinetic term will induce a metric. Suppose we would like to compute the value of the potential on some point in C. Points in C are equivalence classes [Wi ] of points Wi in A: two points in A belong to the same equivalence class if and only if they are related by a gauge transformation in G; that is, if they lie on the same G-orbit. To dene a potential on C we can simply use the potential term (3.5) on A as follows: to nd out the value of the potential on a point [Wi ] in C, we choose some point Wi in the same equivalence class, and we evaluate the potential (3.5) on it. This will only make sense if the value of the potential doesnt depend on which element in the equivalence class we have chosen; that is, if the potential is gauge-invariant. More formally, a function on A will induce a function on C = A/G if and only if it is G-invariant. Luckily this is the case, since in the temporal gauge, the potential and kinetic terms are separately invariant under time-independent gauge-transformations. The kinetic term is trickier since it is not strictly speaking a function on A: it requires not just knowledge of Wi but also of its time-derivative Wi . In other words, it is a function on the tangent bundle T A. The typical bre at a point [Wi ] of the tangent bundle is spanned by the velocities of all smooth curves passing through that point. We may lift such curves to curves in A, but this procedure is not unique. First we have to choose a point Wi in A in the equivalence class [Wi ]. Just as before, this ambiguity is immaterial since the kinetic term is invariant under time-independent gauge transformations. But now we also have to choose a tangent vector Wi . Clearly adding to a tangent vector a vector tangent to the orbits of G does not change the curve in C since every G-orbit in A is identied with a single point in C. Hence the kinetic term should be impervious to such a change. Tangent vectors to G are innitesimal gauge-transformations whose parameters go to zero at spatial innity, therefore the kinetic term (3.4) denes a kinetic energy on C provided that Wi and Wi + Di have the same kinetic energy. Integrating by parts we see that this is a consequence of Gausss law (3.2). In summary, the YangMillsHiggs lagrangian induces a lagrangian in the true conguration space C, whose energy is of course given by E = T + V . The kinetic energy term T denes a metric on C. Motion on C is not free of course, since there is also a potential term, but for motion along the at directions at the bottom of the potential well, this will be a good approximation. We turn to this now.
83
[email protected]
3.1.3
The metric on the moduli space
Let M denote the subspace of C where the energy E attains its minimum. From the explicit expression for T and V given in (3.4) and (3.5), we see that the minimum of the energy is given by d3 x i ( Bi ) = a|g| = 4a |k| , e (3.6)
R3
where g is the magnetic charge, and the integer k is the topological or monopole number. Clearly the minimum is attained by those congurations corresponding to static solutions on the Bogomolnyi equation: BPSmonopoles, and where any two such solutions which are gauge-related are identied. In other words, M is the moduli space of static BPS-monopoles. The monopole number labels dierent connected components of the space A, so that Ak , A=
k
but the gauge group G preserves each component. Therefore we can also decompose the true conguration space C as C=
k
Ck
where Ck = Ak /G .
Finally, let Mk = M Ck . This is then the moduli space of static BPSmonopoles of monopole number k, or BPS-k-monopoles, for short. By denition, the potential is constant on Mk , so that the YangMills Higgs lagrangian is given by L=T 4a |k| . e (3.7)
Therefore Mk corresponds to the manifold of at directions of the potential. Mantons argument given at the beginning of this chapter, can now be proven. The motion of slow monopoles which start o tangent to Mk will consists of the superposition of two kinds of motions: motions along the at directions Mk and small oscillations in the directions normal to Mk . In the limit of zero velocity, the oscillatory motion is suppressed and we are left with motion on Mk . But this motion is governed by the lagrangian (3.7) which only has a kinetic termwhence the motion is free, or in other words geodesic relative to the metric on Mk dened by T .
84
[email protected]
3.1.4
The 1-monopole moduli space
Let us study the moduli space M1 in the 1-monopole sector. The coordinates for M1 can be understood as parameters on which the BPS-monopole solution depends. In the t HooftPolyakov Ansatz (1.15), the monopole is centred at the origin in R3 , but the invariance under translations of the YangMills Higgs lagrangian (1.8) means that we can put the centre of the monopole where we please. This introduces three moduli parameters: X. The time evolution of these parameters corresponds to the BPS-monopole moving as if it were a particle with mass 4a/e. The eective lagrangian for these collective coordinates is then Le = 2a 2 X . e
There is a fourth, more subtle, collective coordinate. Consider a oneparameter family Wi (t) of gauge elds, but where the t-dependence is pure gauge: 1 Wi = Di (t) . (3.8) e Since the potential is gauge invariant, this corresponds to a at direction. But one might think that it is not a physical at direction since it is tangent to the G-orbitsit is an innitesimal gauge transformation, after all. But recall that G is the group of short-range gauge transformations, whence for Di to be tangent to the orbits, has to tend to 0 as we approach innity. Indeed, Di would represent a physical deformation of the BPS-monopole if it would obey Gausss law (3.2), which implies D2 = 0 . (3.9)
Exercise 3.2 (D2 has no normalisable zero modes) Prove that acting on square-integrable functions D2 = Di Di is a negative-denite operator. Deduce from this that any normalisable zero mode must be a zero mode of Di for each i, and deduce from this that the only square-integrable solution to (3.9) is the trivial solution = 0. In other words, there exist no normalisable solutions. 2 is constant.) (Hint: If Di = 0, then
This discussion suggests that we look for a gauge parameter which does not tend to zero asymptotically. For example, let (t) = f (t), where f (t) is an arbitrary function. In the 1-monopole sector, denes a map of degree 1 at innity, hence it certainly does not go to zero. Moreover, using the Bogomolnyi equation, it follows at once that f (t) is a (un-normalisable) 85
[email protected]
zero mode of D2 . It is clearly a true moduli parameter because it costs energy to excite it: 1 T = 2 f2 Di 2 > 0 . (3.10) 2e 3 R We can understand this as follows. Let g = exp(/a) be a timedependent gauge transformation, where all the time-dependence resides in . Such a gauge transformation will move us away from the temporal gauge, but assume that at time t=0, say, we start from a conguration Wi in the temporal gauge, and suppose that g(t=0) = 1. Then from (1.9), 1 Wi (t) = gWi g 1 + i g g 1 , e 1 Wi (t) = Di . ae Comparing with (3.8), we see that f = /a. Using (3.10), the kinetic energy of such a conguration is given by T = 1 2 Di 2 e2 2a R3
2
whence
But notice that since Di = Bi , T = 1 1 2 Di 2 2 e2 2a R3 4a 1 = 2 2 2 2a e e 2 2 = 3 . ae

2
1 2
Bi
Notice that is an angular variable. To see this, let us dene g() = exp(/a). Then it is easy to see that g() and g(+2) are gauge-related in G, that is, via short-range gauge transformations. Indeed, recall that a at innity in the PrasadSommereld limit, whence g(2) = exp(2/a) tends to 1 at innity. Since g( + 2) = g()g(2), we are done. Assuming for the moment (we will prove this later) that there are no other collective coordinates in the 1-monopole sector, we have proven that the moduli space of BPS-1-monopoles is given by M1 R 3 S 1 , =
86
[email protected]
and the metric can be read o from the expression for the eective action 4a 1 Le = 2 gab X a X b e 2a 2 2 2 4a = X + 3 ; e ae e that is gab = (3.11) (3.12)
4a 13 0 , 0 e2 e from where we can see that the radius of the circle is inversely proportional to the electric charge.
3.1.5
The quantisation of the eective action
The eective action (3.12) corresponds to a particle moving freely in R3 S 1 with the at metric. The quantisation of this eective action is straightforward. The canonical momenta (P , Q) given by 4a 4 X and Q = 3 , e ae are conserved, and the hamiltonian is given by P = ae3 2 4a e 2 P + Q + . (3.13) H= 8a 8 e Bound states of minimum energy are given by those eigenstates of the hamiltonian for which P = 0. Since is angular with period 2, the eigenvalues of Q are quantised in units of , whence the spectrum looks like 4a ae3 + (n )2 . (3.14) e 8 Notice that this energy spectrum has the standard form which corresponds to perturbative states around a nonperturbative vacuum. If we think of e as the coupling constant, then the zero-point energy is not analytic in e, hence it corresponds to a non-perturbative state in the theory: the BPS-monopole in this case. The second term in the energy corresponds to excitations around the monopole, which are clearly perturbative since their energy goes to zero as we let the coupling tend to zero. En =
Exercise 3.3 (The electric charge) Prove that the electric eld for classical congurations in which P = 0 is given by Ei = G0i = Wi = e2 QBi /4 . Conclude that e Q can be interpreted as the electric charge.
87
[email protected]
Taking the above exercise into consideration, we see that the spectrum of the quantum eective theory corresponds to dyons of magnetic charge 4/e and electric charge ne , for n Z. According to the classical BPS formula (1.27), the rest mass of such a dyon would be equal to Mn = a 4 e
2
+ (ne )2 =
4a e
1+
ne2 4
which, if we expand the square root assuming that e is small, becomes = En + O(e5 ) , where weve used (3.14). In summary, the energy spectrum obtained from quantising the eective action of the collective coordinates is a small-coupling approximation to the expected BPS energy spectrum. However, even if their energy is only approximately correct, the multiplicity of bound states can be read accurately from the eective action. This is one of the important lessons to be drawn from the collective coordinate expansion. In principle one can repeat this analysis in the k-monopole sector provided that one knows the form of the metric. But at the present moment this is only the case for k=1 and k=2. We will discuss the eective theory for k=2 later on in the lectures in the context of N =4 supersymmetric YangMills.
3.1.6
Some general properties of the monopole moduli space
Quite a lot is known about the properties of the k-monopole1 moduli space Mk , even though its metric (and hence the eective action) is known explicitly only for k = 1, 2. As we saw in the previous section, the metric on M1 can be computed directly from the eld theory. On the other hand, the metric on M2 can only be determined via indirect means. This result as well as much else of what is known about Mk is to be found either explicitly or referenced in the book [AH88] by Atiyah and Hitchin (see also [AH85]) to where we refer the reader for details. We will now state some facts about Mk . Some of them we will be able to prove later with eld-theoretical means, but proving some others would take us too far aeld. The following properties of Mk are known [AH88]:
We will only concern ourselves with positive k: Mk is naturally isomorphic to Mk by performing a parity transformation on the solutions.
1
88
[email protected]
1. Mk is a 4k-dimensional (non-compact) complete riemannian manifold; 2. The natural metric on Mk is hyperkhler; a = 3. Mk Mk /Zk where Mk (R3 S 1 ) M0 as hyperkhler spaces. a = k a 4. M0 is a 4(k1)-dimensional, irreducible, simply-connected, hyperkhler k manifold admitting an action of SO(3) by isometries which rotates the three complex structures; 5. Asymptotically Mk M1 M1 M1 Mk , and Mk Mk /Zk . 1 1 Physically this means that a conguration of well-separated BPS-kmonopoles can be considered as k 1-monopole congurations. The fact that BPS-monopoles are classically indistinguishable is responsible for the Zk -quotient.
k times
3.2
dim Mk = 4k
In this section we compute the dimension of the moduli space of static BPSmonopoles. The strategy is typical of this kind of problems. We x a reference BPS-monopole and ask in how many directions can we deform the solution innitesimally and still remain with a BPS-monopole. Most of these directions will be unphysical: corresponding to innitesimal gauge transformations. Discarding them leaves us with a nite number of physical directions along which to deform the BPS-monopole. In other words, we are computing the dimension of the tangent space at a particular point in the moduli space. If the point is regular (and generic points usually are) then this is the dimension of the moduli space itself. This number is in any case called the formal dimension of the moduli space. Since the number of all deformations and of innitesimal gauge transformations are both innite, it is better to x the gauge before counting: this eliminates the gauge-redundant deformations and leaves us with only a nite formal dimension. With a little extra argument, the counting can then be done via an index theorem. In the case of BPS-monopoles, the relevant index theorem is that of Callias [Cal78] (slightly modied by Weinberg [Wei79]) which is valid for open spaces and for operators with suitable decay properties at innity. Weinbergs calculation contains steps which from a strictly mathematical point of view may be deemed unjustied. The necessary analytic details have been sorted out by Taubes [Tau83], but we will be following Weinbergs heuristic calculation in any case. 89
[email protected]
3.2.1
The dimension as an index
First let set up the problem. We want to nd out in how many physically dierent ways can one deform a given BPS-monopole. It will turn out that these are given by zero modes of a dierential operator. Asking for the number of zero modes will be the same as asking for the dimension of the tangent space at a given BPS-monopole solution. Hence let t (Wi (t), (t)) be a family of static BPS-monopoles (here t is an abstract parameter which has nothing to do with time). This means that (Wi , ) (Wi (0), (0)) is a tangent vector to the moduli space at the point (Wi , ) (Wi (0), (0)). Taking the t-derivative of the Bogomolnyi equation (1.28), we nd that (Wi , ) satises the linearised Bogomolnyi equation: Di + e Wi =
ijk Dj Wk
(3.15)
However, not every solution of the linearised Bogomolnyi equation need be a physical deformation: it could be an innitesimal gauge transformation. To make sure that it isnt, it is necessary to impose in addition Gausss law (3.1). In other words, the dimension of the tangent space of the moduli space of BPS-monopoles is given by the maximum number of linearly independent solutions of both (3.15) and (3.1). In order to count these solutions it will be convenient to rewrite both of these equations in terms of a single matrix-valued equation. We will dene the following 2 2 complex matrix: = 1 + iWj j , and the following linear operator: D = e 1 + iDj j , (3.17) (3.16)
where we follow the convention that all elds which appear in operators are in the adjoint representation; that is, really stands for ad = , etc.
Exercise 3.4 (Two equations in one) Prove that the linearised Bogomolnyi equation (3.15) and Gausss law (3.1) together are equivalent to the equation D = 0.
We want to count the number of linearly independent real normalisable solutions to D = 0. It is easier to compute the index of the operator D. By denition, the index of D is dierence between the number of its normalisable zero modes and the number of normalisable zero modes of its hermitian adjoint D relative to the inner product: d3 x tr = d3 x + Wi Wi 90 ,
[email protected]
where denotes complex conjugation of the elds and hermitian conjugation on the 2 2 matrices, and where tr denotes the 2 2 matrix trace. The expression for the index of D ind D = dim ker D dim ker D can be turned into an inequality dim ker D ind D , which saturates precisely when D has no normalisable zero modes. Happily this is the case, as the next exercise asks you to show.
Exercise 3.5 (D has no normalisable zero modes) Prove that DD is a positive-denite operator, whence it has no normalisable zero modes. Compare with Exercise 3.2. Also prove that for antimonopoles (i.e., the other sign in the Bogomolnyi equation, it is D D that is the positive operator. (Hint: Use the fact that both and Dj are antihermitian operators to prove that D = e 1 + ij Dj , and that DD = e2 2 (Dj )2 . Deduce that this operator is positive-denite.
Therefore the number of normalisable zero modes of D equals the index of the operator D. In the following sections we will compute the index of D acting on two-component complex vectors; that is, on functions R3 C2 . However the (formal) dimension of monopole moduli space is given by the number of normalisable zero modes of D acting on matrices of the form (3.16). To a deformation (, Wi ) there corresponds a matrix = + iW3 W2 + iW1 W2 + iW1 iW3 .
Clearly the rst column of the above matrix determines the matrix. Moreover this rst column is a normalisable zero mode of D acting on vectors if and only if is a normalisable zero mode of D acting on matrices. This would seems to indicate that there is a one-to-one correspondence between the normalisable zero modes of D acting on vectors and of D acting on matrices, but notice that D is a complex linear operator in a complex vector space, hence the space of its normalisable zero modes is complex, of complex dimension ind D. However the matrices and i determine linearly independent tangent vectors to monopole moduli space, hence it is its real dimension which equals the (formal) dimension of monopole moduli space Mk . In other words, dim Mk = 2 ind D . 91
[email protected]
3.2.2
Computing the index of D
Our purpose is then to compute ind D. To this eect consider the following expression: I(M 2 ) = Tr M2 D D + M 2 Tr M2 DD + M 2 , (3.18)
where Tr is the operatorial trace.

Exercise 3.6 (A formula for the index of D) Prove that the index of D is given by ind D = dim ker D D dim ker DD = lim I(M 2 ) .
M 2 0
(Hint: Prove this assuming that there is a gap in the spectrum of these operators. This is not the case, but as argued in [Wei79] the conclusion is unaltered.)
In order to manipulate equation (3.18) it is again convenient to use the reformulation of the BPS-monopole as an instanton, in terms of Wi = (Wi , ), and to dene the following four-dimensional euclidean Dirac matrices: k = obeying {i , j } = 2ij . Letting Di denote the gauge covariant derivative corresponding to Wi , and remembering that 4 0, we nd that D i D i = whence ( D)2 = 0 D D 0 , 0 ik ik 0 4 = 0 1 1 0 5 = 1 2 3 4 =
0 0 1
(3.19)
DD 0 0 D D
Exercise 3.7 (Another formula for I(M 2 )) Prove that M2 I(M 2 ) = Tr 5 , ( D)2 + M 2 where Tr now also includes the spinor trace. More generally, if f is any function for which the traces Tr f (D D) and Tr f (DD ) exist, prove that Tr 5 f (( D)2 ) = Tr f (DD ) Tr f (D D) .
92
[email protected]
Let K be any operator acting on square-integrable (matrix-valued) functions (x). K is dened uniquely by its kernel K(x, y): (K )(x) = d3 y K(x, y)(y) .
If we rewrite this equation using Diracs ket notation, so that (x) = x| , then we see that the above equation becomes x| K | = d3 y x| K |y y| ,
whence we can think of the kernel K(x, y) as x| K |y . We will often use this abbreviation for the kernel of an operator. In particular, its trace is given by Tr K = d3 x tr x| K |x ,
where tr stands for the matrix trace, if any. The rest of this section will concern the calculation of the following expression M2 I(M 2 ) = d3 x tr 5 x| |x , (3.20) ( D)2 + M 2 where tr now stands for both the spinor and matrix traces. Lets focus on the kernel M2 I(x, y) = tr 5 x| |y . ( D)2 + M 2
Exercise 3.8 (Some properties of kernels) Let A and B be operators acting on (matrix-valued) square-integrable functions. Let A be a dierential operator. Then prove the following identities: A(x) B(x, y) = (AB)(x, y) B(x, y) A (y) = (BA)(x, y) where means action of dierential operators, the label on a dierential operator denotes on which variable it acts, and the arrow on A in the second equation means that the derivatives act on B.
Using the results of this exercise and the fact that the trace of an even number of -matrices vanishes, we can rewrite I(x, y) slightly. Writing ( 2 2 D) + M = ( D + M )( D + M ), we have that I(x, y) = tr 5 x| M |y = M tr 5 (x, y) , D+M 93 (3.21)
[email protected]
where we have introduced the propagator (x, y) = x| 1 |y . D+M
Using once again the results of Exercise 3.8, one immediately deduces the following identities: ei Wi (x) + M (x, y) = (x y) xi (x, y) i i ei Wi (y) + M = (x y) ; y i and from them: I(x, y) =
1 2
+ i i x y
tr 5 i (x, y)
e 2 tr 5 i (Wi (x) Wi (y)) (x, y) , (3.22)
which can be understood as a conservation law for the bi-local current Ji (x, y) tr 5 i (x, y) . In order to compute I(M 2 ) we have to rst take the limit y x of I(x, y). From equation (3.22) we nd that I(x, x) =
1 2
e Ji (x, x) lim 2 tr 5 i (Wi (x) Wi (y)) (x, y) . yx xi
(3.23)
Although the last term has a Wi (x) Wi (y) which vanishes as y x, the propagator is singular in this limit and we have to pay careful attention to the nature of these singularities in order to conclude that this term does not contribute. Clearly we can admit at most a logarithmic singularity. The purpose of the following (long) exercise is to show that nothing more singular than that occurs.
Exercise 3.9 (Regularity properties of the propagator) Prove that the following limit has at most a logarithmic singularity:
yx
lim tr 5 i (x, y) ,
where tr now only denotes the spinor trace. (Hint: Notice that (x, y) is the propagator of a three-dimensional spinor in the
94
[email protected]
presence of a background gauge eld. First let us approximate (x, y) perturbatively in the coupling constant e:
(x, y) =
n=0
en n (x, y) .
Imposing the equation ( D + M )(x)(x, y) = (x y) order by order in e, we nd that

n
(x, y) =
n=0
n i=1
d3 zi 0 (x, z1 )
n1
i=1
( W)(zi )0 (zi , zi+1 ) ( W)(zn )0 (zn , y) ,
where 0 (x, y) is the free propagator: 0 (x, y) = x| 1 |y = ( ) + M d3 p eip(xy) . (2)3 i( p) + M
Since we are interested in the behaviour of (x, y) as |x y| 0, we need to make some estimates. Prove that n (x y) |x y|2+n in this limit, whence we the potentially singular contributions come from n = 0 and n = 1. Prove that tr 5 i 0 (x, y) = 0 using the facts that tr 5 i = tr 5 i j = 0. These same identities reduce the computation of tr 5 i 1 (x, y) to tr 5 i j k d3 p d3 q pk q eipx eiqy (2)3 (2)3 (p2 + M 2 )(q 2 + M 2 ) d3 z ei(pq)z Wj (z) .
Compute this (introducing Feynman parameters,...) and show that it vanishes.)

It is instructive to compare this with the calculation of the axial anomaly in four dimensions. The same calculation in four dimensions would have yielded a singularity |xy|1 in the n=2 term of the above calculation. A bit of familiar algebra would then have yielded a multiple of the Pontrjagin density for the second term in equation (3.23). In the four dimensional problem, the gauge elds go to zero at innity and the integral of I(x, x) would have received contributions only from the Pontrjagin term, since the i Ji (x, x) would give a vanishing boundary term. In our case, though, the situation is dierent. The second term in (3.23) vanishes, whereas the boundary term coming from i Ji (x, x) is not zero due to the nontrivial behaviour of the Higgs eld at innity.
From the results of the above exercise, the second term in (3.23) vanishes, and using the expression (3.20) for I(M 2 ), we nd that I(M 2 ) =
1 2
d3 x i Ji (x, x) = 95
1 2
dSi Ji (x, x) ,
[email protected]
where is the sphere at spatial innity. The (formal) dimension of the moduli space of BPS-k-monopole will then be given by dim Mk = lim 2
M 0
dSi Ji (x, x) .
(3.24)
In the remainder of this section, we will compute this integral and show that it is related to the magnetic number of the monopole.
3.2.3
Computing the current Ji (x, x)
We start by rewriting Ji (x, x). Inserting 1 in the form ( D + M )1 ( D + M ) into the denition of Ji (x, x), we get 1 Ji (x, x) = tr 5 i x| ( D + M )|x . ( D)2 + M 2 Using the fact that the trace of an odd number of -matrices vanishes, we remain with 1 Ji (x, x) = tr 5 i x| (j Dj + e4 )|x , (3.25) ( D)2 + M 2 where now ( D)2 = (Di )2 e2 2 + 1 eij Gij + ei 4 Di . 2 We proceed by treating the last terms 1 eij Gij + ei 4 Di as a perturbation 2 and expanding 1 1 = 2 + M2 2 e2 2 + M 2 ( D) (Di ) 1 1 ( 1 eij Gij + ei 4 Di ) + . 2 e2 2 + M 2 2 2 e2 2 + M 2 (Di ) (Di ) It is now time to use the fact that Wi corresponds to a monopole background. For such a background Gij = O(|x|2 ) asymptotically as |x| . Because we are integrating Ji (x, x) on , we are free to discard terms which decay faster than O(|x|2 ) at innity, hence no further terms other than those shown in the above perturbative expansion contribute. Plugging the remaining two terms of the expansion into (3.25), we notice that the rst term vanishes due to the trace identity tr 5 i j = 0. Similar identities leave only the following terms: Ji (x, x) = e tr 5 i j 4 k x| 1 1 Dj Dk |x K K
1 e2 tr 5 i jk 4 x| 2 96
1 1 Gjk |x + O(|x|3 ) , K K
[email protected]
where we have introduced the shorthand K = (Di )2 e2 2 + M 2 . Using the trace identity tr 5 i j k 4 = 4 ijk we can rewrite the above equation as Ji (x, x) = 4e
ijk
tr x|
1 1 Dj Dk |x K K 1 1 2e2 ijk tr x| Gjk |x + O(|x|3 ) , (3.26) K K
where the trace now refers only to the SO(3) adjoint representation. The propagators K 1 are not yet those of a free spinor, thanks to their dependence on and Wi , thus we must treat them perturbatively as well. Since Wi decays at innity, we can eectively put Wi = 0 in the propagators in the above expressions which are already O(|x|2 ). On the other hand, the perturbative treatment of is a bit more subtle, since it doesnt decay at innity but rather behaves as a homogeneous function of degree zero; that is, its behaviour on the radius |x| is constant at innity, but not so its angular dependence, which gives rise to the topological stability of the BPS-monopole. First we notice that in the adjoint representation, 2 v = ( v) = a2 v + ( v) , where we have recalled that = a2 at innity. Hence on we can put K = Q + e2 where Q = 2 + M 2 + a2 e2 and is (up to a factor) the projector onto the direction: (v) = ( v). Because [Q, ] = O(|x|1 ) asymptotically, we can eectively treat these two operators as commuting, whence we can write 1 1 1 = + (e2 )n + O(|x|1 ) . n+1 K Q n1 Q Notice moreover that as operators in the adjoint representation of SO(3), ( )v = ( v) = ( v) = 0 . (3.27)
We are now in a position to prove that the rst term in equation (3.26) doesnt contribute.
Exercise 3.10 (The rst term doesnt contribute) Prove that 1 1 tr x| Dj Dk |x = O(|x|3 ) K K
97
[email protected]
whence it doesnt contribute to the integral over . (Hint: First notice that Dk = k + O(|x|1 ), whence up to O(|x|3 ) we can simply substitute k for Dk in the above expression. Now use equation (3.27) and the fact that and Dj are parallel in , to argue that one can substitute K for Q in the above expression (again up to terms of order O(|x|3 )). Then simply take the trace to obtain the result.)
Hence we are left with Ji (x, x) = 2e2

ijk
tr x|
1 1 Gjk |x + O(|x|3 ) . K K
(3.28)
Since Gjk is parallel to on , Gjk = 0 by (3.27). Thus we are free to substitute the free propagator Q1 for K 1 in the above expression, to obtain: Ji (x, x) = 2e2
ijk
d3 x tr (x)Gjk (x ) Q1 (x, x )Q1 (x , x) + O(|x|3 ) ,
where the free propagator is given by Q1 (x, y) = d3 p eip(xy) . (2)3 p2 + M 2 + a2 e2
Changing variables x y = x x , and using tr (x)Gjk (x y) = tr (x)Gjk (x) + O(|x|3 ) , we remain with Ji (x, x) = 2e2
ijk tr Gjk (x)(x)
d3 p d3 q i(pq)y e (2)3 (2)3 1 2 + O(|x|3 ) . 2 + a2 e2 )(q 2 + M 2 + a2 e2 ) (p + M d3 y
The y-integral gives 2 3 (p q), which gets rid of the q-integral and we remain with Ji (x, x) = 2e2
ijk
tr (x)Gjk (x)
1 d3 p + O(1/|x|3 ) . 3 (p2 + M 2 + a2 e2 )2 (2)
The p-integral is readily evaluated d3 p 1 1 1 = , 3 (p2 + M 2 + a2 e2 )2 2 + a2 e2 (2) 8 M 98
[email protected]
whence dSi Ji (x, x) =
1 e2 2 M 2 + a2 e2
dSi
ijk
Gjk ,
where have also used that the trace in the adjoint representation is normalised so that tr AB = 2A B.
Exercise 3.11 (Another expression for the degree of the map ) Prove that on , 1 Gjk = 2 (j k ) , ea and, comparing with equation (1.22), deduce that the degree of the map from to the sphere of radius a in R3 is given by deg = e 8a
dSi
ijk
Gjk .
(Hint: Use that Dj = 0 on , and expand 0 = (Dj Dk ).)
From the results of this exercise and the fact that for a k-monopole solution, the degree of is k, we can write dSi Ji (x, x) =
4aek , M 2 + a2 e2
whence plugging this into the equation (3.24) for the formal dimension of Mk , we nd that dim Mk = 4k .
If you are familiar with the calculation of the number of instanton parameters, you may be surprised by the explicit M 2 dependence in the expression dSi Ji (x, x). This is due to the asymptotic behaviour of , which prevents the above calculation from being tackled by the methods usually applied to index theorems on compact spaces. For the index of an operator on a compact space, or similarly for elds which decay at innity, one can prove that the result of the above integral is actually independent of M 2 , hence one can compute the integral for already in the limit M 2 0 (cf. the Witten index). Here we are in fact faced essentially with the calculation of the index of an operator on a manifold with boundary, for which a satisfactory Witten-index treatment is lacking, to the best of my knowledge.
3.3
A quick motivation of hyperkhler geoma etry
In the next section we will prove that the natural metric on Mk induced by the YangMillsHiggs functional is hyperkhler; but rst we will briey rea view the necessary notions from riemannian geometry leading to hyperkhler a manifolds. The reader familiar with this topic can easily skip this section. 99
[email protected]
Hyperkhler geometry is probably best understood from the point of view a of holonomy groups in riemannian geometry. In this section we review the basic notions. Sadly, the classic treatises on the holonomy approach to riemannian geometry [KN63, KN69, Lic76] stop just short of hyperkhler gea ometry; but two more recent books [Bes86, Sal89] on the subject do treat the hyperkhler case, albeit from a slightly dierent point of view than the one a adopted here. We direct the mathematically inclined reader to the classics for the basic results on riemannian and Khler geometry, which we will only a have time to review ever so briey in these notes; and to the newer references for a more thorough discussion of hyperkhler manifolds. All our manifolds a will be assumed dierentiable, as will be any geometric object dened on them, unless otherwise stated.
3.3.1
Riemannian geometry
Any manifold M admits a riemannian metric. Fix one such metric g. On the riemannian manifold (M, g) there exists a unique linear connection which is torsion-free [X, Y ] =
X
for any vector elds X, Y on M
(3.29)
and preserves the metric g = 0. It is called the Levi-Civita connection and relative to a local chart xa , it is dened by the Christoel symbols ab c which in turn are dened by c a b = ab c , where we have used the shorthand a = a . The dening properties of the Levi-Civita connection are sucient to express the Christoel symbols in terms of the components gab of the metric:
1 ab c = 2 g cd (a gdb + b gad d gab ) ,
(3.30)
which proves the uniqueness of the Levi-Civita connection.

Exercise 3.12 (A coordinate-free expression for ) Using the dening conditions of the Levi-Civita connection 2 Z,
XY
, prove that
= X Y, Z + Y Z, X Z X, Y [X, Z], Y [Y, Z], X [X, Y ], Z , (3.31)
where we have used the notation X, Y = g(X, Y ). Equation (3.30) follows after substituting a , b and c for X, Y , and Z respectively.
100
[email protected]
With we can give meaning to the notion of parallel transport. Given a curve t (t) on M with velocity vector , we say that a vector eld X is parallel along if X = 0. Relative to a local coordinate chart xa , we can write this equation as DX b a dt
aX b
= a a X b + ac b X c = X b + ac b a X c = 0 .

(3.32) = 0. In (3.33)
A curve is a geodesic if its velocity vector is self-parallel: terms of (3.32) we arrive at the celebrated geodesic equation: c + ab c a b = 0 . This equation follows by extremising the action with lagrangian L(x, x) = 1 gab (x) xa xb , 2
whence our claim at the end of section 3.1.3 that free motion on a riemannian manifold is geodesic. We can integrate equation (3.32) and arrive at the concept of paralleltransport. More concretely, associated with any curve : [0, 1] M there is a linear map P : T(0) M T(1) M taking vectors tangent to M at (0) to vectors tangent to M at (1). If X Tp M is a tangent vector to M at p (0) we dene its parallel transport P (X) relative to by rst extending X to a vector eld along in a way that solves (3.32), and then simply evaluating the vector eld at (1). Now x a point p M and let be a piecewise dierentiable loop based at p, that is, a piecewise dierentiable curve which starts and ends at p. Then P is a linear map Tp M Tp M . We can compose these maps: if and are two loops based at p, then P P is the linear map corresponding to parallel transport on the loop based at p obtained by rst tracing the path and then . (Notice that this new loop may not be dierentiable even if and are: but it is certainly piecewise dierentiable, hence the need to consider such loops from the outset.) Also P is invertible: simply trace the path backwards in time. Therefore the transformations {P } form a group. If we restrict ourselves to loops which are contractible, the group of linear transformations: H(p) = {P | a contractible loop based at p} is called the (restricted) holonomy group at p of the connection . It can be shown that to be a Lie group. One should hasten to add that there is no reason to restrict ourselves to the Levi-Civita connection. We will be mostly interested in the classical 101
[email protected]
case, where is the Levi-Civita connection, but these denitions make sense in more generality.
Exercise 3.13 (The holonomy group of a connected manifold) Prove that if two points p and q in M can be joined by a path in M , their holonomy groups H(p) and H(q) are conjugate and therefore isomorphic. (Hint: Use parallel transport along the path joining p and q to provide the conjugation.)
Hence it makes sense to speak of the holonomy group of a connected manifold M . From now on all we will only concern ourselves with connected manifolds. A further useful restriction that one can impose on the type of manifolds we consider is that of irreducibility. A manifold is said to be (ir)reducible relative to a linear connection if the tangent space at any point is an (ir)reducible representation of the holonomy group. Clearly the holonomy group (relative to the Levi-Civita connection) of product manifold M M with the product metric acts reducibly. A famous theorem of de Rhams provides a converse. This theorem states that if a simplyconnected complete riemannian manifold M is reducible relative to the LeviCivita connection, then M = M M isometrically. We will restrict ourselves in what follows to irreducible manifolds. For a generic linear connection on an irreducible manifold M , the holonomy group is (isomorphic to) GL(m), where m = dim M . However, the Levi-Civita connection is far from generic as the following exercise shows.
Exercise 3.14 (The holonomy group of a riemannian manifold) Prove that the holonomy group of an m-dimensional riemannian manifold (relative to the Levi-Civita connection) is actually in SO(m). (Hint: Show that g = 0 implies that the parallel transport operation P preserves the norm of the vectors, whence the holonomy group is in O(m). Argue that since we consider only contractible loops, the holonomy group is connected and hence it must be in SO(m). By the way, the same would hold for orientable manifolds even if considering non-contractible loops.)
A celebrated theorem of Ambrose and Singer tells us that the Lie algebra of the holonomy group is generated by the Riemann curvature tensor in the following way. Recall that the Riemann curvature tensor is dened as follows. Fix vector elds X and Y on M , and dene a linear map from vector elds to vector elds as follows: R(X, Y ) = [
X, Y]
[X,Y ]
It is easy to prove that this map is actually tensorial in X and Y . Indeed, relative to a coordinate basis, it may be written out as a tensor Rabc d dened 102
[email protected]
by R(a , b ) c = Rabc d d , and therefore has components Rabc d = a bc d + bc e ae d b ac d ac e be d . (3.34)
Then the Lie algebra of the holonomy group is the Lie subalgebra of gl(m) spanned by the curvature operators Rab : c Rabc d d .
Exercise 3.15 (The holonomy algebra of a riemannian manifold) Using the AmbroseSinger theorem this time, prove a second time that the holonomy group of a riemannian manifold lies in SO(m), by showing that its Lie algebra lies in so(m). In other words, prove that each curvature operator Rab (for xed a and b) is antisymmetric: Rabcd = Rabdc where Rabcd = Rabc e ged .
Adding more structure to a riemannian manifold in a way that is consistent with the metric restricts the holonomy group further. Next we will discuss what happens when we add a complex structure.
3.3.2
Khler geometry a
An almost complex structure is a linear map I : T M T M which obeys I 2 = 1. This gives each tangent space Tp M the structure of a complex vector space, since we can multiply a tangent vector X by a complex number z = x + i y simply by z X = x X + y I(X). In particular, it means that the (real) dimension of each Tp M and hence of M must be even: 2n, say. We will also assume that the complex structure I is compatible with the metric in the sense that g(IX, IY ) = g(X, Y ) for all vector elds X and Y . Another way to say this is that the metric g is hermitian relative to the complex structure I. If we complexify the tangent space, we can diagonalise the complex structure. Clearly the eigenvalues of I are i. Complex vector elds Z for which I Z = iZ are said to be of type (1, 0), whereas those for which I Z = iZ are of type (0, 1). If we can introduce local complex coordinates (z , z ), , = 1, . . . n, relative to which a basis for the (1, 0) (resp. (0, 1)) vector elds is given by (resp. ) and if when we change charts the local com plex coordinates are related by biholomorphic transformations, then we say that I is integrable. A hard theorem due to Newlander and Nirenberg translates this into a beautiful local condition on the complex structure. According to the 103
[email protected]
NewlanderNirenberg theorem, an almost complex structure I is integrable if and only if the Lie bracket of any two vector elds of type (1, 0) is again of type (1, 0). This in turns translates into the vanishing of a tensor.
Exercise 3.16 (The Nijenhuis tensor) Using the NewlanderNirenberg theorem prove that I is integrable if and only if the following tensor vanishes: NI (X, Y ) = I[IX, IY ] + [X, IY ] + [IX, Y ] I[X, Y ] . NI is known as the Nijenhuis tensor of the complex structure I. It is easy to prove that the Nijenhuis tensor NI vanishes in a complex manifold (do it!)it is the converse that is hard to prove.
Now suppose that is a linear connection relative to which I is parallel: I = 0. Lets call this a complex connection.
Exercise 3.17 (The holonomy group of a complex connection) Let H denote the holonomy group of a complex connection on a complex manifold M . Prove that H GL(n, C) GL(2n, R). (Hint: Its probably easiest to prove the equivalent statement that the holonomy algebra is a subalgebra of gl(n, C). Choose a basis for Tp M R2n in which the = complex structure I has the form I= 0 1 1 0 ,
where 1 is the nn unit matrix. Argue that the curvature operators Rab commute with I, whence in this basis, they are of the form A B B A ,
where A and B are arbitrary real n n matrices. This then corresponds to the real 2n-dimensional representation of the matrix A + i B gl(n, C).)
If is the Levi-Civita connection, then the holonomy lies in the intersection GL(n, C) SO(2n) GL(2n, R).
Exercise 3.18 (The unitary group) Prove that GL(n, C) SO(2n) GL(2n, R) is precisely the image of the real 2ndimensional representation of the unitary group U (n). (Hint: Prove the equivalent statement for Lie algebras. In the basis of the previous exercise, prove that a matrix in so(2n) has the form A B B t D ,
104
[email protected]
where At = A, Dt = D and B are otherwise arbitrary real n n matrices. If the matrix is also in gl(n, C), we know that A = D and that B = B t . Thus matrices in gl(n, C) so(2n) gl(2n, R) are of the form A B B A ,
where At = A and B t = B, which corresponds to the complex matrix A + i B gl(n, C). Prove that this matrix is anti-hermitian, whence in u(n).)
If the Levi-Civita connection is complex, so that the holonomy lies in U (n), the manifold (M, g, I) is said to be Khler. In other words, Khler a a geometry is the intersection, so to speak, of riemannian and complex geometries. There is another perhaps more familiar denition of Khler manifolds, a involving the Khler form. a
Exercise 3.19 (The Khler form) a Given a complex structure I relative to which g is hermitian, we dene a 2-form by (X, Y ) = g(X, IY ) or equivalently ab = Ia c gbc . Prove that (X, Y ) = (Y, X), so that it is in fact a form. Prove that it is actually of type (1, 1).
An equivalent denition of a Khler manifold is that (M, g, I) is Khler a a if and only if is closed. These two denitions can of course be reconciled. We review this now.
Exercise 3.20 (Khler is Khler is Khler) a a a Let (M, g, I) be a complex riemannian manifold, where g is hermitian relative to I. Let be the corresponding Khler form and the Levi-Civita connection. Prove a that the following three conditions are equivalent (and are themselves equivalent to (M, g, I) being a Khler manifold): a (a) (b) I = 0; = 0; and
(c) d = 0. (Hint: (a) (b) is obvious. For (b) (c) simply antisymmetrise and use the fact that is torsion-less, which implies the symmetry of the Christoel symbols in the lower two indices. The trickiest calculation is (c) (b), and we break this up into several steps: 1. From (3.29), deduce that I is integrable if and only if
a bc
+ Ic d Ib e
e ad
(a b) = 0 .
105
[email protected]
2. From the fact that g is hermitian relative to I, deduce that Ic d

a bd
= Ib d
a cd
3. Using the previous two steps, show that

a bc
= Ib d Ic e
a de
. = 0.)
4. Finally, use these formulae to show that d = 0
Conversely, one can show that if the holonomy group of a 2n-dimensional riemannian manifold is contained in U (n), then the manifold is Khler. The a proof is paradigmatic of the more algebraic approach to the study of holonomy, which has begotten some of the more remarkable results in this eld. We will therefore allow ourselves a brief digression. We urge the reader to take a look at the books [Bes86, Sal89] for a more thorough treatment. For simplicity, we start with a torsionless connection . The fundamental elementary observation is that there is a one-to-one correspondence between covariantly constant tensors and singlets of the holonomy group. (Clearly if t = 0, then t is invariant under the holonomy group; conversely, if t is invariant under the holonomy group, taking the derivative of the parallel transport of t along the path is zero, but to rst order this is precisely t.) In turn, singlets of the holonomy group determine to a large extent the geometry of M . For example, suppose that M is an m-dimensional irreducible manifold with holonomy group G GL(m, R). Irreducibility means that the fundamental m-dimensional representation of GL(m, R) remains irreducible under G. Let us call this representation T the T stands for tangent space. Under the action of G, tensors on M will transform according to tensor powers of the representation T . For example, 1-forms will transform according to the dual representation T , symmetric rank p tensors will transform as S p T , whereas p-forms will transform as p T , and so on. We can then break up all these tensorial representations in terms of irreducibles and, in particular, exhibit all the singlets. These singlets will correspond, by the observation made above, in a one-to-one fashion with covariantly constant tensors. Let us run through some examples. Suppose that G = SO(m). Then T is the fundamental m-dimensional representation of SO(m). We know that, in particular, there is a singlet g S 2 T and moreover that the map T T dened by this g is non degenerate. Hence by the fundamental observation, there exists a covariantly constant tensor g which can be thought of as a riemannian metric. By the uniqueness of the Levi-Civita connection, it follows that is the Levi-Civita connection associated to g. In other words, manifolds with SO(m) holonomy 106
[email protected]
relative to a torsionless connection are simply riemannian manifolds. Well, not just any riemannian manifold. There is another SO(m)-invariant tensor m T . The covariantly constant m-form denes an orientation on M . In fact, one can show that there are no other invariant tensors which are algebraic independent from these ones, so that manifolds with SO(m) holonomy (again, relative to a torsionless connection) are precisely orientable riemannian manifolds. Now suppose that the dimension of M is even: m = 2n, say, and that G = U (n) GL(2n, R). Then T is the real 2n irreducible representation of U (n), whose complexication breaks up as T C = T T , where T is the complex n-dimensional (fundamental) representation of U (n), and T = T is its conjugate. Since G SO(2n), we know from the previous paragraph that M is riemannian and orientable, and that we can think of as the LeviCivita connection of this metric. However there is also a singlet 2 T . The resulting covariantly constant 2-form is precisely the Khler form. a Hence manifolds with U (n)-holonomy are precisely the Khler manifolds. a
3.3.3
Ricci atness
We can now restrict the holonomy of a Khler manifold a little bit further by a imposing constraints on the curvature: namely that it be Ricci-at. As we will see, this is equivalent to demanding that the holonomy lie in SU (n) U (n). As Lie groups, U (n) = U (1) SU (n). If we think of U (n) as unitary matrices, the U (1) factor is simply the determinant. Hence the manifold will have SU (n) holonomy provided that the determinant of every parallel transport operator P is equal to 1. Geometrically, the determinant can be understood as follows. Suppose that M is a Khler manifold and lets look at how forms of type (n, 0) (or a (0, n)) transform under parallel transport. At a xed point p in M , the space of such forms is 1-dimensional. Hence if is an (n, 0)-form, then P = where is a complex number of unit norm.
Exercise 3.21 (The determinant of P ) Prove that is the determinant of the linear map
P : Tp M Tp M .
Therefore the holonomy lies in SU (n) if and only if = 1 for all . By our previous discussion, it means that there is a nonzero parallel (n, 0) form . Since parallel forms have constant norm, this form if nonzero at some point is nowhere vanishing, hence the bundle of (n, 0)-forms is trivial. Equivalently this means that the rst Chern class of the manifold vanishes. Such Khler manifolds are known as CalabiYau manifolds, after the celea brated conjecture of Calabi, proven by Yau. Calabis conjecture stated that 107
[email protected]
given a xed Khler manifold with vanishing rst Chern class, there exists a a unique Ricci-at Khler metric in the same Khler class. The Calabi cona a jecture (now theorem) allows us to construct manifolds admitting Ricci-at Khler metrics, by the simpler procedure of constructing Khler manifolds a a with vanishing rst Chern class. Algebraic geometry provides us with many constructions of such manifolds: as algebraic varieties of complex projective space, for example. The catch is that the Ricci-at Khler metric is most a denitely not the induced metric. In fact, the form of the metric is very difcult to determine. Even for relatively simple examples like K3, the metric is not known. It follows from this conjecture (now theorem) that an irreducible Khler a manifold has SU (n) holonomy if and only if it is Ricci-at. We dont need to appeal to the Calabi conjecture to prove this result, though, as we now begin to show. Lets rst recall how the Ricci tensor is dened. If X and Y are vector elds on M , Ric(X, Y ) is dened as the trace of the map V R(V, X) Y , or relative to a local chart Sab Ric(a , b ) = Rcab c .
Exercise 3.22 (The Ricci tensor is symmetric) Prove that Sab = Sba .
In a Khler manifold and relative to complex coordinates adapted to I, a many of the components of the Ricci and Riemann curvature tensors are zero.
Exercise 3.23 (Curvature tensors in Khler manifolds) a , z ) be complex coordinates adapted to the complex structure I; that is, the Let (z corresponding vector elds are of type (1, 0) and (0, 1) respectively: I( ) = i and I( ) = i . Prove that the metric has components g , and that the Christoel symbols have components and . Prove that the only nonzero components of the Riemann curvature are R = R and R = R . Finally deduce that the Ricci tensor has components S , so that S = S = 0.

The holonomy algebra of a Khler manifold is u(n), whence for xed a a and b, the curvature operator Rab belongs to u(n) = u(1) su(n). How can one extract the u(1)-component? For this we need to recall Exercise 3.18. If A + i B u(n), exp(A + iB) U (n) and the U (1) component is the determinant: det exp(A + i B) = exp(tr(A + i B)). Since At = A, it is traceless, whence det exp(A + i B) = exp(i tr B). Hence if we let i1 u(n) be the generator of the u(1) subalgebra, the u(1) component of a matrix in u(n) is just its trace. In a Khler manifold, the holonomy representation of a 108
[email protected]
U (n) is real and 2n-dimensional, which means that a matrix A + i B u(n) is represented by a real 2n 2n matrix Q= A B B A ,
and therefore its u(1)-component, tr B, is simply given by tr B = 1 tr 2 0 1 1 0 A B B A

1 = 2 tr(I Q) .
From this it follows that the u(1)-component Fab of the curvature operator is given by Fab = 1 tr(I Rab ) = 1 Rabc d Id c . The next exercise asks you to 2 2 show that this is essentially the Ricci curvature, from where it follows that Ricci-at Khler manifolds have SU (n) holonomy and viceversa. a
Exercise 3.24 (An equivalent expression for the Ricci curvature) Prove that the Ricci curvature on a Khler manifold can be also be dened by a Ric(X, Y ) = or equivalently, Sac Ib c = 1 tr(I Rab ) = Fab , 2 which, relative to complex coordinates, becomes S = iF . Using the above results, give another proof of the symmetry of the Ricci tensor: Sab = Sba . (Compare with Exercise 3.22.)
1 2
tr(V I R(X, IY ) V ) ;
3.3.4
Hyperkhler geometry a
Finally we dene hyperkhler manifolds. In a hyperkhler manifold we have a a not one but three parallel almost complex structures I, J, and K which satisfy the quaternion algebra: IJ = K = JI, JK = I = KJ, KI = J = IK 2 2 2 I = J = K = 1 ,
and such that the metric is hermitian relative to all three. Notice that we dont demand that the complex structures be integrable. This is a consequence of the denition.
109
[email protected]
Exercise 3.25 (Hyperkhler implies integrability) a Let (M, g, I, J, K) be a hyperkhler manifold. Prove that I, J and K are integrable a complex structures. (Hint: Associated to each of the almost complex structures there is a 2-form: I , J and K . Because I = 0, Exercise 3.20 implies that dI = 0, and similarly for J and K. Notice that J (X, Y ) = g(X, JY ) = g(X, KIY ) = K (X, IY ), whence2 X J = IX K . A complex vector eld X is of type (1, 0) with respect to I, if and only if X J = i X K . By the NewlanderNirenberg theorem, it is sucient to prove that the Lie bracket of two such complex vector elds also obeys the same relation. But this is a simple computation, where the fact that the forms J and K are closed is used heavily. The same proof holds mutatis mutandis for J and K.)
Just like an almost complex structure I on a manifold allows us to multiply vector elds by complex numbers and hence turn each tangent space into a complex vector space, the three complex structure in a hyperkhler manifold a allow us to multiply by quaternions. Concretely, if q = x + i y + j z + k w H is a quaternion, and X is a vector eld on M , then we dene q X x X + y I(X) + z J(X) + w K(X) . This turns each tangent space into a quaternionic vector space (a left Hmodule, to be precise) and, in particular, this means that hyperkhler mana ifolds are 4k-dimensional. One can prove, just as we did with complex manifolds, that the holonomy group of a hyperkhler manifold lies in3 U Sp(2k) SU (2k) GL(4k). In a particular, hyperkhler manifolds are Ricci-at. a
Exercise 3.26 (The holonomy group of a hyperkhler manifold) a Prove that the holonomy group of a hyperkhler manifold is a subgroup of U Sp(2k). a (Hint: Depending on how one looks at this, there may be nothing that needs proving. If we take as denition of U Sp(2k) GL(2k, C) those matrices which commute with the natural action of the quaternions on C2k Hk , then the result = is immediate since the fact that I, J, and K are parallel means that and hence the curvature operators commute with multiplication by H. If you have another denition of U Sp(2k) in mind, then the exercise is to reconcile both denitions.)
Conversely, if the holonomy group of a manifold M is a subgroup of U Sp(2k) SO(4k), then decomposing tensor powers of the fundamental 4kdimensional representation T of SO(2k) into U Sp(2k)-irreducibles, we nd that 2 T possesses three singlets. The resulting covariantly constant 2forms are of course the three Khler forms of M . A little bit closer inspection a
2 3
The conventions for the interior product X are summarised in Exercise 3.27. Mathematicians call Sp(k) what we call U Sp(2k).
110
[email protected]
shows that the associated complex structures obey the quaternion algebra, so that M is hyperkhler. a
3.4
Mk is hyperkhler a
In this section we prove that the metric on Mk dened by the kinetic term in the YangMillsHiggs functional is hyperkhler. We will prove this in two a ways. First we will prove that the conguration space A is hyperkhler and a that Mk is its hyperkhler quotient. In order to do this we will review the a notion of a Khler quotient which should be very familiar as a special case a of the symplectic quotient of MarsdenWeinstein which appears in physics whenever we want to reduce a phase space with rst-class constraints. In a nutshell, the group G of gauge transformations acts on A preserving the hyperkhler structure and the resulting moment mapping is nothing but a the Bogomolnyi equation. We will also give a computationally more involved proof of the hyperkhler nature of Mk , which is independent of the a hyperkhler quotient, at least on the face of it. a
3.4.1
Symplectic quotients
Let (M, ) be a symplectic manifoldthat is, is a nondegenerate closed 2-formand let G be a Lie group acting on M in a way that preserves . Let g denote the Lie algebra of G and let {ea } be a basis for g which we x once and for all. To each ea there is an associated vector eld Xa on M . The fact that G preserves means that LXa = 0 for every Xa , where LXa is the Lie derivative along Xa . We will often abbreviate LXa by La .
Exercise 3.27 (The Lie derivative acting on forms) Prove that if is a dierential form on M and X is any vector eld, then the Lie derivative LX is given by LX = (dX + X d) where d is the exterior derivative and X is the contraction operator characterised uniquely by the following properties: (a) X f = 0 for all functions f ; (b) X = (X) for all one-forms ; and (c) X ( ) = (X ) + (1)p X , for a p-form and any form.
111
[email protected]
Because d = 0, La = 0 is equivalent to the one-form a being closed, where a Xa . Let us assume that this form is also exact, so that there is a function a such that a = da . This would be guaranteed, for example, if M were simply-connected, or if g were semi-simple. More precise conditions on the absence of this obstruction can be written down but we wont need them in what follows. The functions a allow us to dene a moment mapping : M g by (p) = a (p)ea for every p M , where {ea } is the canonically dual basis for g . In other words, (p)(ea ) = a (p).
Exercise 3.28 (The Poisson bracket) Prove that the Poisson bracket dened by: {f, g} = ij i f j g , where ij jk = k i , is antisymmetric and obeys the Jacobi identity. Using the above denition of the Poisson bracket (or otherwise) prove the following identity: d{a , b } = fab c dc , (3.35)
where fab c are the structure constants of g in the chosen basis. (Hint: Prove rst that Xa b = {a , b } and take d of this expression. You may wish you use the following properties of the contraction: [La , b ] = fab c c .)
Now notice that since a are dened by their gradients, they are dened up to a constant. If these constants can be chosen so that equation (3.35) can be integrated to {a , b } = fab c c , (3.36) then the moment mapping is equivariant and the action is called Poisson. In other words, the moment mapping intertwines between the action of G on M and the coadjoint action of G on g . Again one can write down precise conditions under which this is the caseconditions which would be met, for example, if g were semisimple. We will assume henceforth that the necessary conditions are met and that the moment mapping is equivariant. The components a of an equivariant moment mapping can be understood as rst-class constraints. It is well-known that if the constraints are irreducible, so that their gradients are linearly independent almost everywhere on the constraint submanifold, one can reduce the original symplectic manifold to a smaller symplectic manifold (or, more generally, orbifold). More precisely, the irreducibility condition on the constraints means that their zero locus 1 (0) is an embedded submanifold of M . The fact that the constraints are rst class means that the vector elds Xa , when restricted to 1 (0), are tangent to 1 (0). 112
[email protected]
Equivalently, one may deduce from the equivariance of the moment mapping that G acts on 1 (0). Provided that it does so nicely (that is, freely and properly discontinuous) the space 1 (0)/G of G-orbits is a manifold, and a celebrated theorem of Marsden and Weinstein tells us that it is symplectic. Indeed, if we let : 1 (0) 1 (0)/G denote the natural projection, then the MarsdenWeinstein theorem says that there is a unique symplectic form on 1 (0)/G such that its pullback to 1 (0) coincides with 1 the restriction to (0) of the symplectic form on M . The symplectic manifold (1 (0)/G, ) is known as the symplectic quotient of (M, ) by the action of G. It wouldnt be too dicult to sketch a proof of this theorem, but since we will only need the special case of a Khler quotient, we will omit it. a
3.4.2
Khler quotients a
Now suppose that (M, g, I) is Khler with Khler form . Then in particular a a (M, ) is symplectic. Assume that the action of G on M is not just Poisson, but that G also acts by isometries, that is, preserving g. Because G preserves both g and , it also preserves I. On 1 (0) we have the induced metric: the restriction to 1 (0) of the metric on M . This gives rise to a a metric on 1 (0)/G which we will discuss below. For every p 1 (0), the tangent space Tp M breaks up as Tp M = Tp 1 (0) Np 1 (0) , where Tp 1 (0) is the tangent space to 1 (0) and the normal space Np 1 (0) is dened as its orthogonal complement (Tp 1 (0)) . Globally this means that the restriction to 1 (0) of the tangent bundle of M decomposes as: T M = T 1 (0) N 1 (0) , (3.37)
where the normal bundle N 1 (0) is dened as (T 1 (0)) . As the next exercise shows, the normal bundle is trivial because 1 (0) is dened globally by irreducible constraints.
Exercise 3.29 (Triviality of the normal bundle) Prove that the normal space Np 1 (0) is spanned by the gradients gradp a of the constraints; or globally, that the gradients of the constraints {grad a } trivialise the normal bundle. In fact, the converse is also true. If you feel up to it, prove that the normal bundle to a submanifold is trivial if and only if the submanifold can be described globally as the zero locus of some irreducible constraints. (Hint: A vector eld X is tangent to 1 (0) if and only if it preserves the constraints: da (X) = 0, but this is precisely g(grad a , X) = 0, by denition of grad a .)
113
[email protected]
Both the metric and the symplectic form restrict to 1 (0), but whereas g is nondegenerate on 1 (0), the symplectic form isnt. Thus in order to obtain a Khler manifold it is necessary to perform a quotient. We will a describe this quotient locally. To this eect, let us split the tangent space Tp 1 (0) further as: Tp 1 (0) = Hp Vp , where the vertical vectors Vp are those vectors tangent to the G-orbits and the horizontal vectors Hp = Vp are their orthogonal complement. The vertical subspace is spanned by the Killing vectors Xa , whereas the horizontal space Hp can be identied with the tangent space to 1 (0)/G at the G-orbit of p. Indeed, given any vector eld X on 1 (0)/G we dene its horizontal lift to be the unique horizontal vector eld X on 1 (0) which projects down to = X. X: X Now given any two vector elds on 1 (0)/G, we dene their inner product to be the inner product of their horizontal lifts. This is independent on the point in the orbit to where we lift, because the metric is constant on the orbits. Hence it is a well-dened metric on 1 (0)/G. In fancier language, this is the unique metric on 1 (0)/G which makes the projection a riemannian submersion. (The reader will surely recognise this construction as the one which in section 3.1.2 yielded the metric on the physical conguration space C of the YangMillsHiggs system.) We claim that there is also a symplectic form on 1 (0)/G which makes this metric Khler. We prefer to work with the complex structure. a By denition, if Y is any vector eld tangent to M , its inner product with grad a is given by g(grad a , Y ) = da (Y ) = (Xa , Y ) = g(IXa , Y ) , whence grad a = IXa . Hence if we decompose the restriction of T M to 1 (0) as T M = T 1 (0) N 1 (0) = H V N 1 (0) , and we choose as bases for V and N 1 (0), {Xa } and {grad a } respectively, the complex structure I has the following form: I 0 0 I = 0 0 1 , 0 1 0 whence H is a complex subspace relative to the restriction I of I. In other words, the complex structure commutes with the horizontal projection, or a 114
[email protected]
little bit more precisely, if Y is a vector eld on 1 (0)/G and Y its horizontal lift, then I Y = IY . The next exercise asks you to prove that this complex structure is integrable, whence 1 (0)/G is a complex manifold.
Exercise 3.30 (Integrability of the restricted complex structure) Use the NewlanderNirenberg theorem to deduce that I is integrable. to that of I, which vanishes since I is (Hint: Relate the Nijenhuis tensor NI of I integrable.)
To prove that I is parallel, we need to know how the Levi-Civita connection of 1 (0)/G is related to the one on M . The next exercise asks you to prove the relevant relation.
Exercise 3.31 (ONeills formula) Let X and Y be vector elds on 1 (0)/G, and let X and Y be their horizontal lifts. Prove the following formula
XY
= X Y + 1 [X, Y ]v , 2
(3.38)
where is the Levi-Civita connection on 1 (0)/G, and v denotes the projection onto the vertical subspace. In other words, the horizontal projection of X Y is XY . precisely the horizontal lift of (Hint: Use expressions (3.31) and (3.29) to evaluate the horizontal and vertical components of X Y .)
In other words, formula (3.38) says that if we identify H with the tangent space to 1 (0)/G, then the Levi-Civita connection on 1 (0)/G is given simply by the horizontal projection of the Levi-Civita connection on M . Or, said dierently, that the covariant derivative commutes with the horizontal projection. Since the complex structures also commute with the projection, we see that I = 0 on M implies that I = 0 on 1 (0)/G. Therefore using 1 Exercise 3.20, (0)/G is Khler. a Notice that the above decomposition (3.37) can be thought of as T M T (1 (0)/G) gC , = where gC is the complexication of the Lie algebra of G. Therefore, morally speaking, it would seem that 1 (0)/G is the quotient of M by the action of GC . In some circumstances this is actually an accurate description of the Khler quotient; for instance, the construction of complex projective space a CPn as a Khler quotient of Cn+1 . a
115
[email protected]
3.4.3
Hyperkhler quotients a
Now let (M, g, I, J, K) be a hyperkhler manifold. We have three Khler a a (I) (J) (K) forms: , , and . Suppose that G acts on M via isometries and preserving the three complex structures, hence the three Khler forms. Asa sume moreover that the action of G gives rise to three equivariant moment mappings: (I) , (J) and (K) ; which we can combine into a single map : M g R 3 . Equivariance implies that 1 (0) is acted on by G. Assuming that 1 (0)/G is a manifold, we claim that it is actually hyperkhler. a Fix one of the complex structures, I, say; and consider the function = (J) + i(K) : M g C . For each Killing vector eld Xa and any vector eld Y , da (Y ) = (J) (Xa , Y ) + i (K) (Xa , Y ) = g(JXa , Y ) + ig(KXa , Y ) da (IY ) = g(JXa , IY ) + ig(KXa , IY ) = g(KX, Y ) + ig(JX, Y ) , whence da (IY ) = ida (Y ) ; or in other words, a = 0, so that is a holomorphic function (relative to I). This means that 1 (0) is a complex submanifold of a Khler manifold a 1 and hence its induced metric is Khler. Now G acts on (0) in such a way a that it preserves the Khler structure, and the resulting moment mapping a is clearly the restriction of (I) to 1 (0). We may therefore perform the Khler quotient of 1 (0) by the action of G, and obtain a manifold: a 1 (0) ((I) )1 (0)/G = 1 (0)/G , whose metric is Khler relative to (the complex structure induced by) I. To a nish the proof that 1 (0)/G is hyperkhler, we repeat the above for J and a K. This construction is called the hyperkhler quotient, and was described a for the rst time in [HKLR87].
3.4.4
Mk as a hyperkhler quotient a
Now we will prove that Mk is a hyperkhler quotient of the conguraa tion space Ak of elds Wi corresponding to nite-energy congurations with monopole number k. We can think of Wi as maps R3 R4 so(3), and R4 116
[email protected]
can be thought of as a quaternionic vector space in two inequivalent ways: we rst identify R4 = H, but then we have to choose whether H acts by left or right multiplication. Since quaternionic multiplication is not commutative, the two actions are dierent. Since we will be dealing with monopoles, we choose the right actionleft multiplication would correspond to antimonopoles. Let I, J, and K denote the linear maps R4 R4 representing right multiplication on H R4 by the conjugate quaternion units i, j, = and k respectively. The next exercise asks you to work out the explicit expressions for I, J, and K relative to a chosen basis.
Exercise 3.32 (Hyperkhler structure of R4 ) a Choose a basis {1, i, j, k} for H. Then relative to this basis, prove that the linear maps I, J, and K are given by the matrices: I= i 2 0 0 i 2 J= 0 1 1 0 K = IJ = 0 i 2 i 2 0 .
Notice that together with the euclidean metric on R4 , I, J, K make R4 into a (linear) hyperkhler manifold. a (Hint: Remember that the matrix associated to a linear transformation is dened by Iei = ej Iji . This choice makes composition of linear transformations correspond with matrix multiplication.)
We may now dene a hyperkhler structure on Ak as follows. If Wi is a a vector eld on Ak , then we dene (I W)i (x) = Iij Wj (x) , and similarly for J and K. Clearly they obey the quaternion algebra I J = K, etc because I, J and K do. Moreover since they are constant (and so is the metric) they are certainly parallel relative to the Levi-Civita connection on Ak with the metric given by the YangMillsHiggs functional. Hence Ak is an innite-dimensional (ane) hyperkhler manifold. a Let G denote the group of nite-range time- and x4 -independent gauge transformations. Since the metric is gauge invariant, G acts on Ak via isometries. We also claim that G preserves the three complex structures and gives rise to three equivariant moment mappings. In fact, we will prove this in one go by constructing the moment mappings from the start. The Killing vectors of the G action are just the innitesimal gauge transformations and they are parametrised by square-integrable functions : R3 so(3). The resulting Killing vector eld is X Wi = Di . For every such , dene the following function: (I) =
1 2
d3 x Iij (Gij ) , 117
[email protected]
and the same for J and K. be any tangent vector eld on Ak . (Here and in the sequel we Now let W will suppress the indices i whenever they dont play a role in an expression.) Then
(I) (W) = (I) (D , W)
= g(ID , W) = = d3 x Iij Dj Wi d3 x Iij Di Wj (integrating by parts)
= d(I) (W) .
Hence,
(I) = d(I) . Naturally, the same holds also for J and K. Hence we can construct a (I) (J) (K) moment mapping such that = ( , , ). The next exercise asks you to prove that it is equivariant.
Exercise 3.33 (Equivariance of the moment mapping) Prove that the moment mapping = ((I) , (J) , (K) ) is equivariant. In other words, if and are gauge parameters, prove that X = d (X ) = and the same for J and K.
(I) (I) (I)
Therefore we can apply the preceding discussion about the hyperkhler a 1 quotient to deduce that (0)/G is a hyperkhler manifold. But what is a (I) 1 1 (0)? Congurations Wi belonging in (0) are those for which = 0 for all , and the same for J and K. Since is arbitrary, this is equivalent to demanding that Iij Gij = 0, and the same for J and K. From the explicit expressions for the matrices I, J, and K found in Exercise 3.32, we nd that Iij Gij = 0 G12 = G34 , Jij Gij = 0 G13 = G24 , Kij Gij = 0 G14 = G23 . But these make up precisely the self-duality condition on Gij , that is, the Bogomolnyi equation! 118
[email protected]
Therefore 1 (0) is the submanifold of static solutions of the Bogomolnyi equation with monopole number k (the BPS-k-monopoles) and 1 (0)/G is their moduli space Mk . In summary, Mk is a 4k-dimensional hyperkhler a manifold, obtained as an innite-dimensional hyperkhler quotient of Ak by a the action of the gauge group G. This proof, although conceptually clear and oering a natural explanation of why Mk should be a hyperkhler manifold in the rst place, relies a rather heavily on dierential geometry. Therefore a more pedestrian proof might be helpful, and we now turn to one such proof.
3.4.5
Another proof that Mk is hyperkhler a
We start by expanding the YangMillsHiggs action in terms of collective coordinates in order to obtain an expression for the metric. Let X a , a = 1, . . . , 4k, denote the collective coordinates on the moduli space Mk of BPSk-monopoles. Let Wi (x, X(t)) be a family of BPS-monopoles whose t-dependence is only through the t-dependence of the collective coordinates; that is, Wi = a Wi X a . (3.39) Notice that a Wi need not be perpendicular to the gauge orbits. Indeed, generically, we will have a decomposition a Wi = a Wi + Di
a
(3.40)
where a Wi is the component perpendicular to the gauge orbits and Di a is the component tangent to the gauge orbits, hence an innitesimal gaugetransformation. The gauge parameters a are determined uniquely by a Wi . Indeed, simply apply Di and use the fact that D2 Di Di is negative-denite (hence invertible) to solve for a :
a
= D2 Di a Wi .
Exercise 3.34 (The YangMillsHiggs eective action) Compute the eective action for such a conguration of BPS-monopoles, and show that provided one sets W0 = X a a , it is given by 4a 1 |k| , Le = 2 gab X a X b e where the metric on Mk is given by gab = d3 x a Wi b Wi , (3.41)
where, by construction, a Wi are perpendicular to the gauge orbits and satisfy the linearised Bogomolnyi equation.
119
[email protected]
One way to systematise the above expansion is in terms of t-derivatives. The zeroth order term is given by the potential, which is a constant since the motion is purely along the at directions. The rst order term vanishes due to our choice for W0 ,4 while the quadratic term, which describes the motion of such monopoles in the limit of zero velocity, corresponds precisely to geodesic motion on Mk relative to the induced metricthat is, as a particle moving freely on Mk or, somewhat pedantically, as a (1 + 0)-dimensional -model with Mk as its target space. It is convenient to think of a as the components of a connection. We dene Da a e( a ), whence we can think of (Wi , a ) as the components of a connection on R4 Mk . This allows us to interpret a Wi as the mixed components of the curvature: Gai = a Wi i
a
W i = a W i D i
= a Wi .
The other components Gab of the curvature may be formally computed from the Bianchi identity: Di Gab = 2D[a b] Wi = Da b Wi + Db a Wi , by applying Di and inverting D2 as before.
Exercise 3.35 (A somewhat more explicit formula for Gab ) Prove that Gab = 2eD2 (a Wi b Wi ) . (Hint: Apply Di to (3.42), and use that a Wi is perpendicular to the gauge orbits.)
(3.42)
Using these formulae it is possible to write a formal expression for the Christoel symbols of the Levi-Civita connection. Naturally this is left as an exercise.
Exercise 3.36 (The Christoel symbols) Prove that abc = gcd ab d = d3 x Da b Wi c Wi . (3.43)
Notice that abc = bac , since D[b c] Wi = 1 Di Gab which is orthogonal to c Wi . 2 (Hint: Use the explicit expressions (3.30) and (3.41) and compute.)
Using the explicit expressions found in Exercise 3.32 for the hyperkhler a 4 structure in R we now dene the following two-forms on Mk : ab =
4
(I)
d3 x Iij a Wi b Wj ,
Notice that W0 is not zero for generic choices of Wi (x, X(t)), but it can be made to vanish after a t-dependent gauge transformation.
120
[email protected]
and similarly for J and K, and their corresponding almost complex structures Ia b = g bc ac
(I)
Ja b = g bc ac
(J)
Ka b = g bc ac .
(K)
Exercise 3.37 (Explicit expressions for the complex structures) (a) Prove that Iij a Wj is orthogonal to the gauge orbits, and the same for J and K. (b) Then derive the following formula: Ia b b Wi = Iij a Wj , and the same for Ja b and Ka b . (c) Using these expressions, prove that Ia b , Ja b and Ka b obey the quaternion algebra. (Hints: (a) This is equivalent to the linearised Bogomolnyi equation, in the form Iij Gij = 0, etc. (b) Argue that since Ia b b Wi is orthogonal to the gauge orbits, d3 x Ia b b Wi c Wi denes it uniquely. Then just compute the integral and use (a).) (3.44)
We claim that the forms (I) , (J) and (K) are parallel. Lets see this for one of them (I) , the other cases being identical. By denition,
a bc
= a bc ab d dc ac d bd .
We now compute this in steps. First of all we have: a bc = d3 x Iij Da b Wi c Wj + b Wi Da c Wj . (3.45)
Now we notice that ab d dc = Ic e abe . Using the explicit expression (3.43), we arrive at ab d dc = Ic e d3 x Da b Wi e Wi , (3.46) and using (3.44) we can rewrite this as ab d dc = d3 x Iij Da b Wi c Wj , (3.47)
with a similar expression for ac d bd = ac d db . Adding it all together we nd that a bc = 0. But this means that Ia b , Ja b , and Ka b are also parallel, whence Mk is hyperkhler. a
121
Chapter 4 The Eective Action for N =2 Supersymmetric YangMills

In this chapter we will perform the collective coordinate expansion of the N =2 supersymmetric SO(3) YangMills theory dened by equation (2.13). We will also discuss the quantisation of the corresponding eective action. As we saw in the discussion in section 3.1.5 on the eective theory for the 1monopole sector, the eective theory oers a qualitatively faithful description of the dyonic spectrum, even though quantitatively it is only an approximation. Of course, in the non-supersymmetric theory there is no reason to expect that the true quantum spectrum should resemble the classical spectrum given by the Bogomolnyi formula, but as we saw in Chapter 2, supersymmetry protects both the formula for the bound from quantum corrections and also the saturation of the bound. Hence it makes sense to expect that in the supersymmetric theory, the quantisation of the eective action should teach us something about the full quantum theory. As we shall soon discuss, this will have a chance of holding true only in the N =4 theory, but we can already learn something from the N =2 theory we have just studied. We will therefore rst discuss the fermionic collective coordinates and then the eective quantum theory in the k-monopole sector. We will see that there are 2k fermionic collective coordinates and that the resulting eective theory is to lowest order a (0 + 1) supersymmetric -model admitting N =4 supersymmetry consistent with the fact that Mk is hyperkhler. The quantisation a of this theory then leads to a geometric interpretation of the Hilbert space as the square-integrable (0, q)-forms on Mk , and of the hamiltonian as the laplacian. This chapter is based on the work of Gauntlett [Gau94].
122
[email protected]
4.1
Fermionic collective coordinates
As we saw in section 3.2, there are 4k bosonic collective coordinates in the k-monopole sector. The purpose of this section is to compute the number of fermionic collective coordinates: we will see that there are 2k of them. We will prove this in two ways. First we can set up this problem as the computation of the index of an operator, as we did for the bosonic collective coordinates; then essentially the same calculation that was done in section 3.2 yields the answer. Alternatively, and following Zumino [Zum77], we will exhibit a supersymmetry between the bosonic and fermionic collective coordinates which will also allow us to count them. Suppose that we start with an N =2 BPS-monopole obtained, say, in the manner of Exercise 2.17. Fermionic collective coordinates are simply fermionic at directions of the potential; that is, fermionic congurations which do not change the potential. In order to see what this means, let us rst write down the potential for a general eld conguration. To this eect let us break up the lagrangian density (2.13) into kinetic minus potential terms L = T V, where T= and V= Di P 2 + 1 Di S 2 + 1 Gij Gij + 1 e2 P S 2 4 2 5 P ie S . ie
1 2 2 1 2
G0i
1 2
D0 P
1 2
D0 S
+ i 0 D0
(4.1)
+ i i Di (4.2)
The potential is then the integral V = R3 V. For an N =2 BPS-monopole, with W0 = 0, S = , P = with 2 + 2 = 1, the potential is given by V =
1 2
Di
1 4
Gij Gij + i
(i Di e( + 5 )) ,
where in the last term is in the adjoint representation; that is, = . The rst two terms in the potential already reproduce the potential energy of a nonsupersymmetric BPS-monopole: 4a |k|, for a k-monopole. Therefore e turning on the fermions will not change the potential provided that the third term vanishes; in other words, provided that satises the Dirac equation in the presence of the BPS-monopole. In other words, fermionic collective coordinates are in one-to-one correspondence with zero modes of the Dirac operator. We will now count the number of zero modes in two ways.
123
[email protected]
4.1.1
Computing the index
In order to count the zero modes it is again convenient to use the reformulation of the BPS-monopole as an instanton, in terms of Wi = (Wi , ), and to dene the following four-dimensional euclidean Dirac matrices: i = 0 i and 4 = 0 ( + 5 ). In terms of these, the fermion term in the potential is given by i R3 i Di , keeping in mind that 4 0. We want to compute the number of normalisable solutions to the equation i Di = 0. Let us choose a Weyl basis in which 5 = 1 2 3 4 is diagonal. In such a basis, a convenient representation of the euclidean Dirac matrices is the one given in equation (3.19). In that representation the euclidean Dirac equation becomes: 0 ii Di e 1 ii Di e 1 0 + =0.
But notice that we have seen these very operators before, in the computation of the number of bosonic collective coordinates in section 3.2! In fact, in terms of the operator D dened in equation (3.17), the above Dirac equation breaks up into two equations, one for each chirality: D = 0 and D + = 0 . But now in Exercise 3.5 you showed that the operator DD is positive, whence it has no normalisable zero modes, hence neither does D . Therefore we notice that fermionic zero modes in the presence of a BPS-monopole necessarily have negative chirality. (For antimonopoles, it would have been D which has no normalisable zero modes, and fermion zero modes would have positive chirality.) We can arrive at the same result in a dierent way which doesnt use the explicit realisation of the i -matrices. In fact, it is an intrinsic property of fermions coupled to instantons (in four-dimensions). The next exercise takes you through it.
Exercise 4.1 (Fermion zero modes are chiral) Consider solutions of the four-dimensional euclidean Dirac equation i Di = 0 in the presence of an (anti)self-dual gauge eld. Prove that if the gauge eld is self-dual (respectively, antiselfdual), then fermion zero modes have negative (respectively, positive) chirality. (Hint: Compute the Dirac laplacian i j Di Dj and use the fact that D2 = Di Di is negative-denite and has not normalisable zero modes.)
Finally, just as in section 3.2.1, the number of normalisable zero modes of D is given by its index, which was computed in section 3.2.3 to be 2k, where k is the monopole number. 124
[email protected]
4.1.2
Using supersymmetry
We can reproduce this result in a dierent, but more useful way by exhibiting a supersymmetry between the bosonic and fermionic zero modes. This is based on work by Zumino [Zum77]. Let Wi be a bosonic zero mode; that is, Wi satises the linearised Bogomolnyi equation (3.15) and Gausss law (3.1). Let + be a constant, commuting spinor of positive chirality, normalised to + + = 1. Dene Wi i + . (4.3)
It is clear that has negative chirality and, as the next exercise asks you to show, satises the Dirac equation.
Exercise 4.2 (From bosonic to fermionic zero modes) Let Wi be a bosonic zero mode as above. With dened as above, prove that D = 0. (Hint: Use Exercise 3.4.)
Conversely, suppose that is a fermionic zero mode with negative chirality; that is, 5 = and D = 0. Then dene
Wi i+ i i i + .
The next exercise asks you to prove that Wi is a bosonic zero mode.
Exercise 4.3 (... and back) With Wi dened as above, prove that it satises the linearised Bogomolnyi equation (3.15) and Gausss law (3.1).
The above result seems to suggest that there is a one-to-one correspondence between the bosonic and fermionic zero modes, but this is ctitious, since not all the fermionic zero modes obtained in this fashion are independent. Indeed, as we will now see, they are related by the complex structure. Let a Wi for a = 1, . . . , 4k denote the bosonic zero modes, and let a = a Wi i + be the corresponding fermionic zero modes. We will prove that b Ia b = i a , where I is one of the complex structures of Mk . Hence comparing with the discussion at the end of section 3.2.1, we see that unlike bosonic zero modes, a and i a are not linearly independent. Let + be a commuting spinor of positive chirality normalised to + + = 1. Dene a 4 4 matrix A with entries
Aij + ij + .
(4.4)
We start by listing some properties of this matrix. 125
[email protected]
Exercise 4.4 (A complex structure) Let A be the 4 4 matrix with entries Aij given by (4.4). Prove that A satises the following properties: (1) A is antisymmetric; (2) iA is real;
1 (3) A is antiselfdual: Aij = 2 ijk
Ak ;
(4) A2 = 1, so that iA is a complex structure; and (5) Aij j + = i + . (Hint: This requires the Fierz identity:
1 + + = 4 (1 + 5 ) 1 Aij ij , 8
which you should prove.)
We will now prove that we can choose + in such a way that that iA agrees with any one of the complex structures on R4 dened in Exercise 3.32, and that such an + is unique up to a phase. We start by noticing that the 4 4 matrix iA dened above is real and antisymmetric, hence it belongs to so(4). As a Lie algebra, so(4) is isomorphic to so(3) so(3) (see Exercise 2.32). The fact that iA is antiselfdual, means that iA belongs to one of these so(3)s. In fact, it is the so(3) spanned by the complex structures I, J, and K of Exercise 3.32. (Check that they are antiselfdual!) In fact, R4 has a two-sphere worth of complex structures: {aI + bJ + cK|a2 + b2 + c2 = 1}, and from the above exercise, we see that iA denes a point in this two-sphere. In the next exercise we see this explicitly.
Exercise 4.5 (iA lives on the sphere) Compute the matrix iA explicitly in the representation of the Dirac matrices given by equation (3.19), and show that it is given by 0 q3 q2 q1 q 0 q1 q2 = q1 K q2 J + q3 I , (4.5) iA = 3 q2 q1 0 q3 q1 q2 q3 0 where I, J, and K are the complex structures in R4 dened in Exercise 3.32 and qi = i , where is a complex Weyl spinor normalised to = 1. (In the Weyl 2 basis above + = .) Prove that qi are real and that they satisfy i qi = 1, 0 hence iA denes a point in the unit two-sphere in R3 .
126
[email protected]
. Any other nor0 malised positive chirality spinor + will have the same form with replacing . This new Weyl spinor will be related to by an element of U (2): = U . The matrix iA obtained from has the form given by (4.5) but with qi replaced by qi i = U 1 i U . Now, in the Weyl basis introduced above, + =
Exercise 4.6 (Adjoint transformation) In the notation above, prove that qi = Uij qj , where Uij is the three-dimensional adjoint representation of U (2). (Notice that because the U (1) subgroup corresponding to the scalar matrices act trivially, only the SU (2) subgroup acts eectively in this representation.)
Therefore the action of U (2) on induces the adjoint action on the (qi ). 2 This action is transitive on the unit sphere i qi = 1, hence any two points (qi ) and (qi ) are related by an element of U (2). Notice that U (2) = SU (2) U (1) and that the SU (2) subgroup acts freely, whereas the U (1) acts trivially. Hence once a complex structure has been chosen, is unique up to the action of U (1); that is, a phase. Let us then exercise our right to choose + . We do so in such a way that
+ ij + = iIij ,
where Iij is given by Exercise 3.32. Then using equation (3.44), we have Ia b b = Ia b b Wi i + = Iij a Wj i + , = ia Wj j + , = i a , (4.6)
where in the next to last line we have used (5) in Exercise 4.5. Therefore there are only half as many linearly independent fermionic zero modes as there are bosonic ones, in agreement with the index calculation in the previous section.
4.2
The eective action
In this section we will write down the eective action governing the dynamics of the collective coordinates to lowest order. Let us introduce bosonic collective coordinates X a , for a = 1, . . . , 4k. These coordinates parametrise the moduli space Mk of BPS-monopoles with topological number k. In addition there will be fermionic collective coordinates a , for a = 1, . . . , 4k satisfying the condition a Ia b = ib . We now expand the supersymmetric YangMills action (2.13) in terms of the collective coordinates {X a , a } and keep only 127
[email protected]
the lowest nontrivial order. In order to count the order of an expression we take the conventions that a has order 1 , X a has order 0, but time derivatives 2 have order 1. These conventions are such that a free theory of bosons X a and fermions a is of quadratic order. We start by performing an SO(2) transformation which puts S = and P = 0, and choosing an appropriate parametrisation for the elds Wi , P, W0 , and in terms of the collective coordinates. As in the nonsupersymmetric theory, we leave W0 and, in this case also P, undetermined for the moment. We choose to parametrise Wi as Wi (x, X(t)), where all the time dependence comes from the collective coordinates. For this reason, equation (3.39) still holds where, as before, a Wi need not be perpendicular to the gauge orbits. Nevertheless we can project out a part which does: a Wi as in (3.40). Because a Wi is perpendicular to the gauge orbits, we know that a given by (4.3) is a fermion zero mode. We therefore parametrise = a a . Notice that since a is commuting (because so is + ), a is anticommuting as expected.
There is no reason, in principle, to expect Wi not to depend also on the fermionic collective coordinates. In fact, using that these are odd, we can expand Wi as follows: Wi (x, X, ) = Wi (x, X) + a Wi,a (x, X) + a b Wi,ab (x, X) + ; but it is not hard to see that all terms but the rst in the above expansion contribute only higher order terms to the eective action.
Because our choice of is a zero mode of the Dirac equation in the presence of a BPS-monopole, the discussion of section 4.1 applies and provided that P = 0, the potential remains at its minimum value: 4a k. However, P e need not remain zero for this to be case. It can evolve along a at direction as we now show. The P-dependent terms in the potential (4.2) can be written as follows: 1 Di P 2 + ie (P ) , 2
R3 R3
where we have used that 5 = 0 5 = . Integrating the rst term by parts, and using the invariance of the inner product in the second term, the above expression becomes: 1 2 where D2 = Di Di .
Exercise 4.7 (Computing ) Prove the following identity: 1 = a b D2 Gab , e
R
P D2 P + 2ie ,
128
[email protected]
where Gab are some of the components of the curvature of the connection (Wi , on R4 Mk , which were computed in Exercise 3.35. (Hint: You might want to use the identity:
+ i j + = ij + iIij ,
a)
(4.7)
which is (up to a factor) the projector onto the I-antiholomorphic subspace of the complexication C4 of R4 .)
Therefore we see that the condition that the potential remains constant demands that we either set P to zero or else P = 2ia b Gab . Next we tackle the kinetic terms. Notice that either of the choices for P allow us to discard P from the kinetic terms. Indeed, if P is nonzero, then the above expression shows that it is already of order 1 and hence its contribution to the kinetic term (4.1) will be of order higher than quadratic. Having discarded P from the kinetic terms, we remain with
1 2
R3
G0i
+i
R3
D0 .
The rst term is computed in the following exercise.

Exercise 4.8 (The rst kinetic term) Prove that the rst term in the kinetic energy above is given by
1 2
R3
G0i
= 1 gab X a X b + 2
1 2
R3
Di ( a X a W0 )
where gab was dened in (3.41). (Hint: Use the fact that a Wi is perpendicular to the gauge orbits!)
Finally we come to the second kinetic term. Plugging in the expression for and using equations (4.7), (3.44), and (4.6) we can rewrite the second term as i D0 = 2igab a b + 2ia b X c a Wi c b Wi
R3
R3
R3
+ 2iea b The next exercise nishes o the calculation.
W0 a Wi b Wj
129
[email protected]
Exercise 4.9 (... and the second kinetic term) Prove that the second kinetic term can be written as i D0 = 2igab a b + cd b X c d ia b W0
cX
c D2 Gab ,
where the Christoel symbols cd b were dened in equation (3.43).
Putting the results of Exercises 4.8 and 4.9, we nd that the kinetic terms of the action are (to lowest order) given by:
1 g X aX b 2 ab
+ 2igab a b + 2igab cd b a X c d ia b
R
W0
cX
c D2 Gab +
1 2
Di ( a X a W0 )
We see that we can cancel the last two terms provided that we set W0 =
aX
a 2ia b Gab .
With this choice, and to lowest order, the eective action then becomes: 4a Le = 1 gab X a X b + 2igab a b + cd b X c d k. 2 e (4.8)
Ignoring the constant term, this action describes a (0+1)-dimensional supersymmetric (as we shall see shortly) -model with target Mk .
4.3
N =4 supersymmetry of the eective action
In general, symmetries of the theory under consideration play important roles in the eective action. Broken symmetries give rise to collective coordinates, whereas unbroken symmetries remain symmetries of the eective action. As we have seen in section 2.3.4, N =2 BPS-monopoles preserve one half of the four-dimensional N =2 supersymmetry. This supersymmetry must be present in the eective action. In 0+1 dimensions, supersymmetry charges are one-component Majorana spinors, hence one supersymmetry charge in four dimensions gives rise to four supersymmetry charges in 0+1. In this section we will prove that the eective action given by (4.8) admits N =4 supersymmetry. From the proof it follows that the same is true for any supersymmetric -model with hyperkhler target manifoldwhich is, of a course, a well-known fact. 130
[email protected]
We start by discarding the constant term in the action (4.8) and rewriting the remaining terms in terms of complex coordinates adapted to the complex structure I of Mk . To this end we dene complex coordinates (Z , Z ) which diagonalise the complex structure; that is, such that I = i1 and I = i1 . As for the fermions, equation (4.6) implies that a Ia b = ib , whence = 0. Similarly, = 0. This prompts us to dene new fermions such that = 2 and = 2 . In terms of these new variables, the eective action remains
Le = g Z Z + ig + Z
(4.9)
where we have used that for a Khler metric in complex coordinates, the only a nonzero components of the metric and the Christoel symbols are g = g and and , as was proven in Exercise 3.23.
4.3.1
N =4 supersymmetry in R4 : a toy model
In order to understand the supersymmetry of the action (4.9), let us rst discuss the case of R4 with the standard euclidean at metric. We can think of R4 as C2 and introduce complex coordinates Z , Z where = 1, 2. The analogous action to (4.9) in this case is given simply by Le =
where and are the accompanying fermions. This toy action has four real supersymmetries. Two of them are manifest, as the next exercise asks you to show. Z Z + i
(4.10)
Exercise 4.10 (N =2 supersymmetry in at space) Let 1 and I be the supersymmetries dened as follows: 1 Z =
1 Z =
1 = iZ = iZ
1
I Z = i
I Z = i
I = Z I = Z .
(The names chosen for these transformations will appear more natural below.) Prove that they are invariances of the toy action (4.10), and that they satisfy the following algebra:
2 1 = I 2 = i
d dt
and
1 I = I 1 .
131
[email protected]
We can rewrite the second of these supersymmetries in a way that makes its generalisation obvious. If we let I denote the complex structure in R4 = C2 which is diagonalised by our choice of complex coordinates, then the second supersymmetry I can be rewritten as follows: I Z = I I Z = I I = iI Z I = iI Z ,
which explains the notation. It now doesnt take much imagination to write down the remaining two supersymmetries. We simply replace I in turn by each of the other two complex structures J and K of Exercise 3.32. The fact that I, J, and K satisfy the quaternion algebra is instrumental in showing that these transformations obey the right supersymmetry algebra.
Exercise 4.11 (The N =4 supersymmetry algebra) Let 1 and I be the supersymmetries given in Exercise 4.10. Dene J and K in the obvious way. Let be any of these supersymmetries and = be a second of these supersymmetries. Prove that the following algebra is obeyed: 2 = i d dt and + = 0 .
This is the N =4 supersymmetry algebra.
4.3.2
N =4 supersymmetry in hyperkhler manifolds a
We now abandon our toy model and return to the action Le given by (4.9). We expect that the supersymmetry 1 dened in Exercise 4.10 should also be an invariance of Le and, given our choice of coordinates, that so should I. This is because I is diagonal and constant on the chosen basis. In fact, leaving aside for the moment the issue of the invariance of Le under these transformations, Exercise 4.10 shows that they obey the right supersymmetry algebra. On the other hand, the other two complex structures J and K will not be constant in this basis, and hence the transformations J and K dened above will not obey the supersymmetry algebra. We will have to modify them appropriately. To see this we will investigate the supersymmetry transformations associated to a a covariantly constant complex structure I. Let us not work on a complex basis adapted to I, but rather on some arbitrary basis (X a , a ). We will attempt to write down a supersymmetry transformation using I. Because the has order 1 (being essentially a square root of d/dt), X a is 2 determined up to an inconsequential overall constant: X a = Ib a b . 132 (4.11)
[email protected]
Computing 2 we nd 2 X a = c Ib a Id c d b + Ib a b . If we now use the fact that I is covariantly constant, so that c Ib a = cb d Id a cd a Ib d , we can solve for b by demanding that 2 X a = iX a : a = iIb a X b bc a Id b d c , (4.12)
where we have discarded a term 1 Ib a Tcd b Ie c If d e f where Tcd b cd b dc b 2 is the torsion of the connection, which in our case is zero. In order to show that 2 = id/dt on a , two approaches present themselves. One can use the fact that a = Ib a X b and use the fact that on (any function of) X, 2 = id/dt: 2 a = 2 (Ib a X b ) = ic Ib a X c X b Ib a 2 X b = ic Ib a X c X b Ib a 2 X b = ic Ib a X c X b iIb a X b = ic Ib a X c X b iIb a (Ic b c ) = ic Ib a Id b X c d iIb a d Ic b X d c + i a = i a , where in the last line we have used an identity resulting from taking the derivative of I 2 = 1. Alternatively, one can compute 2 a directly. This is naturally left as an exercise.
Exercise 4.12 (Another proof that 2 a = i a ) a , prove that 2 a = i a . By taking of (Hint: You might nd it necessary to use two properties of the Riemann curvature tensor: Rabc d + Rbca d + Rcab d = 0; and Rabc d Id e = Rabd e Ic d . You are encouraged to prove these identities. The rst one is the (rst) Bianchi identity, the other one follows from the fact that I is covariantly constant, and hence commutes with the curvature operator.)
since 2 = 3 = 2
133
[email protected]
Let I denote the supersymmetry transformation associated to the com plex structure I. If we dene 1 as above: 1 X a = a and 1 a = iX a , then just as before 1 I = I 1 . In other words, 1 and I generate an N =2 supersymmetry algebra. Therefore this result holds for any Khler manifold, a 4 and not just for R as in the previous section. Now let J be a second covariantly constant complex structure. It will give rise to its own supersymmetry transformation given by equations (4.11) and (4.12), but with J replacing I. Let us call this supersymmetry transformation J . When will I and J (anti)commute? The next exercise provides the answer.
Exercise 4.13 (Commuting supersymmetries) Let I and J denote the supersymmetries generated by covariantly constant complex structure I and J. Prove that I J X a = iIc b Jb a X c bc a Id b Je c d e . Conclude that (I J + J I )X a = 0 if and only if IJ = JI. Prove that if this is the case, then (I J + J I ) a = 0 as well, so that the two supersymmetries (anti)commute. (Hint: To compute I J + J I on a , you might nd it easier to exhibit a = Ib a I X b , say, and then use that I J + J I is zero on (functions of) X a .)
This means that if {X a } denote the coordinates of a hyperkhler manifold a a and { } are the accompanying fermions, then the four supersymmetries 1 , I , J and K satisfy an N =4 supersymmetry algebra.
4.3.3
N =4 supersymmetry of Le
It remains to show that the four supersymmetries dened above, are indeed symmetries of the eective action Le . This will be easier to ascertain if we rst rewrite the supersymmetry transformations (4.11) and (4.12) in complex coordinates adapted to one of the complex structures: I, say. Let us therefore choose coordinates (Z , Z , , ) adapted to the complex structure I. Because the the metric is hermitian relative to this complex structure (in fact, relative to all three), we can rewrite the equations (4.11) and (4.12) using the results of Exercise 3.23. Because I is constant relative to these basis, I precisely agrees with I in Exercise 4.10. Now consider the second complex structure J. Because IJ = JI, J maps vectors of type (0, 1) relative to I to vectors of type (1, 0) and viceversa. In other words, relative to the above basis adapted to I, J has com ponents J and J . Therefore J generates the following supersymmetry
134
[email protected]
transformation: J Z = iJ J Z = J
J = iJ Z J J = iJ Z J ,

where we have used (see Exercise 3.23) that and are the only nonzero components of the Christoel symbols. Similar formulas hold for K . As they stand, these supersymmetries are fermionic transformations. We can make them bosonic by introducing an anticommuting parameter and dening the following transformations: 1 Z = 1 Z = I Z = i I Z = i J Z = iJ J Z = J K Z = iK K Z = K 1 = iZ = iZ I = Z I = Z J = iJ Z J = iJ Z J J K = iK Z K = iK Z K K
(4.13) (4.14) (4.15) . (4.16)
The task ahead is now straightforwardalbeit a little time consuming. Taking each of these supersymmetries in turn, and letting depend on time, we vary the action Le . Invariance of the action implies that Le = Q + X , where X is arbitrary, and Q is the charge generating the supersymmetry. The next exercise summarises the results of this calculation.
Exercise 4.14 (The supersymmetry charges) Prove that Le is invariant under the supersymmetries given by equations (4.13) (4.16), with the following associated supersymmetry charges: Q1 = g Z + g Z Q = ig Z ig Z
I QJ = J Z + J Z QK = K Z + K Z ,
135
[email protected]
where J = J g and J = J g , and similarly for K. (Hint: The calculation uses the two fundamental identities described in the hint to Exercise 4.12. In complex coordinates, and using the results of Exercise 3.23, they now look as follows:
R = R R J and similarly for K.)

R = R
= R J
R J = R J ,
Having established the N =4 supersymmetry of the eective action Le it is now time to quantise the theory. It turns out that supersymmetry will provide a geometric description of the Hilbert space and of the hamiltonian. This will require some basic concepts of harmonic theory on Khler mania folds. The purpose of the next section is to provide a brief review for those who are not familiar with this topic.
4.4
A brief review of harmonic theory
This section contains a brief scholium on the harmonic theory of orientable riemannian manifolds and in particular of Khler manifolds. The reader a familiar with these results can easily skip this section. Other readers are encouraged to read on. We will of necessity be brief: details can be found in many ne books on the subject [Gol62, GH78, War83, Wel80].
4.4.1
Harmonic theory for riemannian manifolds
Let M be a smooth manifold. We will let E = p Ep denote the algebra of dierential forms on M . The de Rham operator d : Ep Ep+1 obeys d2 = 0 and hence one can dene its cohomology (the de Rham cohomology of M ) as follows: ker d : Ep Ep+1 p HdR (M ) = . im d : Ep1 Ep In other words, the p-th de Rham cohomology is a vector space whose elements are equivalence classes of closed p-forms (d = 0)two closed p-forms 1 and 2 being equivalent if their dierence is exact: 1 2 = d, for some (p 1)-form . The crown jewel of harmonic theory is the decomposition theorem of Hodge, which states that if M is a compact orientable manifold there exists a privileged representative for each de Rham cohomology class. This representative is obtained by introducing more structure on M namely a riemannian metric. From the above denition, it is clear that 136
[email protected]
the de Rham cohomology does not depend on any geometric properties of the manifold. It is precisely this reason why the Hodge theorem is of fundamental importance: because it establishes a link between the topological and the geometric properties of riemannian manifolds. (Actually, the fact that the de Rham cohomology is a topological invariant of M is not obvious. It is called the de Rham theorem and it is proven in [War83, BT81].) We therefore let (M, g) be an m-dimensional orientable riemannian manifold. Let {ei } for i = 1, . . . , m be a local orthonormal basis for the 1-forms. Orthonormality means that the line element is locally ds2 = i ei ei . In general such a basis will of course not exist globally, but will transform under a local O(m) transformation when we change coordinate charts. In this basis the volume form is given by vol = e1 e2 em . This volume form denes a local orientation in M . Orientability simply means that, unlike the 1-forms {ei }, the volume formand hence the orientationdoes exist globally. It also means that upon changing charts, the {ei } will change by a local SO(m) transformation. (Prove this!) A local basis for the dierential forms E on M is given by wedge products of these 1-forms. It is convenient to introduce multi-indices I = (i1 , i2 , . . . , ip ) where 1 i1 < i2 < < ip m. We say that I has length p or that |I| = p. We then dene eI ei1 ei2 eip . In this notation, {eI |I, |I| = p} is a local basis for Ep ; that is, any p-form on M can be written locally like |I|=p I eI , where the coecients I are smooth functions. If I = (i1 , i2 , . . . , ip ) is a multi-index of length p, we let I c = (ip+1 , ip+2 , . . . , im ) denote the multi-index of length m p uniquely dened by the fact that {i1 , . . . , ip } {ip+1 , . . . , im } = {1, 2, . . . , m}. We can now dene the Hodge -operation. This is a linear map : Ep mp E dened by c eI = sign eI , where if I = (i1 , . . . , ip ) and I c = (ip+1 , . . . , im ), then sign is the sign of the permutation i i im = 1 2 . 1 2 m In particular 1 = vol. The following result is important for calculations.
Exercise 4.15 (The square of the Hodge ) Prove that acting on Ep , 2 = ()(mp)p .
The Hodge -operator allows us to dene a pointwise metric , on forms as follows: = , vol . (4.17)
137
[email protected]
The properties of this pointwise metric are summarised in the following exercise.
Exercise 4.16 (The pointwise metric on forms) Prove that the basis {eI } is orthonormal relative to the pointwise metric dened in (4.17) and therefore that it agrees on 1-forms with the one induced by the riemannian metric on M . Conclude that the pointwise metric is positive-denite.
If, in addition, M is compact we can dene an honest metric (called the Hodge metric) on forms by integrating the pointwise metric over the manifold relative to the volume form: (, ) =
M
, vol =
M
If M is not compact, then we can restrict ourselves to compactly supported forms or to forms for which the Hodge norm 2 (, ) is nite. Such forms are often called square-integrable. The Hodge metric allows us to dene the adjoint d to the de Rham operator, with which the following exercise concerns itself.
Exercise 4.17 (The adjoint de Rham operator) Dene the adjoint d of the de Rham operator by (d, ) = (, d ) , for all forms , E. Prove that d satises the following properties: (1) d : Ep Ep+1 ; (2) (d )2 = 0; and (3) d = ()mp+m+1 d acting on Ep .
Now let us dene the Hodge laplacian : Ep Ep by We say that a p-form is harmonic if = 0.
dd + d d.
Exercise 4.18 (Harmonic forms) Prove that a form is harmonic if and only if d = d = 0. Prove that harmonic forms have minimal Hodge norm in their cohomology class. That is, if is harmonic, then prove that the Hodge norm of + d is strictly greater than that of .
The Hodge decomposition theorem states that in a compact orientable manifold each de Rham cohomology class has a unique harmonic representative; that is, that there is a vector space isomorphism
p HdR (M ) harmonic p-forms . =
138
[email protected]
The proof of this theorem is rather technical. The idea is to use the normminimising property to dene the harmonic representative; but one then has to prove that this form is smooth. This calls for the use of regularity theorems which are beyond the scope of these notes. A proof can be found, for example, in [War83]. The Hodge decomposition theorem has a very important corollary, which the following exercise asks you to prove.
Exercise 4.19 (Poincar duality) e Prove that the Hodge -operator commutes with the Hodge laplacian. Use the Hodge decomposition theorem to conclude that for M an m-dimensional compact orientable manifold, there is an isomorphism
mp p HdR (M ) HdR (M ) . =
This isomorphism is known as Poincar duality. e
4.4.2
Harmonic theory for Khler manifolds a
Now suppose that M is a complex manifold of complex dimension n. As explained in section 3.3.2, on a complex manifold one has local coordinates (z , z ), where , = 1, 2, . . . , n. This allows us to rene the grading of the complex dierential forms. We say that a complex dierential form is of type (p, q) if it can be written in local complex coordinates as = 1 p 1 q (z, z )dz 1 dz p d1 dq , z z where 1 p 1 q (z, z ) are smooth functions. The algebra of complex dif ferential forms is then bigraded as follows: E=
0p,qn
Ep,q ,
(4.18)
where Ep,q is the space of (p, q)-forms. The de Rham operator d also breaks up into a type (1,0) piece and a type (0,1) piece: : Ep,q Ep+1,q . d=+ where : Ep,q Ep,q+1 Breaking d2 = 0 into types we nd that 2 = 2 = + = 0. We call the Dolbeault operator, and its cohomology
p,q H (M ) =
ker : Ep,q Ep,q+1 im : Ep,q1 Ep,q 139
[email protected]
the Dolbeault cohomology. Now suppose that we give M a hermitian metric h; that is, a riemannian metric compatible with the complex structure: h(IX, IY ) = h(X, Y ). Such metrics always exist: one simply takes any riemannian metric g, say, and 1 averages it over the nite group generated by I: h(X, Y ) 2 g(X, Y ) + 1 g(IX, IY ). If we forget the complex structure for a moment, M is an 2 orientable riemannian manifold of (real) dimension 2n. Therefore we have a Hodge -operator dened as in the previous section. The next exercise asks you to show how interacts with the complex structure.
Exercise 4.20 (The Hodge and the complex structure) Prove that the Hodge -operator maps (p, q)-forms to (n q, n p)-forms: : Ep,q Enq,np , and that acting on (p, q)-forms, 2 = ()p+q . (Hint: The rst part is computationally quite involved, but the idea is easy. We can always nd a local basis {i } for the (1,0)-forms on M such that the line element (relative to the hermitian metric) has the form
n
ds2 =
i=1
i i + i i
where {i } are the complex conjugate (0,1)-forms. We can decompose these forms 1 into their real and imaginary parts as follows: j = 2 e2j1 + ie2j and j = other words, they form an orthonormal basis. Therefore we know the action of the Hodge -operator on the {eI }. Your mission, should you decide to accept it, is to nd out what it is in terms of the I J . The second part simply uses Exercise 4.15.)
1 e2j1 ie2j 2 2n j j j=1 e e ; in
. In terms of these real 1-forms, the line element becomes ds2 =
Another operation that we have on a complex manifold is complex conjugation, which exchanges (p, q)-forms with (q, p)-forms. Using the Hodge -operator and complex conjugation we can dene a pointwise hermitian metric for the complex forms, also denoted , as in the real case treated in the previous section. This metric is dened by = , vol . Notice that relative to this metric, the decomposition in equation (4.18) is orthogonal: if is a (p, q)-form, then is a (q, p)-form, and is a (n p, n q)-form. The only way one can obtain the volume form, which is an (n, n)-form, is to wedge with another (p, q)-form. 140
[email protected]
Exercise 4.21 (The pointwise hermitian metric) Prove that the basis {I J } is orthonormal relative to the pointwise hermitian metric, and conclude that it is positive-denite.
If M is compact, we can then integrate this pointwise metric relative to the volume form and dene an honest hermitian metric on complex forms: (, ) =
M
, vol =
M
(4.19)
This metric is again called the Hodge metric. As in the real case, if M is not compact, then we can still make sense of this provided we restrict our attention to square-integrable forms. It follows from Exercise 4.17 that on a complex manifold, d = d regardless on which forms it is acting. Breaking d into types we nd d = + where : Ep,q Ep1,q . : Ep,q Ep,q1
On the other hand, breaking d into types, and comparing we see that = and = .
We can therefore dene two laplacian operators: = + and = + ,
both of which map Ep,q Ep,q . Just as for de Rham cohomology, there is a Hodge decomposition theorem for Dolbeault cohomology, which says that the -cohomology on (p, q)-forms is isomorphic to the space of -harmonic (p, q)-forms:
p,q H -harmonic (p, q)-forms . =
In a generic complex manifold there is no reason to expect any relation between the Dolbeault laplacians and the Hodge laplacian = d d + dd ; but the magic of Khler geometry is that if M is Khler, then a a =2 = 2 . (4.20)
This is not a hard result to obtain, but it requires quite a bit of formalism that we will not need in the remainder of this course, hence we leave it unproven and refer the interested reader to the literature [Gol62, GH78, Wel80]. 141
[email protected]
As an immediate corollary of equation (4.20) and of the Hodge decomposition theorems for de Rham and Dolbeault cohomologies, we have
r HdR (M ) = p+q=r p,q H (M ) ;
and the following exercise describes another immediate corollary of equation (4.20).
Exercise 4.22 (Serre duality) Prove that both the Hodge -operator and complex conjugation commute with the laplacian. Use this to conclude that for M a compact Khler manifold of complex a dimension n, there exist isomorphisms:
p,q nq,np np,nq H (M ) H (M ) H (M ) . = =
These isomorphisms are known collectively as Serre duality.
Finally, a curiosity. If we dene the r-th Betti number br of a manifold as the real dimension of the r-th de Rham cohomology, we have as a consequence of Serre duality that for a compact Khler manifold all the odd Betti numbers a are even.
4.4.3
Explicit formulas for and
The purpose of this section is simply to derive some explicit expressions for the dierential operators and . These are the expressions by which we will be able to recognise them when we quantise the eective action. Throughout this section M shall be a Khler manifold of complex dimension n. a Let us rst start by deriving formulas for d and d . For this we can forget momentarily the complex structure and think of M simply as an orientable riemannian manifold of dimension 2n. Let {ei } denote a local orthonormal basis for the vector elds, and let {ei } denote the canonical dual basis for the 1-forms. They are also orthonormal relative to the induced metric. Let denote the Levi-Civita connection. We claim that d can be written as
2n
d=
i=1
ei
ei
(4.21)
Proving this will be the purpose of the following exercise.

Exercise 4.23 (An explicit expression for d) Let d? denote the right-hand side of equation (4.21).
142
[email protected]
1. Prove that d? is independent of the orthonormal basis chosen so that it is well-dened. 2. Let ei = ei dxa and ei = ea a . Prove that d? = a i that d? = d.
a a dx
a , and conclude
(Hint: Use the fact that the Levi-Civita connection is torsionless.)
With this result we can now describe a similar formula for d . Letting {ei } and {ei } be as above, we will prove that
2n
d =
i=1
(ei )
ei
(4.22)
where (ei ) is the contraction operation, dened by: 1. (ei ) f = 0 for f a function; 2. (ei ) ej = ij ; and 3. (ei ) (ej ) = ij ej (ei ) . The next exercise asks you to prove equation (4.22).
Exercise 4.24 (An explicit expression for d ) Let d stand for the right-hand side of equation (4.22). Prove that d = d , ? ? whence it agrees with d . (We are using (3) in Exercise 4.17, with m = 2n.) (Hint: Prove rst that d is well-dened; that is, it is independent of the choice of ? orthonormal frame. Because of this and by linearity, conclude that it is sucient to compare d and d on a p-form of the form f e1 e2 ep . Moreover ? argue that it is sucient to compute this at a point where ei ej = 0. Then do it.)
As a corollary of the previous exercise, it follows that relative to a coordinate basis, we can write d =
a
(dxa ) a ,
(4.23)
where now (dxa ) dxb = g ab . We can now re-introduce the complex structure. Let {i , i } be a complex basis for the complex vector elds, and let {i , i } be the canonical dual basis for the complex 1-forms. In terms of the above basis {ei }, i is given as in Exercise 4.20. The canonical dual basis for the vector elds are related by 1 i = (e2i1 ie2i ) 2 143 1 i = (e2i1 + ie2i ) . 2
[email protected]
Inverting this change of basis, and using equations (4.21) and (4.22), we nd
n
d=
i=1
+ i
and
n
d =
i=1
(i )
+ (i )
Breaking up into types, one concludes that

n n
=
i=1
and
=
i=1
(i )
Or in a coordinate basis,
n n
=
=1
d z
and
=
=1
(dz ) .
(4.24)
These equations will be important in the sequel.
4.5
Quantisation of the eective action
In this section we discuss the canonical quantisation of the eective action (4.8). We will be able to identify the Hilbert space with the square-integrable (0, q)-forms on the moduli space Mk . We will exploit the supersymmetry to write the hamiltonian as the anticommutator of supersymmetry charges which, under the aforementioned isomorphism, will be identied as the Dol beault operator and its adjoint under the Hodge metric. This will then allow us to identify the ground states of the eective quantum theory as the harmonic (0, q)-forms on the moduli space.
4.5.1
Canonical analysis
The rst step in this direction is to nd the expression for the canonical momenta. Then we write the hamiltonian and the supersymmetry charges in terms of the momenta. We write down the Poisson brackets and make sure that the classical algebra is indeed the N =4 supersymmetry algebra. Most of these calculations are routine, and are therefore left as exercises. The rst exercise starts you in this path by asking you to compute the canonical momenta. 144
[email protected]
Exercise 4.25 (The canonical momenta) Prove that the canonical momenta dened by Le take the following form: Le Z Le P = Z Le = Le = P = where = g .
The fact that = 0 is not very important. It is simply a consequence of the fact that the fermionic part of the eective lagrangian is already in rst order form, so that morally speaking { } play the role of momenta while { } are coordinates. = g Z + i
= g Z = ig =0,
The eective hamiltonian He is dened as usual by: He = Z P + Z P + Le . The next exercise asks you to compute it.
Exercise 4.26 (The eective hamiltonian) Prove that the eective hamiltonian is given by He = g P P + g P .

Next we write the supersymmetry charges obtained in Exercise 4.14 in terms of momenta. This is another easy exercise.
Exercise 4.27 (The supersymmetry charges revisited) Prove that the supersymmetry charges obtained in Exercise 4.14 have the following form:
Q1 = P + P = P + ig P QI = i P i P = i P + g P QJ = J P + J P iJ QK = K P + K P iK .
The canonical Poisson brackets are dened to be the following:

{P , Z } = {P , Z } = { , } = .
145
[email protected]
Exercise 4.28 (Some checks) As a check on our calculations, show that the supersymmetry transformations given in equations (4.13)(4.16) are indeed generated via Poisson brackets by the supersymmetry charges computed in the previous exercise.
Finally, we are ready to verify that we have a classical realisation of the N =4 supersymmetry algebra. Let q = q1 i + q2 j + q3 k + q4 H be a quaternion. Let Qq q1 QI + q2 QJ + q3 QK + q4 Q1 . The next exercise asks you to prove that the supersymmetry charges obey the N =4 supersymmetry algebra.
Exercise 4.29 (Classical N =4 supersymmetry algebra) Let q, q H be quaternions. Then prove that under Poisson bracket: {Qq , Qq } = i(q )He , q where q = q1 i q2 j q3 k + q4 is the conjugate quaternion and q q = the quaternionic product.
i qi qi
is
Because the supersymmetry charges generate under Poisson bracket the supersymmetry transformations, the above exercise implies that the eective hamiltonian indeed generates time translation. If you feel up to it you can check this directly from the expression of the hamiltonian.
4.5.2
The quantisation of the eective hamiltonian
To quantise the eective hamiltonian we rst need to identify the Hilbert space. Let us quickly quantise the bosons. We choose to realise Z and Z as will be realised as derivatives: multiplication operators and hence P and P P i Z and P i . Z
For the fermions, we notice that the canonical Poisson brackets can be rewrit ten in terms of and as follows: { , } = ig . Upon quantisation this gives rise to the following anticommutation relations + = g , with all other anticommutators vanishing. Of course, g is a function of Z , Z ; but for each point (Z , Z ) in Mk , the above anticommutation relations dene a Cliord algebra. In other words, this denes a Cliord bundle on Mk . Fixing a point in Mk , we have a standard Cliord algebra of 146

[email protected]
the type studied in section 2.2.2. It has a unique irreducible representation constructed as follows. We choose a Cliord vacuum | , dened by the condition | = 0 for all .
The representation is then built on | by acting with the . We now tensor together the representations of the bosons and the fermions and what we have is linear combinations of objects of the form f (Z, Z) | .
If we take the f (Z, Z) smooth, this space is clearly isomorphic to the space 0,p 0p2k E of dierential forms of type (0, p) on Mk :
f (Z, Z) | f (Z, Z)dZ dZ dZ .
Of course the Hilbert space will consist of (the completion of) the subspace formed by those forms which are square integrable relative to a suitable inner product. As we saw in section 4.4.2, the natural inner product to consider is the Hodge metric given by (4.19). Therefore we have the following geometric interpretation of the Hilbert space H of the quantum eective theory: H =
0p2k
E0,p , L2
(4.25)
where E0,p denotes the space of (0, p)-forms on Mk with nite Hodge norm; L2 that is, square-integrable. In order to identify the hamiltonian we will use supersymmetry. The expressions for the supersymmetry charges and the hamiltonian, being polynomial, suer from ordering ambiguities. One way to get around this problem is to dene the quantisation in a way that the N =4 supersymmetry algebra is realised quantum-mechanically, and in such a way that we can identify the resulting operators geometrically. The hamiltonian can be dened as the square of any of the supersymmetry charges, but we nd it more convenient to take complex linear combinations of Q1 and QI . Indeed, let us dene
1 Q = i 2 (Q1 + iQI )
and
Q = i 1 (Q1 iQI ) . 2
The classical expressions for these charges are very simple

Q = i P
and
Q = i P ,
147
[email protected]
and they obey the following algebra {Q, Q } = i 1 He . 2

Q Z
(4.26)
The quantisation is now clear. Quantise the charges Q and Q as follows: and Q . Z (4.27)
But notice that we have seen these operators before. Indeed, acting on f f | , we nd that
Qf = f | .
Under the isomorphism (4.25), this corresponds to the form f . In other words, Q . How about Q ? Acting on a (0, 0)-form, Q is zero, since annihilates the Cliord vacuum. Acting on a (0, 1)-form f | , we nd
Q f | = f g | .
In other words, up to a sign, it is given by the divergence. This fact persists to higher (0, p)-forms. Indeed, the next exercise asks you to show that Q = , the adjoint of under the Hodge metric.
Exercise 4.30 (Q is ) Show that under the isomorphism (4.25), the quantisation of Q given by (4.27) agrees with = , the adjoint of under the Hodge metric. (Hint: Compare with equation (4.24).)
Finally, we quantise the hamiltonian by demanding that the N =4 supersymmetry algebra be preserved quantum-mechanically. In other words, and taking into account equation (4.26), we quantise the hamiltonian as follows: He 2 (QQ + Q Q) . Under the identication Q and Q , the quantum eective hamiltonian agrees with twice the Dolbeault laplacian orsince Mk is (hyper)Khlerwith the Hodge laplacian . a This result doesnt just provide a beautiful geometric interpretation of the eective quantum theory, but also allows us to use geometric information to derive physical results. For example, the ground states of the theory will be in one-to-one correspondence with (square-integrable) harmonic (0, p)-forms. This sort of reasoning will play a crucial role in the test of MontonenOlive duality in N =4 supersymmetric YangMills theory, which shall be the focus of the next chapter. 148
Chapter 5 The Eective Action for N =4 Supersymmetric YangMills

In the previous chapter we found that the low energy eective action for the collective coordinates of N =2 supersymmetric YangMills was given by supersymmetric quantum mechanics on the moduli space of BPS-monopoles. In this chapter we will do the same for N =4 super YangMills. As we saw when we discussed that theory in Chapter 2, N =4 super YangMills is a prime candidate to exhibit MontonenOlive duality: not just are the masses and the structure of the multiplets protected by supersymmetry, but the massive vector bosons and the BPS-monopole belong to isomorphic multiplets. Therefore it would be possible for this theory to aord two inequivalent descriptions: one the standard one and a dual description where the perturbative elds are those in the multiplet containing the BPS-monopole. The structure of this chapter is therefore very similar to that of the previous chapter. We will rst count the number of fermionic collective coordinates and will perform the collective coordinate expansion of the action up to second order. The resulting theory is again a (0 + 1) supersymmetric -model, this time admitting N =8 supersymmetry due to the fact that there are twice as many fermionic collective coordinates as in the N =2 case. The quantisation of the eective action will proceed along lines similar to the previous chapter: this time the Hilbert space will be isomorphic to square integrable forms on the monopole moduli space, and the hamiltonian will once again be given by the laplacian. This chapter is based on the work of Blum [Blu94].
149
[email protected]
5.1
Fermionic collective coordinates
We saw in section 3.2 that there are 4k bosonic coordinates in the k-monopole sector; and, as we saw in section 4.1, N =2 supersymmetry contributed 2k fermionic collective coordinates. In this section we will show that for N =4 supersymmetric YangMills the number of fermionic collective coordinates will double. We can understand this heuristically in a very simple matter. It follows from the discussion in section 4.1, that fermionic collective coordinates are in one-to-one correspondence with zero modes of the Dirac equation in the monopole background. The Dirac operator is the same in both the N =2 and the N =4 theories, but it acts on dierent types of fermions. In the N =2 theory, was an unconstrained Dirac spinor (it came from a Weyl spinor in six dimensions); whereas in the N =4 theory, it acts on a quartet of Majorana fermions (the dimensional reduction from ten-dimensions of a MajoranaWeyl fermion). But now four Majorana spinors have twice the number of degrees of freedom that an unconstrained Dirac spinor does: 4 4 = 16 real components to only 4 complex. To make this argument precise, we need to look in detail at how the Dirac operator breaks up. We start with a monopole background like the one in 2.4.3. Namely, we choose W0 = 0, SI = aI , and PJ = bJ , where aI and bJ are real numbers satisfying I (a2 + b2 ) = 1, and where (Wi , ) dene a I I k-monopole. Because the scalar elds are collinear, the potential remains at the minimum provided that the fermions satisfy the Dirac equation: i D i = 0 , where i = 0 i , and 4 = i0 aI I + bJ J 5 .
Exercise 5.1 (Euclidean Cliord algebra) Prove that the matrices i dened above satisfy a euclidean Cliord algebra in four-dimensions.
From Exercise 4.1 we know that the normalisable zero modes of the Dirac operator i Di will have negative chirality with respect to 5 . But remember that is also Majorana. We now check what chirality with respect to 5 and the Majorana condition imply on a spinor. Recall that our choice (2.32) of ten-dimensional -matrices is such that the Majorana condition in ten-dimensions translates directly into the Majorana condition in four-dimensions. In four-dimensional Minkowski spacetime, there cannot be MajoranaWeyl spinors, but the euclidean -matrices preserve the Majorana condition as the next exercise asks you to show.
150
[email protected]
Exercise 5.2 (i and the Majorana condition) Let be a quartet of Majorana spinors. Prove that i is again Majorana. Deduce that one can simultaneously impose the Majorana and 5 -chirality conditions.
Let us now start by choosing a explicit realisation for the -matrices: 0 = 0 1 1 0 i = ii 0 0 ii 5 =
0 1 0
(5.1)
The next exercise asks you to compute the charge conjugation matrix in this realisation.
Exercise 5.3 (The charge conjugation matrix explicitly) Prove that the charge conjugation matrix C in the above realisation can be chosen to be i2 0 C= . 0 i2
t (Hint: Using that C t = C and that C = C determine C up to a constant multiple. A possible choice for this multiple is then one for which C C = 1. That is the choice exhibited above.)
Now let denote a quartet of Majorana spinors which in addition obey 5 = . Because 5 = 5 4 (prove it!), the chirality condition on means that 4 = 5 . This means that the euclidean Dirac equation i Di = 0 becomes (i Di 5 D4 ) = 0. For the explicit realisation (5.1), this has the virtue that the Dirac operator doesnt see the internal SU (4) indices. Indeed, the Dirac operator is given by: i D i = 0 D 14 = (i2 D) 14 ) , D 0 (5.2)
where 14 is the identity matrix in the internal SU (4) space, and D = iDi i + e1 is the operator introduced in (3.17).
Notice that if had the opposite chirality with respect to 5 , then it would have been D which would have appeared. This is as expected from the results of Exercise 3.5 and Exercise 4.1.
We are now ready to count the zero modes of the euclidean Dirac operator, by relating them to zero modes of D, which we have already calculated to be 2k. We rst choose the explicit realisation for the I and J matrices found in Exercise 2.33: I = e+ and J = e . Next we exploit the internal SU (4) I J invariance to x a1 = 1 and all the other aI and bJ to zero. This means that 5 = i0 5 1 = (3 1) (2 1). From Exercise 5.3 we know that the charge conjugation matrix is given by C = (3 i2 ) 14 . The next exercise asks you to write down the typical quartet of Majorana spinors which in addition are chiral with respected to 5 . 151
[email protected]
Exercise 5.4 (MajoranaWeyl spinors) Prove that every quartet of Majorana spinors obeying 5 = is of the form: i2 i2 i 2 i 2 ,
where and are complex two-component spinors.
Finally we count the zero mode of the euclidean Dirac operator i Di .

Exercise 5.5 (Counting zero modes) Show that is a MajoranaWeyl zero mode of the euclidean Dirac operator, in the sense of the previous exercise, if and only if and are zero modes of D. Therefore if a for a = 1, . . . , 2k is a basis for the normalisable zero modes of D, then the 4k spinors a i2 a 0 0 ia 2 a 0 0 0 0 a i2 a 0 0 ia 2 a
form a basis for the normalisable zero modes of i Di .
In summary, there are 4k fermionic collective coordinates for N =4 supersymmetric YangMills with gauge group SO(3).
152
Part II Arbitrary Gauge Groups
153
Chapter 6 Monopoles for Arbitrary Gauge Groups

In this chapter we start the study of electromagnetic duality in supersymmetric gauge theories with an arbitrary gauge group. We will be interested in this part of the notes only on N =4 super YangMills. Our principal aim is to frame an analogue of the MontonenOlive duality conjecture for these theories, to develop testable predictions and then to test them. This will occupy several chapters, but in this one we will start with the analysis of the kind of monopole solutions that can exist in a YangMillsHiggs theory with gauge group G, taken to be a compact, connected Lie group, and a Higgs eld with values in the adjoint representation. We will cover the homotopy classication of topologically stable solutions and the generalised Dirac quantisation condition. This chapters borrows quite a lot from the magnicent lectures of Coleman [Col77], and from the paper of Goddard, Nuyts and Olive [GNO77].
6.1
Topologically stable solutions
Let G be a compact connected Lie group, and a scalar eld taking values in some nite-dimensional representation V of G. We will assume that there is a G-invariant potential V () which is positive semi-denite and also that V admits a G-invariant metric. This is necessary in order to write down the kinetic term for in the action. We will let g denote the Lie algebra of G. We will x once and for all an invariant metric on g. As the next exercise shows, such a metrics always exists.
Exercise 6.1 (Invariant metrics exist) Prove that there exists a G-invariant metric in the Lie algebra of a compact Lie
154
[email protected]
group. (Hint: Start with any metric and average over the group with respect to the Haar measure. Does this argument work for any representation?)
We will denote both metrics on g and on V by (, ), hoping that no confusion will arise. The lagrangian density of the YangMillsHiggs system is given by 1 L = 4 (G , G ) + 1 (D , D ) V () , (6.1) 2 where D = eW , G = W W e[W , W ], where W are the g-valued gauge potentials, and by we mean the action of g on the representation V. Let M0 denote the manifold of vacua: those values of for which V () = 0. Because V is G-invariant, G will map M0 to M0 ; in other words, G stabilises M0 . We now choose the temporal gauge W0 = 0. As the next exercise shows, this can always be done and leaves intact the freedom of performing timeindependent gauge transformations.
Exercise 6.2 (The temporal gauge) Prove that the temporal gauge exists by exhibiting a gauge transformation which makes W0 = 0. Prove that this gauge is preserved by time-independent gauge transformations. (Hint: Use path-ordered exponentials.)
In this gauge, the energy density corresponding to the lagrangian density (6.1) is given by
1 1 H = 1 (Wi , Wi ) + 1 (, ) + 4 (Gij , Gij ) + 2 (Di , Di ) + V () , 2 2
where a dot indicates the time derivative, and where repeated indices are summed. The energy is of course the integral over space R3 of the energy density H , and hence nite-energy congurations must obey the following asymptotic conditions as |r| : Wi = 0 and = 0, whence elds are asymptotically static; Gij = 0 faster than O(1/r); Di = 0 faster than O(1/r); and 155
[email protected]
V () = 0. In particular this last condition says that denes a map from the asymp2 totic 2-sphere S R3 to the manifold of vacua M0 . Because M0 is stabilised by G, it will be foliated by orbits of G. For example, in G = SO(3) and the Higgs is in the adjoint, the leaves of the foliation of SO(3) in R3 are the round spheres centred at the origin, with a singular orbit corresponding to the sphere of zero size. A priori there is no reason to expect that the mapping 2 S M0 dened by the asymptotics of the Higgs eld should lie on only one of the orbits, but because Di = 0 in this limit, this is actually the case. The proof is left as an exercise.
2 Exercise 6.3 ((S ) M0 lies in a single orbit) 2 Prove that the image of S lies in a single orbit in M0 . 2 (Hint: Integrate the equation Di = 0 on S .)
A sucientbut as shown by Coleman [Col77]) not necessarycondition for a nite-energy conguration to be non-dissipative is that it should be topologically stable. As explained in section 1.2.2, a way to guarantee the 2 topological stability of a eld conguration is for the map : S M0 to belong to a nontrivial homotopy class. It therefore behoves us to study the 2 homotopy classes of maps from the asymptotic two-sphere S to the manifold 2 of vacua M0 or, more precisely, to the orbit to which (S ) belongs. To this eect we nd it useful to set down some basic notions about homotopy groups. Readers familiar with this material can easily skip the next section.
6.1.1
Some elements of homotopy
This section contains a brief review of homotopy theory. Homotopy theory is the study of continuous change, and is particularly concerned with the determination of quantities which are impervious to such changes. Roughly speaking a homotopy is a continuous deformation parametrised by the unit interval I = [0, 1]. We will have to be a little bit more precise than this in what follows, but we will avoid getting too technical. In particular, we will not give many proofs. Luckily for us, the aspects of homotopy theory that we will need in these notes can be understood quite intuitively. Proofs can be given but they rely quite a bit on point-set topology. Since that is not the main point of these notes, we simply point the reader who wishes to look at the proofs of the statements made in this section to the old but still excellent book by Steenrod [Ste51]. The useful objects in homotopy theory are not just topological spaces, but spaces with a privileged point called the basepoint. A map between two 156
[email protected]
such spaces is understood to be a continuous function which sends basepoint to basepoint. Let X and Y be two topological spaces with basepoints x0 and y0 respectively, and let f0 and f1 be two continuous functions X Y taking x0 to y0 . We say that these two functions are homotopic if there exists a family of functions parametrised by the interval which interpolates continuously between them. More precisely, f0 is homotopic to f1 (written f0 f1 ) if there exists a continuous function F : X I Y , such that for all x X, F (x, 0) = f0 (x) and F (x, 1) = f1 (x) and such that for all t I, F (x0 , t) = y0 . This last condition says that the homotopy is relative to the basepoint. The fundamental group A good example with which to visualise these denitions is to take X to be the circle. We can think of the circle as the unit interval with endpoints identied. Then a map from the circle to Y as a map f : I Y with f (0) = f (1) = y0 . That is, a continuous loop based at y0 . Then two such loops are homotopic if they can be continuously deformed to each other through loops which are based at y0 . The set of homotopy equivalence classes of maps f : X Y with f (x0 ) = y0 is written [X, x0 ; Y, y0 ]. In the special case above where X is the circle, the set of homotopy equivalence classes is written 1 (Y, y0 ). But 1 (Y, y0 ) is more than just a set: indeed, based loops can be composed. Given two loops f1 and f2 based at y0 , we can form a third loop f1 f2 by simply going rst along f1 and then along f2 at twice the speed. In other words, (f1 f2 )(t) = f1 (2t) for t [0, 1 ], 2 1 f2 (2t 1) for t [ 2 , 1].
Notice that composition is not just dened for loops but also for paths, provided that the rst path ends where the second begins. As the following exercise shows, composition of based loops induces a well-dened operation on homotopy classes, which makes 1 (Y, y0 ) into a group.
Exercise 6.4 (1 (Y, y0 ) is a group) In this exercise we prove that 1 (Y, y0 ) is a group, with group multiplication given by composition of loops. The proof consists of several steps which are all very easy. The idea is to rst prove that makes sense in 1 (Y, y0 ) and then that on loops satises all the properties of a group up to homotopy. This means that in 1 (Y, y0 ) they are satised exactly. is well-dened in homotopy. Prove that if f0 f1 and g0 g1 are loops, then f0 g0 f1 g1 . (This allows us to work with loops, knowing that up
157
[email protected]
to homotopy it doesnt really matter which loop we choose to represent its homotopy class.) is associative in 1 (Y, y0 ). Prove that if f1 , f2 , and f3 are loops then (f1 f2 ) f3 f1 (f2 f3 ). In other words, is associative up to homotopy. 1 (Y, y0 ) has an identity. Prove that the constant loop sending all the circle to y0 is an identity for up to homotopy; that is, if k denotes the constant loop, then k f f k f for any loop f . Inverses exist. Let f be a loop, and let f denote the loop obtained by following f backwards in time: f (t) = f (1 t). Prove that f f f f k, where k is the constant loop. (Hint: It may be convenient to devise a pictorial way to denote loops and homotopies. For instance a loop f based at y0 can be depicted as a unit interval whose endpoints are marked y0 :
y0
y0
Similarly if f and g are two such loops, a homotopy H between them can be depicted as a square whose left and right edges are marked y0 and whose top and bottom edges correspond to f and g: g y0 H y0
f Composition of loops can then be depicted simply as pasting the intervals together by their endpoints and contracting (reparametrising time) so that the resulting interval has again unit length. The following picture illustrates this:
y0
y0
y0
y0
y0
g y0
158
[email protected]
In this language, the group properties become almost self-evident. For example, the associativity property of simply becomes f
gg g
g
gg g g g
y0
y0
g g g
f and similarly for the other axioms.)
The group 1 (Y, y0 ) is known as the rst homotopy group of the pointed space (Y, y0 ). If Y is path-connected, so that any two points in Y can be joined by a continuous path, then the rst homotopy group does not depend (up to isomorphism) on the basepoint. This fact has a simple proof which we leave to the next exercise. Incidentally, the condition of connectedness and path-connectedness are not equivalent, but they do agree for manifolds, and hence for all the spaces we will be considering in these notes.
Exercise 6.5 (1 (Y, y0 ) 1 (Y, y1 )) = Let Y be path-connected and y0 and y1 be two points in Y . Fix a path : I Y with (0) = y0 and (1) = y1 . Because Y is path-connected, exists. We can use this path to turn any loop f1 based at y1 into a loop based on y0 : one simply composes f1 , where (t) = (1 t). Prove that this denes a group isomorphism 1 (Y, y1 ) 1 (Y, y0 ). =
Therefore when Y is connected, it makes sense to talk about 1 (Y ) without reference to a basepoint. This group is called the fundamental group of Y . If this group is trivial, so that all loops are homotopic to the constant map, then Y is said to be simply-connected.
Notice that the isomorphism in Exercise 6.5 depends on the choice of path joining the two basepoints. How does the isomorphism depend on ? It is easy to show (do it!) that if is any other path which is homotopic to with endpoints xed, then the isomorphisms induced by and agree. On the other hand, paths in dierent homotopy classes generally dene dierent isomorphisms. This can be formulated in a way that shows an action of the fundamental group of the space on itself by conjugation. This will play a role later and we will discuss this further we study the addition of topological charges for largely separated monopoles.
159
[email protected]
The fundamental groups of some manifolds are well known. Here are some examples. Rn is simply-connected for any n. The punctured plane R2 \{0} is no longer simply-connected; in fact, 1 (R2 \{0}) Z. The isomorphism is given by the following well-known = integral formula from complex analysis. To see notice that R2 \{0} = C is the punctured complex plane. Let be a loop in C , and compute the contour integral 1 dz . 2i z It is well known that this is an integer and is a homotopy invariant of the loop. 1 (S 1 ) Z. This is just the above example in disguise. We can think = 1 of S as the unit circle in the complex plane S 1 C . Any loop (or homotopy for that matter) in C can be projected onto the unit circle by t (t) (t)/|(t)|. The isomorphism 1 (S 1 ) Z is known as = the degree of the map. It basically counts the number of times one circle winds around another. Puncturing Rn , for n > 2, does not alter the fundamental group. In fact, any loop in Rn \{0} is homotopic to a loop on its unit sphere S n1 Rn , again by projecting. But for n > 2, it is intuitively clear that any loop on S n1 is homotopic to a constant. For n = 3 it is the well-known principle that you cannot lasso an orange. You cannot lasso higher-dimensional oranges either. You can make a non-simply-connected space out of R3 by removing a circle (or a knot), say, or an innite line. However it is not only by making holes in a space that we can generate nontrivial loops. We can also identify points. For example, if we take the sphere S n , for n > 1, and identify antipodal points, we describe a space which is not simply-connected. The space in question is the space of lines through the origin in Rn+1 : since every line through the origin will intersect the unit sphere in two antipodal points. We call this space the real projective space RPn . We can lift a loop in RPn up to S n . This procedure is locally well-dened once we choose a starting point in S n . There is no further choice and when we are done with the lift, we are either at the starting point or at its antipodal point, since both points map down to the same point in RPn . 160
[email protected]
(a)
(b)
Figure 6.1: The two possible lifts to S n of a loop in RPn . In the former case, the loop has lifted to an honest loop in S n , which is depicted by (a) in Figure 6.1. Since S n is simply-connected, we can project the homotopy to RPn and this gives a homotopy for the original loop. On the other hand, if the loop ends at the antipodal point, as shown in (b) in Figure 6.1, there is clearly no way to deform it to the constant map while keeping endpoints xed, so it denes a nontrivial loop in RPn . However notice that the loop obtained by going twice around the loop lifts to an honest loop in the sphere, and is hence trivial. This shows that 1 (RPn ) Z2 . The two-to-one = map : S n RPn is a covering map, in that it is a local homeomorphism and every point p in RPn has a neighbourhood U such that its inverse image by the covering map 1 (U ) S n consists of two disconnected neighbourhoods. Because S n is simply-connected, we say that S n is the universal covering space of RPn . All reasonable spaces X (certainly all manifolds and hence all spaces considered in these notes) possess a universal covering space X. This space is simply-connected and is such that it admits a free action of the fundamental group of X. In the case of S n , it admits an action of Z2 , sending a point on the sphere to its antipodal point. The case n = 3 is particularly interesting, because it is intimately related with two of our favourite Lie groups: SU (2) and SO(3).
Exercise 6.6 (SU (2) and SO(3)) Prove that the Lie group SU (2) of 22 special unitary matrices is parametrised by a three-sphere S 3 and that the group SO(3) of 3 3 special orthogonal matrices is parametrised by the real projective space RP3 . Prove that there is a group homomorphism SU (2) SO(3) which sends both 1 and 1 in SU (2) to 1 in SO(3). Notice that 1 and 1 generate the centre of SU (2), which is isomorphic to Z2 . Hence SU (2) is the universal covering group of SO(3).
This situation persists for other Lie groups. Every semisimple compact 161
[email protected]
Lie group G has a universal covering group G, sharing the same Lie algebra g. The fundamental group 1 (G) is naturally identied with a subgroup of the centre of G. We will be able to compute 1 (G) by comparing the nite-dimensional irreducible representations of G with those of G, which are those of g. For example, not every irreducible representation of SU (2) is a representation of SO(3): only those with integer spin. Since representations of SU (2) can have integer or half-integer spin, this means that SU (2) has twice as many irreducible representations as SO(3)which is precisely the order of 1 (SO(3)) Z2 . This is no accident, as we will see later on. = Before we abandon the subject of the fundamental group, we mention one last fact. Notice that all the fundamental groups that we have discussed so far are abelian. This is not always the case. In fact, the fundamental group of any compact Riemann surface of genus g > 1 is non-abelian. However, there is an important class of manifolds for which the fundamental group is abelian.
Exercise 6.7 (1 (G) is abelian) Let G be a connected Lie group. Prove that 1 (G) is abelian. (Hint: In a Lie group there are two ways to compose loops. We can use the loop composition dened above, or we can use pointwise group multiplication, provided that the loops are based at the identity. Indeed, if f and g are loops in G based at the identity, one can dene (f g)(t) = f (t) g(t). Prove that f g f g, so that we can use group multiplication to dene the multiplication in the fundamental group. Use this to write down a homotopy between f g and g f , for any two loops f and g in G.)
Higher homotopy groups The fundamental group has higher dimensional analogues obtained by substituting the circle by a sphere. Just like we could think of the circle as the interval I with edges identied, we can think of the n-sphere as the multiinterval I n with its boundary I n identied to one point. There is a rich theory for all n, but we will only need n = 2 in these notes, so we will concentrate mainly on this case. Any map S 2 X can be thought of as a map I 2 X which sends the boundary I 2 to the basepoint x0 X. Just as in Exercise 6.4, we choose to depict such a map f as a rectangle I 2 with the basepoint x0 along the edges to remind us that x0 is where I 2 gets mapped to:
162
[email protected]
x0 x0 f x0 . We will denote the homotopy classes of such maps by 2 (X, x0 ). Just like for 1 , we can turn this space into a group. We rst discuss composition. Two maps S 2 X can be composed by adjoining the squares, just like we did for loops. However in this case there seems to be an ambiguity: we can adjoin the squares horizontally or vertically. We will see that there is indeed no such ambiguity, but for the present we choose to resolve it by composing them horizontally. In other words, if f and g are two maps I 2 X, we dene the composition f g by (f g)(t1 , t2 ) =
1 f (2t1 , t2 ) for t1 [0, 2 ], g(2t1 1, t2 ) for t1 [ 1 , 1]. 2
x0
Clearly the resulting map is continuous, since the boundary conditions agree: f (1, t2 ) = g(0, t2 ) = x0 for all t2 . Pictorially this composition corresponds to the following diagram: x0 x0 f x0 x0 x0 x0 g x0 x0 = x0 f x0 x0 g x0
Just like we did for 1 (X, x0 ), we can show that 2 (X, x0 ) is a group.
Exercise 6.8 (2 (X, x0 ) is a group) This exercise follows similar steps to Exercise 6.4. Prove the following: is well-dened in homotopy. is associative up to homotopy.
163
[email protected]
2 (X, x0 ) has an identity. Prove that the constant map sending all of I 2 to x0 is an identity for up to homotopy. Inverses exist. Let f be a map I 2 X, let f denote the map obtained by following f backwards in the rst of the two times: f (t1 , t2 ) = f (1 t1 , t2 ). is the inverse of f up to homotopy. Prove that f
If X is path-connected, it follows that 2 (X, x0 ) doesnt depend on the basepoint (up to isomorphism). Indeed, let f represent a homotopy class in 2 (X, x0 ). Given any other basepoint x1 X, let be a path from x0 to x1 . The following diagram represents a homotopy class in 2 (X, x1 ): x1
d s s d d s d
x0 f

x1
x0

x0
d d d
x1
x0 d x1
where the arrows represent the path going from x0 to x1 . The next exercise asks you to show that this map is an isomorphism.
Exercise 6.9 (2 (X, x0 ) 2 (X, x1 )) = Prove that the above map is an isomorphism 2 (X, x0 ) 2 (X, x1 ); and that the = isomorphism only depends on the homotopy class of the path used to dene it.
Hence for path-connected X it makes sense to talk about 2 (X) without reference to the basepoint, provided that we are only interested in its isomorphism class. This group is a higher-dimensional analogue of the fundamental group 1 (X). Unlike the fundamental group, 2 (X) is always abelian. The
164
[email protected]
following sequence of homotopies proves this assertion: x0 x0 f x0 g x0 = x0 f x0 x0 x0 = x0 x0 g x0 f x0 x0 x0 = x0 g x0 x0 g x0 = x0 x0 f g x0 x0 f x0 . (6.2) x0
This proof also shows that there is no ambiguity in the composition after all. Similarly one proves that also k (X) for k > 2 are abelian. Finally we discuss some examples of higher homotopy groups. Unlike the fundamental group, for which there are theorems (for instance, the Van Kampen theorem) allowing us to compute 1 (X) starting from a decomposition of X into simpler spaces, the computation of the higher homotopy groups is a very dicult problem. For example, not all the homotopy groups of the 2-sphere S 2 are known! Nevertheless, here are some examples of higher homotopy groups: k (S n ) = 0 for k < n, and n (S n ) Z. The rst equality is the = fact that one cannot k-lasso an n-orange, for k < n; whereas the last isomorphism is given by the degree of the map, as was the case for n=1. k (S 1 ) = 0 for k > 1. This follows because S 1 can be covered by a contractible space: S 1 R/Z. = k (T n ) = 0 for k > 1, where T n is an n-torus. This follows for the same reason as the above example: T n Rn /, where is some lattice. = k (g ) = 0 for k > 1, where g is a compact Riemann surface of genus g. Again, the proof is as above, since g can be written as the quotient of the Poincar upper half plane by a Fuchsian group: g H/. e = 165
[email protected]
2 (G) = 0, for any topological group G. This result of E. Cartan will play a very important role in the next section. The homotopy groups of the spaces determining a bration F E B, where F is the typical bre, E is the total space, and B is the base, are related by a useful gadget known as the exact homotopy sequence of the bration: n (B) n1 (F ) n1 (E) n1 (B) 2 (E) 2 (B) 1 (F ) 1 (E) 1 (B) . The exactness of this sequence simply means that every arrow is a group homomorphism such that its kernel (the normal subgroup sent to the identity) precisely agrees with the image of the preceding arrow. If the bration is principal, so that F is a Lie group, then the sequence extends one more term to include a map 1 (B) 0 (F ). Where 0 (F ) is the set of connected components of the typical bre. One can dene 0 (X) in this way for any space X, but for a general X, 0 (X) is only a set. It is when X = G is a group, that 0 (G) also inherits a group operation. Indeed, 0 (G) G/G0 , where G0 is the connected = component of the identity. It is in this case that it makes sense to speak of a group homomorphism 1 (B) 0 (G). A lot more could be said about higher homotopy groups, but this about covers all that we will need in the sequel.
6.1.2
Homotopy classication of nite-energy congurations
After this brief review of homotopy theory, we return to the problem at hand. Let us x a basepoint r 0 in the two-sphere at innity; for example, we could choose the north pole. Let (r 0 ) = 0 M0 . The G-orbit of 0 will be the set G 0 = {g 0 |g G} M0 . If we let H = H0 G denote the stability subgroup of 0 : H0 = {h G|h 0 = 0 }, then G 0 G/H. = The asymptotics of the Higgs eld dene a map from the two-sphere to G/H taking the basepoint to 0 . In other words it denes an element in the second homotopy group 2 (G/H, 0 ). Because G is connected, this class is gauge-invariant, as the next exercise asks you to show. This shows that the homotopy class is physical.
Exercise 6.10 (Gauge invariance of the homotopy class) Prove that gauge related Higgs eld congurations dene asymptotics which are homotopic.
166
[email protected]
(Hint: Use the fact that G is connected to write down an explicit homotopy between the two congurations.)
As shown in the previous section, 2 (G/H, 0 ) is an abelian group, and because G/H is connected it does not depend (up to isomorphism) on 0 . Because of this fact we will drop the reference to the basepoint when unnecessary. We will now prove that 2 (G/H) is isomorphic to the subgroup of 1 (H) given by those homotopy classes of loops in H which are null-homotopic in G. From what was said in the previous section, you will immediately recognise this statement as part of the exact homotopy sequence associated to the bration H G G/H. But rather than appealing to such heavy machinery, we will prove most of this statement here using more pedestrian methods. We rst associate a loop in H with each . Let denote an open cover 2 for the asymptotic two-sphere S ; more concretely, we take their union to 2 be S and their intersection to be a small band around the equator. Since are homeomorphic to disks (hence contractible), we can nd local gauge transformations g : G such that (x) = g (x) 0 for x .
Then on the intersection + , we have g+ (x) 0 = g (x) 0 , whence g+ (x)1 g (x) 0 = 0 , whence g+ (x)1 g (x) denes an element of H. Restricting to the equator, we have a continuous map x h(x) = g+ (x)1 g (x) H; that is, a loop in H. Because g (x) are dened only up to right multiplication by H, we can always arrange so that h(x) is the identity for some x in the equator. This way h denes an element in 1 (H, 1). Furthermore, this loop is trivial in G. Indeed, since G is path-connected, there exist paths t g (t, x) from the identity to g (x). Dening h(t, x) = g+ (t, x)1 g (t, x) provides a homotopy in G from the identity to the loop h(x). Alternatively we can understand this in a more explicit way. A map 2 : S G/H can be thought of as a loop of loops: In the above picture, there is a family parametrised by the interval s [0, 1] of loops based at 0 and each loop in the family is in turn parametrised by the interval t [0, 1], with the condition that the initial and nal loops are trivial. Therefore we can redraw Figure 6.2 as a map from the square to G/H where the edges are mapped to 0 :
167
[email protected]
Figure 6.2: A map S 2 G/H as a loop of loops.
0 0
T Et E E E
where the three horizontal lines are precisely the three loops depicted above. Consider now a xed loop, that is, a xed value of s. Because Di = 0 on S 2 , we can solve for .
Exercise 6.11 (Solving for ) Prove that for a xed s, (s, t) given by
t
(s, t) = P exp e
0
dt Wi (s, t )
xi t
(6.3)
is a solution of Di with the boundary conditions (s, 0) = 0 . Here s, t are 2 coordinates for S and x(s, t) are the coordinates in R3 . Similarly Wi (s, t) is short for Wi (x(s, t)).
Let g(s, t) G be the group element dened by (s, t) = g(s, t) 0 in i (6.3). From its denition it follows that g(s, 0) = 1 and since x = 0 at t s = 0 and s = 1, it follows that g(0, t) = g(1, t) = 1. How about g(s, 1)? Because (s, 1) = 0 , g(s, 1) 0 = 0 , whence g(s, 1) = h(s) is in H. Since h(0) = h(1) = 1, as s varies, h(s) denes a loop in H:
1
h(s) P exp e
0
dt Wi (s, t)
xi t
(6.4)
168
[email protected]
Moreover, this loop in H is trivial in G, the homotopy being given by g(s, t) itself, as the following gure shows:
1 1
g(s, t) h(s)
1
This approach has the added benet that it is very easy to see that the map 2 (G/H) 1 (H) sending the class of (s, t) to the class of h(s) is a group homomorphism. Indeed rotating the squares, we have the composition: h(s) h (s) (h h )(s)
g(s, t)
g (s, t)
from where we see that if h and h , then h h . In summary we have a map 2 (G/H) 1 (H), sending (s, t) to h(s), which is a homomorphism of groups. Notice that both groups are abelian by Exercise 6.7 and equation (6.2). Also, because any loop in H can be thought of as a loop in G, we have a natural map 1 (H) 1 (G), which is also a homomorphism of abelian groups. We can thus compose these two maps, and what we have shown is that the composition: 2 (G/H) 1 (H) 1 (G) is zero. (6.5)
Conversely it follows readily from what we said above that any loop in H which is trivial in G of necessity comes from a map in 2 (G/H). Indeed, if h(s) is a loop in H which is null-homotopic in G, let g(s, t) denote the homotopy in G. We can then dene (s, t) = g(s, t) 0 . It is not hard to convince oneself that this (s, t) gives rise to a loop in H which is homotopic to the one from which we obtained it. In other words, we have proven that the above sequence (6.5) is exact at 1 (H), since the kernel of the arrow leaving 1 (H) coincides with the image 169
[email protected]
of the arrow entering 1 (H). We are still not done, though: for we still have to show that the map 2 (G/H) 1 (H) is one-to-one; that is, that if the loop h(s) in H dened by (s, t) is null-homotopic in H, then the map (s, t) was already null-homotopic in G/H. Indeed, suppose that there exists a homotopy in H interpolating between h(s) and the constant loop based at the identity; that is, that there exists a map h(s, t):
1
h(s) h(s, t)
1
Then we can compose this with g(s, t) as follows:
1 1
g(s, t) h(s) h(s)
1
h(s, t)
1 1
=
g (s, t)
Now dene (s, t) = g (s, t) 0 . Because h(s, t) H, this map is homotopic to (s, t). In fact, acting on 0 with the above maps we nd: 0 0 0 0 = 0 0 0 0 0 = 0 0 0 0
But now notice that g (s, t) denes an element in 2 (G). It is now that we must invoke the result of E. Cartan mentioned in the previous section, that
170
[email protected]
2 (G) = 0.1 This means that there exists a homotopy H(s, t, u) interpolating continuously between g (s, t) and 1. Acting on 0 , we see that H(s, t, u) 0 provides the desired homotopy between (s, t) and the constant map 0 . In summary we have proven that 2 (G/H) ker (1 (H) 1 (G)) , = or equivalently the exactness of the sequence: 0 2 (G/H) 1 (H) 1 (G) . In particular, if G is simply-connected, as is often the case, then every loop in G is null-homotopic, and we nd that 2 (G/H) 1 (H). = We can illustrate this theorem with a simple example. Suppose that G = SU (2) and H = U (1), then we have that SU (2)/U (1) S 2 and U (1) S 1 , and indeed 2 (S 2 ) 1 (S 1 ) Z. As an abelian group, Z is freely = = generated by 1, hence it suces to determine where 1 gets sent to under the map. In the above case the map 2 (SU (2)/U (1)) 1 (U (1)) sends the generator to the generator since the two groups are isomorphic, SU (2) being simply-connected. On the other hand now consider G = SO(3) and H = SO(2). Again we have that SO(3)/SO(2) S 2 and SO(2) S 1, hence as abstract abelian groups 2 (SO(3)/SO(2)) and 1 (SO(2)) are both isomorphic to Z, but the theorem says more. It says that the generator of 2 (SO(3)/SO(2)) cannot be sent to the generator of 1 (SO(2)), since the image of 2 (SO(3)/SO(2)) is not all of 1 (SO(2)) but only the kernel of the map 1 (SO(2)) Z Z2 1 (SO(3)). This map is simply reduction = = modulo 2 and its kernel consists of the even integers, that is, the subgroup generated by 2. Hence the generator of 2 (SO(3)/SO(2)) must get sent to twice the generator of 1 (SO(2)). This can also be understood more pictorially from the discussion surrounding Figure 6.1 and from Exercise 6.6, and it is a good exercise to do so. Adding topological charges We now briey discuss to what extent we can add the topological charges of distant monopoles. It is physically intuitive, despite the fact that the equations describing monopoles are nonlinear, that one should be able to patch distant monopoles together to form a multi-monopole solution. It also seems physically intuitive that the charge of this monopole solution should
There is to my knowledge no simple proof of this fact, but the interested reader is encouraged to go through the one in the book of Brcker and tom Dieck [BtD85]. o
1
(6.6)
171
[email protected]
be given purely in terms of the charges of the constituents and not depend on the details on how the solutions were patched together. We will now see, however that this is not quite right. We will see that when the unbroken gauge group H is disconnected there is an ambiguity in the addition of the monopole charges. Consider rst a monopole conguration. For convenience we will draw monopole congurations in such a way that the asymptotic sphere at spatial innity is brought forth to a nite distance from the origin. Physically, we are assuming that the elds reach their asymptotic values to a good approximation in a nite distance. More formally, we are Figure 6.3: A performing a conformal compactication of our space. This will be perfectly valid for our purposes, monopole. since many physical quantities (e.g., the monopole charge) will turn out to be conformally invariant. The Higgs conguration is gauge related to one which is constant almost everywhere on the asymptotic sphere. This is the so-called unitary gauge. More precisely, the unitary gauge is one where the Higgs eld is constant throughout the sphere. It follows that the unitary gauge is singular whenever the Higgs conguration has nontrivial topological charge, since we have seen that regular gauge transformations are homotopies. For our purposes it will be sucient to consider gauges in which the Higgs is constant almost everywhere on the sphere. Such congurations are depicted in Figure 6.3, where the Higgs is constant everywhere but in the shaded region at the south pole. Now suppose that we have two monopoles. They are assumed to be so separated that their asymptotic spheres do not intersect. In other words in the space on and outside their two asymptotic spheres, the elds have already attained their asymptotic values. It is as if the monopoles were non-intersecting bubbles in the Higgs vacuum. We can make a 2-monopole solution by patching together two monopoles in the following way. Let us denote by 1 and 2 the Higgs elds for each of the monopoles. It is of course necessary that the image of the Higgs elds lie in the same G-orbit of the manifold of vacua. In a sense dierent G-orbits are like dierent superselection sectors. We start by gauge transforming the Higgs elds in such a way that they are equal to 0 almost everywhere on their asymptotic spheres. We can further orient the monopoles in such a way that the regions in which 1 and 2 are allowed to uctuate do not intersect. We can do this independently for each monopole because we can always perform gauge transformations which are compactly supported in the sense that they are the identity far away from the centre of the monopole. After these 172
[email protected]
gauge transformations, we have a conguration where on and outside the asymptotic spheres (except for the shaded regions in Figure 6.4) the Higgs eld is constant and equal to 0 . In particular we have continuity in the Higgs eld along the dotted line in Figure 6.4. It is clear that the resulting eld conguration is continuous, but is this procedure unambiguous? Suppose that we had used a dierent set of gauge transformations in order to make 1 and 2 equal almost everywhere asymptotically. Once the Higgs are set to 0 we are still allowed to make an gauge transformation in the stability subgroup H. Would the resulting two-monopole conguration be homoFigure 6.4: Patching two monopoles. topic (i.e, gauge-equivalent) to the one resulting from our rst attempt at patching? Clearly if H is connected, then we can always make a homotopy so that any discontinuity in the H-gauge transformation can be undone. But how about if H is disconnected? In this case there is a potential ambiguity in the patching prescription. Another way to understand this is as follows. The question boils down to whether 2 (G/H, 1 ) and 2 (G/H, 2 ) can be composed meaningfully. Because G is path-connected, we know that G/H is path-connected, hence 2 (G/H, 1 ) 2 (G/H, 1 ). But remember that this isomorphism depends = on the path used to connect 1 and 2 . If all such paths were homotopic that is, if G/H were simply-connectedthen all such isomorphisms would be one and the same and we could unambiguously compose elements in 2 (G/H, 1 ) and 2 (G/H, 2 ). In other words, if G/H were simply-connected, then we could add without ambiguity the topological charges of each of the monopoles constituting a given two-monopole solution to derive its charge. It turns out, thanks to Theorem 16.11 in [Ste51], that it is enough to check that H be connected. If H is connected, the theorem states, that the isomorphism 2 (G/H, 1 ) 2 (G/H, 2 ) is independent of the path used to go between = 1 and 2 . If H is not connected, however, there is a potential ambiguity. We can patch separated monopoles together, but the topological charge of the resulting two-monopole conguration will not be given simply in terms of the topological charges of its constituents. We need more information: namely the details on how the solutions were put together. 173
[email protected]
6.2
The Dirac quantisation condition
In this section we start the analysis of the generalised Dirac quantisation condition obeyed by these monopole solutions. The results in this section are based on the seminal paper of Goddard, Nuyts and Olive [GNO77]. We start by considering a monopole in the unitary gauge, where the Higgs eld is constant and equal to 0 , say, almost everywhere on the asymptotic sphere. Looking back at Figure 6.3, we have = 0 everywhere on the sphere but on the shaded region around the south pole. Because Di = 0 everywhere on the sphere, on the part where is constant, this condition becomes Wi = 0, hence Wi takes values in the Lie algebra h of the stability subgroup H G, and so hence so does the path-ordered exponential in equation (6.3). We will now assume that the eld strength Gij has the following asymptotic form: xk Q(x) Gij = ijk 3 , (6.7) |x| 4 where the magnetic charge Q(x) is Lie algebra valued, and hence takes values 2 in h almost everywhere on S . It may seem surprising at rst that the magnetic charge is not constant; but in the presence of a non-abelian gauge symmetry, constancy is not a gauge-invariant statement. The correct nonabelian generalisation of constancy is covariantly constant; and, as the next exercise asks you to show, this is indeed the case.
Exercise 6.12 (The magnetic charge is covariantly constant) Prove that Di Q(x) = 0 on the sphere. (Hint: Analyse the Bianchi identity and the equations of motion on the asymptotic sphere in the Ansatz (6.7) and show that Bianchi identity xk Dk Q(x) = 0 equation of motion
k ijk x Dj Q(x)
=0.
Deduce that these two equations together imply that Dk Q = 0.)
The quantisation condition will come from demanding that the map h(s) dened in (6.4) does indeed trace a loop in H, so that h(1) = h(0). To make this condition into something amenable to computation we will derive another expression for h(1) in the unitary gauge. This will bring to play a non-abelian version of Stokes theorem. Let us dene the group element g(s, t) by (s, t) = g(s, t) 0 in (6.3). Because Di = 0, it follows that Di g(s, t) = 0, where the covariant derivative is now in the adjoint representation. Dene the covariant derivative along 174
[email protected]
i
the curves of constant s, by Dt = x Di , and the covariant derivative along t xi the curves of constant t, by Ds = s Di . The next exercise asks you to prove the non-abelian Stokes theorem.
Exercise 6.13 (The non-abelian Stokes theorem) The point of this exercise is to prove the following formula. h(s)1 We proceed in steps: 1. Use the fact that Dt (s, t) = 0 implies Dt g(s, t) = 0, provided the curve s = constant lies in the region where (s, t) = 0 is constant, and prove that this is equivalent to Dt g(s, t) = g(s, t) t as operators. 2. Show that t g(s, t)1 Ds g(s, t) = g(s, t)1 [Dt , Ds ]g(s, t) = e xi xj g(s, t)1 Gij g(s, t) . t s dh(s) = e ds
1 0
dt g(s, t)1 Gij g(s, t)
xi xj . t s
(6.8)
3. Finally integrate the above expression over t [0, 1] and use the fact that g(s, t)1 Ds g(s, t) t=0 = 0 because g(s, 0) = 1 and xi (s, 0) is constant for all s, and that g(s, t)1 Ds g(s, t)
t=0
= h(s)1 dh(s) . ds
With the Ansatz (6.7) for Gij we now have that g(s, t)1 Gij g(s, t) = 1 4
ijk
xk g(s, t)1 Q(x)g(s, t) . 3 |x|
But now notice that because Q(x) is covariantly constant, g(s, t)1 Q(x(s, t))g(s, t) = Q(x(0, 0)) Q h . In other words, h(s)
1 dh(s)
ds
e = Q 4
1 0
xk xi xj dt ijk 3 , |x| t s
which can be trivially solved for h(s) to yield: h(s) = exp e Q 4

s 1
ds
0 0
dt
xk xi xj h(0) , ijk |x|3 t s
175
[email protected]
and in particular e h(1) = exp Q 4

1 1
ds
0 0
dt
ijk
xk xi xj h(0) , |x|3 t s
(6.9)
where exp : h H is the exponential map. Alternatively, if you are more familiar with matrix groups, you can embed H inside a matrix group (every Lie group has a faithful nite-dimensional matrix representation) and then the above dierential equation for the matrix h(s) is solved by the above expression, but where exp of a matrix is now dened by its power series. Let us rst compute the above integral. Notice that because the integrand is invariant under rescalings of x we can evaluate it on the unit sphere in R3 ; that is, we take |x| = 1. We then rewrite it in a more invariant looking form. To this end, it suces to notice that the integrand is the pull back via the embedding S 2 R3 , (s, t) xi (s, t) of the form = 1 ijk xk dxi dxj , whose 2 1 exterior derivative d = 2 ijk dxi dxj dxk is precisely 3 times the volume form in R3 relative to the standard euclidean metric. Therefore using Stokes theorem and understanding the unit sphere in R3 as the boundary of the unit ball S 2 = B 3 , we have that =
B 3 B3
d = 3 vol(B 3 ) = 4 .
We can then rewrite equation (6.9) as follows h(1) = exp [eQ] h(0) , whence the the Dirac quantisation condition h(1) = h(0), becomes exp eQ = 1 H . (6.10)
Before undertaking a general analysis of this equation, let us make sure that it reduces to the familiar condition (1.6) for G = SO(3) and a nonzero Higgs eld in the adjoint representation, as was the case in Chapter 1. The adjoint representation of SO(3) is three-dimensional. Choose the Higgs eld to point in the z-direction. The stability subgroup if the subgroup of rotations about the z-axis. It is the SO(2) subgroup consisting of matrices of the form cos sin 0 sin cos 0 , 0 0 1
176
[email protected]
where runs from 0 to 2. The magnetic charge Q is given by: 0 1 0 Q = g 1 0 0 , 0 0 1 where g is what appears in (1.5). With these denitions, we see that the Dirac quantisation condition (6.10) becomes 0 1 0 cos(eg) sin(eg) 0 exp eg 1 0 0 = sin(eg) cos(eg) 0 = 1 , 0 0 1 0 0 1 whence eg 2 Z in complete agreement with (1.6). We are clearly on the right track. In order to analyse the Dirac quantisation condition properly we will need quite a bit of technology concerning compact Lie groups. This is the purpose of the following section. Readers who already know this material are encouraged to skim through the section for notation.
6.3
Some facts about compact Lie groups and Lie algebras
In this section we collect without proof those results from the theory of compact Lie groups that are relevant for the analysis of the Dirac quantisation condition. There are many ne books on the subject. A quick and ecient introduction to the main results can be found in the second chapter of Pressley and Segals book on loop groups [PS86]. A fuller treatment of the parts we will need can be can be found in the book by Adams [Ada69] and also in the more comprehensive book by Brcker and tom Dieck [BtD85]. For the o results on Lie algebras we have followed the book by Humphreys [Hum72].
6.3.1
Compact Lie groups
Suppose that G is a compact connected Lie group. Any connected abelian subgroup is clearly a torus. Let T be a xed maximal connected abelian subgroup of G; that is, a maximal torus. Maximal tori obviously exist because any one-parameter subgroup is a connected abelian subgroup. One of the key theorems in the structure of compact Lie groups is the fact that all maximal tori are conjugate in G. This implies, in particular, that the dimension of all maximal tori are the same: it is an invariant of G known as the rank of G. Another way to rephrase this theorem is that any element in G is conjugate 177
[email protected]
to an element in T , or simply that every group element in G lies in some maximal torus. Generic elements will lie in just one maximal torus: these are called regular elements, whereas there exist also singular elements which lie in more than one. The prototypical compact connected Lie group is U (n) and many of the results in the theory of compact Lie groups, when restricted to U (n), reduce to well-known facts. For instance, a maximal torus in U (n) can be taken to be the set of diagonal matrices; hence the rank of U (n) is n, and the rank for the SU (n) subgroup is n1. The theorem about maximal tori being conjugate, is simply the fact that any set of commuting unitary matrices can be simultaneously diagonalised by a unitary transformation. The regular elements are those matrices which have distinct eigenvalues. Let g and t denote the Lie algebras of G and T , respectively. t is a maximal toral subalgebra. A lot can be learned about G by studying the action of T on g. Because T is abelian, any nite-dimensional complex representation is completely reducible into one-dimensional representations. But g is a real representation, so we complexify it rst: dene gC = g C, and extend the action of G (and hence the one of T ) complex-linearly. We can now decompose gC as representations of T as follows: gC = tC
where tC = t C is the subspace on which T acts trivially, and g is the subspace of gC dened as follows: v g exp X v = ei(X) v , where X t and : t R is a real linear function. The s appearing in the above decomposition are known as the (innitesimal) roots of G. Notice that if is a root, so is since if v g its complex conjugate v g . The complexied Lie algebra of U (n) is the Lie algebra of all nn complex matrices. The roots are given by ij where 1 i, j n, i = j, and the root subspace corresponding to ij is spanned by the matrices Eij with a 1 in the (ij) entry and zeroes everywhere else. Acting on the diagonal matrix X = diag(x1 , x2 , . . . , xn ) t, ij (X) = xi xj . Let us think of U (1) as the group of complex numbers of unit norm. A homomorphism : T U (1) is called a character of T . Characters can be multiplied pointwise and indeed form a group called the character group of T . Characters are uniquely determined by their derivatives at the identity. In other words, if is a character and exp X belongs to T , then (exp X) = eiw(X) , 178 (6.11)
[email protected]
where w t is an innitesimal character or a weight. The set of innitesimal characters dene a lattice in t called the weight lattice of G and denoted w (G). The roots are particular examples of weights, and taking integer linear combinations of the roots, we obtain a sublattice of the weight lattice known as the root lattice and denoted r (G). The root lattice only depends on the Lie algebra, whence Lie groups sharing the same Lie algebra have the same root lattice. On the other hand, the weight lattice identies the Lie group. If G is semisimple, then both the weight and root lattices span t . It means that the quotient w (G)/r (G) is a nite abelian group. We will see later that it is the fundamental group of the dual group of G. We can illustrate this with SU (2) and SO(3). The weights of SU (2) form a one-dimensional integral lattice isomorphic to Z, shown below. The weight m Z corresponds to twice the magnetic quantum number, since for SU (2) the magnetic quantum number, like the spin, can be half-integral. In the case of SO(3) only integral spin representations can occur, hence its weight lattice (shown below with lled circles) corresponds to the sublattice consisting of even integers:
3 e 2
In the semisimple case, t is called the root space of g. On the other hand, if G is not semisimple, the roots will only span a subspace of t which is then the root space of its maximal semisimple subalgebra [g, g] g. This shows that the Lie algebra of a compact Lie group is reductive; that is, the direct product of a semisimple Lie algebra and an abelian algebranamely, its centre. From the denition of the roots of U (n) above we see that they dont span t , since they annihilate the scalar matrices. This is to be expected since the scalar matrices are in the centre of the Lie algebra u(n) of U (n). The traceless matrices in u(n) span the complement of the scalars matrices and generate the Lie algebra su(n) of SU (n), which is semisimple (in fact, simple). The space spanned by the roots is the root space of su(n). The root subspaces g are one-dimensional. Choose vectors e g such that e = e . Then e , e and their bracket h = i[e , e ] t dene an embedding of sl(2, C) in gC : [h , e ] = 2ie [h , e ] = 2ie and [e , e ] = ih .
Explicitly the embedding is given by e e , f e and h h , where e= 0 1 0 0 f= 0 0 1 0 179 and h= i 0 0 i .
[email protected]
It therefore follows that exp(2h ) = 1. It also follows from the representation theory of sl(2, C) that for any root t , (h ) Z and that, in particular, (h ) = 2. The h are known as coroots and their integer linear combinations span a lattice in t called the coroot lattice and denoted (G). r If is a lattice, then the dual lattice is the set of linear functions Z and is denoted . This relation is reexive because = . In this notation we now see that the coroot lattice is a sublattice of the dual root lattice: (G) r (G) . We will see later that the two lattices will agree when G r is simply-connected. Despite the fact that the coroot lattice lives naturally in t, one often sees in the literature where the coroot lattice is a lattice in t , just like the root and weight lattices. In my opinion this causes more confusion than it is worth, but for the sake of comparison let us see how this goes. In order to identify t and t we need a new piece of information: namely, a metric. We saw in Exercise 6.1 that the Lie algebra of every compact Lie group has an invariant metric, so we will x one such G-invariant metric (, ) on g. Its restriction to T will also be denoted (, ). Because this metric is invariant and non-degenerate, we can use it to identify t and t . In particular there is an element t such that for all X t, (X) = (h , X). In terms of the root , we have that = 2/(, ). We call the inverse root corresponding to the root . Taking integer linear combinations of the coroots, we span the inverse root lattice of G. The weight, root and inverse root lattices are all subsets of t , but notice that whereas the weight and root lattices are intrinsic, the inverse root lattice depends on the chosen metric. In particular the inverse root lattice cannot be meaningfully compared with either the root or weight lattices, since we can scale it at will by rescaling the metric. As we do not wish to advocate its use, we will not give it a symbol, but the reader should beware that sometimes the symbol we use for the coroot lattice is reserved for the inverse root lattice, relative to some standard metric. Using the metric on t one can measure lengths of roots, and it can be proven that if g is simple, then there are at most two lengths of roots, called long and short roots. Simple Lie algebras for which all roots are the same length are called simply-laced . For these simple Lie algebras, we can choose the metric so that (, ) = 2 for all roots. Under this metric, the roots and the inverse roots agree.
6.3.2
The Weyl group
Because the maximal torus T is abelian, conjugation by elements of T is trivial. Moreover generic elements of G will conjugate T to another maximal 180
[email protected]
torus. However there are some elements of G which conjugate T back to T . The largest such subgroup of G is called the normaliser of T and is denoted N (T ); that is, N (T ) = h G | hT h1 = T . It follows from this denition that N (T ) is indeed a subgroup of G and that T is contained in N (T ) as a normal subgroup. Because T N (T ) is a normal subgroup, it follows that N (T )/T is a group. This group is the Weyl group of G relative to the maximal torus T . It is the group of symmetries of the maximal torus. Although it is dened relative to T , the Weyl group N (T )/T corresponding to any other maximal torus T is conjugate (and hence isomorphic) to N (T )/T . Hence it makes sense to talk about the Weyl group W of G, up to isomorphism. The Weyl group W is a nite group, generated by reections corresponding to the roots. More precisely, if is a root, then consider the group element exp (e + e ) N (T ) . 2 The adjoint action of this group element on t corresponds to a reection on the reection hyperplane H t dened by H = {X t | (X) = 0}. Indeed, one computes that for all X t, (X) = X (X)h . It can be proven that the generate W . For example, the Weyl group of U (n) is Sn , the symmetric group in n objects, and it acts by permuting the entries of the diagonal matrices in t. This is also the Weyl group of SU (n) since the roots of U (n) are the roots of SU (n). Elements of t not belonging to any hyperplane H are called regular ; whereas those who are not regular are called singular . Regular elements fall into connected components called Weyl chambers. The Weyl group permutes the Weyl chambers and no two elements in the same Weyl chamber are Weylrelated. Fix a Weyl chamber C and call it positive. The roots can then be split into two sets, positive and negative roots, according to whether they are positive or negative on Cthey cannot be zero, because C does not intersect any hyperplane H . A positive root is called simple if H is a wall of C. If G is a simple group of rank , then there are simple roots. Every positive root is a linear combination of the simple roots with nonnegative integer coecients, hence the simple roots generate the root lattice, and their associated reections generate the Weyl group. The positive Weyl chamber is sometimes called the fundamental Weyl chamber . Its closure 181
[email protected]
(that is, including the walls) is a fundamental domain for the action of the Weyl group on t: every point in t is Weyl-related to a unique point in the closure of the fundamental Weyl chamber. In the case of U (n), a choice of fundamental Weyl chamber consists in choosing diagonal matrices whose entries are ordered in a particular way. For instance, we can choose a descending order, in which case the positive roots of U (n) are the ij with i < j. The simple roots are then clearly the i,i+1 . Again using the metric on t there is a dual picture of this construction in t , where the hyperplanes H now are dened as the hyperplanes perpendicular to the roots. This picture is independent of the metric since the notion of perpendicularity does not depend on the choice of G-invariant metric on t. The Weyl group acts on t and it is once again generated by reections associated to every root. If and are roots, we have () = (, ) = (, ) , (6.12)
where = 2/(, ) is the inverse root. Once again the complement of the hyperplanes is divided into connected components called the dual Weyl chambers and one any one of them can be chosen to be the positive or fundamental dual Weyl chamber. The walls of the fundamental dual Weyl chamber are the hyperplanes perpendicular to the simple roots. Once again the closure of the fundamental dual Weyl chamber is a fundamental domain for the action of the Weyl group in t . Those weights of G which lie in the closure of the fundamental dual Weyl chamber are called dominant. We write this set + (G). It is a semigroup of w w (W ); that is, if w1 and w2 are dominant, so is their sum w1 + w2 , but there are no inverses. Every irreducible representation of G has a unique highest weight which is dominant. Therefore + (W ) is in one-to-one correspondence with the set of nite-dimensional irreducible representations of G.
6.3.3
Root systems and simple Lie algebras
We have seen that the Lie algebra of a compact Lie group is reductive. Because semisimple Lie algebras split in turn into their simple factors, we see that the Lie algebra of a compact Lie group is a direct sum of abelian and simple Lie algebras. This does not mean that any compact Lie group is the direct product of simple Lie groups and a torus, but it turns out that it is covered nitely by a compact Lie group of this type. Hence to a large extent it is enough to study simple Lie groups and abelian Lie groups separately and only at the end put the structures together. Let us therefore assume that G is a simple Lie group. It is a remarkable fact that compact simple Lie groups 182
[email protected]
are essentially classied up to nite ambiguity by their root systems. We will begin to describe this process now. First of all we need to axiomatise the notion of a root system. A subset of a euclidean space E is a root system if the following conditions are obeyed: 1. is nite, spans E and does not contain the origin; 2. If , then and no other multiples of are in ; 3. The reections (see (6.12)) leave invariant; and 4. For all , , ( , ) Z, where (, ) is the metric in E. This last condition is extremely restrictive. It essentially says that only very few angles can occur between roots. Indeed, notice that ( , ) = 2||/|| cos , where is the angle between and . Now, ( , )( , ) = 4 cos2 is a non-negative integer. Taking into account that ( , ) and ( , ) have the same sign we are left with the possibilities listed in Table 6.1, where we have chosen (, ) < (, ) for deniteness and have omitted the trivial case = . (, ) 0 1 1 1 1 1 1 ( , ) 0 1 1 2 2 3 3 /2 /3 2/3 /4 3/4 /6 5/6 (, )/(, ) undetermined 1 1 2 2 3 3
Table 6.1: Allowed angles between roots in a root system. If a root system admits a split = 1 2 into disjoint sets so that every element of 1 is orthogonal to every element of 2 , we say that it is reducible; otherwise it is simple. The simple root systems have been classied. There are four innite families: A , for 1, B and C for 2, and D for 3; and ve exceptional root systems G2 , F4 , E6 , E7 and E8 . There are two accidental isomorphisms in the above list: B2 = C2 and A3 = D3 . In all cases, the subscript indicates the rank. The simply-laced root systems are those in the A, D and E series. They are listed in Table 6.2 in a graphical notation that will be explained shortly. 183
[email protected]
Notice that the above denition of a root system is symmetrical with respect with the interchange of a root and the inverse root. In particular this shows that the set E consisting of the dual roots is again a root system, and that it is simple if is. In this case must be again one of the simple root systems listed above. From the denition of , it follows that for simply-laced root systems (where all roots have the same length) we can choose the metric so that = , hence simply laced root systems are self-dual. More generally, since ( , ) = 4/(, ), long and short roots are interchanged. A quick look at Table 6.2 reveals that G2 and F4 are also self-dual, whereas B = C , and viceversa, since = . Let 1 , 2 , . . . , be a set of simple roots. Let i = 2i /(i , i ). Then the inner product aij = (i , j ) is an integer. The set {aij } of all such integers are called the Cartan integers and the matrix (aij ) is known as the Cartan matrix . They are independent of the invariant metric chosen for t. If two complex simple Lie algebras have the same Cartan matrix, then they are isomorphic. The Cartan matrices of the simple Lie algebras are listed in [Hum72], for example. There is also a graphical notation for root systems. Let be a root system of rank , and let 1 , 2 , . . . , be a set of simple roots. The Coxeter graph of is the graph consisting of vertices and such that the ith vertex is joined to the jth vertex by (i , j )(i , j ) lines. From Table 6.1, we know that this number can be 0, 1, 2 or 3. The Coxeter graph can be shown to determine the Weyl group, but does not determine the root system because when two vertices are connected by more than one line, it fails to tell us which of the two vertices corresponds to the shorter root. In other words, the Coxeter graph cannot tell between and . In order to distinguish them it is necessary to decorate the diagram further: we colour those vertices which corresponds to the short roots, if any are present. The resulting diagram is called the Dynkin diagram. The Dynkin diagrams corresponding to the simple root systems are listed in Table 6.2, the vertex labelled i corresponds to i , and the lled vertices corresponds to the short roots. Reconstructing the group From the above discussion about compact Lie groups it follows that the root system associated to a compact simple Lie group is simple. Hence it has to be one of the roots systems listed above. This prompts the question of the reconstruction of the group from the root system. It turns out that this is possible up to a nite ambiguity. In a nutshell, given the root system of a compact group, one can obtain a nite covering group of the group in question. In this section we will consider only simple Lie groups. 184
[email protected]
A B C
1
Dynkin diagram
e e u
2
e 1 e
1
e u e
e 2 e u 2 u
e u
e 3 e e
2
e2
E6
4 e
2 5
E7
4 e
2 5
E8 F4 G2
e e u
e e e
4 e 3
e u
Table 6.2: Dynkin diagrams of the simple root systems. Given an simple root system of rank , one can construct a unique simple complex Lie algebra. Associated with each simple root i there exist three generators ei = ei , fi = ei and hi = hi . These 3 elements, subject to the so-called Serre relations (see [Hum72], for example) generate a complex simple Lie algebra gC , where tC is spanned by the hi and whose root system relative to this maximal torus is . As a linear space, the Lie algebra will be generated by elements hi spanning the Cartan subalgebra and generators e for each root , whose Lie brackets can be written down as follows: n e+ if + is a root; [e , e ] = h if = ; and 0 otherwise, where h belongs to the Cartan subalgebra. Furthermore, for every h in the Cartan subalgebra, [h, e ] = (h)e . It is possible to choose a basis in which the n are all nonzero integers. These integers obey n = n, . 185
[email protected]
Every complex simple Lie algebra has in general several real forms; that is, real subalgebras. Among these real forms there is a unique compact real form; that is, one for which the Killing form is negative-denite. It is easy to write this form down explicitly. It is generated over the reals by the ih , and the combinations i (e + e ) 2 and 1 (e e ) . 2
It is easy to check that the real linear combinations of these generators close under the Lie bracket. It is also easy to compute the Killing form and see that it is indeed negative-denite. Next, each compact real form g of a simple Lie algebra is the Lie algebra of a unique simply-connected compact simple Lie group G, whose maximal torus T is obtained by exponentiating the {ih }, and whose root system relative to T agrees with the one of G. Therefore we have almost come full circle. I say almost, because we are left with a simply-connected compact simple Lie group, even though we started with a compact simple Lie group G which was not assumed to be simply-connected. Therefore we need more information. The information we need is of course the fundamental group 1 (G) of G, which is a nite subgroup of the centre of G. Remarkably, the centre of G can be read o simply from Lie algebraic data. We review this now. The centre of G Let {i } for i = 1, . . . , denote the simple roots of G relative to a maximal torus T . We dene fundamental weights {i } by the requirement: (i , j ) = ij . Alternatively, i (hj ) = ij . In other words, the fundamental weights generate a lattice which is dual to the coroot lattice (G). The r lattice generated by the fundamental weights is the weight lattice of the simply-connected Lie group G: w (G). This lattice contains the root lattice as a sublattice, and the quotient w (G)/r (G) is a nite abelian group r (G) isomorphic to the centre Z(G) of G. It is sometimes called the fundamental group of the root system, since it is the fundamental group of the adjoint group, the Lie group whose weight lattice agrees with its root lattice. Let us now explain why w (G)/r (G) is isomorphic to the centre Z(G) First of all, notice that Z(G) is contained in every maximal torus of of G. G. In fact, Z(G) is the intersection of all the maximal tori of G.
Exercise 6.14 (The centre is the intersection of all the maximal tori) Prove that the centre Z(G) is the intersection of all the maximal tori of G.
186
[email protected]
(Hint: Use that any element can be conjugated to any given maximal torus and the fact that an element of the centre is invariant under conjugation.)
Fix a maximal torus T of G and let exp : t T be the restriction of is abelian, the exponential map is a the exponential map to t. Because T homomorphism of abelian groups. We nd it convenient in what follows to include a factor of 2 in the exponential map. We will introduce then the reduced exponential map, denoted exp, and dened by exp X = exp 2i X. Clearly exp is also a group homomorphism and, in particular, its kernel is a lattice I (G), called the integer lattice of G. The reduced exponential map t/I (G), whence we see that the integer lattice yields an isomorphism T = is the lattice of periods of the maximal torus T . It follows from (6.11) that h belongs to the integer lattice if and only if for every weight w of G, w(h) Z. In other words, the integer lattice and the weight lattice are dual: I (G) = w (G) . Let Z (G) = exp1 Z(G) denote those elements of t which the reduced ex ponential map sends to the centre of G. Z (G) too is a lattice called the which by denition contains the integer lattice. Because central lattice of G, exp is a group homomorphism, we have that Z(G) is canonically isomorphic to Z (G)/I (G). We now claim that the central lattice is the dual of the root lattice.
Exercise 6.15 (The central and root lattices are dual) Prove that X t belongs to the central lattice if and only if for every root t , (X) Z. (Hint: X belongs to the central lattice if and only exp 2i X is central in G, which in turn is equivalent to the statement that for every root , exp(2i X) exp(te ) = exp(te ) exp(2i X), for all t. Now use that [X, e ] = (X)e .)
= Therefore Z(G) Z (G)/I (G) = r (G) /w (G) , which as the next exercise asks you to show is isomorphic to w (G)/r (G) as we had claimed.
Exercise 6.16 (Some facts about lattices) Let 1 2 be lattices. Prove the following: Duality reverses inclusions: ; 1 2 1 /2 / ; and = 2 1 if 3 2 is a third lattice, then 1 /2 (1 /3 ) / (2 /3 ) . =
187
[email protected]
Since the root lattice is contained in the fundamental weight lattice, we can write i = j Mij j , for some integers Mij . Now, taking the inner product with j and using the denition of the fundamental weights, we nd that (i , j ) = Mij . In other words, (Mij ) is the Cartan matrix. Hence in order to write the fundamental weights in terms of the roots, it is necessary to invert the Cartan matrix. If the Cartan matrix has unit determinant then its inverse has integer entries and the fundamental weights belong to the root lattice. In this case the root lattice and the fundamental weight lattice agree, and G has no centre. In general, the order of the group Z(G) is given by the determinant of the Cartan matrix, since this is the only denominator in which we incur in the process of expressing the fundamental weights in terms of the roots. In many cases, the order of Z(G) is enough to determine the group uniquely. For example, the Cartan matrices of G2 , F4 and E8 have unit determinant, the ones of B , C and E7 have determinant 2, and the one of E6 has determinant 3. Hence the fundamental groups of the roots systems are respectively 1, Z2 and Z3 . In the other cases, the order does not generally determine the group, and one has to work a little harder: the Cartan matrix of A has determinant + 1, from where it follows that if + 1 is prime, then the fundamental group of A is Z +1 , since the only nite abelian group of prime order is the cyclic group. (Proof: Take any element not equal to the identity. It generates a cyclic subgroup whose order must divide the order of the group.) In fact, this persists for all , but this requires an explicit computation. Finally the Cartan matrix of D has determinant 4, which again does not determine the fundamental group. It turns out that for even the fundamental group is Z2 Z2 , whereas for odd it is Z4 . A useful mnemonic in this case is to remember that D3 = A3 . An example: A3 = D3 Let us in fact work out this example to see how to go about these calculations. Let us consider the root system A3 = D3 whose simply-connected compact Lie group is SU (4) Spin(6). We can read o the Cartan matrix from its = Dynkin diagram listed in Table 6.2: 2 1 0 1 2 1 , 0 1 2
188
[email protected]
which has indeed determinant 4. Inverting this matrix we can read o the expression for the fundamental weights in terms of the roots: 1 = 3 1 + 1 2 + 1 3 4 2 4 1 1 2 = 2 1 + 2 + 2 3 3 = 1 1 + 1 2 + 3 3 . 4 2 4 We can now compute the factor group w /r . Its elements are the cosets 0+r , 1 +r , 2 +r and 3 +r , which possess the following multiplication table: 0 1 2 3 0 0 1 2 3 1 1 2 3 0 , 2 2 3 0 1 3 3 0 1 2 where all entries are understood modulo r . It follows clearly that this is the cyclic group Z4 . Another example: D4 Finally let us work a second example. We pick now one of our favourite root systems: D4 , whose simply-connected compact Lie group is Spin(8). From Table 6.2 we can read o the Cartan matrix: 2 1 0 0 1 2 1 1 , 0 1 2 0 0 1 0 2 and inverting it, we can read o the expression of the fundamental weights in terms of the roots: 1 = 1 + 2 + 1 3 + 1 4 2 2 1 1 3 = 2 1 + 2 + 3 + 2 4 2 = 1 + 22 + 3 + 4 4 = 1 1 + 2 + 1 3 + 4 . 2 2
We can now compute the factor group w /r . Because 2 r , it has as elements the cosets of 0, 1 , 3 and 4 . The multiplication table for this group can be read o easily: 0 1 3 4 0 0 1 3 4 1 1 0 4 3 189 3 3 4 0 1 4 4 3 , 1 0
[email protected]
where all entries are understood modulo r . It is clear that this group is Z2 Z2 . It has three proper Z2 subgroups, each one generated by one of the cosets 1 + r , 3 + r and 4 + r . The representations with highest weights 1 , 3 and 4 are all eight-dimensional. They correspond to the vector and the two spinor representations of Spin(8). Alternatively, they correspond to the three inequivalent embeddings Spin(7) Spin(8). Given any one of these eight-dimensional representations there exists an Spin(7) subgroup of Spin(8) under which the representation remains irreducible and can be identied with the unique spinorial representation of Spin(7). The dihedral group D3 of automorphisms of the Dynkin diagram is the group of outer automorphisms of Spin(8). It is called the triality group in the physics literature, and it permutes the three inequivalent Spin(7) subgroups and thus the three eight-dimensional representations. In terms of the weights, it permutes 1 , 3 and 4 . All the connected compact simple Lie groups It is now time to summarise what we have learned so far in a table. Table 6.3 lists the simple root systems, their Weyl groups, the associated simple complex Lie algebras, their simply-connected simple compact Lie groups, and their centres. An eternal thorny issue about the notation in Table 6.3: the compact Lie group associated to the root system of type C is called in the physics literature U Sp(2 ) and in the mathematics literature Sp( ). From here until the end of this section, all Lie groups are connected, compact and simple unless otherwise explicitly stated . We start by associating with every Lie group G a subgroup of the centre of its universal cover G and viceversa. Representations of G are also repre sentations of G, whence the weight lattice w (G) is contained in the weight lattice w (G). We thus have the following inclusions of lattices in t : r (G) w (G) w (G) . Dualising and keeping in mind Exercise 6.16, we have in t the following lattices: Z (G) w (G) I (G) , where we have used that I (G) = w (G) , and that Z (G) = r (G) . Ap to these lattices and rememplying the reduced exponential map exp : t G bering that exp is a group homomorphism when restricted to the maximal torus, we nd that exp w (G) = G Z(G) is a subgroup of the centre. The subgroup G is naturally isomorphic to w (G) /I (G), since the integral lattice is the kernel of the reduced exponential map. Using the fact that 190
[email protected]
A B C D G2 F4 E6 E7 E8
W or |W | S +1 (Z2 ) S (Z2 ) S (Z2 ) 1 S
gC sl( + 1, C) so(2 + 1, C) sp(2 , C) so(2 , C) G2 F4 E6 E7 E8
G SU ( + 1) Spin(2 + 1) U Sp(2 ) Spin(2 ) G2 F4 E6 E7 E8
Z(G)
Z +1 Z2 Z2 Z4 Z2 Z2
1 1 is odd is even
D6 27 32 27 34 5 210 34 5 7 210 35 52 7
Z3 Z2
1
Table 6.3: Simple root systems, their Weyl groups, their complex Lie algebras, compact Lie groups and their centres. I (G) = w (G) , we have = G = w (G) /w (G) w (G)/w (G) , where we have again used Exercise 6.16. Since G is determined by its weight lattice, this actually tells us that G G/G . Since G is simply-connected, = G . this implies that 1 (G) = Conversely, if Z(G) is a subgroup of the centre of G. The preimage is a sublattice of the of via the reduced exponential map exp : t G, central lattice and contains the integer lattice: I (G) Z (G) , which upon dualising gives in t the following series of lattices: w (G) r (G) . (6.14) (6.13)
It is not hard to see that is the weight lattice of the group G dened by G/. In summary, there is a one-to-one correspondence between Lie groups with the same universal covering group G and subgroups of the centre Z(G); or, equivalently, between Lie groups with the same Lie algebra g and lattices containing the root lattice and contained in the fundamental weight lattice. Since the centre Z(G) is nite, it has a nite number of subgroups, and hence there are only a nite number of Lie groups covered by the same simplyconnected Lie group. This is what we meant earlier by nite ambiguity. 191
[email protected]
Minding Table 6.3, we can now list all the connected compact simple Lie groups. For the root systems E8 , F4 and G2 , the centre is trivial, so they are the only groups with that root system. Similarly, the centres of E6 , E7 , B and C are not trivial but have no proper nontrivial subgroups, hence there are only two groups associated with each of those root systems: the simply-connected group and the adjoint group: E6 and E6 /Z3 , E7 and E7 /Z2 , Spin(2 + 1) and SO(2 +1) = Spin(2 +1)/Z2 , and U Sp(2 ) and U Sp(2 )/Z2 . A similar story holds for A with +1 prime: there are only two groups with that root system, SU ( +1) and SU ( +1)/Z +1 . For general , however, the centre of SU ( +1) has subgroups corresponding to the divisors of +1: Zm Z +1 if and only if m divides + 1. So we have a whole hierarchy of groups SU ( + 1)/Zm where m runs over the divisors or + 1, interpolating between the simply-connected SU ( + 1) and the adjoint group SU ( + 1)/Z +1 . For the root system D2 +1 , the centre is Z4 which has a single nontrivial proper subgroup isomorphic to Z2 . Hence there are three groups: Spin(4 +2), SO(4 +2) = Spin(4 +2)/Z2 and Spin(4 + 2)/Z4 . Finally, the root systems D2 has centre Z2 Z2 which has three proper subgroups isomorphic to Z2 . Hence there are ve groups in this family: Spin(4 ), SO(4 ) = Spin(4 )/Z2 , Spin(4 )/Z2 , Spin(4 )/Z2 , and the adjoint group Spin(4 )/(Z2 Z2 ). In the next section we will see that many of these groups are mapped to each other by a duality transformation.
6.3.4
Some simple examples
We illustrate some of the results above with some simple examples: the simple root systems of rank 2: A2 , B2 = C2 and G2 . The simple root system A2 The root system A2 is dened by the Cartan matrix (Aij ) = 2 1 1 2 .
Therefore the simple roots are given in terms of the fundamental weights as follows: 1 = 21 2 and 2 = 1 + 22 . Inverting these relations we see 1 2 that 1 = 3 1 + 1 2 and 2 = 3 1 + 2 2 . This clearly shows that the order 3 3 of the fundamental group of A2 is 3, and hence that w /r Z3 . Indeed, = notice that this group has as elements the cosets 0 + r and 1 + r and
192
[email protected]
2 12 1 1 2
21
121
212
Figure 6.5: The root system A2 . 2 + r , with multiplication table: 0 1 2 0 1 2 0 1 2 1 2 0 2 0 1
where all entries are to be understood modulo r . We can choose a euclidean metric on R2 and represent these lattices pictorially. This is done in Figure 6.5, which also shows the hyperplanes perpendicular to the roots as dashed lines, and the positive dual Weyl chamber as shaded. The Weyl group is the dihedral group D3 S3 , the symmetries of an equilateral triangle, and it = clearly permutes the dual Weyl chambers. Indeed, in Figure 6.5 all chambers but the fundamental are labelled with the element of the Weyl with which it is associated. Since the Weyl group is generated by reections on the hyperplanes perpendicular to the simple roots, I have chosen to write the Weyl group elements in this way: the notation i means the reection i and ijk = i j k . The lled circles denes the root lattice r and these together with the open circle dene the weight lattice w . The fundamental 193
[email protected]
weights and simple roots are also shown. w is the weight lattice of the group SU (3), whereas r is the weight lattice of the adjoint group SU (3)/Z3 . The simple root systems B2 = C2
1 12
2 1 1
121 1212 = 2121 212
21
Figure 6.6: The root system B2 . The root system B2 is dened by the Cartan matrix (Aij ) = 2 1 2 2 .
Therefore the simple roots are given in terms of the fundamental weights as follows: 1 = 21 2 and 2 = 21 + 22 . Inverting these relations we see that 1 = 1 + 1 2 and 2 = 1 + 2 . This clearly shows that the order of 2 the fundamental group of B2 is 2, and hence that w /r Z2 . Again, one = can see this directly: the cosets 0 + r and 1 + r are the elements of the fundamental group with multiplication table 0 1 0 1 0 1 1 0 194
[email protected]
where all entries are again to be understood modulo r . We can choose a euclidean metric on R2 and represent these lattices pictorially. This is done in Figure 6.6, which also shows the hyperplanes perpendicular to the roots as dashed lines, and the positive dual Weyl chamber as shaded. The Weyl group is isomorphic to the dihedral group D4 of symmetries of the square, and again the Weyl chambers have been decorated with the corresponding element of the Weyl group. Once again the lled circles dene the root lattice r and these together with the open circles dene the weight lattice w . The fundamental weights and simple roots are also shown. w is the weight lattice of the group Spin(5) U Sp(4), whereas r is the weight lattice of the = group SO(5). The weight 1 is the highest weight of the irreducible spinorial representation of Spin(5) obtained as the unique irreducible representation of the Cliord algebra in ve-dimensional euclidean space. The dual root system C2 has as Cartan matrix the transpose of the Cartan matrix of B2 . They are of course isomorphic root systems, but the isomorphism interchanges long and short roots: 1 2 . This essentially rotates the root diagram by /4, and chooses a dierent fundamental dual Weyl chamber. The simple root system G2 The root system G2 is dened by the Cartan matrix (Aij ) = 2 1 3 2 .
Therefore the simple roots are given in terms of the fundamental weights as follows: 1 = 21 2 and 2 = 31 + 22 . Inverting these relations we see that 1 = 21 + 2 and 2 = 31 + 22 . Hence the root and weight lattices agree. We can choose a euclidean metric on R2 and represent this lattice pictorially. This is done in Figure 6.7, which also shows the hyperplanes perpendicular to the roots as dashed lines, and the positive dual Weyl chamber as shaded. The Weyl group is now D6 , the symmetries of the regular hexagon, and it permutes the Weyl chambers as shown in the gure. Now the open circles dene the root/weight lattice . The fundamental weights and simple roots are also shown. Notice that the long roots form a root system of type A2 , indicative of the fact that SU (3) is a maximal subgroup of G2 .
6.4
The magnetic dual of a compact Lie group
We now start to analyse the Dirac quantisation condition (6.10) in more detail. The punch-line is that the Dirac quantisation condition says that the 195
[email protected]
12
2 121 2 1 21
1212
212
12121 2121
121212 = 212121
21212
Figure 6.7: The root system G2 . magnetic charge (suitably normalised) is a dominant weight of a connected compact Lie group H , the (magnetic) dual group of H. First of all notice that it is irrelevant for these purposes that H be connected, since the image of the exponential map lies in the connected component of the identity. (Proof : If g = exp X, then g(t) = exp(1 t)X is a path to the identity.) Therefore we will assume from now on that H is connected. It is also compact since it is a closed subgroup of a compact Lie group. So we are in the situation that we have just discussed. Because physics is gauge invariant, we have to identify dierent charges Q which are gauge related via the unbroken gauge group H. Q belongs to the Lie algebra h of H and H acts on its Lie algebra via conjugation. One way to x this gauge invariance is to to choose a xed maximal torus T in H, with Lie algebra t and use our gauge freedom to conjugate Q to lie in t. As discussed above, this does not x the gauge completely, because there will be elements of H which stabilise T ; in other words, we have to still take into account the action of the Weyl group. The action of the Weyl group is xed by choosing Q in the closure of the fundamental Weyl chamber C, since this is a fundamental domain for the action of the Weyl group. 196
[email protected]
Therefore in the Dirac quantisation condition (6.10), we can take eQ to lie in the closure C of the fundamental Weyl chamber in t. The exponential map in (6.10) is then the exponential map t T , and (6.10) says that eQ/2 belongs to the integer lattice I (H) of H. We saw above that I (H) = w (H) , whence the Dirac quantisation condition becomes: eQ/2 w (H) /W w (H) C , = where W is the Weyl group. On the other hand, the integer lattice can be thought of as the weight lattice of a connected compact Lie group H known as the (magnetic) dual group of H. As we will see below, this group is a quotient of the simply-connected compact simple Lie group whose root system is dual to the root system of H. Now, dual root systems share the same Weyl group. This follows from the fact that the Weyl group is generated by those reections in (6.12) corresponding to simple roots. But from (6.12) it follows that = . Therefore we can x the Weyl symmetry by going to the fundamental dual Weyl chamber of H . In other words, the Dirac quantisation condition can be rewritten as eQ/2 w (H )/W + (H ) , = w where + (H ) are the dominant weights of H , which are in one-to-one w correspondence with the nite-dimensional irreducible representations of H . We now turn to a more detailed description of the dual group.
6.4.1
Some lattices and dual groups
In our ash review of compact Lie groups, we have already encountered several lattices. We will now review their interrelations and in particular how they can be used to describe the dual of a connected compact Lie group. For the purposes of studying the Dirac quantisation condition, we will take the unbroken gauge group H to be a compact connected Lie group. Such a group is covered nitely by a compact group H = K S, where K is a simply-connected compact Lie group (hence semisimple, and in turn the product of simple factors K1 Kp ) and S is a torus. S is the connected component of the identity of the centre of H. We dene the dual group H of H to be the compact connected Lie group whose weight lattice is dual to the weight lattice of H. From this it follows that just like H is a nite quotient of H, so will H be a nite quotient of H = K1 Kp S . It is impractical to treat the general case, so we will discuss separately the cases of H abelian and H simple. From these ingredients it should be possible to treat the case of general H should the urge arise. All Lie groups in this section are compact and connected unless stated otherwise. 197
[email protected]
H abelian If H is abelian, then it is a torus. Let h be its Lie algebra. The (reduced) exponential map is surjective and denes a dieomorphism H h/, where = h is the lattice of periods of H. As we reviewed above, this lattice is dual to the weight lattice w (H) h . By denition this is the weight lattice of the dual group H . Hence we have a dieomorphism H h /w (H). = Notice that h is identied with h . H simple Let H be a simple Lie group, H its universal covering group, and h its Lie algebra. Let T be a xed maximal torus and t h its Lie algebra. We let t be the space of linear forms t R. The root lattices r (H) and r (H) in t agree, since as explained above they only depend on the Lie algebra. We will therefore write it as r (h). The weight lattices w (H) and w (H) are dierent, with w (H) depending only on the Lie algebra again, since it is the lattice of fundamental weights. We will then often write it as w (h). We have the following inclusions: r (h) w (H) w (h) , (6.15)
where the rst inclusion is an equality when H is the adjoint group, and the last inclusion is an equality when H = H. From Table 6.3 we see that for E8 , F4 and H2 , the adjoint group is simply-connected, so that in these cases, and in these cases only, are both inclusions equalities. The dual of these lattices give rise to lattices in t. Dualising the lattices reverses the inclusions in (6.15), so we have r (h) w (H) w (h) . (6.16)
We have met some of these lattices before. w (h) = I (H) = (h), which r again only depends on the Lie algebra. Similarly, w (H) = I (H) is the integer lattice of H: those elements h T such that 2h lies in the kernel of the exponential map exp : t T . It clearly depends on H, as it will be dierent for H and for H. Now with the same notation as above, let us consider the inverse root system . As mentioned above, it is a simple root system. Therefore by the construction outlined above, there will be a complex simple Lie algebra h associated to , which has a unique compact real form h , which can be C integrated to a unique connected compact simply-connected Lie group H . by a nite We will now exhibit the dual group H of H as a quotient of H subgroup of its centre. 198
[email protected]
The inverse root system requires for its very denition the existence of the metric: = 2/(, ). We can undo this dependence by using the metric to map each t to a unique t such that if t , ( ) = (, ). We have met these before: they are nothing but the coroots h . The coroots generate a root system in t whose root lattice is the coroot lattice (h) of h and whose fundamental weight lattice is the dual lattice to the root lattice of h.
Exercise 6.17 (The dual fundamental weights) Let w (h ) t denote the lattice of fundamental weights of the dual root system. Prove that w (h ) = r (h) .
In other words, if we let t = t , then on t = t, we have a root lattice r (h ) = (h) = w (h) and a fundamental weight lattice w (h ) = r (h) . r On t = t there is also a notion of reduced exponential map exp : t T which is given by the canonical projection t t /r (h). The centre of H is given by = Z(H ) w (h )/r (h ) r (h) /w (h) w (h)/r (h) Z(H) , = = = where we have used Exercise 6.16. Now, by denition, the weight lattice w (H ) of H is dual of the weight lattice w (H) of H: w (H ) w (H) , which sits between the above two lattices in t: r (h ) w (H ) w (h ) . From the above discussion surrounding equations (6.13) and (6.14), we know that H is given by H / where Z(H ) is the subgroup of the centre of H dened by w (H ) /w (h ) = w (H)/r (h). But consider now the subgroup H Z(H) which denes H = H/. Taking into account Exercise 6.16, we nd that it is given by w (h)/w (H) (w (h)/r (h)) / (w (H)/r (h)) Z(H)/ , = = = whence ||| | = |Z(H)| . (6.17)
Let us now look at examples of dual groups. Above we listed the connected compact simple Lie groups. We now do the same for their duals. This has been done in [GNO77]. We list the results in Table 6.4. Most cases 199
[email protected]
H SU (pq)/Zp Spin(2 + 1) SO(2 + 1) SO(2 ) Spin(4 + 2) Spin(4 ) Spin(8 )/Z2 Spin(8 )/Z2 Spin(8 + 4)/Z2 G2 F4 E6 E7 E8
H SU (pq)/Zq U Sp(2 )/Z2 U Sp(2 ) SO(2 ) Spin(4 + 2)/Z4 Spin(4 )/(Z2 Z2 ) Spin(8 )/Z2 Spin(8 )/Z2 Spin(8 + 4)/Z2 G2 F4 E6 /Z3 E7 /Z2 E8
Table 6.4: The connected compact simple Lie groups and their duals. can be determined without any computation, but some of the D series turn out to be subtle, and require an explicit description of the weight and root lattices. They are listed, for example, in [Hum72]. Equation (6.17) tells us that the orders of and are complementary in |Z(H)| = |Z(H )|. This means that the dual of the simply-connected group H is the adjoint group H /Z(H ). This already tells us the last ve entries of Table 6.4 as well as the second, third, fth and sixth entries, and the special case p = 1, in the rst entry. But, in fact, the rest of the rst entry also requires no further calculation. Since any subgroup of a cyclic group is cyclic and is moreover unique, the dual of SU (pq)/Zp has to be SU (pq)/Zq , since given Zp Zpq there is a unique subgroup of Zpq of order q, and it is Zq . The same argument also applies to D2 +1 , since the centre is cyclic in this case: whence SO(4 + 2) is self-dual. For the groups with root system D2 one has to work harder. An example: Spin(8) and its quotients As an example we will work out the example of Spin(8) and its factor groups. The root system of Spin(8) is D4 and we worked out the Cartan matrix, the centre and the fundamental weights above. The lattice of fundamental weights w = Z i is the integer span of the fundamental weights i . The 200
[email protected]
root lattice is the sublattice of the fundamental weights generated by the combinations: 1 = 21 2 3 = 2 + 23 2 = 1 + 22 3 4 4 = 2 + 24 .
Equivalently it is the lattice consisting of elements 4 ni i where ni Z i=1 such that n1 , n3 and n4 are either all even or all odd. There are three intermediate lattices corresponding to the weight lattices of the three subgroups SO(8), Spin(8)/Z2 and Spin(8)/Z2 : 1 = r (r +1 ), 3 = r (r +3 ), and 4 = r (r + 4 ). Equivalently,
4
1 = {
i=1 4
ni i | n3 n4 (2)} ni i | n1 n4 (2)}
i=1 4
3 = { 4 = {
i=1
ni i | n1 n3 (2)} ,
all other integers ni unconstrained. We can easily nd a Z-basis for these lattices as follows: 1 = Z 1 , 2 , 3 4 3 = Z 2 , 3 , 1 4 4 = Z 2 , 4 , 1 3 . The dual picture is as follows. Take as a basis the canonical dual basis { } to the roots: i (j ) = i j . Their Z-span is the lattice and all r lattices of interest are contained in it, so their elements will be integer linear combinations of the i . Given a sublattice w described as the Z-span of some vectors vi in the weight lattice w , the dual lattice will be the sublattice given by the Z-span of the canonical dual basis v i to the vi . Let r j j vi = j Mi j , where Mi Z since is a sublattice of w . Similarly v i = j N i j j , where N i j Z. We can solve for N in terms of M and the Cartan matrix C as follows. By denition, v i (vj ) = i j , whence
i
i j = vj (v i ) = N i k Mj (k ) = N i k Mj (C 1 ) 201
k
[email protected]
k
where we have used that = (C 1 ) k . In other words, N = (CM 1 )t . Computing this for each of the lattices above, we nd: = Z 21 2 , 1 2 + 3 + 4 , 2 3 4 , 3 4 1 3 = Z 1 2 + 4 , 1 2 + 3 + 4 , 2 23 , 1 4 = Z 1 2 + 3 , 1 2 + 3 + 4 , 1 3 , 2 24 . 4 We can understand these lattices as sublattices of by changing basis to r the i and constraining the coecients. We nd
4
= { 1
i=1 4
ni i | n3 n4 (2)} ni i | n1 n4 (2)}
i=1 4
= { 3 = { 4
i=1
ni i | n1 n3 (2)} ,
whence we conclude that all three lattices are self-dual, in agreement with Table 6.4. Another example: Spin(12) and its quotients As a nal example, and to illustrate the other behaviour of the D2 series, we will work out the example of Spin(12) and its factor groups. The root system of Spin(12) is D6 , whose Cartan matrix follows from Table 6.2: 2 1 0 0 0 0 1 2 1 0 0 0 0 1 2 1 0 0 . 0 0 1 2 1 1 0 0 0 1 2 0 0 0 0 1 0 2 The fundamental weights are given by 1 2 3 4 5 6 = 1 + 2 + 3 + 4 + 1 5 + 1 6 2 2 = 1 + 22 + 23 + 24 + 5 + 6 = 1 + 22 + 33 + 34 + 3 5 + 3 6 2 2 = 1 + 22 + 33 + 44 + 25 + 26 = 1 1 + 2 + 3 3 + 24 + 3 5 + 6 2 2 2 1 3 = 2 1 + 2 + 2 3 + 24 + 5 + 3 6 . 2 202
[email protected]
It follows that the centre w /r Z2 Z2 consists of the following r -cosets: = 0, 1 , 5 and 6 , with multiplication table: 0 1 5 6 0 0 1 5 6 1 1 0 6 5 5 5 6 0 1 6 6 5 , 1 0
where as usual all entries are modulo r . Letting r = Z i , the root lattice is the sublattice r = Z i spanned by the following combinations: 1 = 21 2 3 = 2 + 23 4 5 = 4 + 25 2 = 1 + 22 3 4 = 3 + 24 5 6 6 = 4 + 26
6 Equivalently it is the lattice consisting of elements i=1 ni i where ni Z such that n1 + n3 , n5 and n6 are either all even or all odd. There are three intermediate lattices corresponding to the weight lattices of the three subgroups SO(12), Spin(12)/Z2 and Spin(12)/Z2 : 1 = r (r + 1 ), 5 = r (r + 5 ), and 6 = r (r + 6 ). Equivalently, 6
1 = {
i=1 6
ni i | n5 n6 (2)} ni i | n1 + n3 n6 (2)}
i=1 6
5 = { 6 = {
i=1
ni i | n1 + n3 n5 (2)} ,
all other integers ni unconstrained. We can easily nd a Z-basis for these lattices as follows: 1 = Z 1 , 2 , 3 , 4 , 5 6 5 = Z 2 , 4 , 5 , 1 + 3 , 1 + 6 , 3 + 6 6 = Z 2 , 4 , 6 , 1 + 3 , 1 + 5 , 3 + 5 .
203
[email protected]
Following the discussion given in the previous example, the dual lattices are given by = Z 21 2 , 1 22 + 3 , 2 23 + 4 , 3 24 + 5 + 6 , 1 5 6 , 4 5 6 = Z 1 2 + 3 6 , 1 3 + 4 6 , 3 24 + 5 + 6 , 5 1 22 + 3 , 5 26 , 1 3 + 6 , = Z 1 2 + 3 5 , 1 3 + 4 5 , 3 24 + 5 + 6 , 6 1 22 + 3 , 1 3 + 5 , 4 26 . We can understand these lattices as sublattices of by changing basis to r the i and constraining the coecients. We nd
6
={
i=1 6
ni i | n5 n6 (2)} ni i | n1 + n3 n5 (2)}
i=1 6
= { 5 = { 6
i=1
ni i | n1 + n3 n6 (2)} ,
whence we conclude that 1 is self-dual, whereas duality interchanges the groups whose weight lattices are 5 and 6 . It can be shown that the group whose weight lattice is 1 is SO(12). Again this is in agreement with Table 6.4.
204
Bibliography
[Ada69] [AH85] [AH88] [Bes86] [Blu94] [Bog76] [BT81] [BtD85] [Cal78] [Col77] J. F. Adams, Lectures on Lie groups, W. A. Benjamin, 1969. M. F. Atiyah and N. J. Hitchin, Low energy scattering of nonabelian monopoles, Phys. Lett. 107A (1985), 2125. , The geometry and dynamics of magnetic monopoles, Princeton University Press, 1988. A. L. Besse, Einstein manifolds, SpringerVerlag, 1986. J. D. Blum, Supersymmetric quantum mechanics of monopoles in N =4 YangMills theory, Phys. Lett. B333 (1994), 9297. E. B. Bogomolnyi, The stability of classical solutions, Soviet J. Nuc. Phys. 24 (1976), 449454. R. Bott and L. W. Tu, Dierential forms in algebraic topology, SpringerVerlag, 1981. T. Brcker and T. tom Dieck, Representations of compact Lie o groups, SpringerVerlag, 1985. C. Callias, Axial anomalies and index theorems on open spaces, Comm. Math. Phys. 62 (1978), 213. S. Coleman, Classical lumps and their quantum descendants, New Phenomena in Subnuclear Physics (A. Zichichi, ed.), Plenum, 1977, Erice lectures on t Hooft Polyakov monopoles, larger gauge groups, homotopy considerations; comparison with sine-Gordon.
[CPNS76] S. Coleman, S. Parke, A. Neveu, and C. M. Sommereld, Can one dent a dyon?, Phys. Rev. D15 (1976), 544545. [DHdV78] A. DAdda, R. Horsley, and P. di Vecchia, Supersymmetric magnetic monopoles and dyons, Phys. Lett. 76B (1978), 298302. 205
[email protected]
[Dir31] [Gau94] [GH78] [GM86] [GNO77] [Gol62]
P. A. M. Dirac, Quantised singularities in the electromagnetic eld, Proc. R. Soc. A133 (1931), 6072. J. P. Gauntlett, Low energy dynamics of N =2 supersymmetric monopoles, Nuc. Phys. 411B (1994), 433460. P. Griths and J. Harris, Principles of algebraic geometry, Wiley, 1978. G. W. Gibbons and N. S. Manton, Classical and quantum dynamics of BPS monopoles, Nuc. Phys. 274B (1986), 183224. P. Goddard, J. Nuyts, and D. Olive, Gauge theories and magnetic charge, Nuc. Phys. B125 (1977), 128. S. I. Goldberg, Curvature and homology, Academic Press, 1962.
[HKLR87] N. J. Hitchin, A. Karlhede, U. Lindstrm, and M. Roek, Hyo c perkhler metrics and supersymmetry, Comm. Math. Phys. 108 a (1987), 535589. [Hum72] [JT80] [JZ75] [KN63] [KN69] 1969. [Lic76] [Man82] [MO77] A. Lichnerowicz, Global theory of connections and holonomy groups, Noordho International Pub., 1976. N. S. Manton, A remark on the scattering of BPS monopoles, Phys. Lett. 110B (1982), 5456. C. Montonen and D. Olive, Magnetic monopoles as gauge particles, Phys. Lett. 72B (1977), 117120. J. E. Humphreys, Introduction to lie algebras and representation theory, vol. 9, SpringerVerlag, 1972. A. Jae and C. Taubes, Vortices and monopoles, Birkhuser, a 1980. B. Julia and A. Zee, Poles with both magnetic and electric charges in non-abelian gauge theories, Phys. Rev. D11 (1975), 22272232. S. Kobayashi and K. Nomizu, Foundations of dierential geometry, vol. 1, Interscience, 1963. , Foundations of dierential geometry, vol. 2, Interscience,
206
[email protected]
[Oli79] [Osb79]
D. Olive, The electric and magnetic charges as extra components of four-momentum, Nuc. Phys. B153 (1979), 112. H. Osborn, Topological charges for N =4 supersymmetric gauge theories and monopoles of spin 1, Phys. Lett. 83B (1979), 321 326. A. M. Polyakov, Particle spectrum in quantum eld theory, JETP Letters 20 (1974), 19495. M. K. Prasad and C. M. Sommereld, Exact classical solution for the t Hooft monopole and the JuliaZee dyon, Phys. Rev. Lett. 35 (1975), 760762. A. Pressley and G. Segal, Loop groups, Clarendon, 1986. S. Salamon, Riemannian geometry and holonomy groups, Longman Scientic and Technical, 1989. J. P. Serre, A course in arithmetic, SpringerVerlag, 1973. M. F. Sohnius, Introducing supersymmetry, Phys. Rep. 128 (1985), 39204. N. E. Steenrod, The topology of bre bundles, Princeton University Press, 1951. C. H. Taubes, Stability in YangMills theories, Comm. Math. Phys. 91 (1983), 473540. G. t Hooft, Magnetic monopoles in unied gauge theories, Nuc. Phys. B79 (1974), 276284. F. W. Warner, Foundations of dierentiable manifolds and lie groups, vol. 94, SpringerVerlag, 1983. E. J. Weinberg, Parameter counting for multimonopole solutions, Phys. Rev. D20 (1979), 936944. R. O. Wells, Jr., Dierential analysis on complex manifolds, SpringerVerlag, 1980. E. Witten, Dyons of charge e/2, Phys. Lett. 86B (1979), 283 287.
[Pol74] [PS75]
[PS86] [Sal89] [Ser73] [Soh85] [Ste51] [Tau83] [tH74] [War83] [Wei79] [Wel80] [Wit79]
207
[email protected]
[WO78] [Zum77]
E. Witten and D. Olive, Supersymmetry algebras that include topological charges, Phys. Lett. 78B (1978), 97101. B. Zumino, Euclidean supersymmetry and the many-instanton problem, Phys. Lett. 69B (1977), 369371.
208

Electromagnetic Duality For Children

Uploaded by

Copyright:

Available Formats

Electromagnetic Duality For Children

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Electromagnetic Duality For Children

Uploaded by

Copyright:

Available Formats

Electromagnetic Duality for Children

JM Figueroa-OFarrill [email protected] Version of 8 October 1998

draft version of 8/10/1998

N =2 2.3.1 2.3.2 2.3.3 2.3.4 2.3.5

N =4 2.4.1 2.4.2 2.4.3 2.4.4

draft version of 8/10/1998

Arbitrary Gauge Groups

draft version of 8/10/1998

192 194 195 195 197 198 198 200 202

draft version of 8/10/1998

draft version of 8/10/1998

draft version of 8/10/1998

Part I The Simplest Example: SO(3)

Chapter 1 Classical Electromagnetic Duality

The Dirac Monopole

And in the beginning there was Maxwell...

Maxwells equations in vacuo, given by

draft version of 8/10/1998

In terms of F , Maxwells equations (1.1) become F = 0 where F =

draft version of 8/10/1998

The Dirac quantisation condition

draft version of 8/10/1998

draft version of 8/10/1998

Dyons and the ZwanzigerSchwinger quantisation condition

draft version of 8/10/1998

where the E-eld is the one due to the charged particle.

draft version of 8/10/1998

The t HooftPolyakov Monopole

The bosonic part of the GeorgiGlashow model

draft version of 8/10/1998

we can write the energy density as

draft version of 8/10/1998

Table 1.1: The perturbative spectrum after higgsing.

Finite-energy solutions: the t HooftPolyakov Ansatz

draft version of 8/10/1998

for some arbitrary functions H and K. 21

draft version of 8/10/1998

draft version of 8/10/1998

draft version of 8/10/1998

The topological origin of the magnetic charge

draft version of 8/10/1998

A . Computing the components perpendicular to we nd that W = 1 1 + A . 2e a a

draft version of 8/10/1998

draft version of 8/10/1998

Estimating the mass of a monopole: the Bogomolnyi bound

where we have introduced the obvious shorthand Vi notice that

by the Bianchi identity (1.11) by Stokes (1.24)

draft version of 8/10/1998

Saturating the bound: the BPS-monopole

draft version of 8/10/1998

draft version of 8/10/1998

The MontonenOlive conjecture

Particle Photon Higgs W boson M monopole

draft version of 8/10/1998

The Witten eect

draft version of 8/10/1998

Using equation (1.29), and hence that = 0, we can rewrite N as N = 1 L Di . ae R3 0 Wi

Since the conjugate momentum to Wi is G0i = Ei = Ei , we nd that N= 1 q Ei Di = , ae R3 e (1.30)

draft version of 8/10/1998