IST Quantum Field Theory Notes

INSTITUTO SUPERIOR TÉCNICO
Advanced Quantum Field Theory

(Version of Thursday 7th June, 2012)
Jorge C. Romão
Physics Department
2012
2
Contents
1 Free Field Quantization 9

1.1 General formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.1 Canonical quantization for particles . . . . . . . . . . . . . . . . . . 9
1.1.2 Canonical quantization for fields . . . . . . . . . . . . . . . . . . . . 12
1.1.3 Symmetries and conservation laws . . . . . . . . . . . . . . . . . . . 15
1.2 Quantization of scalar fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.1 Real scalar field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.2.2 Microscopic causality . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.2.3 Vacuum fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.2.4 Charged scalar field . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.2.5 Time ordered product and the Feynman propagator . . . . . . . . . 26
1.3 Second quantization of the Dirac field . . . . . . . . . . . . . . . . . . . . . 27
1.3.1 Canonical formalism for the Dirac field . . . . . . . . . . . . . . . . 27
1.3.2 Microscopic causality . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.3.3 Feynman propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.4 Electromagnetic field quantization . . . . . . . . . . . . . . . . . . . . . . . 32
1.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.4.2 Undefined metric formalism . . . . . . . . . . . . . . . . . . . . . . . 34
1.4.3 Feynman propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.5 Discrete Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.5.1 Parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.5.2 Charge conjugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.5.3 Time reversal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
1.5.4 The T CP theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Problems for Chapter 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2 Physical States. S Matrix. LSZ Reduction. 51

2.1 Physical states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2 In states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.3 Spectral representation for scalar fields . . . . . . . . . . . . . . . . . . . . . 54
2.4 Out states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.5 S matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.6 Reduction formula for scalar fields . . . . . . . . . . . . . . . . . . . . . . . 58
2.7 Reduction formula for fermions . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.7.1 States in and out . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1
2 CONTENTS
2.7.2 Spectral representation fermions . . . . . . . . . . . . . . . . . . . . 62

2.7.3 Reduction formula fermions . . . . . . . . . . . . . . . . . . . . . . . 64
2.8 Reduction formula for photons . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.9 Cross sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3 Covariant Perturbation Theory 73

3.1 U matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.2 Perturbative expansion of Green functions . . . . . . . . . . . . . . . . . . . 76
3.3 Wick’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.4 Vacuum–Vacuum amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.5 Feynman rules for λϕ4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.6 Feynman rules for QED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.6.1 Compton scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3.6.2 Electron–positron elastic scattering (Bhabha scattering) . . . . . . . 93
3.6.3 Fermion Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.6.4 Feynman rules for QED . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.7 General formalism for getting the Feynman rules . . . . . . . . . . . . . . . 99
3.7.1 Example: scalar electrodynamics . . . . . . . . . . . . . . . . . . . . 100
4 Radiative Corrections 105

4.1 QED Renormalization at one-loop . . . . . . . . . . . . . . . . . . . . . . . 105
4.1.1 Vacuum Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.1.2 Self-energy of the electron . . . . . . . . . . . . . . . . . . . . . . . . 115
4.1.3 The Vertex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.2 Ward-Takahashi identities in QED . . . . . . . . . . . . . . . . . . . . . . . 124
4.2.1 Transversality of the photon propagator n = 0, p = 1 . . . . . . . . . 125
4.2.2 Identity for the vertex n = 1, p = 0 . . . . . . . . . . . . . . . . . . . 127
4.3 Counterterms and power counting . . . . . . . . . . . . . . . . . . . . . . . 129
4.4 Finite contributions from RC to physical processes . . . . . . . . . . . . . . 133
4.4.1 Anomalous electron magnetic moment . . . . . . . . . . . . . . . . . 133
4.4.2 Cancellation of IR divergences in Coulomb scattering . . . . . . . . 135
5 Functional Methods 141

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.2 Generating functional for Green’s functions . . . . . . . . . . . . . . . . . . 142
5.2.1 Green’s functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.2.2 Connected Green’s functions . . . . . . . . . . . . . . . . . . . . . . 142
5.2.3 Truncated Green’s functions . . . . . . . . . . . . . . . . . . . . . . . 144
5.2.4 Irreducible diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.3 Generating functionals for Green’s functions . . . . . . . . . . . . . . . . . . 146
5.4 Feynman rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.5 Path integral for generating functionals . . . . . . . . . . . . . . . . . . . . 153
5.5.1 Quantum mechanics of n degrees of freedom . . . . . . . . . . . . . . 153
5.5.2 Field theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
CONTENTS 3
5.5.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

5.5.4 Example: perturbation theory for λφ4 . . . . . . . . . . . . . . . . . 158
5.5.5 Symmetry factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
5.5.6 A comment on the normal ordering . . . . . . . . . . . . . . . . . . . 165
5.5.7 Generating functionals for fermions . . . . . . . . . . . . . . . . . . . 167
5.6 Change of variables in path integrals. Applications . . . . . . . . . . . . . . 168
5.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.6.2 Dyson-Schwinger equations . . . . . . . . . . . . . . . . . . . . . . . 169
5.6.3 Ward identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6 Non Abelian Gauge Theories 179

6.1 Classical theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.1.2 Covariant derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.1.3 Tensor Fµν . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.1.4 Choice of gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
6.1.5 The action and the equations of motion . . . . . . . . . . . . . . . . 182
6.1.6 Energy–momentum tensor . . . . . . . . . . . . . . . . . . . . . . . . 183
6.1.7 Hamiltonian formalism . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.2 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
6.2.1 Systems with n degrees of freedom . . . . . . . . . . . . . . . . . . . 186
6.2.2 QED as a simple example . . . . . . . . . . . . . . . . . . . . . . . . 189
6.2.3 Non abelian gauge theories. Non covariant gauges . . . . . . . . . . 190
6.2.4 Non abelian gauge theories in covariant gauges . . . . . . . . . . . . 193
6.2.5 Gauge invariance of the S matrix . . . . . . . . . . . . . . . . . . . . 198
6.2.6 Fadeev-Popov ghosts . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
6.2.7 Feynman rules in the Lorenz gauge . . . . . . . . . . . . . . . . . . . 204
6.2.8 Feynamn rules for the interaction with matter . . . . . . . . . . . . 207
6.3 Ward identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.3.1 BRS transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.3.2 Ward-Takahashi-Slavnov-Taylor identities . . . . . . . . . . . . . . . 213
6.3.3 Example: Transversality of vacuum polarization . . . . . . . . . . . 217
6.3.4 Gauge invariance of the S matrix . . . . . . . . . . . . . . . . . . . . 222
6.4 Ward-Takahashi identities in QED . . . . . . . . . . . . . . . . . . . . . . . 224
6.4.1 Ward-Takahashi identities for the funcional Z[J] . . . . . . . . . . . 224
6.4.2 Ward-Takahashi identities for the functionals W and Γ. . . . . . . . 225
6.4.3 Example: Ward identity for the QED vertex . . . . . . . . . . . . . 226
6.4.4 Ghosts in QED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.5 Unitarity and Ward identities . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.5.1 Optical theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.5.2 Cutkosky rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
6.5.3 Example of Unitariedade: scalars and fermions . . . . . . . . . . . . 234
6.5.4 Unitarity and gauge fields . . . . . . . . . . . . . . . . . . . . . . . . 237
4 CONTENTS
7 Renormalization Group 247

7.1 Callan -Symanzik equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
7.1.1 Renormalization scheme with momentum subtraction . . . . . . . . 247
7.1.2 Renormalization group . . . . . . . . . . . . . . . . . . . . . . . . . . 249
7.1.3 Callan - Symanzik equation . . . . . . . . . . . . . . . . . . . . . . . 250
7.1.4 Weinberg’s theorem and the solution of the RG equations . . . . . . 253
7.1.5 Asymptotic solution of the RG equations . . . . . . . . . . . . . . . 253
7.2 Minimal subtraction (MS) scheme . . . . . . . . . . . . . . . . . . . . . . . 255
7.2.1 Renormalization group equations for MS . . . . . . . . . . . . . . . . 255
7.2.2 Minimal subtraction scheme . . . . . . . . . . . . . . . . . . . . . . . 256
7.2.3 Physical parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
7.2.4 Renomalization group functions in minimal subtraction . . . . . . . 260
7.2.5 β and γ properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
7.2.6 Gauge independence of β and γm in MS . . . . . . . . . . . . . . . . 265
7.3 Effective gauge couplings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
7.3.1 Fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
7.3.2 β function for theories with scalars, fermions and gauge fields . . . . 270
7.3.3 The vacuum of a N AGT as a paramagnetic medium (µ > 1) . . . . 275
7.4 Renormalization group applications . . . . . . . . . . . . . . . . . . . . . . . 276
7.4.1 Scale MX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
7.4.2 Scale MZ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
A Path Integral in Quantum Mechanics 291

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
A.2 Configuration space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
A.2.1 Matrix elements of operators . . . . . . . . . . . . . . . . . . . . . . 292
A.2.2 Time ordered product of operators . . . . . . . . . . . . . . . . . . . 293
A.2.3 Exact results I: harmonic oscillator . . . . . . . . . . . . . . . . . . . 293
A.2.4 Exact results II: external force . . . . . . . . . . . . . . . . . . . . . 293
A.2.5 Perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
A.3 Phase space formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
A.4 Bargmann-Fock space (coherent states) . . . . . . . . . . . . . . . . . . . . 294
A.4.1 Normal form for an operator . . . . . . . . . . . . . . . . . . . . . . 296
A.4.2 Evolution operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
A.4.3 Exact results I: harmonic oscillator . . . . . . . . . . . . . . . . . . . 298
A.4.4 Exact results II: external force . . . . . . . . . . . . . . . . . . . . . 301
A.5 Fermion systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
A.5.1 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
A.5.2 Dot product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
A.5.3 Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
A.5.4 Representation of operators . . . . . . . . . . . . . . . . . . . . . . . 303
A.5.5 Normal form for operators . . . . . . . . . . . . . . . . . . . . . . . . 304
Problems for Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
CONTENTS 5
B Path Integral in Quantum Field Theory 307

B.1 Path integral quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
B.2 Path integral for generating functionals . . . . . . . . . . . . . . . . . . . . 310
B.3 Fermion systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
Problems Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
C Useful techniques for renormalization 321

C.1 µ parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
C.2 Feynman parametrization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
C.3 Wick Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
C.4 Scalar integrals in dimensional regularization . . . . . . . . . . . . . . . . . 325
C.5 Tensor integrals in dimensional regularization . . . . . . . . . . . . . . . . . 326
C.6 Γ function and useful relations . . . . . . . . . . . . . . . . . . . . . . . . . 327
C.7 Explicit formulæ for the 1–loop integrals . . . . . . . . . . . . . . . . . . . . 328
C.7.1 Tadpole integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
C.7.2 Self–Energy integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
C.7.3 Triangle integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
C.7.4 Box integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
C.8 Divergent part of 1–loop integrals . . . . . . . . . . . . . . . . . . . . . . . . 331
C.8.1 Tadpole integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
C.8.2 Self–Energy integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
C.8.3 Triangle integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
C.8.4 Box integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
C.9 Passarino-Veltman Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
C.9.1 The general definition . . . . . . . . . . . . . . . . . . . . . . . . . . 332
C.9.2 The divergences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
C.9.3 Useful results for PV integrals . . . . . . . . . . . . . . . . . . . . . 335
C.10 Examples of 1-loop calculations with PV functions . . . . . . . . . . . . . . 345
C.10.1 Vaccum Polarization in QED . . . . . . . . . . . . . . . . . . . . . . 345
C.10.2 Electron Self-Energy in QED . . . . . . . . . . . . . . . . . . . . . . 347
C.10.3 QED Vertex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
C.11 Modern techniques in a real problem: µ → eγ . . . . . . . . . . . . . . . . . 355
C.11.1 Neutral scalar charged fermion loop . . . . . . . . . . . . . . . . . . 355
C.11.2 Charged scalar neutral fermion loop . . . . . . . . . . . . . . . . . . 366
D Feynman Rules for the Standard Model 373

D.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
D.2 The Standard Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
D.2.1 Gauge Group SU (3)c . . . . . . . . . . . . . . . . . . . . . . . . . . 373
D.2.2 Gauge Group SU (2)L . . . . . . . . . . . . . . . . . . . . . . . . . . 374
D.2.3 Gauge Group U (1)Y . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
D.2.4 The Gauge Field Lagrangian . . . . . . . . . . . . . . . . . . . . . . 375
D.2.5 The Fermion Fields Lagrangian . . . . . . . . . . . . . . . . . . . . . 376
D.2.6 The Higgs Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . 376
D.2.7 The Yukawa Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . 377
D.2.8 The Gauge Fixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
6 CONTENTS
D.2.9 The Ghost Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . 377

D.2.10 The Complete SM Lagrangian . . . . . . . . . . . . . . . . . . . . . 378
D.3 The Feynman Rules for QCD . . . . . . . . . . . . . . . . . . . . . . . . . . 379
D.3.1 Propagators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
D.3.2 Triple Gauge Interactions . . . . . . . . . . . . . . . . . . . . . . . . 379
D.3.3 Quartic Gauge Interactions . . . . . . . . . . . . . . . . . . . . . . . 379
D.3.4 Fermion Gauge Interactions . . . . . . . . . . . . . . . . . . . . . . . 379
D.3.5 Ghost Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
D.4 The Feynman Rules for the Electroweak Theory . . . . . . . . . . . . . . . 380
D.4.1 Propagators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
D.4.2 Triple Gauge Interactions . . . . . . . . . . . . . . . . . . . . . . . . 381
D.4.3 Quartic Gauge Interactions . . . . . . . . . . . . . . . . . . . . . . . 381
D.4.4 Charged Current Interaction . . . . . . . . . . . . . . . . . . . . . . 382
D.4.5 Neutral Current Interaction . . . . . . . . . . . . . . . . . . . . . . . 382
D.4.6 Fermion-Higgs and Fermion-Goldstone Interactions . . . . . . . . . . 382
D.4.7 Triple Higgs-Gauge and Goldstone-Gauge Interactions . . . . . . . . 382
D.4.8 Quartic Higgs-Gauge and Goldstone-Gauge Interactions . . . . . . . 384
D.4.9 Triple Higgs and Goldstone Interactions . . . . . . . . . . . . . . . . 386
D.4.10 Quartic Higgs and Goldstone Interactions . . . . . . . . . . . . . . . 387
D.4.11 Ghost Propagators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
D.4.12 Ghost Gauge Interactions . . . . . . . . . . . . . . . . . . . . . . . . 388
D.4.13 Ghost Higgs and Ghost Goldstone Interactions . . . . . . . . . . . . 389
Preface
This is a text for an Advanced Quantum Field Theory course that I have been teaching
for many years at Instituto Superior Técnico. This course was first written in Portuguese.
Then, at a latter stage, I added some text in one-loop techniques in English. Then,
I realized that this text could be more useful if it was all in English. As the process of
revising the whole text shall take a long time, I decided to make available a mixed language
text. At the moment, around 60% are in English, the rest in Portuguese. My goal is to
change this gradually.
This last semester an effort was made to correct known misprints in the equations
before the full translation gets done. However, I am certain that many more still remain.
If you find errors or misprints, please send me an email.
IST, May 2012

Jorge C. Romão
jorge.romao@ist.utl.pt
7
8 CONTENTS
Chapter 1
Free Field Quantization
1.1 General formalism

1.1.1 Canonical quantization for particles
Before we study the canonical quantization of systems with an infinite number of degrees
of freedom, as it is the case with fields, we will review briefly the quantization of systems
with a finite number of degrees of freedom, like a system of particles.
Let us start with a system that consists of one particle with just one degree of freedom,
like a particle moving in one space dimension. The classical equations of motion are
obtained from the action,
Z t2
S= dtL(q, q̇) . (1.1)
t1
The condition for the minimization of the action, δS = 0, gives the Euler-Lagrange equa-
tions,
dL ∂ ∂L
− =0 (1.2)
dt ∂ q̇ ∂q
which are the equations of motion.
Before proceeding to the quantization, it is convenient to change to the Hamiltonian
formulation. We start by defining the conjugate momentum p, to the coordinate q, by
∂L
p= (1.3)
∂ q̇
Then we introduce the Hamiltonian using the Legendre transform
H(p, q) = pq̇ − L(q, q̇) (1.4)
In terms of H the equations of motion are,
∂H
{H, q}PB = = q̇ (1.5)
∂p
9
10 CHAPTER 1. FREE FIELD QUANTIZATION
∂H
{H, p}PB = − = ṗ (1.6)
∂q
where the Poisson Bracket (PB) is defined by
∂f ∂g ∂f ∂g
{f (p, q), g(p, q)}PB = − (1.7)
∂p ∂q ∂q ∂p
obviously satisfying
{p, q}PB = 1 . (1.8)

The quantization is done by promoting p and q to hermitian operators that instead of
Eq. (1.8) will satisfy the commutation relation (h̄ = 1),
[p, q] = −i (1.9)
∂
which is trivially satisfied in the coordinate representation where p = −i . The dynamics
∂q
is the given by the Schrödinger equation
∂
H(p, q) |ΨS (t)i = i |ΨS (t)i (1.10)
∂t
If we know the state of the system in t = 0, |ΨS (0)i, then Eq. (1.10) completely
determines the state |Ψs (t)i and therefore the value of any physical observable. This
description, where the states are time dependent and the operators, on the contrary, do not
depend on time, is known as the Schrödinger representation. There exits and alternative
description, where the time dependence goes to the operators and the states are time
independent. This is called the Heisenberg representation. To define this representation,
we formally integrate Eq. (1.10) to obtain
|ΨS (t)i = e−iHt |ΨS (0)i = e−iHt |ΨH i . (1.11)

The state in the Heisenberg representation, |ΨH i, is defined as the state in the Schrödinger
representation for t = 0. The unitary operator e−iHt allows us to go from one representa-
tion to the other. If we define the operators in the Heisenberg representation as,
OH (t) = eiHt OS e−iHt (1.12)

then the matrix elements are representation independent. In fact,

hΨS (t)|OS |ΨS (t)i = ΨS (0)|eiHt OS e−iHt |ΨS (0) (1.13)
= hΨH |OH (t)|ΨH i . (1.14)
The time evolution of the operator OH (t) is then given by the equation
dOH (t) ∂OH

= i[H, OH (t)] + (1.15)
dt ∂t
which can easily be obtained from Eq. (1.12). The last term in Eq. (1.15) is only present
if OS explicitly depends on time.
1.1. GENERAL FORMALISM 11
In the non-relativistic theory the difference between the two representations is very
small if we work with energy eigenfunctions. If ψn (q, t) = e−iωn t un (q) is a Schrödinger
wave function, then the Heisenberg wave function is simply un (q). For the relativistic
theory, the Heisenberg representation is more convenient, because it is easier to describe
the time evolution of operators than that of states. Also, Lorentz covariance is more
easily handled in the Heisenberg representation, because time and spatial coordinates are
together in the field operators.
In the Heisenberg representation the fundamental commutation relation is now
[p(t), q(t)] = −i (1.16)

The dynamics is now given by
dp(t) dq(t)
= i[H, p(t)] ; = i[H, q(t)] (1.17)
dt dt
Notice that in this representation the fundamental equations are similar to the classical
equations with the substitution,
{, }P B =⇒ i[, ] (1.18)
In the case of a system with n degrees of freedom Eqs. (1.16) and (1.17) are generalized
to
[pi (t), qj (t)] = −iδij (1.19)

[pi (t), pj (t)] = 0 (1.20)
[qi (t), qj (t)] = 0 (1.21)
and
ṗi (t) = i[H, pi (t)] ; q̇i (t) = i[H, qi (t)] (1.22)
Because it is an important example let us look at the harmonic oscillator. The Hamil-
tonian is
1 2
H= (p + ω02 q 2 ) (1.23)
2
The equations of motion are
ṗ = i[H, p] = −ω02 q (1.24)

q̇ = i[H, q] = p =⇒ q̈ + ω02 q = 0 . (1.25)
It is convenient to introduce the operators

1 1
a= √ (ω0 q + ip) ; a† = √ (ω0 q − ip) (1.26)
2ω0 2ω0
The equations of motion for a and a† are very simple:
ȧ(t) = −iω0 a(t) e ȧ† (t) = iω0 a† (t) . (1.27)

They have the solution
a(t) = a0 e−iω0 t ; a† (t) = a†0 eiω0 t (1.28)

and obey the commutation relations
[a, a† ] = [a0 , a†0 ] = 1 (1.29)

[a, a] = [a0 , a0 ] = 0 (1.30)
[a† , a† ] = [a†0 , a†0 ] = 0 (1.31)
In terms of a, a† the Hamiltonian reads

1 1
H = ω0 (a† a + aa†) = ω0 (a†0 a0 + a0 a†0 ) (1.32)
2 2
1
= ω0 a†0 a0 + ω0 (1.33)
2
where we have used
[H, a0 ] = −ω0 a0 , [H, a†0 ] = ω0 a†0 (1.34)
We see that a0 decreases the energy of a state by the quantity ω0 while a†0 increases
the energy by the same amount. As the Hamiltonian is a sum of squares the eigenvalues
must be positive. Then it should exist a ground state (state with the lowest energy), |0i,
defined by the condition
a0 |0i = 0 (1.35)
n
The state |ni is obtained by the application of a†0 . If we define
1 † n
|ni = √ a0 |0i (1.36)
n!
then
hm|ni = δmn (1.37)
and
1
H |ni = n+ ω0 |ni (1.38)
2
We will see that, in the quantum field theory, the equivalent of a0 and a†0 are the
creation and annihilation operators.
1.1.2 Canonical quantization for fields

Let us move now to field theory, that is, systems with an infinite number of degrees of
freedom. To specify the state of the system, we must give for all space-time points one
number (or more if we are not dealing with a scalar field). The equivalent of the coordinates
qi (t) and velocities, q̇i , are here the fields ϕ(~x, t) and their derivatives, ∂ µ ϕ(~x, t). The action
is now Z
S = d4 xL(ϕ, ∂ µ ϕ) (1.39)
where the Lagrangian density L, is a functional of the fields ϕ and their derivatives ∂ µ ϕ.
Let us consider closed systems for which L does not depend explicitly on the coordinates
xµ (energy and linear momentum are therefore conserved). For simplicity let us consider
systems described by n scalar fields ϕr (x), r = 1, 2, · · · n. The stationarity of the action,
δS = 0, implies the equations of motion, the so-called Euler-Lagrange equations,
∂L ∂L
∂µ − =0 r = 1, · · · n (1.40)
∂(∂µ ϕr ) ∂ϕr
For the case of real scalar fields with no interactions that we are considering, we can
easily see that the Lagrangian density should be,
n
X
1 1
L= ∂ ϕr ∂µ ϕr − m2 ϕr ϕr
µ
(1.41)
2 2
r=1
in order to obtain the Klein-Gordon equations as the equations of motion,
⊓ + m2 )ϕr = 0 ;
(⊔ r = 1, · · · n (1.42)
To define the canonical quantization rules we have to change to the Hamiltonian for-
malism, in particular we need to define the conjugate momentum π(x) for the field ϕ(x).
To make an analogy with systems with n degrees of freedom, we divide the 3-dimensional
space in cells with elementary volume ∆Vi . Then we introduce the coordinate ϕi (t) as the
average of ϕ(~x, t) in the volume element ∆Vi , that is,
Z
1
ϕi (t) ≡ d3 xϕ(~x, t) (1.43)
∆Vi (∆Vi )
and also Z
1
ϕ̇i (t) ≡ d3 xϕ̇(~x, t) . (1.44)
∆Vi (∆Vi )
Then Z X
L= d3 xL → ∆Vi Li . (1.45)
i
Therefore the canonical momentum is now
∂L ∂Li
pi (t) = = ∆Vi ≡ ∆Vi πi (t) (1.46)
∂ ϕ̇i (t) ∂ ϕ̇i (t)
and the Hamiltonian

X X
H= pi ϕ̇i − L = ∆Vi (πi ϕ̇i − Li ) (1.47)
i i
Going now into the limit of the continuum, we define the conjugate momentum,
∂L(ϕ, ϕ̇)
π(~x, t) ≡ (1.48)
∂ ϕ̇(~x, t)
in such a way that its average value in ∆Vi is πi (t) defined in Eq. (1.46). Eq. (1.47)
suggests the introduction of an Hamiltonian density such that
Z
H = d3 xH (1.49)
H = π ϕ̇ − L . (1.50)
To define the rules of the canonical quantization we start with the coordinates ϕi (t)
and conjugate momenta pi (t). We have
[pi (t), ϕj (t)] = −iδij
[ϕi (t), ϕj (t)] = 0
[pi (t), pj (t)] = 0 (1.51)
In terms of momentum πi (t) we have
δij
[πi (t), ϕj (t)] = −i . (1.52)
∆Vi
Going into the continuum limit, ∆Vi → 0, we obtain
[ϕ(~x, t), ϕ(~x′ , t)] = 0 (1.53)

[π(~x, t), π(~x′ , t)] = 0 (1.54)
[π(~x, t), ϕ(~x′ , t)] = −iδ(~x − ~x′ ) (1.55)
These relations are the basis of the canonical quantization. For the case of n scalar
fields, the generalization is:
[ϕr (~x, t), ϕs (~x′ , t)] = 0 (1.56)

′
[πr (~x, t), πs (~x , t)] = 0 (1.57)
[πr (~x, t), ϕs (~x′ , t)] = −iδrs δ(~x − ~x′ ) (1.58)
where
∂L
πr (~x, t) = (1.59)
∂ ϕ̇r (~x, t)
and the Hamiltonian is Z
H= d3 xH (1.60)
with
n
X
H= πr ϕ̇r − L . (1.61)
r=1
1.1.3 Symmetries and conservation laws

The Lagrangian formalism gives us a powerful method to relate symmetries and conser-
vation laws. At the classical level the fundamental result is the following theorem.
Noether’s Theorem
To each continuous symmetry transformation that leaves L and the equations of
motion invariant, corresponds one conservation law.
Proof:
Instead of making the proof for all cases, we will consider three very important par-
ticular cases:
i) Translations
Let us consider an infinitesimal translation
x′µ = xµ + εµ (1.62)
Then
∂L
δL = L′ − L = εµ (1.63)
∂xµ
and L′ leads to the same equations of motion as L, as they differ only by a 4-
divergence. If L is invariant for translations, then Eq. (1.63) tells us that it can not
depend explicitly on the coordinates xµ . Therefore
X ∂L ∂L

δL = δϕr + δ(∂µ ϕr )
r
∂ϕr ∂(∂µ ϕr )
" #
X ∂L
= ∂µ εν ∂ν ϕr (1.64)
r
∂(∂ µ ϕr )
where we have used the equations of motion, that is Eq. (1.40) and δϕr = εν ∂ν ϕr .
From Eqs. (1.63) and (1.64) and using the fact that εµ is arbitrary we get
∂µ T µν = 0 (1.65)
where T µν is the energy-momentum tensor defined by
X ∂L
T µν = −gµν L + ∂ ν ϕr (1.66)
r
∂(∂µ ϕr )
Using these relations we can define the conserved quantities

Z
P ≡ d3 xT 0µ
µ
dP µ
=0 (1.67)
dt
Noticing that T 00 = H, it is easy to realize that P µ should be the 4-momentum vector.
Therefore we conclude that invariance for translations leads to the conservation of
energy and momentum.
ii) Lorentz transformations

Consider the infinitesimal Lorentz transformations
x′µ = xµ + ω µ ν xν (1.68)
The transformation (1.68) for the coordinates will induce the following transforma-
tion for the fields,
ϕ′r (x′ ) = Srs (ω)ϕs (x) (1.69)
For the case of scalar fields Srs = δrs and for spinors we know that Srs = δrs +
1 µν
8 [γµ , γν ]rs ω . In general the variation of ϕr comes from two different effects. We
have
δϕr (x) ≡ ϕ′r (x) − ϕr (x) = Srs

−1
(ω)ϕs (x′ ) − ϕr (x)
1 h i
= − ωαβ (xα ∂ β − xβ ∂ α )δrs + Σαβ rs ϕs (1.70)
2
where we have defined
1
Srs (ω) = δrs + ωαβ Σαβ
rs . (1.71)
2
Then
∂L
δL = ∂µ δϕr (1.72)
∂(∂µ ϕr )
which gives
∂µ M µαβ = 0 (1.73)
with
∂L
M µαβ = xα T µβ − xβ T µα + Σαβ ϕs . (1.74)
∂(∂µ ϕr ) rs
The conserved angular momentum is then

Z Z " #
X
αβ 3 0αβ 3 α 0β β 0α
M = d xM = d x x T −x T + πr Σαβ
rs ϕs (1.75)
r,s
with
dM αβ
=0. (1.76)
dt
iii) Internal symmetries
Let us consider that the Lagrangian is invariant for an infinitesimal internal sym-
metry transformation
δϕr (x) = −iελrs ϕs (x) (1.77)
Then we can easily show that (see Problem 1.2)
∂µ J µ = 0 (1.78)
∂L
J µ = −i λrs ϕs (1.79)
∂(∂µ ϕr )
which leads to the conserved charge

Z
dQ
Q(λ) = −i d3 xπr λrs ϕs ; =0 (1.80)
dt
These relations between symmetries and conservation laws were derived for the classical
theory. Let us see now what happens when we quantize the theory. In the quantum theory
the fields ϕr (x) become operators acting on the Hilbert space of the states. The physical
observables are related with the matrix elements of these operators. We have therefore
to require Lorentz covariance for those matrix elements. This in turn requires that the
operators have to fulfill certain conditions.
This means that the classical fields relation
ϕ′r (x′ ) = Srs (a)ϕs (x) (1.81)
should be in the quantum theory

′
Φα |ϕr (x′ )|Φ′β = Srs (a) hΦα |ϕs (x)|Φβ i (1.82)
It should exist an unitary transformation U (a, b) that should relate the two inertial frames
′
Φ = U (a, b) |Φi (1.83)
where aµ ν e bµ are defined by

x′µ = aµ ν xν + bµ (1.84)
Using Eq. (1.83) in Eq. (1.82) we get that the field operators should transform as
U (a, b)ϕr (x)U −1 (a, b) = Srs

(−1)
(a)ϕs (ax + b) (1.85)
Let us look at the consequences of this relation for translations and Lorentz transfor-
mations. We consider first the translations. Eq. (1.85) is then
U (b)ϕr (x)U −1 (b) = ϕr (x + b) (1.86)
For infinitesimal translations we can write

µ
U (ε) ≡ eiεµ P ≃ 1 + iεµ P µ (1.87)
where P µ is an hermitian operator. Then Eq. (1.86) gives
i[P µ , ϕr (x)] = ∂ µ ϕr (x) (1.88)
The correspondence with classical mechanics and non relativistic quantum theory sug-
gests that we identify P µ with the 4-momentum, that is, P µ ≡ P µ where P µ has been
defined in Eq. (1.67).
As we have an explicit expression for P µ and we know the commutation relations of
the quantum theory, the Eq. (1.88) becomes an additional requirement that the theory
has to verify in order to be invariant under translations. We will see explicitly that this is
indeed the case for the theories in which we are interested.
For Lorentz transformations x′µ = aµ ν xν , we write for an infinitesimal transformation
aµ ν = gνµ + ω µ ν + O(ω 2 ) (1.89)
and therefore
i
U (ω) ≡ 1 − ωµν Mµν (1.90)
2
We then obtain from Eq. (1.85) the requirement
i[Mµν , ϕr (x)] = xµ ∂ ν ϕr − xν ∂ µ ϕr + Σµν

rs ϕs (x) (1.91)
Once more the classical correspondence lead us to identify Mµν = M µν where the
angular momentum M µν is defined in Eq. (1.75). For each theory we will have to verify
Eq. (1.91) for the theory to be invariant under Lorentz transformations. We will see that
this is true for the cases of interest.
1.2 Quantization of scalar fields

1.2.1 Real scalar field
The real scalar field described by the Lagrangian density
1 1
L = ∂ µ ϕ∂µ ϕ − m2 ϕϕ (1.92)
2 2
to which corresponds the Klein-Gordon equation
⊓ + m2 )ϕ = 0
(⊔ (1.93)
is the simplest example, and in fact was already used to introduce the general formalism.
As we have seen the conjugate momentum is
∂L
π= = ϕ̇ (1.94)
∂ ϕ̇
and the commutation relations are
[ϕ(~x, t), ϕ(~x′ , t)] = [π(~x, t), π(~x′ , t)] = 0
[π(~x, t), ϕ(~x′ , t)] = −iδ3 (~x − ~x′ ) (1.95)
The Hamiltonian is given by,

Z
H = P0 = d3 xH
Z
1 1 ~ 2 1 2 2
= 3
d x π 2 + |∇ϕ| + m ϕ (1.96)
2 2 2
1.2. QUANTIZATION OF SCALAR FIELDS 19
and the linear momentum is Z

P~ = − ~
d3 xπ ∇ϕ (1.97)
Using Eqs. (1.96) and (1.97) it is easy to verify that
i[P µ , ϕ] = ∂ µ ϕ (1.98)
showing the invariance of the theory for the translations. In the same way we can verify
the invariance under Lorentz transformations, Eq. (1.91), with Σµν rs = 0 (spin zero).
In order to define the states of the theory it is convenient to have eigenstates of energy
and momentum. To build these states we start by making a spectral Fourier decomposition
of ϕ(~x, t) in plane waves:
Z h i
f a(k)e−ik·x + a† (k)eik·x
ϕ(~x, t) = dk (1.99)
where q
f≡ d3 k
dk ; ωk = + |~k|2 + m2 (1.100)
(2π)3 2ωk
is the Lorentz invariant integration measure. As in the quantum theory ϕ is an operator,
also a(k) e a† (k) should be operators. As ϕ is real, then a† (k) should be the hermitian
conjugate to a(k). In order to determine their commutation relations we start be solving
Eq. (1.99) in order to a(k) and a† (k). Using the properties of the delta function, we get
Z
↔
a(k) = i d3 xeik·x ∂ 0 ϕ(x)
Z
↔
†
a (k) = −i d3 xe−ik·x ∂ 0 ϕ(x) (1.101)
where we have introduced the notation

↔ ∂b ∂a
a∂ 0 b = a − b (1.102)
∂t ∂t
The second member of Eq. (1.101) is time independent as can be checked explicitly
(see Problem 1.3). This observation is important in order to be able to choose equal times
in the commutation relations. We get
Z Z h i
↔ ′ ↔
† ′
[a(k), a (k )] = d x d3 y eik·x ∂ 0 ϕ(~x, t), e−ik ·y ∂ 0 ϕ(~y , t)
3
= (2π)3 2ωk δ3 (~k − ~k′ ) (1.103)
and
[a(k), a(k′ )] = [a† (k), a† (k′ )] = 0 (1.104)
We then see that, except for a small difference in the normalization, a(k) e a† (k) should
be interpreted as annihilation and creation operators of states with momentum kµ . To
show this, we observe that
Z h i
1 f
H = dk ωk a† (k)a(k) + a(k)a† (k) (1.105)
2
Z h i
1 f ~k a† (k)a(k) + a(k)a† (k)
P~ = dk (1.106)
2
Using these explicit forms we can then obtain
[P µ , a† (k)] = kµ a† (k) (1.107)

µ µ
[P , a(k)] = −k a(k) (1.108)
showing that a† (k) adds momentum kµ and that a(k) destroys momentum kµ . That the
quantization procedure has produced an infinity number of oscillators should come as no
surprise. In fact a(k), a† (k) correspond to the quantization of the normal modes of the
classical Klein-Gordon field.
By analogy with the harmonic oscillator, we are now in position of finding the eigen-
states of H. We start by defining the base state, that in quantum field theory is called
the vacuum. We have
a(k) |0ik = 0 ; ∀k (1.109)
Then the vacuum, that we will denote by |0i, will be formally given by
|0i = Πk |0ik (1.110)
and we will assume that is normalized, that is h0|0i = 1. If now we calculate the vacuum
energy, we find immediately the first problem with infinities in Quantum Field Theory
(QFT). In fact
Z D h i E
1 f
h0|H|0i = dk ωk 0| a† (k)a(k) + a(k)a† (k) |0
2
Z D h i E
1 f
= dk ωk 0| a(k), a† (k) |0
2
Z
1 d3 k
= ωk (2π)3 2ωk δ3 (0)
2 (2π)3 2ωk
Z
1
= d3 k ωk δ3 (0) = ∞ (1.111)
2
This infinity can be understood as the the (infinite) sum
Pof1the zero point energy of all
quantum oscillators. In the discrete case we would have, k 2 ωk = ∞. This infinity can
be easily removed. We start by noticing that we only measure energies as differences with
respect to the vacuum energy, and those will be finite. We will then define the energy of
the vacuum as been zero. Technically this is done as follows. We define a new operator
µ
PN.O. as
Z
µ 1 f µh † i
PN.O. ≡ dk k a (k)a(k) + a(k)a† (k)
2
Z
1 f µD h † i E
− dk k 0| a (k)a(k) + a(k)a† (k) |0
2
Z
= f k µ a† (k)a(k)
dk (1.112)

µ
Now 0|PN.O. |0 = 0. The ordering of operators where the annihilation operators appear
on the right of the creation operators is called normal ordering and the usual notation is
1
: (a† (k)a(k) + a(k)a† (k)) :≡ a† (k)a(k) (1.113)
2
Therefore to remove the infinity of the energy and momentum corresponds to choose the
normal ordering to our operators. We will adopt this convention in the following dropping
the subscript ”N.O.” to simplify the notation. This should not appear as an ad hoc
procedure. In fact, in going from the classical theory where we have products of fields
into the quantum theory where the fields are operators, we should have a prescription for
the correct ordering of such products. We have just seen that this should be the normal
ordering.
Once we have the vacuum we can build the states by applying the the creation operators
†
a (k). As in the case of the harmonic oscillator, we can define the number operator,
Z
N = dk f a† (k)a(k) (1.114)
It is easy to see that N commutes with H and therefore the eigenstates of H are also
eigenstates of N . The state with one particle of momentum kµ is obtained as a† (k) |0i. In
fact we have
Z
P µ a† (k) |0i = f ′ k′µ a† (k′ )a(k′ )a† (k) |0i
dk
Z
= d3 k′ k′µ δ3 (~k − ~k′ )a† (k) |0i
= kµ a† (k) |0i (1.115)
and
N a† (k) |0i = a† (k) |0i (1.116)
a† (k1 )...a† (kn ) |0i
In a similar way, the state would be a state with n particles. However,
the sates that we have just defined have a problem. They are not normalizable and
therefore they can not form a basis for the Hilbert space of the quantum field theory, the
so-called Fock space. The origin of the problem is related to the use of plane waves and
states with exact momentum. This can be solved forming states that are superpositions
of plane waves Z
f
|1i = λ dkC(k)a †
(k) |0i (1.117)
Then
Z D E
h1|1i = λ 2 f 1 dk
dk f 2 C ∗ (k1 )C(k2 ) 0|a(k1 )a† (k2 )|0
Z
= λ2 f
dk|C(k)|2
=1 (1.118)
and therefore Z −1/2

λ= f |C(k)|2
dk (1.119)
R
f |C(k)|2 < ∞. If k is only different from zero in a neighborhood
with the condition that dk
of a given 4-momentum kµ , then the state will have a well defined momentum (within some
experimental error).
A basis for the Fock space can then be constructed from the n–particle normalized
states
Z −1/2
|ni = n! dkf 1 · · · dk
f n |C(k1 , · · · kn )|2
Z
f 1 · · · dk
dk f n C(k1 , · · · kn )a† (k1 ) · · · a† (kn ) |0i (1.120)
that satisfy
hn|ni = 1 (1.121)
N |ni = n |ni (1.122)
Due to the commutation relations of the operators a† (k) in Eq. (1.120), the functions
C(k1 · · · kn ) are symmetric, that is,
C(· · · ki , · · · kj , · · · ) = C(· · · kj · · · ki · · · ) (1.123)
This shows that the quanta that appear in the canonical quantization of real scalar fields
obey the Bose–Einstein statistics. This interpretation in terms of particles, with creation
and annihilation operators, that results from the canonical quantization, is usually called
second quantization, as opposed to the description in terms of wave functions (the first
quantization.
1.2.2 Microscopic causality

Classically, the fields can be measured with an arbitrary precision. In a relativistic quan-
tum theory we have several problems. The first, results from the fact that the fields are
now operators. This means that the observables should be connected with the matrix
elements of the operators and not with the operators. Besides this question, we can only
speak of measuring ϕ in two space-time points x and y if [ϕ(x), ϕ(y)] vanishes. Let us
look at the conditions needed for this to occur.
Z nh i h i o
[ϕ(x), ϕ(y)] = f 1 dk
dk f 2 a(k1 ), a† (k2 ) e−ik1 ·x+ik2·y + a† (k1 ), a(k2 ) eik1 x−ik2 ·y
Z
= f 1 e−ik1 ·(x−y) − eik1 ·(x−y)
dk
≡ i∆(x − y) (1.124)
The function ∆(x − y) is Lorentz invariant and satisfies the relations
⊓x + m2 )∆(x − y) = 0
(⊔ (1.125)
∆(x − y) = −∆(y − x) (1.126)
∆(~x − ~y , 0) = 0 (1.127)
The last relation ensures that the equal time commutator of two fields vanishes.
Lorentz invariance implies then,
∆(x − y) = 0 ; ∀ (x − y)2 < 0 (1.128)
This means that for two points that can not be physically connected, that is for which
(x − y)2 < 0, the fields interpreted as physical observables, can then be independently
measured. This result is known as Microscopic Causality. We note that
∂ 0 ∆(x − y)|x0 =y0 = −δ3 (~x − ~y ) (1.129)
which ensures the canonical commutation relation, Eq. (1.95).
1.2.3 Vacuum fluctuations

It is well known from Quantum Mechanics that, in an harmonic oscillator, the coordinate
is not well defined for the energy eigenstates, that is

n|q 2 |n > (hn|q|ni)2 = 0 (1.130)
In Quantum Field Theory, we deal with an infinite set of oscillators, and therefore we will
have the same behavior, that is,
h0|ϕ(x)ϕ(y)|0i 6= 0 (1.131)
although
h0|ϕ(x)|0i = 0 (1.132)
We can calculate Eq. (1.131). We have
Z D E
h0|ϕ(x)ϕ(y)|0i = f 1 dk
dk f 2 e−ik1 ·x eik2 ·y 0|a(k1 )a† (k2 )|0
Z
= f 1 e−ik·(x−y) ≡ ∆+ (x − y)
dk (1.133)
The function ∆+ (x − y) corresponds to the positive frequency part of ∆(x − y). When
y → x this expression diverges quadratically,
Z Z

2

f1 = d3 k1
0|ϕ (x)|0 = ∆+ (0) = dk (1.134)
(2π)3 2ωk1
This divergence can not be eliminated in the way we did with the energy of the vacuum.
In fact these vacuum fluctuations, as they are known, do have observable consequences
like, for instance, the Lamb shift. We will be less worried with the result of Eq. (1.134),
if we notice that for measuring the square of the operator ϕ at x we need frequencies
arbitrarily large, that is, an infinite amount of energy. Physically only averages over a
finite space-time region have meaning.
1.2.4 Charged scalar field

The description in terms of real fields does not allow the distinction between particles
and anti-particles. It applies only the those cases were the particle and anti-particle
are identical, like the π 0 . For the more usual case where particles and anti-particles are
distinct, it is necessary to have some charge (electric or other) that allows us to distinguish
them. For this we need complex fields.
The theory for the scalar complex field can be easily obtained from two real scalar
fields ϕ1 and ϕ2 with the same mass. If we denote the complex field ϕ by,
ϕ1 + iϕ2
ϕ= √ (1.135)
2
then
L = L(ϕ1 ) + L(ϕ2 ) =: ∂ µ ϕ† ∂µ ϕ − m2 ϕ† ϕ : (1.136)
which leads to the equations of motion
⊓ + m2 )ϕ = 0 ; (⊔
(⊔ ⊓ + m2 )ϕ† = 0 (1.137)
The classical theory given in Eq. (1.136) has, at the classical level, a conserved current,
∂µ J µ = 0, with
↔µ
J µ = iϕ† ∂ ϕ (1.138)
Therefore we expect, at the quantum level, the charge Q
Z
Q = d3 x : i(ϕ† ϕ̇ − ϕ̇† ϕ) : (1.139)
to be conserved, that is, [H, Q] = 0. To show this we need to know the commutation
relations for the field ϕ. The definition Eq. (1.135), and the commutation relations for ϕ1
and ϕ2 allow us to obtain the following relations for ϕ and ϕ† :
[ϕ(x), ϕ(y)] = [ϕ† (x), ϕ† (y)] = 0 (1.140)

†
[ϕ(x), ϕ (y)] = i∆(x − y) (1.141)
For equal times we can get from Eq. (1.141)
[π(~x, t), ϕ(~y , t)] = [π † (~x, t), ϕ† (~y , t)] = −iδ3 (~x − ~y ) (1.142)
where
π = ϕ̇† ; π † = ϕ̇ (1.143)
The plane waves expansion is then
Z h i
ϕ(x) = dk f a+ (k)e−ik·x + a† (k)eik·x
−
Z h i
ϕ† (x) = f
dk a− (k)e−ik·x + a†+ (k)eik·x (1.144)
where the definition of a± (k) is
a1 (k) ± ia2 (k) † a†1 (k) ∓ ia†2 (k)

a± (k) = √ ; a± = √ (1.145)
2 2
The algebra of the operators a± it is easily obtained from the algebra of the operators
ai′ s. We get the following non-vanishing commutators:
[a+ (k), a†+ (k′ )] = [a− (k), a†− (k′ )] = (2π)3 2ωk δ3 (~k − ~k′ ) (1.146)
therefore allowing us to interpret a+ and a†+ as annihilation and creation operators of

quanta of type +, and similarly for the quanta of type −. We can construct the number
operators for those quanta: Z
N± = dk f a† (k)a± (k) (1.147)
±
One can easily verify that

N+ + N− = N1 + N2 (1.148)
where Z
Ni = f a† (k)ai (k)
dk (1.149)
i
The energy-momentum operator can be written in terms of the + and − operators,

Z h i
µ
P = dk f k µ a† (k)a+ (k) + a† (k)a− (k) (1.150)
+ −
where we have already considered the normal ordering. Using the decomposition in
Eq. (1.144), we obtain for the charge Q:
Z
Q = d3 x : i(ϕ† ϕ̇ − ϕ̇† ϕ̇) :
Z h i
= f
dk a†+ (k)a+ (k) − a†− (k)a− (k)
= N+ − N− (1.151)
Using the commutation relation in (Eq. (1.146)) one can easily verify that
[H, Q] = 0 (1.152)
showing that the charge Q is conserved. The Eq. (1.151) allows us to interpret the ± quanta
as having charge ±1. However, before introducing interactions, the theory is symmetric,
and we can not distinguish between the two types of quanta. From the commutation
relations (1.146) we obtain,
[P µ , a†+ (k)] = kµ a†+ (k)

[Q, a†+ (k)] = +a†+ (k) (1.153)
showing that a†+ (k) creates a quanta with 4-momentum kµ and charge +1. In a similar
way we can show that a†− creates a quanta with charge −1 and that a± (k) annihilate
quanta of charge ±1, respectively.
1.2.5 Time ordered product and the Feynman propagator

The operator ϕ† creates a particle with charge +1 or annihilates a particle with charge −1.
In both cases it adds a total charge +1. In a similar way ϕ annihilates one unit of charge.
Let us construct a state of one particle (not normalized) with charge +1 by application of
ϕ† in the vacuum:
|Ψ+ (~x, t)i ≡ ϕ† (~x, t) |0i (1.154)
The amplitude to propagate the state |Ψ+ i into the future to the point (~x′ , t′ ) with t′ > t
is given by

D E
θ(t′ − t) Ψ+ (~x′ , t′ )|Ψ+ (~x, t) = θ(t′ − t) 0|ϕ(~x′ , t′ )ϕ† (~x, t)|0 (1.155)
In ϕ† (~x, t) |0i only the operator a†+ (k) is active, while in h0| ϕ(~x′ , t′ ) the same happens to
a+ (k). Therefore Eq. (1.155) is the matrix element that creates a quanta of charge +1 in
(~x, t) and annihilates it in (~x′ , t′ ) with t′ > t.
There exists another way of increasing the charge by +1 unit in (~x, t) and decreasing
it by −1 in (~x′ , t′ ). This is achieved if we create a quanta of charge −1 in ~x′ at time t′ and
let it propagate to ~x where it is absorbed at time t > t′ . The amplitude is then,

D E
θ(t − t′ ) Ψ− (~x, t)|Ψ− (~x′ , t′ ) = 0|ϕ† (~x, t)ϕ(~x′ , t′ )|0 θ(t − t′ ) (1.156)
The sum of the two amplitudes in Eqs. (1.155) and (1.156) is the so-called Feynman
propagator. It can be written in a more compact way if we introduce the time ordered
product. Given two operators a(x) and b(x′ ) we define the time ordered product T by,
T a(x)b(x′ ) = θ(t − t′ )a(x)b(x′ ) + θ(t′ − t)b(x′ )a(x) (1.157)
In this prescription the older times are always to the right of the more recent times. It
can be applied to an arbitrary number of operators. With this definition, the Feynman
propagator reads, D E
∆F (x′ − x) = 0|T ϕ(x′ )ϕ† (x)|0 (1.158)
Using the ϕ and ϕ† decomposition we can calculate ∆F (for free fields, of course)
Z h i
′
∆F (x − x) = f θ(t′ − t)e−ik·(x′ −x) + θ(t − t′ )eik·(x′ −x)
dk (1.159)
Z
d4 k i ′
= e−ik·(x −x) (1.160)
(2π)4 k2 − m2 + iε
Z
d4 k ′
≡ ∆F (k)e−ik·(x −x)
(2π)4
where
i
∆F (k) ≡ (1.161)
k2 − m2 + iε
∆F (k) is the propagator in momenta space (Fourier transform). The equivalence be-
tween Eq. (1.160) and Eq. (1.159) is done using integration in the complex plane of the
1.3. SECOND QUANTIZATION OF THE DIRAC FIELD 27
time component k0 , with the help of the residue theorem. The contour is defined by the
⊓′x + m2 ) to ∆F (x′ − x)
iε prescription, as indicated in Fig. (1.1). Applying the operator (⊔
in any of the forms of Eq. (1.159) one can show that
⊓′x + m2 )∆F (x′ − x) = −iδ4 (x′ − x)

(⊔ (1.162)
that is, ∆F (x′ − x) is the Green’s function for the Klein-Gordon equation with Feynman
boundary conditions.
In the presence of interactions, Feynman propagator looses the simple form of Eq. (1.161).
However, as we will see, the free propagator plays a key role in perturbation theory.
1.3 Second quantization of the Dirac field

Let us now apply the formalism of second quantization to the Dirac field. As we will
see, something has to be changed, otherwise we would be led to a theory obeying Bose
statistics, while we know that electrons have spin 1/2 and obey Fermi statistics.
1.3.1 Canonical formalism for the Dirac field

The Lagrangian density that leads to the Dirac equation is
L = iψγ µ ∂µ ψ − mψψ (1.163)
The conjugate momentum to ψα is
∂L
πα = = iψα† (1.164)
∂ ψ̇α
while the conjugate momentum to ψα† vanishes. The Hamiltonian density is then
H = π ψ̇ − L = ψ † (−i~ ~ + βm)ψ
α·∇ (1.165)
The requirement of translational and Lorentz invariance for L leads to the tensors T µν
and M µνλ . Using the obvious generalizations of Eqs. (1.66) and (1.74) we get
T µν = iψγ µ ∂ ν ψ − gµν L (1.166)
Im ko
x
x
Re ko
Figure 1.1: Integration in the complex k0 plane.

and
M µνλ = iψγ µ (xν ∂ λ − xλ ∂ ν + σ νλ )ψ − xν gµλ − xµ gνλ L (1.167)
where
1
σ νλ = [γ ν , γ λ ] (1.168)
4
The 4-momentum P µ and the angular momentum tensor M νλ are then given by,
Z
Pµ ≡ d3 xT 0µ
Z
µλ
M ≡ d3 xM 0νλ (1.169)
or
Z
H ≡ d3 xψ † (−i~ ~ + βm)ψ
α·∇
Z
P~ ≡ ~
d3 xψ † (−i∇)ψ (1.170)
If we define the angular momentum vector J~ ≡ (M 23 , M 31 , M 12 ) we get

Z
~ 3 † 1~ 1
J = d xψ ~r × ∇ + ~σ ψ (1.171)
i 2
which has the familiar aspect J~ = L ~ + S.~ We can also identify a conserved current,
∂µ j µ = 0, with j µ = ψγ µ ψ, which will give the conserved charge
Z
Q = d3 xψ † ψ (1.172)
All that we have done so far is at the classical level. To apply the canonical formalism
we have to enforce commutation relations and verify the Lorentz invariance of the theory.
This will lead us into problems. To see what are the problems and how to solve them, we
will introduce the plane wave expansions,
Z Xh i
ψ(x) = dp f b(p, s)u(p, s)e−ip·x + d† (p, s)v(p, s)eip·x (1.173)
s
Z Xh i
ψ † (x) = f
dp b† (p, s)u† (p, s)e+ip·x + d(p, s)v † (p, s)e−ip·x (1.174)
s
where u(p, s) and v(p, s) are the spinors for positive and negative energy, respectively,
introduced in the study of the Dirac equation and b, b† , d and d† are operators. To see
what are the problems with the canonical quantization of fermions, let us calculate P µ .
We get Z Xh i
P µ = dk f kµ b† (k, s)b(k, s) − d(k, s)d† (k, s) (1.175)
s
where we have used the orthogonality and closure relations for the spinors u(p, s) and
v(p, s). From Eq. (1.175) we realize that if we define the vacuum as b(k, s) |0i = d(k, s) |0i =
0 and if we quantize with commutators then particles b and particles d will contribute with
opposite signs to the energy and the theory will not have a stable ground state. In fact,
this was the problem already encountered in the study of the negative energy solutions of
the Dirac equation, and this is the reason for the negative sign in Eq. (1.175). Dirac’s hole
theory required Fermi statistics for the electrons and we will see how spin and statistics
are related.
To discover what are the relations that b, b† , d and d† should obey, we recall that at
the quantum level it is always necessary to verify Lorentz invariance. This gives,
i[Pµ , ψ(x)] = ∂µ ψ ; i[Pµ , ψ(x)] = ∂µ ψ (1.176)
Instead of imposing canonical quantization commutators and, as a consequence, verifying
Eq. (1.176) we will do the other way around. We start with Eq. (1.176) and we will
discover the appropriate relations for the operators. Using the expansions Eqs. (1.173)
and (1.174) we can show that Eq. (1.176) leads to
[Pµ , b(k, s)] = −kµ b(k, s) ; [Pµ , b† (k, s)] = kµ b† (k, s) (1.177)
[Pµ , d(k, s)] = −kµ d(k, s) ; [Pµ , d† (k, s)] = kµ d† (k, s) (1.178)
Using Eq. (1.175) for Pµ we get
X h i
b† (p, s′ )b(p, s′ ) − d(p, s′ )d† (p, s′ ) , b(k, s) = −(2π)3 2k0 δ3 (~k − p~)b(k, s) (1.179)
s′
and three other similar relations. If we assume that

[d† (p, s′ )d(p, s′ ), b(k, s)] = 0 (1.180)
the condition from Eq. (1.179) reads
Xh i
b† (p, s′ ){b(p, s′ ), b(k, s)} − {b† (p, s′ ), b(k, s)}b(p, s′ ) =
s′
p − ~k)b(k, s)
= −(2π)3 2k0 δ3 (~ (1.181)
where the parenthesis {, } denote anti-commutators. It is easy to see that Eq. (1.181) is
verified if we impose the canonical commutation relations. We should have
p − ~k)δss′
{b† (p, s), b(k, s)} = (2π)3 2k0 δ3 (~
p − ~k)δss′
{d† (p, s′ ), d(k, s)} = (2π)3 2k0 δ3 (~ (1.182)
and all the other anti-commutators vanish. Note that as b anti-commutes with d and d† ,
then it commutes with d† d and therefore Eq. (1.180) is verified.
With the anti-commutator relations both contributions to P µ in Eq. (1.175) are posi-
tive. As in boson case we have to subtract the zero point energy. This is done, as usual,
by taking all quantities normal ordered. Therefore we have for P µ ,
Z X
Pµ = f kµ
dk : b† (k, s)b(k, s) − d(k, s)d† (k, s) :
s
Z X
= f kµ
dk : b† (k, s)b(k, s) + d† (k, s)d(k, s) : (1.183)
s
and for the charge

Z
Q = d3 x : ψ † (x)ψ(x) :
Z Xh i
= f
dk b† (k, s)b(k, s) − d† (k, s)d(k, s) (1.184)
s
which means that the quanta of b type have charge +1 while those of d type have charge
−1. It is interesting to note that was the second quantization of the Dirac field that
introduced the − sign in Eq. (1.184), making the charge operator without a definite sign,
while in Dirac theory was the probability density that was positive defined. The reverse
is true for bosons. We can easily show that
[Q, b† (k, s)] = b† (k, s) [Q, d(k, s)] = d(k, s)

[Q, b(k, s)] = −b(k, s) [Q, d† (k, s)] = −d† (k, s)
and then
[Q, ψ] = −ψ ; [Q, ψ] = ψ (1.185)
In QED the charge is given by eQ (e < 0). Therefore we see that ψ creates positrons and
annihilates electrons and the opposite happens with ψ.
We can introduce the number operators
N + (p, s) = b† (p, s)b(p, s) ; N − (p, s) = d† (p, s)d(p, s) (1.186)
and we can rewrite

Z X
P µ
= f kµ
dk (N + (k, s) + N − (k, s)) (1.187)
s
Z X
Q = f
dk (N + (k, s) − N − (k, s)) (1.188)
s
Using the anti-commutator relations in Eq. (1.182) it is now easy to verify that the theory
is Lorentz invariant, that is (see Problem 1.4),
i[M µν , ψ] = (xµ ∂ ν − xν ∂ µ )ψ + Σµν ψ . (1.189)
1.3.2 Microscopic causality

The anti-commutation relations in Eq. (1.182) can be used to find the anti-commutation
relations at equal times for the fields. We get
{ψα (~x, t), ψβ† (~y , t)} = δ3 (~x − ~y)δαβ (1.190)
and
{ψα (~x, t), ψβ (~y , t)} = {ψα† (~x, t), ψβ† (~y , t)} = 0 (1.191)
These relations can be generalized to unequal times

Z h i

{ψα (x), ψβ† (y)} = f
dp (p/ + m)γ 0 αβ
e−ip·(x−y) − (−p/ + m)γ 0 αβ eip·(x−y)

= (i∂/x + m)γ 0 αβ
i∆(x − y) (1.192)
where the ∆(x − y) function was defined in Eq. (1.124) for the scalar field. The fact that
γ 0 appears in Eq. (1.192) is due to the fact that in Eq. (1.192) we took ψ † and not ψ. In
fact, if we multiply on the right by γ 0 we get
{ψα (x), ψ β (y)} = (i∂/x + m)αβ i∆(x − y) (1.193)
and
{ψα (x), ψβ (y)} = {ψ α (x), ψ β (y)} = 0 (1.194)
We can easily verify the covariance of Eq. (1.193). We use
U (a, b)ψ(x)U −1 (a, b) = S −1 (a)ψ(ax + b)
U (a, b)ψ(x)U −1 (a, b) = ψ(ax + b)S(a)
S −1 γ µ S = aµ ν γ ν (1.195)
to get
U (a, b){ψα (x), ψ β (y)}U −1 (a, b) =

−1
= Sατ (a){ψτ (ax + b), ψ λ (ay + b)}Sλβ (a)
−1
= Sατ (a)(i∂/ax + m)τ λ i∆(ax − ay)Sλβ (a)
= (i∂/ + m)αβ i∆(x − y) (1.196)
where we have used the invariance of ∆(x − y) and the result S −1 i∂/ax S = i∂/x . For
(x − y)2 < 0 the anti-commutators vanish, because ∆(x − y) also vanishes. This result
allows us to show that any two observables built as bilinear products of ψ e ψ commute
for two spacetime points for which (x − y)2 < 0. Therefore

ψ α (x)ψβ (x), ψ λ (y)ψτ (y) =
= ψ α (x){ψβ (x), ψ λ (y)}ψτ (y) − {ψ α (x), ψ λ (y)}ψβ (x)ψτ (y)
+ψ λ (y)ψ α (x){ψβ (x), ψτ (y)} − ψ λ (y){ψτ (y), ψ α (x)}ψβ (x)
= 0 (1.197)
for (x − y)2 < 0. In this way the microscopic causality is satisfied for the physical observ-
ables, such as the charge density or the momentum density.
1.3.3 Feynman propagator

For the Dirac field, as in the case of the charged scalar field, there are two ways of increasing
the charge by one unit in x′ and decrease it by one unit in x (note that the electron has
negative charge). These ways are
D E
θ(t′ − t) 0|ψβ (x′ )ψα† (x)|0 (1.198)
D E
θ(t − t′ ) 0|ψα† (x)ψβ (x′ )|0 (1.199)
In Eq. (1.198) an electron of positive energy is created at ~x in the instant t, propagates

until ~x′ where is annihilated at time t′ > t. In Eq. (1.199) a positron of positive energy
is created in x′ and annihilated at x with t > t′ . The Feynman propagator is obtained
summing the two amplitudes. Due the exchange of ψβ and ψ α there must be a minus sign
between these two amplitudes. Multiplying by γ 0 , in order to get ψ instead of ψ † , we get
for the Feynman propagator,

SF (x′ − x)αβ = θ(t′ − t) 0|ψα (x′ )ψ β (x)|0

−θ(t − t′ ) 0|ψ β (x)ψα (x′ )|0

≡ 0|T ψα (x′ )ψ β (x)|0 (1.200)
where we have defined the time ordered product for fermion fields,
T η(x)χ(y) ≡ θ(x0 − y 0 )η(x)χ(y) − θ(y 0 − x0 )χ(y)η(x) . (1.201)
Inserting in Eq. (1.200) the expansions for ψ and ψ we get,

Z h i
′
SF (x − x)αβ = dk f (p/ + m)αβ θ(t′ − t)e−ik·(x′ −x) + (−p/ + m)αβ θ(t − t′ )eik·(x′ −x)
Z
d4 k i(k/ + m)αβ −ik·(x′ −x)
= e
(2π)4 k2 − m2 + iε
Z
d4 k ′
≡ 4
SF (k)αβ e−ik·(x −x) (1.202)
(2π)
where SF (k) is the Feynman propagator in momenta space. We can also verify that
Feynman’s propagator is the Green function for the Dirac equation, that is (see Problem
1.7),
(i∂/ − m)λα SF (x′ − x)αβ = iδλβ δ4 (x′ − x) (1.203)
1.4 Electromagnetic field quantization

1.4.1 Introduction
The free electromagnetic field is described by the classical Lagrangian,
1
L = − Fµν F µν (1.204)
4
1.4. ELECTROMAGNETIC FIELD QUANTIZATION 33
where
Fµν = ∂µ Aν − ∂ν Aµ (1.205)
The free field Maxwell equations are
∂α F αβ = 0 (1.206)
that corresponds to the usual equations in 3-vector notation,

~
~ ·E
∇ ~ =0 ; ~ = ∂E
~ ×B
∇ (1.207)
∂t
The other Maxwell equations are a consequence of Eq. (1.205) and can written as,
1
∂α Feαβ = 0 ; Feαβ = εαβµν Fµν (1.208)
2
corresponding to
~
~ ·B
∇ ~ =0 ; ~ = − ∂B
~ ×E
∇ (1.209)
∂t
~ e B,
Classically, the quantities with physical significance are the fields E ~ and the
µ
potentials A are auxiliary quantities that are not unique due to the gauge invariance of
the theory. In the quantum theory are the potentials Aµ that play the leading role as, for
instance in the minimal prescription. We have therefore to formulate the quantum fields
theory in terms of Aµ and not of E ~ and B.
~
When we try to apply the canonical quantization to the potentials Aµ we immediately
run into difficulties. For instance, if we define the conjugate momentum as,
∂L
πµ = (1.210)
∂(Ȧµ)
we get
∂L ∂A0
πk = = −Ȧk − k
= Ek
∂(Ȧk ) ∂x
∂L
π0 = =0 (1.211)
∂ Ȧ0
Therefore the conjugate momentum to the coordinate A0 vanishes and does not allow
us to use directly the canonical formalism. The problem has its origin in the fact that the
photon, that we want to describe, has only two degrees of freedom (positive or negative
helicity) but we are using a field Aµ with four degrees of freedom. In fact, we have to
impose constraints on Aµ in such a way that it describes the photon. This problem can
be addressed in three different ways:
i) Radiation Gauge
Historically, this was the first method to be used. It is based in the fact that it is
always possible to choose a gauge, called the radiation gauge, where
A0 = 0 ; ~ ·A
∇ ~=0 (1.212)
~ is transverse. The conditions in Eq. (1.212) reduce the num-

that is, the potential A
ber of degrees of freedom to two, the transverse components of A. ~ It is then possible
to apply the canonical formalism to these transverse components and quantize the
electromagnetic field in this way. The problem with this method is that we loose
explicit Lorentz covariance. It is then necessary to show that this is recovered in the
final result. This method is followed in many text books, for instance in Bjorken
and Drell [1].
ii) Quantization of systems with constraints

It can be shown that the electromagnetism is an example of an Hamilton generalized
system, that is a system where there are constraints among the variables. The way
to quantize these systems was developed by Dirac for systems of particles with n
degrees of freedom. The generalization to quantum field theories is done using the
formalism of path integrals. We will study this method in Chapter 6, as it will be
shown, this is the only method that can be applied to non-abelian gauge theories,
like the Standard Model.
iii) Undefined metric formalism

There is another method that works for the electromagnetism, called the formalism
of the undefined metric, developed by Gupta and Bleuler [2, 3]. In this formalism,
that we will study below, Lorentz covariance is kept, that is we will always work
with the 4-vector Aµ , but the price to pay is the appearance of states with negative
norm. We have then to define the Hilbert space of the physical states as a sub-space
where the norm is positive. We see that in all cases, in order to maintain the explicit
Lorentz covariance, we have to complicate the formalism. We will follow the book
of Silvan Schweber [4].
1.4.2 Undefined metric formalism

To solve the difficulty of the vanishing of π 0 , we will start by modifying the Maxwell
Lagrangian introducing a new term,
1 1
L = − Fµν F µν − (∂ · A)2 (1.213)
4 2ξ
where ξ is a dimensionless parameter. The equations of motion are now,

µ 1
⊓A − 1 −
⊔ ∂ µ (∂ · A) = 0 (1.214)
ξ
and the conjugate momenta
∂L 1
πµ = = F µ0 − gµ0 (∂ · A) (1.215)
∂ Ȧµ ξ
that is (
π 0 = − 1ξ (∂ · A)
πk = E k
We remark that the Lagrangian of Eq. (1.212) and the equations of motion, Eq. (1.214),
reduce to Maxwell theory in the gauge ∂ · A = 0. This why we say that the choice of
Eq. (1.212) corresponds to a class of Lorenz gauges with parameter ξ. With this abuse of
language (in fact we are not setting ∂ · A = 0, otherwise the problems would come back)
the value of ξ = 1 is known as the Feynman gauge and ξ = 0 as the Landau gauge.
From Eq. (1.214) we get
⊓
⊔(∂ · A) = 0 (1.216)
implying that (∂ · A) is a massless scalar field. Although it would be possible to continue
with a general ξ, from now on we will take the case of the so-called Feynman gauge, where
ξ = 1. Then the equation of motion coincides with the Maxwell theory in the Lorenz
gauge. As we do not have anymore π 0 = 0, we can impose the canonical commutation
relations at equal times:
[π µ (~x, t), Aν (~y , t)] = −ig µ ν δ3 (~x − ~y)
[Aµ (~x, t), Aν (~y , t)] = [πµ (~x, t), πν (~y , t)] = 0 (1.217)
Knowing that [Aµ (~x, t), Aµ (~y , t)] = 0 at equal times, we can conclude that the space
derivatives of Aµ also commute at equal times. Then, noticing that
π µ = −Ȧµ + space derivatives (1.218)
we can write instead of Eq. (1.217)
[Aµ (~x, t), Aν (~y , t)] = [Ȧµ (~x, t), Ȧµ (~y , t)] = 0
[Ȧµ (~x, t), Aν (~y , t)] = igµν δ3 (~x − ~y) (1.219)
If we compare these relations with the corresponding ones for the real scalar field, where
the only one non-vanishing is,
[ϕ̇(~x, t), ϕ(~y , t)] = −iδ3 (~x − ~y) (1.220)
we see (gµν = diag(+, −, −, −) that the relations for space components are equal but they
differ for the time component. This sign will be the source of the difficulties previously
mentioned.
If, for the moment, we do not worry about this sign, we expand Aµ (x) in plane waves,
Z 3 h
X i
µ
A (x) = f
dk a(k, λ)εµ (k, λ)e−ik·x + a† (k, λ)εµ∗ (k, λ)eik·x (1.221)
λ=0
where εµ (k, λ) are a set of four independent 4-vectors that we assume to real, without loss
of generality. We will now make a choice for these 4-vectors. We choose εµ (1) and εµ (2)
orthogonal to kµ and nµ , such that
εµ (k, λ)εµ (k, λ′ ) = −δλλ′ for λ, λ′ = 1, 2 (1.222)
After, we choose εµ (k, 3) in the plane (kµ , nµ ) orthogonal to nµ and normalized, that is
εµ (k, 3)nµ = 0 ; εµ (k, 3)εµ (k, 3) = −1 (1.223)

Finally we choose εµ (k, 0) = nµ . The vectors εµ (k, 1) and εµ (k, 2) are called transverse
polarizations, while εµ (k, 3) and εµ (k, 0) longitudinal and scalar polarizations, respec-
tively. We can give an example. In the frame where nµ = (1, 0, 0, 0) and ~k is along the z
axis we have
εµ (k, 0) ≡ (1, 0, 0, 0) ; εµ (k, 1) ≡ (0, 1, 0, 0)
εµ (k, 2) ≡ (0, 0, 1, 0) ; εµ (k, 3) ≡ (0, 0, 0, 1) (1.224)
In general we can show that

′
ε(k, λ) · ε∗ (k, λ′ ) = g λλ
X
g λλ εµ (k, λ)ε∗ν (k, λ) = gµν (1.225)
λ
Inserting the expansion (1.221) in (1.219) we get

′
[a(k, λ), a† (k′ , λ′ )] = −g λλ 2k0 (2π)3 δ3 (~k − ~k′ ) (1.226)
showing, once more, that the quanta associated with λ = 0 has a commutation relation
with the wrong sign. Before addressing this problem, we can verify that the generalization
of Eq. (1.219) for arbitrary times is
[Aµ (x), Aν (y)] = −igµν ∆(x, y) (1.227)
showing the covariance of the theory. The function ∆(x − y) is the same that was intro-
duced before for scalar fields.
Therefore, up to this point, everything is as if we had 4 scalar fields. There is, however,
the problem of the sign difference in one of the commutators. Let us now see what are
the consequences of this sign. For that we introduce the vacuum state defined by
a(k, λ) |0i = 0 λ = 0, 1, 2, 3 (1.228)
To see the problem with the sign we construct the one-particle state with scalar polariza-
tion, that is Z
|1i = dkf f (k)a† (k, 0) |0i (1.229)
and calculate its norm

Z D E
h1|1i = f 1 dk
dk f 2 f ∗ (k1 )f (k2 ) 0|a(k1 , 0)a† (k2 , 0)|0
Z
= − h0|0i f |f (k)|2
dk (1.230)
where we have used Eq. (1.226) for λ = 0. The state |1i has a negative norm. The
same calculation for the other polarization would give well behaved positive norms. We
therefore conclude that the Fock space of the theory has indefinite metric. What happens
then to the probabilistic interpretation of quantum mechanics?
To solve this problem we note that we are not working anymore with the classical
Maxwell theory because we modified the Lagrangian. What we would like to do is to
impose the condition ∂ · A = 0, but that is impossible as an equation for operators, as
that would bring us back to the initial problems with π 0 = 0. We can, however, require
that condition on a weaker form, as a condition only to be verified by the physical states.
More specifically, we require that the part of ∂ · A that contains the annihilation operator
(positive frequencies) annihilates the physical states,
∂ µ A(+)
µ |ψi = 0 (1.231)
The states |ψi can be written in the form
|ψi = |ψT i |φi (1.232)
where |ψT i is obtained from the vacuum with creation operators with transverse polar-
ization and |φi with scalar and longitudinal polarization. This decomposition depends,
of course, on the choice of polarization vectors. To understand the consequences of
(+)
Eq. (1.231) is enough to analyze the states |φi as ∂ µ Aµ contains only scalar and longi-
tudinal polarizations,
Z X
i∂ · A(+) f e−ik·x
= dk a(k, λ) ε(k, λ) · k (1.233)
λ=0,3
and therefore Eq. (1.231) becomes

X
k · ε(k, λ) a(k, λ) |φi = 0 (1.234)
λ=0,3
Condition (1.234) does not determine completely |φi. In fact, there much arbitrariness
in the choice of the transverse polarization vectors, to which we can always add a term
proportional to kµ because k · k = 0. This arbitrariness must reflect itself on the choice of
|φi. Condition (1.234) is equivalent to,
[a(k, 0) − a(k, 3)] |φi = 0 . (1.235)
We can construct |φi as a linear combination of states |φn i with n scalar or longitudinal
photons:
|φi = C0 |φ0 i + C1 |φ1 i + · · · + Cn |φn i + · · ·
|φ0 i ≡ |0i (1.236)
The states |φn i are eigenstates of the operator number for scalar or longitudinal pho-
tons,
N ′ |φn i = n |φn i (1.237)
where Z h i
N′ = f
dk a† (k, 3)a(k, 3) − a† (k, 0)a(k, 0) (1.238)
Then

n hφn |φn i = φn |N ′ |φn = 0 (1.239)
where we have used Eq. (1.235). This means that
hφn |φn i = δn0 (1.240)
that is, for n 6= 0, the state |φn i has zero norm. We have then for the general state |φi,
hφ|φi = |C0 |2 ≥ 0 (1.241)
and the coefficients Ci , i = 1, · · · n · · · are arbitrary. We have to show that this arbitrariness
does not affect the physical observables. The Hamiltonian is
Z
H = d3 x : π µ Ȧµ − L :
Z 3 h
X i
1 3 ~ 0 )2 :
~ i )2 − Ȧ20 − (∇A
= d x: Ȧ2i + (∇A
2
i=1
Z " 3
#
X
= fk
dk 0 † †
a (k, λ)a(k, λ) − a (k, 0)a(k, 0) (1.242)
λ=1
It is easy to check that if |ψi is a physical state we have

D R P2 E
ψ | f
dk k 0 a† (k, λ)a(k, λ)|ψ
hψ|H|ψi T λ=1 T
= (1.243)
hψ|ψi hψT |ψT i
and the arbitrariness on the physical states completely disappears when we take average
values. Besides that, only the physical transverse polarizations contribute to the result.
One can show (see Problem 1.10) that the arbitrariness in |φi is related with a gauge
transformation within the class of Lorenz gauges.
It is important to note that although for the average values of the physical observables
only the transverse polarizations contribute, the scalar and longitudinal polarizations are
necessary for the consistency of the theory. In particular they show up when we consider
complete sums over the intermediate states.
Invariance for translations is readily verified. For that we write,
Z 3
X
P = µ f kµ
dk (−gλλ )a† (k, λ)a(k, λ) (1.244)
λ=0
Then
Z X nh i
i[P µ , Aν ] = dk f ′ ikµ
f dk (−gλλ ) a† (k, λ)a(k, λ), a(k′ , λ′ ) εν (k′ , λ′ )e−ik·x
λ,λ′
h i ′
o
+ a† (x, λ)a(k, λ), a† (k′ , λ′ ) ε∗ν (k′ , λ′ )eik ·x
Z Xh i
= f ikµ
dk a(k, λ)εν (k, λ)e−ik·x − a† (k, λ)εν (k, λ)eik·x
λ
= ∂ µ Aν (1.245)
showing the invariance under translations. In a similar way, it can be shown the invariance
for Lorentz transformations (see Problem 1.11). For that we have to show that
Z h i
jk
M = d3 x : xj T 0k − xk T 0j + E j Ak − E k Aj : (1.246)
Z
0i

M = d3 x : x0 T 0i − xi T 00 − (∂ · A)Ai − E i A0 : (1.247)
where (ξ = 1)
T 0i = −(∂ · A) ∂ i A0 − E k ∂ i Ak
3 h
X i
T 00 = Ȧ2i + (∇A ~ 0 )2
~ j )2 − Ȧ20 − (∇A (1.248)
i=1
Using these expressions one can show that the photon has helicity ±1, corresponding
therefore to spin one. For that we start by choosing the direction of ~k along the axis 3
(z axis) and take the polarization vector with the choice of Eq. (1.224). A one-photon
physical state will then be (not normalized),
|k, λi = a† (k, λ) |0i λ = 1, 2 (1.249)
Let us now calculate the angular momentum along the axis 3. This is given by
M 12 |k, λi = M 12 a† (k, λ) |0i
= [M 12 , a† (k, λ)] |0i (1.250)
where we have used the fact that the vacuum state satisfies M 12 |0i = 0. The operator M 12
has one part corresponding the orbital angular momenta and another corresponding to the
spin. The contribution of the orbital angular momenta vanishes (angular momenta in the
direction of motion) as one can see calculating the commutator. In fact the commutator
with the orbital angular momenta is proportional to k1 or k2 , which are zero by hypothesis.
Let us then calculate the spin part. Using the notation,
Aµ = Aµ(+) + Aµ(−) (1.251)
where Aµ(+) (Aµ(−) ) correspond to the positive (negative) frequencies, we get
: E 1 A2 − E 2 A1 := E 1(+) A2(+) + E 1(−) A2(+) + A2(−) E 1(+) + E 1(−) A2(−) − (1 ↔ 2) (1.252)
Then
h i
: E 1 A2 − E 2 A1 :, a† (k, λ) =
h i h i
= E 1(+) A2 (+), a† (k, λ) + E 1(+) , a† (k, λ) A2(+)
h i h i
+E 1 (−) A2 (+), a† (k, λ) + A2(−) E 1(+) , a† (k, λ) − (1 ↔ 2)
h i h i
= E 1 A2(+) , a† (k, λ) + A2 E 1(+) , a† (k, λ) − (1 ↔ 2) (1.253)
Now (recall that λ = 1, 2)

Z X h i
[A2(+) †
, a (k, λ)] = dkf′ ′
ε2 (k′ , λ′ ) a(k′ , λ′ ), a† (k, λ) e−ik ·x
λ′
= ε2 (k, λ)e−ik·x
Z X h i
[E 1(+) †
, a (k, λ)] = f′
dk
′
ik ′0 ε0 (k′ , λ′ ) + ik′1 ε0 (k′ , λ′ ) a(k′ , λ′ ), a† (k, λ) e−ik ·x
λ′
= ik0 ε1 (k, λ)e−ik·x (1.254)

Therefore
Z h i
d3 x : E 1 A2 − E 2 A1 :, a† (k, λ)
Z

= d3 xe−ik·x E 1 ε2 (k, λ) + A2 ik0 ε1 (k, λ) − E 2 ε1 (k, λ) + A1 ik 0 ε2 (k, λ)
Z h i
↔ ↔
= d3 xe−ik·x ε1 (k, λ)∂ 0 A2 (x) − ε2 (k, λ)∂ 0 A1 (x) (1.255)
where we have used the fact that E i = −Ȧi , i = 1, 2, for our choice of frame and
polarization vectors. On the other hand
Z
↔
a(k, λ) = −i d3 xeik·x ∂ 0 εµ (k, λ)Aµ (x)
Z
↔
†
a (k, λ) = i d3 xe−ik·x ∂ 0 εµ (k, λ)Aµ (x) (1.256)
For our choice we get

Z
↔
†
a (k, 1) = −i d3 xe−ik·x ∂ 0 A1 (x)
Z
↔
†
a (k, 2) = −i d3 xe−ik·x ∂ 0 A2 (x) (1.257)
and therefore
[M 12 , a† (k, λ)] = iε1 (k, λ)a† (k, 2) − iε2 (k, λ)a† (k, 1) (1.258)
We find that the state a† (k, λ) |0i , λ = 1, 2 is not an eigenstate of the operator M 12 .
However the linear combinations,
1 h i
a†R (k) = √ a† (k, 1) + ia† (k, 2)
2
1 h i
a†L (k) = √ a† (k, 1) − ia† (k, 2) (1.259)
2
which correspond to right and left circular polarization, verify
[M 12 , a†R (k)] = a†R (k) ; [M 12 , a†L (k)] = −a†L (k) (1.260)
showing that the photon has spin 1 with right or left circular polarization (negative or
positive helicity).
1.4.3 Feynman propagator

The Feynman propagator is defined as the vacuum expectation value of the time ordered
product of the fields, that is
Gµν (x, y) ≡ h0|T Aµ (x)Aν (y)|0i
= θ(x0 − y 0 ) h0|Aµ (x)Aν (y)|0i + θ(y 0 − x0 ) h0|Aν (y)Aµ (x)|0i (1.261)

Inserting the expansions for Aµ (x) and Aν (y) we get
Z h i
Gµν (x − y) = −gµν dk f e−ik·(x−y) θ(x0 − y 0 ) + eik·(x−y)θ(y0 −x0 )
Z
d4 k i
= −gµν 4 2
e−ik·(x−y)
(2π) k + iε
Z
d4 k
≡ Gµν (k)e−ik·(x−y) (1.262)
(2π)4
where Gµν (k) is the Feynman propagator on the momentum space
−igµν
Gµν (k) ≡ (1.263)
k2 + iε
It is easy to verify that Gµν (x − y) is the Green’s function of the equation of motion, that
for ξ = 1 is the wave equation, that is
⊓x Gµν (x − y) = igµν δ4 (x − y)
⊔ (1.264)
These expressions for Gµν (x−y) and Gµν (k) correspond to the particular case of ξ = 1,
the so-called Feynman gauge. For the general case where ξ 6= 0 the equation of motion
reads
µ 1
⊓x gρ − 1 −
⊔ µ
∂ ∂ρ Aρ (x) = 0 (1.265)
ξ
For this case the equal times commutation relations are more complicated (see Problem
1.12). Using those relations one can show that the Feynman propagator is still the Green’s
function of the equation of motion, that is

µ 1
⊓x gρ − 1 −
⊔ µ
∂ ∂ρ h0|T Aρ (x)Aν (y)|0i = ig µν δ4 (x − y) (1.266)
ξ
Using this equation we can then obtain in an arbitrary ξ gauge (of the Lorenz type),
gµν kµ kν
Gµν (k) = −i + i(1 − ξ) 2 . (1.267)
k2 + iε (k + iε)2
1.5 Discrete Symmetries

We know from the study of the Dirac equation the transformations like space inversion
(Parity) and charge conjugation, are symmetries of the Dirac equation. More precisely, if
ψ(x) is a solution of the Dirac equation, then
ψ ′ (x) = ψ ′ (−~x, t) = γ0 ψ(~x, t) (1.268)

T
ψ c (x) = Cψ (x) (1.269)
are also solutions (if we take the charge −e for ψ c ). Similar operations could also be
defined for scalar and vector fields.
With second quantization the fields are no longer functions, they become operators. We
have therefore to find unitary operators P and C that describe those operations within this
formalism. There is another discrete symmetry, time reversal, that in second quantization
will be described by an anti-unitary operator T . We will exemplify with the scalar field
how to get these operators. We will leave the Dirac and Maxwell fields as exercises.
1.5.1 Parity
To define the meaning of the Parity operation we have to put the system in interaction
with the measuring system, considered to be classical. This means that we will consider
the system described by
L −→ L − jµ (x)Aµext (x) (1.270)
where we have considered that the interaction is electromagnetic. jµ (x) is the electromag-
netic current that has the form,
↔
jµ (x) = ie : ϕ∗ ∂ µ ϕ : scalar field
jµ (x) = e : ψγµ ψ : Dirac field (1.271)
In a Parity transformation we invert the coordinates of the measuring system, therefore

the classical fields are now
Aµext = (A0ext (−~x, t)), −A

~ ext (−~x, t) = Aext
µ (−~
x, t) (1.272)
For the dynamics of the new system to be identical to the that of the original system,
which should be the case if Parity is conserved, it is necessary that the equations of
motion remain the same. This is true if
PL(~x, t)P −1 = L(−~x, t) (1.273)
Pjµ (~x, t)P −1 = j µ (−~x, t) (1.274)
Eqs. (1.273) and (1.274) are the conditions that a theory should obey in order to be
invariant under Parity. Furthermore P should leave the commutation relations unchanged,
so that the quantum dynamics is preserved. For each theory that conserves Parity should
be possible to find an unitary operator P that satisfies these conditions.
1.5. DISCRETE SYMMETRIES 43
Now we will find such an operator P for the scalar field. It is easy to verify that the
condition
Pϕ(~x, t)P −1 = ±ϕ(−~x, t) (1.275)
satisfies all the requirements. The sign ± is the intrinsic parity of the particle described
by the field ϕ, (+ for scalar and − for pseudo-scalar). In terms of the expansion of the
momentum, Eq. (1.275) requires
Pa(k)P −1 = ±a(−k) ; Pa† (k)P −1 = ±a† (−k) (1.276)

~ ~ 0 0
q −k means that we have changed k into −k (but k remains intact, that is, k =
where
+ |~k|2 + m2 ). It is easier to solve Eq. (1.276) in the momentum space. As P should be
unitary, we write
P = eiP (1.277)
Then
in
Pa(k)P −1 = a(k) + i[P, a(k)] + · · · + [P, [· · · , [P, a(k)] · · · ] + · · ·
n!
= −a(−k) (1.278)
where we have chosen the case of the pseudo-scalar field.

Eq. (1.278) suggests the form
λ
[P, a(k)] = [a(k) + εa(−k)] (1.279)
2
where λ and ε = ±1 are to be determined. We get
λ2
[P, [P, a(k)]] = [a(k) + εa(−k)] (1.280)
2
and therefore

−1 1 (iλ)2 (iλ)4
Pa(k)P = a(k) + iλ + + ··· + + · · · (a(k) + εa(−k))
2 2! n!
1 1
= [a(k) − εa(−k)] + eiλ [a(k) + εa(−k)]
2 2
= −a(−k) (1.281)
We solve Eq. (1.281) if we choose λ = π and ε = +1 (λ = π and ε = −1 for the scalar

case). It is easy to check that
Z h i
π f a† (k)a(k) + a† (k)a(−k) = P †
Pps = − dk ps (1.282)
2
and it is solution of Eq. (1.279) for λ = π and ε = +1. Therefore,

Z h i
π f † †
Pps = exp −i dk a (k)a(k) + a (k)a(−k) (1.283)
2
and for the scalar field

Z h i
π f † †
Ps = exp −i dk a (k)a(k) − a (k)a(−k) (1.284)
2
For the case of the Dirac field, the condition equivalent to Eq. (1.275) is now
Pψ(~x, t)P −1 = γ 0 ψ(−~x, t) (1.285)
Repeating the same steps we get

( Z
π f Xh †
PDirac = exp −i dp b (p, s)b(p, s) − b† (p, s)b(−p, s)
2 s

† †
+ d (p, s)d(p, s) + d (p, s)d(−p, s) (1.286)
The case of the Maxwell field is left as an exercise.
1.5.2 Charge conjugation

The conditions for charge conjugation invariance are now
CL(x)C −1 = L ; Cjµ C −1 = −jµ (1.287)
where jµ is the electromagnetic current. Conditions (1.287) are verified for the charged
scalar fields if
Cϕ(x)C −1 = ϕ∗ (x) ; Cϕ∗ (x)C −1 = ϕ(x) (1.288)
and for the Dirac field if
Cψα (x)C −1 = Cαβ ψ β (x)

−1
Cψ α (x)C −1 = −ψβ (x)Cβα (1.289)
where C is the charge conjugation matrix.

Finally from the invariance of jµ Aµ we obtain the condition for the electromagnetic
field,
CAµ C −1 = −Aµ (1.290)
By using a method similar to the one used in the case of the Parity we can get the
operator C for the different theories. For instance, for the scalar field we get
Z
π f † †
Cs = exp i dk (a+ − a− )(a+ − a− ) (1.291)
2
and for the Dirac field
C = C1 C2 (1.292)
with
( Z )
X h i
C1 = exp −i f
dp † †
φ(p, s) b (p, s)b(p, s) − d (p, s)d(p, s)
s
1.5. DISCRETE SYMMETRIES 45
( Z )
π Xh i
C2 = exp i f
dp b† (p, s) − d† (p, s) [b(p, s) − d(p, s)] (1.293)
2 s
with
v(p, s) = eiφ(p,s) uc (p, s)
u(p, s) = eiφ(p,s) v c (p, s) (1.294)
where the phase φ(p, s) is arbitrary (see [5]).
1.5.3 Time reversal

Classically the meaning of the time reversal invariance it is clear. We change the sign of
the time, the velocities change direction and the system goes from what was the final state
to the initial state. This exchange between the initial and final state has as consequence, in
quantum mechanics, that the corresponding operator must be anti-linear or anti-unitary.
In fact hf |ii = hi|f i∗ and therefore if we want hT ϕf |T ϕi i = hϕi |ϕf i then T must include
the complex conjugation operation. We can write
T = UK (1.295)
where U is unitary and K is the instruction to tale the complex conjugate of all c-numbers.
Then
hT ϕf |T ϕi i = hU Kϕf |U Kϕi i
= hU ϕf |U ϕi i∗
= hϕf |ϕi i∗ = hϕi |ϕf i (1.296)
as we wanted. A theory will be invariant under time reversal if
T L(~x, t)T −1 = L(~x, −t)
T jµ (~x, t)T −1 = j µ (~x, −t) (1.297)
For the scalar field this condition will be verified if
T ϕ(~x, t)T −1 = ±ϕ(~x, −t) (1.298)
and for the electromagnetic field we must have.
T Aµ (~x, t)T −1 = Aµ (~x, −t) (1.299)
making j µ Aµ invariant. For the case of the Dirac field the transformation is
T ψα (~x, t)T −1 = Tαβ ψβ (~x, −t) (1.300)
In order that Eq. (1.297) is satisfied, the T matrix must satisfy
T γµ T −1 = γµT = γ µ∗ (1.301)
with a solution, in the Dirac representation,
T = iγ 1 γ 3 (1.302)
Applying the same type of reasoning already used for P and C we can find T , or
equivalently, U . For the Dirac field, noticing that
T u(p, s) = u∗ (−p, −s)eiα+ (p,s)
T v(p, s) = v ∗ (−p, −s)eiα−(p,s) (1.303)
we can write U = U1 U2 and obtain

( Z )
Xh i
U1 = exp −i dp f α+ b† (p, s)b(p, s) − α− d† (p, s)d(p, s) (1.304)
s
and
( Z
π Xh
U2 = exp −i f
dp b† (p, s)b(p, s) + b† (p, s)b(−p − s)
2 s
#)
† †
− d (p, s)d(p, s) − d (p, s)d(−p, −s) (1.305)
1.5.4 The T CP theorem

It is a fundamental theorem in Quantum Field Theory that the product T CP is an invari-
ance of any theory that satisfies the following general conditions:
• The theory is local and covariant for Lorentz transformations.
• The theory is quantized using the usual relation between spin and statistics, that is,
commutators for bosons and anti-commutators for fermions.
This theorem due to Lüdus, Zumino, Pauli e Schwinger has an important consequence
that if one of the discrete symmetries is not preserve then another one must also be violated
to preserve the invariance of the product. For a proof of the theorem see the books of
Bjorken and Drell[1, 6] and Itzykson and Zuber[7].
Problems 47
Problems for Chapter 1
1.1 Verify, for the scalar field, the covariant relations for translations and Lorentz trans-
formations,
i[P µ , ϕ] = ∂ µ ϕ
i[M µν , ϕ] = (xµ ∂ ν − xν ∂ µ )ϕ (1.306)
1.2 Show that

∂ 0 ∆(x − y)|x0 =y0 = −δ3 (~x − ~y ) (1.307)
1.3 Show that

Z
d4 k i
e−ik·(x−y) =
(2π)4 k2 − m2 + iε
Z h i
= f θ(x0 − y 0 )e−ik·(x−y) + θ(y 0 − x0 )eik·(x−y)
dk (1.308)
f ≡ d3 k
where dk . Hint: Integrate in the complex plane of the variable dk0 and use
(2π)3 2k0
the prescription iε to define the contours.
1.4 Show that for the Dirac theory the requirements of Lorentz invariance are satisfied,
1
i[M µν , ϕ] = (xµ ∂ ν − xν ∂ µ )ϕ + Σµν ϕ ; Σµν = [γ µ , γ ν ] (1.309)
4
1.5 Show that

{ψα (~x, t), ψβ† (~y , t)} = δαβ δ3 (~x − ~y ) (1.310)
1.6 Show that

SF (x − y)αβ = θ(x0 − y 0 ) 0|ψα (x)ψ β (y)|0

−θ(y 0 − x0 ) 0|ψ β (y)ψα (x)|0 (1.311)
corresponds to
Z
d4 p i(p/ + m)αβ −ip·(x−y)
SF (x − y)αβ = e (1.312)
(2π)4 p2 − m2 + iε
Hint: Expand ψα e ψ β in plane waves.
1.7 Show that
(i∂/x − m)αβ SF (x − y)βγ = iδαγ δ4 (x − y) (1.313)
1.8 Show that is is always possible to choose the electromagnetic potential Aµ such that
~ ·A
A0 = 0 , ∇ ~=0 (Radiation gauge) (1.314)
1.9 Show that we have

[Aµ (x), Aν (y)] = −igµν ∆(x − y) (1.315)
1.10 Consider the indefinite metric formalism for the electromagnetic field.
a) Consider the expectation value of Aµ in the state |φi. Show that
Z
hφ|Aµ |φi = C0∗ C1 f e−ik·x h0| [εµ (k, 3)a(k, 3) + εµ (k, 0)a(k, 0)] |φ1 i
dk
+h.c. (1.316)
b) Choose the state |φ1 i in the form

Z h i
|φ1 i = f f (k) a† (k, 3) − a† (k, 0) |0i
dk (1.317)
Show that
Z
hφ|Aµ |φi = f [εµ (k, 3) + εµ (k, 0)] (C ∗ C1 e−ik·x f (k) + c.c.)
dk (1.318)
0
c) Choose εµ (k, λ) to be real. Show that
kµ
εµ (k, 3) + εµ (k, 0) = (1.319)
(n · k)
d) Show that
hφ|Aµ |φi = ∂µ Λ(x) (1.320)
where
⊓Λ = 0
⊔ (1.321)
Comment the result.

Problems 49
1.11 Show the covariance of the electromagnetism for the Lorentz transformations,
i[M µν , Aλ ] = (xµ ∂ ν − xν ∂ µ )Aλ + Σµν,λ σ Aσ (1.322)

where
Σµν,λσ = gµλ gνσ − g µσ gλν (1.323)
1.12 Show that for the general case of ξ 6= 1 we have

[Aµ (~x, t), Aν (~y , t)] = 0
[Ȧµ (~x, t), Aν (~y , t)] = igµν [1 − (1 − ξ)gµ0 ] δ3 (~x − ~y )
[Ȧi (~x, t), Ȧj (~y , t)] = [Ȧ0 (~x, t), Ȧ0 (~y , t)] = 0
[Ȧ0 (~x, t), Ȧi (~y , t)] = i(1 − ξ)∂i δ3 (~x − ~y) (1.324)
1.13 Use the results of Problem 1.12 to show that, in the general gauge with ξ 6= 1 we
have
µ 1
⊓x g ρ − 1 −
⊔ µ
∂ ∂ρ h0|T Aρ (x)Aν (y)|0i = ig µν δ4 (x − y) (1.325)
ξ
where
µ 1
⊓g ρ − 1 −
⊔ µ
∂ ∂ρ Aρ = 0 (1.326)
ξ
1.14 Find the operator P for the Dirac and Maxwell fields.
1.15 Find the operator C for the Dirac and Maxwell fields.
1.16 Show that
T ψα (~x, t)T −1 = Tαβ ψβ (~x, −t) (1.327)
ensures that
T L(~x, t)T −1 = L(~x, −t) (1.328)
if there is a matrix T such that T γµ T −1 = γ µ∗ . Find T in the Dirac representation.
1.17 Find the operator T for the Dirac and Maxwell fields.
1.18 Consider the Lagrangian
L = ψiγ µ Dµ PL ψ − mψψ (1.329)
where
τa
Dµ = ∂µ + iAaµ
2
1 − γ5
PL = (1.330)
2
Show that the theory is neither invariant under P nor under C but it is invariant for the
product CP.
Chapter 2
Physical States. S Matrix. LSZ

Reduction.
2.1 Physical states

In the previous chapter we saw, for the case of free fields, how to construct the space of
states, the so-called Fock space of the theory. When we consider the real physical case,
with interactions, we are no longer able to solve the problem exactly. For instance, the
interaction between electrons and photons is given by a set of nonlinear coupled equations,
(i∂/ − m)ψ = eA
/ψ
∂µ F µν = eψγ ν ψ (2.1)
that do not have an exact solution. In practice we have to resort to approximation meth-
ods. In the following chapter we will learn how to develop a covariant perturbation theory.
Here we are going just to study the general properties of the theory.
Let us start by the physical states. As we do not know how to solve the problem exactly,
we can not prove the assumptions we are going to make about these states. However, these
are reasonable assumptions, based essentially on Lorentz covariance. We choose our states
to be eigenstates of energy and momentum, and of all the other observables that commute
with P µ . Besides that, we will also assume that
i) The eigenvalues of p2 are non-negative and p0 > 0.
ii) There exists one non-degenerate base state, with the minimum of energy, which is
Lorentz invariant. This state is called the vacuum state |0i and by convention
pµ |0i = 0 (2.2)

iii) There exist one particle states p(i) , such that,
p(i)
µ p
(i)µ
= m2i (2.3)
for each stable particle with mass mi .
iv) The vacuum and the one-particle states constitute the discrete spectrum of pν .
51
52 CHAPTER 2. PHYSICAL STATES. S MATRIX. LSZ REDUCTION.
2.2 In states
As we are mainly interested in scattering problems, we should construct states that have
a simple interpretation in the limit t → −∞. At that time, the particles that are going
to participate in the scattering process have not interacted yet (we assume that the in-
teractions are adiabatically switched off when |t| → ∞ which is appropriate for scattering
problems).
We look for operators that create one particle states with the physical mass. To be
explicit, we start by an hermitian scalar field given by the Lagrangian
1 1
L = ∂ µ ϕ∂µ ϕ − m2 ϕ2 + j(x)ϕ(x) (2.4)
2 2
where the current j(x) is an operator made of the interacting fields ϕ at point x. For
instance, those interactions can be self-interactions of the type
λ 4
j(x)ϕ(x) = ϕ (x) (2.5)
4!
that is
λ 3
j(x) = ϕ (x) (2.6)
4!
The field ϕ satisfies the following equation of motion
⊓ + m2 )ϕ(x) = j(x)
(⊔ (2.7)
and the equal time canonical commutation relations,
[ϕ(~x, t)ϕ(~y , t)] = [π(~x, t)π(~y , t)] = 0
[π(~x, t), ϕ(~y , t)] = −iδ3 (~x − ~y) (2.8)
where
π(x) = ϕ̇(x) (2.9)
if we assume that j(x) has no derivatives. We designate by ϕin (x) the operator that creates
one-particle states. It will be a functional of the fields ϕ(x) and other fields present on
j(x). Its existence will be shown by explicit construction. We require that ϕin (x) must
satisfy the conditions:
i) ϕin (x) and ϕ(x) transform in the same way for translations and Lorentz transforma-
tions. For translations we have then
i [P µ , ϕin (x)] = ∂ µ ϕin (x) (2.10)
ii) The spacetime evolution of ϕin (x) corresponds to that of a particle of mass m, that is
⊓ + m2 )ϕin (x) = 0
(⊔ (2.11)
2.2. IN STATES 53
From these definitions it follows that ϕin (x) creates one-particle states from the vac-
uum. In fact, let us consider a state |ni, such that,
P µ = pµn |ni . (2.12)
Then
∂ µ hn|ϕin (x)|0i = i hn| [P µ , ϕin (x)] |0i
= ipµn hn|ϕin (x)|0i (2.13)
and therefore
⊓ hn|ϕin (x)|0i = −p2n hn|ϕin (x)|0i
⊔ (2.14)
Then
⊓ + m2 ) hn|ϕin (x)|0i = (m2 − p2n ) hn|ϕin (x)|0i = 0
(⊔ (2.15)
where we have used the fact that ϕin (x) is a free field, Eq. (2.11). Therefore the states
created from the vacuum by ϕin are those for which p2n = m2 , that is, the one-particle
states of mass m.
The Fourier decomposition of ϕin (x) is then the same as for free fields, that is,
Z h i
ϕin (x) = dkf ain (k)e−ik·x + a† (k)eik·x (2.16)
in
where ain (k) and a†in (k) satisfy the usual algebra for creation and annihilation operators.
In particular, by repeated use of a†in (k) we can create one state of n particles.
To express ϕin (x) in terms of ϕ(x) and j(x) we start by introducing the retarded
Green’s function of the Klein-Gordon operator,
⊓x + m2 )∆ret (x − y; m) = δ4 (x − y)
(⊔ (2.17)
where
∆ret (x − y; m) = 0 se x0 < y 0 (2.18)
We can then write
√ Z
Zϕin (x) = ϕ(x) − d4 y∆ret(x − y; m)j(y) (2.19)
The field ϕin (x), defined by Eq. (2.16), satisfies the two initial conditions. The constant
√
Z was introduced to normalize ϕin in such a way that it has amplitude 1 to create
one-particle
√ states from the vacuum. The fact that ∆ret = 0 for x0 → −∞, suggests that
Zϕin (x) is, in some way, the limit of ϕ(x) when x0 → −∞. In fact, as ϕ and ϕin are
operators, the correct asymptotic condition must be set on the matrix elements of the
operators. Let |αi and |βi be two normalized states. We define the operators
Z
↔
ϕ (t) = i d3 xf ∗ (x) ∂ 0 ϕ(x)
f
Z
↔
ϕfin = i d3 xf ∗ (x) ∂ 0 ϕin (x) (2.20)
where f (x) is a normalized solution of the Klein-Gordon equation. By Green’s theorem,

ϕfin does not depend on time (for plane waves f = e−ik·x and ϕfin = ain ). Then the
asymptotic condition of Lehmann, Symanzik e Zimmermann (LSZ) [8], is
√
lim hα| ϕf (t) |βi = Z hα| ϕfin |βi (2.21)
t→−∞
2.3 Spectral representation for scalar fields

We saw that Z had a physical meaning as the square of the amplitude for the field ϕ(x)
to create one-particle states from the vacuum. Let us now find a formal expression for Z
and show that 0 ≤ Z ≤ 1.
We start by calculating the expectation value in the vacuum of the commutator of two
fields,
i∆′ (x, y) ≡ h0| [ϕ(x), ϕ(y)] |0i (2.22)
As we do not know how to solve the equations for the interacting fields ϕ, we can not solve
exactly the problem of finding the ∆′ , in contrast with the free field case. We can, however,
determine its form using general arguments of Lorentz invariance and the assumed spectra
for the physical states. We introduce a complete set of states between the two operators
in Eq. (2.22) and we use the invariance under translations in order to obtain,
hn|ϕ(y)|mi = hn| eiP ·y ϕ(0)e−iP ·y |mi
= ei(pn −pm )·y hn|ϕ(0)|mi (2.23)
Therefore we get
X
∆′ (x, y) = −i h0|ϕ(0)|ni hn|ϕ(0)|0i (e−ipn ·(x−y) − eipn ·(x−y) )
n
≡ ∆′ (x − y) (2.24)
that is, like in the free field case, ∆′ is only a function of the difference x − y. Introducing
now Z
1 = d4 q δ4 (q − pn ) (2.25)
we get
Z " #
d4 q X
∆′ (x − y) = −i (2π)3 δ4 (pn − q)| h0|ϕ(0)|ni |2 (e−iq·(x−y) − eiq·(x−y) )
(2π)3 n
Z
d4 q
= −i ρ(q)(e−iq·(x−y) − eiq·(x−y) ) (2.26)
(2π)3
where we have defined the density ρ(q) (spectral amplitude),

X
ρ(q) = (2π)3 δ4 (pn − q)| h0|ϕ|0)|ni |2 (2.27)
n
2.3. SPECTRAL REPRESENTATION FOR SCALAR FIELDS 55
This spectral amplitude measures the contribution to ∆′ of the states with 4-momentum
q µ . ρ(q) is Lorentz invariant (as can be shown using the invariance of ϕ(x) and the
properties of the vacuum and of the states |ni) and vanishes when q is not in future light
cone, due the assumed properties of the physical states. Then we can write
ρ(q) = ρ(q 2 )θ(q 0 ) (2.28)
and we get
Z
d4 q
∆′ (x − y) = −i ρ(q 2 )θ(q 0 )(e−iq·(x−y) − eiq·(x−y) )
(2π)3
Z Z h i
d4 q 2 2 2 2 0 −iq·(x−y) iq·(x−y)
= −i dσ δ(q − σ )ρ(σ )θ(q ) e − e
(2π)3
Z ∞
= dσ 2 ρ(σ 2 )∆(x − y; σ) (2.29)
0
where Z
d4 q
∆(x − y; σ) = −i δ(q 2 − σ 2 )θ(q 0 )(e−iq·(x−y) − eiq·(x−y) ) (2.30)
(2π)3
is the invariant function defined for the commutator of free fields with mass σ.
The Eq. (2.29) is known as the spectral decomposition of the commutator of two fields.
This expression will allow us to show that 0 ≤ Z < 1. To show that, we separate the states
of one-particle from the sum in Eq. (2.27). Let |pi be a one-particle state with momentum
p. Then
√ Z
h0|ϕ(x)|pi = Z h0|ϕin (x)|pi + d4 y∆ret (x − y; m) h0|j(y)|pi
√
= Z h0|ϕin (x)|pi (2.31)
where we have used

h0|j(y)|pi = ⊓ + m2 )ϕ(y)|p =
0|(⊔
⊓ + m2 )e−ip·y h0|ϕ(0)|pi
= (⊔
= (m2 − p2 )e−ip·y h0|ϕ(0)|pi = 0 (2.32)
On the other hand

Z
d3 k
h0|ϕin (x)|pi = e−ik·x h0|ain (k)|pi
(2π)3 2ωk
= e−ip·x (2.33)
and therefore
Z
ρ(q) = (2π)3 f δ4 (p − q)Z +
dp contributions from more than one particle
= Zδ(q 2 − m2 )θ(q 0 ) + · · · (2.34)

Therefore Z ∞
′
∆ (x − y) = Z∆(x − y; m) + dσ 2 ρ(σ 2 )∆(x − y; σ) (2.35)
m21
where m1 is the mass of the lightest state of two or more particles. Finally noticing that
∂ ∂
∆′ (x − y)|x0 =y0 = ∆(x − y; σ)|x0 =y0 = −δ3 (~x − ~y) (2.36)
∂x0 ∂x0
we get the relation Z ∞
1=Z+ dσ 2 ρ(σ 2 ) (2.37)
m21
which means
0≤Z<1 (2.38)
where this last step results from the assumed positivity of ρ(σ 2 ).
2.4 Out states

In the same way as we reduced the dynamics of t → −∞ to the free fields ϕin , it is also
possible to define in the limit t → +∞ the corresponding free fields, ϕout (x). These free
fields will be the final state of a scattering problem. The formalism is copied from the
case of ϕin , and therefore we will present the results without going into the details of the
derivations. ϕout (x) obey the following relations:
i [P µ , ϕout ] = ∂ µ ϕout
⊓ + m2 )ϕout = 0
(⊔ (2.39)
and has the expansion
Z h i
ϕout (x) = f aout (k)e−ik·x + a† (k)eik·x
dk (2.40)
out
The asymptotic condition is now

√
lim hα| ϕf (t) |βi = Z hα| ϕfout |βi (2.41)
t→∞
and Z
√
Zϕout (x) = ϕ(x) − d4 y∆adv (x − y; m)j(y) (2.42)
where the Green’s functions ∆adv satisfy

⊓x + m2 )∆adv (x − y; m) = δ4 (x − y)
(⊔
∆adv (x − y; m) = 0 ; x0 > y 0 . (2.43)

For one-particle states we get
√
h0|ϕ(x)|pi = Z h0|ϕout (x)|pi
√
= Z h0|ϕin (x)|pi
√ −ip·x
= Ze (2.44)
2.5. S MATRIX 57
2.5 S matrix
We have now all the formalism needed to study the transition amplitudes from one initial
state to a given final state, the so-called S matrix elements. Let us start by an initial state
with n non-interacting particles (we suppose that initially they are well separated),
|p1 · · · pn ; ini ≡ |α ; ini (2.45)
where p1 · · · pn are the 4-momenta of the n particles. Other quantum numbers are assumed
but not explicitly written. The final state will be, in general, a state with m particles
′
p1 · · · p′m ; out ≡ |β ; outi (2.46)
The S matrix element Sβα is defined by the amplitude
Sβα ≡ hβ ; out|α ; ini (2.47)
The S matrix is an operator that induces an isomorphism between the in and out states,
that by assumption are a complete set of states,
hβ ; out| = hβ ; in| S
hβ ; in| = hβ ; out| S −1
hβ ; out|α ; ini = hβ ; in|S|α ; ini = hβ ; out|S|α; outi (2.48)
From the assumed properties for the states we can show the following results for the
S matrix.
i) h0|S|0i = h0|0i = 1 (stability and unicity of the vacuum)
ii) The stability of the one-particle states gives
hp ; in|S|p ; ini = hp ; out|p ; ini = hp ; in|p ; outi = 1 (2.49)
because |p ; ini = |p ; outi.
iii) ϕin (x) = Sϕout (x)S −1
iv) The S matrix is unitary. To show this we have
D E
δβα = hβ ; out|α ; outi = β ; in|SS † |α ; in (2.50)
and therefore
SS † = 1 (2.51)
v) The S matrix is Lorentz invariant. In fact we have
ϕin (ax + b) = U (a, b)ϕin (x)U −1 (a, b) = U Sϕout (x)S −1 U −1
= U SU −1 ϕout (ax + b)U S −1 U −1 . (2.52)
But
ϕin (ax + b) = Sϕout (ax + b)S −1 , (2.53)
and therefore we get finally
S = U (a, b)SU −1 (a, b) . (2.54)
2.6 Reduction formula for scalar fields

The S matrix elements are the quantities that are directly connected to the experiment.
In fact, |Sβα |2 represents the transition probability from the initial state |α ; ini to the
final |β ; outi. We are going in this section to use the previous formalism to express these
matrix elements in terms of the so-called Green functions for the interacting fields. In
this way the problem of the calculation of these probabilities is transferred to the problem
of calculating these Green functions. These, of course, can not be evaluated exactly, but
we will learn in the next chapter how to develop a covariant perturbation theory for that
purpose.
Let us then proceed to the derivation of the relation between the S matrix elements
and the the Green functions of the theory. This technique is known as the LSZ reduction
from the names of Lehmann, Symanzik e Zimmermann [8] that have introduced it. By
definition D E
hp1 · · · ; out|q1 · · · ; ini = p1 , · · · ; out|a†in (q1 )|q2 , · · · ; in (2.55)
Using Z
↔
a†in (q1 ) = −i d3 xe−iq1 x ∂ 0 ϕin (x) (2.56)
t
where the integral is time-independent, and therefore can be calculated for an arbitrary
time t. If we take t → −∞ and use the asymptotic condition for the in fields, Eq. (2.21),
we get
Z
↔
−1/2
hp1 · · · ; out|q1 · · · ; ini = − lim iZ d3 xe−iq1 −·x ∂ 0 hp1 · · · ; out|ϕ(x)|q2 · · · ; ini
t→−∞ t
(2.57)
In a similar way one can show that
D E
p1 · · · ; out|a†out (q1 )|q2 · · · ; in =
Z
↔
−1/2
= − lim iZ d3 xe−iq1 ·x ∂ 0 hp1 · · · ; out|ϕ(x)|q2 · · · ; ini . (2.58)
t→∞ t
2.6. REDUCTION FORMULA FOR SCALAR FIELDS 59
Then, using the result,

Z Z tf Z
∂
lim − lim d3 xf (~x, t) = lim dt d3 xf (~x, t)
t→∞ t→−∞ tf →∞,ti →−∞ t
i
∂t
Z
= d4 x∂0 f (~x, t) (2.59)
and subtracting Eq. (2.58) from Eq. (2.57) we get

D E
hp1 · · · ; out|q1 · · · ; ini = p1 · · · ; out|a†out (q1 )|q2 · · · ; in
Z h i
↔
−1/2
+iZ d4 x ∂0 e−iq1 ·x ∂ 0 hp1 · · · ; out|ϕ(x)|q2 · · · ; ini (2.60)
The first term on the right-hand side of Eq. (2.60) corresponds to a sum of disconnected
terms, in which at least one of the particles is not affected by the interaction (it will vanish
if none of the initial momenta coincides with one of the final momenta). Its form is
D E
p1 · · · ; out|a†out (q1 )|q2 · · · ; in =
n
X
= (2π)3 2p0k δ3 (~
pk − ~q1 ) hp1 , · · · , pbk , · · · ; out|q2 , · · · ; ini (2.61)
k=1
where pbk means that this momentum was taken out from that state. For the second term
we write,
Z h i
↔
d4 x ∂0 e−iqi x ∂ 0 hp1 · · · ; out|ϕ(x)|q2 · · · ; ini
Z

= d4 x −∂02 e−iq1 ·x h· · · i + e−iq1 ·x ∂02 h· · · i
Z

= d4 x (−∆2 + m2 )e−iq1 ·x h· · · i + e−iq1 ·x ∂02 h· · · i
Z
= d4 xe−iq1 ·x (⊔
⊓ + m2 ) hp1 · · · ; out|ϕ(x)|q2 · · · ; ini (2.62)
where we have used (⊔ ⊓ + m2 )e−iq1 ·x = 0, and have performed an integration by parts

(whose justification would imply the substitution of plane waves by wave packets).
Therefore, after this first step in the reduction we get,
hp1 , · · · pn ; out|q1 · · · qℓ ; ini =

n
X
= 2p0k (2π)3 δ3 (~
pk − ~
q1 ) hp1 , · · · pbk ; · · · pn ; out|q2 · · · q2 · · · qℓ ; ini
k=1
Z
+iZ −1/2 d4 xe−iq1 x (⊔
⊓ + m2 ) hp1 · · · pn ; out|ϕ(x)|q2 · · · qℓ ; ini (2.63)
We will proceed with the process until all the momenta in the initial and final state are
exchanged by the field operators. To be specific, let us now remove one momentum in
the final state. From now on we will no longer consider the disconnected terms, because
in practice we are only interested in the cases where all the particles interact1 . We have
then
hp1 · · · pn ; out|ϕ(x1 )|q2 · · · qℓ ; ini = hp2 · · · pn ; out|aout (p1 )ϕ(x1 )|q2 · · · qℓ ; ini
Z
↔
−1/2
= lim iZ d3 y1 eip1 ·y1 ∂ y10 hp2 · · · pn ; out|ϕ(y1 )ϕ(x1 )|q2 · · · qℓ ; ini
y10 →∞
= hp2 · · · pn ; out|ϕ(x1 )ain (p1 )|q2 · · · qℓ ; ini

Z
↔
+ lim iZ −1/2 d3 y1 eip1 ·y1 ∂ y10 hp2 · · · pn ; out|ϕ(y1 )ϕ(x1 )|q2 · · · qℓ ; ini
y10 →∞
Z
↔
− lim iZ −1/2 d3 y1 eip1 ·y1 ∂ y10 hp2 · · · pn ; out|ϕ(x1 )ϕ(y1 )|q2 · · · qℓ ; ini
y10 →−∞
= hp2 · · · pn ; out|ϕ(x1 )ain (p1 )|q2 · · · qℓ ; ini

!Z
↔
+iZ −1/2 lim − lim d3 y1 eip1 ·y1 ∂ y10 hp2 · · · pn ; out|T ϕ(y1 )ϕ(x1 )|q2 · · · qℓ ; ini
y10 →∞ y10 →−∞
(2.64)
where we have used the properties of the time-ordered product, Eq. (1.157). Applying the
same procedure that lead to Eq. (2.62) we obtain,
hp1 · · · pn , ; out|ϕ(x1 )|q2 · · · qℓ ; ini = disconnected terms

Z
−1/2
+iZ d4 y1 eip1 ·y1 (⊔
⊓y1 + m2 ) hp2 · · · pn ; out|T ϕ(y1 )ϕ(x1 )|q2 · · · qℓ ; ini (2.65)
It is not very difficult to generalize this method to obtain the final reduction formula
for scalar fields,
hp1 · · · pn ; out|q1 · · · qℓ ; ini = disconnected terms

Z
i n+ℓ Pn Pℓ
+ √ d4 y1 · · · d4 y1 d4 x1 · · · d4 xℓ e[i 1 pk ·yk −i 1 qr ·xr ]
Z
⊓y1 + m2 ) · · · (⊔
(⊔ ⊓xℓ + m2 ) h0|T ϕ(y1 ) · · · ϕ(yn )ϕ(x1 ) · · · ϕ(xℓ )|0i (2.66)
This last equation is the fundamental equation in quantum field theory. It allows us to
relate the transition amplitudes to the Green functions of the theory. The quantity
h0|T ϕ(x1 ) · · · ϕ(xn )|0i ≡ G(x1 · · · xn ) (2.67)

1
Once we know the cases where all the particles interact, we can always calculate situations where some
of the particles do not participate in the scattering.
2.7. REDUCTION FORMULA FOR FERMIONS 61
is known as the complete green function for r = m + ℓ particles and we will introduce the
following diagrammatic representation for it,
xn xl
G(x1 · · · xn ) = (2.68)
x1 xi
⊓ + m2 ) in Eq. (2.66) force the external particles to be on-shell. In fact, in

The factors (⊔
momentum space (⊔ ⊓ + m2 ) → (−p2 + m2 ). Therefore, Eq. (2.66) will vanish unless the
1
propagators of the external legs are on-shell, as in that case they will have a pole, p2 −m2.
Eq. (2.66) will then give the residue of that pole. We conclude that for the transition
amplitudes only the truncated Green functions will contribute, that is the ones with the
external legs removed. In the next chapter we will learn how to evaluate these Green
functions in perturbation theory.
2.7 Reduction formula for fermions

2.7.1 States in and out
The definition of the in and out follows exactly the same steps as in the case of the scalar
fields. We will therefore, for simplicity, just state the results with the details.
The states ψin (x) satisfy the conditions,
(i∂/ − m)ψin (x) = 0
[Pµ , ψin (x)] = −i∂µ ψin (x) . (2.69)
The states ψin (x) will create one-particle states and they are related with the fields ψ(x)
by, Z
p
Z2 ψin (x) = ψ(x) − d4 ySret (x − y, m)j(y) (2.70)
where ψ(x) satisfies the Dirac equation,
(i∂/ − m)ψ(x) = j(x) (2.71)
and Sret is the retarded Green function for the Dirac equation,
(i∂/x − m)Sret (x − y, m) = δ4 (x − y)
Sret(x − y) = 0 ; x0 < y 0 (2.72)

The fields ψin (x), as free fields, have the Fourier expansion,
Z Xh i
ψin (x) = dp f bin (p, s)u(p, s)e−ip·x + d†in (p, s)v(p, s)eip·x (2.73)
s
where the operators bin , din satisfy exactly the same algebra as in the free field case. The
asymptotic condition is now,
p f
lim hα| ψ f (t) |βi = Z2 hα| ψin |βi (2.74)
t→−∞
f
where ψ f (t) and ψin have a meaning similar to Eq. (2.20).
For the ψout fields we get essentially the same expressions with ψin substituted by ψout .
The main difference is in the asymptotic condition that now reads,
p f
lim hα| ψ f (t) |βi = Z2 hα| ψout |βi (2.75)
t→∞
implying the following relation between the fields ψout and ψ,

p Z
Z2 ψout = ψ(x) − d4 ySadv (x − y; m)j(y) (2.76)
where
(i∂/x − m)Sadv (x − y; m) = δ4 (x − y)
Sadv (x − y; m) = 0 x0 > y 0 . (2.77)
2.7.2 Spectral representation fermions

Let us consider the vacuum expectation value of the anti-commutator of two Dirac fields,
′
Sαβ (x, y) ≡ i h0| {ψα (x), ψ β (y)} |0i
X
= i h0| ψα (0) |ni hn| ψ β (0) |0i e−ipn (x−y)
n

ipn ·(x−y)
+ h0| ψ β (0) |ni hn| ψα (0) |0i e
′
≡ Sαβ (x − y) (2.78)
where we have introduced a complete set of eigen-states of the 4-momentum. As before

we introduce the spectral amplitude ραβ (q),
X
ραβ (q) ≡ (2π)3 δ4 (pn − q) h0| ψα (0) |ni hn| ψ β (0) |0i (2.79)
n
We will now find the most general form for ραβ (q) using Lorentz invariance arguments.
ραβ (q) is a 4 × 4 matrix in Dirac space, and can therefore be written as
µ µν 5
ραβ (q) = ρ(q)δαβ + ρµ (q)γαβ + ρµν (q)σαβ + ρ̃(q)γαβ + ρ̃µ (q)(γ µ γ 5 )αβ (2.80)
Lorentz invariance arguments restrict the form of the coefficients ρ(q), ρµ (q), ρµν (q), ρ̃(q)
and ρ̃µ (q). Under Lorentz transformations the fields transform as
−1
U (a)ψα (0)U −1 (a) = Sαλ (a)ψ λ (0)
U (a)ψ α (0)U −1 (a) = ψ λ (0)Sλα (a)
S −1 γ µ S = aµ ν γ ν (2.81)
Then we can show that the matrix (in Dirac space), ραβ must obey the relation,
ρ(q) = S −1 (a)ρ(qa−1 )S(a) (2.82)
where we have used a matrix notation. This relation gives the properties of the different
coefficients on Eq. (2.80). For instance,
ρµ (q) = aµ ν ρν (qa−1 ) (2.83)
which means that ρµ transform as a 4−vector.

Using the fact that ραβ is a function of q and vanishes outside the future light cone,
we can finally write
ραβ (q) = ρ1 (q 2 )q/αβ + ρ2 (q 2 )δαβ + ρ̃1 (q 2 )(q/γ 5 )αβ + ρ̃2 (q 2 )γαβ

5
(2.84)
that is, ραβ (q) is determined up to four scalar functions of q 2 . Requiring invariance under
parity transformations we get, instead of Eq. (2.82),
0
ραβ (~
q , q0 ) = γαλ ρλδ (−~q, q 0 )γδβ
0
(2.85)
and inserting in Eq. (2.84) we obtain,
ρe1 = ρe2 = 0 (2.86)
Therefore for the Dirac theory, that preserves parity, we get,
ραβ (q) = ρ1 (q 2 )q/αβ + ρ2 (q 2 )δαβ (2.87)
Repeating the steps of the scalar case we write,

Z ∞
′

Sαβ (x − y) = dσ 2 ρ1 (σ 2 )Sαβ (x − y; σ)+
0

+ σρ1 (σ 2 ) − ρ2 (σ 2 ) δαβ ∆(x − y; σ) (2.88)
where ∆ and Sαβ are the functions defined for free fields. We can then show that
i) ρ1 e ρ2 are real
ii) ρ1 (σ 2 ) ≥ 0
iii) σρ1 (σ 2 ) − ρ2 (σ 2 ) ≥ 0
Using the previous relations we can extract of the one-particle states from Eq. (2.88). We
get,
′
Sαβ (x − y) = Z2 Sαβ (x − y; m)
Z ∞
+ 2
dσ ρ1 (σ 2 )Sαβ (x − y; σ)
m21

2
2
+ σρ1 (σ ) − ρ2 (σ ) δαβ ∆(x − y; σ) (2.89)
where m1 is the threshold for the production of two or more particles. Evaluating
Eq. (2.89) at equal times we can obtain
Z
1 = Z2 + dσ 2 ρ1 (σ 2 ) (2.90)
m21
that is
0 ≤ Z2 < 1 (2.91)
2.7.3 Reduction formula fermions

To get the reduction formula for fermions we will proceed as in the scalar case. The only
difficulty has to do with the spinor indices. The creation and annihilation operators can
be expressed in terms of the fields ψin by the relations,
Z
bin (p, s) = d3 xu(p, s)eip·x γ 0 ψin (x)
Z
d†in (p, s) = d3 xv(p, s)e−ip·x γ 0 ψin (x)
Z
b†in (p, s) = d3 xψ in (x)γ 0 e−ip·x u(p, s)
Z
din (p, s) = d3 xψ in (x)γ 0 eip·x v(p, s) (2.92)
with the integrals being time dependent. In fact, to be more rigorous we should substitute
the plane wave solutions by wave packets, but as in the scalar case, to simplify matter we
will not do it. To establish the reduction formula we start by extracting one electron from
the initial state,
D E
hβ; out|(ps)α; ini = β; out|b†in (p, s)|α, in
D E
= hβ − (p, s); out|α; ini + β; out|b†in (p, s) − b†out (p, s)|α; in
= disconnected terms
Z

+ d3 x β; out|ψ in (x) − ψ out (x)|α; in γ 0 e−ip·x u(p, s)
Z
1

− lim − lim √ d3 x β; out|ψ(x)|α; in γ 0 e−ip·x u(p, s)
t→+∞ t→−∞ Z2
Z
−1/2

−Z2 d4 x β; out|∂0 ψ(x)|α; in γ 0 e−ip·x u(p, s)

+ β; out|ψ(x)|α; in γ 0 ∂0 (e−ip·x u(p, s)) (2.93)
Using now
(iγ 0 ∂0 + iγ i ∂i − m)(e−ip·x u(p, s)) = 0 (2.94)
we get, after an integration by parts,
D E
β; out|b†in (ρ, s)|α; in = disconnected terms
Z
−1/2
←
−iZ2 d4 x β; out|ψ(x)|α; in (−i∂/x − m)e−ip·x u(p, s) (2.95)
In a similar way the reduction of an anti-particle from the initial state gives,
D E
β; out|d†in (p, s)|α; in = disconnected terms
Z
−1/2
+iZ2 d4 xe−ip·x v(p, s)(i∂/x − m) hβ; out|ψ(x)|α; ini (2.96)
while the reduction of a particle or of an anti-particle from the final state give, respectively,
hβ; out|bout (p, s)|α; ini = disconnected terms

Z
−1/2
−iZ2 d4 xeip·x u(p, s)(i∂/x − m) hβ; out|ψ(x)|α; ini (2.97)
and
hβ; out|dout (p, s)|α; ini = disconnected terms

Z
−1/2
←
+iZ2 d4 x β; out|ψ(x)|α; in (−i∂/x − m)v(p, s)eip·x (2.98)
Notice the formal relation between one electron in the initial state and a positron in the
final state. To go from one to the other one just has to do,
u(p, s)e−ip·x → −v(p, s)eip·x (2.99)
The following steps in the reduction are similar, one only has to pay attention to signs
because of the anti-commutation relations for fermions. To write the final expression we
denote the momenta in the state hin| by pi or pi , respectively for particles or anti-particles,
and those in the state hout| by p′i , p′i . We also make the following conventions (needed to
define the global sign),
|(p1 , s1 ), · · · , (p1 , s1 ); · · · ; ini = b†in (p1 , s1 ) · · · d†in (p1 , s1 ) · · · |0i (2.100)
and

out; (p′1 , s′1 ) · · · , (p′1 , s′1 ) · · · = h0| · · · dout (p′1 , s′1 ), · · · bout (p′1 , s′1 ) (2.101)
Then, if n(n′ ) denotes the total number of particles (anti-particles), we get

out; (p′1 , s′1 ) · · · , (p′1 , s′1 ) · · · |(p1 , s1 ), · · · (p1 , s1 ), · · · ; in = disconnected terms
Z
−1/2 n −1/2 n′
+(−iZ2 ) (iZ2 ) d4 x1 · · · d4 y1 · · · d4 x′1 · · · d4 y1′ · · ·
(pi ·xi )−i (pi ·yi ) +i (p′i ·x′i )+i (p′i ·yi′ )
P P P P
e−i e
u(p′1 , s′1 )(i∂/x′1 − m) · · · v(p1 , s1 )(i∂/y1 − m)
h0| T (· · · ψ(y1′ ) · · · ψ(x′1 )ψ(x1 ) · · · ψ(y1 ) · · · |0i

← ←
(−i∂/x1 − m)u(p1 , s1 ) · · · (−i∂/y1′ − m)v(p′1 , s′1 ) (2.102)
Eq. (2.102) is the fundamental expression that allows ti relate the elements of the S
matrix with the Green functions of the theory. The operators within the time-ordered
product can be reordered, modulo some minus sign. The sign and ordering shown corre-
spond to the conventions in Eqs. (2.100) and (2.101). In terms of diagrams, we represent
the Green function,
′ ′ ′ ′

h0| T ψ(ym ′ ) · · · ψ(y1 )ψ(xℓ′ ) · · · ψ(x1 )ψ(x1 ) · · · ψ(xℓ )ψ(y1 ) · · · ψ(ym ) |0i (2.103)
by the diagram2 of Fig. (2.1).

←
The operators (i∂/ − m) e (−i∂/ − m) force the particles to be on-shell and remove the
propagators from the external lines (truncated Green functions). In the next chapter we
will learn how to determine these functions in perturbation theory.
2.8 Reduction formula for photons

The LSZ formalism for photons, has some difficulties connected with the problems in
quantizing the electromagnetic field. When one adopts a formalism (radiation gauge)
where the only components of the field Aµ are transverse (as in Ref.[6]), the problems
arise in showing the Lorentz and gauge invariance of the S matrix. In the formalism of
the undefined metric, that we adopted in section 1.4.2, the difficulties are connected with
the states of negative norm, besides the gauge invariance.
2
With lepton number conservation, the number of particles minus anti-particles is conserved, that is
ℓ − m = ℓ′ − m′
2.8. REDUCTION FORMULA FOR PHOTONS 67
x′l y1′ ′
x′1 ym
x1 xl y1 ym
Figure 2.1: Green function for fermions.
Here we are going to ignore these difficulties3 and assume that we can define the in
fields by the relation,
p Z
µ µν
Z3 Ain (x) = A (x) − d4 yDret
µ
(x − y)jν (y) (2.104)
and in the same way for the out fields,

p Z
Z3 Aµout (x) = Aµ (x) − µν
d4 yDadv (x − y)jν (y) (2.105)
where
⊓Aµin = ⊔
⊔ ⊓Aµout = 0
⊓Aµ = j µ
⊔
µν
⊓Dadv,
⊔ ret = δµν δµ (x − y) (2.106)
The fields in and out are free fields, and therefore they have a Fourier expansion in
plane waves and creation and annihilation operators of the form
Z 3 h
X i
Aµin (x) = f
dk µ
ain (k, λ)ε (k, λ)e −ik·x
+ a†in (k, λ)εµ∗ (k, λ)eik·x (2.107)
λ=0
and therefore
Z
↔
ain (k, λ) = −i d3 xeik·x ∂ 0 εµ (k, λ)Ain
µ (x)
Z
↔
a†in (k, λ) = i d3 xe−ik·x ∂ 0 εµ∗ (k, λ)Ain
µ (x) (2.108)
3
We will see in chapter 6 a more satisfactory procedure to quantize all gauge theories, including Maxwell
theory of the electromagnetic field. We will see that the resulting perturbation theory coincides with the
one we get here. This is our justification to be less precise here.
where, as usual, ain (k, λ) and a†in (k, λ) are time independent. In Eq. (2.107) all the
polarizations appear, but as the elements of the S matrix are between physical states, we
are sure that the longitudinal and scalar polarizations do not contribute. In this formalism
what is difficult to show is the spectral decomposition. We are not going to enter those
details, just state that we can show that Z3 is gauge independent and satisfies 0 ≤ Z3 < 1.
The reduction formula is easily obtained. We get
D E
hβ; out|(kλ)α; ini = hβ − (k, λ); out|α; ini + β; out|a†in (k, λ) − a†out (k, λ)|α; in
Z
↔
+i d3 xe−ik·x ∂ 0 ε∗µ (k, λ) hβ; out|Aµin (x) − Aµout (x)|α; ini
Z
−1/2 ↔
−i( lim − lim )Z3 d3 xe−ik·x ∂ 0 hβ; out|Aµ (x)|α; ini ε∗µ (k, λ)
t→+∞ t→−∞
Z
−1/2 ↔
−iZ3 d4 xe−ik·x ∂ 0 hβ; out|Aµ (x)|α; ini ε∗µ (k, λ)
Z
−1/2
−iZ3 ⊓x hβ; out|Aµ (x)|α; ini ε∗µ (k, λ)
d4 xe−ik·x ~
⊔ (2.109)
The final formula for photons is then

k1′ · · · kn′ ; out|k1 · · · kℓ ; in = disconnected terms
Z
−i n+ℓ Pn ′
ki ·yi −i ℓ ki ·xi ]
d4 y1 · · · d4 yn d4 x1 · · · d4 xℓ e[i
P
+ √
Z3
′ ′
εµ1 (k1 , λ1 ) · · · εµℓ (kℓ , λℓ )ε∗µ1 (k1′ , λ′1 ) · · · ε∗µn (kn′ , λ′n )
⊓ y1 · · · ⊔
⊔ ⊓xℓ h0| T (Aµ′1 (y1 ) · · · Aµ′n (yn )Aµ1 (x1 ) · · · Aµℓ (xℓ ) |0i (2.110)
and corresponds to the diagram of Fig. (2.2).
2.9 Cross sections

The reduction formulas, Eqs.(2.66), (2.102) and (2.110), are the fundamental results of
this chapter. They relate the transition amplitudes from the initial to the final state with
the Green functions of the theory. In the next chapter we will show how to evaluate these
Green functions setting up the so-called covariant perturbation theory. Before we close
this chapter, let us indicate how these transition amplitudes
Sf i ≡ hf ; out|i; ini (2.111)

2.9. CROSS SECTIONS 69
xn xl
x1 xi
Figure 2.2: Green function for photons.
are related with the quantities that are experimentally accessible, the cross sections. Then
the path between experiment (cross sections) and theory (Green functions) will be estab-
lished.
As we have seen in the reduction formulas there is always a trivial contribution to the
S matrix, that corresponds to the so-called disconnected terms, when the system goes from
the initial to the final state without interaction. The subtraction of this trivial contribution
leads us to introduce the T matrix with the relation,
Sf i = 1f i − i(2π)4 δ4 (Pf − Pi )Tf i (2.112)
where we have factorized explicitly the delta function expressing the 4-momentum conser-
vation. If we neglect the trivial contribution, the transition probability from the initial to
the final state will be given by
2
Wf ←i = (2π)4 δ4 (Pf − Pi )Tf i (2.113)
To proceed we have to deal with the meaning of a square of a delta function. This
appears because we are using plane waves. To solve this problem we can normalize in a
box of volume V and consider that the interaction has a duration of T . Then
Z Z T /2
4 4 3
(2π) δ (Pf − Pi ) = lim d x dx0 ei(Pf −Pi )·x . (2.114)
V →∞ V −T /2
T →∞
However
Z Z
T /2
2 T
F ≡ 3
d x 0 i(Pf −Pi )·x
dx e = V δP~f P~j sin (Ef − Ei )
(2.115)
V −T /2 |Ef − Ei | 2
and the square of the last expression can be done, giving,

4
2 T

2 2
|F | = V δP~f ,P~i 2
sin (Ef − Ei ) . (2.116)
|Ef − Ei | 2
If we want the transition rate by unit of volume (and unit of time) we divide by V T . Then
sin2 T2 (Ef − Ei )
Γf i = lim V δP~f ,P~i 2 T
|Tf i |2 (2.117)
V →∞ 2 (Ef − Ei )2
T →∞
Using now the results
lim V δP~f P~j = (2π)3 δ3 (P~f − P~i )

V →∞
sin2 T2 (Ef − Ei )
lim 2 T
= (2π) δ(Ef − Ei ) (2.118)
T →∞
2 (Ef − Ei )2
we get for the transition rate by unit volume and unit time,
Γf i ≡ (2π)4 δ4 (Pf − Pi )|Tf i |2 (2.119)
To get the cross section we have to further divide by the incident flux, and normalize the
particle densities to one particle per unit volume. Finally, we sum (integrate) over all final
states in a certain energy-momentum range. We get,
n
Y d3 pj
1 1
dσ = Γf i (2.120)
ρ1 ρ2 |~v12 |
j=3
2p0j (2π)3
where
ρ1 = 2E1 ; ρ2 = 2E2 (2.121)
An equivalent way of writing this equation is
n
Y
1 4 4 2
dσ = 1/2 (2π) δ (Pf − Pi )|Tf i | dpj (2.122)
4 (pi · p2 )2 − m21 m22 j=3
that exhibits well the Lorentz invariance of each part that enters the cross section4 . The
incident flux and phase space factors are purely kinematics. The physics, with its inter-
actions, is in the matrix element Tf i .
We note that with our conventions, fermion and boson fields have the same normal-
ization, that is, the one-particle states obey

′
p|p = 2p0 (2π)3 δ3 (~
p − ~p′ ) (2.123)
differing in this way from some older books like Ref.[1].
4
It is assumed that, in the case of two beams they are in the same line. Then the cross section, being
a transverse area, is invariant for Lorentz transformations along that direction.
Problems 71
2.1 Show that the spectral representation for fermions, ραβ (q), satisfies,
a) ρ(q) = S −1 (a)ρ(qa−1 )S(a)
q , q 0 ) = γαλ
b) ραβ (~ 0 ρ (−~
λδ q , q 0 )γδβ
0
2.2 Use the results of the previous problem to show that, in a theory that preserves Parity,
like QED, we have
ραβ (q) = ρ1 (q 2 )q/αβ + ρ2 (q 2 )δαβ (2.124)
2.3 Show that the functions ρ1 and ρ2 defined in problem 2.2 satisfy the following prop-
erties:
i) ρ1 and ρ2 are real
ii) ρ1 (σ 2 ) ≥ 0
iii) σρ1 (σ 2 ) − ρ2 (σ 2 ) ≥ 0
2.4 Show that for the Dirac field we have

Z ∞
1 = Z2 + dσ 2 ρ1 (σ 2 ) (2.125)
m21
2.5 Show that
h0| [ϕin (x), ϕout (y)] |0i = i∆(x − y; m) (2.126)

Chapter 3
Covariant Perturbation Theory
3.1 U matrix
In this chapter we are going to develop a method to evaluate the Green functions of a
given theory. From what we have seen in the two previous chapters, we realize that we
only know how to calculate for free fields, like the in an out fields. However, the Green
functions we are interested in, are given in terms of the physical interacting fields, and we
do not know how to operate with these. We are going to see how to express the physical
fields as perturbative series in terms of free in fields. In this way we will be able to evaluate
the Green functions in perturbation theory.
We start by defining the U matrix. To simplify matters, we will be considering, for
the moment, only scalar fields. In the end we will return to the other cases. The physical
interacting fields ϕ(~x, t) and their conjugate momenta π(~x, t), satisfy the same equal time
commutation relations than the in fields, ϕin (~x, t) and their πin (~x, t). Also, both ϕ and
ϕin form a complete set of operators, in the sense that any state, free or interacting, can
be obtained by application of ϕin or ϕ in the vacuum. This implies that should be an
unitary transformation U (t) that relates ϕ with ϕin , that is,
ϕ(~x, t) = U −1 (t)ϕin (~x, t)U (t)
π(~x, t) = U −1 (t)πin (~x, t)U (t) (3.1)
The dynamics of U can be obtained from the equations of motion for ϕ(x) and ϕin (x).
These are,
∂ϕin
(x) = i[Hin (ϕin , πin ), ϕin ]
∂t
∂πin
(x) = i[Hin (ϕin , πin ), πin ] (3.2)
∂t
and
∂ϕ
(x) = i[H(ϕ, π), ϕ]
∂t
∂π
(x) = i[H(ϕ, π), π] (3.3)
∂t
73
74 CHAPTER 3. COVARIANT PERTURBATION THEORY
Then from Eqs. (3.2) and (3.1) we get,
∂
ϕ̇in (x) = U (t)ϕ(x)U −1 (t)
∂t
h i
= U̇ (t)U −1 (t), ϕin + i [H(ϕin , πin ), ϕin (x)]
h i
= ϕ̇in (x) + U̇ U −1 + iHI (ϕin , πin ), ϕin (3.4)
where
HI (ϕin , πin ) = H(ϕin , πin ) − Hin (ϕin , πin ) ≡ HI (t) (3.5)
and in a similar way
h i
π̇in (x) = π̇in + U̇ U −1 + iHI (ϕin , πin ), πin (3.6)
From Eqs. (3.4) and (3.6) we obtain,
iU̇ U −1 = HI (t) + E0 (t) (3.7)
where E0 (t) commutes with ϕin and πin and is therefore a time dependent c-number, not
an operator. Defining
HI′ (t) = HI (t) + E0 (t) (3.8)
we get a differential equation for U (t), that reads,
∂U (t)
i = HI′ (t)U (t) (3.9)
∂t
The solution of this equation in terms of the in fields, is the basis of the covariant pertur-
bation theory.
To integrate Eq. (3.9) we need an initial condition. For that we introduce the operator
U (t, t′ ) ≡ U (t)U −1 (t′ ) (3.10)
where t ≥ t′ , and that obviously satisfies
U (t, t) = 1 (3.11)
It is easy to see that U (t, t′ ) also satisfies Eq. (3.9), that is,
∂U (t, t′ )
i = HI′ (t)U (t, t′ ) (3.12)
∂t
and has the initial condition, Eq. (3.11). To proceed we start by transforming Eq. (3.12)
in an equivalent integral equation, that is,
Z t
′
U (t, t ) = 1 − i dt1 HI′ (t1 )U (t1 , t′ ) (3.13)
t′
3.1. U MATRIX 75
Notice that we have not solved the problem because U (t, t′ ) appears on both sides of the
equation. However, we can iterate the equation to get the expansion,
Z t Z t Z t1
′
U (t, t ) = 1 − i dt1 HI′ (t1 ) + (−i) 2
dt1 HI′ (t1 ) dt2 HI′ (t2 )
t′ t′ t′
Z t Z t1 Z tn−1
n
+ · · · + (−i) dt1 dt2 · · · dtn HI′ (t1 ) · · · HI′ (tn )
t′ t′ t′
+··· (3.14)
Of course this expansion can only be useful if HI contains a small parameter and, because
of that, we can truncate the expansion at certain order in that parameter. Coming back
to Eq. (3.14), as t1 ≥ t2 ≥, · · · tn , the product is time-ordered and we can therefore write
∞
X Z t Z t1 Z tn−1
′ n
U (t, t ) = 1 + (−i) dt1 dt2 · · · dtn T (HI′ (t1 ) · · · HI′ (tn )) (3.15)
n=1 t′ t′ t′
Using the symmetry t1 , t2 we can write,

Z t Z t1 Z t Z t2
dt1 dt2 T (HI′ (t1 )HI′ (t2 )) = dt2 dt1 T (HI′ (t1 )HI′ (t2 ))
t′ t′ t′ t′
Z t Z t
1
= dt1 dt2 T (HI′ (t1 )H2′ (t2 )) (3.16)
2 t′ t′
1 1
In general, for n integrations, instead of 2 we will have n! , and we get,
∞
X Z t Z t
(−i)n
U (t, t′ ) = 1 + dt1 · · · dtn T (HI′ (t1 ) · · · HI′ (tn ))
n! t′ t′
n=1
Z t
≡ T exp[−i dtHI′ (t)]
t′
Z t
4
= T exp[−i d xHI (ϕin )] (3.17)
t′
where the time-ordered product is to be interpreted expanding the exponential.

The operators U satisfy the following multiplication rule
U (t, t′ ) = U (t, t′′ )U (t′′ , t′ ) (3.18)
which can seen using the definition, Eq. (3.10), or from the explicit expression, Eq. (3.17).
From Eq. (3.18), we can obtain,
U (t, t′ ) = U −1 (t′ , t) (3.19)

3.2 Perturbative expansion of Green functions

As we saw in the previous chapter, the LSZ technique reduces the evaluation of the
elements of the S matrix to a basic ingredient, the so-called Green functions of the theory.
These are expectation values of time-ordered products of the Heisenberg fields, ϕ(x),
G(x1 , · · · , xn ) ≡ h0| T ϕ(x1 )ϕ(x2 ) · · · ϕ(xn ) |0i (3.20)
The basic idea for the evaluation of the Green functions consists in expressing the fields
ϕ(x) in terms of the fields ϕin (x), using the operator U . We get
G(x1 , · · · , xn ) = h0| T (U −1 (t1 )ϕin (x1 )U (t1 , t2 )ϕin (x2 )U (t2 , t3 ) · · ·
· · · U (tn−1 , tn )ϕin (xn )U (tn )) |0i
= h0| T (U −1 (t)U (t, t1 )ϕin (x1 )U (t1 , t2 ) · · ·
· · · U (tn−1 , tn )ϕin (xn )U (tn , −t)U (−t)) |0i (3.21)
where t is a time that we will let go to ∞. When t → ∞, t is later than all the ti and −t is
earlier than all the times ti . Therefore we can take U −1 (t) e U (−t) out of the time-ordered
product. Using the multiplicative property of the operator U we can then write,
Z t
−1 ′ ′ ′
G(x1 , · · · , xn ) = h0| U (t)T ϕin (x1 ) · · · ϕin (xn ) exp[−i HI (t )dt ] U (−t) |0i (3.22)
−t
where the time-ordered product T is meant to be applied after expanding the exponential.
If it were not for the presence of the operators U −1 (t) and U (−t), we would have been
successful in expressing the Green function G(x1 · · · xn ) completely in terms if the in fields.
Now we are going to show that the vacuum is an eigenstate of the operator U (t). For that
we consider an arbitrary state |αp; ini that contains one particle of momentum p, all
the other quantum numbers being denoted collectively by α. To simplify, we continue
considering the case of the scalar field. We have then,
hαp; in|U (−t)|0i = hα; in|ain (p)U (−t)|0i

Z → ← !
3 ∗ ′ ∂ ∂
= −i d xfp (~x, −t ) ′
− ′ hα; in| ϕin (~x, −t′ )U (−t) |0i
∂t ∂t
(3.23)
where fp (~x, t) = e−ip·x . We use now Eq. (3.1) to express ϕin (~x, −t) in terms of ϕ(~x, −t).
We get,
hαp; in|U (−t)|0i =

Z
↔′

= −i d3 xfp∗ (~x, −t′ )∂ 0 α; in|U (−t′ )ϕ(~x, −t′ )U −1 (−t′ )U (−t)|0
Z h ←
= −i d3 xfp∗ (~x, −t′ ) − ∂ 0′ hα; in| U (−t′ )ϕ(~x, −t′ )U −1 (−t′ )U (−t) |0i
3.2. PERTURBATIVE EXPANSION OF GREEN FUNCTIONS 77
i
+ hα; in| U (−t′ )ϕ̇(~x, −t′ )U −1 (−t′ )U (−t) |0i
Z ′
+i d3 xfp∗ (~x, −t′ ) hα; in| U̇ (−t′ )ϕ(~x, −t′ )U −1 (−t′ )U (−t) |0i
Z
+i d3 xfp∗ (~x, −t′ ) hα; in| U (−t′ )ϕ(~x, −t′ )U̇ −1 (−t′ )U (−t) |0i (3.24)
We take now the t = t′ → ∞ limit. Then

√
hαp; in| U (−t) |0i = Z hα; in| U (−t)ain (p) |0i
+ hα; in| U̇ (−t)ϕ(~x, −t) + U (−t)ϕ(~x, −t)U̇ −1 (−t)U (−t) |0i (3.25)
Now the first term in Eq. (3.25) vanishes because ain (p) |0i = 0. The second term also
vanishes because we have (we omit the arguments to simplify the notation),
U̇ ϕ + U ϕU̇ −1 U = U̇ U −1 ϕin U + ϕin U U̇ −1 U
= U̇ U −1 ϕin U − ϕin U̇ U −1 U
= [U̇ U −1 , ϕin ]U = −i[HI , ϕin ]U = 0 (3.26)
where we have used Eq. (3.7) and assumed that the interactions have no derivative1 . We
conclude then that,
lim hαp; in|U (−t)|0i = 0 (3.27)
t→∞
for all states in that contain at least one particle. This means that,
lim U (−t) |0i = λ− |0i (3.28)

t→∞
In a similar way we could show that,
lim U (t) |0i = λ+ |0i (3.29)

t→∞
Returning now to the expression for the Green function, we can write,
Z t
∗ ′ ′ ′
G(x1 , · · · xn ) = λ− λ+ h0| T ϕin (x1 ) · · · ϕin (xn ) exp −i HI (t )dt |0i (3.30)
−t
The dependence in the operator U disappeared from the expectation value. To proceed,
let us evaluate the constants λ± , or more to the point, the combination λ− λ∗+ that appears
in Eq. (3.30). We get (in the limit → ∞),
λ− λ∗+ = h0| U (−t) |0i h0| U −1 (t) |0i

1
The study of theories with derivatives was not trivial before the quantization via path integral was
introduced. As we will be viewing this method for gauge theories, we can avoid here the complications of
the derivatives. The quantization via path integral is the only method that is available for non-abelian
gauge theories as we be discussing in chapter 6.
= h0| U (−t)U −1 (t) |0i = h0| U (−t, t) |0i

Z t
′ ′ ′
= h0| T exp +i dt HI (t ) |0i
−t
Z t
= h0| T exp −i dt ′
HI′ (t′ ) |0i−1 (3.31)
−t
Using this result we can write the Green function of Eq. (3.30) in the form,
Rt
h0| T (ϕin (x1 ) · · · ϕin (xn ) exp[−i −t dt′ HI′ (t′ )]) |0i
G(x1 , · · · , xn ) = Rt (3.32)
h0| T (exp[−i −t dt′ HI′ (t′ )) |0i
when t → ∞. Before we write the final expression, we can now introduce the number
E0 (t). For that we recall that,
HI′ = HI + E0 (3.33)
Rt
and noticing that E0 is not an operator, we get a factor exp[−i −t dt′ E0 (t′ )] both in the
numerator and denominator, canceling out in the final result. The final result can then
be obtained from Eq. (3.32), just substituting HI′ by HI . We get,
Rt
h0| T (ϕin (x1 ) · · · ϕin (xn ) exp[−i −t dt′ HI (t′ )]) |0i
G(x1 · · · , xn ) = Rt
h0| T (exp[−i −t dt′ HI (t′ )) |0i
P∞ (−i)m R ∞ 4 4
m=0 m! −∞ d y1 · · · d ym h0| T (ϕin (x1 ) · · · ϕin (xm )HI (y1 ) · · · HI (ym ) |0i
= P∞ (−i)n R +∞ 4 4
n=0 n! −∞ d y1 · · · d yn h0| T (HI (y1 ) · · · HI (yn )) |0i
(3.34)
This equation is the fundamental result. The Green functions have been expressed in
terms of the in fields whose algebra we know. It is therefore possible to reduce Eq. (3.34)
to known quantities. In this reduction plays an important role the Wick’s theorem, to
which we now turn.
3.3 Wick’s theorem

To evaluate the amplitudes that appear in Eq. (3.34) we have to move the annihilation
operators to the right until they act on the vacuum. The final result from these manipu-
lations can be stated in the form of a theorem, known as Wick’s theorem, which reads,
T (ϕin (x1 ) · · · ϕin (xn )) =
= : ϕin (x1 ) · · · ϕin (xn ) : +[h0| T (ϕin (x1 )ϕin (x2 )) |0i : ϕin (x3 ) · · · ϕin (xn ) : +perm.
= h0| T (ϕin (x1 )ϕin (x2 )) |0i h0| T (ϕin (x3 )ϕin (x4 )) |0i : ϕin (x5 ) · · · ϕin (xn ) : +perm.
= ···
3.3. WICK’S THEOREM 79


 h0| T (ϕin (x1 )ϕin (x2 )) |0i · · · h0| T (ϕin (xn−1 )ϕin (xn )) |0i + perm.



 n even
=



 h0| T (ϕin (x1 )ϕin (x2 ) |0i · · · h0| T (ϕin (xn−2 )ϕin (xn−1 )) |0i ϕin (xn ) + perm.


n odd
(3.35)
Proof:
The proof of the theorem is done by induction. For n = 1 is certainly true (and trivial).
Also for n = 2 we can shown that
T (ϕin (x1 )ϕin (x2 )) =: ϕin (x1 )ϕin (x2 ) : +c-number (3.36)
where the c-number comes from the commutations that are needed to move the annihila-
tion operators to the right. To find this constant, we do not have to do any calculation,
just to notice that
h0| : · · · : |0i = 0 (3.37)
Then
T (ϕin (x1 )ϕin (x2 ) =: ϕin (x1 )ϕin (x2 ) : + h0| T (ϕin (x1 )ϕin (x2 )) |0i (3.38)
what is in agreement with Eq. (3.35).
Continuing with the induction, let us assume that Eq. (3.35) is valid for a given n. We
have to show that it remains valid for n + 1. Let us consider then T (ϕin (x1 ) · · · ϕin (xn+1 ))
and let us assume that tn+1 is the earliest time. Then
T (ϕin (x1 ) · · · ϕin (xn+1 )) =
= T (ϕin (x1 ) · · · ϕin (xn ))ϕin (xn+1 )
= : ϕin (x1 ) · · · ϕin (xn ) : ϕin (xn+1 )

X
+ h0| T (ϕin (x1 )ϕin (x2 )) |0i : ϕin (x3 ) · · · ϕin (xn ) : ϕin (xn+1 )
perm
+··· (3.39)
To write Eq. (3.39) in the form of Eq. (3.35) it is necessary to find the rule showing how
to introduce ϕin (xn+1 ) inside the normal product. For that, we introduce the notation,
(+) (−)
ϕin (x) = ϕin (x) + ϕin (x) (3.40)
(+) (−)
where ϕin (x) contains the annihilation operator and ϕin (x) the creation operator. Then
we can write,
X Y (−) Y (+)
: ϕin (x1 ) · · · ϕin (xn ) := ϕin (xi ) ϕin (xj ) (3.41)
A,B i∈A j∈B
where the sum runs over all the sets A, B that constitute partitions of the n indices. Then
: ϕin (x1 ) · · · ϕin (xn ) : ϕin (xn+1 ) =

X Y (−) Y (+) (+) (−)
= ϕin (xi ) ϕin (xj )[ϕin (xn+1 ) + ϕin (xn+1 )]
A,B i∈A j∈B
XY (−)
Y (+) (+)
= ϕin (xi ) ϕin (xj )ϕin (xn+1 )
A,B iǫA j∈B
XY (−) (−)
Y (+)
+ ϕin (xi )ϕin (xn+1 ) ϕin (xj )
A,B i∈A j∈B
XY (−)
X Y (+) (+) (−)
+ ϕin (xi ) ϕin (xj ) h0| ϕin (xk )ϕin (xn+1 ) |0i (3.42)
A,B i∈A k∈B j∈Bj6=k
we can now write,

(+) (−)
h0| ϕin (xk )ϕin (xn+1 ) |0i = h0| ϕin (xk )ϕin (xn+1 ) |0i
= h0| T (ϕin (xk )ϕin (xn+1 )) |0i (3.43)
where we have used the fact that tn+1 is the earliest time. We can then write Eq. (3.42)
in the form,
: ϕin (x1 ) · · · ϕin (xn ) : ϕin (xn+1 ) =: ϕin (x1 ) · · · ϕin (xn+1 ) :
X
+ : ϕin (x1 ) · · · ϕin (xk−1 )ϕin (xk+1 ) · · · ϕin (xn ) : h0| T (ϕin (xk )ϕin (xn+1 )) |0i
k
(3.44)
With this result, Eq. (3.39) takes the general form of Eq. (3.35) for the n + 1 case,
ending the proof of the theorem. To fully understand the theorem, it is important to do
in detail the case n = 4, to see how things work. The importance of the Wick’s theorem
for the applications comes from the following two corollaries.
Corollary 1 : If n is odd, then h0| T (ϕin (x1 ) · · · ϕin (xn )) |0i = 0, as results trivially
from Eqs. (3.35) and (3.37) and from,
h0| ϕin (x) |0i = 0 (3.45)
Corollary 2: If n is even
h0| T (ϕin (x1 ) · · · ϕin (xn )) |0i =

X
= δp h0| T (ϕin (x1 )ϕin (x2 )) |0i · · · h0| T (ϕin (xn−1 )ϕin (xn )) |0i (3.46)
perm
where δp is the sign of the permutation that is necessary to introduce in case of fermion
fields. This result, that in practice is the most important one, also results from Eqs.
(3.35), (3.37) and (3.45).
3.3. WICK’S THEOREM 81
Therefore the vacuum expectation value of the time-ordered product of n operators

that appear in the general formula, Eq. (3.34), are obtained considering all the vacuum
expectation values of the fields taken two by two (contractions) in all possible ways. Now
these contractions are nothing else than the Feynman propagators for free fields. For
instance,
h0| T (ϕin (x1 )ϕin (x2 )ϕin (x3 )ϕin (x4 )) |0i
= h0| T (ϕin (x1 )ϕin (x2 )) |0i h0| T (ϕin (x3 )ϕin (x4 ) |0i
+ h0| T (ϕin (x1 )ϕin (x3 )) |0i h0| T (ϕin (x2 )ϕin (x4 )) |0i
+ h0| T (ϕin (x1 )ϕin (x4 )) |0i h0| T (ϕin (x2 )ϕin (x3 )) |0i
= ∆F (x1 − x2 )∆F (x3 − x4 ) + ∆F (x1 − x3 )∆F (x2 − x4 )
+∆F (x1 − x4 )∆F (x2 − x3 ) (3.47)

where Z
d4 k i
∆F (x − y) = e−ik(x−y) (3.48)
(2π) k − m2 + iǫ
4 2
is the Feynman propagator for the free field theory in the case of scalar fields.
It is convenient to use a graphical (diagrammatic) representation for these propagators.
We have in configuration space,
Z
d4 k i
∆F (x − y) = e−ik·(x−y) (3.49)
p (2π) k − m2 + iǫ
4 2
Z
β α d4 p i(p/ + m)αβ −ip·(x−y)
SF (x − y)αβ = e (3.50)
p (2π)4 p2 − m2 + iǫ
Z
µ d4 k −igµν −ik·(x−y)
ν DFµν (x − y) = e (3.51)
p (2π)4 k2 + iǫ
respectively for scalar, spinor and photon (in the Feynman gauge) fields.
As the interaction Hamiltonian is normal ordered, there will be no contractions between
the fields that appear in HI . The fields in HI can only contract with fields outside. In this
way the contractions will connect the points corresponding to HI , the so-called vertices
to either external points or points in another HI , corresponding to another vertex. To
illustrate this point let us consider the λϕ4 theory where,
1
HI (x) = λ : ϕ4in (x) : (3.52)
4!
Then a contribution of order λ2 to G(x1 , x2 , x3 , x4 ) comes from the term,
λ2
h0| T (ϕin (x1 )ϕin (x2 )ϕin (x3 )ϕin (x4 ) : ϕ4in (y1 ) :: ϕ4in (y2 ) : |0i (3.53)
(4!)2
x3 x4 x3 x4
y1 y2 y2 y1
x2
x1 x2 x1 x2
x3 x3 x4
y2
y1 y2
y1
x1 x2 x1 x2
Figure 3.1: Some of the diagrams resulting from Eq (3.53).
and leads to the diagrams in Fig. (3.1). In these diagrams, the interaction is represented
by four lines coming from one point, y1 or y2 . These lines are contractions between one
field from one HI with other field that might belong either to another HI , or be one
of the external fields in G(x1 · · · x4 ). To obtain the Feynman rules we are left with a
combinatorial problem. We are not going to find them, here as they are much easier to
express in momentum space, as we will see in the following.
In Fig. (3.1) the diagrams a), b) and d) are called connected while the diagram c) is
called disconnected. One diagram is disconnected when there is a part of the diagram that
is not connected in any way to an external line. We will see in the following that these
diagrams do not contribute to the Green functions. Diagram d) is connected but is also
called reducible because it can be obtained by multiplication of simpler Green functions.
As we will see only the irreducible diagrams are important.
3.4 Vacuum–Vacuum amplitudes

We have seen in the previous section examples of the numerator of Eq. (3.34). Let us
now look at the denominator, the so-called vacuum-vacuum amplitudes. Continuing with
the example of λϕ4 , some of the diagrams contributing for these amplitudes are shown in
Fig. (3.2). The diagrams associated with the numerator of Eq. (3.34) can be separated
into connected and disconnected parts. For all diagrams that have as connected part a
contribution of order s in the interaction HI , the numerator of G(x1 · · · xn ) takes the form,
∞
X Z
(−i)p
d4 y1 · · · d4 yp h0| T (ϕin (x1 ) · · · ϕin (xn )HI (y1 ) · · · HI (ys )) |0ic
p!
p=0
3.4. VACUUM–VACUUM AMPLITUDES 83
y3 y4 y3 y4
y5
y1 y2
y1 y2 y1 y2
a) b) c)
Figure 3.2: Some vacuum–vacuum amplitudes in Eq. (3.54).
p!
× h0| T (HI (ys+1 ) · · · HI (yp )) |0i (3.54)
s!(p − s)!
where the subscript c indicates that only the connected parts are included. The combina-
torial factor
 
p p!
 = (3.55)
s s!(p − s)!
is the number of ways in which we can extract s terms HI from a set of p terms. We write
then Eq. (3.54) in the form,
Z
(−i)s
d4 y1 · · · d4 ys h0| T (ϕin (x1 ) · · · ϕin (xn )HI (y1 ) · · · H(ys )) |0ic
s!
X∞ Z
(−i)r
× d4 z1 · · · d4 zr h0| T (HI (z1 ) · · · HI (zr )) |0i (3.56)
r!
r=0
This equation has the form of a connected diagram of order s times an infinite series
of vacuum-vacuum amplitudes, that cancels exactly against the denominator. This is true
for all orders, and therefore we can write,
P P P
i (x1 · · · xn )
i GP ( i Gci (x1 , · · · xn ))( k Dk )
G(x1 , · · · xn ) = = P
k Dk k Dk
X
= Gci (x1 · · · xn ) (3.57)
i
where Gci are the connected diagrams and Dk the disconnected ones. This result means
that we can simply ignore completely the disconnected diagrams and consider only the
connected ones when evaluating the Green functions. These are simply the sum of all
connected diagrams, simplifying enormously the structure of Eq. (3.34).
3.5 Feynman rules for λϕ4

To understand how the Feynman rules appear, let us consider the case of a real scalar
field with an interaction of the form,
λ
HI = : ϕ4in := −LI (3.58)
4!
To be more precise we consider two particles in the initial and final state. Then the S
matrix element is,

Sf i = p′1 p′2 ; out|p1 p2 ; in
Z
′ ′
= (i)4 d4 x1 d4 x2 d4 x3 d4 x4 e−ip1 ·x1 −ip2 ·x2 +ipi x3 +ip2 ·x4
⊓x1 + m2 ) · · · (⊔
(⊔ ⊓x4 + m2 ) h0| T (ϕ(x1 )ϕ(x2 )ϕ(x3 )ϕ(x4 )) |0i (3.59)
for the Green function we use the expressions in Eqs. (3.34) and (3.57) and we obtain,
∞
X Z
(−iλ)p
G(x1 , x2 , x3 , x4 ) = d4 z1 · · · d4 zp
p!
0
ϕ4in (z1 ) ϕ4 (zp )
h0| T (ϕin (x1 )ϕin (x2 )ϕin (x3 )ϕin (x4 ) : : · · · : in :) |0i (3.60)
4! 4!
As the case p = 0 is trivial (there is no interaction) we begin by the p = 1 case.
• p=1
Then the Green function is,
Z
4 ϕ4in (z)
G(x1 , x2 , x3 , x4 ) = (−iλ) d z h0| T ϕin (x1 )ϕin (x2 )ϕin (x3 )ϕin (x4 ) : : |0i
4!
Z
4!
= (−iλ) d4 z∆F (x1 − z)∆F (x2 − z)∆F (x3 − z)∆F (x4 − z)
4!
(3.61)
to which corresponds, in the configuration space, the diagram of Fig. (3.3). To proceed,
we introduce the Fourier transform of the propagators, that is,
Z 4
d q1 −iq1 ·(x1 −z)
∆F (x1 − z) = e ∆F (q1 ) (3.62)
(2π)4
where
i
∆F (q1 ) = (3.63)
q12 − m2
then
Z
d4 q1 d4 q4 −iq1 ·x1 −iq2 x2 −iq3 x3 −iq4 x4 +i(q1 +q2 +q3 +q4 )·z
G(x1 , · · · x4 ) = (−iλ) d4 z · · · e
(2π)4 (2π)4
3.5. FEYNMAN RULES FOR λϕ4 85
x3 x4
x1 x2
Figure 3.3: Vertex in the λφ4 theory.
p′1 p′2
−i λ
p1 p2
Figure 3.4: Vertex in momentum space.
∆F (q1 )∆F (q2 )∆F (q3 )∆F (q4 )

Z
d4 q1 d4 q4 −iq1 ·x1 −iq2 ·x2 −iq3 ·x3 −iq4 ·x4
= (−iλ) · · · e
(2π)4 (2π)4
(2π)4 δ4 (q1 + q2 + q3 + q4 )∆F (q1 )∆F (q2 )∆F (q3 )∆F (q4 )
(3.64)
If we now introduce the T matrix transition amplitude, defined by
Sf i = δf i − i(2π)4 δ(Pf − Pi ) Tf i (3.65)
we obtain
−iTf i = (−iλ) (3.66)
for this amplitude we draw the Feynman diagram of Fig. (3.4), and we associate to the
vertex the number (−iλ).
• p=2
Let us consider now a more complicated case, the evaluation of G(x1 · · · x4 ) in second
order in the coupling λ. After this exercise we will be in position to be able to state the
Feynman rules in momentum space with all generality. From Eq. (3.60) we get in second
order in λ,
G(x1 , · · · x4 ) =
Z
(−iλ)2 4 4 ϕ4in (z1 ) ϕ4in (z2 )
= d z1 d z2 h0| T ϕin (x1 )ϕin (x2 )ϕin (x3 )ϕin (x4 ) : :: : |0ic
2! 4! 4!
Z
(−iλ)2 4 4 4×3 4×3
= d z1 d z2 × ×2
2! 4! 4!

∆F (x1 − z1 )∆F (x2 − z1 )∆F (z1 − z2 )∆F (z1 − z2 )∆F (z2 − x3 )∆F (z2 − x4 )
+∆F (x1 − z1 )∆F (x2 − z2 )∆F (z1 − z2 )∆F (z1 − z2 )∆F (z1 − x3 )∆F (z2 − x4 )

+ ∆F (x1 − z2 )∆F (x2 − z1 )∆F (z1 − z2 )∆F (z1 − z2 )∆F (z1 − x4 )∆F (z1 − x3 )
Z
(−iλ)2
= d4 z1 d4 z2
2!
(
x3 x4 x3 x4 x3 x4
1
__ z1 z2 1 z
__ z2 1 z
__ z2
1 1
2 2 2
x1 x2 x1 x2 x1 x2
)
+ (z1 ↔ z2 ) (3.67)
Let us now go into momentum space, by introducing the Fourier transform of the
propagators. We start by diagram a),
G(a) (x1 , x2 , x3 , x4 ) =
Z
(−iλ)2 1
= d4 z1 d4 z2 ∆F (x1 − z1 )∆F (x2 − z1 )∆F (z1 − z2 )∆F (z1 − z2 )
2! 2
3.5. FEYNMAN RULES FOR λϕ4 87
∆F (z2 − x3 )∆F (z2 − x4 )

Z
(−iλ)2 1 d4 q1 d4 q2 d4 q3 d4 q4 d4 q5 d4 q6
= d4 z1 d4 z2
2! 2 (2π)4 (2π)4 (2π)4 (2π)4 (2π)4 (2π)4
ei[(q1 ·x1 +q2 ·x2 −q3 ·x3 −q4 ·x4 )+z1 ·(q5 −q1 −q2 +q6 )+z2 ·(q3 +q4 −q5 −q6 )]
∆F (q1 )∆F (q2 )∆F (q3 )∆F (q4 )∆F (q5 )∆F (q6 )
Z
(−iλ)2 1 d4 q1 d4 q5 4
= (2π)4 · · · δ (q1 + q2 − q3 − q4 ) ei[q1 ·x1 +q2 ·x2 −q2 ·x3−q4 ·x4 ]
2! 2 (2π)4 (2π)4
∆F (q1 )∆F (q2 )∆F (q3 )∆F (q4 )∆F (q5 )∆F (q1 + q2 − q5 ) (3.68)
Now we insert the last equation into the reduction formula. We get
Z
(a) 4 ′ ′
Sf i = (i) d4 x1 · · · d4 x4 e−i[p1 ·x1 +p2 ·x2 −p1 ·x3−p2 ·x4 ]
⊓x1 + m2 ) · · · (⊔
(⊔ ⊓x4 + m2 )G(a) (x1 , · · · , x4 ) (3.69)
The only dependence of G(a) in the coordinates, xi (i = 1, · · · 4), is in the exponential,
therefore,
⊓xi + m2 ) → (−qi2 + m2 )
(⊔ (3.70)
and using
(−qi2 + m2 )∆F (qi ) = −i (3.71)
we get
Z Z
(a) (−iλ)2 1 d4 q1 d4 q5
Sf i = d4 x1 · · · d4 x4 4
··· (2π)4 δ4 (q1 + q2 − q3 − q4 )
2! 2 (2π) (2π)4
′ ′
e−i[x1 ·(p1 −q1 )+x2 ·(p2 −q2 )−x3 ·(p1 −q3 )−x4 ·(p2 −q4)] ∆F (q5 )∆F (q1 + q2 − q5 )
Z 4
(−iλ)2 1 d q1 d4 q5
= · · · (2π)4 δ4 (q1 + q2 − q3 − q4 )(2π)4 δ4 (p1 − q1 )
2! 2 (2π)4 (2π)4
(2π)4 δ4 (p2 − q2 )(2π)4 δ4 (p′1 − q3 )(2π)4 δ4 (p′2 − q4 )∆F (q5 )∆F (q1 + q2 − q5 )
Z 4
(−iλ)2 1 d q5
= (2π)4 δ4 (p1 + p2 − p′1 − p′2 )∆F (q5 )∆F (p1 + p2 − q5 ) (3.72)
2! 2 (2π)4
This expression can be written in the form
Z
(a) 4 4 (−iλ)2 1 d4 q
Sf i = (2π) δ (p1 + p2 − p′1 − p′2 ) ∆F (q)∆F (p1 + p2 − q) (3.73)
2! 2 (2π)4
If we denote by a′ ) the diagram a) with the interchange z1 ↔ z2 and redo the calculation
we get exactly the same result as in Eq. (3.73). Therefore,
Z
(a+a′ ) 4 4 ′ ′ 21 d4 q
Sf i = (2π) δ (p1 + p2 − p1 − p2 )(−iλ) ∆F (q)∆F (p1 + p2 − q) (3.74)
2 (2π)4
p′1 p′2
1 q q − p1 − p2
2
p1 p2
Figure 3.5:
p′1 p′2
p′1 q−p1 +p′1 p′2
1 1 q−p1 +p′2
2 2
p1 q p2 p1 q p2
Figure 3.6:
or in terms of the Tf i matrix,

Z
(a+a′ ) 21 d4 q
−iTf i = (−iλ) ∆F (q)∆F (p1 + p2 − q) (3.75)
2 (2π)4
To encode this result we draw the Feynman diagram of Fig. (3.5), that has the same
topology as a) and a′ ) but in momentum space. We find that in order to evaluate the −iT
matrix, we associate to each vertex a factor (−iλ), to each internal line a propagator ∆F
R d4 q
and for each loop the integral (2π) 4 . Besides that we have 4-momentum conservation at
each vertex. Finally there is a symmetry factor (see below) which takes the value 21 for
this diagram.
If we repeat the calculations for diagrams b) + b′ ) and c) + c′ ) it is easy to see that we
get,
Z
(b+b′ ) 21 d4 q
−iTf i = (−iλ) ∆F (q)∆F (q − p1 + p′1 ) (3.76)
2 (2π)4
and Z
(c+c′ ) 21 d4 q
−iTf i = (−iλ) ∆F (q)∆F (q − p1 + p′2 ) (3.77)
2 (2π)4
to which correspond the diagrams of Fig. (3.6).
3.6. FEYNMAN RULES FOR QED 89
After this exercise we are in position to state the Feynman rules with all generality for
the λϕ4 theory. These are rules for the −iT matrix, that is, after we factorize (2π)4 δ4 (· · · ).
These are (for a process with n external legs):
1. Draw all topologically distinct diagrams with n external legs.
2. At each vertex multiply by the factor (−iλ).
3. To each internal line associate a propagator ∆F (q).
R d4 q
4. For each loop include an integral (2π) 4 . The direction of this momentum is irrele-
vant, but we have to respect 4-momentum conservation ate each vertex.

5. Multiply by the symmetry factor of the diagram. This is defined by,
# of distinct ways of connecting the vertices to the external legs

S= (3.78)
Permutations of each vertex × Permutations of equal vertices
6. Add the contributions of all the topologically distinct diagrams. The result is the
−iT matrix amplitude that enters the formula for the cross section.
3.6 Feynman rules for QED

We now turn to the case of QED. Like λφ4 , it is a theory without derivatives and therefore,
LI = −HI = −e Qψ in γ µ ψin Ain
µ (3.79)
where e is the absolute value of the electron charge, or the proton charge. For the electron
the sign enters explicitly in Q = −1. This way of writing in Eq. (3.79), allows for obvious
generalizations for particles with other charges, like for instance the quarks. For QED we
have then,
LQED
I = e ψ in γ µ ψin Ain
µ (3.80)
Due to the electric charge conservation, the Green functions that we have to deal with
have an equal number of ψ and ψ fields. In general we have,
G(x1 · · · xn xn+1 · · · x2n ; y1 · · · yp ) =
= h0| T (ψ(x1 ) · · · ψ(xn )ψ(xn+1 ) · · · ψ(x2n )Aµ1 (y1 ) · · · Aµp (yp )) |0i (3.81)
where, for simplicity, we omit the spinorial indices in the fermion fields. This equation is
written in terms of the physical fields. Following a similar procedure to the scalar field
case, we can obtain an expression for G in terms of the in fields. This will be,
µ
h0| T ψin (x1 ) · · · ψ in (x2n )Aµin1 (y1 ) · · · Ainp (yp ) e[i d4 zLI (z)] |0i
R
G(x1 · · · x2n ; y1 · · · yp ) = R
h0| T exp[i d4 zLI (z)] |0i
µ d4 tLI (z)]
R
= h0| T ψin (x1 ) · · · ψ in (x2n )Aµin1 (y1 ) · · · Ainp (yp ) e[i |0ic
(3.82)
where the fields in LI are normal ordered, and h0| · · · |0ic means that we only consider the
connected diagrams. To get the Feynman rule we will evaluate a few simple processes.
γ γ
k k′
p p′
e− e−
Figure 3.7: Kinematics for the Compton scattering.
3.6.1 Compton scattering

Compton scattering corresponds to the following process,
e− + γ → e− + γ (3.83)
and we choose the kinematics in Fig. (3.7). The S matrix element to evaluate is therefore,

Sf i = (p′ , s′ ), k′ ; out|(p, s), k; in (3.84)
Using Eq. (2.102) and Eq. (2.110) we can write,

Z Z
4 4 ′ ′ ′ ′ ′ ′
Sf i = d xd x d4 yd4 y ′ e−i[p·x+k·y−p ·x −k ·y ] εµ (k)ε∗µ (k′ )
→ ←
u(p′ , s′ )α′ (i∂/x′ −m)α′ β ′ ⊔ ~ y′ h0| T (ψβ ′ (x′ )ψ β (x)Aµ (y)Aµ′ (y ′ )) |0i (−i∂/x −m)βα uα (p, s)
~ y⊔
⊓ ⊓
(3.85)
Our task is therefore to evaluate the Green function
G(x′ , x, y, y ′ ) ≡ h0| T (ψβ ′ (x′ )ψ β (x)Aµ (y)Aµ′ (y ′ )) |0i (3.86)
If we use Eq. (3.82) and the fact that the interaction has an odd number of fields, we find
that the lowest contribution is quadratic in the interaction2 . We get
G(x, x′ , y, y ′ ) =
Z
(ie)2 in
= d4 z1 d4 z2 h0| T (ψβin′ (x′ )ψ β (x)Ain in ′
µ (y)Aµ′ (y )
2!
in
: ψ in (z1 )γ σ ψin (z1 )Ain ρ in in
σ (z1 ) :: ψ (z2 )γ ψ (z2 )Aρ (z2 ) :) |0i
Z
(ie)2 σ in
= (γ )γδ (γ ρ )γ ′ δ′ d4 z1 d4 z2 h0| T (ψβin′ (x′ )ψ β (x)Ain in ′
µ (y)Aµ′ (y )
2!
2
By Wick’s theorem the expectation value of an odd number of fields vanishes.
in in
: ψ γ (z1 )ψδin (z1 )Ain in in
σ (z1 ) :: ψ γ ′ (z2 )ψδ′ (z2 )Aρ (z2 )) |0i (3.87)
Now we use Wick’s theorem to write h0| T (· · · ) |0i in terms of the propagators, We get,
in in in
h0| T (ψβin′ (x′ )ψ β (x)Ain in ′ in in in in
µ (y)Aµ (y ) : ψ γ (z1 )ψδ (z1 )Aσ (z1 ) :: ψ γ ′ (z2 )ψδ′ (z2 )Aρ (z2 )) :) |0i
in in
= h0| T ψβin′ (x′ )ψ γ (z1 ) |0i h0| T ψδin′ (z2 )ψ β (x) |0i h0| T ψδin (z1 )ψ γ ′ (z2 ) |0i
h0| T (Ain in in ′ in
µ (y)Aσ (z1 )) |0i h0| T Aµ′ (y )Aρ (z2 ) |0i
in in
+ h0| T ψβin′ (x′ )ψ γ (z1 ) |0i h0| T ψδin′ (z2 )ψ β (x) |0i h0| T ψδin (z1 )ψ γ ′ (z2 )|0
h0| T Ain in in ′ in
µ (y)Aρ (z2 ) |0i h0| T Aµ′ (y )Aσ (z1 ) |0i
in in in
+ h0| T ψβin′ (x′ )ψ γ ′ (z2 ) |0i h0| T ψδin (z1 )ψ β (x) |0i h0| T ψδin′ (z2 )ψ γ (z1 ) |0i
µ (y)Aσ (z1 ) |0i h0| T Aµ′ (y )Aρ (z2 ) |0i
in in in
+ h0| T ψβin′ (x′ )ψ γ ′ (z2 ) |0i h0| T ψδin (z1 )ψ β (x) |0i h0| T ψδin′ (z2 )ψ γ (z1 ) |0i
µ (y)Aρ (z2 ) |0i h0| T Aµ′ (y )Aσ (z1 ) |0i
= SF β ′ γ (x′ − z1 )SF δ′ β (z2 − x)SF δγ ′ (z1 − z2 )DF µσ (y − z1 )DF µ′ ρ (y ′ − z2 )
+SF β ′ γ (x′ − z1 )SF δ′ β (z2 − x)SF δγ ′ (z1 − z2 )DF µρ (y − z2 )DF µ′ σ (y ′ − z1 )
+SF β ′ γ ′ (x′ − z2 )SF δβ (z1 − x)SF δ′ γ (z2 − z1 )DF µσ (y − z1 )DF µ′ ρ (y ′ − z2 )
+SF β ′ γ ′ (x′ − z2 )SF δβ (z1 − x)SF δ′ γ (z2 − z1 )DF µρ (y − z2 )DF µσ (y ′ − z1 ) (3.88)
To better understand Eq. (3.88) it is useful to draw the corresponding diagrams in

configuration space. We show them in Fig. (3.8). From this figure it is clear that b) ≡ c)
and a) ≡ d) because z1 and z2 are irrelevant labels. From this we get a factor of 2 that
1
is going to cancel the 2! in Eq. (3.87)3 . We have then only two distinct diagrams that we
take as c) and d). Then, including already the factor of 2, we get
Z
G(2) (x, x′ , y, y ′ ) = (ie)2 (γ σ )γδ (γ ρ )γ ′ δ′ d4 z1 d4 z2 SF β ′ γ ′ (x′ − z2 )SF δβ (z1 − x)
SF δ′ γ (z2 − z1 )DF µσ (y − z1 )DF µ′ ρ (y ′ − z2 ) (3.89)
To proceed we could, like in the case of λϕ4 , introduce the Fourier transform of the
propagators. However, it is easier to get rid of the external legs using the results,
(i∂/x − m)αλ SF λβ (x − y) = iδαβ δ4 (x − y)

←
SF αλ (x − y)(−i∂/y − m)λβ = iδαβ δ4 (x − y) (3.90)
3 1
In fact this result is general, for n vertices we have n! that cancels against the factor n!
from the
expansion of the exponential.
y y′ y y′
x z2 z1 x′ x z2 z1 x′
a) b)
y y′ y y′
x z1 z2 x′ x z1 z2 x′
c) d)
Figure 3.8: Diagrams for Compton scattering in configuration space.
and
⊓x DF µν (x − y) = igµν δ4 (x − y)
⊔ (3.91)
We get therefore,
Z
(c) ′ ′ ′ ′ ′
Sf i = (ie)2 d4 xd4 x′ d4 yd4 y ′ e−i(p·x+k·y−p ·x −k ·y ) εµ (k)gµσ ε∗µ (k′ )gµ′ ρ
(γ σ )γδ (γ ρ )γ ′ δ′ u(p′ , s′ )α′ δα′ γ ′ uα (p, s) δδα

Z
d4 z1 d4 z2 δ4 (x′ − z2 )δ4 (x − z1 )δ4 (y − z1 )δ4 (y ′ − z2 ) SF δ′ γ (z2 − z1 )
Z
2 ′ ′ ′
= (ie) d4 z1 d4 z2 e−i(p·z1 +k·z1 −p ·z2 −k ·z2 ) εµ (k)ε∗µ (k′ )
u(p′ , s′ )α′ (γµ′ )α′ δ′ SF δ′ γ (z2 − z1 )(γµ )γα uα (p, s) (3.92)
Finally we use
Z
d4 q i(q/ + m) −iq·(z2 −z1 )
SF (z2 − z1 ) = e
(2π) q − m2 + iǫ
4 2
Z
d4 q
≡ SF (q)e−iq·(z2 −z1 ) (3.93)
(2π)4
to get
Z
(c) d4 q 4 4 −iz1 ·(p+k−q)+iz2 ·(p′ +k′ −q)
Sf i = d z1 d z2 e
(2π)4
′
εµ (k)εµ ∗ (k′ )u(p′ , s′ )(ieγµ′ )SF (q)(ieγµ )u(p, s)
µ ν µ ν
k k′ k k′
p a) p′ p b) p′
Figure 3.9: Diagrams for Compton scattering.
= (2π)4 δ(4) (p + k − p′ − k)·

′
εµ (k)εµ ∗ (k′ )u(p′ , s′ )(ieγµ′ )SF (p + k)(ieγµ )u(p, s) (3.94)
Therefore, the T matrix transition amplitude is,

(c) ′
−iTf i = εµ (k)εµ ∗ (k′ )u(p′ , s′ )(ieγµ′ )SF (p + k)(ieγµ )u(p, s) (3.95)
corresponding to the diagram on the left panel of Fig. (3.9). In Eq. (3.95) we factor out
the quantity (ieγµ ), because it will be clear that this quantity will be the Feynman rule for
the vertex. The arrows in these diagrams correspond to the flow of electric charge. Notice
that to an electron in the initial state we associate a spinor u(p, s) and for an electron in
the final state we associate the spinor u(p′ , s′ ). As the result of the electron line as to be
a c-number, we start writing the line in the reverse order of that of the arrows.
In a similar way for diagram d) we will get the diagram represented in Fig. (3.9), that
corresponds to the following expression,
(d) ′
−iTf i = εµ (k)ε∗µ (k′ )u(p′ , s′ )(ieγµ )SF (p − k′ )(ieγµ′ )u(p, s) (3.96)
Looking at Eqs. (3.95) and (3.96) we are almost in a position to state the Feynman rules
for QED. Before that we will look at a case where we have positrons.
3.6.2 Electron–positron elastic scattering (Bhabha scattering)

We will consider electron-positron elastic scattering, the so-called Bhabha scattering,
e− (p) + e+ (q) → e− (p′ ) + e+ (q ′ ) (3.97)
This example will teach us two things. First, how do positrons (that is the anti-particles)
enter in the amplitudes. Secondly we will learn that, sometimes, due the anti-commutation
rules of the fermions, we will get relative minus signs between different diagrams. We have,

Sf i = (p′ , s′ ), (q ′ , s′ ); out|(p, s), (q, s); in (3.98)
corresponding to the kinematics in Fig. (3.10). Notice that the arrows are in the direction
of flow of charge of the electron, but the momenta do correspond to the real momenta
of the particles or antiparticles in that frame: p entering and p′ exiting for the electron,
p p′
q q′
Figure 3.10: Diagram with positrons
and q entering and q ′ exiting for the positron. In the following we will not show the spin
dependence in order to simplify the notation. Then using Eq. (3.98) we write,
Z
′ ′ ′ ′
Sf i = d4 xd4 yd4 x′ d4 y ′ e−i[p·x+q·y−p ·x −q ·y ]
→ →
u(p′ )α (i∂/x′ − m)αβ v γ (q)(i∂/y − m)γδ
h0| T ψ δ′ (y ′ )ψβ (x′ )ψ β ′ (x)ψδ (y) |0i

← ←
(−i∂/x − m)β ′ α′ uα′ (p)(−i∂/y′ − m)δ′ γ ′ vγ ′ (q ′ ) (3.99)
We have, therefore, to evaluate the Green function
G(y ′ , x′ , x, y) ≡ h0| T ψ δ′ (y ′ )ψβ (x′ )ψ β ′ (x)ψδ (y) |0i (3.100)
The lowest order contribution is of second order in the coupling4 . We have (to simplify
we omit the label in),
Z
(ie)2 µ
G(y ′ , x′ , x, y) = (γ )ǫǫ′ (γ ν )ϕϕ′ d4 z1 d4 z2
2
h0| T ψ δ′ (y ′ )ψβ (x′ )ψ β ′ (x)ψδ (y) : ψ ǫ (z1 )ψǫ′ (z1 )Aµ (z1 ) :: ψ ϕ (z2 )ψϕ′ (z2 )Aν (z2 ) : |0i
Z
(ie)2 µ
= ν
(γ )εε′ (γ )ϕϕ′ d4 z1 d4 z2
2
h
− SF βǫ (x′ − z1 )SF ǫ′ β ′ (z1 − x)SF δϕ (y − z2 )SF ϕ′ δ′ (z2 − y ′ )DF µν (z1 − z2 )
+SF δǫ (y − z1 )SF ǫ′ β ′ (z1 − x)SF βϕ (x′ − z2 )SF ϕ′ δ′ (z2 − y ′ )DF µν (z1 − z2 )

i
+(z1 ↔ z2 ) (3.101)
Once more the exchange (z1 ↔ z2 ) compensates for the factor 2!1 and we have two diagrams
with a relative minus sign, as it is shown in Fig. (3.11). Let us look at the contribution of
diagram a),
Z
(a) ′ ′ ′ ′
Sf i = − d4 xd4 yd4 x′ d4 y ′ d4 z1 d4 z2 (ie)2 (γ µ )εε′ (γ ν )ϕϕ′ e−i[p·x+q·y−p ·x −q ·y ]
4
There is, of course, a contribution without interaction, but that corresponds to disconnected terms in
which we are not interested.
x z1 x’ x x’
__ + z1 z2
z2
y y’ y y’
a) b)
Figure 3.11:
→ →
u(p′ )α (i∂/′x − m)αβ v γ (q)(i∂/y − m)γδ
SF βε (x′ − z1 )SF ε′ β ′ (z1 − x)SF δϕ (y − z2 )SF ϕ′ δ′ (z2 − y ′ )

← ←
(−i∂/x − m)β ′ α′ uα′ (p)(−i∂/′y − m)δ′ γ ′ vγ ′ (q ′ )DF µν (z1 − z2 )
Z
′ ′
= − d4 z1 d4 z2 e−i[p·z1+q·z2 −p ·z1 −q ·z2 ]
u(p′ )(ieγ µ )u(p)v(q)(ieγ ν )v(q ′ )DF µν (z1 − z2 ) (3.102)

Using now the Fourier transform of the photon propagator,
Z
d4 k −igµν −ik·(z1 −z2 )
DF µν (z1 − z2 ) = e
(2π)4 k2 + iε
Z
d4 k
≡ DF µν (k)e−ik·(z1 −z2 ) (3.103)
(2π)u
we get
(a)
Sf i = −u(p′ )(ieγ µ )u(ρ)v(q)(ieγ ν )v(q ′ )
Z
d4 k ′ ′
d4 z1 d4 z2 DF µν (k)e−iz1 ·(p−p +k) e−iz2 ·(q−q −k)
(2π)4
= −(2π)4 δ4 (p + q − p′ − q ′ )u(p′ )(ieγ ν )u(p)v(q)(ieγ µ )v(q ′ )DF µν (p′ − p)
(3.104)
and therefore the T matrix element is,
(a)
−iTf i = −u(p′ )(ieγ µ )u(p)DF µν (p′ − p)v(q)(ieγ ν )v(q ′ ) (3.105)
to which corresponds the Feynman diagram of Fig. (3.12).
In a similar way we would get
(b)
−iTf i = v(q)(ieγ µ )u(p)DF µν (p + q)u(p′ )(ieγ ν )v(q ′ ) (3.106)
that corresponds to the diagram of Fig. (3.13). Which of the diagrams has the minus
sign is irrelevant. It depends on the conventions how to build the in in state that lead to
Eq. (3.98). Only the relative sign is important.
µ
p p’
p’-p
q ν q’
Figure 3.12:
p p’
p+q
µ ν
q q’
Figure 3.13:
3.6.3 Fermion Loops

Before we summarize the Feynman rules for QED let us look at what happens with fermion
loops. One such example is the second order correction to the photon propagator shown
in Fig. (3.14). First of all, the loop orientation it is only relevant if it leads to topological
Figure 3.14: Vacuum polarization
different diagrams. Therefore the diagrams of Fig. (3.15) are topologically equivalent and
only one should be considered. However the diagrams in Fig. (3.16) are topologically
distinct and both should be considered.
The second aspect that is relevant is a possible sign coming from the anti-commutation
of the fermion fields, that should affect some diagrams, and in particular the fermion loop.
To understand this sign we should note that by definition of loop, the internal lines are
not connected to external fermion lines, they should originate only in the interaction.
Figure 3.15: Topologically equivalent diagrams
1 4 1 4
2 3 2 3
Figure 3.16: Topologically distinct diagrams
Therefore they should come from terms of the form
h0| T · · · : ψ(z1 )A/(z1 )ψ(z1 ) : · · · : ψ(zn )A/(zn )ψ(zn ) : · · · |0i . (3.107)
Now it is clear that in order to make the appropriate contractions of the fermion fields
to bring them to the form of the Feynman propagator, h0| T ψ(z1 )ψ(z2 ) |0i, it is necessary
to make an odd number of permutations of the fermion fields, and therefore we get a (−)
sign for the loops.
3.6.4 Feynman rules for QED

We are now in position to state the Feynman rules for QED
1. For a given process, draw all topologically distinct diagrams.
2. For each electron entering a diagram a factor u(p, s). If it leaves the diagram a factor
u(p, s).
3. For each positron leaving the diagram (final state) a factor v(p, s). It it enters the
diagram (initial state) then we have a factor v(p, s).
4. For each photon in the initial state we have the vector εµ (k) and in the final state
ε∗µ (k).
5. For each internal fermionic line the propagator

(p/ + m)αβ
β p α SFαβ (p) = i (3.108)
p2 − m2 + iε
6. For each virtual photon the propagator (Feynman gauge)
gµν
µ ν DF µν (k) = −i (3.109)
k k2
7. For each vertex the factor
µ
ieγ µ (3.110)
8. For each internal momentum, not fixed by conservation of momenta, as in the case
of loops, a factor
Z
d4 q
(3.111)
(2π)4
9. For each loop of fermions a −1 sign.
10. A factor of −1 between diagrams that differ by exchange of fermionic lines. In doubt,
revert to Wick’s theorem.
Comments
• In QED there are no symmetry factors, that is, they are always equal to 1.
• In our discussion we did not consider the Z factors that come in the reduction
formulas, like in Eq. (2.66). This is true in lowest order in perturbation theory.
They can be calculated also in perturbation theory. Their definition is (for instance
for the electron),
lim SF′ (p) = Z2 SF (p) (3.112)

p
/→m
where SF′ (p) is the propagator of the theory with interactions. Then we can obtain,
in perturbation theory,
Z2 = 1 + O(α) + · · · (3.113)
√
In higher orders it is necessary to correct the external lines with these Z factors.
3.7. GENERAL FORMALISM FOR GETTING THE FEYNMAN RULES 99
3.7 General formalism for getting the Feynman rules

After showing how to obtain the Feynman rules for λφ4 and QED, we are going to present
here, without proof, a general method to obtain the Feynman rules of any theory, including
the case when the interactions have derivatives, that we have excluded up to now, and
that is very important for the Standard Model. This method can only be fully justified
with the methods of Chapter 5. For simplicity we will consider only scalar fields.
The starting point is the action taken as a functional of the fields,
Z
Γ0 [ϕ] ≡ d4 xL[ϕ]· (3.114)
In fact, Γ0 [ϕ] is the generating functional of the one particle irreducible Green functions
in lowest order, as we will see in Chapter 5. The rules are as follows:
Propagators
(2) δ2 Γ0 [ϕ]
1. Start by evaluating Γ0 (xi , xj ) ≡
δϕ(xi )δϕ(xj )
(2)
2. Then evaluate the Fourier Transform (FT) to get Γ0 (pi , pj ) defined by the relation
Z
(2) (2)
(2π)4 δ4 (pi + pj )Γ0 (pi , pj ) ≡ d4 xi d4 xj e−i(pi ·xi +pj ·xj ) Γ0 (xi , xj ) (3.115)
where all the momenta are incoming.
3. The Feynman propagator is then
(0) (2)
GF ij = i[Γ0 (pi , pj )]−1 . (3.116)
Do not forget that pi = −pj .
Vertices
(n) δn Γ0 [ϕ]
1. Evaluate Γ0 (x1 · · · xn ) =
δϕ(x1 ) · · · δϕ(xn )
2. Then take the Fourier Transform to obtain
(n)
(2π)4 δ4 (p1 + p2 + · · · + pn )Γ0 (p1 · · · pn )
Z
(n)
≡ d4 x1 · · · d4 xn e−i(p1 ·x1 +···pn ·xn ) Γ0 (x1 · · · xn ) (3.117)
3. The vertex in momenta space is then given by the rule
(n)
iΓ0 (p1 , · · · pn ) (3.118)
Comments
• For fermionic fields it is necessary to take care with the order of the derivation. The
convention that we take is
δ2
ψ(z)Γψ(z) ≡ Γβα δ4 (z − x)δ4 (z − y) (3.119)
δψα (x)δψ β (y)
ψα (x) e ψβ (x) are here taken as classical anti-commuting fields (Grassmann variables,
see Chapter 5).
• The functional derivatives are defined by
δϕi (x)
≡ δik δ4 (x − y) (3.120)
δϕk (y)
3.7.1 Example: scalar electrodynamics

The Lagrangian is
λ ∗ 2
L = (∂µ − ieQAµ )ϕ∗ (∂ µ + ieQAµ )ϕ − mϕ∗ ϕ + LMaxwell − (ϕ ϕ) (3.121)
4
Therefore ↔
Lint = −ieQϕ∗ ∂ µ ϕAµ + e2 Q2 ϕ∗ ϕAµ Aµ (3.122)
The propagators are the usual ones, let us consider only the vertices. There are two
vertices. The cubic vertex is
ϕ∗
p
k
µ
q
ϕ
Figure 3.17: Cubic vertex in scalar QED.
Z
→ ←
Γ(3)
µ (x1 , x2 , x3 ) = −ieQ d4 zδ4 (z − x1 )(∂ µ − ∂ µ )δ4 (z − x2 )δ4 (z − x3 ) (3.123)
therefore
Z
4 4
(2π) δ (p + k + q)Γ(3)
µ (p, q, k) ≡ −ieQ d4 zd4 x1 d4 x2 d4 x3 e−i(x1 ·p+x2 ·q+x3·k)
→ ←
δ4 (z − x1 ) (∂ µ − ∂ µ )δ4 (z − x2 )δ4 (z − x3 )
3.7. GENERAL FORMALISM FOR GETTING THE FEYNMAN RULES 101
Z
= −ieQ d4 zd4 x2 e−i[(p+k)·z+q·x2] ∂µ δ4 (z − x2 )
Z
+ieQ d4 zd4 x1 e−i[p·x1+(q+k)·z] ∂µ δ4 (z − x1 )
= −ieQ(ipµ − iqµ )(2π)4 δ4 (p + q + k) (3.124)
Therefore we obtain for this vertex
iΓµ (p, q, k) = ieQ (pµ − qµ ) = −ieQ (qµ − pµ ) (3.125)
The other vertex is
µ ν
k1 k2
p q
Figure 3.18: Quartic vertex in scalar QED (seagull).
We obtain,
Γ(4) 2 2 4 4 4
µν (x1 , x2 , x3 , x4 ) = 2e Q δ (x1 − x2 )δ (x1 − x3 )δ (x1 − x4 )gµν (3.126)
and
Γ(4) 2
µν (p, q, k1 , k2 ) = 2(eQ) gµν (3.127)
and we finally get for the Feynman rule.
i2e2 Q2 gµν (3.128)
Comment
• From the above results we can enunciate a simple rule for interactions that have
derivatives of fields.
Consider that we have one field in the Lagrangian that has a derivative,
say ∂µ φ. Then the rule is
∂µ φ → −i (incoming momentum)µ (3.129)
In the end do not forget to multiply the result by i.

• As an example consider the following term in the Lagrangian for scalar electrody-
namics
L = ieQ∂µ ϕ∗ ϕAµ + · · · (3.130)
If p is the incoming momentum of the line associated with the field ϕ∗ , see Fig. 3.17,
we have
Vertex = i × (ieQ) × (−ipµ ) = i e Q pµ (3.131)
in agreement with Eq. (3.125).
Problems 103
3.1 Show explicitly that Wick’s theorem is valid for the case of 4 fields, that is
T (ϕin (x1 )ϕin (x2 )ϕin (x3 )ϕin (x4 )) =: ϕin (x1 ) · · · ϕin (x4 ) : + · · · (3.132)
3.2 For the case of the λϕ4 theory verify the Feynman rules for the diagrams
p′1 p′2
p′1 q−p1 +p′1 p′2
1 1 q−p1 +p′2
2 2
p1 q p2 p1 q p2
3.3 Consider a theory with the following interaction Lagrangian
λ
LI = − ϕ3in (3.133)
3!
• a) Find the Feynman rules for this theory.
• b) Find the symmetry factor for the diagram
3.4 Verify that for Compton scattering the diagram

µ ν
k k′
p b p′
gives the result of Eq. (3.96).

3.5 Verify Eq. (3.106).
3.6 Show that in QED the symmetry factors are always 1.
3.7 Explicitly calculate the T matrix element for the process e+ e− → γγ and verify that
is in agreement with the general rules.
3.8 Show that the amplitudes for e+ e− → γγ and e− γ → e− γ are related. How can one
obtain one from the other?
Chapter 4
Radiative Corrections
4.1 QED Renormalization at one-loop

We will consider the theory described by the Lagrangian
1 1
LQED = − Fµν F µν − (∂ · A)2 + ψ(i∂/ + eA
/ − m)ψ . (4.1)
4 2ξ
The free propagators are

i
β α ≡ SF0 βα (p) (4.2)
p p/ − m + iε βα

gµν kµ kν
µ ν −i 2 + (1 − ξ) 2
k k + iε (k + iε)2

kµ kν 1 kµ kν
= −i gµν − 2 +ξ 4
k k2 + iε k
≡ G0F µν (k) (4.3)
and the vertex
e−
γ
+ie(γµ )βα e = |e| > 0 (4.4)
e−
105
106 CHAPTER 4. RADIATIVE CORRECTIONS
We will now consider the one-loop corrections to the propagators and to the vertex. We
will work in the Feynman gauge (ξ = 1).
4.1.1 Vacuum Polarization

In first order the contribution to the photon propagator is given by the diagram of Fig. 4.1
that we write in the form
p
k k
p+k
Figure 4.1: Vacuum polarization.
′ ′
G(1) 0
µν (k) ≡ Gµµ′ i Π
µν
(k)G0ν ′ ν (k) (4.5)
where
Z
2 d4 p Tr[γµ (p/ + m)γν (p/ + k/ + m)]
i Πµν = −(+ie) (4.6)
(2π)4 (p2 − m2 + iε)((p + k)2 − m2 + iε)
Z
2 d4 p [2pµ pν + pµ kν + pν kµ − gµν (p2 + p · k − m2 )
= −4e
(2π)4 (p2 − m2 + iε)((p + k)2 − m2 + iε)
Simple power counting indicates that this integral is quadratically divergent for large
values of the internal loop momenta. In fact the divergence is milder, only logarithmic. The
integral being divergent we have first to regularize it and then to define a renormalization
procedure to cancel the infinities. For this purpose we will use the method of dimensional
regularization. For a value of d small enough the integral converges. If we define ǫ = 4 − d,
in the end we will have a divergent result in the limit ǫ → 0. We get therefore1
Z
2 ǫ dd p [2pµ pν + pµ kν + pν kµ − gµν (p2 + p · k − m2 )]
i Πµν (k, ǫ) = −4e µ
(2π)d (p2 − m2 + iε)((p + k)2 − m2 + iε)
Z
dd p Nµν (p, k)
= −4e2 µǫ (4.7)
(2π) (p − m + iε)((p + k)2 − m2 + iε)
d 2 2
where
Nµν (p, k) = 2pµ pν + pµ kν + pν kµ − gµν (p2 + p · k − m2 ) (4.8)
To evaluate this integral we first use the Feynman parameterization to rewrite the denom-
inator as a single term. For that we use (see Appendix)
Z 1
1 dx
= 2 (4.9)
ab 0 [ax + b(1 − x)]
1
Where µ is a parameter with dimensions of a mass that is introduced to ensure the correct dimensions
ǫ
of the coupling constant in dimension d, that is, [e] = 4−d
2
= 2ǫ . We take then e → eµ 2 . For more details
see the Appendix.
4.1. QED RENORMALIZATION AT ONE-LOOP 107
to get
Z 1 Z
dd p Nµν (p, k)
i Πµν (k, ǫ) = −4e2 µǫ dx
0 (2π) [x(p + k)2 − xm2 + (1 − x)(p2 − m2 ) + iε]2
d
Z 1 Z
2 ǫ dd p Nµν (p, k)
= −4e µ dx
0 (2π)d [p2 + 2k · px + xk 2 − m2 + iε]2
Z 1 Z
2 ǫ dd p Nµν (p, k)
= −4e µ dx (4.10)
0 (2π) [(p + kx) + k2 x(1 − x) − m2 + iε]2
d 2
For dimension d sufficiently small this integral converges and we can change variables
p → p − kx (4.11)
We then get Z Z
1
dd p Nµν (p − kx, k)
i Πµν (k, ǫ) = −4e2 µǫ dx (4.12)
0 (2π)d [p2 − C + iǫ]2
where
C = m2 − k2 x(1 − x) (4.13)
Nµν is a polynomial of second degree in the loop momenta as can be seen from Eq. (4.8).
However as the denominator in Eq. (4.12) only depends on p2 is it easy to show that
Z
dd p pµ
=0
(2π)d [p2 − C + iǫ]2
Z Z
dd p pµ pν 1 µν dd p p2
= g (4.14)
(2π)d [p2 − C + iǫ]2 d (2π)d [p2 − C + iǫ]2
This means that we only have to calculate integrals of the form

Z
dd p (p2 )r
Ir,m =
(2π) [p − C + iǫ]m
d 2
Z d−1 Z
d p 0 (p2 )r
= dp (4.15)
(2π)d [p2 − C + iǫ]m
To make this integration we will use integration in the plane of the complex variable p0 as
described in Fig. 4.2. The deformation of the contour corresponds to the so called Wick
rotation, Z Z +∞ +∞
p0 → ip0E ; →i dp0E (4.16)
−∞ −∞
and p2 = (p0 )2 − |~
p|2 = −(p0E )2 − |~
p|2 ≡ −p2E , where pE = (p0E , p~) is an euclidean vector,
that is
p2E = (p0E )2 + |~
p|2 (4.17)
We can then write (see the Appendix for more details),
Z d r
r−m d pE p2E
Ir,m = i(−1) (4.18)
(2π)d p2E + C m
Im p0
x Re p0
Figure 4.2: Wick rotation.
where we do not need the iǫ anymore because the denominator is positive definite2 (C > 0).
To proceed with the evaluation of Ir,m we write,
Z Z
dd pE = dp p d−1 dΩd−1 (4.19)
q
where p = (p0E )2 + |~
p|2 is the length of of vector pE in the euclidean space with d
dimensions and dΩd−1 is the solid angle that generalizes spherical coordinates. We can
show (see Appendix) that
Z d
π2
dΩd−1 = 2 d (4.20)
Γ( 2 )
The p integral is done using the result,
Z ∞
xp Γ( p+1
n )
dx n = π(−1)q−1 ap+1−nq p+1 p+1 (4.21)
0 (x + an )q n sin(π n ) Γ( 2 − q + 1)
and we finally get
d (−1)r−m Γ(r + d2 ) Γ(m − r − d2 )

Ir,m = iC r−m+ 2 (4.22)
Γ( d2 )
d
(4π) 2 Γ(m)
Note that the integral representation of Ir,m , Eq. (4.15) is only valid for d < 2(m − r) to
ensure the convergence of the integral when p → ∞. However the final form of Eq. (4.22)
can be analytically continued for all the values of d except for those where the function
Γ(m − r − d/2) has poles, which are (see section C.6),
d
m−r− 6= 0, −1, −2, . . . (4.23)
2
For the application to dimensional regularization it is convenient to write Eq. (4.22) after
making the substitution d = 4 − ǫ. We get
ǫ
(−1)r−m 4π 2 2+r−m Γ(2 + r − 2ǫ ) Γ(m − r − 2 + 2ǫ )
Ir,m = i C (4.24)
(4π)2 C Γ(2 − 2ǫ ) Γ(m)
2
The case when C < 0 is obtained by analytical continuation of the final result.
that has poles for m − r − 2 ≤ 0 (see section C.6).

We now go back to calculate Πµν . First we notice that after the change of variables of
Eq. (4.11) we get, neglecting terms that vanish due to Eq. (4.14),

Nµν (p − kx, k) = 2pµ pν + 2x2 kµ kν − 2xkµ kν − gµν p2 + x2 k2 − xk 2 − m2 (4.25)
and therefore
Z
ǫ dd p Nµν (p − kx, k)
Nµν ≡ µ
(2π)d [p2 − C + iǫ]2

2
= ǫ 2 2
− 1 gµν µ I1,2 + − 2x(1 − x)kµ kν +x(1 − x)k gµν + gµν m µǫ I0,2 (4.26)
d
Using now Eq. (4.24) we can write

2ǫ
ǫ i 4πµ2 Γ( 2ǫ )
µ I0,2 =
16π 2 C Γ(2)

i C
= ∆ǫ − ln 2 + O(ǫ) (4.27)
16π 2 µ
where we have used the expansion of the Γ function, Eq. (C.47),

ǫ 2
Γ = − γ + O(ǫ) (4.28)
2 ǫ
γ being the Euler constant and we have defined, Eq. (C.50),
2
∆ǫ = − γ + ln 4π (4.29)
ǫ
In a similar way
2ǫ
ǫ i 4πµ2 Γ(3 − 2ǫ ) Γ(−1 + 2ǫ )
µ I1,2 = − C
16π 2 C Γ(2 − 2ǫ ) Γ(2)

i C
= 2
C 1 + 2∆ǫ − 2 ln 2 + O(ǫ) (4.30)
16π µ
Due to the existence of a pole in 1/ǫ in the previous equations we have to expand all
quantities up to O(ǫ). This means for instance, that
2 2 1 1
−1= − 1 = − + ǫ + O(ǫ2 ) (4.31)
d 4−ǫ 2 8
Substituting back into Eq. (4.26), and using Eq. (4.13), we obtain

1 1 2 i C
Nµν = gµν − + ǫ + O(ǫ ) C 1 + 2∆ǫ − 2 ln 2 + O(ǫ)
2 8 16π 2 µ

2 2 i C
+ − 2x(1 − x)kµ kν +x(1 − x)k gµν + gµν m ∆ǫ − ln 2 + O(ǫ)
16π 2 µ

i C
= − kµ kν ∆ǫ − ln 2x(1 − x)
16π 2 µ2

i 2 C
+ gµν k ∆ǫ x(1 − x) + x(1 − x) + ln 2 − x(1 − x) − x(1 − x)
16π 2 µ
#
1 1
+ x(1 − x) −
2 2

i 2 C 1 1
+ gµν m ∆ǫ (−1 + 1) + ln 2 (1 − 1) + (− + ) (4.32)
16π 2 µ 2 2
and finally

i C
Nµν = ∆ǫ − ln gµν k2 − kµ kν 2x(1 − x) (4.33)
16π 2 µ2
Now using Eq. (4.7) we get
Z 1
12 2
C
Πµν = −4e gµν k − kµ kν dx 2x(1 − x) ∆ǫ − ln 2
16π 2 0 µ

= − gµν k2 − kµ kν Π(k2 , ǫ) (4.34)
where
Z 1
2α m2 − x(1 − x)k2
Π(k2 , ǫ) ≡ dx x(1 − x) ∆ǫ − ln (4.35)
π 0 µ2
This expression clearly diverges as ǫ → 0. Before we show how to renormalize it

let us discuss the meaning of Πµν (k). The full photon propagator is given by the series
represented in Fig. 4.3, where
Gµν =
= + +
+ + ...
Figure 4.3: Full photon propagator.

≡ i Πµν (k) = sum of all one-particle irreducible

(4.36)
(proper ) diagrams to all orders
In lowest order we have the contribution represented in Fig. 4.4, which is what we have
Figure 4.4: Lowest order contribution.
just calculated. To continue it is convenient to rewrite the free propagator of the photon
(in an arbitrary gauge ξ) in the following form

0 kµ kν 1 kµ kν T 1 kµ kν
iGµν = gµν − 2 2
+ ξ 4 = Pµν 2
+ξ 4
k k k k k
≡ iG0T 0L
µν + iGµν (4.37)
T defined by
where we have introduced the transversal projector Pµν

T kµ kν
Pµν = gµν − 2 (4.38)
k
obviously satisfying the relations,

 kµ Pµν
T =0
(4.39)
 P T νP T = P T
µ νρ µρ
The full photon propagator can also in general be written separating its transversal an
longitudinal parts
Gµν = GTµν + GL
µν (4.40)
where GTµν satisfies
GTµν = Pµν
T
Gµν (4.41)
We have obtained, in first order, that the vacuum polarization tensor is transversal,
that is
i Πµν (k) = −ik 2 Pµν

T
Π(k) (4.42)
This result is in fact valid to all orders of perturbation theory, a result that can be shown
using the Ward-Takahashi identities. This means that the longitudinal part of the photon
propagator is not renormalized,
GL 0L
µν = Gµν (4.43)
For the transversal part we obtain
1 T 1 ′ ′ 1
iGTµν T
= Pµν 2
+ Pµµ ′
2
(−i)k2 P T µ ν Π(k)(−i)PνT′ ν 2
k k k
T 1 1 T 1
+Pµρ 2
(−i)k2 P T ρλ Π(k)(−i)Pλτ T
2
(−i)k2 P T τ σ Π(k)(−i)Pσν + ···
k k k2
T 1

= Pµν 2
1 − Π(k) + Π2 (k2 ) + · · · (4.44)
k
which gives, after summing the geometric series,
1
iGTµν = Pµν
T (4.45)
k2 1 + Π(k)
All that we have done up to this point is formal because the function Π(k) diverges.
The most satisfying way to solve this problem is the following. The initial lagrangian
from which we started has been obtained from the classical theory and nothing tell us
that it should be exactly the same in quantum theory. In fact, as we have just seen, the
normalization of the wave functions is changed when we calculate one-loop corrections,
and the same happens to the physical parameters of the theory, the charge and the mass.
Therefore we can think that the correct lagrangian is obtained by adding corrections to the
classical lagrangian, order by order in perturbation theory, so that we keep the definitions
of charge and mass as well as the normalization of the wave functions. The terms that we
add to the lagrangian are called counterterms 3 . The total lagrangian is then,
Ltotal = L(e, m, ...) + ∆L (4.46)

Counterterms are defined from the normalization conditions that we impose on the fields
and other parameters of the theory. In QED we have at our disposal the normalization
of the electron and photon fields and of the two physical parameters, the electric charge
and the electron mass. The normalization conditions are, to a large extent, arbitrary. It
is however convenient to keep the expressions as close as possible to the free field case,
that is, without radiative corrections. We define therefore the normalization of the photon
field as,
lim k2 iGRT T
µν = 1 · Pµν (4.47)
k→0
where GRT
µν is the renormalized propagator (the transversal part) obtained from the la-
grangian Ltotal . The justification for this definition comes from the following argument.
3
This interpretation in terms of quantum corrections makes sense. In fact we can show that an expansion
in powers of the coupling constant can be interpreted as an expansion in h̄L , where L is the number of the
loops in the expansion term.
Consider the Coulomb scattering to all orders of perturbation theory. We have then the
situation described in Fig. 4.5. Using the Ward-Takahashi identities one can show that
+ + +
Figure 4.5: Corrections to Coulomb scattering.
the last three diagrams cancel in the limit q = p′ − p → 0. Then the normalization con-
dition, Eq. (4.47), means that we have the situation described in Fig. 4.6, that is, the
experimental value of the electric charge is determined in the limit q → 0 of the Coulomb
scattering.
lim =
q→0
Figure 4.6: Definition of the electric charge.
The counterterm lagrangian has to have the same form as the classical lagrangian to
respect the symmetries of the theory. For the photon field it is traditional to write
1 1
∆L = − (Z3 − 1)Fµν F µν = − δZ3 Fµν F µν (4.48)
4 4
corresponding to the following Feynman rule

k k 2 kµ kν
µ ν − i δZ3 k gµν − 2 (4.49)
k
We have then

kµ kν
iΠµν = iΠloop
µν − i δZ3 k 2
gµν − 2
k
= −i (Π(k, ǫ) + δZ3 ) k2 Pµν

T
(4.50)
Therefore we should make the substitution
Π(k, ǫ) → Π(k, ǫ) + δZ3 (4.51)
in the photon propagator. We obtain,

1 1
iGTµν = Pµν
T
2
(4.52)
k 1 + Π(k, ǫ) + δZ3
The normalization condition, Eq. (4.47), implies
Π(k, ǫ) + δZ3 = 0 (4.53)
from which one determines the constant δZ3 . We get

Z
2α 1 m2
δZ3 = −Π(0, ǫ) = − dx x(1 − x) ∆ǫ − ln 2
π 0 µ

α m2
= − ∆ǫ − ln 2 (4.54)
3π µ
The renormalized photon propagator can then be written as 4
PµνT
iGµν (k) = 2 + i GL
µν (4.55)
k [1 + Π(k, ǫ) − Π(0, ǫ)]
The finite radiative corrections are given through the function
ΠR (k2 ) ≡ Π(k2 , ǫ) − Π(0, ǫ)

Z 2
2α 1 m − x(1 − x)k2
=− dx x(1 − x) ln
π 0 m2
( "
2 1/2 2 1/2 #)
α 1 2m2 4m 4m
=− +2 1+ 2 −1 cot−1 −1 −1 (4.56)
3π 3 k k2 k2
where the last equation is valid for k2 < 4m2 . For values k2 > 4m2 the result for ΠR (k2 )
can be obtained from Eq. (4.56) by analytical continuation. Using (k2 > 4m2 )

−1 −1 iπ
cot iz = i − tanh z + (4.57)
2
and 1/2 r
4m2 4m2
− 1 → i 1 − (4.58)
k2 k2
4
Notice that the photon mass is not renormalized, that is the pole of the photon propagator still is at
k2 = 0.
we get
( " r 1/2
R 2 α 1 2m2 4m2 −1 4m2
Π (k ) = − +2 1+ 2 −1 + 1 − 2 tanh 1− 2 (4.59)
3π 3 k k k
r #)
π 4m2
−i 1− 2 (4.60)
2 k
The imaginary part of ΠR is given by

r
R α 2 2m2 4m2 4m2
Im Π (k ) = 1+ 2 1− 2 θ 1− 2 (4.61)
3 k k k
and it is related to the pair production that can occur 5 for k2 > 4m2 .
4.1.2 Self-energy of the electron

The electron full propagator is given by the diagrammatic series of Fig. 4.7, which can be
written as,
= + +
+ + ...
Figure 4.7: Full electron propagator

S(p) = S (p) + S (p) − i Σ(p) S 0 (p) + · · ·
0 0

0
= S (p) 1 − i Σ(p)S(p) (4.62)
where we have identified

≡ −i Σ(p) (4.63)
Multiplying on the left with S0−1 (p) and on the right with S −1 (p) we get
S0−1 (p) = S −1 (p) − i Σ(p) (4.64)

which we can rewrite as
5
For k2 > 4m2 there is the possibility of producing one pair e+ e− . Therefore on top of a virtual process
(vacuum polarization) there is a real process (pair production).
S −1 (p) = S0−1 (p) + i Σ(p) (4.65)

Using the expression for the free field propagator,
i
S0 (p) = =⇒ S0−1 (p) = −i(p/ − m) (4.66)
p/ − m
we can then write
S −1 (p) = S0−1 (p) + iΣ(p)

= −i p/ − (m + Σ(p)) (4.67)
We conclude that it is enough to calculate Σ(p) to all orders of perturbation theory to

obtain the full electron propagator. The name self-energy given to Σ(p) comes from the
fact that, as can be seen in Eq. (4.67), it comes as an additional (momentum dependent)
contribution to the mass.
In lowest order there is only the diagram of Fig. 4.8 contributing Σ(p) and therefore
we get, Z
2 d4 k gµν i
−iΣ(p) = (+ie) 4
(−i) 2 2
γµ γν (4.68)
(2π) k − λ + iε p/ + k/ − m + iε
p p+k p
Figure 4.8: Self-energy of the electron
where we have chosen the Feynman gauge (ξ = 1) for the photon propagator and we have
introduced a small mass for the photon λ, in order to control the infrared divergences
(IR) that will appear when k2 → 0 (see below). Using dimensional regularization and the
results of the Dirac algebra in dimension d,
γµ (p/ + k/)γ µ = −(p/ + k/)γµ γ µ + 2(p/ + k/) = −(d − 2)(p/ + k/)
mγµ γ µ = m d (4.69)
we get
Z
ǫ 2 dd k 1 p/ + k/ + m
−i Σ(p) = −µ e d 2 2
γµ γµ
(2π) k − λ + iε (p + k)2 − m2 + iε
Z
ǫ 2 dd k −(d − 2)(p/ + k/) + m d
= −µ e
(2π) [k − λ2 + iε] [(p + k)2 − m2 + iε]
d 2
Z 1 Z
ǫ 2 dd k −(d − 2)(p/ + k/) + m d
= −µ e dx
0 (2π) [(k − λ ) (1 − x) + x(p + k)2 − xm2 + iε]2
d 2 2
Z 1 Z
dd k −(d − 2)(p/ + k/) + m d
= −µǫ e2 dx
0 (2π)d [(k + px)2 + p2 x(1 − x) − λ2 (1 − x) − xm2 + iε]2
Z 1 Z
ǫ 2 dd k −(d − 2) [p/(1 − x) + k/] + m d
= −µ e dx
0 (2π) d
[k + p2 x(1 − x) − λ2 (1 − x) − xm2 + iε]2
2
Z 1 h i
ǫ 2
= −µ e dx − (d − 2)p/(1 − x) + m d I0,2 (4.70)
0
where6
i
I0,2 =
2
∆ǫ − ln −p2 x(1 − x) + m2 x + λ2 (1 − x) (4.71)
16π
The contribution from the loop in Fig. 4.8 to the electron self-energy Σ(p) can then be
written in the form,
Σ(p)loop = A(p2 ) + B(p2 ) p/ (4.72)
with
Z 1
2 ǫ 1
A = e µ (4 − ǫ)m dx ∆ǫ − ln −p2 x(1 − x) + m2 x + λ2 (1 − x)
16π 2 0
Z 1
2 ǫ 1
B = −e µ (2 − ǫ) dx (1 − x) ∆ǫ
16π 2 0

− ln −p2 x(1 − x) + m2 x + λ2 (1 − x) (4.73)
Using now the expansions

ǫ 1 2
µ (4 − ǫ) = 4 1 + ǫ ln µ − + O(ǫ )
4

1
µǫ (4 − ǫ)∆ǫ = 4 ∆ǫ + 2 ln µ − + O(ǫ)
4

1
µǫ (2 − ǫ) = 2 1 + ǫ ln µ − 2
+ O(ǫ )
2

1
µǫ (2 − ǫ)∆ǫ = 2 ∆ǫ + 2 ln µ − + O(ǫ) (4.74)
2
we can finally write,

Z 1 2
2 4 e2 m 1 −p x(1 − x) + m2 x + λ2 (1 − x)
A(p ) = dx ∆ǫ − − ln (4.75)
16π 2 0 2 µ2
and
Z 1
2 2 e2 −p2 x(1 − x) + m2 x + λ2 (1 − x)
B(p ) = − dx (1 − x) ∆ǫ − 1 − ln (4.76)
16π 2 0 µ2
6
The linear term in k vanishes.
To continue with the renormalization program we have to introduce the counterterm la-
grangian and define the normalization conditions. We have
∆L = i (Z2 − 1) ψγ µ ∂µ ψ − (Z2 − 1) m ψψ + Z2 δm ψψ + (Z1 − 1)e ψγ µ ψAµ (4.77)
and therefore we get for the self-energy
−i Σ(p) = −i Σloop (p) + i (p/ − m) δZ2 + i δm (4.78)
Contrary to the case of the photon we see that we have two constants to determine. In
the on-shell renormalization scheme that is normally used in QED the two constants are
obtained by requiring that the pole of the propagator corresponds to the physical mass
(hence the name of on-shell renormalization), and that the residue of the pole of the
renormalized electron propagator has the same value as the free field propagator. This
implies,
Σ(p/ = m) = 0 → δm = Σloop (p/ = m)

∂Σ ∂Σloop
= 0 → δZ2 = (4.79)
∂p/ p/=m ∂p/ p/=m
We then get for δm,
δm = A(m2 ) + m B(m2 )
Z 2 2
2 me2 1 m x + λ2 (1 − x)
= dx 2∆ ǫ − 1 − 2 ln
16π 2 0 µ2
2 2
m x + λ2 (1 − x)
−(1 − x) ∆ǫ − 1 − ln
µ2
Z 1 2 2
2 m e2 3 1 m x + λ2 (1 − x)
= ∆ǫ − − dx (1 + x) ln
16π 2 2 2 0 µ2
Z 2 2
3αm 1 2 1 m x
= ∆ǫ − − dx (1 + x) ln (4.80)
4π 3 3 0 µ2
where in the last step in Eq. (4.80) we have taken the limit λ → 0 because the integral
does not diverge in that limit7 . In a similar way we get for δZ2 ,

∂Σloop ∂A ∂B
δZ2 = = +B+m (4.81)
∂p/ p/=m ∂p/ p/=m ∂p/ p/=m
where
Z 1
∂A 4 e2 m2 2(1 − x)x
= dx
∂p/ p/=m 16π 2 0 −m2 x(1 − x) + m2 x + λ2 (1 − x)
Z 1
2 α m2 (1 − x)x
= dx
π 0 m2 x2 + λ2 (1 − x)
7
δm is not IR divergent.
Z 1
α m2 x2 + λ2 (1 − x)
B = − dx (1 − x) ∆ǫ − 1 − ln
2π 0 µ2
Z
∂B α 2 1 2x(1 − x)2
m = − m dx (4.82)
∂p/ p/=m 2π 0 m2 x2 + λ2 (1 − x)
Substituting Eq. (4.82) in Eq. (4.81) we get,

Z 1 2 2 Z 1
α 1 1 m x (1 + x)(1 − x)xm2
δZ2 = − ∆ǫ − − dx (1 − x) ln −2 dx
2π 2 2 0 µ2 0 m2 x2 + λ2 (1 − x)

α m2 λ2
= −∆ǫ − 4 + ln 2 − 2 ln 2 (4.83)
4π µ m
where we have taken the λ → 0 limit in all cases that was possible. It is clear that the
final result in Eq. (4.83) diverges in that limit, therefore implying that Z2 is IR divergent.
This is not a problem for the theory because δZ2 is not a physical parameter. We will see
in section 4.4.2 that the IR diverges cancel for real processes. If we had taken a general
gauge (ξ 6= 1) we will find out that δm would not be changed but that Z2 would show a
gauge dependence. Again, in physical processes this should cancel in the end.
4.1.3 The Vertex

The diagram contributing to the QED vertex at one-loop is the one shown in Fig. 4.9. In
p′
k
µ
Figure 4.9: The QED vertex.
the Feynman gauge (ξ = 1) this gives a contribution,
Z
dd k gρσ
ie µǫ/2 Λloop ′
µ (p , p) = (ie µ
ǫ/2 3
) d
(−i) 2
(2π) k − λ2 + iε
i[(p/′ + k/) + m] i[(p/ + k/) + m]
γσ γµ γρ (4.84)
(p + k) − m + iε (p + k)2 − m2 + iε
′ 2 2
where Λµ is related to the full vertex Γµ through the relation
iΓµ = ie (γµ + Λloop

µ + γµ δZ1 )

= ie γµ + ΛR µ (4.85)
The integral that defines Λloop ′

µ (p , p) is divergent. As before we expect to solve this problem
by regularizing the integral, introducing counterterms and normalization conditions. The
counterterm has the same form as the vertex and is already included in Eq. (4.85). The
normalization constant is determined by requiring that in the limit q = p′ − p → 0 the
vertex reproduces the tree level vertex because this is what is consistent with the definition
of the electric charge in the q → 0 limit of the Coulomb scattering. Also this should be
defined for on-shell electrons. We have therefore that the normalization condition gives,

u(p) Λloop
µ + γ µ δZ1 u(p) =0 (4.86)
p
/=m
If we are interested only in calculating δZ1 and in showing that the divergences can be
removed with the normalization condition then the problem is simpler. It can be done in
two ways.
1st method
We use the fact that δZ1 is to be calculated on-shell and for p = p′ . Then
Z
dd k 1 1 1
iΛloop 2 ǫ
µ (p, p) = e µ d 2 2
γρ γµ γρ (4.87)
(2π) k − λ + iε p/ + k/ − m + iε p/ + k/ − m + iε
However we have
1 1 ∂ 1
γµ =− µ (4.88)
p/ + k/ − m + iε p/ + k/ − m + iε ∂p p/ + k/ − m + iε
and therefore
Z
∂ dd k 1 p/ + k/ + m
iΛloop
µ (p, p) = −e µ 2 ǫ
γρ γρ (4.89)
∂pµ (2π) k − λ + iε (p + k)2 − m2 + iε
d 2 2
∂
= −i Σloop (p) (4.90)
∂pµ
We conclude then, that Λloop 8

µ (p, p) is related to the self-energy of the electron ,
∂
Λloop
µ (p, p) = − Σloop (4.91)
∂pµ
On-shell we have

∂Σloop
Λloop
µ (p, p) = − = −δZ2 γµ (4.92)
p
/=m ∂pµ p/=m
and the normalization condition, Eq. (4.86), gives
δZ1 = δZ2 (4.93)

As we have already calculated δZ2 in Eq. (4.83), then δZ1 is determined.
8
This result is one of the forms of the Ward-Takahashi identity.
2nd method
In this second method we do not rely in the Ward identity but just calculate the integrals
for the vertex in Eq. (4.84). For the moment we do not put p′ = p but we will assume
that the vertex form factors are to be evaluated for on-shell spinors. Then we have
Z
′ loop 2 ǫ dd k u(p)γρ [p/′ + k/ + m)] γµ [p/ + k/ + m)] γ ρ u(p)
i u(p )Λµ u(p) = e µ
(2π)d D0 D1 D2
Z
dd k Nµ
= e2 µǫ d
(4.94)
(2π) D0 D1 D2
where

Nµ = u(p) (−2 + d)k2 γµ + 4p · p′ γµ + 4(p + p′ ) · k γµ + 4m kµ

′
− 4k/ (p + p )µ + 2(2 − d)k/kµ u(p) (4.95)
D0 = k2 − λ2 + iǫ (4.96)
D1 = (k + p′ )2 − m2 + iǫ (4.97)
D2 = (k + p)2 − m2 + iǫ (4.98)
Now using the results of section C.7.3 with
r1µ = p′µ ; r2µ = pµ (4.99)
P µ = x1 p′µ + x2 pµ (4.100)
C = (x1 + x2 )2 m2 − x1 x2 q 2 + (1 − x1 − x2 )λ2 (4.101)
where
q = p′ − p (4.102)
we get,
Z 1 Z 1−x1
′ α 1
i u(p )Λloop
µ u(p) = i Γ(3) dx1 dx2
4π 0 0 2C
( "
u(p′ )γµ u(p) − (−2 + d)(x21 m2 + x22 m2 + 2x1 x2 p′ · p) − 4p′ · p

′ (2 − d)2
′ C
+4(p + p ) · (x1 p + x2 p) + C ∆ǫ − ln 2
2 µ

+u(p)u(p)m 4(x1 p′ + x2 p)µ − 4(p′ + p)µ (x1 + x2 )
)
− 2(2 − d)(x1 + x2 )(x1 p′ + x2 p)µ (4.103)

= i u(p) G(q 2 ) γµ + H(q 2 ) (p + p′ ) u(p) (4.104)
where we have defined9 ,

Z 1 Z 1−x1
2 α (x1 + x2 )2 m2 − x1 x2 q 2 + (1 − x1 − x2 )λ2
G(q ) ≡ ∆ǫ − 2 − 2 dx1 dx2 ln
4π 0 0 µ2
Z 1 Z 1−x1
−2(x1 + x2 )2 m2 − x1 x2 q 2 − 4m2 + 2q 2
+ dx1 dx2
0 0 (x1 + x2 )2 m2 − x1 x2 q 2 + (1 − x1 − x2 )λ2

2(x1 + x2 )(4m2 − q 2 )
+ (4.105)
(x1 + x2 )2 m2 − x1 x2 q 2 + (1 − x1 − x2 )λ2
Z 1 Z 1−x1
2 α −2m (x1 + x2 ) + 2m (x1 + x2 )2
H(q ) ≡ dx1 dx2 (4.106)
4π 0 0 (x1 + x2 )2 m2 − x1 x2 q 2 + (1 − x1 − x2 )λ2
Now, using the definition of Eq. (4.85), we get for the renormalized vertex,

u(p′ )ΛR ′ ′ 2 2 ′
µ (p , p)u(p) = u(p ) G(q ) + δZ1 γµ + H(q ) (p + p )µ u(p) (4.107)
As δZ1 is calculated in the limit of q = p′ − p → 0 it is convenient to use the Gordon

identity to get rid of the (p′ + p)µ term. We have
h i
u(p′ ) p′ + p µ u(p) = u(p′ ) 2mγµ − iσµν q ν u(p) (4.108)
and therefore,
h i
u(p′ )ΛR ′ ′
µ (p , p)u(p) = u(p ) G(q 2 ) + 2mH(q 2 ) + δZ1 γµ − i H(q 2 ) σµν q ν u(p)

′ i 2 ν 2
= u(p ) γµ F1 (q ) + σµν q F2 (q ) u(p) (4.109)
2m
where we have introduced the usual notation for the vertex form factors,
F1 (q 2 ) ≡ G(q 2 ) + 2mH(q 2 ) + δZ1 (4.110)
F2 (q 2 ) ≡ −2mH(q 2 ) (4.111)
The normalization condition of Eq. (4.86) implies F1 (0) = 0, that is,
δZ1 = −G(0) − 2m H(0) (4.112)
We have therefore to calculate G(0) and H(0). In this limit the integrals of Eqs. (4.105)
and (4.106) are much simpler. We get (we change variables x1 + x2 → y),
Z 1 Z 1
α y 2 m2 + (1 − y)λ2
G(0) = ∆ǫ − 2 − 2 dx1 dy ln
4π 0 x1 µ2
9
To obtain Eq. (4.106) one has to show that the symmetry of the integrals in x1 ↔ x2 implies that the
coefficient of p is equal to the coefficient of p′ .
Z 1 Z 1
−2y 2 m2 − 4m2 + 8ym2
+ dx1 dy (4.113)
0 x1 y 2 m2 + (1 − y)λ2
Z 1 Z 1
α −2m y + 2m y 2
H(0) = dx1 dy (4.114)
4π 0 x1 y 2 m2 + (1 − y)λ2
Now using
Z 1 Z 1
y 2 m2 + (1 − y)λ2 1 m2
dx1 dy ln = ln − 1 (4.115)
0 x1 µ2 2 µ2
Z 1 Z 1
−2y 2 m2 − 4m2 + 8ym2 λ2
dx1 dy = 7 + 2 ln (4.116)
0 x1 y 2 m2 + (1 − y)λ2 m2
Z 1 Z 1
−2m y + 2m y 2 1
dx1 dy 2 2 2
=− (4.117)
0 x1 y m + (1 − y)λ m
(where we took the limit λ → 0 if possible) we get,

α m2 λ2
G(0) = ∆ǫ + 6 − ln 2 + 2 ln 2 (4.118)
4π µ m
α 1
H(0) = − (4.119)
4π m
Substituting the previous expressions in Eq. (4.112) we get finally,

α m2 λ2
δZ1 = −∆ǫ − 4 + ln 2 − 2 ln 2 (4.120)
4π µ m
in agreement with Eq. (4.83) and Eq. (4.93). The general form of the form factors Fi (q 2 ),
for q 2 6= 0, is quite complicated. We give here only the result for q 2 < 0 (in section C.10.3
we will give a general formula for numerical evaluation of these functions),
( Z θ/2 )
2 α λ2 θ
F1 (q ) = 2 ln 2 + 4 (θ coth θ − 1) − θ tanh − 8 coth θ β tanh βdβ
4π m 2 0
α θ
F2 (q 2 ) = (4.121)
2π sinh θ
where
θ q2
= − 2· sinh2 (4.122)
2 4m
In the limit of zero transferred momenta (q = p′ − p = 0) we get

 F1 (0) = 0
(4.123)
 F2 (0) = α
2π
a result that we will use in section 4.4.1 while discussing the anomalous magnetic moment
of the electron.
4.2 Ward-Takahashi identities in QED

In the study of the QED vertex, in one of the methods, we used the Ward identity
∂
Λµ (p, p) = − Σ(p) (4.124)
∂pµ
We are going to derive here the general form for these identities. The following discussion
is formal in the sense that the various Green functions are divergent. We have to prove
that we can find a regularization scheme that preserves the identities. This happens when
one uses a regularization that preserves the gauge invariance of the theory. Examples are
dimensional regularization and the Pauli-Villars regularization.
Ward identities are a consequence of the gauge invariance of the theory, as will be
fully discussed in chapters 5 and 6. Here we are only going to use the fact that there is a
conserved current,
jµ = eψγµ ψ
(4.125)
∂µ j µ = 0
We are interested in calculating the quantity
∂xµ h0| T jµ (x)ψ(x1 )ψ(y1 ) · · · ψ(xn )ψ(yn )Aν1 (z1 ) · · · Aνp (zp ) |0i (4.126)
This quantity does not vanish, despite the fact that ∂ µ jµ = 0. This happens because
in the time ordered product we have θ functions that depend on the coordinate x0 . For
instance, for the field ψ(xi ) we should have a contribution of the form,

∂x0 θ(x0 − x0i )j0 (x)ψ(xi ) + θ(x0i − x0 )ψ(xi )j0 (x)
= δ(x0 − x0i )j0 (x)ψ(xi ) − δ(x0 − x0i )ψ(xi )j0 (x)

= j0 (x), ψ(xi ) δ(x0 − x0i ) (4.127)
In this way we get ( b means that we omit that term from the sum),
∂xµ h0| T jµ (x)ψ(x1 ) · · · ψ(yn )Aν1 (z1 ) · · · Aνp (zp ) |0i

n
X
= h0| T [j0 (x), ψ(xi )] δ(x0 − x0i )ψ(yi )
i=1

+ψ(xi ) j0 (x), ψ(yi ) δ(x0 − yi0 ) ψ(x1 )ψ(y1 ) · · · ψ(xd
i )ψ(yi ) · · · Aνp (zp ) |0i
p
X
+ h0| T ψ(x1 ) · ψ(yn )Aνp (z1 ) · · · [j0 (x), Aνj (zj )]δ(x0 − zj0 ) · · · Aνp (zp ) |0i (4.128)
j=1
Using now the equal time commutation relations,

j0 (x), ψ(x′ ) δ(x0 − x′0 ) = −eψ(x)δ4 (x − x′ )

j0 (x), ψ(x′ ) δ(x0 − x′0 ) = eψ(x)δ4 (x − x′ ) (4.129)

j0 (x), Aµ (x′ ) δ(x0 − x′0 ) = 0
4.2. WARD-TAKAHASHI IDENTITIES IN QED 125
R
that express that ψ, ψ and Aµ create quanta with charge Q = d3 xj 0 (x) equal to −e, +e
and zero, respectively, we get,
∂xµ h0| T jµ (x)ψ(x1 ) · · · ψ(yn )Aν1 (z1 ) · · · Aνp (zp ) |0i

n
X 4
= e h0| T ψ(y1 ) · · · Aνp (zp ) |0i δ (x − yi ) − δ4 (x − xi ) (4.130)
i=1
Taking different values for n and p we get different relations among the Green functions
of the theory. We will consider in the following, two important cases.
4.2.1 Transversality of the photon propagator n = 0, p = 1

The Green function h0| T jµ (x)Aν (y) |0i corresponds to the Feynman diagram of Fig. 4.10,
and it is related with the full photon propagator shown in Fig. 4.11, by the diagrammatic
µ, x ν, y
Figure 4.10: Green function h0| T jµ (x)Aν (y) |0i.
relation shown in Fig. 4.12 known as the Dyson-Schwinger equation for QED. It can be
written as
Z
Gµν (x − y) = Gµν (x − y) − i d4 x′ G0µρ (x − x′ ) h0| T j ρ (x′ )Aν (y) |0i
0
(4.131)
We apply now the derivative ∂xµ to get,

Z
∂x Gµν (x − y) = ∂x Gµν (x − y) − i d4 x′ ∂xµ G0µρ (x − x′ ) h0| T j ρ (x′ )Aµ (y) |0i
µ µ 0
(4.132)
The free photon propagator is given by,

Z
d4 p −i(x−x′ )·p 0
G0µρ (x − x′ ) = e Gµρ (p) (4.133)
(2π)4
where
pµ pν 1 pµ pν
G0µν (p) = −i gµν − 2 +ξ 4 . (4.134)
p p2 p
µ, x ν, y
Figure 4.11: Photon propagator

µ, x ν, y = µ, x ν, y + µ, x ν, y
Figure 4.12: Dyson-Schwinger equation.
Therefore
Z
d4 p −i(x−x′ )·p
∂xµ G0µρ (x ′
−x) = e (−ipµ )G0µρ (p)
(2π)4
Z
d4 p −i(x−x′ )·p
= e (−ipρ )F (p2 )
(2π)4
Z
x′ d4 p −i(x−x′ )·p
= −∂ρ e F (p2 )
(2π)4
′
= −∂ρx Fe(x − x′ ) (4.135)
and we get
Z
∂xµ Gµν (x − y) = ∂xµ G0µν (x − y) + i d4 x′ ∂xρ′ Fe(x − x′ ) h0| T jρ (x′ )Aν (y) |0i
Z
′
= ∂xµ G0µν (x − y) − i d4 x′ Fe(x − x′ )∂ρx h0| T j ρ (x′ )Aν (y) |0i
= ∂xµ G0µν (x − y) (4.136)
where we have made an integration by parts and used the Ward-Takashashi identity for
n = 0, p = 1. We have then
∂xµ Gµν (x − y) = ∂xµ G0µν (x − y) (4.137)
which in momenta space implies
pµ Gµν (p) = pµ G0µν (p) (4.138)
This means that the longitudinal part of the photon propagator is not renormalized,
or in other words, that the self-energy of the photon (vacuum polarization) is transverse.
In fact
pν
pµ G0µν (p) = −iξ 2 (4.139)
p
or
pν pν
pµ = −iξ 2 G−1 νµ (p) = −ξ 2 Γνµ (p) (4.140)
p p
But, in agreement with our conventions, we have
1
Γνµ (p) = −(gνµ p2 − pν pµ ) − pν pµ + Πνµ (p2 ) (4.141)
ξ
and therefore
pν 1 1 ν
−ξ 2
Γνµ (p) = pµ − p Πνµ (p2 ) = pµ (4.142)
p ξ p2
which gives
pν Πνµ (p2 ) = 0 (4.143)
that is, the self-energy is transverse.
4.2.2 Identity for the vertex n = 1, p = 0

We are now interested in the Green function,
h0| T jµ (x)ψβ (x1 )ψ α (y1 ) |0i (4.144)
to which corresponds the diagram of Fig. 4.13. This Green function can be related with
µ, x
β, x1 α, y1
Figure 4.13: Green function h0| T jµ (x)ψβ (x1 )ψ α (y1 ) |0i.
the vertex h0| T Aµ (x)ψβ (x1 )ψ α (y1 ) |0i corresponding to the diagram of Fig. 4.14, through
µ, x
β, x1 α, y1
Figure 4.14: Full vertex.
the following diagrammatic equation,

µ, x µ, x
= (4.145)
β, x1 α, y1 β, x1 α, y1
that we can write as,

Z
h0| T Aµ (x)ψβ (x1 )ψ α (y1 ) |0i = −i d4 x′ G0µν (x − x′ ) h0| T j ν (x′ )ψβ (x1 )ψ α (y1 ) |0i (4.146)
Taking the Fourier transform,

Z
′
d4 xd4 x1 d4 y1 ei(p ·x1 −p·y1−q·x) h0| T Aµ (x)ψβ (x1 )ψ α (y1 ) |0i
Z
′
= −iG0µν (q) d4 xd4 x1 d4 y1 ei(p ·x1 −p·y1−q·x) h0| T j ν (x)ψβ (x1 )ψ α (y1 ) |0i (4.147)
where the direction of the momenta are shown in Fig. 4.15, and the momentum transfered
q = p′ − q
p′ p
β α
Figure 4.15: Definition of the momenta in the vertex.
is q = p′ − p.
On the other side, using the definition of Γµ , we have,
Z
′
d4 xd4 x1 d4 y1 ei(p ·x1 −p·y1 −q·x) h0| T Aµ (x)ψβ (x1 )ψ α (y1 ) |0i
4.3. COUNTERTERMS AND POWER COUNTING 129

= (2π)4 δ4 (p′ − p − q)Gµν (q) S(p′ )iΓν (p′ , p)S(p) βα (4.148)
Therefore we get,
(2π)4 δ(p′ − p − q)Gµν (q)S(p′ )iΓν (p′ , p)S(p)

Z
′
= −iGµν (q) d4 xd4 x1 d4 y1 ei(p ·x1 −p·y1−q·x) h0| T j ν (x)ψ(x1 )ψ(y1 ) |0i (4.149)
0
Multiplying by q µ and using the result,

qν
q µ Gµν (q) = q µ G0µν (q) = −iξ (4.150)
q2
we can then write (using the Ward identity for n = 1, p = 0)
(2π)4 δ(p′ − q − p)S(p′ )q ν Γν (p′ , p)S(p)

Z
′
= i d4 xd4 x1 d4 y1 ∂xν h0| T jν (x)ψ(x1 )ψ(y1 ) |0i ei(p ·x1 −p·y1−q·x)
Z
′
= ie d4x d4 x1 d4 y1 ei(p ·x1−p·y1 −q·x) h0| T ψ(x1 )ψ(y1 ) |0i δ(x − y1 ) − δ(x − x1 )

= ie(2π)4 δ(p′ − p − q) S(p′ ) − S(p) (4.151)
or
q ν Γν (p′ , p) = ie S −1 (p) − S −1 (p′ ) (4.152)
As q ν = (p′ − p)ν we get in the limit p′ = p,
∂S −1
Γν (p, p) = −ie
∂pν

∂Σ
= −e γν − ν (4.153)
∂p
Using Γν = −e(γν + Λν ) we finally get the Ward identity in the form used before,
∂Σ
Λν (p, p) = − . (4.154)
∂pν
4.3 Counterterms and power counting

All that we have shown in the previous sections can be interpreted as follows. The initial
Lagrangian L(e, m, · · · ) has been obtained from a correspondence between classical and
quantum theory. It is then natural that the initial Lagrangian has to be modified by
quantum corrections. The total Lagrangian is then given by,
Ltotal = L(e, m, · · · ) + ∆L (4.155)

and
∆L = ∆L(1) + ∆L[2] + · · · (4.156)
i
where ∆L[i] is the ith − loops correction. This also correspond to order h̄ as counting
in terms of loops is equivalent to counting in terms of h̄10 . This interpretation is quite
attractive because in the limit h̄ → 0 the total Lagrangian reduces to the classical one.
With the Lagrangian Ltot we can then obtain finite results, although Ltot is divergent
because of the counter-terms in ∆L.
With this language the results up to the first order in h̄ can be written as,
1 λ2 1
L(e, m, · · · ) = − Fµν F µν + Aµ Aµ − (∂ · A)2
4 2 2ξ
+iψ∂/ψ − mψψ − eψA

/ψ (4.157)
1
∆L(1) = − (Z3 − 1)Fµν F µν + (Z2 − 1)(iψ∂/ψ − mψψ)
4
+Z2 δmψψ − e(Z1 − 1)ψA
/ψ (4.158)
The Lagrangian
1 λ2 1
Ltotal = − Z3 Fµν F µν + Aµ Aµ − (∂ · A)2
4 2 2ξ
+Z2 (iψ∂/ψ − mψψ + δmψψ)
−eZ1 ψA
/ψ (4.159)
will give the renormalized Green’s functions up the the order h̄.
In fact, we have only shown that the two-point and three-point Green’s functions (self-
energies and vertex) were finite. It is important to verify that all the Green’s functions,
with an arbitrary number of external legs are finite, as we have already used all our freedom
in the renormalization of those Green’s functions. This leads us to the so-called power
counting.
Let us consider a Feynman diagram G, with L loops, IB bosonic and IF fermionic
internal lines. If there are vertices with derivatives, δv is the number of derivatives in
that vertex. We define then the superficial degree of divergence of the diagram (note that
L = IB + IF + 1 − V ) by,
X
ω(G) = 4L + δv − IF − 2IB
v
X
= 4 + 3IF + 2IB + (δv − 4) (4.160)
v
For large values of the momenta the diagram will be divergent as
Λω (G) se ω(G) > 0 (4.161)

10 E−1+L E V
h̄ = h̄ 2 + 2 . We have the following relations L = I − V + 1 ; 3V = E + 2I (this only for QED).
4.3. COUNTERTERMS AND POWER COUNTING 131
and as
ln Λ se ω(G) = 0 (4.162)
where Λ is a cutoff. The origin of the different terms can be seen in the following corre-
spondence, Z
d4 q
(for each loop) → 4L
(2π)4
∂µ ⇔ kµ → δv
(4.163)
i → −IF
/q − m
i → −2IB
q 2 − m2
The expression for ω(G) is more useful when expressed in terms of the number of ex-
ternal legs and of the dimensionality of the vertices of the theory. Let ωv be the dimension,
in terms of mass, of the vertex v, that is,
3
ωv = δv + #campos bosónicos + #campos fermiónicos (4.164)
2
Then, if we denote by fv (bv ) the number of fermionic (bosonic) internal lines that join at
the vertex v, we can write,
X X 3 3
ωv = (δv + fv + bv ) + EF + EB (4.165)
v v
2 2
where EF (EB ) are the total number of external fermionic (bosonic) lines of the diagram.
As we have,
1X
IF = fv
2 v
1X
IB = bv (4.166)
2 v
we get
X X 3
ωv = δv + 3IF + 2IB + EF + EB (4.167)
v v
2
Substituting in the expression for ω(G) we get finally,
3 X
ω(G) =4 − EF − EB + (ωv − 4)
2 v
3 X
=4 − EF − EB − [gv ] (4.168)
2 v
where [gv ] denotes the dimension in terms of mass of the coupling constant of vertex v,
satisfying,
ωv + [gv ] = 4 . (4.169)
From the previous expression for the superficial degree of divergence, Eq. (4.168), we
can then classify theories in three classes,
i) Non-renormalizable Theories
They have at least one vertex with ωv > 4 (or [gv ] < 0). The superficial degree of di-
vergence increases with the number of vertices, that is, with the order of perturbation
theory. For an order high enough all the Green functions will diverge.
ii) Renormalizable Theories

All the vertices have ωv ≤ 4 and at least one has ωv = 4. If all vertices have ωv = 4
then
3
ω(G) = 4 − EF − EB (4.170)
2
and all the diagrams contributing to a given Green function have the same degree
of divergence. Only a finite number of Green functions are divergent.
iii) Super-Renormalizable Theories

All the vertices have ωv < 4. Only a finite number of diagrams are divergent11
Coming back to our question of knowing which are the divergent diagram in QED, we
can now summarize the situation in Table 4.1. All the other diagrams are superficially
EF EB ω(G) Effective degree

of divergence
0 2 2 0 (Current Conservation (CC))
0 3 0 (Furry’s Theorem)
0 4 0 Convergent (CC)
2 0 1 0 (Current Conservation)
2 1 0 0
Table 4.1: Superficial and effective degree of divergence in QED.
convergent. We have therefore a situation where there are only a finite number of divergent
diagrams, exactly the ones that we considered before. This analysis shows that, up to order
h̄, the Lagrangian
1 1 1
Ltotal = − Z3 Fµν F µν + µAµ Aµ − (∂ · A)2
4 2 2ξ
+Z2 (iψ∂/ψ − mψψ + δmψψ)
−eZ1 ψA
/ψ (4.171)
gives Green functions that are finite and renormalized with an arbitrary number of external
legs. It remains to be shown that this Lagrangian is still valid up an arbitrary order in
h̄, with the only modification that the renormalization constants Z1 , Z2 , Z3 e δm are now
given by power series,
(1) (2)
Z1 = Z1 + Z1 + · · · (4.172)
11
The effective degree of divergence it is sometimes smaller than the superficial degree because of sym-
metries of the theory. This what happens for gauge theories like QED (see Table 4.1).
4.4. FINITE CONTRIBUTIONS FROM RC TO PHYSICAL PROCESSES 133
The previous Lagrangian, Eq. (4.171), allows for another interpretation that it is also
useful. The fields A, ψ and ψ are the renormalized fields that give the residues equal to
1 for the poles of the propagators and the constants e, m are the physical electric charge
and mass of the electron. Let us define the non-renormalized fields ψ0 , ψ 0 and A0 and the
bare (cutoff dependent) µ20 , m0 through the definitions,
√
ψ0 = Z2 ψ m0 = m − δm
√
ψ 0 = Z2 ψ µ20 = Z3−1 µ2
√ q (4.173)
−1
A0 = Z3 A e0 = Z1 Z2 Z3−1 e = √1Z e
3
ξ 0 = Z3 ξ
Then the Lagrangian written in terms of the bare quantities is identical to the original
Lagrangian12
1 1 1
L = − F0µν F0µν + µ0 A0µ Aµ0 − (∂ · A0 )2
4 2 2ξ0
+i(ψ 0 ∂/ψ0 − m0 ψ 0 ψ0 ) − e0 ψ 0 A
/0 ψ0 (4.174)
Finally we notice that the bare Green functions are related to the renormalized ones
by
Gn,ℓ
0 (p1 , · · · p2n , k1 , · · · kℓ , µ0 , m0 , ℓ0 , ξ0 , Λ)
ℓ/2
= Z2n (Λ)Z3 Gn,ℓ
R (p1 , · · · p2n , k1 · · · kℓ , µ, m, e, ξ) (4.175)
where p1 · · · p2n (k1 · · · kℓ ) are the fermion (boson) momenta. We will come back to these
relations in the study of the renormalization group, in chapter 7.
4.4 Finite contributions from RC to physical processes

4.4.1 Anomalous electron magnetic moment
We will show here, for the case of the electron anomalous moment, how the finite part
of the radiative corrections can be compared with experiment, given credibility to the
renormalization program. In fact we will just consider the first order, while to compare
with the present experimental limit one has to go to fourth order in QED and to include
also the weak and QCD corrections. The electron magnetic moment is given by
e ~σ
g ~µ = (4.176)
2m 2
where e = −|e| for the electron. One of the biggest achievements of the Dirac equation
was precisely to predict the value g = 2. Experimentally we know that g is close to, but
2 µ2
12
The terms µ2 A2 = 20 A20 and 2ξ 1
(∂ · A)2 = 2ξ10 (∂ · A0 )2 are not renormalized. This a consequence of
the Ward-Takashashi identities for QED. The Ward identity Z1 = Z2 is crucial for the equality e0 A0 = eA
given a meaning to the electric charge independently of the renormalization scheme.
not exactly, 2. It is usual to define this difference as the anomalous magnetic moment.
More precisely,
g = 2(1 + a) (4.177)
or
g
a=−1 (4.178)
2
Our task is to calculate a from the radiative corrections that we have computed in the
previous sections. To do that let us start to show how a value a 6= 0 will appear in
non relativistic quantum mechanics. Schrödinger’s equation for a charged particle in an
exterior field is, " #
∂ϕ (~ ~ 2
p − eA) e
i = + eφ − ~ ϕ
(1 + a)~σ · B (4.179)
∂t 2m 2m
~ =∇
Now we consider that the external field is a magnetic field B ~ × A.
~ Then keeping only
terms first order in e we get
p2 ~+A
p~ · A ~ · ~p e
H = −e − ~ ×A
(1 + a)~σ · ∇ ~
2m 2m 2m
≡ H0 + Hint (4.180)
With this interaction Hamiltonian we calculate the transition amplitude between two
electron states of momenta p and p′ . We get
Z

′ e d3 x χ† −i~p′ ·~x ~ ~
p Hint |pi = − e p
~ · A + A · p
~ + (1 + a)~σ × ~ ·A
∇ ~ ei~p·x χ
2m (2π)3
Z
e d3 x χ† ′
~ + i(1 + a)σ i ǫijk q j Ak e−i~q·x χ
= − 3
p + ~p) · A
(~
2m (2π)
e χ† ′
= − (p + p)k + i(1 + a)σ i ǫijk q j Ak (q)χ (4.181)
2m
This is the result that we want to compare with the non relativistic limit of the renormal-
ized vertex. The amplitude is given by,
A = eu(p′ )(γµ + ΛR µ
µ )u(p)A (q)

′ 2 i
= eu(p ) γµ (1 + F1 (q )) + σµν q F2 (q ) u(p)Aµ (q)
ν 2
2m
e n o
= u(p′ ) (p′ + p)µ 1 + F1 (q 2 ) + iσµν q ν 1 + F1 (q 2 ) + F2 (q 2 ) u(p)Aµ (q)(4.182)
2m
~ =∇
where we have used Gordon’s identity. For an external magnetic field B ~ ×A
~ and in
2
the limit q → 0 this expression reduces to
e
A = u(p′ ) (p′ + p)k 1 + F1 (0) + iσkj q j 1 + F1 (0) + F2 (0) u(p)Ak (q)
2m
e h α i
= u(p′ ) −(p′ + p)k + iΣi ǫkij q j 1 + u(p)Ak (q) (4.183)
2m 2π
where we have used the results of Eq. (4.123),


 F1 (0) = 0
(4.184)
 F2 (0) = α
2π
Using the explicit form for the spinors u
 
χ
 
u(p) = 


 (4.185)
~σ · (~ ~
p − eA) χ
2m
we can write in the non relativistic limit,
e χ† h ′ α i ijk j i χ k
A=− (p + p)k + i 1 + σǫ q A (4.186)
2m 2π
which after comparing with Eq. (4.181) leads to
α
aeth = (4.187)
2π
This result obtained for the first time by Schwinger and experimentally confirmed, was
very important in the acceptance of the renormalization program of Feynman, Dyson and
Schwinger for QED.
4.4.2 Cancellation of IR divergences in Coulomb scattering

In this section we will show how the IR divergences cancel in physical processes. We will
take as an example the Coulomb scattering from a fixed nucleus. This is better done if we
start from first principles. Coulomb scattering corresponds to the diagram of Fig. 4.16,
which gives the following matrix element for the S matrix,
Aµc
pi pf
Figure 4.16: Lowest order diagram to Coulomb scattering.
1
Sf i = iZe2 (2π)δ(Ei − Ef ) u(pf )γ 0 u(pi ) (4.188)
|~q|2
We will now study the radiative corrections to this result in lowest order in perturbation
theory. Due to the IR divergences it is convenient to introduce a mass λ for the photon.
For a classical field, as we are considering, this means a screening. If we take,
e−λ|~x|
A0c (x) = Ze (4.189)
4π|~x|
the Fourier transform will be,
1
A0c (q) = Ze (4.190)
|~q|2 + λ2
that shows that the screening is equivalent to a mass for the photon. With these modifi-
cations we have,
i
Sf i = iZe2 (2π)δ(Ef − Ei ) u(pf )γ 0 u(pi ) (4.191)
|~q|2 + λ2
We are interested in calculating the corrections up to order e3 in the amplitude. To
this contribute13 the diagrams of Fig. 4.17. Diagram 1 is of order e2 while diagrams 2, 3, 4
Aµc
Aµc Aµc Aµc Aµc
pi pf pi pf pi pf pi pf
Figure 4.17: Coulomb scattering up to second order.
are of order e4 . Therefore the interference between 1 and (2 + 3 + 4) is of order α3 and

should be added to the result of the bremsstrahlung in a Coulomb field. The contribution
from 1 + 2 + 3 can be easily obtained by noticing that
eAµc γµ → eAµc (γµ + ΛR R νρ

µ + Πµν G γρ ) (4.192)
where ΛR R
µ e Πµν have been calculated before. We get

(1+2+3) 2 1 0 α 1
Sf i = iZe (2π)δ(Ei − Ef ) 2 u(pf )γ 1 + − ϕ tanh ϕ
|~q| + λ2 π 2
Z ϕ
λ
1 + ln (2ϕ coth 2ϕ − 1) − 2 coth 2ϕ β tanh βdβ
m 0

coth2 ϕ 1 /q α ϕ
+ 1− (ϕ coth ϕ − 1) + − u(pi ) (4.193)
β 9 2m π sinh 2ϕ
where
|~q|2
= sinh2 ϕ . (4.194)
4m
Finally the fourth diagram gives
13
We do not have to consider the self-energies of the external legs of the electron because they are
on-shell.
Z
(4) 2 2 d4 k 2πδ(Ef − k0 ) 0 i 0
0 2πδ(k − Ei )
Sf i = (iZe) (e) u(pf ) γ γ
(2π)4 (pf − k)2 − λ2 k/ − m + iε (k − pi )2 − λ2
Z 2 α2
= −2i 2πδ(Ef − Ei ) u(pf ) m(I1 − I2 ) + γ 0 Ei (I1 + I2 ) u(pi ) (4.195)
π
with
Z
1
I1 = d3 k (4.196)
pf − ~k)2 + λ2 ][(~
[(~ pi − ~k)2 + λ2 ][(~
p)2 − (~k)2 + iε]
and
Z ~k
1
pi + p~f )I2 ≡
(~ d3 k . (4.197)
2 pf − ~k)2 + λ2 ][(~
[(~ pi − ~k)2 + λ2 ][(~
p)2 − (~k)2 + iε]
In the limit λ → 0 it can be shown that

π2 2p sin(θ/2)
I1 = ln (4.198)
2ip3 sin2 θ/2 λ

π2 π 1 1 2p sin θ/2 λ
I2 = 1 − − i ln + ln
2p3 cos2 θ/2 2 sin θ/2 sin2 θ/2 λ 2p
(4.199)
With these expressions we get for the cross section
dσ Z 2 α2 1 X
= |u(pf )Γu(pi )|2 (4.200)
dΩ q |2 2
|~
pol
where
/q
Γ = γ 0 (1 + A) + γ 0 B+C (4.201)
2m
and
ϕ Z
α λ ϕ
A = (2ϕ coth 2ϕ − 1) − 2 coth 2ϕ
1 + ln dββ tanh β − tanh ϕ
π m 0 2

1 1 Zα
+ 1 − coth2 ϕ (ϕ coth ϕ − 1) + − 2 |~q|2 E(I1 + I2 ) (4.202)
3 9 2π
α ϕ
B = − (4.203)
π sinh 2ϕ
Zα
C = − q |2 (I1 − I2 )
m|~ (4.204)
2π 2
Therefore
1X 1
|u(pf )pu(pi )|2 = Tr[Γ(p/i + m)Γ(p/f + m)]
4 4
pol
θ
= 2E 2 (1 − β 2 sin2 θ/2) + 2E 2 2Bβ 2 sin2
2

2 2 2 θ
+2E 2ReA 1 − β sin + 2ReC(2mE) + O(α2 ) (4.205)
2
Notice that A, B e C are of order α. Therefore the final result is, up to order α3 :

dσ dσ 2α λ ϕ
= 1+ 1 + ln (2ϕ coth ϕ − 1) − tanh ϕ
dΩ elastic dΩ
Mott π m 2
Z ϕ
coth2 ϕ 1
−2 coth 2ϕ dββ tanh β + − (ϕ coth ϕ − 1) +
0 3 9
)
ϕ B 2 sin2 θ/2 β sin 2θ [1 − sin θ/2]
− + Zαπ (4.206)
sinh 2ϕ 1 − β 2 sin2 θ/2 1 − β 2 sin2 θ/2
As we had said before the result is IR in the limit λ → 0. This divergence is not physical
and can be removed in the following way. The detectors have an energy threshold, below
which they can not detect. Therefore in the limit ω → 0 bremsstrahlung in a Coulomb
field and Coulomb scattering can not be distinguished. This means that we have to add
both results. If we consider an energy interval ∆E with λ ≤ ∆E ≤ E we get
Z
dσ dσ d3 k 2 2pi · pf m2 m2
(∆E) = e − −
dΩ BR dΩ Mott ω≤∆E 2ω(2π)3 ki · pi k · pf (k · p·)2 (k · pf )2
(4.207)
~ 2 2 1/2
Giving a mass to the photon (that is ω = (|k| + λ ) ) the integral can be done with the
result,

dσ dσ 2α 2∆E 1 1+β
(∆E) = (2ϕ coth 2ϕ − 1) ln + ln
dΩ BR dΩ Mott π λ 2β 1 − β
Z )
1
1 1 − β2 1 1 + βξ
− cosh 2ϕ dξ ln
2 β sin θ/2 cos θ/2 (1 − β 2 ξ 2 )[ξ − cos2 θ/2]1/2 1 − βξ
(4.208)
The inclusive cross section can now be written as

dσ dσ dσ
(∆E) = + (∆E)
dΩ dΩ elastic dΩ BR

dσ
= (1 − δR + δB ) (4.209)
dΩ Mott
where δR and δB are complicated expressions that depend on the resolution of the detector
∆E but do not depend on λ that can be finally put to zero. One can show that in QED all
the IR divergences can be treated in a similar way. One should note that the final effect
of the bremsstrahlung is finite and can be important.
Chapter 5
Functional Methods
5.1 Introduction
In this chapter, called Functional Methods, we are going to present the path integral
quantization. For systems that are not described by gauge theories this method may
seem unnecessary, as the canonical quantization works without problems. However, for
non-abelian gauge theories, as we shall see in the next chapter, this is the only known
method. Besides this fundamental point, the quantization done using functional methods
and the path integral formalism is very elegant and allows us to obtain the results much
faster, even for the cases where the canonical quantization works. Examples of this are
the Ward-Takahashi identities and the Dyson-Schwinger equations, as we will discuss at
the end of the chapter.
We are going to assume that the reader is familiar with the path-integral quantization
for systems of N particles in non-relativistic quantum mechanics. Therefore only a brief
sumamry of the results will be given. A more detailed account is given in Appendix A.
The step from the quantization of a system with N particles to the quantization of a field
theory will be done heuristically. A more rigorous treatment will be given in Appendix B.
Before we start, let us clarify some questions related with the notation. Let us assume
that we have real scalar field φa (x) where a = 1, ...N . In the following we will encounter
expressions of the type, Z
I1 = d4 x φa (x)φa (x) (5.1)
or Z
I2 = d4 xd4 y φa (x)M ab (x, y)φb (y) (5.2)
where M ab (x, y) is normally a differential operator. According to the rules for functional
derivation, we have,
δI1
= 2φb (y) (5.3)
δφb (y)
where we used the result
δφa (x)
= δab δ4 (x − y) (5.4)
δφb (y)
If we keep all the indices in the previous expressions (and in some much more compli-
cated that we will encounter soon), we will get a very complicated situation with respect
141
142 CHAPTER 5. FUNCTIONAL METHODS
to the notation. Therefore it will be useful to make use of a more compact notation. To
this end we identify,
φi ⇐⇒ φa (x) (5.5)
that is, the indice i will represent both the discrete indice a as weel as the continuous x,
i ⇐⇒ {a, x} (5.6)
In the case that the fields have further indices we will assume that i will always reprent
them collectively. We also use the Einstein convention meaing a sum for discrete indices
and an integration for continuous indices. With these conventions Eq. (5.1) and Eq. (5.4)
can be writtem as
I1 = φi φi I2 = φi Mij φj
(5.7)
δI1 = 2φ δφi
j = δij
δφj δφj
In the following we will use these conventions, returning to the more usual notation when
convenient or in case of a possible confusion.
5.2 Generating functional for Green’s functions

5.2.1 Green’s functions
Os objectos básicos em Teoria Quântica dos Campos são as funções de Green. Para evitar
complicações desnecessárias vamos aqui considerar quase exclusivamente campos escalares.
As generalizações são no entanto imediatas. Assim a função de Green de ordem n é dada
por
G(n) (x1 , . . . , xn ) ≡ h0|T φ(x1 ) · · · φ(xn )|0i . (5.8)

As funções de Green definidas por 5.8 são, por vezes, designadas completas por oposição
às chamadas funções de Green conexas, truncadas e irredutı́veis (ou próprias) que passamos
a definir.
5.2.2 Connected Green’s functions

Chamam-se funções de Green conexas aquelas em que nenhuma das linhas exteriores passa
através do diagrama sem interagir. Por exemplo na Figura 5.1 está representada uma
contribuição desconexa para G4 (x1 , . . . , x4 ), enquanto que a Figura 5.2 representa uma
contribuição conexa para a mesma função.
Está implı́cito que as funções de Green são calculadas em teoria das perturbações us-
(4)
ando diagramas de Feynman. Assim as funções de Green conexas Gc (x1 , . . . , xn ), são
obtidas somando todos os diagramas conexos. As funções de Green desconexas, correspon-
dendo aos diagramas desconexos, podem ser obtidas a partir de funções de Green conexas
de ordem mais baixa, pelo que as quantidades relevantes são as funções de Green conexas
Gnc (x1 , . . . , xn ). É claro que temos
Gnc (x1 , . . . , xn ) = Gn (x1 , . . . , xn ) − partes desconexas, (5.9)

5.2. GENERATING FUNCTIONAL FOR GREEN’S FUNCTIONS 143
x1 x2
x3 x4
Figure 5.1: Disconneted contribution to G4 (x1 , . . . , x4 ).
x1 x2
x3 x4
Figure 5.2: Connected contribution to G4 (x1 , . . . , x4 ).
e ainda
G2c (x1 , x2 ) = G2 (x1 , x2 ) . (5.10)

Convencionalmente representamos Gnc (x1 , . . . , xn ) pelo diagrama da Figura 5.3.
x1 x2
xn xi
Figure 5.3: Graphical representation for Gnc (x1 , . . . , xn ).
Por vezes interessa considerar as funções de Green no espaço dos momentos. Definimos
então Gnc (p1 , . . . , pn ) através da relação
(2π)4 δ4 (p1 + p2 + · · · pn ) Gnc (p1 , . . . , pn )

Z
≡ d4 x1 · · · d4 xn e−i(p1 ·x1+···+pn ·xn ) Gnc (x1 , . . . , xn ) , (5.11)
onde os momentos p1 , . . . , pn estão a entrar no diagrama (incoming momenta), conforme

indicado na Figura 5.4. Notar ainda que na definição 5.11 se factorizou a função delta
p1 p2
pn pi
Figure 5.4: Graphical representation for Gnc (p1 , . . . , pn )
que assegura a conservação de 4-momento. Com estas convenções G2 (p, −p) ≡ G2 (p) é o
propagador completo representado na Figura 5.5.
p −p
Figure 5.5: Full propagator G2 (p).
5.2.3 Truncated Green’s functions

Para n > 2 definem-se as funções de Green truncadas através da relação
n
Y −1 n
Gntrunc (p1 , . . . , pn ) = G2 (pk ) Gc (p1 , . . . , pn ) (5.12)
k=1
isto é, multiplica-se cada linha exterior pelo inverso do propagador completo referente a
essa linha. São estas funções que representam um papel fundamental na Teoria pois são
elas que estão relacionadas com os elementos da matriz S. De facto a fórmula de redução
LSZ para campos escalares é
hp1 , . . . , pn out|q1 , . . . , qℓ ini = hp1 , . . . , pn in|S|q1 , . . . , qℓ ini
= termos desconexos
" !#
n+ℓ Z n
X ℓ
X
−1/2 4 4
+ iZ d y1 · · · d xℓ exp i pk · y k − q k · xk
k=1 k=1
⊓y1 + m2 ) · · · (⊔
×(⊔ ⊓xℓ + m2 ) h0|T φ(y1 ) · · · φ(xℓ )|0ic (5.13)
5.2. GENERATING FUNCTIONAL FOR GREEN’S FUNCTIONS 145
ou seja
hp1 , . . . , pn out|q1 , . . . , qℓ ini = hp1 , . . . , pn in|S|q1 , . . . , qℓ ini
= termos desconexos
X X
n+ℓ
+Z −(n+ℓ)/2 (2π)4 δ pi − qj Gtrunc (−p1 , . . . , −pn , q1 , . . . , qℓ ) (5.14)
5.2.4 Irreducible diagrams

Vimos na Eq. 5.14 que os elementos da matriz S, relacionados com as secções eficazes,
são expressos em termos dos diagramas truncados. De entre os diagramas truncados
desempenha um papel importante o subconjunto dos diagramas próprios ou irredutı́veis
(em inglês 1-Particle Irreducible), que são os diagramas truncados que permanecem ligados
quando uma linha interna arbitrária é cortada. Por exemplo, o diagrama da Figura 5.61 é
truncado mas não é próprio enquanto que o diagrama da Figura 5.7 é próprio (na teoria
λφ3 ). Nestas figuras as barras indicam que as linhas exteriores estão truncadas.
Figure 5.6: Example of a truncated diagram that is not proper.
Figure 5.7: Example of a proper diagram.
A razão pela qual os diagramas truncados não irredutı́veis não são importantes é que
estes se podem escrever sempre em termos de diagramas irredutı́veis de ordem mais baixa
(recordar a série que conduz à definição de self-energy). É conveniente introduzir uma
notação para as funções de Green irredutı́veis (soma de todos os diagramas irredutı́veis
para determinado número de pernas exteriores) onde o factor i é introduzido por con-
veniência. Na Figura 5.8 as pernas externas estão desenhadas para tornar a figura mais
clara. Elas estão de facto truncadas. Dentro desta notação pode por vezes ser conveniente
1
As barras indicam que as linhas exteriores estão truncadas.
x1 x2
iΓ(n) (x1 , . . . , xn ) =
xn xi
Figure 5.8: Irreducible Green functions.
definir um diagrama para as funções de Green truncadas de ordem n. Este está repre-
sentado na Figura 5.9 ou doutra forma na Figura 5.10. Diagramas semelhantes podem-se
x1 x2 x1 x2
(n)
Gtrunc (x1 , . . . , xn ) ≡ ≡ =
xn xi xn xi
Figure 5.9: Graphical representation of truncated Green functions.
definir no espaço dos momentos.
5.3 Generating functionals for Green’s functions

Os funcionais geradores (FG) das funções de Green representam um papel muito impor-
tante em teoria quântica dos campos. De facto a partir deles, por derivação funcional em
relação a fontes exteriores podem-se obter todas as funções de Green. Permitem assim
tratar ao mesmo tempo um número infinito de funções de Green. O FG das funções de
Green completas é dado por
D E
Z(J) ≡ 0|T eiJi φi |0 , (5.15)
onde estamos a usar a notação condensada explicada na introdução

5.3. GENERATING FUNCTIONALS FOR GREEN’S FUNCTIONS 147
x1 x2 x1 x2
xn xi xn xi
Figure 5.10: Another representation of truncated Green functions.
Z
Ji φi ≡ d4 x J(x)φ(x) . (5.16)
É fácil de ver que Z(J) gera todas as funções de Green. Se expandirmos a exponencial
em 5.15 obtemos
P∞ in
Z(J) = n=0 n! Ji1 · · · Jin h0|T φi1 · · · φin |0i
P∞ in
= n=0 n! Ji1 · · · Jin Gni1 ···in (5.17)
As funções de Green são então dadas por

δn Z
Gni1 ···in = (5.18)
iδJi1 · · · iδJin Ji =0
O F G das funções de Green conexas é definido através da relação
Z(J) = eiW (J) (5.19)

ou ainda
W (J) = −i ln Z(J) . (5.20)

As funções de Green conexas são então obtidas por derivação funcional

δn W
n
Gc i1 ···in = i (5.21)
iδJi1 · · · iδJin Ji =0
Antes de mostrarmos que isto é de facto verdade, vamos definir o funcional gerador das
funções de Green irredutı́veis. Este é definido através da primeira transformada da Leg-
endre de W (J), isto é
Γ(φ) ≡ W (J) − Ji φi (5.22)

onde

 φi ≡ δW (J)

δJi
(5.23)

 Ji = − δΓ(φ)
δφi
As funções de Green próprias são então dadas por

δn Γ(φ)
Γni1 ···in = . (5.24)
δφi1 · · · δφin φ=0
Dadas as definições falta-nos agora mostrar que W (J) e Γ(φ) geram efectivamente as
funções de Green conexas e próprias. Comecemos por W (J). A demonstração faz-se
calculando Gnc i1 ···in dada por 5.21 e usando a relação 5.20. Vamos fazer somente para
n = 2, n = 3 e n = 4. As generalizações são imediatas.
• n=2

δ2 W δ2 ln Z δ 1 δZ
G2c i1 i2 = i = =
iδJi1 iδJi2 Ji =0 iδJi1 iδJi2 Ji =0 iδJi1 Z iδJi2 Ji =0

1 δ2 Z 1 δZ 1 δZ
= −
Z iδJi1 iδJi2 Ji =0 Z iδJi1 Z iδJi2 Ji =0

δ2 Z
= (5.25)
iδJi1 iδJi2 Ji =0
ou seja
G2c i1 i2 = G2i1 i2 (5.26)

Para se obter 5.25 fez-se uso dos seguintes resultados
Z(0) = 1 O vácuo está normalizado
δZ
= h0|T φi |0i = 0 Não há quebra de simetria (5.27)
iδJi
• n=3

1 δ3 Z 1 δ2 Z 1 δZ
G3c i1 i2 i3 = −
Z iδJi1 iδJi2 iδJi3 Z iδJi1 iδJi2 Z iδJi3
1 δ2 Z 1 δZ 1 δ2 Z 1 δZ
− −
Z iδJi2 iδJi3 Z iδJi1 Z iδJi1 iδJi3 Z iδJi2

1 δZ 1 δZ 1 δZ
+2 (5.28)
Z iδJi1 Z iδJi2 Z iδJi3 |J =0
i
logo
5.3. GENERATING FUNCTIONALS FOR GREEN’S FUNCTIONS 149
G3i1 i2 i3 = G3c i1 i2 i3 (5.29)

O caso n = 4 é deixado como problema (ver Problema 1.1) A extensão a n > 4 é ime-
diata. Mostrámos assim que W (J) dado por 5.20 é o funcional gerador das funções de
Green conexas. Mostremos agora que Γ(φ) é o funcional gerador das funções de Green
próprias, ou irredutı́veis. Para isto necessitamos de dois resultados prévios que passamos
a demonstrar. O primeiro baseia-se na relação
δJi
= δik (5.30)
δJk
Esta relação é evidente mas podemos obter a partir dela uma relação importante. De facto
δJi δJi δφℓ δ2 Γ δ2 W

= =− = −iΓiℓ Gℓk (5.31)
δJk δφℓ δJk δφi δφℓ δJℓ δJk
ou ainda
Γiℓ Gℓk = iδik (5.32)

Esta relação fundamental exprime que Γ2 é o inverso do propagador G2 (à parte o factor
i que tem que ver com convenções). É também útil escrevê-la duma forma diagramática
conforme indicado na Figura 5.11.
= −1
Figure 5.11: Graphical representation of Eq.5.32.
Notar que
(2) i k
iΓik ≡ (5.33)
o que explica o desaparecimento do i na equação 5.32.

O segundo resultado diz respeito à seguinte derivada funcional
δ
. (5.34)
iδJi
Pretende-se derivar em ordem a Ji quantidades que dependem de Ji indirectamente através
de φk (ver definições 5.23). Obtemos então
δ δφk δ δ2 W δ (2) δ
= = = Gik (5.35)
iδJi iδJi δφk δJk iδJi δφk δφk
e portanto
δ δ
= Gik (5.36)
iδJi δφk
As equações 5.32 e 5.36 permitem obter todas as relações entre as funções de Green próprias
e as funções de Green conexas. Esta análise é mais fácil em termos de diagramas desde
que se notem as seguintes identidades
m
δ
k m k (5.37)
iδJi =
j
δ
i i (5.38)
δφk =
e
i
m
δ
k j k j k j (5.39)
iδJi = Gim δ =
δφm
onde se usou 5.36 para estabelecer 5.39. Em todas estas manipulações está subentendido
que no final se faz J = 0 e φ = 0 nos sı́tios convenientes. Usemos agora estes métodos
para relacionar as funções de Green próprias e conexas para n = 3 e n = 4.
• n=3
δ
O ponto de partida é a equação 5.32. Aplicamos iδJℓ a 5.32 e obtemos
δ
i =k0 (5.40)
iδJℓ =0
Usando as equações 5.37 e 5.39 obtemos então o diagrama da Figura 5.12. Multiplicando
(2)
à esquerda por Gmi e usando 5.32 obtemos
5.4. FEYNMAN RULES 151
k
i i k
+ =0
Figure 5.12: Graphical result of Eq.5.40.
k
(3) m m
iΓmkl ≡ =
(5.41)
l
l
(3)
o que mostra que Γmkl é de facto a função de Green própria com 3 pernas exteriores porque
para 3 pernas as funções próprias e truncadas coincidem. Para se ver que se trata de facto
de funções próprias e não somente truncadas, é preciso ir para n = 4 pois aı́ é que começa
a diferença.
• n=4
Partimos da equação 5.41 e derivamos em relação a iδJδ n . Usando os métodos anteriores
(3)
obtemos a equação representada na Figura 5.13. Se usarmos 5.41 para expressar Gkml em
(3)
termos de Γkml obtemos a equação diagramática da Figura 5.14 que mostra claramente
(4)
que Γmkln é de facto a função de Green própria ou irredutı́vel de 4 pernas.
• n>4
É agora trivial continuar o processo para n > 4. Para um dado n parte-se da relação
para n − 1 e aplicam-se as equações 5.37 e 5.39. Estes resultados mostram de facto que os
objectos mais importantes são as funções de Green irredutı́veis, todas as outras se podem
obter a partir delas. Isto é um resultado importante porque reduz imenso o número de
diagramas de Feynman que têm que ser calculados.
5.4 Feynman rules

The formalism of functional generators allows us to obtain the Feynman rules of any theory
with all the correct conventions. We have already shown how to get the Feynman rules in
m m k k
k
m
= +
n n
n l l l
k
k
n
m m
+
l l
n
Figure 5.13: Graphical representation of Gmkln in terms of Γmkln .
section 3.7. There we used the result that in lowest order (tree level) we have
Z
Γtree (φ) = d4 xL[φ] ≡ Γ0 (φ) (5.42)
Here we are going just to show this result. For the interaction terms (n > 2) this is clear.
For instance for n = 3 we have
(3)
iΓ(3) = Gtree (5.43)
while for n = 4 we get
iΓ(4) = G(4) − irreducible parts (5.44)
(4)
and it is obvious that iΓtree generates the vertices.
The i factor is in agreement
R 4 with the usual conventions for the Feynman rules as it
comes from the term exp i d xLint in the calculation of the Green functions. For the
quadratic terms we have, Eq. 5.32,
(2)
Γtree (p) = p2 − m2 (5.45)
therefore, doing the inverse Fourier transform,

Z
(2) d4 p1 d4 p2 i(p1 ·x+p2·y)
Γtree (x, y) = e (2π)4 δ4 (p1 + p2 ) (p21 − m2 ) (5.46)
(2π)4 (2π)4
and
Z
1 (2)
d4 xd4 y φ(x)Γtree (x, y)φ(y) =
2
Z
1 d4 p1 d4 p2 i(p1 ·x+p2·y)
= d4 xd4 y e (2π)4 δ(p1 + p2 )(p21 − m2 )φ(x)φ(y)
2 (2π)4 (2π)4
5.5. PATH INTEGRAL FOR GENERATING FUNCTIONALS 153
m k
m m k
k
= +
n l n l
n l
m k
m k
+ +
n l n l
Figure 5.14: Graphical representation of Gmkln in terms of Γmkln .
Z Z
1 1
= d4 x φ(x)(−⊔
⊓ − m2 )φ(x) = d4 x ∂µ φ∂ µ φ − m2 φ2 (5.47)
2 2
which shows that Γtree is in fact the action. In getting to Eq. (5.47) we have done an
integration by parts and discarded, usual, the boundary term. We refer the reader to
Section 3.7 for the actual recipes on how to determine the Feynman rules of any theory.
5.5 Path integral for generating functionals

5.5.1 Quantum mechanics of n degrees of freedom
Comecemos por recordar os resultados conhecidos para sistemas com 1 grau de liber-
dade. No Apêndice A faz-se uma introdução à quantificação via integral de caminho. Lá
poderão ser encontradas as justificações para os resultados que usaremos no seguimento.
O resultado fundamental é para a amplitude de transição
Z Z

R t′
q ′ ; t′ |q; t = N D(q)ei t
dtL(q,q̇)
=N D(q)eiS (5.48)
onde N é um factor de normalização e D(q) é uma forma simbólica de representar a

medida de integração que é de facto um limite complicado (ver Apêndice A). Outro
resultado importante diz respeito aos elementos de matriz do produto ordenado no tempo
de operadores. Seja
O(t1 , . . . , tn ) = T [O1H (t1 )O2H (t2 )...OnH (tn )] (5.49)
tal que
t′ ≥ (t1 , t2 , . . . , tn ) ≥ t (5.50)
Então
Z

′ ′

q ; t O(t1 , . . . , tn ) |q; ti = N D(q)O1 (q(t1 )) · · · On (q(tn ))eiS (5.51)
onde se admitiu que os operadores Oi são diagonais no espaço das coordenadas. Para
a generalização à Teoria do Campo os objectos importantes não são as amplitudes de
transição mas as funções de Green e os seus funcionais geradores. Consideremos por
exemplo a função de Green
G(t1 , t2 ) ≡ h0| T (QH (t1 )QH (t2 )) |0i (5.52)

onde |0i é o estado de base e QH (t) é o operador coordenada na representação de Heisen-
berg. Para escrevermos a Eq. (5.52) em termos dum integral de caminho introduzimos
conjuntos completos de estados e escrevemos
Z

G(t1 , t2 ) = dq dq ′ 0|q ′ ; t′ q ′ ; t′ T (QH (t1 )QH (t2 )) |q; ti hq; t|0i
Z Z R t′
= dqdq ′ φ0 (q ′ , t′ )φ∗0 (q, t) D(q)q(t1 )q(t2 )ei t Ldτ (5.53)
onde
φ0 (q, t) = h0|q; ti = φ0 (q)e−iE0 t (5.54)

A presença na expressão 5.53 das funções onda do estado base torna a expressão pouco
prática. Podemos removê-los do modo seguinte. Consideremos o elemento de matriz

q ′ ; t′ O(t1 , t2 ) |q; ti
Z

= dQ dQ′ q ′ ; t′ |Q′ ; T ′ Q′ ; T ′ O(t1 , t2 ) |Q; T i hQ; T |q; ti (5.55)
onde
O(t1 , t2 ) = T (QH (t1 )QH (t2 ))
t′ ≥ T ′ ≥ (t1 , t2 ) ≥ T ≥ t (5.56)
Sejam |ni os estados próprios com energia En e função de onda φn (q) isto é
H |ni = En |ni
hq|ni = φ∗n (q) (5.57)
Então

′ −iH(t′ −T ′ ) ′
q ′ ; t′ |Q′ ; T ′ = q e Q
X
′ ′
= q ′ |n hn| e−iH(t −T ) Q′
n
X ′ ′)
= φ∗n (q ′ )φn (Q′ )e−iEn (t −T (5.58)
n
Consideremos agora o limite t′ → −i∞. Então

′ ′
lim q ′ ; t′ |Q′ ; T ′ = φ∗0 (q ′ )φ0 (Q′ )e−E0 |t | eiE0 T (5.59)
t′ →−i∞
De modo semelhante
lim hQ; T |q; ti = φ0 (q)φ∗0 (Q)e−E0 |t| e−iE0 T (5.60)

t→i∞
Aplicando estes limites à Eq. (5.55) obtemos

lim lim q ′ ; t′ O(t1 t2 ) |q; ti
t′ →−i∞ t→i∞
Z
′ ′
= dQdQ′ φ∗0 (q ′ )φ0 (Q′ )e−E0 |t | eiE0 T

Q′ ; T ′ O(t1 , t2 ) |Q; T i φ0 (q)φ∗0 (Q)e−E0 |t| e−iE0 T
′
= φ∗0 (q ′ )φ0 (q)e−E0 |t | e−E0 |t|
Z

dQdQ′ φ0 (Q′ , T ′ )φ∗0 (Q, T ) Q′ ; T ′ O(t1 , t2 ) |Q; T i (5.61)
Usando 5.53 obtemos o resultado importante

′
lim lim q ′ ; t′ O(t1 t2 ) |q; ti = φ∗0 (q ′ )φ0 (q)e−E0 |t | e−E0 |t| G(t1 , t2 ) (5.62)
t′ →−i∞ t→i∞
Por outro lado

′
′
lim lim q ′ ; t′ |q; t = φ∗0 (q ′ )φ0 (q)e−E0 |t | eE0 |t| (5.63)
t →−i∞ t→i∞
pelo que finalmente podemos escrever

hq ′ ; t′ | T (QH (t1 )QH (t2 )) |q; ti
G(t1 , t2 ) = lim lim (5.64)
t′ →−i∞ t→i∞ hq ′ ; t′ |q; ti
Usando agora a expressão 5.51 podemos finalmente escrever G(t1 , t2 ) em termos dum
integral de caminho
Z
1 R ′
i tt Ldτ
G(t1 , t2 ) = ′ lim lim D(q)q(t 1 )q(t 2 )e (5.65)
t →−i∞ t→i∞ hq ′ ; t′ |q; ti
Este resultado é facilmente generalizado para funções de Green com n-pontos,
G(t1 , . . . , tn ) = h0| T (q(t1 ) · · · q(tn )) |0i

Z
1 R ′
i tt Ldτ
= ′ lim lim D(q)q(t 1 ) · · · q(t n )e (5.66)
t →−i∞ t→i∞ hq ′ ; t′ |q; ti
É agora fácil de ver que todas as funções de Green podem ser obtidos a partir do
funcional gerador
Z
1 R ′
i tt [L(q,q̇)+Jq]dτ
Z[J] = ′ lim lim D(q)e (5.67)
t →−i∞ t→i∞ hq ′ ; t′ |q; ti
por derivação funcional

δn Z[J]
G(t1 , . . . , tn ) = (5.68)
iδJ(t1 ) · · · iδJ(tn ) J=0
A expressão 5.67 para o funcional gerador mostra que ele é a amplitude de transição
entre o estado base no instante t e o estado base no instante t′ na presença duma fonte
exterior
Z[J] = h0|0iJ (5.69)

com a normalização Z[J = 0] = 1.
Para um sistema com N graus de liberdade temos a generalização de 5.67
Z R t′ PN
Z[J1 , . . . , Jn ] = ′ lim lim N D(qi )ei t dτ [L(qi ,q̇i )+ i=1 Ji qi ] (5.70)
t →−i∞ t→i∞
Notas
• Na equação anterior os limites nos tempos t e t′ são imaginários. Isto quer dizer
que as funções de Green bem definidas são as funções de Green Euclidianas. Para a
teoria de campos isto corresponde à prescrição m2 − iǫ.
• Na equação 5.70 não escrevemos explicitamente a normalização. Ela é obviamente

escolhida para que Z[0, . . . , 0] = 1 mas como veremos, para as funções de Green
conexas em teoria dos campos a normalização não é relevante pelo que não nos
vamos preocupar mais com ela.
5.5.2 Field theory

Para obter o funcional gerador das funções de Green em Teoria dos Campos procedemos
da forma heurı́stica usual
t → xµ
q(t) → φ(x)
D(q) → D(φ)
Z
L(qi , q̇i ) → d3 xL(φ, ∂µ φ) (5.71)
Então obtemos
Z
d4 x[L(φ,∂µ φ)+J(x)φ(x)]
R
Z[J] = N D(φ)ei (5.72)
O limite em 5.70 recorda-nos que os integrais têm que se continuar analiticamente para
o espaço euclidiano ou equivalentemente que se tem que fazer a prescrição m2 − iǫ.
A expressão 5.66 é a expressão fundamental que procurávamos para o funcional gerador
das funções de Green completas. É fácil de ver que para o funcional gerador das funções
de Green conexas
W (J) = −i ln Z(J) (5.73)

a normalização é irrelevante (pois não depende de J). Uma demonstração mais rigorosa
da expressão 5.72 pode ser encontrada no Apêndice B.
5.5.3 Applications
Uma vez conhecido o funcional gerador Z(J) são conhecidas todas as funções de Green
e portanto qualquer problema em Teoria dos Campos. Pode-se então perguntar em que
condições é possı́vel calcular 5.72. Como só se sabem fazer exactamente integrais gaus-
sianos a resposta é que só se pode calcular 5.66 em situações triviais, sem interacções no
Lagrangeano. Contudo as vantagens de 5.72 resultam de dois aspectos:
• Manipulações formais
Relações entre funções de Green que tenham a ver com propriedades de simetria
(identidades de Ward) são muito facilmente deduzidas por manipulações dos fun-
cionais geradores. Aqui a forma 5.66 é particularmente útil como veremos nas secções
4 e 5.
• Teoria de perturbações
A expressão 5.72 permite imediatamente desenvolver a teoria de perturbações.
Como exemplo do ponto ii) consideremos o Lagrangeano para um campo escalar que
por simplicidade tomaremos real,
L(φ) = L0 (φ) + LI (φ) (5.74)

onde L0 (φ) é quadrático, isto é
1 1
L0 (φ) = ∂µ φ∂ µ φ − m2 φ2 (5.75)
2 2
Então podemos escrever
Z
d4 x[L0 (φ)+LI (φ)+Jφ]
R
Z[J] = N D(φ)ei
Z
d4 x[LI (φ)] d4 x[L0 (φ)+Jφ]
R R
= N D(φ)ei ei (5.76)
ou seja
Z
4 δ
Z[J] = exp i d xLI Z0 [J] (5.77)
iδJ
onde
Z
d4 x[L0 +Jφ]
R
Z0 [J] = N D(φ)ei . (5.78)
A utilidade desta expressão resulta do facto de que por um lado Z0 [J] pode ser cal-
culado exactamente, porque é quadrático nos campos, e por outro se LI (φ) tiver um
parâmetro pequeno a exponencial pode ser desenvolvida em série nesse parâmetro e o
funcional gerador Z[J] obtido até à ordem que se pretender em teoria das perturbações.
5.5.4 Example: perturbation theory for λφ4

Para se ver a ligação com os resultados usuais vamos considerar como exemplo a teoria
dum campo escalar em que a interacção é
λ 4
LI = − φ . (5.79)
4!
O funcional gerador Z[J] é dado por
( Z 4 )
1 δ
Z[J] = N exp (−iλ) d4 x Z0 [J] (5.80)
4! iδJ
onde (ver problema 1.2)

Z
1 4 4
Z0 [J] = exp − d xd yJ(x)∆(x, y)J(y) (5.81)
2
A normalização N é escolhida para que Z[0] = 1, como veremos adiante. Desenvolvemos
a exponencial em série na constante de acoplamento:

Z[J] = N Z0 [J] 1 + (−iλ)Z1′ [J] + (−iλ)2 Z2′ [J] + · · · (5.82)
onde
( Z 4 )
1 δ
Z1′ [J] ≡ Z0−1 [J] d x4
Z 0 [J] (5.83)
4! δJ
e
( Z 4 ) 2
1 −1 1 δ
Z2′ [J] ≡ Z [J] d4 x Z 0 [J]
2 0 4! δJ
( Z 4 )
1 −1 1 δ
= Z [J] d4 x (Z0 Z1′ )
2 0 4! δJ
Z
1 ′ 2 1 −1 1 4 δ3 Z0 δZ1′
= Z1 [J] + Z0 [J] d x 4
2 2 4! δJ 3 (x) δJ(x)

δ2 Z0 δ2 Z1′ δZ0 δ3 Z1′ δ4 Z1′
+6 2 + 4 + Z0 (5.84)
δJ (x) δJ 2 (x) δJ(x) δJ 3 (x) δJ 4 (x)
Obtemos
Z1′ [J] =
Z Z
1 4
= d x 3∆(x, x)∆(x, x) − 3!∆(x, x) d4 y1 d4 y2 ∆(x, y1 )∆(x, y2 )J(y1 )J(y2 )
4!
Z
4 4
+ d y1 · · · d y4 ∆(x, y1 )∆(x, y2 )∆(x, y3 )∆(x, y4 )J(y1 )J(y2 )J(y3 )J(y4 )
(5.85)
Este resultado pode ser representado diagramaticamente na forma seguinte
1 1 1
Z1′ = − + (5.86)
8 4 4!
Para Z2′ virá
Z2′ [J]
Z
1 ′ 2 1 1 2
= Z1 [J] + 4! d4 x1 d4 x2 ∆(x1 , x2 )∆(x1 , x2 )∆(x1 , x2 )∆(x1 , x2 )
2 2 4!
2 Z Z Z
1
+ −72 d x2 d x1 ∆(x1 , x2 ) d4 y1 ∆(x1 , y1 )J(y1 )
4 4
4!
Z
∆(x1 , x2 )∆(x2 , x2 ) d4 y2 ∆(x2 , y2 )J(y2 )
Z Z
4 4
+24 d x2 d x1 ∆(x1 , x1 ) d4 y1 ∆(x1 , y1 )J(y1 )∆(x1 , y2 )
Z Z Z
d4 y2 ∆(x2 , y1 )J(y2 ) d4 y3 ∆(x2 , y3 )J(y2 ) d4 y4 ∆(x1 , y4 )J(y4 )
Z Z Z
+24 d4 x2 d4 x1 d4 y1 d4 y2 d4 y3 d4 y4 ∆(x1 , x2 )∆(x1 , y1 )
∆(x1 , y2 )∆(x1 , y2 )∆(x2 , x2 )∆(x2 , y4 )J(y1 ) · · · J(y4 )

Z Z Z
4 2 4
−8 d x d x1 d4 y1 · · · d4 y6 ∆(x1 , x2 )∆(x1 , y1 )∆(x1 y2 )
∆(x1 , y3 )∆(x2 , y4 )∆(x2 , y5 )∆(x2 , y6 )J(y1 ) · · · J(y6 )

Z
+36 d4 x2 d4 x1 ∆(x1 , x1 )∆(x1 , x2 )∆(x1 , x2 )∆(x2 , x2 )
Z Z
4 4
−36 d x2 d x1 ∆(x1 , x1 )∆(x1 , x2 )∆(x1 , x2 ) d4 y1 d4 y2 ∆(x2 , y1 )
∆(x2 , y2 )J(y1 )J(y2 )

Z
−36 d4 x2 d4 x1 d4 y1 d4 y2 ∆(x1 , x2 )∆(x1 , x2 )∆(x2 , x1 )
∆(x1 , y1 )∆(x1 , y2 )J(y1 )J(y2 )

Z
+36 d4 x2 d4 x1 d4 y1 · · · d4 y4 ∆(x1 , x2 )∆(x1 , x2 )∆(x1 , y1 )
∆(x1 , y2 )∆(x2 , y3 )∆(x2 , y4 )J(y1 ) · · · J(y4 )

Z
−48 d4 x2 dy1 d4 y1 d4 y2 ∆(x1 , x2 )∆(x1 , x2 )∆(x1 , x2 )

∆(x1 , y1 )∆(x2 , y2 )J(y1 )J(y2 ) (5.87)
ou seja:
Z2′ [J]
Z
1 ′ 2 1
= Z [J] + d4 x1 d4 x2 ∆4 (x1 , x2 )
2 1 2 · 4!
Z
3
+ d4 x1 d4 x2 ∆(x1 , x1 )∆2 (x1 , x2 )∆(x2 , x2 )
2 · 4!
Z
1
− d4 x1 d4 x2 d4 y1 · · · d4 y6 ∆(y1 , x1 )∆(y2 , x1 )∆(y3 , x1 )
2 · 3! · 3!
∆(x1 , x2 )∆(x2 , y4 )∆(x2 , y5 )∆(x2 , y6 )J(y1 ) · · · J(y6 )

Z
2
+ d4 x1 d4 x2 d4 y1 · · · d4 y4 ∆(y1 , x1 )∆(x1 , x1 )∆(x1 , x2 )
4!
∆(x2 , y2 )∆(x2 , y3 )∆(x2 , y4 )J(y1 ) · · · J(y4 )

Z
3
+ d4 x1 d4 x2 d4 y1 · · · d4 y4 ∆(y1 , x1 )∆(y2 , x1 )∆2 (x1 , x2 )
2 · 4!
∆(x2 , y3 )∆(x2 , y4 )J(y1 ) · · · J(y4 )

Z
1
− d4 x1 d4 x2 d4 y1 d4 y2 ∆(y1 , x1 )∆(x1 , x1 )∆(x1 , x2 )∆(x2 , x2 )
8
∆(x2 , y2 )J(y1 )J(y2 )

Z
1
− d4 x1 d4 x2 d4 y1 d4 y2 ∆(y1 , x1 )∆2 (x1 , x2 )∆(x2 , x2 )∆(x1 , y2 )J(y1 )J(y2 )
8
Z
1
− d4 x1 d4 x2 d4 y1 d4 y2 ∆(y1 , x1 )∆3 (x1 , x2 )∆(x2 , y2 )J(y1 )J(y2 ) (5.88)
12
Vamos agora calcular a normalização até à segunda ordem em teoria de perturbações.
Para isso a condição Z[0] = 1 dá:

1 = N 1 + (−iλ)n1 + (−iλ)2 n2 + · · · (5.89)
onde
1
n1 = (5.90)
8
1 2 1 3
n2 = n1 + + (5.91)
2 2 · 4! 2 · 4!
Obtemos portanto
1
N =
1 + (−iλ)n1 + (−iλ)2 n2 + · · ·
= 1 − (−iλ)n1 − (−iλ)2 (n2 − n21 ) + · · · (5.92)
Então

Z[J] = Z0 [J] 1 − (−iλ)n1 − (−iλ)2 (n2 − n21 ) + · · ·

1 + (−iλ)Z1′ + (−iλ)2 Z2′ + · · ·

= Z0 [J] 1 + (−iλ)(Z1′ − n1 ) + (−iλ)2 (Z2′ − n2 + n21 − n1 Z1′ ) + · · · (5.93)
Definindo agora
Z1 ≡ Z1′ − n1
Z2 ≡ Z2′ − n2 + n21 − n1 Z1′ = Z2′ − n2 − n1 Z1 (5.94)
obtemos
1 1
Z1 [J] = − + (5.95)
4 4!
Z2 [J]
1 −1
= (Z1 [J])2 +
2 2! 3! 3!
3 2
+ +
2! 4! 4!
1 1 1
− − − (5.96)
8 8 12
com Z1 [0] = Z2 [0] = 0. Logo o funcional gerador


Z[J] = Z0 [J] 1 + (−iλ)Z1 [J] + (−iλ)2 Z2 [J] + · · · (5.97)
é automaticamente normalizado se desprezarmos todas as amplitudes vácuo-vácuo, de-
signadas por bubles. Para verificarmos que a expressão anterior reproduz os resultados
da teoria de perturbações usual calculemos como exemplo o propagador até à ordem λ2 .
Obtemos

δ2 Z[J]
′
∆ (x1 , x2 ) =
iδJ(x1 )iδJ(x2 ) J=0

δ2 Z0 [J] δ2 Z1 [J] 2 δ2 Z2 [J]
=− − (−iλ) − (−iλ)
δJ(x1 )δJ(x2 ) J=0 δJ(x1 )δJ(x2 ) J=0 δJ(x1 )δJ(x2 ) J=0
Z
1
= ∆(x1 , x2 ) + (−iλ) d4 y∆(x1 , y)∆(x2 , y)∆(y, y)
2
Z
2 4 4 1
+(−iλ) d y1 d y2 ∆(x1 , y1 )∆(y1 , y1 )∆(y1 , y2 )∆(y2 , y2 )∆(y2 , x2 )
4

1 2 1 3
+ ∆(x1 , y1 )∆ (y1 , y2 )∆(y2 , y2 )∆(y1 , x2 ) + ∆(x1 , y1 )∆ (y1 , y2 )∆(y2 , x2 ) (5.98)
4 6
Em termos diagramáticos temos a situação da Figura 5.15
1
= +
x1 x2 x1 x2 2 x1 x2
1 1 1
+ + +
4 x1 x2 4 x1 x2 6 x1 x2
Figure 5.15:
Continuando a estudar o exemplo da teoria λφ4 passemos a analisar o funcional gerador

das funções de Green conexas W [J]. É fácil de ver que termos do tipo Z12 [J] correspondem
a diagramas disconexos contidos em Z[J]. Vamos ver que elas desaparecem no funcional
W [J]. Temos
iW [J] = ln Z[J] =

= ln Z0 [J] + ln 1 + (−iλ)Z1 [J] + (−iλ)2 Z2 [J] + · · ·
1
= iW0 [J] + (−iλ)Z1 [J] − (−iλ)2 (Z1 [J])2 + (−iλ)2 Z2 [J] + · · ·
2

2 1 2
= iW0 [J] + (−iλ)Z1 [J] + −iλ) (Z2 [J] − (Z1 [J]) + · · ·
2

≡ i W0 [J] + (−iλ)W1 [J] + (−iλ)2 W2 [J] + · · · (5.99)
com
iW1 [J] = Z1 [J]
1
iW2 [J] = Z2 [J] − (Z1 [J])2 (5.100)
2
Portanto os diagramas disconexos contidos em Z2 [J] são subtraı́dos e W1 e W2 contém
somente diagramas conexos como seria de esperar.
5.5.5 Symmetry factors

Depois de efectuar as derivadas em ordem a J para obter numa dada função de Green os
números que resultam são os chamados factores de simetria. Por exemplo para a correcção
a 1-loop ao propagador obtemos

δ2 Z
′
∆ (x1 , x2 ) = =
iδJ(x1 )iδJ(x2 ) J=0

δ 2 Z0 δ2 Z1
= + (−iλ) + ···

iδJ(x1 )iδJ(x2 ) J=0 iδJ(x1 )iδJ(x2 ) J=0
1
= ∆(x1 , x2 ) + + ··· (5.101)
2
O factor 21 é o factor de simetria correspondente àquele diagrama. Como vimos o método

de obter as funções de Green a partir do funcional gerador automaticamente dá os estes
factores correctos. Contudo na maior parte das aplicações é mais fácil aplicar directamente
as regras de Feynman e então uma regra para os factores de simetria deve ser fornecida.
Regra para os factores de simetria:
O factor de simetria S dum dado diagrama é dado por
N
S= (5.102)
D
onde N é o # de maneiras diferentes de formar o diagrama e D é o produto dos factores

de simetrias de cada vértice e do número de permutações de vértices iguais.
Como exemplo consideremos o diagrama que contribui para o propagador a 1-loop,
representado na Figura 5.16.
p p
Figure 5.16:
Então de acordo com a regra obtemos
4×3 1
S= = (5.103)
4! 2
5.5.6 A comment on the normal ordering

No exemplo anterior diagramas como o da Figura 5.16, designados por tadpoles, apare-
cem enquanto que no formalismo canónico usual estão excluı́dos devido ao ordenamento
normal. Esta diferença deve-se ao facto de não termos sido muito rigorosos na definição
do integral de caminho. Se o tivéssemos Rfeito chegarı́amos à conclusão que para obter a
4
expressão do Lagrangeano a incluir em ei d xL(φ) terı́amos primeiro que ordenar normal-
mente o Lagrangeano Quântico e só então efectuar a transcrição para os campos clássicos
do integral de caminho. Isto faz com que o Lagrangeano L(φ) usado no integral de caminho
seja diferente do Lagrangeano para a teoria clássica. Vejamos o exemplo de φ4 . Vamos
usar as relações2
φ̂(x)φ̂(x) =: φ̂(x)φ̂(x) : + h0| φ̂(x)φ̂(x) |0i (5.104)

ou, mais simbolicamente,
: φ̂2 := φ̂2 − h0| φ̂2 |0i . (5.105)

De modo semelhante
φ̂4 =: φ̂4 : +6 : φ̂2 : h0| φ̂2 |0i + 6 h0| φ̂2 |0i h0| φ̂2 |0i (5.106)
e portanto obtemos
2
: φ̂4 := φ̂4 − 6 : φ̂2 (x) : h0| φ̂2 (x) |0i − 6 h0| φ̂2 |0i (5.107)
ou ainda
2
Usaremos nesta secção a notação φ̂ para o campo quântico (operador) para o distinguir do campo
clássico φ.
: φ̂4 := φ̂4 − 6 φ̂2 h0| φ̂2 |0i (5.108)

Isto quer dizer que o Lagrangeano quântico se escreve
λ
LQ
Int = − : φ̂4 :
4!
λ λ
= − φ̂4 + φ̂2 I (5.109)
4! 4
onde
I ≡ h0| φ̂2 (x) |0i

Z
= ˜ 1 dk
dk ˜ 2 h0| a† (k1 )eik1 ·x + a(k1 )e−ik1 ·x a† (k2 )eik2 ·x + a(k2 )e−ik2 ·x |0i
Z
= ˜ 1 dk
dk ˜ 2 h0| a(k1 )a† (k2 ) |0i ei(k2 −k1 )·x
Z Z
˜1 = d3 k 1
= dk (5.110)
(2π)3 2ωk
Na expressão anterior usámos as relações

a(k), a+ (k′ ) = (2π)3 2ωk δ3 (~k − ~k′ ) (5.111)
q
ωk = k02 + |~k|2 (5.112)
O integral I é divergente e é igual ao integral do loop da Figura 5.16. De facto
Z
d4 k i
(2π) k − m2 + iε
4 2
Z Z +∞
d3 k i
= dk0
(2π)3 −∞ (k0 − ωk )(k0 + ωk )
Z
d3 k 1
= =I (5.113)
(2π)3 2ωk
Portanto se tivéssemos sido cuidadosos terı́amos que incluir o termo λ4 φ2 I na in-

teracção. O Lagrangeano a introduzir na exponencial do integral de caminho seria então
λ 4 λ 2
LIC
Int = − φ + φ I (5.114)
4! 4
É fácil de ver que o termo adicional cancela o tadpole. De facto temos
+ =
1/2

i λ λ i
= 2 2
−i I + i I
p −m 2 2 p − m2
2
= 0 (5.115)
e portanto os tadpoles não apareceriam. Contudo muitas vezes não nos preocupamos
em usar o Lagrangeano correcto e usamos simplesmente o Lagrangeano clássico pois a
contribuição do tadpole é uma renormalização da massa (infinita) e pode assim ser sempre
reabsorvida no processo de renormalização.
Para QED o mesmo se passa, isto é, devı́amos usar como Lagrangeano de interacção
LIC µ µ
Int = −eψγ ψAµ + eAµ h0| ψγ ψ |0i (5.116)
e o segundo termo removeria o tadpole representado na Figura 5.17,
Figure 5.17: Tadpole for QED.
No entanto, devido à invariância de Lorentz da teoria, pode-se mostrar que este tadpole é
zero em todas as ordens e portanto não nos temos que preocupar.
5.5.7 Generating functionals for fermions

Para sistemas de fermiões introduzimos variáveis de Grassman. Estas variáveis antico-
mutativas são de alguma forma o limite clássico dos campos quânticos fermiónicos. Os
detalhes desta construção estão explicados nos Apêndices A e B. Aqui apenas recordamos
as nossas convenções. Devido ao carácter anticomutativo é necessário explicitar a ordem
da derivação. Assim
• As derivadas são esquerdas
Z
δ
d4 yη(y)ψ(y) = ψ(x)
δη(x)
Z
δ
d4 yψ(y)η(y) = −ψ(x) (5.117)
δη(x)
• Nas funções de Green a ordem de derivação é
G2n (x1 , . . . , yn ) = h0| T ψ(x1 ) · · · ψ(xn )ψ(y1 ) · · · ψ(yn ) |0i
δ2n Z[η, η]
≡
iδη(yn ) · · · iδη(y1 )iδη(xn ) · · · iδη(x1 )
δ δ
≡ ··· Z[η, η] (5.118)
iδη(yn ) iδη(x1 )
onde
d4 x[η(x)ψ(x)+ψ(x)η(x)]
R
Z[η, η] = h0| T ei |0i
Z
d4 x[L+η(x)ψ(x)+ψ(x)η(x)]
R
= D(ψ, ψ)ei (5.119)
Exemplos de aplicação destes resultados serão dados nos problemas.
5.6 Change of variables in path integrals. Applications

5.6.1 Introduction
Uma das grandes vantagens de ter uma expressão para o funcional gerador Z(J) em termos
dum integral de caminho é que um grande número de manipulações familiares para inte-
grais usuais (mudança de variáveis de integração, integração por partes ...) podem agora
ser aqui aplicadas. Vamos ver as consequências da mudança de variáveis de integração.
Consideremos uma transformação infinitesimal da forma
φi → φi + εFi (φ) (5.120)

onde
Fi (φ) = fi + fij φj + · · · (5.121)

Então devemos ter

δFi
D(φ) → D(φ) det δij + ε δφ j

= D(φ) 1 + ε δF i
δφi (5.122)
5.6. CHANGE OF VARIABLES IN PATH INTEGRALS. APPLICATIONS 169
e

i(S(φ)+Ji φi ) i(S(φ)+Ji φi ) δS
e →e 1 + iε + Ji Fi (φ) (5.123)
δφi
Como Z(J) deverá ser independente de transformação de variáveis obtemos
Z
δS δFi i(S[φ]+Ji φi )
0 = D(φ) i + Ji Fi + e (5.124)
δφi δφi
δ
Usando φi → obtemos a expressão mais compacta
iδJi

δS δ δ δFi δ
+ Ji Fi + Z(J) = 0 (5.125)
δφi iδJi iδJi δφi iδJ
Esta é a expressão geral que vamos aplicar a dois casos particulares importantes, as
equações de Dyson-Schwinger e as identidades de Ward.
5.6.2 Dyson-Schwinger equations

Seja Fi = fi independente de φi , isto é uma simples translação dos campos. Então a
equação anterior escreve-se

δS δ
+ Ji Z(J) = 0 (5.126)
δφi iδJ
que, como veremos, é a expressão da equação de Dyson-Schwinger (DS) para o funcional
gerador das funções de Green completas. Assim as equações de DS não são mais do que
uma consequência da invariância dos integrais de caminho para translações. A equação
5.126 pode-se ainda escrever

1 δ
Jk = − E Z[J] (5.127)
Z iδJk
onde o funcional E[φ] é a equação de movimento,
δS
E [φk ] ≡ . (5.128)
δφk
Para muitas aplicações é mais conveniente escrever as equações de Dyson-Schwinger
para as funções de Green conexas e próprias. Para isso temos de escrever a equação
equivalente a 5.126 para os funcionais W e Γ
i) Funções de Green conexas
Usando a identidade

1 δ δiW δ
(Z[J]f [J]) = + f [J] (5.129)
Z iδJk iδJk iδJk
Podemos escrever

1 δ δW δ
E Z[J] = E i + 1 (5.130)
Z iδJk iδJk iδJk
Portanto a equação de DS para o funcional gerador das funções de Green conexas

escreve-se simplesmente

δW δ
Jk = −E i + 1 (5.131)
iδJk iδJk
ii) Funções de Green próprias
Mais útil é a equação de DS para as funções de Green próprias ou irredutı́veis. Para
isso utilizamos as relações

 δW δΓ

 φk = iδJ Jk = −
k δφk

 δ δ δ δ
 = Gkm = −iΓkr
iδJk δφm δφk iδJr
e obtemos

δΓ δ
= E φk + Gkm 1 (5.132)
δφk δφm
É na forma 5.132 que as equações de DS são mais úteis.
Example : Self-energy in φ3
A acção para esta teoria escreve-se, usando a notação compacta
λ
⊓ − m2 )δkm φm −
S[φ] = φk (−⊔ (φk )3 (5.133)
3!
logo
λ
⊓ − m2 )φk −
E[φk ] = (−⊔ (φk )2 (5.134)
2
e portanto

δ 2 λ δ
E φk + Gkm 1 = −(⊔
⊓ + m )φk − φk + Gkr φk (5.135)
δφm 2 δφr
dá
δΓ λ 2
⊓ + m2 )φk −
= −(⊔ φk + Gkr δrk (5.136)
δφk 2
Derivando funcionalmente em relação a φm obtemos as equações de DS para diversas
funções de Green. Por exemplo para a self-energy obtemos
δ2 Γ λ
⊓ + m2 )δkm − (2φk δkm − iΓmn Gkrn δrk )
= −(⊔ (5.137)
δφk δφm 2
Pondo φk = 0 obtemos
λ
⊓ − m2 )δkm = i
Γkm − (−⊔ Γmn Gkrn δrk
2
λ
= Γmn Gkrn Γrs Gsk
2
λ
= i Γmn Γrs Gsk Gkk′ Grr′ Gnn′ Γk′ r′ n′
2
λ
= −i Gkk′ Gks Γk′ sm (5.138)
2
onde se usou repetidamente a equação 5.32 e a definição da Figura 5.11. Mas por definição
de self-energy
⊓ − m2 )δkm ≡ −Σkm
Γkm − (−⊔ (5.139)
e portanto
λ
−iΣkm = −i
Gkk′ Gks iΓk′ sm (5.140)
2
ou em termos de diagramas da Figura 5.18
k m k m
=
Figure 5.18: Dyson-Schwinger equation for φ3 .

λ 3
Como vemos a equação de DS não é mais que a afirmação que o vértice na teoria é 3! φ .
Example: Self-energy in φ4
Neste caso a acção escreve-se
λ
⊓ − m2 )δkm φm −
S[φ] = φk (−⊔ (φk )4 . (5.141)
4!
Logo a equação de movimento é
λ
⊓ − m2 )φk −
E[φ] = (−⊔ (φk )3 . (5.142)
3!
Portanto

δ 2 λ δ δ
E φk + Gkm 1 = −(⊔
⊓ + m )φk − φk + Gkm φk + Gkn φk
δφm 3! δφm δφn

λ δ
⊓ + m2 )φk −
= −(⊔ φk + Gkm φ2k + Gkn δnk
3! δφm
λ 3
⊓ + m2 )φk −
= −(⊔ φ + φk Gkn δnk + 2Gkm φk δkm
3! k

− iGkm Γmℓ Gknℓ δnk (5.143)
e obtemos
δΓ λ
⊓ + m2 )φk −
= −(⊔ φ3k + φk Gkn δnk + 2Gkm φk δkm − iGkm Γmℓ Gknℓ δnk (5.144)
δφk 3!
Para obter a equação de DS para a self-energy derivamos em ordem a φj e fazemos
φi = 0 depois de derivar. Obtemos assim
Γkj − (−⊔⊓ − m2 )δkj

λ
= − (Gkn δnk δkj + 2Gkm δkm δkj − iGkm Γmℓj Gknℓ δnk
3!
−Gkmp Γpj Γmℓ Gknℓ δnk − Gkm Γmℓ Gknℓp Γpj δnk ) (5.145)
logo
λ λ
−iΣkj = −i Gkk δkj + i Gkm iΓmℓj Gkk′ Gnn′ Gℓℓ′ iΓk′ n′ ℓ′ δnk
2 3!
λ
+i Gkk′ Gmm′ Gpp′ iΓk′ m′ p′ Γpj Γmℓ Gkk′′ Gnn′ Gℓℓ′ iΓk′′ n′ ℓ′ δnk
3!
λ
+i Gkm Γmℓ Gknℓp Γpj δnk
3!
λ λ
= −i Gkk δkj + i δkℓ δnk Gknℓp iΓpj (5.146)
2 3!
Para a teoria φ4 temos Γijk = 0 pelo que
Gknℓp = Gkk′ Gnn′ Gℓℓ′ Gpp′ iΓk′ n′ ℓ′ p′ (5.147)

e obtemos
λ λ
−iΣkj = −i Gkk δkj − i Gkk′ Gkn′ Gkℓ′ iΓk′ n′ ℓ′ j (5.148)
2 3!
que representamos diagramaticamente na Figura 5.19
k m 1 k m 1 k m
= +
2 3!
Figure 5.19: Dyson-Schwinger equation for φ4 .

5.6.3 Ward identities

Seja uma teoria com uma simetria qualquer. Essa simetria é expressa pela invariância da
acção
δS[φ]
Fi (φ) = 0 (5.149)
δφi
onde considerámos uma transformação do tipo 5.120. Se esta expressão deixar invariante
a medida D(φ) então a expressão geral 5.125 reduz-se a

δ
Ji Fi Z(J) = 0 (5.150)
iδJ
Esta expressão é conhecida por identidade de Ward. Derivação em ordem às fontes conduz
a relações entre as funções de Green que expressam as simetrias da teoria.
Para teorias de gauge a expressão é um pouco mais complicada. A razão é que no
processo de quantificação das teorias de gauge é normalmente necessário introduzir termos
que quebram a simetria para fixar a gauge. Assim podemos escrever
Sef f = SI + SN I (5.151)
onde δS δSNI
δφi Fi = 0 e δφi Fi 6= 0. Então se a medida continuar a ser invariante, devemos ter
I
agora a identidade de Ward na forma

δSN I δ δ
+ Ji Fi Z(J) = 0 (5.152)
δφi iδJ iδJ
Na próxima secção vamos aplicar esta expressão para obter as identidades de Ward
em QED. Para as teorias de gauge não abelianas a questão da invariância da medida é um
pouco mais delicada e será analisada no próximo capı́tulo depois de mostrarmos como se
quantificam estas teorias.
5.1 Calcule G4c a partir de 5.21 e mostre que é de facto a função de Green conexa de
quatro pernas.
5.2 Mostre que para um campo escalar real temos
Z
1 4 4 (0)
Z0 [J] = exp − d xd yJ(x)∆ (x, y)J(y) (5.153)
2
onde
Z
d4 k ik·(x−y) i
∆(0) (x − y) = e (5.154)
(2π)4 k2 − m2 + iǫ
Sugestão: Use a generalização do resultado
Z +∞
1 1 −1
dx1 · · · dxN e− 2 xi Mij xj +bi xi = π N/2 (det M )−1/2 e 2 bi (M )ij bj (5.155)
−∞
5.3 Determine os factores de simetria dos diagramas seguintes:
λ 3
5.4 Considere a teoria φ3 , isto é V (φ) = 3! φ . Usando
Z
4 δ
Z[J] = exp −i d xV Z0 (J) (5.156)
iδJ
onde
Z
1 (0)
Z0 (J) = exp[− d4 xd4 x′ J(x)GF (x − x′ )J(x′ )] (5.157)
2
e
Z
(0) ′ 1
GF (x − x′ ) = i d4 keik(x−x ) (5.158)
k2 − m2 + iε
Problems 175
mostre que o factor de simetria do diagrama
é S = 21 .
5.5 Dado o Lagrangeano de Dirac (teoria livre)
L0 = ψ(iγ µ ∂µ − m)ψ , (5.159)
mostre que o funcional gerador das funções de Green é dado por

0 (x,y)η(y)
d4 xd4 y η(x)SF
R
Z0 [η, η] = e− (5.160)
onde
Z
d4 p −ip·(x−y) i
SF0 αβ (x, y) = e
(2π)4 p/ − m + iε αβ
δ2 Z0
=
iδηα (y) iδη β (x)
= h0| T ψβ (x)ψ α (y) |0i .
5.6 Como mostraremos no Capı́tulo 6 o funcional gerador das funções de Green para
QED é dado por
Z R 4 µ
Z(Jµ , η, η) = D(Aµ , ψ, ψ) ei d x(LQED +LGF +J Aµ +ηψ+ψη) . (5.161)
onde
1
LQED = − Fµν F µν + ψ(iD
/ − m)ψ
4
1
LGF = − (∂ · A)2
2ξ
Dµ = ∂µ + ieAµ .
a) Calcule Z0 [J µ , η, η]
b) Mostre que
Z
δ δ δ
Z[J µ , η, η] = exp (−ie) d4 x (γ µ )αβ Z0 [J µ , η, η] . (5.162)
δηα (x) δη β (x) δJµ (x)
c) Expanda
Z = Z0 1 + (−ie)Z1 + (−ie)2 Z2 + · · · (5.163)
onde se retiraram as amplitudes vácuo-vácuo em Zi , isto é, Zi [0] = 0 → Z[0] = 1. Mostre
que
Z1 = −i (5.164)
1 1
Z2 = Z12 + +
2 2
1
− + (5.165)
2
d) Discuta os factores numéricos e os sinais das expressões anteriores.

e) Calcule em ordem mais baixa
δ3 Z
h0| T Aµ (x)ψβ (y)ψ α (z) |0i = (5.166)
iδηα (z)iδη β (y)iδJµ (x)
e verifique que coincide com as regras de Feynman para o vértice.
f) Calcule a amplitude para o efeito de Compton em ordem mais baixa, isto é
δ4 Z
h0| T Aµ (x)Aν (y)ψβ (z)ψ α (w) |0i = (5.167)
iδηα (w)iδη β (z)iδJν (y)iδJµ
e verifique que reproduz o que se obtém usando as regras de Feynman usuais.
5.7 As identidades de Ward para QED deduzidas na secção 1.7 não têm a forma

∂
Ji Fi Z(J) = 0 (5.168)
i∂J
onde δφi = Fi [φ] pois Z

4 1
SGF = d x − (∂ · A)2 (5.169)
2ξ
não é invariante para transformações de gauge. Introduza o funcional
Z R 4 µ
Z (Jµ , η, η) = D(Aµ , ψ, ψ, ω, ω) ei d x(Lef f +J Aµ +ηψ+ψη)
′
(5.170)
onde
Lef f = LQED + LGF + LG (5.171)
e
⊓ω .
LG = −ω⊔ (5.172)
onde ω e ω são campos escalares anticomutativos.
Problems 177
a) Mostre que
Z ′ (Jµ , η, η) = N Z(Jµ , η, η) (5.173)
onde N não depende das fontes nem dos campos. Explique porque é que esta renor-
malização (infinita) não afecta o cálculo das funções de Green. Assim tanto
R Z como Z ′
4
servem para o cálculo destas. b) Mostre que a medida D(Aµ , ψ, ψ, ω, ω) e d xLef f são
invariantes para a transformação
δψ = −ieωθψ δψ = ieψωθ
δAµ = ∂µ ωθ (5.174)
δω = 1ξ (∂ · A)θ δω = 0
onde θ é um parâmetro anticomutativo constante (variável de Grassman). c) Introduza
fontes anticomutativas para os campos ω e ω, isto é
Z R 4 µ
Z(Jµ , η, η, ζ, ζ) = D(Aµ , ψ, ψ, ω, ω)ei d x(Lef f +J Aµ +ηψ+ψη+ωζ+ζω) (5.175)
Mostre que
Z(Jµ , η, η, ζ, ζ) = ZG (ζ, ζ) Z(Jµ , η, η) (5.176)

onde Z
d4 x(LQED +LGF +J µ Aµ +ηψ+ψη)
R
Z(Jµ , η, η) = D(Aµ , ψ, ψ) ei . (5.177)
Considere os funcionais W , WG e W e ainda Γ, ΓG e Γ definidos de maneira semelhante.

Qual a relação entre W , WG e W e entre Γ, ΓG e Γ . d) Mostre que a equação de Dyson
Schwinger para os campos ω e ω é
δΓ
= −⊔
⊓ω . (5.178)
δω
e) Mostre que as identidades de Ward se podem agora escrever na forma
δ
Ji Fi [ ]Z = 0 . (5.179)
iδJ
Escreva as identidades de Ward para Γ(Aµ , ψ, ψ, ω, ω). Mostre que conduzem aos resulta-
dos conhecidos f) Mostre que um termo de massa para o fotão, embora quebre a simetria
de gauge, não estraga as identidades de Ward desde que os fantasmas ω adquiram massa.
Se o termo de massa do fotão for 21 µ2 Aµ Aµ qual a massa dos fantasmas?
Chapter 6
Non Abelian Gauge Theories
6.1 Classical theory

6.1.1 Introduction
Vamos brevemente rever como se constrói a acção clássica para uma teoria de gauge
não abeliana (Yang-Mills). Consideremos um grupo compacto G correspondendo a uma
simetria interna. Seja φi , (i = 1, · · · , N ) um conjunto de campos que se transformam de
acordo com uma representação de dimensão N de G, isto é
φ(x) → φ′ (x) = U (g)φ(x) (6.1)

onde U (g) é uma matriz N × N . Numa transformação infinitesimal
g = 1 − iαa ta a = 1, · · · , r (6.2)
onde os parâmetros αa são infinitesimais e ta são os geradores do grupo satisfazendo as
relações, para a representação fundamental
h i
ta , tb = if abc tc
1
T r ta tb = δab (6.3)
2
Exemplos destes geradores são
σa
SU (2) ta = ; a = 1, 2, 3
2
λ a
SU (3) ta = ; a = 1, · · · , 8 (6.4)
2
onde σ a e λa são as matrizes de Pauli e Gell-Mann respectivamente.
Na representação associada aos campos φ, as matrizes T a de dimensão (N ×N ) formam
uma representação de álgebra de Lie, isto é,
[T a , T b ] = if abc T c (6.5)
179
180 CHAPTER 6. NON ABELIAN GAUGE THEORIES
A sua normalização é dada por
T r(T a T b ) = δab T (R) (6.6)

onde T (R) é um número caracterı́stico da representação R. Para uma dada representação
mostra-se a identidade (ver problema 2.1)
T (R) r = d(R)C2 (R) (6.7)

onde r é a dimensão do Grupo G e d(R) é a dimensão da representação R. Numa trans-
formação infinitesimal
δφ = −iαa T a φ ≡ −iα
∼φ (6.8)
α ≡ αa T a .
onde introduzimos a notação ∼
6.1.2 Covariant derivative

Para resolver o problema da derivada não se transformar como os campos, isto é,
∂µ φ′ 6= U ∂µ φ (6.9)
quando os parâmetros dependem de xα , introduz-se a derivada covariante
a a
Dµ φ = (∂µ − igA
∼µ )φ ; A
∼µ = Aµ T (6.10)
a
onde Aµ são os campos de gauge (tantos quantos os geradores do grupo). As propriedades
de transformação de Aaµ são obtidas exigindo que Dµ φ se transforme como φ, isto é,
(Dµ φ)′ = (∂µ − igA ′ ′

∼µ )φ = (∂µ − igA
′
∼µ )U φ
′
= ∂µ U φ + U ∂µ φ − igA
∼µ U φ
′
∼µ − igA
= U Dµ φ + (igU A ∼µ U + ∂µ U )φ (6.11)
Portanto (Dµ φ)′ = U (Dµ φ) requer
′ −1 i −1
A ∼µ U − g ∂µ U U
∼µ = U A (6.12)
Infinitesimalmente U ≃ 1 − iα
∼ e obtemos
h i 1
δA
∼µ = −i ∼ ∼µ − g ∂µ∼
α , A α . (6.13)
o que se escreve em componentes
1
δAaµ = − ∂µ αa + f bca αb Acµ
g
1
= − (∂µ αa − gf bca αb Acµ ) (6.14)
g
6.1. CLASSICAL THEORY 181
Como na representação adjunta (T c )ab = −if bca então
1
δAaµ = − ∂µ δab − ig(T c )ab Acµ αb (6.15)
g
ou ainda
1
δAaµ = − (Dµ α)a (6.16)
g
6.1.3 Tensor Fµν

Calculemos o comutador de duas derivadas covariantes
h i
[Dµ , Dν ] φ = ∂µ − igA
∼ µ , ∂ν − igA∼ φ
ν
h i
= −ig ∂µ A∼ ν − ∂ν A
∼ µ − ig A µ
∼ ∼ , A ν φ
≡ −ig ∼
F µν φ (6.17)
a T a designado por curvatura,
F µν ≡ Fµν
onde se definiu o tensor ∼
h i
F
∼ µν = ∂µ A
∼ ν − ∂ν A
∼ µ − ig A µ
∼ ∼ , A ν (6.18)
ou em componentes
a
Fµν = ∂µ Aaν − ∂ν Aaµ + gf abc Abµ Acν (6.19)
Vejamos como se transforma Fµν numa transformação de gauge.
h i
F ′µν
∼ = ∂µ A
∼ν
′
− ∂ A
ν ∼µ
′
− ig A ′
,
∼µ ∼ν A ′

−1 i −1
= ∂µ (U A∼ν U ) − g ∂µ (∂ν U U ) − (µ ↔ ν)
h i
−1 −1 −1
−igU [A ,
∼µ ∼νA ]U − ∂µ U U , U A
∼ν U
h i i
−1 −1 −1 −1
− UA
∼ µ U , ∂ν U U + ∂µ U U , ∂ν U U (6.20)
g
Usando
∂µ U −1 = −U −1 ∂µ U U −1 (6.21)
obtemos
∼ F µν U −1
F ′µν = U ∼ (6.22)
ou infinitesimalmente
h i
∼µν = −i ∼
δF α,∼
F µν (6.23)
É fácil de ver que com o tensor Fµν é possı́vel construir um invariante. De facto a
quantidade
′ ′µν 1 a aµν
T r(F
∼µν F∼ ) = T r(F F µν ) =
∼µν ∼ F F (6.24)
2 µν
é invariante e pode ser usada para construir uma acção, generalizando a acção de Maxwell
para as teorias abelianas.
6.1.4 Choice of gauge

Chama-se gauge pura ao potencial A µ F µν = 0. É fácil de mostrar que
∼ tal que ∼
−1
Fµν = 0 ⇐⇒ ∃U : A ∼µ = ∂µ U U (6.25)
Para evitar a arbitrariedade de gauge é conveniente por vezes impor condições de
gauge. Exemplos são a Gauge Axial definida por
nµ Aµ (x) = 0 (6.26)
onde nµ é um quadrivector constante, e a Gauge Lorentz
∂µ Aµ (x) = 0 (6.27)
6.1.5 The action and the equations of motion

A acção para a teoria de gauge pura é
Z
1
S = − d4 xT r(F F µν )
∼µν ∼
2
Z
1
= − d4 xFµν
a
F µνa (6.28)
4
Devido ao resultado anterior sobre T r(F F µν ), Eq. (6.24), a acção é invariante para
∼µν ∼
transformações do grupo de gauge G. A equação de Euler-Lagrange
δL δL
∂µ − =0 (6.29)
δ(∂µ Aν ) δAaν
a
obtém-se facilmente notando que

b
δL δL δFρσ
= = −F aµν (6.30)
δ(∂µ Aaν ) b δ(∂ Aa )
δFρσ µ ν
e
b
δL δL δFρσ
= = gf bca Abµ F cµν (6.31)
δAaν b δAa
δFρσ ν
Então
∂µ F µνa + gf bca Abµ F µνc = 0 (6.32)

ou ainda atendendo a que na representação adjunta (T c )ab = −if bca
(∂µ δab − ig(T c )ab Acµ )F µνb = 0 (6.33)

isto é
Dµab F µνb = 0 (6.34)
6.1.6 Energy–momentum tensor

Como para o Electromagnetismo o tensor canónico não é invariante de gauge. De facto 1
δL
θ̃ µν = − ∂ ν Aaρ + g µν L
δ(∂µ Aaρ )
1
= F µρa ∂ ν Aaρ − gµν F ρσa Fρσ
a
(6.35)
4
Para o tornar invariante de gauge procedemos como no electromagnetismo. Subtraı́mos a
θ̃ µν uma quantidade que seja uma divergência para que as leis de conservação não venham
alteradas. A quantidade relevante é
∆θ µν = ∂ρ (F µρa Aνa )
= ∂ρ F µρa Aνa + F µρa ∂ρ Aνa
= gf bca Abρ F ρµc Aνa + F µρa ∂ρ Aνa
= F µρa (−Fρνa + ∂ ν Aaρ ) (6.36)
logo
θ µν ≡ θ̃ µν − ∆θ µν
1
= F µρa Fρνa − gµν F ρσa Fρσ
a
(6.37)
4
expressão análoga à obtida no electromagnetismo. Introduzindo os análogos dos campos
eléctricos e magnéticos
1
Eai = Fai0 ; Bak = − εijk Faij i, j, k = 1, 2, 3 (6.38)
2
1
One should note an overall sign difference with respect to the general definition of Eq. (1.66). This is
to maintain the component θ00 with the meaning of a positive energy density. Obvsiously, the overall sign
in Eq. (1.66), as no meaning prior to make contact with the model.
obtemos

 θ 00 = 1 ~a ~a + B
~a · B
~ a)
2 (E ·E
(6.39)
 θ 0i ~a × B
= (E ~ a )i
com uma interpretação semelhante ao caso do electromagnetismo.
6.1.7 Hamiltonian formalism

Da expressão para θ 00 é claro que o Hamiltoniano é
Z
1 ~a ~a ~a ~a
H = d3 x (E ·E +B ·B )
2
Z
= d3 xH (6.40)
onde a H é a densidade Hamiltoniana.

Vamos ver no entanto que devido à invariância de gauge a relação entre o Hamiltoni-
ano e o Lagrangeano não é a usual. Para isso é conveniente partir da acção escrita em
formalismo de 1a ordem.
Z
4 1 a a abc b c µνa 1 a µνa
S = d x − (∂µ Aν − ∂ν Aµ + gf Aµ Aν )F + Fµν F (6.41)
2 4
a são agora variáveis independentes. É fácil de ver que a variação de S em
onde Aaµ e Fµν
ordem a Fµνa dá a sua definição
a
Fµν = ∂µ Aaν − ∂ν Aaν + gf abc Abµ Acν (6.42)
e que por sua vez substituindo 6.42 em S obtemos a acção usual. Usando as definições de
~a e B
E ~ a obtemos
Z
S = ~ a + ∇A
d4 x−(∂ 0 A ~ 0a − gf abc A0b A ~ a − 1 (E
~c) · E ~a · E
~a + B ~a · B
~ a)
2
Z
4 0 ~a ~ a 1 ~2 ~2 0a ~ ~ a abc ~ b ~ c
= d x −∂ A · E − (E + B ) + A (∇ · E − gf A · E ) (6.43)
2
A densidade Lagrangeana escreve-se, portanto
L = −E ka ∂ 0 Aka − H(E ka , Aka ) + A0a C a (6.44)

onde

 ~a · E
H ≡ 12 (E ~a + B~a · B
~ a)



B ka ≡ − 21 ǫkmn F mna (6.45)



 a ~ ~a
C = ∇ · E − gf abc A ~b · E
~c
As variáveis Aak e −Eka são as variáveis conjugadas, H(Eka , Aak ) é a densidade Hamil-
toniana. As variáveis A0a desempenham o papel de multiplicadores de Lagrange para as
condições,
~ ·E
∇ ~ a − gf abc A
~b · E
~c = 0 (6.46)
que não são mais do que as equações de movimento 6.34 para ν = 0.

Se introduzirmos os parêntesis de Poisson a tempo igual
{Aia (x), E jb (y)}x0 =y0 = δij δab δ3 (~x − ~y) (6.47)
é fácil de mostrar que
{C a (x), C b (y)}x0 =y0 = −gf abc C c (x)δ3 (~x − ~y )
{H, C a (x)} = 0 (6.48)
Isto mostra que as teorias de gauge (não abelianas neste caso, mas a afirmação é
igualmente verdadeira para teorias abelianas) representam um exemplo daquilo a que se
chama Sistemas de Hamilton Generalizados primeiro introduzidos por Dirac. Para definir
estes sistemas consideremos um sistema com as variáveis canónicas (pi , qi ) que geram o
espaço de fase Γ2n (i = 1, . . . , n). Então a acção destes sistemas é escrita na forma
Z
S= L(t)dt (6.49)
onde
n
X m
X
L(t) = pi q̇i − h(p, q) − λα ϕα (p, q) . (6.50)
i=1 α=1
As variáveis λα (α = 1, ...m) são multiplicadores de Lagrange e ϕα são as ligações. Para

que este sistema seja um sistema de Hamilton generalizado é necessário que as condições
seguintes sejam verificadas
X
{ϕα , ϕβ } = f αβγ (p, q)ϕγ
α
{h, ϕα } = f αβ (p, q)ϕβ (6.51)
O caso das teorias de gauge é um caso particular com f αβ = 0. Portanto para a quan-
tificação das teorias de gauge temos primeiro de aprender a quantificar sistemas Hamilton
generalizados.
6.2 Quantization
6.2.1 Systems with n degrees of freedom
Consideremos um sistema de Hamilton generalizado descrito na última secção. O seu
Lagrangeano é
L(t) = pi q̇i − h(p, q) − λα ϕα (p, q) (6.52)

que conduz às equações de movimento

∂ϕα


 q̇i = ∂h + λα

 ∂pi ∂pi
 α
ṗ = − ∂h − λα ∂ϕ (6.53)
 i ∂qi ∂qi




 α
ϕ (p, q) = 0 α = 1, . . . , m
Pode-se mostrar que um sistema de Hamilton generalizado (SHG) é equivalente a um
sistema de Hamilton usual (SH) definido um espaço de fase Γ∗2(n−m) . Isto é um SHG é
equivalente a um SH com n − m graus de liberdade. O SH Γ∗ pode ser construı́do da
maneira seguinte. Sejam m condições
χα (p, q) = 0 ; α = 1, . . . , m (6.54)
que satisfaçam
n o
χα , χβ = 0 (6.55)
e

det {ϕα , χβ } 6= 0 (6.56)
Então o subespaço de Γ2n definido pelas condições


 χα (p, q) = 0
α = 1, . . . , m (6.57)
 ϕα (p, q) = 0
é o espaço Γ∗2(n−m) pretendido. As variáveis canónicas p∗ e q ∗ em Γ∗2(n−m) podem ser

encontradas da maneira seguinte. Devido à condição 6.55 podemos escolher as variáveis qi
em Γ2n de tal forma que os χα coincidam com as primeiras m variáveis do tipo coordenada,
isto é,
q ≡ ( χα , q ∗ ) (6.58)
|{z} |{z} |{z}
n m n−m
Sejam p = (pα , p∗ )
os correspondentes momentos conjugados. Nestas variáveis a condição
6.56 toma a forma
α
∂ϕ
det β 6= 0 (6.59)
∂p
6.2. QUANTIZATION 187
portanto as condições ϕα (p, q) = 0 podem ser resolvidas para pα , isto é
pα = pα (p∗ , q ∗ ) (6.60)
O subespaço Γ∗ é portanto definido pelas condições

 χα ≡ q α = 0
(6.61)
 pα = pα (p∗ , q ∗ )
As variáveis p∗ e q ∗ são canónicas e o Hamiltoniano é dado por

h∗ (p∗ , q ∗ ) = h(p, q) (χ=0 ; ϕ=0) (6.62)
e as equações de movimento são agora
∂h∗ ∂h∗
q̇ ∗ = ṗ∗ = − (6.63)
∂p∗ ∂q ∗
num total de 2(n − m) equações. O resultado fundamental pode ser enunciado na forma
dum teorema.
Teorema 2.1
As duas representações são equivalentes, isto é, conduzem às mesmas equações de
movimento.
Dem:
As relações q α = 0 =⇒ q̇ α = 0 ou seja na descrição (p, q)
∂h ∂ϕβ
+ λβ =0 ; α = 1, . . . , m (6.64)
∂pα ∂pα
Consideremos agora as equações de movimento para as coordenadas q ∗ nas duas
representações
∂h ∂ϕα
q̇ ∗ = ∗
+ λα ∗
∂p ∂p
∂h∗ ∂h ∂h ∂pα
q̇ ∗ = = ∗+ α ∗ (6.65)
∂p∗ ∂p ∂p ∂p
As duas equações serão equivalentes se
∂ϕα ∂h ∂pα
λα ∗
= α ∗ (6.66)
∂p ∂p ∂p
ou seja usando as relações 6.64

α ∂ϕα ∂ϕα ∂pβ
λ + =0 (6.67)
∂p∗ ∂pβ ∂p∗
Ora esta relação é verdadeira em virtude da ligação ϕα = 0. Portanto as duas

representações são equivalentes o que demonstra2 o teorema.
Para efectuar a quantificação destes sistemas podemos usar as expressões para o opera-
dor evolução em termos dum integral de caminho nas variáveis (p∗ , q ∗ ) pois estas formam
um sistema Hamiltoniano normal. Temos
Z Y ∗ ∗ R
dp dq i [p∗ q̇∗ −h(p∗ ,q∗ )]dt
U (qf∗ , qi∗ ) = e (6.68)
t
(2π)
Embora este seja um modo possı́vel de proceder à quantificação, não é o mais con-
veniente em muitas situações onde é difı́cil inverter as relações ϕα = 0 para obter pα =
pα (p∗ , q ∗ ). Será mais conveniente usar as variáveis (p, q) com restrições apropriadas. Isto
pode ser feito facilmente substituindo
Y dp∗ dq ∗ Y dpdq Y
→ δ(q ∗ )δ(pα − pα (p∗ , q ∗ )) (6.69)
t
(2π) t
2π t
Então
Z Y
dpdq Y R
U (qf , qi ) = δ(q α )δ(pα − pα (p∗ , q ∗ ))ei dt(pq̇−h(p,q)) (6.70)
t
2π t
Esta expressão pode ainda ser escrita em termos das ligações se recordarmos que
δ(q α ) = δ(χα )

∂ϕα
α α ∗ ∗
δ(p − p (p , q )) = δ(ϕ ) det α
(6.71)
∂pβ
Então
Y Y
δ(q α )δ(pα − pα (p∗ , q ∗ )) = δ(ϕα )δ(χα ) det |{ϕα , χβ }| (6.72)
t t
Finalmente se usarmos a identidade

Z
α dλ −i R dtλα ϕα
δ(ϕ ) = e (6.73)
2π
obtemos
Z Y
dpdq dλ Y
U (qf , qi ) = δ(χα ) det |{ϕα , χβ }| eiS(p,q,λ) (6.74)
t
2π 2π t,x
onde
Z
S(p, q, λ) = [pq̇ − h(p, q) − λϕ]dt (6.75)
2
As equações para as variáveis p∗ tratavam-se de modo semelhante.
Esta é a expressão que iremos aplicar às teorias de gauge. Notar que a expressão dentro do
parêntesis recto é precisamente a do Lagrangeano para sistemas de Hamilton generalizados,
6.52. Pode-se mostrar (ver problema 2.2) que os resultados fı́sicos não dependem da escolha
das condições auxiliares χα = 0. Em teorias de gauge, fala-se da escolha de gauge.
6.2.2 QED as a simple example

Consideremos o campo electromagnético acoplado a uma corrente conservada J µ = (p, J~),
∂µ J µ = 0. O Lagrangeano é
1
L = − Fµν F µν − J µ Aµ (6.76)
4
A acção pode ser escrita na seguinte forma equivalente3
Z " #
~2 ~2
S= ~ · (∇A
d4 x −E ~˙ − B
~ 0 + A) ~ ·∇ ~ + B − E − ρA0 + J~ · A
~ ×A ~ (6.77)
2
~ eB
As equações do movimento são, variando em ordem E ~
 
~ = −(∇A
 E ~˙
~ 0 + A) 
 ~ ~
∇·B = 0
→ (6.78)
 ~ ~ ×A
~ 
 ~
B = ∇ ∇×E~ = − ∂ B~
∂t
~
e, variando em ordem a A0 e A,

~ ·E
 ∇ ~ =ρ
(6.79)
 ∇
~ ×B
~− ∂E
= J~
∂t
~ =∇
Se substituirmos B ~ ×A
~ obtemos, depois de uma integração por partes,
Z ( ! )
~ 2 + (∇
E ~ × A)2
S= 4
d x −E ~˙ −
~ ·A − J~ · A
~ + A (∇
0 ~ ·E
~ − ρ) (6.80)
2
É claro que A0 desempenha o papel dum multiplicador de Lagrange. As variáveis

~ e E
canónicas são A ~ mas elas não são livres pois existe uma ligação que tem que ser
~ E
respeitada, ∇· ~ = ρ. Esta ligação é linear nos campos. Aqui reside a grande simplificação
do electromagnetismo. Isto porque se escolhermos uma condição de gauge também linear
então o det{ϕα , χβ } não dependerá de E ~ eA ~ e será uma constante que apenas afectará
a normalização. Uma tal condição de gauge é obtida, por exemplo, da forma seguinte
(gauge de Lorentz)
χ = ∂µ Aµ − c(~x, t) (6.81)
3
A acção escrita na forma 6.77 é designada por formalismo de 1a ordem. Comparar com a equação 6.43.
onde c(~x, t) é uma função arbitrária. Então é fácil de ver que a expressão para o funcional
gerador das funções de Green é
Z Y
µ
Z[J ] = D(E, ~ A,
~ A0 ) δ(∂µ Aµ − c(x))eiS (6.82)
x
onde
Z " ( # )
˙ E ~ × A)2
2 + (∇
S = ~ ·A
d4 x −E ~− + (J~ · A)
~ + A0 (∇~ ·E
~ − ρ)
2
Z ( )
E2 ˙ ( ~ × A)2
∇
= 4
d x − −E~ · (∇A
~ 0 ~ −
+ A) − Jµ Aµ
(6.83)
2 2
~ é gaussiana e pode ser imediatamente efectuada obtendo-se

A integração em E
Z Y
Z[Jµ] = D(Aµ ) δ(∂µ Aµ − c(x))eiS (6.84)
x
onde agora

Z
1
S = d4 x − (∂µ Aν − ∂ν Aµ )(∂ µ Aν − ∂ ν Aµ ) − Jµ Aµ
4
Z
4 1 µν µ
= d x − Fµν F − Jµ A (6.85)
4
Como as funções c(x) são arbitrárias podemos integrar sobre elas com um peso
Z
1 4 2
exp − d xc (x) (6.86)
2ξ
obtendo então o resultado familiar
Z h i
1
i d4 x − 14 F 2 − 2ξ (∂·A)2 −J·A
R
µ
Z[J ] = D(Aµ )e (6.87)
Como veremos adiante se tivéssemos escolhido uma condição de gauge não linear, o
det |{q, χ}| já dependeria de E~ ou A~ e não teria sido possı́vel absorvê-lo na normalização
(que é irrelevante pois escolhemos sempre a normalização de forma que Z[0] = 1) . Nesse
caso será necessário utilizar os métodos das teorias de gauge não abelianas que vamos
estudar na próxima secção.
6.2.3 Non abelian gauge theories. Non covariant gauges

Vimos anteriormente que a acção para as teorias de gauge não abelianas (TGNA), se podia
escrever na forma
Z
4 ~ 0~ 1 ~2 ~2 0 ~ ~ ~ ~
S =2 ∼·∂ A
d x Tr E ∼ +B
∼ + 2 (E ∼ (∇ · E
∼ )−A ∼ + g[A
∼, E
∼]) (6.88)
Z

= d4 x −Eka ∂ 0 Aak − H(Ek , Ak ) + A0a C a (6.89)
onde
~ ·E
Ca = ∇ ~ a − gf abc A
~b · E
~c (6.90)
Se introduzirmos os parêntesis de Poisson a tempo igual
n o
−Eai (x), Ajb (y) 0 0 = δij δab δ3 (~x − ~y ) (6.91)
x =y
é fácil mostrar que
n o
C a (x), C b (y) = −gf abc C c (x)δ3 (~x − ~y)
x0 =y0
{H, C a (x)} = 0 (6.92)
onde
Z Z h i
1
H= d3 xH(Ek , Ak ) = d3 x (E ka )2 + (B ka )2 (6.93)
2
Assim as teorias de gauge não abelianas representam um exemplo de sistemas da Hamilton
generalizados. As variáveis do género coordenada são Aak , os momentos conjugados são
−Eka . As varáveis A0a são multiplicadores de Lagrange para garantir as ligações
~ ·E
∇ ~ a − gf abc A
~b · E
~c = 0 (6.94)
que são parte das equações de movimento.
Para procedermos à quantificação podemos usar o formalismo da secção 6.2.1 para
SHG. Temos para isso que impor r condições auxiliares (r é a dimensão do grupo de
Lie), isto é, tantas como as condições de ligação C a (x) = 0, a = 1, . . . , r. Escolher estas
condições é aquilo que se chama escolher, ou fixar, a gauge. Esta escolha é arbitrária,
os resultados fı́sicos não devem depender dela (ver problema 2.2). No entanto expressões
intermédias como sejam, por exemplo, as regras de Feynman, dependem fortemente da
gauge.
Como vimos no exemplo do electromagnetismo se for possı́vel fixar uma condição de
gauge linear nas variáveis dinâmicas, A ~a e E~ a , então a expressão do integral de caminho
simplifica-se bastante devido ao facto do determinante não depender dessas variáveis e
poder ser absorvido numa constante de normalização. Uma gauge em que isto é possı́vel
é a chamada gauge axial que passaremos a estudar.
• Gauge Axial
É sempre possı́vel efectuar uma transformação de gauge tal que a componente de A ~a
segundo uma direcção espacial seja nula em todos os pontos, isto é, escolhendo a direcção
segundo o eixo dos zz
A3a = 0 a = 1, . . . , r (6.95)
Estas r condições constituem as nossas condições auxiliares necessárias para se proceder à

quantificação da teoria. A vantagem desta escolha de gauge é a seguinte. Se calcularmos
{C a , A3b } obtemos
{Ca (x), A3b (y)} = {∂k Eak (x), A3b (y)} − gCadc Akd {Eck (x), A3b (y)}
∂ 3 δ
= −g δab δ (~x − ~y ) = (δA3b (y)) (6.96)
∂x3 δαa (x)
onde se usou o facto de A3b = 0. Vemos assim que {C a , A3b } não depende de A ~a e E~a e
o determinante que aparece na expressão do integral de caminho pode ser absorvido na
normalização. Podemos assim escrever para o funcional gerador das funções de Green
Z Y
~ A,
~ A0 ) ~ ~ 0 µ
Z[J µa ] = D(E, δ(A3 )eiS(E,A,A ,J ) (6.97)
x
onde
Z
~ A,
~ A ,J ) =
0 µ 4 ~a 1 h ~a 2
0 ~a
i
~ a )2 + A0a C a + Aa · J a
S(E, d x −E · ∂ A − (E ) + (B (6.98)
2
e
~ ·E
Ca = ∇ ~ a − gf abc A
~b · E
~c (6.99)
Como a integração em E~ aparece na forma duma integração gaussiana obtemos facil-
mente
Z Y
3 i d4 x L(x)+Aa ·J a
R
µa µ
ZA [J ] = D(A ) δ(A )e (6.100)
x
onde
1 a aµν
L = − (Fµν F ) (6.101)
4
O ı́ndice A em ZA [J µa ] realça o facto deste funcional corresponder à escolha de gauge
axial. Embora a expressão para o funcional gerador se escreva facilmente nesta gauge ela
tem a desvantagem das regras de Feynman não serem covariantes. Antes de introduzirmos
as gauge covariantes mostraremos ainda outra gauge não covariante a chamada gauge de
Coulomb:
• Gauge de Coulomb
Esta gauge é definida pelas condições auxiliares
~ ·A
∇ ~a = 0 a = 1, . . . , r (6.102)
Estas condições auxiliares têm um parêntesis de Poisson não trivial com as ligações C a (x).
De facto ( ver problema 2.3)
Z n o
δA~ a = − 1 d3 y A ~ a (x), αb (y)Cb (y) (6.103)
g x0 =y0
logo
n o δ
~ a (x), Cb (y)
A = −g ~ a (x))
(δA (6.104)
x0 =y0 δαb (y)
e
n o δ ~
∇~ ·A
~ a (x), Cb (y) = −g ~ a (x))
∇ · (δA (6.105)
x0 =y0 δαb (y)
Como
~ a (x) = 1~ ~ c (x)
δA ∇αa (x) + Cabc αb (x)A (6.106)
g
~ ·A
obtemos (com a condição ∇ ~ a = 0)
~ · (δA
−g∇ ~ a (x)) = −∇2x αa (x) − gCabc A
~ c (x) · ∇α
~ b (x) (6.107)
e finalmente
n o h i
~ ·A
∇ ~ a (x), Cb (y) ~ c (x) · ∇
= −∇2x δab − gCabc A ~ x δ3 (~x − ~y )
≡ Mcab (x, y) (6.108)

~ não depende de E,
Como det M embora dependendo de A ~ a integração gaussiana em E ~
pode ainda ser feita e obtemos
Z Y Y
~ i d4 x[L+Aa ·J a ]
R
µa
ZC [J ] = D(Aµ ) [det MC ~ · A)]e
δ(∇ (6.109)
x x
Agora não é possı́vel absorver det M na normalização. As regras de Feynman que
podem ser obtidas a partir de ZC [J µ ] também não são covariantes.
6.2.4 Non abelian gauge theories in covariant gauges

As condições de gauge escolhidas até aqui (gauge axial e gauge de Coulomb) conduzem
a regras de Feynman onde a covariância de Lorentz é perdida. Claro que os resultados
finais não devem depender desta escolha, mas a não covariância dos cálculos intermédios é
normalmente uma complicação. Vamos aqui generalizar os resultados anteriores a gauges
covariantes. O método a seguir surgirá como um subproduto da resposta a um outro
problema: Como mostrar a equivalência das gauges axial e de Coulomb?
Para o argumento que se segue é conveniente trabalhar com quantidades invariantes de
gauge. Assim em vez do funcional ZA [J µ ] vamos por agora considerar o integral ZA [J = 0]
que como vimos tem o significado duma amplitude transição vácuo → vácuo na ausência
das fontes exteriores,
Z Y
ZA [0] = D(Aµ ) δ(A3a (x)) exp{iS[Aµ ]} (6.110)
x,a
onde S[Aµ ] é a acção. Numa transformação de gauge

′ g −1 i −1
∼µ → A
A ∼µ = A ∼µ U (g) − g ∂µ U U
∼µ = U (g)A (6.111)
A acção S[Aµ ] e a medida D(Aµ ) são invariantes, pelo que obtemos

Z Y
ZA (J = 0) = D(Aµ ) δ(gA3a (x)) exp{iS[Aµ ]} . (6.112)
x,a
Definimos agora o funcional ∆C [Aµ ] através da relação

Z Y
−1
∆C (Aµ ) = D(g) δ(∇ ~ · gA
~a) (6.113)
x,a
onde D(g) representa o produto infinito de medidas invariantes para o grupo G em cada
ponto, isto é
Y
D(g) = dg(x) . (6.114)
x
A invariância da medida da integração do grupo G, dg′ = d(gg ′ ) tem como consequência

que ∆C é invariante de gauge. De facto
Z Y ′
∆−1 g
D(g′ ) ~ · g gA
δ(∇ ~a)
C ( Aµ ) =
x,a
Z Y ′
= D(g′ g) ~ · g gA
δ(∇ ~a)
x,a
= ∆−1
C [Aµ ] (6.115)
Introduzimos agora na expressão de ZA [J = 0] a identidade

Z Y
1 = ∆C [Aµ ] D(g) ~ · gA
δ(∇ ~a) (6.116)
x,a
Obtemos então
Z Y Z Y

ZA (J = 0) = DAµ eiS[Aµ] δ A3a (x) ∆C [Aµ ] D(g) ~ · gA
δ(∇ ~b)
x,a y,b
Z Y Z Y −1
= DAµ eiS[Aµ ]
∆C [Aµ ] ~
δ ∇·A~ b
D(g) δ(g A3a ) (6.117)
y,b x,a
onde usámos a invariância de D, S[Aµ ] e ∆C [Aµ ]. Como a medida é invariante podemos

escrever no último integral g−1 → gg0 . Então
Z Y −1 Z Y
g 3a

D(g) δ A (x) = D(g) δ gg0A3a (x) (6.118)
x,a x,a
~ ·A
onde g0 é a transformação de gauge que leva da gauge ∇ ~ = 0 para a gauge A′3 = 0,
isto é
A′3 = g0A3 = 0 (6.119)

~ ·A
com ∇ ~ = 0. Falta-nos portanto calcular o integral sobre o grupo que agora se escreve
Z Y
D(g) δ gA′3a (x) (6.120)
x,a
com A′3a = 0. Como A′3a = 0 basta considerar transformações infinitesimais, na vizi-

nhança da identidade,
a a
g(x) = e − iα
∼(x) = e − iα (x)t (6.121)
onde αa (x) são infinitesimais. Nestas condições a medida de integração dg(x) vem dada
por
Y
dg(x) = dαa (x) (6.122)
a
Por outro lado em primeira ordem em αa temos
g ′3a 1 ∂αa
A (x) = (6.123)
g ∂x3
pelo que o integral virá
Z Y
Z Y 1 ∂αa
g ′3a
D(g) δ A (x) = D(α) δ =N (6.124)
x,a x,a
g ∂x3
O integral é independente de Aµ pelo que pode ser absorvido na normalização. Obtemos

assim
Z Y
ZA [J = 0] = N D(Aµ )∆C [Aµ ] ~ · A)e
δ(∇ ~ iS[Aµ ] (6.125)
x
Na secção anterior obtivemos uma expressão para ZC [J = 0], que é

Z Y Y
ZC [J = 0] = D(Aµ ) det MC δ(∇~ · A)e
~ iS[Aµ ] (6.126)
x x
Portanto para mostrar a equivalência dos integrais representando a amplitude vácuo →
vácuo na ausência de fontes exteriores nas duas gauges consideradas falta-nos mostrar que
∆C [Aµ ] = det MC . De facto
Z Y
∆−1 D(g) δ ∇~ · gA
~a
f [Aµ ] =
x,a
Z Y
1

= D(α) ~ ·
δ ∇ ~ (x) + f α A
∇αa abc b ~ c
x,a
g
Z Y 1
= D(α) δ ~ b·A
∇2x αa (x) + f abc ∇α ~c
x,a
g
∝ det−1 MC (6.127)
onde
δ ~ g ~a
Mab
f (x, y) = −g ∇· A
δαb (y) α=0

= ~c · ∇
−∇2x δab − gCabc A ~ x δ3 (~x − ~y ) (6.128)
Portanto ∆C [Aµ ] ∝ det MC e à parte uma normalização irrelevante ZA [0] = ZC [0].

A maneira como se demonstrou a equivalência entre as gauges de Coulomb e axial
sugere a forma de definir a amplitude vácuo → vácuo para uma gauge arbitrária definida
pela condição
F a [Aµ ] = 0 a = 1, . . . , r (6.129)
Para isso definimos ∆F [Aµ ] pela expressão
Z Y
∆−1
F [Aµ ] = D(g) δ (F a [gAµ ]) (6.130)
x,a
e como anteriormente introduzimos

Z Y
1 = ∆F [Aµ ] D(g) δ (F a [gAµ ]) (6.131)
x,a
na expressão para ZA [J = 0]. Obtemos
Z Y Z Y

ZA [J = 0] = D(Aµ ) δ A (x) eiS[Aµ ] ∆F [Aµ ]
3a
D(g) δ F b [g Aµ ]
x,a y,b
Z Y Z Y −1
b iS[Aµ ]
= D(Aµ ) δ F [Aµ ] ∆F [Aµ ]e D(g) δ g A3a (x)
y,b x,a
Z Y
= N D(Aµ )∆F [Aµ ] δ F [Aµ ] eiS[Aµ ]
b
x,a
= N ZF [J = 0] (6.132)
mostrando que as gauges axial e as gauges do tipo F são equivalentes. A amplitude vácuo
→ vácuo na gauge F a = 0 é portanto dada por
Z Y
ZF [J = 0] = D(Aµ )∆F [Aµ ] δ (F a [Aµ ]) eiS[Aµ ] (6.133)
x,a
Falta-nos calcular ∆F [Aµ ]. Como na definição anterior ∆F [Aµ ] aparece multiplicado por
Q
δ (F a [Aµ ]), basta-nos conhecer ∆F [Aµ ] para Aµ que satisfaz F a [Aµ ] = 0. Então para g
perto da identidade obtemos
δF a b
F a [gAbµ ] = F a [Abµ ] + δA
δAbµ µ
1 δF a
= − b
(Dµ α)b (6.134)
g δAµ
onde se usou F a [Abµ ] = 0 e δAbµ = − g1 (Dµ α)b . Calculemos então ∆F . Obtemos
Z Y
∆−1
F [Aµ ] = D(g) δ F a [gAbµ ]
x,a
Z Y 1 δF a
b
= D(α) δ − (Dµ α)
x,a
g δAbµ
∝ det−1 MF (6.135)
onde
δF a δF a [g A(x)]
Mab
F (x, y) = D cb 4
δ (x − y) = −g (6.136)
δAcµ (x) µ δαb (y)
e portanto

δF a (x)
∆F [Aµ ] = det MF = det −g (6.137)
δ(αb (y))
Já sabemos como escrever a amplitude vácuo → vácuo na ausência de fontes exteriores
para uma gauge arbitrária. De facto não é esta quantidade a mais interessante, mas sim a
amplitude vácuo → vácuo na presença de fontes, ZF [J] pois será esta que gera as funções
de Green. Em toda a discussão
R 4 a até aqui se fizeram as fontes exteriores nulas. A razão é
que o termo das fontes, d xJµ A , não é invariante de gauge. Se definirmos ZF [Jµa ] pela
µa
relação
Z Y R 4 a µa
ZF [Jµa ] ≡ D(Aµ )∆F [Aµ ] δ(F a [Abµ (x)])ei(S[Aµ ]+ d xJµ A ) (6.138)
x,a
então é claro que os funcionais ZF não serão equivalentes para diferentes escolhas de
F a = 0. Isto quer dizer que as funções de Green calculadas a partir de ZF [J µ ] vão
depender de gauge F a = 0. Na secção 2.2.5 mostraremos que embora as funções de Green
dependam da escolha de gauge, este não é um problema importante porque os resultados
fisicamente relevantes (mensuráveis) estão relacionados com os elementos da matriz S
renormalizada e esta é independente da gauge conforme aı́ mostraremos.
Antes de acabarmos esta secção façamos uma transformação no funcional ZF [Jµa ] para
nos vermos livres
Q da função δ que aı́ intervém. Para os cálculos é de toda a conveniência
exponenciar δ(F a [Aµ ]). Isto pode fazer-se do seguinte modo. Definamos uma condição
de gauge mais geral
F a [Abµ ] − ca (x) = 0 (6.139)

onde ca (x)
são funções arbitrárias do espaço-tempo, mas não dependem dos campos. Então
∆F [A] não vem alterado e escrevemos
Z Y R 4 a µ
a
ZF [Jµ ] = N D(Aµ )∆F [Aµ ] δ(F a [Aµ ] − ca )ei(S[Aµ ]+ d xJµ A ) (6.140)
O lado esquerdo da equação não depende de ca (x) pelo que podemos integrar em ca (x)
com um peso conveniente, especificamente com
Z
i 4 2
exp − d x ca (x) (6.141)
2
onde x é um parâmetro real. Obtemos então
Z
d4 x(− 12 Fa2 +J µa Aa
D(Aµ )∆F [Aµ ]ei(S[Aµ ]+ µ ))
R
ZF [Jµa ] = N
Z
d4 x[L(x)− 21 Fa2 +J µa Aa
R
= N D(Aµ )∆F [Aµ ]ei α] (6.142)
Esta expressão é o ponto de partida para o cálculo das funções de Green numa gauge
arbitrária definida pela função F a . Para sermos capazes de estabelecer as regras de Feyn-
man para esta teoria teremos ainda de exponenciar ∆F [Aµ ]. Isto será feito numa das
secções seguintes com a introdução dos chamados fantasmas de Fadeev-Popov.
6.2.5 Gauge invariance of the S matrix

Na secção anterior definimos o funcional gerador das funções de Green ZF [Jµa ], para uma
gauge dada pela função F a [Abµ ], através da relação
Z Y R 4 a µa
a
ZF [Jµ ] ≡ N D(Aµ )∆F [A] δ(F a [Abµ (x)])ei(S[Aµ ]+ d xJµ A ) (6.143)
x,a
e mostrámos a equivalência das diferentes gauges quando as fontes eram nulas. Vamos
agora mostrar o que acontece quando Jµa 6= 0. Para isso vamos refazer a demonstração da
equivalência entre duas gauges na presença das fontes. Escolhemos para esta demonstração
as gauges de Coulomb e Lorentz4 definidas por
 a
 F = ∇ ~ ·A~a gauge de Coulomb
(6.144)
 a
F = ∂µ Aµa gauge de Lorentz
Definimos então os funcionais geradores ZC [jµa ] e ZL [Jµa ] por
Z Y
~ a )ei(S[A]+ d4 xjµa Aµa )
R
ZC [jµa ] ≡ N D(Aµ )∆c [A] ~ ·A
δ(∇ (6.145)
x,a
4
Depois de estudarmos as identidades de Ward-Takahashi faremos uma demonstração geral (ver
secção 6.3.4).
e
Z Y
d4 xJµa Aµa )
R
ZL [Jµa ] =N D(Aµ )∆L [A] δ(∂µ Aµa )ei(S[A]+ (6.146)
x,a
Vamos mostrar a relação entre eles. Seguindo os métodos da secção anterior introduzimos
em ZC [jµa ] a identidade dada por
Z Y
1 = ∆L [A] D(g) (∂µ g Aµa ) (6.147)
x,a
e obtemos
ZC [jµa ]
R Q R
~ a )ei(S[A]+ d4 xjµa Aµa ) ∆L [A] D(g) Q δ(∂µ gAµa )
R
~ ·A
= N D(Aµ )∆C [A] x,a δ(∇ y,b
R Q R Q R
d4 xjµ
a g −1Aµa
=N D(Aµ )∆L [A] µb iS[A] ∆ [A] D(g) ~ · ~ a )ei
g −1A
y,b δ(∂µ A )e C x,a δ(∇
R Q R Q gg 0A
R
d4 xjµ
a gg 0Aµa
=N D(Aµ )∆L [A] µb iS[A] ∆ [A] D(g) ~ · ~ a )ei
y,b δ(∂µ A )e C x,a δ(∇
(6.148)
~ ·A
onde g0 é a transformação de gauge que leva de gauge ∂µ Aµa = 0 para a gauge ∇ ~ ′a = 0,
~ ′a = gA
A ~ e é portanto obtida resolvendo a equação

~ ~ ′ ~ 0 ~ −1 0 i~ 0 −1 0
∇ · A = ∇ · U (g )AU (g ) − ∇U (g )U (g ) = 0 (6.149)
g
Q ~ · gA
~ ′ ) só nos interessam as transformações g
onde ∂µ Aµa = 0. Devido ao factor x δ(∇
infinitesimais pelo que
Z Y
d4 xjµ
a g 0Aµ
R
ZC [jµa ] =N D(Aµ )∆L [A] δ(∂µ Aµb )eiS[A] ei (6.150)
y,b
onde se usou o resultado

Z Y
D(g) ~ · gA
δ(∇ ~ ′ ) = ∆−1 [A] . (6.151)
C
x,a
0
Para comparar com ZL [Jµa ] é necessário escrever g Aµ em função de Aµ , resolvendo a
equação para g0 . Isto pode ser feito formalmente em série de potenciais de Aµ . É fácil de
ver que devemos ter

1
A′i = δij − ∇i 2 ∇j Aj + O(A2λ ) (6.152)
∇
~ · ~j = 0
Se restringirmos a fonte na gauge de Coulomb a ser transversal j 0 = 0 e ∇
podemos então escrever
Z Y
d4 xFµa j µa
R
ZC [jµa ] =N D(Aµ )∆L [A] δ(∂µ Aµb )eiS[A]+ (6.153)
y,b
onde FµC [A] = Aaµ + O(A2λ ). Comparando com a expressão de ZL [Jµa ] obtemos finalmente
Z
a 4 a µa δ
ZC [jµ ] = exp i d xjµ F ZL [Jµa ] (6.154)
iδJ b
Esta é a expressão que relaciona ZC com ZL . Como Fµ [A] é um funcional complicado
é fácil de ver que as funções de Green nas duas gauges vão ser diferentes. Mas o que
tem significado fı́sico (comparável com a experiência) são os elementos de matriz S renor-
malizada. O teorema da equivalência que a seguir demonstramos mostra que a matriz S
renormalizada é invariante de gauge. Por simplicidade demonstraremos o teorema para a
teoria λφ4 mas o raciocı́nio é análogo para o caso das teorias de gauge.
Teorema 2.2
Se dois funcionais geradores Z e Z̃ diferem somente nos termos das fontes exteriores
então eles conduzem à mesma matriz S renormalizada.
Dem. Consideremos o funcional gerador das funções de Green
Z
d4 xJφ)
R
Z[J] = N D(φ)ei(S[φ]+ (6.155)
onde
Z Z
4 4 1 µ 1 2 2 λ 4
S[φ] + d xJφ = d x ∂µ φ∂ φ − m φ − φ + Jφ . (6.156)
2 2 4!
Que acontece se acoplarmos a fonte exterior a φ + φ3 em vez de ser somente a φ ?

Podemos escrever o funcional gerador Z̃[j] dado por
Z
d4 xj(φ+φ3 )]
R
Z̃[j] = N D(φ)ei[S[φ]+ (6.157)
e podemos escrever Z̃[j] em termos de Z[J],

Z
4 δ
Z̃[j] = exp i d xj(x)F Z[J] (6.158)
iδJ
onde F [φ] = φ + φ3 . Consideremos a função de 4- pontos, G̃(1, 2, 3, 4) gerada por

Z̃[j]
δ4 Z̃[j]
G̃(1, 2, 3, 4) = (−i)4 (6.159)
δj(1)δj(2)δj(3)δj(4)
Um diagrama tı́pico que contribui para G̃(1, 2, 3, 4) é o da Figura 6.1, onde a parte
do diagrama dentro do quadrado a tracejado é uma função de Green gerada por Z[J].
Figure 6.1: Green functions generated by Z[J] and Z̃[j].
Consideremos agora os propagadores G̃(1, 2) e G(1, 2) gerados por Z̃[j] e Z[J] re-
spectivamente. Obtemos a seguinte expansão de G̃(1, 2) em termos de G(1, 2)
~
G = + +
+ + + (6.160)
Se examinarmos os propagadores junto do pólo na massa fı́sica, obtemos (Z2 e Z̃2

são as constantes de renormalização nos dois esquemas)
iZ̃2 iZ2
lim G̃ = ; lim G = . (6.161)
p2 →m2R p2 − m2R p2 →m2R p2 − m2R
Então multiplicando a expansão da equação 6.160 por p2 − m2R e tomando o limite

p2 → m2R obtemos
2
Z̃2 = Z2 1 + 2 + + ··· (6.162)
ou seja
!1/2
Z̃2
σ≡ =1+ +··· (6.163)
Z2
A matriz S não renormalizada é dada por
n
Y
S NR (k1 , . . . , kn ) = lim (ki2 − m2R )G(k1 , . . . , kn ) (6.164)
ki2 →m2R
i=1
para as funções de Green calculadas a partir de Z[J]. Definimos de igual modo
n
Y
S̃ NR (k1 , . . . , kn ) = lim (ki2 − m2R )G̃(k1 , . . . , kn ) (6.165)
ki2 →m2R
i=1
para as funções de Green calculadas a partir do funcional Z̃[j], sendo n o número de

partı́culas exteriores. Do argumento usado para relacionar lim(k2 −m Q
2 )G̃(k , . . . , k )
R 1 n
com Q lim(k2 − m2R )G(k1 , . . . , kn ) é fácil de ver que para relacionar lim(ki2 − m2R )G̃
com lim(ki2 − m2R )G somente contribuem os diagramas que tiverem pólos em todas
as variáveis ki2 . Então
n
Y −n Y n
Z
lim (ki2 − m2R )G̃ = lim (ki2 − m2R )G
i=1
ki2 →m2R Z̃ i=1
ki2 →m2R
n
n Y
= σ2 lim (ki2 − m2R )G (6.166)
ki2 →m2R
i=1
Portanto obtemos uma relação entre os elementos da matriz S não normalizada
n
NR NR
S̃ = σ2S (6.167)
ou ainda
1 NR 1
n S̃ (k1 , . . . , kn ) = n S NR (k1 , . . . , kn ) (6.168)
Z̃ 2 Z2
1
Mas n S NR (k1 , . . . , kn ) é precisamente a definição da matriz S renormalizada pelo
Z2
que
S̃ R = S R . (6.169)
Concluı́mos assim que dois funcionais geradores que defiram pelo acoplamento à
fonte exterior produzem os mesmos elementos de matriz da matriz S renormalizada.
Isto completa a demonstração do teorema da equivalência.
A aplicação deste resultado ao nosso caso é agora imediata pois

Z
δ
ZC [jµa ] 4 a µa
= exp i d xjµ F ZL [Jµc ] (6.170)
iδJx
onde Fµa [A] = Aaµ + O(A2λ ). A diferença entre ZC [jµ ] e ZL [Jµ ] reside no acoplamento
à fonte exterior, pelo que embora as funções de Green dependam da gauge, a matriz S
renormalizada deverá ser invariante.
6.2.6 Fadeev-Popov ghosts

Tendo demonstrado a invariância de gauge de matriz S renormalizada, voltemos ao fun-
cional gerador numa gauge arbitrária definida pela condição F a [Abµ ]. No final de secção 6.2.4
tı́nhamos escrito este funcional na forma
Z
1
i d4 x[L(x)− 2ξ (F a )2 +Jµa Aµa ]
R
ZF [Jµa ] = N D(Aµ )∆F [A]e (6.171)
onde

δF a (x)
∆F [A] = det MF = det −g b (6.172)
δα (y)
Nesta forma as regras de Feynman são complicadas porque o det MF conduz a in-
teracções não locais entre os campos de gauge. Se de alguma forma pudéssemos exponen-
ciar detMF e metê-lo numa acção efectiva terı́amos o nosso problema resolvido.
Ora no nosso estudo dos integrais gaussianos sobre variáveis de Grassman obtivemos
o resultado
Z R 4
D(ω, ω)e− d xωMF ω = det MF (6.173)
usando este resultado e mudando por conveniência MF → iMF (uma mudança na nor-
malização irrelevante) obtemos
Z R 4 a µa
ZF [Jµ ] = N D(Aµ , ω, ω)ei d x[Lef f +Jµ A ]
a
(6.174)
onde ω e ω são campos escalares anticomutativos e o Lef f é definido por
Lef f = L + LGF + LG (6.175)

onde
1 a µνa
L = − Fµν F
4
1
LGF = − (F a )2
2ξ
LG = −ωa Mab
F ω
b
(6.176)
Os campos ω e ω são campos auxiliares não fı́sicos e chamam-se fantasmas de Fadeev-

Popov. Como não são fı́sicos não há problema com o teorema que relaciona o spin com a
estatı́stica5 .
Calculemos agora duma forma mais explı́cita o Lagrangeano do fantasmas. Como
δF a (x) δF a [A(x)] cb
Mab
F (x, y) = −g = Dµ (6.177)
δαb (y) δAcµ (y)
5
Campos fı́sicos com spin inteiro são bosões (comutativos) e campos fı́sicos com spin semi- inteiro são
fermiões (anticomutativos).
obtemos
Z Z Z
4 4 a δF a (x) cb
d xd yω (x)Mab b
F (x, y)ω (y) = 4
d x d4 y ω a (x) D ωb (y) (6.178)
δAcµ (y) µ
ou seja
Z
δF a (x) bc
LG (x) = − d4 y ω a (x) D ωc (y) (6.179)
δAbµ (y) µ
Para termos uma forma mais explı́cita temos que especificar a gauge. Na gauge de
Lorentz F a = ∂µ Aµa e portanto
Z

LG (x) = − d4 yω a (x)∂xµ δ4 (x − y) Dµab ω b (y)
= ∂ µ ω a (x)Dµab ω b (x) (6.180)
onde se efectuou uma integração por partes e a derivada covariante na representação

adjunta6 é
Dµab = ∂µ δab − gf abc Acµ (6.181)
6.2.7 Feynman rules in the Lorenz gauge

Estamos agora em posição de escrever as regras de Feynman para calcular, em teoria
das perturbações, qualquer processo que envolva partı́culas cujas interacções possam ser
descritas por uma teoria de gauge não abeliana referente a um dado grupo de simetria.
Todo o nosso trabalho até aqui se pode resumir na procura do Lagrangeano efectivo a
partir do qual as regras de Feynman podem ser obtidas como se se tratasse duma teoria
normal sem graus de liberdade a mais. O nosso Lagrangeano efectivo é, como vimos, dado
por
Lef f = L + LGF + LG (6.182)

onde
1 a µνa a
L = − Fµν F ; Fµν = ∂µ Aaν − ∂ν Aaν + gf bca Abµ Acν
4
1
LGF = − (Fa )2
2ξ
Z
δF a
LG = −ωa d4 y b Dµbc ωc (6.183)
δAµ
As constantes f abc são definidas pela comutação dos geradores do grupo, sendo as nossas
convenções
6
Os fantasmas, tal como os campos de gauge estão na representação adjunta do grupo G.
[ta , tb ] = if abc tc
1
T r(ta tb ) = δab (6.184)
2
Para fixar ideias vamos considerar a gauge de Lorentz definida por
F a [A] = ∂µ Aµa (x) . (6.185)

Obtemos então
1 a µνa 1
Lef f = − Fµν F − (∂µ Aµa )2 + ∂ µ ω a Dµab ω b (6.186)
4 2ξ
onde
Dµab ω b = (∂µ δab − gf abc Acµ )ω b (6.187)

e usámos o facto de que os fantasmas se encontram na representação adjunta pelo que

(Dµ ω)a = ∂µ δab − igAcµ (T c )ab ω b (6.188)
com
(T c )ab ≡ −if bca = −if abc (6.189)

Podemos escrever portanto
Lef f = Lcin + Lint (6.190)

Onde
1 1
Lcin = − (∂µ Aaν − ∂ν Aaµ )2 − (∂µ Aµa )2 + ∂µ ω a ∂ µ ω a
4 2ξ

1 µa 1
= A ⊓gµν − 1 −
⊔ ∂µ ∂ν δab Aνb − ω a ⊔
⊓ δab ω b (6.191)
2 ξ
onde se desprezaram divergências totais. O Lagrangeano de interacção é
1
Lint = −gf abc ∂µ Aaν Aµb Aνc − g2 f abc f ade Abµ Acν Aµd Aνe + gf abc ∂ µ ω a Abµ ω c . (6.192)
4
Com as convenções usuais (ver capı́tulo 5) obtemos as seguintes regras de Feynman
• Propagadores:
i) Campos de gauge
µν
µ ν g kµ kν
a k b −iδab 2 − (1 − ξ) 2 (6.193)
k + iǫ (k + iǫ)2
ii) Fantasmas
i
a b δab (6.194)
k k2 + iǫ
• Vértices:
i) Vértice triplo dos bosões de gauge
ρ, c
−gf abc [ gµν (p1 − p2 )ρ + g νρ (p2 − p3 )µ
p3
+g ρµ (p3 − p1 )ν ]
p2
p1
µ, a ν, b p1 + p2 + p3 = 0
(6.195)
ii) Vértice quártico dos bosões de gauge
h
σ, d ρ, c −ig 2 Ceab Cecd (gµρ gνσ − gµσ gνρ )
p4 p3 +Ceac Cedb (gµσ gρν − gµν gρσ )

i
p1 p2 +Cead Cebc (gµν gρσ − gµρ gνσ )
µ, a ν, b
p1 + p2 + p3 + p4 = 0
(6.196)
iii) Interacção Fantasmas-Bosões de Gauge
µ, c
p3
g f abc pµ1
p2 (6.197)
p1 p1 +p2 + p3 = 0
a b
Notas:
1. O ponto no vértice entre os fantasmas e os bosões de gauge refere-se à perna que

tem a derivada e que corresponde à linha de saı́da (as linhas dos fantasmas são
orientadas)
2. As outras regras são as usuais não esquecendo o sinal − por cada loop de fantasmas.
6.2.8 Feynamn rules for the interaction with matter

Na secção anterior vimos as regras de Feynman para a teoria de gauge pura, sem interacção
com a matéria. A interacção com a matéria faz-se da forma habitual passando as derivadas
usuais a derivadas covariantes. Em geral a matéria é descrita por partı́culas escalares
φi ; i = 1, ...M (6.198)
e partı́culas spinoriais
ψj ; j = 1, ...N (6.199)
pertencendo a representações de dimensão M e N , respectivamente. O Lagrangeano será

dado por
Lmatéria = (Dµ φ)† D µ φ − m2φ φ† φ − V (φ)
+iψD µ γµ ψ − mψ ψψ
≡ Lcin + Lint . (6.200)
O Lagrangeano de interacção entre a matéria e os campos de gauge obtém-se facilmente

a partir da derivada covariante
µ
Dij = ∂µ δij − igAaµ Tija (6.201)
onde Tija são os geradores nas representações adequadas para os campos φ e ψ. Assim
obtemos
→ ←
b
Lint = igφ∗i (∂ − ∂ )µ φj Tija Aµa + g 2 φ∗i Tija Tjk φk Aaµ Aµb
+gψ i γ µ ψj Tija Aaµ (6.202)
o que conduz aos seguintes vértices

µ, a
p3
p2 ig(γ µ )βα Tija (6.203)
p1
β, i α, j
µ, a
p3
p2 ig(p1 − p2 )µ Tija (6.204)
p1
i j
µ, a ν, b
ig2 gµν {T a , T b }ij (6.205)

i j
• Factores do Grupo
Os factores f abc e Tija que aparecem nos vértices não precisam de facto de ser conhe-
cidos. Nos cálculos aparecem, como veremos, combinações daqueles factores que podem
ser expressas em termos de quantidades invariantes que caracterizam o grupo e a repre-
sentação. Por conveniência resumimos aqui os resultados mais usados.
Os nossos geradores são hermitı́cos (T a+ = +T c ) satisfazendo as relações
[T a , T b ] = if abc T c
T r(T a T b ) = δab T (R) (6.206)

onde T (R) é um número caracterizando a representação R. Outra quantidade frequente-
mente usada é o operador de Casimir da representação definido por
X
a a
Tik Tkj = δij C2 (R) (6.207)
a,k
Para a representação adjunta obtemos
f acd f bcd = δab C2 (G) . (6.208)

T (R) e C2 (R) não são independentes obedecendo à relação
T (R)r = d(R)C2 (R) (6.209)

onde r é a dimensão do grupo G e d(R) é a dimensão da representação R.
Em muitas aplicações estamos interessados em grupos SU (N ). Para estes temos os
seguintes resultados
r = N 2 − 1 ; d(N ) = N ; d(adj) ≡ d(G) = r (6.210)
1 N2 − 1
T (N ) = ; C2 (N ) = (6.211)
2 2N
T (G) = C2 (G) = N (6.212)

• Factores de Simetria
Para o cálculo de diagramas envolvendo partı́culas idênticas é necessário multiplicar o
resultado de aplicar as regras de Feynman por um factor de simetria conveniente. Estes
factores de simetria foram discutidos na secção 5.5.5 . Por conveniência reproduzimos
aqui a regra lá deduzida. O factor de simetria é o # de maneiras diferentes em que
as linhas podem ser ligadas com o mesmo resultado topológico a dividir pelos factores
permutacionais dos vértices e pelo número de permutações de pontos com vértices iguais.
6.3. WARD IDENTITIES 209
6.3 Ward identities

6.3.1 BRS transformation
Vamos aqui deduzir as identidades de Ward7 para as teorias de gauge não abelianas. O
método mais conveniente é o chamado método das transformações de Becchi, Rouet e
Stora (BRS) que já introduzimos no capı́tulo 5 para o caso de QED. As transformações de
BRS são uma generalização das transformações de gauge que tornam invariante a acção
efectiva.
Como vimos para uma teoria de gauge não abeliana a acção efectiva é dada por (A =
campos de gauge ; φ = campos de matéria)
Z Z
1
Sef f [A, φ] = S[A, φ] − d xFa [A, φ] − d4 xω a Mab ω b
4 2
(6.213)
2ξ
onde S[A, φ] é a acção clássica, invariante para transformações de gauge
1
δAaµ = − Dµab αb
g
δφi = −i(T a )ij φj αa , (6.214)
Fa [A, φ] são as condições de gauge e o operador Mab é tal que
δFa cb b δFa
Mab ω b = D ω + ig(T b )ij φj ω b . (6.215)
δAcµ µ δφi
Sef f não é invariante para as transformações de gauge devido à não invariância do
termo que fixa a gauge e do Lagrangeano dos fantasmas. Esta não invariância pode
desaparecer se escolhermos transformações
R apropriadas para os fantasmas para compensar
a não invariância do termo d4 xFa2 . Estas transformações são


 δBRS Aaµ = Dµab ω b θ





 δBRS φi = ig(T b )ij φj ω b θ
(6.216)

 δBRS ω a = 1

 ξ Fa [A, φ]θ



 1
δBRS ω a = 2 gf
abc ω b ω c θ
onde θ é um parâmetro anticomutativo independente do ponto do espaço-tempo (variável

de Grassman). Vemos que as transformações BRS para os campos Aaµ e φi são trans-
formações de gauge com parâmetro αa (x) = −gω a (x)θ. Notar que o carácter anticomuta-
tivo de θ é necessário para que o produto ω a θ tenha um carácter bosónico (comutativo).
Para demonstrar a invariância de Sef f [A, φ] vamos demonstrar uma série de teoremas
que são necessários para a prova geral. Antes é no entanto conveniente introduzir o
operador de Slavnov s, definido pelas relações seguintes,
7
Designamos pelo nome genérico de identidades de Ward as identidades que foram descobertas por
Ward, Takahashi, Slavnov e Taylor.
δBRS Aaµ = sAaµ θ δBRS ω a = sω a θ

(6.217)
δBRS φi = sφi θ δBRS ωa = sω a θ
Este operador é distributivo em relação à multiplicação verificando-se as relações
seguintes
s(B1 B2 ) = sB1 B2 + B1 sB2
s(F1 B2 ) = sF1 B2 + F1 sB2
s(B1 F2 ) = −sB1 F2 + B1 sF2
s(F1 F2 ) = −sF1 F2 + F1 sF2 (6.218)

que podem ser demonstradas a partir da definição.
Teorema 2.3
O operador s é nilpotente nos campos Aaµ , φi e ω a , isto é s2 Aaµ = s2 φi = s2 ω a = 0.
Dem.
Demonstremos para cada um dos casos. Obtemos
a) s2 Aaµ = 0
2 δDµab c b
s Aaµ = s(Dµab ω b )
=− sAν ω + Dµab sω b
δAcν
1
= −δµν (−gf abc )Dνcd ω d ω b + gf bcd Dµab (ω c ω d )
2

1 1
= gf abc ∂µ ω c ω b + gf acd ∂µ ω c ω d + gf acd ω c ∂µ ω d
2 2

abc cde e d b 1 bcd abe e c d
+ gf (−g)f Aµ ω ω + g(−g)f f Aµ ω ω
2
= (gf abc ∂µ ω c ω b − gf abc ∂µ ω c ω b )

1
− g 2 (f abc f cde − f adc f cbe + f cdb f ace )Aeµ ω d ω b
2
= 0 (6.219)
onde se usou a identidade de Jacobi e a antisimetria das constantes de estrutura do

grupo.
b) s2 φi = 0
s2 φi = s [ig(T a )ij φj ω a ]
= −ig(T a )ij sφj ω a + ig(T a )ij φj sω a

1
= g2 (T a )ij (T b )jk φk ω b ω a + ig(T a )ij φj gf abc ω b ω c
2
1 2 c b i
= g [T , T ]ik ω b ω c φk + g2 (T a )ij φj f abc ω b ω c
2 2
i 2 a
= g (T )ij φj (f acb + f abc )ω b ω c
2
= 0 (6.220)
c) s2 ω a = 0

2 a 1 abc b c
s ω = s gf ω ω
2
1 1
= − gf abc sω b ω c + gf abc ω b sω c
2 2
= −gf abc sω b ω c
1
= − g2 f abc f bef ω e ω t ω c
2
1
= − g2 (f abc f bef + f abe f bf c + f abf f bce )ω e ω t ω c
6
= 0 (6.221)
onde se usou a anticomutatividade dos fantasmas e a identidade de Jacobi.
Fica assim demonstrado o teorema 2.3. Para gauges lineares pode-se demonstrar um
resultado importante a que daremos a forma de teorema.
Teorema 2.4
Para gauges lineares o operador de Slavnov verifica a relação
s(Mab ω b ) = 0 (6.222)
Dem:
Vimos anteriormente que
Z
δFa (x) cb b δFa (x)
Mab ω b (x) = d4 y D ω (y) + ig(T b
) φ
ij j ω b
(y) (6.223)
δAcµ (y) µ δφi (y)
Se usarmos as definições de δBRS e do operador de Slavnov podemos escrever

Z
b 4 δFa (x) c δFa (x)
Mab ω (x) = d y sA (y) + sφi (y) (6.224)
δAcµ (y) µ δφi (y)
δFa δFa
Se a gauge for linear δAcµ e δφi não dependem dos campos e portanto
h i Z
b 4 δFa (x) 2 c δFa (x) 2
s Mab ω (x) = d y s Aµ (y) + s φi (y) = 0 (6.225)
δAµ (y) δφi (y)
onde se usaram os resultados do teorema 2.3.
Usando os teoremas 2.3 e 2.4 podemos agora mostrar que a acção efectiva é invariante
para transformações de BRS. Vamos apresentar este resultado também sobre a forma de
teorema.
Teorema 2.5
A acção Sef f é invariante para as transformações de BRS.
Dem.
A acção efectiva é
Z
1
Sef f [A, φ] = S[A, φ] + 4
d x − Fa2 [A, φ] − ω a Mab ω b (6.226)
2ξ
Como a acção clássica é invariante para transformações de gauge devemos ter
s (S[A, φ]) = 0 . (6.227)
Para os outros termos obtemos

1 1
s − Fa2 − ω a Mab ω b = − Fa sFa + sωa Mab ω b − ω a s(Mab ω b ) (6.228)
2ξ ξ
Mas
Z
4 δFa b δFa
sFa (x) = d y sA (y) + sφi (y) = Mab ω b (x) (6.229)
δAbµ (y) µ δφi (y)
e pelo teorema 2.4
s(Mab ω b ) = 0 (6.230)
logo

1 1
s − Fa2 − ω a Mab ω b = − Fa + sω a Mab ω b = 0 (6.231)
2ξ ξ
onde se usou sω a = 1ξ Fa . Pondo tudo junto obtemos portanto
sSef f [A, φ] = 0 . (6.232)

Para o seguimento é ainda importante um outro teorema,
Teorema 2.6
A medida D(Aµ , φi , ω a , ω b ) é invariante para transformações BRS.
Dem:
Cálculos simples conduzem às seguintes relações:
δ(sAaµ )
= −gf aba δµµ ω b = 0
δAaµ
δ(sφi )
= ig(T a )ii ω a = 0 ; (T r(T a ) = 0)
δφi
δ(sω a )
= gf aac ω c = 0
δω a
δ(sω a )
=0 (6.233)
δω a
Como vimos no capı́tulo 5 estas relações implicam que a medida é invariante, o que
demonstra o teorema.
6.3.2 Ward-Takahashi-Slavnov-Taylor identities

Vamos aqui deduzir a generalização das identidades de Ward-Takahashi para as teorias
de gauge não abelianas. Esse trabalho foi feito, entre outros, por Slavnov e Taylor mas
usaremos com frequência o nome de identidades de Ward mesmo para as teorias não
abelianas. Duma forma genérica, as identidades de Ward são relações entre as funções
de Green que resultam da simetria de gauge da teoria. De acordo com o que vimos no
capı́tulo 5 a maneira mais conveniente das expressar é usar os funcionais geradores das
funções de Green.
Consideremos então uma teoria de gauge não abeliana. Por simplicidade consideramos
que a matéria é constituı́da por campos escalares φi . A introdução de fermiões é imediata.
O funcional gerador das funções de Green é então
Z
d4 x[Lef f +Jµa Aµa +Ji φi +ηa ω a +ω a ηa
R
Z[Jµa , Ji , η a , η a ] = D(Aµ , φi , ω, ω)ei ] (6.234)
onde introduzimos também fontes para os fantasmas. Uma transformação de BRS é uma
mudança de variável no integral. O valor do integral não deve ser alterado por essa
mudança de variáveis. Como Sef f e a medida são invariantes devemos ter o seguinte
teorema:
Teorema 2.7
Dada uma função de Green qualquer
G(x1 , ..., y1 , ..., z1 , ..., w1 , ...)

= h0| T Aaµ1 (x1 ) · · · φi1 (y1 ) · · · ω a (z1 ) · · · ω b (w1 ) · · · |0i (6.235)
temos as relações
i) s h0| T Aaµ (x1 ) · · · φi1 (y1 ) · · · ω a (z1 ) · · · ω b1 (w1 ) |0i = 0
ii) 0 = h0| T sAaµ (x1 ) · · · |0i + · · · + h0| T · · · sφi · · · |0i + · · ·
+ h0| T · · · sωa · · · |0i · · · + h0| T · · · sω a · · · |0i (6.236)
Dem:
A demonstração é imediata se escrevermos
h0| T Aaµ (x1 ) · · · φi1 (y1 ) · · · ω a (z1 ) · · · ω b (w1 ) |0i =

Z
= D(Aµ , φi , ω, ω)Aaµ (x1 ) · · · φi1 (y1 ) · · · ω a (z1 ) · · · ω b (w1 )eiSef f (6.237)
Então a transformação de BRS deve deixar o valor do integral invariante pelo que a
primeira relação é imediata. A segunda relação resulta da primeira e da invariância
de medida e da acção efectiva.
Este teorema constitui uma forma expedita de estabelecer relações entre as funções de
Green para casos particulares e é muito útil em cálculos práticos, como veremos no segui-
mento. Contudo para estabelecer resultados gerais sobre a renormalização e invariância de
gauge da matriz S interessa-nos as identidades de Ward expressas em termos dos funcionais
geradores. Usando a invariância dum integral numa mudança de variáveis, a invariância
da medida D e de Sef f obtemos a identidade de Ward para o funcional gerador Z
Z Z
0= D(Aµ , φi , ω, ω) d4 x(J µa sAaµ + Ji sφi + η a sω a − sω a η a )ei(Sef f +fontes) (6.238)
Como vimos em QED as identidades de Ward mais úteis são para o funcional Γ. A
expressão anterior não permite passar para o funcional Γ porque sAaµ , sφi e sω a são não
lineares nos campos. Para resolver este problema introduzimos fontes para estes operadores
não lineares. Generalizamos assim a acção efectiva definindo uma nova quantidade Σ tal
que
Σ[Aaµ , φi , ω a , ω a , Kµa , Ki , La ]
Z
≡ Sef f [Aaµ , φi , ω a , ω a ] + d4 x(K aµ sAaµ + K i sφi + La sω a ) (6.239)
onde K aµ , Ki e La são fontes para os operadores compostos sAaµ , sφi e sω a respectivamente.

Usando os teoremas 2.3 e 2.5 é imediato mostrar que Σ é invariante para transformações
BRS, isto é,
sΣ = 0 . (6.240)
Consideremos agora que o funcional gerador das funções de Green na presença das
fontes Jµa , Ji , η a , η a , K µa , K i e La , isto é
Z R
i Σ+ d4 x(Jµa Aµa +Ji φi +ηω+ωη)
Z[Jµa , Ji , η, η, Kµ , Ki , L] = D(Aµ , φi , ω, ω)e (6.241)
Podemos agora repetir o raciocı́nio da invariância para as transformações de BRS.

Como anteriormente obtemos (recordar que sΣ = 0)
Z Z
0 = D(· · · ) d4 x[Jaµ sAaµ + J i sφi + η a sω a − sωa η a ]ei(Σ+fontes) , (6.242)
mas agora temos operadores compostos sA, sφ e sω, isto é
δΣ δΣ
sAaµ = sφi =
δK µa δKi
δΣ 1
sω a = sωa = F a . (6.243)
δLa ξ
Obtemos então
Z Z
δΣ i δΣ a δΣ 1 a a i(Σ+fontes)
D(· · · ) d4 x Jaµ + J + η − F η e =0 (6.244)
δK µa δK i δLa ξ
ou ainda
Z
4 δ i δ a δ 1 a δ δ a
d x Jaµ +J +η − F , η a eiW [Jµ ,Ji ,η,η,Kµ ,Ki ,L] = 0
iδK µa iδK i iδLa ξ iδJµ iδJi
(6.245)
Para uma condição de gauge, linear todos os operadores diferenciais dentro do parêntesis
recto são de 1a ordem e portanto podemos escrever
Z
4 µ δ i δ a δ 1 a
d x Ja +J +η − Fa η W = 0 . (6.246)
δK µa δKi δLa ξ
Esta é a expressão da identidade de Ward para o funcional gerador das funções de
Green conexas. Normalmente as identidades de Ward são mais úteis para o funcional
gerador das funções de Green irredutı́veis que é definido por
Z
Γ[Aµ , φi , ω, ω, Kµ , Ki , L] ≡ W [Jµ , Ji , η, η, Kµ , Ki , L] − d4 x[Jµa Aaµ + Ji φi + ηω + ωη]
(6.247)
com as relações habituais
δW ωa = δW
φi = δη a
δJi (6.248)
Aaµ = δW ω a = − δWa
δJ µa δη
e as relações inversas
Ji = − δΓ δΓ ηa =
δφi δω a (6.249)
a
Jµ = − µa δΓ δΓ
δA ηa = − a
δω
Como a transformada de Legendre deixa inertes as fontes Kµ , Ki e La devemos ter
a
δW δΓ δW δΓ δW δΓ
a
= ; = ; = (6.250)
δKµ δKµa δKi δKi δL a δLa
Obtemos então facilmente
Z
4 δΓ δΓ δΓ δΓ δΓ δΓ 1 a δΓ
d x + − − F = 0 (6.251)
δKµa (x) δAµa (x) δKi (x) δφi (x) δLa (x) δω a (x) ξ δω a (x)
Esta equação é o funcional gerador das identidades de Ward para uma teoria de gauge
não abeliana numa gauge linear. As identidades de Ward para funções de Green especı́ficas
obtém-se por derivação funcional em ordem aos campos apropriados.
Na prática a equação anterior usa-se em ligação com outra identidade funcional, a
equação de movimento (ou de Dyson- Schwinger) para os fantasmas. Esta pode ser obtida
fazendo a seguinte mudança de variáveis no integral funcional (ver capı́tulo 5),

 δAaµ = δφi = δω a = 0
(6.252)
 δω a = f a = constante infinitesimal
Então
Z
δΣ
δZ = 0 = a
D(· · · ) i a + iη f a ei(Σ+fontes) (6.253)
δω
mas
δΣ
= −Mab ω b (x) = −sFa (x)
δω a (x)
Z
δFa (x) b
4 δFa (x)
= − d y sA (y) + sφi (y)
δAbµ (y) µ δφi (y)
Z
4 δFa (x) δΣ δFa (x) δΣ
= − d y + (6.254)
δAbµ (y) δK bµ (y) δφi (y) δKi (y)
e portanto obtemos
Z Z
4 δFa (x) δΣ δFa (x) δΣ
0 = D(· · · ) −i d y + + iη (x) ei(Σ+fontes)
a
Z
4 δFa (x) δ δFa (x) δ
= − d y + + iη (x) eiW .
a
(6.255)
Usando agora
δΓ
ηa = − (6.256)
δω a
obtemos finalmente (para gauges lineares)
Z
4 δFa (x) δΓ δFa (x) δΓ δΓ
d y b µb
+ =− a (6.257)
δAµ (y) δK (y) δφi (y) δKi (y) δω (x)
que é o funcional gerador das equações de Dyson-Schwinger para os fantasmas.
6.3.3 Example: Transversality of vacuum polarization

Vamos aqui dar um exemplo de aplicação das identidades de Ward mostrando que a
polarização do vácuo é transversal. Como a teoria de gauge pura já é não trivial vamos
somente considerar este caso, as generalizações são imediatas. Para mostrar os detalhes
dos cálculos vamos fazer este exemplo usando dois métodos. O primeiro, que chamaremos
método formal, consiste na aplicação das identidades de Ward para o funcional Γ que
acabámos de mostrar, o segundo é o método prático que resulta da aplicação dos resultados
do teorema 2.7. A comparação dos dois métodos será importante para a compreensão das
expressões.
i) Método formal
Como estamos a considerar uma teoria de gauge pura a expressão para o funcional
gerador das identidades de Ward para o funcional Γ é
Z
δΓ δΓ δΓ δΓ 1 a δΓ
d4 x − − F (x) =0 (6.258)
δKµa (x) δAµa (x) δLa (x) δω a (x) ξ δω a (x)
onde vamos escolher a gauge covariante
F a (x) = ∂µ Aaµ (x) (6.259)

δΓ δΓ
Para prosseguir é necessário saber o que representam δKµa e δLa . Da forma como foram
introduzidos temos
δΓ δW δ 1 δZ
= = ln Z =
δKµa (x) δKµa iδKµ a Z iδKµa (x)
Z
1
= D(· · · )sAaµ (x)ei(Σ+fontes) (6.260)
Z
Como sAaµ (x) = Dµab ω b = ∂µ ω a (x) − gf abc ω b (x)Acµ (x), obtemos então
δΓ x1 δZ abc 1 δ2 Z
= ∂µ − gf (6.261)
δKµa (x) Z iδη a (x) Z iδJµc (x)iδη b (x)
Introduzindo Z ≡ exp(iW ), a equação 6.261 escreve-se

" #
δΓ µ δ(iW ) abc δ2 iW δiW δiW
= ∂x − gf + (6.262)
δKµa (x) iδη a (x) iδJµc (x)iδη b (x) iδJµc (x) iδη b (x)
que tem a seguinte representação diagramática:
iW
b
δΓ b
a
= ∂xµ iW − gf abc µ iW − gf abc µ (6.263)
δKµa (x) c
c
iW
onde W é o funcional gerador das funções de Green conexas. De igual modo se pode
mostrar
δΓ 1 1 δ2 Z
= g f abc
δLa (x) 2 Z iδη c (x)iδη b (x)

1 abc δ2 (iW ) δ(iW ) δ(iW )
= gf + (6.264)
2 iδη c (x)iδη b (x) iδη c (x) iδη b (x)
ou diagramaticamente
iW
b
δΓ 1 b 1
= g f abc µ iW + g f abc µ (6.265)
δLa (x) 2 c 2
c
iW
Posto isto voltemos ao problema de provar a transversabilidade do vácuo. Olhando

para a expressão inicial é fácil de ver que temos que aplicar δ2 à equação de
δω (y)δAcν (z)
b
partida. Temos sucessivamente

δ2 δΓ δΓ δ2 Γ δ2 Γ
= (6.266)
δω b (y)δAcν (z) δKµa (x) δAµa (x) =0 δω b (y)δKµa (x) =0 δAcν (z)δAµa (x) =0
mas

δ2 Γ
=0
δω (y)δKµ (x) =0
b a
Z
δ2 Γ δ2 Γ
4
= d w −i b
f
δω (y)δω (w) iδη (w)δKµ (x) =0
f a
Z
δ2 Γ δ2 (iW )
= ∂xµ d4 w −i b
δω (y)δω f (w) iδη f (w)iδη (x) =0
a
Z !
δ 2Γ δ 3 iW
′
−g f ab c d4 w −i b ′
δω (y)δω f (w) iδη f (w)iδη b (x)iδJµc (x)
=0
Z
′ δ2 Γ
= ∂xµ δ4 (x − y)δab − gf ab c d4 w −i
δω b (y)δω f (w)
!
δ3 iW

′ (6.267)
iδη f (w)iδη b (x)iδJµc (x)
=0
De modo semelhante

δ2 δΓ δΓ
=0 (6.268)
δω b (y)δAcν (z) δLa δω a =0
e

δ2 1 δΓ 1 ν 4 δ2 Γ
ρa
∂ρ A (x) a = ∂x δ (x − z) (6.269)
δω b (y)δAcν (z) ξ
δω (x) =0 ξ δω (y)δω (x) =0
b a
Usando estes resultados obtemos
Z
δ2 Γ δ2 Γ
−∂µy − gf ade
d4
xd 4
w −i
δAbµ (y)δAcν (z) δω b (y)δω f (w)
!
δ3 iW δ2 Γ 1 ν δ2 Γ
+ ∂ =0
iδη f (w)iδη d (x)iδJµe (x) δAaµ (x)δAcν (z) ξ z δω b (y)δω c (z)
(6.270)
Aplicando transformadas de Fourier, com a convenção da Figura 6.2, obtemos
i
−ipµ (i)G−1cb
νµ (p) − gf
ade
iG−1ca
νµ (p)∆
−1f b µdef
X + (−ipν ) ∆−1cb (p) = 0 (6.271)
ξ
ou ainda
1 −1cb
pµ G−1cb
νµ = − ∆ pν + ig f ade G−1ca
νµ (p) ∆
−1f b µdef
X (6.272)
ξ
p
z y
Figure 6.2: Momentum definition for the Fourier Transform.
onde
h i
X µdef = TF < 0|T ω d (x)ω f (w)Aµe (x)|0 >c
d f
≡ µ iW (6.273)
e
Para demonstrar a transversabilidade precisamos ainda da equação de movimento para

os fantasmas que é, para o nosso caso,
δΓ δΓ
= −∂zµ (6.274)
δω a (z) δK µa (z)
Aplicando o operador δ , obtemos

δω b (y)
δ2 Γ
⊓ δab δ4 (y − z)
= −⊔
δω b (y)δω a (z)
Z !
δ2 Γ δ3 iW
+gf adc d4 w −i b ∂zµ
δω (y)δω f (w) iδJµc (z)iδη f (w)iδη d (z)
(6.275)
Aplicando a transformada de Fourier, obtemos
i∆−1ab = p2 δab + gf adc (−ipµ )Xµdcf ∆−1f b (6.276)

As equações 6.272 e 6.276 permitem mostrar a transversabilidade do vácuo. Para isso
escrevamos
a ab
G−1ab −1 ab
µν = GT µν + i δ pµ pν (6.277)
ξ
onde pµ G−1 ab
T µν = 0. Para o propagador livre a = 1. Para mostrar que a polarização do
vácuo é transversal basta mostrar que a parte longitudinal não é renormalizada e portanto
que o valor de a continuar a ser a = 1. Usando
a ab 2
pµ G−1ab
µν = i δ p pν (6.278)
ξ
e multiplicando a equação 6.272 por pν obtemos
a 1 a
i p4 δcb = − p2 ∆−1cb − p2 g f cde pµ X µdef ∆−1f b (6.279)
ξ ξ ξ
Usando agora a equação 6.276 obtemos depois de alguma álgebra trivial
1 a
0 = − p2 ∆−1cb + p2 ∆−1cb (6.280)
ξ ξ
o que implica
a=1 (6.281)
como querı́amos mostrar.
ii) Método prático
Vamos agora mostrar a transversabilidade da polarização do vácuo usando o método
prático baseado nos resultados do Teorema 2.7. Como
1
sω b (x) = ∂µ Aµb (x) (6.282)
ξ
e
sAaν = ∂ν ω a − gf adc ω d Acν (6.283)

é fácil de ver que a função de Green de partida deverá ser < 0|T Aaν (x)ω b (y)|0 >. Então o
teorema diz-nos que
s < 0|T Aaµ (x)ω b (y)|0 >= 0 (6.284)

ou seja
1
< 0|T Aaν (x)∂µ Aµb (y)|0 > = < 0|T ∂ν ω a (x)ω b (y)|0 >
ξ
−gf adc < 0|T ω d (x)Acν (x)ω b (y)|0 > (6.285)
Aplicando a transformação de Fourier obtemos
i ρ ab
p Gνρ (p) = −ipν ∆ab (p) − gf adc Xνdcb (6.286)
ξ
onde Xνdcb foi definido anteriormente. Multiplicando por G−1νµ ∆−1 obtemos
1
pµ G−1ac
νµ = − pν ∆−1ac + igf f de χdeb
ν ∆
−1bc −iνµaf
G (6.287)
ξ
que é precisamente a equação 6.272.

A equação 6.276 pode ser obtida facilmente sabendo que o único vértice dos fantasmas
é
µ, c
a b
gf abc pµ (6.288)
p
Então
a b a b a d b
iW = + iW (6.289)
p µ
c
ou seja
i ab i
∆ab (p) = 2
δ + 2 gf adc pµ Xµdcb (6.290)
p p
ou ainda
′ ′
i∆−1ab = p2 δab − igf adc pµ Xµdcb ∆−1b b (6.291)
que é precisamente a equação 6.276. A demonstração da transversabilidade é agora igual
a i).
6.3.4 Gauge invariance of the S matrix

Mostrámos na secção 6.2.5 a invariância da gauge da matriz S, usando o teorema da
equivalência e o facto que os funcionais geradores correspondentes a condições de gauge
diferentes diferiam somente no termo das fontes. A demonstração que fizemos usava as
propriedades especı́ficas de gauge de Coulomb e podia levantar alguma dúvida quanto à
sua validade geral.
Vamos aqui mostrar, usando as identidades de Ward, que os funcionais ZF e ZF +∆F
correspondentes às condições de gauge F e F + ∆F , respectivamente, diferem somente no
termo das fontes. Como F e ∆F são arbitrários, e demonstração é geral. Temos
Z R 4 a µa
ZF [Jµ ] = D(· · · )ei[Sef f + d x(Jµ A +Ji φi )]
a
(6.292)
Então
Z Z " Z a
1 δ∆F(x)
ZF +∆F − ZF = d x i − F a ∆F a − ω a d4 y b
D(· · · ) 4
sAbµ (y)
ξ δAµ (y)
Z a

a 4 δ∆F (x)
−ω d y sφi (y) ei(Sef f +fontes) (6.293)
δφi (y)
Usamos agora as identidades de Ward na forma correspondente ao funcional Z, isto é
Z Z
a µa +J
0= D(· · · ) d4 x[J µa sAaµ + J i sφi + ηsω − sωη] e{i(Sef f +Jµ A i φi +ωη+ηω)}
(6.294)
Derivando em ordem a η a (x) e pondo as fontes dos fantasmas nulas obtemos
Z Z
1 R 4 a µi
0= D(· · · ) F a (x) + iω a (x) 4
d y[J µb
sAbµ + J sφi ] ei[Sef f + d x(Jµ A +Ji φi )]
i
ξ
(6.295)
ou ainda
h i Z
−1F a δ D(· · · )ei(Sef f +fontes) =
ξ iδJ
Z Z (6.296)
= D(· · · )iω (x) d4 y[J µb sAbµ + Ji sφi ]ei(Sef f +fontes)
a
Então
Z
D(· · · ) − 1 F a ∆F a ei(Sef f +fontes) =
ξ
h i h i Z
= ∆F a δ 1
− F a δ D(· · · )ei(Sef f +fontes)
iδJ ξ iδJ
h iZ Z
= ∆F a δ D(· · · )iω a (x) d4 y[J µb sAbµ + J i sφi ]ei(Sef f +fontes) (6.297)
iδJ
Z ( Z " #
a a
δ∆F (x) δ∆F (x)
= D(· · · ) ω a (x) d4 y sAbµ (y) + sφi (y)
δAbµ (y) δφi (y)
Z
4 µb b i
+iω (x)∆F (x) d y[J sAµ + J sφi ] ei(Sef f +fontes)
a a
Portanto
Z Z " #!
1 a a a 4 δ∆F a (x) b δ∆F a
D(· · · ) − F ∆F −ω (x) d y b
sAµ (η)+ sφi (y) ei(Sef f +fontes)
ξ δAµ (y) δφ i (y)
Z Z h i
= D(· · · )iω a (x)∆F a (x) d4 y J µb sAbµ + J i sφi ei(Sef f +fontes)
(6.298)
Podemos então escrever
ZF +∆F −ZF
Z Z h Z i
4
= D(· · · )i d x iω (x)∆F (x) d4 y(J µb sAbµ + Ji sφi ) ei(Sef f +fontes)
a a
Z
Z
i Sef f + d4 y[Jµa (y)Aµa (y) + Ji Φi (y)]
= D(· · · )e
(6.299)
onde
Z
Φi (y) ≡ φi (y) + i d4 x[ω a (x)∆F a (x)sφi (y)] (6.300)
e
Z
Aaµ (y) ≡ Aaµ (y) + i d4 x[ω b (x)∆F b (x)sAaµ (y)] (6.301)
A diferença entre os funcionais geradores ZF +∆F e ZF é apenas na forma funcional

dos termos das fontes. Podemos portanto usar o teorema da equivalência para mostrar
que as matrizes S renormalizadas são iguais nos dois casos
SFR+∆F = SFR . (6.302)
6.4 Ward-Takahashi identities in QED

6.4.1 Ward-Takahashi identities for the funcional Z[J]
Vamos aqui tornar a derivar as identidades de Ward para QED já encontradas no estudo
da renormalização, mas utilizando agora os métodos funcionais.
Pode-se mostrar que para o funcional gerador das funções de Green completas se pode
escrever, numa gauge linear,
Z R 4 µ
Z(Jµ , η, η) = D(Aµ , ψ, ψ)ei(Sef f + d x(Jµ A +ηψ+ψη) (6.303)
onde Jµ , η e η são as fontes associadas a Aµ , ψ e ψ respectivamente. A acção efectiva é

dada por
Z
4 1 2
Sef f = d x LQED − (∂ · A) = SQED + SGF (6.304)
2ξ
onde
1
LQED = − Fµν F µν + ψ(iγ µ Dµ − m)ψ . (6.305)
4
SQED é invariante para as transformações de gauge do grupo U (1)

 δAµ = ∂µ Λ
δψ = −ieΛψ

δψ = ieΛψ
enquanto que Sef f contém a parte de gauge fixing que não é invariante nas transformações
6.306. Portanto as identidades de Ward tomam aqui a forma

δSGF δ δ
+ Ji Fi Z(J) = 0 (6.306)
δφi iδJ iδJ
o que se escreve no nosso caso, reintroduzido as integrações,,
Z
1 δ δ δ
0= d x ∂ µ ∂ν
4 µ
∂µ Λ + J ∂µ Λ − ieΛη + ieΛη Z(J µ , η, η) (6.307)
ξ iδJν iδη iδη
ou seja, integrando por partes,
Z
4 1 δ µ δ δ
d xΛ − ⊔⊓∂ν − ∂µ J − ieη + ieη Z(J µ , η, η) = 0 (6.308)
ξ iδJν iδη iδη
Podemos ainda escrever

1 δ µ δ δ
⊔
⊓∂µ + ∂µ J + ieη − ieη Z(J, η, η) = 0 (6.309)
ξ iδJµ iδη iδη
6.4.2 Ward-Takahashi identities for the functionals W and Γ

Do ponto de vista das aplicações é mais útil a identidade de Ward em termos do funcional
gerador das funções de Green próprias. Este problema para as Teorias de Gauge não
abelianas é bastante difı́cil de resolver, conforme veremos no próximo capı́tulo. Aqui o
problema é simples pois a equação acima é linear nas derivadas funcionais em relação às
diversas fontes (note-se que se se tivesse escolhido uma condição de gauge não linear isto
já não seria verdade, mesmo em QED). Esta linearidade permite escrever imediatamente

µ 1 δ δ δ
∂µ J + ⊔ ⊓∂µ + ieη − ieη W (J, η, η) = 0 (6.310)
ξ iδJµ iδη iδη
onde W é o funcional gerador das funções de Green conexas
µ ,η,η)
Z(J µ , η, η) ≡ eiW (J (6.311)
Como vimos, o funcional gerador das funções de Green próprias é
Z
Γ(Aµ , ψ, ψ) = W (Jµ , η, η) − d4 x[J µ Aµ + ηψ + ψη] (6.312)
onde
δW δW δW
Aµ = µ
; ψ= ; ψ=− (6.313)
iδJ iδη iδη
e
δΓ δΓ δΓ
Jµ = − µ
; η=− ; η= (6.314)
δA δψ δψ
onde, como habitualmente, as derivadas fermiónicas são derivadas esquerdas. Então a

equação 6.310 pode ser escrita
1 δΓ δΓ δΓ
⊓∂µ Aµ − ∂µ
⊔ + ie ψ + ieψ =0 (6.315)
ξ δAµ δψ δψ
Esta equação é o ponto de partida para gerar todas as identidades de Ward em QED.
A sua aplicação é muito mais fácil do que a expressão equivalente usada no estudo da
renormalização e que foi demonstrada utilizando o formalismo canónico. Os métodos
funcionais tornam estas expressões particularmente simples.
6.4.3 Example: Ward identity for the QED vertex

Para nos convencermos que a equação 6.315 conduz às identidades de Ward já nossas con-
δ2
hecidas, vamos obter a identidade de Ward para o vértice em QED. Apliquemos δψ (y)δψ (z)
α β
a 6.315. Obtemos então
δ3 Γ
∂xµ
δψα (y)δψ β (z)δAµ (x)
" #
δ2 Γ 4 δ2 Γ 4
= ie δ (z − x) − δ (y − x) (6.316)
δψα (y)δψ β (x) δψα (x)δψ β (z)
ou seja

∂xµ Γµβα (x, z, y) = ie Γβα (x, y)δ4 (z − x) − Γβα (z, x)δ4 (y − x) (6.317)
Aplicando agora a transformada de Fourier a ambos os membros, com os momentos
definidos de acordo com a Figura 6.3
q = p′ − p
p′ p
β α
Figure 6.3: Definition of the momenta for the vertex.
obtemos, omitindo os ı́ndices spinoriais,
q µ Γµ (p′ , p) = ie[S −1 (p) − S −1 (p′ )] (6.318)

que é exactamente a identidade pretendida.
6.4.4 Ghosts in QED

Anteriormente dissemos que para QED o funcional gerador é dado por
Z R 4 µ
Z(Jµ , η, η) = D(Aµ , ψ, ψ)ei d x[LQED +LGF +Jµ A +ηψ+ψη] (6.319)
onde LQED é o Lagrangeano usual de QED e o termo que fixa a gauge é dado por
1
LGF = − (∂ · A)2 . (6.320)
2ξ
De facto isto não é estritamente verdade. Como veremos o funcional gerador que
obterı́amos se usássemos a prescrição para teorias de gauge seria
Z
d4 x[Lef f +J µ Aµ +ηψ+ψη+ωζ+ζω]
R
Z̃(Jµ , η, η, ζ, ζ) = D(Aµ , ψ, ψ, ω, ω)ei (6.321)
onde ω e ω são campos escalares anticomutativos. Estas partı́culas fictı́cias são designadas
por fantasmas de Fadeev-Popov e desempenham um papel fulcral em teorias de gauge não
abelianas. Embora não apareçam como estados finais em processos fı́sicos, a introdução
de fontes para elas é conveniente para discutir as identidades de Ward. No Lagrangeano
anterior Lef f é dado por
Leff = LQED + LGF + LG (6.322)

onde
LG = −ω ⊔
⊓ω (6.323)
A razão pela qual em QED podemos trabalhar com o funcional gerador Z e não Z̃ tem
a ver com o facto de os fantasmas em QED não terem acoplamentos com as partı́culas
fı́sicas e poderem ser omitidos completamente da teoria (ver problema 1.6).
Vamos introduzir agora as transformações de Becchi Rouet e Stora (BRS). O objectivo
destas transformações é fazer com que Lef f seja invariante. É fácil de ver que para QED
obtemos esse resultado com as transformações


 δψ = −ieωθψ





 δψ = ieψωθ

δAµ = δµ ωθ (6.324)





 δω = 1ξ (∂ · A)θ


 δω = 0
onde o parâmetro θ é anticomutativo (variável de Grassman). As transformações nos cam-

pos fı́sicos são transformações de gauge de parâmetro Λ = ωθ pelo que LQED é invariante.
As transformações em ω e ω são tais que a variação de LGF cancela a de LG . A invariância
da medida de integração e de Sef f permite imediatamente escrever as identidades de Ward,
para os funcionais geradores (ver problema 1.6).
As transformações BRS permitem obter as identidades de Ward duma forma expedita

sem ter que recorrer à derivação funcional de Γ̃. Este método baseia-se no facto de que o
operador δBRS aplicado a qualquer função de Green é zero (ver problema 1.6), isto é
δBRS h0| T Aµ1 · · · ω · · · ω · · · ψ · · · ψ · · · |0i = 0 (6.325)

Vejamos duas aplicações simples do método:
i) A parte longitudinal do propagador do fotão não é renormalizada
Este resultado é equivalente, como é sabido, a dizer que a polarização do vácuo é
transversal. Prova-se facilmente partindo da função de Green h0| T Aµ ω |0i e fazendo uso
de 6.325.
δBRS h0| T Aµ ω |0i = 0 (6.326)

ou seja
1
h0| |T Aµ ∂ ν Aν |0i θ − h0| T ∂µ ωω |0i θ = 0 (6.327)
ξ
Portanto
1 µ
k Gµν (k) = −kν ∆(k) (6.328)
ξ
usando para propagador dos fantasmas
i
∆(k) = (6.329)
k2
pois o fantasma não tem interacções. Multiplicando pelo inverso do propagador do fotão
obtemos
1 µ kν
k = −i 2 G−1νµ (k) (6.330)
ξ k
e portanto
i
kν G−1νµ (k) = kµ k2 = kν G−1νµ
(0) (k) (6.331)
ξ
o que mostra que a parte longitudinal não é renormalizada.
ii) Identidade de Ward para o vértice
Partimos de
δBRS h0| T ωψψ |0i = 0 (6.332)

Então
1
h0| T ∂ µ Aµ ψψ |0i = ie h0| T ωωψψ |0i − ie h0| T ωψψω |0i (6.333)
ξ
ou ainda
i µ
q Tµ = T (6.334)
ξ
6.5. UNITARITY AND WARD IDENTITIES 229
onde
p′
µ q
iTµ = = Gµν (q)S(p′ )iΓν S(p) (6.335)
p
p′ p′
q q
iT = −ie + ie
p p
= −ie∆(q)S(p) + ie∆(q)S(p′ ) (6.336)
A última igualdade resulta dos fantasmas não terem interacções em QED numa gauge
linear. Pondo tudo junto obtemos
i µ
q Gµν (q)S(p′ )iΓν S(p) = −ie∆(q)S(p) + ie∆(q)S(p′ ) (6.337)
ξ
Usando
1 µ
k Gµν (k) = −kν ∆(k) (6.338)
ξ
e multiplicando pelos inversos dos propagadores dos fermiões obtemos finalmente a iden-
tidade pretendida

qµ Γµ (p′ , p) = ie S −1 (p) − S −1 (p′ ) (6.339)
6.5 Unitarity and Ward identities

6.5.1 Optical theorem
A matriz S pode-se escrever na forma
S = 1 + iT (6.340)
Então a unitariedade, SS † = 1 implica
2 Im T = T T † (6.341)
Se inserirmos esta relação entre o mesmo estado inicial e final obtemos
2 Im < i|T |i > = < i|T T † |i >

X
= | < f |T |i > |2 (6.342)
f
onde introduzimos um conjunto completo de estados. Esta relação pode ainda escrever-se
na forma
σtotal = 2 Im T elástica
frente (6.343)
conhecida por teorema óptico. O que chamamos aqui σtotal não é exactamente a secção
eficaz porque faltam os factores de fluxo. É rigorosamente a quantidade definida por
X
σtotal ≡ | < f |T |i > |2 (6.344)
f
A unitariedade estabelece portanto uma relação entre a secção eficaz total e a parte
imaginária da amplitude elástica na direcção frontal (o estado inicial e final têm que ser o
mesmo).
6.5.2 Cutkosky rules

Para mostrar que a unitariedade é respeitada num dado processo é necessário saber calcular
a parte imaginária de diagramas. Claro que há sempre a possibilidade de fazer as contas
explı́citas até ao fim e ver qual foi a parte imaginária que ficou, mas este processo não é
muito conveniente para diagramas complicados.
Assim existem regras, chamadas regras de Cutkosky que nos dão simplesmente a parte
imaginária duma amplitude qualquer. Estas regras são:
Regra 1:
A parte imaginária duma amplitude obtém-se através da expressão
X
2 Im T = − T (6.345)
cortes
Regra 2:
O corte obtém-se escrevendo a amplitude iT = · · · e substituindo nesta expressão os
propagadores das linhas cortadas pelas seguintes expressões:
i) Campos Escalares
∆(p) =⇒ 2πθ(po )δ(p2 − m2 ) (6.346)

ii) Campos Spinoriais
S(p) ⇒ (p/ + m)2πθ(po )δ(p2 − m2 ) (6.347)
iii) Campos Spin 1 (na gauge de Feynman)
Gµν (p) ⇒ −gµν 2πθ(p0 )δ(p2 − m2 ) (6.348)
Nestas expressões as funções θ asseguram o fluxo da energia. As regras de Cutkosky

são um pouco complicados de mostrar em geral8 mas nós vamos aqui mostrar dois casos
e verificá-las explicitamente
Exemplo 2.1 Propagador livre
A amplitude é
i
iT = (6.349)
p2 − m2 + iε
A parte imaginária obtém-se explicitamente usando

1 1
=P − iπδ(x) (6.350)
x + iε x
logo

1
T =P − iπδ(p2 − m2 ) (6.351)
p 2 − m2
e portanto
2 Im T = −2πδ(p2 − m2 ) (6.352)
Pela regra de Cutkosky obtemos imediatamente
2 Im T = −2πδ(p2 − m2 )θ(p0 ) (6.353)

que é o mesmo resultado. A função θ(p0 ) assegura que o fluxo da energia é da esquerda
para a direita.
λ 3
Exemplo 2.2: Self-energy em 3! φ
Consideremos a self-energy na teoria dada por
1 1 λ
L = ∂µ φ∂ µ φ − m2 φ2 − φ3 (6.354)
2 2 3!
O diagrama de self-energy é o representado na Figura 6.4. A amplitude correspondente é
Z
2 d4 p i i
iT = (iλ) (6.355)
(2π)4 p2 − m2 + iε (p − k)2 − m2 + iε
Calculemos a parte imaginária de T por dois métodos, primeiro explicitamente e depois
usando a regra de Cutkosky.
8
Para um tratamento mais completo ver G. ’t Hooft, ”Diagrammar”, CERN Report 1972.
k k
p−k
Figure 6.4: Self-energy
i) Cálculo explı́cito
Z
2 d4 p 1
iT = λ
(2π) (p − m + iε)[(p − k)2 − m2 + iε]
4 2 2
Z Z 1
2 d4 p 1
= λ 4
dx 2
(2π) 0 (p + 2p · P − M 2 + iε)2
Z Z 1
2 d4 p 1
= λ dx (6.356)
(2π)4 0 [(p + P )2 − ∆]2
onde

 P = −x k
(6.357)
 ∆ = P 2 + M 2 = m2 − k2 x(1 − x) − iε
Então
Z Z 1
d4 p 1
iT = λ2 dx (6.358)
(2π)4 0 (p2 − ∆)2
O integral é divergente. Fazendo regularização dimensional obtemos finalmente
Z 1
λ ε d d
T = 2
µ Γ 2− dx∆−(2− 2 ) (6.359)
16π 2 0
Para prosseguir temos que impor um esquema de renormalização. Fazendo renormali-
zação on-shell, TR (k2 = m2 ) = 0, obtemos
TR = T − T (k2 = m2 ) "
Z − 2ε − 2ε #
λ2 ε 1 ∆(k2 ) ∆(k2 = m2 )
= Γ dx −
16π 2 2 0 µ2 µ2
Z 1
λ2 2 ε m2 − k2 x(1 − x) − iε
= − C + O(ε) dx 1 − 1 − ln 2
16π 2 ε 0 2 m − m2 x(1 − x) − iε
Z 1
λ2 1 − βx(1 − x) − iε
= − dx ln
16π 2 0 1 − x(1 − x) − iε
λ2
= − [L(β) − L(1)] (6.360)
16π 2
k2
onde β = m2 e a função L(β) é definida por
Z 1
L(β) ≡ dx ln 1 − β(1 − x)x − iε (6.361)
0
e satisfaz
r
4
Im L(β) = −π 1− θ(β − 4) (6.362)
β
Então
λ2
Im T = − 2
Im L(β) − Im L(1) (6.363)
16π
e obtemos finalmente
r
λ2 4m2 4m2
Im T = 1− 2 θ 1− 2 (6.364)
16π k k
A função θ assegura que só há parte imaginária quando o estado intermédio puder ser
final (produção de 2 partı́culas de massa m).
ii) Cálculo usando as Regras de Cutkosky
Usando as regras obtemos
Z
2 d4 p
2 Im T = −(iλ) (2π)2 θ(p0 )θ(k0 − p0 )δ(p2 − m2 )δ((p2 − k2 ) − m2 )
(2π)4
Z
d4 p 4 ′
= λ2 d p (2π)2 θ(p0 )θ(k0 − p0 )δ(p2 − m2 )δ(p′2 − m2 )δ4 (p′ − k + p)
(2π)4
(6.365)
Usando agora
Z Z
4 0 2 2 1
d pθ(p )δ(p − m ) = d3 p (6.366)
2p0
obtemos
Z
d3 p 3 ′ 1 1
2 Im T = λ2 d p 0 ′0 2πδ4 (p′ − k + p) (6.367)
(2π)3 2p 2p
ou ainda
Z
2 d3 p 1 1
2 Im T = λ 2πδ(k0 − p0 − p′0 ) (6.368)
(2π)3 2p0 2p′0
No referencial do centro de massa
√ p p
k = ( s, ~0) ; p = ( |~
p|2 + m2 , p~) ; p′ = ( |~
p′ |2 + m2 , −~
p) (6.369)
e portanto
Z p
d3 p 1 √
2 Im T = λ2 3 2 2
2πδ( s − 2 |~ p|2 + m2 )
(2π) 4(|~ p| + m )
Z p
λ2 p|2
|~ p| − 4s − m2 )
δ(|~ 4m2
= p| 2
d|~ θ 1−
4π p | + m2
|~ √ 2|~p| s
p|2 +m2
|~
r
λ2 4m2 4m2
= 1− θ 1− (6.370)
8π s s
Logo usando k2 = s obtemos

r
λ2 4m2 4m2
Im T = 1− 2 θ 1− 2 (6.371)
16π k k
que é de facto o mesmo resultado que 6.364.
6.5.3 Example of Unitariedade: scalars and fermions

Consideremos a teoria descrita pelo seguinte Lagrangeano
1 1
L = iψ∂/ψ − mψψ + ∂µ φ∂ µ φ − M 2 φ2 + gψψφ (6.372)
2 2
Vamos mostrar a unitariedade em 2 casos em que as linhas cortadas são fermiónicas
i) Self-energy dos escalares
A self energy dos escalares e dada pelo diagrama da Figura 6.5, a que corresponde a
amplitude
k k
p−k
Figure 6.5: Fermion contribution to the scalar self-energy.
Z
2 d4 p i i
iT = g Tr (6.373)
(2π)4 p/ − m + iε p/ − k/ − m + iε
logo
X
2 Im T = − T
cuts
Z
d4 p
= −g2 T r[(p/ + m)(p/ − k/ + m)](2π)θ(p0 )δ(p2 − m2 )·
(2π)4
(2π)θ(k0 − p0 )δ((p − k)2 − m2 ) (6.374)
Para mostrarmos a unitariedade calculemos
2
p

X k
σ= (6.375)

f

p′
ou seja
X X
σ= |igu(p)v(p′ )|2 = −g 2 T r[(p/ + m)(−p/′ + m)] (6.376)
f f
P P
onde se usou spins v(p′ )v(p) = −(−p/′ + m) e spins u(p)u(p
′) = p/ + m. Logo
Z
2
σ = −g dρ2 Tr[(p/ + m)(−p/′ + m)] (6.377)
onde dρ2 é o espaço de fase de duas partı́culas, isto é
Z Z 3
d p d3 p′ 1 1 4 4 ′
dρ2 ≡ 3 3 0 ′0 (2π) δ (k − p − p )
(2π) (2π) 2p 2p
Z 4
d p d4 p′
= (2π)θ(p0 )δ(p2 − m2 )(2π)θ(p′0 )δ(p′2 − m2 )(2π)4 δ4 (k − p − p′ )
(2π)4 (2π)4
(6.378)
Daqui se conclui que
Z
2 d4 p
σ = −g (2π)θ(p0 )δ(p2 − m2 )(2π)θ(k0 − p0 )δ((1 − k)2 − m2 )Tr[(p/ + m)(p/ − k/ + m)]
(2π)4
(6.379)
e obtemos portanto finalmente
2 Im T = σ (6.380)
ii) Caso geral
Consideremos o caso geral com 2 linhas internas de fermiões. A amplitude iT é repre-
sentada pelo seguinte diagrama
k1 k1
k2 k2
p
≡ iT
−p′
n
X
kn ′ kn
p = ki − p (6.381)
i=1
A amplitude iT escreve-se
Z h i
d4 p ′ S(p)T ′ S(−p′ )
iT = − Tr T (6.382)
(2π)4
onde a amplitude iT ′ é, por sua vez, definida pelo diagrama seguinte
k1
k2
p
≡ u(p)iT ′ v(p′ ) (6.383)
−p′
kn
Então
Z
d4 p
2 Im T = − (2π)2 δ(p2 − m2 )θ(p0 )δ(p′2 − m2 )θ(p′0 )·
(2π)4
h i
Tr T ′ (p/ + m)T ′ (−p/′ + m)
Z h i
= − dρ2 Tr T ′ (p/ + m)T ′ (−p/′ + m) (6.384)
Por outro lado

2

k1

k2
X p

σ =

f
−p′

kn
X
= |u(p)T ′ v(p′ )|2
f
Z h i
= − dρ2 Tr (p/ + m)T ′ (−p/ + m)T ′ (6.385)
e portanto
σ = 2 Im T (6.386)
Se as linhas cortadas fossem de escalares em vez de fermiões o resultado seria o mesmo,

não haveria o sinal menos do loop mas também não haveria o sinal menos da soma dos
spins (ver problema 2.8).
6.5.4 Unitarity and gauge fields
Na secção 6.5.3 demonstrou-se a unitariedade das teorias com escalares e spinores. Vamos
aqui ver que a demonstração da unitariedade para o caso dos campos de gauge é mais
complicada e exige o uso das identidades de Ward. O problema reside no facto que os
campos de gauge em linhas internas podem ter polarizações não fı́sicas enquanto que
no estado final o não podem. Esta diferença levaria a uma violação da unitariedade se
as linhas internas não pudessem ser também de fantasmas que compensam os graus de
liberdade a mais. Consideremos as seguintes amplitudes
p1 k1 p1 p1 k1 p1
iT = +
p2 k2 p2 p2 k2 p2
p1 k1
ab µ, a
iTµν =
ν, b
p2 k2
p1 k1
a
iT ab = b (6.387)
p2 k2
onde
k2 = p1 + p2 − k1 (6.388)
Então a amplitude escreve-se9
Z
d4 k1 1 ab aa′ ′ ∗a′ b′ µ′ ν ′ ′ ′ ′ ′
iT = Tµν Gµµ′ (k1 )Gbb
νν ′ (k2 )T − T ab ∆aa (k1 )∆bb (k2 )T ∗a b (6.389)
(2π)4 2
Aplicando as regras de Cutkosky a parte imaginária é
Z
d4 k1 1 ab ∗abµν
2 Im T = (2π)2 θ(k10 )θ(k20 )δ(k12 )δ(k22 ) T T − T ab T ∗ab
(2π)4 2 µν
Z
1 ab ∗abµν ab ∗ab
≡ dρ2 Tµν T −T T (6.390)
2
Calculemos agora σtot . Como os fantasmas não são fı́sicos teremos
9
O factor 1/2 é o factor de simetria dum loop com escalares. O sinal − é devido ao loop de fantasmas.
2

k1
X p1
µ, a
σ =
ν, b
p2 k2

Z X 2
1 µ ν ab
= dρ2 ε (k1 )ε (k )T
2 µν (6.391)
2
P ol
onde o factor 1/2 se deve agora a haver partı́culas idênticas no estado final. Pondo
X ′ ′
εµ (k1 )εµ ∗ (k1 ) = P µµ (k1 ) (6.392)
P ol
obtemos
Z
1 ab ∗ab µµ′ ′
σ= dρ2 Tµν Tµ′ ν ′ P (k1 )P νν (k2 ) . (6.393)
2
Usando agora o resultado do problema 2.10
kµ ην + kν ηµ
P µν (k) = −gµν + (6.394)
k·η
onde η µ é um 4-vector que satisfaz η · ε e η 2 = 0, obtemos
1 ab ∗ab µµ′ ′
Tµν Tµ′ ν ′ P (k1 )P νν (k2 ) =
2
1 ab ∗abµν 1 ab 1
= Tµν T − (T · k2 ) · (T ∗ab · η)
2 2 k2 · η
1 1 1 1
− (T ab · η) · (T ∗ab · k2 ) − (k1 · T ab ) · (η · T ∗ab )
2 k2 · η 2 k1 · η

1 ab ∗ab 1 1
− (η · T ) · (k1 · T ) + (k1 · T ab · η)(η · T ∗ab · k2 )+
2 k1 · η 2
1 1
+ (k1 · T ab · k2 )(η · T ∗ab · η) + (η · T ab · η)(k1 T ∗ab · k2 )
2 2

1 1
+ (η · T ab · k2 )(k1 · T ∗ab · η) (6.395)
2 (k1 · η)(k2 · η)
Fazendo uso das identidades de Ward (ver problema 2.11),
k1µ Tµν
ab = k T ab
2ν
=⇒ k1 · T ab · k2 = 0 (6.396)
k2µ Tµν
ab = k1ν T ab
obtemos
1 ab ∗ab ′ ′
T T ′ ′ P µµ (k1 )P νν (k2 ) =
2 µν µ ν
1 ab ∗abµν 1 ab 1
= Tµν T − T (k1 · T ∗ab · η)
2 2 k2 · η
1 1 1 1
− T ∗ab (k1 · T ab · η) − T ab (η · T ∗ab · k2 )
2 k2 · η 2 k1 · η
1 1 1 1
− (η · T ab · k2 )T ∗ab + T ab T ∗ab + T ab T ∗ab
2 k1 · η 2 2
1 ab ∗abµν
= T T − T ab T ∗ab (6.397)
2 µν
e portanto
Z
1 ab ∗abµν
σ= dρ2 T T − T ab T ∗ab (6.398)
2 µν
o que comparando com 6.390 dá
σ = 2 Im T (6.399)
Problems 241
2.1 Mostre que T (R) está relacionado com o operador de Casimir da representação R
C2 (R) através de
T (R)r = d(R)C2 (R) (6.400)
onde r é a dimensão do Grupo G e d(R) é a dimensão da representação R. O Casimir

C2 (R) é definido por
X
a a
Tik Tkj = δij ...C2 (R) . (6.401)
a,k
2.2 Mostrar que numa escolha diferente de condições auxiliares χiα = 0 conduz ao mesmo
resultado.
Sugestão: considere uma variação infinitesimal
χα + δχα = 0 α = 1, ...m (6.402)
Mostre então que
πα δ(ϕα )δ(χa ) det({ϕ, χ}) → πα δ(ϕα δ(χα + δχα ) det({ϕ, χ + δχ}) . (6.403)
2.3 Mostrar que para transformações infinitesimais
Z
~ a (x) = − 1 ~ a (x), αb (y)Cb (y)}
δE d3 y{E
y xo =yo
Z
~ a (x) = − 1 ~ a (x), αb (y)Cb (y)}
δA d3 y{A (6.404)
y xo =yo
isto é, as ligações Ca são os geradores infinitesimais das transformações de gauge

independentes do tempo.
2.4 Mostre que é sempre possı́vel encontrar uma gauge onde A3a = 0 a = 1, ...r .
2.5 Mostre os resultados expressos na equação 6.92.

2.6 Mostre que a parte imaginária não depende do esquema de renormalização, calculan-
do-a em M S e M S para o exemplo 2.2, isto é para a teoria descrita pelo Lagrangeano
da equação 6.354.
λ 3
2.7 Considere a teoria 3! φ do problema 2.6. Faça a demonstração da unitariedade para
a self-energy dessa teoria, isto é
2

X

2 Im = (6.405)

f

2.8 Considere a teoria descrita pelo Lagrangeano da Eq. (6.372). Refaça a demonstração
geral da unitariedade no caso dos estados intermédios serem escalares, isto é
k1 k1 2
k1

k2 k2 k2
p p
X
2 Im = (6.406)
−p′ f
−p
′

kn kn k
n
2.9 Mostre que o integral que resulta de cortar n linhas internas é igual ao integral do
espaço de fase de n partı́culas. Use este resultado para fazer uma demonstração
geral da unitariedade.
2.10 Mostre que
kµ ην + kν ηµ
P µν (k) = −gµν + (6.407)
k·η
onde kµ , εν (k, 1), ερ (k, 2) e η σ são quatro 4-vectores independentes e satisfazendo
η · ε(k, σ) = 0 σ = 1, 2
ε(k, 1) · ε(k, 2) = 0
k · ε(k, σ) = 0 σ = 1, 2
k2 = 0
η2 = 0 (escolha conveniente)
ε2 (k, σ) = −1 σ = 1, 2 (6.408)
Problems 243
Sugestão: A expressão mais geral para P µν é
P µν = ag µν + bk µ kν + cη µ η ν + d(kµ η ν + kν η µ ) . (6.409)
Use as relações anteriores para determinar a, b, c, d.

2.11 Demonstrar as identidades de Ward,
k1µ Tµν
ab = k T ab
2ν
=⇒ k1 · T ab · k2 = 0 (6.410)
k2µ Tµν
ab = k1ν T ab
ab e T ab são definidas em 6.387.
onde Tµν
a dos campos de Yang-Mills satisfaz as identidades de Bianchi:
2.12 Mostre que o tensor Fµν
b
Dµab Fρσ + Dρab Fσµ
b b
+ Dσab Fµρ =0 (6.411)
ou
Dµab ∗F µν b
=0 (6.412)
onde
∗ 1 µνρσ a
F µν a
= ε Fρσ (6.413)
2
2.13 Explique o significado geométrico da Identidades de Bianchi.
Sugestão: Veja o artigo de R.P. Feynman em Les Houches, Session XXIX, 1976,
North Holland, 1977, Pags: 135-140.
2.14 Considere a teoria de Yang-Mills (YM) sem matéria.
a) Mostre que as eqs. de YM sem matéria se podem escrever na forma

 ~ ·E
∇ ~ a = ρa


 ∇~ ·B
~ a = ∗ρa
~ ×E~ a = − ∂ B~ a + J~a (6.414)

 ∇

 ~ ∂t
∇×B ~ a = − ∂ E~ a + ∗J~a
∂t
calcule ρa , ∗ ρa , J~a e ∗ J~a .

b) Mostre que as 4-correntes jµa ≡ (ρa , J~a ) e ∗ jµa ≡ (∗ ρa , ∗ J~a ) são conservadas.
2.15 Mostre que Tr (∗Fµν F µν ) é uma 4-divergência. Comente sobre a sua inclusão na
acção.
2.16 Mostre que o seguinte Ansatze (S. Coleman, Phys. Lett70B (77), 59)
A1a = A2a = 0
A0a = −A3a = x1 f a (x0 + x3 ) + x2 ga (x0 + x3 ) (6.415)
onde f a e ga são funções arbitrárias, são soluções das equações de YM sem matéria.
Discuta esta solução.
2.17 Considere o Ansatze de Wu-Yang para soluções estáticas em SU(2) YM.

G(r) F (r)
A0a = xa Aia = εaij xj (6.416)
r2 r2
a) Deduza as equações a que F e G devem obedecer.

b) Mostre que elas são satisfeitas para F = −1/g e G = constante. Mostre que estas
soluções correspondem a ρa = ∗ ρa = 0 e J~a = ∗ J~a = 0. (ρa , ... são definidos no
problema 2.14).
c) Para as soluções da alı́nea b) descreva o potencial e os campos e calcule a energia.
2.18 Considere QED com a condição de gauge não linear

λ
F = ∂µ Aµ + Aµ Aµ . (6.417)
2
a) Escreva Lef f e mostre que sLef f = 0, onde s é o operador de Slavnov.

b) Calcule a polarização do vácuo a 1-loop. Discuta o programa de renormalização
dando especial atenção aos vértices proporcionais a λ. Pode aqui considerar a teoria
sem fermiões.
c) Demonstre a invariância da matriz S renormalizada em relação ao parâmetro λ.
d) Verifique o resultado anterior mostrando que o diagrama da figura junta, poten-
cialmente perigoso para o momento magnético anómalo do electrão, não dá con-
tribuição (seria proporcional a λ).
e e
e) Deduza as identidades de Ward desta teoria para os funcionais Z e Γ. Escreva o

funcional gerador das equações de Dyson-Schwinger para os fantasmas, isto é,
δΓ
= ··· (6.418)
δω
f) Calcule ao nı́vel árvore γ + γ → γ + γ. Compare com o resultado na gauge linear.
g) Calcule ao nı́vel árvore a amplitude T µν para e+ + e− → γ + γ. Verifique que

k1µ T µν 6= 0 e k2µ T µν 6= 0 onde k1 e k2 são os 4- momentos dos fotões. Utilize
as identidades de Ward para verificar os resultados. Há algum problema com este
resultado?
Problems 245
2.19 Considere a teoria que descreve as interacções dos quarks com os gluões (Cromodinâ-
mica Quântica) descrita pelo Lagrangeano
n
1 a µνa X α
LQCD = − Fµν F + / − mα )ij ψjα
ψ i (iD (6.419)
4
α=1
onde
a
Fµν = ∂µ Aaν − ∂ν Aaµ + gf abc Abµ Acν
a
λ
(Dµ )ij = δij ∂µ − ig Aa . (6.420)
2 ij µ
O ı́ndice α = 1, 2, . . . , n indica os diferentes sabores de quarks (up, down, · · · , top).

Para quantificar a teoria considere a condição de gauge
1
LGF = − (∂µ Aµa )2 , (6.421)
2ξ
para a qual resulta o Lagrangeano dos fantasmas
LG = ∂µ ω a ∂ µ ω a + gf abc ∂ µ ω a Abµ ω c . (6.422)
Para renormalizar a teoria necessitamos do seguinte Lagrangeano de contratermos:

1 2
∆L = − (Z3 − 1) ∂µ Aaν − ∂ν Aaµ − (Z4 − 1)gf abc ∂µ Aaν Aµb Aνc
4
1 X α
− g 2 (Z5 − 1)f abc f ade Abµ Acν Aµd Aνe + (Z2 − 1)iψ i γ µ ∂µ ψiα
4 α
X X a
α α α µ λ
− mα (Zmα − 1)ψ i ψi + (Z1 − 1)g ψi γ ψjα Aaµ
α α
2 ij
+(Z6 − 1)∂µ ω a ∂ µ ω a + (Z7 − 1)gf abc ∂ µ ω a Abµ ω c . (6.423)
a) Verifique a expressão para LG .

b) Considere a amplitude
p1 k1
µ, a
ab
iTµν ≡ ν, b (6.424)
p2 k2
ab . Verifique que k µ T ab 6= 0.
Calcule ao nı́vel árvore Tµν 1 µν
c) Verifique as contas da alı́nea anterior calculando k1µ Tµν

ab através das identidades
de Ward.
d) Supondo que os gluões possam ser estados finais, a amplitude para o processo
fı́sico q + q → g + g onde g é o gluão, é dada pela expressão
M = εµ (k1 )sa Tµν

ab ν
ε (k2 )sb , (6.425)
onde εµ (k1 ) e sa são os vectores de polarização de spin e de cor, respectivamente (o

mesmo para εν (k2 ) e sb ). Sabe-se que num processo fı́sico, M se deve anular quando
se faz a substituição εµ (k) → kµ . Como é que este resultado é compatı́vel com as
alı́neas anteriores?
e) Mostre que se devem verificar as relações
√
Z1 Z4 Z7 Z5
= = =√ (6.426)
Z2 Z3 Z6 Z3
f) Calcule Z1 , Z2 , Z3 , Z6 e Z7 , usando subtracção mı́nima (MS), (isto é, calcule só

a parte divergente dos diagramas) e verifique explicitamente que Z1 Z6 = Z2 Z7 .
g) Calcule a contribuição dos fermiões para Z4 e Z5 e verifique que também obedecem
às relações da alı́nea a).
h) Calcule as funções do Grupo de Renormalização β, γA e γF .
Chapter 7
Renormalization Group
7.1 Callan -Symanzik equation

7.1.1 Renormalization scheme with momentum subtraction
Em teoria quântica dos campos um esquema de renormalização tem duas componentes.
Primeiro há um processo de regularização que isola os infinitos que aparecem nos diagramas
de Feynman. A regularização é arbitrária desde que mantenha as simetrias da teoria. Para
teorias sem campos de gauge há muitos processos alternativos. Para teorias de gauge o
melhor processo parece ser a regularização dimensional.
Depois de regularizada a teoria teremos que especificar um método sistemático para
remover as divergências e definir os parâmetros renormalizados de teoria. A este processo
chamamos esquema de renormalização. Há uma grande arbitrariedade na escolha do
processo de subtração. A fı́sica contudo não pode depender desta escolha. Este é o
conteúdo do grupo de renormalização: O conteúdo fı́ sico de teoria deve ser invariante para
transformações que apenas mudem as condições de normalização. Esta afirmação trivial
põe no entanto, como veremos, constrangimentos altamente não triviais no comportamento
assimptótico da teoria.
Vamos começar por estudar os chamados esquemas com subtração de momento. Con-
forme o ponto no espaço dos momentos externos que serve de definição às funções de Green
irredutı́veis, podemos ter várias formas deste esquema. Vamos exemplificar com a teoria
λφ4 .
On–shell renormalization
Isto corresponde a uma série de Taylor para os momentos exteriores on - shell. Para a
self-energy isto dá
e 2)
Σ(p2 ) = Σ(m2 ) + (p2 − m2 )Σ′ (m2 ) + Σ(p (7.1)
com as condições

 e 2) = 0
 Σ(m
e 2)
∂ Σ(p (7.2)

 ∂p2 2 2 = 0
p =m
247
248 CHAPTER 7. RENORMALIZATION GROUP
(2)
Em termos de ΓR (p2 ) dado por
e 2)
Γ2R (p) = p2 − m2 − Σ(p (7.3)
temos
 (2)

 Γ (m2 ) = 0
 R
(2) (7.4)
 ∂Γ R
 ∂p2
 =1
2
p =m 2
(4)
Para ΓR uma escolha conveniente é
 2 2
 pi = m
(4)
ΓR (p1 , p2 , p3 ) = −λ para (7.5)
 2
s = t = u = 4m
3
Neste caso os parâmetros m2 e λ são a massa fı́sica e, a menos de factores cinemáticos, a
secção eficaz para s = t = u = 34 m2 respectivamente.
Intermediate renormalization
Este esquema corresponde a uma expansão de Taylor em torno de momentos nulos.
e 2)
Σ(p2 ) = Σ(0) + Σ′ (0)p2 + Σ(p (7.6)
e 2 ) obdece às condições
A parte finita Σ(p

 e
 Σ(0) = 0


e (7.7)

 ∂ Σ
 ∂p2 2 = 0
p =0
(2)
que traduzidas em termos de ΓR se escrevem
 (2)
 2
 ΓR (0) = m
(2) (7.8)
 ∂ΓR2 = 1

∂p
(4)
Para ΓR a condição é
(4)
ΓR (p1 , p2 , p3 ) = −λ para p1 = p2 = p3 = 0 (7.9)
Neste esquema m2 não é a massa fı́sica e λ não é nenhuma quantidade mensurável pois
os pontos pi = 0 não pertencem à região fı́sica. Podemos no entanto exprimir todas as
quantidades mensuráveis em termos destes dois parâmetros, como veremos na secç ão 3.3.
7.1. CALLAN -SYMANZIK EQUATION 249
General case
Os dois exemplos anteriores são casos particulares do esquema geral onde as condições de
normalização podem ser funções de vários momentos de referência ξ1 , ξ2 ... tais que

 (2)

 ΓR (ξ12 ) = m2



 ∂Γ(2)
 R =1 (7.10)
 ∂p2 2 2

 p =ξ2




 Γ(4) (ξ , ξ , ξ ) = −λ
R 3 4 5
7.1.2 Renormalization group

Consideremos dois esquemas de renormalização R e R′ . Como ambos partem do mesmo
Lagrangeano não renormalizado
L = LR + ∆LR = LR′ + ∆LR′ (7.11)

devemos ter
−1/2 −1/2
φR = Zφ (R)φ0 ; φ′R = Zφ (R′ )φ0 . (7.12)
Logo
−1/2
φ′R = Zφ (R′ , R)φR (7.13)
onde
Zφ (R′ )
Zφ (R′ , R) = (7.14)
Zφ (R)
Estas relações indicam que os campos renormalizados em diferentes esquemas estão
relacionados por uma constante multiplicativa. Esta constante é finita pois tanto φR′
como φR são finitos. De modo semelhante
λR′ = Zλ−1 (R′ , R)Zφ2 (R′ , R)λR
m2R′ = m2R + δm2 (R′ , R) (7.15)

onde
Zλ (R′ )
Zλ (R′ , R) =
Zλ (R)
δm2 (R′ , R) = δm2 (R′ ) − δm2 (R) (7.16)

são quantidades finitas. A operação que leva as quantidades num esquema de renormal-
ização R para outro esquema R′ pode ser vista como uma transformação de R em R′ . O
conjunto de todas estas transformaçoes forma o Grupo de Renormalização.
7.1.3 Callan - Symanzik equation

Vamos agora ver como dar uma expressão analı́tica à invariância para transformações
do grupo de renormalização. A forma da equação do grupo de renormalização depende
do esquema de renormalização utilizado. Vamos aqui obter as equações do GR para o
esquema com substração de momento, a chamada equação de Callan - Symanzik.
Notemos primeiro que

∂ i i i
= (−i) 2 (7.17)
∂m20 p − m20 + iε
2 p2 2
− m0 + iε p − m20 + iε
isto é, a derivação duma função de Green não renormalizada em relação à massa despida
é equivalente à inserção dum operador composto 12 φ2 levando momento zero, isto é
∂Γ(n) (pi ) (n)

2 = −iΓφ2 (0, pi ) (7.18)
∂m0
As funções Green irredutı́veis renormalizadas são dadas por

(n) (n/2) (n)

 ΓR (pi ; λ; m) = Zφ Γ (pi ; λ0 ; m0 )
(7.19)

 Γ(n) (p; p ; λ; m) = Z −1 Z n/2 Γ(n) (p; p ; λ ; m )
φ2 R i φ2 φ φ2 i 0 0
Então a equação anterior escreve-se
∂ h −n/2 (n) i
(n)
Z Γ R (p i , λ, m) = −iZφ2 Z −n/2 Γφ2 R (0, pi , λ, m) (7.20)
∂m20 φ
e portanto
n ∂Zφ −n/2 (n) −n/2 ∂ (n) −n/2 (n)

− Zφ−1 2 Zφ ΓR + Zφ 2 ΓR = −iZφ2 Zφ Γφ2 R (0, pi , λ, m) (7.21)
2 ∂m0 ∂m0
ou seja

∂ n ∂ ln Zφ (n) (n)
− ΓR = −i Zφ2 Γφ2 R
∂m20 2 ∂m20

∂m2 ∂m ∂ ∂λ ∂ n ∂ ln Zφ (n) (n)
2 2
+ 2 − 2 ΓR = −iZφ2 Γφ2 R (7.22)
∂m0 ∂m ∂m ∂m0 ∂λ 2 ∂m0
ou ainda

∂ ∂ (n) (n)
m +β − nγ ΓR = −im2 αΓφ2 R (7.23)
∂m ∂λ
que é a equação de Callan - Symanzik para a teoria φ4 , onde α, β e γ são funções sem
dimensões
∂λ
2
∂m
β = 2m2 0
(7.24)
∂m2
∂m20
∂ ln Zφ
∂m20
γ = m2 (7.25)
∂m2
∂m20
Zφ 2
α = 2 (7.26)
∂m
∂m20
A função α está relacionada com γ. De facto se escolhermos as condições de normal-

ização a pi = 0 
 (2)
 ΓR (0, λ, m) = −m2
(7.27)

 Γ(2) (0, 0, λ, m) = i
φ2 R
obtemos
α = 2(γ − 1) (7.28)
(n) (n)
Como as quantidades ΓR e Γφ2 R não dependem do cut - off, esperamos também que
α, β e γ sejam independentes do cut - off. Para vermos isso pomos n = 2 e diferenciamos
em ordem a p2

∂ ∂ ∂ (2) ∂ (2)
m +β − 2γ 2
ΓR (p, λ, m) = −im2 α 2 Γφ2 R (0, p, λ, m) (7.29)
∂m ∂λ ∂p ∂p
Pondo p2 = 0 e usando

(2)
∂ΓR
=1 (7.30)
∂p2
p2 =0
obtemos

2 ∂ (2)
γ = im (γ − 1) Γ 2 (0, p, λ, m) (7.31)
∂p2 φ R p2 =0
o que demonstra que γ é independente do cut - off. Então α = 2(γ − 1) também o é e

todas as funções excepto β são agora independentes do cut - off. Portanto β também o é.
Como α, β e γ são sem dimensões e não dependem do cut - off, então são somente funções
da constante de acoplamento que também não tem dimensões, isto é
α = α(λ)
β = β(λ)
γ = γ(λ) (7.32)
Nós vamos sobretudo estar interessados no esquema de subtracção mı́ nima, por isso
não vamos agora calcular as funções α, β e γ para todas as teorias, faremos isso na secção
3.3. Indicaremos no entanto um método expedito para o seu cálculo. Seja por exemplo a
função β(λ). Notando que
∂λ ∂m2 ∂
(λ0 , Λ/m) = λ(λ0 , Λ/m)
∂m20 ∂m20 ∂m2
∂m2 1 ∂
= λ(λ0 , Λ/m) (7.33)
∂m20 2m ∂m
obtemos da definição 7.24
∂ ∂ ∂
β=m λ(λ0 , Λ/m) = m [Z(λ0 , Λ/m)λ0 ] = −λ0 Λ [Z(λ0 , Λ/m)] (7.34)
∂m ∂m ∂Λ
ou
∂
β = −λ [ln Z(λ0 , Λ/m)] (7.35)
∂ ln Λ
onde1 Z = Zλ−1 Zφ2 . O resultado de 1 - loop dá
3λ0 Λ2
Zλ = 1 + ln + O(λ20 )
32π 2 m2
Zφ = 1 + O(λ20 ) (7.36)
logo
3λ0 Λ2
Z =1− ln + ... (7.37)
32π 2 m2
e
3λ0 Λ
ln Z = 2
ln + · · · (7.38)
16π m
Portanto para φ4
3λ2
β(λ) = + O(λ3 ) . (7.39)
16π 2
1
Por definição λ=Zλ0 .
7.1.4 Weinberg’s theorem and the solution of the RG equations

O teorema de Weinberg diz respeito ao comportamento assimptótico das funções de Green
1P I na região Euclediana (p2i < 0) e para valores não excepcionais dos momentos (nen-
huma soma parcial é nula).
Teorema 3.1
Se os momentos não forem excepcionais e se os parametrizarmos com pi = σki as
(n)
funções de Green irredutı́ veis de e partı́cula ΓR comportam-se na região euclediana
profunda (σ → ∞ e ki fixos, p2i < 0) do modo seguinte
lim Γ(n) (σki , λ, m)σ 4−n [a0 (ln σ)b0 + a1 (ln σ)b1 + · · · ] (7.40)
σ→∞
(n) ′ ′
lim Γφ2 (σki , λ, m)σ 2−n [a′0 (ln σ)b0 + a′1 (ln σ)b1 + · · · ] . (7.41)
σ→∞
Não faremos a demonstração (ver Bjorken and Drell) mas notemos que as potências de
σ são as dimensões canónicas das funções de Green (em termos da massa). Se este com-
portamento é o verificado assimptoticamente depende da soma da série dos logaritmos. Se
esta somar para uma potência de σ, por exemplo σ −γ , então assimptóticamente o com-
portamento canónico σ 4−n é modificado para σ 4−n−γ .γ é chamada a dimensão anómala.
Como vamos ver o GR vai efectuar esta soma de logaritmos e dar-nos qual a dimensão
anómala.
7.1.5 Asymptotic solution of the RG equations

(n) (n)
Do teorema de Weinberg temos que ΓR ≫ Γφ2 R para qualquer ordem (finita) em λ na
região euclediana profunda (σ → ∞). Se admitirmos que isto continua verdade mesmo
depois de somar todas as ordens de teoria de perturbações, então podemos desprezar
o segundo membro da equação de Callan-Symanzik e obtemos uma equação diferencial
homogénea

∂ ∂
m + β(λ) − nγ(λ) Γ(n)
as (pi , λ, m) = 0 (7.42)
∂m ∂λ
(n) (n)
onde Γas é a forma assimptótica de ΓR . O significado desta equação é que nesta região
assimptótica, uma mudança no parâmetro de massa pode ser sempre compensada por
mudanças apropriadas do acoplamento e da escala dos campos.
(n)
Para resolver esta equação começamos por definir uma quantidade ΓR sem dimensões,
usando análise dimensional
(n) (n)
Γas (pi , λ, m) = m4−n ΓR (pi /m, λ) . (7.43)
(n)
ΓR satisfaz
p
∂ ∂ (n) i
m +σ ΓR σ ,λ = 0 . (7.44)
∂m ∂σ m
Então

∂ ∂
m +σ mn−4 Γ(n)
as (σpi , λ, m) = 0 (7.45)
∂m ∂σ
ou seja

∂ ∂
m +σ + (n − 4) Γ(n)
as (σpi , λ, m) = 0 (7.46)
∂m ∂σ
Usando esta equação podemos trocar a derivação em ordem à massa pela derivação
em ordem à escala na equação de Callan-Symanzik para obter

∂ ∂
σ − β(λ) + nγ(λ) + (n − 4) Γ(n)
as (σpi , λ, m) = 0 (7.47)
∂σ ∂λ
Para resolver esta equação removemos os termos sem derivadas com a transformação
Rλγ(x)
n 0 β(x) dx
Γ(n)
as (σpi , λ, m) = σ
4−n
e F (n) (σpi , λ, m) . (7.48)
Substituindo na equação diferencial vemos que os termos sem derivadas desaparecem e
obtemos uma equação diferencial para F (n)

∂ ∂
σ − β(λ) F (n) (σp, λ, m) = 0 (7.49)
∂σ ∂λ
Introduzindo t = ln σ podemos escrever

∂ ∂
− β(λ) F (n) (et p, λ, m) = 0 (7.50)
∂t ∂λ
Para resolver esta equação introduzimos a constante de acoplamento efectiva λ(t, λ)
como solução da equação
∂λ(t, λ)
= β(λ) (7.51)
∂t
com a condição fronteira λ(0, λ) = λ. Para vermos que esta definição nos vai dar a solução,
escrevemos
Z λ(t,λ)
dx
t= (7.52)
λ β(x)
e diferenciamos em ordem a λ
1 ∂λ 1
0= − (7.53)
β(λ) ∂λ β(λ)
ou ainda
7.2. MINIMAL SUBTRACTION (MS) SCHEME 255
∂λ
β(λ) − β(λ) =0 (7.54)
∂λ
Usando agora a definição de λ obtemos

∂ ∂
− β(λ) λ(t, λ) = 0 (7.55)
∂t ∂λ
O operador diferencial é exactamente o mesmo da equação para F (n) (et p, λ, m). Por-
tanto F (n) satisfaz aquela equação se depender da t e λ através de λ(t, λ). Portanto a
(n)
solução geral de Γas é
Rλ γ(x)
(n) n dx
Γas (σpi , λ, m) = σ 4−n e 0 β(x) F (n) (pi , λ(t, λ), m) (7.56)
Para se obter uma interpretação fı́sica desta solução notemos que
Rλγ(x) Rλγ(x) Rλ λ(x)

n 0 β(x) dx n 0 β(x) dx n dx
e = e e λ β(x)
Rλγ(x) Rλ γ(x)
n 0 β(x) dx −n λ β(x) dx
= e e
Rλγ(x) Rt
n dx γ(λ(t′ ,λ))dt′
= e 0 β(x) e−n 0 (7.57)
Portanto
Rt Rλ γ(x)
(n) γ(λ(t′ ,λ))dt′ −n dx
Γas (σpi , λ, m) = σ 4−n e−n 0 e β(x) F (n) (pi , λ(t, λ), m) (7.58)
Rλ γ
(n)
Se pusermos σ = 1(t = 0), vemos que en 0 β dx F (n) é Γas . Então obtemos finalmente a
solução da equação do GR.
Rt
γ(λ(t′ ,λ))dt′
Γ(n)
as (σpi , λ, m) = σ
4−n −n
e 0 Γ(n)
as (pi , λ(t, λ), m) (7.59)
Nesta forma a solução tem uma interpretação simples. O efeito de efectuar uma mu-
(n)
dança de escala nos momentos pi nas funções de Green ΓR é equivalente a substituir a
constante de acoplamento λ, pela constante de acoplamento efectiva λ à parte factores mul-
(n)
tiplicativos. O primeiro é simplesmente resultante do facto de ΓR ter dimensão canónica
4 − n em unidades de massa. O factor exponencial é o termo da dimensão anómala que
resultou de somar todos os logaritmos em teoria de perturbações. Este factor é controlado
por γ, a dimensão anómala. Veremos à frente como calcular a dimensão anómala, numa
teoria qualquer.
7.2 Minimal subtraction (MS) scheme

7.2.1 Renormalization group equations for MS
Vamos agora ver outras formas que pode tomar a equação do grupo de renormalização. A
afirmação que a renormalização é multiplicativa pode ser escrita na forma
−n/2 (n)
Γ(n) (pi , λ0 , m0 ) = Zφ ΓR (pi , λ, m, µ) (7.60)
onde µ é a escala usada para definir a normalização das funções de Green. O lado esquerdo
da equação não depende de µ, mas o lado direito depende explicitamente e implicitamente
através de λ e m. Então temos
∂ h −n/2 (n) i
µ Zφ ΓR (pi , λ, m, µ) = 0 (7.61)
∂µ
ou seja

∂ ∂ ∂ (n)
µ +β + γm m − nγ ΓR = 0 (7.62)
∂µ ∂λ ∂m
com

m ∂λ
β λ, = µ
µ ∂µ

m ∂ ln m
γm λ, = µ
µ ∂µ

m 1 ∂ ln Zφ
γ λ, = µ (7.63)
µ 2 ∂µ
Esta equação tem a vantagem sobre a equação de Callan - Symanzik de ser homogénea.
A dificuldade reside nas funções β e γ dependerem de duas variáveis λ e m µ e portanto a
equação ser de difı́cil resolução. Existe contudo um esquema de renormalização em que a
dependência em m µ desaparece e portanto a equação é simples de resolver. É o chamado
esquema de subtracção mı́nima que passamos a expôr.
7.2.2 Minimal subtraction scheme

O esquema de subtracção mı́nima (MS) está relacionado com o método de regularização
dimensional. As divergências dos integrais aparecem neste método como pólos em 1ε onde
ε = 4 − d. O esquema de subtracção mı́nima consiste em escolher os contratermos para
cancelar somente os pólos.
Vamos exemplificar com a self-energy em λφ4 , a que corresponde o diagrama da Fig. 7.1
p p
Figure 7.1: Diagram for self-energy in φ4 .
Temos
Z
1 dd k i
−iΣ(p) = (−iλ)µε
2 (2π)d p2 − m2 + iε
1 ε Γ(1 − d/2) ε ε/2
= −iλ µ 2 π (7.64)
32π 2 m2−d
onde ε = 4 − d. Então
1 ε Γ(−1 + ε/2) √
Σ(p2 ) = λ 2
µ −2+ε
· (2 π)ε
32π m
m2
µ ε √ ε
= λ Γ(−1 + ε/2) · (2 π) (7.65)
32π 2 m
Usando2
 
ψ(2)
ε z }| {
2 
Γ −1 + = −  + 1 − γ +O(ε) (7.66)
2 ε
e
µ ε µ
= 1 + ε ln (7.67)
m m
obtemos

λm2 2 √
Σ(p2 ) = − + ψ(2) + 2 ln(µ/m) + 2 ln 2 π + O(ε) (7.68)
32π 2 ε
Portanto no esquema de subtracção mı́nima devemos adicionar um contratermo
λm2 1 2
∆LM S
φ2 = − φ (7.69)
32π 2 ε
Se tivéssemos feito subtracção de momento à escala µ, isto é ΣR (p2 = µ2 ) = 0 terı́amos
o contratermo

M OM λm2 1 1 √
∆Lφ2 =− + ψ(2) + ln(µ/m) + ln 2 π φ2 (7.70)
32π 2 ε 2
Vemos assim que o Lagrangeano de contratermos no esquema de subtracção mı́nima
quando expandido em série de Laurent em ε contém só termos divergentes. Portanto
p
φ0 = Zφ φ
m0 = Zm m
2
γ é a constante de Euler e ψ(x) a derivada logaritmica da função Γ. Ver o Apêndice da Mecânica
Quântica Relativista.
λ0 = µ ε Z λ λ (7.71)
tendo as constantes de renormalização Zφ , Zm e Zλ a forma
∞
X
Zλ = 1 + ar (λ)/εr
r=1
∞
X
Zm = 1 + br (λ)/εr
r=1
∞
X
Zφ = 1 + cr (λ)/εr (7.72)
r=1
Assim os coeficientes da equação do grupo de renormalização são independentes de µ, e

como são adimensionais, também são independentes de m dependendo somente da cons-
tante de acoplamento. Isto simplifica a solução da equação do grupo de renormalização

∂ ∂ ∂ (n)
µ +β + γm m − nγ ΓR = 0 (7.73)
∂µ ∂λ ∂m
Usando análise dimensional

∂ ∂ ∂
m + (n − 4) + µ +σ ΓR (σp, m, λ, µ) = 0 (7.74)
∂m ∂µ ∂σ
e podemos escrever

∂ ∂ ∂
σ −β − (γm − 1) m + nγ + (n − 4) ΓR (σp, m, λ, µ) = 0 (7.75)
∂σ ∂λ ∂m
que tem a solução
Rt
γ(λ(t′ ))dt′ (n)
ΓR (σpi , m, λ, µ) = σ 4−n e−n 0 ΓR (pi , m(t), λ(t), µ) (7.76)
onde se introduziram a massa efectiva m(t) e a constante de acoplamento efectiva λ(t)
definidas por

 dλ
 dt = β(λ) ; λ(t = 0) = λ
h i (7.77)

 dm(t) = γ (λ) − 1 m(t)
m ; m(t = 0) = m
dt
A solução desta equação é
Rt ′ ′
m(t) = m e 0 [γm (λ(t ))−1]dt
Rt
γm (λ(t′ ))dt′
= m e−t e 0
R λ(t) γm (x)
dx
= m e−t e λ β(x) (7.78)
7.2.3 Physical parameters

Os parâmetros definidos pelo esquema de subtracção mı́nima não são parâmetros fı́sicos.
Os parâmetros fı́sicos podem no entanto ser calculados em função deles. Por parâmetro
fı́sico entendemos um elemento de matriz S ou a posição do pólo no propagador. Para eles
é válido o teorema seguinte,
Teorema 3.2:
Qualquer parâmetro fı́sico P (λ, m, µ) satisfaz a seguinte equação do grupo de renor-
malização

∂ ∂ ∂
DP (λ, m, µ) ≡ µ + β(λ) + γm m P (λ, m, µ) = 0 (7.79)
∂µ ∂λ ∂m
Dem: Consideremos primeiro o propagador escalar ∆(p2 ) que satisfaz a equação do

grupo de renormalização
[D + 2γ]∆(p2 , λ, m, µ) = 0 (7.80)
Podemos escrever uma série de Laurent em voltado pólo em p2 = m2p
R2 e
∆(p2 , λ, m, µ) = +∆ (7.81)
p2 − m2p
A posição do pólo mp (λ, m, µ) e o resı́duo R2 (λ, m, µ) satisfazem as equações do

grupo de renormalização que podem ser obtidas por aplicação do operador (D + 2γ)
à equação anterior. Igualando os resı́duos dos pólos obtemos
Dmp (λ, m, µ) = 0 (7.82)
[D + γ(λ)]R(λ, m, µ) = 0 (7.83)
Demonstrámos portanto o teorema para a massa fı́sica. Para um elemento da matriz

S temos (SR = Rn Γ(n) )
D lim Rn Γ(n) = lim D(Rn Γn )

p2i →m2p p2i →m2p
= limp2 →m2p [nDRRn−1 Γn + Rn DΓn ]
i
= limp2 →m2p [−nγ + nγ]Rn Γn = 0 (7.84)
i
o que completa a demonstração.
Veremos à frente como estes resultados podem ser usados para relacionar os parâmetros
fı́sicos com os parâmetros da teoria.
7.2.4 Renomalization group functions in minimal subtraction

Vimos anteriormente que
 p

 φ0 = Zφ φ
 m =Z m
0 m
(7.85)



λ0 = µε Zλ λ
e que as constantes de renormalização têm a forma.
 P
 Zλ = 1 + ∞ r=1 ar (λ)/ε
r



 P
Zm = 1 + ∞ r=1 br (λ)/ε
r (7.86)




 P
Zφ = 1 + ∞ r
r=1 cr (λ)/ε .
Vejamos como se calculam β, γm e γ.
i) Cálculo de β(λ)
Por definição
∂λ
β(λ) = µ (7.87)
∂µ
Esta quantidade é finita no limite ε → 0. Isto quer dizer que antes de fazermos ε → 0
deve ser uma função analı́tica em ε. É então conveniente definir
β(λ) = β̂(λ, ε = 0) = d0 (7.88)

onde
β̂(λ, ε) = d0 + d1 ε + d2 ε2 + · · · (7.89)
com coeficientes dr a determinar. Posto isto, usamos o facto de λ0 não depender da escala
µ. Então
∂ ε
0 = µ (µ Zλ λ)
∂µ
∂Zλ
= εµε Zλ λ + µε β̂(λ, ε)λ + µε Zλ β̂(λ, ε) (7.90)
∂λ
Então

∂Zλ
ελZλ + β̂(λ, ε) Zλ + λ =0 (7.91)
∂λ
Usando as expressões de Zλ e β̂ obtemos
∞
" ∞ #
X ar+1 X 1 dar
ελ + a1 λ + λ + (d0 + d1 ε + d2 ε2 + · · · ) 1 + ar + λ = 0 (7.92)
r=1
εr r=1
εr dλ
Então dr = 0 para r > 1 e
X 1
da1 dar
ε(λ + d1 ) + a1 λ + d0 + d1 a1 + λ + ar+1 λ + d0 ar + λ
dλ r
εr dλ

dar+1
+ d1 ar+1 + λ =0 (7.93)
dλ
logo
λ + d1 = 0

da1
a1 λ + d0 + d1 a1 + λ =0
dλ

dar dar+1
ar+1 λ + d0 ar + λ + d1 ar+1 + λ =0 (7.94)
dλ dλ
Estes cálculos dão
d1 = −λ (7.95)
da1
β(λ) = d0 = λ2
dλ
d d
λ2 (ar+1 ) = β(λ) (λar ) (7.96)
dλ dλ
Portanto a função β(λ) depende somente do coeficiente em 1ε de Zλ que se calcula
fácilmente em teoria de perturbações. Além disso vemos que os resı́duos dos pólos de
ordem superior se podem calcular em termos do resı́duo do pólo simples. Para λφ4 é fácil
de ver que
3λ 1
Zλ = 1 + + ··· (7.97)
16π 2 ε
e portanto

da1 d 3λ 3λ2
β(λ) = λ2 = λ2 = (7.98)
dλ dλ 16π 2 16π 2
como tı́nhamos obtido anteriormente. Para teorias de gauge há uma pequena modificação
pois g0 = µε/2 Zg g. Um cálculo trivial dá neste caso
d1 = −g/2 (7.99)
e
1 2 da1
β(g) = g
2 dg
1 2 dar+1 d
g = β(g) (gar ) (7.100)
2 dg dg
onde, como anteriormente
∞
X
Zg = 1 + ar (g)/εr . (7.101)
r=1
ii) Cálculo de γm (λ)

∂
Partimos de m0 = Zm m. Aplicando µ ∂µ obtemos
∂Zm ∂m
0 = µ m + Zm µ
∂µ ∂µ
∂Zm ∂ ln m
= β̂(λ, ε) m + mZm µ (7.102)
∂λ ∂µ
Como µ ∂ ∂µ
ln m
= γm , obtemos a equação

∂
β̂(λ, ε) + γm Zm = 0 (7.103)
∂λ
ou seja
X∞
db1 1 dbr dbr+1
γm + d1 + d0 + γm br + d1 =0 (7.104)
dλ ε
r=1 r
dλ dλ
Portanto
db1
γm = −d1 (7.105)
dλ
dbr+1 dbr
−d1 = β(λ) + γ m br (7.106)
dλ dλ
onde

 −λ teoria λφ4
d1 = (7.107)

− g/2 teorias de gauge
Mais uma vez γm depende somente do resı́duo do pólo simples.
iii) Cálculo de γ(λ)
Aqui é mais fácil partir da definição de γ(λ)
1 ∂ 1 ∂ 1
γ(λ) = µ ln Zφ = µ Zφ (7.108)
2 ∂µ 2 ∂µ Zφ
logo

∂
β̂(λ, ε) − 2γ(λ) Zφ = 0 (7.109)
∂λ
o que dá
∞
dc1 X 1 dcr dcr+1
−2γ(λ) + d1 + d0 − 2γcr + d1 =0 (7.110)
dλ εr dλ dλ
r=1
Então
1 dc1
γ(λ) = d1 (7.111)
2 dλ
dcr+1 dcr
−d1= β(λ) − 2γcr (7.112)
dλ dλ
sendo o coeficiente d1 dado anteriormente, Eq. (7.107).
Podemos concluir dizendo que o coeficiente do pólo simples nas constantes de renor-
malização, determina univocamente as funções β, γm e γ e também os valores dos pólos
de ordem superior.
7.2.5 β and γ properties

Nós adoptamos um esquema particular de renormalização. Se tivéssemos adoptado outro
esquema terı́amos outra definição dos parâmetros da teoria e funções β, γm e γ diferentes.
Vamos aqui discutir os aspectos do grupo de renormalização que são independentes do
esquema usado.
Consideremos então dois esquemas (ambos independentes da massa). Então
g′ = gFg (g) Fg (g) = 1 + O(g 2 )
′ (g ′ ) = Z (g)F (g)
Zm Fm (g) = 1 + O(g 2 )
m m (7.113)
Zφ′ (g′ ) = Zφ (g)Fφ (g) Fφ (g) = 1 + O(g ′ )
O 1 nas funções F expressa o facto que ao nı́vel árvore não há ambiguidades. Usando
as relações acima podemos ver como estão relacionadas as funções β, γm e γ em dois
esquemas. Obtemos (estamos a considerar o caso duma teoria de gauge)

∂ ′ ∂ ∂Fg
β ′ (g′ ) = µ g = µ (gFg (g)) = β(g) Fg + g
∂µ ∂µ ∂g
′ ∂ ∂ ln −1 ∂
γm (g′ ) = µ ln m′ = µ (Fm (g)m) = γm (g) − β(g) ln Fm
∂µ ∂µ ∂g
1 ∂ 1 ∂
γ ′ (g′ ) = µ ln Zφ′ (g′ ) = γ(g) + β(g) ln Fφ (7.114)
2 ∂µ 2 ∂g
As funções β, γm e γ só coincidirão se os esquemas forem o mesmo, isto é Fg = Fm =
Fφ = 1. Contudo as propriedades seguintes são ainda independentes do esquema.
i) A existência de um zero de β(g).
Se β(g0 ) = 0 então β ′ (g0′ ) = 0 para g0′ = g0 Fg (g0 ). Notar que em geral g0 depende do
esquema, isto é g0 6= g0′ .
ii) A primeira derivada de β(g) no zero.

Seja β(g0 ) = 0. Então

∂β ′ (g0′ ) ∂g ∂ ∂Fg
= β(g) Fg + g
∂g′ ∂g′ ∂g ∂g g0
 
∂F
∂Fg ∂β 1 ∂ Fg + g ∂gg
= Fg + g +g + β(g) ∂F

∂g ∂g Fg + g ∂gg ∂g
g0
∂β
= (g0 ) · (7.115)
∂g
iii) Os primeiros dois termos em β(g).
Seja β(g) = b0 g3 + b1 g5 + O(g 7 ), e
Fg (g) = 1 + ag 2 + O(g 4 ) · (7.116)

Então
g ′ = g + ag 3 + O(g 5 ) (7.117)
e
g = g ′ − ag ′3 + O(g 5 ) (7.118)
Portanto
∂
β ′ (g′ ) = β(g) (gFg ) = (b0 g 3 + b1 g 5 + O(g 7 ))(1 + 3ag 2 + O(g 4 ))
∂g
= b0 g3 + (3ab0 + b1 )g5 + O(g 7 )
= b0 (g′3 − 3ag ′5 + O(g ′7 ) + (3ab0 + b1 )(g′5 + O(g ′7 ))
= b0 g′3 + b1 g′5 + O(g ′7 ) (7.119)
iv) O primeiro termo em γ(g) e γm (g).

Seja
γ(g) = cg2 + O(g 4 )
γm (g) = dg 2 + O(g 4 ) (7.120)
Então como β(g) = O(g 3 ) é evidente que
γ ′ (g′ ) = cg′2 + O(g ′4 )

′
γm (g′ ) = dg′2 + O(g ′4 ) . (7.121)
v) O valor de γ(g0 ) e γm (g0 ) se β(g0 ) = 0.

Este resultado é imediato. Como veremos na secção seguinte todos estes resultados são
necessários pois eles controlam resultados fı́sicos e estes não podem depender do esquema
de renormalização.
7.2.6 Gauge independence of β and γm in MS

A equação do grupo de renormalização em MS foi escrita para a teoria λφ4 . Vamos agora
considerar teorias da gauge (abelianas ou não abelianas). Para a quantificação destas
teorias é necessário introduzir um termo que fixe a gauge
1
LGF = − (∂ · A)2 (7.122)
2ξ
onde escolhemos as gauges do tipo de Lorentz. Como não há correcções radiativas para
a parte longitudinal do propagador, não é necessário nenhum contratermo para o termo
que fixa a gauge. Portanto se pusermos, como habitualmente,
−1/2
Aµ = ZA Aµ0 (7.123)
obtemos
1 1 1
(∂ · A)2 = (∂ · A0 )2 = (∂ · A0 )2 (7.124)
2ξ 2ξZA 2ξ0
o que quer dizer que o parâmetro de gauge é renormalização de acordo com
ξ 0 = ZA ξ . (7.125)
As funções de Green irredutı́veis renormalizadas, dependem em geral de ξ, isto é
(n) n/2 (n)
ΓR (g, m, ξ, µ) = ZA Γ0 (g0 , m0 , ξ0 , ε) (7.126)
A equação do grupo de renormalização é então

∂ ∂ ∂ ∂ (n)
µ + β(g, ξ) + γm (g, ξ)m + δ(g, ξ) − γA (g, ξ) ΓR (g, m, ξ, µ) = 0 (7.127)
∂µ ∂g ∂m ∂ξ
onde
∂ ∂ −1
δ(g, ξ) = µ ξ = µ (ZA ξ0 ) =
∂µ ∂µ
1 ∂
= −ξ0 2 ∂µ ZA
ZA
= −2ξγA (g, ξ) (7.128)

e se admitiu a possibilidade de β, γm e γA dependerem do parâmetro ξ. Contudo a de-

pendência em ξ não é arbitrária, obedece a certos constrangimentos. Para vermos isso
consideremos uma função de Green sem dimensões e correspondendo a operadores invari-
antes de gauge. Então
∂
G0 (g0 , m0 , ξ0 , ε) = 0 (independente de gauge) (7.129)
∂ξ0
e
G0 (g0 , m0 , ξ0 , ε) = G(g, m, ξ, µ) (sem dimensões (7.130)

e portanto
∂
G=0 (7.131)
∂ξ
ou seja

∂ ∂ ∂
DG G ≡ + ρ(g, ξ) + σ(g, ξ)m G(g, m, ξ, µ) = 0 (7.132)
∂ξ ∂g ∂m
onde
∂g ∂
ρ(g, ξ) = ; σ(g, ξ) = ln m (7.133)
∂ξ ∂ξ
Mas G obedece à equação do grupo de renormalização

∂ ∂ ∂ ∂
DG ≡ µ +β + γm m +δ G=0 (7.134)
∂µ ∂g ∂m ∂ξ
Usando a equação para DG G = 0 podemos substituir a derivada em ordem a ξ por
derivadas em ordem aos outros parâmetros, obtendo uma equação do grupo de renor-
malização semelhante à das teorias que não têm campos de gauge, isto é

∂ ∂ ∂
µ +β + γmm G=0 (7.135)
∂µ ∂g ∂m
onde
β ≡ β − ρδ γ m = γm − σδ (7.136)
Calculemos agora o comutador [DG , D]G = 0. Obtemos

∂β ∂β ∂ρ ∂ρ ∂ ∂δ ∂δ ∂
+β −β −δ + +ρ
∂ξ ∂g ∂g ∂ξ ∂g ∂ξ ∂g ∂ξ

∂γm ∂γm ∂σ ∂σ ∂
+ +ρ −β −δ m G=0 (7.137)
∂ξ ∂g ∂g ∂ξ ∂m
Introduzindo as funções β e γ m e o operador

∂ ∂
D≡ +ρ (7.138)
∂ξ ∂g
a equação anterior escreve-se

∂ ∂ρ ∂
(Dδ) + D β + D(ρδ) − β − δDρ
∂ξ ∂g ∂g

∂σ ∂
+ Dγ m + D(σδ) − β − − δDσ m G=0 (7.139)
∂g ∂m
Multiplicando a equação DG G = 0 por (Dδ) obtemos

∂ ∂ ∂
(Dδ) + ρ(Dδ) + σ(Dδ)m G=0 (7.140)
∂ξ ∂g ∂m
Comparando as duas equações vemos que
∂ρ ∂σ
Dβ=β e D γm = β (7.141)
∂g ∂g
Estas equações asseguram que resultados fı́sicos sejam independentes de gauge. Assim
β = 0 tem consequências fı́sicas. Então D β = 0 e D γ m = 0 dizendo que a existência dos
zeros de β e a dimensão anómala da massa γ m são independentes de gauge. Também se
β = 0 obtemos

∂β ∂ ∂
D = D β + D, β
∂g ∂g ∂g
∂ ∂ρ ∂β
= Dβ− =0 (7.142)
∂g ∂g ∂g
e portanto a primeira derivada de β no zero é independente da gauge. Finalmente como

ρ = O(g 3 ) e δ = O(g 2 ) temos então
β = β + O(g 5 ) . (7.143)
Estes resultados não dependem de se ter adoptado subtracção minı́ma ou não. Se
adoptarmos subtracção mı́nima temos então
Teorema 3.3
No esquema de subtracção mı́nima temos ρ = σ = 0 e portanto
∂
D= ; β=β e γ m = γm (7.144)
∂ξ
e β e γm são independentes da gauge em todas as ordens.

Dem: Demonstramos só para ρ, para σ é igual.
∂ g ∂Zg
ρ=g ln g = − (7.145)
∂ξ Zg ∂ξ
Então
∂ a1 a2
0 = Zg ρ + g 1+ + 2 + ···
∂ξ ε ε

1 ∂a1
= ρ+ ρa1 + g + O(1/ε2 ) (7.146)
ε ∂ξ
obtemos portanto
ρ=0 . (7.147)
7.3 Effective gauge couplings

7.3.1 Fixed points
Como vimos na secção anterior, o comportamento assimptótico das funções de Green
irredutı́veis depende do comportamento assimptótico das soluções das equações para a
constante de acoplamento efectivo λ(t) e para a massa efectiva, que como vimos são

 dλ

dt = β(λ) ; λ(0) = λ
(7.148)

 dm
= γm (λ) − 1 m(t) ; m(0) = m
dt
Destas equações resulta que as variações da constante de acoplamento efectiva e da
massa efectiva com uma variação de escala de energia são controladas pelas funções β e
γm , respectivamente. Para estudar o comportamento assimptótico de λ vamos admitir
que β(λ) tem a forma da figura 7.2.
β(λ)
0 λ1 λ2 λ
Figure 7.2: β(λ) as function pf λ.
Os pontos, 0, λ1 e λ2 onde β(λ) se anula são chamadas pontos fixos, pois se λ seencontra

num desses pontos em t = 0 então ficará aı́ para todos os valores do momento dλ dt = 0 .
Os pontos fixos podem ser de dois tipos:
7.3. EFFECTIVE GAUGE COUPLINGS 269
i) Ponto fixo estável ultravioleta(UV)

São aqueles em que β ′ (λ) < 0. É o caso do ponto λ1 na figura 7.2. Neste caso β(λ) > 0
para λ < λ1 e β(λ) < 0 para λ > λ1 . Então se para t = 0 0 < λ < λ1 então quando
t → ∞ λ → λ1 . Por outro lado se λ1 < λ < λ2 quando t → ∞ também λ → λ1 .
Poratnto no intervalo 0 < λ < λ2 a constante de acoplamento é sempre conduzida para
λ1 quanto t → ∞, isto é, para momentos grandes.
ii) Ponto fixo estável infravermelho(IR)
São aqueles em que β ′ (λ) > 0. É o caso dos pontos 0 e λ2 da figura. É fácil de ver que
quando t → ∞ a constante de acoplamento se afasta de 0 e λ2 , mas que no limite t → 0
se aproxima deles.
Podemos agora estudar o comportamento assimptótico das soluções do grupo de renor-
malização. Supomos, por exemplo 0 < λ < λ2 . Então (ver figura 7.2)
lim λ(t, λ) = λ1 (7.149)

t→0
A maneira como tende para λ1 depende da primeira derivada de β(λ). Suponhamos

que na vizinhança de λ1 temos
β(λ) = a(λ1 − λ) ; a>0
β ′ (λ1 ) = −a < 0 (7.150)
Então
λ(t, λ) = λ1 + (λ − λ1 )e−at (7.151)

isto é, a aproximação do ponto fixo é exponencial na variável t. Será tanto maior quanto
maior for |β ′ (λ1 )| = a. Vimos anteriormente que a solução da equação da massa efectiva
era
Rt
γm (λ)dt′
m(t) = me−t e 0 (7.152)
Se lim λ = λ1 então temos para t → ∞
t→∞
m = me−t(1−γm (λ1 )) (7.153)

o que mostra que se γm (λ1 ) < 1 então m(t) → 0 quando t → ∞. Na mesma aproximação
Z t
γ(λ(t′ ))dt′ ≃ γ(λ1 )t (7.154)
0
e portanto a solução assimptótica é
lim Γn (σpi , m, λ, µ) = σ 4−n[1+γ(λ1 )] Γ(n) (pi , m, λ1 , µ) (7.155)

σ→∞
o que mostra que a dimensão dos campos não é 1 mas 1 + γ(λ1 ). Daı́ o nome de dimensão
anómala para γ(λ).
Em geral é difı́cil calcular os zeros da função β, pois requere normalmente resultados
para além da teoria de perturbações. Contudo β(λ), γm (λ) e γ(λ) têm um zero trivial
na origem. Se acontecer que a origem seja um ponto fixo estável U V então quer dizer
que quando a escala da energia aumenta a constante de acoplamento diminui. No limite
t → ∞, λ → 0 e por isso se diz destas teorias que são assimptoticamente livres. É fácil de
ver que isso acontece se β ′ (0) < 0. Na secção seguinte vamos ver quais as teorias em que
isso pode acontecer.
7.3.2 β function for theories with scalars, fermions and gauge fields
Vamos nesta secção mostrar que só as teorias de gauge não abelianas podem ser assimp-
toticamente livres, isto é, só estas verificam a propriedade β ′ (0) < 0.
i) Teorias com escalares
Já vimos anteriormente que para a teoria escalar mais simples, λφ4 , temos
3λ2
β(λ) = + O(λ4 ) (7.156)
16π 2
e portanto não é assimptoticamente livre. Consideramos agora a teoria escalar mais geral
com campos φi e acoplamento
LI = −λijkℓ φi φj φk φℓ (7.157)
onde se somam os ı́ndices repetidos. Então
dλijkℓ (t)
βijkℓ = = A(λiℓmn λkjmn + λijmn λkℓmn + λikmn λjℓmn ) (7.158)
dt
com A > 0. A teoria não é assimptoticamente livre pois há sempre funções β com derivadas
positivas. Por exemplo
dλ1111
= β1111 = 3A|λ11mn |2 > 0 ; ∀t (7.159)
dt
ii) Teorias com escalares + fermiões + acoplamentos de Yukawa
O termo da interacção mais geral para uma teoria com escalares e fermiões é
X X a b
LI = − λijkℓ φi φj φk φℓ + k
ψ (Akab + iBab γ5 )ψ φk (7.160)
i,j,k,ℓ a,b,k
onde A e B são matrizes reais. Agora já não é possı́vel mostrar que dλdtiiii > 0 por
causa do loop de fermiões de ordem A2 ou B 2 com um sinal negativo. Se definirmos
(gi )ab ≡ Aiab + iBab
i , obtemos
dg i
16π 2 = (Trgi gj† )gj + Tr(gi† gj )g j + M ij gj
dt
1 1
+ g i g †j g j + gj g†j gi + 2gj g†i gj (7.161)
2 2
onde M ij ≡ 14 λikℓm λjkℓm . Usando este resultado é possı́vel demonstrar o teorema seguinte:
Teorema 3.4
A teoria mais geral com escalares e fermiões não é assimptoticamente livre pois
d i† i
dt Tr(g g ) > 0 e portanto não é possivel gi → 0 quando t → ∞.
Dem:
d d X i 2
8π 2 Tr(gi† g i ) = 8π 2 |gab |
dt dt
a,b,i
= Tr(gi g j† )Tr(gi† g j ) + Tr(gi gj† )(Trgi gj† )

1 1
+ Tr(gi gi† gj gj† ) + Tr(gi† gi gj† gj )
2 2
+2Tr(gi gj† gi gj† ) + M ij Tr(gi† gj ) (7.162)
Agora o último termo é positivo, assim como o terceiro e o quarto. O primeiro é

maior que o segundo e portanto
d h i
8π 2 Tr(gi† gi ) ≥ 2 Tr(gi gj† )Tr(gi gj† ) + Tr(gi gj† g i g j† ) (7.163)
dt
e o segundo membro é positivo pois pode ser escrito
d j† j† j† j†
8π 2 Tr(gi† g i ) ≥ (gab
i i i i
gcd + gad gcd )(gba gdc + gda gbc ) ≥ 0 (7.164)
dt
como querı́amos demonstrar.
iii) Teorias gauge abelianas

Consideremos o caso de QED. Temos
−1/2 −1/2
Ze = Z1 Z2−1 Z3 = Z3 (7.165)
Z3 pode ser calculado do diagrama de polarização do vácuo representado na Fig. 7.3, e o
resultado é
Figure 7.3: Vacuum polarization in QED.
−1/2 e2 1
Z3 =1+ + ··· (7.166)
12π 2 ε
logo
1 da1 e3
β(e) = e2 = >0 (7.167)
2 de 12π 2
Se tivéssemos electrodinâmica escalar Z3 seria obtido a partir dos diagramas da figura 7.4,
e o resultado seria
Figure 7.4: Vacuum polarization in scalar electrodynamics.
−1/2 e2 1
Z3 =1+ (7.168)
48π 2 ε
e3
o que dá neste caso β(e) = 48π 2 > 0. Portanto as teorias de gauge abelianas não são livres
assimptoticamente.
iv) Teorias de gauge não abelianas
Comecemos pela teoria de gauge pura definida no capı́tulo 2. A renormalização da
função de onda para os campos de gauge é obtida a partir dos diagramas da figura 7.5.
Em subtracção mı́nima obtemos,
Figure 7.5: Vacuum polarization in a pure non-abelian gauge theory.

g2 13 1
ZA = 1 + − ξ C2 (V ) (7.169)
16π 2 3 ε
onde C2 (V ) é o operador de Casimir definido no capı́tulo 2 para a representação adjunta,
a que pertencem os campos da gauge (vectores). A constante de renormalização do vértice
triplo, Z1 , é obtida a partir dos diagramas da figura 7.6.
Obtemos

g2 17 3ξ 1
Z1 = 1 + 2
− C2 (V ) + · · · (7.170)
16π 6 2 ε
Então
Figure 7.6: Vertex in a pure non-abelian gauge theory.

−3/2 g2 11 1
Zg ≡ Z1 ZA =1− C2 (V ) + ··· (7.171)
16π 2 3 ε
Usando ZA e Zg e as definições de β e γ obtemos
g3 11
β=− C2 (V ) < 0 (7.172)
16π 2 3
e

g2 1 13
γA = − − ξ C2 (V ) (7.173)
16π 2 2 3
Portanto as teorias de gauge não abelianas sem campos de matéria são assimptoticamente
livres. Notar que a dependência da gauge (em ξ) desapareceu de β de acordo com o
resultado demonstrado anteriormente.
A inclusão de fermiões e escalares acoplados mı́nimamente é agora trivial. O la-
grangeano de interacção é
Lint = gψ i γ µ ψj TFa ij Aaµ

↔
+igφ∗i ∂ µ φj TSij
a
Aµa
+g2 φ∗i TSij

a b
TSjk φk Aaµ Aµb (7.174)
onde TFa e TSa são os geradores na representação em que se encontram os fermiões e os

escalares respectivamente. Para se encontrar a contribuição destas partı́culas para a função
β temos que calcular a contribuição delas para Zg . O mais fácil é usar os resultados de
QED e electrodinâmica escalar que dizem que
−1/2
Zg = ZA (7.175)
e calcular a contribuição para ZA dos fermiões e escalares devido aos diagramas da figura
7.7. O resultado é
Figure 7.7: Contribution from fermions and scalars to vacuum polarization.

g2 4 1 1
Zg (fermiões + escalares) = 1 + T (R F ) + T (R S ) + ··· (7.176)
16π 2 3 3 ε
pelo que
g3 4
β(fermiões) = T (RF ) (7.177)
16π 2 3
e
g3 1
β(escalares) = T (RS ) (7.178)
16π 2 3
Pondo tudo junto obtemos

g3 11 4 1
β= − C2 (V ) + T (RF ) + T (Rs ) (7.179)
16π 2 3 3 3
onde as quantidades T (R) são definidas para uma dada representação por
Tr(T a T b ) = T (R)δab (7.180)

Se a teoria contém fermiões de Majorana (ou spinores de Weyl) ou campos escalares,
os coeficientes em frente de T (RF ) e T (RS ) são multiplicados por um factor adicional de
1/2. Consideremos agora um exemplo simples:
Exemplo 3.1:
QCD (SU (3)) com as três familias de quarks.
Para SU (N ) temos
C2 (V ) = N (7.181)
e como os quarks se encontram na representação fundamental
1
T (RF ) = (7.182)
2
Então

g 33 4 1
β= − + × × 2N g (7.183)
16π 2 3 3 2
ou seja (Ng = número de gerações ou famı́lias)

g3 33 − 4Ng
β= − (7.184)
16π 2 3
Portanto SU (3) será assimptoticamente livre se
33 − 4Ng > 0 (7.185)
ou ainda
33
Ng < → Ng ≤ 8 (7.186)
4
São portanto permitidas 8 famı́lias ou seja 16 tripletos de SU (3).
7.3.3 The vacuum of a NAGT as a paramagnetic medium (µ > 1)

Um argumento recente (Nielsen 1981, Hughs 1981) permite compreender melhor o que
se passa de diferente nas teorias de gauge não abelianas para que elas tenham liberdade
assimptotica. Primeiro o facto de a carga diminuir a curta distância pode ser interpretado
como um anti - shielding do vácuo, isto é
ε<1 (7.187)
O problema em compreender o que se passa resulta do facto de não conhecermos
substâncias3 com ε < 1. Contudo o vácuo deve ser invariante relativista e portanto deve
ter uma permeabilidade µ tal que (estamos a fazer c = 1)
µε = 1 (7.188)
Assim o antiscreening corresponde a µ > 1. Portanto o vácuo duma teoria de gauge não
abeliana é um paramagnético e este conceito pode ser compreendido mais facilmente.
A permeabilidade magnética pode ser calculada calculando a densidade de energia do
vácuo num campo exterior
1 2
u0 = B (7.189)
2µ ext
Nielsen e Hughes mostraram que µ = 1 + χ onde a susceptibilidade χ é dada por
X 1
2s 2 2 2
χ ∼ (−1) q − + γ s3 (7.190)
s
3
3
3
Em QED a carga aumenta a curta distância e portanto o vácuo é um dieléctrico normal ε>1.
onde s é o spin, q a carga, γ a razão giromagnética e s3 a projecção de spin na direcção

do campo magnético externo. Assim para escalares, fermiões e campos de gauge obtemos
Escalares
1
χS ∼ − qS2 < 0 (diamagnético) (7.191)
3
Fermiões (γF = 2)

1 4
χF ∼ (−1)qF2 2 − + 1 = − qF2 (diamagnético) (7.192)
3 3
Bosões de gauge (γV = 2)

1 22 2
χV ∼ qV2 2 − + 4 = q (paramagnético) (7.193)
3 3 V
e portanto
22 2 4 1
χTotal ∼ qV − qF2 − qS2 (7.194)
3 3 3
Comparando com a função β podemos fazer a correspondência
1
qV2 → C2 (V )
2
qF2 → T (RF )
qS2 → T (RS ) (7.195)
o que permite compreender o vácuo das teorias de gauge não abelianas como um meio
paramagnético.
7.4 Renormalization group applications

We consider the Grand Unified Theory (GUT) with the gauge group SU (5), that is
SU (5) ⊃ SUc (3) × SUL (2) × UY (1) . (7.196)
The unification takes place at the GUT scale MX . Using the renormalization group
equations and the low energy data on the coupling constants, it is possible to determine
the scale MX as well as other predictions for the theory at the low scale, which we take
to be the scale MZ . For this we need to know how the different coupling constants evolve
with the scale.
7.4. RENORMALIZATION GROUP APPLICATIONS 277
7.4.1 Scale MX
We start by writing the covariant derivatives for the unified theory and for the theory with
the broken symmetry.
23
X λa
SU (5) : Dµ = ∂µ + ig5 Aaµ (7.197)
a=0
2
8
X λa
SU (3) × SU (2) × U (1) : Dµ = ∂µ + ig3 Gaµ
α
22
3
X σa Y
+ig2 Aaµ + ig′ Bµ (7.198)
α
22 2
At the scale MX where the unification takes place we have
g5 = g3 = g2 = g1 (7.199)
where g1 is the coupling constant of the abelian subgroup of SU (5). However for the
abelian groups there are no constraints in the normalization of the generators, and there-
fore the the generator λ0 of that U (1) can be normalized in a different way from the
hypercharge. We must have
g1 λ0 = g ′ Y (7.200)
As λ0 is a generator of SU (5) it is normalized according to
TF (λa λb ) = 2δab (7.201)
that is, for the fundamental representation we must have

 
2
 2 
1  
λ =√ 
0
2  (7.202)
15 
 −3


−3
Now, for the fundamental representation, we have

 
 d1 
 
 
 d2 
 
 
5=


d3  (7.203)
 
 
 + 
 e 
 
νec R
and the hypercharge4 can be read directly. We obtain,

 
−2/3
 −2/3 
 
Y =  −2/3  (7.204)

 1 
1
q q
5 0 3
Therefore Y = − 3λ and g′ = − 5 g1 . This allows to determine sin2 θW at the GUT
scale MX ,
3
g′2 5 g1 3
sin2 θW (MX ) = 2 ′2
= 3 = (7.205)
g +g g2 + 5 g1 8
Also, for future reference, we note that
3 2
g′2 = g . (7.206)
5 1
7.4.2 Scale MZ
Let us look now at what happens at the scale MZ . The evolution of the coupling constants
is governed by the RGE equations for the three gauge groups in the broken phase
dgi
= βi (7.207)
dt
These β functions are given by
 
gi3  11 X4 X1
βi = − C2 (V ) + T (RFj ) + T (RSk ) (7.208)
16π 2 3 3 3
j k
where the sums are over all the fermion and scalar physical states of the theory at a given
scale. Given the form of Eq. (7.208), it is usual to define
1
βi ≡ bi gi3 (7.209)
16π 2
and therefore the bi are defined by the bracket in Eq. (7.208). Before we evaluate them
let us introduce Eq. (7.209) into Eq. (7.207). We get
dgi bi 3
= g (7.210)
dt 16π 2 i
Let us solve this equations before we evaluate the beta function coefficients bi . For that
it is usual to introduce the generalization of the fine structure constant, that is, we define
gi2
αi ≡ (7.211)
4π
4 Y
Remember that our convention is such that Q = T3 + .
2
Multiplying both sides of Eq. (7.210) by gi and doing some trivial algebra we get,
dαi bi 2
= α (7.212)
dt 2π i
Rearranging and integrating between some initial (µi ), and final scale (µf ), we get
Z f Z f
dαi bi
= dt (7.213)
i α2i 2π i
or5 f
1 bi bi µf
− = (tf − ti ) = ln (7.214)
αi i 2π 2π µi
and finally !
bi µ2f
α−1 −1
i (µf ) = αi (µi ) − ln (7.215)
4π µ2i
As at the unification scale MX we have, by definition (see Eq. (7.199)), that

α1 = α2 = α2 = α5 (7.216)
where α5 is the SU (5) unified value, and we can write the final solution
2
−1 −1 bi MX
αi (µ) = α5 + ln , i = 1, 2, 3 (7.217)
4π µ2
We can rewrite these equations in terms of electromagnetic fine structure constant α(µ)
and of the strong coupling equivalent αs µ), that are measured at the weak scale, to obtain
 2
 −1 b3 MX

 α −1 (µ) = α + ln


s 5
4π µ2



 2
−1 2 −1 b2 MX
α (µ) sin θW (µ) = α5 + ln (7.218)



4π µ2

 2

 b1 MX

 3 cos2 θW (µ)α−1 (µ) = α−1
5 5 + ln
4π µ2
From these equations we obtain,
2
MX 12π 1 8 1
ln 2 = − (7.219)
µ −8b3 + 3b2 + 5b1 α(µ) 3 αs (µ)
That allows to determine MX , once α(µ) and αs (µ) are known, at a given scale µ, and
3(b2 − b3 ) 5(b1 − b2 ) α(µ)
sin2 θW (µ) = + (7.220)
5b1 + 3b2 − 8b3 5b1 + 3b2 − 8b3 αS (µ)
which allows to determine sin2 θW at the scale µ = MZ , once α(MZ ) and αs (MZ ) are
known. Finally we can also solve for the value of α−1
5 . We get

−1 −1 1 α(µ)
α5 = α (µ) −3b3 + (5b1 + 3b2 ) (7.221)
5b1 + 3b2 − 8b3 αS (µ)
Now we turn to the evaluation of the coefficients bi first in the Standard Model (SM)
and the in the Minimal Supersymmetric Standard Model (MSSM).
5
Remember that t = ln(µ).
Standard Model
In the SM we have the gauge fields, Ng = 3 families of leptons, NF = 2Ng = 6 quark
flavours and one Higgs. With this information we can find the coefficients bi for the SM
using the definition
11 X2 X1
bi = − C2 (Vi ) + T (RFj ) + T (RSk ) (7.222)
3 3 3
j k
where we have modified Eq. (7.208), as the sum in the fermions is done separately for
each quirality. This is important for the SM as the model is described in terms of left and
right-handed fermions.
• SU (3)
For SU (3), we have C2 (V3 ) = 3 and the quarks are in the fundamental representation,
therefore T (RFj ) = 1/2. Then the counting goes as follows,
 
11  2 1 
b3 = − × 3 + Ng × 
 × × (2 + 1 + 1) 
 = −7 (7.223)
| 3{z } |3 2 {z }
Gauge quarks
where the meaning of (2 + 1 + 1) is that we count the up and down components of each
(SU (2)L ) doublet and then the corresponding right-handed quarks for each generation.
• SU (2)
For the SU (2) we get

 
11  2 1 2 1  1 1 19
b2 = − × 2 + Ng × 
 Nc × × + ×   + × =− (7.224)
| 3{z } | {z3 2} |3 {z 2} |3 {z 2} 6
Gauge quarksL leptons Higgs
where Nc = 3 is the number of colours.
• U (1)
Finally for the U (1) part, with the correct normalization, we have
 
X 2 X 2
3 2 Y 1 Y
b1 = ×  × + ×  (7.225)
5 3 2 3 2
fL ,fR scalars
and therefore,
 
3   2 1 2 2 2
2
1 2
2
2
  2
b1 = × Ng ×  × − × 2 + × (−1) +Nc × × × 2 +Nc × ×
5  3 2 3 3 6 3 3
| {z } | {z } | {z } | {z }
LeptonsL LeptonsR QuarksL Up−QuarksR
 
2  1 1 2 
2 1  
+Nc × × + × ×2 
3 3  3 2 
| {z } | {z }
Down−QuarksR Higgs
1 41
= 4+ = (7.226)
10
|{z} 10
Higgs
So in summary we have for the SM,

41 19
b1 = , b2 = − , b3 = −7 (7.227)
10 6
Now let us look to see what are the results for MX , sin2 θW (MZ ) and α−1
5 . We will
use the current values from the Particle Data Group. These are (without worrying about
errors)6
α−1 (MZ ) = 127.916, αs (MZ ) = 0.118, MZ = 91.1896 GeV (7.228)
we get
MX = 6.7 × 1014 GeV, sin2 θW (MZ ) = 0.208, α−1

5 = 41.48 (7.229)
At the time that this GUT model was proposed by the first time, the constants were
not known so precisely as today. Also the bound on the lifetime of the proton was much
lower than today. So at that time the model was completely consistent. However after
many years of dedicated experiments for find the decay of the proton, the lower limit was
substantially improved and also after LEP the coupling constants are known with greater
precision. So today the value for MX is too low, the same being true for the value of
sin2 θW (MZ ) (the best value today is around sin2 θW (MZ ) = 0.230 7 ).
This can be seen very clearly if we use Eq. (7.215), with µi = MZ and plot the α−1
i as
2 2
a function of ln(µ /MZ ). This is shown in Fig. 7.8. We clearly see that the agreement is
quite poor with today’s values.
Minimal Supersymmetric Standard Model

Let us now turn to the MSSM. Below the GUT scale the gauge group is the same as in
the SM, but the particle content is larger, more than duplicated in relation to the SM.
We summarize in the Table 7.1 the particle content and their quantum numbers under
G = SUc (3) ⊗ SUL (2) ⊗ UY (1).
With the values in Table 7.1 we can calculate the contribution of the various particles
to the bi coefficients. We will do it in succession for the three groups and for the different
supermultiplets.
• SU (3)
6
Not only errors but also the difference between different renormalization schemes. Also this discussion
is at one-loop level.
7
Again, without discussing the very small errors and the dependence on the renormalization scheme.
60
50 α1-1
40
αi-1
30 α2-1
20
10 α3-1
0
100 104 108 1012 1016
µ (GeV)
Figure 7.8: Evolution of αi as function of the scale µ in the Standard Model for a SU (5)
minimal GUT theory.
– Gauge Supermultiplet
We first do it in general for any gauge group and then apply it to the cases of interest.
The gauge multiplet has a gauge boson contributing with
11
b gauge boson
=− C2 (V ) (7.230)
3
and the left-handed gauginos in the adjoint representation of the gauge group. These
therefore contribute
2
b gauginos = C2 (V ) (7.231)
3
and therefore
b gauge SM = −3 C2 (V ) (7.232)
where SM stands here for super-multiplet. Applying now to SU (3) we get
b3gauge SM
= −9 (7.233)
– Left-handed Lepton Supermultiplet
b3LeptonsL SM = 0 (7.234)
– Right-handed Lepton Supermultiplet
b3LeptonsR SM = 0 (7.235)
Supermultiplet SUc (3) ⊗ SUL (2) ⊗ UY (1)

Quantum Numbers
Vb1 ≡ (λ′ , W1µ ) (1, 1, 1)
Vb2 ≡ (λa , W2µa ) (1, 3, 0)
V3 ≡ (egb , W3µb ) (8, 1, 0)
b i ≡ (L,
L e L)i (1, 2, −1)
bi ≡ (ℓeR , ℓc )i
R (1, 1, 2)
L
b e
Qi ≡ (Q, Q)i (3, 2, 13 )
b i ≡ (deR , dc )i
D (3, 1, 23 )
L
b
Ui ≡ (e c
uR , u )i (3, 1, − 43 )
L
b d ≡ (Hd , H
H ed ) (1, 2, −1)
b u ≡ (Hu , H
H eu) (1, 2, 1)
Table 7.1: Particle content of the MSSM. Note that Q = T3 + Y /2.
– Left-handed Quark Supermultiplet
2 1 1 1
b3QuarksL SM = × ×2 + × ×2 =1 (7.236)
|3 {z
2 } |3 {z
2 }
QuarksL SquarksL
– Right-handed Up-Quark Supermultiplet
2 1 1 1 1
b3Up−QuarkR SM = × + × = (7.237)
| {z 2}
3 |3 {z 2} 2
Up−QuarksR Up−SquarksR
– Right-handed Down-Quark Supermultiplet
2 1 1 1 1
b3Down−QuarkR SM = × + × = (7.238)
|3 {z 2} | {z 2}
3 2
Down−QuarksR Down−SquarksR
– Up type Higgs Supermultiplet
b3Up−Higgs SM = 0 (7.239)
– Down type Higgs Supermultiplet
b3Down−Higgs SM = 0 (7.240)
• SU (2)
Gauge Supermultiplet
We get
b2gauge SM
= −6 (7.241)
2 1 1 1 1
b2LeptonsL SM = × + × = (7.242)
|3 {z 2} |3 {z 2} 2
LeptonsL SleptonsL
b2LeptonsR SM = 0 (7.243)
2 1 1 1 1 3
b2QuarksL SM = Nc × +Nc × = Nc = (7.244)
|3 {z 2} |3 {z 2} 2 2
QuarksL SquarksL
b2Up−QuarkR SM = 0 (7.245)
b2Down−QuarkR SM = 0 (7.246)
1 1 2 1 1
b2Up−Higgs SM = × + × = (7.247)
3 2
| {z } 3 2
| {z } 2
Higgsu Higgsinou
1 1 2 1 1
b2Down−Higgs SM = × + × = (7.248)
3 2
| {z } 3 2
| {z } 2
Higgsd Higgsinod
• U (1)
Gauge Supermultiplet
We get
b1gauge SM
=0 (7.249)
 
2 2
3 
2 1 1 1 
 3
b1LeptonsL SM = × × − ×2 + × − × 2 = (7.250)
5 3 2 3 2  10
| {z } | {z }
LeptonsL SleptonsL
 
3  2 1  3
b1LeptonsR SM = × × (−1) 2
+ × (−1) 2
= 5 (7.251)
5 |3 {z } |3 {z }
LeptonsR SleptonsR
 
3  2 1 2 1
2
1  3 1 1
 
b1QuarksL SM = × Nc ×  × ×2 + × × 2  = Nc × × = (7.252)
5 3 6 3 6  5 18 10
| {z } | {z }
QuarksL SquarksL
 
3  2 2 2 1
2 
2  3 4 4

b1Up−QuarksR SM = × Nc ×  × + ×  = Nc × × = (7.253)
5 3 3 3 3  5 9 5
| {z } | {z }
Up−QuarksR Up−SquarksR
 
3  2 1 2 1
2 
1  3 1 1

b1Down−QuarksR SM = × Nc ×  × − + × −  = Nc × × = (7.254)
5 3 3 3 3  5 9 5
| {z } | {z }
Down−QuarksR Down−SquarksR
 
2 2
3 
1 1 2 1  3 1
 3
b1Higgsu SM = × × ×2 + × × 2 = × = (7.255)
5 3 2 3 2  5 2 10
| {z } | {z }
Higgsu Higgsu
 
2 2
3 
1 1 2 1  3 1
 3
b1Higgsd SM = × × − ×2 + × − × 2 = × = (7.256)
5 3 2 3 2  5 2 10
| {z } | {z }
Higgsd Higgsd
Now we put everything together to obtain fro the MSSM,

3 3 1 4 1 3 3 3 33
b1 =Ng × + + + + + + =3×2+ =
10 5 10 5 5 10 10 5 5

1 3 1 1
b2 = − 6 + Ng × + + + =1
2 2 2 2

1 1
b3 = − 9 + Ng × 1 + + = −3 (7.257)
2 2
Now let us look to see what are the results for MX , sin2 θW (MZ ) and α−1
5 in the
8
MSSM. Using the same inputs as for the SM, Eq. (7.228) , we get
MX = 2.1 × 1016 GeV, sin2 θW (MZ ) = 0.231, α−1

5 = 24.27 (7.258)
we immediately see that these values are quite good. This can be seen very clearly if we
use Eq. (7.215), with µi = MZ and plot the α−1 2 2
i as a function of ln(µ /MZ ). This is shown
in Fig. 7.9 and the agreement is excellent.
We can still go a step further. We know that supersymmetry must be broken above
the electroweak scale, so what we have done in Fig. 7.9 is not quite correct because we
are running with the MSSM content down to the weak scale. Of course each particle will
decouple at its mass, but assuming that their masses are not much different we can assume
that there will a scale MSUSY , below which we will have the SM RGEs. We can redo the
calculation taking now the evolved SM values at MSUSY as the boundary conditions for
the MSSM evolution. In Fig. and Fig. the results are shown for various values of the
SUSY scale. We see from these results that if the SUSY scale is much higher than, say
8
Again we do not take into account errors and the difference between different renormalization schemes.
Also this discussion is at one-loop level and the effects of the supersymmetric particles not decoupling at
the same scale (thresholds) are not taken in account.
60
50
α1-1
40
αi-1
30
α2-1
20
10
α3-1
0
100 104 108 1012 1016
µ (GeV)
Figure 7.9: Evolution of αi as function of the scale µ in the MSSM for a SU (5) minimal
GUT theory.
1 TeV, the good agreement starts do disappear. Before we end we should emphasize that
these are one loop results, without many fine details, like thresholds (talking in account
that not all the supersymmetric particles decouple at the same scale) and the important
two-loop effects.
60 60
50 50 -1
α1
-1 α1
40 40
-1
-1
30 30
αi
αi
-1
α2
-1 α2
20 20
-1
10
α3
-1 10 α3
0 0
100 104 108 1012 1016 100 104 108 1012 1016
µ (GeV) µ (GeV)
Figure 7.10: Evolution of αi as function of the scale in the MSSM for a SU (5) minimal
GUT theory. On the left panel MSUSY = 500 GeV and on the right panel MSUSY = 1000
GeV.
60 60
-1
50 α1
-1 50 α1
40 40
-1
-1
30 30 -1
α2
αi
αi
-1
α2
20 20
-1
α3
-1 α3
10 10
0 0
100 104 108 1012 1016 100 104 108 1012 1016
µ (GeV) µ (GeV)
Figure 7.11: Evolution of αi as function of the scale in the MSSM for a SU (5) minimal
GUT theory. On the left panel MSUSY = 104 GeV and on the right panel MSUSY = 108
GeV.
dg i
3.1 Verifique a equação 7.161. Para isso note que β i = dt onde β i é calculada a partir
dos diagramas seguintes
Problems 289
i i i
i i
3.2 Calcule em subtração mı́nima (MS) a constante de renormalização Z3 para QED,

equação 7.166.
3.3 Calcule em MS a constante de renormalização Z3 em electrodinâmica escalar, equação

7.168.
3.4 Considere uma teoria não abeliana com grupo de simetria G e sem matéria. Calcule
as constantes de renormalização do propagador ZA , e do vértice triplo Z1 .
3.5 Considere uma teoria não abeliana em interacção com campos escalares e fermióni-
cos. Calcule a contribuição destes campos para ZA e Z1 . Utilize estes resultados
juntamente com os resultados do problema 3.4 para determinar a função β do grupo
de renormalização para essa teoria.
3.6 Considere o modelo padrão das interacções electrofracas e fortes. Considerando todos
os campos do modelo (incluindo os Higgs) calcule os coeficientes b1 , b2 e b3 definidos
na equação (tc2-3.201).
3.7 Considere agora o modelo padrão supersimétrico mı́nimo (MSSM). Calcule os coefi-
cientes b1 , b2 e b3 definidos na equação (tc2-3.201). Refaça a análise da convergência
das constantes de acoplamento no quadro duma teoria de grande unificação super-
simétrica com o grupo SU (5).
Appendix A
Path Integral in Quantum

Mechanics
A.1 Introduction
A formulação usual é dada pela equação de Schrödinger
∂
ih̄ |a(t)i = H |a(t)i (A.1)
∂t
onde
P2
H= + V (Q) (A.2)
2m
e
[Q, P ] = ih̄ (A.3)
Esta formulação é equivalente a uma outra definida em termos de integrais de caminho.

Para isso baseamo-nos na observação que em mecânica quântica sabemos responder a
qualquer pergunta sobre o sistema se soubermos calcular as amplitudes de transição

′
b(t′ )|a(t) = hb| e−iH(t −t) |ai (A.4)
São estas amplitudes de transição que são definidas em termos de integrais de caminho.
Conforme a representação escolhida para os estados |ai e |bi as expressões para o integral de
caminho vêm diferentes. Assim vamos analisar separadamente os casos das representações
no espaço das configurações (coordenadas), no espaço de fase e por meio de estados coe-
rentes (espaço de Bargmann-Fock).
A.2 Configuration space

Introduzimos os estados |qi e |pi tais que
291
292 APPENDIX A. PATH INTEGRAL IN QUANTUM MECHANICS
Q |qi = q |qi ; P |pi = p |pi
hq ′ |qi = δ(q ′ − q) ; hp′ |pi = δ(p′ − p)
hq|pi = hp|qi∗ = √1 eipq

2π
(A.5)
Então
Z R tf
i
dtL(q,q̇)
hqf , tf |qi , ti i = D(q)e h̄ ti
(A.6)
onde D é uma medida de integração definida pelo limite
n−1
" #n
Y nme−iπ/2 2
D(q) = lim dqp (A.7)

n→∞ 2π(tf − ti )
1
sendo n o número de intervalos em que se fez a partição do intervalo (ti , tf ). O limite

n → ∞ é bastante complicado e só existe prova matemática para certas classes de potenci-
ais. A Eq. (A.6) permite uma interpretação da mecânica clássica como limite da mecânica
quântica. De facto quando h̄ → 0 a maior contribuição para a amplitude vem das tra-
jectórias que minimizam a ac cão, isto é, as trajectórias clássicas. A mecânica quântica é
então vista como o estudo das flutuações à volta da trajectória clássica.
A.2.1 Matrix elements of operators

Usando a propriedade dos integrais de caminho
Z Z Z Z
iS(f,i) iS(f,t)
D(q)e = dq(t) D(q)e D(q)eiS(t,i) (A.8)
onde ti < t < tf , é fácil mostrar que
Z Z
′ ′′ ′′ ,t)
hqf , tf | O(t) |qi , ti i = dq dq D(q)eiS(qf ,tf ;q
Z

′′ ′ ′
q O q D(q)eiS(q ,t;qi ,ti ) (A.9)
Então se O for diagonal no espaço das coordenadas, isto é, se

′′ ′
q O q = O(q ′ )δ(q ′ − q ′′ ) (A.10)
obtemos
Z
hqf , tf | O |qi , ti i = D(q) eiS(f,i) O(q(t)) (A.11)
A.2. CONFIGURATION SPACE 293
A.2.2 Time ordered product of operators

Seja O1 (t1 )O2 (t2 ) · · · On (tn ) com t1 ≥ t2 ≥ · · · ≥ tn . Então é fácil de mostrar que a
ordenação no tempo é automática no integral de caminho, isto é,
Z
hqf , tf | O1 (t1 )O2 (t2 ) · · · On (tn ) |qi , ti i = D(q)eiS(f,i) O1 O2 · · · On (A.12)
Este resultado é particularmente importante, pois permitirá escrever as funções de Green

de produtos de operadores ordenados no tempo como simples integrais de caminho de
produtos dos equivalentes clássicos desses operadores.
A.2.3 Exact results I: harmonic oscillator

Para alguns potenciais é possı́vel calcular exactamente o limite introduzido em (A.5). Para
esses casos o integral de caminho é portanto perfeitamente bem definido. Esses potenciais
não são muitos, mas são particularmente importantes. Para o seguimento interessa-nos
discutir dois deles. O primeiro é o oscilador harmónico definido pelo potencial
ω2 2
V (Q) = mQ (A.13)
2
Para este caso obtém-se, (os integrais são gaussianos e por isso podem ser explicitamente
calculados)
!1
mωe−iπ/2 2
mω 2 2 2qf qi
hf |ii = exp i (qf + qi ) cot ωt − (A.14)
2π sin ωt 2 sin ωt
Este resultado vai ser útil adiante.
A.2.4 Exact results II: external force

Consideremos agora uma força exterior tal que o potencial é dado por
V (Q) = −QF (t) (A.15)

Neste caso obtemos
" #1
me−iπ/2 2
hf |iiF = eiS(f,i) (A.16)

2π(tf − ti )
onde S(f, i) é a acção calculada ao longo da trajectória clássica,
Z tf
m (qf − qi )2 t − ti tf − t
S(f, i) = + dtF (t) qf + qi
2 tf − ti ti tf − ti tf − ti
Z tf Z tf
1
+ dt′ dt′′ F (t′ )G(t′ , t′′ )F (t′′ ) (A.17)
2m ti ti
′ ′′ ..
onde G(t′ , t′′ ) = t Tt −Inf(t′ , t′′ ) é a função de Green simétrica para o problema q = F (t)/m
com as condições na fronteira G(0, t′′ ) = G(t′ , 0) = 0.
A.2.5 Perturbation theory

A importância do resultado exacto para a força exterior deve-se ao facto que usando
esse resultado podemos formalmente resolver o problema dum potencial qualquer. Para
isso notemos que a derivação funcional em realação à fonte F (t) faz baixar Q(t). Mais
explicitamente
δ
hf | Q(t) |iiF = hf |iiF (A.18)
iδF (t)
onde h|iF significa calculado na presença da fonte exterior (i.e. para o hamiltoniano
H = P 2 /2m − QF (t)). Então para um potencial arbitrário V (q) temos
Z R tf
dt[ 21 mq̇ 2 −V (q)]
hf |ii = D(q)ei ti
Z tf
δ
= exp −i dtV hf |iiF (A.19)
ti iδF (t) F =0
Esta expressão formal torna-se muito útil quando a exponencial é expandida em série.
Então todos os integrais são do tipo gaussiano e podem ser exactamente executados.
Obtemos assim a teoria das perturbações. Claro que só terá significado se houver um
parâmetro pequeno no potencial. É importante notar que enquanto se faça teoria das per-
turbações não há qualquer problema com a indefinição matemática do integral de caminho,
pois todas as integrações são gaussianas.
A.3 Phase space formulation

Para este caso obtemos
Z R tf
i dt pq̇−h(p,q)
hqf , tf |qi , ti i = D(p, q)e ti
(A.20)
onde h(p, q) é o hamiltoniano clássico e a medida é dada pelo limite

n
Y n−1
Y
dps dqr
1 1
D(p, q) = lim (A.21)
n→∞ (2π)n
A fase da exponencial é novamente a acção clássica expressa nas variáveis canónicas p
e q. Se h(p, q) depender quadraticamente de p como é usual, pode-se fazer a integração
gaussiana em p e a expressão reduz-se à do caso anterior, equação (A.4).
A.4 Bargmann-Fock space (coherent states)

Nesta representação usamos funções analı́ticas de variável complexa para descrevermos os
operadores a e a† ( [a, a† ] = 1). A correspondência é feita do modo seguinte. As funções
analı́ticas geram um despaço de Hilbert com o produto interno definido por
A.4. BARGMANN-FOCK SPACE (COHERENT STATES) 295
Z
dzdz −zz
hg|f i ≡ e g(z)f (z) (A.22)
2πi
Os operadores a e a† são representados neste espaço por
∂
a→
∂z
a† → z (A.23)
Dado um estado |f i, representado pela função f (z), a acção do operador A em |f i produz

outro estado que também pode ser representado por funções analı́ticas. Se designarmos
por g(z) essa função temos
Z
dξdξ −ξξ
g(z) ≡ hz| A |f i ≡ e A(z, ξ)f (ξ) (A.24)
2πi
onde A(z, ξ) é o kernel do operador A. Uma representação explicita para o kernel é fácil
de obter. Para isso introduzimos os estados |ni, definidos por
a†n
|ni = √ |0i (A.25)
n!
É fácil de verificar (ver Problema A.2) que com a definição de produto interno acima
introduzida estes estados são ortonormados, isto é hfm |fn i = δmn .
Então
X zn
hz| A |f i = √ hn| A |mi hm|f i
n,m n!
X zn Z
dξdξ −ξξ ξ m
= √ An,m e √ f (ξ)
n,m n! 2πi m!
Z " #
dξdξ −ξξ X z n ξm
= e √ An,m √ f (ξ) (A.26)
2πi n,m n! m!
Portanto
X zn ξm
A(z, ξ) ≡ √ An,m √ (A.27)
n,m n! m!
O kernel de qualquer operador é assim obtido desde que se conheçam os seus elementos
de matriz na base |ni.
Já vimos como se representam estados e operadores. Vamos ver como representar
produtos de operadores. Sejam dois operadores A1 e A2 e um estado |f i. Seja ainda
Z
dξdξ −ξξ
g(η) = hη| A2 |f i = e A2 (η, ξ)f (ξ) (A.28)
2πi
Então
Z
dξdξ −ξξ
hz| A1 |gi = e A1 (z, η)g(η)
2πi
Z
dηdη dξdξ −ξξ −ηη
= e e A1 (z, η)A2 (η, ξ)f (ξ)
2πi 2πi
Z Z
dξdξ −ξξ dηdη −ηη
= e e A1 (z, η)A2 (η, ξ) f (ξ)
2πi 2πi
Z
dξdξ −ξξ
= e A3 (z, ξ)f (ξ) (A.29)
2πi
Portanto o kernel do operador A3 = A1 A2 é obtido por convolução dos kernéis de A1 e
A2 , isto é
Z
dηdη −ηη
A3 (z, η) = e A1 (z, η)A2 (η, ξ) (A.30)
2πi
A.4.1 Normal form for an operator

Como já sabemos representar estados, operadores e produtos de operadores já temos toos
os ingredientes para fazer mecânica quântica neste espaço. Há contudo um outro assunto
que é importante tendo em atenção que pretendemos aplicar este formalismo em teoria
quântica dos campos. Trata-se da forma normal dum operador1 . O operador A na sua
forma normal é definido por
X a†n am
A= AN
n,m √ √ (A.31)
n,m n! m!
isto é, os operadores de destruição estão à direita dos operadores de criação. O kernel
normal é definido por
X zn zm
AN (z, z) ≡ √ ANn,m √ (A.32)
n,m n! m!
isto é, é obtido por subtituição directa dos operadores de destruição por z e dos de criação
por z. Para um operador dado na sua forma normal este é o kernel imediato de obter. É
contudo diferente do kernel atrás definido. Para ver a relação entre eles notemos a seguinte
relação
X z
f (z) = √ hn|f i
n n!
Z
X dξdξ znξn
= e−ξξ f (ξ)
n
2πi n!
1
Notar que em teoria quântica dos campos tem que se proceder ao ordenamento normal do hamiltoniano
para definir o zero da energia.
Z
dξdξ −ξξ zξ
= e e f (ξ) (A.33)
2πi
O kernel ezξ é portanto uma função delta neste espaço. Usando este resultado obtemos
ANX
n,m dm
hz| A |f i = √ √ zn f (z)
n,m n! m! dz m
X AN Z
n,m dξdξ −ξξ zξ n m
= √ √ e e z ξ f (ξ)
n,m n! m! 2πi
Z
dξdξ −ξξ zξ N
= e e A (z, ξ)f (ξ) (A.34)
2πi
donde resulta
A(z, ξ) = ezξ AN (z, ξ) (A.35)

Esta relação é muito importante pois permite imediatamente escrever o kernel dum ope-
rador qualquer uma vez que seja conhecida a sua forma normal. Isto é particularmente
útil em teoria quântica dos campos onde o hamiltoniano é dado na sua forma normal.
A.4.2 Evolution operator

Podemos obter agora a expressão para o operador de evolução nesta representação. De
acordo com aquilo que acabámos de dizer, para um intervalo infinitesimal, devemos ter
para o kernel de U
U (z, ξ, ∆t) = ezξ e−i∆t h(z,ξ)

(A.36)
onde h(z, ξ) é o kernel normal obtido por substituição directa dos operadores a† e a pelas
variáveis complexas z e ξ. Notar que quando ∆t → 0 o kernel do operador de evolução se
reduz ao kernel da identidade, ezξ , que como vimos é a função δ neste espaço.
Para um intervalode tempo finito t = tf − ti , dividimos o intervalo em n intervalos
t
∆t =
n
z0 z1 z2 ··· zn−1 zn (A.37)
Então
U (z 1 , z0 ) ≃ ez 1 z0 −i∆th(z 1 ,z0 )
U (z 2 , z1 ) ≃ ez 2 z1 −i∆th(z 2 ,z1 )
.. .. ..
. . .
U (z n , zn−1 ) ≃ ez n zn−1 −i∆th(z n ,zn−1 ) (A.38)
Aplicando agora a regra de multiplicação dos kernéis obtemos
Z n−1
Y dzk dz k hX
n
U (z f , tf ; zi , ti ) = lim exp z k zk−1
n→∞ 2πi
k=1 k=1
n−1
X n
X i
− z k zk − i h(z k , zk−1 )∆t (A.39)
k=1 k=1
ou seja
Z R tf
1
U (z f , tf ; zi , ti ) ≡ D(z, z) e 2 (z f zf +z i zi )+i ti [ 2i1 (żz−zż)−h(z,z)]dt (A.40)
Nesta expressão z f (tf ) e zi (ti ) são fixados pelas condições fronteiras mas z f (ti ) e zi (tf )
são arbitrários. A fase da exponencial é novamente a acção, agora escrita nas variáveis
complexas z e z. Para ver isso basta lembrar que
1 1
(pdq + qdp) = (zdz − zdz) (A.41)
2 2i
A.4.3 Exact results I: harmonic oscillator

Também aqui vamos analisar os casos importantes em que há resultados exactos, nomeada-
mente o oscilador hamónico e o caso das fontes externas. Comecemos pelo oscilador
harmónico. O hamiltoniano é dado por
H 0 = ω a† a (A.42)
Trata-se portanto dum caso em que o hamiltoniano é dado na forma normal. Este problema
pode ser resolvido exactamente. Temos
Z n−1
Y dzk dz k hX
n
U (z f , zi , t) = lim exp z k zk−1
n→∞ 2πi
k=1 k=1
n−1
X n
t X i
− z k zk − iω z k zk−1
n
k=1 k=1
Z n−1
Y dzk dz k
= lim e[−XAX+XB+BX ] (A.43)
n→∞ 2πi
k=1
onde
 
z1
 z2 
 
X= ..  ; X = (z 1 , z 2 , · · · , z n−1 ) (A.44)
 . 
zn−1
e
 
z0 a
 0 
 
B= .  ; B = (0, 0, · · · , 0, zn a) (A.45)
 .. 
0
com z0 = zi e zn = zf . A matriz A de dimensão (n − 1) × (n − 1) é dada por
 
1 0 ··· ··· ··· ···
−a 1 0 ··· ··· ···
 
 0 −a 1 0 ··· ···
 
 .. .. .. .. .. ..  (A.46)
 . . . . . . 
 
· · · · · · 0 −a 1 0 
··· ··· ··· 0 −a 1
onde se definiu
t
a ≡ 1 − ih̄ω (A.47)
n
As (n − 1) integrações gaussianas podem ser facilmente feitas usando o resultado (ver
Problema A.3),
Z Y
dzk dz k −zAz+uz+zu −1
e = (det A)−1 euA u (A.48)
2πi
obtemos então
h −1
i
U0 (z f , zi ; t) = lim (det A)−1 eBA B
n→∞
h 2 −1
i
= lim (det A)−1 ez f zi a (A )n−1,1 (A.49)
n→∞
É fácil de verificar que para a matriz A se tem

 k−m
a se k ≥ m
−1
(A )k,m = (A.50)

0 se k < m
e portanto
det A = 1 (A.51)
e
A−1 n
n−1,1 = (−1) (−a)
n−2
(A.52)
donde se conclui que
n
2 iωt
lim a (A−1 )n−1,1 = lim 1− = e−iωt (A.53)
n→∞ n→∞ n
Obtemos então finalmente

U0 (z f , zi ; t) = exp z f zi e−iωt (A.54)
Podemos verificar que este resultado é da forma eiS onde S é a acção calculada ao longo
da trajectória clássica. De facto a estacionaridade do expoente da exponencial dá
Z tf
1 żz − z ż
δ (z f z(tf ) + z(ti )zi ) + − iωzz dt
2 ti 2
1 1 1 1
= z f δz(tf ) + zi δz(ti ) − z f δz(tf ) − zi δz(ti )
2 2 2 2
Z tf

+ δz(ż − iωz) − δz(ż + iωz) dt (A.55)
ti
pois δz f = δzi = 0. As equações de movimento são portanto

 
 ż − iωz = 0  z(tf ) = zf
com (A.56)
 
ż + iωz = 0 z(ti ) = zi
que têm como solução

 z(t) = zi eiω(ti −t)
(A.57)

z(t) = z f eiωt−tf )
Substituindo estas soluções no expoente obtemos
Z tf
1 1
[z f z(tf ) + zi z(ti )] + (żz − z ż) − iωzz dt
2 ti 2
= z f zi eiω(ti −tf )
= z f zi e−iωt (A.58)
para t ≡ tf − ti , como querı́amos mostrar.

Um outro resultado importante do oscilador harmónico é que a evolução dum estado
sob a acção de H0 = ωa† a é particularmente simples neste espaço das funções de variável
complexa. Seja f (z) a representação do estado |f i. A evolução debaixo de H0 é dada por
Z
dξdξ −ξξ zξe−iωt
U0 (t)f (z) = e e f (ξ)
2πi
= f (ze−iωt ) (A.59)
isto é, é reduzida à multiplicação por e−iωt

A.5. FERMION SYSTEMS 301
z → z e−iωt (A.60)
Isto é importante para descrever a matriz S, em que os estados assimptóticos evoluem de
acordo com o hamiltoniano livre.
A.4.4 Exact results II: external force

Seja o hamiltoniano
H = ωa† a − f (t)a† − f (t)a (A.61)

Este hamiltoniano também conduz a um resultado exacto. Usando os mesmos métodos
que foram utilizados para o oscilador harmónico pode-se mostrar que neste caso também
temos
U (z f , zi ; t) = eiS(f,i) (A.62)
onde S(f, i) é a acção calculada ao longo das trajectórias clássicas (ver Problema A.1).
A.5 Fermion systems

Vamos generalizar os resultados anteriores ao caso de sistemas de fermiões. Começamos
com sistemas com dois nı́veis com os operadores a† e a tais que
n o
a† , a = 1 ; a2 = a2† = 0 (A.63)
Para efectuar a construção anterior vamos tentar representar estes operadores num espaço
de Hilbert de funções analı́ticas. Isto é possı́vel se considerarmos funções (de facto polinó-
mios) com coeficientes complexos em duas variáveis que anticomutam η e η, designadas
por variáveis de Grassmann e que obedecem a
ηη + ηη = 0 ; η2 = η2 = 0 (A.64)
Então qualquer função P (η, η) terá a forma
∼
P (η, η) = p0 + p1 η + p 1 η + p12 ηη (A.65)
A.5.1 Derivatives
Neste espaço a derivação é definida por (as derivadas são esquerdas)
∂P ∼
= p 1 + p12 η
∂η
∂P
= p1 − p12 η (A.66)
∂η
De entre todas as funções nas variáveis η e η definimos o subconjunto das funções analı́ticas
tais que
∂
f =0 (A.67)
∂η
isto é as funções analı́ticas têm a forma
f = f0 + f1 η (A.68)
A.5.2 Dot product

No espaço das funções analı́ticas define-se o produto interno
(g, f ) = g 0 f0 + g 1 f1 (A.69)
este produto interno pode ser representado por um integral desde que definamos a inte-
gração convenientemente (ver equação (A.74)) .
A.5.3 Integration
A integração nas variáveis de Grassmann é definida pelas relações
Z Z
dη η = dη η = 1
Z Z
dη 1 = dη 1 = 0 (A.70)
Notar que a integração assim definida é semelhante à derivação. De facto2
Z Z
dη P = ∂P ; dη P = ∂P
Z
dηdη P = ∂∂P (A.72)
Devido à forma da equação A.62 é claro que se tem

2
∂ = ∂2 = 0 (A.73)
e que portanto o integral duma derivada é zero. Consideremos agora a mudança de
variáveis nos integrais. Seja

η ξ
=A (A.74)
η ξ
Então obtemos
2
Estamos a usar a notação compacta
∂
∂≡ ∂η ; ∂
∂≡ ∂η (A.71)
ηη = (A11 ξ + A12 ξ)(A21 ξ + A22 ξ)
= (A11 A22 − A12 A21 )ξξ
= det A ξξ (A.75)
Pelo que
Z Z
dηdηP (η, η) = dξdξ(det A)−1 Q(ξ, ξ) (A.76)
onde Q(ξ, ξ) é o polinómio que se obtém de P (η, η) por substituição de η e η por ξ e ξ.

Finalmente notemos que se definirmos a conjugação complexa de f por
f = f 0 + f 1η (A.77)
então podemos encontrar uma representação integral para o produto interno dada por
Z
(g, f ) ≡ dηdη e−ηη g f (A.78)
Para vermos isso calculemos o integral. Obtemos
Z
dηdη e−ηη g f
Z
= dηdη(1 − ηη)(g 0 + g 1 η)(f0 + f1 η)
= g 0 f0 + g 1 f1
= (g, f ) (A.79)
A.5.4 Representation of operators

Os operadores a e a† podem ser representados por
a→∂
a† → η (A.80)
É fácil de ver que com estas definições temos a2 = a†2 = 0 e aa† + a† a = 1.

Consideremos agora os estados |0i e |1i = a† |0i a que correspondem as funções 1 e η.
Então podemos encontrar o kernel de qualquer operador
X
A= |ni An,m hm| (A.81)
n,m
De facto
X
(Af )η = η n An,m hm|f i
n,m
Z X
= dξdξ eξξ η n An,m ξ m f (ξ)
n,m
Z
= dξdξ e−ξξ A(η, ξ) f (ξ) (A.82)
onde
X
A(η, ξ) ≡ η n An,m ξ m ; n, m = 0, 1 (A.83)
n,m
Para o produto de operadores é fácil de ver que temos como anteriormente

Z
A1 A2 (η, η) = dξdξ e−ξξ A1 (η, ξ) A2 (ξ, η) (A.84)
A.5.5 Normal form for operators

Seja um operador definido por
X X
A= |ni An,m hm| = a†n |0i h0| am An,m (A.85)
n,m n,m
O projector do estado base é

†a
|0i h0| =: e−a : (A.86)
logo
X †a
A = An,m : a†n e−a am :
n,m
X
≡ AN †n m
n,m a a (A.87)
n,m
O kernel normal é então definido pela substituição a† → η e a → η, isto é

X
AN (η, η) = AN n m
n,m η η (A.88)
n,m
O kernel da identidade é eηη , isto é

Z
f (η) = dξdξ e−ξξ+ηξ f (ξ) (A.89)
o que permite obter a relação entre o kernel usual e kernel normal. De facto
h i ∂m
a†n am f η = η n f (η)
∂η m
Z
= dξdξ e−ξξ eηξ η n ξ m f (ξ) (A.90)
o que permite escrever a relação procurada
A(η, η) = eηη AN (η, η) (A.91)

Finalmente seguindo um raciocı́nio análogo ao do sistema de bosões é fácil obter o kernel
do operador de evolução
Z R tf 1
U (η f , tf ; ηi , ti ) = D(η, η) e 2 (ηf ηf +ηi ηi ) ei ti dt[ 2i (ηη̇−η̇)−h(η,η)]
1
(A.92)
Problemas do Apêndice A
Problems for Appendix A
A.1 Mostre o resultado da equação A.59. Em particular mostre que
Z tf h i
−iω(tf −ti )
iS(f, i) = z f e zi + i dt z f e−iω(tf −t) f (t) + f (t) e−iω(t−ti ) zi
ti
Z tf Z tf
′
− dt dt′ f (t) e−iω(t−t ) f (t′ )θ(t − t′ ) (A.93)
ti ti
Este resultado é útil em muitas aplicações (ver equação B.9).

zn zm
A.2 Mostre que os representantes dos estados |ni e |mi, √ e √ , respectivamente, são
n! m!
ortonormados, isto é, hfn |fm i = δn,m .
A.3 Mostrar que para integrais gaussianos se tem

Z Y
dzk dz k −1 u
e−zAz+uz+zu = (det A)−1 euA (A.94)
2πi
Notar que o expoente é o valor no ponto de estacionaridade.
A.4 Mostrar que para integrais gaussianos se obtém
Z Y
n P P
ηk Akℓ ηℓ + (η k ξk +ξk ηk )
dη k dηk e
1
ξk (A−1 )kℓ ξℓ
P
= det A e (A.95)
Comparar com o resultado do problema A.3.

Appendix B
Path Integral in Quantum Field

Theory
B.1 Path integral quantization

Vamos aqui generalizar os resultados do apêndice A para o caso de sistemas com um
número infinito de graus de liberdade que são os que interessam em teoria quântica dos
campos. Para evitar complicações com ı́ndices e com problemas decorrentes da invariância
de gauge vamos estudar o caso do campo escalar cuja acção clássica em presença duma
fonte exterior é
Z
S(φ, J) = S0 (φ, J) + d4 x V (x) (B.1)
onde
Z
4 1 µ 1 2 2
S0 (φ, J) = d x ∂µ φ∂ φ − m φ + Jφ (B.2)
2 2
é a acção do campo escalar livre acoplada a uma fonte exterior. Vamos primeiro estudar
este caso, isto é, supor que V = 0. O caso geral é fácil, de obter a partir deste, como
veremos mais à frente. O hamiltoniano é dado por
Z
1 2 1 1
H= 3
d x πop + (∇φop )2 + m2 φ2op − Jφop (B.3)
2 2 2
e podemos introduzir os operadores a(k) e a† (k) tais que num certo instante
Z h i
φop = ˜ a(k) ei~k·~x + a† (k) e−i~k·~x
dk (B.4)
e
Z h i
πop = −i ˜ a(k) ei~k·~x − a† (k) e−i~k·~x ω(k)
dk (B.5)
então
307
308 APPENDIX B. PATH INTEGRAL IN QUANTUM FIELD THEORY
Z h i
H= ˜ ω(k)a† (k)a(k) − f (t, ~k)a† (k) − f (t, ~k)a(k)
dk (B.6)
onde introduzimos a transformada de fourier espacial da fonte,

Z
~ ~
f (t, k) = d3 x e−ik·~x j(t, ~x) (B.7)
e onde definimos
˜ ≡ d3 k d4 k
dk = 2πδ(k2 − m2 )θ(k0 ) (B.8)
(2π)3 2ωk (2π)4
usando os resultados do problema A.1 podemos escrever imediatamente o kernel do oper-
ador de evolução
Z
U (z f , tf ; zi , ti ) = exp ˜
dk z f (k) e−iω(k)(tf −ti ) zi (k)
Z tf h i
+i dt z f (k) e −iω(k)(tf −t) ~ ~
f (t, k) + f (t, k) e−iω(k)(t−ti )
zi (k)
ti
Z tf Z tf
1 ~
′ −iω(k)(t−t′ ) ′ ~
− dt dt f (t, k) e f (t , k) (B.9)
2 ti ti
A matriz S é então definida como o limite
lim eitf H0 U (tf , ti ) e−iti H0 (B.10)

−ti ,tf →∞
onde H0 é obtido a partir de H fazendo J = 0. Na representação que estamos a usar a

acção de e−itH0 é uma simples multiplicação (ver eq. A.60).
z → z e−iωt (B.11)
Portanto o kernel da matriz S é
Z Z
S(z f , zi ) = lim exp ˜
dkz f (k)zi (k) exp ˜
dk
−ti ,tf →∞
Z tf
i z f (k) eiω(k)t f (t, ~k) + f (t, ~k) e−iω(k)t zi (k)
ti
Z tf Z tf
1 ′ ~ −iω(k)(t−t′ ) ′ ~
− dtdt f (t, k) e f (t , k) (B.12)
2 ti ti
O primeiro factor é aquilo que é necessário para passar do kernel usual para o kernel
normal. O restante pode ser interpretado se definirmos
Z h i
φas ≡ dk ˜ zi (k) e−ik·x + z f (k) eik·x (B.13)
B.1. PATH INTEGRAL QUANTIZATION 309
Como z f não é o complexo conjugado de zi então φas é dado em termos de condições na

fronteira com frequências positivas para t → −∞ e frequências negativas para t → ∞.
Estas são precisamente as condições na fronteira de Feynman. Com estas convenções e
notações obtemos para o primeiro termo
Z Z tf h i
˜
dk dt z f (k) eiω(k)t f (t, ~k) + f (t, ~k) e−iω(k)t zi (k)
ti
Z Z h i
˜ ~ ~
= d4 x dkJ(x) z f eiω(k)t−ik·~x + zi (k) e−iω(k)t+ik·~x
Z
= d4 xJ(x)φas (x) (B.14)
e para o segundo
Z Z Z
′
˜
dk dt dt′ f (t, ~k) e−iω(k)(t−t ) f (t′ , ~k)
Z Z
= 4 4 ′
d xd x J(x)J(x ) ′ ˜ e−iω(k)(t−t′ )+i~k·(~x−~x′ )
dk
Z
= d4 xd4 x′ J(x)GF (x − x′ )J(x′ ) (B.15)
pois
Z
˜ e−iω(k)(t−t′ )+i~k·(~x−~x′ )
dk
Z
= ˜ e−iω(k)(t−t′ )+i~k·(~x−~x′ ) θ(t − t′ )
dk
Z
+ ˜ eiω(k)(t−t′ )+i~k·(~x−~x′ ) θ(t′ − t)
dk
Z
′ 1
=i d4 k e−ik·(x−x )
k2 − m2 + iε
= GF (x − x′ ) (B.16)
Notar que as condições na fronteira mistas conduzem ao propagador de Feynman. Podemos

portanto finalmente escrever o kernel normal da matriz S na presença da fonte J,
R 4 1
R
d4 xd4 x′ J(x)G0F (x−x′ )J(x′ )
S N (z f , zi )J = ei d x j(x) φas (x)
e− 2 (B.17)
Para se obter o operador S substituimos φas por φop e fazemos o ordenamento normal,
isto é
1
d4 x J(x)φop (x) d4 xd4 x′ J(x)G0F (x−x′ )J(x′ )
R R
S0 (J) =: ei : e− 2 (B.18)
Como o funcional gerador das funções de green é h0|S0 (J)|0i obtemos imediatamente
1
d4 xd4 x′ J(x)G0F (x−x′ )J(x′ )
R
Z0 (J) = e− 2 (B.19)
Este resultado permite resolver o problema de qualquer potencial V (x). De facto é fácil
de mostrar que no caso geral os kernéis estão relacionadoss por
Z
δ
S N
= exp −i 4
d xV S N (J)J=0 (B.20)
iδJ(x)
e o operador S é
Z
R
d4 xJ(x)φop (x) δ
S =: e : exp −i 4
d xV Z0 (J)J=0 (B.21)
iδJ(x)
ou seja
Z
4 δ
Z(J) = exp −i d xV Z0 (J) (B.22)
iδJ(x)
com
1
d4 xd4 x′ J(x)G0F (x−x′ )J(x′ )
R
Z0 (J) = e− 2 (B.23)
Estas expressões permitem calcular qualquer função de green com as regras usuais da
teoria das perturbações. A quantificação usando os integrais de caminho conduziu aos mes-
mos resultados (em teoria das perturbações) que a quantificação canónica. As expressões
para os funcionais geradores embora dêem resultados perturbativos duma forma imediata
não são as mais úteis quando estamos interessados em encontrar resultados válidos para
além da teoria das perturbações. Para esses casos (identidades de Ward, etc) é mais útil
ter uma expressão formal em termos dum integral de caminho. É isso que vamos agora
estudar.
B.2 Path integral for generating functionals

O ponto de partida é a expressão para o kernel da matriz S,
Z
1 ˜
R
dk[z(k,tf )z(k,tf )+z(k,ti )z(k,ti )]
S(z f , zi ) = lim D(z, z) e 2
−ti ,tf →∞
Z tf Z
˜ 1
exp i dt dk (ż(k, t)z(k, t) − z(k, t)ż(k, t))
ti 2i

−ω(k)z(k, t)z(k, t) − V (z, z) (B.24)
com as condições na fronteira1

1
Não há restrições em z(k,ti ) e z(k,tf ).
B.2. PATH INTEGRAL FOR GENERATING FUNCTIONALS 311

 z(k, tf ) = z f (k) eiωtf
(B.25)

z(k, ti ) = zi (k) e−iωti
Em vez das variáveis z(k, t) e z(k, t) vamos introduzir os campos clássicos φ(~x, t) e π(~x, t)
definidos por
Z h i
φ(~x, t) = dk ˜ z(k, t) ei~k·~x + z(k, t) e−i~k·~x (B.26)
e
Z h i
π(~x, t) = −i ˜ ω(k) z(k, t) ei~k·~x − z(k, t) e−i~k·~x
dk (B.27)
Estas fórmulas são obviamente sugeridas pelas relações entre φop , πop e a(k), a† (k) expres-
sas nas equações B.4 e B.5, só que aqui não se trata de operadores mas sim de campos
clássicos. Comecemos por escrever a acção em termos das novas variáveis,
Z tf Z
˜ 1
dt dk (ż(k, t)z(k, t) − z(k, t)ż(k, t))
ti 2i
−ω(k)z(k, t)z(k, t) − V (z, z)]

Z Z tf
1 1
= d3 x dt (π∂0 φ − ∂0 πφ) − π 2
ti 2 2

1 2 1 2 2
− (∂k φ) − m φ − V (φ) (B.28)
2 2
Introduzimos agora novas variáveis φ1 (~x, t) e π1 (~x, t) definidas do modo seguinte


 φ(~x, t) ≡ φas (~x, t) + φ1 (~x, t)
(B.29)

π(~x, t) ≡ ∂0 φ(~x, t) + φ1 (~x, t)
onde
Z h i
φas,in (~x, t) = ˜ z in (k) eik·x + zin (k) e−ik·x
dk (B.30)
e
Z h i
φas,out (~x, t) = ˜ z out (k) eik·x + zout (k) e−ik·x
dk (B.31)
onde in ≡ t → −∞ e out ≡ t → +∞ com


 z out (k) ≡ z f (k)
(B.32)

zin (k) ≡ zi (k)
O campo φas tem portanto as condições fronteira apropriadas para o problema e satisfaz
à equação de Klein-Gordon
⊓ + m2 )φas = 0
(⊔ (B.33)
Escrevemos a acção nas novas variáveis
Z Z tf
3 1 1 2 1 2 1 2 2
d x dt (π∂0 φ − ∂0 πφ) − π − (∂k φ) − m φ − V (φ)
ti 2 2 2 2
Z tf
1
= d3 x − πφ
2 ti
Z Z tf
1 1 1
+ d3 x dt π∂0 φ − (π12 + 2π∂0 φ − (∂0 φ)2 ) − (∂k φas )2 − (∂k φ1 )2
ti 2 2 2

1 1
−∂k φas ∂k φ1 − m2 φ2as − m2 φ21 − m2 φas φ1 − V (φ)
2 2
Z tf
1
= d3 x − πφ
2 ti
Z Z tf
1 1 1 1
+ d3 x dt − π12 + (∂0 φas )2 + ∂0 φas ∂0 φ1 + (∂0 φ1 )2 − (∂k φas )2
ti 2 2 2 2

1 2 1 2 2 1 2 2 2
− (∂k φ1 ) − ∂k φas ∂k φ1 − m φas − m φ1 − m φas φ1 − V (φ)
2 2 2
Z tf
1 1
= d3 x ∂0 φas φas + ∂0 φas φ1 − πφ
2 2 ti
Z Z tf
3 1 2 1 µ 1 2 2
+ d x dt − π1 + ∂µ φ1 ∂ φ1 − m φ1 − V (φ)
ti 2 2 2
Z tf
1 1
= d3 x ∂0 φas φ − ∂0 φas φas − πφ
2 2 ti
Z Z tf
1 1 1
+ d3 x dt − π12 + ∂µ φ1 ∂ µ φ1 − m2 φ21 − V (φ) (B.34)
ti 2 2 2
Vemos que no segundo termo as variáveis φ1 e π1 estão separadas e π1 aparece quadrati-
camente. Isto permitirá eliminar π1 como veremos no seguimento. Analisemos contudo
primeiro o termo que tem as condições na fronteira. Usando as definições de φas , φ e π
podemos escrever
Z tf
3 1 1
i d x ∂0 φas φ − ∂0 φas φas − πφ
2 2 ti
Z

= dk˜ z f (k)zi (k) − 1 z(k, tf )z(k, tf ) + z(k, ti )z(k, ti )
2

1 2 1 2
− z(k, tf ) − zi (k) e−iωtf − z(k, ti ) − z f (k) eiωti (B.35)
4 4
Nesta expressão o primeiro termo dá a passagem do kernel usual para o kernel normal,
o segundo cancela exactamente o termo na fronteira na definição inicial de S(z f , zi ) e os
últimos têm que ser estudados em detalhe. Reunindo tudo até este ponto a expressão do
kernel normal da matriz S é
Z Z h
1 2
N
S (φas ) = lim D(φ, π) exp − ˜
dk z(k, tf ) − zi (k) e−iωtf
−ti ,tf →∞ 4
2 io
+ z(k, ti ) − zf (k) eiωti
Z Z tf
3 1 1 2 2
exp d x dt −π12 µ
+ ∂µ φ1 ∂ φ1 − m φ1 − V (φ) (B.36)
ti 2 2
Esta expressão já está próxima do resultado final. Falta só mostrar que os termos dentro
da primeira exponencial tendem para zero quando −ti , tf → ∞. Esta é a parte mais
delicada do argumento. Vamos expô-lo por passos:
i) Funções rapidamente decrescentes
R
Queremos que I(t) = d3 x π12 (~x, t) seja integrável. Dizemos então que funções como
π1 (~x, t) são rapidamente decrescentes (RD) quando |t| → ∞.
ii) Informação sobre z 1,out (k, t) e z1,in (k, t)
Da definição φ = φas + φ1 resultam as definições

 z(k, t) = zi (k) e−iωt + z1 (k, t)
(B.37)
 iωt
z(k, t) = z f (k) e + z 1 (k, t)
As condições na fronteira dizem-nos que z 1,out (k, t) e z1,in (k, t) são funções RD quando
t → +∞ e t → −∞ respectivamente, mas não nos dizem nada sobre z 1,in e z1,out , que são
precisamente os limites que precisamos.
iii) Informação sobre os limites z1,out e z 1,in
Informação sobre os limites z1,out e z 1,in obtém-se a partir do seguinte raciocı́nio,
π1 = π − ∂0 φ
Z n
= ˜ [iω(k)z(k, t) − ∂0 z(k, t)] e−i~k·~x
dk
o
~
− [iω(k)z(k, t) + ∂0 z(k, t)] eik·~x
Z h i
≡ − ˜ z 2 (k, t) e−i~k·~x + z2 (k, t) ei~k·~x
dk (B.38)
Para que π1 seja do tipo RD quando |t| → ∞ também teremos que ter z2 (k, t) e z 2 (k, t)
RD nesses limites. Vejamos qual a informação contida neste resultado.
• t → +∞
Obtemos então que a função
z2 (k, t) ≡ ∂0 z(k, t) + iω(k)z(k, t) (B.39)

é RD quando t → +∞. A informação sobre z 2 (k, t) não trás nada de novo já que está
contida nas condições fronteiras. De facto
lim z 2 (k, t) ≡ lim [∂0 z(k, t) − iω(k)z(k, t)]

t→+∞ t→+∞
= iω(k)z f (k) eiωt + ∂0 z 1,out (k, t)
−iω(k)z f (k) eiωt − iω(k)z 1,out (k, t)
= RD t → +∞ (B.40)
• t → −∞
A informação contida nas condições na fronteira é
z 2 (k, t) ≡ ∂0 z(k, t) − iω(k)z(k, t) = RD t → −∞ (B.41)

iv) Demonstração que z1,out e z 1,in são RD
Da definição
φ(~x, t) = φas + φ1 (B.42)

resulta


φ(~x, t) = φas,in (~x, t) + φ1,in (~x, t) t → −∞
(B.43)


φ(~x, t) = φas,out (~x, t) + φ1,out (~x, t) t → +∞
ou seja

 iωt t → −∞
z(k, t) = z in (k) e + z 1,in (k, t)
(B.44)


z(k, t) = zout (k) e−iωt + z1,out (k, t) t → +∞
Mas usando os resultados anteriores
∂0 z(k, t) − iω(k)z(k, t) = RD t → −∞
= iω(k)z in (k) eiωt + ∂0 z 1,in (k, t)
−iω(k)z in (k) eiωt − iω(k)z 1,in (k, t) (B.45)
ou seja
z 1,in (k, t) = RD t → −∞ (B.46)

e igualmente
z1,out (k, t) = RD t → +∞ (B.47)

Isto quer dizer que


φ1,in = RD t → −∞
(B.48)


φ1,out = RD t → +∞
isto é, assimptoticamente
φ = φas + RD (B.49)
v) Resultado final
Estamos agora em condições de atacar o nosso problema. Temos
Z h i2
lim ˜ z(k, ti ) − zf (k) eiωti
dk
ti →−∞
Z

= lim ˜ (z in (k) − z f (k)) eiωti + z 1,in (k, ti ) 2
dk
ti →−∞
Z h
= lim ˜ (z in (k) − z f (k))2 e2iωti + 2 (z in (k) − z f (k)) eiωti z 1,in (k, ti )
dk
ti →−∞
i
+z 21,in (k, ti )
= 0 (B.50)
Para o outro termo obter-se-ia o mesmo resultado. Chegamos portanto ao resultado
Z Z
N i 4 2
S (φas ) = D(φ, π) exp − d x π1
2
Z
4 1 µ 1 2 2
× exp i d x ∂µ φ1 ∂ φ1 − m φ1 − V (φ) (B.51)
2 2
onde a integração é feita sobre os campos φ = φas + φ1 com as condições fronteiras
apropriadas. Fazendo a integração sobre π1 obtemos ( a menos duma normalização)
Z
d4 x[ 12 ∂µ φ1 ∂ µ φ1 − 21 m2 φ21 −V (φ)]
R
S N (φas ) = D(φ) ei
Z
d4 x[L(φ1 )−(V (φ)−V (φ1 ))]
R
= D(φ) ei (B.52)
φ=φas +φ1
Na presença de fontes exteriores obtemos

Z R 4
S N (φas , J) = D(φ) ei d x[L(φ1 )−(V (φ)−V (φ1 ))+Jφ] (B.53)
φ=φas +φ1
Normalmente não estamos interessados na matriz S mas no funcional gerador das

funções de Green. Por definição

Z(J) ≡ S(φas , J) (B.54)
φas =0
Obtemos portanto a expressão fundamental

Z R 4
Z(J) = D(φ) ei d x[L(φ)+Jφ] (B.55)
B.3 Fermion systems

O uso de variáveis de Grassmann permite escrever expressões de integrais de caminho para
a matriz S e para o funcional gerador das funções de Green Z para este caso. Não vamos
aqui repetir os cálculos que fizémos para os sistemas de bosões, mas antes apresentar
somente os resultados deixando as demonstrações para os problemas.
O ponto de partida é a definição do funcional gerador das funções de Green em presença
das fontes exteriores fermiónicas. Este é dado por2
Z
4

Z[η, η] = h0| T exp i d x η(x)ψ(x) + ψ(x)η(x) |0i (B.56)
Então as funções de Green
G2n (x1 , . . . , yn ) ≡ h0| T ψ(x1 ) · · · ψ(xn )ψ(y1 ) · · · ψ(yn ) |0i (B.57)

são dadas por
δ2n Z
G2n (x1 , . . . , yn ) = (B.58)
onde as derivadas são esquerdas, isto é
Z
δ
d4 yη(y)ψ(y) = ψ(x)
δη(x)
Z
δ
d4 yψ(y)η(y) = −ψ(x) (B.59)
δη(x)
e por convenção a ordem da derivação é a indicada, isto é
δ δ
··· (B.60)
iδη(yn ) iδη(x1 )
Consideramos agora o lagrangeano de Dirac livre
2
Comparar com a definição do caso bosónico, equação 5.15.
B.3. FERMION SYSTEMS 317
L = ψ(i∂/ − m)ψ (B.61)

Pode-se mostrar (ver problema B.2) que o funcional gerador é neste caso dado por
d4 xd4 y η(x)SF
0 (x−y)η(y)
R
Z0 [η, η] = e− (B.62)
onde SF0 (x − y) é o propagador de Feynman para a teoria de Dirac livre, dado por
Z
0 d4 p −ip·(x−y) i
SF (x − y) = 4
e (B.63)
(2π) p/ − m + iε
Seguindo métodos semelhantes ao do caso bosónico podemos também mostrar que este
funcional gerador pode ser representado pelo integral de caminho,
Z
Z0 [η, η] = D(ψ, ψ) ei d x [L(x)+ηψ+ψη]
R 4
(B.64)
Tendo o funcional gerador para a teoria livre podemos formalmente escrever o funcional
gerador para qualquer teoria fermiónica com interacções. Um exemplo é dado no Problema
B.4.
Problems Appendix B
B.1 Mostre que as funções de Green
G2n (x1 , . . . , yn ) ≡ h0| T ψ(x1 ) · · · ψ(xn )ψ(y1 ) · · · ψ(yn ) |0i (B.65)
são dadas por
δ2n Z
G2n (x1 , . . . , yn ) = (B.66)
B.2 Mostre que o funcional gerador das funções de Green para a teoria de Dirac livre é
dado por
0 (x−y)η(y)
d4 xd4 y η(x)SF
R
Z0 [η, η] = e− (B.67)
B.3 Mostre que o funcional gerador das funções de Green para a teoria de Dirac livre se
pode representar pelo seguinte integral de caminho
Z
d4 x [L(x)+ηψ+ψη]
R
Z0 [η, η] = D(ψ, ψ) ei (B.68)
B.4 Considere o lagrangiano seguinte,
1 1
L(x) = ψ(i∂/ − m)ψ + ∂µ φ∂ µ φ − mφ2
2 2
−gψ(x)ψ(x)φ(x) (B.69)
que descreve a interacção dum campo de Dirac com um campo escalar.

Problems 319
a) Mostre que
Z
4 δ δ δ
Z[η, η, J] = exp −ig d x Z0 [η, η, J] (B.70)
iδη iδη iδJ
onde
0 (x−y)η(y)+ 1 J(x)∆ (x−y)J(y)

d4 xd4 y [η(x)SF ]
R
Z0 [η, η, J] = e− 2 F
(B.71)
e ∆F é o propagador livre do campo escalar.

b) Mostre que Z[η, η, J] se pode exprimir por meio do integral de caminho
Z
d4 x[L(x)+Jφ+ηψ+ψη]
R
Z[η, η, J] = D(ψ, ψ, φ) ei (B.72)
Appendix C
Useful techniques for

renormalization
C.1 µ parameter
The reason for the µ parameter introduced in section 4.1.1 is the following. In dimension
d = 4 − ǫ, the fields Aµ and ψ have dimensions given by the kinetic terms in the action,
Z
d 1 2
d x − (∂µ Aν − ∂ν Aµ ) + ψγ · ∂ψ (C.1)
4
We have therefore
0 = −d + 2 + 2[Aµ ] ⇒ [Aµ ] = 12 (d − 2) = 1 − ǫ
2
(C.2)
0 = −d + 1 + 2[ψ] ⇒ [ψ] = 12 (d − 1) = 3
2 − ǫ
2
Using these dimensions in the interaction term

Z
SI = dd x eψγµ ψAµ (C.3)
we get
[SI ] = −d + [e] + 2[ψ] + [A]

ǫ
= −4 + ǫ + [e] + 3 − ǫ + 1 −
2
ǫ
= [e] − (C.4)
2
Therefore, if we want the action to be dimensionless (remember that we use the system
where h̄ = c = 1), we have to set
ǫ
[e] = (C.5)
2
321
322 APPENDIX C. USEFUL TECHNIQUES FOR RENORMALIZATION
We see then that in dimensions d 6= 4 the coupling constant has dimensions. As it is more
convenient to work with a dimensionless coupling constant we introduce a parameter µ
with dimensions of a mass and in d 6= 4 we will make the substitution
ǫ
e → eµ 2 (ǫ = 4 − d) (C.6)
while keeping e dimensionless.
C.2 Feynman parametrization

The most general form for a 1–loop is 1
Z
µ1 ···µp dd k kµ1 · · · kµp
T̂n ≡ (C.7)
(2π)d D0 D1 · · · Dn−1
where
Di = (k + ri )2 − m2i + iǫ (C.8)
and the momenta ri are related with the external momenta (all taken to be incoming)
through the relations,
j
X
rj = pi ; j = 1, . . . , n − 1
i=1
Xn
r0 = pi = 0 (C.9)
i=1
as indicated in Fig. (C.1). In these expressions there appear in the denominators products
pn
p1 k
p2 k+r1 pn-1
p3 k+r3
pi
Figure C.1:
of the denominators of the propagators of the particles in the loop. It is convenient to

combine these products in just one common denominator. This is achieved by a technique
due to Feynman. Let us exemplify with two denominators.
Z 1
1 dx
= 2 (C.10)
ab 0 [ax + b(1 − x)]
1
We introduce here the notation T̂ to distinguish from a more standard notation that will be explained
in subsection C.9.
C.2. FEYNMAN PARAMETRIZATION 323
The proof is trivial. In fact

Z
1 x
dx 2 = b [(a − b)x + b] (C.11)
[ax + b(1 − x)]
and therefore Eq. (C.10) immediately follows. Taking successive derivatives with respect
to a and b we get Z
1 Γ(p + q) 1 xp−1 (1 − x)q−1
= dx (C.12)
a p bq Γ(p)Γ(q) 0 [ax + b(1 − x)]p+q
and using induction we obtain a general formula
Z 1 Z 1−x1
1
= Γ(n) dx1 dx2 · · ·
a1 a2 · · · an 0 0
Z 1−x1 −···−xn−2
dxn−1
(C.13)
0 [a1 x1 + a2 x2 + · · · + an (1 − x1 − · · · − xn−1 )]n
Before closing the section let us give an example that will be useful in the self-energy case.
Consider the situation with the kinematics described in Fig. (C.2).
p p
p+k
Figure C.2:
We get
Z
dd k 1
I =
d
(2π) (k + p) − m1 + iǫ k2 − m22 + iǫ
2 2
Z 1 Z
dd k 1
= dx
0 (2π) k + 2p · k x + p x − m2 x − m2 (1 − x) + iǫ 2
d 2 2
1 2
Z 1 Z
dd k 1
= dx
0 (2π) [k + 2P · k − M 2 + iǫ]2
d 2
Z 1 Z
dd k 1
= dx (C.14)
0 (2π) d
[(k + P ) − P 2 − M 2 + iǫ]2
2
where in the last line we have completed the square in the term with the loop momenta
k. The quantities P and M 2 are, in this case, defined by
P = xp (C.15)
and
M 2 = −x p2 + m21 x + m22 (1 − x) (C.16)
They depend on the masses, external momenta and Feynman parameters, but not in the
loop momenta. Now changing variables k → k − P we get rid of the linear terms in k and
finally obtain Z 1 Z
dd k 1
I= dx (C.17)
0 (2π) [k2 − C + iǫ]2
d
where C is independent of the loop momenta k and it is given by
C = P2 + M2 (C.18)
Notice that the iǫ factors will add correctly and can all be put as in Eq. (C.17).
C.3 Wick Rotation

From the example of the last section we can conclude that all the scalar integrals can be
reduced to the form Z r
dd k k2
Ir,m = (C.19)
(2π)d [k2 − C + iǫ]m
It is also easy to realize that also all the tensor integrals can be obtained from the scalar
integrals. For instance
Z
dd k kµ
=0
(2π)d [k2 − C + iǫ]m
Z Z
dd k kµ kν 1 µν dd k k2
= g (C.20)
(2π)d [k2 − C + iǫ]m d (2π)d [k2 − C + iǫ]m
and so on. Therefore the integrals Ir,m are the important quantities to evaluate. We will
consider that C > 0. The case C < 0 can be done by analytical continuation of the final
formula for C > 0.
To evaluate the integral Ir,m we will use integration in the complex plane of the variable
k0 as described in Fig. C.3. We can then write
Z d−1 Z r
d k 0 h k2
Ir,m = dk im (C.21)
(2π)d k02 − |~k|2 − C + iǫ
The function under the integral has poles for

q
0
k =± |~k| + C − iǫ
2 (C.22)
as shown in Fig. C.3. Using the properties of functions of complex variables (Cauchy the-
orem) we can deform the contour, changing the integration from the real to the imaginary
axis plus the two arcs at infinity. This can be done because in deforming the contour we
do not cross any pole. Notice the importance of the iǫ prescription to be able to do this.
The contribution from the arcs at infinity vanishes in dimension sufficiently low for the
C.4. SCALAR INTEGRALS IN DIMENSIONAL REGULARIZATION 325
Im k0
x Re k0
Figure C.3:
integral to converge, as we assume in dimensional regularization. We have then changed

the integration along the real axis into an integration along the imaginary axis in the plane
of the complex variable k0 . If we write
Z +∞ Z +∞
k0 = ikE
0
com dk 0 → i dkE0
(C.23)
−∞ −∞
and k2 = (k0 )2 − |~k|2 = −(kE

0 )2 − |~
k|2 ≡ −kE 2 , where k 0 ~
E = (kE , k) is an euclidean
vector. By this we mean that we calculate the scalar product using the euclidean metric
diag(+, +, +, +),
2
kE = (kE ) + |~k|2
0 2
(C.24)
We can them write Z r
2
r−m dd kE kE
Ir,m = i(−1)
2 +C m
(C.25)
(2π)d kE
where we do not need the iǫ because the denominator is strictly positive (C > 0). This
procedure is known as Wick Rotation. We note that the Feynman prescription for the
propagators that originated the iǫ rule for the denominators is crucial for the Wick rotation
to be possible.
C.4 Scalar integrals in dimensional regularization

We have seen in the last section that the scalar integrals to be calculated with dimensional
regularization had the general form of Eq. (C.25). We are now going to find a general
formula for Ir,m . We begin by writing
Z Z
d d−1
d kE = dk k dΩd−1 (C.26)
q
where k = (kE 0 )2 + |~
k|2 is the length of the vector kE in the euclidean space in d dimen-
sions and dΩd−1 is the solid angle that generalizes spherical coordinates in that euclidean
space. The angles are defined by
kE = k(cos θ1 , sin θ1 cos θ2 , sin θ1 sin θ2 , sin θ1 sin θ2 cos θ3 , . . . , sin θ1 · · · sin θd−1 ) (C.27)
We can then write Z Z Z

π 2π
dΩd−1 = sin θ1d−2 dθ1 · · · dθd−1 (C.28)
0 0
Using now
Z π √ Γ( m+1
2 )
sin θ m dθ = π m+2 (C.29)
0 Γ( 2 )
where Γ(z) is the gamma function (see section C.6) we get
Z d
π2
dΩd−1 =2 d (C.30)
Γ( 2 )
The integration in k is done using the result

Z ∞
xp Γ( p+1
n )
dx n = π(−1)q−1 ap+1−nq p+1 p+1 (C.31)
0 (x + an )q n sin(π n ) Γ( 2 − q + 1)
and we finally get
d (−1)r−m Γ(r + d2 ) Γ(m − r − d2 )

Ir,m = iC r−m+ 2 (C.32)
Γ( d2 )
d
(4π) 2 Γ(m)
Before ending the section we note that the integral representation for Ir,m , Eq. (C.19), is
valid only for d < 2(m − r) to ensure convergence when k → ∞. However the final form
in Eq. (C.32) can be analytically continued for all values of d except for those where the
function Γ(m − r − d/2) has poles, that is for (see section C.6),
d
m−r− 6= 0, −1, −2, . . . (C.33)
2
For the application in dimensional regularization it is convenient to rewrite Eq. (C.32)
using the relation d = 4 − ǫ. we get
ǫ
(−1)r−m 4π 2 Γ(2 + r − 2ǫ ) Γ(m − r − 2 + 2ǫ )
Ir,m =i C 2+r−m (C.34)
(4π)2 C Γ(2 − 2ǫ ) Γ(m)
C.5 Tensor integrals in dimensional regularization

We are frequently faced with the task of evaluating the tensor integrals of the form of
Eq. (C.7),
Z
µ ···µ dd k kµ1 · · · kµp
T̂n 1 p ≡ (C.35)
(2π)d D0 D1 · · · Dn−1
The first step is to reduce to one common denominator using the Feynman parameteriza-
tion technique. The result is,
Z 1 Z 1−x1 −···−xn−2 Z
µ ···µ dd k kµ1 · · · kµp
T̂n 1 p = Γ(n) dx1 · · · dxn−1
0 0 (2π)d [k2 + 2k · P − M 2 + iǫ]n
C.6. Γ FUNCTION AND USEFUL RELATIONS 327
Z 1 Z 1−x1 −···−xn−2
µ ···µp
= Γ(n) dx1 · · · dxn−1 In 1 (C.36)
0 0
where we have defined

Z
µ ···µp dd k kµ1 · · · kµp
In 1 ≡ (C.37)
(2π)d [k2 + 2k · P − M 2 + iǫ]n
that we call, from now on, the tensor integral. In principle all these integrals can be
written in terms of scalar integrals. It is however convenient to have a general formula for
them. This formula can be obtained by noticing that
∂ 1 2kµ
n = −n (C.38)
µ 2 2
∂P [k + 2k · P − M + iǫ] [k + 2k · P − M 2 + iǫ]n+1
2
Using the last equation one can write the final result
Z ∞
µ ···µ i (4π)ǫ/2 dt n−3+ǫ/2 ∂ ∂
In 1 p = 2 p
t ··· e−t C (C.39)
16π Γ(n) 0 (2t) ∂Pµ1 ∂Pµp
where C = P 2 + M 2 . After doing the derivatives the remaining integrals can be done
using the properties of the Γ function (see section C.6). Notice that P , M 2 and therefore
also C depend not only in the Feynman parameters but also in the exterior momenta.
The advantage of having a general formula is that it can be programmed [9] and all the
integrals can then be obtained automatically.
C.6 Γ function and useful relations

The Γ function is defined by the integral
Z ∞
Γ(z) = tz−1 e−t dt (C.40)
0
or equivalently
Z ∞
tz−1 e−µt dt = µ−z Γ(z) (C.41)
0
The function Γ(z) has the following important properties
Γ(z + 1) = zΓ(z)
Γ(n + 1) = n! (C.42)
Another related function is the logarithmic derivative of the Γ function, with the proper-
ties,
d
ψ(z) = ln Γ(z) (C.43)
dz
ψ(1) = −γ (C.44)
1
ψ(z + 1) = ψ(z) + (C.45)
z
where γ is the Euler constant. The function Γ(z) has poles for z = 0, −1, −2, · · · . Near
the pole z = −m we have
(−1)m 1 (−1)m
Γ(z) = + ψ(m + 1) + O(z + m) (C.46)
m! m + z m!
From this we conclude that when ǫ → 0
ǫ
2 (−1)n 2
Γ = + ψ(1) + O(ǫ) Γ(−n + ǫ) = + ψ(n + 1) + 1 (C.47)
2 ǫ n! ǫ
and
π2 2 ǫ2
Γ(1 + ǫ) = 1 − γǫ + γ + + ··· , ǫ→0 (C.48)
6 2!
Using these results we can expand our integrals in powers of ǫ and separate the divergent
and finite parts. For instance for the one of the integrals of the self-energy,
ǫ
i 4π 2 2 Γ(1 + 2ǫ )
I0,2 =
(4π)2 C ǫ

i 2
= − γ + ln 4π − ln C + O(ǫ)
16π 2 ǫ
i
= [∆ǫ − ln C + O(ǫ)] (C.49)
16π 2
where we have introduced the notation

2
∆ǫ = − γ + ln 4π (C.50)
ǫ
for a combination that will appear in all expressions.
C.7 Explicit formulæ for the 1–loop integrals

Although we have presented in the previous sections the general formulæ for all the in-
tegrals that appear in 1–loop, Eqs. (C.34) and (C.39), in pratice it is convenient to have
expressions for the most important cases with the expansion on the ǫ already done. The
results presented below were generated with the Mathematica package OneLoop [9] from
the general expressions. In these results the integration on the Feynman parameters has
still to be done. This is in general a difficult problem and we will present in section C.9 an
alternative way of expressing these integrals more convenient for a numerical evaluation.
C.7. EXPLICIT FORMULÆ FOR THE 1–LOOP INTEGRALS 329
C.7.1 Tadpole integrals

With the definitions of Eqs. (C.34) and (C.39) we get
i
I0,1 = C(1 + ∆ǫ − ln C)
16π 2
I1µ = 0
i 1 2 µν
I1µν = C g (3 + 2∆ǫ − 2 ln C) (C.51)
16π 2 8
where for the tadpole integrals
P =0 ; C = m2 (C.52)
because there are no Feynman parameters and there is only one mass. In this case the
above results are final.
C.7.2 Self–Energy integrals

For the integrals with two denominators we get,
i
I0,2 = (∆ǫ − ln C)
16π 2
i
I2µ = (−∆ǫ + ln C)P µ
16π 2

i 1
I2µν = µν
Cg (1 + ∆ǫ − ln C) + 2(∆ǫ − ln C)P P µ ν
16π 2 2

i 1
I2µνα = − Cg µν (1 + ∆ǫ − ln C)P α − Cg να (1 + ∆ǫ − ln C)P µ
16π 2 2

µα ν α µ α µ ν
− Cg (1 + ∆ǫ − ln C)P − (2∆ǫ P P − 2 ln CP P )P (C.53)
where, with the notation and conventions of Fig. (C.1), we have
P µ = x r1µ ; C = x2 r12 + (1 − x) m22 + x m21 − x r12 (C.54)
C.7.3 Triangle integrals

For the integrals with three denominators we get,
i −1
I0,3 =
16π 2 2C
i 1
I3µ = 2
Pµ
16π 2C

i 1
I3µν = Cg µν (∆ǫ − ln C) − 2P µ P ν
16π 2 4C

i 1
I3µνα = 2
Cg µν (−∆ǫ + ln C)P α + Cg να (−∆ǫ + ln C)P µ
16π 4C

µα ν α µ ν
+ Cg (−∆ǫ + ln C)P + 2P P P

i 1
I3µναβ = 2
C 2 (1 + ∆ǫ − ln C) gµα gνβ + gµβ gνα + gαβ gµν
16π 8C

+ 2C (∆ǫ − ln C) gµν P α P β + g νβ P α P µ + gνα P β P µ + g µα P β P ν

µβ α ν αβ µ ν α β µ ν
+g P P +g P P − 4P P P P (C.55)
where
P µ = x1 r1µ + x2 r2µ
C = x21 r12 + x22 r22 + 2x1 x2 r1 · r2 + x1 m21 + x2 m22
+(1 − x1 − x2 ) m23 − x1 r12 − x2 r22 (C.56)
C.7.4 Box integrals
i 1
I0,4 =
16π 2 6C 2
i −1 µ
I4µ = P
16π 2 6C 2

i −1
I4µν = µν
Cg − 2P P µ ν
16π 2 12C 2

i 1
I4µνα = µν α να µ µα ν
C (g P + g P + g P ) − 2P P P α µ ν
16π 2 12C 2

i 1
I4µναβ = C 2
(∆ ǫ − ln C) g µα νβ
g + g µβ να
g + g αβ µν
g
16π 2 24C 2

− 2C gµν P α P β + gνβ P α P µ + g να P β P µ + gµα P β P ν

µβ α ν αβ µ ν α β µ ν
+g P P +g P P + 4P P P P (C.57)
where
C.8. DIVERGENT PART OF 1–LOOP INTEGRALS 331
P µ = x1 r1µ + x2 r2µ + x3 r3µ
C = x21 r12 + x22 r22 + x23 r32 + 2x1 x2 r1 · r2 + 2x1 x3 r1 · r3 + 2x2 x3 r2 · r3
+x1 m21 + x2 m22 + x3 m23 + (1 − x1 − x2 − x3 ) m24
−x1 r12 − x2 r22 − x3 r32 (C.58)
C.8 Divergent part of 1–loop integrals

When we want to study the renomalization of a given theory it is often convenient to have
expressions for the divergent part of the one-loop integrals, with the integration on the
Feynman parameters already done. We present here the results for the most important
cases. These divergent parts were calculated with the help of the package OneLoop [9].
µ µ ,···µ
The results are for the functions T̂n , 2 n defined in Eq. (C.35).
C.8.1 Tadpole integrals

h i i
Div T̂1 = ∆ ǫ m2
16π 2
h i
Div T̂1µ = 0
h i i 1
Div T̂1µν = ∆ǫ m4 gµν (C.59)
16π 2 4
C.8.2 Self–Energy integrals

h i i
Div T̂2 = ∆ǫ
16π 2
h i
i 1
Div T̂2µ = − ∆ǫ r1µ
16π 2 2
h i
i 1
Div T̂2µν = 2 2 2 µν
∆ǫ (3m1 + 3m2 − r1 )g + 4r1 r1 µ ν
16π 2 12
h i
i 1
Div T̂2µνα = − ∆ǫ (4m21 + 2m22 − r12 ) (g µν r1α + gνα r1µ + gµα r1ν )
16π 2 24

α µ ν
+ 6 r1 r1 r1 (C.60)
C.8.3 Triangle integrals

h i
Div T̂3 = 0
h i
Div T̂3µ = 0
h i i 1
Div T̂3µν = ∆ǫ gµν
16π 2 4
h i
i 1
Div T̂3µνα = − ∆ ǫ g µν α
(r1 + r2
α
) + g να µ
(r1 + r µ
2 ) + g µα ν
(r 1 + r2
ν
)
16π 2 12
h i
µναβ i 1 2 2 2 µα νβ αβ µν µβ να
Div T̂3 = ∆ ǫ (2m 1 + 2m 2 + 2m 3 ) g g + g g + g g
16π 2 48
h i h i
+g αβ 2r1µ r1ν + r1µ r2ν + (r1 ↔ r2 ) + g µβ 2r1α r1ν + r1α r2ν + (r1 ↔ r2 )
h i h i
+g νβ 2r1α r1µ + r1α r2µ + (r1 ↔ r2 ) + gµν 2r1α r1β + r1α r2β + (r1 ↔ r2 )
h i h i
+g µα 2r1β r1ν + r1β r2ν + (r1 ↔ r2 ) + g να 2r1β r1µ + r1β r2µ + (r1 ↔ r2 )
µα νβ
+ −r12 + r1 · r2 − r22 αβ µν
g g +g g +g gµβ να
(C.61)
C.8.4 Box integrals

h i h i h i h i
Div T̂4 = Div T̂4µ = Div T̂4µν = Div T̂4µνα = 0
h i
i 1
Div T̂4µναβ = µν αβ µβ αν µα νβ
∆ǫ g g + g g + g g (C.62)
16π 2 24
C.9 Passarino-Veltman Integrals

C.9.1 The general definition
The description of the previous sections works well if one just wants to calculate the
divergent part of a diagram or to show the cancellation of divergences in a set of diagrams.
If one actually wants to numerically calculate the integrals the task is normally quite
complicated. Except for the self-energy type of diagrams the integration over the Feynman
parameters is normally quite difficult.
To overcome this problem a scheme was first proposed by Passarino and Veltman [10].
These scheme with the conventions of [11, 12] was latter implemented in the Mathematica
package FeynCalc [13] and, for numerical evaluation, in the LoopTools package [14, 15].
The numerical evaluation follows the code developed earlier by van Oldenborgh [16].
We will now describe this scheme. We will write the generic one-loop tensor integral
as Z
µ1 ···µp (2πµ)4−d d kµ1 · · · kµp
Tn ≡ d k (C.63)
iπ 2 D0 D1 D2 · · · Dn−1
where we follow for the momenta the conventions of section C.2 and Fig. C.1 and defined
D0 ≡ Dn and mn = m0 so that D0 = k2 − m20 (remember that rn ≡ r0 = 0. The
C.9. PASSARINO-VELTMAN INTEGRALS 333
main difference between this definition and the previous one Eq. (C.7) is that a factor of
i
16π 2
is taken out. This is because, as we have seen in section C.3 these integrals always
give that prefactor. So with our new convention that prefactor has to included in the
end. Factoring out the i has also the convenience of dealing with real functions in many
cases.2 From all those integrals in Eq. (C.63) the scalar integrals are, has we have seen,
of particular importance and deserve a special notation. It can be shown that there are
only four independent such integrals, namelly (4 − d = ǫ)
Z
(2πµ)ǫ 1
A0 (m20 ) = dd k (C.64)
iπ 2 k2 − m20
Z 1
Y
2 (2πµ)ǫ 1
B0 (r10 , m21 , m22 ) = dd k (C.65)
iπ 2
i=0
(k + ri )2 − m2i
Z Y2
2 2 2 (2πµ)ǫ 1
C0 (r10 , r12 , r20 , m21 , m22 , m23 ) = dd k (C.66)
iπ 2 (k + ri )2 − m2i
i=0
Z 3
Y
2 2 2 2 2 2 (2πµ)ǫ 1
D0 (r10 , r12 , r23 , r30 , r20 , r13 , m21 , . . . , m23 ) = dd k (C.67)
iπ 2 (k + ri )2 − m2i
i=0
where
2
rij = (ri − rj )2 ; ∀ i, j = (0, n − 1) (C.68)
Remember that with our conventions r0 = 0 so ri0 2 = r 2 . In all these expressions the iǫ
i
part of the denominator factors is supressed. The general one-loop tensor integrals are not
independent. Their decomposition is not unique. We follow the conventions of [13, 15] to
write
Z 1
Y
µ (2πµ)4−d 1
B ≡ dd k kµ (C.69)
iπ 2 (k + ri )2 − m2i
i=0
Z 1
Y
(2πµ)4−d 1
B µν ≡ dd k kµ kν (C.70)
iπ 2 (k + ri )2 − m2i
i=0
Z 2
Y
(2πµ)4−d 1
Cµ ≡ dd k kµ (C.71)
iπ 2 (k + ri )2 − m2i
i=0
Z 2
Y
µν (2πµ)4−d d µ ν 1
C ≡ d kk k (C.72)
iπ 2 (k + ri )2 − m2i
i=0
Z 2
Y
µνρ (2πµ)4−d 1
C ≡ dd k kµ kν kρ (C.73)
iπ 2 (k + ri )2 − m2i
i=0
Z 3
Y
µ (2πµ)4−d d µ 1
D ≡ d kk (C.74)
iπ 2 (k + r i )2 − m2
i
i=0
2
The one loop functions are in general complex, but in some cases they can be real. These cases
correspond to the situation where cutting the diagram does not corresponding to a kinematically allowed
process.
Z 3
Y
(2πµ)4−d 1
D µν ≡ dd k kµ kν (C.75)
iπ 2 (k + ri )2 − m2i
i=0
Z 3
Y
µνρ (2πµ)4−d d µ ν ρ 1
D ≡ d kk k k (C.76)
iπ 2
i=0
(k + ri )2 − m2i
Z 3
Y
(2πµ)4−d 1
D µνρσ ≡ dd k kµ kν kρ kσ (C.77)
iπ 2 (k + ri )2 − m2i
i=0
These integrals can be decomposed in terms of (reducible) functions in the following way:
B µ = r1µ B1 (C.78)
B µν
= g B00 + r1µ r1ν B11
µν
(C.79)
C µ
= r1µ C1 + r2µ C2 (C.80)
X 2
C µν = gµν C00 + riµ rjν Cij (C.81)
i=1
X 2 2
X
C µνρ
= (gµν riρ + g νρ riµ + gρµ riν ) C00i + riµ rjν rkρ Cijk (C.82)
i=1 i,j,k=1
3
X
D µ
= riµ Di (C.83)
i=1
3
X
D µν = gµν D00 + riµ rjν Dij (C.84)
i=1
3
X 2
X
D µνρ
= (gµν riρ + g νρ riµ + gρµ riν ) D00i + riµ rjν rkρ Dijk (C.85)
i=1 i,j,k=1
D µνρσ = (gµν gρσ + gµρ gνσ + gµσ gνρ ) D0000
X3
+ gµν riρ rjσ + g νρ riµ rjσ + gµρ riν rjσ + g µσ riν rjρ (C.86)
i,j=1

+g νσ riµ rjρ + gρσ riµ rjν D00ij
3
X
+ riµ rjν rkρ rlσ Cijkl (C.87)
i,j,k,l=1
All coefficient functions have the same arguments as the corresponding scalar functions
and are totally symmetric in their indices. In the FeynCalc [13] package one generic
notation is used,
PaVe i, j, . . . , {r210 , r212 , . . .}, {m20 , m21 , . . .} (C.88)
for instance
2
B11 (r10 , m20 , m21 ) = PaVe 1, 1, {r210 }, {m20 , m21 } (C.89)
All these coefficient functions are not independent and can be reduced to the scalar func-
tions. FeynCalc provides the command PaVeREduce[...] to acomplish that. This is very
useful if one wants to check for cancellation of divergences or for gauge invariance where
a number of diagrams have to cancel.
C.9.2 The divergences

The package LoopTools provides ways to numerically check for the cancellation of diver-
gences. However it is useful to know the divergent part of the Passarino-Veltman integrals.
Only a small number of these integrals are divergent. They are

Div A0 (m20 ) = ∆ǫ m20 (C.90)
2 2 2

Div B0 (r10 , m0 , m1 ) = ∆ǫ (C.91)
1
Div B1 (r210 , m20 , m21 ) = − ∆ǫ (C.92)
2
1
Div B00 (r210 , m20 , m21 ) = ∆ǫ 3m20 + 3m21 − r10
2
(C.93)
12
1
Div B11 (r210 , m20 , m21 ) = ∆ǫ (C.94)
3
1
Div C00 (r210 , r212 , r220 , m20 , m21 , m22 ) = ∆ǫ (C.95)
4
1
Div C001 (r210 , r212 , r220 , m20 , m21 , m22 ) = − ∆ǫ (C.96)
12
1
Div C002 (r210 , r212 , r220 , m20 , m21 , m22 ) = − ∆ǫ (C.97)
12
1
Div D0000 (r210 , . . . , m20 , . . .) = ∆ǫ (C.98)
24
(C.99)
These results were obtained with the package LoopTools, after reducing to the scalar
integrals with the command PaVeReduce, but they can be verified by comparing with our
results of section C.8, after factoring out the i/(16π 2 ).
C.9.3 Useful results for PV integrals

Although the PV approach is intended primarily to be used numerically there are situations
where one wants to have explicit results. These can be useful to check cancellation of
divergences or because in some simple cases the integrals can be done analytically. We
note that as our conventions for the momenta are the same in sections C.9 and C.7 one
can read immediately the integral representation of the PV in terms of the Feynman
parameters just by comparing both expressions, not forgetting to take out the i/(16π 2 )
factor. For instance, from Eq. (C.81) for C µν and Eq. (C.55) for I3µν we get
Z Z 1−x1
2 1 x1 x2
C12 (r12 , r12
2
, r22 , m20 , m21 , m22 ) = −Γ(3) dx1 dx2 (C.100)
4 0 0 C
with
C = x21 r12 + x22 r22 + x1 x2 (r12 + r22 − r12

2
) + x1 m21 + x2 m22
+(1 − x1 − x2 ) m20 − x1 r12 − x2 r22 (C.101)

Explicit expression for A0

This integral is trivial. There is no Feynman parameter and the integral can be read from
Eq. (C.51). We get, after factoring out the i/(16π 2 ),

m2
A0 (m2 ) = m2 ∆ǫ + 1 − ln 2 (C.102)
µ
Explicit expressions for the B functions

Function B0
The general form of the integral B0 (p2 , m21 , m22 ) can be read from Eq. (C.53). We obtain
Z 1
−x(1 − x)p2 + xm21 + (1 − x)m20
B0 (p2 , m20 , m21 ) = ∆ǫ − dx ln (C.103)
0 µ2
From this expression one can easily get the following results,
m20 m21
m20 ln µ2
− m21 ln µ2
B0 (0, m20 , m21 ) = ∆ǫ + 1 − (C.104)
m20 − m21
A0 (m20 ) − A0 (m21 )
B0 (0, m20 , m21 ) = (C.105)
m20 − m21
m2 A0 (m2 )
B0 (0, m2 , m2 ) = ∆ǫ − ln 2
= −1 (C.106)
µ m2
m2 A0 (m2 )
B0 (m2 , 0, m2 ) = ∆ǫ + 2 − ln 2
= +1 (C.107)
µ m2
m2
B0 (0, 0, m2 ) = ∆ǫ + 1 − ln (C.108)
µ2
Function B0′
The derivative of the B0 function with respect to p2 appears many times. From Eq. (C.103)
one can derive an integral representation,
Z 1
x(1 − x)
B0′ (p2 , m20 , m21 ) =− dx (C.109)
0 −p2 x(1 − x) + xm21 + (1 − x)m20
An important particular case corresponds to B0′ (m2 , m20 , m2 ) that appears in the self-
energy of the electron. In this case m is the electron mass and m0 = λ is the photon mass
that one has to introduce to regularize the IR divergent integral. The integral in this case
reduces to
Z 1
x(1 − x)
B0′ (m2 , λ2 , m2 ) = − dx
0 m 2 x2+ (1 − x)λ2
1 1 λ2
= + ln (C.110)
m2 2m2 m2
It is clear that in the limit λ → 0 this integral diverges.
Function B1
The explicit expression can be read from Eq. (C.53). We have
Z 1
2 1 −x(1 − x)p2 + xm21 + (1 − x)m20
B1 (p , m20 , m21 ) = − ∆ǫ + dxx ln (C.111)
2 0 µ2
For p2 = 0 this integral can be easily evaluated to give

2
2 2 1 1 m0 −3 + 4t − t2 − 4t ln t + 2t2 ln t
B1 (0, m0 , m1 ) = − ∆ǫ + ln + (C.112)
2 2 µ2 4(−1 + t)2
where we defined
m21
t= (C.113)
m20
From Eq. (C.112) one can shown that even for p2 = 0 B1 is not a symmetric function of
the masses,
B1 (p2 , m20 , m21 ) 6= B1 (p2 , m21 , m20 ) (C.114)
As this might appear strange let us show with one example how the coefficient functions
are tied to our conventions about the order of the momenta and Feynman parameters. Let
us consider the contribution to the self-energy of a fermion of mass mf of the exchange of
a scalar with mass ms . We can consider the two choices in Fig. C.4,
q q+r1
p p p p
q+r1 q
(r1=p) (r1=-p)
Figure C.4:
Now with the first choice (diagram on the left of Fig. C.4) we have
i h 2 2 2 2 2 2
i
−iΣ1 = (p
/ + m f )B 0 (p , m s , m f ) + p
/ B 1 (p , m s , m f )
16π 2
i h 2 2 2 2 2 2
2 2 2
i
= p
/ B 0 (p , m s , m f ) + B 1 (p , m s , m f ) + m B
f 0 (p , m s , m f (C.115)
)
16π 2
while with the second choice we have
i h 2 2 2 2 2 2
i
−iΣ2 = − p
/ B 1 (p , m f , m s ) + m f B 0 (p , m f , m s ) (C.116)
16π 2
How can these two expressions be equal? The reason has precisely to do with the non
symmetry of B1 with respect to the mass entries. In fact from Eq. (C.111) we have
Z 1
2 1 −x(1 − x)p2 + xm21 + (1 − x)m20
B1 (p , m20 , m21 ) = − ∆ǫ + dxx ln
2 0 µ2
Z 1
1 −x(1 − x)p2 + (1 − x)m21 + xm20
= − ∆ǫ + dx(1 − x) ln
2 0 µ2

1 1
= − ∆ǫ + ∆ǫ − B0 (p2 , m21 , m20 ) − ∆ǫ + B1 (p2 , m21 , m20 )
2 2

= − B0 (p2 , m21 , m20 ) + B1 (p2 , m21 , m20 ) (C.117)
where we have changed variables (x → 1 − x) in the integral and used the definitions of
B0 and B1 . We have then, remembering that B0 (p2 , m2s , m2f ) = B0 (p2 , m2f , m2s ),

B1 (p2 , m2f , m2s ) = − B0 (p2 , m2s , m2f ) + B1 (p2 , m2s , m2f ) (C.118)
and therefore Eqs. (C.115) and (C.116) are equivalent.
Explicit expressions for the C functions

In Eq. (C.100) we have already given the general form of C12 . The other functions are
very similar. In the following we just present the results for the particular case of p2 = 0.
This case is important in many situations where it is a good approximation to neglect the
external momenta in comparison with the masses of the particles in the loop. We also
warn the reader that the coefficient functions Ci , Cij obtained from LoopTools are not
well defined in this limit. Hence there is some utility in given them here.
Function C0
Z Z 1−x1
1 1 1
C0 (0, 0, 0, m20 , m21 , m22 ) = −Γ(3) dx1 dx2 2 + x m2 + (1 − x − x )m2
2 0 0 x m
1 1 2 2 1 2 0
Z 1 Z 1−x1
1 1
= − 2 dx1 dx2
m0 0 0 x1 t1 + x2 t2 + (1 − x1 − x2 )
1 −t1 ln t1 + t1 t2 ln t1 + t2 ln t2 − t1 t2 ln t2
= − (C.119)
m20 (−1 + t1 )(t1 − t2 )(−1 + t2 )
where
m21 m22
t1 = ; t2 = (C.120)
m20 m20
Using the properties of the logarithms one can show that in this limit C0 is a symmetric
function of the masses. This expression is further simplified when two of the masses are
equal, as it happens in the µ → eγ problem. Then t = t1 = t2 ,
1 −1 + t − ln t
C0 (0, 0, 0, m20 , m21 , m21 ) = − (C.121)
m20 (−1 + t)2
in agreement with Eq.(20) of [17]. In the case of equal masses for all the loop particles we
have
1
C0 (0, 0, 0, m20 , m20 , m20 ) = − 2 (C.122)
2m0
Before we close this section on C0 there is another particular case when it is useful to have
an explicit case for it. This in the case when it is IR divergent as in the QED vertex. The
functions needed is C0 (m2 , m2 , 0, m2 , λ2 , m2 ). Using the definition we have
Z 1 Z 1−x1
2 2 2 2 2 1
C0 (m , m , 0, m , λ , m ) = − dx1 dx2
0 0 m2 (1 − 2x1 + x21 ) + x1 λ2
Z 1
1 − x1
= − dx1
0 m2 (1 − x1 )2 + (1 − x1 )λ2
Z 1
x
= − dx
0 m 2 x2 + xλ2
1 λ2 1
= 2
ln 2
= B0′ (m2 , λ2 , m2 ) − 2 (C.123)
2m m m
Function C00
Z 1 Z 1−x1
1 C
C00 (0, 0, 0, m20 , m21 , m22 ) = Γ(3) dx1 dx2 ∆ǫ − ln
4 0 0 µ2
Z Z 1−x1
1 1 1 x1 m21 + x2 m22 + (1 − x1 − x2 )m20
= ∆ǫ − dx1 dx2 ln
4 2 0 0 µ2

1 m20 3 t21
= ∆ǫ − ln 2 + − ln t1
4 µ 8 4(t1 − 1)(t1 − t2 )
t22
+ ln t2 (C.124)
4(t2 − 1)(t1 − t2 )
where, as before
m21 m22
t1 = ; t2 = (C.125)
m20 m20
Using the properties of the logarithms one can show that in this limit C00 is a symmetric
function of the masses. This expression is further simplified when two of the masses are
equal. Then t = t1 = t2 ,

1 m2 −3 + 4t − t2 − 4t ln t + 2t2 ln t
C00 (0, 0, 0, m20 , m21 , m21 ) = ∆ǫ − ln 20 −
4 µ 8(t − 1)2
1
= − B1 (0, m20 , m21 ) (C.126)
2
Functions Ci and Cij
We recall that the definition of the coefficient functions is not unique, it is tied to a
particular convention for assigning the loop momenta and Feynman parameters, as shown
in Fig. C.1. For the particular case of the C functions we show our conventions in Fig. C.5.
p1 (q,m0) p2
1-x1-x2
x1 x2
(q+r1,m1) (q+r2,m2)
p3
Figure C.5:
With the same techniques we obtain,
Z 1 Z 1−x1
1 x1
C1 (0, 0, 0, m20 , m21 , m22 ) = dx1 dx2
m20 0 0 x1 t1 + x2 t2 + (1 − x1 − x2 )

1 t1 t1 (t1 − 2t2 + t1 t2 )
= − − ln t1
m20 2(−1 + t1 )(t1 − t2 ) 2(−1 + t1 )2 (t1 − t2 )2

t2 2 − 2t1 t2 2 + t1 2 t2 2
+ ln t2 (C.127)
2(−1 + t1 )2 (t1 − t2 )2 (−1 + t2 )
Z 1 Z 1−x1
1 x2
C2 (0, 0, 0, m20 , m21 , m22 ) = dx1 dx2
m20 0 0 x1 t1 + x2 t2 + (1 − x1 − x2 )

1 t2 ln t1
= − 2 − +
m0 2(t1 − t2 )(−1 + t2 ) 2(−1 + t1 )(−1 + t2 )2

2t1 t2 − 2t1 2 t2 − t2 2 + t1 2 t2 2 t1
+ 2 2 ln t (C.128)
2(−1 + t1 )(t1 − t2 ) (−1 + t2 ) 2
Z 1 Z 1−x1
1 xi xj
Cij (0, 0, 0, m20 , m21 , m22 ) = − 2 dx1 dx2 (C.129)
m0 0 0 x1 t1 + x2 t2 + (1 − x1 − x2 )
where we have not written explicitly the Cij for i, j = 1, 2 because they are rather lengthy.
However a simple Fortran program can be developed [9] to calculate all the three point
functions in the zero external limit case. This is useful because in this case some of
the functions from LoopTools will fail. Notice that the Ci and Cij functions are not
symmetric in their arguments. This a consequence of their non-uniqueness, they are tied
to a particular convention. This is very important when ones compares with other results.
However using their definition one can get some relations. For instance we can show
C1 (0, 0, 0, m20 , m21 , m22 ) = C1 (0, 0, 0, m22 , m21 , m20 ) (C.130)
C2 (0, 0, 0, m20 , m21 , m22 ) = −C0 (0, 0, 0, m22 , m21 , m20 ) − C1 (0, 0, 0, m22 , m21 , m20 )
−C2 (0, 0, 0, m22 , m21 , m20 ) (C.131)
In the limit m1 = m2 we get the simple expressions,
C1 (0, 0, 0, m20 , m21 , m21 ) = C2 (0, 0, 0, m20 , m21 , m21 )

1 3 − 4t + t2 + 2 ln t
= − (C.132)
m20 4(−1 + t)3
C11 (0, 0, 0, m20 , m21 , m21 ) = C22 (0, 0, 0, m20 , m21 , m21 ) = 2 C12 (0, 0, 0, m20 , m21 , m21 )
1 −11 + 18t − 9t2 + 2t3 − 6 ln t

= − (C.133)
m20 18(−1 + t)4
in agreement with Eqs. (21-22) of [17]. The case of masses equal gives
1
C1 (0, 0, 0, m20 , m20 , m20 ) = C2 (0, 0, 0, m20 , m20 , m20 ) = (C.134)
6m20
1
C11 (0, 0, 0, m20 , m20 , m20 ) = C22 (0, 0, 0, m20 , m20 , m20 ) = − (C.135)
12m20
1
C12 (0, 0, 0, m20 , m20 , m20 ) = − (C.136)
24m20
The package PVzem

As we said before, in many situations it is a good approximation to neglect the external
momenta. In this case, the loop functions are easier to evaluate and one approach is
for each problem to evaluate them. However our approach here is more in the direction
of automatically evaluating the one-loop amplitudes. If one does that with the use of
FeynCalc, has we have been doing, then the result is given in terms of standard functions
that can be numerically evaluated with the package LoopTools. However this package has
problems with this limit. This is because this limit is unphysical. Let us illustrate this
point calculating the functions C1 (m2 , 0, 0, m2S , m2F , m2F ) and C2 (m2 , 0, 0, m2S , m2F , m2F ) for
mB = 100 GeV, mF = 80 GeV and m2 ranging from 10−6 to 100 GeV. To better illustrate
our point we show two plots with different scales on the axis.
In these plots, CiEx are the exact Ci functions calculated with LoopTools and CiAp are
the Ci calculated in the zero momenta limit. We can see that only for external momenta
(in this case corresponding to the mass m2 ) close enough to the masses of the particles
in the loop, the exact result deviates from the approximate one. However for very small
mS=100 GeV, mF=80 GeV mS=100 GeV, mF=80 GeV

0.30 4.00
m B Ci
m B Ci
2
2
Ex
0.28 Ex C2
C1
2.00
0.26 Ex
C2 Ex
C1
0.24
0.00
0.22 Ap Ap Ap Ap
C1 = C2 C1 = C2
0.20 0 1 2
-2.00 -6 -4 -2 0 2
10 10 10 10 10 10 10 10
m2 (GeV) m2 (GeV)
Figure C.6:
values of the external momenta, LoopTools has numerical problems as shown in the right
panel of Fig. C.6. To overcome this problem I have developed a Fortran package that
evaluates all the C functions in the zero external momenta limit. There are no restrictions
on the masses being equal or different and the conventions are the same as in FeynCalc
and LoopTools, for instance,
c12zem(m02, m12, m22) = c0i(cc12, 0, 0, 0, m02, m12, m22) (C.137)
where c0i(cc12, · · · ) is the LoopTools notation and c12zem(· · · ) is the notation of my

package, called PVzem. It can be obtained from the address indicated in Ref.[9]. The
approximate functions shown in Fig. C.6 were calculated using that package. We include
here the Fortran code used to produce that figure.
*************************************************************
* *
* Program LoopToolsExample *
* *
* This program calculates the values used in the plots *
* of Figure 20. For the exact results the LoopTools *
* package was used. The package PVzem was used for the *
* approximate results. *
* *
* Version of 14/05/2012 *
* *
* Author: Jorge C. Romao *
* e-mail: jorge.romao@ist.utl.pt *
*************************************************************
program LoopToolsExample
implicit none
*
* LoopTools has to be used with FORTRAN programs with the
* extension .F in order to have the header file "looptools.h"
* preprocessed. This file includes all the definitions used
* by LoopTools.
*
* Functions c1zem and c2zem are provided by the package PVzem.
*
#include "looptools.h"
integer i
real*8 m2,mF2,mS2,m
real*8 lgmmin,lgmmax,lgm,step
real*8 rc1,rc2
real*8 c1zem,c2zem
mS2=100.d0**2
mF2=80.d0**2
*
* Initialize LoopTools. See the LoopTools manual for further
* details. There you can also learn how to set the scale MU
* and how to handle the UR and IR divergences.
*
call ltini
lgmmax=log10(100.d0)
lgmmin=log10(1.d-6)
step=(lgmmax-lgmmin)/100.d0
lgm=lgmmin-step
open(10,file=’plot.dat’,status=’unknown’)
do i=1,101
lgm=lgm+step
m=10.d0**lgm
m2=m**2
*
* In LoopTools the c0i(...) are complex functions. For the
* kinematics chosen here they are real, so we take the real
* part for comparison.
*
rc1=dble(c0i(cc1,m2,0.d0,0.d0,mS2,mF2,mF2))
rc2=dble(c0i(cc2,m2,0.d0,0.d0,mS2,mF2,mF2))
write(10,100)m,rc1*mS2,rc2*mS2,c1zem(mS2,mF2,mF2)*mS2,
& c2zem(mS2,mF2,mF2)*mS2
enddo
100 format(5(e22.14))
call ltexi
end
************** End of Program LoopToolsExample.F ************
When the above program is compiled, the location of the header file looptools.h
must be known by the compiler. This is best achieved by using a Makefile. We give
below, as an example, the one that was used with the above program. Depending on the
installation details of LoopTools the paths might be different.
FC =
LT = /usr/local/lib/LoopTools
FFLAGS = -c -O -I$(LT)/include
LDFLAGS =
LINKER = $(FC)
LIB = -L$(LT)/lib
LIBS = -looptools
.f.o:
$(FC) $(FFLAGS) $*.F
files = LoopToolsExample.o PVzem.o
all: $(files)
$(LINKER) $(LDFLAGS) -o Example $(files) $(LIB) $(LIBS)
Explicit expressions for the D functions

Function D0
The various D functions can be calculated in a similar way. However they are rather
lengthy and have to handled numerically [9]. Here we just give D0 for the equal masses
case.
C.10. EXAMPLES OF 1-LOOP CALCULATIONS WITH PV FUNCTIONS 345
Z Z 1−x1 Z 1−x1 −x2

1 1 1
D0 (0, · · · , 0, m2 , m2 , m2 , m2 ) = Γ(4) dx1 dx2 dx3
6 0 0 0 (m2 )2
Z 1 Z 1−x1 Z 1−x1 −x2
1
= dx 1 dx 2 dx3
m4 0 0 0
1
= (C.138)
6m4
C.10 Examples of 1-loop calculations with PV functions

In this section we will work out in detail a few examples of one-loop calculations using the
FeynCalc package and the Passarino-Veltman scheme.
C.10.1 Vaccum Polarization in QED

We have done this example in section 4.1.1 using the techniques described in sections C.3,
C.4 and C.5. Now we will use FeynCalc. The first step is to write the Matematica
program. We list it below:
(*********************** Program VacPol.m **************************)
(* First input FeynCalc *)
<< FeynCalc.m
(* These are some shorthands for the FeynCalc notation *)
dm[mu_]:=DiracMatrix[mu,Dimension->D]
dm[5]:=DiracMatrix[5]
ds[p_]:=DiracSlash[p]
mt[mu_,nu_]:=MetricTensor[mu,nu]
fv[p_,mu_]:=FourVector[p,mu]
epsilon[a_,b_,c_,d_]:=LeviCivita[a,b,c,d]
id[n_]:=IdentityMatrix[n]
sp[p_,q_]:=ScalarProduct[p,q]
li[mu_]:=LorentzIndex[mu]
L:=dm[7]
R:=dm[6]
(* Now write the numerator of the Feynman diagram. We define the

constant
C=alpha/(4 pi)
*)
num:= - C Tr[dm[mu] . (ds[q] + m) . dm[nu] . (ds[q]+ds[k]+m)]
(* Tell FeynCalc to evaluate the integral in dimension D *)
SetOptions[OneLoop,Dimension->D]
(* Define the amplitude *)
amp:=num * FeynAmpDenominator[PropagatorDenominator[q+k,m], \
PropagatorDenominator[q,m]]
(* Calculate the result *)
res:=(-I / Pi^2) OneLoop[q,amp]
(***************** End of Program VacPol.m *********************)
The output from Mathematica is:
2 2 2 2 2 2 2 2 2
Out[4]= (4 C (k + 6 m B0[0, m , m ] - 3 (k + 2 m ) B0[k , m , m ])
2 2
(k g[mu, nu] - k[mu] k[nu])) / (9 k )
Now remembering that,

α
C= (C.139)
4π
and
i Πµν (k, ε) = −i k 2 Pµν
T
Π(k, ε) (C.140)
we get

α 4 8 m2 2 2 4 2m2 2 2 2
Π(k, ε) = − − B0 (0, m , m ) + 1+ 2 B0 (k , m , m ) (C.141)
4π 9 3 k2 3 k
To obtain the renormalized vacuum polarization one needs to know the value of Π(0, ε).
To do that one has to take the limit k → 0 in Eq. (C.141). For that one uses the derivative
of the B0 function
∂
B0′ (p2 , m21 , m22 ) ≡ 2 B0 (p2 , m21 , m22 ) (C.142)
∂p
to obtain
α 4 4 2 2 8 2 ′ 2 2
Π(0, ε) = − + B0 (0, m , m ) + m B0 (0, m , m ) (C.143)
4π 9 3 3
Using
1
B0′ (0, m2 , m2 ) = (C.144)
6m2
we finally get
α 4 2 2
Π(0, ε) = −δZ3 = B0 (0, m , m ) (C.145)
4π 3
and the final result for the renormalized vertex is:

R α 1 2m2 2 2 2 2 2

Π (k) = − + 1+ 2 B0 (k , m , m ) − B0 (0, m , m ) (C.146)
3π 3 k
If we want to compare with our earlier analytical results we need to know that
m2
B0 (0, m2 , m2 ) = ∆ε − ln (C.147)
µ2
Then Eq. (C.146) reproduces the result of Eq. (4.54). The comparison between Eq. (C.146)
and Eq. (4.56) can be done numerically using the package LoopTools[15].
C.10.2 Electron Self-Energy in QED

In this section we repeat the calculation of section 4.1.2 using the Passarino-Veltman
scheme. We start with the Mathematica program,
(********************* Program SelfEnergy.m ***********************)
<< FeynCalc.m

L:=dm[7]
R:=dm[6]
(* Tell FeynCalc to reduce the result to scalar functions *)
SetOptions[{B0,B1,B00,B11},BReduce->True]

constant
C= - alpha/(4 pi)
The minus sign comes from the photon propagator. The factor
i/(16 pi^2) is already included in this definition.
*)
num:= C dm[mu] . (ds[p]+ds[k]+m) . dm[mu]
(* Tell FeynCalc to evaluate the one-loop integral in dimension D *)
(* Define the amplitude *)
amp:= num \
FeynAmpDenominator[PropagatorDenominator[p+k,m], \
PropagatorDenominator[k]]
(* Calculate the result *)
res:=(-I / Pi^2) OneLoop[k,amp]
ans=-res;
(*
The minus sign in ans comes from the fact that -i \Sigma = diagram
*)
(* Calculate the functions A(p^2) and B(p^2) *)
A=Coefficient[ans,DiracSlash[p],0];
B=Coefficient[ans,DiracSlash[p],1];
(* Calculate deltm *)
delm=A + m B /. p->m
(* Calculate delZ2 *)
Ap2 = A /. ScalarProduct[p,p]->p2
Bp2 = B /. ScalarProduct[p,p]->p2
aux=2 m D[Ap2,p2] + Bp2 \

+ 2 m^2 D[Bp2,p2] /. D[B0[p2,0,m^2],p2]->DB0[p2,0,m^2]
aux2= aux /. p2->m^2
aux3= aux2 /. A0[m^2]->m^2 (B0[m^2,0,m^2] -1)
delZ2=Simplify[aux3]
(***************** End of Program SelfEnergy.m ********************)
The output from Mathematica is:
2 2
A = -(C (-2 m + 4 m B0[p , 0, m ]))
2 2 2 2 2 2
C (p + A0[m ] - (m + p ) B0[p , 0, m ])
B = -(-----------------------------------------)
2
p
2 2 2 2 2
C (m - A0[m ] - 2 m B0[m , 0, m ])
delm = ------------------------------------
m
2 2 2 2 2
delZ2= C (-2 + B0[m , 0, m ] - 4 m DB0[m , 0, m ])
We therefore get

αm 1 2 2
A = − + B0 (p , 0, m ) (C.148)
π 2

α 1 2 m2 2 2
B = 1 + 2 A0 (m ) − 1 + 2 B0 (p , 0, m ) (C.149)
4π p p

3αm 1 1 2 2 2 2
δm = − + A0 (m ) + B0 (m , 0, m ) (C.150)
4π 3 3m2 3
One can check that Eq. (C.150) is in agreement with Eq. (4.80). For that one needs the
following relations,

A0 (m2 ) = m2 B0 (m2 , 0, m2 ) − 1 (C.151)
m2
B0 (m2 , 0, m2 ) = ∆ε + 2 − ln (C.152)
µ2
Z 1
m 2 x2 5 3 m2
dx(1 + x) ln = − + ln 2 (C.153)
0 µ2 2 2 µ
For δZ2 we get
α
δZ2 = 2 − B0 (m2 , 0, m2 ) − 4m2 B0′ (m2 , λ2 , m2 ) (C.154)
4π
This expression can be shown to be equal to Eq. (4.83) although this is not trivial. The
reason is that B0′ is IR divergent, hence the parameter λ that controls the divergence.
C.10.3 QED Vertex

In this section we repeat the calculation of section 4.1.3 for the QED vertex using the
Passarino-Veltman scheme. The Mathematica program should by now be easy to under-
stand. We just list it here,
(********************* Program QEDVertex.m ***********************)
<< FeynCalc.m

L:=dm[7]
R:=dm[6]
(* Tell FeynCalc to reduce the result to scalar functions *)
SetOptions[{B1,B00,B11},BReduce->True]

constant
C= alpha/(4 pi)
The kinematics is: q = p1 -p2 and the internal momenta is k.
*)
num:=Spinor[p1,m] . dm[ro] . (ds[p1]-ds[k]+m) . ds[a] \

. (ds[p2]-ds[k]+m) . dm[ro] . Spinor[p2,m]
amp:=C num \
FeynAmpDenominator[PropagatorDenominator[k,lbd], \
PropagatorDenominator[k-p1,m], \
PropagatorDenominator[k-p2,m]]
(* Define the on-shell kinematics *)
onshell={ScalarProduct[p1,p1]->m^2,ScalarProduct[p2,p2]->m^2, \
ScalarProduct[p1,p2]->m^2-q2/2}
(* Define the divergent part of the relevant PV functions*)
div={B0[0,0,m^2]->Div,B0[0,m^2,m^2]->Div, \
B0[m^2,0,m^2]->Div,B0[m^2,lbd^2,m^2]->Div,\
B0[q2,m^2,m^2]->Div,B0[0,lbd^2,m^2]->Div}
res1:=(-I / Pi^2) OneLoop[k,amp]
res:=res1 /. onshell
auxV1:= res /.onshell

auxV2:= PaVeReduce[auxV1]
auxV3:= auxV2 /. div
divV:=Simplify[Div*Coefficient[auxV3,Div]]
(* Check that the divergences do not cancel *)
testdiv:=Simplify[divV]
ans1=res;
var=Select[Variables[ans1],(Head[#]===StandardMatrixElement)&]
Set @@ {var, {ME[1],ME[2]}}
(* Extract the different Matrix Elements
Mathematica writes the result in terms of 2 Standard Matrix

Elements. To have a simpler result we substitute these elements
by simpler expressions (ME[1],ME[2]).
{StandardMatrixElement[u[p1, m1] . u[p2, m2]],
StandardMatrixElement[u[p1, m1] . ga[mu] . u[p2, m2]]}
*)
ans2=Simplify[PaVeReduce[ans1]]
CE11=Coefficient[ans2,ScalarProduct[a,p1] ME[1]]
CE12=Coefficient[ans2,ScalarProduct[a,p2] ME[1]]
CE2=Coefficient[ans2, ME[2]]
ans3=CE11 (ScalarProduct[a,p1]+ScalarProduct[a,p2]) ME[1] + \

CE2 ME[2]
test1:=Simplify[CE11-CE12]
test2:=Simplify[ans2-ans3]
(* ME[2] is \overline{u}(p’)\gamma_{\mu}u(p) and ME[3] is

\frac{i}{2m} \overline{u}(p’)\sigma_{\mu\nu} q^{\u} u(p)
*)
ans4= ans3 /. {(ScalarProduct[a,p1]+ScalarProduct[a,p2]) \

ME[1] -> 2 m ME[2] -2 m ME[3]}
CGamma:=Coefficient[ans4,ME[2]]
CSigmaAux:=Coefficient[ans4,ME[3]]
test3:=Simplify[ans4-CGamma ME[2] -CSigmaAux ME[3]]

F2:=Simplify[CSigmaAux /. lbd->0]
delZ1aux= - CGamma /. q2->0
delZ1:= delZ1aux /. lbd->0
F1:=CGamma + delZ1 /. lbd->0
(***************** End of Program QEDVertex.m ********************)
From this program we can obtain first the value of δZ1 . We get
2 2 2 2 2
delZ1= C(1 - B0[0, 0, m ] + 2 B0[0, m , m ] - 2 B0[m , 0, m ] -
2 2 2 2 2
4 m C0[m , m , 0, m , 0, m ])
which can be written as

α
δZ1 = 1 − B0 (0, 0, m2 ) + 2B0 (0, m2 , m2 ) − 2B0 (m2 , m2 )
2π

−4m2 C0 (m2 , m2 , 0, m2 , λ2 , m2 ) (C.155)
where we have introduced a small mass for the photon in the function C0 (m2 , m2 , 0, m2 , λ2 , m2 )
because it is IR divergent when λ → 0 (see Eq. (C.123)). Using the results of Eqs. (C.106),
(C.107), (C.108) and Eq. (C.123) we can show the important result
δZ1 = δZ2 (C.156)
where δZ2 was defined in Eq. (C.154). After performing the renormalization the coefficient
F1 (k2 ) is finite and given by
2 2 2
q2 q2 B0[0, 0, m ] 2 q2 B0[0, m , m ]
F1 = C (-(---------) - --------------- + ------------------ -
2 2 2
q2 - 4 m q2 - 4 m q2 - 4 m
2 2 2 2 2 2 2 2
8 m B0[0, m , m ] 3 q2 B0[q2, m , m ] 8 m B0[q2, m , m ]
------------------ - ------------------- + -------------------
2 2 2
q2 - 4 m q2 - 4 m q2 - 4 m
2 2 2 2 2 2 2
2 q2 B0[m , 0, m ] 4 q2 m C0[m , m , 0, m , 0, m ]
+ ------------------ - -------------------------------- +
2 2
q2 - 4 m q2 - 4 m
4 2 2 2 2 2 2 2 2 2
16 m C0[m , m , 0, m , 0, m ] 2 q2 C0[m , m , q2, m , 0, m ]
------------------------------ - ------------------------------- +
2 2
q2 - 4 m q2 - 4 m
2 2 2 2 2 4 2 2 2 2
12 q2 m C0[m , m , q2, m , 0, m ] 16 m C0[m , m , q2, m , 0, m ]
---------------------------------- - ------------------------------- )
2 2
q2 - 4 m q2 - 4 m
In[5]:= F1 /. q2->0
Out[5]= 0
while the coefficient F2 (q 2 ) does not need renormalization and it is given by,
2 2 2 2 2 2
4 m (1 + B0[0, 0, m ] + B0[q2, m , m ] - 2 B0[m , 0, m ])
F2= C ---------------------------------------------------------
2
q2 - 4 m
and for F2 (0) we get
2 2 2 2 2
F2[0]= -(C (1 + B0[0, 0, m ] + B0[0, m , m ] - 2 B0[m , 0, m ]))
Using the results of the Appendix we can show that,

α
F2 (0) = (C.157)
2π
a well known result, first obtained by Schwinger even before the renormalization program
was fully understood (F2 (q 2 ) is finite).
C.11. MODERN TECHNIQUES IN A REAL PROBLEM: µ → Eγ 355
C.11 Modern techniques in a real problem: µ → eγ

In the previous sections we have redone most of the QED standard textbook examples
using the PV decomposition and automatic tools. Here we want to present a more complex
example, the calculation of the partial width µ → eγ in an arbitrary theory where the
charged leptons couple to scalars and fermions, charged or neutral. This has been done in
Ref.[17] for fermions and bosons of arbitrary charge QF and QB , but for simplicity I will
consider here separately the cases of neutral and charged scalars.
C.11.1 Neutral scalar charged fermion loop

We will consider a theory with the following interactions,
l- F-
S 0 i (A P + A P ) S 0 i ( B P +B P )
L L R R L L R R
F- l-
where F − is a fermion with mass mF and S 0 a neutral scalar with mass mS . In fact
BL,R are not independent of AL,R but it is easier for our programming to consider them
completely general. The Feynman rule for the coupling of the photon with the lepton is
−i e Qℓ γ µ where e is the positron charge (for an electron Qℓ = −1). ℓ−i can be any of the
leptons but we will omit all indices in the program, the lepton being identified by its mass
and from the assumed kinematics
ℓ2 (p2 ) → ℓ1 (p1 ) + γ(k) (C.158)
The diagrams contributing to the process are given in Fig. C.7,
k k
D1 D1 D1
p2 q p1 p2 q p1 p2 q p1
D5 D7
D2 D3
D4 D6
k
1) 2) 3)
Figure C.7:
where
D1 = q 2 − m2S ; D2 = (p2 + q)2 − m2F ; D3 = (q + p2 − k)2 − m2F (C.159)

D4 = D3 ; D6 = D2 ; D5 = (p2 − k)2 − m22 = −2p2 · k (C.160)
D7 = (p1 + k)2 − m21 = 2p1 · k = −D5 (C.161)
The amplitudes are
e Qℓ
iM1 = u(p1 ) (AL PL + AR PR ) (q/ + p/2 − k/ + mF ) γ µ (q/ + p/2 + mF )
D1 D2 D3
(BL PL + BR PR ) u(p2 ) εµ (k) (C.162)
e Qℓ
iM2 = u(p1 ) (AL PL + AR PR ) (q/ + p/2 − k/ + mF ) (BL PL + BR PR )
D1 D4 D5
(p/2 − k/2 + m2 ) γ µ u(p2 ) εµ (k) (C.163)
e Qℓ
iM3 = u(p1 )γ µ (p/1 + k/ + mF ) (AL PL + AR PR ) (q/ + p/2 + m1 )
D1 D6 D7
(BL PL + BR PR ) u(p2 ) εµ (k) (C.164)
On-shell the amplitude will take the form (we have p1 · k = p2 · k)
h i
iM = 2p2 · ε(k) CL u(p1 )PL u(p2 ) + CR u(p1 )PR u(p2 )
+DL u(p1 )ε/PL u(p2 ) + DR u(p1 )ε/PR u(p2 ) (C.165)
If we write the amplitude as
M = Mµ εµ (k) (C.166)
then gauge invariance implies
Mµ kµ = 0 (C.167)
Imposing this condition on Eq. (C.165) we get the relations
DL = −m2 CR − m1 CL (C.168)
DR = −m1 CR − m2 CL (C.169)
Assuming these relations the amplitude can be written as
iM = CL [2p2 · ε(k)u(p1 )PL u(p2 ) − m1 u(p1 )ε/(k)PL u(p2 ) − m2 u(p1 )ε/(k)PR u(p2 )]
+CR [2p2 · ε(k)u(p1 )PR u(p2 ) − m2 u(p1 )ε/(k)PL u(p2 ) − m1 u(p1 )ε/(k)PR u(p2 )](C.170)
and the decay width will be

1 3
Γ= m22 − m21 |CL |2 + |CR |2 (C.171)
16πm32
As the coefficient of p2 · ε(k) only comes from the 3-point function (amplitude M1 ) this
justifies the usual procedure of just calculating that coefficient and forgetting about the
self-energies (amplitudes M2 and M3 ). However these amplitudes are crucial for the can-
cellation of divergences and for gauge invariance. Now we will show the power of the auto-
matic FeynCalc [13] program and calculate both the coefficients CL,R and DL,R , showing
the cancellation of the divergences and that the relations, Eqs. (C.168) and (C.169) needed
for gauge invariance are satisfied. We start by writing the mathematica program:
(************************ Program mueg-ns.m **************************)

(*
This program calculates the COMPLETE (both the 3 point amplitude and
the two self energy type on each external line) amplitudes for
\mu -> e \gamma when the fermion line in the loop is charged and the
neutral line is a scalar. The \mu has momentum p2 and mass m2, the
electron (p1,m1) and the photon momentum k. The momentum in the loop
is q.
The assumed vertices are,
1) Electron-Scalar-Fermion:
Spinor[p1,m1] (AL P_L + AR P_R) Spinor [pf,mf]
2) Fermion-Scalar-Muon:
Spinor[pf,mf] (BL P_L + BR P_R) Spinor [p2,m2]

*)
L:=dm[7]
R:=dm[6]
(*
*)
gA:= AL DiracMatrix[7] + AR DiracMatrix[6]

gB:= BL DiracMatrix[7] + BR DiracMatrix[6]
num1:=Spinor[p1,m1] . gA . (ds[q]+ds[p2]-ds[k]+mf) . ds[Polarization[k]]\

. (ds[q]+ds[p2]+mf) . gB . Spinor[p2,m2]
num2:=Spinor[p1,m1] . gA . (ds[q]+ds[p1]+mf) . gB . (ds[p1]+m2) . \

ds[Polarization[k]] . Spinor[p2,m2]
num3:=Spinor[p1,m1] . ds[Polarization[k]] . (ds[p2]+m1) . gA . \

(ds[q]+ds[p2]+mf) . gB . Spinor[p2,m2]
amp1:=num1 \
FeynAmpDenominator[PropagatorDenominator[q+p2-k,mf], \
PropagatorDenominator[q+p2,mf], \
PropagatorDenominator[q,ms]]
amp2:=num2 \
FeynAmpDenominator[PropagatorDenominator[q+p1,mf], \
PropagatorDenominator[p2-k,m2], \
amp3:=num3 \
FeynAmpDenominator[PropagatorDenominator[p1+k,m1], \
onshell={ScalarProduct[p1,p1]->m1^2,ScalarProduct[p2,p2]->m2^2, \
ScalarProduct[k,k]->0,ScalarProduct[p1,k]->(m2^2-m1^2)/2,\
ScalarProduct[p2,k]->(m2^2-m1^2)/2, \
ScalarProduct[p2,Polarization[k]]->p2epk, \
ScalarProduct[p1,Polarization[k]]->p2epk}

div={B0[m1^2,mf^2,ms^2]->Div,B0[m2^2,mf^2,ms^2]->Div, \
B0[0,mf^2,ms^2]->Div,B0[0,mf^2,mf^2]->Div,B0[0,ms^2,ms^2]->Div}
res1:=(-I / Pi^2) OneLoop[q,amp1]

res:=res1+res2+res3 /. onshell
auxT1:= res1 /.onshell

auxT2:= PaVeReduce[auxT1]
auxT3:= auxT2 /. div
divT:=Simplify[Div*Coefficient[auxT3,Div]]
auxS1:= res2 + res3 /.onshell

auxS2:= PaVeReduce[auxS1]
auxS3:= auxS2 /. div
divS:=Simplify[Div*Coefficient[auxS3,Div]]
(* Check cancellation of divergences

testdiv should be zero because divT=-divS *)
testdiv:=Simplify[divT + divS]
Mathematica writes the result in terms of 8 Standard Matrix Elements.

To have a simpler result we substitute these elements by simpler
expressions (ME[1],...ME[8]). But they are not all independent. The
final result can just be written in terms of 4 Matrix Elements.
{StandardMatrixElement[p2epk u[p1,m1] . ga[6] . u[p2,m2]],
StandardMatrixElement[p2epk u[p1,m1] . ga[7] . u[p2,m2]],
StandardMatrixElement[p2epk u[p1,m1] . gs[k] . ga[6] . u[p2,m2]],
StandardMatrixElement[p2epk u[p1,m1] . gs[k] . ga[7] . u[p2,m2]],
StandardMatrixElement[u[p1,m1] . gs[ep[k]] . ga[6] . u[p2,m2]],
StandardMatrixElement[u[p1,m1] . gs[ep[k]] . ga[7] . u[p2,m2]],
StandardMatrixElement[u[p1,m1] . gs[k] . gs[ep[k]] . ga[6] . u[p2,m2]],
StandardMatrixElement[u[p1,m1] . gs[k] . gs[ep[k]] . ga[7]. u[p2,m2]]} *)

ans1=res;
Set @@ {var, {ME[1],ME[2],ME[3],ME[4],ME[5],ME[6],ME[7],ME[8]}}
identities={ME[3]->-m1 ME[1] + m2 ME[2], ME[4]->-m1 ME[2] + m2 ME[1],
ME[7]->-m1 ME[5] - m2 ME[6] + 2 ME[1],
ME[8]->-m1 ME[6] - m2 ME[5] + 2 ME[2]}
ans2 =ans1 /. identities ;

ans=Simplify[ans2];
CR=Coefficient[ans,ME[1]]/2;
CL=Coefficient[ans,ME[2]]/2;
DR=Coefficient[ans,ME[5]];
DL=Coefficient[ans,ME[6]];
(* Test to see if we did not forget any term *)
test1:=Simplify[ans-2 CR*ME[1]-2 CL*ME[2]-DR*ME[5]-DL*ME[6]]
(* Test that the divergences cancel term by term *)
auxCL=PaVeReduce[CL] /. div ;
testdivCL:=Simplify[Coefficient[auxCL,Div]]
auxCR=PaVeReduce[CR] /. div ;
testdivCR:=Simplify[Coefficient[auxCR,Div]]
auxDL=PaVeReduce[DL] /. div ;
testdivDL:=Simplify[Coefficient[auxDL,Div]]
auxDR=PaVeReduce[DR] /. div ;
testdivDR:=Simplify[Coefficient[auxDR,Div]]
(* Test the gauge invariance relations *)

testGI1:=Simplify[PaVeReduce[(m2^2-m1^2)*CR - DR*m1 + DL*m2]]
testGI2:=Simplify[PaVeReduce[(m2^2-m1^2)*CL + DR*m2 - DL*m1]]
(********************** End Program mueg-ns.m *************************)
We first do the tests. The output of mathematica is

(********************** Mathematica output *************************)

In[3]:= << FeynCalc.m
FeynCalc4.1.0.3b Type ?FeynCalc for help or visit

http://www.feyncalc.org
In[4]:= << mueg-ns.m
In[5]:= test1
Out[5]= 0
In[6]:= testdiv
Out[6]= 0
In[7]:= testdivCL
Out[7]= 0
In[8]:= testdivCR
Out[8]= 0
In[9]:= testdivDL
Out[9]= 0
In[10]:= testdivDR
Out[10]= 0
In[11]:= testGI1
Out[11]= 0
In[12]:= testGI2
Out[12]= 0
(******************* End of Mathematica output *********************)
Now we obtain the results for CL


In[13]:= CL
2 2 2 2 2
Out[13]= (-4 AL BL mf C0[0, m2 , m1 , mf , mf , ms ] +
2 2 2 2 2
4 AL BR m2 PaVe[2, {0, m1 , m2 }, {mf , mf , ms }] -
22 2 2 2
4 AL BL mf PaVe[2, {0, m1 , m2 }, {mf , mf , ms }] -
2 2 2 2 2
4 AR BL m1 PaVe[1, 2, {0, m1 , m2 }, {mf , mf , ms }] +
2 2 2 2 2
4 AL BR m2 PaVe[1, 2, {0, m1 , m2 }, {mf , mf , ms }] +
2 2 2 2 2
4 AL BR m2 PaVe[2, 2, {0, m1 , m2 }, {mf , mf , ms }]) / 4
and for CR
In[15]:= CR
2 2 2 2 2
Out[15]= (-4 AR BR mf C0[0, m2 , m1 , mf , mf , ms ] +
2 2 2 2 2
4 AR BL m2 PaVe[2, {0, m1 , m2 }, {mf , mf , ms }] -
2 2 2 2 2
4 AR BR mf PaVe[2, {0, m1 , m2 }, {mf , mf , ms }] -
2 2 2 2 2
4 AL BR m1 PaVe[1, 2, {0, m1 , m2 }, {mf , mf , ms }] +
2 2 2 2 2
4 AR BL m2 PaVe[1, 2, {0, m1 , m2 }, {mf , mf , ms }] +
2 2 2 2 2
4 AR BL m2 PaVe[2, 2, {0, m1 , m2 }, {mf , mf , ms }]) / 4
(******************* End of Mathematica output **********************)
The expressions for DL,R are quite complicated. They are not normally calculated because
they can be related to CL,R by gauge invariance. However the power of this automatic
program can be illustrated by asking for these functions. As they are very long we calculate
them by pieces. We just calculate DL because one can easily check that DR = DL (L ↔ R).
In[12]:= Coefficient[PaVeReduce[DL],AL BL]
2 2 2 2 2 2
m1 mf B0[m1 , mf , ms ] m1 mf B0[m2 , mf , ms ]
Out[12]= ----------------------- - ----------------------- +
2 2 2 2
m1 - m2 m1 - m2
2 2 2 2 2
m1 mf C0[m1 , m2 , 0, mf , ms , mf ]
In[13]:= Coefficient[PaVeReduce[DL],AL BR]
2 2 2 2
(mf - ms ) B0[0, mf , ms ]
Out[13]= --------------------------- -
2 m1 m2
2 2 2 2 2 2
(m1 m2 - m2 mf + m2 ms ) B0[m1 , mf , ms ]
-------------------------------------------- +
2 2
2 m1 (m1 - m2 )
2 2 2 2 2 2
(m1 m2 - m1 mf + m1 ms ) B0[m2 , mf , ms ]
--------------------------------------------
2 2
2 m2 (m1 - m2 )
In[14]:= Coefficient[PaVeReduce[DL],AR BL]
2 2 2 2 2
1 (-2 m1 mf + 2 m1 ms ) B0[m1 , mf , ms ]
Out[14]= - - ---------------------------------------- +
2 2 2
2 m1 (m1 - m2 )
2 2 2 2 2
(-2 m2 mf + 2 m2 ms ) B0[m2 , mf , ms ]
----------------------------------------
2 2
2 m2 (m1 - m2 )
2 2 2 2 2 2
+ mf C0[m1 , m2 , 0, mf , ms , mf ]
In[15]:= Coefficient[PaVeReduce[DL],AR BR]
2 2 2 2 2 2
m2 mf B0[m1 , mf , ms ] m2 mf B0[m2 , mf , ms ]
Out[15]= ----------------------- - -----------------------
2 2 2 2
m1 - m2 m1 - m2
2 2 2 2 2
+ m2 mf C0[m1 , m2 , 0, mf , ms , mf ]
(******************* End of Mathematica output **********************)

From these expressions one can immediately verify that the divergences cancel in DL,R
and that they are not present in CL,R . To finish this section we just rewrite the CL,R in
our usual notation. We get

e Qℓ 2 2 2 2 2 2 2 2 2 2

CL = AL B L m F −C 0 (0, m 2 , m 1 , m F , m F , m S ) − C 2 (0, m 1 , m 2 , m F , m F , m S )
16π 2

+AL BR m2 C2 (0, m21 , m22 , m2F , m2F , m2S ) + C12 (0, m21 , m22 , m2F , m2F , m2S )

+C22 (0, m21 , m22 , m2F , m2F , m2S )

2 2 2 2 2
+ AR BL m1 C12 (0, m1 , m2 , mF , mF , mS ) (C.172)
CR = CL (L ↔ R) (C.173)
These equations are in agreement with Eqs. (32-34) and Eqs. (38-39) of Ref. [17], although
some work has to be done in order to verify that3 . This has to do with the fact that the
PV decomposition functions are not independent (see the Appendix for further details on
this point). We can however use the power of FeynCalc to verify this. We list below a
simple program to accomplish that.
(******************** Program lavoura-ns.m **************************)

(*
This program tests the results of my program mueg-ns.m against the
results obtained by L. Lavoura (hepph/0302221).
*)
(* First load FeynCalc.m and mueg-ns.m *)
3
An important difference between our conventions and those of Ref.[17] is that p1 and p2 (and obviously
m1 and m2 ) are interchanged.
<< FeynCalc.m
<< mueg-ns.m
(*
Now write Lavoura integrals in the notation of FeynCalc. Be careful
with the order of the entries.
*)
c1:=PaVe[1,{m2^2,0,m1^2},{ms^2,mf^2,mf^2}]
c2:=PaVe[2,{m2^2,0,m1^2},{ms^2,mf^2,mf^2}]
d1:=PaVe[1,1,{m2^2,0,m1^2},{ms^2,mf^2,mf^2}]
d2:=PaVe[2,2,{m2^2,0,m1^2},{ms^2,mf^2,mf^2}]
f:=PaVe[1,2,{m2^2,0,m1^2},{ms^2,mf^2,mf^2}]
(* Write Eqs. (32)-(34) of hepph/0302221 in our notation *)
k1:=PaVeReduce[m2*(c1+d1+f)]
k2:=PaVeReduce[m1*(c2+d2+f)]
k3:=PaVeReduce[mf*(c1+c2)]
(*
Now test the results. For this we should use the equivalences:
\rho -> AL BR
\lambda -> AR BL
\xi -> AR BR
\nu -> AL BL
*)
testCLALBR:=Simplify[PaVeReduce[Coefficient[CL, AL BR]-k1]]
testCLARBL:=Simplify[PaVeReduce[Coefficient[CL, AR BL]-k2]]
testCLALBL:=Simplify[PaVeReduce[Coefficient[CL, AL BL]-k3]]
testCRALBR:=Simplify[PaVeReduce[Coefficient[CR, AL BR]-k2]]
testCRARBL:=Simplify[PaVeReduce[Coefficient[CR, AR BL]-k1]]
testCRARBR:=Simplify[PaVeReduce[Coefficient[CR, AR BR]-k3]]
(****************** End of Program lavoura-ns.m **********************)
One can easily check that the output of the six tests is zero, showing the equivalence
between our results. And all this is done in a few seconds.
C.11.2 Charged scalar neutral fermion loop

We consider now the case of the scalar being charged and the scalar neutral. The general
case of both charged [17] can also be easily implemented, but for simplicity we do not
consider it here. The couplings are now
l- F0
S -i (A P + A P ) S + i ( B P +B P )
L L R R L L R R
F0 l-
and the diagrams contributing to the process are given in Fig. C.8, where all the denomi-
k
k k
D1 D1
D’3 D’2
p2 p1 p2 q p1 p2 q p1
q D5 D7
D’1 D4 D6
1) 2) 3)
Figure C.8:
nators are as in Eqs. (C.159)- (C.161) except that
D1′ = q 2 − m2F ; D2′ = (q − p1 )2 − m2S ; D3′ = (q − p1 − k)2 − m2S (C.174)
Also the coupling of the photon to the charged scalar is, in our notation,
−ie Qℓ (−2q + p1 + p2 )µ (C.175)
The procedure is very similar to the neutral scalar case and we just present here the
mathematica program and the final result. All the checks of finiteness and gauge invariance
can be done as before.
(************************ Program mueg-cs.m ***************************)

(*
This program calculates the COMPLETE (both the 3 point amplitude and
the two self energy type on each external line) amplitudes for
\mu -> e \gamma when the fermion line in the loop is neutral and the
charged line is a scalar. The \mu has momentum p2 and mass m2, the
electron (p1,m1) and the photon momentum k. The momentum in the loop
is q.
The assumed vertices are,
1) Electron-Scalar-Fermion:
Spinor[p1,m1] (AL P_L + AR P_R) Spinor [pf,mf]
2) Fermion-Scalar-Muon:
Spinor[pf,mf] (BL P_L + BR P_R) Spinor [p2,m2]

*)
dm[mu_]:=DiracMatrix[mu,Dimension->4]
L:=dm[7]
R:=dm[6]
(*
*)
gA:= AL DiracMatrix[7] + AR DiracMatrix[6]

gB:= BL DiracMatrix[7] + BR DiracMatrix[6]
num1:= Spinor[p1,m1] . gA . (ds[q]+mf) . gB . Spinor[p2,m2] \

PolarizationVector[k,mu] ( - 2 fv[q,mu] + fv[p1,mu] + fv[p2,mu] )
num11:=DiracSimplify[num1];
num2:=Spinor[p1,m1] . gA . (ds[q]+ds[p1]+mf) . gB . (ds[p1]+m2) . \

ds[Polarization[k]] . Spinor[p2,m2]
num3:=Spinor[p1,m1] . ds[Polarization[k]] . (ds[p2]+m1) . gA . \

(ds[q]+ds[p2]+mf) . gB . Spinor[p2,m2]
amp1:=num1 \
FeynAmpDenominator[PropagatorDenominator[q,mf],\
PropagatorDenominator[q-p1,ms],\
PropagatorDenominator[q-p1-k,ms]]
amp2:=num2 \
FeynAmpDenominator[PropagatorDenominator[q+p1,mf], \
PropagatorDenominator[p2-k,m2], \
amp3:=num3 \
FeynAmpDenominator[PropagatorDenominator[p1+k,m1], \
onshell={ScalarProduct[p1,p1]->m1^2,ScalarProduct[p2,p2]->m2^2, \
ScalarProduct[k,k]->0,ScalarProduct[p1,k]->(m2^2-m1^2)/2, \
ScalarProduct[p2,k]->(m2^2-m1^2)/2, \
ScalarProduct[p2,Polarization[k]]->p2epk, \
ScalarProduct[p1,Polarization[k]]->p2epk}
div={B0[m1^2,mf^2,ms^2]->Div,B0[m2^2,mf^2,ms^2]->Div, \
B0[0,mf^2,ms^2]->Div,B0[0,mf^2,mf^2]->Div,B0[0,ms^2,ms^2]->Div}

res:=res1+res2+res3 /. onshell
auxT1:= res1 /.onshell

auxT2:= PaVeReduce[auxT1]
auxT3:= auxT2 /. div
divT:=Simplify[Div*Coefficient[auxT3,Div]]
auxS1:= res2 + res3 /.onshell

auxS2:= PaVeReduce[auxS1]
auxS3:= auxS2 /. div
divS:=Simplify[Div*Coefficient[auxS3,Div]]
(* Check cancellation of divergences
testdiv should be zero because divT=-divS
*)
testdiv:=Simplify[divT + divS]
Mathematica writes the result in terms of 6 Standard Matrix Elements.

To have a simpler result we substitute these elements by simpler
expressions (ME[1],...ME[6]). Not all are independent.
{StandardMatrixElement[p2epk u[p1, m1] . ga[6] . u[p2, m2]],
StandardMatrixElement[p2epk u[p1, m1] . ga[7] . u[p2, m2]],
StandardMatrixElement[p2epk u[p1, m1] . gs[k] . ga[6] . u[p2, m2]],
StandardMatrixElement[p2epk u[p1, m1] . gs[k] . ga[7] . u[p2, m2]],
StandardMatrixElement[u[p1, m1] . gs[ep[k]] . ga[6] . u[p2, m2]],
StandardMatrixElement[u[p1, m1] . gs[ep[k]] . ga[7] . u[p2, m2]]}

*)
ans1=res;
Set @@ {var, {ME[1],ME[2],ME[3],ME[4],ME[5],ME[6]}}

identities={ME[3]->-m1 ME[1] + m2 ME[2],ME[4]->-m1 ME[2] + m2 ME[1]}
ans2 =ans1 /. identities ;

ans=Simplify[ans2];
CR=Coefficient[ans,ME[1]]/2;
CL=Coefficient[ans,ME[2]]/2;
DR=Coefficient[ans,ME[5]];
DL=Coefficient[ans,ME[6]];
(* Test to see if we did not forget any term *)
test1:=Simplify[ans-2*CR*ME[1]-2*CL*ME[2]-DR*ME[5]-DL*ME[6]]
(* Test that the divergences cancel term by term *)
auxCL:=PaVeReduce[CL] /. div ;
testdivCL:=Simplify[Coefficient[auxCL,Div]]
auxCR:=PaVeReduce[CR] /. div ;
testdivCR:=Simplify[Coefficient[auxCR,Div]]
auxDL:=PaVeReduce[DL] /. div ;
testdivDL:=Simplify[Coefficient[auxDL,Div]]
auxDR:=PaVeReduce[DR] /. div ;
testdivDR:=Simplify[Coefficient[auxDR,Div]]
(* Test the gauge invariance relations *)

testGI1:=PaVeReduce[(m2^2-m1^2)*CR - DR*m1 + DL*m2]
testGI2:=PaVeReduce[(m2^2-m1^2)*CL + DR*m2 - DL*m1]
(********************** End Program mueg-cs.m ***********************)
Note that although these programs look large, in fact they are very simple. Most of it are
comments and tests. The output of this program gives,
(********************* Mathematica output ************************)

In[3]:= CL
2 2 2 2 2
Out[3]= (-2 AR BL m1 C0[0, m1 , m2 , ms , ms , mf ] -
2 2 2 2 2
2 AR BL m1 PaVe[1, {m1 , 0, m2 }, {mf , ms , ms }] -
2 2 2 2 2
4 AR BL m1 PaVe[1, {m1 , m2 , 0}, {ms , mf , ms }] -
2 2 2 2 2
2 AL BL mf PaVe[1, {m1 , m2 , 0}, {ms , mf , ms }] -
2 2 2 2 2
2 AL BR m2 PaVe[2, {m1 , 0, m2 }, {mf , ms , ms }] -
2 2 2 2 2
2 AR BL m1 PaVe[2, {m1 , m2 , 0}, {ms , mf , ms }] +
2 2 2 2 2
2 AL BR m2 PaVe[2, {m1 , m2 , 0}, {ms , mf , ms }] -
2 2 2 2 2
2 AR BL m1 PaVe[1, 1, {m1 , m2 , 0}, {ms , mf , ms }] -
2 2 2 2 2
2 AR BL m1 PaVe[1, 2, {m1 , m2 , 0}, {ms , mf , ms }] +
2 2 2 2 2
2 AL BR m2 PaVe[1, 2, {m1 , m2 , 0}, {ms , mf , ms }]) / 2
(******************* End of Mathematica output ********************)
To finish this section we just rewrite the CL,R in our usual notation. We get

e Qℓ
CL = 2
AL BL mF −C1 (m21 , m22 , 0, m2S , m2F , m2S )
16π

+AL BR m2 − C2 (m21 , 0, m22 , m2F , m2S , m2S ) + C2 (m21 , m22 , 0, m2S , m2F , m2S )

+C12 (m21 , m22 , 0, m2S , m2F , m2S )
+ AR BL m1 −C0 (0, m21 , m22 , m2S , m2S , m2F ) − C1 (m21 , 0, m22 , m2F , m2S , m2S )
−2C1 (m21 , m22 , 0, m2S , m2F , m2S ) − C2 (m21 , m22 , 0, m2S , m2F , m2S )

−C11 (m21 , m22 , 0, m2S , m2F , m2S ) − C12 (m21 , m22 , 0, m2S , m2F , m2S )
CR = CL (L ↔ R) (C.176)
It is left as an exercise to write a mathematica program that proves that these equations
are in agreement with Eqs. (35-37) and Eqs. (38-39) of Ref. [17].
Appendix D
Feynman Rules for the Standard

Model
D.1 Introduction
To do actual calculations it is very important to have all the Feynman rules with consistent
conventions. In this Appendix we will give the complete Feynman rules for the Standard
Model in the general Rξ gauge.
D.2 The Standard Model

One of the most difficult problems in having a consistent set of of Feynman rules are the
conventions. We give here those that are important for building the SM. We will separate
them by gauge group.
D.2.1 Gauge Group SU(3)c

Here the important conventions are for the field strengths and the covariant derivatives.
We have
Gaµν = ∂µ Gaν − ∂ν Gaµ + gf abc Gbµ Gcν , a = 1, . . . , 8 (D.1)
where f abc are the group structure constants, satisfying

h i
T a , T b = if abc T c (D.2)
and T a are the generators of the group. The covariant derivative of a (quark) field q in
some representation T a of the gauge group is given by

Dµ q = ∂µ − i g Gaµ T a q (D.3)
In QCD the quarks are in the fundamental representation and T a = λa /2 where λa are
the Gell-Mann matrices. A gauge transformation is given by a matrix
a αa
U = e−iT (D.4)
373
374 APPENDIX D. FEYNMAN RULES FOR THE STANDARD MODEL
and the fields transform as

a αa
q → e−iT q δq = −iT a αa q
i 1
Gaµ T a → U Gaµ T a U −1 − ∂µ U U −1 δGaµ = − ∂µ αa + f abc αb Gcµ (D.5)
g g
where the second column is for infinitesimal transformations. With these definitions one
can verify that the covariant derivative transforms like the field itself,
δ(Dµ q) = −i T a αa (Dµ q) (D.6)
ensuring the gauge invariance of the Lagrangian.
D.2.2 Gauge Group SU(2)L

This is similar to the previous case. We have
a
Wµν = ∂µ Wνa − ∂ν Wµa + gǫabc Wµb Wνc , a = 1, . . . , 3 (D.7)
where, for the fundamental representation of SU (2)L we have T a = σ a /2 and ǫabc is the
completely anti-symmetric tensor in 3 dimensions. The covariant derivative for any field
ψL transforming non-trivially under this group is,

Dµ ψL = ∂µ − i g Wµa T a ψL (D.8)
D.2.3 Gauge Group U(1)Y

In this case the group is abelian and we have
Bµν = ∂µ Bν − ∂ν Bµ (D.9)
with the covariant derivative given by

Dµ ψR = ∂µ + i g ′ Y Bµ ψR (D.10)
where Y is the hypercharge of the field. Notice the different sign convention between
Eq. (D.8) and Eq. (D.9). This is to have the usual definition1
Q = T3 + Y . (D.12)
It is useful to write the covariant derivative in terms of the mass eigenstates Aµ and
Zµ . These are defined by the relations,
( (
Wµ3 = Zµ cos θW − Aµ sin θW Zµ = Wµ3 cos θW + Bµ sin θW
, . (D.13)
Bµ = Zµ sin θW + Aµ cos θW Aµ = −Wµ3 sin θW + Bµ cos θW
1
For this to be consistent one must also have, under hypercharge transformations, for a field of hyper-
charge Y ,
1
ψ ′ = e+iY αY ψ, Bµ′ = Bµ − ′ ∂µ αY . (D.11)
g
This is important when finding the ghost interactions. It would have been possible to have a minus sign
in Eq. (D.10), with a definition θW → θW + π. This would also mean reversing the sign in the exponent
of the hypercharge transformation in Eq. (D.11) maintaining the similarity with Eq. (D.5).
D.2. THE STANDARD MODEL 375
Field ℓL ℓR νL uL dL uR dR φ+ φ0
T3 − 21 0 1
2
1
2 − 12 0 0 1
2 − 12
Y − 21 −1 − 12 1
6
1
6
2
3 − 13 1
2
1
2
2
Q −1 −1 0 3 − 13 2
3 − 13 1 0
Table D.1: Values of T3f , Q and Y for the SM particles.
For a field ψL , with hypercharge Y , we get,

g + + − −
g 3 ′
Dµ ψL = ∂µ − i √ τ Wµ + τ Wµ − i τ3 Wµ + ig Y Bµ ψL (D.14)
2 2

g + + − −
g τ3 2
= ∂µ − i √ τ Wµ + τ Wµ + ie Q Aµ − i − Q sin θW Zµ ψL
2 cos θW 2
where, as usual, τ ± = (τ1 ± iτ2 )/2 and the charge operator is defined by
1 
2 +Y 0
Q=  , (D.15)
0 − 21 + Y
and we have used the relations,
e = g sin θW = g ′ cos θW , (D.16)
and the usual definition,

Wµ1 ∓ i Wµ2
Wµ± = √ . (D.17)
2
For a singlet of SU (2)L , ψR we have,

Dµ ψR = ∂µ + ig ′ Y Bµ ψR

g 2
= ∂µ + ie Q Aµ + i Q sin θW Zµ ψR . (D.18)
cos θW
We collect in Table D.1 the quantum number of the SM particles.
D.2.4 The Gauge Field Lagrangian

For completeness we write the gauge field Lagrangian. We have
1 1 a aµν 1
Lgauge = − Gaµν Gaµν − Wµν W − Bµν B µν (D.19)
4 4 4
where the field strengths are given in Eqs. (D.1), and (D.9).
D.2.5 The Fermion Fields Lagrangian

Here we give the kinetic part and gauge interaction, leaving the Yukawa interaction for a
next section. We have
X X X
LFermion = iqγ µ Dµ q + iψL γ µ Dµ ψL + iψR γ µ Dµ ψR (D.20)
quarks ψL ψR
where the covariant derivatives are obtained with the rules in Eqs. (D.3), (D.14) and
(D.18).
D.2.6 The Higgs Lagrangian

In the SM we use an Higgs doublet with the following assignments,
 
φ+
 
Φ =  v + H + iϕZ  (D.21)
√
2
The hypercharge of this doublet is 1/2 and therefore the covariant derivative reads

g + + − −
g 3 g′
Dµ Φ = ∂µ − i √ τ Wµ + τ Wµ − i τ3 Wµ + i Bµ Φ (D.22)
2 2 2

g + + − −
g τ3 2
= ∂µ − i √ τ Wµ τ Wµ + ie Q Aµ − i − Q sin θW Zµ Φ
2 cos θW 2
The Higgs Lagrangian is then
2
LHiggs = (Dµ Φ)† Dµ Φ + µ2 Φ† Φ − λ Φ† Φ (D.23)
If we expand this Lagrangian we find the following terms

1 1 1 1
LHiggs = · · · + g2 v 2 Wµ3 W µ3 + g′2 v 2 Bµ B µ + gg′ v 2 Wµ3 B µ + g2 v 2 Wµ+ W −µ
8 8 4 4
1 i i
+ v ∂ µ ϕZ g′ Bµ + gWµ3 + gvWµ− ∂ µ ϕ+ − gvWµ+ ∂ µ ϕ− (D.24)
2 2 2
The first three terms give, after diagonalization, a massless field, the photon, and a massive
one, the Z, with the relations given in Eq. (D.13), while the fourth gives the mass to the
charged Wµ± boson. Using Eq. (D.13) we get,
1
LHiggs = · · · + MZ2 Zµ Z µ + MW
2
Wµ+ W −µ
2

+ MZ Zµ ∂ µ ϕZ + iMW Wµ− ∂ µ ϕ+ − Wµ+ ∂ µ ϕ− (D.25)
where
1 1 1 1
MW = gv, MZ = gv = MW (D.26)
2 cos θW 2 cos θW
By looking at Eq. (D.25) we realize that besides finding a realistic spectra for the gauge
bosons, we also got a problem. In fact the terms in the last line are quadratic in the fields
and complicate the definition of the propagators. We now see how one can use the needed
gauge fixing to solve also this problem.
D.2. THE STANDARD MODEL 377
D.2.7 The Yukawa Lagrangian

Now we have to spell out the interaction between the fermions and the Higgs doublet that
after spontaneous symmetry breaking gives masses to the elementary fermions. We have,
e uR + h.c.
LYukawa = − Yl L Φ ℓR − Yd Q Φ dR − Yu Q Φ (D.27)
where sum is implied over generations, L (Q) are the lepton (quark) doublets and,
 
v + H − iϕZ
e = i σ2 Φ ∗ =  √ 
Φ 2 (D.28)
−ϕ−
D.2.8 The Gauge Fixing

As it is well known, we have to gauge fix the gauge part of the Lagrangian to be able to
define the propagators. We will use a generalization of the class of Lorenz gauges, the
so-called Rξ gauges. With this choice the gauge fixing Lagrangian reads
1 2 1 1 1
LGF = − FG − FA2 − FZ2 − F− F+ (D.29)
2ξ 2ξ 2ξ ξ
where
FGa =∂ µ Gaµ , FA = ∂ µ Aµ , FZ = ∂ µ Zµ − ξMZ ϕZ
F+ =∂ µ Wµ+ − iξMW ϕ+ , F− = ∂ µ Wµ− + iξMW ϕ− (D.30)
One can easily verify that with these definitions we cancel the quadratic terms in Eq. (D.25).
D.2.9 The Ghost Lagrangian

The last piece in writing the SM Lagrangian is the ghost Lagrangian. As it is well known,
this is given by the Fadeev-Popov prescription,
4
X
∂(δF+ ) ∂(δF+ ) ∂(δFZ ) ∂(δFA )
LGhost = c+ + c− + cZ + cA ci
∂αi ∂αi ∂αi ∂αi
i=1
8
X ∂(δFGa ) b
+ ωa ω (D.31)
∂β b
a,b=1
where we have denoted by ω a the ghosts associated with the SU (3)c transformations
defined by,
a a
U = e−iT β , a = 1, . . . , 8 (D.32)
and by c± , cA , cZ the electroweak ghosts associated with the gauge transformations,
a αa 4
U = e−iT , a = 1, . . . , 3, U = eiY α (D.33)
For completeness we write here the gauge transformations of the gauge fixing terms needed
to find the Lagrangian in Eq. (D.31). It is convenient to redefine the parameters as
α1 ∓ α2
α± = √
2
αZ =α3 cos θW + α4 sin θW
αA = − α3 sin θW + α4 cos θW (D.34)
We then get
δFGa = − ∂µ β a + gs f abc β b Gcµ
δFA = − ∂µ αA
δFZ =∂µ (δZ µ ) − MZ δϕZ
δF+ =∂µ (δWµ+ ) − iMW δϕ+
δF− =∂µ (δWµ− ) + iMW δϕ− (D.35)
Using the explicit form of the gauge transformations we can finally find the missing pieces,

δZµ = − ∂µ αZ + ig cos θW Wµ+ α− − Wµ− α+ (D.36)

δWµ+ = − ∂µ α+ + ig α+ (Zµ cos θW − Aµ sin θw ) − (αZ cos θw − αA sin θW ) Wµ+

δWµ− = − ∂µ α− − ig α− (Zµ cos θW − Aµ sin θw ) − (αZ cos θw − αA sin θW ) Wµ−
and
1 g
δϕZ = − g α− ϕ+ + α+ ϕ− + αZ (v + H)
2 2 cos θW
g g cos 2θW +
δϕ+ = − i (v + H + iϕZ )α+ − i ϕ αZ + ie ϕ+ αA
2 2 cos θW
g g cos 2θW −
δϕ− =i (v + H − iϕZ )α− + i ϕ αZ − ie ϕ− αA (D.37)
2 2 cos θW
D.2.10 The Complete SM Lagrangian

Finally the complete Lagrangian for the Standard Model is obtained putting together all
the pieces. We have,
LSM = Lgauge + LFermion + LHiggs + LYukawa + LGF + LGhost (D.38)
where the different terms were given in Eqs. (D.19), (D.20), (D.23), (D.27), (D.29), (D.31).
D.3. THE FEYNMAN RULES FOR QCD 379
D.3 The Feynman Rules for QCD

We give separately the Feynman Rules for QCD and the electroweak part of the Standard
Model.
D.3.1 Propagators
g
gµν kµ kν
µ, a ν, b −iδab − (1 − ξ) 2 2 (D.39)
2
k + iǫ (k )
ω i
a b δab (D.40)
k2 + iǫ
D.3.2 Triple Gauge Interactions

ρ, c
gf abc [ gµν (p1 − p2 )ρ + gνρ (p2 − p3 )µ
p3
p2 +gρµ (p3 − p1 )ν ]
p1
µ, a ν, b p1 + p2 + p3 = 0
(D.41)
D.3.3 Quartic Gauge Interactions

ii) Vértice quártico dos bosões de gauge
σ, d ρ, c
h
p4 p3 −ig 2 feab fecd (gµρ gνσ − gµσ gνρ )
p1 p2 +feac fedb (gµσ gρν − gµν gρσ )

i (D.42)
µ, a ν, b +fead febc (gµν gρσ − gµρ gνσ )
p1 + p2 + p3 + p4 = 0
D.3.4 Fermion Gauge Interactions

µ, a
p3
p2 ig(γ µ )βα Tija (D.43)
p1
β, i α, j
D.3.5 Ghost Interactions
µ, c
p3
g C abc pµ1
p2 (D.44)
p1 p1 +p2 + p3 = 0
a b
D.4 The Feynman Rules for the Electroweak Theory

D.4.1 Propagators
γ
gµν kµ kν
µ ν −i 2 − (1 − ξ) 2 2 (D.45)
k + iǫ (k )
W −igµν
µ ν 2 + iǫ (D.46)
k2 − MW
Z −igµν
µ ν (D.47)
k2 − MZ2 + iǫ
i(p/ + mf )
(D.48)
p p2 − m2f + iǫ
h i
(D.49)
p p2 − Mh2 + iǫ
ϕZ i
(D.50)
p p2 − ξm2Z + iǫ
ϕ± i
(D.51)
p p2 − ξm2W + iǫ
D.4. THE FEYNMAN RULES FOR THE ELECTROWEAK THEORY 381
D.4.2 Triple Gauge Interactions
Wα−
p q
Aµ
−ie [gαβ (p − k)µ + gβµ (k − q)α + gµα (q − p)β ] (D.52)
k
Wβ+
Wα−
p q
Zµ
ig cos θW [gαβ (p − k)µ + gβµ (k − q)α + gµα (q − p)β ] (D.53)
k
Wβ+
D.4.3 Quartic Gauge Interactions
Wα+ Wβ−
−ie2 [2gαβ gµµ − gαµ gβν − gαν gβµ ] (D.54)
Aµ Aν
Wα+ Wβ−
−ig 2 cos2 θW [2gαβ gµν − gαµ gβν − gαν gβµ ] (D.55)
Zµ Zν
Wα+ Wβ−
ieg cos θW [2gαβ gµν − gαµ gβν − gαν gβµ ] (D.56)
Aµ Zν
Wα+ Wβ−
ig 2 [2gαµ gβν − gαβ gµν − gαν gβµ ] (D.57)
Wµ+ Wν−
D.4.4 Charged Current Interaction
ψu,d
Wµ±
g 1 − γ5
i √ γµ (D.58)
2 2
ψd,u
D.4.5 Neutral Current Interaction

ψf ψf
Zµ g Aµ
i γµ gVf − gA
f
γ5 −ieQf γµ (D.59)
cos θW
ψf ψf
where
1 3 1
gVf = T − Qf sin2 θW , f
gA = Tf3 . (D.60)
2 f 2
D.4.6 Fermion-Higgs and Fermion-Goldstone Interactions
f
h g mf
−i (D.61)
2 mW
f
f
ϕZ mf
−g Tf3 γ5 (D.62)
mW
f
ψd,u
ϕ∓
g mu md
i√ PR,L − PL,R (D.63)
2 mW mW
ψu,d
D.4.7 Triple Higgs-Gauge and Goldstone-Gauge Interactions

ϕ+
p+
Aµ
−i e (p+ − p− )µ (D.64)
p−
ϕ−
ϕ+
p+
Zµ cos 2θW
ig (p+ − p− )µ (D.65)
2 cos θW
p−
ϕ−
h
p
Wµ± i
∓ g (k − p)µ (D.66)
2
k
ϕ∓
ϕZ
p
Wµ± g
(k − p)µ (D.67)
2
k
ϕ∓
h
p
Zµ g
(k − p)µ (D.68)
2 cos θ
k
ϕZ
ϕ∓
Aµ
−ie mW gµν (D.69)
Wν±
ϕ∓
Zµ
−ig mZ sin2 θW gµν (D.70)
Wν±
Wµ±
ig mW gµν (D.71)
Wν∓
Zµ g
i mZ gµν (D.72)
cos θW
Zν
D.4.8 Quartic Higgs-Gauge and Goldstone-Gauge Interactions
h Wµ±
i 2
g gµν (D.73)
2
h Wν∓
ϕZ Wµ±
i 2
g gµν (D.74)
2
ϕZ Wν∓
h Zµ
i g2
gµν (D.75)
2 cos2 θW
h Zν
ϕZ Zµ
i g2
gµν (D.76)
2 cos2 θW
ϕZ Zν
ϕ+ Aµ
2i e2 gµν (D.77)
ϕ− Aν
ϕ+ Zµ
2
i g cos 2θW
gµν (D.78)
2 cos θW
ϕ− Zν
ϕ+ Wµ+
i 2
g gµν (D.79)
2
ϕ− Wν−
ϕ∓ Wµ±
sin2 θW
−i g 2 gµν (D.80)
2 cos θW
h Zν
ϕ± Wµ∓
sin2 θW
∓ g2 gµν (D.81)
2 cos θW
ϕZ Zν
ϕ± Wµ∓
i
− eg gµν (D.82)
2
h Aν
ϕ∓ Wµ±
1
± eg gµν (D.83)
2
ϕZ Aν
ϕ+ Zµ
cos 2θW
−i eg gµν (D.84)
cos θW
ϕ− Aν
D.4.9 Triple Higgs and Goldstone Interactions

ϕ+
h
i m2
− g h (D.85)
2 mW
ϕ−
h
h
3 m2
− i g 2h (D.86)
2 mW
h
ϕZ
h
i m2
− g 2h (D.87)
2 mW
ϕZ
D.4.10 Quartic Higgs and Goldstone Interactions
ϕ+ ϕ−
i m2
− g2 2h (D.88)
2 mW
ϕ+ ϕ−
ϕ+ h
i m2
− g2 2h (D.89)
4 mW
ϕ− h
ϕ+ ϕZ
i m2
− g2 2h (D.90)
4 mW
ϕ− ϕZ
h h
3 m2
− i g2 2h (D.91)
4 mW
h h
ϕZ h
i m2
− g 2 2h (D.92)
4 mW
ϕZ h
ϕZ ϕZ
3 m2
− i g2 2h (D.93)
4 mW
ϕZ ϕZ
D.4.11 Ghost Propagators
cA i
(D.94)
k2 + iǫ
c± i
(D.95)
k2 − ξm2W + iǫ
cZ i
(D.96)
k2 − ξm2Z + iǫ
D.4.12 Ghost Gauge Interactions
c±
p
Aµ
∓ie pµ (D.97)
c±
c±
p
Zµ
±ig cos θW pµ (D.98)
c±
c±
p
Wµ±
∓ ig cos θW pµ (D.99)
cZ
c±
p
Wµ±
± ie pµ (D.100)
cA
cZ
p
Wµ∓
∓ ig cos θW pµ (D.101)
c±
cA
p
Wµ∓
± ie pµ (D.102)
c±
D.4.13 Ghost Higgs and Ghost Goldstone Interactions
c±
ϕZ g
± ξ mW (D.103)
2
c±
c±
h i
− g ξ mW (D.104)
2
c±
cZ
h ig
− ξ mZ (D.105)
2 cos θW
cZ
cZ
ϕ∓ i
g ξ mZ (D.106)
2
c±
c±
ϕ± cos 2θW
−ig ξ mW (D.107)
2 cos θW
cZ
c±
ϕ±
ie ξ mW (D.108)
cA
Bibliography
[1] J. Bjorken and S. Drell, Relativistic Quantum Mechanics (McGraw-Hill, New York,
1962).
[2] S. N. Gupta, Proc. Phys. Soc. 63, 681 (1950).
[3] K. Bleuler, Helv. Phys. Acta 23, 567 (1950).
[4] S. S. Schweber, An Introduction to Relativistic Quantum Field Theory (Dover Pub-

lications, 1962).
[5] J. C. Romão, Introdução à Teoria do Campo (IST, 2012), Available online at

http://porthos.ist.utl.pt/ftp/textos/itc.pdf.
[6] J. D. Bjorken and S. D. Drell, Relativistic Quantum Fields (McGraw-Hill, N.Y.,

1964).
[7] C. Itzykson and J. B. Zuber, Quantum Field Theory (McGraw-Hill, New York, 1980).
[8] H. Lehmann, K. Symanzik and W. Zimmerman, Nuovo Cimento 1, 2005 (1955).
[9] J. C. Romão, Utilities for One Loop Calculations (IST, 2004), Available online at
http://porthos.ist.utl.pt/OneLoop/.
[10] G. Passarino and M. J. G. Veltman, Nucl. Phys. B160, 151 (1979).
[11] A. Denner, Fortschr. Phys. 41, 307 (1993).
[12] R. Mertig, M. Bohm and A. Denner, Comput. Phys. Commun. 64, 345 (1991).
[13] R. Mertig, http://www.feyncalc.org .
[14] T. Hahn and M. Perez-Victoria, Comput. Phys. Commun. 118, 153 (1999), [hep-
ph/9807565].
[15] T. Hahn, http://www.feynarts.de/looptools .
[16] G. J. van Oldenborgh, Comput. Phys. Commun. 66, 1 (1991).
[17] L. Lavoura, Eur. Phys. J. C29, 191 (2003), [hep-ph/0302221].
391

IST Quantum Field Theory Notes

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

IST Quantum Field Theory Notes

Uploaded by

Copyright:

Available Formats

INSTITUTO SUPERIOR TÉCNICO

Advanced Quantum Field Theory

1 Free Field Quantization 9

2 Physical States. S Matrix. LSZ Reduction. 51

2.7.2 Spectral representation fermions . . . . . . . . . . . . . . . . . . . . 62

3 Covariant Perturbation Theory 73

4 Radiative Corrections 105

5 Functional Methods 141

5.5.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

6 Non Abelian Gauge Theories 179

7 Renormalization Group 247

A Path Integral in Quantum Mechanics 291

B Path Integral in Quantum Field Theory 307

C Useful techniques for renormalization 321

D Feynman Rules for the Standard Model 373

D.2.9 The Ghost Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . 377

IST, May 2012

Free Field Quantization

1.1 General formalism

H(p, q) = pq̇ − L(q, q̇) (1.4)

In terms of H the equations of motion are,

{p, q}PB = 1 . (1.8)

|ΨS (t)i = e−iHt |ΨS (0)i = e−iHt |ΨH i . (1.11)

OH (t) = eiHt OS e−iHt (1.12)

dOH (t) ∂OH

[p(t), q(t)] = −i (1.16)

[pi (t), qj (t)] = −iδij (1.19)

ṗ = i[H, p] = −ω02 q (1.24)

It is convenient to introduce the operators

The equations of motion for a and a† are very simple:

ȧ(t) = −iω0 a(t) e ȧ† (t) = iω0 a† (t) . (1.27)

They have the solution

a(t) = a0 e−iω0 t ; a† (t) = a†0 eiω0 t (1.28)

[a, a† ] = [a0 , a†0 ] = 1 (1.29)

In terms of a, a† the Hamiltonian reads

1.1.2 Canonical quantization for fields

in order to obtain the Klein-Gordon equations as the equations of motion,

Therefore the canonical momentum is now

and the Hamiltonian

[pi (t), ϕj (t)] = −iδij

[ϕi (t), ϕj (t)] = 0

[pi (t), pj (t)] = 0 (1.51)

In terms of momentum πi (t) we have

Going into the continuum limit, ∆Vi → 0, we obtain

[ϕ(~x, t), ϕ(~x′ , t)] = 0 (1.53)

[ϕr (~x, t), ϕs (~x′ , t)] = 0 (1.56)

1.1.3 Symmetries and conservation laws

Using these relations we can define the conserved quantities

ii) Lorentz transformations

δϕr (x) ≡ ϕ′r (x) − ϕr (x) = Srs

The conserved angular momentum is then

which leads to the conserved charge

ϕ′r (x′ ) = Srs (a)ϕs (x) (1.81)

should be in the quantum theory

where aµ ν e bµ are deﬁned by

U (a, b)ϕr (x)U −1 (a, b) = Srs

U (b)ϕr (x)U −1 (b) = ϕr (x + b) (1.86)

For inﬁnitesimal translations we can write

where P µ is an hermitian operator. Then Eq. (1.86) gives

i[P µ , ϕr (x)] = ∂ µ ϕr (x) (1.88)

aµ ν = gνµ + ω µ ν + O(ω 2 ) (1.89)

i[Mµν , ϕr (x)] = xµ ∂ ν ϕr − xν ∂ µ ϕr + Σµν

1.2 Quantization of scalar fields

and therefore Z −1/2